Merge branch 'develop' of https://github.com/ROCm/composable_kernel into develop

example/01_gemm/gemm_xdl_streamk.cpp

@@ -27,11 +27,15 @@ using DeviceGemmStreamK = ck::tensor_operation::device::DeviceGemmXdlStreamK
 // ######|      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|
 // ######|          |          |          |            |        |        |        |            |            |            |       |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+#if defined(CK_USE_AMD_MFMA_GFX950)
+    < ADataType, BDataType, CDataType, AccDataType, ALayout, BLayout, CLayout,  AElementOp,  BElementOp,  CElementOp,    256,   256,   128,     4, 16,   32,   32,    4,    2,     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,              4,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>;
+#else // defined(CK_USE_AMD_MFMA_GFX950)
     < 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>;
+#endif // defined(CK_USE_AMD_MFMA_GFX950)
 
 
 

example/20_grouped_conv_bwd_weight/CMakeLists.txt

@@ -13,3 +13,9 @@ add_example_dependencies(example_grouped_conv_bwd_weight example_grouped_conv_bw
 
 add_example_executable(example_grouped_conv_bwd_weight_dl_fp16 grouped_conv_bwd_weight_dl_fp16.cpp)
 add_example_dependencies(example_grouped_conv_bwd_weight example_grouped_conv_bwd_weight_dl_fp16)
+
+add_example_executable(example_grouped_conv_bwd_weight_v3_xdl_bf16 grouped_conv_bwd_weight_v3_xdl_bf16.cpp)
+add_example_dependencies(example_grouped_conv_bwd_weight example_grouped_conv_bwd_weight_v3_xdl_bf16)
+
+add_example_executable(example_grouped_conv_bwd_weight_v3_xdl_fp16 grouped_conv_bwd_weight_v3_xdl_fp16.cpp)
+add_example_dependencies(example_grouped_conv_bwd_weight example_grouped_conv_bwd_weight_v3_xdl_fp16)

example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_v3_xdl_bf16.cpp

@@ -0,0 +1,102 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2025, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_xdl_cshuffle_v3.hpp"
+
+using InDataType = BF16;
+// bf16 kernel use fp32 atomic add to accumulate Weight tensor into global memory
+using WeiDataType = F32;
+using OutDataType = BF16;
+using AccDataType = F32;
+
+using InElementOp  = PassThrough;
+using WeiElementOp = PassThrough;
+using OutElementOp = PassThrough;
+
+template <ck::index_t NDimSpatial>
+using DeviceConvBwdWeightInstance =
+    // clang-format on
+    ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffleV3<
+        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
+        32,                   // K0PerBlock
+        8,                    // K1
+        32,                   // MPerXdl
+        32,                   // NPerXdl
+        2,                    // MXdlPerWave
+        2,                    // NXdlPerWave
+        S<4, 16, 1>,          // ABlockTransferThreadClusterLengths_K0_M_K1
+        S<2, 0, 1>,           // ABlockTransferThreadClusterArrangeOrder
+        S<1, 0, 2>,           // ABlockTransferSrcAccessOrder
+        2,                    // ABlockTransferSrcVectorDim
+        1,                    // ABlockTransferSrcScalarPerVector
+        2,                    // ABlockTransferDstScalarPerVector_K1
+        true,                 // ABlockLdsAddExtraM
+        S<4, 16, 1>,          // BBlockTransferThreadClusterLengths_K0_N_K1
+        S<2, 0, 1>,           // BBlockTransferThreadClusterArrangeOrder
+        S<1, 0, 2>,           // BBlockTransferSrcAccessOrder
+        2,                    // BBlockTransferSrcVectorDim
+        1,                    // BBlockTransferSrcScalarPerVector
+        2,                    // BBlockTransferDstScalarPerVector_K1
+        true,                 // BBlockLdsAddExtraN
+        1,                    // CShuffleMXdlPerWavePerShuffle
+        1,                    // CShuffleNXdlPerWavePerShuffle
+        S<1, 32, 1, 4>,       // CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock
+        128 / (sizeof(WeiDataType) * CHAR_BIT)>; // CBlockTransferScalarPerVector_NWaveNPerXdl
+                                                 // clang-format off
+
+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 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;
+}

example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_v3_xdl_fp16.cpp

@@ -0,0 +1,99 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2025, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_xdl_cshuffle_v3.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_Xdl_CShuffleV3<
+        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
+        32,                   // K0PerBlock
+        8,                    // K1
+        32,                   // MPerXdl
+        32,                   // NPerXdl
+        2,                    // MXdlPerWave
+        2,                    // NXdlPerWave
+        S<4, 16, 1>,          // ABlockTransferThreadClusterLengths_K0_M_K1
+        S<2, 0, 1>,           // ABlockTransferThreadClusterArrangeOrder
+        S<1, 0, 2>,           // ABlockTransferSrcAccessOrder
+        2,                    // ABlockTransferSrcVectorDim
+        1,                    // ABlockTransferSrcScalarPerVector
+        2,                    // ABlockTransferDstScalarPerVector_K1
+        false,                // ABlockLdsAddExtraM
+        S<4, 16, 1>,          // BBlockTransferThreadClusterLengths_K0_N_K1
+        S<2, 0, 1>,           // BBlockTransferThreadClusterArrangeOrder
+        S<1, 0, 2>,           // BBlockTransferSrcAccessOrder
+        2,                    // BBlockTransferSrcVectorDim
+        1,                    // BBlockTransferSrcScalarPerVector
+        2,                    // BBlockTransferDstScalarPerVector_K1
+        false,                // BBlockLdsAddExtraN
+        1,                    // CShuffleMXdlPerWavePerShuffle
+        1,                    // CShuffleNXdlPerWavePerShuffle
+        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[])
+{
+    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;
+}

example/67_gemm_microscaling/gemm_mx_common.hpp

@@ -13,7 +13,9 @@
 #include "ck/utility/blkgemmpipe_scheduler.hpp"
 #include "ck/utility/data_type.hpp"
 #include "ck/utility/sequence.hpp"
+
 #include "ck/library/reference_tensor_operation/cpu/reference_mx_gemm.hpp"
+
 #include "ck/library/utility/check_err.hpp"
 #include "ck/library/utility/device_memory.hpp"
 #include "ck/library/utility/fill.hpp"

example/ck_tile/01_fmha/codegen/ops/fmha_bwd.py

@@ -492,6 +492,7 @@ def get_bwd_dq_dk_dv_blobs(kernel_filter : Optional[str], receipt, mask_impl) ->
             if kernel_filter != None:
                 if not fnmatch.fnmatch(k.name, kernel_filter):
                     continue
+            # Flash attention integration
             if receipt == 2:
                     cond = dtype in ['fp16', 'bf16']
                     cond &= bias in ['no', 'alibi']
@@ -499,14 +500,15 @@ def get_bwd_dq_dk_dv_blobs(kernel_filter : Optional[str], receipt, mask_impl) ->
                     cond &= dpad == dvpad
                     if not cond:
                         continue
-            if receipt == 3:
+            elif receipt == 3:
                     cond = dtype in ['fp16', 'bf16']
                     cond &= bias in ['no', 'alibi']
                     cond &= dpad == dvpad
                     cond &= deterministic == "f"
                     if not cond:
                         continue
-            if receipt == 4:
+            # PyTorch integration
+            elif receipt == 4:
                     cond = dtype in ['fp16', 'bf16']
                     cond &= bias in ['no', 'bias']
                     cond &= dropout in ['no', 'dropout_wg32',  'dropout_wg16']
@@ -514,6 +516,26 @@ def get_bwd_dq_dk_dv_blobs(kernel_filter : Optional[str], receipt, mask_impl) ->
                     cond &= deterministic == "f"
                     if not cond:
                         continue
+            # Aiter (mha_bwd) integration
+            elif receipt == 10:
+                    cond = dtype in ['fp16', 'bf16']
+                    cond &= mode == "batch"
+                    cond &= bias in ['no', 'alibi']
+                    cond &= dropout in ['no', 'dropout_wg32',  'dropout_wg16']
+                    cond &= dpad == dvpad
+                    cond &= deterministic == "t"
+                    if not cond:
+                        continue
+            # Aiter (mha_varlen_bwd) integration
+            elif receipt == 11:
+                    cond = dtype in ['fp16', 'bf16']
+                    cond &= mode == "group"
+                    cond &= bias in ['no', 'alibi']
+                    cond &= dropout in ['no', 'dropout_wg32',  'dropout_wg16']
+                    cond &= dpad == dvpad
+                    cond &= deterministic == "t"
+                    if not cond:
+                        continue
             api_pool.register_dq_dk_dv_traits(k.api_trait())
             gen.append(k)
 

example/ck_tile/01_fmha/codegen/ops/fmha_fwd.py

@@ -487,6 +487,7 @@ def get_fwd_blobs(kernel_filter : Optional[str], receipt, mask_impl) -> Tuple[Fm
                 if kernel_filter != None:
                     if not fnmatch.fnmatch(k.name, kernel_filter):
                         continue
+                # 2 - Flash attention integration
                 if receipt in (2, 3):
                     cond = dtype in ['fp16', 'bf16']
                     cond &= pipeline.F_vlayout == 'row'
@@ -494,13 +495,32 @@ def get_fwd_blobs(kernel_filter : Optional[str], receipt, mask_impl) -> Tuple[Fm
                     cond &= pipeline.F_squant == 'f'
                     if not cond:
                         continue
-                if receipt == 4:
+                # PyTorch integration
+                elif receipt == 4:
                     cond = dtype in ['fp16', 'bf16']
                     cond &= pipeline.F_vlayout == 'row'
                     cond &= pipeline.F_bias in ['no', 'bias']
                     cond &= pipeline.F_squant == 'f'
                     if not cond:
                         continue
+                # Aiter(mha_fwd) integration
+                elif receipt == 10:
+                    cond = dtype in ['fp16', 'bf16']
+                    cond &= mode == "batch"
+                    cond &= pipeline.F_vlayout == 'row'
+                    cond &= pipeline.F_bias in ['no', 'alibi']
+                    cond &= pipeline.F_squant == 'f'
+                    if not cond:
+                        continue
+                # Aiter(mha_varlen_fwd) integration
+                elif receipt == 11:
+                    cond = dtype in ['fp16', 'bf16']
+                    cond &= mode == "group"
+                    cond &= pipeline.F_vlayout == 'row'
+                    cond &= pipeline.F_bias in ['no', 'alibi']
+                    cond &= pipeline.F_squant == 'f'
+                    if not cond:
+                        continue
                 api_pool.register_traits(k.api_trait())
                 gen.append(k)
 

example/ck_tile/01_fmha/codegen/ops/fmha_fwd_appendkv.py

@@ -326,7 +326,9 @@ def get_fwd_appendkv_blobs(kernel_filter : Optional[str], receipt, mask_impl) ->
                 if kernel_filter != None:
                     if not fnmatch.fnmatch(k.name, kernel_filter):
                         continue
-                if receipt == 2:
+                # 2 - Flash attention integration
+                # 12 - Aiter(mha_fwd_kvcache) integration
+                if receipt in (2, 12):
                     cond = dtype in ['fp16', 'bf16']
                     cond &= pipeline.F_vlayout == 'row'
                     if not cond:

example/ck_tile/01_fmha/codegen/ops/fmha_fwd_splitkv.py

@@ -268,7 +268,7 @@ float fmha_fwd_splitkv(fmha_fwd_splitkv_traits t, fmha_fwd_splitkv_args a, const
 FMHA_FWD_SPLITKV_API_INNER_DISPATCH="""            {F_if}((t.is_group_mode == {F_mode}) && (t.is_v_rowmajor == {F_vlayout}) && ({F_mask_check}) && (t.bias_type == {F_bias_check}) && (t.do_fp8_static_quant == {F_squant}) &&
                         ((a.block_table_ptr != nullptr) == {F_pagedkv}) && ({F_scheck}) && ({F_skcheck}) && ({F_dcheck}) && ({F_dvcheck})) {{
                 using traits_ = fmha_fwd_splitkv_traits_<{F_hdim}, {F_dtype}, {F_mode}, {F_bm0}, {F_bn0}, {F_bk0}, {F_bn1}, {F_bk1}, {F_bk0max}, {F_vlayout}, {F_pipeline_enum}, {F_mask}, {F_bias}, true, {F_squant}, {F_pagedkv}, {F_spad}, {F_skpad}, {F_dpad}, {F_dvpad}>;
-                
+
                 // get combine kernel tile sizes
                 using OaccDataType = typename FmhaFwdTypeConfig<{F_dtype}>::OaccDataType;
                 constexpr ck_tile::index_t kM0 = ck_tile::BlockFmhaSplitKVCombinePipelineTileSizes<OaccDataType, /*F_bn1=*/32>::kM0;
@@ -705,6 +705,7 @@ def get_fwd_splitkv_blobs(kernel_filter : Optional[str], receipt, mask_impl) ->
                 if kernel_filter != None:
                     if not fnmatch.fnmatch(k.name, kernel_filter):
                         continue
+                # Flash attention integration
                 if receipt == 2:
                     cond = dtype in ['fp16', 'bf16']
                     cond &= pipeline.F_vlayout == 'row'
@@ -712,6 +713,24 @@ def get_fwd_splitkv_blobs(kernel_filter : Optional[str], receipt, mask_impl) ->
                     cond &= pipeline.F_squant == 'f'
                     if not cond:
                         continue
+                # Aiter(mha_varlen_fwd) integration
+                elif receipt == 11:
+                    cond = dtype in ['fp16', 'bf16']
+                    cond &= mode == "group"
+                    cond &= pipeline.F_vlayout == 'row'
+                    cond &= pipeline.F_bias in ['no', 'alibi']
+                    cond &= pipeline.F_squant == 'f'
+                    if not cond:
+                        continue
+                # Aiter(mha_fwd_kvcache) integration
+                elif receipt == 12:
+                    cond = dtype in ['fp16', 'bf16']
+                    cond &= mode == "batch"
+                    cond &= pipeline.F_vlayout == 'row'
+                    cond &= pipeline.F_bias in ['no', 'alibi']
+                    cond &= pipeline.F_squant == 'f'
+                    if not cond:
+                        continue
                 api_pool.register_traits(k.api_trait())
                 gen.append(k)
 

example/ck_tile/01_fmha/generate.py

@@ -17,7 +17,7 @@ class HandlerId(IntEnum):
     LIST_BLOBS = 0
     WRITE_BLOBS = 1
 
-# inspect all modules under 'codegen.ops' and register API handlers 
+# inspect all modules under 'codegen.ops' and register API handlers
 ops = []
 for importer, module_name, _ in pkgutil.iter_modules(codegen.ops.__path__):
     full_module_name = '%s.%s' % (codegen.ops.__name__, module_name)
@@ -104,7 +104,11 @@ if __name__ == "__main__":
         help="codegen receipt. 0: generate only 8xhdim coverage\n"  + \
              "  1: generate more instance to cover all hdim\n"  + \
              "  2: Only generate instance for Flash attention integration\n"  + \
-             "  4: Only generate instance for PyTorch integration"
+             "  4: Only generate instance for PyTorch integration\n" + \
+             " 10: Only generate instance for Aiter(mha_fwd, mha_bwd) integration\n" + \
+             " 11: Only generate instance for Aiter(mha_varlen_fwd, mha_varlen_bwd) integration\n" + \
+             " 12: Only generate instance for Aiter(mha_fwd_kvcache) integration"
+      
     )
 
     args = parser.parse_args()

example/ck_tile/03_gemm/run_gemm_example.inc

@@ -107,10 +107,10 @@ int run_gemm_example_with_layouts(int argc,
     ck_tile::index_t stride_B = arg_parser.get_int("stride_b");
     ck_tile::index_t stride_C = arg_parser.get_int("stride_c");
 
-    ck_tile::index_t kbatch = arg_parser.get_int("split_k");
-    int n_warmup            = arg_parser.get_int("warmup");
-    int n_repeat            = arg_parser.get_int("repeat");
-    ck_tile::index_t init_method = arg_parser.get_int("init");  
+    ck_tile::index_t kbatch      = arg_parser.get_int("split_k");
+    int n_warmup                 = arg_parser.get_int("warmup");
+    int n_repeat                 = arg_parser.get_int("repeat");
+    ck_tile::index_t init_method = arg_parser.get_int("init");
 
     stride_A = ck_tile::get_default_stride(M, K, stride_A, is_row_major(a_layout));
     stride_B = ck_tile::get_default_stride(K, N, stride_B, is_row_major(b_layout));
@@ -123,16 +123,23 @@ int run_gemm_example_with_layouts(int argc,
     ck_tile::HostTensor<CDataType> c_m_n_dev_result(
         ck_tile::host_tensor_descriptor(M, N, stride_C, is_row_major(CLayout{})));
 
-    if (init_method == 0) {
+    if(init_method == 0)
+    {
         ck_tile::FillUniformDistribution<ADataType>{-1.f, 1.f}(a_m_k);
         ck_tile::FillUniformDistribution<BDataType>{-1.f, 1.f}(b_k_n);
-    } else if (init_method == 1) {
+    }
+    else if(init_method == 1)
+    {
         ck_tile::FillMonotonicSeq<ADataType>{}(a_m_k);
         ck_tile::FillMonotonicSeq<BDataType>{}(b_k_n);
-    } else if (init_method == 2) {
+    }
+    else if(init_method == 2)
+    {
         ck_tile::FillConstant<ADataType>{static_cast<ADataType>(1)}(a_m_k);
         ck_tile::FillConstant<BDataType>{static_cast<BDataType>(1)}(b_k_n);
-    } else {
+    }
+    else
+    {
         a_m_k.SetZero();
         b_k_n.SetZero();
     }

include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_streamk_v3.hpp

@@ -230,6 +230,23 @@ struct GridwiseGemm_xdl_cshuffle_streamk_v3
             }
         }();
 
+        // Pad both M and K to be multiples of the block sizes
+        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;
+#if 0
         using GemmSpecialization = tensor_operation::device::GemmSpecialization;
 
         if constexpr(GemmSpec == GemmSpecialization::MKPadding ||
@@ -296,6 +313,7 @@ struct GridwiseGemm_xdl_cshuffle_streamk_v3
 
             return a_grid_desc_ak0_m_ak1;
         }
+#endif
     }
 
     __device__ static auto MakeBGridDescriptor_BK0_N_BK1(
@@ -312,6 +330,23 @@ struct GridwiseGemm_xdl_cshuffle_streamk_v3
             }
         }();
 
+        // Pad both N and K to be multiples of the block sizes
+        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;
+#if 0     
         using GemmSpecialization = tensor_operation::device::GemmSpecialization;
 
         if constexpr(GemmSpec == GemmSpecialization::NKPadding ||
@@ -378,6 +413,7 @@ struct GridwiseGemm_xdl_cshuffle_streamk_v3
 
             return b_grid_desc_bk0_n_bk1;
         }
+#endif
     }
 
     template <typename ABlockDesc_AK0_M_AK1>
@@ -412,6 +448,13 @@ struct GridwiseGemm_xdl_cshuffle_streamk_v3
             }
         }();
 
+        // Pad both M and N to be multiples of the block sizes
+        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>{}));
+#if 0
         using GemmSpecialization = tensor_operation::device::GemmSpecialization;
 
         if constexpr(GemmSpec == GemmSpecialization::MNPadding ||
@@ -449,6 +492,7 @@ struct GridwiseGemm_xdl_cshuffle_streamk_v3
             // not pad M or N
             return c_grid_desc_mraw_nraw;
         }
+#endif
     }
 
     struct Problem
@@ -953,7 +997,8 @@ struct GridwiseGemm_xdl_cshuffle_streamk_v3
         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))
+                       GemmSpec == tensor_operation::device::GemmSpecialization::MNKPadding) &&
+                     !(is_same<tensor_layout::gemm::RowMajor, ALayout>::value))
         {
             if(!(karg.M % MPerBlock == 0))
             {
@@ -970,7 +1015,8 @@ struct GridwiseGemm_xdl_cshuffle_streamk_v3
         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))
+                       GemmSpec == tensor_operation::device::GemmSpecialization::MNKPadding) &&
+                     (is_same<tensor_layout::gemm::RowMajor, BLayout>::value))
         {
             if(!(karg.N % NPerBlock == 0))
             {
@@ -1036,6 +1082,7 @@ struct GridwiseGemm_xdl_cshuffle_streamk_v3
                               << ABlockTransferSrcScalarPerVector << " )! " << __FILE__ << ":"
                               << __LINE__ << ", in function: " << __func__ << std::endl;
                 }
+
                 return false;
             }
         }
@@ -1051,6 +1098,10 @@ struct GridwiseGemm_xdl_cshuffle_streamk_v3
                               << BBlockTransferSrcScalarPerVector << " )! " << __FILE__ << ":"
                               << __LINE__ << ", in function: " << __func__ << std::endl;
                 }
+                std::cout << "Arg N (" << karg.N
+                          << ") value is not a multiple of BBlockTransferSrcScalarPerVector ("
+                          << BBlockTransferSrcScalarPerVector << " )! " << __FILE__ << ":"
+                          << __LINE__ << ", in function: " << __func__ << std::endl;
                 return false;
             }
         }
@@ -1065,6 +1116,7 @@ struct GridwiseGemm_xdl_cshuffle_streamk_v3
                               << BBlockTransferSrcScalarPerVector << " )! " << __FILE__ << ":"
                               << __LINE__ << ", in function: " << __func__ << std::endl;
                 }
+
                 return false;
             }
         }
@@ -1082,6 +1134,7 @@ struct GridwiseGemm_xdl_cshuffle_streamk_v3
                               << __FILE__ << ":" << __LINE__ << ", in function: " << __func__
                               << std::endl;
                 }
+
                 return false;
             }
         }
@@ -1098,17 +1151,8 @@ struct GridwiseGemm_xdl_cshuffle_streamk_v3
                               << __FILE__ << ":" << __LINE__ << ", in function: " << __func__
                               << std::endl;
                 }
-                return false;
-            }
-        }
 
-        if constexpr(is_same<remove_cvref_t<CDataType>, bhalf_t>::value)
-        {
-            if(ck::EnvIsEnabled(CK_ENV(CK_LOGGING)))
-            {
-                std::cout << " Grid size: " << karg.Grid_size << " > 1 is not support yet"
-                          << __FILE__ << ":" << __LINE__ << ", in function: " << __func__
-                          << std::endl;
+                return false;
             }
         }
 

include/ck/tensor_operation/operator_transform/transform_conv_bwd_weight_to_gemm_v2.hpp

@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2024-2025, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -34,6 +34,94 @@ struct TransformConvBwdWeightToGemmV2
     static constexpr auto I0 = Number<0>{};
     static constexpr auto I1 = Number<1>{};
 
+    template <index_t NDim, typename enable_if<NDim == 1, bool>::type = false>
+    constexpr static auto
+    make_out_grid_desc(const index_t N,
+                       const index_t Wo,
+                       const index_t K,
+                       const std::array<index_t, NDimSpatial + 3>& output_strides)
+    {
+        const index_t BatchStride = output_strides[0];
+        const index_t WoStride    = output_strides[3];
+        const auto KStride        = Number<1>{};
+        return make_naive_tensor_descriptor(make_tuple(N * Wo, NumGroupsToMerge, K),
+                                            make_tuple(WoStride, BatchStride, KStride));
+    }
+
+    template <index_t NDim, typename enable_if<NDim == 1, bool>::type = false>
+    constexpr static auto
+    make_in_grid_desc(const index_t N,
+                      const index_t Wi,
+                      const index_t C,
+                      const std::array<index_t, NDimSpatial + 3>& input_strides)
+    {
+        const index_t BatchStride = input_strides[0];
+        const index_t NStride     = input_strides[1];
+        const index_t WiStride    = input_strides[3];
+        const auto CStride        = input_strides[2];
+        if constexpr(ConvBackwardWeightSpecialization ==
+                     device::ConvolutionBackwardWeightSpecialization::Filter1x1Stride1Pad0)
+        {
+            return make_naive_tensor_descriptor(make_tuple(N * Wi, NumGroupsToMerge, C),
+                                                make_tuple(WiStride, BatchStride, CStride));
+        }
+        else
+        {
+            return make_naive_tensor_descriptor(
+                make_tuple(N, Wi, NumGroupsToMerge, C),
+                make_tuple(NStride, WiStride, BatchStride, CStride));
+        }
+    }
+
+    template <index_t NDim, typename enable_if<NDim == 1, bool>::type = false>
+    constexpr static auto
+    make_wei_grid_desc(const index_t K,
+                       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];
+        const auto XStride     = weights_strides[3];
+        const auto BatchStride = weights_strides[0];
+        // Add NumGroupsToMerge for Batch+M dimension and, 1 as a placehorder
+        // for Batch+N dimension
+        const auto desc = make_naive_tensor_descriptor(
+            make_tuple(NumGroupsToMerge, K, X, 1, C),
+            make_tuple(BatchStride, KStride, XStride, BatchStride, CStride));
+        // Padd 1 to NumGroupsToMerge
+        const auto padded_desc = transform_tensor_descriptor(
+            desc,
+            make_tuple(make_pass_through_transform(NumGroupsToMerge),
+                       make_pass_through_transform(K),
+                       make_pass_through_transform(X),
+                       make_pad_transform(1, 0, NumGroupsToMerge - 1),
+                       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>{}));
+        // We need only matrices from diagonal. Xor returns 0 for the same
+        // values. So if matrices is not on diagonal then it will be stored in padding.
+        // To avoid use of modulo after xor we assume that NumBatch to merge is power of 2.
+        static_assert(NumGroupsToMerge == 1 || NumGroupsToMerge == 2 || NumGroupsToMerge == 4 ||
+                      NumGroupsToMerge == 8 || NumGroupsToMerge == 16 || NumGroupsToMerge == 32 ||
+                      NumGroupsToMerge == 64);
+        const auto unmerged_padded_desc = transform_tensor_descriptor(
+            padded_desc,
+            make_tuple(make_xor_transform(make_tuple(NumGroupsToMerge, NumGroupsToMerge)),
+                       make_pass_through_transform(K),
+                       make_pass_through_transform(X),
+                       make_pass_through_transform(C)),
+            make_tuple(Sequence<0, 3>{}, Sequence<1>{}, Sequence<2>{}, Sequence<4>{}),
+            make_tuple(Sequence<0, 3>{}, Sequence<1>{}, Sequence<2>{}, Sequence<4>{}));
+        // Merge To M, N
+        return transform_tensor_descriptor(
+            unmerged_padded_desc,
+            make_tuple(make_merge_transform(make_tuple(NumGroupsToMerge, K)),
+                       make_merge_transform(make_tuple(X, NumGroupsToMerge, C))),
+            make_tuple(Sequence<0, 1>{}, Sequence<2, 3, 4>{}),
+            make_tuple(Sequence<0>{}, Sequence<1>{}));
+    }
+
     template <index_t NDim, typename enable_if<NDim == 2, bool>::type = false>
     constexpr static auto
     make_out_grid_desc(const index_t N,
@@ -221,6 +309,187 @@ struct TransformConvBwdWeightToGemmV2
             make_tuple(Sequence<0>{}, Sequence<1>{}));
     }
 
+    template <index_t NDim, typename enable_if<NDim == 1, 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,
+        const index_t batch_k)
+    {
+        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 ConvStrideW = conv_filter_strides[0];
+
+        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 index_t GemmKTotal = N * Wo;
+        const index_t GemmM      = K * NumGroupsToMerge;
+        const index_t GemmN      = C * X * NumGroupsToMerge;
+
+        const auto PadGemmM = MPerBlock - GemmM % MPerBlock;
+        const auto PadGemmN = NPerBlock - GemmN % NPerBlock;
+
+        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, Wo, K, output_strides);
+        const auto in_grid_desc  = make_in_grid_desc<NDim>(N, Wi, C, input_strides);
+        const auto wei_grid_desc = make_wei_grid_desc<NDim>(K, X, C, weights_strides);
+
+        if constexpr(ConvBackwardWeightSpecialization ==
+                     device::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_merge_transform(make_tuple(NumGroupsToMerge, GemmM / NumGroupsToMerge))),
+                make_tuple(Sequence<0>{}, Sequence<1, 2>{}),
+                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, 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_merge_transform(make_tuple(NumGroupsToMerge, GemmN / NumGroupsToMerge))),
+                make_tuple(Sequence<0>{}, Sequence<1, 2>{}),
+                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, 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_merge_transform(make_tuple(NumGroupsToMerge, GemmM / NumGroupsToMerge))),
+                make_tuple(Sequence<0>{}, Sequence<1, 2>{}),
+                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, 2>{}, Sequence<1>{}));
+
+            // B: input tensor
+            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(Wi, InLeftPadW, InRightPadW),
+                           make_pass_through_transform(NumGroupsToMerge),
+                           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(X, Wo), make_tuple(ConvDilationW, ConvStrideW)),
+                    make_pass_through_transform(NumGroupsToMerge),
+                    make_pass_through_transform(C)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
+                make_tuple(Sequence<0>{}, Sequence<1, 2>{}, Sequence<3>{}, Sequence<4>{}));
+
+            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(X, NumGroupsToMerge, C)),
+                           make_merge_transform(make_tuple(N, Wo))),
+                make_tuple(Sequence<1, 3, 4>{}, 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_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, 2>{}, Sequence<1>{}));
+
+            // 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 * 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(GemmKBatch * 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);
+        }
+
+    } // function end
+
     template <index_t NDim, typename enable_if<NDim == 2, bool>::type = false>
     static auto MakeABCGridDescriptor_A_K0_M_K1_B_K0_N_K1_C_M_N(
         const index_t N,

include/ck_tile/ops/fmha/block/block_masking.hpp

@@ -310,7 +310,7 @@ struct SimplifiedGenericAttentionMask
 
         const index_t x_per_split = ck_tile::max(1, integer_divide_ceil(x_total, num_splits));
         const index_t split_start = x_per_split * i_split;
-        const index_t split_end   = split_start + x_per_split;
+        const index_t split_end   = ck_tile::min(x_total, split_start + x_per_split);
 
         return ck_tile::make_tuple(ck_tile::max(origin_start, split_start),
                                    ck_tile::min(origin_end, split_end));

include/ck_tile/ops/fmha/kernel/fmha_fwd_splitkv_kernel.hpp

@@ -742,7 +742,7 @@ struct FmhaFwdSplitKVKernel
                 return pad_tensor_view(
                     v_dram_transposed,
                     make_tuple(number<FmhaPipeline::kN1>{}, number<FmhaPipeline::kK1>{}),
-                    sequence<kPadHeadDimV, false>{});
+                    sequence<kPadHeadDimV, kPadSeqLenK>{});
             }
             else
             {

library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_v3_xdl_instance.hpp

@@ -0,0 +1,112 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2025, 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_v3.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;
+
+#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>
+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,
+          BlockGemmPipelineScheduler Scheduler,
+          BlockGemmPipelineVersion PipelineVersion>
+using device_grouped_conv_bwd_weight_v3_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| BlockGemm| BlockGemm|
+        //#########################################|     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|  Pipeline|  Pipeline|
+        //#########################################| 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| Scheduler|   Version|
+        //#########################################|        |         |          |          |       |        |        |        |            |            |            |                          |      |      |      |      |   |     |     |     |     |                |                 |               |               |               |               |          |                |               |               |              |               |               |          |            |            | NBlock_NPerBlock|                |          |          |
+        // generic instance
+        DeviceGroupedConvBwdWeight_Xdl_CShuffleV3< NDimSpatial,  ALayout,   BLayout,   ELayout,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,    64,    16,    16,     32,   8,   16,   16,    1,    1,  S<4, 8,  1>, S<2, 0, 1>,  S<1, 0, 2>,                   1,              1,              4,      false,  S<4, 8,  1>,  S<2, 0, 1>,  S<1, 0, 2>,                1,              1,              4,      false,           1,           1,   S<1, 8, 1, 8>,                  2, Scheduler, PipelineVersion>
+    // clang-format on
+    >;
+
+template <ck::index_t NDimSpatial,
+          typename ALayout,
+          typename BLayout,
+          typename ELayout,
+          ConvolutionBackwardWeightSpecialization ConvSpec,
+          BlockGemmPipelineScheduler Scheduler,
+          BlockGemmPipelineVersion PipelineVersion>
+using device_grouped_conv_bwd_weight_v3_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| BlockGemm| BlockGemm|
+        //#########################################|     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|  Pipeline|  Pipeline|
+        //#########################################| 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| Scheduler|   Version|
+        //#########################################|        |         |          |          |       |        |        |        |            |            |            |                          |      |      |      |      |   |     |     |     |     |                |                 |               |               |               |               |          |                |               |               |              |               |               |          |            |            | NBlock_NPerBlock|                |          |          |
+        // generic instance
+        DeviceGroupedConvBwdWeight_Xdl_CShuffleV3< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,    64,    32,    32,     32,   8,   32,   32,    1,    1,  S<4, 8,  1>, S<2, 0, 1>,  S<1, 0, 2>,                   1,              2,              2,      false,  S<4, 16, 1>,  S<2, 0, 1>,  S<1, 0, 2>,                1,              2,              2,      false,           1,           1,   S<1, 8, 1, 8>,                  2, Scheduler, PipelineVersion>,
+        DeviceGroupedConvBwdWeight_Xdl_CShuffleV3< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,    64,    32,    64,     32,   8,   32,   32,    1,    2,  S<4, 8,  1>, S<2, 0, 1>,  S<1, 0, 2>,                   1,              4,              4,      false,  S<4, 16, 1>,  S<2, 0, 1>,  S<1, 0, 2>,                1,              4,              4,      false,           1,           1,   S<1, 8, 1, 8>,                  2, Scheduler, PipelineVersion>,
+        DeviceGroupedConvBwdWeight_Xdl_CShuffleV3< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,    64,    32,   128,     32,   8,   32,   32,    1,    4,  S<4, 4,  1>, S<2, 0, 1>,  S<1, 0, 2>,                   1,              8,              8,      false,  S<4, 16, 1>,  S<2, 0, 1>,  S<1, 0, 2>,                1,              8,              8,      false,           1,           1,   S<1, 8, 1, 8>,                  2, Scheduler, PipelineVersion>,
+        DeviceGroupedConvBwdWeight_Xdl_CShuffleV3< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,    64,    64,    32,     32,   8,   32,   32,    2,    1,  S<4, 16, 1>, S<2, 0, 1>,  S<1, 0, 2>,                   1,              4,              4,      false,  S<4, 8,  1>,  S<2, 0, 1>,  S<1, 0, 2>,                1,              4,              4,      false,           1,           1,   S<1, 8, 1, 8>,                  2, Scheduler, PipelineVersion>,
+        DeviceGroupedConvBwdWeight_Xdl_CShuffleV3< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,    64,   128,    32,     32,   8,   32,   32,    4,    1,  S<4, 16, 1>, S<2, 0, 1>,  S<1, 0, 2>,                   1,              8,              8,      false,  S<4, 4,  1>,  S<2, 0, 1>,  S<1, 0, 2>,                1,              8,              8,      false,           1,           1,   S<1, 8, 1, 8>,                  2, Scheduler, PipelineVersion>,       
+        DeviceGroupedConvBwdWeight_Xdl_CShuffleV3< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,    64,    64,    80,     32,   8,   16,   16,    4,    5,  S<4, 16, 1>, S<2, 0, 1>,  S<2, 0, 1>,                   1,              4,              4,      false,  S<4, 16, 1>,  S<2, 0, 1>,  S<2, 0, 1>,                1,              5,              4,      false,           1,           1,   S<1, 8, 1, 8>,                  2, Scheduler, PipelineVersion>,
+        DeviceGroupedConvBwdWeight_Xdl_CShuffleV3< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,    64,    64,   112,     32,   8,   16,   16,    4,    7,  S<4, 16, 1>, S<2, 0, 1>,  S<2, 0, 1>,                   1,              4,              4,      false,  S<4, 16, 1>,  S<2, 0, 1>,  S<2, 0, 1>,                1,              7,              4,      false,           1,           1,   S<1, 8, 1, 8>,                  2, Scheduler, PipelineVersion>
+    // clang-format on
+    >;
+
+template <ck::index_t NDimSpatial,
+          typename ALayout,
+          typename BLayout,
+          typename ELayout,
+          ConvolutionBackwardWeightSpecialization ConvSpec,
+          BlockGemmPipelineScheduler Scheduler,
+          BlockGemmPipelineVersion PipelineVersion>
+using device_grouped_conv_bwd_weight_v3_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| BlockGemm| BlockGemm|
+        //#########################################|     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|  Pipeline|  Pipeline|
+        //#########################################| 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| Scheduler|   Version|
+        //#########################################|        |         |          |          |       |        |        |        |            |            |            |                          |      |      |      |      |   |     |     |     |     |                |                 |               |               |               |               |          |                |               |               |              |               |               |          |            |            | NBlock_NPerBlock|                |          |          |
+        // generic instance
+        DeviceGroupedConvBwdWeight_Xdl_CShuffleV3< NDimSpatial,  ALayout,   BLayout,   ELayout,   BF16,    BF16,    BF16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,    64,    32,    32,     32,   8,   32,   32,    1,    1,  S<4, 8,  1>,  S<2, 0, 1>,  S<1, 0, 2>,                  1,              2,              2,      false,  S<4, 16,  1>, S<2, 0, 1>,  S<1, 0, 2>,                1,              2,              2,      false,           1,           1,   S<1, 8, 1, 8>,                  2, Scheduler, PipelineVersion>,
+        DeviceGroupedConvBwdWeight_Xdl_CShuffleV3< NDimSpatial,  ALayout,   BLayout,   ELayout,   BF16,    BF16,    BF16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,    64,    32,    64,     32,   8,   32,   32,    1,    2,  S<4, 8,  1>,  S<2, 0, 1>,  S<1, 0, 2>,                  1,              4,              4,      false,  S<4, 16,  1>, S<2, 0, 1>,  S<1, 0, 2>,                1,              4,              4,      false,           1,           1,   S<1, 8, 1, 8>,                  2, Scheduler, PipelineVersion>,
+        DeviceGroupedConvBwdWeight_Xdl_CShuffleV3< NDimSpatial,  ALayout,   BLayout,   ELayout,   BF16,    BF16,    BF16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,    64,    32,   128,     32,   8,   32,   32,    1,    4,  S<4, 4,  1>,  S<2, 0, 1>,  S<1, 0, 2>,                  1,              8,              8,      false,  S<4, 16,  1>, S<2, 0, 1>,  S<1, 0, 2>,                1,              8,              8,      false,           1,           1,   S<1, 8, 1, 8>,                  2, Scheduler, PipelineVersion>,
+        DeviceGroupedConvBwdWeight_Xdl_CShuffleV3< NDimSpatial,  ALayout,   BLayout,   ELayout,   BF16,    BF16,    BF16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,    64,    64,    32,     32,   8,   32,   32,    2,    1,  S<4, 16, 1>,  S<2, 0, 1>,  S<1, 0, 2>,                  1,              4,              4,      false,  S<4, 8,  1>,  S<2, 0, 1>,  S<1, 0, 2>,                1,              4,              4,      false,           1,           1,   S<1, 8, 1, 8>,                  2, Scheduler, PipelineVersion>,
+        DeviceGroupedConvBwdWeight_Xdl_CShuffleV3< NDimSpatial,  ALayout,   BLayout,   ELayout,   BF16,    BF16,    BF16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,    64,   128,    32,     32,   8,   32,   32,    4,    1,  S<4, 16, 1>,  S<2, 0, 1>,  S<1, 0, 2>,                  1,              8,              8,      false,  S<4, 4,  1>,  S<2, 0, 1>,  S<1, 0, 2>,                1,              8,              8,      false,           1,           1,   S<1, 8, 1, 8>,                  2, Scheduler, PipelineVersion>,
+        DeviceGroupedConvBwdWeight_Xdl_CShuffleV3< NDimSpatial,  ALayout,   BLayout,   ELayout,   BF16,    BF16,    BF16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,    64,    64,    80,     32,   8,   16,   16,    4,    5,  S<4, 16, 1>,  S<2, 0, 1>,  S<2, 0, 1>,                  1,              4,              4,      false,  S<4, 16,  1>,  S<2, 0, 1>,  S<2, 0, 1>,               1,              5,              4,      false,           1,           1,   S<1, 8, 1, 8>,                  2, Scheduler, PipelineVersion>,
+        DeviceGroupedConvBwdWeight_Xdl_CShuffleV3< NDimSpatial,  ALayout,   BLayout,   ELayout,   BF16,    BF16,    BF16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,    64,    64,   112,     32,   8,   16,   16,    4,    7,  S<4, 16, 1>,  S<2, 0, 1>,  S<2, 0, 1>,                  1,              4,              4,      false,  S<4, 16,  1>,  S<2, 0, 1>,  S<2, 0, 1>,               1,              7,              4,      false,           1,           1,   S<1, 8, 1, 8>,                  2, Scheduler, PipelineVersion>
+    //clang-format on
+    >;
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_comp_instance.hpp

@@ -61,8 +61,9 @@ using device_grouped_conv_fwd_xdl_bf16_comp_instances = std::tuple<
         //########################################|    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|
         //########################################|           |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-#if defined(__gfx950__)
-#else
+#if defined(CK_USE_AMD_MFMA_GFX950)
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,   256,   128,   128,    64,  16,  16,  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,  BlockGemmPipelineScheduler::Interwave, BlockGemmPipelineVersion::v1>
+#else // defined(CK_USE_AMD_MFMA_GFX950)
         // Compute friendly
         DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,   256,   256,   256,    32,   8,   8,  32,   32,    4,    4,     S<4, 64, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<4, 64, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v4>,
         DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,   256,   128,   128,    64,   8,   8,  32,   32,    2,    2,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v4>,
@@ -81,7 +82,7 @@ using device_grouped_conv_fwd_xdl_bf16_comp_instances = std::tuple<
         DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,   256,   128,    64,    64,   8,   8,  32,   32,    2,    1,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3>,
         DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,   256,    64,   128,    64,   8,   8,  32,   32,    1,    2,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3>,
         DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,   256,    64,    64,    64,   8,   8,  32,   32,    1,    1,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3>
-#endif // defined(__gfx950__)
+#endif // defined(CK_USE_AMD_MFMA_GFX950)
     // clang-format on
     >;
 
@@ -97,8 +98,9 @@ using device_grouped_conv_fwd_xdl_f16_comp_instances = std::tuple<
         //########################################|    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|
         //########################################|           |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-#if defined(__gfx950__)
-#else
+#if defined(CK_USE_AMD_MFMA_GFX950)
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,   256,   128,   128,    64,  16,  16,  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,  BlockGemmPipelineScheduler::Interwave, BlockGemmPipelineVersion::v1>
+#else // defined(CK_USE_AMD_MFMA_GFX950)
         DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,   256,   256,   256,    32,   8,   8,  32,   32,    4,    4,     S<4, 64, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<4, 64, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v4>,
         DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,   256,   128,   128,    64,   8,   8,  32,   32,    2,    2,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v4>,
         DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,   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,          0,    S<4, 64, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v4>,
@@ -113,7 +115,7 @@ using device_grouped_conv_fwd_xdl_f16_comp_instances = std::tuple<
         DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,   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,          0,    S<4, 64, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Interwave, BlockGemmPipelineVersion::v1>,
         DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,   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,          0,    S<4, 64, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Interwave, BlockGemmPipelineVersion::v1>,
         DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,   256,   128,   128,    64,   8,   8,  32,   32,    2,    2,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Interwave, BlockGemmPipelineVersion::v1>
-#endif // defined(__gfx950__)
+#endif // defined(CK_USE_AMD_MFMA_GFX950)
     // clang-format on
     >;
 
@@ -148,8 +150,9 @@ using device_grouped_conv_fwd_xdl_int8_comp_instances = std::tuple<
         //########################################|    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|
         //########################################|           |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-#if defined(__gfx950__)
-#else
+#if defined(CK_USE_AMD_MFMA_GFX950)
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout,DsLayout,ELayout,int8_t,int8_t,int32_t,   int8_t,    DsLayout,int8_t, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,             256,   128,   128,   128,  32,  32,   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>,              16,  BlockGemmPipelineScheduler::Interwave, BlockGemmPipelineVersion::v1>
+#else // defined(CK_USE_AMD_MFMA_GFX950)
         DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   int8_t,   int8_t,     int32_t,      int8_t,    DsLayout,   int8_t, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,   256,   256,   256,    32,   8,   8,  32,   32,    4,    4,     S<4, 64, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<4, 64, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v4>,
         DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   int8_t,   int8_t,     int32_t,      int8_t,    DsLayout,   int8_t, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,   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,          0,    S<4, 64, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v4>,
         DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   int8_t,   int8_t,     int32_t,      int8_t,    DsLayout,   int8_t, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,   256,   256,   256,    32,   8,   8,  32,   32,    4,    4,     S<4, 64, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<4, 64, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3>,
@@ -160,7 +163,7 @@ using device_grouped_conv_fwd_xdl_int8_comp_instances = std::tuple<
         DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   int8_t,   int8_t,     int32_t,      int8_t,    DsLayout,   int8_t, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,   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,          0,    S<4, 64, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Interwave, BlockGemmPipelineVersion::v1>,
         DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   int8_t,   int8_t,     int32_t,      int8_t,    DsLayout,   int8_t, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,   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,          0,    S<4, 64, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Interwave, BlockGemmPipelineVersion::v1>,
         DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   int8_t,   int8_t,     int32_t,      int8_t,    DsLayout,   int8_t, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,   256,   128,   128,    64,   8,   8,  32,   32,    2,    2,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Interwave, BlockGemmPipelineVersion::v1>
-#endif // defined(__gfx950__)
+#endif // defined(CK_USE_AMD_MFMA_GFX950)
     // clang-format on
     >;
 

library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_merged_groups_instance.hpp

@@ -45,13 +45,16 @@ using device_grouped_conv_fwd_xdl_merged_groups_bf16_instances = std::tuple<
         //########################################|    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|     Type|     Type|  Pipeline|   ToMerge|
         //########################################|           |       |       |            |       |      |      |        |         |            |      |   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|         |         | Scheduler|          |
         //########################################|           |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |         |         |          |          |
-#if defined(__gfx950__)
-#else
+#if defined(CK_USE_AMD_MFMA_GFX950)
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,                  ConvSpec, GemmMNKPadding,  1,  64,    64,    16,     32,   8, 4,  16,   16,    4,    1,  S< 4, 16,  1>, S<0, 2, 1>,     S<0, 2, 1>,                   1,              4,              4,      1,  S< 4, 16,  1>,   S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              4,      1,           1,           1,   S<1, 16, 1, 4>,                  1, BF16, BF16, LoopScheduler::Default, 8>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,                  ConvSpec, GemmMNKPadding,  1,  64,    64,    16,     32,   8, 4,  16,   16,    4,    1,  S< 4, 16,  1>, S<0, 2, 1>,     S<0, 2, 1>,                   1,              4,              4,      1,  S< 4, 16,  1>,   S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              4,      1,           1,           1,   S<1, 16, 1, 4>,                  1, BF16, BF16, LoopScheduler::Default, 16>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,                  ConvSpec, GemmMNKPadding,  1,  64,    64,    16,     32,   8, 4,  16,   16,    4,    1,  S< 4, 16,  1>, S<0, 2, 1>,     S<0, 2, 1>,                   1,              4,              4,      1,  S< 4, 16,  1>,   S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              4,      1,           1,           1,   S<1, 16, 1, 4>,                  1, BF16, BF16, LoopScheduler::Default, 32>
+#else // defined(CK_USE_AMD_MFMA_GFX950)
         // Instances with NumGroupsPerBatch > 1
         DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,                  ConvSpec, GemmMNKPadding,  1,  64,    64,    16,     16,   4, 4,  16,   16,    4,    1,  S< 4, 16,  1>, S<0, 2, 1>,     S<0, 2, 1>,                   1,              4,              4,      1,  S< 4, 16,  1>,   S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              4,      1,           1,           1,   S<1, 16, 1, 4>,                  1, BF16, BF16, LoopScheduler::Default, 8>,
         DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,                  ConvSpec, GemmMNKPadding,  1,  64,    64,    16,     16,   4, 4,  16,   16,    4,    1,  S< 4, 16,  1>, S<0, 2, 1>,     S<0, 2, 1>,                   1,              4,              4,      1,  S< 4, 16,  1>,   S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              4,      1,           1,           1,   S<1, 16, 1, 4>,                  1, BF16, BF16, LoopScheduler::Default, 16>,
         DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,                  ConvSpec, GemmMNKPadding,  1,  64,    64,    16,     16,   4, 4,  16,   16,    4,    1,  S< 4, 16,  1>, S<0, 2, 1>,     S<0, 2, 1>,                   1,              4,              4,      1,  S< 4, 16,  1>,   S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              4,      1,           1,           1,   S<1, 16, 1, 4>,                  1, BF16, BF16, LoopScheduler::Default, 32>
-#endif // defined(__gfx950__)
+#endif // defined(CK_USE_AMD_MFMA_GFX950)
     // clang-format on
     >;
 
@@ -67,13 +70,17 @@ using device_grouped_conv_fwd_xdl_merged_groups_f16_instances = std::tuple<
         //########################################|    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|
         //########################################|           |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-#if defined(__gfx950__)
-#else
+#if defined(CK_USE_AMD_MFMA_GFX950)
+        // Instances with NumGroupsPerBatch > 1
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,                  ConvSpec, GemmMNKPadding,  1,  64,    64,    16,     32,   8, 4,  16,   16,    4,    1,  S< 4, 16,  1>, S<0, 2, 1>,     S<0, 2, 1>,                   1,              4,              4,      1,  S< 4, 16,  1>,   S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              4,      1,           1,           1,   S<1, 16, 1, 4>,                  1, F16, F16, LoopScheduler::Default, 8>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,                  ConvSpec, GemmMNKPadding,  1,  64,    64,    16,     32,   8, 4,  16,   16,    4,    1,  S< 4, 16,  1>, S<0, 2, 1>,     S<0, 2, 1>,                   1,              4,              4,      1,  S< 4, 16,  1>,   S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              4,      1,           1,           1,   S<1, 16, 1, 4>,                  1, F16, F16, LoopScheduler::Default, 16>,
+        DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,                  ConvSpec, GemmMNKPadding,  1,  64,    64,    16,     32,   8, 4,  16,   16,    4,    1,  S< 4, 16,  1>, S<0, 2, 1>,     S<0, 2, 1>,                   1,              4,              4,      1,  S< 4, 16,  1>,   S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              4,      1,           1,           1,   S<1, 16, 1, 4>,                  1, F16, F16, LoopScheduler::Default, 32>
+#else // defined(CK_USE_AMD_MFMA_GFX950)
         // Instances with NumGroupsPerBatch > 1
         DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,                  ConvSpec, GemmMNKPadding,  1,  64,    64,    16,     16,   4, 4,  16,   16,    4,    1,  S< 4, 16,  1>, S<0, 2, 1>,     S<0, 2, 1>,                   1,              4,              4,      1,  S< 4, 16,  1>,   S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              4,      1,           1,           1,   S<1, 16, 1, 4>,                  1, F16, F16, LoopScheduler::Default, 8>,
         DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,                  ConvSpec, GemmMNKPadding,  1,  64,    64,    16,     16,   4, 4,  16,   16,    4,    1,  S< 4, 16,  1>, S<0, 2, 1>,     S<0, 2, 1>,                   1,              4,              4,      1,  S< 4, 16,  1>,   S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              4,      1,           1,           1,   S<1, 16, 1, 4>,                  1, F16, F16, LoopScheduler::Default, 16>,
         DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,                  ConvSpec, GemmMNKPadding,  1,  64,    64,    16,     16,   4, 4,  16,   16,    4,    1,  S< 4, 16,  1>, S<0, 2, 1>,     S<0, 2, 1>,                   1,              4,              4,      1,  S< 4, 16,  1>,   S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              4,      1,           1,           1,   S<1, 16, 1, 4>,                  1, F16, F16, LoopScheduler::Default, 32>
-#endif // defined(__gfx950__)
+#endif // defined(CK_USE_AMD_MFMA_GFX950)
     // clang-format on
     >;
 

library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_weight.hpp

@@ -311,6 +311,11 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                 {
                     add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_instances(
                         op_ptrs);
+
+                    add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_default_pipev1_instances(
+                        op_ptrs);
+                    add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_pad0_pipev1_instances(
+                        op_ptrs);
                 }
 #endif
 #ifdef CK_ENABLE_FP16
@@ -320,6 +325,11 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                 {
                     add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_instances(
                         op_ptrs);
+
+                    add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_default_pipev1_instances(
+                        op_ptrs);
+                    add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_pad0_pipev1_instances(
+                        op_ptrs);
                 }
 #endif
 #ifdef CK_ENABLE_BF16
@@ -343,6 +353,15 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                 {
                     add_device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f32_instances(
                         op_ptrs);
+
+                    add_device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f32_default_pipev2_instances(
+                        op_ptrs);
+                    add_device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f32_default_pipev5_instances(
+                        op_ptrs);
+                    add_device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f32_pad0_pipev2_instances(
+                        op_ptrs);
+                    add_device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f32_pad0_pipev5_instances(
+                        op_ptrs);
                 }
 #endif
 #ifdef CK_ENABLE_FP16
@@ -352,6 +371,16 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                 {
                     add_device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f16_instances(
                         op_ptrs);
+
+                    add_device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f16_default_pipev2_instances(
+                        op_ptrs);
+                    add_device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f16_default_pipev5_instances(
+                        op_ptrs);
+                    add_device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f16_pad0_pipev2_instances(
+                        op_ptrs);
+                    add_device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f16_pad0_pipev5_instances(
+                        op_ptrs);
+
                     add_device_grouped_conv2d_bwd_weight_two_stage_xdl_nhwgc_gkyxc_nhwgk_f16_pipev1_instances(
                         op_ptrs);
                     add_device_grouped_conv2d_bwd_weight_two_stage_xdl_nhwgc_gkyxc_nhwgk_f16_pipev2_instances(
@@ -381,6 +410,16 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                 {
                     add_device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_bf16_instances(
                         op_ptrs);
+
+                    add_device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_bf16_default_pipev2_instances(
+                        op_ptrs);
+                    add_device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_bf16_default_pipev5_instances(
+                        op_ptrs);
+                    add_device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_bf16_pad0_pipev2_instances(
+                        op_ptrs);
+                    add_device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_bf16_pad0_pipev5_instances(
+                        op_ptrs);
+
                     add_device_grouped_conv2d_bwd_weight_two_stage_xdl_nhwgc_gkyxc_nhwgk_bf16_pipev1_instances(
                         op_ptrs);
                     add_device_grouped_conv2d_bwd_weight_two_stage_xdl_nhwgc_gkyxc_nhwgk_bf16_pipev2_instances(
@@ -471,6 +510,15 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                 {
                     add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instances(
                         op_ptrs);
+
+                    add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f32_default_pipev2_instances(
+                        op_ptrs);
+                    add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f32_default_pipev5_instances(
+                        op_ptrs);
+                    add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f32_pad0_pipev2_instances(
+                        op_ptrs);
+                    add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f32_pad0_pipev2_instances(
+                        op_ptrs);
                 }
 #endif
 #ifdef CK_ENABLE_FP16
@@ -480,6 +528,16 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                 {
                     add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instances(
                         op_ptrs);
+
+                    add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f16_default_pipev2_instances(
+                        op_ptrs);
+                    add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f16_default_pipev5_instances(
+                        op_ptrs);
+                    add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f16_pad0_pipev2_instances(
+                        op_ptrs);
+                    add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f16_pad0_pipev5_instances(
+                        op_ptrs);
+
                     add_device_grouped_conv3d_bwd_weight_two_stage_xdl_ndhwgc_gkzyxc_ndhwgk_f16_pipev1_instances(
                         op_ptrs);
                     add_device_grouped_conv3d_bwd_weight_two_stage_xdl_ndhwgc_gkzyxc_ndhwgk_f16_pipev2_instances(
@@ -509,6 +567,16 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                 {
                     add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instances(
                         op_ptrs);
+
+                    add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_default_pipev2_instances(
+                        op_ptrs);
+                    add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_default_pipev5_instances(
+                        op_ptrs);
+                    add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_pad0_pipev2_instances(
+                        op_ptrs);
+                    add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_pad0_pipev5_instances(
+                        op_ptrs);
+
                     add_device_grouped_conv3d_bwd_weight_two_stage_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_pipev1_instances(
                         op_ptrs);
                     add_device_grouped_conv3d_bwd_weight_two_stage_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_pipev2_instances(

library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_weight_xdl.inc

@@ -74,6 +74,30 @@ void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_instances(
                                                            PassThrough,
                                                            PassThrough,
                                                            PassThrough>>>& instances);
+
+void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_default_pipev1_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_xdl_gnhwc_gkyxc_gnhwk_f16_pad0_pipev1_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
+                                                           GNHWC,
+                                                           GKYXC,
+                                                           GNHWK,
+                                                           F16,
+                                                           F16,
+                                                           F16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
 #endif
 #ifdef CK_ENABLE_FP32
 void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_instances(
@@ -87,6 +111,30 @@ void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_instances(
                                                            PassThrough,
                                                            PassThrough,
                                                            PassThrough>>>& instances);
+
+void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_default_pipev1_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_xdl_gnhwc_gkyxc_gnhwk_f32_pad0_pipev1_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
+                                                           GNHWC,
+                                                           GKYXC,
+                                                           GNHWK,
+                                                           F32,
+                                                           F32,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
 #endif
 #ifdef CK_ENABLE_BF16
 void add_device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_bf16_instances(
@@ -112,6 +160,53 @@ void add_device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_bf16_f32_bf16_in
                                                            PassThrough,
                                                            PassThrough,
                                                            PassThrough>>>& instances);
+void add_device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_bf16_default_pipev2_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
+                                                           NHWGC,
+                                                           GKYXC,
+                                                           NHWGK,
+                                                           BF16,
+                                                           BF16,
+                                                           BF16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
+
+void add_device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_bf16_default_pipev5_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
+                                                           NHWGC,
+                                                           GKYXC,
+                                                           NHWGK,
+                                                           BF16,
+                                                           BF16,
+                                                           BF16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
+
+void add_device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_bf16_pad0_pipev2_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
+                                                           NHWGC,
+                                                           GKYXC,
+                                                           NHWGK,
+                                                           BF16,
+                                                           BF16,
+                                                           BF16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
+
+void add_device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_bf16_pad0_pipev5_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
+                                                           NHWGC,
+                                                           GKYXC,
+                                                           NHWGK,
+                                                           BF16,
+                                                           BF16,
+                                                           BF16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
 
 void add_device_grouped_conv2d_bwd_weight_two_stage_xdl_nhwgc_gkyxc_nhwgk_bf16_pipev1_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
@@ -222,6 +317,54 @@ void add_device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f16_instances(
                                                            PassThrough,
                                                            PassThrough>>>& instances);
 
+void add_device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f16_default_pipev2_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_f16_default_pipev5_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_f16_pad0_pipev2_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_f16_pad0_pipev5_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_two_stage_xdl_nhwgc_gkyxc_nhwgk_f16_pipev1_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
                                                            NHWGC,
@@ -330,6 +473,54 @@ void add_device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f32_instances(
                                                            PassThrough,
                                                            PassThrough,
                                                            PassThrough>>>& instances);
+
+void add_device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f32_default_pipev2_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
+                                                           NHWGC,
+                                                           GKYXC,
+                                                           NHWGK,
+                                                           F32,
+                                                           F32,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
+
+void add_device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f32_default_pipev5_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
+                                                           NHWGC,
+                                                           GKYXC,
+                                                           NHWGK,
+                                                           F32,
+                                                           F32,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
+
+void add_device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f32_pad0_pipev2_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
+                                                           NHWGC,
+                                                           GKYXC,
+                                                           NHWGK,
+                                                           F32,
+                                                           F32,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
+
+void add_device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f32_pad0_pipev5_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
+                                                           NHWGC,
+                                                           GKYXC,
+                                                           NHWGK,
+                                                           F32,
+                                                           F32,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
 #endif
 // conv3d backward weight
 #ifdef CK_ENABLE_BF16
@@ -384,6 +575,54 @@ void add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance
                                                            PassThrough,
                                                            PassThrough>>>& instances);
 
+void add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_default_pipev2_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
+                                                           NDHWGC,
+                                                           GKZYXC,
+                                                           NDHWGK,
+                                                           BF16,
+                                                           BF16,
+                                                           BF16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
+
+void add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_default_pipev5_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
+                                                           NDHWGC,
+                                                           GKZYXC,
+                                                           NDHWGK,
+                                                           BF16,
+                                                           BF16,
+                                                           BF16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
+
+void add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_pad0_pipev2_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
+                                                           NDHWGC,
+                                                           GKZYXC,
+                                                           NDHWGK,
+                                                           BF16,
+                                                           BF16,
+                                                           BF16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
+
+void add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_pad0_pipev5_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
+                                                           NDHWGC,
+                                                           GKZYXC,
+                                                           NDHWGK,
+                                                           BF16,
+                                                           BF16,
+                                                           BF16,
+                                                           PassThrough,
+                                                           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,
@@ -505,6 +744,54 @@ void add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instances
                                                            PassThrough,
                                                            PassThrough>>>& instances);
 
+void add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f16_default_pipev2_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_f16_default_pipev5_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_f16_pad0_pipev2_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_f16_pad0_pipev5_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_two_stage_xdl_ndhwgc_gkzyxc_ndhwgk_f16_pipev1_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
                                                            NDHWGC,
@@ -613,6 +900,54 @@ void add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instances
                                                            PassThrough,
                                                            PassThrough,
                                                            PassThrough>>>& instances);
+
+void add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f32_default_pipev2_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
+                                                           NDHWGC,
+                                                           GKZYXC,
+                                                           NDHWGK,
+                                                           F32,
+                                                           F32,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
+
+void add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f32_default_pipev5_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
+                                                           NDHWGC,
+                                                           GKZYXC,
+                                                           NDHWGK,
+                                                           F32,
+                                                           F32,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
+
+void add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f32_pad0_pipev2_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
+                                                           NDHWGC,
+                                                           GKZYXC,
+                                                           NDHWGK,
+                                                           F32,
+                                                           F32,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
+
+void add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f32_pad0_pipev2_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
+                                                           NDHWGC,
+                                                           GKZYXC,
+                                                           NDHWGK,
+                                                           F32,
+                                                           F32,
+                                                           F32,
+                                                           PassThrough,
+                                                           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(

library/src/tensor_operation_instance/gpu/gemm_multiply_multiply/device_gemm_multiply_multiply_xdl_f8_f8_bf16/device_gemm_multiply_multiply_xdl_f8_f8_bf16_mk_nk_mn.hpp

@@ -53,6 +53,7 @@ using device_gemm_multiply_multiply_xdl_f8_f8_bf16_mk_nk_mn_comp_instances = std
         DeviceGemmMultiD_Xdl_CShuffle_V3<  Row,     Col,     Tuple<Row, Col>,  Row,    F8,    F8,    Tuple<F32, F32>, BF16,  F32,     F32,     PassThrough, PassThrough, MultiplyMultiply,    GemmSpec,   256,   256,   160,   128,  16,  16,  32,   32,    2,    5,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,    S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,          2,           1,                   S<1, 64, 1, 4>,     S<8, 8, 1>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3, F8>,
         DeviceGemmMultiD_Xdl_CShuffle_V3<  Row,     Col,     Tuple<Row, Col>,  Row,    F8,    F8,    Tuple<F32, F32>, BF16,  F32,     F32,     PassThrough, PassThrough, MultiplyMultiply,    GemmSpec,   256,   256,   128,   128,  16,  16,  32,   32,    4,    2,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,    S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,          1,           1,                   S<1, 32, 1, 8>,     S<8, 8, 1>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3, F8>,
         DeviceGemmMultiD_Xdl_CShuffle_V3<  Row,     Col,     Tuple<Row, Col>,  Row,    F8,    F8,    Tuple<F32, F32>, BF16,  F32,     F32,     PassThrough, PassThrough, MultiplyMultiply,    GemmSpec,   256,   256,    96,   128,  16,  16,  32,   32,    2,    3,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,    S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,          2,           1,                   S<1, 64, 1, 4>,     S<8, 8, 1>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3, F8>,
+        DeviceGemmMultiD_Xdl_CShuffle_V3<  Row,     Col,     Tuple<Row, Col>,  Row,    F8,    F8,    Tuple<F32, F32>, BF16,  F32,     F32,     PassThrough, PassThrough, MultiplyMultiply,    GemmSpec,   256,   256,    96,   128,  16,  16,  16,   16,    8,    3,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,    S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,          2,           1,                   S<1, 64, 1, 4>,     S<8, 8, 1>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3, F8>,
         DeviceGemmMultiD_Xdl_CShuffle_V3<  Row,     Col,     Tuple<Row, Col>,  Row,    F8,    F8,    Tuple<F32, F32>, BF16,  F32,     F32,     PassThrough, PassThrough, MultiplyMultiply,    GemmSpec,   256,   256,    64,   128,  16,  16,  32,   32,    4,    1,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,    S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,          1,           1,                   S<1, 32, 1, 8>,     S<8, 8, 1>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3, F8>,
         DeviceGemmMultiD_Xdl_CShuffle_V3<  Row,     Col,     Tuple<Row, Col>,  Row,    F8,    F8,    Tuple<F32, F32>, BF16,  F32,     F32,     PassThrough, PassThrough, MultiplyMultiply,    GemmSpec,   256,   128,   256,   128,  16,  16,  32,   32,    2,    4,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,    S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,          1,           1,                   S<1, 32, 1, 8>,     S<8, 8, 1>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3, F8>,
         DeviceGemmMultiD_Xdl_CShuffle_V3<  Row,     Col,     Tuple<Row, Col>,  Row,    F8,    F8,    Tuple<F32, F32>, BF16,  F32,     F32,     PassThrough, PassThrough, MultiplyMultiply,    GemmSpec,   256,   128,   224,   128,  16,  16,  32,   32,    1,    7,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,    S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,          1,           1,                   S<1, 64, 1, 4>,     S<8, 8, 1>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3, F8>,
@@ -62,6 +63,13 @@ using device_gemm_multiply_multiply_xdl_f8_f8_bf16_mk_nk_mn_comp_instances = std
         DeviceGemmMultiD_Xdl_CShuffle_V3<  Row,     Col,     Tuple<Row, Col>,  Row,    F8,    F8,    Tuple<F32, F32>, BF16,  F32,     F32,     PassThrough, PassThrough, MultiplyMultiply,    GemmSpec,   256,   128,   128,   128,  16,  16,  32,   32,    2,    2,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,    S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,          1,           1,                   S<1, 32, 1, 8>,     S<8, 8, 1>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3, F8>,
         DeviceGemmMultiD_Xdl_CShuffle_V3<  Row,     Col,     Tuple<Row, Col>,  Row,    F8,    F8,    Tuple<F32, F32>, BF16,  F32,     F32,     PassThrough, PassThrough, MultiplyMultiply,    GemmSpec,   256,   128,    96,   256,  16,  16,  32,   32,    1,    3,     S<16, 16, 1>,    S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,    S<16, 16, 1>,    S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,          1,           1,                   S<1, 64, 1, 4>,     S<8, 8, 1>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3, F8>,
         // Compute friendly
+        DeviceGemmMultiD_Xdl_CShuffle_V3<  Row,     Col,     Tuple<Row, Col>,  Row,    F8,    F8,    Tuple<F32, F32>, BF16,  F32,     F32,     PassThrough, PassThrough, MultiplyMultiply,    GemmSpec,   256,   192,    128,  128,  16,  16,  32,   32,    3,    2,     S<4, 64, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,    S<4, 64, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,          1,           1,                   S<1, 32, 1, 8>,     S<8, 8, 1>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3, F8>,
+        DeviceGemmMultiD_Xdl_CShuffle_V3<  Row,     Col,     Tuple<Row, Col>,  Row,    F8,    F8,    Tuple<F32, F32>, BF16,  F32,     F32,     PassThrough, PassThrough, MultiplyMultiply,    GemmSpec,   256,   160,   128,   128,  16,  16,  16,   16,    5,    4,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,    S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,          1,           2,                   S<1, 32, 1, 8>,     S<8, 8, 1>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3, F8>,
+        DeviceGemmMultiD_Xdl_CShuffle_V3<  Row,     Col,     Tuple<Row, Col>,  Row,    F8,    F8,    Tuple<F32, F32>, BF16,  F32,     F32,     PassThrough, PassThrough, MultiplyMultiply,    GemmSpec,   256,   192,    96,   128,  16,  16,  16,   16,    6,    3,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,    S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,          2,           1,                   S<1, 64, 1, 4>,     S<8, 8, 1>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3, F8>,
+
+        DeviceGemmMultiD_Xdl_CShuffle_V3<  Row,     Col,     Tuple<Row, Col>,  Row,    F8,    F8,    Tuple<F32, F32>, BF16,  F32,     F32,     PassThrough, PassThrough, MultiplyMultiply,    GemmSpec,   256,    96,   128,   128,  16,  16,  16,   16,    3,    4,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,    S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,          1,           2,                   S<1, 32, 1, 8>,     S<8, 8, 1>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3, F8>,
+        DeviceGemmMultiD_Xdl_CShuffle_V3<  Row,     Col,     Tuple<Row, Col>,  Row,    F8,    F8,    Tuple<F32, F32>, BF16,  F32,     F32,     PassThrough, PassThrough, MultiplyMultiply,    GemmSpec,   256,    96,   128,   256,  16,  16,  16,   16,    3,    4,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,    S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,          1,           2,                   S<1, 32, 1, 8>,     S<8, 8, 1>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3, F8>,
+        DeviceGemmMultiD_Xdl_CShuffle_V3<  Row,     Col,     Tuple<Row, Col>,  Row,    F8,    F8,    Tuple<F32, F32>, BF16,  F32,     F32,     PassThrough, PassThrough, MultiplyMultiply,    GemmSpec,   256,    80,   128,   256,  16,  16,  16,   16,    5,    2,     S<16, 16, 1>,    S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,    S<16, 16, 1>,    S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,          1,           2,                   S<1, 16, 1, 16>,    S<8, 8, 1>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3, F8>,
         DeviceGemmMultiD_Xdl_CShuffle_V3<  Row,     Col,     Tuple<Row, Col>,  Row,    F8,    F8,    Tuple<F32, F32>, BF16,  F32,     F32,     PassThrough, PassThrough, MultiplyMultiply,    GemmSpec,   256,    64,   256,   128,  16,  16,  32,   32,    1,    4,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,    S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,          1,           1,                   S<1, 32, 1, 8>,     S<8, 8, 1>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3, F8>,
         DeviceGemmMultiD_Xdl_CShuffle_V3<  Row,     Col,     Tuple<Row, Col>,  Row,    F8,    F8,    Tuple<F32, F32>, BF16,  F32,     F32,     PassThrough, PassThrough, MultiplyMultiply,    GemmSpec,   256,    64,   224,   128,  16,  16,  16,   16,    2,    7,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,    S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,          2,           1,                   S<1, 64, 1, 4>,     S<8, 8, 1>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3, F8>,
         DeviceGemmMultiD_Xdl_CShuffle_V3<  Row,     Col,     Tuple<Row, Col>,  Row,    F8,    F8,    Tuple<F32, F32>, BF16,  F32,     F32,     PassThrough, PassThrough, MultiplyMultiply,    GemmSpec,   256,    64,   192,   256,  16,  16,  32,   32,    1,    3,     S<16, 16, 1>,    S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,    S<16, 16, 1>,    S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,          1,           1,                   S<1, 32, 1, 8>,     S<8, 8, 1>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3, F8>,

library/src/tensor_operation_instance/gpu/gemm_universal/device_gemm_xdl_universal_bf16_bf16_bf16/device_gemm_xdl_universal_bf16_bf16_bf16_km_nk_mn.hpp

@@ -44,7 +44,9 @@ using device_gemm_xdl_universal_bf16_bf16_bf16_km_nk_mn_comp_instances = std::tu
 #if defined(CK_USE_AMD_MFMA_GFX950)
 #endif // defined(CK_USE_AMD_MFMA_GFX950)
         // Compute friendly
+#if !defined(CK_USE_AMD_MFMA_GFX950)
         DeviceGemm_Xdl_CShuffleV3<  Col,     Col,     Row,     BF16,   BF16,  BF16,   F32,     BF16,  PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   256,   256,    32,   4,   8,  32,   32,    4,    4,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              4,          0,    S<4, 64, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 16, 1, 16>,               4,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v4>,
+#endif // !defined(CK_USE_AMD_MFMA_GFX950)
         DeviceGemm_Xdl_CShuffleV3<  Col,     Col,     Row,     BF16,   BF16,  BF16,   F32,     BF16,  PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   256,   256,    32,   4,   4,  32,   32,    4,    4,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,          0,    S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              4,              4,          0,          1,           1,                   S<1, 16, 1, 16>,               4,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v4>,
         DeviceGemm_Xdl_CShuffleV3<  Col,     Col,     Row,     BF16,   BF16,  BF16,   F32,     BF16,  PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   256,   256,    32,   2,   2,  32,   32,    4,    4,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              2,          0,    S<16,16, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              2,              2,          0,          1,           1,                   S<1, 16, 1, 16>,               4,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v4>,
         DeviceGemm_Xdl_CShuffleV3<  Col,     Col,     Row,     BF16,   BF16,  BF16,   F32,     BF16,  PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   128,   128,    64,   4,   8,  32,   32,    2,    2,     S<16, 16, 1>,    S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              4,          0,    S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 16, 1, 16>,               4,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v4>,
@@ -54,7 +56,9 @@ using device_gemm_xdl_universal_bf16_bf16_bf16_km_nk_mn_comp_instances = std::tu
         DeviceGemm_Xdl_CShuffleV3<  Col,     Col,     Row,     BF16,   BF16,  BF16,   F32,     BF16,  PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   256,   256,    32,   4,   8,  32,   32,    4,    4,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              4,          0,    S<4, 64, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 16, 1, 16>,               4,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v5>,
         DeviceGemm_Xdl_CShuffleV3<  Col,     Col,     Row,     BF16,   BF16,  BF16,   F32,     BF16,  PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   256,   256,    32,   4,   4,  32,   32,    4,    4,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,          0,    S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              4,              4,          0,          1,           1,                   S<1, 16, 1, 16>,               4,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v5>,
         DeviceGemm_Xdl_CShuffleV3<  Col,     Col,     Row,     BF16,   BF16,  BF16,   F32,     BF16,  PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   256,   256,    32,   2,   2,  32,   32,    4,    4,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              2,          0,    S<16,16, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              2,              2,          0,          1,           1,                   S<1, 16, 1, 16>,               4,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v5>,
+#if !defined(CK_USE_AMD_MFMA_GFX950)
         DeviceGemm_Xdl_CShuffleV3<  Col,     Col,     Row,     BF16,   BF16,  BF16,   F32,     BF16,  PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   256,   224,    64,   8,   8,  16,   16,    8,    7,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              8,          0,    S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          2,           1,                   S<1, 32, 1, 8>,                4,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3>,
+#endif // !defined(CK_USE_AMD_MFMA_GFX950)
         DeviceGemm_Xdl_CShuffleV3<  Col,     Col,     Row,     BF16,   BF16,  BF16,   F32,     BF16,  PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   128,   128,    64,   4,   8,  32,   32,    2,    2,     S<16, 16, 1>,    S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              4,          0,    S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 16, 1, 16>,               4,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3>,
         DeviceGemm_Xdl_CShuffleV3<  Col,     Col,     Row,     BF16,   BF16,  BF16,   F32,     BF16,  PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   128,   128,    64,   4,   8,  32,   32,    2,    2,     S<16, 16, 1>,    S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              4,          0,    S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 16, 1, 16>,               4,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v5>,
         DeviceGemm_Xdl_CShuffleV3<  Col,     Col,     Row,     BF16,   BF16,  BF16,   F32,     BF16,  PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   128,   128,    64,   4,   8,  32,   32,    2,    2,     S<16, 16, 1>,    S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              4,          0,    S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 16, 1, 16>,               4,  BlockGemmPipelineScheduler::Interwave, BlockGemmPipelineVersion::v1>,
@@ -86,7 +90,9 @@ using device_gemm_xdl_universal_bf16_bf16_bf16_km_nk_mn_mem_instances = std::tup
         DeviceGemm_Xdl_CShuffleV3<  Col,     Col,     Row,     BF16,   BF16,  BF16,   F32,     BF16,  PassThrough, PassThrough, PassThrough,       GemmSpec,   128,    32,   16,    64,   4,   8,  16,   16,    1,    1,     S<16, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,          0,    S<8, 16, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 16, 1, 8>,               2,  BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
         DeviceGemm_Xdl_CShuffleV3<  Col,     Col,     Row,     BF16,   BF16,  BF16,   F32,     BF16,  PassThrough, PassThrough, PassThrough,       GemmSpec,    64,    16,   16,    64,   4,   8,  16,   16,    1,    1,     S<16, 4, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,          0,    S<8,  8, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 16, 1, 4>,               4,  BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
         DeviceGemm_Xdl_CShuffleV3<  Col,     Col,     Row,     BF16,   BF16,  BF16,   F32,     BF16,  PassThrough, PassThrough, PassThrough,       GemmSpec,   128,    16,   32,    64,   4,   8,  16,   16,    1,    1,     S<16, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,          0,    S<8, 16, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 16, 1, 8>,               4,  BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
+#if !defined(CK_USE_AMD_MFMA_GFX950)
         DeviceGemm_Xdl_CShuffleV3<  Col,     Col,     Row,     BF16,   BF16,  BF16,   F32,     BF16,  PassThrough, PassThrough, PassThrough,       GemmSpec,   128,    16,   64,    64,   4,   8,  16,   16,    1,    2,     S<16, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,          0,    S<8, 16, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 16, 1, 8>,               4,  BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
+#endif // !defined(CK_USE_AMD_MFMA_GFX950)
         DeviceGemm_Xdl_CShuffleV3<  Col,     Col,     Row,     BF16,   BF16,  BF16,   F32,     BF16,  PassThrough, PassThrough, PassThrough,       GemmSpec,   128,    16,   64,    64,   4,   4,  16,   16,    1,    2,     S<16, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,          0,    S<16, 8, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              4,              4,          0,          1,           1,                   S<1, 16, 1, 8>,               4,  BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
         DeviceGemm_Xdl_CShuffleV3<  Col,     Col,     Row,     BF16,   BF16,  BF16,   F32,     BF16,  PassThrough, PassThrough, PassThrough,       GemmSpec,   128,    16,  128,    64,   4,   8,  16,   16,    1,    4,     S<16, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,          0,    S<8, 16, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 16, 1, 8>,               4,  BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
         DeviceGemm_Xdl_CShuffleV3<  Col,     Col,     Row,     BF16,   BF16,  BF16,   F32,     BF16,  PassThrough, PassThrough, PassThrough,       GemmSpec,   256,    16,  256,    64,   2,   8,  16,   16,    1,    4,     S<32, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              2,          0,    S<8, 16, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 16, 1, 16>,              4,  BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,

library/src/tensor_operation_instance/gpu/gemm_universal/device_gemm_xdl_universal_bf16_bf16_bf16/device_gemm_xdl_universal_bf16_bf16_bf16_mk_kn_mn.hpp

@@ -43,13 +43,17 @@ using device_gemm_xdl_universal_bf16_bf16_bf16_mk_kn_mn_comp_instances = std::tu
         //#########################|        |        |        |     |      |      |        |         |            |            |            |              |      |      |      |      |    |    |    |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                                 |                |                                       |                             |
 #if defined(CK_USE_AMD_MFMA_GFX950)
 #endif // defined(CK_USE_AMD_MFMA_GFX950)
+#if !defined(CK_USE_AMD_MFMA_GFX950)
         DeviceGemm_Xdl_CShuffleV3<  Row,     Row,     Row,     BF16,   BF16,  BF16,   F32,     BF16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   256,   256,    32,   8,   4,  32,   32,    4,    4,     S<4, 64, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              4,          0,          1,           1,                   S<1, 16, 1, 16>,               4,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v4>,
+#endif // !defined(CK_USE_AMD_MFMA_GFX950)
         DeviceGemm_Xdl_CShuffleV3<  Row,     Row,     Row,     BF16,   BF16,  BF16,   F32,     BF16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   256,   256,    32,   4,   4,  32,   32,    4,    4,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              4,              4,          0,    S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              4,          0,          1,           1,                   S<1, 16, 1, 16>,               4,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v4>,
         DeviceGemm_Xdl_CShuffleV3<  Row,     Row,     Row,     BF16,   BF16,  BF16,   F32,     BF16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   256,   256,    32,   2,   2,  32,   32,    4,    4,     S<16,16, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              2,              2,          0,    S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              2,          0,          1,           1,                   S<1, 16, 1, 16>,               4,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v4>,
         DeviceGemm_Xdl_CShuffleV3<  Row,     Row,     Row,     BF16,   BF16,  BF16,   F32,     BF16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   128,   128,    64,   8,   4,  32,   32,    2,    2,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<16, 16, 1>,    S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              4,          0,          1,           1,                   S<1, 16, 1, 16>,               4,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v4>,
         DeviceGemm_Xdl_CShuffleV3<  Row,     Row,     Row,     BF16,   BF16,  BF16,   F32,     BF16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   256,   256,    32,   8,   4,  32,   32,    4,    4,     S<4, 64, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              4,          0,          1,           1,                   S<1, 16, 1, 16>,               4,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v5>,
         DeviceGemm_Xdl_CShuffleV3<  Row,     Row,     Row,     BF16,   BF16,  BF16,   F32,     BF16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   256,   256,    32,   8,   4,  32,   32,    4,    4,     S<4, 64, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              4,          0,          1,           1,                   S<1, 16, 1, 16>,               4,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3>,
+#if !defined(CK_USE_AMD_MFMA_GFX950)
         DeviceGemm_Xdl_CShuffleV3<  Row,     Row,     Row,     BF16,   BF16,  BF16,   F32,     BF16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   224,   256,    64,   8,   8,  16,   16,    7,    8,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,          0,          1,           2,                   S<1, 16, 1, 16>,               4,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3>,
+#endif // !defined(CK_USE_AMD_MFMA_GFX950)
         DeviceGemm_Xdl_CShuffleV3<  Row,     Row,     Row,     BF16,   BF16,  BF16,   F32,     BF16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   128,   128,    64,   8,   4,  32,   32,    2,    2,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<16, 16, 1>,    S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              4,          0,          1,           1,                   S<1, 16, 1, 16>,               4,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3>,
         DeviceGemm_Xdl_CShuffleV3<  Row,     Row,     Row,     BF16,   BF16,  BF16,   F32,     BF16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   128,   128,    64,   8,   4,  32,   32,    2,    2,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<16, 16, 1>,    S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              4,          0,          1,           1,                   S<1, 16, 1, 16>,               4,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v5>,
         DeviceGemm_Xdl_CShuffleV3<  Row,     Row,     Row,     BF16,   BF16,  BF16,   F32,     BF16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   128,   128,    64,   8,   4,  32,   32,    2,    2,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<16, 16, 1>,    S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              4,          0,          1,           1,                   S<1, 16, 1, 16>,               4,  BlockGemmPipelineScheduler::Interwave, BlockGemmPipelineVersion::v1>
@@ -80,8 +84,10 @@ using device_gemm_xdl_universal_bf16_bf16_bf16_mk_kn_mn_mem_instances = std::tup
         DeviceGemm_Xdl_CShuffleV3<  Row,     Row,     Row,     BF16,   BF16,  BF16,   F32,     BF16,      PassThrough, PassThrough, PassThrough,       GemmSpec,    64,    16,   16,    64,   8,   4,  16,   16,    1,    1,     S<8,  8, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<16, 4, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,          0,          1,           1,                   S<1, 16, 1, 4>,               4,  BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
         DeviceGemm_Xdl_CShuffleV3<  Row,     Row,     Row,     BF16,   BF16,  BF16,   F32,     BF16,      PassThrough, PassThrough, PassThrough,       GemmSpec,    64,    16,   16,    64,   4,   4,  16,   16,    1,    1,     S<16, 4, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              4,              4,          0,    S<16, 4, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,          0,          1,           1,                   S<1, 16, 1, 4>,               4,  BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
         DeviceGemm_Xdl_CShuffleV3<  Row,     Row,     Row,     BF16,   BF16,  BF16,   F32,     BF16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   128,    16,   32,    64,   8,   4,  16,   16,    1,    1,     S<8, 16, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<16, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,          0,          1,           1,                   S<1, 16, 1, 8>,               4,  BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
+#if !defined(CK_USE_AMD_MFMA_GFX950)
         DeviceGemm_Xdl_CShuffleV3<  Row,     Row,     Row,     BF16,   BF16,  BF16,   F32,     BF16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   128,    16,   64,    64,   8,   4,  16,   16,    1,    2,     S<8, 16, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<16, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              4,          0,          1,           1,                   S<1, 16, 1, 8>,               4,  BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
         DeviceGemm_Xdl_CShuffleV3<  Row,     Row,     Row,     BF16,   BF16,  BF16,   F32,     BF16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   128,    16,  128,    64,   8,   4,  16,   16,    1,    4,     S<8, 16, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              4,          0,          1,           1,                   S<1, 16, 1, 8>,               4,  BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
+#endif // !defined(CK_USE_AMD_MFMA_GFX950)
         DeviceGemm_Xdl_CShuffleV3<  Row,     Row,     Row,     BF16,   BF16,  BF16,   F32,     BF16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,    16,  256,    64,   8,   4,  16,   16,    1,    4,     S<8, 16, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              4,          0,          1,           1,                   S<1, 16, 1, 16>,              4,  BlkGemmPipeSched, BlockGemmPipelineVersion::v2>
     // clang-format on
     >;

library/src/tensor_operation_instance/gpu/gemm_universal/device_gemm_xdl_universal_bf16_bf16_bf16/device_gemm_xdl_universal_bf16_bf16_bf16_mk_nk_mn.hpp

@@ -44,16 +44,20 @@ using device_gemm_xdl_universal_bf16_bf16_bf16_mk_nk_mn_comp_instances = std::tu
 #if defined(CK_USE_AMD_MFMA_GFX950)
 #endif // defined(CK_USE_AMD_MFMA_GFX950)
         // Compute friendly
+#if !defined(CK_USE_AMD_MFMA_GFX950)
         DeviceGemm_Xdl_CShuffleV3<  Row,     Col,     Row,     BF16,   BF16,  BF16,   F32,     BF16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   256,   256,    32,   8,   8,  32,   32,    4,    4,     S<4, 64, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<4, 64, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 16, 1, 16>,               4,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v4>,
+#endif // !defined(CK_USE_AMD_MFMA_GFX950)
         DeviceGemm_Xdl_CShuffleV3<  Row,     Col,     Row,     BF16,   BF16,  BF16,   F32,     BF16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   256,   256,    32,   4,   4,  32,   32,    4,    4,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              4,              4,          0,    S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              4,              4,          0,          1,           1,                   S<1, 16, 1, 16>,               4,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v4>,
         DeviceGemm_Xdl_CShuffleV3<  Row,     Col,     Row,     BF16,   BF16,  BF16,   F32,     BF16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   256,   256,    32,   2,   2,  32,   32,    4,    4,     S<16,16, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              2,              2,          0,    S<16,16, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              2,              2,          0,          1,           1,                   S<1, 16, 1, 16>,               4,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v4>,
         DeviceGemm_Xdl_CShuffleV3<  Row,     Col,     Row,     BF16,   BF16,  BF16,   F32,     BF16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   256,   256,    32,   8,   8,  32,   32,    4,    4,     S<4, 64, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<4, 64, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 16, 1, 16>,               4,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3>,
+#if !defined(CK_USE_AMD_MFMA_GFX950)
         DeviceGemm_Xdl_CShuffleV3<  Row,     Col,     Row,     BF16,   BF16,  BF16,   F32,     BF16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   256,   256,    32,   8,   8,  32,   32,    4,    4,     S<4, 64, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<4, 64, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 16, 1, 16>,               4,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v5>,
         // AGPR Spill
         // DeviceGemm_Xdl_CShuffleV3<  Row,     Col,     Row,     BF16,   BF16,  BF16,   F32,     BF16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   256,   256,    32,   8,   8,  16,   16,    8,    8,     S<4, 64, 1>,     S<1, 0, 2>,    S<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,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v5>,
         // AGPR Spill when use permuted lds layout. so, use padding for these two.
         DeviceGemm_Xdl_CShuffleV3<  Row,     Col,     Row,     BF16,   BF16,  BF16,   F32,     BF16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   224,   256,    64,   8,   8,  16,   16,    7,    8,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           2,                   S<1, 16, 1, 16>,               4,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3>,
         DeviceGemm_Xdl_CShuffleV3<  Row,     Col,     Row,     BF16,   BF16,  BF16,   F32,     BF16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   256,   224,    64,   8,   8,  16,   16,    8,    7,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          2,           1,                   S<1, 32, 1, 8>,               4,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3>,
+#endif // !defined(CK_USE_AMD_MFMA_GFX950)
         DeviceGemm_Xdl_CShuffleV3<  Row,     Col,     Row,     BF16,   BF16,  BF16,   F32,     BF16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   128,   128,    64,   8,   8,  32,   32,    2,    2,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 16, 1, 16>,               4,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3>,
         DeviceGemm_Xdl_CShuffleV3<  Row,     Col,     Row,     BF16,   BF16,  BF16,   F32,     BF16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   128,   128,    64,   8,   8,  32,   32,    2,    2,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 16, 1, 16>,               4,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v5>,
         DeviceGemm_Xdl_CShuffleV3<  Row,     Col,     Row,     BF16,   BF16,  BF16,   F32,     BF16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   128,   128,    64,   8,   8,  32,   32,    2,    2,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 16, 1, 16>,               4,  BlockGemmPipelineScheduler::Interwave, BlockGemmPipelineVersion::v1>
@@ -84,8 +88,10 @@ using device_gemm_xdl_universal_bf16_bf16_bf16_mk_nk_mn_mem_instances = std::tup
         DeviceGemm_Xdl_CShuffleV3<  Row,     Col,     Row,     BF16,   BF16,  BF16,   F32,     BF16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   128,    32,   16,    64,   8,   8,  16,   16,    1,    1,     S<8, 16, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<8, 16, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 16, 1, 8>,               2,  BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
         DeviceGemm_Xdl_CShuffleV3<  Row,     Col,     Row,     BF16,   BF16,  BF16,   F32,     BF16,      PassThrough, PassThrough, PassThrough,       GemmSpec,    64,    16,   16,    64,   8,   8,  16,   16,    1,    1,     S<8,  8, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<8,  8, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 16, 1, 4>,               4,  BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
         DeviceGemm_Xdl_CShuffleV3<  Row,     Col,     Row,     BF16,   BF16,  BF16,   F32,     BF16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   128,    16,   32,    64,   8,   8,  16,   16,    1,    1,     S<8, 16, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<8, 16, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 16, 1, 8>,               4,  BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
+#if !defined(CK_USE_AMD_MFMA_GFX950)
         DeviceGemm_Xdl_CShuffleV3<  Row,     Col,     Row,     BF16,   BF16,  BF16,   F32,     BF16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   128,    16,   64,    64,   8,   8,  16,   16,    1,    2,     S<8, 16, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<8, 16, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 16, 1, 8>,               4,  BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
         DeviceGemm_Xdl_CShuffleV3<  Row,     Col,     Row,     BF16,   BF16,  BF16,   F32,     BF16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   128,    16,  128,    64,   8,   8,  16,   16,    1,    4,     S<8, 16, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<8, 16, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 16, 1, 8>,               4,  BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
+#endif // !defined(CK_USE_AMD_MFMA_GFX950)
         DeviceGemm_Xdl_CShuffleV3<  Row,     Col,     Row,     BF16,   BF16,  BF16,   F32,     BF16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,    16,  256,    64,   8,   8,  16,   16,    1,    4,     S<8, 16, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<8, 16, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 16, 1, 16>,              4,  BlkGemmPipeSched, BlockGemmPipelineVersion::v2>
     // clang-format on
     >;

library/src/tensor_operation_instance/gpu/gemm_universal/device_gemm_xdl_universal_f16_f16_f16/device_gemm_xdl_universal_f16_f16_f16_mk_kn_mn.hpp

@@ -47,7 +47,9 @@ using device_gemm_xdl_universal_f16_f16_f16_mk_kn_mn_comp_instances = std::tuple
         DeviceGemm_Xdl_CShuffleV3<  Row,     Row,     Row,     F16,   F16,  F16,   F32,     F16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   128,   128,    64,   8,   4,  32,   32,    2,    2,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<16, 16, 1>,    S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              4,          0,          1,           1,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v4>,
         DeviceGemm_Xdl_CShuffleV3<  Row,     Row,     Row,     F16,   F16,  F16,   F32,     F16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   256,   256,    32,   8,   4,  32,   32,    4,    4,     S<4, 64, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              4,          0,          1,           1,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v5>,
         DeviceGemm_Xdl_CShuffleV3<  Row,     Row,     Row,     F16,   F16,  F16,   F32,     F16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   256,   256,    32,   8,   4,  32,   32,    4,    4,     S<4, 64, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              4,          0,          1,           1,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3>,
+#if !defined(CK_USE_AMD_MFMA_GFX950)
         DeviceGemm_Xdl_CShuffleV3<  Row,     Row,     Row,     F16,   F16,  F16,   F32,     F16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   224,   256,    64,   8,   8,  16,   16,    7,    8,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,          0,          1,           2,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3>,
+#endif // !defined(CK_USE_AMD_MFMA_GFX950)
         DeviceGemm_Xdl_CShuffleV3<  Row,     Row,     Row,     F16,   F16,  F16,   F32,     F16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   128,   128,    64,   8,   4,  32,   32,    2,    2,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<16, 16, 1>,    S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              4,          0,          1,           1,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3>,
         DeviceGemm_Xdl_CShuffleV3<  Row,     Row,     Row,     F16,   F16,  F16,   F32,     F16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   128,   128,    64,   8,   4,  32,   32,    2,    2,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<16, 16, 1>,    S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              4,          0,          1,           1,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v5>,
         DeviceGemm_Xdl_CShuffleV3<  Row,     Row,     Row,     F16,   F16,  F16,   F32,     F16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   128,   256,    32,   8,   4,  32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              4,          0,          1,           1,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Interwave, BlockGemmPipelineVersion::v1>,

library/src/tensor_operation_instance/gpu/gemm_universal/device_gemm_xdl_universal_f16_f16_f16/device_gemm_xdl_universal_f16_f16_f16_mk_nk_mn.hpp

@@ -57,11 +57,13 @@ using device_gemm_xdl_universal_f16_f16_f16_mk_nk_mn_comp_instances = std::tuple
         DeviceGemm_Xdl_CShuffleV3<  Row,     Col,     Row,     F16,   F16,  F16,   F32,     F16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   256,   256,    32,   8,   8,  16,   16,    8,    8,     S<4, 64, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<4, 64, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           2,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3>,
         DeviceGemm_Xdl_CShuffleV3<  Row,     Col,     Row,     F16,   F16,  F16,   F32,     F16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   256,   256,    32,   4,   4,  16,   16,    8,    8,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              4,              4,          0,    S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              4,              4,          0,          1,           2,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3>,
         DeviceGemm_Xdl_CShuffleV3<  Row,     Col,     Row,     F16,   F16,  F16,   F32,     F16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   256,   256,    32,   2,   2,  16,   16,    8,    8,     S<16,16, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              2,              2,          0,    S<16,16, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              2,              2,          0,          1,           2,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3>,
+#if !defined(CK_USE_AMD_MFMA_GFX950)
         // AGPR Spill
         // DeviceGemm_Xdl_CShuffleV3<  Row,     Col,     Row,     F16,   F16,  F16,   F32,     F16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   256,   256,    32,   8,   8,  16,   16,    8,    8,     S<4, 64, 1>,     S<1, 0, 2>,    S<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,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v5>,
         // AGPR Spill when use permuted lds layout. so, use padding for these two.
         DeviceGemm_Xdl_CShuffleV3<  Row,     Col,     Row,     F16,   F16,  F16,   F32,     F16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   224,   256,    64,   8,   8,  16,   16,    7,    8,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           2,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3>,
         DeviceGemm_Xdl_CShuffleV3<  Row,     Col,     Row,     F16,   F16,  F16,   F32,     F16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   256,   224,    64,   8,   8,  16,   16,    8,    7,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          2,           1,                   S<1, 64, 1, 4>,               8,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3>,
+#endif // !defined(CK_USE_AMD_MFMA_GFX950)
         DeviceGemm_Xdl_CShuffleV3<  Row,     Col,     Row,     F16,   F16,  F16,   F32,     F16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   128,   128,    64,   8,   8,  32,   32,    2,    2,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3>,
         DeviceGemm_Xdl_CShuffleV3<  Row,     Col,     Row,     F16,   F16,  F16,   F32,     F16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   128,   128,    64,   8,   8,  32,   32,    2,    2,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v5>,
         DeviceGemm_Xdl_CShuffleV3<  Row,     Col,     Row,     F16,   F16,  F16,   F32,     F16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   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,          0,    S<4, 64, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Interwave, BlockGemmPipelineVersion::v1>,
@@ -105,9 +107,11 @@ using device_gemm_xdl_universal_f16_f16_f16_mk_nk_mn_mem_instances = std::tuple<
         DeviceGemm_Xdl_CShuffleV3<  Row,     Col,     Row,     F16,   F16,  F16,   F32,     F16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   128,    16,   32,    64,   8,   8,  16,   16,    1,    1,     S<8, 16, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<8, 16, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 16, 1, 8>,               4,  BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
         DeviceGemm_Xdl_CShuffleV3<  Row,     Col,     Row,     F16,   F16,  F16,   F32,     F16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   128,    16,   32,    64,   4,   4,  16,   16,    1,    1,     S<16, 8, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              4,              4,          0,    S<16, 8, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              4,              4,          0,          1,           1,                   S<1, 16, 1, 8>,               4,  BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
         DeviceGemm_Xdl_CShuffleV3<  Row,     Col,     Row,     F16,   F16,  F16,   F32,     F16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   128,    16,   32,    64,   2,   2,  16,   16,    1,    1,     S<32, 4, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              2,              2,          0,    S<32, 4, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              2,              2,          0,          1,           1,                   S<1, 16, 1, 8>,               4,  BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
+#if !defined(CK_USE_AMD_MFMA_GFX950)
         DeviceGemm_Xdl_CShuffleV3<  Row,     Col,     Row,     F16,   F16,  F16,   F32,     F16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   128,    16,   64,    64,   8,   8,  16,   16,    1,    2,     S<8, 16, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<8, 16, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 16, 1, 8>,               4,  BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
-        DeviceGemm_Xdl_CShuffleV3<  Row,     Col,     Row,     F16,   F16,  F16,   F32,     F16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   128,    32,   64,    64,   8,   8,  32,   32,    1,    1,     S<8, 16, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<8, 16, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 16, 1, 8>,               8,  BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
         DeviceGemm_Xdl_CShuffleV3<  Row,     Col,     Row,     F16,   F16,  F16,   F32,     F16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   128,    16,  128,    64,   8,   8,  16,   16,    1,    4,     S<8, 16, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<8, 16, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 16, 1, 8>,               4,  BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
+#endif // !defined(CK_USE_AMD_MFMA_GFX950)
+        DeviceGemm_Xdl_CShuffleV3<  Row,     Col,     Row,     F16,   F16,  F16,   F32,     F16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   128,    32,   64,    64,   8,   8,  32,   32,    1,    1,     S<8, 16, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<8, 16, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 16, 1, 8>,               8,  BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
         DeviceGemm_Xdl_CShuffleV3<  Row,     Col,     Row,     F16,   F16,  F16,   F32,     F16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   128,    32,  128,    64,   8,   8,  32,   32,    1,    2,     S<8, 16, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<8, 16, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 16, 1, 8>,               8,  BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
         DeviceGemm_Xdl_CShuffleV3<  Row,     Col,     Row,     F16,   F16,  F16,   F32,     F16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,    16,  256,    64,   8,   8,  16,   16,    1,    4,     S<8, 16, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<8, 16, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 16, 1, 16>,              4,  BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
         DeviceGemm_Xdl_CShuffleV3<  Row,     Col,     Row,     F16,   F16,  F16,   F32,     F16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,    32,  256,    64,   8,   8,  32,   32,    1,    2,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 16, 1, 16>,              8,  BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,

library/src/tensor_operation_instance/gpu/gemm_universal/device_gemm_xdl_universal_f16_f8_f16/device_gemm_xdl_universal_f16_f8_f16_mk_kn_mn.hpp

@@ -47,7 +47,9 @@ using device_gemm_xdl_universal_f16_f8_f16_mk_kn_mn_comp_instances = std::tuple<
         // Disable due to test failure
         // DeviceGemm_Xdl_CShuffleV3<  Row,     Row,     Row,     F16,   F8,  F16,   F32,     F16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   256,   256,    32,   8,   4,  32,   32,    4,    4,     S<4, 64, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,               8,              4,          0,          1,           1,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v5>,
         DeviceGemm_Xdl_CShuffleV3<  Row,     Row,     Row,     F16,   F8,  F16,   F32,     F16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   256,   256,    32,   8,   4,  32,   32,    4,    4,     S<4, 64, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,               8,              4,          0,          1,           1,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3>,
+#if !defined(CK_USE_AMD_MFMA_GFX950)
         DeviceGemm_Xdl_CShuffleV3<  Row,     Row,     Row,     F16,   F8,  F16,   F32,     F16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   224,   256,    64,   8,   4,  16,   16,    7,    8,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<16, 16, 1>,    S<0, 2, 1>,     S<0, 2, 1>,             1,              16,              4,          0,          1,           2,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3>,
+#endif // !defined(CK_USE_AMD_MFMA_GFX950)
         DeviceGemm_Xdl_CShuffleV3<  Row,     Row,     Row,     F16,   F8,  F16,   F32,     F16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   128,   128,    64,   8,   4,  32,   32,    2,    2,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<16, 16, 1>,    S<0, 2, 1>,     S<0, 2, 1>,             1,               8,              4,          0,          1,           1,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3>,
         DeviceGemm_Xdl_CShuffleV3<  Row,     Row,     Row,     F16,   F8,  F16,   F32,     F16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   128,   256,    32,   8,   4,  32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,               8,              4,          0,          1,           1,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Interwave, BlockGemmPipelineVersion::v1>,
         DeviceGemm_Xdl_CShuffleV3<  Row,     Row,     Row,     F16,   F8,  F16,   F32,     F16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   128,   128,    64,   8,   4,  32,   32,    2,    2,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<16, 16, 1>,    S<0, 2, 1>,     S<0, 2, 1>,             1,               8,              4,          0,          1,           1,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Interwave, BlockGemmPipelineVersion::v1>

library/src/tensor_operation_instance/gpu/gemm_universal/device_gemm_xdl_universal_f16_f8_f16/device_gemm_xdl_universal_f16_f8_f16_mk_nk_mn.hpp

@@ -44,7 +44,9 @@ using device_gemm_xdl_universal_f16_f8_f16_mk_nk_mn_comp_instances = std::tuple<
 #endif // defined(CK_USE_AMD_MFMA_GFX950)
         // Compute friendly
         DeviceGemm_Xdl_CShuffleV3<  Row,     Col,     Row,     F16,    F8,  F16,   F32,     F16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   128,   128,    64,   8,  16,  32,   32,    2,    2,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<4, 64, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,            16,             16,          0,          1,           1,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v4>,
+#if !defined(CK_USE_AMD_MFMA_GFX950)
         DeviceGemm_Xdl_CShuffleV3<  Row,     Col,     Row,     F16,    F8,  F16,   F32,     F16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   224,   256,    64,   8,  16,  16,   16,    7,    8,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<4, 64, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,            16,             16,          0,          1,           2,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3>,
+#endif // !defined(CK_USE_AMD_MFMA_GFX950)
         DeviceGemm_Xdl_CShuffleV3<  Row,     Col,     Row,     F16,    F8,  F16,   F32,     F16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   128,   128,    64,   8,  16,  32,   32,    2,    2,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<4, 64, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,            16,             16,          0,          1,           1,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3>,
         DeviceGemm_Xdl_CShuffleV3<  Row,     Col,     Row,     F16,    F8,  F16,   F32,     F16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   128,   128,    64,   8,  16,  32,   32,    2,    2,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<4, 64, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,            16,             16,          0,          1,           1,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v5>,
         DeviceGemm_Xdl_CShuffleV3<  Row,     Col,     Row,     F16,    F8,  F16,   F32,     F16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   128,   128,    64,   8,  16,  32,   32,    2,    2,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<4, 64, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,            16,             16,          0,          1,           1,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Interwave, BlockGemmPipelineVersion::v1>

library/src/tensor_operation_instance/gpu/gemm_universal/device_gemm_xdl_universal_f8_f8_bf16/device_gemm_xdl_universal_f8_f8_bf16_mk_kn_mn.hpp

@@ -47,9 +47,11 @@ using device_gemm_xdl_universal_f8_f8_bf16_mk_kn_mn_comp_instances = std::tuple<
         DeviceGemm_Xdl_CShuffleV3<  Row,     Row,     Row,     F8,     F8,    BF16,   F32,     BF16,  PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   128,   128,   128,  16,   4,  32,   32,    2,    2,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,    S<32,  8, 1>,    S<0, 2, 1>,    S<0, 2, 1>,               1,             16,              4,          0,          1,           1,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v4, F8>,
         DeviceGemm_Xdl_CShuffleV3<  Row,     Row,     Row,     F8,     F8,    BF16,   F32,     BF16,  PassThrough, PassThrough, PassThrough,       GemmSpec,   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,          0,    S<16, 16, 1>,    S<0, 2, 1>,    S<0, 2, 1>,               1,              8,              4,          0,          1,           1,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v4, F8>,
         DeviceGemm_Xdl_CShuffleV3<  Row,     Row,     Row,     F8,     F8,    BF16,   F32,     BF16,  PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   256,   256,   128,  16,   8,  16,   16,    8,    8,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,    S<16, 16, 1>,    S<0, 2, 1>,    S<0, 2, 1>,               1,             16,              8,          0,          1,           2,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3, F8>,
+#if !defined(CK_USE_AMD_MFMA_GFX950)
         DeviceGemm_Xdl_CShuffleV3<  Row,     Row,     Row,     F8,     F8,    BF16,   F32,     BF16,  PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   256,   256,    64,  16,   4,  16,   16,    8,    8,     S<4, 64, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,    S<16, 16, 1>,    S<0, 2, 1>,    S<0, 2, 1>,               1,             16,              4,          0,          1,           2,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3, F8>,
         DeviceGemm_Xdl_CShuffleV3<  Row,     Row,     Row,     F8,     F8,    BF16,   F32,     BF16,  PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   224,   256,   128,  16,   8,  16,   16,    7,    8,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,    S<16, 16, 1>,    S<0, 2, 1>,    S<0, 2, 1>,               1,             16,              8,          0,          1,           2,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3, F8>,
         DeviceGemm_Xdl_CShuffleV3<  Row,     Row,     Row,     F8,     F8,    BF16,   F32,     BF16,  PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   128,   128,   128,  16,   4,  32,   32,    2,    2,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,    S<32,  8, 1>,    S<0, 2, 1>,    S<0, 2, 1>,               1,             16,              4,          0,          1,           1,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3, F8>,
+#endif // !defined(CK_USE_AMD_MFMA_GFX950)
         DeviceGemm_Xdl_CShuffleV3<  Row,     Row,     Row,     F8,     F8,    BF16,   F32,     BF16,  PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   128,   128,   128,  16,   4,  32,   32,    2,    2,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,    S<32,  8, 1>,    S<0, 2, 1>,    S<0, 2, 1>,               1,             16,              4,          0,          1,           1,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v5, F8>,
         DeviceGemm_Xdl_CShuffleV3<  Row,     Row,     Row,     F8,     F8,    BF16,   F32,     BF16,  PassThrough, PassThrough, PassThrough,       GemmSpec,   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,          0,    S<16, 16, 1>,    S<0, 2, 1>,    S<0, 2, 1>,               1,             16,              4,          0,          1,           1,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Interwave, BlockGemmPipelineVersion::v1, F8>,
         DeviceGemm_Xdl_CShuffleV3<  Row,     Row,     Row,     F8,     F8,    BF16,   F32,     BF16,  PassThrough, PassThrough, PassThrough,       GemmSpec,   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,          0,    S<16, 16, 1>,    S<0, 2, 1>,    S<0, 2, 1>,               1,              8,              4,          0,          1,           1,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Interwave, BlockGemmPipelineVersion::v1, F8>,

library/src/tensor_operation_instance/gpu/gemm_universal/device_gemm_xdl_universal_f8_f8_bf16/device_gemm_xdl_universal_f8_f8_bf16_mk_nk_mn.hpp

@@ -47,7 +47,9 @@ using device_gemm_xdl_universal_f8_f8_bf16_mk_nk_mn_comp_instances = std::tuple<
         DeviceGemm_Xdl_CShuffleV3<  Row,     Col,     Row,     F8,     F8,    BF16,   F32,     BF16,  PassThrough, PassThrough, PassThrough,       GemmSpec,   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,          0,    S<4, 64, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,          1,           1,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v4, F8>,
         DeviceGemm_Xdl_CShuffleV3<  Row,     Col,     Row,     F8,     F8,    BF16,   F32,     BF16,  PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   256,   256,   128,  16,  16,  16,   16,    8,    8,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,    S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,          1,           2,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3, F8>,
         DeviceGemm_Xdl_CShuffleV3<  Row,     Col,     Row,     F8,     F8,    BF16,   F32,     BF16,  PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   256,   256,    64,  16,  16,  16,   16,    8,    8,     S<4, 64, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,    S<4, 64, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,          1,           2,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3, F8>,
+#if !defined(CK_USE_AMD_MFMA_GFX950)
         DeviceGemm_Xdl_CShuffleV3<  Row,     Col,     Row,     F8,     F8,    BF16,   F32,     BF16,  PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   224,   256,    128, 16,  16,  16,   16,    7,    8,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,    S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,          1,           2,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3, F8>,
+#endif // !defined(CK_USE_AMD_MFMA_GFX950)
         DeviceGemm_Xdl_CShuffleV3<  Row,     Col,     Row,     F8,     F8,    BF16,   F32,     BF16,  PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   256,   224,    128, 16,  16,  16,   16,    8,    7,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,    S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,          2,           1,                   S<1, 64, 1, 4>,               8,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3, F8>,
         DeviceGemm_Xdl_CShuffleV3<  Row,     Col,     Row,     F8,     F8,    BF16,   F32,     BF16,  PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   128,   128,    128, 16,  16,  32,   32,    2,    2,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,    S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,          1,           1,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3, F8>,
         DeviceGemm_Xdl_CShuffleV3<  Row,     Col,     Row,     F8,     F8,    BF16,   F32,     BF16,  PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   128,   128,    128, 16,  16,  32,   32,    2,    2,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,    S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,          1,           1,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v5, F8>,

library/src/tensor_operation_instance/gpu/gemm_universal_batched/device_batched_gemm_xdl_universal_bf16_bf16_bf16/device_batched_gemm_xdl_universal_bf16_bf16_bf16_mk_nk_mn.hpp

@@ -50,7 +50,9 @@ using device_batched_gemm_xdl_universal_bf16_bf16_bf16_mk_nk_mn_comp_instances =
         DeviceBatchedGemmMultiD_Xdl_CShuffle_V3<  Row,     Col, DsLayout,     Row,     BF16,   BF16, DsDataType,  BF16,     F32,     BF16, PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   256,   256,    32,   2,   2,  32,   32,    4,    4,     S<16,16, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              2,              2,          0,    S<16,16, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              2,              2,          0,          1,           1,                   S<1, 16, 1, 16>,         S<4>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v4>,
         DeviceBatchedGemmMultiD_Xdl_CShuffle_V3<  Row,     Col, DsLayout,     Row,     BF16,   BF16, DsDataType,  BF16,     F32,     BF16, PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   256,   256,    32,   8,   8,  32,   32,    4,    4,     S<4, 64, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<4, 64, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 16, 1, 16>,         S<4>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3>,
         DeviceBatchedGemmMultiD_Xdl_CShuffle_V3<  Row,     Col, DsLayout,     Row,     BF16,   BF16, DsDataType,  BF16,     F32,     BF16, PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   256,   256,    32,   8,   8,  32,   32,    4,    4,     S<4, 64, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<4, 64, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 16, 1, 16>,         S<4>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v5>,
+#if !defined(CK_USE_AMD_MFMA_GFX950)
         DeviceBatchedGemmMultiD_Xdl_CShuffle_V3<  Row,     Col, DsLayout,     Row,     BF16,   BF16, DsDataType,  BF16,     F32,     BF16, PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   224,   256,    64,   8,   8,  16,   16,    7,    8,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           2,                   S<1, 16, 1, 16>,         S<4>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3>,
+#endif // !defined(CK_USE_AMD_MFMA_GFX950)
         DeviceBatchedGemmMultiD_Xdl_CShuffle_V3<  Row,     Col, DsLayout,     Row,     BF16,   BF16, DsDataType,  BF16,     F32,     BF16, PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   256,   224,    64,   8,   8,  16,   16,    8,    7,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          2,           1,                   S<1, 32, 1, 8>,          S<4>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3>,
         DeviceBatchedGemmMultiD_Xdl_CShuffle_V3<  Row,     Col, DsLayout,     Row,     BF16,   BF16, DsDataType,  BF16,     F32,     BF16, PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   256,   160,    64,   8,   8,  16,   16,    8,    5,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          2,           1,                   S<1, 32, 1, 8>,          S<4>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3>,
         DeviceBatchedGemmMultiD_Xdl_CShuffle_V3<  Row,     Col, DsLayout,     Row,     BF16,   BF16, DsDataType,  BF16,     F32,     BF16, PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   128,   160,    64,   8,   8,  32,   32,    1,    5,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 64, 1, 4>,          S<8>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3>,

library/src/tensor_operation_instance/gpu/gemm_universal_reduce/device_gemm_xdl_universal_bf16_bf16_bf16/device_gemm_xdl_universal_bf16_bf16_bf16_mk_kn_mn.hpp

@@ -50,7 +50,9 @@ using device_gemm_xdl_universal_reduce_bf16_bf16_bf16_mk_kn_mn_comp_instances =
         DeviceGemm_Xdl_CShuffleV3R1<     Row,     Row,  DsLayout,     Row,  BF16,   BF16, DsDataType,  BF16,   F32,     BF16,   PassThrough, PassThrough, PassThrough,      GemmSpec,   256,   128,   128,    64,   8,   4,  32,   32,    2,    2,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<16, 16, 1>,    S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              4,          0,          1,           1,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v4>,
         DeviceGemm_Xdl_CShuffleV3R1<     Row,     Row,  DsLayout,     Row,  BF16,   BF16, DsDataType,  BF16,   F32,     BF16,   PassThrough, PassThrough, PassThrough,      GemmSpec,   256,   256,   256,    32,   8,   4,  32,   32,    4,    4,     S<4, 64, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              4,          0,          1,           1,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v5>,
         DeviceGemm_Xdl_CShuffleV3R1<     Row,     Row,  DsLayout,     Row,  BF16,   BF16, DsDataType,  BF16,   F32,     BF16,   PassThrough, PassThrough, PassThrough,      GemmSpec,   256,   256,   256,    32,   8,   4,  32,   32,    4,    4,     S<4, 64, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              4,          0,          1,           1,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3>,
+#if !defined(CK_USE_AMD_MFMA_GFX950)
         DeviceGemm_Xdl_CShuffleV3R1<     Row,     Row,  DsLayout,     Row,  BF16,   BF16, DsDataType,  BF16,   F32,     BF16,   PassThrough, PassThrough, PassThrough,      GemmSpec,   256,   224,   256,    64,   8,   8,  16,   16,    7,    8,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,          0,          1,           2,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3>,
+#endif // !defined(CK_USE_AMD_MFMA_GFX950)
         DeviceGemm_Xdl_CShuffleV3R1<     Row,     Row,  DsLayout,     Row,  BF16,   BF16, DsDataType,  BF16,   F32,     BF16,   PassThrough, PassThrough, PassThrough,      GemmSpec,   256,   128,   128,    64,   8,   4,  32,   32,    2,    2,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<16, 16, 1>,    S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              4,          0,          1,           1,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3>,
         DeviceGemm_Xdl_CShuffleV3R1<     Row,     Row,  DsLayout,     Row,  BF16,   BF16, DsDataType,  BF16,   F32,     BF16,   PassThrough, PassThrough, PassThrough,      GemmSpec,   256,   128,   128,    64,   8,   4,  32,   32,    2,    2,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<16, 16, 1>,    S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              4,          0,          1,           1,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v5>,
         DeviceGemm_Xdl_CShuffleV3R1<     Row,     Row,  DsLayout,     Row,  BF16,   BF16, DsDataType,  BF16,   F32,     BF16,   PassThrough, PassThrough, PassThrough,      GemmSpec,   256,   128,   256,    32,   8,   4,  32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              4,          0,          1,           1,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Interwave, BlockGemmPipelineVersion::v1>,

library/src/tensor_operation_instance/gpu/gemm_universal_reduce/device_gemm_xdl_universal_f16_f16_f16/device_gemm_xdl_universal_f16_f16_f16_mk_kn_mn.hpp

@@ -50,7 +50,9 @@ using device_gemm_xdl_universal_reduce_f16_f16_f16_mk_kn_mn_comp_instances = std
         DeviceGemm_Xdl_CShuffleV3R1<     Row,     Row,  DsLayout,     Row,  F16,   F16, DsDataType,   F16,   F32,     F16,    PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   128,   128,    64,   8,   4,  32,   32,    2,    2,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<16, 16, 1>,    S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              4,          0,          1,           1,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v4>,
         DeviceGemm_Xdl_CShuffleV3R1<     Row,     Row,  DsLayout,     Row,  F16,   F16, DsDataType,   F16,   F32,     F16,    PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   256,   256,    32,   8,   4,  32,   32,    4,    4,     S<4, 64, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              4,          0,          1,           1,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v5>,
         DeviceGemm_Xdl_CShuffleV3R1<     Row,     Row,  DsLayout,     Row,  F16,   F16, DsDataType,   F16,   F32,     F16,    PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   256,   256,    32,   8,   4,  32,   32,    4,    4,     S<4, 64, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              4,          0,          1,           1,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3>,
+#if !defined(CK_USE_AMD_MFMA_GFX950)
         DeviceGemm_Xdl_CShuffleV3R1<     Row,     Row,  DsLayout,     Row,  F16,   F16, DsDataType,   F16,   F32,     F16,    PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   224,   256,    64,   8,   8,  16,   16,    7,    8,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,          0,          1,           2,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3>,
+#endif // !defined(CK_USE_AMD_MFMA_GFX950)
         DeviceGemm_Xdl_CShuffleV3R1<     Row,     Row,  DsLayout,     Row,  F16,   F16, DsDataType,   F16,   F32,     F16,    PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   128,   128,    64,   8,   4,  32,   32,    2,    2,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<16, 16, 1>,    S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              4,          0,          1,           1,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3>,
         DeviceGemm_Xdl_CShuffleV3R1<     Row,     Row,  DsLayout,     Row,  F16,   F16, DsDataType,   F16,   F32,     F16,    PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   128,   128,    64,   8,   4,  32,   32,    2,    2,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<16, 16, 1>,    S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              4,          0,          1,           1,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v5>,
         DeviceGemm_Xdl_CShuffleV3R1<     Row,     Row,  DsLayout,     Row,  F16,   F16, DsDataType,   F16,   F32,     F16,    PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   128,   256,    32,   8,   4,  32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              4,          0,          1,           1,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Interwave, BlockGemmPipelineVersion::v1>,

library/src/tensor_operation_instance/gpu/gemm_universal_streamk/device_gemm_xdl_universal_streamk_bf16_bf16_bf16/device_gemm_xdl_universal_streamk_bf16_bf16_bf16_mk_kn_mn.hpp

@@ -48,7 +48,9 @@ using device_gemm_xdl_universal_streamk_bf16_bf16_bf16_mk_kn_mn_comp_instances =
         DeviceGemm_Xdl_CShuffle_Streamk_V3<  Row,     Row,     Row,     BF16,   BF16,  BF16,   F32,     BF16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   128,   128,    64,   8,   4,  32,   32,    2,    2,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<16, 16, 1>,    S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              4,          0,          1,           1,                   S<1, 16, 1, 16>,               4,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v4>,
         DeviceGemm_Xdl_CShuffle_Streamk_V3<  Row,     Row,     Row,     BF16,   BF16,  BF16,   F32,     BF16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   256,   256,    32,   8,   4,  32,   32,    4,    4,     S<4, 64, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              4,          0,          1,           1,                   S<1, 16, 1, 16>,               4,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v5>,
         DeviceGemm_Xdl_CShuffle_Streamk_V3<  Row,     Row,     Row,     BF16,   BF16,  BF16,   F32,     BF16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   256,   256,    32,   8,   4,  32,   32,    4,    4,     S<4, 64, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              4,          0,          1,           1,                   S<1, 16, 1, 16>,               4,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3>,
+#if !defined(CK_USE_AMD_MFMA_GFX950)
         DeviceGemm_Xdl_CShuffle_Streamk_V3<  Row,     Row,     Row,     BF16,   BF16,  BF16,   F32,     BF16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   224,   256,    64,   8,   8,  16,   16,    7,    8,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,          0,          1,           2,                   S<1, 16, 1, 16>,               4,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3>,
+#endif // !defined(CK_USE_AMD_MFMA_GFX950)
         DeviceGemm_Xdl_CShuffle_Streamk_V3<  Row,     Row,     Row,     BF16,   BF16,  BF16,   F32,     BF16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   128,   128,    64,   8,   4,  32,   32,    2,    2,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<16, 16, 1>,    S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              4,          0,          1,           1,                   S<1, 16, 1, 16>,               4,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3>,
         DeviceGemm_Xdl_CShuffle_Streamk_V3<  Row,     Row,     Row,     BF16,   BF16,  BF16,   F32,     BF16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   128,   128,    64,   8,   4,  32,   32,    2,    2,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<16, 16, 1>,    S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              4,          0,          1,           1,                   S<1, 16, 1, 16>,               4,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v5>,
         DeviceGemm_Xdl_CShuffle_Streamk_V3<  Row,     Row,     Row,     BF16,   BF16,  BF16,   F32,     BF16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   128,   128,    64,   8,   4,  32,   32,    2,    2,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<16, 16, 1>,    S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              4,          0,          1,           1,                   S<1, 16, 1, 16>,               4,  BlockGemmPipelineScheduler::Interwave, BlockGemmPipelineVersion::v1>

library/src/tensor_operation_instance/gpu/gemm_universal_streamk/device_gemm_xdl_universal_streamk_bf16_bf16_bf16/device_gemm_xdl_universal_streamk_bf16_bf16_bf16_mk_nk_mn.hpp

@@ -51,11 +51,18 @@ using device_gemm_xdl_universal_streamk_bf16_bf16_bf16_mk_nk_mn_comp_instances =
         // AGPR Spill
         // DeviceGemm_Xdl_CShuffle_Streamk_V3<  Row,     Col,     Row,     BF16,   BF16,  BF16,   F32,     BF16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   256,   256,    32,   8,   8,  16,   16,    8,    8,     S<4, 64, 1>,     S<1, 0, 2>,    S<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,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v5>,
         // AGPR Spill when use permuted lds layout. so, use padding for these two.
+#if !defined(CK_USE_AMD_MFMA_GFX950)
         DeviceGemm_Xdl_CShuffle_Streamk_V3<  Row,     Col,     Row,     BF16,   BF16,  BF16,   F32,     BF16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   224,   256,    64,   8,   8,  16,   16,    7,    8,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           2,                   S<1, 16, 1, 16>,               4,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3>,
+
+#endif // !defined(CK_USE_AMD_MFMA_GFX950)
         DeviceGemm_Xdl_CShuffle_Streamk_V3<  Row,     Col,     Row,     BF16,   BF16,  BF16,   F32,     BF16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   256,   224,    64,   8,   8,  16,   16,    8,    7,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          2,           1,                   S<1, 32, 1, 8>,               4,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3>,
+
         DeviceGemm_Xdl_CShuffle_Streamk_V3<  Row,     Col,     Row,     BF16,   BF16,  BF16,   F32,     BF16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   128,   128,    64,   8,   8,  32,   32,    2,    2,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 16, 1, 16>,               4,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3>,
         DeviceGemm_Xdl_CShuffle_Streamk_V3<  Row,     Col,     Row,     BF16,   BF16,  BF16,   F32,     BF16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   128,   128,    64,   8,   8,  32,   32,    2,    2,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 16, 1, 16>,               4,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v5>,
         DeviceGemm_Xdl_CShuffle_Streamk_V3<  Row,     Col,     Row,     BF16,   BF16,  BF16,   F32,     BF16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   128,   128,    64,   8,   8,  32,   32,    2,    2,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 16, 1, 16>,               4,  BlockGemmPipelineScheduler::Interwave, BlockGemmPipelineVersion::v1>
+       
+        // DeviceGemm_Xdl_CShuffle_Streamk_V3<  Row,     Col,     Row,     BF16,   BF16,  BF16,   F32,     BF16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   128,   128,    64,   8,   8,  32,   32,    2,    2,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              2,              8,          0,    S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              2,              8,          0,          1,           1,                   S<1, 16, 1, 16>,               2,  BlockGemmPipelineScheduler::Interwave, BlockGemmPipelineVersion::v1>
+
     // clang-format on
     >;
 
@@ -85,6 +92,8 @@ using device_gemm_xdl_universal_streamk_bf16_bf16_bf16_mk_nk_mn_mem_instances =
         DeviceGemm_Xdl_CShuffle_Streamk_V3<  Row,     Col,     Row,     BF16,   BF16,  BF16,   F32,     BF16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   128,    16,   64,    64,   8,   8,  16,   16,    1,    2,     S<8, 16, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<8, 16, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 16, 1, 8>,               4,  BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
         DeviceGemm_Xdl_CShuffle_Streamk_V3<  Row,     Col,     Row,     BF16,   BF16,  BF16,   F32,     BF16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   128,    16,  128,    64,   8,   8,  16,   16,    1,    4,     S<8, 16, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<8, 16, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 16, 1, 8>,               4,  BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
         DeviceGemm_Xdl_CShuffle_Streamk_V3<  Row,     Col,     Row,     BF16,   BF16,  BF16,   F32,     BF16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,    16,  256,    64,   8,   8,  16,   16,    1,    4,     S<8, 16, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<8, 16, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 16, 1, 16>,              4,  BlkGemmPipeSched, BlockGemmPipelineVersion::v2>
+
+        // DeviceGemm_Xdl_CShuffle_Streamk_V3<  Row,     Col,     Row,     BF16,   BF16,  BF16,   F32,     BF16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,    16,  256,    64,   8,   8,  16,   16,    1,    4,     S<8, 16, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              2,              8,          0,    S<8, 16, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              2,              8,          0,          1,           1,                   S<1, 16, 1, 16>,              2,  BlkGemmPipeSched, BlockGemmPipelineVersion::v2>
     // clang-format on
     >;
 } // namespace instance

library/src/tensor_operation_instance/gpu/gemm_universal_streamk/device_gemm_xdl_universal_streamk_f16_f16_f16/device_gemm_xdl_universal_streamk_f16_f16_f16_mk_nk_mn.hpp

@@ -58,7 +58,9 @@ using device_gemm_xdl_universal_streamk_f16_f16_f16_mk_nk_mn_comp_instances = st
         DeviceGemm_Xdl_CShuffle_Streamk_V3<  Row,     Col,     Row,     F16,   F16,  F16,   F32,     F16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   256,   256,    32,   4,   4,  16,   16,    8,    8,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              4,              4,          0,    S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              4,              4,          0,          1,           2,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3>,
         DeviceGemm_Xdl_CShuffle_Streamk_V3<  Row,     Col,     Row,     F16,   F16,  F16,   F32,     F16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   256,   256,    32,   2,   2,  16,   16,    8,    8,     S<16,16, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              2,              2,          0,    S<16,16, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              2,              2,          0,          1,           2,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3>,
      
+#if !defined(CK_USE_AMD_MFMA_GFX950)
         DeviceGemm_Xdl_CShuffle_Streamk_V3<  Row,     Col,     Row,     F16,   F16,  F16,   F32,     F16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   224,   256,    64,   8,   8,  16,   16,    7,    8,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           2,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3>,
+#endif // !defined(CK_USE_AMD_MFMA_GFX950)
         DeviceGemm_Xdl_CShuffle_Streamk_V3<  Row,     Col,     Row,     F16,   F16,  F16,   F32,     F16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   256,   224,    64,   8,   8,  16,   16,    8,    7,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          2,           1,                   S<1, 64, 1, 4>,               8,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3>,
         DeviceGemm_Xdl_CShuffle_Streamk_V3<  Row,     Col,     Row,     F16,   F16,  F16,   F32,     F16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   128,   128,    64,   8,   8,  32,   32,    2,    2,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3>,
         DeviceGemm_Xdl_CShuffle_Streamk_V3<  Row,     Col,     Row,     F16,   F16,  F16,   F32,     F16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   128,   128,    64,   8,   8,  32,   32,    2,    2,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,    S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v5>,

library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/CMakeLists.txt

@@ -7,6 +7,22 @@ set(GROUPED_CONV2D_BWD_WEIGHT
     xdl/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp
     xdl/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_bf16_f32_bf16_instance.cpp
     xdl/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp
+    xdl/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_default_pipev1_instance.cpp
+    xdl/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_pad0_pipev1_instance.cpp
+    xdl/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_default_pipev1_instance.cpp
+    xdl/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_pad0_pipev1_instance.cpp
+    xdl/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_bf16_default_pipev2_instance.cpp
+    xdl/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_bf16_default_pipev5_instance.cpp
+    xdl/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_bf16_pad0_pipev2_instance.cpp
+    xdl/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_bf16_pad0_pipev5_instance.cpp
+    xdl/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f16_default_pipev2_instance.cpp
+    xdl/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f16_default_pipev5_instance.cpp
+    xdl/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f16_pad0_pipev2_instance.cpp
+    xdl/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f16_pad0_pipev5_instance.cpp
+    xdl/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f32_default_pipev2_instance.cpp
+    xdl/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f32_default_pipev5_instance.cpp
+    xdl/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f32_pad0_pipev2_instance.cpp
+    xdl/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f32_pad0_pipev5_instance.cpp
     xdl/device_grouped_conv2d_bwd_weight_two_stage_xdl_nhwgc_gkyxc_nhwgk_f16_pipev2_instance.cpp
     xdl/device_grouped_conv2d_bwd_weight_two_stage_xdl_nhwgc_gkyxc_nhwgk_f16_pipev5_instance.cpp
     xdl/device_grouped_conv2d_bwd_weight_two_stage_xdl_ngchw_gkyxc_ngkhw_f16_pipev2_instance.cpp

library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/xdl/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_default_pipev1_instance.cpp

@@ -0,0 +1,38 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2025, 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_v3_xdl_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_xdl_gnhwc_gkyxc_gnhwk_f16_default_pipev1_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
+                                                           GNHWC,
+                                                           GKYXC,
+                                                           GNHWK,
+                                                           F16,
+                                                           F16,
+                                                           F16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_bwd_weight_v3_xdl_c_shuffle_f16_instances<
+                                       2,
+                                       GNHWC,
+                                       GKYXC,
+                                       GNHWK,
+                                       ConvBwdWeightDefault,
+                                       BlockGemmPipelineScheduler::Intrawave,
+                                       BlockGemmPipelineVersion::v1>{});
+}
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/xdl/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_pad0_pipev1_instance.cpp

@@ -0,0 +1,39 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2025, 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_v3_xdl_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_xdl_gnhwc_gkyxc_gnhwk_f16_pad0_pipev1_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
+                                                           GNHWC,
+                                                           GKYXC,
+                                                           GNHWK,
+                                                           F16,
+                                                           F16,
+                                                           F16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_bwd_weight_v3_xdl_c_shuffle_f16_instances<
+                                       2,
+                                       GNHWC,
+                                       GKYXC,
+                                       GNHWK,
+                                       ConvBwdWeightFilter1x1Stride1Pad0,
+                                       BlockGemmPipelineScheduler::Intrawave,
+                                       BlockGemmPipelineVersion::v1>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/xdl/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_default_pipev1_instance.cpp

@@ -0,0 +1,39 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2025, 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_v3_xdl_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_xdl_gnhwc_gkyxc_gnhwk_f32_default_pipev1_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
+                                                           GNHWC,
+                                                           GKYXC,
+                                                           GNHWK,
+                                                           F32,
+                                                           F32,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_bwd_weight_v3_xdl_c_shuffle_f32_instances<
+                                       2,
+                                       GNHWC,
+                                       GKYXC,
+                                       GNHWK,
+                                       ConvBwdWeightDefault,
+                                       BlockGemmPipelineScheduler::Intrawave,
+                                       BlockGemmPipelineVersion::v1>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/xdl/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_pad0_pipev1_instance.cpp

@@ -0,0 +1,39 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2025, 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_v3_xdl_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_xdl_gnhwc_gkyxc_gnhwk_f32_pad0_pipev1_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
+                                                           GNHWC,
+                                                           GKYXC,
+                                                           GNHWK,
+                                                           F32,
+                                                           F32,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_bwd_weight_v3_xdl_c_shuffle_f32_instances<
+                                       2,
+                                       GNHWC,
+                                       GKYXC,
+                                       GNHWK,
+                                       ConvBwdWeightFilter1x1Stride1Pad0,
+                                       BlockGemmPipelineScheduler::Intrawave,
+                                       BlockGemmPipelineVersion::v1>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/xdl/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_bf16_default_pipev2_instance.cpp

@@ -0,0 +1,39 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2025, 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_v3_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_default_pipev2_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
+                                                           NHWGC,
+                                                           GKYXC,
+                                                           NHWGK,
+                                                           BF16,
+                                                           BF16,
+                                                           BF16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_bwd_weight_v3_xdl_c_shuffle_bf16_instances<
+                                       2,
+                                       NHWGC,
+                                       GKYXC,
+                                       NHWGK,
+                                       ConvBwdWeightDefault,
+                                       BlockGemmPipelineScheduler::Intrawave,
+                                       BlockGemmPipelineVersion::v2>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/xdl/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_bf16_default_pipev5_instance.cpp

@@ -0,0 +1,39 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2025, 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_v3_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_default_pipev5_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
+                                                           NHWGC,
+                                                           GKYXC,
+                                                           NHWGK,
+                                                           BF16,
+                                                           BF16,
+                                                           BF16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_bwd_weight_v3_xdl_c_shuffle_bf16_instances<
+                                       2,
+                                       NHWGC,
+                                       GKYXC,
+                                       NHWGK,
+                                       ConvBwdWeightDefault,
+                                       BlockGemmPipelineScheduler::Intrawave,
+                                       BlockGemmPipelineVersion::v5>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/xdl/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_bf16_pad0_pipev2_instance.cpp

@@ -0,0 +1,40 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2025, 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_v3_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_pad0_pipev2_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
+                                                           NHWGC,
+                                                           GKYXC,
+                                                           NHWGK,
+                                                           BF16,
+                                                           BF16,
+                                                           BF16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances)
+{
+
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_bwd_weight_v3_xdl_c_shuffle_bf16_instances<
+                                       2,
+                                       NHWGC,
+                                       GKYXC,
+                                       NHWGK,
+                                       ConvBwdWeightFilter1x1Stride1Pad0,
+                                       BlockGemmPipelineScheduler::Intrawave,
+                                       BlockGemmPipelineVersion::v2>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/xdl/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_bf16_pad0_pipev5_instance.cpp

@@ -0,0 +1,40 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2025, 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_v3_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_pad0_pipev5_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
+                                                           NHWGC,
+                                                           GKYXC,
+                                                           NHWGK,
+                                                           BF16,
+                                                           BF16,
+                                                           BF16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances)
+{
+
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_bwd_weight_v3_xdl_c_shuffle_bf16_instances<
+                                       2,
+                                       NHWGC,
+                                       GKYXC,
+                                       NHWGK,
+                                       ConvBwdWeightFilter1x1Stride1Pad0,
+                                       BlockGemmPipelineScheduler::Intrawave,
+                                       BlockGemmPipelineVersion::v5>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/xdl/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f16_default_pipev2_instance.cpp

@@ -0,0 +1,40 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2025, 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_v3_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_default_pipev2_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
+                                                           NHWGC,
+                                                           GKYXC,
+                                                           NHWGK,
+                                                           F16,
+                                                           F16,
+                                                           F16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_bwd_weight_v3_xdl_c_shuffle_f16_instances<
+                                       2,
+                                       NHWGC,
+                                       GKYXC,
+                                       NHWGK,
+                                       ConvBwdWeightDefault,
+                                       BlockGemmPipelineScheduler::Intrawave,
+                                       BlockGemmPipelineVersion::v2>{});
+    ;
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/xdl/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f16_default_pipev5_instance.cpp

@@ -0,0 +1,40 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2025, 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_v3_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_default_pipev5_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
+                                                           NHWGC,
+                                                           GKYXC,
+                                                           NHWGK,
+                                                           F16,
+                                                           F16,
+                                                           F16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_bwd_weight_v3_xdl_c_shuffle_f16_instances<
+                                       2,
+                                       NHWGC,
+                                       GKYXC,
+                                       NHWGK,
+                                       ConvBwdWeightDefault,
+                                       BlockGemmPipelineScheduler::Intrawave,
+                                       BlockGemmPipelineVersion::v5>{});
+    ;
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/xdl/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f16_pad0_pipev2_instance.cpp

@@ -0,0 +1,39 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2025, 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_v3_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_pad0_pipev2_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
+                                                           NHWGC,
+                                                           GKYXC,
+                                                           NHWGK,
+                                                           F16,
+                                                           F16,
+                                                           F16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_bwd_weight_v3_xdl_c_shuffle_f16_instances<
+                                       2,
+                                       NHWGC,
+                                       GKYXC,
+                                       NHWGK,
+                                       ConvBwdWeightFilter1x1Stride1Pad0,
+                                       BlockGemmPipelineScheduler::Intrawave,
+                                       BlockGemmPipelineVersion::v2>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/xdl/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f16_pad0_pipev5_instance.cpp

@@ -0,0 +1,39 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2025, 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_v3_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_pad0_pipev5_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
+                                                           NHWGC,
+                                                           GKYXC,
+                                                           NHWGK,
+                                                           F16,
+                                                           F16,
+                                                           F16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_bwd_weight_v3_xdl_c_shuffle_f16_instances<
+                                       2,
+                                       NHWGC,
+                                       GKYXC,
+                                       NHWGK,
+                                       ConvBwdWeightFilter1x1Stride1Pad0,
+                                       BlockGemmPipelineScheduler::Intrawave,
+                                       BlockGemmPipelineVersion::v5>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/xdl/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f32_default_pipev2_instance.cpp

@@ -0,0 +1,39 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2025, 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_v3_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_default_pipev2_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
+                                                           NHWGC,
+                                                           GKYXC,
+                                                           NHWGK,
+                                                           F32,
+                                                           F32,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_bwd_weight_v3_xdl_c_shuffle_f32_instances<
+                                       2,
+                                       NHWGC,
+                                       GKYXC,
+                                       NHWGK,
+                                       ConvBwdWeightDefault,
+                                       BlockGemmPipelineScheduler::Intrawave,
+                                       BlockGemmPipelineVersion::v2>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/xdl/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f32_default_pipev5_instance.cpp

@@ -0,0 +1,39 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2025, 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_v3_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_default_pipev5_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
+                                                           NHWGC,
+                                                           GKYXC,
+                                                           NHWGK,
+                                                           F32,
+                                                           F32,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_bwd_weight_v3_xdl_c_shuffle_f32_instances<
+                                       2,
+                                       NHWGC,
+                                       GKYXC,
+                                       NHWGK,
+                                       ConvBwdWeightDefault,
+                                       BlockGemmPipelineScheduler::Intrawave,
+                                       BlockGemmPipelineVersion::v5>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/xdl/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f32_pad0_pipev2_instance.cpp

@@ -0,0 +1,39 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2025, 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_v3_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_pad0_pipev2_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
+                                                           NHWGC,
+                                                           GKYXC,
+                                                           NHWGK,
+                                                           F32,
+                                                           F32,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_bwd_weight_v3_xdl_c_shuffle_f32_instances<
+                                       2,
+                                       NHWGC,
+                                       GKYXC,
+                                       NHWGK,
+                                       ConvBwdWeightFilter1x1Stride1Pad0,
+                                       BlockGemmPipelineScheduler::Intrawave,
+                                       BlockGemmPipelineVersion::v2>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/xdl/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f32_pad0_pipev5_instance.cpp

@@ -0,0 +1,39 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2025, 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_v3_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_pad0_pipev5_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
+                                                           NHWGC,
+                                                           GKYXC,
+                                                           NHWGK,
+                                                           F32,
+                                                           F32,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_bwd_weight_v3_xdl_c_shuffle_f32_instances<
+                                       2,
+                                       NHWGC,
+                                       GKYXC,
+                                       NHWGK,
+                                       ConvBwdWeightFilter1x1Stride1Pad0,
+                                       BlockGemmPipelineScheduler::Intrawave,
+                                       BlockGemmPipelineVersion::v5>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/CMakeLists.txt

@@ -7,22 +7,34 @@ set(GROUPED_CONV3D_BWD_WEIGHT
      xdl/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp
      xdl/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_f32_bf16_instance.cpp
      xdl/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp
-    xdl/device_grouped_conv3d_bwd_weight_two_stage_xdl_ndhwgc_gkzyxc_ndhwgk_f16_pipev2_instance.cpp
-    xdl/device_grouped_conv3d_bwd_weight_two_stage_xdl_ndhwgc_gkzyxc_ndhwgk_f16_pipev5_instance.cpp
-    xdl/device_grouped_conv3d_bwd_weight_two_stage_xdl_ngcdhw_gkzyxc_ngkdhw_f16_pipev2_instance.cpp
-    xdl/device_grouped_conv3d_bwd_weight_two_stage_xdl_ngcdhw_gkzyxc_ngkdhw_f16_pipev5_instance.cpp
-    xdl/device_grouped_conv3d_bwd_weight_two_stage_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_pipev2_instance.cpp
-    xdl/device_grouped_conv3d_bwd_weight_two_stage_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_pipev5_instance.cpp
-    xdl/device_grouped_conv3d_bwd_weight_two_stage_xdl_ngcdhw_gkzyxc_ngkdhw_bf16_pipev2_instance.cpp
-    xdl/device_grouped_conv3d_bwd_weight_two_stage_xdl_ngcdhw_gkzyxc_ngkdhw_bf16_pipev5_instance.cpp
-    xdl/device_grouped_conv3d_bwd_weight_two_stage_xdl_ndhwgc_gkzyxc_ndhwgk_f16_pipev1_instance.cpp
-    xdl/device_grouped_conv3d_bwd_weight_two_stage_xdl_ngcdhw_gkzyxc_ngkdhw_f16_pipev1_instance.cpp
-    xdl/device_grouped_conv3d_bwd_weight_two_stage_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_pipev1_instance.cpp
-    xdl/device_grouped_conv3d_bwd_weight_two_stage_xdl_ngcdhw_gkzyxc_ngkdhw_bf16_pipev1_instance.cpp
-    xdl/device_grouped_conv3d_bwd_weight_two_stage_xdl_ndhwgc_gkzyxc_ndhwgk_f16_pipev2_irregular_instance.cpp
-    xdl/device_grouped_conv3d_bwd_weight_two_stage_xdl_ndhwgc_gkzyxc_ndhwgk_f16_pipev5_irregular_instance.cpp
-    xdl/device_grouped_conv3d_bwd_weight_two_stage_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_pipev2_irregular_instance.cpp
-    xdl/device_grouped_conv3d_bwd_weight_two_stage_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_pipev5_irregular_instance.cpp
+     xdl/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_default_pipev2_instance.cpp
+     xdl/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_default_pipev5_instance.cpp
+     xdl/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_pad0_pipev2_instance.cpp
+     xdl/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_pad0_pipev5_instance.cpp
+     xdl/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f16_default_pipev2_instance.cpp
+     xdl/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f16_default_pipev5_instance.cpp
+     xdl/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f16_pad0_pipev2_instance.cpp
+     xdl/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f16_pad0_pipev5_instance.cpp
+     xdl/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f32_default_pipev2_instance.cpp
+     xdl/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f32_default_pipev5_instance.cpp
+     xdl/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f32_pad0_pipev2_instance.cpp
+     xdl/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f32_pad0_pipev5_instance.cpp
+     xdl/device_grouped_conv3d_bwd_weight_two_stage_xdl_ndhwgc_gkzyxc_ndhwgk_f16_pipev2_instance.cpp
+     xdl/device_grouped_conv3d_bwd_weight_two_stage_xdl_ndhwgc_gkzyxc_ndhwgk_f16_pipev5_instance.cpp
+     xdl/device_grouped_conv3d_bwd_weight_two_stage_xdl_ngcdhw_gkzyxc_ngkdhw_f16_pipev2_instance.cpp
+     xdl/device_grouped_conv3d_bwd_weight_two_stage_xdl_ngcdhw_gkzyxc_ngkdhw_f16_pipev5_instance.cpp
+     xdl/device_grouped_conv3d_bwd_weight_two_stage_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_pipev2_instance.cpp
+     xdl/device_grouped_conv3d_bwd_weight_two_stage_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_pipev5_instance.cpp
+     xdl/device_grouped_conv3d_bwd_weight_two_stage_xdl_ngcdhw_gkzyxc_ngkdhw_bf16_pipev2_instance.cpp
+     xdl/device_grouped_conv3d_bwd_weight_two_stage_xdl_ngcdhw_gkzyxc_ngkdhw_bf16_pipev5_instance.cpp
+     xdl/device_grouped_conv3d_bwd_weight_two_stage_xdl_ndhwgc_gkzyxc_ndhwgk_f16_pipev1_instance.cpp
+     xdl/device_grouped_conv3d_bwd_weight_two_stage_xdl_ngcdhw_gkzyxc_ngkdhw_f16_pipev1_instance.cpp
+     xdl/device_grouped_conv3d_bwd_weight_two_stage_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_pipev1_instance.cpp
+     xdl/device_grouped_conv3d_bwd_weight_two_stage_xdl_ngcdhw_gkzyxc_ngkdhw_bf16_pipev1_instance.cpp
+     xdl/device_grouped_conv3d_bwd_weight_two_stage_xdl_ndhwgc_gkzyxc_ndhwgk_f16_pipev2_irregular_instance.cpp
+     xdl/device_grouped_conv3d_bwd_weight_two_stage_xdl_ndhwgc_gkzyxc_ndhwgk_f16_pipev5_irregular_instance.cpp
+     xdl/device_grouped_conv3d_bwd_weight_two_stage_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_pipev2_irregular_instance.cpp
+     xdl/device_grouped_conv3d_bwd_weight_two_stage_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_pipev5_irregular_instance.cpp
     )
 
 if(DL_KERNELS)

library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/xdl/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_default_pipev2_instance.cpp

@@ -0,0 +1,39 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2025, 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_v3_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_default_pipev2_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
+                                                           NDHWGC,
+                                                           GKZYXC,
+                                                           NDHWGK,
+                                                           BF16,
+                                                           BF16,
+                                                           BF16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_bwd_weight_v3_xdl_c_shuffle_bf16_instances<
+                                       3,
+                                       NDHWGC,
+                                       GKZYXC,
+                                       NDHWGK,
+                                       ConvBwdWeightDefault,
+                                       BlockGemmPipelineScheduler::Intrawave,
+                                       BlockGemmPipelineVersion::v2>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/xdl/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_default_pipev5_instance.cpp

@@ -0,0 +1,39 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2025, 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_v3_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_default_pipev5_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
+                                                           NDHWGC,
+                                                           GKZYXC,
+                                                           NDHWGK,
+                                                           BF16,
+                                                           BF16,
+                                                           BF16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_bwd_weight_v3_xdl_c_shuffle_bf16_instances<
+                                       3,
+                                       NDHWGC,
+                                       GKZYXC,
+                                       NDHWGK,
+                                       ConvBwdWeightDefault,
+                                       BlockGemmPipelineScheduler::Intrawave,
+                                       BlockGemmPipelineVersion::v5>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/xdl/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_pad0_pipev2_instance.cpp

@@ -0,0 +1,39 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2025, 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_v3_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_pad0_pipev2_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
+                                                           NDHWGC,
+                                                           GKZYXC,
+                                                           NDHWGK,
+                                                           BF16,
+                                                           BF16,
+                                                           BF16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_bwd_weight_v3_xdl_c_shuffle_bf16_instances<
+                                       3,
+                                       NDHWGC,
+                                       GKZYXC,
+                                       NDHWGK,
+                                       ConvBwdWeightFilter1x1Stride1Pad0,
+                                       BlockGemmPipelineScheduler::Intrawave,
+                                       BlockGemmPipelineVersion::v2>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/xdl/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_pad0_pipev5_instance.cpp

@@ -0,0 +1,39 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2025, 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_v3_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_pad0_pipev5_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
+                                                           NDHWGC,
+                                                           GKZYXC,
+                                                           NDHWGK,
+                                                           BF16,
+                                                           BF16,
+                                                           BF16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_bwd_weight_v3_xdl_c_shuffle_bf16_instances<
+                                       3,
+                                       NDHWGC,
+                                       GKZYXC,
+                                       NDHWGK,
+                                       ConvBwdWeightFilter1x1Stride1Pad0,
+                                       BlockGemmPipelineScheduler::Intrawave,
+                                       BlockGemmPipelineVersion::v5>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/xdl/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f16_default_pipev2_instance.cpp

@@ -0,0 +1,39 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2025, 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_v3_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_default_pipev2_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_v3_xdl_c_shuffle_f16_instances<
+                                       3,
+                                       NDHWGC,
+                                       GKZYXC,
+                                       NDHWGK,
+                                       ConvBwdWeightDefault,
+                                       BlockGemmPipelineScheduler::Intrawave,
+                                       BlockGemmPipelineVersion::v2>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/xdl/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f16_default_pipev5_instance.cpp

@@ -0,0 +1,39 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2025, 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_v3_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_default_pipev5_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_v3_xdl_c_shuffle_f16_instances<
+                                       3,
+                                       NDHWGC,
+                                       GKZYXC,
+                                       NDHWGK,
+                                       ConvBwdWeightDefault,
+                                       BlockGemmPipelineScheduler::Intrawave,
+                                       BlockGemmPipelineVersion::v5>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/xdl/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f16_pad0_pipev2_instance.cpp

@@ -0,0 +1,39 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2025, 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_v3_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_pad0_pipev2_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_v3_xdl_c_shuffle_f16_instances<
+                                       3,
+                                       NDHWGC,
+                                       GKZYXC,
+                                       NDHWGK,
+                                       ConvBwdWeightFilter1x1Stride1Pad0,
+                                       BlockGemmPipelineScheduler::Intrawave,
+                                       BlockGemmPipelineVersion::v2>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/xdl/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f16_pad0_pipev5_instance.cpp

@@ -0,0 +1,39 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2025, 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_v3_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_pad0_pipev5_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_v3_xdl_c_shuffle_f16_instances<
+                                       3,
+                                       NDHWGC,
+                                       GKZYXC,
+                                       NDHWGK,
+                                       ConvBwdWeightFilter1x1Stride1Pad0,
+                                       BlockGemmPipelineScheduler::Intrawave,
+                                       BlockGemmPipelineVersion::v5>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/xdl/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f32_default_pipev2_instance.cpp

@@ -0,0 +1,39 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2025, 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_v3_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_default_pipev2_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
+                                                           NDHWGC,
+                                                           GKZYXC,
+                                                           NDHWGK,
+                                                           F32,
+                                                           F32,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_bwd_weight_v3_xdl_c_shuffle_f32_instances<
+                                       3,
+                                       NDHWGC,
+                                       GKZYXC,
+                                       NDHWGK,
+                                       ConvBwdWeightDefault,
+                                       BlockGemmPipelineScheduler::Intrawave,
+                                       BlockGemmPipelineVersion::v2>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/xdl/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f32_default_pipev5_instance.cpp

@@ -0,0 +1,39 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2025, 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_v3_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_default_pipev5_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
+                                                           NDHWGC,
+                                                           GKZYXC,
+                                                           NDHWGK,
+                                                           F32,
+                                                           F32,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_bwd_weight_v3_xdl_c_shuffle_f32_instances<
+                                       3,
+                                       NDHWGC,
+                                       GKZYXC,
+                                       NDHWGK,
+                                       ConvBwdWeightDefault,
+                                       BlockGemmPipelineScheduler::Intrawave,
+                                       BlockGemmPipelineVersion::v5>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/xdl/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f32_pad0_pipev2_instance.cpp

@@ -0,0 +1,39 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2025, 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_v3_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_pad0_pipev2_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
+                                                           NDHWGC,
+                                                           GKZYXC,
+                                                           NDHWGK,
+                                                           F32,
+                                                           F32,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_bwd_weight_v3_xdl_c_shuffle_f32_instances<
+                                       3,
+                                       NDHWGC,
+                                       GKZYXC,
+                                       NDHWGK,
+                                       ConvBwdWeightFilter1x1Stride1Pad0,
+                                       BlockGemmPipelineScheduler::Intrawave,
+                                       BlockGemmPipelineVersion::v2>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/xdl/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f32_pad0_pipev5_instance.cpp

@@ -0,0 +1,39 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2025, 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_v3_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_pad0_pipev5_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
+                                                           NDHWGC,
+                                                           GKZYXC,
+                                                           NDHWGK,
+                                                           F32,
+                                                           F32,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_bwd_weight_v3_xdl_c_shuffle_f32_instances<
+                                       3,
+                                       NDHWGC,
+                                       GKZYXC,
+                                       NDHWGK,
+                                       ConvBwdWeightFilter1x1Stride1Pad0,
+                                       BlockGemmPipelineScheduler::Intrawave,
+                                       BlockGemmPipelineVersion::v5>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

profiler/src/profile_gemm_universal_streamk.cpp

@@ -56,6 +56,26 @@ int profile_gemm_universal_streamk(int argc, char* argv[])
         exit(1);
     }
 
+    int M;
+    int N;
+    int StrideA;
+    int StrideB;
+    // Analyze the unsupported matrix shapes, switch the M and N number
+    if(std::stoi(argv[9]) % 8 != 0 && std::stoi(argv[8]) % 8 == 0)
+    {
+        M       = std::stoi(argv[9]);
+        StrideA = std::stoi(argv[12]);
+        N       = std::stoi(argv[8]);
+        StrideB = std::stoi(argv[11]);
+    }
+    else
+    {
+        M       = std::stoi(argv[8]);
+        StrideA = std::stoi(argv[11]);
+        N       = std::stoi(argv[9]);
+        StrideB = std::stoi(argv[12]);
+    }
+
     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]);
@@ -63,12 +83,8 @@ int profile_gemm_universal_streamk(int argc, char* argv[])
     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 Streamk_sel = std::stoi(argv[14]);
     const int Grid_size   = std::stoi(argv[15]);

test/grouped_convnd_bwd_weight/CMakeLists.txt

@@ -9,6 +9,10 @@ add_gtest_executable(test_grouped_convnd_bwd_weight_interface_xdl test_grouped_c
 if(result EQUAL 0)
    target_link_libraries(test_grouped_convnd_bwd_weight_interface_xdl PRIVATE utility)
 endif()
+add_gtest_executable(test_grouped_convnd_bwd_weight_v3_interface_xdl test_grouped_convnd_bwd_weight_v3_interface_xdl.cpp)
+if(result EQUAL 0)
+   target_link_libraries(test_grouped_convnd_bwd_weight_v3_interface_xdl PRIVATE utility)
+endif()
 add_gtest_executable(test_grouped_convnd_bwd_weight_interface_wmma test_grouped_convnd_bwd_weight_interface_wmma.cpp)
 if(result EQUAL 0)
    target_link_libraries(test_grouped_convnd_bwd_weight_interface_wmma PRIVATE utility)

test/grouped_convnd_bwd_weight/test_grouped_convnd_bwd_weight_v3_interface_xdl.cpp

@@ -0,0 +1,179 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2025, 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_v3.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_CShuffleV3
+        //##########|     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,     32,  8,   32,   32,    1,    2,  S<4, 4,  1>,  S<2, 0, 1>,  S<1, 0, 2>,              2,              8,              1,      true,  S<4, 4,  1>,  S<2, 0, 1>,  S<1, 0, 2>,             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 + 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{},
+                                          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 TestGroupedConvndBwdWeightV3Default
+    : public TestGroupedConvndBwdWeight<Tuple, ConvBwdWeightDefault>
+{
+};
+
+template <typename Tuple>
+class TestGroupedConvndBwdWeightV3Filter1x1
+    : public TestGroupedConvndBwdWeight<Tuple, Filter1x1Stride1Pad0>
+{
+};
+
+TYPED_TEST_SUITE(TestGroupedConvndBwdWeightV3Default, KernelTypes);
+TYPED_TEST_SUITE(TestGroupedConvndBwdWeightV3Filter1x1, KernelTypes);
+
+TYPED_TEST(TestGroupedConvndBwdWeightV3Filter1x1, 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);
+
+    // Not 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_FALSE(is_supported);
+}
+
+TYPED_TEST(TestGroupedConvndBwdWeightV3Default, 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);
+}