Merge commit 'ce99cab6056d1ffef5acb6f4ad7ede87a46a3cfc' into develop

example/65_gemm_multiply_multiply/CMakeLists.txt

@@ -77,3 +77,5 @@ example_compile_options(example_moe_gemm1_xdl_fp8_blockscale PRIVATE ${BLOCKSCAL
 add_example_executable(example_gemm_add_add_wmma_fp16 gemm_add_add_wmma_fp16.cpp)
 add_example_executable(example_gemm_multiply_multiply_wmma_fp16_bpreshuffle gemm_multiply_multiply_wmma_fp16_bpreshuffle.cpp)
 add_example_executable(example_gemm_multiply_multiply_wmma_fp8_bpreshuffle gemm_multiply_multiply_wmma_fp8_bpreshuffle.cpp)
+add_example_executable(example_gemm_multiply_multiply_wmma_fp8_ab_scale gemm_multiply_multiply_wmma_fp8_ab_scale.cpp)
+add_example_executable(example_gemm_multiply_multiply_wmma_fp8_blockscale_bpreshuffle gemm_multiply_multiply_wmma_fp8_blockscale_bpreshuffle.cpp)

example/65_gemm_multiply_multiply/gemm_multiply_multiply_wmma_fp8_ab_scale.cpp

@@ -0,0 +1,345 @@
+// Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
+// SPDX-License-Identifier: MIT
+
+#include <iostream>
+#include <numeric>
+#include <initializer_list>
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_wmma_cshuffle_v3_ab_scale.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp"
+
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/literals.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
+#include "ck/library/utility/check_err.hpp"
+
+#include "ck/utility/blkgemmpipe_scheduler.hpp"
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using BF16 = ck::bhalf_t;
+using FP8  = ck::f8_t;
+using F32  = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+using A0DataType       = FP8;
+using A1DataType       = F32;
+using B0DataType       = FP8;
+using B1DataType       = F32;
+using AccDataType      = F32;
+using CShuffleDataType = F32;
+using DsDataType       = ck::Tuple<>;
+using EDataType        = BF16;
+
+using A0Layout = Row;
+using B0Layout = Col;
+using DsLayout = ck::Tuple<>;
+using ELayout  = Row;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+using AElementOp   = PassThrough;
+using BElementOp   = PassThrough;
+using CDEElementOp = PassThrough;
+
+static constexpr auto GemmSpec = ck::tensor_operation::device::GemmSpecialization::Default;
+
+static constexpr ck::index_t Scale_Block_M = 1;
+static constexpr ck::index_t Scale_Block_N = 128;
+static constexpr ck::index_t Scale_Block_K = 128;
+
+using DeviceOpInstance = ck::tensor_operation::device::DeviceGemmMultiD_ABScale_Wmma_CShuffle_V3
+    // clang-format off
+         <Row, Col, DsLayout, ELayout,
+          A0DataType, A1DataType, B0DataType, B1DataType, DsDataType, EDataType,
+          AccDataType, CShuffleDataType, 
+          AElementOp,  BElementOp, CDEElementOp, GemmSpec,
+          256, Scale_Block_M, Scale_Block_N, Scale_Block_K,
+          128, 128, 128,
+          16, 16,
+          16, 16,
+          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>,
+          ck::BlockGemmPipelineScheduler::Intrawave, ck::BlockGemmPipelineVersion::v3, FP8>;
+// clang-format on
+
+int main(int argc, char* argv[])
+{
+    bool do_verification = true;
+    int init_method      = 1;
+    bool time_kernel     = false;
+    bool flush_cache     = true;
+
+    // GEMM shape
+    ck::index_t M = 128;
+    ck::index_t N = 1024;
+    ck::index_t K = 1024;
+
+    ck::index_t StrideA = K;
+    ck::index_t StrideB = K;
+    ck::index_t StrideE = N;
+
+    ck::index_t KBatch = 1;
+
+    if(argc == 1)
+    {
+        // use default case
+    }
+    else if(argc == 4)
+    {
+        do_verification = std::stoi(argv[1]);
+        init_method     = std::stoi(argv[2]);
+        time_kernel     = std::stoi(argv[3]);
+    }
+    else if(argc == 8 || argc == 9)
+    {
+        do_verification = std::stoi(argv[1]);
+        init_method     = std::stoi(argv[2]);
+        time_kernel     = std::stoi(argv[3]);
+
+        M = std::stoi(argv[4]);
+        N = std::stoi(argv[5]);
+        K = std::stoi(argv[6]);
+
+        flush_cache = std::stoi(argv[7]);
+
+        if(argc == 9)
+        {
+            KBatch = std::stoi(argv[8]);
+        }
+
+        StrideA = K;
+        StrideB = K;
+        StrideE = N;
+    }
+    else
+    {
+        printf("arg1: verification (0=no, 1=yes)\n");
+        printf("arg2: initialization (0=no init, 1=integer value, 2=decimal value)\n");
+        printf("arg3: time kernel (0=no, 1=yes)\n");
+        printf("arg4 to 6: M, N, K\n");
+        printf("arg7: flush both I$ and L2$ (0=no, 1=yes)\n");
+        printf("arg8: KBatch (default: 1)\n");
+        exit(0);
+    }
+
+    ck::index_t Scale_Stride_AM = (K + Scale_Block_K - 1) / Scale_Block_K;
+    ck::index_t Scale_Stride_BN = (K + Scale_Block_K - 1) / Scale_Block_K;
+
+    auto f_host_tensor_descriptor =
+        [](std::size_t row, std::size_t col, std::size_t stride, auto layout) {
+            using namespace ck::literals;
+
+            if(std::is_same<decltype(layout), ck::tensor_layout::gemm::RowMajor>::value)
+            {
+                return ck::HostTensorDescriptor({row, col}, {stride, 1_uz});
+            }
+            else
+            {
+                return ck::HostTensorDescriptor({row, col}, {1_uz, stride});
+            }
+        };
+
+    ck::Tensor<A0DataType> a0_m_k(f_host_tensor_descriptor(M, K, StrideA, A0Layout{}));
+    ck::Tensor<A1DataType> a1_m_k(f_host_tensor_descriptor((M + Scale_Block_M - 1) / Scale_Block_M,
+                                                           (K + Scale_Block_K - 1) / Scale_Block_K,
+                                                           Scale_Stride_AM,
+                                                           A0Layout{}));
+    ck::Tensor<B0DataType> b0_k_n(f_host_tensor_descriptor(K, N, StrideB, B0Layout{}));
+    ck::Tensor<B1DataType> b1_k_n(f_host_tensor_descriptor((K + Scale_Block_K - 1) / Scale_Block_K,
+                                                           (N + Scale_Block_N - 1) / Scale_Block_N,
+                                                           Scale_Stride_BN,
+                                                           B0Layout{}));
+    ck::Tensor<EDataType> e_m_n_host_result(f_host_tensor_descriptor(M, N, StrideE, ELayout{}));
+    ck::Tensor<EDataType> e_m_n_device_result(f_host_tensor_descriptor(M, N, StrideE, ELayout{}));
+
+    std::cout << "a0_m_k: " << a0_m_k.mDesc << std::endl;
+    std::cout << "a1_m_k: " << a1_m_k.mDesc << std::endl;
+    std::cout << "b0_k_n: " << b0_k_n.mDesc << std::endl;
+    std::cout << "b1_k_n: " << b1_k_n.mDesc << std::endl;
+    std::cout << "e_m_n: " << e_m_n_host_result.mDesc << std::endl;
+
+    switch(init_method)
+    {
+    case 0: break;
+    case 1:
+        a0_m_k.GenerateTensorValue(GeneratorTensor_2<A0DataType>{-2, 2});
+        b0_k_n.GenerateTensorValue(GeneratorTensor_2<B0DataType>{-2, 2});
+        a1_m_k.GenerateTensorValue(GeneratorTensor_2<A1DataType>{-1, 1});
+        b1_k_n.GenerateTensorValue(GeneratorTensor_2<B1DataType>{-1, 1});
+        break;
+    case 2:
+        a0_m_k.GenerateTensorValue(GeneratorTensor_1<A0DataType>{});
+        b0_k_n.GenerateTensorValue(GeneratorTensor_1<B0DataType>{});
+        a1_m_k.GenerateTensorValue(GeneratorTensor_1<A1DataType>{});
+        b1_k_n.GenerateTensorValue(GeneratorTensor_1<B1DataType>{});
+        break;
+    case 3:
+        a0_m_k.GenerateTensorValue(GeneratorTensor_2<A0DataType>{-2, 2});
+        b0_k_n.GenerateTensorValue(GeneratorTensor_2<B0DataType>{-2, 2});
+        a1_m_k.GenerateTensorValue(GeneratorTensor_1<A1DataType>{});
+        b1_k_n.GenerateTensorValue(GeneratorTensor_1<B1DataType>{});
+        break;
+    case 4:
+        a0_m_k.GenerateTensorValue(GeneratorTensor_2<A0DataType>{-2, 2});
+        b0_k_n.GenerateTensorValue(GeneratorTensor_2<B0DataType>{-2, 2});
+        a1_m_k.GenerateTensorValue(GeneratorTensor_3<A1DataType>{0, 1.0});
+        b1_k_n.GenerateTensorValue(GeneratorTensor_1<B1DataType>{});
+        break;
+    case 5:
+        a0_m_k.GenerateTensorValue(GeneratorTensor_2<A0DataType>{-2, 2});
+        b0_k_n.GenerateTensorValue(GeneratorTensor_2<B0DataType>{-2, 2});
+        a1_m_k.GenerateTensorValue(GeneratorTensor_1<A1DataType>{});
+        b1_k_n.GenerateTensorValue(GeneratorTensor_3<B1DataType>{0, 1.0});
+        break;
+    default:
+        a0_m_k.GenerateTensorValue(GeneratorTensor_3<A0DataType>{-0.5, 0.5});
+        b0_k_n.GenerateTensorValue(GeneratorTensor_3<B0DataType>{-0.5, 0.5});
+        a1_m_k.GenerateTensorValue(GeneratorTensor_3<A1DataType>{0, 1.0});
+        b1_k_n.GenerateTensorValue(GeneratorTensor_3<B1DataType>{0, 1.0});
+    }
+
+    ck::DeviceMem a0_device_buf(sizeof(A0DataType) * a0_m_k.mDesc.GetElementSpaceSize());
+    ck::DeviceMem a1_device_buf(sizeof(A1DataType) * a1_m_k.mDesc.GetElementSpaceSize());
+    ck::DeviceMem b0_device_buf(sizeof(B0DataType) * b0_k_n.mDesc.GetElementSpaceSize());
+    ck::DeviceMem b1_device_buf(sizeof(B1DataType) * b1_k_n.mDesc.GetElementSpaceSize());
+    ck::DeviceMem e_device_buf(sizeof(EDataType) * e_m_n_device_result.mDesc.GetElementSpaceSize());
+
+    a0_device_buf.ToDevice(a0_m_k.mData.data());
+    a1_device_buf.ToDevice(a1_m_k.mData.data());
+    b0_device_buf.ToDevice(b0_k_n.mData.data());
+    b1_device_buf.ToDevice(b1_k_n.mData.data());
+
+    auto a_element_op   = AElementOp{};
+    auto b_element_op   = BElementOp{};
+    auto cde_element_op = CDEElementOp{};
+
+    constexpr ck::index_t NumDTensor = DsDataType::Size();
+
+    // do GEMM
+    auto device_op      = DeviceOpInstance{};
+    std::string op_name = device_op.GetTypeString();
+    auto invoker        = device_op.MakeInvoker();
+    auto argument =
+        device_op.MakeArgument(static_cast<A0DataType*>(a0_device_buf.GetDeviceBuffer()),
+                               static_cast<B0DataType*>(b0_device_buf.GetDeviceBuffer()),
+                               std::array<const void*, NumDTensor>{},
+                               static_cast<EDataType*>(e_device_buf.GetDeviceBuffer()),
+                               M,
+                               N,
+                               K,
+                               StrideA,
+                               StrideB,
+                               std::array<ck::index_t, NumDTensor>{},
+                               StrideE,
+                               static_cast<const A1DataType*>(a1_device_buf.GetDeviceBuffer()),
+                               static_cast<const B1DataType*>(b1_device_buf.GetDeviceBuffer()),
+                               a_element_op,
+                               b_element_op,
+                               cde_element_op,
+                               KBatch);
+
+    if(!device_op.IsSupportedArgument(argument))
+    {
+        throw std::runtime_error(
+            "wrong! device_gemm with the specified compilation parameters does "
+            "not support this GEMM problem");
+    }
+
+    std::size_t flop = std::size_t(2) * M * N * K;
+    std::size_t num_btype =
+        sizeof(A0DataType) * M * K + sizeof(B0DataType) * K * N + sizeof(EDataType) * M * N;
+
+    float ave_time = .0;
+
+    ave_time = invoker.Run(argument, StreamConfig{nullptr, false, 0, 50, 100});
+
+    int pass = 0;
+
+    if(do_verification)
+    {
+        ck::Tensor<AccDataType> c_m_n({M, N});
+        ck::Tensor<float> a_m_k({M, K});
+        ck::Tensor<float> b_k_n({K, N});
+
+        for(int m = 0; m < M; m++)
+        {
+            for(int k = 0; k < K; k++)
+            {
+                a_m_k(m, k) = ck::type_convert<float>(a0_m_k(m, k)) *
+                              a1_m_k(m / Scale_Block_M, k / Scale_Block_K);
+            }
+        }
+
+        for(int n = 0; n < N; n++)
+        {
+            for(int k = 0; k < K; k++)
+            {
+                b_k_n(k, n) = ck::type_convert<float>(b0_k_n(k, n)) *
+                              b1_k_n(k / Scale_Block_K, n / Scale_Block_N);
+            }
+        }
+
+        using ReferenceGemmInstance = ck::tensor_operation::host::ReferenceGemm<float,
+                                                                                float,
+                                                                                CShuffleDataType,
+                                                                                AccDataType,
+                                                                                PassThrough,
+                                                                                PassThrough,
+                                                                                PassThrough>;
+        auto ref_gemm               = ReferenceGemmInstance{};
+        auto ref_invoker            = ref_gemm.MakeInvoker();
+
+        auto ref_argument =
+            ref_gemm.MakeArgument(a_m_k, b_k_n, c_m_n, PassThrough{}, PassThrough{}, PassThrough{});
+
+        ref_invoker.Run(ref_argument);
+
+        for(int m = 0; m < M; ++m)
+        {
+            for(int n = 0; n < N; ++n)
+            {
+                e_m_n_host_result(m, n) = ck::type_convert<EDataType>(c_m_n(m, n));
+            }
+        }
+
+        e_device_buf.FromDevice(e_m_n_device_result.mData.data());
+
+        pass = ck::utils::check_err(
+                   e_m_n_device_result, e_m_n_host_result, "Error: Incorrect results!", 5e-2, 5e-2)
+                   ? 0
+                   : 1;
+    }
+
+    if(flush_cache)
+    {
+        int rotating_buf = (512 * 1024 * 1024 + num_btype - 1) / num_btype;
+
+        ave_time = invoker.Run(argument,
+                               StreamConfig{nullptr, time_kernel, 0, 50, 100, true, rotating_buf});
+    }
+    else
+    {
+        ave_time = invoker.Run(argument, StreamConfig{nullptr, time_kernel, 0, 50, 100});
+    }
+
+    float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
+
+    float gb_per_sec = num_btype / 1.E6 / ave_time;
+
+    std::cout << "Perf: " << ave_time << " ms, " << tflops << " TFlops, " << gb_per_sec << " GB/s, "
+              << op_name << ", KBatch " << KBatch << std::endl;
+
+    return pass;
+}

example/65_gemm_multiply_multiply/gemm_multiply_multiply_wmma_fp8_blockscale_bpreshuffle.cpp

@@ -0,0 +1,357 @@
+// Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
+// SPDX-License-Identifier: MIT
+
+#include <iostream>
+#include <numeric>
+#include <initializer_list>
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_wmma_cshuffle_v3_blockscale_bpreshuffle.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp"
+
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/literals.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
+#include "ck/library/utility/check_err.hpp"
+
+#include "ck/utility/blkgemmpipe_scheduler.hpp"
+
+#include "common.hpp"
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using BF16 = ck::bhalf_t;
+using FP8  = ck::f8_t;
+using F32  = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+using A0DataType       = FP8;
+using A1DataType       = F32;
+using B0DataType       = FP8;
+using B1DataType       = F32;
+using AccDataType      = F32;
+using CShuffleDataType = F32;
+using DsDataType       = ck::Tuple<>;
+using EDataType        = BF16;
+
+using A0Layout = Row;
+using A1Layout = Col;
+using B0Layout = Col;
+using D0Layout = Row;
+using D1Layout = Col;
+using DsLayout = ck::Tuple<>;
+using ELayout  = Row;
+
+static constexpr int KPack = 16;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+using AElementOp   = PassThrough;
+using BElementOp   = PassThrough;
+using CDEElementOp = PassThrough;
+
+static constexpr auto GemmSpec = ck::tensor_operation::device::GemmSpecialization::Default;
+
+static constexpr ck::index_t Scale_Block_M = 1;
+static constexpr ck::index_t Scale_Block_N = 128;
+static constexpr ck::index_t Scale_Block_K = 128;
+
+using DeviceOpInstance =
+    ck::tensor_operation::device::DeviceGemmMultiD_BlockScale_Wmma_CShuffle_V3_BPreshuffle
+    // clang-format off
+         <Row, Col, DsLayout, ELayout,
+          A0DataType, A1DataType, B0DataType, B1DataType, DsDataType, EDataType, AccDataType, CShuffleDataType, 
+          AElementOp,  BElementOp, CDEElementOp, GemmSpec,
+          256, Scale_Block_M, Scale_Block_N, Scale_Block_K,
+          128, 128, 128,
+          16, 16,
+          16, 16,
+          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>,
+          ck::BlockGemmPipelineScheduler::Intrawave, ck::BlockGemmPipelineVersion::v1, FP8>;
+// clang-format on
+
+int main(int argc, char* argv[])
+{
+    bool do_verification = true;
+    int init_method      = 1;
+    bool time_kernel     = false;
+    bool flush_cache     = true;
+
+    // GEMM shape
+    ck::index_t M = 128;
+    ck::index_t N = 1024;
+    ck::index_t K = 1024;
+
+    ck::index_t StrideA = K;
+    ck::index_t StrideB = K;
+    ck::index_t StrideE = N;
+
+    ck::index_t KBatch = 1;
+
+    if(argc == 1)
+    {
+        // use default case
+    }
+    else if(argc == 4)
+    {
+        do_verification = std::stoi(argv[1]);
+        init_method     = std::stoi(argv[2]);
+        time_kernel     = std::stoi(argv[3]);
+    }
+    else if(argc == 8 || argc == 9)
+    {
+        do_verification = std::stoi(argv[1]);
+        init_method     = std::stoi(argv[2]);
+        time_kernel     = std::stoi(argv[3]);
+
+        M = std::stoi(argv[4]);
+        N = std::stoi(argv[5]);
+        K = std::stoi(argv[6]);
+
+        flush_cache = std::stoi(argv[7]);
+
+        if(argc == 9)
+        {
+            KBatch = std::stoi(argv[8]);
+        }
+
+        StrideA = K;
+        StrideB = K;
+        StrideE = N;
+    }
+    else
+    {
+        printf("arg1: verification (0=no, 1=yes)\n");
+        printf("arg2: initialization (0=no init, 1=integer value, 2=decimal value)\n");
+        printf("arg3: time kernel (0=no, 1=yes)\n");
+        printf("arg4 to 6: M, N, K\n");
+        printf("arg7: flush both I$ and L2$ (0=no, 1=yes)\n");
+        printf("arg8: KBatch (default: 1)\n");
+        exit(0);
+    }
+
+    // Transpose the AScale tensor for better performance
+    ck::index_t Scale_Stride_AK = (M + Scale_Block_M - 1) / Scale_Block_M;
+    ck::index_t Scale_Stride_BN = (K + Scale_Block_K - 1) / Scale_Block_K;
+
+    auto f_host_tensor_descriptor =
+        [](std::size_t row, std::size_t col, std::size_t stride, auto layout) {
+            using namespace ck::literals;
+
+            if(std::is_same<decltype(layout), ck::tensor_layout::gemm::RowMajor>::value)
+            {
+                return ck::HostTensorDescriptor({row, col}, {stride, 1_uz});
+            }
+            else
+            {
+                return ck::HostTensorDescriptor({row, col}, {1_uz, stride});
+            }
+        };
+
+    ck::Tensor<A0DataType> a0_m_k(f_host_tensor_descriptor(M, K, StrideA, A0Layout{}));
+    ck::Tensor<A1DataType> a1_m_k(f_host_tensor_descriptor((M + Scale_Block_M - 1) / Scale_Block_M,
+                                                           (K + Scale_Block_K - 1) / Scale_Block_K,
+                                                           Scale_Stride_AK,
+                                                           A1Layout{}));
+    ck::Tensor<B0DataType> b0_k_n(f_host_tensor_descriptor(K, N, StrideB, B0Layout{}));
+    ck::Tensor<B0DataType> b0_preshuffled(
+        f_host_tensor_descriptor(K, N, StrideB, B0Layout{})); // use laout only for size
+    ck::Tensor<B1DataType> b1_k_n(f_host_tensor_descriptor((K + Scale_Block_K - 1) / Scale_Block_K,
+                                                           (N + Scale_Block_N - 1) / Scale_Block_N,
+                                                           Scale_Stride_BN,
+                                                           B0Layout{}));
+    ck::Tensor<EDataType> e_m_n_host_result(f_host_tensor_descriptor(M, N, StrideE, ELayout{}));
+    ck::Tensor<EDataType> e_m_n_device_result(f_host_tensor_descriptor(M, N, StrideE, ELayout{}));
+
+    std::cout << "a0_m_k: " << a0_m_k.mDesc << std::endl;
+    std::cout << "a1_m_k: " << a1_m_k.mDesc << std::endl;
+    std::cout << "b0_k_n: " << b0_k_n.mDesc << std::endl;
+    std::cout << "b1_k_n: " << b1_k_n.mDesc << std::endl;
+    std::cout << "e_m_n: " << e_m_n_host_result.mDesc << std::endl;
+
+    switch(init_method)
+    {
+    case 0: break;
+    case 1:
+        a0_m_k.GenerateTensorValue(GeneratorTensor_2<A0DataType>{-2, 2});
+        b0_k_n.GenerateTensorValue(GeneratorTensor_2<B0DataType>{-2, 2});
+        a1_m_k.GenerateTensorValue(GeneratorTensor_3<A1DataType>{0, 1.0});
+        b1_k_n.GenerateTensorValue(GeneratorTensor_3<B1DataType>{0, 1.0});
+        break;
+    case 2:
+        a0_m_k.GenerateTensorValue(GeneratorTensor_1<A0DataType>{});
+        b0_k_n.GenerateTensorValue(GeneratorTensor_1<B0DataType>{});
+        a1_m_k.GenerateTensorValue(GeneratorTensor_1<A1DataType>{});
+        b1_k_n.GenerateTensorValue(GeneratorTensor_1<B1DataType>{});
+        break;
+    case 3:
+        a0_m_k.GenerateTensorValue(GeneratorTensor_2<A0DataType>{-2, 2});
+        b0_k_n.GenerateTensorValue(GeneratorTensor_2<B0DataType>{-2, 2});
+        a1_m_k.GenerateTensorValue(GeneratorTensor_1<A1DataType>{});
+        b1_k_n.GenerateTensorValue(GeneratorTensor_1<B1DataType>{});
+        break;
+    case 4:
+        a0_m_k.GenerateTensorValue(GeneratorTensor_2<A0DataType>{-2, 2});
+        b0_k_n.GenerateTensorValue(GeneratorTensor_2<B0DataType>{-2, 2});
+        a1_m_k.GenerateTensorValue(GeneratorTensor_3<A1DataType>{0, 1.0});
+        b1_k_n.GenerateTensorValue(GeneratorTensor_1<B1DataType>{});
+        break;
+    case 5:
+        a0_m_k.GenerateTensorValue(GeneratorTensor_2<A0DataType>{-2, 2});
+        b0_k_n.GenerateTensorValue(GeneratorTensor_2<B0DataType>{-2, 2});
+        a1_m_k.GenerateTensorValue(GeneratorTensor_1<A1DataType>{});
+        b1_k_n.GenerateTensorValue(GeneratorTensor_3<B1DataType>{0, 1.0});
+        break;
+    default:
+        a0_m_k.GenerateTensorValue(GeneratorTensor_3<A0DataType>{-0.5, 0.5});
+        b0_k_n.GenerateTensorValue(GeneratorTensor_3<B0DataType>{-0.5, 0.5});
+        a1_m_k.GenerateTensorValue(GeneratorTensor_3<A1DataType>{0, 1.0});
+        b1_k_n.GenerateTensorValue(GeneratorTensor_3<B1DataType>{0, 1.0});
+    }
+
+    ck::DeviceMem a0_device_buf(sizeof(A0DataType) * a0_m_k.mDesc.GetElementSpaceSize());
+    ck::DeviceMem a1_device_buf(sizeof(A1DataType) * a1_m_k.mDesc.GetElementSpaceSize());
+    ck::DeviceMem b0_device_buf(sizeof(B0DataType) * b0_k_n.mDesc.GetElementSpaceSize());
+    ck::DeviceMem b1_device_buf(sizeof(B1DataType) * b1_k_n.mDesc.GetElementSpaceSize());
+    ck::DeviceMem e_device_buf(sizeof(EDataType) * e_m_n_device_result.mDesc.GetElementSpaceSize());
+
+    a0_device_buf.ToDevice(a0_m_k.mData.data());
+    a1_device_buf.ToDevice(a1_m_k.mData.data());
+    b1_device_buf.ToDevice(b1_k_n.mData.data());
+
+    auto a_element_op   = AElementOp{};
+    auto b_element_op   = BElementOp{};
+    auto cde_element_op = CDEElementOp{};
+
+    constexpr ck::index_t NumDTensor = DsDataType::Size();
+
+    // do GEMM
+    auto device_op      = DeviceOpInstance{};
+    std::string op_name = device_op.GetTypeString();
+    int NPerWmma        = device_op.GetPreShuffleParameters();
+
+    preShuffleBuffer<KPack>(b0_k_n.mData.data(), b0_preshuffled.mData.data(), N, K, NPerWmma);
+
+    b0_device_buf.ToDevice(b0_preshuffled.mData.data());
+    auto invoker  = device_op.MakeInvoker();
+    auto argument = device_op.MakeArgument(a0_device_buf.GetDeviceBuffer(),
+                                           b0_device_buf.GetDeviceBuffer(),
+                                           std::array<const void*, NumDTensor>{},
+                                           e_device_buf.GetDeviceBuffer(),
+                                           M,
+                                           N,
+                                           K,
+                                           StrideA,
+                                           StrideB,
+                                           std::array<ck::index_t, NumDTensor>{},
+                                           StrideE,
+                                           a1_device_buf.GetDeviceBuffer(),
+                                           b1_device_buf.GetDeviceBuffer(),
+                                           a_element_op,
+                                           b_element_op,
+                                           cde_element_op,
+                                           KBatch);
+
+    if(!device_op.IsSupportedArgument(argument))
+    {
+        throw std::runtime_error(
+            "wrong! device_gemm with the specified compilation parameters does "
+            "not support this GEMM problem");
+    }
+
+    std::size_t flop = std::size_t(2) * M * N * K;
+    std::size_t num_btype =
+        sizeof(A0DataType) * M * K + sizeof(B0DataType) * K * N + sizeof(EDataType) * M * N;
+
+    float ave_time = 0.0f;
+
+    if(flush_cache)
+    {
+        int rotating_buf = (512 * 1024 * 1024 + num_btype - 1) / num_btype;
+
+        ave_time = invoker.Run(argument,
+                               StreamConfig{nullptr, time_kernel, 0, 50, 100, true, rotating_buf});
+    }
+    else
+    {
+        ave_time = invoker.Run(argument, StreamConfig{nullptr, time_kernel, 0, 50, 100});
+    }
+
+    float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
+
+    float gb_per_sec = num_btype / 1.E6 / ave_time;
+
+    std::cout << "Perf: " << ave_time << " ms, " << tflops << " TFlops, " << gb_per_sec << " GB/s, "
+              << op_name << ", KBatch " << KBatch << std::endl;
+
+    if(do_verification)
+    {
+        ck::Tensor<AccDataType> c_m_n({M, N});
+        ck::Tensor<float> a_m_k({M, K});
+        ck::Tensor<float> b_k_n({K, N});
+
+        for(int m = 0; m < M; m++)
+        {
+            for(int k = 0; k < K; k++)
+            {
+                a_m_k(m, k) = ck::type_convert<float>(a0_m_k(m, k)) *
+                              a1_m_k(m / Scale_Block_M, k / Scale_Block_K);
+            }
+        }
+
+        for(int n = 0; n < N; n++)
+        {
+            for(int k = 0; k < K; k++)
+            {
+                b_k_n(k, n) = ck::type_convert<float>(b0_k_n(k, n)) *
+                              b1_k_n(k / Scale_Block_K, n / Scale_Block_N);
+            }
+        }
+
+        using ReferenceGemmInstance = ck::tensor_operation::host::ReferenceGemm<float,
+                                                                                float,
+                                                                                CShuffleDataType,
+                                                                                AccDataType,
+                                                                                PassThrough,
+                                                                                PassThrough,
+                                                                                PassThrough>;
+        auto ref_gemm               = ReferenceGemmInstance{};
+        auto ref_invoker            = ref_gemm.MakeInvoker();
+
+        auto ref_argument =
+            ref_gemm.MakeArgument(a_m_k, b_k_n, c_m_n, PassThrough{}, PassThrough{}, PassThrough{});
+
+        ref_invoker.Run(ref_argument);
+
+        for(int m = 0; m < M; ++m)
+        {
+            for(int n = 0; n < N; ++n)
+            {
+                e_m_n_host_result(m, n) = ck::type_convert<EDataType>(c_m_n(m, n));
+            }
+        }
+
+        e_device_buf.FromDevice(e_m_n_device_result.mData.data());
+
+        return ck::utils::check_err(
+                   e_m_n_device_result, e_m_n_host_result, "Error: Incorrect results!", 5e-2, 5e-2)
+                   ? 0
+                   : 1;
+    }
+
+    return 0;
+}

experimental/builder/include/ck_tile/builder/README.md

@@ -4,14 +4,16 @@ This directory contains the builder framework for Composable Kernel, which provi
 
 ## Table of Contents
 
-- [Convolution Signature Design](#convolution-signature-design)
+- [Convolution Signature](#convolution-signature)
   - [Overview](#overview)
   - [Architecture](#architecture)
   - [Core Components](#core-components)
   - [Concepts and Validation](#concepts-and-validation)
+- [Convolution Algorithm](#convolution-algorithm)
+- [Convolution Factory](#convolution-factory)
 ---
 
-## Convolution Signature Design
+## Convolution Signature
 
 ### Overview
 
@@ -220,25 +222,9 @@ Several fields in the signature are optional:
 
 This design follows the principle of "make the common case simple, the complex case possible."
 
-#### Union-Based Layout Representation
+## Convolution Algorithm
 
-The `ConvLayout` type uses unions to support dimension-agnostic code:
+## Convolution Factory
 
-```cpp
-struct ConvLayout {
-    union {
-        ConvInputLayout _input_layout;
-        ConvWeightLayout _weight_layout;
-        ConvOutputLayout _output_layout;
-        ConvAuxiliaryTensorLayout _aux_tensor_layout;
-    };
-    // ... constructors for each type
-};
-```
-
-This allows:
-- Single type to represent all layout variants
-- Type-safe construction through overloaded constructors
-- Compile-time enforcement of valid combinations through concepts
-
----
+Convolution factory builds the instance based on the convolution signature and convolution algorithm.
+The signature and the algorithm descriptions are dispatched to the relevant algorithm specific factory for instance creation. The convolution factory design is described in a separate [Readme](factory/README.md).

experimental/builder/include/ck_tile/builder/factory/helpers/ck/conv_tensor_type.hpp

@@ -65,17 +65,19 @@ consteval auto GetTensorDataAndComputeTypes()
     constexpr auto data_type    = Config.data_type;
     constexpr auto compute_type = Config.compute_type;
 
-    if constexpr(data_type == DataType::UNDEFINDED && compute_type == DataType::UNDEFINDED)
+    using enum DataType;
+
+    if constexpr(data_type == UNDEFINED_DATA_TYPE && compute_type == UNDEFINED_DATA_TYPE)
     {
         return std::make_pair(ConvertDataTypeToCK<SignatureDataType>(),
                               ConvertDataTypeToCK<SignatureDataType>());
     }
-    else if constexpr(data_type == DataType::UNDEFINDED)
+    else if constexpr(data_type == UNDEFINED_DATA_TYPE)
     {
         return std::make_pair(ConvertDataTypeToCK<SignatureDataType>(),
                               ConvertDataTypeToCK<compute_type>());
     }
-    else if constexpr(compute_type == DataType::UNDEFINDED)
+    else if constexpr(compute_type == UNDEFINED_DATA_TYPE)
     {
         return std::make_pair(ConvertDataTypeToCK<data_type>(),
                               ConvertDataTypeToCK<SignatureDataType>());
@@ -91,7 +93,7 @@ template <DataType SignatureAccDataType, DataType SignatureDataType>
 consteval auto GetTensorAccumulationType()
 {
     constexpr auto data_type = SignatureAccDataType;
-    if constexpr(data_type == DataType::UNDEFINDED)
+    if constexpr(data_type == DataType::UNDEFINED_DATA_TYPE)
     {
         return ConvertDataTypeToCK<SignatureDataType>();
     }
@@ -105,7 +107,7 @@ template <auto Config, DataType SignatureDataType>
 consteval auto GetAuxiliaryTensorDataTypeValue()
 {
     constexpr auto data_type = Config.data_type;
-    if constexpr(data_type == DataType::UNDEFINDED)
+    if constexpr(data_type == DataType::UNDEFINED_DATA_TYPE)
     {
         return ConvertDataTypeToCK<SignatureDataType>();
     }

experimental/builder/include/ck_tile/builder/reflect/instance_traits_device_grouped_conv_fwd_multiple_abd_xdl_cshuffle.hpp

@@ -316,7 +316,7 @@ struct InstanceTraits<ck::tensor_operation::device::DeviceGroupedConvFwdMultiple
         oss << "," << kABlockTransferSrcScalarPerVector; // 32. ABlockTransferSrcScalarPerVector
         oss << ","
             << kABlockTransferDstScalarPerVectorK1; // 33. ABlockTransferDstScalarPerVector_AK1
-        oss << "," << kABlockLdsExtraM;             // 34. ABlockLdsExtraM
+        oss << "," << (kABlockLdsExtraM ? "true" : "false"); // 34. ABlockLdsExtraM
         oss << ","
             << detail::array_to_string(
                    kBThreadClusterLengths); // 35. BBlockTransferThreadClusterLengths
@@ -329,10 +329,10 @@ struct InstanceTraits<ck::tensor_operation::device::DeviceGroupedConvFwdMultiple
         oss << "," << kBBlockTransferSrcVectorDim;       // 38. BBlockTransferSrcVectorDim
         oss << "," << kBBlockTransferSrcScalarPerVector; // 39. BBlockTransferSrcScalarPerVector
         oss << ","
-            << kBBlockTransferDstScalarPerVectorK1;   // 40. BBlockTransferDstScalarPerVector_BK1
-        oss << "," << kBBlockLdsExtraN;               // 41. BBlockLdsExtraN
-        oss << "," << kCShuffleMXdlPerWavePerShuffle; // 42. CShuffleMXdlPerWavePerShuffle
-        oss << "," << kCShuffleNXdlPerWavePerShuffle; // 43. CShuffleNXdlPerWavePerShuffle
+            << kBBlockTransferDstScalarPerVectorK1; // 40. BBlockTransferDstScalarPerVector_BK1
+        oss << "," << (kBBlockLdsExtraN ? "true" : "false"); // 41. BBlockLdsExtraN
+        oss << "," << kCShuffleMXdlPerWavePerShuffle;        // 42. CShuffleMXdlPerWavePerShuffle
+        oss << "," << kCShuffleNXdlPerWavePerShuffle;        // 43. CShuffleNXdlPerWavePerShuffle
         oss << ","
             << detail::array_to_string(
                    kCThreadClusterLengths); // 44. CDEBlockTransferClusterLengths

experimental/builder/include/ck_tile/builder/reflect/instance_traits_device_grouped_conv_fwd_multiple_abd_xdl_cshuffle_v3.hpp

@@ -316,7 +316,7 @@ struct InstanceTraits<ck::tensor_operation::device::DeviceGroupedConvFwdMultiple
         oss << "," << kABlockTransferSrcScalarPerVector; // 31. ABlockTransferSrcScalarPerVector
         oss << ","
             << kABlockTransferDstScalarPerVectorK1; // 32. ABlockTransferDstScalarPerVector_AK1
-        oss << "," << kABlockLdsExtraM;             // 33. ABlockLdsExtraM
+        oss << "," << (kABlockLdsExtraM ? "true" : "false"); // 33. ABlockLdsExtraM
         oss << ","
             << detail::array_to_string(
                    kBThreadClusterLengths); // 34. BBlockTransferThreadClusterLengths
@@ -329,10 +329,10 @@ struct InstanceTraits<ck::tensor_operation::device::DeviceGroupedConvFwdMultiple
         oss << "," << kBBlockTransferSrcVectorDim;       // 37. BBlockTransferSrcVectorDim
         oss << "," << kBBlockTransferSrcScalarPerVector; // 38. BBlockTransferSrcScalarPerVector
         oss << ","
-            << kBBlockTransferDstScalarPerVectorK1;   // 39. BBlockTransferDstScalarPerVector_BK1
-        oss << "," << kBBlockLdsExtraN;               // 40. BBlockLdsExtraN
-        oss << "," << kCShuffleMXdlPerWavePerShuffle; // 41. CShuffleMXdlPerWavePerShuffle
-        oss << "," << kCShuffleNXdlPerWavePerShuffle; // 42. CShuffleNXdlPerWavePerShuffle
+            << kBBlockTransferDstScalarPerVectorK1; // 39. BBlockTransferDstScalarPerVector_BK1
+        oss << "," << (kBBlockLdsExtraN ? "true" : "false"); // 40. BBlockLdsExtraN
+        oss << "," << kCShuffleMXdlPerWavePerShuffle;        // 41. CShuffleMXdlPerWavePerShuffle
+        oss << "," << kCShuffleNXdlPerWavePerShuffle;        // 42. CShuffleNXdlPerWavePerShuffle
         oss << ","
             << detail::array_to_string(
                    kCThreadClusterLengths); // 43. CDEBlockTransferClusterLengths

experimental/builder/include/ck_tile/builder/reflect/instance_traits_device_grouped_conv_fwd_multiple_d_xdl_large_tensor_cshuffle.hpp

@@ -311,7 +311,7 @@ struct InstanceTraits<
         oss << "," << kABlockTransferSrcScalarPerVector; // 32. ABlockTransferSrcScalarPerVector
         oss << ","
             << kABlockTransferDstScalarPerVectorK1; // 33. ABlockTransferDstScalarPerVector_AK1
-        oss << "," << kABlockLdsExtraM;             // 34. ABlockLdsExtraM
+        oss << "," << (kABlockLdsExtraM ? "true" : "false"); // 34. ABlockLdsExtraM
         oss << ","
             << detail::array_to_string(
                    kBThreadClusterLengths); // 35. BBlockTransferThreadClusterLengths
@@ -324,10 +324,10 @@ struct InstanceTraits<
         oss << "," << kBBlockTransferSrcVectorDim;       // 38. BBlockTransferSrcVectorDim
         oss << "," << kBBlockTransferSrcScalarPerVector; // 39. BBlockTransferSrcScalarPerVector
         oss << ","
-            << kBBlockTransferDstScalarPerVectorK1;   // 40. BBlockTransferDstScalarPerVector_BK1
-        oss << "," << kBBlockLdsExtraN;               // 41. BBlockLdsExtraN
-        oss << "," << kCShuffleMXdlPerWavePerShuffle; // 42. CShuffleMXdlPerWavePerShuffle
-        oss << "," << kCShuffleNXdlPerWavePerShuffle; // 43. CShuffleNXdlPerWavePerShuffle
+            << kBBlockTransferDstScalarPerVectorK1; // 40. BBlockTransferDstScalarPerVector_BK1
+        oss << "," << (kBBlockLdsExtraN ? "true" : "false"); // 41. BBlockLdsExtraN
+        oss << "," << kCShuffleMXdlPerWavePerShuffle;        // 42. CShuffleMXdlPerWavePerShuffle
+        oss << "," << kCShuffleNXdlPerWavePerShuffle;        // 43. CShuffleNXdlPerWavePerShuffle
         oss << ","
             << detail::array_to_string(
                    kCThreadClusterLengths); // 44. CDEBlockTransferClusterLengths

experimental/builder/include/ck_tile/builder/types.hpp

@@ -13,7 +13,7 @@ namespace ck_tile::builder {
 
 enum class DataType
 {
-    UNDEFINDED = 0,
+    UNDEFINED_DATA_TYPE = 0,
     FP32,
     FP16,
     BF16,
@@ -25,7 +25,7 @@ enum class DataType
 
 enum class TensorLayout
 {
-    UNDEFINED,
+    UNDEFINED_TENSOR_LAYOUT = 0,
 
     // Bias tensors
     GC,
@@ -212,219 +212,269 @@ enum class ConvAlgorithmSpecialization
     LARGE_TENSOR
 };
 
-// ostream operator overloads for enum classes
-inline std::ostream& operator<<(std::ostream& os, DataType dt)
+// toString methods for enum classes
+inline std::string_view toString(DataType dt)
 {
     using enum DataType;
     switch(dt)
     {
-    case FP16: return os << "FP16";
-    case FP32: return os << "FP32";
-    case BF16: return os << "BF16";
-    case FP8: return os << "FP8";
-    case INT32: return os << "INT32";
-    case I8: return os << "I8";
-    case U8: return os << "U8";
-    case UNDEFINDED: return os << "UNDEFINDED";
-    default: return os << "Unknown";
+    case FP16: return "FP16";
+    case FP32: return "FP32";
+    case BF16: return "BF16";
+    case FP8: return "FP8";
+    case INT32: return "INT32";
+    case I8: return "I8";
+    case U8: return "U8";
+    case UNDEFINED_DATA_TYPE: return "UNDEFINED_DATA_TYPE";
+    default: return "Unknown";
     }
 }
 
-inline std::ostream& operator<<(std::ostream& os, ConvDirection dir)
+inline std::string_view toString(ConvDirection dir)
 {
     using enum ConvDirection;
     switch(dir)
     {
-    case FORWARD: return os << "Forward";
-    case BACKWARD_DATA: return os << "Backward Data";
-    case BACKWARD_WEIGHT: return os << "Backward Weight";
-    default: return os << "Unknown";
+    case FORWARD: return "Forward";
+    case BACKWARD_DATA: return "Backward Data";
+    case BACKWARD_WEIGHT: return "Backward Weight";
+    default: return "Unknown";
     }
 }
 
-inline std::ostream& operator<<(std::ostream& os, ElementwiseOperation op)
+inline std::string_view toString(ElementwiseOperation op)
 {
     using enum ElementwiseOperation;
     switch(op)
     {
-    case CLAMP: return os << "CLAMP";
-    case SCALE: return os << "SCALE";
-    case PASS_THROUGH: return os << "PASS_THROUGH";
-    case BIAS_BNORM_CLAMP: return os << "BIAS_BNORM_CLAMP";
-    case SCALEADD_SCALEADD_RELU: return os << "SCALEADD_SCALEADD_RELU";
-    default: return os << "Unknown";
+    case CLAMP: return "CLAMP";
+    case SCALE: return "SCALE";
+    case PASS_THROUGH: return "PASS_THROUGH";
+    case BIAS_BNORM_CLAMP: return "BIAS_BNORM_CLAMP";
+    case SCALEADD_SCALEADD_RELU: return "SCALEADD_SCALEADD_RELU";
+    default: return "Unknown";
     }
 }
 
-inline std::ostream& operator<<(std::ostream& os, PipelineVersion ver)
+inline std::string_view toString(PipelineVersion ver)
 {
     using enum PipelineVersion;
     switch(ver)
     {
-    case V1: return os << "V1";
-    case V2: return os << "V2";
-    case V3: return os << "V3";
-    case V4: return os << "V4";
-    case V5: return os << "V5";
-    case WEIGHT_ONLY: return os << "WEIGHT_ONLY";
-    default: return os << "Unknown";
+    case V1: return "V1";
+    case V2: return "V2";
+    case V3: return "V3";
+    case V4: return "V4";
+    case V5: return "V5";
+    case WEIGHT_ONLY: return "WEIGHT_ONLY";
+    default: return "Unknown";
     }
 }
 
-inline std::ostream& operator<<(std::ostream& os, GemmSpecialization spec)
+inline std::string_view toString(GemmSpecialization spec)
 {
     using enum GemmSpecialization;
     switch(spec)
     {
-    case Default: return os << "Default";
-    case MPadding: return os << "MPadding";
-    case NPadding: return os << "NPadding";
-    case KPadding: return os << "KPadding";
-    case MNPadding: return os << "MNPadding";
-    case MKPadding: return os << "MKPadding";
-    case NKPadding: return os << "NKPadding";
-    case MNKPadding: return os << "MNKPadding";
-    case OPadding: return os << "OPadding";
-    case MOPadding: return os << "MOPadding";
-    case NOPadding: return os << "NOPadding";
-    case KOPadding: return os << "KOPadding";
-    case MNOPadding: return os << "MNOPadding";
-    case MKOPadding: return os << "MKOPadding";
-    case NKOPadding: return os << "NKOPadding";
-    case MNKOPadding: return os << "MNKOPadding";
-    default: return os << "Unknown";
+    case Default: return "Default";
+    case MPadding: return "MPadding";
+    case NPadding: return "NPadding";
+    case KPadding: return "KPadding";
+    case MNPadding: return "MNPadding";
+    case MKPadding: return "MKPadding";
+    case NKPadding: return "NKPadding";
+    case MNKPadding: return "MNKPadding";
+    case OPadding: return "OPadding";
+    case MOPadding: return "MOPadding";
+    case NOPadding: return "NOPadding";
+    case KOPadding: return "KOPadding";
+    case MNOPadding: return "MNOPadding";
+    case MKOPadding: return "MKOPadding";
+    case NKOPadding: return "NKOPadding";
+    case MNKOPadding: return "MNKOPadding";
+    default: return "Unknown";
     }
 }
 
-inline std::ostream& operator<<(std::ostream& os, ConvFwdSpecialization spec)
+inline std::string_view toString(ConvFwdSpecialization spec)
 {
     using enum ConvFwdSpecialization;
     switch(spec)
     {
-    case DEFAULT: return os << "DEFAULT";
-    case FILTER_1X1_PAD0: return os << "FILTER_1X1_PAD0";
-    case FILTER_1X1_STRIDE1_PAD0: return os << "FILTER_1X1_STRIDE1_PAD0";
-    case FILTER_3x3: return os << "FILTER_3x3";
-    default: return os << "Unknown";
+    case DEFAULT: return "DEFAULT";
+    case FILTER_1X1_PAD0: return "FILTER_1X1_PAD0";
+    case FILTER_1X1_STRIDE1_PAD0: return "FILTER_1X1_STRIDE1_PAD0";
+    case FILTER_3x3: return "FILTER_3x3";
+    default: return "Unknown";
     }
 }
 
-inline std::ostream& operator<<(std::ostream& os, ConvBwdDataSpecialization spec)
+inline std::string_view toString(ConvBwdDataSpecialization spec)
 {
     using enum ConvBwdDataSpecialization;
     switch(spec)
     {
-    case DEFAULT: return os << "DEFAULT";
-    case FILTER_1X1_STRIDE1_PAD0: return os << "FILTER_1X1_STRIDE1_PAD0";
-    default: return os << "Unknown";
+    case DEFAULT: return "DEFAULT";
+    case FILTER_1X1_STRIDE1_PAD0: return "FILTER_1X1_STRIDE1_PAD0";
+    default: return "Unknown";
     }
 }
 
-inline std::ostream& operator<<(std::ostream& os, ConvBwdWeightSpecialization spec)
+inline std::string_view toString(ConvBwdWeightSpecialization spec)
 {
     using enum ConvBwdWeightSpecialization;
     switch(spec)
     {
-    case DEFAULT: return os << "DEFAULT";
-    case FILTER_1X1_STRIDE1_PAD0: return os << "FILTER_1X1_STRIDE1_PAD0";
-    case FILTER_1X1_PAD0: return os << "FILTER_1X1_PAD0";
-    case ODD_C: return os << "ODD_C";
-    default: return os << "Unknown";
+    case DEFAULT: return "DEFAULT";
+    case FILTER_1X1_STRIDE1_PAD0: return "FILTER_1X1_STRIDE1_PAD0";
+    case FILTER_1X1_PAD0: return "FILTER_1X1_PAD0";
+    case ODD_C: return "ODD_C";
+    default: return "Unknown";
     }
 }
 
-inline std::ostream& operator<<(std::ostream& os, GemmPadding padding)
+inline std::string_view toString(GemmPadding padding)
 {
     using enum GemmPadding;
     switch(padding)
     {
-    case DEFAULT: return os << "DEFAULT";
-    case M_PADDING: return os << "M_PADDING";
-    case N_PADDING: return os << "N_PADDING";
-    case K_PADDING: return os << "K_PADDING";
-    case MN_PADDING: return os << "MN_PADDING";
-    case MK_PADDING: return os << "MK_PADDING";
-    case NK_PADDING: return os << "NK_PADDING";
-    case MNK_PADDING: return os << "MNK_PADDING";
-    case O_PADDING: return os << "O_PADDING";
-    case MO_PADDING: return os << "MO_PADDING";
-    case NO_PADDING: return os << "NO_PADDING";
-    case KO_PADDING: return os << "KO_PADDING";
-    case MNO_PADDING: return os << "MNO_PADDING";
-    case MKO_PADDING: return os << "MKO_PADDING";
-    case NKO_PADDING: return os << "NKO_PADDING";
-    case MNKO_PADDING: return os << "MNKO_PADDING";
-    default: return os << "Unknown";
+    case DEFAULT: return "DEFAULT";
+    case M_PADDING: return "M_PADDING";
+    case N_PADDING: return "N_PADDING";
+    case K_PADDING: return "K_PADDING";
+    case MN_PADDING: return "MN_PADDING";
+    case MK_PADDING: return "MK_PADDING";
+    case NK_PADDING: return "NK_PADDING";
+    case MNK_PADDING: return "MNK_PADDING";
+    case O_PADDING: return "O_PADDING";
+    case MO_PADDING: return "MO_PADDING";
+    case NO_PADDING: return "NO_PADDING";
+    case KO_PADDING: return "KO_PADDING";
+    case MNO_PADDING: return "MNO_PADDING";
+    case MKO_PADDING: return "MKO_PADDING";
+    case NKO_PADDING: return "NKO_PADDING";
+    case MNKO_PADDING: return "MNKO_PADDING";
+    default: return "Unknown";
     }
 }
 
-inline std::ostream& operator<<(std::ostream& os, PipelineScheduler sched)
+inline std::string_view toString(PipelineScheduler sched)
 {
     using enum PipelineScheduler;
     switch(sched)
     {
-    case DEFAULT: return os << "DEFAULT";
-    case INTRAWAVE: return os << "INTRAWAVE";
-    case INTERWAVE: return os << "INTERWAVE";
-    default: return os << "Unknown";
+    case DEFAULT: return "DEFAULT";
+    case INTRAWAVE: return "INTRAWAVE";
+    case INTERWAVE: return "INTERWAVE";
+    default: return "Unknown";
     }
 }
 
-inline std::ostream& operator<<(std::ostream& os, TensorLayout layout)
+inline std::string_view toString(TensorLayout layout)
 {
     using enum TensorLayout;
     switch(layout)
     {
-    case GNCW: return os << "GNCW";
-    case GNWC: return os << "GNWC";
-    case NWGC: return os << "NWGC";
-    case NGCW: return os << "NGCW";
-    case G_NW_C_strided: return os << "G_NW_C_strided";
-    case GNCHW: return os << "GNCHW";
-    case GNHWC: return os << "GNHWC";
-    case NHWGC: return os << "NHWGC";
-    case NGCHW: return os << "NGCHW";
-    case G_NHW_C_strided: return os << "G_NHW_C_strided";
-    case GNCDHW: return os << "GNCDHW";
-    case GNDHWC: return os << "GNDHWC";
-    case NDHWGC: return os << "NDHWGC";
-    case NGCDHW: return os << "NGCDHW";
-    case G_NDHW_C_strided: return os << "G_NDHW_C_strided";
-    case GKXC: return os << "GKXC";
-    case GKCX: return os << "GKCX";
-    case KXGC: return os << "KXGC";
-    case G_K_X_C_strided: return os << "G_K_X_C_strided";
-    case GKYXC: return os << "GKYXC";
-    case GKCYX: return os << "GKCYX";
-    case KYXGC: return os << "KYXGC";
-    case G_K_YX_C_strided: return os << "G_K_YX_C_strided";
-    case GKZYXC: return os << "GKZYXC";
-    case GKCZYX: return os << "GKCZYX";
-    case KZYXGC: return os << "KZYXGC";
-    case G_K_ZYX_C_strided: return os << "G_K_ZYX_C_strided";
-    case GNKW: return os << "GNKW";
-    case GNWK: return os << "GNWK";
-    case NWGK: return os << "NWGK";
-    case NGKW: return os << "NGKW";
-    case G_NW_K_strided: return os << "G_NW_K_strided";
-    case GNKHW: return os << "GNKHW";
-    case GNHWK: return os << "GNHWK";
-    case NHWGK: return os << "NHWGK";
-    case NGKHW: return os << "NGKHW";
-    case G_NHW_K_strided: return os << "G_NHW_K_strided";
-    case GNKDHW: return os << "GNKDHW";
-    case GNDHWK: return os << "GNDHWK";
-    case NDHWGK: return os << "NDHWGK";
-    case NGKDHW: return os << "NGKDHW";
-    case G_NDHW_K_strided: return os << "G_NDHW_K_strided";
-    case GC: return os << "GC";
-    case G_C_strided: return os << "G_C_strided";
-    case G_K_strided: return os << "G_K_strided";
-    case UNDEFINED: return os << "UNDEFINED";
-    default: return os << "Unknown";
+    case GNCW: return "GNCW";
+    case GNWC: return "GNWC";
+    case NWGC: return "NWGC";
+    case NGCW: return "NGCW";
+    case G_NW_C_strided: return "G_NW_C_strided";
+    case GNCHW: return "GNCHW";
+    case GNHWC: return "GNHWC";
+    case NHWGC: return "NHWGC";
+    case NGCHW: return "NGCHW";
+    case G_NHW_C_strided: return "G_NHW_C_strided";
+    case GNCDHW: return "GNCDHW";
+    case GNDHWC: return "GNDHWC";
+    case NDHWGC: return "NDHWGC";
+    case NGCDHW: return "NGCDHW";
+    case G_NDHW_C_strided: return "G_NDHW_C_strided";
+    case GKXC: return "GKXC";
+    case GKCX: return "GKCX";
+    case KXGC: return "KXGC";
+    case G_K_X_C_strided: return "G_K_X_C_strided";
+    case GKYXC: return "GKYXC";
+    case GKCYX: return "GKCYX";
+    case KYXGC: return "KYXGC";
+    case G_K_YX_C_strided: return "G_K_YX_C_strided";
+    case GKZYXC: return "GKZYXC";
+    case GKCZYX: return "GKCZYX";
+    case KZYXGC: return "KZYXGC";
+    case G_K_ZYX_C_strided: return "G_K_ZYX_C_strided";
+    case GNKW: return "GNKW";
+    case GNWK: return "GNWK";
+    case NWGK: return "NWGK";
+    case NGKW: return "NGKW";
+    case G_NW_K_strided: return "G_NW_K_strided";
+    case GNKHW: return "GNKHW";
+    case GNHWK: return "GNHWK";
+    case NHWGK: return "NHWGK";
+    case NGKHW: return "NGKHW";
+    case G_NHW_K_strided: return "G_NHW_K_strided";
+    case GNKDHW: return "GNKDHW";
+    case GNDHWK: return "GNDHWK";
+    case NDHWGK: return "NDHWGK";
+    case NGKDHW: return "NGKDHW";
+    case G_NDHW_K_strided: return "G_NDHW_K_strided";
+    case GC: return "GC";
+    case G_C_strided: return "G_C_strided";
+    case G_K_strided: return "G_K_strided";
+    case UNDEFINED_TENSOR_LAYOUT: return "UNDEFINED_TENSOR_LAYOUT";
+    default: return "Unknown";
     }
 }
 
+// ostream operator overloads for enum classes
+inline std::ostream& operator<<(std::ostream& os, DataType dt) { return os << toString(dt); }
+
+inline std::ostream& operator<<(std::ostream& os, ConvDirection dir) { return os << toString(dir); }
+
+inline std::ostream& operator<<(std::ostream& os, ElementwiseOperation op)
+{
+    return os << toString(op);
+}
+
+inline std::ostream& operator<<(std::ostream& os, PipelineVersion ver)
+{
+    return os << toString(ver);
+}
+
+inline std::ostream& operator<<(std::ostream& os, GemmSpecialization spec)
+{
+    return os << toString(spec);
+}
+
+inline std::ostream& operator<<(std::ostream& os, ConvFwdSpecialization spec)
+{
+    return os << toString(spec);
+}
+
+inline std::ostream& operator<<(std::ostream& os, ConvBwdDataSpecialization spec)
+{
+    return os << toString(spec);
+}
+
+inline std::ostream& operator<<(std::ostream& os, ConvBwdWeightSpecialization spec)
+{
+    return os << toString(spec);
+}
+
+inline std::ostream& operator<<(std::ostream& os, GemmPadding padding)
+{
+    return os << toString(padding);
+}
+
+inline std::ostream& operator<<(std::ostream& os, PipelineScheduler sched)
+{
+    return os << toString(sched);
+}
+
+inline std::ostream& operator<<(std::ostream& os, TensorLayout layout)
+{
+    return os << toString(layout);
+}
+
 // ostream operator overload for std::variant of convolution specializations
 inline std::ostream& operator<<(std::ostream& os,
                                 const std::variant<ConvFwdSpecialization,

experimental/builder/test/conv/ck/test_ckb_conv_fwd_1d_bf16.cpp

@@ -3,6 +3,7 @@
 
 #include "utils/ckb_conv_test_configs.hpp"
 #include "utils/ckb_conv_test_utils.hpp"
+#include "utils/conv_algorithm_type_utils.hpp"
 
 namespace {
 
@@ -13,15 +14,19 @@ using namespace ck_tile::builder::test_utils;
 TEST(FwdConvInstances,
      Create_DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3_Instance_1D_BF16_ChannelsFirst_scale)
 {
+    using enum ck_tile::builder::ConvDirection;
+    using enum ck_tile::builder::DataType;
+    using enum ck_tile::builder::TensorLayout;
+    using enum ck_tile::builder::ElementwiseOperation;
+
     constexpr ConvSignature FwdConvSignature{
         .spatial_dim            = 1,
-        .direction              = ConvDirection::FORWARD,
-        .data_type              = DataType::BF16,
-        .accumulation_data_type = DataType::FP32,
-        .input                  = {.config = {.layout = TensorLayout::NGCW}},
-        .weight                 = {.config = {.layout = TensorLayout::GKXC}},
-        .output                 = {.config    = {.layout = TensorLayout::NGKW},
-                                   .operation = {.elementwise_operation = ElementwiseOperation::SCALE}}};
+        .direction              = FORWARD,
+        .data_type              = BF16,
+        .accumulation_data_type = FP32,
+        .input                  = {.config = {.layout = NGCW}},
+        .weight                 = {.config = {.layout = GKXC}},
+        .output = {.config = {.layout = NGKW}, .operation = {.elementwise_operation = SCALE}}};
 
     constexpr auto FwdConvAlgorithm =
         ConvAlgorithm_DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3{}
@@ -33,8 +38,10 @@ TEST(FwdConvInstances,
             .with_block_gemm(BlockGemmDesc_v2_intrawave);
 
     using Builder = ConvBuilder<FwdConvSignature, FwdConvAlgorithm>;
+
+    const auto expected_transfer_parameters = to_string(FwdConvAlgorithm);
     run_test<Builder>({"DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3",
-                       "256,256,256,32",
+                       expected_transfer_parameters,
                        "NGCW,GKXC,EmptyTuple,NGKW",
                        "PassThrough,PassThrough,Scale",
                        "Filter1x1Stride1Pad0",

experimental/builder/test/conv/ck/test_ckb_conv_fwd_1d_fp16.cpp

@@ -3,6 +3,7 @@
 
 #include "utils/ckb_conv_test_configs.hpp"
 #include "utils/ckb_conv_test_utils.hpp"
+#include "utils/conv_algorithm_type_utils.hpp"
 
 namespace {
 
@@ -12,13 +13,17 @@ using namespace ck_tile::builder::test_utils;
 TEST(FwdConvInstances,
      Create_DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_Instance_1D_FP16_ChannelsFirst)
 {
+    using enum ck_tile::builder::ConvDirection;
+    using enum ck_tile::builder::DataType;
+    using enum ck_tile::builder::TensorLayout;
+
     constexpr ConvSignature FwdConvSignature{.spatial_dim            = 1,
-                                             .direction              = ConvDirection::FORWARD,
-                                             .data_type              = DataType::FP16,
-                                             .accumulation_data_type = DataType::FP32,
-                                             .input  = {.config = {.layout = TensorLayout::NWGC}},
-                                             .weight = {.config = {.layout = TensorLayout::GKXC}},
-                                             .output = {.config = {.layout = TensorLayout::NWGK}}};
+                                             .direction              = FORWARD,
+                                             .data_type              = FP16,
+                                             .accumulation_data_type = FP32,
+                                             .input                  = {.config = {.layout = NWGC}},
+                                             .weight                 = {.config = {.layout = GKXC}},
+                                             .output = {.config = {.layout = NWGK}}};
 
     constexpr auto FwdConvAlgorithm =
         ConvAlgorithm_DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle{}
@@ -29,11 +34,13 @@ TEST(FwdConvInstances,
             .with_prefetch_config(1, 2, PipelineScheduler::DEFAULT);
 
     using Builder = ConvBuilder<FwdConvSignature, FwdConvAlgorithm>;
+
+    const auto expected_transfer_parameters = to_string(FwdConvAlgorithm);
     run_test<Builder>({"DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle",
+                       expected_transfer_parameters,
                        "NWGC,GKXC,EmptyTuple,NWGK",
                        "PassThrough,PassThrough,PassThrough",
                        "MNKPadding",
-                       "64,64,32,32",
                        "Default"});
 }
 

experimental/builder/test/conv/ck/test_ckb_conv_fwd_1d_i8.cpp

@@ -3,6 +3,7 @@
 
 #include "utils/ckb_conv_test_configs.hpp"
 #include "utils/ckb_conv_test_utils.hpp"
+#include "utils/conv_algorithm_type_utils.hpp"
 
 namespace {
 
@@ -14,13 +15,17 @@ using namespace ck_tile::builder::test_utils;
 TEST(FwdConvInstances,
      Create_DeviceGroupedConvFwdMultipleD_Wmma_CShuffle_Instance_1D_FP32_ChannelsFirst_scale)
 {
+    using enum ck_tile::builder::ConvDirection;
+    using enum ck_tile::builder::DataType;
+    using enum ck_tile::builder::TensorLayout;
+
     constexpr ConvSignature FwdConvSignature{.spatial_dim            = 1,
-                                             .direction              = ConvDirection::FORWARD,
-                                             .data_type              = DataType::I8,
-                                             .accumulation_data_type = DataType::INT32,
-                                             .input  = {.config = {.layout = TensorLayout::GNWC}},
-                                             .weight = {.config = {.layout = TensorLayout::GKXC}},
-                                             .output = {.config = {.layout = TensorLayout::GNWK}}};
+                                             .direction              = FORWARD,
+                                             .data_type              = I8,
+                                             .accumulation_data_type = INT32,
+                                             .input                  = {.config = {.layout = GNWC}},
+                                             .weight                 = {.config = {.layout = GKXC}},
+                                             .output = {.config = {.layout = GNWK}}};
 
     constexpr auto FwdConvAlgorithm =
         ConvAlgorithm_DeviceGroupedConvFwdMultipleD_Wmma_CShuffle{}
@@ -31,8 +36,10 @@ TEST(FwdConvInstances,
             .with_prefetch_config(1, 0, PipelineScheduler::DEFAULT);
 
     using Builder = ConvBuilder<FwdConvSignature, FwdConvAlgorithm>;
+
+    const auto expected_transfer_parameters = to_string(FwdConvAlgorithm);
     run_test<Builder>({"DeviceGroupedConvFwdMultipleD_Wmma_CShuffle",
-                       "128,64,64,64",
+                       expected_transfer_parameters,
                        "GNWC,GKXC,EmptyTuple,GNWK",
                        "PassThrough,PassThrough,PassThrough",
                        "Default"});

experimental/builder/test/conv/ck/test_ckb_conv_fwd_2d_bf16.cpp

@@ -3,6 +3,7 @@
 
 #include "utils/ckb_conv_test_configs.hpp"
 #include "utils/ckb_conv_test_utils.hpp"
+#include "utils/conv_algorithm_type_utils.hpp"
 
 namespace {
 
@@ -12,13 +13,17 @@ using namespace ck_tile::builder::test_utils;
 TEST(FwdConvInstances,
      Create_DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3_Instance_2D_BF16_ChannelsLast)
 {
+    using enum ck_tile::builder::ConvDirection;
+    using enum ck_tile::builder::DataType;
+    using enum ck_tile::builder::TensorLayout;
+
     constexpr ConvSignature FwdConvSignature{.spatial_dim            = 2,
-                                             .direction              = ConvDirection::FORWARD,
-                                             .data_type              = DataType::BF16,
-                                             .accumulation_data_type = DataType::FP32,
-                                             .input  = {.config = {.layout = TensorLayout::NHWGC}},
-                                             .weight = {.config = {.layout = TensorLayout::GKYXC}},
-                                             .output = {.config = {.layout = TensorLayout::NHWGK}}};
+                                             .direction              = FORWARD,
+                                             .data_type              = BF16,
+                                             .accumulation_data_type = FP32,
+                                             .input  = {.config = {.layout = NHWGC}},
+                                             .weight = {.config = {.layout = GKYXC}},
+                                             .output = {.config = {.layout = NHWGK}}};
 
     constexpr auto FwdConvAlgorithm =
         ConvAlgorithm_DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3{}
@@ -29,8 +34,10 @@ TEST(FwdConvInstances,
             .with_block_gemm(BlockGemmDesc_v1_intrawave);
 
     using Builder = ConvBuilder<FwdConvSignature, FwdConvAlgorithm>;
+
+    const auto expected_transfer_parameters = to_string(FwdConvAlgorithm);
     run_test<Builder>({"DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3",
-                       "256,256,256,32",
+                       expected_transfer_parameters,
                        "Default",
                        "NHWGC,GKYXC,EmptyTuple,NHWGK",
                        "PassThrough,PassThrough,PassThrough",
@@ -43,13 +50,17 @@ TEST(FwdConvInstances,
 TEST(FwdConvInstances,
      Create_DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3_Instance_2D_BF16_NHWGC_Filter3x3)
 {
+    using enum ck_tile::builder::ConvDirection;
+    using enum ck_tile::builder::DataType;
+    using enum ck_tile::builder::TensorLayout;
+
     constexpr ConvSignature FwdConvSignature{.spatial_dim            = 2,
-                                             .direction              = ConvDirection::FORWARD,
-                                             .data_type              = DataType::BF16,
-                                             .accumulation_data_type = DataType::FP32,
-                                             .input  = {.config = {.layout = TensorLayout::NHWGC}},
-                                             .weight = {.config = {.layout = TensorLayout::GKYXC}},
-                                             .output = {.config = {.layout = TensorLayout::NHWGK}}};
+                                             .direction              = FORWARD,
+                                             .data_type              = BF16,
+                                             .accumulation_data_type = FP32,
+                                             .input  = {.config = {.layout = NHWGC}},
+                                             .weight = {.config = {.layout = GKYXC}},
+                                             .output = {.config = {.layout = NHWGK}}};
 
     constexpr auto FwdConvAlgorithm =
         ConvAlgorithm_DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3{}
@@ -60,7 +71,10 @@ TEST(FwdConvInstances,
             .with_block_gemm(BlockGemmDesc_v5_intrawave);
 
     using Builder = ConvBuilder<FwdConvSignature, FwdConvAlgorithm>;
+
+    const auto expected_transfer_parameters = to_string(FwdConvAlgorithm);
     run_test<Builder>({"DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3",
+                       expected_transfer_parameters,
                        "Filter3x3",
                        "NHWGC,GKYXC,EmptyTuple,NHWGK",
                        "PassThrough,PassThrough,PassThrough",

experimental/builder/test/conv/ck/test_ckb_conv_fwd_2d_bf16_scaleadd_relu.cpp

@@ -3,6 +3,7 @@
 
 #include "utils/ckb_conv_test_configs.hpp"
 #include "utils/ckb_conv_test_utils.hpp"
+#include "utils/conv_algorithm_type_utils.hpp"
 
 namespace {
 
@@ -12,19 +13,22 @@ using namespace ck_tile::builder::test_utils;
 TEST(FwdConvInstances,
      Create_DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_Instance_2D_BF16_scale_add_relu)
 {
+    using enum ck_tile::builder::ConvDirection;
+    using enum ck_tile::builder::DataType;
+    using enum ck_tile::builder::TensorLayout;
+    using enum ck_tile::builder::ElementwiseOperation;
+
     constexpr ConvSignature FwdConvSignature{
         .spatial_dim            = 2,
-        .direction              = ConvDirection::FORWARD,
-        .data_type              = DataType::BF16,
-        .accumulation_data_type = DataType::FP32,
-        .input                  = {.config = {.layout = TensorLayout::NHWGC}},
-        .weight = {.config = {.layout = TensorLayout::GKYXC, .data_type = DataType::BF16}},
-        .output = ConvolutionTensor{
-            .config    = {.layout = TensorLayout::NHWGK},
-            .operation = TensorOperation<>{.elementwise_operation =
-                                               ElementwiseOperation::SCALEADD_SCALEADD_RELU}
-                             .with_auxiliary_operand_configs<TensorLayout::NHWGK,
-                                                             TensorLayout::G_K_strided>()}};
+        .direction              = FORWARD,
+        .data_type              = BF16,
+        .accumulation_data_type = FP32,
+        .input                  = {.config = {.layout = NHWGC}},
+        .weight                 = {.config = {.layout = GKYXC, .data_type = BF16}},
+        .output                 = ConvolutionTensor{
+                            .config    = {.layout = NHWGK},
+                            .operation = TensorOperation<>{.elementwise_operation = SCALEADD_SCALEADD_RELU}
+                             .with_auxiliary_operand_configs<NHWGK, G_K_strided>()}};
 
     constexpr auto FwdConvAlgorithm =
         ConvAlgorithm_DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle{}
@@ -35,10 +39,12 @@ TEST(FwdConvInstances,
             .with_prefetch_config(1, 1, PipelineScheduler::DEFAULT);
 
     using Builder = ConvBuilder<FwdConvSignature, FwdConvAlgorithm>;
+
+    const auto expected_transfer_parameters = to_string(FwdConvAlgorithm);
     run_test<Builder>({"DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle",
+                       expected_transfer_parameters,
                        "NHWGC,GKYXC,Tuple(NHWGK,G_K),NHWGK",
                        "PassThrough,PassThrough,ScaleAddScaleAddRelu",
-                       "64,64,32,32",
                        "MNKPadding",
                        "Default"});
 }

experimental/builder/test/conv/ck/test_ckb_conv_fwd_2d_dl_fp16.cpp

@@ -3,6 +3,7 @@
 
 #include "utils/ckb_conv_test_configs.hpp"
 #include "utils/ckb_conv_test_utils.hpp"
+#include "utils/conv_algorithm_type_utils.hpp"
 
 namespace {
 
@@ -10,13 +11,17 @@ using namespace ck_tile::builder::test_utils;
 
 TEST(FwdConvInstances, Create_DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK_Instance_2D_FP16_GNHWC)
 {
+    using enum ck_tile::builder::ConvDirection;
+    using enum ck_tile::builder::DataType;
+    using enum ck_tile::builder::TensorLayout;
+
     constexpr ConvSignature FwdConvSignature{.spatial_dim            = 2,
-                                             .direction              = ConvDirection::FORWARD,
-                                             .data_type              = DataType::FP16,
-                                             .accumulation_data_type = DataType::FP32,
-                                             .input  = {.config = {.layout = TensorLayout::GNHWC}},
-                                             .weight = {.config = {.layout = TensorLayout::GKYXC}},
-                                             .output = {.config = {.layout = TensorLayout::GNHWK}}};
+                                             .direction              = FORWARD,
+                                             .data_type              = FP16,
+                                             .accumulation_data_type = FP32,
+                                             .input  = {.config = {.layout = GNHWC}},
+                                             .weight = {.config = {.layout = GKYXC}},
+                                             .output = {.config = {.layout = GNHWK}}};
 
     constexpr auto FwdConvAlgorithm =
         ConvAlgorithm_DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK{}
@@ -27,8 +32,10 @@ TEST(FwdConvInstances, Create_DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK_Ins
             .with_dl_transfer(DlFwdTransfer);
 
     using Builder = ConvBuilder<FwdConvSignature, FwdConvAlgorithm>;
+
+    const auto expected_transfer_parameters = to_string(FwdConvAlgorithm);
     run_test<Builder>({"DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK",
-                       "256,128,128,16",
+                       expected_transfer_parameters,
                        "Default",
                        "MNKPadding",
                        "GNHWC,GKYXC,EmptyTuple,GNHWK",
@@ -38,13 +45,17 @@ TEST(FwdConvInstances, Create_DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK_Ins
 TEST(FwdConvInstances,
      Create_DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK_Instance_2D_FP16_FILTER_1X1_PAD0)
 {
+    using enum ck_tile::builder::ConvDirection;
+    using enum ck_tile::builder::DataType;
+    using enum ck_tile::builder::TensorLayout;
+
     constexpr ConvSignature FwdConvSignature{.spatial_dim            = 2,
-                                             .direction              = ConvDirection::FORWARD,
-                                             .data_type              = DataType::FP16,
-                                             .accumulation_data_type = DataType::FP32,
-                                             .input  = {.config = {.layout = TensorLayout::GNHWC}},
-                                             .weight = {.config = {.layout = TensorLayout::GKYXC}},
-                                             .output = {.config = {.layout = TensorLayout::GNHWK}}};
+                                             .direction              = FORWARD,
+                                             .data_type              = FP16,
+                                             .accumulation_data_type = FP32,
+                                             .input  = {.config = {.layout = GNHWC}},
+                                             .weight = {.config = {.layout = GKYXC}},
+                                             .output = {.config = {.layout = GNHWK}}};
 
     constexpr auto FwdConvAlgorithm =
         ConvAlgorithm_DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK{}
@@ -56,8 +67,10 @@ TEST(FwdConvInstances,
             .with_dl_transfer(DlFwdTransfer);
 
     using Builder = ConvBuilder<FwdConvSignature, FwdConvAlgorithm>;
+
+    const auto expected_transfer_parameters = to_string(FwdConvAlgorithm);
     run_test<Builder>({"DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK",
-                       "256,128,128,16",
+                       expected_transfer_parameters,
                        "Filter1x1Pad0",
                        "MNKPadding",
                        "GNHWC,GKYXC,EmptyTuple,GNHWK",

experimental/builder/test/conv/ck/test_ckb_conv_fwd_2d_fp16.cpp

@@ -3,6 +3,7 @@
 
 #include "utils/ckb_conv_test_configs.hpp"
 #include "utils/ckb_conv_test_utils.hpp"
+#include "utils/conv_algorithm_type_utils.hpp"
 
 namespace {
 
@@ -11,13 +12,17 @@ using namespace ck_tile::builder::test_utils;
 TEST(FwdConvInstances,
      Create_DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3_Instance_2D_FP16_GNHWC)
 {
+    using enum ck_tile::builder::ConvDirection;
+    using enum ck_tile::builder::DataType;
+    using enum ck_tile::builder::TensorLayout;
+
     constexpr ConvSignature FwdConvSignature{.spatial_dim            = 2,
-                                             .direction              = ConvDirection::FORWARD,
-                                             .data_type              = DataType::FP16,
-                                             .accumulation_data_type = DataType::FP32,
-                                             .input  = {.config = {.layout = TensorLayout::GNHWC}},
-                                             .weight = {.config = {.layout = TensorLayout::GKYXC}},
-                                             .output = {.config = {.layout = TensorLayout::GNHWK}}};
+                                             .direction              = FORWARD,
+                                             .data_type              = FP16,
+                                             .accumulation_data_type = FP32,
+                                             .input  = {.config = {.layout = GNHWC}},
+                                             .weight = {.config = {.layout = GKYXC}},
+                                             .output = {.config = {.layout = GNHWK}}};
 
     constexpr auto FwdConvAlgorithm =
         ConvAlgorithm_DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3{}
@@ -29,8 +34,10 @@ TEST(FwdConvInstances,
             .with_block_gemm(BlockGemmDesc_v3_intrawave);
 
     using Builder = ConvBuilder<FwdConvSignature, FwdConvAlgorithm>;
+
+    const auto expected_transfer_parameters = to_string(FwdConvAlgorithm);
     run_test<Builder>({"DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3",
-                       "256,256,256,32",
+                       expected_transfer_parameters,
                        "Filter1x1Pad0",
                        "Intrawave",
                        "v3",

experimental/builder/test/conv/ck/test_ckb_conv_fwd_2d_fp32.cpp

@@ -3,6 +3,7 @@
 
 #include "utils/ckb_conv_test_configs.hpp"
 #include "utils/ckb_conv_test_utils.hpp"
+#include "utils/conv_algorithm_type_utils.hpp"
 
 namespace {
 
@@ -11,13 +12,17 @@ using namespace ck_tile::builder::test_utils;
 TEST(FwdConvInstances,
      Create_DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3_Instance_2D_FP32_NGCHW_GKCYX)
 {
+    using enum ck_tile::builder::ConvDirection;
+    using enum ck_tile::builder::DataType;
+    using enum ck_tile::builder::TensorLayout;
+
     constexpr ConvSignature FwdConvSignature{.spatial_dim            = 2,
-                                             .direction              = ConvDirection::FORWARD,
-                                             .data_type              = DataType::FP32,
-                                             .accumulation_data_type = DataType::FP32,
-                                             .input  = {.config = {.layout = TensorLayout::NGCHW}},
-                                             .weight = {.config = {.layout = TensorLayout::GKCYX}},
-                                             .output = {.config = {.layout = TensorLayout::NGKHW}}};
+                                             .direction              = FORWARD,
+                                             .data_type              = FP32,
+                                             .accumulation_data_type = FP32,
+                                             .input  = {.config = {.layout = NGCHW}},
+                                             .weight = {.config = {.layout = GKCYX}},
+                                             .output = {.config = {.layout = NGKHW}}};
 
     constexpr auto FwdConvAlgorithm =
         ConvAlgorithm_DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3{}
@@ -29,8 +34,10 @@ TEST(FwdConvInstances,
             .with_block_gemm(BlockGemmDesc_v4_intrawave);
 
     using Builder = ConvBuilder<FwdConvSignature, FwdConvAlgorithm>;
+
+    const auto expected_transfer_parameters = to_string(FwdConvAlgorithm);
     run_test<Builder>({"DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3",
-                       "256,128,128,32",
+                       expected_transfer_parameters,
                        "Filter1x1Stride1Pad0",
                        "Intrawave",
                        "v4",

experimental/builder/test/conv/ck/test_ckb_conv_fwd_2d_fp8.cpp

@@ -3,6 +3,7 @@
 
 #include "utils/ckb_conv_test_configs.hpp"
 #include "utils/ckb_conv_test_utils.hpp"
+#include "utils/conv_algorithm_type_utils.hpp"
 
 namespace {
 
@@ -12,13 +13,17 @@ using namespace ck_tile::builder::test_utils;
 TEST(FwdConvInstances,
      Create_DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_Instance_2D_FP8_ChannelsLast)
 {
+    using enum ck_tile::builder::ConvDirection;
+    using enum ck_tile::builder::DataType;
+    using enum ck_tile::builder::TensorLayout;
+
     constexpr ConvSignature FwdConvSignature{.spatial_dim            = 2,
-                                             .direction              = ConvDirection::FORWARD,
-                                             .data_type              = DataType::FP8,
-                                             .accumulation_data_type = DataType::FP32,
-                                             .input  = {.config = {.layout = TensorLayout::NHWGC}},
-                                             .weight = {.config = {.layout = TensorLayout::GKYXC}},
-                                             .output = {.config = {.layout = TensorLayout::NHWGK}}};
+                                             .direction              = FORWARD,
+                                             .data_type              = FP8,
+                                             .accumulation_data_type = FP32,
+                                             .input  = {.config = {.layout = NHWGC}},
+                                             .weight = {.config = {.layout = GKYXC}},
+                                             .output = {.config = {.layout = NHWGK}}};
 
     constexpr auto FwdConvAlgorithm =
         ConvAlgorithm_DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle{}
@@ -29,8 +34,10 @@ TEST(FwdConvInstances,
             .with_prefetch_config(1, 1, PipelineScheduler::DEFAULT);
 
     using Builder = ConvBuilder<FwdConvSignature, FwdConvAlgorithm>;
+
+    const auto expected_transfer_parameters = to_string(FwdConvAlgorithm);
     run_test<Builder>({"DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle",
-                       "256,256,128,32",
+                       expected_transfer_parameters,
                        "Default",
                        "NHWGC,GKYXC,EmptyTuple,NHWGK",
                        "PassThrough,PassThrough,PassThrough",

experimental/builder/test/conv/ck/test_ckb_conv_fwd_2d_large_tensor_fp16.cpp

@@ -3,6 +3,7 @@
 
 #include "utils/ckb_conv_test_configs.hpp"
 #include "utils/ckb_conv_test_utils.hpp"
+#include "utils/conv_algorithm_type_utils.hpp"
 
 namespace {
 
@@ -11,13 +12,17 @@ using namespace ck_tile::builder::test_utils;
 TEST(FwdConvInstances,
      Create_DeviceGroupedConvFwdMultipleD_Xdl_CShuffle_Large_Tensor_Instance_2D_FP16_GNHWC)
 {
+    using enum ck_tile::builder::ConvDirection;
+    using enum ck_tile::builder::DataType;
+    using enum ck_tile::builder::TensorLayout;
+
     constexpr ConvSignature FwdConvSignature{.spatial_dim            = 2,
-                                             .direction              = ConvDirection::FORWARD,
-                                             .data_type              = DataType::FP16,
-                                             .accumulation_data_type = DataType::FP32,
-                                             .input  = {.config = {.layout = TensorLayout::GNHWC}},
-                                             .weight = {.config = {.layout = TensorLayout::GKYXC}},
-                                             .output = {.config = {.layout = TensorLayout::GNHWK}}};
+                                             .direction              = FORWARD,
+                                             .data_type              = FP16,
+                                             .accumulation_data_type = FP32,
+                                             .input  = {.config = {.layout = GNHWC}},
+                                             .weight = {.config = {.layout = GKYXC}},
+                                             .output = {.config = {.layout = GNHWK}}};
 
     constexpr auto FwdConvAlgorithm =
         ConvAlgorithm_DeviceGroupedConvFwdMultipleD_Xdl_CShuffle_Large_Tensor{
@@ -30,8 +35,10 @@ TEST(FwdConvInstances,
                                   .with_prefetch_config(1, 1, PipelineScheduler::DEFAULT)};
 
     using Builder = ConvBuilder<FwdConvSignature, FwdConvAlgorithm>;
+
+    const auto expected_transfer_parameters = to_string(FwdConvAlgorithm);
     run_test<Builder>({"DeviceGroupedConvFwdMultipleD_Xdl_CShuffle_Large_Tensor",
-                       "256,256,128,32",
+                       expected_transfer_parameters,
                        "Default",
                        "GNHWC,GKYXC,EmptyTuple,GNHWK",
                        "PassThrough,PassThrough,PassThrough",
@@ -42,13 +49,17 @@ TEST(
     FwdConvInstances,
     Create_DeviceGroupedConvFwdMultipleD_Xdl_CShuffle_Large_Tensor_Instance_2D_FP16_GNHWC_Filter1x1Pad0)
 {
+    using enum ck_tile::builder::ConvDirection;
+    using enum ck_tile::builder::DataType;
+    using enum ck_tile::builder::TensorLayout;
+
     constexpr ConvSignature FwdConvSignature{.spatial_dim            = 2,
-                                             .direction              = ConvDirection::FORWARD,
-                                             .data_type              = DataType::FP16,
-                                             .accumulation_data_type = DataType::FP32,
-                                             .input  = {.config = {.layout = TensorLayout::GNHWC}},
-                                             .weight = {.config = {.layout = TensorLayout::GKYXC}},
-                                             .output = {.config = {.layout = TensorLayout::GNHWK}}};
+                                             .direction              = FORWARD,
+                                             .data_type              = FP16,
+                                             .accumulation_data_type = FP32,
+                                             .input  = {.config = {.layout = GNHWC}},
+                                             .weight = {.config = {.layout = GKYXC}},
+                                             .output = {.config = {.layout = GNHWK}}};
 
     constexpr auto FwdConvAlgorithm =
         ConvAlgorithm_DeviceGroupedConvFwdMultipleD_Xdl_CShuffle_Large_Tensor{
@@ -61,8 +72,10 @@ TEST(
                                   .with_prefetch_config(1, 1, PipelineScheduler::DEFAULT)};
 
     using Builder = ConvBuilder<FwdConvSignature, FwdConvAlgorithm>;
+
+    const auto expected_transfer_parameters = to_string(FwdConvAlgorithm);
     run_test<Builder>({"DeviceGroupedConvFwdMultipleD_Xdl_CShuffle_Large_Tensor",
-                       "128,128,128,32",
+                       expected_transfer_parameters,
                        "Filter1x1Pad0",
                        "GNHWC,GKYXC,EmptyTuple,GNHWK",
                        "PassThrough,PassThrough,PassThrough",

experimental/builder/test/conv/ck/test_ckb_conv_fwd_3d_bf16.cpp

@@ -3,6 +3,7 @@
 
 #include "utils/ckb_conv_test_configs.hpp"
 #include "utils/ckb_conv_test_utils.hpp"
+#include "utils/conv_algorithm_type_utils.hpp"
 
 namespace {
 
@@ -12,14 +13,17 @@ using namespace ck_tile::builder::test_utils;
 TEST(FwdConvInstances,
      Create_DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3_Instance_3D_BF16_GNDHWC)
 {
-    constexpr ConvSignature FwdConvSignature{
-        .spatial_dim            = 3,
-        .direction              = ConvDirection::FORWARD,
-        .data_type              = DataType::BF16,
-        .accumulation_data_type = DataType::FP32,
-        .input                  = {.config = {.layout = TensorLayout::GNDHWC}},
-        .weight                 = {.config = {.layout = TensorLayout::GKZYXC}},
-        .output                 = {.config = {.layout = TensorLayout::GNDHWK}}};
+    using enum ck_tile::builder::ConvDirection;
+    using enum ck_tile::builder::DataType;
+    using enum ck_tile::builder::TensorLayout;
+
+    constexpr ConvSignature FwdConvSignature{.spatial_dim            = 3,
+                                             .direction              = FORWARD,
+                                             .data_type              = BF16,
+                                             .accumulation_data_type = FP32,
+                                             .input  = {.config = {.layout = GNDHWC}},
+                                             .weight = {.config = {.layout = GKZYXC}},
+                                             .output = {.config = {.layout = GNDHWK}}};
 
     constexpr auto FwdConvAlgorithm =
         ConvAlgorithm_DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3{}
@@ -30,8 +34,10 @@ TEST(FwdConvInstances,
             .with_block_gemm(BlockGemmDesc_v3_intrawave);
 
     using Builder = ConvBuilder<FwdConvSignature, FwdConvAlgorithm>;
+
+    const auto expected_transfer_parameters = to_string(FwdConvAlgorithm);
     run_test<Builder>({"DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3",
-                       "256,256,256,32",
+                       expected_transfer_parameters,
                        "Default",
                        "Intrawave",
                        "v3",

experimental/builder/test/conv/ck/test_ckb_conv_fwd_3d_fp16.cpp

@@ -3,6 +3,7 @@
 
 #include "utils/ckb_conv_test_configs.hpp"
 #include "utils/ckb_conv_test_utils.hpp"
+#include "utils/conv_algorithm_type_utils.hpp"
 
 namespace {
 
@@ -12,14 +13,17 @@ using namespace ck_tile::builder::test_utils;
 TEST(FwdConvInstances,
      Create_DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3_Instance_3D_FP16_NDHWGC_ChannelsLast)
 {
-    constexpr ConvSignature FwdConvSignature{
-        .spatial_dim            = 3,
-        .direction              = ConvDirection::FORWARD,
-        .data_type              = DataType::FP16,
-        .accumulation_data_type = DataType::FP32,
-        .input                  = {.config = {.layout = TensorLayout::NDHWGC}},
-        .weight                 = {.config = {.layout = TensorLayout::GKZYXC}},
-        .output                 = {.config = {.layout = TensorLayout::NDHWGK}}};
+    using enum ck_tile::builder::ConvDirection;
+    using enum ck_tile::builder::DataType;
+    using enum ck_tile::builder::TensorLayout;
+
+    constexpr ConvSignature FwdConvSignature{.spatial_dim            = 3,
+                                             .direction              = FORWARD,
+                                             .data_type              = FP16,
+                                             .accumulation_data_type = FP32,
+                                             .input  = {.config = {.layout = NDHWGC}},
+                                             .weight = {.config = {.layout = GKZYXC}},
+                                             .output = {.config = {.layout = NDHWGK}}};
 
     constexpr auto FwdConvAlgorithm =
         ConvAlgorithm_DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3{}
@@ -31,8 +35,10 @@ TEST(FwdConvInstances,
             .with_block_gemm(BlockGemmDesc_v4_intrawave);
 
     using Builder = ConvBuilder<FwdConvSignature, FwdConvAlgorithm>;
+
+    const auto expected_transfer_parameters = to_string(FwdConvAlgorithm);
     run_test<Builder>({"DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3",
-                       "256,128,128,32",
+                       expected_transfer_parameters,
                        "Filter1x1Pad0",
                        "Intrawave",
                        "v4",

experimental/builder/test/conv/ck/test_ckb_conv_fwd_3d_fp32.cpp

@@ -3,6 +3,7 @@
 
 #include "utils/ckb_conv_test_configs.hpp"
 #include "utils/ckb_conv_test_utils.hpp"
+#include "utils/conv_algorithm_type_utils.hpp"
 
 namespace {
 
@@ -12,14 +13,17 @@ using namespace ck_tile::builder::test_utils;
 TEST(FwdConvInstances,
      Create_DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3_Instance_3D_FP32_ChannelsFirst)
 {
-    constexpr ConvSignature FwdConvSignature{
-        .spatial_dim            = 3,
-        .direction              = ConvDirection::FORWARD,
-        .data_type              = DataType::FP32,
-        .accumulation_data_type = DataType::FP32,
-        .input                  = {.config = {.layout = TensorLayout::NGCDHW}},
-        .weight                 = {.config = {.layout = TensorLayout::GKCZYX}},
-        .output                 = {.config = {.layout = TensorLayout::NGKDHW}}};
+    using enum ck_tile::builder::ConvDirection;
+    using enum ck_tile::builder::DataType;
+    using enum ck_tile::builder::TensorLayout;
+
+    constexpr ConvSignature FwdConvSignature{.spatial_dim            = 3,
+                                             .direction              = FORWARD,
+                                             .data_type              = FP32,
+                                             .accumulation_data_type = FP32,
+                                             .input  = {.config = {.layout = NGCDHW}},
+                                             .weight = {.config = {.layout = GKCZYX}},
+                                             .output = {.config = {.layout = NGKDHW}}};
 
     constexpr auto FwdConvAlgorithm =
         ConvAlgorithm_DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3{}
@@ -31,8 +35,10 @@ TEST(FwdConvInstances,
             .with_block_gemm(BlockGemmDesc_v1_intrawave);
 
     using Builder = ConvBuilder<FwdConvSignature, FwdConvAlgorithm>;
+
+    const auto expected_transfer_parameters = to_string(FwdConvAlgorithm);
     run_test<Builder>({"DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3",
-                       "256,256,256,32",
+                       expected_transfer_parameters,
                        "Filter1x1Pad0",
                        "Intrawave",
                        "v1",

experimental/builder/test/conv/ck/test_conv_traits.cpp

@@ -12,6 +12,12 @@
 
 namespace {
 
+using ck_tile::builder::ConvDirection;
+using ck_tile::builder::DataType;
+using ck_tile::builder::ElementwiseOperation;
+using ck_tile::builder::PipelineScheduler;
+using ck_tile::builder::PipelineVersion;
+using ck_tile::builder::TensorLayout;
 using ::testing::ElementsAre;
 
 // Test fixture for ConvTraits tests
@@ -84,15 +90,13 @@ TEST_F(ConvTraitsTest, ConvFwdTraitsExtraction)
 
     // Verify signature information
     EXPECT_EQ(Traits::spatial_dim, 2);
-    EXPECT_EQ(Traits::direction, ck_tile::builder::ConvDirection::FORWARD);
+    EXPECT_EQ(Traits::direction, ConvDirection::FORWARD);
     EXPECT_THAT(Traits::layout,
-                ::testing::ElementsAre(ck_tile::builder::TensorLayout::GNHWC,
-                                       ck_tile::builder::TensorLayout::GKYXC,
-                                       ck_tile::builder::TensorLayout::GNHWK));
-    EXPECT_EQ(Traits::data_type, ck_tile::builder::DataType::FP16);
-    EXPECT_EQ(Traits::input_element_op, ck_tile::builder::ElementwiseOperation::PASS_THROUGH);
-    EXPECT_EQ(Traits::weight_element_op, ck_tile::builder::ElementwiseOperation::PASS_THROUGH);
-    EXPECT_EQ(Traits::output_element_op, ck_tile::builder::ElementwiseOperation::PASS_THROUGH);
+                ElementsAre(TensorLayout::GNHWC, TensorLayout::GKYXC, TensorLayout::GNHWK));
+    EXPECT_EQ(Traits::data_type, DataType::FP16);
+    EXPECT_EQ(Traits::input_element_op, ElementwiseOperation::PASS_THROUGH);
+    EXPECT_EQ(Traits::weight_element_op, ElementwiseOperation::PASS_THROUGH);
+    EXPECT_EQ(Traits::output_element_op, ElementwiseOperation::PASS_THROUGH);
 
     // Verify specializations
     EXPECT_EQ(Traits::gemm_padding, ck_tile::builder::GemmPadding::DEFAULT);
@@ -145,8 +149,8 @@ TEST_F(ConvTraitsTest, ConvFwdTraitsExtraction)
     EXPECT_EQ(Traits::c_tile_transfer.scalar_per_vector, 8);
 
     // Verify pipeline configuration
-    EXPECT_EQ(Traits::pipeline_scheduler, ck_tile::builder::PipelineScheduler::INTRAWAVE);
-    EXPECT_EQ(Traits::pipeline_version, ck_tile::builder::PipelineVersion::V1);
+    EXPECT_EQ(Traits::pipeline_scheduler, PipelineScheduler::INTRAWAVE);
+    EXPECT_EQ(Traits::pipeline_version, PipelineVersion::V1);
 }
 
 // Test ConvTraits with DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle
@@ -214,15 +218,13 @@ TEST_F(ConvTraitsTest, ConvFwdBaseTraitsExtraction)
 
     // Verify signature information
     EXPECT_EQ(Traits::spatial_dim, 2);
-    EXPECT_EQ(Traits::direction, ck_tile::builder::ConvDirection::FORWARD);
+    EXPECT_EQ(Traits::direction, ConvDirection::FORWARD);
     EXPECT_THAT(Traits::layout,
-                ::testing::ElementsAre(ck_tile::builder::TensorLayout::GNHWC,
-                                       ck_tile::builder::TensorLayout::GKYXC,
-                                       ck_tile::builder::TensorLayout::GNHWK));
-    EXPECT_EQ(Traits::data_type, ck_tile::builder::DataType::FP16);
-    EXPECT_EQ(Traits::input_element_op, ck_tile::builder::ElementwiseOperation::PASS_THROUGH);
-    EXPECT_EQ(Traits::weight_element_op, ck_tile::builder::ElementwiseOperation::PASS_THROUGH);
-    EXPECT_EQ(Traits::output_element_op, ck_tile::builder::ElementwiseOperation::PASS_THROUGH);
+                ElementsAre(TensorLayout::GNHWC, TensorLayout::GKYXC, TensorLayout::GNHWK));
+    EXPECT_EQ(Traits::data_type, DataType::FP16);
+    EXPECT_EQ(Traits::input_element_op, ElementwiseOperation::PASS_THROUGH);
+    EXPECT_EQ(Traits::weight_element_op, ElementwiseOperation::PASS_THROUGH);
+    EXPECT_EQ(Traits::output_element_op, ElementwiseOperation::PASS_THROUGH);
 
     // Verify specializations
     EXPECT_EQ(Traits::gemm_padding, ck_tile::builder::GemmPadding::DEFAULT);
@@ -300,15 +302,13 @@ TEST_F(ConvTraitsTest, ConvFwdLargeTensorTraitsExtraction)
 
     // Verify signature information
     EXPECT_EQ(Traits::spatial_dim, 2);
-    EXPECT_EQ(Traits::direction, ck_tile::builder::ConvDirection::FORWARD);
+    EXPECT_EQ(Traits::direction, ConvDirection::FORWARD);
     EXPECT_THAT(Traits::layout,
-                ::testing::ElementsAre(ck_tile::builder::TensorLayout::GNHWC,
-                                       ck_tile::builder::TensorLayout::GKYXC,
-                                       ck_tile::builder::TensorLayout::GNHWK));
-    EXPECT_EQ(Traits::data_type, ck_tile::builder::DataType::FP16);
-    EXPECT_EQ(Traits::input_element_op, ck_tile::builder::ElementwiseOperation::PASS_THROUGH);
-    EXPECT_EQ(Traits::weight_element_op, ck_tile::builder::ElementwiseOperation::PASS_THROUGH);
-    EXPECT_EQ(Traits::output_element_op, ck_tile::builder::ElementwiseOperation::PASS_THROUGH);
+                ElementsAre(TensorLayout::GNHWC, TensorLayout::GKYXC, TensorLayout::GNHWK));
+    EXPECT_EQ(Traits::data_type, DataType::FP16);
+    EXPECT_EQ(Traits::input_element_op, ElementwiseOperation::PASS_THROUGH);
+    EXPECT_EQ(Traits::weight_element_op, ElementwiseOperation::PASS_THROUGH);
+    EXPECT_EQ(Traits::output_element_op, ElementwiseOperation::PASS_THROUGH);
 
     // Verify specializations
     EXPECT_EQ(Traits::gemm_padding, ck_tile::builder::GemmPadding::DEFAULT);

experimental/builder/test/impl/conv_signature_types.hpp

@@ -14,8 +14,8 @@ struct TensorConfig
 {
     TensorLayout layout;
     // Optional data types, override the type defined in the signature if provided.
-    DataType data_type{DataType::UNDEFINDED};
-    DataType compute_type{DataType::UNDEFINDED};
+    DataType data_type{DataType::UNDEFINED_DATA_TYPE};
+    DataType compute_type{DataType::UNDEFINED_DATA_TYPE};
 };
 
 template <TensorConfig... Configs>

experimental/builder/test/test_ck_factory_grouped_convolution_forward_convscale.cpp

@@ -76,54 +76,54 @@ struct F8_ConvScale
 
     constexpr static auto expected = {
         // clang-format off
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,128,128,128,32,8,8,32,32,4,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,128,128,32,32,8,8,32,32,2,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,128,128,64,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,128,32,128,32,8,8,32,32,1,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,128,64,128,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,256,128,256,32,8,8,32,32,2,4,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,256,128,64,32,8,8,32,32,2,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,256,64,128,32,8,8,32,32,1,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,64,32,64,32,8,8,32,32,1,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,128,128,128,32,8,8,32,32,4,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,128,128,32,32,8,8,32,32,2,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,128,128,64,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,128,32,128,32,8,8,32,32,1,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,128,64,128,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,256,128,256,32,8,8,32,32,2,4,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,256,128,64,32,8,8,32,32,2,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,256,64,128,32,8,8,32,32,1,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,64,32,64,32,8,8,32,32,1,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,128,128,128,32,8,8,32,32,4,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,128,128,32,32,8,8,32,32,2,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,128,128,64,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,128,32,128,32,8,8,32,32,1,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,128,64,128,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,256,128,256,32,8,8,32,32,2,4,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,256,128,64,32,8,8,32,32,2,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,256,64,128,32,8,8,32,32,1,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,64,32,64,32,8,8,32,32,1,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>"
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,128,128,128,32,8,8,32,32,4,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,128,128,32,32,8,8,32,32,2,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,128,128,64,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,128,32,128,32,8,8,32,32,1,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,128,64,128,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,256,128,256,32,8,8,32,32,2,4,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,256,128,64,32,8,8,32,32,2,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,256,64,128,32,8,8,32,32,1,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,64,32,64,32,8,8,32,32,1,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,128,128,128,32,8,8,32,32,4,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,128,128,32,32,8,8,32,32,2,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,128,128,64,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,128,32,128,32,8,8,32,32,1,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,128,64,128,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,256,128,256,32,8,8,32,32,2,4,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,256,128,64,32,8,8,32,32,2,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,256,64,128,32,8,8,32,32,1,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,64,32,64,32,8,8,32,32,1,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,128,128,128,32,8,8,32,32,4,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,128,128,32,32,8,8,32,32,2,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,128,128,64,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,128,32,128,32,8,8,32,32,1,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,128,64,128,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,256,128,256,32,8,8,32,32,2,4,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,256,128,64,32,8,8,32,32,2,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,256,64,128,32,8,8,32,32,1,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,64,32,64,32,8,8,32,32,1,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>"
         // clang-format on
     };
 };
@@ -141,54 +141,54 @@ struct F8_BF8_comb1_ConvScale
 
     constexpr static auto expected = {
         // clang-format off
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,128,128,128,32,8,8,32,32,4,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,bf8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,128,128,32,32,8,8,32,32,2,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,4),8,bf8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,128,128,64,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,4),8,bf8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,128,32,128,32,8,8,32,32,1,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,bf8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,128,64,128,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,bf8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,32,1,8),8,bf8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,bf8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,256,128,256,32,8,8,32,32,2,4,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,bf8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,256,128,64,32,8,8,32,32,2,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,bf8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,bf8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,256,64,128,32,8,8,32,32,1,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,bf8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,64,32,64,32,8,8,32,32,1,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,bf8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),1,bf8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,bf8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,16,1,4),1,bf8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,bf8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,128,128,128,32,8,8,32,32,4,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,bf8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,128,128,32,32,8,8,32,32,2,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,4),8,bf8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,128,128,64,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,4),8,bf8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,128,32,128,32,8,8,32,32,1,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,bf8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,128,64,128,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,bf8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,32,1,8),8,bf8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,bf8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,256,128,256,32,8,8,32,32,2,4,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,bf8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,256,128,64,32,8,8,32,32,2,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,bf8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,bf8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,256,64,128,32,8,8,32,32,1,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,bf8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,64,32,64,32,8,8,32,32,1,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,bf8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),1,bf8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,bf8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,16,1,4),1,bf8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,bf8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,128,128,128,32,8,8,32,32,4,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,bf8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,128,128,32,32,8,8,32,32,2,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,4),8,bf8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,128,128,64,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,4),8,bf8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,128,32,128,32,8,8,32,32,1,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,bf8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,128,64,128,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,bf8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,32,1,8),8,bf8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,bf8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,256,128,256,32,8,8,32,32,2,4,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,bf8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,256,128,64,32,8,8,32,32,2,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,bf8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,bf8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,256,64,128,32,8,8,32,32,1,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,bf8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,64,32,64,32,8,8,32,32,1,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,bf8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),1,bf8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,bf8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,16,1,4),1,bf8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,bf8,bf8,Default,1>"
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,128,128,128,32,8,8,32,32,4,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,bf8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,128,128,32,32,8,8,32,32,2,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,4),8,bf8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,128,128,64,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,4),8,bf8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,128,32,128,32,8,8,32,32,1,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,bf8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,128,64,128,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,bf8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,32,1,8),8,bf8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,bf8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,256,128,256,32,8,8,32,32,2,4,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,bf8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,256,128,64,32,8,8,32,32,2,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,bf8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,bf8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,256,64,128,32,8,8,32,32,1,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,bf8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,64,32,64,32,8,8,32,32,1,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,bf8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),1,bf8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,bf8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,16,1,4),1,bf8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,bf8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,128,128,128,32,8,8,32,32,4,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,bf8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,128,128,32,32,8,8,32,32,2,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,4),8,bf8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,128,128,64,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,4),8,bf8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,128,32,128,32,8,8,32,32,1,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,bf8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,128,64,128,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,bf8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,32,1,8),8,bf8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,bf8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,256,128,256,32,8,8,32,32,2,4,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,bf8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,256,128,64,32,8,8,32,32,2,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,bf8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,bf8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,256,64,128,32,8,8,32,32,1,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,bf8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,64,32,64,32,8,8,32,32,1,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,bf8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),1,bf8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,bf8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,16,1,4),1,bf8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,bf8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,128,128,128,32,8,8,32,32,4,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,bf8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,128,128,32,32,8,8,32,32,2,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,4),8,bf8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,128,128,64,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,4),8,bf8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,128,32,128,32,8,8,32,32,1,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,bf8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,128,64,128,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,bf8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,32,1,8),8,bf8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,bf8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,256,128,256,32,8,8,32,32,2,4,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,bf8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,256,128,64,32,8,8,32,32,2,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,bf8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,bf8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,256,64,128,32,8,8,32,32,1,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,bf8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,64,32,64,32,8,8,32,32,1,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,bf8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),1,bf8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,bf8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,16,1,4),1,bf8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,bf8,bf8,Default,1>"
         // clang-format on
     };
 };
@@ -206,54 +206,54 @@ struct F8_BF8_comb2_ConvScale
 
     constexpr static auto expected = {
         // clang-format off
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,128,128,128,32,8,8,32,32,4,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,fp8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,128,128,32,32,8,8,32,32,2,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,4),8,fp8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,128,128,64,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,4),8,fp8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,128,32,128,32,8,8,32,32,1,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,fp8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,128,64,128,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,fp8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,32,1,8),8,fp8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,256,128,256,32,8,8,32,32,2,4,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,256,128,64,32,8,8,32,32,2,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,256,64,128,32,8,8,32,32,1,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,64,32,64,32,8,8,32,32,1,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,fp8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),1,fp8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,fp8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,16,1,4),1,fp8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,fp8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,128,128,128,32,8,8,32,32,4,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,fp8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,128,128,32,32,8,8,32,32,2,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,4),8,fp8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,128,128,64,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,4),8,fp8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,128,32,128,32,8,8,32,32,1,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,fp8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,128,64,128,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,fp8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,32,1,8),8,fp8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,256,128,256,32,8,8,32,32,2,4,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,256,128,64,32,8,8,32,32,2,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,256,64,128,32,8,8,32,32,1,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,64,32,64,32,8,8,32,32,1,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,fp8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),1,fp8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,fp8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,16,1,4),1,fp8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,fp8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,128,128,128,32,8,8,32,32,4,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,fp8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,128,128,32,32,8,8,32,32,2,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,4),8,fp8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,128,128,64,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,4),8,fp8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,128,32,128,32,8,8,32,32,1,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,fp8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,128,64,128,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,fp8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,32,1,8),8,fp8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,256,128,256,32,8,8,32,32,2,4,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,256,128,64,32,8,8,32,32,2,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,256,64,128,32,8,8,32,32,1,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,64,32,64,32,8,8,32,32,1,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,fp8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),1,fp8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,fp8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,16,1,4),1,fp8,bf8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,fp8,bf8,Default,1>"
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,128,128,128,32,8,8,32,32,4,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,fp8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,128,128,32,32,8,8,32,32,2,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,4),8,fp8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,128,128,64,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,4),8,fp8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,128,32,128,32,8,8,32,32,1,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,fp8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,128,64,128,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,fp8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,32,1,8),8,fp8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,256,128,256,32,8,8,32,32,2,4,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,256,128,64,32,8,8,32,32,2,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,256,64,128,32,8,8,32,32,1,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,64,32,64,32,8,8,32,32,1,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,fp8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),1,fp8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,fp8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,16,1,4),1,fp8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,fp8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,128,128,128,32,8,8,32,32,4,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,fp8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,128,128,32,32,8,8,32,32,2,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,4),8,fp8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,128,128,64,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,4),8,fp8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,128,32,128,32,8,8,32,32,1,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,fp8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,128,64,128,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,fp8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,32,1,8),8,fp8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,256,128,256,32,8,8,32,32,2,4,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,256,128,64,32,8,8,32,32,2,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,256,64,128,32,8,8,32,32,1,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,64,32,64,32,8,8,32,32,1,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,fp8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),1,fp8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,fp8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,16,1,4),1,fp8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,fp8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,128,128,128,32,8,8,32,32,4,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,fp8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,128,128,32,32,8,8,32,32,2,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,4),8,fp8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,128,128,64,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,4),8,fp8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,128,32,128,32,8,8,32,32,1,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,fp8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,128,64,128,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,fp8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,32,1,8),8,fp8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,256,128,256,32,8,8,32,32,2,4,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,256,128,64,32,8,8,32,32,2,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,256,64,128,32,8,8,32,32,1,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,64,32,64,32,8,8,32,32,1,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,fp8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),1,fp8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,fp8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,16,1,4),1,fp8,bf8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,bf8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,fp8,bf8,Default,1>"
         // clang-format on
     };
 };
@@ -271,54 +271,54 @@ struct F8_BF8_comb3_ConvScale
 
     constexpr static auto expected = {
         // clang-format off
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,128,128,128,32,8,8,32,32,4,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,bf8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,128,128,32,32,8,8,32,32,2,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,4),8,bf8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,128,128,64,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,4),8,bf8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,128,32,128,32,8,8,32,32,1,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,bf8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,128,64,128,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,bf8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,32,1,8),8,bf8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,bf8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,256,128,256,32,8,8,32,32,2,4,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,bf8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,256,128,64,32,8,8,32,32,2,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,bf8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,bf8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,256,64,128,32,8,8,32,32,1,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,bf8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,64,32,64,32,8,8,32,32,1,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,bf8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),1,bf8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,bf8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,16,1,4),1,bf8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,bf8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,128,128,128,32,8,8,32,32,4,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,bf8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,128,128,32,32,8,8,32,32,2,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,4),8,bf8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,128,128,64,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,4),8,bf8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,128,32,128,32,8,8,32,32,1,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,bf8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,128,64,128,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,bf8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,32,1,8),8,bf8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,bf8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,256,128,256,32,8,8,32,32,2,4,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,bf8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,256,128,64,32,8,8,32,32,2,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,bf8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,bf8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,256,64,128,32,8,8,32,32,1,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,bf8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,64,32,64,32,8,8,32,32,1,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,bf8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),1,bf8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,bf8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,16,1,4),1,bf8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,bf8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,128,128,128,32,8,8,32,32,4,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,bf8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,128,128,32,32,8,8,32,32,2,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,4),8,bf8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,128,128,64,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,4),8,bf8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,128,32,128,32,8,8,32,32,1,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,bf8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,128,64,128,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,bf8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,32,1,8),8,bf8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,bf8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,256,128,256,32,8,8,32,32,2,4,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,bf8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,256,128,64,32,8,8,32,32,2,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,bf8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,bf8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,256,64,128,32,8,8,32,32,1,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,bf8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,64,32,64,32,8,8,32,32,1,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,bf8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),1,bf8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,bf8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,16,1,4),1,bf8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,bf8,fp8,Default,1>"
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,128,128,128,32,8,8,32,32,4,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,bf8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,128,128,32,32,8,8,32,32,2,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,4),8,bf8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,128,128,64,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,4),8,bf8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,128,32,128,32,8,8,32,32,1,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,bf8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,128,64,128,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,bf8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,32,1,8),8,bf8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,bf8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,256,128,256,32,8,8,32,32,2,4,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,bf8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,256,128,64,32,8,8,32,32,2,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,bf8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,bf8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,256,64,128,32,8,8,32,32,1,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,bf8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,64,32,64,32,8,8,32,32,1,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,bf8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),1,bf8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,bf8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,16,1,4),1,bf8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Default,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,bf8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,128,128,128,32,8,8,32,32,4,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,bf8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,128,128,32,32,8,8,32,32,2,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,4),8,bf8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,128,128,64,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,4),8,bf8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,128,32,128,32,8,8,32,32,1,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,bf8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,128,64,128,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,bf8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,32,1,8),8,bf8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,bf8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,256,128,256,32,8,8,32,32,2,4,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,bf8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,256,128,64,32,8,8,32,32,2,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,bf8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,bf8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,256,64,128,32,8,8,32,32,1,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,bf8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,64,32,64,32,8,8,32,32,1,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,bf8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),1,bf8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,bf8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,16,1,4),1,bf8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,bf8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,128,128,128,32,8,8,32,32,4,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,bf8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,128,128,32,32,8,8,32,32,2,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,4),8,bf8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,128,128,64,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,4),8,bf8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,128,32,128,32,8,8,32,32,1,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,bf8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,128,64,128,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,bf8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,32,1,8),8,bf8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,bf8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,256,128,256,32,8,8,32,32,2,4,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,bf8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,256,128,64,32,8,8,32,32,2,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,bf8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,bf8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,256,64,128,32,8,8,32,32,1,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,bf8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,64,32,64,32,8,8,32,32,1,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,bf8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),1,bf8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,bf8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,16,1,4),1,bf8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScale,Filter1x1Stride1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,bf8,fp8,Default,1>"
         // clang-format on
     };
 };
@@ -336,54 +336,54 @@ struct F8_float_CombConvScale
 
     constexpr static auto expected = {
         // clang-format off
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Default,MNKPadding,1,128,128,128,32,8,8,32,32,4,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Default,MNKPadding,1,128,128,32,32,8,8,32,32,2,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Default,MNKPadding,1,128,128,64,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Default,MNKPadding,1,128,32,128,32,8,8,32,32,1,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Default,MNKPadding,1,128,64,128,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Default,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Default,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Default,MNKPadding,1,256,128,256,32,8,8,32,32,2,4,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Default,MNKPadding,1,256,128,64,32,8,8,32,32,2,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Default,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Default,MNKPadding,1,256,64,128,32,8,8,32,32,1,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Default,MNKPadding,1,64,32,64,32,8,8,32,32,1,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Default,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Default,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Default,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Default,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Pad0,MNKPadding,1,128,128,128,32,8,8,32,32,4,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Pad0,MNKPadding,1,128,128,32,32,8,8,32,32,2,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Pad0,MNKPadding,1,128,128,64,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Pad0,MNKPadding,1,128,32,128,32,8,8,32,32,1,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Pad0,MNKPadding,1,128,64,128,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Pad0,MNKPadding,1,256,128,256,32,8,8,32,32,2,4,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Pad0,MNKPadding,1,256,128,64,32,8,8,32,32,2,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Pad0,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Pad0,MNKPadding,1,256,64,128,32,8,8,32,32,1,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Pad0,MNKPadding,1,64,32,64,32,8,8,32,32,1,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Stride1Pad0,MNKPadding,1,128,128,128,32,8,8,32,32,4,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Stride1Pad0,MNKPadding,1,128,128,32,32,8,8,32,32,2,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Stride1Pad0,MNKPadding,1,128,128,64,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Stride1Pad0,MNKPadding,1,128,32,128,32,8,8,32,32,1,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Stride1Pad0,MNKPadding,1,128,64,128,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Stride1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Stride1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Stride1Pad0,MNKPadding,1,256,128,256,32,8,8,32,32,2,4,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Stride1Pad0,MNKPadding,1,256,128,64,32,8,8,32,32,2,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Stride1Pad0,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Stride1Pad0,MNKPadding,1,256,64,128,32,8,8,32,32,1,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Stride1Pad0,MNKPadding,1,64,32,64,32,8,8,32,32,1,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Stride1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Stride1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Stride1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Stride1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>"
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Default,MNKPadding,1,128,128,128,32,8,8,32,32,4,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Default,MNKPadding,1,128,128,32,32,8,8,32,32,2,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Default,MNKPadding,1,128,128,64,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Default,MNKPadding,1,128,32,128,32,8,8,32,32,1,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Default,MNKPadding,1,128,64,128,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Default,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Default,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Default,MNKPadding,1,256,128,256,32,8,8,32,32,2,4,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Default,MNKPadding,1,256,128,64,32,8,8,32,32,2,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Default,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Default,MNKPadding,1,256,64,128,32,8,8,32,32,1,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Default,MNKPadding,1,64,32,64,32,8,8,32,32,1,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Default,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Default,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Default,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Default,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Pad0,MNKPadding,1,128,128,128,32,8,8,32,32,4,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Pad0,MNKPadding,1,128,128,32,32,8,8,32,32,2,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Pad0,MNKPadding,1,128,128,64,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Pad0,MNKPadding,1,128,32,128,32,8,8,32,32,1,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Pad0,MNKPadding,1,128,64,128,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Pad0,MNKPadding,1,256,128,256,32,8,8,32,32,2,4,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Pad0,MNKPadding,1,256,128,64,32,8,8,32,32,2,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Pad0,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Pad0,MNKPadding,1,256,64,128,32,8,8,32,32,1,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Pad0,MNKPadding,1,64,32,64,32,8,8,32,32,1,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Stride1Pad0,MNKPadding,1,128,128,128,32,8,8,32,32,4,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Stride1Pad0,MNKPadding,1,128,128,32,32,8,8,32,32,2,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Stride1Pad0,MNKPadding,1,128,128,64,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Stride1Pad0,MNKPadding,1,128,32,128,32,8,8,32,32,1,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Stride1Pad0,MNKPadding,1,128,64,128,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Stride1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Stride1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Stride1Pad0,MNKPadding,1,256,128,256,32,8,8,32,32,2,4,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Stride1Pad0,MNKPadding,1,256,128,64,32,8,8,32,32,2,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Stride1Pad0,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Stride1Pad0,MNKPadding,1,256,64,128,32,8,8,32,32,1,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Stride1Pad0,MNKPadding,1,64,32,64,32,8,8,32,32,1,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Stride1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Stride1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Stride1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Stride1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>"
         // clang-format on
     };
 };
@@ -401,54 +401,54 @@ struct F8_ConvScaleRelu
 
     constexpr static auto expected = {
         // clang-format off
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Default,MNKPadding,1,128,128,128,32,8,8,32,32,4,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Default,MNKPadding,1,128,128,32,32,8,8,32,32,2,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Default,MNKPadding,1,128,128,64,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Default,MNKPadding,1,128,32,128,32,8,8,32,32,1,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Default,MNKPadding,1,128,64,128,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Default,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Default,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Default,MNKPadding,1,256,128,256,32,8,8,32,32,2,4,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Default,MNKPadding,1,256,128,64,32,8,8,32,32,2,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Default,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Default,MNKPadding,1,256,64,128,32,8,8,32,32,1,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Default,MNKPadding,1,64,32,64,32,8,8,32,32,1,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Default,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Default,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Default,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Default,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Filter1x1Pad0,MNKPadding,1,128,128,128,32,8,8,32,32,4,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Filter1x1Pad0,MNKPadding,1,128,128,32,32,8,8,32,32,2,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Filter1x1Pad0,MNKPadding,1,128,128,64,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Filter1x1Pad0,MNKPadding,1,128,32,128,32,8,8,32,32,1,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Filter1x1Pad0,MNKPadding,1,128,64,128,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Filter1x1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Filter1x1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Filter1x1Pad0,MNKPadding,1,256,128,256,32,8,8,32,32,2,4,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Filter1x1Pad0,MNKPadding,1,256,128,64,32,8,8,32,32,2,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Filter1x1Pad0,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Filter1x1Pad0,MNKPadding,1,256,64,128,32,8,8,32,32,1,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Filter1x1Pad0,MNKPadding,1,64,32,64,32,8,8,32,32,1,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Filter1x1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Filter1x1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Filter1x1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Filter1x1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Filter1x1Stride1Pad0,MNKPadding,1,128,128,128,32,8,8,32,32,4,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Filter1x1Stride1Pad0,MNKPadding,1,128,128,32,32,8,8,32,32,2,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Filter1x1Stride1Pad0,MNKPadding,1,128,128,64,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Filter1x1Stride1Pad0,MNKPadding,1,128,32,128,32,8,8,32,32,1,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Filter1x1Stride1Pad0,MNKPadding,1,128,64,128,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Filter1x1Stride1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Filter1x1Stride1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Filter1x1Stride1Pad0,MNKPadding,1,256,128,256,32,8,8,32,32,2,4,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Filter1x1Stride1Pad0,MNKPadding,1,256,128,64,32,8,8,32,32,2,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Filter1x1Stride1Pad0,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Filter1x1Stride1Pad0,MNKPadding,1,256,64,128,32,8,8,32,32,1,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Filter1x1Stride1Pad0,MNKPadding,1,64,32,64,32,8,8,32,32,1,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Filter1x1Stride1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Filter1x1Stride1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Filter1x1Stride1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Filter1x1Stride1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>"
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Default,MNKPadding,1,128,128,128,32,8,8,32,32,4,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Default,MNKPadding,1,128,128,32,32,8,8,32,32,2,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Default,MNKPadding,1,128,128,64,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Default,MNKPadding,1,128,32,128,32,8,8,32,32,1,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Default,MNKPadding,1,128,64,128,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Default,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Default,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Default,MNKPadding,1,256,128,256,32,8,8,32,32,2,4,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Default,MNKPadding,1,256,128,64,32,8,8,32,32,2,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Default,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Default,MNKPadding,1,256,64,128,32,8,8,32,32,1,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Default,MNKPadding,1,64,32,64,32,8,8,32,32,1,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Default,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Default,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Default,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Default,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Filter1x1Pad0,MNKPadding,1,128,128,128,32,8,8,32,32,4,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Filter1x1Pad0,MNKPadding,1,128,128,32,32,8,8,32,32,2,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Filter1x1Pad0,MNKPadding,1,128,128,64,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Filter1x1Pad0,MNKPadding,1,128,32,128,32,8,8,32,32,1,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Filter1x1Pad0,MNKPadding,1,128,64,128,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Filter1x1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Filter1x1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Filter1x1Pad0,MNKPadding,1,256,128,256,32,8,8,32,32,2,4,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Filter1x1Pad0,MNKPadding,1,256,128,64,32,8,8,32,32,2,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Filter1x1Pad0,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Filter1x1Pad0,MNKPadding,1,256,64,128,32,8,8,32,32,1,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Filter1x1Pad0,MNKPadding,1,64,32,64,32,8,8,32,32,1,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Filter1x1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Filter1x1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Filter1x1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Filter1x1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Filter1x1Stride1Pad0,MNKPadding,1,128,128,128,32,8,8,32,32,4,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Filter1x1Stride1Pad0,MNKPadding,1,128,128,32,32,8,8,32,32,2,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Filter1x1Stride1Pad0,MNKPadding,1,128,128,64,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Filter1x1Stride1Pad0,MNKPadding,1,128,32,128,32,8,8,32,32,1,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Filter1x1Stride1Pad0,MNKPadding,1,128,64,128,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Filter1x1Stride1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Filter1x1Stride1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Filter1x1Stride1Pad0,MNKPadding,1,256,128,256,32,8,8,32,32,2,4,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Filter1x1Stride1Pad0,MNKPadding,1,256,128,64,32,8,8,32,32,2,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Filter1x1Stride1Pad0,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Filter1x1Stride1Pad0,MNKPadding,1,256,64,128,32,8,8,32,32,1,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Filter1x1Stride1Pad0,MNKPadding,1,64,32,64,32,8,8,32,32,1,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Filter1x1Stride1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Filter1x1Stride1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Filter1x1Stride1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvScaleRelu,Filter1x1Stride1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>"
         // clang-format on
     };
 };
@@ -466,54 +466,54 @@ struct F8_CombConvScaleRelu
 
     constexpr static auto expected = {
         // clang-format off
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Default,MNKPadding,1,128,128,128,32,8,8,32,32,4,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Default,MNKPadding,1,128,128,32,32,8,8,32,32,2,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Default,MNKPadding,1,128,128,64,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Default,MNKPadding,1,128,32,128,32,8,8,32,32,1,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Default,MNKPadding,1,128,64,128,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Default,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Default,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Default,MNKPadding,1,256,128,256,32,8,8,32,32,2,4,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Default,MNKPadding,1,256,128,64,32,8,8,32,32,2,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Default,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Default,MNKPadding,1,256,64,128,32,8,8,32,32,1,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Default,MNKPadding,1,64,32,64,32,8,8,32,32,1,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Default,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Default,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Default,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Default,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Pad0,MNKPadding,1,128,128,128,32,8,8,32,32,4,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Pad0,MNKPadding,1,128,128,32,32,8,8,32,32,2,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Pad0,MNKPadding,1,128,128,64,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Pad0,MNKPadding,1,128,32,128,32,8,8,32,32,1,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Pad0,MNKPadding,1,128,64,128,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Pad0,MNKPadding,1,256,128,256,32,8,8,32,32,2,4,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Pad0,MNKPadding,1,256,128,64,32,8,8,32,32,2,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Pad0,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Pad0,MNKPadding,1,256,64,128,32,8,8,32,32,1,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Pad0,MNKPadding,1,64,32,64,32,8,8,32,32,1,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Stride1Pad0,MNKPadding,1,128,128,128,32,8,8,32,32,4,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Stride1Pad0,MNKPadding,1,128,128,32,32,8,8,32,32,2,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Stride1Pad0,MNKPadding,1,128,128,64,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Stride1Pad0,MNKPadding,1,128,32,128,32,8,8,32,32,1,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Stride1Pad0,MNKPadding,1,128,64,128,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Stride1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Stride1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Stride1Pad0,MNKPadding,1,256,128,256,32,8,8,32,32,2,4,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Stride1Pad0,MNKPadding,1,256,128,64,32,8,8,32,32,2,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Stride1Pad0,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Stride1Pad0,MNKPadding,1,256,64,128,32,8,8,32,32,1,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Stride1Pad0,MNKPadding,1,64,32,64,32,8,8,32,32,1,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Stride1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Stride1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Stride1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Stride1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>"
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Default,MNKPadding,1,128,128,128,32,8,8,32,32,4,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Default,MNKPadding,1,128,128,32,32,8,8,32,32,2,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Default,MNKPadding,1,128,128,64,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Default,MNKPadding,1,128,32,128,32,8,8,32,32,1,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Default,MNKPadding,1,128,64,128,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Default,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Default,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Default,MNKPadding,1,256,128,256,32,8,8,32,32,2,4,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Default,MNKPadding,1,256,128,64,32,8,8,32,32,2,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Default,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Default,MNKPadding,1,256,64,128,32,8,8,32,32,1,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Default,MNKPadding,1,64,32,64,32,8,8,32,32,1,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Default,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Default,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Default,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Default,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Pad0,MNKPadding,1,128,128,128,32,8,8,32,32,4,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Pad0,MNKPadding,1,128,128,32,32,8,8,32,32,2,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Pad0,MNKPadding,1,128,128,64,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Pad0,MNKPadding,1,128,32,128,32,8,8,32,32,1,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Pad0,MNKPadding,1,128,64,128,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Pad0,MNKPadding,1,256,128,256,32,8,8,32,32,2,4,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Pad0,MNKPadding,1,256,128,64,32,8,8,32,32,2,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Pad0,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Pad0,MNKPadding,1,256,64,128,32,8,8,32,32,1,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Pad0,MNKPadding,1,64,32,64,32,8,8,32,32,1,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Stride1Pad0,MNKPadding,1,128,128,128,32,8,8,32,32,4,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Stride1Pad0,MNKPadding,1,128,128,32,32,8,8,32,32,2,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Stride1Pad0,MNKPadding,1,128,128,64,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Stride1Pad0,MNKPadding,1,128,32,128,32,8,8,32,32,1,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Stride1Pad0,MNKPadding,1,128,64,128,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Stride1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Stride1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Stride1Pad0,MNKPadding,1,256,128,256,32,8,8,32,32,2,4,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Stride1Pad0,MNKPadding,1,256,128,64,32,8,8,32,32,2,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Stride1Pad0,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Stride1Pad0,MNKPadding,1,256,64,128,32,8,8,32,32,1,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Stride1Pad0,MNKPadding,1,64,32,64,32,8,8,32,32,1,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Stride1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Stride1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Stride1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,UnaryCombinedOp,Filter1x1Stride1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>"
         // clang-format on
     };
 };
@@ -531,54 +531,54 @@ struct F8_ConvScaleAdd
 
     constexpr static auto expected = {
         // clang-format off
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Default,MNKPadding,1,128,128,128,32,8,8,32,32,4,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Default,MNKPadding,1,128,128,32,32,8,8,32,32,2,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Default,MNKPadding,1,128,128,64,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Default,MNKPadding,1,128,32,128,32,8,8,32,32,1,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Default,MNKPadding,1,128,64,128,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Default,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Default,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Default,MNKPadding,1,256,128,256,32,8,8,32,32,2,4,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Default,MNKPadding,1,256,128,64,32,8,8,32,32,2,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Default,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Default,MNKPadding,1,256,64,128,32,8,8,32,32,1,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Default,MNKPadding,1,64,32,64,32,8,8,32,32,1,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Default,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Default,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Default,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Default,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Filter1x1Pad0,MNKPadding,1,128,128,128,32,8,8,32,32,4,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Filter1x1Pad0,MNKPadding,1,128,128,32,32,8,8,32,32,2,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Filter1x1Pad0,MNKPadding,1,128,128,64,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Filter1x1Pad0,MNKPadding,1,128,32,128,32,8,8,32,32,1,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Filter1x1Pad0,MNKPadding,1,128,64,128,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Filter1x1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Filter1x1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Filter1x1Pad0,MNKPadding,1,256,128,256,32,8,8,32,32,2,4,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Filter1x1Pad0,MNKPadding,1,256,128,64,32,8,8,32,32,2,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Filter1x1Pad0,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Filter1x1Pad0,MNKPadding,1,256,64,128,32,8,8,32,32,1,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Filter1x1Pad0,MNKPadding,1,64,32,64,32,8,8,32,32,1,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Filter1x1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Filter1x1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Filter1x1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Filter1x1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Filter1x1Stride1Pad0,MNKPadding,1,128,128,128,32,8,8,32,32,4,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Filter1x1Stride1Pad0,MNKPadding,1,128,128,32,32,8,8,32,32,2,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Filter1x1Stride1Pad0,MNKPadding,1,128,128,64,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Filter1x1Stride1Pad0,MNKPadding,1,128,32,128,32,8,8,32,32,1,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Filter1x1Stride1Pad0,MNKPadding,1,128,64,128,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Filter1x1Stride1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Filter1x1Stride1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Filter1x1Stride1Pad0,MNKPadding,1,256,128,256,32,8,8,32,32,2,4,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Filter1x1Stride1Pad0,MNKPadding,1,256,128,64,32,8,8,32,32,2,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Filter1x1Stride1Pad0,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Filter1x1Stride1Pad0,MNKPadding,1,256,64,128,32,8,8,32,32,1,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Filter1x1Stride1Pad0,MNKPadding,1,64,32,64,32,8,8,32,32,1,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Filter1x1Stride1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Filter1x1Stride1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Filter1x1Stride1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Filter1x1Stride1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>"
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Default,MNKPadding,1,128,128,128,32,8,8,32,32,4,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Default,MNKPadding,1,128,128,32,32,8,8,32,32,2,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Default,MNKPadding,1,128,128,64,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Default,MNKPadding,1,128,32,128,32,8,8,32,32,1,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Default,MNKPadding,1,128,64,128,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Default,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Default,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Default,MNKPadding,1,256,128,256,32,8,8,32,32,2,4,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Default,MNKPadding,1,256,128,64,32,8,8,32,32,2,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Default,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Default,MNKPadding,1,256,64,128,32,8,8,32,32,1,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Default,MNKPadding,1,64,32,64,32,8,8,32,32,1,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Default,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Default,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Default,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Default,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Filter1x1Pad0,MNKPadding,1,128,128,128,32,8,8,32,32,4,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Filter1x1Pad0,MNKPadding,1,128,128,32,32,8,8,32,32,2,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Filter1x1Pad0,MNKPadding,1,128,128,64,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Filter1x1Pad0,MNKPadding,1,128,32,128,32,8,8,32,32,1,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Filter1x1Pad0,MNKPadding,1,128,64,128,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Filter1x1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Filter1x1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Filter1x1Pad0,MNKPadding,1,256,128,256,32,8,8,32,32,2,4,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Filter1x1Pad0,MNKPadding,1,256,128,64,32,8,8,32,32,2,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Filter1x1Pad0,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Filter1x1Pad0,MNKPadding,1,256,64,128,32,8,8,32,32,1,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Filter1x1Pad0,MNKPadding,1,64,32,64,32,8,8,32,32,1,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Filter1x1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Filter1x1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Filter1x1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Filter1x1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Filter1x1Stride1Pad0,MNKPadding,1,128,128,128,32,8,8,32,32,4,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Filter1x1Stride1Pad0,MNKPadding,1,128,128,32,32,8,8,32,32,2,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Filter1x1Stride1Pad0,MNKPadding,1,128,128,64,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Filter1x1Stride1Pad0,MNKPadding,1,128,32,128,32,8,8,32,32,1,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Filter1x1Stride1Pad0,MNKPadding,1,128,64,128,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Filter1x1Stride1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Filter1x1Stride1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Filter1x1Stride1Pad0,MNKPadding,1,256,128,256,32,8,8,32,32,2,4,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Filter1x1Stride1Pad0,MNKPadding,1,256,128,64,32,8,8,32,32,2,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Filter1x1Stride1Pad0,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Filter1x1Stride1Pad0,MNKPadding,1,256,64,128,32,8,8,32,32,1,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Filter1x1Stride1Pad0,MNKPadding,1,64,32,64,32,8,8,32,32,1,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Filter1x1Stride1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Filter1x1Stride1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Filter1x1Stride1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK),NDHWGK,fp8,fp8,fp32,fp32,Tuple(fp32),fp8,PassThrough,PassThrough,ConvScaleAdd,Filter1x1Stride1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>"
         // clang-format on
     };
 };
@@ -596,54 +596,54 @@ struct F8_ConvInvscale
 
     constexpr static auto expected = {
         // clang-format off
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Default,MNKPadding,1,128,128,128,32,8,8,32,32,4,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Default,MNKPadding,1,128,128,32,32,8,8,32,32,2,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Default,MNKPadding,1,128,128,64,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Default,MNKPadding,1,128,32,128,32,8,8,32,32,1,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Default,MNKPadding,1,128,64,128,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Default,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Default,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Default,MNKPadding,1,256,128,256,32,8,8,32,32,2,4,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Default,MNKPadding,1,256,128,64,32,8,8,32,32,2,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Default,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Default,MNKPadding,1,256,64,128,32,8,8,32,32,1,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Default,MNKPadding,1,64,32,64,32,8,8,32,32,1,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Default,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Default,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Default,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Default,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Filter1x1Pad0,MNKPadding,1,128,128,128,32,8,8,32,32,4,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Filter1x1Pad0,MNKPadding,1,128,128,32,32,8,8,32,32,2,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Filter1x1Pad0,MNKPadding,1,128,128,64,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Filter1x1Pad0,MNKPadding,1,128,32,128,32,8,8,32,32,1,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Filter1x1Pad0,MNKPadding,1,128,64,128,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Filter1x1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Filter1x1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Filter1x1Pad0,MNKPadding,1,256,128,256,32,8,8,32,32,2,4,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Filter1x1Pad0,MNKPadding,1,256,128,64,32,8,8,32,32,2,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Filter1x1Pad0,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Filter1x1Pad0,MNKPadding,1,256,64,128,32,8,8,32,32,1,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Filter1x1Pad0,MNKPadding,1,64,32,64,32,8,8,32,32,1,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Filter1x1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Filter1x1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Filter1x1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Filter1x1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Filter1x1Stride1Pad0,MNKPadding,1,128,128,128,32,8,8,32,32,4,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Filter1x1Stride1Pad0,MNKPadding,1,128,128,32,32,8,8,32,32,2,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Filter1x1Stride1Pad0,MNKPadding,1,128,128,64,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Filter1x1Stride1Pad0,MNKPadding,1,128,32,128,32,8,8,32,32,1,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Filter1x1Stride1Pad0,MNKPadding,1,128,64,128,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Filter1x1Stride1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Filter1x1Stride1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Filter1x1Stride1Pad0,MNKPadding,1,256,128,256,32,8,8,32,32,2,4,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Filter1x1Stride1Pad0,MNKPadding,1,256,128,64,32,8,8,32,32,2,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Filter1x1Stride1Pad0,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Filter1x1Stride1Pad0,MNKPadding,1,256,64,128,32,8,8,32,32,1,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Filter1x1Stride1Pad0,MNKPadding,1,64,32,64,32,8,8,32,32,1,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Filter1x1Stride1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Filter1x1Stride1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Filter1x1Stride1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Filter1x1Stride1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>"
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Default,MNKPadding,1,128,128,128,32,8,8,32,32,4,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Default,MNKPadding,1,128,128,32,32,8,8,32,32,2,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Default,MNKPadding,1,128,128,64,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Default,MNKPadding,1,128,32,128,32,8,8,32,32,1,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Default,MNKPadding,1,128,64,128,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Default,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Default,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Default,MNKPadding,1,256,128,256,32,8,8,32,32,2,4,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Default,MNKPadding,1,256,128,64,32,8,8,32,32,2,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Default,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Default,MNKPadding,1,256,64,128,32,8,8,32,32,1,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Default,MNKPadding,1,64,32,64,32,8,8,32,32,1,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Default,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Default,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Default,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Default,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Filter1x1Pad0,MNKPadding,1,128,128,128,32,8,8,32,32,4,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Filter1x1Pad0,MNKPadding,1,128,128,32,32,8,8,32,32,2,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Filter1x1Pad0,MNKPadding,1,128,128,64,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Filter1x1Pad0,MNKPadding,1,128,32,128,32,8,8,32,32,1,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Filter1x1Pad0,MNKPadding,1,128,64,128,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Filter1x1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Filter1x1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Filter1x1Pad0,MNKPadding,1,256,128,256,32,8,8,32,32,2,4,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Filter1x1Pad0,MNKPadding,1,256,128,64,32,8,8,32,32,2,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Filter1x1Pad0,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Filter1x1Pad0,MNKPadding,1,256,64,128,32,8,8,32,32,1,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Filter1x1Pad0,MNKPadding,1,64,32,64,32,8,8,32,32,1,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Filter1x1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Filter1x1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Filter1x1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Filter1x1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Filter1x1Stride1Pad0,MNKPadding,1,128,128,128,32,8,8,32,32,4,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Filter1x1Stride1Pad0,MNKPadding,1,128,128,32,32,8,8,32,32,2,1,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Filter1x1Stride1Pad0,MNKPadding,1,128,128,64,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Filter1x1Stride1Pad0,MNKPadding,1,128,32,128,32,8,8,32,32,1,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Filter1x1Stride1Pad0,MNKPadding,1,128,64,128,32,8,8,32,32,2,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Filter1x1Stride1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Filter1x1Stride1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Filter1x1Stride1Pad0,MNKPadding,1,256,128,256,32,8,8,32,32,2,4,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Filter1x1Stride1Pad0,MNKPadding,1,256,128,64,32,8,8,32,32,2,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Filter1x1Stride1Pad0,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Filter1x1Stride1Pad0,MNKPadding,1,256,64,128,32,8,8,32,32,1,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Filter1x1Stride1Pad0,MNKPadding,1,64,32,64,32,8,8,32,32,1,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Filter1x1Stride1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Filter1x1Stride1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Filter1x1Stride1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,16,1,4),1,fp8,fp8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp8,fp8,fp32,fp32,EmptyTuple,fp8,PassThrough,PassThrough,ConvInvscale,Filter1x1Stride1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),8,fp8,fp8,Default,1>"
         // clang-format on
     };
 };

experimental/builder/test/test_ck_factory_grouped_convolution_forward_dynamic_op.cpp

@@ -85,9 +85,9 @@ struct DyOp_F32_2
 
     constexpr static auto expected = {
         // clang-format off
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<2,NHWGC,GKYXC,EmptyTuple,NHWGK,fp32,fp32,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,DynamicUnaryOp,Default,MNKPadding,1,256,128,128,16,4,4,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,4,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,4,1,1,1,Seq(1,16,1,16),4,fp32,fp32,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<2,NHWGC,GKYXC,EmptyTuple,NHWGK,fp32,fp32,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,DynamicUnaryOp,Default,MNKPadding,1,64,64,32,16,4,4,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,4,4,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,4,4,1,1,1,Seq(1,8,1,8),1,fp32,fp32,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<2,NHWGC,GKYXC,EmptyTuple,NHWGK,fp32,fp32,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,DynamicUnaryOp,Default,MNKPadding,1,64,64,64,16,4,4,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,4,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,4,1,1,1,Seq(1,8,1,8),1,fp32,fp32,Default,1>"
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<2,NHWGC,GKYXC,EmptyTuple,NHWGK,fp32,fp32,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,DynamicUnaryOp,Default,MNKPadding,1,256,128,128,16,4,4,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,4,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,4,true,1,1,Seq(1,16,1,16),4,fp32,fp32,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<2,NHWGC,GKYXC,EmptyTuple,NHWGK,fp32,fp32,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,DynamicUnaryOp,Default,MNKPadding,1,64,64,32,16,4,4,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,4,4,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,4,4,true,1,1,Seq(1,8,1,8),1,fp32,fp32,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<2,NHWGC,GKYXC,EmptyTuple,NHWGK,fp32,fp32,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,DynamicUnaryOp,Default,MNKPadding,1,64,64,64,16,4,4,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,4,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,4,true,1,1,Seq(1,8,1,8),1,fp32,fp32,Default,1>"
         // clang-format on
     };
 };
@@ -98,9 +98,9 @@ struct DyOp_F32_3
 
     constexpr static auto expected = {
         // clang-format off
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp32,fp32,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,DynamicUnaryOp,Default,MNKPadding,1,256,128,128,16,4,4,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,4,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,4,1,1,1,Seq(1,16,1,16),4,fp32,fp32,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp32,fp32,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,DynamicUnaryOp,Default,MNKPadding,1,64,64,32,16,4,4,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,4,4,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,4,4,1,1,1,Seq(1,8,1,8),1,fp32,fp32,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp32,fp32,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,DynamicUnaryOp,Default,MNKPadding,1,64,64,64,16,4,4,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,4,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,4,1,1,1,Seq(1,8,1,8),1,fp32,fp32,Default,1>"
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp32,fp32,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,DynamicUnaryOp,Default,MNKPadding,1,256,128,128,16,4,4,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,4,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,4,true,1,1,Seq(1,16,1,16),4,fp32,fp32,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp32,fp32,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,DynamicUnaryOp,Default,MNKPadding,1,64,64,32,16,4,4,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,4,4,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,4,4,true,1,1,Seq(1,8,1,8),1,fp32,fp32,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp32,fp32,fp32,fp32,EmptyTuple,fp32,PassThrough,PassThrough,DynamicUnaryOp,Default,MNKPadding,1,64,64,64,16,4,4,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,4,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,4,true,1,1,Seq(1,8,1,8),1,fp32,fp32,Default,1>"
         // clang-format on
     };
 };
@@ -111,9 +111,9 @@ struct DyOp_F16_2
 
     constexpr static auto expected = {
         // clang-format off
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<2,NHWGC,GKYXC,EmptyTuple,NHWGK,fp16,fp16,fp32,fp16,EmptyTuple,fp16,PassThrough,PassThrough,DynamicUnaryOp,Default,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,32,1,8),8,fp16,fp16,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<2,NHWGC,GKYXC,EmptyTuple,NHWGK,fp16,fp16,fp32,fp16,EmptyTuple,fp16,PassThrough,PassThrough,DynamicUnaryOp,Default,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),1,fp16,fp16,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<2,NHWGC,GKYXC,EmptyTuple,NHWGK,fp16,fp16,fp32,fp16,EmptyTuple,fp16,PassThrough,PassThrough,DynamicUnaryOp,Default,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,16,1,4),1,fp16,fp16,Default,1>"
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<2,NHWGC,GKYXC,EmptyTuple,NHWGK,fp16,fp16,fp32,fp16,EmptyTuple,fp16,PassThrough,PassThrough,DynamicUnaryOp,Default,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,32,1,8),8,fp16,fp16,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<2,NHWGC,GKYXC,EmptyTuple,NHWGK,fp16,fp16,fp32,fp16,EmptyTuple,fp16,PassThrough,PassThrough,DynamicUnaryOp,Default,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),1,fp16,fp16,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<2,NHWGC,GKYXC,EmptyTuple,NHWGK,fp16,fp16,fp32,fp16,EmptyTuple,fp16,PassThrough,PassThrough,DynamicUnaryOp,Default,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,16,1,4),1,fp16,fp16,Default,1>"
         // clang-format on
     };
 };
@@ -124,9 +124,9 @@ struct DyOp_F16_3
 
     constexpr static auto expected = {
         // clang-format off
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp16,fp16,fp32,fp16,EmptyTuple,fp16,PassThrough,PassThrough,DynamicUnaryOp,Default,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,32,1,8),8,fp16,fp16,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp16,fp16,fp32,fp16,EmptyTuple,fp16,PassThrough,PassThrough,DynamicUnaryOp,Default,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),1,fp16,fp16,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp16,fp16,fp32,fp16,EmptyTuple,fp16,PassThrough,PassThrough,DynamicUnaryOp,Default,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,16,1,4),1,fp16,fp16,Default,1>"
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp16,fp16,fp32,fp16,EmptyTuple,fp16,PassThrough,PassThrough,DynamicUnaryOp,Default,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,32,1,8),8,fp16,fp16,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp16,fp16,fp32,fp16,EmptyTuple,fp16,PassThrough,PassThrough,DynamicUnaryOp,Default,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),1,fp16,fp16,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,fp16,fp16,fp32,fp16,EmptyTuple,fp16,PassThrough,PassThrough,DynamicUnaryOp,Default,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,16,1,4),1,fp16,fp16,Default,1>"
         // clang-format on
     };
 };
@@ -137,9 +137,9 @@ struct DyOp_BF16_2
 
     constexpr static auto expected = {
         // clang-format off
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<2,NHWGC,GKYXC,EmptyTuple,NHWGK,bf16,bf16,fp32,bf16,EmptyTuple,bf16,PassThrough,PassThrough,DynamicUnaryOp,Default,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,32,1,8),8,bf16,bf16,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<2,NHWGC,GKYXC,EmptyTuple,NHWGK,bf16,bf16,fp32,bf16,EmptyTuple,bf16,PassThrough,PassThrough,DynamicUnaryOp,Default,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),1,bf16,bf16,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<2,NHWGC,GKYXC,EmptyTuple,NHWGK,bf16,bf16,fp32,bf16,EmptyTuple,bf16,PassThrough,PassThrough,DynamicUnaryOp,Default,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,16,1,4),1,bf16,bf16,Default,1>"
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<2,NHWGC,GKYXC,EmptyTuple,NHWGK,bf16,bf16,fp32,bf16,EmptyTuple,bf16,PassThrough,PassThrough,DynamicUnaryOp,Default,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,32,1,8),8,bf16,bf16,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<2,NHWGC,GKYXC,EmptyTuple,NHWGK,bf16,bf16,fp32,bf16,EmptyTuple,bf16,PassThrough,PassThrough,DynamicUnaryOp,Default,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),1,bf16,bf16,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<2,NHWGC,GKYXC,EmptyTuple,NHWGK,bf16,bf16,fp32,bf16,EmptyTuple,bf16,PassThrough,PassThrough,DynamicUnaryOp,Default,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,16,1,4),1,bf16,bf16,Default,1>"
         // clang-format on
     };
 };
@@ -150,9 +150,9 @@ struct DyOp_BF16_3
 
     constexpr static auto expected = {
         // clang-format off
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf16,bf16,fp32,bf16,EmptyTuple,bf16,PassThrough,PassThrough,DynamicUnaryOp,Default,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,32,1,8),8,bf16,bf16,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf16,bf16,fp32,bf16,EmptyTuple,bf16,PassThrough,PassThrough,DynamicUnaryOp,Default,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),1,bf16,bf16,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf16,bf16,fp32,bf16,EmptyTuple,bf16,PassThrough,PassThrough,DynamicUnaryOp,Default,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,16,1,4),1,bf16,bf16,Default,1>"
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf16,bf16,fp32,bf16,EmptyTuple,bf16,PassThrough,PassThrough,DynamicUnaryOp,Default,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,32,1,8),8,bf16,bf16,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf16,bf16,fp32,bf16,EmptyTuple,bf16,PassThrough,PassThrough,DynamicUnaryOp,Default,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),1,bf16,bf16,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,bf16,bf16,fp32,bf16,EmptyTuple,bf16,PassThrough,PassThrough,DynamicUnaryOp,Default,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,16,1,4),1,bf16,bf16,Default,1>"
         // clang-format on
     };
 };
@@ -163,9 +163,9 @@ struct DyOp_INT8_2
 
     constexpr static auto expected = {
         // clang-format off
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<2,NHWGC,GKYXC,EmptyTuple,NHWGK,s8,s8,s32,s8,EmptyTuple,s8,PassThrough,PassThrough,DynamicUnaryOp,Default,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,32,1,8),8,s8,s8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<2,NHWGC,GKYXC,EmptyTuple,NHWGK,s8,s8,s32,s8,EmptyTuple,s8,PassThrough,PassThrough,DynamicUnaryOp,Default,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),1,s8,s8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<2,NHWGC,GKYXC,EmptyTuple,NHWGK,s8,s8,s32,s8,EmptyTuple,s8,PassThrough,PassThrough,DynamicUnaryOp,Default,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,16,1,4),1,s8,s8,Default,1>"
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<2,NHWGC,GKYXC,EmptyTuple,NHWGK,s8,s8,s32,s8,EmptyTuple,s8,PassThrough,PassThrough,DynamicUnaryOp,Default,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,32,1,8),8,s8,s8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<2,NHWGC,GKYXC,EmptyTuple,NHWGK,s8,s8,s32,s8,EmptyTuple,s8,PassThrough,PassThrough,DynamicUnaryOp,Default,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),1,s8,s8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<2,NHWGC,GKYXC,EmptyTuple,NHWGK,s8,s8,s32,s8,EmptyTuple,s8,PassThrough,PassThrough,DynamicUnaryOp,Default,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,16,1,4),1,s8,s8,Default,1>"
         // clang-format on
     };
 };
@@ -176,9 +176,9 @@ struct DyOp_INT8_3
 
     constexpr static auto expected = {
         // clang-format off
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,s8,s8,s32,s8,EmptyTuple,s8,PassThrough,PassThrough,DynamicUnaryOp,Default,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,32,1,8),8,s8,s8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,s8,s8,s32,s8,EmptyTuple,s8,PassThrough,PassThrough,DynamicUnaryOp,Default,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),1,s8,s8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,s8,s8,s32,s8,EmptyTuple,s8,PassThrough,PassThrough,DynamicUnaryOp,Default,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,16,1,4),1,s8,s8,Default,1>"
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,s8,s8,s32,s8,EmptyTuple,s8,PassThrough,PassThrough,DynamicUnaryOp,Default,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,32,1,8),8,s8,s8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,s8,s8,s32,s8,EmptyTuple,s8,PassThrough,PassThrough,DynamicUnaryOp,Default,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),1,s8,s8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,s8,s8,s32,s8,EmptyTuple,s8,PassThrough,PassThrough,DynamicUnaryOp,Default,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,16,1,4),1,s8,s8,Default,1>"
         // clang-format on
     };
 };

experimental/builder/test/test_ck_factory_grouped_convolution_forward_scaleadd_ab.cpp

@@ -53,18 +53,18 @@ struct F32
 
     constexpr static auto expected = {
         // clang-format off
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(fp32,fp32),Tuple(fp32,fp32),fp32,fp32,EmptyTuple,fp32,ScaleAdd,ScaleAdd,PassThrough,Default,MNKPadding,1,256,128,128,16,4,4,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,1,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,1,1,1,1,Seq(1,16,1,16),4,fp32,fp32,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(fp32,fp32),Tuple(fp32,fp32),fp32,fp32,EmptyTuple,fp32,ScaleAdd,ScaleAdd,PassThrough,Default,MNKPadding,1,256,256,128,16,4,4,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,4,4,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,4,4,1,1,1,Seq(1,16,1,16),4,fp32,fp32,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(fp32,fp32),Tuple(fp32,fp32),fp32,fp32,EmptyTuple,fp32,ScaleAdd,ScaleAdd,PassThrough,Default,MNKPadding,1,64,64,32,16,4,4,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,4,4,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,4,4,1,1,1,Seq(1,8,1,8),1,fp32,fp32,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(fp32,fp32),Tuple(fp32,fp32),fp32,fp32,EmptyTuple,fp32,ScaleAdd,ScaleAdd,PassThrough,Default,MNKPadding,1,64,64,64,16,4,4,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,1,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,1,1,1,1,Seq(1,8,1,8),1,fp32,fp32,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(fp32,fp32),Tuple(fp32,fp32),fp32,fp32,EmptyTuple,fp32,ScaleAdd,ScaleAdd,PassThrough,Filter1x1Pad0,MNKPadding,1,256,128,128,16,4,4,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,1,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,1,1,1,1,Seq(1,16,1,16),4,fp32,fp32,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(fp32,fp32),Tuple(fp32,fp32),fp32,fp32,EmptyTuple,fp32,ScaleAdd,ScaleAdd,PassThrough,Filter1x1Pad0,MNKPadding,1,256,256,128,16,4,4,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,4,4,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,4,4,1,1,1,Seq(1,16,1,16),4,fp32,fp32,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(fp32,fp32),Tuple(fp32,fp32),fp32,fp32,EmptyTuple,fp32,ScaleAdd,ScaleAdd,PassThrough,Filter1x1Pad0,MNKPadding,1,64,64,32,16,4,4,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,4,4,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,4,4,1,1,1,Seq(1,8,1,8),1,fp32,fp32,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(fp32,fp32),Tuple(fp32,fp32),fp32,fp32,EmptyTuple,fp32,ScaleAdd,ScaleAdd,PassThrough,Filter1x1Pad0,MNKPadding,1,64,64,64,16,4,4,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,1,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,1,1,1,1,Seq(1,8,1,8),1,fp32,fp32,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(fp32,fp32),Tuple(fp32,fp32),fp32,fp32,EmptyTuple,fp32,ScaleAdd,ScaleAdd,PassThrough,Filter1x1Stride1Pad0,MNKPadding,1,256,128,128,16,4,4,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,1,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,1,1,1,1,Seq(1,16,1,16),4,fp32,fp32,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(fp32,fp32),Tuple(fp32,fp32),fp32,fp32,EmptyTuple,fp32,ScaleAdd,ScaleAdd,PassThrough,Filter1x1Stride1Pad0,MNKPadding,1,256,256,128,16,4,4,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,4,4,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,4,4,1,1,1,Seq(1,16,1,16),4,fp32,fp32,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(fp32,fp32),Tuple(fp32,fp32),fp32,fp32,EmptyTuple,fp32,ScaleAdd,ScaleAdd,PassThrough,Filter1x1Stride1Pad0,MNKPadding,1,64,64,32,16,4,4,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,4,4,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,4,4,1,1,1,Seq(1,8,1,8),1,fp32,fp32,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(fp32,fp32),Tuple(fp32,fp32),fp32,fp32,EmptyTuple,fp32,ScaleAdd,ScaleAdd,PassThrough,Filter1x1Stride1Pad0,MNKPadding,1,64,64,64,16,4,4,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,1,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,1,1,1,1,Seq(1,8,1,8),1,fp32,fp32,Default,1>"
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(fp32,fp32),Tuple(fp32,fp32),fp32,fp32,EmptyTuple,fp32,ScaleAdd,ScaleAdd,PassThrough,Default,MNKPadding,1,256,128,128,16,4,4,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,1,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,1,true,1,1,Seq(1,16,1,16),4,fp32,fp32,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(fp32,fp32),Tuple(fp32,fp32),fp32,fp32,EmptyTuple,fp32,ScaleAdd,ScaleAdd,PassThrough,Default,MNKPadding,1,256,256,128,16,4,4,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,4,4,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,4,4,true,1,1,Seq(1,16,1,16),4,fp32,fp32,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(fp32,fp32),Tuple(fp32,fp32),fp32,fp32,EmptyTuple,fp32,ScaleAdd,ScaleAdd,PassThrough,Default,MNKPadding,1,64,64,32,16,4,4,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,4,4,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,4,4,true,1,1,Seq(1,8,1,8),1,fp32,fp32,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(fp32,fp32),Tuple(fp32,fp32),fp32,fp32,EmptyTuple,fp32,ScaleAdd,ScaleAdd,PassThrough,Default,MNKPadding,1,64,64,64,16,4,4,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,1,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,1,true,1,1,Seq(1,8,1,8),1,fp32,fp32,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(fp32,fp32),Tuple(fp32,fp32),fp32,fp32,EmptyTuple,fp32,ScaleAdd,ScaleAdd,PassThrough,Filter1x1Pad0,MNKPadding,1,256,128,128,16,4,4,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,1,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,1,true,1,1,Seq(1,16,1,16),4,fp32,fp32,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(fp32,fp32),Tuple(fp32,fp32),fp32,fp32,EmptyTuple,fp32,ScaleAdd,ScaleAdd,PassThrough,Filter1x1Pad0,MNKPadding,1,256,256,128,16,4,4,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,4,4,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,4,4,true,1,1,Seq(1,16,1,16),4,fp32,fp32,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(fp32,fp32),Tuple(fp32,fp32),fp32,fp32,EmptyTuple,fp32,ScaleAdd,ScaleAdd,PassThrough,Filter1x1Pad0,MNKPadding,1,64,64,32,16,4,4,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,4,4,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,4,4,true,1,1,Seq(1,8,1,8),1,fp32,fp32,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(fp32,fp32),Tuple(fp32,fp32),fp32,fp32,EmptyTuple,fp32,ScaleAdd,ScaleAdd,PassThrough,Filter1x1Pad0,MNKPadding,1,64,64,64,16,4,4,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,1,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,1,true,1,1,Seq(1,8,1,8),1,fp32,fp32,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(fp32,fp32),Tuple(fp32,fp32),fp32,fp32,EmptyTuple,fp32,ScaleAdd,ScaleAdd,PassThrough,Filter1x1Stride1Pad0,MNKPadding,1,256,128,128,16,4,4,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,1,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,1,true,1,1,Seq(1,16,1,16),4,fp32,fp32,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(fp32,fp32),Tuple(fp32,fp32),fp32,fp32,EmptyTuple,fp32,ScaleAdd,ScaleAdd,PassThrough,Filter1x1Stride1Pad0,MNKPadding,1,256,256,128,16,4,4,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,4,4,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,4,4,true,1,1,Seq(1,16,1,16),4,fp32,fp32,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(fp32,fp32),Tuple(fp32,fp32),fp32,fp32,EmptyTuple,fp32,ScaleAdd,ScaleAdd,PassThrough,Filter1x1Stride1Pad0,MNKPadding,1,64,64,32,16,4,4,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,4,4,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,4,4,true,1,1,Seq(1,8,1,8),1,fp32,fp32,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(fp32,fp32),Tuple(fp32,fp32),fp32,fp32,EmptyTuple,fp32,ScaleAdd,ScaleAdd,PassThrough,Filter1x1Stride1Pad0,MNKPadding,1,64,64,64,16,4,4,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,1,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,1,true,1,1,Seq(1,8,1,8),1,fp32,fp32,Default,1>"
         // clang-format on
     };
 };
@@ -75,18 +75,18 @@ struct F16
 
     constexpr static auto expected = {
         // clang-format off
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(fp16,fp16),Tuple(fp16,fp16),fp32,fp16,EmptyTuple,fp16,ScaleAdd,ScaleAdd,PassThrough,Default,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,1,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,1,1,1,1,Seq(1,32,1,8),8,fp16,fp16,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(fp16,fp16),Tuple(fp16,fp16),fp32,fp16,EmptyTuple,fp16,ScaleAdd,ScaleAdd,PassThrough,Default,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp16,fp16,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(fp16,fp16),Tuple(fp16,fp16),fp32,fp16,EmptyTuple,fp16,ScaleAdd,ScaleAdd,PassThrough,Default,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),1,fp16,fp16,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(fp16,fp16),Tuple(fp16,fp16),fp32,fp16,EmptyTuple,fp16,ScaleAdd,ScaleAdd,PassThrough,Default,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,1,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,1,1,1,1,Seq(1,16,1,4),1,fp16,fp16,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(fp16,fp16),Tuple(fp16,fp16),fp32,fp16,EmptyTuple,fp16,ScaleAdd,ScaleAdd,PassThrough,Filter1x1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,1,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,1,1,1,1,Seq(1,32,1,8),8,fp16,fp16,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(fp16,fp16),Tuple(fp16,fp16),fp32,fp16,EmptyTuple,fp16,ScaleAdd,ScaleAdd,PassThrough,Filter1x1Pad0,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp16,fp16,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(fp16,fp16),Tuple(fp16,fp16),fp32,fp16,EmptyTuple,fp16,ScaleAdd,ScaleAdd,PassThrough,Filter1x1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),1,fp16,fp16,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(fp16,fp16),Tuple(fp16,fp16),fp32,fp16,EmptyTuple,fp16,ScaleAdd,ScaleAdd,PassThrough,Filter1x1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,1,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,1,1,1,1,Seq(1,16,1,4),1,fp16,fp16,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(fp16,fp16),Tuple(fp16,fp16),fp32,fp16,EmptyTuple,fp16,ScaleAdd,ScaleAdd,PassThrough,Filter1x1Stride1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,1,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,1,1,1,1,Seq(1,32,1,8),8,fp16,fp16,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(fp16,fp16),Tuple(fp16,fp16),fp32,fp16,EmptyTuple,fp16,ScaleAdd,ScaleAdd,PassThrough,Filter1x1Stride1Pad0,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp16,fp16,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(fp16,fp16),Tuple(fp16,fp16),fp32,fp16,EmptyTuple,fp16,ScaleAdd,ScaleAdd,PassThrough,Filter1x1Stride1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),1,fp16,fp16,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(fp16,fp16),Tuple(fp16,fp16),fp32,fp16,EmptyTuple,fp16,ScaleAdd,ScaleAdd,PassThrough,Filter1x1Stride1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,1,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,1,1,1,1,Seq(1,16,1,4),1,fp16,fp16,Default,1>"
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(fp16,fp16),Tuple(fp16,fp16),fp32,fp16,EmptyTuple,fp16,ScaleAdd,ScaleAdd,PassThrough,Default,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,1,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,1,true,1,1,Seq(1,32,1,8),8,fp16,fp16,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(fp16,fp16),Tuple(fp16,fp16),fp32,fp16,EmptyTuple,fp16,ScaleAdd,ScaleAdd,PassThrough,Default,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp16,fp16,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(fp16,fp16),Tuple(fp16,fp16),fp32,fp16,EmptyTuple,fp16,ScaleAdd,ScaleAdd,PassThrough,Default,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),1,fp16,fp16,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(fp16,fp16),Tuple(fp16,fp16),fp32,fp16,EmptyTuple,fp16,ScaleAdd,ScaleAdd,PassThrough,Default,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,1,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,1,true,1,1,Seq(1,16,1,4),1,fp16,fp16,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(fp16,fp16),Tuple(fp16,fp16),fp32,fp16,EmptyTuple,fp16,ScaleAdd,ScaleAdd,PassThrough,Filter1x1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,1,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,1,true,1,1,Seq(1,32,1,8),8,fp16,fp16,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(fp16,fp16),Tuple(fp16,fp16),fp32,fp16,EmptyTuple,fp16,ScaleAdd,ScaleAdd,PassThrough,Filter1x1Pad0,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp16,fp16,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(fp16,fp16),Tuple(fp16,fp16),fp32,fp16,EmptyTuple,fp16,ScaleAdd,ScaleAdd,PassThrough,Filter1x1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),1,fp16,fp16,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(fp16,fp16),Tuple(fp16,fp16),fp32,fp16,EmptyTuple,fp16,ScaleAdd,ScaleAdd,PassThrough,Filter1x1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,1,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,1,true,1,1,Seq(1,16,1,4),1,fp16,fp16,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(fp16,fp16),Tuple(fp16,fp16),fp32,fp16,EmptyTuple,fp16,ScaleAdd,ScaleAdd,PassThrough,Filter1x1Stride1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,1,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,1,true,1,1,Seq(1,32,1,8),8,fp16,fp16,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(fp16,fp16),Tuple(fp16,fp16),fp32,fp16,EmptyTuple,fp16,ScaleAdd,ScaleAdd,PassThrough,Filter1x1Stride1Pad0,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp16,fp16,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(fp16,fp16),Tuple(fp16,fp16),fp32,fp16,EmptyTuple,fp16,ScaleAdd,ScaleAdd,PassThrough,Filter1x1Stride1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),1,fp16,fp16,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(fp16,fp16),Tuple(fp16,fp16),fp32,fp16,EmptyTuple,fp16,ScaleAdd,ScaleAdd,PassThrough,Filter1x1Stride1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,1,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,1,true,1,1,Seq(1,16,1,4),1,fp16,fp16,Default,1>"
         // clang-format on
     };
 };
@@ -97,18 +97,18 @@ struct BF16
 
     constexpr static auto expected = {
         // clang-format off
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(bf16,bf16),Tuple(bf16,bf16),fp32,bf16,EmptyTuple,bf16,ScaleAdd,ScaleAdd,PassThrough,Default,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,1,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,1,1,1,1,Seq(1,32,1,8),8,bf16,bf16,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(bf16,bf16),Tuple(bf16,bf16),fp32,bf16,EmptyTuple,bf16,ScaleAdd,ScaleAdd,PassThrough,Default,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,bf16,bf16,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(bf16,bf16),Tuple(bf16,bf16),fp32,bf16,EmptyTuple,bf16,ScaleAdd,ScaleAdd,PassThrough,Default,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),1,bf16,bf16,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(bf16,bf16),Tuple(bf16,bf16),fp32,bf16,EmptyTuple,bf16,ScaleAdd,ScaleAdd,PassThrough,Default,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,1,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,1,1,1,1,Seq(1,16,1,4),1,bf16,bf16,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(bf16,bf16),Tuple(bf16,bf16),fp32,bf16,EmptyTuple,bf16,ScaleAdd,ScaleAdd,PassThrough,Filter1x1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,1,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,1,1,1,1,Seq(1,32,1,8),8,bf16,bf16,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(bf16,bf16),Tuple(bf16,bf16),fp32,bf16,EmptyTuple,bf16,ScaleAdd,ScaleAdd,PassThrough,Filter1x1Pad0,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,bf16,bf16,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(bf16,bf16),Tuple(bf16,bf16),fp32,bf16,EmptyTuple,bf16,ScaleAdd,ScaleAdd,PassThrough,Filter1x1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),1,bf16,bf16,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(bf16,bf16),Tuple(bf16,bf16),fp32,bf16,EmptyTuple,bf16,ScaleAdd,ScaleAdd,PassThrough,Filter1x1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,1,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,1,1,1,1,Seq(1,16,1,4),1,bf16,bf16,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(bf16,bf16),Tuple(bf16,bf16),fp32,bf16,EmptyTuple,bf16,ScaleAdd,ScaleAdd,PassThrough,Filter1x1Stride1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,1,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,1,1,1,1,Seq(1,32,1,8),8,bf16,bf16,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(bf16,bf16),Tuple(bf16,bf16),fp32,bf16,EmptyTuple,bf16,ScaleAdd,ScaleAdd,PassThrough,Filter1x1Stride1Pad0,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,bf16,bf16,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(bf16,bf16),Tuple(bf16,bf16),fp32,bf16,EmptyTuple,bf16,ScaleAdd,ScaleAdd,PassThrough,Filter1x1Stride1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),1,bf16,bf16,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(bf16,bf16),Tuple(bf16,bf16),fp32,bf16,EmptyTuple,bf16,ScaleAdd,ScaleAdd,PassThrough,Filter1x1Stride1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,1,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,1,1,1,1,Seq(1,16,1,4),1,bf16,bf16,Default,1>"
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(bf16,bf16),Tuple(bf16,bf16),fp32,bf16,EmptyTuple,bf16,ScaleAdd,ScaleAdd,PassThrough,Default,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,1,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,1,true,1,1,Seq(1,32,1,8),8,bf16,bf16,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(bf16,bf16),Tuple(bf16,bf16),fp32,bf16,EmptyTuple,bf16,ScaleAdd,ScaleAdd,PassThrough,Default,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,bf16,bf16,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(bf16,bf16),Tuple(bf16,bf16),fp32,bf16,EmptyTuple,bf16,ScaleAdd,ScaleAdd,PassThrough,Default,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),1,bf16,bf16,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(bf16,bf16),Tuple(bf16,bf16),fp32,bf16,EmptyTuple,bf16,ScaleAdd,ScaleAdd,PassThrough,Default,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,1,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,1,true,1,1,Seq(1,16,1,4),1,bf16,bf16,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(bf16,bf16),Tuple(bf16,bf16),fp32,bf16,EmptyTuple,bf16,ScaleAdd,ScaleAdd,PassThrough,Filter1x1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,1,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,1,true,1,1,Seq(1,32,1,8),8,bf16,bf16,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(bf16,bf16),Tuple(bf16,bf16),fp32,bf16,EmptyTuple,bf16,ScaleAdd,ScaleAdd,PassThrough,Filter1x1Pad0,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,bf16,bf16,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(bf16,bf16),Tuple(bf16,bf16),fp32,bf16,EmptyTuple,bf16,ScaleAdd,ScaleAdd,PassThrough,Filter1x1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),1,bf16,bf16,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(bf16,bf16),Tuple(bf16,bf16),fp32,bf16,EmptyTuple,bf16,ScaleAdd,ScaleAdd,PassThrough,Filter1x1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,1,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,1,true,1,1,Seq(1,16,1,4),1,bf16,bf16,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(bf16,bf16),Tuple(bf16,bf16),fp32,bf16,EmptyTuple,bf16,ScaleAdd,ScaleAdd,PassThrough,Filter1x1Stride1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,1,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,1,true,1,1,Seq(1,32,1,8),8,bf16,bf16,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(bf16,bf16),Tuple(bf16,bf16),fp32,bf16,EmptyTuple,bf16,ScaleAdd,ScaleAdd,PassThrough,Filter1x1Stride1Pad0,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,bf16,bf16,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(bf16,bf16),Tuple(bf16,bf16),fp32,bf16,EmptyTuple,bf16,ScaleAdd,ScaleAdd,PassThrough,Filter1x1Stride1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),1,bf16,bf16,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(bf16,bf16),Tuple(bf16,bf16),fp32,bf16,EmptyTuple,bf16,ScaleAdd,ScaleAdd,PassThrough,Filter1x1Stride1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,1,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,1,true,1,1,Seq(1,16,1,4),1,bf16,bf16,Default,1>"
         // clang-format on
     };
 };
@@ -119,18 +119,18 @@ struct S8
 
     constexpr static auto expected = {
         // clang-format off
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(s8,s8),Tuple(s8,s8),s32,s8,EmptyTuple,s8,ScaleAdd,ScaleAdd,PassThrough,Default,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,1,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,1,1,1,1,Seq(1,32,1,8),8,s8,s8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(s8,s8),Tuple(s8,s8),s32,s8,EmptyTuple,s8,ScaleAdd,ScaleAdd,PassThrough,Default,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,s8,s8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(s8,s8),Tuple(s8,s8),s32,s8,EmptyTuple,s8,ScaleAdd,ScaleAdd,PassThrough,Default,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),1,s8,s8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(s8,s8),Tuple(s8,s8),s32,s8,EmptyTuple,s8,ScaleAdd,ScaleAdd,PassThrough,Default,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,1,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,1,1,1,1,Seq(1,16,1,4),1,s8,s8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(s8,s8),Tuple(s8,s8),s32,s8,EmptyTuple,s8,ScaleAdd,ScaleAdd,PassThrough,Filter1x1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,1,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,1,1,1,1,Seq(1,32,1,8),8,s8,s8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(s8,s8),Tuple(s8,s8),s32,s8,EmptyTuple,s8,ScaleAdd,ScaleAdd,PassThrough,Filter1x1Pad0,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,s8,s8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(s8,s8),Tuple(s8,s8),s32,s8,EmptyTuple,s8,ScaleAdd,ScaleAdd,PassThrough,Filter1x1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),1,s8,s8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(s8,s8),Tuple(s8,s8),s32,s8,EmptyTuple,s8,ScaleAdd,ScaleAdd,PassThrough,Filter1x1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,1,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,1,1,1,1,Seq(1,16,1,4),1,s8,s8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(s8,s8),Tuple(s8,s8),s32,s8,EmptyTuple,s8,ScaleAdd,ScaleAdd,PassThrough,Filter1x1Stride1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,1,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,1,1,1,1,Seq(1,32,1,8),8,s8,s8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(s8,s8),Tuple(s8,s8),s32,s8,EmptyTuple,s8,ScaleAdd,ScaleAdd,PassThrough,Filter1x1Stride1Pad0,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,s8,s8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(s8,s8),Tuple(s8,s8),s32,s8,EmptyTuple,s8,ScaleAdd,ScaleAdd,PassThrough,Filter1x1Stride1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),1,s8,s8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(s8,s8),Tuple(s8,s8),s32,s8,EmptyTuple,s8,ScaleAdd,ScaleAdd,PassThrough,Filter1x1Stride1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,1,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,1,1,1,1,Seq(1,16,1,4),1,s8,s8,Default,1>"
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(s8,s8),Tuple(s8,s8),s32,s8,EmptyTuple,s8,ScaleAdd,ScaleAdd,PassThrough,Default,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,1,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,1,true,1,1,Seq(1,32,1,8),8,s8,s8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(s8,s8),Tuple(s8,s8),s32,s8,EmptyTuple,s8,ScaleAdd,ScaleAdd,PassThrough,Default,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,s8,s8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(s8,s8),Tuple(s8,s8),s32,s8,EmptyTuple,s8,ScaleAdd,ScaleAdd,PassThrough,Default,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),1,s8,s8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(s8,s8),Tuple(s8,s8),s32,s8,EmptyTuple,s8,ScaleAdd,ScaleAdd,PassThrough,Default,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,1,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,1,true,1,1,Seq(1,16,1,4),1,s8,s8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(s8,s8),Tuple(s8,s8),s32,s8,EmptyTuple,s8,ScaleAdd,ScaleAdd,PassThrough,Filter1x1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,1,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,1,true,1,1,Seq(1,32,1,8),8,s8,s8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(s8,s8),Tuple(s8,s8),s32,s8,EmptyTuple,s8,ScaleAdd,ScaleAdd,PassThrough,Filter1x1Pad0,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,s8,s8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(s8,s8),Tuple(s8,s8),s32,s8,EmptyTuple,s8,ScaleAdd,ScaleAdd,PassThrough,Filter1x1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),1,s8,s8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(s8,s8),Tuple(s8,s8),s32,s8,EmptyTuple,s8,ScaleAdd,ScaleAdd,PassThrough,Filter1x1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,1,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,1,true,1,1,Seq(1,16,1,4),1,s8,s8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(s8,s8),Tuple(s8,s8),s32,s8,EmptyTuple,s8,ScaleAdd,ScaleAdd,PassThrough,Filter1x1Stride1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,1,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,1,true,1,1,Seq(1,32,1,8),8,s8,s8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(s8,s8),Tuple(s8,s8),s32,s8,EmptyTuple,s8,ScaleAdd,ScaleAdd,PassThrough,Filter1x1Stride1Pad0,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,s8,s8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(s8,s8),Tuple(s8,s8),s32,s8,EmptyTuple,s8,ScaleAdd,ScaleAdd,PassThrough,Filter1x1Stride1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),1,s8,s8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,EmptyTuple,NDHWGK,Tuple(s8,s8),Tuple(s8,s8),s32,s8,EmptyTuple,s8,ScaleAdd,ScaleAdd,PassThrough,Filter1x1Stride1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,1,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,1,true,1,1,Seq(1,16,1,4),1,s8,s8,Default,1>"
         // clang-format on
     };
 };

experimental/builder/test/test_ck_factory_grouped_convolution_forward_scaleadd_scaleadd_relu.cpp

@@ -54,18 +54,18 @@ struct F32
 
     constexpr static auto expected = {
         // clang-format off
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,fp32,fp32,fp32,fp32,Tuple(fp32,fp32),fp32,PassThrough,PassThrough,ScaleAddScaleAddRelu,Default,MNKPadding,1,256,128,128,16,4,4,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,4,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,4,1,1,1,Seq(1,16,1,16),4,fp32,fp32,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,fp32,fp32,fp32,fp32,Tuple(fp32,fp32),fp32,PassThrough,PassThrough,ScaleAddScaleAddRelu,Default,MNKPadding,1,256,256,128,16,4,4,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,4,4,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,4,4,1,1,1,Seq(1,16,1,16),4,fp32,fp32,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,fp32,fp32,fp32,fp32,Tuple(fp32,fp32),fp32,PassThrough,PassThrough,ScaleAddScaleAddRelu,Default,MNKPadding,1,64,64,32,16,4,4,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,4,4,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,4,4,1,1,1,Seq(1,8,1,8),1,fp32,fp32,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,fp32,fp32,fp32,fp32,Tuple(fp32,fp32),fp32,PassThrough,PassThrough,ScaleAddScaleAddRelu,Default,MNKPadding,1,64,64,64,16,4,4,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,4,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,4,1,1,1,Seq(1,8,1,8),1,fp32,fp32,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,fp32,fp32,fp32,fp32,Tuple(fp32,fp32),fp32,PassThrough,PassThrough,ScaleAddScaleAddRelu,Filter1x1Pad0,MNKPadding,1,256,128,128,16,4,4,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,4,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,4,1,1,1,Seq(1,16,1,16),4,fp32,fp32,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,fp32,fp32,fp32,fp32,Tuple(fp32,fp32),fp32,PassThrough,PassThrough,ScaleAddScaleAddRelu,Filter1x1Pad0,MNKPadding,1,256,256,128,16,4,4,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,4,4,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,4,4,1,1,1,Seq(1,16,1,16),4,fp32,fp32,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,fp32,fp32,fp32,fp32,Tuple(fp32,fp32),fp32,PassThrough,PassThrough,ScaleAddScaleAddRelu,Filter1x1Pad0,MNKPadding,1,64,64,32,16,4,4,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,4,4,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,4,4,1,1,1,Seq(1,8,1,8),1,fp32,fp32,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,fp32,fp32,fp32,fp32,Tuple(fp32,fp32),fp32,PassThrough,PassThrough,ScaleAddScaleAddRelu,Filter1x1Pad0,MNKPadding,1,64,64,64,16,4,4,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,4,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,4,1,1,1,Seq(1,8,1,8),1,fp32,fp32,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,fp32,fp32,fp32,fp32,Tuple(fp32,fp32),fp32,PassThrough,PassThrough,ScaleAddScaleAddRelu,Filter1x1Stride1Pad0,MNKPadding,1,256,128,128,16,4,4,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,4,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,4,1,1,1,Seq(1,16,1,16),4,fp32,fp32,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,fp32,fp32,fp32,fp32,Tuple(fp32,fp32),fp32,PassThrough,PassThrough,ScaleAddScaleAddRelu,Filter1x1Stride1Pad0,MNKPadding,1,256,256,128,16,4,4,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,4,4,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,4,4,1,1,1,Seq(1,16,1,16),4,fp32,fp32,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,fp32,fp32,fp32,fp32,Tuple(fp32,fp32),fp32,PassThrough,PassThrough,ScaleAddScaleAddRelu,Filter1x1Stride1Pad0,MNKPadding,1,64,64,32,16,4,4,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,4,4,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,4,4,1,1,1,Seq(1,8,1,8),1,fp32,fp32,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,fp32,fp32,fp32,fp32,Tuple(fp32,fp32),fp32,PassThrough,PassThrough,ScaleAddScaleAddRelu,Filter1x1Stride1Pad0,MNKPadding,1,64,64,64,16,4,4,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,4,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,4,1,1,1,Seq(1,8,1,8),1,fp32,fp32,Default,1>"
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,fp32,fp32,fp32,fp32,Tuple(fp32,fp32),fp32,PassThrough,PassThrough,ScaleAddScaleAddRelu,Default,MNKPadding,1,256,128,128,16,4,4,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,4,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,4,true,1,1,Seq(1,16,1,16),4,fp32,fp32,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,fp32,fp32,fp32,fp32,Tuple(fp32,fp32),fp32,PassThrough,PassThrough,ScaleAddScaleAddRelu,Default,MNKPadding,1,256,256,128,16,4,4,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,4,4,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,4,4,true,1,1,Seq(1,16,1,16),4,fp32,fp32,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,fp32,fp32,fp32,fp32,Tuple(fp32,fp32),fp32,PassThrough,PassThrough,ScaleAddScaleAddRelu,Default,MNKPadding,1,64,64,32,16,4,4,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,4,4,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,4,4,true,1,1,Seq(1,8,1,8),1,fp32,fp32,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,fp32,fp32,fp32,fp32,Tuple(fp32,fp32),fp32,PassThrough,PassThrough,ScaleAddScaleAddRelu,Default,MNKPadding,1,64,64,64,16,4,4,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,4,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,4,true,1,1,Seq(1,8,1,8),1,fp32,fp32,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,fp32,fp32,fp32,fp32,Tuple(fp32,fp32),fp32,PassThrough,PassThrough,ScaleAddScaleAddRelu,Filter1x1Pad0,MNKPadding,1,256,128,128,16,4,4,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,4,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,4,true,1,1,Seq(1,16,1,16),4,fp32,fp32,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,fp32,fp32,fp32,fp32,Tuple(fp32,fp32),fp32,PassThrough,PassThrough,ScaleAddScaleAddRelu,Filter1x1Pad0,MNKPadding,1,256,256,128,16,4,4,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,4,4,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,4,4,true,1,1,Seq(1,16,1,16),4,fp32,fp32,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,fp32,fp32,fp32,fp32,Tuple(fp32,fp32),fp32,PassThrough,PassThrough,ScaleAddScaleAddRelu,Filter1x1Pad0,MNKPadding,1,64,64,32,16,4,4,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,4,4,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,4,4,true,1,1,Seq(1,8,1,8),1,fp32,fp32,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,fp32,fp32,fp32,fp32,Tuple(fp32,fp32),fp32,PassThrough,PassThrough,ScaleAddScaleAddRelu,Filter1x1Pad0,MNKPadding,1,64,64,64,16,4,4,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,4,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,4,true,1,1,Seq(1,8,1,8),1,fp32,fp32,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,fp32,fp32,fp32,fp32,Tuple(fp32,fp32),fp32,PassThrough,PassThrough,ScaleAddScaleAddRelu,Filter1x1Stride1Pad0,MNKPadding,1,256,128,128,16,4,4,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,4,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,4,true,1,1,Seq(1,16,1,16),4,fp32,fp32,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,fp32,fp32,fp32,fp32,Tuple(fp32,fp32),fp32,PassThrough,PassThrough,ScaleAddScaleAddRelu,Filter1x1Stride1Pad0,MNKPadding,1,256,256,128,16,4,4,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,4,4,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,4,4,true,1,1,Seq(1,16,1,16),4,fp32,fp32,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,fp32,fp32,fp32,fp32,Tuple(fp32,fp32),fp32,PassThrough,PassThrough,ScaleAddScaleAddRelu,Filter1x1Stride1Pad0,MNKPadding,1,64,64,32,16,4,4,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,4,4,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,4,4,true,1,1,Seq(1,8,1,8),1,fp32,fp32,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,fp32,fp32,fp32,fp32,Tuple(fp32,fp32),fp32,PassThrough,PassThrough,ScaleAddScaleAddRelu,Filter1x1Stride1Pad0,MNKPadding,1,64,64,64,16,4,4,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,4,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,4,true,1,1,Seq(1,8,1,8),1,fp32,fp32,Default,1>"
         // clang-format on
     };
 };
@@ -76,18 +76,18 @@ struct F16
 
     constexpr static auto expected = {
         // clang-format off
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,fp16,fp16,fp32,fp16,Tuple(fp16,fp16),fp16,PassThrough,PassThrough,ScaleAddScaleAddRelu,Default,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,32,1,8),8,fp16,fp16,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,fp16,fp16,fp32,fp16,Tuple(fp16,fp16),fp16,PassThrough,PassThrough,ScaleAddScaleAddRelu,Default,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp16,fp16,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,fp16,fp16,fp32,fp16,Tuple(fp16,fp16),fp16,PassThrough,PassThrough,ScaleAddScaleAddRelu,Default,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),1,fp16,fp16,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,fp16,fp16,fp32,fp16,Tuple(fp16,fp16),fp16,PassThrough,PassThrough,ScaleAddScaleAddRelu,Default,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,16,1,4),1,fp16,fp16,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,fp16,fp16,fp32,fp16,Tuple(fp16,fp16),fp16,PassThrough,PassThrough,ScaleAddScaleAddRelu,Filter1x1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,32,1,8),8,fp16,fp16,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,fp16,fp16,fp32,fp16,Tuple(fp16,fp16),fp16,PassThrough,PassThrough,ScaleAddScaleAddRelu,Filter1x1Pad0,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp16,fp16,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,fp16,fp16,fp32,fp16,Tuple(fp16,fp16),fp16,PassThrough,PassThrough,ScaleAddScaleAddRelu,Filter1x1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),1,fp16,fp16,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,fp16,fp16,fp32,fp16,Tuple(fp16,fp16),fp16,PassThrough,PassThrough,ScaleAddScaleAddRelu,Filter1x1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,16,1,4),1,fp16,fp16,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,fp16,fp16,fp32,fp16,Tuple(fp16,fp16),fp16,PassThrough,PassThrough,ScaleAddScaleAddRelu,Filter1x1Stride1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,32,1,8),8,fp16,fp16,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,fp16,fp16,fp32,fp16,Tuple(fp16,fp16),fp16,PassThrough,PassThrough,ScaleAddScaleAddRelu,Filter1x1Stride1Pad0,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,fp16,fp16,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,fp16,fp16,fp32,fp16,Tuple(fp16,fp16),fp16,PassThrough,PassThrough,ScaleAddScaleAddRelu,Filter1x1Stride1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),1,fp16,fp16,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,fp16,fp16,fp32,fp16,Tuple(fp16,fp16),fp16,PassThrough,PassThrough,ScaleAddScaleAddRelu,Filter1x1Stride1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,16,1,4),1,fp16,fp16,Default,1>"
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,fp16,fp16,fp32,fp16,Tuple(fp16,fp16),fp16,PassThrough,PassThrough,ScaleAddScaleAddRelu,Default,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,32,1,8),8,fp16,fp16,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,fp16,fp16,fp32,fp16,Tuple(fp16,fp16),fp16,PassThrough,PassThrough,ScaleAddScaleAddRelu,Default,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp16,fp16,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,fp16,fp16,fp32,fp16,Tuple(fp16,fp16),fp16,PassThrough,PassThrough,ScaleAddScaleAddRelu,Default,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),1,fp16,fp16,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,fp16,fp16,fp32,fp16,Tuple(fp16,fp16),fp16,PassThrough,PassThrough,ScaleAddScaleAddRelu,Default,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,16,1,4),1,fp16,fp16,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,fp16,fp16,fp32,fp16,Tuple(fp16,fp16),fp16,PassThrough,PassThrough,ScaleAddScaleAddRelu,Filter1x1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,32,1,8),8,fp16,fp16,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,fp16,fp16,fp32,fp16,Tuple(fp16,fp16),fp16,PassThrough,PassThrough,ScaleAddScaleAddRelu,Filter1x1Pad0,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp16,fp16,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,fp16,fp16,fp32,fp16,Tuple(fp16,fp16),fp16,PassThrough,PassThrough,ScaleAddScaleAddRelu,Filter1x1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),1,fp16,fp16,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,fp16,fp16,fp32,fp16,Tuple(fp16,fp16),fp16,PassThrough,PassThrough,ScaleAddScaleAddRelu,Filter1x1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,16,1,4),1,fp16,fp16,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,fp16,fp16,fp32,fp16,Tuple(fp16,fp16),fp16,PassThrough,PassThrough,ScaleAddScaleAddRelu,Filter1x1Stride1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,32,1,8),8,fp16,fp16,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,fp16,fp16,fp32,fp16,Tuple(fp16,fp16),fp16,PassThrough,PassThrough,ScaleAddScaleAddRelu,Filter1x1Stride1Pad0,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,fp16,fp16,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,fp16,fp16,fp32,fp16,Tuple(fp16,fp16),fp16,PassThrough,PassThrough,ScaleAddScaleAddRelu,Filter1x1Stride1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),1,fp16,fp16,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,fp16,fp16,fp32,fp16,Tuple(fp16,fp16),fp16,PassThrough,PassThrough,ScaleAddScaleAddRelu,Filter1x1Stride1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,16,1,4),1,fp16,fp16,Default,1>"
         // clang-format on
     };
 };
@@ -98,18 +98,18 @@ struct BF16
 
     constexpr static auto expected = {
         // clang-format off
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,bf16,bf16,fp32,bf16,Tuple(bf16,bf16),bf16,PassThrough,PassThrough,ScaleAddScaleAddRelu,Default,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,32,1,8),8,bf16,bf16,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,bf16,bf16,fp32,bf16,Tuple(bf16,bf16),bf16,PassThrough,PassThrough,ScaleAddScaleAddRelu,Default,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,bf16,bf16,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,bf16,bf16,fp32,bf16,Tuple(bf16,bf16),bf16,PassThrough,PassThrough,ScaleAddScaleAddRelu,Default,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),1,bf16,bf16,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,bf16,bf16,fp32,bf16,Tuple(bf16,bf16),bf16,PassThrough,PassThrough,ScaleAddScaleAddRelu,Default,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,16,1,4),1,bf16,bf16,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,bf16,bf16,fp32,bf16,Tuple(bf16,bf16),bf16,PassThrough,PassThrough,ScaleAddScaleAddRelu,Filter1x1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,32,1,8),8,bf16,bf16,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,bf16,bf16,fp32,bf16,Tuple(bf16,bf16),bf16,PassThrough,PassThrough,ScaleAddScaleAddRelu,Filter1x1Pad0,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,bf16,bf16,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,bf16,bf16,fp32,bf16,Tuple(bf16,bf16),bf16,PassThrough,PassThrough,ScaleAddScaleAddRelu,Filter1x1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),1,bf16,bf16,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,bf16,bf16,fp32,bf16,Tuple(bf16,bf16),bf16,PassThrough,PassThrough,ScaleAddScaleAddRelu,Filter1x1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,16,1,4),1,bf16,bf16,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,bf16,bf16,fp32,bf16,Tuple(bf16,bf16),bf16,PassThrough,PassThrough,ScaleAddScaleAddRelu,Filter1x1Stride1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,32,1,8),8,bf16,bf16,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,bf16,bf16,fp32,bf16,Tuple(bf16,bf16),bf16,PassThrough,PassThrough,ScaleAddScaleAddRelu,Filter1x1Stride1Pad0,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,bf16,bf16,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,bf16,bf16,fp32,bf16,Tuple(bf16,bf16),bf16,PassThrough,PassThrough,ScaleAddScaleAddRelu,Filter1x1Stride1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),1,bf16,bf16,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,bf16,bf16,fp32,bf16,Tuple(bf16,bf16),bf16,PassThrough,PassThrough,ScaleAddScaleAddRelu,Filter1x1Stride1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,16,1,4),1,bf16,bf16,Default,1>"
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,bf16,bf16,fp32,bf16,Tuple(bf16,bf16),bf16,PassThrough,PassThrough,ScaleAddScaleAddRelu,Default,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,32,1,8),8,bf16,bf16,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,bf16,bf16,fp32,bf16,Tuple(bf16,bf16),bf16,PassThrough,PassThrough,ScaleAddScaleAddRelu,Default,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,bf16,bf16,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,bf16,bf16,fp32,bf16,Tuple(bf16,bf16),bf16,PassThrough,PassThrough,ScaleAddScaleAddRelu,Default,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),1,bf16,bf16,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,bf16,bf16,fp32,bf16,Tuple(bf16,bf16),bf16,PassThrough,PassThrough,ScaleAddScaleAddRelu,Default,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,16,1,4),1,bf16,bf16,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,bf16,bf16,fp32,bf16,Tuple(bf16,bf16),bf16,PassThrough,PassThrough,ScaleAddScaleAddRelu,Filter1x1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,32,1,8),8,bf16,bf16,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,bf16,bf16,fp32,bf16,Tuple(bf16,bf16),bf16,PassThrough,PassThrough,ScaleAddScaleAddRelu,Filter1x1Pad0,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,bf16,bf16,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,bf16,bf16,fp32,bf16,Tuple(bf16,bf16),bf16,PassThrough,PassThrough,ScaleAddScaleAddRelu,Filter1x1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),1,bf16,bf16,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,bf16,bf16,fp32,bf16,Tuple(bf16,bf16),bf16,PassThrough,PassThrough,ScaleAddScaleAddRelu,Filter1x1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,16,1,4),1,bf16,bf16,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,bf16,bf16,fp32,bf16,Tuple(bf16,bf16),bf16,PassThrough,PassThrough,ScaleAddScaleAddRelu,Filter1x1Stride1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,32,1,8),8,bf16,bf16,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,bf16,bf16,fp32,bf16,Tuple(bf16,bf16),bf16,PassThrough,PassThrough,ScaleAddScaleAddRelu,Filter1x1Stride1Pad0,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,bf16,bf16,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,bf16,bf16,fp32,bf16,Tuple(bf16,bf16),bf16,PassThrough,PassThrough,ScaleAddScaleAddRelu,Filter1x1Stride1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),1,bf16,bf16,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,bf16,bf16,fp32,bf16,Tuple(bf16,bf16),bf16,PassThrough,PassThrough,ScaleAddScaleAddRelu,Filter1x1Stride1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,16,1,4),1,bf16,bf16,Default,1>"
         // clang-format on
     };
 };
@@ -120,18 +120,18 @@ struct S8
 
     constexpr static auto expected = {
         // clang-format off
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,s8,s8,s32,s8,Tuple(fp32,fp32),s8,PassThrough,PassThrough,ScaleAddScaleAddRelu,Default,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,32,1,8),8,s8,s8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,s8,s8,s32,s8,Tuple(fp32,fp32),s8,PassThrough,PassThrough,ScaleAddScaleAddRelu,Default,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,s8,s8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,s8,s8,s32,s8,Tuple(fp32,fp32),s8,PassThrough,PassThrough,ScaleAddScaleAddRelu,Default,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),1,s8,s8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,s8,s8,s32,s8,Tuple(fp32,fp32),s8,PassThrough,PassThrough,ScaleAddScaleAddRelu,Default,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,16,1,4),1,s8,s8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,s8,s8,s32,s8,Tuple(fp32,fp32),s8,PassThrough,PassThrough,ScaleAddScaleAddRelu,Filter1x1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,32,1,8),8,s8,s8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,s8,s8,s32,s8,Tuple(fp32,fp32),s8,PassThrough,PassThrough,ScaleAddScaleAddRelu,Filter1x1Pad0,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,s8,s8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,s8,s8,s32,s8,Tuple(fp32,fp32),s8,PassThrough,PassThrough,ScaleAddScaleAddRelu,Filter1x1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),1,s8,s8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,s8,s8,s32,s8,Tuple(fp32,fp32),s8,PassThrough,PassThrough,ScaleAddScaleAddRelu,Filter1x1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,16,1,4),1,s8,s8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,s8,s8,s32,s8,Tuple(fp32,fp32),s8,PassThrough,PassThrough,ScaleAddScaleAddRelu,Filter1x1Stride1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,32,1,8),8,s8,s8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,s8,s8,s32,s8,Tuple(fp32,fp32),s8,PassThrough,PassThrough,ScaleAddScaleAddRelu,Filter1x1Stride1Pad0,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,32,1,8),8,s8,s8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,s8,s8,s32,s8,Tuple(fp32,fp32),s8,PassThrough,PassThrough,ScaleAddScaleAddRelu,Filter1x1Stride1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,4),1,s8,s8,Default,1>",
-        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,s8,s8,s32,s8,Tuple(fp32,fp32),s8,PassThrough,PassThrough,ScaleAddScaleAddRelu,Filter1x1Stride1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,1,1,1,Seq(1,16,1,4),1,s8,s8,Default,1>"
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,s8,s8,s32,s8,Tuple(fp32,fp32),s8,PassThrough,PassThrough,ScaleAddScaleAddRelu,Default,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,32,1,8),8,s8,s8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,s8,s8,s32,s8,Tuple(fp32,fp32),s8,PassThrough,PassThrough,ScaleAddScaleAddRelu,Default,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,s8,s8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,s8,s8,s32,s8,Tuple(fp32,fp32),s8,PassThrough,PassThrough,ScaleAddScaleAddRelu,Default,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),1,s8,s8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,s8,s8,s32,s8,Tuple(fp32,fp32),s8,PassThrough,PassThrough,ScaleAddScaleAddRelu,Default,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,16,1,4),1,s8,s8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,s8,s8,s32,s8,Tuple(fp32,fp32),s8,PassThrough,PassThrough,ScaleAddScaleAddRelu,Filter1x1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,32,1,8),8,s8,s8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,s8,s8,s32,s8,Tuple(fp32,fp32),s8,PassThrough,PassThrough,ScaleAddScaleAddRelu,Filter1x1Pad0,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,s8,s8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,s8,s8,s32,s8,Tuple(fp32,fp32),s8,PassThrough,PassThrough,ScaleAddScaleAddRelu,Filter1x1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),1,s8,s8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,s8,s8,s32,s8,Tuple(fp32,fp32),s8,PassThrough,PassThrough,ScaleAddScaleAddRelu,Filter1x1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,16,1,4),1,s8,s8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,s8,s8,s32,s8,Tuple(fp32,fp32),s8,PassThrough,PassThrough,ScaleAddScaleAddRelu,Filter1x1Stride1Pad0,MNKPadding,1,256,128,128,32,8,8,32,32,2,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,32,1,8),8,s8,s8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,s8,s8,s32,s8,Tuple(fp32,fp32),s8,PassThrough,PassThrough,ScaleAddScaleAddRelu,Filter1x1Stride1Pad0,MNKPadding,1,256,256,128,32,8,8,32,32,4,2,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,64,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,32,1,8),8,s8,s8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,s8,s8,s32,s8,Tuple(fp32,fp32),s8,PassThrough,PassThrough,ScaleAddScaleAddRelu,Filter1x1Stride1Pad0,MNKPadding,1,64,64,32,32,8,8,32,32,2,1,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,4),1,s8,s8,Default,1>",
+        "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<3,NDHWGC,GKZYXC,Tuple(NDHWGK,G_K),NDHWGK,s8,s8,s32,s8,Tuple(fp32,fp32),s8,PassThrough,PassThrough,ScaleAddScaleAddRelu,Filter1x1Stride1Pad0,MNKPadding,1,64,64,64,32,8,8,32,32,2,2,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,Seq(4,16,1),Seq(1,0,2),Seq(1,0,2),2,1,8,true,1,1,Seq(1,16,1,4),1,s8,s8,Default,1>"
         // clang-format on
     };
 };

experimental/builder/test/test_conv_description.cpp

@@ -30,8 +30,8 @@ static_assert(!ckb::TensorOperatorDescriptor<InvalidTensorOp>);
 struct TensorConfig
 {
     ckb::TensorLayout layout;
-    ckb::DataType data_type{ckb::DataType::UNDEFINDED};
-    ckb::DataType compute_type{ckb::DataType::UNDEFINDED};
+    ckb::DataType data_type{ckb::DataType::UNDEFINED_DATA_TYPE};
+    ckb::DataType compute_type{ckb::DataType::UNDEFINED_DATA_TYPE};
 };
 
 struct ConvTensorSimple
@@ -55,39 +55,49 @@ struct ConvTensorWithInvalidOp
 // This includes dimensionality, direction, data layout, and data type.
 struct ConvSignature
 {
+    using enum ckb::DataType;
+    using enum ckb::TensorLayout;
+
     int spatial_dim                      = 2;
-    ckb::DataType data_type              = ckb::DataType::FP16;
-    ckb::DataType accumulation_data_type = ckb::DataType::FP32;
-    ConvTensorSimple input               = {.config = {ckb::TensorLayout::GNHWC}};
-    ConvTensorSimple weight              = {.config = {ckb::TensorLayout::GKYXC}};
-    ConvTensorSimple output              = {.config = {ckb::TensorLayout::GNHWK}};
+    ckb::DataType data_type              = FP16;
+    ckb::DataType accumulation_data_type = FP32;
+    ConvTensorSimple input               = {.config = {GNHWC}};
+    ConvTensorSimple weight              = {.config = {GKYXC}};
+    ConvTensorSimple output              = {.config = {GNHWK}};
 };
 static_assert(ckb::ConvSignatureDescriptor<ConvSignature>);
 
 // Compile time tests for concepts
 struct ConvSignatureWithOptionalParams
 {
+    using enum ckb::DataType;
+    using enum ckb::TensorLayout;
+    using enum ckb::ConvDirection;
+    using enum ckb::ElementwiseOperation;
+
     int spatial_dim                      = 2;
-    ckb::DataType data_type              = ckb::DataType::FP16;
-    ckb::DataType accumulation_data_type = ckb::DataType::FP32;
-    ckb::ConvDirection direction         = ckb::ConvDirection::FORWARD;
+    ckb::DataType data_type              = FP16;
+    ckb::DataType accumulation_data_type = FP32;
+    ckb::ConvDirection direction         = FORWARD;
     ConvTensorWithOp input               = {
-                      .config = {ckb::TensorLayout::GNHWC, ckb::DataType::FP16},
+                      .config = {GNHWC, FP16},
     };
-    ConvTensorWithOp weight = {.config = {ckb::TensorLayout::GKYXC, ckb::DataType::FP16}};
-    ConvTensorWithOp output = {.config    = {ckb::TensorLayout::GNHWK, ckb::DataType::FP16},
-                               .operation = {ckb::ElementwiseOperation::SCALE}};
+    ConvTensorWithOp weight = {.config = {GKYXC, FP16}};
+    ConvTensorWithOp output = {.config = {GNHWK, FP16}, .operation = {SCALE}};
 };
 static_assert(ckb::ConvSignatureDescriptor<ConvSignatureWithOptionalParams>);
 
 struct ConvSignatureWithInvalidOptionalParams
 {
+    using enum ckb::DataType;
+    using enum ckb::TensorLayout;
+
     int spatial_dim                      = 2;
-    ckb::DataType data_type              = ckb::DataType::FP16;
-    ckb::DataType accumulation_data_type = ckb::DataType::FP32;
-    ConvTensorWithInvalidOp input        = {.config = {ckb::TensorLayout::GNHWC}};
-    ConvTensorWithInvalidOp weight       = {.config = {ckb::TensorLayout::GKYXC}};
-    ConvTensorWithInvalidOp output       = {.config = {ckb::TensorLayout::GNHWK}};
+    ckb::DataType data_type              = FP16;
+    ckb::DataType accumulation_data_type = FP32;
+    ConvTensorWithInvalidOp input        = {.config = {GNHWC}};
+    ConvTensorWithInvalidOp weight       = {.config = {GKYXC}};
+    ConvTensorWithInvalidOp output       = {.config = {GNHWK}};
 };
 static_assert(!ckb::ConvSignatureDescriptor<ConvSignatureWithInvalidOptionalParams>);
 

experimental/builder/test/test_fwd_instance_traits.cpp

@@ -262,14 +262,14 @@ TEST(InstanceTraits, V3InstanceStringReturnsCorrectFormat)
                                ",2"             // ABlockTransferSrcVectorDim
                                ",8"             // ABlockTransferSrcScalarPerVector
                                ",8"             // ABlockTransferDstScalarPerVector_AK1
-                               ",1"             // ABlockLdsExtraM
+                               ",true"          // ABlockLdsExtraM
                                ",Seq(4,64,1)"   // BBlockTransferThreadClusterLengths
                                ",Seq(1,0,2)"    // BBlockTransferThreadClusterArrangeOrder
                                ",Seq(1,0,2)"    // BBlockTransferSrcAccessOrder
                                ",2"             // BBlockTransferSrcVectorDim
                                ",8"             // BBlockTransferSrcScalarPerVector
                                ",8"             // BBlockTransferDstScalarPerVector_BK1
-                               ",1"             // BBlockLdsExtraN
+                               ",true"          // BBlockLdsExtraN
                                ",1"             // CShuffleMXdlPerWavePerShuffle
                                ",1"             // CShuffleNXdlPerWavePerShuffle
                                ",Seq(1,32,1,8)" // CDEBlockTransferClusterLengths
@@ -377,14 +377,14 @@ TEST(InstanceTraits, BaseInstanceStringReturnsCorrectFormat)
                                ",2"             // ABlockTransferSrcVectorDim
                                ",8"             // ABlockTransferSrcScalarPerVector
                                ",8"             // ABlockTransferDstScalarPerVector_AK1
-                               ",1"             // ABlockLdsExtraM
+                               ",true"          // ABlockLdsExtraM
                                ",Seq(4,64,1)"   // BBlockTransferThreadClusterLengths
                                ",Seq(1,0,2)"    // BBlockTransferThreadClusterArrangeOrder
                                ",Seq(1,0,2)"    // BBlockTransferSrcAccessOrder
                                ",2"             // BBlockTransferSrcVectorDim
                                ",8"             // BBlockTransferSrcScalarPerVector
                                ",8"             // BBlockTransferDstScalarPerVector_BK1
-                               ",1"             // BBlockLdsExtraN
+                               ",true"          // BBlockLdsExtraN
                                ",1"             // CShuffleMXdlPerWavePerShuffle
                                ",1"             // CShuffleNXdlPerWavePerShuffle
                                ",Seq(1,32,1,8)" // CDEBlockTransferClusterLengths
@@ -492,14 +492,14 @@ TEST(InstanceTraits, LargeTensorInstanceStringReturnsCorrectFormat)
                                ",2"             // ABlockTransferSrcVectorDim
                                ",8"             // ABlockTransferSrcScalarPerVector
                                ",8"             // ABlockTransferDstScalarPerVector_AK1
-                               ",1"             // ABlockLdsExtraM
+                               ",true"          // ABlockLdsExtraM
                                ",Seq(4,64,1)"   // BBlockTransferThreadClusterLengths
                                ",Seq(1,0,2)"    // BBlockTransferThreadClusterArrangeOrder
                                ",Seq(1,0,2)"    // BBlockTransferSrcAccessOrder
                                ",2"             // BBlockTransferSrcVectorDim
                                ",8"             // BBlockTransferSrcScalarPerVector
                                ",8"             // BBlockTransferDstScalarPerVector_BK1
-                               ",1"             // BBlockLdsExtraN
+                               ",true"          // BBlockLdsExtraN
                                ",1"             // CShuffleMXdlPerWavePerShuffle
                                ",1"             // CShuffleNXdlPerWavePerShuffle
                                ",Seq(1,32,1,8)" // CDEBlockTransferClusterLengths

experimental/builder/test/test_instance_string_fwd_grp_conv.cpp

@@ -60,14 +60,14 @@ std::string expected_str = "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle"
                            ",2"             // ABlockTransferSrcVectorDim
                            ",1"             // ABlockTransferSrcScalarPerVector
                            ",8"             // ABlockTransferDstScalarPerVector_AK1
-                           ",1"             // ABlockLdsExtraM
+                           ",true"          // ABlockLdsExtraM
                            ",Seq(4,16,1)"   // BBlockTransferThreadClusterLengths
                            ",Seq(1,0,2)"    // BBlockTransferThreadClusterArrangeOrder
                            ",Seq(1,0,2)"    // BBlockTransferSrcAccessOrder
                            ",2"             // BBlockTransferSrcVectorDim
                            ",1"             // BBlockTransferSrcScalarPerVector
                            ",8"             // BBlockTransferDstScalarPerVector_BK1
-                           ",1"             // BBlockLdsExtraN
+                           ",true"          // BBlockLdsExtraN
                            ",1"             // CShuffleMXdlPerWavePerShuffle
                            ",1"             // CShuffleNXdlPerWavePerShuffle
                            ",Seq(1,16,1,4)" // CDEBlockTransferClusterLengths

experimental/builder/test/test_instance_string_fwd_grp_conv_large_tensor.cpp

@@ -60,14 +60,14 @@ std::string expected_str = "DeviceGroupedConvFwdMultipleD_Xdl_CShuffle_Large_Ten
                            ",2"             // ABlockTransferSrcVectorDim
                            ",1"             // ABlockTransferSrcScalarPerVector
                            ",8"             // ABlockTransferDstScalarPerVector_AK1
-                           ",1"             // ABlockLdsExtraM
+                           ",true"          // ABlockLdsExtraM
                            ",Seq(4,16,1)"   // BBlockTransferThreadClusterLengths
                            ",Seq(1,0,2)"    // BBlockTransferThreadClusterArrangeOrder
                            ",Seq(1,0,2)"    // BBlockTransferSrcAccessOrder
                            ",2"             // BBlockTransferSrcVectorDim
                            ",1"             // BBlockTransferSrcScalarPerVector
                            ",8"             // BBlockTransferDstScalarPerVector_BK1
-                           ",1"             // BBlockLdsExtraN
+                           ",true"          // BBlockLdsExtraN
                            ",1"             // CShuffleMXdlPerWavePerShuffle
                            ",1"             // CShuffleNXdlPerWavePerShuffle
                            ",Seq(1,16,1,4)" // CDEBlockTransferClusterLengths

experimental/builder/test/test_instance_string_fwd_grp_conv_v3.cpp

@@ -60,14 +60,14 @@ std::string expected_str = "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3"
                            ",2"             // ABlockTransferSrcVectorDim
                            ",8"             // ABlockTransferSrcScalarPerVector
                            ",8"             // ABlockTransferDstScalarPerVector_AK1
-                           ",0"             // ABlockLdsExtraM
+                           ",false"         // ABlockLdsExtraM
                            ",Seq(8,32,1)"   // BBlockTransferThreadClusterLengths
                            ",Seq(1,0,2)"    // BBlockTransferThreadClusterArrangeOrder
                            ",Seq(1,0,2)"    // BBlockTransferSrcAccessOrder
                            ",2"             // BBlockTransferSrcVectorDim
                            ",8"             // BBlockTransferSrcScalarPerVector
                            ",8"             // BBlockTransferDstScalarPerVector_BK1
-                           ",0"             // BBlockLdsExtraN
+                           ",false"         // BBlockLdsExtraN
                            ",1"             // CShuffleMXdlPerWavePerShuffle
                            ",1"             // CShuffleNXdlPerWavePerShuffle
                            ",Seq(1,32,1,8)" // CDEBlockTransferClusterLengths

experimental/builder/test/test_testing_utils.cpp

@@ -34,8 +34,8 @@ TEST(InstanceSet, FromFactory)
     const auto* el =
         "DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<2,NHWGC,GKYXC,EmptyTuple,NHWGK,fp16,fp16,"
         "fp32,fp16,EmptyTuple,fp16,PassThrough,PassThrough,PassThrough,Default,MNKPadding,1,128,"
-        "128,128,32,8,8,32,32,4,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,1,Seq(4,32,1),Seq(1,0,2),"
-        "Seq(1,0,2),2,8,8,1,1,1,Seq(1,16,1,8),8,fp16,fp16,Default,1>";
+        "128,128,32,8,8,32,32,4,2,Seq(4,32,1),Seq(1,0,2),Seq(1,0,2),2,8,8,true,Seq(4,32,1),"
+        "Seq(1,0,2),Seq(1,0,2),2,8,8,true,1,1,Seq(1,16,1,8),8,fp16,fp16,Default,1>";
     EXPECT_THAT(instances.instances, testing::Contains(el));
 }
 

experimental/builder/test/unit_conv_tensor_layout.cpp

@@ -9,27 +9,34 @@
 
 namespace {
 
-namespace ckb = ::ck_tile::builder;
-using ::ck_tile::builder::DataType;
-using ::ck_tile::builder::ElementwiseOperation;
-using ::ck_tile::builder::TensorLayout;
-using ::ck_tile::builder::factory::internal::AuxiliaryTensorLayouts;
-using ::ck_tile::builder::factory::internal::ConvTensorLayouts;
-using ::ck_tile::builder::factory::internal::LayoutToCK;
+namespace ckb = ck_tile::builder;
+using ck_tile::builder::DataType;
+using ck_tile::builder::ElementwiseOperation;
+using ck_tile::builder::TensorLayout;
+using ck_tile::builder::factory::internal::AuxiliaryTensorLayouts;
+using ck_tile::builder::factory::internal::ConvTensorLayouts;
+using ck_tile::builder::factory::internal::LayoutToCK;
+using ck_tile::builder::test::ConvolutionTensor;
+using ck_tile::builder::test::ConvSignature;
+using ck_tile::builder::test::TensorConfig;
+using ck_tile::builder::test::TensorOperation;
 
-using namespace ::ck_tile::builder::test;
-using enum ::ck_tile::builder::ConvDirection;
+namespace enums {
+using enum ck_tile::builder::ConvDirection;
+using enum ck_tile::builder::TensorLayout;
+using enum ck_tile::builder::DataType;
+} // namespace enums
 
 TEST(ConvTensorLayout, AssignsLayoutsFor1D_NWGC_GKXC_NWGK)
 {
-    static constexpr auto sig =
-        ConvSignature<>{.spatial_dim            = 1,
-                        .direction              = FORWARD,
-                        .data_type              = DataType::FP16,
-                        .accumulation_data_type = DataType::FP32,
-                        .input                  = {.config = {.layout = TensorLayout::NWGC}},
-                        .weight                 = {.config = {.layout = TensorLayout::GKXC}},
-                        .output                 = {.config = {.layout = TensorLayout::NWGK}}};
+    using namespace enums;
+    static constexpr auto sig = ConvSignature<>{.spatial_dim            = 1,
+                                                .direction              = FORWARD,
+                                                .data_type              = FP16,
+                                                .accumulation_data_type = FP32,
+                                                .input  = {.config = {.layout = NWGC}},
+                                                .weight = {.config = {.layout = GKXC}},
+                                                .output = {.config = {.layout = NWGK}}};
 
     using TensorLayouts = ConvTensorLayouts<sig, 1, FORWARD>;
 
@@ -41,14 +48,14 @@ TEST(ConvTensorLayout, AssignsLayoutsFor1D_NWGC_GKXC_NWGK)
 
 TEST(ConvTensorLayout, AssignsLayoutsFor1D_NGCW_GKXC_NGKW)
 {
-    static constexpr auto sig =
-        ConvSignature<>{.spatial_dim            = 1,
-                        .direction              = FORWARD,
-                        .data_type              = DataType::FP16,
-                        .accumulation_data_type = DataType::FP32,
-                        .input                  = {.config = {.layout = TensorLayout::NGCW}},
-                        .weight                 = {.config = {.layout = TensorLayout::GKXC}},
-                        .output                 = {.config = {.layout = TensorLayout::NGKW}}};
+    using namespace enums;
+    static constexpr auto sig = ConvSignature<>{.spatial_dim            = 1,
+                                                .direction              = FORWARD,
+                                                .data_type              = FP16,
+                                                .accumulation_data_type = FP32,
+                                                .input  = {.config = {.layout = NGCW}},
+                                                .weight = {.config = {.layout = GKXC}},
+                                                .output = {.config = {.layout = NGKW}}};
 
     using TensorLayouts = ConvTensorLayouts<sig, 1, FORWARD>;
 
@@ -60,14 +67,14 @@ TEST(ConvTensorLayout, AssignsLayoutsFor1D_NGCW_GKXC_NGKW)
 
 TEST(ConvTensorLayout, AssignsLayoutsFor1D_GNWC_GKXC_GNWK)
 {
-    static constexpr auto sig =
-        ConvSignature<>{.spatial_dim            = 1,
-                        .direction              = FORWARD,
-                        .data_type              = DataType::FP16,
-                        .accumulation_data_type = DataType::FP32,
-                        .input                  = {.config = {.layout = TensorLayout::GNWC}},
-                        .weight                 = {.config = {.layout = TensorLayout::GKXC}},
-                        .output                 = {.config = {.layout = TensorLayout::GNWK}}};
+    using namespace enums;
+    static constexpr auto sig = ConvSignature<>{.spatial_dim            = 1,
+                                                .direction              = FORWARD,
+                                                .data_type              = FP16,
+                                                .accumulation_data_type = FP32,
+                                                .input  = {.config = {.layout = GNWC}},
+                                                .weight = {.config = {.layout = GKXC}},
+                                                .output = {.config = {.layout = GNWK}}};
 
     using TensorLayouts = ConvTensorLayouts<sig, 1, FORWARD>;
 
@@ -79,14 +86,14 @@ TEST(ConvTensorLayout, AssignsLayoutsFor1D_GNWC_GKXC_GNWK)
 
 TEST(ConvTensorLayout, AssignsLayoutsFor1D_NGCW_GKCX_NGKW)
 {
-    static constexpr auto sig =
-        ConvSignature<>{.spatial_dim            = 1,
-                        .direction              = FORWARD,
-                        .data_type              = DataType::FP16,
-                        .accumulation_data_type = DataType::FP32,
-                        .input                  = {.config = {.layout = TensorLayout::NGCW}},
-                        .weight                 = {.config = {.layout = TensorLayout::GKCX}},
-                        .output                 = {.config = {.layout = TensorLayout::NGKW}}};
+    using namespace enums;
+    static constexpr auto sig = ConvSignature<>{.spatial_dim            = 1,
+                                                .direction              = FORWARD,
+                                                .data_type              = FP16,
+                                                .accumulation_data_type = FP32,
+                                                .input  = {.config = {.layout = NGCW}},
+                                                .weight = {.config = {.layout = GKCX}},
+                                                .output = {.config = {.layout = NGKW}}};
 
     using TensorLayouts = ConvTensorLayouts<sig, 1, FORWARD>;
 
@@ -98,14 +105,14 @@ TEST(ConvTensorLayout, AssignsLayoutsFor1D_NGCW_GKCX_NGKW)
 
 TEST(ConvTensorLayout, AssignsLayoutsFor2D_NGCHW_GKYXC_NGKHW)
 {
-    static constexpr auto sig =
-        ConvSignature<>{.spatial_dim            = 2,
-                        .direction              = FORWARD,
-                        .data_type              = DataType::FP16,
-                        .accumulation_data_type = DataType::FP32,
-                        .input                  = {.config = {.layout = TensorLayout::NGCHW}},
-                        .weight                 = {.config = {.layout = TensorLayout::GKYXC}},
-                        .output                 = {.config = {.layout = TensorLayout::NGKHW}}};
+    using namespace enums;
+    static constexpr auto sig = ConvSignature<>{.spatial_dim            = 2,
+                                                .direction              = FORWARD,
+                                                .data_type              = FP16,
+                                                .accumulation_data_type = FP32,
+                                                .input  = {.config = {.layout = NGCHW}},
+                                                .weight = {.config = {.layout = GKYXC}},
+                                                .output = {.config = {.layout = NGKHW}}};
 
     using TensorLayouts = ConvTensorLayouts<sig, 2, FORWARD>;
 
@@ -117,14 +124,14 @@ TEST(ConvTensorLayout, AssignsLayoutsFor2D_NGCHW_GKYXC_NGKHW)
 
 TEST(ConvTensorLayout, AssignsLayoutsFor2D_NHWGC_GKYXC_NHWGK)
 {
-    static constexpr auto sig =
-        ConvSignature<>{.spatial_dim            = 2,
-                        .direction              = FORWARD,
-                        .data_type              = DataType::FP16,
-                        .accumulation_data_type = DataType::FP32,
-                        .input                  = {.config = {.layout = TensorLayout::NHWGC}},
-                        .weight                 = {.config = {.layout = TensorLayout::GKYXC}},
-                        .output                 = {.config = {.layout = TensorLayout::NHWGK}}};
+    using namespace enums;
+    static constexpr auto sig = ConvSignature<>{.spatial_dim            = 2,
+                                                .direction              = FORWARD,
+                                                .data_type              = DataType::FP16,
+                                                .accumulation_data_type = DataType::FP32,
+                                                .input  = {.config = {.layout = NHWGC}},
+                                                .weight = {.config = {.layout = GKYXC}},
+                                                .output = {.config = {.layout = NHWGK}}};
 
     using TensorLayouts = ConvTensorLayouts<sig, 2, FORWARD>;
 
@@ -136,14 +143,14 @@ TEST(ConvTensorLayout, AssignsLayoutsFor2D_NHWGC_GKYXC_NHWGK)
 
 TEST(ConvTensorLayout, AssignsLayoutsFor2D_GNHWC_GKYXC_GNHWK)
 {
-    static constexpr auto sig =
-        ConvSignature<>{.spatial_dim            = 2,
-                        .direction              = FORWARD,
-                        .data_type              = DataType::FP16,
-                        .accumulation_data_type = DataType::FP32,
-                        .input                  = {.config = {.layout = TensorLayout::GNHWC}},
-                        .weight                 = {.config = {.layout = TensorLayout::GKYXC}},
-                        .output                 = {.config = {.layout = TensorLayout::GNHWK}}};
+    using namespace enums;
+    static constexpr auto sig = ConvSignature<>{.spatial_dim            = 2,
+                                                .direction              = FORWARD,
+                                                .data_type              = DataType::FP16,
+                                                .accumulation_data_type = DataType::FP32,
+                                                .input  = {.config = {.layout = GNHWC}},
+                                                .weight = {.config = {.layout = GKYXC}},
+                                                .output = {.config = {.layout = GNHWK}}};
 
     using TensorLayouts = ConvTensorLayouts<sig, 2, FORWARD>;
 
@@ -155,14 +162,14 @@ TEST(ConvTensorLayout, AssignsLayoutsFor2D_GNHWC_GKYXC_GNHWK)
 
 TEST(ConvTensorLayout, AssignsLayoutsFor2D_NGCHW_GKCYX_NGKHW)
 {
-    static constexpr auto sig =
-        ConvSignature<>{.spatial_dim            = 2,
-                        .direction              = FORWARD,
-                        .data_type              = DataType::FP16,
-                        .accumulation_data_type = DataType::FP32,
-                        .input                  = {.config = {.layout = TensorLayout::NGCHW}},
-                        .weight                 = {.config = {.layout = TensorLayout::GKCYX}},
-                        .output                 = {.config = {.layout = TensorLayout::NGKHW}}};
+    using namespace enums;
+    static constexpr auto sig = ConvSignature<>{.spatial_dim            = 2,
+                                                .direction              = FORWARD,
+                                                .data_type              = DataType::FP16,
+                                                .accumulation_data_type = DataType::FP32,
+                                                .input  = {.config = {.layout = NGCHW}},
+                                                .weight = {.config = {.layout = GKCYX}},
+                                                .output = {.config = {.layout = NGKHW}}};
 
     using TensorLayouts = ConvTensorLayouts<sig, 2, FORWARD>;
 
@@ -174,14 +181,14 @@ TEST(ConvTensorLayout, AssignsLayoutsFor2D_NGCHW_GKCYX_NGKHW)
 
 TEST(ConvTensorLayout, AssignsLayoutsFor3D_NGCDHW_GKCZYX_NGKDHW)
 {
-    static constexpr auto sig =
-        ConvSignature<>{.spatial_dim            = 3,
-                        .direction              = FORWARD,
-                        .data_type              = DataType::FP16,
-                        .accumulation_data_type = DataType::FP32,
-                        .input                  = {.config = {.layout = TensorLayout::NGCDHW}},
-                        .weight                 = {.config = {.layout = TensorLayout::GKCZYX}},
-                        .output                 = {.config = {.layout = TensorLayout::NGKDHW}}};
+    using namespace enums;
+    static constexpr auto sig = ConvSignature<>{.spatial_dim            = 3,
+                                                .direction              = FORWARD,
+                                                .data_type              = DataType::FP16,
+                                                .accumulation_data_type = DataType::FP32,
+                                                .input  = {.config = {.layout = NGCDHW}},
+                                                .weight = {.config = {.layout = GKCZYX}},
+                                                .output = {.config = {.layout = NGKDHW}}};
 
     using TensorLayouts = ConvTensorLayouts<sig, 3, FORWARD>;
 
@@ -193,14 +200,14 @@ TEST(ConvTensorLayout, AssignsLayoutsFor3D_NGCDHW_GKCZYX_NGKDHW)
 
 TEST(ConvTensorLayout, AssignsLayoutsFor3D_NDHWGC_GKZYXC_NDHWGK)
 {
-    static constexpr auto sig =
-        ConvSignature<>{.spatial_dim            = 3,
-                        .direction              = FORWARD,
-                        .data_type              = DataType::FP16,
-                        .accumulation_data_type = DataType::FP32,
-                        .input                  = {.config = {.layout = TensorLayout::NDHWGC}},
-                        .weight                 = {.config = {.layout = TensorLayout::GKZYXC}},
-                        .output                 = {.config = {.layout = TensorLayout::NDHWGK}}};
+    using namespace enums;
+    static constexpr auto sig = ConvSignature<>{.spatial_dim            = 3,
+                                                .direction              = FORWARD,
+                                                .data_type              = DataType::FP16,
+                                                .accumulation_data_type = DataType::FP32,
+                                                .input  = {.config = {.layout = NDHWGC}},
+                                                .weight = {.config = {.layout = GKZYXC}},
+                                                .output = {.config = {.layout = NDHWGK}}};
 
     using TensorLayouts = ConvTensorLayouts<sig, 3, FORWARD>;
 
@@ -212,14 +219,14 @@ TEST(ConvTensorLayout, AssignsLayoutsFor3D_NDHWGC_GKZYXC_NDHWGK)
 
 TEST(ConvTensorLayout, AssignsLayoutsFor3D_GNDHWC_GKZYXC_GNDHWK)
 {
-    static constexpr auto sig =
-        ConvSignature<>{.spatial_dim            = 3,
-                        .direction              = FORWARD,
-                        .data_type              = DataType::FP16,
-                        .accumulation_data_type = DataType::FP32,
-                        .input                  = {.config = {.layout = TensorLayout::GNDHWC}},
-                        .weight                 = {.config = {.layout = TensorLayout::GKZYXC}},
-                        .output                 = {.config = {.layout = TensorLayout::GNDHWK}}};
+    using namespace enums;
+    static constexpr auto sig = ConvSignature<>{.spatial_dim            = 3,
+                                                .direction              = FORWARD,
+                                                .data_type              = DataType::FP16,
+                                                .accumulation_data_type = DataType::FP32,
+                                                .input  = {.config = {.layout = GNDHWC}},
+                                                .weight = {.config = {.layout = GKZYXC}},
+                                                .output = {.config = {.layout = GNDHWK}}};
 
     using TensorLayouts = ConvTensorLayouts<sig, 3, FORWARD>;
 
@@ -261,8 +268,10 @@ struct MockAuxiliaryTensorConfig
 
 TEST(AuxiliaryTensorLayoutIntegration, SingleBiasTensorWithG_K_Layout)
 {
+    using namespace enums;
+
     static constexpr std::array<MockAuxiliaryTensorConfig, 1> aux_configs = {
-        MockAuxiliaryTensorConfig{.layout = TensorLayout::G_K_strided}};
+        MockAuxiliaryTensorConfig{.layout = G_K_strided}};
 
     using AuxLayouts = AuxiliaryTensorLayouts<aux_configs, 2, FORWARD>;
 
@@ -273,6 +282,8 @@ TEST(AuxiliaryTensorLayoutIntegration, SingleBiasTensorWithG_K_Layout)
 
 TEST(AuxiliaryTensorLayoutIntegration, SingleBiasTensorWithGC_Layout)
 {
+    using namespace enums;
+
     static constexpr std::array<MockAuxiliaryTensorConfig, 1> aux_configs = {
         MockAuxiliaryTensorConfig{.layout = TensorLayout::GC}};
 
@@ -285,8 +296,10 @@ TEST(AuxiliaryTensorLayoutIntegration, SingleBiasTensorWithGC_Layout)
 
 TEST(AuxiliaryTensorLayoutIntegration, SingleBiasTensorWithG_C_Layout)
 {
+    using namespace enums;
+
     static constexpr std::array<MockAuxiliaryTensorConfig, 1> aux_configs = {
-        MockAuxiliaryTensorConfig{.layout = TensorLayout::G_C_strided}};
+        MockAuxiliaryTensorConfig{.layout = G_C_strided}};
 
     using AuxLayouts = AuxiliaryTensorLayouts<aux_configs, 2, FORWARD>;
 
@@ -297,9 +310,11 @@ TEST(AuxiliaryTensorLayoutIntegration, SingleBiasTensorWithG_C_Layout)
 
 TEST(AuxiliaryTensorLayoutIntegration, TwoAuxiliaryTensors)
 {
+    using namespace enums;
+
     static constexpr std::array<MockAuxiliaryTensorConfig, 2> aux_configs = {
         MockAuxiliaryTensorConfig{.layout = TensorLayout::G_K_strided},
-        MockAuxiliaryTensorConfig{.layout = TensorLayout::GC}};
+        MockAuxiliaryTensorConfig{.layout = GC}};
 
     using AuxLayouts = AuxiliaryTensorLayouts<aux_configs, 2, FORWARD>;
 
@@ -311,10 +326,12 @@ TEST(AuxiliaryTensorLayoutIntegration, TwoAuxiliaryTensors)
 
 TEST(AuxiliaryTensorLayoutIntegration, ThreeAuxiliaryTensors)
 {
+    using namespace enums;
+
     static constexpr std::array<MockAuxiliaryTensorConfig, 3> aux_configs = {
-        MockAuxiliaryTensorConfig{.layout = TensorLayout::G_K_strided},
-        MockAuxiliaryTensorConfig{.layout = TensorLayout::GC},
-        MockAuxiliaryTensorConfig{.layout = TensorLayout::G_C_strided}};
+        MockAuxiliaryTensorConfig{.layout = G_K_strided},
+        MockAuxiliaryTensorConfig{.layout = GC},
+        MockAuxiliaryTensorConfig{.layout = G_C_strided}};
 
     using AuxLayouts = AuxiliaryTensorLayouts<aux_configs, 2, FORWARD>;
 
@@ -327,8 +344,10 @@ TEST(AuxiliaryTensorLayoutIntegration, ThreeAuxiliaryTensors)
 
 TEST(AuxiliaryTensorLayoutIntegration, WorksWith1DConvolution)
 {
+    using namespace enums;
+
     static constexpr std::array<MockAuxiliaryTensorConfig, 1> aux_configs = {
-        MockAuxiliaryTensorConfig{.layout = TensorLayout::G_K_strided}};
+        MockAuxiliaryTensorConfig{.layout = G_K_strided}};
 
     using AuxLayouts = AuxiliaryTensorLayouts<aux_configs, 1, FORWARD>;
 
@@ -339,8 +358,10 @@ TEST(AuxiliaryTensorLayoutIntegration, WorksWith1DConvolution)
 
 TEST(AuxiliaryTensorLayoutIntegration, WorksWith3DConvolution)
 {
+    using namespace enums;
+
     static constexpr std::array<MockAuxiliaryTensorConfig, 1> aux_configs = {
-        MockAuxiliaryTensorConfig{.layout = TensorLayout::GC}};
+        MockAuxiliaryTensorConfig{.layout = GC}};
 
     using AuxLayouts = AuxiliaryTensorLayouts<aux_configs, 3, FORWARD>;
 
@@ -351,7 +372,8 @@ TEST(AuxiliaryTensorLayoutIntegration, WorksWith3DConvolution)
 
 TEST(ConvTensorLayoutsWithAuxiliary, Conv2DWithSingleBiasG_K)
 {
-    using OutputOp = TensorOperation<TensorConfig{.layout = TensorLayout::G_K_strided}>;
+    using namespace enums;
+    using OutputOp = TensorOperation<TensorConfig{.layout = G_K_strided}>;
 
     static constexpr auto sig =
         ConvSignature<ConvolutionTensor<>, ConvolutionTensor<>, ConvolutionTensor<OutputOp>>{
@@ -359,9 +381,9 @@ TEST(ConvTensorLayoutsWithAuxiliary, Conv2DWithSingleBiasG_K)
             .direction              = FORWARD,
             .data_type              = DataType::FP16,
             .accumulation_data_type = DataType::FP32,
-            .input                  = {.config = {.layout = TensorLayout::NGCHW}},
-            .weight                 = {.config = {.layout = TensorLayout::GKYXC}},
-            .output                 = {.config = {.layout = TensorLayout::NGKHW},
+            .input                  = {.config = {.layout = NGCHW}},
+            .weight                 = {.config = {.layout = GKYXC}},
+            .output                 = {.config = {.layout = NGKHW},
                                        .operation =
                                            OutputOp{.elementwise_operation = ElementwiseOperation::SCALE}}};
 
@@ -377,7 +399,8 @@ TEST(ConvTensorLayoutsWithAuxiliary, Conv2DWithSingleBiasG_K)
 
 TEST(ConvTensorLayoutsWithAuxiliary, Conv2DWithSingleBiasGC)
 {
-    using OutputOp = TensorOperation<TensorConfig{.layout = TensorLayout::GC}>;
+    using namespace enums;
+    using OutputOp = TensorOperation<TensorConfig{.layout = GC}>;
 
     static constexpr auto sig =
         ConvSignature<ConvolutionTensor<>, ConvolutionTensor<>, ConvolutionTensor<OutputOp>>{
@@ -385,9 +408,9 @@ TEST(ConvTensorLayoutsWithAuxiliary, Conv2DWithSingleBiasGC)
             .direction              = FORWARD,
             .data_type              = DataType::BF16,
             .accumulation_data_type = DataType::FP32,
-            .input                  = {.config = {.layout = TensorLayout::NHWGC}},
-            .weight                 = {.config = {.layout = TensorLayout::GKYXC}},
-            .output                 = {.config = {.layout = TensorLayout::NHWGK},
+            .input                  = {.config = {.layout = NHWGC}},
+            .weight                 = {.config = {.layout = GKYXC}},
+            .output                 = {.config = {.layout = NHWGK},
                                        .operation =
                                            OutputOp{.elementwise_operation = ElementwiseOperation::SCALE}}};
 
@@ -403,8 +426,9 @@ TEST(ConvTensorLayoutsWithAuxiliary, Conv2DWithSingleBiasGC)
 
 TEST(ConvTensorLayoutsWithAuxiliary, Conv2DWithTwoAuxiliaryTensors)
 {
-    using OutputOp = TensorOperation<TensorConfig{.layout = TensorLayout::G_K_strided},
-                                     TensorConfig{.layout = TensorLayout::GC}>;
+    using namespace enums;
+    using OutputOp =
+        TensorOperation<TensorConfig{.layout = G_K_strided}, TensorConfig{.layout = GC}>;
 
     static constexpr auto sig =
         ConvSignature<ConvolutionTensor<>, ConvolutionTensor<>, ConvolutionTensor<OutputOp>>{
@@ -412,9 +436,9 @@ TEST(ConvTensorLayoutsWithAuxiliary, Conv2DWithTwoAuxiliaryTensors)
             .direction              = FORWARD,
             .data_type              = DataType::FP16,
             .accumulation_data_type = DataType::FP32,
-            .input                  = {.config = {.layout = TensorLayout::GNHWC}},
-            .weight                 = {.config = {.layout = TensorLayout::GKYXC}},
-            .output                 = {.config    = {.layout = TensorLayout::GNHWK},
+            .input                  = {.config = {.layout = GNHWC}},
+            .weight                 = {.config = {.layout = GKYXC}},
+            .output                 = {.config    = {.layout = GNHWK},
                                        .operation = OutputOp{.elementwise_operation =
                                                  ElementwiseOperation::SCALEADD_SCALEADD_RELU}}};
 
@@ -431,7 +455,8 @@ TEST(ConvTensorLayoutsWithAuxiliary, Conv2DWithTwoAuxiliaryTensors)
 
 TEST(ConvTensorLayoutsWithAuxiliary, Conv1DWithBias)
 {
-    using OutputOp = TensorOperation<TensorConfig{.layout = TensorLayout::G_K_strided}>;
+    using namespace enums;
+    using OutputOp = TensorOperation<TensorConfig{.layout = G_K_strided}>;
 
     static constexpr auto sig =
         ConvSignature<ConvolutionTensor<>, ConvolutionTensor<>, ConvolutionTensor<OutputOp>>{
@@ -439,9 +464,9 @@ TEST(ConvTensorLayoutsWithAuxiliary, Conv1DWithBias)
             .direction              = FORWARD,
             .data_type              = DataType::FP32,
             .accumulation_data_type = DataType::FP32,
-            .input                  = {.config = {.layout = TensorLayout::NWGC}},
-            .weight                 = {.config = {.layout = TensorLayout::GKXC}},
-            .output                 = {.config = {.layout = TensorLayout::NWGK},
+            .input                  = {.config = {.layout = NWGC}},
+            .weight                 = {.config = {.layout = GKXC}},
+            .output                 = {.config = {.layout = NWGK},
                                        .operation =
                                            OutputOp{.elementwise_operation = ElementwiseOperation::SCALE}}};
 
@@ -457,7 +482,8 @@ TEST(ConvTensorLayoutsWithAuxiliary, Conv1DWithBias)
 
 TEST(ConvTensorLayoutsWithAuxiliary, Conv3DWithBias)
 {
-    using OutputOp = TensorOperation<TensorConfig{.layout = TensorLayout::G_C_strided}>;
+    using namespace enums;
+    using OutputOp = TensorOperation<TensorConfig{.layout = G_C_strided}>;
 
     static constexpr auto sig =
         ConvSignature<ConvolutionTensor<>, ConvolutionTensor<>, ConvolutionTensor<OutputOp>>{
@@ -465,9 +491,9 @@ TEST(ConvTensorLayoutsWithAuxiliary, Conv3DWithBias)
             .direction              = FORWARD,
             .data_type              = DataType::FP16,
             .accumulation_data_type = DataType::FP32,
-            .input                  = {.config = {.layout = TensorLayout::NDHWGC}},
-            .weight                 = {.config = {.layout = TensorLayout::GKZYXC}},
-            .output                 = {.config    = {.layout = TensorLayout::NDHWGK},
+            .input                  = {.config = {.layout = NDHWGC}},
+            .weight                 = {.config = {.layout = GKZYXC}},
+            .output                 = {.config    = {.layout = NDHWGK},
                                        .operation = OutputOp{.elementwise_operation =
                                                  ElementwiseOperation::BIAS_BNORM_CLAMP}}};
 

experimental/builder/test/utils/conv_algorithm_type_utils.hpp

@@ -0,0 +1,346 @@
+// Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
+// SPDX-License-Identifier: MIT
+
+#pragma once
+
+#include "../impl/conv_algorithm_types.hpp"
+#include <sstream>
+#include <array>
+
+namespace ck_tile::builder::test {
+
+namespace ckb = ck_tile::builder;
+
+// Helper function to convert arrays to Seq(...) format
+template <typename T, size_t N>
+std::string array_to_seq(const std::array<T, N>& arr)
+{
+    std::ostringstream oss;
+    oss << "Seq(";
+    for(size_t i = 0; i < N; ++i)
+    {
+        if(i > 0)
+            oss << ",";
+        oss << arr[i];
+    }
+    oss << ")";
+    return oss.str();
+}
+
+// Base template - will cause compilation error for unsupported types
+template <typename T>
+std::string to_string(T)
+{
+    static_assert(sizeof(T) == 0, "Unsupported type");
+    return "";
+}
+
+// Template specializations for enum types
+
+template <>
+inline std::string to_string<PipelineVersion>(PipelineVersion t)
+{
+    std::ostringstream oss;
+    oss << t;
+    return oss.str();
+}
+
+template <>
+inline std::string to_string<PipelineScheduler>(PipelineScheduler t)
+{
+    std::ostringstream oss;
+    oss << t;
+    return oss.str();
+}
+
+template <>
+inline std::string to_string<ConvFwdSpecialization>(ConvFwdSpecialization t)
+{
+    std::ostringstream oss;
+    oss << t;
+    return oss.str();
+}
+
+template <>
+inline std::string to_string<GemmSpecialization>(GemmSpecialization t)
+{
+    std::ostringstream oss;
+    oss << t;
+    return oss.str();
+}
+
+// Template specializations for struct types
+
+template <>
+inline std::string to_string<MNK<size_t>>(MNK<size_t> t)
+{
+    return array_to_seq(std::array<size_t, 3>{t.m, t.n, t.k});
+}
+
+template <>
+inline std::string to_string<ThreadBlock>(ThreadBlock t)
+{
+    std::ostringstream oss;
+    oss << t.block_size << "," << t.tile_size.m << "," << t.tile_size.n << "," << t.tile_size.k;
+    return oss.str();
+}
+
+template <>
+inline std::string to_string<GridwiseXdlGemm>(GridwiseXdlGemm t)
+{
+    std::ostringstream oss;
+    oss << t.ak1 << "," << t.bk1 << "," << t.m_per_xdl << "," << t.n_per_xdl << ","
+        << t.m_xdl_per_wave << "," << t.n_xdl_per_wave;
+    return oss.str();
+}
+
+template <>
+inline std::string to_string<GridwiseWmmaGemm>(GridwiseWmmaGemm t)
+{
+    std::ostringstream oss;
+    oss << t.k1 << "," << t.m_per_wmma << "," << t.n_per_wmma << "," << t.m_wmma_per_wave << ","
+        << t.n_wmma_per_wave;
+    return oss.str();
+}
+
+template <>
+inline std::string to_string<BlockGemm>(BlockGemm t)
+{
+    std::ostringstream oss;
+    oss << to_string(t.scheduler) << "," << to_string(t.pipeline_version);
+    return oss.str();
+}
+
+template <>
+inline std::string to_string<BlockTransfer>(BlockTransfer t)
+{
+    return array_to_seq(std::array<size_t, 3>{t.k0, t.m_n, t.k1});
+}
+
+template <>
+inline std::string to_string<ThreadCluster>(ThreadCluster t)
+{
+    return array_to_seq(
+        std::array<size_t, 4>{t.m_block, t.m_wave_per_xdl, t.n_block, t.n_wave_per_xdl});
+}
+
+template <>
+inline std::string to_string<LdsTransfer>(LdsTransfer t)
+{
+    std::ostringstream oss;
+    oss << t.src_vector_dim << "," << t.src_scalar_per_vector << "," << t.lds_dst_scalar_per_vector
+        << "," << (t.lds_padding ? "true" : "false") << ","
+        << (t.is_direct_load ? "true" : "false");
+    return oss.str();
+}
+
+template <>
+inline std::string to_string<AccessOrder>(AccessOrder t)
+{
+    return array_to_seq(t.order);
+}
+
+template <>
+inline std::string to_string<TransferAB>(TransferAB t)
+{
+    std::ostringstream oss;
+    oss << to_string(t.block_transfer) << "," << to_string(t.block_transfer_access_order) << ","
+        << to_string(t.src_access_order) << "," << t.lds_transfer.src_vector_dim << ","
+        << t.lds_transfer.src_scalar_per_vector << "," << t.lds_transfer.lds_dst_scalar_per_vector
+        << "," << (t.lds_transfer.lds_padding ? "true" : "false");
+    return oss.str();
+}
+
+template <>
+inline std::string to_string<TransferC>(TransferC t)
+{
+    std::ostringstream oss;
+    oss << t.epilogue.m_xdl_per_wave_per_shuffle << "," << t.epilogue.n_per_wave_per_shuffle << ","
+        << to_string(t.thread_cluster_dims) << "," << t.epilogue.scalar_per_vector;
+    return oss.str();
+}
+
+template <>
+inline std::string to_string<TransferABC>(TransferABC t)
+{
+    std::ostringstream oss;
+    oss << to_string(t.a) << "," << to_string(t.b) << "," << to_string(t.c);
+    return oss.str();
+}
+
+template <>
+inline std::string to_string<DlThreadConfig>(DlThreadConfig t)
+{
+    std::ostringstream oss;
+    oss << t.k1 << "," << t.m1_per_thread << "," << t.n1_per_thread << "," << t.k_per_thread;
+    return oss.str();
+}
+
+template <>
+inline std::string to_string<DlThreadCluster>(DlThreadCluster t)
+{
+    std::ostringstream oss;
+    oss << array_to_seq(t.m1_xs) << "," << array_to_seq(t.n1_xs);
+    return oss.str();
+}
+
+template <>
+inline std::string to_string<DlBlockTransfer>(DlBlockTransfer t)
+{
+    std::ostringstream oss;
+    oss << array_to_seq(t.thread_slice_lengths) << "," << array_to_seq(t.thread_cluster_lengths)
+        << "," << array_to_seq(t.thread_cluster_arrange_order) << ","
+        << array_to_seq(t.src_access_order) << "," << array_to_seq(t.src_vector_tensor_lengths)
+        << "," << array_to_seq(t.src_vector_tensor_contiguous_dim_order) << ","
+        << array_to_seq(t.dst_vector_tensor_lengths);
+    return oss.str();
+}
+
+template <>
+inline std::string to_string<DlEpilogue>(DlEpilogue t)
+{
+    std::ostringstream oss;
+    oss << array_to_seq(t.src_dst_access_order) << "," << t.src_dst_vector_dim << ","
+        << t.dst_scalar_per_vector;
+    return oss.str();
+}
+
+template <>
+inline std::string to_string<DlBlockTransferAB>(DlBlockTransferAB t)
+{
+    return to_string(t.block_transfer);
+}
+
+template <>
+inline std::string to_string<DlBlockTransferC>(DlBlockTransferC t)
+{
+    return to_string(t.epilogue);
+}
+
+template <>
+inline std::string to_string<DlTransferABC>(DlTransferABC t)
+{
+    std::ostringstream oss;
+    oss << to_string(t.a) << "," << to_string(t.b) << "," << to_string(t.c);
+    return oss.str();
+}
+
+// Template specializations for factory wrapper types
+
+template <>
+inline std::string to_string<ThreadBlock_>(ThreadBlock_ t)
+{
+    return to_string(t.thread_block);
+}
+
+template <>
+inline std::string to_string<XdlGemm_>(XdlGemm_ t)
+{
+    return to_string(t.gridwise_gemm);
+}
+
+template <>
+inline std::string to_string<WmmaGemm_>(WmmaGemm_ t)
+{
+    return to_string(t.gridwise_gemm);
+}
+
+template <>
+inline std::string to_string<Transfer_>(Transfer_ t)
+{
+    return to_string(t.transfer);
+}
+
+template <>
+inline std::string to_string<ConvSpecialization_>(ConvSpecialization_ t)
+{
+    std::ostringstream oss;
+    oss << to_string(t.fwd_specialization) << "," << to_string(t.gemm_specialization);
+    return oss.str();
+}
+
+template <>
+inline std::string to_string<Prefetch_>(Prefetch_ t)
+{
+    std::ostringstream oss;
+    oss << t.num_gemm_k_prefetch_stages << "," << t.num_groups_to_merge << ","
+        << to_string(t.loop_scheduler);
+    return oss.str();
+}
+
+template <>
+inline std::string to_string<BlockGemm_>(BlockGemm_ t)
+{
+    return to_string(t.block_gemm);
+}
+
+template <>
+inline std::string to_string<DlThreadConfig_>(DlThreadConfig_ t)
+{
+    return to_string(t.thread_config);
+}
+
+template <>
+inline std::string to_string<DlThreadCluster_>(DlThreadCluster_ t)
+{
+    return to_string(t.thread_cluster);
+}
+
+template <>
+inline std::string to_string<DlTransfer_>(DlTransfer_ t)
+{
+    return to_string(t.transfer);
+}
+
+// Template specializations for algorithm types
+
+template <>
+inline std::string to_string<ConvAlgorithm_DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle>(
+    ConvAlgorithm_DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle t)
+{
+    std::ostringstream oss;
+    oss << to_string(static_cast<ThreadBlock_>(t)) << "," << to_string(static_cast<XdlGemm_>(t))
+        << "," << to_string(static_cast<Transfer_>(t));
+    return oss.str();
+}
+
+template <>
+inline std::string to_string<ConvAlgorithm_DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3>(
+    ConvAlgorithm_DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3 t)
+{
+    std::ostringstream oss;
+    oss << to_string(static_cast<ThreadBlock_>(t)) << "," << to_string(static_cast<XdlGemm_>(t))
+        << "," << to_string(static_cast<Transfer_>(t));
+    return oss.str();
+}
+
+template <>
+inline std::string to_string<ConvAlgorithm_DeviceGroupedConvFwdMultipleD_Wmma_CShuffle>(
+    ConvAlgorithm_DeviceGroupedConvFwdMultipleD_Wmma_CShuffle t)
+{
+    std::ostringstream oss;
+    oss << to_string(static_cast<ThreadBlock_>(t)) << "," << to_string(static_cast<WmmaGemm_>(t))
+        << "," << to_string(static_cast<Transfer_>(t));
+    return oss.str();
+}
+
+template <>
+inline std::string to_string<ConvAlgorithm_DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK>(
+    ConvAlgorithm_DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK t)
+{
+    std::ostringstream oss;
+    oss << to_string(static_cast<ThreadBlock_>(t)) << ","
+        << to_string(static_cast<DlThreadConfig_>(t)) << ","
+        << to_string(static_cast<DlThreadCluster_>(t)) << ","
+        << to_string(static_cast<DlTransfer_>(t));
+    return oss.str();
+}
+
+template <>
+inline std::string to_string<ConvAlgorithm_DeviceGroupedConvFwdMultipleD_Xdl_CShuffle_Large_Tensor>(
+    ConvAlgorithm_DeviceGroupedConvFwdMultipleD_Xdl_CShuffle_Large_Tensor t)
+{
+    return to_string(t.base_algorithm);
+}
+
+} // namespace ck_tile::builder::test

include/ck/tensor_operation/gpu/block/blockwise_gemm_pipeline_wmmaops_base.hpp

@@ -109,65 +109,145 @@ struct BlockwiseGemmWmmaops_pipeline_base
         }
     };
 
-    template <index_t ScaleSliceSizeN,
+    template <index_t ScaleSliceSizeMN,
+              index_t ScaleSliceStrideMN,
               index_t ScaleSliceSizeK,
-              index_t NWaves,
-              index_t ScaleBlockK,
               index_t NumberOfBuffers,
+              index_t RegSizePerWmma,
               typename GridDesc,
               typename ThreadCopy,
               typename GridBuffer,
               typename ThreadStaticBuffer,
-              typename BScaleThreadDesc>
-    struct BScale
+              typename ThreadDesc>
+    struct ABScale
     {
-        __device__ BScale(GridDesc b_scale_grid_desc_,
-                          ThreadCopy b_scale_thread_copy_,
-                          GridBuffer b_scale_grid_buf_)
-            : b_scale_thread_copy(b_scale_thread_copy_),
-              b_scale_grid_desc(b_scale_grid_desc_),
-              b_scale_grid_buf(b_scale_grid_buf_) {};
+        __device__ ABScale(GridDesc scale_grid_desc_,
+                           ThreadCopy scale_thread_copy_,
+                           GridBuffer scale_grid_buf_)
+            : scale_thread_copy(scale_thread_copy_),
+              scale_grid_desc(scale_grid_desc_),
+              scale_grid_buf(scale_grid_buf_) {};
 
-        static constexpr index_t num_scale_k_block = BScaleThreadDesc{}.GetLength(Number<1>{});
+        static constexpr index_t num_scale_k_block = ThreadDesc{}.GetLength(Number<1>{});
         static constexpr index_t num_scale_krepeat = KRepeat / num_scale_k_block;
 
-        static constexpr auto b_scale_thread_desc = BScaleThreadDesc{};
+        static constexpr index_t num_slice_mn      = ScaleSliceSizeMN;
+        static constexpr index_t num_slice_k       = ScaleSliceSizeK;
+        static constexpr index_t reg_size_per_wmma = RegSizePerWmma;
 
-        static constexpr auto b_scale_thread_copy_step =
-            make_tuple(make_multi_index(NWaves * NPerWmma, 0),
-                       make_multi_index(-NPerBlock, 0),
-                       make_multi_index(-NPerBlock, (KPerBlock + ScaleBlockK - 1) / ScaleBlockK));
+        static constexpr auto scale_thread_desc = ThreadDesc{};
+
+        static constexpr auto scale_thread_copy_step =
+            make_tuple(make_multi_index(ScaleSliceStrideMN, 0),
+                       make_multi_index(-ScaleSliceSizeMN / RegSizePerWmma * ScaleSliceStrideMN, 0),
+                       make_multi_index(-ScaleSliceSizeMN / RegSizePerWmma * ScaleSliceStrideMN,
+                                        ScaleSliceSizeK));
 
         template <index_t NBuffer>
         __device__ void GlobalLoad(bool cond)
         {
-            static_for<0, NRepeat, 1>{}([&](auto n0) {
-                b_scale_thread_copy.Run(b_scale_grid_desc,
-                                        b_scale_grid_buf,
-                                        b_scale_thread_desc,
-                                        make_tuple(n0, Number<0>{}),
-                                        b_scale_thread_bufs(Number<NBuffer>{}));
+            static_for<0, ScaleSliceSizeMN / RegSizePerWmma, 1>{}([&](auto m0) {
+                scale_thread_copy.Run(scale_grid_desc,
+                                      scale_grid_buf,
+                                      scale_thread_desc,
+                                      make_tuple(m0 * Number<RegSizePerWmma>{}, Number<0>{}),
+                                      scale_thread_bufs(Number<NBuffer>{}));
 
-                b_scale_thread_copy.MoveSrcSliceWindow(b_scale_grid_desc,
-                                                       b_scale_thread_copy_step.At(Number<0>{}));
+                scale_thread_copy.MoveSrcSliceWindow(scale_grid_desc,
+                                                     scale_thread_copy_step.At(Number<0>{}));
             });
 
             if(cond)
             {
-                b_scale_thread_copy.MoveSrcSliceWindow(b_scale_grid_desc,
-                                                       b_scale_thread_copy_step.At(Number<2>{}));
+                scale_thread_copy.MoveSrcSliceWindow(scale_grid_desc,
+                                                     scale_thread_copy_step.At(Number<2>{}));
             }
             else
             {
-                b_scale_thread_copy.MoveSrcSliceWindow(b_scale_grid_desc,
-                                                       b_scale_thread_copy_step.At(Number<1>{}));
+                scale_thread_copy.MoveSrcSliceWindow(scale_grid_desc,
+                                                     scale_thread_copy_step.At(Number<1>{}));
             }
         }
 
-        ThreadCopy b_scale_thread_copy;
-        GridDesc b_scale_grid_desc;
-        GridBuffer b_scale_grid_buf;
-        StaticallyIndexedArray<ThreadStaticBuffer, Number<NumberOfBuffers>{}> b_scale_thread_bufs;
+        ThreadCopy scale_thread_copy;
+        GridDesc scale_grid_desc;
+        GridBuffer scale_grid_buf;
+        StaticallyIndexedArray<ThreadStaticBuffer, Number<NumberOfBuffers>{}> scale_thread_bufs;
+    };
+
+    template <typename AScaleStruct, typename BScaleStruct>
+    struct CScale
+    {
+        __device__ CScale() {}
+
+        static constexpr auto reg_size_per_wmma =
+            ck::is_same_v<BScaleStruct, Empty> && ck::is_same_v<AScaleStruct, Empty>
+                ? 1
+                : wmma_gemm.GetRegSizePerWmma();
+        static constexpr auto c_scale_thread_desc = make_naive_tensor_descriptor_packed(make_tuple(
+            Number<ck::math::max(AScaleStruct::num_slice_k, BScaleStruct::num_slice_k)>{},
+            Number<AScaleStruct::num_slice_mn>{},
+            Number<BScaleStruct::num_slice_mn>{}));
+        using CScaleThreadDesc                    = decltype(c_scale_thread_desc);
+        static constexpr auto num_scale_k_block   = CScaleThreadDesc{}.GetLength(Number<0>{});
+        static constexpr auto num_scale_m_block   = CScaleThreadDesc{}.GetLength(Number<1>{});
+        static constexpr auto num_scale_n_block   = CScaleThreadDesc{}.GetLength(Number<2>{});
+        using ThreadStaticBuffer = decltype(make_static_buffer<AddressSpaceEnum::Vgpr, AccDataType>(
+            c_scale_thread_desc.GetElementSpaceSize()));
+
+        __device__ void Load(AScaleStruct& a_scale_struct, BScaleStruct& b_scale_struct)
+        {
+            using AScaleThreadDesc = decltype(AScaleStruct::scale_thread_desc);
+            using BScaleThreadDesc = decltype(BScaleStruct::scale_thread_desc);
+
+            static_for<0, num_scale_m_block, 1>{}([&](auto m0) {
+                static_for<0, num_scale_n_block, 1>{}([&](auto n0) {
+                    static_for<0, num_scale_k_block, 1>{}([&](auto k0) {
+                        constexpr index_t c_offset =
+                            CScaleThreadDesc{}.CalculateOffset(make_tuple(k0, m0, n0));
+                        constexpr index_t a_offset =
+                            AScaleThreadDesc{}.CalculateOffset(make_tuple(m0, k0));
+                        constexpr index_t b_offset =
+                            BScaleThreadDesc{}.CalculateOffset(make_tuple(n0, k0));
+
+                        c_scale_thread_bufs(I0)(Number<c_offset>{}) =
+                            a_scale_struct.scale_thread_bufs(I0)[Number<a_offset>{}] *
+                            b_scale_struct.scale_thread_bufs(I0)[Number<b_offset>{}];
+                    });
+                });
+            });
+        }
+
+        __device__ void Clear()
+        {
+            static_for<0, reg_size_per_wmma, 1>{}([&](auto t) {
+                c_thread_buf_per_scale.GetVectorTypeReference(Number<0>{})
+                    .template AsType<AccDataType>()(Number<t>{}) = 0;
+            });
+        }
+
+        template <index_t k_index, index_t m_index, index_t n_index, typename CThreadBuf>
+        __device__ void UpdateCThreadBuf(CThreadBuf& c_thread_buf)
+        {
+            static_for<0, reg_size_per_wmma, 1>{}([&](auto t) {
+                constexpr index_t c_offset =
+                    c_thread_desc_.CalculateOffset(make_tuple(m_index, n_index, t));
+                constexpr index_t cscale_offset = CScaleThreadDesc{}.CalculateOffset(make_tuple(
+                    k_index,
+                    (m_index * num_scale_m_block / MRepeat) % num_scale_m_block +
+                        (Number<t / (reg_size_per_wmma / AScaleStruct::reg_size_per_wmma)>{}) %
+                            AScaleStruct::reg_size_per_wmma,
+                    (n_index * num_scale_n_block / NRepeat) % num_scale_n_block));
+                c_thread_buf(Number<c_offset>{}) +=
+                    c_thread_buf_per_scale.GetVectorTypeReference(Number<0>{})
+                        .template AsType<AccDataType>()[Number<t>{}] *
+                    type_convert<AccDataType>(c_scale_thread_bufs(I0)[Number<cscale_offset>{}]);
+            });
+        }
+
+        StaticallyIndexedArray<ThreadStaticBuffer, Number<1>{}> c_scale_thread_bufs;
+        StaticBufferTupleOfVector<AddressSpaceEnum::Vgpr, AccDataType, 1, reg_size_per_wmma, true>
+            c_thread_buf_per_scale;
     };
 
     __host__ __device__ constexpr auto& GetCThreadBuffer() { return c_thread_buf_; }

include/ck/tensor_operation/gpu/block/blockwise_gemm_pipeline_wmmaops_v1.hpp

@@ -174,7 +174,9 @@ struct BlockwiseGemmWmmaops_pipeline_v1<BlockGemmPipelineScheduler::Intrawave,
               typename BBlockBuffer,
               typename BBlockTransferStep,
               typename CThreadBuffer,
-              typename BScaleStruct>
+              typename AScaleStruct,
+              typename BScaleStruct,
+              typename enable_if<ck::is_same_v<AScaleStruct, Empty>, bool>::type = false>
     __device__ void Run(const AGridDesc& a_grid_desc,
                         const ABlockDesc& a_block_desc,
                         ABlockTransfer& a_blockwise_copy,
@@ -188,7 +190,7 @@ struct BlockwiseGemmWmmaops_pipeline_v1<BlockGemmPipelineScheduler::Intrawave,
                         BBlockBuffer& b_block_buf,
                         const BBlockTransferStep& b_block_copy_step,
                         CThreadBuffer& c_thread_buf,
-                        // BScaleThreadCopy
+                        AScaleStruct&,
                         BScaleStruct& b_scale_struct,
                         index_t num_loop,
                         index_t num_loop_per_scale) const
@@ -207,6 +209,7 @@ struct BlockwiseGemmWmmaops_pipeline_v1<BlockGemmPipelineScheduler::Intrawave,
         a_blockwise_copy.MoveSrcSliceWindow(a_grid_desc, a_block_copy_step);
         b_blockwise_copy.MoveSrcSliceWindow(b_grid_desc, b_block_copy_step);
 
+        // Scales global load
         b_scale_struct.template GlobalLoad<0>(num_loop_per_scale == 1);
 
         // Local prefill 1
@@ -217,6 +220,7 @@ struct BlockwiseGemmWmmaops_pipeline_v1<BlockGemmPipelineScheduler::Intrawave,
         c_thread_buf.Clear();
 
         auto blockwise_gemm_func = [&]() {
+            // Local load
             static_for<0, KRepeat, 1>{}([&](auto k0) {
                 static_for<0, MRepeat, 1>{}([&](auto m0) {
                     a_thread_copy_.Run(a_block_desc_k0_m0_m1_m2_k1,
@@ -245,7 +249,7 @@ struct BlockwiseGemmWmmaops_pipeline_v1<BlockGemmPipelineScheduler::Intrawave,
                                     b_block_desc_k0_n0_n1_n2_k1,
                                     make_tuple(I0, n0, k0, I0, I0, I0, I0),
                                     b_block_buf,
-                                    b_scale_struct.b_scale_thread_bufs(
+                                    b_scale_struct.scale_thread_bufs(
                                         I0)[Number<n0 * BScaleStruct::num_scale_k_block +
                                                    k0 / BScaleStruct::num_scale_krepeat>{}],
                                     b_thread_desc_,
@@ -366,6 +370,189 @@ struct BlockwiseGemmWmmaops_pipeline_v1<BlockGemmPipelineScheduler::Intrawave,
         }
     }
 
+    template <bool HasMainLoop,
+              TailNumber TailNum,
+              typename AGridDesc,
+              typename ABlockDesc,
+              typename ABlockTransfer,
+              typename AGridBuffer,
+              typename ABlockBuffer,
+              typename ABlockTransferStep,
+              typename BGridDesc,
+              typename BBlockDesc,
+              typename BBlockTransfer,
+              typename BGridBuffer,
+              typename BBlockBuffer,
+              typename BBlockTransferStep,
+              typename CThreadBuffer,
+              typename AScaleStruct,
+              typename BScaleStruct,
+              typename enable_if<!ck::is_same_v<AScaleStruct, Empty> &&
+                                     !ck::is_same_v<BScaleStruct, Empty>,
+                                 bool>::type = false>
+    __device__ void Run(const AGridDesc& a_grid_desc,
+                        const ABlockDesc& a_block_desc,
+                        ABlockTransfer& a_blockwise_copy,
+                        const AGridBuffer& a_grid_buf,
+                        ABlockBuffer& a_block_buf,
+                        const ABlockTransferStep& a_block_copy_step,
+                        const BGridDesc& b_grid_desc,
+                        const BBlockDesc& b_block_desc,
+                        BBlockTransfer& b_blockwise_copy,
+                        const BGridBuffer& b_grid_buf,
+                        BBlockBuffer& b_block_buf,
+                        const BBlockTransferStep& b_block_copy_step,
+                        CThreadBuffer& c_thread_buf,
+                        AScaleStruct& a_scale_struct,
+                        BScaleStruct& b_scale_struct,
+                        index_t num_loop,
+                        index_t num_loop_per_scale) const
+    {
+        constexpr index_t KPerWaveBlock = wmma_gemm.GetKPerWaveBlk();
+        static constexpr auto NumScaleKBlock =
+            Number<ck::math::max(AScaleStruct::num_slice_k, BScaleStruct::num_slice_k)>{};
+
+        auto a_thread_buf = make_static_buffer<AddressSpaceEnum::Vgpr, ComputeTypeA>(
+            Base::a_thread_desc_.GetElementSpaceSize());
+        auto b_thread_buf = make_static_buffer<AddressSpaceEnum::Vgpr, ComputeTypeB>(
+            Base::b_thread_desc_.GetElementSpaceSize());
+
+        using CScaleStruct  = typename Base::template CScale<AScaleStruct, BScaleStruct>;
+        auto c_scale_struct = CScaleStruct{};
+
+        // Global prefetch 1
+        a_blockwise_copy.RunRead(a_grid_desc, a_grid_buf);
+        b_blockwise_copy.RunRead(b_grid_desc, b_grid_buf);
+
+        a_blockwise_copy.MoveSrcSliceWindow(a_grid_desc, a_block_copy_step);
+        b_blockwise_copy.MoveSrcSliceWindow(b_grid_desc, b_block_copy_step);
+
+        // Scales global load
+        a_scale_struct.template GlobalLoad<0>(num_loop_per_scale == 1);
+        b_scale_struct.template GlobalLoad<0>(num_loop_per_scale == 1);
+
+        c_scale_struct.Load(a_scale_struct, b_scale_struct);
+
+        // Local prefill 1
+        a_blockwise_copy.RunWrite(a_block_desc, a_block_buf);
+        b_blockwise_copy.RunWrite(b_block_desc, b_block_buf);
+
+        // Initialize C
+        c_thread_buf.Clear();
+
+        auto blockwise_gemm_func = [&]() {
+            // Local load
+            static_for<0, KRepeat, 1>{}([&](auto k0) {
+                static_for<0, MRepeat, 1>{}([&](auto m0) {
+                    Base::a_thread_copy_.Run(a_block_desc_k0_m0_m1_m2_k1,
+                                             make_tuple(I0, m0, k0, I0, I0, I0, I0),
+                                             a_block_buf,
+                                             Base::a_thread_desc_,
+                                             make_tuple(I0, m0, k0, I0, I0, I0, I0),
+                                             a_thread_buf);
+                });
+                static_for<0, NRepeat, 1>{}([&](auto n0) {
+                    Base::b_thread_copy_.Run(b_block_desc_k0_n0_n1_n2_k1,
+                                             make_tuple(I0, n0, k0, I0, I0, I0, I0),
+                                             b_block_buf,
+                                             Base::b_thread_desc_,
+                                             make_tuple(I0, n0, k0, I0, I0, I0, I0),
+                                             b_thread_buf);
+                });
+            });
+
+            static_for<0, MRepeat, 1>{}([&](auto m0) {
+                static_for<0, NRepeat, 1>{}([&](auto n0) {
+                    static_for<0, NumScaleKBlock, 1>{}([&](auto kscale0) {
+                        c_scale_struct.Clear();
+                        static_for<0, KRepeat / NumScaleKBlock, 1>{}([&](auto k0) {
+                            vector_type<ComputeTypeA, KPack / A_KRow / KInner> a_thread_vec;
+                            vector_type<ComputeTypeB, KPack / B_KRow / KInner> b_thread_vec;
+
+                            static_for<0, KInner, 1>{}([&](auto k_inner) {
+                                static_for<0, KPack / A_KRow / KInner, 1>{}([&](auto ik) {
+                                    constexpr index_t kk = ik + k_inner * KPerWaveBlock;
+                                    constexpr index_t k_index =
+                                        kscale0 * (KRepeat / NumScaleKBlock) + k0;
+                                    a_thread_vec.template AsType<ComputeTypeA>()(ik) =
+                                        a_thread_buf[Number<Base::a_thread_desc_.CalculateOffset(
+                                            make_tuple(Number<kk / A_K1>{},
+                                                       m0,
+                                                       k_index,
+                                                       I0,
+                                                       I0,
+                                                       I0,
+                                                       Number<kk % A_K1>{}))>{}];
+                                });
+                                static_for<0, KPack / B_KRow / KInner, 1>{}([&](auto ik) {
+                                    constexpr index_t kk = ik + k_inner * KPerWaveBlock;
+                                    constexpr index_t k_index =
+                                        kscale0 * (KRepeat / NumScaleKBlock) + k0;
+                                    b_thread_vec.template AsType<ComputeTypeB>()(ik) =
+                                        b_thread_buf[Number<Base::b_thread_desc_.CalculateOffset(
+                                            make_tuple(Number<kk / B_K1>{},
+                                                       n0,
+                                                       k_index,
+                                                       I0,
+                                                       I0,
+                                                       I0,
+                                                       Number<kk % B_K1>{}))>{}];
+                                });
+
+                                using wmma_input_type_a =
+                                    typename vector_type<ComputeTypeA, WmmaK / A_KRow>::type;
+                                using wmma_input_type_b =
+                                    typename vector_type<ComputeTypeB, WmmaK / B_KRow>::type;
+
+                                wmma_gemm.Run(
+                                    a_thread_vec.template AsType<wmma_input_type_a>(),
+                                    b_thread_vec.template AsType<wmma_input_type_b>(),
+                                    c_scale_struct.c_thread_buf_per_scale.GetVectorTypeReference(
+                                        Number<0>{}));
+                            });
+                        });
+                        c_scale_struct.template UpdateCThreadBuf<kscale0, m0, n0>(c_thread_buf);
+                    });
+                });
+            });
+        };
+
+        // main body
+        if constexpr(HasMainLoop)
+        {
+            index_t i = 0;
+            do
+            {
+                a_blockwise_copy.RunRead(a_grid_desc, a_grid_buf);
+                b_blockwise_copy.RunRead(b_grid_desc, b_grid_buf);
+
+                a_blockwise_copy.MoveSrcSliceWindow(a_grid_desc, a_block_copy_step);
+                b_blockwise_copy.MoveSrcSliceWindow(b_grid_desc, b_block_copy_step);
+
+                a_scale_struct.template GlobalLoad<0>((i + 2) % num_loop_per_scale == 0);
+                b_scale_struct.template GlobalLoad<0>((i + 2) % num_loop_per_scale == 0);
+
+                block_sync_lds();
+                blockwise_gemm_func();
+
+                block_sync_lds();
+                c_scale_struct.Load(a_scale_struct, b_scale_struct);
+
+                a_blockwise_copy.RunWrite(a_block_desc, a_block_buf);
+                b_blockwise_copy.RunWrite(b_block_desc, b_block_buf);
+
+                i += 1;
+            } while(i < (num_loop - 1));
+        }
+
+        // tail
+        if constexpr(TailNum == TailNumber::Full)
+        {
+            block_sync_lds();
+            blockwise_gemm_func();
+        }
+    }
+
     protected:
     // A[MRepeat, I1, I1, KPack]
     static constexpr auto a_thread_desc_ = make_naive_tensor_descriptor_packed(
@@ -528,6 +715,23 @@ struct BlockwiseGemmWmmaops_pipeline_v1<BlockGemmPipelineScheduler::Interwave,
         return TailNumber::Full;
     }
 
+    template <typename AScaleStruct, typename BScaleStruct>
+    struct KLoopParams
+    {
+        static constexpr auto KRepeatNoScale = 1;
+        static constexpr auto NumScaleKBlock =
+            Number<ck::math::max(AScaleStruct::num_slice_k, BScaleStruct::num_slice_k)>{};
+        static constexpr auto KRepeatPerNumScaleKBlock = KRepeatPerCluster / NumScaleKBlock;
+    };
+
+    template <>
+    struct KLoopParams<Empty, Empty>
+    {
+        static constexpr index_t KRepeatNoScale           = KRepeatPerCluster;
+        static constexpr index_t NumScaleKBlock           = 1;
+        static constexpr index_t KRepeatPerNumScaleKBlock = 1;
+    };
+
     template <bool HasMainLoop,
               TailNumber TailNum,
               typename AGridDesc,
@@ -543,7 +747,9 @@ struct BlockwiseGemmWmmaops_pipeline_v1<BlockGemmPipelineScheduler::Interwave,
               typename BBlockBuffer,
               typename BBlockTransferStep,
               typename CThreadBuffer,
-              typename BScaleStruct>
+              typename AScaleStruct,
+              typename BScaleStruct,
+              typename enable_if<ck::is_same_v<AScaleStruct, Empty>, bool>::type = false>
     __device__ void Run(const AGridDesc& a_grid_desc,
                         const ABlockDesc& a_block_desc,
                         ABlockTransfer& a_blockwise_copy,
@@ -557,7 +763,7 @@ struct BlockwiseGemmWmmaops_pipeline_v1<BlockGemmPipelineScheduler::Interwave,
                         BBlockBuffer& b_block_buf,
                         const BBlockTransferStep& b_block_copy_step,
                         CThreadBuffer& c_thread_buf,
-                        // BScaleThreadCopy
+                        AScaleStruct&,
                         BScaleStruct& b_scale_struct,
                         index_t num_loop,
                         index_t num_loop_per_scale) const
@@ -576,6 +782,7 @@ struct BlockwiseGemmWmmaops_pipeline_v1<BlockGemmPipelineScheduler::Interwave,
         a_blockwise_copy.MoveSrcSliceWindow(a_grid_desc, a_block_copy_step);
         b_blockwise_copy.MoveSrcSliceWindow(b_grid_desc, b_block_copy_step);
 
+        // Scales global load
         b_scale_struct.template GlobalLoad<0>(num_loop_per_scale == 1);
 
         // Local prefill 1
@@ -615,7 +822,7 @@ struct BlockwiseGemmWmmaops_pipeline_v1<BlockGemmPipelineScheduler::Interwave,
                                 b_block_desc_k0_n0_n1_n2_k1,
                                 make_tuple(I0, n0, k0_offset + k0_inner, I0, I0, I0, I0),
                                 b_block_buf,
-                                b_scale_struct.b_scale_thread_bufs(I0)[Number<
+                                b_scale_struct.scale_thread_bufs(I0)[Number<
                                     n0 * BScaleStruct::num_scale_k_block +
                                     (k0_offset + k0_inner) / BScaleStruct::num_scale_krepeat>{}],
                                 b_thread_desc_,
@@ -704,6 +911,7 @@ struct BlockwiseGemmWmmaops_pipeline_v1<BlockGemmPipelineScheduler::Interwave,
                         });
                     });
                 });
+
                 __builtin_amdgcn_sched_barrier(0);
                 __builtin_amdgcn_s_setprio(0);
                 __builtin_amdgcn_sched_barrier(0);
@@ -982,7 +1190,9 @@ struct BlockwiseGemmWmmaops_pipeline_v1<BlockGemmPipelineScheduler::Intrawave,
               typename BBlockBuffer,
               typename BBlockTransferStep,
               typename CThreadBuffer,
-              typename BScaleStruct>
+              typename AScaleStruct,
+              typename BScaleStruct,
+              typename enable_if<ck::is_same_v<AScaleStruct, Empty>, bool>::type = false>
     __device__ void Run(const AGridDesc& a_grid_desc,
                         const ABlockDesc& a_block_desc,
                         ABlockTransfer& a_blockwise_copy,
@@ -996,7 +1206,7 @@ struct BlockwiseGemmWmmaops_pipeline_v1<BlockGemmPipelineScheduler::Intrawave,
                         BBlockBuffer&,
                         const BBlockTransferStep& b_block_copy_step,
                         CThreadBuffer& c_thread_buf,
-                        // BScaleThreadCopy
+                        AScaleStruct&,
                         BScaleStruct&,
                         index_t num_loop,
                         index_t) const
@@ -1319,6 +1529,248 @@ struct BlockwiseGemmWmmaops_pipeline_v1<BlockGemmPipelineScheduler::Intrawave,
         }
     }
 
+    template <bool HasMainLoop,
+              TailNumber TailNum,
+              typename AGridDesc,
+              typename ABlockDesc,
+              typename ABlockTransfer,
+              typename AGridBuffer,
+              typename ABlockBuffer,
+              typename ABlockTransferStep,
+              typename BGridDesc,
+              typename BBlockDesc,
+              typename BBlockTransfer,
+              typename BGridBuffer,
+              typename BBlockBuffer,
+              typename BBlockTransferStep,
+              typename CThreadBuffer,
+              typename AScaleStruct,
+              typename BScaleStruct,
+              typename enable_if<!ck::is_same_v<AScaleStruct, Empty> &&
+                                     !ck::is_same_v<BScaleStruct, Empty>,
+                                 bool>::type = false>
+    __device__ void Run(const AGridDesc& a_grid_desc,
+                        const ABlockDesc& a_block_desc,
+                        ABlockTransfer& a_blockwise_copy,
+                        const AGridBuffer& a_grid_buf,
+                        ABlockBuffer& a_block_buf,
+                        const ABlockTransferStep& a_block_copy_step,
+                        const BGridDesc& b_grid_desc,
+                        const BBlockDesc&,
+                        BBlockTransfer& b_blockwise_copy,
+                        const BGridBuffer& b_grid_buf,
+                        BBlockBuffer&,
+                        const BBlockTransferStep& b_block_copy_step,
+                        CThreadBuffer& c_thread_buf,
+                        AScaleStruct& a_scale_struct,
+                        BScaleStruct& b_scale_struct,
+                        index_t num_loop,
+                        index_t num_loop_per_scale) const
+    {
+        __builtin_amdgcn_sched_barrier(0);
+        constexpr index_t KPerWaveBlock = wmma_gemm.GetKPerWaveBlk();
+        static constexpr auto NumScaleKBlock =
+            Number<ck::math::max(AScaleStruct::num_slice_k, BScaleStruct::num_slice_k)>{};
+
+        auto a_thread_buf = make_static_buffer<AddressSpaceEnum::Vgpr, ComputeTypeA>(
+            a_thread_desc_.GetElementSpaceSize());
+        auto b_thread_buf = make_static_buffer<AddressSpaceEnum::Vgpr, ComputeTypeB>(
+            b_thread_desc_.GetElementSpaceSize());
+
+        StaticallyIndexedArray<decltype(b_thread_buf), Number<2>{}> b_thread_bufs;
+        constexpr auto b_block_origin_idx = make_tuple(I0, I0, I0, I0, I0, I0, I0);
+
+        using CScaleStruct  = typename Base::template CScale<AScaleStruct, BScaleStruct>;
+        auto c_scale_struct = CScaleStruct{};
+
+        auto gemm_core_func = [&](auto reg_buf) {
+            static_for<0, MRepeat, 1>{}([&](auto m0) {
+                static_for<0, NRepeat, 1>{}([&](auto n0) {
+                    static_for<0, NumScaleKBlock, 1>{}([&](auto kscale0) {
+                        c_scale_struct.Clear();
+                        static_for<0, KRepeat / NumScaleKBlock, 1>{}([&](auto k0) {
+                            vector_type<ComputeTypeA, KPack / A_KRow / KInner> a_thread_vec;
+                            vector_type<ComputeTypeB, KPack / B_KRow / KInner> b_thread_vec;
+                            static_for<0, KInner, 1>{}([&](auto k_inner) {
+                                static_for<0, KPack / A_KRow / KInner, 1>{}([&](auto ik) {
+                                    constexpr index_t kk = ik + k_inner * KPerWaveBlock;
+                                    constexpr index_t k_index =
+                                        kscale0 * (KRepeat / NumScaleKBlock) + k0;
+                                    a_thread_vec.template AsType<ComputeTypeA>()(ik) =
+                                        a_thread_buf[Number<a_thread_desc_.CalculateOffset(
+                                            make_tuple(Number<kk / A_K1>{},
+                                                       m0,
+                                                       k_index,
+                                                       I0,
+                                                       I0,
+                                                       I0,
+                                                       Number<kk % A_K1>{}))>{}];
+                                });
+                                static_for<0, KPack / B_KRow / KInner, 1>{}([&](auto ik) {
+                                    constexpr index_t kk = ik + k_inner * KPerWaveBlock;
+                                    constexpr index_t k_index =
+                                        kscale0 * (KRepeat / NumScaleKBlock) + k0;
+                                    b_thread_vec.template AsType<ComputeTypeB>()(ik) =
+                                        b_thread_bufs[reg_buf]
+                                                     [Number<b_thread_desc_.CalculateOffset(
+                                                         make_tuple(Number<kk / B_K1>{},
+                                                                    I0,
+                                                                    I0,
+                                                                    n0,
+                                                                    I0,
+                                                                    k_index,
+                                                                    Number<kk % B_K1>{}))>{}];
+                                });
+                                using wmma_input_type_a =
+                                    typename vector_type<ComputeTypeA, WmmaK / A_KRow>::type;
+                                using wmma_input_type_b =
+                                    typename vector_type<ComputeTypeB, WmmaK / B_KRow>::type;
+                                wmma_gemm.Run(
+                                    a_thread_vec.template AsType<wmma_input_type_a>(),
+                                    b_thread_vec.template AsType<wmma_input_type_b>(),
+                                    c_scale_struct.c_thread_buf_per_scale.GetVectorTypeReference(
+                                        Number<0>{}));
+                            });
+                        });
+                        c_scale_struct.template UpdateCThreadBuf<kscale0, m0, n0>(c_thread_buf);
+                    });
+                });
+            });
+        };
+
+        auto a_local_prefetch_func = [&]() {
+            static_for<0, MRepeat, 1>{}([&](auto m0) {
+                static_for<0, KRepeat, 1>{}([&](auto k0) {
+                    a_thread_copy_.Run(a_block_desc_k0_m0_m1_m2_k1,
+                                       make_tuple(I0, m0, k0, I0, I0, I0, I0),
+                                       a_block_buf,
+                                       a_thread_desc_,
+                                       make_tuple(I0, m0, k0, I0, I0, I0, I0),
+                                       a_thread_buf);
+                });
+            });
+        };
+
+        // Global prefetch A1 B1
+        a_blockwise_copy.RunRead(a_grid_desc, a_grid_buf);
+        b_blockwise_copy.Run(b_grid_desc,
+                             b_grid_buf,
+                             b_block_desc_k0_n0_n1_n2_k1,
+                             b_block_origin_idx,
+                             b_thread_bufs(I0));
+
+        a_blockwise_copy.MoveSrcSliceWindow(a_grid_desc, a_block_copy_step);
+        b_blockwise_copy.MoveSrcSliceWindow(b_grid_desc, b_block_copy_step);
+
+        // Scales global load
+        a_scale_struct.template GlobalLoad<0>(num_loop_per_scale == 1);
+        b_scale_struct.template GlobalLoad<0>(num_loop_per_scale == 1);
+
+        __builtin_amdgcn_sched_barrier(0);
+
+        c_scale_struct.Load(a_scale_struct, b_scale_struct);
+
+        // Local prefill A1
+        a_blockwise_copy.RunWrite(a_block_desc, a_block_buf);
+
+        // Global prefetch A2
+        a_blockwise_copy.RunRead(a_grid_desc, a_grid_buf, I0);
+        a_blockwise_copy.MoveSrcSliceWindow(a_grid_desc, a_block_copy_step);
+
+        // Local prefetch A1
+        block_sync_lds();
+        a_local_prefetch_func();
+
+        // Initialize C
+        c_thread_buf.Clear();
+
+        __builtin_amdgcn_sched_barrier(0);
+
+        // main body
+        if constexpr(HasMainLoop)
+        {
+            index_t i = 0;
+            do
+            {
+                auto LoopFunc = [&](auto wmma_reg_buf, auto local_read_buf) {
+                    b_blockwise_copy.Run(b_grid_desc,
+                                         b_grid_buf,
+                                         b_block_desc_k0_n0_n1_n2_k1,
+                                         b_block_origin_idx,
+                                         b_thread_bufs(local_read_buf));
+
+                    b_blockwise_copy.MoveSrcSliceWindow(b_grid_desc, b_block_copy_step);
+
+                    block_sync_lds();
+
+                    a_blockwise_copy.RunWrite(a_block_desc, a_block_buf, wmma_reg_buf);
+
+                    a_blockwise_copy.RunRead(a_grid_desc, a_grid_buf, local_read_buf);
+                    a_blockwise_copy.MoveSrcSliceWindow(a_grid_desc, a_block_copy_step);
+
+                    a_scale_struct.template GlobalLoad<0>(
+                        (i + 2 + wmma_reg_buf) % num_loop_per_scale == 0);
+                    b_scale_struct.template GlobalLoad<0>(
+                        (i + 2 + wmma_reg_buf) % num_loop_per_scale == 0);
+
+                    gemm_core_func(wmma_reg_buf);
+
+                    block_sync_lds();
+
+                    // loop prefetch copy
+                    a_local_prefetch_func();
+
+                    c_scale_struct.Load(a_scale_struct, b_scale_struct);
+
+                    // HotLoopScheduler();
+                    __builtin_amdgcn_sched_barrier(0);
+                };
+
+                LoopFunc(I0, I1);
+                LoopFunc(I1, I0);
+
+                i += 2;
+            } while(i < (num_loop - 2));
+        }
+
+        // tail
+        if constexpr(TailNum == TailNumber::Even)
+        {
+            b_blockwise_copy.Run(b_grid_desc,
+                                 b_grid_buf,
+                                 b_block_desc_k0_n0_n1_n2_k1,
+                                 b_block_origin_idx,
+                                 b_thread_bufs(I1));
+
+            block_sync_lds();
+
+            a_blockwise_copy.RunWrite(a_block_desc, a_block_buf);
+
+            a_scale_struct.template GlobalLoad<0>(num_loop % num_loop_per_scale == 0);
+            b_scale_struct.template GlobalLoad<0>(num_loop % num_loop_per_scale == 0);
+
+            gemm_core_func(I0);
+
+            block_sync_lds();
+
+            // tail Local Prefetch A1
+            a_local_prefetch_func();
+
+            c_scale_struct.Load(a_scale_struct, b_scale_struct);
+
+            __builtin_amdgcn_sched_barrier(0);
+
+            gemm_core_func(I1);
+            // Let's leak last WMMA block to epilogue region, cover the potential lds-shuffle
+            // latency
+            // __builtin_amdgcn_sched_barrier(0);
+        }
+        else if constexpr(TailNum == TailNumber::Odd)
+        {
+            gemm_core_func(I0);
+        }
+    }
+
     protected:
     static constexpr auto b_thread_desc_ =
         make_naive_tensor_descriptor_packed(make_tuple(Number<KPack / B_K1 / B_KRow>{},

include/ck/tensor_operation/gpu/block/blockwise_gemm_pipeline_wmmaops_v3.hpp

@@ -123,6 +123,9 @@ struct BlockwiseGemmWmmaops_pipeline_v3<BlockGemmPipelineScheduler::Intrawave,
                                                     KInner,
                                                     TransposeC>;
     using Base::I0;
+    using Base::I1;
+    using Base::I2;
+    using Base::I3;
 
     using Base::A_K1;
     using Base::A_KRow;
@@ -322,7 +325,7 @@ struct BlockwiseGemmWmmaops_pipeline_v3<BlockGemmPipelineScheduler::Intrawave,
                     b_thread_copy_.Run(b_block_desc_k0_n0_n1_n2_k1,
                                        make_tuple(I0, n0, k0, I0, I0, I0, I0),
                                        b_block_buf,
-                                       b_scale_struct.b_scale_thread_bufs(
+                                       b_scale_struct.scale_thread_bufs(
                                            I0)[Number<n0 * BScaleStruct::num_scale_k_block +
                                                       k0 / BScaleStruct::num_scale_krepeat>{}],
                                        b_thread_desc_,
@@ -348,7 +351,9 @@ struct BlockwiseGemmWmmaops_pipeline_v3<BlockGemmPipelineScheduler::Intrawave,
               typename BBlockBuffer,
               typename BBlockTransferStep,
               typename CThreadBuffer,
-              typename BScaleStruct>
+              typename AScaleStruct,
+              typename BScaleStruct,
+              typename enable_if<ck::is_same_v<AScaleStruct, Empty>, bool>::type = false>
     __device__ void Run(const AGridDesc& a_grid_desc,
                         const ABlockDesc& a_block_desc,
                         ABlockTransfer& a_blockwise_copy,
@@ -362,7 +367,7 @@ struct BlockwiseGemmWmmaops_pipeline_v3<BlockGemmPipelineScheduler::Intrawave,
                         BBlockBuffer& b_block_buf,
                         const BBlockTransferStep& b_block_copy_step,
                         CThreadBuffer& c_thread_buf,
-                        // BScaleThreadCopy
+                        AScaleStruct&,
                         BScaleStruct& b_scale_struct,
                         index_t num_loop,
                         index_t num_loop_per_scale) const
@@ -383,6 +388,7 @@ struct BlockwiseGemmWmmaops_pipeline_v3<BlockGemmPipelineScheduler::Intrawave,
         a_blockwise_copy.MoveSrcSliceWindow(a_grid_desc, a_block_copy_step);
         b_blockwise_copy.MoveSrcSliceWindow(b_grid_desc, b_block_copy_step);
 
+        // Scales global load
         b_scale_struct.template GlobalLoad<0>(num_loop_per_scale == 1);
 
         // Local prefill 1
@@ -611,6 +617,339 @@ struct BlockwiseGemmWmmaops_pipeline_v3<BlockGemmPipelineScheduler::Intrawave,
         }
     }
 
+    template <bool HasMainLoop,
+              TailNumber TailNum,
+              typename AGridDesc,
+              typename ABlockDesc,
+              typename ABlockTransfer,
+              typename AGridBuffer,
+              typename ABlockBuffer,
+              typename ABlockTransferStep,
+              typename BGridDesc,
+              typename BBlockDesc,
+              typename BBlockTransfer,
+              typename BGridBuffer,
+              typename BBlockBuffer,
+              typename BBlockTransferStep,
+              typename CThreadBuffer,
+              typename AScaleStruct,
+              typename BScaleStruct,
+              typename enable_if<!ck::is_same_v<AScaleStruct, Empty> &&
+                                     !ck::is_same_v<BScaleStruct, Empty>,
+                                 bool>::type = false>
+    __device__ void Run(const AGridDesc& a_grid_desc,
+                        const ABlockDesc& a_block_desc,
+                        ABlockTransfer& a_blockwise_copy,
+                        const AGridBuffer& a_grid_buf,
+                        ABlockBuffer& a_block_buf,
+                        const ABlockTransferStep& a_block_copy_step,
+                        const BGridDesc& b_grid_desc,
+                        const BBlockDesc& b_block_desc,
+                        BBlockTransfer& b_blockwise_copy,
+                        const BGridBuffer& b_grid_buf,
+                        BBlockBuffer& b_block_buf,
+                        const BBlockTransferStep& b_block_copy_step,
+                        CThreadBuffer& c_thread_buf,
+                        AScaleStruct& a_scale_struct,
+                        BScaleStruct& b_scale_struct,
+                        index_t num_loop,
+                        index_t num_loop_per_scale) const
+    {
+        __builtin_amdgcn_sched_barrier(0);
+
+        constexpr index_t KPerWaveBlock = wmma_gemm.GetKPerWaveBlk();
+        static constexpr auto NumScaleKBlock =
+            Number<ck::math::max(AScaleStruct::num_slice_k, BScaleStruct::num_slice_k)>{};
+
+        auto a_thread_buf = make_static_buffer<AddressSpaceEnum::Vgpr, ComputeTypeA>(
+            a_thread_desc_.GetElementSpaceSize());
+        auto b_thread_buf = make_static_buffer<AddressSpaceEnum::Vgpr, ComputeTypeB>(
+            b_thread_desc_.GetElementSpaceSize());
+
+        using CScaleStruct  = typename Base::template CScale<AScaleStruct, BScaleStruct>;
+        auto c_scale_struct = CScaleStruct{};
+
+        // Global prefetch 1
+        a_blockwise_copy.RunRead(a_grid_desc, a_grid_buf);
+        b_blockwise_copy.RunRead(b_grid_desc, b_grid_buf);
+
+        a_blockwise_copy.MoveSrcSliceWindow(a_grid_desc, a_block_copy_step);
+        b_blockwise_copy.MoveSrcSliceWindow(b_grid_desc, b_block_copy_step);
+
+        // Scales global load
+        a_scale_struct.template GlobalLoad<0>(num_loop_per_scale == 1);
+        b_scale_struct.template GlobalLoad<0>(num_loop_per_scale == 1);
+
+        c_scale_struct.Load(a_scale_struct, b_scale_struct);
+
+        // Local prefill 1
+        a_blockwise_copy.RunWrite(a_block_desc, a_block_buf);
+        b_blockwise_copy.RunWrite(b_block_desc, b_block_buf);
+
+        // Global prefetch 2, perform when at least 2 loops exist.
+        if constexpr(TailNum == TailNumber::Even || TailNum == TailNumber::Full)
+        {
+            a_blockwise_copy.RunRead(a_grid_desc, a_grid_buf);
+            b_blockwise_copy.RunRead(b_grid_desc, b_grid_buf);
+
+            a_blockwise_copy.MoveSrcSliceWindow(a_grid_desc, a_block_copy_step);
+            b_blockwise_copy.MoveSrcSliceWindow(b_grid_desc, b_block_copy_step);
+        }
+
+        // Initialize C
+        c_thread_buf.Clear();
+
+        // Local prefetch 1
+        block_sync_lds();
+
+        auto local_load_func = [&]() {
+            static_for<0, KRepeat, 1>{}([&](auto k0) {
+                static_for<0, MRepeat, 1>{}([&](auto m0) {
+                    a_thread_copy_.Run(a_block_desc_k0_m0_m1_m2_k1,
+                                       make_tuple(I0, m0, k0, I0, I0, I0, I0),
+                                       a_block_buf,
+                                       a_thread_desc_,
+                                       make_tuple(I0, m0, k0, I0, I0, I0, I0),
+                                       a_thread_buf);
+                });
+                static_for<0, NRepeat, 1>{}([&](auto n0) {
+                    b_thread_copy_.Run(b_block_desc_k0_n0_n1_n2_k1,
+                                       make_tuple(I0, n0, k0, I0, I0, I0, I0),
+                                       b_block_buf,
+                                       b_thread_desc_,
+                                       make_tuple(I0, n0, k0, I0, I0, I0, I0),
+                                       b_thread_buf);
+                });
+            });
+        };
+
+        local_load_func();
+
+        __builtin_amdgcn_sched_barrier(0);
+
+        // Main body, perform when at least 3 loops exist.
+        if constexpr(HasMainLoop)
+        {
+            index_t i = 0;
+            do
+            {
+                block_sync_lds();
+
+                a_blockwise_copy.RunWrite(a_block_desc, a_block_buf);
+                b_blockwise_copy.RunWrite(b_block_desc, b_block_buf);
+
+                a_blockwise_copy.RunRead(a_grid_desc, a_grid_buf);
+                b_blockwise_copy.RunRead(b_grid_desc, b_grid_buf);
+
+                a_blockwise_copy.MoveSrcSliceWindow(a_grid_desc, a_block_copy_step);
+                b_blockwise_copy.MoveSrcSliceWindow(b_grid_desc, b_block_copy_step);
+
+                a_scale_struct.template GlobalLoad<0>((i + 2) % num_loop_per_scale == 0);
+                b_scale_struct.template GlobalLoad<0>((i + 2) % num_loop_per_scale == 0);
+
+                static_for<0, MRepeat, 1>{}([&](auto m0) {
+                    static_for<0, NRepeat, 1>{}([&](auto n0) {
+                        static_for<0, NumScaleKBlock, 1>{}([&](auto kscale0) {
+                            c_scale_struct.Clear();
+                            static_for<0, KRepeat / NumScaleKBlock, 1>{}([&](auto k0) {
+                                static_for<0, KInner, 1>{}([&](auto k_inner) {
+                                    vector_type<ComputeTypeA, KPack / A_KRow / KInner> a_thread_vec;
+                                    vector_type<ComputeTypeB, KPack / B_KRow / KInner> b_thread_vec;
+
+                                    static_for<0, KPack / A_KRow / KInner, 1>{}([&](auto ik) {
+                                        constexpr index_t kk = ik + k_inner * KPerWaveBlock;
+                                        constexpr index_t k_index =
+                                            kscale0 * (KRepeat / NumScaleKBlock) + k0;
+                                        a_thread_vec.template AsType<ComputeTypeA>()(ik) =
+                                            a_thread_buf[Number<a_thread_desc_.CalculateOffset(
+                                                make_tuple(Number<kk / A_K1>{},
+                                                           m0,
+                                                           k_index,
+                                                           I0,
+                                                           I0,
+                                                           I0,
+                                                           Number<kk % A_K1>{}))>{}];
+                                    });
+                                    static_for<0, KPack / B_KRow / KInner, 1>{}([&](auto ik) {
+                                        constexpr index_t kk = ik + k_inner * KPerWaveBlock;
+                                        constexpr index_t k_index =
+                                            kscale0 * (KRepeat / NumScaleKBlock) + k0;
+                                        b_thread_vec.template AsType<ComputeTypeB>()(ik) =
+                                            b_thread_buf[Number<b_thread_desc_.CalculateOffset(
+                                                make_tuple(Number<kk / B_K1>{},
+                                                           n0,
+                                                           k_index,
+                                                           I0,
+                                                           I0,
+                                                           I0,
+                                                           Number<kk % B_K1>{}))>{}];
+                                    });
+
+                                    using wmma_input_type_a =
+                                        typename vector_type<ComputeTypeA, WmmaK / A_KRow>::type;
+                                    using wmma_input_type_b =
+                                        typename vector_type<ComputeTypeB, WmmaK / B_KRow>::type;
+
+                                    wmma_gemm.Run(a_thread_vec.template AsType<wmma_input_type_a>(),
+                                                  b_thread_vec.template AsType<wmma_input_type_b>(),
+                                                  c_scale_struct.c_thread_buf_per_scale
+                                                      .GetVectorTypeReference(Number<0>{}));
+                                });
+                            });
+                            c_scale_struct.template UpdateCThreadBuf<kscale0, m0, n0>(c_thread_buf);
+                        });
+                    });
+                });
+
+                c_scale_struct.Load(a_scale_struct, b_scale_struct);
+                block_sync_lds();
+
+                local_load_func();
+
+                HotLoopScheduler();
+                __builtin_amdgcn_sched_barrier(0);
+
+                i += 1;
+            } while(i < (num_loop - 2));
+        }
+
+        // Pre-tail, perform when at least 2 loops exist.
+        if constexpr(TailNum == TailNumber::Even || TailNum == TailNumber::Full)
+        {
+            block_sync_lds();
+
+            a_blockwise_copy.RunWrite(a_block_desc, a_block_buf);
+            b_blockwise_copy.RunWrite(b_block_desc, b_block_buf);
+
+            // No RunRead or MoveSrcSliceWindow here, already finished them all!
+            a_scale_struct.template GlobalLoad<0>(num_loop % num_loop_per_scale == 0);
+            b_scale_struct.template GlobalLoad<0>(num_loop % num_loop_per_scale == 0);
+
+            static_for<0, MRepeat, 1>{}([&](auto m0) {
+                static_for<0, NRepeat, 1>{}([&](auto n0) {
+                    static_for<0, NumScaleKBlock, 1>{}([&](auto kscale0) {
+                        c_scale_struct.Clear();
+                        static_for<0, KRepeat / NumScaleKBlock, 1>{}([&](auto k0) {
+                            static_for<0, KInner, 1>{}([&](auto k_inner) {
+                                vector_type<ComputeTypeA, KPack / A_KRow / KInner> a_thread_vec;
+                                vector_type<ComputeTypeB, KPack / B_KRow / KInner> b_thread_vec;
+
+                                static_for<0, KPack / A_KRow / KInner, 1>{}([&](auto ik) {
+                                    constexpr index_t kk = ik + k_inner * KPerWaveBlock;
+                                    constexpr index_t k_index =
+                                        kscale0 * (KRepeat / NumScaleKBlock) + k0;
+                                    a_thread_vec.template AsType<ComputeTypeA>()(ik) =
+                                        a_thread_buf[Number<a_thread_desc_.CalculateOffset(
+                                            make_tuple(Number<kk / A_K1>{},
+                                                       m0,
+                                                       k_index,
+                                                       I0,
+                                                       I0,
+                                                       I0,
+                                                       Number<kk % A_K1>{}))>{}];
+                                });
+                                static_for<0, KPack / B_KRow / KInner, 1>{}([&](auto ik) {
+                                    constexpr index_t kk = ik + k_inner * KPerWaveBlock;
+                                    constexpr index_t k_index =
+                                        kscale0 * (KRepeat / NumScaleKBlock) + k0;
+                                    b_thread_vec.template AsType<ComputeTypeB>()(ik) =
+                                        b_thread_buf[Number<b_thread_desc_.CalculateOffset(
+                                            make_tuple(Number<kk / B_K1>{},
+                                                       n0,
+                                                       k_index,
+                                                       I0,
+                                                       I0,
+                                                       I0,
+                                                       Number<kk % B_K1>{}))>{}];
+                                });
+
+                                using wmma_input_type_a =
+                                    typename vector_type<ComputeTypeA, WmmaK / A_KRow>::type;
+                                using wmma_input_type_b =
+                                    typename vector_type<ComputeTypeB, WmmaK / B_KRow>::type;
+
+                                wmma_gemm.Run(
+                                    a_thread_vec.template AsType<wmma_input_type_a>(),
+                                    b_thread_vec.template AsType<wmma_input_type_b>(),
+                                    c_scale_struct.c_thread_buf_per_scale.GetVectorTypeReference(
+                                        Number<0>{}));
+                            });
+                        });
+                        c_scale_struct.template UpdateCThreadBuf<kscale0, m0, n0>(c_thread_buf);
+                    });
+                });
+            });
+
+            c_scale_struct.Load(a_scale_struct, b_scale_struct);
+            block_sync_lds();
+
+            local_load_func();
+
+            HotLoopScheduler();
+            __builtin_amdgcn_sched_barrier(0);
+        }
+
+        // Tail, always perform.
+        {
+            static_for<0, MRepeat, 1>{}([&](auto m0) {
+                static_for<0, NRepeat, 1>{}([&](auto n0) {
+                    static_for<0, NumScaleKBlock, 1>{}([&](auto kscale0) {
+                        c_scale_struct.Clear();
+                        static_for<0, KRepeat / NumScaleKBlock, 1>{}([&](auto k0) {
+                            vector_type<ComputeTypeA, KPack / A_KRow / KInner> a_thread_vec;
+                            vector_type<ComputeTypeB, KPack / B_KRow / KInner> b_thread_vec;
+                            static_for<0, KInner, 1>{}([&](auto k_inner) {
+                                static_for<0, KPack / A_KRow / KInner, 1>{}([&](auto ik) {
+                                    constexpr index_t kk = ik + k_inner * KPerWaveBlock;
+                                    constexpr index_t k_index =
+                                        kscale0 * (KRepeat / NumScaleKBlock) + k0;
+                                    a_thread_vec.template AsType<ComputeTypeA>()(ik) =
+                                        a_thread_buf[Number<a_thread_desc_.CalculateOffset(
+                                            make_tuple(Number<kk / A_K1>{},
+                                                       m0,
+                                                       k_index,
+                                                       I0,
+                                                       I0,
+                                                       I0,
+                                                       Number<kk % A_K1>{}))>{}];
+                                });
+                                static_for<0, KPack / B_KRow / KInner, 1>{}([&](auto ik) {
+                                    constexpr index_t kk = ik + k_inner * KPerWaveBlock;
+                                    constexpr index_t k_index =
+                                        kscale0 * (KRepeat / NumScaleKBlock) + k0;
+                                    b_thread_vec.template AsType<ComputeTypeB>()(ik) =
+                                        b_thread_buf[Number<b_thread_desc_.CalculateOffset(
+                                            make_tuple(Number<kk / B_K1>{},
+                                                       n0,
+                                                       k_index,
+                                                       I0,
+                                                       I0,
+                                                       I0,
+                                                       Number<kk % B_K1>{}))>{}];
+                                });
+
+                                using wmma_input_type_a =
+                                    typename vector_type<ComputeTypeA, WmmaK / A_KRow>::type;
+                                using wmma_input_type_b =
+                                    typename vector_type<ComputeTypeB, WmmaK / B_KRow>::type;
+
+                                wmma_gemm.Run(
+                                    a_thread_vec.template AsType<wmma_input_type_a>(),
+                                    b_thread_vec.template AsType<wmma_input_type_b>(),
+                                    c_scale_struct.c_thread_buf_per_scale.GetVectorTypeReference(
+                                        Number<0>{}));
+                            });
+                        });
+                        c_scale_struct.template UpdateCThreadBuf<kscale0, m0, n0>(c_thread_buf);
+                    });
+                });
+            });
+            // Let's leak last WMMA block to epilogue region, cover the potential lds-shuffle
+            // latency
+            // __builtin_amdgcn_sched_barrier(0);
+        }
+    }
+
     protected:
     using Base::a_thread_copy_;
     using Base::a_thread_desc_;

include/ck/tensor_operation/gpu/device/device_gemm_multiple_d_ab_scale.hpp

@@ -105,6 +105,353 @@ struct DeviceGemmMultipleD_BlockScale_BPreshuffle : public BaseOperator
     virtual int GetPreShuffleParameters() = 0;
 };
 
+template <typename ALayout,
+          typename BLayout,
+          typename DsLayout,
+          typename ELayout,
+          typename ADataType,
+          typename AScaleType,
+          typename BDataType,
+          typename BScaleType,
+          typename DsDataType,
+          typename EDataType,
+          index_t ScaleBlockM,
+          index_t ScaleBlockN,
+          index_t ScaleBlockK,
+          typename AElementwiseOperation,
+          typename BElementwiseOperation,
+          typename CDEElementwiseOperation>
+struct DeviceGemmMultipleD_BlockScale_BPreshuffleSplitK : public BaseOperator
+{
+    static constexpr index_t NumDTensor = DsDataType::Size();
+
+    virtual std::unique_ptr<BaseArgument>
+    MakeArgumentPointer(const void* p_a,
+                        const void* p_b,
+                        std::array<const void*, NumDTensor> p_ds,
+                        void* p_e,
+                        const ck::index_t M,
+                        const ck::index_t N,
+                        const ck::index_t K,
+                        const ck::index_t StrideA,
+                        const ck::index_t StrideB,
+                        const std::array<ck::index_t, NumDTensor> StrideDs,
+                        const ck::index_t StrideE,
+                        const void* p_a_scale,
+                        const void* p_b_scale,
+                        AElementwiseOperation a_element_op,
+                        BElementwiseOperation b_element_op,
+                        CDEElementwiseOperation cde_element_op,
+                        index_t KBatch) = 0;
+
+    virtual std::unique_ptr<BaseInvoker> MakeInvokerPointer() = 0;
+
+    virtual int GetPreShuffleParameters() = 0;
+};
+
+/// @brief Wrapper for backward compatibility that allows to use instances of
+///        DeviceGemmMultipleD_BlockScale_BPreshuffleSplitK in contexts where
+//         DeviceGemmMultipleD_BlockScale_BPreshuffle is expected.
+///
+/// @note  The main area where it can be used is DeviceOperationInstanceFactory::GetInstances().
+///        The only difference between API of DeviceGemmMultipleD_BlockScale_BPreshuffle and
+//         DeviceGemmMultipleD_BlockScale_BPreshuffleSplitK is
+///        that DeviceGemmMultipleD_BlockScale_BPreshuffleSplitK::MakeArgumentPointer requires
+//         an additional parameter KBatch which is explicitly passed as 1 by this wrapper.
+template <typename ALayout,
+          typename BLayout,
+          typename DsLayout,
+          typename ELayout,
+          typename ADataType,
+          typename AScaleType,
+          typename BDataType,
+          typename BScaleType,
+          typename DsDataType,
+          typename EDataType,
+          index_t ScaleBlockM,
+          index_t ScaleBlockN,
+          index_t ScaleBlockK,
+          typename AElementwiseOperation,
+          typename BElementwiseOperation,
+          typename CDEElementwiseOperation>
+struct DeviceGemmMultipleD_BlockScale_BPreshuffleWrapper
+    : public DeviceGemmMultipleD_BlockScale_BPreshuffle<ALayout,
+                                                        BLayout,
+                                                        DsLayout,
+                                                        ELayout,
+                                                        ADataType,
+                                                        AScaleType,
+                                                        BDataType,
+                                                        BScaleType,
+                                                        DsDataType,
+                                                        EDataType,
+                                                        ScaleBlockM,
+                                                        ScaleBlockN,
+                                                        ScaleBlockK,
+                                                        AElementwiseOperation,
+                                                        BElementwiseOperation,
+                                                        CDEElementwiseOperation>
+{
+    using DeviceOp = DeviceGemmMultipleD_BlockScale_BPreshuffleSplitK<ALayout,
+                                                                      BLayout,
+                                                                      DsLayout,
+                                                                      ELayout,
+                                                                      ADataType,
+                                                                      AScaleType,
+                                                                      BDataType,
+                                                                      BScaleType,
+                                                                      DsDataType,
+                                                                      EDataType,
+                                                                      ScaleBlockM,
+                                                                      ScaleBlockN,
+                                                                      ScaleBlockK,
+                                                                      AElementwiseOperation,
+                                                                      BElementwiseOperation,
+                                                                      CDEElementwiseOperation>;
+
+    static constexpr index_t NumDTensor = DsDataType::Size();
+
+#ifndef __HIPCC_RTC__
+
+    explicit DeviceGemmMultipleD_BlockScale_BPreshuffleWrapper(std::unique_ptr<DeviceOp> p_op)
+        : p_op_(std::move(p_op))
+    {
+    }
+
+    bool IsSupportedArgument(const BaseArgument* p_arg) override
+    {
+        return p_op_->IsSupportedArgument(p_arg);
+    }
+    std::unique_ptr<BaseArgument>
+    MakeArgumentPointer(const void* p_a,
+                        const void* p_b,
+                        std::array<const void*, NumDTensor> p_ds,
+                        void* p_e,
+                        const ck::index_t M,
+                        const ck::index_t N,
+                        const ck::index_t K,
+                        const ck::index_t StrideA,
+                        const ck::index_t StrideB,
+                        const std::array<ck::index_t, NumDTensor> StrideDs,
+                        const ck::index_t StrideE,
+                        const void* p_a_scale,
+                        const void* p_b_scale,
+                        AElementwiseOperation a_element_op,
+                        BElementwiseOperation b_element_op,
+                        CDEElementwiseOperation cde_element_op) override
+    {
+        return p_op_->MakeArgumentPointer(p_a,
+                                          p_b,
+                                          p_ds,
+                                          p_e,
+                                          M,
+                                          N,
+                                          K,
+                                          StrideA,
+                                          StrideB,
+                                          StrideDs,
+                                          StrideE,
+                                          p_a_scale,
+                                          p_b_scale,
+                                          a_element_op,
+                                          b_element_op,
+                                          cde_element_op,
+                                          1); // KBatch
+    }
+
+    std::unique_ptr<BaseInvoker> MakeInvokerPointer() override
+    {
+        return p_op_->MakeInvokerPointer();
+    }
+
+    int GetPreShuffleParameters() override { return p_op_->GetPreShuffleParameters(); }
+
+    std::string GetTypeString() const override { return p_op_->GetTypeString(); }
+
+    private:
+    std::unique_ptr<DeviceOp> p_op_;
+
+#endif // __HIPCC_RTC__
+};
+
+// GEMM:
+//   input : A[M, K], B[K, N],
+//   input : D0[M, N], D1[M, N], ...
+//   output : E[M, N]
+//   C = a_op(A) * b_op(B)
+//   E = cde_op(C, D0, D1, ...)
+// Assume:
+//   D0, D1, ... and E have the same layout
+template <typename ALayout,
+          typename BLayout,
+          typename DsLayout,
+          typename ELayout,
+          typename ADataType,
+          typename AScaleType,
+          typename BDataType,
+          typename BScaleType,
+          typename DsDataType,
+          typename EDataType,
+          index_t ScaleBlockM,
+          index_t ScaleBlockN,
+          index_t ScaleBlockK,
+          typename AElementwiseOperation,
+          typename BElementwiseOperation,
+          typename CDEElementwiseOperation>
+struct DeviceGemmMultipleD_ABScaleSplitK : public BaseOperator
+{
+    static constexpr index_t NumDTensor = DsDataType::Size();
+
+    virtual std::unique_ptr<BaseArgument>
+    MakeArgumentPointer(const void* p_a,
+                        const void* p_b,
+                        std::array<const void*, NumDTensor> p_ds,
+                        void* p_e,
+                        const ck::index_t M,
+                        const ck::index_t N,
+                        const ck::index_t K,
+                        const ck::index_t StrideA,
+                        const ck::index_t StrideB,
+                        const std::array<ck::index_t, NumDTensor> StrideDs,
+                        const ck::index_t StrideE,
+                        const void* p_a_scale,
+                        const void* p_b_scale,
+                        AElementwiseOperation a_element_op,
+                        BElementwiseOperation b_element_op,
+                        CDEElementwiseOperation cde_element_op,
+                        index_t KBatch) = 0;
+
+    virtual std::unique_ptr<BaseInvoker> MakeInvokerPointer() = 0;
+
+    virtual void SetKBatch(BaseArgument* arg, int KBatch) const = 0;
+};
+
+/// @brief Wrapper for backward compatibility that allows to use instances of
+///        DeviceGemmMultipleD_ABScaleSplitK in contexts where DeviceGemmMultipleD_ABScale is
+///        expected.
+///
+/// @note  The main area where it can be used is DeviceOperationInstanceFactory::GetInstances().
+///        The only difference between API of DeviceGemmMultipleD_ABScale and
+///        DeviceGemmMultipleD_ABScaleSplitK is that
+///        DeviceGemmMultipleD_ABScaleSplitK::MakeArgumentPointer requires a additional parameter
+///        KBatch which is explicitly passed as 1 by this wrapper.
+template <typename ALayout,
+          typename BLayout,
+          typename DsLayout,
+          typename ELayout,
+          typename ADataType,
+          typename AScaleType,
+          typename BDataType,
+          typename BScaleType,
+          typename DsDataType,
+          typename EDataType,
+          index_t ScaleBlockM,
+          index_t ScaleBlockN,
+          index_t ScaleBlockK,
+          typename AElementwiseOperation,
+          typename BElementwiseOperation,
+          typename CDEElementwiseOperation>
+struct DeviceGemmMultipleD_ABScaleSplitKWrapper
+    : public DeviceGemmMultipleD_ABScale<ALayout,
+                                         BLayout,
+                                         DsLayout,
+                                         ELayout,
+                                         ADataType,
+                                         AScaleType,
+                                         BDataType,
+                                         BScaleType,
+                                         DsDataType,
+                                         EDataType,
+                                         ScaleBlockM,
+                                         ScaleBlockN,
+                                         ScaleBlockK,
+                                         AElementwiseOperation,
+                                         BElementwiseOperation,
+                                         CDEElementwiseOperation>
+{
+
+    using DeviceOp = DeviceGemmMultipleD_ABScaleSplitK<ALayout,
+                                                       BLayout,
+                                                       DsLayout,
+                                                       ELayout,
+                                                       ADataType,
+                                                       AScaleType,
+                                                       BDataType,
+                                                       BScaleType,
+                                                       DsDataType,
+                                                       EDataType,
+                                                       ScaleBlockM,
+                                                       ScaleBlockN,
+                                                       ScaleBlockK,
+                                                       AElementwiseOperation,
+                                                       BElementwiseOperation,
+                                                       CDEElementwiseOperation>;
+
+    static constexpr index_t NumDTensor = DsDataType::Size();
+
+#ifndef __HIPCC_RTC__
+
+    explicit DeviceGemmMultipleD_ABScaleSplitKWrapper(std::unique_ptr<DeviceOp> p_op)
+        : p_op_(std::move(p_op))
+    {
+    }
+
+    bool IsSupportedArgument(const BaseArgument* p_arg) override
+    {
+        return p_op_->IsSupportedArgument(p_arg);
+    }
+    std::unique_ptr<BaseArgument>
+    MakeArgumentPointer(const void* p_a,
+                        const void* p_b,
+                        std::array<const void*, NumDTensor> p_ds,
+                        void* p_e,
+                        const ck::index_t M,
+                        const ck::index_t N,
+                        const ck::index_t K,
+                        const ck::index_t StrideA,
+                        const ck::index_t StrideB,
+                        const std::array<ck::index_t, NumDTensor> StrideDs,
+                        const ck::index_t StrideE,
+                        const void* p_a_scale,
+                        const void* p_b_scale,
+                        AElementwiseOperation a_element_op,
+                        BElementwiseOperation b_element_op,
+                        CDEElementwiseOperation cde_element_op) override
+    {
+        return p_op_->MakeArgumentPointer(p_a,
+                                          p_b,
+                                          p_ds,
+                                          p_e,
+                                          M,
+                                          N,
+                                          K,
+                                          StrideA,
+                                          StrideB,
+                                          StrideDs,
+                                          StrideE,
+                                          p_a_scale,
+                                          p_b_scale,
+                                          a_element_op,
+                                          b_element_op,
+                                          cde_element_op,
+                                          1); // KBatch
+    }
+
+    void SetKBatch(BaseArgument* arg, int KBatch) const override { p_op_->SetKBatch(arg, KBatch); }
+
+    std::unique_ptr<BaseInvoker> MakeInvokerPointer() override
+    {
+        return p_op_->MakeInvokerPointer();
+    }
+
+    std::string GetTypeString() const override { return p_op_->GetTypeString(); }
+
+    private:
+    std::unique_ptr<DeviceOp> p_op_;
+
+#endif // __HIPCC_RTC__
+};
+
 } // namespace device
 } // namespace tensor_operation
 } // namespace ck

include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_wmma_cshuffle_v3_b_scale.hpp

@@ -12,7 +12,7 @@
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
 #include "ck/tensor_operation/gpu/device/device_batched_gemm.hpp"
 #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/grid/gridwise_gemm_wmma_cshuffle_v3_b_scale.hpp"
+#include "ck/tensor_operation/gpu/grid/gridwise_gemm_wmma_cshuffle_v3_ab_scale.hpp"
 #include "ck/host_utility/device_prop.hpp"
 #include "ck/host_utility/kernel_launch.hpp"
 #include "ck/host_utility/flush_cache.hpp"
@@ -93,7 +93,8 @@ __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, MinimumOccupancy)
             p_bs_grid_shift,
             karg.p_ds_grid,
             karg.p_e_grid + splitk_batch_offset.c_reduce_offset + c_batch_offset,
-            karg.p_b_scale_grid + b_scale_batch_offset + splitk_batch_offset.scale_k_split_offset,
+            karg.p_a_scale_grid,
+            karg.p_b_scale_grid + b_scale_batch_offset + splitk_batch_offset.scale_b_k_split_offset,
             p_shared,
             karg,
             karg.a_element_op,
@@ -315,12 +316,13 @@ struct DeviceBatchedGemm_Wmma_CShuffleV3_BScale
     };
 
     // GridwiseGemm
-    using GridwiseGemm = GridwiseGemm_wmma_cshuffle_v3_b_scale<
+    using GridwiseGemm = GridwiseGemm_wmma_cshuffle_v3_ab_scale<
         ALayout,
         BLayout,
         Tuple<>, // DsLayout
         CLayout,
         Tuple<ADataType>,
+        void, // AScaleType
         Tuple<BDataType>,
         BScaleDataType,
         AccDataType,
@@ -332,6 +334,7 @@ struct DeviceBatchedGemm_Wmma_CShuffleV3_BScale
         CElementwiseOperation,
         GemmSpec,
         BlockSize,
+        0, // ScaleBlockM
         ScaleBlockN,
         ScaleBlockK,
         MPerBlock,
@@ -405,7 +408,9 @@ struct DeviceBatchedGemm_Wmma_CShuffleV3_BScale
                                      std::array<index_t, 1>{StrideB_},
                                      std::array<index_t, 0>{}, // StrideDs_
                                      StrideC_,
+                                     0, // StrideScaleA
                                      StrideScaleB_,
+                                     nullptr,
                                      p_b_scale_grid_,
                                      k_batch_,
                                      a_element_op_,

include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_wmma_cshuffle_v3_ab_scale.hpp

@@ -0,0 +1,362 @@
+// Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
+// SPDX-License-Identifier: MIT
+
+#pragma once
+
+#include <iostream>
+#include <sstream>
+
+#include "ck/utility/common_header.hpp"
+#include "ck/tensor_description/tensor_descriptor.hpp"
+#include "ck/tensor_description/tensor_descriptor_helper.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_gemm_multiple_d_ab_scale.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/grid/gridwise_gemm_wmma_cshuffle_v3_ab_scale.hpp"
+#include "ck/host_utility/device_prop.hpp"
+#include "ck/host_utility/kernel_launch.hpp"
+#include "ck/host_utility/flush_cache.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_wmma_cshuffle_v3_common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+
+template <typename ALayout,
+          typename BLayout,
+          typename DsLayout,
+          typename CLayout,
+          typename ADataType,
+          typename AScaleDataType,
+          typename BDataType,
+          typename BScaleDataType,
+          typename DsDataType,
+          typename CDataType,
+          typename AccDataType,
+          typename CShuffleDataType,
+          typename AElementwiseOperation,
+          typename BElementwiseOperation,
+          typename CElementwiseOperation,
+          GemmSpecialization GemmSpec,
+          index_t BlockSize,
+          index_t ScaleBlockM, // scale block for M
+          index_t ScaleBlockN, // scale block for N
+          index_t ScaleBlockK, // scale block for K
+          index_t MPerBlock,
+          index_t NPerBlock,
+          index_t KPerBlock,
+          index_t AK1,
+          index_t BK1,
+          index_t MPerWmma,
+          index_t NPerWmma,
+          index_t MRepeat,
+          index_t NRepeat,
+          typename ABlockTransferThreadClusterLengths_AK0_M_AK1,
+          typename ABlockTransferThreadClusterArrangeOrder,
+          typename ABlockTransferSrcAccessOrder,
+          index_t ABlockTransferSrcVectorDim,
+          index_t ABlockTransferSrcScalarPerVector,
+          index_t ABlockTransferDstScalarPerVector_AK1,
+          bool ABlockLdsExtraM,
+          typename BBlockTransferThreadClusterLengths_BK0_N_BK1,
+          typename BBlockTransferThreadClusterArrangeOrder,
+          typename BBlockTransferSrcAccessOrder,
+          index_t BBlockTransferSrcVectorDim,
+          index_t BBlockTransferSrcScalarPerVector,
+          index_t BBlockTransferDstScalarPerVector_BK1,
+          bool BBlockLdsExtraN,
+          index_t CShuffleMRepeatPerShuffle,
+          index_t CShuffleNRepeatPerShuffle,
+          typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
+          typename CShuffleBlockTransferScalarPerVectors,
+          BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave,
+          BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v1,
+          typename ComputeTypeA                       = CDataType,
+          typename ComputeTypeB                       = ComputeTypeA,
+          bool PermuteA                               = false,
+          bool PermuteB                               = false>
+struct DeviceGemmMultiD_ABScale_Wmma_CShuffle_V3
+    : public DeviceGemmMultipleD_ABScaleSplitK<ALayout,
+                                               BLayout,
+                                               DsLayout,
+                                               CLayout,
+                                               ADataType,
+                                               AScaleDataType,
+                                               BDataType,
+                                               BScaleDataType,
+                                               DsDataType,
+                                               CDataType,
+                                               ScaleBlockM,
+                                               ScaleBlockN,
+                                               ScaleBlockK,
+                                               AElementwiseOperation,
+                                               BElementwiseOperation,
+                                               CElementwiseOperation>
+{
+    static constexpr index_t NumDTensor = DsDataType::Size();
+
+    // GridwiseGemm
+    using GridwiseGemm = GridwiseGemm_wmma_cshuffle_v3_ab_scale<
+        ALayout,
+        BLayout,
+        DsLayout,
+        CLayout,
+        Tuple<ADataType>,
+        AScaleDataType,
+        Tuple<BDataType>,
+        BScaleDataType,
+        AccDataType,
+        CShuffleDataType,
+        DsDataType,
+        CDataType,
+        AElementwiseOperation,
+        BElementwiseOperation,
+        CElementwiseOperation,
+        GemmSpec,
+        BlockSize,
+        ScaleBlockM,
+        ScaleBlockN,
+        ScaleBlockK,
+        MPerBlock,
+        NPerBlock,
+        KPerBlock,
+        AK1,
+        BK1,
+        MPerWmma,
+        NPerWmma,
+        MRepeat,
+        NRepeat,
+        ABlockTransferThreadClusterLengths_AK0_M_AK1,
+        ABlockTransferThreadClusterArrangeOrder,
+        ABlockTransferSrcAccessOrder,
+        ABlockTransferSrcVectorDim,
+        ABlockTransferSrcScalarPerVector,
+        ABlockTransferDstScalarPerVector_AK1,
+        false,
+        ABlockLdsExtraM,
+        BBlockTransferThreadClusterLengths_BK0_N_BK1,
+        BBlockTransferThreadClusterArrangeOrder,
+        BBlockTransferSrcAccessOrder,
+        BBlockTransferSrcVectorDim,
+        BBlockTransferSrcScalarPerVector,
+        BBlockTransferDstScalarPerVector_BK1,
+        false,
+        BBlockLdsExtraN,
+        CShuffleMRepeatPerShuffle,
+        CShuffleNRepeatPerShuffle,
+        CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
+        CShuffleBlockTransferScalarPerVectors,
+        BlkGemmPipeSched,
+        BlkGemmPipelineVer,
+        ComputeTypeA,
+        ComputeTypeB,
+        PermuteA,
+        PermuteB>;
+
+    using Argument = typename GridwiseGemm::Argument;
+
+    using DeviceGemmCommon =
+        DeviceGemm_Wmma_CShuffleV3_Common<GridwiseGemm,
+                                          Tuple<ADataType>,
+                                          Tuple<BDataType>,
+                                          DsDataType,
+                                          CDataType,
+                                          MPerBlock,
+                                          NPerBlock,
+                                          KPerBlock,
+                                          BlockSize,
+                                          AK1,
+                                          BK1,
+                                          GemmSpec,
+                                          CShuffleBlockTransferScalarPerVectors,
+                                          BlkGemmPipeSched,
+                                          BlkGemmPipelineVer,
+                                          ComputeTypeA,
+                                          ComputeTypeB>;
+
+    // Invoker
+    using Invoker = typename DeviceGemmCommon::Invoker;
+
+    static bool IsSupportedArgument(const Argument& arg)
+    {
+        // with splitk the implementation doesn't work
+        // when KRead % ScaleBlockK != 0, independently of K padding
+        if(arg.KBatch > 1 && arg.KRead % ScaleBlockK != 0)
+        {
+            return false;
+        }
+
+        return DeviceGemmCommon::IsSupportedArgument(arg);
+    }
+
+    // polymorphic
+    bool IsSupportedArgument(const BaseArgument* p_arg) override
+    {
+        return IsSupportedArgument(*dynamic_cast<const Argument*>(p_arg));
+    }
+
+    void SetKBatch(BaseArgument* base_arg, int KBatch) const override
+    {
+        auto& arg   = *dynamic_cast<Argument*>(base_arg);
+        arg.KBatch  = KBatch;
+        arg.KRead   = GridwiseGemm::CalculateKRead(arg.K, KBatch);
+        arg.KPadded = GridwiseGemm::CalculateKPadded(arg.K, KBatch);
+        arg.AK0     = GridwiseGemm::CalculateAK0Padded(arg.K, KBatch);
+        arg.BK0     = GridwiseGemm::CalculateBK0Padded(arg.K, KBatch);
+    }
+
+    static auto MakeArgument(const ADataType* p_a,
+                             const BDataType* p_b,
+                             std::array<const void*, NumDTensor> p_ds,
+                             CDataType* p_c,
+                             index_t M,
+                             index_t N,
+                             index_t K,
+                             index_t StrideA,
+                             index_t StrideB,
+                             const std::array<index_t, NumDTensor> StrideDs,
+                             index_t StrideC,
+                             const BScaleDataType* p_a_scale,
+                             const BScaleDataType* p_b_scale,
+                             AElementwiseOperation a_element_op,
+                             BElementwiseOperation b_element_op,
+                             CElementwiseOperation cde_element_op,
+                             index_t KBatch = 1)
+    {
+        index_t StrideScaleA = ck::is_same_v<ALayout, tensor_layout::gemm::RowMajor>
+                                   ? math::integer_divide_ceil(K, ScaleBlockK)
+                                   : math::integer_divide_ceil(M, ScaleBlockM);
+
+        index_t StrideScaleB = ck::is_same_v<BLayout, ck::tensor_layout::gemm::ColumnMajor>
+                                   ? math::integer_divide_ceil(K, ScaleBlockK)
+                                   : math::integer_divide_ceil(N, ScaleBlockN);
+
+        return Argument{std::array<const void*, 1>{p_a},
+                        std::array<const void*, 1>{p_b},
+                        p_ds,
+                        p_c,
+                        M,
+                        N,
+                        K,
+                        std::array<index_t, 1>{StrideA},
+                        std::array<index_t, 1>{StrideB},
+                        StrideDs,
+                        StrideC,
+                        StrideScaleA,
+                        StrideScaleB,
+                        p_a_scale,
+                        p_b_scale,
+                        KBatch,
+                        a_element_op,
+                        b_element_op,
+                        cde_element_op};
+    }
+
+    static auto MakeInvoker() { return Invoker{}; }
+
+    // polymorphic
+    std::unique_ptr<BaseArgument>
+    MakeArgumentPointer(const void* p_a,
+                        const void* p_b,
+                        std::array<const void*, NumDTensor> p_ds,
+                        void* p_c,
+                        index_t M,
+                        index_t N,
+                        index_t K,
+                        index_t StrideA,
+                        index_t StrideB,
+                        const std::array<ck::index_t, NumDTensor> StrideDs,
+                        index_t StrideC,
+                        const void* p_a_scale,
+                        const void* p_b_scale,
+                        AElementwiseOperation a_element_op,
+                        BElementwiseOperation b_element_op,
+                        CElementwiseOperation c_element_op,
+                        index_t KBatch = 1) override
+    {
+        index_t StrideScaleA = ck::is_same_v<ALayout, tensor_layout::gemm::RowMajor>
+                                   ? math::integer_divide_ceil(K, ScaleBlockK)
+                                   : math::integer_divide_ceil(M, ScaleBlockM);
+
+        index_t StrideScaleB = ck::is_same_v<BLayout, ck::tensor_layout::gemm::ColumnMajor>
+                                   ? math::integer_divide_ceil(K, ScaleBlockK)
+                                   : math::integer_divide_ceil(N, ScaleBlockN);
+
+        return std::make_unique<Argument>(std::array<const void*, 1>{p_a},
+                                          std::array<const void*, 1>{p_b},
+                                          p_ds,
+                                          static_cast<CDataType*>(p_c),
+                                          M,
+                                          N,
+                                          K,
+                                          std::array<index_t, 1>{StrideA},
+                                          std::array<index_t, 1>{StrideB},
+                                          StrideDs,
+                                          StrideC,
+                                          StrideScaleA,
+                                          StrideScaleB,
+                                          static_cast<const AScaleDataType*>(p_a_scale),
+                                          static_cast<const BScaleDataType*>(p_b_scale),
+                                          KBatch,
+                                          a_element_op,
+                                          b_element_op,
+                                          c_element_op);
+    }
+
+    // polymorphic
+    std::unique_ptr<BaseInvoker> MakeInvokerPointer() override
+    {
+        return std::make_unique<Invoker>(Invoker{});
+    }
+
+    // polymorphic
+    std::string GetTypeString() const override
+    {
+        auto str = std::stringstream();
+
+        std::map<BlockGemmPipelineScheduler, std::string> BlkGemmPipelineSchedulerToString{
+            {BlockGemmPipelineScheduler::Intrawave, "Intrawave"},
+            {BlockGemmPipelineScheduler::Interwave, "Interwave"}};
+
+        std::map<BlockGemmPipelineVersion, std::string> BlkGemmPipelineVersionToString{
+            {BlockGemmPipelineVersion::v1, "v1"},
+            {BlockGemmPipelineVersion::v2, "v2"},
+            {BlockGemmPipelineVersion::v3, "v3"},
+            {BlockGemmPipelineVersion::v4, "v4"},
+            {BlockGemmPipelineVersion::v5, "v5"}};
+
+        // clang-format off
+        str << "DeviceGemm_ABScale_Wmma_CShuffleV3"
+            << "<"
+            << getGemmSpecializationString(GemmSpec) << ", "
+            << std::string(ALayout::name)[0]
+            << std::string(BLayout::name)[0]
+            << std::string(CLayout::name)[0]
+            << ">"
+            << " BlkSize: "
+            << BlockSize << ", "
+            << "BlkTile: "
+            << MPerBlock<<"x"<<NPerBlock<<"x"<<KPerBlock << ", "
+            << "WaveTile: "
+            << MPerWmma<<"x"<<NPerWmma << ", "
+            << "WaveMap: "
+            << MRepeat<<"x" << NRepeat<<", "
+            << "VmemReadVec: "
+            << ABlockTransferSrcScalarPerVector<<"x"<<BBlockTransferSrcScalarPerVector<<", "
+            << "BlkGemmPipelineScheduler: "
+            << BlkGemmPipelineSchedulerToString[BlkGemmPipeSched] << ", "
+            << "BlkGemmPipelineVersion: "
+            << BlkGemmPipelineVersionToString[BlkGemmPipelineVer] << ", "
+            << "BlkGemmPipelinePrefetchStages: "
+            << GridwiseGemm::BlockwiseGemmPipe::PrefetchStages << ", "
+            << "KPack: "
+            << GridwiseGemm::KPack;
+        // clang-format on
+
+        return str.str();
+    }
+};
+
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_wmma_cshuffle_v3_b_preshuffle.hpp

@@ -18,52 +18,6 @@
 #include "ck/host_utility/flush_cache.hpp"
 #include "ck/tensor_operation/gpu/device/impl/device_gemm_wmma_cshuffle_v3_common.hpp"
 
-namespace ck {
-
-template <typename GridwiseGemm,
-          bool HasMainKBlockLoop,
-          InMemoryDataOperationEnum EGlobalMemoryDataOperation,
-          index_t MinimumOccupancy = 1,
-          TailNumber TailNum       = TailNumber::Full>
-__global__ void
-#if CK_USE_LAUNCH_BOUNDS
-__launch_bounds__(CK_MAX_THREAD_PER_BLOCK, MinimumOccupancy)
-#endif
-    kernel_gemm_b_preshuffle_wmma_cshuffle_v3(typename GridwiseGemm::Argument karg)
-{
-#if(defined(__gfx11__) || defined(__gfx12__))
-#if defined(__gfx11__)
-    // gfx11 does not support *_atomic_pk_add_f16/bf16 instructions
-    using e_data_type = remove_cvref_t<remove_pointer_t<decltype(karg.p_e_grid)>>;
-    if constexpr(!(EGlobalMemoryDataOperation == InMemoryDataOperationEnum::AtomicAdd &&
-                   (std::is_same_v<e_data_type, ck::half_t> ||
-                    std::is_same_v<e_data_type, ck::bhalf_t>)))
-    {
-#endif
-        constexpr index_t LDS_size = GridwiseGemm::template GetSharedMemoryNumberOfByte<
-            typename GridwiseGemm::EpilogueCShuffle>();
-        __shared__ char p_shared[LDS_size];
-
-        auto splitk_batch_offset = typename GridwiseGemm::SplitKBatchOffset(karg, blockIdx.z);
-
-        const index_t num_k_per_block = math::integer_divide_ceil(karg.K, GridwiseGemm::KPack);
-        const index_t k_id            = blockIdx.z * num_k_per_block;
-
-        auto epilogue_args = typename GridwiseGemm::EpilogueCShuffle{};
-
-        GridwiseGemm::template Run<HasMainKBlockLoop, EGlobalMemoryDataOperation, TailNum>(
-            p_shared, splitk_batch_offset, karg, epilogue_args, k_id);
-
-#if defined(__gfx11__)
-    }
-#endif
-#else
-    ignore = karg;
-#endif
-}
-
-} // namespace ck
-
 namespace ck {
 namespace tensor_operation {
 namespace device {
@@ -202,270 +156,14 @@ struct DeviceGemmMultiD_Wmma_CShuffle_V3_BPreshuffle
                                           BlkGemmPipeSched,
                                           BlkGemmPipelineVer,
                                           ComputeTypeA,
-                                          ComputeTypeB>;
+                                          ComputeTypeB,
+                                          true>; // IsBPreshuffle
 
     // Invoker
-    struct Invoker : public BaseInvoker
-    {
-        /// @brief  This function issues GPU kernel execution.
-        /// @param arg           The GPU kernel arguments.
-        /// @param stream_config The HIP stream configuration helper structure.
-        /// @return              The kernel's average execution time (if time measurement is
-        ///                      enabled).
-        float Run(const Argument& arg, const StreamConfig& stream_config = StreamConfig{})
-        {
-            if(stream_config.log_level_ > 0)
-            {
-                arg.Print();
-                GridwiseGemm::BlockwiseGemmPipe::HotLoopInstList::Print();
-            }
-
-            if(!GridwiseGemm::CheckValidity(arg))
-            {
-                throw std::runtime_error("wrong! GridwiseGemm has invalid setting");
-            }
-
-            index_t gdx, gdy, gdz;
-            std::tie(gdx, gdy, gdz) = GridwiseGemm::CalculateGridSize(arg.M, arg.N, arg.KBatch);
-
-            float ave_time = 0;
-
-            index_t k_grain = arg.KBatch * KPerBlock;
-            index_t K_split = (arg.K + k_grain - 1) / k_grain * KPerBlock;
-
-            const bool has_main_k_block_loop = GridwiseGemm::CalculateHasMainKBlockLoop(K_split);
-
-            const auto Run = [&](const auto& kernel) {
-                if(stream_config.flush_cache)
-                {
-                    Argument arg_ = arg;
-
-                    const auto a_grid_desc_ak0_m_ak1 = GridwiseGemm::MakeAsGridDescriptor_AK0_M_AK1(
-                        arg_.M, arg_.MPadded, arg_.K, arg_.KPadded, arg_.StrideAs, arg_.AK0);
-                    const auto b_grid_desc_bk0_n_bk1 = GridwiseGemm::MakeBsGridDescriptor_BK0_N_BK1(
-                        arg_.K, arg_.KPadded, arg_.N, arg_.NPadded, arg_.StrideBs, arg_.BK0);
-
-                    std::array<std::size_t, 1> size_as_buffers;
-                    size_as_buffers[Number<0>{}] =
-                        a_grid_desc_ak0_m_ak1[Number<0>{}].GetElementSpaceSize() *
-                        sizeof(ADataType) / GridwiseGemm::APackedSize;
-
-                    std::array<std::size_t, 1> size_bs_buffers;
-                    size_bs_buffers[Number<0>{}] =
-                        b_grid_desc_bk0_n_bk1[Number<0>{}].GetElementSpaceSize() *
-                        sizeof(BDataType) / GridwiseGemm::BPackedSize;
-
-                    const auto ds_grid_desc_m_n = GridwiseGemm::MakeDsGridDescriptor_M_N(
-                        arg_.M, arg_.MPadded, arg_.N, arg_.NPadded, arg_.StrideDs);
-
-                    std::array<std::size_t, GridwiseGemm::NumDTensor> size_ds_buffers;
-                    static_for<0, GridwiseGemm::NumDTensor, 1>{}([&](auto i) {
-                        using DDataType = remove_cvref_t<tuple_element_t<i.value, DsDataType>>;
-                        size_ds_buffers[i] =
-                            ds_grid_desc_m_n[i].GetElementSpaceSize() * sizeof(DDataType);
-                    });
-
-                    ck::utility::RotatingMemWrapperMultiABD<Argument,
-                                                            Tuple<ADataType>,
-                                                            Tuple<BDataType>,
-                                                            DsDataType>
-                        rotating_mem(arg_,
-                                     stream_config.rotating_count,
-                                     size_as_buffers,
-                                     size_bs_buffers,
-                                     size_ds_buffers);
-                    rotating_mem.Print();
-
-                    auto run_flush_cache = [&]() {
-                        // flush icache
-                        ck::utility::flush_icache();
-                        // rotating mem
-                        rotating_mem.Next();
-                        // clear c mem
-                        if(arg_.KBatch > 1)
-                            HIP_CHECK_ERROR(hipMemsetAsync(arg_.p_e_grid,
-                                                           0,
-                                                           arg_.M * arg_.N * sizeof(EDataType),
-                                                           stream_config.stream_id_));
-                    };
-
-                    ave_time = ck::utility::launch_and_time_kernel_with_preprocess<false>(
-                        stream_config,
-                        run_flush_cache,
-                        kernel,
-                        dim3(gdx, gdy, gdz),
-                        dim3(BlockSize),
-                        0,
-                        arg_);
-                }
-                else
-                {
-                    if(arg.KBatch > 1)
-                        HIP_CHECK_ERROR(hipMemsetAsync(arg.p_e_grid,
-                                                       0,
-                                                       arg.M * arg.N * sizeof(EDataType),
-                                                       stream_config.stream_id_));
-
-                    ave_time = launch_and_time_kernel(
-                        stream_config, kernel, dim3(gdx, gdy, gdz), dim3(BlockSize), 0, arg);
-                }
-            };
-
-            constexpr index_t minimum_occupancy = []() {
-                if constexpr(BlkGemmPipeSched == BlockGemmPipelineScheduler::Interwave)
-                {
-                    return 2;
-                }
-                else if constexpr(BlkGemmPipelineVer == BlockGemmPipelineVersion::v3)
-                {
-                    return (MPerBlock * NPerBlock / BlockSize <= 128) ? 2 : 1;
-                }
-                else
-                {
-                    return 1;
-                }
-            }();
-
-            // ThreadwiseTensorSliceTransfer_v7r3 (used in ThreadGroupTensorSliceTransfer_v7r3) is
-            // currently implemented in such a way that all SrcScalarPerVectors must be the same, so
-            // if one of D matrices is column-major, then all SrcScalarPerVectors must be 1. On the
-            // other hand, Split K for 16-bit outputs uses packed atomics so ScalarPerVectors cannot
-            // be odd.
-            constexpr bool AtomicsImplementationExists =
-                !(std::is_same_v<EDataType, ck::half_t> || std::is_same_v<EDataType, ck::bhalf_t> ||
-                  std::is_same_v<EDataType, int8_t>) ||
-                (CDEShuffleBlockTransferScalarPerVectors{}[0] % 2 == 0);
-
-            if(has_main_k_block_loop)
-            {
-                // Tail number always full
-                if constexpr(BlkGemmPipelineVer == BlockGemmPipelineVersion::v1)
-                {
-                    if(arg.KBatch > 1)
-                    {
-                        if constexpr(AtomicsImplementationExists)
-                        {
-                            if(GridwiseGemm::CalculateKBlockLoopTailNum(K_split) == TailNumber::Odd)
-                            {
-                                const auto kernel = kernel_gemm_b_preshuffle_wmma_cshuffle_v3<
-                                    GridwiseGemm,
-                                    true,
-                                    InMemoryDataOperationEnum::AtomicAdd,
-                                    minimum_occupancy,
-                                    TailNumber::Odd>;
-                                Run(kernel);
-                            }
-                            else
-                            {
-                                const auto kernel = kernel_gemm_b_preshuffle_wmma_cshuffle_v3<
-                                    GridwiseGemm,
-                                    true,
-                                    InMemoryDataOperationEnum::AtomicAdd,
-                                    minimum_occupancy,
-                                    TailNumber::Even>;
-                                Run(kernel);
-                            }
-                        }
-                    }
-                    else
-                    {
-                        if(GridwiseGemm::CalculateKBlockLoopTailNum(K_split) == TailNumber::Odd)
-                        {
-                            const auto kernel = kernel_gemm_b_preshuffle_wmma_cshuffle_v3<
-                                GridwiseGemm,
-                                true,
-                                InMemoryDataOperationEnum::Set,
-                                minimum_occupancy,
-                                TailNumber::Odd>;
-                            Run(kernel);
-                        }
-                        else
-                        {
-                            const auto kernel = kernel_gemm_b_preshuffle_wmma_cshuffle_v3<
-                                GridwiseGemm,
-                                true,
-                                InMemoryDataOperationEnum::Set,
-                                minimum_occupancy,
-                                TailNumber::Even>;
-                            Run(kernel);
-                        }
-                    }
-                }
-            }
-            else
-            {
-                // Tail number always 1
-                if constexpr(BlkGemmPipelineVer == BlockGemmPipelineVersion::v1)
-                {
-                    if(arg.KBatch > 1)
-                    {
-                        if constexpr(AtomicsImplementationExists)
-                        {
-                            if(GridwiseGemm::CalculateKBlockLoopTailNum(K_split) == TailNumber::Odd)
-                            {
-                                const auto kernel = kernel_gemm_b_preshuffle_wmma_cshuffle_v3<
-                                    GridwiseGemm,
-                                    false,
-                                    InMemoryDataOperationEnum::AtomicAdd,
-                                    minimum_occupancy,
-                                    TailNumber::Odd>;
-                                Run(kernel);
-                            }
-                            else
-                            {
-                                const auto kernel = kernel_gemm_b_preshuffle_wmma_cshuffle_v3<
-                                    GridwiseGemm,
-                                    false,
-                                    InMemoryDataOperationEnum::AtomicAdd,
-                                    minimum_occupancy,
-                                    TailNumber::Even>;
-                                Run(kernel);
-                            }
-                        }
-                    }
-                    else
-                    {
-                        if(GridwiseGemm::CalculateKBlockLoopTailNum(K_split) == TailNumber::Odd)
-                        {
-                            const auto kernel = kernel_gemm_b_preshuffle_wmma_cshuffle_v3<
-                                GridwiseGemm,
-                                false,
-                                InMemoryDataOperationEnum::Set,
-                                minimum_occupancy,
-                                TailNumber::Odd>;
-                            Run(kernel);
-                        }
-                        else
-                        {
-                            const auto kernel = kernel_gemm_b_preshuffle_wmma_cshuffle_v3<
-                                GridwiseGemm,
-                                false,
-                                InMemoryDataOperationEnum::Set,
-                                minimum_occupancy,
-                                TailNumber::Even>;
-                            Run(kernel);
-                        }
-                    }
-                }
-            }
-
-            return ave_time;
-        }
-
-        // polymorphic
-        float Run(const BaseArgument* p_arg,
-                  const StreamConfig& stream_config = StreamConfig{}) override
-        {
-            return Run(*dynamic_cast<const Argument*>(p_arg), stream_config);
-        }
-    };
+    using Invoker = typename DeviceGemmCommon::Invoker;
 
     static bool IsSupportedArgument(const Argument& arg)
     {
-        if(arg.N % NPerBlock != 0 || arg.K % KPerBlock != 0)
-        {
-            return false;
-        }
         return DeviceGemmCommon::IsSupportedArgument(arg);
     }
 

include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_wmma_cshuffle_v3_blockscale_bpreshuffle.hpp

@@ -0,0 +1,360 @@
+// Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
+// SPDX-License-Identifier: MIT
+
+#pragma once
+
+#include <iostream>
+#include <sstream>
+
+#include "ck/utility/common_header.hpp"
+#include "ck/tensor_description/tensor_descriptor.hpp"
+#include "ck/tensor_description/tensor_descriptor_helper.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_gemm_multiple_d_ab_scale.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/grid/gridwise_gemm_wmma_cshuffle_v3_ab_scale.hpp"
+#include "ck/host_utility/device_prop.hpp"
+#include "ck/host_utility/kernel_launch.hpp"
+#include "ck/host_utility/flush_cache.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_wmma_cshuffle_v3_common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+
+template <typename ALayout,
+          typename BLayout,
+          typename DsLayout,
+          typename CLayout,
+          typename ADataType,
+          typename AScaleDataType,
+          typename BDataType,
+          typename BScaleDataType,
+          typename DsDataType,
+          typename CDataType,
+          typename AccDataType,
+          typename CShuffleDataType,
+          typename AElementwiseOperation,
+          typename BElementwiseOperation,
+          typename CElementwiseOperation,
+          GemmSpecialization GemmSpec,
+          index_t BlockSize,
+          index_t ScaleBlockM, // scale block for M
+          index_t ScaleBlockN, // scale block for N
+          index_t ScaleBlockK, // scale block for K
+          index_t MPerBlock,
+          index_t NPerBlock,
+          index_t KPerBlock,
+          index_t AK1,
+          index_t BK1,
+          index_t MPerWmma,
+          index_t NPerWmma,
+          index_t MRepeat,
+          index_t NRepeat,
+          typename ABlockTransferThreadClusterLengths_AK0_M_AK1,
+          typename ABlockTransferThreadClusterArrangeOrder,
+          typename ABlockTransferSrcAccessOrder,
+          index_t ABlockTransferSrcVectorDim,
+          index_t ABlockTransferSrcScalarPerVector,
+          index_t ABlockTransferDstScalarPerVector_AK1,
+          bool ABlockLdsExtraM,
+          typename BBlockTransferThreadClusterLengths_BK0_N_BK1,
+          typename BBlockTransferThreadClusterArrangeOrder,
+          typename BBlockTransferSrcAccessOrder,
+          index_t BBlockTransferSrcVectorDim,
+          index_t BBlockTransferSrcScalarPerVector,
+          index_t BBlockTransferDstScalarPerVector_BK1,
+          bool BBlockLdsExtraN,
+          index_t CShuffleMRepeatPerShuffle,
+          index_t CShuffleNRepeatPerShuffle,
+          typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
+          typename CShuffleBlockTransferScalarPerVectors,
+          BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave,
+          BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v1,
+          typename ComputeTypeA                       = CDataType,
+          typename ComputeTypeB                       = ComputeTypeA,
+          bool PermuteA                               = false,
+          bool PermuteB                               = false>
+struct DeviceGemmMultiD_BlockScale_Wmma_CShuffle_V3_BPreshuffle
+    : public DeviceGemmMultipleD_BlockScale_BPreshuffleSplitK<ALayout,
+                                                              BLayout,
+                                                              DsLayout,
+                                                              CLayout,
+                                                              ADataType,
+                                                              AScaleDataType,
+                                                              BDataType,
+                                                              BScaleDataType,
+                                                              DsDataType,
+                                                              CDataType,
+                                                              ScaleBlockM,
+                                                              ScaleBlockN,
+                                                              ScaleBlockK,
+                                                              AElementwiseOperation,
+                                                              BElementwiseOperation,
+                                                              CElementwiseOperation>
+{
+    static constexpr index_t NumDTensor = DsDataType::Size();
+
+    using AScaleLayout = tensor_layout::gemm::ColumnMajor;
+    using BScaleLayout = BLayout;
+
+    // GridwiseGemm
+    using GridwiseGemm = GridwiseGemm_wmma_cshuffle_v3_ab_scale<
+        ALayout,
+        BLayout,
+        DsLayout,
+        CLayout,
+        Tuple<ADataType>,
+        AScaleDataType,
+        Tuple<BDataType>,
+        BScaleDataType,
+        AccDataType,
+        CShuffleDataType,
+        DsDataType,
+        CDataType,
+        AElementwiseOperation,
+        BElementwiseOperation,
+        CElementwiseOperation,
+        GemmSpec,
+        BlockSize,
+        ScaleBlockM,
+        ScaleBlockN,
+        ScaleBlockK,
+        MPerBlock,
+        NPerBlock,
+        KPerBlock,
+        AK1,
+        BK1,
+        MPerWmma,
+        NPerWmma,
+        MRepeat,
+        NRepeat,
+        ABlockTransferThreadClusterLengths_AK0_M_AK1,
+        ABlockTransferThreadClusterArrangeOrder,
+        ABlockTransferSrcAccessOrder,
+        ABlockTransferSrcVectorDim,
+        ABlockTransferSrcScalarPerVector,
+        ABlockTransferDstScalarPerVector_AK1,
+        false,
+        ABlockLdsExtraM,
+        BBlockTransferThreadClusterLengths_BK0_N_BK1,
+        BBlockTransferThreadClusterArrangeOrder,
+        BBlockTransferSrcAccessOrder,
+        BBlockTransferSrcVectorDim,
+        BBlockTransferSrcScalarPerVector,
+        BBlockTransferDstScalarPerVector_BK1,
+        false,
+        BBlockLdsExtraN,
+        CShuffleMRepeatPerShuffle,
+        CShuffleNRepeatPerShuffle,
+        CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
+        CShuffleBlockTransferScalarPerVectors,
+        BlkGemmPipeSched,
+        BlkGemmPipelineVer,
+        ComputeTypeA,
+        ComputeTypeB,
+        PermuteA,
+        PermuteB,
+        true,
+        AScaleLayout,
+        BScaleLayout>;
+
+    using Argument = typename GridwiseGemm::Argument;
+    int GetPreShuffleParameters() override { return NPerWmma; }
+
+    using DeviceGemmCommon =
+        DeviceGemm_Wmma_CShuffleV3_Common<GridwiseGemm,
+                                          Tuple<ADataType>,
+                                          Tuple<BDataType>,
+                                          DsDataType,
+                                          CDataType,
+                                          MPerBlock,
+                                          NPerBlock,
+                                          KPerBlock,
+                                          BlockSize,
+                                          AK1,
+                                          BK1,
+                                          GemmSpec,
+                                          CShuffleBlockTransferScalarPerVectors,
+                                          BlkGemmPipeSched,
+                                          BlkGemmPipelineVer,
+                                          ComputeTypeA,
+                                          ComputeTypeB,
+                                          true>; // IsBPreshuffle
+
+    // Invoker
+    using Invoker = typename DeviceGemmCommon::Invoker;
+
+    static bool IsSupportedArgument(const Argument& arg)
+    {
+        // with splitk the implementation doesn't work
+        // when KRead % ScaleBlockK != 0, independently of K padding
+        if(arg.KBatch > 1 && arg.KRead % ScaleBlockK != 0)
+        {
+            return false;
+        }
+
+        return DeviceGemmCommon::IsSupportedArgument(arg);
+    }
+
+    // polymorphic
+    bool IsSupportedArgument(const BaseArgument* p_arg) override
+    {
+        return IsSupportedArgument(*dynamic_cast<const Argument*>(p_arg));
+    }
+
+    static auto MakeArgument(const void* p_a,
+                             const void* p_b,
+                             std::array<const void*, NumDTensor> p_ds,
+                             void* p_e,
+                             index_t M,
+                             index_t N,
+                             index_t K,
+                             index_t StrideA,
+                             index_t StrideB,
+                             const std::array<index_t, NumDTensor> StrideDs,
+                             index_t StrideC,
+                             const void* p_a_scale,
+                             const void* p_b_scale,
+                             AElementwiseOperation a_element_op,
+                             BElementwiseOperation b_element_op,
+                             CElementwiseOperation cde_element_op,
+                             index_t KBatch)
+    {
+        index_t StrideScaleA = ck::is_same_v<AScaleLayout, tensor_layout::gemm::RowMajor>
+                                   ? math::integer_divide_ceil(K, ScaleBlockK)
+                                   : math::integer_divide_ceil(M, ScaleBlockM);
+
+        index_t StrideScaleB = ck::is_same_v<BScaleLayout, ck::tensor_layout::gemm::ColumnMajor>
+                                   ? math::integer_divide_ceil(K, ScaleBlockK)
+                                   : math::integer_divide_ceil(N, ScaleBlockN);
+
+        return Argument{std::array<const void*, 1>{p_a},
+                        std::array<const void*, 1>{p_b},
+                        p_ds,
+                        static_cast<CDataType*>(p_e),
+                        M,
+                        N,
+                        K,
+                        std::array<index_t, 1>{StrideA},
+                        std::array<index_t, 1>{StrideB},
+                        StrideDs,
+                        StrideC,
+                        StrideScaleA,
+                        StrideScaleB,
+                        static_cast<const AScaleDataType*>(p_a_scale),
+                        static_cast<const BScaleDataType*>(p_b_scale),
+                        KBatch,
+                        a_element_op,
+                        b_element_op,
+                        cde_element_op};
+    }
+
+    static auto MakeInvoker() { return Invoker{}; }
+
+    // polymorphic
+    std::unique_ptr<BaseArgument>
+    MakeArgumentPointer(const void* p_a,
+                        const void* p_b,
+                        std::array<const void*, NumDTensor> p_ds,
+                        void* p_e,
+                        index_t M,
+                        index_t N,
+                        index_t K,
+                        index_t StrideA,
+                        index_t StrideB,
+                        const std::array<ck::index_t, NumDTensor> StrideDs,
+                        index_t StrideC,
+                        const void* p_a_scale,
+                        const void* p_b_scale,
+                        AElementwiseOperation a_element_op,
+                        BElementwiseOperation b_element_op,
+                        CElementwiseOperation c_element_op,
+                        index_t KBatch) override
+    {
+        index_t StrideScaleA = ck::is_same_v<AScaleLayout, tensor_layout::gemm::RowMajor>
+                                   ? math::integer_divide_ceil(K, ScaleBlockK)
+                                   : math::integer_divide_ceil(M, ScaleBlockM);
+
+        index_t StrideScaleB = ck::is_same_v<BScaleLayout, ck::tensor_layout::gemm::ColumnMajor>
+                                   ? math::integer_divide_ceil(K, ScaleBlockK)
+                                   : math::integer_divide_ceil(N, ScaleBlockN);
+
+        return std::make_unique<Argument>(std::array<const void*, 1>{p_a},
+                                          std::array<const void*, 1>{p_b},
+                                          p_ds,
+                                          static_cast<CDataType*>(p_e),
+                                          M,
+                                          N,
+                                          K,
+                                          std::array<index_t, 1>{StrideA},
+                                          std::array<index_t, 1>{StrideB},
+                                          StrideDs,
+                                          StrideC,
+                                          StrideScaleA,
+                                          StrideScaleB,
+                                          static_cast<const AScaleDataType*>(p_a_scale),
+                                          static_cast<const BScaleDataType*>(p_b_scale),
+                                          KBatch,
+                                          a_element_op,
+                                          b_element_op,
+                                          c_element_op);
+    }
+
+    // polymorphic
+    std::unique_ptr<BaseInvoker> MakeInvokerPointer() override
+    {
+        return std::make_unique<Invoker>(Invoker{});
+    }
+
+    // polymorphic
+    std::string GetTypeString() const override
+    {
+        auto str = std::stringstream();
+
+        std::map<BlockGemmPipelineScheduler, std::string> BlkGemmPipelineSchedulerToString{
+            {BlockGemmPipelineScheduler::Intrawave, "Intrawave"},
+            {BlockGemmPipelineScheduler::Interwave, "Interwave"}};
+
+        std::map<BlockGemmPipelineVersion, std::string> BlkGemmPipelineVersionToString{
+            {BlockGemmPipelineVersion::v1, "v1"},
+            {BlockGemmPipelineVersion::v2, "v2"},
+            {BlockGemmPipelineVersion::v3, "v3"},
+            {BlockGemmPipelineVersion::v4, "v4"},
+            {BlockGemmPipelineVersion::v5, "v5"}};
+
+        // clang-format off
+        str << "DeviceGemmMultiD_BlockScale_Wmma_CShuffle_V3_BPreshuffle"
+            << "<"
+            << getGemmSpecializationString(GemmSpec) << ", "
+            << std::string(ALayout::name)[0]
+            << std::string(BLayout::name)[0]
+            << std::string(CLayout::name)[0]
+            << ">"
+            << " BlkSize: "
+            << BlockSize << ", "
+            << "BlkTile: "
+            << MPerBlock<<"x"<<NPerBlock<<"x"<<KPerBlock << ", "
+            << "WaveTile: "
+            << MPerWmma<<"x"<<NPerWmma << ", "
+            << "WaveMap: "
+            << MRepeat<<"x" << NRepeat<<", "
+            << "VmemReadVec: "
+            << ABlockTransferSrcScalarPerVector<<"x"<<BBlockTransferSrcScalarPerVector<<", "
+            << "BlkGemmPipelineScheduler: "
+            << BlkGemmPipelineSchedulerToString[BlkGemmPipeSched] << ", "
+            << "BlkGemmPipelineVersion: "
+            << BlkGemmPipelineVersionToString[BlkGemmPipelineVer] << ", "
+            << "BlkGemmPipelinePrefetchStages: "
+            << GridwiseGemm::BlockwiseGemmPipe::PrefetchStages << ", "
+            << "KPack: "
+            << GridwiseGemm::KPack;
+        // clang-format on
+
+        return str.str();
+    }
+};
+
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_xdl_cshuffle_v3_ab_scale.hpp

@@ -262,6 +262,16 @@ struct DeviceGemmMultiD_ABScale_Xdl_CShuffle_V3
                 if constexpr(BlkGemmPipelineVer == BlockGemmPipelineVersion::v1 ||
                              BlkGemmPipelineVer == BlockGemmPipelineVersion::v3)
                 {
+                    if(arg.KBatch > 1)
+                    {
+                        const auto kernel =
+                            kernel_gemm_xdl_cshuffle_v3<GridwiseGemm,
+                                                        true,
+                                                        InMemoryDataOperationEnum::AtomicAdd,
+                                                        minimum_occupancy>;
+                        Run(kernel);
+                    }
+                    else
                     {
                         const auto kernel =
                             kernel_gemm_xdl_cshuffle_v3<GridwiseGemm,
@@ -279,22 +289,47 @@ struct DeviceGemmMultiD_ABScale_Xdl_CShuffle_V3
                 {
                     if(GridwiseGemm::CalculateKBlockLoopTailNum(K_split) == TailNumber::Full)
                     {
-                        const auto kernel =
-                            kernel_gemm_xdl_cshuffle_v3<GridwiseGemm,
-                                                        false,
-                                                        InMemoryDataOperationEnum::Set,
-                                                        minimum_occupancy>;
-                        Run(kernel);
+                        if(arg.KBatch > 1)
+                        {
+                            const auto kernel =
+                                kernel_gemm_xdl_cshuffle_v3<GridwiseGemm,
+                                                            false,
+                                                            InMemoryDataOperationEnum::AtomicAdd,
+                                                            minimum_occupancy>;
+                            Run(kernel);
+                        }
+                        else
+                        {
+                            const auto kernel =
+                                kernel_gemm_xdl_cshuffle_v3<GridwiseGemm,
+                                                            false,
+                                                            InMemoryDataOperationEnum::Set,
+                                                            minimum_occupancy>;
+                            Run(kernel);
+                        }
                     }
                     else if(GridwiseGemm::CalculateKBlockLoopTailNum(K_split) == TailNumber::Odd)
                     {
-                        const auto kernel =
-                            kernel_gemm_xdl_cshuffle_v3<GridwiseGemm,
-                                                        false,
-                                                        InMemoryDataOperationEnum::Set,
-                                                        minimum_occupancy,
-                                                        TailNumber::Odd>;
-                        Run(kernel);
+                        if(arg.KBatch > 1)
+                        {
+                            const auto kernel =
+                                kernel_gemm_xdl_cshuffle_v3<GridwiseGemm,
+                                                            false,
+                                                            InMemoryDataOperationEnum::AtomicAdd,
+                                                            minimum_occupancy,
+                                                            TailNumber::Odd>;
+                            Run(kernel);
+                        }
+                        else
+                        {
+                            const auto kernel =
+                                kernel_gemm_xdl_cshuffle_v3<GridwiseGemm,
+                                                            false,
+                                                            InMemoryDataOperationEnum::Set,
+                                                            minimum_occupancy,
+                                                            TailNumber::Odd>;
+                            Run(kernel);
+                        }
                     }
                 }
             }
@@ -315,6 +350,20 @@ struct DeviceGemmMultiD_ABScale_Xdl_CShuffle_V3
     {
         auto& arg  = *dynamic_cast<Argument*>(base_arg);
         arg.KBatch = KBatch;
+        if(get_warp_size() == 64)
+        {
+            arg.KRead   = GridwiseGemm64::CalculateKRead(arg.K, KBatch);
+            arg.KPadded = GridwiseGemm64::CalculateKPadded(arg.K, KBatch);
+            arg.AK0     = GridwiseGemm64::CalculateAK0Padded(arg.K, KBatch);
+            arg.BK0     = GridwiseGemm64::CalculateBK0Padded(arg.K, KBatch);
+        }
+        else
+        {
+            arg.KRead   = GridwiseGemm32::CalculateKRead(arg.K, KBatch);
+            arg.KPadded = GridwiseGemm32::CalculateKPadded(arg.K, KBatch);
+            arg.AK0     = GridwiseGemm32::CalculateAK0Padded(arg.K, KBatch);
+            arg.BK0     = GridwiseGemm32::CalculateBK0Padded(arg.K, KBatch);
+        }
     }
 
     static constexpr bool IsValidCompilationParameter()
@@ -325,6 +374,13 @@ struct DeviceGemmMultiD_ABScale_Xdl_CShuffle_V3
 
     static bool IsSupportedArgument(const Argument& arg)
     {
+        // with splitk the implementation doesn't work
+        // when KRead % ScaleBlockK != 0, independently of K padding
+        if(arg.KBatch > 1 && arg.KRead % ScaleBlockK != 0)
+        {
+            return false;
+        }
+
         if(!ck::is_xdl_wmma_supported<ComputeTypeA, ComputeTypeB, MPerXDL, NPerXDL>())
         {
             return false;
@@ -385,6 +441,14 @@ struct DeviceGemmMultiD_ABScale_Xdl_CShuffle_V3
                              BElementwiseOperation b_element_op,
                              CElementwiseOperation c_element_op)
     {
+        index_t StrideScaleA = ck::is_same_v<ALayout, tensor_layout::gemm::RowMajor>
+                                   ? math::integer_divide_ceil(K, ScaleBlockK)
+                                   : math::integer_divide_ceil(M, ScaleBlockM);
+
+        index_t StrideScaleB = ck::is_same_v<BLayout, ck::tensor_layout::gemm::ColumnMajor>
+                                   ? math::integer_divide_ceil(K, ScaleBlockK)
+                                   : math::integer_divide_ceil(N, ScaleBlockN);
+
         return Argument{static_cast<const ADataType*>(p_a),
                         static_cast<const BDataType*>(p_b),
                         p_ds,
@@ -396,6 +460,8 @@ struct DeviceGemmMultiD_ABScale_Xdl_CShuffle_V3
                         StrideB,
                         StrideDs,
                         StrideC,
+                        StrideScaleA,
+                        StrideScaleB,
                         static_cast<const AScaleDataType*>(p_a_scale),
                         static_cast<const BScaleDataType*>(p_b_scale),
                         1,
@@ -425,6 +491,14 @@ struct DeviceGemmMultiD_ABScale_Xdl_CShuffle_V3
                         BElementwiseOperation b_element_op,
                         CElementwiseOperation c_element_op) override
     {
+        index_t StrideScaleA = ck::is_same_v<ALayout, tensor_layout::gemm::RowMajor>
+                                   ? math::integer_divide_ceil(K, ScaleBlockK)
+                                   : math::integer_divide_ceil(M, ScaleBlockM);
+
+        index_t StrideScaleB = ck::is_same_v<BLayout, ck::tensor_layout::gemm::ColumnMajor>
+                                   ? math::integer_divide_ceil(K, ScaleBlockK)
+                                   : math::integer_divide_ceil(N, ScaleBlockN);
+
         return std::make_unique<Argument>(static_cast<const ADataType*>(p_a),
                                           static_cast<const BDataType*>(p_b),
                                           p_ds,
@@ -436,6 +510,8 @@ struct DeviceGemmMultiD_ABScale_Xdl_CShuffle_V3
                                           StrideB,
                                           StrideDs,
                                           StrideC,
+                                          StrideScaleA,
+                                          StrideScaleB,
                                           static_cast<const AScaleDataType*>(p_a_scale),
                                           static_cast<const BScaleDataType*>(p_b_scale),
                                           1,

include/ck/tensor_operation/gpu/device/impl/device_gemm_wmma_cshuffle_v3_b_scale.hpp

@@ -12,7 +12,7 @@
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
 #include "ck/tensor_operation/gpu/device/device_gemm_v2.hpp"
 #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/grid/gridwise_gemm_wmma_cshuffle_v3_b_scale.hpp"
+#include "ck/tensor_operation/gpu/grid/gridwise_gemm_wmma_cshuffle_v3_ab_scale.hpp"
 #include "ck/host_utility/device_prop.hpp"
 #include "ck/host_utility/kernel_launch.hpp"
 #include "ck/host_utility/flush_cache.hpp"
@@ -86,12 +86,13 @@ struct DeviceGemm_BScale_Wmma_CShuffleV3 : public DeviceGemmV2BScale<ALayout,
 {
 
     // GridwiseGemm
-    using GridwiseGemm = GridwiseGemm_wmma_cshuffle_v3_b_scale<
+    using GridwiseGemm = GridwiseGemm_wmma_cshuffle_v3_ab_scale<
         ALayout,
         BLayout,
         Tuple<>, // DsLayout
         CLayout,
         Tuple<ADataType>,
+        void, // AScaleType
         Tuple<BDataType>,
         BScaleDataType,
         AccDataType,
@@ -103,6 +104,7 @@ struct DeviceGemm_BScale_Wmma_CShuffleV3 : public DeviceGemmV2BScale<ALayout,
         CElementwiseOperation,
         GemmSpec,
         BlockSize,
+        0, // ScaleBlockM
         ScaleBlockN,
         ScaleBlockK,
         MPerBlock,
@@ -207,7 +209,9 @@ struct DeviceGemm_BScale_Wmma_CShuffleV3 : public DeviceGemmV2BScale<ALayout,
                         std::array<index_t, 1>{StrideB},
                         std::array<index_t, 0>{}, // StrideDs_
                         StrideC,
+                        0, // StrideScaleA
                         StrideScaleB,
+                        nullptr,
                         p_b_scale,
                         KBatch,
                         a_element_op,
@@ -245,7 +249,9 @@ struct DeviceGemm_BScale_Wmma_CShuffleV3 : public DeviceGemmV2BScale<ALayout,
                                           std::array<index_t, 1>{StrideB},
                                           std::array<index_t, 0>{}, // StrideDs_
                                           StrideC,
+                                          0, // StrideScaleA
                                           StrideScaleB,
+                                          nullptr, // p_a_scale
                                           static_cast<const BScaleDataType*>(p_b_scale),
                                           KBatch,
                                           a_element_op,

include/ck/tensor_operation/gpu/device/impl/device_gemm_wmma_cshuffle_v3_common.hpp

@@ -38,7 +38,8 @@ template <typename GridwiseGemm,
           BlockGemmPipelineScheduler BlkGemmPipeSched,
           BlockGemmPipelineVersion BlkGemmPipelineVer,
           typename ComputeTypeA,
-          typename ComputeTypeB>
+          typename ComputeTypeB,
+          bool IsBPreShuffled = false>
 struct DeviceGemm_Wmma_CShuffleV3_Common
 {
 
@@ -189,61 +190,174 @@ struct DeviceGemm_Wmma_CShuffleV3_Common
             if(has_main_k_block_loop)
             {
                 // Tail number always full
-                if constexpr(BlkGemmPipelineVer == BlockGemmPipelineVersion::v1 ||
-                             BlkGemmPipelineVer == BlockGemmPipelineVersion::v3)
+                if constexpr(IsBPreShuffled)
                 {
-                    if(arg.KBatch > 1)
+                    if constexpr(BlkGemmPipelineVer == BlockGemmPipelineVersion::v1)
                     {
-                        if constexpr(AtomicsImplementationExists)
+                        if(arg.KBatch > 1)
                         {
-                            const auto kernel =
-                                kernel_gemm_wmma_cshuffle_v3<GridwiseGemm,
-                                                             true,
-                                                             InMemoryDataOperationEnum::AtomicAdd,
-                                                             minimum_occupancy>;
-                            Run(kernel);
+                            if constexpr(AtomicsImplementationExists)
+                            {
+                                if(GridwiseGemm::CalculateKBlockLoopTailNum(K_split) ==
+                                   TailNumber::Odd)
+                                {
+                                    const auto kernel = kernel_gemm_b_preshuffle_wmma_cshuffle_v3<
+                                        GridwiseGemm,
+                                        true,
+                                        InMemoryDataOperationEnum::AtomicAdd,
+                                        minimum_occupancy,
+                                        TailNumber::Odd>;
+                                    Run(kernel);
+                                }
+                                else
+                                {
+                                    const auto kernel = kernel_gemm_b_preshuffle_wmma_cshuffle_v3<
+                                        GridwiseGemm,
+                                        true,
+                                        InMemoryDataOperationEnum::AtomicAdd,
+                                        minimum_occupancy,
+                                        TailNumber::Even>;
+                                    Run(kernel);
+                                }
+                            }
+                        }
+                        else
+                        {
+                            if(GridwiseGemm::CalculateKBlockLoopTailNum(K_split) == TailNumber::Odd)
+                            {
+                                const auto kernel = kernel_gemm_b_preshuffle_wmma_cshuffle_v3<
+                                    GridwiseGemm,
+                                    true,
+                                    InMemoryDataOperationEnum::Set,
+                                    minimum_occupancy,
+                                    TailNumber::Odd>;
+                                Run(kernel);
+                            }
+                            else
+                            {
+                                const auto kernel = kernel_gemm_b_preshuffle_wmma_cshuffle_v3<
+                                    GridwiseGemm,
+                                    true,
+                                    InMemoryDataOperationEnum::Set,
+                                    minimum_occupancy,
+                                    TailNumber::Even>;
+                                Run(kernel);
+                            }
                         }
-                    }
-                    else
-                    {
-                        const auto kernel =
-                            kernel_gemm_wmma_cshuffle_v3<GridwiseGemm,
-                                                         true,
-                                                         InMemoryDataOperationEnum::Set,
-                                                         minimum_occupancy>;
-                        Run(kernel);
                     }
                 }
                 else
                 {
-                    // TODO: Implement
-                }
-            }
-            else
-            {
-                // Tail number always 1
-                if constexpr(BlkGemmPipelineVer == BlockGemmPipelineVersion::v1)
-                {
-                    if(arg.KBatch > 1)
+                    if constexpr(BlkGemmPipelineVer == BlockGemmPipelineVersion::v1 ||
+                                 BlkGemmPipelineVer == BlockGemmPipelineVersion::v3)
                     {
-                        if constexpr(AtomicsImplementationExists)
+                        if(arg.KBatch > 1)
+                        {
+                            if constexpr(AtomicsImplementationExists)
+                            {
+                                const auto kernel = kernel_gemm_wmma_cshuffle_v3<
+                                    GridwiseGemm,
+                                    true,
+                                    InMemoryDataOperationEnum::AtomicAdd,
+                                    minimum_occupancy>;
+                                Run(kernel);
+                            }
+                        }
+                        else
                         {
                             const auto kernel =
                                 kernel_gemm_wmma_cshuffle_v3<GridwiseGemm,
-                                                             false,
-                                                             InMemoryDataOperationEnum::AtomicAdd,
+                                                             true,
+                                                             InMemoryDataOperationEnum::Set,
                                                              minimum_occupancy>;
                             Run(kernel);
                         }
                     }
-                    else
+                }
+            }
+            else
+            {
+                if constexpr(IsBPreShuffled)
+                {
+                    if constexpr(BlkGemmPipelineVer == BlockGemmPipelineVersion::v1)
                     {
-                        const auto kernel =
-                            kernel_gemm_wmma_cshuffle_v3<GridwiseGemm,
-                                                         false,
-                                                         InMemoryDataOperationEnum::Set,
-                                                         minimum_occupancy>;
-                        Run(kernel);
+                        if(arg.KBatch > 1)
+                        {
+                            if constexpr(AtomicsImplementationExists)
+                            {
+                                if(GridwiseGemm::CalculateKBlockLoopTailNum(K_split) ==
+                                   TailNumber::Odd)
+                                {
+                                    const auto kernel = kernel_gemm_b_preshuffle_wmma_cshuffle_v3<
+                                        GridwiseGemm,
+                                        false,
+                                        InMemoryDataOperationEnum::AtomicAdd,
+                                        minimum_occupancy,
+                                        TailNumber::Odd>;
+                                    Run(kernel);
+                                }
+                                else
+                                {
+                                    const auto kernel = kernel_gemm_b_preshuffle_wmma_cshuffle_v3<
+                                        GridwiseGemm,
+                                        false,
+                                        InMemoryDataOperationEnum::AtomicAdd,
+                                        minimum_occupancy,
+                                        TailNumber::Even>;
+                                    Run(kernel);
+                                }
+                            }
+                        }
+                        else
+                        {
+                            if(GridwiseGemm::CalculateKBlockLoopTailNum(K_split) == TailNumber::Odd)
+                            {
+                                const auto kernel = kernel_gemm_b_preshuffle_wmma_cshuffle_v3<
+                                    GridwiseGemm,
+                                    false,
+                                    InMemoryDataOperationEnum::Set,
+                                    minimum_occupancy,
+                                    TailNumber::Odd>;
+                                Run(kernel);
+                            }
+                            else
+                            {
+                                const auto kernel = kernel_gemm_b_preshuffle_wmma_cshuffle_v3<
+                                    GridwiseGemm,
+                                    false,
+                                    InMemoryDataOperationEnum::Set,
+                                    minimum_occupancy,
+                                    TailNumber::Even>;
+                                Run(kernel);
+                            }
+                        }
+                    }
+                }
+                else
+                {
+                    if constexpr(BlkGemmPipelineVer == BlockGemmPipelineVersion::v1)
+                    {
+                        if(arg.KBatch > 1)
+                        {
+                            if constexpr(AtomicsImplementationExists)
+                            {
+                                const auto kernel = kernel_gemm_wmma_cshuffle_v3<
+                                    GridwiseGemm,
+                                    false,
+                                    InMemoryDataOperationEnum::AtomicAdd,
+                                    minimum_occupancy>;
+                                Run(kernel);
+                            }
+                        }
+                        else
+                        {
+                            const auto kernel =
+                                kernel_gemm_wmma_cshuffle_v3<GridwiseGemm,
+                                                             false,
+                                                             InMemoryDataOperationEnum::Set,
+                                                             minimum_occupancy>;
+                            Run(kernel);
+                        }
                     }
                 }
             }
@@ -299,6 +413,14 @@ struct DeviceGemm_Wmma_CShuffleV3_Common
             return false;
         }
 
+        if constexpr(IsBPreShuffled)
+        {
+            if(arg.N % NPerBlock != 0 || arg.K % KPerBlock != 0)
+            {
+                return false;
+            }
+        }
+
         return GridwiseGemm::CheckValidity(arg);
     }
 };

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

@@ -388,11 +388,11 @@ struct ABTransferThreadTiles
             // 1 - MNRepeat - K0 / KRow - MNWaves - KRow - MNPerWmma - K1
             return transform_tensor_descriptor(
                 BlockDesc{},
-                make_tuple(
-                    make_unmerge_transform(make_tuple(Number<ABK0 / KRow>{}, KRow, Number<1>{})),
-                    make_unmerge_transform(
-                        make_tuple(Number<MNRepeat>{}, Number<MNWaves>{}, Number<MNPerWmma>{})),
-                    make_pass_through_transform(Number<ABK1>{})),
+                make_tuple(make_unmerge_transform(make_tuple(
+                               Number<ABK0 / KRow>{}, KRow, Number<KPack / KRow / ABK1>{})),
+                           make_unmerge_transform(make_tuple(
+                               Number<MNRepeat>{}, Number<MNWaves>{}, Number<MNPerWmma>{})),
+                           make_pass_through_transform(Number<ABK1>{})),
                 make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}),
                 make_tuple(Sequence<2, 4, 0>{}, Sequence<1, 3, 5>{}, Sequence<6>{}));
         }

include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_gemm_wmma_cshuffle_v3.hpp

@@ -895,8 +895,9 @@ struct GridwiseBatchedGemmGemm_wmma_cshuffle_v3
         c_thread_buf.Clear();
 
         // Empty BScale struct for the blockwise pipeline.
-        using BScale        = typename BlockwiseGemmPipe::Empty;
-        auto b_scale_struct = BScale{};
+        using ABScale       = typename BlockwiseGemmPipe::Empty;
+        auto a_scale_struct = ABScale{};
+        auto b_scale_struct = ABScale{};
 
 /*******************************************************************************/
         // 
@@ -919,6 +920,7 @@ struct GridwiseBatchedGemmGemm_wmma_cshuffle_v3
                                                                               b0_block_buf,
                                                                               b0_block_slice_copy_step,
                                                                               acc0_thread_buf,
+                                                                              a_scale_struct,
                                                                               b_scale_struct,
                                                                               KBlockMainLoop,
                                                                               1); // num_k_block_per_scale

include/ck/tensor_operation/gpu/grid/gridwise_gemm_wmma_cshuffle_v3.hpp

@@ -618,8 +618,9 @@ struct GridwiseGemm_wmma_cshuffle_v3
             __builtin_amdgcn_readfirstlane(block_work_idx[Number<BlockMapNBlockIndex>{}]);
 
         // BScale struct (Empty)
-        using BScale        = typename BlockwiseGemmPipe::Empty;
-        auto b_scale_struct = BScale{};
+        using Scale         = typename BlockwiseGemmPipe::Empty;
+        auto a_scale_struct = Scale{};
+        auto b_scale_struct = Scale{};
 
         const index_t num_k_block_per_scale = GetKBlockPerScale();
 
@@ -627,6 +628,7 @@ struct GridwiseGemm_wmma_cshuffle_v3
                            decltype(bs_grid_desc_bk0_n_bk1),
                            decltype(ds_grid_desc_mblock_mperblock_nblock_nperblock),
                            decltype(e_grid_desc_mblock_mperblock_nblock_nperblock),
+                           decltype(a_scale_struct),
                            decltype(b_scale_struct),
                            decltype(epilogue_args),
                            HasMainKBlockLoop,
@@ -646,6 +648,7 @@ struct GridwiseGemm_wmma_cshuffle_v3
                                     block_m_id,
                                     block_n_id,
                                     num_k_block_per_scale,
+                                    a_scale_struct,
                                     b_scale_struct,
                                     epilogue_args,
                                     k_id);

include/ck/tensor_operation/gpu/grid/gridwise_gemm_wmma_cshuffle_v3_ab_scale.hpp

@@ -23,6 +23,7 @@ template <typename ALayout,
           typename DsLayout,
           typename ELayout,
           typename AsDataType,
+          typename AScaleType,
           typename BsDataType,
           typename BScaleType,
           typename AccDataType,
@@ -34,6 +35,7 @@ template <typename ALayout,
           typename CDEElementwiseOperation,
           tensor_operation::device::GemmSpecialization GemmSpec,
           index_t BlockSize,
+          index_t ScaleBlockM,
           index_t ScaleBlockN, // scale N
           index_t ScaleBlockK, // scale K
           index_t MPerBlock,
@@ -65,13 +67,16 @@ template <typename ALayout,
           index_t CShuffleNRepeatPerShuffle,
           typename CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
           typename CDEShuffleBlockTransferScalarPerVectors,
-          BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave,
-          BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v4,
-          typename ComputeTypeA                       = EDataType,
-          typename ComputeTypeB                       = ComputeTypeA,
-          bool PermuteA                               = false,
-          bool PermuteB                               = false>
-struct GridwiseGemm_wmma_cshuffle_v3_b_scale
+          BlockGemmPipelineScheduler BlkGemmPipeSched,
+          BlockGemmPipelineVersion BlkGemmPipelineVer,
+          typename ComputeTypeA,
+          typename ComputeTypeB,
+          bool PermuteA,
+          bool PermuteB,
+          bool IsBPreShuffled   = false,
+          typename AScaleLayout = ALayout,
+          typename BScaleLayout = BLayout>
+struct GridwiseGemm_wmma_cshuffle_v3_ab_scale
     : GridwiseGemm_wmma_cshuffle_v3_base<
           ALayout,
           BLayout,
@@ -123,7 +128,7 @@ struct GridwiseGemm_wmma_cshuffle_v3_b_scale
           ComputeTypeB,
           PermuteA,
           PermuteB,
-          false,
+          IsBPreShuffled,
           true>
 {
     using Base = GridwiseGemm_wmma_cshuffle_v3_base<
@@ -177,7 +182,7 @@ struct GridwiseGemm_wmma_cshuffle_v3_b_scale
         ComputeTypeB,
         PermuteA,
         PermuteB,
-        false,
+        IsBPreShuffled,
         true>;
 
     using Base::I0;
@@ -233,6 +238,7 @@ struct GridwiseGemm_wmma_cshuffle_v3_b_scale
                          std::array<index_t, NumBTensor> StrideBs_,
                          std::array<index_t, NumDTensor> StrideDs_,
                          index_t StrideE_,
+                         index_t StrideScaleA_,
                          index_t StrideScaleB_,
                          index_t KBatch_)
             : M{M_},
@@ -242,6 +248,7 @@ struct GridwiseGemm_wmma_cshuffle_v3_b_scale
               StrideBs{StrideBs_},
               StrideDs{StrideDs_},
               StrideE{StrideE_},
+              StrideScaleA{StrideScaleA_},
               StrideScaleB{StrideScaleB_},
               KBatch{KBatch_},
               MPadded{CalculateMPadded(M_)},
@@ -251,7 +258,8 @@ struct GridwiseGemm_wmma_cshuffle_v3_b_scale
               AK0{CalculateAK0Padded(K_, KBatch_)},
               BK0{CalculateBK0Padded(K_, KBatch_)},
               MBlock{CalculateMBlock(M_)},
-              NBlock{CalculateNBlock(N_)}
+              NBlock{CalculateNBlock(N_)},
+              Kt{K_}
         {
         }
 
@@ -275,11 +283,11 @@ struct GridwiseGemm_wmma_cshuffle_v3_b_scale
                 });
                 std::cout << " }, ";
             }
-            std::cout << "SE:" << StrideE << ", " << "SScaleB:" << StrideScaleB << ", "
-                      << "MP:" << MPadded << ", " << "NP:" << NPadded << ", " << "KRead:" << KRead
-                      << ", " << "KP:" << KPadded << ", " << "AK0:" << AK0 << ", " << "BK0:" << BK0
-                      << ", " << "MBlock: " << MBlock << ", " << "NBlock: " << NBlock << "}"
-                      << std::endl;
+            std::cout << "SE:" << StrideE << ", " << "SScaleA:" << StrideScaleA << ", "
+                      << "SScaleB:" << StrideScaleB << ", " << "MP:" << MPadded << ", "
+                      << "NP:" << NPadded << ", " << "KRead:" << KRead << ", " << "KP:" << KPadded
+                      << ", " << "AK0:" << AK0 << ", " << "BK0:" << BK0 << ", "
+                      << "MBlock: " << MBlock << ", " << "NBlock: " << NBlock << "}" << std::endl;
         }
 
         index_t M;
@@ -289,6 +297,7 @@ struct GridwiseGemm_wmma_cshuffle_v3_b_scale
         std::array<index_t, NumBTensor> StrideBs;
         std::array<index_t, NumDTensor> StrideDs;
         index_t StrideE;
+        index_t StrideScaleA;
         index_t StrideScaleB;
         index_t KBatch;
         index_t MPadded;
@@ -299,6 +308,7 @@ struct GridwiseGemm_wmma_cshuffle_v3_b_scale
         index_t BK0;
         index_t MBlock;
         index_t NBlock;
+        index_t Kt;
     };
 
     // Argument
@@ -315,7 +325,9 @@ struct GridwiseGemm_wmma_cshuffle_v3_b_scale
                           std::array<index_t, NumBTensor> StrideBs_,
                           std::array<index_t, NumDTensor> StrideDs_,
                           index_t StrideE_,
+                          index_t StrideScaleA_,
                           index_t StrideScaleB_,
+                          const AScaleType* p_a_scale_grid_,
                           const BScaleType* p_b_scale_grid_,
                           index_t k_batch_,
                           AElementwiseOperation a_element_op_,
@@ -329,12 +341,14 @@ struct GridwiseGemm_wmma_cshuffle_v3_b_scale
                       StrideBs_,
                       StrideDs_,
                       StrideE_,
+                      StrideScaleA_,
                       StrideScaleB_,
                       k_batch_},
               p_as_grid{},
               p_bs_grid{},
               p_ds_grid{},
               p_e_grid{p_e_grid_},
+              p_a_scale_grid{p_a_scale_grid_},
               p_b_scale_grid{p_b_scale_grid_},
               a_element_op{a_element_op_},
               b_element_op{b_element_op_},
@@ -379,6 +393,7 @@ struct GridwiseGemm_wmma_cshuffle_v3_b_scale
         DsGridPointer p_ds_grid;
         EDataType* p_e_grid;
 
+        const AScaleType* p_a_scale_grid;
         const BScaleType* p_b_scale_grid;
         const AElementwiseOperation a_element_op;
         const BElementwiseOperation b_element_op;
@@ -407,34 +422,52 @@ struct GridwiseGemm_wmma_cshuffle_v3_b_scale
                     [&](auto i) { a_k_split_offset[i] = k_id * karg.KRead * karg.StrideAs[i]; });
             }
 
-            if constexpr(is_same_v<tensor_layout::gemm::RowMajor, BLayout>)
+            if constexpr(IsBPreShuffled)
             {
-                static_for<0, NumBTensor, 1>{}(
-                    [&](auto i) { b_k_split_offset[i] = k_id * karg.KRead * karg.StrideBs[i]; });
+                static_for<0, NumBTensor, 1>{}([&](auto i) { b_k_split_offset[i] = 0; });
             }
-            else if constexpr(is_same_v<tensor_layout::gemm::ColumnMajor, BLayout>)
+            else
             {
-                if constexpr(!PermuteB)
+                if constexpr(is_same_v<tensor_layout::gemm::RowMajor, BLayout>)
                 {
-                    static_for<0, NumBTensor, 1>{}(
-                        [&](auto i) { b_k_split_offset[i] = k_id * karg.KRead / BPackedSize; });
+                    static_for<0, NumBTensor, 1>{}([&](auto i) {
+                        b_k_split_offset[i] = k_id * karg.KRead * karg.StrideBs[i];
+                    });
                 }
-                else
+                else if constexpr(is_same_v<tensor_layout::gemm::ColumnMajor, BLayout>)
                 {
-                    const int k0_offset = karg.KRead * karg.N;
-                    static_for<0, NumBTensor, 1>{}(
-                        [&](auto i) { b_k_split_offset[i] = k_id * k0_offset / BPackedSize; });
+                    if constexpr(!PermuteB)
+                    {
+                        static_for<0, NumBTensor, 1>{}(
+                            [&](auto i) { b_k_split_offset[i] = k_id * karg.KRead / BPackedSize; });
+                    }
+                    else
+                    {
+                        const int k0_offset = karg.KRead * karg.N;
+                        static_for<0, NumBTensor, 1>{}(
+                            [&](auto i) { b_k_split_offset[i] = k_id * k0_offset / BPackedSize; });
+                    }
                 }
             }
 
-            // Calculate B scale offset
-            if constexpr(is_same_v<tensor_layout::gemm::RowMajor, BLayout>)
+            // Calculate A scale offset
+            if constexpr(is_same_v<tensor_layout::gemm::RowMajor, AScaleLayout>)
             {
-                scale_k_split_offset = k_id * (karg.KRead / ScaleBlockK) * karg.StrideB;
+                scale_a_k_split_offset = k_id * (karg.KRead / ScaleBlockK);
             }
-            else if constexpr(is_same_v<tensor_layout::gemm::ColumnMajor, BLayout>)
+            else if constexpr(is_same_v<tensor_layout::gemm::ColumnMajor, AScaleLayout>)
             {
-                scale_k_split_offset = k_id * (karg.KRead / ScaleBlockK);
+                scale_a_k_split_offset = k_id * (karg.KRead / ScaleBlockK) * karg.StrideScaleA;
+            }
+
+            // Calculate B scale offset
+            if constexpr(is_same_v<tensor_layout::gemm::RowMajor, BScaleLayout>)
+            {
+                scale_b_k_split_offset = k_id * (karg.KRead / ScaleBlockK) * karg.StrideScaleB;
+            }
+            else if constexpr(is_same_v<tensor_layout::gemm::ColumnMajor, BScaleLayout>)
+            {
+                scale_b_k_split_offset = k_id * (karg.KRead / ScaleBlockK);
             }
 
             if(k_id < karg.KBatch - 1)
@@ -458,77 +491,225 @@ struct GridwiseGemm_wmma_cshuffle_v3_b_scale
 
         std::array<index_t, NumATensor> a_k_split_offset;
         std::array<index_t, NumBTensor> b_k_split_offset;
-        index_t scale_k_split_offset; // New member for scale matrix offset
+        index_t scale_a_k_split_offset; // A scale matrix offset
+        index_t scale_b_k_split_offset; // B scale matrix offset
         index_t c_reduce_offset;
     };
 
     using BlockwiseGemmPipe = typename Base::BlockwiseGemmPipe;
 
     // return block_id to C matrix tile idx (m0, n0) mapping
-    // if arch = gfx942
     using Block2CTileMap = BlockToCTileMap_Grouped_M00_N0_M01Adapt<8, MPerBlock, NPerBlock>;
-    // using Block2CTileMap = BlockToCTileMap_3DGrid_KSplit<MPerBlock, NPerBlock>;
 
-    template <index_t NumberOfBuffers, typename BScaleGridDesc_BN_AK>
-    __device__ static auto MakeBScale(const BScaleGridDesc_BN_AK& b_scale_grid_desc_bn_ak,
-                                      const BScaleType* p_b_scale_grid,
-                                      index_t block_n_id)
+    __device__ static constexpr auto
+    MakeAScaleGridDesciptor_M_K(index_t M, index_t K, index_t StrideScaleA)
     {
-        const auto b_scale_grid_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
-            p_b_scale_grid, b_scale_grid_desc_bn_ak.GetElementSpaceSize());
+        const auto BM = math::integer_divide_ceil(M, ScaleBlockM);
+        const auto BK = math::integer_divide_ceil(K, ScaleBlockK);
+        if constexpr(is_same<tensor_layout::gemm::RowMajor, AScaleLayout>::value)
+        {
+            return make_naive_tensor_descriptor(make_tuple(BM, BK), make_tuple(StrideScaleA, I1));
+        }
+        else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, AScaleLayout>::value)
+        {
+            return make_naive_tensor_descriptor(make_tuple(BM, BK), make_tuple(I1, StrideScaleA));
+        }
+    }
 
-        static constexpr auto wmma =
-            WmmaSelector<ComputeTypeA, ComputeTypeB, AccDataType, MPerWmma, NPerWmma>{};
-        static constexpr auto KPerThread = wmma.selected_wmma.k_per_wmma;
+    template <index_t NumberOfBuffers>
+    __device__ static auto
+    MakeAScale(const Problem& problem, const AScaleType* p_a_scale_grid, index_t block_m_id)
+    {
+        if constexpr(ck::is_same_v<AScaleType, void>)
+        {
+            using AScale = typename BlockwiseGemmPipe::Empty;
+            return AScale{};
+        }
+        else
+        {
+#if defined(__gfx11__)
+            // TODO: remove this restriction
+            static_assert(ScaleBlockM >= MPerWmma,
+                          "ScaleBlockM must be greater equal than MPerWmma");
+#endif
+            static_assert(
+                ScaleBlockK >=
+                    WmmaSelector<ComputeTypeA, ComputeTypeB, AccDataType, MPerWmma, NPerWmma>::
+                        selected_wmma.k_per_wmma,
+                "ScaleBlockK must be greater equal than KPerWmma");
 
-        static constexpr auto ScaleSliceSizeN = NRepeat;
-        static constexpr auto ScaleSliceSizeK = (KPerThread + ScaleBlockK - 1) / ScaleBlockK;
+            const auto a_scale_grid_desc_am_ak =
+                MakeAScaleGridDesciptor_M_K(problem.M, problem.K, problem.StrideScaleA);
 
-        constexpr auto b_scale_thread_desc = make_naive_tensor_descriptor_packed(
-            make_tuple(Number<ScaleSliceSizeN>{}, Number<ScaleSliceSizeK>{}));
+            const auto a_scale_grid_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+                p_a_scale_grid, a_scale_grid_desc_am_ak.GetElementSpaceSize());
 
-        constexpr index_t NWaves = NPerBlock / (NRepeat * NPerWmma);
+            constexpr auto wmma =
+                WmmaSelector<ComputeTypeA, ComputeTypeB, AccDataType, MPerWmma, NPerWmma>{};
+            constexpr auto RegSizePerWmmaFull =
+                wmma.selected_wmma.num_acc_vgprs_per_wave * wmma.selected_wmma.acc_pack_number;
+            constexpr auto RegSizePerWmma =
+                math::integer_divide_ceil(RegSizePerWmmaFull, ScaleBlockM);
 
-        auto b_thread_offset_n = get_thread_local_1d_id() % NPerWmma +
-                                 (get_thread_local_1d_id() / 32) % NWaves * NPerWmma;
-        auto b_thread_offset_k = (get_thread_local_1d_id() % 32) / NPerWmma * KPerThread;
+            constexpr index_t MWaves = MPerBlock / (MRepeat * MPerWmma);
+            constexpr index_t NWaves = NPerBlock / (NRepeat * NPerWmma);
 
-        auto b_scale_thread_copy =
-            ThreadwiseTensorSliceTransfer_v2<BScaleType,
-                                             BScaleType,
-                                             decltype(b_scale_grid_desc_bn_ak),
-                                             decltype(b_scale_thread_desc),
-                                             Sequence<1, ScaleSliceSizeK>,
-                                             Sequence<0, 1>,
-                                             1,
-                                             ScaleSliceSizeK,
-                                             1,
-                                             false>(
-                b_scale_grid_desc_bn_ak,
-                make_multi_index(block_n_id * NPerBlock / ScaleBlockN + b_thread_offset_n,
-                                 b_thread_offset_k / ScaleBlockK));
+            constexpr auto ScaleSliceSizeM =
+                ScaleBlockM < MPerWmma ? MRepeat * RegSizePerWmma
+                                       : math::integer_divide_ceil(MPerBlock, ScaleBlockM);
+            constexpr auto ScaleSliceStrideM =
+                math::integer_divide_ceil(MWaves * MPerWmma, ScaleBlockM);
+            constexpr auto ScaleSliceSizeK = math::integer_divide_ceil(KPerBlock, ScaleBlockK);
 
-        auto b_scale_thread_buf = make_static_buffer<AddressSpaceEnum::Vgpr, ComputeTypeB>(
-            b_scale_thread_desc.GetElementSpaceSize());
+            constexpr auto a_scale_thread_desc = make_naive_tensor_descriptor_packed(
+                make_tuple(Number<ScaleSliceSizeM>{}, Number<ScaleSliceSizeK>{}));
 
-        using BScale =
-            typename BlockwiseGemmPipe::template BScale<ScaleSliceSizeN,
-                                                        ScaleSliceSizeK,
-                                                        NWaves,
-                                                        ScaleBlockK,
-                                                        NumberOfBuffers,
-                                                        decltype(b_scale_grid_desc_bn_ak),
-                                                        decltype(b_scale_thread_copy),
-                                                        decltype(b_scale_grid_buf),
-                                                        decltype(b_scale_thread_buf),
-                                                        decltype(b_scale_thread_desc)>;
+            auto a_thread_offset_m =
+                ((get_thread_local_1d_id() % 32) / MPerWmma * RegSizePerWmma) /
+                    math::integer_divide_ceil(ScaleBlockM, RegSizePerWmmaFull) +
+                (get_thread_local_1d_id() / 32) / NWaves * MPerWmma / ScaleBlockM;
 
-        return BScale{b_scale_grid_desc_bn_ak, b_scale_thread_copy, b_scale_grid_buf};
+            constexpr index_t VectorDim =
+                is_same<tensor_layout::gemm::ColumnMajor, AScaleLayout>::value ? 0 : 1;
+            constexpr index_t VectorSize =
+                is_same<tensor_layout::gemm::ColumnMajor, AScaleLayout>::value ? RegSizePerWmma
+                                                                               : ScaleSliceSizeK;
+
+            auto a_scale_thread_copy =
+                ThreadwiseTensorSliceTransfer_v2<AScaleType,
+                                                 AScaleType,
+                                                 decltype(a_scale_grid_desc_am_ak),
+                                                 decltype(a_scale_thread_desc),
+                                                 Sequence<RegSizePerWmma, ScaleSliceSizeK>,
+                                                 Sequence<0, 1>,
+                                                 VectorDim,
+                                                 VectorSize,
+                                                 1,
+                                                 true>(
+                    a_scale_grid_desc_am_ak,
+                    make_multi_index(block_m_id * MPerBlock / ScaleBlockM + a_thread_offset_m, 0));
+
+            auto a_scale_thread_buf = make_static_buffer<AddressSpaceEnum::Vgpr, AScaleType>(
+                a_scale_thread_desc.GetElementSpaceSize());
+
+            using AScale =
+                typename BlockwiseGemmPipe::template ABScale<ScaleSliceSizeM,
+                                                             ScaleSliceStrideM,
+                                                             ScaleSliceSizeK,
+                                                             NumberOfBuffers,
+                                                             RegSizePerWmma,
+                                                             decltype(a_scale_grid_desc_am_ak),
+                                                             decltype(a_scale_thread_copy),
+                                                             decltype(a_scale_grid_buf),
+                                                             decltype(a_scale_thread_buf),
+                                                             decltype(a_scale_thread_desc)>;
+
+            return AScale{a_scale_grid_desc_am_ak, a_scale_thread_copy, a_scale_grid_buf};
+        }
+    }
+
+    __device__ static constexpr auto
+    MakeBScaleGridDesciptor_N_K(index_t N, index_t K, index_t StrideScaleB)
+    {
+        const auto BN = math::integer_divide_ceil(N, ScaleBlockN);
+        const auto BK = math::integer_divide_ceil(K, ScaleBlockK);
+        if constexpr(is_same<tensor_layout::gemm::ColumnMajor, BScaleLayout>::value)
+        {
+            return make_naive_tensor_descriptor(make_tuple(BN, BK), make_tuple(StrideScaleB, I1));
+        }
+        else if constexpr(is_same<tensor_layout::gemm::RowMajor, BScaleLayout>::value)
+        {
+            return make_naive_tensor_descriptor(make_tuple(BN, BK), make_tuple(I1, StrideScaleB));
+        }
+    }
+
+    template <index_t NumberOfBuffers>
+    __device__ static auto
+    MakeBScale(const Problem& problem, const BScaleType* p_b_scale_grid, index_t block_n_id)
+    {
+        if constexpr(ck::is_same_v<BScaleType, void>)
+        {
+            using BScale = typename BlockwiseGemmPipe::Empty;
+            return BScale{};
+        }
+        else
+        {
+            static_assert(
+                ScaleBlockK >=
+                    WmmaSelector<ComputeTypeA, ComputeTypeB, AccDataType, MPerWmma, NPerWmma>::
+                        selected_wmma.k_per_wmma,
+                "ScaleBlockK must be greater equal than KPerWmma");
+
+            const auto b_scale_grid_desc_bn_ak =
+                MakeBScaleGridDesciptor_N_K(problem.N, problem.K, problem.StrideScaleB);
+
+            const auto b_scale_grid_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+                p_b_scale_grid, b_scale_grid_desc_bn_ak.GetElementSpaceSize());
+
+            constexpr index_t NWaves = NPerBlock / (NRepeat * NPerWmma);
+
+            constexpr auto ScaleSliceSizeN =
+                ScaleBlockN < NPerWmma ? NRepeat
+                                       : math::integer_divide_ceil(NPerBlock, ScaleBlockN);
+            constexpr auto ScaleSliceStrideN =
+                math::integer_divide_ceil(NWaves * NPerWmma, ScaleBlockN);
+            constexpr auto ScaleSliceSizeK = math::integer_divide_ceil(KPerBlock, ScaleBlockK);
+
+            constexpr auto b_scale_thread_desc = make_naive_tensor_descriptor_packed(
+                make_tuple(Number<ScaleSliceSizeN>{}, Number<ScaleSliceSizeK>{}));
+
+            auto b_thread_offset_n = (get_thread_local_1d_id() % NPerWmma +
+                                      (get_thread_local_1d_id() / 32) % NWaves * NPerWmma) /
+                                     ScaleBlockN;
+
+            constexpr index_t VectorDim =
+                is_same<tensor_layout::gemm::RowMajor, BScaleLayout>::value ? 0 : 1;
+            constexpr index_t VectorSize =
+                is_same<tensor_layout::gemm::RowMajor, BScaleLayout>::value ? 1 : ScaleSliceSizeK;
+
+            auto b_scale_thread_copy =
+                ThreadwiseTensorSliceTransfer_v2<BScaleType,
+                                                 BScaleType,
+                                                 decltype(b_scale_grid_desc_bn_ak),
+                                                 decltype(b_scale_thread_desc),
+                                                 Sequence<1, ScaleSliceSizeK>,
+                                                 Sequence<0, 1>,
+                                                 VectorDim,
+                                                 VectorSize,
+                                                 1,
+                                                 true>(
+                    b_scale_grid_desc_bn_ak,
+                    make_multi_index(block_n_id * NPerBlock / ScaleBlockN + b_thread_offset_n, 0));
+
+            auto b_scale_thread_buf = make_static_buffer<AddressSpaceEnum::Vgpr, BScaleType>(
+                b_scale_thread_desc.GetElementSpaceSize());
+
+            using BScale =
+                typename BlockwiseGemmPipe::template ABScale<ScaleSliceSizeN,
+                                                             ScaleSliceStrideN,
+                                                             ScaleSliceSizeK,
+                                                             NumberOfBuffers,
+                                                             1,
+                                                             decltype(b_scale_grid_desc_bn_ak),
+                                                             decltype(b_scale_thread_copy),
+                                                             decltype(b_scale_grid_buf),
+                                                             decltype(b_scale_thread_buf),
+                                                             decltype(b_scale_thread_desc)>;
+
+            return BScale{b_scale_grid_desc_bn_ak, b_scale_thread_copy, b_scale_grid_buf};
+        }
     }
 
     __device__ static index_t GetKBlockPerScale()
     {
-        return (ScaleBlockK + KPerBlock - 1) / KPerBlock;
+        if constexpr(ck::is_same_v<AScaleType, void> && ck::is_same_v<BScaleType, void>)
+        {
+            return 0;
+        }
+        else
+        {
+            return (ScaleBlockK + KPerBlock - 1) / KPerBlock;
+        }
     }
 
     template <bool HasMainKBlockLoop,
@@ -539,18 +720,21 @@ struct GridwiseGemm_wmma_cshuffle_v3_b_scale
                                BsGridPointer& p_bs_grid,
                                DsGridPointer& p_ds_grid,
                                EDataType* p_e_grid,
+                               const AScaleType* p_a_scale_grid,
                                const BScaleType* p_b_scale_grid,
                                void* p_shared,
                                const Problem& problem,
                                AElementwiseOperation a_element_op,
                                BElementwiseOperation b_element_op,
                                CDEElementwiseOperation cde_element_op,
-                               EpilogueArgument& epilogue_args)
+                               EpilogueArgument& epilogue_args,
+                               const index_t k_id = 0)
     {
         const auto as_grid_desc_ak0_m_ak1 = MakeAsGridDescriptor_AK0_M_AK1(
             problem.M, problem.MPadded, problem.K, problem.KPadded, problem.StrideAs, problem.AK0);
+        const index_t K_b                 = IsBPreShuffled ? problem.Kt : problem.K;
         const auto bs_grid_desc_bk0_n_bk1 = MakeBsGridDescriptor_BK0_N_BK1(
-            problem.K, problem.KPadded, problem.N, problem.NPadded, problem.StrideBs, problem.BK0);
+            K_b, problem.KPadded, problem.N, problem.NPadded, problem.StrideBs, problem.BK0);
         const auto ds_grid_desc_m_n = MakeDsGridDescriptor_M_N(
             problem.M, problem.MPadded, problem.N, problem.NPadded, problem.StrideDs);
         const auto e_grid_desc_m_n = Base::template MakeDEGridDescriptor_M_N<ELayout>(
@@ -562,12 +746,6 @@ struct GridwiseGemm_wmma_cshuffle_v3_b_scale
             MakeDEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
                 e_grid_desc_m_n, problem.MBlock, problem.NBlock);
 
-        // B Scale grid
-        const auto b_scale_grid_desc_bn_ak = make_naive_tensor_descriptor(
-            make_tuple(math::integer_divide_ceil(problem.N, ScaleBlockN),
-                       math::integer_divide_ceil(problem.K, ScaleBlockK)),
-            make_tuple(problem.StrideScaleB, 1));
-
         // divide block work by [M, N]
         const auto block_2_ctile_map = Block2CTileMap{problem.M, problem.N, 4};
 
@@ -585,8 +763,11 @@ struct GridwiseGemm_wmma_cshuffle_v3_b_scale
         const index_t block_m_id = __builtin_amdgcn_readfirstlane(block_work_idx[I0]);
         const index_t block_n_id = __builtin_amdgcn_readfirstlane(block_work_idx[I1]);
 
+        // AScale struct
+        auto a_scale_struct = MakeAScale<1>(problem, p_a_scale_grid, block_m_id);
+
         // BScale struct
-        auto b_scale_struct = MakeBScale<1>(b_scale_grid_desc_bn_ak, p_b_scale_grid, block_n_id);
+        auto b_scale_struct = MakeBScale<1>(problem, p_b_scale_grid, block_n_id);
 
         const index_t num_k_block_per_scale = GetKBlockPerScale();
 
@@ -594,6 +775,7 @@ struct GridwiseGemm_wmma_cshuffle_v3_b_scale
                            decltype(bs_grid_desc_bk0_n_bk1),
                            decltype(ds_grid_desc_mblock_mperblock_nblock_nperblock),
                            decltype(e_grid_desc_mblock_mperblock_nblock_nperblock),
+                           decltype(a_scale_struct),
                            decltype(b_scale_struct),
                            decltype(epilogue_args),
                            HasMainKBlockLoop,
@@ -613,8 +795,10 @@ struct GridwiseGemm_wmma_cshuffle_v3_b_scale
                                     block_m_id,
                                     block_n_id,
                                     num_k_block_per_scale,
+                                    a_scale_struct,
                                     b_scale_struct,
-                                    epilogue_args);
+                                    epilogue_args,
+                                    k_id);
     }
 
     // NOTE: Wrapper function to have __global__ function in common
@@ -626,7 +810,8 @@ struct GridwiseGemm_wmma_cshuffle_v3_b_scale
     __device__ static void Run(void* p_shared,
                                const SplitKBatchOffset& splitk_batch_offset,
                                Argument& karg,
-                               EpilogueArgument& epilogue_args)
+                               EpilogueArgument& epilogue_args,
+                               const index_t k_id = 0)
     {
         // shift A matrices pointer for splitk
         AsGridPointer p_as_grid_splitk;
@@ -644,18 +829,40 @@ struct GridwiseGemm_wmma_cshuffle_v3_b_scale
                                   splitk_batch_offset.b_k_split_offset[i];
         });
 
+        const AScaleType* p_a_scale_grid_ptr;
+        if constexpr(ck::is_same_v<AScaleType, void>)
+        {
+            p_a_scale_grid_ptr = karg.p_a_scale_grid;
+        }
+        else
+        {
+            p_a_scale_grid_ptr = karg.p_a_scale_grid + splitk_batch_offset.scale_a_k_split_offset;
+        }
+
+        const BScaleType* p_b_scale_grid_ptr;
+        if constexpr(ck::is_same_v<BScaleType, void>)
+        {
+            p_b_scale_grid_ptr = karg.p_b_scale_grid;
+        }
+        else
+        {
+            p_b_scale_grid_ptr = karg.p_b_scale_grid + splitk_batch_offset.scale_b_k_split_offset;
+        }
+
         Run<HasMainKBlockLoop, EGlobalMemoryDataOperation, TailNum>(
             p_as_grid_splitk,
             p_bs_grid_splitk,
             karg.p_ds_grid,
             karg.p_e_grid + splitk_batch_offset.c_reduce_offset,
-            karg.p_b_scale_grid + splitk_batch_offset.scale_k_split_offset,
+            p_a_scale_grid_ptr,
+            p_b_scale_grid_ptr,
             p_shared,
             karg,
             karg.a_element_op,
             karg.b_element_op,
             karg.cde_element_op,
-            epilogue_args);
+            epilogue_args,
+            k_id);
     }
 };
 

include/ck/tensor_operation/gpu/grid/gridwise_gemm_wmma_cshuffle_v3_common.hpp

@@ -69,6 +69,48 @@ __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, MinimumOccupancy)
 #endif
 }
 
+template <typename GridwiseGemm,
+          bool HasMainKBlockLoop,
+          InMemoryDataOperationEnum EGlobalMemoryDataOperation,
+          index_t MinimumOccupancy = 1,
+          TailNumber TailNum       = TailNumber::Full>
+__global__ void
+#if CK_USE_LAUNCH_BOUNDS
+__launch_bounds__(CK_MAX_THREAD_PER_BLOCK, MinimumOccupancy)
+#endif
+    kernel_gemm_b_preshuffle_wmma_cshuffle_v3(typename GridwiseGemm::Argument karg)
+{
+#if(defined(__gfx11__) || defined(__gfx12__))
+#if defined(__gfx11__)
+    // gfx11 does not support *_atomic_pk_add_f16/bf16 instructions
+    using e_data_type = remove_cvref_t<remove_pointer_t<decltype(karg.p_e_grid)>>;
+    if constexpr(!(EGlobalMemoryDataOperation == InMemoryDataOperationEnum::AtomicAdd &&
+                   (std::is_same_v<e_data_type, ck::half_t> ||
+                    std::is_same_v<e_data_type, ck::bhalf_t>)))
+    {
+#endif
+        constexpr index_t LDS_size = GridwiseGemm::template GetSharedMemoryNumberOfByte<
+            typename GridwiseGemm::EpilogueCShuffle>();
+        __shared__ char p_shared[LDS_size];
+
+        auto splitk_batch_offset = typename GridwiseGemm::SplitKBatchOffset(karg, blockIdx.z);
+
+        const index_t num_k_per_block = math::integer_divide_ceil(karg.K, GridwiseGemm::KPack);
+        const index_t k_id            = blockIdx.z * num_k_per_block;
+
+        auto epilogue_args = typename GridwiseGemm::EpilogueCShuffle{};
+
+        GridwiseGemm::template Run<HasMainKBlockLoop, EGlobalMemoryDataOperation, TailNum>(
+            p_shared, splitk_batch_offset, karg, epilogue_args, k_id);
+
+#if defined(__gfx11__)
+    }
+#endif
+#else
+    ignore = karg;
+#endif
+}
+
 template <typename ALayout,
           typename BLayout,
           typename DsLayout,
@@ -162,7 +204,7 @@ struct GridwiseGemm_wmma_cshuffle_v3_base
 
     static constexpr index_t KInnerB = ck::math::integer_divide_ceil(BK1Value, KPerWmmaBlk);
 
-    static constexpr index_t KInner = ck::math::min(KInnerA, KInnerB);
+    static constexpr index_t KInner = IsBPreShuffled ? KInnerB : ck::math::min(KInnerA, KInnerB);
 
     static constexpr index_t KPack =
         KInner *
@@ -966,6 +1008,7 @@ struct GridwiseGemm_wmma_cshuffle_v3_base
               typename BGridDesc_BK0_N_K1,
               typename DsGridDesc_MBlock_MPerBlock_NBlock_NPerBlock,
               typename EGridDesc_MBlock_MPerBlock_NBlock_NPerBlock,
+              typename AScaleStruct,
               typename BScaleStruct,
               typename EpilogueArgument,
               bool HasMainKBlockLoop,
@@ -988,6 +1031,7 @@ struct GridwiseGemm_wmma_cshuffle_v3_base
                                const index_t& block_m_id,
                                const index_t& block_n_id,
                                const index_t& num_k_block_per_scale,
+                               AScaleStruct& a_scale_struct,
                                BScaleStruct& b_scale_struct,
                                EpilogueArgument& epilogue_args,
                                const index_t k_id = 0)
@@ -1072,6 +1116,7 @@ struct GridwiseGemm_wmma_cshuffle_v3_base
             b_block_buf,
             b_block_slice_copy_step,
             c_thread_buf,
+            a_scale_struct,
             b_scale_struct,
             num_k_block_main_loop,
             num_k_block_per_scale);

include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_v3_multi_d_ab_scale.hpp

@@ -43,13 +43,15 @@ __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, MinimumOccupancy)
     {
         __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
+        auto splitk_batch_offset = typename GridwiseGemm::SplitKBatchOffset(karg);
+
         GridwiseGemm::template Run<HasMainKBlockLoop, CGlobalMemoryDataOperation, TailNum>(
-            karg.p_a_grid,
-            karg.p_b_grid,
+            karg.p_a_grid + splitk_batch_offset.a_k_split_offset,
+            karg.p_b_grid + splitk_batch_offset.b_k_split_offset,
             karg.p_ds_grid,
             karg.p_c_grid,
-            karg.p_a_scale_grid,
-            karg.p_b_scale_grid,
+            karg.p_a_scale_grid + splitk_batch_offset.scale_a_k_split_offset,
+            karg.p_b_scale_grid + splitk_batch_offset.scale_b_k_split_offset,
             p_shared,
             karg,
             karg.a_element_op,
@@ -405,31 +407,33 @@ struct GridwiseGemmMultiD_ABScale_xdl_cshuffle_v3
         }
     }
 
-    __host__ __device__ static constexpr auto MakeAScaleGridDesciptor_M_K(index_t M, index_t K)
+    __host__ __device__ static constexpr auto
+    MakeAScaleGridDesciptor_M_K(index_t M, index_t K, index_t StrideScaleA)
     {
         const auto BM = math::integer_divide_ceil(M, ScaleBlockM);
         const auto BK = math::integer_divide_ceil(K, ScaleBlockK);
         if constexpr(is_same<tensor_layout::gemm::RowMajor, ALayout>::value)
         {
-            return make_naive_tensor_descriptor(make_tuple(BM, BK), make_tuple(BK, I1));
+            return make_naive_tensor_descriptor(make_tuple(BM, BK), make_tuple(StrideScaleA, I1));
         }
         else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, ALayout>::value)
         {
-            return make_naive_tensor_descriptor(make_tuple(BM, BK), make_tuple(I1, BM));
+            return make_naive_tensor_descriptor(make_tuple(BM, BK), make_tuple(I1, StrideScaleA));
         }
     }
 
-    __host__ __device__ static constexpr auto MakeBScaleGridDesciptor_N_K(index_t N, index_t K)
+    __host__ __device__ static constexpr auto
+    MakeBScaleGridDesciptor_N_K(index_t N, index_t K, index_t StrideScaleB)
     {
         const auto BN = math::integer_divide_ceil(N, ScaleBlockN);
         const auto BK = math::integer_divide_ceil(K, ScaleBlockK);
         if constexpr(is_same<tensor_layout::gemm::ColumnMajor, BLayout>::value)
         {
-            return make_naive_tensor_descriptor(make_tuple(BN, BK), make_tuple(BK, I1));
+            return make_naive_tensor_descriptor(make_tuple(BN, BK), make_tuple(StrideScaleB, I1));
         }
         else if constexpr(is_same<tensor_layout::gemm::RowMajor, BLayout>::value)
         {
-            return make_naive_tensor_descriptor(make_tuple(BN, BK), make_tuple(I1, BN));
+            return make_naive_tensor_descriptor(make_tuple(BN, BK), make_tuple(I1, StrideScaleB));
         }
     }
 
@@ -548,6 +552,8 @@ struct GridwiseGemmMultiD_ABScale_xdl_cshuffle_v3
                          index_t StrideB_,
                          std::array<index_t, NumDTensor> StrideDs_,
                          index_t StrideC_,
+                         index_t StrideScaleA_,
+                         index_t StrideScaleB_,
                          index_t KBatch_)
             : M{M_},
               N{N_},
@@ -556,6 +562,8 @@ struct GridwiseGemmMultiD_ABScale_xdl_cshuffle_v3
               StrideB{StrideB_},
               StrideDs{StrideDs_},
               StrideC{StrideC_},
+              StrideScaleA{StrideScaleA_},
+              StrideScaleB{StrideScaleB_},
               KBatch{KBatch_},
               MPadded{CalculateMPadded(M_)},
               NPadded{CalculateNPadded(N_)},
@@ -585,7 +593,8 @@ struct GridwiseGemmMultiD_ABScale_xdl_cshuffle_v3
         index_t StrideB;
         std::array<index_t, NumDTensor> StrideDs;
         index_t StrideC;
-
+        index_t StrideScaleA;
+        index_t StrideScaleB;
         index_t KBatch;
         index_t MPadded;
         index_t NPadded;
@@ -611,13 +620,24 @@ struct GridwiseGemmMultiD_ABScale_xdl_cshuffle_v3
                           index_t StrideB_,
                           std::array<index_t, NumDTensor> StrideDs_,
                           index_t StrideC_,
+                          index_t StrideScaleA_,
+                          index_t StrideScaleB_,
                           const AScaleType* p_a_scale_grid_,
                           const BScaleType* p_b_scale_grid_,
                           index_t k_batch_,
                           AElementwiseOperation a_element_op_,
                           BElementwiseOperation b_element_op_,
                           CElementwiseOperation c_element_op_)
-            : Problem{M_, N_, K_, StrideA_, StrideB_, StrideDs_, StrideC_, k_batch_},
+            : Problem{M_,
+                      N_,
+                      K_,
+                      StrideA_,
+                      StrideB_,
+                      StrideDs_,
+                      StrideC_,
+                      StrideScaleA_,
+                      StrideScaleB_,
+                      k_batch_},
               p_a_grid{p_a_grid_},
               p_b_grid{p_b_grid_},
               p_ds_grid{},
@@ -673,6 +693,28 @@ struct GridwiseGemmMultiD_ABScale_xdl_cshuffle_v3
                 b_k_split_offset = blockIdx.z * karg.KRead;
             }
 
+            // Calculate A scale offset
+            if constexpr(is_same_v<tensor_layout::gemm::RowMajor, ALayout>)
+            {
+                scale_a_k_split_offset = blockIdx.z * (karg.KRead / ScaleBlockK);
+            }
+            else if constexpr(is_same_v<tensor_layout::gemm::ColumnMajor, ALayout>)
+            {
+                scale_a_k_split_offset =
+                    blockIdx.z * (karg.KRead / ScaleBlockK) * karg.StrideScaleA;
+            }
+
+            // Calculate B scale offset
+            if constexpr(is_same_v<tensor_layout::gemm::RowMajor, BLayout>)
+            {
+                scale_b_k_split_offset =
+                    blockIdx.z * (karg.KRead / ScaleBlockK) * karg.StrideScaleB;
+            }
+            else if constexpr(is_same_v<tensor_layout::gemm::ColumnMajor, BLayout>)
+            {
+                scale_b_k_split_offset = blockIdx.z * (karg.KRead / ScaleBlockK);
+            }
+
             if(blockIdx.z < static_cast<uint32_t>(karg.KBatch - 1))
             {
                 karg.K = karg.KRead;
@@ -685,6 +727,8 @@ struct GridwiseGemmMultiD_ABScale_xdl_cshuffle_v3
 
         index_t a_k_split_offset;
         index_t b_k_split_offset;
+        index_t scale_a_k_split_offset; // A scale matrix offset
+        index_t scale_b_k_split_offset; // B scale matrix offset
     };
 
     __device__ static constexpr auto GetABlockDescriptor_AK0PerBlock_MPerBlock_AK1()
@@ -1221,8 +1265,10 @@ struct GridwiseGemmMultiD_ABScale_xdl_cshuffle_v3
         const auto c_grid_desc_m_n = MakeCGridDescriptor_M_N<CLayout>(
             problem.M, problem.MPadded, problem.N, problem.NPadded, problem.StrideC);
 
-        const auto a_scale_grid_desc_am_ak = MakeAScaleGridDesciptor_M_K(problem.M, problem.K);
-        const auto b_scale_grid_desc_bn_ak = MakeBScaleGridDesciptor_N_K(problem.N, problem.K);
+        const auto a_scale_grid_desc_am_ak =
+            MakeAScaleGridDesciptor_M_K(problem.M, problem.K, problem.StrideScaleA);
+        const auto b_scale_grid_desc_bn_ak =
+            MakeBScaleGridDesciptor_N_K(problem.N, problem.K, problem.StrideScaleB);
 
         const auto c_grid_desc_mblock_mperblock_nblock_nperblock =
             MakeCGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(

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

@@ -16,7 +16,231 @@ namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
 #if(defined(CK_ENABLE_BF16) || defined(CK_ENABLE_FP8))
+#ifdef CK_USE_WMMA_FP8
+// Row, Col
+void add_device_gemm_ab_scale_wmma_f8_f8_bf16_mk_nk_mn_1_128_128_comp_default_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD_ABScaleSplitK<Row,
+                                                                  Col,
+                                                                  Tuple<>,
+                                                                  Row,
+                                                                  F8,
+                                                                  F32,
+                                                                  F8,
+                                                                  F32,
+                                                                  Tuple<>,
+                                                                  BF16,
+                                                                  1,
+                                                                  128,
+                                                                  128,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances);
+
+void add_device_gemm_ab_scale_wmma_f8_f8_bf16_mk_nk_mn_1_128_128_comp_kpadding_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD_ABScaleSplitK<Row,
+                                                                  Col,
+                                                                  Tuple<>,
+                                                                  Row,
+                                                                  F8,
+                                                                  F32,
+                                                                  F8,
+                                                                  F32,
+                                                                  Tuple<>,
+                                                                  BF16,
+                                                                  1,
+                                                                  128,
+                                                                  128,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances);
+
+void add_device_gemm_ab_scale_wmma_f8_f8_bf16_mk_nk_mn_1_128_128_mem_v1_default_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD_ABScaleSplitK<Row,
+                                                                  Col,
+                                                                  Tuple<>,
+                                                                  Row,
+                                                                  F8,
+                                                                  F32,
+                                                                  F8,
+                                                                  F32,
+                                                                  Tuple<>,
+                                                                  BF16,
+                                                                  1,
+                                                                  128,
+                                                                  128,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances);
+
+void add_device_gemm_ab_scale_wmma_f8_f8_bf16_mk_nk_mn_1_128_128_mem_v1_kpadding_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD_ABScaleSplitK<Row,
+                                                                  Col,
+                                                                  Tuple<>,
+                                                                  Row,
+                                                                  F8,
+                                                                  F32,
+                                                                  F8,
+                                                                  F32,
+                                                                  Tuple<>,
+                                                                  BF16,
+                                                                  1,
+                                                                  128,
+                                                                  128,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances);
+
+// Row, Row
+void add_device_gemm_ab_scale_wmma_f8_f8_bf16_mk_kn_mn_1_128_128_comp_default_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD_ABScaleSplitK<Row,
+                                                                  Row,
+                                                                  Tuple<>,
+                                                                  Row,
+                                                                  F8,
+                                                                  F32,
+                                                                  F8,
+                                                                  F32,
+                                                                  Tuple<>,
+                                                                  BF16,
+                                                                  1,
+                                                                  128,
+                                                                  128,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances);
+
+void add_device_gemm_ab_scale_wmma_f8_f8_bf16_mk_kn_mn_1_128_128_comp_kpadding_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD_ABScaleSplitK<Row,
+                                                                  Row,
+                                                                  Tuple<>,
+                                                                  Row,
+                                                                  F8,
+                                                                  F32,
+                                                                  F8,
+                                                                  F32,
+                                                                  Tuple<>,
+                                                                  BF16,
+                                                                  1,
+                                                                  128,
+                                                                  128,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances);
+
+void add_device_gemm_ab_scale_wmma_f8_f8_bf16_mk_kn_mn_1_128_128_mem_v1_default_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD_ABScaleSplitK<Row,
+                                                                  Row,
+                                                                  Tuple<>,
+                                                                  Row,
+                                                                  F8,
+                                                                  F32,
+                                                                  F8,
+                                                                  F32,
+                                                                  Tuple<>,
+                                                                  BF16,
+                                                                  1,
+                                                                  128,
+                                                                  128,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances);
+
+void add_device_gemm_ab_scale_wmma_f8_f8_bf16_mk_kn_mn_1_128_128_mem_v1_kpadding_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD_ABScaleSplitK<Row,
+                                                                  Row,
+                                                                  Tuple<>,
+                                                                  Row,
+                                                                  F8,
+                                                                  F32,
+                                                                  F8,
+                                                                  F32,
+                                                                  Tuple<>,
+                                                                  BF16,
+                                                                  1,
+                                                                  128,
+                                                                  128,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances);
+
+// Col, Row
+void add_device_gemm_ab_scale_wmma_f8_f8_bf16_km_kn_mn_1_128_128_comp_default_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD_ABScaleSplitK<Col,
+                                                                  Row,
+                                                                  Tuple<>,
+                                                                  Row,
+                                                                  F8,
+                                                                  F32,
+                                                                  F8,
+                                                                  F32,
+                                                                  Tuple<>,
+                                                                  BF16,
+                                                                  1,
+                                                                  128,
+                                                                  128,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances);
+
+void add_device_gemm_ab_scale_wmma_f8_f8_bf16_km_kn_mn_1_128_128_comp_kpadding_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD_ABScaleSplitK<Col,
+                                                                  Row,
+                                                                  Tuple<>,
+                                                                  Row,
+                                                                  F8,
+                                                                  F32,
+                                                                  F8,
+                                                                  F32,
+                                                                  Tuple<>,
+                                                                  BF16,
+                                                                  1,
+                                                                  128,
+                                                                  128,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances);
+
+void add_device_gemm_ab_scale_wmma_f8_f8_bf16_km_kn_mn_1_128_128_mem_v1_default_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD_ABScaleSplitK<Col,
+                                                                  Row,
+                                                                  Tuple<>,
+                                                                  Row,
+                                                                  F8,
+                                                                  F32,
+                                                                  F8,
+                                                                  F32,
+                                                                  Tuple<>,
+                                                                  BF16,
+                                                                  1,
+                                                                  128,
+                                                                  128,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances);
+
+void add_device_gemm_ab_scale_wmma_f8_f8_bf16_km_kn_mn_1_128_128_mem_v1_kpadding_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD_ABScaleSplitK<Col,
+                                                                  Row,
+                                                                  Tuple<>,
+                                                                  Row,
+                                                                  F8,
+                                                                  F32,
+                                                                  F8,
+                                                                  F32,
+                                                                  Tuple<>,
+                                                                  BF16,
+                                                                  1,
+                                                                  128,
+                                                                  128,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances);
+#endif
+#ifdef CK_USE_XDL
 // Row, Col
 void add_device_gemm_ab_scale_xdl_f8_f8_bf16_mk_nk_mn_1_128_128_comp_default_instances(
     std::vector<std::unique_ptr<DeviceGemmMultipleD_ABScale<Row,
@@ -236,6 +460,7 @@ void add_device_gemm_ab_scale_xdl_f8_f8_bf16_km_kn_mn_1_128_128_mem_v1_kpadding_
                                                             PassThrough,
                                                             PassThrough>>>& instances);
 #endif
+#endif
 
 template <typename A0DataType,
           typename A1DataType,
@@ -245,23 +470,124 @@ template <typename A0DataType,
           typename ALayout,
           typename BLayout,
           typename CLayout>
-struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGemmMultipleD_ABScale<
-    ALayout,
-    BLayout,
-    Tuple<>,
-    CLayout,
-    A0DataType,
-    A1DataType,
-    B0DataType,
-    B1DataType,
-    Tuple<>,
-    CDataType,
-    1,
-    128,
-    128,
-    ck::tensor_operation::element_wise::PassThrough,
-    ck::tensor_operation::element_wise::PassThrough,
-    ck::tensor_operation::element_wise::PassThrough>>
+struct DeviceOperationInstanceFactory<
+    ck::tensor_operation::device::DeviceGemmMultipleD_ABScaleSplitK<ALayout,
+                                                                    BLayout,
+                                                                    Tuple<>,
+                                                                    CLayout,
+                                                                    A0DataType,
+                                                                    A1DataType,
+                                                                    B0DataType,
+                                                                    B1DataType,
+                                                                    Tuple<>,
+                                                                    CDataType,
+                                                                    1,
+                                                                    128,
+                                                                    128,
+                                                                    PassThrough,
+                                                                    PassThrough,
+                                                                    PassThrough>>
+{
+    using DeviceOp = DeviceGemmMultipleD_ABScaleSplitK<ALayout,
+                                                       BLayout,
+                                                       Tuple<>,
+                                                       CLayout,
+                                                       A0DataType,
+                                                       A1DataType,
+                                                       B0DataType,
+                                                       B1DataType,
+                                                       Tuple<>,
+                                                       CDataType,
+                                                       1,
+                                                       128,
+                                                       128,
+                                                       PassThrough,
+                                                       PassThrough,
+                                                       PassThrough>;
+
+    static auto GetInstances()
+    {
+        std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
+
+#if(defined(CK_ENABLE_BF16) || defined(CK_ENABLE_FP8))
+#ifdef CK_USE_XDL
+        // No XDL instances for DeviceGemmMultipleABDSplitK with Add at the moment
+#endif
+#ifdef CK_USE_WMMA_FP8
+        if constexpr(is_same_v<A0DataType, f8_t> && is_same_v<B0DataType, f8_t> &&
+                     is_same_v<CDataType, bhalf_t>)
+        {
+            if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Col> &&
+                         is_same_v<CLayout, Row>)
+            {
+                add_device_gemm_ab_scale_wmma_f8_f8_bf16_mk_nk_mn_1_128_128_comp_default_instances(
+                    op_ptrs);
+                add_device_gemm_ab_scale_wmma_f8_f8_bf16_mk_nk_mn_1_128_128_comp_kpadding_instances(
+                    op_ptrs);
+
+                add_device_gemm_ab_scale_wmma_f8_f8_bf16_mk_nk_mn_1_128_128_mem_v1_default_instances(
+                    op_ptrs);
+                add_device_gemm_ab_scale_wmma_f8_f8_bf16_mk_nk_mn_1_128_128_mem_v1_kpadding_instances(
+                    op_ptrs);
+            }
+            else if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Row> &&
+                              is_same_v<CLayout, Row>)
+            {
+                add_device_gemm_ab_scale_wmma_f8_f8_bf16_mk_kn_mn_1_128_128_comp_default_instances(
+                    op_ptrs);
+                add_device_gemm_ab_scale_wmma_f8_f8_bf16_mk_kn_mn_1_128_128_comp_kpadding_instances(
+                    op_ptrs);
+
+                add_device_gemm_ab_scale_wmma_f8_f8_bf16_mk_kn_mn_1_128_128_mem_v1_default_instances(
+                    op_ptrs);
+                add_device_gemm_ab_scale_wmma_f8_f8_bf16_mk_kn_mn_1_128_128_mem_v1_kpadding_instances(
+                    op_ptrs);
+            }
+            else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Row> &&
+                              is_same_v<CLayout, Row>)
+            {
+                add_device_gemm_ab_scale_wmma_f8_f8_bf16_km_kn_mn_1_128_128_comp_default_instances(
+                    op_ptrs);
+                add_device_gemm_ab_scale_wmma_f8_f8_bf16_km_kn_mn_1_128_128_comp_kpadding_instances(
+                    op_ptrs);
+
+                add_device_gemm_ab_scale_wmma_f8_f8_bf16_km_kn_mn_1_128_128_mem_v1_default_instances(
+                    op_ptrs);
+                add_device_gemm_ab_scale_wmma_f8_f8_bf16_km_kn_mn_1_128_128_mem_v1_kpadding_instances(
+                    op_ptrs);
+            }
+        }
+#endif
+#endif
+        return op_ptrs;
+    }
+};
+
+template <typename A0DataType,
+          typename A1DataType,
+          typename B0DataType,
+          typename B1DataType,
+          typename CDataType,
+          typename ALayout,
+          typename BLayout,
+          typename CLayout>
+struct DeviceOperationInstanceFactory<
+    ck::tensor_operation::device::DeviceGemmMultipleD_ABScale<ALayout,
+                                                              BLayout,
+                                                              Tuple<>,
+                                                              CLayout,
+                                                              A0DataType,
+                                                              A1DataType,
+                                                              B0DataType,
+                                                              B1DataType,
+                                                              Tuple<>,
+                                                              CDataType,
+                                                              1,
+                                                              128,
+                                                              128,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>
 {
     using DeviceOp = DeviceGemmMultipleD_ABScale<ALayout,
                                                  BLayout,
@@ -276,15 +602,16 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGemmMu
                                                  1,
                                                  128,
                                                  128,
-                                                 ck::tensor_operation::element_wise::PassThrough,
-                                                 ck::tensor_operation::element_wise::PassThrough,
-                                                 ck::tensor_operation::element_wise::PassThrough>;
+                                                 PassThrough,
+                                                 PassThrough,
+                                                 PassThrough>;
 
     static auto GetInstances()
     {
         std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
 
 #if(defined(CK_ENABLE_BF16) || defined(CK_ENABLE_FP8))
+#ifdef CK_USE_XDL
         if constexpr(is_same_v<A0DataType, f8_t> && is_same_v<B0DataType, f8_t> &&
                      is_same_v<CDataType, bhalf_t>)
         {
@@ -328,6 +655,33 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGemmMu
                     op_ptrs);
             }
         }
+#endif
+
+#ifdef CK_USE_WMMA_FP8
+        using Wrapper = DeviceGemmMultipleD_ABScaleSplitKWrapper<ALayout,
+                                                                 BLayout,
+                                                                 Tuple<>,
+                                                                 CLayout,
+                                                                 A0DataType,
+                                                                 A1DataType,
+                                                                 B0DataType,
+                                                                 B1DataType,
+                                                                 Tuple<>,
+                                                                 CDataType,
+                                                                 1,
+                                                                 128,
+                                                                 128,
+                                                                 PassThrough,
+                                                                 PassThrough,
+                                                                 PassThrough>;
+
+        auto new_op_ptrs =
+            DeviceOperationInstanceFactory<typename Wrapper::DeviceOp>::GetInstances();
+        for(auto& op_ptr : new_op_ptrs)
+        {
+            op_ptrs.emplace_back(std::make_unique<Wrapper>(std::move(op_ptr)));
+        }
+#endif // CK_USE_WMMA_FP8
 #endif
         return op_ptrs;
     }

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

@@ -17,6 +17,47 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 #if(defined(CK_ENABLE_BF16) || defined(CK_ENABLE_FP8))
+#if(defined(CK_USE_WMMA) && defined(CK_USE_WMMA_FP8))
+void add_device_gemm_blockscale_wp_wmma_f8_f8_bf16_mk_nk_mn_1_128_128_comp_default_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD_BlockScale_BPreshuffleSplitK<Row,
+                                                                                 Col,
+                                                                                 Tuple<>,
+                                                                                 Row,
+                                                                                 F8,
+                                                                                 F32,
+                                                                                 F8,
+                                                                                 F32,
+                                                                                 Tuple<>,
+                                                                                 BF16,
+                                                                                 1,
+                                                                                 128,
+                                                                                 128,
+                                                                                 PassThrough,
+                                                                                 PassThrough,
+                                                                                 PassThrough>>>&
+        instances);
+
+void add_device_gemm_blockscale_wp_wmma_f8_f8_bf16_mk_nk_mn_1_128_128_mem_default_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD_BlockScale_BPreshuffleSplitK<Row,
+                                                                                 Col,
+                                                                                 Tuple<>,
+                                                                                 Row,
+                                                                                 F8,
+                                                                                 F32,
+                                                                                 F8,
+                                                                                 F32,
+                                                                                 Tuple<>,
+                                                                                 BF16,
+                                                                                 1,
+                                                                                 128,
+                                                                                 128,
+                                                                                 PassThrough,
+                                                                                 PassThrough,
+                                                                                 PassThrough>>>&
+        instances);
+#endif // CK_USE_WMMA && CK_USE_WMMA_FP8
+
+#ifdef CK_USE_XDL
 void add_device_gemm_blockscale_wp_xdl_f8_f8_bf16_mk_nk_mn_1_128_128_comp_default_instances(
     std::vector<std::unique_ptr<DeviceGemmMultipleD_BlockScale_BPreshuffle<Row,
                                                                            Col,
@@ -93,6 +134,82 @@ void add_device_gemm_blockscale_wp_xdl_f8_f8_bf16_mk_nk_mn_1_128_128_mem_v1_kpad
                                                                            PassThrough>>>&
         instances);
 #endif
+#endif
+
+template <typename A0DataType,
+          typename A1DataType,
+          typename B0DataType,
+          typename B1DataType,
+          typename CDataType,
+          typename ALayout,
+          typename BLayout,
+          typename CLayout>
+struct DeviceOperationInstanceFactory<
+    ck::tensor_operation::device::DeviceGemmMultipleD_BlockScale_BPreshuffleSplitK<
+        ALayout,
+        BLayout,
+        Tuple<>,
+        CLayout,
+        A0DataType,
+        A1DataType,
+        B0DataType,
+        B1DataType,
+        Tuple<>,
+        CDataType,
+        1,
+        128,
+        128,
+        ck::tensor_operation::element_wise::PassThrough,
+        ck::tensor_operation::element_wise::PassThrough,
+        ck::tensor_operation::element_wise::PassThrough>>
+{
+    using DeviceOp = DeviceGemmMultipleD_BlockScale_BPreshuffleSplitK<
+        ALayout,
+        BLayout,
+        Tuple<>,
+        CLayout,
+        A0DataType,
+        A1DataType,
+        B0DataType,
+        B1DataType,
+        Tuple<>,
+        CDataType,
+        1,
+        128,
+        128,
+        ck::tensor_operation::element_wise::PassThrough,
+        ck::tensor_operation::element_wise::PassThrough,
+        ck::tensor_operation::element_wise::PassThrough>;
+
+    static auto GetInstances()
+    {
+        std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
+
+#if defined(CK_USE_XDL)
+        // No XDL instances for DeviceGemmMultipleD_BlockScale_BPreshuffleSplitK at the moment
+#endif // CK_USE_XDL
+
+#if(defined(CK_USE_WMMA) && defined(CK_USE_WMMA_FP8))
+#if(defined(CK_ENABLE_BF16) || defined(CK_ENABLE_FP8))
+        if constexpr(is_same_v<A0DataType, f8_t> && is_same_v<B0DataType, f8_t> &&
+                     is_same_v<CDataType, bhalf_t>)
+        {
+            if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Col> &&
+                         is_same_v<CLayout, Row>)
+            {
+                add_device_gemm_blockscale_wp_wmma_f8_f8_bf16_mk_nk_mn_1_128_128_comp_default_instances(
+                    op_ptrs);
+
+                add_device_gemm_blockscale_wp_wmma_f8_f8_bf16_mk_nk_mn_1_128_128_mem_default_instances(
+                    op_ptrs);
+            }
+        }
+#endif
+#endif // CK_USE_WMMA && CK_USE_WMMA_FP8
+
+        return op_ptrs;
+    }
+};
 
 template <typename A0DataType,
           typename A1DataType,
@@ -143,6 +260,7 @@ struct DeviceOperationInstanceFactory<
     {
         std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
 
+#ifdef CK_USE_XDL
 #if(defined(CK_ENABLE_BF16) || defined(CK_ENABLE_FP8))
         if constexpr(is_same_v<A0DataType, f8_t> && is_same_v<B0DataType, f8_t> &&
                      is_same_v<CDataType, bhalf_t>)
@@ -162,6 +280,35 @@ struct DeviceOperationInstanceFactory<
             }
         }
 #endif
+#endif
+
+#if defined(CK_USE_WMMA)
+        // Reuse DeviceGemmMultipleD_BlockScale_BPreshuffleSplitK instances
+        using Wrapper = DeviceGemmMultipleD_BlockScale_BPreshuffleWrapper<
+            ALayout,
+            BLayout,
+            Tuple<>,
+            CLayout,
+            A0DataType,
+            A1DataType,
+            B0DataType,
+            B1DataType,
+            Tuple<>,
+            CDataType,
+            1,
+            128,
+            128,
+            ck::tensor_operation::element_wise::PassThrough,
+            ck::tensor_operation::element_wise::PassThrough,
+            ck::tensor_operation::element_wise::PassThrough>;
+        auto new_op_ptrs =
+            DeviceOperationInstanceFactory<typename Wrapper::DeviceOp>::GetInstances();
+        for(auto& op_ptr : new_op_ptrs)
+        {
+            op_ptrs.emplace_back(std::make_unique<Wrapper>(std::move(op_ptr)));
+        }
+#endif // CK_USE_WMMA
+
         return op_ptrs;
     }
 };

library/src/tensor_operation_instance/gpu/CMakeLists.txt

@@ -103,6 +103,16 @@ function(add_instance_library INSTANCE_NAME)
             message(DEBUG "removing gemm_universal_preshuffle_f8 instance ${source} ")
             list(REMOVE_ITEM ARGN "${source}")
         endif()
+        # Do not build gemm_ab_scale_f8 for any targets except gfx94, gfx95 and gfx12
+        if(NOT (INST_TARGETS MATCHES "gfx942" OR INST_TARGETS MATCHES "gfx950" OR INST_TARGETS MATCHES "gfx12") AND (source_name MATCHES "gemm_ab_scale") AND (source_name MATCHES "_f8_f8_"))
+            message(DEBUG "removing gemm_ab_scale_f8 instance ${source} ")
+            list(REMOVE_ITEM ARGN "${source}")
+        endif()
+        # Do not build gemm_blockscale_wp_f8 for any targets except gfx94, gfx95 and gfx12
+        if(NOT (INST_TARGETS MATCHES "gfx942" OR INST_TARGETS MATCHES "gfx950" OR INST_TARGETS MATCHES "gfx12") AND (source_name MATCHES "gemm_blockscale_wp") AND (source_name MATCHES "_f8_f8_"))
+            message(DEBUG "removing gemm_blockscale_wp_f8 instance ${source} ")
+            list(REMOVE_ITEM ARGN "${source}")
+        endif()
         # Only build tf32 instances for gfx942 & gfx950
         if(source_name MATCHES "_tf32_")
             if(NOT ((INST_TARGETS MATCHES "gfx942|gfx950") AND CK_ENABLE_TF32))
@@ -300,7 +310,7 @@ FOREACH(subdir_path ${dir_list})
             message(DEBUG "Found gemm_multiply_multiply instances, but gfx94/gfx95/gfx11/gfx12 not on the target list. Skipping. ${cmake_instance}")
             set(add_inst 0)
         endif()
-        if(("${cmake_instance}" MATCHES "gemm_universal_preshuffle|gemm_blockscale" AND "${cmake_instance}" MATCHES "_f8_" ) AND (NOT INST_TARGETS MATCHES "gfx94|gfx95|gfx12") AND (NOT CK_USE_FP8_ON_UNSUPPORTED_ARCH))
+        if(("${cmake_instance}" MATCHES "gemm_universal_preshuffle|gemm_blockscale|gemm_ab_scale" AND "${cmake_instance}" MATCHES "_f8_" ) AND (NOT INST_TARGETS MATCHES "gfx94|gfx95|gfx12") AND (NOT CK_USE_FP8_ON_UNSUPPORTED_ARCH))
             message(DEBUG "Found gemm_f8 instances, but gfx94/gfx95 not on the target list. Skipping.")
             set(add_inst 0)
         endif()

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

@@ -1,21 +1,38 @@
 # Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
 # SPDX-License-Identifier: MIT
 
-# ONLY XDL_KERNELS
+# ONLY XDL_AND_WMMA_KERNELS
 set(GEMM_AB_SCALE_INSTANCES)
 
 list(APPEND GEMM_AB_SCALE_INSTANCES 
         # Row, Col
+        device_gemm_ab_scale_wmma_f8_f8_bf16/device_gemm_ab_scale_wmma_f8_f8_bf16_mk_nk_mn_128_128_128_comp_default_instance.cpp
+        device_gemm_ab_scale_wmma_f8_f8_bf16/device_gemm_ab_scale_wmma_f8_f8_bf16_mk_nk_mn_128_128_128_comp_kpadding_instance.cpp
+        device_gemm_ab_scale_wmma_f8_f8_bf16/device_gemm_ab_scale_wmma_f8_f8_bf16_mk_nk_mn_128_128_128_mem_v1_default_instance.cpp
+        device_gemm_ab_scale_wmma_f8_f8_bf16/device_gemm_ab_scale_wmma_f8_f8_bf16_mk_nk_mn_128_128_128_mem_v1_kpadding_instance.cpp
+
         device_gemm_ab_scale_xdl_f8_f8_bf16/device_gemm_ab_scale_xdl_f8_f8_bf16_mk_nk_mn_128_128_128_comp_default_instance.cpp
         device_gemm_ab_scale_xdl_f8_f8_bf16/device_gemm_ab_scale_xdl_f8_f8_bf16_mk_nk_mn_128_128_128_comp_kpadding_instance.cpp
         device_gemm_ab_scale_xdl_f8_f8_bf16/device_gemm_ab_scale_xdl_f8_f8_bf16_mk_nk_mn_128_128_128_mem_v1_default_instance.cpp
         device_gemm_ab_scale_xdl_f8_f8_bf16/device_gemm_ab_scale_xdl_f8_f8_bf16_mk_nk_mn_128_128_128_mem_v1_kpadding_instance.cpp
+
         # Row, Row
+        device_gemm_ab_scale_wmma_f8_f8_bf16/device_gemm_ab_scale_wmma_f8_f8_bf16_mk_kn_mn_128_128_128_comp_default_instance.cpp
+        device_gemm_ab_scale_wmma_f8_f8_bf16/device_gemm_ab_scale_wmma_f8_f8_bf16_mk_kn_mn_128_128_128_comp_kpadding_instance.cpp
+        device_gemm_ab_scale_wmma_f8_f8_bf16/device_gemm_ab_scale_wmma_f8_f8_bf16_mk_kn_mn_128_128_128_mem_v1_default_instance.cpp
+        device_gemm_ab_scale_wmma_f8_f8_bf16/device_gemm_ab_scale_wmma_f8_f8_bf16_mk_kn_mn_128_128_128_mem_v1_kpadding_instance.cpp
+
         device_gemm_ab_scale_xdl_f8_f8_bf16/device_gemm_ab_scale_xdl_f8_f8_bf16_mk_kn_mn_128_128_128_comp_default_instance.cpp
         device_gemm_ab_scale_xdl_f8_f8_bf16/device_gemm_ab_scale_xdl_f8_f8_bf16_mk_kn_mn_128_128_128_comp_kpadding_instance.cpp
         device_gemm_ab_scale_xdl_f8_f8_bf16/device_gemm_ab_scale_xdl_f8_f8_bf16_mk_kn_mn_128_128_128_mem_v1_default_instance.cpp
         device_gemm_ab_scale_xdl_f8_f8_bf16/device_gemm_ab_scale_xdl_f8_f8_bf16_mk_kn_mn_128_128_128_mem_v1_kpadding_instance.cpp
+
         # Col, Row
+        device_gemm_ab_scale_wmma_f8_f8_bf16/device_gemm_ab_scale_wmma_f8_f8_bf16_km_kn_mn_128_128_128_comp_default_instance.cpp
+        device_gemm_ab_scale_wmma_f8_f8_bf16/device_gemm_ab_scale_wmma_f8_f8_bf16_km_kn_mn_128_128_128_comp_kpadding_instance.cpp
+        device_gemm_ab_scale_wmma_f8_f8_bf16/device_gemm_ab_scale_wmma_f8_f8_bf16_km_kn_mn_128_128_128_mem_v1_default_instance.cpp
+        device_gemm_ab_scale_wmma_f8_f8_bf16/device_gemm_ab_scale_wmma_f8_f8_bf16_km_kn_mn_128_128_128_mem_v1_kpadding_instance.cpp
+
         device_gemm_ab_scale_xdl_f8_f8_bf16/device_gemm_ab_scale_xdl_f8_f8_bf16_km_kn_mn_128_128_128_comp_default_instance.cpp
         device_gemm_ab_scale_xdl_f8_f8_bf16/device_gemm_ab_scale_xdl_f8_f8_bf16_km_kn_mn_128_128_128_comp_kpadding_instance.cpp
         device_gemm_ab_scale_xdl_f8_f8_bf16/device_gemm_ab_scale_xdl_f8_f8_bf16_km_kn_mn_128_128_128_mem_v1_default_instance.cpp
@@ -27,11 +44,13 @@ set_source_files_properties(device_gemm_ab_scale_xdl_f8_f8_bf16/device_gemm_ab_s
 set_source_files_properties(device_gemm_ab_scale_xdl_f8_f8_bf16/device_gemm_ab_scale_xdl_f8_f8_bf16_mk_nk_mn_128_128_128_comp_kpadding_instance.cpp PROPERTIES COMPILE_OPTIONS ";-mllvm;-greedy-reverse-local-assignment=1")
 set_source_files_properties(device_gemm_ab_scale_xdl_f8_f8_bf16/device_gemm_ab_scale_xdl_f8_f8_bf16_mk_nk_mn_128_128_128_mem_v1_default_instance.cpp PROPERTIES COMPILE_OPTIONS ";-mllvm;-greedy-reverse-local-assignment=1")
 set_source_files_properties(device_gemm_ab_scale_xdl_f8_f8_bf16/device_gemm_ab_scale_xdl_f8_f8_bf16_mk_nk_mn_128_128_128_mem_v1_kpadding_instance.cpp PROPERTIES COMPILE_OPTIONS ";-mllvm;-greedy-reverse-local-assignment=1")
+
 # Row, Row
 set_source_files_properties(device_gemm_ab_scale_xdl_f8_f8_bf16/device_gemm_ab_scale_xdl_f8_f8_bf16_mk_kn_mn_128_128_128_comp_default_instance.cpp PROPERTIES COMPILE_OPTIONS ";-mllvm;-greedy-reverse-local-assignment=1")
 set_source_files_properties(device_gemm_ab_scale_xdl_f8_f8_bf16/device_gemm_ab_scale_xdl_f8_f8_bf16_mk_kn_mn_128_128_128_comp_kpadding_instance.cpp PROPERTIES COMPILE_OPTIONS ";-mllvm;-greedy-reverse-local-assignment=1")
 set_source_files_properties(device_gemm_ab_scale_xdl_f8_f8_bf16/device_gemm_ab_scale_xdl_f8_f8_bf16_mk_kn_mn_128_128_128_mem_v1_default_instance.cpp PROPERTIES COMPILE_OPTIONS ";-mllvm;-greedy-reverse-local-assignment=1")
 set_source_files_properties(device_gemm_ab_scale_xdl_f8_f8_bf16/device_gemm_ab_scale_xdl_f8_f8_bf16_mk_kn_mn_128_128_128_mem_v1_kpadding_instance.cpp PROPERTIES COMPILE_OPTIONS ";-mllvm;-greedy-reverse-local-assignment=1")
+
 # Col, Row
 set_source_files_properties(device_gemm_ab_scale_xdl_f8_f8_bf16/device_gemm_ab_scale_xdl_f8_f8_bf16_km_kn_mn_128_128_128_comp_default_instance.cpp PROPERTIES COMPILE_OPTIONS ";-mllvm;-greedy-reverse-local-assignment=1")
 set_source_files_properties(device_gemm_ab_scale_xdl_f8_f8_bf16/device_gemm_ab_scale_xdl_f8_f8_bf16_km_kn_mn_128_128_128_comp_kpadding_instance.cpp PROPERTIES COMPILE_OPTIONS ";-mllvm;-greedy-reverse-local-assignment=1")

library/src/tensor_operation_instance/gpu/gemm_ab_scale/device_gemm_ab_scale_wmma_f8_f8_bf16/device_gemm_ab_scale_wmma_f8_f8_bf16_km_kn_mn_128_128_128.hpp

@@ -0,0 +1,79 @@
+// Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
+// SPDX-License-Identifier: MIT
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_wmma_cshuffle_v3_ab_scale.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F8   = f8_t;
+using BF16 = bhalf_t;
+using F32  = float;
+
+using Row = tensor_layout::gemm::RowMajor;
+using Col = tensor_layout::gemm::ColumnMajor;
+
+template <index_t... Is>
+using S = Sequence<Is...>;
+
+using PassThrough = element_wise::PassThrough;
+using PassThrough = element_wise::PassThrough;
+
+static constexpr auto GemmDefault    = GemmSpecialization::Default;
+static constexpr auto GemmKPadding   = GemmSpecialization::KPadding;
+static constexpr auto GemmMNPadding  = GemmSpecialization::MNPadding;
+static constexpr auto GemmMNKPadding = GemmSpecialization::MNKPadding;
+
+static constexpr auto Intrawave = BlockGemmPipelineScheduler::Intrawave;
+
+template <GemmSpecialization GemmSpec>
+using device_gemm_ab_scale_wmma_f8_f8_bf16_km_kn_mn_1_128_128_comp_instances = std::tuple<
+    // clang-format off
+        //#######################################| ALayout| BLayout|       DsLayout| ELayout|      AData|      BData|     DsData| EData| AccData| Cshuffle|           A|           B|           C|           GEMM| Block| Scale| Scale| Scale|  MPer|  NPer|  KPer| AK1| BK1|MPer| NPer| MRepeat| NRepeat|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|   BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle   | CShuffle   |     CBlockTransferClusterLengths  |  CShuffleBlockTransfer|                         Block-wiseGemm|               Block-wiseGemm|
+        //#######################################|        |        |               |        |       Type|       Type|       Type|  Type|    Type|     Type| Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block| Block| Block| Block|    |    |Wmma| Wmma|        |        |   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|    ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MRepeat    | NRepeat    | _MBlock_MPerBlock_NBlock_NPerBlock|        ScalarPerVector|                               Pipeline|                     Pipeline|
+        //#######################################|        |        |               |        |           |           |           |      |        |         |   Operation|   Operation|   Operation|               |      |     M|     N|     K|      |      |      |    |    |    |     |        |        | Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          |  Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          | PerShuffle | PerShuffle |                                   |                       |                              Scheduler|                      Version|
+        //#######################################|        |        |               |        |           |           |           |      |        |         |            |            |            |               |      |      |      |      |      |      |      |    |    |    |     |        |        |                |               |               |               |               |               |          |                 |               |               |              |               |               |          |            |            |                                   |                       |                                       |                             |
+        DeviceGemmMultiD_ABScale_Wmma_CShuffle_V3<     Col,     Row,        Tuple<>,     Row,    F8, F32,    F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough, PassThrough,       GemmSpec,   256,     1,   128,   128,   128,   128,   128,  16,  16,  16,   16,       4,       2,    S< 8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,             16,         0,     S< 8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,             16,         0,           1,           1,                     S<1, 32, 1, 8>,                   S<8>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1, F8>,
+        DeviceGemmMultiD_ABScale_Wmma_CShuffle_V3<     Col,     Row,        Tuple<>,     Row,    F8, F32,    F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough, PassThrough,       GemmSpec,   256,     1,   128,   128,   128,    64,   128,  16,  16,  16,   16,       2,       2,    S< 8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,             16,         0,     S< 8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,             16,         0,           1,           1,                     S<1, 64, 1, 4>,                   S<8>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1, F8>,
+        DeviceGemmMultiD_ABScale_Wmma_CShuffle_V3<     Col,     Row,        Tuple<>,     Row,    F8, F32,    F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough, PassThrough,       GemmSpec,   256,     1,   128,   128,    64,   128,   128,  16,  16,  16,   16,       2,       2,    S< 8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,             16,         0,     S< 8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,             16,         0,           1,           1,                     S<1, 32, 1, 8>,                   S<8>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1, F8>,
+        DeviceGemmMultiD_ABScale_Wmma_CShuffle_V3<     Col,     Row,        Tuple<>,     Row,    F8, F32,    F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough, PassThrough,       GemmSpec,   256,     1,   128,   128,    64,    64,   128,  16,  16,  16,   16,       2,       1,    S< 8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,             16,         0,     S< 8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,             16,         0,           1,           1,                     S<1, 32, 1, 8>,                   S<8>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1, F8>,
+
+        DeviceGemmMultiD_ABScale_Wmma_CShuffle_V3<     Col,     Row,        Tuple<>,     Row,    F8, F32,    F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough, PassThrough,       GemmSpec,   256,     1,   128,   128,   128,   128,   128,  16,  16,  16,   16,       4,       2,    S< 8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,             16,         0,     S< 8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,             16,         0,           1,           1,                     S<1, 32, 1, 8>,                   S<8>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3, F8>,
+        DeviceGemmMultiD_ABScale_Wmma_CShuffle_V3<     Col,     Row,        Tuple<>,     Row,    F8, F32,    F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough, PassThrough,       GemmSpec,   256,     1,   128,   128,   128,    64,   128,  16,  16,  16,   16,       2,       2,    S< 8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,             16,         0,     S< 8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,             16,         0,           1,           1,                     S<1, 64, 1, 4>,                   S<8>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3, F8>,
+        DeviceGemmMultiD_ABScale_Wmma_CShuffle_V3<     Col,     Row,        Tuple<>,     Row,    F8, F32,    F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough, PassThrough,       GemmSpec,   256,     1,   128,   128,    64,   128,   128,  16,  16,  16,   16,       2,       2,    S< 8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,             16,         0,     S< 8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,             16,         0,           1,           1,                     S<1, 32, 1, 8>,                   S<8>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3, F8>,
+        DeviceGemmMultiD_ABScale_Wmma_CShuffle_V3<     Col,     Row,        Tuple<>,     Row,    F8, F32,    F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough, PassThrough,       GemmSpec,   256,     1,   128,   128,    64,    64,   128,  16,  16,  16,   16,       2,       1,    S< 8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,             16,         0,     S< 8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,             16,         0,           1,           1,                     S<1, 32, 1, 8>,                   S<8>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3, F8>
+    // clang-format on
+    >;
+
+template <BlockGemmPipelineScheduler BlkGemmPipeSched, GemmSpecialization GemmSpec>
+using device_gemm_ab_scale_wmma_f8_f8_bf16_km_kn_mn_1_128_128_mem_instances = std::tuple<
+    // clang-format off
+        //#######################################| ALayout| BLayout|       DsLayout| ELayout|      AData|      BData|     DsData| EData| AccData| Cshuffle|           A|           B|           C|           GEMM| Block| Scale| Scale| Scale|  MPer|  NPer|  KPer| AK1| BK1|MPer| NPer| MRepeat| NRepeat|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|   BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle   | CShuffle   |     CBlockTransferClusterLengths  |  CShuffleBlockTransfer|                         Block-wiseGemm|               Block-wiseGemm|
+        //#######################################|        |        |               |        |       Type|       Type|       Type|  Type|    Type|     Type| Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block| Block| Block| Block|    |    |Wmma| Wmma|        |        |   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|    ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MRepeat    | NRepeat    | _MBlock_MPerBlock_NBlock_NPerBlock|        ScalarPerVector|                               Pipeline|                     Pipeline|
+        //#######################################|        |        |               |        |           |           |           |      |        |         |   Operation|   Operation|   Operation|               |      |     M|     N|     K|      |      |      |    |    |    |     |        |        | Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          |  Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          | PerShuffle | PerShuffle |                                   |                       |                              Scheduler|                      Version|
+        //#######################################|        |        |               |        |           |           |           |      |        |         |            |            |            |               |      |      |      |      |      |      |      |    |    |    |     |        |        |                |               |               |               |               |               |          |                 |               |               |              |               |               |          |            |            |                                   |                       |                                       |                             |
+        DeviceGemmMultiD_ABScale_Wmma_CShuffle_V3<     Col,     Row,        Tuple<>,     Row,    F8, F32,    F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough, PassThrough,       GemmSpec,   256,     1,   128,   128,    16,   256,   128,   8,  16,  16,   16,       1,       2,    S<16, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,         0,     S< 8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,             16,         0,           1,           1,                    S<1, 16, 1, 16>,                   S<8>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1, F8>,
+        DeviceGemmMultiD_ABScale_Wmma_CShuffle_V3<     Col,     Row,        Tuple<>,     Row,    F8, F32,    F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough, PassThrough,       GemmSpec,   256,     1,   128,   128,    16,   128,   128,   8,  16,  16,   16,       1,       1,    S<16, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,         0,     S< 8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,             16,         0,           1,           1,                    S<1, 16, 1, 16>,                   S<8>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1, F8>,
+        DeviceGemmMultiD_ABScale_Wmma_CShuffle_V3<     Col,     Row,        Tuple<>,     Row,    F8, F32,    F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough, PassThrough,       GemmSpec,   128,     1,   128,   128,    16,    64,   128,   8,  16,  16,   16,       1,       1,    S<16,  8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,         0,     S< 8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,             16,         0,           1,           1,                    S<1, 16, 1,  8>,                   S<8>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1, F8>,
+
+        DeviceGemmMultiD_ABScale_Wmma_CShuffle_V3<     Col,     Row,        Tuple<>,     Row,    F8, F32,    F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough, PassThrough,       GemmSpec,   256,     1,   128,   128,    32,   256,   128,  16,  16,  16,   16,       2,       2,    S< 8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,             16,         0,     S< 8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,             16,         0,           1,           1,                    S<1, 16, 1, 16>,                   S<8>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1, F8>,
+        DeviceGemmMultiD_ABScale_Wmma_CShuffle_V3<     Col,     Row,        Tuple<>,     Row,    F8, F32,    F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough, PassThrough,       GemmSpec,   256,     1,   128,   128,    32,   128,   128,  16,  16,  16,   16,       2,       1,    S< 8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,             16,         0,     S< 8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,             16,         0,           1,           1,                    S<1, 16, 1, 16>,                   S<8>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1, F8>,
+        DeviceGemmMultiD_ABScale_Wmma_CShuffle_V3<     Col,     Row,        Tuple<>,     Row,    F8, F32,    F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough, PassThrough,       GemmSpec,   256,     1,   128,   128,    32,    64,   128,  16,  16,  16,   16,       1,       1,    S< 8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,             16,         0,     S< 8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,             16,         0,           1,           1,                    S<1, 32, 1,  8>,                   S<8>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1, F8>,
+
+        DeviceGemmMultiD_ABScale_Wmma_CShuffle_V3<     Col,     Row,        Tuple<>,     Row,    F8, F32,    F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough, PassThrough,       GemmSpec,   256,     1,   128,   128,    64,   256,   128,  16,  16,  16,   16,       4,       2,    S< 8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,             16,         0,     S< 8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,             16,         0,           1,           1,                    S<1, 16, 1, 16>,                   S<8>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1, F8>,
+        DeviceGemmMultiD_ABScale_Wmma_CShuffle_V3<     Col,     Row,        Tuple<>,     Row,    F8, F32,    F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough, PassThrough,       GemmSpec,   256,     1,   128,   128,    64,   128,   128,  16,  16,  16,   16,       4,       1,    S< 8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,             16,         0,     S< 8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,             16,         0,           1,           1,                    S<1, 16, 1, 16>,                   S<8>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1, F8>,
+        DeviceGemmMultiD_ABScale_Wmma_CShuffle_V3<     Col,     Row,        Tuple<>,     Row,    F8, F32,    F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough, PassThrough,       GemmSpec,   256,     1,   128,   128,    64,    64,   128,  16,  16,  16,   16,       2,       1,    S< 8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,             16,         0,     S< 8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,             16,         0,           1,           1,                    S<1, 32, 1,  8>,                   S<8>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1, F8>
+    // clang-format on
+    >;
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/gemm_ab_scale/device_gemm_ab_scale_wmma_f8_f8_bf16/device_gemm_ab_scale_wmma_f8_f8_bf16_km_kn_mn_128_128_128_comp_default_instance.cpp

@@ -0,0 +1,37 @@
+// Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
+// SPDX-License-Identifier: MIT
+
+#include "device_gemm_ab_scale_wmma_f8_f8_bf16_km_kn_mn_128_128_128.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_gemm_ab_scale_wmma_f8_f8_bf16_km_kn_mn_1_128_128_comp_default_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD_ABScaleSplitK<Col,
+                                                                  Row,
+                                                                  Tuple<>,
+                                                                  Row,
+                                                                  F8,
+                                                                  F32,
+                                                                  F8,
+                                                                  F32,
+                                                                  Tuple<>,
+                                                                  BF16,
+                                                                  1,
+                                                                  128,
+                                                                  128,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_gemm_ab_scale_wmma_f8_f8_bf16_km_kn_mn_1_128_128_comp_instances<GemmDefault>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/gemm_ab_scale/device_gemm_ab_scale_wmma_f8_f8_bf16/device_gemm_ab_scale_wmma_f8_f8_bf16_km_kn_mn_128_128_128_comp_kpadding_instance.cpp

@@ -0,0 +1,37 @@
+// Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
+// SPDX-License-Identifier: MIT
+
+#include "device_gemm_ab_scale_wmma_f8_f8_bf16_km_kn_mn_128_128_128.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_gemm_ab_scale_wmma_f8_f8_bf16_km_kn_mn_1_128_128_comp_kpadding_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD_ABScaleSplitK<Col,
+                                                                  Row,
+                                                                  Tuple<>,
+                                                                  Row,
+                                                                  F8,
+                                                                  F32,
+                                                                  F8,
+                                                                  F32,
+                                                                  Tuple<>,
+                                                                  BF16,
+                                                                  1,
+                                                                  128,
+                                                                  128,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_gemm_ab_scale_wmma_f8_f8_bf16_km_kn_mn_1_128_128_comp_instances<GemmKPadding>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/gemm_ab_scale/device_gemm_ab_scale_wmma_f8_f8_bf16/device_gemm_ab_scale_wmma_f8_f8_bf16_km_kn_mn_128_128_128_mem_v1_default_instance.cpp

@@ -0,0 +1,38 @@
+// Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
+// SPDX-License-Identifier: MIT
+
+#include "device_gemm_ab_scale_wmma_f8_f8_bf16_km_kn_mn_128_128_128.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_gemm_ab_scale_wmma_f8_f8_bf16_km_kn_mn_1_128_128_mem_v1_default_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD_ABScaleSplitK<Col,
+                                                                  Row,
+                                                                  Tuple<>,
+                                                                  Row,
+                                                                  F8,
+                                                                  F32,
+                                                                  F8,
+                                                                  F32,
+                                                                  Tuple<>,
+                                                                  BF16,
+                                                                  1,
+                                                                  128,
+                                                                  128,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_gemm_ab_scale_wmma_f8_f8_bf16_km_kn_mn_1_128_128_mem_instances<Intrawave,
+                                                                              GemmDefault>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/gemm_ab_scale/device_gemm_ab_scale_wmma_f8_f8_bf16/device_gemm_ab_scale_wmma_f8_f8_bf16_km_kn_mn_128_128_128_mem_v1_kpadding_instance.cpp

@@ -0,0 +1,38 @@
+// Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
+// SPDX-License-Identifier: MIT
+
+#include "device_gemm_ab_scale_wmma_f8_f8_bf16_km_kn_mn_128_128_128.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_gemm_ab_scale_wmma_f8_f8_bf16_km_kn_mn_1_128_128_mem_v1_kpadding_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD_ABScaleSplitK<Col,
+                                                                  Row,
+                                                                  Tuple<>,
+                                                                  Row,
+                                                                  F8,
+                                                                  F32,
+                                                                  F8,
+                                                                  F32,
+                                                                  Tuple<>,
+                                                                  BF16,
+                                                                  1,
+                                                                  128,
+                                                                  128,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_gemm_ab_scale_wmma_f8_f8_bf16_km_kn_mn_1_128_128_mem_instances<Intrawave,
+                                                                              GemmKPadding>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/gemm_ab_scale/device_gemm_ab_scale_wmma_f8_f8_bf16/device_gemm_ab_scale_wmma_f8_f8_bf16_mk_kn_mn_128_128_128.hpp

@@ -0,0 +1,80 @@
+// Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
+// SPDX-License-Identifier: MIT
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_wmma_cshuffle_v3_ab_scale.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F8   = f8_t;
+using BF16 = bhalf_t;
+using F32  = float;
+
+using Row = tensor_layout::gemm::RowMajor;
+using Col = tensor_layout::gemm::ColumnMajor;
+
+template <index_t... Is>
+using S = Sequence<Is...>;
+
+using PassThrough = element_wise::PassThrough;
+using PassThrough = element_wise::PassThrough;
+
+static constexpr auto GemmDefault    = GemmSpecialization::Default;
+static constexpr auto GemmKPadding   = GemmSpecialization::KPadding;
+static constexpr auto GemmMNPadding  = GemmSpecialization::MNPadding;
+static constexpr auto GemmMNKPadding = GemmSpecialization::MNKPadding;
+
+static constexpr auto Intrawave = BlockGemmPipelineScheduler::Intrawave;
+
+template <GemmSpecialization GemmSpec>
+using device_gemm_ab_scale_wmma_f8_f8_bf16_mk_kn_mn_1_128_128_comp_instances = std::tuple<
+    // clang-format off
+        //#######################################| ALayout| BLayout|       DsLayout| ELayout|      AData|      BData|     DsData| EData| AccData| Cshuffle|           A|           B|           C|           GEMM| Block| Scale| Scale| Scale|  MPer|  NPer|  KPer| AK1| BK1|MPer| NPer| MRepeat| NRepeat|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|   BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle   | CShuffle   |     CBlockTransferClusterLengths  |  CShuffleBlockTransfer|                         Block-wiseGemm|               Block-wiseGemm|
+        //#######################################|        |        |               |        |       Type|       Type|       Type|  Type|    Type|     Type| Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block| Block| Block| Block|    |    |Wmma| Wmma|        |        |   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|    ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MRepeat    | NRepeat    | _MBlock_MPerBlock_NBlock_NPerBlock|        ScalarPerVector|                               Pipeline|                     Pipeline|
+        //#######################################|        |        |               |        |           |           |           |      |        |         |   Operation|   Operation|   Operation|               |      |     M|     N|     K|      |      |      |    |    |    |     |        |        | Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          |  Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          | PerShuffle | PerShuffle |                                   |                       |                              Scheduler|                      Version|
+        //#######################################|        |        |               |        |           |           |           |      |        |         |            |            |            |               |      |      |      |      |      |      |      |    |    |    |     |        |        |                |               |               |               |               |               |          |                 |               |               |              |               |               |          |            |            |                                   |                       |                                       |                             |
+        DeviceGemmMultiD_ABScale_Wmma_CShuffle_V3<     Row,     Row,        Tuple<>,     Row,    F8, F32,    F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough, PassThrough,       GemmSpec,   256,     1,   128,   128,   128,   128,   128,  16,  16,  16,   16,       4,       2,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         0,      S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,             16,         0,           1,           1,                     S<1, 32, 1, 8>,                   S<8>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1, F8>,
+        DeviceGemmMultiD_ABScale_Wmma_CShuffle_V3<     Row,     Row,        Tuple<>,     Row,    F8, F32,    F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough, PassThrough,       GemmSpec,   256,     1,   128,   128,   128,    64,   128,  16,  16,  16,   16,       2,       2,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         0,      S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,             16,         0,           1,           1,                     S<1, 64, 1, 4>,                   S<8>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1, F8>,
+        DeviceGemmMultiD_ABScale_Wmma_CShuffle_V3<     Row,     Row,        Tuple<>,     Row,    F8, F32,    F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough, PassThrough,       GemmSpec,   256,     1,   128,   128,    64,   128,   128,  16,  16,  16,   16,       2,       2,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         0,      S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,             16,         0,           1,           1,                     S<1, 32, 1, 8>,                   S<8>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1, F8>,
+        DeviceGemmMultiD_ABScale_Wmma_CShuffle_V3<     Row,     Row,        Tuple<>,     Row,    F8, F32,    F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough, PassThrough,       GemmSpec,   256,     1,   128,   128,    64,    64,   128,  16,  16,  16,   16,       2,       1,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         0,      S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,             16,         0,           1,           1,                     S<1, 32, 1, 8>,                   S<8>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1, F8>,
+
+        DeviceGemmMultiD_ABScale_Wmma_CShuffle_V3<     Row,     Row,        Tuple<>,     Row,    F8, F32,    F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough, PassThrough,       GemmSpec,   256,     1,   128,   128,   128,   128,   128,  16,  16,  16,   16,       4,       2,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         0,      S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,             16,         0,           1,           1,                     S<1, 32, 1, 8>,                   S<8>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3, F8>,
+        DeviceGemmMultiD_ABScale_Wmma_CShuffle_V3<     Row,     Row,        Tuple<>,     Row,    F8, F32,    F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough, PassThrough,       GemmSpec,   256,     1,   128,   128,   128,    64,   128,  16,  16,  16,   16,       2,       2,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         0,      S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,             16,         0,           1,           1,                     S<1, 64, 1, 4>,                   S<8>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3, F8>,
+        DeviceGemmMultiD_ABScale_Wmma_CShuffle_V3<     Row,     Row,        Tuple<>,     Row,    F8, F32,    F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough, PassThrough,       GemmSpec,   256,     1,   128,   128,    64,   128,   128,  16,  16,  16,   16,       2,       2,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         0,      S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,             16,         0,           1,           1,                     S<1, 32, 1, 8>,                   S<8>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3, F8>,
+        DeviceGemmMultiD_ABScale_Wmma_CShuffle_V3<     Row,     Row,        Tuple<>,     Row,    F8, F32,    F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough, PassThrough,       GemmSpec,   256,     1,   128,   128,    64,    64,   128,  16,  16,  16,   16,       2,       1,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         0,      S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,             16,         0,           1,           1,                     S<1, 32, 1, 8>,                   S<8>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3, F8>
+    // clang-format on
+    >;
+
+template <BlockGemmPipelineScheduler BlkGemmPipeSched, GemmSpecialization GemmSpec>
+using device_gemm_ab_scale_wmma_f8_f8_bf16_mk_kn_mn_1_128_128_mem_instances = std::tuple<
+    // clang-format off
+        //#######################################| ALayout| BLayout|       DsLayout| ELayout|      AData|      BData|     DsData| EData| AccData| Cshuffle|           A|           B|           C|           GEMM| Block| Scale| Scale| Scale|  MPer|  NPer|  KPer| AK1| BK1|MPer| NPer| MRepeat| NRepeat|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|   BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle   | CShuffle   |     CBlockTransferClusterLengths  |  CShuffleBlockTransfer|                         Block-wiseGemm|               Block-wiseGemm|
+        //#######################################|        |        |               |        |       Type|       Type|       Type|  Type|    Type|     Type| Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block| Block| Block| Block|    |    |Wmma| Wmma|        |        |   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|    ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MRepeat    | NRepeat    | _MBlock_MPerBlock_NBlock_NPerBlock|        ScalarPerVector|                               Pipeline|                     Pipeline|
+        //#######################################|        |        |               |        |           |           |           |      |        |         |   Operation|   Operation|   Operation|               |      |     M|     N|     K|      |      |      |    |    |    |     |        |        | Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          |  Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          | PerShuffle | PerShuffle |                                   |                       |                              Scheduler|                      Version|
+        //#######################################|        |        |               |        |           |           |           |      |        |         |            |            |            |               |      |      |      |      |      |      |      |    |    |    |     |        |        |                |               |               |               |               |               |          |                 |               |               |              |               |               |          |            |            |                                   |                       |                                       |                             |
+        // Memory friendly 
+        DeviceGemmMultiD_ABScale_Wmma_CShuffle_V3<     Row,     Row,        Tuple<>,     Row,    F8, F32,    F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough, PassThrough,       GemmSpec,   256,     1,   128,   128,    16,   256,   128,   8,  16,  16,   16,       1,       2,    S<16, 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,             16,         0,           1,           1,                    S<1, 16, 1, 16>,                   S<8>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1, F8>,
+        DeviceGemmMultiD_ABScale_Wmma_CShuffle_V3<     Row,     Row,        Tuple<>,     Row,    F8, F32,    F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough, PassThrough,       GemmSpec,   256,     1,   128,   128,    16,   128,   128,   8,  16,  16,   16,       1,       1,    S<16, 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,              4,             16,         0,           1,           1,                    S<1, 16, 1, 16>,                   S<8>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1, F8>,
+        DeviceGemmMultiD_ABScale_Wmma_CShuffle_V3<     Row,     Row,        Tuple<>,     Row,    F8, F32,    F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough, PassThrough,       GemmSpec,   128,     1,   128,   128,    16,    64,   128,   8,  16,  16,   16,       1,       1,    S<16,  8, 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,              4,             16,         0,           1,           1,                    S<1, 16, 1,  8>,                   S<8>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1, F8>,
+
+        DeviceGemmMultiD_ABScale_Wmma_CShuffle_V3<     Row,     Row,        Tuple<>,     Row,    F8, F32,    F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough, PassThrough,       GemmSpec,   256,     1,   128,   128,    32,   256,   128,  16,  16,  16,   16,       2,       2,    S< 8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         0,      S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,             16,         0,           1,           1,                    S<1, 16, 1, 16>,                   S<8>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1, F8>,
+        DeviceGemmMultiD_ABScale_Wmma_CShuffle_V3<     Row,     Row,        Tuple<>,     Row,    F8, F32,    F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough, PassThrough,       GemmSpec,   256,     1,   128,   128,    32,   128,   128,  16,  16,  16,   16,       2,       1,    S< 8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         0,      S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,             16,         0,           1,           1,                    S<1, 16, 1, 16>,                   S<8>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1, F8>,
+        DeviceGemmMultiD_ABScale_Wmma_CShuffle_V3<     Row,     Row,        Tuple<>,     Row,    F8, F32,    F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough, PassThrough,       GemmSpec,   256,     1,   128,   128,    32,    64,   128,  16,  16,  16,   16,       1,       1,    S< 8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         0,      S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,             16,         0,           1,           1,                    S<1, 32, 1,  8>,                   S<8>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1, F8>,
+
+        DeviceGemmMultiD_ABScale_Wmma_CShuffle_V3<     Row,     Row,        Tuple<>,     Row,    F8, F32,    F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough, PassThrough,       GemmSpec,   256,     1,   128,   128,    64,   256,   128,  16,  16,  16,   16,       4,       2,    S< 8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         0,      S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,             16,         0,           1,           1,                    S<1, 16, 1, 16>,                   S<8>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1, F8>,
+        DeviceGemmMultiD_ABScale_Wmma_CShuffle_V3<     Row,     Row,        Tuple<>,     Row,    F8, F32,    F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough, PassThrough,       GemmSpec,   256,     1,   128,   128,    64,   128,   128,  16,  16,  16,   16,       4,       1,    S< 8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         0,      S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,             16,         0,           1,           1,                    S<1, 16, 1, 16>,                   S<8>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1, F8>,
+        DeviceGemmMultiD_ABScale_Wmma_CShuffle_V3<     Row,     Row,        Tuple<>,     Row,    F8, F32,    F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough, PassThrough,       GemmSpec,   256,     1,   128,   128,    64,    64,   128,  16,  16,  16,   16,       2,       1,    S< 8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         0,      S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,             16,         0,           1,           1,                    S<1, 32, 1,  8>,                   S<8>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1, F8>
+    // clang-format on
+    >;
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/gemm_ab_scale/device_gemm_ab_scale_wmma_f8_f8_bf16/device_gemm_ab_scale_wmma_f8_f8_bf16_mk_kn_mn_128_128_128_comp_default_instance.cpp

@@ -0,0 +1,37 @@
+// Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
+// SPDX-License-Identifier: MIT
+
+#include "device_gemm_ab_scale_wmma_f8_f8_bf16_mk_kn_mn_128_128_128.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_gemm_ab_scale_wmma_f8_f8_bf16_mk_kn_mn_1_128_128_comp_default_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD_ABScaleSplitK<Row,
+                                                                  Row,
+                                                                  Tuple<>,
+                                                                  Row,
+                                                                  F8,
+                                                                  F32,
+                                                                  F8,
+                                                                  F32,
+                                                                  Tuple<>,
+                                                                  BF16,
+                                                                  1,
+                                                                  128,
+                                                                  128,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_gemm_ab_scale_wmma_f8_f8_bf16_mk_kn_mn_1_128_128_comp_instances<GemmDefault>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/gemm_ab_scale/device_gemm_ab_scale_wmma_f8_f8_bf16/device_gemm_ab_scale_wmma_f8_f8_bf16_mk_kn_mn_128_128_128_comp_kpadding_instance.cpp

@@ -0,0 +1,37 @@
+// Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
+// SPDX-License-Identifier: MIT
+
+#include "device_gemm_ab_scale_wmma_f8_f8_bf16_mk_kn_mn_128_128_128.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_gemm_ab_scale_wmma_f8_f8_bf16_mk_kn_mn_1_128_128_comp_kpadding_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD_ABScaleSplitK<Row,
+                                                                  Row,
+                                                                  Tuple<>,
+                                                                  Row,
+                                                                  F8,
+                                                                  F32,
+                                                                  F8,
+                                                                  F32,
+                                                                  Tuple<>,
+                                                                  BF16,
+                                                                  1,
+                                                                  128,
+                                                                  128,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_gemm_ab_scale_wmma_f8_f8_bf16_mk_kn_mn_1_128_128_comp_instances<GemmKPadding>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/gemm_ab_scale/device_gemm_ab_scale_wmma_f8_f8_bf16/device_gemm_ab_scale_wmma_f8_f8_bf16_mk_kn_mn_128_128_128_mem_v1_default_instance.cpp

@@ -0,0 +1,38 @@
+// Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
+// SPDX-License-Identifier: MIT
+
+#include "device_gemm_ab_scale_wmma_f8_f8_bf16_mk_kn_mn_128_128_128.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_gemm_ab_scale_wmma_f8_f8_bf16_mk_kn_mn_1_128_128_mem_v1_default_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD_ABScaleSplitK<Row,
+                                                                  Row,
+                                                                  Tuple<>,
+                                                                  Row,
+                                                                  F8,
+                                                                  F32,
+                                                                  F8,
+                                                                  F32,
+                                                                  Tuple<>,
+                                                                  BF16,
+                                                                  1,
+                                                                  128,
+                                                                  128,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_gemm_ab_scale_wmma_f8_f8_bf16_mk_kn_mn_1_128_128_mem_instances<Intrawave,
+                                                                              GemmDefault>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/gemm_ab_scale/device_gemm_ab_scale_wmma_f8_f8_bf16/device_gemm_ab_scale_wmma_f8_f8_bf16_mk_kn_mn_128_128_128_mem_v1_kpadding_instance.cpp

@@ -0,0 +1,38 @@
+// Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
+// SPDX-License-Identifier: MIT
+
+#include "device_gemm_ab_scale_wmma_f8_f8_bf16_mk_kn_mn_128_128_128.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_gemm_ab_scale_wmma_f8_f8_bf16_mk_kn_mn_1_128_128_mem_v1_kpadding_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD_ABScaleSplitK<Row,
+                                                                  Row,
+                                                                  Tuple<>,
+                                                                  Row,
+                                                                  F8,
+                                                                  F32,
+                                                                  F8,
+                                                                  F32,
+                                                                  Tuple<>,
+                                                                  BF16,
+                                                                  1,
+                                                                  128,
+                                                                  128,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_gemm_ab_scale_wmma_f8_f8_bf16_mk_kn_mn_1_128_128_mem_instances<Intrawave,
+                                                                              GemmKPadding>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/gemm_ab_scale/device_gemm_ab_scale_wmma_f8_f8_bf16/device_gemm_ab_scale_wmma_f8_f8_bf16_mk_nk_mn_128_128_128.hpp

@@ -0,0 +1,95 @@
+// Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
+// SPDX-License-Identifier: MIT
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_wmma_cshuffle_v3_ab_scale.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F8   = f8_t;
+using BF16 = bhalf_t;
+using F32  = float;
+
+using Row = tensor_layout::gemm::RowMajor;
+using Col = tensor_layout::gemm::ColumnMajor;
+
+template <index_t... Is>
+using S = Sequence<Is...>;
+
+using PassThrough = element_wise::PassThrough;
+using PassThrough = element_wise::PassThrough;
+
+static constexpr auto GemmDefault    = GemmSpecialization::Default;
+static constexpr auto GemmKPadding   = GemmSpecialization::KPadding;
+static constexpr auto GemmMNPadding  = GemmSpecialization::MNPadding;
+static constexpr auto GemmMNKPadding = GemmSpecialization::MNKPadding;
+
+static constexpr auto Intrawave = BlockGemmPipelineScheduler::Intrawave;
+
+template <GemmSpecialization GemmSpec>
+using device_gemm_ab_scale_wmma_f8_f8_bf16_mk_nk_mn_1_128_128_comp_instances = std::tuple<
+    // clang-format off
+        //#######################################| ALayout| BLayout|       DsLayout| ELayout|      AData|      BData|     DsData| EData| AccData| Cshuffle|           A|           B|              C|          GEMM| Block| Scale| Scale| Scale|  MPer|  NPer|  KPer| AK1| BK1|MPer| NPer| MRepeat| NRepeat|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|   BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle   | CShuffle   |     CBlockTransferClusterLengths  |  CShuffleBlockTransfer|                         Block-wiseGemm|               Block-wiseGemm|
+        //#######################################|        |        |               |        |       Type|       Type|       Type|  Type|    Type|     Type| Elementwise| Elementwise|    Elementwise|Specialization|  Size| Block| Block| Block| Block| Block| Block|    |    |Wmma| Wmma|        |        |   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|    ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MRepeat    | NRepeat    | _MBlock_MPerBlock_NBlock_NPerBlock|        ScalarPerVector|                               Pipeline|                     Pipeline|
+        //#######################################|        |        |               |        |           |           |           |      |        |         |   Operation|   Operation|      Operation|              |      |     M|     N|     K|      |      |      |    |    |    |     |        |        | Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          |  Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          | PerShuffle | PerShuffle |                                   |                       |                              Scheduler|                      Version|
+        //#######################################|        |        |               |        |           |           |           |      |        |         |            |            |               |              |      |      |      |      |      |      |      |    |    |    |     |        |        |                |               |               |               |               |               |          |                 |               |               |              |               |               |          |            |            |                                   |                       |                                       |                             |
+
+        // Compute friendly
+        DeviceGemmMultiD_ABScale_Wmma_CShuffle_V3<     Row,     Col,        Tuple<>,     Row,    F8, F32,    F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough,    PassThrough,      GemmSpec,   256,     1,   128,   128,   128,   128,   128,  16,  16,  16,   16,       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>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1, F8>,
+        DeviceGemmMultiD_ABScale_Wmma_CShuffle_V3<     Row,     Col,        Tuple<>,     Row,    F8, F32,    F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough,    PassThrough,      GemmSpec,   256,     1,   128,   128,   128,    64,   128,  16,  16,  16,   16,       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,  4>,                   S<8>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1, F8>,
+        DeviceGemmMultiD_ABScale_Wmma_CShuffle_V3<     Row,     Col,        Tuple<>,     Row,    F8, F32,    F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough,    PassThrough,      GemmSpec,   256,     1,   128,   128,    64,   128,   128,  16,  16,  16,   16,       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>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1, F8>,
+        DeviceGemmMultiD_ABScale_Wmma_CShuffle_V3<     Row,     Col,        Tuple<>,     Row,    F8, F32,    F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough,    PassThrough,      GemmSpec,   256,     1,   128,   128,    64,    64,   128,  16,  16,  16,   16,       2,       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>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1, F8>,
+
+        DeviceGemmMultiD_ABScale_Wmma_CShuffle_V3<     Row,     Col,        Tuple<>,     Row,    F8, F32,    F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough,    PassThrough,      GemmSpec,   256,     1,   128,   128,   128,   128,   128,  16,  16,  16,   16,       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>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3, F8>,
+        DeviceGemmMultiD_ABScale_Wmma_CShuffle_V3<     Row,     Col,        Tuple<>,     Row,    F8, F32,    F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough,    PassThrough,      GemmSpec,   256,     1,   128,   128,   128,    64,   128,  16,  16,  16,   16,       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,  4>,                   S<8>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3, F8>,
+        DeviceGemmMultiD_ABScale_Wmma_CShuffle_V3<     Row,     Col,        Tuple<>,     Row,    F8, F32,    F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough,    PassThrough,      GemmSpec,   256,     1,   128,   128,    64,   128,   128,  16,  16,  16,   16,       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>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3, F8>,
+        DeviceGemmMultiD_ABScale_Wmma_CShuffle_V3<     Row,     Col,        Tuple<>,     Row,    F8, F32,    F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough,    PassThrough,      GemmSpec,   256,     1,   128,   128,    64,    64,   128,  16,  16,  16,   16,       2,       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>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3, F8>,
+
+        DeviceGemmMultiD_ABScale_Wmma_CShuffle_V3<     Row,     Col,        Tuple<>,     Row,    F8, F32,    F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough,    PassThrough,      GemmSpec,   256,     1,   128,   128,   128,   128,    64,  16,  16,  16,   16,       4,       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>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3, F8>,
+        DeviceGemmMultiD_ABScale_Wmma_CShuffle_V3<     Row,     Col,        Tuple<>,     Row,    F8, F32,    F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough,    PassThrough,      GemmSpec,   256,     1,   128,   128,    64,    64,    64,  16,  16,  16,   16,       2,       1,    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>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3, F8>,
+
+        DeviceGemmMultiD_ABScale_Wmma_CShuffle_V3<     Row,     Col,        Tuple<>,     Row,    F8, F32,    F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough,    PassThrough,      GemmSpec,   256,     1,   128,   128,   128,   128,    64,  16,  16,  16,   16,       4,       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>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1, F8>,
+        DeviceGemmMultiD_ABScale_Wmma_CShuffle_V3<     Row,     Col,        Tuple<>,     Row,    F8, F32,    F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough,    PassThrough,      GemmSpec,   256,     1,   128,   128,    64,    64,    64,  16,  16,  16,   16,       2,       1,    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>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1, F8>
+    // clang-format on
+    >;
+
+template <BlockGemmPipelineScheduler BlkGemmPipeSched, GemmSpecialization GemmSpec>
+using device_gemm_ab_scale_wmma_f8_f8_bf16_mk_nk_mn_1_128_128_mem_instances = std::tuple<
+    // clang-format off
+        //#######################################| ALayout| BLayout|       DsLayout| ELayout|      AData|      BData|     DsData| EData| AccData| Cshuffle|           A|           B|              C|          GEMM| Block| Scale| Scale| Scale|  MPer|  NPer|  KPer| AK1| BK1|MPer| NPer| MRepeat| NRepeat|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|   BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle   | CShuffle   |     CBlockTransferClusterLengths  |  CShuffleBlockTransfer|    Block-wiseGemm|               Block-wiseGemm|
+        //#######################################|        |        |               |        |       Type|       Type|       Type|  Type|    Type|     Type| Elementwise| Elementwise|    Elementwise|Specialization|  Size| Block| Block| Block| Block| Block| Block|    |    |Wmma| Wmma|        |        |   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|    ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MRepeat    | NRepeat    | _MBlock_MPerBlock_NBlock_NPerBlock|        ScalarPerVector|          Pipeline|                     Pipeline|
+        //#######################################|        |        |               |        |           |           |           |      |        |         |   Operation|   Operation|      Operation|              |      |     M|     N|     K|      |      |      |    |    |    |     |        |        | Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          |  Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          | PerShuffle | PerShuffle |                                   |                       |         Scheduler|                      Version|
+        //#######################################|        |        |               |        |           |           |           |      |        |         |            |            |               |              |      |      |      |      |      |      |      |    |    |    |     |        |        |                |               |               |               |               |               |          |                 |               |               |              |               |               |          |            |            |                                   |                       |                  |                             |
+
+        // Memory friendly
+        DeviceGemmMultiD_ABScale_Wmma_CShuffle_V3<     Row,     Col,        Tuple<>,     Row,    F8, F32,    F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough,    PassThrough,      GemmSpec,   256,     1,   128,   128,    16,   256,   128,   8,  16,  16,   16,       1,       2,    S<16, 16, 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,             16,             16,         0,           1,           2,                    S<1, 16, 1,  8>,                   S<8>,  BlkGemmPipeSched, BlockGemmPipelineVersion::v1, F8>,
+        DeviceGemmMultiD_ABScale_Wmma_CShuffle_V3<     Row,     Col,        Tuple<>,     Row,    F8, F32,    F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough,    PassThrough,      GemmSpec,   256,     1,   128,   128,    16,   128,   128,   8,  16,  16,   16,       1,       1,    S<16, 16, 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,             16,             16,         0,           1,           1,                    S<1, 16, 1,  8>,                   S<8>,  BlkGemmPipeSched, BlockGemmPipelineVersion::v1, F8>,
+        DeviceGemmMultiD_ABScale_Wmma_CShuffle_V3<     Row,     Col,        Tuple<>,     Row,    F8, F32,    F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough,    PassThrough,      GemmSpec,   128,     1,   128,   128,    16,    64,   128,   8,  16,  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,             16,             16,         0,           1,           1,                    S<1, 16, 1,  8>,                   S<4>,  BlkGemmPipeSched, BlockGemmPipelineVersion::v1, F8>,
+        DeviceGemmMultiD_ABScale_Wmma_CShuffle_V3<     Row,     Col,        Tuple<>,     Row,    F8, F32,    F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough,    PassThrough,      GemmSpec,   256,     1,   128,   128,    16,   128,   256,   8,  16,  16,   16,       1,       1,    S<16, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         0,     S<16, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         0,           1,           1,                    S<1, 16, 1, 16>,                   S<8>,  BlkGemmPipeSched, BlockGemmPipelineVersion::v1, F8>,
+        DeviceGemmMultiD_ABScale_Wmma_CShuffle_V3<     Row,     Col,        Tuple<>,     Row,    F8, F32,    F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough,    PassThrough,      GemmSpec,   128,     1,   128,   128,    16,    64,   256,   8,  16,  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<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         0,           1,           1,                    S<1, 16, 1,  8>,                   S<4>,  BlkGemmPipeSched, BlockGemmPipelineVersion::v1, F8>,
+
+        DeviceGemmMultiD_ABScale_Wmma_CShuffle_V3<     Row,     Col,        Tuple<>,     Row,    F8, F32,    F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough,    PassThrough,      GemmSpec,   256,     1,   128,   128,    32,   256,   128,  16,  16,  16,   16,       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>,  BlkGemmPipeSched, BlockGemmPipelineVersion::v1, F8>,
+        DeviceGemmMultiD_ABScale_Wmma_CShuffle_V3<     Row,     Col,        Tuple<>,     Row,    F8, F32,    F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough,    PassThrough,      GemmSpec,   256,     1,   128,   128,    32,   128,   128,  16,  16,  16,   16,       2,       1,    S< 8, 16, 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, 16, 1, 8>,                    S<8>,  BlkGemmPipeSched, BlockGemmPipelineVersion::v1, F8>,
+        DeviceGemmMultiD_ABScale_Wmma_CShuffle_V3<     Row,     Col,        Tuple<>,     Row,    F8, F32,    F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough,    PassThrough,      GemmSpec,   256,     1,   128,   128,    32,    64,   128,  16,  16,  16,   16,       1,       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>,  BlkGemmPipeSched, BlockGemmPipelineVersion::v1, F8>,
+        DeviceGemmMultiD_ABScale_Wmma_CShuffle_V3<     Row,     Col,        Tuple<>,     Row,    F8, F32,    F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough,    PassThrough,      GemmSpec,   256,     1,   128,   128,    32,   128,   256,  16,  16,  16,   16,       1,       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,           1,                    S<1, 32, 1, 8>,                    S<8>,  BlkGemmPipeSched, BlockGemmPipelineVersion::v1, F8>,
+        DeviceGemmMultiD_ABScale_Wmma_CShuffle_V3<     Row,     Col,        Tuple<>,     Row,    F8, F32,    F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough,    PassThrough,      GemmSpec,   128,     1,   128,   128,    32,    64,   256,  16,  16,  16,   16,       2,       1,    S<16,  8, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         0,     S<16,  8, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         0,           1,           1,                    S<1, 16, 1, 8>,                    S<8>,  BlkGemmPipeSched, BlockGemmPipelineVersion::v1, F8>,
+
+        DeviceGemmMultiD_ABScale_Wmma_CShuffle_V3<     Row,     Col,        Tuple<>,     Row,    F8, F32,    F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough,    PassThrough,      GemmSpec,   256,     1,   128,   128,    64,   256,   128,  16,  16,  16,   16,       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>,  BlkGemmPipeSched, BlockGemmPipelineVersion::v1, F8>,
+        DeviceGemmMultiD_ABScale_Wmma_CShuffle_V3<     Row,     Col,        Tuple<>,     Row,    F8, F32,    F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough,    PassThrough,      GemmSpec,   256,     1,   128,   128,    64,   128,   128,  16,  16,  16,   16,       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>,  BlkGemmPipeSched, BlockGemmPipelineVersion::v1, F8>,
+        DeviceGemmMultiD_ABScale_Wmma_CShuffle_V3<     Row,     Col,        Tuple<>,     Row,    F8, F32,    F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough,    PassThrough,      GemmSpec,   256,     1,   128,   128,    64,    64,   128,  16,  16,  16,   16,       2,       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>,  BlkGemmPipeSched, BlockGemmPipelineVersion::v1, F8>,
+        DeviceGemmMultiD_ABScale_Wmma_CShuffle_V3<     Row,     Col,        Tuple<>,     Row,    F8, F32,    F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough,    PassThrough,      GemmSpec,   256,     1,   128,   128,    64,   128,   256,  16,  16,  16,   16,       2,       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,           1,                    S<1, 32, 1, 8>,                    S<8>,  BlkGemmPipeSched, BlockGemmPipelineVersion::v1, F8>,
+        DeviceGemmMultiD_ABScale_Wmma_CShuffle_V3<     Row,     Col,        Tuple<>,     Row,    F8, F32,    F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough,    PassThrough,      GemmSpec,   256,     1,   128,   128,    64,    64,   256,  16,  16,  16,   16,       2,       1,    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>,  BlkGemmPipeSched, BlockGemmPipelineVersion::v1, F8>
+    // clang-format on
+    >;
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/gemm_ab_scale/device_gemm_ab_scale_wmma_f8_f8_bf16/device_gemm_ab_scale_wmma_f8_f8_bf16_mk_nk_mn_128_128_128_comp_default_instance.cpp

@@ -0,0 +1,37 @@
+// Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
+// SPDX-License-Identifier: MIT
+
+#include "device_gemm_ab_scale_wmma_f8_f8_bf16_mk_nk_mn_128_128_128.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_gemm_ab_scale_wmma_f8_f8_bf16_mk_nk_mn_1_128_128_comp_default_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD_ABScaleSplitK<Row,
+                                                                  Col,
+                                                                  Tuple<>,
+                                                                  Row,
+                                                                  F8,
+                                                                  F32,
+                                                                  F8,
+                                                                  F32,
+                                                                  Tuple<>,
+                                                                  BF16,
+                                                                  1,
+                                                                  128,
+                                                                  128,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_gemm_ab_scale_wmma_f8_f8_bf16_mk_nk_mn_1_128_128_comp_instances<GemmDefault>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/gemm_ab_scale/device_gemm_ab_scale_wmma_f8_f8_bf16/device_gemm_ab_scale_wmma_f8_f8_bf16_mk_nk_mn_128_128_128_comp_kpadding_instance.cpp

@@ -0,0 +1,37 @@
+// Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
+// SPDX-License-Identifier: MIT
+
+#include "device_gemm_ab_scale_wmma_f8_f8_bf16_mk_nk_mn_128_128_128.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_gemm_ab_scale_wmma_f8_f8_bf16_mk_nk_mn_1_128_128_comp_kpadding_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD_ABScaleSplitK<Row,
+                                                                  Col,
+                                                                  Tuple<>,
+                                                                  Row,
+                                                                  F8,
+                                                                  F32,
+                                                                  F8,
+                                                                  F32,
+                                                                  Tuple<>,
+                                                                  BF16,
+                                                                  1,
+                                                                  128,
+                                                                  128,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_gemm_ab_scale_wmma_f8_f8_bf16_mk_nk_mn_1_128_128_comp_instances<GemmKPadding>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/gemm_ab_scale/device_gemm_ab_scale_wmma_f8_f8_bf16/device_gemm_ab_scale_wmma_f8_f8_bf16_mk_nk_mn_128_128_128_mem_v1_default_instance.cpp

@@ -0,0 +1,38 @@
+// Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
+// SPDX-License-Identifier: MIT
+
+#include "device_gemm_ab_scale_wmma_f8_f8_bf16_mk_nk_mn_128_128_128.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_gemm_ab_scale_wmma_f8_f8_bf16_mk_nk_mn_1_128_128_mem_v1_default_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD_ABScaleSplitK<Row,
+                                                                  Col,
+                                                                  Tuple<>,
+                                                                  Row,
+                                                                  F8,
+                                                                  F32,
+                                                                  F8,
+                                                                  F32,
+                                                                  Tuple<>,
+                                                                  BF16,
+                                                                  1,
+                                                                  128,
+                                                                  128,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_gemm_ab_scale_wmma_f8_f8_bf16_mk_nk_mn_1_128_128_mem_instances<Intrawave,
+                                                                              GemmDefault>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/gemm_ab_scale/device_gemm_ab_scale_wmma_f8_f8_bf16/device_gemm_ab_scale_wmma_f8_f8_bf16_mk_nk_mn_128_128_128_mem_v1_kpadding_instance.cpp

@@ -0,0 +1,38 @@
+// Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
+// SPDX-License-Identifier: MIT
+
+#include "device_gemm_ab_scale_wmma_f8_f8_bf16_mk_nk_mn_128_128_128.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_gemm_ab_scale_wmma_f8_f8_bf16_mk_nk_mn_1_128_128_mem_v1_kpadding_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD_ABScaleSplitK<Row,
+                                                                  Col,
+                                                                  Tuple<>,
+                                                                  Row,
+                                                                  F8,
+                                                                  F32,
+                                                                  F8,
+                                                                  F32,
+                                                                  Tuple<>,
+                                                                  BF16,
+                                                                  1,
+                                                                  128,
+                                                                  128,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_gemm_ab_scale_wmma_f8_f8_bf16_mk_nk_mn_1_128_128_mem_instances<Intrawave,
+                                                                              GemmKPadding>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

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

@@ -1,7 +1,7 @@
 # Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
 # SPDX-License-Identifier: MIT
 
-# ONLY XDL_KERNELS
+# ONLY XDL_AND_WMMA_KERNELS
 if(SUPPORTED_GPU_TARGETS MATCHES "gfx9[45]|gfx12")
         set(GEMM_BLOCKSCALE_WP_INSTANCES)
 
@@ -10,6 +10,9 @@ if(SUPPORTED_GPU_TARGETS MATCHES "gfx9[45]|gfx12")
                 device_gemm_blockscale_wp_xdl_f8_f8_bf16/device_gemm_blockscale_wp_xdl_f8_f8_bf16_mk_nk_mn_128_128_128_comp_kpadding_instance.cpp
                 device_gemm_blockscale_wp_xdl_f8_f8_bf16/device_gemm_blockscale_wp_xdl_f8_f8_bf16_mk_nk_mn_128_128_128_mem_v1_default_instance.cpp
                 device_gemm_blockscale_wp_xdl_f8_f8_bf16/device_gemm_blockscale_wp_xdl_f8_f8_bf16_mk_nk_mn_128_128_128_mem_v1_kpadding_instance.cpp
+
+                device_gemm_blockscale_wp_wmma_f8_f8_bf16/device_gemm_blockscale_wp_wmma_f8_f8_bf16_mk_nk_mn_128_128_128_comp_default_instance.cpp
+                device_gemm_blockscale_wp_wmma_f8_f8_bf16/device_gemm_blockscale_wp_wmma_f8_f8_bf16_mk_nk_mn_128_128_128_mem_default_instance.cpp
                 )
         check_cxx_compiler_flag("-mllvm --misched-bottomup=1" HAS_MISCHED_BOTTOMUP)
         check_cxx_compiler_flag("-mllvm --misched-prera-direction=bottomup" HAS_MISCHED_PRERA_DIRECTION)

library/src/tensor_operation_instance/gpu/gemm_blockscale_wp/device_gemm_blockscale_wp_wmma_f8_f8_bf16/device_gemm_blockscale_wp_wmma_f8_f8_bf16_mk_nk_mn_128_128_128.hpp

@@ -0,0 +1,77 @@
+// Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
+// SPDX-License-Identifier: MIT
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_wmma_cshuffle_v3_blockscale_bpreshuffle.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F8   = f8_t;
+using BF16 = bhalf_t;
+using F32  = float;
+
+using Row = tensor_layout::gemm::RowMajor;
+using Col = tensor_layout::gemm::ColumnMajor;
+
+template <index_t... Is>
+using S = Sequence<Is...>;
+
+using PassThrough = element_wise::PassThrough;
+
+static constexpr auto GemmDefault    = GemmSpecialization::Default;
+static constexpr auto GemmKPadding   = GemmSpecialization::KPadding;
+static constexpr auto GemmMNPadding  = GemmSpecialization::MNPadding;
+static constexpr auto GemmMNKPadding = GemmSpecialization::MNKPadding;
+
+static constexpr auto Intrawave = BlockGemmPipelineScheduler::Intrawave;
+
+template <GemmSpecialization GemmSpec>
+using device_gemm_blockscale_wp_wmma_f8_f8_bf16_mk_nk_mn_1_128_128_comp_instances = std::tuple<
+    // clang-format off
+        //######################################################| ALayout| BLayout| DsLayout| ELayout|   AData|   BData|     DsData| EData| AccData| Cshuffle|           A|           B|              C|          GEMM| Block| Scale| Scale| Scale|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MRepeat| NRepeat|    ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer|  ABlockTransfer| ABlockLds|    BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|     CDEBlockTransferClusterLengths|  CShuffleBlockTransfer|                         Block-wiseGemm|               Block-wiseGemm|
+        //######################################################|        |        |         |        |    Type|    Type|       Type|  Type|    Type|     Type| Elementwise| Elementwise|    Elementwise|Specialization|  Size| Block| Block| Block| Block| Block| Block|    |    | Wmma| Wmma|        |        |     ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|       DstScalar| AddExtraM|     ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN|     MRepeat|     NRepeat| _MBlock_MPerBlock_NBlock_NPerBlock|       ScalarPerVectors|                               Pipeline|                     Pipeline|
+        //######################################################|        |        |         |        |        |        |           |      |        |         |   Operation|   Operation|      Operation|              |      |     M|     N|     K|      |      |      |    |    |     |     |        |        | Lengths_AK0_M_AK1|   ArrangeOrder|               |               |      PerVector|   PerVector_AK1|          | Lengths_BK0_N_BK1|   ArrangeOrder|               |              |      PerVector|  PerVector_BK1|          |  PerShuffle|  PerShuffle|                                   |                       |                              Scheduler|                     Verision|
+        //######################################################|        |        |         |        |        |        |           |      |        |         |            |            |               |              |      |      |      |      |      |      |      |    |    |     |     |        |        |                  |               |               |               |               |                |          |                  |               |               |              |               |               |          |            |            |                                   |                       |                                       |                             |
+        DeviceGemmMultiD_BlockScale_Wmma_CShuffle_V3_BPreshuffle<     Row,     Col,  Tuple<>,     Row, F8, F32, F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough,    PassThrough,      GemmSpec,   256,     1,   128,   128,   128,   128,   128,  16,  16,   16,   16,       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>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1, F8>,
+        DeviceGemmMultiD_BlockScale_Wmma_CShuffle_V3_BPreshuffle<     Row,     Col,  Tuple<>,     Row, F8, F32, F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough,    PassThrough,      GemmSpec,   256,     1,   128,   128,    64,   128,   128,  16,  16,   16,   16,       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>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1, F8>,
+        DeviceGemmMultiD_BlockScale_Wmma_CShuffle_V3_BPreshuffle<     Row,     Col,  Tuple<>,     Row, F8, F32, F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough,    PassThrough,      GemmSpec,   256,     1,   128,   128,    64,    64,   128,  16,  16,   16,   16,       2,       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>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1, F8>
+    // clang-format on
+    >;
+
+template <GemmSpecialization GemmSpec>
+using device_gemm_blockscale_wp_wmma_f8_f8_bf16_mk_nk_mn_1_128_128_mem_instances = std::tuple<
+    // clang-format off
+        //######################################################| ALayout| BLayout| DsLayout| ELayout|   AData|   BData|     DsData| EData| AccData| Cshuffle|           A|           B|              C|          GEMM| Block| Scale| Scale| Scale|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MRepeat| NRepeat|    ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer|  ABlockTransfer| ABlockLds|    BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|     CDEBlockTransferClusterLengths|  CShuffleBlockTransfer|                         Block-wiseGemm|               Block-wiseGemm|
+        //######################################################|        |        |         |        |    Type|    Type|       Type|  Type|    Type|     Type| Elementwise| Elementwise|    Elementwise|Specialization|  Size| Block| Block| Block| Block| Block| Block|    |    | Wmma| Wmma|        |        |     ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|       DstScalar| AddExtraM|     ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN|     MRepeat|     NRepeat| _MBlock_MPerBlock_NBlock_NPerBlock|       ScalarPerVectors|                               Pipeline|                     Pipeline|
+        //######################################################|        |        |         |        |        |        |           |      |        |         |   Operation|   Operation|      Operation|              |      |     M|     N|     K|      |      |      |    |    |     |     |        |        | Lengths_AK0_M_AK1|   ArrangeOrder|               |               |      PerVector|   PerVector_AK1|          | Lengths_BK0_N_BK1|   ArrangeOrder|               |              |      PerVector|  PerVector_BK1|          |  PerShuffle|  PerShuffle|                                   |                       |                              Scheduler|                     Verision|
+        //######################################################|        |        |         |        |        |        |           |      |        |         |            |            |               |              |      |      |      |      |      |      |      |    |    |     |     |        |        |                  |               |               |               |               |                |          |                  |               |               |              |               |               |          |            |            |                                   |                       |                                       |                             |
+        DeviceGemmMultiD_BlockScale_Wmma_CShuffle_V3_BPreshuffle<     Row,     Col,  Tuple<>,     Row, F8, F32, F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough,    PassThrough,      GemmSpec,   256,     1,   128,   128,    16,   256,   128,   8,  16,   16,   16,       1,       2,      S<16, 16, 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,             16,             16,         0,           1,           1,                    S<1, 16, 1, 16>,                   S<8>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1, F8>,
+        DeviceGemmMultiD_BlockScale_Wmma_CShuffle_V3_BPreshuffle<     Row,     Col,  Tuple<>,     Row, F8, F32, F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough,    PassThrough,      GemmSpec,   256,     1,   128,   128,    16,   128,   128,   8,  16,   16,   16,       1,       1,      S<16, 16, 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,             16,             16,         0,           1,           1,                    S<1, 16, 1, 16>,                   S<8>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1, F8>,
+        DeviceGemmMultiD_BlockScale_Wmma_CShuffle_V3_BPreshuffle<     Row,     Col,  Tuple<>,     Row, F8, F32, F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough,    PassThrough,      GemmSpec,   128,     1,   128,   128,    16,    64,   128,   8,  16,   16,   16,       1,       1,      S<16,  8, 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,             16,             16,         0,           1,           1,                    S<1, 16, 1,  8>,                   S<8>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1, F8>,
+        DeviceGemmMultiD_BlockScale_Wmma_CShuffle_V3_BPreshuffle<     Row,     Col,  Tuple<>,     Row, F8, F32, F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough,    PassThrough,      GemmSpec,   256,     1,   128,   128,    16,   128,   256,  16,  16,   16,   16,       1,       1,      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, 16, 1, 16>,                   S<8>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1, F8>,
+        DeviceGemmMultiD_BlockScale_Wmma_CShuffle_V3_BPreshuffle<     Row,     Col,  Tuple<>,     Row, F8, F32, F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough,    PassThrough,      GemmSpec,   128,     1,   128,   128,    16,    64,   256,  16,  16,   16,   16,       1,       1,      S<16,  8, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,              16,         0,      S<16,  8, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         0,           1,           1,                    S<1, 16, 1,  8>,                   S<8>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1, F8>,
+
+        DeviceGemmMultiD_BlockScale_Wmma_CShuffle_V3_BPreshuffle<     Row,     Col,  Tuple<>,     Row, F8, F32, F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough,    PassThrough,      GemmSpec,   256,     1,   128,   128,    32,   256,   128,  16,  16,   16,   16,       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, 16, 1, 16>,                   S<8>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1, F8>,
+        DeviceGemmMultiD_BlockScale_Wmma_CShuffle_V3_BPreshuffle<     Row,     Col,  Tuple<>,     Row, F8, F32, F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough,    PassThrough,      GemmSpec,   256,     1,   128,   128,    32,   128,   128,  16,  16,   16,   16,       2,       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, 16, 1, 16>,                   S<8>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1, F8>,
+        DeviceGemmMultiD_BlockScale_Wmma_CShuffle_V3_BPreshuffle<     Row,     Col,  Tuple<>,     Row, F8, F32, F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough,    PassThrough,      GemmSpec,   256,     1,   128,   128,    32,    64,   128,  16,  16,   16,   16,       1,       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>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1, F8>,
+        DeviceGemmMultiD_BlockScale_Wmma_CShuffle_V3_BPreshuffle<     Row,     Col,  Tuple<>,     Row, F8, F32, F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough,    PassThrough,      GemmSpec,   256,     1,   128,   128,    32,   128,   256,  16,  16,   16,   16,       2,       1,      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, 16, 1, 16>,                   S<8>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1, F8>,
+        DeviceGemmMultiD_BlockScale_Wmma_CShuffle_V3_BPreshuffle<     Row,     Col,  Tuple<>,     Row, F8, F32, F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough,    PassThrough,      GemmSpec,   256,     1,   128,   128,    32,    64,   256,  16,  16,   16,   16,       1,       1,      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>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1, F8>,
+
+        DeviceGemmMultiD_BlockScale_Wmma_CShuffle_V3_BPreshuffle<     Row,     Col,  Tuple<>,     Row, F8, F32, F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough,    PassThrough,      GemmSpec,   256,     1,   128,   128,    64,   256,   128,  16,  16,   16,   16,       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, 16, 1, 16>,                   S<8>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1, F8>,
+        DeviceGemmMultiD_BlockScale_Wmma_CShuffle_V3_BPreshuffle<     Row,     Col,  Tuple<>,     Row, F8, F32, F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough,    PassThrough,      GemmSpec,   256,     1,   128,   128,    64,   128,   128,  16,  16,   16,   16,       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, 16, 1, 16>,                   S<8>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1, F8>,
+        DeviceGemmMultiD_BlockScale_Wmma_CShuffle_V3_BPreshuffle<     Row,     Col,  Tuple<>,     Row, F8, F32, F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough,    PassThrough,      GemmSpec,   256,     1,   128,   128,    64,    64,   128,  16,  16,   16,   16,       2,       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>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1, F8>,
+        DeviceGemmMultiD_BlockScale_Wmma_CShuffle_V3_BPreshuffle<     Row,     Col,  Tuple<>,     Row, F8, F32, F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough,    PassThrough,      GemmSpec,   256,     1,   128,   128,    64,   128,   256,  16,  16,   16,   16,       4,       1,      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, 16, 1, 16>,                   S<8>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1, F8>,
+        DeviceGemmMultiD_BlockScale_Wmma_CShuffle_V3_BPreshuffle<     Row,     Col,  Tuple<>,     Row, F8, F32, F8, F32,    Tuple<>,  BF16,     F32,      F32, PassThrough, PassThrough,    PassThrough,      GemmSpec,   256,     1,   128,   128,    64,    64,   256,  16,  16,   16,   16,       2,       1,      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>,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v1, F8>
+    // clang-format on
+    >;
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/gemm_blockscale_wp/device_gemm_blockscale_wp_wmma_f8_f8_bf16/device_gemm_blockscale_wp_wmma_f8_f8_bf16_mk_nk_mn_128_128_128_comp_default_instance.cpp

@@ -0,0 +1,38 @@
+// Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
+// SPDX-License-Identifier: MIT
+
+#include "device_gemm_blockscale_wp_wmma_f8_f8_bf16_mk_nk_mn_128_128_128.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_gemm_blockscale_wp_wmma_f8_f8_bf16_mk_nk_mn_1_128_128_comp_default_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD_BlockScale_BPreshuffleSplitK<Row,
+                                                                                 Col,
+                                                                                 Tuple<>,
+                                                                                 Row,
+                                                                                 F8,
+                                                                                 F32,
+                                                                                 F8,
+                                                                                 F32,
+                                                                                 Tuple<>,
+                                                                                 BF16,
+                                                                                 1,
+                                                                                 128,
+                                                                                 128,
+                                                                                 PassThrough,
+                                                                                 PassThrough,
+                                                                                 PassThrough>>>&
+        instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_gemm_blockscale_wp_wmma_f8_f8_bf16_mk_nk_mn_1_128_128_comp_instances<GemmDefault>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/gemm_blockscale_wp/device_gemm_blockscale_wp_wmma_f8_f8_bf16/device_gemm_blockscale_wp_wmma_f8_f8_bf16_mk_nk_mn_128_128_128_mem_default_instance.cpp

@@ -0,0 +1,38 @@
+// Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
+// SPDX-License-Identifier: MIT
+
+#include "device_gemm_blockscale_wp_wmma_f8_f8_bf16_mk_nk_mn_128_128_128.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_gemm_blockscale_wp_wmma_f8_f8_bf16_mk_nk_mn_1_128_128_mem_default_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD_BlockScale_BPreshuffleSplitK<Row,
+                                                                                 Col,
+                                                                                 Tuple<>,
+                                                                                 Row,
+                                                                                 F8,
+                                                                                 F32,
+                                                                                 F8,
+                                                                                 F32,
+                                                                                 Tuple<>,
+                                                                                 BF16,
+                                                                                 1,
+                                                                                 128,
+                                                                                 128,
+                                                                                 PassThrough,
+                                                                                 PassThrough,
+                                                                                 PassThrough>>>&
+        instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_gemm_blockscale_wp_wmma_f8_f8_bf16_mk_nk_mn_1_128_128_mem_instances<GemmDefault>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

profiler/include/profiler/profile_gemm_ab_scale_impl.hpp

@@ -109,8 +109,8 @@ bool profile_gemm_ab_scale_impl(int do_verification,
     case 1:
         a0_m_k.GenerateTensorValue(GeneratorTensor_2<A0DataType>{-2, 2});
         b0_k_n.GenerateTensorValue(GeneratorTensor_2<B0DataType>{-2, 2});
-        a1_m_k.GenerateTensorValue(GeneratorTensor_3<A1DataType>{0, 1.0});
-        b1_k_n.GenerateTensorValue(GeneratorTensor_3<B1DataType>{0, 1.0});
+        a1_m_k.GenerateTensorValue(GeneratorTensor_2<A1DataType>{-1, 2});
+        b1_k_n.GenerateTensorValue(GeneratorTensor_2<B1DataType>{-1, 2});
         break;
     default:
         a0_m_k.GenerateTensorValue(GeneratorTensor_3<A0DataType>{-0.5, 0.5});
@@ -302,7 +302,7 @@ bool profile_gemm_ab_scale_impl(int do_verification,
             float gb_per_sec = num_btype / 1.E6 / ave_time;
 
             std::cout << "Perf: " << std::setw(10) << ave_time << " ms, " << tflops << " TFlops, "
-                      << gb_per_sec << " GB/s, " << op_name << std::endl;
+                      << gb_per_sec << " GB/s, " << op_name << ", KBatch " << KBatch << std::endl;
 
             if(tflops > best_tflops)
             {

profiler/include/profiler/profile_gemm_blockscale_wp_impl.hpp

@@ -29,7 +29,7 @@ void preShuffleBuffer(const InOutDataType* src, InOutDataType* dst, int N, int K
 {
     int KPack = 16;
     int NLane = NXdl;
-    int KLane = 64 / NLane;
+    int KLane = ck::get_warp_size() / NLane;
 
     int K0 = K / (KLane * KPack);
     // K -> K0 KLane KPack

test/CMakeLists.txt

@@ -261,6 +261,7 @@ add_subdirectory(gemm_multiply_multiply_wp)
 add_subdirectory(gemm_split_k)
 add_subdirectory(gemm_universal)
 add_subdirectory(gemm_universal_preshuffle)
+add_subdirectory(gemm_ab_scale)
 add_subdirectory(gemm_b_scale)
 add_subdirectory(gemm_universal_streamk)
 add_subdirectory(gemm_reduce)

test/gemm_ab_scale/CMakeLists.txt

@@ -0,0 +1,9 @@
+# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
+# SPDX-License-Identifier: MIT
+
+if(GPU_TARGETS MATCHES "gfx9[45]|gfx12")
+    add_gtest_executable(test_gemm_ab_scale test_gemm_ab_scale.cpp)
+    if(result EQUAL 0)
+        target_link_libraries(test_gemm_ab_scale PRIVATE utility device_gemm_ab_scale_instance)
+    endif()
+endif()

test/gemm_ab_scale/test_gemm_ab_scale.cpp

@@ -0,0 +1,236 @@
+// Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
+// SPDX-License-Identifier: MIT
+
+#include <tuple>
+
+#include "gtest/gtest.h"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "test_gemm_ab_scale_util.hpp"
+
+using BF16 = ck::bhalf_t;
+using F32  = float;
+using F8   = ck::f8_t;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+namespace {
+
+template <typename X, typename Y>
+struct tuple_concat;
+
+template <typename... Xs, typename... Ys>
+struct tuple_concat<std::tuple<Xs...>, std::tuple<Ys...>>
+{
+    using type = std::tuple<Xs..., Ys...>;
+};
+
+} // namespace
+
+template <typename Tuple>
+class TestGemmABScale_MK_NK : public ck::test::TestGemmABScale<
+                                  typename tuple_concat<std::tuple<Row, Col, Row>, Tuple>::type>
+{
+};
+
+template <typename Tuple>
+class TestGemmABScale_MK_KN : public ck::test::TestGemmABScale<
+                                  typename tuple_concat<std::tuple<Row, Row, Row>, Tuple>::type>
+{
+};
+
+template <typename Tuple>
+class TestGemmABScale_KM_KN : public ck::test::TestGemmABScale<
+                                  typename tuple_concat<std::tuple<Col, Row, Row>, Tuple>::type>
+{
+};
+
+// clang-format off
+using KernelTypes = ::testing::Types<
+    //         ADataType, BDataType, ComputeDataType, EDataType
+    std::tuple<  F8, F32,   F8, F32,              F8,      BF16>
+    >;
+// clang-format on
+
+TYPED_TEST_SUITE(TestGemmABScale_MK_NK, KernelTypes);
+TYPED_TEST_SUITE(TestGemmABScale_MK_KN, KernelTypes);
+TYPED_TEST_SUITE(TestGemmABScale_KM_KN, KernelTypes);
+
+// Row Col
+TYPED_TEST(TestGemmABScale_MK_NK, SmallM)
+{
+    std::vector<int> Ms{1, 2, 3, 4, 5, 6};
+    constexpr int N = 512;
+    constexpr int K = 1024;
+
+    constexpr int StrideA = K;
+    constexpr int StrideB = K;
+    constexpr int StrideC = N;
+
+    for(int M : Ms)
+        this->Run(M, N, K, StrideA, StrideB, StrideC);
+}
+
+TYPED_TEST(TestGemmABScale_MK_NK, SmallMPadK)
+{
+    std::vector<int> Ms{1, 2, 3, 4, 5, 6};
+    constexpr int N = 512;
+    constexpr int K = 704;
+
+    constexpr int StrideA = K;
+    constexpr int StrideB = K;
+    constexpr int StrideC = N;
+
+    for(int M : Ms)
+        this->Run(M, N, K, StrideA, StrideB, StrideC);
+}
+
+TYPED_TEST(TestGemmABScale_MK_NK, MidLargeM)
+{
+    std::vector<int> Ms{127, 255, 312, 799, 1573};
+    constexpr int N = 512;
+    constexpr int K = 1024;
+
+    constexpr int StrideA = K;
+    constexpr int StrideB = K;
+    constexpr int StrideC = N;
+
+    for(int M : Ms)
+        this->Run(M, N, K, StrideA, StrideB, StrideC);
+}
+
+TYPED_TEST(TestGemmABScale_MK_NK, Regular)
+{
+    std::vector<int> Ms{512};
+    constexpr int N = 512;
+    constexpr int K = 1024;
+
+    constexpr int StrideA = K;
+    constexpr int StrideB = K;
+    constexpr int StrideE = N;
+
+    for(int M : Ms)
+        this->Run(M, N, K, StrideA, StrideB, StrideE);
+}
+
+// Row Row
+TYPED_TEST(TestGemmABScale_MK_KN, SmallM)
+{
+    std::vector<int> Ms{1, 2, 3, 4, 5, 6};
+    constexpr int N = 512;
+    constexpr int K = 1024;
+
+    constexpr int StrideA = K;
+    constexpr int StrideB = N;
+    constexpr int StrideC = N;
+
+    for(int M : Ms)
+        this->Run(M, N, K, StrideA, StrideB, StrideC);
+}
+
+TYPED_TEST(TestGemmABScale_MK_KN, SmallMPadK)
+{
+    std::vector<int> Ms{1, 2, 3, 4, 5, 6};
+    constexpr int N = 512;
+    constexpr int K = 704;
+
+    constexpr int StrideA = K;
+    constexpr int StrideB = N;
+    constexpr int StrideC = N;
+
+    for(int M : Ms)
+        this->Run(M, N, K, StrideA, StrideB, StrideC);
+}
+
+TYPED_TEST(TestGemmABScale_MK_KN, MidLargeM)
+{
+    std::vector<int> Ms{127, 255, 312, 799, 1573};
+    constexpr int N = 512;
+    constexpr int K = 1024;
+
+    constexpr int StrideA = K;
+    constexpr int StrideB = N;
+    constexpr int StrideC = N;
+
+    for(int M : Ms)
+        this->Run(M, N, K, StrideA, StrideB, StrideC);
+}
+
+TYPED_TEST(TestGemmABScale_MK_KN, Regular)
+{
+    std::vector<int> Ms{512};
+    constexpr int N = 512;
+    constexpr int K = 1024;
+
+    constexpr int StrideA = K;
+    constexpr int StrideB = N;
+    constexpr int StrideE = N;
+
+    for(int M : Ms)
+        this->Run(M, N, K, StrideA, StrideB, StrideE);
+}
+
+// Col Row
+TYPED_TEST(TestGemmABScale_KM_KN, SmallM)
+{
+    std::vector<int> Ms{16, 32};
+    constexpr int N = 512;
+    constexpr int K = 1024;
+
+    constexpr int StrideB = N;
+    constexpr int StrideC = N;
+
+    for(int M : Ms)
+    {
+        int StrideA = M;
+        this->Run(M, N, K, StrideA, StrideB, StrideC);
+    }
+}
+
+TYPED_TEST(TestGemmABScale_KM_KN, SmallMPadK)
+{
+    std::vector<int> Ms{16, 32};
+    constexpr int N = 512;
+    constexpr int K = 704;
+
+    constexpr int StrideB = N;
+    constexpr int StrideC = N;
+
+    for(int M : Ms)
+    {
+        int StrideA = M;
+        this->Run(M, N, K, StrideA, StrideB, StrideC);
+    }
+}
+
+TYPED_TEST(TestGemmABScale_KM_KN, MidLargeM)
+{
+    std::vector<int> Ms{128, 256};
+    constexpr int N = 512;
+    constexpr int K = 1024;
+
+    constexpr int StrideB = N;
+    constexpr int StrideC = N;
+
+    for(int M : Ms)
+    {
+        int StrideA = M;
+        this->Run(M, N, K, StrideA, StrideB, StrideC);
+    }
+}
+
+TYPED_TEST(TestGemmABScale_KM_KN, Regular)
+{
+    std::vector<int> Ms{512};
+    constexpr int N = 512;
+    constexpr int K = 1024;
+
+    constexpr int StrideB = N;
+    constexpr int StrideE = N;
+
+    for(int M : Ms)
+    {
+        int StrideA = M;
+        this->Run(M, N, K, StrideA, StrideB, StrideE);
+    }
+}

test/gemm_ab_scale/test_gemm_ab_scale_util.hpp

@@ -0,0 +1,102 @@
+// Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
+// SPDX-License-Identifier: MIT
+
+#pragma once
+
+#include <string>
+#include <sstream>
+#include <tuple>
+#include <vector>
+#include <gtest/gtest.h>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "include/ck/utility/data_type.hpp"
+#include "profiler/profile_gemm_ab_scale_impl.hpp"
+
+namespace ck {
+namespace test {
+
+template <typename Tuple>
+class TestGemmABScale : public testing::Test
+{
+    using F32 = float;
+
+    protected:
+    using ALayout         = std::tuple_element_t<0, Tuple>;
+    using BLayout         = std::tuple_element_t<1, Tuple>;
+    using ELayout         = std::tuple_element_t<2, Tuple>;
+    using A0DataType      = std::tuple_element_t<3, Tuple>;
+    using A1DataType      = std::tuple_element_t<4, Tuple>;
+    using B0DataType      = std::tuple_element_t<5, Tuple>;
+    using B1DataType      = std::tuple_element_t<6, Tuple>;
+    using ComputeDataType = std::tuple_element_t<7, Tuple>;
+    using EDataType       = std::tuple_element_t<8, Tuple>;
+
+    public:
+    static constexpr ck::index_t ScaleBlockM = 1;
+    static constexpr ck::index_t ScaleBlockN = 128;
+    static constexpr ck::index_t ScaleBlockK = 128;
+    static constexpr bool verify_            = true;
+    static constexpr int init_method_        = 1;
+    static constexpr bool log_               = false;
+    static constexpr bool bench_             = false;
+    std::vector<int> k_batches_;
+
+    void SetUp() override { k_batches_ = {1, 2}; }
+
+    void Run(const int M,
+             const int N,
+             const int K,
+             const int StrideA,
+             const int StrideB,
+             const int StrideE)
+    {
+        for(auto kb : k_batches_)
+        {
+            RunSingle(M, N, K, StrideA, StrideB, StrideE, kb);
+        }
+    }
+
+    void RunSingle(const int M,
+                   const int N,
+                   const int K,
+                   const int StrideA,
+                   const int StrideB,
+                   const int StrideE,
+                   int kbatch   = 1,
+                   int n_warmup = 1,
+                   int n_iter   = 10)
+    {
+        bool pass = ck::profiler::profile_gemm_ab_scale_impl<A0DataType,
+                                                             A1DataType,
+                                                             B0DataType,
+                                                             B1DataType,
+                                                             ComputeDataType,
+                                                             F32,
+                                                             EDataType,
+                                                             ScaleBlockM,
+                                                             ScaleBlockN,
+                                                             ScaleBlockK,
+                                                             ALayout,
+                                                             BLayout,
+                                                             ELayout>(verify_,
+                                                                      init_method_,
+                                                                      log_,
+                                                                      bench_,
+                                                                      M,
+                                                                      N,
+                                                                      K,
+                                                                      StrideA,
+                                                                      StrideB,
+                                                                      StrideE,
+                                                                      kbatch,
+                                                                      n_warmup,
+                                                                      n_iter);
+        EXPECT_TRUE(pass);
+    }
+};
+
+} // namespace test
+} // namespace ck

test/gemm_blockscale_wp/CMakeLists.txt

@@ -2,8 +2,8 @@
 # SPDX-License-Identifier: MIT
 
 if(GPU_TARGETS MATCHES "gfx9[45]|gfx12")
-    add_gtest_executable(test_gemm_blockscale_wp_xdl_fp8 test_gemm_blockscale_wp_xdl_fp8.cpp)
+    add_gtest_executable(test_gemm_blockscale_wp_fp8 test_gemm_blockscale_wp_fp8.cpp)
     if(result EQUAL 0)
-        target_link_libraries(test_gemm_blockscale_wp_xdl_fp8 PRIVATE utility device_gemm_blockscale_wp_instance)
+        target_link_libraries(test_gemm_blockscale_wp_fp8 PRIVATE utility device_gemm_blockscale_wp_instance)
     endif()
 endif()