Merge branch 'amd-develop' into amd-master

library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "device_grouped_conv2d_bwd_data_xdl_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+// Compilation parameters for out[n, hi, wi, g, c] * wei[g, k, y, x, c] = in[n, ho, wo, g, k]
+void add_device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f32_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
+                                                                  NHWGK,
+                                                                  GKYXC,
+                                                                  Empty_Tuple,
+                                                                  NHWGC,
+                                                                  F32,
+                                                                  F32,
+                                                                  Empty_Tuple,
+                                                                  F32,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances)
+{
+    // 1. Default
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv2d_bwd_data_xdl_f32_instances<NHWGK,
+                                                         GKYXC,
+                                                         Empty_Tuple,
+                                                         NHWGC,
+                                                         ConvBwdDataDefault>{});
+    // 2. Filter1x1Stride1Pad0
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv2d_bwd_data_xdl_f32_instances<NHWGK,
+                                                         GKYXC,
+                                                         Empty_Tuple,
+                                                         NHWGC,
+                                                         ConvBwdDataFilter1x1Stride1Pad0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_instance.hpp

 // SPDX-License-Identifier: MIT
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
-#include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
 #include "device_grouped_conv2d_fwd_common.hpp"
 
 namespace ck {

library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_dl_int8_instance.hpp

@@ -4,7 +4,7 @@
 #pragma once
 
 #include "conv2d_quantization_common.hpp"
-#include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
 
 namespace ck {
 namespace tensor_operation {

profiler/README.md

@@ -102,4 +102,42 @@ arg.b_grid_desc_k0_n0_n1_k1_{2048, 4096, 2}
 arg.e_grid_desc_m_n_{ 4096, 4096}
 ....
 Best Perf: 58.0306 ms, 37.8942 TFlops, 27.7545 GB/s
+## Profile grouped convolution backward data kernels
+```bash
+# arg1: tensor operation (grouped_conv_bwd_data: Grouped Convolution Backward Data)
+# arg2: data type (0: Output fp32, Weight fp32, Input fp32
+#                  1: Output fp16, Weight fp16, Input fp16
+#                  2: Output bf16, Weight bf16, Input bf16
+# arg3: tensor layout (0: Output[G, N, Hi, Wi, C], Weight[G, K, Y, X, C], Input[G, N, Ho, Wo, K]
+#                      1: Output[N, Hi, Wi, G, C], Weight[G, K, Y, X, C], Input[N, Ho, Wo, G, K])
+# arg4: verification (0: no, 1: yes)
+# arg5: initialization (0: no init, 1: integer value, 2: decimal value)
+# arg6: print tensor value (0: no; 1: yes)
+# arg7: time kernel (0: no, 1: yes)
+# Following arguments (depending on number of spatial dims):
+#  Number of spatial dimensions (1=Conv1d, 2=Conv2d, 3=Conv3d)
+#  G, N, K, C, 
+#  <filter spatial dimensions>, (ie Y, X for 2D)
+#  <input image spatial dimensions>, (ie Hi, Wi for 2D)
+#  <strides>, (ie Sy, Sx for 2D)
+#  <dilations>, (ie Dy, Dx for 2D)
+#  <left padding>, (ie LeftPy, LeftPx for 2D)
+#  <right padding>, (ie RightPy, RightPx for 2D)
+
+ ################                   op   datatype  layout  verify  init  log  time  Ndims  G  N   K   C  Y  X  Hi  Wi  Sy  Sx  Dy  Dx  LeftPy  LeftPx  RightPy  RightPx
+./bin/ckProfiler grouped_conv_bwd_data          1       0       1     1    0     1      2 32  4 192 192  3  3  28  28   1   1   1   1       1       1        1        1
+
+ ```
+
+Result (MI100, FP16, GNHWC_GKYXC_GNHWK)
+```
+out: dim 5, lengths {32, 4, 192, 28, 28}, strides {602112, 150528, 1, 5376, 192}
+wei: dim 5, lengths {32, 192, 192, 3, 3}, strides {331776, 1728, 1, 576, 192}
+in: dim 5, lengths {32, 4, 192, 28, 28}, strides {602112, 150528, 1, 5376, 192}
+....
+Best configuration parameters:
+name: DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<256, 128, 256, 32, 8, 2, Default, 32, 32, 2, 4, 8, 4, 1, 1>
+avg_time: 0.768321
+tflops: 86.6679
+GB/s: 127.947
 ```

profiler/include/profiler/profile_grouped_conv_bwd_data_impl.hpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iostream>
+#include <numeric>
+#include <initializer_list>
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/tensor_operation/gpu/device/device_grouped_conv_bwd_data_multiple_d.hpp"
+
+#include "ck/library/utility/check_err.hpp"
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/convolution_parameter.hpp"
+#include "ck/library/utility/convolution_host_tensor_descriptor_helper.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_conv_bwd_data.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_data.hpp"
+
+namespace ck {
+namespace profiler {
+
+template <ck::index_t NDimSpatial,
+          typename OutLayout,
+          typename WeiLayout,
+          typename InLayout,
+          typename OutDataType,
+          typename WeiDataType,
+          typename InDataType>
+bool profile_grouped_conv_bwd_data_impl(int do_verification,
+                                        int init_method,
+                                        bool do_log,
+                                        bool time_kernel,
+                                        const ck::utils::conv::ConvParam& conv_param)
+{
+    using OutElementOp = ck::tensor_operation::element_wise::PassThrough;
+    using WeiElementOp = ck::tensor_operation::element_wise::PassThrough;
+    using InElementOp  = ck::tensor_operation::element_wise::PassThrough;
+
+    const auto out_element_op = OutElementOp{};
+    const auto wei_element_op = WeiElementOp{};
+    const auto in_element_op  = InElementOp{};
+
+    const auto out_g_n_k_wos_desc =
+        ck::utils::conv::make_output_host_tensor_descriptor_g_n_k_wos_packed<OutLayout>(conv_param);
+
+    const auto wei_g_k_c_xs_desc =
+        ck::utils::conv::make_weight_host_tensor_descriptor_g_k_c_xs_packed<WeiLayout>(conv_param);
+
+    const auto in_g_n_c_wis_desc =
+        ck::utils::conv::make_input_host_tensor_descriptor_g_n_c_wis_packed<InLayout>(conv_param);
+
+    Tensor<OutDataType> out(out_g_n_k_wos_desc);
+    Tensor<WeiDataType> wei(wei_g_k_c_xs_desc);
+    Tensor<InDataType> in_host(in_g_n_c_wis_desc);
+    Tensor<InDataType> in_device(in_g_n_c_wis_desc);
+
+    std::cout << "out: " << out.mDesc << std::endl;
+    std::cout << "wei: " << wei.mDesc << std::endl;
+    std::cout << "in: " << in_host.mDesc << std::endl;
+
+    switch(init_method)
+    {
+    case 0: break;
+    case 1:
+        out.GenerateTensorValue(GeneratorTensor_2<OutDataType>{-5, 5});
+        wei.GenerateTensorValue(GeneratorTensor_2<WeiDataType>{-5, 5});
+        break;
+    case 2:
+        out.GenerateTensorValue(GeneratorTensor_3<OutDataType>{0.0, 1.0});
+        wei.GenerateTensorValue(GeneratorTensor_3<WeiDataType>{-0.5, 0.5});
+        break;
+    default:
+        out.GenerateTensorValue(GeneratorTensor_1<OutDataType>{1});
+        wei.GenerateTensorValue(GeneratorTensor_1<WeiDataType>{1});
+    }
+
+    DeviceMem out_device_buf(sizeof(OutDataType) * out.mDesc.GetElementSpaceSize());
+    DeviceMem wei_device_buf(sizeof(WeiDataType) * wei.mDesc.GetElementSpaceSize());
+    DeviceMem in_device_buf(sizeof(InDataType) * in_device.mDesc.GetElementSpaceSize());
+
+    out_device_buf.ToDevice(out.mData.data());
+    wei_device_buf.ToDevice(wei.mData.data());
+
+    // reset input to zero
+    in_device_buf.SetZero();
+
+    if(do_verification)
+    {
+        auto ref_conv = ck::tensor_operation::host::ReferenceConvBwdData<NDimSpatial,
+                                                                         InDataType,
+                                                                         WeiDataType,
+                                                                         OutDataType,
+                                                                         InElementOp,
+                                                                         WeiElementOp,
+                                                                         OutElementOp>();
+
+        auto ref_invoker = ref_conv.MakeInvoker();
+
+        in_host.SetZero();
+
+        auto ref_argument = ref_conv.MakeArgument(in_host,
+                                                  wei,
+                                                  out,
+                                                  conv_param.conv_filter_strides_,
+                                                  conv_param.conv_filter_dilations_,
+                                                  conv_param.input_left_pads_,
+                                                  conv_param.input_right_pads_,
+                                                  out_element_op,
+                                                  wei_element_op,
+                                                  in_element_op);
+
+        ref_invoker.Run(ref_argument);
+    }
+
+    std::string best_op_name;
+    float best_avg_time   = 0;
+    float best_tflops     = 0;
+    float best_gb_per_sec = 0;
+
+    // profile device op instances
+    bool pass = true;
+
+    auto run_impl = [&](auto& op_ptr, auto& argument_ptr) {
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            // re-init output to zero before profiling next kernel
+            in_device_buf.SetZero();
+
+            std::string op_name = op_ptr->GetTypeString();
+
+            auto invoker_ptr = op_ptr->MakeInvokerPointer();
+
+            float avg_time =
+                invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, time_kernel});
+
+            std::size_t flop      = conv_param.GetFlops();
+            std::size_t num_btype = conv_param.GetByte<InDataType, WeiDataType, OutDataType>();
+
+            float tflops = static_cast<float>(flop) / 1.E9 / avg_time;
+
+            float gb_per_sec = num_btype / 1.E6 / avg_time;
+
+            std::cout << "Perf: " << std::setw(10) << avg_time << " ms, " << tflops << " TFlops, "
+                      << gb_per_sec << " GB/s, " << op_name << std::endl;
+
+            if(tflops > best_tflops)
+            {
+                best_op_name    = op_name;
+                best_tflops     = tflops;
+                best_avg_time   = avg_time;
+                best_gb_per_sec = gb_per_sec;
+            }
+
+            if(do_verification)
+            {
+                in_device_buf.FromDevice(in_device.mData.data());
+
+                pass = pass & ck::utils::check_err(in_device, in_host);
+
+                if(do_log)
+                {
+                    LogRangeAsType<float>(std::cout << "output : ", out.mData, ",") << std::endl;
+                    LogRangeAsType<float>(std::cout << "weight: ", wei.mData, ",") << std::endl;
+                    LogRangeAsType<float>(std::cout << "in_host  : ", in_host.mData, ",")
+                        << std::endl;
+                    LogRangeAsType<float>(std::cout << "in_device: ", in_device.mData, ",")
+                        << std::endl;
+                }
+            }
+        }
+        else
+        {
+            std::cout << op_ptr->GetTypeString() << " does not support this problem" << std::endl;
+        }
+    };
+
+    // do GEMM
+    using DeviceOp = ck::tensor_operation::device::DeviceGroupedConvBwdDataMultipleD<NDimSpatial,
+                                                                                     OutLayout,
+                                                                                     WeiLayout,
+                                                                                     ck::Tuple<>,
+                                                                                     InLayout,
+                                                                                     OutDataType,
+                                                                                     WeiDataType,
+                                                                                     ck::Tuple<>,
+                                                                                     InDataType,
+                                                                                     OutElementOp,
+                                                                                     WeiElementOp,
+                                                                                     InElementOp>;
+
+    // get device op instances
+    const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+        DeviceOp>::GetInstances();
+
+    std::array<ck::index_t, NDimSpatial + 3> out_lengths{};
+    std::array<ck::index_t, NDimSpatial + 3> out_strides{};
+    std::array<ck::index_t, NDimSpatial + 3> wei_lengths{};
+    std::array<ck::index_t, NDimSpatial + 3> wei_strides{};
+    std::array<ck::index_t, NDimSpatial + 3> in_lengths{};
+    std::array<ck::index_t, NDimSpatial + 3> in_strides{};
+    std::array<ck::index_t, NDimSpatial> conv_filter_strides{};
+    std::array<ck::index_t, NDimSpatial> conv_filter_dilations{};
+    std::array<ck::index_t, NDimSpatial> input_left_pads{};
+    std::array<ck::index_t, NDimSpatial> input_right_pads{};
+
+    auto copy = [](const auto& x, auto& y) { ck::ranges::copy(x, y.begin()); };
+
+    copy(out_g_n_k_wos_desc.GetLengths(), out_lengths);
+    copy(out_g_n_k_wos_desc.GetStrides(), out_strides);
+    copy(wei_g_k_c_xs_desc.GetLengths(), wei_lengths);
+    copy(wei_g_k_c_xs_desc.GetStrides(), wei_strides);
+    copy(in_g_n_c_wis_desc.GetLengths(), in_lengths);
+    copy(in_g_n_c_wis_desc.GetStrides(), in_strides);
+    copy(conv_param.conv_filter_strides_, conv_filter_strides);
+    copy(conv_param.conv_filter_dilations_, conv_filter_dilations);
+    copy(conv_param.input_left_pads_, input_left_pads);
+    copy(conv_param.input_right_pads_, input_right_pads);
+
+    for(auto& op_ptr : op_ptrs)
+    {
+        auto argument_ptr =
+            op_ptr->MakeArgumentPointer(static_cast<OutDataType*>(out_device_buf.GetDeviceBuffer()),
+                                        static_cast<WeiDataType*>(wei_device_buf.GetDeviceBuffer()),
+                                        {},
+                                        static_cast<InDataType*>(in_device_buf.GetDeviceBuffer()),
+                                        out_lengths,
+                                        out_strides,
+                                        wei_lengths,
+                                        wei_strides,
+                                        {},
+                                        {},
+                                        in_lengths,
+                                        in_strides,
+                                        conv_filter_strides,
+                                        conv_filter_dilations,
+                                        input_left_pads,
+                                        input_right_pads,
+                                        out_element_op,
+                                        wei_element_op,
+                                        in_element_op);
+
+        run_impl(op_ptr, argument_ptr);
+    }
+
+    std::cout << "Best configuration parameters:"
+              << "\nname: " << best_op_name << "\navg_time: " << best_avg_time
+              << "\ntflops: " << best_tflops << "\nGB/s: " << best_gb_per_sec << std::endl;
+
+    return pass;
+}
+
+} // namespace profiler
+} // namespace ck

profiler/src/CMakeLists.txt

@@ -35,6 +35,7 @@ set(PROFILER_SOURCES
     profile_contraction_bilinear.cpp
     profile_contraction_scale.cpp
     profile_batched_gemm_multi_d.cpp
+    profile_grouped_conv_bwd_data.cpp
 )
 
 set(PROFILER_EXECUTABLE ckProfiler)
@@ -79,4 +80,5 @@ target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_contraction_bilinear
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_contraction_scale_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_pool_fwd_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_batched_gemm_multi_d_instance)
+target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv2d_bwd_data_instance)
 rocm_install(TARGETS ${PROFILER_EXECUTABLE} COMPONENT profiler)

profiler/src/profile_batchnorm_fwd.cpp

@@ -148,7 +148,7 @@ int profile_batchnorm_forward(int argc, char* argv[])
     {
         if(arg_parser.inLengths.size() == 4 && arg_parser.reduceDims.size() == 3)
         {
-            profile_batchnorm_forward_impl<F16, F16, F32, F16, F16, F16, 4, 3>(
+            profile_batchnorm_forward_impl<F16, F16, F32, F16, F16, F32, 4, 3>(
                 arg_parser.do_verification,
                 arg_parser.init_method,
                 arg_parser.do_dumpout,

profiler/src/profile_grouped_conv_bwd_data.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iostream>
+#include <numeric>
+#include <initializer_list>
+#include <cstdlib>
+
+#include "profiler/profile_grouped_conv_bwd_data_impl.hpp"
+#include "profiler_operation_registry.hpp"
+
+namespace {
+
+enum struct ConvLayout
+{
+    GNHWC_GKYXC_GNHWK, // 0
+    NHWGC_GKYXC_NHWGK, // 1
+};
+
+enum struct ConvDataType
+{
+    F32_F32_F32,    // 0
+    F16_F16_F16,    // 1
+    BF16_BF16_BF16, // 2
+};
+
+#define OP_NAME "grouped_conv_bwd_data"
+#define OP_DESC "Grouped Convolution Backward Data"
+
+static void print_helper_msg()
+{
+    std::cout
+        // clang-format off
+        << "arg1: tensor operation (" OP_NAME ": " OP_DESC ")\n"
+        << "arg2: data type (0: Output fp32, Weight fp32, Input fp32\n"
+        << "                 1: Output fp16, Weight fp16, Input fp16\n"
+        << "                 2: Output bf16, Weight bf16, Input bf16\n"
+        << "arg3: tensor layout (0: Output[G, N, Hi, Wi, C], Weight[G, K, Y, X, C], Input[G, N, Ho, Wo, K]\n"
+        << "                     1: Output[N, Hi, Wi, G, C], Weight[G, K, Y, X, C], Input[N, Ho, Wo, G, K])\n"
+        << "arg4: verification (0: no, 1: yes)\n"
+        << "arg5: initialization (0: no init, 1: integer value, 2: decimal value)\n"
+        << "arg6: print tensor value (0: no; 1: yes)\n"
+        << "arg7: time kernel (0: no, 1: yes)\n"
+        << ck::utils::conv::get_conv_param_parser_helper_msg() << std::endl;
+    // clang-format on
+}
+
+} // namespace
+
+int profile_grouped_conv_bwd_data(int argc, char* argv[])
+{
+    // 8 for control, 1 for num_dim_spatial
+    if(argc < 9)
+    {
+        print_helper_msg();
+        return 1;
+    }
+
+    const auto data_type       = static_cast<ConvDataType>(std::stoi(argv[2]));
+    const auto layout          = static_cast<ConvLayout>(std::stoi(argv[3]));
+    const bool do_verification = std::stoi(argv[4]);
+    const int init_method      = std::stoi(argv[5]);
+    const bool do_log          = std::stoi(argv[6]);
+    const bool time_kernel     = std::stoi(argv[7]);
+    const int num_dim_spatial  = std::stoi(argv[8]);
+
+    // 8 for control, 1 for num_dim_spatial, 4 for G/N/K/C, and 6 * num_dim_spatial
+    if(argc != 8 + 1 + 4 + 6 * num_dim_spatial)
+    {
+        print_helper_msg();
+        return 1;
+    }
+
+    const auto params = ck::utils::conv::parse_conv_param(num_dim_spatial, 9, argv);
+
+    using F32  = float;
+    using F16  = ck::half_t;
+    using BF16 = ck::bhalf_t;
+
+    using GNHWC = ck::tensor_layout::convolution::GNHWC;
+    using NHWGC = ck::tensor_layout::convolution::NHWGC;
+
+    using GKYXC = ck::tensor_layout::convolution::GKYXC;
+
+    using GNHWK = ck::tensor_layout::convolution::GNHWK;
+    using NHWGK = ck::tensor_layout::convolution::NHWGK;
+
+    constexpr auto I2 = ck::Number<2>{};
+
+    auto profile = [&](auto num_dim_spatial_tmp,
+                       auto out_layout,
+                       auto wei_layout,
+                       auto in_layout,
+                       auto wei_type,
+                       auto out_type,
+                       auto in_type) {
+        constexpr ck::index_t NDimSpatial = num_dim_spatial_tmp.value;
+
+        using OutLayout = decltype(out_layout);
+        using WeiLayout = decltype(wei_layout);
+        using InLayout  = decltype(in_layout);
+
+        using OutDataType = decltype(out_type);
+        using WeiDataType = decltype(wei_type);
+        using InDataType  = decltype(in_type);
+
+        bool pass = ck::profiler::profile_grouped_conv_bwd_data_impl<NDimSpatial,
+                                                                     OutLayout,
+                                                                     WeiLayout,
+                                                                     InLayout,
+                                                                     OutDataType,
+                                                                     WeiDataType,
+                                                                     InDataType>(
+            do_verification, init_method, do_log, time_kernel, params);
+
+        return pass ? 0 : 1;
+    };
+
+    // GNHWC_GKYXC_GNHWK
+    if(num_dim_spatial == 2 && layout == ConvLayout::GNHWC_GKYXC_GNHWK)
+    {
+        if(data_type == ConvDataType::F32_F32_F32)
+        {
+            return profile(I2, GNHWK{}, GKYXC{}, GNHWC{}, F32{}, F32{}, F32{});
+        }
+        else if(data_type == ConvDataType::F16_F16_F16)
+        {
+            return profile(I2, GNHWK{}, GKYXC{}, GNHWC{}, F16{}, F16{}, F16{});
+        }
+        else if(data_type == ConvDataType::BF16_BF16_BF16)
+        {
+            return profile(I2, GNHWK{}, GKYXC{}, GNHWC{}, BF16{}, BF16{}, BF16{});
+        }
+    }
+    // NHWGC_GKYXC_NHWGK
+    else if(num_dim_spatial == 2 && layout == ConvLayout::NHWGC_GKYXC_NHWGK)
+    {
+        if(data_type == ConvDataType::F32_F32_F32)
+        {
+            return profile(I2, NHWGK{}, GKYXC{}, NHWGC{}, F32{}, F32{}, F32{});
+        }
+        else if(data_type == ConvDataType::F16_F16_F16)
+        {
+            return profile(I2, NHWGK{}, GKYXC{}, NHWGC{}, F16{}, F16{}, F16{});
+        }
+        else if(data_type == ConvDataType::BF16_BF16_BF16)
+        {
+            return profile(I2, NHWGK{}, GKYXC{}, NHWGC{}, BF16{}, BF16{}, BF16{});
+        }
+    }
+
+    std::cout << "this data_type & layout is not implemented" << std::endl;
+
+    return 1;
+}
+
+REGISTER_PROFILER_OPERATION(OP_NAME, OP_DESC, profile_grouped_conv_bwd_data);

script/check_copyright_year.sh

+#!/bin/bash
+
+current_year=$(date +%Y)
+exit_code=0
+
+for file in $@; do
+    if grep -q "Copyright (c)" $file
+    then
+        if ! grep -q "Copyright (c).*$current_year" $file
+        then
+            echo "ERROR: File $file has a copyright notice without the current year ($current_year)."
+            exit_code=1
+        fi
+    fi
+done
+
+exit $exit_code

script/install_precommit.sh

+#!/bin/bash
+
+run_and_check() {
+    "$@"
+    status=$?
+    if [ $status -ne 0 ]; then
+        echo "Error with \"$@\": Exited with status $status"
+        exit $status
+    fi
+    return $status
+}
+
+echo "I: Installing tools required for pre-commit checks..."
+run_and_check apt install clang-format-10
+
+echo "I: Installing pre-commit itself..."
+run_and_check pip3 install pre-commit
+run_and_check pre-commit install
+
+echo "I: Installation successful."

script/uninstall_precommit.sh

+pre-commit uninstall

test/CMakeLists.txt

@@ -59,6 +59,7 @@ add_subdirectory(batchnorm)
 add_subdirectory(contraction)
 add_subdirectory(pool_fwd)
 add_subdirectory(batched_gemm_multi_d)
+add_subdirectory(grouped_convnd_bwd_data)
 if(GPU_TARGETS MATCHES "gfx1100")
     add_subdirectory(wmma_op)
 endif()

test/batched_gemm_multi_d/test_batched_gemm_multi_d.cpp

 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <gtest/gtest.h>

test/data_type/CMakeLists.txt

@@ -2,3 +2,6 @@ if (USE_BITINT_EXTENSION_INT4)
   add_gtest_executable(test_int4 int4.cpp)
   target_link_libraries(test_int4 PRIVATE utility)
 endif()
+
+add_gtest_executable(test_fp8 fp8.cpp)
+target_link_libraries(test_fp8 PRIVATE utility)

test/data_type/fp8.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "gtest/gtest.h"
+#include "ck/utility/data_type.hpp"
+#include "ck/utility/type_convert.hpp"
+
+using ck::f8_convert_sr;
+using ck::f8_t;
+using ck::half_t;
+using ck::type_convert;
+
+TEST(FP8, NumericLimits)
+{
+    EXPECT_EQ(ck::NumericLimits<f8_t>::Min(), 0x08);
+    EXPECT_EQ(ck::NumericLimits<f8_t>::Max(), 0x77);
+    EXPECT_EQ(ck::NumericLimits<f8_t>::Lowest(), 0xF7);
+    EXPECT_EQ(ck::NumericLimits<f8_t>::QuietNaN(), 0x80);
+}
+
+TEST(FP8, ConvertFP32Nearest)
+{
+    // fix the tolerance value
+    float abs_tol = 1e-6;
+    // convert 0 float to fp8 and back, check if holds
+    ASSERT_NEAR(0.0f, type_convert<float>(type_convert<f8_t>(0.0f)), abs_tol);
+    // convert minimal float to fp8 and back, check if holds
+    ASSERT_NEAR(std::numeric_limits<float>::min(),
+                type_convert<float>(type_convert<f8_t>(std::numeric_limits<float>::min())),
+                abs_tol);
+    // convert maximal f8_t to float and check if equal to 240.0
+    ASSERT_NEAR(240.0f, type_convert<float>(type_convert<f8_t>(240.0f)), abs_tol);
+    // convert maximal float to fp8 and back, check if clipped to 240.0
+    ASSERT_NEAR(240.0f,
+                type_convert<float>(type_convert<f8_t>(std::numeric_limits<float>::max())),
+                abs_tol);
+    // convert inf float to f8_t and check if it is qNan
+    ASSERT_NEAR(0x80, type_convert<f8_t>(std::numeric_limits<float>::infinity()), abs_tol);
+    // positive float value to fp8 and back, check if holds
+    float pos_float = 0.0078125f;
+    ASSERT_NEAR(pos_float, type_convert<float>(type_convert<f8_t>(pos_float)), abs_tol);
+    // negative float value to fp8 and back, check if holds
+    float neg_float = -0.0156250f;
+    ASSERT_NEAR(neg_float, type_convert<float>(type_convert<f8_t>(neg_float)), abs_tol);
+}
+
+TEST(FP8, ConvertFP32Stochastic)
+{
+    // fix the tolerance value
+    float abs_tol = 1e-6;
+    // convert 0 float to fp8 and back, check if holds
+    ASSERT_NEAR(0.0f, type_convert<float>(f8_convert_sr<f8_t>(0.0f)), abs_tol);
+    // convert minimal float to fp8 and back, check if holds
+    ASSERT_NEAR(std::numeric_limits<float>::min(),
+                type_convert<float>(f8_convert_sr<f8_t>(std::numeric_limits<float>::min())),
+                abs_tol);
+    // convert maximal f8_t to float and check if equal to 240.0
+    ASSERT_NEAR(240.0f, type_convert<float>(f8_convert_sr<f8_t>(240.0f)), abs_tol);
+    // convert maximal float to fp8 and back, check if clipped to 240.0
+    ASSERT_NEAR(240.0f,
+                type_convert<float>(f8_convert_sr<f8_t>(std::numeric_limits<float>::max())),
+                abs_tol);
+    // convert inf float to f8_t and check if it is qNan
+    ASSERT_NEAR(0x80, f8_convert_sr<f8_t>(std::numeric_limits<float>::infinity()), abs_tol);
+    // positive float value to fp8 and back, check if holds
+    float pos_float = 0.0078125f;
+    ASSERT_NEAR(pos_float, type_convert<float>(f8_convert_sr<f8_t>(pos_float)), abs_tol);
+    // negative float value to fp8 and back, check if holds
+    float neg_float = -0.0156250f;
+    ASSERT_NEAR(neg_float, type_convert<float>(f8_convert_sr<f8_t>(neg_float)), abs_tol);
+}
+
+TEST(FP8, ConvertFP16Nearest)
+{
+    // fix the tolerance value
+    float abs_tol = 1e-3;
+    // convert 0 fp16 to fp8 and back, check if holds
+    ASSERT_NEAR(half_t{0.0}, type_convert<half_t>(type_convert<f8_t>(half_t{0.0})), abs_tol);
+    // convert minimal fp16 to fp8 and back, check if holds
+    ASSERT_NEAR(ck::NumericLimits<half_t>::Min(),
+                type_convert<half_t>(type_convert<f8_t>(ck::NumericLimits<half_t>::Min())),
+                abs_tol);
+    // convert maximal f8_t to fp16 and check if equal to 240.0
+    ASSERT_NEAR(half_t{240.0}, type_convert<half_t>(type_convert<f8_t>(half_t{240.0})), abs_tol);
+    // convert maximal fp16 to fp8 and back, check if clipped to 240.0
+    ASSERT_NEAR(half_t{240.0},
+                type_convert<half_t>(type_convert<f8_t>(ck::NumericLimits<half_t>::Max())),
+                abs_tol);
+    // convert QuietNaN fp16 to f8_t and check if it is QuietNaN
+    ASSERT_NEAR(0x80, type_convert<f8_t>(ck::NumericLimits<half_t>::QuietNaN()), abs_tol);
+    // positive fp16 value to fp8 and back, check if holds
+    half_t pos_half = half_t{0.0078125};
+    ASSERT_NEAR(pos_half, type_convert<half_t>(type_convert<f8_t>(pos_half)), abs_tol);
+    // negative fp16 value to fp8 and back, check if holds
+    half_t neg_half = half_t{-0.0156250};
+    ASSERT_NEAR(neg_half, type_convert<half_t>(type_convert<f8_t>(neg_half)), abs_tol);
+}
+
+TEST(FP8, ConvertFP16Stochastic)
+{
+    // fix the tolerance value
+    float abs_tol = 1e-3;
+    // convert 0 fp16 to fp8 and back, check if holds
+    ASSERT_NEAR(half_t{0.0}, type_convert<half_t>(f8_convert_sr<f8_t>(half_t{0.0})), abs_tol);
+    // convert minimal fp16 to fp8 and back, check if holds
+    ASSERT_NEAR(ck::NumericLimits<half_t>::Min(),
+                type_convert<half_t>(f8_convert_sr<f8_t>(ck::NumericLimits<half_t>::Min())),
+                abs_tol);
+    // convert maximal f8_t to fp16 and check if equal to 240.0
+    ASSERT_NEAR(half_t{240.0}, type_convert<half_t>(f8_convert_sr<f8_t>(half_t{240.0})), abs_tol);
+    // convert maximal fp16 to fp8 and back, check if clipped to 240.0
+    ASSERT_NEAR(half_t{240.0},
+                type_convert<half_t>(f8_convert_sr<f8_t>(ck::NumericLimits<half_t>::Max())),
+                abs_tol);
+    // convert QuietNaN fp16 to f8_t and check if it is QuietNaN
+    ASSERT_NEAR(0x80, f8_convert_sr<f8_t>(ck::NumericLimits<half_t>::QuietNaN()), abs_tol);
+    // positive fp16 value to fp8 and back, check if holds
+    half_t pos_half = half_t{0.0078125};
+    ASSERT_NEAR(pos_half, type_convert<half_t>(f8_convert_sr<f8_t>(pos_half)), abs_tol);
+    // negative fp16 value to fp8 and back, check if holds
+    half_t neg_half = half_t{-0.0156250};
+    ASSERT_NEAR(neg_half, type_convert<half_t>(f8_convert_sr<f8_t>(neg_half)), abs_tol);
+}

test/gemm/instance/gemm_wavelet_f16_tn_instance.cpp

@@ -5,7 +5,7 @@
 
 #include "ck/ck.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-#include "ck/tensor_operation/gpu/device/device_gemm_xdl_waveletmodel_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl_waveletmodel_cshuffle.hpp"
 #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"

test/grouped_convnd_bwd_data/CMakeLists.txt

+if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS MATCHES "gfx940")
+    add_gtest_executable(test_grouped_convnd_bwd_data test_grouped_convnd_bwd_data.cpp)
+    target_link_libraries(test_grouped_convnd_bwd_data PRIVATE utility device_grouped_conv2d_bwd_data_instance)
+    add_gtest_executable(test_grouped_convnd_bwd_data_interface test_grouped_convnd_bwd_data_interface.cpp)
+    target_link_libraries(test_grouped_convnd_bwd_data_interface PRIVATE utility device_grouped_conv2d_bwd_data_instance)
+endif()
\ No newline at end of file

test/grouped_convnd_bwd_data/test_grouped_convnd_bwd_data.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+#include <iostream>
+#include <initializer_list>
+#include <tuple>
+#include <vector>
+
+#include <gtest/gtest.h>
+
+#include "profiler/profile_grouped_conv_bwd_data_impl.hpp"
+
+template <typename Tuple>
+class TestGroupedConvndBwdData : public ::testing::Test
+{
+    protected:
+    using DataType  = std::tuple_element_t<0, Tuple>;
+    using OutLayout = std::tuple_element_t<1, Tuple>;
+    using WeiLayout = std::tuple_element_t<2, Tuple>;
+    using InLayout  = std::tuple_element_t<3, Tuple>;
+
+    std::vector<ck::utils::conv::ConvParam> conv_params;
+
+    template <ck::index_t NDimSpatial>
+    void Run()
+    {
+        EXPECT_FALSE(conv_params.empty());
+        bool pass = true;
+        for(auto& param : conv_params)
+        {
+            pass = pass && ck::profiler::profile_grouped_conv_bwd_data_impl<NDimSpatial,
+                                                                            OutLayout,
+                                                                            WeiLayout,
+                                                                            InLayout,
+                                                                            DataType,
+                                                                            DataType,
+                                                                            DataType>(
+                               true,  // do_verification
+                               1,     // init_method: integer value
+                               false, // do_log
+                               false, // time_kernel
+                               param);
+        }
+        EXPECT_TRUE(pass);
+    }
+};
+
+using GNHWC = ck::tensor_layout::convolution::GNHWC;
+using NHWGC = ck::tensor_layout::convolution::NHWGC;
+
+using GKYXC = ck::tensor_layout::convolution::GKYXC;
+
+using GNHWK = ck::tensor_layout::convolution::GNHWK;
+using NHWGK = ck::tensor_layout::convolution::NHWGK;
+
+using KernelTypes = ::testing::Types<std::tuple<float, GNHWK, GKYXC, GNHWC>,
+                                     std::tuple<ck::half_t, GNHWK, GKYXC, GNHWC>,
+                                     std::tuple<ck::bhalf_t, GNHWK, GKYXC, GNHWC>,
+                                     std::tuple<float, NHWGK, GKYXC, NHWGC>,
+                                     std::tuple<ck::half_t, NHWGK, GKYXC, NHWGC>,
+                                     std::tuple<ck::bhalf_t, NHWGK, GKYXC, NHWGC>>;
+TYPED_TEST_SUITE(TestGroupedConvndBwdData, KernelTypes);
+
+TYPED_TEST(TestGroupedConvndBwdData, Test2D)
+{
+    this->conv_params.clear();
+
+    this->conv_params.push_back(
+        {2, 2, 4, 192, 192, {3, 3}, {28, 28}, {1, 1}, {1, 1}, {1, 1}, {1, 1}});
+    this->conv_params.push_back(
+        {2, 2, 128, 128, 256, {3, 3}, {14, 14}, {1, 1}, {1, 1}, {1, 1}, {1, 1}});
+    this->conv_params.push_back(
+        {2, 2, 128, 128, 256, {1, 1}, {7, 7}, {2, 2}, {1, 1}, {0, 0}, {0, 0}});
+    this->conv_params.push_back(
+        {2, 2, 128, 128, 256, {1, 1}, {3, 3}, {1, 1}, {1, 1}, {0, 0}, {0, 0}});
+    this->template Run<2>();
+}

test/grouped_convnd_bwd_data/test_grouped_convnd_bwd_data_interface.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+#include <iostream>
+#include <initializer_list>
+#include <tuple>
+#include <vector>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/convolution_backward_data_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp"
+
+#include "ck/library/utility/convolution_parameter.hpp"
+#include "ck/library/utility/algorithm.hpp"
+#include "ck/library/utility/convolution_host_tensor_descriptor_helper.hpp"
+
+#include <gtest/gtest.h>
+
+using DataType    = ck::half_t;
+using AccDataType = float;
+using Pass        = ck::tensor_operation::element_wise::PassThrough;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+using ConvBackwardDataSpecialization =
+    ck::tensor_operation::device::ConvolutionBackwardDataSpecialization;
+
+static constexpr auto ConvBwdDataDefault   = ConvBackwardDataSpecialization::Default;
+static constexpr auto Filter1x1Stride1Pad0 = ConvBackwardDataSpecialization::Filter1x1Stride1Pad0;
+
+template <typename Tuple, ConvBackwardDataSpecialization ConvSpec>
+class TestGroupedConvndBwdData : public ::testing::Test
+{
+    protected:
+    static constexpr ck::index_t NDimSpatial = 2;
+
+    using OutLayout = std::tuple_element_t<0, Tuple>;
+    using WeiLayout = std::tuple_element_t<1, Tuple>;
+    using InLayout  = std::tuple_element_t<2, Tuple>;
+
+    // clang-format off
+    using GroupedConvBwdDataDeviceInstance = ck::tensor_operation::device::DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1
+    // ######| NDimSpatial|   ALayout|   BLayout|   DsLayout|  ELayout|       AData|       BData|     AccData|         CShuffle|       DsData|      EData| AElementwise| BElementwise| CDEElementwise| ConvolutionBackward| DoPad| DoPad|      NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer|    MXdl|    NXdl|    ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|    BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffleMXdl| CShuffleNXdl|   CDEBlockTransfer| CDEBlockTransfer|
+    // ######|            |          |          |           |         |        Type|        Type|        Type|         DataType|         Type|       Type|    Operation|    Operation|      Operation|  DataSpecialization| GemmM| GemmN| PrefetchStage|  Size| Block| Block| Block|    |    |  XDL|  XDL| PerWave| PerWave|     ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar|    ExtraM|     ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar|    ExtraN|      PerWave|      PerWave|  _MBlock_MPerBlock|  ScalarPerVector|
+    // ######|            |          |          |           |         |            |            |            |                 |             |           |             |             |               |                    |      |      |              |      |      |      |      |    |    |     |     |        |        | Lengths_AK0_M_AK1|   ArrangeOrder|               |               |      PerVector|  PerVector_AK1|          | Lengths_BK0_N_BK1|   ArrangeOrder|               |               |      PerVector|  PerVector_BK1|          |   PerShuffle|   PerShuffle|  _NBlock_NPerBlock|       _NPerBlock|
+    // ######|            |          |          |           |         |            |            |            |                 |             |           |             |             |               |                    |      |      |              |      |      |      |      |    |    |     |     |        |        |                  |               |               |               |               |               |          |                  |               |               |               |               |               |          |             |             |                   |                 |
+    < NDimSpatial, OutLayout, WeiLayout, ck::Tuple<>, InLayout, DataType, DataType, AccDataType, DataType,   ck::Tuple<>, DataType, Pass, Pass,    Pass,  ConvSpec,  true,  true,             1,   256,   128,   256,    32,   8,   2,   32,   32,       2,       4,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,            1,            1,     S<1, 32, 1, 8>,                8>;
+    // clang-format on
+
+    ck::utils::conv::ConvParam conv_param;
+
+    template <ck::index_t NDimSpatial>
+    bool Run()
+    {
+
+        const auto out_g_n_k_wos_desc =
+            ck::utils::conv::make_output_host_tensor_descriptor_g_n_k_wos_packed<OutLayout>(
+                conv_param);
+
+        const auto wei_g_k_c_xs_desc =
+            ck::utils::conv::make_weight_host_tensor_descriptor_g_k_c_xs_packed<WeiLayout>(
+                conv_param);
+
+        const auto in_g_n_c_wis_desc =
+            ck::utils::conv::make_input_host_tensor_descriptor_g_n_c_wis_packed<InLayout>(
+                conv_param);
+
+        std::array<ck::index_t, NDimSpatial + 3> out_lengths{};
+        std::array<ck::index_t, NDimSpatial + 3> out_strides{};
+        std::array<ck::index_t, NDimSpatial + 3> wei_lengths{};
+        std::array<ck::index_t, NDimSpatial + 3> wei_strides{};
+        std::array<ck::index_t, NDimSpatial + 3> in_lengths{};
+        std::array<ck::index_t, NDimSpatial + 3> in_strides{};
+        std::array<ck::index_t, NDimSpatial> conv_filter_strides{};
+        std::array<ck::index_t, NDimSpatial> conv_filter_dilations{};
+        std::array<ck::index_t, NDimSpatial> input_left_pads{};
+        std::array<ck::index_t, NDimSpatial> input_right_pads{};
+
+        auto copy = [](const auto& x, auto& y) { ck::ranges::copy(x, y.begin()); };
+
+        copy(out_g_n_k_wos_desc.GetLengths(), out_lengths);
+        copy(out_g_n_k_wos_desc.GetStrides(), out_strides);
+        copy(wei_g_k_c_xs_desc.GetLengths(), wei_lengths);
+        copy(wei_g_k_c_xs_desc.GetStrides(), wei_strides);
+        copy(in_g_n_c_wis_desc.GetLengths(), in_lengths);
+        copy(in_g_n_c_wis_desc.GetStrides(), in_strides);
+        copy(conv_param.conv_filter_strides_, conv_filter_strides);
+        copy(conv_param.conv_filter_dilations_, conv_filter_dilations);
+        copy(conv_param.input_left_pads_, input_left_pads);
+        copy(conv_param.input_right_pads_, input_right_pads);
+
+        auto conv = GroupedConvBwdDataDeviceInstance{};
+
+        auto argument = conv.MakeArgument(nullptr,
+                                          nullptr,
+                                          std::array<const void*, 0>{},
+                                          nullptr,
+                                          out_lengths,
+                                          out_strides,
+                                          wei_lengths,
+                                          wei_strides,
+                                          {},
+                                          {},
+                                          in_lengths,
+                                          in_strides,
+                                          conv_filter_strides,
+                                          conv_filter_dilations,
+                                          input_left_pads,
+                                          input_right_pads,
+                                          Pass{},
+                                          Pass{},
+                                          Pass{});
+        return conv.IsSupportedArgument(argument);
+    }
+};
+
+using GNHWC = ck::tensor_layout::convolution::GNHWC;
+using NHWGC = ck::tensor_layout::convolution::NHWGC;
+
+using GKYXC = ck::tensor_layout::convolution::GKYXC;
+
+using GNHWK = ck::tensor_layout::convolution::GNHWK;
+using NHWGK = ck::tensor_layout::convolution::NHWGK;
+
+using KernelTypes =
+    ::testing::Types<std::tuple<GNHWK, GKYXC, GNHWC>, std::tuple<NHWGK, GKYXC, NHWGC>>;
+
+template <typename Tuple>
+class TestGroupedConvndBwdDataDefault : public TestGroupedConvndBwdData<Tuple, ConvBwdDataDefault>
+{
+};
+
+template <typename Tuple>
+class TestGroupedConvndBwdDataFilter1x1
+    : public TestGroupedConvndBwdData<Tuple, Filter1x1Stride1Pad0>
+{
+};
+
+TYPED_TEST_SUITE(TestGroupedConvndBwdDataDefault, KernelTypes);
+TYPED_TEST_SUITE(TestGroupedConvndBwdDataFilter1x1, KernelTypes);
+
+TYPED_TEST(TestGroupedConvndBwdDataFilter1x1, SpecializationCheck)
+{
+    // Check filter 3,3 instead of 1,1
+    this->conv_param  = {2, 2, 4, 192, 192, {3, 3}, {28, 28}, {1, 1}, {1, 1}, {0, 0}, {0, 0}};
+    bool is_supported = this->template Run<2>();
+    EXPECT_FALSE(is_supported);
+
+    // Check strides 2,2 instead of 1,1
+    this->conv_param = {2, 2, 4, 192, 192, {1, 1}, {28, 28}, {2, 2}, {1, 1}, {0, 0}, {0, 0}};
+    is_supported     = this->template Run<2>();
+    EXPECT_FALSE(is_supported);
+
+    // Check with pad
+    this->conv_param = {2, 2, 4, 192, 192, {1, 1}, {28, 28}, {1, 1}, {1, 1}, {1, 1}, {1, 1}};
+    is_supported     = this->template Run<2>();
+    EXPECT_FALSE(is_supported);
+
+    // Supported version
+    this->conv_param = {2, 2, 4, 192, 192, {1, 1}, {28, 28}, {1, 1}, {1, 1}, {0, 0}, {0, 0}};
+    is_supported     = this->template Run<2>();
+    EXPECT_TRUE(is_supported);
+}
+
+TYPED_TEST(TestGroupedConvndBwdDataDefault, VectorLoadCheck)
+{
+    // vector load for A
+    this->conv_param  = {2, 2, 128, 129, 256, {1, 1}, {7, 7}, {2, 2}, {1, 1}, {0, 0}, {0, 0}};
+    bool is_supported = this->template Run<2>();
+    EXPECT_FALSE(is_supported);
+    // vector load for B, E, Ds
+    this->conv_param = {2, 2, 128, 128, 257, {1, 1}, {7, 7}, {2, 2}, {1, 1}, {0, 0}, {0, 0}};
+    is_supported     = this->template Run<2>();
+    EXPECT_FALSE(is_supported);
+}