Merge branch 'develop' into sphinx_doc

example/09_convnd_fwd/convnd_fwd_dl_common.hpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+#include <iostream>
+#include <numeric>
+#include <type_traits>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/utility/algorithm.hpp"
+#include "ck/library/utility/check_err.hpp"
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/convolution_parameter.hpp"
+#include "ck/library/utility/convolution_host_tensor_descriptor_helper.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_conv_fwd.hpp"
+
+void print_helper_msg()
+{
+    std::cout << "arg1: verification (0=no, 1=yes)\n"
+              << "arg2: initialization (0=no init, 1=integer value, 2=decimal value)\n"
+              << "arg3: time kernel (0=no, 1=yes)\n"
+              << ck::utils::conv::get_conv_param_parser_helper_msg() << std::endl;
+}
+
+template <ck::index_t NDimSpatial,
+          typename InDataType,
+          typename WeiDataType,
+          typename DsDataType,
+          typename OutDataType,
+          typename InElementOp,
+          typename WeiElementOp,
+          typename OutElementOp,
+          typename DeviceConvNDFwdInstance>
+bool run_grouped_conv_fwd_dl(bool do_verification,
+                             int init_method,
+                             bool time_kernel,
+                             const ck::utils::conv::ConvParam& conv_param,
+                             const HostTensorDescriptor& in_g_n_c_wis_desc,
+                             const HostTensorDescriptor& wei_g_k_c_xs_desc,
+                             const HostTensorDescriptor& out_g_n_k_wos_desc,
+                             const InElementOp& in_element_op,
+                             const WeiElementOp& wei_element_op,
+                             const OutElementOp& out_element_op)
+{
+    using DDataType = ck::remove_cvref_t<ck::tuple_element_t<0, DsDataType>>;
+    Tensor<InDataType> in(in_g_n_c_wis_desc);
+    Tensor<WeiDataType> wei(wei_g_k_c_xs_desc);
+    Tensor<DDataType> bias(out_g_n_k_wos_desc);
+    Tensor<OutDataType> out_host(out_g_n_k_wos_desc);
+    Tensor<OutDataType> out_device(out_g_n_k_wos_desc);
+
+    std::cout << "in: " << in.mDesc << std::endl;
+    std::cout << "wei: " << wei.mDesc << std::endl;
+    std::cout << "out: " << out_host.mDesc << std::endl;
+
+    switch(init_method)
+    {
+    case 0: break;
+    case 1:
+        in.GenerateTensorValue(GeneratorTensor_2<InDataType>{-2, 3});
+        wei.GenerateTensorValue(GeneratorTensor_2<WeiDataType>{-2, 3});
+        bias.GenerateTensorValue(GeneratorTensor_2<DDataType>{-2, 3});
+        break;
+    case 2:
+        in.GenerateTensorValue(GeneratorTensor_3<InDataType>{0.0, 1.0});
+        wei.GenerateTensorValue(GeneratorTensor_3<WeiDataType>{-0.5, 0.5});
+        bias.GenerateTensorValue(GeneratorTensor_3<DDataType>{-0.5, 0.5});
+        break;
+    default:
+        in.GenerateTensorValue(GeneratorTensor_1<InDataType>{1});
+        wei.GenerateTensorValue(GeneratorTensor_1<WeiDataType>{-1});
+        bias.GenerateTensorValue(GeneratorTensor_1<DDataType>{1});
+    }
+
+    DeviceMem in_device_buf(sizeof(InDataType) * in.mDesc.GetElementSpaceSize());
+    DeviceMem wei_device_buf(sizeof(WeiDataType) * wei.mDesc.GetElementSpaceSize());
+    DeviceMem bias_device_buf(sizeof(DDataType) * bias.mDesc.GetElementSpaceSize());
+    DeviceMem out_device_buf(sizeof(OutDataType) * out_device.mDesc.GetElementSpaceSize());
+
+    in_device_buf.ToDevice(in.mData.data());
+    wei_device_buf.ToDevice(wei.mData.data());
+    bias_device_buf.ToDevice(bias.mData.data());
+
+    std::array<ck::index_t, NDimSpatial + 3> a_g_n_c_wis_lengths{};
+    std::array<ck::index_t, NDimSpatial + 3> a_g_n_c_wis_strides{};
+    std::array<ck::index_t, NDimSpatial + 3> b_g_k_c_xs_lengths{};
+    std::array<ck::index_t, NDimSpatial + 3> b_g_k_c_xs_strides{};
+    std::array<ck::index_t, NDimSpatial + 3> d_g_n_k_wos_lengths{};
+    std::array<ck::index_t, NDimSpatial + 3> d_g_n_k_wos_strides{};
+    std::array<ck::index_t, NDimSpatial + 3> e_g_n_k_wos_lengths{};
+    std::array<ck::index_t, NDimSpatial + 3> e_g_n_k_wos_strides{};
+    std::array<ck::index_t, NDimSpatial> conv_filter_strides{};
+    std::array<ck::index_t, NDimSpatial> conv_filter_dilations{};
+    std::array<ck::index_t, NDimSpatial> input_left_pads{};
+    std::array<ck::index_t, NDimSpatial> input_right_pads{};
+
+    auto copy = [](auto& x, auto& y) { ck::ranges::copy(x, y.begin()); };
+
+    copy(in_g_n_c_wis_desc.GetLengths(), a_g_n_c_wis_lengths);
+    copy(in_g_n_c_wis_desc.GetStrides(), a_g_n_c_wis_strides);
+    copy(wei_g_k_c_xs_desc.GetLengths(), b_g_k_c_xs_lengths);
+    copy(wei_g_k_c_xs_desc.GetStrides(), b_g_k_c_xs_strides);
+    copy(out_g_n_k_wos_desc.GetLengths(), d_g_n_k_wos_lengths);
+    copy(out_g_n_k_wos_desc.GetStrides(), d_g_n_k_wos_strides);
+    copy(out_g_n_k_wos_desc.GetLengths(), e_g_n_k_wos_lengths);
+    copy(out_g_n_k_wos_desc.GetStrides(), e_g_n_k_wos_strides);
+    copy(conv_param.conv_filter_strides_, conv_filter_strides);
+    copy(conv_param.conv_filter_dilations_, conv_filter_dilations);
+    copy(conv_param.input_left_pads_, input_left_pads);
+    copy(conv_param.input_right_pads_, input_right_pads);
+
+    // do Conv
+    auto conv     = DeviceConvNDFwdInstance{};
+    auto invoker  = conv.MakeInvoker();
+    auto argument = conv.MakeArgument(
+        in_device_buf.GetDeviceBuffer(),
+        wei_device_buf.GetDeviceBuffer(),
+        std::array<const void*, 1>{bias_device_buf.GetDeviceBuffer()},
+        out_device_buf.GetDeviceBuffer(),
+        a_g_n_c_wis_lengths,
+        a_g_n_c_wis_strides,
+        b_g_k_c_xs_lengths,
+        b_g_k_c_xs_strides,
+        std::array<std::array<ck::index_t, NDimSpatial + 3>, 1>{{d_g_n_k_wos_lengths}},
+        std::array<std::array<ck::index_t, NDimSpatial + 3>, 1>{{d_g_n_k_wos_strides}},
+        e_g_n_k_wos_lengths,
+        e_g_n_k_wos_strides,
+        conv_filter_strides,
+        conv_filter_dilations,
+        input_left_pads,
+        input_right_pads,
+        in_element_op,
+        wei_element_op,
+        out_element_op);
+
+    if(!conv.IsSupportedArgument(argument))
+    {
+        std::cout << "wrong! device_conv with the specified compilation parameters does not "
+                     "support this Conv problem"
+                  << std::endl;
+        return true;
+    }
+
+    float avg_time = invoker.Run(argument, StreamConfig{nullptr, time_kernel});
+
+    std::size_t flop      = conv_param.GetFlops();
+    std::size_t num_btype = conv_param.GetByte<InDataType, WeiDataType, OutDataType>();
+
+    float tflops     = static_cast<float>(flop) / 1.E9 / avg_time;
+    float gb_per_sec = num_btype / 1.E6 / avg_time;
+    std::cout << "Perf: " << avg_time << " ms, " << tflops << " TFlops, " << gb_per_sec << " GB/s, "
+              << conv.GetTypeString() << std::endl;
+
+    if(do_verification)
+    {
+        auto ref_conv = ck::tensor_operation::host::ReferenceConvFwd<
+            NDimSpatial,
+            InDataType,
+            WeiDataType,
+            OutDataType,
+            InElementOp,
+            WeiElementOp,
+            ck::tensor_operation::element_wise::PassThrough>();
+
+        auto ref_invoker = ref_conv.MakeInvoker();
+        auto ref_argument =
+            ref_conv.MakeArgument(in,
+                                  wei,
+                                  out_host,
+                                  conv_param.conv_filter_strides_,
+                                  conv_param.conv_filter_dilations_,
+                                  conv_param.input_left_pads_,
+                                  conv_param.input_right_pads_,
+                                  in_element_op,
+                                  wei_element_op,
+                                  ck::tensor_operation::element_wise::PassThrough{});
+
+        ref_invoker.Run(ref_argument);
+
+        // cde_elementwise
+        out_host.ForEach(
+            [&](auto&, auto idx) { out_element_op(out_host(idx), out_host(idx), bias(idx)); });
+
+        out_device_buf.FromDevice(out_device.mData.data());
+
+        return ck::utils::check_err(
+            out_device.mData, out_host.mData, "Error: incorrect results!", 1e-5f, 1e-4f);
+    }
+
+    return true;
+}

example/09_convnd_fwd/convnd_fwd_dl_fp16.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "convnd_fwd_dl_common.hpp"
+
+#include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
+
+#include "ck/library/utility/convolution_host_tensor_descriptor_helper.hpp"
+
+using InDataType  = ck::half_t;
+using WeiDataType = ck::half_t;
+using AccDataType = float;
+using DsDataType  = ck::Tuple<ck::half_t>;
+using OutDataType = ck::half_t;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using InElementOp  = ck::tensor_operation::element_wise::PassThrough;
+using WeiElementOp = ck::tensor_operation::element_wise::PassThrough;
+using OutElementOp = ck::tensor_operation::element_wise::AddRelu;
+
+static constexpr auto ConvSpec =
+    ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
+
+static constexpr auto GemmPadingSpec = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+template <ck::index_t NDimSpatial, typename InLayout, typename WeiLayout, typename OutLayout>
+// clang-format off
+using DeviceGroupedConvNDFwdInstance = ck::tensor_operation::device::DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK
+// ######|        NDim|     InData|     WeiData|   MultpleD|     OutData|     AccData| InLayout| WeiLayout|            MultipleD| OutLayout|           In|           Wei|           Out|    Convolution|              GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|      ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|      BBlockTransfer|     CThreadTransfer| CThreadTransfer|    CThreadTransfer|
+// ######|     Spatial|       Type|        Type|       Type|        Type|        Type|         |          |               Layout|          |  Elementwise|   Elementwise|   Elementwise|        Forward|    Spacialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor|        SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
+// ######|            |           |            |           |            |            |         |          |                     |          |    Operation|     Operation|     Operation| Specialization|                  |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1|               Order|                |                   |
+// ######|            |           |            |           |            |            |         |          |                     |          |             |              |              |               |                  |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                    |                   |                     |               |               |                    |                   |                    |                    |                |                   |
+         < NDimSpatial, InDataType, WeiDataType, DsDataType, OutDataType, AccDataType, InLayout, WeiLayout, ck::Tuple<OutLayout>, OutLayout,  InElementOp,  WeiElementOp,  OutElementOp,       ConvSpec,    GemmPadingSpec,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,               5,                 4>;
+// clang-format on
+
+#include "run_convnd_fwd_dl_example.inc"
+
+int main(int argc, char* argv[]) { return run_convnd_fwd_dl_example(argc, argv) ? 0 : 1; }

example/09_convnd_fwd/convnd_fwd_dl_fp32.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "convnd_fwd_dl_common.hpp"
+
+#include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
+
+#include "ck/library/utility/convolution_host_tensor_descriptor_helper.hpp"
+
+using InDataType  = float;
+using WeiDataType = float;
+using AccDataType = float;
+using DsDataType  = ck::Tuple<float>;
+using OutDataType = float;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using InElementOp  = ck::tensor_operation::element_wise::PassThrough;
+using WeiElementOp = ck::tensor_operation::element_wise::PassThrough;
+using OutElementOp = ck::tensor_operation::element_wise::AddRelu;
+
+static constexpr auto ConvSpec =
+    ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
+
+static constexpr auto GemmPadingSpec = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+template <ck::index_t NDimSpatial, typename InLayout, typename WeiLayout, typename OutLayout>
+// clang-format off
+using DeviceGroupedConvNDFwdInstance = ck::tensor_operation::device::DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK
+// ######|        NDim|     InData|     WeiData|   MultpleD|     OutData|     AccData| InLayout| WeiLayout|            MultipleD| OutLayout|           In|           Wei|           Out|    Convolution|              GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|      ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|      BBlockTransfer|     CThreadTransfer| CThreadTransfer|    CThreadTransfer|
+// ######|     Spatial|       Type|        Type|       Type|        Type|        Type|         |          |               Layout|          |  Elementwise|   Elementwise|   Elementwise|        Forward|    Spacialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor|        SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
+// ######|            |           |            |           |            |            |         |          |                     |          |    Operation|     Operation|     Operation| Specialization|                  |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1|               Order|                |                   |
+// ######|            |           |            |           |            |            |         |          |                     |          |             |              |              |               |                  |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                    |                   |                     |               |               |                    |                   |                    |                    |                |                   |
+         < NDimSpatial, InDataType, WeiDataType, DsDataType, OutDataType, AccDataType, InLayout, WeiLayout, ck::Tuple<OutLayout>, OutLayout,  InElementOp,  WeiElementOp,  OutElementOp,       ConvSpec,    GemmPadingSpec,   256,   128,   128,    16,  1,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 1>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 1>,      S<1, 2, 0, 3>,       S<1, 1, 1, 1>,      S<8, 1, 1, 1>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 1>,      S<1, 2, 0, 3>,       S<1, 1, 1, 1>, S<0, 1, 2, 3, 4, 5>,               5,                 4>;
+// clang-format on
+
+#include "run_convnd_fwd_dl_example.inc"
+
+int main(int argc, char* argv[]) { return run_convnd_fwd_dl_example(argc, argv) ? 0 : 1; }

example/09_convnd_fwd/convnd_fwd_dl_int8.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "convnd_fwd_dl_common.hpp"
+
+#include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
+
+#include "ck/library/utility/convolution_host_tensor_descriptor_helper.hpp"
+
+using InDataType  = int8_t;
+using WeiDataType = int8_t;
+using AccDataType = int32_t;
+using DsDataType  = ck::Tuple<int8_t>;
+using OutDataType = int8_t;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using InElementOp  = ck::tensor_operation::element_wise::PassThrough;
+using WeiElementOp = ck::tensor_operation::element_wise::PassThrough;
+using OutElementOp = ck::tensor_operation::element_wise::AddRelu;
+
+static constexpr auto ConvSpec =
+    ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
+
+static constexpr auto GemmPadingSpec = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+template <ck::index_t NDimSpatial, typename InLayout, typename WeiLayout, typename OutLayout>
+// clang-format off
+using DeviceGroupedConvNDFwdInstance = ck::tensor_operation::device::DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK
+// ######|        NDim|     InData|     WeiData|   MultpleD|     OutData|     AccData| InLayout| WeiLayout|            MultipleD| OutLayout|           In|           Wei|           Out|    Convolution|              GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|      ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|      BBlockTransfer|     CThreadTransfer| CThreadTransfer|    CThreadTransfer|
+// ######|     Spatial|       Type|        Type|       Type|        Type|        Type|         |          |               Layout|          |  Elementwise|   Elementwise|   Elementwise|        Forward|    Spacialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor|        SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
+// ######|            |           |            |           |            |            |         |          |                     |          |    Operation|     Operation|     Operation| Specialization|                  |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1|               Order|                |                   |
+// ######|            |           |            |           |            |            |         |          |                     |          |             |              |              |               |                  |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                    |                   |                     |               |               |                    |                   |                    |                    |                |                   |
+         < NDimSpatial, InDataType, WeiDataType, DsDataType, OutDataType, AccDataType, InLayout, WeiLayout, ck::Tuple<OutLayout>, OutLayout,  InElementOp,  WeiElementOp,  OutElementOp,       ConvSpec,    GemmPadingSpec,   256,   128,   128,    16,  4,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,       S<1, 1, 1, 4>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,       S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,               5,                 4>;
+// clang-format on
+
+#include "run_convnd_fwd_dl_example.inc"
+
+int main(int argc, char* argv[]) { return run_convnd_fwd_dl_example(argc, argv) ? 0 : 1; }

example/09_convnd_fwd/run_convnd_fwd_dl_example.inc

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+bool run_convnd_fwd_dl_example(int argc, char* argv[])
+{
+    print_helper_msg();
+    bool do_verification = true;
+    int init_method      = 1;
+    bool time_kernel     = false;
+
+    ck::utils::conv::ConvParam conv_param{
+        2, 1, 128, 256, 192, {3, 3}, {71, 71}, {2, 2}, {1, 1}, {1, 1}, {1, 1}};
+
+    if(argc == 1)
+    {
+        // use default
+    }
+    else if(argc == 4)
+    {
+        do_verification = std::stoi(argv[1]);
+        init_method     = std::stoi(argv[2]);
+        time_kernel     = std::stoi(argv[3]);
+    }
+    else
+    {
+        do_verification                   = std::stoi(argv[1]);
+        init_method                       = std::stoi(argv[2]);
+        time_kernel                       = std::stoi(argv[3]);
+        const ck::index_t num_dim_spatial = std::stoi(argv[4]);
+
+        conv_param = ck::utils::conv::parse_conv_param(num_dim_spatial, 5, argv);
+    }
+
+    const auto in_element_op  = InElementOp{};
+    const auto wei_element_op = WeiElementOp{};
+    const auto out_element_op = OutElementOp{};
+
+    const auto run = [&](auto ndim_spatial, auto in_layout, auto wei_layout, auto out_layout) {
+        constexpr ck::index_t ndim_spatial_value = ndim_spatial.value;
+        std::cout << "ndim_spatial_value: " << ndim_spatial_value << std::endl;
+
+        using InLayout  = decltype(in_layout);
+        using WeiLayout = decltype(wei_layout);
+        using OutLayout = decltype(out_layout);
+
+        const auto in_g_n_c_wis_desc =
+            ck::utils::conv::make_input_host_tensor_descriptor_g_n_c_wis_packed<InLayout>(
+                conv_param);
+
+        const auto wei_g_k_c_xs_desc =
+            ck::utils::conv::make_weight_host_tensor_descriptor_g_k_c_xs_packed<WeiLayout>(
+                conv_param);
+
+        const auto out_g_n_k_wos_desc =
+            ck::utils::conv::make_output_host_tensor_descriptor_g_n_k_wos_packed<OutLayout>(
+                conv_param);
+
+        return run_grouped_conv_fwd_dl<
+            ndim_spatial_value,
+            InDataType,
+            WeiDataType,
+            DsDataType,
+            OutDataType,
+            InElementOp,
+            WeiElementOp,
+            OutElementOp,
+            DeviceGroupedConvNDFwdInstance<ndim_spatial_value, InLayout, WeiLayout, OutLayout>>(
+            do_verification,
+            init_method,
+            time_kernel,
+            conv_param,
+            in_g_n_c_wis_desc,
+            wei_g_k_c_xs_desc,
+            out_g_n_k_wos_desc,
+            in_element_op,
+            wei_element_op,
+            out_element_op);
+    };
+
+    namespace ctc = ck::tensor_layout::convolution;
+
+    if(conv_param.num_dim_spatial_ == 1)
+    {
+        return run(ck::Number<1>{}, ctc::GNWC{}, ctc::GKXC{}, ctc::GNWK{});
+    }
+    else if(conv_param.num_dim_spatial_ == 2)
+    {
+        return run(ck::Number<2>{}, ctc::GNHWC{}, ctc::GKYXC{}, ctc::GNHWK{});
+    }
+    else if(conv_param.num_dim_spatial_ == 3)
+    {
+        return run(ck::Number<3>{}, ctc::GNDHWC{}, ctc::GKZYXC{}, ctc::GNDHWK{});
+    }
+
+    return true;
+}

example/10_convnd_fwd_multiple_d_multiple_reduce/common.hpp

@@ -16,6 +16,7 @@
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
+#include "ck/library/utility/algorithm.hpp"
 #include "ck/library/utility/check_err.hpp"
 #include "ck/library/utility/convolution_parameter.hpp"
 #include "ck/library/utility/convolution_host_tensor_descriptor_helper.hpp"
@@ -140,9 +141,7 @@ make_r0_host_tensor_descriptor(const ck::utils::conv::ConvParam& problem_size)
 {
     std::vector<ck::index_t> dimensions{problem_size.G_, problem_size.N_};
 
-    std::copy(begin(problem_size.output_spatial_lengths_),
-              end(problem_size.output_spatial_lengths_),
-              std::back_inserter(dimensions));
+    ck::ranges::copy(problem_size.output_spatial_lengths_, std::back_inserter(dimensions));
 
     return HostTensorDescriptor(dimensions);
 }
@@ -158,10 +157,3 @@ void unpack_host_tensor_descriptor(const HostTensorDescriptor& descriptor,
     assert(size(descriptor.GetStrides()) == size(strides));
     std::copy_n(begin(descriptor.GetStrides()), size(descriptor.GetStrides()), begin(strides));
 }
-
-template <typename Range, typename OutputIterator>
-auto copy(const Range& range, OutputIterator iter)
-    -> decltype(std::copy(std::begin(range), std::end(range), iter))
-{
-    return std::copy(std::begin(range), std::end(range), iter);
-}

example/10_convnd_fwd_multiple_d_multiple_reduce/run_convnd_fwd_max_example.inc

@@ -77,15 +77,12 @@ bool run_convnd_fwd_max(const ck::utils::conv::ConvParam& problem_size,
     {
     case 0: break;
     case 1:
-        ck::utils::FillUniformDistributionIntegerValue<ADataType>{-8, 7}(conv_input.begin(),
-                                                                         conv_input.end());
-        ck::utils::FillUniformDistributionIntegerValue<BDataType>{-8, 7}(conv_weight.begin(),
-                                                                         conv_weight.end());
+        ck::utils::FillUniformDistributionIntegerValue<ADataType>{-8, 7}(conv_input);
+        ck::utils::FillUniformDistributionIntegerValue<BDataType>{-8, 7}(conv_weight);
         break;
     default:
-        ck::utils::FillUniformDistribution<ADataType>{-5, 5}(conv_input.begin(), conv_input.end());
-        ck::utils::FillUniformDistribution<BDataType>{-5, 5}(conv_weight.begin(),
-                                                             conv_weight.end());
+        ck::utils::FillUniformDistribution<ADataType>{-5, 5}(conv_input);
+        ck::utils::FillUniformDistribution<BDataType>{-5, 5}(conv_weight);
     }
 
     DeviceMem conv_input_device_buf(sizeof(ADataType) * conv_input.mDesc.GetElementSpaceSize());
@@ -123,10 +120,10 @@ bool run_convnd_fwd_max(const ck::utils::conv::ConvParam& problem_size,
         conv_output_g_n_k_wos_desc, conv_output_g_n_k_wos_lengths, conv_output_g_n_k_wos_strides);
     unpack_host_tensor_descriptor(r0_desc, r0_lengths, r0_strides);
 
-    copy(problem_size.conv_filter_strides_, begin(conv_filter_strides));
-    copy(problem_size.conv_filter_dilations_, begin(conv_filter_dilations));
-    copy(problem_size.input_left_pads_, begin(input_left_pads));
-    copy(problem_size.input_right_pads_, begin(input_right_pads));
+    ck::ranges::copy(problem_size.conv_filter_strides_, begin(conv_filter_strides));
+    ck::ranges::copy(problem_size.conv_filter_dilations_, begin(conv_filter_dilations));
+    ck::ranges::copy(problem_size.input_left_pads_, begin(input_left_pads));
+    ck::ranges::copy(problem_size.input_right_pads_, begin(input_right_pads));
 
     // run Conv + Reduction on device
     auto conv     = DeviceInstance<NDimSpatial>{};
@@ -276,16 +273,13 @@ bool run_convnd_fwd_max(const ck::utils::conv::ConvParam& problem_size,
         conv_output_device_buf.FromDevice(conv_output_device.mData.data());
         r0_device_buf.FromDevice(r0_device.mData.data());
 
-        return ck::utils::check_err(conv_output_device.mData,
-                                    conv_output_host.mData,
+        return ck::utils::check_err(conv_output_device,
+                                    conv_output_host,
                                     "Error: incorrect results! (Matrix E)",
                                     1e-5f,
                                     1e-4f) &&
-               ck::utils::check_err(r0_device.mData,
-                                    r0_host.mData,
-                                    "Error: incorrect results! (Matrix R0)",
-                                    1e-5f,
-                                    1e-4f);
+               ck::utils::check_err(
+                   r0_device, r0_host, "Error: incorrect results! (Matrix R0)", 1e-5f, 1e-4f);
     }
 
     return true;

example/12_reduce/reduce_blockwise.cpp

@@ -142,7 +142,7 @@ bool reduce_blockwise_test(bool do_verification,
 
         std::array<int, ShapeType::NumReduceDim_> arrReduceDims;
 
-        std::copy(reduceDims.begin(), reduceDims.end(), arrReduceDims.begin());
+        ck::ranges::copy(reduceDims, arrReduceDims.begin());
 
         result = reduce_blockwise_impl<InOutDataType,
                                        AccDataType,

example/12_reduce/reduce_blockwise_impl.hpp

@@ -9,13 +9,14 @@
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/tensor_operation/gpu/device/reduction_operator_mapping.hpp"
 #include "ck/tensor_operation/gpu/device/impl/device_reduce_multiblock.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_reduce.hpp"
 
+#include "ck/library/utility/algorithm.hpp"
 #include "ck/library/utility/check_err.hpp"
 #include "ck/library/utility/device_memory.hpp"
 #include "ck/library/utility/host_tensor.hpp"
 #include "ck/library/utility/host_tensor_generator.hpp"
 #include "ck/library/utility/host_common_util.hpp"
-#include "ck/library/utility/host_reduction.hpp"
 
 #include "reduce_example_common.hpp"
 
@@ -235,38 +236,57 @@ int reduce_blockwise_impl(bool do_verification,
         reduce_unary_operator<ReduceOpId, true, true>::GetElementwiseOperator(
             static_cast<int32_t>(reduce_total_length));
 
-    if(do_verification)
-    {
-        ReductionHost<InOutDataType,
-                      AccDataType,
-                      InOutDataType,
-                      ReduceOperation,
-                      InElementwiseOperation,
-                      AccElementwiseOperation,
-                      Rank,
-                      NumReduceDim,
-                      PropagateNan,
-                      OutputIndex>
-            hostReduce(in.mDesc, out_ref.mDesc, invariantDims, reduceDims);
-
-        hostReduce.Run(alpha,
-                       in.mData.data(),
-                       beta,
-                       out_ref.mData.data(),
-                       out_indices_ref.mData.data(),
-                       in_elementwise_op,
-                       acc_elementwise_op);
-    };
-
     std::array<index_t, Rank> arrInLengths;
     std::array<index_t, Rank> arrInStrides;
     std::array<index_t, NumOutDim> arrOutLengths;
     std::array<index_t, NumOutDim> arrOutStrides;
 
-    std::copy(inLengths.begin(), inLengths.end(), arrInLengths.begin());
-    std::copy(inStrides.begin(), inStrides.end(), arrInStrides.begin());
-    std::copy(outLengths.begin(), outLengths.end(), arrOutLengths.begin());
-    std::copy(outStrides.begin(), outStrides.end(), arrOutStrides.begin());
+    ck::ranges::copy(inLengths, arrInLengths.begin());
+    ck::ranges::copy(inStrides, arrInStrides.begin());
+    ck::ranges::copy(outLengths, arrOutLengths.begin());
+    ck::ranges::copy(outStrides, arrOutStrides.begin());
+
+    if(do_verification)
+    {
+        using ReferenceReduceInstance =
+            ck::tensor_operation::host::ReferenceReduce<InOutDataType,
+                                                        AccDataType,
+                                                        InOutDataType,
+                                                        Rank,
+                                                        NumReduceDim,
+                                                        ReduceOperation,
+                                                        InElementwiseOperation,
+                                                        AccElementwiseOperation,
+                                                        PropagateNan,
+                                                        OutputIndex>;
+
+        auto reduce_ref = ReferenceReduceInstance{};
+
+        auto argument_ptr_ref = reduce_ref.MakeArgumentPointer(arrInLengths,
+                                                               arrInStrides,
+                                                               arrOutLengths,
+                                                               arrOutStrides,
+                                                               reduceDims,
+                                                               static_cast<double>(alpha),
+                                                               static_cast<double>(beta),
+                                                               in.mData.data(),
+                                                               nullptr,
+                                                               out_ref.mData.data(),
+                                                               out_indices_ref.mData.data(),
+                                                               in_elementwise_op,
+                                                               acc_elementwise_op);
+
+        if(!reduce_ref.IsSupportedArgument(argument_ptr_ref.get()))
+        {
+            std::cout << "The runtime parameters not supported by the reduce reference, exiting!"
+                      << std::endl;
+            return (false);
+        };
+
+        auto invoker_ptr_ref = reduce_ref.MakeInvokerPointer();
+
+        invoker_ptr_ref->Run(argument_ptr_ref.get());
+    };
 
     auto reduce = DeviceReduceInstance{};
 
@@ -275,8 +295,8 @@ int reduce_blockwise_impl(bool do_verification,
                                                    arrOutLengths,
                                                    arrOutStrides,
                                                    reduceDims,
-                                                   alpha,
-                                                   beta,
+                                                   static_cast<double>(alpha),
+                                                   static_cast<double>(beta),
                                                    in_dev.GetDeviceBuffer(),
                                                    nullptr,
                                                    out_dev.GetDeviceBuffer(),
@@ -286,9 +306,8 @@ int reduce_blockwise_impl(bool do_verification,
 
     if(!reduce.IsSupportedArgument(argument_ptr.get()))
     {
-        std::cerr
-            << "The runtime parameters seems not supported by the DeviceReduce instance, exiting!"
-            << std::endl;
+        std::cerr << "The runtime parameters not supported by the DeviceReduce instance, exiting!"
+                  << std::endl;
 
         return (-2);
     };
@@ -324,12 +343,12 @@ int reduce_blockwise_impl(bool do_verification,
 #endif
             out_dev.FromDevice(out.mData.data());
 
-        pass = pass && ck::utils::check_err(out.mData, out_ref.mData);
+        pass = pass && ck::utils::check_err(out, out_ref);
 
         if(OutputIndex)
         {
             out_index_dev.FromDevice(out_indices.mData.data());
-            pass = pass && ck::utils::check_err(out_indices.mData, out_indices_ref.mData);
+            pass = pass && ck::utils::check_err(out_indices, out_indices_ref);
         };
     };
 

example/12_reduce/reduce_blockwise_two_call.cpp

@@ -12,13 +12,13 @@
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/tensor_operation/gpu/device/reduction_operator_mapping.hpp"
 #include "ck/tensor_operation/gpu/device/impl/device_reduce_multiblock.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_reduce.hpp"
 
 #include "ck/library/utility/check_err.hpp"
 #include "ck/library/utility/device_memory.hpp"
 #include "ck/library/utility/host_tensor.hpp"
 #include "ck/library/utility/host_tensor_generator.hpp"
 #include "ck/library/utility/host_common_util.hpp"
-#include "ck/library/utility/host_reduction.hpp"
 
 using namespace ck;
 using namespace ck::tensor_operation::device;
@@ -97,8 +97,8 @@ int main(int argc, char* argv[])
     // const std::array<int, 3> invariantDims_2 = {0, 1, 2};
 
     // used by the host reduction
-    const std::array<int, 2> reduceDims    = {3, 4};
-    const std::array<int, 3> invariantDims = {0, 1, 2};
+    const std::array<int, 2> reduceDims = {3, 4};
+    // const std::array<int, 3> invariantDims = {0, 1, 2};
 
     const std::vector<size_t> inLengths_1 = {64, 320, 80, 4, 128};
 
@@ -191,29 +191,6 @@ int main(int argc, char* argv[])
         reduce_unary_operator<ReduceOpId, true, true>::GetElementwiseOperator(
             static_cast<int32_t>(reduce_total_length));
 
-    if(do_verify)
-    {
-        ReductionHost<InOutDataType,
-                      AccDataType,
-                      InOutDataType,
-                      ReduceOperation,
-                      InElementwiseOperation,
-                      AccElementwiseOperation,
-                      5, // Rank
-                      2, // NumReduceDim
-                      PropagateNan,
-                      OutputIndex>
-            hostReduce(in_1.mDesc, out_ref.mDesc, invariantDims, reduceDims);
-
-        hostReduce.Run(alpha,
-                       in_1.mData.data(),
-                       beta,
-                       out_ref.mData.data(),
-                       nullptr,
-                       in_elementwise_op,
-                       acc_elementwise_op);
-    };
-
     std::array<index_t, 5> arrInLengths_1;
     std::array<index_t, 5> arrInStrides_1;
     std::array<index_t, 4> arrInLengths_2;
@@ -221,12 +198,54 @@ int main(int argc, char* argv[])
     std::array<index_t, 3> arrOutLengths;
     std::array<index_t, 3> arrOutStrides;
 
-    std::copy(inLengths_1.begin(), inLengths_1.end(), arrInLengths_1.begin());
-    std::copy(inStrides_1.begin(), inStrides_1.end(), arrInStrides_1.begin());
-    std::copy(inLengths_2.begin(), inLengths_2.end(), arrInLengths_2.begin());
-    std::copy(inStrides_2.begin(), inStrides_2.end(), arrInStrides_2.begin());
-    std::copy(outLengths.begin(), outLengths.end(), arrOutLengths.begin());
-    std::copy(outStrides.begin(), outStrides.end(), arrOutStrides.begin());
+    ck::ranges::copy(inLengths_1, arrInLengths_1.begin());
+    ck::ranges::copy(inStrides_1, arrInStrides_1.begin());
+    ck::ranges::copy(inLengths_2, arrInLengths_2.begin());
+    ck::ranges::copy(inStrides_2, arrInStrides_2.begin());
+    ck::ranges::copy(outLengths, arrOutLengths.begin());
+    ck::ranges::copy(outStrides, arrOutStrides.begin());
+
+    if(do_verify)
+    {
+        using ReferenceReduceInstance =
+            ck::tensor_operation::host::ReferenceReduce<InOutDataType,
+                                                        AccDataType,
+                                                        InOutDataType,
+                                                        5,
+                                                        2,
+                                                        ReduceOperation,
+                                                        InElementwiseOperation,
+                                                        AccElementwiseOperation,
+                                                        PropagateNan,
+                                                        OutputIndex>;
+
+        auto reduce_ref = ReferenceReduceInstance{};
+
+        auto argument_ptr_ref = reduce_ref.MakeArgumentPointer(arrInLengths_1,
+                                                               arrInStrides_1,
+                                                               arrOutLengths,
+                                                               arrOutStrides,
+                                                               reduceDims,
+                                                               static_cast<double>(alpha),
+                                                               static_cast<double>(beta),
+                                                               in_1.mData.data(),
+                                                               nullptr,
+                                                               out_ref.mData.data(),
+                                                               nullptr,
+                                                               in_elementwise_op,
+                                                               acc_elementwise_op);
+
+        if(!reduce_ref.IsSupportedArgument(argument_ptr_ref.get()))
+        {
+            std::cout << "The runtime parameters not supported by the reduce reference, exiting!"
+                      << std::endl;
+            return (false);
+        };
+
+        auto invoker_ptr_ref = reduce_ref.MakeInvokerPointer();
+
+        invoker_ptr_ref->Run(argument_ptr_ref.get());
+    };
 
     auto reduce_1 = DeviceReduceInstance_1{};
 
@@ -235,8 +254,8 @@ int main(int argc, char* argv[])
                                                        arrInLengths_2,
                                                        arrInStrides_2,
                                                        reduceDims_1,
-                                                       1.0f,
-                                                       0.0f,
+                                                       1.0,
+                                                       0.0,
                                                        in_1_dev.GetDeviceBuffer(),
                                                        nullptr,
                                                        in_2_dev.GetDeviceBuffer(),
@@ -246,9 +265,8 @@ int main(int argc, char* argv[])
 
     if(!reduce_1.IsSupportedArgument(argument_ptr_1.get()))
     {
-        std::cout
-            << "The runtime parameters seems not supported by the DeviceReduce instance, exiting!"
-            << std::endl;
+        std::cout << "The runtime parameters seems supported by the DeviceReduce instance, exiting!"
+                  << std::endl;
     };
 
     auto invoker_ptr_1 = reduce_1.MakeInvokerPointer();
@@ -260,8 +278,8 @@ int main(int argc, char* argv[])
                                                        arrOutLengths,
                                                        arrOutStrides,
                                                        reduceDims_2,
-                                                       alpha,
-                                                       beta,
+                                                       static_cast<double>(alpha),
+                                                       static_cast<double>(beta),
                                                        in_2_dev.GetDeviceBuffer(),
                                                        nullptr,
                                                        out_dev.GetDeviceBuffer(),
@@ -294,7 +312,7 @@ int main(int argc, char* argv[])
     if(do_verify)
     {
         out_dev.FromDevice(out.mData.data());
-        pass = pass && ck::utils::check_err(out.mData, out_ref.mData);
+        pass = pass && ck::utils::check_err(out, out_ref);
     };
 
     return (pass ? 0 : 1);

example/12_reduce/reduce_multiblock_atomic_add.cpp

@@ -140,7 +140,7 @@ bool reduce_multiblock_atomic_add_test(bool do_verification,
 
         std::array<int, ShapeType::NumReduceDim_> a_reduceDims;
 
-        std::copy(reduceDims.begin(), reduceDims.end(), a_reduceDims.begin());
+        ck::ranges::copy(reduceDims, a_reduceDims.begin());
 
         result = reduce_multiblock_atomic_add_impl<InOutDataType,
                                                    AccDataType,

example/12_reduce/reduce_multiblock_atomic_add_impl.hpp

@@ -9,13 +9,14 @@
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/tensor_operation/gpu/device/reduction_operator_mapping.hpp"
 #include "ck/tensor_operation/gpu/device/impl/device_reduce_multiblock.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_reduce.hpp"
 
+#include "ck/library/utility/algorithm.hpp"
 #include "ck/library/utility/check_err.hpp"
 #include "ck/library/utility/device_memory.hpp"
 #include "ck/library/utility/host_tensor.hpp"
 #include "ck/library/utility/host_tensor_generator.hpp"
 #include "ck/library/utility/host_common_util.hpp"
-#include "ck/library/utility/host_reduction.hpp"
 
 #include "reduce_example_common.hpp"
 
@@ -148,38 +149,57 @@ int reduce_multiblock_atomic_add_impl(bool do_verification,
         reduce_unary_operator<ReduceOpId, true, true>::GetElementwiseOperator(
             static_cast<int32_t>(reduce_total_length));
 
-    if(do_verification)
-    {
-        ReductionHost<InOutDataType,
-                      AccDataType,
-                      InOutDataType,
-                      ReduceOperation,
-                      InElementwiseOperation,
-                      AccElementwiseOperation,
-                      Rank,
-                      NumReduceDim,
-                      PropagateNan,
-                      false>
-            hostReduce(in.mDesc, out_ref.mDesc, invariantDims, reduceDims);
-
-        hostReduce.Run(alpha,
-                       in.mData.data(),
-                       beta,
-                       out_ref.mData.data(),
-                       nullptr,
-                       in_elementwise_op,
-                       acc_elementwise_op);
-    };
-
     std::array<index_t, Rank> arrInLengths;
     std::array<index_t, Rank> arrInStrides;
     std::array<index_t, NumOutDim> arrOutLengths;
     std::array<index_t, NumOutDim> arrOutStrides;
 
-    std::copy(inLengths.begin(), inLengths.end(), arrInLengths.begin());
-    std::copy(inStrides.begin(), inStrides.end(), arrInStrides.begin());
-    std::copy(outLengths.begin(), outLengths.end(), arrOutLengths.begin());
-    std::copy(outStrides.begin(), outStrides.end(), arrOutStrides.begin());
+    ck::ranges::copy(inLengths, arrInLengths.begin());
+    ck::ranges::copy(inStrides, arrInStrides.begin());
+    ck::ranges::copy(outLengths, arrOutLengths.begin());
+    ck::ranges::copy(outStrides, arrOutStrides.begin());
+
+    if(do_verification)
+    {
+        using ReferenceReduceInstance =
+            ck::tensor_operation::host::ReferenceReduce<InOutDataType,
+                                                        AccDataType,
+                                                        InOutDataType,
+                                                        Rank,
+                                                        NumReduceDim,
+                                                        ReduceOperation,
+                                                        InElementwiseOperation,
+                                                        AccElementwiseOperation,
+                                                        PropagateNan,
+                                                        false>;
+
+        auto reduce_ref = ReferenceReduceInstance{};
+
+        auto argument_ptr_ref = reduce_ref.MakeArgumentPointer(arrInLengths,
+                                                               arrInStrides,
+                                                               arrOutLengths,
+                                                               arrOutStrides,
+                                                               reduceDims,
+                                                               static_cast<double>(alpha),
+                                                               static_cast<double>(beta),
+                                                               in.mData.data(),
+                                                               nullptr,
+                                                               out_ref.mData.data(),
+                                                               nullptr,
+                                                               in_elementwise_op,
+                                                               acc_elementwise_op);
+
+        if(!reduce_ref.IsSupportedArgument(argument_ptr_ref.get()))
+        {
+            std::cout << "The runtime parameters not supported by the reduce reference, exiting!"
+                      << std::endl;
+            return (false);
+        };
+
+        auto invoker_ptr_ref = reduce_ref.MakeInvokerPointer();
+
+        invoker_ptr_ref->Run(argument_ptr_ref.get());
+    };
 
     auto reduce = DeviceReduceInstance{};
 
@@ -188,8 +208,8 @@ int reduce_multiblock_atomic_add_impl(bool do_verification,
                                                    arrOutLengths,
                                                    arrOutStrides,
                                                    reduceDims,
-                                                   alpha,
-                                                   beta,
+                                                   static_cast<double>(alpha),
+                                                   static_cast<double>(beta),
                                                    in_dev.GetDeviceBuffer(),
                                                    nullptr,
                                                    out_dev.GetDeviceBuffer(),
@@ -199,9 +219,8 @@ int reduce_multiblock_atomic_add_impl(bool do_verification,
 
     if(!reduce.IsSupportedArgument(argument_ptr.get()))
     {
-        std::cerr
-            << "The runtime parameters seems not supported by the DeviceReduce instance, exiting!"
-            << std::endl;
+        std::cerr << "The runtime parameters not supported by the DeviceReduce instance, exiting!"
+                  << std::endl;
 
         return (-2);
     };
@@ -225,7 +244,7 @@ int reduce_multiblock_atomic_add_impl(bool do_verification,
     if(do_verification)
     {
         out_dev.FromDevice(out.mData.data());
-        pass = pass && ck::utils::check_err(out.mData, out_ref.mData);
+        pass = pass && ck::utils::check_err(out, out_ref);
     };
 
     return (pass ? 0 : 1);

example/13_pool2d_fwd/pool2d_fwd_common.hpp

@@ -16,6 +16,7 @@
 #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"
 
 template <typename InDataType,
           typename OutDataType,
@@ -172,16 +173,16 @@ bool pool_test(bool do_verification,
     // tensor layout
     auto f_host_tensor_descriptor =
         [](std::size_t N_, std::size_t C_, std::size_t H, std::size_t W, auto layout) {
+            using namespace ck::literals;
+
             if constexpr(ck::is_same<decltype(layout), ck::tensor_layout::convolution::NCHW>::value)
             {
-                return HostTensorDescriptor(std::vector<std::size_t>({N_, C_, H, W}),
-                                            std::vector<std::size_t>({C_ * H * W, H * W, W, 1}));
+                return HostTensorDescriptor({N_, C_, H, W}, {C_ * H * W, H * W, W, 1_uz});
             }
             else if constexpr(ck::is_same<decltype(layout),
                                           ck::tensor_layout::convolution::NHWC>::value)
             {
-                return HostTensorDescriptor(std::vector<std::size_t>({N_, C_, H, W}),
-                                            std::vector<std::size_t>({C_ * H * W, 1, W * C_, C_}));
+                return HostTensorDescriptor({N_, C_, H, W}, {C_ * H * W, 1_uz, W * C_, C_});
             }
         };
 
@@ -267,14 +268,14 @@ bool pool_test(bool do_verification,
 
         out_device_buf.FromDevice(out_n_c_ho_wo_device.mData.data());
 
-        pass = pass && ck::utils::check_err(out_n_c_ho_wo_device.mData, out_n_c_ho_wo_host.mData);
+        pass = pass && ck::utils::check_err(out_n_c_ho_wo_device, out_n_c_ho_wo_host);
 
         if constexpr(OutputIndex)
         {
             out_indices_device_buf.FromDevice(out_indices_n_c_ho_wo_device.mData.data());
 
-            pass = pass && ck::utils::check_err(out_indices_n_c_ho_wo_device.mData,
-                                                out_indices_n_c_ho_wo_host.mData);
+            pass = pass &&
+                   ck::utils::check_err(out_indices_n_c_ho_wo_device, out_indices_n_c_ho_wo_host);
         };
     }
 

example/14_gemm_quantization/CMakeLists.txt

+add_example_executable(example_gemm_xdl_bias_relu_quantization_int8 gemm_xdl_bias_relu_quantization_int8.cpp)
+add_example_executable(example_gemm_xdl_quantization_int8 gemm_xdl_quantization_int8.cpp)
\ No newline at end of file

example/14_gemm_xdl_requant_relu_requant/gemm_xdl_requant_relu_requant_int8.cpp → example/14_gemm_quantization/gemm_xdl_bias_relu_quantization_int8.cpp

@@ -9,80 +9,74 @@
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
 #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_xdl_cshuffle.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/utility/device_memory.hpp"
 #include "ck/library/utility/host_tensor.hpp"
 #include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/literals.hpp"
 #include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
 #include "ck/library/utility/check_err.hpp"
 
-struct RequantReluRequant
-{
-    // FIXME: We just need one scale for Relu / Leaky Relu / PRelu
-    RequantReluRequant(float scaleGemm, float scaleRelu)
-        : scaleGemm_(scaleGemm), scaleRelu_(scaleRelu)
-    {
-    }
-
-    __host__ __device__ constexpr void operator()(float& y, const float& x) const
-    {
-        float gemm_requant = scaleGemm_ * x;
-        float relu         = gemm_requant > 0 ? gemm_requant : 0;
-        float relu_requant = scaleRelu_ * relu;
-        y                  = relu_requant > 127 ? 127 : relu_requant < -128 ? -128 : relu_requant;
-    }
-
-    float scaleGemm_;
-    float scaleRelu_;
-};
-
 template <ck::index_t... Is>
 using S = ck::Sequence<Is...>;
 
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-
-using ADataType        = int8_t;
-using BDataType        = int8_t;
-using CDataType        = int8_t;
-using AccDataType      = int32_t;
-using CShuffleDataType = float;
-
-using ALayout = ck::tensor_layout::gemm::RowMajor;
-using BLayout = ck::tensor_layout::gemm::ColumnMajor;
-using CLayout = ck::tensor_layout::gemm::RowMajor;
+using I8  = int8_t;
+using I32 = int32_t;
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+using PassThrough  = ck::tensor_operation::element_wise::PassThrough;
+using ActivationOp = ck::tensor_operation::element_wise::Relu;
+using CDEElementOp = ck::tensor_operation::element_wise::Add_Activation_Mul_Clamp<ActivationOp>;
+
+using ADataType        = I8;
+using BDataType        = I8;
+using AccDataType      = I32;
+using CShuffleDataType = I32;
+using BiasDataType     = I32;
+using DsDataType       = ck::Tuple<BiasDataType>;
+using EDataType        = I8;
+
+using ALayout    = Row;
+using BLayout    = Col;
+using BiasLayout = Row;
+using DsLayout   = ck::Tuple<BiasLayout>;
+using ELayout    = Row;
 
 static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
 
 // clang-format off
-using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemm_Xdl_CShuffle<
-     ALayout,                    // typename ALayout,
-     BLayout,                    // typename BLayout,
-     CLayout,                    // typename CLayout,
-     ADataType,                  // typename ADataType,
-     BDataType,                  // typename BDataType,
-     CDataType,                  // typename CDataType,
-     AccDataType,                // typename GemmAccDataType,
-     CShuffleDataType,           // typename CShuffleDataType,
-     PassThrough,                // typename AElementwiseOperation,
-     PassThrough,                // typename BElementwiseOperation,
-     RequantReluRequant,         // typename CElementwiseOperation,
+using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemmMultipleD_Xdl_CShuffle<
+     ALayout,
+     BLayout,
+     DsLayout,
+     ELayout,
+     ADataType,
+     BDataType,
+     AccDataType,
+     CShuffleDataType,
+     DsDataType,
+     EDataType,
+     PassThrough,                // AElementwiseOperation,
+     PassThrough,                // BElementwiseOperation,
+     CDEElementOp,               // CDEElementwiseOperation,
      GemmDefault,                // GemmSpecialization GemmSpec,
-     1,                          // index_t NumGemmKPrefetchStage,
-     256,                        // index_t BlockSize,
-     256,                        // index_t MPerBlock,
-     128,                        // index_t NPerBlock,
-     64,                         // index_t KPerBlock,
-     16,                         // index_t AK1,
-     16,                         // index_t BK1,
-     32,                         // index_t MPerXDL,
-     32,                         // index_t NPerXDL,
-     4,                          // index_t MXdlPerWave,
-     2,                          // index_t NXdlPerWave,
-     S<4, 64, 1>,                // typename ABlockTransferThreadClusterLengths_AK0_M_AK1,
-     S<1, 0, 2>,                 // typename ABlockTransferThreadClusterArrangeOrder,
-     S<1, 0, 2>,                 // typename ABlockTransferSrcAccessOrder,
+     1,                          // NumGemmKPrefetchStage,
+     256,                        // BlockSize,
+     256,                        // MPerBlock,
+     128,                        // NPerBlock,
+     64,                         // KPerBlock,
+     16,                         // AK1,
+     16,                         // BK1,
+     32,                         // MPerXDL,
+     32,                         // NPerXDL,
+     4,                          // MXdlPerWave,
+     2,                          // NXdlPerWave,
+     S<4, 64, 1>,                // ABlockTransferThreadClusterLengths_AK0_M_AK1,
+     S<1, 0, 2>,                 // ABlockTransferThreadClusterArrangeOrder,
+     S<1, 0, 2>,                 // ABlockTransferSrcAccessOrder,
      2,                          // index_t ABlockTransferSrcVectorDim,
      16,                         // index_t ABlockTransferSrcScalarPerVector,
      16,                         // index_t ABlockTransferDstScalarPerVector_AK1,
@@ -97,125 +91,100 @@ using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemm_Xdl_CShuffle
      1,                          // index_t CShuffleMXdlPerWavePerShuffle,
      1,                          // index_t CShuffleNXdlPerWavePerShuffle,
      S<1, 64, 1, 4>,             // typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
-     16>;                        // index_t CShuffleBlockTransferScalarPerVector_NPerBlock>
+     8>;                         // index_t CShuffleBlockTransferScalarPerVector_NPerBlock>
 // clang-format on
 
 using ReferenceGemmInstance = ck::tensor_operation::host::ReferenceGemm<ADataType,
                                                                         BDataType,
-                                                                        CDataType,
-                                                                        float,
+                                                                        AccDataType,
+                                                                        AccDataType,
                                                                         PassThrough,
                                                                         PassThrough,
-                                                                        RequantReluRequant>;
+                                                                        PassThrough>;
 
-int main(int argc, char* argv[])
+int main()
 {
     bool do_verification = true;
-    int init_method      = 1;
     bool time_kernel     = false;
 
     // GEMM shape
-    ck::index_t M = 3840;
-    ck::index_t N = 4096;
-    ck::index_t K = 4096;
-
-    ck::index_t StrideA = 4096;
-    ck::index_t StrideB = 4096;
-    ck::index_t StrideC = 4096;
+    ck::index_t M = 1024;
+    ck::index_t N = 1024;
+    ck::index_t K = 1024;
 
-    float scale_gemm = 0.03;
-    float scale_relu = 1;
+    ck::index_t StrideA    = 1024;
+    ck::index_t StrideB    = 1024;
+    ck::index_t StrideBias = 0;
+    ck::index_t StrideE    = 1024;
 
-    if(argc == 4)
-    {
-        do_verification = std::stoi(argv[1]);
-        init_method     = std::stoi(argv[2]);
-        time_kernel     = std::stoi(argv[3]);
-    }
-    else if(argc == 10)
-    {
-        do_verification = std::stoi(argv[1]);
-        init_method     = std::stoi(argv[2]);
-        time_kernel     = std::stoi(argv[3]);
+    float requant_scale = 0.03;
 
-        M = std::stoi(argv[4]);
-        N = std::stoi(argv[5]);
-        K = std::stoi(argv[6]);
-
-        StrideA = std::stoi(argv[7]);
-        StrideB = std::stoi(argv[8]);
-        StrideC = std::stoi(argv[9]);
-    }
-    else
-    {
-        printf("arg1: verification (0=no, 1=yes)\n");
-        printf("arg2: initialization (0=no init, 1=integer value, 2=decimal value)\n");
-        printf("arg3: time kernel (0=n0, 1=yes)\n");
-        printf("arg4 to 9: M (256x), N(128x), K(32x), StrideA, StrideB, StrideC\n");
-        exit(0);
-    }
-
-    auto f_host_tensor_descriptor =
+    auto f_host_tensor_descriptor2d =
         [](std::size_t row, std::size_t col, std::size_t stride, auto layout) {
+            using namespace ck::literals;
+
             if(std::is_same<decltype(layout), ck::tensor_layout::gemm::RowMajor>::value)
             {
                 return HostTensorDescriptor(std::vector<std::size_t>({row, col}),
-                                            std::vector<std::size_t>({stride, 1}));
+                                            std::vector<std::size_t>({stride, 1_uz}));
             }
             else
             {
                 return HostTensorDescriptor(std::vector<std::size_t>({row, col}),
-                                            std::vector<std::size_t>({1, stride}));
+                                            std::vector<std::size_t>({1_uz, stride}));
             }
         };
 
-    Tensor<ADataType> a_m_k(f_host_tensor_descriptor(M, K, StrideA, ALayout{}));
-    Tensor<BDataType> b_k_n(f_host_tensor_descriptor(K, N, StrideB, BLayout{}));
-    Tensor<CDataType> c_m_n_host_result(f_host_tensor_descriptor(M, N, StrideC, CLayout{}));
-    Tensor<CDataType> c_m_n_device_result(f_host_tensor_descriptor(M, N, StrideC, CLayout{}));
+    auto f_host_tensor_descriptor1d = [](std::size_t len, std::size_t stride) {
+        return HostTensorDescriptor(std::vector<std::size_t>({len}),
+                                    std::vector<std::size_t>({stride}));
+    };
+
+    Tensor<ADataType> a_m_k(f_host_tensor_descriptor2d(M, K, StrideA, ALayout{}));
+    Tensor<BDataType> b_k_n(f_host_tensor_descriptor2d(K, N, StrideB, BLayout{}));
+    Tensor<BiasDataType> bias_n(f_host_tensor_descriptor1d(N, 1));
+    Tensor<EDataType> e_m_n_host_result(f_host_tensor_descriptor2d(M, N, StrideE, ELayout{}));
+    Tensor<EDataType> e_m_n_device_result(f_host_tensor_descriptor2d(M, N, StrideE, ELayout{}));
 
     std::cout << "a_m_k: " << a_m_k.mDesc << std::endl;
     std::cout << "b_k_n: " << b_k_n.mDesc << std::endl;
-    std::cout << "c_m_n: " << c_m_n_host_result.mDesc << std::endl;
+    std::cout << "bias_n: " << bias_n.mDesc << std::endl;
+    std::cout << "e_m_n: " << e_m_n_host_result.mDesc << std::endl;
 
-    switch(init_method)
-    {
-    case 0: break;
-    case 1:
-        a_m_k.GenerateTensorValue(GeneratorTensor_2<ADataType>{-5, 5});
-        b_k_n.GenerateTensorValue(GeneratorTensor_2<BDataType>{-5, 5});
-        break;
-    default:
-        a_m_k.GenerateTensorValue(GeneratorTensor_3<ADataType>{0.0, 1.0});
-        b_k_n.GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5});
-    }
+    a_m_k.GenerateTensorValue(GeneratorTensor_2<ADataType>{-128, 127});
+    b_k_n.GenerateTensorValue(GeneratorTensor_2<BDataType>{-128, 127});
+    bias_n.GenerateTensorValue(GeneratorTensor_2<BiasDataType>{-128, 127});
 
-    DeviceMem a_m_k_device_buf(sizeof(ADataType) * a_m_k.mDesc.GetElementSpaceSize());
-    DeviceMem b_k_n_device_buf(sizeof(BDataType) * b_k_n.mDesc.GetElementSpaceSize());
-    DeviceMem c_m_n_device_buf(sizeof(CDataType) * c_m_n_device_result.mDesc.GetElementSpaceSize());
+    DeviceMem a_device_buf(sizeof(ADataType) * a_m_k.mDesc.GetElementSpaceSize());
+    DeviceMem b_device_buf(sizeof(BDataType) * b_k_n.mDesc.GetElementSpaceSize());
+    DeviceMem bias_device_buf(sizeof(BiasDataType) * bias_n.mDesc.GetElementSpaceSize());
+    DeviceMem e_device_buf(sizeof(EDataType) * e_m_n_device_result.mDesc.GetElementSpaceSize());
 
-    a_m_k_device_buf.ToDevice(a_m_k.mData.data());
-    b_k_n_device_buf.ToDevice(b_k_n.mData.data());
+    a_device_buf.ToDevice(a_m_k.mData.data());
+    b_device_buf.ToDevice(b_k_n.mData.data());
+    bias_device_buf.ToDevice(bias_n.mData.data());
 
-    auto a_element_op = PassThrough{};
-    auto b_element_op = PassThrough{};
-    auto c_element_op = RequantReluRequant{scale_gemm, scale_relu};
+    auto a_element_op   = PassThrough{};
+    auto b_element_op   = PassThrough{};
+    auto cde_element_op = CDEElementOp{requant_scale, ActivationOp{}};
 
     // do GEMM
     auto gemm     = DeviceGemmInstance{};
     auto invoker  = gemm.MakeInvoker();
-    auto argument = gemm.MakeArgument(static_cast<ADataType*>(a_m_k_device_buf.GetDeviceBuffer()),
-                                      static_cast<BDataType*>(b_k_n_device_buf.GetDeviceBuffer()),
-                                      static_cast<CDataType*>(c_m_n_device_buf.GetDeviceBuffer()),
+    auto argument = gemm.MakeArgument(a_device_buf.GetDeviceBuffer(),
+                                      b_device_buf.GetDeviceBuffer(),
+                                      {bias_device_buf.GetDeviceBuffer()},
+                                      e_device_buf.GetDeviceBuffer(),
                                       M,
                                       N,
                                       K,
                                       StrideA,
                                       StrideB,
-                                      StrideC,
+                                      {StrideBias},
+                                      StrideE,
                                       a_element_op,
                                       b_element_op,
-                                      c_element_op);
+                                      cde_element_op);
 
     if(!gemm.IsSupportedArgument(argument))
     {
@@ -228,7 +197,7 @@ int main(int argc, char* argv[])
 
     std::size_t flop = std::size_t(2) * M * N * K;
     std::size_t num_btype =
-        sizeof(ADataType) * M * K + sizeof(BDataType) * K * N + sizeof(CDataType) * M * N;
+        sizeof(ADataType) * M * K + sizeof(BDataType) * K * N + sizeof(EDataType) * M * N;
 
     float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
 
@@ -237,19 +206,29 @@ int main(int argc, char* argv[])
     std::cout << "Perf: " << ave_time << " ms, " << tflops << " TFlops, " << gb_per_sec << " GB/s, "
               << gemm.GetTypeString() << std::endl;
 
-    c_m_n_device_buf.FromDevice(c_m_n_device_result.mData.data());
+    e_device_buf.FromDevice(e_m_n_device_result.mData.data());
 
     if(do_verification)
     {
+        Tensor<AccDataType> c_m_n(HostTensorDescriptor{M, N});
+
         auto ref_gemm    = ReferenceGemmInstance{};
         auto ref_invoker = ref_gemm.MakeInvoker();
 
-        auto ref_argument = ref_gemm.MakeArgument(
-            a_m_k, b_k_n, c_m_n_host_result, a_element_op, b_element_op, c_element_op);
+        auto ref_argument =
+            ref_gemm.MakeArgument(a_m_k, b_k_n, c_m_n, a_element_op, b_element_op, PassThrough{});
 
         ref_invoker.Run(ref_argument);
 
-        return ck::utils::check_err(c_m_n_device_result.mData, c_m_n_host_result.mData) ? 0 : 1;
+        for(int m = 0; m < M; ++m)
+        {
+            for(int n = 0; n < N; ++n)
+            {
+                cde_element_op(e_m_n_host_result(m, n), c_m_n(m, n), bias_n(n));
+            }
+        }
+
+        return ck::utils::check_err(e_m_n_device_result, e_m_n_host_result) ? 0 : 1;
     }
 
     return 0;

example/14_gemm_quantization/gemm_xdl_quantization_int8.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iostream>
+#include <numeric>
+#include <initializer_list>
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/literals.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
+#include "ck/library/utility/check_err.hpp"
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using I8  = int8_t;
+using I32 = int32_t;
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+using PassThrough  = ck::tensor_operation::element_wise::PassThrough;
+using ActivationOp = PassThrough;
+using CDEElementOp = ck::tensor_operation::element_wise::Activation_Mul_Clamp<ActivationOp>;
+
+using ADataType        = I8;
+using BDataType        = I8;
+using AccDataType      = I32;
+using CShuffleDataType = I32;
+using DsDataType       = ck::Tuple<>;
+using EDataType        = I8;
+
+using ALayout  = Row;
+using BLayout  = Col;
+using DsLayout = ck::Tuple<>;
+using ELayout  = Row;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+
+// clang-format off
+using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemmMultipleD_Xdl_CShuffle<
+     ALayout,
+     BLayout,
+     DsLayout,
+     ELayout,
+     ADataType,
+     BDataType,
+     AccDataType,
+     CShuffleDataType,
+     DsDataType,
+     EDataType,
+     PassThrough,                // AElementwiseOperation,
+     PassThrough,                // BElementwiseOperation,
+     CDEElementOp,               // CDEElementwiseOperation,
+     GemmDefault,                // GemmSpecialization GemmSpec,
+     1,                          // NumGemmKPrefetchStage,
+     256,                        // BlockSize,
+     256,                        // MPerBlock,
+     128,                        // NPerBlock,
+     64,                         // KPerBlock,
+     16,                         // AK1,
+     16,                         // BK1,
+     32,                         // MPerXDL,
+     32,                         // NPerXDL,
+     4,                          // MXdlPerWave,
+     2,                          // NXdlPerWave,
+     S<4, 64, 1>,                // ABlockTransferThreadClusterLengths_AK0_M_AK1,
+     S<1, 0, 2>,                 // ABlockTransferThreadClusterArrangeOrder,
+     S<1, 0, 2>,                 // ABlockTransferSrcAccessOrder,
+     2,                          // index_t ABlockTransferSrcVectorDim,
+     16,                         // index_t ABlockTransferSrcScalarPerVector,
+     16,                         // index_t ABlockTransferDstScalarPerVector_AK1,
+     1,                          // bool ABlockLdsExtraM,
+     S<4, 64, 1>,                // typename BBlockTransferThreadClusterLengths_BK0_N_BK1,
+     S<1, 0, 2>,                 // typename BBlockTransferThreadClusterArrangeOrder,
+     S<1, 0, 2>,                 // typename BBlockTransferSrcAccessOrder,
+     2,                          // index_t BBlockTransferSrcVectorDim,
+     8,                          // index_t BBlockTransferSrcScalarPerVector,
+     8,                          // index_t BBlockTransferDstScalarPerVector_BK1,
+     1,                          // bool BBlockLdsExtraN,
+     1,                          // index_t CShuffleMXdlPerWavePerShuffle,
+     1,                          // index_t CShuffleNXdlPerWavePerShuffle,
+     S<1, 64, 1, 4>,             // typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
+     16>;                        // index_t CShuffleBlockTransferScalarPerVector_NPerBlock>
+// clang-format on
+
+using ReferenceGemmInstance = ck::tensor_operation::host::
+    ReferenceGemm<ADataType, BDataType, EDataType, float, PassThrough, PassThrough, CDEElementOp>;
+
+int main()
+{
+    bool do_verification = true;
+    bool time_kernel     = false;
+
+    // GEMM shape
+    ck::index_t M = 1024;
+    ck::index_t N = 1024;
+    ck::index_t K = 1024;
+
+    ck::index_t StrideA = 1024;
+    ck::index_t StrideB = 1024;
+    ck::index_t StrideE = 1024;
+
+    float requant_scale = 0.03;
+
+    auto f_host_tensor_descriptor =
+        [](std::size_t row, std::size_t col, std::size_t stride, auto layout) {
+            using namespace ck::literals;
+
+            if(std::is_same<decltype(layout), ck::tensor_layout::gemm::RowMajor>::value)
+            {
+                return HostTensorDescriptor(std::vector<std::size_t>({row, col}),
+                                            std::vector<std::size_t>({stride, 1_uz}));
+            }
+            else
+            {
+                return HostTensorDescriptor(std::vector<std::size_t>({row, col}),
+                                            std::vector<std::size_t>({1_uz, stride}));
+            }
+        };
+
+    Tensor<ADataType> a_m_k(f_host_tensor_descriptor(M, K, StrideA, ALayout{}));
+    Tensor<BDataType> b_k_n(f_host_tensor_descriptor(K, N, StrideB, BLayout{}));
+    Tensor<EDataType> e_m_n_host_result(f_host_tensor_descriptor(M, N, StrideE, ELayout{}));
+    Tensor<EDataType> e_m_n_device_result(f_host_tensor_descriptor(M, N, StrideE, ELayout{}));
+
+    std::cout << "a_m_k: " << a_m_k.mDesc << std::endl;
+    std::cout << "b_k_n: " << b_k_n.mDesc << std::endl;
+    std::cout << "e_m_n: " << e_m_n_host_result.mDesc << std::endl;
+
+    a_m_k.GenerateTensorValue(GeneratorTensor_2<ADataType>{-128, 127});
+    b_k_n.GenerateTensorValue(GeneratorTensor_2<BDataType>{-128, 127});
+
+    DeviceMem a_device_buf(sizeof(ADataType) * a_m_k.mDesc.GetElementSpaceSize());
+    DeviceMem b_device_buf(sizeof(BDataType) * b_k_n.mDesc.GetElementSpaceSize());
+    DeviceMem e_device_buf(sizeof(EDataType) * e_m_n_device_result.mDesc.GetElementSpaceSize());
+
+    a_device_buf.ToDevice(a_m_k.mData.data());
+    b_device_buf.ToDevice(b_k_n.mData.data());
+
+    auto a_element_op   = PassThrough{};
+    auto b_element_op   = PassThrough{};
+    auto cde_element_op = CDEElementOp{requant_scale, ActivationOp{}};
+
+    // do GEMM
+    auto gemm     = DeviceGemmInstance{};
+    auto invoker  = gemm.MakeInvoker();
+    auto argument = gemm.MakeArgument(a_device_buf.GetDeviceBuffer(),
+                                      b_device_buf.GetDeviceBuffer(),
+                                      {},
+                                      e_device_buf.GetDeviceBuffer(),
+                                      M,
+                                      N,
+                                      K,
+                                      StrideA,
+                                      StrideB,
+                                      {},
+                                      StrideE,
+                                      a_element_op,
+                                      b_element_op,
+                                      cde_element_op);
+
+    if(!gemm.IsSupportedArgument(argument))
+    {
+        throw std::runtime_error(
+            "wrong! device_gemm with the specified compilation parameters does "
+            "not support this GEMM problem");
+    }
+
+    float ave_time = invoker.Run(argument, StreamConfig{nullptr, time_kernel});
+
+    std::size_t flop = std::size_t(2) * M * N * K;
+    std::size_t num_btype =
+        sizeof(ADataType) * M * K + sizeof(BDataType) * K * N + sizeof(EDataType) * M * N;
+
+    float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
+
+    float gb_per_sec = num_btype / 1.E6 / ave_time;
+
+    std::cout << "Perf: " << ave_time << " ms, " << tflops << " TFlops, " << gb_per_sec << " GB/s, "
+              << gemm.GetTypeString() << std::endl;
+
+    e_device_buf.FromDevice(e_m_n_device_result.mData.data());
+
+    if(do_verification)
+    {
+        auto ref_gemm    = ReferenceGemmInstance{};
+        auto ref_invoker = ref_gemm.MakeInvoker();
+
+        auto ref_argument = ref_gemm.MakeArgument(
+            a_m_k, b_k_n, e_m_n_host_result, a_element_op, b_element_op, cde_element_op);
+
+        ref_invoker.Run(ref_argument);
+
+        return ck::utils::check_err(e_m_n_device_result, e_m_n_host_result) ? 0 : 1;
+    }
+
+    return 0;
+}

example/14_gemm_xdl_requant_relu_requant/CMakeLists.txt

-add_example_executable(example_gemm_xdl_requant_relu_requant_int8 gemm_xdl_requant_relu_requant_int8.cpp)
\ No newline at end of file

example/15_grouped_gemm/grouped_gemm_xdl_bfp16.cpp

@@ -16,6 +16,7 @@
 #include "ck/library/utility/device_memory.hpp"
 #include "ck/library/utility/host_tensor.hpp"
 #include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/literals.hpp"
 #include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
 
 template <ck::index_t... Is>

example/15_grouped_gemm/grouped_gemm_xdl_fp16.cpp

@@ -16,6 +16,7 @@
 #include "ck/library/utility/device_memory.hpp"
 #include "ck/library/utility/host_tensor.hpp"
 #include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/literals.hpp"
 #include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
 
 template <ck::index_t... Is>

example/15_grouped_gemm/grouped_gemm_xdl_fp32.cpp

@@ -16,6 +16,7 @@
 #include "ck/library/utility/device_memory.hpp"
 #include "ck/library/utility/host_tensor.hpp"
 #include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/literals.hpp"
 #include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
 
 template <ck::index_t... Is>