merge develop

client_example/24_grouped_conv_activation/grouped_convnd_bwd_data_bilinear/grouped_conv_bwd_data_bilinear_residual_fp16.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <tuple>
+#include <cstdlib>
+#include <iomanip>
+#include <iostream>
+#include <iterator>
+#include <numeric>
+#include <vector>
+
+#include "ck/utility/data_type.hpp"
+#include "ck/utility/tuple.hpp"
+#include "ck/ck.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_data_bilinear.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+using InDataType  = ck::half_t;
+using WeiDataType = ck::half_t;
+using OutDataType = ck::half_t;
+// Use std tuple instead of ck tuple to avoid clang
+// implicit instantiation of undefined template error.
+using DDataTypes = std::tuple<ck::half_t>;
+
+using InLayout    = ck::tensor_layout::convolution::NDHWGC;
+using WeiLayout   = ck::tensor_layout::convolution::GKZYXC;
+using OutLayout   = ck::tensor_layout::convolution::NDHWGK;
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Bilinear    = ck::tensor_operation::element_wise::Bilinear;
+
+static constexpr ck::index_t NumDimSpatial = 3;
+static constexpr ck::index_t G             = 32;
+static constexpr ck::index_t N             = 64; // batch size
+static constexpr ck::index_t K             = 64; // output channel
+static constexpr ck::index_t C             = 32; // input channel (per group)
+static constexpr ck::index_t Z             = 3;  // filter D
+static constexpr ck::index_t Y             = 3;  // filter H
+static constexpr ck::index_t X             = 3;  // filter W
+static constexpr ck::index_t Di            = 14; // input D
+static constexpr ck::index_t Hi            = 14; // input H
+static constexpr ck::index_t Wi            = 14; // input W
+static constexpr ck::index_t Do            = 14; // output D
+static constexpr ck::index_t Ho            = 14; // output H
+static constexpr ck::index_t Wo            = 14; // output W
+
+struct SimpleDeviceMem
+{
+    SimpleDeviceMem() = delete;
+
+    SimpleDeviceMem(std::size_t mem_size) : p_mem_{}
+    {
+        (void)hipMalloc(static_cast<void**>(&p_mem_), mem_size);
+    }
+
+    void* GetDeviceBuffer() { return p_mem_; }
+
+    ~SimpleDeviceMem() { (void)hipFree(p_mem_); }
+
+    void* p_mem_;
+};
+
+int execute_conv_bwd_data_bilinear()
+{
+    std::array<ck::index_t, NumDimSpatial + 3> in_lengths{G, N, C, Di, Hi, Wi};
+    std::array<ck::index_t, NumDimSpatial + 3> in_strides{
+        C, Di * Hi * Wi * G * C, 1, Hi * Wi * G * C, Wi * G * C, G * C};
+
+    std::array<ck::index_t, NumDimSpatial + 3> wei_lengths{G, K, C, Z, Y, X};
+    std::array<ck::index_t, NumDimSpatial + 3> wei_strides{
+        K * Z * Y * X * C, Z * Y * X * C, 1, Y * X * C, X * C, C};
+
+    std::array<ck::index_t, NumDimSpatial + 3> out_lengths{G, N, K, Do, Ho, Wo};
+    std::array<ck::index_t, NumDimSpatial + 3> out_strides{
+        K, Do * Ho * Wo * G * K, 1, Ho * Wo * G * K, Wo * G * K, G * K};
+
+    std::array<ck::index_t, NumDimSpatial> filter_strides{1, 1, 1};
+    std::array<ck::index_t, NumDimSpatial> filter_dilations{1, 1, 1};
+    std::array<ck::index_t, NumDimSpatial> input_left_pads{1, 1, 1};
+    std::array<ck::index_t, NumDimSpatial> input_right_pads{1, 1, 1};
+
+    SimpleDeviceMem in(sizeof(InDataType) * G * N * Di * Hi * Wi * C);
+    SimpleDeviceMem wei(sizeof(WeiDataType) * G * K * Z * Y * X * C);
+    SimpleDeviceMem out(sizeof(OutDataType) * G * N * Do * Ho * Wo * K);
+
+    using DeviceOp =
+        ck::tensor_operation::device::DeviceGroupedConvBwdDataMultipleD<NumDimSpatial,
+                                                                        OutLayout,
+                                                                        WeiLayout,
+                                                                        ck::Tuple<InLayout>,
+                                                                        InLayout,
+                                                                        OutDataType,
+                                                                        WeiDataType,
+                                                                        ck::Tuple<InDataType>,
+                                                                        InDataType,
+                                                                        PassThrough,
+                                                                        PassThrough,
+                                                                        Bilinear>;
+
+    // get device op instances
+    const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+        DeviceOp>::GetInstances();
+
+    std::cout << "found " << op_ptrs.size() << " instances" << std::endl;
+
+    std::string best_op_name;
+    int best_op_id        = -1;
+    float best_avg_time   = std::numeric_limits<float>::max();
+    float best_gb_per_sec = 0;
+    float best_tflops     = 0;
+
+    // profile device operation instances
+    std::cout << "Run all instances and do timing" << std::endl;
+
+    for(int i = 0; i < op_ptrs.size(); ++i)
+    {
+        auto& op_ptr        = op_ptrs[i];
+        auto argument_ptr   = op_ptr->MakeArgumentPointer(out.GetDeviceBuffer(),
+                                                        wei.GetDeviceBuffer(),
+                                                        {in.GetDeviceBuffer()},
+                                                        in.GetDeviceBuffer(),
+                                                        out_lengths,
+                                                        out_strides,
+                                                        wei_lengths,
+                                                        wei_strides,
+                                                        {in_lengths},
+                                                        {in_strides},
+                                                        in_lengths,
+                                                        in_strides,
+                                                        filter_strides,
+                                                        filter_dilations,
+                                                        input_left_pads,
+                                                        input_right_pads,
+                                                        PassThrough{},
+                                                        PassThrough{},
+                                                        Bilinear{2.f, 2.f});
+        auto invoker_ptr    = op_ptr->MakeInvokerPointer();
+        std::string op_name = op_ptr->GetTypeString();
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            float avg_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, true});
+
+            std::size_t flop = std::size_t(2) * G * N * K * C * Do * Ho * Wo * Y * X +
+                               3 * G * N * Di * Hi * Wi * C;
+            std::size_t num_bytes = 2 * sizeof(InDataType) * G * N * Di * Hi * Wi * C +
+                                    sizeof(WeiDataType) * G * K * Z * Y * X * C +
+                                    sizeof(OutDataType) * G * N * Do * Ho * Wo * K;
+
+            float tflops     = static_cast<float>(flop) / 1.E9 / avg_time;
+            float gb_per_sec = num_bytes / 1.E6 / avg_time;
+
+            std::cout << "Perf: " << std::setw(10) << avg_time << " ms, " << tflops << " TFlops, "
+                      << gb_per_sec << " GB/s, " << op_name << std::endl;
+
+            if(tflops > best_tflops)
+            {
+                best_op_id      = i;
+                best_op_name    = op_name;
+                best_avg_time   = avg_time;
+                best_gb_per_sec = gb_per_sec;
+                best_tflops     = tflops;
+            }
+        }
+        else
+        {
+            std::cerr << op_name << " does not support this problem" << std::endl;
+        }
+    }
+
+    if(best_op_id < 0)
+    {
+        std::cerr << "no suitable instance" << std::endl;
+        return EXIT_FAILURE;
+    }
+
+    std::cout << "Best Perf: " << std::setw(10) << best_avg_time << " ms, " << best_tflops
+              << " TFlops, " << best_gb_per_sec << " GB/s, " << best_op_name << std::endl;
+
+    // run the best intance
+    {
+        auto& op_ptr = op_ptrs[best_op_id];
+        std::cout << "Run the best instance without timing: " << op_ptr->GetTypeString()
+                  << std::endl;
+        auto argument_ptr = op_ptr->MakeArgumentPointer(out.GetDeviceBuffer(),
+                                                        wei.GetDeviceBuffer(),
+                                                        {in.GetDeviceBuffer()},
+                                                        in.GetDeviceBuffer(),
+                                                        out_lengths,
+                                                        out_strides,
+                                                        wei_lengths,
+                                                        wei_strides,
+                                                        {in_lengths},
+                                                        {in_strides},
+                                                        in_lengths,
+                                                        in_strides,
+                                                        filter_strides,
+                                                        filter_dilations,
+                                                        input_left_pads,
+                                                        input_right_pads,
+                                                        PassThrough{},
+                                                        PassThrough{},
+                                                        Bilinear{2.f, 2.f});
+
+        auto invoker_ptr = op_ptr->MakeInvokerPointer();
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, false});
+        }
+
+        std::cout << "Done" << std::endl;
+    }
+    return 0;
+}
+
+int main() { return execute_conv_bwd_data_bilinear(); }

client_example/24_grouped_conv_activation/grouped_convnd_fwd_bilinear/grouped_conv_fwd_bilinear_residual_fp16.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <tuple>
+#include <cstdlib>
+#include <iomanip>
+#include <iostream>
+#include <iterator>
+#include <numeric>
+#include <vector>
+
+#include "ck/utility/data_type.hpp"
+#include "ck/utility/tuple.hpp"
+#include "ck/ck.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_convolution_forward_bilinear.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+using InDataType  = ck::half_t;
+using WeiDataType = ck::half_t;
+using OutDataType = ck::half_t;
+// Use std tuple instead of ck tuple to avoid clang
+// implicit instantiation of undefined template error.
+using DDataTypes = std::tuple<ck::half_t>;
+
+using InLayout    = ck::tensor_layout::convolution::NDHWGC;
+using WeiLayout   = ck::tensor_layout::convolution::GKZYXC;
+using OutLayout   = ck::tensor_layout::convolution::NDHWGK;
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Bilinear    = ck::tensor_operation::element_wise::Bilinear;
+
+static constexpr ck::index_t NumDimSpatial = 3;
+static constexpr ck::index_t G             = 32;
+static constexpr ck::index_t N             = 64; // batch size
+static constexpr ck::index_t K             = 64; // output channel
+static constexpr ck::index_t C             = 32; // input channel (per group)
+static constexpr ck::index_t Z             = 3;  // filter D
+static constexpr ck::index_t Y             = 3;  // filter H
+static constexpr ck::index_t X             = 3;  // filter W
+static constexpr ck::index_t Di            = 14; // input D
+static constexpr ck::index_t Hi            = 14; // input H
+static constexpr ck::index_t Wi            = 14; // input W
+static constexpr ck::index_t Do            = 14; // output D
+static constexpr ck::index_t Ho            = 14; // output H
+static constexpr ck::index_t Wo            = 14; // output W
+
+struct SimpleDeviceMem
+{
+    SimpleDeviceMem() = delete;
+
+    SimpleDeviceMem(std::size_t mem_size) : p_mem_{}
+    {
+        (void)hipMalloc(static_cast<void**>(&p_mem_), mem_size);
+    }
+
+    void* GetDeviceBuffer() { return p_mem_; }
+
+    ~SimpleDeviceMem() { (void)hipFree(p_mem_); }
+
+    void* p_mem_;
+};
+
+int execute_conv_fwd_bilinear()
+{
+    // We have NHWGC/GKYXC/NHWGK (x, weight, y) in memory space.
+    // However, CK's API only accepts lengths and strides with order of GNCDHW/GKCZYX/GNKDHW.
+    // Hence, we need to adjust the order of strides.
+    std::array<ck::index_t, 6> in_lengths{G, N, C, Di, Hi, Wi};
+    std::array<ck::index_t, 6> in_strides{
+        C, Di * Hi * Wi * G * C, 1, Hi * Wi * G * C, Wi * G * C, G * C};
+    std::array<ck::index_t, 6> wei_lengths{G, K, C, Z, Y, X};
+    std::array<ck::index_t, 6> wei_strides{
+        K * Z * Y * X * C, Z * Y * X * C, 1, Y * X * C, X * C, C};
+    std::array<ck::index_t, 6> out_lengths{G, N, K, Do, Ho, Wo};
+    std::array<ck::index_t, 6> out_strides{
+        K, Do * Ho * Wo * G * K, 1, Ho * Wo * G * K, Wo * G * K, G * K};
+    // Logical broadcast bias (we have to pass bias lengths in the same format as output - GNKDHW)
+    std::array<ck::index_t, 6> bias_lengths{G, 1, K, 1, 1, 1};
+    std::array<ck::index_t, 6> bias_strides{K, 0, 1, 0, 0, 0};
+
+    std::array<ck::index_t, NumDimSpatial> filter_strides{1, 1, 1};
+    std::array<ck::index_t, NumDimSpatial> filter_dilations{1, 1, 1};
+    std::array<ck::index_t, NumDimSpatial> input_left_pads{1, 1, 1};
+    std::array<ck::index_t, NumDimSpatial> input_right_pads{1, 1, 1};
+
+    SimpleDeviceMem in(sizeof(InDataType) * N * Di * Hi * Wi * G * C);
+    SimpleDeviceMem wei(sizeof(WeiDataType) * G * K * Z * Y * X * C);
+    SimpleDeviceMem out(sizeof(OutDataType) * N * Do * Ho * Wo * G * K);
+
+    using DeviceOp =
+        ck::tensor_operation::device::DeviceGroupedConvFwdMultipleABD<NumDimSpatial,
+                                                                      InLayout,
+                                                                      WeiLayout,
+                                                                      ck::Tuple<OutLayout>,
+                                                                      OutLayout,
+                                                                      InDataType,
+                                                                      WeiDataType,
+                                                                      ck::Tuple<OutDataType>,
+                                                                      OutDataType,
+                                                                      PassThrough,
+                                                                      PassThrough,
+                                                                      Bilinear>;
+
+    // get device op instances
+    const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+        DeviceOp>::GetInstances();
+
+    std::cout << "found " << op_ptrs.size() << " instances" << std::endl;
+
+    std::string best_op_name;
+    int best_op_id        = -1;
+    float best_avg_time   = std::numeric_limits<float>::max();
+    float best_gb_per_sec = 0;
+    float best_tflops     = 0;
+
+    // profile device operation instances
+    std::cout << "Run all instances and do timing" << std::endl;
+
+    for(int i = 0; i < op_ptrs.size(); ++i)
+    {
+        auto& op_ptr        = op_ptrs[i];
+        auto argument_ptr   = op_ptr->MakeArgumentPointer(in.GetDeviceBuffer(),
+                                                        wei.GetDeviceBuffer(),
+                                                        {out.GetDeviceBuffer()},
+                                                        out.GetDeviceBuffer(),
+                                                        in_lengths,
+                                                        in_strides,
+                                                        wei_lengths,
+                                                        wei_strides,
+                                                        {out_lengths},
+                                                        {out_strides},
+                                                        out_lengths,
+                                                        out_strides,
+                                                        filter_strides,
+                                                        filter_dilations,
+                                                        input_left_pads,
+                                                        input_right_pads,
+                                                        PassThrough{},
+                                                        PassThrough{},
+                                                        Bilinear{2.f, 2.f});
+        auto invoker_ptr    = op_ptr->MakeInvokerPointer();
+        std::string op_name = op_ptr->GetTypeString();
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            float avg_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, true});
+
+            std::size_t flop =
+                std::size_t(2) * G * N * K * C * Ho * Wo * Y * X + 3 * N * Ho * Wo * G * K;
+            std::size_t num_bytes = sizeof(InDataType) * N * Hi * Wi * G * C +
+                                    sizeof(WeiDataType) * G * K * Y * X * C +
+                                    sizeof(OutDataType) * 2 * N * Ho * Wo * G * K;
+
+            float tflops     = static_cast<float>(flop) / 1.E9 / avg_time;
+            float gb_per_sec = num_bytes / 1.E6 / avg_time;
+
+            std::cout << "Perf: " << std::setw(10) << avg_time << " ms, " << tflops << " TFlops, "
+                      << gb_per_sec << " GB/s, " << op_name << std::endl;
+
+            if(tflops > best_tflops)
+            {
+                best_op_id      = i;
+                best_op_name    = op_name;
+                best_avg_time   = avg_time;
+                best_gb_per_sec = gb_per_sec;
+                best_tflops     = tflops;
+            }
+        }
+        else
+        {
+            std::cerr << op_name << " does not support this problem" << std::endl;
+        }
+    }
+
+    if(best_op_id < 0)
+    {
+        std::cerr << "no suitable instance" << std::endl;
+        return EXIT_FAILURE;
+    }
+
+    std::cout << "Best Perf: " << std::setw(10) << best_avg_time << " ms, " << best_tflops
+              << " TFlops, " << best_gb_per_sec << " GB/s, " << best_op_name << std::endl;
+
+    // run the best intance
+    {
+        auto& op_ptr = op_ptrs[best_op_id];
+        std::cout << "Run the best instance without timing: " << op_ptr->GetTypeString()
+                  << std::endl;
+        auto argument_ptr = op_ptr->MakeArgumentPointer(in.GetDeviceBuffer(),
+                                                        wei.GetDeviceBuffer(),
+                                                        {out.GetDeviceBuffer()},
+                                                        out.GetDeviceBuffer(),
+                                                        in_lengths,
+                                                        in_strides,
+                                                        wei_lengths,
+                                                        wei_strides,
+                                                        {out_lengths},
+                                                        {out_strides},
+                                                        out_lengths,
+                                                        out_strides,
+                                                        filter_strides,
+                                                        filter_dilations,
+                                                        input_left_pads,
+                                                        input_right_pads,
+                                                        PassThrough{},
+                                                        PassThrough{},
+                                                        Bilinear{2.f, 2.f});
+
+        auto invoker_ptr = op_ptr->MakeInvokerPointer();
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, false});
+        }
+
+        std::cout << "Done" << std::endl;
+    }
+    return 0;
+}
+
+int main() { return execute_conv_fwd_bilinear(); }

client_example/24_grouped_conv_activation/grouped_convnd_fwd_scaleadd_ab/grouped_conv_fwd_scaleadd_ab.inc

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023-2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+#include <iomanip>
+#include <iostream>
+#include <iterator>
+#include <numeric>
+#include <vector>
+
+#include "ck/ck.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_convolution_forward_scaleadd_ab.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+using InLayout    = ck::tensor_layout::convolution::NDHWGC;
+using WeiLayout   = ck::tensor_layout::convolution::GKZYXC;
+using OutLayout   = ck::tensor_layout::convolution::NDHWGK;
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using ScaleAdd    = ck::tensor_operation::element_wise::ScaleAdd;
+
+static constexpr ck::index_t NumDimSpatial = 3;
+static constexpr ck::index_t G             = 32;
+static constexpr ck::index_t N             = 64; // batch size
+static constexpr ck::index_t K             = 64; // output channel
+static constexpr ck::index_t C             = 32; // input channel (per group)
+static constexpr ck::index_t Z             = 3;  // filter D
+static constexpr ck::index_t Y             = 3;  // filter H
+static constexpr ck::index_t X             = 3;  // filter W
+static constexpr ck::index_t Di            = 14; // input D
+static constexpr ck::index_t Hi            = 14; // input H
+static constexpr ck::index_t Wi            = 14; // input W
+static constexpr ck::index_t Do            = 14; // output D
+static constexpr ck::index_t Ho            = 14; // output H
+static constexpr ck::index_t Wo            = 14; // output W
+
+struct SimpleDeviceMem
+{
+    SimpleDeviceMem() = delete;
+
+    SimpleDeviceMem(std::size_t mem_size) : p_mem_{}
+    {
+        (void)hipMalloc(static_cast<void**>(&p_mem_), mem_size);
+    }
+
+    void* GetDeviceBuffer() { return p_mem_; }
+
+    ~SimpleDeviceMem() { (void)hipFree(p_mem_); }
+
+    void* p_mem_;
+};
+
+int execute_conv_fwd_scaleadd_ab()
+{
+    constexpr ck::index_t NumAs = 2;
+    constexpr ck::index_t NumBs = 2;
+
+    constexpr float scale = 1.5f;
+
+    // We have NHWGC/GKYXC/NHWGK (x, weight, y) in memory space.
+    // However, CK's API only accepts lengths and strides with order of GNCDHW/GKCZYX/GNKDHW.
+    // Hence, we need to adjust the order of strides.
+    std::array<ck::index_t, 6> in_lengths{G, N, C, Di, Hi, Wi};
+    std::array<ck::index_t, 6> in_strides{
+        C, Di * Hi * Wi * G * C, 1, Hi * Wi * G * C, Wi * G * C, G * C};
+    std::array<ck::index_t, 6> wei_lengths{G, K, C, Z, Y, X};
+    std::array<ck::index_t, 6> wei_strides{
+        K * Z * Y * X * C, Z * Y * X * C, 1, Y * X * C, X * C, C};
+    std::array<ck::index_t, 6> out_lengths{G, N, K, Do, Ho, Wo};
+    std::array<ck::index_t, 6> out_strides{
+        K, Do * Ho * Wo * G * K, 1, Ho * Wo * G * K, Wo * G * K, G * K};
+
+    std::array<ck::index_t, NumDimSpatial> filter_strides{1, 1, 1};
+    std::array<ck::index_t, NumDimSpatial> filter_dilations{1, 1, 1};
+    std::array<ck::index_t, NumDimSpatial> input_left_pads{1, 1, 1};
+    std::array<ck::index_t, NumDimSpatial> input_right_pads{1, 1, 1};
+
+    using InputDtype      = ck::tuple_element_t<0, InDataType>;
+    using InputBiasDtype  = ck::tuple_element_t<1, InDataType>;
+    using WeightDtype     = ck::tuple_element_t<0, WeiDataType>;
+    using WeightBiasDtype = ck::tuple_element_t<1, WeiDataType>;
+
+    SimpleDeviceMem in(sizeof(InputDtype) * N * Di * Hi * Wi * G * C);
+    SimpleDeviceMem in_bias(sizeof(InputBiasDtype) * N * Di * Hi * Wi * G * C);
+    SimpleDeviceMem wei(sizeof(WeightDtype) * G * K * Z * Y * X * C);
+    SimpleDeviceMem wei_bias(sizeof(WeightBiasDtype) * G * K * Z * Y * X * C);
+    SimpleDeviceMem out(sizeof(OutDataType) * N * Do * Ho * Wo * G * K);
+
+    using DeviceOp = ck::tensor_operation::device::DeviceGroupedConvFwdMultipleABD<NumDimSpatial,
+                                                                                   InLayout,
+                                                                                   WeiLayout,
+                                                                                   ck::Tuple<>,
+                                                                                   OutLayout,
+                                                                                   InDataType,
+                                                                                   WeiDataType,
+                                                                                   ck::Tuple<>,
+                                                                                   OutDataType,
+                                                                                   ScaleAdd,
+                                                                                   ScaleAdd,
+                                                                                   PassThrough>;
+
+    // get device op instances
+    const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+        DeviceOp>::GetInstances();
+
+    std::cout << "found " << op_ptrs.size() << " instances" << std::endl;
+
+    std::string best_op_name;
+    int best_op_id        = -1;
+    float best_avg_time   = std::numeric_limits<float>::max();
+    float best_gb_per_sec = 0;
+    float best_tflops     = 0;
+
+    // profile device operation instances
+    std::cout << "Run all instances and do timing" << std::endl;
+
+    std::array<const void*, NumAs> as = {in.GetDeviceBuffer(), in_bias.GetDeviceBuffer()};
+    std::array<const void*, NumBs> bs = {wei.GetDeviceBuffer(), wei_bias.GetDeviceBuffer()};
+    std::array<const void*, 0> ds{};
+
+    for(int i = 0; i < op_ptrs.size(); ++i)
+    {
+        auto& op_ptr        = op_ptrs[i];
+        auto argument_ptr   = op_ptr->MakeArgumentPointer(as,
+                                                        bs,
+                                                        ds,
+                                                        out.GetDeviceBuffer(),
+                                                        in_lengths,
+                                                        in_strides,
+                                                        wei_lengths,
+                                                        wei_strides,
+                                                        {},
+                                                        {},
+                                                        out_lengths,
+                                                        out_strides,
+                                                        filter_strides,
+                                                        filter_dilations,
+                                                        input_left_pads,
+                                                        input_right_pads,
+                                                        ScaleAdd{scale},
+                                                        ScaleAdd{scale},
+                                                        PassThrough{});
+        auto invoker_ptr    = op_ptr->MakeInvokerPointer();
+        std::string op_name = op_ptr->GetTypeString();
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            float avg_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, true});
+
+            std::size_t flop = std::size_t(2) * G * N * K * C * Do * Ho * Wo * Z * Y * X +
+                               N * Di * Hi * Wi * G * C + G * K * Z * Y * X * C;
+            std::size_t num_bytes = 2 * sizeof(InDataType) * N * Di * Hi * Wi * G * C +
+                                    2 * sizeof(WeiDataType) * G * K * Z * Y * X * C +
+                                    sizeof(OutDataType) * N * Do * Ho * Wo * G * K;
+
+            float tflops     = static_cast<float>(flop) / 1.E9 / avg_time;
+            float gb_per_sec = num_bytes / 1.E6 / avg_time;
+
+            std::cout << "Perf: " << std::setw(10) << avg_time << " ms, " << tflops << " TFlops, "
+                      << gb_per_sec << " GB/s, " << op_name << std::endl;
+
+            if(tflops > best_tflops)
+            {
+                best_op_id      = i;
+                best_op_name    = op_name;
+                best_avg_time   = avg_time;
+                best_gb_per_sec = gb_per_sec;
+                best_tflops     = tflops;
+            }
+        }
+        else
+        {
+            std::cerr << op_name << " does not support this problem" << std::endl;
+        }
+    }
+
+    if(best_op_id < 0)
+    {
+        std::cerr << "no suitable instance" << std::endl;
+        return EXIT_FAILURE;
+    }
+
+    std::cout << "Best Perf: " << std::setw(10) << best_avg_time << " ms, " << best_tflops
+              << " TFlops, " << best_gb_per_sec << " GB/s, " << best_op_name << std::endl;
+
+    // run the best intance
+    {
+        auto& op_ptr = op_ptrs[best_op_id];
+        std::cout << "Run the best instance without timing: " << op_ptr->GetTypeString()
+                  << std::endl;
+        auto argument_ptr = op_ptr->MakeArgumentPointer(as,
+                                                        bs,
+                                                        ds,
+                                                        out.GetDeviceBuffer(),
+                                                        in_lengths,
+                                                        in_strides,
+                                                        wei_lengths,
+                                                        wei_strides,
+                                                        {},
+                                                        {},
+                                                        out_lengths,
+                                                        out_strides,
+                                                        filter_strides,
+                                                        filter_dilations,
+                                                        input_left_pads,
+                                                        input_right_pads,
+                                                        ScaleAdd{scale},
+                                                        ScaleAdd{scale},
+                                                        PassThrough{});
+
+        auto invoker_ptr = op_ptr->MakeInvokerPointer();
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, false});
+        }
+
+        std::cout << "Done" << std::endl;
+    }
+    return 0;
+}

client_example/24_grouped_conv_activation/grouped_convnd_fwd_scaleadd_ab/grouped_conv_fwd_scaleadd_ab_bf16.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023-2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/utility/data_type.hpp"
+#include "ck/utility/tuple.hpp"
+
+using InDataType  = ck::Tuple<ck::bhalf_t, ck::bhalf_t>;
+using WeiDataType = ck::Tuple<ck::bhalf_t, ck::bhalf_t>;
+using OutDataType = ck::bhalf_t;
+
+#include "grouped_conv_fwd_scaleadd_ab.inc"
+
+int main() { return execute_conv_fwd_scaleadd_ab(); }

client_example/24_grouped_conv_activation/grouped_convnd_fwd_scaleadd_ab/grouped_conv_fwd_scaleadd_ab_fp16.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023-2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/utility/data_type.hpp"
+#include "ck/utility/tuple.hpp"
+
+using InDataType  = ck::Tuple<ck::half_t, ck::half_t>;
+using WeiDataType = ck::Tuple<ck::half_t, ck::half_t>;
+using OutDataType = ck::half_t;
+
+#include "grouped_conv_fwd_scaleadd_ab.inc"
+
+int main() { return execute_conv_fwd_scaleadd_ab(); }

client_example/24_grouped_conv_activation/grouped_convnd_fwd_scaleadd_ab/grouped_conv_fwd_scaleadd_ab_fp32.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023-2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/utility/data_type.hpp"
+#include "ck/utility/tuple.hpp"
+
+using InDataType  = ck::Tuple<float, float>;
+using WeiDataType = ck::Tuple<float, float>;
+using OutDataType = float;
+
+#include "grouped_conv_fwd_scaleadd_ab.inc"
+
+int main() { return execute_conv_fwd_scaleadd_ab(); }

client_example/24_grouped_conv_activation/grouped_convnd_fwd_scaleadd_ab/grouped_conv_fwd_scaleadd_ab_int8.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023-2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/utility/data_type.hpp"
+#include "ck/utility/tuple.hpp"
+
+using InDataType  = ck::Tuple<int8_t, int8_t>;
+using WeiDataType = ck::Tuple<int8_t, int8_t>;
+using OutDataType = int8_t;
+
+#include "grouped_conv_fwd_scaleadd_ab.inc"
+
+int main() { return execute_conv_fwd_scaleadd_ab(); }

client_example/24_grouped_conv_activation/grouped_convnd_fwd_scaleadd_scaleadd_relu/grouped_conv_fwd_scaleadd_scaleadd_relu.inc

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023-2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+#include <iomanip>
+#include <iostream>
+#include <iterator>
+#include <numeric>
+#include <vector>
+
+#include "ck/ck.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_convolution_forward_scaleadd_scaleadd_relu.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+using InLayout             = ck::tensor_layout::convolution::NDHWGC;
+using WeiLayout            = ck::tensor_layout::convolution::GKZYXC;
+using OutLayout            = ck::tensor_layout::convolution::NDHWGK;
+using BiasLayout           = ck::tensor_layout::convolution::G_K;
+using PassThrough          = ck::tensor_operation::element_wise::PassThrough;
+using ScaleAddScaleAddRelu = ck::tensor_operation::element_wise::ScaleAddScaleAddRelu;
+
+static constexpr ck::index_t NumDimSpatial = 3;
+static constexpr ck::index_t G             = 32;
+static constexpr ck::index_t N             = 64; // batch size
+static constexpr ck::index_t K             = 64; // output channel
+static constexpr ck::index_t C             = 32; // input channel (per group)
+static constexpr ck::index_t Z             = 3;  // filter D
+static constexpr ck::index_t Y             = 3;  // filter H
+static constexpr ck::index_t X             = 3;  // filter W
+static constexpr ck::index_t Di            = 14; // input D
+static constexpr ck::index_t Hi            = 14; // input H
+static constexpr ck::index_t Wi            = 14; // input W
+static constexpr ck::index_t Do            = 14; // output D
+static constexpr ck::index_t Ho            = 14; // output H
+static constexpr ck::index_t Wo            = 14; // output W
+
+struct SimpleDeviceMem
+{
+    SimpleDeviceMem() = delete;
+
+    SimpleDeviceMem(std::size_t mem_size) : p_mem_{}
+    {
+        (void)hipMalloc(static_cast<void**>(&p_mem_), mem_size);
+    }
+
+    void* GetDeviceBuffer() { return p_mem_; }
+
+    ~SimpleDeviceMem() { (void)hipFree(p_mem_); }
+
+    void* p_mem_;
+};
+
+int execute_conv_fwd_scaleadd_scaleadd_relu()
+{
+    // We have NHWGC/GKYXC/NHWGK (x, weight, y) in memory space.
+    // However, CK's API only accepts lengths and strides with order of GNCDHW/GKCZYX/GNKDHW.
+    // Hence, we need to adjust the order of strides.
+    std::array<ck::index_t, 6> in_lengths{G, N, C, Di, Hi, Wi};
+    std::array<ck::index_t, 6> in_strides{
+        C, Di * Hi * Wi * G * C, 1, Hi * Wi * G * C, Wi * G * C, G * C};
+    std::array<ck::index_t, 6> wei_lengths{G, K, C, Z, Y, X};
+    std::array<ck::index_t, 6> wei_strides{
+        K * Z * Y * X * C, Z * Y * X * C, 1, Y * X * C, X * C, C};
+    std::array<ck::index_t, 6> out_lengths{G, N, K, Do, Ho, Wo};
+    std::array<ck::index_t, 6> out_strides{
+        K, Do * Ho * Wo * G * K, 1, Ho * Wo * G * K, Wo * G * K, G * K};
+    // Logical broadcast bias (we have to pass bias lengths in the same format as output - GNKDHW)
+    std::array<ck::index_t, 6> bias_lengths{G, 1, K, 1, 1, 1};
+    std::array<ck::index_t, 6> bias_strides{K, 0, 1, 0, 0, 0};
+
+    std::array<ck::index_t, NumDimSpatial> filter_strides{1, 1, 1};
+    std::array<ck::index_t, NumDimSpatial> filter_dilations{1, 1, 1};
+    std::array<ck::index_t, NumDimSpatial> input_left_pads{1, 1, 1};
+    std::array<ck::index_t, NumDimSpatial> input_right_pads{1, 1, 1};
+
+    SimpleDeviceMem in(sizeof(InDataType) * N * Di * Hi * Wi * G * C);
+    SimpleDeviceMem wei(sizeof(WeiDataType) * G * K * Z * Y * X * C);
+    SimpleDeviceMem out(sizeof(OutDataType) * N * Do * Ho * Wo * G * K);
+    SimpleDeviceMem d0(sizeof(std::tuple_element_t<0, DDataTypes>) * N * Do * Ho * Wo * G * K);
+    SimpleDeviceMem d1(sizeof(std::tuple_element_t<1, DDataTypes>) * G * K);
+
+    using DeviceOp = ck::tensor_operation::device::DeviceGroupedConvFwdMultipleABD<
+        NumDimSpatial,
+        InLayout,
+        WeiLayout,
+        ck::Tuple<OutLayout, BiasLayout>,
+        OutLayout,
+        InDataType,
+        WeiDataType,
+        ck::Tuple<std::tuple_element_t<0, DDataTypes>, std::tuple_element_t<1, DDataTypes>>,
+        OutDataType,
+        PassThrough,
+        PassThrough,
+        ScaleAddScaleAddRelu>;
+
+    // get device op instances
+    const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+        DeviceOp>::GetInstances();
+
+    std::cout << "found " << op_ptrs.size() << " instances" << std::endl;
+
+    std::string best_op_name;
+    int best_op_id        = -1;
+    float best_avg_time   = std::numeric_limits<float>::max();
+    float best_gb_per_sec = 0;
+    float best_tflops     = 0;
+
+    // profile device operation instances
+    std::cout << "Run all instances and do timing" << std::endl;
+
+    for(int i = 0; i < op_ptrs.size(); ++i)
+    {
+        auto& op_ptr = op_ptrs[i];
+        auto argument_ptr =
+            op_ptr->MakeArgumentPointer(in.GetDeviceBuffer(),
+                                        wei.GetDeviceBuffer(),
+                                        {d0.GetDeviceBuffer(), d1.GetDeviceBuffer()},
+                                        out.GetDeviceBuffer(),
+                                        in_lengths,
+                                        in_strides,
+                                        wei_lengths,
+                                        wei_strides,
+                                        {out_lengths, bias_lengths},
+                                        {out_strides, bias_strides},
+                                        out_lengths,
+                                        out_strides,
+                                        filter_strides,
+                                        filter_dilations,
+                                        input_left_pads,
+                                        input_right_pads,
+                                        PassThrough{},
+                                        PassThrough{},
+                                        ScaleAddScaleAddRelu{2.f, 2.f});
+        auto invoker_ptr    = op_ptr->MakeInvokerPointer();
+        std::string op_name = op_ptr->GetTypeString();
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            float avg_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, true});
+
+            std::size_t flop =
+                std::size_t(2) * G * N * K * C * Ho * Wo * Y * X + 2 * N * Ho * Wo * G * K;
+            std::size_t num_bytes =
+                sizeof(InDataType) * N * Hi * Wi * G * C + sizeof(WeiDataType) * G * K * Y * X * C +
+                (sizeof(OutDataType) + sizeof(std::tuple_element_t<0, DDataTypes>) +
+                 sizeof(std::tuple_element_t<1, DDataTypes>)) *
+                    N * Ho * Wo * G * K;
+
+            float tflops     = static_cast<float>(flop) / 1.E9 / avg_time;
+            float gb_per_sec = num_bytes / 1.E6 / avg_time;
+
+            std::cout << "Perf: " << std::setw(10) << avg_time << " ms, " << tflops << " TFlops, "
+                      << gb_per_sec << " GB/s, " << op_name << std::endl;
+
+            if(tflops > best_tflops)
+            {
+                best_op_id      = i;
+                best_op_name    = op_name;
+                best_avg_time   = avg_time;
+                best_gb_per_sec = gb_per_sec;
+                best_tflops     = tflops;
+            }
+        }
+        else
+        {
+            std::cerr << op_name << " does not support this problem" << std::endl;
+        }
+    }
+
+    if(best_op_id < 0)
+    {
+        std::cerr << "no suitable instance" << std::endl;
+        return EXIT_FAILURE;
+    }
+
+    std::cout << "Best Perf: " << std::setw(10) << best_avg_time << " ms, " << best_tflops
+              << " TFlops, " << best_gb_per_sec << " GB/s, " << best_op_name << std::endl;
+
+    // run the best intance
+    {
+        auto& op_ptr = op_ptrs[best_op_id];
+        std::cout << "Run the best instance without timing: " << op_ptr->GetTypeString()
+                  << std::endl;
+        auto argument_ptr =
+            op_ptr->MakeArgumentPointer(in.GetDeviceBuffer(),
+                                        wei.GetDeviceBuffer(),
+                                        {d0.GetDeviceBuffer(), d1.GetDeviceBuffer()},
+                                        out.GetDeviceBuffer(),
+                                        in_lengths,
+                                        in_strides,
+                                        wei_lengths,
+                                        wei_strides,
+                                        {out_lengths, bias_lengths},
+                                        {out_strides, bias_strides},
+                                        out_lengths,
+                                        out_strides,
+                                        filter_strides,
+                                        filter_dilations,
+                                        input_left_pads,
+                                        input_right_pads,
+                                        PassThrough{},
+                                        PassThrough{},
+                                        ScaleAddScaleAddRelu{2.f, 2.f});
+
+        auto invoker_ptr = op_ptr->MakeInvokerPointer();
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, false});
+        }
+
+        std::cout << "Done" << std::endl;
+    }
+    return 0;
+}

client_example/24_grouped_conv_activation/grouped_convnd_fwd_scaleadd_scaleadd_relu/grouped_conv_fwd_scaleadd_scaleadd_relu_bf16.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023-2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <tuple>
+
+#include "ck/utility/data_type.hpp"
+#include "ck/utility/tuple.hpp"
+
+using InDataType  = ck::bhalf_t;
+using WeiDataType = ck::bhalf_t;
+using OutDataType = ck::bhalf_t;
+// Use std tuple instead of ck tuple to avoid clang
+// implicit instantiation of undefined template error.
+using DDataTypes = std::tuple<ck::bhalf_t, ck::bhalf_t>;
+
+#include "grouped_conv_fwd_scaleadd_scaleadd_relu.inc"
+
+int main() { return execute_conv_fwd_scaleadd_scaleadd_relu(); }

client_example/24_grouped_conv_activation/grouped_convnd_fwd_scaleadd_scaleadd_relu/grouped_conv_fwd_scaleadd_scaleadd_relu_fp16.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023-2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <tuple>
+
+#include "ck/utility/data_type.hpp"
+#include "ck/utility/tuple.hpp"
+
+using InDataType  = ck::half_t;
+using WeiDataType = ck::half_t;
+using OutDataType = ck::half_t;
+// Use std tuple instead of ck tuple to avoid clang
+// implicit instantiation of undefined template error.
+using DDataTypes = std::tuple<ck::half_t, ck::half_t>;
+
+#include "grouped_conv_fwd_scaleadd_scaleadd_relu.inc"
+
+int main() { return execute_conv_fwd_scaleadd_scaleadd_relu(); }

client_example/24_grouped_conv_activation/grouped_convnd_fwd_scaleadd_scaleadd_relu/grouped_conv_fwd_scaleadd_scaleadd_relu_fp32.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023-2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <tuple>
+
+#include "ck/utility/data_type.hpp"
+#include "ck/utility/tuple.hpp"
+
+using InDataType  = float;
+using WeiDataType = float;
+using OutDataType = float;
+// Use std tuple instead of ck tuple to avoid clang
+// implicit instantiation of undefined template error.
+using DDataTypes = std::tuple<float, float>;
+
+#include "grouped_conv_fwd_scaleadd_scaleadd_relu.inc"
+
+int main() { return execute_conv_fwd_scaleadd_scaleadd_relu(); }

client_example/24_grouped_conv_activation/grouped_convnd_fwd_scaleadd_scaleadd_relu/grouped_conv_fwd_scaleadd_scaleadd_relu_int8.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023-2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <tuple>
+
+#include "ck/utility/data_type.hpp"
+#include "ck/utility/tuple.hpp"
+
+using InDataType  = int8_t;
+using WeiDataType = int8_t;
+using OutDataType = int8_t;
+// Use std tuple instead of ck tuple to avoid clang
+// implicit instantiation of undefined template error.
+using DDataTypes = std::tuple<float, float>;
+
+#include "grouped_conv_fwd_scaleadd_scaleadd_relu.inc"
+
+int main() { return execute_conv_fwd_scaleadd_scaleadd_relu(); }

client_example/25_wrapper/CMakeLists.txt

+add_executable(client_tensor_transform_using_wrapper tensor_transform_using_wrapper.cpp)
+target_link_libraries(client_tensor_transform_using_wrapper PRIVATE composable_kernel::device_other_operations)
+add_executable(client_wrapper_img2col wrapper_img2col.cpp)
+target_link_libraries(client_wrapper_img2col PRIVATE composable_kernel::device_other_operations)
+if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR
+   GPU_TARGETS MATCHES "gfx940" OR GPU_TARGETS MATCHES "gfx941" OR
+   GPU_TARGETS MATCHES "gfx942")
+    add_executable(client_wrapper_basic_gemm wrapper_basic_gemm.cpp)
+    target_link_libraries(client_wrapper_basic_gemm PRIVATE composable_kernel::device_other_operations)
+    add_executable(client_wrapper_optimized_gemm wrapper_optimized_gemm.cpp)
+    target_link_libraries(client_wrapper_optimized_gemm PRIVATE composable_kernel::device_other_operations)
+endif()

client_example/25_wrapper/README.md

+# Composable Kernel wrapper GEMM tutorial
+
+This tutorial demonstrates how to implement matrix multiplication using Composable Kernel (CK)
+wrapper. We present the base version of GEMM without most of the available optimizations; however,
+it's worth noting that CK has kernels with different optimizations.
+
+To implement these optimizations, you can use the CK wrapper or directly use available instances in
+CK. You can also refer to the
+[optimized GEMM example](https://github.com/ROCm/composable_kernel/blob/develop/client_example/25_wrapper/wrapper_optimized_gemm.cpp),
+that uses CK wrapper based on the
+[`gridwise_gemm_xdlops_v2r3`](https://github.com/ROCm/composable_kernel/blob/develop/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r3.hpp) implementation.
+
+The kernel definition should look similar to:
+
+```cpp
+template <typename DataType,
+          typename GemmTraits,
+          ck::index_t scalar_per_vector,
+          typename BlockShape,
+          typename ThreadLayout>
+__global__ void __CK_WRAPPER_LAUNCH_BOUNDS__ DeviceGemm(const void* p_a,
+                                                        const void* p_b,
+                                                        void* p_c,
+                                                        const ck::index_t M,
+                                                        const ck::index_t N,
+                                                        const ck::index_t K,
+                                                        const BlockShape tile_shape,
+                                                        const ThreadLayout thread_layout)
+```
+
+We pass pointers to global memory and matrix dimensions via arguments. Additionally, we pass
+selected lengths of processed data through each block (`tile_shape`) and thread layout
+(`thread_layout`). For compilation time parameters, we define the data type,
+[traits for the GEMM operation](https://github.com/ROCm/composable_kernel/blob/develop/include/ck/wrapper/traits/blockwise_gemm_xdl_traits.hpp)
+and scalar per vector value during copy.
+
+Step 1: Create layouts for global and LDS memory.
+
+```cpp
+    // Specify layouts for global memory.
+    const auto a_global_layout =
+        ck::wrapper::make_layout(ck::make_tuple(M, K), ck::make_tuple(K, 1));
+    const auto b_global_layout =
+        ck::wrapper::make_layout(ck::make_tuple(N, K), ck::make_tuple(K, 1));
+    const auto c_global_layout =
+        ck::wrapper::make_layout(ck::make_tuple(M, N), ck::make_tuple(N, 1));
+
+    // Specify layouts for tiles.
+    constexpr auto a_tile_layout = ck::wrapper::make_layout(
+        ck::make_tuple(MPerBlock, KPerBlock), ck::make_tuple(KPerBlock, ck::Number<1>{}));
+    constexpr auto b_tile_layout = ck::wrapper::make_layout(
+        ck::make_tuple(NPerBlock, KPerBlock), ck::make_tuple(KPerBlock, ck::Number<1>{}));
+    constexpr auto c_tile_layout = ck::wrapper::make_layout(
+        ck::make_tuple(MPerBlock, NPerBlock), ck::make_tuple(NPerBlock, ck::Number<1>{}));
+
+    // Apply padding for global memory.
+    auto a_global_layout_padded = ck::wrapper::pad(a_global_layout, shape(a_tile_layout));
+    auto b_global_layout_padded = ck::wrapper::pad(b_global_layout, shape(b_tile_layout));
+    auto c_global_layout_padded = ck::wrapper::pad(c_global_layout, shape(c_tile_layout));
+```
+
+We pad layouts for global tensors in case M, N, and K are not divisible by `MPerBlock`, `NPerBlock`, or
+`KPerBlock`.
+
+Step 2: Create tensors for global and LDS memory.
+
+```cpp
+    // Make tensors for global memory.
+    auto a_global_tensor = ck::wrapper::make_tensor<ck::wrapper::MemoryTypeEnum::Global>(
+        static_cast<const DataType*>(p_a), a_global_layout_padded);
+    auto b_global_tensor = ck::wrapper::make_tensor<ck::wrapper::MemoryTypeEnum::Global>(
+        static_cast<const DataType*>(p_b), b_global_layout_padded);
+    auto c_global_tensor = ck::wrapper::make_tensor<ck::wrapper::MemoryTypeEnum::Global>(
+        static_cast<DataType*>(p_c), c_global_layout_padded);
+
+    // Allocate LDS memory.
+    __shared__ DataType lds_a[ck::wrapper::size(a_tile_layout)];
+    __shared__ DataType lds_b[ck::wrapper::size(b_tile_layout)];
+
+    // Make tensors for lds memory.
+    auto a_lds_tensor = ck::wrapper::make_tensor<ck::wrapper::MemoryTypeEnum::Lds>(
+        static_cast<DataType*>(lds_a), a_tile_layout);
+    auto b_lds_tensor = ck::wrapper::make_tensor<ck::wrapper::MemoryTypeEnum::Lds>(
+        static_cast<DataType*>(lds_b), b_tile_layout);
+```
+
+We must specify parameters for copy and convert block indexes to tuple:
+
+```cpp
+    // Specify block index as tuple.
+    const auto block_idxs = ck::make_tuple(static_cast<ck::index_t>(blockIdx.x),
+                                           static_cast<ck::index_t>(blockIdx.y),
+                                           ck::wrapper::slice());
+    // Specify access parameters for copy.
+    using DimAccessOrder             = ck::Tuple<ck::Number<0>, ck::Number<1>>;
+    constexpr ck::index_t vector_dim = 1;
+```
+
+We create a local tile (per block) and local partitions (per thread) for the global memory (`C`). We also
+define and clear an output register (`c_vgpr_reg`) for the accumulation.
+
+```cpp
+    auto c_global_local_tile = ck::wrapper::make_local_tile(
+        c_global_tensor,
+        tile_shape,
+        block_idxs,
+        make_tuple(ck::Number<1>{}, ck::Number<1>{}, ck::wrapper::slice(KPerBlock)));
+    auto c_global_local_partition =
+        ck::wrapper::make_blockwise_gemm_xdl_c_local_partition<DataType,
+                                                               decltype(a_tile_layout),
+                                                               decltype(b_tile_layout),
+                                                               ck::wrapper::size(thread_layout),
+                                                               GemmTraits>(c_global_local_tile);
+    // Create C vgpr to accumulate results.
+    auto c_vgpr_reg = ck::wrapper::make_blockwise_gemm_xdl_c_vgpr<DataType,
+                                                                  decltype(a_tile_layout),
+                                                                  decltype(b_tile_layout),
+                                                                  ck::wrapper::size(thread_layout),
+                                                                  GemmTraits>();
+    // Clear C vgpr.
+    ck::wrapper::clear(c_vgpr_reg);
+```
+
+We use two specific functions for `blockwise_gemm`: `make_blockwise_gemm_xdl_c_local_partition` and
+`make_blockwise_gemm_xdl_c_vgpr`. This helps to choose the appropriate partition for the `C` output
+and define tensors with specific layouts for `blockwise_gemm`. In the following step, we use only
+generic functions for the CK wrapper.
+
+Step 3: Create the compute loop.
+
+```cpp
+    const ck::index_t num_loop = ck::math::integer_divide_ceil(K, KPerBlock);
+    ck::index_t i              = 0;
+    do
+    {
+        // Get KPerBlock slice.
+        const auto k_slice           = ck::wrapper::slice(i * KPerBlock, (i + 1) * KPerBlock);
+        auto a_global_tensor_k_slice = a_global_tensor(ck::wrapper::slice(), k_slice);
+        auto b_global_tensor_k_slice = b_global_tensor(ck::wrapper::slice(), k_slice);
+        // Create local tiles for A and B.
+        auto a_global_local_tile = ck::wrapper::make_local_tile(
+            a_global_tensor_k_slice,
+            tile_shape,
+            block_idxs,
+            make_tuple(ck::Number<1>{}, ck::wrapper::slice(N), ck::Number<1>{}));
+        auto b_global_local_tile = ck::wrapper::make_local_tile(
+            b_global_tensor_k_slice,
+            tile_shape,
+            block_idxs,
+            make_tuple(ck::wrapper::slice(M), ck::Number<1>{}, ck::Number<1>{}));
+        // Copy from global to LDS.
+        ck::wrapper::blockwise_copy<DimAccessOrder, vector_dim, scalar_per_vector>(
+            a_global_local_tile, a_lds_tensor, thread_layout);
+        ck::wrapper::blockwise_copy<DimAccessOrder, vector_dim, scalar_per_vector>(
+            b_global_local_tile, b_lds_tensor, thread_layout);
+        // Synchronize lds.
+        ck::block_sync_lds();
+        // Execute blockwise GEMM.
+        ck::wrapper::blockwise_gemm_xdl<DataType, ck::wrapper::size(thread_layout), GemmTraits>(
+            a_lds_tensor, b_lds_tensor, c_vgpr_reg);
+
+        ++i;
+    } while(i < num_loop);
+```
+
+Loop iterate over `K / KPerBlock`. Each time a local tile is created for A and B tensors (tensor per block),
+data is copied from global memory to LDS. The `blockwise_gemm` function performs the GEMM
+operation on `a_lds_tensor` and `b_lds_tensor`, and stores results in `c_vgpr_reg`.
+
+The end result from `c_vgpr_reg` is stored in the `C` local partition (tensor per thread):
+
+```cpp
+    ck::wrapper::copy(c_vgpr_reg, c_global_local_partition);
+```
+
+If you want to dive deep into the details, you can find the entire example
+[here](https://github.com/ROCm/composable_kernel/blob/develop/client_example/25_wrapper/wrapper_basic_gemm.cpp).

client_example/25_wrapper/tensor_transform_using_wrapper.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023-2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iostream>
+
+#include "ck/ck.hpp"
+
+#include "ck/utility/number.hpp"
+#include "ck/utility/tuple.hpp"
+#include "ck/utility/sequence.hpp"
+
+#include "ck/wrapper/layout.hpp"
+
+using DataType = int;
+
+template <typename Layout>
+void Print1d(const Layout& layout)
+{
+    std::cout << "Print1d" << std::endl;
+    for(ck::index_t w = 0; w < ck::wrapper::size(layout); w++)
+    {
+        std::cout << layout(ck::make_tuple(w)) << " ";
+    }
+    std::cout << std::endl;
+}
+
+template <typename Layout>
+void Print2d(const Layout& layout)
+{
+    std::cout << "Print2d" << std::endl;
+    for(ck::index_t h = 0; h < ck::wrapper::size<0>(layout); h++)
+    {
+        for(ck::index_t w = 0; w < ck::wrapper::size<1>(layout); w++)
+        {
+            std::cout << layout(ck::make_tuple(h, w)) << " ";
+        }
+        std::cout << std::endl;
+    }
+}
+
+// Print in (x,y),z pattern
+template <typename Layout>
+void Print3dCustom(const Layout& layout)
+{
+    std::cout << "Print3dCustom" << std::endl;
+    for(ck::index_t d = 0; d < ck::wrapper::size<0>(ck::wrapper::get<0>(layout)); d++)
+    {
+        for(ck::index_t h = 0; h < ck::wrapper::size<1>(ck::wrapper::get<0>(layout)); h++)
+        {
+            for(ck::index_t w = 0; w < ck::wrapper::size<1>(layout); w++)
+            {
+                std::cout << layout(ck::make_tuple(ck::make_tuple(d, h), w)) << " ";
+            }
+            std::cout << std::endl;
+        }
+        std::cout << std::endl;
+    }
+}
+
+int main()
+{
+    // Layout traverse in row-major
+    std::cout << "Note: Layout traverse in column-major" << std::endl;
+    // Basic descriptor 0, 1, 2, ... 30, 31 (compile-time descriptor)
+    // (dims:4,8 strides:1,4)
+    const auto shape_4x8       = ck::make_tuple(ck::Number<4>{}, ck::Number<8>{});
+    const auto layout_4x8_s1x4 = ck::wrapper::make_layout(shape_4x8);
+    std::cout << "dims:4,8 strides:1,4" << std::endl;
+    Print2d(layout_4x8_s1x4);
+    using Cord1x1Type                = ck::Tuple<ck::Number<1>, ck::Number<1>>;
+    constexpr ck::index_t offset_1x1 = layout_4x8_s1x4.template operator()<Cord1x1Type>();
+    std::cout << "Constexpr calculated [1, 1] offset:" << offset_1x1 << std::endl;
+
+    // Basic descriptor 0, 1, 8, 9, 16, 17, ... 30, 31 (runtime descriptor)
+    // dims:4,(2,4) strides:2,(1,8)
+    const auto shape_4x2x4         = ck::make_tuple(4, ck::make_tuple(2, 4));
+    const auto strides_s2x1x8      = ck::make_tuple(2, ck::make_tuple(1, 8));
+    const auto layout_4x2x4_s2x1x8 = ck::wrapper::make_layout(shape_4x2x4, strides_s2x1x8);
+
+    std::cout << "dims:4,(2,4) strides:2,(1,8)" << std::endl;
+    Print2d(layout_4x2x4_s2x1x8);
+
+    // Basic descriptor 0, 1, 8, 9, 16, 17, ... 30, 31 (compile-time descriptor)
+    // dims:(2,2),(2,4) strides:((1,4),(2,8)
+    const auto shape_2x2x2x4    = ck::make_tuple(ck::make_tuple(ck::Number<2>{}, ck::Number<2>{}),
+                                              ck::make_tuple(ck::Number<2>{}, ck::Number<4>{}));
+    const auto strides_s1x4x2x8 = ck::make_tuple(ck::make_tuple(ck::Number<1>{}, ck::Number<4>{}),
+                                                 ck::make_tuple(ck::Number<2>{}, ck::Number<8>{}));
+    static const auto layout_2x2x2x4_s1x4x2x8 =
+        ck::wrapper::make_layout(shape_2x2x2x4, strides_s1x4x2x8);
+
+    std::cout << "dims:(2,2),(2,4) strides:(1,4),(2,8)" << std::endl;
+    Print2d(layout_2x2x2x4_s1x4x2x8);
+    Print3dCustom(layout_2x2x2x4_s1x4x2x8);
+
+    // Basic descriptor 0, 1, 8, 9, 16, 17, ... 30, 31 (compile-time descriptor)
+    // dims:((2,2),2),4 strides:((1,4),2),8
+    // Transform to 2d
+    const auto shape_2x2x2x4_nested = ck::make_tuple(
+        ck::make_tuple(ck::make_tuple(ck::Number<2>{}, ck::Number<2>{}), ck::Number<2>{}),
+        ck::Number<4>{});
+    const auto strides_s1x4x2x8_nested = ck::make_tuple(
+        ck::make_tuple(ck::make_tuple(ck::Number<1>{}, ck::Number<4>{}), ck::Number<2>{}),
+        ck::Number<8>{});
+    static const auto layout_2x2x2x4_s1x4x2x8_nested =
+        ck::wrapper::make_layout(shape_2x2x2x4_nested, strides_s1x4x2x8_nested);
+
+    std::cout << "dims:((2,2),2),4 strides:((1,4),2),8" << std::endl;
+    Print1d(layout_2x2x2x4_s1x4x2x8_nested);
+    Print2d(layout_2x2x2x4_s1x4x2x8_nested);
+    Print3dCustom(layout_2x2x2x4_s1x4x2x8_nested);
+
+    return 0;
+}

client_example/25_wrapper/wrapper_basic_gemm.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <numeric>
+#include <cstdlib>
+#include <iostream>
+#include <initializer_list>
+#include <vector>
+
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+
+#include "ck/host_utility/kernel_launch.hpp"
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/check_err.hpp"
+#include "ck/utility/common_header.hpp"
+#include "ck/library/utility/fill.hpp"
+#include "ck/wrapper/layout.hpp"
+#include "ck/wrapper/tensor.hpp"
+#include "ck/wrapper/operations/copy.hpp"
+#include "ck/wrapper/operations/gemm.hpp"
+#include "ck/wrapper/utils/kernel_utils.hpp"
+
+struct SimpleDeviceMem
+{
+    SimpleDeviceMem() = delete;
+
+    SimpleDeviceMem(std::size_t mem_size) : p_mem_{}
+    {
+        (void)hipMalloc(static_cast<void**>(&p_mem_), mem_size);
+    }
+
+    void* GetDeviceBuffer() { return p_mem_; }
+
+    ~SimpleDeviceMem() { (void)hipFree(p_mem_); }
+
+    void* p_mem_;
+};
+
+template <typename DataType,
+          typename GemmTraits,
+          ck::index_t scalar_per_vector,
+          typename BlockShape,
+          typename ThreadLayout>
+__global__ void __CK_WRAPPER_LAUNCH_BOUNDS__ DeviceGemm(const void* p_a,
+                                                        const void* p_b,
+                                                        void* p_c,
+                                                        const ck::index_t M,
+                                                        const ck::index_t N,
+                                                        const ck::index_t K,
+                                                        const BlockShape tile_shape,
+                                                        const ThreadLayout thread_layout)
+{
+    constexpr auto MPerBlock = ck::wrapper::size<0>(tile_shape);
+    constexpr auto NPerBlock = ck::wrapper::size<1>(tile_shape);
+    constexpr auto KPerBlock = ck::wrapper::size<2>(tile_shape);
+
+    // Specify layouts for global memory.
+    const auto a_global_layout =
+        ck::wrapper::make_layout(ck::make_tuple(M, K), ck::make_tuple(K, 1));
+    const auto b_global_layout =
+        ck::wrapper::make_layout(ck::make_tuple(N, K), ck::make_tuple(K, 1));
+    const auto c_global_layout =
+        ck::wrapper::make_layout(ck::make_tuple(M, N), ck::make_tuple(N, 1));
+    // Specify layouts for tiles.
+    constexpr auto a_tile_layout = ck::wrapper::make_layout(
+        ck::make_tuple(MPerBlock, KPerBlock), ck::make_tuple(KPerBlock, ck::Number<1>{}));
+    constexpr auto b_tile_layout = ck::wrapper::make_layout(
+        ck::make_tuple(NPerBlock, KPerBlock), ck::make_tuple(KPerBlock, ck::Number<1>{}));
+    constexpr auto c_tile_layout = ck::wrapper::make_layout(
+        ck::make_tuple(MPerBlock, NPerBlock), ck::make_tuple(NPerBlock, ck::Number<1>{}));
+    // Apply padding for global memory.
+    auto a_global_layout_padded = ck::wrapper::pad(a_global_layout, shape(a_tile_layout));
+    auto b_global_layout_padded = ck::wrapper::pad(b_global_layout, shape(b_tile_layout));
+    auto c_global_layout_padded = ck::wrapper::pad(c_global_layout, shape(c_tile_layout));
+    // Make tensors for global memory.
+    auto a_global_tensor = ck::wrapper::make_tensor<ck::wrapper::MemoryTypeEnum::Global>(
+        static_cast<const DataType*>(p_a), a_global_layout_padded);
+    auto b_global_tensor = ck::wrapper::make_tensor<ck::wrapper::MemoryTypeEnum::Global>(
+        static_cast<const DataType*>(p_b), b_global_layout_padded);
+    auto c_global_tensor = ck::wrapper::make_tensor<ck::wrapper::MemoryTypeEnum::Global>(
+        static_cast<DataType*>(p_c), c_global_layout_padded);
+    // Allocate lds memory.
+    __shared__ DataType lds_a[ck::wrapper::size(a_tile_layout)];
+    __shared__ DataType lds_b[ck::wrapper::size(b_tile_layout)];
+    // Make tensors for lds memory.
+    auto a_lds_tensor = ck::wrapper::make_tensor<ck::wrapper::MemoryTypeEnum::Lds>(
+        static_cast<DataType*>(lds_a), a_tile_layout);
+    auto b_lds_tensor = ck::wrapper::make_tensor<ck::wrapper::MemoryTypeEnum::Lds>(
+        static_cast<DataType*>(lds_b), b_tile_layout);
+    // Specify block index as tuple.
+    const auto block_idxs = ck::make_tuple(static_cast<ck::index_t>(blockIdx.x),
+                                           static_cast<ck::index_t>(blockIdx.y),
+                                           ck::wrapper::slice());
+    // Specify access parameters for copy.
+    using DimAccessOrder             = ck::Tuple<ck::Number<0>, ck::Number<1>>;
+    constexpr ck::index_t vector_dim = 1;
+    // Create tile and partition for C. Use specific function for blockwise_gemm to assign the
+    // appropriate partitions.
+    auto c_global_local_tile = ck::wrapper::make_local_tile(
+        c_global_tensor,
+        tile_shape,
+        block_idxs,
+        make_tuple(ck::Number<1>{}, ck::Number<1>{}, ck::wrapper::slice(KPerBlock)));
+    auto c_global_local_partition =
+        ck::wrapper::make_blockwise_gemm_xdl_c_local_partition<DataType,
+                                                               decltype(a_tile_layout),
+                                                               decltype(b_tile_layout),
+                                                               ck::wrapper::size(thread_layout),
+                                                               GemmTraits>(c_global_local_tile);
+    // Create C vgpr to accumulate results.
+    auto c_vgpr_reg = ck::wrapper::make_blockwise_gemm_xdl_c_vgpr<DataType,
+                                                                  decltype(a_tile_layout),
+                                                                  decltype(b_tile_layout),
+                                                                  ck::wrapper::size(thread_layout),
+                                                                  GemmTraits>();
+    // Clear C vgpr.
+    ck::wrapper::clear(c_vgpr_reg);
+
+    // Iterate over K with KPerBlock step.
+    const ck::index_t num_loop = ck::math::integer_divide_ceil(K, KPerBlock);
+    ck::index_t i              = 0;
+    do
+    {
+        // Get KPerBlock slice.
+        const auto k_slice           = ck::wrapper::slice(i * KPerBlock, (i + 1) * KPerBlock);
+        auto a_global_tensor_k_slice = a_global_tensor(ck::wrapper::slice(), k_slice);
+        auto b_global_tensor_k_slice = b_global_tensor(ck::wrapper::slice(), k_slice);
+        // Create local tiles for A and B.
+        auto a_global_local_tile = ck::wrapper::make_local_tile(
+            a_global_tensor_k_slice,
+            tile_shape,
+            block_idxs,
+            make_tuple(ck::Number<1>{}, ck::wrapper::slice(N), ck::Number<1>{}));
+        auto b_global_local_tile = ck::wrapper::make_local_tile(
+            b_global_tensor_k_slice,
+            tile_shape,
+            block_idxs,
+            make_tuple(ck::wrapper::slice(M), ck::Number<1>{}, ck::Number<1>{}));
+        // Copy from global to lds.
+        ck::wrapper::blockwise_copy<DimAccessOrder, vector_dim, scalar_per_vector>(
+            a_global_local_tile, a_lds_tensor, thread_layout);
+        ck::wrapper::blockwise_copy<DimAccessOrder, vector_dim, scalar_per_vector>(
+            b_global_local_tile, b_lds_tensor, thread_layout);
+        // Synchronize lds.
+        ck::block_sync_lds();
+        // Execute blockwise gemm.
+        ck::wrapper::blockwise_gemm_xdl<DataType, ck::wrapper::size(thread_layout), GemmTraits>(
+            a_lds_tensor, b_lds_tensor, c_vgpr_reg);
+
+        ++i;
+    } while(i < num_loop);
+    // Copy vgpr results to C global memory.
+    ck::wrapper::copy(c_vgpr_reg, c_global_local_partition);
+}
+
+template <typename DataType,
+          typename GemmTraits,
+          ck::index_t scalar_per_vector,
+          typename BlockShape,
+          typename ThreadLayout>
+void PerformGemm(const ck::index_t M,
+                 const ck::index_t N,
+                 const ck::index_t K,
+                 const BlockShape& tile_shape,
+                 const ThreadLayout& thread_layout)
+{
+    // Global memory buffers
+    SimpleDeviceMem a_mem(M * K * sizeof(DataType));
+    SimpleDeviceMem b_mem(K * N * sizeof(DataType));
+    SimpleDeviceMem c_mem(M * N * sizeof(DataType));
+
+    const ck::index_t grid_size_x =
+        ck::math::integer_divide_ceil(M, ck::wrapper::size<0>(tile_shape));
+    const ck::index_t grid_size_y =
+        ck::math::integer_divide_ceil(N, ck::wrapper::size<1>(tile_shape));
+
+    const auto kernel =
+        DeviceGemm<DataType, GemmTraits, scalar_per_vector, BlockShape, ThreadLayout>;
+    const float avg_time = launch_and_time_kernel(StreamConfig{nullptr, true},
+                                                  kernel,
+                                                  dim3(grid_size_x, grid_size_y, 1),
+                                                  dim3(ck::wrapper::size(thread_layout)),
+                                                  0,
+                                                  a_mem.GetDeviceBuffer(),
+                                                  b_mem.GetDeviceBuffer(),
+                                                  c_mem.GetDeviceBuffer(),
+                                                  M,
+                                                  N,
+                                                  K,
+                                                  tile_shape,
+                                                  thread_layout);
+
+    std::size_t flop = std::size_t(2) * M * N * K;
+    std::size_t num_btype =
+        sizeof(DataType) * M * K + sizeof(DataType) * K * N + sizeof(DataType) * M * N;
+
+    float tflops     = static_cast<float>(flop) / 1.E9 / avg_time;
+    float gb_per_sec = num_btype / 1.E6 / avg_time;
+
+    std::cout << "Perf: " << std::setw(10) << avg_time << " ms, " << tflops << " TFlops, "
+              << gb_per_sec << " GB/s, " << std::endl;
+}
+
+int main(int argc, char* argv[])
+{
+    using DataType = ck::half_t;
+    const auto thread_layout =
+        ck::wrapper::make_layout(ck::make_tuple(ck::Number<64>{}, ck::Number<4>{}),
+                                 ck::make_tuple(ck::Number<4>{}, ck::Number<1>{}));
+    const auto tile_shape = ck::make_tuple(ck::Number<256>{}, ck::Number<128>{}, ck::Number<32>{});
+    PerformGemm<DataType, ck::wrapper::BlockwisGemmXdlTraits_32x32Xdl_4x2XdlPerWave_8K1, 8>(
+        3840, 4096, 4096, tile_shape, thread_layout);
+    return 0;
+}
+// MI300X Perf:   0.471337 ms, 273.369 TFlops, 204.671 GB/s,

client_example/25_wrapper/wrapper_img2col.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <numeric>
+#include <cstdlib>
+#include <iomanip>
+#include <iostream>
+#include <initializer_list>
+#include <vector>
+
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+
+#include "ck/host_utility/kernel_launch.hpp"
+#include "ck/utility/common_header.hpp"
+#include "ck/wrapper/layout.hpp"
+#include "ck/wrapper/tensor.hpp"
+#include "ck/wrapper/operations/copy.hpp"
+#include "ck/wrapper/utils/kernel_utils.hpp"
+
+static constexpr ck::index_t NumDimSpatial = 3;
+using DataType                             = float;
+using InputLayout                          = ck::tensor_layout::convolution::NDHWGC;
+
+struct SimpleDeviceMem
+{
+    SimpleDeviceMem() = delete;
+
+    SimpleDeviceMem(std::size_t mem_size) : p_mem_{}
+    {
+        (void)hipMalloc(static_cast<void**>(&p_mem_), mem_size);
+    }
+
+    void* GetDeviceBuffer() { return p_mem_; }
+
+    ~SimpleDeviceMem() { (void)hipFree(p_mem_); }
+
+    void* p_mem_;
+};
+
+template <typename InputTensor, typename OutputTensor, typename BlockShape, typename ThreadLayout>
+__global__ void __CK_WRAPPER_LAUNCH_BOUNDS__
+DeviceImageToColumnPad0(InputTensor input_tensor,
+                        OutputTensor output_tensor,
+                        const BlockShape tile_shape,
+                        const ThreadLayout thread_layout)
+{
+    // grid layout (dim1, dim0)
+    const auto block_idxs =
+        ck::make_tuple(static_cast<ck::index_t>(blockIdx.y), static_cast<ck::index_t>(blockIdx.x));
+
+    // Get local tiles for global memory
+    auto input_local_tile  = ck::wrapper::make_local_tile(input_tensor, tile_shape, block_idxs);
+    auto output_local_tile = ck::wrapper::make_local_tile(output_tensor, tile_shape, block_idxs);
+
+    // Get partition per thread
+    const auto input_local_partition =
+        ck::wrapper::make_local_partition(input_local_tile, thread_layout, threadIdx.x);
+    auto output_local_partition =
+        ck::wrapper::make_local_partition(output_local_tile, thread_layout, threadIdx.x);
+
+    // Perform copy
+    using DimAccessOrder                    = ck::Tuple<ck::Number<0>, ck::Number<1>>;
+    constexpr ck::index_t vector_dim        = 1;
+    constexpr ck::index_t scalar_per_vector = 4;
+    ck::wrapper::copy<DimAccessOrder, vector_dim, scalar_per_vector>(input_local_partition,
+                                                                     output_local_partition);
+}
+
+void PerformImageToColumnPad0(const ck::index_t G,
+                              const ck::index_t N,
+                              const ck::index_t Di,
+                              const ck::index_t Hi,
+                              const ck::index_t Wi,
+                              const ck::index_t Do,
+                              const ck::index_t Ho,
+                              const ck::index_t Wo,
+                              const ck::index_t C,
+                              const ck::index_t Z,
+                              const ck::index_t Y,
+                              const ck::index_t X,
+                              std::array<ck::index_t, NumDimSpatial> filter_strides,
+                              std::array<ck::index_t, NumDimSpatial> filter_dilations)
+{
+    const ck::index_t ZYXC = Z * Y * X * C;
+    const ck::index_t GC   = G * C;
+
+    // shape: (G, (Wo, Ho, Do, N)), (C, X, Y, Z))
+    const auto shape = ck::make_tuple(ck::make_tuple(G, ck::make_tuple(Wo, Ho, Do, N)),
+                                      ck::make_tuple(C, X, Y, Z));
+    const auto in_strides =
+        ck::make_tuple(ck::make_tuple(C,
+                                      ck::make_tuple(filter_strides[2] * GC,
+                                                     filter_strides[1] * Wi * GC,
+                                                     filter_strides[0] * Hi * Wi * GC,
+                                                     Di * Hi * Wi * GC)),
+                       ck::make_tuple(1,
+                                      filter_dilations[2] * GC,
+                                      filter_dilations[1] * Wi * GC,
+                                      filter_dilations[0] * Hi * Wi * GC));
+    const auto in_layout = ck::wrapper::make_layout(shape, in_strides);
+
+    const auto out_strides = ck::make_tuple(
+        ck::make_tuple(
+            ZYXC,
+            ck::make_tuple(ZYXC * G, Wo * ZYXC * G, Ho * Wo * ZYXC * G, Do * Ho * Wo * ZYXC * G)),
+        ck::make_tuple(1, C, X * C, Y * X * C));
+    const auto out_layout = ck::wrapper::make_layout(shape, out_strides);
+
+    const ck::index_t input_size = N * Di * Hi * Wi * GC;
+    // Global memory buffers
+    SimpleDeviceMem in_buf(input_size * sizeof(DataType));
+    SimpleDeviceMem out_buf(ck::wrapper::size(out_layout) * sizeof(DataType));
+
+    // User can choose appropriate number of threads and sizes per block
+    const auto thread_layout =
+        ck::wrapper::make_layout(ck::make_tuple(ck::Number<8>{}, ck::Number<16>{}),
+                                 ck::make_tuple(ck::Number<16>{}, ck::Number<1>{}));
+    // This example doesn't support padding, user should select tile sizes
+    // which are divisible by the shape.
+    const auto tile_shape = ck::make_tuple(ck::Number<32>{}, ck::Number<64>{});
+
+    // Create buffers for global memory
+    auto input_tensor_global = ck::wrapper::make_tensor<ck::wrapper::MemoryTypeEnum::Global>(
+        static_cast<const DataType*>(in_buf.GetDeviceBuffer()), in_layout);
+    auto output_tensor_global = ck::wrapper::make_tensor<ck::wrapper::MemoryTypeEnum::Global>(
+        static_cast<DataType*>(out_buf.GetDeviceBuffer()), out_layout);
+
+    // grid layout (dim1, dim0)
+    const ck::index_t grid_size_x = ck::math::integer_divide_ceil(ck::wrapper::size<1>(in_layout),
+                                                                  ck::wrapper::size<1>(tile_shape));
+    const ck::index_t grid_size_y = ck::math::integer_divide_ceil(ck::wrapper::size<0>(in_layout),
+                                                                  ck::wrapper::size<0>(tile_shape));
+
+    const auto kernel    = DeviceImageToColumnPad0<decltype(input_tensor_global),
+                                                decltype(output_tensor_global),
+                                                decltype(tile_shape),
+                                                decltype(thread_layout)>;
+    const float avg_time = launch_and_time_kernel(StreamConfig{nullptr, true},
+                                                  kernel,
+                                                  dim3(grid_size_x, grid_size_y, 1),
+                                                  dim3(ck::wrapper::size(thread_layout)),
+                                                  0,
+                                                  input_tensor_global,
+                                                  output_tensor_global,
+                                                  tile_shape,
+                                                  thread_layout);
+
+    std::size_t num_btype = G * N * Do * Ho * Wo * ZYXC * 2 * sizeof(DataType);
+    float gb_per_sec      = num_btype / 1.E6 / avg_time;
+    std::cout << "Perf: " << std::setw(10) << avg_time << " ms, " << gb_per_sec << " GB/s, "
+              << std::endl;
+}
+
+int main(int argc, char* argv[])
+{
+    constexpr ck::index_t G  = 4;  // number of groups
+    constexpr ck::index_t N  = 32; // batch
+    constexpr ck::index_t C  = 64; // input channel (per group)
+    constexpr ck::index_t Z  = 3;  // filter D
+    constexpr ck::index_t Y  = 3;  // filter H
+    constexpr ck::index_t X  = 3;  // filter W
+    constexpr ck::index_t Di = 9;  // input D
+    constexpr ck::index_t Hi = 9;  // input H
+    constexpr ck::index_t Wi = 7;  // input W
+    constexpr ck::index_t Do = 7;  // output D
+    constexpr ck::index_t Ho = 7;  // output H
+    constexpr ck::index_t Wo = 5;  // output W
+    PerformImageToColumnPad0(G,
+                             N,
+                             Di,
+                             Hi,
+                             Wi,
+                             Do,
+                             Ho,
+                             Wo,
+                             C,
+                             Z,
+                             Y,
+                             X,
+                             {1, 1, 1} /*filter_strides*/,
+                             {1, 1, 1} /*filter_dilations*/);
+    return 0;
+}
+// MI100 Perf:   0.255178 ms, 1698.9 GB/s,

client_example/25_wrapper/wrapper_optimized_gemm.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <numeric>
+#include <cstdlib>
+#include <iostream>
+#include <initializer_list>
+#include <vector>
+
+#include "ck/library/utility/host_tensor.hpp"
+
+#include "ck/host_utility/kernel_launch.hpp"
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/check_err.hpp"
+#include "ck/utility/common_header.hpp"
+#include "ck/library/utility/fill.hpp"
+#include "ck/wrapper/layout.hpp"
+#include "ck/wrapper/tensor.hpp"
+#include "ck/wrapper/operations/copy.hpp"
+#include "ck/wrapper/operations/gemm.hpp"
+#include "ck/wrapper/utils/kernel_utils.hpp"
+
+struct SimpleDeviceMem
+{
+    SimpleDeviceMem() = delete;
+
+    SimpleDeviceMem(std::size_t mem_size) : p_mem_{}
+    {
+        (void)hipMalloc(static_cast<void**>(&p_mem_), mem_size);
+    }
+
+    void* GetDeviceBuffer() { return p_mem_; }
+
+    ~SimpleDeviceMem() { (void)hipFree(p_mem_); }
+
+    void* p_mem_;
+};
+
+template <bool DoPad, typename Layout, typename PaddingDims>
+__device__ auto ApplyPadding(const Layout& layout, const PaddingDims& padding_dims)
+{
+    if constexpr(DoPad)
+    {
+        return ck::wrapper::pad(layout, padding_dims);
+    }
+    else
+    {
+        return layout;
+    }
+}
+
+template <typename DataType,
+          typename GemmTraits,
+          ck::index_t scalar_per_vector,
+          typename BlockShape,
+          typename ThreadLayout,
+          bool DoPadding>
+__global__ void __CK_WRAPPER_LAUNCH_BOUNDS__ DeviceGemm(const void* p_a,
+                                                        const void* p_b,
+                                                        void* p_c,
+                                                        const ck::index_t M,
+                                                        const ck::index_t N,
+                                                        const ck::index_t K,
+                                                        const BlockShape tile_shape,
+                                                        const ThreadLayout thread_layout)
+{
+    constexpr auto MPerBlock  = ck::wrapper::size<0>(tile_shape);
+    constexpr auto NPerBlock  = ck::wrapper::size<1>(tile_shape);
+    constexpr auto KPerBlock  = ck::wrapper::size<2>(tile_shape);
+    constexpr auto K1         = GemmTraits::K1;
+    constexpr auto K0PerBlock = KPerBlock / K1;
+    const auto K0             = ck::math::integer_divide_ceil(K, K1);
+
+    const auto tile_shape_k0_m_n_k1 = ck::make_tuple(K0PerBlock, MPerBlock, NPerBlock, K1);
+    // Create layouts for global memory
+    const auto a_global_layout =
+        ck::wrapper::make_layout(ck::make_tuple(M, K), ck::make_tuple(K, 1));
+    const auto b_global_layout =
+        ck::wrapper::make_layout(ck::make_tuple(N, K), ck::make_tuple(K, 1));
+    const auto c_global_layout =
+        ck::wrapper::make_layout(ck::make_tuple(M, N), ck::make_tuple(N, 1));
+    // Apply padding
+    auto a_padded_global_layout =
+        ApplyPadding<DoPadding>(a_global_layout, ck::make_tuple(MPerBlock, KPerBlock));
+    auto b_padded_global_layout =
+        ApplyPadding<DoPadding>(b_global_layout, ck::make_tuple(NPerBlock, KPerBlock));
+    auto c_padded_global_layout =
+        ApplyPadding<DoPadding>(c_global_layout, ck::make_tuple(MPerBlock, NPerBlock));
+    // Reshape from M,K to K0,M,K1
+    const auto reshaped_dims_idxs =
+        ck::make_tuple(ck::Number<1>{}, ck::make_tuple(ck::Number<0>{}, ck::Number<2>{}));
+    auto a_padded_unmerged_global_layout =
+        ck::wrapper::unmerge<1>(a_padded_global_layout, ck::make_tuple(K0, K1), reshaped_dims_idxs);
+    auto b_padded_unmerged_global_layout =
+        ck::wrapper::unmerge<1>(b_padded_global_layout, ck::make_tuple(K0, K1), reshaped_dims_idxs);
+    // Create tensors for global memory
+    auto a_global_tensor = ck::wrapper::make_tensor<ck::wrapper::MemoryTypeEnum::Global>(
+        static_cast<const DataType*>(p_a), a_padded_unmerged_global_layout);
+    auto b_global_tensor = ck::wrapper::make_tensor<ck::wrapper::MemoryTypeEnum::Global>(
+        static_cast<const DataType*>(p_b), b_padded_unmerged_global_layout);
+    auto c_global_tensor = ck::wrapper::make_tensor<ck::wrapper::MemoryTypeEnum::Global>(
+        static_cast<DataType*>(p_c), c_padded_global_layout);
+    // Create layouts and tensors for lds memory.
+    constexpr auto a_tile_layout = ck::wrapper::make_layout(
+        ck::make_tuple(K0PerBlock, MPerBlock, K1),
+        ck::make_tuple((MPerBlock + ck::Number<1>{}) * K1, K1, ck::Number<1>{}));
+    constexpr auto b_tile_layout = ck::wrapper::make_layout(
+        ck::make_tuple(K0PerBlock, NPerBlock, K1),
+        ck::make_tuple((NPerBlock + ck::Number<1>{}) * K1, K1, ck::Number<1>{}));
+
+    __shared__ DataType lds_a[ck::wrapper::size(a_tile_layout) + K0PerBlock];
+    __shared__ DataType lds_b[ck::wrapper::size(b_tile_layout) + K0PerBlock];
+
+    auto a_lds_tensor = ck::wrapper::make_tensor<ck::wrapper::MemoryTypeEnum::Lds>(
+        static_cast<DataType*>(lds_a), a_tile_layout);
+    auto b_lds_tensor = ck::wrapper::make_tensor<ck::wrapper::MemoryTypeEnum::Lds>(
+        static_cast<DataType*>(lds_b), b_tile_layout);
+
+    const auto block_idxs            = ck::make_tuple(ck::wrapper::slice(),
+                                           static_cast<ck::index_t>(blockIdx.x),
+                                           static_cast<ck::index_t>(blockIdx.y),
+                                           ck::wrapper::slice());
+    using DimAccessOrder             = ck::Tuple<ck::Number<1>, ck::Number<0>, ck::Number<2>>;
+    constexpr ck::index_t vector_dim = 2;
+
+    // Create tile and partition for C global memory. Use specific gemm
+    // functions to get appropriate layouts.
+    auto c_global_local_tile =
+        ck::wrapper::make_local_tile(c_global_tensor,
+                                     tile_shape_k0_m_n_k1,
+                                     block_idxs,
+                                     make_tuple(ck::wrapper::slice(K0PerBlock),
+                                                ck::Number<1>{},
+                                                ck::Number<1>{},
+                                                ck::wrapper::slice(K1)));
+    auto c_global_local_partition =
+        ck::wrapper::make_blockwise_gemm_xdl_c_local_partition<DataType,
+                                                               decltype(a_tile_layout),
+                                                               decltype(b_tile_layout),
+                                                               ck::wrapper::size(thread_layout),
+                                                               GemmTraits>(c_global_local_tile);
+    // Define and clear c vgpr register
+    auto c_vgpr_reg = ck::wrapper::make_blockwise_gemm_xdl_c_vgpr<DataType,
+                                                                  decltype(a_tile_layout),
+                                                                  decltype(b_tile_layout),
+                                                                  ck::wrapper::size(thread_layout),
+                                                                  GemmTraits>();
+    ck::wrapper::clear(c_vgpr_reg);
+    // Local partitions for lds memory
+    auto a_lds_tensor_local_partition =
+        ck::wrapper::make_local_partition(a_lds_tensor, thread_layout, threadIdx.x);
+    auto b_lds_tensor_local_partition =
+        ck::wrapper::make_local_partition(b_lds_tensor, thread_layout, threadIdx.x);
+    // Lamda to slice tensor, then create local tile and partition
+    auto make_global_partition = [&](auto tensor, auto projection, ck::index_t i) {
+        const auto k_slice =
+            ck::make_tuple(ck::wrapper::slice(i * K0PerBlock, (i + 1) * K0PerBlock),
+                           ck::wrapper::slice(),
+                           ck::wrapper::slice());
+        auto local_tile = ck::wrapper::make_local_tile(
+            tensor(k_slice), tile_shape_k0_m_n_k1, block_idxs, projection);
+        return ck::wrapper::make_local_partition(local_tile, thread_layout, threadIdx.x);
+    };
+
+    auto a_global_local_partition = make_global_partition(
+        a_global_tensor,
+        make_tuple(ck::Number<1>{}, ck::Number<1>{}, ck::wrapper::slice(N), ck::Number<1>{}),
+        0);
+    auto b_global_local_partition = make_global_partition(
+        b_global_tensor,
+        make_tuple(ck::Number<1>{}, ck::wrapper::slice(M), ck::Number<1>{}, ck::Number<1>{}),
+        0);
+
+    // (row-major vgpr layout)
+    auto a_vgpr_tensor =
+        ck::wrapper::make_register_tensor<ck::wrapper::MemoryTypeEnum::Vgpr, DataType>(
+            ck::wrapper::make_layout(
+                shape(a_global_local_partition),
+                ck::make_tuple(ck::wrapper::size<1>(a_global_local_partition) *
+                                   ck::wrapper::size<2>(a_global_local_partition),
+                               ck::wrapper::size<2>(a_global_local_partition),
+                               ck::Number<1>{})));
+    auto b_vgpr_tensor =
+        ck::wrapper::make_register_tensor<ck::wrapper::MemoryTypeEnum::Vgpr, DataType>(
+            ck::wrapper::make_layout(
+                shape(b_global_local_partition),
+                ck::make_tuple(ck::wrapper::size<1>(a_global_local_partition) *
+                                   ck::wrapper::size<2>(a_global_local_partition),
+                               ck::wrapper::size<2>(a_global_local_partition),
+                               ck::Number<1>{})));
+    // Copy first values to lds
+    ck::wrapper::copy<DimAccessOrder, vector_dim, scalar_per_vector>(a_global_local_partition,
+                                                                     a_vgpr_tensor);
+    ck::wrapper::copy<DimAccessOrder, vector_dim, scalar_per_vector>(b_global_local_partition,
+                                                                     b_vgpr_tensor);
+    ck::wrapper::copy<DimAccessOrder, vector_dim, scalar_per_vector>(a_vgpr_tensor,
+                                                                     a_lds_tensor_local_partition);
+    ck::wrapper::copy<DimAccessOrder, vector_dim, scalar_per_vector>(b_vgpr_tensor,
+                                                                     b_lds_tensor_local_partition);
+    // Pipeline loop
+    const ck::index_t num_loop =
+        __builtin_amdgcn_readfirstlane(ck::math::integer_divide_ceil(K, KPerBlock));
+    // Skip if only tile should be processed
+    if(num_loop > 1)
+    {
+        ck::index_t i = 0;
+        do
+        {
+            auto a_global_local_partition_i = make_global_partition(
+                a_global_tensor,
+                make_tuple(
+                    ck::Number<1>{}, ck::Number<1>{}, ck::wrapper::slice(N), ck::Number<1>{}),
+                i + 1);
+            auto b_global_local_partition_i = make_global_partition(
+                b_global_tensor,
+                make_tuple(
+                    ck::Number<1>{}, ck::wrapper::slice(M), ck::Number<1>{}, ck::Number<1>{}),
+                i + 1);
+            // Copy data to A vgpr.
+            ck::wrapper::copy<DimAccessOrder, vector_dim, scalar_per_vector>(
+                a_global_local_partition_i, a_vgpr_tensor);
+            // Synchronize.
+            ck::block_sync_lds();
+            // Copy data to B vgpr.
+            ck::wrapper::copy<DimAccessOrder, vector_dim, scalar_per_vector>(
+                b_global_local_partition_i, b_vgpr_tensor);
+            // Perform gemm.
+            ck::wrapper::blockwise_gemm_xdl<DataType, ck::wrapper::size(thread_layout), GemmTraits>(
+                a_lds_tensor, b_lds_tensor, c_vgpr_reg);
+            // Synchronize
+            ck::block_sync_lds();
+            // Copy data to A and B lds tiles.
+            ck::wrapper::copy<DimAccessOrder, vector_dim, scalar_per_vector>(
+                a_vgpr_tensor, a_lds_tensor_local_partition);
+            ck::wrapper::copy<DimAccessOrder, vector_dim, scalar_per_vector>(
+                b_vgpr_tensor, b_lds_tensor_local_partition);
+
+            ++i;
+        } while(i < (num_loop - 1));
+    }
+    // Handle tail.
+    ck::block_sync_lds();
+    ck::wrapper::blockwise_gemm_xdl<DataType, ck::wrapper::size(thread_layout), GemmTraits>(
+        a_lds_tensor, b_lds_tensor, c_vgpr_reg);
+    // Store data from C vgpr to C global memory.
+    ck::wrapper::copy(c_vgpr_reg, c_global_local_partition);
+}
+
+template <typename DataType,
+          typename GemmTraits,
+          ck::index_t scalar_per_vector,
+          bool DoPadding,
+          typename BlockShape,
+          typename ThreadLayout>
+void PerformGemm(const ck::index_t M,
+                 const ck::index_t N,
+                 const ck::index_t K,
+                 const BlockShape& tile_shape,
+                 const ThreadLayout& thread_layout)
+{
+    // Global memory buffers
+    SimpleDeviceMem a_mem(M * K * sizeof(DataType));
+    SimpleDeviceMem b_mem(K * N * sizeof(DataType));
+    SimpleDeviceMem c_mem(M * N * sizeof(DataType));
+
+    const ck::index_t grid_size_x =
+        ck::math::integer_divide_ceil(M, ck::wrapper::size<0>(tile_shape));
+    const ck::index_t grid_size_y =
+        ck::math::integer_divide_ceil(N, ck::wrapper::size<1>(tile_shape));
+
+    const auto kernel =
+        DeviceGemm<DataType, GemmTraits, scalar_per_vector, BlockShape, ThreadLayout, DoPadding>;
+    const float avg_time = launch_and_time_kernel(StreamConfig{nullptr, true},
+                                                  kernel,
+                                                  dim3(grid_size_x, grid_size_y, 1),
+                                                  dim3(ck::wrapper::size(thread_layout)),
+                                                  0,
+                                                  a_mem.GetDeviceBuffer(),
+                                                  b_mem.GetDeviceBuffer(),
+                                                  c_mem.GetDeviceBuffer(),
+                                                  M,
+                                                  N,
+                                                  K,
+                                                  tile_shape,
+                                                  thread_layout);
+    std::size_t flop     = std::size_t(2) * M * N * K;
+    std::size_t num_btype =
+        sizeof(DataType) * M * K + sizeof(DataType) * K * N + sizeof(DataType) * M * N;
+
+    float tflops     = static_cast<float>(flop) / 1.E9 / avg_time;
+    float gb_per_sec = num_btype / 1.E6 / avg_time;
+
+    std::cout << "Perf: " << std::setw(10) << avg_time << " ms, " << tflops << " TFlops, "
+              << gb_per_sec << " GB/s, " << std::endl;
+}
+
+int main(int argc, char* argv[])
+{
+    using DataType = ck::half_t;
+    const auto thread_layout =
+        ck::wrapper::make_layout(ck::make_tuple(ck::Number<4>{}, ck::Number<64>{}, ck::Number<1>{}),
+                                 ck::make_tuple(ck::Number<1>{}, ck::Number<4>{}, ck::Number<1>{}));
+    const auto tile_shape = ck::make_tuple(ck::Number<256>{}, ck::Number<128>{}, ck::Number<32>{});
+    PerformGemm<DataType, ck::wrapper::BlockwisGemmXdlTraits_32x32Xdl_4x2XdlPerWave_8K1, 8, false>(
+        3840, 4096, 4096, tile_shape, thread_layout);
+    return 0;
+}
+// MI300X Perf:   0.411552 ms, 313.081 TFlops, 234.403 GB/s,

client_example/CMakeLists.txt

@@ -2,7 +2,53 @@ cmake_minimum_required(VERSION 3.15)
 project(ck_app)
 add_compile_options(-std=c++17)
 
-find_package(composable_kernel 1.0.0 COMPONENTS device_operations)
+if (DTYPES)
+    add_definitions(-DDTYPES)
+    if (DTYPES MATCHES "int8")
+        add_definitions(-DCK_ENABLE_INT8)
+        if(NOT DEFINED ${CK_ENABLE_INT8})
+            set(CK_ENABLE_INT8 "ON")
+        endif()
+    endif()
+    if (DTYPES MATCHES "fp8")
+        add_definitions(-DCK_ENABLE_FP8)
+        if(NOT DEFINED ${CK_ENABLE_FP8})
+            set(CK_ENABLE_FP8 "ON")
+        endif()
+    endif()
+    if (DTYPES MATCHES "fp16")
+        add_definitions(-DCK_ENABLE_FP16)
+        if(NOT DEFINED ${CK_ENABLE_FP16})
+            set(CK_ENABLE_FP16 "ON")
+        endif()
+    endif()
+    if (DTYPES MATCHES "fp32")
+        add_definitions(-DCK_ENABLE_FP32)
+        if(NOT DEFINED ${CK_ENABLE_FP32})
+            set(CK_ENABLE_FP32 "ON")
+        endif()
+    endif()
+    if (DTYPES MATCHES "fp64")
+        add_definitions(-DCK_ENABLE_FP64)
+        if(NOT DEFINED ${CK_ENABLE_FP64})
+            set(CK_ENABLE_FP64 "ON")
+        endif()
+    endif()
+    if (DTYPES MATCHES "bf16")
+        add_definitions(-DCK_ENABLE_BF16)
+        if(NOT DEFINED ${CK_ENABLE_BF16})
+            set(CK_ENABLE_BF16 "ON")
+        endif()
+    endif()
+    message("DTYPES macro set to ${DTYPES}")
+else()
+    add_definitions(-DCK_ENABLE_INT8 -DCK_ENABLE_FP8 -DCK_ENABLE_FP16 -DCK_ENABLE_FP32 -DCK_ENABLE_FP64 -DCK_ENABLE_BF16)
+    if(NOT DEFINED ${CK_ENABLE_ALL_DTYPES})
+        set(CK_ENABLE_ALL_DTYPES "ON")
+    endif()
+endif()
+
+find_package(composable_kernel COMPONENTS device_other_operations device_gemm_operations device_conv_operations device_contraction_operations device_reduction_operations)
 find_package(hip REQUIRED PATHS /opt/rocm)
 message(STATUS "Build with HIP ${hip_VERSION}")
 

cmake/ClangTidy.cmake

@@ -149,7 +149,7 @@ function(clang_tidy_check TARGET)
             add_custom_target(${tidy_target}
                 # for some targets clang-tidy not able to get information from .clang-tidy
                 DEPENDS ${SOURCE}
-                COMMAND ${CLANG_TIDY_COMMAND} "-config=\{CheckOptions: \[\{key: bugprone-reserved-identifier.AllowedIdentifiers,value: __HIP_PLATFORM_HCC__\; __HIP_ROCclr__\}\]\}" ${SOURCE} "-export-fixes=${CLANG_TIDY_FIXIT_DIR}/${TARGET}-${tidy_file}.yaml"
+                COMMAND ${CLANG_TIDY_COMMAND} "-config=\{CheckOptions: \[\{key: bugprone-reserved-identifier.AllowedIdentifiers,value: __HIP_PLATFORM_HCC__\; __HIP_PLATFORM_AMD__\; __HIP_ROCclr__\}\]\}" ${SOURCE} "-export-fixes=${CLANG_TIDY_FIXIT_DIR}/${TARGET}-${tidy_file}.yaml"
                 WORKING_DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR}
                 COMMENT "clang-tidy: Running clang-tidy on target ${SOURCE}..."
             )