Add bias version of example

composable_kernel/include/tensor_operation/element_wise_operation.hpp

 #ifndef CK_ELEMENT_WISE_OPERATION_HPP
 #define CK_ELEMENT_WISE_OPERATION_HPP
 
+#include "data_type.hpp"
+
 namespace ck {
 namespace tensor_operation {
 namespace element_wise {
@@ -21,6 +23,20 @@ struct PassThrough
     }
 };
 
+// TODO - Move bias functor to the kernel
+struct Bias
+{
+    __host__ __device__ constexpr void operator()(float& y_plus_bias, const float& y, const float& bias) const
+    {
+        y_plus_bias = y + bias;
+    }
+
+    __host__ __device__ constexpr void operator()(half_t& y_plus_bias, const half_t& y, const float& bias) const
+    {
+        y_plus_bias = y + bias;
+    }
+};
+
 struct AddRelu
 {
     template <typename T>

example/2_gemm_xdl_bias/README.md

+# Instructions for ```gemm_xdl``` Example
+
+## Docker script
+```bash
+docker run                                                                   \
+-it                                                                          \
+--rm                                                                         \
+--privileged                                                                 \
+--group-add sudo                                                             \
+-w /root/workspace                                                           \
+-v ${PATH_TO_LOCAL_WORKSPACE}:/root/workspace                                \
+rocm/tensorflow:rocm4.3.1-tf2.6-dev                                          \
+/bin/bash
+```
+
+## Build ```gemm_xdl```
+```bash
+mkdir build && cd build
+```
+
+```bash
+# Need to specify target ID, example below is gfx908
+cmake                                                                  \
+-D BUILD_DEV=OFF                                                       \
+-D CMAKE_BUILD_TYPE=Release                                            \
+-D CMAKE_CXX_FLAGS="-DCK_AMD_GPU_GFX908 --amdgpu-target=gfx908 -O3 "   \
+-D CMAKE_CXX_COMPILER=/opt/rocm/bin/hipcc                              \
+-D CMAKE_PREFIX_PATH=/opt/rocm                                         \
+..
+```
+
+```bash
+ make -j gemm_xdl
+```
+
+## Run ```gemm_xdl```
+```bash
+#arg1: verification (0=no, 1=yes)
+#arg2: initialization (0=no init, 1=integer value, 2=decimal value)
+#arg3: run kernel # of times (>1)
+./example/gemm_xdl 0 1 5
+```
+
+Result (MI100 @ 1087Mhz, 133.5TFlops peak FP16)
+```
+a_m_k: dim 2, lengths {3840, 4096}, strides {4096, 1}
+b_k_n: dim 2, lengths {4096, 4096}, strides {1, 4096}
+c_m_n: dim 2, lengths {3840, 4096}, strides {4096, 1}
+arg.a_grid_desc_k0_m_k1_{512, 3840, 8}
+arg.b_grid_desc_k0_n_k1_{512, 4096, 8}
+arg.c_grid_desc_m_n_{ 3840, 4096}
+launch_and_time_kernel: grid_dim {480, 1, 1}, block_dim {256, 1, 1}
+Warm up
+Start running 5 times...
+Perf: 1.19685 ms, 107.657 TFlops, 78.8501 GB/s
+```

example/2_gemm_xdl_bias/gemm_xdl_bias.cpp

+#include <iostream>
+#include <numeric>
+#include <initializer_list>
+#include <cstdlib>
+#include <stdlib.h>
+#include <half.hpp>
+#include "config.hpp"
+#include "print.hpp"
+#include "device.hpp"
+#include "host_tensor.hpp"
+#include "host_tensor_generator.hpp"
+#include "host_gemm.hpp"
+#include "device_tensor.hpp"
+#include "device_base.hpp"
+#include "device_gemm_xdl_c_shuffle_bias.hpp"
+#include "element_wise_operation.hpp"
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using ADataType   = ck::half_t;
+using BDataType   = ck::half_t;
+using CDataType   = ck::half_t;
+using AccDataType = float;
+// TODO
+// using C0DataType  = float;
+
+using ALayout = ck::tensor_layout::gemm::RowMajor;
+using BLayout = ck::tensor_layout::gemm::ColumnMajor;
+using CLayout = ck::tensor_layout::gemm::RowMajor;
+
+using AElementOp = ck::tensor_operation::element_wise::PassThrough;
+using BElementOp = ck::tensor_operation::element_wise::PassThrough;
+using CElementOp = ck::tensor_operation::element_wise::Bias;
+
+// clang-format off
+using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemmXdl_C_Shuffle_Bias<
+    ADataType,              // ADataType
+    BDataType,              // BDataType
+    CDataType,              // CDataType
+    AccDataType,            // AccDataType
+    ALayout,                // ALayout
+    BLayout,                // BLayout
+    CLayout,                // CLayout
+    AElementOp,             // AElementwiseOperation
+    BElementOp,             // BElementwiseOperation
+    CElementOp,             // CElementwiseOperation
+    256,                    // BlockSize
+    256,                    // MPerBlock
+    128,                    // NPerBlock
+    4,                      // K0PerBlock
+    8,                      // K1
+    32,                     // MPerXDL
+    32,                     // NPerXDL
+    4,                      // MXdlPerWave
+    2,                      // NXdlPerWave
+    S<4, 64, 1>,            // ABlockTransferThreadClusterLengths_K0_M_K1
+    S<1, 0, 2>,             // ABlockTransferThreadClusterArrangeOrder
+    S<1, 0, 2>,             // ABlockTransferSrcAccessOrder
+    2,                      // ABlockTransferSrcVectorDim
+    8,                      // ABlockTransferSrcScalarPerVector
+    8,                      // ABlockTransferDstScalarPerVector_K1
+    true,                   // ABlockLdsAddExtraM
+    S<4, 64, 1>,            // BBlockTransferThreadClusterLengths_K0_N_K1
+    S<1, 0, 2>,             // BBlockTransferThreadClusterArrangeOrder
+    S<1, 0, 2>,             // BBlockTransferSrcAccessOrder
+    2,                      // BBlockTransferSrcVectorDim
+    8,                      // BBlockTransferSrcScalarPerVector
+    8,                      // BBlockTransferDstScalarPerVector_K1
+    true,                   // BBlockLdsAddExtraN
+    1,                      // CShuffleMXdlPerWavePerShuffle
+    1,                      // CShuffleNXdlPerWavePerShuffle
+    S<1, 1, 32, 1, 1, 8>,   // CBlockTransferClusterLengths_MBlock_MXdlPerWave_MWaveMPerXdl_NBlock_NXdlPerWave_NWaveNPerXdl
+    8>;                     // CBlockTransferScalarPerVector_NWaveNPerXdl
+// clang-format on
+
+template <typename AType,
+          typename BType,
+          typename CType,
+          typename C0Type,
+          typename AElementwiseOperation,
+          typename BElementwiseOperation,
+          typename CElementwiseOperation>
+static void host_verify(const Tensor<AType>& a_m_k,
+                        const Tensor<BType>& b_k_n,
+                        const Tensor<C0Type>& c0_k_n,
+                        Tensor<CType>& c_m_n,
+                        const AElementwiseOperation& a_element_op,
+                        const BElementwiseOperation& b_element_op,
+                        const CElementwiseOperation& c_element_op)
+{
+    auto f_mk_kn_mn = [&](auto m, auto n) {
+        const int K = a_m_k.mDesc.GetLengths()[1];
+
+        AccDataType v = 0;
+
+        for(int k = 0; k < K; ++k)
+        {
+            v += static_cast<AccDataType>(a_element_op(a_m_k(m, k))) *
+                 static_cast<AccDataType>(b_element_op(b_k_n(k, n)));
+        }
+
+        c_element_op(c_m_n(m, n), v, c0_k_n(m, n));
+    };
+
+    make_ParallelTensorFunctor(f_mk_kn_mn,
+                               c_m_n.mDesc.GetLengths()[0],
+                               c_m_n.mDesc.GetLengths()[1])(std::thread::hardware_concurrency());
+}
+
+int main(int argc, char* argv[])
+{
+    bool do_verification = 0;
+    int init_method      = 0;
+    int nrepeat          = 5;
+
+    // GEMM shape
+    ck::index_t M = 3840;
+    ck::index_t N = 4096;
+    ck::index_t K = 4096;
+
+    ck::index_t StrideA = 4096;
+    ck::index_t StrideB = 4096;
+    ck::index_t StrideC = 4096;
+
+    if(argc == 4)
+    {
+        do_verification = std::stoi(argv[1]);
+        init_method     = std::stoi(argv[2]);
+        nrepeat         = std::stoi(argv[3]);
+    }
+    else if(argc == 10)
+    {
+        do_verification = std::stoi(argv[1]);
+        init_method     = std::stoi(argv[2]);
+        nrepeat         = std::stoi(argv[3]);
+
+        M = std::stoi(argv[4]);
+        N = std::stoi(argv[5]);
+        K = std::stoi(argv[6]);
+
+        StrideA = std::stoi(argv[7]);
+        StrideB = std::stoi(argv[8]);
+        StrideC = std::stoi(argv[9]);
+    }
+    else
+    {
+        printf("arg1: verification (0=no, 1=yes)\n");
+        printf("arg2: initialization (0=no init, 1=integer value, 2=decimal value)\n");
+        printf("arg3: run kernel # of times (>1)\n");
+        printf("arg4 to 9: M (256x), N(128x), K(32x), StrideA, StrideB, StrideC\n");
+        exit(0);
+    }
+
+    auto f_host_tensor_descriptor =
+        [](std::size_t row, std::size_t col, std::size_t stride, auto layout) {
+            if(std::is_same<decltype(layout), ck::tensor_layout::gemm::RowMajor>::value)
+            {
+                return HostTensorDescriptor(std::vector<std::size_t>({row, col}),
+                                            std::vector<std::size_t>({stride, 1}));
+            }
+            else
+            {
+                return HostTensorDescriptor(std::vector<std::size_t>({row, col}),
+                                            std::vector<std::size_t>({1, stride}));
+            }
+        };
+
+    auto f_bias_host_tensor_descriptor = [](std::size_t row, std::size_t col) {
+        // TODO - stride with {0, 1}
+        return HostTensorDescriptor(std::vector<std::size_t>({row, col}),
+                                    std::vector<std::size_t>({1, 0}));
+    };
+
+    Tensor<ADataType> a_m_k(f_host_tensor_descriptor(M, K, StrideA, ALayout{}));
+    Tensor<BDataType> b_k_n(f_host_tensor_descriptor(K, N, StrideB, BLayout{}));
+    Tensor<BDataType> c0_m_n(f_bias_host_tensor_descriptor(M, N));
+    Tensor<BDataType> c_m_n_host_result(f_host_tensor_descriptor(M, N, StrideC, CLayout{}));
+    Tensor<BDataType> c_m_n_device_result(f_host_tensor_descriptor(M, N, StrideC, CLayout{}));
+
+    std::cout << "a_m_k: " << a_m_k.mDesc << std::endl;
+    std::cout << "b_k_n: " << b_k_n.mDesc << std::endl;
+    std::cout << "c0_m_n: " << c0_m_n.mDesc << std::endl;
+    std::cout << "c_m_n: " << c_m_n_host_result.mDesc << std::endl;
+
+    switch(init_method)
+    {
+    case 0: break;
+    case 1:
+        a_m_k.GenerateTensorValue(GeneratorTensor_2<ADataType>{-5, 5});
+        b_k_n.GenerateTensorValue(GeneratorTensor_2<BDataType>{-5, 5});
+        c0_m_n.GenerateTensorValue(GeneratorTensor_2<BDataType>{-5, 5});
+        break;
+    default:
+        a_m_k.GenerateTensorValue(GeneratorTensor_3<ADataType>{0.0, 1.0});
+        b_k_n.GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5});
+        c0_m_n.GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5});
+    }
+
+    DeviceMem a_m_k_device_buf(sizeof(ADataType) * a_m_k.mDesc.GetElementSpace());
+    DeviceMem b_k_n_device_buf(sizeof(BDataType) * b_k_n.mDesc.GetElementSpace());
+    DeviceMem c0_m_n_device_buf(sizeof(CDataType) * c0_m_n.mDesc.GetElementSpace());
+    DeviceMem c_m_n_device_buf(sizeof(CDataType) * c_m_n_device_result.mDesc.GetElementSpace());
+
+    a_m_k_device_buf.ToDevice(a_m_k.mData.data());
+    b_k_n_device_buf.ToDevice(b_k_n.mData.data());
+    c0_m_n_device_buf.ToDevice(c0_m_n.mData.data());
+    c_m_n_device_buf.ToDevice(c_m_n_device_result.mData.data());
+
+    // do GEMM
+    auto gemm     = DeviceGemmInstance{};
+    auto invoker  = gemm.MakeInvoker();
+    auto argument = gemm.MakeArgument(static_cast<ADataType*>(a_m_k_device_buf.GetDeviceBuffer()),
+                                      static_cast<BDataType*>(b_k_n_device_buf.GetDeviceBuffer()),
+                                      static_cast<CDataType*>(c0_m_n_device_buf.GetDeviceBuffer()),
+                                      static_cast<CDataType*>(c_m_n_device_buf.GetDeviceBuffer()),
+                                      M,
+                                      N,
+                                      K,
+                                      StrideA,
+                                      StrideB,
+                                      StrideC,
+                                      AElementOp{},
+                                      BElementOp{},
+                                      CElementOp{});
+
+    if(!gemm.IsSupportedArgument(argument))
+    {
+        throw std::runtime_error(
+            "wrong! device_gemm with the specified compilation parameters does "
+            "not support this GEMM problem");
+    }
+
+    float ave_time = invoker.Run(argument, nrepeat);
+
+    std::size_t flop = std::size_t(2) * M * N * K;
+    std::size_t num_btype =
+        sizeof(ADataType) * M * K + sizeof(BDataType) * K * N + sizeof(CDataType) * M * N;
+
+    float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
+
+    float gb_per_sec = num_btype / 1.E6 / ave_time;
+
+    std::cout << "Perf: " << ave_time << " ms, " << tflops << " TFlops, " << gb_per_sec << " GB/s"
+              << std::endl;
+
+    c_m_n_device_buf.FromDevice(c_m_n_device_result.mData.data());
+
+    if(do_verification)
+    {
+        host_verify(
+            a_m_k, b_k_n, c0_m_n, c_m_n_host_result, AElementOp{}, BElementOp{}, CElementOp{});
+
+        check_error(c_m_n_host_result, c_m_n_device_result);
+    }
+}

example/CMakeLists.txt

@@ -12,6 +12,7 @@ include_directories(BEFORE
 )
 
 set(GEMM_XDL_SOURCE 1_gemm_xdl/gemm_xdl.cpp)
+set(GEMM_XDL_BIAS_SOURCE 2_gemm_xdl_bias/gemm_xdl_bias.cpp)
 set(GEMM_XDL_BIAS_RELU_ADD_SOURCE 3_gemm_xdl_bias_relu_add/gemm_xdl_bias_relu_add.cpp)
 set(CONV2D_FWD_XDL_SOURCE 4_conv2d_fwd_xdl/conv2d_fwd_xdl.cpp)
 set(CONV2D_FWD_XDL_BIAS_RELU_SOURCE 5_conv2d_fwd_xdl_bias_relu/conv2d_fwd_xdl_bias_relu.cpp)
@@ -19,6 +20,7 @@ set(CONV2D_FWD_XDL_BIAS_RELU_ADD_SOURCE 6_conv2d_fwd_xdl_bias_relu_add/conv2d_fw
 set(CONV2D_FWD_XDL_BIAS_RELU_ATOMIC_ADD_SOURCE 7_conv2d_fwd_xdl_bias_relu_atomic_add/conv2d_fwd_xdl_bias_relu_atomic_add.cpp)
 
 add_executable(gemm_xdl ${GEMM_XDL_SOURCE})
+add_executable(gemm_xdl_bias ${GEMM_XDL_BIAS_SOURCE})
 add_executable(gemm_xdl_bias_relu_add ${GEMM_XDL_BIAS_RELU_ADD_SOURCE})
 add_executable(conv2d_fwd_xdl ${CONV2D_FWD_XDL_SOURCE})
 add_executable(conv2d_fwd_xdl_bias_relu ${CONV2D_FWD_XDL_BIAS_RELU_SOURCE})
@@ -26,6 +28,7 @@ add_executable(conv2d_fwd_xdl_bias_relu_add ${CONV2D_FWD_XDL_BIAS_RELU_ADD_SOURC
 add_executable(conv2d_fwd_xdl_bias_relu_atomic_add ${CONV2D_FWD_XDL_BIAS_RELU_ATOMIC_ADD_SOURCE})
 
 target_link_libraries(gemm_xdl PRIVATE host_tensor)
+target_link_libraries(gemm_xdl_bias PRIVATE host_tensor)
 target_link_libraries(gemm_xdl_bias_relu_add PRIVATE host_tensor)
 target_link_libraries(conv2d_fwd_xdl PRIVATE host_tensor)
 target_link_libraries(conv2d_fwd_xdl_bias_relu PRIVATE host_tensor)