initial commit

client_example/08_fused_attention/CMakeLists.txt

+add_executable(client_fused_attention fused_attention.cpp)
+target_link_libraries(client_fused_attention PRIVATE composable_kernel::device_operations)
+
+add_executable(client_fused_attention_bias fused_attention_bias.cpp)
+target_link_libraries(client_fused_attention_bias PRIVATE composable_kernel::device_operations)

client_example/09_quantization/CMakeLists.txt

+if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
+add_executable(client_conv2d_fwd_bias_tanh_perchannel_quantization conv2d_fwd_bias_tanh_perchannel_quantization.cpp)
+target_link_libraries(client_conv2d_fwd_bias_tanh_perchannel_quantization PRIVATE composable_kernel::device_operations)
+
+add_executable(client_conv2d_fwd_bias_relu_perchannel_quantization conv2d_fwd_bias_relu_perchannel_quantization.cpp)
+target_link_libraries(client_conv2d_fwd_bias_relu_perchannel_quantization PRIVATE composable_kernel::device_operations)
+
+add_executable(client_conv2d_fwd_bias_tanh_perlayer_quantization conv2d_fwd_bias_tanh_perlayer_quantization.cpp)
+target_link_libraries(client_conv2d_fwd_bias_tanh_perlayer_quantization PRIVATE composable_kernel::device_operations)
+
+add_executable(client_conv2d_fwd_bias_relu_perlayer_quantization conv2d_fwd_bias_relu_perlayer_quantization.cpp)
+target_link_libraries(client_conv2d_fwd_bias_relu_perlayer_quantization PRIVATE composable_kernel::device_operations)
+
+add_executable(client_conv2d_fwd_perchannel_quantization conv2d_fwd_perchannel_quantization.cpp)
+target_link_libraries(client_conv2d_fwd_perchannel_quantization PRIVATE composable_kernel::device_operations)
+
+add_executable(client_conv2d_fwd_perlayer_quantization conv2d_fwd_perlayer_quantization.cpp)
+target_link_libraries(client_conv2d_fwd_perlayer_quantization PRIVATE composable_kernel::device_operations)
+
+add_executable(client_gemm_quantization gemm_quantization.cpp)
+target_link_libraries(client_gemm_quantization PRIVATE composable_kernel::device_operations)
+endif()

client_example/10_grouped_conv2d_bwd_data/CMakeLists.txt

+add_executable(client_grouped_conv2d_bwd_data grouped_conv2d_bwd_data.cpp)
+target_link_libraries(client_grouped_conv2d_bwd_data PRIVATE composable_kernel::device_operations)

client_example/11_grouped_conv_bwd_weight/CMakeLists.txt

+add_executable(client_grouped_conv1d_bwd_weight_fp16 grouped_conv1d_bwd_weight_fp16.cpp)
+add_executable(client_grouped_conv2d_bwd_weight_fp16 grouped_conv2d_bwd_weight_fp16.cpp)
+add_executable(client_grouped_conv3d_bwd_weight_fp16 grouped_conv3d_bwd_weight_fp16.cpp)
+add_executable(client_grouped_conv3d_bwd_weight_fp32 grouped_conv3d_bwd_weight_fp32.cpp)
+
+target_link_libraries(client_grouped_conv1d_bwd_weight_fp16 PRIVATE composable_kernel::device_operations)
+target_link_libraries(client_grouped_conv2d_bwd_weight_fp16 PRIVATE composable_kernel::device_operations)
+target_link_libraries(client_grouped_conv3d_bwd_weight_fp16 PRIVATE composable_kernel::device_operations)
+target_link_libraries(client_grouped_conv3d_bwd_weight_fp32 PRIVATE composable_kernel::device_operations)

client_example/11_grouped_conv_bwd_weight/grouped_conv1d_bwd_weight_fp16.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+
+using InDataType  = ck::half_t;
+using WeiDataType = ck::half_t;
+using OutDataType = ck::half_t;
+
+using InLayout  = ck::tensor_layout::convolution::GNWC;
+using WeiLayout = ck::tensor_layout::convolution::GKXC;
+using OutLayout = ck::tensor_layout::convolution::GNWK;
+
+static constexpr ck::index_t NumDimSpatial = 1;
+static constexpr ck::index_t G             = 32;
+static constexpr ck::index_t N             = 256;
+static constexpr ck::index_t K             = 192;
+static constexpr ck::index_t C             = 192;
+static constexpr ck::index_t X             = 3;
+static constexpr ck::index_t Wi            = 28;
+static constexpr ck::index_t Wo            = 28;
+static constexpr std::array<ck::index_t, NumDimSpatial + 3> input_lengths{G, N, C, Wi};
+static constexpr std::array<ck::index_t, NumDimSpatial + 3> filter_lengths{G, K, C, X};
+static constexpr std::array<ck::index_t, NumDimSpatial + 3> output_lengths{G, N, K, Wo};
+static constexpr std::array<ck::index_t, NumDimSpatial + 3> input_strides{N * Wi * C, Wi* C, 1, C};
+static constexpr std::array<ck::index_t, NumDimSpatial + 3> weights_strides{K * X * C, X* C, 1, C};
+static constexpr std::array<ck::index_t, NumDimSpatial + 3> output_strides{N * Wo * K, Wo* K, 1, K};
+static constexpr std::array<ck::index_t, NumDimSpatial> conv_filter_strides{1};
+static constexpr std::array<ck::index_t, NumDimSpatial> conv_filter_dilations{1};
+static constexpr std::array<ck::index_t, NumDimSpatial> input_left_pads{1};
+static constexpr std::array<ck::index_t, NumDimSpatial> input_right_pads{1};
+
+int main()
+{
+    return run_grouped_conv_bwd_weight<NumDimSpatial,
+                                       InDataType,
+                                       WeiDataType,
+                                       OutDataType,
+                                       InLayout,
+                                       WeiLayout,
+                                       OutLayout>(input_lengths,
+                                                  input_strides,
+                                                  filter_lengths,
+                                                  weights_strides,
+                                                  output_lengths,
+                                                  output_strides,
+                                                  conv_filter_strides,
+                                                  conv_filter_dilations,
+                                                  input_left_pads,
+                                                  input_right_pads)
+               ? EXIT_SUCCESS
+               : EXIT_FAILURE;
+}

client_example/12_elementwise_normalization/CMakeLists.txt

+add_executable(client_elementwise_layernorm2d elementwise_layernorm2d.cpp)
+target_link_libraries(client_elementwise_layernorm2d PRIVATE composable_kernel::device_operations)