Skip to content

GitLab

Commit 6b9a4bd5 authored by Jun Liu's avatar Jun Liu
Browse files

Merge branch 'amd-develop-staging-0423' into amd-master

parents 56de337f c5f1cdf7
Hide whitespace changes
Inline Side-by-side
Showing with 282 additions and 84 deletions
.github/CODEOWNERS
View file @ 6b9a4bd5
* @zjing14 @junliume @illsilin @carlushuang @aosewski
* @zjing14 @junliume @illsilin @carlushuang @aosewski @yigex
# Documentation files
docs/* @ROCm/rocm-documentation
*.md @ROCm/rocm-documentation
......
CMakeLists.txt
View file @ 6b9a4bd5
......@@ -81,9 +81,6 @@ endif()
include(getopt)
# CK config file to record supported datatypes, etc.
configure_file(include/ck/config.h.in ${CMAKE_CURRENT_BINARY_DIR}/include/ck/config.h)
# CK version file to record release version as well as git commit hash
find_package(Git REQUIRED)
execute_process(COMMAND "${GIT_EXECUTABLE}" rev-parse HEAD OUTPUT_VARIABLE COMMIT_ID OUTPUT_STRIP_TRAILING_WHITESPACE)
......@@ -143,6 +140,25 @@ if(GPU_TARGETS)
else()
message("Building CK for the following targets: ${AMDGPU_TARGETS}")
endif()
if (GPU_TARGETS)
if (GPU_TARGETS MATCHES "gfx9")
add_definitions(-DCK_USE_XDL)
set(CK_USE_XDL "ON")
endif()
if (GPU_TARGETS MATCHES "gfx11")
add_definitions(-DCK_USE_WMMA)
set(CK_USE_WMMA "ON")
endif()
else()
add_definitions(-DCK_USE_WMMA -DCK_USE_XDL)
set(CK_USE_XDL "ON")
set(CK_USE_WMMA "ON")
endif()
# CK config file to record supported datatypes, etc.
configure_file(include/ck/config.h.in ${CMAKE_CURRENT_BINARY_DIR}/include/ck/config.h)
find_package(hip)
# No assumption that HIP kernels are launched with uniform block size for backward compatibility
# SWDEV-413293 and https://reviews.llvm.org/D155213
......
Jenkinsfile
View file @ 6b9a4bd5
......@@ -619,6 +619,8 @@ def process_results(Map conf=[:]){
dir("script"){
if (params.RUN_FULL_QA){
// unstash perf files to master
unstash "ckprofiler_0.2.0_amd64.deb"
sh "sshpass -p ${env.ck_deb_pw} scp -o StrictHostKeyChecking=no ckprofiler_0.2.0_amd64.deb ${env.ck_deb_user}@${env.ck_deb_ip}:/var/www/html/composable_kernel/"
unstash "perf_gemm.log"
unstash "perf_resnet50_N256.log"
unstash "perf_resnet50_N4.log"
......@@ -632,8 +634,6 @@ def process_results(Map conf=[:]){
unstash "perf_onnx_gemm.log"
unstash "perf_mixed_gemm.log"
sh "./process_qa_data.sh"
unstash "ckprofiler_0.2.0_amd64.deb"
sh "sshpass -p ${env.ck_deb_pw} scp -o StrictHostKeyChecking=no ckprofiler_0.2.0_amd64.deb ${env.ck_deb_user}@${env.ck_deb_ip}:/var/www/html/composable_kernel/"
}
else{
// unstash perf files to master
......@@ -645,10 +645,13 @@ def process_results(Map conf=[:]){
}
}
catch(e){
echo "throwing error exception while processing performance test results"
echo "Throwing error exception while processing performance test results"
echo 'Exception occurred: ' + e.toString()
throw e
}
finally{
echo "Finished processing performance test results"
}
}
}
}
......
client_example/02_gemm_add_add_fastgelu/CMakeLists.txt
View file @ 6b9a4bd5
add_custom_target(client_gemm_fastgelu_examples)
if(GPU_TARGETS MATCHES "gfx9")
add_custom_target(client_gemm_fastgelu_examples)
add_executable(client_gemm_add_add_fastgelu gemm_add_add_fastgelu.cpp)
target_link_libraries(client_gemm_add_add_fastgelu PRIVATE composable_kernel::device_gemm_operations)
add_executable(client_gemm_add_add_fastgelu gemm_add_add_fastgelu.cpp)
target_link_libraries(client_gemm_add_add_fastgelu PRIVATE composable_kernel::device_gemm_operations)
add_executable(client_gemm_add_fastgelu gemm_add_fastgelu.cpp)
target_link_libraries(client_gemm_add_fastgelu PRIVATE composable_kernel::device_gemm_operations)
add_executable(client_gemm_add_fastgelu gemm_add_fastgelu.cpp)
target_link_libraries(client_gemm_add_fastgelu PRIVATE composable_kernel::device_gemm_operations)
add_executable(client_gemm_fastgelu gemm_fastgelu.cpp)
target_link_libraries(client_gemm_fastgelu PRIVATE composable_kernel::device_gemm_operations)
add_executable(client_gemm_fastgelu gemm_fastgelu.cpp)
target_link_libraries(client_gemm_fastgelu PRIVATE composable_kernel::device_gemm_operations)
add_dependencies(client_gemm_fastgelu_examples client_gemm_add_add_fastgelu client_gemm_add_fastgelu
add_dependencies(client_gemm_fastgelu_examples client_gemm_add_add_fastgelu client_gemm_add_fastgelu
client_gemm_fastgelu)
add_custom_target(client_gemm_fastgelu_generic_examples)
add_custom_target(client_gemm_fastgelu_generic_examples)
add_executable(client_gemm_add_add_fastgelu_generic gemm_add_add_fastgelu_generic.cpp)
target_link_libraries(client_gemm_add_add_fastgelu_generic composable_kernel::device_gemm_operations)
add_executable(client_gemm_add_add_fastgelu_generic gemm_add_add_fastgelu_generic.cpp)
target_link_libraries(client_gemm_add_add_fastgelu_generic composable_kernel::device_gemm_operations)
add_executable(client_gemm_add_fastgelu_generic gemm_add_fastgelu_generic.cpp)
target_link_libraries(client_gemm_add_fastgelu_generic PRIVATE composable_kernel::device_gemm_operations)
add_executable(client_gemm_add_fastgelu_generic gemm_add_fastgelu_generic.cpp)
target_link_libraries(client_gemm_add_fastgelu_generic PRIVATE composable_kernel::device_gemm_operations)
add_executable(client_gemm_fastgelu_generic gemm_fastgelu_generic.cpp)
target_link_libraries(client_gemm_fastgelu_generic PRIVATE composable_kernel::device_gemm_operations)
add_executable(client_gemm_fastgelu_generic gemm_fastgelu_generic.cpp)
target_link_libraries(client_gemm_fastgelu_generic PRIVATE composable_kernel::device_gemm_operations)
add_dependencies(client_gemm_fastgelu_generic_examples client_gemm_add_add_fastgelu_generic
add_dependencies(client_gemm_fastgelu_generic_examples client_gemm_add_add_fastgelu_generic
client_gemm_add_fastgelu_generic client_gemm_fastgelu_generic)
endif()
client_example/03_gemm_layernorm/CMakeLists.txt
View file @ 6b9a4bd5
add_executable(client_gemm_add_add_layernorm_naive gemm_add_add_layernorm_naive.cpp)
target_link_libraries(client_gemm_add_add_layernorm_naive PRIVATE composable_kernel::device_gemm_operations composable_kernel::device_other_operations)
if(GPU_TARGETS MATCHES "gfx9")
add_executable(client_gemm_add_add_layernorm_naive gemm_add_add_layernorm_naive.cpp)
target_link_libraries(client_gemm_add_add_layernorm_naive PRIVATE composable_kernel::device_gemm_operations composable_kernel::device_other_operations)
add_executable(client_gemm_add_relu_add_layernorm_welford gemm_add_relu_add_layernorm_welford.cpp)
target_link_libraries(client_gemm_add_relu_add_layernorm_welford PRIVATE composable_kernel::device_gemm_operations composable_kernel::device_other_operations)
add_executable(client_gemm_add_relu_add_layernorm_welford gemm_add_relu_add_layernorm_welford.cpp)
target_link_libraries(client_gemm_add_relu_add_layernorm_welford PRIVATE composable_kernel::device_gemm_operations composable_kernel::device_other_operations)
endif()
client_example/04_contraction/CMakeLists.txt
View file @ 6b9a4bd5
add_executable(client_contraction_scale_fp32 contraction_scale_fp32.cpp)
target_link_libraries(client_contraction_scale_fp32 PRIVATE composable_kernel::device_other_operations composable_kernel::device_contraction_operations composable_kernel::device_gemm_operations)
if(GPU_TARGETS MATCHES "gfx9")
add_executable(client_contraction_scale_fp32 contraction_scale_fp32.cpp)
target_link_libraries(client_contraction_scale_fp32 PRIVATE composable_kernel::device_other_operations composable_kernel::device_contraction_operations composable_kernel::device_gemm_operations)
add_executable(client_contraction_bilinear_fp32 contraction_bilinear_fp32.cpp)
target_link_libraries(client_contraction_bilinear_fp32 PRIVATE composable_kernel::device_other_operations composable_kernel::device_contraction_operations composable_kernel::device_gemm_operations)
add_executable(client_contraction_bilinear_fp32 contraction_bilinear_fp32.cpp)
target_link_libraries(client_contraction_bilinear_fp32 PRIVATE composable_kernel::device_other_operations composable_kernel::device_contraction_operations composable_kernel::device_gemm_operations)
add_executable(client_contraction_scale_fp64 contraction_scale_fp64.cpp)
target_link_libraries(client_contraction_scale_fp64 PRIVATE composable_kernel::device_other_operations composable_kernel::device_contraction_operations composable_kernel::device_gemm_operations)
add_executable(client_contraction_scale_fp64 contraction_scale_fp64.cpp)
target_link_libraries(client_contraction_scale_fp64 PRIVATE composable_kernel::device_other_operations composable_kernel::device_contraction_operations composable_kernel::device_gemm_operations)
add_executable(client_contraction_bilinear_fp64 contraction_bilinear_fp64.cpp)
target_link_libraries(client_contraction_bilinear_fp64 PRIVATE composable_kernel::device_other_operations composable_kernel::device_contraction_operations composable_kernel::device_gemm_operations)
add_executable(contraction_g1m2n3k1_add_xdl_fp16 contraction_g1m2n3k1_add_xdl_fp16.cpp)
target_link_libraries(contraction_g1m2n3k1_add_xdl_fp16 PRIVATE composable_kernel::device_other_operations composable_kernel::device_contraction_operations composable_kernel::device_gemm_operations)
add_executable(client_contraction_bilinear_fp64 contraction_bilinear_fp64.cpp)
target_link_libraries(client_contraction_bilinear_fp64 PRIVATE composable_kernel::device_other_operations composable_kernel::device_contraction_operations composable_kernel::device_gemm_operations)
add_executable(contraction_g1m2n3k1_add_xdl_fp16 contraction_g1m2n3k1_add_xdl_fp16.cpp)
target_link_libraries(contraction_g1m2n3k1_add_xdl_fp16 PRIVATE composable_kernel::device_other_operations composable_kernel::device_contraction_operations composable_kernel::device_gemm_operations)
endif()
client_example/07_grouped_convnd_fwd/CMakeLists.txt
View file @ 6b9a4bd5
add_executable(client_grouped_conv2d_fwd grouped_conv2d_fwd.cpp)
target_link_libraries(client_grouped_conv2d_fwd PRIVATE composable_kernel::device_conv_operations)
if(GPU_TARGETS MATCHES "gfx9")
add_executable(client_grouped_conv2d_fwd grouped_conv2d_fwd.cpp)
target_link_libraries(client_grouped_conv2d_fwd PRIVATE composable_kernel::device_conv_operations)
add_executable(client_grouped_conv1d_fwd grouped_conv1d_fwd.cpp)
target_link_libraries(client_grouped_conv1d_fwd PRIVATE composable_kernel::device_conv_operations)
add_executable(client_grouped_conv1d_fwd grouped_conv1d_fwd.cpp)
target_link_libraries(client_grouped_conv1d_fwd PRIVATE composable_kernel::device_conv_operations)
endif()
\ No newline at end of file
client_example/08_fused_attention/CMakeLists.txt
View file @ 6b9a4bd5
add_executable(client_fused_attention fused_attention.cpp)
target_link_libraries(client_fused_attention PRIVATE composable_kernel::device_other_operations composable_kernel::device_gemm_operations)
if(GPU_TARGETS MATCHES "gfx9")
add_executable(client_fused_attention fused_attention.cpp)
target_link_libraries(client_fused_attention PRIVATE composable_kernel::device_other_operations composable_kernel::device_gemm_operations)
add_executable(client_fused_attention_bias fused_attention_bias.cpp)
target_link_libraries(client_fused_attention_bias PRIVATE composable_kernel::device_other_operations composable_kernel::device_gemm_operations)
add_executable(client_fused_attention_bias fused_attention_bias.cpp)
target_link_libraries(client_fused_attention_bias PRIVATE composable_kernel::device_other_operations composable_kernel::device_gemm_operations)
endif()
client_example/09_quantization/CMakeLists.txt
View file @ 6b9a4bd5
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_conv_operations composable_kernel::device_other_operations composable_kernel::device_gemm_operations)
if(GPU_TARGETS MATCHES "gfx9" AND (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_conv_operations composable_kernel::device_other_operations composable_kernel::device_gemm_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_conv_operations composable_kernel::device_other_operations composable_kernel::device_gemm_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_conv_operations composable_kernel::device_other_operations composable_kernel::device_gemm_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_conv_operations composable_kernel::device_other_operations composable_kernel::device_gemm_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_conv_operations composable_kernel::device_other_operations composable_kernel::device_gemm_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_conv_operations composable_kernel::device_other_operations composable_kernel::device_gemm_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_conv_operations composable_kernel::device_other_operations composable_kernel::device_gemm_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_conv_operations composable_kernel::device_other_operations composable_kernel::device_gemm_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_conv_operations composable_kernel::device_other_operations composable_kernel::device_gemm_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_conv_operations composable_kernel::device_other_operations composable_kernel::device_gemm_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_conv_operations composable_kernel::device_other_operations composable_kernel::device_gemm_operations)
add_executable(client_gemm_quantization gemm_quantization.cpp)
target_link_libraries(client_gemm_quantization PRIVATE composable_kernel::device_conv_operations composable_kernel::device_other_operations composable_kernel::device_gemm_operations)
add_executable(client_gemm_quantization gemm_quantization.cpp)
target_link_libraries(client_gemm_quantization PRIVATE composable_kernel::device_conv_operations composable_kernel::device_other_operations composable_kernel::device_gemm_operations)
endif()
client_example/15_convnd_bwd_data/CMakeLists.txt
View file @ 6b9a4bd5
add_executable(client_conv3d_bwd_data_fp16 conv3d_bwd_data_fp16.cpp)
add_executable(client_conv3d_bwd_data_fp32 conv3d_bwd_data_fp32.cpp)
if(GPU_TARGETS MATCHES "gfx9")
add_executable(client_conv3d_bwd_data_fp16 conv3d_bwd_data_fp16.cpp)
add_executable(client_conv3d_bwd_data_fp32 conv3d_bwd_data_fp32.cpp)
target_link_libraries(client_conv3d_bwd_data_fp16 PRIVATE composable_kernel::device_conv_operations)
target_link_libraries(client_conv3d_bwd_data_fp32 PRIVATE composable_kernel::device_conv_operations)
target_link_libraries(client_conv3d_bwd_data_fp16 PRIVATE composable_kernel::device_conv_operations)
target_link_libraries(client_conv3d_bwd_data_fp32 PRIVATE composable_kernel::device_conv_operations)
endif()
client_example/15_gemm_add_multiply/CMakeLists.txt
View file @ 6b9a4bd5
add_executable(client_gemm_add_multiply gemm_add_multiply.cpp)
target_link_libraries(client_gemm_add_multiply PRIVATE composable_kernel::device_gemm_operations)
\ No newline at end of file
if(GPU_TARGETS MATCHES "gfx9")
add_executable(client_gemm_add_multiply gemm_add_multiply.cpp)
target_link_libraries(client_gemm_add_multiply PRIVATE composable_kernel::device_gemm_operations)
endif()
client_example/16_convnd_fwd/CMakeLists.txt
View file @ 6b9a4bd5
......@@ -12,6 +12,19 @@ if((DTYPES MATCHES "fp8") OR NOT DEFINED DTYPES)
target_link_libraries(client_conv3d_fwd_fp8 PRIVATE composable_kernel::device_conv_operations)
endif()
if((DTYPES MATCHES "bf8") OR NOT DEFINED DTYPES)
add_executable(client_conv3d_fwd_bf8 conv3d_fwd_bf8.cpp)
target_link_libraries(client_conv3d_fwd_bf8 PRIVATE composable_kernel::device_conv_operations)
endif()
if((DTYPES MATCHES "fp8" AND DTYPES MATCHES "bf8") OR NOT DEFINED DTYPES)
add_executable(client_conv3d_fwd_fp8_bf8 conv3d_fwd_fp8_bf8.cpp)
target_link_libraries(client_conv3d_fwd_fp8_bf8 PRIVATE composable_kernel::device_conv_operations)
add_executable(client_conv3d_fwd_bf8_fp8 conv3d_fwd_bf8_fp8.cpp)
target_link_libraries(client_conv3d_fwd_bf8_fp8 PRIVATE composable_kernel::device_conv_operations)
endif()
if((DTYPES MATCHES "fp32") OR NOT DEFINED DTYPES)
add_executable(client_conv3d_fwd_fp32 conv3d_fwd_fp32.cpp)
target_link_libraries(client_conv3d_fwd_fp32 PRIVATE composable_kernel::device_conv_operations)
......
client_example/16_convnd_fwd/common.hpp
View file @ 6b9a4bd5
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#include <iomanip>
......@@ -95,7 +95,8 @@ template <ck::index_t NumDimSpatial,
typename WeiLayout,
typename OutLayout,
ck::index_t NumNonSpatialDim = 3,
typename ComputeType = InDataType>
typename AComputeType = InDataType,
typename BComputeType = AComputeType>
bool run_grouped_conv_fwd(std::array<ck::index_t, NumDimSpatial + NumNonSpatialDim> in_lengths,
std::array<ck::index_t, NumDimSpatial + NumNonSpatialDim> wei_lengths,
std::array<ck::index_t, NumDimSpatial + NumNonSpatialDim> out_lengths)
......@@ -186,7 +187,8 @@ bool run_grouped_conv_fwd(std::array<ck::index_t, NumDimSpatial + NumNonSpatialD
PassThrough,
PassThrough,
PassThrough,
ComputeType>;
AComputeType,
BComputeType>;
// get device op instances
const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
DeviceOp>::GetInstances();
......
client_example/16_convnd_fwd/conv3d_fwd_bf8.cpp 0 → 100644
View file @ 6b9a4bd5
// SPDX-License-Identifier: MIT
// Copyright (c) 2024, 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::bf8_t;
using WeiDataType = ck::bf8_t;
using OutDataType = ck::f8_t;
using InLayout = ck::tensor_layout::convolution::NDHWGC;
using WeiLayout = ck::tensor_layout::convolution::GKZYXC;
using OutLayout = ck::tensor_layout::convolution::NDHWGK;
static constexpr ck::index_t NumDimSpatial = 3;
static constexpr ck::index_t G = 1;
static constexpr ck::index_t N = 64;
static constexpr ck::index_t K = 128;
static constexpr ck::index_t C = 64;
static constexpr ck::index_t Z = 3;
static constexpr ck::index_t Y = 3;
static constexpr ck::index_t X = 3;
static constexpr ck::index_t Di = 28;
static constexpr ck::index_t Hi = 28;
static constexpr ck::index_t Wi = 3;
static constexpr ck::index_t Do = 28;
static constexpr ck::index_t Ho = 28;
static constexpr ck::index_t Wo = 3;
int main()
{
return run_grouped_conv_fwd<NumDimSpatial,
InDataType,
WeiDataType,
OutDataType,
InLayout,
WeiLayout,
OutLayout,
3,
ck::bf8_t>(
{N, Di, Hi, Wi, G, C}, {G, K, Z, Y, X, C}, {N, Do, Ho, Wo, G, K})
? EXIT_SUCCESS
: EXIT_FAILURE;
}
client_example/16_convnd_fwd/conv3d_fwd_bf8_fp8.cpp 0 → 100644
View file @ 6b9a4bd5
// SPDX-License-Identifier: MIT
// Copyright (c) 2024, 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::bf8_t;
using WeiDataType = ck::f8_t;
using OutDataType = ck::f8_t;
using InLayout = ck::tensor_layout::convolution::NDHWGC;
using WeiLayout = ck::tensor_layout::convolution::GKZYXC;
using OutLayout = ck::tensor_layout::convolution::NDHWGK;
using AComputeType = ck::bf8_t;
using BComputeType = ck::f8_t;
static constexpr ck::index_t NumDimSpatial = 3;
static constexpr ck::index_t G = 1;
static constexpr ck::index_t N = 64;
static constexpr ck::index_t K = 128;
static constexpr ck::index_t C = 64;
static constexpr ck::index_t Z = 3;
static constexpr ck::index_t Y = 3;
static constexpr ck::index_t X = 3;
static constexpr ck::index_t Di = 28;
static constexpr ck::index_t Hi = 28;
static constexpr ck::index_t Wi = 3;
static constexpr ck::index_t Do = 28;
static constexpr ck::index_t Ho = 28;
static constexpr ck::index_t Wo = 3;
int main()
{
return run_grouped_conv_fwd<NumDimSpatial,
InDataType,
WeiDataType,
OutDataType,
InLayout,
WeiLayout,
OutLayout,
3,
AComputeType,
BComputeType>(
{N, Di, Hi, Wi, G, C}, {G, K, Z, Y, X, C}, {N, Do, Ho, Wo, G, K})
? EXIT_SUCCESS
: EXIT_FAILURE;
}
client_example/16_convnd_fwd/conv3d_fwd_fp8_bf8.cpp 0 → 100644
View file @ 6b9a4bd5
// SPDX-License-Identifier: MIT
// Copyright (c) 2024, 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::f8_t;
using WeiDataType = ck::bf8_t;
using OutDataType = ck::f8_t;
using InLayout = ck::tensor_layout::convolution::NDHWGC;
using WeiLayout = ck::tensor_layout::convolution::GKZYXC;
using OutLayout = ck::tensor_layout::convolution::NDHWGK;
using AComputeType = ck::f8_t;
using BComputeType = ck::bf8_t;
static constexpr ck::index_t NumDimSpatial = 3;
static constexpr ck::index_t G = 1;
static constexpr ck::index_t N = 64;
static constexpr ck::index_t K = 128;
static constexpr ck::index_t C = 64;
static constexpr ck::index_t Z = 3;
static constexpr ck::index_t Y = 3;
static constexpr ck::index_t X = 3;
static constexpr ck::index_t Di = 28;
static constexpr ck::index_t Hi = 28;
static constexpr ck::index_t Wi = 3;
static constexpr ck::index_t Do = 28;
static constexpr ck::index_t Ho = 28;
static constexpr ck::index_t Wo = 3;
int main()
{
return run_grouped_conv_fwd<NumDimSpatial,
InDataType,
WeiDataType,
OutDataType,
InLayout,
WeiLayout,
OutLayout,
3,
AComputeType,
BComputeType>(
{N, Di, Hi, Wi, G, C}, {G, K, Z, Y, X, C}, {N, Do, Ho, Wo, G, K})
? EXIT_SUCCESS
: EXIT_FAILURE;
}
client_example/17_grouped_gemm_fastgelu/CMakeLists.txt
View file @ 6b9a4bd5
add_executable(client_grouped_gemm_fastgelu grouped_gemm_fastgelu.cpp)
target_link_libraries(client_grouped_gemm_fastgelu PRIVATE composable_kernel::device_gemm_operations)
\ No newline at end of file
if(GPU_TARGETS MATCHES "gfx9")
add_executable(client_grouped_gemm_fastgelu grouped_gemm_fastgelu.cpp)
target_link_libraries(client_grouped_gemm_fastgelu PRIVATE composable_kernel::device_gemm_operations)
endif()
client_example/20_splitk_gemm/CMakeLists.txt
View file @ 6b9a4bd5
if((DTYPES MATCHES "fp8" AND DTYPES MATCHES "fp16") OR NOT DEFINED DTYPES)
if(GPU_TARGETS MATCHES "gfx9" AND ((DTYPES MATCHES "fp8" AND DTYPES MATCHES "fp16") OR NOT DEFINED DTYPES))
add_executable(client_splitK_gemm splitK_gemm_fp16_f8.cpp)
target_link_libraries(client_splitK_gemm PRIVATE composable_kernel::device_gemm_operations)
endif()
client_example/21_grouped_gemm_bias/CMakeLists.txt
View file @ 6b9a4bd5
add_executable(client_grouped_gemm_fixed_nk_bias_fp16 grouped_gemm_fixed_nk_bias_fp16.cpp)
target_link_libraries(client_grouped_gemm_fixed_nk_bias_fp16 PRIVATE composable_kernel::device_gemm_operations)
if(GPU_TARGETS MATCHES "gfx9")
add_executable(client_grouped_gemm_fixed_nk_bias_fp16 grouped_gemm_fixed_nk_bias_fp16.cpp)
target_link_libraries(client_grouped_gemm_fixed_nk_bias_fp16 PRIVATE composable_kernel::device_gemm_operations)
endif()
client_example/22_grouped_gemm/CMakeLists.txt
View file @ 6b9a4bd5
add_executable(client_grouped_gemm_fixed_nk_fp16 grouped_gemm_fixed_nk_fp16.cpp)
target_link_libraries(client_grouped_gemm_fixed_nk_fp16 PRIVATE composable_kernel::device_gemm_operations)
if(GPU_TARGETS MATCHES "gfx9")
add_executable(client_grouped_gemm_fixed_nk_fp16 grouped_gemm_fixed_nk_fp16.cpp)
target_link_libraries(client_grouped_gemm_fixed_nk_fp16 PRIVATE composable_kernel::device_gemm_operations)
add_executable(client_grouped_gemm_fixed_nk_fp8 grouped_gemm_fixed_nk_fp8.cpp)
target_link_libraries(client_grouped_gemm_fixed_nk_fp8 PRIVATE composable_kernel::device_gemm_operations)
add_executable(client_grouped_gemm_fixed_nk_fp8 grouped_gemm_fixed_nk_fp8.cpp)
target_link_libraries(client_grouped_gemm_fixed_nk_fp8 PRIVATE composable_kernel::device_gemm_operations)
add_executable(client_grouped_gemm_fixed_nk_i8 grouped_gemm_fixed_nk_i8.cpp)
target_link_libraries(client_grouped_gemm_fixed_nk_i8 PRIVATE composable_kernel::device_gemm_operations)
add_executable(client_grouped_gemm_fixed_nk_i8 grouped_gemm_fixed_nk_i8.cpp)
target_link_libraries(client_grouped_gemm_fixed_nk_i8 PRIVATE composable_kernel::device_gemm_operations)
add_executable(client_grouped_gemm_fixed_nk_bf16 grouped_gemm_fixed_nk_bf16.cpp)
target_link_libraries(client_grouped_gemm_fixed_nk_bf16 PRIVATE composable_kernel::device_gemm_operations)
add_executable(client_grouped_gemm_fixed_nk_bf16 grouped_gemm_fixed_nk_bf16.cpp)
target_link_libraries(client_grouped_gemm_fixed_nk_bf16 PRIVATE composable_kernel::device_gemm_operations)
endif()
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment