Skip to content

GitLab

"magic_pdf/git@developer.sourcefind.cn:wangsen/mineru.git" did not exist on "99055af3316ffe7b29a8e560fa1c472a2edf7f2e"
Commit e00a943e authored by myamlak's avatar myamlak
Browse files

Merge remote-tracking branch 'origin/develop' into myamlak/cgemm

parents ffe12e2e 9f71ff48
Hide whitespace changes
Inline Side-by-side
Showing with 427 additions and 356 deletions
CMakeLists.txt
View file @ e00a943e
......@@ -27,6 +27,8 @@ set(CMAKE_CXX_STANDARD_REQUIRED ON)
set(CMAKE_CXX_EXTENSIONS OFF)
message("CMAKE_CXX_COMPILER_ID: ${CMAKE_CXX_COMPILER_ID}")
option(CK_TIME_KERNEL "Turning off will disable kernel timing globally" ON)
## OpenMP
if(CMAKE_CXX_COMPILER_ID MATCHES "Clang")
# workaround issue hipcc in rocm3.5 cannot find openmp
......@@ -72,8 +74,9 @@ message(STATUS "Build with HIP ${HIP_VERSION}")
rocm_create_package(
NAME CK-${CK_BACKEND}
NAME composablekernel
DESCRIPTION "High Performance Composable Kernel for AMD GPUs"
MAINTAINER "MIOpen Kernels Dev Team <dl.MIOpen@amd.com>"
LDCONFIG
)
......@@ -226,7 +229,7 @@ set(CMAKE_LIBRARY_OUTPUT_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}/lib)
set(CMAKE_ARCHIVE_OUTPUT_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}/lib)
set(CMAKE_RUNTIME_OUTPUT_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}/bin)
configure_file("${PROJECT_SOURCE_DIR}/include/ck/hip_version.hpp.in" "${PROJECT_BINARY_DIR}/include/ck/hip_version.hpp")
configure_file("${PROJECT_SOURCE_DIR}/include/ck/options.hpp.in" "${PROJECT_BINARY_DIR}/include/ck/options.hpp")
include_directories(BEFORE
${PROJECT_SOURCE_DIR}/include
......@@ -234,7 +237,6 @@ include_directories(BEFORE
${PROJECT_SOURCE_DIR}/library/include
)
include(googletest)
SET(BUILD_DEV ON CACHE BOOL "BUILD_DEV")
if(BUILD_DEV)
......@@ -243,7 +245,31 @@ if(BUILD_DEV)
endif()
message("CMAKE_CXX_FLAGS: ${CMAKE_CXX_FLAGS}")
add_custom_target(check COMMAND ${CMAKE_CTEST_COMMAND} --output-on-failure -C ${CMAKE_CFG_INTDIR})
add_subdirectory(library)
add_subdirectory(example)
add_subdirectory(test)
add_subdirectory(profiler)
#Create an interface target for the include only files and call it "composablekernels"
include(CMakePackageConfigHelpers)
set(version 1.0.0)
write_basic_package_version_file(
"${CMAKE_CURRENT_BINARY_DIR}/composable_kernelConfigVersion.cmake"
VERSION "${version}"
COMPATIBILITY AnyNewerVersion
)
configure_package_config_file(${CMAKE_CURRENT_SOURCE_DIR}/Config.cmake.in
"${CMAKE_CURRENT_BINARY_DIR}/composable_kernelConfig.cmake"
INSTALL_DESTINATION ${CMAKE_INSTALL_LIBDIR}/cmake/composable_kernel
NO_CHECK_REQUIRED_COMPONENTS_MACRO
)
install(FILES
"${CMAKE_CURRENT_BINARY_DIR}/composable_kernelConfig.cmake"
"${CMAKE_CURRENT_BINARY_DIR}/composable_kernelConfigVersion.cmake"
DESTINATION ${CMAKE_INSTALL_LIBDIR}/cmake/composable_kernel
)
Config.cmake.in 0 → 100644
View file @ e00a943e
@PACKAGE_INIT@
set(_composable_kernel_supported_components device_operations host_tensor)
foreach(_comp ${composable_kernel_FIND_COMPONENTS})
if(NOT _comp IN_LIST _composable_kernel_supported_components)
set(composable_kernel_FOUND False)
set(composable_kernel_NOT_FOUND_MESSAGE "Unsupported component: ${_comp}")
endif()
include("${CMAKE_CURRENT_LIST_DIR}/composable_kernel${_comp}Targets.cmake")
endforeach()
Dockerfile
View file @ e00a943e
......@@ -11,13 +11,7 @@ ARG DEB_ROCM_REPO=http://repo.radeon.com/rocm/apt/.apt_$ROCMVERSION/
RUN apt-get update
RUN apt-get install -y wget gnupg
RUN wget -qO - http://repo.radeon.com/rocm/rocm.gpg.key | apt-key add -
RUN if ! [ -z $OSDB_BKC_VERSION ]; then \
echo "Using BKC VERISION: $OSDB_BKC_VERSION";\
sh -c "echo deb [arch=amd64 trusted=yes] http://compute-artifactory.amd.com/artifactory/list/rocm-osdb-deb/ compute-rocm-dkms-no-npi-hipclang ${OSDB_BKC_VERSION} > /etc/apt/sources.list.d/rocm.list" ;\
cat /etc/apt/sources.list.d/rocm.list;\
else \
sh -c "echo deb [arch=amd64] $DEB_ROCM_REPO ubuntu main > /etc/apt/sources.list.d/rocm.list" ;\
fi
RUN sh -c "echo deb [arch=amd64] $DEB_ROCM_REPO ubuntu main > /etc/apt/sources.list.d/rocm.list"
RUN wget --no-check-certificate -qO - https://apt.kitware.com/keys/kitware-archive-latest.asc 2>/dev/null | apt-key add -
RUN sh -c "echo deb https://apt.kitware.com/ubuntu/ bionic main | tee -a /etc/apt/sources.list"
......@@ -25,18 +19,15 @@ RUN sh -c "echo deb https://apt.kitware.com/ubuntu/ bionic main | tee -a /etc/ap
# Install dependencies
RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y --allow-unauthenticated \
apt-utils \
sshpass \
build-essential \
cmake-data=3.15.1-0kitware1 \
cmake=3.15.1-0kitware1 \
curl \
doxygen \
g++ \
gdb \
git \
hip-rocclr \
jq \
lcov \
libelf-dev \
libncurses5-dev \
libnuma-dev \
......@@ -62,8 +53,6 @@ RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y --allow-
apt-get clean && \
rm -rf /var/lib/apt/lists/*
# RUN pip3 install --default-timeout=100000 -r requirements.txt
# Setup ubsan environment to printstacktrace
RUN ln -s /usr/bin/llvm-symbolizer-3.8 /usr/local/bin/llvm-symbolizer
ENV UBSAN_OPTIONS=print_stacktrace=1
......@@ -92,5 +81,3 @@ ADD rbuild.ini /rbuild.ini
ADD dev-requirements.txt dev-requirements.txt
RUN rbuild prepare -s develop -d $PREFIX
RUN groupadd -f render
# RUN cget install -f min-requirements.txt
# RUN CXXFLAGS='-isystem $PREFIX/include' cget install -f ./mlir-requirements.txt
Jenkinsfile
View file @ e00a943e
......@@ -320,7 +320,7 @@ pipeline {
{
agent{ label rocmnode("gfx908")}
environment{
setup_args = """ -D CMAKE_CXX_FLAGS="--offload-arch=gfx908 -O3 " -DBUILD_DEV=On """
setup_args = """ -D CMAKE_CXX_FLAGS=" --offload-arch=gfx900 --offload-arch=gfx906 --offload-arch=gfx908 --offload-arch=gfx90a -O3 " -DBUILD_DEV=On """
}
steps{
buildHipClangJobAndReboot(setup_args:setup_args, config_targets: "check", no_reboot:true, build_type: 'Release')
......@@ -341,6 +341,23 @@ pipeline {
}
}
stage("Client App")
{
parallel
{
stage("Run Client App")
{
agent{ label rocmnode("gfx908")}
environment{
setup_args = """ -D -DBUILD_DEV=Off -DCMAKE_INSTALL_PREFIX=../install CMAKE_CXX_FLAGS="--offload-arch=gfx908 -O3 " """
execute_args = """ cd ../test/client_app && rm -rf build && mkdir build && cd build && cmake -DCMAKE_PREFIX_PATH="${env.WORKSPACE}/install;/opt/rocm" .. && make """
}
steps{
buildHipClangJobAndReboot(setup_args: setup_args, config_targets: "install", no_reboot:true, build_type: 'Release', execute_cmd: execute_args, prefixpath: '/usr/local')
}
}
}
}
stage("Performance Tests")
{
parallel
......
README.md
View file @ e00a943e
......@@ -43,3 +43,13 @@ Instructions for running each individual examples are under ```example/```
make -j ckProfiler
```
Instructions for running ckProfiler are under ```profiler/```
## Caveat
### Kernel Timing and Verification
CK's own kernel timer will warn up kernel once, and then run it multiple times
to get average kernel time. For some kernels that use atomic add, this will cause
output buffer to be accumulated multiple times, causing verfication failure.
To work around it, do not use CK's own timer and do verification at the same time.
CK's own timer and verification in each example and ckProfiler can be enabled or
disabled from command line.
cmake/googletest.cmake
View file @ e00a943e
......@@ -18,6 +18,8 @@ list(APPEND GTEST_CMAKE_CXX_FLAGS
-Wno-switch-enum
-Wno-zero-as-null-pointer-constant
-Wno-unused-member-function
-Wno-comma
-Wno-old-style-cast
)
message(STATUS "Suppressing googltest warnings with flags: ${GTEST_CMAKE_CXX_FLAGS}")
......@@ -33,4 +35,5 @@ FetchContent_MakeAvailable(googletest)
target_compile_options(gtest PRIVATE ${GTEST_CMAKE_CXX_FLAGS})
target_compile_options(gtest_main PRIVATE ${GTEST_CMAKE_CXX_FLAGS})
target_compile_options(gmock PRIVATE ${GTEST_CMAKE_CXX_FLAGS})
target_compile_options(gmock_main PRIVATE ${GTEST_CMAKE_CXX_FLAGS})
example/01_gemm/gemm_xdl_bf16.cpp
View file @ e00a943e
......@@ -88,9 +88,9 @@ using ReferenceGemmInstance = ck::tensor_operation::host::
int main(int argc, char* argv[])
{
bool do_verification = 0;
int init_method = 0;
int nrepeat = 5;
bool do_verification = true;
int init_method = 1;
bool time_kernel = false;
// GEMM shape
ck::index_t M = 3840;
......@@ -105,13 +105,13 @@ int main(int argc, char* argv[])
{
do_verification = std::stoi(argv[1]);
init_method = std::stoi(argv[2]);
nrepeat = std::stoi(argv[3]);
time_kernel = 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]);
time_kernel = std::stoi(argv[3]);
M = std::stoi(argv[4]);
N = std::stoi(argv[5]);
......@@ -125,7 +125,7 @@ int main(int argc, char* argv[])
{
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("arg3: time kernel (0=n0, 1=yes)\n");
printf("arg4 to 9: M (256x), N(128x), K(32x), StrideA, StrideB, StrideC\n");
exit(0);
}
......@@ -198,7 +198,7 @@ int main(int argc, char* argv[])
"not support this GEMM problem");
}
float ave_time = invoker.Run(argument, nrepeat);
float ave_time = invoker.Run(argument, StreamConfig{nullptr, time_kernel});
std::size_t flop = std::size_t(2) * M * N * K;
std::size_t num_btype =
......@@ -232,7 +232,7 @@ int main(int argc, char* argv[])
ref_invoker.Run(ref_argument);
ck::utils::check_err(c_m_n_device_f32_result.mData, c_m_n_host_result.mData);
return ck::utils::check_err(c_m_n_device_f32_result.mData, c_m_n_host_result.mData) ? 0 : 1;
}
return 0;
......
example/01_gemm/gemm_xdl_fp16.cpp
View file @ e00a943e
......@@ -56,9 +56,9 @@ using ReferenceGemmInstance = ck::tensor_operation::host::
int main(int argc, char* argv[])
{
bool do_verification = 0;
int init_method = 0;
int nrepeat = 5;
bool do_verification = true;
int init_method = 1;
bool time_kernel = false;
// GEMM shape
ck::index_t M = 3840;
......@@ -73,13 +73,13 @@ int main(int argc, char* argv[])
{
do_verification = std::stoi(argv[1]);
init_method = std::stoi(argv[2]);
nrepeat = std::stoi(argv[3]);
time_kernel = 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]);
time_kernel = std::stoi(argv[3]);
M = std::stoi(argv[4]);
N = std::stoi(argv[5]);
......@@ -93,7 +93,7 @@ int main(int argc, char* argv[])
{
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("arg3: time kernel (0=n0, 1=yes)\n");
printf("arg4 to 9: M (256x), N(128x), K(32x), StrideA, StrideB, StrideC\n");
exit(0);
}
......@@ -171,7 +171,7 @@ int main(int argc, char* argv[])
"not support this GEMM problem");
}
float ave_time = invoker.Run(argument, nrepeat);
float ave_time = invoker.Run(argument, StreamConfig{nullptr, time_kernel});
std::size_t flop = std::size_t(2) * M * N * K;
std::size_t num_btype =
......@@ -196,7 +196,7 @@ int main(int argc, char* argv[])
ref_invoker.Run(ref_argument);
ck::utils::check_err(c_m_n_device_result.mData, c_m_n_host_result.mData);
return ck::utils::check_err(c_m_n_device_result.mData, c_m_n_host_result.mData) ? 0 : 1;
}
return 0;
......
example/01_gemm/gemm_xdl_int8.cpp
View file @ e00a943e
......@@ -83,9 +83,9 @@ using ReferenceGemmInstance = ck::tensor_operation::host::
int main(int argc, char* argv[])
{
bool do_verification = 0;
int init_method = 0;
int nrepeat = 5;
bool do_verification = true;
int init_method = 1;
bool time_kernel = false;
// GEMM shape
ck::index_t M = 3840;
......@@ -100,13 +100,13 @@ int main(int argc, char* argv[])
{
do_verification = std::stoi(argv[1]);
init_method = std::stoi(argv[2]);
nrepeat = std::stoi(argv[3]);
time_kernel = 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]);
time_kernel = std::stoi(argv[3]);
M = std::stoi(argv[4]);
N = std::stoi(argv[5]);
......@@ -120,7 +120,7 @@ int main(int argc, char* argv[])
{
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("arg3: time kernel (0=n0, 1=yes)\n");
printf("arg4 to 9: M (256x), N(128x), K(32x), StrideA, StrideB, StrideC\n");
exit(0);
}
......@@ -194,7 +194,7 @@ int main(int argc, char* argv[])
"not support this GEMM problem");
}
float ave_time = invoker.Run(argument, nrepeat);
float ave_time = invoker.Run(argument, StreamConfig{nullptr, time_kernel});
std::size_t flop = std::size_t(2) * M * N * K;
std::size_t num_btype =
......@@ -219,7 +219,7 @@ int main(int argc, char* argv[])
ref_invoker.Run(ref_argument);
ck::utils::check_err(c_m_n_device_result.mData, c_m_n_host_result.mData);
return ck::utils::check_err(c_m_n_device_result.mData, c_m_n_host_result.mData) ? 0 : 1;
}
return 0;
......
example/02_gemm_alpha_beta/gemm_xdl_alpha_beta.cpp
View file @ e00a943e
......@@ -86,9 +86,9 @@ using ReferenceGemmInstance = ck::tensor_operation::host::ReferenceGemmBias2D<AD
int main(int argc, char* argv[])
{
bool do_verification = 0;
int init_method = 0;
int nrepeat = 5;
bool do_verification = true;
int init_method = 1;
bool time_kernel = false;
// GEMM shape
ck::index_t M = 3840;
......@@ -106,13 +106,13 @@ int main(int argc, char* argv[])
{
do_verification = std::stoi(argv[1]);
init_method = std::stoi(argv[2]);
nrepeat = std::stoi(argv[3]);
time_kernel = std::stoi(argv[3]);
}
else if(argc == 6)
{
do_verification = std::stoi(argv[1]);
init_method = std::stoi(argv[2]);
nrepeat = std::stoi(argv[3]);
time_kernel = std::stoi(argv[3]);
alpha = std::stof(argv[4]);
beta = std::stof(argv[5]);
......@@ -121,7 +121,7 @@ int main(int argc, char* argv[])
{
do_verification = std::stoi(argv[1]);
init_method = std::stoi(argv[2]);
nrepeat = std::stoi(argv[3]);
time_kernel = std::stoi(argv[3]);
M = std::stoi(argv[4]);
N = std::stoi(argv[5]);
......@@ -138,7 +138,7 @@ int main(int argc, char* argv[])
{
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("arg3: time kernel (0=n0, 1=yes)\n");
printf("arg4 to 9: M (256x), N(128x), K(32x), StrideA, StrideB, StrideC, alpha, beta\n");
exit(0);
}
......@@ -216,7 +216,7 @@ int main(int argc, char* argv[])
"not support this GEMM problem");
}
float ave_time = invoker.Run(argument, nrepeat);
float ave_time = invoker.Run(argument, StreamConfig{nullptr, time_kernel});
std::size_t flop = std::size_t(2) * M * N * K;
std::size_t num_btype =
......@@ -246,6 +246,8 @@ int main(int argc, char* argv[])
ref_invoker.Run(ref_argument);
ck::utils::check_err(c_m_n_device_result.mData, c_m_n_host_result.mData);
return ck::utils::check_err(c_m_n_device_result.mData, c_m_n_host_result.mData) ? 0 : 1;
}
return 0;
}
example/03_gemm_bias_relu/gemm_xdl_bias_relu.cpp
View file @ e00a943e
......@@ -83,9 +83,9 @@ using ReferenceGemmInstance = ck::tensor_operation::host::ReferenceGemmBiasActiv
int main(int argc, char* argv[])
{
bool do_verification = 0;
int init_method = 0;
int nrepeat = 5;
bool do_verification = true;
int init_method = 1;
bool time_kernel = false;
// GEMM shape
ck::index_t M = 3840;
......@@ -100,13 +100,13 @@ int main(int argc, char* argv[])
{
do_verification = std::stoi(argv[1]);
init_method = std::stoi(argv[2]);
nrepeat = std::stoi(argv[3]);
time_kernel = 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]);
time_kernel = std::stoi(argv[3]);
M = std::stoi(argv[4]);
N = std::stoi(argv[5]);
......@@ -120,7 +120,7 @@ int main(int argc, char* argv[])
{
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("arg3: time kernel (0=n0, 1=yes)\n");
printf("arg4 to 9: M (256x), N(128x), K(32x), StrideA, StrideB, StrideC\n");
exit(0);
}
......@@ -206,7 +206,7 @@ int main(int argc, char* argv[])
"not support this GEMM problem");
}
float ave_time = invoker.Run(argument, nrepeat);
float ave_time = invoker.Run(argument, StreamConfig{nullptr, time_kernel});
std::size_t flop = std::size_t(2) * M * N * K;
......@@ -232,6 +232,8 @@ int main(int argc, char* argv[])
ref_invoker.Run(ref_argument);
ck::utils::check_err(c_m_n_device_result.mData, c_m_n_host_result.mData);
return ck::utils::check_err(c_m_n_device_result.mData, c_m_n_host_result.mData) ? 0 : 1;
}
return 0;
}
example/04_gemm_bias_relu_add/gemm_xdl_bias_relu_add.cpp
View file @ e00a943e
......@@ -83,9 +83,9 @@ using ReferenceGemmInstance =
CElementOp>;
int main(int argc, char* argv[])
{
bool do_verification = 0;
int init_method = 0;
int nrepeat = 5;
bool do_verification = true;
int init_method = 1;
bool time_kernel = false;
// GEMM shape
ck::index_t M = 3840;
......@@ -101,13 +101,13 @@ int main(int argc, char* argv[])
{
do_verification = std::stoi(argv[1]);
init_method = std::stoi(argv[2]);
nrepeat = std::stoi(argv[3]);
time_kernel = std::stoi(argv[3]);
}
else if(argc == 11)
{
do_verification = std::stoi(argv[1]);
init_method = std::stoi(argv[2]);
nrepeat = std::stoi(argv[3]);
time_kernel = std::stoi(argv[3]);
M = std::stoi(argv[4]);
N = std::stoi(argv[5]);
......@@ -122,7 +122,7 @@ int main(int argc, char* argv[])
{
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("arg3: time kernel (0=n0, 1=yes)\n");
printf("arg4 to 10: M (256x), N(128x), K(32x), StrideA, StrideB, StrideC, StrideC1\n");
exit(0);
}
......@@ -218,7 +218,7 @@ int main(int argc, char* argv[])
"not support this GEMM problem");
}
float ave_time = invoker.Run(argument, nrepeat);
float ave_time = invoker.Run(argument, StreamConfig{nullptr, time_kernel});
std::size_t flop = std::size_t(2) * M * N * K;
std::size_t num_btype = sizeof(ADataType) * M * K + sizeof(BDataType) * K * M +
......@@ -250,6 +250,8 @@ int main(int argc, char* argv[])
ref_invoker.Run(ref_argument);
ck::utils::check_err(c_m_n_device_result.mData, c_m_n_host_result.mData);
return ck::utils::check_err(c_m_n_device_result.mData, c_m_n_host_result.mData) ? 0 : 1;
}
return 0;
}
example/06_conv2d_fwd_bias_relu/CMakeLists.txt
View file @ e00a943e
add_example_executable(example_conv2d_fwd_xdl_bias_relu conv2d_fwd_xdl_bias_relu.cpp)
target_link_libraries(example_conv2d_fwd_xdl_bias_relu PRIVATE conv_fwd_util)
target_link_libraries(example_conv2d_fwd_xdl_bias_relu PRIVATE conv_util)
example/06_conv2d_fwd_bias_relu/conv2d_fwd_xdl_bias_relu.cpp
View file @ e00a943e
......@@ -7,7 +7,7 @@
#include "check_err.hpp"
#include "config.hpp"
#include "conv_fwd_util.hpp"
#include "conv_util.hpp"
#include "device.hpp"
#include "device_conv2d_fwd_xdl_c_shuffle_bias_activation_nhwc_kyxc_nhwk.hpp"
#include "device_tensor.hpp"
......@@ -93,7 +93,7 @@ void PrintUseMsg()
{
std::cout << "arg1: verification (0=no, 1=yes)\n"
<< "arg2: initialization (0=no init, 1=integer value, 2=decimal value)\n"
<< "arg3: run kernel # of times (>1)\n"
<< "arg3: time kernel (0=n0, 1=yes)\n"
<< "Following arguments:\n"
<< " N, K, C, \n"
<< " <filter spatial dimensions>, (ie Y, X for 2D)\n"
......@@ -120,40 +120,40 @@ ck::utils::conv::ConvParams ParseConvParams(int argc, char* argv[])
ck::utils::conv::ConvParams params;
int arg_idx = 4;
params.num_dim_spatial = num_dim_spatial;
params.N = std::stoi(argv[arg_idx++]);
params.K = std::stoi(argv[arg_idx++]);
params.C = std::stoi(argv[arg_idx++]);
params.num_dim_spatial_ = num_dim_spatial;
params.N_ = std::stoi(argv[arg_idx++]);
params.K_ = std::stoi(argv[arg_idx++]);
params.C_ = std::stoi(argv[arg_idx++]);
params.filter_spatial_lengths.resize(num_dim_spatial);
params.filter_spatial_lengths_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i)
{
params.filter_spatial_lengths[i] = std::stoi(argv[arg_idx++]);
params.filter_spatial_lengths_[i] = std::stoi(argv[arg_idx++]);
}
params.input_spatial_lengths.resize(num_dim_spatial);
params.input_spatial_lengths_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i)
{
params.input_spatial_lengths[i] = std::stoi(argv[arg_idx++]);
params.input_spatial_lengths_[i] = std::stoi(argv[arg_idx++]);
}
params.conv_filter_strides.resize(num_dim_spatial);
params.conv_filter_strides_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i)
{
params.conv_filter_strides[i] = std::stoi(argv[arg_idx++]);
params.conv_filter_strides_[i] = std::stoi(argv[arg_idx++]);
}
params.conv_filter_dilations.resize(num_dim_spatial);
params.conv_filter_dilations_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i)
{
params.conv_filter_dilations[i] = std::stoi(argv[arg_idx++]);
params.conv_filter_dilations_[i] = std::stoi(argv[arg_idx++]);
}
params.input_left_pads.resize(num_dim_spatial);
params.input_left_pads_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i)
{
params.input_left_pads[i] = std::stoi(argv[arg_idx++]);
params.input_left_pads_[i] = std::stoi(argv[arg_idx++]);
}
params.input_right_pads.resize(num_dim_spatial);
params.input_right_pads_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i)
{
params.input_right_pads[i] = std::stoi(argv[arg_idx++]);
params.input_right_pads_[i] = std::stoi(argv[arg_idx++]);
}
return params;
......@@ -165,9 +165,9 @@ int main(int argc, char* argv[])
{
using namespace ck::utils::conv;
bool do_verification = 0;
int init_method = 0;
int nrepeat = 5;
bool do_verification = true;
int init_method = 1;
bool time_kernel = false;
const int num_dim_spatial = 2;
ck::utils::conv::ConvParams params;
......@@ -176,7 +176,7 @@ int main(int argc, char* argv[])
{
do_verification = std::stoi(argv[1]);
init_method = std::stoi(argv[2]);
nrepeat = std::stoi(argv[3]);
time_kernel = std::stoi(argv[3]);
}
if(argc >= 5)
......@@ -184,21 +184,21 @@ int main(int argc, char* argv[])
params = ParseConvParams(argc, argv);
}
std::vector<std::size_t> input_dims{static_cast<std::size_t>(params.N),
static_cast<std::size_t>(params.C)};
std::vector<std::size_t> input_dims{static_cast<std::size_t>(params.N_),
static_cast<std::size_t>(params.C_)};
input_dims.insert(std::end(input_dims),
std::begin(params.input_spatial_lengths),
std::end(params.input_spatial_lengths));
std::begin(params.input_spatial_lengths_),
std::end(params.input_spatial_lengths_));
std::vector<std::size_t> filter_dims{static_cast<std::size_t>(params.K),
static_cast<std::size_t>(params.C)};
std::vector<std::size_t> filter_dims{static_cast<std::size_t>(params.K_),
static_cast<std::size_t>(params.C_)};
filter_dims.insert(std::end(filter_dims),
std::begin(params.filter_spatial_lengths),
std::end(params.filter_spatial_lengths));
std::begin(params.filter_spatial_lengths_),
std::end(params.filter_spatial_lengths_));
const std::vector<ck::index_t>& output_spatial_lengths = params.GetOutputSpatialLengths();
std::vector<std::size_t> output_dims{static_cast<std::size_t>(params.N),
static_cast<std::size_t>(params.K)};
std::vector<std::size_t> output_dims{static_cast<std::size_t>(params.N_),
static_cast<std::size_t>(params.K_)};
output_dims.insert(std::end(output_dims),
std::begin(output_spatial_lengths),
std::end(output_spatial_lengths));
......@@ -211,7 +211,7 @@ int main(int argc, char* argv[])
get_output_host_tensor_descriptor(output_dims, num_dim_spatial));
// bias: assume contiguous 1d vector
Tensor<OutDataType> bias(
HostTensorDescriptor(std::vector<std::size_t>({static_cast<std::size_t>(params.K)})));
HostTensorDescriptor(std::vector<std::size_t>({static_cast<std::size_t>(params.K_)})));
std::cout << "input: " << input.mDesc << std::endl;
std::cout << "weights: " << weights.mDesc << std::endl;
......@@ -248,16 +248,16 @@ int main(int argc, char* argv[])
static_cast<const WeiDataType*>(wei_device_buf.GetDeviceBuffer()),
static_cast<OutDataType*>(out_device_buf.GetDeviceBuffer()),
static_cast<const OutDataType*>(bias_device_buf.GetDeviceBuffer()),
params.N,
params.K,
params.C,
params.input_spatial_lengths,
params.filter_spatial_lengths,
params.N_,
params.K_,
params.C_,
params.input_spatial_lengths_,
params.filter_spatial_lengths_,
output_spatial_lengths,
params.conv_filter_strides,
params.conv_filter_dilations,
params.input_left_pads,
params.input_right_pads,
params.conv_filter_strides_,
params.conv_filter_dilations_,
params.input_left_pads_,
params.input_right_pads_,
InElementOp{},
WeiElementOp{},
OutElementOp{});
......@@ -269,18 +269,18 @@ int main(int argc, char* argv[])
"not support this problem");
}
float ave_time = invoker.Run(argument, nrepeat);
float ave_time = invoker.Run(argument, StreamConfig{nullptr, time_kernel});
std::size_t flop = get_flops(
params.N, params.C, params.K, params.filter_spatial_lengths, output_spatial_lengths);
params.N_, params.C_, params.K_, params.filter_spatial_lengths_, output_spatial_lengths);
std::size_t num_btype =
get_btype<InDataType, WeiDataType, OutDataType>(params.N,
params.C,
params.K,
params.input_spatial_lengths,
params.filter_spatial_lengths,
get_btype<InDataType, WeiDataType, OutDataType>(params.N_,
params.C_,
params.K_,
params.input_spatial_lengths_,
params.filter_spatial_lengths_,
output_spatial_lengths) +
sizeof(OutDataType) * (params.K);
sizeof(OutDataType) * (params.K_);
float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
float gb_per_sec = num_btype / 1.E6 / ave_time;
......@@ -296,16 +296,17 @@ int main(int argc, char* argv[])
weights,
host_output,
bias,
params.conv_filter_strides,
params.conv_filter_dilations,
params.input_left_pads,
params.input_right_pads,
params.conv_filter_strides_,
params.conv_filter_dilations_,
params.input_left_pads_,
params.input_right_pads_,
InElementOp{},
WeiElementOp{},
OutElementOp{});
ref_invoker.Run(ref_argument);
out_device_buf.FromDevice(device_output.mData.data());
ck::utils::check_err(
host_output.mData, device_output.mData, "Error: incorrect results!", 1e-5f, 1e-4f);
return ck::utils::check_err(device_output.mData, host_output.mData) ? 0 : 1;
}
return 0;
}
example/07_conv2d_fwd_bias_relu_add/CMakeLists.txt
View file @ e00a943e
add_example_executable(example_conv2d_fwd_xdl_bias_relu_add conv2d_fwd_xdl_bias_relu_add.cpp)
target_link_libraries(example_conv2d_fwd_xdl_bias_relu_add PRIVATE conv_fwd_util)
# FIXME: should fix validation failure
add_example_executable_no_testing(example_conv2d_fwd_xdl_bias_relu_add conv2d_fwd_xdl_bias_relu_add.cpp)
target_link_libraries(example_conv2d_fwd_xdl_bias_relu_add PRIVATE conv_util)
example/07_conv2d_fwd_bias_relu_add/conv2d_fwd_xdl_bias_relu_add.cpp
View file @ e00a943e
......@@ -7,7 +7,7 @@
#include "check_err.hpp"
#include "config.hpp"
#include "conv_fwd_util.hpp"
#include "conv_util.hpp"
#include "device.hpp"
#include "device_conv2d_fwd_xdl_c_shuffle_bias_activation_add_nhwc_kyxc_nhwk.hpp"
#include "device_tensor.hpp"
......@@ -90,7 +90,7 @@ void PrintUseMsg()
{
std::cout << "arg1: verification (0=no, 1=yes)\n"
<< "arg2: initialization (0=no init, 1=integer value, 2=decimal value)\n"
<< "arg3: run kernel # of times (>1)\n"
<< "arg3: time kernel (0=n0, 1=yes)\n"
<< "Following arguments:\n"
<< " N, K, C, \n"
<< " <filter spatial dimensions>, (ie Y, X for 2D)\n"
......@@ -117,40 +117,40 @@ ck::utils::conv::ConvParams ParseConvParams(int argc, char* argv[])
ck::utils::conv::ConvParams params;
int arg_idx = 4;
params.num_dim_spatial = num_dim_spatial;
params.N = std::stoi(argv[arg_idx++]);
params.K = std::stoi(argv[arg_idx++]);
params.C = std::stoi(argv[arg_idx++]);
params.num_dim_spatial_ = num_dim_spatial;
params.N_ = std::stoi(argv[arg_idx++]);
params.K_ = std::stoi(argv[arg_idx++]);
params.C_ = std::stoi(argv[arg_idx++]);
params.filter_spatial_lengths.resize(num_dim_spatial);
params.filter_spatial_lengths_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i)
{
params.filter_spatial_lengths[i] = std::stoi(argv[arg_idx++]);
params.filter_spatial_lengths_[i] = std::stoi(argv[arg_idx++]);
}
params.input_spatial_lengths.resize(num_dim_spatial);
params.input_spatial_lengths_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i)
{
params.input_spatial_lengths[i] = std::stoi(argv[arg_idx++]);
params.input_spatial_lengths_[i] = std::stoi(argv[arg_idx++]);
}
params.conv_filter_strides.resize(num_dim_spatial);
params.conv_filter_strides_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i)
{
params.conv_filter_strides[i] = std::stoi(argv[arg_idx++]);
params.conv_filter_strides_[i] = std::stoi(argv[arg_idx++]);
}
params.conv_filter_dilations.resize(num_dim_spatial);
params.conv_filter_dilations_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i)
{
params.conv_filter_dilations[i] = std::stoi(argv[arg_idx++]);
params.conv_filter_dilations_[i] = std::stoi(argv[arg_idx++]);
}
params.input_left_pads.resize(num_dim_spatial);
params.input_left_pads_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i)
{
params.input_left_pads[i] = std::stoi(argv[arg_idx++]);
params.input_left_pads_[i] = std::stoi(argv[arg_idx++]);
}
params.input_right_pads.resize(num_dim_spatial);
params.input_right_pads_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i)
{
params.input_right_pads[i] = std::stoi(argv[arg_idx++]);
params.input_right_pads_[i] = std::stoi(argv[arg_idx++]);
}
return params;
......@@ -162,9 +162,9 @@ int main(int argc, char* argv[])
{
using namespace ck::utils::conv;
bool do_verification = 0;
int init_method = 0;
int nrepeat = 5;
bool do_verification = true;
int init_method = 1;
bool time_kernel = false;
const int num_dim_spatial = 2;
ck::utils::conv::ConvParams params;
......@@ -173,7 +173,7 @@ int main(int argc, char* argv[])
{
do_verification = std::stoi(argv[1]);
init_method = std::stoi(argv[2]);
nrepeat = std::stoi(argv[3]);
time_kernel = std::stoi(argv[3]);
}
if(argc >= 5)
......@@ -181,21 +181,21 @@ int main(int argc, char* argv[])
params = ParseConvParams(argc, argv);
}
std::vector<std::size_t> input_dims{static_cast<std::size_t>(params.N),
static_cast<std::size_t>(params.C)};
std::vector<std::size_t> input_dims{static_cast<std::size_t>(params.N_),
static_cast<std::size_t>(params.C_)};
input_dims.insert(std::end(input_dims),
std::begin(params.input_spatial_lengths),
std::end(params.input_spatial_lengths));
std::begin(params.input_spatial_lengths_),
std::end(params.input_spatial_lengths_));
std::vector<std::size_t> filter_dims{static_cast<std::size_t>(params.K),
static_cast<std::size_t>(params.C)};
std::vector<std::size_t> filter_dims{static_cast<std::size_t>(params.K_),
static_cast<std::size_t>(params.C_)};
filter_dims.insert(std::end(filter_dims),
std::begin(params.filter_spatial_lengths),
std::end(params.filter_spatial_lengths));
std::begin(params.filter_spatial_lengths_),
std::end(params.filter_spatial_lengths_));
const std::vector<ck::index_t>& output_spatial_lengths = params.GetOutputSpatialLengths();
std::vector<std::size_t> output_dims{static_cast<std::size_t>(params.N),
static_cast<std::size_t>(params.K)};
std::vector<std::size_t> output_dims{static_cast<std::size_t>(params.N_),
static_cast<std::size_t>(params.K_)};
output_dims.insert(std::end(output_dims),
std::begin(output_spatial_lengths),
std::end(output_spatial_lengths));
......@@ -209,7 +209,7 @@ int main(int argc, char* argv[])
// bias: assume contiguous 1d vector
Tensor<OutDataType> bias(
HostTensorDescriptor(std::vector<std::size_t>({static_cast<std::size_t>(params.K)})));
HostTensorDescriptor(std::vector<std::size_t>({static_cast<std::size_t>(params.K_)})));
// residual: assume same layout as output tensor
Tensor<OutDataType> residual(get_output_host_tensor_descriptor(output_dims, num_dim_spatial));
......@@ -259,16 +259,16 @@ int main(int argc, char* argv[])
static_cast<OutDataType*>(out_device_buf.GetDeviceBuffer()),
static_cast<const OutDataType*>(bias_device_buf.GetDeviceBuffer()),
static_cast<const OutDataType*>(resi_device_buf.GetDeviceBuffer()),
params.N,
params.K,
params.C,
params.input_spatial_lengths,
params.filter_spatial_lengths,
params.N_,
params.K_,
params.C_,
params.input_spatial_lengths_,
params.filter_spatial_lengths_,
output_spatial_lengths,
params.conv_filter_strides,
params.conv_filter_dilations,
params.input_left_pads,
params.input_right_pads,
params.conv_filter_strides_,
params.conv_filter_dilations_,
params.input_left_pads_,
params.input_right_pads_,
in_element_op,
wei_element_op,
out_element_op);
......@@ -280,20 +280,20 @@ int main(int argc, char* argv[])
"not support this problem");
}
float ave_time = invoker.Run(argument, nrepeat);
float ave_time = invoker.Run(argument, StreamConfig{nullptr, time_kernel});
std::size_t flop = get_flops(
params.N, params.C, params.K, params.filter_spatial_lengths, output_spatial_lengths);
params.N_, params.C_, params.K_, params.filter_spatial_lengths_, output_spatial_lengths);
std::size_t num_btype =
get_btype<InDataType, WeiDataType, OutDataType>(params.N,
params.C,
params.K,
params.input_spatial_lengths,
params.filter_spatial_lengths,
get_btype<InDataType, WeiDataType, OutDataType>(params.N_,
params.C_,
params.K_,
params.input_spatial_lengths_,
params.filter_spatial_lengths_,
output_spatial_lengths) +
sizeof(OutDataType) * (params.K) +
sizeof(OutDataType) * (params.K_) +
sizeof(OutDataType) *
(params.N * params.K * output_spatial_lengths[0] * output_spatial_lengths[1]);
(params.N_ * params.K_ * output_spatial_lengths[0] * output_spatial_lengths[1]);
float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
float gb_per_sec = num_btype / 1.E6 / ave_time;
......@@ -310,17 +310,18 @@ int main(int argc, char* argv[])
host_output,
bias,
residual,
params.conv_filter_strides,
params.conv_filter_dilations,
params.input_left_pads,
params.input_right_pads,
params.conv_filter_strides_,
params.conv_filter_dilations_,
params.input_left_pads_,
params.input_right_pads_,
in_element_op,
wei_element_op,
out_element_op);
ref_invoker.Run(ref_argument);
out_device_buf.FromDevice(device_output.mData.data());
ck::utils::check_err(
host_output.mData, device_output.mData, "Error: incorrect results!", 1e-5f, 1e-4f);
return ck::utils::check_err(device_output.mData, host_output.mData) ? 0 : 1;
}
return 0;
}
example/09_convnd_fwd/CMakeLists.txt
View file @ e00a943e
add_example_executable(example_convnd_fwd_xdl convnd_fwd_xdl.cpp)
target_link_libraries(example_convnd_fwd_xdl PRIVATE conv_fwd_util)
add_example_executable(example_convnd_fwd_xdl_fp32 convnd_fwd_xdl_fp32.cpp)
add_example_executable(example_convnd_fwd_xdl_int8 convnd_fwd_xdl_int8.cpp)
target_link_libraries(example_convnd_fwd_xdl_int8 PRIVATE conv_fwd_util)
add_example_executable(example_convnd_fwd_xdl_fp16 convnd_fwd_xdl_fp16.cpp)
target_link_libraries(example_convnd_fwd_xdl_fp16 PRIVATE conv_fwd_util)
target_link_libraries(example_convnd_fwd_xdl_fp32 PRIVATE conv_util)
target_link_libraries(example_convnd_fwd_xdl_int8 PRIVATE conv_util)
target_link_libraries(example_convnd_fwd_xdl_fp16 PRIVATE conv_util)
example/09_convnd_fwd/convnd_fwd_xdl_fp16.cpp
View file @ e00a943e
......@@ -5,7 +5,7 @@
#include "check_err.hpp"
#include "config.hpp"
#include "conv_fwd_util.hpp"
#include "conv_util.hpp"
#include "device.hpp"
#include "device_tensor.hpp"
#include "device_convnd_fwd_xdl_nhwc_kyxc_nhwk.hpp"
......@@ -43,10 +43,10 @@ template <ck::index_t NumDimSpatial>
using DeviceConvNDFwdInstance = ck::tensor_operation::device::
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K<
// clang-format off
InDataType, //
InDataType, //
WeiDataType, //
OutDataType, //
AccDataType, //
AccDataType, //
InElementOp, // Input Elementwise Operation
WeiElementOp, // Weights Elementwise Operation
OutElementOp, // Output Elementwise Operation
......@@ -110,7 +110,7 @@ void print_use_msg()
{
std::cout << "arg1: verification (0=no, 1=yes)\n"
<< "arg2: initialization (0=no init, 1=integer value, 2=decimal value)\n"
<< "arg3: run kernel # of times (>1)\n"
<< "arg3: time kernel (0=n0, 1=yes)\n"
<< "arg4: N spatial dimensions (default 2)\n"
<< "Following arguments (depending on number of spatial dims):\n"
<< " N, K, C, \n"
......@@ -137,40 +137,40 @@ ck::utils::conv::ConvParams parse_conv_params(int num_dim_spatial, int argc, cha
ck::utils::conv::ConvParams params;
int arg_idx = 5;
params.num_dim_spatial = num_dim_spatial;
params.N = std::stoi(argv[arg_idx++]);
params.K = std::stoi(argv[arg_idx++]);
params.C = std::stoi(argv[arg_idx++]);
params.num_dim_spatial_ = num_dim_spatial;
params.N_ = std::stoi(argv[arg_idx++]);
params.K_ = std::stoi(argv[arg_idx++]);
params.C_ = std::stoi(argv[arg_idx++]);
params.filter_spatial_lengths.resize(num_dim_spatial);
params.filter_spatial_lengths_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i)
{
params.filter_spatial_lengths[i] = std::stoi(argv[arg_idx++]);
params.filter_spatial_lengths_[i] = std::stoi(argv[arg_idx++]);
}
params.input_spatial_lengths.resize(num_dim_spatial);
params.input_spatial_lengths_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i)
{
params.input_spatial_lengths[i] = std::stoi(argv[arg_idx++]);
params.input_spatial_lengths_[i] = std::stoi(argv[arg_idx++]);
}
params.conv_filter_strides.resize(num_dim_spatial);
params.conv_filter_strides_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i)
{
params.conv_filter_strides[i] = std::stoi(argv[arg_idx++]);
params.conv_filter_strides_[i] = std::stoi(argv[arg_idx++]);
}
params.conv_filter_dilations.resize(num_dim_spatial);
params.conv_filter_dilations_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i)
{
params.conv_filter_dilations[i] = std::stoi(argv[arg_idx++]);
params.conv_filter_dilations_[i] = std::stoi(argv[arg_idx++]);
}
params.input_left_pads.resize(num_dim_spatial);
params.input_left_pads_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i)
{
params.input_left_pads[i] = std::stoi(argv[arg_idx++]);
params.input_left_pads_[i] = std::stoi(argv[arg_idx++]);
}
params.input_right_pads.resize(num_dim_spatial);
params.input_right_pads_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i)
{
params.input_right_pads[i] = std::stoi(argv[arg_idx++]);
params.input_right_pads_[i] = std::stoi(argv[arg_idx++]);
}
return params;
......@@ -182,9 +182,9 @@ int main(int argc, char* argv[])
{
using namespace ck::utils::conv;
bool do_verification = 0;
int init_method = 0;
int nrepeat = 5;
bool do_verification = true;
int init_method = 1;
bool time_kernel = false;
int num_dim_spatial = 2;
ck::utils::conv::ConvParams params;
......@@ -193,7 +193,7 @@ int main(int argc, char* argv[])
{
do_verification = std::stoi(argv[1]);
init_method = std::stoi(argv[2]);
nrepeat = std::stoi(argv[3]);
time_kernel = std::stoi(argv[3]);
num_dim_spatial = std::stoi(argv[4]);
}
......@@ -202,21 +202,21 @@ int main(int argc, char* argv[])
params = parse_conv_params(num_dim_spatial, argc, argv);
}
std::vector<std::size_t> input_dims{static_cast<std::size_t>(params.N),
static_cast<std::size_t>(params.C)};
std::vector<std::size_t> input_dims{static_cast<std::size_t>(params.N_),
static_cast<std::size_t>(params.C_)};
input_dims.insert(std::end(input_dims),
std::begin(params.input_spatial_lengths),
std::end(params.input_spatial_lengths));
std::begin(params.input_spatial_lengths_),
std::end(params.input_spatial_lengths_));
std::vector<std::size_t> filter_dims{static_cast<std::size_t>(params.K),
static_cast<std::size_t>(params.C)};
std::vector<std::size_t> filter_dims{static_cast<std::size_t>(params.K_),
static_cast<std::size_t>(params.C_)};
filter_dims.insert(std::end(filter_dims),
std::begin(params.filter_spatial_lengths),
std::end(params.filter_spatial_lengths));
std::begin(params.filter_spatial_lengths_),
std::end(params.filter_spatial_lengths_));
const std::vector<ck::index_t>& output_spatial_lengths = params.GetOutputSpatialLengths();
std::vector<std::size_t> output_dims{static_cast<std::size_t>(params.N),
static_cast<std::size_t>(params.K)};
std::vector<std::size_t> output_dims{static_cast<std::size_t>(params.N_),
static_cast<std::size_t>(params.K_)};
output_dims.insert(std::end(output_dims),
std::begin(output_spatial_lengths),
std::end(output_spatial_lengths));
......@@ -256,16 +256,16 @@ int main(int argc, char* argv[])
conv->MakeArgumentPointer(static_cast<InDataType*>(in_device_buf.GetDeviceBuffer()),
static_cast<WeiDataType*>(wei_device_buf.GetDeviceBuffer()),
static_cast<OutDataType*>(out_device_buf.GetDeviceBuffer()),
params.N,
params.K,
params.C,
params.input_spatial_lengths,
params.filter_spatial_lengths,
params.N_,
params.K_,
params.C_,
params.input_spatial_lengths_,
params.filter_spatial_lengths_,
output_spatial_lengths,
params.conv_filter_strides,
params.conv_filter_dilations,
params.input_left_pads,
params.input_right_pads,
params.conv_filter_strides_,
params.conv_filter_dilations_,
params.input_left_pads_,
params.input_right_pads_,
InElementOp{},
WeiElementOp{},
OutElementOp{});
......@@ -277,16 +277,16 @@ int main(int argc, char* argv[])
"not support this Conv problem");
}
float ave_time = invoker->Run(argument.get(), nrepeat);
float ave_time = invoker->Run(argument.get(), StreamConfig{nullptr, time_kernel});
std::size_t flop = get_flops(
params.N, params.C, params.K, params.filter_spatial_lengths, output_spatial_lengths);
params.N_, params.C_, params.K_, params.filter_spatial_lengths_, output_spatial_lengths);
std::size_t num_btype = get_btype<InDataType, WeiDataType, OutDataType>(
params.N,
params.C,
params.K,
params.input_spatial_lengths,
params.filter_spatial_lengths,
params.N_,
params.C_,
params.K_,
params.input_spatial_lengths_,
params.filter_spatial_lengths_,
output_spatial_lengths);
float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
......@@ -302,18 +302,18 @@ int main(int argc, char* argv[])
auto ref_argument = ref_conv.MakeArgument(input,
weights,
host_output,
params.conv_filter_strides,
params.conv_filter_dilations,
params.input_left_pads,
params.input_right_pads,
params.conv_filter_strides_,
params.conv_filter_dilations_,
params.input_left_pads_,
params.input_right_pads_,
InElementOp{},
WeiElementOp{},
OutElementOp{});
ref_invoker.Run(ref_argument);
out_device_buf.FromDevice(device_output.mData.data());
ck::utils::check_err(
host_output.mData, device_output.mData, "Error: incorrect results!", 1e-5f, 1e-4f);
return ck::utils::check_err(
host_output.mData, device_output.mData, "Error: incorrect results!", 1e-5f, 1e-4f) ? 0 : 1;
};
switch(num_dim_spatial)
......@@ -338,4 +338,5 @@ int main(int argc, char* argv[])
}
}
}
return 0;
}
example/09_convnd_fwd/convnd_fwd_xdl.cpp → example/09_convnd_fwd/convnd_fwd_xdl_fp32.cpp
View file @ e00a943e
......@@ -5,7 +5,7 @@
#include "check_err.hpp"
#include "config.hpp"
#include "conv_fwd_util.hpp"
#include "conv_util.hpp"
#include "device.hpp"
#include "device_tensor.hpp"
#include "device_convnd_fwd_xdl_nhwc_kyxc_nhwk.hpp"
......@@ -39,10 +39,10 @@ template <ck::index_t NumDimSpatial>
using DeviceConvNDFwdInstance = ck::tensor_operation::device::
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K<
// clang-format off
InDataType, //
InDataType, //
WeiDataType, //
OutDataType, //
AccDataType, //
AccDataType, //
InElementOp, // Input Elementwise Operation
WeiElementOp, // Weights Elementwise Operation
OutElementOp, // Output Elementwise Operation
......@@ -107,7 +107,7 @@ void print_use_msg()
{
std::cout << "arg1: verification (0=no, 1=yes)\n"
<< "arg2: initialization (0=no init, 1=integer value, 2=decimal value)\n"
<< "arg3: run kernel # of times (>1)\n"
<< "arg3: time kernel (0=n0, 1=yes)\n"
<< "arg4: N spatial dimensions (default 2)\n"
<< "Following arguments (depending on number of spatial dims):\n"
<< " N, K, C, \n"
......@@ -134,40 +134,40 @@ ck::utils::conv::ConvParams parse_conv_params(int num_dim_spatial, int argc, cha
ck::utils::conv::ConvParams params;
int arg_idx = 5;
params.num_dim_spatial = num_dim_spatial;
params.N = std::stoi(argv[arg_idx++]);
params.K = std::stoi(argv[arg_idx++]);
params.C = std::stoi(argv[arg_idx++]);
params.num_dim_spatial_ = num_dim_spatial;
params.N_ = std::stoi(argv[arg_idx++]);
params.K_ = std::stoi(argv[arg_idx++]);
params.C_ = std::stoi(argv[arg_idx++]);
params.filter_spatial_lengths.resize(num_dim_spatial);
params.filter_spatial_lengths_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i)
{
params.filter_spatial_lengths[i] = std::stoi(argv[arg_idx++]);
params.filter_spatial_lengths_[i] = std::stoi(argv[arg_idx++]);
}
params.input_spatial_lengths.resize(num_dim_spatial);
params.input_spatial_lengths_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i)
{
params.input_spatial_lengths[i] = std::stoi(argv[arg_idx++]);
params.input_spatial_lengths_[i] = std::stoi(argv[arg_idx++]);
}
params.conv_filter_strides.resize(num_dim_spatial);
params.conv_filter_strides_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i)
{
params.conv_filter_strides[i] = std::stoi(argv[arg_idx++]);
params.conv_filter_strides_[i] = std::stoi(argv[arg_idx++]);
}
params.conv_filter_dilations.resize(num_dim_spatial);
params.conv_filter_dilations_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i)
{
params.conv_filter_dilations[i] = std::stoi(argv[arg_idx++]);
params.conv_filter_dilations_[i] = std::stoi(argv[arg_idx++]);
}
params.input_left_pads.resize(num_dim_spatial);
params.input_left_pads_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i)
{
params.input_left_pads[i] = std::stoi(argv[arg_idx++]);
params.input_left_pads_[i] = std::stoi(argv[arg_idx++]);
}
params.input_right_pads.resize(num_dim_spatial);
params.input_right_pads_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i)
{
params.input_right_pads[i] = std::stoi(argv[arg_idx++]);
params.input_right_pads_[i] = std::stoi(argv[arg_idx++]);
}
return params;
......@@ -179,9 +179,9 @@ int main(int argc, char* argv[])
{
using namespace ck::utils::conv;
bool do_verification = 0;
int init_method = 0;
int nrepeat = 5;
bool do_verification = true;
int init_method = 1;
bool time_kernel = false;
int num_dim_spatial = 2;
ck::utils::conv::ConvParams params;
......@@ -190,7 +190,7 @@ int main(int argc, char* argv[])
{
do_verification = std::stoi(argv[1]);
init_method = std::stoi(argv[2]);
nrepeat = std::stoi(argv[3]);
time_kernel = std::stoi(argv[3]);
num_dim_spatial = std::stoi(argv[4]);
}
......@@ -199,21 +199,21 @@ int main(int argc, char* argv[])
params = parse_conv_params(num_dim_spatial, argc, argv);
}
std::vector<std::size_t> input_dims{static_cast<std::size_t>(params.N),
static_cast<std::size_t>(params.C)};
std::vector<std::size_t> input_dims{static_cast<std::size_t>(params.N_),
static_cast<std::size_t>(params.C_)};
input_dims.insert(std::end(input_dims),
std::begin(params.input_spatial_lengths),
std::end(params.input_spatial_lengths));
std::begin(params.input_spatial_lengths_),
std::end(params.input_spatial_lengths_));
std::vector<std::size_t> filter_dims{static_cast<std::size_t>(params.K),
static_cast<std::size_t>(params.C)};
std::vector<std::size_t> filter_dims{static_cast<std::size_t>(params.K_),
static_cast<std::size_t>(params.C_)};
filter_dims.insert(std::end(filter_dims),
std::begin(params.filter_spatial_lengths),
std::end(params.filter_spatial_lengths));
std::begin(params.filter_spatial_lengths_),
std::end(params.filter_spatial_lengths_));
const std::vector<ck::index_t>& output_spatial_lengths = params.GetOutputSpatialLengths();
std::vector<std::size_t> output_dims{static_cast<std::size_t>(params.N),
static_cast<std::size_t>(params.K)};
std::vector<std::size_t> output_dims{static_cast<std::size_t>(params.N_),
static_cast<std::size_t>(params.K_)};
output_dims.insert(std::end(output_dims),
std::begin(output_spatial_lengths),
std::end(output_spatial_lengths));
......@@ -255,16 +255,16 @@ int main(int argc, char* argv[])
conv->MakeArgumentPointer(static_cast<InDataType*>(in_device_buf.GetDeviceBuffer()),
static_cast<WeiDataType*>(wei_device_buf.GetDeviceBuffer()),
static_cast<OutDataType*>(out_device_buf.GetDeviceBuffer()),
params.N,
params.K,
params.C,
params.input_spatial_lengths,
params.filter_spatial_lengths,
params.N_,
params.K_,
params.C_,
params.input_spatial_lengths_,
params.filter_spatial_lengths_,
output_spatial_lengths,
params.conv_filter_strides,
params.conv_filter_dilations,
params.input_left_pads,
params.input_right_pads,
params.conv_filter_strides_,
params.conv_filter_dilations_,
params.input_left_pads_,
params.input_right_pads_,
InElementOp{},
WeiElementOp{},
OutElementOp{});
......@@ -276,16 +276,16 @@ int main(int argc, char* argv[])
"not support this Conv problem");
}
float ave_time = invoker->Run(argument.get(), nrepeat);
float ave_time = invoker->Run(argument.get(), StreamConfig{nullptr, time_kernel});
std::size_t flop = get_flops(
params.N, params.C, params.K, params.filter_spatial_lengths, output_spatial_lengths);
params.N_, params.C_, params.K_, params.filter_spatial_lengths_, output_spatial_lengths);
std::size_t num_btype =
get_btype<InDataType, WeiDataType, OutDataType>(params.N,
params.C,
params.K,
params.input_spatial_lengths,
params.filter_spatial_lengths,
get_btype<InDataType, WeiDataType, OutDataType>(params.N_,
params.C_,
params.K_,
params.input_spatial_lengths_,
params.filter_spatial_lengths_,
output_spatial_lengths);
float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
......@@ -301,18 +301,23 @@ int main(int argc, char* argv[])
auto ref_argument = ref_conv.MakeArgument(input,
weights,
host_output,
params.conv_filter_strides,
params.conv_filter_dilations,
params.input_left_pads,
params.input_right_pads,
params.conv_filter_strides_,
params.conv_filter_dilations_,
params.input_left_pads_,
params.input_right_pads_,
InElementOp{},
WeiElementOp{},
OutElementOp{});
ref_invoker.Run(ref_argument);
out_device_buf.FromDevice(device_output.mData.data());
ck::utils::check_err(
host_output.mData, device_output.mData, "Error: incorrect results!", 1e-5f, 1e-4f);
return ck::utils::check_err(device_output.mData,
host_output.mData,
"Error: incorrect results!",
1e-5f,
1e-4f)
? 0
: 1;
};
switch(num_dim_spatial)
......@@ -337,4 +342,5 @@ int main(int argc, char* argv[])
}
}
}
return 0;
}
example/09_convnd_fwd/convnd_fwd_xdl_int8.cpp
View file @ e00a943e
......@@ -5,7 +5,7 @@
#include "check_err.hpp"
#include "config.hpp"
#include "conv_fwd_util.hpp"
#include "conv_util.hpp"
#include "device.hpp"
#include "device_tensor.hpp"
#include "device_convnd_fwd_xdl_nhwc_kyxc_nhwk.hpp"
......@@ -45,10 +45,10 @@ template <ck::index_t NumDimSpatial>
using DeviceConvNDFwdInstance = ck::tensor_operation::device::
DeviceConvNDFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K<
// clang-format off
InDataType, //
InDataType, //
WeiDataType, //
OutDataType, //
AccDataType, //
AccDataType, //
InElementOp, // Input Elementwise Operation
WeiElementOp, // Weights Elementwise Operation
OutElementOp, // Output Elementwise Operation
......@@ -112,7 +112,7 @@ void print_use_msg()
{
std::cout << "arg1: verification (0=no, 1=yes)\n"
<< "arg2: initialization (0=no init, 1=integer value, 2=decimal value)\n"
<< "arg3: run kernel # of times (>1)\n"
<< "arg3: time kernel (0=n0, 1=yes)\n"
<< "arg4: N spatial dimensions (default 2)\n"
<< "Following arguments (depending on number of spatial dims):\n"
<< " N, K, C, \n"
......@@ -139,40 +139,40 @@ ck::utils::conv::ConvParams parse_conv_params(int num_dim_spatial, int argc, cha
ck::utils::conv::ConvParams params;
int arg_idx = 5;
params.num_dim_spatial = num_dim_spatial;
params.N = std::stoi(argv[arg_idx++]);
params.K = std::stoi(argv[arg_idx++]);
params.C = std::stoi(argv[arg_idx++]);
params.num_dim_spatial_ = num_dim_spatial;
params.N_ = std::stoi(argv[arg_idx++]);
params.K_ = std::stoi(argv[arg_idx++]);
params.C_ = std::stoi(argv[arg_idx++]);
params.filter_spatial_lengths.resize(num_dim_spatial);
params.filter_spatial_lengths_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i)
{
params.filter_spatial_lengths[i] = std::stoi(argv[arg_idx++]);
params.filter_spatial_lengths_[i] = std::stoi(argv[arg_idx++]);
}
params.input_spatial_lengths.resize(num_dim_spatial);
params.input_spatial_lengths_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i)
{
params.input_spatial_lengths[i] = std::stoi(argv[arg_idx++]);
params.input_spatial_lengths_[i] = std::stoi(argv[arg_idx++]);
}
params.conv_filter_strides.resize(num_dim_spatial);
params.conv_filter_strides_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i)
{
params.conv_filter_strides[i] = std::stoi(argv[arg_idx++]);
params.conv_filter_strides_[i] = std::stoi(argv[arg_idx++]);
}
params.conv_filter_dilations.resize(num_dim_spatial);
params.conv_filter_dilations_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i)
{
params.conv_filter_dilations[i] = std::stoi(argv[arg_idx++]);
params.conv_filter_dilations_[i] = std::stoi(argv[arg_idx++]);
}
params.input_left_pads.resize(num_dim_spatial);
params.input_left_pads_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i)
{
params.input_left_pads[i] = std::stoi(argv[arg_idx++]);
params.input_left_pads_[i] = std::stoi(argv[arg_idx++]);
}
params.input_right_pads.resize(num_dim_spatial);
params.input_right_pads_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i)
{
params.input_right_pads[i] = std::stoi(argv[arg_idx++]);
params.input_right_pads_[i] = std::stoi(argv[arg_idx++]);
}
return params;
......@@ -184,9 +184,9 @@ int main(int argc, char* argv[])
{
using namespace ck::utils::conv;
bool do_verification = 0;
int init_method = 0;
int nrepeat = 5;
bool do_verification = true;
int init_method = 1;
bool time_kernel = false;
int num_dim_spatial = 2;
ck::utils::conv::ConvParams params;
......@@ -195,7 +195,7 @@ int main(int argc, char* argv[])
{
do_verification = std::stoi(argv[1]);
init_method = std::stoi(argv[2]);
nrepeat = std::stoi(argv[3]);
time_kernel = std::stoi(argv[3]);
num_dim_spatial = std::stoi(argv[4]);
}
......@@ -204,21 +204,21 @@ int main(int argc, char* argv[])
params = parse_conv_params(num_dim_spatial, argc, argv);
}
std::vector<std::size_t> input_dims{static_cast<std::size_t>(params.N),
static_cast<std::size_t>(params.C)};
std::vector<std::size_t> input_dims{static_cast<std::size_t>(params.N_),
static_cast<std::size_t>(params.C_)};
input_dims.insert(std::end(input_dims),
std::begin(params.input_spatial_lengths),
std::end(params.input_spatial_lengths));
std::begin(params.input_spatial_lengths_),
std::end(params.input_spatial_lengths_));
std::vector<std::size_t> filter_dims{static_cast<std::size_t>(params.K),
static_cast<std::size_t>(params.C)};
std::vector<std::size_t> filter_dims{static_cast<std::size_t>(params.K_),
static_cast<std::size_t>(params.C_)};
filter_dims.insert(std::end(filter_dims),
std::begin(params.filter_spatial_lengths),
std::end(params.filter_spatial_lengths));
std::begin(params.filter_spatial_lengths_),
std::end(params.filter_spatial_lengths_));
const std::vector<ck::index_t>& output_spatial_lengths = params.GetOutputSpatialLengths();
std::vector<std::size_t> output_dims{static_cast<std::size_t>(params.N),
static_cast<std::size_t>(params.K)};
std::vector<std::size_t> output_dims{static_cast<std::size_t>(params.N_),
static_cast<std::size_t>(params.K_)};
output_dims.insert(std::end(output_dims),
std::begin(output_spatial_lengths),
std::end(output_spatial_lengths));
......@@ -258,16 +258,16 @@ int main(int argc, char* argv[])
conv->MakeArgumentPointer(static_cast<InDataType*>(in_device_buf.GetDeviceBuffer()),
static_cast<WeiDataType*>(wei_device_buf.GetDeviceBuffer()),
static_cast<OutDataType*>(out_device_buf.GetDeviceBuffer()),
params.N,
params.K,
params.C,
params.input_spatial_lengths,
params.filter_spatial_lengths,
params.N_,
params.K_,
params.C_,
params.input_spatial_lengths_,
params.filter_spatial_lengths_,
output_spatial_lengths,
params.conv_filter_strides,
params.conv_filter_dilations,
params.input_left_pads,
params.input_right_pads,
params.conv_filter_strides_,
params.conv_filter_dilations_,
params.input_left_pads_,
params.input_right_pads_,
InElementOp{},
WeiElementOp{},
OutElementOp{});
......@@ -279,16 +279,16 @@ int main(int argc, char* argv[])
"not support this Conv problem");
}
float ave_time = invoker->Run(argument.get(), nrepeat);
float ave_time = invoker->Run(argument.get(), StreamConfig{nullptr, time_kernel});
std::size_t flop = get_flops(
params.N, params.C, params.K, params.filter_spatial_lengths, output_spatial_lengths);
params.N_, params.C_, params.K_, params.filter_spatial_lengths_, output_spatial_lengths);
std::size_t num_btype = get_btype<InDataType, WeiDataType, OutDataType>(
params.N,
params.C,
params.K,
params.input_spatial_lengths,
params.filter_spatial_lengths,
params.N_,
params.C_,
params.K_,
params.input_spatial_lengths_,
params.filter_spatial_lengths_,
output_spatial_lengths);
float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
......@@ -304,18 +304,18 @@ int main(int argc, char* argv[])
auto ref_argument = ref_conv.MakeArgument(input,
weights,
host_output,
params.conv_filter_strides,
params.conv_filter_dilations,
params.input_left_pads,
params.input_right_pads,
params.conv_filter_strides_,
params.conv_filter_dilations_,
params.input_left_pads_,
params.input_right_pads_,
InElementOp{},
WeiElementOp{},
OutElementOp{});
ref_invoker.Run(ref_argument);
out_device_buf.FromDevice(device_output.mData.data());
ck::utils::check_err(
host_output.mData, device_output.mData, "Error: incorrect results!", 1e-5f, 1e-4f);
return ck::utils::check_err(
host_output.mData, device_output.mData, "Error: incorrect results!", 1e-5f, 1e-4f) ? 0 : 1;
};
switch(num_dim_spatial)
......@@ -340,4 +340,5 @@ int main(int argc, char* argv[])
}
}
}
return 0;
}
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