saymrwulf/onnxruntime
1
0
Fork 0
mirror of https://github.com/saymrwulf/onnxruntime.git synced 2026-07-12 17:57:38 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Add build option to enable cuda profiling (#9875)

Browse source
This commit is contained in:
RandySheriffH 2021-11-29 22:44:50 -08:00 committed by GitHub
parent 16bfd3c771
commit 9345894c82
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
8 changed files with 24 additions and 7 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

6
cmake/CMakeLists.txt
View file

@ -175,6 +175,8 @@ option(onnxruntime_PREBUILT_PYTORCH_PATH "Path to pytorch installation dir")
# external transformer src path
option(onnxruntime_EXTERNAL_TRANSFORMER_SRC_PATH "Path to external transformer src dir")
option(onnxruntime_ENABLE_CUDA_PROFILING "Enable CUDA kernel profiling" OFF)
if (onnxruntime_USE_CUDA)
set(onnxruntime_DISABLE_RTTI OFF)
endif()
@ -1700,6 +1702,10 @@ if (onnxruntime_ENABLE_TRAINING_OPS)
add_compile_definitions(ENABLE_TRAINING_OPS)
endif()
if (onnxruntime_ENABLE_CUDA_PROFILING)
add_compile_definitions(ENABLE_CUDA_PROFILING)
endif()
if (onnxruntime_ENABLE_TRAINING)
add_compile_definitions(ENABLE_TRAINING)
add_compile_definitions(ENABLE_TRAINING_OPS)

11
cmake/onnxruntime_providers.cmake
View file

@ -353,13 +353,18 @@ if (onnxruntime_USE_CUDA)
endif()
add_dependencies(onnxruntime_providers_cuda onnxruntime_providers_shared ${onnxruntime_EXTERNAL_DEPENDENCIES} ${onnxruntime_tvm_dependencies})
target_link_directories(onnxruntime_providers_cuda PRIVATE ${onnxruntime_CUDA_HOME}/extras/CUPTI/lib64)
target_link_libraries(onnxruntime_providers_cuda PRIVATE cublas cudnn curand cufft cupti ${ONNXRUNTIME_PROVIDERS_SHARED})
target_include_directories(onnxruntime_providers_cuda PRIVATE ${ONNXRUNTIME_ROOT} ${CMAKE_CURRENT_BINARY_DIR} ${onnxruntime_CUDNN_HOME}/include ${eigen_INCLUDE_DIRS} ${TVM_INCLUDES} PUBLIC ${CMAKE_CUDA_TOOLKIT_INCLUDE_DIRECTORIES} ${onnxruntime_CUDA_HOME}/extras/CUPTI/include)
target_link_libraries(onnxruntime_providers_cuda PRIVATE cublas cudnn curand cufft ${ONNXRUNTIME_PROVIDERS_SHARED})
target_include_directories(onnxruntime_providers_cuda PRIVATE ${ONNXRUNTIME_ROOT} ${CMAKE_CURRENT_BINARY_DIR} ${onnxruntime_CUDNN_HOME}/include ${eigen_INCLUDE_DIRS} ${TVM_INCLUDES} PUBLIC ${CMAKE_CUDA_TOOLKIT_INCLUDE_DIRECTORIES})
# ${CMAKE_CURRENT_BINARY_DIR} is so that #include "onnxruntime_config.h" inside tensor_shape.h is found
set_target_properties(onnxruntime_providers_cuda PROPERTIES LINKER_LANGUAGE CUDA)
set_target_properties(onnxruntime_providers_cuda PROPERTIES FOLDER "ONNXRuntime")
if (onnxruntime_ENABLE_CUDA_PROFILING) # configure cupti for cuda profiling
target_include_directories(onnxruntime_providers_cuda PRIVATE ${onnxruntime_CUDA_HOME}/extras/CUPTI/include)
target_link_directories(onnxruntime_providers_cuda PRIVATE ${onnxruntime_CUDA_HOME}/extras/CUPTI/lib64)
target_link_libraries(onnxruntime_providers_cuda PRIVATE cupti)
endif()
if (onnxruntime_ENABLE_NVTX_PROFILE)
target_link_libraries(onnxruntime_providers_cuda PRIVATE nvToolsExt)
endif()

2
onnxruntime/core/providers/cuda/cuda_profiler.cc
View file

@ -1,6 +1,6 @@
// Copyright (c) Microsoft Corporation. All rights reserved.
// Licensed under the MIT License.
#if !(defined(USE_ROCM) || defined(ENABLE_TRAINING))
#if defined(USE_CUDA) && defined(ENABLE_CUDA_PROFILING)
#include "cuda_profiler.h"
#include <map>

2
onnxruntime/core/providers/cuda/cuda_profiler.h
View file

@ -2,7 +2,7 @@
// Licensed under the MIT License.
#include "core/common/profiler_common.h"
#if !(defined(USE_ROCM) || defined(ENABLE_TRAINING))
#if defined(USE_CUDA) && defined(ENABLE_CUDA_PROFILING)
#include "core/platform/ort_mutex.h"
#include <cupti.h>

2
onnxruntime/test/framework/inference_session_test.cc
View file

@ -661,7 +661,7 @@ TEST(InferenceSessionTests, CheckRunProfilerWithSessionOptions) {
}
}
#if defined(USE_CUDA) && !defined(ENABLE_TRAINING) && defined(CUDA_VERSION) && CUDA_VERSION >= 11000
#if defined(USE_CUDA) && defined(ENABLE_CUDA_PROFILING)
ASSERT_TRUE(has_kernel_info);
#endif
}

5
tools/ci_build/build.py
View file

@ -567,6 +567,10 @@ def parse_arguments():
"--test_external_transformer_example", action='store_true',
help="run the example external transformer test, mainly used in CI pipeline.")
parser.add_argument(
"--enable_cuda_profiling", action='store_true', help="enable cuda kernel profiling, \
cupti library must be added to PATH beforehand.")
return parser.parse_args()
@ -817,6 +821,7 @@ def generate_build_tree(cmake_path, source_dir, build_dir, cuda_home, cudnn_home
"-Donnxruntime_ENABLE_EXTERNAL_CUSTOM_OP_SCHEMAS=" + ("ON" if args.enable_external_custom_op_schemas
else "OFF"),
"-Donnxruntime_NVCC_THREADS=" + str(args.parallel),
"-Donnxruntime_ENABLE_CUDA_PROFILING=" + ("ON" if args.enable_cuda_profiling else "OFF"),
]
if args.external_graph_transformer_path:
cmake_args.append("-Donnxruntime_EXTERNAL_TRANSFORMER_SRC_PATH=" + args.external_graph_transformer_path)

1
tools/ci_build/github/azure-pipelines/linux-gpu-ci-pipeline.yml
View file

@ -38,6 +38,7 @@ jobs:
--parallel \
--build_wheel \
--enable_onnx_tests --use_cuda --cuda_version=11.4 --cuda_home=/usr/local/cuda-11.4 --cudnn_home=/usr/local/cuda-11.4 \
--enable_cuda_profiling \
--enable_pybind --build_java \
--cmake_extra_defines CMAKE_CUDA_HOST_COMPILER=/opt/rh/devtoolset-10/root/usr/bin/cc CMAKE_CUDA_ARCHITECTURES=52
workingDirectory: $(Build.SourcesDirectory)

2
tools/ci_build/github/azure-pipelines/win-gpu-ci-pipeline.yml
View file

@ -13,7 +13,7 @@ stages:
strategy:
matrix:
cuda:
additionalBuildFlags: --build_java --build_nodejs --use_cuda --cuda_version=11.4 --cuda_home="C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.4" --cudnn_home="C:\local\cudnn-11.4-windows-x64-v8.2.2.26\cuda" --cmake_extra_defines CMAKE_CUDA_ARCHITECTURES=52 --gen_doc validate
additionalBuildFlags: --build_java --build_nodejs --use_cuda --cuda_version=11.4 --cuda_home="C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.4" --cudnn_home="C:\local\cudnn-11.4-windows-x64-v8.2.2.26\cuda" --cmake_extra_defines CMAKE_CUDA_ARCHITECTURES=52 --gen_doc validate --enable_cuda_profiling
EnvSetupScript: setup_env_cuda_11.bat
ORT_EP_NAME: CUDA
dml: