saymrwulf/onnxruntime
1
0
Fork 0
mirror of https://github.com/saymrwulf/onnxruntime.git synced 2026-07-29 20:14:01 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Enable LTO on Linux

Browse source
This commit is contained in:
Changming Sun 2019-03-21 15:53:06 -07:00
parent 12ecd77ffb
commit a26696fb0e
15 changed files with 56 additions and 24 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

22
cmake/CMakeLists.txt
View file

@ -6,6 +6,9 @@ cmake_minimum_required(VERSION 3.11)
# Project
project(onnxruntime C CXX)
cmake_policy(SET CMP0069 NEW)
set(CMAKE_POLICY_DEFAULT_CMP0069 NEW)
include(CheckCXXCompilerFlag)
include(CheckLanguage)
@ -65,6 +68,7 @@ option(onnxruntime_ENABLE_MICROSOFT_INTERNAL "Use this option to enable/disable
option(onnxruntime_USE_NUPHAR "Build with Nupha" OFF)
option(onnxruntime_USE_BRAINSLICE "Build with BrainSlice" OFF)
option(onnxruntime_USE_TENSORRT "Build with TensorRT support" OFF)
option(onnxruntime_ENABLE_LTO "Enable link time optimization, which is not stable on older GCCs" OFF)
option(onnxruntime_CROSS_COMPILING "Cross compiling onnx runtime" OFF)
option(onnxruntime_USE_FULL_PROTOBUF "Use full protobuf" OFF)
@ -73,6 +77,24 @@ set(protobuf_BUILD_TESTS OFF CACHE BOOL "Build protobuf tests" FORCE)
set(NSYNC_ENABLE_TESTS OFF CACHE BOOL "Build protobuf tests" FORCE)
set(ONNX_ML 1)
if(onnxruntime_ENABLE_LTO)
#LTO(or LTCG) is great, in our case it can reduce binary size by 1/3.
#cmake can only help us check if the compiler support LTO or not, it can't tell us if the feature works well
#Don't enable this option in Ubuntu 16.04, protoc will crash
include(CheckIPOSupported)
check_ipo_supported(RESULT ipo_enabled OUTPUT ipo_output)
#TODO: figure out why nsync doesn't work
if(NOT onnxruntime_USE_NSYNC)
if(ipo_enabled)
set(CMAKE_INTERPROCEDURAL_OPTIMIZATION_RELEASE ON)
set(CMAKE_INTERPROCEDURAL_OPTIMIZATION_RELWITHDEBINFO ON)
else()
message(WARNING "IPO is not supported: ${ipo_output}")
endif()
endif()
endif()
set(REPO_ROOT ${PROJECT_SOURCE_DIR}/..)
set(ONNXRUNTIME_ROOT ${PROJECT_SOURCE_DIR}/../onnxruntime)
file (STRINGS "${REPO_ROOT}/VERSION_NUMBER" VERSION_NUMBER)

3
cmake/onnxruntime_unittests.cmake
View file

@ -315,12 +315,15 @@ endif() # SingleUnitTestProject
# standalone test for inference session without environment
# the normal test executables set up a default runtime environment, which we don't want here
if(NOT ipo_enabled)
#TODO: figure out why this test doesn't work with gcc LTO
AddTest(
TARGET onnxruntime_test_framework_session_without_environment_standalone
SOURCES "${TEST_SRC_DIR}/framework/inference_session_without_environment/inference_session_without_environment_standalone_test.cc" "${TEST_SRC_DIR}/framework/test_main.cc"
LIBS onnxruntime_test_utils ${ONNXRUNTIME_TEST_LIBS}
DEPENDS ${onnxruntime_EXTERNAL_DEPENDENCIES}
)
endif()
#
# onnxruntime_ir_graph test data

4
include/onnxruntime/core/session/onnxruntime_cxx_api.h
View file

@ -117,13 +117,13 @@ class SessionOptionsWrapper {
}
#ifdef _WIN32
OrtSession* OrtCreateSession(_In_ const wchar_t* model_path) {
OrtSession* ret;
OrtSession* ret = nullptr;
ORT_THROW_ON_ERROR(::OrtCreateSession(env_, model_path, value.get(), &ret));
return ret;
}
#else
OrtSession* OrtCreateSession(_In_ const char* model_path) {
OrtSession* ret;
OrtSession* ret = nullptr;
ORT_THROW_ON_ERROR(::OrtCreateSession(env_, model_path, value.get(), &ret));
return ret;
}

3
onnxruntime/core/framework/cblas.h
View file

@ -13,8 +13,7 @@ enum CBLAS_ORDER { CblasRowMajor = 101,
enum CBLAS_TRANSPOSE {
CblasNoTrans = 111,
CblasTrans = 112,
CblasConjTrans = 113,
AtlasConj = 114
CblasConjTrans = 113
};
enum CBLAS_UPLO { CblasUpper = 121,
CblasLower = 122 };

2
onnxruntime/core/framework/session_state_initializer.cc
View file

@ -327,7 +327,7 @@ common::Status SaveInitializedTensors(const Env& env, const std::basic_string<PA
id_to_initialized_tensor[mlvalue_index] = entry.second;
}
for (const auto& entry : id_to_initialized_tensor) {
size_t len;
size_t len = 0;
ORT_RETURN_IF_ERROR(utils::GetSizeInBytesFromTensorProto<alignment>(*entry.second, &len));
ORT_RETURN_IF_ERROR(planner.TraceAllocation(entry.first, len));
}

8
onnxruntime/core/platform/posix/ort_mutex.cc
View file

@ -57,7 +57,7 @@ OrtCondVar::~OrtCondVar() {
void OrtCondVar::notify_one() noexcept {
#ifdef USE_NSYNC
nsync_cv_signal(&native_cv_object);
nsync::nsync_cv_signal(&native_cv_object);
#else
pthread_cond_signal(&native_cv_object);
#endif
@ -65,7 +65,7 @@ void OrtCondVar::notify_one() noexcept {
void OrtCondVar::notify_all() noexcept {
#ifdef USE_NSYNC
nsync_cv_broadcast(&native_cv_object);
nsync::nsync_cv_broadcast(&native_cv_object);
#else
pthread_cond_broadcast(&native_cv_object);
#endif
@ -75,7 +75,7 @@ void OrtCondVar::wait(std::unique_lock<OrtMutex>& lk) {
if (!lk.owns_lock())
throw std::runtime_error("OrtCondVar wait failed: mutex not locked");
#ifdef USE_NSYNC
nsync_cv_wait(&native_cv_object, lk.mutex()->native_handle());
nsync::nsync_cv_wait(&native_cv_object, lk.mutex()->native_handle());
#else
int ec = pthread_cond_wait(&native_cv_object, lk.mutex()->native_handle());
if (ec) {
@ -106,7 +106,7 @@ void OrtCondVar::timed_wait_impl(std::unique_lock<OrtMutex>& lk,
abs_deadline.tv_nsec = 999999999;
}
#ifdef USE_NSYNC
nsync_cv_wait_with_deadline(&native_cv_object, lk.mutex()->native_handle(), abs_deadline, nullptr);
nsync::nsync_cv_wait_with_deadline(&native_cv_object, lk.mutex()->native_handle(), abs_deadline, nullptr);
#else
int ec = pthread_cond_timedwait(&native_cv_object, lk.mutex()->native_handle(), &abs_deadline);
if (ec != 0 && ec != ETIMEDOUT) {

2
onnxruntime/core/providers/cuda/math/binary_elementwise_ops.h
View file

@ -58,7 +58,7 @@ struct BinaryElementwisePreparation {
// When N > 1: out[id] = op(lhs[id], rhs[id / H % C])
if (lhs_shape == output_shape) {
const auto& rhs_dims = rhs_shape.GetDims();
int64_t C;
int64_t C = 0;
if (1 == std::count_if(rhs_dims.begin(), rhs_dims.end(), [&C](int64_t dim) { if (dim > 1) C = dim; return (dim > 1); })) {
auto dim_C = std::find(rhs_dims.begin(), rhs_dims.end(), C) - rhs_dims.begin() + output_shape.NumDimensions() - rhs_shape.NumDimensions();
int64_t N = output_shape.SizeToDimension(dim_C);

5
onnxruntime/core/session/onnxruntime_c_api.cc
View file

@ -126,8 +126,11 @@ ORT_API_STATUS_IMPL(OrtCreateEnv, OrtLoggingLevel default_warning_level,
&name);
std::unique_ptr<Environment> env;
Status status = Environment::Create(env);
if (status.IsOK())
if (status.IsOK()) {
*out = new OrtEnv(env.release(), default_logging_manager.release());
return nullptr;
}
*out = nullptr;
return ToOrtStatus(status);
API_IMPL_END
}

4
onnxruntime/test/onnx/TestCase.cc
View file

@ -458,9 +458,9 @@ Status OnnxTestCase::ParseModel() {
Status st = Status::OK();
std::call_once(model_parsed_, [this, &st]() {
//parse model
ONNX_NAMESPACE::ModelProto* model_pb;
ONNX_NAMESPACE::ModelProto* model_pb = nullptr;
st = loadModelFile(model_url_.c_str(), &model_pb);
if (!st.IsOK()) return;
if (!st.IsOK() || model_pb == nullptr) return;
const ONNX_NAMESPACE::GraphProto& graph = model_pb->graph();
if (graph.node().size() == 1) {
node_name_ = graph.node()[0].op_type();

2
onnxruntime/test/providers/cpu/controlflow/if_test.cc
View file

@ -16,12 +16,14 @@ using namespace ONNX_NAMESPACE;
namespace onnxruntime {
namespace test {
namespace {
struct RunOptions {
bool include_dim_values_in_main_graph = false;
int symbolic_dim_value_in_main_graph = -1;
bool include_dim_values_in_subgraph = true;
bool mixed_execution_providers = false;
};
}
static const ONNX_NAMESPACE::GraphProto CreateSubgraph(bool then_branch, const RunOptions& options);

3
onnxruntime/test/providers/cpu/controlflow/loop_test.cc
View file

@ -17,13 +17,14 @@ using namespace ONNX_NAMESPACE;
namespace onnxruntime {
namespace test {
namespace {
struct RunOptions {
bool include_dim_values_in_main_graph = true;
bool include_dim_values_in_subgraph = false;
bool include_types_in_subgraph = false;
bool mixed_execution_providers = false;
};
}
static const ONNX_NAMESPACE::GraphProto CreateSubgraph(const RunOptions& options);
static const float kOuterNodeAddValue = 3.f;

2
onnxruntime/test/util/compare_mlvalue.cc
View file

@ -355,7 +355,7 @@ std::pair<COMPARE_RESULT, std::string> VerifyValueInfo(const ONNX_NAMESPACE::Val
//}
std::unique_ptr<OrtTensorTypeAndShapeInfo> info;
{
OrtTensorTypeAndShapeInfo* t1;
OrtTensorTypeAndShapeInfo* t1 = nullptr;
ORT_THROW_ON_ERROR(OrtGetTensorShapeAndType(o, &t1));
info.reset(t1);
}

2
tools/ci_build/github/linux/docker/scripts/install_deps.sh
View file

@ -65,5 +65,7 @@ tar -jxf /tmp/src/eigen-eigen-323c052e1731.tar.bz2 -C /usr/include
mv /usr/include/eigen-eigen-323c052e1731 /usr/include/eigen3
rm -rf /tmp/src
rm -rf /usr/include/google
rm -rf /usr/lib/libproto*

8
tools/ci_build/github/linux/run_build.sh
View file

@ -15,17 +15,16 @@ done
if [ $BUILD_DEVICE = "gpu" ]; then
_CUDNN_VERSION=$(echo $CUDNN_VERSION | cut -d. -f1-2)
python3 $SCRIPT_DIR/../../build.py --build_dir /home/onnxruntimedev \
python3 $SCRIPT_DIR/../../build.py --build_dir /build \
--config Debug Release \
--skip_submodule_sync --enable_onnx_tests \
--parallel --build_shared_lib \
--use_cuda --use_openmp \
--cuda_home /usr/local/cuda \
--cudnn_home /usr/local/cudnn-$_CUDNN_VERSION/cuda --build_shared_lib $BUILD_EXTR_PAR
/home/onnxruntimedev/Release/onnx_test_runner -e cuda /data/onnx
elif [ $BUILD_DEVICE = "tensorrt" ]; then
_CUDNN_VERSION=$(echo $CUDNN_VERSION | cut -d. -f1-2)
python3 $SCRIPT_DIR/../../build.py --build_dir /home/onnxruntimedev \
python3 $SCRIPT_DIR/../../build.py --build_dir /build \
--config Release \
--enable_onnx_tests \
--parallel --build_shared_lib \
@ -34,10 +33,9 @@ elif [ $BUILD_DEVICE = "tensorrt" ]; then
--cuda_home /usr/local/cuda \
--cudnn_home /usr/local/cuda --build_shared_lib $BUILD_EXTR_PAR
else
python3 $SCRIPT_DIR/../../build.py --build_dir /home/onnxruntimedev \
python3 $SCRIPT_DIR/../../build.py --build_dir /build \
--config Debug Release --build_shared_lib \
--skip_submodule_sync --enable_onnx_tests \
--build_wheel \
--parallel --use_openmp --build_shared_lib $BUILD_EXTR_PAR
/home/onnxruntimedev/Release/onnx_test_runner /data/onnx
fi

10
tools/ci_build/github/linux/run_dockerbuild.sh
View file

@ -49,15 +49,16 @@ else
fi
set +e
mkdir -p ~/.cache/onnxruntime
mkdir -p ~/.onnx
if [ $BUILD_DEVICE = "cpu" ]; then
docker rm -f "onnxruntime-$BUILD_DEVICE" || true
docker run -h $HOSTNAME \
--rm \
--name "onnxruntime-$BUILD_DEVICE" \
--volume "$SOURCE_ROOT:/onnxruntime_src" \
--volume "$BUILD_DIR:/home/onnxruntimedev" \
--volume "$BUILD_DIR:/build" \
--volume "$HOME/.cache/onnxruntime:/home/onnxruntimedev/.cache/onnxruntime" \
--volume "$HOME/.onnx:/home/onnxruntimedev/.onnx" \
"onnxruntime-$IMAGE" \
/bin/bash /onnxruntime_src/tools/ci_build/github/linux/run_build.sh \
-d $BUILD_DEVICE -x "$BUILD_EXTR_PAR" &
@ -67,8 +68,9 @@ else
--rm \
--name "onnxruntime-$BUILD_DEVICE" \
--volume "$SOURCE_ROOT:/onnxruntime_src" \
--volume "$BUILD_DIR:/home/onnxruntimedev" \
--volume "$BUILD_DIR:/build" \
--volume "$HOME/.cache/onnxruntime:/home/onnxruntimedev/.cache/onnxruntime" \
--volume "$HOME/.onnx:/home/onnxruntimedev/.onnx" \
"onnxruntime-$IMAGE" \
/bin/bash /onnxruntime_src/tools/ci_build/github/linux/run_build.sh \
-d $BUILD_DEVICE -x "$BUILD_EXTR_PAR" &