Expose build information in dynamic lib (#15643)

### Description
<!-- Describe your changes. -->
1. Add Build Info API to onnx.
2. Fix compile error while building onnxruntime_benchmark in MacOs.


### Motivation and Context
<!-- - Why is this change required? What problem does it solve?
- If it fixes an open issue, please link to the issue here. -->
1. When Onnxruntime lib is serving online, we need a way to detect how
this lib is built. This PR helps the developer to get the build
information using `strings` such as git branch, git commit id, build
type and cmake cxx flags, which is showed as follows.


![image](https://user-images.githubusercontent.com/19584326/233794371-b2f95a2c-27fb-4709-a6dd-bf4bb12b0b5b.png)


![image](https://user-images.githubusercontent.com/19584326/233794360-f96f5d2e-332c-405c-83f1-370ccc2b86f8.png)

If the build env has no git, there will be no git related infor:


![image](https://user-images.githubusercontent.com/19584326/234558596-298c1b01-9a90-41bf-9372-7259a8f8e5be.png)


3. Fix the following compile error while building benchmark in MacOs.

![image](https://user-images.githubusercontent.com/19584326/233793571-c261ac1f-47b2-434d-a293-7e9edc6c8a66.png)

---------

Co-authored-by: Yuhong Guo <yuhong.gyh@antgroup.com>

cmake/CMakeLists.txt

@@ -1236,6 +1236,19 @@ if (onnxruntime_USE_VITISAI)
   endif()
 endif()
 
+set(ORT_BUILD_INFO "ORT Build Info: ")
+find_package(Git)
+if (Git_FOUND)
+  execute_process(COMMAND ${GIT_EXECUTABLE} log -1 --format=%h
+      OUTPUT_VARIABLE ORT_GIT_COMMIT)
+  string(STRIP "${ORT_GIT_COMMIT}" ORT_GIT_COMMIT)
+  execute_process(COMMAND ${GIT_EXECUTABLE} rev-parse --abbrev-ref HEAD
+      OUTPUT_VARIABLE ORT_GIT_BRANCH)
+  string(STRIP "${ORT_GIT_BRANCH}" ORT_GIT_BRANCH)
+  string(APPEND ORT_BUILD_INFO "git-branch=${ORT_GIT_BRANCH}, git-commit-id=${ORT_GIT_COMMIT}, ")
+endif()
+string(APPEND ORT_BUILD_INFO "build type=${CMAKE_BUILD_TYPE}")
+string(APPEND ORT_BUILD_INFO ", cmake cxx flags: ${CMAKE_CXX_FLAGS}")
 configure_file(onnxruntime_config.h.in ${CMAKE_CURRENT_BINARY_DIR}/onnxruntime_config.h)
 if (WIN32)
   configure_file(../requirements.txt.in ${CMAKE_CURRENT_BINARY_DIR}/Debug/requirements.txt)

cmake/onnxruntime_config.h.in

@@ -22,3 +22,4 @@
 #cmakedefine HAS_BITWISE_INSTEAD_OF_LOGICAL
 #cmakedefine HAS_REALLOCARRAY
 #cmakedefine ORT_VERSION "@ORT_VERSION@"
+#cmakedefine ORT_BUILD_INFO "@ORT_BUILD_INFO@"

include/onnxruntime/core/session/onnxruntime_c_api.h

@@ -98,8 +98,11 @@ extern "C" {
 #ifndef ORT_TSTR
 #ifdef _WIN32
 #define ORT_TSTR(X) L##X
+// When X is a macro, L##X is not defined. In this case, we need to use ORT_TSTR_ON_MACRO.
+#define ORT_TSTR_ON_MACRO(X) L"" X
 #else
 #define ORT_TSTR(X) X
+#define ORT_TSTR_ON_MACRO(X) X
 #endif
 #endif
 
@@ -626,7 +629,8 @@ struct OrtApiBase {
    *   older than the version created with this header file.
    */
   const OrtApi*(ORT_API_CALL* GetApi)(uint32_t version)NO_EXCEPTION;
-  const char*(ORT_API_CALL* GetVersionString)(void)NO_EXCEPTION;  ///< Returns a null terminated string of the version of the Onnxruntime library (eg: "1.8.1")
+  const ORTCHAR_T*(ORT_API_CALL* GetVersionString)(void)NO_EXCEPTION;    ///< Returns a null terminated string of the version of the Onnxruntime library (eg: "1.8.1")
+  const ORTCHAR_T*(ORT_API_CALL* GetBuildInfoString)(void)NO_EXCEPTION;  ///< Returns a null terminated string of the build info including git info and cxx flags
 };
 typedef struct OrtApiBase OrtApiBase;
 

include/onnxruntime/core/session/onnxruntime_cxx_api.h

@@ -125,7 +125,14 @@ inline const OrtApi& GetApi() noexcept { return *Global<void>::api_; }
 /// This function returns the onnxruntime version string
 /// </summary>
 /// <returns>version string major.minor.rev</returns>
-std::string GetVersionString();
+std::basic_string<ORTCHAR_T> GetVersionString();
+
+/// <summary>
+/// This function returns the onnxruntime build information: including git branch,
+/// git commit id, build type(Debug/Release/RelWithDebInfo) and cmake cpp flags.
+/// </summary>
+/// <returns>string</returns>
+std::basic_string<ORTCHAR_T> GetBuildInfoString();
 
 /// <summary>
 /// This is a C++ wrapper for OrtApi::GetAvailableProviders() and

include/onnxruntime/core/session/onnxruntime_cxx_inline.h

@@ -1987,8 +1987,13 @@ inline void CustomOpApi::ReleaseKernelInfo(_Frees_ptr_opt_ OrtKernelInfo* info_c
   api_.ReleaseKernelInfo(info_copy);
 }
 
-inline std::string GetVersionString() {
-  std::string result = OrtGetApiBase()->GetVersionString();
+inline std::basic_string<ORTCHAR_T> GetVersionString() {
+  std::basic_string<ORTCHAR_T> result = OrtGetApiBase()->GetVersionString();
+  return result;
+}
+
+inline std::basic_string<ORTCHAR_T> GetBuildInfoString() {
+  std::basic_string<ORTCHAR_T> result = OrtGetApiBase()->GetBuildInfoString();
   return result;
 }
 

java/src/main/native/ai_onnxruntime_OnnxRuntime.c

@@ -76,8 +76,13 @@ JNIEXPORT jobjectArray JNICALL Java_ai_onnxruntime_OnnxRuntime_getAvailableProvi
 JNIEXPORT jstring JNICALL Java_ai_onnxruntime_OnnxRuntime_initialiseVersion
   (JNIEnv * jniEnv, jclass clazz) {
   (void)clazz;  // required JNI parameter not needed by functions which don't access their host class.
-  const char* version = OrtGetApiBase()->GetVersionString();
+  const ORTCHAR_T* version = OrtGetApiBase()->GetVersionString();
   assert(version != NULL);
+#ifdef _WIN32
+  jsize len = (jsize)(wcslen(version));
+  jstring versionStr = (*jniEnv)->NewString(jniEnv, (const jchar*)version, len);
+#else
   jstring versionStr = (*jniEnv)->NewStringUTF(jniEnv, version);
+#endif
   return versionStr;
 }

onnxruntime/__init__.py

@@ -38,6 +38,7 @@ try:
     from onnxruntime.capi._pybind_state import enable_telemetry_events  # noqa: F401
     from onnxruntime.capi._pybind_state import get_all_providers  # noqa: F401
     from onnxruntime.capi._pybind_state import get_available_providers  # noqa: F401
+    from onnxruntime.capi._pybind_state import get_build_info  # noqa: F401
     from onnxruntime.capi._pybind_state import get_device  # noqa: F401
     from onnxruntime.capi._pybind_state import get_version_string  # noqa: F401
     from onnxruntime.capi._pybind_state import set_default_logger_severity  # noqa: F401

onnxruntime/core/session/onnxruntime_c_api.cc

@@ -2381,6 +2381,7 @@ ORT_API(const OrtTrainingApi*, OrtApis::GetTrainingApi, uint32_t version) {
 static constexpr OrtApiBase ort_api_base = {
     &OrtApis::GetApi,
     &OrtApis::GetVersionString,
+    &OrtApis::GetBuildInfoString,
 };
 
 /* Rules on how to add a new Ort API version
@@ -2763,8 +2764,12 @@ ORT_API(const OrtApi*, OrtApis::GetApi, uint32_t version) {
   return nullptr;  // Unsupported version
 }
 
-ORT_API(const char*, OrtApis::GetVersionString) {
-  return ORT_VERSION;
+ORT_API(const ORTCHAR_T*, OrtApis::GetVersionString) {
+  return ORT_TSTR_ON_MACRO(ORT_VERSION);
+}
+
+ORT_API(const ORTCHAR_T*, OrtApis::GetBuildInfoString) {
+  return ORT_TSTR_ON_MACRO(ORT_BUILD_INFO);
 }
 
 const OrtApiBase* ORT_API_CALL OrtGetApiBase(void) NO_EXCEPTION {

onnxruntime/core/session/ort_apis.h

@@ -4,7 +4,8 @@
 namespace OrtApis {
 
 ORT_API(const OrtApi*, GetApi, uint32_t version);
-ORT_API(const char*, GetVersionString);
+ORT_API(const ORTCHAR_T*, GetVersionString);
+ORT_API(const ORTCHAR_T*, GetBuildInfoString);
 
 ORT_API(void, ReleaseEnv, OrtEnv*);
 ORT_API(void, ReleaseStatus, _Frees_ptr_opt_ OrtStatus*);

onnxruntime/python/onnxruntime_pybind_module.cc

@@ -22,6 +22,7 @@ PYBIND11_MODULE(onnxruntime_pybind11_state, m) {
       "from highest to lowest.");
 
   m.def("get_version_string", []() -> std::string { return ORT_VERSION; });
+  m.def("get_build_info", []() -> std::string { return ORT_BUILD_INFO; });
 }
 }  // namespace python
 }  // namespace onnxruntime

onnxruntime/test/onnx/main.cc

@@ -29,6 +29,7 @@ using namespace onnxruntime;
 
 namespace {
 void usage() {
+  auto version_string = ToUTF8String(OrtGetApiBase()->GetVersionString());
   printf(
       "onnx_test_runner [options...] <data_root>\n"
       "Options:\n"
@@ -64,7 +65,7 @@ void usage() {
       "\t-h: help\n"
       "\n"
       "onnxruntime version: %s\n",
-      OrtGetApiBase()->GetVersionString());
+      version_string.c_str());
 }
 
 static TestTolerances LoadTestTolerances(bool enable_cuda, bool enable_openvino, bool useCustom, double atol, double rtol) {

onnxruntime/test/onnx/microbenchmark/activation.cc

@@ -97,7 +97,7 @@ class MyIExecutionFrame : public IExecutionFrame {
   Status CreateNodeOutputMLValueImpl(OrtValue& /*ort_value*/, int /*ort_value_idx*/, const TensorShape* /*shape*/) override {
     abort();
   }
-  AllocatorPtr GetAllocatorImpl(const OrtMemoryInfo& info) const {
+  AllocatorPtr GetAllocatorImpl(const OrtMemoryInfo& info) const override {
     return a_.GetAllocator(info.mem_type);
   }
 
@@ -126,7 +126,7 @@ class MyIExecutionFrame : public IExecutionFrame {
     return Status::OK();
   }
 
-  Status CopyTensor(const Tensor& /*src*/, Tensor& /*dest*/) const {
+  Status CopyTensor(const Tensor& /*src*/, Tensor& /*dest*/) const override {
     return Status::OK();
   }
 };

onnxruntime/test/onnx/microbenchmark/batchnorm.cc

@@ -2,14 +2,14 @@
 #include "core/util/thread_utils.h"
 #include <benchmark/benchmark.h>
 
-#if defined(__GNUC__)
+#if defined(__GNUC__) && !defined(__clang__)
 #pragma GCC diagnostic push
 #pragma GCC diagnostic ignored "-Wmaybe-uninitialized"
 #endif
 
 #include "core/common/eigen_common_wrapper.h"
 
-#if defined(__GNUC__)
+#if defined(__GNUC__) && !defined(__clang__)
 #pragma GCC diagnostic pop
 #endif
 

onnxruntime/test/onnx/microbenchmark/common.h

@@ -2,7 +2,9 @@
 
 #include <algorithm>
 #include <cstdlib>
+#ifdef _WIN32
 #include <malloc.h>
+#endif
 #include <new>
 #include <random>
 
@@ -46,4 +48,4 @@ T* GenerateArrayWithRandomValue(size_t batch_size, T low, T high) {
     data[i] = static_cast<T>(dist(gen));
   }
   return data;
-}
+}

onnxruntime/test/onnx/microbenchmark/eigen.cc

@@ -1,4 +1,4 @@
-#if defined(__GNUC__)
+#if defined(__GNUC__) && !defined(__clang__)
 #pragma GCC diagnostic push
 #if __GNUC__ >= 6
 #pragma GCC diagnostic ignored "-Wignored-attributes"
@@ -25,7 +25,7 @@
 #include <unsupported/Eigen/CXX11/ThreadPool>
 #include <unsupported/Eigen/CXX11/Tensor>
 #include <unsupported/Eigen/CXX11/src/Tensor/TensorDeviceThreadPool.h>
-#if defined(__GNUC__)
+#if defined(__GNUC__) && !defined(__clang__)
 #pragma GCC diagnostic pop
 #elif defined(_MSC_VER)
 #pragma warning(pop)

onnxruntime/test/onnx/microbenchmark/gelu.cc

@@ -137,10 +137,9 @@ static void BM_ScaledTanhParallelFor(benchmark::State& state) {
   std::unique_ptr<concurrency::ThreadPool> tp(
       concurrency::CreateThreadPool(&onnxruntime::Env::Default(), tpo, concurrency::ThreadPoolType::INTRA_OP));
   const float alpha_ = 0.3f;
-  const float beta_ = 0.6f;
   for (auto _ : state) {
     ThreadPool::TryParallelFor(tp.get(), batch_size, cost,
-                               [alpha_, beta_, data, output](ptrdiff_t first, ptrdiff_t last) {
+                               [alpha_, data, output](ptrdiff_t first, ptrdiff_t last) {
                                  ptrdiff_t len = last - first;
                                  float* output_ptr = output + first;
                                  onnxruntime::ConstEigenVectorArrayMap<float> xm(data + first, len);

onnxruntime/test/onnx/microbenchmark/tptest.cc

@@ -30,7 +30,7 @@ BENCHMARK(BM_CreateThreadPool)
 // On Xeon W-2123 CPU, it takes about 2ns for each iteration
 #ifdef _WIN32
 #pragma optimize("", off)
-#else
+#elif defined(__GNUC__) && !defined(__clang__)
 #pragma GCC push_options
 #pragma GCC optimize("O0")
 #endif
@@ -42,7 +42,7 @@ void SimpleForLoop(ptrdiff_t first, ptrdiff_t last) {
 }
 #ifdef _WIN32
 #pragma optimize("", on)
-#else
+#elif defined(__GNUC__) && !defined(__clang__)
 #pragma GCC pop_options
 #endif
 

onnxruntime/test/python/onnxruntime_test_python.py

@@ -69,6 +69,10 @@ class TestInferenceSession(unittest.TestCase):
     def testGetVersionString(self):  # noqa: N802
         self.assertIsNot(onnxrt.get_version_string(), None)
 
+    def testGetBuildInfo(self):  # noqa: N802
+        self.assertIsNot(onnxrt.get_build_info(), None)
+        self.assertIn("Build Info", onnxrt.get_build_info())
+
     def testModelSerialization(self):  # noqa: N802
         try:
             so = onnxrt.SessionOptions()

onnxruntime/test/shared_lib/test_inference.cc

@@ -2208,9 +2208,9 @@ TEST(CApiTest, get_available_providers_cpp) {
 }
 
 TEST(CApiTest, get_version_string_cpp) {
-  std::string version_string = Ort::GetVersionString();
+  std::basic_string<ORTCHAR_T> version_string = Ort::GetVersionString();
   ASSERT_FALSE(version_string.empty());
-  ASSERT_EQ(version_string, ORT_VERSION);
+  ASSERT_EQ(version_string, std::basic_string<ORTCHAR_T>(ORT_TSTR_ON_MACRO(ORT_VERSION)));
 }
 
 TEST(CApiTest, TestSharedAllocators) {

orttraining/orttraining/python/orttraining_python_module.cc

@@ -360,6 +360,8 @@ PYBIND11_MODULE(onnxruntime_pybind11_state, m) {
 
   m.def("get_version_string", []() -> std::string { return ORT_VERSION; });
 
+  m.def("get_build_info", []() -> std::string { return ORT_BUILD_INFO; });
+
   m.def(
       "clear_training_ep_instances", []() -> void {
         GetTrainingEnv().ClearExecutionProviderInstances();