Merge remote-tracking branch 'origin/main' into fs-eire/string-template-prototype

Author: fs-eire

cmake/onnxruntime.cmake

@@ -22,13 +22,14 @@ endif()
 function(get_c_cxx_api_headers HEADERS_VAR)
   set(_headers
     "${REPO_ROOT}/include/onnxruntime/core/session/onnxruntime_c_api.h"
-    "${REPO_ROOT}/include/onnxruntime/core/session/onnxruntime_ep_c_api.h"
     "${REPO_ROOT}/include/onnxruntime/core/session/onnxruntime_cxx_api.h"
     "${REPO_ROOT}/include/onnxruntime/core/session/onnxruntime_cxx_inline.h"
+    "${REPO_ROOT}/include/onnxruntime/core/session/onnxruntime_ep_c_api.h"
+    "${REPO_ROOT}/include/onnxruntime/core/session/onnxruntime_ep_device_ep_metadata_keys.h"
     "${REPO_ROOT}/include/onnxruntime/core/session/onnxruntime_float16.h"
+    "${REPO_ROOT}/include/onnxruntime/core/session/onnxruntime_lite_custom_op.h"
     "${REPO_ROOT}/include/onnxruntime/core/session/onnxruntime_run_options_config_keys.h"
     "${REPO_ROOT}/include/onnxruntime/core/session/onnxruntime_session_options_config_keys.h"
-    "${REPO_ROOT}/include/onnxruntime/core/session/onnxruntime_lite_custom_op.h"
   )
 
   if (onnxruntime_ENABLE_TRAINING_APIS)

cmake/onnxruntime_providers_openvino.cmake

@@ -30,11 +30,6 @@
   endif()
 
   list(APPEND OPENVINO_LIB_LIST openvino::frontend::onnx openvino::runtime ${PYTHON_LIBRARIES})
-  if ((DEFINED ENV{OPENCL_LIBS}) AND (DEFINED ENV{OPENCL_INCS}) AND onnxruntime_USE_OPENVINO_GPU)
-    add_definitions(-DIO_BUFFER_ENABLED=1)
-    list(APPEND OPENVINO_LIB_LIST $ENV{OPENCL_LIBS})
-  endif()
-
   source_group(TREE ${ONNXRUNTIME_ROOT}/core FILES ${onnxruntime_providers_openvino_cc_srcs})
   onnxruntime_add_shared_library_module(onnxruntime_providers_openvino ${onnxruntime_providers_openvino_cc_srcs} "${ONNXRUNTIME_ROOT}/core/dll/onnxruntime.rc")
 

cmake/onnxruntime_unittests.cmake

@@ -767,6 +767,13 @@ if(onnxruntime_USE_AZURE)
   list(APPEND onnxruntime_test_providers_libs onnxruntime_providers_azure)
 endif()
 
+if (onnxruntime_USE_OPENVINO)
+  list(APPEND onnxruntime_test_framework_src_patterns ${TEST_SRC_DIR}/providers/openvino/*)
+  list(APPEND onnxruntime_test_framework_libs onnxruntime_providers_openvino)
+  list(APPEND onnxruntime_test_providers_dependencies onnxruntime_providers_openvino)
+  list(APPEND onnxruntime_test_providers_dependencies onnxruntime_providers_shared)
+endif()
+
 file(GLOB onnxruntime_test_framework_src CONFIGURE_DEPENDS
   ${onnxruntime_test_framework_src_patterns}
   )
@@ -1111,7 +1118,7 @@ if (NOT IOS)
 
     target_link_libraries(onnx_test_runner PRIVATE onnx_test_runner_common ${GETOPT_LIB_WIDE} ${onnx_test_libs} nlohmann_json::nlohmann_json)
     target_include_directories(onnx_test_runner PRIVATE ${ONNXRUNTIME_ROOT})
-    
+
     if (onnxruntime_ENABLE_TRAINING_TORCH_INTEROP)
       target_link_libraries(onnx_test_runner PRIVATE Python::Python)
     endif()
@@ -1232,7 +1239,7 @@ if (NOT onnxruntime_ENABLE_TRAINING_TORCH_INTEROP)
     target_include_directories(onnxruntime_perf_test PRIVATE   ${onnx_test_runner_src_dir} ${ONNXRUNTIME_ROOT}
           ${onnxruntime_graph_header} ${onnxruntime_exec_src_dir}
           ${CMAKE_CURRENT_BINARY_DIR})
-    
+
     if (WIN32)
       target_compile_options(onnxruntime_perf_test PRIVATE ${disabled_warnings})
       if (NOT DEFINED SYS_PATH_LIB)
@@ -1338,7 +1345,7 @@ endif()
     if (onnxruntime_USE_CUDA)
       list(APPEND onnxruntime_shared_lib_test_LIBS)
     endif()
-    
+
     if (onnxruntime_USE_TENSORRT)
       list(APPEND onnxruntime_shared_lib_test_LIBS ${TENSORRT_LIBRARY_INFER})
     endif()
@@ -1372,7 +1379,7 @@ endif()
     if (onnxruntime_USE_NV)
       target_include_directories(onnxruntime_shared_lib_test PRIVATE ${CUDAToolkit_INCLUDE_DIRS})
     endif()
-    
+
 
     if (CMAKE_SYSTEM_NAME STREQUAL "Android")
       target_sources(onnxruntime_shared_lib_test PRIVATE
@@ -1429,7 +1436,7 @@ endif()
       DEPENDS ${all_dependencies}
     )
 
-    
+
 
     target_compile_definitions(onnxruntime_test_debug_node_inputs_outputs
       PRIVATE DEBUG_NODE_INPUTS_OUTPUTS)
@@ -1983,6 +1990,11 @@ if (onnxruntime_BUILD_SHARED_LIB AND NOT CMAKE_SYSTEM_NAME STREQUAL "Emscripten"
           LIBS ${onnxruntime_ep_graph_test_LIBS}
           DEPENDS ${all_dependencies}
   )
+  if (UNIX AND (onnxruntime_USE_TENSORRT OR onnxruntime_USE_NV))
+    # The test_main.cc includes NvInfer.h where it has many deprecated declarations
+    # simply ignore them for TensorRT EP build
+    set_property(TARGET onnxruntime_ep_graph_test APPEND_STRING PROPERTY COMPILE_FLAGS "-Wno-deprecated-declarations")
+  endif()
 endif()
 
 include(onnxruntime_fuzz_test.cmake)

include/onnxruntime/core/framework/allocator.h

@@ -86,7 +86,7 @@ class Stream;
 namespace synchronize {
 class Notification;
 }
-using WaitNotificationFn = std::function<void(Stream&, synchronize::Notification&)>;
+using WaitNotificationFn = std::function<void(Stream*, synchronize::Notification&)>;
 void* AllocateBufferWithOptions(IAllocator& allocator, size_t size, bool use_reserve, Stream* stream, WaitNotificationFn wait_fn);
 
 template <typename T>

include/onnxruntime/core/framework/stream_handles.h

@@ -172,12 +172,16 @@ class IStreamCommandHandleRegistry {
   // i.e., for an cuda event what notify the memory copy, it could be wait on a CPU stream, or on another cuda stream.
   [[nodiscard]] virtual WaitNotificationFn GetWaitHandle(const OrtDevice& notification_owner_device,
                                                          const OrtDevice& executor_device) const = 0;
-  // Get the stream creation function registered on the given device type.
+
+  // Get the stream creation function registered for the given device type.
   [[nodiscard]] virtual CreateStreamFn GetCreateStreamFn(OrtDevice::DeviceType execution_device_type) const = 0;
-  // register a wait methond which will be invoked when we wait a notification (created by 'notification_device_type' device) on a stream at 'device_type' device.
+
+  // register a wait method which will be invoked to await a notification that is
+  // created by 'notification_device_type' device on a stream at 'device_type' device.
   virtual void RegisterWaitFn(OrtDevice::DeviceType notification_device_type,
                               OrtDevice::DeviceType device_type,
                               WaitNotificationFn fn) = 0;
+
   // register a handle about how to create stream on given device type.
   virtual void RegisterCreateStreamFn(OrtDevice::DeviceType device_type, CreateStreamFn f) = 0;