Merge branch 'master' into sync_ort_main_19_7_25

Author: ankitm3k

.github/workflows/windows_openvino.yml

@@ -54,12 +54,12 @@ jobs:
             core.exportVariable('ACTIONS_CACHE_URL', process.env.ACTIONS_CACHE_URL || '');
             core.exportVariable('ACTIONS_RUNTIME_TOKEN', process.env.ACTIONS_RUNTIME_TOKEN || '');
 
-      - name: Download OpenVINO Toolkit v2025.0.0
+      - name: Download OpenVINO Toolkit v2025.2.0
         env:
-          OpenVINOVersion: 2025.0.0
+          OpenVINOVersion: 2025.2.0
         shell: pwsh
         run: |
-          $Url = "https://storage.openvinotoolkit.org/repositories/openvino/packages/2025.0/windows/openvino_toolkit_windows_2025.0.0.17942.1f68be9f594_x86_64.zip"
+          $Url ="https://storage.openvinotoolkit.org/repositories/openvino/packages/2025.2/windows/openvino_toolkit_windows_2025.2.0.19140.c01cd93e24d_x86_64.zip"
           $OutputPath = "$env:RUNNER_TEMP\openvino.zip"
           $ExtractPath = "$env:RUNNER_TEMP\openvino-v$env:OpenVINOVersion"
           $TempExtractPath = "$env:RUNNER_TEMP\openvino_temp"
@@ -102,7 +102,7 @@ jobs:
         shell: pwsh
         # Use $GITHUB_ENV to set the variable for subsequent steps
         run: |
-          $openVinoRootDir = Join-Path $env:RUNNER_TEMP "openvino-v2025.0.0"
+          $openVinoRootDir = Join-Path $env:RUNNER_TEMP "openvino-v2025.2.0"
           echo "OpenVINORootDir=$openVinoRootDir" >> $env:GITHUB_ENV
 
       - name: Print OpenVINORootDir after downloading OpenVINO

cmake/onnxruntime_unittests.cmake

@@ -498,17 +498,19 @@ set (ONNXRUNTIME_AUTOEP_TEST_SRC_DIR "${TEST_SRC_DIR}/autoep")
 set (ONNXRUNTIME_EP_GRAPH_TEST_SRC_DIR "${TEST_SRC_DIR}/ep_graph")
 
 set (onnxruntime_shared_lib_test_SRC
-          ${ONNXRUNTIME_SHARED_LIB_TEST_SRC_DIR}/test_fixture.h
-          ${ONNXRUNTIME_SHARED_LIB_TEST_SRC_DIR}/test_session_options.cc
-          ${ONNXRUNTIME_SHARED_LIB_TEST_SRC_DIR}/test_run_options.cc
+          ${ONNXRUNTIME_SHARED_LIB_TEST_SRC_DIR}/custom_op_utils.h
+          ${ONNXRUNTIME_SHARED_LIB_TEST_SRC_DIR}/custom_op_utils.cc
           ${ONNXRUNTIME_SHARED_LIB_TEST_SRC_DIR}/test_allocator.cc
-          ${ONNXRUNTIME_SHARED_LIB_TEST_SRC_DIR}/test_nontensor_types.cc
+          ${ONNXRUNTIME_SHARED_LIB_TEST_SRC_DIR}/test_data_copy.cc
+          ${ONNXRUNTIME_SHARED_LIB_TEST_SRC_DIR}/test_fixture.h
           ${ONNXRUNTIME_SHARED_LIB_TEST_SRC_DIR}/test_model_loading.cc
+          ${ONNXRUNTIME_SHARED_LIB_TEST_SRC_DIR}/test_nontensor_types.cc
           ${ONNXRUNTIME_SHARED_LIB_TEST_SRC_DIR}/test_ort_format_models.cc
+          ${ONNXRUNTIME_SHARED_LIB_TEST_SRC_DIR}/test_run_options.cc
+          ${ONNXRUNTIME_SHARED_LIB_TEST_SRC_DIR}/test_session_options.cc
           ${ONNXRUNTIME_SHARED_LIB_TEST_SRC_DIR}/utils.h
           ${ONNXRUNTIME_SHARED_LIB_TEST_SRC_DIR}/utils.cc
-          ${ONNXRUNTIME_SHARED_LIB_TEST_SRC_DIR}/custom_op_utils.h
-          ${ONNXRUNTIME_SHARED_LIB_TEST_SRC_DIR}/custom_op_utils.cc)
+          )
 
 if (NOT onnxruntime_MINIMAL_BUILD)
   list(APPEND onnxruntime_shared_lib_test_SRC ${ONNXRUNTIME_SHARED_LIB_TEST_SRC_DIR}/test_inference.cc)
@@ -722,6 +724,7 @@ endif()
 if(onnxruntime_USE_QNN AND NOT onnxruntime_MINIMAL_BUILD AND NOT onnxruntime_REDUCED_OPS_BUILD)
   list(APPEND onnxruntime_test_framework_src_patterns ${TEST_SRC_DIR}/providers/qnn/*)
   list(APPEND onnxruntime_test_framework_src_patterns ${TEST_SRC_DIR}/providers/qnn/qnn_node_group/*)
+  list(APPEND onnxruntime_test_framework_src_patterns ${TEST_SRC_DIR}/providers/qnn/optimizer/*)
   list(APPEND onnxruntime_test_framework_libs onnxruntime_providers_qnn)
   list(APPEND onnxruntime_test_providers_dependencies onnxruntime_providers_qnn)
   if(NOT onnxruntime_BUILD_QNN_EP_STATIC_LIB)

include/onnxruntime/core/providers/tensorrt/tensorrt_provider_options.h

@@ -89,7 +89,12 @@ struct OrtTensorRTProviderOptionsV2 {
   size_t trt_onnx_bytestream_size{0};               // size of the byte stream provided as "trt_onnx_bytestream"
                                                     // can be updated using: UpdateTensorRTProviderOptionsWithValue
 
-  const char* trt_engine_cache_prefix{nullptr};  // specify engine cache prefix
-  int trt_engine_hw_compatible{0};               // Enable hardware compatibility. Default 0 = false, nonzero = true
-  const char* trt_op_types_to_exclude{};         // Exclude specific ops from running on TRT.
+  const void* trt_external_data_bytestream{nullptr};  // The byte stream containing the weights to override the ones provided in the ONNX model.
+                                                      // can be updated using: UpdateTensorRTProviderOptionsWithValue
+  size_t trt_external_data_bytestream_size{0};        // size of the byte stream provided as "trt_external_data_bytestream"
+                                                      // can be updated using: UpdateTensorRTProviderOptionsWithValue
+  const char* trt_engine_cache_prefix{nullptr};       // specify engine cache prefix
+  int trt_engine_hw_compatible{0};                    // Enable hardware compatibility. Default 0 = false, nonzero = true
+  const char* trt_op_types_to_exclude{};              // Exclude specific ops from running on TRT.
+  int trt_load_user_initializer{0};                   // Save initializers locally instead of to disk. Default 0 = false, nonzero = true
 };

include/onnxruntime/core/providers/utils/ort_graph_to_proto.h

@@ -665,11 +665,11 @@ static Ort::Status OrtOpAttrToProto(const OrtOpAttr& ort_attr, onnx::AttributePr
       Ort::Status status{ort_api.ReadOpAttr(&ort_attr, attr_type, nullptr, 0, &total_attr_bytes)};
       std::string* str = attr_proto.mutable_s();
 
-      str->resize(total_attr_bytes, '\0');
+      str->resize(total_attr_bytes);
       ORT_EP_UTILS_C_RETURN_IF_ERROR(ort_api.ReadOpAttr(&ort_attr, attr_type, str->data(), total_attr_bytes,
                                                         &total_attr_bytes));
 
-      str->resize(total_attr_bytes - 1);  // remove extra ending terminating '\0' character.
+      str->resize(total_attr_bytes);
       break;
     }
     case OrtOpAttrType::ORT_OP_ATTR_STRINGS: {

include/onnxruntime/core/session/environment.h

@@ -15,6 +15,7 @@
 #include "core/common/status.h"
 #include "core/framework/allocator.h"
 #include "core/framework/execution_provider.h"
+#include "core/framework/data_transfer_manager.h"
 #include "core/platform/device_discovery.h"
 #include "core/platform/threadpool.h"
 
@@ -140,6 +141,10 @@ class Environment {
                                OrtDeviceMemoryType mem_type, OrtAllocatorType allocator_type,
                                const OrtKeyValuePairs* allocator_options, OrtAllocator** allocator);
   Status ReleaseSharedAllocator(const OrtEpDevice& ep_device, OrtDeviceMemoryType mem_type);
+
+  const DataTransferManager& GetDataTransferManager() const {
+    return data_transfer_mgr_;
+  }
 #endif  // !defined(ORT_MINIMAL_BUILD)
 
   // return a shared allocator from a plugin EP or custom allocator added with RegisterAllocator
@@ -185,6 +190,23 @@ class Environment {
 
   using OrtAllocatorUniquePtr = std::unique_ptr<OrtAllocator, std::function<void(OrtAllocator*)>>;
 
+  // if the user calls CreateSharedAllocator and wraps the plugin EP's allocator with an arena we end up with
+  // OrtAllocator from EP -> wrapped in IAllocatorImplWrappingOrtAllocator -> inside a BFCArena IAllocator.
+  // we can put that in shared_allocators_ for sessions to use, but to have an OrtAllocator available in
+  // shared_ort_allocators_ that can be used outside of a session we need to additionally wrap that in an
+  // OrtAllocatorImplWrappingIAllocator. way too many levels of indirection but that is what it is currently.
+  // we need something to own that final OrtAllocator, so we add it to arena_ort_allocators_.
+  //
+  // TODO: we could split out the BFCArena implementation so it can be plugged into either an IAllocator
+  // or an OrtAllocator instance to reduce the indirection a little.
+  // with that we get an OrtAllocator from the EP, wrap it with an OrtAllocator based BFCArena, and wrap that with the
+  // IAllocatorImplWrappingOrtAllocator which takes ownership of the OrtAllocator and is in shared_allocators_.
+  //
+  // Alternatively we can disable wrapping an EP's allocator with a BFCArena and say the EP should provide the arena
+  // implementation directly. They're free to copy BFCArena as it came from TF originally. Or we could provide a
+  // cut-and-paste BFCArena implementation that works using the EP API that can be included in the EP source.
+  std::unordered_map<const OrtMemoryInfo*, std::unique_ptr<OrtAllocatorImplWrappingIAllocator>> arena_ort_allocators_;
+
 #if !defined(ORT_MINIMAL_BUILD)
   // register EPs that are built into the ORT binary so they can take part in AutoEP selection
   // added to ep_libraries
@@ -207,7 +229,9 @@ class Environment {
 
     std::unique_ptr<EpLibrary> library;
     std::vector<std::unique_ptr<OrtEpDevice>> execution_devices;
-    std::vector<EpFactoryInternal*> internal_factories;  // factories that can create IExecutionProvider instances
+    std::vector<OrtEpFactory*> factories;
+    std::vector<EpFactoryInternal*> internal_factories;    // factories that can create IExecutionProvider instances
+    std::vector<plugin_ep::DataTransfer*> data_transfers;  // data transfer instances for this EP.
 
    private:
     EpInfo() = default;
@@ -223,6 +247,9 @@ class Environment {
 
   // lookup set for internal EPs so we can create an IExecutionProvider directly
   std::unordered_set<EpFactoryInternal*> internal_ep_factories_;
+
+  DataTransferManager data_transfer_mgr_;  // plugin EP IDataTransfer instances
+
 #endif  // !defined(ORT_MINIMAL_BUILD)
 };