intel · ankitm3k · Aug 7, 2025 · Aug 5, 2025 · Aug 5, 2025 · Aug 6, 2025
diff --git a/.github/workflows/windows_webgpu.yml b/.github/workflows/windows_webgpu.yml
@@ -128,7 +128,7 @@ jobs:
             ${{ matrix.vcpkg_option == 'vcpkg' && '--use_vcpkg' || '' }} `
             --cmake_extra_defines `
             onnxruntime_BUILD_UNIT_TESTS=ON `
-            onnxruntime_BUILD_DAWN_MONOLITHIC_LIBRARY=ON
+            onnxruntime_BUILD_DAWN_SHARED_LIBRARY=ON
           if ($lastExitCode -ne 0) {
             exit $lastExitCode
           }

diff --git a/cmake/CMakeLists.txt b/cmake/CMakeLists.txt
@@ -140,7 +140,7 @@ option(onnxruntime_USE_WEBGPU "Build with WebGPU support. Enable WebGPU via C/C+
 option(onnxruntime_WGSL_TEMPLATE "Specify the code generator for WGSL template. Default is static." "static")
 option(onnxruntime_USE_EXTERNAL_DAWN "Build with treating Dawn as external dependency. Will not link Dawn at build time." OFF)
 option(onnxruntime_CUSTOM_DAWN_SRC_PATH "Path to custom Dawn src dir.")
-option(onnxruntime_BUILD_DAWN_MONOLITHIC_LIBRARY "Build Dawn as a monolithic library" OFF)
+option(onnxruntime_BUILD_DAWN_SHARED_LIBRARY "Build Dawn as a shared library" OFF)
 option(onnxruntime_ENABLE_PIX_FOR_WEBGPU_EP "Adding frame present for PIX to capture a frame" OFF)
 # The following 2 options are only for Windows
 option(onnxruntime_ENABLE_DAWN_BACKEND_VULKAN "Enable Vulkan backend for Dawn (on Windows)" OFF)
@@ -899,8 +899,8 @@ if (onnxruntime_USE_WEBGPU)
   #
   # if (onnxruntime_USE_VCPKG AND NOT CMAKE_SYSTEM_NAME STREQUAL "Emscripten")
   if (FALSE)
-    if (NOT onnxruntime_BUILD_DAWN_MONOLITHIC_LIBRARY)
-      message(FATAL_ERROR "onnxruntime_USE_VCPKG is not supported with onnxruntime_BUILD_DAWN_MONOLITHIC_LIBRARY=OFF")
+    if (onnxruntime_BUILD_DAWN_SHARED_LIBRARY)
+      message(FATAL_ERROR "onnxruntime_USE_VCPKG is not supported with onnxruntime_BUILD_DAWN_SHARED_LIBRARY=ON")
     endif()
     if (onnxruntime_USE_EXTERNAL_DAWN)
       message(FATAL_ERROR "onnxruntime_USE_VCPKG is not supported with onnxruntime_USE_EXTERNAL_DAWN=ON")
@@ -921,8 +921,8 @@ if (onnxruntime_USE_WEBGPU)
     endif()
   endif()
 
-  if (onnxruntime_BUILD_DAWN_MONOLITHIC_LIBRARY)
-    list(APPEND ORT_PROVIDER_FLAGS -DBUILD_DAWN_MONOLITHIC_LIBRARY=1)
+  if (onnxruntime_BUILD_DAWN_SHARED_LIBRARY)
+    list(APPEND ORT_PROVIDER_FLAGS -DBUILD_DAWN_SHARED_LIBRARY=1)
   endif()
   if (onnxruntime_USE_EXTERNAL_DAWN)
     list(APPEND ORT_PROVIDER_FLAGS -DUSE_EXTERNAL_DAWN=1)

diff --git a/cmake/deps.txt b/cmake/deps.txt
@@ -55,6 +55,6 @@ cutlass;https://github.com/NVIDIA/cutlass/archive/refs/tags/v3.9.2.zip;b7f8dc4a8
 extensions;https://github.com/microsoft/onnxruntime-extensions/archive/c24b7bab0c12f53da76d0c31b03b9f0f8ec8f3b4.zip;239063aee4946a9af147b473a4c3da78ba7413b4
 directx_headers;https://github.com/microsoft/DirectX-Headers/archive/refs/tags/v1.613.1.zip;47653509a3371eabb156360f42faf582f314bf2e
 cudnn_frontend;https://github.com/NVIDIA/cudnn-frontend/archive/refs/tags/v1.12.0.zip;7e733cfdc410d777b76122d64232499205589a96
-dawn;https://github.com/google/dawn/archive/9733be39e18186961d503e064874afe3e9ceb8d1.zip;2a4017c32892b90d072a9102eba90ae691fae36d
+dawn;https://github.com/google/dawn/archive/794b6fadc4171f7b853a77ffdf0948fbec431f41.zip;77bb02deace0d140411f02a2fb8f5f925ea6a1b6
 kleidiai;https://github.com/ARM-software/kleidiai/archive/refs/tags/v1.9.0.tar.gz;a2765979f64efb173a4b8ba4de39dcba9c655786
 duktape;https://github.com/svaarala/duktape/releases/download/v2.7.0/duktape-2.7.0.tar.xz;8200c8e417dbab7adcc12c4dbdef7651cfc55794
diff --git a/cmake/external/abseil-cpp.cmake b/cmake/external/abseil-cpp.cmake
@@ -43,9 +43,6 @@ onnxruntime_fetchcontent_makeavailable(abseil_cpp)
 FetchContent_GetProperties(abseil_cpp)
 if(abseil_cpp_SOURCE_DIR)
   set(ABSEIL_SOURCE_DIR ${abseil_cpp_SOURCE_DIR})
-  if(onnxruntime_USE_WEBGPU)
-    set(DAWN_ABSEIL_DIR ${abseil_cpp_SOURCE_DIR})
-  endif()
 endif()
 
 # abseil_cpp_SOURCE_DIR is non-empty if we build it from source

diff --git a/cmake/external/onnxruntime_external_deps.cmake b/cmake/external/onnxruntime_external_deps.cmake
@@ -224,11 +224,6 @@ onnxruntime_fetchcontent_makeavailable(Protobuf)
 if(Protobuf_FOUND)
   message(STATUS "Using protobuf from find_package(or vcpkg). Protobuf version: ${Protobuf_VERSION}")
 else()
-  if(protobuf_SOURCE_DIR)
-    if(onnxruntime_USE_WEBGPU)
-      set(DAWN_PROTOBUF_DIR ${protobuf_SOURCE_DIR})
-    endif()
-  endif()
   # Adjust warning flags
   if (TARGET libprotoc)
     if (NOT MSVC)
@@ -645,19 +640,28 @@ if (onnxruntime_USE_WEBGPU)
     set(DAWN_BUILD_SAMPLES OFF CACHE BOOL "" FORCE)
     set(DAWN_ENABLE_NULL OFF CACHE BOOL "" FORCE)
     set(DAWN_FETCH_DEPENDENCIES ON CACHE BOOL "" FORCE)
+    set(DAWN_BUILD_PROTOBUF OFF CACHE BOOL "" FORCE)
     set(DAWN_BUILD_TESTS OFF CACHE BOOL "" FORCE)
     if (NOT CMAKE_SYSTEM_NAME STREQUAL "Emscripten")
-      if (onnxruntime_BUILD_DAWN_MONOLITHIC_LIBRARY)
-        set(DAWN_BUILD_MONOLITHIC_LIBRARY ON CACHE BOOL "" FORCE)
+      if (onnxruntime_BUILD_DAWN_SHARED_LIBRARY)
+        set(DAWN_BUILD_MONOLITHIC_LIBRARY SHARED CACHE BOOL "" FORCE)
         set(DAWN_ENABLE_INSTALL ON CACHE BOOL "" FORCE)
 
         if (onnxruntime_USE_EXTERNAL_DAWN)
-          message(FATAL_ERROR "onnxruntime_USE_EXTERNAL_DAWN and onnxruntime_BUILD_DAWN_MONOLITHIC_LIBRARY cannot be enabled at the same time.")
+          message(FATAL_ERROR "onnxruntime_USE_EXTERNAL_DAWN and onnxruntime_BUILD_DAWN_SHARED_LIBRARY cannot be enabled at the same time.")
         endif()
       else()
         # use dawn::dawn_native and dawn::dawn_proc instead of the monolithic dawn::webgpu_dawn to minimize binary size
         set(DAWN_BUILD_MONOLITHIC_LIBRARY OFF CACHE BOOL "" FORCE)
         set(DAWN_ENABLE_INSTALL OFF CACHE BOOL "" FORCE)
+
+        # use the same protobuf/abseil for ORT and Dawn when static linking
+        if(abseil_cpp_SOURCE_DIR)
+          set(DAWN_ABSEIL_DIR ${abseil_cpp_SOURCE_DIR})
+        endif()
+        if(protobuf_SOURCE_DIR)
+          set(DAWN_PROTOBUF_DIR ${protobuf_SOURCE_DIR})
+        endif()
       endif()
 
       if (onnxruntime_ENABLE_PIX_FOR_WEBGPU_EP)
@@ -714,6 +718,7 @@ if (onnxruntime_USE_WEBGPU)
         set(DAWN_ENABLE_D3D11 OFF CACHE BOOL "" FORCE)
       endif()
     endif()
+
     if (onnxruntime_CUSTOM_DAWN_SRC_PATH)
       # use the custom dawn source path if provided
       #
@@ -766,7 +771,7 @@ if (onnxruntime_USE_WEBGPU)
   endif()
 
   if (NOT CMAKE_SYSTEM_NAME STREQUAL "Emscripten")
-    if (onnxruntime_BUILD_DAWN_MONOLITHIC_LIBRARY)
+    if (onnxruntime_BUILD_DAWN_SHARED_LIBRARY)
       list(APPEND onnxruntime_EXTERNAL_LIBRARIES dawn::webgpu_dawn)
     else()
       if (NOT onnxruntime_USE_EXTERNAL_DAWN)

diff --git a/cmake/onnxruntime_java.cmake b/cmake/onnxruntime_java.cmake
@@ -197,7 +197,7 @@ if (WIN32)
           )
         endif()
       endif()
-      if (onnxruntime_BUILD_DAWN_MONOLITHIC_LIBRARY)
+      if (onnxruntime_BUILD_DAWN_SHARED_LIBRARY)
         add_custom_command(TARGET onnxruntime4j_jni POST_BUILD COMMAND ${CMAKE_COMMAND} -E copy_if_different $<TARGET_FILE:dawn::webgpu_dawn> ${JAVA_PACKAGE_LIB_DIR}/$<TARGET_FILE_NAME:dawn::webgpu_dawn>)
       endif()
     endif()
@@ -223,7 +223,7 @@ else()
   if (onnxruntime_USE_QNN AND NOT onnxruntime_BUILD_QNN_EP_STATIC_LIB)
     add_custom_command(TARGET onnxruntime4j_jni POST_BUILD COMMAND ${CMAKE_COMMAND} -E copy_if_different $<TARGET_FILE:onnxruntime_providers_qnn> ${JAVA_PACKAGE_LIB_DIR}/$<TARGET_LINKER_FILE_NAME:onnxruntime_providers_qnn>)
   endif()
-  if (onnxruntime_USE_WEBGPU AND onnxruntime_BUILD_DAWN_MONOLITHIC_LIBRARY)
+  if (onnxruntime_USE_WEBGPU AND onnxruntime_BUILD_DAWN_SHARED_LIBRARY)
     add_custom_command(TARGET onnxruntime4j_jni POST_BUILD COMMAND ${CMAKE_COMMAND} -E copy_if_different $<TARGET_FILE:dawn::webgpu_dawn> ${JAVA_PACKAGE_LIB_DIR}/$<TARGET_LINKER_FILE_NAME:dawn::webgpu_dawn>)
   endif()
 endif()

diff --git a/cmake/onnxruntime_nodejs.cmake b/cmake/onnxruntime_nodejs.cmake
@@ -65,7 +65,7 @@ if (onnxruntime_USE_WEBGPU)
             list(APPEND NODEJS_DLL_DEPS "$<TARGET_FILE_DIR:dxcompiler>/dxcompiler.dll")
         endif()
     endif()
-    if (onnxruntime_BUILD_DAWN_MONOLITHIC_LIBRARY)
+    if (onnxruntime_BUILD_DAWN_SHARED_LIBRARY)
         list(APPEND NODEJS_DLL_DEPS "$<TARGET_FILE:dawn::webgpu_dawn>")
     endif()
 endif()

diff --git a/cmake/onnxruntime_providers_webgpu.cmake b/cmake/onnxruntime_providers_webgpu.cmake
@@ -63,7 +63,7 @@
 
     set(onnxruntime_providers_webgpu_dll_deps)
 
-    if (onnxruntime_BUILD_DAWN_MONOLITHIC_LIBRARY)
+    if (onnxruntime_BUILD_DAWN_SHARED_LIBRARY)
       target_link_libraries(onnxruntime_providers_webgpu dawn::webgpu_dawn)
 
       if (WIN32)

diff --git a/cmake/onnxruntime_python.cmake b/cmake/onnxruntime_python.cmake
@@ -1100,7 +1100,7 @@ if (onnxruntime_USE_WEBGPU)
       )
     endif()
   endif()
-  if (onnxruntime_BUILD_DAWN_MONOLITHIC_LIBRARY)
+  if (onnxruntime_BUILD_DAWN_SHARED_LIBRARY)
     add_custom_command(
       TARGET onnxruntime_pybind11_state POST_BUILD
       COMMAND ${CMAKE_COMMAND} -E copy

diff --git a/cmake/onnxruntime_unittests.cmake b/cmake/onnxruntime_unittests.cmake
@@ -1258,7 +1258,7 @@ if (NOT onnxruntime_ENABLE_TRAINING_TORCH_INTEROP)
             onnx_test_runner_common onnxruntime_test_utils onnxruntime_common
             onnxruntime onnxruntime_flatbuffers onnx_test_data_proto
             ${onnxruntime_EXTERNAL_LIBRARIES}
-            ${GETOPT_LIB_WIDE} ${SYS_PATH_LIB} ${CMAKE_DL_LIBS})
+            absl::flags absl::flags_parse ${SYS_PATH_LIB} ${CMAKE_DL_LIBS})
       if(NOT WIN32)
         if(onnxruntime_USE_SNPE)
           list(APPEND onnxruntime_perf_test_libs onnxruntime_providers_snpe)
@@ -1278,7 +1278,7 @@ if (NOT onnxruntime_ENABLE_TRAINING_TORCH_INTEROP)
         target_link_libraries(onnxruntime_perf_test PRIVATE debug dbghelp advapi32)
       endif()
     else()
-      target_link_libraries(onnxruntime_perf_test PRIVATE onnx_test_runner_common ${GETOPT_LIB_WIDE} ${onnx_test_libs})
+      target_link_libraries(onnxruntime_perf_test PRIVATE onnx_test_runner_common absl::flags absl::flags_parse ${onnx_test_libs})
     endif()
     set_target_properties(onnxruntime_perf_test PROPERTIES FOLDER "ONNXRuntimeTest")
 

diff --git a/csharp/test/Microsoft.ML.OnnxRuntime.Tests.BrowserStack.Android/browserstack.yml b/csharp/test/Microsoft.ML.OnnxRuntime.Tests.BrowserStack.Android/browserstack.yml
@@ -3,11 +3,11 @@ platforms:
   - platformName: android
     deviceName: Samsung Galaxy S22 Ultra
     platformVersion: 12.0
-browserstackLocal: true
+browserstackLocal: false
 buildName: ORT android test
 buildIdentifier: ${BUILD_NUMBER}
 projectName: ORT-UITests
 debug: true
 networkLogs: false
 testContextOptions:
-    skipSessionStatus: true   
+    skipSessionStatus: true
diff --git a/include/onnxruntime/core/framework/execution_provider.h b/include/onnxruntime/core/framework/execution_provider.h
@@ -179,7 +179,12 @@ class IExecutionProvider {
   /**
      Get the device id of current execution provider
   */
-  virtual int GetDeviceId() const { return default_device_.Id(); };
+  virtual int GetDeviceId() const { return default_device_.Id(); }
+
+  /**
+   * Get the OrtDevice the execution provider was registered with.
+   */
+  const OrtDevice& GetDevice() const { return default_device_; }
 
   /**
      Get execution provider's configuration options.

diff --git a/include/onnxruntime/core/framework/ortdevice.h b/include/onnxruntime/core/framework/ortdevice.h
@@ -150,6 +150,13 @@ struct OrtDevice {
     return alignment < other.alignment;
   }
 
+  bool EqualIgnoringAlignment(const OrtDevice& other) const {
+    return device_type == other.device_type &&
+           memory_type == other.memory_type &&
+           vendor_id == other.vendor_id &&
+           device_id == other.device_id;
+  }
+
  private:
   // Device type.
   int32_t device_type : 8;

diff --git a/include/onnxruntime/core/session/environment.h b/include/onnxruntime/core/session/environment.h
@@ -106,6 +106,15 @@ class Environment {
     return shared_allocators_;
   }
 
+  /**
+   * Returns an AllocatorPtr for a shared IAllocator based allocator if it matches the memory info.
+   * The OrtMemoryInfo name and whether it's an arena or device allocator is ignored in the lookup, as is the
+   * alignment.
+   * The user calling this function is not expected to know the alignment, and we expect the allocator instance to be
+   * created with a valid alignment for the device.
+   */
+  AllocatorPtr GetRegisteredSharedAllocator(const OrtMemoryInfo& mem_info) const;
+
   /**
    * Removes registered allocator that was previously registered for sharing between multiple sessions.
    */
@@ -171,7 +180,7 @@ class Environment {
   std::unique_ptr<onnxruntime::concurrency::ThreadPool> inter_op_thread_pool_;
   bool create_global_thread_pools_{false};
 
-  std::mutex mutex_;
+  mutable std::mutex mutex_;
 
   // shared allocators from various sources.
   // CreateAndRegisterAllocator[V2]: IAllocator allocators created by ORT

diff --git a/onnxruntime/core/dll/delay_load_hook.cc b/onnxruntime/core/dll/delay_load_hook.cc
@@ -21,10 +21,10 @@
 // - https://learn.microsoft.com/en-us/windows/win32/dlls/dynamic-link-library-search-order#alternate-search-order-for-unpackaged-apps
 //
 // The DLL DelayLoad hook is only enabled when the compiler is MSVC and at least one of the following is True:
-// - both USE_WEBGPU and BUILD_DAWN_MONOLITHIC_LIBRARY are defined
+// - both USE_WEBGPU and BUILD_DAWN_SHARED_LIBRARY are defined
 // - USE_DML is defined
 //
-#if defined(USE_WEBGPU) && defined(BUILD_DAWN_MONOLITHIC_LIBRARY)
+#if defined(USE_WEBGPU) && defined(BUILD_DAWN_SHARED_LIBRARY)
 #define ORT_DELAY_LOAD_WEBGPU_DAWN_DLL 1
 #else
 #define ORT_DELAY_LOAD_WEBGPU_DAWN_DLL 0
@@ -45,7 +45,7 @@
 
 namespace {
 
-#define DEFINE_KNOWN_DLL(name) {#name ".dll", L#name L".dll"}
+#define DEFINE_KNOWN_DLL(name) {#name ".dll", L## #name L".dll"}
 
 constexpr struct {
   const char* str;

diff --git a/onnxruntime/core/providers/cpu/ml/tree_ensemble_attribute.h b/onnxruntime/core/providers/cpu/ml/tree_ensemble_attribute.h
@@ -389,8 +389,10 @@ struct TreeEnsembleAttributesV5 {
     int64_t curr_treeid = 0;
     for (const int64_t& tree_root : tree_roots) {
       size_t tree_root_size_t = onnxruntime::narrow<size_t>(tree_root);
+      bool is_leaf = (nodes_falsenodeids[tree_root_size_t] == nodes_truenodeids[tree_root_size_t] &&
+                      nodes_falseleafs[tree_root_size_t] && nodes_trueleafs[tree_root_size_t]);
       transformInputOneTree(tree_root_size_t, curr_treeid, 0,
-                            nodes_falsenodeids[tree_root_size_t] == nodes_truenodeids[tree_root_size_t],
+                            is_leaf,
                             membership_values_by_id, output);
       curr_treeid++;
     }

diff --git a/onnxruntime/core/providers/cuda/cuda_provider_factory.cc b/onnxruntime/core/providers/cuda/cuda_provider_factory.cc
@@ -734,6 +734,10 @@
     }
     */
 
+    // guard against bad device discovery. max devices we expect to add is num_cuda_devices. if we're attempting
+    // to add more than that we have duplicates in the `devices` array.
+    max_ep_devices = std::min(max_ep_devices, static_cast<size_t>(num_cuda_devices));
+
     int16_t device_id = 0;
     for (size_t i = 0; i < num_devices && num_ep_devices < max_ep_devices; ++i) {
       const OrtHardwareDevice& device = *devices[i];