Merge branch 'master' into sync_msft_14112025

Author: Jaswanth51

.github/workflows/reusable_linux_build.yml

@@ -78,6 +78,7 @@ jobs:
         uses: actions/checkout@v5
 
       - name: Set up Python ${{ inputs.python_version }}
+        if: inputs.architecture != 'arm64'
         uses: actions/setup-python@v6
         with:
           python-version: ${{ inputs.python_version }}

cmake/CMakeLists.txt

@@ -1454,7 +1454,12 @@ if (onnxruntime_USE_CUDA)
     message(STATUS "CUDA Toolkit version is greater or equal than 12.8, enable -DENABLE_FP4 flag")
   endif()
 
-  set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -Xfatbin=-compress-all")
+  if(CMAKE_CUDA_COMPILER_VERSION VERSION_GREATER_EQUAL "13.0")
+    set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -Xfatbin=-compress-all -compress-mode=size")
+  else()
+    set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -Xfatbin=-compress-all")
+  endif()
+
   set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} --expt-relaxed-constexpr")
   set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} --Werror default-stream-launch")
 
@@ -1654,9 +1659,25 @@ if ("${CMAKE_SYSTEM_NAME}" STREQUAL "Linux")
   endif()
 endif()
 
-#Now the 'onnxruntime_EXTERNAL_LIBRARIES' variable should be sealed. It will be used in onnxruntime.cmake which will be included in the next.
-#The order of the following targets matters. Right depends on left. If target A appears before target B. Then A.cmake can not use variables defined in B.cmake.
-set(ONNXRUNTIME_CMAKE_FILES onnxruntime_flatbuffers onnxruntime_common onnxruntime_mlas onnxruntime_graph onnxruntime_lora onnxruntime_framework onnxruntime_util onnxruntime_providers onnxruntime_optimizer onnxruntime_session ${ONNXRUNTIME_EAGER_CMAKE_FILE_NAME})
+# From this point on, onnxruntime_EXTERNAL_LIBRARIES should be no longer be modified. It will be used in
+# onnxruntime.cmake which will be included next.
+
+# The order of the CMake file names (which exclude the ".cmake" suffix) in ONNXRUNTIME_CMAKE_FILES matters. Later CMake
+# files may depend on earlier ones but earlier ones cannot depend on later ones. For example, if A appears before B,
+# then A.cmake cannot use variables defined in B.cmake.
+set(ONNXRUNTIME_CMAKE_FILES
+    onnxruntime_flatbuffers
+    onnxruntime_common
+    onnxruntime_mlas
+    onnxruntime_graph
+    onnxruntime_lora
+    onnxruntime_framework
+    onnxruntime_util
+    onnxruntime_providers
+    onnxruntime_optimizer
+    onnxruntime_session
+    ${ONNXRUNTIME_EAGER_CMAKE_FILE_NAME}
+    )
 
 if (onnxruntime_USE_WINML)
   # WINML uses and depends on the shared lib.  Note:  You can build WINML without DML and you will get a

cmake/deps.txt

@@ -56,5 +56,5 @@ extensions;https://github.com/microsoft/onnxruntime-extensions/archive/c24b7bab0
 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/13c1635a14574ebb7116b56a69f5519301417fda.zip;0aadd28fc385cf7d657d5fc70a352372d2d3c76a
-kleidiai;https://github.com/ARM-software/kleidiai/archive/refs/tags/v1.10.0.tar.gz;11b62149cb2514b3b9069cc435c3aa7a4e82b97a
+kleidiai;https://github.com/ARM-software/kleidiai/archive/refs/tags/v1.15.0.tar.gz;62ccd24ab60bcef68766440fb42d79071ac2a5d2
 duktape;https://github.com/svaarala/duktape/releases/download/v2.7.0/duktape-2.7.0.tar.xz;8200c8e417dbab7adcc12c4dbdef7651cfc55794

cmake/external/onnxruntime_external_deps.cmake

@@ -751,9 +751,18 @@ if (onnxruntime_USE_WEBGPU)
           #   Some build warnings are not allowed to be disabled in project level.
           ${Patch_EXECUTABLE} --binary --ignore-whitespace -p1 < ${PROJECT_SOURCE_DIR}/patches/dawn/dawn_binskim.patch &&
 
-          # Android devices doesn't seem to allow fp16 in uniforms so the WebGPU EP has to manually handle passing an fp32
-          # in the uniform and converting to fp16 before using.
-          ${Patch_EXECUTABLE} --binary --ignore-whitespace -p1 < ${PROJECT_SOURCE_DIR}/patches/dawn/uniform_and_storage_buffer_16_bit_access.patch)
+          # The uniform_and_storage_buffer_16_bit_access.patch contains the following changes:
+          #
+          # - (private) Android devices don't seem to allow fp16 in uniforms so the WebGPU EP has to manually handle passing an fp32
+          #   in the uniform and converting to fp16 before using.
+          ${Patch_EXECUTABLE} --binary --ignore-whitespace -p1 < ${PROJECT_SOURCE_DIR}/patches/dawn/uniform_and_storage_buffer_16_bit_access.patch &&
+
+          # The safari_polyfill.patch contains the following changes:
+          #
+          # - (private) Fix compatibility issues with Safari. Contains the following changes:
+          #   - Polyfill for `device.AdapterInfo` (returns `undefined` in Safari v26.0)
+          #
+          ${Patch_EXECUTABLE} --binary --ignore-whitespace -p1 < ${PROJECT_SOURCE_DIR}/patches/dawn/safari_polyfill.patch)
 
       onnxruntime_fetchcontent_declare(
         dawn

cmake/onnxruntime_mlas.cmake

@@ -5,6 +5,7 @@ set(MLAS_ROOT ${ONNXRUNTIME_ROOT}/core/mlas)
 set(MLAS_SRC_DIR ${MLAS_ROOT}/lib)
 set(MLAS_INC_DIR ${MLAS_ROOT}/inc)
 
+
 # mlas_private_compile_definitions contains compile definitions that are private to onnxruntime_mlas and targets which
 # use internal MLAS headers like mlasi.h.
 set(mlas_private_compile_definitions)
@@ -285,6 +286,15 @@ function(setup_kleidiai)
   list(APPEND onnxruntime_EXTERNAL_LIBRARIES kleidiai)
   set(onnxruntime_EXTERNAL_LIBRARIES ${onnxruntime_EXTERNAL_LIBRARIES} PARENT_SCOPE)
 
+  # If KLEIDIAI_DEBUG is enabled that implies both DEBUG and KERNEL messages.
+  if(onnxruntime_KLEIDIAI_DEBUG_LOGGING)
+    target_compile_definitions(onnxruntime_mlas PRIVATE KLEIDIAI_DEBUG=1)
+    target_compile_definitions(onnxruntime_mlas PRIVATE KLEIDIAI_KERNEL=1)
+  endif()
+  if(onnxruntime_KLEIDIAI_KERNEL_LOGGING)
+    target_compile_definitions(onnxruntime_mlas PRIVATE KLEIDIAI_KERNEL=1)
+  endif()
+
   if (NOT onnxruntime_BUILD_SHARED_LIB)
     install(TARGETS kleidiai EXPORT ${PROJECT_NAME}Targets
     ARCHIVE DESTINATION ${CMAKE_INSTALL_LIBDIR}

cmake/onnxruntime_unittests.cmake

@@ -588,6 +588,7 @@ set (onnxruntime_shared_lib_test_SRC
           ${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_runtime_path.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
@@ -1518,9 +1519,48 @@ endif()
     target_link_libraries(onnxruntime_mocked_allocator PRIVATE ${GSL_TARGET})
     set_target_properties(onnxruntime_mocked_allocator PROPERTIES FOLDER "ONNXRuntimeTest")
 
+    # onnxruntime_runtime_path_test_shared_library
+    block()
+      set(onnxruntime_runtime_path_test_shared_library_src
+          "${TEST_SRC_DIR}/shared_lib/runtime_path_test_shared_library/runtime_path_test_shared_library.h"
+          "${TEST_SRC_DIR}/shared_lib/runtime_path_test_shared_library/runtime_path_test_shared_library.cc")
+
+      onnxruntime_add_shared_library(onnxruntime_runtime_path_test_shared_library
+                                     ${onnxruntime_runtime_path_test_shared_library_src})
+
+      target_link_libraries(onnxruntime_runtime_path_test_shared_library PRIVATE
+                            onnxruntime_common cpuinfo ${CMAKE_DL_LIBS})
+      target_include_directories(onnxruntime_runtime_path_test_shared_library PRIVATE ${ONNXRUNTIME_ROOT})
+
+      if(UNIX)
+        if (APPLE)
+          set(onnxruntime_runtime_path_test_shared_library_link_flags "-Xlinker -dead_strip")
+        elseif (NOT CMAKE_SYSTEM_NAME MATCHES "AIX")
+          string(CONCAT onnxruntime_runtime_path_test_shared_library_link_flags
+                 "-Xlinker --version-script=${TEST_SRC_DIR}/shared_lib/runtime_path_test_shared_library/runtime_path_test_shared_library.lds "
+                 "-Xlinker --no-undefined -Xlinker --gc-sections -z noexecstack")
+        endif()
+      else()
+        set(onnxruntime_runtime_path_test_shared_library_link_flags
+            "-DEF:${TEST_SRC_DIR}/shared_lib/runtime_path_test_shared_library/runtime_path_test_shared_library.def")
+      endif()
+
+      set_property(TARGET onnxruntime_runtime_path_test_shared_library APPEND_STRING PROPERTY LINK_FLAGS
+                   ${onnxruntime_runtime_path_test_shared_library_link_flags})
+
+      set_target_properties(onnxruntime_runtime_path_test_shared_library PROPERTIES FOLDER "ONNXRuntimeTest")
+      source_group(TREE ${TEST_SRC_DIR} FILES ${onnxruntime_runtime_path_test_shared_library_src})
+    endblock()
+
     #################################################################
     # test inference using shared lib
-    set(onnxruntime_shared_lib_test_LIBS onnxruntime_mocked_allocator onnxruntime_test_utils onnxruntime_common onnx_proto)
+    set(onnxruntime_shared_lib_test_LIBS
+        onnxruntime_mocked_allocator
+        onnxruntime_test_utils
+        onnxruntime_common
+        onnx_proto
+        onnxruntime_runtime_path_test_shared_library)
+
     if(NOT WIN32)
       if(onnxruntime_USE_SNPE)
         list(APPEND onnxruntime_shared_lib_test_LIBS onnxruntime_providers_snpe)
@@ -1576,6 +1616,10 @@ endif()
       target_compile_definitions(onnxruntime_shared_lib_test PRIVATE USE_DUMMY_EXA_DEMANGLE=1)
     endif()
 
+    if (CMAKE_SYSTEM_NAME MATCHES "AIX" AND CMAKE_CXX_COMPILER_ID STREQUAL "GNU")
+      set_target_properties(onnxruntime_shared_lib_test PROPERTIES ENABLE_EXPORTS 1)
+    endif()
+
     if (IOS)
       add_custom_command(
         TARGET onnxruntime_shared_lib_test POST_BUILD
@@ -1978,8 +2022,7 @@ endif()
 
 # Build library that can be used with RegisterExecutionProviderLibrary and automatic EP selection
 # We need a shared lib build to use that as a dependency for the test library
-# Currently we only have device discovery on Windows so no point building the test app on other platforms.
-if (WIN32 AND onnxruntime_BUILD_SHARED_LIB AND
+if (onnxruntime_BUILD_SHARED_LIB AND
     NOT CMAKE_SYSTEM_NAME STREQUAL "Emscripten" AND
     NOT onnxruntime_MINIMAL_BUILD)
 
@@ -1991,7 +2034,7 @@ if (WIN32 AND onnxruntime_BUILD_SHARED_LIB AND
                                                 "${TEST_SRC_DIR}/autoep/library/plugin_ep_utils.h")
   onnxruntime_add_shared_library_module(example_plugin_ep ${onnxruntime_autoep_test_library_src})
   target_include_directories(example_plugin_ep PRIVATE ${REPO_ROOT}/include/onnxruntime/core/session)
-  target_link_libraries(example_plugin_ep PRIVATE onnxruntime)
+  target_link_libraries(example_plugin_ep PRIVATE onnxruntime ${GSL_TARGET})
 
   if(UNIX)
     if (APPLE)
@@ -2024,7 +2067,7 @@ if (WIN32 AND onnxruntime_BUILD_SHARED_LIB AND
           "${TEST_SRC_DIR}/autoep/library/example_plugin_ep_virt_gpu/ep.cc")
   onnxruntime_add_shared_library_module(example_plugin_ep_virt_gpu ${onnxruntime_autoep_test_example_plugin_ep_virt_gpu_src})
   target_include_directories(example_plugin_ep_virt_gpu PRIVATE ${REPO_ROOT}/include/onnxruntime/core/session)
-  target_link_libraries(example_plugin_ep_virt_gpu PRIVATE onnxruntime)
+  target_link_libraries(example_plugin_ep_virt_gpu PRIVATE onnxruntime ${GSL_TARGET})
 
   if(UNIX)
     if (APPLE)

cmake/patches/dawn/safari_polyfill.patch

@@ -0,0 +1,17 @@
+diff --git a/third_party/emdawnwebgpu/pkg/webgpu/src/library_webgpu.js b/third_party/emdawnwebgpu/pkg/webgpu/src/library_webgpu.js
+index a3ce6dc732..7119eae424 100644
+--- a/third_party/emdawnwebgpu/pkg/webgpu/src/library_webgpu.js
++++ b/third_party/emdawnwebgpu/pkg/webgpu/src/library_webgpu.js
+@@ -903,6 +903,12 @@ var LibraryWebGPU = {
+           stackRestore(sp);
+       };
+ 
++      // Polyfill `device.AdapterInfo` if not present (Safari v26.0).
++      // See https://bugs.webkit.org/show_bug.cgi?id=301878
++      if (!('adapterInfo' in device)) {
++        device.adapterInfo = adapter.info;
++      }
++
+       _emwgpuOnRequestDeviceCompleted(futureId, {{{ gpu.RequestDeviceStatus.Success }}},
+         {{{ gpu.passAsPointer('devicePtr') }}}, {{{ gpu.NULLPTR }}});
+     }, (ex) => {

cmake/vcpkg-ports/onnx/portfile.cmake

@@ -36,7 +36,6 @@ vcpkg_cmake_configure(
         "-DProtobuf_PROTOC_EXECUTABLE:FILEPATH=${PROTOC}"
         -DONNX_ML=ON
         -DONNX_USE_PROTOBUF_SHARED_LIBS=${USE_PROTOBUF_SHARED}
-        -DONNX_USE_LITE_PROTO=OFF
         -DONNX_USE_MSVC_STATIC_RUNTIME=${USE_STATIC_RUNTIME}
         -DONNX_BUILD_TESTS=OFF
         -DONNX_BUILD_BENCHMARKS=OFF

csharp/src/Microsoft.ML.OnnxRuntime/NativeMethods.shared.cs

@@ -866,13 +866,14 @@ static NativeMethods()
         internal class NativeLib
         {
 #if __ANDROID__
-            // define the library name required for android
+            // Define the library name required for Android
             internal const string DllName = "libonnxruntime.so";
 #elif __IOS__
-            // define the library name required for iOS
+            // Define the library name required for iOS
             internal const string DllName = "__Internal";
 #else
-            internal const string DllName = "onnxruntime";
+            // Note: the file name in ONNX Runtime nuget package must be onnxruntime.dll instead of onnxruntime.DLL(Windows filesystem can be case sensitive)
+            internal const string DllName = "onnxruntime.dll";
 #endif
         }
 
@@ -2951,7 +2952,9 @@ internal static class OrtExtensionsNativeMethods
 #elif __IOS__
         internal const string ExtensionsDllName = "__Internal";
 #else
-        internal const string ExtensionsDllName = "ortextensions";
+        // For desktop platforms, explicitly specify the DLL name with extension to avoid
+        // issues on case-sensitive filesystems. See NativeLib.DllName for detailed explanation.
+        internal const string ExtensionsDllName = "ortextensions.dll";
 #endif
 
         [DllImport(ExtensionsDllName, CharSet = CharSet.Ansi,

docs/ContribOperators.md

@@ -4534,10 +4534,19 @@ This version of the operator has been available since version 1 of the 'com.micr
 
   Quantized mixture of experts (MoE).
 
-        Only weights are quantized with symmetric quantization.
         The quantized weights are stored in column major order per expert.
         The quantization block size can be specified. If not provided, column wise quantization is used.
 
+        The formula of linear dequantization of the quantized weights using scale and (optionally) zero-point is:
+          dequantized_weight = (quantized_weight - zero_point) * scale
+        When zero_point is not provided, the default value is 2^(bits-1): 8 for 4 bits, 128 for 8 bits.
+  
+        If block_size is provided, both hidden_size and inter_size must be divisible by the block size, and
+        the dequantization is performed per block of size block_size along the K (input feature) dimension.
+  
+        If block_size and zero_point are provided, both hidden_size and inter_size must be divisible by block_size * pack_size,
+        where pack_size = 8 / expert_weight_bits.
+  
         The SwiGLU (Swish-Gated Linear Unit) activation function is like:
            g = xW + b
            l = xV + c
@@ -4579,7 +4588,7 @@ This version of the operator has been available since version 1 of the 'com.micr
 <dd>Whether to use sparse mixer</dd>
 </dl>
 
-#### Inputs (7 - 11)
+#### Inputs (7 - 14)
 
 <dl>
 <dt><tt>input</tt> : T</dt>
@@ -4604,6 +4613,12 @@ This version of the operator has been available since version 1 of the 'com.micr
 <dd>2D optional tensor with shape (num_experts, inter_size), or 3D optional tensor with shape (num_experts, inter_size, hidden_size / block_size) when block_size is provided.</dd>
 <dt><tt>fc3_experts_bias</tt> (optional) : T</dt>
 <dd>2D optional tensor with shape (num_experts, inter_size)</dd>
+<dt><tt>fc1_zero_points</tt> (optional) : T1</dt>
+<dd>2D tensor with shape (num_experts, fusion_size * inter_size / pack_size), or 3D tensor with shape (num_experts, fusion_size * inter_size, hidden_size / block_size / pack_size) when block_size is provided.</dd>
+<dt><tt>fc2_zero_points</tt> (optional) : T1</dt>
+<dd>2D tensor with shape (num_experts, hidden_size / pack_size), or 3D tensor with shape (num_experts, hidden_size, inter_size / block_size / pack_size) when block_size is provided.</dd>
+<dt><tt>fc3_zero_points</tt> (optional) : T1</dt>
+<dd>2D optional tensor with shape (num_experts, inter_size / pack_size), or 3D optional tensor with shape (num_experts, inter_size, hidden_size / block_size / pack_size) when block_size is provided.</dd>
 </dl>
 
 #### Outputs