Sync with Microsoft ONNX Runtime - 20032026

cmake/CMakeLists.txt

@@ -992,10 +992,10 @@ if (onnxruntime_USE_WEBGPU)
   if (onnxruntime_USE_EXTERNAL_DAWN)
     list(APPEND ORT_PROVIDER_FLAGS -DUSE_EXTERNAL_DAWN=1)
   endif()
-  if (onnxruntime_ENABLE_DAWN_BACKEND_VULKAN)
+  if ((onnxruntime_ENABLE_DAWN_BACKEND_VULKAN AND WIN32) OR LINUX OR ANDROID)
     list(APPEND ORT_PROVIDER_FLAGS -DDAWN_ENABLE_VULKAN=1)
   endif()
-  if (onnxruntime_ENABLE_DAWN_BACKEND_D3D12)
+  if (onnxruntime_ENABLE_DAWN_BACKEND_D3D12 AND WIN32)
     list(APPEND ORT_PROVIDER_FLAGS -DDAWN_ENABLE_D3D12=1)
   endif()
   if (onnxruntime_ENABLE_PIX_FOR_WEBGPU_EP)
@@ -1189,7 +1189,9 @@ function(onnxruntime_set_compile_flags target_name)
     endif()
     if (onnxruntime_USE_CUDA)
       foreach(FLAG ${ORT_WARNING_FLAGS})
-        target_compile_options(${target_name} PRIVATE "$<$<COMPILE_LANGUAGE:CUDA>:SHELL:--compiler-options ${FLAG}>")
+        if (NOT "${FLAG}" STREQUAL "-Wshorten-64-to-32")
+          target_compile_options(${target_name} PRIVATE "$<$<COMPILE_LANGUAGE:CUDA>:SHELL:--compiler-options ${FLAG}>")
+        endif()
       endforeach()
       if (NVCC_HAS_STRICT_ALIASING AND "${target_name}" MATCHES "cuda")
         target_compile_options(${target_name} PRIVATE "$<$<COMPILE_LANGUAGE:CUDA>:-Wno-strict-aliasing>")

cmake/onnxruntime_mlas.cmake

@@ -284,6 +284,7 @@ function(setup_kleidiai)
   target_sources(onnxruntime_mlas PRIVATE
     ${MLAS_SRC_DIR}/kai_ukernel_interface.cpp
     ${MLAS_SRC_DIR}/kleidiai/sgemm_kleidiai.cpp
+    ${MLAS_SRC_DIR}/kleidiai/sbgemm_kleidiai.cpp
     ${MLAS_SRC_DIR}/kleidiai/convolve_kleidiai.cpp
     ${MLAS_SRC_DIR}/kleidiai/qgemm_kleidiai.cpp
   )

cmake/onnxruntime_providers_cuda.cmake

@@ -181,7 +181,9 @@
     endif()
 
     foreach(ORT_FLAG ${ORT_WARNING_FLAGS})
+      if (NOT "${ORT_FLAG}" STREQUAL "-Wshorten-64-to-32")
         target_compile_options(${target} PRIVATE "$<$<COMPILE_LANGUAGE:CUDA>:SHELL:-Xcompiler \"${ORT_FLAG}\">")
+      endif()
     endforeach()
 
     # Note: The minimum required CUDA version is greater than 11.3.
@@ -222,6 +224,38 @@
                   "$<$<NOT:$<COMPILE_LANGUAGE:CUDA>>:-Wno-reorder>")
       target_compile_options(${target} PRIVATE "$<$<COMPILE_LANGUAGE:CUDA>:SHELL:-Xcompiler -Wno-error=sign-compare>"
                   "$<$<NOT:$<COMPILE_LANGUAGE:CUDA>>:-Wno-error=sign-compare>")
+      if (CMAKE_CXX_COMPILER_ID STREQUAL "Clang" OR CMAKE_CXX_COMPILER_ID STREQUAL "AppleClang" OR CMAKE_CXX_COMPILER_ID STREQUAL "IBMClang")
+        foreach(CLANG_WARNING
+          braced-scalar-init
+          defaulted-function-deleted
+          inconsistent-missing-override
+          instantiation-after-specialization
+          logical-op-parentheses
+          mismatched-tags
+          shorten-64-to-32
+          unneeded-internal-declaration
+          unknown-warning-option
+          unused-private-field
+          unused-variable)
+          target_compile_options(${target} PRIVATE "$<$<NOT:$<COMPILE_LANGUAGE:CUDA>>:-Wno-error=${CLANG_WARNING}>")
+        endforeach()
+        if (CMAKE_CUDA_HOST_COMPILER_ID STREQUAL "Clang" OR CMAKE_CUDA_HOST_COMPILER_ID STREQUAL "AppleClang" OR CMAKE_CUDA_HOST_COMPILER_ID STREQUAL "IBMClang")
+          foreach(CLANG_WARNING
+            braced-scalar-init
+            defaulted-function-deleted
+            inconsistent-missing-override
+            instantiation-after-specialization
+            logical-op-parentheses
+            mismatched-tags
+            shorten-64-to-32
+            unneeded-internal-declaration
+            unknown-warning-option
+            unused-private-field
+            unused-variable)
+            target_compile_options(${target} PRIVATE "$<$<COMPILE_LANGUAGE:CUDA>:SHELL:-Xcompiler -Wno-error=${CLANG_WARNING}>")
+          endforeach()
+        endif()
+      endif()
     else()
       #mutex.cuh(91): warning C4834: discarding return value of function with 'nodiscard' attribute
       target_compile_options(${target} PRIVATE "$<$<COMPILE_LANGUAGE:CUDA>:SHELL:-Xcompiler /wd4834>")