Merge branch 'microsoft:main' into jclohess_kleidiai_gemv_implementation

Author: JonathanC-ARM

cmake/deps.txt

@@ -47,7 +47,7 @@ protoc_mac_universal;https://github.com/protocolbuffers/protobuf/releases/downlo
 psimd;https://github.com/Maratyszcza/psimd/archive/072586a71b55b7f8c584153d223e95687148a900.zip;1f5454b01f06f9656b77e4a5e2e31d7422487013
 pthreadpool;https://github.com/google/pthreadpool/archive/dcc9f28589066af0dbd4555579281230abbf74dd.zip;533a77943203ef15ca608bcd9dbe2c94da7451d2
 pybind11;https://github.com/pybind/pybind11/archive/refs/tags/v2.13.6.zip;f780292da9db273c8ef06ccf5fd4b623624143e9
-pytorch_cpuinfo;https://github.com/pytorch/cpuinfo/archive/de0ce7c7251372892e53ce9bc891750d2c9a4fd8.zip;c45b8d3619b9bccbd26dc5f657959aee38b18b7a
+pytorch_cpuinfo;https://github.com/pytorch/cpuinfo/archive/877328f188a3c7d1fa855871a278eb48d530c4c0.zip;9152d4bf6b8bde9f19b116de3bd8a745097ed9df
 re2;https://github.com/google/re2/archive/refs/tags/2024-07-02.zip;646e1728269cde7fcef990bf4a8e87b047882e88
 safeint;https://github.com/dcleblanc/SafeInt/archive/refs/tags/3.0.28.zip;23f252040ff6cb9f1fd18575b32fa8fb5928daac
 tensorboard;https://github.com/tensorflow/tensorboard/archive/373eb09e4c5d2b3cc2493f0949dc4be6b6a45e81.zip;67b833913605a4f3f499894ab11528a702c2b381

cmake/external/onnxruntime_external_deps.cmake

@@ -440,7 +440,7 @@ if(onnxruntime_USE_VCPKG)
   find_package(flatbuffers REQUIRED)
 else()
 # We do not need to build flatc for iOS or Android Cross Compile
-if (CMAKE_SYSTEM_NAME STREQUAL "iOS" OR CMAKE_SYSTEM_NAME STREQUAL "Android" OR CMAKE_SYSTEM_NAME STREQUAL "Emscripten")
+if (CMAKE_SYSTEM_NAME STREQUAL "iOS" OR CMAKE_SYSTEM_NAME STREQUAL "tvOS" OR CMAKE_SYSTEM_NAME STREQUAL "visionOS" OR CMAKE_SYSTEM_NAME STREQUAL "Android" OR CMAKE_SYSTEM_NAME STREQUAL "Emscripten")
   set(FLATBUFFERS_BUILD_FLATC OFF CACHE BOOL "FLATBUFFERS_BUILD_FLATC" FORCE)
 endif()
 set(FLATBUFFERS_BUILD_TESTS OFF CACHE BOOL "FLATBUFFERS_BUILD_TESTS" FORCE)

cmake/onnxruntime_python.cmake

@@ -962,13 +962,33 @@ if (onnxruntime_USE_TENSORRT)
 endif()
 
 if (onnxruntime_USE_NV)
+  if (WIN32 OR ${CMAKE_SYSTEM_NAME} STREQUAL "Linux")
+      file(GLOB NV_LIB_FILES LIST_DIRECTORIES false "${TENSORRT_RTX_ROOT}/lib/tensorrt_*.dll"
+                                             "${TENSORRT_RTX_ROOT}/bin/tensorrt_*.dll"
+                                             "${TENSORRT_RTX_ROOT}/lib/libtensorrt_*.so"
+                                             "${TENSORRT_RTX_ROOT}/bin/libtensorrt_*.so")
+    message(STATUS "NV lib files: " ${NV_LIB_FILES})
+  endif()
   add_custom_command(
     TARGET onnxruntime_pybind11_state POST_BUILD
     COMMAND ${CMAKE_COMMAND} -E copy
+        ${NV_LIB_FILES}
         $<TARGET_FILE:onnxruntime_providers_nv_tensorrt_rtx>
         $<TARGET_FILE:onnxruntime_providers_shared>
         $<TARGET_FILE_DIR:${build_output_target}>/onnxruntime/capi/
   )
+  if (EXISTS "${TENSORRT_RTX_ROOT}/doc/LICENSE.txt")
+    add_custom_command(
+      TARGET onnxruntime_pybind11_state POST_BUILD
+        COMMAND ${CMAKE_COMMAND} -E copy "${TENSORRT_RTX_ROOT}/doc/LICENSE.txt" $<TARGET_FILE_DIR:${build_output_target}>/onnxruntime/TRT_RTX_LICENSE.txt
+    )
+  endif()
+  if (EXISTS "${TENSORRT_RTX_ROOT}/doc/Acknowledgements.txt")
+    add_custom_command(
+      TARGET onnxruntime_pybind11_state POST_BUILD
+        COMMAND ${CMAKE_COMMAND} -E copy "${TENSORRT_RTX_ROOT}/doc/Acknowledgements.txt" $<TARGET_FILE_DIR:${build_output_target}>/onnxruntime/TRT_RTX_Acknowledgements.txt
+    )
+  endif()
 endif()
 
 if (onnxruntime_USE_MIGRAPHX)

cmake/vcpkg-ports/cpuinfo/portfile.cmake

@@ -6,8 +6,8 @@ endif()
 vcpkg_from_github(
     OUT_SOURCE_PATH SOURCE_PATH
     REPO pytorch/cpuinfo
-    REF de0ce7c7251372892e53ce9bc891750d2c9a4fd8
-    SHA512 0fde9210b700d2648d37c8deeb0d5c0d007d8ca5689578dd3bce4c460886b20d7649f0194d2ea06b02238fe9d4f06193599ec3ab5cafb19f1f860b00404264fa
+    REF 877328f188a3c7d1fa855871a278eb48d530c4c0
+    SHA512 b6d5a9ce9996eee3b2f09f39115f7ae178fe4d4814cc35b049a59d04a82228e268aa52d073c307ccb56a427428622940e1c77f004c99851dfca0d3a5d803658b
     HEAD_REF master
     PATCHES
         patch_cpuinfo_h_for_arm64ec.patch

cmake/vcpkg-ports/cpuinfo/vcpkg.json

@@ -1,7 +1,7 @@
 {
   "name": "cpuinfo",
-  "version-date": "2024-12-09",
-  "port-version": 3,
+  "version-date": "2025-10-23",
+  "port-version": 4,
   "description": "CPU INFOrmation library (x86/x86-64/ARM/ARM64, Linux/Windows/Android/macOS/iOS)",
   "homepage": "https://github.com/pytorch/cpuinfo",
   "license": "BSD-2-Clause",

js/web/package.json

@@ -115,7 +115,15 @@
         "default": "./dist/ort.jspi.bundle.min.mjs"
       },
       "require": "./dist/ort.jspi.min.js"
-    }
+    },
+    "./ort-wasm-simd-threaded.wasm": "./dist/ort-wasm-simd-threaded.wasm",
+    "./ort-wasm-simd-threaded.jsep.wasm": "./dist/ort-wasm-simd-threaded.jsep.wasm",
+    "./ort-wasm-simd-threaded.jspi.wasm": "./dist/ort-wasm-simd-threaded.jspi.wasm",
+    "./ort-wasm-simd-threaded.asyncify.wasm": "./dist/ort-wasm-simd-threaded.asyncify.wasm",
+    "./ort-wasm-simd-threaded.mjs": "./dist/ort-wasm-simd-threaded.mjs",
+    "./ort-wasm-simd-threaded.jsep.mjs": "./dist/ort-wasm-simd-threaded.jsep.mjs",
+    "./ort-wasm-simd-threaded.jspi.mjs": "./dist/ort-wasm-simd-threaded.jspi.mjs",
+    "./ort-wasm-simd-threaded.asyncify.mjs": "./dist/ort-wasm-simd-threaded.asyncify.mjs"
   },
   "types": "./types.d.ts",
   "description": "A Javascript library for running ONNX models on browsers"

js/web/test/e2e/package.json

@@ -20,6 +20,7 @@
     "rollup": "^4.13.2",
     "rollup-plugin-copy": "^3.5.0",
     "tree-kill": "^1.2.2",
+    "vite": "^7.1.12",
     "webpack-cli": "^5.1.4"
   },
   "scripts": {
@@ -32,7 +33,10 @@
     "build:p:esmjs": "parcel build --no-autoinstall --target esm && node ./bundler.esm.postprocess.js ./dist/parcel_esm_js/main.js",
     "build:p:umdjs": "parcel build --no-autoinstall --target umd",
     "build:p": "npm run build:p:esmjs && npm run build:p:umdjs",
-    "build": "npm run build:w && npm run build:r && npm run build:p"
+    "build:v:esmjs": "vite build -c vite.config.esm-js.js",
+    "build:v:umdjs": "vite build -c vite.config.umd-js.js",
+    "build:v": "npm run build:v:esmjs && npm run build:v:umdjs",
+    "build": "npm run build:v && npm run build:w && npm run build:r && npm run build:p"
   },
   "@parcel/resolver-default": {
     "packageExports": true

js/web/test/e2e/run-data.js

@@ -77,6 +77,8 @@ const BUNDLER_TEST_CASES = [
   ['./dist/rollup_umd_js/ort-test-e2e.bundle.js', 'iife'],
   ['./dist/parcel_esm_js/main.js', 'esm'],
   ['./dist/parcel_umd_js/main.js', 'iife'],
+  ['./dist/vite_esm_js/ort-test-e2e.bundle.mjs', 'esm'],
+  ['./dist/vite_umd_js/ort-test-e2e.bundle.js', 'iife'],
 ];
 
 module.exports = {

js/web/test/e2e/src/cjs-js/vite-main.js

@@ -0,0 +1,13 @@
+require('onnxruntime-web/ort-wasm-simd-threaded.wasm?url&no-inline');
+require('onnxruntime-web/ort-wasm-simd-threaded.mjs?url&no-inline');
+
+require('onnxruntime-web/ort-wasm-simd-threaded.jsep.wasm?url&no-inline');
+require('onnxruntime-web/ort-wasm-simd-threaded.jsep.mjs?url&no-inline');
+
+require('onnxruntime-web/ort-wasm-simd-threaded.jspi.wasm?url&no-inline');
+require('onnxruntime-web/ort-wasm-simd-threaded.jspi.mjs?url&no-inline');
+
+require('onnxruntime-web/ort-wasm-simd-threaded.asyncify.wasm?url&no-inline');
+require('onnxruntime-web/ort-wasm-simd-threaded.asyncify.mjs?url&no-inline');
+
+require('./main');

js/web/test/e2e/src/esm-js/vite-main.js

@@ -0,0 +1,13 @@
+import 'onnxruntime-web/ort-wasm-simd-threaded.wasm?url&no-inline';
+import 'onnxruntime-web/ort-wasm-simd-threaded.mjs?url&no-inline';
+
+import 'onnxruntime-web/ort-wasm-simd-threaded.jsep.wasm?url&no-inline';
+import 'onnxruntime-web/ort-wasm-simd-threaded.jsep.mjs?url&no-inline';
+
+import 'onnxruntime-web/ort-wasm-simd-threaded.jspi.wasm?url&no-inline';
+import 'onnxruntime-web/ort-wasm-simd-threaded.jspi.mjs?url&no-inline';
+
+import 'onnxruntime-web/ort-wasm-simd-threaded.asyncify.wasm?url&no-inline';
+import 'onnxruntime-web/ort-wasm-simd-threaded.asyncify.mjs?url&no-inline';
+
+import './main.js';