Upgrade emsdk version to v4.0.3 (#23633)

### Description

Upgrade EMSDK to v4.0.3

### File Size Comparison:

Baseline: e666503
[[Build]](https://dev.azure.com/onnxruntime/onnxruntime/_build/results?buildId=1613209)
This change: bb9f747
[[Build]](https://dev.azure.com/onnxruntime/onnxruntime/_build/results?buildId=1613180)

|File | Baseline | This Change | Diff |
|--|--|---|---|
| ort-wasm-simd-threaded.jsep.mjs | 49,240 | 44,230 | -5,010 |
| ort-wasm-simd-threaded.jsep.wasm | 22,921,505 | 21,236,866 |
-1,684,639 |
| ort-wasm-simd-threaded.mjs | 25,539 | 20,890 | -4,649 |
| ort-wasm-simd-threaded.wasm | 11,771,161 | 10,942,193 | -828,968 |
| ort.min.mjs | 345,608 | 345,608 | 0 |
| ort.bundle.min.mjs | 392,960 | 388,426 | -4,534 |

### Motivation and Context
<!-- - Why is this change required? What problem does it solve?
- If it fixes an open issue, please link to the issue here. -->


- EMSDK 3.1.62-3.1.74 is not working because of an linker issue
(wasm-ld) that causes out-of-memory or crash.
- EMSDK > 3.1.69 requires a change as a post-process of the generated JS
file otherwise it will fails onnxruntime-web (as a NPM package) to be
consumed in user's web application because of an unexpected webpack
behavior.
- EMSDK >= 4.0.1 fixes the out-of-memory issue but brought new issues
causes build break of an indirect dependency of ORT (ORT ->
ORT-extensions -> dlib).
- EMSDK 4.0.3 introduces a change that allows a previous workaround to
be removed.

In this PR, all of the above issues are addressed:
- the linker issue (wasm-ld) is fixed in 4.0.1+
- the post-process is added in `onnxruntime_webassembly.cmake` as a
`POST_BUILD` event of target `onnxruntime_webassembly`. It's done by a
Node.js script that uses Regex match and replace.
- `dlib` not working with EMSDK v4:
- the issue is tracked at davisking/dlib#3045
- Currently no active user is using ORT extension. In this PR, removed
ORT extension from the build.
- file `gen_struct_info.py` is moved from
`${DAWN_EMSCRIPTEN_TOOLCHAIN}/tools/maint` to
`${DAWN_EMSCRIPTEN_TOOLCHAIN}/tools`. This helped to remove the
patch.tgz for emscripten. This change also requires a change in Dawn, so
patched the change needed for Dawn in this PR.

.gitmodules

@@ -7,4 +7,4 @@
 [submodule "cmake/external/emsdk"]
 	path = cmake/external/emsdk
 	url = https://github.com/emscripten-core/emsdk.git
-	branch = 3.1.59
+	branch = 4.0.3

cmake/external/onnxruntime_external_deps.cmake

@@ -510,7 +510,7 @@ else()
   message("Setting pybind11_dep")
   set(pybind11_dep pybind11::pybind11)
 endif()
-  
+
 endif()
 onnxruntime_fetchcontent_declare(
   onnx
@@ -559,7 +559,7 @@ if (onnxruntime_USE_XNNPACK)
      find_library(xnnpack_LIBRARY NAMES XNNPACK)
      find_library(microkernels_prod_LIBRARY NAMES microkernels-prod)
      find_package(unofficial-pthreadpool CONFIG REQUIRED)
-     
+
      target_include_directories(xnnpack INTERFACE "${XNNPACK_HDR}")
      set(XNNPACK_INCLUDE_DIR ${XNNPACK_DIR}/include)
      set(onnxruntime_EXTERNAL_LIBRARIES_XNNPACK ${xnnpack_LIBRARY} ${microkernels_prod_LIBRARY} unofficial::pthreadpool unofficial::pthreadpool_interface)
@@ -643,10 +643,12 @@ if (onnxruntime_USE_WEBGPU)
       dawn
       URL ${DEP_URL_dawn}
       URL_HASH SHA1=${DEP_SHA1_dawn}
-      # All previous patches are merged into the upstream dawn project. We don't need to apply any patches right now.
-      # if we need to apply patches in the future, we can uncomment the following line.
-
-      # PATCH_COMMAND ${Patch_EXECUTABLE} --binary --ignore-whitespace -p1 < ${PROJECT_SOURCE_DIR}/patches/dawn/dawn.patch
+      # # All previous patches are merged into the upstream dawn project. We don't need to apply any patches right now.
+      # # if we need to apply patches in the future, we can uncomment the following line.
+      #
+      # The dawn.patch contains the following changes:
+      # - https://dawn-review.googlesource.com/c/dawn/+/225514
+      PATCH_COMMAND ${Patch_EXECUTABLE} --binary --ignore-whitespace -p1 < ${PROJECT_SOURCE_DIR}/patches/dawn/dawn.patch
       EXCLUDE_FROM_ALL
     )
   endif()
@@ -657,20 +659,6 @@ if (onnxruntime_USE_WEBGPU)
 
   if (CMAKE_SYSTEM_NAME STREQUAL "Emscripten")
     set(DAWN_EMSCRIPTEN_TOOLCHAIN "${REPO_ROOT}/cmake/external/emsdk/upstream/emscripten" CACHE STRING "" FORCE)
-
-    # Add the missing files from the emsdk installation
-    #
-    # For a "standard" emscripten build, the folder "${DAWN_EMSCRIPTEN_TOOLCHAIN}/tools/maint/" is not used. This is the
-    # reason why EMSDK installation does not include it.
-    # However, currently the WebGPU support in Emscripten is still being developed and the Dawn project is maintaining
-    # a fork of the Emscripten toolchain. As an extra build step, Dawn needs to generate some files using the file
-    # "${DAWN_EMSCRIPTEN_TOOLCHAIN}/tools/maint/gen_struct_info.py" from emscripten, which is missing in the emscripten
-    # installed by emsdk.
-    #
-    # We keep a copy of the missing file(s) in ${PROJECT_SOURCE_DIR}/patches/emscripten/, and now we extract them to the
-    # emscripten toolchain folder.
-    execute_process(COMMAND ${CMAKE_COMMAND} -E tar x "${PROJECT_SOURCE_DIR}/patches/emscripten/patch_3.1.74.tgz"
-                    WORKING_DIRECTORY ${DAWN_EMSCRIPTEN_TOOLCHAIN})
   else()
     if (onnxruntime_BUILD_DAWN_MONOLITHIC_LIBRARY)
       set(DAWN_BUILD_MONOLITHIC_LIBRARY ON CACHE BOOL "" FORCE)

cmake/onnxruntime_webassembly.cmake

@@ -472,4 +472,59 @@ jsepDownload:_pp_")
   endif()
 
   set_target_properties(onnxruntime_webassembly PROPERTIES OUTPUT_NAME ${target_name} SUFFIX ".mjs")
+
+  #
+  # The following POST_BUILD script is a workaround for enabling:
+  # - using onnxruntime-web with Multi-threading enabled when import from CDN
+  # - using onnxruntime-web when consumed in some frameworks like Vite
+  #
+  # In the use case mentioned above, the file name of the script may be changed. So we need to replace the line:
+  # `new Worker(new URL("ort-wasm-*.mjs", import.meta.url),`
+  # with
+  # `new Worker(new URL(import.meta.url),`
+  #
+  # This behavior is introduced in https://github.com/emscripten-core/emscripten/pull/22165. Since it's unlikely to be
+  # reverted, and there is no config to disable this behavior, we have to use a post-build script to workaround it.
+  #
+
+  # Generate a script to do the post-build work
+  file(WRITE ${CMAKE_CURRENT_BINARY_DIR}/wasm_post_build.js "
+    const fs = require('fs');
+    const path = require('path');
+
+    // node wasm_post_build.js <mjsFilePath>
+    const mjsFilePath = process.argv[2];
+    let contents = fs.readFileSync(mjsFilePath).toString();
+
+    const regex = 'new Worker\\\\(new URL\\\\(\".+?\", ?import\\\\.meta\\\\.url\\\\),';
+    const matches = [...contents.matchAll(new RegExp(regex, 'g'))];
+    if (matches.length !== 1) {
+      throw new Error(
+        `Unexpected number of matches for \"${regex}\" in \"${filepath}\": ${matches.length}.`,
+      );
+    }
+
+    // Replace the only occurrence.
+    contents = contents.replace(
+      new RegExp(regex),
+      `new Worker(new URL(import.meta.url),`,
+    );
+
+    fs.writeFileSync(mjsFilePath, contents);
+  "
+  )
+
+  find_program(NODE_EXECUTABLE node required)
+  if (NOT NODE_EXECUTABLE)
+    message(FATAL_ERROR "Node is required to run the post-build script")
+  endif()
+
+  add_custom_command(
+    TARGET onnxruntime_webassembly
+    POST_BUILD
+    # Backup file at $<TARGET_FILE_NAME:onnxruntime_webassembly>.bak
+    COMMAND ${CMAKE_COMMAND} -E copy_if_different "$<TARGET_FILE_NAME:onnxruntime_webassembly>" "$<TARGET_FILE_NAME:onnxruntime_webassembly>.bak"
+    COMMAND ${CMAKE_COMMAND} -E echo "Performing post-process for $<TARGET_FILE_NAME:onnxruntime_webassembly>"
+    COMMAND ${NODE_EXECUTABLE} "${CMAKE_CURRENT_BINARY_DIR}/wasm_post_build.js" "$<TARGET_FILE_NAME:onnxruntime_webassembly>"
+  )
 endif()

cmake/patches/dawn/dawn.patch

@@ -0,0 +1,23 @@
+diff --git a/src/emdawnwebgpu/CMakeLists.txt b/src/emdawnwebgpu/CMakeLists.txt
+index 6e8ae37593..633af91eef 100644
+--- a/src/emdawnwebgpu/CMakeLists.txt
++++ b/src/emdawnwebgpu/CMakeLists.txt
+@@ -77,9 +77,17 @@ if (${DAWN_ENABLE_EMSCRIPTEN})
+                 "${arg_UNPARSED_ARGUMENTS}")
+         endif()
+
++        # since Emscripten 4.0.3, file gen_struct_info.py is moved to outside of directory maint.
++        if (EXISTS "${DAWN_EMSCRIPTEN_TOOLCHAIN}/tools/gen_struct_info.py")
++            set(EM_GEN_STRUCT_INFO_SCRIPT "${DAWN_EMSCRIPTEN_TOOLCHAIN}/tools/gen_struct_info.py")
++        elseif (EXISTS "${DAWN_EMSCRIPTEN_TOOLCHAIN}/tools/maint/gen_struct_info.py")
++            set(EM_GEN_STRUCT_INFO_SCRIPT "${DAWN_EMSCRIPTEN_TOOLCHAIN}/tools/maint/gen_struct_info.py")
++        else()
++            message(FATAL_ERROR "Dawn: Failed to locate file gen_struct_info.py from Emscripten.")
++        endif()
+         set(ARGS
+             ${Python3_EXECUTABLE}
+-            "${DAWN_EMSCRIPTEN_TOOLCHAIN}/tools/maint/gen_struct_info.py"
++            "${EM_GEN_STRUCT_INFO_SCRIPT}"
+             -q
+             "${EM_BUILD_GEN_DIR}/struct_info_webgpu.json"
+             "-I=${EM_BUILD_GEN_DIR}/include"

tools/ci_build/build.py

@@ -488,7 +488,7 @@ def convert_arg_line_to_args(self, arg_line):
     # WebAssembly build
     parser.add_argument("--build_wasm", action="store_true", help="Build for WebAssembly")
     parser.add_argument("--build_wasm_static_lib", action="store_true", help="Build for WebAssembly static library")
-    parser.add_argument("--emsdk_version", default="3.1.59", help="Specify version of emsdk")
+    parser.add_argument("--emsdk_version", default="4.0.3", help="Specify version of emsdk")
 
     parser.add_argument("--enable_wasm_simd", action="store_true", help="Enable WebAssembly SIMD")
     parser.add_argument("--enable_wasm_threads", action="store_true", help="Enable WebAssembly multi-threads support")

tools/ci_build/github/azure-pipelines/templates/linux-wasm-ci.yml

@@ -45,7 +45,7 @@ jobs:
     name: ${{ parameters.PoolName }}
   variables:
     buildArch: x64
-    CommonBuildArgs: '--parallel --config ${{ parameters.BuildConfig }} --skip_submodule_sync --build_wasm --enable_wasm_simd --enable_wasm_threads --use_extensions --cmake_extra_defines onnxruntime_WEBASSEMBLY_DEFAULT_EXTENSION_FLAGS=ON ${{ parameters.ExtraBuildArgs }}'
+    CommonBuildArgs: '--parallel --config ${{ parameters.BuildConfig }} --skip_submodule_sync --build_wasm --enable_wasm_simd --enable_wasm_threads ${{ parameters.ExtraBuildArgs }}'
     runCodesignValidationInjection: false
     TODAY: $[format('{0:dd}{0:MM}{0:yyyy}', pipeline.startTime)]
     ORT_CACHE_DIR: $(Agent.TempDirectory)/ort_ccache
@@ -82,15 +82,15 @@ jobs:
       - script: |
           set -ex
           cd '$(Build.SourcesDirectory)/cmake/external/emsdk'
-          ./emsdk install 3.1.59 ccache-git-emscripten-64bit
-          ./emsdk activate 3.1.59 ccache-git-emscripten-64bit
+          ./emsdk install 4.0.3 ccache-git-emscripten-64bit
+          ./emsdk activate 4.0.3 ccache-git-emscripten-64bit
         displayName: 'emsdk install and activate ccache for emscripten'
   - ${{if eq(parameters.WithCache, false)}}:
       - script: |
           set -ex
           cd '$(Build.SourcesDirectory)/cmake/external/emsdk'
-          ./emsdk install 3.1.59
-          ./emsdk activate 3.1.59
+          ./emsdk install 4.0.3
+          ./emsdk activate 4.0.3
         displayName: 'emsdk install and activate ccache for emscripten'
 
   - template: build-linux-wasm-step.yml