fix runtime failure

cmake/external/onnxruntime_external_deps.cmake

@@ -644,10 +644,8 @@ if (onnxruntime_USE_WEBGPU)
 
     # file "${DAWN_EMSCRIPTEN_TOOLCHAIN}/tools/maint/gen_struct_info.py" is missing from emsdk installation.
     # we should copy if from ${PROJECT_SOURCE_DIR}/patches/
-    file(COPY_FILE
-         "${PROJECT_SOURCE_DIR}/patches/emscripten/tools/maint/gen_struct_info.py"
-         "${DAWN_EMSCRIPTEN_TOOLCHAIN}/tools/maint/gen_struct_info.py"
-         ONLY_IF_DIFFERENT)
+    file(COPY "${PROJECT_SOURCE_DIR}/patches/emscripten/tools/maint/gen_struct_info.py"
+         DESTINATION "${DAWN_EMSCRIPTEN_TOOLCHAIN}/tools/maint/")
   else()
     if (onnxruntime_BUILD_DAWN_MONOLITHIC_LIBRARY)
       set(DAWN_BUILD_MONOLITHIC_LIBRARY ON CACHE BOOL "" FORCE)

cmake/onnxruntime_providers_webgpu.cmake

@@ -32,6 +32,7 @@
     endif()
 
     target_link_libraries(onnxruntime_providers_webgpu PUBLIC emdawnwebgpu_cpp)
+    target_link_options(onnxruntime_providers_webgpu PUBLIC "SHELL:-s ASYNCIFY=1")
   else()
     onnxruntime_add_include_to_target(onnxruntime_providers_webgpu dawn::dawncpp_headers dawn::dawn_headers)
 

cmake/onnxruntime_unittests.cmake

@@ -984,12 +984,15 @@ if (CMAKE_SYSTEM_NAME STREQUAL "Emscripten")
   set_target_properties(onnxruntime_test_all PROPERTIES LINK_DEPENDS ${ONNXRUNTIME_ROOT}/wasm/pre.js)
   set_target_properties(onnxruntime_test_all PROPERTIES LINK_FLAGS "-s STACK_SIZE=5242880 -s INITIAL_MEMORY=536870912 -s ALLOW_MEMORY_GROWTH=1 -s MAXIMUM_MEMORY=4294967296 -s INCOMING_MODULE_JS_API=[preRun,locateFile,arguments,onExit,wasmMemory,buffer,instantiateWasm] --pre-js \"${TEST_SRC_DIR}/wasm/onnxruntime_test_all_adapter.js\" --pre-js \"${ONNXRUNTIME_ROOT}/wasm/pre.js\" -s \"EXPORTED_RUNTIME_METHODS=['FS']\" --preload-file ${CMAKE_CURRENT_BINARY_DIR}/testdata@/testdata -s EXIT_RUNTIME=1")
   if (onnxruntime_ENABLE_WEBASSEMBLY_THREADS)
-    set_property(TARGET onnxruntime_test_all APPEND_STRING PROPERTY LINK_FLAGS " -s DEFAULT_PTHREAD_STACK_SIZE=131072 -s PROXY_TO_PTHREAD=1")
+    set_property(TARGET onnxruntime_test_all APPEND_STRING PROPERTY LINK_FLAGS " -s DEFAULT_PTHREAD_STACK_SIZE=131072 -s PROXY_TO_PTHREAD=1 -s PTHREAD_POOL_SIZE=globalThis.PTHREAD_POOL_SIZE")
   endif()
   if (onnxruntime_USE_JSEP)
     set_target_properties(onnxruntime_test_all PROPERTIES LINK_DEPENDS ${ONNXRUNTIME_ROOT}/wasm/pre-jsep.js)
     set_property(TARGET onnxruntime_test_all APPEND_STRING PROPERTY LINK_FLAGS " --pre-js \"${ONNXRUNTIME_ROOT}/wasm/pre-jsep.js\"")
   endif()
+  if (onnxruntime_USE_WEBGPU)
+    set_property(TARGET onnxruntime_test_all APPEND_STRING PROPERTY LINK_FLAGS " -s ASYNCIFY_STACK_SIZE=131072")
+  endif()
 
   ###
   ### if you want to investigate or debug a test failure in onnxruntime_test_all, replace the following line.

cmake/patches/dawn/dawn.patch

@@ -10,3 +10,109 @@ index de673537d3..c98dc46de7 100644
              "${DAWN_EMSCRIPTEN_TOOLCHAIN}/tools/maint/gen_struct_info.py"
              -q
              "${EM_BUILD_GEN_DIR}/struct_info_webgpu.json"
+diff --git a/third_party/emdawnwebgpu/library_webgpu.js b/third_party/emdawnwebgpu/library_webgpu.js
+index d1835cb090..df03ea2f94 100644
+--- a/third_party/emdawnwebgpu/library_webgpu.js
++++ b/third_party/emdawnwebgpu/library_webgpu.js
+@@ -16,10 +16,19 @@
+     throw new Error("To use Dawn's library_webgpu.js, disable -sUSE_WEBGPU and first include Dawn's library_webgpu_struct_info.js and library_webgpu_enum_tables.js (before library_webgpu.js)");
+   }
+ 
++  if (MEMORY64) {
++    throw new Error("The current implementation of Dawn's library_webgpu.js does not support MEMORY64 yet");
++  }
++
+   // Helper functions for code generation
+   globalThis.gpu = {
+-    convertSentinelToUndefined: function(name) {
+-      return `if (${name} == -1) ${name} = undefined;`;
++    convertSentinelToUndefined: function(name, isPtr = false) {
++      // When `CAN_ADDRESS_2GB` is true, value `-1` is normalized to `0xFFFFFFFF` for pointer.
++      if (CAN_ADDRESS_2GB && isPtr) {
++        return `if (${name} == 0xFFFFFFFF) ${name} = undefined;`;
++      } else {
++        return `if (${name} == -1) ${name} = undefined;`;
++      }
+     },
+ 
+     makeGetBool: function(struct, offset) {
+@@ -700,6 +709,7 @@ var LibraryWebGPU = {
+     {{{ makeSetValue('info', C_STRUCTS.WGPUAdapterInfo.adapterType, 'adapterType', 'i32') }}};
+     {{{ makeSetValue('info', C_STRUCTS.WGPUAdapterInfo.vendorID, '0', 'i32') }}};
+     {{{ makeSetValue('info', C_STRUCTS.WGPUAdapterInfo.deviceID, '0', 'i32') }}};
++    return 1;
+   },
+ 
+   wgpuAdapterGetLimits: (adapterPtr, limitsOutPtr) => {
+@@ -882,7 +892,7 @@ var LibraryWebGPU = {
+ 
+     if (size === 0) warnOnce('getMappedRange size=0 no longer means WGPU_WHOLE_MAP_SIZE');
+ 
+-    {{{ gpu.convertSentinelToUndefined('size') }}}
++    {{{ gpu.convertSentinelToUndefined('size', true) }}}
+ 
+     var mapped;
+     try {
+@@ -909,7 +919,7 @@ var LibraryWebGPU = {
+ 
+     if (size === 0) warnOnce('getMappedRange size=0 no longer means WGPU_WHOLE_MAP_SIZE');
+ 
+-    {{{ gpu.convertSentinelToUndefined('size') }}}
++    {{{ gpu.convertSentinelToUndefined('size', true) }}}
+ 
+     var mapped;
+     try {
+@@ -950,7 +960,7 @@ var LibraryWebGPU = {
+     var buffer = WebGPU.getJsObject(bufferPtr);
+     WebGPU.Internals.bufferOnUnmaps[bufferPtr] = [];
+ 
+-    {{{ gpu.convertSentinelToUndefined('size') }}}
++    {{{ gpu.convertSentinelToUndefined('size', true) }}}
+ 
+     {{{ runtimeKeepalivePush() }}}
+     WebGPU.Internals.futureInsert(futureId, buffer.mapAsync(mode, offset, size).then(() => {
+@@ -1145,7 +1155,7 @@ var LibraryWebGPU = {
+ 
+   wgpuCommandEncoderClearBuffer: (encoderPtr, bufferPtr, offset, size) => {
+     var commandEncoder = WebGPU.getJsObject(encoderPtr);
+-    {{{ gpu.convertSentinelToUndefined('size') }}}
++    {{{ gpu.convertSentinelToUndefined('size', true) }}}
+ 
+     var buffer = WebGPU.getJsObject(bufferPtr);
+     commandEncoder.clearBuffer(buffer, offset, size);
+@@ -2103,7 +2113,7 @@ var LibraryWebGPU = {
+   wgpuRenderBundleEncoderSetIndexBuffer: (passPtr, bufferPtr, format, offset, size) => {
+     var pass = WebGPU.getJsObject(passPtr);
+     var buffer = WebGPU.getJsObject(bufferPtr);
+-    {{{ gpu.convertSentinelToUndefined('size') }}}
++    {{{ gpu.convertSentinelToUndefined('size', true) }}}
+     pass.setIndexBuffer(buffer, WebGPU.IndexFormat[format], offset, size);
+   },
+ 
+@@ -2116,7 +2126,7 @@ var LibraryWebGPU = {
+   wgpuRenderBundleEncoderSetVertexBuffer: (passPtr, slot, bufferPtr, offset, size) => {
+     var pass = WebGPU.getJsObject(passPtr);
+     var buffer = WebGPU.getJsObject(bufferPtr);
+-    {{{ gpu.convertSentinelToUndefined('size') }}}
++    {{{ gpu.convertSentinelToUndefined('size', true) }}}
+     pass.setVertexBuffer(slot, buffer, offset, size);
+   },
+ 
+@@ -2211,7 +2221,7 @@ var LibraryWebGPU = {
+   wgpuRenderPassEncoderSetIndexBuffer: (passPtr, bufferPtr, format, offset, size) => {
+     var pass = WebGPU.getJsObject(passPtr);
+     var buffer = WebGPU.getJsObject(bufferPtr);
+-    {{{ gpu.convertSentinelToUndefined('size') }}}
++    {{{ gpu.convertSentinelToUndefined('size', true) }}}
+     pass.setIndexBuffer(buffer, WebGPU.IndexFormat[format], offset, size);
+   },
+ 
+@@ -2234,7 +2244,7 @@ var LibraryWebGPU = {
+   wgpuRenderPassEncoderSetVertexBuffer: (passPtr, slot, bufferPtr, offset, size) => {
+     var pass = WebGPU.getJsObject(passPtr);
+     var buffer = WebGPU.getJsObject(bufferPtr);
+-    {{{ gpu.convertSentinelToUndefined('size') }}}
++    {{{ gpu.convertSentinelToUndefined('size', true) }}}
+     pass.setVertexBuffer(slot, buffer, offset, size);
+   },
+ 

js/build_webgpu.bat

@@ -17,7 +17,8 @@ set BUILD_DIR=%ROOT%build_webgpu
 :arg1
 if ["%~1"]==["d"] (
     set CONFIG=Debug
-    set CONFIG_EXTRA_FLAG=--enable_wasm_profiling --cmake_extra_defines onnxruntime_ENABLE_WEBASSEMBLY_OUTPUT_OPTIMIZED_MODEL=1
+    set CONFIG_EXTRA_FLAG=--enable_wasm_profiling --wasm_run_tests_in_browser
+    @rem --cmake_extra_defines onnxruntime_ENABLE_WEBASSEMBLY_OUTPUT_OPTIMIZED_MODEL=1
     @rem --enable_wasm_debug_info
     goto :arg2
 )

onnxruntime/core/providers/webgpu/webgpu_context.cc

@@ -696,9 +696,13 @@ WebGpuContext& WebGpuContextFactory::CreateContext(const WebGpuContextConfig& co
 #endif
 
       // Step.2 - Create wgpu::Instance
+#if !defined(__wasm__)
       wgpu::InstanceDescriptor instance_desc{};
       instance_desc.features.timedWaitAnyEnable = true;
       default_instance_ = wgpu::CreateInstance(&instance_desc);
+#else
+      default_instance_ = wgpu::CreateInstance(nullptr);
+#endif
 
       ORT_ENFORCE(default_instance_ != nullptr, "Failed to create wgpu::Instance.");
     });

onnxruntime/test/wasm/karma.conf.js

@@ -7,7 +7,7 @@ const THREADS = process.argv.includes('--wasm-threads');
 
 const files = [];
 const proxies = {};
-const chromeFlags = ['--no-sandbox'];
+const chromeFlags = ['--no-sandbox', '--js-flags="--max_old_space_size=8192"'];
 for (const arg of process.argv) {
   if (arg.startsWith('--entry=')) {
     const entry = arg.substring('--entry='.length);
@@ -38,6 +38,10 @@ const gtestReporter = {'reporter:gtest': ['type', function() {
   };
 }]};
 
+// In Node.js v16 and below, 'localhost' is using IPv4, so need to listen to '0.0.0.0'
+// In Node.js v17+, 'localhost' is using IPv6, so need to listen to '::'
+const listenAddress = Number.parseInt(process.versions.node.split('.')[0]) >= 17 ? '::' : '0.0.0.0';
+
 module.exports = function(config) {
     config.set({
       basePath: '.',
@@ -55,11 +59,12 @@ module.exports = function(config) {
       client: {
         captureConsole: true,
         // Pass the gtest flags to the test runner
-        args: process.argv.filter(arg => arg.startsWith('--gtest_'))
+        args: process.argv.filter(arg => arg.startsWith('--gtest_') || arg.startsWith('--wasm-threads='))
       },
       browserDisconnectTimeout: 600000,
       // allow running tests for 30 minutes
       browserNoActivityTimeout: 30 * 60 * 1000,
+      listenAddress,
       customLaunchers: {
         ChromeTest: {
           base: 'ChromeCanary',

onnxruntime/test/wasm/onnxruntime_test_all_adapter.js

@@ -15,6 +15,7 @@
     if (typeof process !== 'undefined') {
         // In Node.js
         args = process.argv;
+        globalThis.PTHREAD_POOL_SIZE = Math.min(8, require('os').cpus().length);
         onTestComplete = function () {
             if (gtestOutputFilepath) {
                 require('fs').writeFileSync(gtestOutputFilepath, gtestOutputFiledata);
@@ -23,6 +24,7 @@
     } else if (typeof __karma__ !== 'undefined') {
         // In browser (launched by karma)
         args = __karma__.config.args;
+        globalThis.PTHREAD_POOL_SIZE = Math.min(8, navigator.hardwareConcurrency);
         onTestComplete = function (exitCode) {
             __karma__.result({
                 id: '',
@@ -42,10 +44,14 @@
     // check for flag "--gtest_output=xml:"
     const argGtestOutputPrefix = '--gtest_output=xml:';
 
+    // check for flag "--wasm-threads="
+    const argWasmThreads = '--wasm-threads=';
+
     for (const arg of args) {
         if (arg.startsWith(argGtestOutputPrefix)) {
             gtestOutputFilepath = arg.substring(argGtestOutputPrefix.length);
-            break;
+        } else if (arg.startsWith(argWasmThreads)) {
+            globalThis.PTHREAD_POOL_SIZE = Number.parseInt(arg.substring(argWasmThreads.length));
         }
     }