[NativeAOT-LLVM] Wasm managed threads - build new nupkg

eng/native/gen-buildsys.cmd

@@ -65,7 +65,7 @@ if /i "%__Arch%" == "wasm" (
         set "WASI_SDK_PATH=!WASI_SDK_PATH:\=/!"
         if not "!WASI_SDK_PATH:~-1!" == "/" set "WASI_SDK_PATH=!WASI_SDK_PATH!/"
         set __CmakeGenerator=Ninja
-        set __ExtraCmakeParams=%__ExtraCmakeParams% -DCLR_CMAKE_TARGET_OS=wasi -DCLR_CMAKE_TARGET_ARCH=wasm "-DWASI_SDK_PREFIX=!WASI_SDK_PATH!" "-DCMAKE_TOOLCHAIN_FILE=!WASI_SDK_PATH!/share/cmake/wasi-sdk-p2.cmake" "-DCMAKE_SYSROOT=!WASI_SDK_PATH!share/wasi-sysroot" "-DCMAKE_CROSSCOMPILING_EMULATOR=node --experimental-wasm-bigint --experimental-wasi-unstable-preview1"
+        set __ExtraCmakeParams=%__ExtraCmakeParams% -DCLR_CMAKE_TARGET_OS=wasi -DCLR_CMAKE_TARGET_ARCH=wasm "-DWASI_SDK_PREFIX=!WASI_SDK_PATH!" "-DCMAKE_TOOLCHAIN_FILE=!WASI_SDK_PATH!/share/cmake/wasi-sdk-p2.cmake" "-DCMAKE_SYSROOT=!WASI_SDK_PATH!share/wasi-sysroot" "-DCMAKE_CROSSCOMPILING_EMULATOR=node --experimental-wasm-bigint --experimental-wasi-unstable-preview1 -D_WASI_EMULATED_PTHREAD"
     )
 ) else (
     set __ExtraCmakeParams=%__ExtraCmakeParams%  "-DCMAKE_SYSTEM_VERSION=10.0"

eng/pipelines/common/global-build-job.yml

@@ -219,7 +219,7 @@ jobs:
       - script: pwsh $(Build.SourcesDirectory)/eng/pipelines/runtimelab/install-nodejs.ps1 $(Build.SourcesDirectory)/wasm-tools
         displayName: Install NodeJS
 
-    - ${{ if and(eq(parameters.runtimeFlavor, 'coreclr'), or(eq(parameters.platform, 'wasi_wasm_win'), eq(parameters.platform, 'wasi_wasm_linux_x64_naot_llvm'))) }}:
+    - ${{ if and(eq(parameters.runtimeFlavor, 'coreclr'), in(parameters.platform, 'wasi_wasm_win', 'wasi_wasm_linux_x64_naot_llvm', 'wasi_multithread_wasm_win')) }}:
       # Install Wasi Wasm dependencies: wasi-sdk, wasmtime
       - script: pwsh $(Build.SourcesDirectory)/eng/pipelines/runtimelab/install-wasi-sdk.ps1 -CI -InstallDir $(Build.SourcesDirectory)/wasm-tools
         displayName: Install wasi-sdk
@@ -228,7 +228,7 @@ jobs:
       - script: pwsh $(Build.SourcesDirectory)/eng/pipelines/runtimelab/install-jco.ps1 $(Build.SourcesDirectory)
         displayName: Install Jco
 
-    - ${{ if or(eq(parameters.platform, 'browser_wasm_win'), and(eq(parameters.platform, 'wasi_wasm_win'), not(eq(parameters.runtimeFlavor, 'coreclr')))) }}:
+    - ${{ if or(eq(parameters.platform, 'browser_wasm_win'), and(in(parameters.platform, 'wasi_wasm_win', 'wasi_multithread_wasm_win'), not(eq(parameters.runtimeFlavor, 'coreclr')))) }}:
       # Update machine certs
       - task: PowerShell@2
         displayName: Update machine certs

eng/pipelines/common/platform-matrix.yml

@@ -500,6 +500,44 @@ jobs:
         buildConfig: ${{ parameters.buildConfig }}
         ${{ insert }}: ${{ parameters.jobParameters }}
 
+- ${{ if containsValue(parameters.platforms, 'wasi_multithread_wasm_win') }}:
+  - template: xplat-setup.yml
+    parameters:
+      jobTemplate: ${{ parameters.jobTemplate }}
+      helixQueuesTemplate: ${{ parameters.helixQueuesTemplate }}
+      variables: ${{ parameters.variables }}
+      osGroup: wasi
+      osSubgroup: multithread
+      archType: wasm
+      targetRid: wasi-wasm
+      platform: wasi_multithread_wasm_win
+      shouldContinueOnError: ${{ parameters.shouldContinueOnError }}
+      jobParameters:
+        hostedOs: windows
+        runtimeFlavor: ${{ parameters.runtimeFlavor }}
+        stagedBuild: ${{ parameters.stagedBuild }}
+        buildConfig: ${{ parameters.buildConfig }}
+        ${{ insert }}: ${{ parameters.jobParameters }}
+
+- ${{ if containsValue(parameters.platforms, 'browser_multithread_wasm_win') }}:
+  - template: xplat-setup.yml
+    parameters:
+      jobTemplate: ${{ parameters.jobTemplate }}
+      helixQueuesTemplate: ${{ parameters.helixQueuesTemplate }}
+      variables: ${{ parameters.variables }}
+      osGroup: browser
+      osSubgroup: multithread
+      archType: wasm
+      targetRid: browser-wasm
+      platform: browser_multithread_wasm_win
+      shouldContinueOnError: ${{ parameters.shouldContinueOnError }}
+      jobParameters:
+        hostedOs: windows
+        runtimeFlavor: ${{ parameters.runtimeFlavor }}
+        stagedBuild: ${{ parameters.stagedBuild }}
+        buildConfig: ${{ parameters.buildConfig }}
+        ${{ insert }}: ${{ parameters.jobParameters }}
+
 # Browser WebAssembly Linux X64 for NAOT-LLVM
 # Use a different name to differentiate from upstream as we need the -sanitizer image to get build tools that are not present in the stock browser_wasm image.
 

eng/pipelines/libraries/helix-queues-setup.yml

@@ -175,7 +175,7 @@ jobs:
       - (Ubuntu.2204.Amd64)[email protected]/dotnet-buildtools/prereqs:ubuntu-22.04-helix-webassembly
 
     # Browser WebAssembly windows
-    - ${{ if in(parameters.platform, 'browser_wasm_win', 'wasi_wasm_win') }}:
+    - ${{ if in(parameters.platform, 'browser_wasm_win', 'wasi_wasm_win', 'browser_multithread_wasm_win', 'wasi_multithread_wasm_win') }}:
       - (Windows.Amd64.Server2022.Open)[email protected]/dotnet-buildtools/prereqs:windowsservercore-ltsc2022-helix-webassembly
       - (Windows.Server2025.Amd64.Open)[email protected]/dotnet-buildtools/prereqs:windowsservercore-ltsc2025-helix-webassembly-amd64
 

eng/pipelines/runtimelab.yml

@@ -71,6 +71,25 @@ extends:
                   parameters:
                     librariesConfiguration: Debug
 
+        #
+        # Build and test Wasm Debug multithreaded libraries and Debug runtime
+        #
+        - template: /eng/pipelines/common/platform-matrix.yml
+          parameters:
+            jobTemplate: /eng/pipelines/common/global-build-job.yml
+            helixQueuesTemplate: /eng/pipelines/coreclr/templates/helix-queues-setup.yml
+            buildConfig: debug
+            platforms:
+            - browser_multithread_wasm_win
+            - wasi_multithread_wasm_win
+            jobParameters:
+              timeoutInMinutes: 300
+              buildArgs: -s clr.aot+libs -c debug -rc $(_BuildConfig) -cmakeargs -DCLR_CMAKE_TARGET_OS_SUBGROUP=multithread '/p:WasmEnableThreads=true'
+              postBuildSteps:
+                - template: /eng/pipelines/runtimelab/runtimelab-post-build-steps.yml
+                  parameters:
+                    librariesConfiguration: Debug
+
         #
         # Build and test with Debug libraries and Checked runtime
         #

eng/pipelines/runtimelab/runtimelab-post-build-steps.yml

@@ -37,6 +37,14 @@ steps:
       - script: |
           $(Build.SourcesDirectory)/src/tests/build$(scriptExt) nativeaot $(buildConfigUpper) ${{ parameters.archType }} $(crossArg) $(_officialBuildParameter) ci wasi tree nativeaot /p:LibrariesConfiguration=${{ parameters.librariesConfiguration }}
         displayName: Build WebAssembly tests
+    - ${{ elseif eq(parameters.platform, 'browser_multithread_wasm_win') }}:
+      - script: |
+          $(Build.SourcesDirectory)/src/tests/build$(scriptExt) nativeaot $(buildConfigUpper) ${{ parameters.archType }} $(crossArg) $(_officialBuildParameter) ci browser tree nativeaot /p:LibrariesConfiguration=${{ parameters.librariesConfiguration }} /p:FeatureWasmManagedThreads=true
+        displayName: Build WebAssembly browser Mutlithread tests
+    - ${{ elseif eq(parameters.platform, 'wasi_multithread_wasm_win') }}:
+      - script: |
+          $(Build.SourcesDirectory)/src/tests/build$(scriptExt) nativeaot $(buildConfigUpper) ${{ parameters.archType }} $(crossArg) $(_officialBuildParameter) ci wasi tree nativeaot /p:LibrariesConfiguration=${{ parameters.librariesConfiguration }} /p:FeatureWasmManagedThreads=true
+        displayName: Build WebAssembly wasi Mutlithread tests
     - ${{ elseif eq(parameters.platform, 'Browser_wasm_linux_x64_naot_llvm') }}:
       - script: |
           source $(Build.SourcesDirectory)/wasm-tools/emsdk/emsdk_env.sh
@@ -59,6 +67,9 @@ steps:
           call $(Build.SourcesDirectory)\wasm-tools\emsdk\emsdk_env
           $(Build.SourcesDirectory)/src/tests/run$(scriptExt) runnativeaottests $(buildConfigUpper) ${{ parameters.archType }} ${{ parameters.osGroup }}
         displayName: Run WebAssembly tests in single file mode
+    - ${{ elseif in(parameters.platform, 'browser_multithread_wasm_win', 'wasi_multithread_wasm_win') }}:
+      - script: $(Build.SourcesDirectory)/src/tests/run$(scriptExt) runnativeaottests $(buildConfigUpper) ${{ parameters.archType }} ${{ parameters.osGroup }}
+        displayName: Run WebAssembly multithread tests in single file mode
     - ${{ elseif eq(parameters.osGroup, 'windows') }}:
       - script: $(Build.SourcesDirectory)/src/tests/run$(scriptExt) runnativeaottests $(buildConfigUpper) ${{ parameters.archType }} ${{ parameters.osGroup }}
         displayName: Run tests in single file mode

src/coreclr/nativeaot/CMakeLists.txt

@@ -30,8 +30,18 @@ endif (CLR_CMAKE_HOST_UNIX)
 
 if(CLR_CMAKE_TARGET_OS STREQUAL wasi)
   add_definitions(-DTARGET_UNIX)
+
+  if(CLR_CMAKE_TARGET_OS_SUBGROUP STREQUAL multithread)
+    add_definitions(-DFEATURE_WASM_MANAGED_THREADS)
+  endif(CLR_CMAKE_TARGET_OS_SUBGROUP STREQUAL multithread)
 endif(CLR_CMAKE_TARGET_OS STREQUAL wasi)
 
+if((CLR_CMAKE_TARGET_OS STREQUAL wasi OR CLR_CMAKE_TARGET_OS STREQUAL browser) 
+  AND CLR_CMAKE_TARGET_OS_SUBGROUP STREQUAL multithread)
+  add_definitions(-DFEATURE_WASM_MANAGED_THREADS)
+endif((CLR_CMAKE_TARGET_OS STREQUAL wasi OR CLR_CMAKE_TARGET_OS STREQUAL browser) 
+  AND CLR_CMAKE_TARGET_OS_SUBGROUP STREQUAL multithread)
+
 if(CLR_CMAKE_TARGET_ANDROID)
     add_definitions(-DFEATURE_EMULATED_TLS)
 endif()

src/coreclr/nativeaot/Runtime/unix/PalRedhawkUnix.cpp

@@ -1151,6 +1151,8 @@ REDHAWK_PALEXPORT bool PalGetMaximumStackBounds(_Out_ void** ppStackLowOut, _Out
     {
 #if defined(HOST_WASM) && !defined(FEATURE_WASM_MANAGED_THREADS)
         PalGetMaximumStackBounds_SingleThreadedWasm(&pStackLowOut, &pStackHighOut);
+#elif defined(HOST_WASM) && defined(FEATURE_WASM_MANAGED_THREADS)
+        PalGetMaximumStackBounds_MultiThreadedWasm(&pStackLowOut, &pStackHighOut);
 #elif defined(__APPLE__)
         // This is a Mac specific method
         pStackHighOut = pthread_get_stackaddr_np(pthread_self());

src/coreclr/nativeaot/Runtime/wasm/PalRedhawkWasm.cpp

@@ -191,6 +191,182 @@ extern "C" int __cxa_thread_atexit(Dtor dtor, void* obj, void*)
 #endif // TARGET_WASI
 #endif // !FEATURE_WASM_MANAGED_THREADS
 
+// TODO-LLVM: For now, a copy of the single threaded implementation.
+#ifdef FEATURE_WASM_MANAGED_THREADS
+int __cxa_thread_atexit(void (*func)(), void *obj, void *dso_symbol)
+{}
+
+//
+// Note that we return the native stack bounds here, not shadow stack ones. Currently this functionality is mainly
+// used for RuntimeHelpers.TryEnsureSufficientExecutionStack, and we do use the native stack in codegen, so this
+// is an acceptable approximation.
+//
+extern "C" unsigned char __stack_low;
+extern "C" unsigned char __stack_high;
+void PalGetMaximumStackBounds_MultiThreadedWasm(void** ppStackLowOut, void** ppStackHighOut)
+{
+    // See https://github.com/emscripten-core/emscripten/pull/18057 and https://reviews.llvm.org/D135910.
+    unsigned char* pStackLow = &__stack_low;
+    unsigned char* pStackHigh = &__stack_high;
+
+    // Sanity check that we have the expected memory layout.
+    ASSERT((pStackHigh - pStackLow) >= 64 * 1024);
+    if (pStackLow >= pStackHigh)
+    {
+        PalPrintFatalError("\nFatal error. Unexpected stack layout.\n");
+        RhFailFast();
+    }
+
+    *ppStackLowOut = pStackLow;
+    *ppStackHighOut = pStackHigh;
+}
+
+#ifdef TARGET_WASI
+// No-op stubs that assume a single-threaded environment.
+int pthread_mutex_init(pthread_mutex_t *, const pthread_mutexattr_t *)
+{
+    return 0;
+}
+
+int pthread_mutexattr_init(pthread_mutexattr_t *)
+{
+    return 0;
+}
+
+int pthread_mutexattr_settype(pthread_mutexattr_t *, int)
+{
+    return 0;
+}
+
+int pthread_mutex_destroy(pthread_mutex_t *)
+{
+    return 0;
+}
+
+int pthread_mutexattr_destroy(pthread_mutexattr_t *)
+{
+    return 0;
+}
+
+int pthread_cond_init(pthread_cond_t *, const pthread_condattr_t *)
+{
+    return 0;
+}
+
+int pthread_cond_destroy(pthread_cond_t *)
+{
+    return 0;
+}
+
+int pthread_cond_wait(pthread_cond_t *, pthread_mutex_t *)
+{
+    return 0;
+}
+
+int pthread_cond_timedwait(pthread_cond_t *, pthread_mutex_t *, const struct timespec *)
+{
+    return 0;
+}
+
+int pthread_condattr_init(pthread_condattr_t *)
+{
+    return 0;
+}
+
+int pthread_mutex_lock(pthread_mutex_t *)
+{
+    return 0;
+}
+
+int pthread_mutex_unlock(pthread_mutex_t *)
+{
+    return 0;
+}
+
+pthread_t pthread_self(void)
+{
+    return (pthread_t)0;
+}
+
+int pthread_equal(pthread_t, pthread_t)
+{
+    return 1; // only one thread
+}
+
+int pthread_attr_init(pthread_attr_t *)
+{
+    return 0;
+}
+
+int pthread_attr_destroy(pthread_attr_t *)
+{
+    return 0;
+}
+
+int pthread_condattr_destroy(pthread_condattr_t *)
+{
+    return 0;
+}
+
+int pthread_cond_broadcast(pthread_cond_t *)
+{
+    return 0;
+}
+
+int pthread_attr_setdetachstate(pthread_attr_t *, int)
+{
+    return 0;
+}
+
+using Dtor = void(*)(void*);
+
+// Due to a bug in the toolchain, we have to provide an implementation of thread-local destruction.
+// Since this is the single-threaded case, we simply delegate to the static destruction mechanism.
+// Reference: https://github.com/llvm/llvm-project/blob/main/libcxxabi/src/cxa_thread_atexit.cpp.
+//
+extern "C" int __cxa_thread_atexit(Dtor dtor, void* obj, void*)
+{
+    struct DtorList
+    {
+        Dtor dtor;
+        void* obj;
+        DtorList* next;
+    };
+
+    struct DtorsManager
+    {
+        DtorList* m_dtors = nullptr;
+
+        ~DtorsManager()
+        {
+            while (DtorList* head = m_dtors)
+            {
+                m_dtors = head->next;
+                head->dtor(head->obj);
+                free(head);
+            }
+        }
+    };
+
+    // The linked list of "thread-local" destructors to run.
+    static DtorsManager s_dtorsManager;
+
+    DtorList* head = static_cast<DtorList*>(malloc(sizeof(DtorList)));
+    if (head == nullptr)
+    {
+        return -1;
+    }
+
+    head->dtor = dtor;
+    head->obj = obj;
+    head->next = s_dtorsManager.m_dtors;
+    s_dtorsManager.m_dtors = head;
+
+    return 0;
+}
+#endif // TARGET_WASI
+#endif // FEATURE_WASM_MANAGED_THREADS
+
 // Recall that WASM's model is extremely simple: we have one linear memory, which can only be grown, in chunks
 // of 64K pages. Thus, "mmap"/"munmap" fundamentally cannot be faithfully recreated and the Unix emulators we
 // layer on top of (Emscripten/WASI libc) reflect this by not supporting the scenario. Fortunately, the current

src/coreclr/nativeaot/Runtime/wasm/PalRedhawkWasm.h

@@ -1,6 +1,9 @@
 // Licensed to the .NET Foundation under one or more agreements.
 // The .NET Foundation licenses this file to you under the MIT license.
 
-#ifndef FEATURE_WASM_MANAGED_THREADS
+#ifdef FEATURE_WASM_MANAGED_THREADS
+int __cxa_thread_atexit(void (*func)(), void *obj, void *dso_symbol);
+void PalGetMaximumStackBounds_MultiThreadedWasm(void** ppStackLowOut, void** ppStackHighOut);
+#else
 void PalGetMaximumStackBounds_SingleThreadedWasm(void** ppStackLowOut, void** ppStackHighOut);
 #endif // !FEATURE_WASM_MANAGED_THREADS

src/coreclr/nativeaot/System.Private.CoreLib/src/System.Private.CoreLib.csproj

@@ -35,6 +35,10 @@
   <PropertyGroup Condition="'$(TargetsSingleThreadedWasm)' != 'true'">
     <FeaturePortableThreadPool>true</FeaturePortableThreadPool>
     <FeaturePortableTimer>true</FeaturePortableTimer>
+    <FeatureWasmManagedThreads Condition="'$(WasmEnableThreads)' == 'true'">true</FeatureWasmManagedThreads>
+    <FeatureWasmPerfTracing Condition="('$(TargetsBrowser)' == 'true' or '$(TargetsWasi)' == 'true') and ('$(WasmEnableThreads)' == 'true')">true</FeatureWasmPerfTracing>
+    <DefineConstants Condition="'$(FeatureWasmManagedThreads)' == 'true'">$(DefineConstants);FEATURE_WASM_MANAGED_THREADS</DefineConstants>
+    <DefineConstants Condition="'$(FeatureWasmPerfTracing)' == 'true'">$(DefineConstants);FEATURE_WASM_PERFTRACING</DefineConstants>
   </PropertyGroup>
   <PropertyGroup>
     <FeatureHardwareIntrinsics>true</FeatureHardwareIntrinsics>
@@ -347,6 +351,8 @@
     <Compile Include="System\Diagnostics\StackFrame.NativeAot.Wasm.cs" />
     <Compile Include="System\Diagnostics\StackFrame.NativeAot.Wasi.cs" Condition="'$(TargetsWasi)' == 'true'" />
     <Compile Include="System\Diagnostics\StackFrame.NativeAot.Browser.cs" Condition="'$(TargetsBrowser)' == 'true'" />
+    <Compile Include="System\Threading\PortableThreadPool.NativeAot.Browser.cs" Condition="'$(TargetsBrowser)' == 'true' and '$(FeatureWasmManagedThreads)' == 'true'" />
+    <Compile Include="System\Threading\ThreadPool.NativeAot.Browser.cs" Condition="'$(TargetsBrowser)' == 'true' and '$(FeatureWasmManagedThreads)' == 'true'" />
   </ItemGroup>
 
   <!-- WASM-specific things. Keep in sync with Mono. -->

src/coreclr/nativeaot/System.Private.CoreLib/src/System/Threading/Monitor.NativeAot.cs

@@ -143,7 +143,9 @@ public static bool IsEntered(object obj)
 
         #region Public Wait/Pulse methods
 
+#if !FEATURE_WASM_MANAGED_THREADS
         [UnsupportedOSPlatform("browser")]
+#endif
         public static bool Wait(object obj, int millisecondsTimeout)
         {
             return GetCondition(obj).Wait(millisecondsTimeout, obj);