diff --git a/.github/workflows/macos.yml b/.github/workflows/macos.yml
index 160fe0fafaf3..dffe12a3e23b 100644
--- a/.github/workflows/macos.yml
+++ b/.github/workflows/macos.yml
@@ -196,3 +196,135 @@ jobs:
       run: |
         printf "[Processor]\nThreadCount=1\n" > build-x86_64/tests/SwiftShader.ini
         cd build-x86_64 && ctest --output-on-failure -j 4
+
+  test-arm:
+    runs-on: macos-latest
+    env:
+      OPENMP_VERSION: '18.1.2'
+      OPENMP_CMAKE_OPTIONS: |
+        -DCMAKE_TOOLCHAIN_FILE=../../toolchains/ios.toolchain.cmake \
+        -DDEPLOYMENT_TARGET=$MAC_DEPLOYMENT_TARGET \
+        -DENABLE_BITCODE=$ENABLE_BITCODE \
+        -DENABLE_ARC=$ENABLE_ARC \
+        -DENABLE_VISIBILITY=$ENABLE_VISIBILITY \
+        -DCMAKE_INSTALL_PREFIX=install \
+        -DCMAKE_BUILD_TYPE=Release \
+        -DPERL_EXECUTABLE=/usr/bin/perl \
+        -DLIBOMP_ENABLE_SHARED=OFF \
+        -DLIBOMP_OMPT_SUPPORT=OFF \
+        -DLIBOMP_USE_HWLOC=OFF \
+
+      MOLTENVK_VERSION: 'v1.2.8'
+      NCNN_CMAKE_OPTIONS: |
+        -DCMAKE_TOOLCHAIN_FILE=../toolchains/ios.toolchain.cmake \
+        -DDEPLOYMENT_TARGET=$MAC_DEPLOYMENT_TARGET \
+        -DENABLE_BITCODE=$ENABLE_BITCODE \
+        -DENABLE_ARC=$ENABLE_ARC \
+        -DENABLE_VISIBILITY=$ENABLE_VISIBILITY \
+        -DCMAKE_INSTALL_PREFIX=install \
+        -DCMAKE_BUILD_TYPE=Release \
+        -DOpenMP_C_FLAGS="-Xclang -fopenmp" -DOpenMP_CXX_FLAGS="-Xclang -fopenmp" \
+        -DOpenMP_C_LIB_NAMES="libomp" -DOpenMP_CXX_LIB_NAMES="libomp" \
+        -DOpenMP_libomp_LIBRARY="libomp.a" \
+        -DNCNN_VULKAN=ON \
+
+    steps:
+    - uses: actions/checkout@v4
+      with:
+        submodules: true
+
+    - name: cache-openmp
+      id: cache-openmp
+      uses: actions/cache@v4
+      with:
+        path: openmp-install
+        key: openmp-macos-install-20240403
+    - name: openmp
+      if: steps.cache-openmp.outputs.cache-hit != 'true'
+      run: |
+        wget https://github.com/llvm/llvm-project/releases/download/llvmorg-${{ env.OPENMP_VERSION }}/cmake-${{ env.OPENMP_VERSION }}.src.tar.xz
+        tar -xf cmake-${{ env.OPENMP_VERSION }}.src.tar.xz
+        wget https://github.com/llvm/llvm-project/releases/download/llvmorg-${{ env.OPENMP_VERSION }}/openmp-${{ env.OPENMP_VERSION }}.src.tar.xz
+        tar -xf openmp-${{ env.OPENMP_VERSION }}.src.tar.xz
+        mv cmake-${{ env.OPENMP_VERSION }}.src/Modules/* openmp-${{ env.OPENMP_VERSION }}.src/cmake/
+        cd openmp-${{ env.OPENMP_VERSION }}.src
+        wget https://github.com/nihui/llvm-project/commit/ef8c35bcf5d9cfdb0764ffde6a63c04ec715bc37.patch
+        patch -p2 -i ef8c35bcf5d9cfdb0764ffde6a63c04ec715bc37.patch
+        wget https://github.com/nihui/llvm-project/commit/5c12711f9a21f41bea70566bf15a4026804d6b20.patch
+        patch -p2 -i 5c12711f9a21f41bea70566bf15a4026804d6b20.patch
+    - name: openmp-x86_64
+      if: steps.cache-openmp.outputs.cache-hit != 'true'
+      run: |
+        cd openmp-${{ env.OPENMP_VERSION }}.src
+        mkdir -p build-x86_64 && cd build-x86_64
+        cmake ${{ env.OPENMP_CMAKE_OPTIONS }} -DPLATFORM=MAC -DARCHS="x86_64" ..
+        cmake --build . -j 4
+        cmake --build . --target install
+    - name: openmp-arm64
+      if: steps.cache-openmp.outputs.cache-hit != 'true'
+      run: |
+        cd openmp-${{ env.OPENMP_VERSION }}.src
+        mkdir -p build-arm64 && cd build-arm64
+        cmake ${{ env.OPENMP_CMAKE_OPTIONS }} -DPLATFORM=MAC_ARM64 -DARCHS="arm64" ..
+        cmake --build . -j 4
+        cmake --build . --target install
+    - name: openmp-merge-fat-library
+      if: steps.cache-openmp.outputs.cache-hit != 'true'
+      run: |
+        mkdir -p $GITHUB_WORKSPACE/openmp-install
+        cp -a openmp-${{ env.OPENMP_VERSION }}.src/build-x86_64/install/include $GITHUB_WORKSPACE/openmp-install
+        mkdir -p $GITHUB_WORKSPACE/openmp-install/lib
+        lipo -create \
+            openmp-${{ env.OPENMP_VERSION }}.src/build-x86_64/install/lib/libomp.a \
+            openmp-${{ env.OPENMP_VERSION }}.src/build-arm64/install/lib/libomp.a \
+            -o $GITHUB_WORKSPACE/openmp-install/lib/libomp.a
+
+    - name: install-openmp
+      run: |
+        sudo cp $GITHUB_WORKSPACE/openmp-install/include/* $DEVELOPER_DIR/Platforms/MacOSX.platform/Developer/SDKs/MacOSX.sdk/usr/include
+        sudo cp $GITHUB_WORKSPACE/openmp-install/lib/libomp.a $DEVELOPER_DIR/Platforms/MacOSX.platform/Developer/SDKs/MacOSX.sdk/usr/lib
+
+    - name: cache-swiftshader
+      id: cache-swiftshader
+      uses: actions/cache@v4
+      with:
+        path: swiftshader-install
+        key: swiftshader-macos-install-20240622
+    - name: checkout-swiftshader
+      if: steps.cache-swiftshader.outputs.cache-hit != 'true'
+      uses: actions/checkout@v4
+      with:
+        repository: google/swiftshader
+        path: swiftshader
+        ref: de870ac7518fe2b6bb651ecc22fc36647cf7b986
+    - name: checkout-swiftshader-submodules
+      if: steps.cache-swiftshader.outputs.cache-hit != 'true'
+      run: |
+        cd swiftshader
+        git -c submodule."third_party/git-hooks".update=none submodule update --init --recursive
+    - name: swiftshader
+      if: steps.cache-swiftshader.outputs.cache-hit != 'true'
+      run: |
+        cd swiftshader
+        mkdir -p build; cd build
+        cmake -DCMAKE_INSTALL_PREFIX=install -DSWIFTSHADER_BUILD_EGL=FALSE -DSWIFTSHADER_BUILD_GLESv2=FALSE -DSWIFTSHADER_BUILD_GLES_CM=FALSE -DSWIFTSHADER_BUILD_VULKAN=TRUE -DSWIFTSHADER_BUILD_PVR=FALSE -DSWIFTSHADER_BUILD_TESTS=FALSE -DSWIFTSHADER_ENABLE_ASTC=FALSE -DSWIFTSHADER_WARNINGS_AS_ERRORS=FALSE -DREACTOR_BACKEND=Subzero -DREACTOR_DEFAULT_OPT_LEVEL=Default -DCMAKE_BUILD_TYPE=Release ..
+        cmake --build . -j 3
+        mkdir $GITHUB_WORKSPACE/swiftshader-install
+        cp Darwin/* $GITHUB_WORKSPACE/swiftshader-install
+
+    - name: moltenvk
+      run: |
+        wget -q https://github.com/KhronosGroup/MoltenVK/releases/download/${{ env.MOLTENVK_VERSION }}/MoltenVK-all.tar
+        tar -xf MoltenVK-all.tar
+
+    - name: arm64
+      run: |
+        mkdir build-arm64 && cd build-arm64
+        cmake ${{ env.NCNN_CMAKE_OPTIONS }} -DPLATFORM=MAC_ARM64 -DARCHS="arm64" -DNCNN_BUILD_TESTS=ON \
+            -DVulkan_LIBRARY=$GITHUB_WORKSPACE/MoltenVK/MoltenVK/static/MoltenVK.xcframework/macos-arm64_x86_64/libMoltenVK.a ..
+        cmake --build . -j 4
+
+    - name: arm64-test
+      run: |
+        printf "[Processor]\nThreadCount=1\n" > build-arm64/tests/SwiftShader.ini
+        cd build-arm64 && ctest --output-on-failure -j 4