windows.yml

name: Windows build
on: [push, pull_request]
concurrency:
  group: ${{ github.workflow }}-${{ github.job }}-${{ github.ref }}
  cancel-in-progress: true

jobs:
  build-windows-x86:
    name: 'MSVC ${{ matrix.os }}, ${{ matrix.target }} ${{ matrix.arch }}'
    defaults:
      run:
        shell: bash {0}
    strategy:
      matrix:
        os:
          - 2022
          - 2025
        target:
          - x86
          - x64
        arch:
          - sse2
          - avx
          - avx2
          - avx512bw
        exclude:
          # AVX on both platforms has a codegen error
          # On 2019 in _mm256_rsqrt_ps, on 2022 in _mm256_blend_p*
          - { arch: avx }
          # On both platforms x86 + AVX512 triggers a compiler crash
          - { target: x86, arch: avx512bw }
          # /arch:SSE2 is not available on x64 platforms (SSE2 is enabled by default)
          - { target: x64, arch: sse2 }
    runs-on: windows-${{ matrix.os }}
    steps:
    - name: Setup compiler
      uses: ilammy/msvc-dev-cmd@v1
      with:
        arch: ${{ matrix.target }}
    - name: Setup Ninja
      run: |
        python3 -m pip install --upgrade pip setuptools wheel
        python3 -m pip install ninja
    - name: Checkout xsimd
      uses: actions/checkout@v6
    - name: Setup
      run: |
        cmake -B _build \
              -DBUILD_TESTS=ON \
              -DDOWNLOAD_DOCTEST=ON \
              -DBUILD_BENCHMARK=ON \
              -DBUILD_EXAMPLES=ON \
              -DCMAKE_BUILD_TYPE=Release \
              -DTARGET_ARCH=${{ matrix.arch }} \
              -G Ninja
    - name: Build
      run: cmake --build _build
    - name: Testing xsimd
      if: ${{ !startsWith(matrix.arch, 'avx512') }}
      env:
        # Set CPU feature test expectations
        # Assuming the runner always has AVX2 (independent of compilation option)
        XSIMD_TEST_CPU_ASSUME_NEON64: "0"
        XSIMD_TEST_CPU_ASSUME_SSE4_2: "1"
        XSIMD_TEST_CPU_ASSUME_AVX2: "1"
        XSIMD_TEST_CPU_ASSUME_MANUFACTURER: "intel,amd"
      run: ./_build/test/test_xsimd

  build-windows-mingw:
    name: 'MSYS2 ${{ matrix.msystem }}'
    runs-on: windows-2022
    defaults:
      run:
        shell: msys2 {0}
    strategy:
      matrix:
        # Temporarily remove MINGW64 and UCRT64 builds because
        # GCC 12 gives an unexpected overflow warning for __builtin_memmove
        # see https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106199
        msystem: [ MINGW32, CLANG64 ]
        #msystem: [ MINGW32, MINGW64, UCRT64, CLANG32, CLANG64 ]
      fail-fast: false
    steps:
      - name: Use MinGW from MSYS2
        uses: msys2/setup-msys2@v2
        with:
          msystem: ${{ matrix.msystem }}
          update: true
          path-type: minimal
          pacboy: >-
            cc:p
            cmake:p
            ninja:p
      - name: Checkout xsimd
        uses: actions/checkout@v6
      - name: Configure
        run: |
          cmake -B _build \
                -DBUILD_TESTS=ON \
                -DBUILD_BENCHMARK=ON \
                -DBUILD_EXAMPLES=ON \
                -DCMAKE_BUILD_TYPE=Release \
                -DDOWNLOAD_DOCTEST=ON \
                -G Ninja
      - name: Build
        run: cmake --build _build
      - name: Test
        run: ./_build/test/test_xsimd

  build-windows-clang-cl:
    name: 'clang-cl x64 ${{ matrix.config.name }}'
    defaults:
      run:
        shell: bash {0}
    strategy:
      matrix:
        config:
          - { name: "AVX2",     arch: "avx2",    flags: "",         benchmark: "ON",  examples: "ON" }
          - { name: "/fp:fast", arch: "generic", flags: "/fp:fast", benchmark: "OFF", examples: "OFF" }
    runs-on: windows-2025
    steps:
    - name: Setup compiler
      uses: ilammy/msvc-dev-cmd@v1
      with:
        arch: amd64
    - name: Check clang-cl
      run: |
        command -v clang-cl
        clang-cl --version
    - name: Setup Ninja
      run: |
        python3 -m pip install --upgrade pip setuptools wheel
        python3 -m pip install ninja
    - name: Checkout xsimd
      uses: actions/checkout@v3
    - name: Setup
      run: |
        cmake -B _build \
              -DBUILD_TESTS=ON \
              -DDOWNLOAD_DOCTEST=ON \
              -DBUILD_BENCHMARK=${{ matrix.config.benchmark }} \
              -DBUILD_EXAMPLES=${{ matrix.config.examples }} \
              -DCMAKE_BUILD_TYPE=Release \
              -DCMAKE_C_COMPILER=clang-cl \
              -DCMAKE_CXX_COMPILER=clang-cl \
              -DTARGET_ARCH=${{ matrix.config.arch }} \
              -DCMAKE_CXX_FLAGS="${{ matrix.config.flags }} -DXSIMD_REASSOCIATIVE_MATH=1" \
              -G Ninja
    - name: Build
      run: cmake --build _build
    - name: Testing xsimd
      run: ./_build/test/test_xsimd

  build-windows-arm64:
    name: 'MSVC arm64 ${{ matrix.abi.name }}'
    defaults:
      run:
        shell: bash {0}
    strategy:
      matrix:
        abi:
          - { name: "native", cflags: "" }
          - { name: "arm64ec", cflags: " /arm64EC" }
    runs-on: windows-11-arm
    steps:
    - name: Setup compiler
      uses: ilammy/msvc-dev-cmd@v1
      with:
        arch: arm64
        # MSVC 19.51 miscompiles the float log1p kernel on neon64
        toolset: 14.44
    - name: Setup Ninja
      run: |
        python3 -m pip install --upgrade pip setuptools wheel
        python3 -m pip install ninja
    - name: Checkout xsimd
      uses: actions/checkout@v6
    - name: Setup
      run: |
        cmake -B _build \
              -DBUILD_TESTS=ON \
              -DDOWNLOAD_DOCTEST=ON \
              -DBUILD_BENCHMARK=ON \
              -DBUILD_EXAMPLES=ON \
              -DCMAKE_BUILD_TYPE=Release \
              -DTARGET_ARCH=neon64 \
              -G Ninja
      env:
        CFLAGS: ${{ matrix.abi.cflags }}
        CXXFLAGS: ${{ matrix.abi.cflags }}
    - name: Build
      run: cmake --build _build
    - name: Testing xsimd
      run: ./_build/test/test_xsimd