name: Build llama.cpp runtime binaries

# Cross-platform prebuilt binaries for the FunASR llama.cpp / GGUF runtime,
# like whisper.cpp's whisper-bin-*. Push a `runtime-llamacpp-v*` tag to publish
# a GitHub Release with the binaries; or run manually (workflow_dispatch) to test
# the build without creating a release.
on:
  workflow_dispatch:
  push:
    tags:
      - 'runtime-llamacpp-v*'

permissions:
  contents: write

jobs:
  build:
    name: build-${{ matrix.name }}
    strategy:
      fail-fast: false
      matrix:
        include:
          - os: ubuntu-latest
            name: linux-x64
            cmake_flags: >-
              -DGGML_NATIVE=OFF
              -DGGML_AVX=OFF
              -DGGML_AVX2=OFF
              -DGGML_AVX_VNNI=OFF
              -DGGML_AVX512=OFF
              -DGGML_AVX512_VBMI=OFF
              -DGGML_AVX512_VNNI=OFF
              -DGGML_AVX512_BF16=OFF
              -DGGML_FMA=OFF
              -DGGML_F16C=OFF
              -DGGML_BMI2=OFF
          - os: ubuntu-latest
            name: linux-x64-avx2
            cmake_flags: >-
              -DGGML_NATIVE=OFF
              -DGGML_AVX=ON
              -DGGML_AVX2=ON
              -DGGML_AVX_VNNI=OFF
              -DGGML_AVX512=OFF
              -DGGML_AVX512_VBMI=OFF
              -DGGML_AVX512_VNNI=OFF
              -DGGML_AVX512_BF16=OFF
              -DGGML_FMA=ON
              -DGGML_F16C=ON
              -DGGML_BMI2=ON
          - os: ubuntu-24.04-arm
            name: linux-arm64
            cmake_flags: -DGGML_NATIVE=OFF
          - os: macos-latest
            name: macos-arm64
            cmake_flags: -DGGML_NATIVE=OFF
          - os: windows-latest
            name: windows-x64
            cmake_flags: >-
              -DGGML_NATIVE=OFF
              -DGGML_AVX=OFF
              -DGGML_AVX2=OFF
              -DGGML_AVX_VNNI=OFF
              -DGGML_AVX512=OFF
              -DGGML_AVX512_VBMI=OFF
              -DGGML_AVX512_VNNI=OFF
              -DGGML_AVX512_BF16=OFF
              -DGGML_FMA=OFF
              -DGGML_F16C=OFF
              -DGGML_BMI2=OFF
          - os: windows-latest
            name: windows-x64-avx2
            cmake_flags: >-
              -DGGML_NATIVE=OFF
              -DGGML_AVX=ON
              -DGGML_AVX2=ON
              -DGGML_AVX_VNNI=OFF
              -DGGML_AVX512=OFF
              -DGGML_AVX512_VBMI=OFF
              -DGGML_AVX512_VNNI=OFF
              -DGGML_AVX512_BF16=OFF
              -DGGML_FMA=ON
              -DGGML_F16C=ON
              -DGGML_BMI2=ON
    runs-on: ${{ matrix.os }}
    defaults:
      run:
        shell: bash
        working-directory: runtime/llama.cpp
    steps:
      - uses: actions/checkout@v4
      - name: Configure and build
        run: |
          # Release binaries must be portable across user machines, not tuned to
          # the ephemeral CI runner CPU. A native ggml build may emit AVX512 or
          # AVX-VNNI instructions and then crash with SIGILL on ordinary CPUs.
          # Keep the default x64 package conservative, and publish explicit
          # x64-avx2 assets for machines that support AVX2/FMA/F16C/BMI2.
          cmake -B build -DCMAKE_BUILD_TYPE=Release ${{ matrix.cmake_flags }}
          cmake --build build --config Release -j 2
      - name: Package
        run: |
          mkdir -p pkg
          cp build/bin/llama-funasr-* pkg/ 2>/dev/null || true
          cp build/bin/Release/llama-funasr-* pkg/ 2>/dev/null || true
          cp README.md download-funasr-model.sh pkg/ 2>/dev/null || true
          echo "--- packaged binaries ---"; ls -la pkg
          if [ "${{ runner.os }}" = "Windows" ]; then
            7z a "funasr-llamacpp-${{ matrix.name }}.zip" ./pkg/*
          else
            tar czf "funasr-llamacpp-${{ matrix.name }}.tar.gz" -C pkg .
          fi
      - uses: actions/upload-artifact@v4
        with:
          name: funasr-llamacpp-${{ matrix.name }}
          path: runtime/llama.cpp/funasr-llamacpp-${{ matrix.name }}.*
          if-no-files-found: error

  release:
    needs: build
    if: startsWith(github.ref, 'refs/tags/')
    runs-on: ubuntu-latest
    steps:
      - uses: actions/download-artifact@v4
        with:
          path: dist
          merge-multiple: true
      - name: Create GitHub Release
        env:
          GH_TOKEN: ${{ github.token }}
        run: |
          gh release create "${{ github.ref_name }}" dist/* \
            --repo "${{ github.repository }}" \
            --title "FunASR llama.cpp runtime ${{ github.ref_name }}" \
            --notes "Prebuilt self-contained binaries for the FunASR llama.cpp / GGUF runtime — SenseVoice, Paraformer and Fun-ASR-Nano with built-in FSMN-VAD (a whisper.cpp-style on-device ASR, strong on Chinese). Get a model with \`bash download-funasr-model.sh <sensevoice|paraformer|nano>\`, then run \`llama-funasr-cli\` / \`llama-funasr-sensevoice\` / \`llama-funasr-paraformer\`. Use the default x64 asset for maximum CPU compatibility; use the x64-avx2 asset on CPUs with AVX2/FMA/F16C/BMI2 for higher throughput. No Python, no build. Docs: runtime/llama.cpp/README.md"