Brunobkr/llama.cpp_AlgMor24_github

1name: CI (CUDA, windows)2 3# TODO: this workflow is only triggered manually because it is very heavy on the CI4#       when we provision dedicated windows runners, we can enable it for pushes too5# note: running this workflow manually will populate the ccache for the release builds6#       this can be used before merging a PR to speed up the release workflow7on:8  workflow_dispatch: # allows manual triggering9 10# note: this will run in queue with the release workflow11concurrency:12  group: release13  queue: max14 15env:16  GH_TOKEN: ${{ github.token }}17  GGML_NLOOP: 318  GGML_N_THREADS: 119  LLAMA_ARG_LOG_COLORS: 120  LLAMA_ARG_LOG_PREFIX: 121  LLAMA_ARG_LOG_TIMESTAMPS: 122 23jobs:24  cuda:25    runs-on: windows-202226 27    permissions:28      actions: write29 30    strategy:31      matrix:32        cuda: ['12.4', '13.3']33 34    steps:35      - name: Clone36        id: checkout37        uses: actions/checkout@v638 39      - name: ccache40        uses: ggml-org/ccache-action@v1.2.2141        with:42          key: release-windows-2022-x64-cuda-${{ matrix.cuda }}43 44      - name: Install Cuda Toolkit45        uses: ./.github/actions/windows-setup-cuda46        with:47          cuda_version: ${{ matrix.cuda }}48 49      - name: Install Ninja50        id: install_ninja51        run: |52          choco install ninja53 54      - name: Build55        id: cmake_build56        shell: cmd57        # TODO: Remove GGML_CUDA_CUB_3DOT2 flag once CCCL 3.2 is bundled within CTK and that CTK version is used in this project58        run: |59          call "C:\Program Files\Microsoft Visual Studio\2022\Enterprise\VC\Auxiliary\Build\vcvarsall.bat" x6460          cmake -S . -B build -G "Ninja Multi-Config" ^61            -DLLAMA_BUILD_SERVER=ON ^62            -DLLAMA_BUILD_BORINGSSL=ON ^63            -DGGML_NATIVE=OFF ^64            -DGGML_BACKEND_DL=ON ^65            -DGGML_CPU_ALL_VARIANTS=ON ^66            -DGGML_CUDA=ON ^67            -DGGML_RPC=ON ^68            -DGGML_CUDA_CUB_3DOT2=ON69          set /A NINJA_JOBS=%NUMBER_OF_PROCESSORS%-170          cmake --build build --config Release -j %NINJA_JOBS% -t ggml71          cmake --build build --config Release72 73      - name: ccache-clear74        uses: ./.github/actions/ccache-clear75        with:76          key: release-windows-2022-x64-cuda-${{ matrix.cuda }}77 78  hip:79    runs-on: windows-202280 81    permissions:82      actions: write83 84    env:85      # Make sure this is in sync with build-cache.yml86      ROCM_VERSION: "7.14.0"87 88    strategy:89      matrix:90        include:91          # sync with release.yml92          - name: "radeon"93            gpu_targets: "gfx1150;gfx1151;gfx1200;gfx1201;gfx1100;gfx1101;gfx1102;gfx1030;gfx1031;gfx1032"94 95    steps:96      - name: Clone97        id: checkout98        uses: actions/checkout@v699 100      - name: Cache ROCm Installation101        uses: actions/cache@v5102        id: cache-rocm103        with:104          path: C:\TheRock\build105          key: rocm-wheels-${{ env.ROCM_VERSION }}-multi-arch-${{ runner.os }}106 107      - name: Setup ROCm108        if: steps.cache-rocm.outputs.cache-hit != 'true'109        uses: ./.github/actions/windows-setup-rocm110        with:111          version: ${{ env.ROCM_VERSION }}112 113      - name: Setup ROCm Environment114        run: |115          $ErrorActionPreference = "Stop"116 117          # Activate venv from cache or fresh install118          & C:\TheRock\build\.venv\Scripts\Activate.ps1119 120          # Expand the devel tree (idempotent; no-op if already done during install)121          rocm-sdk init122          if ($LASTEXITCODE -ne 0) { throw "rocm-sdk init failed with exit code $LASTEXITCODE" }123 124          # Get ROCm installation paths using the rocm-sdk CLI tool125          $rocmPath = (rocm-sdk path --root)126          if (-not $rocmPath) { throw "rocm-sdk path --root returned empty - devel package may not be installed" }127          $rocmPath = $rocmPath.Trim()128          $cmakePath = (rocm-sdk path --cmake).Trim()129          $binPath = (rocm-sdk path --bin).Trim()130          write-host "ROCm root: $rocmPath"131 132          echo "HIP_PATH=$rocmPath" >> $env:GITHUB_ENV133          echo "CMAKE_PREFIX_PATH=$cmakePath" >> $env:GITHUB_ENV134          echo "HIP_DEVICE_LIB_PATH=$rocmPath\lib\llvm\amdgcn\bitcode" >> $env:GITHUB_ENV135          echo "HIP_PLATFORM=amd" >> $env:GITHUB_ENV136          echo "LLVM_PATH=$rocmPath\lib\llvm" >> $env:GITHUB_ENV137          echo "$binPath" >> $env:GITHUB_PATH138 139          # Keep venv in PATH for subsequent steps140          echo "C:\TheRock\build\.venv\Scripts" >> $env:GITHUB_PATH141 142      - name: Verify ROCm143        id: verify144        run: |145          # Test the ROCm clang shipped in the installed wheel146          & "${env:HIP_PATH}\lib\llvm\bin\clang.exe" --version147 148      - name: ccache149        uses: ggml-org/ccache-action@v1.2.21150        with:151          # TODO: this build does not match the build in release.yml, so we use a different cache key152          #       ideally, the builds should match, similar to the CUDA build above so that we would be able153          #       to populate the ccache for the release with manual runs of this workflow154          #key: release-windows-2022-x64-hip-${{ env.ROCM_VERSION }}-${{ matrix.name }}155          key: cuda-windows-2022-x64-hip-${{ env.ROCM_VERSION }}-${{ matrix.name }}156 157      - name: Build158        id: cmake_build159        run: |160          cmake -G "Unix Makefiles" -B build -S . `161            -DCMAKE_PREFIX_PATH="${env:HIP_PATH}" `162            -DCMAKE_C_COMPILER="${env:HIP_PATH}\lib\llvm\bin\clang.exe" `163            -DCMAKE_CXX_COMPILER="${env:HIP_PATH}\lib\llvm\bin\clang++.exe" `164            -DCMAKE_HIP_COMPILER="${env:HIP_PATH}\lib\llvm\bin\clang.exe" `165            -DCMAKE_BUILD_TYPE=Release `166            -DLLAMA_BUILD_BORINGSSL=ON `167            -DHIP_PATH="${env:HIP_PATH}" `168            -DGGML_HIP=ON `169            -DGPU_TARGETS="gfx1100" `170            -DGGML_RPC=ON171          cmake --build build -j ${env:NUMBER_OF_PROCESSORS}172 173      - name: ccache-clear174        uses: ./.github/actions/ccache-clear175        with:176          #key: release-windows-2022-x64-hip-${{ env.ROCM_VERSION }}-${{ matrix.name }}177          key: cuda-windows-2022-x64-hip-${{ env.ROCM_VERSION }}-${{ matrix.name }}178