build-deploy-linux-cuda-12.6.yml

on:
  workflow_dispatch:
    inputs:
      buildThreads:
        description: 'Build threads for libnd4j. Used to control memory usage of builds.'
        required: true
        default: 4

      deployToReleaseStaging:
        description: 'Whether to deploy to release staging or not.'
        required: false
        default: 0

      releaseVersion:
        description: 'Release version target'
        required: false
        default: 1.0.0-M3

      snapshotVersion:
        description: 'Snapshot version target'
        required: false
        default: 1.0.0-SNAPSHOT

      releaseRepoId:
        description: 'Release repository id'
        required: false
        default:

      serverId:
        description: 'Server id to publish to'
        required: false
        default: central


      mvnFlags:
        description: "Extra maven flags (must escape input yourself if used)"
        required: false
        default:

      libnd4jUrl:
        description: 'Whether to download libnd4j using  https://github.com/KonduitAI/gh-actions-libnd4j-urls/ for this build. LIBND4J_HOME will automatically be set. Should be used when only needing to build other modules.'
        required: false
        default:


      runsOn:
        description: 'System to run on'
        required: false
        default: ubuntu-22.04

      debug_enabled:
        description: 'Run the build with tmate debugging enabled (https://github.com/marketplace/actions/debugging-with-tmate)'
        required: false
        default: false
jobs:
  linux-x86_64-cuda-12-6:
    strategy:
      fail-fast: false
      matrix:
        helper: [ cudnn, "" ]
        extension: [ ""  ]
        include:
          - mvn_ext: ${{ github.event.inputs.mvnFlags }}
            experimental: true
            name: Extra maven flags
          - debug_enabled:  ${{ github.event.inputs.debug_enabled }}
            experimental: true
            name: Debug enabled

          - runs_on: ${{ github.event.inputs.runsOn }}
            experimental: true
            name: OS to run on

          - libnd4j_file_download: ${{ github.event.inputs.libnd4jUrl }}
            experimental: true
            name: OS to run on

          - deploy_to_release_staging: ${{ github.event.inputs.deployToReleaseStaging }}
            experimental: true
            name: Whether to deploy to release staging or not

          - release_version: ${{ github.event.inputs.releaseVersion }}
            experimental: true
            name: Release version

          - snapshot_version: ${{ github.event.inputs.snapshotVersion }}
            experimental: true
            name: Snapshot version

          - server_id: ${{ github.event.inputs.serverId }}
            experimental: true
            name: Server id

          - release_repo_id: ${{ github.event.inputs.releaseRepoId }}
            experimental: true
            name: The release repository to run on

          - mvn_flags: ${{ github.event.inputs.mvnFlags }}
            experimental: true
            name: Extra maven flags to use as part of the build

          - build_threads: ${{ github.event.inputs.buildThreads }}
            experimental: true
            name: The number of threads to build libnd4j with

    runs-on: ${{ matrix.runs_on }}
    env:
      CUDA_PATH: /usr/local/cuda-12.6
      CUDNN_ROOT_DIR: /usr/local/cuda-12.6
    steps:
      - name: Cancel Previous Runs
        uses: styfle/cancel-workflow-action@0.8.0
        with:
          access_token: ${{ github.token }}
      
      - name: Free Disk Space
        uses: jlumbroso/free-disk-space@main
        with:
          tool-cache: false
          android: true
          dotnet: true
          haskell: true
          large-packages: true
          docker-images: true
          swap-storage: true

      - name: Configure swap space
        shell: bash
        run: |
          echo "=== Initial system status ==="
          free -h
          df -h
          echo "Available disk space on root:"
          df -h /
          
          # Check if swap is already enabled
          if swapon --show | grep -q "/"; then
            echo "Swap already configured:"
            swapon --show
          else
            echo "Configuring swap space..."
          
            # Use input parameter for swap size, default to 12GB like Android workflow
            SWAP_SIZE="12G"
            echo "Creating ${SWAP_SIZE} swap file..."
          
            # Create swap file
            sudo fallocate -l ${SWAP_SIZE} /swapfile || {
              sudo dd if=/dev/zero of=/swapfile bs=1G count=12 status=progress
            }
          
            # Set up swap
            sudo chmod 600 /swapfile
            sudo mkswap /swapfile
            sudo swapon /swapfile
          
            # Verify swap is active
            echo "=== Swap configuration completed ==="
            free -h
            swapon --show
          fi
          
          # Tune swappiness for build workloads - using Android workflow settings
          # Lower values (10-20) prefer RAM, higher values (60+) use swap more aggressively
          echo "Current swappiness: $(cat /proc/sys/vm/swappiness)"
          sudo sysctl vm.swappiness=80
          echo "Adjusted swappiness: $(cat /proc/sys/vm/swappiness)"
          sudo sysctl vm.overcommit_memory=2        # Allow overcommit
          sudo sysctl vm.overcommit_ratio=80
          sudo sysctl vm.vfs_cache_pressure=50
          sudo sysctl vm.dirty_ratio=10
          sudo sysctl vm.dirty_background_ratio=5
          sudo sysctl kernel.shmmax=68719476736
          # Show final memory status
          echo "=== Final memory status ==="
          free -h

      - uses: actions/checkout@v2
      - name: Set mvn build command based on matrix
        shell: bash
        run: |
            if  [  "${{ matrix.libnd4j_file_download }}" != '' ]; then
                 modules=':nd4j-cuda-12.6,:nd4j-cuda-12.6-preset'
            elif [ "${{ matrix.helper }}" == '' ]; then
               echo "Building libnd4j from source"
               modules=':nd4j-cuda-12.6,:nd4j-cuda-12.6-preset,:libnd4j,:nd4j-cuda-12.6-platform'
            else 
                 echo "Building libnd4j from source"
                 modules=':nd4j-cuda-12.6,:nd4j-cuda-12.6-preset,:libnd4j'
            fi
            command="mvn ${{ matrix.mvn_ext }} -Pcuda -Dlibnd4j.generate.flatc=ON --no-transfer-progress -Dlibnd4j.cuda.compile.skip=false -Dlibnd4j.chip=cuda -pl ${modules} -Dlibnd4j.compute='8.6 9.0' -Dlibnd4j.cpu.compile.skip=true -Dhttp.keepAlive=false -Dmaven.wagon.http.pool=false -Dmaven.wagon.http.retryHandler.count=3 -Possrh -Dlibnd4j.buildthreads=${{ matrix.build_threads }} -Djavacpp.platform=linux-x86_64 -Dlibnd4j.chip=cuda --also-make clean --batch-mode package deploy -DskipTests"
            libnd4j_download_file_url=""
            if [ "${{ matrix.helper }}" != '' ] && [ "${{ matrix.extension }}" != '' ]; then
               mvn_ext=" -Djavacpp.platform.extension=-${{ matrix.helper }}-${{ matrix.extension }} -Dlibnd4j.helper=${{ matrix.helper }} -Dlibnd4j.extension=${{ matrix.extension }} -Dlibnd4j.classifier=linux-x86_64-cuda-12.6-${{ matrix.helper }}-${{matrix.extension}}"
               libnd4j_download_file_url="linux-cuda-12.6-${{ matrix.extension }}-${{ matrix.helper }}"
            elif [ "${{ matrix.helper }}" != '' ]; then
               mvn_ext=" -Djavacpp.platform.extension=-${{ matrix.helper }} -Dlibnd4j.helper=${{ matrix.helper }} -Dlibnd4j.classifier=linux-x86_64-cuda-12.6-${{ matrix.helper }}"
               libnd4j_download_file_url="linux-cuda-12.6-${{ matrix.extension }}-${{ matrix.helper }}"
            else
               mvn_ext=" -Dlibnd4j.classifier=linux-x86_64-cuda-12.6"
               libnd4j_download_file_url="linux-cuda-12.6-${{ matrix.extension }}-${{ matrix.helper }}"

            fi
            command="${command} ${mvn_ext}"
            echo "Setting command for helper ${{ matrix.helper }} and extension ${{ matrix.extension }} to ${command}"
            echo "COMMAND=${command}" >> $GITHUB_ENV
      - name: Cache protobuf install
        uses: actions/cache@v4
        id: cache-protobuf
        with:
          path: /opt/protobuf
          key: ${{ matrix.runs_on }}-protobuf
          restore-keys: ${{ matrix.runs_on }}-protobuf
      - uses: ./.github/actions/install-protobuf-linux
      - name: Set up Java for publishing to GitHub Packages
        uses: actions/setup-java@v4
        with:
          java-version: 11
          distribution: 'temurin'
          server-id:  ${{ github.event.inputs.serverId }}
          server-username: MAVEN_USERNAME
          server-password: MAVEN_PASSWORD
          gpg-private-key: ${{ secrets.SONATYPE_GPG_KEY }}
          gpg-passphrase: MAVEN_GPG_PASSPHRASE
          cache: 'maven'

      - uses: konduitai/cuda-install/.github/actions/install-cuda-ubuntu@master
        env:
          cuda: 12.6.0
          GCC:  11
        if: steps.cache-cuda-126.outputs.cache-hit != 'true'

      # Set up CUDA environment paths
      - name: Setup CUDA PATH
        run: |
          echo "PATH=/usr/local/cuda-12.6/bin:$PATH" >> $GITHUB_ENV
          echo "LD_LIBRARY_PATH=/usr/local/cuda-12.6/lib64:$LD_LIBRARY_PATH" >> $GITHUB_ENV

      # Verify the CUDA installation and PATH setup
      - name: Verify CUDA Setup
        run: |
          echo "Verifying CUDA installation and PATH setup"
          echo $PATH
          ls -la /usr/local/cuda-12.6/bin
          which nvcc || echo "nvcc not found in PATH"
          nvcc --version || echo "nvcc command failed"

      - name: Run cuda compilation on linux-x86_64
        shell: bash
        env:
          MAVEN_GPG_KEY: ${{ secrets.SONATYPE_GPG_KEY }}
          DEBIAN_FRONTEND: noninteractive
          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
          PUBLISH_TO: central
          MAVEN_USERNAME: ${{ secrets.CENTRAL_SONATYPE_TOKEN_USERNAME }}
          MAVEN_PASSWORD: ${{ secrets.CENTRAL_SONATYPE_TOKEN_PASSWORD }}
          MAVEN_GPG_PASSPHRASE: ${{ secrets.PACKAGES_GPG_PASS }}
          PERFORM_RELEASE: ${{ matrix.deploy_to_release_staging }}
          RELEASE_VERSION: ${{ matrix.release_version }}
          SNAPSHOT_VERSION: ${{ matrix.snapshot_version }}
          RELEASE_REPO_ID: ${{ matrix.release_repo_id }}
          MODULES: ${{ matrix.mvn_flags }}
          LIBND4J_HOME_SUFFIX: cuda
          MAVEN_OPTS: -Xmx2g
          HELPER: ${{ matrix.helper }}
          EXTENSION: ${{ matrix.extension }}
          LIBND4J_FILE_NAME: ${{ matrix.libnd4j_file_download }}
        run: |
          echo "libnd4j build threads ${{ matrix.build_threads }}"
          echo "deploy to release staging repo or not ${{ matrix.deploy_to_release_staging }}"
          echo "release version ${{ matrix.release_version }}"
          echo "snapshot version ${{ matrix.snapshot_version }}"
          echo "debug enabled ${{ matrix.debug_enabled }}"
          echo "libnd4j url ${{ matrix.libnd4j_file_download }}"
          echo "maven flags ${{ matrix.mvn_flags }}"
          echo "snapshot version ${{ matrix.snapshot_version }}"
          echo "server id ${{ matrix.server_id }}"
          echo "release repo id ${{ matrix.release_repo_id }}"

          # Explicitly set PATH and other environment variables for this step
          export PATH=/usr/local/cuda-12.6/bin:/opt/protobuf/bin:/opt/cmake/bin:$PATH
          export LD_LIBRARY_PATH=/usr/local/cuda-12.6/lib64:$LD_LIBRARY_PATH
          
          # Test that nvcc is properly in the PATH
          echo "Checking nvcc availability:"
          which nvcc
          nvcc --version
          
          mvn --version
          cmake --version
          protoc --version
          sudo apt-get autoremove
          sudo apt-get clean
          bash ./change-cuda-versions.sh 12.6
      
          # Note: we need this for the cudnn helpers, our cmake can't find it otherwise.
          # See here: https://github.com/eclipse/deeplearning4j/blob/master/libnd4j/CMakeLists.txt#L298
          if [ "$PERFORM_RELEASE" == 1 ]; then
                    echo "Performing release"
                    bash ${GITHUB_WORKSPACE}/release-specified-component.sh  "${RELEASE_VERSION}" "${SNAPSHOT_VERSION}" "${RELEASE_REPO_ID}" "${COMMAND}"
                   else
                       echo "Running build and deploying to snapshots"
                       eval "${COMMAND}"
          fi