[gpu][alpaka] Add initial integration of ALPAKA

.github/workflows/benchmark.yml

@@ -0,0 +1,158 @@
+name: Benchmark
+
+on:
+  pull_request:
+    branches: [main]
+  workflow_dispatch:
+    inputs:
+      warmup:
+        description: "Warmup iterations"
+        default: "5"
+      iterations:
+        description: "Timed iterations"
+        default: "100"
+
+concurrency:
+  group: benchmark-${{ github.ref }}
+  cancel-in-progress: true
+
+env:
+  LCG_VIEW:     /cvmfs/sft.cern.ch/lcg/views/LCG_106a/x86_64-el9-gcc13-opt
+  CUDA_CVMFS:   /cvmfs/sft.cern.ch/lcg/contrib/cuda/12.4/x86_64-el9
+  CUDA_ARCH:    "90"
+  BUILD_TYPE:   Release
+  BENCH_WARMUP: ${{ github.event.inputs.warmup || '5' }}
+  BENCH_ITERS:  ${{ github.event.inputs.iterations || '100' }}
+  DEPS_CACHE:   /tmp/sofie-cmake-deps
+
+jobs:
+  benchmark:
+    name: Benchmark Comparison (H100)
+    runs-on: ml4ep-h100
+    container: registry.cern.ch/ngt/lxplus-like:9
+    timeout-minutes: 120
+
+    steps:
+      - name: GPU check
+        run: nvidia-smi
+
+      - name: Setup build environment
+        run: |
+          set -euo pipefail
+
+          if [ -f "${{ env.LCG_VIEW }}/setup.sh" ]; then
+            set +u; source "${{ env.LCG_VIEW }}/setup.sh"; set -u
+          else
+            echo "LCG view not found — installing from dnf"
+            dnf install -y epel-release
+            dnf install -y cmake ninja-build gcc-c++ python3 git \
+                           protobuf-devel openblas-devel
+          fi
+
+          if [ -x "${{ env.CUDA_CVMFS }}/bin/nvcc" ]; then
+            echo "${{ env.CUDA_CVMFS }}/bin" >> "$GITHUB_PATH"
+            echo "CUDA_HOME=${{ env.CUDA_CVMFS }}" >> "$GITHUB_ENV"
+          elif [ -x /usr/local/cuda/bin/nvcc ]; then
+            echo "/usr/local/cuda/bin" >> "$GITHUB_PATH"
+            echo "CUDA_HOME=/usr/local/cuda" >> "$GITHUB_ENV"
+          else
+            dnf config-manager --add-repo \
+              https://developer.download.nvidia.com/compute/cuda/repos/rhel9/x86_64/cuda-rhel9.repo
+            dnf install -y cuda-compiler-12-4 cuda-cudart-devel-12-4 cuda-libraries-devel-12-4
+            echo "/usr/local/cuda-12.4/bin" >> "$GITHUB_PATH"
+            echo "CUDA_HOME=/usr/local/cuda-12.4" >> "$GITHUB_ENV"
+          fi
+
+          echo "PATH=$PATH"                               >> "$GITHUB_ENV"
+          echo "LD_LIBRARY_PATH=${LD_LIBRARY_PATH:-}"    >> "$GITHUB_ENV"
+          echo "CMAKE_PREFIX_PATH=${CMAKE_PREFIX_PATH:-}" >> "$GITHUB_ENV"
+
+      - name: Checkout PR branch
+        uses: actions/checkout@v4
+        with:
+          path: sofie-pr
+
+      - name: Checkout main branch
+        if: github.event_name == 'pull_request'
+        uses: actions/checkout@v4
+        with:
+          ref: main
+          path: sofie-main
+
+      - name: Cache FetchContent dependencies
+        uses: actions/cache@v4
+        with:
+          path: ${{ env.DEPS_CACHE }}
+          key: cmake-deps-bench-${{ hashFiles('sofie-pr/benchmark/CMakeLists.txt') }}
+          restore-keys: cmake-deps-bench-
+
+      - name: Configure PR build
+        run: |
+          cmake -B sofie-pr/build -S sofie-pr \
+            -DCMAKE_BUILD_TYPE=${{ env.BUILD_TYPE }} \
+            -DSOFIE_WITH_ROOT=OFF \
+            -DSOFIE_BENCHMARK=ON \
+            -DSOFIE_BENCHMARK_BACKEND=CUDA \
+            "-DSOFIE_BENCHMARK_CUDA_ARCH=${{ env.CUDA_ARCH }}" \
+            "-DCMAKE_CUDA_ARCHITECTURES=${{ env.CUDA_ARCH }}" \
+            "-DFETCHCONTENT_BASE_DIR=${{ env.DEPS_CACHE }}"
+
+      - name: Build PR benchmark
+        run: cmake --build sofie-pr/build --target sofie_benchmark -j$(nproc)
+
+      - name: Run PR benchmark
+        working-directory: sofie-pr/build/benchmark
+        run: |
+          ./sofie_benchmark \
+            -w ${{ env.BENCH_WARMUP }} \
+            -n ${{ env.BENCH_ITERS }} \
+            | tee benchmark_pr.txt
+
+      - name: Configure main build
+        if: github.event_name == 'pull_request'
+        run: |
+          cmake -B sofie-main/build -S sofie-main \
+            -DCMAKE_BUILD_TYPE=${{ env.BUILD_TYPE }} \
+            -DSOFIE_WITH_ROOT=OFF \
+            -DSOFIE_BENCHMARK=ON \
+            -DSOFIE_BENCHMARK_BACKEND=CUDA \
+            "-DSOFIE_BENCHMARK_CUDA_ARCH=${{ env.CUDA_ARCH }}" \
+            "-DCMAKE_CUDA_ARCHITECTURES=${{ env.CUDA_ARCH }}" \
+            "-DFETCHCONTENT_BASE_DIR=${{ env.DEPS_CACHE }}"
+
+      - name: Build main benchmark
+        if: github.event_name == 'pull_request'
+        run: cmake --build sofie-main/build --target sofie_benchmark -j$(nproc)
+
+      - name: Run main benchmark
+        if: github.event_name == 'pull_request'
+        working-directory: sofie-main/build/benchmark
+        run: |
+          ./sofie_benchmark \
+            -w ${{ env.BENCH_WARMUP }} \
+            -n ${{ env.BENCH_ITERS }} \
+            | tee benchmark_main.txt
+
+      - name: Summarise PR vs main
+        if: github.event_name == 'pull_request'
+        run: |
+          echo "### Benchmark comparison: PR vs main" >> "$GITHUB_STEP_SUMMARY"
+          echo "" >> "$GITHUB_STEP_SUMMARY"
+          echo '```' >> "$GITHUB_STEP_SUMMARY"
+          echo "── PR ──────────────────────────────────────────────────────────────" \
+            >> "$GITHUB_STEP_SUMMARY"
+          cat sofie-pr/build/benchmark/benchmark_pr.txt   >> "$GITHUB_STEP_SUMMARY"
+          echo "" >> "$GITHUB_STEP_SUMMARY"
+          echo "── main ────────────────────────────────────────────────────────────" \
+            >> "$GITHUB_STEP_SUMMARY"
+          cat sofie-main/build/benchmark/benchmark_main.txt >> "$GITHUB_STEP_SUMMARY"
+          echo '```' >> "$GITHUB_STEP_SUMMARY"
+
+      - name: Upload benchmark results
+        if: always()
+        uses: actions/upload-artifact@v4
+        with:
+          name: benchmark-results-${{ github.run_id }}
+          path: |
+            sofie-pr/build/benchmark/benchmark_pr.txt
+            sofie-main/build/benchmark/benchmark_main.txt

.github/workflows/tests.yml

@@ -0,0 +1,113 @@
+name: Unit Tests
+
+on:
+  push:
+    branches: [main]
+  pull_request:
+    branches: [main]
+  workflow_dispatch:
+
+concurrency:
+  group: tests-${{ github.ref }}
+  cancel-in-progress: true
+
+env:
+  LCG_VIEW:   /cvmfs/sft.cern.ch/lcg/views/LCG_106a/x86_64-el9-gcc13-opt
+  CUDA_CVMFS: /cvmfs/sft.cern.ch/lcg/contrib/cuda/12.4/x86_64-el9
+  CUDA_ARCH:  "90"
+  BUILD_TYPE: Release
+  DEPS_CACHE: /tmp/sofie-cmake-deps
+
+jobs:
+  gpu-tests:
+    name: GPU Unit Tests (NVIDIA/H100)
+    runs-on: ml4ep-h100
+    container: registry.cern.ch/ngt/lxplus-like:9
+    timeout-minutes: 60
+
+    steps:
+      - name: GPU check
+        run: nvidia-smi
+
+      - name: Setup build environment
+        run: |
+          set -euo pipefail
+
+          # LCG view (cmake, gcc-13, protobuf, openblas)
+          if [ -f "${{ env.LCG_VIEW }}/setup.sh" ]; then
+            set +u; source "${{ env.LCG_VIEW }}/setup.sh"; set -u
+          else
+            echo "LCG view not found — installing from dnf"
+            dnf install -y epel-release
+            dnf install -y cmake ninja-build gcc-c++ python3 git \
+                           protobuf-devel openblas-devel
+          fi
+
+          # CUDA toolkit (nvcc + headers)
+          if [ -x "${{ env.CUDA_CVMFS }}/bin/nvcc" ]; then
+            echo "${{ env.CUDA_CVMFS }}/bin" >> "$GITHUB_PATH"
+            echo "CUDA_HOME=${{ env.CUDA_CVMFS }}" >> "$GITHUB_ENV"
+          elif [ -x /usr/local/cuda/bin/nvcc ]; then
+            echo "/usr/local/cuda/bin" >> "$GITHUB_PATH"
+            echo "CUDA_HOME=/usr/local/cuda" >> "$GITHUB_ENV"
+          else
+            echo "nvcc not found — installing CUDA toolkit from NVIDIA repo"
+            dnf config-manager --add-repo \
+              https://developer.download.nvidia.com/compute/cuda/repos/rhel9/x86_64/cuda-rhel9.repo
+            dnf install -y cuda-compiler-12-4 cuda-cudart-devel-12-4 cuda-libraries-devel-12-4
+            echo "/usr/local/cuda-12.4/bin" >> "$GITHUB_PATH"
+            echo "CUDA_HOME=/usr/local/cuda-12.4" >> "$GITHUB_ENV"
+          fi
+
+          # GTest
+          dnf install -y gtest-devel 2>/dev/null || \
+          dnf install -y googletest-devel 2>/dev/null || (
+            cd /tmp
+            git clone --depth 1 -b v1.14.0 https://github.com/google/googletest.git
+            cmake -B gtest-build -S googletest \
+                  -DCMAKE_INSTALL_PREFIX=/usr/local -DBUILD_SHARED_LIBS=ON
+            cmake --build gtest-build -j$(nproc)
+            cmake --install gtest-build
+          )
+
+          echo "PATH=$PATH"                               >> "$GITHUB_ENV"
+          echo "LD_LIBRARY_PATH=${LD_LIBRARY_PATH:-}"    >> "$GITHUB_ENV"
+          echo "CMAKE_PREFIX_PATH=${CMAKE_PREFIX_PATH:-}" >> "$GITHUB_ENV"
+
+      - name: Checkout
+        uses: actions/checkout@v4
+
+      - name: Cache FetchContent dependencies
+        uses: actions/cache@v4
+        with:
+          path: ${{ env.DEPS_CACHE }}
+          key: cmake-deps-tests-${{ hashFiles('test/CMakeLists.txt') }}
+          restore-keys: cmake-deps-tests-
+
+      - name: Configure
+        run: |
+          cmake -B build -S . \
+            -DCMAKE_BUILD_TYPE=${{ env.BUILD_TYPE }} \
+            -DSOFIE_WITH_ROOT=OFF \
+            -Dtesting=ON \
+            -DENABLE_ALPAKA_TESTS=ON \
+            -DALPAKA_BACKEND=cuda \
+            "-DCMAKE_CUDA_ARCHITECTURES=${{ env.CUDA_ARCH }}" \
+            "-DFETCHCONTENT_BASE_DIR=${{ env.DEPS_CACHE }}"
+
+      - name: Build tests
+        run: |
+          cmake --build build \
+            --target TestCustomModelsFromONNXForAlpakaCuda \
+            -j$(nproc)
+
+      - name: Run tests
+        working-directory: build
+        run: ctest --output-on-failure -j1
+
+      - name: Upload test log
+        if: always()
+        uses: actions/upload-artifact@v4
+        with:
+          name: test-log-${{ github.run_id }}
+          path: build/Testing/Temporary/LastTest.log