CI: Split up our monolithic build workflow

.github/workflows/ais-ci.yml

@@ -24,36 +24,56 @@ jobs:
         run: |
           echo "changed_dockerfile=$(./util/files-changed.sh origin/${{ github.base_ref }} HEAD 'util/docker/DOCKERFILE.*')" >> ${GITHUB_OUTPUT}
   build_AIS_CI_image:
-    if: ${{ needs.AIS_CI_Pre-check.outputs.changed_dockerfile == '1' }}
     needs: AIS_CI_Pre-check
     permissions:
       contents: read
       packages: write
     uses: ./.github/workflows/build-ais-ci-image.yml
     with:
+      dockerfile_changed: ${{ needs.AIS_CI_Pre-check.outputs.changed_dockerfile == '1' }}
       platform: ${{ inputs.platform }}
   build_and_test:
-    # Run after build_AIS_CI_image but only if build_AIS_CI_image passed or was skipped.
-    # success() implicitly added if no other status check present.
-    if: >-
-      ${{
-        !cancelled() &&
-        (
-          needs.build_AIS_CI_image.result == 'skipped' ||
-          (
-            needs['AIS_CI_Pre-check'].outputs.changed_dockerfile == '1' &&
-            needs.build_AIS_CI_image.result == 'success'
-          )
-        )
-      }}
+    # cancelled() needed. Otherwise success() implicitly added and fails if build_AIS_CI_image skipped.
+    if: ${{ !cancelled() && needs.build_AIS_CI_image.outputs.ci_image_build_satisfied == 'true' }}
     needs: [AIS_CI_Pre-check, build_AIS_CI_image]
     uses: ./.github/workflows/build-ais.yml
     with:
-      # If Dockerfile changed, use the new CI image. Otherwise, use 'latest'.
-      ci_image: >-
-        ${{
-          needs.AIS_CI_Pre-check.outputs.changed_dockerfile == '1' &&
-          needs.build_AIS_CI_image.outputs.ci_image ||
-          format('ghcr.io/rocm/hipfile/ais_ci_{0}:latest', inputs.platform)
-        }}
+      ci_image: ${{ needs.build_AIS_CI_image.outputs.ci_image }}
+      cxx_compiler: amdclang++
+      platform: ${{ inputs.platform }}
+      upload_artifacts: true
+  build_and_test_other_compilers:
+    # Try building on other compilers, but keep as separate job.
+    # Removes issues regarding matrix job overwriting outputs and
+    # does not propagate failures.
+    if: ${{ !cancelled() && needs.build_AIS_CI_image.outputs.ci_image_build_satisfied == 'true' }}
+    needs: [AIS_CI_Pre-check, build_AIS_CI_image]
+    strategy:
+      fail-fast: false
+      matrix:
+        cxx_compiler:
+          - clang++
+          - g++
+    uses: ./.github/workflows/build-ais.yml
+    with:
+      ci_image: ${{ needs.build_AIS_CI_image.outputs.ci_image }}
+      cxx_compiler: ${{ matrix.cxx_compiler }}
+      platform: ${{ inputs.platform }}
+  AIS_system_tests:
+    # cancelled() needed. Otherwise success() implicitly added and fails if the chained
+    # dependency build_AIS_CI_image is skipped.
+    if: ${{ !cancelled() && needs.build_and_test.result == 'success'}}
+    needs: [build_and_test]
+    uses: ./.github/workflows/test-ais-system.yml
+    with:
+      ais_hipfile_pkg_dev_filename: ${{ needs.build_and_test.outputs.ais_hipfile_pkg_dev_filename }}
+      ais_hipfile_pkg_filename: ${{ needs.build_and_test.outputs.ais_hipfile_pkg_filename }}
+      ci_image: ${{ needs.build_and_test.outputs.ci_image }}
+      platform: ${{ inputs.platform }}
+  build_and_test_Nvidia:
+    if: ${{ !cancelled() && needs.build_AIS_CI_image.outputs.ci_image_build_satisfied == 'true' }}
+    needs: [AIS_CI_Pre-check, build_AIS_CI_image]
+    uses: ./.github/workflows/build-ais-nvidia.yml
+    with:
+      ci_image: ${{ needs.build_AIS_CI_image.outputs.ci_image }}
       platform: ${{ inputs.platform }}

.github/workflows/build-ais-ci-image.yml

@@ -6,18 +6,51 @@ env:
 on:
   workflow_call:
     inputs:
+      dockerfile_changed:
+        # While this seems like a smell to not make this check in this workflow,
+        # it allows us to skip the job entirely. This yields the benefit of
+        # letting downstream jobs start sooner and keeping the job result as
+        # 'skipped' if a new CI image is not needed.
+        description: Hint for if this workflow actually needs to run.
+        required: true
+        type: boolean
       platform:
         required: true
         type: string
     outputs:
       ci_image:
-        description: The full name & tag of the newly built image.
-        value: ${{ jobs.build_AIS_image.outputs.new_ci_image}}
+        description: |
+          The full name & tag of the CI image to use.
+          If a new CI image is produced, returns the new CI image name & tag.
+          Otherwise, returns the current CI image name & tag.
+        value: >-
+          ${{ 
+            jobs.build_AIS_image.result == 'skipped' && format('ghcr.io/rocm/hipfile/ais_ci_{0}:latest', inputs.platform) ||
+            jobs.build_AIS_image.result == 'success' && jobs.build_AIS_image.outputs.new_ci_image ||
+            null
+          }}
+      ci_image_build_satisfied:
+        description: |
+          Indicates if the CI image is in an acceptable state.
+          Set to 'true' if the CI image is unchanged, or if a new CI image was
+          built successfully. Otherwise 'false'.
+        value: >-
+          ${{
+            (
+              !inputs.dockerfile_changed &&
+              jobs.build_AIS_image.result == 'skipped'
+            ) ||
+            (
+              inputs.dockerfile_changed &&
+              jobs.build_AIS_image.result == 'success'
+            )
+          }}
 permissions:
   contents: read
   packages: write
 jobs:
   build_AIS_image:
+    if: ${{ inputs.dockerfile_changed }}
     env:
       AIS_CI_IMAGE_NAME: ais_ci_${{ inputs.platform }}
     runs-on: [ubuntu-24.04]

.github/workflows/build-ais-nvidia.yml

@@ -0,0 +1,107 @@
+name: build_ais_nvidia
+run-name: Build AIS for Nvidia platforms
+env:
+  AIS_MOUNT_PATH: /mnt/ais/ext4
+on:
+  workflow_call:
+    inputs:
+      ci_image:
+        required: true
+        type: string
+      platform:
+        required: true
+        type: string
+permissions:
+  contents: read
+  packages: read
+jobs:
+  compile_on_NVIDIA:
+    runs-on: [ubuntu-24.04]
+    strategy:
+      fail-fast: false
+      matrix:
+        supported_compilers:
+          - g++
+          - clang++
+    steps:
+      - name: Set early AIS CI environment variables
+        run: |
+          echo "AIS_PR_NUMBER=$(echo ${{ github.ref }} | sed 's|[^0-9]||g')" >> "${GITHUB_ENV}"
+          echo "AIS_SAFE_COMPILER_NAME=$(echo ${{ matrix.supported_compilers }} | sed 's|[\+]|plus|g')" >> "${GITHUB_ENV}"
+      - name: Set AIS CI container name
+        run: |
+          echo "AIS_CONTAINER_NAME=${AIS_PR_NUMBER}_${{ github.job }}_${{ inputs.platform }}_${AIS_SAFE_COMPILER_NAME}" >> "${GITHUB_ENV}"
+      - name: Fetching code repository...
+        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8 #v6.0.1
+        with:
+          path: hipFile
+      - name: Authenticating to GitHub Container Registry
+        uses: docker/login-action@5e57cd118135c172c3672efd75eb46360885c0ef #v3.6.0
+        with:
+          registry: ghcr.io
+          username: ${{ github.actor }}
+          password: ${{ secrets.GITHUB_TOKEN }}
+      # Detach the container and run separate commands to it.
+      # Thus we can make separate explicit steps in the Github CI
+      # as if we were able to parameterize the container image in the first place.
+      - name: Starting Docker Container
+        run: |
+          docker run \
+            -dt \
+            --rm \
+            --pull always \
+            -v ${GITHUB_WORKSPACE}:/mnt/ais:ro \
+            -v ${{ env.AIS_MOUNT_PATH }}:/mnt/ais-fs \
+            --name ${AIS_CONTAINER_NAME} \
+            ${{ inputs.ci_image }}
+      - name: Make copy of the code repository and create build directories
+        run: |
+          docker exec \
+            -t \
+            ${AIS_CONTAINER_NAME} \
+            /bin/bash -c '
+              cp -R /mnt/ais /ais
+              mkdir /ais/hipFile/build
+            '
+      - name: Generate build files for hipFile targeting the NVIDIA platform (${{ matrix.supported_compilers }})
+        run: |
+          docker exec \
+            -t \
+            -w /ais/hipFile/build \
+            ${AIS_CONTAINER_NAME} \
+            /bin/bash -c '
+              cmake \
+                -DCMAKE_CXX_COMPILER=${{ matrix.supported_compilers }} \
+                -DCMAKE_CXX_FLAGS="-Werror" \
+                -DCMAKE_HIP_PLATFORM=nvidia \
+                -DAIS_BUILD_DOCS=ON \
+                ..
+            '
+      - name: Build hipFile for the NVIDIA platform (${{ matrix.supported_compilers }})
+        run: |
+          docker exec \
+            -t \
+            -w /ais/hipFile/build \
+            ${AIS_CONTAINER_NAME} \
+            /bin/bash -c '
+              cmake --build . --parallel
+            '
+      - name: Clean CMake build directories
+        run: |
+          docker exec \
+            -t \
+            -w /ais/hipFile \
+            ${AIS_CONTAINER_NAME} \
+            /bin/bash -c '
+              rm -rf build/*
+            '
+      - name: Cleanup & Stop the Docker container
+        if: ${{ always() }}
+        run: |
+          docker stop ${AIS_CONTAINER_NAME}
+### Disable until NVIDIA runners come back
+#  Run_hipFile_NVIDIA:
+#    uses: ./.github/workflows/hipfile-nvidia.yml
+#    needs: build_AIS_image
+#    with:
+#      platform: ${{inputs.platform}}