Improve CI

.github/workflows/__build_base.yaml

@@ -1,19 +1,24 @@
-# Generate base images for each backend. Images are stored on Github registry
-# as Docker format inconditionnaly and as Singularity format only if requested.
-# Docker images are used for building Kokkos FFT, while Singularity images are
-# used for test execution on the righteous hardware. This workflow can only be
-# invoked through another workflows.
+# Generate base images for each compiler environment. Images are stored on
+# Github registry as Docker format inconditionnaly and as Singularity format
+# only if requested.  Docker images are used for building Kokkos FFT, while
+# Singularity images are used for test execution on the righteous hardware.
+# This workflow can only be invoked through another workflows.
 
 name: Build base images
 
 on:
   workflow_call:
     inputs:
-      # suffix of the Docker and Singularity images
-      image_name_suffix:
+      image_suffix:
+        description: "Suffix of the Docker and Singularity images"
         required: false
         default: main
         type: string
+      image_tag:
+        description: "Tag of the Docker and Singularity images"
+        required: false
+        default: latest
+        type: string
 
 env:
   # Force the use of BuildKit for Docker
@@ -25,14 +30,14 @@ jobs:
 
     strategy:
       matrix:
-        backend:
-          - name: openmp
+        image:
+          - name: gcc
             use_singularity: false
-          - name: cuda
+          - name: nvcc
             use_singularity: true
-          - name: hip
+          - name: rocm
             use_singularity: false
-          - name: sycl
+          - name: intel
             use_singularity: false
 
     steps:
@@ -61,19 +66,19 @@ jobs:
       - name: Build Docker image
         run: |
           docker build \
-            -t ghcr.io/cexa-project/kokkos-fft/base_${{ matrix.backend.name }}_${{ inputs.image_name_suffix }} \
-            --cache-from ghcr.io/cexa-project/kokkos-fft/base_${{ matrix.backend.name }}_main \
+            -t ghcr.io/cexa-project/kokkos-fft/base_${{ matrix.image.name }}_${{ inputs.image_suffix }}:${{ inputs.image_tag }} \
+            --cache-from ghcr.io/cexa-project/kokkos-fft/base_${{ matrix.image.name }}_main:latest \
             --build-arg BUILDKIT_INLINE_CACHE=1 \
             --progress=plain \
-            docker/${{ matrix.backend.name }}
+            docker/${{ matrix.image.name }}
 
       - name: Push Docker image
-        run: docker push ghcr.io/cexa-project/kokkos-fft/base_${{ matrix.backend.name }}_${{ inputs.image_name_suffix }}
+        run: docker push ghcr.io/cexa-project/kokkos-fft/base_${{ matrix.image.name }}_${{ inputs.image_suffix }}:${{ inputs.image_tag }}
 
       - name: Convert Docker image to Singularity
-        run: singularity build base.sif docker://ghcr.io/cexa-project/kokkos-fft/base_${{ matrix.backend.name }}_${{ inputs.image_name_suffix }}
-        if: ${{ matrix.backend.use_singularity }}
+        run: singularity build base.sif docker://ghcr.io/cexa-project/kokkos-fft/base_${{ matrix.image.name }}_${{ inputs.image_suffix }}:${{ inputs.image_tag }}
+        if: ${{ matrix.image.use_singularity }}
 
       - name: Push Singularity image
-        run: singularity push base.sif oras://ghcr.io/cexa-project/kokkos-fft/base_${{ matrix.backend.name }}_singularity_${{ inputs.image_name_suffix }}
-        if: ${{ matrix.backend.use_singularity }}
+        run: singularity push base.sif oras://ghcr.io/cexa-project/kokkos-fft/base_${{ matrix.image.name }}_singularity_${{ inputs.image_suffix }}:${{ inputs.image_tag }}
+        if: ${{ matrix.image.use_singularity }}

.github/workflows/__check_docker_files.yaml

@@ -0,0 +1,57 @@
+# Check if Docker images have changed. In that case, proposes images suffix and
+# tag.
+
+name: Check Docker files
+
+on:
+  workflow_call:
+    inputs:
+      event_name:
+        description: "Event name of the calling workflow"
+        required: true
+        type: string
+
+    outputs:
+      docker_files_have_changed:
+        description: "True if any Docker file was modified"
+        value: ${{ jobs.check_docker_files.outputs.docker_files_have_changed }}
+      image_suffix:
+        description: "Suffix of the images"
+        value: ${{ jobs.check_docker_files.outputs.image_suffix }}
+      image_tag:
+        description: "Tag of the images"
+        value: ${{ jobs.check_docker_files.outputs.image_tag }}
+
+jobs:
+  check_docker_files:
+    runs-on: ubuntu-latest
+
+    outputs:
+      # true if any Docker file was modified in the PR (PR mode) or since last pushed commit (push mode)
+      docker_files_have_changed: ${{ steps.get_changed_docker_files.outputs.any_changed == 'true' }}
+      # use "pr" as image name suffix if on PR mode and if any Docker file was modified, otherwise use "main";
+      # this is intended to avoid a PR to alter Docker images for other PRs or for the main branch
+      image_suffix: ${{ steps.get_changed_docker_files.outputs.any_changed == 'true' && inputs.event_name == 'pull_request' && 'pr' || 'main' }}
+      # use "<hash>" as image name tag if on PR mode and if any Docker file was modified, otherwise use "latest";
+      # this is intended to distinguish PR images from each other
+      image_tag: ${{ steps.get_changed_docker_files.outputs.any_changed == 'true' && inputs.event_name == 'pull_request' && github.sha || 'latest' }}
+
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          fetch-depth: 0
+
+      - name: Get changed Docker files
+        id: get_changed_docker_files
+        uses: tj-actions/changed-files@v42
+        with:
+          files: docker/**/Dockerfile
+
+      - name: List changed Docker files
+        if: ${{ steps.get_changed_docker_files.outputs.any_changed == 'true' }}
+        env:
+          ALL_CHANGED_FILES: ${{ steps.get_changed_docker_files.outputs.all_changed_files }}
+        run: |
+          for file in $ALL_CHANGED_FILES; do
+            echo "$file was changed"
+          done

.github/workflows/build_test.yaml

@@ -29,34 +29,10 @@ jobs:
         clangFormatVersion: 12
 
   check_docker_files:
-    runs-on: ubuntu-latest
-
-    outputs:
-      # true if any Docker file was modified in the PR (PR mode) or since last pushed commit (push mode)
-      docker_files_have_changed: ${{ steps.get_changed_docker_files.outputs.any_changed == 'true' }}
-      # use "pr" as image name suffix if on PR mode and if any Docker file was modified, otherwise use "main"
-      # this is intended to avoid a PR test to alter Docker images for other PRs or for the main branch
-      image_name_suffix: ${{ steps.get_changed_docker_files.outputs.any_changed == 'true' && github.event_name == 'pull_request' && 'pr' || 'main' }}
-
-    steps:
-      - uses: actions/checkout@v4
-        with:
-          fetch-depth: 0
+    uses: ./.github/workflows/__check_docker_files.yaml
 
-      - name: Get changed Dockerfiles
-        id: get_changed_docker_files
-        uses: tj-actions/changed-files@v42
-        with:
-          files: docker/**/Dockerfile
-
-      - name: List changed Dockerfiles
-        if: ${{ steps.docker_files_have_changed.outputs.any_changed == 'true' }}
-        env:
-          ALL_CHANGED_FILES: ${{ steps.docker_files_have_changed.outputs.all_changed_files }}
-        run: |
-          for file in "$ALL_CHANGED_FILES"; do
-            echo "$file was changed"
-          done
+    with:
+      event_name: ${{ github.event_name }}
 
   build_base:
     needs: check_docker_files
@@ -66,7 +42,8 @@ jobs:
     uses: ./.github/workflows/__build_base.yaml
 
     with:
-      image_name_suffix: ${{ needs.check_docker_files.outputs.image_name_suffix }}
+      image_suffix: ${{ needs.check_docker_files.outputs.image_suffix }}
+      image_tag: ${{ needs.check_docker_files.outputs.image_tag }}
 
   build:
     runs-on: ubuntu-latest
@@ -82,18 +59,22 @@ jobs:
       matrix:
         backend:
           - name: openmp
+            image: gcc
             c_compiler: gcc
             cxx_compiler: g++
             cmake_flags: -DKokkos_ENABLE_OPENMP=ON
           - name: cuda
+            image: nvcc
             c_compiler: gcc
             cxx_compiler: g++
             cmake_flags: -DKokkos_ENABLE_CUDA=ON -DKokkos_ARCH_AMPERE80=ON
           - name: hip
+            image: rocm
             c_compiler: hipcc
             cxx_compiler: hipcc
             cmake_flags: -DKokkos_ENABLE_HIP=ON -DKokkos_ARCH_VEGA90A=ON
           - name: sycl
+            image: intel
             c_compiler: icx
             cxx_compiler: icpx
             # building for Intel PVC was unsuccessful without the proper device
@@ -124,7 +105,7 @@ jobs:
 
       - name: Configure
         run: |
-          docker run -v ${{ github.workspace }}:/work ghcr.io/cexa-project/kokkos-fft/base_${{ matrix.backend.name }}_${{ needs.check_docker_files.outputs.image_name_suffix }} \
+          docker run -v ${{ github.workspace }}:/work ghcr.io/cexa-project/kokkos-fft/base_${{ matrix.backend.image }}_${{ needs.check_docker_files.outputs.image_suffix }}:${{ needs.check_docker_files.outputs.image_tag }} \
             cmake -B build \
             -DCMAKE_INSTALL_PREFIX=/work/install \
             -DCMAKE_BUILD_TYPE=${{ env.BUILD_TYPE }} \
@@ -139,7 +120,7 @@ jobs:
 
       - name: Build
         run: |
-          docker run -v ${{ github.workspace }}:/work ghcr.io/cexa-project/kokkos-fft/base_${{ matrix.backend.name }}_${{ needs.check_docker_files.outputs.image_name_suffix }} \
+          docker run -v ${{ github.workspace }}:/work ghcr.io/cexa-project/kokkos-fft/base_${{ matrix.backend.image }}_${{ needs.check_docker_files.outputs.image_suffix }}:${{ needs.check_docker_files.outputs.image_tag }} \
             cmake --build build -j $(( $(nproc) * 2 + 1 ))
 
       - name: Prepare artifacts
@@ -157,21 +138,21 @@ jobs:
 
       - name: Install
         run: |
-          docker run -v ${{ github.workspace }}:/work ghcr.io/cexa-project/kokkos-fft/base_${{ matrix.backend.name }}_${{ needs.check_docker_files.outputs.image_name_suffix }} \
+          docker run -v ${{ github.workspace }}:/work ghcr.io/cexa-project/kokkos-fft/base_${{ matrix.backend.image }}_${{ needs.check_docker_files.outputs.image_suffix }}:${{ needs.check_docker_files.outputs.image_tag }} \
             cmake --install build
 
       - name: Configure and build test code
         # Use the built and installed Kokkos FFT library to build a test code
         run: |
-          docker run -v ${{ github.workspace }}:/work ghcr.io/cexa-project/kokkos-fft/base_${{ matrix.backend.name }}_${{ needs.check_docker_files.outputs.image_name_suffix }} \
+          docker run -v ${{ github.workspace }}:/work ghcr.io/cexa-project/kokkos-fft/base_${{ matrix.backend.image }}_${{ needs.check_docker_files.outputs.image_suffix }}:${{ needs.check_docker_files.outputs.image_tag }} \
             cmake -B build_test \
             -DCMAKE_BUILD_TYPE=${{env.BUILD_TYPE}} \
             -DCMAKE_C_COMPILER=${{ matrix.backend.c_compiler }} \
             -DCMAKE_CXX_COMPILER=${{ matrix.backend.cxx_compiler }} \
             -DCMAKE_CXX_STANDARD=17 \
             -DCMAKE_PREFIX_PATH=/work/install \
             install_test
-          docker run -v ${{ github.workspace }}:/work ghcr.io/cexa-project/kokkos-fft/base_${{ matrix.backend.name }}_${{ needs.check_docker_files.outputs.image_name_suffix }} \
+          docker run -v ${{ github.workspace }}:/work ghcr.io/cexa-project/kokkos-fft/base_${{ matrix.backend.image }}_${{ needs.check_docker_files.outputs.image_suffix }}:${{ needs.check_docker_files.outputs.image_tag }} \
             cmake --build build_test -j $(( $(nproc) * 2 + 1 ))
 
   test:
@@ -189,9 +170,11 @@ jobs:
         backend:
           # run CUDA tests on Ruche supercomputer
           - name: cuda
+            image: nvcc
             runner: [self-hosted, cuda]
           # run OpenMP tests on Azure server
           - name: openmp
+            image: gcc
             runner: ubuntu-latest
 
     steps:
@@ -206,18 +189,18 @@ jobs:
 
       - name: Pull Singularity image
         # pulling the image in advance seems necessary as sometimes invoking `singularity run` on the image URL fails because it cannot find ghcr.io
-        run: singularity pull oras://ghcr.io/cexa-project/kokkos-fft/base_${{ matrix.backend.name }}_singularity_${{ needs.check_docker_files.outputs.image_name_suffix }}:latest
+        run: singularity pull oras://ghcr.io/cexa-project/kokkos-fft/base_${{ matrix.backend.image }}_singularity_${{ needs.check_docker_files.outputs.image_suffix }}:${{ needs.check_docker_files.outputs.image_tag }}
         if: ${{ matrix.backend.name == 'cuda' }}
 
       - name: Run CUDA tests within Slurm job and Singularity image
         run: |
           srun --nodes=1 --time=01:00:00 -p gpua100 --gres=gpu:1 \
-          singularity run --nv --bind $PWD/build:/work/build -H /work/build base_${{ matrix.backend.name }}_singularity_${{ needs.check_docker_files.outputs.image_name_suffix }}_latest.sif \
-          ctest
+          singularity run --nv --bind $PWD/build:/work/build -H /work/build base_${{ matrix.backend.image }}_singularity_${{ needs.check_docker_files.outputs.image_suffix }}_${{ needs.check_docker_files.outputs.image_tag }}.sif \
+          ctest --output-on-failure
         if: ${{ matrix.backend.name == 'cuda' }}
 
       - name: Run OpenMP tests within Docker image
         run: |
-          docker run -v $PWD/build:/work/build -w /work/build ghcr.io/cexa-project/kokkos-fft/base_${{ matrix.backend.name }}_${{ needs.check_docker_files.outputs.image_name_suffix }} \
-          ctest
+          docker run -v $PWD/build:/work/build -w /work/build ghcr.io/cexa-project/kokkos-fft/base_${{ matrix.backend.image }}_${{ needs.check_docker_files.outputs.image_suffix }}:${{ needs.check_docker_files.outputs.image_tag }} \
+          ctest --output-on-failure
         if: ${{ matrix.backend.name == 'openmp' }}

.github/workflows/pre_build_base.yaml

@@ -1,14 +1,28 @@
-# Recreate the base images inconditionnaly on a regural basis. This is to
-# ensure that anybody using the Docker files for local development does not
-# encounter weird bugs unnoticed by the CI, because the images would be too
-# old.
+# Recreate the base images inconditionnaly on a regural basis and on push on
+# the main branch. This is to ensure that anybody using the Docker files for
+# local development does not encounter weird bugs unnoticed by the CI, because
+# the images would be too old.
 
 name: Pre-build base images
 
 on:
   schedule:
     - cron: "0 1 2,16 * *" # every 2nd and 16th of the month at 1am UTC
+  push:
+    branches:
+      - main
 
 jobs:
+  check_docker_files:
+    uses: ./.github/workflows/__check_docker_files.yaml
+
+    with:
+      event_name: ${{ github.event_name }}
+
   build_base:
+    needs: check_docker_files
+
+    # run inconditionnaly on schedule mode or if Docker files changed on other modes
+    if: ${{ github.event_name == 'schedule' || needs.check_docker_files.outputs.docker_files_have_changed == 'true' }}
+
     uses: ./.github/workflows/__build_base.yaml

docker/hip/Dockerfile → docker/rocm/Dockerfile

@@ -1,4 +1,4 @@
-ARG BASE=rocm/dev-ubuntu-20.04:5.4
+ARG BASE=rocm/dev-ubuntu-20.04
 FROM $BASE
 
 ARG ADDITIONAL_PACKAGES