CI job script refactoring (#582)

Author: ChristopherBignamini

ci/cuda12/run_debug_sm90.yml

@@ -1,238 +1,46 @@
 include:
-  - local: 'ci/common.yml'
-
-variables:
-  TEST_INSTALL_DIR: 'sphexa+spack/store/linux-neoverse_v2/sphexa-*/sbin'
-  APP_INSTALL_DIR: 'sphexa+spack/store/linux-neoverse_v2/sphexa-*/bin'
+  - local: 'ci/cuda12/run_template_sm90.yml'
 
 sph-run-cuda12-sm90-debug-1rank:
-  stage: sphexa_run
+  extends: [.sph-run-cuda12-sm90-1rank]
   needs: ['sph-build-cuda12-sm90-debug']
-  image: prgenv-gnu/25.11:v1
-  extends: [.uenv-runner-daint-gh200]
-  variables:
-    SLURM_TIMELIMIT: '01:00:00'
-    WITH_UENV_VIEW: 'default'
-    # SLURM_PARTITION: 'debug'
-    SLURM_JOB_NUM_NODES: 1
-    SLURM_NTASKS: 1
-  before_script:
-    - pwd
-    - date
-    - $APP_INSTALL_DIR/sphexa --help
-  script:
-    - $TEST_INSTALL_DIR/coord_samples/coordinate_test
-    - $TEST_INSTALL_DIR/hydro/turbulence_tests
-    - ln -fs $TEST_INSTALL_DIR/hydro/example_data.txt .
-    - $TEST_INSTALL_DIR/hydro/sph_tests
-    - $TEST_INSTALL_DIR/performance/octree_perf
-    - $TEST_INSTALL_DIR/performance/peers_perf
-    - $TEST_INSTALL_DIR/ryoanji/cpu_unit_tests/ryoanji_cpu_unit_tests
-    - $TEST_INSTALL_DIR/ryoanji/unit_tests/ryoanji_unit_tests
-    # NOTE:slow    - $TEST_INSTALL_DIR/performance/hilbert_perf
-    - $TEST_INSTALL_DIR/unit/component_units_omp
-    - $TEST_INSTALL_DIR/unit/component_units
-    # --- grackle
-    # https://github.com/grackle-project/grackle_data_files/blob/928696482fbe15d9bac4382de6134d95568f099c/input/CloudyData_UVB%3DHM2012.h5
-    - wget https://jfrog.svc.cscs.ch/artifactory/cscs-reframe-tests/sphexa/CloudyData_UVB%3DHM2012.h5
-    - mkdir -p extern/grackle/grackle_repo/input
-    - mv CloudyData_UVB=HM2012.h5 extern/grackle/grackle_repo/input/
-    - export GRACKLE_DATA_FILE="$PWD/extern/grackle/grackle_repo/input/CloudyData_UVB=HM2012.h5"
-    - test -f "$GRACKLE_DATA_FILE"
-    - $TEST_INSTALL_DIR/physics/cooling_tests
-    # GPU tests:
-    - $TEST_INSTALL_DIR/hydro/sph_gpu_tests
-    - $TEST_INSTALL_DIR/performance/hilbert_perf_gpu
-    - $TEST_INSTALL_DIR/performance/neighbors_test_gpu
-    - $TEST_INSTALL_DIR/performance/octree_perf_gpu
-    - $TEST_INSTALL_DIR/unit_cuda/component_units_cuda
-    - date
 
 sph-run-cuda12-sm90-debug-2ranks:
-  stage: sphexa_run
+  extends: [.sph-run-cuda12-sm90-2ranks]
   needs: ['sph-build-cuda12-sm90-debug']
-  image: prgenv-gnu/25.11:v1
-  extends: [.uenv-runner-daint-gh200]
-  variables:
-    WITH_UENV_VIEW: 'default'
-    # SLURM_PARTITION: 'debug'
-    SLURM_JOB_NUM_NODES: 1
-    SLURM_NTASKS: 2
-  before_script:
-    - pwd
-    - date
-  script:
-    - $TEST_INSTALL_DIR/integration_mpi/box_mpi
-    - $TEST_INSTALL_DIR/integration_mpi/domain_2ranks
-    - $TEST_INSTALL_DIR/integration_mpi/domain_resize
-    - $TEST_INSTALL_DIR/integration_mpi/exchange_focus
-    - $TEST_INSTALL_DIR/integration_mpi/exchange_halos
-    - $TEST_INSTALL_DIR/integration_mpi/exchange_keys
-    - $TEST_INSTALL_DIR/integration_mpi/globaloctree
-    - $TEST_INSTALL_DIR/integration_mpi/hdf5io
-    - $TEST_INSTALL_DIR/ryoanji/global_upsweep_cpu
-    - $TEST_INSTALL_DIR/ryoanji/ryoanji_demo/ryoanji_demo
-    - $TEST_INSTALL_DIR/ryoanji/ryoanji_demo_mpi
-    - $TEST_INSTALL_DIR/physics/disk_tests
-    # GPU tests:
-    - $TEST_INSTALL_DIR/integration_mpi/assignment_gpu
-    - $TEST_INSTALL_DIR/integration_mpi/domain_gpu
-    - $TEST_INSTALL_DIR/integration_mpi/exchange_domain_gpu
-    - $TEST_INSTALL_DIR/integration_mpi/exchange_halos_gpu
-    - date
 
 sph-run-cuda12-sm90-debug-4ranks:
-  stage: sphexa_run
+  extends: [.sph-run-cuda12-sm90-4ranks]
   needs: ['sph-build-cuda12-sm90-debug']
-  image: prgenv-gnu/25.11:v1
-  extends: [.uenv-runner-daint-gh200]
-  variables:
-    WITH_UENV_VIEW: 'default'
-    # SLURM_PARTITION: 'debug'
-    SLURM_JOB_NUM_NODES: 1
-    SLURM_NTASKS: 4
-  before_script:
-    - pwd
-    - date
-  script:
-    - $TEST_INSTALL_DIR/integration_mpi/exchange_general_gpu # was -n5
-    - $TEST_INSTALL_DIR/ryoanji/global_forces_gpu
-    - $TEST_INSTALL_DIR/ryoanji/global_upsweep_gpu
-    - date
 
 sph-run-cuda12-sm90-debug-12ranks:
-  stage: sphexa_run
+  extends: [.sph-run-cuda12-sm90-12ranks]
   needs: ['sph-build-cuda12-sm90-debug']
-  image: prgenv-gnu/25.11:v1
-  extends: [.uenv-runner-daint-gh200]
-  variables:
-    WITH_UENV_VIEW: 'default'
-    # SLURM_PARTITION: 'debug'
-    SLURM_JOB_NUM_NODES: 1
-    SLURM_NTASKS: 12
-  before_script:
-    - pwd
-    - date
-  script:
-    - $TEST_INSTALL_DIR/integration_mpi/exchange_domain
-    - $TEST_INSTALL_DIR/integration_mpi/exchange_general
-    - $TEST_INSTALL_DIR/integration_mpi/focus_tree
-    - $TEST_INSTALL_DIR/integration_mpi/treedomain
-    - $TEST_INSTALL_DIR/integration_mpi/domain_nranks
-    - date
 
 sph-run-cuda12-sm90-debug-sphexa-cuda:
-  stage: sphexa_run
-  needs:
+  extends: [.sph-run-cuda12-sm90-sphexa-cuda]
+  needs: 
     - job: sph-build-cuda12-sm90-debug
     - job: sph-test-venv-setup
       artifacts: true
-  image: prgenv-gnu/25.11:v1
-  extends: [.uenv-runner-daint-gh200]
-  variables:
-    WITH_UENV_VIEW: 'default'
-    # SLURM_PARTITION: 'debug'
-    SLURM_JOB_NUM_NODES: 1
-    SLURM_NTASKS: 4
-    SLURM_CPUS_PER_TASK: 71
-  before_script:
-    - pwd
-    - date
-    - export OMP_NUM_THREADS=64
-  script:
-    - wget --quiet -O 50c.h5 https://zenodo.org/records/8369645/files/50c.h5
-    - $APP_INSTALL_DIR/sphexa-cuda --glass ./50c.h5 --init sedov --G 1.0 -n 40 -s 100 -w 10
-    - if [ "$SLURM_PROCID" -eq 0 ]; then mv constants.txt constants_ref.txt; fi
-    - wait
-    - $APP_INSTALL_DIR/sphexa-cuda --init dump_sedov.h5:4 -s 100
-    - |
-      if [ "$SLURM_PROCID" -eq 0 ]; then
-        awk 'start||$1==50 {print; start=1}' constants_ref.txt > constants_ref_tail.txt
-        . constant_comparison_venv/bin/activate
-        python ci/scripts/compare_constants.py constants_ref_tail.txt constants.txt "7,8"
-        if [ $? -ne 0 ]; then
-          exit 1
-        fi
-        deactivate
-      fi
-    - date
 
 sph-run-cuda12-sm90-debug-sphexa-cpu:
-  stage: sphexa_run
-  needs:
+  extends: [.sph-run-cuda12-sm90-sphexa-cpu]
+  needs: 
     - job: sph-build-cuda12-sm90-debug
     - job: sph-test-venv-setup
       artifacts: true
-  image: prgenv-gnu/25.11:v1
-  extends: [.uenv-runner-daint-gh200]
-  variables:
-    WITH_UENV_VIEW: 'default'
-    # SLURM_PARTITION: 'debug'
-    SLURM_JOB_NUM_NODES: 1
-    SLURM_NTASKS: 4
-    SLURM_CPUS_PER_TASK: 71
-  before_script:
-    - pwd
-    - date
-    - export OMP_NUM_THREADS=64
-  script:
-    - wget --quiet -O 50c.h5 https://zenodo.org/records/8369645/files/50c.h5
-    - $APP_INSTALL_DIR/sphexa --glass ./50c.h5 --init sedov --G 1.0 -n 40 -s 100 -w 10
-    - if [ "$SLURM_PROCID" -eq 0 ]; then mv constants.txt constants_ref.txt; fi
-    - wait
-    - $APP_INSTALL_DIR/sphexa --init dump_sedov.h5:4 -s 100
-    - |
-      if [ "$SLURM_PROCID" -eq 0 ]; then
-        awk 'start||$1==50 {print; start=1}' constants_ref.txt > constants_ref_tail.txt
-        . constant_comparison_venv/bin/activate
-        python ci/scripts/compare_constants.py constants_ref_tail.txt constants.txt "7,8"
-        if [ $? -ne 0 ]; then
-          exit 1
-        fi
-        deactivate
-      fi
-    - date
 
 sph-compare-constants-cuda12-sm90-debug-sphexa-cuda:
-  stage: sphexa_run
-  needs:
+  extends: [.sph-compare-constants-cuda12-sm90-sphexa-cuda]
+  needs: 
     - job: sph-build-cuda12-sm90-debug
     - job: sph-test-venv-setup
       artifacts: true
-  image: prgenv-gnu/25.11:v1
-  extends: [.uenv-runner-daint-gh200]
-  variables:
-    WITH_UENV_VIEW: 'default'
-    # SLURM_PARTITION: 'debug'
-    SLURM_JOB_NUM_NODES: 1
-    SLURM_NTASKS: 4
-    SLURM_CPUS_PER_TASK: 71
-  before_script:
-    - pwd
-    - date
-    - export OMP_NUM_THREADS=71
-  script:
-    - bash ci/scripts/run_constants_comparison.sh "$APP_INSTALL_DIR/sphexa-cuda"
-    - date
 
 sph-compare-constants-cuda12-sm90-debug-sphexa-cpu:
-  stage: sphexa_run
-  needs:
+  extends: [.sph-compare-constants-cuda12-sm90-sphexa-cpu]
+  needs: 
     - job: sph-build-cuda12-sm90-debug
     - job: sph-test-venv-setup
       artifacts: true
-  image: prgenv-gnu/25.11:v1
-  extends: [.uenv-runner-daint-gh200]
-  variables:
-    WITH_UENV_VIEW: 'default'
-    # SLURM_PARTITION: 'debug'
-    SLURM_JOB_NUM_NODES: 1
-    SLURM_NTASKS: 4
-    SLURM_CPUS_PER_TASK: 71
-  before_script:
-    - pwd
-    - date
-    - export OMP_NUM_THREADS=71
-  script:
-    - bash ci/scripts/run_constants_comparison.sh "$APP_INSTALL_DIR/sphexa"
-    - date