common-ci.yml

#
# Distributed Linear Algebra with Future (DLAF)
#
# Copyright (c) ETH Zurich
# All rights reserved.
#
# Please, refer to the LICENSE file in the root directory.
# SPDX-License-Identifier: BSD-3-Clause
#
include:
  - remote: 'https://gitlab.com/cscs-ci/recipes/-/raw/master/templates/v2/.ci-ext.yml'
  - local: 'ci/ci-ext-custom.yml'

stages:
  - build_deps
  - build
  - test

variables:
  FF_TIMESTAMPS: true

## BUILD DEPS

.build_deps_common_base:
  stage: build_deps
  timeout: 2 hours
  before_script:
    - echo $DOCKERHUB_TOKEN | podman login docker.io -u $DOCKERHUB_USERNAME --password-stdin
    - TAG_IMAGE=`echo ${BASE_IMAGE##*/} | sed 's/[:]//g'`
    - TAG_APTGET=`echo ${EXTRA_APTGET} | sha256sum - | head -c 6`
    - TAG_COMPILER=`echo ${COMPILER}_CXX${CXXSTD} | sed 's/[@]//g'`
    - TAG_DOCKERFILE=`sha256sum $DOCKERFILE | head -c 12`
    - TAG_SPACK=`echo $SPACK_SHA | sed "s/develop-//g" | head -c 8`
    - TAG_SPACK_PACKAGES=`echo $SPACK_PACKAGES_SHA | sed "s/develop-//g" | head -c 8`
    - TAG_ALPS_CLUSTER_CONFIG=`echo $ALPS_CLUSTER_CONFIG_SHA | head -c 8`
    - TAG_REPO=`find $SPACK_DLAF_REPO -type f -exec sha256sum {} \; | sha256sum - | head -c 12`
    - TAG_ENVIRONMENT=`cat $SPACK_ENVIRONMENT $COMMON_SPACK_ENVIRONMENT | sha256sum | head -c 12`
    - TAG=${TAG_IMAGE}-${TAG_APTGET}-${TAG_COMPILER}-MKL${USE_MKL}-${TAG_DOCKERFILE}-${TAG_SPACK}-${TAG_SPACK_PACKAGES}-${TAG_ALPS_CLUSTER_CONFIG}-${TAG_REPO}-${TAG_ENVIRONMENT}
    - export PERSIST_IMAGE_NAME=$DEPS_IMAGE:$TAG
    - echo "DEPS_IMAGE=$PERSIST_IMAGE_NAME" > build.env
    - echo "USE_MKL=$USE_MKL" >> build.env
    - echo "USE_CODECOV=$USE_CODECOV" >> build.env
    - 'echo "INFO: Building image $PERSIST_IMAGE_NAME"'
    - 'echo "INFO: Using NUM_CORES_BUILD_DEPS=$NUM_CORES_BUILD_DEPS"'
  artifacts:
    reports:
      dotenv: build.env
  variables:
    SPACK_SHA: fbadf557433ae3d8cb8a2b63a81303de42a9fd6f
    SPACK_PACKAGES_SHA: bf71e784ba28818a05468ce4ffef405fa7c3d151
    ALPS_CLUSTER_CONFIG_SHA: 3c26eb25bdfa3151b637c040154a55bb9da31ba1
    SPACK_DLAF_REPO: ./spack
    DOCKER_BUILD_ARGS: '[
        "BASE_IMAGE",
        "SPACK_SHA",
        "SPACK_PACKAGES_SHA",
        "ALPS_CLUSTER_CONFIG_SHA",
        "EXTRA_APTGET",
        "COMPILER",
        "CXXSTD",
        "SPACK_ENVIRONMENT",
        "COMMON_SPACK_ENVIRONMENT",
        "SPACK_DLAF_REPO",
        "USE_MKL",
        "NUM_PROCS=$NUM_CORES_BUILD_DEPS"
      ]'
    # default configuration variables
    # can be overwritten in the configuration as needed
    DOCKERFILE: ci/docker/build.Dockerfile
    BASE_IMAGE: docker.io/ubuntu:24.04
    EXTRA_APTGET: ""
    CXXSTD: 17
    USE_MKL: "OFF"
    USE_CODECOV: "false"

.build_deps_common:
  extends:
    - .container-builder-cscs-zen2
    - .build_deps_common_base
  variables:
    COMMON_SPACK_ENVIRONMENT: ci/docker/common.yaml

.build_deps_common_amdgpu:
  extends: .build_deps_common
  variables:
    COMMON_SPACK_ENVIRONMENT: ci/docker/common-amdgpu.yaml

.build_deps_common_gh200:
  timeout: 6 hours
  extends:
    - .container-builder-cscs-gh200
    - .build_deps_common_base
  variables:
    COMMON_SPACK_ENVIRONMENT: ci/docker/common-gh200.yaml
    SLURM_TIMELIMIT: '00:45:00'

## BUILD DLAF

.build_common_base:
  stage: build
  timeout: 2 hours
  before_script:
    - 'echo $DOCKERHUB_TOKEN | podman login docker.io -u $DOCKERHUB_USERNAME --password-stdin'
    - 'echo "INFO: Using NUM_CORES_BUILD_DLAF=$NUM_CORES_BUILD_DLAF"'
  after_script:
    - podman run -v $PWD/ci/ctest_to_gitlab.sh:/ctest_to_gitlab.sh $DLAF_IMAGE /ctest_to_gitlab.sh "$DLAF_IMAGE" "$USE_CODECOV" "$THREADS_MAX_PER_TASK" "$THREADS_PER_NODE" "$RUNNER" > pipeline.yml
  variables:
    PERSIST_IMAGE_NAME: $DLAF_IMAGE
    DOCKER_BUILD_ARGS: '[
        "DEPS_IMAGE",
        "DLAF_LD_PRELOAD",
        "PIP_OPTS",
        "NUM_PROCS=$NUM_CORES_BUILD_DLAF"
      ]'
    # default configuration variables
    # can be overwritten in the configuration as needed
    DOCKERFILE: ci/docker/deploy.Dockerfile
    PIP_OPTS: ""
  artifacts:
    paths:
      - pipeline.yml

.build_common:
  extends:
    - .container-builder-cscs-zen2
    - .build_common_base
  variables:
    DLAF_LD_PRELOAD: "/lib/x86_64-linux-gnu/libSegFault.so"

.build_common_gh200:
  timeout: 6 hours
  extends:
    - .container-builder-cscs-gh200
    - .build_common_base
  variables:
    DLAF_LD_PRELOAD: "/lib/aarch64-linux-gnu/libSegFault.so"
    SLURM_TIMELIMIT: '00:45:00'

.build_for_eiger:
  variables:
    RUNNER: ".dlaf-container-runner-eiger"
    THREADS_MAX_PER_TASK: 32
    THREADS_PER_NODE: 256

.build_for_beverin_mi200:
  variables:
    RUNNER: ".dlaf-container-runner-beverin-mi200"
    THREADS_MAX_PER_TASK: 16
    THREADS_PER_NODE: 256

.build_for_alps_gh200:
  variables:
    RUNNER: ".dlaf-container-runner-daint-gh200"
    # 64 / 2 to avoid ranks on multiple sockets for RANK6
    THREADS_MAX_PER_TASK: 32
    THREADS_PER_NODE: 256

## RUN

.run_common:
  stage: test
  trigger:
    strategy: depend
    forward:
      pipeline_variables: true
  variables:
    PIKA_USE_GUARD_PAGES: 1