Add Cray psmp H100 tester (#4868)

ci/cscs_daint_spack_psmp_h100.yaml

@@ -0,0 +1,59 @@
+include:
+  - remote: 'https://gitlab.com/cscs-ci/recipes/-/raw/master/templates/v2/.ci-ext.yml'
+
+stages:
+  - deps
+  - cp2k
+
+variables:
+  REPOSITORY: cp2k_daint
+  BUILDER: spack
+  VERSION: psmp
+  TAG: "${BUILDER}_${VERSION}"
+  UBUNTU_IMAGE_NAME: "nvidia/cuda:12.4.1-devel-ubuntu22.04"
+  CSCS_IMAGE_NAME: "${CSCS_REGISTRY_PATH}/${REPOSITORY}:${TAG}_${CI_COMMIT_SHORT_SHA}"
+  DOCKERHUB_IMAGE_NAME: "docker.io/${DOCKERHUB_USERNAME}/${REPOSITORY}:${TAG}"
+
+build deps daint:
+  extends: .container-builder-cscs-gh200
+  stage: deps
+  timeout: 1h
+  before_script:
+    - DEPS_IMAGE_TAG=`cat ci/cscs_daint_${TAG}.yaml ci/docker/build_${CI_JOB_STAGE}_${BUILDER}.Dockerfile $(find tools/spack -type f) $(find cmake -type f) | sha256sum - | head -c 16`
+    - export PERSIST_IMAGE_NAME=${CSCS_REGISTRY_PATH}/base/${REPOSITORY}:${TAG}_${DEPS_IMAGE_TAG}
+    - echo "DEPS_IMAGE_NAME=${PERSIST_IMAGE_NAME}" | tee build.env
+  artifacts:
+    reports:
+      dotenv: build.env
+  variables:
+    DOCKER_BUILD_ARGS: '[
+      "BASE_IMAGE",
+      "CP2K_VERSION",
+      "FEATURE_FLAGS"
+    ]'
+    DOCKERFILE: ci/docker/build_${CI_JOB_STAGE}_${BUILDER}.Dockerfile
+    BASE_IMAGE: ${UBUNTU_IMAGE_NAME}
+    CP2K_VERSION: ${VERSION}
+    FEATURE_FLAGS: "-gpu H100 -df all""
+
+build cp2k daint:
+  extends: .container-builder-cscs-gh200
+  stage: cp2k
+  needs: ["build deps daint"]
+  timeout: 1h
+  before_script:
+    - echo ${DOCKERHUB_TOKEN} | podman login docker.io --username ${DOCKERHUB_USERNAME} --password-stdin
+  after_script:
+    - podman images
+    - podman push ${PERSIST_IMAGE_NAME} ${DOCKERHUB_IMAGE_NAME}
+  variables:
+    DOCKER_BUILD_ARGS: '[
+      "BASE_IMAGE",
+      "DEPS_IMAGE",
+      "FEATURE_FLAGS"
+    ]'
+    DOCKERFILE: ci/docker/build_${CI_JOB_STAGE}_${BUILDER}.Dockerfile
+    BASE_IMAGE: ${UBUNTU_IMAGE_NAME}
+    DEPS_IMAGE: ${DEPS_IMAGE_NAME}
+    FEATURE_FLAGS: "-gpu H100 -df all"
+    PERSIST_IMAGE_NAME: ${CSCS_IMAGE_NAME}

make_cp2k.sh

@@ -569,6 +569,7 @@ echo "INSTALL_MESSAGE     = ${INSTALL_MESSAGE}"
 echo "IN_CONTAINER        = ${IN_CONTAINER}"
 echo "MPI_MODE            = ${MPI_MODE}"
 echo "NUM_PROCS           = ${NUM_PROCS} (processes)"
+echo "Physical cores      = $(lscpu -p=Core,Socket | grep -v '#' | sort -u | wc -l) (host view)"
 echo "REBUILD_CP2K        = ${REBUILD_CP2K}"
 echo "RUN_TEST            = ${RUN_TEST}"
 if [[ "${RUN_TEST}" == "yes" ]]; then
@@ -636,7 +637,8 @@ if ((CUDA_ARCH > 0)); then
   # Check if the selected CUDA arch is valid when the nvidia-smi command is available
   if command -v nvidia-smi &> /dev/null; then
     nvidia-smi
-    HOST_CUDA_ARCH=$(nvidia-smi --query-gpu=compute_cap --format=csv,noheader | awk '{print 10*$1}')
+    CUDA_VERSION=$(nvidia-smi | grep "CUDA Version:" | awk '{print $9}')
+    HOST_CUDA_ARCH=$(nvidia-smi --query-gpu=compute_cap --format=csv,noheader | tail -n 1 | awk '{print 10*$1}')
     if ((CUDA_ARCH > HOST_CUDA_ARCH)); then
       echo "ERROR: The requested CUDA arch (${CUDA_ARCH}) is larger than the maximum"
       echo "       CUDA arch (${HOST_CUDA_ARCH}) supported by the host system"
@@ -657,6 +659,7 @@ if ((CUDA_ARCH > 0)); then
   CMAKE_CUDA_FLAGS="$(printf '%s\n' "${out[*]}")"
   echo "CMAKE_CUDA_FLAGS    = ${CMAKE_CUDA_FLAGS}"
   [[ -n "${CUDA_VERSION:-}" ]] && echo "CUDA_VERSION        = ${CUDA_VERSION}"
+  [[ -n "${CUDA_HOME:-}" ]] && echo "CUDA_HOME           = ${CUDA_HOME}"
   echo ""
 else
   CMAKE_CUDA_FLAGS="-DCP2K_USE_ACCEL=OFF"
@@ -814,6 +817,12 @@ if [[ ! -d "${SPACK_BUILD_PATH}" ]]; then
   # Activate CUDA in the spack configuration file if requested
   if ((CUDA_ARCH > 0)); then
     sed -E -e "0,/~cuda/s//+cuda cuda_arch=${CUDA_ARCH}/" -i "${CP2K_CONFIG_FILE}"
+    if [[ -n "${CUDA_VERSION:-}" ]]; then
+      sed -E -e "s/spec:\s+cuda@[.0-9]*/spec: cuda@${CUDA_VERSION}/" -i "${CP2K_CONFIG_FILE}"
+    fi
+    if [[ -n "${CUDA_HOME:-}" ]]; then
+      sed -E -e "s|prefix: /usr/local/cuda|prefix: ${CUDA_HOME}|" -i "${CP2K_CONFIG_FILE}"
+    fi
   fi
 
   # Apply Cray specific adaptation of the spack configuration if requested (CSCS)

tools/dashboard/dashboard.conf

@@ -5,6 +5,13 @@ host:        Alps Daint (CSCS)
 report_url:  https://storage.googleapis.com/cp2k-ci/dashboard_cscs-daint-psmp_report.txt
 info_url:    https://gitlab.com/cscs-ci/ci-testing/webhook-ci/mirrors/3546923033037693/3295760759903023/-/pipelines
 
+[cscs-daint-psmp-h100]
+sortkey:     121
+name:        CSCS Daint (psmp, H100)
+host:        Alps Daint (CSCS)
+report_url:  https://storage.googleapis.com/cp2k-ci/dashboard_cscs-daint-psmp-h100_report.txt
+info_url:    https://gitlab.com/cscs-ci/ci-testing/webhook-ci/mirrors/3546923033037693/3295760759903023/-/pipelines
+
 [cscs-eiger-ssmp]
 sortkey:     201
 name:        CSCS Eiger (ssmp)

tools/docker/cp2k-ci.conf

@@ -5,6 +5,13 @@ tags:          daily
 cpu:           0.1
 nodepools:     default-pool
 
+[cscs-daint-psmp-h100]
+display_name:  CSCS Daint (psmp, H100)
+cscs_pipeline: daint-spack-psmp-h100
+tags:          daily
+cpu:           0.1
+nodepools:     default-pool
+
 [cscs-eiger-ssmp]
 display_name:  CSCS Eiger (ssmp)
 cscs_pipeline: eiger-spack-ssmp