gpu to gh200

ci/.gitlab-ci.yml

@@ -8,9 +8,9 @@ include:
 #  - local: 'ci/cpu/gcc11_release_stdexec.yml'
 #  - local: 'ci/cpu/gcc11_debug_stdexec.yml'
 #  - local: 'ci/cpu/gcc12_release_cxx20.yml'
-  - local: 'ci/cpu/gcc13_codecov.yml'
+#  - local: 'ci/cpu/gcc13_codecov.yml'
 #  - local: 'ci/cpu/gcc13_release.yml'
-  - local: 'ci/cuda/gcc11_release.yml'
+  - local: 'ci/cuda/gcc13_release.yml'
 #  - local: 'ci/cuda/gcc11_release_scalapack.yml'
 #  - local: 'ci/cuda/gcc11_codecov.yml'
 #  - local: 'ci/cuda/gcc11_debug_scalapack.yml'

ci/common-ci.yml

@@ -10,8 +10,7 @@ stages:
 ## BUILDS
 ##
 
-.build_deps_common:
-  extends: .container-builder
+.build_deps_common_base:
   stage: build_deps
   timeout: 6 hours
   before_script:
@@ -61,7 +60,17 @@ stages:
     COMMON_SPACK_ENVIRONMENT: ci/docker/common.yaml
     USE_CODECOV: "false"
 
-.build_common:
+.build_deps_common:
+  extends:
+    - .container-builder-cscs-zen2
+    - .build_deps_common_base
+
+.build_deps_common_gh200:
+  extends:
+    - .container-builder-cscs-gh200
+    - .build_deps_common_base
+
+.build_common_base:
   extends: .container-builder
   stage: build
   timeout: 2 hours
@@ -91,6 +100,16 @@ stages:
     paths:
       - pipeline.yml
 
+.build_common:
+  extends:
+    - .container-builder-cscs-zen2
+    - .build_common_base
+
+.build_common_gh200:
+  extends:
+    - .container-builder-cscs-gh200
+    - .build_common_base
+
 .build_for_daint-mc:
   variables:
     RUNNER: ".container-runner-daint"
@@ -112,6 +131,14 @@ stages:
     THREADS_MAX_PER_TASK: 32
     THREADS_PER_NODE: 256
 
+.build_for_alps_gh200:
+  variables:
+    RUNNER: ".container-runner-todi"
+    SLURM_CONSTRAINT: gpu
+    # 64 / 2 to avoid ranks on multiple sockets for RANK6
+    THREADS_MAX_PER_TASK: 32
+    THREADS_PER_NODE: 256
+
 .run_common:
   stage: test
   trigger:

ci/cuda/gcc13_release.yml

@@ -0,0 +1,29 @@
+include:
+  - local: 'ci/common-ci.yml'
+
+cuda gcc13 release deps:
+  extends: .build_deps_common_gh200
+  variables:
+    BASE_IMAGE: docker.io/nvidia/12.6.1-devel-ubuntu24.04
+    COMPILER: gcc@13
+    SPACK_ENVIRONMENT: ci/docker/release-cuda-gh200.yaml
+    BUILD_IMAGE: $CSCS_REGISTRY_PATH/cuda-gh200-gcc13-release/build
+
+cuda gcc13 release build:
+  extends:
+    - .build_common_gh200
+    - .build_for_alps_gh200
+  needs:
+    - cuda gcc13 release deps
+  variables:
+    DEPLOY_BASE_IMAGE: docker.io/ubuntu:24.04
+    DEPLOY_IMAGE: $CSCS_REGISTRY_PATH/cuda-gh200-gcc13-release/deploy:$CI_COMMIT_SHA
+
+cuda gcc13 release test:
+  extends: .run_common
+  needs:
+    - cuda gcc13 release build
+  trigger:
+    include:
+      - artifact: pipeline.yml
+        job: cuda gcc13 release build

ci/docker/common-gh200.yaml

@@ -0,0 +1,51 @@
+#
+# Distributed Linear Algebra with Future (DLAF)
+#
+# Copyright (c) 2018-2024, ETH Zurich
+# All rights reserved.
+#
+# Please, refer to the LICENSE file in the root directory.
+# SPDX-License-Identifier: BSD-3-Clause
+#
+
+packages:
+  all:
+    target: [x86_64]
+  # Set intel MKL as default blas, lapack and scalapack provider.
+  # Can be overwritten in environments if needed.
+  blas:
+    require: 'nvpl-blas'
+  lapack:
+    require: 'nvpl-lapack'
+  scalapack:
+    require: 'netlib-scalapack'
+  mpi:
+    require: 'mpich'
+  blaspp:
+    variants:
+      - '~cuda'
+      - '~openmp'
+      - '~rocm'
+  nvpl-blas:
+    require:
+      - 'threads=openmp'
+  nvpl-lapack:
+    require:
+      - 'threads=openmp'
+  openblas:
+    variants:
+      - 'threads=openmp'
+  mpich:
+    # Fix version to have better control.
+    require:
+      - '@4.2.1'
+    variants:
+      - '~fortran'
+      - '~libxml2'
+  hwloc:
+    variants:
+      - '~libxml2'
+  git:
+    # Force git as non-buildable to allow deprecated versions in environments
+    # https://github.com/spack/spack/pull/30040
+    buildable: false