Add CSCS CI

.gitignore

@@ -25,6 +25,7 @@ __pycache__
 /*
 
 # except
+!ci
 !cmake
 !docs
 !docs_src

.python_package/.gitignore

@@ -4,3 +4,4 @@ dist/
 setup.cfg
 *.egg-info/
 src/gridtools_cpp/data
+build/

ci/base.Dockerfile

@@ -0,0 +1,63 @@
+ARG UBUNTU_VERSION=24.04
+ARG CUDA_VERSION
+FROM docker.io/nvidia/cuda:${CUDA_VERSION}-devel-ubuntu${UBUNTU_VERSION}
+ENV LANG C.UTF-8
+ENV LC_ALL C.UTF-8
+
+ARG DEBIAN_FRONTEND=noninteractive
+RUN apt-get update -qq && \
+    apt-get install -qq -y --no-install-recommends \
+    gfortran \
+    g++ \
+    gcc \
+    strace \
+    build-essential \
+    tar \
+    wget \
+    curl \
+    cmake \
+    ca-certificates \
+    zlib1g-dev \
+    libssl-dev \
+    libbz2-dev \
+    libsqlite3-dev \
+    llvm \
+    libncurses5-dev \
+    libncursesw5-dev \
+    xz-utils \
+    tk-dev \
+    libffi-dev \
+    liblzma-dev \
+    libreadline-dev \
+    python3-dev \
+    python3-pip \
+    git \
+    rustc \
+    htop && \
+    rm -rf /var/lib/apt/lists/*
+
+ARG MPICH_VERSION=3.3.2
+ARG MPICH_PATH=/usr/local
+RUN wget -q https://www.mpich.org/static/downloads/${MPICH_VERSION}/mpich-${MPICH_VERSION}.tar.gz && \
+    tar -xzf mpich-${MPICH_VERSION}.tar.gz && \
+    cd mpich-${MPICH_VERSION} && \
+    ./configure \
+    --disable-fortran \
+    --prefix=$MPICH_PATH && \
+    make install -j32 && \
+    rm -rf /root/mpich-${MPICH_VERSION}.tar.gz /root/mpich-${MPICH_VERSION}
+RUN echo "${MPICH_PATH}/lib" >> /etc/ld.so.conf.d/cscs.conf && ldconfig
+
+ENV CXX=${MPICH_PATH}/bin/mpicxx
+ENV CC=${MPICH_PATH}/bin/mpicc
+
+RUN wget --quiet https://archives.boost.io/release/1.85.0/source/boost_1_85_0.tar.gz && \
+    echo be0d91732d5b0cc6fbb275c7939974457e79b54d6f07ce2e3dfdd68bef883b0b boost_1_85_0.tar.gz > boost_hash.txt && \
+    sha256sum -c boost_hash.txt && \
+    tar xzf boost_1_85_0.tar.gz && \
+    mv boost_1_85_0/boost /usr/local/include/ && \
+    rm boost_1_85_0.tar.gz boost_hash.txt
+ENV BOOST_ROOT /usr/local/
+
+ENV CUDA_HOME /usr/local/cuda
+ENV CUDA_ARCH=${CUDA_ARCH}

ci/build.Dockerfile

@@ -0,0 +1,23 @@
+ARG BASE_IMAGE
+FROM $BASE_IMAGE
+
+COPY . /gridtools
+
+ARG BUILD_TYPE
+
+ENV GTRUN_BUILD_COMMAND='make -j 32'
+ENV GTCMAKE_Boost_NO_BOOST_CMAKE=ON
+ENV GTCMAKE_Boost_NO_SYSTEM_PATHS=ON
+ENV GTCMAKE_GT_TESTS_REQUIRE_FORTRAN_COMPILER=ON
+ENV GTCMAKE_GT_TESTS_REQUIRE_C_COMPILER=ON
+ENV GTCMAKE_GT_TESTS_REQUIRE_OpenMP=ON
+ENV GTCMAKE_GT_TESTS_REQUIRE_GPU=ON
+ENV GTCMAKE_GT_TESTS_MPI_WITH_MPI_EXECUTABLE=OFF
+ENV GTCMAKE_GT_TESTS_REQUIRE_Python=ON
+ENV GT_ENABLE_STENCIL_DUMP=ON
+ENV GTCMAKE_CMAKE_EXPORT_NO_PACKAGE_REGISTRY=ON
+
+RUN curl -LsSf https://astral.sh/uv/install.sh | sh
+ENV PATH="/root/.local/bin:${PATH}"
+
+RUN uv run /gridtools/pyutils/driver.py -v build -b ${BUILD_TYPE} -o build -i install -t install

ci/cscs.yml

@@ -0,0 +1,81 @@
+include:
+  - remote: "https://gitlab.com/cscs-ci/recipes/-/raw/master/templates/v2/.ci-ext.yml"
+
+stages:
+  - baseimage
+  - build
+  - test
+
+.build_baseimage:
+  stage: baseimage
+  # we create a tag that depends on the SHA value of ci/base.Dockerfile, this way
+  # a new base image is only built when the SHA of this file changes
+  # If there are more dependency files that should change the tag-name of the base container
+  # image, they can be added too.
+  # Since the base image name is runtime dependent, we need to carry the value of it to
+  # the following jobs via a dotenv file.
+  before_script:
+  # include build arguments in hash since we use a parameterized Docker file
+  - DOCKER_TAG=`echo "$(cat $DOCKERFILE) $DOCKER_BUILD_ARGS" | sha256sum | head -c 16`
+  - export PERSIST_IMAGE_NAME=$CSCS_REGISTRY_PATH/public/$ARCH/base/gridtools-ci:$DOCKER_TAG
+  - echo "BASE_IMAGE=$PERSIST_IMAGE_NAME" >> build.env
+  artifacts:
+    reports:
+      dotenv: build.env
+  variables:
+    DOCKERFILE: ci/base.Dockerfile
+    # change to 'always' if you want to rebuild, even if target tag exists already (if-not-exists is the default, i.e. we could also skip the variable)
+    CSCS_REBUILD_POLICY: if-not-exists
+    DOCKER_BUILD_ARGS: '["CUDA_VERSION=$CUDA_VERSION", "UBUNTU_VERSION=$UBUNTU_VERSION"]'
+build_baseimage_aarch64:
+  extends: [.container-builder-cscs-gh200, .build_baseimage]
+  variables:
+    CUDA_VERSION: 12.6.2
+    CUDA_ARCH: sm_90
+    UBUNTU_VERSION: 24.04
+    SLURM_TIMELIMIT: 10
+
+
+.build_image:
+  stage: build
+  variables:
+    # make sure we use a unique name here, otherwise we could create a race condition, when multiple pipelines
+    # are running.
+    PERSIST_IMAGE_NAME: $CSCS_REGISTRY_PATH/public/$ARCH/gridtools/gridtools-ci:$CI_COMMIT_SHA
+    DOCKERFILE: ci/build.Dockerfile
+    DOCKER_BUILD_ARGS: '["BASE_IMAGE=${BASE_IMAGE}", "BUILD_TYPE=release"]'
+build_image_aarch64:
+  extends: [.container-builder-cscs-gh200, .build_image]
+  variables:
+    SLURM_TIMELIMIT: 10
+
+.test_helper:
+  stage: test
+  image: $CSCS_REGISTRY_PATH/public/$ARCH/gridtools/gridtools-ci:$CI_COMMIT_SHA
+  variables:
+    GTRUN_WITH_SLURM: False # since we are already in a SLURM job
+    SLURM_JOB_NUM_NODES: 1
+    SLURM_TIMELIMIT: 10
+    CSCS_CUDA_MPS: 0
+
+test_aarch64:
+  extends: [.container-runner-daint-gh200, .test_helper]
+  script:
+  - cd /build && ctest -LE mpi --output-on-failure
+  variables:
+    SLURM_NTASKS: 1
+
+test_aarch64_mpi:
+  extends: [.container-runner-daint-gh200, .test_helper]
+  script:
+  - export LD_LIBRARY_PATH=/usr/lib64:$LD_LIBRARY_PATH
+  - export LD_PRELOAD=/usr/lib64/libmpi_gtl_cuda.so
+  - cd /build && ctest -L mpi --output-on-failure
+  variables:
+    NVIDIA_VISIBLE_DEVICES: all
+    SLURM_NTASKS: 4
+    SLURM_GPUS_PER_TASK: 1
+    MPICH_GPU_SUPPORT_ENABLED: 1
+    USE_MPI: "YES"
+    SLURM_MPI_TYPE: cray_shasta
+    CSCS_ADDITIONAL_MOUNTS: '["/opt/cray/pe/mpich/8.1.28/ofi/gnu/12.3/lib/libmpi.so:/usr/local/lib/libmpi.so.12.1.8", "/opt/cray/pe/lib64/libpmi.so.0:/usr/lib64/libpmi.so.0", "/opt/cray/pe/lib64/libpmi2.so.0:/usr/lib64/libpmi2.so.0",  "/opt/cray/pals/1.4/lib/libpals.so.0:/usr/lib64/libpals.so.0",  "/usr/lib64/libgfortran.so.5:/usr/lib64/libgfortran.so.5", "/opt/cray/pe/mpich/8.1.28/gtl/lib/libmpi_gtl_cuda.so:/usr/lib64/libmpi_gtl_cuda.so"]'

examples/CMakeLists.txt

@@ -36,17 +36,6 @@ if(GT_INSTALL_EXAMPLES)
 
     install_example(DIRECTORY boundaries SOURCES boundaries boundaries_provided)
 
-    configure_file(c_bindings/CMakeLists.txt.in
-        ${CMAKE_CURRENT_BINARY_DIR}${CMAKE_FILES_DIRECTORY}/c_bindings/CMakeLists.txt @ONLY)
-    install(FILES ${CMAKE_CURRENT_BINARY_DIR}${CMAKE_FILES_DIRECTORY}/c_bindings/CMakeLists.txt
-        DESTINATION ${GT_INSTALL_EXAMPLES_PATH}/c_bindings)
-    install(
-        DIRECTORY c_bindings
-        DESTINATION ${GT_INSTALL_EXAMPLES_PATH}
-        PATTERN "CMakeLists.txt.in" EXCLUDE
-        )
-    list(APPEND enabled_examples c_bindings)
-
     configure_file(cmake_skeletons/CMakeLists.txt.driver.in ${CMAKE_CURRENT_BINARY_DIR}${CMAKE_FILES_DIRECTORY}/CMakeLists.txt @ONLY)
     install(FILES ${CMAKE_CURRENT_BINARY_DIR}${CMAKE_FILES_DIRECTORY}/CMakeLists.txt DESTINATION ${GT_INSTALL_EXAMPLES_PATH})
 

pyutils/driver.py

@@ -1,5 +1,13 @@
 #!/usr/bin/env python3
 
+# /// script
+# dependencies = [
+#   "matplotlib",
+#   "numpy",
+#   "python-dateutil",
+# ]
+# ///
+
 import json
 import os
 

pyutils/pyutils/env.py

@@ -10,20 +10,36 @@
 env = os.environ.copy()
 
 
+def env_flag_to_bool(name: str, default: bool) -> bool:
+    """Recognize true or false signaling string values."""
+    flag_value = None
+    if name in env:
+        flag_value = env[name].lower()
+    if flag_value is None:
+        return default
+    elif flag_value in ("0", "false", "off"):
+        return False
+    elif flag_value in ("1", "true", "on"):
+        return True
+    else:
+        raise ValueError(
+            "Invalid environment flag value: use '0 | false | off' or '1 | true | on'."
+        )
+
+
 def load(envfile):
     if not os.path.exists(envfile):
         raise FileNotFoundError(f'Could find environment file "{envfile}"')
-    env['GTCMAKE_PYUTILS_ENVFILE'] = os.path.abspath(envfile)
+    env["GTCMAKE_PYUTILS_ENVFILE"] = os.path.abspath(envfile)
 
     envdir, envfile = os.path.split(envfile)
     output = runtools.run(
-        ['bash', '-c', f'set -e && source {envfile} && env -0'],
-        cwd=envdir).strip('\0')
-    env.update(line.split('=', 1) for line in output.split('\0'))
+        ["bash", "-c", f"set -e && source {envfile} && env -0"], cwd=envdir
+    ).strip("\0")
+    env.update(line.split("=", 1) for line in output.split("\0"))
 
-    log.info(f'Loaded environment from {os.path.join(envdir, envfile)}')
-    log.debug(f'New environment',
-              '\n'.join(f'{k}={v}' for k, v in sorted(env.items())))
+    log.info(f"Loaded environment from {os.path.join(envdir, envfile)}")
+    log.debug(f"New environment", "\n".join(f"{k}={v}" for k, v in sorted(env.items())))
 
 
 try:
@@ -36,39 +52,43 @@ def load(envfile):
 
 
 def _items_with_tag(tag):
-    return {k[len(tag):]: v for k, v in env.items() if k.startswith(tag)}
+    return {k[len(tag) :]: v for k, v in env.items() if k.startswith(tag)}
 
 
 def cmake_args():
     args = []
-    for k, v in _items_with_tag('GTCMAKE_').items():
-        if v.strip().upper() in ('ON', 'OFF'):
-            k += ':BOOL'
+    for k, v in _items_with_tag("GTCMAKE_").items():
+        if v.strip().upper() in ("ON", "OFF"):
+            k += ":BOOL"
         else:
-            k += ':STRING'
-        args.append(f'-D{k}={v}')
+            k += ":STRING"
+        args.append(f"-D{k}={v}")
     return args
 
 
 def set_cmake_arg(arg, value):
     if isinstance(value, bool):
-        value = 'ON' if value else 'OFF'
-    env['GTCMAKE_' + arg] = value
+        value = "ON" if value else "OFF"
+    env["GTCMAKE_" + arg] = value
 
 
 def sbatch_options(mpi):
-    options = _items_with_tag('GTRUN_SBATCH_')
+    options = _items_with_tag("GTRUN_SBATCH_")
     if mpi:
-        options.update(_items_with_tag('GTRUNMPI_SBATCH_'))
+        options.update(_items_with_tag("GTRUNMPI_SBATCH_"))
 
     return [
-        '--' + k.lower().replace('_', '-') + ('=' + v if v else '')
+        "--" + k.lower().replace("_", "-") + ("=" + v if v else "")
         for k, v in options.items()
     ]
 
 
 def build_command():
-    return env.get('GTRUN_BUILD_COMMAND', 'make').split()
+    return env.get("GTRUN_BUILD_COMMAND", "make").split()
+
+
+def run_with_slurm() -> bool:
+    return env_flag_to_bool("GTRUN_WITH_SLURM", True)
 
 
 def hostname():
@@ -90,9 +110,10 @@ def clustername():
         'kesch'
     """
     try:
-        output = runtools.run(['scontrol', 'show', 'config'])
-        m = re.compile(r'.*ClusterName\s*=\s*(\S*).*',
-                       re.MULTILINE | re.DOTALL).match(output)
+        output = runtools.run(["scontrol", "show", "config"])
+        m = re.compile(r".*ClusterName\s*=\s*(\S*).*", re.MULTILINE | re.DOTALL).match(
+            output
+        )
         if m:
             return m.group(1)
     except FileNotFoundError: