Skip to content

build-hpc: specify OMP_STACKSIZE for GPU build #1161

build-hpc: specify OMP_STACKSIZE for GPU build

build-hpc: specify OMP_STACKSIZE for GPU build #1161

Workflow file for this run

name: build
# Controls when the action will run
on:
# Trigger the workflow on all pushes, except on tag creation
push:
branches:
- '**'
tags-ignore:
- '**'
# Trigger the workflow on all pull requests
pull_request: ~
# Allow workflow to be dispatched on demand
workflow_dispatch: ~
env:
ECWAM_TOOLS: ${{ github.workspace }}/.github/tools
CTEST_PARALLEL_LEVEL: 1
CACHE_SUFFIX: v1 # Increase to force new cache to be created
jobs:
ci:
name: ci
strategy:
fail-fast: false # false: try to complete all jobs
matrix:
build_type: [Bit, Debug]
prec: ['DP', 'SP']
name:
- linux gnu-14
- linux nvhpc-25.9
- linux intel-classic
- linux intel-modern
- macos
include:
- name: linux gnu-14
os: ubuntu-24.04
compiler: gnu-14
compiler_cc: gcc-14
compiler_cxx: g++-14
compiler_fc: gfortran-14
python-version: '3.11'
caching: true
- name: linux nvhpc-25.9
os: ubuntu-22.04
compiler: nvhpc-25.9
compiler_cc: nvc
compiler_cxx: nvc++
compiler_fc: nvfortran
cmake_options: -DCMAKE_CXX_FLAGS=--diag_suppress177 -DECBUILD_Fortran_FLAGS=-tp=haswell
# nvc++ 25.9 has a codegen bug where an AVX-512 SCALEFS node is
# emitted by an idiom-recognition pass even on non-AVX-512 targets,
# causing llc to ICE on some GH runner CPUs. Pin -tp=haswell to
# constrain the target features so the pass cannot emit it.
dep_eccodes_extra_options: -DCMAKE_CXX_FLAGS=-tp=haswell
python-version: '3.8'
caching: true
- name : linux intel-classic
os: ubuntu-22.04
compiler: intel-classic
compiler_cc: icc
compiler_cxx: icpc
compiler_fc: ifort
python-version: '3.8'
caching: true
- name: linux intel-modern
os: ubuntu-24.04
compiler: intel-modern
compiler_cc: icx
compiler_cxx: icpx
compiler_fc: ifx
python-version: '3.11'
caching: true
- name: macos
# Xcode compiler requires empty environment variables, so we pass null (~) here
os: macos-15
compiler: clang-17
compiler_cc: ~
compiler_cxx: ~
compiler_fc: gfortran-15
python-version: '3.11'
caching: true
runs-on: ${{ matrix.os }}
steps:
- name: Set up Python ${{ matrix.python-version }}
uses: actions/setup-python@v5
with:
python-version: ${{ matrix.python-version }}
- name: Checkout Repository
uses: actions/checkout@v2
- name: Environment
run: |
echo "DEPS_DIR=${{ runner.temp }}/deps" >> $GITHUB_ENV
echo "CC=${{ matrix.compiler_cc }}" >> $GITHUB_ENV
echo "CXX=${{ matrix.compiler_cxx }}" >> $GITHUB_ENV
echo "FC=${{ matrix.compiler_fc }}" >> $GITHUB_ENV
if [[ "${{ matrix.os }}" =~ macos ]]; then
export HOMEBREW_NO_INSTALLED_DEPENDENTS_CHECK=1
export HOMEBREW_NO_AUTO_UPDATE=1
export HOMEBREW_NO_INSTALL_CLEANUP=1
export SDKROOT=$(xcrun --show-sdk-path)
export ECWAM_LAUNCH_ARGS=--oversubscribe
echo "HOMEBREW_NO_INSTALLED_DEPENDENTS_CHECK=1" >> $GITHUB_ENV
echo "HOMEBREW_NO_AUTO_UPDATE=1" >> $GITHUB_ENV
echo "HOMEBREW_NO_INSTALL_CLEANUP=1" >> $GITHUB_ENV
echo "SDKROOT=$(xcrun --show-sdk-path)" >> $GITHUB_ENV
echo "ECWAM_LAUNCH_ARGS=--oversubscribe" >> $GITHUB_ENV
brew install ninja
brew install libomp
brew install libaec
brew install coreutils
else
sudo apt-get update
sudo apt-get install libaec-dev
sudo apt-get install ninja-build
fi
pip3 install numpy
printenv
- name: Cache Dependencies
# There seems to be a problem with cached NVHPC dependencies, leading to SIGILL perhaps due to slightly different architectures
if: matrix.caching
id: deps-cache
uses: pat-s/always-upload-cache@v2.1.5
with:
path: ${{ env.DEPS_DIR }}
key: deps-${{ matrix.os }}-${{ matrix.compiler }}-${{ matrix.build_type }}-${{ env.CACHE_SUFFIX }}
# Free up disk space for nvhpc
- name: Free Disk Space (Ubuntu)
uses: jlumbroso/free-disk-space@main
if: contains( matrix.compiler, 'nvhpc' )
continue-on-error: true
with:
# this might remove tools that are actually needed,
# if set to "true" but frees about 6 GB
tool-cache: false
# all of these default to true, but feel free to set to
# "false" if necessary for your workflow
android: true
dotnet: true
haskell: true
large-packages: true
docker-images: true
swap-storage: true
- name: Install NVHPC compiler
if: contains( matrix.compiler, 'nvhpc' )
shell: bash -eux {0}
run: |
${ECWAM_TOOLS}/install-nvhpc.sh --prefix /opt/nvhpc --version 25.9
source /opt/nvhpc/env.sh
echo "${NVHPC_DIR}/compilers/bin" >> $GITHUB_PATH
[ -z ${MPI_HOME+x} ] || echo "MPI_HOME=${MPI_HOME}" >> $GITHUB_ENV
echo "localhost slots=72" >> ${MPI_HOME}/etc/openmpi-default-hostfile
echo "ECWAM_LAUNCH_ARGS=--oversubscribe" >> $GITHUB_ENV
echo "ECWAM_LAUNCH_SERIAL_MPI=1" >> $GITHUB_ENV
- name: Install Intel oneAPI compiler
if: contains( matrix.compiler, 'intel' )
run: |
${ECWAM_TOOLS}/install-intel-oneapi.sh --compiler ${{ matrix.compiler }}
source /opt/intel/oneapi/setvars.sh
printenv >> $GITHUB_ENV
echo "CACHE_SUFFIX=$CC-$($CC -dumpversion)" >> $GITHUB_ENV
- name: Install MPI
shell: bash -eux {0}
run: |
FCFLAGS=-fPIC CFLAGS=-fPIC FFLAGS=-fPIC ${ECWAM_TOOLS}/install-mpi.sh --mpi openmpi --prefix ${DEPS_DIR}/openmpi
[ -f ${DEPS_DIR}/openmpi/env.sh ] && source ${DEPS_DIR}/openmpi/env.sh
[ -z ${MPI_HOME+x} ] || echo "MPI_HOME=${MPI_HOME}" >> $GITHUB_ENV
- name: Set Build & Test Environment
run: |
# Add mpirun to path for testing
[ -z ${MPI_HOME+x} ] || echo "${MPI_HOME}/bin" >> $GITHUB_PATH
# Force UCX (and thus OpenMPI's UCX PML) onto TCP + shared-memory transports.
# The GitHub-hosted Azure runners intermittently expose a MANA (mana_0) RDMA/
# InfiniBand device on which UCX fails to create UD queue-pairs, causing
# MPI_Init to abort ("failed to create UD QP ... Operation not supported").
# These are single-node tests that never need RDMA, so restrict transports to
# loopback TCP + shared memory to make MPI startup deterministic.
echo "UCX_TLS=tcp,self,sm" >> $GITHUB_ENV
echo "UCX_NET_DEVICES=lo" >> $GITHUB_ENV
echo "UCX_LOG_LEVEL=error" >> $GITHUB_ENV
- name: Build & Test
id: build-test
uses: ecmwf-actions/build-package@v2
with:
self_coverage: false
force_build: true
cache_suffix: "${{ matrix.build_type }}-${{ env.CACHE_SUFFIX }}"
recreate_cache: ${{ matrix.caching == false }}
dependencies: |
ecmwf/ecbuild@refs/tags/3.14.2
ecmwf/eccodes@refs/tags/2.44.0
ecmwf/fckit@refs/tags/0.13.0
ecmwf-ifs/fiat
ecmwf-ifs/field_api@refs/tags/v0.3.4
ecmwf-ifs/loki@refs/heads/main
dependency_branch: develop
dependency_cmake_options: |
ecmwf/fckit: "-G Ninja -DCMAKE_BUILD_TYPE=${{ matrix.build_type }} -DENABLE_TESTS=OFF -DENABLE_FCKIT_VENV=ON"
ecmwf-ifs/fiat: "-G Ninja -DCMAKE_BUILD_TYPE=${{ matrix.build_type }} -DENABLE_TESTS=OFF -DENABLE_SINGLE_PRECISION=${{ matrix.prec == 'SP' }} -DENABLE_DOUBLE_PRECISION=${{ matrix.prec == 'DP' }} -DENABLE_DR_HOOK_NVTX=OFF"
ecmwf/eccodes: "-G Ninja -DCMAKE_BUILD_TYPE=${{ matrix.build_type }} -DENABLE_MEMFS=ON -DENABLE_JPG=OFF -DENABLE_PNG=OFF ${{ matrix.dep_eccodes_extra_options }}"
ecmwf-ifs/field_api: "-G Ninja -DCMAKE_BUILD_TYPE=${{ matrix.build_type }} -DENABLE_TESTS=OFF -DENABLE_ACC=OFF -DENABLE_SINGLE_PRECISION=${{ matrix.prec == 'SP' }} -DENABLE_DOUBLE_PRECISION=${{ matrix.prec == 'DP' }}"
ecmwf-ifs/loki: "-G Ninja -DCMAKE_BUILD_TYPE=${{ matrix.build_type }} -DENABLE_TESTS=OFF"
cmake_options: "-G Ninja -DCMAKE_BUILD_TYPE=${{ matrix.build_type }} ${{ matrix.cmake_options }} -DENABLE_MPI=ON -DENABLE_LOKI=ON -DLOKI_MODE=idem-stack -DENABLE_SINGLE_PRECISION=${{ matrix.prec == 'SP' }} -DENABLE_DOUBLE_PRECISION=${{ matrix.prec == 'DP' }}"
ctest_options: "${{ matrix.ctest_options }} ${{ matrix.build_type == 'Debug' && '-R etopo1_oper_an_fc_O48' || '' }}"
- name: Verify tools
run: |
export PATH=${{ steps.build-test.outputs.bin_path }}:$PATH
echo "+ ecwam --info"
ecwam --info
# - name: Codecov Upload
# if: steps.build-test.outputs.coverage_file
# uses: codecov/codecov-action@v2
# with:
# files: ${{ steps.build-test.outputs.coverage_file }}