Skip to content

New function for domain decomposition (#4183) #304

New function for domain decomposition (#4183)

New function for domain decomposition (#4183) #304

Workflow file for this run

name: hip
on: [push, pull_request]
concurrency:
group: ${{ github.ref }}-${{ github.head_ref }}-hip
cancel-in-progress: true
jobs:
tests-hip:
name: HIP ROCm Flang C++17 [tests]
runs-on: ubuntu-20.04
steps:
- uses: actions/checkout@v4
- name: Dependencies
run: |
.github/workflows/dependencies/dependencies_hip.sh
.github/workflows/dependencies/dependencies_ccache.sh
- name: Set Up Cache
uses: actions/cache@v4
with:
path: ~/.cache/ccache
key: ccache-${{ github.workflow }}-${{ github.job }}-git-${{ github.sha }}
restore-keys: |
ccache-${{ github.workflow }}-${{ github.job }}-git-
- name: Build & Install
# Have to remove "-fno-operator-names to avoid
# /opt/rocm-6.1.0/include/rocprim/device/detail/device_adjacent_difference.hpp:198:26: error: token is not a valid binary operator in a preprocessor subexpression
# 198 | #if defined(__gfx1102__) or defined(__gfx1030__)
# | ~~~~~~~~~~~~~~~~~~~~~^~
env: {CXXFLAGS: "-Werror -Wall -Wextra -Wpedantic -Wnull-dereference -Wfloat-conversion -Wshadow -Woverloaded-virtual -Wextra-semi -Wunreachable-code -Wnon-virtual-dtor"}
run: |
export CCACHE_COMPRESS=1
export CCACHE_COMPRESSLEVEL=10
export CCACHE_MAXSIZE=150M
ccache -z
source /etc/profile.d/rocm.sh
hipcc --version
which clang
which clang++
which flang
# "mpic++ --showme" forgets open-pal in Ubuntu 20.04 + OpenMPI 4.0.3
# https://bugs.launchpad.net/ubuntu/+source/openmpi/+bug/1941786
# https://github.com/open-mpi/ompi/issues/9317
export LDFLAGS="-lopen-pal"
cmake -S . -B build \
-DCMAKE_VERBOSE_MAKEFILE=ON \
-DAMReX_EB=ON \
-DAMReX_ENABLE_TESTS=ON \
-DAMReX_FORTRAN=ON \
-DAMReX_LINEAR_SOLVERS=ON \
-DAMReX_GPU_BACKEND=HIP \
-DAMReX_AMD_ARCH=gfx1010 \
-DAMReX_ROCTX=ON \
-DCMAKE_C_COMPILER=$(which clang) \
-DCMAKE_CXX_COMPILER=$(which clang++) \
-DCMAKE_Fortran_COMPILER=$(which flang) \
-DCMAKE_CXX_STANDARD=17 \
-DCMAKE_CXX_COMPILER_LAUNCHER=ccache
cmake --build build -j 4
ccache -s
du -hs ~/.cache/ccache
tests-hip-wrapper:
name: HIP ROCm [email protected] C++17 [tests-hipcc]
runs-on: ubuntu-20.04
steps:
- uses: actions/checkout@v4
- name: Dependencies
run: |
.github/workflows/dependencies/dependencies_hip.sh
.github/workflows/dependencies/dependencies_ccache.sh
- name: Set Up Cache
uses: actions/cache@v4
with:
path: ~/.cache/ccache
key: ccache-${{ github.workflow }}-${{ github.job }}-git-${{ github.sha }}
restore-keys: |
ccache-${{ github.workflow }}-${{ github.job }}-git-
- name: Build & Install
# Have to remove "-fno-operator-names to avoid
# /opt/rocm-6.1.0/include/rocprim/device/detail/device_adjacent_difference.hpp:198:26: error: token is not a valid binary operator in a preprocessor subexpression
# 198 | #if defined(__gfx1102__) or defined(__gfx1030__)
# | ~~~~~~~~~~~~~~~~~~~~~^~
env: {CXXFLAGS: "-Werror -Wall -Wextra -Wpedantic -Wnull-dereference -Wfloat-conversion -Wshadow -Woverloaded-virtual -Wextra-semi -Wunreachable-code -Wnon-virtual-dtor"}
run: |
export CCACHE_COMPRESS=1
export CCACHE_COMPRESSLEVEL=10
export CCACHE_MAXSIZE=100M
ccache -z
source /etc/profile.d/rocm.sh
hipcc --version
# "mpic++ --showme" forgets open-pal in Ubuntu 20.04 + OpenMPI 4.0.3
# https://bugs.launchpad.net/ubuntu/+source/openmpi/+bug/1941786
# https://github.com/open-mpi/ompi/issues/9317
export LDFLAGS="-lopen-pal"
cmake -S . -B build_full_legacywrapper \
-DCMAKE_VERBOSE_MAKEFILE=ON \
-DAMReX_EB=OFF \
-DAMReX_ENABLE_TESTS=ON \
-DAMReX_FORTRAN=ON \
-DAMReX_GPU_RDC=OFF \
-DAMReX_LINEAR_SOLVERS=ON \
-DAMReX_GPU_BACKEND=HIP \
-DAMReX_AMD_ARCH=gfx908 \
-DAMReX_ROCTX=ON \
-DCMAKE_C_COMPILER=$(which clang) \
-DCMAKE_CXX_COMPILER=$(which hipcc) \
-DCMAKE_Fortran_COMPILER=$(which gfortran) \
-DCMAKE_CXX_STANDARD=17 \
-DCMAKE_CXX_COMPILER_LAUNCHER=ccache
cmake --build build_full_legacywrapper -j 4
ccache -s
du -hs ~/.cache/ccache
# Build 2D libamrex hip build with configure
configure-2d-single-hip:
name: HIP EB [configure 2D]
runs-on: ubuntu-20.04
steps:
- uses: actions/checkout@v4
- name: Dependencies
run: |
.github/workflows/dependencies/dependencies_hip.sh
.github/workflows/dependencies/dependencies_ccache.sh
- name: Set Up Cache
uses: actions/cache@v4
with:
path: ~/.cache/ccache
key: ccache-${{ github.workflow }}-${{ github.job }}-git-${{ github.sha }}
restore-keys: |
ccache-${{ github.workflow }}-${{ github.job }}-git-
- name: Build & Install
run: |
export CCACHE_COMPRESS=1
export CCACHE_COMPRESSLEVEL=10
export CCACHE_MAXSIZE=100M
ccache -z
./configure --dim 2 --with-hip yes --enable-eb yes --enable-xsdk-defaults yes --with-mpi no --with-omp no --single-precision yes --single-precision-particles yes
make -j4 WARN_ALL=TRUE AMD_ARCH=gfx90a CCACHE=ccache
make install
ccache -s
du -hs ~/.cache/ccache
# Build 3D EB hip with gnu make
hip-3d-eb-gmake:
name: HIP EB 3D GMake
runs-on: ubuntu-20.04
steps:
- uses: actions/checkout@v4
- name: Dependencies
run: |
.github/workflows/dependencies/dependencies_hip.sh
.github/workflows/dependencies/dependencies_ccache.sh
- name: Set Up Cache
uses: actions/cache@v4
with:
path: ~/.cache/ccache
key: ccache-${{ github.workflow }}-${{ github.job }}-git-${{ github.sha }}
restore-keys: |
ccache-${{ github.workflow }}-${{ github.job }}-git-
- name: Build & Install
run: |
export CCACHE_COMPRESS=1
export CCACHE_COMPRESSLEVEL=10
export CCACHE_MAXSIZE=120M
ccache -z
cd Tests/LinearSolvers/NodeEB
make -j4 USE_HIP=TRUE USE_MPI=FALSE BL_NO_FORT=TRUE WARN_ALL=TRUE WARN_ERROR=TRUE AMD_ARCH=gfx90a CCACHE=ccache
ccache -s
du -hs ~/.cache/ccache