Parallel implementations of Shallow Water Equations based on the teaching code(https://github.com/TUM-I5/SWE).
The documentation is available in the Wiki
- CMake >= 3.10
- NetCDF
- OpenMP
- HPX >= 1.4 (https://github.com/STEllAR-GROUP/hpx)
- Charm++ (https://github.com/UIUC-PPL/charm)
- UPC++ >= 2020.3.0 (https://bitbucket.org/berkeleylab/upcxx)
- ASAGI (only required for usage with geographic scenarios) (https://github.com/TUM-I5/ASAGI)
- Chameleon (https://github.com/chameleon-hpc/chameleon)
- Clone the repository with following command:
git clone --recursive
- Checkout the sc_paw-atm_workshop :
git checkout tags/sc_paw-atm_workshop -b master
- Create build directory
mkdir build && cd build
- Set necessary environment variables
UPC++:
export UPCXX_PATH=\path\to\upcxx
Chameleon:
export CHAMELEON_PATH=\path\to\chameleon
HPX:
export HPX_PATH=\path\to\hpx
\ - Specify cmake targets and compile options
cmake -DBUILD_SWE_MPIOVERDECOMP=On\ -DBUILD_SWE_MPIOVERDECOMPTASKING=On\ \-DBUILD_SWE_MPI=On \-DBUILD_SWE_UPCXX=On \-DBUILD_SWE_CHAMELEON=On \-DBUILD_SWE_HPX=On ..
For a complete list of configurable options call
cmake -LAH ..
- Compile targets
make
orcmake --build . --target swe_benchmark_<target>
Unfortunately Charm++ still requires the old scons build system.
- Set environment variable
export CHARM_PATH=\path\to\charm
- Compile using scons
scons writeNetCDF=True compiler=intel solver=hybrid openmp=false parallelization=charm asagi=false copyenv=true vectorize=true
Note, that the below provided examples may vary depending on the system architecture and configuration of the frameworks. The examples execute the compiled scenario with a 2048x2048 cell resolution,80 seconds simulation duration, 20 checkpoints, global time stepping and file output enabled.
- MPI:
mpirun -np 56 ./build/swe_benchmark_mpi --simulation-duration 80 --checkpoint-count 20 --resolution-horizontal 2048 --resolution-vertical 2048 --output-basepath ./output/mpi_gts --local-timestepping 0 --write 1
- UPC++:
$UPCXX_PATH/bin/upcxx-run -np 56 ./build/swe_benchmark_upcxx --simulation-duration 80 --checkpoint-count 20 --resolution-horizontal 2048 --resolution-vertical 2048 --output-basepath ./output/upcxx_gts --local-timestepping 0 --write 1
- Charm++:
$CHARM_PATH/bin/charmrun +p56 ./build/SWE_intel_release_charm_hybrid_vec --simulation-duration 80 --checkpoint-count 20 --resolution-horizontal 2048 --resolution-vertical 2048 --output-basepath ./output/charm_gts --local-timestepping 0 --write 1 --chares 56
- Chameleon:
I_MPI_PIN=1 I_MPI_PIN_DOMAIN=auto OMP_NUM_THREADS=27 OMP_PLACES=cores OMP_PROC_BIND=close mpirun -np 56 ./build/swe_benchmark_upcxx --simulation-duration 80 --checkpoint-count 20 --resolution-horizontal 2048 --resolution-vertical 2048 --output-basepath ./output/chameleon_gts --local-timestepping 0 --write 1 --blocks 28
- HPX:
mpirun -np 56 ./build/swe_benchmark_hpx --simulation-duration 80 --checkpoint-count 20 --resolution-horizontal 2048 --resolution-vertical 2048 --output-basepath ./output/hpx_gts --local-timestepping 0 --write 1 --blocks 28
More examples of execution scripts can be found at https://github.com/TUM-I5/SWE-PPM/tree/master/jobs
- Michael Bader
- Martin Bogusz
- Philipp Samfaß
- Alexander Pöppl
SWE is release unter GPLv3 (see gpl.txt)