-
Notifications
You must be signed in to change notification settings - Fork 0
Meeting on ESPResSo test for test suite (2023‐07‐14)
Caspar van Leeuwen edited this page Aug 3, 2023
·
1 revision
In Espresso github:
-
maintainer/benchmarks has simple cases. Best way to make sure that Espresso is working fine. Run with
make -j<corecount> check_benchmarks
- TO run independenly
mpiexec -n 16 ./pypresso ../maintainer/benchmarks/lj.py
- p3m, lb and lj scale to many cores (4000 or so).
- Do they have GPU support? => They don't at the moment
- Jean-Noel will open a PR to add GPU support, it is not there now => He'll push it to the development branch
- with --help you get options. You can reduce number of particles per core
--particles-per-core
-
n_iterations
(in the benchmark file) can also be reduced to reduce the runtime -
-DWITH_CUDA=OFF
to make a CPU only build for Espresso -
cmake . -DESPRESSO_CTEST_ARGS="-j$(nproc)"
to allow ctest to use all cores (when calling from CMake's Makefiles) - It's pure MPI, so if you want to run on 4000 cores, you'll have to launch 4000 ranks.
- TO run independenly
-
testsuite: 3 min, tests individual features of Espresso (run with
make check_python
andcheck_unit_tests
).check_python
covers everything, using the python interface. End users use this anyway.check_unit_tests
uses CPP unit tests. They often bring issues to light quickly, but they don't cover the full codebase. -
samples has more complex cases
-
prepare an easyconfig without CUDA support.( toolchain:
foss
)- Make Mesa / IPython optional dependency for Notebooks. (Add comments to the easyconfig)