MPI-CorrBench [CORRBE21] enables a structured comparison of the different available tools for MPI correctness checking w.r.t. various types of errors.
MPI-CorrBench is automatically executed by our CI pipeline. You can find the automatically generated visualizations in our Visualization Dashboard. for the MPI OpenMP testcases, the CI only executes one configuration for a better assesment it may be a good idea to execute it multiple times variing the amount of threads. For more detailes refer [HYBE22]
Provide the Following environment Variables:
MPI_CORRECTNESS_BM_DIR
full path to the local directory of this repoMPI_CORRECTNESS_BM_EXPERIMENT_DIR
full path to the directory where the tools should be executed
- Adapt the SLURM configuration in
scripts/SLURM_header.in
- for each tool a script file
scripts/<TOOL>/execute_tool.sh
exists. You may need to adopt it to use the tool on your system - run
scripts/run_measurement.sh 1
to generate the jobscript and submit them via sbatch (the argument given is the number of repetition for each test case) - after all jobs are finished use
scripts/gather_results.sh
to gather all the results, which will override previous data inoutput
(such as our results)
The visualizations presented in [CORRBE21] can be generated by the usage of the provided scripts. The scripts don't take any arguments and read the data in output
to generate the visualization.
[CORRBE21] | Lehr, Jan-Patrick and Jammer, Tim and Bischof, Christian: MPI-CorrBench: Towards an MPI Correctness Benchmark Suite 2021. https://dl.acm.org/doi/abs/10.1145/3431379.3460652 |
[HYBE22] | Jammer, Tim and Hück, Alexander and Lehr, Jan-Patrick and Protze, Joachim and Schwitanski, Simon and Bischof, Christian: Towards a Hybrid MPI Correctness Benchmark Suite 2022. https://dl.acm.org/doi/10.1145/3555819.3555853 |