Investigating the Real-World Applicability of MPI Correctness Benchmarks.

The MPI correctness benchmarks MPI-Corrbench and the MPI Bugs Initiative contain standardized test cases of correct and erroneous use of MPI, allowing MPI correctness tool developers to assess their tools’ performance and guide further development of their checking capabilities. Hence, the correctness benchmarks should encompass representative MPI (mis-)usage that mirrors real-world codes. To that end, we analyze the MPI usage of these correctness benchmarks at argument granularity and compare it to a previously collected data set of 96 HPC codes. This assessment measures the benchmarks’ proximity to real-world MPI usage patterns and offers insights for enhancing their coverage.