Continuous Benchmarking of Numerical Algorithms Implemented in M++ via Gitlab CI/CD and Google Benchmark∗

Abstract

We present an automated framework for benchmarking numerical algorithms that solve partial differential equations under consistent and reproducible conditions using the parallel finite element software M++. This framework integrates GitLab CI/CD, Google Benchmark, and the HoreKa supercomputing system to enable continuous integration and benchmarking. By incorporating ongoing software development, the framework supports improving performance and reliability, which are vital for various scientific computing applications, including wave propagation, cardiovascular simulations, dislocation dynamics, and uncertainty quantification. These applications motivate the two benchmarking examples presented in this text. We further outline the benchmarking workflow as well as the use of a research database storing comprehensive performance data, facilitating reproducibility for future studies.