FAROS: A Framework to Analyze OpenMP Compilation Through Benchmarking and Compiler Optimization Analysis

Compilers optimize OpenMP programs differently than their serial elision. Early outlining of parallel regions and invocation of parallel code via OpenMP runtime functions are two of the most profound differences. Understanding the interplay between compiler optimizations, OpenMP compilation, and application performance is hard and usually requires specialized benchmarks and compilation analysis tools. To this end, we present FAROS, an extensible framework to automate and structure the analysis of compiler optimization of OpenMP programs. FAROS provides a generic configuration interface to profile and analyze OpenMP applications with their native build configurations. Using FAROS on a set of 39 OpenMP programs, including HPC applications and kernels, we show that OpenMP compilation hinders optimization for the majority of programs. Comparing single-threaded OpenMP execution to its sequential counterpart, we observed slowdowns as much as 135.23%. In some cases, however, OpenMP compilation speeds up execution as much as 25.48% when OpenMP semantics help compiler optimization. Following analysis on compiler optimization reports enables us to pinpoint the reasons without in-depth knowledge of the compiler. The information can be used to improve compilers and also to bring performance on par through manual code refactoring.