LLOV: A Fast Static Data-Race Checker for OpenMP Programs

In the era of Exascale computing, writing efficient parallel programs is indispensable and at the same time, writing sound parallel programs is highly difficult. While parallel programming is easier with frameworks such as OpenMP, the possibility of data races in these programs still persists. In this paper, we propose a fast, lightweight, language agnostic, and static data race checker for OpenMP programs based on the LLVM compiler framework. We compare our tool with other state-of-the-art data race checkers on a variety of well-established benchmarks. We show that the precision, accuracy, and the F1 score of our tool is comparable to other checkers while being orders of magnitude faster. To the best of our knowledge, this work is the only tool among the state-of-the-art data race checkers that can verify a FORTRAN program to be data race free