Dynamic instrumentation for determining logical concurrency between threads of OpenMP Programs

Abstract

OpenMP provides a portable way to achieve high performance and simple compiler directives to transform a sequential program into parallel program. It is important to determine the logical concurrency of OpenMP programs, because it helps detecting data races among concurrent threads in an execution of the program. The previous work, RaceStand, has been developed to detect on-the-fly data races for OpenMP applications. Unfortunately, RaceStand does not detect data races or report false positives since it creates incorrect logical thread identifiers for OpenMP programs with nested parallelism which uses function calls for the execution of parallel regions due to its defective source code instrumentation. This paper presents a dynamic binary instrumentation technique to determine logical concurrency of parallel threads based on Pin software framework. We implemented a Pin-tool as data race detector including our instrumentation technique and empirically compared the correctness of the Pin-tool with RaceStand using a set of OpenMP synthetic programs and real-world applications. The results show that our technique precisely determines logical concurrency among the execution of concurrent threads in nested fork-join parallel programs. ? 2014 International Information Institute.