Abstract: Data race probably occurs when many threads concurrently access the same memory location and at least one is a write thread. Data race detection suffers from false negatives and false positives. How to detect data race and avoid false negatives and false positives has become a hot topic. This paper proposes a static data race detection methodology to eliminate false negatives and false positives. We use Soot to conduct intra-thread and inter-thread analysis. Our data race detection focuses on variable access events that are collected from call graphs. Several program analysis technologies, such as alias variable analysis, alias lock analysis, happens-before analysis, constraint graph, and slicing analysis, are used to improve the coverage and precision of the detection results. In the experimentation, several benchmarks, such as raytracer, sor, and mergesort, have been selected to evaluate our methodology. Experimental results show that SRD can not only eliminate fake races but also identify potential races. Furthermore, SRD can detect more positive races than the existing tool RVPredict.

Key words: data race, intra/inter-thread analysis, alias analysis, happens-before, concurrent program