Detecting Correlation Violations and Data Races by Inferring Non-deterministic Reads

摘要: With the introduction of multicore systems and parallel programs concurrency bugs have become more common. A notorious class these are data races that violate correlations between variables. This happens, for example, when programmer does not update correlated variables atomically, which is needed to maintain their semantic relationship. The detection such challenging because among usually escape traditional race detectors oblivious relationships. In this paper, we present an effective method dynamically identifying together with a detector based on notion non-deterministic reads identifies malicious eight 190 micro benchmarks, found than 100 were overlooked by other detectors. Furthermore, identified about 300 variable violated races.