A Loop Filtering Technique for Reducing Time Overhead of Dynamic Data Race Detection

Abstract

Data races are the hardest defect to handle in multithread programs due to their nondeterministic interleaving of concurrent threads. The main drawback of data race detection using dynamic techniques is the additional overhead of monitoring program execution and analyzing every conflicting memory operation. Thus, it is important to reduce the additional overheads for debugging data races. This paper presents a loop filtering technique that rules out repeatedly execution regions of loops from the monitoring targets in the multithread programs. The empirical results using multithread programs show that the filtering technique reduces the average runtime overhead to 60% of that of dynamic data race detection.