Enabling semantics to improve detection of data races and misuses of lock-free data structures

摘要

The rapid progress of multi/many-core architectures has caused data-intensive parallel applicationsnot yet fully optimized to deliver thebest performance. Intheadvent of concurrent programming,frameworks offering structured patterns have alleviated developers' burden adapting suchapplications to multithreaded architectures. Whilesomeof these patterns are implemented usingsynchronization primitives, others avoid them by means of lock-free data mechanisms. However,lock-free programming is not straightforward, ensuring an appropriate use of their interfaces canbe challenging, since different memory models plus instruction reordering at compiler/processorlevels can interfere in the occurrence of data races. The benefits of race detectors are formidablein this sense; however, they may emit false positives if are unaware of the underlying lock-freestructure semantics. To mitigate this issue, this paper extends ThreadSanitizer, a race detectiontool, with the semantics of 2 lock-free data structures: the single-producer/single-consumer andthemultiple-producer/multiple-consumer queues. With it, we are able to drop false positives anddetect potential semantic violations. The experimental evaluation, using different queue implementationson a set of 𜇠benchmarks and real applications, demonstrates that it is possible toreduce, on average, 60% the number of data race warnings and detect wrong uses of thesestructures.