New Search

Export article


Aditya K. Kamath, Arkaprava Basu

Abstract: Newer use cases of GPU (Graphics Processing Unit) computing, e.g., graph analytics, look less like traditional bulk-synchronous GPU programs. To cater to the needs of emerging applications with semantically richer and finer grain sharing patterns, GPU vendors have been introducing advanced programming features, e.g., scoped synchronization and independent thread scheduling. While these features can speed up many applications and enable newer use cases, they can also introduce subtle synchronization errors if used incorrectly. We present iGUARD, a runtime software tool to detect races in GPU programs due to incorrect use of such advanced features. A key need for a race detector to be practical is to accurately detect races at reasonable overheads. We thus perform the race detection on the GPU itself without relying on the CPU. The GPU's parallelism helps speed up race detection by 15x over a closely related prior work. Importantly, iGUARD detects newer types of races that were hitherto not possible for any known tool. It detected previously unknown subtle bugs in popular GPU programs, including three in NVIDIA supported commercial libraries. In total, iGUARD detected 57 races in 21 GPU programs, without false positives.
Keywords: GPU program correctness / Data races / Debugging

Scifeed alert for new publications

Never miss any articles matching your research from any publisher
  • Get alerts for new papers matching your research
  • Find out the new papers from selected authors
  • Updated daily for 49'000+ journals and 6000+ publishers
  • Define your Scifeed now

Share this article

References (27)
    Back to Top Top