Georgia Institute of Technology
As highly-parallel accelerators such as graphics processing units become more important in high-performance computing, so does the need to ensure their reliable operation. In response, research has been directed at several efforts to characterize and understand the hardware vulnerability of GPU micro-architecture structures, as well as to detecting and correcting such vulnerabilities. In this position paper, the authors advocate a transparent, customizable, and dynamic software approach to reduce the vulnerability of GPU applications. Specifically, they propose to realize these enhancements via transparent code patching of GPU applications to increase protection of critical program regions.