POSTER: MAPA: An Automatic Memory Access Pattern Analyzer for GPU Applications

摘要: Various existing optimization and memory consistency management techniques for GPU applications rely on access patterns of kernels. However, they suffer from poor practicality because require explicit user interventions to extract kernel patterns. This paper proposes an automatic memory-access-pattern analysis framework called MAPA. MAPA is based a source-level technique derived traditional symbolic analyses run-time pattern selection technique. The experimental results show that properly analyzes 116 real-world OpenCL kernels Rodinia Parboil.