On the cache access behavior of openMP applications

Abstract

The widening gap between memory and processor speed results in increasing requirements to improve the cache utility. This issue is especially critical for OpenMP execution which usually explores fine-grained parallelism. The work presented in this paper studies the cache behavior of OpenMP applications in order to detect potential optimizations with respect to cache locality. This study is based on a simulation environment that models the parallel execution of OpenMP programs and provides comprehensive information about the runtime data accesses. This performance data enables a detailed analysis and an easy understanding of the cache operations performed on-line during the execution.