Extending OpenMP to Facilitate Loop Optimization

Abstract

OpenMP provides several mechanisms to specify parallel source-code transformations. Unfortunately, many compilers perform these transformations early in the translation process, often before performing traditional sequential optimizations, which can limit the effectiveness of those optimizations. Further, OpenMP semantics preclude performing those transformations in some cases prior to the parallel transformations, which can limit overall application performance.

Access paper here