Pattern-Based Refactoring Process of Sequential Source Code

Software refactoring is a very well-studied subject, but with the huge gap between omnipresent multicore processors and the vast majority of software that has not been developed with multicore in mind, it gains new and important significance: At what locations should the sequential code be refactored? How should this be done for a given location? Is the code still correct? And finally: Does it execute faster? In this paper we present a refactoring concept to exploit parallelizable regions in legacy software. Our concept relies on the presence of recurring patterns and identifies potential regions, transforms them to parallel versions, tests them for correctness and tunes their parallel performance. We show early implementation results.