Differences in algorithmic parallelism in control flow and call multigraphs

Abstract

Our parallel hybrid analysis methods facilitate the parallelization of the analysis phase of a software transformation system, by enabling deeper semantic analyses to be accomplished more efficiently than if performed sequentially. Our previous empirical studies profiled these hybrid techniques on the Reaching Definitions problem [LMR91, LR92a, LR92b]. Recently, we have applied our method to the Interprocedural May _Alias Problem for Fortran programs in a prototype implementation. The interpretation of our results suggested further performance studies, comparing our region partition algorithms (both Bottom_Up and Forward partitioning [LRF94]) on call multigraphs and control flow graphs. These comparisons yielded statistically significant differences in performance of our graph partitioning techniques applied to these two graph populations. This research is part of a larger effort to calculate the modification side effects problem (MOD) for Fortran programs using our parallel hybrid techniques.

The work presented here was supported, in part, by ARPA contract DABT-63-93-C-0064 and by the National Science Foundation grants CCR92-13518 and CCR90-23628. The content of the information herein does not necessarily reflect the position of the Government and official endorsement should not be inferred.