A tool for evaluating compiler-based parallelization strategies

This paper describes a knowledge-based tool being developed to study the selection of a sequence of transformations that if applied to a sequential program will result in an efficient parallel schedule. This work concentrates on parallelization of scientific programs. We expect that the experience gained from using such a tool will provide valuable insight into the interactions among the transformations. The results can then be incorporated into a restructuring compiler or an interactive tool. Various strategies are being developed to efficiently search for a fast parallel schedule.