Design and Evaluation of an Environment APE for Automatic Parallelization of Programs

In this paper, we have presented the design and evalu-ation of a compiler system, called APE, f o r automatic parallelization of scientific and engineering applications o n distributed m e m o r y computers. APE i s built o n top of S U I F compiler. It extends S U I F with capabilities in parallelizing loops with non-uniform cross-iteration dependencies, and in handling loops that have indirect access patterns. W e have evaluated the effectiveness of S U I F with several CFD test codes, and found that S U I F handles uni form loops over dense and regular data structures very well. For non-uniform loops, an innova-tive and efficient parallelization approach based o n con-vex theory have been proposed and i s being implemented. W e have also presented a class of scalable algorithms f o r parallel distribution and redistribution of unstructured data structures during parallelizing irregular loops.