Abstract
Efficient methods of partitioning nested for-loops for parallel execution on multicomputers are presented. The authors seek to identify appropriate partition schemes systematically and automatically without users specifying data partition schemes explicitly. The grouping method, which takes advantage of the regularity of nested for-loops, is very efficient and uses only simple algebraic manipulations of loop dependence vectors. Grouping is inherent in techniques for synthesizing systolic arrays and is augmented with strategies for merging computations to perform loop partition. The results point out a new direction for developing highly automatic parallelizing compilers.