Maximizing parallelism for non-uniform dependence Loops using two parallel region partitioning method

Abstract

The existing parallelizing compilers can parallelize most of the loops with uniform dependences, but they do not satisfactorily handle loops with non-uniform dependences. Most of the time, the compiler leaves such loops running sequentially. Unfortunately, loops with non-uniform dependences are not so uncommon in the real world. This paper presents the two parallel region partitioning method of nested loops with non-uniform dependences for maximizing parallelism. By parallelizing anti dependence region using variable renaming, we will divide the iteration space into two parallel regions by a line in case that FDT (Flow Dependence Tail set) does not overlap FDH (Flow Dependence Head set). Comparison with some related works shows more parallelism than other existing methods. © Springer-Verlag Berlin Heidelberg 2005.