In this paper, we propose a loop parallelization technique and iteration allocation technique using cosets. This technique targets loops that contain linear dependence vectors, rather than only constant dependence vectors. Previous work in loop parallelization with linear dependence vectors limits parallelism and ignores the effects of interprocessor communication. The technique we propose is based on cosets. It extracts more parallelism and considers interprocessor communication. Loops of n iterations are parallelized with O(log(n)) steps. Iterations are allocated to processors to reduce interprocessor communication cost.
|Number of pages||5|
|Publication status||Published - 1995|
All Science Journal Classification (ASJC) codes