Pipeline Optimization in an HPF Compiler

Summary Generally, in scientific applications, parallelism is extracted from loops because they spend the most of execution time. Iterations of such loops are executed in parallel to achieve speedup over the sequential program. However, a number of these scientific applications exhibit recurrences that give rise to data dependencies across processors or nodes. These dependencies tend to slow down parallel execution and sometimes even serialize the loop. This paper proposes pipeline technology to resolve such problems, which breaks the computation into blocks. Each processor performs the computation of a block, which enables the next processor in the pipeline to compute its corresponding block. Once the pipeline is filled, the computation of blocks on different processors proceeds in parallel. We describe the design and implementation of the pipelining in an HPF compiler and show that the computation achieves better parallel performance using our method.