AbstractIn this paper, we propose a systematic method to synthesize the alignment and realignment directives for data-parallel languages. By using several heuristic rules, this method can determine the multiple alignment relation of the given program. This method can be carried out in polynomial time. An NAS parallel benchmark program was selected as a benchmark and compiled for execution on a Transputer system. The result shows that the execution time of the multiple alignment version of the given program obtained by the proposed method is smaller than the single alignment version of the given program.
Categories and Subject Descriptors: D.3.4 [Programming Languages]: Processors; I.2.2 [Artificial Intelligence]: Automatic Programming; D.3.2 [Programming Languages]: Language Classifications
Additional Key Words and Phrases: parallel processing, compilers, data management, performance evaluation, multiprocessors