作者投稿和查稿 主编审稿 专家审稿 编委审稿 远程编辑

计算机工程 ›› 2008, Vol. 34 ›› Issue (16): 95-97. doi: 10.3969/j.issn.1000-3428.2008.16.033

• 软件技术与数据库 • 上一篇    下一篇

程序自动并行化中的数组终写关系分析

罗 勇,张 平,龚雪容   

  1. (信息工程大学信息工程学院,郑州 450002)
  • 收稿日期:1900-01-01 修回日期:1900-01-01 出版日期:2008-08-20 发布日期:2008-08-20

Array Last-write Relation Analysis for Automatic Program Parallelization

LUO Yong, ZHANG Ping, GONG Xue-rong   

  1. (College of Information Engineering, University of Information Engineering, Zhengzhou 450002)
  • Received:1900-01-01 Revised:1900-01-01 Online:2008-08-20 Published:2008-08-20

摘要: 在程序自动并行化中过程中,数据收集阶段可能产生冗余通信,该文利用数组终写关系分析的方法来消除冗余通信,实现嵌套循环中数组数据最后写关系的快速求解,并将结果提供给编译器后端,生成精确数据收集代码。描述数组终写关系的研究目的和内容,将所处理的嵌套循环根据其结构特征进行分类,给出实现算法的过程。测试结果证明了该算法的正确性和高效性,所产生的精确数据收集代码能够有效地消除部分冗余通信,从而优化和提高了并行化程序的性能。

关键词: 并行编译, 数组终写关系分析, 精确数据收集, 自干扰, 线性不等式

Abstract: In order to eliminate redundant communication generated in data collection of automatic program parallelization, array last-write relation analysis is proposed to implement fast computation of array data last-write relation in loop nest, whose result is necessary condition for generating accurate data collection code in backend of compiler. This paper describes the aim and content of array last-write analysis, sorts the loop nests into different types according to their construction characteristic, and presents the algorithm implementation process in detail for each type. The test result shows the high performance and accuracy of the algorithm. And the accurate data collection code can efficiently eliminate part of redundant communication, which optimizes the parallel programs.

Key words: parallel compiling, array last-write relation analysis, accurate data collection, self-interfere, linear inequality

中图分类号: