一个有效的并行分析算法  

AN EFFICIENT PARALLEL ANALYSIS ALGORITHM

作　　者：胡永刚[1,2] 乔如良[1,2] 

机构地区：[1]国家智能计算机研究开发中心 [2]中国科学院计算技术研究所

出　　处：《计算机学报》1999年第2期134-140,共7页Chinese Journal of Computers

摘　　要：并行分析在并行编译系统中有着很重要的作用，它的优劣直接影响到编译系统的成败．随着机群系统及其并行开发环境的发展，多数的并行系统可支持多重并行循环的运行．而对只支持一重并行循环的编程系统，选择并行运行效率最高的循环，也是很重要的．为此，本文提出了一个有效的循环并行分析方案．它不但能给出多层循环的并行性，而且能够处理绝大部分实际应用中的并行性问题．本文对传统的并行分析算法进行了修改，并给出了一个有效的并行分析算法．该方案已在一个并行Ｃ编译中实现．Parallel analysis is a very important technique in parallel compiler. A good parallel analysis algorithm will make a parallel compiler success. By the development of computer cluster and its parallel program exploitation environment, many parallel computer systems support multi level parallel loop execution. Moreover, those parallel programming systems that only support one level parallel loop may face the fact of selecting a perfect parallel loop to enhance the efficiency of the loop. After analyzing the serial loops, the parallel analysis pass may tell code generation pass which loop is parallel and which loop is serial. Marking all paralleling loops can reduce the spending of code generation. Therefor, this paper proposes a practical parallel analysis scheme. It not only can generate multiple parallel loops, but also can process many loops of practical programs. It improves traditional parallel analysis algorithm and gives an efficient parallel analysis algorithm. It will not end the processing of analyzing loops after find a parallel loop, and can find all parallel loops in all levels of a nested loop. This scheme has implemented in a parallelizing compiler system with C language.

关 键 词：依赖图 并行分析算法 C语言 并行编译系统 

分 类 号：TP314[自动化与计算机技术—计算机软件与理论]