基于数组元素视图的并行化和特征  

Parallelization and Parallelism Characteristic Based on Array Element Graph

作　　者：曾国荪[1] 陆鑫达[2] 

机构地区：[1]同济大学计算机科学及工程系,上海200092 [2]上海交通大学计算机科学及工程系,上海200030

出　　处：《计算机科学》2001年第5期15-19,70,共6页Computer Science

基　　金：国家自然科学基金(69773014)

摘　　要：1 引言近年高性能并行计算研究领域已转到异构计算[1,2],因为一个复杂的计算问题,如Grand challenge问题[3],往往存在多种计算需求,任何单一的计算模式,如: SISD, SIMD, MIMD,数据流等,都不能使求解过程顺利进行.只有按需分配计算模式,才能取得计算性能最优.异构计算的关键是并行性特征的提取.我们已经提出了基于程序结构和基于程序运行性能分析的两种方法[4～6],但还不完善.本文希望通过数组元素视图作深入的研究.In order to realize heterogeneous computing,one must firstly find out the parallelism characteristic for a segment of program. But the existing theory of parallel processing is not sophisticated,and prevents parallel computing from becoming the mainstream in computer applications. Therefore,it is necessary to achieve breakthrough in schemes of research. In this paper,a novel approach to data dependence analysis,based array element graph,is proposed. By means of graph theory and algebra abstract, the authors infer and prove what parallelism characteristic types of single nested loop program have. The paper also establishes a foundation for further research in such area.

关 键 词：数组元素视图 并行化 异构计算 算法 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]