自动提取程序的SIMD和MIMD异构性  

AUTOMATICALLY EXTRACTING THE SIMD/MIMD HETEROGENEITY HIDING A PROGRAM

作　　者：曾国荪[1,2] 陆鑫达[1] 

机构地区：[1]上海交通大学计算机科学与工程系 [2]江西师范大学计算机科学系,南昌330027

出　　处：《计算机研究与发展》2000年第11期1397-1403,共7页Journal of Computer Research and Development

基　　金：国家自然科学基金资助!(项目编号 6 97730 14 )

摘　　要：异构计算是并行处理的一个新领域 ,可望达到超级线性加速比 .提取程序的异构性是异构计算的一个重要步骤 ,这方面的工作难度大 ,概念和术语含糊不清 .从程序结构和程序运行的观点出发 ,清楚地给出了 SIMD,MIMD并行性的形式定义 ,它是提取程序异构性的依据 .还提出了基于程序结构变换和基于性能分析的两种方法 ,该方法可作为开创自动提取程序异构性的框架 .Heterogeneous computing is a new branch of parallel processing and is expected to reach super speedup. It is an important step for heterogeneous computing to extract the heterogeneity from a program. Such work is very difficult and some concepts and terminologies are ambiguous. From the point of view of program structure and program running, the formal definitions about SIMD and MIMD parallelism are presented clearly, which are criteria of extracting program heterogeneity. Also proposed are two approaches that are based on program structure transformation and performance analysis. Such approaches can serve as the originating framework of automatically extracting the heterogeneity from a program.

关 键 词：程序异构性 异构计算 SIMD MIMD 

分 类 号：TP31[自动化与计算机技术—计算机软件与理论]