论文部分内容阅读
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，但是直到目前为止，还没有任何自动提取程序并行特征的研究报导。这方面的工作难度大，概念和术语含糊不清，本文从程序结构和程序运行的观点出发，清楚地给出了单指令多数据流（ＳＩＭＤ）、多指令多数据流（ＭＩＭＤ）并行性的形式定义，它是提取程序异构性的依据。文中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 area of parallel processing and is expected to achieve super linear speedup. Heterogeneity of extraction programs is an important step in heterogeneous computing, but until now, there has not been any research report on the parallelism of automatic extraction programs. In this aspect, the work is difficult and the concepts and terminology are ambiguous. In this paper, from the point of view of program structure and program running, this paper clearly presents the parallelism of SIMD and MIMD Formal definition, it is the basis for extracting the heterogeneity of the program. In the paper, two methods based on program structure transformation and performance-based analysis are also proposed, which can be used as a framework to create the heterogeneity of automatic extraction procedures.