摘要
针对 CFD( computational fluid dynamics)问题中的点松弛和线松弛迭代算法 ,研究其帧内和帧间数据相关性 ,提出一种数据相关性分析算法和通信策略。与相应的程序重构技术相结合 ,实现这类程序的 SPMD模式的自动并行化。该算法与平台无关 ,能够适用于消息传递 ( MP)和共享变量的通信机制 ,目前已在 PVM环境中实现。经测试 ,基于该算法自动生成的并行程序能够达到很高的并行效率 ,对于绝大多数算例 ,其相关性和通信点的识别和归约可达到手工处理的效果。
In this paper, we present an algorithm for data dependency analysis of CFD programs, and a related communication strategy. In section 2, based on the features of CFD programs, we proposed an algorithm (13 steps) for recognizing in frame and inter frame data dependency. In section 3, we proposed a communication strategy (20 steps) for reducing communication spots. This algorithm is platform independent and suitable for both message passing (MP) and shared variable mechanisms. This algorithm has been implemented in PVM (parallel virtual machine). The test results of several practical CFD samples are given in tables 1 through 3. They show that this algorithm achieves rather high parallel efficiency. For most of the samples, this algorithm is as efficient as manual parallelization both in data dependency recognition and in the reduction of communication spots.
出处
《西北工业大学学报》
EI
CAS
CSCD
北大核心
2000年第3期341-344,共4页
Journal of Northwestern Polytechnical University
基金
国防科技实验室项目!(99JS94.6 .1)
关键词
相关性分析
自动并行化
计算流体动力学
程序
computational fluid dynamics (CFD), data dependency analysis, reduction of communication spots, auto parallelization
