线性方程组迭代法在流处理器上的映射与分析  

Mapping and Analysis of the Linear Algebraic Equations Iteration Algorithm on Stream Processors

作　　者：王桂彬[1] 唐沛蓉[1] 唐玉华[1] 

机构地区：[1]国防科技大学计算机学院,湖南长沙410073

出　　处：《计算机工程与科学》2007年第7期88-90,95,共4页Computer Engineering & Science

基　　金：校预研"VIM"项目专项(CX04-06-018)

摘　　要：斯坦福大学的Imagine流处理器具有很强的计算能力,如何将该体系结构应用在科学计算领域是当前研究的热点。解线性方程组的迭代法在工程和科学计算的各个领域中有着十分广泛的应用,该算法具有较好的计算密集性和并行性,十分适合流处理器的计算模型。本文分别针对系数矩阵的规模大小和稠密程度,介绍了Jacobi和Seidel迭代在流处理器上的映射。实验结果表明,迭代算法能高效地开发Imagine的计算能力,取得较高的性能加速。The Imagine stream processor of Stanford University has powerful computing ability, and the key point is how to apply this architecture into the domain of scientific computing. The iteration method in resolving linear algebraic equations has broad applications in engineering and scientific computing. The algorithm has great computing intensity and parallelism, which are suitable for the computing model, in stream processors. In this paper, we introduce the mapping of the Jacobi iteration and the Seidel iteration onto the Imagine stream processor according to the size and denseness of the coefficient matrix respectively. The result shows that the iteration algorithm can exploit the computing ability of Imagine effectively and get high performance speedup.

关 键 词：IMAGINE Jaeobi迭代 Seidel迭代 流编程模型 

分 类 号：TP314[自动化与计算机技术—计算机软件与理论]