科学计算应用与流处理器  

A Study of the Applicability of Stream Processors to Scientific Computing Applications

作　　者：张英[1] 李根[1] 庞征斌[1] 李永进[1] 张俊[1] 夏军[1] 杨学军[1] 

机构地区：[1]并行与分布处理国家重点实验室,湖南长沙410073

出　　处：《计算机工程与科学》2009年第8期73-76,81,共5页Computer Engineering & Science

基　　金：国家自然科学基金资助项目(60621003)

摘　　要：研究科学计算应用在流处理器上的适用性已成为当前研究热点之一。本文首先分析了流处理器处理科学计算应用的优势以及在流处理器上开发科学计算面临的重大挑战;然后针对不同类型的科学计算应用给出了将科学计算应用映射到流处理器上的关键与优化方法;最后将八种具有不同性能特征的典型科学计算应用映射到流处理器上,并比较和分析这些流程序在时钟精确模拟器运行性能和在处理科学计算应用的相应Fortran程序在主流Itanium 2处理器上的运行性能。实验结果表明,流处理器能有效处理科学计算应用。Stream processors, developed for the stream programming model, perform well on media applications. This paper examines the applicability of stream processors to scientific computing applications. First, the advantages of stream processors for scientific computing applications and the challenges to developing scientific computing applications on the stream processors are analyzed. Then, the rules and methods to map scientific computing applications with different performance features to the stream processors are presented. Finally, eight testbenches of typical scientific computing applications are mapped to the stream processors, each having different performance characteristics. Their performances are compared on a cycle-accurate simulator to their FORTRAN versions on a/tanium 2 processor. The experimental results show the applicability of the stream processors to scientific computing applications.

关 键 词：流处理器 科学计算应用 流 SRF 

分 类 号：TP314[自动化与计算机技术—计算机软件与理论]