利用循环分割和循环展开避免Cache代价.pdfVIP

ISSN 1000-9825, CODEN RUXUEW E-mail: jos@ Journal of Software , Vol.19, No.9, September 2008, pp.2228−2242 DOI: 10.3724/SP.J.1001.2008.02228 Tel/Fax: +86-10 © 2008 by Journal of Software . All rights reserved. ∗ 利用循环分割和循环展开避免 Cache 代价 + 刘 利 , 陈 彧, 乔 林, 汤志忠 (清华大学 计算机科学与技术系,北京 100084) Optimization to Prevent Cache Penalty by Loop Partition and Loop Unrolling + LIU Li , CHEN Yu, QIAO Lin, TANG Zhi-Zhong (Department of Computer Science and Technology, Tsinghua University, Beijing 100084, China) + Corresponding author: E-mail: liuli03@ Liu L, Chen Y, Qiao L, Tang ZZ. Optimization to prevent cache penalty by loop partition and loop unrolling. Journal of Software, 2008,19(9):2228−2242. /1000-9825/19/2228.htm Abstract : Due to the increasing speed gap between memory system and processor, cache hierarchies have been implemented into memory system, but additional latency (cache penalty) is introduced. This paper presents an algorithm named as prevent cache penalty by loop partition-unrolling (PCPLPU), which can prevent cache penalty in loops by the combination of loop partition and unrolling. Experimental results show that PCPLPU can prevent cache penalty and improve the performance of programs. Key words: loop partition; loop unrolling; cache penalty; bank conflict 摘 要: 存储系统与处理器之间的速度差距逐渐变大,为此,cache 使用了分级机制,但这也带来了额外的存储延迟 (cache 代价).提出一种利用循环分割和循环展开相结合避免 cache 代价的 PCPLPU(prevent cache penalty by loop partition-unrolling)算法.实验结果表明,PCPLPU 算法能够有效避免循环代价,提高程序性能. 关键词: 循环分割;循环展开;cache 代价;bank 冲突 中图法分类号: TP311 文献标识码: A 由于存储系统与处理机之间的速度差距变大,分级机制应用到 cache