Thread+id+Cache%3a一种多线程共享低功耗高速缓存结构.pdfVIP

 Thread id Cache ：一种多线程共享的低功耗高速缓存结构 1+ 1,2 1 1 1 范灵俊 , 徐远超 , 唐士斌 ，郑亚松 ，范东睿 1( 中国科学院计算技术研究所 计算机体系结构国家重点实验室,北京 100190) 2(首都师范大学,信息工程学院,北京,100048) * Thread id Cache: A Low Power Cache Architecture Shared by Multiple Threads 1+ 1,2 1 1 1 FAN Ling-Jun , XU Yuan-Chao , TANG Shi-Bin , ZHENG Ya-Song , FAN Dong-Rui 1(State Key Laboratory of Computer Architecture, ICT, CAS, Beijing 100190, China) 2(College of Information Engineering, Capital Normal University, Beijing 100048, China) + Corresponding author: Phn: +86- 10 Fax: +86-10 E-mail: fanlingjun@ict.ac.cn Abstract : Power has been a big issue in processor design for several years. This paper presents Thread id Cache, which is shared by multiple threads, and by which unnecessary way accesses to private data of other threads from the request thread can be eliminated , to reduce dynamic power consumption. Experiments conducted on 128KB-8Way Thread id Cache for both testing conventional parallel programs and parallel programs for processing big data show that, 69.8%-87.5% way accesses could be eliminated due to different applications and number of executing threads, and equivalent energy caused by way accesses could be saved accordingly. Key words: Set-associative Cache; Dynamical Power; Multiple threads; Thread id Table 摘 要: 近年来，功耗是处理器设计领域的关键问题之一。本文提出了一种针对多线程共享缓存通过过滤不必 要缓存路 （Way ）访问来降低CPU 动态功耗的缓存结构，称为Thread id Cache 。通过设置和预先访问Thread id 过滤器来排除每次访问发生时不同线程对组相联缓存同一组（Set）中其它线程私有数据块的访问。在 128KB 8 路Thread id Cache 上对传统并行程序和大数据处理并行程序进行的测试表明，对于不同应用、不同执行线程数， 69.8%-87.5%的缓存路访问可以得到消除，相应地，由于访问操作产生的同等的动态功耗可以被有效降低。 关键词: 组相联缓存; 动态功耗; 多线程; Thre