期刊文献+

以基本块为单位的非顺序指令预取 被引量:4

The Non-Sequential Instruction Prefetching Based on Basic Blocks
下载PDF
导出
摘要 取指令能力的高低对微处理器的性能有很大影响。指令预取技术能够有效地降低指令Cache的访问失效率,提高微处理器的取指令能力,进而提高微处理器的性能。本文提出了一种由分支指令指导的、以基本块为单位的非顺序指令预取技术,每次预取将一个完整的基本块读入指令Cache。这种方法使用静态策略分析程序行为,实现所需的硬件复杂度低。模拟结果显示,该方法能够有效地提高指令Cache访问的命中率。 Instruction supply can influence processor performance greatly.Instruction prefetching is an effective mechanism to reduce the instruction cache miss rate.This paper proposes a nonsequential instruction prefetching mechanism,which is directed by branch instructions and prefetches a whole basic block each time.Experimental results show it can increase the instruction cache hit rate effectively.In addition,it predicts program behaviors statically so that it does not need any complex hardware to predict and restore.
出处 《计算机工程与科学》 CSCD 2003年第4期94-98,共5页 Computer Engineering & Science
基金 国家自然科学基金资助项目(60173040 6993303)
关键词 微处理器 指令处理模块 基本块 非顺序指令预取 执行模块 basic block instruction cache instruction prefetching branch prediction
  • 相关文献

参考文献6

  • 1R Colwell, R Nix, J O' Donnell, et al. A VLIW Architecture for a Trace Scheduling Compiler[ A ]. Proc of the 2nd Int' 1 Conf on Architectural Support for Programming Languages and Operating Systems[C]. 1987.180 - 192.
  • 2W Hwu, S Mahlke, W Chen, et al. The Superblock: An Effective Technique for VLIW and Superscalar Compilation[ J]. The Journal of Supercomputing, 1993,7:229 - 248.
  • 3C Xia, J Torrellas. Instruction Prefetching of Systems Codes with Layout Optimized for Reduced Cache Misses[ A] .23rd Annual Int' l Symp on Computer Architecture[ C]. 1996.
  • 4N Jouppi. Improving Direct-Mapped Cache Performance by the Addition of a Small Fully Associative Cache and Prefetch Buffers[ A].Proc of the 17th Annual Int'l Symp on Computer Architecture[ C].1990.
  • 5J Pieroe, T Mudge. Wrong-Path Instruction Prefetching[ A ]. 29th Int'l Symp on Microarchitecture[ C]. 1996.165 - 175.
  • 6D Joseph, D Grunwald. Prefetching Using Markov Predictors[ A ].24th Annual Int'l Symp on Computer Architecture[C]. 1990.

同被引文献17

  • 1International technology roadmap for semiconductors[EB/OL]. [2012-09-07].http ://www.itrs.net.
  • 2Jacobsen E,Rotenberg E, Smith J E.Assigning confidence to conditional branch predictions[C]//29th Annual International Symposium on Microarchitecture.Paris, France: ACM, IEEE, 1996: 142-152.
  • 3Grunwald D, Klauser A, Manne S, et al.Confidence esimation for speculation control[C]//25th International Symposium on Computer Architecture.Barcelona, Spain: ACM, IEEE, 1998 : 122-131.
  • 4Reungsang P, Park S K, Jeong S W, et al.Reducing cache pollution of prefetching in a small data cache[C]//Proceedings of the International Conference on Computer Design:VLSI in Computers & Processors.Washington, DC, USA: IEEE Computer Society,2001 : 530-533.
  • 5Sendag R,Lilja D J,Kunkel S R.Exploiting the prefetching effect provided by executing mispredicted load instructions[C]// Proceedings of the 8th International Euro-Par Conference on Parallel Processing.London, UK: Springer-Verlag, 2002 : 468-480.
  • 6Chen Y, Sendag R,Lilja D J.Using incorrect speculation to prefetch data in a concurrent multithreaded processor[C//Pro- ceedings of the International Parallel and Distributed Pro- cessing Symposium.Washington, DC, USA: IEEE Computer Society, 2003.
  • 7Armstrong D N, Kim H, Mutlu O, Patt Y N.Wrong path events: exploiting unusual and illegal program behavior for early mispredietion detection and recovery[C]//Proceedings of the 37th annual IEEE/ACM International Symposium on Micro- architecture.Washington, DC, USA: IEEE Computer Society, 2004: 119-128.
  • 8Sendag R, Yilmazer A, Yi J J, et al.The impact of wrong-path memory references in cache-coherent multiprocessor systems[J]. J Parallel Distrib Comput,2007(67) : 1256-1269.
  • 9李瑛.同时多线程结构指令流特性及取指技术研究[D].西安:西北工业大学,2004.
  • 10Henessy J L,Patterson D A.Computer architecture-a quanti- tative approach[M].4th ed.[S.1.]: Morgan Kaufmann Publishers Inc, 2007.

引证文献4

二级引证文献7

相关作者

内容加载中请稍等...

相关机构

内容加载中请稍等...

相关主题

内容加载中请稍等...

浏览历史

内容加载中请稍等...
;
使用帮助 返回顶部