首页 | 本学科首页   官方微博 | 高级检索  
     

一种Runahead执行的改进算法
引用本文:刘德峰,谢伦国. 一种Runahead执行的改进算法[J]. 计算机工程与科学, 2009, 31(Z1). DOI: 10.3969/j.issn.1007-130X.2009.A1.025
作者姓名:刘德峰  谢伦国
作者单位:国防科技大学计算机学院,湖南,长沙,410073
摘    要:Runahead执行技术能够显著地提高计算机系统的存储级并行,而无需对处理器结构做出较大改动。但Runahead执行处理器要比传统处理器多执行很多指令,最多是正常执行指令数的三倍以上,大大增加了处理器的功耗。本文通过分析发现Runahead执行在预执行阶段会执行大量的无效指令,据此提出一种减少无效指令的方法来提高Runa-head执行处理器的效率。通过实验分析,在性能影响较小的情况下,该方法最多可以减少50%的Runahead执行处理器在预执行阶段执行的无效指令。

关 键 词:Runahead执行  存储级并行  cache不命中  检查点

An Improvement Algorithm to Runahead Execution
LIU De-feng,XIE Lun-guo. An Improvement Algorithm to Runahead Execution[J]. Computer Engineering & Science, 2009, 31(Z1). DOI: 10.3969/j.issn.1007-130X.2009.A1.025
Authors:LIU De-feng  XIE Lun-guo
Abstract:Runahead execution can significantly improve the memory-level parallelism, and processor needn't be modified a lot. But Runahead execution processor excute more instructions than traditional processor, even triple the number of instructions of the common execution, so it severely increases processors' power waste. After analyzing, we find there are many useless instructions executed when Runahead processor is in runahead's execting stage, We present a new algorithm to reduce the useless instructions. The results of the experimentions show that this improvement algorithm mostly can reduce 50% useless instructions in runahead execution processor's pre-excute stage, and basically don't affect the performance a lot.
Keywords:Runahead execution  memory-level parallelism  cache miss  checkpoint
本文献已被 万方数据 等数据库收录!
设为首页 | 免责声明 | 关于勤云 | 加入收藏

Copyright©北京勤云科技发展有限公司  京ICP备09084417号