• 中国计算机学会会刊
  • 中国科技核心期刊
  • 中文核心期刊

J4 ›› 2006, Vol. 28 ›› Issue (4): 126-129.

• 论文 • 上一篇    下一篇

值预测技术中基本值预测模型的功耗分析

冀蓉 张民选 陈怒兴   

  • 出版日期:2006-04-01 发布日期:2010-05-20

  • Online:2006-04-01 Published:2010-05-20

摘要:

值预测是通过推断执行来提高ILP的有效技术之一,但实现其预测模型的硬件成本和功耗却是制约它的两个主要因素。本文从实现值预测的底层电路入手,着重探究了启用值预测时功耗的主要来源,并从性能与功耗比率的角度对目前值预测中采用的几种预测模型进行了比较和评估。

关键词: ILP VP 值预测 功耗 IPC Last预测 Stride预测 Context预测

Abstract:

Value prediction is one of the effective techniques that increase ILP by speculative execution, but its hardware cost and power consumption are two ma  in limitations. This paper focuses on studying the main sources of power consumption during the use of value prediction, and compare several value prediction models from the angle of power consumption.

Key words: ILP, VP, value prediction, power consumption, IPC, Last prediction, Stride prediction, Cont ext prediction