首页> 外文会议>International conference on computer engineering and network >Leakage Power Reduction of Instruction Cache Based on Tag Prediction and Drowsy Cache
【24h】

Leakage Power Reduction of Instruction Cache Based on Tag Prediction and Drowsy Cache

机译:基于标签预测和困倦缓存的指令缓存泄漏功率降低

获取原文

摘要

Tag prediction is proposed to reduce the leakage power consumption of instruction cache and the power consumption of branch prediction that represent a sizeable fraction of the total power consumption of embedded processors in this chapter. By extending the architectural control mechanism of the drowsy cache to predict the cache line read in the next access, the tag prediction wakes up the necessary cache line in advance, while the rest of cache line is in the drowsy mode. Empirical results show that the tag prediction reduces the 77 % power consumption compared to the policy adopting branch prediction, and the accuracy of tag prediction is roughly same with the accuracy of BTB prediction. By removing the BTB and adopting the technique of drowsy cache, the tag prediction effectively reduces the power consumption without significant impact on performance of processors.
机译:在本章中,提出了标记预测以减少指令高速缓存的泄漏功耗和分支预测的功耗,这些功耗占嵌入式处理器总功耗的很大一部分。通过扩展困倦缓存的体系结构控制机制以预测下一次访问中读取的缓存行,标签预测可以提前唤醒必要的缓存行,而其余缓存行则处于困倦模式。实验结果表明,与采用分支预测的策略相比,标签预测的功耗降低了77%,标签预测的准确性与BTB预测的准确性大致相同。通过删除BTB并采用困倦的缓存技术,标签预测可以有效地降低功耗,而不会显着影响处理器的性能。

著录项

相似文献

  • 外文文献
  • 中文文献
  • 专利
获取原文

客服邮箱:kefu@zhangqiaokeyan.com

京公网安备:11010802029741号 ICP备案号:京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有
  • 客服微信

  • 服务号