Stride Equality Prediction for Value Speculation

Stride Equality Prediction for Value Speculation
复制标题

价值投机的跨步平等预测

DOI:
10.1109/lca.2022.3195411
复制
发表时间:
2022-07
影响因子:
2.3
通讯作者:
Weixia Xu
Weixia Xu
中科院分区:
计算机科学3区
文献类型:
--
作者:
Ling Yang;Libo Huang;Run Yan;Nong Xiao;Sheng Ma;Li Shen;Weixia Xu

文献摘要

参考文献

相似文献

步幅预测是价值预测的一个不可分割的部分。大多数混合预测器都无法消除这种模式。现有的步幅预测器可以处理规则步幅模式,但很难很好地适应间隔步幅模式。为了解决这个问题,我们引入了步幅相等预测(SEP)的概念,它预测当前指令的步幅特征等于上一个提交的相同指令的步幅特征。SEP可以很好地处理间隔式步幅模式,其在间隔期间总是执行恒定步幅,尽管在不同的间隔中可能存在不同的端点和步幅。SEP根据指令窗口中同一指令的最后一次提交出现和同一指令的数量来预测跨距相等指令的值。评估结果表明,SEP是有效的步幅值预测。它比增强步幅预测器平均高5.3%,比最先进的计算预测器基于上下文的计算值TAGE(CBC-VTAGE)平均高1.5%。此外,通过应用SEP更新条件,CBC-VTAGE可以获得性能增益,而无需额外的成本。
Stride prediction is a nonignorable part of value prediction. Most hybrid predictors cannot eliminate this pattern. Existing stride predictors can handle the regular stride pattern, but it is hard to fit well with the interval stride pattern. To deal with it, we introduce the notion of Stride Equality Prediction (SEP), which predicts the stride feature of the current instruction is equal to that of the last committed same instruction. SEP can deal well with interval-style stride patterns, which always perform a constant stride during an interval, although there may be different endpoints and strides in different intervals. SEP predicts the value of stride equality instructions from the last committed occurrence of the same instruction and the number of the same instruction in the instruction window. Evaluation results show that SEP is effective in stride value prediction. It outperforms the Enhanced Stride predictor for 5.3% and state-of-art computational predictor Context-based Computational Value TAGE (CBC-VTAGE) for 1.5% on average. Moreover, by applying the SEP update condition, CBC-VTAGE can obtain performance gain without extra cost.
DOI: --
发表时间: 2006
期刊: J. Instr. Level Parallelism
影响因子: --
作者:
André Seznec;P. Michaud
通讯作者: André Seznec;P. Michaud
DOI: --
发表时间: 2018
期刊: --
影响因子: --
作者:
Y. Ishii;Arm;Y. Ishii
通讯作者: Y. Ishii;Arm;Y. Ishii
DOI: 10.1109/micro.1997.645815
发表时间: 1997-12
期刊: Proceedings of 30th Annual International Symposium on Microarchitecture
影响因子: --
作者:
Yiannakis Sazeides;James E. Smith
通讯作者: Yiannakis Sazeides;James E. Smith
DOI: --
发表时间: 2018-06
期刊: --
影响因子: --
作者:
André Seznec
通讯作者: André Seznec
DOI: 10.1109/iccd46524.2019.00101
发表时间: 2019-11
期刊: 2019 IEEE 37th International Conference on Computer Design (ICCD)
影响因子: --
作者:
Kleovoulos Kalaitzidis;André Seznec
通讯作者: Kleovoulos Kalaitzidis;André Seznec