Increasing the lookahead of multilevel branch prediction

A. Veidenbaum
{"title":"Increasing the lookahead of multilevel branch prediction","authors":"A. Veidenbaum","doi":"10.1109/IWIA.1998.779074","DOIUrl":null,"url":null,"abstract":"Many techniques have been proposed for tolemting memory latency in future systems, including prefetching and Decoupled-Access DRAM (DA-DRAM) architectures. In order for these techniques to be effective they need to have a suficient lookahead, i.e. to be far enough ahead of processor execution in requesting data. Bmnch prediction has been utilized before to achieve this but only small degrees of lookahead have been studied. This is not enough for latencies of tens to hundreds of clock cycles. In particular, DA-DRAM systems need a higher degree of decoupling because the access processor is placed in memory. This paper investigates the potential of multilevel prediction for increased lookahead, beyond the 2-3 level predictors that have been reported. A new multilevel predictor is defined and up to 8 levels of lookahead are studied. We find that the predictor has a high degree of accuracy, over 80%, even for integer benchmarks such as gcc and Oracle. Branch outcome as well as the address of the next instruction after the branch are predicted. A comparison of singleand multiple lookup prediction is presented. In general, accuracy decreases with lookahead. However, given sufficient predictor size it can be made quite high. With the DA-DRAM predictor located in DRAM, a purely software predictor with storage in the DRAM can be implemented. A similar predictor can be used for instruction and L2 cache prefetching or trace scheduling.","PeriodicalId":418008,"journal":{"name":"Innovative Architecture for Future Generation High-Performance Processors and Systems","volume":"70 1","pages":"0"},"PeriodicalIF":0.0000,"publicationDate":"1900-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Innovative Architecture for Future Generation High-Performance Processors and Systems","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1109/IWIA.1998.779074","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 0

Abstract

Many techniques have been proposed for tolemting memory latency in future systems, including prefetching and Decoupled-Access DRAM (DA-DRAM) architectures. In order for these techniques to be effective they need to have a suficient lookahead, i.e. to be far enough ahead of processor execution in requesting data. Bmnch prediction has been utilized before to achieve this but only small degrees of lookahead have been studied. This is not enough for latencies of tens to hundreds of clock cycles. In particular, DA-DRAM systems need a higher degree of decoupling because the access processor is placed in memory. This paper investigates the potential of multilevel prediction for increased lookahead, beyond the 2-3 level predictors that have been reported. A new multilevel predictor is defined and up to 8 levels of lookahead are studied. We find that the predictor has a high degree of accuracy, over 80%, even for integer benchmarks such as gcc and Oracle. Branch outcome as well as the address of the next instruction after the branch are predicted. A comparison of singleand multiple lookup prediction is presented. In general, accuracy decreases with lookahead. However, given sufficient predictor size it can be made quite high. With the DA-DRAM predictor located in DRAM, a purely software predictor with storage in the DRAM can be implemented. A similar predictor can be used for instruction and L2 cache prefetching or trace scheduling.
增加多级分支预测的前瞻性
在未来的系统中,已经提出了许多技术来缓解内存延迟,包括预取和解耦访问DRAM (DA-DRAM)架构。为了使这些技术有效,它们需要有足够的前瞻性,即在请求数据的处理器执行之前足够远。在此之前,已经使用了大量预测来实现这一点,但只研究了很小程度的前瞻性。对于数十到数百个时钟周期的延迟来说,这是不够的。特别是,DA-DRAM系统需要更高程度的解耦,因为访问处理器位于内存中。本文研究了多层预测的潜力,以增加前瞻性,超过2-3级预测已经报道。定义了一种新的多水平预测器,并研究了多达8个水平的前瞻性。我们发现该预测器具有很高的准确度,超过80%,甚至对于整数基准测试(如gcc和Oracle)也是如此。分支结果以及分支后的下一条指令的地址被预测。对单查找预测和多查找预测进行了比较。一般来说,准确性随着前视而降低。然而,给定足够的预测器大小,它可以相当高。由于DA-DRAM预测器位于DRAM中,因此可以实现一个存储在DRAM中的纯软件预测器。类似的预测器可以用于指令和L2缓存预取或跟踪调度。
本文章由计算机程序翻译,如有差异,请以英文原文为准。
求助全文
约1分钟内获得全文 求助全文
来源期刊
自引率
0.00%
发文量
0
×
引用
GB/T 7714-2015
复制
MLA
复制
APA
复制
导出至
BibTeX EndNote RefMan NoteFirst NoteExpress
×
提示
您的信息不完整,为了账户安全,请先补充。
现在去补充
×
提示
您因"违规操作"
具体请查看互助需知
我知道了
×
提示
确定
请完成安全验证×
copy
已复制链接
快去分享给好友吧!
我知道了
右上角分享
点击右上角分享
0
联系我们:info@booksci.cn Book学术提供免费学术资源搜索服务,方便国内外学者检索中英文文献。致力于提供最便捷和优质的服务体验。 Copyright © 2023 布克学术 All rights reserved.
京ICP备2023020795号-1
ghs 京公网安备 11010802042870号
Book学术文献互助
Book学术文献互助群
群 号:604180095
Book学术官方微信
小红书