基于重用距离的访存指令优化  

Optimization of Memory Access Instructions Based on Reuse Distance

作　　者：魏雨桐 顾乃杰[1] 黄章进[1] 苏俊杰[1] 齐东升 WEI Yutong;GU Naijie;HUANG Zhangjin;SU Junjie;QI Dongsheng(School of Computer Science and Technology,University of Science and Technology of China,Hefei 230027,China)

机构地区：[1]中国科学技术大学计算机科学与技术学院,合肥230027

出　　处：《小型微型计算机系统》2024年第11期2784-2789,共6页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(U20A20229)资助.

摘　　要：随着机器学习、推荐系统和社交网络等数据驱动类技术的发展,数据正在以流的形式呈现.传统的缓存替换算法无法有效适应应用程序的流式访问行为,导致数据流程序带来了大量的缓存未命中与严重的缓存污染问题.本文依据数据流程序变化带来的新的局部性优化挑战,提出了一种基于重用距离和非时态访存指令的优化方法RDNT.该方法首先筛选内存访问指令,然后计算重用距离,最后用非时态内存访问指令替换重用距离过大的常规内存访问指令.在SPEC CPU 2017测试集的实验结果表明,RDNT能够有效提高程序性能,与常规访存方式相比产生了8%的加速比,降低了程序的运行时间.As data-driven technologies such as machine learning and social networks become ubiquitous,the method of presenting data in the form of a continuous stream becomes more and more common.However,traditional cache replacement algorithms are unable to manage the streaming memory access pattern exhibited by modern applications,resulting in mandatory loss and cache pollution.As a result of this challenge in optimizing locality for data streaming programs,an optimization method RDNT based on reuse distance and non-temporal instructions is proposed.This method first filters memory access instructions,then calculates the reuse distance.and finally replaces the regular memory access instructions with non-temporal instructions.Results from experiments conducted using the SPEC CPU 2017 reveal that RDNT can effectively improve program performance,yielding a 8%speedup compared to the regular memory access method,reducing the running time of the programs.

关 键 词：LLVM编译器 缓存污染 内存访问指令 编译器优化 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]