摘要
指令流发射和指令Cache失效是处理器能量耗散的两个重要原因。松耦合的RISC指令集所产生的程序加剧了这样的能耗,而在片上Cache有限的网络设备如路由器、交换机中,因为指令流而遭受的性能下降和功耗增加更为严重。面向网络报文转发这一重要的网络功能服务,分析了网络报文转发的指令特性,并基于RISC-V指令集架构,重定制了RV32C压缩指令扩展集。经过Spike模拟器测试,优化后压缩率缩减至70%,动态指令压缩率为90%,同时在同等Cache条件下,使用定制压缩指令的指令Cache失效率比标准RISC-V降低了30%~70%。
Instruction stream delivery and instruction Cache failure are two of the important reasons for processor energy dissipation.Programs based on loosely coupled RISC instruction sets exacerbate such energy consumption.Cache-limited network devices such as routers and switches suffer more per-formance degradation and power consumption due to instruction related processing.This paper focuses on network packet forwarding,which is one of the most important network functions.Thought analy-zing instruction characteristics of network packet forwarding,we redefine the RV32 compressed instruc-tion extension set based on the RISC-V instruction set architecture and test it by Spike simulator.Ex-perimental results show that the optimized compression rate is reduced to 70%and the dynamic instruc-tion compression rate is 90%.Meanwhile,under the same Cache conditions,our RV32C set can reduce the instruction Cache failure by 30%?70%in comparison to the standard RISC-V set.
作者
吕倩茹
王彦鹏
曹壮
文梅
LU Qian-ru;WANG Yan-peng;CAO Zhuang;WEN Mei(College of Computer,National University of Defense Technology,Changsha 410073,China)
出处
《计算机工程与科学》
CSCD
北大核心
2018年第3期381-387,共7页
Computer Engineering & Science
基金
"十三五"国家重点研发计划(2016YFB1000401)
核高基重大转型"超级计算机处理器研发"(HGJ20152X01028101)
作者简介
吕倩茹(993-),女,重庆人,硕士生,研究方向为计算机体系结构。E-mail:theotuck@126.com;王彦鹏(1993-),男,河南许昌人,硕士生,研究方向为计算机体系结构。Email:wangyanpeng11@nudt.edu.cn;曹壮(1982-),男,湖南岳阳人,博士生,研究方向为计算机体系结构。E-mail:Caozh0031@163.com;文梅(1975-),女,湖南常德人,博士,研究员,研究方向为高性能微处理器体系结构和并行处理技术。E-mail:meiwen@nudt.edu.cn
