一种基于微基准程序和理想上限的处理器性能分析方法

A Microprocessor Performance Analysis Approach Based on Microbench and Ideal Bound

随着现代高性能通用处理器结构的不断发展,处理器的性能分析已经变得越来越困难.基于大工作负载和单纯依靠模拟器的性能分析方法复杂度高,且难以直观地反映微体系结构特征.本文针对超标量处理器的特点,提出一种新的处理器性能分析方法,具体包括:设计一个微基准程序集Godson-Microbench,并提出相应的理想性能上限计算公式.这种方法扩充了过去基于约束的性能分析方法,可以更加完整地评估流水线性能并有效地发现性能瓶颈.本文使用这种方法分析比较了龙芯2号处理器和Alpha21264处理器,并依此改进了龙芯2号的结构,使得微基准程序的平均性能提高了13.8%,SPEC程序集的IPC提高了28.8%.本文提出的性能分析方法在龙芯2号的结构优化工作中发挥了重要作用.

The performance analysis process of microprocessor has become more and more difficult with the rapid development of modem processor architecture. The traditional performance analysis approach based on macro benchmarks and simulator is complex and time-consuming. Moreover, it can not analyze the micro-architectural characteristics systematically. In this paper, a novel performance analysis approach based on micro-benchmark （Godson-Microbench） and ideal-bound computation formulas is proposed.This approach extends the bounds based performance analysis approach.Compared with the traditional approach,this ap- proach can be used to analyze the pipeline efficiency and identify the performance bottleneck. With this approach, the micro-archi- tecture of the Godson-2 processor and Alpha21264 processor are analyzed and compared. On the basis of this work, we improve the architecture of Godson-2,resulting a 13.8% increase in performance of the micro-benchmarks and 28.8% increase in IPC of the SPEC CPU2000. This performance analysis approach is a complement of the cycle-accurate simulation and plays an important role in the architectural optimization process of Godson-2.