分支预测对超标量流水线性能影响分析

The Impact Analysis of Branch Prediction for the Performance of the Superscalar Pipeline

随着RSIC处理器体系结构的发展,指令级并行度的提高成为微处理器架构研究热点,分支预测技术作为解决流水线中控制相关问题的关键技术,对指令级并行度有着重要影响.通过对超标量处理器中流水线性能与分支预测相关的因素的关系分析,以及和标量处理器性能的对比,基于合理的假设条件得到它们之间关系的解析函数,从不同的角度深入分析了分支预测算法中分支预测正确率、分支指令所占比例、分支预测错误时刻、程序指令数大小、流水线深度以及指令并行发射宽度对流水线性能影响,为现代RISC处理器的研究以及分支预测技术的研究提供参考.

With the development of RSIC processor architecture,instruction-level parallelism has become a research focus.Branch prediction techniques is a key technology to solve control-related,and has an important impact on ILP.This paper analyzes the relationship between the superscalar pipeline performance and the factors of branch prediction techniques,and the deference of the relationship between superscalar pipeline and scalar pipeline,then gets some analytic functions expressing their relationship based on reasonable assumptions.Finally,this paper forms an in-depth impact analysis between the branch prediction parameters and the pipeline performance from different perspectives to provide a reference for the study of RSIC processor and branch prediction techniques,and these parameters include the accuracy of prediction,the proportion of branch instruction,the error moment,the program size,the pipeline depth and the instruction parallel issue width.