动态分支预测技术分析与量化研究

Dynamic branch prediction analysis and quantitative research

随着现代微处理器的流水线变得越来越深,指令发射宽度越来越大,预测失败所带来的时间代价和空间(功能部件)有效利用率损失也越来越大,预测准确率对流水线的性能的影响也越来越重要。为了量化研究不同动态分支预测技术的准确率,采用计算机体系结构模拟软件Simple Scalar和SPEC 2000 Benchmark,对2bC,GAg,PAg,GAp,PAp,G-share六种动态预测技术的分支方向和分支地址预测准确率进行了性能模拟与测试,最后对模拟结果进行统计分析,得出预测准确率与硬件实现复杂度之间的量化关系,为设计出具有高准确率和低复杂度的动态分支预测器提供了参考。

With the pipeline of modern microprocessors becoming deeper,and the width of dispatched instructions increasing continuously,the performance loss（including delay and efficient utilization of hardware） caused by prediction failure become so significant that the accuracy of prediction on the pipeline＇s performance is also increasingly important.In order to quantify the prediction accuracy of different dynamic branch methods,this paper uses computer-architecture emulation software： Simple Scalar,and SPEC 2000 Benchmark,to evaluate the performance of both branch addresses and branch direction on six different methods： 2bC,GAg,PAg,GAp,PAp,and G-share.The statistical data demonstrate the quantitative relationship between the prediction accuracy and hardware implementation complexity,which is valuable for the design of high accuracy and low complexity of the dynamic branch predictor.