反馈式编译优化在转移预测中的研究

Study on Feedback Directed Optimization for Branch Prediction

处理器性能的发挥常常受到转移指令的限制,所以转移预测的成功与否对于处理器的性能影响至关重要。反馈式编译优化是一种基于程序当前和以前运行时的趋势来改变程序以后执行动作的技术,能够提供给编译器一些有用的优化信息。本文针对ALPHA中的结构特点,利用反馈式编译优化技术,提高了ALPHA中的转移预测命中率,实验结果表明,加速比效果较为明显。

The performance of processor is restricted by the branch instructions,and whether the prediction is right or not has a strong impact on it.Feedback directed optimization is a technique which can execute operation after changing programs based on tendencies of current and previous program running,and it can supply useful information for compiler optimizations.This passage adopts this technique according to the features of ALPHA structure and improves the correction of branch prediction schemes.The experiment result shows that the effect of this optimization is evident.