摘要
提出了一种基于反向支配关系的指令重用优化,在编译器中计算条件分支的反向支配基本块以及基本块内的控制无关语句,将这些信息编码成一条指令放置在分支指令之前.修改过的寄存器重命名阶段根据指令提供的信息保留部分错路指令的结果以便恢复后重用,降低了处理器分支预测错误的代价.实验表明该方法对典型应用有性能提升,与纯硬件方法相比硬件开销更小.
We propose a method for instruction reuse based on post-dominance.In compiler there's a pass analyze post-dominator of conditional branches and control-independent instructions in post-dominator,result will be encoded as a 'hint'instruction before corresponding branch.Modified rename stage in CPU salvages instruction results selectively according to'hint'instruction,which will reduce penalty of branch misprediction.Our evaluation showed that a short hardware table could provide performance improvements of up to 1.97%.
出处
《微电子学与计算机》
CSCD
北大核心
2015年第3期6-10,共5页
Microelectronics & Computer
基金
国家自然科学基金面上项目(61170022
61373039)
关键词
寄存器集成
反向支配关系
Y-行为
控制无关
指令重用
register integration
post-dominance
Y-behavior
control-independence
instruction reuse