A branch prediction algorithm and predictor called Pshare is proposed, which has the benefit of Pas and the Gshare predictor,and modifies their flaws due to using a global history pattern by u-sing a separate branch history shift register(BHSR). Moreover it also can reduce the area and delay overhead than Pas by accessing a small PHT table using the hash of address and history patterns instead of accessing two dimensions PHT table using the address and history patterns. The Pshare predictor can obtain the branch prediction accuracy of Pas , but the implementation overhead is near to Gshare. This predictor is applied to high performance processors which have a superscalar deep pipeline and focuses more on higher core performance. It reduces the pipeline's vertical waste and improves the core performance and efficiency by using the Pshare predictor.%本文提出了一种结合PAs和Gshare两者优点,同时尽可能消除两者缺点的两级自适应分支预测算法及其实现-Pshare分支预测方法.通过使用单独的BHSR来跟踪不同分支指令的历史轨迹,克服了GAs和Gshare机制中统一历史模式带来的缺点；同时,通过使用分支指令部分地址和历史模式位的杂凑运算来索引共享的一维PHT小表,减少了PAs机制中二维PHT大表的面积开销和索引延时开销.Pshare两级自适应分支预测器用接近Gshare的实现代价,达到了近似PAs的预测精度.该分支预测技术适用于强调单核性能的超标量超流水结构的高性能处理器中,有助于减少垂直浪费,提高单核心的性能和效率.
展开▼
