摘要
"龙腾R"是西北工业大学自主研制的32位高性能微处理器.该处理器的分支处理单元(BPU)能有效降低控制相关带来的延迟.通过分析已有的分支方向预测算法和分支目标地址预测策略,在分支处理单元总体约束下,合理分配分支方向预测和目标地址预测的实现代价,提出了一种基于混合分支预测器和经过改进的目标地址缓冲(BTB)结构的分支处理单元结构.该结构不仅比传统的由gshare分支方向预测器构成的分支处理单元预测准确率平均高出1%～2%,并具有面积小、功耗低的特点.
＂Longtium R＂ is a 32-bit high performance microprocessor developed by Northwestern Polytechnical University.Branch process unit（BPU）of the processor can remarkably reduce the latency caused by control hazard.Through analyzing some algorithms of branch direction predict and some strategies of branch target address predict,under the overall constraints of BPU,appropriately allocating the hardware cost,this paper proposes a new structure of branch process unit,based on a combined branch predictor and improved branch target buffer（BTB）.Comparing to the traditional BPU which is comprised of gshare branch predictor,this one increases prediction accuracy by 1%～2%,with characteristics of small area and low power consumption.
出处
《微电子学与计算机》
CSCD
北大核心
2010年第6期122-127,共6页
Microelectronics & Computer
基金
国家自然科学基金项目(60773223
60736012)
关键词
分支预测
分支处理
混合分支预测器
目标地址缓冲
branch prediction
branch process
combined branch predictor
branch target buffer
