一种基于综合历史信息的SMT结构分支预测算法  

An Intergrated History Information Branch Prediction Policy for Simultaneous Multi-threading Architecture

作　　者：王晶[1] 樊晓桠[1] 叶曾[1] 

机构地区：[1]西北工业大学计算机学院,西安710072

出　　处：《计算机科学》2008年第2期259-262,共4页Computer Science

基　　金：国家自然科学基金(60573143);新世纪优秀人才支持计划资助

摘　　要：在SMT结构中,可以同时从多个线程中取指。当可取指线程个数较少时,分支预测的重要性与在超标量处理器中的相比有增无减,因为SMT结构中转移误预测的代价更大了。影响分支预测准确率的关键因素是历史信息的组织方式和更新方式。本文仿真分析了这些因素对分支预测准确率的影响,提出了一种基于综合历史信息的分支预测算法——IHBP,把全局信息和局部信息结合在一起预测转移,解决了SMT结构中分支预测信息过时、混乱等问题,使得预测的准确率更具备鲁棒性。仿真结果表明:在8线程结构中,该算法与目前国际普遍采用的Gshare算法和Pag算法相比,分支预测准确率分别提高了8.5%和2.3%。By converting thread-level parallelism to instruction level parallelism, Simultaneous Multithreaded （SMT） processors are emerging as effective ways to utilize the resources of modern superscalar architectures. However, branch prediction is still very important because of high wrong branch prediction penalty. The organization and modification method of branch history information are believed main factors affect the accuracy of branch predcitor In this paper, the various factors are simulated on 1,2,4 and 8 threads condition, where the branch prediction accuracy is compared and analyzed. And then, a new branch prediction method, called IHBP （Integrated History Information Branch Prediction）, is proposed. It integrates both global history information and local history information to index Pattern History Table （PHT）. The simulation results show that this new branch prediction method can improve branch prediction accuracy by 8. 5% and 2. 3% over per-thread Gshare and Pag method respectively for 8 threads. This speedup is enhanced by an advantage of overcoming the adverse influence of outdated and scrambled branch history and to make the prediction more stable and more accurate.

关 键 词：线程级并行 同时多线程 分支预测 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] F231.5[自动化与计算机技术—控制科学与工程]