间接转移指令在动态二进制翻译器中的预测方法  

Indirect Branch Prediction in Dynamic Binary Translator

作　　者：徐超豪[1] 臧斌宇[1] 

机构地区：[1]复旦大学计算机科学与工程系,上海200433

出　　处：《小型微型计算机系统》2007年第11期2080-2084,共5页Journal of Chinese Computer Systems

基　　金：博士点基金项目(20050246020)资助.

摘　　要：动态二进制翻译器中间接转移指令的预测优化对其性能至关重要.在研究了动态二进制翻译器的取值信息收集框架后,给出如何在程序运行起始阶段对间接转移指令的目标地址进行取值信息收集的方法,并结合统计原理,提出对目标地址的预测应该结合取值信息进行区间估计.最后,提出了一个基于取值信息收集分析结果进行目标地址预测的优化决策函数并且通过在动态二进制翻译器"IA-32 Execution Layer"上进行实验验证了该决策函数对于间接转移指令预测的效果,并讨论了其准确性和精确度.The prediction and optimization of indirect branch is analyzing the framework of value profile within dynamic binary critical to the performance of dynamic binary translator. After translator, this paper suggested a method of how to collect value profile of indirect branch target addresses in program＇s initial execution. Based on theory of statistics, this paper suggested an innovative way of indirect branch prediction with the method of interval estimation and proposed an optimization decision function based on profiling and code specialization. Finally, experiments have been conducted on IA-32 Execution Layer and results showed that the optimization decision function was not only correct but also accurate.

关 键 词：动态二进制翻译 间接转移指令预测 取值信息收集 特殊代码处理优化 

分 类 号：TP303[自动化与计算机技术—计算机系统结构]