间接转移指令在动态二进制翻译器中的预测方法

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.

Indirect Branch Prediction in Dynamic Binary Translator

The prediction and optimization of indirect branch is critical to the performance of dynamic binary translator.After analyzing the framework of value profile within dynamic binary 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.