二进制翻译中冗余指令优化算法

发布时间：2018-07-05 04:45

本文选题：二进制翻译 + 冗余指令　；参考：《计算机研究与发展》2017年09期

【摘要】：二进制翻译是实现软件移植的主要方法.动态二进制翻译受动态执行限制而不能深度优化导致效率较低而传统的静态二进制翻译难以处理间接分支,且现有的优化方法大部分集中在中间代码层,对目标码中存在的大量冗余指令较少关注.针对这一现状,提出一种静态二进制翻译框架SQEMU,基于该框架提出了一种对目标码冗余指令进行删除的优化算法.该算法通过分析目标码生成指令特定数据依赖图(instruction-specific data dependence graph,IDDG),再利用该图将活性分析和窥孔优化的2种理论相结合,有效删除目标码中的冗余指令.实验结果表明,利用该算法对目标码优化后,其执行效率得到显著提升,最大提升可达42%,整体性能测试表明,优化后nbench测试集翻译效率提高约20%,SPEC CINT2006测试集翻译效率提高约17%.
[Abstract]:Binary translation is the main method to implement software transplanting. Dynamic binary translation is limited by dynamic execution and can not be optimized by depth. Traditional static binary translation is difficult to handle indirect branches, and most of the existing optimization methods are concentrated in the middle code layer, and a large number of redundant instructions exist in the target code. In view of this situation, a static binary translation framework SQEMU is proposed, based on which an optimization algorithm for deletion of target code redundancy instructions is proposed. This algorithm is used to generate instruction specific data dependency graph (instruction-specific data dependence graph, IDDG) by analyzing target code, and then using the graph to analyze the activity and the activity analysis. The 2 theories of peephole optimization are combined to effectively delete the redundant instructions in the target code. The experimental results show that the efficiency of the algorithm is improved significantly and the maximum lift up to 42% after the optimization of the target code. The overall performance test shows that the translation efficiency of the nbench test set is improved by about 20% after the optimization, and the translation efficiency of the SPEC CINT2006 test set is raised. High about 17%.
【作者单位】：数学工程与先进计算国家重点实验室;
【基金】：国家自然科学基金项目(61472447) 国家“八六三”高技术研究发展计划基金项目(2009AA012201) “核高基”国家科技重大专项基金项目(2009ZX01036-001-001)~~
【分类号】：TP311.54

【相似文献】