基于多面体模型的数据依赖分析方法  

Data-Dependence Analysis Method Based on Polyhedral Model

作　　者：李川[1] 陈朝晖[1] 

机构地区：[1]北京控制工程研究所,北京100190

出　　处：《空间控制技术与应用》2015年第5期43-47,共5页Aerospace Control and Application

基　　金：国家自然科学基金资助项目(91118007)

摘　　要：设计一种基于多面体模型的静态数据依赖分析方法,对程序中的循环体进行分析,将生存周期思想引入到数据的依赖分析中.数据的依赖关系中只有流依赖是无法消除的固有依赖,必须保持变换前的执行顺序,而输出依赖和反依赖可以通过标量扩展及向前替换等方法消去.对传统数据依赖分析进行改进,通过分析内存单元的生存周期,摒除不必要的伪依赖,从而可以对更多的循环体进行变换.通过实验表明了该方法的可行性和有效性.A static data-dependence analysis method for loops based on polyhedral model is designed. The concept of live range is introduced into analysis. Only flow dependences must keep consistent with the or- der that they appears in the original execution of the program. Output dependences and anti-dependences can be eliminated by scalar expansion or forward substitution. This analysis method reforms the traditional analysis by introducing live range and eliminating unnecessary false dependence, via which more loops can be transformed. The validity and efficiency of the presented method are demonstrated by experiment.

关 键 词：依赖分析 多面体模型 生存周期 循环变换 

分 类 号：TP31[自动化与计算机技术—计算机软件与理论]