分支嵌套循环的自动并行化研究  

Auto-parallelization Research Based on Branch Nested Loops

作　　者：丁丽丽[1] 李雁冰[1] 张素平[1] 王鹏翔[1] 张庆花[1] 

机构地区：[1]解放军信息工程大学数学工程与先进计算国家重点实验室,郑州450002

出　　处：《计算机科学》2017年第5期14-19,52,共7页Computer Science

基　　金：国家高技术研究发展计划(863计划)(2009AA01220);"核高基"重大专项(2009zx01036-001-001-2)资助

摘　　要：GCC编译器是一种受广大研究者青睐的开源优化编译器,但它仅仅能够对完美嵌套循环进行依赖分析。为了更好地挖掘嵌套循环粗粒度的并行,深入研究了GCC5.1数据依赖分析过程,提出了一种能够处理分支嵌套循环的依赖测试方法。首先识别出分支嵌套循环,然后分析数组下标与分支嵌套循环外层索引变量的关系,最后计算出外层循环索引变量的距离向量,并通过检测距离向量判断循环是否存在依赖。实验结果表明,该方法能够正确、有效地分析出分支嵌套循环的依赖关系。GCC compiler is an open source compiler system which has won favour among many researchers, however, it is only able to analyze the dependence of perfect nested loop. In order to efficiently explore the granularity parallelism of nested loop, we deeply analyzed the data dependence of GCCS. 1 and put forward a dependence testing method of han- dling branch nested loop. At first, the branch nested loop is recognized. Then, the relationship between array index and index variable of outer loop is identfied. At last, the distance vector of outer loop is computed, and whether the loop has carried dependence or not is decided by examining the distance vector. The experimental results show that the proposed method can effectively recognize the dependence of branch nested loop.

关 键 词：数据依赖分析 GCC 完美嵌套循环 分支嵌套循环 距离向量 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]