摘要
本文对Wolfe86年提出的循环扭曲转换技术进行了重新认识,通过引入相关距离矩阵和相关方向矩阵概念,给出了扭曲变换多重紧嵌套循环的一般化方法,然后分析了循环扭曲对并行性和数据局部性的影响,最后讨论了它和其它转换技术之间的相互关系.
We revisit,in this paper,the loop skewing transformation technique which was previously proposed by M. Wolfe in 1986. By means of dependence distance matrix and dependence direction matrix, we first present a generalized method of skewing multiple level perfectly nested loops.Then we analyze the effect of loop skewing on parallelism and data locality. Finally, we discuss its relation with other transformation techniques.
出处
《电子学报》
EI
CAS
CSCD
北大核心
1994年第5期25-31,共7页
Acta Electronica Sinica
基金
军用机"八.五"预研资助
关键词
循环扭曲
相关距离矩阵
数据通信
Loop skewing
Dependence distance matrix
Dependence direction matrix
Parallelism
Date locality
Compiler
FORTRAN