摘要
如何利用循环体诸如初值、终值、步长和下标式等数学对象的诸如恒等、正负、单调性一类的数学特征,按照严格推导出来的简单规则,绕过各种复杂的数据依赖关系图及其上诸如拓扑排序或等价变换等复杂的操作,直接从事向量化处理,这是极富实用价值的研究方向.文献[2,3,5]给出的可原形/准原形算法,以及本文的可反原形与拟离散性,正是这样的捷径.它们能使相当部分的常见循环,直接向量化.
It is a research orientation of great practical value to skip various complicated data dependence relation graphs and complex operations such as topologically ordering, equivalent transformation[1], based upon the simple rules strictly derived, and make a direct vectorization treatment by means of the math properties such as identity, positive/negative and monotonicity of the math objects in a loop body such as initial/final values, step and subscript expreeeion. The original-shape/quasi-orginal-shape algorithms presented in Ref. [2,3,5] , and the anti-original-shape and quasi-discreteness in this paper are very successful ways that enable a considerable part of common loops to be directly vectorized.
出处
《计算机学报》
EI
CSCD
北大核心
1991年第10期741-747,共7页
Chinese Journal of Computers
关键词
软件
反原形
向量化
拟离散性
Subscript tracing method, direct vectorization, anti-original-shape, quasi-discreteness.
