软件流水领域多维数组数据相关性的测试

Testing Data Dependencies of Multi-Dimensional Arrays in Software Pipelining Field

体差不等式测试为适用于软件流水领域的高维数组数据相关性分析算法,通过使用更严格的限制条件,该算法可以获得比传统数据相关性分析方法更精确的结果。文章展开该算法的实验研究。实验表明,虽然体差不等式测试算法只能针对实可行解域进行数据相关性分析,但所得到的结果仍然与实际情况相吻合。对于科学计算循环中出现的大多数数据相关性判定问题,使用体差不等式测试算法可以获得很好的效果。

The dependence difference inequality test for analyzing data dependencies of multi-dimensional arrays in software pipelining field is more efficient and accurate than other traditional data dependence analysis methods by using more strict inequality limitations.This paper gives experimental results of the algorithm.Experimental results show that the algorithm coincides with reality even though it can only work with real feasible solution region.Experimental results also show that the dependence difference inequality test algorithm works very well for most of data dependencies appeared in scientific computation programs.