A precise dependence analysis for multi-dimensional arrays under specific dependence direction

Abstract In process of automatic parallelizing/vectorizing constant-bound loops with multi-dimensional arrays under specific dependence direction, the Lambda test is claimed to be an efficient and precise data dependence analysis method that can check whether there exist generally inexact ` real-valued ' solutions to the derived dependence equations. In this paper, we propose a precise data dependence analysis method – the multi-dimensional direction vector I test. The multi-dimensional direction vector I test can be applied towards testing whether there exist generally accurate ` integer-valued ' solutions to the dependence equations derived from multi-dimensional arrays under specific dependence direction in constant-bound loops. Experiments with benchmark showed that the accuracy rate and the improvement rate for the proposed method are approximately 33.3% and 21.6%, respectively.