期刊名称：International Journal of Grid and Distributed Computing
印刷版ISSN：2005-4262
出版年度：2014
卷号：7
期号：6
页码：257-274
DOI：10.14257/ijgdc.2014.7.6.21
出版社：SERSC
摘要：According to the traditional stabilized biconjugate gradient algorithm (BiCGSTAB) deficiency in data locality, this paper proposed a QTBiCGSTAB algorithm whose core idea is that recursively divides sparse matrix with quarter tree into sub-matrix and reorders them, to improve the hit ratio of cache and enhance the algorithm's efficiency. And the idea is good for algorithm being parallized, that is proved by the numerical experiments later. It mainly shows, firstly, QTBiCGSTAB algorithm is more efficiency than BiCGSTAB, and the speedup would reach 1:330. The target division length would be influnced on the algorithm's performance; Secondly, for large linear system, parallized QTBiCGSTAB is more efficiency than serial's.
