针对求解大型稀疏复对称线性方程组,提出了1种适合于分布式并行计算的并行化COCR(Conju—gateA-Orthogonal Conjugate Residual)方法,简记为PCOCR.在保证计算次序、矩阵向量乘积和向量校正不变的情况下,通过利用等价的数学推导,PCOCR方法将COCR方法每个迭代步所需的2次全局通讯降为了1次,同时,2种方法具有相同的数值稳定性.性能分析部分表明,所提出的PCOCR.方法比COCR方法具有更好的并行可扩展性,同时并行通讯性能改进比率趋于50%.
In this paper, a (parallelized Conjugate A-Orthogonal Conjugate Residual) (PCOCR method, in brief) method for solving large sparse linear systems with complex symmetric linear systems is proposed for distributed parallel computing. The method reduces two global synchronization points to one by using mathematical deduction. Meanwhile, it can keep the number of vector updates and matrix-vector multiplications constant and has the same numerical stability with COCR method. Performance analysis shows that PCOCR method has better parallelism and scalability than COCR method and the communica- tion performance can be improved by a factor of about 2.