Block Gauss-Huard algorithm with column pivoting on a hybrid CPU-GPU architecture

摘要

In this paper, different variants of the block Gauss-Huard algorithm with column pivoting and correspondingly interchanging of columns are presented and implemented on a hybrid CPU-GPU architecture. The study gives numerical evidence that the algorithms yield numerical solutions as good as those obtained by Gaussian elimination with partial pivoting and performance evidence that they are more suitable for parallel architectures. This is confirmed by solving systems of linear equations for a set of randomly generated matrices. Copyright (C) 2016 John Wiley & Sons, Ltd.