Mixed precision iterative refinement techniques for the WZ factorization
Beata Bylina, Jarosław Bylina · 2013
Abstract—The aim of the paper is to analyze the potential of the mixed precision iterative refinement technique for the WZ factorization. We design and implement a mixed precision iterative refinement algorithm for the WZ factorization with the use of the single (a.k.a. float), double and long double precision. For random dense square matrices with the dominant diagonal we report the performance and the speedup of the solvers using different machines and we investigate the accuracy of obtained solutions. Additionally, the results (performance, speedup and accuracy) for our mixed precision implementation based on the WZ factorization were compared to the similar ones based on the LU factorization. I.