Solving linear systems with vectorized WZ factorization

Abstract

In the paper we present a vectorized algorithm for WZ factorization of a matrix which was implemented with the BLAS1 library. We present the results of numerical experiments which show that vectorization accelerates the sequential WZ factorization. Next, we parallelized both algorithms for a two-processor shared memory machine using the OpenMP standard. We present performances of these four algorithms on a two-Pentium III machine with a Linux system (the parallelized sequential algorithm is better than the normal sequential one, but the parallelized vectorized algorithm is very similar in its performance to the non-parallelized vectorized one).