A Vectorizable Variant of some ICCG Methods
Henk A. van der Vorst · SIAM Journal on Scientific and Statistical Computing · 1982
The preconditioned conjugate gradient method can be a useful tool in solving certain very large sparse linear systems. If this is done on a vector machine like the CRAY-1, then it appears that some of the most effective preconditionings are difficult to vectorize. In this paper it is shown how a class of preconditionings can be modified in such a way that they become highly vectorizable while still easy to program. Numerical experiments that show the improvement in performance of the preconditioned conjugate gradient algorithm have been included.