From: Jed B. <je...@je...> - 2019-05-02 18:04:38
|
John Peterson <jwp...@gm...> writes: > On Thu, May 2, 2019 at 12:35 PM Yuxiang Wang <yw...@vi...> wrote: > >> Dear all, >> >> A quick question - when using libmesh, should I consciously do bandwidth >> minimization when I number my node IDs? Or, are those taken care of at >> pre-processing stage of PETSc and other solvers? >> > > I wouldn't try and do anything manually since PETSc has a number of > renumbering algorithms that I think you can play around with pretty easily > to see if they make any difference... You can run factorizations in a different ordering, but low-bandwidth orderings make a difference for cache reuse in MatMult. PETSc can't reorder that without making a copy. But don't put too much effort into it. |