[petsc-users] Solving multiple linear systems with similar matrix sequentially

[hong at aspiritech.org]

ML and Hypre both use RAP, as I recall.

PtAP uses outer product which cannot be as fast as RAP, based on an
analysis of memory access.

Sequential RARt might be fast if ARt has small number of colors. We do not
have parallel RARt.

Hong

On Mon, Feb 23, 2015 at 9:47 PM, Barry Smith <bsmith at mcs.anl.gov> wrote:

>
> > On Feb 23, 2015, at 9:41 PM, Jed Brown <jed at jedbrown.org> wrote:
> >
> > Barry Smith <bsmith at mcs.anl.gov> writes:
> >>   Freeze the hierarchy and coarse grid interpolations of GAMG but
> >>   compute the new coarse grid operators RAP for each new linear
> >>   system (this is a much cheaper operation). Use
> >>   PCGAMGSetReuseInterpolation() to freeze/unfreeze the hierarchy.
> >
> > The RAP is often more than 50% of PCSetUp, so this might not save much.
>
>    Hee,hee  in some of my recent runs of GAMG the RAP was less than 25% of
> the time. Thus skipping the other portions could really pay off.*
>
>   Barry
>
> * this could just be because the RAP portion has been optimized much more
> than the other parts but ...
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20150223/578674ad/attachment-0001.html>