[petsc-dev] LU for MPIAIJ

[Mark F. Adams]

"Redundant" is an interesting place in the design space but the important thing is to reduce the active number of processors to where your accurate coarse grid solver is happy (e.g., one processor).  "Redundant" just confuses the issue.  I think we should purge "redundant" from the repo, remove the logic that puts the coarse grid on one processor from GAMG, have LU/MPI do the to coalescing by default (as Jed suggested).

Then if we really want to have the redundant option then add this on top of the processor reduction.  GAMG coarsens w/o doing redundant before the coarsest grid so doing redundant on the coarsest grid make no sense -- thats why I try to defeat the redundant coarse grid solves.  I could add code to make the recursive processor reduction redundant (I've done it before in another code) but have not interest in doing it for GAMG (it is an interesting point in the design space but I don't want to implement/deploy/maintain every interesting place in the design space).


On Jan 7, 2013, at 9:55 PM, Jed Brown <jedbrown at mcs.anl.gov> wrote:

> On Mon, Jan 7, 2013 at 8:51 PM, Barry Smith <bsmith at mcs.anl.gov> wrote:
> > That wouldn't be redundant at all, so if we put it there, we'd have to at least rename it.
> 
>    I can teach you how to use etags or you can teach me how to use PERL :-)
> 
> It's not a technical problem, but we have to decide what this component really does. Perhaps merge with Redistribute?

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-dev/attachments/20130107/ab844122/attachment.html>