[petsc-users] Convert MatNest to MatAIJ

Dmitry Karpeev karpeev at mcs.anl.gov
Wed Oct 10 11:57:36 CDT 2012


On Wed, Oct 10, 2012 at 11:38 AM, Barry Smith <bsmith at mcs.anl.gov> wrote:

>
> On Oct 10, 2012, at 8:50 AM, Jed Brown <jedbrown at mcs.anl.gov> wrote:
>
> > How did you get the MatNest?
> >
> > What you describe is the right approach, but proper preallocation is
> messy in the general case. There is a partial implementation in petsc-dev,
> but it needs to be finished/debugged.
>
>    Thomas,
>
>       We should have a reasonable MatConvert_MatNest_MATAIJ. Perhaps you
> can look at what Jed has started in petsc-dev and improve it enough to suit
> your purpose.
>
There is this commit from last summer:
http://petsc.cs.iit.edu/petsc/petsc-dev/rev/5289fd118d48
It's untested, but I pushed it back then hoping someone could beat it into
shape.
I also have a patch that eliminates ISAllGather() in favor of PetscSF
operations, but that's even
less tested (if such a thing is possible).  I can clean it up and push, if
somebody wants to play with it.

Dmitry.

>
>     Barry
>
>
>
>
> >
> > On Oct 10, 2012 8:45 AM, "Thomas Witkowski" <
> thomas.witkowski at tu-dresden.de> wrote:
> > You are right. But what if a have a MATNEST and i must convert it to
> MATAIJ. Any better way than what I wrote in my first mail?
> >
> > Thomas
> >
> > Am 10.10.2012 15:33, schrieb Jed Brown:
> >> The best way, by far, is to assemble the matrix as AIJ if that is what
> you want to use. See MatGetLocalSubMatrix and SNES ex28. Using MATNEST
> should he a run-time choice.
> >>
> >> On Oct 10, 2012 8:30 AM, "Thomas Witkowski" <
> thomas.witkowski at tu-dresden.de> wrote:
> >> Is there a fast way to convert a MatNest to a MatAIJ? I tried to use
> MatConvert, but it seems that this convert is not yet implemented. As I
> need this functionality to factorize the nested matrix with MUMPS, is there
> any better way than to extract the nested matrices, iterate over all rows
> of these matrices and use MatSetValues to add them again to the final
> MatAIJ?
> >>
> >> Thomas
> >
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20121010/20e53519/attachment.html>


More information about the petsc-users mailing list