> On Aug 4, 2017, at 2:52 PM, Xiangdong wrote:
>
> Hello everyone,
>
> I have a quick questions on the MatMPIAIJSetPreallocation. If we do not pass
> the exact number of nonzeros for o_nnz and d_nnz, (for example,
> MAT_INFO_NZ_ALLOCATED is 50% more than
Coming back to this after a while. It would still be good if I
could provide a callback for user-defined adjacency. Does
https://bitbucket.org/petsc/petsc/pull-requests/690/plex-support-user-defined-adjacencies-via/diff
Hello everyone,
I have a quick questions on the MatMPIAIJSetPreallocation. If we do not
pass the exact number of nonzeros for o_nnz and d_nnz, (for example,
MAT_INFO_NZ_ALLOCATED is 50% more than MAT_INFO_NZ_USED), will the
performance of MatMult be affected? What effect will these allocated but
On Fri, Aug 4, 2017 at 9:37 AM, Lawrence Mitchell <
lawrence.mitch...@imperial.ac.uk> wrote:
>
> > On 6 Jun 2017, at 09:01, Lawrence Mitchell ac.uk> wrote:
> >
> >> If you do really want to prune them, then I guess overriding the
> DMPlexGetAdjacency() as you
On Thu, Aug 3, 2017 at 10:52 PM, Adrian Croucher
wrote:
> Would it also work if I just didn't use DMPlexGetHybridBounds() to
> differentiate between partition ghost cells and boundary ghost cells, but
> instead checked the value of the "ghost" DMLabel?
>
Yes.
Matt
> On 6 Jun 2017, at 09:01, Lawrence Mitchell
> wrote:
>
>> If you do really want to prune them, then I guess overriding the
>> DMPlexGetAdjacency() as you propose is probably the best way. I would
>> be willing to put it in. Please send me a reminder email