[petsc-dev] [petsc-users] MatGetSubMatrices or VecCreateGhost

[Barry Smith]

> On Oct 16, 2015, at 1:29 PM, Jed Brown <jed at jedbrown.org> wrote:
> 
> Barry Smith <bsmith at mcs.anl.gov> writes:
>>   We don't have a parallel 
> 
> Elemental makes sense here; the 2D distribution will scale much better.

  For the multiple sure
> 
> Note that MatTransposeMatMult_MPIDense_MPIDense is fairly trivial and
> very useful for tall-skinny matrices, which is the case in which the
> MPIDense distribution ([VC,*] in Elemental vernacular) actually makes
> sense.

   Presumably his "local" calculation using the ghosted parts of his "vectors" is simpler with PETSCMPIDENSE?

  Barry