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

[Jed Brown]

Barry Smith <bsmith at mcs.anl.gov> writes:
>    Presumably his "local" calculation using the ghosted parts of his "vectors" is simpler with PETSCMPIDENSE?

That may be, but the square dense MatMult will eventually be expensive
enough to justify redistribution and then putting the result back.
(This is different from the MatTransposeMatMult that Hong is looking at,
for which there is no communication except the unavoidable
MPI_Reduce_scatter.)
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 818 bytes
Desc: not available
URL: <http://lists.mcs.anl.gov/pipermail/petsc-dev/attachments/20151016/65297772/attachment.sig>