[petsc-dev] Generality of VecScatter

[Jed Brown]

On Thu, Nov 24, 2011 at 13:28, Barry Smith <bsmith at mcs.anl.gov> wrote:

> It sounds to me like you are proposing deprecating point to point in MPI
> and using the one-sided operations as the primitives? I  think this is
> still wrong because you still have no ability to combine different
> one-sided to do proper scheduling,
>

Use neighborhood collectives for that.

I think point-to-point is still occasionally useful for setup and probably
for certain kinds of unstructured communication, but usually just used once.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-dev/attachments/20111124/5fce90d5/attachment.html>