[petsc-dev] Generality of VecScatter

[Jed Brown]

On Thu, Nov 24, 2011 at 13:55, Matthew Knepley <knepley at gmail.com> wrote:

> No, I agree with Mark here. I originally wrote this way and had to rewrite
> EVERYTHING later on to get this information.


You were using different primitives. With this model, it is one
non-synchronizing gather if you want it. In a high-level interface, it
might be cached the first time you request it. But the presence of this
data is only there when it is needed by application logic, not to define
the communication. (The two-sided graph might be created temporarily if you
are setting up a neighborhood collective to perform a related operation.
You can decide whether a given communication pattern is going to be used
enough times or in a way that warrants this setup).
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-dev/attachments/20111124/6217869c/attachment.html>