<div class="gmail_quote">On Mon, May 16, 2011 at 13:52, Dmitry Karpeev <span dir="ltr"><<a href="mailto:karpeev@mcs.anl.gov">karpeev@mcs.anl.gov</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex;">
<div id=":2w0">True, although MatSetValuesLocal (not the default implementation in<br>
the interface)<br>
is a way to extend the mechanism that relies on ISLocalToGlobalMapping.<br>
This can include whatever semantics are appropriate to handle the overlaps, etc.<br>
I think this is a reasonable splitting of functionality between<br>
PetscLayout and MatSetValuesLocal().</div></blockquote></div><br><div>Yes, I agree that the MatGetLocalSubMatrix()/MatSetValuesLocal() paradigm is a good way to manage overlap. With more general transformations, we could consider extending the interface (or having a new interface that is similar in spirit).</div>