On Tue, Dec 7, 2010 at 22:52, Dmitry Karpeev <karpeev at mcs.anl.gov> wrote:
> Actually, I don't think this would require DM to know that much than it > does now: > I don't see how this is DM-generic, it seems to be specialized to certain DMs. Yes, sometimes I would like to have "one DM to rule them all", but I think there will always be value in specialization, for performance or for special applications. Maybe, if we had a proven system for traversal that we wanted to promote, it could be moved up from DMOne to DM, but I'm skeptical about this. DMDA has GlobalToLocal as well as VecGetArray, which provides both > "restriction to local" (a coarse-granularity, heavy-weight operation), and > "restrict to (a bunch of) stencil(s)" (a fine granularity lightweight > "scatter"). > This encapsulates your notions of a "local space" and an "extended space", > which, clearly, require > different access patterns. > A distinction is that we need iterators on the expanded space to be non-collective. Jed -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.mcs.anl.gov/pipermail/petsc-dev/attachments/20101207/79634404/attachment.html>