<div class="gmail_quote">On Mon, Oct 18, 2010 at 15:23, Barry Smith <span dir="ltr"><<a href="mailto:bsmith@mcs.anl.gov">bsmith@mcs.anl.gov</a>></span> wrote:<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex;">
You are both wrong on this issue. In 1995 most people thought the flexibility was totally unneeded and actually harmful for matrices; they were wrong. In five years you will realize that you were wrong on the issue for DMs in 2010. Just wait, this is going to be damn powerful.</blockquote>
<div><br></div><div>To elucidate this power, can you explain what benefit *user code* obtains from having a DM while still calling DMGetLocalInfo and rolling loops over the structured local ranges?</div><div><br></div><div>
I fully understand the need for DM to be flexible from the solver perspective, in its uses for multigrid and such. But that's what the interface in _DMOps is for. Do you really envision user code that provides a discretization on an arbitrary DM (same code works with a DA, Sieve-mesh, particle DM)? What might that look like?</div>
<div><br></div><div>Jed</div></div>