<div class="gmail_quote">On Thu, Nov 25, 2010 at 05:08, Barry Smith <span dir="ltr"><<a href="mailto:bsmith@mcs.anl.gov">bsmith@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 class="im">> 2. Not necessarily make copies of everything within FieldSplit (I think all users of FieldSplit complain that it costs so much memory at present).<br>
<br>
</div> They think it takes too much, but if we had real profiling of where the memory is used they would see that it is actually trivial.</blockquote></div><br><div>Calling MatGetSubMatrix to break an assembled matrix into pieces triples the amount of memory for column indices and values, plus has the non-scalable MPI_Allgather. You routinely complain about other packages that don't properly support 64-bit indices. If you use FieldSplit with a globally assembled matrix, that will never be a problem unless you have 16 GB per process to dedicate solely to storing redundant indices. It also loses block structure and symmetry, why even bother with (S)BAIJ if you can't use it for coupled problems?</div>
<div><br></div><div>Jed</div>