<div class="gmail_quote">On Mon, Aug 29, 2011 at 11:10, John Fettig <span dir="ltr"><<a href="mailto:john.fettig@gmail.com">john.fettig@gmail.com</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex;">
<div id=":2gw">I think the way to do it is similar to the way it is done in the<br>
function ML_Gen_MultiLevelHierarchy_UsingSmoothedAggr_ReuseExistingAgg,<br>
although that function rebuilds R, A, and P. So we just have to<br>
rebuild A. Here's some code that does that, it seems to work as well<br>
as the previous code I sent that only worked in serial (this works in<br>
parallel too). </div></blockquote><div><br></div><div>Would you be willing to make this into a Mercurial patch against petsc-dev? Just commit your changes locally, hg export them, and send us the patch.</div><div> </div>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex;"><div id=":2gw"> Could this be made to be the default if you set the<br>
flag "SAME_NONZERO_PATTERN"? Then if you want to rebuild the<br>
preconditioner, call with "DIFFERENT_NONZERO_PATTERN".</div></blockquote></div><br><div>Perhaps it can be made the default, but there should be a runtime option to recompute even when SAME_NONZERO_PATTERN is set (because the nonzero pattern of the operator, which is what that flag is supposed to indicate, did not change).</div>