<p>snes_mf_operator has one MFFD (almost free), your assembled matrix, and the factored matrix created by the preconditioner.</p>
<div class="gmail_quote">On Mar 1, 2012 10:16 AM, "Derek Gaston" <<a href="mailto:friedmud@gmail.com">friedmud@gmail.com</a>> wrote:<br type="attribution"><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
In trying to track down where all of our memory usage is coming from I noticed that PETSc is saying that more matrices than I expect are getting created. For instance, with -snes_mf I would expect to see just one matrix (we always create one, even with -snes_mf)... but instead I'm seeing 2 (from -log_summary):<div>
<br></div><div><div> Matrix 2 2 1634161540 0</div></div><div><br></div><div>Then when using -snes_mf_operator with ilu as the preconditioner I'm actually seeing _3_!</div><div><br>
</div>
<div> Matrix 3 3 3080557036 0</div><div><br></div><div>As you can see, our matrices are fairly heavy (cubic hermites in 3d have a heavy sparsity pattern) so it would be awesome to keep the number of them down. I've double checked that we're not calling MatCreate more than once in our code. So is this expected behavior or am I doing something wrong somewhere else? What's the best way to track this down... maybe print a stack trace inside MatCreate?</div>
<div><br></div><div>Particulars:</div><div>PETSc 3.1-p8 (yes yes old... ;-)</div><div>Using it through libMesh</div><div><br></div><div>Thanks,</div><div>Derek</div>
</blockquote></div>