<div dir="ltr"><br><div class="gmail_extra"><br><br><div class="gmail_quote">On Fri, May 3, 2013 at 10:20 PM, Jed Brown <span dir="ltr"><<a href="mailto:jedbrown@mcs.anl.gov" target="_blank">jedbrown@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">Dharmendar Reddy <<a href="mailto:dharmareddy84@gmail.com">dharmareddy84@gmail.com</a>> writes:<br>
<br>
> Hello,<br>
> I see an interesting behavior when i call EPSSolve in a loop. Can<br>
> you help me figure out whats going on ?<br>
><br>
> I have a setup like this to solve a A x = lambda B x (Generalized<br>
> Hermitian problem)<br>
><br>
> type eigenSolver_t<br>
> EPS :: eps<br>
><br>
> end type eigenSolver_t<br>
><br>
> the type has bound procedure which calls EPSSetOperators and EPSSolve when<br>
> eigenSolver%solve() is called<br>
><br>
> Now i run a for loop<br>
><br>
> do ic=1,111<br>
> call eigenSolver(ic)%solve()<br>
> end do<br>
><br>
> I print the time for each solve. The operators A,B =A1,B1 for ic=1 to 50<br>
> A2,B2 for ic=51 to 80 and A3,B3 for ic=81 to 111<br>
><br>
> Now i see that time per solve per ic is almost constant when i use<br>
> -eps_type lapack.<br>
<br>
</div>How do you figure? It looks like it increases by more than an order of<br>
magnitude.<br>
<br></blockquote><div> You are right, The time per solve increases in both case, see the attached plot. <br><br></div><div> I have attached the stage wise log_summary . <br></div><div> <br></div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
Always send -log_summary output when asking about performance. In this<br>
case, it would be nice to use a different stage to log each solve.<br>
<div class="im"><br>
> But for, defualt solver, time per solve seem to increase with increasing<br>
> ic. Have look at the attached timing information.<br>
><br>
> Also, Time per solve using lapack is lower than any of the iterative<br>
> solvers i have tried. Problem size is about 100 x 100, operators are<br>
> tri-diagonal.<br>
<br>
</div>I'm not surprised. That problem is tiny.<br>
</blockquote></div><br><br clear="all"><br>-- <br>-----------------------------------------------------<br>Dharmendar Reddy Palle<br>Graduate Student<br>Microelectronics Research center,<br>University of Texas at Austin,<br>
10100 Burnet Road, Bldg. 160<br>MER 2.608F, TX 78758-4445<br>e-mail: <a href="mailto:dharmareddy84@gmail.com" target="_blank">dharmareddy84@gmail.com</a><br>Phone: +1-512-350-9082<br>United States of America.<br>Homepage: <a href="https://webspace.utexas.edu/~dpr342" target="_blank">https://webspace.utexas.edu/~dpr342</a><br>
</div></div>