<html><head><style type='text/css'>p { margin: 0; }</style></head><body><div style='font-family: Arial; font-size: 12pt; color: #000000'>I understand the issue of some uses of blas being too small to justify shipping to the gpu.<br>So I was thinking maybe petsc could choose at runtime based on problem size whether<br>to use a cpu implementation of the blas or a gpu implementation of the blas. Seems like<br>that should be feasible but more complex than just having a petsc interface to substitute<br>use of cublas for everything.<br><br>Yes, I will back off tonight and try to work with a petsc example problem. I had to add<br>an additional library to CHOLMOD.py to satisfy an unstatisfied external. I think it was<br>named libsuitesparseconfig.a and was needed to resolve the SuiteSparse_time symbol.<br>Would probably be good to have a SuiteSparse.py module. I'll send more info if I can<br>reproduce the problem with a petsc example.<br><br>Thanks,<br><br>Dave<br><br><hr id="zwchr"><b>From: </b>"Jed Brown" <jedbrown@mcs.anl.gov><br><b>To: </b>"Dave Nystrom" <dnystrom1@comcast.net><br><b>Cc: </b>"For users of the development version of PETSc" <petsc-dev@mcs.anl.gov><br><b>Sent: </b>Wednesday, June 13, 2012 7:41:05 AM<br><b>Subject: </b>Re: [petsc-dev] SuiteSparse 4.0<br><br><p>You can't use cublas everywhere because it's better to do small sizes on the CPU and different threads need to be able to operate independently (without multiple devices).</p>
<p>Yes, try a PETSc example before your own code. Did petsc build without any changes to the cholmod interface? ALWAYS send the error message/stack trace---even if we don't have an answer, it usually tells us how difficult the problem is likely to be to fix.</p>
<div class="gmail_quote">On Jun 13, 2012 8:09 AM, "Dave Nystrom" <<a href="mailto:dnystrom1@comcast.net" target="_blank">dnystrom1@comcast.net</a>> wrote:<br><blockquote class="gmail_quote" style="margin: 0pt 0pt 0pt 0.8ex; border-left: 1px solid rgb(204, 204, 204); padding-left: 1ex;">
Well, I tried it last night but without success. I got a memory corruption<br>
error when I tried running my app. I suppose I should try first to get it<br>
working with a petsc example. I'm building SuiteSparse but only using<br>
cholmod out of that package. Several of the SuiteSparse packages use blas<br>
but cholmod is the only one so far that has the option to use cublas. Tim<br>
gets about a 10x speedup over the single core result when using cublas on an<br>
example problem that spends a lot of time in blas. So seems worth trying.<br>
Anyway, I think I must be doing something wrong so far. But for the same<br>
problem, Tim gets about a 20+ percent speedup over using multi-threaded Goto<br>
blas.<br>
<br>
Probably a better solution for petsc would be support for using cublas for<br>
all of the petsc blas needs. Why should just one petsc blas client have<br>
access to cublas? But not sure how much work is involved for that - I see it<br>
is on the petsc ToDo list. I will try again tonight but would welcome advice<br>
or experiences from anyone else who has tried the new cholmod.<br>
<br>
Dave<br>
<br>
Jed Brown writes:<br>
> Nope, why don't you try it and send us a patch if you get it working.<br>
><br>
> On Wed, Jun 13, 2012 at 12:49 AM, Dave Nystrom <<a href="mailto:dnystrom1@comcast.net" target="_blank">dnystrom1@comcast.net</a>>wrote:<br>
><br>
> > Has anyone tried building petsc-dev to use cholmod-2.0 which is part of<br>
> > SuiteSparse-4.0 with the cholmod support for using cublas enabled? I am<br>
> > interested in trying the cublas support for cholmod to see how it compares<br>
> > with mkl and goto for my solves.<br>
> ><br>
> > Thanks,<br>
> ><br>
> > Dave<br>
> ><br>
</blockquote></div>
</div></body></html>