On Thu, Apr 5, 2012 at 10:19 AM, Barry Smith <span dir="ltr"><<a href="mailto:bsmith@mcs.anl.gov">bsmith@mcs.anl.gov</a>></span> wrote:<br><div class="gmail_quote"><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
<div class="HOEnZb"><div class="h5"><br>
On Apr 5, 2012, at 9:54 AM, Jack Poulson wrote:<br>
<br>
> On Thu, Apr 5, 2012 at 9:20 AM, Jack Poulson <<a href="mailto:jack.poulson@gmail.com">jack.poulson@gmail.com</a>> wrote:<br>
> On Thu, Apr 5, 2012 at 8:02 AM, Barry Smith <<a href="mailto:bsmith@mcs.anl.gov">bsmith@mcs.anl.gov</a>> wrote:<br>
><br>
> Parallel dense we know very little about and cannot write those routines, sadly there are no decently supported parallel dense matrix general purpose libraries out there that we can use (and no Scalapack, plapack and elemental do not count as decent AND supported) so it is unlikely WE will write the MPIDENSE versions of these routines. Though if someone else writes them we would be happy to include them. So basically for parallel dense I have no suggestions.<br>
><br>
><br>
> If you let me know what your major complaints are, I'll look into them.<br>
><br>
> Jack<br>
><br>
> Forwarding to Jed and Matt; seriously, let me know what your complaints are. If I can, I will fix them.<br>
><br>
> Jack<br>
<br>
</div></div> Jack,<br>
<br>
I apologize, I was too quick to dismiss elemental. We went to the effort to do the MPIDENSE interface for PLAPACK and that ended up being a complete waste and left a bad taste in my mouth since I consider that wasted time when we could have done something else.<br>
<br></blockquote><div><br></div><div>No offense taken; it is understandable that PLAPACK did not meet PETSc's rigorous development standards, as it was closer to a research code. With that said, the reason I wrote Elemental is that I wanted a robust linear algebra package that made use of PLAPACK's insights (ala exposing matrix distributions as objects).</div>
<div><br></div><div>Given that you had a bad experience with PLAPACK, perhaps the best assurance I can give you is that I am actively dogfooding Elemental in my other projects (see <a href="http://bitbucket.org/poulson/clique">http://bitbucket.org/poulson/clique</a> and <a href="http://bitbucket.org/poulson/psp">http://bitbucket.org/poulson/psp</a>). In addition, van de Geijn has a student working on extending Elemental's matrix-matrix multiplication code to more sophisticated distribution schemes. I _highly_ doubt that you will find a bug in Elemental's matrix-matrix multiplication, but, if you do, I can assure you that I will fix it within a few hours. The code is quite simple:</div>
<div><a href="http://code.google.com/p/elemental/source/browse/#hg%2Finclude%2Felemental%2Fbasic%2Flevel3%2FGemm">http://code.google.com/p/elemental/source/browse/#hg%2Finclude%2Felemental%2Fbasic%2Flevel3%2FGemm</a></div>
<div> </div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
So we could try again with Elemental but its a question of having a competent person to do the interface efficiently. Perhaps you have a student or someone you know that knows elemental we could hire for a short time to do an interface and see how it goes?<br>
<span class="HOEnZb"><font color="#888888"><br></font></span></blockquote><div><br></div><div>Unfortunately I am still just a lowly Ph.D. student (but hopefully that will change in December). I would be happy to try to help, and _may_ find the time to write the interface myself, but, until graduation, I can't promise more than maintenance. I can talk with a few folks that might be interested, but I would think that this should only take a few days of work, not anything that warranted an appointment.</div>
<div> </div></div>Jack