[CP2K:6462] Re: Memory problem when using ELPA

[Dominik 'Rathann' Mierzejewski]

Hello, Marcella and Jean-Marie,

On Monday, 04 May 2015 at 16:07, Marcella Iannuzzi wrote:
> On Monday, May 4, 2015 at 3:06:26 PM UTC+2, Jean-Marie wrote:
> > I have recently started to use the ELPA library and I was very happy when 
> > I obtained a nice increase of the calculation speed (~10-20%), even when 
> > using as few as 20 processors (the number of cores per node on the 
> > supercalculator I am using). However, I have also noticed that when I run a 
> > geometry optimization, the memory requirement increases at each geometry 
> > step, until the calculation finally crashes! Has anyone else encountered 
> > such problem and managed to fix it?
> > I am currently using cp2k-2.7 (rel. 14892 but I had the same problem with 
> > a previous release) and ELPA-2013-11 (but I had the same problem with other 
> > versions) compiled with gcc-4.9.2 (but I had the same problem with 4.6.0), 
> > Intel mkl-15.0 and Intel MPI 5.0.1.
> >
> > I thank in advance anyone who can help...
> 
> yes we know the problem. It is unfortunately a memory leak appearing in the 
> latest versions of library and we did not solve it yet.
> It is possibly related to the modifications introduced to enable  OMP. 
> However, we could  compile the latest cp2k with the ELPA-2013-08 with gcc 
> 4.8.2 obtaining a working executable, without leaks.
> Have you tried this combination?

ELPA can be compiled with or without OpenMP. Have you tried either
version to narrow down the source of the leaks? Also, are you using
ELPA-2015.02.002 release?

Lastly, have you reported this issue to ELPA developers? They are
quite responsive to bug reports.

Regards,
Dominik
-- 
Fedora http://fedoraproject.org/wiki/User:Rathann
RPMFusion http://rpmfusion.org
"Faith manages."
        -- Delenn to Lennier in Babylon 5:"Confessions and Lamentations"