[Pw_forum] newbie update

Axel Kohlmeyer akohlmey at cmm.chem.upenn.edu
Wed Mar 4 21:06:37 CET 2009


On Wed, 4 Mar 2009, Lex Kemper wrote:

LK> Out of curiosity, what behaviour do you see from MKL that it doesn't 
LK> obey OMP_NUM_THREADS? I ask, because that's what I'm using, and would 
LK> like to make sure it's behaving right.

lex,

it doubt it is mkl that is to blame, but rather the mpirun
script/wrapper not exporting environment variables to all tasks,
even on remote nodes. if you have a good sysadmin running your
machine her or she will have already set OMP_NUM_THREADS=1
in the default environment.

with openmpi, for example, the environment is not exported
unless explicitly set it via a flag.

linking the sequential version of mkl (see mkl docs for)
details, is the solution for paranoid people (like me).

cheers,
   axel.

LK> 
LK> Thanks,
LK> 
LK> Lex Kemper
LK> University of Florida
LK> 
LK> Axel Kohlmeyer wrote:
LK> > On Wed, 4 Mar 2009, JR Schmidt wrote:
LK> > 
LK> > JRS> I had problems with MKL not obeyig OMP_NUM_THREADS.  I found it was 
LK> > JRS> easier to simply link with the MKL serial (non-threaded) library, since 
LK> > JRS> that is what you want anyway.
LK> > 
LK> > yes. but in carlo's case it _does_ work correctly.
LK> > 
LK> > axel.
LK> > 
LK> > JRS> > I tried to run the example01 with the following settings:
LK> > JRS> >
LK> > JRS> > OMP_NUM_THREADS=1
LK> > JRS> > mpiexec -n 8
LK> > JRS> >
LK> > JRS> > The example is executed, but with a very long time. All 8
LK> > JRS> > CPUs are running
LK> > JRS> >   
LK> > JRS> 
LK> > 
LK> _______________________________________________
LK> Pw_forum mailing list
LK> Pw_forum at pwscf.org
LK> http://www.democritos.it/mailman/listinfo/pw_forum
LK> 

-- 
=======================================================================
Axel Kohlmeyer   akohlmey at cmm.chem.upenn.edu   http://www.cmm.upenn.edu
   Center for Molecular Modeling   --   University of Pennsylvania
Department of Chemistry, 231 S.34th Street, Philadelphia, PA 19104-6323
tel: 1-215-898-1582,  fax: 1-215-573-6233,  office-tel: 1-215-898-5425
=======================================================================
If you make something idiot-proof, the universe creates a better idiot.



More information about the users mailing list