[Pw_forum] failure with npool

Eduardo Ariel Menendez Proupin eariel99 at gmail.com
Sat Aug 1 15:08:48 CEST 2009


Dear Paolo,
Here is the answer to your suggestion a few days ago, concerning the
use of versions of  zhpev_drv.f90

echo "4.0.4/bin-hpmpi/pw.x"
/opt/hpmpi/bin/mpirun -np ${NCPUS} -hostfile nodelist /discolocal/app/Espresso/e
spresso-4.0.4/bin-hpmpi/pw.x -in ${inputfile} >> ${outputfile}
     PWSCF        : 34m15.70s CPU time,    36m14.28s wall time


echo "bin-speed/pw-1.4.x"
/opt/hpmpi/bin/mpirun -np ${NCPUS} -hostfile nodelist  /discolocal/emenendez/Che
mUtils/Espresso/espresso-4.1/bin-speed/pw-1.3.x -in ${inputfile} >> ${outputfile
}
     PWSCF        : 35m 4.30s CPU time,    40m17.14s wall time

echo "bin-speed/pw-1.3.x"
/opt/hpmpi/bin/mpirun -np ${NCPUS} -hostfile nodelist  /discolocal/emenendez/Che
mUtils/Espresso/espresso-4.1/bin-speed/pw-1.3.x -in ${inputfile} >> ${outputfile
}
     PWSCF        : 32m15.69s CPU time,    37m13.17s wall time

echo "bin-speed/pw-1.2.x"
/opt/hpmpi/bin/mpirun -np ${NCPUS} -hostfile nodelist  /discolocal/emenendez/Che
mUtils/Espresso/espresso-4.1/bin-speed/pw-1.2.x -in ${inputfile} >> ${outputfile
}
     PWSCF        : 30m13.87s CPU time,    35m22.95s wall time


>< cdiaghg : 605.99s CPU ( 1419 calls, 0.427 s avg)
>
>>* cdiaghg : 893.34s CPU ( 1593 calls, 0.561 s avg)
*>
>here it is: subspace diagonalization. Are you sure that the two
>codes were compiled in the same way? and that you are running
>under the same conditions? If so, please get the previous version
>of Modules/zhpev_drv.f90:
>http://qe-forge.org/cgi-bin/cvstrac/q-e/fileview?f=espresso/Modules/zhpev_drv.f90&v=1.4 <http://qe-forge.org/cgi-bin/cvstrac/q-e/fileview?f=espresso/Modules/zhpev_drv.f90&v=1.4>
>then 1.3 and 1.2, and please let us know if and when you notice a
>difference.  If you get errors related to nonexistent routines
>'numroc' and 'indxl2g', just remove the line
>      INTEGER     :: numroc, INDXL2G
>Some changes that were needed to ensure statibility on weird
>machines may have slowed down the general performances. If
>you have scalapack, it mighy be interesting to try it.







-- 
Eduardo Menendez
Departamento de Fisica
Facultad de Ciencias
Universidad de Chile
Phone: (56)(2)9787439
URL: http://fisica.ciencias.uchile.cl/~emenendez
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20090801/e95cdf4a/attachment.html>


More information about the users mailing list