[CP2K:3240] status of hybrid OpenMP+MPI version

Axel akoh... at gmail.com
Tue May 17 16:44:02 UTC 2011


perhaps the following helps to track down where things go wrong.

i just updated cp2k to the latest cvs and ran with 2 MPI tasks the 
H2O-32.inp 
benchmark file. output is identical between OMP_NUM_THREADS=1 and
OMP_NUM_THREADS=2 until i reach this output.


with one thread:
  ----------------------------------- OT 
---------------------------------------

  Step     Update method      Time    Convergence         Total energy   
 Change
  
------------------------------------------------------------------------------

  Electronic density on regular grids:       -255.9999972509       
 0.0000027491
  Core density on regular grids:              255.9999999997       
-0.0000000003
  Total charge density on r-space grids:        0.0000027487
  Total charge density g-space grids:           0.0000027487


  Core Hamiltonian energy:                                       
 424.9326981392
  Hartree energy:                                                 
577.5530255740
  Exchange-correlation energy:                                   
-129.1348019384
  Coulomb (electron-electron) energy:                             
261.5717237448
        Maximum deviation from MO S-orthonormality                   
 0.4229E-12
        Minimum/Maximum MO magnitude              0.3562E+00         
 0.3521E+01
     1 OT DIIS     0.15E+00    1.4     0.01909729      -529.3015728815 
-5.29E+02


with two threads:

  ----------------------------------- OT 
---------------------------------------

  Step     Update method      Time    Convergence         Total energy   
 Change
  
------------------------------------------------------------------------------

  Electronic density on regular grids:       -140.0786039192     
 115.9213960808
  Core density on regular grids:              255.9999999997       
-0.0000000003
  Total charge density on r-space grids:      115.9213960805
  Total charge density g-space grids:         -82.9977475763


  Core Hamiltonian energy:                                       
 424.9326981392
  Hartree energy:                                               
 1368.0021602823
  Exchange-correlation energy:                                   
-391.6132000948
  Coulomb (electron-electron) energy:                             
197.3468594016
        Maximum deviation from MO S-orthonormality                   
 0.4229E-12
        Minimum/Maximum MO magnitude              0.3562E+00         
 0.3521E+01
     1 OT DIIS     0.15E+00    1.0     0.10264462        -1.3308363296 
-1.33E+00


the funny thing is that 140.0786039192+115.9213960808 = 256.0000000000
so it looks as if some information is accumulated in the wrong way.

can anybody lend me a hand in locating where this could happen?

as a reminder. this only goes wrong if using more than one thread and
using more than one MPI task.

thanks,
    axel.





-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20110517/a00ba586/attachment.htm>


More information about the CP2K-user mailing list