[Pw_forum] Problems in running check-pw.x.j

[Lorenzo Paulatto]

On Tue, 27 Mar 2012 10:41:46 +0200, Carlo Nervi <carlo.nervi at unito.it>  
wrote:
> By using "mpirun -n 48" it crash at dipole calculation. Too fast for 48
> cores?

Some test are too mall to be run with so many processors. I.e. most tests  
only have one or a few k-points, so they can only be run with one or a few  
pools.

In particular, the dipole test a FFT grid with only 38 points along the Z  
axis, hence it cannot parallelized on more then 38 processors per pool  
(without using bands, taskground or openmp). You can see it at the  
beginning:

      sticks:   dense  smooth     PW     G-vecs:    dense   smooth      PW
      Min           7       7      1                  728      728      22
      Max           8       8      2                  816      816     146
      Sum         303     303     75                29755    29755    3661
      Tot         152     152     38

You see that the "tot" value for PW says 38. You can also notice that  
dimension constraints of the DFFT algorithm make you waste a lot of cpu  
time (every cpu doubles the size of its FFT grid along z!).

Running with 39 cpus you do actually get this error:
  %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
      task #        38
      from n_plane_waves : error #         1
      No plane waves found: running on too many processors?
  %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
for some reason this message does not get printed to stdout, but you can  
get it from the CRASH file.

bests

-- 
Lorenzo Paulatto IdR CNRS @ IMPMC/UPMC - Paris 6
phone: +33 (0)1 44275 084 / skype: paulatz
www:   http://www.impmc.upmc.fr/~paulatto/
mail:  23-24/4é14 Boîte courrier 115, 4 place Jussieu 75252 Paris Cédex 05