--- Paolo Giannozzi <giannozz at nest.sns.it> wrote: > On Friday 12 May 2006 14:09, Gabriele Sclauzero wrote: > > > from read_file : error # 1 > > problem reading file /local_scratch/sclauzer/tmp/prefix.save
> I am quite sure this has been fixed in the meantime. Actually, I encountered the same issue too with one of the very recent CVS versions (2 days ago). While PW itself proceeds fine from the distributed *.wfc files, the post-processing codes crash (pw2wannier90 for me). I tracked it a bit to the subroutine read_file.f90, where things crash in these calls CALL pw_readfile( 'reset', ierr ) CALL pw_readfile( 'dim', ierr ) even when the *.save directory itself is perfectly visible. Like when running on 2 cpus of a dual box. I ran under x86-64 Linux with mpich 1.2.7 and ifort 9.0 > Note that > > - there is no guarantee that the processor reading the data file is > the same that wrote it in a previous run. One has no control on > which processor holds which MPI process. Some queueing systems read the cpu list from a node file ( $PBS_NODEFILE under PBS ), and so the chance that the order is the same is high. > - you need to set wf_collect to .true. in order to save wavefunctions > into a single file But then is post-processing smart enough to read the collected file and distribute it? I could not verify this due to crashes. > - if I remember correctly, pseudopotentials are still read by every > processor. There is no special reason for this, other than the time > and boredom needed to implement it in the same way as the rest > of I/O (i.e. read on one processor, broadcast all variables to all > other processors) But these come from a separate pseudo-directory, do they not? Kostya __________________________________________________ Do You Yahoo!? Tired of spam? Yahoo! Mail has the best spam protection around http://mail.yahoo.com
