[Pw_forum] Re: Re: error while parallel running

Lin Zhuang lzhuang at whu.edu.cn
Sun Apr 16 08:47:11 CEST 2006


Hi, Eyvaz:

Thank you for your suggestion.

I had succesfully run the input file with serial version pw.x and parallel
version pw.x if only one CPU had been specified, i.e.,
  $ mpirun c0 pw.x -in inputfile > outputfile
But when I tried to run the same input file with more than one CPU, like
  $ mpirun c0-1 pw.x -in inputfile > outputfile   # CPU c0-1 are on the same
computer
or
  $ mpirun -np 4 pw.x -in inputfile > outputfile  # 4 CPUs cross 2 computers
I got error like
%%%%%%%%%%%%%%%%%%%%%%%%%%%
from read_namelists : error #    17
  reading namelist control
%%%%%%%%%%%%%%%%%%%%%%%%%%%

Acturally, the input file was just Example 01 in espresso 3.0,
si.scf.david.in, and I just modified those pseudo/scratch directories.

By the way, the MPI environment is LAM/MPI 7.1.2, and was complied with
Intel icc/ifort 9.0, the same compilers as I used for espresso 3.0, and was
fully tested with the test suit of LAM/MPI. I had other ab initio MD
software, e.g., Dacapo, run successfully on the same computers and MPI
environment.

I used to add wt_collect = .TRUE. in &control in the input file, although
above error message disappeared, but pw.x still crashed with a message like
invalid argument (rank 0, MPI_COMM_WORLD).

I totally lost here. Any comment would be highly appreciated.

best,
Lin
-------------- next part --------------
An HTML attachment was scrubbed...
URL: /pipermail/attachments/20060416/05ff7ac2/attachment.htm 


More information about the Pw_forum mailing list