The ABINIT Users Mailing List ( CLOSED )

["Matteo Giantomassi" <Matteo.Giantomassi@uclouvain.be>]

> Hi there,
>
> I've upgraded my abinit 5.6.3 to 5.7.2, but when carrying out GW
> calculation, I
> found the message in log file as below. It seems that GWPARA=2 didn't work
> any
> more, because it didn't run in parallel on bands.

  Dear Aaron,

  What do you mean with "it doesn't work anymore".
  It it crashing? Did you run the automatic tests
  for the parallel version?

  Just go to tests/, create a directory named paral/
  (if it's not already existing) and run the following script

  tests/Scripts/wrap-parallel-tests.sh generic seqpar > log_para 2> err_para

  The gwpara=2 algorithm is tested in TN2, TN4, so you can easily check
  whether the code runs correctly on your architecture.

  The fact that you are getting warnings in the log files saying
  that the number of processors is larger that nkpt is not an issue.
  For GW calculations, the MPI distribution of the tasks is redefined inside
  the GW part such that both the CPU load and the memory consumption are
  distributed over bands.
  This kind of distribution strongly differs from the MPI approach used
  in GS calculations. Due to technical reasons, it's not yet possible
  to perform the setup of the parallelism at the beginning of the run.
  The GW part defines its own distribution at each dataset, so the initial
  warnings related to the k-point distribution are meaningless.

  For gwpara = 2, the only limiting factor is the number of bands (nband).
  The number of processor must be <= nband and the load distribution is
  optimal when the number of CPU's divides nbands,

  Best Regards
  Matteo Giantomassi