[MITgcm-support] MPI_COMM_WORLD question

chris hill cnh at mit.edu
Mon Jan 16 08:03:56 EST 2006


gianmaria & sandro,

  The exch_jam bits are not used unless you have some obscure hardware 
that you don't have - so don't worry about those.

  The eedie routine is meant to be that way. The COMM_WORLD barrier
is there so that you don't drop through to Finalize without first 
syncing everyone up (Finalize is a global shut down that doesn't take a 
communicator). This can be useful with some mpi implementations in the 
case for example where an atmosphere takes several internal timesteps 
for each ocean timestep. At the end of the run the ocean may get to 
eedie while the atmosphere still has several internal timesteps left 
(depending on how you chose to couple). In that case you want to hold 
off calling Finalize from the ocean until the atmosphere is also done. 
the COMM_WORLD Barrier in eedie makes sure of this. However, it means in 
your atmosphere driver you will also need a Barrier(COMM_WORLD) at the 
very end.

chris
Gianmaria Sannino wrote:
> Hi folks,
> 
> In eeboot_minimal.F the default MPI_COMM_WORLD is passed to 
> MPI_COMM_MODEL. Throughout the code MPI_COMM_MODEL is used as the local 
> communicator. However there are few routines still using MPI_COMM_WORLD 
> (i.e. exch_jam.F, gsum_jam.F, eedie.F). Is there a particular reason for 
> that?
> thanks
> gianmaria & sandro
> 
> 
> 
> ******************************************************
> Gianmaria Sannino   ENEA (Italian Agency for Energy and Environment)
> Climate Project - Ocean Modelling Unit SP 91 - via Anguillarese 301
> S.M. di Galeria, 00060 ROMA, ITALY
> Voice: (+39) 06 3048 6799   Fax: (+39) 06 3048 4264
> e-mail: gianmaria.sannino at casaccia.enea.it
> URL: http://clima.casaccia.enea.it/staff/sannino/
> ******************************************************
> _______________________________________________
> MITgcm-support mailing list
> MITgcm-support at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-support
> 




More information about the MITgcm-support mailing list