AMBER Archive (2008)

Subject: Re: AMBER: Build a MPI NAB

From: David A. Case (case_at_scripps.edu)
Date: Tue Apr 08 2008 - 16:35:41 CDT

On Tue, Apr 08, 2008, Hu, Shaowen (JSC-SK)[USRA] wrote:
>
>
> MPI_Init: unclassified: MPI already initialized (rank 3, MPI_COMM_WORLD)

Aargh! Section 6.4 of the Users' Manual is outdated. The mpi_init()
and mpi_finalize() commands are now taken care of automatically.

The result is that you don't need to change anything to run MPI codes.
For example, go to amber10/test/nab, and type:

    nab -c gbrna_long.nab

Then look at the gbrna_long.c file. You will see what is happening.

Bottom line is that you should ignore the part in Section 6.4 where it tells
you to modify your nab programs -- this is now done for you.

You also need to pick up bugfix.2 from http://ambermd.org/bugfixes.html. Then,
try the following:

     cd amber10/test/nab
     nab -o gbrna_long gbrna_long.nab
     mpirun -np 2 ./gbrna_long

Sorry for the glitches: we mostly use the openmpi version of NAB, and it is
clear that the final testing did not check the MPI implementation.
-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber_at_scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo_at_scripps.edu