Hi Dave
> The one area in particular that may need to be updated is in the
> Ambertools/src/configure -help output. It implies that -mpi and
> -openmp can be used at the same time for sander QM/MM, but are
> mutually exclusive in nab. Also src/sander/ifdefs reads that openmp
> is experimental which is not mentioned elsewhere.
>
> Just checking with the SDSC guys: are we really saying we support
> simultaneous MPI and openMP? Or should we not disable this until we
> are sure
> that it works, and is useful for real problems?
As far as I can tell the 'ONLY' mention of this is in the configure help
output:
-openmp Use OpenMP pragmas to parallelize (not available for pgi;
only affects NAB and sander QM/MM. For NAB, do not set
-openmp and -mpi at the same time.)
There is also mention of the OPENMP flag in src/sander/ifdefs which lists
experimental. I don't see where it is documented anywhere else so I think
the above poster has a case of probably grepping the code and then assuming
that no documentation implies it is fully supported etc etc.
Maybe we should just remove the mention of sander QM/MM from the configure
help line. I guess it is like most things, if people see a single mention of
it then they assume it is totally complete, fully works, works as well as
all the snake oil salesmen claim it does etc etc. This is the real problem.
You have tons of sales guys and people who don't actually write code going
around telling everyone how great hybrid MPI / OpenMP is, how everyone
should use it etc etc. Then I think these people look at a 'real' code like
sander and assume we followed their advice about it being easy, efficient
etc etc...
As for mixing openmp with mpi and sander it works fine. In fact it is
probably safer because it has code at the beginning to force the number of
openmp threads to 1 unless the user explicitly asks for more in the
namelist. So it is perfectly fine to have it here. Unless you really know
what you are doing (use very specific MPI libraries etc) you don't get any
performance increase but that said I don't think we made any claims anywhere
about it actually giving better performance over pure MPI. Again people
believing the PR they are being fed trying to justify that mulit-core is
actually a good idea....
Oh well, ho hum... What do you want to do?
All the best
Ross
/\
\/
|\oss Walker
| Assistant Research Professor |
| San Diego Supercomputer Center |
| Tel: +1 858 822 0854 | EMail:- ross.rosswalker.co.uk |
|
http://www.rosswalker.co.uk |
http://www.wmd-lab.org/ |
Note: Electronic Mail is not secure, has no guarantee of delivery, may not
be read every day, and should not be used for urgent or sensitive issues.
_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers
Received on Fri Jul 09 2010 - 08:00:06 PDT
