Hi,
On Sun, Apr 28, 2013 at 04:34:08PM -0400, David A Case wrote:
> On Sun, Apr 28, 2013, M. L. Dodson wrote:
> >
> > I can't see how openmp, on its own, would impact queue
> > handling. It just implements multiple threads on the same CPU within the same
> > process.
>
> Depends on what you mean by "the same cpu": an openmp job *could* be using
> multiple cores at times when that conflicted with other expectations, either
> from the user or from the queuing system.
>
> But I agree with the general point that we could/should do more to encourage
> use of the openmp capabilities of cpptraj and nab. Certainly promoting it
> more would be one strategy, short of making openmp the default.
Our current openmp installation model is broken imo:
http://archive.ambermd.org/201301/0431.html
A simple fix is to follow our MPI model - openmp executables could have
.OPENMP or .OMP suffixes. Users that know about openmp and want to use
it will know what to do; users that don't know will be insulated from
unintended consequences.
A start for promoting openmp executables would be improvements to the manual
and benchmarks that underscore openmp advantages.
scott
education
_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers
Received on Sun Apr 28 2013 - 16:00:02 PDT
