OMP threads for PME should be 1 for everything except MIC.
It should be cores per socket for GB simulations.
-Charlie
________________________________________
From: Jason Swails [jason.swails.gmail.com]
Sent: Saturday, March 05, 2016 4:35 AM
To: AMBER Developers Mailing List
Subject: Re: [AMBER-Developers] pmemd.MPI build broken
On Sat, Mar 5, 2016 at 2:50 AM, Charles Lin <clin92.ucsd.edu> wrote:
> So for the new Intel optimizations in PMEMD, openmp is currently required
> (I believe this is especially important for the GB portion). When we were
> looking through the configure2 file it looked like the dragonegg compiler
> didn't support openmp, which would make it incompatible with the new pmemd
> code. Its been thoroughly tested with intel, gnu, and mpich though.
>
It looks like the number of OpenMP threads is hard-coded to 1 for the CPU
version, is that correct? If that's the case, it kind of begs the question
why is it required for pmemd.MPI (and not just the MIC builds)?
Also, when I switch to using OpenMPI *without* dragonegg, the linker line
still needs -lgomp to complete successfully, so the build doesn't really
work in general yet.
_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers
_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers
Received on Sat Mar 05 2016 - 08:30:03 PST
