> On Oct 16, 2015, at 17:20, Jason Swails <jason.swails.gmail.com> wrote:
>
>> On Fri, Oct 16, 2015 at 5:13 PM, Daniel Roe <daniel.r.roe.gmail.com> wrote:
>>
>> On Fri, Oct 16, 2015 at 2:51 PM, Novosielski, Ryan
>> <novosirj.ca.rutgers.edu> wrote:
>>>
>>> Thanks. Then I guess doing a serial build and then building over the top
>> with OpenMP to only replace those parts that support OpenMP is a good
>> solution? Thanks again!
>>
>> The OpenMP build of cpptraj doesn't actually replace the serial
>> binary; they are two separate entities (cpptraj and cpptraj.OMP). I
>> think that nab does get replaced (although I may be wrong about that).
>
> Yes, nab gets replaced. And nab benefits nicely from OpenMP
> parallelization when you're using it to compute nonbonded interactions
> (like GB, for example). The most common improvement here is through
> cpptraj.
Thanks again! Have a nice weekend.
> P.S., as a side-note to Dan, what do you think about trying to get both
> the serial and OpenMP versions built in the serial build? It seems madness
> *not* to use cpptraj.OMP when it's never slower, and quite often a bit
> faster :)
Might it not make more sense to build all of the parts that cannot be built with OpenMB as part of the OpenMP build? Sort of making the switch "OpenMP, if possible." I don't know what one would do to get a serial billed under your scenario, I should they have wanted one. Under the OpenMP-only scenario, is it really better for someone to not get binaries at all? I would think a binary and a warning would be more useful.
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Fri Oct 16 2015 - 14:30:05 PDT
