Re: [AMBER] pmemd with hybrid mpi/mp

From: Bill Abbott <babbott.rutgers.edu>
Date: Tue, 27 Oct 2020 16:13:32 -0400

Understood, thanks very much.

I noticed during this task that the new cmake procedure with Amber20
works well for both gcc and Intel, no issues other than MP, easy and
straightforward, if you're still collecting info on that.


Bill

On 10/27/20 4:04 PM, David A Case wrote:
> On Tue, Oct 27, 2020, Bill Abbott wrote:
>>
>> I'm trying to compile a hybrid pmemd with both MPI and OpenMP and it's
>> not acting the way I expect. It seems to build ok, the run_cmake step
>> indicates that OpenMP is included and the pmemd.MPI binary acts exactly
>> as I would expect with no OpenMP - exactly one core per rank.
>
> What you see is correct and expected. There was some effort to use
> OpenMP and MPI together on the Xeon Phi architecture, but it died when
> the enthusiams for Xeon Phi did.
>
> We should have some warning message about this. It's legal to use both
> OpenMP and MPI together, but cpptraj (the analysis program) is the only
> code that makes use of this, as far as I know. pmemd.MPI is not set up
> for OpenMP.
>
> Sorry for the confusion....regards....dave
>
> [Others on the list should chime in if the above description is wrong or
> incomplete.]
>
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
>

_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Tue Oct 27 2020 - 13:30:03 PDT
Custom Search