Re: [AMBER-Developers] MPI Amber build is broken

From: Yinglong Miao <yinglong.miao.gmail.com>
Date: Tue, 7 Nov 2017 07:16:36 -0600

Hi All,

Thanks a lot for spotting the problem. I tested the serial cpu/gpu pmemd, which runs fine. I will try to look into the parallel version in more detail and fix it today …

Sorry for any inconvenience,
Yinglong

> On Nov 7, 2017, at 3:20 AM, David Cerutti <dscerutti.gmail.com> wrote:
>
> At risk of sounding repetitive, that looks like something related to
> parallel pmemd and GaMD. I've seen the subroutine call in question, and
> the my_atm_lst argument is a staple of parallel pmemd workload
> distribution. Probably he just needs to update his code to handle the
> complete set of arguments, including my_atm_lst, needed by the parallel
> code. One could go in there and patch it up, but there are a lot of things
> with parallel pmemd, particularly the cuda code, that saddle the master
> node with small but significant tasks. Calculating weights could well be
> one of them, so I'd suggest leaving it to Yinglong to correct.
>
> Dave
>
>
> On Tue, Nov 7, 2017 at 3:46 AM, Jamie Smith <
> jsmith.crackofdawn.onmicrosoft.com> wrote:
>
>> Hello All,
>>
>> I’m not sure if anyone else noticed, but since Yinglong Miao’s merge of
>> the new GaMD code, MPI PMEMD has been failing to build. I first spotted
>> this on the CMake Jenkins server, but it affects the Makefile build system
>> as well. Serial PMEMD builds fine, but the MPI version dies during
>> compilation with the following error:
>>
>> mpif90 -DMPI -DBINTRAJ -DEMIL -DPUBFFT -mtune=native
>> -I/home/jamie/amber-cmake/amber/include -c pme_force.F90
>> pme_force.F90:1315:81:
>>
>> pot_ene%total,pot_ene%dihedral,pot_ene%gamd_boost,
>> frc,crd)
>>
>> 1
>> Error: Missing actual argument for argument ‘my_atm_lst’ at (1)
>> pme_force.F90:1319:110:
>>
>> call calculate_gamd_nb_weights(atm_cnt,
>> pot_ene_nb,pot_ene%dihedral,pot_ene%gamd_boost,nb_frc_gamd,crd)
>>
>> 1
>> Error: Missing actual argument for argument ‘my_atm_lst’ at (1)
>>
>> It looks like the “my_atm_lst” argument does not exist for
>> calculate_gamd_nb_weights() when building for MPI.
>>
>> Can we see about fixing this issue? Or is it some kind of build system
>> problem?
>>
>> Thanks,
>> Jamie
>> _______________________________________________
>> 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


_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers
Received on Tue Nov 07 2017 - 05:30:02 PST
Custom Search