I'm guessing mdgx is using some MPI-2 features that Intel MPI does not have
implemented. OpenMPI (at least as of version 1.5.x) also does not
implement a large number of MPI-2 features (such as those used by the
sander/terachem interface and pmemd.cuda.MPI). And apparently MDGX, now,
as well.
If you don't intend to use mdgx, you can just skip that building (change
MDGX from yes to no in config.h).
Also, I always suggest upgrading to the latest developer version (via 'git
pull') before reporting errors, as nobody will attempt to debug an old tree
state ;).
HTH,
Jason
On Fri, Oct 26, 2012 at 11:50 AM, Dickson, Callum <
callum.dickson09.imperial.ac.uk> wrote:
> Thanks Jason, that first option worked fine (installing Amber without
> AmberTools).
>
> I did previously try OpenMPI without success, there isn't a version of
> mpich2 available on the cluster.
>
> Cheers,
> Callum
>
> ------------------------------
> *From:* Jason Swails [jason.swails.gmail.com]
> *Sent:* 26 October 2012 16:28
> *To:* Dickson, Callum; AMBER Developers Mailing List
> *Subject:* Re: [AMBER] Compiling amber parallel
>
> I'm moving this discussion over to the developers list since it's a
> developer version issue.
>
> On Fri, Oct 26, 2012 at 11:11 AM, Dickson, Callum <
> callum.dickson09.imperial.ac.uk> wrote:
>
>> Hi,
>>
>> I am having some issues trying to compile the development version of
>> amber12 parallel on the cluster at my university. I am able to compile the
>> serial version, however I get an MPI related error when I then try to
>> compile the parallel version. The configuration of amber12 parallel seems
>> to proceed without a problem.
>>
>> I am using intel compilers and intel MPI. Here is the exit output error
>> when I try to compile:
>>
>>
>> FILE_SOURCE -DBINTRAJ -I/home/cd709/include -DMPI
>> -I/apps/intel/2013/mkl/include -shared-intel \
>> -o mdgx.MPI CompFrc.o BSpline.o Grid.o Random.o mdgxVector.o Matrix.o
>> ChargeMap.o pmeRecip.o mleRecip.o pmeDirect.o CrdManip.o CellManip.o
>> Topology.o Trajectory.o SpecialMath.o Nonbonded.o Bonded.o Parse.o
>> Command.o Constraints.o Thermostats.o Barostats.o Integrator.o Timings.o
>> Manual.o VirtualSites.o Buckingham.o ThermoDyn.o ChargeFit.o ParamFit.o
>> Debug.o BroadcastCommand.o MPIMap.o MPITypeCast.o ptrajmask.o
>> ../sff/AmberNetcdf.o mdgx.o -lfftw3 -L/home/cd709/amber12/lib
>> -L/home/cd709/amber12/lib -lnetcdf -lm
>> /apps/intel/ict/mpi/3.1.038/lib/libmpigf.so: undefined reference to
>> `MPIR_Err_return_comm'
>>
>> /apps/intel/ict/mpi/3.1.038/lib/libmpigf.so: undefined reference to
>> `MPIR_Grequest_set_lang_f77'
>>
>> /apps/intel/ict/mpi/3.1.038/lib/libmpigf.so: undefined reference to
>> `b_use_gettimeofday'
>>
>> /apps/intel/ict/mpi/3.1.038/lib/libmpigf.so: undefined reference to
>> `MPID_Wtick'
>>
>> /apps/intel/ict/mpi/3.1.038/lib/libmpigf.so: undefined reference to
>> `MPIR_Err_create_code'
>>
>> /apps/intel/ict/mpi/3.1.038/lib/libmpigf.so: undefined reference to
>> `i_malloc'
>>
>> /apps/intel/ict/mpi/3.1.038/lib/libmpigf.so: undefined reference to
>> `MPIR_Keyval_set_fortran90'
>>
>> /apps/intel/ict/mpi/3.1.038/lib/libmpigf.so: undefined reference to
>> `i_free'
>>
>> /apps/intel/ict/mpi/3.1.038/lib/libmpigf.so: undefined reference to
>> `MPIR_Keyval_set_fortran'
>>
>> /apps/intel/ict/mpi/3.1.038/lib/libmpigf.so: undefined reference to
>> `MPID_Wtime_todouble'
>>
>> make[2]: *** [mdgx.MPI] Error 1
>>
>> make[2]: Leaving directory
>> `/export71/home/cd709/amber12/AmberTools/src/mdgx'
>>
>> make[1]: *** [parallel] Error 2
>>
>> make[1]: Leaving directory `/export71/home/cd709/amber12/AmberTools/src'
>>
>> make: *** [install] Error 2
>>
>
> Yikes. Looks like the MPI is borked. Maybe mdgx is not strictly
> adhering to some standard. Or maybe MDGX is using some MPI-2 standard
> features that Intel has not implemented in their MPI. (Dave C. can weigh
> in on this if he knows any of them to be true).
>
> In any case, I don't know of many devs that regularly test Intel's MPI,
> so you have a couple options (and if you try them, we'd appreciate hearing
> back):
>
> 1) compile the Amber programs (sander.MPI and pmemd.MPI) without doing
> any of the AmberTools installs: cd $AMBERHOME/src && make install) and see
> if that works.
> 2) try mpich2 instead of intel MPI (or OpenMPI, either one).
>
> Also, I don't know of any devs that have been using Intel 13 compilers
> yet, so you could also try downgrading to an older (more-tested) version if
> the 2 options above don't work.
>
> Good luck,
> Jason
>
> --
> Jason M. Swails
> Quantum Theory Project,
> University of Florida
> Ph.D. Candidate
> 352-392-4032
>
--
Jason M. Swails
Quantum Theory Project,
University of Florida
Ph.D. Candidate
352-392-4032
_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers
Received on Fri Oct 26 2012 - 09:30:04 PDT
