Re: [AMBER] amber11 intel mpi compile on altix 450 problem

From: Lisa Perez <mouse.mail.chem.tamu.edu>
Date: Wed, 16 Mar 2011 11:45:03 -0500

For anyone who might need it:

To compile amber11 with intel 11.1.037 and mkl 10.2.7 on my SGI Altix 450 (ia64) I had to make the following modifications:

remove the -lsvml flags from $AMBERHOME/AmberTools/src/configure (it should only be on 2 lines that you can comment out)
#libsvml is not available for ia64 intel compilers so you need to remove the -lsvml flag before you configure.

I also added the -shared-intel flag to the LDFLAGS and PMEMD_LD lines in config.h (if it wasn't already there) before each make since I ran into final link errors without it (see http://archive.ambermd.org/201012/0085.html) I also had to use the -shared-intel flag on our x86_64 machine with the same version of ifort, icc and mkl.

To get the parallel version to compile, I had to change every instance of mpif90 to ifort and mpicc to icc and add the -lmpi flag to LDFLAGS and PMEMD_LD in $AMBERHOME/src/config.h

Lastly, to run the parallel test jobs: make test.parallel, I had to change 1 line in the $AMBERHOME/test/test_amber_parallel.sh file:

#numprocs=`${DO_PARALLEL} echo "Testing number of processors" | wc -l`
numprocs=4

for some reason mpirun did not like the echo command so, I just set numprocs manually to 2 (I also set export DO_PARALLEL='mpirun -np 2') This problem may be specific my altix and you may not have this problem. On my machine, running: ${DO_PARALLEL} echo "Testing number of processors" | wc -l returns the following error:
MPI: could not run executable (case #3)
MPI: No details available, no log files found
Error: Parallel jobs on one CPU are not supported. Exiting.

note: 2 ambertools serial test jobs for chamber segmentation faulted:
cd ./chamber/dhfr_gas && ./Run.dhfr_gas_chamber
Segmentation fault
  ./Run.dhfr_gas_chamber: Program error
make[1]: [test.chamber] Error 1 (ignored)
cd ./chamber/dhfr_cmap_pbc && ./Run.dhfr_cmap_pbc_chamber
/bin/rm: No match.
Segmentation fault
  ./Run.dhfr_cmap_pbc_chamber: Program error

I didn't try to fix the seg fault problem on my altix machine. I did not get the seg fault on our x86_64 machine.

Sincerely,

Lisa M. Pérez, Ph.D.

Texas A&M University
Laboratory for Molecular Simulation, Manager
Graduate Faculty
http://lms.chem.tamu.edu/

Department of Chemistry
580 Ross Street
College Station, TX 77842

Chemistry Bldg. RM. 2109
Ofc: (979) 845-9384
Fax: (979) 845-2971



On Mar 15, 2011, at 9:38 AM, Jason Swails wrote:

> Hello,
>
> You have to make sure that you provide the location of the libraries either
> via a -L/path/to/mpi/lib or by adding /path/to/mpi/lib to LD_LIBRARY_PATH,
> so that the loader knows where to find it (these variables can be added to a
> LDFLAGS line. The second thing you have to make sure of is that the proper
> MPI headers are found at compile time for each source file. You'll have to
> either add -I/path/to/mpi/include or add path/to/mpi/include to INCLUDE_PATH
> (I prefer the first way, since I don't know how general the second way is)
> to the FFLAGS/CFLAGS variables in the config.h.
>
> Other than this, there is no *easy*, automated way of building an MPI
> version in which mpif90/mpicc DON'T exist. In order to provide any more,
> detailed, help, I would have to see the error messages you're getting along
> with the exact commands that generated them.
>
> Hope this helps,
> Jason
>
> On Mon, Mar 14, 2011 at 10:09 PM, Lisa Perez <mouse.mail.chem.tamu.edu>wrote:
>
>>
>> I am trying to install parallel (mpi) amber on our 64 processor altix 450
>> using the intel 11.1 compilers (mkl 10.2) and am running into problems.
>> ./configure -mpi -altix intel specifies the use of mpicc and mpif90 which
>> is not available in the ia64 version of the intel compilers. I have been
>> tinkering with the configure (changing mpicc -> icc, mpif90 -> ifort and
>> adding the -lmpi flag) but I wanted to know if anyone had a quick fix for
>> this problem since my modification have yet to be successful.
>>
>> I have already successfully compiled parallel (mpi) amber11 with the same
>> version of the intel compilers and mkl on an x86_64 machine.
>>
>> Thank you,
>>
>> Lisa M. Pérez, Ph.D.
>>
>> Texas A&M University
>> Laboratory for Molecular Simulation, Manager
>> Graduate Faculty
>> http://lms.chem.tamu.edu/
>>
>> Department of Chemistry
>> 580 Ross Street
>> College Station, TX 77842
>>
>> Chemistry Bldg. RM. 2109
>> Ofc: (979) 845-9384
>> Fax: (979) 845-2971
>>
>>
>>
>> _______________________________________________
>> AMBER mailing list
>> AMBER.ambermd.org
>> http://lists.ambermd.org/mailman/listinfo/amber
>>
>
>
>
> --
> Jason M. Swails
> Quantum Theory Project,
> University of Florida
> Ph.D. Candidate
> 352-392-4032
> _______________________________________________
> 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 Wed Mar 16 2011 - 10:00:04 PDT
Custom Search