AMBER Archive (2009)

Subject: Re: [AMBER] Compilation and usage of NAB in parallel ?

From: Marek Malý (maly_at_sci.ujep.cz)
Date: Fri Jan 30 2009 - 08:59:07 CST


Dear prof. Case,

thank you very much for your info !

I finally succeeded with -openmp option.
Nevertheless I am really curious why especially configuration with
-scalapack
flag but also with -mpi flag doesn't work. It is clear from the attached
header
files, which was created after config. commands.

./config_at -mpi icc (config_mpi.h)
./config_at -openmp icc (config_openmp.h)
./config_at -scalapack icc (config_scalapack.h)

that only in the case of -openmp this flag properly propagated into
config header file. I think that especially in the case of -scalapack
there should be no problem since as I learned recently ScaLAPACK is
part of our Intel MKL libraries as you rightly indicated in your mail.
Strange is that in case -mpi nor in the case -scalapack I didn't
obtain any error messages during configuration preocess. Below is screen
output from the configuration process.
It is identical in all 3 cases. Also files "config_mpi.h" and
"config_scalapack.h"
are identical except the first row (I checked it). It is not so crucial
issue for me
but if any explanation of this strange behaviour available I would
appreciated it much.

       Marek

----The screen output during the
configuration---------------------------------------------------

mmaly_at_enode12 ~/amber/src $ ./configure_at -scalapack icc
Setting AMBERHOME to /home/mmaly/amber

Using Intel MKL libraries in /opt/intel/mkl/10.0.011
Warning: the X11 libraries are not in the usual location !
     To search for them try the command: locate libXt
     On Fedora Core 5 install an xorg-x11-devel package.
     On RedHat8 install an XFree86-devel package.
     For the moment Amber will be configured not to build XLEaP.

Testing the C compiler:
       icc -wd117,177,266,880,1011 -o testp testp.c
OK

Testing the g77 compiler:
      g77 -O2 -fno-automatic -finit-local-zero -o testp testp.f
./configure_at: line 538: g77: command not found
./configure_at: line 539: ./testp: No such file or directory
Unable to compile a Fortran program using g77 -O2 -fno-automatic
-finit-local-zero

Testing the gfortran compiler:
      gfortran -O1 -fno-automatic -o testp testp.f
OK

Testing flex:
OK

Configuring netcdf; (may be time-consuming)

NETCDF configure succeeded.

The configuration file, config.h, was successfully created.

The next step is to type 'make -f Makefile_at'

Dne Thu, 29 Jan 2009 14:22:08 +0100 David A. Case
<case_at_biomaps.rutgers.edu> napsal/-a:

> On Wed, Jan 28, 2009, Marek Mal? wrote:
>>
>> I need urgently parallel version of NAB to use it for huge system (15000
>> atoms) analysis.
>>
>> I found just a little about this topic in AmberTools manual pages
>> 115-116.
>>
>> Our platform is "SGI AltixXE 310" with "Intel Xeon Quad-core 5365"
>> processors.
>
> With this platform, I think you should give the "-openmp" flag to get a
> parallel NAB.
>
>> But it just loaded 8 serial jobs in one time nothing more.
>
> Since you compiled NAB in a serial fashion, this result makes sense.
>
>>
>> The main problem is probably that I was not able to compile NAB in
>> parallel way.
>> I tried -mpi flag but using this option was absolutely useless it even
>> didn't add
>> the proper flags into config.h.
>
> You give few details here. The "-mpi" flag in NAB assumes that you have
> a correctly configured "mpicc" in you PATH.
>
>> -openmp successfully propagated into config.h in this two lines:
>> -L$(MKL_HOME)/lib/em64t -lguide -lpthread -openmp
>> OCFLAGS=-wd117,177,266,880,1011 -O2 -DBINTRAJ -openmp -DOPENMP
>> $(AMBERBUILDFLAGS)
>
> Now you should set OMP_NUM_THREADS and see what happens.
>
> Don't confuse openmp with MPI: they are distinct and not compatible with
> each other. If you use the -openmpi flag, don't use mpiexec (etc.)
>
>>
>> If "-intelmpi way" (IntelMPI) is really unpossible could anyone who has
>> some experience with parallel compilation/run of NAB with -openmp (under
>> OpenMPI)
>
> This is an example of the last point: openmp has nothing to do with
> openMPI...
>
> Admittedly, this is hard stuff, and not for the faint of heart. I've
> never run a truly parallel normal mode calculations (via Scalapak). If
> you have an altix, you may well have the Intel MKL libraries. If so,
> try using them, since I think(?) they automagically will parallelize
> some of the linear algebra.
>
> ...good luck...dac
>
>
> _______________________________________________
> AMBER mailing list
> AMBER_at_ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
>
> __________ Informace od NOD32 3810 (20090129) __________
>
> Tato zprava byla proverena antivirovym systemem NOD32.
> http://www.nod32.cz
>
>

-- 
Tato zpráva byla vytvořena převratným poštovním klientem Opery:  
http://www.opera.com/mail/




_______________________________________________
AMBER mailing list
AMBER_at_ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber