Re: [AMBER] regarding compiling amber10

From: Jason Swails <jason.swails.gmail.com>
Date: Tue, 9 Mar 2010 10:07:15 -0500

Hello,

On Tue, Mar 9, 2010 at 6:54 AM, jani vinod <genomejani.gmail.com> wrote:
> I have configured AMBER with the following option:
>
>
> export MPI_HOME=/opt/mvapich2-1.0.2
> export ACML_DIR=/opt/Libraries/acml/pathscale64_mp/lib
> ./configure_amber -mpich2 pathscale
>
> while building the parallel version i face the following problem:
>
> $ make parallel
> .
> .
> .
> ../lib/nxtsec.o ../lib/sys.a mt19937.o ncsu-abmd-hooks.o ncsu-bbmd-ctxt.o
> ncsu-bbmd-hooks.o ncsu-cftree.o ncsu-colvar-math.o ncsu-colvar-type.o
> ncsu-colvar-utils.o ncsu-colvar.o ncsu-constants.o ncsu-cv-ANGLE.o
> ncsu-cv-DISTANCE.o ncsu-cv-HANDEDNESS.o ncsu-cv-MULTI_RMSD.o
> ncsu-cv-N_OF_BONDS.o ncsu-cv-N_OF_STRUCTURES.o ncsu-cv-R_OF_GYRATION.o
> ncsu-cv-TORSION.o ncsu-lexer.o ncsu-parser.o ncsu-pmd-hooks.o ncsu-rmsd.o
> ncsu-sander-hooks.o ncsu-sander-proxy.o ncsu-smd-hooks.o ncsu-umbrella.o
> ncsu-utils.o ncsu-value.o ../netcdf/lib/libnetcdf.a
> -L/opt/Libraries/acml/pathscale64_mp/lib -lacml
> /opt/Libraries/acml/pathscale64_mp/lib/libacml.so: undefined reference to
> `__ompc_barrier'
> /opt/Libraries/acml/pathscale64_mp/lib/libacml.so: undefined reference to
> `__ompc_get_local_thread_num'
> /opt/Libraries/acml/pathscale64_mp/lib/libacml.so: undefined reference to
> `omp_in_parallel_'

These errors are associated with the ACML library you're trying to
link to. Amber is asking for access to functions that do not exist
(the names of all these functions are above). I don't think many
developers use the ACML (I could be mistaken, but there's not much
traffic regarding that particular math library. Intel MKL and GOTO are
far more common). First, it's important that libacml.so is actually
in the directory it's trying to access. Then, you can use the "nm"
unix utility to see if the functions it's looking for above are
actually in that library.

I haven't done much research regarding the relative performance of the
various math libraries, but I don't think you'll see a significant
performance boost using ACML over even the included BLAS routines that
come with Amber. If you unset ACML_DIR and re-run configure_amber
-mpich2 pathscale, then it will bypass the use of the ACML. If you
want to use ACML, though, you'll have to make sure that Amber can find
the routines it's looking for (which must start, at the very least, by
finding the files it's looking for in the directories it's looking for
them).

Hope this helps,
Jason

> /opt/Libraries/acml/pathscale64_mp/lib/libacml.so: undefined reference to
> `__ompc_fork'
> /opt/Libraries/acml/pathscale64_mp/lib/libacml.so: undefined reference to
> `omp_get_thread_num_'
> /opt/Libraries/acml/pathscale64_mp/lib/libacml.so: undefined reference to
> `__ompc_serialized_parallel'
> /opt/Libraries/acml/pathscale64_mp/lib/libacml.so: undefined reference to
> `omp_get_num_threads_'
> /opt/Libraries/acml/pathscale64_mp/lib/libacml.so: undefined reference to
> `__ompc_can_fork'
> /opt/Libraries/acml/pathscale64_mp/lib/libacml.so: undefined reference to
> `__ompc_end_serialized_parallel'
> collect2: ld returned 1 exit status
> make[1]: *** [sander.MPI] Error 1
> make[1]: Leaving directory `/md1/vinod/Amber10/amber10_1/amber10/src/sander'
> make: *** [parallel] Error 2
>
> Can anyone help in resolving the problem.
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
>



-- 
---------------------------------------
Jason M. Swails
Quantum Theory Project,
University of Florida
Ph.D. Graduate Student
352-392-4032
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Tue Mar 09 2010 - 07:30:02 PST
Custom Search