On Tue, Sep 13, 2011, Lars Skjærven wrote:
> I'm having a rather unsuccessfull attempt to compile amber11 on a IBM
> cluster. The serial version seems to compile nicely, but my parallel
> version crashes with a bunch of "undefined reference" messages when
> trying to compile sander.MPI with mpif90. (see below).
>
> Using my own compiled openmpi to compile sander.MPI seems to work
> fine. However, in that case I'm limited to using only 1 node (8cores)
> at most. Thus, I have to use the mpif90 version in /usr/bin (in which
> I have problems with compiling)
>
> mpif90 has the following specs:
>
> -bash-3.2$ mpif90 -v
> Using built-in specs.
> Target: x86_64-redhat-linux
> Configured with: ../configure --prefix=/usr --mandir=/usr/share/man
> --infodir=/usr/share/info --enable-shared --enable-threads=posix
> --enable-checking=release --with-system-zlib --enable-__cxa_atexit
> --disable-libunwind-exceptions --enable-libgcj-multifile
> --enable-languages=c,c++,objc,obj-c++,java,fortran,ada
> --enable-java-awt=gtk --disable-dssi --enable-plugin
> --with-java-home=/usr/lib/jvm/java-1.4.2-gcj-1.4.2.0/jre
> --with-cpu=generic --host=x86_64-redhat-linux
> Thread model: posix
> gcc version 4.1.2 20080704 (Red Hat 4.1.2-48)
>
>
> ERROR MSG:
>
> mpif90 -o sander.MPI constants.o state.o memory_module.o stack.o
> nose_hoover.o nose_hoover_vars.o nose_hoover_init.o qmmm_module.o trace.o
> lmod.o rand2.o ....... spatial_recip.o spatial_fft.o parms.o
> softcore.o mbar.o molecule.o
> qm_mm.o qm_ewald.o qm2_variable_solvent.o qmmm_qmtheorymodule.o \
>
> .... [Lots of stuff] ....
> error messages like:
> _nose_hoover.f:(.text+0x1219): undefined reference to `mpi_allreduce_'
> qmmm_module.o: In function `__qmmm_module__qmmm_mpi_setup':
> _qmmm_module.f:(.text+0x742): undefined reference to `mpi_bcast_'
> _qmmm_module.f:(.text+0x767): undefined reference to `mpi_bcast_'
> _qmmm_module.f:(.text+0x78c): undefined reference to `mpi_bcast_'
> .... [Lots of stuff] ....
>
> ... [terminates with this error] ....
> /home/skjaerve/software/amber11/src/sander/_ncsu-utils.f:256:
> undefined reference to `mpi_comm_rank_'
> /home/skjaerve/software/amber11/src/sander/_ncsu-utils.f:259:
> undefined reference to `mpi_comm_size_'
> /home/skjaerve/software/amber11/src/sander/_ncsu-utils.f:269:
> undefined reference to `mpi_recv_'
> collect2: ld returned 1 exit status
> mpif90: No such file or directory
> make[1]: *** [sander.MPI] Error 1
> make[1]: Leaving directory
> `/home/skjaerve/software/amber11/src/sander'
> make: *** [parallel] Error 2
>
Typing "mpif90 --show" (or something similar, depends on the exact
distribution) will show what the mpif90 script is doing under the hood. That
might offer a clue. This doesn't look like a problem that has anything to do
with Amber: you might try compiling a simple test MPI program (e.g. that calls
mpi_init and mpi_bcast) to see if that works. Basically, the mpif90 command
should know how to find mpi_bcast() and friends.
Be sure you did the obvious things: use "which mpif90" to be sure your PATH
has the correct script; make sure you did a "make clean" before compiling, so
that nothing is left over from the earlier (successful) MPI compile step.
...good luck...dac
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Tue Sep 13 2011 - 07:00:03 PDT
