OK, a few more data points:
(1) libmpi_f90.so *does* exist within the /global/software/...
directory. libmpi_f90.a *does not* exist, but there is a file called
libmpi_f90.la. This is interesting/weird, as on my Macs and on two
other clusters that I use, the openmpi library directory has both
libmpi_f90.a and libmpi_f90.la.
(2) Trying to build with ./configure -mpi -static -noX11 intel results
in the "ld_ cannot find -lmpi_f90" error for sander.MPI. In other
words, the compiling process dies even sooner.
(3) Nevermind my bugfix 8 idea. That was a bad (i.e., dumb) Friday
afternoon thought!
I am going to try Ross's suggestion to link manually next, but I think
the lack of libmpi_f90.a might be more relevant...
Thanks for your help,
Paul
On Sep 10, 2010, at 5:03 PM, Jason Swails wrote:
> Hello,
>
> In addition to what Ross said, is either of those files he mentioned (
> libmpi_f90.so/libmpi_f90.a) in any of the library files specified by
> -L in
> mpif90 -show? Namely, check
> /global/software/centos-5.x86_64/modules/openmpi/1.4.2-intel/lib to
> see if
> either of those files are present. If not, where are they? In any
> case, it
> seems to be more a problem with your MPI than amber itself (or
> perhaps, like
> Ross suggested, only dynamic or static linking is supported by this
> installation of MPI and you unluckily chose the wrong one).
>
> Good luck!
> Jason
>
> On Fri, Sep 10, 2010 at 7:46 PM, Ross Walker <ross.rosswalker.co.uk>
> wrote:
>
>> Hi Paul,
>>
>> This looks strange since the library is actually included by your
>> mpif90
>> script. Did this successfully build sander.MPI correctly?
>>
>> Take a look and see if libmpi_f90.so exists in
>> /global/software/centos-5.x86_64/modules/openmpi/1.4.2-intel/lib.
>> You may
>> need to be looking in lib64 instead of lib if you are using the
>> INTEL64
>> version of the compilers (x86_64). The other possibility that would
>> cause
>> this is if you tried to link statically and only had dynamic '.so'
>> versions
>> of the library or vice versa, tried to link dynamically but only
>> had static
>> '.a' versions of the library.
>>
>> Try repeating the build with the -static flag given to configure
>> and see if
>> that makes any difference.
>>
>> Alternatively try running the link step manually using the command
>> line
>> mpif90 -show gives you in place of mpif90. Then give it the
>> explicit path
>> to
>> the libmpi_f90.so library.
>>
>> All the best
>> Ross
>>
>>> -----Original Message-----
>>> From: Paul S. Nerenberg [mailto:psn.berkeley.edu]
>>> Sent: Friday, September 10, 2010 4:27 PM
>>> To: AMBER Mailing List
>>> Subject: [AMBER] pmemd 11 (parallel) compilation error
>>>
>>> Hi all,
>>>
>>> I'm compiling AMBER 11 on a new system (using the Intel compilers)
>>> and
>>> running into some trouble at more or less the last step of "make
>>> parallel" -- compiling pmemd.MPI. The serial versions of sander and
>>> pmemd both compile without a problem on this system, and the
>>> parallel
>>> version of sander also compiles correctly (*and* passes the parallel
>>> test cases). Here is the error that I get when compiling pmemd.MPI:
>>>
>>> mpif90 -fast -o pmemd.MPI gbl_constants.o ...bunch of object
>>> files...
>>> charmm_gold.o ../../netcdf/lib/libnetcdf.a -L/global/software/
>>> centos-5.x86_64/modules/intel/Compiler/11.1/072/mkl/lib/em64t -Wl,--
>>> start-group /global/software/centos-5.x86_64/modules/intel/Compiler/
>>> 11.1/072/mkl/lib/em64t/libmkl_intel_lp64.a /global/software/
>>> centos-5.x86_64/modules/intel/Compiler/11.1/072/mkl/lib/em64t/
>>> libmkl_sequential.a /global/software/centos-5.x86_64/modules/intel/
>>> Compiler/11.1/072/mkl/lib/em64t/libmkl_core.a -Wl,--end-group -
>>> lpthread
>>> ipo: remark #11000: performing multi-file optimizations
>>> ipo: remark #11005: generating object file /tmp/ipo_ifortx3AcWQ.o
>>> ld: cannot find -lmpi_f90
>>> make[2]: *** [pmemd.MPI] Error 1
>>> make[2]: Leaving directory `/global/home/users/psn/amber11/src/
>>> pmemd/
>>> src'
>>> make[1]: *** [parallel] Error 2
>>> make[1]: Leaving directory `/global/home/users/psn/amber11/src/
>>> pmemd'
>>> make: *** [parallel] Error 2
>>>
>>> Here is the output of "mpif90 -show":
>>> [psn.ln000 bin]$ mpif90 -show
>>> ifort -I/global/software/centos-5.x86_64/modules/openmpi/1.4.2-
>>> intel/
>>> include -I/global/software/centos-5.x86_64/modules/openmpi/1.4.2-
>>> intel/
>>> lib -L/global/software/centos-5.x86_64/modules/openmpi/1.4.2-intel/
>>> lib
>>> -lmpi_f90 -lmpi_f77 -lmpi -lopen-rte -lopen-pal -ldl -Wl,--export-
>>> dynamic -lnsl -lutil
>>>
>>> And finally, here are relevant details of the system configuration:
>>> * OS: CentOS 5.5
>>> * Compilers: Intel v11.1.072
>>> * Math library: Intel MKL (also v11.1.072)
>>> * OpenMPI 1.4.2
>>> * AMBER was configured using ./configure -mpi -noX11 intel
>>>
>>> Any suggestions? This seems similar to an error that I saw in the
>>> archive for compiling pmemd in AMBER 9, but the resolution to the
>>> problem was never posted (as far as I can tell)...and it's been my
>>> experience that pmemd in AMBER11 is so much more straightforward to
>>> configure that I'm actually quite surprised I ran into any trouble.
>>>
>>> Best,
>>>
>>> Paul
>>>
>>> _______________________________________________
>>> 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
>>
>
>
>
> --
> 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
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Fri Sep 10 2010 - 17:30:06 PDT
