I apologize for the delay in replying -- too much backlogged e-mail...
Did you get this working properly with 7.1.1 or any of the 7.1.2 betas?
Note that LAM defaults to building statically -- have you tried
building it with dynamic libraries (i.e., use "--enable-shared" -- due
to oddness in the OSX linker, you can't build the shared libraries
without also building the static libraries, so "--enable-shared"
enables building both the static and the shared libraries).
On Jun 3, 2005, at 3:59 PM, Jean-Luc Vay wrote:
> I was compiling lam/mpi 7.1.1 with FC=xlf. All this part worked.
> It is only after rebuilding pyMPI and trying it that I got into
> trouble,
> i.e. got
>
> $ ./pyMPI
> dyld: ./pyMPI multiple definitions of symbol _Py_GetVersion
> ./pyMPI definition of _Py_GetVersion
> /sw/lib/python2.3/config/libpython2.3.dylib(getversion.o) definition
> of _Py_GetVersion
> Trace/BPT trap
>
> Since I could build and run pyMPI using lam/mpi 6.5.9
> pre-compiled library, I am assuming that the issue might just be
> related to options passed when configuring lam/mpi before
> compiling it. Interestingly, the multiple definition of _Py_GetVersion
> message, that is caught at run time using lam/mpi 7.1.1 (compiled
> by myself), preceding the 'Trace/BPT trap' message, is caused at
> link time when building pyMPI using the fink pre-compiled
> lam/mpi 6.5.9. So, I am assuming that some flags used when
> configuring before compiling the lam/mpi binary distributed through
> fink caused mpicc to detect the multiple definition at link time but
> somehow ignore it.
>
> Jean-Luc
>
> On Jun 3, 2005, at 5:19 AM, Jeff Squyres wrote:
>
>> On Jun 2, 2005, at 11:25 AM, Jean-Luc Vay wrote:
>>
>>> I would need to know which configure options have
>>> been used to compile the lammpi package version
>>> 6.5.9-1 that is distributed through fink. I can build
>>> and run pyMPI successfully with this package. Unfortunately,
>>> the mpi routines for FORTRAN have been compiled with
>>> the 2underscores options, which makes it impossible to
>>> link with my FORTRAN routines when compiled using xlf.
>>
>> I'm guessing that this is because fink is compiling against g77 and
>> you're trying to use xlf, and that they have different
>> symbol-mangling mechanisms. I confess to not knowing much how fink
>> works -- I see that 6.5.9 is the binary that is available, but the
>> source that is available is 7.0.6; can you use fink to install 7.0.6
>> and override its choice of Fortran compiler (e.g., insert "FC=xlf" in
>> the configure command line)?
>>
>>> I tried to compile lammpi myself but I then get the
>>> following message when starting pyMPI:
>>>
>>> $ ./pyMPI
>>> dyld: ./pyMPI multiple definitions of symbol _Py_GetVersion
>>> ./pyMPI definition of _Py_GetVersion
>>> /sw/lib/python2.3/config/libpython2.3.dylib(getversion.o) definition
>>> of _Py_GetVersion
>>> Trace/BPT trap
>>>
>>> I believe that it is a matter of configure options but do not know
>>> what options need to be set.
>>
>> What version of LAM/MPI were you compiling?
>>
>> --
>> {+} Jeff Squyres
>> {+} jsquyres_at_[hidden]
>> {+} http://www.lam-mpi.org/
>>
>> _______________________________________________
>> This list is archived at http://www.lam-mpi.org/MailArchives/lam/
>>
> -------------------------------------------------
> Jean-Luc Vay
> Accelerator and Fusion Research Division
> Lawrence Berkeley National Laboratory - BLDG 47R0112
> Berkeley, CA 94720, USA
> Tel: (1) 510-486-4934
> Fax: (1) 510-486-5392
> Email: jlvay_at_[hidden]
> -------------------------------------------------
>
> _______________________________________________
> This list is archived at http://www.lam-mpi.org/MailArchives/lam/
>
--
{+} Jeff Squyres
{+} jsquyres_at_[hidden]
{+} http://www.lam-mpi.org/
|
