It's not entirely clear that this means that it is a bug in Open MPI -- there's not really enough information here to say where the problem is. All that is clear is that a seg fault is happening somewhere in LAPACK.

FWIW, I don't see MPI in the call stack of the segv at all. This doesn't definitively mean that MPI is not involved, but it does suggest that the bug is elsewhere. You might want to run this under a debugger or examine the generated corefiles to see exactly what the problem was.


On Aug 15, 2008, at 1:03 PM, C.Y. Lee wrote:

All,

I had a similar problem as James described in an earlier message: 
http://www.open-mpi.org/community/lists/users/2008/07/6204.php
While he was able to recompile openmpi to solve the problem, I had no luck with my RedHat Enterprise 5 system. Here are two other threads with similar issues regarding openmpi on Ubuntu and OSX which were solved: https://bugs.launchpad.net/ubuntu/+source/binutils/+bug/234837
http://www.somewhereville.com/?cat=55

Now...
Here is my story:
I had Quantum Espresso (QE) running without problem using openmpi.
However, when I tried to recompile QE with a recompiled fftw-2.1.5, it compiled without any error. But when I ran QE, it gave me the error below:

*** Process received signal ***
Signal: Segmentation fault (11)
Signal code: Address not mapped (1)
Failing at address: 0x22071b70
[ 0] /lib64/libpthread.so.0 [0x352420de70]
[ 1] /usr/lib64/liblapack.so.3(dsytf2_+0xc43) [0x2aaaaac9f5e3]
[ 2] /usr/lib64/liblapack.so.3(dsytrf_+0x407) [0x2aaaaaca0567]
[ 3] /opt/espresso-4.0.1/bin/pw.x(mix_rho_+0x828) [0x5044b8]
[ 4] /opt/espresso-4.0.1/bin/pw.x(electrons_+0xb37) [0x4eae47]
[ 5] /opt/espresso-4.0.1/bin/pw.x(MAIN__+0xbf) [0x42b3af]
[ 6] /opt/espresso-4.0.1/bin/pw.x(main+0xe) [0x6aad5e]
[ 7] /lib64/libc.so.6(__libc_start_main+0xf4) [0x352361d8a4]
[ 8] /opt/espresso-4.0.1/bin/pw.x [0x42b239]
 *** End of error message ***

From what I read from the above links, it seems to be a bug in openmpi.
Please share your thoughts on this, thank you!

CY
_______________________________________________
users mailing list
us...@open-mpi.org
http://www.open-mpi.org/mailman/listinfo.cgi/users


--
Jeff Squyres
Cisco Systems

Reply via email to