[gmx-developers] grompp and vsites

[Berk Hess]

Michel Cuendet wrote:
>
> Hi David,
>> This doesn't look good. Are you certain compilation went fine? Did
>> you run the test set?
>> Did you compile double and single from exactly the same source? And
>> did you run make distclean in between?
>> The only other thing I can think of is that there were strange
>> characters in the mdp file.
>>   
> After recompiling, all grompp versions suddenly give the same output
> with vsites (without the -e option).  No idea what happened there.  :0S
>
> After that, the other problem kept happening consistently: using the
> -e option to read an edr file would create a Segmentation fault (as
> mentioned in previous email). Gdb output:
> -------------------------------------------------------
> Program received signal SIGSEGV, Segmentation fault.
> 0xb78d64ac in free () from /lib/tls/i686/cmov/libc.so.6
> (gdb) bt
> #0  0xb78d64ac in free () from /lib/tls/i686/cmov/libc.so.6
> #1  0xb7a8fccd in save_free ()
>   from
> /export/home/mcuendet/Gromacs/gromacs-4.0.cvs_12.6.2009//i686_gcc4.2.4//lib/libgmx.so.5
>
> #2  0xb7a8fcb4 in save_free ()
>   from
> /export/home/mcuendet/Gromacs/gromacs-4.0.cvs_12.6.2009//i686_gcc4.2.4//lib/libgmx.so.5
>
> #3  0xb7a40e4f in do_enxnms ()
>   from
> /export/home/mcuendet/Gromacs/gromacs-4.0.cvs_12.6.2009//i686_gcc4.2.4//lib/libgmx.so.5
>
> #4  0x0000003d in ?? ()
> Cannot access memory at address 0xfff5a8ed
> -------------------------------------------------------
>
> I figured out that compiling without the --enable-shared option solves
> the problem!! The grompp on the 64bit machines (which was working, see
> previous email) was compiled without shared libraries, and that was
> the only significant difference. So there is a problem in the way
> shared libraries are called.
>
> Now I think I have all the CVS grompps working properly. Note that
> disabling the shared libraries did NOT improve the SegFault with
> pdb2gmx mentioned in a separate thread.
>
> However, I still see some errors in the gmxtests, depending on which
> CVS version I look at. Could you confirm if the latest CVS should
> actually pass all the gmxtests ?
>
> Thanks,
>
> Michel
Are you really, really sure you don't have any compilations problems?
Did you try to compile from a completely clean tree?
Also the energy file format has changed. Reading a new energy file
with a pre 4.0.4 version of g_energy might give a segv and not a fatal
error.

Berk