[gmx-users] Segmentation fault

[Dmitry Kovalsky]

On Sunday 20 October 2002 13:00, gmx-users-request at gromacs.org wrote:
> >>anything. Is it reproducible? What machine?
> >>    
> >
> >Ya, it happens every time I'm trying to continue it again
> >
> >I'm using Linux on Athlon
>
> Is it a newer type Athlon using SSE? (it says in the mdrun logfile)
here is from log....


Testing x86 3DNow capabilities...
CPU and OS support extended 3DNow.
Using Gromacs 3DNow single precision assembly innerloops.
>
> If you have a small tpr file where the error happens soon (say max an
> hour into the simulation), I'd be
> REALLY interested in a copy.
Generally speaking it is the continuation of crashed run. i mean, initial run 
had crashed due to bond length warning, after I used 
tpbconv -s pr.tpr -f pr.trr -o pra.tpa  
and obtained the "unsuccessfull" pra.tpr file which don't want to work. Its 
size is 
-rw-r--r--    1 dikov    dikov     5055360 Oct 18 12:47 pra.tpr
so if you want I cand send it and it will crash in few houres (on my 850 MHz 
athlon). 
Now I'm restarting the initial run and will try to rebtain the crash.

Sincerely yours
Dima
>
> I have been trying to track down a possible problem on Athlons for a
> while, but I've only seen it a week
> into a long run, so it's been really hard to repeat.
>
> I have no idea what it is since it doesn't show on Intel (which also
> uses SSE), but there might be minor
> architectural differences of course.
>
> You might also want to try setting the environment variable
> "NOASSEMBLYLOOPS" (slow), but since
> we sum things in a slightly different order in that case it might not
> necessarily be a bug in the assembly
> routines if it doesn't crash...
>
>
> Cheers,
>
> Erik

-- 
Sincerely yours,

Dima

Ph.D. Student Dmitry Kovalsky
Institute of Molecular Biology & Genetics
150 Akad. Zabolotnogo Street,
Kiev-143, 03143
UKRAINE

E-mail: dikov at imbg.org.ua
Fax:  +380 (44) 266-0759
Tel.: +380 (44) 266-5589