[gmx-users] Debugging

[Erik Lindahl]

Hi,

Right, then it sounds like a classical crash, but one that we don't 
detect.

The normal reason for NaN is that some coordinate get a high velocity 
(think above speed of light), and eventually the coordinate exceeds the 
available numerical precision.

The easiest way to find exactly where the error (probably segfault) 
happens is to recompile with debugging support by setting

CFLAGS="-O3 -g"

before running ./configure, and then run the program inside gdb as

1. "gdb /path/to/mdrun"
2. in gdb: "run <list of arguments to mdrun>"

After a while you should get an error message telling you exactly where 
it causes the segfault (or other error).

Cheers,

Erik


On Feb 18, 2005, at 11:28 PM, Kirill Titievsky wrote:

> Erik,
> the program does not run -- it's not there if you call it with top.
> When in the simulation it crashes depends on the time step, force
> strengths, etc.    But, BUT, if I restart andy of the crashed
> simulations the available trajectory information, the restart quits
> nicely right away reporting NaN forces.  I use soft potentials that
> make finite forces under all conditions, so it's either the way gmx
> reads the force table or the way it applies boundary conditions (some
> atoms in the restart *.gro file) are outside the box.
>
> thanks
>
> Kirill
>
>
>
> On Fri, 18 Feb 2005 23:00:20 +0100, Erik Lindahl <lindahl at sbc.su.se> 
> wrote:
>> Hi,
>>
>> Hm. not really.
>>
>> Is the program still running if you run a command like "top"?
>>
>> And, can you interrupt it with ctrl+c?
>>
>> (About a year ago we worked around a hardware bug on Athlon CPUs that
>> caused total OS hangs, but I'm not aware of any issues on P4).
>>
>> How long into the simulation does it happen?
>>
>> Cheers,
>>
>> Erik
>>
>> On Feb 18, 2005, at 10:44 PM, Kirill Titievsky wrote:
>>
>>> Erik,
>>>> 0. What Gromacs version, hardware, and operating system version are
>>>> you
>>>> using?
>>> I'm using the current version of GMX posted on the website running
>>> under Linux Kernel 2.6 on an  P4.
>>>> 1. By "stop running", do you mean crash, or a freeze where the step
>>>> count stops updating?
>>> That's exactly it: simulations just stop running.  No error messages,
>>> md.log stops in the middle of a line.
>>>
>>> Does this give any clues?
>>>
>>> thanks
>>>
>>>> 2. If it is a Mac (using Altivec) or x86 (using SSE/SSE2/3DNow), try
>>>> disabling assembly loop support by setting the environment variable
>>>> NOASSEMBLYLOOPS (to anything) before starting mdrun.
>>>>
>>>> Cheers,
>>>>
>>>> Erik
>>>>
>>>> On Feb 17, 2005, at 2:00 PM, Kirill Titievsky wrote:
>>>>
>>>>> Dear All,
>>>>>
>>>>> Gromacs seems to just stop running in the middle of half of my
>>>>> simulations.  No error messages.  Where do I start debugging?  Is
>>>>> there a debug reporting level in Gromacs?
>>>>>
>>>>> Symptoms:
>>>>> The logs get cut off in the middle of a line.  It appears that this
>>>>> has something to do with atoms ending up outside the box -- the
>>>>> trajectory files show some files to be outside and restarting the
>>>>> simulation shows NaN forces on some atoms, even though I'm using
>>>>> soft,
>>>>> finite potentials.
>>>>>
>>>>>
>>>>> Many thanks
>>>>>
>>>>> Kirill Titievsky
>>>>> Chemical Engineering
>>>>> MIT
>>>>> _______________________________________________
>>>>> gmx-users mailing list
>>>>> gmx-users at gromacs.org
>>>>> http://www.gromacs.org/mailman/listinfo/gmx-users
>>>>> Please don't post (un)subscribe requests to the list. Use the
>>>>> www interface or send it to gmx-users-request at gromacs.org.
>>>>
>>>>
>>>>
>>> _______________________________________________
>>> gmx-users mailing list
>>> gmx-users at gromacs.org
>>> http://www.gromacs.org/mailman/listinfo/gmx-users
>>> Please don't post (un)subscribe requests to the list. Use the
>>> www interface or send it to gmx-users-request at gromacs.org.
>>
>>
>>
> _______________________________________________
> gmx-users mailing list
> gmx-users at gromacs.org
> http://www.gromacs.org/mailman/listinfo/gmx-users
> Please don't post (un)subscribe requests to the list. Use the
> www interface or send it to gmx-users-request at gromacs.org.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: smime.p7s
Type: application/pkcs7-signature
Size: 2363 bytes
Desc: not available
URL: <http://maillist.sys.kth.se/pipermail/gromacs.org_gmx-users/attachments/20050218/8803d611/attachment.p7s>