Re: [ESPResSo-users] Error prompt in lbgpu.cu

[Axel Arnold]

On Tuesday 03 January 2012, Farnoosh Farahpoor wrote:
> I forgot to write the last line of the error: "CUDA error: unspecified
> launch failure"

That is the equivalent of a segfault. reset_particle_force is the first check 
for errors after the asynchronous force calculation, so the segfault can 
happen anywhere during the force calculation, e.g. since a particle got a 
position like NaN or so - there are plenty of possibilities. Since the error 
seems to occur at different times, it is quite likely that your system is not 
properly set up and is unstable, which of course shows up much faster if you 
increase the number of particles. You might want to reduce the time steps or 
viscosities.

Many regards,
Axel

> On Tue, Jan 3, 2012 at 8:02 AM, Farnoosh Farahpoor 
<address@hidden>wrote:
> > Dear Espresso Developers
> > 
> > In my simulation with CUDA version of ESPResSo (developers version on
> > git) when I increase the number of particles in my system, sometimes
> > after a while I get this error:
> > "error calling reset_particle_force with #thpb 64 in lbgpu.cu:1534"
> > 
> > Do you know what is the reason? Is there any problem with my code? (I get
> > the error with no order in time, in random time steps and on different
> > machines, frequently. The worst thing is that I face with a crash in
> > system without any error prompt on one of the machines.)
> > 
> > Thank you in advance
> > Farnoush Farahpour


-- 
JP Dr. Axel Arnold      Tel: +49 711 685 67609
ICP, Universität Stuttgart      Email: address@hidden
Pfaffenwaldring 27
70569 Stuttgart, Germany