[BUG#1-remedy_available-LIGGGHTS] PARTICLES FALL THROUGH STL WALL

ckloss's picture
Submitted by ckloss on Thu, 08/04/2011 - 15:27

Bug description:
Particles falling through STL walls - though they should not

Applies to:
Reported for LIGGGHTS 1.2.9, 1.3.3, 1.4.1 only if compiled with icpc 12

Comments:
Compiler related: Occurs with Intel compiler icpc, version 12, works fine with version 11 and older.
After talking to a number of experts, the conclusion is that this is a compiler bug.

Remedy:
Compile with "-fstrict-aliasing" if using Intel icpc 12

Reported by:
Christine Bertsch and Xuetao Wang (Züblin), Klaus Straka and Michael Aigner (JKU), Rami Jabakhanji

Status :
(4 August 2011): Rami Jabakhanji reports he has maybe resolved this issue and is tracking down which steps resolve it
(31 August 2011): Bug is now reproducible on my machine using icc12, which has been identified by Christine Bertsch (Züblin) to reproduce the issue.
(2 September 2011): Bug tracked down, see above for suggested remedy

Forums:

marc | Fri, 08/05/2011 - 18:32

Hi There,

I had the same problem. I don't know if this helps, but what I did was basically change the StL mesh refinement.

I had a coarse StL mesh and the particles were passing through. I went back to my CAD, generated a more refined mesh and ran the simulation again. Apparently the problem disappeard.

No other parameter was changed. I'm using 1.3.3, running on 4 cores shared memory.

Best Regards,

Marc

ckloss's picture

ckloss | Mon, 08/08/2011 - 05:05

Hi Marc,

can you please comment on which system you are running on (OS, compiler, MPI, makefile settings)?

Thanks,
Christoph

studdicb | Mon, 08/08/2011 - 17:13

Hi all,

we did some tests on compilerversions here as we changed from intel 11 to 12 when changing from 1.2.9 (no problems) to 1.3.X.
Our tests show that using intel compiler 11.1.072 the problems do not occur with liggghts 1.4, while using the newer compiler version intel 12.0.4 liggghts shows this strange behaviour.

Maybe anyone else knows anything about the changes within these compiler versions?

Cheers,
Christine

ckloss_ | Mon, 08/08/2011 - 18:25

thanks, that's a very valuable hint! do the problems also occur with the tutorials, such as chute_wear?

thanks,
Christoph

studdicb | Wed, 08/10/2011 - 08:41

Hi Christoph,

I tested the chute_wear example - using the icc12 compiled version of liggghts 1.4 the particles fall directly through the stl-mesh, while using the same liggghts version compiled with icc11 it works fine.

regards,
Christine

ckloss_ | Wed, 08/10/2011 - 17:58

Ok thanks - so I guess for you personally the issue is resolve for now by using icc11 ?
I will work on resolving that issue as I am back in office in sept
Christoph

studdicb | Thu, 08/11/2011 - 15:57

Well, actually we would like to use the new compiler ;) For now it is ok, anyway it would be interesting to know the reasons for the different compiler behaviour. But of course this is not that urgent.
Christine

marc | Tue, 08/09/2011 - 15:49

Hi Cristoph, sorry for the late reply. The requested info are below:

OS: CentOS 5.6
System: VirtualBox 4.0.12 on i7-920 (x64)
MPI: OpenMPI 4.3
Liggghts: 1.3.3 with FFTW (2.1.5)
Command Line: mpirun -np 4 lmp_openmpi < input_file

Best regards,

Marc