compute pe/atom error with gpu package

LAMMPS LAMMPS Mailing List Mirror
1

Dear all,

I am trying to print out the per-atom energies to a dump file using GPU package with the following computes:

compute 1 all pe/atom
dump 2 all cfg 10000 dump.compre_*.cfg id type xs ys zs c_KE c_1

Screen output:

LAMMPS (9 Aug 2013)
Reading data file …
orthogonal box = (0.0334666 -30 0.286126) to (54.3803 300 543.852)
1 by 2 by 3 MPI processor grid
500000 atoms
500000 velocities
Changing box …
orthogonal box = (0.0334666 -100 0.286126) to (54.3803 400 543.852)

in.deform (1.76 KB)

2

Dear all,

I am trying to print out the per-atom energies to a dump file using GPU
package with the following computes:

compute 1 all pe/atom
dump 2 all cfg 10000 dump.compre_*.cfg id type xs ys zs c_KE c_1

[...]

And without that pe/atom compute, everything works well. I think GPU package
does support this "compute". Is that true? Or any way to solve this problem?
I'm attaching my input script(in.deform). Please let me know if you need
any additional information.

attaching the input script without the corresponding data file is
pretty much useless, since it is not possible to repeat your
calculation.

since you specifically mention the GPU package, have you tested the
input running just on the CPU?

axel.

3

All of the computes and fixes should work with the GPU package. In some cases, you have to turn off GPU neighbor list builds, but in those cases, the error message should tell you what to do.

I have verified your issue, but it is not a problem with the GPU package. It will happen when running a simulation with compute pe/atom, newton off, and no kspace solver. You can verify this by adding ‘newton off’ to your CPU-only run.

Steve can probably produce a fix for this quickly (it is line 150 of compute pe/atom, access to force->kspace).

Best,

  • Mike
4

I think I fixed this - will be in a 18Sep patch later today …

Thanks,
Steve