Neighbor listing of a triclinic box on GPU

Hi,

This is an error regarding GPU implementation of neighbor listing. My simulation box is triclinic as I deform the box but lammps does not have neighbor listing for triclinic box for GPU. I have read about Axel saying, on a similar problem, neighbor listing be built on the CPU.

Could you please explain what the complications are in having Lees Edwards boundary conditions on the GPU itself.

Regards,

pk leishangthem

http://www.jncasr.ac.in/sastry/group.htm, Bangalore, India

If the code is saying it doesn’t do it, then it means

no one has coded it yet. As Axel said, if you are

talking about the GPU package, then it has options

for building the neigh lists on the CPU vs GPU, so

just use the CPU option for now.

Steve

Hi Steve,

Thank you for your reply.

Wanted to get help from the developers to avoid the switching between CPU and GPU just for neigh list because

  1. Lammps already has neigh list on GPU for a cubic box and
  2. if I do call cpu for neigh list, it’s slow and computationally costly.
    So, I was curious to know if there was any nontrivial limitation in case of triclinic for which lammps has a GPU restriction for triclinic… Would see if this can be coded for GPU. Your take on this would be very useful!
    Sorry if you find my post to be repeating.

Warm Regards,
pk

If the code is saying it doesn’t do it, then it means
no one has coded it yet. As Axel said, if you are
talking about the GPU package, then it has options
for building the neigh lists on the CPU vs GPU, so
just use the CPU option for now.

Steve

If the code is saying it doesn’t do it, then it means

no one has coded it yet. As Axel said, if you are

talking about the GPU package, then it has options

for building the neigh lists on the CPU vs GPU, so

just use the CPU option for now.

Steve

Stan could comment better if there are limitations on triclinic

neigh lists for GPU. I think its just a matter of someone doing

the coding.

Steve