Hello everyone,
As it is known LAMMPS approach makes use of spatial decomposition , allowing the balance to manually when necessary . My question is: LAMMPS uses decomposition atom and forces at some time or particular operation ? if so, when and why? .
regards
Hello everyone,
As it is known LAMMPS approach makes use of spatial decomposition , allowing
the balance to manually when necessary . My question is: LAMMPS uses
decomposition atom and forces at some time or particular operation ? if so,
when and why? .
decomposition across atoms is done in threaded compute kernels inside
the subdomain, i.e. GPU, KOKKOS, USER-INTEL and USER-OMP packages.
those usually decompose the outer pairwise loop over atoms, and some
also decompose the inner loops across thread blocks.
the answer to why is simple. it is simple to do, the data is readily
available, and since it is orthogonal spatial decomposition, it
complements the (coarse grained) message passing parallelization.
is this the kind of answer you are looking for? otherwise, please
explain in more detail, why you are asking.
axel.
Yeah, thanks for your reply.