The State of the LAMMPS KOKKOS Package

Upcoming supercomputers (NERSC Permutter, ORNL Frontier, and ANL Aurora) feature several different types of GPU hardware (from NVIDIA, AMD, and Intel). Using the Kokkos performance portability library [1] protects developers from having to maintain multiple versions of LAMMPS for each of these (and future) machines. In the LAMMPS KOKKOS package, C++ code is written once using Kokkos abstractions in a form that looks independent of the target hardware, and then target-specific code (i.e. CUDA, HIP or SYCL) is generated at compile-time. An overview of recent improvements to the LAMMPS KOKKOS package will be given, along with performance measurements on different hardware.

1. H. C. Edwards, C. R. Trott, and D. J. Sunderland, Kokkos: Enabling manycore performance portability through polymorphic memory access patterns, Parallel Distr. Com., 74, 12 (2014), 3202-3216.