AutoPas is an auto tuner for N-body simulations. Kokkos is a library that offers hardware independency through abstractions. The goal of this thesis was, an integration of Kokkos into AutoPas to improve performance portability.

This integration for now is limited to the KokkosLinkedCells container, which uses a central particle storage and an index based metadatastructure to link particles to cells in the simulation domain.

However, the implementation for this thesis was not finished to a point where it could be run on systems other than conventional multi core systems. Instead, the performance of the new KokkosLinkedCells container is compared to that of the regular LinkedCells container to determine the overhead that is introduced by this additional layer of abstraction. Overall, both containers perform very similarly in scenarios where work is evenly distributed among all threads, but the absence of algorithmic load balancing or dynamic scheduling harms the performance of the KokkosLinkedCells container in scenarios where particles are focused in one part of the simulation domain.