Molecular Dynamics Simulations are a major field in high-performance computing, as simulations including many particles impose a large computational effort. It is especially important to optimize the force calculations between particles, which take up a large portion of the execution time. A common optimization includes SIMD instructions and SIMD registers to vectorize the force calculations.

AutoPas is a performance-focused particle simulations library in C++. This thesis presents an optimization strategy for selecting execution-time-efficient vectorization pattern in AutoPas. A vectorization pattern is the order in which particles are loaded into a SIMD-register. The optimization strategy, implemented in this thesis, is based on a short benchmark at the beginning of each simulation, which enables a dynamic vectorization pattern selection during the simulation based on the benchmark results. Finally, the performance of the benchmark-based vectorization pattern selection is compared to a different vectorization pattern selection strategy included in AutoPas.