An efficient vectorization of linked-cell particle simulations

Molecular dynamics simulations for short-range potentials represent an important class of applications in scientific computing. While a lot of work has been spent on the efficient implementation of such simulations on vector machines in general, not much effort has been invested into the efficient implementation for current x86 processor architectures' SIMD extensions such as SSE and AVX. We describe an implementation of the linked-cell algorithm for the SSE and AVX instruction set, which achieves the theoretical limit for SSE. Moreover, the proposed scheme will allow the efficient usage of future architectures with wider vector units. We implemented the kernel using intrinsics within a small test program and conducted a number of runs for different setups of the Lennard-Jones fluid on an Intel- and AMD-based cluster, respectively.