Hardware Acceleration of Reaction-Diffusion Systems: A Guide to Optimisation of Pattern Formation Algorithms Using Openacc

Reaction Diffusion Systems (RDS) have widespread applications in computational ecology, biology, computer graphics and the visual arts. For the former applications a major barrier to the development of effective simulation models is their computational complexity - it takes a great deal of processing power to simulate enough replicates such that reliable conclusions can be drawn. Optimizing the computation is thus highly desirable in order to obtain more results with less resources. Existing optimizations of RDS tend to be low-level and GPGPU based. Here we apply the higher-level OpenACC framework to two case studies: a simple RDS to learn the `workings' of OpenACC and a more realistic and complex example. Our results show that simple parallelization directives and minimal data transfer can produce a useful performance improvement. The relative simplicity of porting OpenACC code between heterogeneous hardware is a key benefit to the scientific computing community in terms of speed-up and portability.