GPU implementations for MOOSE

by Wenyan for International Neuroinformatics Coordinating Facility

Two solvers, one for single neurone calculations and one for chemical reaction- diffusion calculations, are both computationally demanding. They are heavily used in MOOSE project and they affect the overall computational efficiency. Previously a basic GPU implementation has been done but its calculation doesn’t give any speedup compared to serial calculation on CPU. A scalable and optimised GPU parallelisation algorithm needs to be implemented based on the previous work.