GPU-accelerated adaptive particle splitting and merging in SPH

Graphical processing unit (GPU) implementation of adaptive particle splitting and merging (APS) in the framework of smoothed particle hydrodynamics (SPH) is presented. Particle splitting and merging process are carried out based on a prescribed criterion. Multiple time stepping technology is used to reduce computational cost further. Detailed implementations on both single- and multi-GPU are discussed. A benchmark test that is a flow past fixed periodic circles is simulated to investigate the accuracy and speed of the algorithm. Comparable precision with uniformly fine simulation is achieved by APS, whereas computational demand is reduced considerably. Satisfactory speedup and acceptable scalability are obtained, demonstrating that GPU-accelerated APS is a promising tool to speed up large-scale particle-based simulations.