Adaptive Mesh Refinement GPU Implementation using Dynamic Memory Arrangement for Interface Advection by Conservative Phase Field Method

摘要

Adaptive Mesh Refinement (AMR) is a widely used method in computational uid dynamics (CFD) to save the memory usage and reduce the number of elements of the computational domain, in order to enlarge the scale of the domain as well as increase the performance. The highly parallelism computation power of graphics processing units (GPU) have been widely utilized in accelerating the computations. However, due to the di culties and the big overheads of dynamic changes of the AMR blocks during the simulation, an e cient AMR implementation using GPU is a challenging task. In this paper, we present an e cient GPU implementation of tree-based AMR using dynamic memory arrangement and apply it to an interface advection problem, solving by a conservative phase eld method. Performance measurement showed an obviously 16.87 speedup of our GPU implementation compares to a OpenMP CPU implementation.