GPU Accelerated Unstructured Mesh Based AMR Method:Investigation And Application

A GPU accelerated inviscid flow solver and its condensation extension (VA2DG and VA2DGC) are developed on an unstructured quadrilateral grid with the cell-based adaptive mesh refinement (AMR) in the present work. The AMR method is one of the most important methods in Computational Fluid Dynamics. Researchers had realized the AMR solver on GPU before and achieved remarkable performance enhancement. However, they avoided the problem of implementing AMR processing on GPU, which may lead the data exchange frequency between GPU and CPU increasing and drag the performance. Thus, for the first time, in present work, the cell-based AMR is fully implemented on GPU for the unstructured grid, which greatly reduces the frequency of data exchange between GPU and CPU. Specifically, the AMR is processed with atomic operations to parallelize list operations, and null memory recycling is realized to improve the utilization efficiency of memory.Three typical two-dimensional flow problems are then simulated with the solver running on CPU (Intel E7300) and on GPUs (Geforce GT9800and Tesla C2050) for comparison. It is found that results obtained by GPUs agree very well with the exact or experimental results in literature. An acceleration ratio of8is obtained between the parallel code running on the old GPU GT9800and the serial code running on E7300. With the optimization of configuring a larger L1cache and adopting Shared Memory based atomic operations on the newer GPU C2050. an acceleration ratio of40is achieved. The parallelized cell-based AMR processes have achieved3x speedup on GT9800and28x on Tesla C2050, which demonstrates that parallel running of the cell-based AMR method on GPU is feasible and efficient. Our results also indicate that the new development of GPU architecture benefits the fluid dynamics computing significantly.With the validation of the VA2DG solver, its expansion version with conden-sation, the VA2DGC solver is adopted for simulating vapor condensation problems in compressible flow in this study. The shock tube problem with condensation is simulated for validating reliability of the solver. The flow in Prandtl-Meyer expan-sion fan with condensation is simulated in two scales:a10m x10m and a100m x100zone. The wave configuration and nucleation area’s distribution in space are investi-gated. The shocks induced by condensation are found distributing discontinuously: one’s head stacking on the next one’s tail. The nucleation areas have irregular pro- tuberances and the increasing rate of these protuberances’distances to the corner is found biquadratic.