面向Ghost光滑粒子动力学模拟的图形处理器快速邻居搜索算法

Fast neighbor search on GPU for Ghost SPH simulation

提出了一种全新的快速邻居搜索方法,该方法可提高基于光滑粒子动力学的流体模拟在图形处理器上的运行效率.此外,这种新的邻居表建立方法可以对两种或者两种以上的粒子进行邻居搜索,使所有粒子能在同一背景网格下拥有独立的粒子属性.在此基础上,引入了Ghost边界粒子以加强光滑粒子动力学方法在边界模拟上的准确性,从而使流体模拟更加真实.实验证明,与传统的基于图形处理器的光滑粒子动力学模拟相比,本文方法效率更高.

This paper presents a novel fast neighbor searching method. By using this method,fluid simulation based on smooth particle hydrodynamics（ SPH） can be parallelized easily and run on graphic processing unit（ GPU） with high efficiency. The neighbor searching method can search two or more kinds of particles,while saving their information in the same background grid. Ghost boundary particles are introduced to improve the accuracy of boundaries,which can enhance the fidelity of the fluid simulation. Experiments show that the proposed method is more efficient than the traditional SPH method based on GPU.