一个SPH流体实时模拟的全GPU实现框架  

A FULL GPU IMPLEMENTATION FRAMEWORK OF SPH FLUID REAL-TIME SIMULATION

作　　者：郭秋雷[1] 唐逸之[1] 刘诗秋[2] 李桂清[1] 

机构地区：[1]华南理工大学计算机科学与工程学院,广东广州510006 [2]华南理工大学软件学院,广东广州510006

出　　处：《计算机应用与软件》2011年第11期69-72,共4页Computer Applications and Software

基　　金：国家自然科学基金项目(60973084)

摘　　要：怎样实时地进行高度逼真的大规模流体模拟是图形学要研究的一个重要内容。流体的模拟由物理计算、碰撞检测、表面重构和渲染几个部分组成,因此有大量工作针对流体模拟中的各个部分算法进行GPU加速。提出一整套基于GPU的SPH流体模拟加速框架。在利用平滑粒子动力学(SPH)求解Navier-Stokes方程的基础上,借助基于GPU的空间划分PSS(Parallel Spatial Subdivision)来大幅度提升粒子碰撞的检测速度。同时,设计一种基于几何着色器(Geometry Shader)的流体表面信息的重建算法,并进一步地实现基于索引的优化,使得在流体表面重建过程无须遍历不包含表面的区域。实验结果表明,该方法能实时模拟出具有较好真实感的流体场景。How to implement timely the high realistic imitation of large-scale fluid simulation is an important element in computer graphics research. Fluid simulation consists of quite a few components including physical calculation, collision detection, surface reconstruction and rendering, so that there are a lot of works with regard to GPU acceleration aiming at the algorithms of each component of the fluid simulation. This paper proposes a set of GPU-based framework for SPH fluid simulation acceleration. On the basis of employing smoothed-particle hydrodynamics to solve Navier-Stokes equation, we speed up greatly the particle collision detection with GPU-based parallel spatial subdivision. Meanwhile, we design a fluid surface information reconstruction algorithm which uses the geometry shader, and the index-based optimisation is further carried out, which makes the reconstruction process of fluid surface avoid from traversing those areas do not contain surfaces. Experimental results show that the method in this paper can simulate timely the fluid scene with perfect reality.

关 键 词：流体模拟 GPGPU 实时 SPH 并行空间划分 几何着色器 MARCHING CUBES 

分 类 号：TP391.77[自动化与计算机技术—计算机应用技术]