一种基于GPU的粒子系统火焰模拟

Fire Simulation with GPU-based Particle System

针对传统火焰模拟耗时、模拟真实性不理想这一问题,提出了一种借助GPU(图形处理器)高通用计算能力进行并行模拟的粒子系统火焰模拟方法。该方法采用基于方位角和仰角的粒子散射器、基于层流火焰轮廓计算公式的火焰外形计算及通过拉格朗日插值方法平滑火焰骨架线等手段提高了紊流火焰的模拟真实度。在提高性能方面,该方法使用全局存储空间存储粒子信息,避免了因使用纹理存储而产生的反复绑定的开销,同时结合GPU强大的通用计算能力,通过采用CUDA(计算统一设备架构)编写的并行算法,实现了基于GPU的拉格朗日插值并行求解及并行计算、更新粒子属性,并就如何确定块内线程数量作了论述。实验结果表明,该方法满足了火焰模拟的真实性和实时性要求,且较传统方法有很大的提高。

This approach aims at improving the performance and reality of fire simulation in virtue of powerful ability of GPUPU. An emitter composed of yaw and pitch was designed to control the particle stream precisely. The formula of laminar flame was modified to outline the shape of turbulent fire. Besides, Lagrange interpolation was used to smooth and get the accurate fire skeleton around which particles move even when it becomes twist. In order to improve the per- formance,global memory was adopted to store particles to prevent the spending of binding texture memory repeatedly; parallelization of Lagrange interpolation and properties of particles update were realized of CUDA. All these measures contribute to reach a desirable real-time simulation speed and acquire an improvement of performance.