Scalable fluid simulation in linear time on shared memory multiprocessors

Abstract

We describe a framework for creating animated special effects such as dust, smoke and fire using a generalized transport system coupled with a fluid dynamics engine. The framework presents artists with an open system for animating volumes with generic transport properties such as diffusion rate, dissipation rate, and contribution to buoyancy and gradient forces. Artists may attach additional volumes representing source terms, external forces and collisions that control the fluid motion. By presenting all inputs to the fluid solver as volumetric data, we can take advantage of a fast, direct FFT-based elliptic solver whose performance and scalability is optimal on shared-memory multiprocessors and is dramatically faster than the best iterative methods.