Acceleration Method Of Fluid Simulation Using GPU

Fluid phenomena(such as smoke, fire, clouds, waves, bubbles, explosions, etc.) is a very common scene in daily life. Fluid simulation is widely used in movie special effects, television advertisements, online computer games and other graphics related fields. It is a very important research subject of computer graphics. In recent years, fluid simulation mainly uses physics-based method, and this method can produce vivid images of fluid phenomena. However, the calculation process of the physics-based method is very complex, results in simulating real-time fluid animation is rather difficult. Therefore, how to improve the simulation speed of fluid animation has become an urgent research topic.This paper first studied physics-based fluid simulation method. Then, proposes a new parallel algorithm for conjugate gradient method (CGM) which implements on CUDA to speed up the most time-consuming part of the N-S Equation’s solution-the solution of the pressure term and the diffusion term-which two terms are actually Poisson Equation. We mainly design efficient parallel algorithms for the most time-consuming operations of the CGM:matrix-vector multiplication and inner-product.A matrix-free formulation of matrix-vector multiplication algorithm is proposed, which can overcome the difficulty in handling the Neumann boundary condition to provide a high degree of parallelization, thus tremendously reducing the computing time. For the inner-product a generalized sum reduction algorithm is proposed, which can reasonable reduce the synchronization operations of the algorithm, resulting in less computing time. The experiments show that this algorithm can effectively accelerate the speed of the Poisson Equation’s solver, to some extent,providing a faster way to simulate fluid phenomena.