GPU accelerated SPH simulation of fluids for VFX

Fluids are important to the Visual Effects Industry but extremely hard to control and simulate because of the complexity of the governing equations. Fluid solvers can be divided into two categories, those of the Eulerian point of reference and those of the Lagrangian. Both categories have different advantages and weaknesses and hybrid methods are popular. This thesis examines Smoothed Particle Hydrodynamics, a Lagrangian method for physically based uid simulations. To allow the artist the exibility given by shorter simulation times and increased number of iterations, the performance of the solver is key. In order to maximize the speed of the solver it is implemented entirely on the GPU, including collisions, volumetric force fields, sinks and other artist tools. To understand the implementation decisions, it is important to be familiar with the CUDA programming model. Thus, a brief explanation of CUDA is given before the exact implementation of the methods are explained. The results are presented along with a performance comparison as well as a discussion of the different parameters which can be fed to the solver. Some thoughts on possible future extensions can be found in the conclusion