An Improved Study of Physically Based Fluid Simulation on GPU

A feasible routine for complex fluid simulation is presented, consisting of the pre-processing and runtime stage. The pre-processing stage generates all computation-aided textures, which speeds up the simulating and rendering at run-time stage. We improve the unreasonable processing in most previous methods and give a correct discretized solution of Poisson equation. To improve the computational accuracy, the residuals are calculated to adoptively control the Jacobian iterations based on ping-pong technique and reduction operation. The potential parallelism of GPU is harnessed by effectively organizing the data and taking advantages of 4-element parallelism of vector operations. The experiment reveals that the proposed routine creates visually convincing results with encouraging FPS and meets the requirements of real-time simulation.