A GPU-accelerated meshless method for two-phase incompressible fluid flows

Abstract

This paper presents the development and implementation of a Meshless twophase incompressible fluid flow solver and its acceleration using the graphics processing unit (GPU). The solver is formulated as a Localized Radial-basis Function Collocation (LRC) Meshless method and the interface of the two-phase flow is captured using an implementation of the Level-Set method. The Compute Unified Device Architecture (CUDA) language for general-purpose computing on the GPU is used to accelerate the solver. Through the combined use of the LRC Meshless method and GPU acceleration this paper seeks to address the issue of robustness and speed in computational fluid dynamics. The LRC Meshless method seeks to mitigate the issue of extensive and time-consuming user input of mesh-based methods by representing the field variables on a set of scattered points that need not meet stringent geometric requirements such as connectivity and poligonalization. The method is shown to render very accurate and stable solutions and the implementation of the solver on the GPU is shown to accelerate the solution process significantly. © 2013 WIT Press.