Research On Acceleration Of OpenFOAM Based On GPU

Computational Fluid Dynamics (CFD) is a subject extensively used in many research and industry area. It demands amount of calculation and takes long time in solving complex CFD problem with high precision requirement. On the other hand, the recent rapidly developing General Purpose computing on GPU (GPGPU) technology makes it possible to manipulate the GPU with strong computational capability, high memory bandwidth and relatively low cost for general usage. So the research on how to accelerate CFD computations using GPU power is of great practical significance.The advantage of GPGPU is the massive parallel hardware capability and good programmability. This dissertation investigates the hardware foundation of GPGPU technology and takes Compute Unified Device Architecture (CUDA) as an example to discuss the programming model on GPU.OpenFOAM is a widely used well designed open source CFD software framework with good expandability. This dissertation analyzes the design feature of OpenFOAM and demonstrates the usage and customizing method of the software.After profiled the hotspots of a typical application icoFoam in OpenFOAM, this thesis research some parallel sparse linear system computation methods on GPU and represents a GPU plug-in for OpenFOAM which provides serveral GPU accelerated sparse linear equations solvers. The GPU parallel methods achieved good speedup and proved to be correct and efficient.