Real-time rendering of large-scale terrain based on OpenCL

Abstract

In this paper, we propose a framework of CPU-GPU coupled computation based on OpenCL (Open Computing Language) for the real-time rendering of large-scale terrain datasets. Firstly, large-scale terrain datasets are divided into terrain chunks with the same size. Then appropriate terrain chunks are loaded into the host memory and the global memory of OpenCL device by 2-level caching mechanism and the quadtree hierarchy of LOD (Level of Detail) model is parallel constructed by OpenCL without intervention of the CPU and the multi-resolution terrain scene is rendered by OpenGL(Open Graphics Library). It is easy for OpenCL to develop program for general purpose computation in the GPU and greatly improve performance in term of execution time. Sharing data between OpenCL and OpenGL can save much memory and reduce data copy or movement. The experimental results have demonstrated that our method greatly reduces the CPU workload and balances the workload between CPU and GPU, and improves the efficiency of constructing of the LOD model and achieves higher frame rates.