摘要
球体退化八叉树格网(Spherical Degenerated Octree Grid,SDOG)及其三元组模型为全球变化及地球系统科学研究提供了一种地表及其内外空间统一表达与建模的全球空间数据模型。然而,现有SDOG模型可视化效率较低,难以满足全球模式下空间数据快速可视化的要求。该文采用GPU并行计算技术实现SDOG模型可视化加速。通过分析SDOG模型可视化过程的并行特征,选取格网码(C)解码及三元组渲染两个过程进行GPU并行化,分别利用数据并行及直接显存渲染策略实现GPU加速;设计了两组实验进行测试,结果表明:1)GPU并行对C解码过程具有明显的加速效果,且随数据量增长加速效果更加突出,加速比达15倍;2)受系统构建、数据传递等因素制约,GPU并行对C解码过程的加速效果与GPU性能呈正相关,但不呈正比;3)直接显存渲染优化策略可有效提高SDOG的GPU并行渲染效率,渲染速度可提高1.5倍。
Spherical Degenerated Octree Grid （SDOG） and its triple model provide a new global spatial data model for integrated representation and modeling of the inner and outer of planet Earth, and new tools are developed for global change and Earth System Science （ESS） research. However, the efficiency of massive SDOG visualization is a little low, and can not satisfy the requirements of rapid rendering of global spatial data. The GPU parallel computing technology was employed to accelerate the speed of SDOG visualization in this paper. The parallel characteristics of SDOG visualization process is analyzed, and the grid code （C） decoding and triple rendering are selected to make parallel implementation on GPU, and video memory is adopted for the two GPU parallel implementations. Two experiments were carried out to test the performance of GPU acceleration. Results show that： 1） obvious acceleration is reached in the C-decoding process with GPU,and the acceleration will be strengthened as data size increase,the acceleration ratio reach 15; 2） owing to the factors of the construction of system environment and data transferring, the accelerating performance is positively correlative with the GPU performance, but got linear correlation; 3） the GPU paralleling process of SDOG visualization can be accelerated by direct rendering on video memory, and the rendering speed can be increased 1. 5 times.
出处
《地理与地理信息科学》
CSCD
北大核心
2013年第4期72-76,共5页
Geography and Geo-Information Science
基金
国家自然科学基金重点项目(40930104)
国家863计划项目(2011AA120302)
