基于CUDA汇流分析并行算法研究与实现.docVIP

基于CUDA汇流分析并行算法研究与实现 　　摘 要:针对基于数字高程模型(DEM)生成流域等流时线的快速运算问题,提出了一种基于统一设备计算架构(CUDA)平台同时可发挥图形处理器(GPU)并行运算特性的汇流分析的快速并行算法。采用改进后的归并排序算法进行数据排序及新的内存分配策略和改进的并行算法进行汇流分析。用该并行算法和CPU上的串行算法, 对生成基于DEM的等流时线运算时间和矩阵乘法运算时间进行分析验证。实验结果表明,基于CUDA的汇流分析并行算法能提高系统的计算效率,具有较好的效果。 　　关键词:并行计算;图形处理器;统一设备计算架构;汇流分析;数字高程模型 　　中图分类号:TP391;TP301.6文献标志码:A 　　文章编号:1001-3695(2010)07-2445-03 　　doi:10.3969/j.issn.1001-3695.2010.07.011 　　 　　Research and realization in parallel algorithm of??confluence analysis based on CUDA 　　 　　ZHAO Xiang-hui????1,2, MIAO Qing?? ??1,2, FU Zhong-liang????1,2, SU Chang????1,2, LI Xin????1,2 　　(1.Chengdu Institute of Computer Applications, Chinese Academy of Sciences, Chengdu610041, China;2.Graduate School, Chinese Academy of Sciences, Beijing 100049, China) 　　 　　Abstract:Aiming at the fast parallel computing of generating isochrones of watersheds that based on digital elevation model(DEM),this paper proposed a fast parallel algorithm of confluence analysis based on compute unified device architecture (CUDA) platform that could use parallel computing of graphic processing unit(GPU).Carried data sorting out by using the improved merge sorting algorithm,adopted the new memory allocation strategy,carried confluence analysis out by using the improved parallel computing algorithm. Using the presented parallel algorithm and the serial algorithm based on CPU to analyze and verify the time consumed when generating isochrones of watersheds based on DEM and executing matrix multiplication. The experiments results illustrate that this parallel algorithm of confluence analysis based on CUDA can improve the computational efficiency of the system and have a better effect. 　　Key words:parallel computing; graphic processing unit(GPU); compute unified device architecture(CUDA); confluence analysis; digital elevation model(DEM) 　　CUDA是NVIDIA公司提供的GPU用于通用计算的开发环境,是一个全新的软硬件架构,可以将GPU视为一个并行数据计算的设备,对进行的计算进行分配和管理。图形显示卡,尤其是高端的显卡,现在几乎是单机必配的重要硬件。其核心部件GPU采用 SIMD体系结构,大量的计算部件使得它很适合处理高数据量的并行运算[1~3]。