基于CUDA并行的全球海洋表面温度场等值线提取算法研究

Research of parallel global sea surface temperature contours extraction algorithm on CUDA platform

在分析GPU并行架构和CUDA灵活可编程性基础上,提出了一种基于区间块搜索的等值线并行提取方法,可应用于全球海洋表面温度格网数据的分析.算法有效减少了等值线追踪过程中重复性的格网遍历及不必要的格网搜索.最后,实验采用了不同规模的海表温度场格网数据进行等值线的提取并比较串并行耗时,结果表明:(1)算法能实现全球海洋表面温度等值线的有效提取并提高其效率,尤其对于大规模格网数据;(2)基于所有实验数据,格网规模大于720×1 440时,相较于串行过程GPU执行体现了其效率上的优势,GPU与CPU串行耗时的加速比最大可达3.124.

Based on the analysis of GPU＇s parallel architecture and CUDA＇s flexible programmability, a contour ex- traction algorithm applied to gridded SST （Sea Surface Temperature） data implemented through block searching is proposed. This algorithm builds the＂block＂to reduce the times of grid traversal and avoids the unnecessary search of excluded cells effectively during the contour tracing procedure. Finally, the gridded products of different sizes are experimented. The results of contours extraction and time-consuming comparison show （1） the effectiveness and ef ficiency of the algorithm are improved especially for the large scale gird and （2） according to experimental data, GPU acceleration ratio increases with the grid＇s scale to 3. 124 in maximun, validating its novel advantage compared to CPU when the grid＇s scale is barger than 720×1 440.