Grid Digital Elevation Model ( DEM) has advantages in structure, storage, management and analysis relative to Triangular Irregular Network ( TIN) . However, traditional serial converting algorithm from TIN to grid DEM has lower efficiency. To improve the efficiency, Graphic Processing Unit ( GPU) parallel computation is used to implement the traditional algorithm. Then, the algorithm is optimized with respect to GPU memory access. At last an experimental system is developed by using C++ and CUDA ( Compute Unified Device Architecture) to verify the algorithm. Results show that the efficiency is improved 72 times after the optimization.%基于规则格网的数字高程模型( DEM)相对于不规则三角网( TIN)具有结构简单,便于存储、管理和分析等优点。针对TIN向规则格网转换的串行算法效率较低的问题,利用图形处理器( GPU)并行编程对一种串行算法进行实现;然后从GPU 全局内存和共享内存的访问方面对算法进行优化;最后用C++语言和统一计算设备架构( CUDA)开发了实验系统,对优化前后算法的效率进行对比。结果表明,优化后的算法效率较优化前最大提高了72倍。
展开▼