基于GPU加速的雷暴追踪外推方法研究  

Research on Thunderstorm Track and Extrapolation Based on GPU-Acceleration

作　　者：王兴[1] 王新[2] 苗春生[1] 王介君[1] 

机构地区：[1]南京信息工程大学大气科学学院 [2]北京华风气象影视信息集团公司

出　　处：《南京师范大学学报（工程技术版）》2015年第1期35-42,共8页Journal of Nanjing Normal University(Engineering and Technology Edition)

基　　金：国家科技支撑计划课题(2012BAH05B01);公益性行业科研计划课题(GYHY201206030);大学生实践创新训练计划项目(201410300119)

摘　　要：基于气象雷达的雷暴识别与追踪是临近预报中重要的方法之一.为解决传统算法实时性差的问题,运用Open CL构建异构计算模型对算法进行并行化改进.通过对算法分支结构优化、Open CL设备内存优化,以及针对VLIW的优化,分步阐述算法优化的过程和原理.这些方法不仅使得基于光流的计算速度大幅提升,还可为其他基于Open CL异构计算的优化提供参考.以AMD两代不同架构的GPU和Intel XEON CPU作为测试平台测试,结果表明,改进后的算法程序在硬件同等功耗的情况下,计算速度提高了10至18倍.The thunderstorms recognition and tracking based on weather radar is one of the important methods in weather nowcasting. In order to solve the problem that the traditional algorithm is poor real-time,this paper discusses the heterogeneous computing model based on GPU,and presents that parallel the algorithm with OpenC L to achieve high performance. By the methods of branching structure optimization,OpenC L device memory optimization,and optimized for VLIW,this paper expounds the optimization of the algorithm step by step. These methods not only boost the speed of optical flow computation,but also provide a reference for other optimization based on OpenC L heterogeneous computing. We use AMD＇s two generations of different architecture of GPU and Intel XEON CPU as a test platform. The tests results show that the computing performance improve 10-18 times under the circumstance of same power consumption.

关 键 词：光流计算 开放运算语言 并行化 临近预报 雷达资料 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]