基于GPU的GRAPES数值预报系统中RRTM模块的并行化研究

GRAPES(Global and Regional Assimilation and Prediction System)是由中国气象科学研究院自主研究开发的中国新一代数值天气预报系统，由于其处理的数据量非常庞大以及对实时性的要求较高，因此一直是并行计算领域研究的热点。首次运用GPU(图形处理器)通用计算及CUDA技术对CRAPES_Meso。模式中物理过程的RRTM(快速辐射传输模式)长波辐射模块进行并行化处理。在性能分析的基础上，针对GPU体系结构的特点，从代码优化、存储器优化、编译选项等方面对程序性能进行优化，并取得了14X倍的加速比。经过测试表明，长波辐射RRTM模块在GPU上并行计算过程正确、稳定而且有效，并为GRAPES系统未来在GPU平台上的并行化发展奠定了一定的基础。

GPU-based Parallel Rearches on RRTM Module of GRAPES Numerical Prediction System

GRAPES(Global and Regional Assimilation and Prediction System) is a new generation of numerical weather prediction(NWP) system of China. As the system processes amount of data and requires high real-time, it is always a hot research field of parallel computing. This is the first time that we use GPU(Graphics Processor Unit) general-pur-pose computing and CUDA technology on RRTM(Rapid Radiative transfer model) long-wave radiation module of GRAPES Meso model for parallel processing. Based on the analysis of computing performance, and according to the characteristics of the GPU architecture, the RRTM module parallel computational efficiency was optimized from the aspect of code tuning,memory,compiler options and etc. The optimization results indicate that the performance obtains a speedup of 14.3 X.Experiments were carried out on the GPU platform. The results show that the parallel computing algorithm is correct, stable and efficient for operational implementation of GRAPES in near future.