Gpu Acceleration of the Long-wave Rapid Radiative Transfer Model in Wrf Using Cuda Fortran

This paper presents the approach and results of porting the Long-Wave Rapid Radiative Transfer Model (RRTM) component of the Weather Research and Forecast (WRF) code to the GPU using CUDA Fortran. After a brief description of the RTTM code, considerations regarding porting the application to the GPU are discussed. Included in the porting discussion are how the data structures have been modified for the GPU architecture, strategies for optimizing data movement, and determining how to partition the code into different kernels and how these kernels are configured. Results for both the original CPU and GPU code are provided in terms of accuracy and speed. Future optimizations using features not currently available in CUDA Fortran will be briefly discussed.