Performance−portability and the Weather Research and Forecast Model

Performance and portability are important but conflicting concerns in the development of the Weather Research and Forecast model, a next−generation community mesoscale model for numerical weather prediction and atmospheric research. Efficiency depends on the ability to realize significant percentages of peak processor performance, yet without hyper−engineering the codes for a particular brand or even type of CPU. This involves first developing a quantitative understanding how aspects of the software, in particular data and looping structures, affect performance and then engineering the code to enable flexible tuning of these aspects across a variety of different platforms in a single source code. This paper describes work to characterize the performance effects of WRF model loop and data structures on representative micro− and vector−processors.