GPU Acceleration of Cloud Microphysics: Porting CASIM to GPUs Using PSyclone

Leveraging GPUs offers both significant opportunities and challenges for weather and climate models. The opportunity lies in the GPUs' massive parallelism, which can greatly enhance computational performance. However, the challenge arises from the need to refactor thousands of lines of existing CPU code and adopt a new programming model to fully harness this parallelism. In this presentation, we will discuss the development of the GPU-optimized cloud microphysics scheme, CASIM, and how we employ PSyclone, a Domain Specific Compiler developed in the Met Office, to keep single source science code whilst targeting multiple programming models for different processor architectures. We will show how PSyclone ports and optimizes CASIM on GPUs following ORNL’s experience in porting and optimizing CASIM using OpenACC.