Exascale Programming Approaches for Accelerated Climate Modeling for Energy

Abstract

The Accelerated Climate Modeling for Energy (ACME) project model is composed of five main components: atmosphere, ocean, land surface, sea ice, and land ice. The atmospheric component, the community atmosphere model-spectral element (CAM-SE) is based on a local, time-explicit spectral element (SE) method on cubed-sphere topology with static adaptation capability. CAM-SE realizes spherical geometry by use of the so-called cubed-sphere grid, which in this case utilizes a nonorthogonal equal-angle gnomonic projection from a cube onto the sphere. The fact that the SE method at its core is essentially made up of 1-D sweeps of matrix-vector multiplies is advantageous for graphics processing units because this means that data fetched from dynamic random-access memory will be reused. For the sake of efficiency, the choice was made at the creation of CAM-SE to structure the OpenMP implementation using parallel regions instead of loop-level OpenMP.