Discussion related to adding support for GPUs in the framework and/or physics

[grantfirl]

Short of rewriting the entire codebase in CUDA, adding openACC directives to offload computationally burdensome computations to the GPU seems like the “path of least resistance”?

Where should this offloading to the GPU occur? In some schemes this is done at the ccpp interface level, in other instances this is done deep in the code at the kernel level? (And we at GSL are going to look at putting this offloading above the scheme, in the caps.)