Stencil operations are used widely in HPC applications and pose an optimization challenge on both CPUs and GPUs. On GPUs, fine-tuned optimizations can be formulated using low-level APIs such as CUDA, but many large established codes prefer a portable, higher-level API such as OpenACC. Although OpenACC lacks the fine-tuning of CUDA, it does allow for some tuning through a variety of parallelization constructs and loop directives. Here, we try various OpenACC directive options to optimize the computationally heaviest stencil operation within our production solar physics research code Magnetohydrodynamics Around a Sphere (MAS).

• Explore numerous OpenACC directive options, including tile, cache, and collapse, and compare their performance over several problem types and sizes;

• Learn which combination of options yields the overall optimal result and see how it appears in a full-scale simulation analyzed with Nsight Systems and Nsight Compute; and

• Be aware of cautionary cases where using the "wrong" directives can yield very poor performance.

Content

content goes here