High-performance xPU Stencil Computations in Julia

We present an efficient approach for writing architecture-agnostic parallel high-performance stencil computations in Julia, which is instantiated in the package ParallelStencil.jl. Powerful metaprogramming, costless abstractions and multiple dispatch enable writing a single code that is suitable for both productive prototyping on a single CPU thread and high performance production runs on GPU or CPU workstations or, if used in combination with distributed parallelization packages as ImplicitGlobalGrid.jl, on supercomputers. We demonstrate performance close to the theoretical upper bound on GPUs for a 3-D heat diffusion solver, which is a massive improvement over reachable performance with CUDA.jl Array programming.