Automatic code generation and tuning for stencil kernels on modern shared memory architectures

Abstract In this paper, we present PATUS, a code gener-ation and auto-tuning framework for stencil computations targeted at multi- and manycore processors, such as mul-ticore CPUs and graphics processing units. PATUS, which stands for “Parallel Autotuned Stencils, ” generates a com-pute kernel from a specification of the stencil operation and a strategy which describes the parallelization and optimiza-tion to be applied, and leverages the autotuning methodol-ogy to optimize strategy-specific parameters for the given hardware architecture