This study focuses on the key numerical technique of stencil computations, used in many different scientific disciplines, and illustrates how auto-tuning can be used to produce very efficient implementations across a diverse set of current multicore architectures.
