Relatore: Prof. Frank Loeffler CCT, Luisiana State University Baton Rouge (Luisiana), USALuogo: Sala FeynmanE-mail organizzatore: roberto.depietri@fis.unipr.it Abstract: Stencil computations are a class of algorithms which perform nearest-neighborcomputation, often on a multi-dimensional grid. This type of calculation forms thebasis for computer simulations across almost every field of science, including numericalrelativity and astrophysics. The increasing computational speed of graphics processingunits (GPUs) make their use for stencil computations an interesting goal.However, achieving highly efficient implementations is often nontrivial, as numerouspublications attest. In this talk, I present our proposal of an analytic performancemodel for stencil codes on GPUs, which both delivers close-to optimal performance,but at the same time does not require extensive tuning at compile or run time.
