A framework that supports in writing performance-optimized stencil-based codes

On modern multicore processors, many applications run much slower than one would expect when looking at the vast congregated computational power. After a discussion of factors determining the performance on such CPUs, a framework concept that simplifies writing performance-optimized codes for stencil-based algorithms and a prototypical implementation are presented. Finally, the suitability of this approach is discussed by analyzing performance results for three different applications which have been optimized within that framework.