The interplay of run-time estimation and granularity in HW/SW partitioning

An important presupposition for HW/SW partitioning are sophisticated estimation algorithms at a high level of abstraction that obtain high quality results. Therefore the granularities of estimation and partitioning have to be adapted adequately. In this paper we discuss the effects that arise when the granularities of partitioning and estimation are not adapted in a necessary way. Furthermore we present our solution that allows to choose different levels of granularities adapted to the estimation and partitioning phase. The experiments show that this refinement in estimation at a high level of abstraction leads to an improvement (in terms of run-time and chip area) of the whole mixed HW/SW system.