Communication-aware HW/SW co-design for heterogeneous multicore platforms

Abstract

QUAD is an open source profiling toolset, which is an integral part of the Q2 profiling framework. In this paper, we extend QUAD to introduce the concept of Unique Data Values regarding the data communication among functions. This feature is important to make a proper partitioning of the application. Mapping a well-known feature tracker application onto the multicore heterogeneous platform at hand is presented as a case study to substantiate the usefulness of the added feature. Experimental results show a speedup of 2.24x by utilizing the new QUAD toolset.