Editorial: Special Issue on Hardware/Software Partitioning

Abstract

graph models. The fourth approach uses coarse-grain task descriptions (as synchronous data-flow graphs) in a targeted simulation environment. For a specific architecture, partitioning of an algorithmic description results in a partition of operations or tasks into hardware and software portions. An exact solution to such a partitioning problem even in the simplest cases requires solution to known intractable problems. This provides motivation for heuristic solutions. The first paper by Kalavade and Lee presents a heuristic approach that uses a combination of global and local measures to guide the partitioning algorithm. These separate measures are used to direct the algorithmic search to good solution while at the same time ensuring the satisfaction of the partitioning goals without excessive backtracking in the design space. These measures effectively compensate for the modeling limitations in system design by providing additional (expert) information to guide the search for a good solution. As mentioned earlier, the partitioning and synthesis subtasks such as scheduling and resource allocation and binding are closely inter-related: that is, a good solution to any one of the subtasks uses solutions for the other subtasks. This dilemma, in the specific case of partitioning and scheduling, is addressed directly in the second paper by Niemann and Marwedel. The authors describe an integer programming (IP) formulation for the partitioning problem that uses an approximate schedule of operations. Solution to this partitioning IP formulation is followed by a solution to the IP formulation of the scheduling problem that updates the schedule. A similar relationship of partitioning and allocation can be seen in the third paper by Madsen and co-authors. The fourth paper by Vahid and Le considers a host of methods to speed up the hardware/software partitioning algorithm using group migration techniques without sacrificing the quality of results. The four papers included in this issue were selected from a total nine submissions after an exhaustive review process in which each of the submissions underwent at least two rounds of reviews. My sincere thanks to the authors for their prompt response and patience with inevitable delays in the review process. My sincere gratitude to all the reviewers for their constructive comments. Finally, I would like to thank Karen Cullen of the journal editorial staff and Mike Casey of the journal production staff for their help in putting this issue