Donald Bren School of Information and Computer Sciences

Given a system behavioral specification, partitioning can be used to distribute among chips the processes, procedures, and storage elements that comprise the specification. We introduce a technique for constant-time recomputation of pin, area, and execution-time estimates for a behavioral partitioning move. The technique permits fast, accurate estimations of a large number of partitionings, thus enabling better results than approaches which attain tractable computation time by using gross estimates or less thorough partitioning algorithms. The key to our technique is the isolation and extraction before partitioning of the basic design attributes needed for estimation, and the updating of this information in constant-time for each move. The estimation models are almost as detailed as those presented in previous estimation approaches not intended for constant-time update. The results we provide indicate the speed and practicality of our estimation approach in conjunction with sophisticated partitioning algorithms.