Discrete-relaxation-based heuristic techniques for video algorithm/architecture matching and system level transformations

System level design and computational transformations have been rapidly establishing themselves as important high impact design steps which often have the most influential impact on the most important final performance parameter, throughput, of a design. In this paper, we introduce an iterative heuristic approach for throughput optimization when algorithm-architecture matching, and any of three behavioral transformations, retiming, rephasing, and pipelining, are considered simultaneously. The effectiveness of the approach and optimization algorithms is demonstrated on several video and image processing examples