Dataflow-driven Memory Allocation For Multi-dimensional Signal Processing Systems

Abstract

Memory cost is responsible for a large amount of the chip and/or board area of customized video and image processing systems. In this paper, a novel background memory allocation and assignment technique is presented. It is intended for a behavioural algorithm specification, where the procedural ordering of the memory related operations is not yet fully fixed. Instead of the more restricted classical scheduling-based explorations, starting from procedurally interpreted specifications in terms of loops, a novel optimization approach—driven by data flow analysis—is proposed. Employing the estimated silicon area as a steering cost, this allocation/assignment technique yields one or (optionally) several distributed (multi-port) memory architecture(s) with fully-determined characteristics, complying with a given clock cycle budget for read/write operations. Moreover, our approach can accurately deal with complex multi-dimensional signals by means of a polyhedral data-flow analysis operating with groups of scalars.