Library insertion and reuse of datapath components in high-level synthesis
Roger Ang · 1996
This thesis presents a formulation for improved reuse of combinatorial datapath components in high-level synthesis. For combinatorial datapath components, I define new abstractions that encapsulate a broader range of components than previous models, and provide a tighter coupling of existing RT component libraries with high-level synthesis. I describe the role of datapath component reuse in high-level synthesis and discuss existing work related to this issue. I define the structures necessary for improved reuse of datapath components which include: a database that abstracts the functionality of combinatorial components, an intermediate representation to describe high-level behavior in terms of datapath component functionality, and a graph-matching mechanism for efficient mapping of abstract operations to component functions. I illustrate the use of these models and representations in two algorithms: an algorithm to allocate function units to a behavioral description, and an algorithm to schedule operations of a behavioral description to a given set of components. I describe the experimental system implemented to demonstrate the models and algorithms, and the experiments conducted. The results of these experiments illustrate some of the advantages of this approach including development of design implementations with improved performance and reduced area.