Open Issues

Abstract

This book has presented the state of the art in SDF scheduling theory. A formal theory for constructing and manipulating looped schedules from SDF representations of digital signal processing algorithms has been given, and based on this theory, techniques have been presented for compiling SDF programs into efficient code for programmable processors. The techniques have focused on the minimization of code size, and the minimization of the buffer memory requirement. We have defined a class of code-size-minimizing schedules called single appearance schedules. There are two central themes. The first is a uniprocessor scheduling framework that constructs single appearance schedules whenever they exist, and when single appearance schedules do not exist, guarantees optimal code size for all actors that are not contained in a certain type of subgraph called a tightly interdependent subgraph. The second theme consists of techniques that construct single appearance schedules for acyclic graphs that minimize data memory size, and an extensive experimental evaluation of these techniques on practical and random graphs. These techniques can easily be incorporated into the scheduling framework described in Chapter 5. Despite the focus on single appearance scheduling, many results of a more general nature, such as clustering theory, and the problem of minimum-buffer scheduling (not necessarily single appearance schedules) have also been presented.