A Compiler Scheduling Framework For Minimizing Memory Requirements Of Multirate Dsp Systems Represented As Dataflow Graphs

menting other scheduling objectives, such as the minimization of buffering requirements, in a manner that is guaranteed not to interfere with code compaction goals. We have defined a class of SDF graphs called tightly interdependent graphs. Schedules for arbitrary SDF graphs can be constructed such that each block that is not contained in a tightly interdependent subgraph appears only once, and thus requires only one instance of its code block to appear in the target program. Our framework defines a broad class of scheduling algorithms that construct such schedules. Our observations suggest that the vast majority of practical SDF graphs do not contain any tightly interdependent subgraphs, and thus that our scheduling framework guarantees optimal program compactness for most cases. However, we are also investigating how to schedule general tightly interdependent subgraphs compactly. New techniques that are developed for tightly interdependent graphs can easily be incorporated within our scheduling framework, since the framework modularizes the scheduling of tightly interdependent subgraphs: the algorithm used to schedule tightly interdependent subgraphs, called the " tight interdependence algorithm " never interacts with other parts of the overall scheduling algorithm, and the tight interdependence algorithm completely determines the amount of program memory required for the tightly interdependent subgraphs.