Renesting Single Appearance Schedules to Minimize Buffer Memory.

摘要

Minimizing memory requirements for program and data are critical objectives when synthesizing software for embedded DSP applications. In prior work, it has been demonstrated that for graphical DSP programs based on the widely-used synchronous dataflow model, an important class of minimum code size implementations can be viewed as parenthesizations of lexical orderings of the computational blocks. Such a parenthesization corresponds to the hierarchy of loops in the software implementation. In this paper, we present a dynamic programming technique for constructing a parenthesization that minimizes data memory cost from a given lexical ordering of a synchronous dataflow graph. For graphs that do not contain delays on the edges, this technique always constructs a parenthesization that has minimum data memory cost from among all parenthesizations for the given lexical ordering. When delays are present, the technique may make refinements to the lexical ordering while it is computing the parenthesization, and the data memory cost of the result is guaranteed to be less than or equal to the data memory cost of all valid parenthesizations for the initial (input) lexical ordering. Thus, our dynamic programming technique can be used to post-optimize the output for any algorithm that schedules SDF graphs into minimum code size loop hierarchies. On several practical examples, we demonstrate that significant improvement can be gained by such post-optimization when applied to two scheduling techniques that have been developed earlier. That is, the result of each scheduling algorithm combined with the post-optimization often requires significantly less data memory than the result of the scheduling algorithm without post-optimization. We also present an adaptation of our dynamic programming technique for post- optimizing an arbitrary (not necessarily minimum code size) schedule to optimally reduce the code size.