| Summary: | 碩士 === 國立臺灣大學 === 電機工程研究所 === 82 === This thesis proposed an approach for rate-optimal scheduling of
recursive DSP programs. A multiprocessor schedule is rate-
optimal if the iteration period equals the iteration bound. The
iteration bound is naturally derived from recursive programs.
Many researches have been devoted to similar work. Retiming and
unfolding transformations are two useful techniques in this
research domain. The retiming transformation can optimize the
computation time of each iteration, but can not guarantee this
schedule is rate-optimal. Unfolding can exploit intra-iteration
and inter-iteration parallelism, and can reduce the iteration
period. We introduce an approach to obtain a rate-optimal
schedule through unfolding transformations. For general DFG's,
Pari and Messerschmitt show that if the unfolding factor is the
least common multiple of all the loop register counts in DFG,
without any retiming a rate-optimal schedule can be achieved.
Our work here is to reduce the unfolding factor via simple
arithmetic computing.
|
|---|
