Generalized ILP scheduling and allocation for high-level DSP synthesis

Integer linear programming (ILP) techniques have recently been proposed to solve the single iteration scheduling problem. In the present work, the integer linear programming formulation is generalized to support unfolded schedules. A new time assignment formulation which automatically pipelines and retimes the dataflow graph (DFG) is presented. A new processor allocation formulation which automatically unfolds the DFG is also presented. It is noted that this is the first time that systematic unfolding is proposed for improving processor utilization on digital signal processing architectures. These models fully optimize the processor utilization for a specific iteration period.