MILP-based models are useful for finding optimal schedules and for proving their optimality. Because of the problem complexity, model improvements have to be investigated. We analyze the constraints necessary for precluding resource conflicts, present novel formulations, and evaluate them. The efficiency of the solution process can be improved significantly by selecting the proper formulation.

Our problem consists of a partially ordered set of tasks communi-cating over a shared bus which are to be mapped to a heterogeneous multiprocessor system. The goal is to minimize the makespan, while satisfying constrains implied by data dependencies and ex-clusive resource usage. We present a new efficient heuristic approach based on list scheduling and genetic algorithms, which finds the optimum in few seconds on average even for large examples (up to 96 tasks) taken from [3]. The superiority of our algorithm compared to some other algorithms is demonstrated.

Most of the existing Digital Signal Processors (DSPs) are optimized for a fast and efficient computation of the Fast Fourier Transform (FFT). However, there are only two floating-point DSPs available, which perform the butterfly operation of a FFT in 4 processor cycles, but no fixed-point DSP is designed that way. The new 24 bit fixed-point DSP DAISY, which is able to execute the butterfly in 4 cycles even using a two-stage pipelined multiplier, is described in this paper. With this pipelined multiplication it is possible to reduce the processor cycle time significantly.