Towards a global solution to high level synthesis problems
Abdelhakim Safir, Bertrand Zavidovique · European Design Automation Conference · 1990
Since the various tasks as scheduling, operator allocation and module selection involved in the high level synthesis are strongly interdependent, a global solution to the high level synthesis problems becomes necessary. Therefore, a global optimization process which simultaneously performs scheduling, operators allocation and module selection, is presented in this paper. The search for a good solution in a realistic enough design space is made possible thanks to a global optimization algorithm which is simulated-annealing-based and improved by a pseudo-deterministic control. In the absence of other challenging global methods for both scheduling-operator allocation and module selection, the proposed global optimization algorithm is compared to a regular simulated annealing. Experimental results are shown in this paper and highlight a significant speed-up over a regular simulated annealing.- ranging from 45% to 73% on medium size problems (for instance the fifth order elliptic digital wave filter).- ranging from 82% to 235% on small size problems (the radix-2 FFT example (pipelined architecture)).- A much more complex example (from line-detection algorithm of [Danielson]) is briefly exposed and used to test the capability of the proposed method for processing such level of complexity.