Multi-schedule design space exploration: an alternative synthesis framework

• Published in: Integration (Amsterdam) Vol. 27; no. 2; pp. 87 - 112
• Main Authors: Dalkilic, M.E., Pitchumani, V.
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 1999
• ISSN: 0167-9260; 1872-7522
• DOI: 10.1016/S0167-9260(99)00007-3

A multi-schedule data-path synthesis framework is described that is an improvement over conventional single-schedule methods. The multi-schedule approach retains all feasible scheduling options and generates complete data-path solutions for each schedule. Constraints on time and resources are used to create a cost-ordered set of feasible scheduling solutions. Hence scheduling optimality is assured within the model of constraints. The algorithms have been converted to parallel form allowing execution on a cluster of workstations or an Intel hypercube. The results show that multi-schedule synthesis approach often outperforms well-known high-level synthesis systems in terms of both solution quality and run time.