Genetic multiway partitioning

This research investigates a new software tool for Genetic Partitioning. The Genetic Algorithm is used to perform the partitioning with a significant improvement in result quality. Furthermore, it can optimize a cost function with multiple objectives and constraints. Separate algorithms have been developed, fine-tuned for bipartitioning and multiway partitioning. The bipartitioning problem is represented as a binary chromosome. Efficient bit-mask operations perform crossover, mutation, and net cut evaluation 32 bits at a time, without unpacking. The multiway partitioning algorithm has a global view of the problem, and generates/optimizes all the necessary partitions simultaneously. The algorithms were tested on the MCNC benchmark circuits, and the cut size obtained was lower than that for the conventional Fiduccia-Mattheyses algorithm.