Section: New Results

Reducing Thread Divergence in GPU-based B&B Applied to the Flow-shop Problem

Participants : I. Chakroun, A. Bendjoudi, N. Melab.

Branch-and-Bound (B&B) algorithms are attractive methods for solving to optimality combinatorial optimization problems. Nevertheless, they are time-intensive when dealing with large problem instances. Therefore, several parallel B&B strategies based on large computer clusters and grids have been proposed in the literature. However, to the best of our knowledge no contribution has been proposed for designing B&B algorithms on GPUs (Graphic Processing Units). Because of their tremendous computing power and remarkable cost efficiency, GPUs have been recently revealed as a powerful way to achieve high performance on long-running scientific applications. In this research work, we propose to revisit the design and implementation of B&B algorithms on GPU. We focus on the parallel evaluation of the bounds since preliminary experiments performed on the Flow-Shop scheduling problem (FSP) have shown that the bounding operation consumes over 98% of the execution time of the B&B algorithm. To deal with thread divergence reduction issue caused by the bounding operation a code refactoring approach have been proposed.