Solving Constraint Satisfaction Problems in GPUs

Abstract

The complex Constraint Satisfaction Problems (CSPs) still require too long to solve even in the most recent computers. Recently the possibility emerged of using the high level of parallelism provided by current Graphics Processing Units (GPUs) to solve such problems. The work in progress presented here is meant to allow the use of the full computational power of GPUs to solve CSPs in a faster manner. This is done by parallelizing the search process between all the available cores on the GPU while attending to its complex memory hierarchy. This paper describes the methodologies used and the challenges encountered while programming GPUs to solve the CSP Costas Array.