A* Algorithm for Graphics Processors

Paper in proceeding, 2010

Today's computer games have thousands of agents moving at the same time in areas inhabited by a large number of obstacles. In such an environment it is important to be able to calculate multiple shortest paths concurrently in an efficient manner. The highly parallel nature of the graphics processor suits this scenario perfectly. We have implemented a graphics processor based version of the A* path finding algorithm together with three algorithmic improvements that allow it to work faster and on bigger maps. The first makes use of pre-calculated paths for commonly used paths. The second use multiple threads that work concurrently on the same path. The third improvement makes use of a scheme that hierarchically breaks down large search spaces. In the latter the algorithm first calculates the path on a high level abstraction of the map, lowering the amount of nodes that needs to be visited. This algorithmic technique makes it possible to calculate more paths concurrently on large map settings compared to what was possible using the standard A* algorithm. Experimental results comparing the efficiency of the algorithmic techniques on a NVIDIA GeForce GTX 260 with 24 multi-processors are also presented in the paper.