A search method based on multiple GPUs

• Main Authors: YAO YAPENG, CHEN HAO, SUN JIANHUA
• Format: Patent
• Language: Chinese; English
• Published: 25.08.2020
• Subjects: CALCULATING; COMPUTING; COUNTING; ELECTRIC DIGITAL DATA PROCESSING; PHYSICS

The invention discloses a parallel A* search method based on multiple GPUs, and relates to the field of distribution. At present, due to the limitation of GPU memory and computing power, when the datascale reaches a certain order of magnitude, the bottleneck of speed and memory occurs in A * search based on the GPU. The method comprises a graph data segmentation method and an execution mode of A* search on a GPU. Aiming at the memory heterogeneous hierarchy problem under the multi-GPU computing architecture condition, the method focuses on the partitioning modes of different data, so that the load balance between GPUs and the effective interaction of the data are effectively realized. And then a parallel A * search method of a plurality of priority queues is adopted on the GPUs, and corresponding optimization is realized for the architecture of the plurality of GPUs. And finally, for the problem of repeated node search possibly occurring in the operation on the GPU, alternative parallel hashing is selected. A