ISSN 1000-1239 CN 11-1777/TP

计算机研究与发展 ›› 2017, Vol. 54 ›› Issue (3): 514-528.doi: 10.7544/issn1000-1239.2017.20150912

• 人工智能 • 上一篇    下一篇

基于GPU的约束网络模型和并行弧相容算法

李哲,李占山,李颖   

  1. (吉林大学计算机科学与技术学院 长春 130012) (符号计算与知识工程教育部重点实验室(吉林大学) 长春 130012) (zslizsli@163.com)
  • 出版日期: 2017-03-01
  • 基金资助: 
    国家自然科学基金项目(61272208,61373052);吉林省自然科学基金项目(20140101200JC)

A Constraint Network Model and Parallel Arc Consistency Algorithms Based on GPU

Li Zhe, Li Zhanshan,Li Ying   

  1. (College of Computer Science and Technology, Jilin University, Changchun 130012) (Key Laboratory of Symbolic Computation and Knowledge Engineering (Jilin University), Ministry of Education, Changchun 130012)
  • Online: 2017-03-01

摘要: 弧相容算法是约束满足问题的基本压缩求解空间算法之一,很多优秀的高级算法都以高性能的弧相容算法作为核心.近年来,以GPU为计算工具加速并行计算被用来尝试解决许多问题.基于GPU和基本的并行算法,提出一种适合GPU运算的约束网络表示模型N-E,给出其生成算法BuildNE.结合细粒度的弧相容算法——AC4,基于N-E模型提出AC4的并行化算法AC4\+GPU与改进算法AC4\+GPU+,使弧相容算法得以扩展到GPU上执行.实验结果验证了该算法的可行性,与AC4算法的比较,其在一些规模较小的问题上取得了10%~50%的加速,在一些规模较大的问题上则加速1~2个数量级.为今后进一步在GPU上以并行形式解决其他约束满足问题提供了一种核心算法方案.

关键词: 人工智能, 约束满足问题, 弧相容, 图形处理器, 计算统一设备架构

Abstract: Constraint satisfaction problem is a popular paradigm to deal with combinatorial problems in artificial intelligence. Arc consistency (AC) is one of basic solution compression algorithms of constraint satisfaction problem, which is also a core algorithm of many excellent advanced algorithms. When constraints are considered independently, AC corresponds to the strongest form of local reasoning. An effective underlying AC can improve the efficiency of reducing the search space. Recent years, GPU has been used for constituting many super computers, which solve many problems in parallel. Based on GPU's computation, this paper proposes a constraint networks presentation model N-E and its parallel generation algorithm BuildNE. According to fine-grained arc consistency AC4, a parallel edition AC4\+GPU and its improved algorithm—AC4\+GPU, are proposed. The two parallel algorithms extend arc consistency to GPU. Experimental results verify the feasibility of these new algorithms. Compared with AC4, the parallel versions have made the 10% to 50% acceleration in some smaller instances, and obtained 1 to 2 orders of magnitude in some bigger instances. They provide a core algorithm to other constraint satisfaction problem solving in parallel for further study.

Key words: artificial intelligence, constraint satisfaction problem (CSP), arc consistency (AC), GPU, compute unified device architecture (CUDA)

中图分类号: