摘要： 针对现有分块算法并行度低、负载不平衡的缺陷，提出一种基于多区域多代码问题的自动分块算法。通过循环分配算法实现计算区域间的处理器分配，基于Block的递归二分法对无向图进行剖分，实现计算区域内的任务分配。实验结果表明，该算法可使整个计算空间分配到的处理器量大致相等，处理器间的通信量最小。

关键词: 多区域多代码, 并行计算, 自动划分, 负载平衡

Abstract: Aiming at the shortage of low parallelism degree and load imbalance in existing partition algorithm, this paper presents an automatic partition algorithm based on multi-region and multi-code problem. This algorithm uses loop distribution algorithm to realize processor distribution between computational domain, uses recursion dichotomy based on Block for splitting undirected graph to realize task distribution in computational domain. Experimental results show that this algorithm can make allocated number of processor equation all of the computational space, and the communication traffic between processors is least.

Key words: multi-region and multi-code, parallel computing, automatic partition, load balance

中图分类号: 

• TP301