摘要：

现有的行波列借拓扑重构算法是基于分级优化思想,把整体优化问题分解为以失效核为中心的局部优化问题,通过局部搜索失效核重构的最优解来求解整体优化问题的最优解。但其在局部邻域进行的是单向搜索,易导致搜索到的解并不是局部最优解,或者前一单元依次占用下一单元最优解而导致连锁列借操作。针对上述情况,构造一种局部邻域双向搜索的优化行波列借算法,可使局部解更优,并避免连锁操作。实验结果表明,该算法在失效核数目较多的情况下,与原有行波列借算法相比,众核处理器虚拟拓扑结构性能明显提高。

关键词: 众核处理器, 微体系结构级冗余, 核级冗余, 虚拟拓扑, 拓扑重构, 行波列借

Abstract:

The existing Row Rippling Column Stealing(RRCS) topology reconfiguration algorithm,which is based on the idea of hierarchical optimization,achieves the overall optimal solution through searching the local optimal solution. However,the local neighborhood searching of the RRCS algorithm is unidirectional, easily leading to the worse suboptimal solutions or the serious chain column stealing operations that the former unit sequentially occupies the optimal solution of next unit. Facing this situation,an optimized RRCS algorithm is proposed to improve the local solution and avoid the chain operations by using local neighborhood bidirectional searching. Experimental results show that when the fault cores on physical topology are more,the optimized algorithm is better.

Key words: many-core processors, microarchitecture-level redundancy, core-level redundancy, virtual topology, topology reconfiguration, Row Rippling Column Stealing (RRCS)

中图分类号: 

• TP391