科研成果详情

We propose and study a mapping algorithm optimized for shared-cache multicore processors. Performance requirement of modern applications is constantly growing. Processing huge amount of data in real-time is a trend even for mobile devices. It is common to find a octa-core processor in mobile phones or tablets. We will be able to see embedded devices with tens of cores in the next few years, if the trend continues. Conventional mapping algorithms are not well designed for shared-cache multicore processors. We discuss the importance of application mapping in terms of inter-application communication and sharedcache access delay. An algorithm is proposed with optimizations of the two aspects. We introduce a method with low computation complexity. First the mapping region is calculated with the congregate degree of nodes, then the region is expanded with a strategy in which the nearest nodes with lowest average cache latency are selected. The comparison with other mapping algorithms shows up to 13.9% improvement in average inter-application communication distance, with near optimal values considering the average cache latency. The results from real applications show that, the execution time and power consumption of the proposed algorithm has improved for 8% and 16.7% respectively, compared with an incremental mapping algorithm.