杭赟, 谢憬, 毛志刚. 基于任务流水化的簇状流处理器专用通信单元的设计[J]. 微电子学与计算机, 2015, 32(5): 91-94. DOI: 10.19304/j.cnki.issn1000-7180.2015.05.019
引用本文: 杭赟, 谢憬, 毛志刚. 基于任务流水化的簇状流处理器专用通信单元的设计[J]. 微电子学与计算机, 2015, 32(5): 91-94. DOI: 10.19304/j.cnki.issn1000-7180.2015.05.019
HANG Yun, XIE Jing, MAO Zhi-gang. Design of Intra Communication Module for the Pipelined Task Scheme in the Cluster-based Stream Processor[J]. Microelectronics & Computer, 2015, 32(5): 91-94. DOI: 10.19304/j.cnki.issn1000-7180.2015.05.019
Citation: HANG Yun, XIE Jing, MAO Zhi-gang. Design of Intra Communication Module for the Pipelined Task Scheme in the Cluster-based Stream Processor[J]. Microelectronics & Computer, 2015, 32(5): 91-94. DOI: 10.19304/j.cnki.issn1000-7180.2015.05.019

基于任务流水化的簇状流处理器专用通信单元的设计

Design of Intra Communication Module for the Pipelined Task Scheme in the Cluster-based Stream Processor

  • 摘要: 针对基于任务流水化机制的高性能DSP流处理器的运算簇簇间数据通信问题,提出了COMM通信单元的设计思想.该单元通过对各个簇输出数据进行重新排布,提升了簇间数据交换的性能,避免了主控核参与所引起的较大的时间开销.基于若干个典型DSP应用对设计方案进行实验,测得平均性能提升可达24.6%,实现了性能优化的目标.

     

    Abstract: The paper introduces an idea of a communicating module, COMM, which is designed to optimize communication performance among computing clusters according to the pipelined-task scheme for the cluster-based DSP stream processor. The design improves the performance of inter-cluster communication by rearranging the output data of clusters, reducing the cost of running time caused by the prior scheme with the management from the main core. With some typical DSP benchmark, the design can help to improve the performance of the stream processor in about 24.6% in average.

     

/

返回文章
返回