Design and Implementation of a Dynamic Task Mapping Algorithm Based on Hotspot Statistics. Issue 1 (1st April 2022)

• Record Type: Journal Article
• Authors: Wang, Jie; Zhang, Duoli; Ge, Hu; Ni, Wei; Song, Yukun
• Is Part Of: Journal of physics. Volume 2245:Issue 1(2022)
• Publication Date: 2022-04-01
• Subjects: Physics -- Congresses; 530.5
• DOI: 10.1088/1742-6596/2245/1/012001
• Languages: English
• ISSNs: 1742-6588

Abstract

Abstract: In a coarse-grained heterogeneous multi-core system, the task map is responsible for allocating the cores for task execution. The existing user static mapping method has low programming friendliness. As the scale and density of system tasks increase, so does the complexity of manual programming by users, resulting in programming inefficiencies. Aiming at this problem, based on the existing heterogeneous multi-core platform, this paper studies a dynamic task mapping method with hotspot statistics, and uses the preprocessing of the mapping to reduce the time spent on the actual mapping. The hardware mapper is designed and implemented in Verilog, and the verification platform is built in the HDL environment. The experimental results show that the new mapping algorithm significantly improves the performance of the system, with an average of 32.50% and a maximum of 47.16%.