Design exploration of efficient implementation on SoC heterogeneous platform: HEVC intra prediction application. (21st December 2016)

• Record Type: Journal Article
• Authors: Kammoun, Manel; Ben Atitallah, Ahmed; Ben Atitallah, Rabie; Masmoudi, Nouri
• Is Part Of: International journal of circuit theory and applications. Volume 45:Number 12(2017:Dec.)
• Page Start: 2243
• Page End: 2259
• Publication Date: 2016-12-21
• Subjects: Xilinx Zynq‐based FPGA -- power consumption -- run‐time -- intra prediction HEVC -- SPSS; Electric circuit analysis -- Periodicals; 621.319205
• DOI: 10.1002/cta.2308
• Languages: English
• ISSNs: 0098-9886

Abstract

Summary: The relationship between CPU and hardware accelerator is critical especially in some systems that require intensive tasks and large amount of data to deal with such as video coding systems. This cooperation provides significant improvements in run‐time speed and power consumption. As software (SW) and hardware (HW) solutions provide better flexibility and performance, HW/SW implementation has emerged as a more efficient and desirable methodology for real‐time implementation. In order to evaluate different implementation methods (SW) and (HW/SW) in terms of power consumption, run‐time and area cost, we choose the Xilinx Zynq‐based FPGA as a target to perform some hardware acceleration tasks. In this case, we choose to accelerate the intra prediction block because it is one of the most complex modules defined in the high efficiency video coding decoder chain. Experimental results show that HW/SW accelerations are more than 50% improved in term of run‐time speed relative to SW modules. Moreover, the power consumption of HW/SW designs is saved by nearly 80% compared with SW cases. Copyright © 2016 John Wiley & Sons, Ltd. Abstract : In this paper, the Xilinx Zynq‐based FPGA is used as a target in order to evaluate the performances of HW and SW implementation of an intra prediction block in the high efficiency video coding standard. Experimental results show an improvement by more than 50% in terms of run‐time and power consumption compared with SW cases.