Research On H.264 Video Multi-Core Decoding Andthe Realization On The Virtual Nuclear Platform

Video codec technology is widely used in the filed of Digital HDTV, network streaming media, video communications etc.The current mainstream standards are H.264/AVC was offically released by 2003.3,and HEVC/H.265 was offically released by 2013.1.Existing large number of H.264 standard video files,and compared to HEVC/H.265,H.264 decoding algorithm is less complex,so this paper chooses H.264 public core decoding as the research starting point.For guarantee high quality, low bit rate,H.264 need to do a lot of calculation,H.264 decoding program is currently High performance CPU / DSP solution, Hardware acceleration module,Dedicated ASIC chip / FPGA solution,Multi-core (2/4 core, ARM + DSP structure) program,Each program has advantages and disadvantages in terms of performance, power consumption, and flexibility.This paper is for H.264 video decoding applications,researching on high-performance, low-power, high-flexibility H.264 video core (16-core and above)decoding implementation,explore the key technical issues in H.264 public core decoding,the results can also be used for H.265 and other video standards.The target public core platform is a virtual platform built by the MCVP-NoC system.MCVP-NoC is a self-designed multi-core virtual platform modeling tool that supports custom NoC (Network-on-Chip).It is using the "SystemC + TLM2.0 + OVP"tool,MCVP-NoC generated virtual platform can run the actual program,supportingreal application drivers,for project early software development, debugging, system architecture exploration, and performance, power consumption, area assessment, etc.The main contents of this paper are:(1) Analysis of H.264 decoding performance bottlenecks;(2) H.264 decoding task division and public core mapping.First by the decoding process to divide the task,Using four assembly line of "stream analysis, slice decoding, filtering, output" ,and then dividing the parts of the decoding process by calculation and Multi-core parallel acceleration.(3) Building the public nuclear virtual platform, using it to run H.264 public core decoding program, evaluating performance.The public core system uses 4x4 2D mesh structure,a total of 16 processor nodes,each node contains 1 orlk processor, 32M bytes of instruction memory, 32M bytes of data memory,all nodes share one 256M bytes of shared-memory.The model can run the actual H.264 decoding code,quantitative analysis of performance,can be used for system architecture optimization, on-chip storage requirements analysis and storage plan optimization;quantitative analysis of inter-core communication traffic.can be used to guide the inter-core NoC interconnect structure and link bandwidth design.