Implementation And Optimization Of HEVC Entropy Coding And Deblocking Filter Algorithm On BWDSP Platform

Compared with H.264,high efficient video coding standard(HEVC)to ensure the video quality under the same compression ratio can be doubled.However,HEVC has achieved high efficient coding performance,and its coding complexity has also increased dramatically.It is necessary for the coding processing terminal to have higher data processing capability to achieve real-time encoding requirements,which has a negative impact on the popularization and application of HEVC.Entropy encoding and deblocking filter are two key modules in HEVC encoder,based on the actual needs of project,based on the BWDSP entropy encoding and deblocking filter module is implemented and optimized,its purpose is to reduce the two modules of the computational complexity,improve the operating speed,lay the foundation for the HEVC real time encoding.This paper first analyzes the complexity of the entropy coding and the block filtering module based on BWDSP.Secondly,based on the hardware resources carried by BWDSP,we optimize the entropy coding and deblocking filter modules from 3 levels of combinatorial optimization,namely,algorithm level optimization,system level optimization and code level optimization.(1)The optimization algorithm,this paper proposes a non termination filtering boundary into subsequent processing method based on the method of dividing depth,without affecting the bit rate is greatly reduced to the deblocking filter process useless work,and proposes an improved transform coefficient entropy encoding structure,the structure reduces a lot of conditions the structure of the original algorithm in the branch,easy hardware implementation.(2)For system optimization,is proposed based on the entropy encoding memory optimization strategy of BWDSP,balance the contradiction between the entropy encoding speed and limited memory,and proposes a optimization method of DMA data transmission based on the ping-pong buffer,this method can improve the data throughput rate,accelerate the speed of entropy encoding.(3)The code level optimization,mainly using loop unrolling and software pipelining and parallel optimization method to strengthen the compilation optimization design,and the scanning process for transform coefficient matrix in the last nonzero coefficient information such as the location of the design of a multi task level single core DSP parallel processing optimization scheme based on the optimization program execution the speed is 7 times higher than the traditional serial assembly optimization method,and the context model initialization process and strong filtering process were designed corresponding parallel processing architecture.Finally,the experimental results show that after optimization,the operation speed of HEVC entropy coding and deblocking filter module is significantly improved,and the average speedup is 15 times and 6 times respectively.