Optimization Of VP10 Decoder

Google has begun to develop VP10 codec standard in August 2015. The bandwidth of 4K video which used by VP10 codec technology is half of which used by the VP9. At present, the mobile device is more popular, but VP10 in the field of mobile applications is still in its infancy. Because of the fact that the performance of mobile devices is limited and the processor is not strong, the problem about the performance optimization of VP10 on mobile devices is proposed.Because of the situation that most of smart mobile phone manufacturer generally use the Cortex-A9 architecture processor to product equipment at home and abroad, this research take the Nexus7 plate as the laboratory equipment. The performance test tool named Oprofile is ported into Nexus7 to test the performance of VP10 decoder. To analysis the result which was tested by Oprofile. The time-consuming modules of the VP10 which include motion compensation, inverse transform, entropy decoding and loop filter are to be found.In order to resolve the problem about the performance bottlenecks, this research analysis those time-consuming modules include motion compensation, inverse transform, entropy decoding and loop filter. And then the conclusion that the data correlation of the motion compensation, inverse transform and loop filter are small and it is feasible in data parallel operation on those modules. Therefore, this research offer an optimization solution about the performance of these modules based on the special instruction set of mobile device platform.This research study optimization solution about VP10 decoder based on the open source project named libvpx. The project offers the standard of VP10 and the code. When study the motion compensation, inverse transform, entropy decoding and loop filter of libvpx, the optimization solution that refactoring libvpx used by NEON come up. The final experimental results show that this optimization solution improves 43% to 68% at the overall performance of VP10 decoder.