Optimization Of VC-1Decoder On Cortex-A9Platform

With the development of mobile platform and the popularity of the mobile smartdevice, the multimedia application on mobile device is at a rapid increment stage. Mostof the mobile devices use ARM processor. NEON technique is able to acceleratemultimedia and signal processing algorithm and is an extension of ARM Cortex-Aprocessor’s SIMD technique, which is aimed to provide consuming multimediaapplication with flexible and powerful accelerating function and improves userexperience. VC-1is a new video codec technique based on WMV9(Windows MediaVideo9) proposed by Microsoft.After analyzing the Cortex-A9processor architecture, understanding the mainfunctions and performance of Cortex-A9and studying NEON technique, this paperstudies the algorithm of modules Inverse-DCT transform and motion compensation thatconsume much time according to the key technique of decoding process and the AmdahlLaw. First of all, this paper analyzes inverse DCT transform formula of8x8andconverts2-dimension Inverse-DCT transform to double horizontal Inverse-DCTtransform with matrix transposition. Then, using NEON instruction’s concurrentcomputation realizes horizontal Inverse-DCT transform and matrix transposition quickly.Secondly, after analyzing the algorithm of quarter pixel Bicubic interpolation of8x8block，this paper uses NEON’s function to implement the concurrent computation ofpixel’s interpolation in consideration of the uncorrelated of pixel data.Finally, the NEON optimization is implemented with hand-written assembly ofNEON instruction being based on the open source of FFmpeg. The optimized decoder istested with multiple video chips and the result is desired. The decoding speed of theoptimized VC-1decoder can improve25%.