| With the rapid development of embedded systems, video decoding technology onmobile devices is becoming widely popular. At present, the optimization of videodecoding is of great significance, because most of the mobile devices run on the ARMprocessor that is limited by the lack of storage capacity, processor speed and otherresources problems.As a new generation video coding standard, H.264has excellent compressionperformance and good network compatibility, but algorithm complexity is multiplying inorder to enhance the compression performance, which leads to difficult to achieve highreal-time video decoding on embedded devices because of insufficient resources.Cortex-A8is the latest processor developed by ARM Company which using ARMv7architecture, having unique NEON media processing engine that can accelerate videocoding, voice, image and other multimedia applications. Therefore, through optimizing theH.264decoding algorithm with the hardware characteristics of Cortex-A8processor, thevideo decoding in embedded platform can better meet the needs of practical applications.In this paper, the process of H.264video decoding and the main features ofCortex-A8processor are introduced in detail at first. And then, it concludes that themainly time-consuming operation is sub-pixel interpolation and loop filter by analyzing itscomplexity in the total decoding process, after H.264decoder of FFmpeg is transplantedto the iOS system. For the sub-pixel interpolation, the location of sub-pixel interpolation iscalculated by using the NEON parallelism instruction set, and the interpolation order isadjusted to avoid repeated access and calculations of pixels. For the loop filter, the filterdirection and order are changed correctly to be operated in parallel. At the same time,optimization of buffer, Cache performance in H.264decoder and the program-leveloptimization are conducted. In the end, these optimization measures are verified on the iPod Touch4with Cortex-A8processor. Through decoding two sequences of differentcomplexity, the results show that the total decoding time reduce about30%than before theoptimization. |
