| As a way of expressing information intuitively and efficiently, video is widely used in today’s society. Video source must be compressed because of huge amount of data. H.264standard, as a new generation of video compression coding standard, has remarkable image quality of recovery and compression performance, but its computational complexity is very high, which makes the encoding speed very slow.In recent years, in the heterogeneous computing platform constituted of CPU and GPU, GPU is used for the general computing instead of rendering of the image. In2008, OpenCL programming model was put forward, so that the applications based on it can work on a variety of heterogeneous platforms widely. Therefore, using GPU to compute parallelly high complexity modules in H.264standard will improve the encoding speed obviously.This paper has designed an intra prediction completely diagonal parallel algorithm based on OpenCL for luma and chroma, and further designed a luma fast mode selection algorithm which greatly reduces the computation of mode selection. For B frame bidirectional motion search in main profile and extend profile, taking advantage of the characteristics that different waves can compute parallelly in OpenCL programming model, we design B frame scalariform parallel bidirectional motion search algorithm. For loop filter, without changing the H.264standard original algorithm, we design two kinds of parallel algorithms, which make full use of the OpenCL programming model’s features.This paper has done a lot of parallel encoding experiments using HD video sequences data, and the experimental results show that compared with CPU serial algorithm, parallel algorithms of I frame and B frame reduce encoding time effectively in the case of ensuring the image quality of recovery and compression performance. The two loop filter parallel algorithms in this paper ensure filter performance the same with CPU serial algorithm, while they reduce the computational time evidently. |
PDF Full Text Request