Design Of H.264 Parallel Decoder Based On GPU

H.264 is the most optimum practical video coding standard in the video compression field nowadays, which was co-developed by the two international organizations ITU-T and ISO/IEC in 2003. H.264 has the incomparable performance comparing with the previous coding standards, however, the computational complexity and the time for coding and decoding have increased.GPU (Graphic Processing Unit) is the core of Graphics Card, which has strong parallel computing ability. In recent years, GPU has been used for general purpose computation increasingly, the GPU for video compression has broad application prospects.In this paper, a parallel H.264 decoder based on GPU is designed, following an evaluation of the traditional H.264 serial decoding model. The parallel decoder uses CPU+ GPU framework and double thread design, where the main thread is responsible for scheduling GPU to process IQ, IDCT, Intra/Inter frame prediction, loop filtering module etc, and the data transmission between CPU and GPU, while the sub thread is focused on the CPU to complete code stream analysis, entropy decoding and reordering. CPU and GPU can work efficiently at the same time in this decoder.By fully researching the critical and time-consuming modules in decoder, some parallel algorithms based on data independence and GPU resource are proposed in this paper. A butterfly parallel algorithm, an enhanced butterfly parallel algorithm, and a direct multiplication parallel algorithm based on left multiplication and transposition are presented for IDCT. A parallel algorithm is applied for IQ. A fixed trapezoid parallel algorithm and an adaptive trapezoid parallel algorithm are proposed for intra frame prediction. An efficient MC parallel algorithm is utilitied for inter frame prediction. A parallel method and a simplified parallel algorithm are porposed for calculating filtering strength in loop filtering. A fixed trapezoid parallel and an improved parallel filter algorithm for the executive parts in loop filtering are presented in this paper.The experimental results show that the proposed CPU+GPU parallel decoder could accelerate the decoding significantly. Though the hardware condition is quite limited, the parallel decoder can make full use of the parallel computing ability of GPU and the serial computing ability of CPU.