Accelerating IEEE 1857 Deblocking Filter on GPU Using CUDA

In IEEE 1857/AVS decoder, the deblocking filter is one of the most time consuming tasks and it accounts for nearly 30% of the decoder complexity. Due to data dependence in the process of filtering, it is a big challenge to execute deblocking filter in parallel efficiently. In this paper, a parallel algorithm is proposed, which hides the data dependence by dividing a frame into several 8 × 8 Intersection-blocks. The proposed parallel deblocking filter algorithm is implemented on GPU using CUDA. Experimental results of processing 1920 × 1080 video streams show that the proposed CUDA deblocking filter implementation accelerates traditional CPU implementation by 6~8 times.