A high-efficient software parallel CAVCL encoder based on GPU

Abstract

This article presents an efficient parallel CAVLC encoder for H.264/AVC based on GPU. By optimizing the architecture of the encoder, three kinds of dependences are eliminated or weakened, including the context dependence, the memory accessing dependence, and the control dependence. We divide the execution of CAVLC into three stages: two scans, component-oriented coding, and lag packing. For each stage, data of a frame can be processed synchronously. Experimental results show that the proposed parallel CAVLC encoder can achieve more than 30 times speedup when compared with the CPU version, and a real-time process for 720p @ 30p can be achieved. The throughput of the presented CAVLC encoder is 11.17 to 6.29 times higher than that of the published software encoders on DSP and multi-core platforms.