Abstract:

A GPU multistream parallel decoding implementation based on compute unified device architecture (CUDA) platform is designed to solve the high CPU load problem in multivideo decoding and display.The data structure of the graphics processing unit (GPU) device and video decoder are defined.By calling the decoding function interface,the decoder can be used in variety of video players efficiently.The experimental results show that this decoder can greatly reduce the CPU utilization.Also,compared with the JM reference software decoder,CPU utilization is remarkably reduced by 30% when decoding the single road 720 p high definition video synchronously.So the decoder based on hardware enjoys higher efficiencyin video parallel decoding.Besides,the system performance and resource reuse rate are improved together with lower energy consumption.

Key words: multi video decoding;CUDA;parallelization;occupancy;energy consumption