A high parallel way for processing IQ/IT part of HEVC decoder based on GPU

摘要

HEVC (High Efficiency Video Coding) is the newest video compression standard. Compared with the previous standards, the Coding efficiency is greatly improved at the cost of much higher codec complexity. So, many people improve the HEVC algorithm from the hardware level and software level. For the IQ/IT (inverse quantization/inverse transform) part, HEVC just processes one TU block one by one. And TU block size can be 4×4, 8×8, 16×16 and 32×32. So, it costs much decoding time by using this processing. In this paper, we present a parallel way for processing IQ/IT part based on GPU. In HEVC decoder, CABAC decoder only decodes one 64×64 size CTU block's parameters and some syntax elements. So, for now, we can process one 64×64 size Luma block and two 32×32 size Chroma block parallelly by using GPU. And from the experiment results, we can see the computation speed of computing the IQ/IT is about 33 times faster than the HM 12.0's. And finally, we can reduce about 40.7% decoding time compared with HM 12.0's.