HEVC in-loop filters GPU parallelization in embedded systems

The added encoding efficiency and visual quality that is offered by the latest HEVC standard is mostly attained at the cost of a significant increase of the computational complexity at both the encoder and decoder. However, such added complexity greatly compromises the implementation of this standard in computational and energy constrained devices, including embedded systems, mobile and battery supplied devices. To circumvent this limitation, this paper proposes the exploitation of embedded GPU devices already equipping many state of the art SoCs to accelerate the HEVC in-loop filters (i.e. deblocking filter and sample adaptive offset). The presented approaches comprehensively exploit both fine and coarse-grained parallelization opportunities of these filters in an NVIDIA Tegra GPU.According to the conducted experimental evaluation, the proposed approach showed to be a remarkable strategy to satisfy the real-time requirements of the HEVC decoder, being able to filter each Ultra HD 4K intra frame in less than 20 ms (about 50 fps).