Vectorized deblocking filter for HD H.264 decoding on Cell/B.E.

For high definition (HD) H.264 decoding, deblocking filter is one of the most time-consuming modules. This paper proposes several vectorization approaches to speed up it on a single synergistic processor element (SPE) of IBM Cell Broadband Engine (Cell/B.E.) processor, by which great performance improvements are achieved. The average deblocking speed is 42.4 frames per second (fps) for 141 1080p H.264 video streams. The steady-going performance is obtained for different streams of various contents and bitrates. With the vectorized deblocking filter, the new HD H.264 decoder is able to decode two 1080p H.264 video streams simultaneously on PlayStation® 3 (PS3) in real-time. The proposed approaches are practical on many other vector processor platforms, but not limited to Cell/B.E. processor.