H.264/AVC inter prediction on accelerator-based multi-core systems

The AVC video coding standard adopts variable block sizes for inter frame coding to increase compression efficiency, among other new features. As a consequence of this, an AVC encoder has to employ a complex mode decision technique that requires high computational complexity. Several techniques aimed at accelerating the inter prediction process have been proposed in the literature in recent years. Recently, with the emergence of many-core processors or accelerators, a new way of supporting inter frame prediction has presented itself. In this paper, we present a step forward in the implementation of an AVC inter prediction algorithm in a graphics processing unit, using Compute Unified Device Architecture. The results show a negligible drop in rate distortion with a time reduction, on average, of over 98.8 % compared with full search and fast full search, and of over 80 % compared with UMHexagonS search.