Computational requirements for media signal processing

Media signal processing is characterized by high computing power and a great deal of parallelism on low precision data, which usually consists of simple integers with 8, 12, or 16 bits of precision. Efficient processing of media signals requires modifications of the traditional data-path found in general purpose processors. In this paper, we address those differences and propose adequate solutions for efficient implementation of datapath that support main classes of media instructions.