| 摘要 |
A video platform architecture provides video processing using parallel vector processing. The video platform architecture includes a plurality of video processing modules, each module including a plurality of processing elements (PEs). Each PE provides parallel vector processing. Specifically, means are provided to read all elements of one or two source vector registers in each PE simultaneously, process the read elements by a set of arithmetic-logical units (ALUs), and write back all results to one of the vector registers, all of which occurs in one PE cycle. To provide such parallel vector processing capabilities, the datapath of each PE is built as a set of identical PE processing slices, each of which includes an integer arithmetic-logical unit (ALU), a vector register bank, and a block register bank. A block/vector register bank holds all I elements of row J in a two-dimensional IxJ data blocks for all block/vector registers provided by the architecture. |
