Abstract :
The fastest parallel and pipeline computers presently being designed use interleaved memory systems with more than 16 individual memory units. A common difficulty in these machines is the alignment of data before it is arithmetically processed. Usually the arithmetic unit is used to preprocess the data. This may increase the computation time by a factor of two or more. This paper proposes a programmed memory preprocessing system which aligns the data before it is passed to the arithmetic processor.
Keywords :
Array processing, data prefetching, interleaved memories, parallel processing, pipeline processing, sparse matrix operations, table lookup.; Arithmetic; Computer aided instruction; Concurrent computing; Logic; Parallel processing; Pipeline processing; Prefetching; Routing; Sparse matrices; Table lookup; Array processing, data prefetching, interleaved memories, parallel processing, pipeline processing, sparse matrix operations, table lookup.;
