There are two concepts in a vector processing method. One is a vector-register method and the other is a memory-memory method. Many mechines operate based on the former concept. The CRAY-series and NEC SX/2 are typical examples. In this study, another...
There are two concepts in a vector processing method. One is a vector-register method and the other is a memory-memory method. Many mechines operate based on the former concept. The CRAY-series and NEC SX/2 are typical examples. In this study, another architecture concept that can operate in a shorter time than the former for the SIMD principle has been proposed. In this concept the data components are transmitted and operated directly in pipeline without staying in the buffer register. It has been proposed that the intermediate results are allowed to stay in the buffer for only escaping the pipeline conflicts and access conflicts and the bottleneck phenomena.