An architecture for high speed scientific calculation-oriented data flow machine

This paper presents the architecture of a data flow processor array system for large scale scientific calculations. High performance level is achieved by adjusting a system structure to application programs. In addition, flexibility is attained through execution based on a data driven control mechanism. This paper also describes several simulation results for the programs which calculate partial differential equations and matrix multiplication. These results clarify that the proposed architecture can efficiently exploit parallelism in the practical and important region. (author)