DATA PROCESSOR AND DATA PROCESSING SYSTEM

摘要

PROBLEM TO BE SOLVED: To provide a data processor for improving computation parallelism and improving computation throughput by making a floating point inner product computing element into SIMD and to provide a computation system for substantially improving the computation parallelism per instruction while maintaining the efficiency of a floating point 4-element vector inner product computing element. SOLUTION: The floating point 4-element vector inner product computing element is defined by the minimum width (32 bits in the case of single precision) even when the utilization of a wide arithmetic system becomes possible and the inner product computing element is made into the SIMD. Thus, the computation parallelism is substantially improved by the synergistic effect of parallelism improvement by both of the inner product computing element and SIMD making. When the floating point computing element for obtaining the sum of the inner product of a 4-element vector and a scalar is made into 4 parallel SIMD, it becomes 32 FLOPS per cycle.