Data processing method and system for semi-precision floating-point divider

Abstract

The invention provides a data processing method of a semi-precision floating-point divider, which comprises the following steps of: respectively carrying out bit separation processing on first data and second data to respectively obtain first bit separation data and second bit separation data, and carrying out data adjustment on the second bit separation data to obtain adjustment data, and performing iterative calculation processing on the adjustment data to obtain iterative data, performing hybrid operation on the first bit separation data and the iterative data to obtain hybrid bit separation data, and performing data reset adjustment on the hybrid bit separation data to obtain calculation result data. According to the data processing method of the semi-precision floating-point divider, through iterative computation processing, the operation period is shortened, and the operation efficiency is improved. The invention also provides a semi-precision floating-point divider data processing system for achieving the semi-precision floating-point divider data processing method.

Description

technical field [0001] The invention relates to the field of half-precision floating-point number dividers, in particular to a data processing method and system for a half-precision floating-point number divider. Background technique [0002] In the design of Convolutional Neural Networks (CNN), as the number of processed image pixels increases and the types of discrimination increase, the number of layers of the network model is also gradually increasing. [0003] From the perspective of easy design, most convolutional neural network designs use fixed-point numbers to process data. The mathematical operation speed is fast, and the algorithm circuit design is mature. The fixed-point division circuit has also been successfully applied to CNN neural network design. However, the out-of-bounds error will continue to increase with the increase in the number of layers. It is easy to judge the final classification of the image, and the floating-point number method is used. Because ...

AI Technical Summary

Problems solved by technology

[0004] At present, the calculation circuit of floating-point numbers is mature, but there are still problems of too many operation cycles and long operation time on the divider. For example, in the divider of Xilinx FPGA, the division of FP16 needs 17 clock cycles, and the cycle is long. low efficiency

Images