Calculation circuit and device, and calculation method

Abstract

An operation circuit, an operation device and an operation method, the method is applied to the operation device, and the method comprises the following steps: calculating a first output result through a zero check module according to initial data to be subjected to convolution operation, the first output result being used for representing whether a prediction result of 0 is obtained by performing convolution operation on the initial data; if the prediction result is not 0, inputting the initial data into an additive operation module through the control module for convolution operation to obtain a second output result; if the prediction result is 0, stopping inputting the initial data into the additive operation module through the control module, and directly taking the default operation result as a second output result; and performing AND operation on the second output result and the first output result through an additive operation module to obtain a convolution operation result. By implementing the embodiment of the invention, the calculation amount of the electronic equipment in the convolution operation process can be reduced, and the calculation efficiency of the convolution operation of the electronic equipment can be improved.

Description

technical field [0001] The present application relates to the technical field of electronic circuits, in particular to an arithmetic circuit, device, and arithmetic method. Background technique [0002] At present, in order to perform tasks such as deep learning on electronic devices, it is often necessary to design specific circuits to complete a large number of convolution operations. However, it is found in practice that conventional convolution operation circuits can only implement direct multiply-accumulate operations (the basic operation units of convolution operations are multiplication and addition), such unoptimized or simple circuit-optimized schemes can easily lead to When the electronic device performs the convolution operation, the calculation amount is too large, and the required power consumption increases, which reduces the calculation efficiency of the electronic device for the convolution operation. Contents of the invention [0003] The embodiment of th...

AI Technical Summary

Problems solved by technology

However, it is found in practice that conventional convolution operation circuits can only implement direct multiply-accumulate operations (the basic operation units of convolution operations are multiplication and addition), such unoptimized or simple circuit-optimized schemes can easily lead to The calculation amount of the electronic equipment is too large when performing the convolution operation, and the required power consumption increases, which reduces the calculation efficiency of the electronic equipment for the convolution operation.

Images