Acceleration calculation method of neural network, accelerator and computer readable storage medium

Abstract

The invention provides an acceleration calculation method of a neural network, an accelerator and a computer readable storage medium. The method comprises the following steps: storing an input matrix in a first data cache region; acquiring a coefficient matrix, decomposing the coefficient matrix into a plurality of coefficient sub-matrixes with a preset size, sliding the coefficient sub-matrixes relative to an input matrix at a preset step length, and performing point multiplication summation calculation on the coefficient sub-matrixes and the input matrix by applying at least one processing unit of a multiply-add array every time the coefficient sub-matrixes slide; obtaining a part and a matrix corresponding to each coefficient sub-matrix; and carrying out accumulation calculation on the corresponding parts of the plurality of coefficient sub-matrixes and the matrix to obtain an accumulation calculation result, carrying out feature calculation according to the accumulation calculation result, and storing a calculation result of the one-layer model in a second data cache region. The invention further provides a neural network accelerator for implementing the method and a computer readable storage medium. The method can reduce the number of data carrying times during neural network calculation, and improves the energy consumption ratio.

Description

technical field [0001] The present invention relates to the technical field of neural networks, in particular to a neural network acceleration calculation method, a neural network accelerator and a computer-readable storage medium for realizing the method. Background technique [0002] In recent years, with the development of artificial intelligence technology, neural network technology has also developed rapidly. Neural network (Neuron Network) is a multi-layer artificial intelligence model, including deep neural network (DNN), convolutional neural network (CNN), recurrent neural network (RNN) and so on. With the increase of the calculation amount of the neural network, the requirements for the computing power of the processor and data storage are also increasing day by day. Neural network accelerators can accelerate the computational efficiency of neural networks at the hardware level, so they are more and more widely used in various fields. [0003] However, the calcula...

AI Technical Summary

Problems solved by technology

However, this kind of accelerator does not consider the problem of multiple data transfers, nor does it solve the problems of increased energy consumption and low computing efficiency caused by data transfers, which affects the computing efficiency of neural networks and leads to low energy consumption of electronic devices.

Images