Row fixed data stream mapping method based on graph segmentation

Abstract

The invention discloses a row fixed data stream mapping method based on graph segmentation, and mainly solves the problems of limited application scene and low utilization rate of a processing array in the existing row fixed data stream mapping method. The method comprises the following implementation steps: 1, acquiring relevant parameters of a convolutional neural network convolutional layer anda processing array; 2, generating a mapping graph according to the parameters of the convolutional layer, and determining relevant parameters of the mapping graph; 3, performing mapping graph segmentation according to the mapping graph parameters and the processing array related parameters; and 4, generating corresponding data flow mapping according to a graph segmentation result. According to the invention, the mapping graph based on the row fixed data flow is segmented and mapped according to the processing array scale; while the high data reusability characteristic of the line fixed data flow is kept, the convolutional layer of any scale can be mapped into the processing array of any scale, and the method has the advantages of being high in flexibility, high in applicability, high in processing unit utilization rate and high in processing performance, and can be used for accelerating the data processing process of the convolutional neural network.

Description

technical field [0001] The invention belongs to the field of communication technology, and in particular relates to a line-fixed data flow mapping method, which can be used for accelerating the process of processing data by a convolutional neural network. Background technique [0002] Neural network NN is the foundation of modern artificial intelligence applications. The number of applications using neural networks has increased dramatically since their breakthrough applications in areas such as speech recognition and image recognition. These neural networks are widely used in a variety of fields including autonomous driving, cancer detection and complex games. In many fields, neural networks have surpassed human accuracy and greatly improved execution efficiency. The good performance of neural network stems from its ability to use statistical learning to obtain an effective representation of the input space from a large amount of data, and then extract advanced features f...

AI Technical Summary

Problems solved by technology

However, the existing mapping method based on row-fixed data flow has a minimum requirement for the size of the processing array in the convolutional neural network processing system. This mapping method requires that the width of the processing array must not be less than the width of the convolution kernel in the convolution layer; for When the array width is larger than the convolution kernel width, the existing mapping method cannot fully utilize all processing units, resulting in idle processing units and low system operation efficiency

These issues limit the further application of row-fixed data streams in convolutional neural networks

Images