A method of calling data based on fpga off-chip memory

Abstract

The invention provides a data calling method based on an FPGA off-chip memory. The method comprises the following steps of storing the data of the feature map in a fifo line by line according to a sequence; during each read-write operation, enabling the first M fifo to output a currently stored first data, writing the first currently stored data back to the tail of the first data which is stored by the fifo and corresponds to the serial number which is smaller than the serial number of the first data by L-M by the M fifo; meanwhile, writing the first data of the L + 1th row of the feature mapinto the tail part of the data stored in the L-1 fifo, writing the first data of the L + 2 row of the feature map into the tail of the data stored in the L fifo, so that when the fifo continuously outputs the data out of the fifo group in sequence, writing the rest data of the feature map into the fifo group in sequence for reading until the data traversal of the whole feature map iscompleted; Therefore, the data of the FPGA off-chip memory are not directly called, complex address hopping is avoided, and the efficiency of calling the data of the FPGA off-chip memory is greatly improved.

Description

technical field [0001] The invention belongs to the technical field of image classification and recognition, and in particular relates to a method for calling data based on an FPGA off-chip memory. Background technique [0002] In the past five years, convolutional neural networks have achieved good results in the fields of image feature extraction, classification and recognition. Due to the flexible and changeable convolutional neural network architecture, the current convolutional neural network is mainly realized by software platforms such as CPU and GPU. However, in current engineering applications, the demand for system real-time performance and low power consumption is becoming more and more prominent. Therefore, the use of hardware platforms to accelerate the calculation of convolutional neural networks and achieve the purpose of reducing system power consumption has become increasingly popular in convolutional neural networks. Research hotspots of network in enginee...

AI Technical Summary

Problems solved by technology

The data in the off-chip memory of the FPGA is stored in banks. The main idea of ​​the current method is to store the different input feature map data in the convolutional neural network in different Banks of the off-chip memory of the FPGA as much as possible. However, this method requires frequent Jumping addresses to access off-chip memory, the reading and writing efficiency is low, especially when encountering large-scale convolutional neural network calculations, it is even more unable to meet the efficiency requirements of data calls

Images