Image data processing method and device, computer equipment and storage medium

Abstract

The invention relates to an image data processing method, which comprises the steps of obtaining image data of N to-be-processed images, and the image data of each to-be-processed image comprising C parts of image channel data; storing the image data into C storage areas of a local memory, wherein N storage units in each storage area store one piece of image channel data of N to-be-processed images respectively; when a data reading instruction occurs, determining a historical address according to a current address carried by the data reading instruction, thereby reading target image channel data from a local memory based on the storage address; arranging the read target image channel data in the form of a two-dimensional matrix, enabling the data corresponding to the same channel to be inthe same matrix row, and enabling the data corresponding to the same to-be-processed image in the adjacent matrix rows to be distributed in two adjacent matrix columns; and sequentially transmitting each column of data arranged in a two-dimensional matrix form to the pulsation array according to a time sequence for operation to obtain an operation result. By adopting the method, the data processing efficiency can be improved.

Description

Technical field [0001] The present application relates to the field of computer technology, and more particularly to an image data processing method, apparatus, computer equipment and storage media. Background technique [0002] With the development of computer technology, the emergence of machine learning techniques, various machine learning models based on machine learning techniques can improve model accuracy by processing a large amount of calculation. In the actual process, for machine learning models involving convolution calculation can often parallel processing to improve computational efficiency-based graphics processor (Graphics Processing Unit, GPU) multithreaded SIMD module. Wherein, SIMD (Single Instruction MultipleData, single instruction multiple data) using a controller to control a plurality of processors, while each of the operations are performed on the same set of data (also known as "vector data") such that technology to achieve spatial parallelism. [0003] ...

AI Technical Summary

Problems solved by technology

When it comes to matrix multiplication and multidimensional convolution in deep learning calculations, the SIMD module often encounters that the speed of data access is much lower than the speed of data processing.

Therefore, when performing computationally intensive tasks such as matrix multiplication or multidimensional convolution, there will be a problem of low data processing efficiency

Images