Deep convolutional neural network acceleration method and system suitable for NPU

Abstract

The embodiment of the invention provides a deep convolutional neural network acceleration method and system suitable for NPU. According to the NPU calculation unit structure, the data arrangement format of the input parameters of the convolutional layer is adjusted in advance; therefore, the problem that the calculation units are not fully utilized in the convolution calculation process is solved, and the throughput of the matrix calculation units is fully utilized. and execution assembly line of the calculation instruction in the NPU is rearranged, the convolution calculation capability of the NPU is mined, and the calculation parallelism in the deep convolutional neural network is enhanced. According to the acceleration method provided by the invention, the arrangement mode of the data streams and the instructions in the deep convolutional neural network is modified, so that the efficient and rapid convolutional neural network can be realized on the NPU.

Description

technical field [0001] Embodiments of the present invention relate to the field of computer technology, and in particular to a deep convolutional neural network acceleration method and system suitable for NPU. Background technique [0002] With the deepening of research in the field of deep learning, convolutional neural networks have been applied to various fields of computer vision. Research and experiments in recent years have shown that convolutional neural networks have shown absolute dominance over traditional computer vision algorithms in many image processing tasks such as target detection, face recognition, image classification, and semantic segmentation. [0003] While the performance of convolutional neural networks in classification and detection tasks continues to improve, the amount of parameters and calculations in neural network models are also increasing significantly. When neural network algorithms are applied, they will face many problems. Although the us...

AI Technical Summary

Problems solved by technology

Although the use of high-performance GPU (Graphics Processing Unit, graphics processing unit) cluster services can meet the requirements of computing speed, it also has some disadvantages: high latency, the road video needs to be transmitted to the computing center for processing, and the delay of transmission greatly limits the speed of road traffic. Real-time monitoring capability; high cost, need to buy or rent a large number of expensive high-performance servers; low stability, network fluctuations directly affect system stability

However, the NPU embedded AI computing node must have the disadvantage of low computing power while satisfying low power consumption and low cost.

Images