Neural network compiling optimization method and system

Abstract

The invention provides a neural network compiling optimization method. The method comprises the steps of obtaining a calculation graph with a set data structure according to a to-be-compiled deep learning model; fusing the one or more preprocessing layers into a plurality of fusion layers; obtaining an operator calculation sequence in the fusion layer according to the inter-operator dependency relationship in the fusion layer; obtaining input and output calling times and a splitting strategy of an in-layer operator; obtaining corresponding system overhead values of the plurality of fusion layers on the simulation hardware platform; taking the fusion layer corresponding to the minimum value in the system overhead values of the plurality of fusion layers as the current fusion layer; and compiling the to-be-compiled deep learning model according to the current fusion layer. By fusing multiple layers of operators of the neural network, a calculated intermediate result is stored on a chip instead of being read and written through a memory, so that the memory access requirement can be effectively reduced, and the execution efficiency of the system is improved. Meanwhile, the invention further provides an optimization system for neural network compilation.

Description

Technical field[0001]The present invention relates to a reconfigurable processor and application field, and is applied to a compiler and compilation process of a reconfigurable compiler. The present invention specifically relates to an optimization method and system of neural network compilation.Background technique[0002]Depth neural networks have been applied to multiple fields, such as face recognition, machine translation, recommendation system and other fields. With the improvement of the complexity of the depth neural network model, the complexity of the calculated complexity is also increased while obtaining a better effect on the corresponding task. In order to improve the calculation efficiency of the depth neural network, it can make the corresponding task more efficiently, and it is necessary to optimize the calculation of complex neural networks based on compilation optimization techniques of neural networks.[0003]In the current computer architecture, the execution speed ...

AI Technical Summary

Problems solved by technology

However, while this method improves the system execution speed, it also increases the memory burden, which does not significantly improve the system execution efficiency, but also increases the system burden.

Images