Memory allocation method and device of neural network

Abstract

The invention provides a memory allocation method and device for a neural network, and the neural network needs to occupy N memory blocks during operation, and the method comprises the following steps: determining the information of the N memory blocks; sorting the N memory blocks from large to small according to the sizes of the N memory blocks to obtain a sorting sequence of the N memory blocks;allocating a first memory block to the neural network for operation of a first node set in the neural network; judging whether the nth memory block can reuse the memory blocks in the allocated set ornot, wherein the allocated set comprises a memory block allocated to the neural network; and if the nth memory block can reuse the memory block in the allocated set, allocating the reusable memory block to the neural network for operation of an nth node set in the neural network, and updating the allocated set; and sequentially executing the judgment from n = 2 to n = N according to the sorting sequence. By adopting the method and the device provided by the invention, the memory allocation of the neural network can be optimized.

Description

Technical field [0001] This application relates to the field of deep learning technology, and in particular to a memory allocation method and device of a neural network. Background technique [0002] In the current field of computer deep learning, in order to achieve better algorithm accuracy, deep learning neural networks are becoming more and more complex, and hardware capabilities limit the development of neural networks in a deeper direction, and memory optimization must be carried out. In order to achieve memory optimization, the industry usually adopts the following memory allocation strategies: [0003] Run the entire neural network, and then allocate memory for the entire neural network in the order in which it runs. For example, the neural network needs to occupy 100M memory block, 10M memory block and 50M memory block in sequence during operation. Then in the prior art, when the neural network applies for a 100M memory block, a 100M memory block can be allocated to the ...

AI Technical Summary

Problems solved by technology

However, if both the requested 10M memory block and the requested 50M memory block can reuse the allocated 100M memory block, the requested 10M memory block will reuse the allocated 100M memory block, and a 50M memory block will be allocated to the neural network. Blocks, the entire neural network needs to occupy a total of 150M memory blocks, resulting in a large amount of memory occupied by the entire neural network, and unreasonable memory allocation

Images