Executable program generation device and method for neural network processor

Abstract

The invention discloses an executable program generation device and method for a neural network processor, and the executable program generation device comprises a source program segmentation module which receives a source file as an input, identifies and extracts the positions of a code segment and a data segment according to the format in the source file, and generates an intermediate file containing the code segment and an intermediate file containing the data segment; a data processing module which is used for inputting an intermediate file containing data, processing placement of the dataand outputting memory allocation information and a data placement mode; and a neural network algorithm mapping module which is used for inputting an intermediate file containing the code segment, mapping a neural network algorithm represented by a block in the code into an algorithm flow composed of macro statements, and mapping the algorithm flow into a hardware-related instruction. The device provides a convenient and fast method for using the neural network processor for a user.

Description

technical field [0001] The present disclosure relates to the field of computers, and further relates to the field of artificial intelligence. Background technique [0002] Deep neural network algorithm is a very popular machine learning algorithm recently, which is widely used in various fields, such as image recognition, speech recognition, natural language processing, etc. Since the deep neural network has achieved good results in various tasks, a variety of network structures and algorithms emerge in an endless stream, which brings challenges to programming development. For the neural network processor, due to its unique hardware structure and the characteristics of intensive calculation and memory access of the neural network algorithm running on it, programming becomes more complicated and difficult. The neural network processors proposed now are programmed by handwritten instructions, which is very time-consuming and labor-intensive, and it is also prone to errors and...

AI Technical Summary

Problems solved by technology

Since the deep neural network has achieved good results in various tasks, a variety of network structures and algorithms emerge in an endless stream, which brings challenges to programming development

For the neural network processor, due to its unique hardware structure and the computationally intensive and memory-intensive characteristics of the neural network algorithm running on it, programming becomes more complicated and difficult.

The neural network processors currently proposed are programmed by handwritten instructions, which is very time-consuming and labor-intensive, and it is also prone to errors and difficult to debug.

[0003] In the process of implementing the present disclosure, the applicant found the following problems in the above-mentioned prior art: the neural network processor lacks effective programming means and efficient code generation device, which makes its programming very difficult and the generated code is inefficient. It is difficult to give full play to the advantages of the neural network processor

Images