A kind of asynchronous processing method and device based on neural network accelerator

Abstract

The invention discloses an asynchronous processing method and device based on a neural network accelerator. The method includes: generating an asynchronous request through a main thread according to algorithmic business requirements, generating a corresponding handle number, and generating a set of command words according to the generated business request, Join the command word queue, query the command word queue, if there is only one set of command words at present, set the set of command words to the execution state, start the hard core of the neural network accelerator, otherwise set the set of command words to the initial state and return. After the execution of the hard core, a hardware interrupt is generated and a soft interrupt is started, and the command word queue is queried. If there is a command word, it is taken out from the queue, and whether it is in the initial state is judged. If yes, it is set to the execution state and the hard core is started. Character. The main thread queries whether the hard core of the operation is completed according to the handle number. The above operation realizes the asynchronous operation process of multiple network algorithms, and handles asynchronous operations through hard core interrupts, which accelerates network performance and reduces processor bandwidth occupation.

Description

technical field [0001] The invention relates to the technical field of deep learning, in particular to a processing method of a neural network accelerator, and an asynchronous processing method and device based on the neural network accelerator. Background technique [0002] With the development of artificial intelligence, the design of deep learning networks is becoming more and more complex. Without neural network accelerator hardware, it is impossible to meet the needs of embedded performance. In the field of security, the main detection objects are people, cars, and objects. At the same time of detection, people, cars, and objects need to be tracked. Therefore, multiple deep learning algorithms must be integrated to solve the problem of detecting people and tracking people. When there are multiple people, it needs to be tracked one by one, and then the feature matching strategy is used to ensure that the same person is tracked; the existing invention patent with publicat...

AI Technical Summary

Problems solved by technology

[0002] With the development of artificial intelligence, the design of deep learning network is becoming more and more complex. Without neural network accelerator hardware, it cannot meet the needs of embedded performance.

In the security field, the main detection objects are people, vehicles, and objects. While detecting people, vehicles, and objects, it is necessary to track people, vehicles, and objects. Then it is necessary to integrate multiple deep learning algorithms to solve the detection and tracking of people. When a frame of image screen When there are many people, it is necessary to track them one by one, and then use the feature matching strategy to ensure that the same person is tracked; the existing invention patent with the publication number of CN111679815A discloses an asynchronous operation method and device, storage medium and related equipment. The invention patent solves the problem of multi-layer nesting, separates the main logic and processing logic, solves the problem of poor flexibility and small scope of application, an asynchronous processing method, device, computer equipment and storage medium for Lua voice, adopts the callback and Promise scheme, and aims at the traditional Asynchronous callback method solves the problem of bloated callback code

Images