Multilayer neural network calculation method and device based on heterogeneous multi-core processor

• Main Authors: SONG HAIFENG, CHEN XIN, GAO XINGYU, LIN RONGFEN, SUN QIANG, SHANG HONGHUI, WANG LIFANG
• Format: Patent
• Language: Chinese; English
• Published: 27.12.2022
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; ELECTRIC DIGITAL DATA PROCESSING; PHYSICS

The invention provides a multi-layer neural network calculation method and device based on a heterogeneous multi-core processor, and relates to the technical field of neural networks, and the method comprises the steps: dividing a memory of each slave core into a calculation region and a storage region, obtaining neural network parameters, and storing the neural network parameters in the storage region of each slave core in a distributed manner; controlling the slave core to read the input data from the main memory, and storing the input data into the calculation area; and controlling the slave core to perform neural network calculation in the calculation area based on the neural network parameters stored in the storage area and the input data, and storing output data obtained by calculation into the main memory. According to the invention, the neural network parameters are stored in the slave core in a distributed manner, and data interaction with the main memory is not required during calculation of each la