US 12,174,746 B2
Data processing method and chip, device, and storage medium
Yu Meng, Shenzhen (CN)
Assigned to TENCENT TECHNOLOGY (SHENZHEN) COMPANY LIMITED, Shenzhen (CN)
Filed by TENCENT TECHNOLOGY (SHENZHEN) COMPANY LIMITED, Guangdong (CN)
Filed on Oct. 15, 2021, as Appl. No. 17/502,218.
Application 17/502,218 is a continuation of application No. PCT/CN2020/118893, filed on Sep. 29, 2020.
Claims priority of application No. 201911235760.X (CN), filed on Dec. 5, 2019.
Prior Publication US 2022/0035745 A1, Feb. 3, 2022
Int. Cl. G06F 12/0875 (2016.01); G06N 3/04 (2023.01)
CPC G06F 12/0875 (2013.01) [G06N 3/04 (2013.01); G06F 2212/608 (2013.01)] 20 Claims
OG exemplary drawing
 
1. A data processing method, performed by a computer device, the method comprising:
reading parallel control code;
reading, as instructed by the parallel control code, first data that has been cached in a data cache space, processing the read first data, and outputting processed first data to the data cache space; and
simultaneously, with processing the read first data, moving second data from a data storage space to the data cache space as instructed by the parallel control code, the second data being next data of the first data;
simultaneously, with processing the second data, moving third data from the data storage space to the data cache space as instructed by the parallel control code, the third data being next data of the second data, and, while moving the third data from the data storage space to the data cache space, outputting the processed first data to the data cache space;
reading, according to a data processing code corresponding to a convolutional layer based on the convolutional layer, the first data that has been cached in the data cache space, performing convolution processing on the first data, and outputting the first data after the convolution processing to the data cache space; and
simultaneously, with reading the first data according to the data processing code corresponding to the convolutional layer, reading, according to a data processing code corresponding to a pooling layer based on the pooling layer, the third data after the convolution processing that has been cached in the data cache space, performing pooling processing on the third data after the convolution processing, and outputting the third data after the pooling processing to the data cache space, the third data being previous data of the first data in the plurality of pieces of data.