US 12,436,769 B2
Computational graph compiling and scheduling methods and related products
Yimin Zhuang, Jiangsu (CN); and Xiaobing Chen, Jiangsu (CN)
Assigned to CAMBRICON SINGGO (NANJING) TECHNOLOGY CO., LTD., Jiangsu (CN)
Appl. No. 18/706,187
Filed by Cambricon SingGo (Nanjing) Technology Co., Ltd., Jiangsu (CN)
PCT Filed Jun. 22, 2022, PCT No. PCT/CN2022/100305
§ 371(c)(1), (2) Date Apr. 30, 2024,
PCT Pub. No. WO2023/071238, PCT Pub. Date May 4, 2023.
Claims priority of application No. 202111291728.0 (CN), filed on Nov. 1, 2021.
Prior Publication US 2024/0419449 A1, Dec. 19, 2024
Int. Cl. G06F 9/38 (2018.01); G06F 9/48 (2006.01)
CPC G06F 9/3836 (2013.01) [G06F 9/4881 (2013.01)] 20 Claims
OG exemplary drawing
 
1. A computing graph runtime scheduling method, wherein a computing graph is a compiled computing graph, and the compiled computing graph comprises a plurality of execution blocks, the execution blocks represent child operations of each node in the computing graph, the method comprising:
storing information about an execution block that is required to be executed in the compiled computing graph in response to the preparatory execution of the execution block when the compiled computing graph is run, so as to construct a runtime computing graph; and
scheduling the stored execution block based on the runtime computing graph according to dependencies, so as to implement pipeline operations.