Data processing apparatus and related products

Patent metadata

Field	Value
Publication number	US-12346697-B2
Application number	US-202318531734-A
Country	US
Kind code	B2
Filing date	Dec 7, 2023
Priority date	Apr 4, 2019
Publication date	Jul 1, 2025
Grant date	Jul 1, 2025

Abstract

The present disclosure provides a data processing apparatus and related products. The products include a control module including an instruction caching unit, an instruction processing unit, and a storage queue unit. The instruction caching unit is configured to store computation instructions associated with an artificial neural network operation; the instruction processing unit is configured to parse the computation instructions to obtain a plurality of operation instructions; and the storage queue unit is configured to store an instruction queue, where the instruction queue includes a plurality of operation instructions or computation instructions to be executed in the sequence of the queue. By adopting the above-mentioned method, the present disclosure can improve the operation efficiency of related products when performing operations of a neural network model.

First claim

The invention claimed is:

1. A data synchronization method performed by a first processor, comprising:
determining synchronization information of tensor data according to a descriptor of the tensor data to be synchronized, wherein the descriptor indicates a shape of the tensor data to be synchronized;
generating a synchronization instruction according to the synchronization information of the tensor data; and
sending the synchronization instruction to a second processor, wherein the synchronization instruction is used to instruct the second processor to obtain the tensor data to be synchronized according to the synchronization instruction.

2. The data synchronization method of claim 1 , wherein the synchronization information includes a storage address of the tensor data to be synchronized, wherein the generating a synchronization instruction according to the synchronization information of the tensor data includes:
when the storage address of the tensor data to be synchronized is in a shared storage space, generating the synchronization instruction according to the storage address of the tensor data to be synchronized, wherein the synchronization instruction is used to instruct the second processor to obtain the tensor data to be synchronized from the shared storage space.

3. The data synchronization method of claim 1 , wherein the synchronization information includes a storage address of the tensor data to be synchronized, wherein the generating a synchronization instruction according to the synchronization information of the tensor data includes:
when the storage address of the tensor data to be synchronized is in a non-shared storage space, storing the tensor data to be synchronized in the shared storage space; and
according to the storage address of the tensor data to be synchronized in the shared storage space, generating the synchronization instruction to instruct the second processor to obtain the tensor data to be synchronized from the shared storage space.

4. The data synchronization method of claim 1 , further comprising:
determining the descriptor of the tensor data to be synchronized according to a synchronization request instruction from the second processor.

5. The data synchronization method of claim 4 , wherein the synchronization request instruction includes data characteristics of the tensor data to be synchronized, wherein the determining the descriptor of the tensor data to be synchronized according to the synchronization request instruction from the second processor includes:
parsing the synchronization request instruction to obtain the data characteristics of the tensor data to be synchronized; and
determining the descriptor of the tensor data to be synchronized according to the data characteristics of the tensor data to be synchronized.

6. A data synchronization method performed by a second processor, comprising:
parsing a synchronization instruction from a first processor to obtain synchronization information of tensor data to be synchronized;
determining a descriptor of the tensor data to be synchronized according to the synchronization information of the tensor data to be synchronized, wherein the descriptor indicates a shape of the tensor data to be synchronized; and
obtaining the tensor data to be synchronized according to the descriptor of the tensor data to be synchronized.

7. The data synchronization method of claim 6 , wherein the synchronization information includes a storage address of the tensor data to be synchronized, wherein
the determining a descriptor of the tensor data to be synchronized according to the synchronization information of the tensor data to be synchronized includes: determining an identifier of the descriptor of the tensor data to be synchronized or content of the descriptor according to the storage address of the tensor data to be synchronized; and
the obtaining the tensor data to be synchronized according to the descriptor of the tensor data to be synchronized includes: according to the content of the descriptor of the tensor data to be synchronized, obtaining the tensor data to be synchronized from a shared storage space.

8. A data synchronization method performed by a second processor, comprising:
when there is tensor data to be synchronized, generating a synchronization request instruction, wherein the synchronization request instruction is used to instruct a first processor to determine a descriptor of the tensor data to be synchronized, and the descriptor indicates a shape of the tensor data to be synchronized; and
sending the synchronization request instruction to the first processor.

9. The data synchronization method of claim 8 , wherein the synchronization request instruction includes data characteristics of the tensor data to be synchronized.

10. The data synchronization method of claim 8 , further comprising:
parsing a synchronization instruction from the first processor to obtain synchronization information of the tensor data to be synchronized;
determining the descriptor of the tensor data to be synchronized according to the synchronization information of the tensor data to be synchronized; and
obtaining the tensor data to be synchronized according to the descriptor of the tensor data to be synchronized.

11. The data synchronization method of claim 10 , wherein the synchronization information includes a storage address of the tensor data to be synchronized, wherein
the determining the descriptor of the tensor data to be synchronized according to the synchronization information of the tensor data to be synchronized includes: determining an identifier of the descriptor of the tensor data to be synchronized or content of the descriptor according to the storage address of the tensor data to be synchronized; and
the obtaining the tensor data to be synchronized according to the descriptor of the tensor data to be synchronized includes: according to the content of the descriptor of the tensor data to be synchronized, obtaining the tensor data to be synchronized from a shared storage space.

12. A data synchronization apparatus, comprising:
a first processor:
configured to determine synchronization information of tensor data according to a descriptor of the tensor data to be synchronized, wherein the descriptor indicates a shape of the tensor data to be synchronized;
generate a synchronization instruction according to the synchronization information of the tensor data; and
send the synchronization instruction to a second processor, where the synchronization instruction is used to instruct the second processor to obtain the tensor data to be synchronized according to the synchronization instruction.

13. The data synchronization apparatus of claim 12 , wherein the synchronization information includes a storage address of the tensor data to be synchronized, wherein
when the storage address of the tensor data to be synchronized is in a shared storage space, the first processor is further configured to:
generate the synchronization instruction according to the storage address of the tensor data to be synchronized, wherein the synchronization instruction is used to instruct the second processor to obtain the tensor data to be synchronized from the shared storage space.

14. The data synchronization apparatus of claim 12 , wherein the synchronization information includes a storage address of the tensor data to be synchronized, wherein the first processor is further configured to:
when the storage address of the tensor data to be synchronized is in a non-shared storage space, store the tensor data to be synchronized in the shared storage space; and
acc

Assignees

• Cambricon Tech Corp Ltd

Inventors

• Liu Shaoli
• Wang Bingrui
• Zhou Xiaoyong
• Zhuang Yimin
• Lan Huiying
• Liang Jun
• Zeng Hongbo

Classifications

• G06F9/30036 Primary

  Instructions to perform operations on packed data, e.g. vector, tile or matrix operations · CPC title

• G06N3/063

  using electronic means · CPC title

• G06F9/3838

  Dependency mechanisms, e.g. register scoreboarding · CPC title

• G06F9/3824

  Operand accessing · CPC title

• G06N3/08

  Learning methods · CPC title