Processing apparatus and processing method

Patent metadata

Field	Value
Publication number	US-11698786-B2
Application number	US-201916697727-A
Country	US
Kind code	B2
Filing date	Nov 27, 2019
Priority date	Apr 19, 2017
Publication date	Jul 11, 2023
Grant date	Jul 11, 2023

Abstract

The present disclosure provides a computation device and method. The device may include an input module configured to acquire input data; a model generation module configured to construct an offline model according to an input network structure and weight data; a neural network operation module configured to generate a computation instruction based on the offline model and cache the computation instruction, and compute the data to be processed based on the computation instruction to obtain a computation result; and an output module configured to output a computation result. The device and method may avoid the overhead caused by running an entire software architecture, which is a problem in a traditional method.

First claim

What is claimed is:

1. A computation device comprising:
an input circuit configured to acquire input data, wherein the input data includes data to be processed;
a model generation circuit configured to construct an offline model according to the input data;
a neural network operation circuit configured to:
determine a computation instruction based on the offline model,
cache the computation instruction, and
compute the data to be processed according to the computation instruction to obtain a computation result; and

a model parsing circuit configured to generate the computation instruction according to the offline model; and
a neural network processor configured to cache the computation instruction for subsequent computation, or retrieve a cached computation instruction based on a determination that the input data only includes the data to be processed, and perform computation on the data to be processed according to the computation instruction to obtain the computation result.

2. The computation device of claim 1 , wherein the input data includes offline model data.

3. The computation device of claim 1 , wherein,
the input data includes a network structure, and weight data.

4. The computation device of claim 3 , wherein, the network structure is a neural network structure including at least one of the following:
AlexNet, GoogleNet, ResNet, VGG, R-CNN, GAN, LSTM, RNN, and ResNe.

5. The computation device of claim 1 , further comprising:
a control circuit configured to determine content of the input data,
wherein, based on a determination that the input data includes a network structure and weight data, the control circuit is configured to:
instruct the input circuit to transmit the network structure and the weight data in the input data to the model generation circuit,

instruct the model generation circuit to generate the offline model based on the weight data and the network structure,
control the neural network operation circuit to compute the data to be processed based on the generated offline model,
wherein, based on a determination that the input data includes the offline model, the control circuit is configured to:
instruct the input circuit to transmit the data to be processed and the offline model to the neural network operation circuit,

control the neural network operation circuit to generate the computation instruction based on the offline model and cache the computation instruction,
instruct the neural network operation circuit to compute the data to be processed based on the computation instruction, and
wherein, based on a determination that the input data only includes the data to be processed, the control circuit is configured to:
instruct the input circuit to transmit the data to be processed to the neural network operation circuit,
instruct the neural network operation circuit to retrieve a cached computation instruction and to compute the data to be processed based on the computation instruction.

6. The computation device of claim 1 , wherein, the neural network operation circuit further includes:
an instruction caching circuit configured to write the computation instruction for the neural network processor to retrieve, and
a data caching circuit configured to write the input data.

7. The computation device of claim 1 , wherein, the offline model is a neural network model including at least one of the following:
Cambricon_model, AlexNet_model, GoogleNet_model, VGG_model, R-CNN_model, GAN_model, LSTM_model, RNN_model, and ResNet_model.

8. A neural network computation method, comprising:
acquiring, by an input circuit, input data that includes data to be processed;
determining, by a model generation circuit, an offline model according to the input data;
determining, by a neural network operation circuit, a computation instruction according to the offline model for subsequent computation to call;
calling, by the neural network operation circuit, the computation instruction;
computing, by the neural network operation circuit, the data to be processed;
generating, by a model parsing circuit, the computation instruction according to the offline model;
caching, by a neural network processor, the computation instruction for subsequent computation, or retrieving, by the neural network processor, a cached computation instruction based on a determination that the input data only includes the data to be processed;
performing, by the neural network processor, computation on the data to be processed according to the computation instruction to obtain the computation result.

9. The method of claim 8 , wherein the input data includes a network structure and weight data.

10. The method of claim 9 , wherein when the input data includes a network structure and weight data, the determining an offline model according to the input data includes:
constructing the offline model according to the network structure and the weight data.

11. The method of claim 9 , wherein, the network structure is a neural network structure including at least one of the following:
AlexNet, GoogleNet, ResNet, VGG, R-CNN, GAN, LSTM, RNN, and ResNe.

12. The method of claim 8 , wherein the input data includes offline model data.

13. The method of claim 8 , wherein the calling the computation instruction includes:
performing network operations according to the computation instruction based on a determination that the input data only includes the data to be processed and does not include an offline model, or
performing network operations according to the computation instruction based on a determination that the offline model is determined.

14. The method of claim 8 , wherein, the offline model is a neural network model, and the neural network model includes at least one of the following:
Cambricon_model, AlexNet_model, GoogleNet_model, VGG_model, R-CNN_model, GAN_model, LSTM_model, RNN_model, and ResNet_model.

Assignees

• Shanghai Cambricon Inf Tech Co Ltd

Inventors

• Liu Shaoli
• Li Wei
• Zhi Tian
• Chen Tianshi

Classifications

• G06F9/30 Primary

  Arrangements for executing machine instructions, e.g. instruction decode (for executing microinstructions G06F9/22) · CPC title

• G06F9/30007

  to perform operations on data operands · CPC title

• G06F7/46

  using electromechanical counter-type accumulators · CPC title

• G06N3/08

  Learning methods · CPC title

• G06N3/04

  Architecture, e.g. interconnection topology · CPC title