Looking for breakthrough ideas for innovation challenges? Try Patsnap Eureka!

Method, device, device and medium for reduction calculation of multidimensional tensors

A protocol and multi-dimensional technology, applied in the direction of multi-program device, calculation, program control design, etc., can solve the problems of unsatisfactory computing power requirements, slow speed, etc., and achieve the effect of improving the calculation speed of the protocol

Active Publication Date: 2019-10-25
上海燧原智能科技有限公司
View PDF2 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0003] In the field of artificial intelligence and high-performance computing, protocol computing is a commonly used computing mode. The computing speed of protocol computing directly affects the overall computing power output of the system. However, the traditional protocol computing method is slow and cannot meet the current rapid growth. computing power demand

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • Method, device, device and medium for reduction calculation of multidimensional tensors
  • Method, device, device and medium for reduction calculation of multidimensional tensors
  • Method, device, device and medium for reduction calculation of multidimensional tensors

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0028] figure 1 It is a flow chart of a multi-dimensional tensor reduction calculation method provided by Embodiment 1 of the present invention. This embodiment is applicable to the case of performing reduction calculations on multi-dimensional tensors. It can be executed by a computing device, which can be implemented in the form of software and / or hardware, and can generally be integrated in a processor.

[0029] Such as figure 1 As shown, the method of this embodiment specifically includes:

[0030] S110. Obtain the multi-dimensional tensor to be reduced and calculated, and the corresponding dimension information and non-reducible dimension information of the multi-dimensional tensor.

[0031] A multidimensional tensor refers to a tensor with multiple dimensions. The dimension to be reduced is predetermined according to the specific application of the multidimensional tensor. Among the multiple dimensions, all the dimensions except the dimension to be reduced are non-redu...

Embodiment 2

[0058] figure 2 It is a flow chart of a multi-dimensional tensor calculation method provided by Embodiment 2 of the present invention. This embodiment is embodied on the basis of the above-mentioned embodiment, wherein, according to the low-dimensional information of the two-dimensional tensor, the two-dimensional tensor The volume is divided into at least two data areas, specifically:

[0059] According to the low-dimensional information of the two-dimensional tensor and the available number of hardware computing units, the two-dimensional tensor is evenly divided into at least two data areas, wherein the number of data areas matches the available number of hardware computing units.

[0060] Such as figure 2 As shown, the method of this embodiment specifically includes:

[0061] S210. Acquire the multidimensional tensor to be reduced and calculated, and the dimension information to be reduced and dimension information not to be reduced corresponding to the multidimensional ...

Embodiment 3

[0083] On the basis of the above-mentioned embodiments, this embodiment provides a specific implementation manner, using multi-dimensional tensor T Dimensions are [ R 1_1, R 1_2, …, R 1_nr1, U 1_1, U 1_2, …, U 1_nu1, R 2_1, R 2_2, …, R 2_nr2, U 2_1, U 2_2, …, U 2_nu2, …, R n_1, R n_2, ..., R n_nrn, U n_1, U n_2, ..., U n_nun] as an example for explanation.

[0084] Such as Figure 5 As shown, the method of this embodiment specifically includes:

[0085] S310. Obtain the multi-dimensional tensor to be reduced and calculated, and the corresponding dimension information and non-reduced dimension information of the multi-dimensional tensor.

[0086] assumed dimension R 1_1, R 1_2,..., R 1_nr1, R 2_1, R 2_2,..., R 2_nr2, ..., R n_1, R n_2,..., R n_nrn is the dimension to be reduced, U 1_1, U 1_2, …, U 1_nu1, U 2_1, U 2_2, …, U 2_nu2, …, U n_1, U n_2, ..., U n_nun is the dimension not to be reduced.

[0087] S320. Merge adjacent dimensions to be...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

PUM

No PUM Login to View More

Abstract

The embodiment of the invention discloses a protocol calculation method and device for multi-dimensional tensor, equipment and a medium. The method comprises the steps of obtaining a multi-dimensionaltensor to be subjected to protocol calculation, and to-be-protocolled dimension information and non-to-be-protocolled dimension information of the multi-dimensional tensor; according to the to-be-protocolled dimension information and the non-to-be-protocolled dimension information, preprocessing the dimension information of the multi-dimensional tensor, and adjusting the multi-dimensional tensorinto a two-dimensional tensor form according to a preprocessing result; wherein the high dimension of the two-dimensional tensor is a to-be-stipulated dimension, and the low dimension of the two-dimensional tensor is a non-to-be-protocolled dimension; dividing the two-dimensional tensor into at least two data areas according to the low-dimensional information of the two-dimensional tensor, and calling a matched number of hardware calculation units to carry out protocol calculation on each data area; and sequentially combining the protocol calculation subarea vectors obtained by the hardware calculation units to obtain a protocol calculation result corresponding to the multi-dimensional tensor. According to the technical scheme, the protocol calculation speed of the multi-dimensional tensoris increased.

Description

technical field [0001] Embodiments of the present invention relate to the technical field of intelligent computing, and in particular, to a method, device, device and medium for multi-dimensional tensor reduction calculation. Background technique [0002] With the vigorous development of artificial intelligence and high-performance computing, people's demand for computer computing power has also increased exponentially. At the same time, the growth rate of computing power has also greatly promoted the development of artificial intelligence and high-performance computing. [0003] In the field of artificial intelligence and high-performance computing, protocol computing is a commonly used computing mode. The computing speed of protocol computing directly affects the overall computing power output of the system. However, the traditional protocol computing method is slow and cannot meet the current rapid growth. computing power requirements. Contents of the invention [0004...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

Application Information

Patent Timeline
no application Login to View More
Patent Type & Authority Patents(China)
IPC IPC(8): G06F9/50
CPCG06F9/5027
Inventor 谢启凯
Owner 上海燧原智能科技有限公司
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Patsnap Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Patsnap Eureka Blog
Learn More
PatSnap group products

Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic, Popular Technical Reports.

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap|About US| Contact US: help@patsnap.com