Vector Extension Storage Across Consecutive Addresses to Reduce Data Overhead

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing image recognition technologies face complex data processing and high data overhead due to inefficient vector handling.

Innovation Solution

A method and apparatus for vector extension and storage using an extension parameter in a vector extension instruction to simplify processing and reduce data overhead.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If traditional vector processing methods are used for image recognition, then data processing can be performed, but the processing complexity is high and data overhead is large

Engineering Contradiction:
Improvedata processing efficiencyVSAvoidprocessing complexity
Core Design Contradiction:
ProductivityVSDevice complexity

Solution Approach 1:

The patent segments vector data into multiple segments and processes each segment independently using vector extension instructions. This divides the complex data processing task into smaller, more manageable units, reducing overall processing complexity while maintaining efficiency

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent introduces a new dimension of vector extension by extending vectors to new consecutive address spaces. This dimensional expansion allows data to be organized and processed more efficiently, reducing both processing complexity and data overhead

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

2Quantity of substance

If traditional vector handling is used, then image recognition tasks can be executed, but data overhead is relatively high

Engineering Contradiction:
Improvedata overheadVSAvoidprocessing efficiency
Core Design Contradiction:
Quantity of substanceVSProductivity

Solution Approach 1:

The patent merges vector extension operations with storage operations into a single unified process. By combining these operations, the patent reduces the number of separate data handling steps, thereby reducing data overhead while maintaining processing efficiency

Inventive Principle:
Principle #5Merging (Combining)

Solution Approach 2:

The patent performs vector extension preliminarily before main processing operations. By pre-extending vectors to appropriate address spaces beforehand, the patent eliminates the need for repeated data handling during processing, thus reducing data overhead

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP4148561B1Data processing method and apparatus, and related product
Publication Date: 2025.07.16 ANHUI CAMBRICON INFORMATION TECH CO LTD
  • EP4148561B1 patent drawingFigure 1~2
  • EP4148561B1 patent drawingFigure 3~4
  • EP4148561B1 patent drawing

AI summary

A data processing method and apparatus, and a related product. The data processing method comprises: when a decoded processing instruction is a vector extension instruction, determining a source data address, a destination data address and an extension parameter of data corresponding to the processing instruction; according to the extension parameter, extending first vector data of the source data address, so as to obtain extended second vector data; storing the second vector data to the destination data address, wherein the source data address and the destination data address comprise consecutive data addresses. Vector extension and storage are implemented by means of an extension parameter in a vector extension instruction, so as to obtain extended vector data, thereby simplifying processing, and reducing data overhead.