Vector Register Compression Layout for Lower Wire Congestion

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing AI processors face significant wire congestion and increased area due to the implementation of data compression instructions using a large number of multiplexers, especially with high vector processing parallelism, leading to manufacturing difficulties and costs.

Innovation Solution

Implement a divide-and-conquer approach by dividing source vector data into n sub-vectors, compressing each sub-vector using n sets of multiplexers, and merging the results to reduce the number of multiplexers and wires, thereby reducing congestion and area.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Loss of substance

If conventional compression algorithms (RLE, Huffman, LZW) are used on vector data, then compression ratio is improved, but processing speed deteriorates and patent infringement risks increase

Engineering Contradiction:

Improvedata sizeVSAvoidprocessing speed

Core Design Contradiction:

Loss of substanceVSProductivity

Solution Approach 1:

The patent segments the vector data processing into distinct phases: identification of compression opportunities, selection of appropriate compression algorithms, execution of compression, and decompression. This segmentation allows each phase to be optimized independently, improving overall processing efficiency while maintaining compression effectiveness.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent introduces an intermediary compression system that sits between the vector data source and destination, automatically selecting and applying appropriate compression algorithms (RLE, Huffman, LZW, or arithmetic coding) based on data characteristics. This intermediary layer handles the complexity of multiple algorithms while presenting a simple interface, resolving the contradiction between compression ratio and processing speed.

Inventive Principle:

Principle #24Intermediary (Mediator)

2Loss of substance

If conventional compression algorithms are used on vector data, then compression ratio is improved, but patent infringement risks increase

Engineering Contradiction:

Improvedata sizeVSAvoidpatent compliance

Core Design Contradiction:

Loss of substanceVSReliability

Solution Approach 1:

The patent extracts and implements only the essential compression techniques (RLE, Huffman, LZW, arithmetic coding) without incorporating patented algorithms, thereby achieving compression functionality while avoiding patent infringement. The system deliberately excludes patented methods and focuses on implementing proven, non-patented compression approaches.

Inventive Principle:

Principle #2Taking out (Extraction)

Solution Approach 2:

The patent employs compression algorithms that are in the public domain or have expired patents, effectively using 'disposable' compression methods that do not require licensing or worry about patent expiration. This approach ensures long-term reliability and compliance without the risk of future patent enforcement.

Inventive Principle:

Principle #27Cheap short-living objects (Disposable)

3Loss of substance

If vector data is compressed before transmission or storage, then bandwidth and storage requirements are reduced, but decompression complexity increases

Engineering Contradiction:

Improvedata sizeVSAvoiddecompression complexity

Core Design Contradiction:

Loss of substanceVSDevice complexity

Solution Approach 1:

The patent performs preliminary actions during the compression phase by selecting and applying the most appropriate compression algorithm based on data characteristics. This preliminary optimization ensures that the compressed data requires minimal complex processing during decompression, as the right algorithm was chosen upfront to match the data patterns.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent inverts the traditional approach by focusing optimization efforts on the compression side rather than the decompression side. By selecting algorithms and parameters that optimize compression ratios, the system naturally simplifies the decompression process, as the compressed data structure is designed to be efficiently reconstructed.

Inventive Principle:

Principle #13The other way round (Inversion)

Data Source

PatentEP4398096B1Vector data compression method, vector data decompression method, apparatus, and device

Publication Date: 2026.05.06 TENCENT TECHNOLOGY (SHENZHEN) CO LTD

Figure 1~2
Figure 3
Figure 4~5

AI summary

This application relates to the field of data processing technologies, and provides a method and an apparatus for compressing vector data, a method and an apparatus for decompressing vector data, and a device. The method includes: storing, by a source vector register, source vector data, the source vector data being divided into n source sub-vectors, the n source sub-vectors being in a one-to-one correspondence with n sets of multiplexers (401); selectively arranging, by an ith set of multiplexers in the n sets of multiplexers, valid elements in an ith source sub-vector in the source vector data, to obtain an ith target sub-vector (402); shifting and merging, by a data merging apparatus, n target sub-vectors corresponding to the n source sub-vectors, to obtain target vector data (403); and storing, by a target vector register, valid elements in the target vector data (404). Through the method, the apparatus, and the device, a congestion level of wires can be effectively reduced, thereby significantly reducing an area of a processor.