Bit-Split Data Compression for 10-Bit GPU Image Values

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current data compression techniques for image data, particularly in GPUs, face challenges in efficiently compressing and decompressing 10-bit data values without significantly increasing silicon area, as existing units are often configured for 8-bit data and lack efficient methods for handling 10-bit values effectively.

Innovation Solution

A method involving dividing n-bit data values into two subsets, where the first subset comprises the most significant bits and the second subset comprises the least significant bits, with different compression schemes applied to each subset, allowing for efficient compression and decompression of 10-bit data using existing 8-bit compression modules with minimal additional hardware.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If compression units are configured in hardware to compress and decompress 10-bit data values in addition to 8-bit data values, then support for 10-bit data is achieved, but silicon area increases significantly

Engineering Contradiction:
Improvesupport for 10-bit dataVSAvoidsilicon area
Core Design Contradiction:
Adaptability or versatilityVSArea of stationary object

Solution Approach 1:

The patent divides the n-bit data value into two subsets: the first subset containing the n-2 most significant bits and the second subset containing the two least significant bits. This segmentation allows existing 8-bit compression modules to handle the first subset while a separate module handles only the two least significant bits, avoiding the need for complete 10-bit compression hardware and thereby reducing silicon area.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent makes existing 8-bit compression modules multi-functional by having them process the most significant n-2 bits of n-bit data values. This universal approach allows the same hardware to handle both 8-bit data (when n=8) and 10-bit data (when n=10), eliminating the need for separate dedicated 10-bit compression units and reducing overall silicon area.

Inventive Principle:
Principle #6Universality (Multi-functionality)

2Productivity

If data compression is applied to reduce memory bandwidth, then latency and power consumption are reduced, but image quality may deteriorate

Engineering Contradiction:
Improvememory bandwidth efficiencyVSAvoidimage quality
Core Design Contradiction:
ProductivityVSManufacturing precision

Solution Approach 1:

The patent applies different compression schemes to different parts of the data: the first compression module handles the most significant n-2 bits with one compression scheme, while the second compression module handles the two least significant bits with a different compression scheme. This local differentiation allows optimization for both compression efficiency and image quality preservation in different bit positions.

Inventive Principle:
Principle #3Local quality

Solution Approach 2:

The patent changes the compression parameters by using different compression schemes for different subsets of bits. The first compression module uses parameters optimized for the most significant bits, while the second compression module uses parameters optimized for the least significant bits, allowing the system to achieve both high compression ratios and preserved image quality.

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentEP4254807A1Data compression and decompression
Publication Date: 2023.10.04 IMAGINATION TECH LTD
  • EP4254807A1 patent drawingFigure 1A
  • EP4254807A1 patent drawingFigure 1B
  • EP4254807A1 patent drawingFigure 2

AI summary

A computer-implemented method for compressing an n-bit data value, the method comprising dividing the n bits of the data value into a first subset of bits and a second subset of bits, the first subset comprising the n - 2 most significant bits of the data value and the second subset comprising the two least significant bits of the data value; performing compression of the first subset using a first compression module; and performing compression of the second subset using a second compression module, the first and second compression modules implementing different compression schemes.