Floating-Point Compression via Shared Exponents and Fixed Mantissas

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Conventional floating-point technologies consume significant storage space and power due to storing and processing multiple floating-point numbers individually, leading to increased operational time and power consumption, and existing compression methods like Microsoft Floating Point (MSFP) compromise accuracy with large compression errors.

Innovation Solution

A method that generates common scaling factors and compresses floating-point numbers into fixed-point mantissas, allowing for efficient compression and decompression while adjusting for error within set thresholds, using algorithms like Heuristic, Randomized, or Brute-force to optimize compression results.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If conventional floating-point technologies store and calculate multiple floating-point numbers individually, then calculation accuracy is maintained, but storage space and power consumption increase significantly

Engineering Contradiction:

Improvecalculation accuracyVSAvoidpower consumption

Core Design Contradiction:

Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The patent combines multiple floating-point numbers into a single compressed representation by extracting common exponents and storing only unique mantissas. This merging approach reduces the number of stored values while maintaining calculation accuracy through selective decomposition during operations.

Inventive Principle:

Principle #5Merging (Combining)

Solution Approach 2:

The patent creates a universal compressed format that can represent multiple floating-point numbers using shared exponent information and a set of mantissas. This multi-functional structure serves both storage efficiency and accurate reconstruction of original values for various calculation scenarios.

Inventive Principle:

Principle #6Universality (Multi-functionality)

2Measurement precision

If conventional floating-point technologies store and calculate multiple floating-point numbers individually, then calculation accuracy is maintained, but operational time increases

Engineering Contradiction:

Improvecalculation accuracyVSAvoidoperational time

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The patent merges multiple floating-point operations into a single compressed representation, reducing the number of individual operations required. By storing common exponents once and only unique mantissas, the system decreases operational time while preserving accuracy through efficient decomposition when needed.

Inventive Principle:

Principle #5Merging (Combining)

3Productivity

If Microsoft Floating Point (MSFP) compresses multiple exponents into a single exponent, then operational speed increases, but accuracy sharply declines due to large compression errors

Engineering Contradiction:

Improveoperational speedVSAvoidcalculation accuracy

Core Design Contradiction:

ProductivityVSMeasurement precision

Solution Approach 1:

The patent segments the exponent information by identifying and extracting common exponents from multiple floating-point numbers. Instead of forcing all numbers into a single exponent like MSFP, it divides them into groups with shared exponents, reducing compression errors while maintaining speed benefits.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent applies local quality by treating different groups of floating-point numbers differently based on their exponent characteristics. Groups with common exponents are compressed together, while maintaining the ability to handle groups with different exponents separately, thus preserving accuracy where needed while achieving compression where possible.

Inventive Principle:

Principle #3Local quality

4Loss of time

If Microsoft Floating Point (MSFP) compresses multiple exponents into a single exponent, then transmission time reduces, but accuracy sharply declines

Engineering Contradiction:

Improvetransmission timeVSAvoidcalculation accuracy

Core Design Contradiction:

Loss of timeVSMeasurement precision

Solution Approach 1:

The patent merges transmission efficiency with accuracy by combining common exponent information with selectively stored mantissas. This approach reduces transmission time by compressing redundant information while maintaining accuracy by preserving unique mantissa values needed for precise reconstruction of original floating-point numbers.

Inventive Principle:

Principle #5Merging (Combining)

Data Source

PatentUS20240231755A1Method for compressing floating-point numbers, computing device and computer-readable medium

Publication Date: 2024.07.11 NATIONAL TSING HUA UNIVERSITY

AI summary

A method of compressing floating points includes using an arithmetic unit to perform following steps: obtaining a plurality of floating-point numbers; producing scaling factors from the floating-point numbers; compressing each of the floating-point numbers into a plurality of fixed-point mantissas; outputting a compression result that includes the plurality of scaling factors and the plurality of fixed-point mantissas.