Multi-format Operand Circuit for Mixed Precision Computing

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing computing systems face inefficiencies and latency issues when performing computations involving different floating-point number formats, particularly in complex systems and machine learning applications.

Innovation Solution

A multi-format source operand circuit that allows direct operation with operands in different number formats without converting them to a common format, using specialized processing circuits and instruction sets to perform operations efficiently.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If software-based support for different floating point number formats is used, then adaptability to different formats is improved, but processing latency increases and performance deteriorates

Engineering Contradiction:

Improvesupport for different floating point number formatsVSAvoidprocessing latency

Core Design Contradiction:

Adaptability or versatilityVSLoss of time

Solution Approach 1:

The patent replaces software-based floating point format support with hardware-based processing circuits. The processing circuit includes dedicated circuitry for detecting and handling different floating point formats (e.g., IEEE 754 single precision, double precision, and custom formats) directly in hardware, eliminating the need for software interpretation and conversion, thereby reducing latency while maintaining format adaptability

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The processing circuit is designed with universal functionality to handle multiple floating point number formats simultaneously. It includes format detection logic and multiple processing paths that can accommodate different precision requirements (single precision, double precision, and custom formats) within a single hardware unit, enabling the system to adapt to various computational needs without format conversion overhead

Inventive Principle:

Principle #6Universality (Multi-functionality)

2Reliability

If operands are converted to a common number format before operation, then computational compatibility is improved, but processing time increases and efficiency deteriorates

Engineering Contradiction:

Improvecomputational compatibilityVSAvoidprocessing efficiency

Core Design Contradiction:

ReliabilityVSProductivity

Solution Approach 1:

The processing circuit performs preliminary detection and classification of operand formats before the computational operation. By detecting the floating point formats of both operands in advance and routing them to appropriate processing paths, the system prepares the operands for direct operation without requiring conversion, thus maintaining compatibility while improving efficiency

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The processing circuit segments the computational path into separate handling mechanisms for different format combinations. Instead of converting all operands to a single common format, the circuit creates dedicated processing paths for matching format pairs (e.g., single-singles, double-doubles, custom-custom), allowing each segment to operate independently and efficiently without unnecessary conversions

Inventive Principle:

Principle #1Segmentation

Data Source

PatentUS20250130794A1Multi-format operand circuit

Publication Date: 2025.04.24 ADVANCED MICRO DEVICES INC

AI summary

The disclosed processing circuit can perform an operation with a first operand having a first number format and a second operand having a second number format by directly using the first operand in the first number format and the second operand in the second number format to produce an output result. Various other methods, systems, and computer-readable media are also disclosed.