18754921. Floating-Point Dynamic Range Expansion simplified abstract (Intel Corporation)

Floating-Point Dynamic Range Expansion

Organization Name

Intel Corporation

Inventor(s)

Bogdan Mihai Pasca of Toulouse (FR)

Martin Langhammer of High Wycombe (GB)

Floating-Point Dynamic Range Expansion - A simplified explanation of the abstract

This abstract first appeared for US patent application 18754921 titled 'Floating-Point Dynamic Range Expansion

The present disclosure discusses techniques for adjusting the number representation of a variable before and/or after performing arithmetic operations on it, particularly scaling the range of a variable to a suitable representation based on available hardware in an integrated circuit device.

• Scaling input from one number format to another allows for arithmetic operations to be performed by circuitry designed for the second number format.
• The output produced by the circuitry can be scaled back to the original number format after the operations are completed.
• This technique enables emulation of arithmetic operations in one format by scaling inputs and outputs to and from another format.

1. 1. 1. Potential Applications:

This technology can be applied in various fields such as computer graphics, artificial intelligence, and scientific computing where different number representations are used for calculations.

1. 1. 1. Problems Solved:

This innovation addresses the challenge of performing arithmetic operations on variables with different number formats efficiently and accurately.

1. 1. 1. Benefits:

- Improved performance in arithmetic operations - Enhanced flexibility in handling different number representations - Increased compatibility with various hardware configurations

1. 1. 1. Commercial Applications:

Optimizing arithmetic operations in integrated circuit devices for applications in graphics processing units, machine learning accelerators, and scientific computing clusters.

1. 1. 1. Prior Art:

Researchers can explore prior patents related to number representation scaling techniques in integrated circuit devices to understand the evolution of this technology.

1. 1. 1. Frequently Updated Research:

Stay updated on the latest advancements in number representation scaling techniques in integrated circuit devices to leverage cutting-edge solutions for arithmetic operations.

1. 1. 1. 1. Questions about Number Representation Scaling Techniques:

1. How does scaling the range of a variable to a suitable representation impact the efficiency of arithmetic operations? 2. What are the key considerations when selecting the appropriate number format for scaling in integrated circuit devices?

Original Abstract Submitted

The present disclosure relates generally to techniques for adjusting the number representation (e.g., format) of a variable before and/or after performing one or more arithmetic operations on the variable. In particular, the present disclosure relates to scaling the range of a variable to a suitable representation based on available hardware (e.g., hard logic) in an integrated circuit device. For example, an input in a first number format (e.g., bfloat16) may be scaled to a second number format (e.g., half-precision floating-point) so that circuitry implemented to receive inputs in the second number format may perform one or more arithmetic operations on the input. Further, the output produced by the circuitry may be scaled back to the first number format. Accordingly, arithmetic operations, such as a dot-product, performed in a first format may be emulated by scaling the inputs to and/or the outputs from arithmetic operations performed in another format.