Low-precision floating-point formats: from general-purpose to application-specific

Amir Sabbagh Molahosseini; Leonel Sousa; Azadeh Alsadat Emrani Zarandi; Hans Vandierendonck

doi:10.1007/978-3-030-98347-5_4

Low-precision floating-point formats: from general-purpose to application-specific

Amir Sabbagh Molahosseini, Leonel Sousa^*, Azadeh Alsadat Emrani Zarandi, Hans Vandierendonck

^*Corresponding author for this work

School of Electronics, Electrical Engineering and Computer Science

Research output: Chapter in Book/Report/Conference proceeding › Chapter (peer-reviewed) › peer-review

Abstract

Low-precision floating-point formats are used to enhance performance by providing the minimal precision required for the application’s requirements. General-purpose formats, such as IEEE Half-Precision floating-point (FP) format, are not practical for all applications. When dealing with small bit widths, careful adjustment of precision and dynamic range is required, and applications require bespoke number format. This chapter provides a comprehensive review of low-precision FP formats identifying: (1) the numerical features (e.g., dynamic range and precision) of each format, (2) their usage in the target applications, (3) their accuracy and performance. Finally, guidelines are provided to design high-performance and efficient applications via customized FP formats.

Original language	English
Title of host publication	Approximate computing
Editors	Weiqiang Liu, Fabrizio Lombardi
Publisher	Springer Nature Switzerland AG
Chapter	4
Pages	77-98
ISBN (Electronic)	9783030983475
ISBN (Print)	9783030983468, 9783030983499
DOIs	https://doi.org/10.1007/978-3-030-98347-5_4
Publication status	Published - 22 Aug 2022

Keywords

Computer Arithmetic
Approximate Computing
Transprecise Computing
Floating-Point Number Formats

Access to Document

10.1007/978-3-030-98347-5_4Licence: Unspecified

Cite this

@inbook{c867649d27b94e96a1f340c34fe31df0,

title = "Low-precision floating-point formats: from general-purpose to application-specific",

abstract = "Low-precision floating-point formats are used to enhance performance by providing the minimal precision required for the application{\textquoteright}s requirements. General-purpose formats, such as IEEE Half-Precision floating-point (FP) format, are not practical for all applications. When dealing with small bit widths, careful adjustment of precision and dynamic range is required, and applications require bespoke number format. This chapter provides a comprehensive review of low-precision FP formats identifying: (1) the numerical features (e.g., dynamic range and precision) of each format, (2) their usage in the target applications, (3) their accuracy and performance. Finally, guidelines are provided to design high-performance and efficient applications via customized FP formats.",

keywords = "Computer Arithmetic, Approximate Computing, Transprecise Computing, Floating-Point Number Formats",

author = "{Sabbagh Molahosseini}, Amir and Leonel Sousa and {Emrani Zarandi}, {Azadeh Alsadat} and Hans Vandierendonck",

year = "2022",

month = aug,

day = "22",

doi = "10.1007/978-3-030-98347-5_4",

language = "English",

isbn = "9783030983468",

pages = "77--98",

editor = "Weiqiang Liu and Fabrizio Lombardi",

booktitle = "Approximate computing",

publisher = "Springer Nature Switzerland AG",

address = "Switzerland",

}

Low-precision floating-point formats: from general-purpose to application-specific. / Sabbagh Molahosseini, Amir; Sousa, Leonel; Emrani Zarandi, Azadeh Alsadat et al.

Approximate computing. ed. / Weiqiang Liu; Fabrizio Lombardi. Springer Nature Switzerland AG, 2022. p. 77-98.

Research output: Chapter in Book/Report/Conference proceeding › Chapter (peer-reviewed) › peer-review

TY - CHAP

T1 - Low-precision floating-point formats: from general-purpose to application-specific

AU - Sabbagh Molahosseini, Amir

AU - Sousa, Leonel

AU - Emrani Zarandi, Azadeh Alsadat

AU - Vandierendonck, Hans

PY - 2022/8/22

Y1 - 2022/8/22

N2 - Low-precision floating-point formats are used to enhance performance by providing the minimal precision required for the application’s requirements. General-purpose formats, such as IEEE Half-Precision floating-point (FP) format, are not practical for all applications. When dealing with small bit widths, careful adjustment of precision and dynamic range is required, and applications require bespoke number format. This chapter provides a comprehensive review of low-precision FP formats identifying: (1) the numerical features (e.g., dynamic range and precision) of each format, (2) their usage in the target applications, (3) their accuracy and performance. Finally, guidelines are provided to design high-performance and efficient applications via customized FP formats.

AB - Low-precision floating-point formats are used to enhance performance by providing the minimal precision required for the application’s requirements. General-purpose formats, such as IEEE Half-Precision floating-point (FP) format, are not practical for all applications. When dealing with small bit widths, careful adjustment of precision and dynamic range is required, and applications require bespoke number format. This chapter provides a comprehensive review of low-precision FP formats identifying: (1) the numerical features (e.g., dynamic range and precision) of each format, (2) their usage in the target applications, (3) their accuracy and performance. Finally, guidelines are provided to design high-performance and efficient applications via customized FP formats.

KW - Computer Arithmetic

KW - Approximate Computing

KW - Transprecise Computing

KW - Floating-Point Number Formats

U2 - 10.1007/978-3-030-98347-5_4

DO - 10.1007/978-3-030-98347-5_4

M3 - Chapter (peer-reviewed)

SN - 9783030983468

SN - 9783030983499

SP - 77

EP - 98

BT - Approximate computing

A2 - Liu, Weiqiang

A2 - Lombardi, Fabrizio

PB - Springer Nature Switzerland AG

ER -

Low-precision floating-point formats: from general-purpose to application-specific

Abstract

Keywords

Access to Document

Fingerprint

Cite this