Templatised soft floating-point for High-level Synthesis
File(s)float_hls.pdf (1.66 MB)
Accepted version
Author(s)
Thomas, David
Type
Conference Paper
Abstract
High-level Synthesis (HLS) tools have greatly in-creased the productivity of FPGA application development,making it possible to easily create highly-parallel application-accelerators. However, while FPGAs are known for the ability tocustomise the number representation of data-paths, most HLSwork only uses custom-precision for fixed-point representations,and for floating-point relies on the 64-, 32-, and 16-bit for-mats provided by vendors. This paper presents a solution forparametrised floating-point in HLS via C++ templates, allowingfor compile-time selection of exponent and fraction widths,including the use of mixed precisions for input arguments andresult types. By using arbitrary width integers and compile-time logic the resulting operators describe the same data-pathas an external floating-point IP generator, while still allowingthe HLS tool to perform detailed optimisation and scheduling ofthe internal components. We show that the resulting custom-width HLS cores provide similar area and performance toplatform-native vendor IP blocks, while adding full support forheterogeneous precision floating-point data-paths to HLS tools.
Date Issued
2019-06-13
Date Acceptance
2019-03-03
Citation
Proceedings of The 27th IEEE International Symposium On Field-Programmable Custom Computing Machines, 2019, pp.227-235
ISBN
9781728111315
ISSN
2576-2621
Publisher
IEEE
Start Page
227
End Page
235
Journal / Book Title
Proceedings of The 27th IEEE International Symposium On Field-Programmable Custom Computing Machines
Copyright Statement
© 2019 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works.
Sponsor
Royal Academy Of Engineering
Grant Number
LTSRF1718\14\38
Source
The 27th IEEE International Symposium On Field-Programmable Custom Computing Machines
Subjects
Science & Technology
Technology
Computer Science, Hardware & Architecture
Computer Science, Theory & Methods
Engineering, Electrical & Electronic
Computer Science
Engineering
Publication Status
Published
Start Date
2019-04-28
Finish Date
2019-05-01
Coverage Spatial
San Diego, CA, USA
Date Publish Online
2019-06-13