Dynamic Compilation for Transprecision Applications on Heterogeneous Platform - CEA - Commissariat à l’énergie atomique et aux énergies alternatives Access content directly
Journal Articles Journal of Low Power Electronics and Applications Year : 2021

Dynamic Compilation for Transprecision Applications on Heterogeneous Platform

Abstract

This article describes a software environment called HybroGen , which helps to experi2 ment binary code generation at run-time. As computing architectures are getting more complex, the application performances become data-dependent. The proposed experimental platform is helpful in programming applications that can be reconfigured at run-time in order adapted for a new data environment. HybroGen platform is adapted to heterogeneous architectures and can generate instructions for different target. This platform allows to go farther than classical JIT compilation in many directions: the code generator is smaller by three orders of magnitude, faster by three orders of magnitude compared to JIT (Just-In-Time) platforms and allows making code transformation that is impossible in traditional compilation scheme like code generation for non Von Neumann accelerators or dynamic code transformations for transprecision. The latter will be illustrated in a code example: the square root with Newton’s algorithm. We also illustre the proposed HybroGen platform with two others examples: a multiplication with a specialization on a value determine at run-time and a convertion of degree Celcius to degree Fahrenheit. This article presents a proof of concept of the proposed HybroGen platform in terms of its functionalities, and demonstrate the working status.
Fichier principal
Vignette du fichier
main.pdf (488.19 Ko) Télécharger le fichier
Origin : Files produced by the author(s)

Dates and versions

cea-03313560 , version 1 (04-08-2021)

Licence

Attribution

Identifiers

Cite

Julie Dumas, Henri-Pierre Charles, Kévin Mambu, Maha Kooli. Dynamic Compilation for Transprecision Applications on Heterogeneous Platform. Journal of Low Power Electronics and Applications, 2021, 11 (3), https://doi.org/10.3390/jlpea11030028. ⟨10.3390/jlpea11030028⟩. ⟨cea-03313560⟩
65 View
62 Download

Altmetric

Share

Gmail Facebook X LinkedIn More