Year: 2024
Type: conference-paper
Abstract: This study presents a new construct in OpenMP designed to facilitate the implementation of approximate computing techniques within parallel programming environments. By integrating approximation methods such as task dropping, loop perforation, and floating-point relaxation, the proposed construct aims to enhance performance and energy efficiency while maintaining acceptable accuracy levels. Experimental results on benchmark applications demonstrate a trade-off of up to 490.83%, with 55.1% of quality loss, highlighting the potential and limitations of approximate computing in parallel contexts.
Authors: João B. Oliveira, Rogério A. Gonçalves, João Fabrício Filho
Institution: Universidade Tecnológica Federal do Paraná
Language: English
Subfield: Hardware and Architecture
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
Award: G198732175
OpenAlex ID: W4403758176