Skip to main navigation Skip to search Skip to main content

Discussion of Device-Device Collective Communication in OpenMP Target Offloading

  • Baodi Shan
  • , Mauricio Araya-Polo
  • , Johannes Doerfert
  • , Barbara Chapman
  • Stony Brook University
  • Total S.A.
  • Lawrence Livermore National Laboratory

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

OpenMP target offloading currently supports only four basic devicetodevice copy routines, a gap that forces programmers to rely on vendor libraries and limits scalability on multiGPU nodes. We introduce a prototype OpenMP-native, topology-aware collective-communication framework that: (i) exposes lightweight APIs for querying link type and peer bandwidth, (ii) offers an optional devicegroup abstraction, and (iii) implements ring and pipelinetree broadcasts whose ordering is optimized at run time. On an eightGPU MI250X node, a topologyaware halo exchange accelerates communication by up to 23.4% and averages 19.4% once the payload exceeds 8MB, and our ring broadcast outperforms RCCL for midsize messages on both AMD and NVIDIA platforms. These results demonstrate that modest runtime extensions can deliver vendor-class performance without sacrificing OpenMPs single-source portability. Our current prototype focuses on broadcast, but the same approach naturally generalizes to the full suite of collective operations.

Original languageEnglish
Title of host publicationOpenMP
Subtitle of host publicationBalancing Productivity and Performance Portability - 21st International Workshop on OpenMP, IWOMP 2025, Proceedings
EditorsYonghong Yan, Erik Saule, Michael Klemm, Bronis R. de Supinski, Jannis Klinkenberg, Swaroop Pophale
PublisherSpringer Science and Business Media Deutschland GmbH
Pages3-17
Number of pages15
ISBN (Print)9783032063427
DOIs
StatePublished - 2026
Event21st International Workshop on OpenMP, IWOMP 2025 - Charlotte, United States
Duration: Oct 1 2025Oct 3 2025

Publication series

NameLecture Notes in Computer Science
Volume16123 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference21st International Workshop on OpenMP, IWOMP 2025
Country/TerritoryUnited States
CityCharlotte
Period10/1/2510/3/25

Keywords

  • Collective Communication
  • Heterogeneous Computing
  • OpenMP

Fingerprint

Dive into the research topics of 'Discussion of Device-Device Collective Communication in OpenMP Target Offloading'. Together they form a unique fingerprint.

Cite this