[Submitted on 29 Jul 2026]
Abstract:OpenMP 5.0はrequiresディレクティブを通じてUnified Shared Memory (USM)機能を導入しました。この機能は、アクセラレータとホスト間のユニークで共通のアドレス空間を提供し、異なるデバイス上で同じメモリアドレスにアクセス(参照)できるようにすることで、OpenMPプログラミングモデルの採用を簡素化します。これにより、アドレス空間間の整合性を維持するための明示的なデータ転送の負担を軽減し、OpenMPによるアクセラレータへのアプリケーションの迅速なプロトタイピングと移植を容易にします。
In this paper, we introduce the Intel implementation for USM. We briefly discuss its implementation in the software stack (OS kernel, compiler, and runtime), then assess its adoption complexity in existing HPC applications using OpenMP for accelerators, and, finally, evaluate the performance of these applications when adopting USM on an Intel Battlemage GPU. USM is not expected to grant performance uplifts to already optimized applications with explicit, granular data-motion control and our results show an overhead with a geometric mean below 1.2x (1.03x seems achievable with further optimizations). Yet, in this paper we show there exist applications that benefit from this feature, making it attractive even for already ported applications.
Submission history
From: Harald Servat [view email]
[v1]
Wed, 29 Jul 2026 08:04:24 UTC (613 KB)
0 Comments
Log in to join the conversation.No comments yet. Be the first to share your thoughts.