<p>With fast advancements in detector and software technologies used in large-scale physics experiments, the requirements for the performance of computing systems used for both online and offline data processing have grown drastically. The industry offers a wide range of hardware devices to select from when designing such systems, but in turn, this imposes a technical challenge. To achieve efficient device utilization, deep knowledge of a particular technology is required. While one can achieve a high optimization level, it yields a hard-to-maintain codebase with limited upgradeability or capability to migrate to other platforms. It becomes a significant challenge, especially if available human resources are limited. In this paper, we present the application of the SYCL heterogeneous programming model that can help overcome those drawbacks. By the introduction of an abstraction layer, the source code is decoupled from the computing device architecture, and the developer can select the compilation target. The same codebase can, therefore, be executed on any supported hardware platform. We use the particle track reconstruction algorithm developed for the Forward Tracker in the <InlineEquation ID="IEq1"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="41781_2025_143_Article_IEq1.gif" Format="GIF" Height="17" Rendition="HTML" Resolution="72" Type="Linedraw" Width="24" /> </InlineMediaObject> <EquationSource Format="TEX">\(\overline{\text{ P }}\)</EquationSource> <EquationSource Format="MATHML"><math> <mover> <mrow> <mspace width="0.333333em" /> <mtext>P</mtext> <mspace width="0.333333em" /> </mrow> <mo>¯</mo> </mover> </math></EquationSource> </InlineEquation>ANDA experiment to demonstrate portability between various computing architectures and performance evaluation of the solution.</p>

错误:搜索内容不能为空,请输入英文关键词
错误:关键词超出字数限制,请精简
高级检索

Performance Portability of the Particle Tracking Algorithm Using SYCL

  • Bartosz Soból,
  • Michael Papenbrock,
  • Tobias Stockmanns,
  • Grzegorz Korcyl

摘要

With fast advancements in detector and software technologies used in large-scale physics experiments, the requirements for the performance of computing systems used for both online and offline data processing have grown drastically. The industry offers a wide range of hardware devices to select from when designing such systems, but in turn, this imposes a technical challenge. To achieve efficient device utilization, deep knowledge of a particular technology is required. While one can achieve a high optimization level, it yields a hard-to-maintain codebase with limited upgradeability or capability to migrate to other platforms. It becomes a significant challenge, especially if available human resources are limited. In this paper, we present the application of the SYCL heterogeneous programming model that can help overcome those drawbacks. By the introduction of an abstraction layer, the source code is decoupled from the computing device architecture, and the developer can select the compilation target. The same codebase can, therefore, be executed on any supported hardware platform. We use the particle track reconstruction algorithm developed for the Forward Tracker in the \(\overline{\text{ P }}\) P ¯ ANDA experiment to demonstrate portability between various computing architectures and performance evaluation of the solution.