Abstract
SYCL is a platform agnostic, single-source, C++ based, parallel programming framework for developing platform independent software for heterogeneous systems.
As an emerging framework, SYCL has been under active development for several years, with multiple implementations available from hardware vendors and others.
A crucial metric for potential adopters is how mature these implementations are;
are they still improving rapidly, indicating that the space is still quite immature, or has performance improvement plateaued, potentially indicating a mature market?
This study presents a historical study of the performance delivered by all major SYCL implementations on a range of supported platforms.
We use existing HPC-style mini-apps written in SYCL, and benchmark these on current and historical revisions of each SYCL implementation, revealing the rate of change of performance improvements over time.
The data indicates that most SYCL implementations are now quite mature, showing rapid performance improvements in the past, slowing to more modest performance improvements more recently.
We also compare the most recent SYCL performance to existing well established frameworks, such as OpenCL and OpenMP.
As an emerging framework, SYCL has been under active development for several years, with multiple implementations available from hardware vendors and others.
A crucial metric for potential adopters is how mature these implementations are;
are they still improving rapidly, indicating that the space is still quite immature, or has performance improvement plateaued, potentially indicating a mature market?
This study presents a historical study of the performance delivered by all major SYCL implementations on a range of supported platforms.
We use existing HPC-style mini-apps written in SYCL, and benchmark these on current and historical revisions of each SYCL implementation, revealing the rate of change of performance improvements over time.
The data indicates that most SYCL implementations are now quite mature, showing rapid performance improvements in the past, slowing to more modest performance improvements more recently.
We also compare the most recent SYCL performance to existing well established frameworks, such as OpenCL and OpenMP.
| Original language | English |
|---|---|
| Title of host publication | Wei-Chen Lin |
| Subtitle of host publication | International Workshop on OpenCL |
| Place of Publication | New York, NY |
| Publisher | Association for Computing Machinery (ACM) |
| Pages | 1–13 |
| ISBN (Electronic) | 9781450390330 |
| DOIs | |
| Publication status | Published - 27 Apr 2021 |
Bibliographical note
Funding Information:This work makes use of Intel’s DevCloud online cluster service available for developers at https://intelsoftwaresites.secure.force. com/devcloud/oneapi. This work used the Isambard UK National Tier-2 HPC Service (http://gw4.ac.uk/isambard/) operated by GW4 and the UK Met Office, and funded by EPSRC (EP/P020224/1).
Publisher Copyright:
© 2021 ACM.