ISC High Performance 2023 is held as an in-person conference and exhibition focusing on HPC, machine learning, data analytics, and quantum computing. Most parts of the event will be available as on-demand content for attendees who cannot travel this year.

 

Khronos Related Presentations

Tutorial: Heterogeneous Programing in Modern C++ with Khronos SYCL

• Date: Sunday, May 21, 2023
• Time: 14:00 - 18:00
• Location: Hall Y5 - 2nd floor
• Speakers: Rod Burns, Codeplay Software; Aksel Alpay, Univeristy of Heidelberg; Igor Baratta, University of Cambridge; Tom Deakin, University of Bristol; Igor Vorobstov, Intel
• Abstract: Parallel programming can be used to take advantage of heterogeneous architectures including GPUs, FPGAs, XPUs, IPUs, TPUs or special units on CPUs to significantly increase the performance of applications. SYCL is an open standard programming model that is defined by the industry and lets developers support many of these processors from different vendors using a single code base and only modern standard C++ code. This tutorial will give software developers the knowledge they need to begin developing parallel applications using C++ and the SYCL programming model. Our goal is to equip attendees with the skills they need to build highly performant applications that can be used in the fields of HPC and AI and deployed to multiple hardware platforms. We will cover the fundamentals of the SYCL programming model before moving to more advanced topics. We will explore how SYCL can be used to write serious applications, covering intermediate to advanced features of SYCL as well as some of the tools and libraries that support SYCL application development. This is a hands-on tutorial, attendees will work through exercises that represent key design patterns encountered by people who program heterogeneous systems and deploy this code to multiple processors from different vendors.

BOF: Heterogeneous Programing in Modern C++ with Khronos SYCL

• Date: Tuesday, May 23, 2023
• Time: 13:00 - 14:00
• Location: Hall G1 - 2nd floor
• Speakers: Aksel Alpay, Univeristy of Heidelberg; Tom Deakin, University of Bristol; Michael Wong, Codeplay Software; James Brodman, Intel
• Abstract: SYCL is an open standard modern C++ programming model for programming common hardware accelerators used in HPC. SYCL provides scientific programmers a vendor neutral way to code key algorithms for offload onto accelerators providing portability between hardware. SYCL is a royalty free standard under Khronos and available for anyone to join. There are multiple implementations currently available, both open source and commercial, supporting a wide variety of hardware. After ISC 2022’s highly successful SYCL BoF where several hundred attended, we will offer updates from the Working Group of current standardization plans, highlight new features, and showcase workloads such as GROMACS which have converted to SYCL before opening the floor to the audience. We will discuss SYCL features, portability, tools and ecosystem as well as what ISC attendees need to drive SYCL forward. There will be a panel of experts with representation from SYCL implementers and application developers, who will take audience questions on all aspects of SYCL including specific implementations. The goal is both to introduce people to SYCL, and serve those who are already working in SYCL to deliver HPC workloads. SYCL has emerged as a standard programming model enabling modern C++ workloads to be dispatched to heterogeneous devices for EXAscale computing. It has gained traction with the rise of implementations targeting accelerators from all major vendors. SYCL is a Khronos Standard language maintained by multiple members leading the delivery of modern C++ code for dispatch and aims to converge with and help define heterogeneous parallelism in ISO C++.

Poster: A SYCL Extension to Enable Approximate Computing on Heterogeneous Systems

• Date: Monday, May 22nd, 3:00 PM to Wednesday, May 24th, 5:00 PM
• Location: Foyer D-G - 2nd Floor
• Presenter: Lorenzo Carpentieri, University of Salerno
• Abstract: Approximate computing tries to balance performance or energy consumption gains and accuracy loss of applications. Several domains, such as image/video processing and neural networks, allow for a loss of accuracy in the final output, enabling the chance to improve performance and energy consumption through approximation. Specifically, in order to enable approximate computing techniques on heterogeneous systems we propose SYprox, a C++ library that enhances the SYCL programming model by providing support for the development of heterogeneous approximate applications. SYprox extends the SYCL buffer class with a pbuffer that supports various perforation strategies to skip data computation. It also implements two types of reconstruction approaches to approximate data skipped by the perforation decreasing the error introduced. Finally, we show a preliminary error and performance evaluation of the approximation techniques on three stencil applications implemented using the SYprox library.