Highlighting SYCL and oneAPI at SC22
10 November 2022
The International Conference for High Performance Computing, Networking, Storage, and Analysis, (known as SC22) returns and promises again to be a huge event.
Hosted in Dallas Convention Center from 13th – 18th November, the program attracts an amazing array of presentations, Birds of a Feather sessions, awards, and poster and panel sessions. While the key theme of the event is around supercomputers, the most powerful processing systems in the world, the topics vary from semiconductors, memory storage, cooling, and ever increasingly, software. Software is where Codeplay gets involved, providing an open standards based platform for portability and performance with industry leaders. Codeplay continues to promote the SYCL programming model as the ideal solution for application developers, especially with oneAPI platform.
A major topic this year is the oneAPI Community Forum, led by Codeplay. This organization is bringing the developer community together to drive an open standards and open source software ecosystem. Participants contribute to the definition of standard interfaces for commonly used libraries used in HPC and AI applications. Ensure you link with someone from Codeplay to discuss how to join the forum, in particular you can join the reception linked below.
Join these free social events to meet and chat with the Codeplay team and wider HPC community:
- Multicore World Happy Hour sponsored by Codeplay, on Sunday 13th November.
- oneAPI Community Forum Reception, Monday 14th November. RSVP here.
There is a strong agenda at SC22 as always, with many papers and presentations covering the latest research from the SYCL community.
Here is a list of the most interesting sessions:
- The tutorial “Hands-On HPC Application Development Using C++ and SYCL” will introduce SYCL and provide programmers with a solid foundation they can build on to gain mastery of the language. Presented by Hugh Delaney (Software Engineer, Codeplay), Dounia Khaldi (Senior HPC/DL Compiler Developer, Intel), Michael Wong (Distinguished Engineer, Codeplay) James Reinders (Chief Evangelist of Intel Software Products), and Ronan Keryell (Fellow Software Development Engineer at AMD AECG Research Labs)
- A Birds of a Feather session will discuss SYCL further. Hosted by Khronos (standards body defining SYCL and other popular open standards), the chairman of SYCL, Michael Wong will discuss “Khronos SYCL: Current and Future Directions”, with Tom Deakin (Assistant Professor in Advanced Computer Systems at the University of Bristol), and James Brodman (Principal Engineer, Intel)
- Another Birds of a Feather session entitled “SYCL Portability: Tips and Tricks for Porting High Performance Libraries and Applications” with Noah Clemons (Developer Tools Customer Engineering Division at Intel), Hartwig Anzt (Junior Professor at the Steinbuch Centre for Computing, Karlsruhe Institute of Technology), Yu-Hsiang Mike Tsai (Ph.D. student in Hartwig Anzt's group at the Karlsruhe Institute of Technology), and Istvan Reguly (Associate Professor with Pazmany Peter Catholic University (PPCU), Budapest)
- “SYCL and RISC-V” by Michael Wong (Fellow, Codeplay) will walk through the process of using SYCL on a RISC-V processor taking advantage of the Vector Extension (RVV). Codeplay cooperated with Intel and Andes to produce a demonstration called Pathfinder running SYCL.
- “The oneAPI Software Abstraction for Accelerated Computing” panel session discusses portability and performance achieved with SYCL and the oneAPI libraries. The panel is hosted by Henry Gabb, James Reinders, Aksel Alpay (researcher and software engineer from Heidelberg University), Kumudha Narasimhan (Senior Software Engineer at Codeplay), Ronan Keryell (Principal Software Engineer at Xilinx) and Zheming Jin (Ph.D. in Computer Science and Engineering from the University of South Carolina).
- “Towards Cross-Platform Portability of Coupled-Cluster Methods with Perturbative Triples using SYCL” by Abhishek Bagusetty, Ajay Panyala, Gordon Brown, Jack Kirk. They present the cross-platform performance achieved using SYCL implementations and bottlenecks when executed on Nvidia and AMD GPUs, with comparison to their own native programming models.
- “Toward Performance Portability of AI Graphs Using SYCL” by Kumudha Narasimhan (Codeplay), Ouadie El Farouki, Mehdi Goli (Codeplay), Muhammad Tanvir, Svetlozar Georgiev, Isaac Ault, presenting a SYCL backend for ONNX Runtime as a possible solution towards the performance portability of deep learning algorithms using existing state-of-the-art SYCL-DNN and SYCL-BLAS libraries to invoke tuned SYCL kernels for DNN operations.
- “A First Step towards Support for MPI Partitioned Communication on SYCL-programmed FPGAs” by Steffen Christgau, Marius Knaust and Thomas Steink, where they demonstrate that the communication concept in v4 of the Message Passing Interface (MPI) can also be implemented for SYCL-programmed FPGAs.
- “Evaluating ISO C++ Parallel Algorithms on Heterogeneous HPC Systems” by Wei-Chen Lin, Tom Deakin and Simon McIntosh-Smith, where review benchmarks on platforms such as OpenMP, CUDA and SYCL to show that C++17 parallel algorithms are able to achieve competitive performance across multiple mini-apps on many platforms, with some notable exceptions.
- “Evaluating Nonuniform Reduction in HIP and SYCL on GPUs” by Zheming Jin and Jeffrey Vetter. They explain the migration experience from CUDA to HIP and SYCL, evaluate the performance of the reduction on an NVIDIA V100 GPU and an AMD MI100 GPU, and provide feedback on improving portability for the development of the SYCL programming model.
- “Runtimes Systems for Extreme Heterogeneity: Challenges and Opportunities” panel session with Pedro Valero-Lara, Jeffrey Vetter, Michael Wong, Johannes Doerfert, Rosa Badia, George Bosilca, Olivier Aumage, Sunita Chandrasekaran and Jesus Labarta. They discuss the latest runtime evolution and the impact on applications i.e. how easily and rapidly a science team can develop or port a workflow to a new platform, and how well the resulting implementation makes use of the platform and its resources