NERSC, ALCF, Codeplay Partner on SYCL for Next-generation Supercomputers
March 1, 2021 – The NERSC (National Energy Research Scientific Computing Center) of the Lawrence Berkeley National Laboratory (Berkeley Lab), in partnership with the Argonne Leadership Computing Facility (ALCF) of the Argonne National Laboratory, has signed a contract with codeplay software to improve the LLVM SYCL GPU compiler functions for NVIDIA A100 GPUs.
This collaboration will help NERSC and ALCF users, and the high-performance computing community in general, build high-performance applications that are portable across multi-vendor computer architectures.
Codeplay is a UK based software company that has a long history of developing compilers and tools for various hardware architectures. The company was the main implementer of SYCL compilers and contributed significantly to the existing open source support for NVIDIA V100 GPUs as part of the DPC ++ project. NVIDIA A100 GPUs are available in the ThetaGPU extension of ALCF’s Theta and will power NERSC’s next-generation supercomputer. NERSC supercomputers are used for scientific research by researchers working in various fields such as alternative energy, the environment, high energy and nuclear physics, advanced computing, materials science, and chemistry.
Last year, 20 research teams were involved in COVID-19 simulations to analyze and develop solutions. ALCF supercomputers enable scientific research and technology by providing supercomputer resources and practical expertise to the research community. These systems have helped advance scientific computing in a number of areas through the convergence of simulation, data science, and machine learning methods.
Over the past year, ALCF supercomputers have accelerated the development of treatments and strategies to fight the COVID-19 pandemic. SYCL (pronounced “sickle”) is an open standard that is maintained under The Khronos Group. Originally released in 2014, it is a royalty-free, cross-platform abstraction layer that allows heterogeneous processor code to be written using standard ISO C ++ with the host and kernel code for an application contained in the same source file . SYCL was closely aligned with OpenCL, but has evolved into a completely different programming model over time. Under the new contract, Berkeley Lab and Argonne researchers will work with code play engineers to improve the LLVM-based open source compiler DPC ++ based on the SYCL 2020 standard to support NVIDIA A100 GPUs.
Today, the SYCL programming model supports a variety of accelerators through multiple implementations. SYCL is supported on the Department of Energy’s upcoming Aurora exascale supercomputer and can be used with mother-of-pearl with this work.
“With thousands of users and a wide variety of applications leveraging NERSC’s resources, we need to support a wide variety of programming models. In addition to policy-based approaches, we see modern language-based C ++ approaches to accelerator programming such as SYCL as an important part of our programming environment for users of Perlmutter, “said Brandon Cook, application performance specialist at NERSC.
“In addition, this work supports the productivity of developers and users of scientific applications through the performance portability of applications between Aurora and Mother of Pearl.” “The ALCF is pleased to see that Perlmutter will support the SYCL programming model through DPC ++,” said Kalyan Kumaran, technology director at ALCF.
“As the key programming model for Argonne’s upcoming exascale system, SYCL and DPC ++ will benefit the broader DOE community by enabling portability of accelerator programming models across all DOE computing facilities.” “We are pleased that the SYCL programming standard is being adopted by the US national laboratories and providing a standardized software platform for scientists developing accelerated C ++,” said Andrew Richards, Founder and CEO of Codeplay Software.
“Codeplay is a firm believer in open standards and has worked hard at Khronos to define and publish SYCL 2020, which includes many new features such as memory handling for better overall system performance.” NERSC and ALCF are user organizations of the US Department of Energy.