Explore the Khronos SYCL language framework for C++ accelerators in this comprehensive conference talk from ACCU 2023. Learn how to leverage all available MIPS in your machine using a single programming language and support various offload devices like FPGAs, GPUs, matrix/tensors, and DSPs. Discover why US National Labs are choosing SYCL as a standard programming model for exascale computing, and how it enables maximum performance through a combination of host and accelerator devices. Delve into SYCL's architecture as a framework built on top of Modern C++, backed by an open Khronos standard, and its ability to enable ML frameworks and Standard C++ code. Examine the various backends supported by SYCL, including CUDA, PTX, OpenMP, AMD, NEC, Huawei, Kokkos, Raja, and TBB. Explore real-world use cases involving complex modern C++, such as PyTorch, Blender, ray-tracing, Flashlight ML, Eigen, Tensorflow, Gromacs, and CERN's ATLAS experiment. Gain insights into the latest features of SYCL 2020 and its relationship to ISO C++, as well as its potential in the era of software and hardware codesign with extreme heterogeneity.
Overview
Syllabus
Khronos Sycl Language Framework for C++ Accelerators: Take Advantage of All the MIPS Michael S Wong
Taught by
ACCU Conference