## Khronos APIs for Heterogeneous Compute and Safety: SYCL and SYCL SC MICHAEL WONG, NEVIN LIBER & VERENA BECKHAM ## Open Standards and Open Source implementations, community driven Open cross-company 18c308786b2683cbd327df/p3_6.jpg) SYCL 1.2 C++11 Single source programming  SYCL 1.2.1 C++11 Single source programming 683cbd327df/p3_8.jpg) SYCL 2020 C++17 Single source programming Many backend options  SYCL 202X C++20 Single source

+21 ## Heterogeneous Modern C++ with SYCL 2020 GORDON BROWN, MICHAEL WONG, NEVIN LIBER, TOM DEAKIN 20 21 October 24-29 ## Distinguished Engineer Chair of SYCL Heterogeneous Programming Language ISO Argonne and Oak Ridge National Laboratories Award Codeplay $ ^{®} $ Software to Further Strengthen SYCL $ ^{™} $ Support Extending the Open Standard Software for AMD GPUs 17 June 2021 ![Image](/uplo Implementing OpenCL™ and SYCL™ for the popular RISC-V processors will make it easier to port existing HPC and AI software for embedded systems ## NERSC, ALCF, Codeplay Partner on SYCL for Next-generation Supercomputers

## +23 ## The Absurdity of Error Handling: Finding a Purpose for Errors in Safety-Critical SYCL ## ERIK TOMUSK ## codeplay $ ^{®} $ Enabling AI & HPC To Be Open, Safe & Accessible To All to expanding the open ecosystem for heterogeneous computing. Through our involvement in oneAPI and SYCL governance, we help to maintain and develop open standards. ![Image](/uploads/documents/d/2/9/1/ involved in two research projects - SYCLOPS and AERO, both funded by the Horizon Europe Project. ## SYCL|SC - SYCL is an abstraction layer for running C++ code on accelerators like GPUs • SC stands for safety-critical

vectorization and numerics. Contribution into SYCL  ## Distinguished Engineer Chair of SYCL Heterogeneous Programming Language ISO Argonne and Oak Ridge National Laboratories Award Codeplay $ ^{®} $ Software to Further Strengthen SYCL $ ^{™} $ Support Extending the Open Standard Software for AMD GPUs 17 June 2021 ![Image](/uplo Implementing OpenCL™ and SYCL™ for the popular RISC-V processors will make it easier to port existing HPC and AI software for embedded systems ## NERSC, ALCF, Codeplay Partner on SYCL for Next-generation Supercomputers

strategies: C-style programming with OpenMP, native mechanisms in modern C++, as well as through Kokkos and Sycl. ## Discussion An important corner of scientific computing is concerned with sparse matrices, in The second extension is Sycl, a Khronos open standard, which can be considered a library built on standard C++, but in practice requires compiler support. The major proponent of Sycl is Intel, which provides work: The ‘AdaptiveCPP’ open source alternative to Sycl is very hard to install. I hope to find time for another attempt. This will allow testing Sycl on two different compilers: Intel with optimized TBB

MDSPAN A DEEP DIVE SPANNING C++, KOKKOS & SYCL NEVIN “:-)” LIBER Computer Scientist nliber@anl.gov MDSPAN: A Deep Dive Spanning C++, Kokkos & SYCL ## motivations: people who know more than me

transform(fun_) } ); } private: V base_; F fun_; }; ## SYCL Codebase (shp) Data automatically distributed using namespace dr::shp; amongst multiple GPUs - Distributed }); return reduce(par_unseq, z.begin(), z.end(), T(0), std::plus()); } ## SYCL Codebase (shp) Data automatically distributed using namespace dr::shp; amongst multiple GPUs -

CPU操作 CUDA GPU操作 OpenCL GPU操作 XLA:CPU XLA:GPU 通信网络 计算设备 gRPC RDMA GPU (StreamExecutor) GPU (SYCL) CPU (Eigen) Android ## TensorFlow 数据流图介绍 ## TensorFlow 数据流图是一种声明式编程范式 声明式编程与命令式编程的多角度对比 |编

☐ Ask questions during the talk □ Audience? □ Only Cuda □ What is Cuda ☐ Do not ask me about SYCL, ROCm, Vulkan, ... ☐ You can tell me about afterwards ## Why write programs for CPU and GPU ## ☐