GPU Performance Portability Using Standard C++ with SYCL - CppCon 2022

Explore GPU performance portability using standard C++ with SYCL in this CppCon 2022 conference talk. Delve into the challenges of developing software for heterogeneous architectures with GPUs from different manufacturers. Learn how SYCL, an open standard programming interface, enables the use of standard C++ syntax and semantics to offload parallel execution to various GPUs, including those from Nvidia, AMD, and Intel. Discover techniques for achieving optimal performance across different processor architectures, including managing memory resources, implementing efficient memory access patterns, and utilizing hardware-specific features like tensor cores. Compare the execution performance of SYCL-based code with proprietary CUDA and HIP environments, and gain valuable insights into compiling C++ code for multiple GPU platforms using SYCL compilers.