Open standards underpin the future of heterogeneous programming, and as Intel Principal Engineer Jeff Hammond points out, SYCL has proven in many ways that “standards don’t cost you a thing and you get portability.” A modern C++ language, SYCL has demonstrated amazing growth and now boasts four implementations—ComputeCpp, triSYCL, hipSYCL and DPC++. hipSYCL Creator Aksel Alpay explains some of the differences between these implementations, and how hipSYCL is evolving to incorporate DPC++ features contributed to the SYCL 2020 specification, including unified shared memory (USM), reductions, group algorithms and subgroups. The two also dive into a discussion of buffers and accessors, and Aksel debunks a few myths around hipSYCL.

Guests:
Aksel Alpay, Engineer, Heidelberg University Computing Center
Jeff Hammond, Principal Engineer at Intel

To learn more:
hipSYCL (GitHub)
Dive into DPC++ (Blog)
DPC++ Data Management across Multiple Architectures (Blog)
Data Parallel C++ Book
oneAPI.com

Read/Download the transcript

Tags: accessors, Aksel Alpay, AMD, backend, buffers, C++, Center of Excellence, Clang, Codeplay, compiler, ComputeCpp, cross-architecture, CUDA, DPC++, FPGA, GPU, Heidelberg University, heterogeneous, hipSYCL, HPC, Intel, Jeff Hammond, kernel, kernel library, Khronos Group, Kokkos, LLVM, MPI, NVIDIA, oneAPI, Open Source, Open Standards, OpenACC, OpenCL, OpenMP, Parallel Research Kernels, parallelism, pointers, queue, software programming, SYCL, SYCL 2020, task graph, tool chain, triSYCL, unified shared memory, Xilinx
 
Posted in: Audio Podcast, Code Together, Intel