SYCL - A Modern Platform for Heterogeneous Architectures

About

Location: Argonne Leadership Computing Facility (ALCF), U.S.A Dates: 30th October 2019 Website: eventbrite.com

Computation offload is one of the fastest growing fields in the IT industry, demanding new levels of productivity, interoperability, and compute device support. OpenCL does a decent job here today, but there are opportunities to do better with modern C++ support. This presentation will go into detail on the SYCL programming model, which is at the core of the new oneAPI initiative. Intel is working with the community to evolve the Khronos SYCL language specification to include extensions that improve GPU offload computing. SYCL leverages all the power of modern C++ to provide productivity and deliver optimal performance yet allowing the new code to interoperate with all the existing OpenCL codebase. SYCL allows the host and device code to be kept together in the same source file enabling better error checking and optimizations. It also extends the OpenCL programming model in a number of ways to let people write efficient code targeting multiple devices.

Speaker: Konstantin (Konst) Bobrovskii, Intel