A Smooth Introduction to SYCL for C++20 Afficionados
summary
Overview of SYCL as a C++20-friendly accelerator layer, showing concrete tooling (Intel oneAPI backend, pre-made Docker images), core APIs (buffers and accessors, ND-range work-groups with 64 work‑items / groups of 8), and how lifetime-driven buffers replace manual copies. Demonstrates a real port of ATLAS track reconstruction (connected‑component labeling, Kalman-style fitting) and reports benchmarks where SYCL and CUDA are essentially tied on most hardware, proving nontrivial code compiles, runs, and achieves near-GPU parity.
AI-written summary · about
chapters (12)
- 0:00Introduction
- 3:15What is SYCL
- 6:49SYCL programming model
- 9:19Buffers and memory
- 13:39Accessors and sync
- 15:24Task graphs with SYCL
- 17:10Retrieving results
- 18:36SYCL and C++ algorithms
- 21:04ND-range execution
- 22:50Case study: Particle physics
- 26:30SYCL performance
- 27:11Try SYCL and Q&A
related talks
41mSYCL building blocks for C++ libraries
01h 01mA Modern C++ Programming Model for GPUs using Khronos SYCL
01h 00mSYCL: Open Your Code to More Hardware Choices
01h 33mGPU programming with modern C++
56m“Towards Heterogeneous Programming in C++"
01h 10mC++17 ParallelSTL: A Standardization Experience Report for CPU and GPU on SYCL
01h 32mHeterogeneous Programming in C++ today
05mMaximising Throughput & Minimising Latency in High-Performance C++ Code