Khronos Sycl Language Framework for C++ Accelerators: Take Advantage of All the MIPS
summary
SYCL 2020 is a single-source C++ framework that compiles to multiple backends (CUDA/PTX, OpenMP, AMD, Level Zero, SPIR‑V) using DPC++/other compilers; it adds buffer/accessor model for implicit data movement and an optional unified shared memory explicit model. The project ships conformance tests, runs in Compiler Explorer (defaults to CPU if no GPU), has ~12 implementations and hundreds of real ports (Gromacs, TensorFlow, CERN workloads), and claims performance portability within ~10% of native on many targets.
AI-written summary · about
chapters (13)
- 0:00Introduction to SYCL
- 1:09Why SYCL for heterogeneous
- 9:22SYCL overview and history
- 11:52Heterogeneous computing history
- 18:15Heterogeneous programming challenges
- 24:22Offloading code to accelerators
- 31:55Data movement in SYCL
- 37:10SYCL compiler and ecosystem
- 41:22Parallelism and memory in SYCL
- 53:48SYCL future directions
- 58:02Hello World in SYCL
- 1:19:23SYCL implementations and portability
- 1:23:40SYCL universal programming
related talks
56mA Smooth Introduction to SYCL for C++20 afficionados
31mGPU Performance Portability Using Standard C++ with SYCL
37mC++ for multi-accelerator, multi-vendor systems
01h 02mC++ Performance Portability - A Decade of Lessons Learned
01h 22mCostless Software Abstractions for Parallel Architectures
55mC++ 20 Innovations: High-Performance Cross-Platform Architecture in C++
01h 08mAn implementer's view of Boost.SIMD
01h 01mDoes my C++ Object Model Work with a GPU and Can I Make It Safe