A Modern C++ Programming Model for GPUs using Khronos SYCL

on YouTube ↗01h 01m

" summary

SYCL is presented as a standard C++ single‑source model layered on OpenCL 1.2 (SYCL 1.2.1) with an implementation (ComputeCpp) that compiles device lambdas to SPIR/PTX and runs them via OpenCL. It uses buffers and accessors to build an implicit task graph and move data, queries device limits (max work‑group size), and executes an nd‑range kernel that does a parallel reduction using local memory with work‑group barriers; a host device fallback lets the same code run on CPU for debugging or as a fallback.

AI-written summary · about

" chapters (50)

1. 0:00Introduction
2. 8:40CPU vs GPU
3. 10:29CPU
4. 10:56GPU
5. 11:46Work Items
6. 12:23Memory
7. 16:19Summary
8. 16:55Writing GPU code
9. 20:45Terminology
10. 21:50Benefits of SYCL
11. 22:37SYCL Ecosystem
12. 25:49Memory Regions
13. 27:54Benefits
14. 28:36Parallel Reduction
15. 29:03Parallel Algorithms
16. 30:01Parallel Villains
17. 32:12concurrency
18. 34:01how would we paralyze
19. 34:31highlevel overview
20. 36:05lowlevel overview
21. 37:30buffer
22. 37:54cue
23. 38:23query
24. 38:58reduction
25. 39:43command group
26. 40:19kernel invocation
27. 40:56accesser
28. 41:12local access
29. 41:58parallel
30. 42:37template parameter
31. 43:11copy data from global memory
32. 43:49create lists
33. 44:22write enough statement
34. 44:34reduction operation
35. 45:12barrier
36. 46:25complete
37. 46:41return
38. 47:10ecosystem
39. 48:07post conference
40. 48:28conclusion
41. 48:57Questions
42. 49:08Is SYCL tied to specific OpenGL version
43. 49:36Accessors and permissions system
44. 50:16Flexibility
45. 51:25Support for SYCL
46. 53:43Compilation
47. 55:00Parallel STL
48. 55:50Facilities
49. 56:29Power Rangers
50. 57:46No GPU

" related talks

• 41mSYCL building blocks for C++ librariesGordon Brown · Meeting C++ 2016
• 01h 33mGPU programming with modern C++Michael S Wong · ACCU 2019
• 56m“Towards Heterogeneous Programming in C++"Gordon Brown, Michael Wong · CppCon 2016
• 28mA Smooth Introduction to SYCL for C++20 AfficionadosJoel Falcou · C++ on Sea 2023
• 01h 00mSYCL: Open Your Code to More Hardware ChoicesIgor Vorobtsov · C++ on Sea 2022
• 01h 32mHeterogeneous Programming in C++ todayMichael S Wong · ACCU 2018
• 01h 10mC++17 ParallelSTL: A Standardization Experience Report for CPU and GPU on SYCLGordon Brown, Michael Wong, Ruyman Reyes · CppCon 2017
• 58mC++ Standard ParallelismBryce Adelstein Lelbach · CppNorth 2022