Does my C++ Object Model Work with a GPU and Can I Make It Safe
summary
Claims that C++’s local reasoning breaks on GPUs unless the API changes: OpenCL and Vulkan expose explicit resource graphs so RAII and invariants get tangled, and naive C++ wrappers leak ownership or lose performance. The uploader shows code with OpenCL and argues SYCL (slides: https slides meetingcpp com) style reference semantics and opaque interfaces let a runtime decouple host objects from scarce backend resources, enabling predictable memory and timing bounds. The claim: a safety‑certifiable GPU API is technically feasible, not yet broadly implemented.
AI-written summary · about
chapters (12)
- 0:00Introduction
- 1:29C++ object model
- 10:45Class invariance limits
- 13:43C++ object model benefits
- 14:03GPU programming basics
- 21:11OpenCL challenges
- 32:10Sickle abstraction layer
- 45:55Sickle’s semantics
- 46:16Functional safety primer
- 50:31Safety-critical determinism
- 55:51C++ and GPU programming
- 1:00:06Conclusion
related talks
37mC++ for multi-accelerator, multi-vendor systems
01h 00mC++ on GPUs done right?
31mGPU Performance Portability Using Standard C++ with SYCL
56mA Smooth Introduction to SYCL for C++20 afficionados
40mMaking GPUs Feel Native in Rust
01h 29mDelivering Safe C++
01h 03mThe Landscape of Parallelism
01h 26mKhronos Sycl Language Framework for C++ Accelerators: Take Advantage of All the MIPS