Concurrency TS2: Improved C++ Concurrency and Lock-free Programming
summary
Concurrency TS2 claims standardized low‑level tools: Hazard Pointers and RCU for safe reclamation with retire semantics (hazard protection ~4 ns, RCU reader construction ~3 ns in hot cache) and bounded unreclaimed objects, Bytewise Atomic Memcpy to enable efficient sequence‑lock style reads, and an Asymmetric Fence that makes the common reader path ~2 ns vs ~32 ns for a full atomic fence on Haswell. The paper is sent as a PDTS toward ISO and some pieces are already moving into C++26; reclamation can be deferred or offloaded to a background thread.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 6:43Hazard pointers and RCU
- 25:59RCU API details
- 31:27Hazard pointer API
- 39:17Asymmetric fences
- 45:41Bitwise atomics
- 50:01Synchronized values
- 54:12C++26 integration
- 56:04Q&A
related talks
01h 08mA lock-free concurrent toolkit for hazard pointers
59mThreads and Locks must Go
01h 02mA Lock-Free Atomic Shared Pointer in Modern Cpp
01h 04m[old version]The C++20 Synchrononization Library
52mBest Practices for Concurrency
01h 04mThe C++20 Synchronization Library
01h 00mLock-Free Programming (or, Juggling Razor Blades), Part I
01h 01mThe Future of C++ Parallel and Concurrency Safety Guidelines
