“High-Radix Concurrent C++”

on YouTube ↗56m

" summary

Demonstrates a C++ radix-26 trie using atomic counts and atomic pointers with a test‑and‑set insertion that yields a starvation-free algorithm, ports to CUDA C++ using a freestanding simt::std::atomic layer and cudaMallocManaged, and launches 100000 concurrent threads on Turing/Volta hardware. Kernel building time is about 650 µs (PCIe transfers dominate) and the GPU run gives roughly 5–10x speedups versus a 40-thread Xeon; the live demo completes successfully.

AI-written summary · about

" chapters (8)

1. 0:00Introduction and trie design
2. 11:32Trie demo execution
3. 13:02Building a concurrent trie
4. 21:06CUDA crash course
5. 33:41Porting to CUDA C++
6. 37:54GPU performance analysis
7. 40:51C++20 concurrency features
8. 44:12Q&A session

" related talks

• 47mConcurrent Programming in C++Venkat Subramaniam · NDC TechTown 2017
• 01h 04mAtomics, Locks, and Tasks (part 2 of 2)Rainer Grimm · CppCon 2019
• 01h 08mBetter Code: ConcurrencySean Parent · NDC London 2017
• 17mArctic: A Practical Lock-Free Adaptive Radix TreeMichal Friedman, Newton Ni, Emmett Witchel, Nicolas Garza, Jenny Stinehour, Michael Goppert · 20th USENIX Symposium on Operating Systems Design and Implementation (OSDI '26)
• 01h 00mConcurrency in C++20: A Deep DiveRainer Grimm · Meeting C++ 2023
• 01h 00mConcurrency Improvements in C++20: A Deep DiveRainer Grimm · NDC TechTown 2023
• 59mAn Introduction to Multithreading in C++20Anthony Williams · C++ on Sea 2022
• 55m“A lock-free concurrency toolkit for deferred reclamation and optimistic speculation"Paul E. McKenney, Michael Wong, Maged Michael · CppCon 2016