The One-Decade Task: Putting std::atomic in CUDA. - Olivier Giroux - CppCon 2019

Channel:CppCon · Length:1:05:38 · Published:2019-09-28 · Views:10899

Watch on YouTube

Summary

Olivier Giroux's CppCon 2019 talk recounts the roughly ten-year effort to bring std::atomic and the C++ memory model to CUDA GPUs, through ten stories from that work.

Why HN recommends it

Cited as a reference when discussing NVIDIA's C++ memory model, libcu++, and GPU C++ resources.

C, C++, Rust, Go & systems languages · Computer architecture & CPUs

Subtitles (excerpt)

good afternoon welcome this talk is the story of the incredible heavy lift that it took to put standard atomic into CUDA it didn't take a decade because we stopped to smell the roses it took a decade because there was an incredibly long dependent chain of things that had to go just right in order for it to happen my name is una voz Lu I'm a GPU architect have been for quite some time and I am also your chair of concurrency and parallelism and in this talk I have the opportunity to really join together my my two jobs I get to talk the whole time about both at the same time instead of having a portion of my talk that's about GPUs and a portion of my talk that's about concurrency and parallelism in C++ the whole talk is only about this all right so I'm a member of a couple of communities first and foremost I'm an Nvidia engineer I've been an engineer at Nvidia for 17 years I've been in the architecture team the entire time I have been working on the SM for the last 10 years so that means you know Kepler Maxwell Pascal Volta Turing next next next next next next and so on I've been a C++ user since 1997 or so I'm a member of dub G 21 I'm the chair of concurrency and parallelism and I'm also a member of this memory model community which is a

On Hacker News

> I really wish that everybody would finally admit that the "fixed function pipeline" is dead There is a great deal of space between 'fixed function pipeline' (which no one is using) and 'all compute all the time' (which pretty much no one is using for real-time graphics either). > GPUs don't yet have a well-defined memory model Nvidia has a memory model. (I have heard--from an nvidia employee--t… — moonchild, 2022-10
Sure, "CppCon 2016: “Bringing Clang and C++ to GPUs: An Open-Source, CUDA-Compatible GPU C++ Compiler" [LINK] "CppCon 2019: Olivier Giroux “The One-Decade Task: Putting std::atomic in CUDA.”" [VIDEO] "Octane render. The Future of GPU Rendering: Real-Time Raytracing" (with CUDA) [LINK] NVIDIA HPC SDK [LINK] — pjmlp, 2021-08
CppCon 2019: Olivier Giroux “The One-Decade Task: Putting std::atomic in CUDA.” [VIDEO] Future of ISO and CUDA C++ [LINK] Thrust, the CUDA C++ template library [LINK] Optix, single-ray shader programming model using C++, and ray Tracing acceleration using RT Cores. [LINK] Phew, so many wrappers yet to list. — pjmlp, 2021-04
Here's a somewhat related talk from CppCon '19: "The One-Decade Task: Putting std::atomic in CUDA" [VIDEO] — davvid, 2020-09
Right now, NVidia is designing their GPGPUs to run C++ code. [LINK] [VIDEO] With Intel just releasing oneAPI (aka Data Parallel C++) last month, and Khronos pushing SYCL as alternative to CUDA C++. It took 30 years for C++ to reach this stage, and in some domains (like embedded) it is still fighting for relevance. That is something to keep in mind whenever to advocate Rust as replacement for Doma… — pjmlp, 2019-12

Related videos