search

Rust-GPU / Rust-CUDA

Ecosystem of libraries and tools for writing and executing fast GPU code fully in Rust.
Apache License 2.0
3.13k stars 120 forks source link

Is this project abandoned? #106

Open jackbackes opened 1 year ago

smallstepman commented 1 year ago

https://github.com/Rust-GPU/Rust-CUDA/pull/98#issuecomment-1495202147

ianrgraham commented 1 year ago

I'm happy to see @RDambrosio016 is OK (hopefully just busy with school and life). But could we try to work towards a solution to clear through open issues and PRs on this project? Many people (myself include) have a lot of interest in this project and where it may lead, but the lack of updates has been a bit saddening.

Maybe a better communication channel (zulip, discord, etc.), and eventually an additional maintainer or two, is all that's needed?

RDambrosio016 commented 1 year ago

I wouldn't be opposed to adding another maintainer for clearing through stuff, unfortunately i am just too busy at the moment :(

David-OConnor commented 1 year ago

Hey - that's OK and normal and you shouldn't feel like this project, which you're doing for free, is a burden.

That said, CUDA is awesome, and Rust is Awesome. nVidia hasn't posted anything I've read indicating Rust will get first-class support. So, what do y'all think is the plan? You can, of course, use Vulkan or W/E compute shaders, but my understanding is CUDA's value is it's easier because you #1: Don't have to use a separate shader language and #2: Don't have to [de]serialize the data as byte arrays. This is important, and why CUDA is so popular.

apriori commented 1 year ago

@David-OConnor Forget compute shaders and/or OpenCL kernels for anything scalable and serious. Unfortunately khronous group never understood that it is virtually inacceptable to build a scaling eco system with a language with just "string templating" as a sharing and reuse system. That is one of THE reasons why both utterly fail. You cant just "import" some foundational lighting or what not library, extend beyond that and build foundational and reusable building blocks that make an eco system thrive.

Cuda C++ did exactly that. You just call imported routines - no insane boilerplate involved.

There is no technological reason why there is no alternative. It is mostly funding and conceptual failure of NVIDIAs competitors.

David-OConnor commented 1 year ago

Hopefully we get a good CUDA rust. I'm working on learning Cuda in C++ now because I can't get this lib to compile, but C++ CUDA seems to work; even though I don't really know C++.

Does anyone have any tips on getting this library to compile?

apriori commented 1 year ago

Well, it does compile, however, on a somewhat older rustc. This PR https://github.com/Rust-GPU/Rust-CUDA/pull/98 started some work on updating to a more current rustc, however was never finished by me. Not sure where other update attempts went. See the discussion in the mentioned PR.

ckissane commented 1 year ago

@RDambrosio016 I believe there exists enough community interest. I think people are making issues such as https://github.com/Rust-GPU/Rust-CUDA/issues/114 , because they simply are not sure how to direct it. As @ianrgraham noted a discord server and roadmap creation could do wonders... I understand the college student grind, would you be opposed to someone else taking initiative on that, eg if @ianrgraham started a server would it have your blessing?

David-OConnor commented 1 year ago

Hey, I really want to use this lib. Is there a for-dummies guide to patching this to use the latest RUSTC?