Rust CUDA Project

Summary, from someone who uses CUDA on rust in several projects (Computational chemistry and cosmology simulations):<p><pre><code> - This lib has been in an unusable and unmaintained state for years. I.e., to get it working, you need to use specific, several-years-old variants of both rustc, and CUDA. - It was recently rebooted. I haven&#x27;t tried the Github branch, but there isn&#x27;t a release yet. Has anyone verified if this is working on current Rustc and CUDA yet? - The Cudarc library (https:&#x2F;&#x2F;github.com&#x2F;coreylowman&#x2F;cudarc) is actively maintained, and works well. It does not, however, let you share host and device data structures; you will [de]serialize as a byte stream, using functions the lib provides. Works on any (within past few years at least) CUDA version and GPU. </code></pre> I highlight this as a trend I see in software libs, in Rust more than others: The projects that are promoted the most are often not the most practical or well-managed ones. It&#x27;s not clear from the description, but maybe rust-CUDA intends to allow shared data structures between host and device? That would be nice.

I’ve been using the cudarc crate professionally for a while to write and call cuda from rust. Can highly recommend. You don’t have to use super old rustc versions. Although I haven’t looked exactly what you do need to use recently.

Very cool to see this project get rebooted. I&#x27;m hoping it will have the critical mass needed to actually take off. Writing CUDA kernels in C++ is a pain.<p>In theory, since the NVVM IR is based on LLVM IR, rust in CUDA should be quite doable. In practice, though, of course it is an extreme amount of work.

Shouldn&#x27;t it be called RUDA?

For this to take off it has to be supported by Nvidia, at feature parity with C++. Otherwise it&#x27;ll remain a niche tool, unfortunately. I do hope Nvidia either supports this or rolls out their own alternative. I&#x27;ve been looking to get off the C++ train for years, but stuff like this keeps me there.

Looks like a dead end. Why CUDA? There should be some way to use Rust for GPU programming in general fashion, without being tied to Nvidia.