Mesa's New OpenCL Stack "Rusticl" Nearing Formal Support For OpenCL 3.0

[karolherbst]

Meme around here is for any project to be suggested to be rewritten in rust, even if the project is already written in rust.

ahh, didn't know about that

[MadeUpName]

Read the article...

They have been crowing about Mesa 22 having 3.0 support since it was in beta. I have it installed and it is only 1.1.

https://www.khronos.org/opencl/

I read Micheals article a couple of months ago about image support being added and that seems to have gone no where.

Just before MESA announced they were getting ready to release 22 some one that works for AMD desperately jumped in and said they were going to start packaging ROCm for Fedora and Debian. That got half done. I have spent weeks and weeks desperately chasing these BS claims. I don't need to read another article about grandiose claims that are never coming true. The only thing that comes from these claims is they string you along until it is supposedly 90% of the way there, then they announce they need to abandon it, reinvent the whole thing from scratch and then put out another press release that working openCL with image support is just around the corner.

If any one ever gets an actual working implementations let us know. But I am done pissing my time away on snake oil claims.

[smitty3268]

If any one ever gets an actual working implementations let us know. But I am done pissing my time away on snake oil claims.

I'm not sure what you count as working, but they've announced a working implementation that passes all tests (on Intel hardware) right here in this article.

I guess time will tell whether it actually works in apps or not.

[Quackdoc]

I'm not sure what you count as working, but they've announced a working implementation that passes all tests (on Intel hardware) right here in this article.

I guess time will tell whether it actually works in apps or not.

time, or compiling it. you can compile the PR if you wish. get the alpha experience

[shmerl]

Very interesting. How exactly is it using llvm (besides rust compiler itself relying on it)?

[karolherbst]

Very interesting. How exactly is it using llvm (besides rust compiler itself relying on it)?

We use clang to compile OpenCL C to LLVM IR, and then use the SPIRV-LLVM-translator to translate that into SPIR-V and from there it's all common/shared code (between OpenGL SPIR-V, Vulkan, and OpenCL kernel SPIR-V)

[karolherbst]

I'm not sure what you count as working, but they've announced a working implementation that passes all tests (on Intel hardware) right here in this article.

I guess time will tell whether it actually works in apps or not.

yeah, that's actually the biggest concern at the moment. I now it can run the luxmark v3.1 luxball demo, but others take forever to compile because we inline everything. I couldn't get Blender to work, because it doesn't enable OpenCL on all plattforms and I wasn't in the mood of compiling Blender myself.

Geekbench seems to run, but can't tell if it actually produces a valid output or not.

So if others would test it and point out where bugs are, that would be very helpful.

[MadeUpName]

The first thing you should check is what does clinfo report. If it won't report what the app needs there isn't even a need to test with an app.

[Quackdoc]

yeah, that's actually the biggest concern at the moment. I now it can run the luxmark v3.1 luxball demo, but others take forever to compile because we inline everything. I couldn't get Blender to work, because it doesn't enable OpenCL on all plattforms and I wasn't in the mood of compiling Blender myself.

Geekbench seems to run, but can't tell if it actually produces a valid output or not.

So if others would test it and point out where bugs are, that would be very helpful.

where do we go to point out bugs anyway? mesa's issue tracker, on the PR, or on your fork's issue tracker? or somewhere else?

[karolherbst]

where do we go to point out bugs anyway? mesa's issue tracker, on the PR, or on your fork's issue tracker? or somewhere else?

on the MR would be good