[Question] was rust-gpu evaluated? about vello HOT 2 CLOSED

I am sorry for opening an issue, it seemed inappropriate to me too, but there was no discussions tab and I am not on zulip (too many different messengers these days, but worry not, once xilem is ready I'll write the chosen one to rule them all! ;))

Thanks a bunch for taking time to answer and in such depth! Feel free to close and again: sorry for opening an issue when it's clearly not!

from vello.

Absolutely. I'd love to have something along the lines of rust-gpu, but it is sadly not mature enough. Another major concern was portability - WGSL is carefully designed to work well even on Vulkan 1.0, while rust-gpu requires more modern GPUs. This is a tradeoff of course, there are also advantages to being able to depend on modern features.

For evaluating shader languages, I strongly recommend implementing prefix sum. That should only be about a hundred lines of code (give or take) but shakes out a lot of the potential problems and also is a good chance to show off the abstraction capabilities of the language. In particular, I'd love to have a Rust trait to express a monoid, and a scan implementation generic on the monoid. For extra points, have that trait include methods for subgroup reduction and scan with a default implementation based on subgroup shuffle.

The currently most promising alternative to WGSL is slang. I've lightly prototyped prefix sum, but run into issues with default trait methods (called "extension" in slang). I should file issues there :)

BTW, there's nothing wrong with using issues to ask questions, but Zulip might be better for such discussions, in this case the #gpu stream.

from vello.