WIP add support for parsing Op::AtomicLoad in spv frontend

[schell]

@cwfitzgerald here is my first stab at making progress on gfx-rs/wgpu#4489. Am I on the right track? Are there more steps besides parsing?

[schell]

I see here I provided a body that might match OpAtomicLoad, not OpAtomicIAdd...

[Patryk27]

fwiw, some time ago I tried to implement support for spir-v atomics as well, but I gave up up due to the inconsistency between the atomic models here:

If I remember correctly, spir-v has untyped atomics (i.e. you can atomic-op almost any piece of memory), while wgsl (and thus Naga) requires typed atomics -- and so supporting them on the spir-v frontend would probably require an extra type promotion late-pass where spir-v int & float types detected to be used in atomic context would be "promoted" into their Naga atomic equivalents across the entire program graph.

(since wgsl and/or Metal forbids castings between atomic and non-atomic types.)

Note that I haven't done much development on Naga, so this is just a couple of loose thoughts - maybe my initial approach was totally wrong and that's why it failed 😄

One way or another, I'd love to see spir-v atomics working - rust-gpu supports them (more or less) and I could use some in my shader 😄

[schell]

@Patryk27 Yes, that's exactly the situation as it stands. It's a really big bite to chew because of having to do that type-promotion. Here's the start of a conversation I had with the gang on the matrix channel (it's pretty long) https://app.element.io/#/room/#naga:matrix.org/$_04dSsaJ6O8BFPxfV61v6Wb9I8xYBy9Zrtm4nvF3PGs

[pema99]

You still attempting this or was it just a quick experiment?

[schell]

Just a quick experiment - I might pick it up again but if anyone else is going to give it a shot I'll defer.

[pema99]

Alright. I'd love to see this come to fruition, but I read a bit up on what is preventing it (the whole untyped atomics vs Atomic thing), and the need for some kind of inference/promotion pass from regular to atomic types seems a bit... daunting for a first time contributor.

[pema99]

In case anyone reads this and is wondering what I am referring to: gpuweb/gpuweb#2377

[cwfitzgerald]

Hello, thank you for your PR against Naga!

As part of gfx-rs/wgpu#4231, we have moved development of Naga into the wgpu repository in the Naga subfolder. We have transferred all issues, but we are unable to automatically transfer PRs.

As such, please recreate your PR against the wgpu repository. We apologize for the inconvenience this causes, but will make contributing to both projects more streamlined going forward.

We are leaving PRs open, but once they are transferred, please close the original Naga PR.

[schell]

I will retarget my PR at wgpu, thanks :)