Expose shader validation

[daxpedda]

I was trying to pre-compile shaders while still running the exact same validation that wgpu does.
Would it be acceptable to expose this module for that purpose?

[Wumpf]

I think the better way to go about this is to call https://docs.rs/naga/0.14.1/naga/valid/struct.Validator.html#method.validate
which is most of what wgpu does. Is there a reason this would miss something in your case?

let module =
    naga::front::wgsl::parse_str(&shader.source).expect("should parse :O");

let mut validator = naga::valid::Validator::new(
    naga::valid::ValidationFlags::all(),
    naga::valid::Capabilities::all(),
);

validator.validate(&module).expect("should be valid!");

The nice thing about this is that you can then at least skip the parsing on native wgpu since you're allowed to pass in a naga module for creating the shader module https://github.com/gfx-rs/wgpu/blob/trunk/wgpu/src/lib.rs#L576

[daxpedda]

What you are describing is a different use-case, one which I've used pretty successfully so far!

I'm talking specifically about passing raw SPIRV modules, which I would generate externally first. Wgpu currently does a bunch of validation, including using the Naga Validator, but that's not all it does. E.g. Interface::check_stage() does a lot of checking against the render pipeline, which is completely skipped when providing a raw SPIRV shader.

The goal is to replicate exactly what Wgpu does internally, most of this stuff is exposed and easy to re-use. But the whole validation going on in wgpu_core::validation will require me to almost copy the whole file. Which, alternatively, I can do as well.

I was told before that large parts of the wgpu-core and wgpu-hal API are not exactly considered stable, so I thought exposing this isn't an issue.

Cc #3103, which would make this pointless as well.

[Wumpf]

ah I see, thanks for elaborating, makes sense! I'd be in favor of exposing then as well, but I'd like to hear what @cwfitzgerald says

[cwfitzgerald]

I don't love it, but can't argue against it.