Missing VertexFormat::Unorm8x3 and VertexFormat::Unorm16x3

[Pauan]

Currently there are x2 and x4 variants of Unorm, but not x3. Is this intentional or is it just a missing feature?

How am i supposed to send RGB data to the GPU (which is [u8; 3])? Do I have to manually pad every array to make it into a [u8; 4]?

The glTF parser crate supports both RGB and RGBA in various formats, and ideally I would like to support all of them.

[kvark]

These vertex formats are not portable, hence missing from WebGPU upstream and wgpu.
What is the use case for providing lots of RGB data to vertex attributes?

[Pauan]

Well, that's what vertex colors are. Though I'm not sure how often glTF models use RGB vertex colors rather than RGBA vertex colors, so perhaps it's a non-issue in practice.

[cwfitzgerald]

In general I'd recommend normalizing things like vertex formats to prevent combinatoric explosions of the pipeline states needed to deal with gltf.

[Pauan]

@cwfitzgerald Yeah, that makes sense, though WGSL already does a lot of normalization.

For example it will auto-convert a vec3 into a vec4 (adding 1 to the final element). And the point of Unorm is to normalize an integer into an f32, so the only thing I'm missing is Unorm8x3 and Unorm16x3.

But since they're never going to be supported, I guess I have to do manual normalization before passing to the buffer. Unfortunately that does have a pretty big performance cost, since I can't just pass the bytes directly into the buffer, I'm forced to allocate a new Vec<u8>.