Support normalized 16-bit texture formats

[kvark]

Is your feature request related to a problem? Please describe.
It's often convenient to process the data in shaders as u16 and store it this way. However, reading it in the shader is problematic: currently we have no 16-bit normalized formats, so it has to be bound as integer. But integer textures can't be sampled. So using them is highly restricted today without the normalization.

Describe the solution you'd like
Exposed a set of 16-bit normalized formats gated by a feature.

Describe alternatives you've considered

Additional context
Requested on matrix.

[aloucks]

I was looking into the details of this to see if I could implement it when I get some free time and noticed that the existing texture format features all have a TEXTURE_COMPRESSION_ prefix, rather than TEXTURE_FORMAT_, and the 16bit formats are not related to compression.

Would you be open to adjusting the naming of the existing texture format features? It would be nice if they all followed the same pattern, such as TEXTURE_FORMAT_{family}.

Currently the compressed formats have features like TEXTURE_COMPRESSION_ETC2 and TEXTURE_COMPRESSION_ASTC_LDR.

I'm suggesting changing these to allow for other (non-compression formats) to all follow the same convention.

e.g.

// Enables ASTC family of compressed textures ...
TEXTURE_FORMAT_ETC2

// Enables ASTC family of compressed textures ...
TEXTURE_FORMAT_ASTC_LDR

// Enable additional 16-bit texture formats ...
TEXTURE_FORMAT_16BIT

[kvark]

We should name the features the same way as WebGPU spec does it, unless there is a really strong reason to diverge.

[VincentJousse]

@kvark Will this become a WebGPU feature any time soon ?

[kvark]

I'm not participating in this standard body any more.
Redirecting to @jimblandy