What limits the available pixel format for HDR

[hmaarrfk]

In https://github.com/pygfx/rendercanvas/pull/75/files I see that

• rgba-u8
• rgba-u16
• rgba-f16
• rgba-f32

I'm wondering if it would also be possible to add support for

• rgba-u32
• rgba-i8
• rgba-i16
• rgba-i32

We use i16 sometimes because u16 doesn't seem to be supported by pytorch in many operations.

Exciting that this has come through!

[almarklein]

Would be nice to improve support on this front, indeed. For now I kept it simple. Here are some thoughts ...

A fragment shader normally produces an rgba color with values 0..1.

• The rgba-u8 format maps to rgba8unorm, where norm means normalized to 0..1.
• I added rgba-f16 and tested it with an example. Since this is a float, it accepts pixel values 0..1 too.
• I also added rgba-f32. I'm pretty sure one needs to enable a wgpu feature to be able to use this. In any case I have not tried it yet. But if it does work, it's float, so 0..1.
• I also added rgba-u16, but I'm not sure how that works out to be honest. Should the fragment shader now produce values 0..65535?
• Same for the other integer formats.

[hmaarrfk]

Right. One of the challenges with the more extended formats is that often they don’t have normalized forms.