I'm on TC39. We've had a request from the ColorWebCG to add a Float16Array to represent float-backed canvases.

In pursuing that proposal within TC39, it would be helpful to know if it would be useful for people integrating with WebGPU as well. I saw that float16 is supported in WebGPU, and I saw an example of wrapping a buffer in a Float32Array. Can I conclude that therefore some buffers are created holding float16 values, and that it would be helpful in some cases to wrap those buffers in a Float16Array? I am guessing so, but I haven't gone through the WebGPU spec before, and it's a bit too large for me to confidently conclude either way.

I've tentatively added that as a use case to the proposal for now, but am happy to remove it if I'm mistaken.

(To be clear, this would be specifically IEEE754 "binary16" floats, not bfloat16.)