It's possible to attempt to put NaN/Infinity/subnormal numbers in rgba16float canvases - though they may not be written, per "equivalent texel representation", which may flush them - but we haven't written much detail on that: https://github.com/gpuweb/gpuweb/issues/2558#issuecomment-2138335450
It's also of course possible to put arbitrary f16s in the alpha channel.
We need to test both of those things in:
[ ] Reftests (I don't know how they should look, it may be undefined, especially if alpha is NaN/infinity, but hopefully not)
[ ] copyExternalImageToTexture() into rgba8unorm and into rgba16float, probably
It's possible to attempt to put NaN/Infinity/subnormal numbers in rgba16float canvases - though they may not be written, per "equivalent texel representation", which may flush them - but we haven't written much detail on that: https://github.com/gpuweb/gpuweb/issues/2558#issuecomment-2138335450
It's also of course possible to put arbitrary f16s in the alpha channel.
We need to test both of those things in:
copyExternalImageToTexture()intorgba8unormand intorgba16float, probably