AMD returns the full 32-bit FP value on native, and NV gives you 24-bit. I guess our equal check could do (v & 0xffffff) == depth_24 to handle the case where the top bits are undefined. They're not undefined on NV apparently, but not 100% sure what spec says here. Copying D24 is a pretty messy topic.
Nothing extracted yet.
Test separately copies depth and stencil aspects of DS image. At the moment on Turnip it fails with:
When copying depth from D24S8 the test doesn't like that in 8 MSBs there is a stencil value. However VK spec states:
https://github.com/HansKristian-Work/vkd3d-proton/blob/5657f799749d60bbc3d536a80b1514ddc8b24c9b/tests/d3d12_copy.c#L523-L533
This comment only confuses me...
On the other hand I don't know what d3d12 expects in this case.