Stencil reads are S,X,X,X. (but expect S,S,S,S or S,0,0,1) #1978
Conversation
kvark
left a comment
kvark
left a comment
There was a problem hiding this comment.
Looks fine, assuming that we can indeed rely on these two possibilities.
spec/index.bs
Outdated
| <td>1 − 4 | ||
| <td>stencil | ||
| <td>{{GPUTextureSampleType/"uint"}} | ||
| <td>`vec4<u32>(S, S, S, S)` or `vec4<u32>(S, 0, 0, 1)` |
There was a problem hiding this comment.
This needs to be S,X,X,X
There was a problem hiding this comment.
I don't think this matches the prose below.
|
Resolution: Make this non-normative because Metal doesn't actually guarantee this. |
More specifically, our consensus is that it's not worth it to try to make this normative. (not just because Metal doesn't guarantee it per se) Making it normative can exert pressure to keep just these two possibilities, so that no one ends up adding a third. (like S,0,0,0?) Probably not a big concern. |
kdashg
changed the title
kainino0x
left a comment
kainino0x
left a comment
There was a problem hiding this comment.
LGTM. I think we also needed to do something like this for binding depth textures as non-depth textures? But we can follow up with it.
I thought the response from Apple was that it's just UB and we aren't supposed to do this at all. (i.e. it's not just the swizzling problem). |
|
Oh yeah, I think you're right. Can't keep all this stuff in my head |
spec/index.bs
Outdated
| <td>1 − 4 | ||
| <td>stencil | ||
| <td>{{GPUTextureSampleType/"uint"}} | ||
| <td>`vec4<u32>(S, S, S, S)` or `vec4<u32>(S, 0, 0, 1)` |
There was a problem hiding this comment.
I don't think this matches the prose below.
But probably S,S,S,S or S,0,0,1.
4 participants
Closes #1198.
Preview | Diff