You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Had a half baked thought worth considering here - what if we continue to use the shader flags analysis in DXContainerGlobals, but we teach the analysis to trust the metadata? That is to say, the analysis can just fill the flags structure if the metadata is present, and since the metadata is always present post-lowering, then it can run at any time. This has the interesting effect that the flags pass can be run on DXIL and get the flags from metadata if we at some point want to reuse the pass in the DXIL->modern LLVM path.
DXContainerGlobals currently grabs shader flags from the analysis. This seems reasonable at a high level, but there are a couple of caveats:
We should update the pass to simply pull the flags in an opaque way from metadata and put them into the appropriate container part.
The text was updated successfully, but these errors were encountered: