jenatali
commented
1 year ago Discussed offline - there's a problem I can see with this where we'll still clear the "need to reapply" bit at the end, meaning that a future draw/dispatch could end up not having a bound table. We're going to try a different approach, where the dirty-ness of a descriptor table will be tracked up to the size of the root signature bucket for that type, rather than the shader decls directly.
The Apply*Helper()s didn't verify that the shader they were trying to bind resources to actually used that type of resource, so there were cases where we would try to bind invalid data to shaders that didn't reference it. This didn't cause issues in applications, but does log a breaking error in the debug layer.