https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96591
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Component|lto |c++
Target Milestone|--- |8.5
Keywords| |lto
--- Comment #2 from Richard Biener <rguenth at gcc dot gnu.org> ---
FE trees should not leak into the middle-end so it looks more like a FE issue
to me. We're chasing TYPE_FIELDS of the context of the scalar_type name
and visiting a TEMPLATE_DECL for 'rebind' in simd_traits.
In the end this might mean free-lang-data doesn't visit DECL_CONTEXT of
the INTEGER_TYPEs TYPE_NAME? At least I do not see it walking
simd_traits.
It might also be that some intermediate optimization somehow introduces
a reference to this type from nowhere (huh).