https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70119
--- Comment #3 from ktkachov at gcc dot gnu.org --- Right, I see we have the same issue on aarch64 as on x86. So what would be the accepted solution here? I've been playing with a patch to simplify-rtx to move the subreg inside the AND and transform the AND into a QImode operation, but there were concerns on it pessimising other code