https://gcc.gnu.org/bugzilla/show_bug.cgi?id=93738
--- Comment #25 from Jakub Jelinek <jakub at gcc dot gnu.org> --- Including that other subreg_lowpart_p call introduced in this PR, I think that is just fine if it is paradoxical subreg, it will just create a different paradoxical subreg or a normal subreg depending on the modes involved.
