https://gcc.gnu.org/bugzilla/show_bug.cgi?id=120360
--- Comment #5 from Linus Torvalds <torva...@linux-foundation.org> ---
(In reply to Jakub Jelinek from comment #4)
>
> Untested fix for #c2/#c3. We already had *cmp<mode>_minus_1 pattern which
> handles the case where result of subtraction is used solely in comparison
> against zero, but that doesn't match when there is a subtraction of a
> constant because we canonicalize that to addition of the negated constant.
While I don't really read the gcc instruction definitions, it looks sane to me.
Except for those odd 0xfa1e0ff3UL / 0xfb1e0ff3UL constants.
I can only assume they are some test-case magic?
Linus
