https://gcc.gnu.org/bugzilla/show_bug.cgi?id=120560
--- Comment #9 from Richard Biener <rguenth at gcc dot gnu.org> ---
(In reply to Andrew Macleod from comment #8)
> Created attachment 62480 [details]
> proposed fix
>
> OK, after more poking, this is the real fix.
>
> The original code during the final comparison used
> /* Check if init + nit * step overflows. Though we checked
> scev {init, step}_loop doesn't wrap, it is not enough
> because the loop may exit immediately. Overflow could
> happen in the plus expression in this case. */
> if ((dir == EV_DIR_DECREASES && compare_values (maxvr.min (), initvr.min
> ()) != -1)
> || (dir == EV_DIR_GROWS && compare_values (maxvr.max (), initvr.max
> ()) != 1))
> return false;
>
> The new code did the same check in a slightly different way. IN this case,
> compare_values must return -1 or 1, which means A < B or A >B.
> if A == B, it returns false.
>
> The new code used wi::gt_p to look for the failure case, which means if the
> 2 values are the same, it will not fail, and we will produce a range.
>
> The fix is to change it to use wi::ge_p which will again return false when
> the 2 values are the same.
That indeed seems to be an unintended difference in how the code behaves.
> Running through regression testing now.
