https://gcc.gnu.org/bugzilla/show_bug.cgi?id=122537
--- Comment #2 from Richard Biener <rguenth at gcc dot gnu.org> ---
number_of_iterations_lt_to_ne turns this into {0, +, 61} != 4 in signed int
and we set ->no_overflow to be true on this IV. Under the assumption
that a.1_16 != -2147483629 && a.1_16 + -20 >= -2147483592 (a.1_16 is zero).
Somehow
/* IV0 < IV1 does not loop if IV0->base >= IV1->base. */
if (mpz_cmp (mmod, bnds->below) < 0)
noloop = boolean_false_node;
else
noloop = fold_build2 (GE_EXPR, boolean_type_node,
iv0->base, iv1->base);
hits the first condition even though clearly
a.1_16 + -20 >= a.1_16 + -72
for a.1_16 == 0 at least. So sth in the computation earlier or the
input to number_of_iterations_lt_to_ne goes off. It isn't very well
documented, so will have to reverse engineer that (again).
