[Bug tree-optimization/127130] [14/15/16/17 Regression] wrong code at -O{2,3} on x86_64-linux-gnu

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=127130

--- Comment #6 from Richard Biener <rguenth at gcc dot gnu.org> ---
I'll note the analysis of _3 is correct to be { 93, +, -97 }_1 given we know
the loop iterates only once, thus the values are 93 and -4 without overflow,
but the peeled result overflows, invoking UB which makes
infer_loop_bounds_from_signedness record a wrong niter bound.