https://gcc.gnu.org/bugzilla/show_bug.cgi?id=120747
--- Comment #10 from Filip Kastl <pheeck at gcc dot gnu.org> --- > Given the nature of the change that caused this (trimming integral ranges > bounds to match the bitmasks) its probable that a smaller range had some > other pass make a different decision. Yeah, I also think that if the numerical error is a problem in GCC then it's not in your patch but rather in some other optimization decision that the patch influenced indirectly.
