[Bug tree-optimization/126538] Compile time hog with gimple-range-phi and bitint

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=126538

--- Comment #4 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
precision is in bits, so it should be likely also something in bits.
Could be e.g. MAX_FIXED_MODE_SIZE, that is maximum precision one can get for
integral types when not using _BitInt (which indeed can have precision up to
65535 bits).
Note, it is certainly useful to have value ranges even for _BitInt(65535)
SSA_NAMEs, we just need to avoid spending way too much time on it like in this
testcase.