[Bug tree-optimization/89789] [9 Regression] Compile time hog during RPO VN

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89789

--- Comment #3 from Richard Biener <rguenth at gcc dot gnu.org> ---
OK, so we oscillate for

# vh_5 = PHI <vh_12(3), vh_8(D)(2)>
vh_12 = vh_8(D) ^ vh_5;

where vh_8(D) is undefined.  First we get zero (vh_8 ^ vh_8) then we get,
with vh_5 == 0 (UNDEF meet 0) we get vh_8(D) and so on.

The issue is the vh_5 lattice changes UNDEF -> constant -> UNDEF where
we shouldn't go back to UNDEF but to VARYING.