https://gcc.gnu.org/bugzilla/show_bug.cgi?id=128672
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Target Milestone|--- |14.5
Assignee|unassigned at gcc dot gnu.org |rguenth at gcc dot
gnu.org
Status|NEW |ASSIGNED
Priority|P3 |P2
Summary|wrong code for (int)-x / 2 |[14/15/16/17 Regression]
|when (int)x / 2 is also |wrong code for (int)-x / 2
|computed |when (int)x / 2 is also
| |computed
--- Comment #2 from Richard Biener <rguenth at gcc dot gnu.org> ---
gimple_match_op match_op (gimple_match_cond::UNCOND,
NEGATE_EXPR, type, rhs[i]);
ops[i] = vn_nary_build_or_lookup_1 (&match_op, false, true);
so while that's not inserting the possibly undefined negate it still
simplifies as if it were present. We have to avoid that here I think.