[Bug tree-optimization/110269] [13/14 Regression] Missed Dead Code Elimination when using __builtin_unreachable since r13-4607-g2dc5d6b1e7e

[rguenth at gcc dot gnu.org via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110269

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Assignee|unassigned at gcc dot gnu.org      |rguenth at gcc dot 
gnu.org
             Status|NEW                         |ASSIGNED

--- Comment #3 from Richard Biener <rguenth at gcc dot gnu.org> ---
It does handle GIMPLE_COND.  What the change did was avoid using GENERIC
fold_const and instead resort to match.pd only.  I can't see how it
arrived at folding it before.  We (now) have

Visiting PHI node: _15 = PHI <a.3_11(2), 6(5)>
        Argument #0 (2 -> 7 executable)
        a.3_11  Value: CONSTANT a.3_11
        Argument #1 (5 -> 7 executable)
        6       Value: CONSTANT 6

    PHI node value: VARYING

Lattice value changed to VARYING.  Adding SSA edges to worklist.

so there's no nonzero bits info.  The input to CCP2 is almost the
same:

 int main ()
 {
-  int D.2014;
+  int D.2786;
   int _1;
   int * * d.0_2;
   int * b.1_3;
@@ -46,7 +46,7 @@
   __assert_fail ();

   <bb 7> [local count: 1073458354]:
-  # RANGE ~[0, 0]
+  # RANGE [irange] int [-INF, -1][1, +INF]
   # _15 = PHI <a.3_11(2), 6(5)>
   _20 = _15;
   _1 = _20;

so it looks like range representation changes make a difference here?

Ah, so match.pd lacks a simple (ne @0 integer_zerop) check using
tree_single_nonzero_warnv_p.  We have

 (simplify
  (cmp (convert? addr@0) integer_zerop)
  (if (tree_single_nonzero_warnv_p (@0, NULL))
   { constant_boolean_node (cmp == NE_EXPR, type); }))