[Bug tree-optimization/79034] [6 Regression] error: missing PHI def in verify_gimple_in_cfg

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=79034

--- Comment #3 from Richard Biener <rguenth at gcc dot gnu.org> ---
The issue is the degenerate PHI that doesn't appear on trunk (we're lucky). 
Thus

      /* We don't want to handle PHIs.  */
      if (EDGE_COUNT (join_tgt_in_edge_from_call->dest->preds) > 1)
        join_tgt_bb = split_edge (join_tgt_in_edge_from_call);

doesn't catch that.  There's the slight complication that virtual PHIs are ok
(we rewrite those).  So the easiest thing is to propagate them out.

Index: gcc/tree-call-cdce.c
===================================================================
--- gcc/tree-call-cdce.c        (revision 244260)
+++ gcc/tree-call-cdce.c        (working copy)
@@ -805,7 +805,18 @@ shrink_wrap_one_built_in_call_with_conds
       if (EDGE_COUNT (join_tgt_in_edge_from_call->dest->preds) > 1)
        join_tgt_bb = split_edge (join_tgt_in_edge_from_call);
       else
-       join_tgt_bb = join_tgt_in_edge_from_call->dest;
+       {
+         join_tgt_bb = join_tgt_in_edge_from_call->dest;
+         /* We may have degenerate PHIs in the destination.  Propagate
+            those out.  */
+         for (gphi_iterator i = gsi_start_phis (join_tgt_bb); !gsi_end_p (i);)
+           {
+             gphi *phi = i.phi ();
+             replace_uses_by (gimple_phi_result (phi),
+                              gimple_phi_arg_def (phi, 0));
+             remove_phi_node (&i, true);
+           }
+       }
     }
   else
     {