https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65655
--- Comment #5 from Jan Hubicka <hubicka at gcc dot gnu.org> ---
Hmm it is interesting case, we try to resolve speculation while we are
duplicating it. I am testing the following
Index: ipa-inline-analysis.c
===================================================================
--- ipa-inline-analysis.c (revision 221845)
+++ ipa-inline-analysis.c (working copy)
@@ -793,7 +793,11 @@ edge_set_predicate (struct cgraph_edge *
{
/* If the edge is determined to be never executed, redirect it
to BUILTIN_UNREACHABLE to save inliner from inlining into it. */
- if (predicate && false_predicate_p (predicate))
+ if (predicate && false_predicate_p (predicate)
+ /* When handling speculative edges, we need to do the redirection
+ just once. Do it always on the direct edge, so we do not
+ attempt to resolve speculation while duplicating the edge. */
+ && (!e->speculative || e->callee))
e = redirect_to_unreachable (e);
struct inline_edge_summary *es = inline_edge_summary (e);
