[Bug ipa/59831] [4.9 Regression] ice in cgraph_speculative_call_info with -O3

[hubicka at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59831

--- Comment #3 from Jan Hubicka <hubicka at gcc dot gnu.org> ---
Easy fix is the following:
Index: cgraph.c
===================================================================
--- cgraph.c    (revision 206617)
+++ cgraph.c    (working copy)
@@ -815,6 +815,7 @@ cgraph_set_call_stmt (struct cgraph_edge

   e->call_stmt = new_stmt;
   if (e->indirect_unknown_callee
+      && !e->speculative
       && (decl = gimple_call_fndecl (new_stmt)))
     {
       /* Constant propagation (and possibly also inlining?) can turn an

I am however not entirely happy about it.  The devirtualization happens while
saving function body for inlining, so it is obviously a missed optimization
here: I assume that we never fold the call pre-IPA.

Also we should resolve speculation here, for that we however need to copy the
associated references...