> > >
> > > The problem with instrumented call is that instrumented function
> > > returns two values and call lhs gets only the first one. Thus we
> > > generate bndret call to get the second one to build own return with
> > > two values.
> >
> > I see, patch is OK then (preferably merging as much as possible with
> > ipa-split)
> >
> > Honza
>
> Here is a refactored version with common code moved to tree-chkp.c.
> Bootstrapped and tested on x86_64-unknown-linux-gnu. Does it look OK?
>
> Thanks,
> Ilya
> --
> gcc/
>
> 2015-04-07 Ilya Enkovich <ilya.enkov...@intel.com>
>
> * tree-chkp.h (chkp_insert_retbnd_call): New.
> * tree-chkp.c (chkp_insert_retbnd_call): New.
> * ipa-split.c (insert_bndret_call_after): Remove.
> (split_function): Use chkp_insert_retbnd_call.
> * cgraphunit.c (cgraph_node::expand_thunk): Build returned
> bounds for instrumented functions.
>
> gcc/testsuite/
>
> 2015-04-07 Ilya Enkovich <ilya.enkov...@intel.com>
>
> * gcc/testsuite/gcc.target/i386/thunk-retbnd.c: New.
OK, thanks!
> @@ -1697,6 +1698,17 @@ cgraph_node::expand_thunk (bool output_asm_thunks,
> bool force_gimple_thunk)
> gsi_insert_after (&bsi, call, GSI_NEW_STMT);
> if (!alias_is_noreturn)
> {
> + if (instrumentation_clone
> + && !DECL_BY_REFERENCE (resdecl)
> + && restmp
> + && BOUNDED_P (restmp))
> + {
> + resbnd = chkp_insert_retbnd_call (NULL, restmp, &bsi);
> + create_edge (get_create (gimple_call_fndecl (gsi_stmt (bsi))),
> + as_a <gcall *> (gsi_stmt (bsi)),
> + callees->count, callees->frequency);
> + }
Is there any reasons the rtbnd builtin call is not gimple_call_internal_p?
That way we would not need to worry about representing this in callgraph.
Honza
