This is the mail archive of the gcc@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: Insertng a call function


Could it be that it being done lower in the pass unlike insertions in
tree-profile, gomp - some ssa defintion information is missing. The
reason I have been led to think is because it fails
register_new_defintions.  I would appreciate any suggestions.

thanks 

On 7/28/05, drizzle drizzle <drizzle76@gmail.com> wrote:
> Hi
>       I updated my function call creating statement based on how
> gomp_parallel_end is inserted in the gomp branch. It still breaks. I
> have provided the function declaration in the file, just want to
> insert the call. Any further suggestions. My whole objective is  to
> mark a few particular
> tree statements (array references actually) and carry them over to RTL
> form. So I am trying to insert a call before these references which
> when seen in RTL will tell  me my references of interest.  Any
> suggestions on some other ways I an try apart from this call
> insertion?
> 
> 
> here is my upated piece of code to insert a function call
>                 tree fn, type;
>                  type = build_function_type_list (void_type_node, void_type_node, NULL_TREE);
>                  tree id = get_identifier ("foo");
>                  tree fn_decl = build_decl (FUNCTION_DECL, id, type);
>                  DECL_EXTERNAL (fn_decl) = 1;
>                  TREE_PUBLIC (fn_decl) = 1;
>                  DECL_ARTIFICIAL (fn_decl) = 1;
>                  TREE_NOTHROW (fn_decl) = 1;
> 
>                  tree call=build_function_call_expr (fn_decl, NULL_TREE);
> 
> 
> thanks
> 
> 
> On 7/28/05, Diego Novillo <dnovillo@redhat.com> wrote:
> > On Thu, Jul 28, 2005 at 01:15:22PM -0400, drizzle drizzle wrote:
> >
> > >       I am inserting a call stmt  in linear_transform_loops. Al though
> > > the call statement gets inserted  , the compilation breaks. Can some
> > > one help me identify if I am missing some information  in the call
> > > node that I create  (thanks Daniel  for all the help until now)
> > >
> > Check gomp-20050608-branch.  gimple-low.c:create_gomp_fn takes a
> > block of code, puts it inside a new function F and replaces the
> > block of code with a call to F.
> >
>


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]