[tree-ssa] GGC friendly function save/restore

[Jan Hubicka]

> On Fri, Nov 28, 2003 at 11:18:16PM +0100, Jan Hubicka wrote:
> >     /* Nonzero if the rtl inliner has saved the function for inlining.  */
> >     unsigned int saved_for_inline : 1;
> > + 
> > +   /* Saved tree and arguments during tree optimization.  Used later for
> > +      inlining */
> > +   tree GTY(()) saved_tree;
> > +   tree GTY(()) saved_args;
> 
> Not ok.
> 
> (1) No need for GTY markings on individual fields.
> (2) You didn't put these at the correct place.  Read the comments.
I see.
> 
> > +   if (cfun->saved_tree)
> > +     {
> > +       /* We might need the body of this function so that we can expand
> > +          it inline somewhere else.  */
> > +       DECL_SAVED_TREE (fndecl) = cfun->saved_tree;
> > +       DECL_ARGUMENTS (fndecl) = cfun->saved_args;
> > +     }
> > +   cfun = 0;
> 
> (3) What do you think this is doing?
> cfun == DECL_SAVED_INSNS (current_function_decl), so this is still
> reachable.  You definitely beed to zap saved_tree/saved_args here.
> Which also means that the zapping of saved_tree earlier should have
> been useless.

DECL_SAVED_INSNS is zapped at the end of rest_of_compilation.  Perhaps I
should move that bit to line just bellow cfun = 0 to make this more
readable.  Or am I missing your point?
> 
> (4) I also don't see that you've replaced the ggc_collect call that
> you removed.  The intent is to get rid of all the rtl that is now
> most definitely garbage.
> 
This got lost during patch separation, sorry.  I have many ggc_collect
calls in tree-optimize now and I killed them all.  I will re-add one
just after cfun = 0.

Honza
> 
> r~