Remove unit-at-a-time checks and flag_inline_trees

[Jan Hubicka]

> > I've set it to -O3 behaviour in the patch as it was most resonable
> > intermediate step (it makes it conservatively correct as it was allways
> > supposed to be) and I plan to drop it as followup.  This patch has
> > snowballed enough changes already.
> 
> Why are you treating Ada and C++ differently here?  Why not just do in 
> ada/utils.c:end_subprog_body what you've done in cp/decl.c:grogfndecl?

I've got c++ grokfndecl update somewhat convoluted, but it will also set
DECL_INLINE flag for every declaration.  

C++ is reason why I am not droping DECL_INLINE right away: there are two
problems here.  At the moment (and for a while) we entirely ignore
-fdefault-inline and also DECL_INLINE is controlling production of
instantiations of template methods that might be inlined.  We need to
find way to model -fdefault-inline for new inliner heuristics or drop
that command line option.  I plan to work on this next.

Honza
> 
> -- 
> Eric Botcazou