[committed] Add __float128 type support to hppa-hpux

[John David Anglin]

On Wed, 26 Jan 2011, Joseph S. Myers wrote:

> On Tue, 25 Jan 2011, John David Anglin wrote:
> 
> > > 
> > > On Tue, 18 Jan 2011, John David Anglin wrote:
> > > 
> > > > +      /* TFmode support builtins.  */
> > > > +      ftype = build_function_type_list (long_double_type_node,
> > > > +					long_double_type_node,
> > > > +					NULL_TREE);
> > > > +      decl = add_builtin_function ("__builtin_fabsq", ftype,
> > > > +				   PA_BUILTIN_FABSQ, BUILT_IN_MD,
> > > > +				   "_U_Qfabs", NULL_TREE);
> > > 
> > > Adding these built-in functions is not appropriate.  The correct model for 
> > > __float128 as an alias for long double is IA64 HP-UX, where these 
> > > functions are not provided, or __float80 on x86, where again no special 
> > > functions are needed or provided.
> > 
> > "_U_Qfabs" is provided by HP on PA-RISC.
> 
> I don't see how that relates to my comment.  The existence of some 
> function in HP's libraries is independent of the uselessness of a built-in 
> function that just duplicates __builtin_fabsl; my point is that no such 
> duplicative built-in function should be added.

Unfortunately, the duplication is needed to provide __builtin_fabsq,
etc.  Operations like addition transparently use the long double builtin
support, but __builtin_fabsq doesn't automatically use the __builtin_fabsl
builtin support when __float128 is an alias for long double.  I suppose
the same is true for __float80.

Dave
-- 
J. David Anglin                                  dave.anglin@nrc-cnrc.gc.ca
National Research Council of Canada              (613) 990-0752 (FAX: 952-6602)