Q: middle-end problem when variadic builtins promote float to double

[Richard Guenther]

On 7/8/07, Kaveh R. GHAZI <ghazi@caip.rutgers.edu> wrote:
> On Sun, 8 Jul 2007, Richard Guenther wrote:
>
> > So type-generic is supposed to apply to scalar floating point types
> > only?
>
> So far, yes.
>
> I don't see anything that requires or prohibits changing that for the
> initial implementation.  If later a GCC developer wants to change it to
> not promote e.g. char, that could be a refinement.  I don't think any of
> the type-generic builtins allows non-floats anyway, so it would have no
> effect to have it or not at this point.

Ok.

>
> >  Btw. you need to add documentation for this function attribute.
>
> The internal "no vops" builtin isn't documented AFAICT, I was just
> following that example.  We don't want to confuse users into thinking they
> can use it IMHO.  What do you suggest?

Ah, I didn't notice that it had a space in its name.  I guess not documenting
it is ok then.

Richard.