Re: [PATCH v3 4/8] float128: Add public _Float128 declarations to libm.

[Joseph Myers <joseph at codesourcery dot com>]

On Tue, 9 May 2017, Joseph Myers wrote:

> > +/* Include the file of declarations again, this time using `_Float128'
> > +   instead of `double' and appending f128 to each function name.  */
> > +
> > +#if __HAVE_DISTINCT_FLOAT128
> 
> That's not the right logically condition for when to declare (subject to 
> __GLIBC_USE (IEC_60559_TYPES_EXT)) most functions.  I think 
> __HAVE_FLOAT128 && !defined _LIBC might be better, by analogy with the 
> conditions for declaring long double functions.  (long double tests 
> _LIBC_TEST as well, but _LIBC_TEST use ought to be obsolete after Zack's 
> patch to avoid internal declarations, including _LIBC, for most of the 
> testsuite).

Correction:

If there's a distinct _Float128, the functions should be declared, both 
inside and outside libc.  If it exists but is not distinct, they should 
only be declared outside libc.  So:

#if __HAVE_DISTINCT_FLOAT128 || (__HAVE_FLOAT128 && !defined _LIBC)

should be right.

Also, for the patch where bits/floatn.h is added for powerpc64le: note 
that GCC 7 supports _Float128 and the f128 constant suffix for C, not for 
C++.  The header will need to do appropriate things for C++: either 
declare the feature not to be supported, or, if you want to be able to use 
the APIs for C++ (which seems sensible), map _Float128 to __float128 for 
C++ even for GCC 7 and later.

-- 
Joseph S. Myers
joseph@codesourcery.com