Re: [PATCH v3 8/8] float128: Add wrappers to override ldbl-128 as float128.

[Joseph Myers <joseph at codesourcery dot com>]

On Tue, 9 May 2017, Gabriel F. T. Gomes wrote:

> 	* math/math.h [__HAVE_FLOAT128 && __GLIBC_USE (IEC_60559_TYPES_EXT)]
> 	(SNANF128): New macro.

As noted, I think this belongs in patch 4 alongside HUGE_VAL_F128.

> 	* sysdeps/ieee754/float128/s_canonicalizef128.c: New file.

You shouldn't need this file.  The type-generic templates machinery should 
build canonicalizef128 from the template automatically when you enable 
float128 support.

> diff --git a/math/math.h b/math/math.h
> index 4fc96ac..ed713df 100644
> --- a/math/math.h
> +++ b/math/math.h
> @@ -63,6 +63,9 @@ __BEGIN_DECLS
>  #  define SNAN (__builtin_nans (""))
>  #  define SNANL (__builtin_nansl (""))
>  # endif
> +# if __HAVE_FLOAT128 && __GLIBC_USE (IEC_60559_TYPES_EXT)
> +#  define SNANF128 (__builtin_nansf128 (""))
> +# endif
>  #endif

This (moved to patch 4) should move outside the __GLIBC_USE 
(IEC_60559_BFP_EXT) conditional (as with the function declarations, macros 
added in TS 18661-1 have their _FloatN versions defined with 
__STDC_WANT_IEC_60559_TYPES_EXT__, whether or not 
__STDC_WANT_IEC_60559_BFP_EXT__ is defined).

> diff --git a/sysdeps/ieee754/float128/s_fromfpf128.c b/sysdeps/ieee754/float128/s_fromfpf128.c
> new file mode 100644
> index 0000000..891de3d
> --- /dev/null
> +++ b/sysdeps/ieee754/float128/s_fromfpf128.c
> @@ -0,0 +1,5 @@
> +#define UNSIGNED 0
> +#define INEXACT 0
> +#define FUNC fromfpf128
> +#include <float128_private.h>
> +#include "../ldbl-128/s_fromfpl_main.c"
> diff --git a/sysdeps/ieee754/float128/s_fromfpxf128.c b/sysdeps/ieee754/float128/s_fromfpxf128.c
> new file mode 100644
> index 0000000..0159ae2
> --- /dev/null
> +++ b/sysdeps/ieee754/float128/s_fromfpxf128.c
> @@ -0,0 +1,5 @@
> +#define UNSIGNED 0
> +#define INEXACT 1
> +#define FUNC fromfpxl

I'd say that for the fromfp functions you should be consistent about 
whether the wrappers define FUNC to the *f128 name, or whether you use a 
separate define of the *l name to the *f128 name and define FUNC to the *l 
name.

> diff --git a/sysdeps/ieee754/float128/w_llogbf128.c b/sysdeps/ieee754/float128/w_llogbf128.c
> new file mode 100644
> index 0000000..f83657d
> --- /dev/null
> +++ b/sysdeps/ieee754/float128/w_llogbf128.c
> @@ -0,0 +1,3 @@
> +#define declare_mgen_alias(from, to) weak_alias (M_SUF (from), M_SUF (to))
> +#include <math-type-macros-float128.h>
> +#include <w_llogb_template.c>

That's the default definition of declare_mgen_alias.  You shouldn't need 
this file; the template machinery should do the right thing without it.

-- 
Joseph S. Myers
joseph@codesourcery.com