This is the mail archive of the mailing list for the GCC project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

LIA-1, denormals and is_iec559 and C++


Below is the answer I got from the SC22/WG11 committee about the
possibility of having has_denormal == denormal_absent and 
is_iec559 == true:  That isn't right.


-- Gaby

--- Begin Message ---

Since IEC 60559 requires subnormals, iec_559_F=TRUE implies that
denorm_F=TRUE (otherwise the requirements of LIA-1 are not
fulfilled).  For a system where subnormals can be turned on/off,
these parameters must vary accordingly (when subnormals are off
it is not an IEC 60559 conforming system at that time).

(On the other hand, LIA gives rather free hands to implementations
to change things, but such changes must be clearly documented,
and preferably well motivated.)

		Kind regards
		/Kent Karlsson
		(editor for the LIA series)

> -----Original Message-----
> From: 
> []On Behalf Of Gabriel Dos Reis
> Sent: den 24 september 2002 04:00
> To:
> Cc:
> Subject: (SC22WG11.979) LIA-1, denormals and is_iec559 and C++
> [Note: this message is primarily sent to WG11 list with a copy to the
>  C++ Library Working Group reflector for record.  Please, when
>  replying, be sure to reply to both (especially from the LWG side) if
>  the message is of interest to both group.  ]
> Hi,
> The following issue pops up while working on the GNU implementation of
> the C++ standard library class std::numeric_limits<>.
>    Basically, it reads as follows:
>        Can is_iec559 be true for a floating-point type system that
>        does not support denormals (subnormals in the new 
> terminology)? 
>    The C++ standard gives an example with is_iec559 = true and
> has_denormal = denormal_absent, but I'm not sure the example is right.
> Thanks,
> -- Gaby

--- End Message ---

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]