This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: Ping Re: Make strtod respect the rounding mode (bug 14518)
On Tue, 11 Sep 2012, Richard Henderson wrote:
> On 2012-09-11 14:25, Joseph S. Myers wrote:
> > +/* Internal type for floating-point rounding modes. */
> > +typedef enum
> > +{
> > + round_downward,
> > + round_tonearest,
> > + round_towardzero,
> > + round_upward
> > +} rounding_mode_t;
>
> Is there a good reason why you're defining a new enumeration instead
> of letting the arch-specific header either (1) include bits/fenv.h,
> or (2) define some other names that are bit-compatible with what comes
> out of the fpu?
The idea is to have the same set of *names* present in the code
everywhere, so that code using these values doesn't need to be conditioned
on which are available on a particular architecture. This includes
setting the mode to round_tonearest in cases where there is no rounding
mode support for an architecture and so no FE_* macros are defined at all.
There's no particular need for the *values* to be
architecture-independent; it just keeps the architecture-independent part
of the code simplest, and the compiler can optimize away the use of the
internal enumeration (and the abort () calls, and any cases that are
unreachable for a given architecture because it doesn't support the
relevant rounding modes).
--
Joseph S. Myers
joseph@codesourcery.com