Re: Serious bug

["Charles M. Hannum" <root at ihack dot net>]

Actually, I have some data on this, regarding the default rounding and
precision modes on various UN*X-like OSes.  As you can see, the only
one I tested which does not use IEEE double precision by default is
Linux.  A bunch of systems leave exceptions unmasked, though, which
means they do not comply with IEEE 754/854 by default.

(I'm not going to make any value judgement as to whether complying
with IEEE 754/854 is a useful goal.)

There is a semi-standard interface to change the rounding mode and
exception mask: fpgetround(), fpsetround(), fpgetmask() and
fpsetmask().  However, there is no portable way of changing the
precision.  On most systems, to get correct rounding for `float's, you
need to either write some assembly language goop to change the
precision, or use -ffloat-store.


This is just pasted from a source file:

/*
 * The i387 defaults to Intel extended precision mode and round to nearest,
 * with all exceptions masked.
 */
#define __INITIAL_NPXCW__       0x037f
/* NetBSD uses IEEE double precision. */
#define __NetBSD_NPXCW__        0x127f
/* FreeBSD leaves some exceptions unmasked as well. */
#define __FreeBSD_NPXCW__       0x1272
/* iBCS2 goes a bit further and leaves the underflow exception unmasked. */
#define __iBCS2_NPXCW__         0x0262
/* Linux just uses the default control word. */
#define __Linux_NPXCW__         0x037f
/* SVR4 uses the same control word as iBCS2. */
#define __SVR4_NPXCW__          0x0262