Date: Thu, 31 Oct 2002 23:36:13 +1100 (EST) From: Bruce Evans <bde@zeta.org.au> To: David Schultz <dschultz@uclink.Berkeley.EDU> Cc: "M. Warner Losh" <imp@bsdimp.com>, <tlambert2@mindspring.com>, <rittle@labs.mot.com>, <rittle@latour.rsch.comm.mot.com>, <current@FreeBSD.ORG> Subject: Re: Lack of real long double support Message-ID: <20021031225719.V11320-100000@gamplex.bde.org> In-Reply-To: <20021031112623.GA28135@HAL9000.homeunix.com>
next in thread | previous in thread | raw e-mail | index | archive | help
On Thu, 31 Oct 2002, David Schultz wrote:
> Thus spake Bruce Evans <bde@zeta.org.au>:
> > $ cc -o z z.c
> > $ ./z
> > LDBL_EPSILON failed test 1 with prec 2
> > $ cc -O -o z z.c.
> > $ ./z
> > LDBL_EPSILON failed test 1 with prec 2
> > DBL_EPSILON failed test 2 with prec 3
> > %%%
> >
> > The full brokenness only shows up with -O.
>
> Actually, the _full_ brokenness of floating point in FreeBSD shows
> up only with -O2. :-( A number of library functions
> (e.g. isinf()) use unions for type punning, which violates C's
> aliasing rules.
>
> It turns out that there's a bad interaction
> between gcc3's -fschedule-insns and -fstrict-aliasing (both
> implied by -O2) that can cause problems. The best fix I know of
> is to use unions in the limited way that gcc's optimizer knows how
> to handle in a reasonable way.
Eeek.
gcc.info says one version is required to work even with -fstrict-aliasing.
The problems seems to be limited, since msun was fixed before rev.1.1
to use this version. From msun/src/math_private.h:
%%%
/*
* The original fdlibm code used statements like:
* n0 = ((*(int*)&one)>>29)^1; * index of high word *
* ix0 = *(n0+(int*)&x); * high word of x *
* ix1 = *((1-n0)+(int*)&x); * low word of x *
* to dig two 32 bit words out of the 64 bit IEEE floating point
* value. That is non-ANSI, and, moreover, the gcc instruction
* scheduler gets it wrong. We instead use the following macros.
* Unlike the original code, we determine the endianness at compile
* time, not at run time; I don't see much benefit to selecting
* endianness at run time.
*/
...
#define EXTRACT_WORDS(ix0,ix1,d) \
do { \
ieee_double_shape_type ew_u; \
ew_u.value = (d); \
(ix0) = ew_u.parts.msw; \
(ix1) = ew_u.parts.lsw; \
} while (0)
...
%%%
This stores the double into the union and doesn't use any pointers to
access the results, so it should work. isinf.c and even the "fixed"
version of strtod.c uses a dubious cast to pun the double to a union.
I think the store gets optimized away as far as possible.
Bruce
To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-current" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20021031225719.V11320-100000>