diff options
Diffstat (limited to 'get_ld.c')
| -rw-r--r-- | get_ld.c | 17 |
1 files changed, 12 insertions, 5 deletions
@@ -1,7 +1,7 @@ /* mpfr_get_ld -- convert a multiple precision floating-point number to a machine long double -Copyright 2002, 2003, 2004 Free Software Foundation, Inc. +Copyright 2002, 2003, 2004, 2005 Free Software Foundation, Inc. This file is part of the MPFR Library. @@ -22,9 +22,16 @@ MA 02111-1307, USA. */ #include <float.h> - #include "mpfr-impl.h" +/* FIXME: rounding direction can be wrong if |x| < LDBL_MIN. */ + +/* The MPFR number rounded to the size MPFR_LDBL_MANT_DIG of a long double + will be scaled so that all its bits are representable in a double. + Therefore the exponent of its LSB must be >= -1074, and the exponent + of the MPFR number must be >= -1074 + MPFR_LDBL_MANT_DIG. */ +#define EMIN (-1074 + MPFR_LDBL_MANT_DIG) + long double mpfr_get_ld (mpfr_srcptr x, mp_rnd_t rnd_mode) { @@ -57,10 +64,10 @@ mpfr_get_ld (mpfr_srcptr x, mp_rnd_t rnd_mode) sh = e - 1023; MPFR_SET_EXP (y, 1023); } - else if (e < -1021) + else if (e < EMIN) { - sh = e + 1021; - MPFR_SET_EXP (y, -1021); + sh = e - EMIN; + MPFR_SET_EXP (y, EMIN); } else { |