xref: /netbsd-src/external/lgpl3/mpfr/dist/src/set_float128.c (revision ba125506a622fe649968631a56eba5d42ff57863)
1 /* mpfr_set_float128 -- convert a machine _Float128 number to
2                         a multiple precision floating-point number
3 
4 Copyright 2012-2023 Free Software Foundation, Inc.
5 Contributed by the AriC and Caramba projects, INRIA.
6 
7 This file is part of the GNU MPFR Library.
8 
9 The GNU MPFR Library is free software; you can redistribute it and/or modify
10 it under the terms of the GNU Lesser General Public License as published by
11 the Free Software Foundation; either version 3 of the License, or (at your
12 option) any later version.
13 
14 The GNU MPFR Library is distributed in the hope that it will be useful, but
15 WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
16 or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU Lesser General Public
17 License for more details.
18 
19 You should have received a copy of the GNU Lesser General Public License
20 along with the GNU MPFR Library; see the file COPYING.LESSER.  If not, see
21 https://www.gnu.org/licenses/ or write to the Free Software Foundation, Inc.,
22 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA. */
23 
24 #define MPFR_NEED_LONGLONG_H
25 #include "mpfr-impl.h"
26 
27 #ifdef MPFR_WANT_FLOAT128
28 
29 #if MPFR_WANT_FLOAT128 == 1
30 /* _Float128 type from ISO/IEC TS 18661 */
31 # define MPFR_FLOAT128_MAX 0x1.ffffffffffffffffffffffffffffp+16383f128
32 #elif MPFR_WANT_FLOAT128 == 2
33 /* __float128 type (GNU C extension) */
34 # define MPFR_FLOAT128_MAX 0x1.ffffffffffffffffffffffffffffp+16383q
35 #else
36 # error "Unsupported value for MPFR_WANT_FLOAT128"
37 #endif
38 
39 int
mpfr_set_float128(mpfr_ptr r,_Float128 d,mpfr_rnd_t rnd_mode)40 mpfr_set_float128 (mpfr_ptr r, _Float128 d, mpfr_rnd_t rnd_mode)
41 {
42   mpfr_t t;
43   mp_limb_t *tp;
44   int inexact, shift_exp, neg, e, i;
45   _Float128 p[14], q[14];
46   MPFR_SAVE_EXPO_DECL (expo);
47 
48   /* Check for NaN */
49   if (MPFR_UNLIKELY (DOUBLE_ISNAN (d)))
50     {
51       MPFR_SET_NAN(r);
52       MPFR_RET_NAN;
53     }
54 
55   /* Check for INF */
56   if (MPFR_UNLIKELY (d > MPFR_FLOAT128_MAX))
57     {
58       MPFR_SET_INF (r);
59       MPFR_SET_POS (r);
60       return 0;
61     }
62   else if (MPFR_UNLIKELY (d < -MPFR_FLOAT128_MAX))
63     {
64       MPFR_SET_INF (r);
65       MPFR_SET_NEG (r);
66       return 0;
67     }
68   /* Check for ZERO */
69   else if (MPFR_UNLIKELY (d == (_Float128) 0.0))
70     return mpfr_set_d (r, (double) d, rnd_mode);
71 
72   shift_exp = 0; /* invariant: remainder to deal with is d*2^shift_exp */
73   neg = d < 0;
74   if (d < 0)
75     d = -d;
76 
77   /* Scaling, avoiding (slow) divisions. Should the tables be cached? */
78   if (d >= 1.0)
79     {
80       p[0] = 2.0;
81       q[0] = 0.5;
82       e = 1;
83       /* p[i] = 2^(2^i), q[i] = 1/p[i] */
84       for (i = 0; i < 13 && d >= p[i]; i++)
85         {
86           p[i+1] = p[i] * p[i];
87           q[i+1] = q[i] * q[i];
88           e <<= 1;
89         }
90       for (; i >= 0; e >>= 1, i--)
91         if (d >= p[i])
92           {
93             d *= q[i];
94             shift_exp += e;
95           }
96       d *= 0.5;
97       shift_exp++;
98     }
99   else if (d < 0.5)
100     {
101       p[0] = 2.0;
102       q[0] = 0.5;
103       e = 1;
104       /* p[i] = 2^(2^i), q[i] = 1/p[i] */
105       for (i = 0; i < 13 && d < q[i]; i++)
106         {
107           p[i+1] = p[i] * p[i];
108           q[i+1] = q[i] * q[i];
109           e <<= 1;
110         }
111       /* The while() may be needed for i = 13 due to subnormals.
112          This can probably be improved without yielding an underflow. */
113       for (; i >= 0; e >>= 1, i--)
114         while (d < q[i])
115           {
116             d *= p[i];
117             shift_exp -= e;
118           }
119     }
120 
121   MPFR_ASSERTD (d >= 0.5 && d < 1.0);
122 
123   mpfr_init2 (t, IEEE_FLOAT128_MANT_DIG);
124   tp = MPFR_MANT (t);
125 
126   MPFR_SAVE_EXPO_MARK (expo);
127   MPFR_SET_EXP (t, shift_exp);
128   MPFR_SET_SIGN (t, neg ? MPFR_SIGN_NEG : MPFR_SIGN_POS);
129 
130   for (i = MPFR_LAST_LIMB (t); i >= 0; i--)
131     {
132       d *= 2 * (_Float128) MPFR_LIMB_HIGHBIT;
133       tp[i] = (mp_limb_t) d;
134       d -= tp[i];
135     }
136 
137   inexact = mpfr_set (r, t, rnd_mode);
138   mpfr_clear (t);
139 
140   MPFR_SAVE_EXPO_FREE (expo);
141   return mpfr_check_range (r, inexact, rnd_mode);
142 }
143 
144 #endif /* MPFR_WANT_FLOAT128 */
145