68
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
1 /* Compute x * y + z as ternary operation.
|
111
|
2 Copyright (C) 2010-2017 Free Software Foundation, Inc.
|
68
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
3 This file is part of the GNU C Library.
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
4 Contributed by Jakub Jelinek <jakub@redhat.com>, 2010.
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
5
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
6 The GNU C Library is free software; you can redistribute it and/or
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
7 modify it under the terms of the GNU Lesser General Public
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
8 License as published by the Free Software Foundation; either
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
9 version 2.1 of the License, or (at your option) any later version.
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
10
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
11 The GNU C Library is distributed in the hope that it will be useful,
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
12 but WITHOUT ANY WARRANTY; without even the implied warranty of
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
14 Lesser General Public License for more details.
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
15
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
16 You should have received a copy of the GNU Lesser General Public
|
111
|
17 License along with the GNU C Library; if not, see
|
|
18 <http://www.gnu.org/licenses/>. */
|
68
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
19
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
20 #include "quadmath-imp.h"
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
21 #include <math.h>
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
22 #include <float.h>
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
23 #ifdef HAVE_FENV_H
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
24 # include <fenv.h>
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
25 # if defined HAVE_FEHOLDEXCEPT && defined HAVE_FESETROUND \
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
26 && defined HAVE_FEUPDATEENV && defined HAVE_FETESTEXCEPT \
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
27 && defined FE_TOWARDZERO && defined FE_INEXACT
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
28 # define USE_FENV_H
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
29 # endif
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
30 #endif
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
31
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
32 /* This implementation uses rounding to odd to avoid problems with
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
33 double rounding. See a paper by Boldo and Melquiond:
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
34 http://www.lri.fr/~melquion/doc/08-tc.pdf */
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
35
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
36 __float128
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
37 fmaq (__float128 x, __float128 y, __float128 z)
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
38 {
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
39 ieee854_float128 u, v, w;
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
40 int adjust = 0;
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
41 u.value = x;
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
42 v.value = y;
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
43 w.value = z;
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
44 if (__builtin_expect (u.ieee.exponent + v.ieee.exponent
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
45 >= 0x7fff + IEEE854_FLOAT128_BIAS
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
46 - FLT128_MANT_DIG, 0)
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
47 || __builtin_expect (u.ieee.exponent >= 0x7fff - FLT128_MANT_DIG, 0)
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
48 || __builtin_expect (v.ieee.exponent >= 0x7fff - FLT128_MANT_DIG, 0)
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
49 || __builtin_expect (w.ieee.exponent >= 0x7fff - FLT128_MANT_DIG, 0)
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
50 || __builtin_expect (u.ieee.exponent + v.ieee.exponent
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
51 <= IEEE854_FLOAT128_BIAS + FLT128_MANT_DIG, 0))
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
52 {
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
53 /* If z is Inf, but x and y are finite, the result should be
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
54 z rather than NaN. */
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
55 if (w.ieee.exponent == 0x7fff
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
56 && u.ieee.exponent != 0x7fff
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
57 && v.ieee.exponent != 0x7fff)
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
58 return (z + x) + y;
|
111
|
59 /* If z is zero and x are y are nonzero, compute the result
|
|
60 as x * y to avoid the wrong sign of a zero result if x * y
|
|
61 underflows to 0. */
|
|
62 if (z == 0 && x != 0 && y != 0)
|
|
63 return x * y;
|
|
64 /* If x or y or z is Inf/NaN, or if x * y is zero, compute as
|
|
65 x * y + z. */
|
68
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
66 if (u.ieee.exponent == 0x7fff
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
67 || v.ieee.exponent == 0x7fff
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
68 || w.ieee.exponent == 0x7fff
|
111
|
69 || x == 0
|
|
70 || y == 0)
|
68
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
71 return x * y + z;
|
111
|
72 /* If fma will certainly overflow, compute as x * y. */
|
|
73 if (u.ieee.exponent + v.ieee.exponent
|
|
74 > 0x7fff + IEEE854_FLOAT128_BIAS)
|
|
75 return x * y;
|
|
76 /* If x * y is less than 1/4 of FLT128_DENORM_MIN, neither the
|
|
77 result nor whether there is underflow depends on its exact
|
|
78 value, only on its sign. */
|
|
79 if (u.ieee.exponent + v.ieee.exponent
|
|
80 < IEEE854_FLOAT128_BIAS - FLT128_MANT_DIG - 2)
|
|
81 {
|
|
82 int neg = u.ieee.negative ^ v.ieee.negative;
|
|
83 __float128 tiny = neg ? -0x1p-16494Q : 0x1p-16494Q;
|
|
84 if (w.ieee.exponent >= 3)
|
|
85 return tiny + z;
|
|
86 /* Scaling up, adding TINY and scaling down produces the
|
|
87 correct result, because in round-to-nearest mode adding
|
|
88 TINY has no effect and in other modes double rounding is
|
|
89 harmless. But it may not produce required underflow
|
|
90 exceptions. */
|
|
91 v.value = z * 0x1p114Q + tiny;
|
|
92 if (TININESS_AFTER_ROUNDING
|
|
93 ? v.ieee.exponent < 115
|
|
94 : (w.ieee.exponent == 0
|
|
95 || (w.ieee.exponent == 1
|
|
96 && w.ieee.negative != neg
|
|
97 && w.ieee.mant_low == 0
|
|
98 && w.ieee.mant_high == 0)))
|
|
99 {
|
|
100 __float128 force_underflow = x * y;
|
|
101 math_force_eval (force_underflow);
|
|
102 }
|
|
103 return v.value * 0x1p-114Q;
|
|
104 }
|
68
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
105 if (u.ieee.exponent + v.ieee.exponent
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
106 >= 0x7fff + IEEE854_FLOAT128_BIAS - FLT128_MANT_DIG)
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
107 {
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
108 /* Compute 1p-113 times smaller result and multiply
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
109 at the end. */
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
110 if (u.ieee.exponent > v.ieee.exponent)
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
111 u.ieee.exponent -= FLT128_MANT_DIG;
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
112 else
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
113 v.ieee.exponent -= FLT128_MANT_DIG;
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
114 /* If x + y exponent is very large and z exponent is very small,
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
115 it doesn't matter if we don't adjust it. */
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
116 if (w.ieee.exponent > FLT128_MANT_DIG)
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
117 w.ieee.exponent -= FLT128_MANT_DIG;
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
118 adjust = 1;
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
119 }
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
120 else if (w.ieee.exponent >= 0x7fff - FLT128_MANT_DIG)
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
121 {
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
122 /* Similarly.
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
123 If z exponent is very large and x and y exponents are
|
111
|
124 very small, adjust them up to avoid spurious underflows,
|
|
125 rather than down. */
|
|
126 if (u.ieee.exponent + v.ieee.exponent
|
|
127 <= IEEE854_FLOAT128_BIAS + FLT128_MANT_DIG)
|
|
128 {
|
|
129 if (u.ieee.exponent > v.ieee.exponent)
|
|
130 u.ieee.exponent += 2 * FLT128_MANT_DIG + 2;
|
|
131 else
|
|
132 v.ieee.exponent += 2 * FLT128_MANT_DIG + 2;
|
|
133 }
|
|
134 else if (u.ieee.exponent > v.ieee.exponent)
|
68
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
135 {
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
136 if (u.ieee.exponent > FLT128_MANT_DIG)
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
137 u.ieee.exponent -= FLT128_MANT_DIG;
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
138 }
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
139 else if (v.ieee.exponent > FLT128_MANT_DIG)
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
140 v.ieee.exponent -= FLT128_MANT_DIG;
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
141 w.ieee.exponent -= FLT128_MANT_DIG;
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
142 adjust = 1;
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
143 }
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
144 else if (u.ieee.exponent >= 0x7fff - FLT128_MANT_DIG)
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
145 {
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
146 u.ieee.exponent -= FLT128_MANT_DIG;
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
147 if (v.ieee.exponent)
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
148 v.ieee.exponent += FLT128_MANT_DIG;
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
149 else
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
150 v.value *= 0x1p113Q;
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
151 }
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
152 else if (v.ieee.exponent >= 0x7fff - FLT128_MANT_DIG)
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
153 {
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
154 v.ieee.exponent -= FLT128_MANT_DIG;
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
155 if (u.ieee.exponent)
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
156 u.ieee.exponent += FLT128_MANT_DIG;
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
157 else
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
158 u.value *= 0x1p113Q;
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
159 }
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
160 else /* if (u.ieee.exponent + v.ieee.exponent
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
161 <= IEEE854_FLOAT128_BIAS + FLT128_MANT_DIG) */
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
162 {
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
163 if (u.ieee.exponent > v.ieee.exponent)
|
111
|
164 u.ieee.exponent += 2 * FLT128_MANT_DIG + 2;
|
68
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
165 else
|
111
|
166 v.ieee.exponent += 2 * FLT128_MANT_DIG + 2;
|
|
167 if (w.ieee.exponent <= 4 * FLT128_MANT_DIG + 6)
|
68
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
168 {
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
169 if (w.ieee.exponent)
|
111
|
170 w.ieee.exponent += 2 * FLT128_MANT_DIG + 2;
|
68
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
171 else
|
111
|
172 w.value *= 0x1p228Q;
|
68
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
173 adjust = -1;
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
174 }
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
175 /* Otherwise x * y should just affect inexact
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
176 and nothing else. */
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
177 }
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
178 x = u.value;
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
179 y = v.value;
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
180 z = w.value;
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
181 }
|
111
|
182
|
|
183 /* Ensure correct sign of exact 0 + 0. */
|
|
184 if (__builtin_expect ((x == 0 || y == 0) && z == 0, 0))
|
|
185 {
|
|
186 x = math_opt_barrier (x);
|
|
187 return x * y + z;
|
|
188 }
|
|
189
|
|
190 #ifdef USE_FENV_H
|
|
191 fenv_t env;
|
|
192 feholdexcept (&env);
|
|
193 fesetround (FE_TONEAREST);
|
|
194 #endif
|
|
195
|
68
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
196 /* Multiplication m1 + m2 = x * y using Dekker's algorithm. */
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
197 #define C ((1LL << (FLT128_MANT_DIG + 1) / 2) + 1)
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
198 __float128 x1 = x * C;
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
199 __float128 y1 = y * C;
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
200 __float128 m1 = x * y;
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
201 x1 = (x - x1) + x1;
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
202 y1 = (y - y1) + y1;
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
203 __float128 x2 = x - x1;
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
204 __float128 y2 = y - y1;
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
205 __float128 m2 = (((x1 * y1 - m1) + x1 * y2) + x2 * y1) + x2 * y2;
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
206
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
207 /* Addition a1 + a2 = z + m1 using Knuth's algorithm. */
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
208 __float128 a1 = z + m1;
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
209 __float128 t1 = a1 - z;
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
210 __float128 t2 = a1 - t1;
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
211 t1 = m1 - t1;
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
212 t2 = z - t2;
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
213 __float128 a2 = t1 + t2;
|
111
|
214 /* Ensure the arithmetic is not scheduled after feclearexcept call. */
|
|
215 math_force_eval (m2);
|
|
216 math_force_eval (a2);
|
|
217 #ifdef USE_FENV_H
|
|
218 feclearexcept (FE_INEXACT);
|
|
219 #endif
|
|
220
|
|
221 /* If the result is an exact zero, ensure it has the correct sign. */
|
|
222 if (a1 == 0 && m2 == 0)
|
|
223 {
|
|
224 #ifdef USE_FENV_H
|
|
225 feupdateenv (&env);
|
|
226 #endif
|
|
227 /* Ensure that round-to-nearest value of z + m1 is not reused. */
|
|
228 z = math_opt_barrier (z);
|
|
229 return z + m1;
|
|
230 }
|
68
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
231
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
232 #ifdef USE_FENV_H
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
233 fesetround (FE_TOWARDZERO);
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
234 #endif
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
235 /* Perform m2 + a2 addition with round to odd. */
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
236 u.value = a2 + m2;
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
237
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
238 if (__builtin_expect (adjust == 0, 1))
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
239 {
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
240 #ifdef USE_FENV_H
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
241 if ((u.ieee.mant_low & 1) == 0 && u.ieee.exponent != 0x7fff)
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
242 u.ieee.mant_low |= fetestexcept (FE_INEXACT) != 0;
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
243 feupdateenv (&env);
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
244 #endif
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
245 /* Result is a1 + u.value. */
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
246 return a1 + u.value;
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
247 }
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
248 else if (__builtin_expect (adjust > 0, 1))
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
249 {
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
250 #ifdef USE_FENV_H
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
251 if ((u.ieee.mant_low & 1) == 0 && u.ieee.exponent != 0x7fff)
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
252 u.ieee.mant_low |= fetestexcept (FE_INEXACT) != 0;
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
253 feupdateenv (&env);
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
254 #endif
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
255 /* Result is a1 + u.value, scaled up. */
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
256 return (a1 + u.value) * 0x1p113Q;
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
257 }
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
258 else
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
259 {
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
260 #ifdef USE_FENV_H
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
261 if ((u.ieee.mant_low & 1) == 0)
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
262 u.ieee.mant_low |= fetestexcept (FE_INEXACT) != 0;
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
263 #endif
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
264 v.value = a1 + u.value;
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
265 /* Ensure the addition is not scheduled after fetestexcept call. */
|
111
|
266 asm volatile ("" : : "m" (v.value));
|
68
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
267 #ifdef USE_FENV_H
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
268 int j = fetestexcept (FE_INEXACT) != 0;
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
269 feupdateenv (&env);
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
270 #else
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
271 int j = 0;
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
272 #endif
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
273 /* Ensure the following computations are performed in default rounding
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
274 mode instead of just reusing the round to zero computation. */
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
275 asm volatile ("" : "=m" (u) : "m" (u));
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
276 /* If a1 + u.value is exact, the only rounding happens during
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
277 scaling down. */
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
278 if (j == 0)
|
111
|
279 return v.value * 0x1p-228Q;
|
68
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
280 /* If result rounded to zero is not subnormal, no double
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
281 rounding will occur. */
|
111
|
282 if (v.ieee.exponent > 228)
|
|
283 return (a1 + u.value) * 0x1p-228Q;
|
|
284 /* If v.value * 0x1p-228Q with round to zero is a subnormal above
|
|
285 or equal to FLT128_MIN / 2, then v.value * 0x1p-228Q shifts mantissa
|
68
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
286 down just by 1 bit, which means v.ieee.mant_low |= j would
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
287 change the round bit, not sticky or guard bit.
|
111
|
288 v.value * 0x1p-228Q never normalizes by shifting up,
|
68
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
289 so round bit plus sticky bit should be already enough
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
290 for proper rounding. */
|
111
|
291 if (v.ieee.exponent == 228)
|
68
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
292 {
|
111
|
293 /* If the exponent would be in the normal range when
|
|
294 rounding to normal precision with unbounded exponent
|
|
295 range, the exact result is known and spurious underflows
|
|
296 must be avoided on systems detecting tininess after
|
|
297 rounding. */
|
|
298 if (TININESS_AFTER_ROUNDING)
|
|
299 {
|
|
300 w.value = a1 + u.value;
|
|
301 if (w.ieee.exponent == 229)
|
|
302 return w.value * 0x1p-228Q;
|
|
303 }
|
68
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
304 /* v.ieee.mant_low & 2 is LSB bit of the result before rounding,
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
305 v.ieee.mant_low & 1 is the round bit and j is our sticky
|
111
|
306 bit. */
|
|
307 w.value = 0.0Q;
|
|
308 w.ieee.mant_low = ((v.ieee.mant_low & 3) << 1) | j;
|
|
309 w.ieee.negative = v.ieee.negative;
|
|
310 v.ieee.mant_low &= ~3U;
|
|
311 v.value *= 0x1p-228Q;
|
|
312 w.value *= 0x1p-2Q;
|
|
313 return v.value + w.value;
|
68
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
314 }
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
315 v.ieee.mant_low |= j;
|
111
|
316 return v.value * 0x1p-228Q;
|
68
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
317 }
|
Nobuyasu Oshiro <dimolto@cr.ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
318 }
|