1 | /* log1pl.c |
2 | * |
3 | * Relative error logarithm |
4 | * Natural logarithm of 1+x for __float128 precision |
5 | * |
6 | * |
7 | * |
8 | * SYNOPSIS: |
9 | * |
10 | * __float128 x, y, log1pl(); |
11 | * |
12 | * y = log1pq( x ); |
13 | * |
14 | * |
15 | * |
16 | * DESCRIPTION: |
17 | * |
18 | * Returns the base e (2.718...) logarithm of 1+x. |
19 | * |
20 | * The argument 1+x is separated into its exponent and fractional |
21 | * parts. If the exponent is between -1 and +1, the logarithm |
22 | * of the fraction is approximated by |
23 | * |
24 | * log(1+x) = x - 0.5 x^2 + x^3 P(x)/Q(x). |
25 | * |
26 | * Otherwise, setting z = 2(w-1)/(w+1), |
27 | * |
28 | * log(w) = z + z^3 P(z)/Q(z). |
29 | * |
30 | * |
31 | * |
32 | * ACCURACY: |
33 | * |
34 | * Relative error: |
35 | * arithmetic domain # trials peak rms |
36 | * IEEE -1, 8 100000 1.9e-34 4.3e-35 |
37 | */ |
38 | |
39 | /* Copyright 2001 by Stephen L. Moshier |
40 | |
41 | This library is free software; you can redistribute it and/or |
42 | modify it under the terms of the GNU Lesser General Public |
43 | License as published by the Free Software Foundation; either |
44 | version 2.1 of the License, or (at your option) any later version. |
45 | |
46 | This library is distributed in the hope that it will be useful, |
47 | but WITHOUT ANY WARRANTY; without even the implied warranty of |
48 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
49 | Lesser General Public License for more details. |
50 | |
51 | You should have received a copy of the GNU Lesser General Public |
52 | License along with this library; if not, write to the Free Software |
53 | Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA */ |
54 | |
55 | |
56 | #include "quadmath-imp.h" |
57 | |
58 | /* Coefficients for log(1+x) = x - x^2 / 2 + x^3 P(x)/Q(x) |
59 | * 1/sqrt(2) <= 1+x < sqrt(2) |
60 | * Theoretical peak relative error = 5.3e-37, |
61 | * relative peak error spread = 2.3e-14 |
62 | */ |
63 | static const __float128 |
64 | P12 = 1.538612243596254322971797716843006400388E-6Q, |
65 | P11 = 4.998469661968096229986658302195402690910E-1Q, |
66 | P10 = 2.321125933898420063925789532045674660756E1Q, |
67 | P9 = 4.114517881637811823002128927449878962058E2Q, |
68 | P8 = 3.824952356185897735160588078446136783779E3Q, |
69 | P7 = 2.128857716871515081352991964243375186031E4Q, |
70 | P6 = 7.594356839258970405033155585486712125861E4Q, |
71 | P5 = 1.797628303815655343403735250238293741397E5Q, |
72 | P4 = 2.854829159639697837788887080758954924001E5Q, |
73 | P3 = 3.007007295140399532324943111654767187848E5Q, |
74 | P2 = 2.014652742082537582487669938141683759923E5Q, |
75 | P1 = 7.771154681358524243729929227226708890930E4Q, |
76 | P0 = 1.313572404063446165910279910527789794488E4Q, |
77 | /* Q12 = 1.000000000000000000000000000000000000000E0Q, */ |
78 | Q11 = 4.839208193348159620282142911143429644326E1Q, |
79 | Q10 = 9.104928120962988414618126155557301584078E2Q, |
80 | Q9 = 9.147150349299596453976674231612674085381E3Q, |
81 | Q8 = 5.605842085972455027590989944010492125825E4Q, |
82 | Q7 = 2.248234257620569139969141618556349415120E5Q, |
83 | Q6 = 6.132189329546557743179177159925690841200E5Q, |
84 | Q5 = 1.158019977462989115839826904108208787040E6Q, |
85 | Q4 = 1.514882452993549494932585972882995548426E6Q, |
86 | Q3 = 1.347518538384329112529391120390701166528E6Q, |
87 | Q2 = 7.777690340007566932935753241556479363645E5Q, |
88 | Q1 = 2.626900195321832660448791748036714883242E5Q, |
89 | Q0 = 3.940717212190338497730839731583397586124E4Q; |
90 | |
91 | /* Coefficients for log(x) = z + z^3 P(z^2)/Q(z^2), |
92 | * where z = 2(x-1)/(x+1) |
93 | * 1/sqrt(2) <= x < sqrt(2) |
94 | * Theoretical peak relative error = 1.1e-35, |
95 | * relative peak error spread 1.1e-9 |
96 | */ |
97 | static const __float128 |
98 | R5 = -8.828896441624934385266096344596648080902E-1Q, |
99 | R4 = 8.057002716646055371965756206836056074715E1Q, |
100 | R3 = -2.024301798136027039250415126250455056397E3Q, |
101 | R2 = 2.048819892795278657810231591630928516206E4Q, |
102 | R1 = -8.977257995689735303686582344659576526998E4Q, |
103 | R0 = 1.418134209872192732479751274970992665513E5Q, |
104 | /* S6 = 1.000000000000000000000000000000000000000E0Q, */ |
105 | S5 = -1.186359407982897997337150403816839480438E2Q, |
106 | S4 = 3.998526750980007367835804959888064681098E3Q, |
107 | S3 = -5.748542087379434595104154610899551484314E4Q, |
108 | S2 = 4.001557694070773974936904547424676279307E5Q, |
109 | S1 = -1.332535117259762928288745111081235577029E6Q, |
110 | S0 = 1.701761051846631278975701529965589676574E6Q; |
111 | |
112 | /* C1 + C2 = ln 2 */ |
113 | static const __float128 C1 = 6.93145751953125E-1Q; |
114 | static const __float128 C2 = 1.428606820309417232121458176568075500134E-6Q; |
115 | |
116 | static const __float128 sqrth = 0.7071067811865475244008443621048490392848Q; |
117 | static const __float128 zero = 0.0Q; |
118 | |
119 | |
120 | __float128 |
121 | log1pq (__float128 xm1) |
122 | { |
123 | __float128 x, y, z, r, s; |
124 | ieee854_float128 u; |
125 | int32_t hx; |
126 | int e; |
127 | |
128 | /* Test for NaN or infinity input. */ |
129 | u.value = xm1; |
130 | hx = u.words32.w0; |
131 | if ((hx & 0x7fffffff) >= 0x7fff0000) |
132 | return xm1 + fabsq (xm1); |
133 | |
134 | /* log1p(+- 0) = +- 0. */ |
135 | if (((hx & 0x7fffffff) == 0) |
136 | && (u.words32.w1 | u.words32.w2 | u.words32.w3) == 0) |
137 | return xm1; |
138 | |
139 | if ((hx & 0x7fffffff) < 0x3f8e0000) |
140 | { |
141 | math_check_force_underflow (xm1); |
142 | if ((int) xm1 == 0) |
143 | return xm1; |
144 | } |
145 | |
146 | if (xm1 >= 0x1p113Q) |
147 | x = xm1; |
148 | else |
149 | x = xm1 + 1.0Q; |
150 | |
151 | /* log1p(-1) = -inf */ |
152 | if (x <= 0.0Q) |
153 | { |
154 | if (x == 0.0Q) |
155 | return (-1.0Q / zero); /* log1p(-1) = -inf */ |
156 | else |
157 | return (zero / (x - x)); |
158 | } |
159 | |
160 | /* Separate mantissa from exponent. */ |
161 | |
162 | /* Use frexp used so that denormal numbers will be handled properly. */ |
163 | x = frexpq (x, &e); |
164 | |
165 | /* Logarithm using log(x) = z + z^3 P(z^2)/Q(z^2), |
166 | where z = 2(x-1)/x+1). */ |
167 | if ((e > 2) || (e < -2)) |
168 | { |
169 | if (x < sqrth) |
170 | { /* 2( 2x-1 )/( 2x+1 ) */ |
171 | e -= 1; |
172 | z = x - 0.5Q; |
173 | y = 0.5Q * z + 0.5Q; |
174 | } |
175 | else |
176 | { /* 2 (x-1)/(x+1) */ |
177 | z = x - 0.5Q; |
178 | z -= 0.5Q; |
179 | y = 0.5Q * x + 0.5Q; |
180 | } |
181 | x = z / y; |
182 | z = x * x; |
183 | r = ((((R5 * z |
184 | + R4) * z |
185 | + R3) * z |
186 | + R2) * z |
187 | + R1) * z |
188 | + R0; |
189 | s = (((((z |
190 | + S5) * z |
191 | + S4) * z |
192 | + S3) * z |
193 | + S2) * z |
194 | + S1) * z |
195 | + S0; |
196 | z = x * (z * r / s); |
197 | z = z + e * C2; |
198 | z = z + x; |
199 | z = z + e * C1; |
200 | return (z); |
201 | } |
202 | |
203 | |
204 | /* Logarithm using log(1+x) = x - .5x^2 + x^3 P(x)/Q(x). */ |
205 | |
206 | if (x < sqrth) |
207 | { |
208 | e -= 1; |
209 | if (e != 0) |
210 | x = 2.0Q * x - 1.0Q; /* 2x - 1 */ |
211 | else |
212 | x = xm1; |
213 | } |
214 | else |
215 | { |
216 | if (e != 0) |
217 | x = x - 1.0Q; |
218 | else |
219 | x = xm1; |
220 | } |
221 | z = x * x; |
222 | r = (((((((((((P12 * x |
223 | + P11) * x |
224 | + P10) * x |
225 | + P9) * x |
226 | + P8) * x |
227 | + P7) * x |
228 | + P6) * x |
229 | + P5) * x |
230 | + P4) * x |
231 | + P3) * x |
232 | + P2) * x |
233 | + P1) * x |
234 | + P0; |
235 | s = (((((((((((x |
236 | + Q11) * x |
237 | + Q10) * x |
238 | + Q9) * x |
239 | + Q8) * x |
240 | + Q7) * x |
241 | + Q6) * x |
242 | + Q5) * x |
243 | + Q4) * x |
244 | + Q3) * x |
245 | + Q2) * x |
246 | + Q1) * x |
247 | + Q0; |
248 | y = x * (z * r / s); |
249 | y = y + e * C2; |
250 | z = y - 0.5Q * z; |
251 | z = z + x; |
252 | z = z + e * C1; |
253 | return (z); |
254 | } |
255 | |