1 /*							log1pl.c
2  *
3  *      Relative error logarithm
4  *	Natural logarithm of 1+x, 128-bit long double precision
5  *
6  *
7  *
8  * SYNOPSIS:
9  *
10  * long double x, y, log1pl();
11  *
12  * y = log1pl( 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, see
53     <https://www.gnu.org/licenses/>.  */
54 
55 
56 #include <float.h>
57 #include <math.h>
58 #include <math_private.h>
59 #include <math-underflow.h>
60 
61 /* Coefficients for log(1+x) = x - x^2 / 2 + x^3 P(x)/Q(x)
62  * 1/sqrt(2) <= 1+x < sqrt(2)
63  * Theoretical peak relative error = 5.3e-37,
64  * relative peak error spread = 2.3e-14
65  */
66 static const _Float128
67   P12 = L(1.538612243596254322971797716843006400388E-6),
68   P11 = L(4.998469661968096229986658302195402690910E-1),
69   P10 = L(2.321125933898420063925789532045674660756E1),
70   P9 = L(4.114517881637811823002128927449878962058E2),
71   P8 = L(3.824952356185897735160588078446136783779E3),
72   P7 = L(2.128857716871515081352991964243375186031E4),
73   P6 = L(7.594356839258970405033155585486712125861E4),
74   P5 = L(1.797628303815655343403735250238293741397E5),
75   P4 = L(2.854829159639697837788887080758954924001E5),
76   P3 = L(3.007007295140399532324943111654767187848E5),
77   P2 = L(2.014652742082537582487669938141683759923E5),
78   P1 = L(7.771154681358524243729929227226708890930E4),
79   P0 = L(1.313572404063446165910279910527789794488E4),
80   /* Q12 = 1.000000000000000000000000000000000000000E0L, */
81   Q11 = L(4.839208193348159620282142911143429644326E1),
82   Q10 = L(9.104928120962988414618126155557301584078E2),
83   Q9 = L(9.147150349299596453976674231612674085381E3),
84   Q8 = L(5.605842085972455027590989944010492125825E4),
85   Q7 = L(2.248234257620569139969141618556349415120E5),
86   Q6 = L(6.132189329546557743179177159925690841200E5),
87   Q5 = L(1.158019977462989115839826904108208787040E6),
88   Q4 = L(1.514882452993549494932585972882995548426E6),
89   Q3 = L(1.347518538384329112529391120390701166528E6),
90   Q2 = L(7.777690340007566932935753241556479363645E5),
91   Q1 = L(2.626900195321832660448791748036714883242E5),
92   Q0 = L(3.940717212190338497730839731583397586124E4);
93 
94 /* Coefficients for log(x) = z + z^3 P(z^2)/Q(z^2),
95  * where z = 2(x-1)/(x+1)
96  * 1/sqrt(2) <= x < sqrt(2)
97  * Theoretical peak relative error = 1.1e-35,
98  * relative peak error spread 1.1e-9
99  */
100 static const _Float128
101   R5 = L(-8.828896441624934385266096344596648080902E-1),
102   R4 = L(8.057002716646055371965756206836056074715E1),
103   R3 = L(-2.024301798136027039250415126250455056397E3),
104   R2 = L(2.048819892795278657810231591630928516206E4),
105   R1 = L(-8.977257995689735303686582344659576526998E4),
106   R0 = L(1.418134209872192732479751274970992665513E5),
107   /* S6 = 1.000000000000000000000000000000000000000E0L, */
108   S5 = L(-1.186359407982897997337150403816839480438E2),
109   S4 = L(3.998526750980007367835804959888064681098E3),
110   S3 = L(-5.748542087379434595104154610899551484314E4),
111   S2 = L(4.001557694070773974936904547424676279307E5),
112   S1 = L(-1.332535117259762928288745111081235577029E6),
113   S0 = L(1.701761051846631278975701529965589676574E6);
114 
115 /* C1 + C2 = ln 2 */
116 static const _Float128 C1 = L(6.93145751953125E-1);
117 static const _Float128 C2 = L(1.428606820309417232121458176568075500134E-6);
118 
119 static const _Float128 sqrth = L(0.7071067811865475244008443621048490392848);
120 /* ln (2^16384 * (1 - 2^-113)) */
121 static const _Float128 zero = 0;
122 
123 _Float128
__log1pl(_Float128 xm1)124 __log1pl (_Float128 xm1)
125 {
126   _Float128 x, y, z, r, s;
127   ieee854_long_double_shape_type u;
128   int32_t hx;
129   int e;
130 
131   /* Test for NaN or infinity input. */
132   u.value = xm1;
133   hx = u.parts32.w0;
134   if ((hx & 0x7fffffff) >= 0x7fff0000)
135     return xm1 + fabsl (xm1);
136 
137   /* log1p(+- 0) = +- 0.  */
138   if (((hx & 0x7fffffff) == 0)
139       && (u.parts32.w1 | u.parts32.w2 | u.parts32.w3) == 0)
140     return xm1;
141 
142   if ((hx & 0x7fffffff) < 0x3f8e0000)
143     {
144       math_check_force_underflow (xm1);
145       if ((int) xm1 == 0)
146 	return xm1;
147     }
148 
149   if (xm1 >= L(0x1p113))
150     x = xm1;
151   else
152     x = xm1 + 1;
153 
154   /* log1p(-1) = -inf */
155   if (x <= 0)
156     {
157       if (x == 0)
158 	return (-1 / zero);  /* log1p(-1) = -inf */
159       else
160 	return (zero / (x - x));
161     }
162 
163   /* Separate mantissa from exponent.  */
164 
165   /* Use frexp used so that denormal numbers will be handled properly.  */
166   x = __frexpl (x, &e);
167 
168   /* Logarithm using log(x) = z + z^3 P(z^2)/Q(z^2),
169      where z = 2(x-1)/x+1).  */
170   if ((e > 2) || (e < -2))
171     {
172       if (x < sqrth)
173 	{			/* 2( 2x-1 )/( 2x+1 ) */
174 	  e -= 1;
175 	  z = x - L(0.5);
176 	  y = L(0.5) * z + L(0.5);
177 	}
178       else
179 	{			/*  2 (x-1)/(x+1)   */
180 	  z = x - L(0.5);
181 	  z -= L(0.5);
182 	  y = L(0.5) * x + L(0.5);
183 	}
184       x = z / y;
185       z = x * x;
186       r = ((((R5 * z
187 	      + R4) * z
188 	     + R3) * z
189 	    + R2) * z
190 	   + R1) * z
191 	+ R0;
192       s = (((((z
193 	       + S5) * z
194 	      + S4) * z
195 	     + S3) * z
196 	    + S2) * z
197 	   + S1) * z
198 	+ S0;
199       z = x * (z * r / s);
200       z = z + e * C2;
201       z = z + x;
202       z = z + e * C1;
203       return (z);
204     }
205 
206 
207   /* Logarithm using log(1+x) = x - .5x^2 + x^3 P(x)/Q(x). */
208 
209   if (x < sqrth)
210     {
211       e -= 1;
212       if (e != 0)
213 	x = 2 * x - 1;	/*  2x - 1  */
214       else
215 	x = xm1;
216     }
217   else
218     {
219       if (e != 0)
220 	x = x - 1;
221       else
222 	x = xm1;
223     }
224   z = x * x;
225   r = (((((((((((P12 * x
226 		 + P11) * x
227 		+ P10) * x
228 	       + P9) * x
229 	      + P8) * x
230 	     + P7) * x
231 	    + P6) * x
232 	   + P5) * x
233 	  + P4) * x
234 	 + P3) * x
235 	+ P2) * x
236        + P1) * x
237     + P0;
238   s = (((((((((((x
239 		 + Q11) * x
240 		+ Q10) * x
241 	       + Q9) * x
242 	      + Q8) * x
243 	     + Q7) * x
244 	    + Q6) * x
245 	   + Q5) * x
246 	  + Q4) * x
247 	 + Q3) * x
248 	+ Q2) * x
249        + Q1) * x
250     + Q0;
251   y = x * (z * r / s);
252   y = y + e * C2;
253   z = y - L(0.5) * z;
254   z = z + x;
255   z = z + e * C1;
256   return (z);
257 }
258