xref: /netbsd-src/lib/libm/noieee_src/n_log.c (revision 23c8222edbfb0f0932d88a8351d3a0cf817dfb9e)
1 /*      $NetBSD: n_log.c,v 1.6 2003/08/07 16:44:51 agc Exp $ */
2 /*
3  * Copyright (c) 1992, 1993
4  *	The Regents of the University of California.  All rights reserved.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  * 3. Neither the name of the University nor the names of its contributors
15  *    may be used to endorse or promote products derived from this software
16  *    without specific prior written permission.
17  *
18  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
19  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
22  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28  * SUCH DAMAGE.
29  */
30 
31 #ifndef lint
32 #if 0
33 static char sccsid[] = "@(#)log.c	8.2 (Berkeley) 11/30/93";
34 #endif
35 #endif /* not lint */
36 
37 #include <math.h>
38 #include <errno.h>
39 
40 #include "mathimpl.h"
41 
42 /* Table-driven natural logarithm.
43  *
44  * This code was derived, with minor modifications, from:
45  *	Peter Tang, "Table-Driven Implementation of the
46  *	Logarithm in IEEE Floating-Point arithmetic." ACM Trans.
47  *	Math Software, vol 16. no 4, pp 378-400, Dec 1990).
48  *
49  * Calculates log(2^m*F*(1+f/F)), |f/j| <= 1/256,
50  * where F = j/128 for j an integer in [0, 128].
51  *
52  * log(2^m) = log2_hi*m + log2_tail*m
53  * since m is an integer, the dominant term is exact.
54  * m has at most 10 digits (for subnormal numbers),
55  * and log2_hi has 11 trailing zero bits.
56  *
57  * log(F) = logF_hi[j] + logF_lo[j] is in tabular form in log_table.h
58  * logF_hi[] + 512 is exact.
59  *
60  * log(1+f/F) = 2*f/(2*F + f) + 1/12 * (2*f/(2*F + f))**3 + ...
61  * the leading term is calculated to extra precision in two
62  * parts, the larger of which adds exactly to the dominant
63  * m and F terms.
64  * There are two cases:
65  *	1. when m, j are non-zero (m | j), use absolute
66  *	   precision for the leading term.
67  *	2. when m = j = 0, |1-x| < 1/256, and log(x) ~= (x-1).
68  *	   In this case, use a relative precision of 24 bits.
69  * (This is done differently in the original paper)
70  *
71  * Special cases:
72  *	0	return signalling -Inf
73  *	neg	return signalling NaN
74  *	+Inf	return +Inf
75 */
76 
77 #if defined(__vax__) || defined(tahoe)
78 #define _IEEE		0
79 #define TRUNC(x)	x = (double) (float) (x)
80 #else
81 #define _IEEE		1
82 #define endian		(((*(int *) &one)) ? 1 : 0)
83 #define TRUNC(x)	*(((int *) &x) + endian) &= 0xf8000000
84 #define infnan(x)	0.0
85 #endif
86 
87 #define N 128
88 
89 /* Table of log(Fj) = logF_head[j] + logF_tail[j], for Fj = 1+j/128.
90  * Used for generation of extend precision logarithms.
91  * The constant 35184372088832 is 2^45, so the divide is exact.
92  * It ensures correct reading of logF_head, even for inaccurate
93  * decimal-to-binary conversion routines.  (Everybody gets the
94  * right answer for integers less than 2^53.)
95  * Values for log(F) were generated using error < 10^-57 absolute
96  * with the bc -l package.
97 */
98 static const double	A1 = 	  .08333333333333178827;
99 static const double	A2 = 	  .01250000000377174923;
100 static const double	A3 =	 .002232139987919447809;
101 static const double	A4 =	.0004348877777076145742;
102 
103 static const double logF_head[N+1] = {
104 	0.,
105 	.007782140442060381246,
106 	.015504186535963526694,
107 	.023167059281547608406,
108 	.030771658666765233647,
109 	.038318864302141264488,
110 	.045809536031242714670,
111 	.053244514518837604555,
112 	.060624621816486978786,
113 	.067950661908525944454,
114 	.075223421237524235039,
115 	.082443669210988446138,
116 	.089612158689760690322,
117 	.096729626458454731618,
118 	.103796793681567578460,
119 	.110814366340264314203,
120 	.117783035656430001836,
121 	.124703478501032805070,
122 	.131576357788617315236,
123 	.138402322859292326029,
124 	.145182009844575077295,
125 	.151916042025732167530,
126 	.158605030176659056451,
127 	.165249572895390883786,
128 	.171850256926518341060,
129 	.178407657472689606947,
130 	.184922338493834104156,
131 	.191394852999565046047,
132 	.197825743329758552135,
133 	.204215541428766300668,
134 	.210564769107350002741,
135 	.216873938300523150246,
136 	.223143551314024080056,
137 	.229374101064877322642,
138 	.235566071312860003672,
139 	.241719936886966024758,
140 	.247836163904594286577,
141 	.253915209980732470285,
142 	.259957524436686071567,
143 	.265963548496984003577,
144 	.271933715484010463114,
145 	.277868451003087102435,
146 	.283768173130738432519,
147 	.289633292582948342896,
148 	.295464212893421063199,
149 	.301261330578199704177,
150 	.307025035294827830512,
151 	.312755710004239517729,
152 	.318453731118097493890,
153 	.324119468654316733591,
154 	.329753286372579168528,
155 	.335355541920762334484,
156 	.340926586970454081892,
157 	.346466767346100823488,
158 	.351976423156884266063,
159 	.357455888922231679316,
160 	.362905493689140712376,
161 	.368325561158599157352,
162 	.373716409793814818840,
163 	.379078352934811846353,
164 	.384411698910298582632,
165 	.389716751140440464951,
166 	.394993808240542421117,
167 	.400243164127459749579,
168 	.405465108107819105498,
169 	.410659924985338875558,
170 	.415827895143593195825,
171 	.420969294644237379543,
172 	.426084395310681429691,
173 	.431173464818130014464,
174 	.436236766774527495726,
175 	.441274560805140936281,
176 	.446287102628048160113,
177 	.451274644139630254358,
178 	.456237433481874177232,
179 	.461175715122408291790,
180 	.466089729924533457960,
181 	.470979715219073113985,
182 	.475845904869856894947,
183 	.480688529345570714212,
184 	.485507815781602403149,
185 	.490303988045525329653,
186 	.495077266798034543171,
187 	.499827869556611403822,
188 	.504556010751912253908,
189 	.509261901790523552335,
190 	.513945751101346104405,
191 	.518607764208354637958,
192 	.523248143765158602036,
193 	.527867089620485785417,
194 	.532464798869114019908,
195 	.537041465897345915436,
196 	.541597282432121573947,
197 	.546132437597407260909,
198 	.550647117952394182793,
199 	.555141507540611200965,
200 	.559615787935399566777,
201 	.564070138285387656651,
202 	.568504735352689749561,
203 	.572919753562018740922,
204 	.577315365035246941260,
205 	.581691739635061821900,
206 	.586049045003164792433,
207 	.590387446602107957005,
208 	.594707107746216934174,
209 	.599008189645246602594,
210 	.603290851438941899687,
211 	.607555250224322662688,
212 	.611801541106615331955,
213 	.616029877215623855590,
214 	.620240409751204424537,
215 	.624433288012369303032,
216 	.628608659422752680256,
217 	.632766669570628437213,
218 	.636907462236194987781,
219 	.641031179420679109171,
220 	.645137961373620782978,
221 	.649227946625615004450,
222 	.653301272011958644725,
223 	.657358072709030238911,
224 	.661398482245203922502,
225 	.665422632544505177065,
226 	.669430653942981734871,
227 	.673422675212350441142,
228 	.677398823590920073911,
229 	.681359224807238206267,
230 	.685304003098281100392,
231 	.689233281238557538017,
232 	.693147180560117703862
233 };
234 
235 static const double logF_tail[N+1] = {
236 	0.,
237 	-.00000000000000543229938420049,
238 	 .00000000000000172745674997061,
239 	-.00000000000001323017818229233,
240 	-.00000000000001154527628289872,
241 	-.00000000000000466529469958300,
242 	 .00000000000005148849572685810,
243 	-.00000000000002532168943117445,
244 	-.00000000000005213620639136504,
245 	-.00000000000001819506003016881,
246 	 .00000000000006329065958724544,
247 	 .00000000000008614512936087814,
248 	-.00000000000007355770219435028,
249 	 .00000000000009638067658552277,
250 	 .00000000000007598636597194141,
251 	 .00000000000002579999128306990,
252 	-.00000000000004654729747598444,
253 	-.00000000000007556920687451336,
254 	 .00000000000010195735223708472,
255 	-.00000000000017319034406422306,
256 	-.00000000000007718001336828098,
257 	 .00000000000010980754099855238,
258 	-.00000000000002047235780046195,
259 	-.00000000000008372091099235912,
260 	 .00000000000014088127937111135,
261 	 .00000000000012869017157588257,
262 	 .00000000000017788850778198106,
263 	 .00000000000006440856150696891,
264 	 .00000000000016132822667240822,
265 	-.00000000000007540916511956188,
266 	-.00000000000000036507188831790,
267 	 .00000000000009120937249914984,
268 	 .00000000000018567570959796010,
269 	-.00000000000003149265065191483,
270 	-.00000000000009309459495196889,
271 	 .00000000000017914338601329117,
272 	-.00000000000001302979717330866,
273 	 .00000000000023097385217586939,
274 	 .00000000000023999540484211737,
275 	 .00000000000015393776174455408,
276 	-.00000000000036870428315837678,
277 	 .00000000000036920375082080089,
278 	-.00000000000009383417223663699,
279 	 .00000000000009433398189512690,
280 	 .00000000000041481318704258568,
281 	-.00000000000003792316480209314,
282 	 .00000000000008403156304792424,
283 	-.00000000000034262934348285429,
284 	 .00000000000043712191957429145,
285 	-.00000000000010475750058776541,
286 	-.00000000000011118671389559323,
287 	 .00000000000037549577257259853,
288 	 .00000000000013912841212197565,
289 	 .00000000000010775743037572640,
290 	 .00000000000029391859187648000,
291 	-.00000000000042790509060060774,
292 	 .00000000000022774076114039555,
293 	 .00000000000010849569622967912,
294 	-.00000000000023073801945705758,
295 	 .00000000000015761203773969435,
296 	 .00000000000003345710269544082,
297 	-.00000000000041525158063436123,
298 	 .00000000000032655698896907146,
299 	-.00000000000044704265010452446,
300 	 .00000000000034527647952039772,
301 	-.00000000000007048962392109746,
302 	 .00000000000011776978751369214,
303 	-.00000000000010774341461609578,
304 	 .00000000000021863343293215910,
305 	 .00000000000024132639491333131,
306 	 .00000000000039057462209830700,
307 	-.00000000000026570679203560751,
308 	 .00000000000037135141919592021,
309 	-.00000000000017166921336082431,
310 	-.00000000000028658285157914353,
311 	-.00000000000023812542263446809,
312 	 .00000000000006576659768580062,
313 	-.00000000000028210143846181267,
314 	 .00000000000010701931762114254,
315 	 .00000000000018119346366441110,
316 	 .00000000000009840465278232627,
317 	-.00000000000033149150282752542,
318 	-.00000000000018302857356041668,
319 	-.00000000000016207400156744949,
320 	 .00000000000048303314949553201,
321 	-.00000000000071560553172382115,
322 	 .00000000000088821239518571855,
323 	-.00000000000030900580513238244,
324 	-.00000000000061076551972851496,
325 	 .00000000000035659969663347830,
326 	 .00000000000035782396591276383,
327 	-.00000000000046226087001544578,
328 	 .00000000000062279762917225156,
329 	 .00000000000072838947272065741,
330 	 .00000000000026809646615211673,
331 	-.00000000000010960825046059278,
332 	 .00000000000002311949383800537,
333 	-.00000000000058469058005299247,
334 	-.00000000000002103748251144494,
335 	-.00000000000023323182945587408,
336 	-.00000000000042333694288141916,
337 	-.00000000000043933937969737844,
338 	 .00000000000041341647073835565,
339 	 .00000000000006841763641591466,
340 	 .00000000000047585534004430641,
341 	 .00000000000083679678674757695,
342 	-.00000000000085763734646658640,
343 	 .00000000000021913281229340092,
344 	-.00000000000062242842536431148,
345 	-.00000000000010983594325438430,
346 	 .00000000000065310431377633651,
347 	-.00000000000047580199021710769,
348 	-.00000000000037854251265457040,
349 	 .00000000000040939233218678664,
350 	 .00000000000087424383914858291,
351 	 .00000000000025218188456842882,
352 	-.00000000000003608131360422557,
353 	-.00000000000050518555924280902,
354 	 .00000000000078699403323355317,
355 	-.00000000000067020876961949060,
356 	 .00000000000016108575753932458,
357 	 .00000000000058527188436251509,
358 	-.00000000000035246757297904791,
359 	-.00000000000018372084495629058,
360 	 .00000000000088606689813494916,
361 	 .00000000000066486268071468700,
362 	 .00000000000063831615170646519,
363 	 .00000000000025144230728376072,
364 	-.00000000000017239444525614834
365 };
366 
367 double
368 log(double x)
369 {
370 	int m, j;
371 	double F, f, g, q, u, u2, v, zero = 0.0, one = 1.0;
372 	volatile double u1;
373 
374 	/* Catch special cases */
375 	if (x <= 0) {
376 		if (_IEEE && x == zero)	/* log(0) = -Inf */
377 			return (-one/zero);
378 		else if (_IEEE)		/* log(neg) = NaN */
379 			return (zero/zero);
380 		else if (x == zero)	/* NOT REACHED IF _IEEE */
381 			return (infnan(-ERANGE));
382 		else
383 			return (infnan(EDOM));
384 	} else if (!finite(x)) {
385 		if (_IEEE)		/* x = NaN, Inf */
386 			return (x+x);
387 		else
388 			return (infnan(ERANGE));
389 	}
390 
391 	/* Argument reduction: 1 <= g < 2; x/2^m = g;	*/
392 	/* y = F*(1 + f/F) for |f| <= 2^-8		*/
393 
394 	m = logb(x);
395 	g = ldexp(x, -m);
396 	if (_IEEE && m == -1022) {
397 		j = logb(g), m += j;
398 		g = ldexp(g, -j);
399 	}
400 	j = N*(g-1) + .5;
401 	F = (1.0/N) * j + 1;	/* F*128 is an integer in [128, 512] */
402 	f = g - F;
403 
404 	/* Approximate expansion for log(1+f/F) ~= u + q */
405 	g = 1/(2*F+f);
406 	u = 2*f*g;
407 	v = u*u;
408 	q = u*v*(A1 + v*(A2 + v*(A3 + v*A4)));
409 
410     /* case 1: u1 = u rounded to 2^-43 absolute.  Since u < 2^-8,
411      * 	       u1 has at most 35 bits, and F*u1 is exact, as F has < 8 bits.
412      *         It also adds exactly to |m*log2_hi + log_F_head[j] | < 750
413     */
414 	if (m | j)
415 		u1 = u + 513, u1 -= 513;
416 
417     /* case 2:	|1-x| < 1/256. The m- and j- dependent terms are zero;
418      * 		u1 = u to 24 bits.
419     */
420 	else
421 		u1 = u, TRUNC(u1);
422 	u2 = (2.0*(f - F*u1) - u1*f) * g;
423 			/* u1 + u2 = 2f/(2F+f) to extra precision.	*/
424 
425 	/* log(x) = log(2^m*F*(1+f/F)) =				*/
426 	/* (m*log2_hi+logF_head[j]+u1) + (m*log2_lo+logF_tail[j]+q);	*/
427 	/* (exact) + (tiny)						*/
428 
429 	u1 += m*logF_head[N] + logF_head[j];		/* exact */
430 	u2 = (u2 + logF_tail[j]) + q;			/* tiny */
431 	u2 += logF_tail[N]*m;
432 	return (u1 + u2);
433 }
434 
435 /*
436  * Extra precision variant, returning struct {double a, b;};
437  * log(x) = a+b to 63 bits, with a is rounded to 26 bits.
438  */
439 struct Double
440 __log__D(double x)
441 {
442 	int m, j;
443 	double F, f, g, q, u, v, u2;
444 	volatile double u1;
445 	struct Double r;
446 
447 	/* Argument reduction: 1 <= g < 2; x/2^m = g;	*/
448 	/* y = F*(1 + f/F) for |f| <= 2^-8		*/
449 
450 	m = logb(x);
451 	g = ldexp(x, -m);
452 	if (_IEEE && m == -1022) {
453 		j = logb(g), m += j;
454 		g = ldexp(g, -j);
455 	}
456 	j = N*(g-1) + .5;
457 	F = (1.0/N) * j + 1;
458 	f = g - F;
459 
460 	g = 1/(2*F+f);
461 	u = 2*f*g;
462 	v = u*u;
463 	q = u*v*(A1 + v*(A2 + v*(A3 + v*A4)));
464 	if (m | j)
465 		u1 = u + 513, u1 -= 513;
466 	else
467 		u1 = u, TRUNC(u1);
468 	u2 = (2.0*(f - F*u1) - u1*f) * g;
469 
470 	u1 += m*logF_head[N] + logF_head[j];
471 
472 	u2 +=  logF_tail[j]; u2 += q;
473 	u2 += logF_tail[N]*m;
474 	r.a = u1 + u2;			/* Only difference is here */
475 	TRUNC(r.a);
476 	r.b = (u1 - r.a) + u2;
477 	return (r);
478 }
479