Home | History | Annotate | Line # | Download | only in complex
catrigf.c revision 1.2
      1 /*	$NetBSD: catrigf.c,v 1.2 2022/04/19 20:32:16 rillig Exp $	*/
      2 /*-
      3  * Copyright (c) 2012 Stephen Montgomery-Smith <stephen (at) FreeBSD.ORG>
      4  * 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  *
     15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
     16  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     17  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     18  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
     19  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     20  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     21  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     22  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     23  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     24  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     25  * SUCH DAMAGE.
     26  */
     27 
     28 /*
     29  * The algorithm is very close to that in "Implementing the complex arcsine
     30  * and arccosine functions using exception handling" by T. E. Hull, Thomas F.
     31  * Fairgrieve, and Ping Tak Peter Tang, published in ACM Transactions on
     32  * Mathematical Software, Volume 23 Issue 3, 1997, Pages 299-335,
     33  * http://dl.acm.org/citation.cfm?id=275324.
     34  *
     35  * See catrig.c for complete comments.
     36  *
     37  * XXX comments were removed automatically, and even short ones on the right
     38  * of statements were removed (all of them), contrary to normal style.  Only
     39  * a few comments on the right of declarations remain.
     40  */
     41 
     42 #include <sys/cdefs.h>
     43 #if 0
     44 __FBSDID("$FreeBSD: head/lib/msun/src/catrigf.c 275819 2014-12-16 09:21:56Z ed $");
     45 #endif
     46 __RCSID("$NetBSD: catrigf.c,v 1.2 2022/04/19 20:32:16 rillig Exp $");
     47 
     48 #include "namespace.h"
     49 #ifdef __weak_alias
     50 __weak_alias(casinf, _casinf)
     51 #endif
     52 #ifdef __weak_alias
     53 __weak_alias(catanf, _catanf)
     54 #endif
     55 
     56 
     57 #include <complex.h>
     58 #include <float.h>
     59 
     60 #include "math.h"
     61 #include "math_private.h"
     62 
     63 #undef isinf
     64 #define isinf(x)	(fabsf(x) == INFINITY)
     65 #undef isnan
     66 #define isnan(x)	((x) != (x))
     67 #define	raise_inexact()	do { volatile float junk __unused = /*LINTED*/1 + tiny; } while (0)
     68 #undef signbit
     69 #define signbit(x)	(__builtin_signbitf(x))
     70 
     71 static const float
     72 A_crossover =		10,
     73 B_crossover =		0.6417,
     74 FOUR_SQRT_MIN =		0x1p-61,
     75 QUARTER_SQRT_MAX =	0x1p61,
     76 m_e =			2.7182818285e0,		/*  0xadf854.0p-22 */
     77 m_ln2 =			6.9314718056e-1,	/*  0xb17218.0p-24 */
     78 pio2_hi =		1.5707962513e0,		/*  0xc90fda.0p-23 */
     79 RECIP_EPSILON =		1 / FLT_EPSILON,
     80 SQRT_3_EPSILON =	5.9801995673e-4,	/*  0x9cc471.0p-34 */
     81 SQRT_6_EPSILON =	8.4572793338e-4,	/*  0xddb3d7.0p-34 */
     82 SQRT_MIN =		0x1p-63;
     83 
     84 static const volatile float
     85 pio2_lo =		7.5497899549e-8,	/*  0xa22169.0p-47 */
     86 tiny =			0x1p-100;
     87 
     88 static float complex clog_for_large_values(float complex z);
     89 
     90 static inline float
     91 f(float a, float b, float hypot_a_b)
     92 {
     93 	if (b < 0)
     94 		return ((hypot_a_b - b) / 2);
     95 	if (b == 0)
     96 		return (a / 2);
     97 	return (a * a / (hypot_a_b + b) / 2);
     98 }
     99 
    100 static inline void
    101 do_hard_work(float x, float y, float *rx, int *B_is_usable, float *B,
    102     float *sqrt_A2my2, float *new_y)
    103 {
    104 	float R, S, A;
    105 	float Am1, Amy;
    106 
    107 	R = hypotf(x, y + 1);
    108 	S = hypotf(x, y - 1);
    109 
    110 	A = (R + S) / 2;
    111 	if (A < 1)
    112 		A = 1;
    113 
    114 	if (A < A_crossover) {
    115 		if (y == 1 && x < FLT_EPSILON * FLT_EPSILON / 128) {
    116 			*rx = sqrtf(x);
    117 		} else if (x >= FLT_EPSILON * fabsf(y - 1)) {
    118 			Am1 = f(x, 1 + y, R) + f(x, 1 - y, S);
    119 			*rx = log1pf(Am1 + sqrtf(Am1 * (A + 1)));
    120 		} else if (y < 1) {
    121 			*rx = x / sqrtf((1 - y) * (1 + y));
    122 		} else {
    123 			*rx = log1pf((y - 1) + sqrtf((y - 1) * (y + 1)));
    124 		}
    125 	} else {
    126 		*rx = logf(A + sqrtf(A * A - 1));
    127 	}
    128 
    129 	*new_y = y;
    130 
    131 	if (y < FOUR_SQRT_MIN) {
    132 		*B_is_usable = 0;
    133 		*sqrt_A2my2 = A * (2 / FLT_EPSILON);
    134 		*new_y = y * (2 / FLT_EPSILON);
    135 		return;
    136 	}
    137 
    138 	*B = y / A;
    139 	*B_is_usable = 1;
    140 
    141 	if (*B > B_crossover) {
    142 		*B_is_usable = 0;
    143 		if (y == 1 && x < FLT_EPSILON / 128) {
    144 			*sqrt_A2my2 = sqrtf(x) * sqrtf((A + y) / 2);
    145 		} else if (x >= FLT_EPSILON * fabsf(y - 1)) {
    146 			Amy = f(x, y + 1, R) + f(x, y - 1, S);
    147 			*sqrt_A2my2 = sqrtf(Amy * (A + y));
    148 		} else if (y > 1) {
    149 			*sqrt_A2my2 = x * (4 / FLT_EPSILON / FLT_EPSILON) * y /
    150 			    sqrtf((y + 1) * (y - 1));
    151 			*new_y = y * (4 / FLT_EPSILON / FLT_EPSILON);
    152 		} else {
    153 			*sqrt_A2my2 = sqrtf((1 - y) * (1 + y));
    154 		}
    155 	}
    156 }
    157 
    158 float complex
    159 casinhf(float complex z)
    160 {
    161 	float x, y, ax, ay, rx, ry, B, sqrt_A2my2, new_y;
    162 	int B_is_usable;
    163 	float complex w;
    164 
    165 	x = crealf(z);
    166 	y = cimagf(z);
    167 	ax = fabsf(x);
    168 	ay = fabsf(y);
    169 
    170 	if (isnan(x) || isnan(y)) {
    171 		if (isinf(x))
    172 			return (CMPLXF(x, y + y));
    173 		if (isinf(y))
    174 			return (CMPLXF(y, x + x));
    175 		if (y == 0)
    176 			return (CMPLXF(x + x, y));
    177 		return (CMPLXF(x + 0.0L + (y + 0), x + 0.0L + (y + 0)));
    178 	}
    179 
    180 	if (ax > RECIP_EPSILON || ay > RECIP_EPSILON) {
    181 		if (signbit(x) == 0)
    182 			w = clog_for_large_values(z) + m_ln2;
    183 		else
    184 			w = clog_for_large_values(-z) + m_ln2;
    185 		return (CMPLXF(copysignf(crealf(w), x),
    186 		    copysignf(cimagf(w), y)));
    187 	}
    188 
    189 	if (x == 0 && y == 0)
    190 		return (z);
    191 
    192 	raise_inexact();
    193 
    194 	if (ax < SQRT_6_EPSILON / 4 && ay < SQRT_6_EPSILON / 4)
    195 		return (z);
    196 
    197 	do_hard_work(ax, ay, &rx, &B_is_usable, &B, &sqrt_A2my2, &new_y);
    198 	if (B_is_usable)
    199 		ry = asinf(B);
    200 	else
    201 		ry = atan2f(new_y, sqrt_A2my2);
    202 	return (CMPLXF(copysignf(rx, x), copysignf(ry, y)));
    203 }
    204 
    205 float complex
    206 casinf(float complex z)
    207 {
    208 	float complex w = casinhf(CMPLXF(cimagf(z), crealf(z)));
    209 
    210 	return (CMPLXF(cimagf(w), crealf(w)));
    211 }
    212 
    213 float complex
    214 cacosf(float complex z)
    215 {
    216 	float x, y, ax, ay, rx, ry, B, sqrt_A2mx2, new_x;
    217 	int sx, sy;
    218 	int B_is_usable;
    219 	float complex w;
    220 
    221 	x = crealf(z);
    222 	y = cimagf(z);
    223 	sx = signbit(x);
    224 	sy = signbit(y);
    225 	ax = fabsf(x);
    226 	ay = fabsf(y);
    227 
    228 	if (isnan(x) || isnan(y)) {
    229 		if (isinf(x))
    230 			return (CMPLXF(y + y, -INFINITY));
    231 		if (isinf(y))
    232 			return (CMPLXF(x + x, -y));
    233 		if (x == 0)
    234 			return (CMPLXF(pio2_hi + pio2_lo, y + y));
    235 		return (CMPLXF(x + 0.0L + (y + 0), x + 0.0L + (y + 0)));
    236 	}
    237 
    238 	if (ax > RECIP_EPSILON || ay > RECIP_EPSILON) {
    239 		w = clog_for_large_values(z);
    240 		rx = fabsf(cimagf(w));
    241 		ry = crealf(w) + m_ln2;
    242 		if (sy == 0)
    243 			ry = -ry;
    244 		return (CMPLXF(rx, ry));
    245 	}
    246 
    247 	if (x == 1 && y == 0)
    248 		return (CMPLXF(0, -y));
    249 
    250 	raise_inexact();
    251 
    252 	if (ax < SQRT_6_EPSILON / 4 && ay < SQRT_6_EPSILON / 4)
    253 		return (CMPLXF(pio2_hi - (x - pio2_lo), -y));
    254 
    255 	do_hard_work(ay, ax, &ry, &B_is_usable, &B, &sqrt_A2mx2, &new_x);
    256 	if (B_is_usable) {
    257 		if (sx == 0)
    258 			rx = acosf(B);
    259 		else
    260 			rx = acosf(-B);
    261 	} else {
    262 		if (sx == 0)
    263 			rx = atan2f(sqrt_A2mx2, new_x);
    264 		else
    265 			rx = atan2f(sqrt_A2mx2, -new_x);
    266 	}
    267 	if (sy == 0)
    268 		ry = -ry;
    269 	return (CMPLXF(rx, ry));
    270 }
    271 
    272 float complex
    273 cacoshf(float complex z)
    274 {
    275 	float complex w;
    276 	float rx, ry;
    277 
    278 	w = cacosf(z);
    279 	rx = crealf(w);
    280 	ry = cimagf(w);
    281 	if (isnan(rx) && isnan(ry))
    282 		return (CMPLXF(ry, rx));
    283 	if (isnan(rx))
    284 		return (CMPLXF(fabsf(ry), rx));
    285 	if (isnan(ry))
    286 		return (CMPLXF(ry, ry));
    287 	return (CMPLXF(fabsf(ry), copysignf(rx, cimagf(z))));
    288 }
    289 
    290 static float complex
    291 clog_for_large_values(float complex z)
    292 {
    293 	float x, y;
    294 	float ax, ay, t;
    295 
    296 	x = crealf(z);
    297 	y = cimagf(z);
    298 	ax = fabsf(x);
    299 	ay = fabsf(y);
    300 	if (ax < ay) {
    301 		t = ax;
    302 		ax = ay;
    303 		ay = t;
    304 	}
    305 
    306 	if (ax > FLT_MAX / 2)
    307 		return (CMPLXF(logf(hypotf(x / m_e, y / m_e)) + 1,
    308 		    atan2f(y, x)));
    309 
    310 	if (ax > QUARTER_SQRT_MAX || ay < SQRT_MIN)
    311 		return (CMPLXF(logf(hypotf(x, y)), atan2f(y, x)));
    312 
    313 	return (CMPLXF(logf(ax * ax + ay * ay) / 2, atan2f(y, x)));
    314 }
    315 
    316 static inline float
    317 sum_squares(float x, float y)
    318 {
    319 
    320 	if (y < SQRT_MIN)
    321 		return (x * x);
    322 
    323 	return (x * x + y * y);
    324 }
    325 
    326 static inline float
    327 real_part_reciprocal(float x, float y)
    328 {
    329 	float scale;
    330 	uint32_t hx, hy;
    331 	int32_t ix, iy;
    332 
    333 	GET_FLOAT_WORD(hx, x);
    334 	ix = hx & 0x7f800000;
    335 	GET_FLOAT_WORD(hy, y);
    336 	iy = hy & 0x7f800000;
    337 #define	BIAS	(FLT_MAX_EXP - 1)
    338 #define	CUTOFF	(FLT_MANT_DIG / 2 + 1)
    339 	if (ix - iy >= CUTOFF << 23 || isinf(x))
    340 		return (1 / x);
    341 	if (iy - ix >= CUTOFF << 23)
    342 		return (x / y / y);
    343 	if (ix <= (BIAS + FLT_MAX_EXP / 2 - CUTOFF) << 23)
    344 		return (x / (x * x + y * y));
    345 	SET_FLOAT_WORD(scale, 0x7f800000 - ix);
    346 	x *= scale;
    347 	y *= scale;
    348 	return (x / (x * x + y * y) * scale);
    349 }
    350 
    351 float complex
    352 catanhf(float complex z)
    353 {
    354 	float x, y, ax, ay, rx, ry;
    355 
    356 	x = crealf(z);
    357 	y = cimagf(z);
    358 	ax = fabsf(x);
    359 	ay = fabsf(y);
    360 
    361 	if (y == 0 && ax <= 1)
    362 		return (CMPLXF(atanhf(x), y));
    363 
    364 	if (x == 0)
    365 		return (CMPLXF(x, atanf(y)));
    366 
    367 	if (isnan(x) || isnan(y)) {
    368 		if (isinf(x))
    369 			return (CMPLXF(copysignf(0, x), y + y));
    370 		if (isinf(y))
    371 			return (CMPLXF(copysignf(0, x),
    372 			    copysignf(pio2_hi + pio2_lo, y)));
    373 		return (CMPLXF(x + 0.0L + (y + 0), x + 0.0L + (y + 0)));
    374 	}
    375 
    376 	if (ax > RECIP_EPSILON || ay > RECIP_EPSILON)
    377 		return (CMPLXF(real_part_reciprocal(x, y),
    378 		    copysignf(pio2_hi + pio2_lo, y)));
    379 
    380 	if (ax < SQRT_3_EPSILON / 2 && ay < SQRT_3_EPSILON / 2) {
    381 		raise_inexact();
    382 		return (z);
    383 	}
    384 
    385 	if (ax == 1 && ay < FLT_EPSILON)
    386 		rx = (m_ln2 - logf(ay)) / 2;
    387 	else
    388 		rx = log1pf(4 * ax / sum_squares(ax - 1, ay)) / 4;
    389 
    390 	if (ax == 1)
    391 		ry = atan2f(2, -ay) / 2;
    392 	else if (ay < FLT_EPSILON)
    393 		ry = atan2f(2 * ay, (1 - ax) * (1 + ax)) / 2;
    394 	else
    395 		ry = atan2f(2 * ay, (1 - ax) * (1 + ax) - ay * ay) / 2;
    396 
    397 	return (CMPLXF(copysignf(rx, x), copysignf(ry, y)));
    398 }
    399 
    400 float complex
    401 catanf(float complex z)
    402 {
    403 	float complex w = catanhf(CMPLXF(cimagf(z), crealf(z)));
    404 
    405 	return (CMPLXF(cimagf(w), crealf(w)));
    406 }
    407