Home | History | Annotate | Download | only in src
      1 /*-
      2  * Copyright (c) 2012 Stephen Montgomery-Smith <stephen (at) FreeBSD.ORG>
      3  * All rights reserved.
      4  *
      5  * Redistribution and use in source and binary forms, with or without
      6  * modification, are permitted provided that the following conditions
      7  * are met:
      8  * 1. Redistributions of source code must retain the above copyright
      9  *    notice, this list of conditions and the following disclaimer.
     10  * 2. Redistributions in binary form must reproduce the above copyright
     11  *    notice, this list of conditions and the following disclaimer in the
     12  *    documentation and/or other materials provided with the distribution.
     13  *
     14  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
     15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     17  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
     18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     24  * SUCH DAMAGE.
     25  */
     26 
     27 /*
     28  * The algorithm is very close to that in "Implementing the complex arcsine
     29  * and arccosine functions using exception handling" by T. E. Hull, Thomas F.
     30  * Fairgrieve, and Ping Tak Peter Tang, published in ACM Transactions on
     31  * Mathematical Software, Volume 23 Issue 3, 1997, Pages 299-335,
     32  * http://dl.acm.org/citation.cfm?id=275324.
     33  *
     34  * See catrig.c for complete comments.
     35  *
     36  * XXX comments were removed automatically, and even short ones on the right
     37  * of statements were removed (all of them), contrary to normal style.  Only
     38  * a few comments on the right of declarations remain.
     39  */
     40 
     41 #include <sys/cdefs.h>
     42 __FBSDID("$FreeBSD: head/lib/msun/src/catrigf.c 275819 2014-12-16 09:21:56Z ed $");
     43 
     44 #include <complex.h>
     45 #include <float.h>
     46 
     47 #include "math.h"
     48 #include "math_private.h"
     49 
     50 #undef isinf
     51 #define isinf(x)	(fabsf(x) == INFINITY)
     52 #undef isnan
     53 #define isnan(x)	((x) != (x))
     54 #define	raise_inexact()	do { volatile float junk = 1 + tiny; } while(0)
     55 #undef signbit
     56 #define signbit(x)	(__builtin_signbitf(x))
     57 
     58 static const float
     59 A_crossover =		10,
     60 B_crossover =		0.6417,
     61 FOUR_SQRT_MIN =		0x1p-61,
     62 QUARTER_SQRT_MAX =	0x1p61,
     63 m_e =			2.7182818285e0,		/*  0xadf854.0p-22 */
     64 m_ln2 =			6.9314718056e-1,	/*  0xb17218.0p-24 */
     65 pio2_hi =		1.5707962513e0,		/*  0xc90fda.0p-23 */
     66 RECIP_EPSILON =		1 / FLT_EPSILON,
     67 SQRT_3_EPSILON =	5.9801995673e-4,	/*  0x9cc471.0p-34 */
     68 SQRT_6_EPSILON =	8.4572793338e-4,	/*  0xddb3d7.0p-34 */
     69 SQRT_MIN =		0x1p-63;
     70 
     71 static const volatile float
     72 pio2_lo =		7.5497899549e-8,	/*  0xa22169.0p-47 */
     73 tiny =			0x1p-100;
     74 
     75 static float complex clog_for_large_values(float complex z);
     76 
     77 static inline float
     78 f(float a, float b, float hypot_a_b)
     79 {
     80 	if (b < 0)
     81 		return ((hypot_a_b - b) / 2);
     82 	if (b == 0)
     83 		return (a / 2);
     84 	return (a * a / (hypot_a_b + b) / 2);
     85 }
     86 
     87 static inline void
     88 do_hard_work(float x, float y, float *rx, int *B_is_usable, float *B,
     89     float *sqrt_A2my2, float *new_y)
     90 {
     91 	float R, S, A;
     92 	float Am1, Amy;
     93 
     94 	R = hypotf(x, y + 1);
     95 	S = hypotf(x, y - 1);
     96 
     97 	A = (R + S) / 2;
     98 	if (A < 1)
     99 		A = 1;
    100 
    101 	if (A < A_crossover) {
    102 		if (y == 1 && x < FLT_EPSILON * FLT_EPSILON / 128) {
    103 			*rx = sqrtf(x);
    104 		} else if (x >= FLT_EPSILON * fabsf(y - 1)) {
    105 			Am1 = f(x, 1 + y, R) + f(x, 1 - y, S);
    106 			*rx = log1pf(Am1 + sqrtf(Am1 * (A + 1)));
    107 		} else if (y < 1) {
    108 			*rx = x / sqrtf((1 - y) * (1 + y));
    109 		} else {
    110 			*rx = log1pf((y - 1) + sqrtf((y - 1) * (y + 1)));
    111 		}
    112 	} else {
    113 		*rx = logf(A + sqrtf(A * A - 1));
    114 	}
    115 
    116 	*new_y = y;
    117 
    118 	if (y < FOUR_SQRT_MIN) {
    119 		*B_is_usable = 0;
    120 		*sqrt_A2my2 = A * (2 / FLT_EPSILON);
    121 		*new_y = y * (2 / FLT_EPSILON);
    122 		return;
    123 	}
    124 
    125 	*B = y / A;
    126 	*B_is_usable = 1;
    127 
    128 	if (*B > B_crossover) {
    129 		*B_is_usable = 0;
    130 		if (y == 1 && x < FLT_EPSILON / 128) {
    131 			*sqrt_A2my2 = sqrtf(x) * sqrtf((A + y) / 2);
    132 		} else if (x >= FLT_EPSILON * fabsf(y - 1)) {
    133 			Amy = f(x, y + 1, R) + f(x, y - 1, S);
    134 			*sqrt_A2my2 = sqrtf(Amy * (A + y));
    135 		} else if (y > 1) {
    136 			*sqrt_A2my2 = x * (4 / FLT_EPSILON / FLT_EPSILON) * y /
    137 			    sqrtf((y + 1) * (y - 1));
    138 			*new_y = y * (4 / FLT_EPSILON / FLT_EPSILON);
    139 		} else {
    140 			*sqrt_A2my2 = sqrtf((1 - y) * (1 + y));
    141 		}
    142 	}
    143 }
    144 
    145 float complex
    146 casinhf(float complex z)
    147 {
    148 	float x, y, ax, ay, rx, ry, B, sqrt_A2my2, new_y;
    149 	int B_is_usable;
    150 	float complex w;
    151 
    152 	x = crealf(z);
    153 	y = cimagf(z);
    154 	ax = fabsf(x);
    155 	ay = fabsf(y);
    156 
    157 	if (isnan(x) || isnan(y)) {
    158 		if (isinf(x))
    159 			return (CMPLXF(x, y + y));
    160 		if (isinf(y))
    161 			return (CMPLXF(y, x + x));
    162 		if (y == 0)
    163 			return (CMPLXF(x + x, y));
    164 		return (CMPLXF(x + 0.0L + (y + 0), x + 0.0L + (y + 0)));
    165 	}
    166 
    167 	if (ax > RECIP_EPSILON || ay > RECIP_EPSILON) {
    168 		if (signbit(x) == 0)
    169 			w = clog_for_large_values(z) + m_ln2;
    170 		else
    171 			w = clog_for_large_values(-z) + m_ln2;
    172 		return (CMPLXF(copysignf(crealf(w), x),
    173 		    copysignf(cimagf(w), y)));
    174 	}
    175 
    176 	if (x == 0 && y == 0)
    177 		return (z);
    178 
    179 	raise_inexact();
    180 
    181 	if (ax < SQRT_6_EPSILON / 4 && ay < SQRT_6_EPSILON / 4)
    182 		return (z);
    183 
    184 	do_hard_work(ax, ay, &rx, &B_is_usable, &B, &sqrt_A2my2, &new_y);
    185 	if (B_is_usable)
    186 		ry = asinf(B);
    187 	else
    188 		ry = atan2f(new_y, sqrt_A2my2);
    189 	return (CMPLXF(copysignf(rx, x), copysignf(ry, y)));
    190 }
    191 
    192 float complex
    193 casinf(float complex z)
    194 {
    195 	float complex w = casinhf(CMPLXF(cimagf(z), crealf(z)));
    196 
    197 	return (CMPLXF(cimagf(w), crealf(w)));
    198 }
    199 
    200 float complex
    201 cacosf(float complex z)
    202 {
    203 	float x, y, ax, ay, rx, ry, B, sqrt_A2mx2, new_x;
    204 	int sx, sy;
    205 	int B_is_usable;
    206 	float complex w;
    207 
    208 	x = crealf(z);
    209 	y = cimagf(z);
    210 	sx = signbit(x);
    211 	sy = signbit(y);
    212 	ax = fabsf(x);
    213 	ay = fabsf(y);
    214 
    215 	if (isnan(x) || isnan(y)) {
    216 		if (isinf(x))
    217 			return (CMPLXF(y + y, -INFINITY));
    218 		if (isinf(y))
    219 			return (CMPLXF(x + x, -y));
    220 		if (x == 0)
    221 			return (CMPLXF(pio2_hi + pio2_lo, y + y));
    222 		return (CMPLXF(x + 0.0L + (y + 0), x + 0.0L + (y + 0)));
    223 	}
    224 
    225 	if (ax > RECIP_EPSILON || ay > RECIP_EPSILON) {
    226 		w = clog_for_large_values(z);
    227 		rx = fabsf(cimagf(w));
    228 		ry = crealf(w) + m_ln2;
    229 		if (sy == 0)
    230 			ry = -ry;
    231 		return (CMPLXF(rx, ry));
    232 	}
    233 
    234 	if (x == 1 && y == 0)
    235 		return (CMPLXF(0, -y));
    236 
    237 	raise_inexact();
    238 
    239 	if (ax < SQRT_6_EPSILON / 4 && ay < SQRT_6_EPSILON / 4)
    240 		return (CMPLXF(pio2_hi - (x - pio2_lo), -y));
    241 
    242 	do_hard_work(ay, ax, &ry, &B_is_usable, &B, &sqrt_A2mx2, &new_x);
    243 	if (B_is_usable) {
    244 		if (sx == 0)
    245 			rx = acosf(B);
    246 		else
    247 			rx = acosf(-B);
    248 	} else {
    249 		if (sx == 0)
    250 			rx = atan2f(sqrt_A2mx2, new_x);
    251 		else
    252 			rx = atan2f(sqrt_A2mx2, -new_x);
    253 	}
    254 	if (sy == 0)
    255 		ry = -ry;
    256 	return (CMPLXF(rx, ry));
    257 }
    258 
    259 float complex
    260 cacoshf(float complex z)
    261 {
    262 	float complex w;
    263 	float rx, ry;
    264 
    265 	w = cacosf(z);
    266 	rx = crealf(w);
    267 	ry = cimagf(w);
    268 	if (isnan(rx) && isnan(ry))
    269 		return (CMPLXF(ry, rx));
    270 	if (isnan(rx))
    271 		return (CMPLXF(fabsf(ry), rx));
    272 	if (isnan(ry))
    273 		return (CMPLXF(ry, ry));
    274 	return (CMPLXF(fabsf(ry), copysignf(rx, cimagf(z))));
    275 }
    276 
    277 static float complex
    278 clog_for_large_values(float complex z)
    279 {
    280 	float x, y;
    281 	float ax, ay, t;
    282 
    283 	x = crealf(z);
    284 	y = cimagf(z);
    285 	ax = fabsf(x);
    286 	ay = fabsf(y);
    287 	if (ax < ay) {
    288 		t = ax;
    289 		ax = ay;
    290 		ay = t;
    291 	}
    292 
    293 	if (ax > FLT_MAX / 2)
    294 		return (CMPLXF(logf(hypotf(x / m_e, y / m_e)) + 1,
    295 		    atan2f(y, x)));
    296 
    297 	if (ax > QUARTER_SQRT_MAX || ay < SQRT_MIN)
    298 		return (CMPLXF(logf(hypotf(x, y)), atan2f(y, x)));
    299 
    300 	return (CMPLXF(logf(ax * ax + ay * ay) / 2, atan2f(y, x)));
    301 }
    302 
    303 static inline float
    304 sum_squares(float x, float y)
    305 {
    306 
    307 	if (y < SQRT_MIN)
    308 		return (x * x);
    309 
    310 	return (x * x + y * y);
    311 }
    312 
    313 static inline float
    314 real_part_reciprocal(float x, float y)
    315 {
    316 	float scale;
    317 	uint32_t hx, hy;
    318 	int32_t ix, iy;
    319 
    320 	GET_FLOAT_WORD(hx, x);
    321 	ix = hx & 0x7f800000;
    322 	GET_FLOAT_WORD(hy, y);
    323 	iy = hy & 0x7f800000;
    324 #define	BIAS	(FLT_MAX_EXP - 1)
    325 #define	CUTOFF	(FLT_MANT_DIG / 2 + 1)
    326 	if (ix - iy >= CUTOFF << 23 || isinf(x))
    327 		return (1 / x);
    328 	if (iy - ix >= CUTOFF << 23)
    329 		return (x / y / y);
    330 	if (ix <= (BIAS + FLT_MAX_EXP / 2 - CUTOFF) << 23)
    331 		return (x / (x * x + y * y));
    332 	SET_FLOAT_WORD(scale, 0x7f800000 - ix);
    333 	x *= scale;
    334 	y *= scale;
    335 	return (x / (x * x + y * y) * scale);
    336 }
    337 
    338 float complex
    339 catanhf(float complex z)
    340 {
    341 	float x, y, ax, ay, rx, ry;
    342 
    343 	x = crealf(z);
    344 	y = cimagf(z);
    345 	ax = fabsf(x);
    346 	ay = fabsf(y);
    347 
    348 	if (y == 0 && ax <= 1)
    349 		return (CMPLXF(atanhf(x), y));
    350 
    351 	if (x == 0)
    352 		return (CMPLXF(x, atanf(y)));
    353 
    354 	if (isnan(x) || isnan(y)) {
    355 		if (isinf(x))
    356 			return (CMPLXF(copysignf(0, x), y + y));
    357 		if (isinf(y))
    358 			return (CMPLXF(copysignf(0, x),
    359 			    copysignf(pio2_hi + pio2_lo, y)));
    360 		return (CMPLXF(x + 0.0L + (y + 0), x + 0.0L + (y + 0)));
    361 	}
    362 
    363 	if (ax > RECIP_EPSILON || ay > RECIP_EPSILON)
    364 		return (CMPLXF(real_part_reciprocal(x, y),
    365 		    copysignf(pio2_hi + pio2_lo, y)));
    366 
    367 	if (ax < SQRT_3_EPSILON / 2 && ay < SQRT_3_EPSILON / 2) {
    368 		raise_inexact();
    369 		return (z);
    370 	}
    371 
    372 	if (ax == 1 && ay < FLT_EPSILON)
    373 		rx = (m_ln2 - logf(ay)) / 2;
    374 	else
    375 		rx = log1pf(4 * ax / sum_squares(ax - 1, ay)) / 4;
    376 
    377 	if (ax == 1)
    378 		ry = atan2f(2, -ay) / 2;
    379 	else if (ay < FLT_EPSILON)
    380 		ry = atan2f(2 * ay, (1 - ax) * (1 + ax)) / 2;
    381 	else
    382 		ry = atan2f(2 * ay, (1 - ax) * (1 + ax) - ay * ay) / 2;
    383 
    384 	return (CMPLXF(copysignf(rx, x), copysignf(ry, y)));
    385 }
    386 
    387 float complex
    388 catanf(float complex z)
    389 {
    390 	float complex w = catanhf(CMPLXF(cimagf(z), crealf(z)));
    391 
    392 	return (CMPLXF(cimagf(w), crealf(w)));
    393 }
    394