xref: /freebsd/lib/msun/ld80/s_asinpil.c (revision ae417b3194e76ce26065dc20281493ee83619879)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2026 Steven G. Kargl
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice unmodified, this list of conditions, and the following
12  *    disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
18  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
19  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
20  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
21  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
22  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
26  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27  */
28 
29 /*
30  * src/s_asinpi.c for implemenation details.
31  */
32 
33 #ifdef __i386__
34 #include <ieeefp.h>
35 #endif
36 #include <stdint.h>
37 
38 #include "fpmath.h"
39 #include "math.h"
40 #include "math_private.h"
41 
42 #define _CC	(0x1p32L + 1)
43 #define _ROOT	sqrtl
44 
45 volatile static const double tiny = 1.e-300;
46 static const double half = 0.5, one = 1.;
47 
48 /* 1/pi split into the leading and trailing 53 bits. */
49 static const double
50 invpihi =  3.1830988618379069e-01,
51 invpilo = -1.9678676675182486e-17;
52 
53 /*
54  * Prior to the leading multiplication by x^2, the rational approximation
55  * has an absolute minimax error less than 1.36e-22 over the [0x1p-32,0.5]
56  * domain (or log2(error) = -72.6).
57  */
58 static inline long double
__r(long double xs)59 __r(long double xs)
60 {
61 	static const union IEEEl2bits
62 	    R0u = LD80C(0xd94caf3dbdb01c38,  -5,  5.30516476972984452564e-02L),
63 	    R1u = LD80C(0x8d346599ebe3212a,  -3, -1.37895190734499743665e-01L),
64 	    R2u = LD80C(0x852c1751d7112655,  -3,  1.30051006670116064731e-01L),
65 	    R3u = LD80C(0xdb55697553742657,  -5, -5.35482520546937456640e-02L),
66 	    R4u = LD80C(0x931ba448fbd6e8c3,  -7,  8.97875827280130762861e-03L),
67 	    R5u = LD80C(0xdbbe8e80762881b0, -12, -4.19129108247665906395e-04L),
68 	    S1u = LD80C(0xc3272074944c6389,   1, -3.04926310906120631911e+00L),
69 	    S2u = LD80C(0xe390d58f48c1a96d,   1,  3.55571497910116337692e+00L),
70 	    S3u = LD80C(0xfccb6664d1fbfc39,   0, -1.97495727465499715865e+00L),
71 	    S4u = LD80C(0x86f4f140e3be416b,  -1,  5.27175024358933283746e-01L),
72 	    S5u = LD80C(0xf218070935962bdb,  -5, -5.91049456446391222176e-02L),
73 	    S6u = LD80C(0xec656d26215b624b, -10,  1.80355985054588957542e-03L);
74 
75 #define	R0	(R0u.e)
76 #define	R1	(R1u.e)
77 #define	R2	(R2u.e)
78 #define	R3	(R3u.e)
79 #define	R4	(R4u.e)
80 #define	R5	(R5u.e)
81 #define	S1	(S1u.e)
82 #define	S2	(S2u.e)
83 #define	S3	(S3u.e)
84 #define	S4	(S4u.e)
85 #define	S5	(S5u.e)
86 #define	S6	(S6u.e)
87 
88 	long double r, s;
89 	r = R0 + (R1 + (R2 + (R3 + (R4 + R5 * xs) * xs) * xs) * xs) * xs;
90 	s =  1 + (S1 + (S2 + (S3 + (S4 + (S5 + S6 * xs) * xs) * xs) *
91 	    xs) * xs) * xs;
92 	return (xs * (r / s));
93 }
94 
95 #define	GREATER(a)	(ix == a && lx >  0x8000000000000000ull)
96 #define	LESSEQ(a)	(ix == a && lx <= 0x8000000000000000ull)
97 
98 long double
asinpil(long double x)99 asinpil(long double x)
100 {
101 	long double ax, hi, lo, xh, xl, y, zh, zl;
102 	uint64_t lx;
103 	uint16_t hx, ix;
104 
105 	EXTRACT_LDBL80_WORDS(hx, lx, x);
106 	ix = hx & 0x7fff;
107 
108 	if (ix >= 0x4000 || GREATER(0x3fff))	/* |x| > 1 */
109 		return ((x - x) / (x - x));
110 
111 	ENTERI();
112 
113 	INSERT_LDBL80_WORDS(ax, ix, lx);
114 
115 	if (ix < 0x3ffe || LESSEQ(0x3ffe)) {	/* |x| <= 0.5 */
116 		if (ix < 0x3fde) {		/* |x| < 0x1p-33 */
117 			if (ix < 0x002b) {	/* |x| < 0x1p-16340 */
118 				if ((ix | lx) == 0)
119 					RETURNI(x);
120 				/* Scale for near subnormal. */
121 				ax *= 0x1p65;
122 				_XMUL(ax, 0, invpihi, invpilo, hi, lo);
123 				y = (hi + lo) * 0x1p-65;
124 			} else {
125 				_XMUL(ax, 0, invpihi, invpilo, hi, lo);
126 				y = hi + lo;
127 			}
128 		} else {
129 			y = __r(ax * ax);
130 			_XADD(invpihi, invpilo, y, 0, xh, xl);
131 			_XMUL(ax, 0, xh, xl, hi, lo);
132 			y = hi + lo;
133 		}
134 	} else if (ix < 0x3fff) {		/* |x| < 1 */
135 		y = 1 - ax;
136 		x = __r(y / 2);
137 		_XADD(invpihi, invpilo, x, 0, xh, xl);	/* 1 / pi + r(t^2) */
138 		_SQRT(2 * y, zh, zl);			/* 2 * t */
139 		_XMUL(xh, xl, zh, zl, hi, lo);
140 		_XADD(half, 0, -hi, -lo, y, x);
141 	} else					/* |x| == 1 */
142 		y = half;
143 
144 	RETURNI((hx & 0x8000) ? -y : y);
145 }
146 
147 /*
148  * See src/s_asinpi.c for implementation details.
149  */
150 
151 long double
acospil(long double x)152 acospil(long double x)
153 {
154 	long double ax, hi, lo, xh, xl, y, zh, zl;
155 	uint64_t lx;
156 	uint16_t hx, ix;
157 
158 	EXTRACT_LDBL80_WORDS(hx, lx, x);
159 	ix = hx & 0x7fff;
160 
161 	if (ix >= 0x4000 || GREATER(0x3fff))	/* |x| > 1 */
162 		return ((x - x) / (x - x));
163 
164 	ENTERI();
165 
166 	if (ix < 0x3ffe || LESSEQ(0x3ffe)) {	/* |x| <= 0.5 */
167 		if (ix < 0x3fe9) {		/* |x| < 0x1p-22 */
168 			y = ((ix | lx) == 0) ? half : (LESSEQ(0x3fbf) ?
169 			    half - tiny : half - x * invpihi);
170 		} else {
171 			y = __r(x * x);
172 			_XADD(invpihi, invpilo, y, 0, xh, xl);
173 			_XMUL(x, 0, xh, xl, hi, lo);
174 			_XADD(half, 0, -hi, -lo, y, ax);
175 		}
176 	} else if (ix < 0x3fff) {		/* |x| < 1 */
177 		INSERT_LDBL80_WORDS(ax, ix, lx);
178 		y = 1 - ax;
179 		ax = __r(y / 2);
180 		_XADD(invpihi, invpilo, ax, 0, xh, xl);	/* 1 / pi + r(t^2) */
181 		_SQRT(2 * y, zh, zl);			/* 2 * t */
182 		_XMUL(xh, xl, zh, zl, hi, lo);
183 		if (hx & 0x8000)
184 			_XADD(one, 0, -hi, -lo, y, ax);
185 		else
186 			y = hi + lo;
187 	} else					/* |x| == 1 */
188 		y = half;
189 
190 	RETURNI(y);
191 }
192