xref: /freebsd/lib/msun/ld80/s_atanpil.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_atanpi.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 #define NBIT	(0x8000000000000000ull)
45 
46 volatile static const double tiny = 1.e-300;
47 static const double half = 0.5, one = 1., qrtr = 0.25;
48 static const double x0 = 0.625, x1 = 0.875, x2 = 1.5;
49 
50 /* 53-bit high and low parts. */
51 static const double
52 invpihi =  3.1830988618379069e-01,	/* 1/pi */
53 invpilo = -1.9678676675182486e-17,	/* 1/pi */
54 a0hi =  1.7780768448935275e-01,		/* atanpi(x0) */
55 a0lo =  6.7223942595197191e-18,		/* atanpi(x0) */
56 a1hi =  2.2881069536505358e-01,		/* atanpi(x1) */
57 a1lo =  8.7193139538130510e-18,		/* atanpi(x1) */
58 a2hi =  3.1283295818900120e-01,		/* atanpi(x2) */
59 a2lo = -1.4076885713501453e-17;		/* atanpi(x2) */
60 
61 /*
62  * Prior to the leading multiplication by x^2, the rational approximation
63  * has an absolute minimax error less than 1.22e-23 over the [0x1p-32,0.5]
64  * domain (or log2(error) = -76.1).
65  */
66 static inline long double
__r(long double xs)67 __r(long double xs)
68 {
69 	static const union IEEEl2bits
70 	    R0u = LD80C(0xd94caf3dbdb01c38,  -4, -1.06103295394596890513e-01L),
71 	    R1u = LD80C(0x845b6d12f35f0ccb,  -2, -2.58510025561632250608e-01L),
72 	    R2u = LD80C(0xe679d75bf98585ae,  -3, -2.25074162472429082081e-01L),
73 	    R3u = LD80C(0xaba63bb6f14e17e4,  -4, -8.38131585316324449243e-02L),
74 	    R4u = LD80C(0xca768e4a7d2fd52e,  -7, -1.23573674736290235872e-02L),
75 	    R5u = LD80C(0x802e09b751d4c6bc, -11, -4.88967268965928487608e-04L),
76 	    S1u = LD80C(0xc2545ef3d335e598,   1,  3.03639959155120062705e+00L),
77 	    S2u = LD80C(0xe0ee42f67b6b24bc,   1,  3.51454233236806805243e+00L),
78 	    S3u = LD80C(0xf7200820c8d271d1,   0,  1.93066503144077077959e+00L),
79 	    S4u = LD80C(0x820ec69663c6f1f3,  -1,  5.08037959781883845928e-01L),
80 	    S5u = LD80C(0xe61a8fde47e2f517,  -5,  5.61776752333507006525e-02L),
81 	    S6u = LD80C(0xe260573083b4d35b, -10,  1.72711433720654801154e-03L);
82 
83 #define	R0	(R0u.e)
84 #define	R1	(R1u.e)
85 #define	R2	(R2u.e)
86 #define	R3	(R3u.e)
87 #define	R4	(R4u.e)
88 #define	R5	(R5u.e)
89 #define	S1	(S1u.e)
90 #define	S2	(S2u.e)
91 #define	S3	(S3u.e)
92 #define	S4	(S4u.e)
93 #define	S5	(S5u.e)
94 #define	S6	(S6u.e)
95 
96 	long double r, s;
97 	r = R0 + (R1 + (R2 + (R3 + (R4 + R5 * xs) * xs) * xs) * xs) * xs;
98 	s =  1 + (S1 + (S2 + (S3 + (S4 + (S5 + S6 * xs) * xs) * xs) *
99 	    xs) * xs) * xs;
100 	return (xs * (r / s));
101 }
102 
103 long double
atanpil(long double x)104 atanpil(long double x)
105 {
106 	long double ax, hi, lo, xh, xl, y, zh, zl;
107 	uint64_t lx;
108 	uint16_t hx, ix;
109 
110 	EXTRACT_LDBL80_WORDS(hx, lx, x);
111 	ix = hx & 0x7fff;
112 
113 	/* x = +-inf, nan */
114 	if (ix >= 0x7fff && lx >= 0x8000000000000000ull) {
115 		if (lx > 0x8000000000000000ull)
116 			return (x + x);
117 		return ((hx & 0x8000) ? -half : half);
118 	}
119 
120 	ENTERI();
121 
122 	INSERT_LDBL80_WORDS(ax, ix, lx);
123 
124 	if (ix < 0x3ffe ) { /* |x| < 0.5 */
125 		if (ix < 0x3fde) {		/* |x| < 0x1p-33 */
126 			if (ix < 0x002b) {	/* |x| < 0x1p-16340 */
127 				if ((ix | lx) == 0)
128 					RETURNI(x);
129 				/* Scale for near subnormal. */
130 				ax *= 0x1p65;
131 				_XMUL(ax, 0, invpihi, invpilo, hi, lo);
132 				y = (hi + lo) * 0x1p-65;
133 			} else {
134 				_XMUL(ax, 0, invpihi, invpilo, hi, lo);
135 				y = hi + lo;
136 			}
137 		} else {
138 			y = __r(ax * ax);
139 			_XADD(invpihi, invpilo, y, 0, xh, xl);
140 			_XMUL(ax, 0, xh, xl, hi, lo);
141 			y = hi + lo;
142 		}
143 	} else if (ax < 1) {	/* |x| < 1 */
144 		/* |x| < 0.75 */
145 		if (ix == 0x3ffe && lx < 0xc000000000000000ull) {
146 			x = (ax - x0) / (1 + x0 * ax);
147 			y = __r(x * x);
148 			_XADD(invpihi, invpilo, y, 0, xh, xl);
149 			_XMUL(x, 0, xh, xl, hi, lo);
150 			_XADD(a0hi, a0lo, hi, lo, y, xl);
151 		} else {
152 			x = (ax - x1) / (1 + x1 * ax);
153 			y = __r(x * x);
154 			_XADD(invpihi, invpilo, y, 0, xh, xl);
155 			_XMUL(x, 0, xh, xl, hi, lo);
156 			_XADD(a1hi, a1lo, hi, lo, y, xl);
157 		}
158 	} else if (ix < 0x4000) {		/* |x| < 2 */
159 		if (ix == 0x3fff && lx == NBIT)
160 			return ((hx & 0x8000) ? -qrtr : qrtr);
161 		x = (ax - x2) / (1 + x2 * ax);
162 		y = __r(x * x);
163 		_XADD(invpihi, invpilo, y, 0, xh, xl);
164 		_XMUL(x, 0, xh, xl, hi, lo);
165 		_XADD(a2hi, a2lo, hi, lo, y, xl);
166 	} else {				/* |x| > 2 */
167 		x = 1 / ax;
168 		y = __r(x * x);
169 		_XADD(invpihi, invpilo, y, 0, xh, xl);
170 		_XMUL(x, 0, xh, xl, hi, lo);
171 		_XADD(half, 0, -hi, -lo, y, x);
172 	}
173 
174 	RETURNI((hx & 0x8000) ? -y : y);
175 }
176