1 /*-
2 * Copyright (c) 2017, 2023 Steven G. Kargl
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 unmodified, this list of conditions, and the following
10 * 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 ``AS IS'' AND ANY EXPRESS OR
16 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
17 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
18 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
19 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
20 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
21 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
22 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
23 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
24 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
25 */
26
27 /**
28 * sinpi(x) computes sin(pi*x) without multiplication by pi (almost). First,
29 * note that sinpi(-x) = -sinpi(x), so the algorithm considers only |x| and
30 * includes reflection symmetry by considering the sign of x on output. The
31 * method used depends on the magnitude of x.
32 *
33 * 1. For small |x|, sinpi(x) = pi * x where a sloppy threshold is used. The
34 * threshold is |x| < 0x1pN with N = -(P/2+M). P is the precision of the
35 * floating-point type and M = 2 to 4. To achieve high accuracy, pi is
36 * decomposed into high and low parts with the high part containing a
37 * number of trailing zero bits. x is also split into high and low parts.
38 *
39 * 2. For |x| < 1, argument reduction is not required and sinpi(x) is
40 * computed by calling a kernel that leverages the kernels for sin(x)
41 * ans cos(x). See k_sinpi.c and k_cospi.c for details.
42 *
43 * 3. For 1 <= |x| < 0x1p(P-1), argument reduction is required where
44 * |x| = j0 + r with j0 an integer and the remainder r satisfies
45 * 0 <= r < 1. With the given domain, a simplified inline floor(x)
46 * is used. Also, note the following identity
47 *
48 * sinpi(x) = sin(pi*(j0+r))
49 * = sin(pi*j0) * cos(pi*r) + cos(pi*j0) * sin(pi*r)
50 * = cos(pi*j0) * sin(pi*r)
51 * = +-sinpi(r)
52 *
53 * If j0 is even, then cos(pi*j0) = 1. If j0 is odd, then cos(pi*j0) = -1.
54 * sinpi(r) is then computed via an appropriate kernel.
55 *
56 * 4. For |x| >= 0x1p(P-1), |x| is integral and sinpi(x) = copysign(0,x).
57 *
58 * 5. Special cases:
59 *
60 * sinpi(+-0) = +-0
61 * sinpi(+-n) = +-0, for positive integers n.
62 * sinpi(+-inf) = nan. Raises the "invalid" floating-point exception.
63 * sinpi(nan) = nan. Raises the "invalid" floating-point exception.
64 */
65
66 #include <sys/cdefs.h>
67
68 #include "namespace.h"
69 __weak_alias(sinpi, _sinpi)
70 #include <float.h>
71 #include "math.h"
72 #include "math_private.h"
73
74 static const double
75 pi_hi = 3.1415926814079285e+00, /* 0x400921fb 0x58000000 */
76 pi_lo =-2.7818135228334233e-08; /* 0xbe5dde97 0x3dcb3b3a */
77
78 #include "k_cospi.h"
79 #include "k_sinpi.h"
80
81 static volatile const double vzero = 0;
82
83 double
sinpi(double x)84 sinpi(double x)
85 {
86 double ax, hi, lo, s;
87 uint32_t hx, ix, j0, lx;
88
89 EXTRACT_WORDS(hx, lx, x);
90 ix = hx & 0x7fffffff;
91 INSERT_WORDS(ax, ix, lx);
92
93 if (ix < 0x3ff00000) { /* |x| < 1 */
94 if (ix < 0x3fd00000) { /* |x| < 0.25 */
95 if (ix < 0x3e200000) { /* |x| < 0x1p-29 */
96 if (x == 0)
97 return (x);
98 /*
99 * To avoid issues with subnormal values,
100 * scale the computation and rescale on
101 * return.
102 */
103 INSERT_WORDS(hi, hx, 0);
104 hi *= 0x1p53;
105 lo = x * 0x1p53 - hi;
106 s = (pi_lo + pi_hi) * lo + pi_lo * hi +
107 pi_hi * hi;
108 return (s * 0x1p-53);
109 }
110
111 s = __kernel_sinpi(ax);
112 return ((hx & 0x80000000) ? -s : s);
113 }
114
115 if (ix < 0x3fe00000) /* |x| < 0.5 */
116 s = __kernel_cospi(0.5 - ax);
117 else if (ix < 0x3fe80000) /* |x| < 0.75 */
118 s = __kernel_cospi(ax - 0.5);
119 else
120 s = __kernel_sinpi(1 - ax);
121 return ((hx & 0x80000000) ? -s : s);
122 }
123
124 if (ix < 0x43300000) { /* 1 <= |x| < 0x1p52 */
125 FFLOOR(x, j0, ix, lx); /* Integer part of ax. */
126 ax -= x;
127 EXTRACT_WORDS(ix, lx, ax);
128
129 if (ix == 0)
130 s = 0;
131 else {
132 if (ix < 0x3fe00000) { /* |x| < 0.5 */
133 if (ix < 0x3fd00000) /* |x| < 0.25 */
134 s = __kernel_sinpi(ax);
135 else
136 s = __kernel_cospi(0.5 - ax);
137 } else {
138 if (ix < 0x3fe80000) /* |x| < 0.75 */
139 s = __kernel_cospi(ax - 0.5);
140 else
141 s = __kernel_sinpi(1 - ax);
142 }
143
144 if (j0 > 30)
145 x -= 0x1p30;
146 j0 = (uint32_t)x;
147 if (j0 & 1) s = -s;
148 }
149
150 return ((hx & 0x80000000) ? -s : s);
151 }
152
153 /* x = +-inf or nan. */
154 if (ix >= 0x7ff00000)
155 return (vzero / vzero);
156
157 /*
158 * |x| >= 0x1p52 is always an integer, so return +-0.
159 */
160 return (copysign(0, x));
161 }
162