xref: /netbsd-src/lib/libm/src/k_tan.c (revision 458288a6268dbcae17521ee6a2825ce41a73ff7b)
1 /* @(#)k_tan.c 5.1 93/09/24 */
2 /*
3  * ====================================================
4  * Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved.
5  *
6  * Developed at SunPro, a Sun Microsystems, Inc. business.
7  * Permission to use, copy, modify, and distribute this
8  * software is freely granted, provided that this notice
9  * is preserved.
10  * ====================================================
11  */
12 
13 #include <sys/cdefs.h>
14 #if defined(LIBM_SCCS) && !defined(lint)
15 __RCSID("$NetBSD: k_tan.c,v 1.12 2004/07/22 18:24:09 drochner Exp $");
16 #endif
17 
18 /* __kernel_tan( x, y, k )
19  * kernel tan function on [-pi/4, pi/4], pi/4 ~ 0.7854
20  * Input x is assumed to be bounded by ~pi/4 in magnitude.
21  * Input y is the tail of x.
22  * Input k indicates whether tan (if k=1) or
23  * -1/tan (if k= -1) is returned.
24  *
25  * Algorithm
26  *	1. Since tan(-x) = -tan(x), we need only to consider positive x.
27  *	2. if x < 2^-28 (hx<0x3e300000 0), return x with inexact if x!=0.
28  *	3. tan(x) is approximated by a odd polynomial of degree 27 on
29  *	   [0,0.67434]
30  *		  	         3             27
31  *	   	tan(x) ~ x + T1*x + ... + T13*x
32  *	   where
33  *
34  * 	        |tan(x)         2     4            26   |     -59.2
35  * 	        |----- - (1+T1*x +T2*x +.... +T13*x    )| <= 2
36  * 	        |  x 					|
37  *
38  *	   Note: tan(x+y) = tan(x) + tan'(x)*y
39  *		          ~ tan(x) + (1+x*x)*y
40  *	   Therefore, for better accuracy in computing tan(x+y), let
41  *		     3      2      2       2       2
42  *		r = x *(T2+x *(T3+x *(...+x *(T12+x *T13))))
43  *	   then
44  *		 		    3    2
45  *		tan(x+y) = x + (T1*x + (x *(r+y)+y))
46  *
47  *      4. For x in [0.67434,pi/4],  let y = pi/4 - x, then
48  *		tan(x) = tan(pi/4-y) = (1-tan(y))/(1+tan(y))
49  *		       = 1 - 2*(tan(y) - (tan(y)^2)/(1+tan(y)))
50  */
51 
52 #include "math.h"
53 #include "math_private.h"
54 
55 static const double xxx[] = {
56 		 3.33333333333334091986e-01,	/* 3FD55555, 55555563 */
57 		 1.33333333333201242699e-01,	/* 3FC11111, 1110FE7A */
58 		 5.39682539762260521377e-02,	/* 3FABA1BA, 1BB341FE */
59 		 2.18694882948595424599e-02,	/* 3F9664F4, 8406D637 */
60 		 8.86323982359930005737e-03,	/* 3F8226E3, E96E8493 */
61 		 3.59207910759131235356e-03,	/* 3F6D6D22, C9560328 */
62 		 1.45620945432529025516e-03,	/* 3F57DBC8, FEE08315 */
63 		 5.88041240820264096874e-04,	/* 3F4344D8, F2F26501 */
64 		 2.46463134818469906812e-04,	/* 3F3026F7, 1A8D1068 */
65 		 7.81794442939557092300e-05,	/* 3F147E88, A03792A6 */
66 		 7.14072491382608190305e-05,	/* 3F12B80F, 32F0A7E9 */
67 		-1.85586374855275456654e-05,	/* BEF375CB, DB605373 */
68 		 2.59073051863633712884e-05,	/* 3EFB2A70, 74BF7AD4 */
69 /* one */	 1.00000000000000000000e+00,	/* 3FF00000, 00000000 */
70 /* pio4 */	 7.85398163397448278999e-01,	/* 3FE921FB, 54442D18 */
71 /* pio4lo */	 3.06161699786838301793e-17	/* 3C81A626, 33145C07 */
72 };
73 #define	one	xxx[13]
74 #define	pio4	xxx[14]
75 #define	pio4lo	xxx[15]
76 #define	T	xxx
77 
78 double
__kernel_tan(double x,double y,int iy)79 __kernel_tan(double x, double y, int iy)
80 {
81 	double z, r, v, w, s;
82 	int32_t ix, hx;
83 
84 	GET_HIGH_WORD(hx, x);	/* high word of x */
85 	ix = hx & 0x7fffffff;			/* high word of |x| */
86 	if (ix < 0x3e300000) {			/* x < 2**-28 */
87 		if ((int) x == 0) {		/* generate inexact */
88 			u_int32_t low;
89 			GET_LOW_WORD(low, x);
90 			if(((ix | low) | (iy + 1)) == 0)
91 				return one / fabs(x);
92 			else {
93 				if (iy == 1)
94 					return x;
95 				else {	/* compute -1 / (x+y) carefully */
96 					double a, t;
97 
98 					z = w = x + y;
99 					SET_LOW_WORD(z, 0);
100 					v = y - (z - x);
101 					t = a = -one / w;
102 					SET_LOW_WORD(t, 0);
103 					s = one + t * z;
104 					return t + a * (s + t * v);
105 				}
106 			}
107 		}
108 	}
109 	if (ix >= 0x3FE59428) {	/* |x| >= 0.6744 */
110 		if (hx < 0) {
111 			x = -x;
112 			y = -y;
113 		}
114 		z = pio4 - x;
115 		w = pio4lo - y;
116 		x = z + w;
117 		y = 0.0;
118 	}
119 	z = x * x;
120 	w = z * z;
121 	/*
122 	 * Break x^5*(T[1]+x^2*T[2]+...) into
123 	 * x^5(T[1]+x^4*T[3]+...+x^20*T[11]) +
124 	 * x^5(x^2*(T[2]+x^4*T[4]+...+x^22*[T12]))
125 	 */
126 	r = T[1] + w * (T[3] + w * (T[5] + w * (T[7] + w * (T[9] +
127 		w * T[11]))));
128 	v = z * (T[2] + w * (T[4] + w * (T[6] + w * (T[8] + w * (T[10] +
129 		w * T[12])))));
130 	s = z * x;
131 	r = y + z * (s * (r + v) + y);
132 	r += T[0] * s;
133 	w = x + r;
134 	if (ix >= 0x3FE59428) {
135 		v = (double) iy;
136 		return (double) (1 - ((hx >> 30) & 2)) *
137 			(v - 2.0 * (x - (w * w / (w + v) - r)));
138 	}
139 	if (iy == 1)
140 		return w;
141 	else {
142 		/*
143 		 * if allow error up to 2 ulp, simply return
144 		 * -1.0 / (x+r) here
145 		 */
146 		/* compute -1.0 / (x+r) accurately */
147 		double a, t;
148 		z = w;
149 		SET_LOW_WORD(z, 0);
150 		v = r - (z - x);	/* z+v = r+x */
151 		t = a = -1.0 / w;	/* a = -1.0/w */
152 		SET_LOW_WORD(t, 0);
153 		s = 1.0 + t * z;
154 		return t + a * (s + t * v);
155 	}
156 }
157