xref: /openbsd-src/lib/libm/src/k_tan.c (revision 2b0358df1d88d06ef4139321dd05bd5e05d91eaf)
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 #if defined(LIBM_SCCS) && !defined(lint)
14 static char rcsid[] = "$NetBSD: k_tan.c,v 1.8 1995/05/10 20:46:37 jtc Exp $";
15 #endif
16 
17 /* __kernel_tan( x, y, k )
18  * kernel tan function on [-pi/4, pi/4], pi/4 ~ 0.7854
19  * Input x is assumed to be bounded by ~pi/4 in magnitude.
20  * Input y is the tail of x.
21  * Input k indicates whether tan (if k=1) or
22  * -1/tan (if k= -1) is returned.
23  *
24  * Algorithm
25  *	1. Since tan(-x) = -tan(x), we need only to consider positive x.
26  *	2. if x < 2^-28 (hx<0x3e300000 0), return x with inexact if x!=0.
27  *	3. tan(x) is approximated by a odd polynomial of degree 27 on
28  *	   [0,0.67434]
29  *		  	         3             27
30  *	   	tan(x) ~ x + T1*x + ... + T13*x
31  *	   where
32  *
33  * 	        |tan(x)         2     4            26   |     -59.2
34  * 	        |----- - (1+T1*x +T2*x +.... +T13*x    )| <= 2
35  * 	        |  x 					|
36  *
37  *	   Note: tan(x+y) = tan(x) + tan'(x)*y
38  *		          ~ tan(x) + (1+x*x)*y
39  *	   Therefore, for better accuracy in computing tan(x+y), let
40  *		     3      2      2       2       2
41  *		r = x *(T2+x *(T3+x *(...+x *(T12+x *T13))))
42  *	   then
43  *		 		    3    2
44  *		tan(x+y) = x + (T1*x + (x *(r+y)+y))
45  *
46  *      4. For x in [0.67434,pi/4],  let y = pi/4 - x, then
47  *		tan(x) = tan(pi/4-y) = (1-tan(y))/(1+tan(y))
48  *		       = 1 - 2*(tan(y) - (tan(y)^2)/(1+tan(y)))
49  */
50 
51 #include "math.h"
52 #include "math_private.h"
53 
54 static const double xxx[] = {
55 		 3.33333333333334091986e-01,	/* 3FD55555, 55555563 */
56 		 1.33333333333201242699e-01,	/* 3FC11111, 1110FE7A */
57 		 5.39682539762260521377e-02,	/* 3FABA1BA, 1BB341FE */
58 		 2.18694882948595424599e-02,	/* 3F9664F4, 8406D637 */
59 		 8.86323982359930005737e-03,	/* 3F8226E3, E96E8493 */
60 		 3.59207910759131235356e-03,	/* 3F6D6D22, C9560328 */
61 		 1.45620945432529025516e-03,	/* 3F57DBC8, FEE08315 */
62 		 5.88041240820264096874e-04,	/* 3F4344D8, F2F26501 */
63 		 2.46463134818469906812e-04,	/* 3F3026F7, 1A8D1068 */
64 		 7.81794442939557092300e-05,	/* 3F147E88, A03792A6 */
65 		 7.14072491382608190305e-05,	/* 3F12B80F, 32F0A7E9 */
66 		-1.85586374855275456654e-05,	/* BEF375CB, DB605373 */
67 		 2.59073051863633712884e-05,	/* 3EFB2A70, 74BF7AD4 */
68 /* one */	 1.00000000000000000000e+00,	/* 3FF00000, 00000000 */
69 /* pio4 */	 7.85398163397448278999e-01,	/* 3FE921FB, 54442D18 */
70 /* pio4lo */	 3.06161699786838301793e-17	/* 3C81A626, 33145C07 */
71 };
72 #define	one	xxx[13]
73 #define	pio4	xxx[14]
74 #define	pio4lo	xxx[15]
75 #define	T	xxx
76 
77 double
78 __kernel_tan(double x, double y, int iy)
79 {
80 	double z, r, v, w, s;
81 	int32_t ix, hx;
82 
83 	GET_HIGH_WORD(hx, x);	/* high word of x */
84 	ix = hx & 0x7fffffff;			/* high word of |x| */
85 	if (ix < 0x3e300000) {			/* x < 2**-28 */
86 		if ((int) x == 0) {		/* generate inexact */
87 			u_int32_t low;
88 			GET_LOW_WORD(low, x);
89 			if(((ix | low) | (iy + 1)) == 0)
90 				return one / fabs(x);
91 			else {
92 				if (iy == 1)
93 					return x;
94 				else {	/* compute -1 / (x+y) carefully */
95 					double a, t;
96 
97 					z = w = x + y;
98 					SET_LOW_WORD(z, 0);
99 					v = y - (z - x);
100 					t = a = -one / w;
101 					SET_LOW_WORD(t, 0);
102 					s = one + t * z;
103 					return t + a * (s + t * v);
104 				}
105 			}
106 		}
107 	}
108 	if (ix >= 0x3FE59428) {	/* |x| >= 0.6744 */
109 		if (hx < 0) {
110 			x = -x;
111 			y = -y;
112 		}
113 		z = pio4 - x;
114 		w = pio4lo - y;
115 		x = z + w;
116 		y = 0.0;
117 	}
118 	z = x * x;
119 	w = z * z;
120 	/*
121 	 * Break x^5*(T[1]+x^2*T[2]+...) into
122 	 * x^5(T[1]+x^4*T[3]+...+x^20*T[11]) +
123 	 * x^5(x^2*(T[2]+x^4*T[4]+...+x^22*[T12]))
124 	 */
125 	r = T[1] + w * (T[3] + w * (T[5] + w * (T[7] + w * (T[9] +
126 		w * T[11]))));
127 	v = z * (T[2] + w * (T[4] + w * (T[6] + w * (T[8] + w * (T[10] +
128 		w * T[12])))));
129 	s = z * x;
130 	r = y + z * (s * (r + v) + y);
131 	r += T[0] * s;
132 	w = x + r;
133 	if (ix >= 0x3FE59428) {
134 		v = (double) iy;
135 		return (double) (1 - ((hx >> 30) & 2)) *
136 			(v - 2.0 * (x - (w * w / (w + v) - r)));
137 	}
138 	if (iy == 1)
139 		return w;
140 	else {
141 		/*
142 		 * if allow error up to 2 ulp, simply return
143 		 * -1.0 / (x+r) here
144 		 */
145 		/* compute -1.0 / (x+r) accurately */
146 		double a, t;
147 		z = w;
148 		SET_LOW_WORD(z, 0);
149 		v = r - (z - x);	/* z+v = r+x */
150 		t = a = -1.0 / w;	/* a = -1.0/w */
151 		SET_LOW_WORD(t, 0);
152 		s = 1.0 + t * z;
153 		return t + a * (s + t * v);
154 	}
155 }
156