xref: /freebsd/lib/msun/src/k_tan.c (revision 2008043f386721d58158e37e0d7e50df8095942d)
1 /* @(#)k_tan.c 1.5 04/04/22 SMI */
2 
3 /*
4  * ====================================================
5  * Copyright 2004 Sun Microsystems, Inc.  All Rights Reserved.
6  *
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 /* INDENT OFF */
14 #include <sys/cdefs.h>
15 /* __kernel_tan( x, y, k )
16  * kernel tan function on ~[-pi/4, pi/4] (except on -0), pi/4 ~ 0.7854
17  * Input x is assumed to be bounded by ~pi/4 in magnitude.
18  * Input y is the tail of x.
19  * Input k indicates whether tan (if k = 1) or -1/tan (if k = -1) is returned.
20  *
21  * Algorithm
22  *	1. Since tan(-x) = -tan(x), we need only to consider positive x.
23  *	2. Callers must return tan(-0) = -0 without calling here since our
24  *	   odd polynomial is not evaluated in a way that preserves -0.
25  *	   Callers may do the optimization tan(x) ~ x for tiny x.
26  *	3. tan(x) is approximated by a odd polynomial of degree 27 on
27  *	   [0,0.67434]
28  *		  	         3             27
29  *	   	tan(x) ~ x + T1*x + ... + T13*x
30  *	   where
31  *
32  * 	        |tan(x)         2     4            26   |     -59.2
33  * 	        |----- - (1+T1*x +T2*x +.... +T13*x    )| <= 2
34  * 	        |  x 					|
35  *
36  *	   Note: tan(x+y) = tan(x) + tan'(x)*y
37  *		          ~ tan(x) + (1+x*x)*y
38  *	   Therefore, for better accuracy in computing tan(x+y), let
39  *		     3      2      2       2       2
40  *		r = x *(T2+x *(T3+x *(...+x *(T12+x *T13))))
41  *	   then
42  *		 		    3    2
43  *		tan(x+y) = x + (T1*x + (x *(r+y)+y))
44  *
45  *      4. For x in [0.67434,pi/4],  let y = pi/4 - x, then
46  *		tan(x) = tan(pi/4-y) = (1-tan(y))/(1+tan(y))
47  *		       = 1 - 2*(tan(y) - (tan(y)^2)/(1+tan(y)))
48  */
49 
50 #include "math.h"
51 #include "math_private.h"
52 static const double xxx[] = {
53 		 3.33333333333334091986e-01,	/* 3FD55555, 55555563 */
54 		 1.33333333333201242699e-01,	/* 3FC11111, 1110FE7A */
55 		 5.39682539762260521377e-02,	/* 3FABA1BA, 1BB341FE */
56 		 2.18694882948595424599e-02,	/* 3F9664F4, 8406D637 */
57 		 8.86323982359930005737e-03,	/* 3F8226E3, E96E8493 */
58 		 3.59207910759131235356e-03,	/* 3F6D6D22, C9560328 */
59 		 1.45620945432529025516e-03,	/* 3F57DBC8, FEE08315 */
60 		 5.88041240820264096874e-04,	/* 3F4344D8, F2F26501 */
61 		 2.46463134818469906812e-04,	/* 3F3026F7, 1A8D1068 */
62 		 7.81794442939557092300e-05,	/* 3F147E88, A03792A6 */
63 		 7.14072491382608190305e-05,	/* 3F12B80F, 32F0A7E9 */
64 		-1.85586374855275456654e-05,	/* BEF375CB, DB605373 */
65 		 2.59073051863633712884e-05,	/* 3EFB2A70, 74BF7AD4 */
66 /* one */	 1.00000000000000000000e+00,	/* 3FF00000, 00000000 */
67 /* pio4 */	 7.85398163397448278999e-01,	/* 3FE921FB, 54442D18 */
68 /* pio4lo */	 3.06161699786838301793e-17	/* 3C81A626, 33145C07 */
69 };
70 #define	one	xxx[13]
71 #define	pio4	xxx[14]
72 #define	pio4lo	xxx[15]
73 #define	T	xxx
74 /* INDENT ON */
75 
76 double
77 __kernel_tan(double x, double y, int iy) {
78 	double z, r, v, w, s;
79 	int32_t ix, hx;
80 
81 	GET_HIGH_WORD(hx,x);
82 	ix = hx & 0x7fffffff;			/* high word of |x| */
83 	if (ix >= 0x3FE59428) {	/* |x| >= 0.6744 */
84 		if (hx < 0) {
85 			x = -x;
86 			y = -y;
87 		}
88 		z = pio4 - x;
89 		w = pio4lo - y;
90 		x = z + w;
91 		y = 0.0;
92 	}
93 	z = x * x;
94 	w = z * z;
95 	/*
96 	 * Break x^5*(T[1]+x^2*T[2]+...) into
97 	 * x^5(T[1]+x^4*T[3]+...+x^20*T[11]) +
98 	 * x^5(x^2*(T[2]+x^4*T[4]+...+x^22*[T12]))
99 	 */
100 	r = T[1] + w * (T[3] + w * (T[5] + w * (T[7] + w * (T[9] +
101 		w * T[11]))));
102 	v = z * (T[2] + w * (T[4] + w * (T[6] + w * (T[8] + w * (T[10] +
103 		w * T[12])))));
104 	s = z * x;
105 	r = y + z * (s * (r + v) + y);
106 	r += T[0] * s;
107 	w = x + r;
108 	if (ix >= 0x3FE59428) {
109 		v = (double) iy;
110 		return (double) (1 - ((hx >> 30) & 2)) *
111 			(v - 2.0 * (x - (w * w / (w + v) - r)));
112 	}
113 	if (iy == 1)
114 		return w;
115 	else {
116 		/*
117 		 * if allow error up to 2 ulp, simply return
118 		 * -1.0 / (x+r) here
119 		 */
120 		/* compute -1.0 / (x+r) accurately */
121 		double a, t;
122 		z = w;
123 		SET_LOW_WORD(z,0);
124 		v = r - (z - x);	/* z+v = r+x */
125 		t = a = -1.0 / w;	/* a = -1.0/w */
126 		SET_LOW_WORD(t,0);
127 		s = 1.0 + t * z;
128 		return t + a * (s + t * v);
129 	}
130 }
131