xref: /netbsd/lib/libm/src/k_tan.c (revision bf9ec67e)
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 1.11 2002/05/26 22:01:54 wiz 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 static const double
55 one   =  1.00000000000000000000e+00, /* 0x3FF00000, 0x00000000 */
56 pio4  =  7.85398163397448278999e-01, /* 0x3FE921FB, 0x54442D18 */
57 pio4lo=  3.06161699786838301793e-17, /* 0x3C81A626, 0x33145C07 */
58 T[] =  {
59   3.33333333333334091986e-01, /* 0x3FD55555, 0x55555563 */
60   1.33333333333201242699e-01, /* 0x3FC11111, 0x1110FE7A */
61   5.39682539762260521377e-02, /* 0x3FABA1BA, 0x1BB341FE */
62   2.18694882948595424599e-02, /* 0x3F9664F4, 0x8406D637 */
63   8.86323982359930005737e-03, /* 0x3F8226E3, 0xE96E8493 */
64   3.59207910759131235356e-03, /* 0x3F6D6D22, 0xC9560328 */
65   1.45620945432529025516e-03, /* 0x3F57DBC8, 0xFEE08315 */
66   5.88041240820264096874e-04, /* 0x3F4344D8, 0xF2F26501 */
67   2.46463134818469906812e-04, /* 0x3F3026F7, 0x1A8D1068 */
68   7.81794442939557092300e-05, /* 0x3F147E88, 0xA03792A6 */
69   7.14072491382608190305e-05, /* 0x3F12B80F, 0x32F0A7E9 */
70  -1.85586374855275456654e-05, /* 0xBEF375CB, 0xDB605373 */
71   2.59073051863633712884e-05, /* 0x3EFB2A70, 0x74BF7AD4 */
72 };
73 
74 double
75 __kernel_tan(double x, double y, int iy)
76 {
77 	double z,r,v,w,s;
78 	int32_t ix,hx;
79 	GET_HIGH_WORD(hx,x);
80 	ix = hx&0x7fffffff;	/* high word of |x| */
81 	if(ix<0x3e300000)			/* x < 2**-28 */
82 	    {if((int)x==0) {			/* generate inexact */
83 	        u_int32_t low;
84 		GET_LOW_WORD(low,x);
85 		if(((ix|low)|(iy+1))==0) return one/fabs(x);
86 		else return (iy==1)? x: -one/x;
87 	    }
88 	    }
89 	if(ix>=0x3FE59428) { 			/* |x|>=0.6744 */
90 	    if(hx<0) {x = -x; y = -y;}
91 	    z = pio4-x;
92 	    w = pio4lo-y;
93 	    x = z+w; y = 0.0;
94 	}
95 	z	=  x*x;
96 	w 	=  z*z;
97     /* Break x^5*(T[1]+x^2*T[2]+...) into
98      *	  x^5(T[1]+x^4*T[3]+...+x^20*T[11]) +
99      *	  x^5(x^2*(T[2]+x^4*T[4]+...+x^22*[T12]))
100      */
101 	r = T[1]+w*(T[3]+w*(T[5]+w*(T[7]+w*(T[9]+w*T[11]))));
102 	v = z*(T[2]+w*(T[4]+w*(T[6]+w*(T[8]+w*(T[10]+w*T[12])))));
103 	s = z*x;
104 	r = y + z*(s*(r+v)+y);
105 	r += T[0]*s;
106 	w = x+r;
107 	if(ix>=0x3FE59428) {
108 	    v = (double)iy;
109 	    return (double)(1-((hx>>30)&2))*(v-2.0*(x-(w*w/(w+v)-r)));
110 	}
111 	if(iy==1) return w;
112 	else {		/* if allow error up to 2 ulp,
113 			   simply return -1.0/(x+r) here */
114      /*  compute -1.0/(x+r) accurately */
115 	    double a,t;
116 	    z  = w;
117 	    SET_LOW_WORD(z,0);
118 	    v  = r-(z - x); 	/* z+v = r+x */
119 	    t = a  = -1.0/w;	/* a = -1.0/w */
120 	    SET_LOW_WORD(t,0);
121 	    s  = 1.0+t*z;
122 	    return t+a*(s+t*v);
123 	}
124 }
125