1141296Sdas/* @(#)k_tan.c 1.5 04/04/22 SMI */
2141296Sdas
32116Sjkh/*
42116Sjkh * ====================================================
5129980Sdas * Copyright 2004 Sun Microsystems, Inc.  All Rights Reserved.
62116Sjkh *
72116Sjkh * Permission to use, copy, modify, and distribute this
88870Srgrimes * software is freely granted, provided that this notice
92116Sjkh * is preserved.
102116Sjkh * ====================================================
112116Sjkh */
122116Sjkh
13141296Sdas/* INDENT OFF */
14176451Sdas#include <sys/cdefs.h>
15176451Sdas__FBSDID("$FreeBSD$");
162116Sjkh
172116Sjkh/* __kernel_tan( x, y, k )
18151969Sbde * kernel tan function on ~[-pi/4, pi/4] (except on -0), pi/4 ~ 0.7854
192116Sjkh * Input x is assumed to be bounded by ~pi/4 in magnitude.
202116Sjkh * Input y is the tail of x.
21141296Sdas * Input k indicates whether tan (if k = 1) or -1/tan (if k = -1) is returned.
222116Sjkh *
232116Sjkh * Algorithm
248870Srgrimes *	1. Since tan(-x) = -tan(x), we need only to consider positive x.
25151969Sbde *	2. Callers must return tan(-0) = -0 without calling here since our
26151969Sbde *	   odd polynomial is not evaluated in a way that preserves -0.
27151969Sbde *	   Callers may do the optimization tan(x) ~ x for tiny x.
28141296Sdas *	3. tan(x) is approximated by a odd polynomial of degree 27 on
292116Sjkh *	   [0,0.67434]
302116Sjkh *		  	         3             27
312116Sjkh *	   	tan(x) ~ x + T1*x + ... + T13*x
322116Sjkh *	   where
338870Srgrimes *
342116Sjkh * 	        |tan(x)         2     4            26   |     -59.2
352116Sjkh * 	        |----- - (1+T1*x +T2*x +.... +T13*x    )| <= 2
368870Srgrimes * 	        |  x 					|
378870Srgrimes *
382116Sjkh *	   Note: tan(x+y) = tan(x) + tan'(x)*y
392116Sjkh *		          ~ tan(x) + (1+x*x)*y
408870Srgrimes *	   Therefore, for better accuracy in computing tan(x+y), let
412116Sjkh *		     3      2      2       2       2
422116Sjkh *		r = x *(T2+x *(T3+x *(...+x *(T12+x *T13))))
432116Sjkh *	   then
442116Sjkh *		 		    3    2
452116Sjkh *		tan(x+y) = x + (T1*x + (x *(r+y)+y))
462116Sjkh *
472116Sjkh *      4. For x in [0.67434,pi/4],  let y = pi/4 - x, then
482116Sjkh *		tan(x) = tan(pi/4-y) = (1-tan(y))/(1+tan(y))
492116Sjkh *		       = 1 - 2*(tan(y) - (tan(y)^2)/(1+tan(y)))
502116Sjkh */
512116Sjkh
522116Sjkh#include "math.h"
532116Sjkh#include "math_private.h"
54141296Sdasstatic const double xxx[] = {
55141296Sdas		 3.33333333333334091986e-01,	/* 3FD55555, 55555563 */
56141296Sdas		 1.33333333333201242699e-01,	/* 3FC11111, 1110FE7A */
57141296Sdas		 5.39682539762260521377e-02,	/* 3FABA1BA, 1BB341FE */
58141296Sdas		 2.18694882948595424599e-02,	/* 3F9664F4, 8406D637 */
59141296Sdas		 8.86323982359930005737e-03,	/* 3F8226E3, E96E8493 */
60141296Sdas		 3.59207910759131235356e-03,	/* 3F6D6D22, C9560328 */
61141296Sdas		 1.45620945432529025516e-03,	/* 3F57DBC8, FEE08315 */
62141296Sdas		 5.88041240820264096874e-04,	/* 3F4344D8, F2F26501 */
63141296Sdas		 2.46463134818469906812e-04,	/* 3F3026F7, 1A8D1068 */
64141296Sdas		 7.81794442939557092300e-05,	/* 3F147E88, A03792A6 */
65141296Sdas		 7.14072491382608190305e-05,	/* 3F12B80F, 32F0A7E9 */
66141296Sdas		-1.85586374855275456654e-05,	/* BEF375CB, DB605373 */
67141296Sdas		 2.59073051863633712884e-05,	/* 3EFB2A70, 74BF7AD4 */
68141296Sdas/* one */	 1.00000000000000000000e+00,	/* 3FF00000, 00000000 */
69141296Sdas/* pio4 */	 7.85398163397448278999e-01,	/* 3FE921FB, 54442D18 */
70141296Sdas/* pio4lo */	 3.06161699786838301793e-17	/* 3C81A626, 33145C07 */
712116Sjkh};
72141296Sdas#define	one	xxx[13]
73141296Sdas#define	pio4	xxx[14]
74141296Sdas#define	pio4lo	xxx[15]
75141296Sdas#define	T	xxx
76141296Sdas/* INDENT ON */
772116Sjkh
7897413Salfreddouble
79141296Sdas__kernel_tan(double x, double y, int iy) {
80141296Sdas	double z, r, v, w, s;
81141296Sdas	int32_t ix, hx;
82141296Sdas
832116Sjkh	GET_HIGH_WORD(hx,x);
84141296Sdas	ix = hx & 0x7fffffff;			/* high word of |x| */
85141296Sdas	if (ix >= 0x3FE59428) {	/* |x| >= 0.6744 */
86141296Sdas		if (hx < 0) {
87141296Sdas			x = -x;
88141296Sdas			y = -y;
89141296Sdas		}
90141296Sdas		z = pio4 - x;
91141296Sdas		w = pio4lo - y;
92141296Sdas		x = z + w;
93141296Sdas		y = 0.0;
942116Sjkh	}
95141296Sdas	z = x * x;
96141296Sdas	w = z * z;
97141296Sdas	/*
98141296Sdas	 * Break x^5*(T[1]+x^2*T[2]+...) into
99141296Sdas	 * x^5(T[1]+x^4*T[3]+...+x^20*T[11]) +
100141296Sdas	 * x^5(x^2*(T[2]+x^4*T[4]+...+x^22*[T12]))
101141296Sdas	 */
102141296Sdas	r = T[1] + w * (T[3] + w * (T[5] + w * (T[7] + w * (T[9] +
103141296Sdas		w * T[11]))));
104141296Sdas	v = z * (T[2] + w * (T[4] + w * (T[6] + w * (T[8] + w * (T[10] +
105141296Sdas		w * T[12])))));
106141296Sdas	s = z * x;
107141296Sdas	r = y + z * (s * (r + v) + y);
108141296Sdas	r += T[0] * s;
109141296Sdas	w = x + r;
110141296Sdas	if (ix >= 0x3FE59428) {
111141296Sdas		v = (double) iy;
112141296Sdas		return (double) (1 - ((hx >> 30) & 2)) *
113141296Sdas			(v - 2.0 * (x - (w * w / (w + v) - r)));
1142116Sjkh	}
115141296Sdas	if (iy == 1)
116141296Sdas		return w;
117141296Sdas	else {
118141296Sdas		/*
119141296Sdas		 * if allow error up to 2 ulp, simply return
120141296Sdas		 * -1.0 / (x+r) here
121141296Sdas		 */
122141296Sdas		/* compute -1.0 / (x+r) accurately */
123141296Sdas		double a, t;
124141296Sdas		z = w;
125141296Sdas		SET_LOW_WORD(z,0);
126141296Sdas		v = r - (z - x);	/* z+v = r+x */
127141296Sdas		t = a = -1.0 / w;	/* a = -1.0/w */
128141296Sdas		SET_LOW_WORD(t,0);
129141296Sdas		s = 1.0 + t * z;
130141296Sdas		return t + a * (s + t * v);
1312116Sjkh	}
1322116Sjkh}
133