1/* From: @(#)k_tan.c 1.5 04/04/22 SMI */
2
3/*
4 * ====================================================
5 * Copyright 2004 Sun Microsystems, Inc.  All Rights Reserved.
6 * Copyright (c) 2008 Steven G. Kargl, David Schultz, Bruce D. Evans.
7 *
8 * Permission to use, copy, modify, and distribute this
9 * software is freely granted, provided that this notice
10 * is preserved.
11 * ====================================================
12 */
13
14#include <sys/cdefs.h>
15__FBSDID("$FreeBSD$");
16
17/*
18 * ld80 version of k_tan.c.  See ../src/k_tan.c for most comments.
19 */
20
21#include "math.h"
22#include "math_private.h"
23
24/*
25 * Domain [-0.67434, 0.67434], range ~[-2.25e-22, 1.921e-22]
26 * |tan(x)/x - t(x)| < 2**-71.9
27 *
28 * See k_cosl.c for more details about the polynomial.
29 */
30#if defined(__amd64__) || defined(__i386__)
31/* Long double constants are slow on these arches, and broken on i386. */
32static const volatile double
33T3hi =  0.33333333333333331,		/*  0x15555555555555.0p-54 */
34T3lo =  1.8350121769317163e-17,		/*  0x15280000000000.0p-108 */
35T5hi =  0.13333333333333336,		/*  0x11111111111112.0p-55 */
36T5lo =  1.3051083651294260e-17,		/*  0x1e180000000000.0p-109 */
37T7hi =  0.053968253968250494,		/*  0x1ba1ba1ba1b827.0p-57 */
38T7lo =  3.1509625637859973e-18,		/*  0x1d100000000000.0p-111 */
39pio4_hi =  0.78539816339744828,		/*  0x1921fb54442d18.0p-53 */
40pio4_lo =  3.0628711372715500e-17,	/*  0x11a80000000000.0p-107 */
41pio4lo_hi = -1.2541394031670831e-20,	/* -0x1d9cceba3f91f2.0p-119 */
42pio4lo_lo =  6.1493048227390915e-37;	/*  0x1a280000000000.0p-173 */
43#define	T3	((long double)T3hi + T3lo)
44#define	T5	((long double)T5hi + T5lo)
45#define	T7	((long double)T7hi + T7lo)
46#define	pio4	((long double)pio4_hi + pio4_lo)
47#define	pio4lo	((long double)pio4lo_hi + pio4lo_lo)
48#else
49static const long double
50T3 =   0.333333333333333333180L,	/*  0xaaaaaaaaaaaaaaa5.0p-65 */
51T5 =   0.133333333333333372290L,	/*  0x88888888888893c3.0p-66 */
52T7 =   0.0539682539682504975744L,	/*  0xdd0dd0dd0dc13ba2.0p-68 */
53pio4 = 0.785398163397448309628L,	/*  0xc90fdaa22168c235.0p-64 */
54pio4lo = -1.25413940316708300586e-20L;	/* -0xece675d1fc8f8cbb.0p-130 */
55#endif
56
57static const double
58T9  =  0.021869488536312216,		/*  0x1664f4882cc1c2.0p-58 */
59T11 =  0.0088632355256619590,		/*  0x1226e355c17612.0p-59 */
60T13 =  0.0035921281113786528,		/*  0x1d6d3d185d7ff8.0p-61 */
61T15 =  0.0014558334756312418,		/*  0x17da354aa3f96b.0p-62 */
62T17 =  0.00059003538700862256,		/*  0x13559358685b83.0p-63 */
63T19 =  0.00023907843576635544,		/*  0x1f56242026b5be.0p-65 */
64T21 =  0.000097154625656538905,		/*  0x1977efc26806f4.0p-66 */
65T23 =  0.000038440165747303162,		/*  0x14275a09b3ceac.0p-67 */
66T25 =  0.000018082171885432524,		/*  0x12f5e563e5487e.0p-68 */
67T27 =  0.0000024196006108814377,	/*  0x144c0d80cc6896.0p-71 */
68T29 =  0.0000078293456938132840,	/*  0x106b59141a6cb3.0p-69 */
69T31 = -0.0000032609076735050182,	/* -0x1b5abef3ba4b59.0p-71 */
70T33 =  0.0000023261313142559411;	/*  0x13835436c0c87f.0p-71 */
71
72long double
73__kernel_tanl(long double x, long double y, int iy) {
74	long double z, r, v, w, s;
75	long double osign;
76	int i;
77
78	iy = (iy == 1 ? -1 : 1);	/* XXX recover original interface */
79	osign = (x >= 0 ? 1.0 : -1.0);	/* XXX slow, probably wrong for -0 */
80	if (fabsl(x) >= 0.67434) {
81		if (x < 0) {
82			x = -x;
83			y = -y;
84		}
85		z = pio4 - x;
86		w = pio4lo - y;
87		x = z + w;
88		y = 0.0;
89		i = 1;
90	} else
91		i = 0;
92	z = x * x;
93	w = z * z;
94	r = T5 + w * (T9 + w * (T13 + w * (T17 + w * (T21 +
95	    w * (T25 + w * (T29 + w * T33))))));
96	v = z * (T7 + w * (T11 + w * (T15 + w * (T19 + w * (T23 +
97	    w * (T27 + w * T31))))));
98	s = z * x;
99	r = y + z * (s * (r + v) + y);
100	r += T3 * s;
101	w = x + r;
102	if (i == 1) {
103		v = (long double) iy;
104		return osign *
105			(v - 2.0 * (x - (w * w / (w + v) - r)));
106	}
107	if (iy == 1)
108		return w;
109	else {
110		/*
111		 * if allow error up to 2 ulp, simply return
112		 * -1.0 / (x+r) here
113		 */
114		/* compute -1.0 / (x+r) accurately */
115		long double a, t;
116		z = w;
117		z = z + 0x1p32 - 0x1p32;
118		v = r - (z - x);	/* z+v = r+x */
119		t = a = -1.0 / w;	/* a = -1.0/w */
120		t = t + 0x1p32 - 0x1p32;
121		s = 1.0 + t * z;
122		return t + a * (s + t * v);
123	}
124}
125