/src/moddable/xs/tools/fdlibm/k_tan.c
Line | Count | Source |
1 | | /* |
2 | | * ==================================================== |
3 | | * Copyright 2004 Sun Microsystems, Inc. All Rights Reserved. |
4 | | * |
5 | | * Permission to use, copy, modify, and distribute this |
6 | | * software is freely granted, provided that this notice |
7 | | * is preserved. |
8 | | * ==================================================== |
9 | | */ |
10 | | |
11 | | /* __kernel_tan( x, y, k ) |
12 | | * kernel tan function on ~[-pi/4, pi/4] (except on -0), pi/4 ~ 0.7854 |
13 | | * Input x is assumed to be bounded by ~pi/4 in magnitude. |
14 | | * Input y is the tail of x. |
15 | | * Input k indicates whether tan (if k = 1) or -1/tan (if k = -1) is returned. |
16 | | * |
17 | | * Algorithm |
18 | | * 1. Since tan(-x) = -tan(x), we need only to consider positive x. |
19 | | * 2. Callers must return tan(-0) = -0 without calling here since our |
20 | | * odd polynomial is not evaluated in a way that preserves -0. |
21 | | * Callers may do the optimization tan(x) ~ x for tiny x. |
22 | | * 3. tan(x) is approximated by a odd polynomial of degree 27 on |
23 | | * [0,0.67434] |
24 | | * 3 27 |
25 | | * tan(x) ~ x + T1*x + ... + T13*x |
26 | | * where |
27 | | * |
28 | | * |tan(x) 2 4 26 | -59.2 |
29 | | * |----- - (1+T1*x +T2*x +.... +T13*x )| <= 2 |
30 | | * | x | |
31 | | * |
32 | | * Note: tan(x+y) = tan(x) + tan'(x)*y |
33 | | * ~ tan(x) + (1+x*x)*y |
34 | | * Therefore, for better accuracy in computing tan(x+y), let |
35 | | * 3 2 2 2 2 |
36 | | * r = x *(T2+x *(T3+x *(...+x *(T12+x *T13)))) |
37 | | * then |
38 | | * 3 2 |
39 | | * tan(x+y) = x + (T1*x + (x *(r+y)+y)) |
40 | | * |
41 | | * 4. For x in [0.67434,pi/4], let y = pi/4 - x, then |
42 | | * tan(x) = tan(pi/4-y) = (1-tan(y))/(1+tan(y)) |
43 | | * = 1 - 2*(tan(y) - (tan(y)^2)/(1+tan(y))) |
44 | | */ |
45 | | |
46 | | #include "math_private.h" |
47 | | static const double xxx[] = { |
48 | | 3.33333333333334091986e-01, /* 3FD55555, 55555563 */ |
49 | | 1.33333333333201242699e-01, /* 3FC11111, 1110FE7A */ |
50 | | 5.39682539762260521377e-02, /* 3FABA1BA, 1BB341FE */ |
51 | | 2.18694882948595424599e-02, /* 3F9664F4, 8406D637 */ |
52 | | 8.86323982359930005737e-03, /* 3F8226E3, E96E8493 */ |
53 | | 3.59207910759131235356e-03, /* 3F6D6D22, C9560328 */ |
54 | | 1.45620945432529025516e-03, /* 3F57DBC8, FEE08315 */ |
55 | | 5.88041240820264096874e-04, /* 3F4344D8, F2F26501 */ |
56 | | 2.46463134818469906812e-04, /* 3F3026F7, 1A8D1068 */ |
57 | | 7.81794442939557092300e-05, /* 3F147E88, A03792A6 */ |
58 | | 7.14072491382608190305e-05, /* 3F12B80F, 32F0A7E9 */ |
59 | | -1.85586374855275456654e-05, /* BEF375CB, DB605373 */ |
60 | | 2.59073051863633712884e-05, /* 3EFB2A70, 74BF7AD4 */ |
61 | | /* one */ 1.00000000000000000000e+00, /* 3FF00000, 00000000 */ |
62 | | /* pio4 */ 7.85398163397448278999e-01, /* 3FE921FB, 54442D18 */ |
63 | | /* pio4lo */ 3.06161699786838301793e-17 /* 3C81A626, 33145C07 */ |
64 | | }; |
65 | | #define one xxx[13] |
66 | 2.51k | #define pio4 xxx[14] |
67 | 2.51k | #define pio4lo xxx[15] |
68 | 125k | #define T xxx |
69 | | /* INDENT ON */ |
70 | | |
71 | | double |
72 | 9.67k | __kernel_tan(double x, double y, int iy) { |
73 | 9.67k | double z, r, v, w, s; |
74 | 9.67k | int32_t ix, hx; |
75 | | |
76 | 9.67k | GET_HIGH_WORD(hx,x); |
77 | 9.67k | ix = hx & 0x7fffffff; /* high word of |x| */ |
78 | 9.67k | if (ix >= 0x3FE59428) { /* |x| >= 0.6744 */ |
79 | 2.51k | if (hx < 0) { |
80 | 1.43k | x = -x; |
81 | 1.43k | y = -y; |
82 | 1.43k | } |
83 | 2.51k | z = pio4 - x; |
84 | 2.51k | w = pio4lo - y; |
85 | 2.51k | x = z + w; |
86 | 2.51k | y = 0.0; |
87 | 2.51k | } |
88 | 9.67k | z = x * x; |
89 | 9.67k | w = z * z; |
90 | | /* |
91 | | * Break x^5*(T[1]+x^2*T[2]+...) into |
92 | | * x^5(T[1]+x^4*T[3]+...+x^20*T[11]) + |
93 | | * x^5(x^2*(T[2]+x^4*T[4]+...+x^22*[T12])) |
94 | | */ |
95 | 9.67k | r = T[1] + w * (T[3] + w * (T[5] + w * (T[7] + w * (T[9] + |
96 | 9.67k | w * T[11])))); |
97 | 9.67k | v = z * (T[2] + w * (T[4] + w * (T[6] + w * (T[8] + w * (T[10] + |
98 | 9.67k | w * T[12]))))); |
99 | 9.67k | s = z * x; |
100 | 9.67k | r = y + z * (s * (r + v) + y); |
101 | 9.67k | r += T[0] * s; |
102 | 9.67k | w = x + r; |
103 | 9.67k | if (ix >= 0x3FE59428) { |
104 | 2.51k | v = (double) iy; |
105 | 2.51k | return (double) (1 - ((hx >> 30) & 2)) * |
106 | 2.51k | (v - 2.0 * (x - (w * w / (w + v) - r))); |
107 | 2.51k | } |
108 | 7.16k | if (iy == 1) |
109 | 3.72k | return w; |
110 | 3.44k | else { |
111 | | /* |
112 | | * if allow error up to 2 ulp, simply return |
113 | | * -1.0 / (x+r) here |
114 | | */ |
115 | | /* compute -1.0 / (x+r) accurately */ |
116 | 3.44k | double a, t; |
117 | 3.44k | z = w; |
118 | 3.44k | SET_LOW_WORD(z,0); |
119 | 3.44k | v = r - (z - x); /* z+v = r+x */ |
120 | 3.44k | t = a = -1.0 / w; /* a = -1.0/w */ |
121 | 3.44k | SET_LOW_WORD(t,0); |
122 | 3.44k | s = 1.0 + t * z; |
123 | 3.44k | return t + a * (s + t * v); |
124 | 3.44k | } |
125 | 7.16k | } |