Go to most recent revision | Details | Last modification | View Log | RSS feed
| Rev | Author | Line No. | Line |
|---|---|---|---|
| 3362 | Serge | 1 | |
| 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 | |||
| 14 | * kernel tan function on [-pi/4, pi/4], pi/4 ~ 0.7854 |
||
| 15 | * Input x is assumed to be bounded by ~pi/4 in magnitude. |
||
| 16 | * Input y is the tail of x. |
||
| 17 | * Input k indicates whether tan (if k=1) or |
||
| 18 | * -1/tan (if k= -1) is returned. |
||
| 19 | * |
||
| 20 | * Algorithm |
||
| 21 | * 1. Since tan(-x) = -tan(x), we need only to consider positive x. |
||
| 22 | * 2. if x < 2^-28 (hx<0x3e300000 0), return x with inexact if x!=0. |
||
| 23 | * 3. tan(x) is approximated by a odd polynomial of degree 27 on |
||
| 24 | * [0,0.67434] |
||
| 25 | * 3 27 |
||
| 26 | * tan(x) ~ x + T1*x + ... + T13*x |
||
| 27 | * where |
||
| 28 | * |
||
| 29 | * |tan(x) 2 4 26 | -59.2 |
||
| 30 | * |----- - (1+T1*x +T2*x +.... +T13*x )| <= 2 |
||
| 31 | * | x | |
||
| 32 | * |
||
| 33 | * Note: tan(x+y) = tan(x) + tan'(x)*y |
||
| 34 | * ~ tan(x) + (1+x*x)*y |
||
| 35 | * Therefore, for better accuracy in computing tan(x+y), let |
||
| 36 | * 3 2 2 2 2 |
||
| 37 | * r = x *(T2+x *(T3+x *(...+x *(T12+x *T13)))) |
||
| 38 | * then |
||
| 39 | * 3 2 |
||
| 40 | * tan(x+y) = x + (T1*x + (x *(r+y)+y)) |
||
| 41 | * |
||
| 42 | * 4. For x in [0.67434,pi/4], let y = pi/4 - x, then |
||
| 43 | * tan(x) = tan(pi/4-y) = (1-tan(y))/(1+tan(y)) |
||
| 44 | * = 1 - 2*(tan(y) - (tan(y)^2)/(1+tan(y))) |
||
| 45 | */ |
||
| 46 | |||
| 47 | |||
| 48 | |||
| 49 | |||
| 50 | |||
| 51 | |||
| 52 | static const double |
||
| 53 | #else |
||
| 54 | static double |
||
| 55 | #endif |
||
| 56 | one = 1.00000000000000000000e+00, /* 0x3FF00000, 0x00000000 */ |
||
| 57 | pio4 = 7.85398163397448278999e-01, /* 0x3FE921FB, 0x54442D18 */ |
||
| 58 | pio4lo= 3.06161699786838301793e-17, /* 0x3C81A626, 0x33145C07 */ |
||
| 59 | T[] = { |
||
| 60 | 3.33333333333334091986e-01, /* 0x3FD55555, 0x55555563 */ |
||
| 61 | 1.33333333333201242699e-01, /* 0x3FC11111, 0x1110FE7A */ |
||
| 62 | 5.39682539762260521377e-02, /* 0x3FABA1BA, 0x1BB341FE */ |
||
| 63 | 2.18694882948595424599e-02, /* 0x3F9664F4, 0x8406D637 */ |
||
| 64 | 8.86323982359930005737e-03, /* 0x3F8226E3, 0xE96E8493 */ |
||
| 65 | 3.59207910759131235356e-03, /* 0x3F6D6D22, 0xC9560328 */ |
||
| 66 | 1.45620945432529025516e-03, /* 0x3F57DBC8, 0xFEE08315 */ |
||
| 67 | 5.88041240820264096874e-04, /* 0x3F4344D8, 0xF2F26501 */ |
||
| 68 | 2.46463134818469906812e-04, /* 0x3F3026F7, 0x1A8D1068 */ |
||
| 69 | 7.81794442939557092300e-05, /* 0x3F147E88, 0xA03792A6 */ |
||
| 70 | 7.14072491382608190305e-05, /* 0x3F12B80F, 0x32F0A7E9 */ |
||
| 71 | -1.85586374855275456654e-05, /* 0xBEF375CB, 0xDB605373 */ |
||
| 72 | 2.59073051863633712884e-05, /* 0x3EFB2A70, 0x74BF7AD4 */ |
||
| 73 | }; |
||
| 74 | |||
| 75 | |||
| 76 | double __kernel_tan(double x, double y, int iy) |
||
| 77 | #else |
||
| 78 | double __kernel_tan(x, y, iy) |
||
| 79 | double x,y; int iy; |
||
| 80 | #endif |
||
| 81 | { |
||
| 82 | double z,r,v,w,s; |
||
| 83 | __int32_t ix,hx; |
||
| 84 | GET_HIGH_WORD(hx,x); |
||
| 85 | ix = hx&0x7fffffff; /* high word of |x| */ |
||
| 86 | if(ix<0x3e300000) /* x < 2**-28 */ |
||
| 87 | {if((int)x==0) { /* generate inexact */ |
||
| 88 | __uint32_t low; |
||
| 89 | GET_LOW_WORD(low,x); |
||
| 90 | if(((ix|low)|(iy+1))==0) return one/fabs(x); |
||
| 91 | else return (iy==1)? x: -one/x; |
||
| 92 | } |
||
| 93 | } |
||
| 94 | if(ix>=0x3FE59428) { /* |x|>=0.6744 */ |
||
| 95 | if(hx<0) {x = -x; y = -y;} |
||
| 96 | z = pio4-x; |
||
| 97 | w = pio4lo-y; |
||
| 98 | x = z+w; y = 0.0; |
||
| 99 | } |
||
| 100 | z = x*x; |
||
| 101 | w = z*z; |
||
| 102 | /* Break x^5*(T[1]+x^2*T[2]+...) into |
||
| 103 | * x^5(T[1]+x^4*T[3]+...+x^20*T[11]) + |
||
| 104 | * x^5(x^2*(T[2]+x^4*T[4]+...+x^22*[T12])) |
||
| 105 | */ |
||
| 106 | r = T[1]+w*(T[3]+w*(T[5]+w*(T[7]+w*(T[9]+w*T[11])))); |
||
| 107 | v = z*(T[2]+w*(T[4]+w*(T[6]+w*(T[8]+w*(T[10]+w*T[12]))))); |
||
| 108 | s = z*x; |
||
| 109 | r = y + z*(s*(r+v)+y); |
||
| 110 | r += T[0]*s; |
||
| 111 | w = x+r; |
||
| 112 | if(ix>=0x3FE59428) { |
||
| 113 | v = (double)iy; |
||
| 114 | return (double)(1-((hx>>30)&2))*(v-2.0*(x-(w*w/(w+v)-r))); |
||
| 115 | } |
||
| 116 | if(iy==1) return w; |
||
| 117 | else { /* if allow error up to 2 ulp, |
||
| 118 | simply return -1.0/(x+r) here */ |
||
| 119 | /* compute -1.0/(x+r) accurately */ |
||
| 120 | double a,t; |
||
| 121 | z = w; |
||
| 122 | SET_LOW_WORD(z,0); |
||
| 123 | v = r-(z - x); /* z+v = r+x */ |
||
| 124 | t = a = -1.0/w; /* a = -1.0/w */ |
||
| 125 | SET_LOW_WORD(t,0); |
||
| 126 | s = 1.0+t*z; |
||
| 127 | return t+a*(s+t*v); |
||
| 128 | } |
||
| 129 | } |
||
| 130 | |||
| 131 | |||
| 132 |