vx32

Local 9vx git repository for patches.
git clone git://r-36.net/vx32
Log | Files | Refs

k_tan.c (4269B)


      1 /* @(#)k_tan.c 5.1 93/09/24 */
      2 /*
      3  * ====================================================
      4  * Copyright 2004 Sun Microsystems, Inc.  All Rights Reserved.
      5  *
      6  * Permission to use, copy, modify, and distribute this
      7  * software is freely granted, provided that this notice
      8  * is preserved.
      9  * ====================================================
     10  */
     11 
     12 #ifndef lint
     13 static char rcsid[] = "$FreeBSD: src/lib/msun/src/k_tan.c,v 1.9 2004/06/02 04:39:29 das Exp $";
     14 #endif
     15 
     16 /* __kernel_tan( x, y, k )
     17  * kernel tan function on [-pi/4, pi/4], pi/4 ~ 0.7854
     18  * Input x is assumed to be bounded by ~pi/4 in magnitude.
     19  * Input y is the tail of x.
     20  * Input k indicates whether tan (if k=1) or
     21  * -1/tan (if k= -1) is returned.
     22  *
     23  * Algorithm
     24  *	1. Since tan(-x) = -tan(x), we need only to consider positive x.
     25  *	2. if x < 2^-28 (hx<0x3e300000 0), return x with inexact if x!=0.
     26  *	3. tan(x) is approximated by an odd polynomial of degree 27 on
     27  *	   [0,0.67434]
     28  *		  	         3             27
     29  *	   	tan(x) ~ x + T1*x + ... + T13*x
     30  *	   where
     31  *
     32  * 	        |tan(x)         2     4            26   |     -59.2
     33  * 	        |----- - (1+T1*x +T2*x +.... +T13*x    )| <= 2
     34  * 	        |  x 					|
     35  *
     36  *	   Note: tan(x+y) = tan(x) + tan'(x)*y
     37  *		          ~ tan(x) + (1+x*x)*y
     38  *	   Therefore, for better accuracy in computing tan(x+y), let
     39  *		     3      2      2       2       2
     40  *		r = x *(T2+x *(T3+x *(...+x *(T12+x *T13))))
     41  *	   then
     42  *		 		    3    2
     43  *		tan(x+y) = x + (T1*x + (x *(r+y)+y))
     44  *
     45  *      4. For x in [0.67434,pi/4],  let y = pi/4 - x, then
     46  *		tan(x) = tan(pi/4-y) = (1-tan(y))/(1+tan(y))
     47  *		       = 1 - 2*(tan(y) - (tan(y)^2)/(1+tan(y)))
     48  */
     49 
     50 #include "math.h"
     51 #include "math_private.h"
     52 static const double
     53 one   =  1.00000000000000000000e+00, /* 0x3FF00000, 0x00000000 */
     54 pio4  =  7.85398163397448278999e-01, /* 0x3FE921FB, 0x54442D18 */
     55 pio4lo=  3.06161699786838301793e-17, /* 0x3C81A626, 0x33145C07 */
     56 T[] =  {
     57   3.33333333333334091986e-01, /* 0x3FD55555, 0x55555563 */
     58   1.33333333333201242699e-01, /* 0x3FC11111, 0x1110FE7A */
     59   5.39682539762260521377e-02, /* 0x3FABA1BA, 0x1BB341FE */
     60   2.18694882948595424599e-02, /* 0x3F9664F4, 0x8406D637 */
     61   8.86323982359930005737e-03, /* 0x3F8226E3, 0xE96E8493 */
     62   3.59207910759131235356e-03, /* 0x3F6D6D22, 0xC9560328 */
     63   1.45620945432529025516e-03, /* 0x3F57DBC8, 0xFEE08315 */
     64   5.88041240820264096874e-04, /* 0x3F4344D8, 0xF2F26501 */
     65   2.46463134818469906812e-04, /* 0x3F3026F7, 0x1A8D1068 */
     66   7.81794442939557092300e-05, /* 0x3F147E88, 0xA03792A6 */
     67   7.14072491382608190305e-05, /* 0x3F12B80F, 0x32F0A7E9 */
     68  -1.85586374855275456654e-05, /* 0xBEF375CB, 0xDB605373 */
     69   2.59073051863633712884e-05, /* 0x3EFB2A70, 0x74BF7AD4 */
     70 };
     71 
     72 double
     73 __kernel_tan(double x, double y, int iy)
     74 {
     75 	double z,r,v,w,s;
     76 	int32_t ix,hx;
     77 	GET_HIGH_WORD(hx,x);
     78 	ix = hx&0x7fffffff;	/* high word of |x| */
     79 	if(ix<0x3e300000) {			/* x < 2**-28 */
     80 		if ((int) x == 0) {		/* generate inexact */
     81 			u_int32_t low;
     82 			GET_LOW_WORD(low,x);
     83 			if (((ix | low) | (iy + 1)) == 0)
     84 				return one / fabs(x);
     85 			else {
     86 				if (iy == 1)
     87 					return x;
     88 				else {	/* compute -1 / (x+y) carefully */
     89 					double a, t;
     90 
     91 					z = w = x + y;
     92 					SET_LOW_WORD(z, 0);
     93 					v = y - (z - x);
     94 					t = a = -one / w;
     95 					SET_LOW_WORD(t, 0);
     96 					s = one + t * z;
     97 					return t + a * (s + t * v);
     98 				}
     99 			}
    100 		}
    101 	}
    102 	if(ix>=0x3FE59428) { 			/* |x|>=0.6744 */
    103 	    if(hx<0) {x = -x; y = -y;}
    104 	    z = pio4-x;
    105 	    w = pio4lo-y;
    106 	    x = z+w; y = 0.0;
    107 	}
    108 	z	=  x*x;
    109 	w 	=  z*z;
    110     /* Break x^5*(T[1]+x^2*T[2]+...) into
    111      *	  x^5(T[1]+x^4*T[3]+...+x^20*T[11]) +
    112      *	  x^5(x^2*(T[2]+x^4*T[4]+...+x^22*[T12]))
    113      */
    114 	r = T[1]+w*(T[3]+w*(T[5]+w*(T[7]+w*(T[9]+w*T[11]))));
    115 	v = z*(T[2]+w*(T[4]+w*(T[6]+w*(T[8]+w*(T[10]+w*T[12])))));
    116 	s = z*x;
    117 	r = y + z*(s*(r+v)+y);
    118 	r += T[0]*s;
    119 	w = x+r;
    120 	if(ix>=0x3FE59428) {
    121 	    v = (double)iy;
    122 	    return (double)(1-((hx>>30)&2))*(v-2.0*(x-(w*w/(w+v)-r)));
    123 	}
    124 	if(iy==1) return w;
    125 	else {		/* if allow error up to 2 ulp,
    126 			   simply return -1.0/(x+r) here */
    127      /*  compute -1.0/(x+r) accurately */
    128 	    double a,t;
    129 	    z  = w;
    130 	    SET_LOW_WORD(z,0);
    131 	    v  = r-(z - x); 	/* z+v = r+x */
    132 	    t = a  = -1.0/w;	/* a = -1.0/w */
    133 	    SET_LOW_WORD(t,0);
    134 	    s  = 1.0+t*z;
    135 	    return t+a*(s+t*v);
    136 	}
    137 }