src/libm/e_pow.c
author Sam Lantinga <slouken@libsdl.org>
Tue, 21 Nov 2017 21:51:33 -0800
changeset 11711 8a982ed61896
parent 11683 48bcba563d9c
child 12420 4a6c91d9cc33
permissions -rw-r--r--
Disabled spurious Visual Studio warnings in the uClibc math code
slouken@2756
     1
/*
slouken@2756
     2
 * ====================================================
slouken@2756
     3
 * Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved.
slouken@2756
     4
 *
slouken@2756
     5
 * Developed at SunPro, a Sun Microsystems, Inc. business.
slouken@2756
     6
 * Permission to use, copy, modify, and distribute this
slouken@2756
     7
 * software is freely granted, provided that this notice
slouken@2756
     8
 * is preserved.
slouken@2756
     9
 * ====================================================
slouken@2756
    10
 */
slouken@2756
    11
slouken@2756
    12
/* __ieee754_pow(x,y) return x**y
slouken@2756
    13
 *
slouken@2756
    14
 *		      n
slouken@2756
    15
 * Method:  Let x =  2   * (1+f)
slouken@2756
    16
 *	1. Compute and return log2(x) in two pieces:
slouken@2756
    17
 *		log2(x) = w1 + w2,
slouken@2756
    18
 *	   where w1 has 53-24 = 29 bit trailing zeros.
slouken@2756
    19
 *	2. Perform y*log2(x) = n+y' by simulating muti-precision
slouken@2756
    20
 *	   arithmetic, where |y'|<=0.5.
slouken@2756
    21
 *	3. Return x**y = 2**n*exp(y'*log2)
slouken@2756
    22
 *
slouken@2756
    23
 * Special cases:
slouken@11683
    24
 *	1.  +-1 ** anything  is 1.0
slouken@11683
    25
 *	2.  +-1 ** +-INF     is 1.0
slouken@11683
    26
 *	3.  (anything) ** 0  is 1
slouken@11683
    27
 *	4.  (anything) ** 1  is itself
slouken@11683
    28
 *	5.  (anything) ** NAN is NAN
slouken@11683
    29
 *	6.  NAN ** (anything except 0) is NAN
slouken@11683
    30
 *	7.  +-(|x| > 1) **  +INF is +INF
slouken@11683
    31
 *	8.  +-(|x| > 1) **  -INF is +0
slouken@11683
    32
 *	9.  +-(|x| < 1) **  +INF is +0
slouken@11683
    33
 *	10  +-(|x| < 1) **  -INF is +INF
slouken@11683
    34
 *	11. +0 ** (+anything except 0, NAN)               is +0
slouken@11683
    35
 *	12. -0 ** (+anything except 0, NAN, odd integer)  is +0
slouken@11683
    36
 *	13. +0 ** (-anything except 0, NAN)               is +INF
slouken@11683
    37
 *	14. -0 ** (-anything except 0, NAN, odd integer)  is +INF
slouken@11683
    38
 *	15. -0 ** (odd integer) = -( +0 ** (odd integer) )
slouken@11683
    39
 *	16. +INF ** (+anything except 0,NAN) is +INF
slouken@11683
    40
 *	17. +INF ** (-anything except 0,NAN) is +0
slouken@11683
    41
 *	18. -INF ** (anything)  = -0 ** (-anything)
slouken@11683
    42
 *	19. (-anything) ** (integer) is (-1)**(integer)*(+anything**integer)
slouken@11683
    43
 *	20. (-anything except 0 and inf) ** (non-integer) is NAN
slouken@2756
    44
 *
slouken@2756
    45
 * Accuracy:
slouken@2756
    46
 *	pow(x,y) returns x**y nearly rounded. In particular
slouken@2756
    47
 *			pow(integer,integer)
slouken@2756
    48
 *	always returns the correct integer provided it is
slouken@2756
    49
 *	representable.
slouken@2756
    50
 *
slouken@2756
    51
 * Constants :
slouken@2756
    52
 * The hexadecimal values are the intended ones for the following
slouken@2756
    53
 * constants. The decimal values may be used, provided that the
slouken@2756
    54
 * compiler will convert from decimal to binary accurately enough
slouken@2756
    55
 * to produce the hexadecimal values shown.
slouken@2756
    56
 */
slouken@2756
    57
slouken@6044
    58
#include "math_libm.h"
slouken@2756
    59
#include "math_private.h"
slouken@2756
    60
slouken@11711
    61
#if defined(_MSC_VER)           /* Handle Microsoft VC++ compiler specifics. */
slouken@11711
    62
/* C4756: overflow in constant arithmetic */
slouken@11711
    63
#pragma warning ( disable : 4756 )
slouken@11711
    64
#endif
slouken@11711
    65
slouken@11683
    66
static const double
slouken@11683
    67
bp[] = {1.0, 1.5,},
slouken@11683
    68
dp_h[] = { 0.0, 5.84962487220764160156e-01,}, /* 0x3FE2B803, 0x40000000 */
slouken@11683
    69
dp_l[] = { 0.0, 1.35003920212974897128e-08,}, /* 0x3E4CFDEB, 0x43CFD006 */
slouken@11683
    70
zero    =  0.0,
slouken@11683
    71
one	=  1.0,
slouken@11683
    72
two	=  2.0,
slouken@11683
    73
two53	=  9007199254740992.0,	/* 0x43400000, 0x00000000 */
slouken@11683
    74
huge	=  1.0e300,
slouken@11683
    75
tiny    =  1.0e-300,
slouken@11683
    76
	/* poly coefs for (3/2)*(log(x)-2s-2/3*s**3 */
slouken@11683
    77
L1  =  5.99999999999994648725e-01, /* 0x3FE33333, 0x33333303 */
slouken@11683
    78
L2  =  4.28571428578550184252e-01, /* 0x3FDB6DB6, 0xDB6FABFF */
slouken@11683
    79
L3  =  3.33333329818377432918e-01, /* 0x3FD55555, 0x518F264D */
slouken@11683
    80
L4  =  2.72728123808534006489e-01, /* 0x3FD17460, 0xA91D4101 */
slouken@11683
    81
L5  =  2.30660745775561754067e-01, /* 0x3FCD864A, 0x93C9DB65 */
slouken@11683
    82
L6  =  2.06975017800338417784e-01, /* 0x3FCA7E28, 0x4A454EEF */
slouken@11683
    83
P1   =  1.66666666666666019037e-01, /* 0x3FC55555, 0x5555553E */
slouken@11683
    84
P2   = -2.77777777770155933842e-03, /* 0xBF66C16C, 0x16BEBD93 */
slouken@11683
    85
P3   =  6.61375632143793436117e-05, /* 0x3F11566A, 0xAF25DE2C */
slouken@11683
    86
P4   = -1.65339022054652515390e-06, /* 0xBEBBBD41, 0xC5D26BF1 */
slouken@11683
    87
P5   =  4.13813679705723846039e-08, /* 0x3E663769, 0x72BEA4D0 */
slouken@11683
    88
lg2  =  6.93147180559945286227e-01, /* 0x3FE62E42, 0xFEFA39EF */
slouken@11683
    89
lg2_h  =  6.93147182464599609375e-01, /* 0x3FE62E43, 0x00000000 */
slouken@11683
    90
lg2_l  = -1.90465429995776804525e-09, /* 0xBE205C61, 0x0CA86C39 */
slouken@11683
    91
ovt =  8.0085662595372944372e-0017, /* -(1024-log2(ovfl+.5ulp)) */
slouken@11683
    92
cp    =  9.61796693925975554329e-01, /* 0x3FEEC709, 0xDC3A03FD =2/(3ln2) */
slouken@11683
    93
cp_h  =  9.61796700954437255859e-01, /* 0x3FEEC709, 0xE0000000 =(float)cp */
slouken@11683
    94
cp_l  = -7.02846165095275826516e-09, /* 0xBE3E2FE0, 0x145B01F5 =tail of cp_h*/
slouken@11683
    95
ivln2    =  1.44269504088896338700e+00, /* 0x3FF71547, 0x652B82FE =1/ln2 */
slouken@11683
    96
ivln2_h  =  1.44269502162933349609e+00, /* 0x3FF71547, 0x60000000 =24b 1/ln2*/
slouken@11683
    97
ivln2_l  =  1.92596299112661746887e-08; /* 0x3E54AE0B, 0xF85DDF44 =1/ln2 tail*/
slouken@11683
    98
slouken@11683
    99
double attribute_hidden __ieee754_pow(double x, double y)
slouken@11683
   100
{
slouken@11683
   101
	double z,ax,z_h,z_l,p_h,p_l;
slouken@11683
   102
	double y1,t1,t2,r,s,t,u,v,w;
slouken@11683
   103
	int32_t i,j,k,yisint,n;
slouken@11683
   104
	int32_t hx,hy,ix,iy;
slouken@11683
   105
	u_int32_t lx,ly;
slouken@11683
   106
slouken@11683
   107
	EXTRACT_WORDS(hx,lx,x);
slouken@11683
   108
    /* x==1: 1**y = 1 (even if y is NaN) */
slouken@11683
   109
	if (hx==0x3ff00000 && lx==0) {
slouken@11683
   110
		return x;
slouken@11683
   111
	}
slouken@11683
   112
	ix = hx&0x7fffffff;
slouken@11683
   113
slouken@11683
   114
	EXTRACT_WORDS(hy,ly,y);
slouken@11683
   115
	iy = hy&0x7fffffff;
slouken@11683
   116
slouken@11683
   117
    /* y==zero: x**0 = 1 */
slouken@11683
   118
	if((iy|ly)==0) return one;
slouken@11683
   119
slouken@11683
   120
    /* +-NaN return x+y */
slouken@11683
   121
	if(ix > 0x7ff00000 || ((ix==0x7ff00000)&&(lx!=0)) ||
slouken@11683
   122
	   iy > 0x7ff00000 || ((iy==0x7ff00000)&&(ly!=0)))
slouken@11683
   123
		return x+y;
slouken@11683
   124
slouken@11683
   125
    /* determine if y is an odd int when x < 0
slouken@11683
   126
     * yisint = 0	... y is not an integer
slouken@11683
   127
     * yisint = 1	... y is an odd int
slouken@11683
   128
     * yisint = 2	... y is an even int
slouken@11683
   129
     */
slouken@11683
   130
	yisint  = 0;
slouken@11683
   131
	if(hx<0) {
slouken@11683
   132
	    if(iy>=0x43400000) yisint = 2; /* even integer y */
slouken@11683
   133
	    else if(iy>=0x3ff00000) {
slouken@11683
   134
		k = (iy>>20)-0x3ff;	   /* exponent */
slouken@11683
   135
		if(k>20) {
slouken@11683
   136
		    j = ly>>(52-k);
slouken@11683
   137
		    if((j<<(52-k))==ly) yisint = 2-(j&1);
slouken@11683
   138
		} else if(ly==0) {
slouken@11683
   139
		    j = iy>>(20-k);
slouken@11683
   140
		    if((j<<(20-k))==iy) yisint = 2-(j&1);
slouken@11683
   141
		}
slouken@11683
   142
	    }
slouken@11683
   143
	}
slouken@11683
   144
slouken@11683
   145
    /* special value of y */
slouken@11683
   146
	if(ly==0) {
slouken@11683
   147
	    if (iy==0x7ff00000) {       /* y is +-inf */
slouken@11683
   148
	        if (((ix-0x3ff00000)|lx)==0)
slouken@11683
   149
		    return one;	        /* +-1**+-inf is 1 (yes, weird rule) */
slouken@11683
   150
	        if (ix >= 0x3ff00000)   /* (|x|>1)**+-inf = inf,0 */
slouken@11683
   151
		    return (hy>=0) ? y : zero;
slouken@11683
   152
	        /* (|x|<1)**-,+inf = inf,0 */
slouken@11683
   153
		return (hy<0) ? -y : zero;
slouken@11683
   154
	    }
slouken@11683
   155
	    if(iy==0x3ff00000) {	/* y is  +-1 */
slouken@11683
   156
		if(hy<0) return one/x; else return x;
slouken@11683
   157
	    }
slouken@11683
   158
	    if(hy==0x40000000) return x*x; /* y is  2 */
slouken@11683
   159
	    if(hy==0x3fe00000) {	/* y is  0.5 */
slouken@11683
   160
		if(hx>=0)	/* x >= +0 */
slouken@11683
   161
		    return __ieee754_sqrt(x);
slouken@11683
   162
	    }
slouken@11683
   163
	}
slouken@11683
   164
slouken@11683
   165
	ax   = fabs(x);
slouken@11683
   166
    /* special value of x */
slouken@11683
   167
	if(lx==0) {
slouken@11683
   168
	    if(ix==0x7ff00000||ix==0||ix==0x3ff00000){
slouken@11683
   169
		z = ax;			/*x is +-0,+-inf,+-1*/
slouken@11683
   170
		if(hy<0) z = one/z;	/* z = (1/|x|) */
slouken@11683
   171
		if(hx<0) {
slouken@11683
   172
		    if(((ix-0x3ff00000)|yisint)==0) {
slouken@11683
   173
			z = (z-z)/(z-z); /* (-1)**non-int is NaN */
slouken@11683
   174
		    } else if(yisint==1)
slouken@11683
   175
			z = -z;		/* (x<0)**odd = -(|x|**odd) */
slouken@11683
   176
		}
slouken@11683
   177
		return z;
slouken@11683
   178
	    }
slouken@11683
   179
	}
slouken@11683
   180
slouken@11683
   181
    /* (x<0)**(non-int) is NaN */
slouken@11683
   182
	if(((((u_int32_t)hx>>31)-1)|yisint)==0) return (x-x)/(x-x);
slouken@11683
   183
slouken@11683
   184
    /* |y| is huge */
slouken@11683
   185
	if(iy>0x41e00000) { /* if |y| > 2**31 */
slouken@11683
   186
	    if(iy>0x43f00000){	/* if |y| > 2**64, must o/uflow */
slouken@11683
   187
		if(ix<=0x3fefffff) return (hy<0)? huge*huge:tiny*tiny;
slouken@11683
   188
		if(ix>=0x3ff00000) return (hy>0)? huge*huge:tiny*tiny;
slouken@11683
   189
	    }
slouken@11683
   190
	/* over/underflow if x is not close to one */
slouken@11683
   191
	    if(ix<0x3fefffff) return (hy<0)? huge*huge:tiny*tiny;
slouken@11683
   192
	    if(ix>0x3ff00000) return (hy>0)? huge*huge:tiny*tiny;
slouken@11683
   193
	/* now |1-x| is tiny <= 2**-20, suffice to compute
slouken@11683
   194
	   log(x) by x-x^2/2+x^3/3-x^4/4 */
slouken@11683
   195
	    t = x-1;		/* t has 20 trailing zeros */
slouken@11683
   196
	    w = (t*t)*(0.5-t*(0.3333333333333333333333-t*0.25));
slouken@11683
   197
	    u = ivln2_h*t;	/* ivln2_h has 21 sig. bits */
slouken@11683
   198
	    v = t*ivln2_l-w*ivln2;
slouken@11683
   199
	    t1 = u+v;
slouken@11683
   200
	    SET_LOW_WORD(t1,0);
slouken@11683
   201
	    t2 = v-(t1-u);
slouken@11683
   202
	} else {
slouken@11683
   203
	    double s2,s_h,s_l,t_h,t_l;
slouken@11683
   204
	    n = 0;
slouken@11683
   205
	/* take care subnormal number */
slouken@11683
   206
	    if(ix<0x00100000)
slouken@11683
   207
		{ax *= two53; n -= 53; GET_HIGH_WORD(ix,ax); }
slouken@11683
   208
	    n  += ((ix)>>20)-0x3ff;
slouken@11683
   209
	    j  = ix&0x000fffff;
slouken@11683
   210
	/* determine interval */
slouken@11683
   211
	    ix = j|0x3ff00000;		/* normalize ix */
slouken@11683
   212
	    if(j<=0x3988E) k=0;		/* |x|<sqrt(3/2) */
slouken@11683
   213
	    else if(j<0xBB67A) k=1;	/* |x|<sqrt(3)   */
slouken@11683
   214
	    else {k=0;n+=1;ix -= 0x00100000;}
slouken@11683
   215
	    SET_HIGH_WORD(ax,ix);
slouken@11683
   216
slouken@11683
   217
	/* compute s = s_h+s_l = (x-1)/(x+1) or (x-1.5)/(x+1.5) */
slouken@11683
   218
	    u = ax-bp[k];		/* bp[0]=1.0, bp[1]=1.5 */
slouken@11683
   219
	    v = one/(ax+bp[k]);
slouken@11683
   220
	    s = u*v;
slouken@11683
   221
	    s_h = s;
slouken@11683
   222
	    SET_LOW_WORD(s_h,0);
slouken@11683
   223
	/* t_h=ax+bp[k] High */
slouken@11683
   224
	    t_h = zero;
slouken@11683
   225
	    SET_HIGH_WORD(t_h,((ix>>1)|0x20000000)+0x00080000+(k<<18));
slouken@11683
   226
	    t_l = ax - (t_h-bp[k]);
slouken@11683
   227
	    s_l = v*((u-s_h*t_h)-s_h*t_l);
slouken@11683
   228
	/* compute log(ax) */
slouken@11683
   229
	    s2 = s*s;
slouken@11683
   230
	    r = s2*s2*(L1+s2*(L2+s2*(L3+s2*(L4+s2*(L5+s2*L6)))));
slouken@11683
   231
	    r += s_l*(s_h+s);
slouken@11683
   232
	    s2  = s_h*s_h;
slouken@11683
   233
	    t_h = 3.0+s2+r;
slouken@11683
   234
	    SET_LOW_WORD(t_h,0);
slouken@11683
   235
	    t_l = r-((t_h-3.0)-s2);
slouken@11683
   236
	/* u+v = s*(1+...) */
slouken@11683
   237
	    u = s_h*t_h;
slouken@11683
   238
	    v = s_l*t_h+t_l*s;
slouken@11683
   239
	/* 2/(3log2)*(s+...) */
slouken@11683
   240
	    p_h = u+v;
slouken@11683
   241
	    SET_LOW_WORD(p_h,0);
slouken@11683
   242
	    p_l = v-(p_h-u);
slouken@11683
   243
	    z_h = cp_h*p_h;		/* cp_h+cp_l = 2/(3*log2) */
slouken@11683
   244
	    z_l = cp_l*p_h+p_l*cp+dp_l[k];
slouken@11683
   245
	/* log2(ax) = (s+..)*2/(3*log2) = n + dp_h + z_h + z_l */
slouken@11683
   246
	    t = (double)n;
slouken@11683
   247
	    t1 = (((z_h+z_l)+dp_h[k])+t);
slouken@11683
   248
	    SET_LOW_WORD(t1,0);
slouken@11683
   249
	    t2 = z_l-(((t1-t)-dp_h[k])-z_h);
slouken@11683
   250
	}
slouken@11683
   251
slouken@11683
   252
	s = one; /* s (sign of result -ve**odd) = -1 else = 1 */
slouken@11683
   253
	if(((((u_int32_t)hx>>31)-1)|(yisint-1))==0)
slouken@11683
   254
	    s = -one;/* (-ve)**(odd int) */
slouken@11683
   255
slouken@11683
   256
    /* split up y into y1+y2 and compute (y1+y2)*(t1+t2) */
slouken@11683
   257
	y1  = y;
slouken@11683
   258
	SET_LOW_WORD(y1,0);
slouken@11683
   259
	p_l = (y-y1)*t1+y*t2;
slouken@11683
   260
	p_h = y1*t1;
slouken@11683
   261
	z = p_l+p_h;
slouken@11683
   262
	EXTRACT_WORDS(j,i,z);
slouken@11683
   263
	if (j>=0x40900000) {				/* z >= 1024 */
slouken@11683
   264
	    if(((j-0x40900000)|i)!=0)			/* if z > 1024 */
slouken@11683
   265
		return s*huge*huge;			/* overflow */
slouken@11683
   266
	    else {
slouken@11683
   267
		if(p_l+ovt>z-p_h) return s*huge*huge;	/* overflow */
slouken@11683
   268
	    }
slouken@11683
   269
	} else if((j&0x7fffffff)>=0x4090cc00 ) {	/* z <= -1075 */
slouken@11683
   270
	    if(((j-0xc090cc00)|i)!=0) 		/* z < -1075 */
slouken@11683
   271
		return s*tiny*tiny;		/* underflow */
slouken@11683
   272
	    else {
slouken@11683
   273
		if(p_l<=z-p_h) return s*tiny*tiny;	/* underflow */
slouken@11683
   274
	    }
slouken@11683
   275
	}
slouken@11683
   276
    /*
slouken@11683
   277
     * compute 2**(p_h+p_l)
slouken@11683
   278
     */
slouken@11683
   279
	i = j&0x7fffffff;
slouken@11683
   280
	k = (i>>20)-0x3ff;
slouken@11683
   281
	n = 0;
slouken@11683
   282
	if(i>0x3fe00000) {		/* if |z| > 0.5, set n = [z+0.5] */
slouken@11683
   283
	    n = j+(0x00100000>>(k+1));
slouken@11683
   284
	    k = ((n&0x7fffffff)>>20)-0x3ff;	/* new k for n */
slouken@11683
   285
	    t = zero;
slouken@11683
   286
	    SET_HIGH_WORD(t,n&~(0x000fffff>>k));
slouken@11683
   287
	    n = ((n&0x000fffff)|0x00100000)>>(20-k);
slouken@11683
   288
	    if(j<0) n = -n;
slouken@11683
   289
	    p_h -= t;
slouken@11683
   290
	}
slouken@11683
   291
	t = p_l+p_h;
slouken@11683
   292
	SET_LOW_WORD(t,0);
slouken@11683
   293
	u = t*lg2_h;
slouken@11683
   294
	v = (p_l-(t-p_h))*lg2+t*lg2_l;
slouken@11683
   295
	z = u+v;
slouken@11683
   296
	w = v-(z-u);
slouken@11683
   297
	t  = z*z;
slouken@11683
   298
	t1  = z - t*(P1+t*(P2+t*(P3+t*(P4+t*P5))));
slouken@11683
   299
	r  = (z*t1)/(t1-two)-(w+z*w);
slouken@11683
   300
	z  = one-(r-z);
slouken@11683
   301
	GET_HIGH_WORD(j,z);
slouken@11683
   302
	j += (n<<20);
slouken@11683
   303
	if((j>>20)<=0) z = scalbn(z,n);	/* subnormal output */
slouken@11683
   304
	else SET_HIGH_WORD(z,j);
slouken@11683
   305
	return s*z;
slouken@11683
   306
}
slouken@11683
   307
slouken@11683
   308
/*
slouken@11683
   309
 * wrapper pow(x,y) return x**y
slouken@11683
   310
 */
slouken@11683
   311
#ifndef _IEEE_LIBM
slouken@11683
   312
double pow(double x, double y)
slouken@11683
   313
{
slouken@11683
   314
	double z = __ieee754_pow(x, y);
slouken@11683
   315
	if (_LIB_VERSION == _IEEE_|| isnan(y))
slouken@11683
   316
		return z;
slouken@11683
   317
	if (isnan(x)) {
slouken@11683
   318
		if (y == 0.0)
slouken@11683
   319
			return __kernel_standard(x, y, 42); /* pow(NaN,0.0) */
slouken@11683
   320
		return z;
slouken@11683
   321
	}
slouken@11683
   322
	if (x == 0.0) {
slouken@11683
   323
		if (y == 0.0)
slouken@11683
   324
	    		return __kernel_standard(x, y, 20); /* pow(0.0,0.0) */
slouken@11683
   325
		if (isfinite(y) && y < 0.0)
slouken@11683
   326
			return __kernel_standard(x,y,23); /* pow(0.0,negative) */
slouken@11683
   327
		return z;
slouken@11683
   328
	}
slouken@11683
   329
	if (!isfinite(z)) {
slouken@11683
   330
		if (isfinite(x) && isfinite(y)) {
slouken@11683
   331
			if (isnan(z))
slouken@11683
   332
				return __kernel_standard(x, y, 24); /* pow neg**non-int */
slouken@11683
   333
			return __kernel_standard(x, y, 21); /* pow overflow */
slouken@11683
   334
		}
slouken@11683
   335
	}
slouken@11683
   336
	if (z == 0.0 && isfinite(x) && isfinite(y))
slouken@11683
   337
		return __kernel_standard(x, y, 22); /* pow underflow */
slouken@11683
   338
	return z;
slouken@11683
   339
}
slouken@2756
   340
#else
slouken@11683
   341
strong_alias(__ieee754_pow, pow)
slouken@2756
   342
#endif
slouken@11683
   343
libm_hidden_def(pow)