src/video/e_pow.h
author Sam Lantinga <slouken@libsdl.org>
Mon, 21 Sep 2009 08:58:51 +0000
branchSDL-1.2
changeset 4214 4250beeb5ad1
parent 1424 7a610f25c12f
child 1662 782fd950bd46
child 1895 c121d94672cb
permissions -rw-r--r--
Oh yeah, we have GLX support too.
slouken@1330
     1
/* @(#)e_pow.c 5.1 93/09/24 */
slouken@1330
     2
/*
slouken@1330
     3
 * ====================================================
slouken@1330
     4
 * Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved.
slouken@1330
     5
 *
slouken@1330
     6
 * Developed at SunPro, a Sun Microsystems, Inc. business.
slouken@1330
     7
 * Permission to use, copy, modify, and distribute this
slouken@1330
     8
 * software is freely granted, provided that this notice
slouken@1330
     9
 * is preserved.
slouken@1330
    10
 * ====================================================
slouken@1330
    11
 */
slouken@1330
    12
slouken@1330
    13
#if defined(LIBM_SCCS) && !defined(lint)
slouken@1330
    14
static char rcsid[] = "$NetBSD: e_pow.c,v 1.9 1995/05/12 04:57:32 jtc Exp $";
slouken@1330
    15
#endif
slouken@1330
    16
slouken@1330
    17
/* __ieee754_pow(x,y) return x**y
slouken@1330
    18
 *
slouken@1330
    19
 *		      n
slouken@1330
    20
 * Method:  Let x =  2   * (1+f)
slouken@1330
    21
 *	1. Compute and return log2(x) in two pieces:
slouken@1330
    22
 *		log2(x) = w1 + w2,
slouken@1330
    23
 *	   where w1 has 53-24 = 29 bit trailing zeros.
slouken@1330
    24
 *	2. Perform y*log2(x) = n+y' by simulating muti-precision
slouken@1330
    25
 *	   arithmetic, where |y'|<=0.5.
slouken@1330
    26
 *	3. Return x**y = 2**n*exp(y'*log2)
slouken@1330
    27
 *
slouken@1330
    28
 * Special cases:
slouken@1330
    29
 *	1.  (anything) ** 0  is 1
slouken@1330
    30
 *	2.  (anything) ** 1  is itself
slouken@1330
    31
 *	3.  (anything) ** NAN is NAN
slouken@1330
    32
 *	4.  NAN ** (anything except 0) is NAN
slouken@1330
    33
 *	5.  +-(|x| > 1) **  +INF is +INF
slouken@1330
    34
 *	6.  +-(|x| > 1) **  -INF is +0
slouken@1330
    35
 *	7.  +-(|x| < 1) **  +INF is +0
slouken@1330
    36
 *	8.  +-(|x| < 1) **  -INF is +INF
slouken@1330
    37
 *	9.  +-1         ** +-INF is NAN
slouken@1330
    38
 *	10. +0 ** (+anything except 0, NAN)               is +0
slouken@1330
    39
 *	11. -0 ** (+anything except 0, NAN, odd integer)  is +0
slouken@1330
    40
 *	12. +0 ** (-anything except 0, NAN)               is +INF
slouken@1330
    41
 *	13. -0 ** (-anything except 0, NAN, odd integer)  is +INF
slouken@1330
    42
 *	14. -0 ** (odd integer) = -( +0 ** (odd integer) )
slouken@1330
    43
 *	15. +INF ** (+anything except 0,NAN) is +INF
slouken@1330
    44
 *	16. +INF ** (-anything except 0,NAN) is +0
slouken@1330
    45
 *	17. -INF ** (anything)  = -0 ** (-anything)
slouken@1330
    46
 *	18. (-anything) ** (integer) is (-1)**(integer)*(+anything**integer)
slouken@1330
    47
 *	19. (-anything except 0 and inf) ** (non-integer) is NAN
slouken@1330
    48
 *
slouken@1330
    49
 * Accuracy:
slouken@1330
    50
 *	pow(x,y) returns x**y nearly rounded. In particular
slouken@1330
    51
 *			pow(integer,integer)
slouken@1330
    52
 *	always returns the correct integer provided it is
slouken@1330
    53
 *	representable.
slouken@1330
    54
 *
slouken@1330
    55
 * Constants :
slouken@1330
    56
 * The hexadecimal values are the intended ones for the following
slouken@1330
    57
 * constants. The decimal values may be used, provided that the
slouken@1330
    58
 * compiler will convert from decimal to binary accurately enough
slouken@1330
    59
 * to produce the hexadecimal values shown.
slouken@1330
    60
 */
slouken@1330
    61
slouken@1330
    62
/*#include "math.h"*/
slouken@1330
    63
#include "math_private.h"
slouken@1330
    64
slouken@1330
    65
#ifdef __STDC__
slouken@1330
    66
static const double
slouken@1330
    67
#else
slouken@1330
    68
static double
slouken@1330
    69
#endif
slouken@1330
    70
bp[] = {1.0, 1.5,},
slouken@1330
    71
dp_h[] = { 0.0, 5.84962487220764160156e-01,}, /* 0x3FE2B803, 0x40000000 */
slouken@1330
    72
dp_l[] = { 0.0, 1.35003920212974897128e-08,}, /* 0x3E4CFDEB, 0x43CFD006 */
slouken@1330
    73
	/* poly coefs for (3/2)*(log(x)-2s-2/3*s**3 */
slouken@1330
    74
L1  =  5.99999999999994648725e-01, /* 0x3FE33333, 0x33333303 */
slouken@1330
    75
L2  =  4.28571428578550184252e-01, /* 0x3FDB6DB6, 0xDB6FABFF */
slouken@1330
    76
L3  =  3.33333329818377432918e-01, /* 0x3FD55555, 0x518F264D */
slouken@1330
    77
L4  =  2.72728123808534006489e-01, /* 0x3FD17460, 0xA91D4101 */
slouken@1330
    78
L5  =  2.30660745775561754067e-01, /* 0x3FCD864A, 0x93C9DB65 */
slouken@1330
    79
L6  =  2.06975017800338417784e-01, /* 0x3FCA7E28, 0x4A454EEF */
slouken@1330
    80
P1   =  1.66666666666666019037e-01, /* 0x3FC55555, 0x5555553E */
slouken@1330
    81
P2   = -2.77777777770155933842e-03, /* 0xBF66C16C, 0x16BEBD93 */
slouken@1330
    82
P3   =  6.61375632143793436117e-05, /* 0x3F11566A, 0xAF25DE2C */
slouken@1330
    83
P4   = -1.65339022054652515390e-06, /* 0xBEBBBD41, 0xC5D26BF1 */
slouken@1330
    84
P5   =  4.13813679705723846039e-08, /* 0x3E663769, 0x72BEA4D0 */
slouken@1330
    85
lg2  =  6.93147180559945286227e-01, /* 0x3FE62E42, 0xFEFA39EF */
slouken@1330
    86
lg2_h  =  6.93147182464599609375e-01, /* 0x3FE62E43, 0x00000000 */
slouken@1330
    87
lg2_l  = -1.90465429995776804525e-09, /* 0xBE205C61, 0x0CA86C39 */
slouken@1330
    88
ovt =  8.0085662595372944372e-0017, /* -(1024-log2(ovfl+.5ulp)) */
slouken@1330
    89
cp    =  9.61796693925975554329e-01, /* 0x3FEEC709, 0xDC3A03FD =2/(3ln2) */
slouken@1330
    90
cp_h  =  9.61796700954437255859e-01, /* 0x3FEEC709, 0xE0000000 =(float)cp */
slouken@1330
    91
cp_l  = -7.02846165095275826516e-09, /* 0xBE3E2FE0, 0x145B01F5 =tail of cp_h*/
slouken@1330
    92
ivln2    =  1.44269504088896338700e+00, /* 0x3FF71547, 0x652B82FE =1/ln2 */
slouken@1330
    93
ivln2_h  =  1.44269502162933349609e+00, /* 0x3FF71547, 0x60000000 =24b 1/ln2*/
slouken@1330
    94
ivln2_l  =  1.92596299112661746887e-08; /* 0x3E54AE0B, 0xF85DDF44 =1/ln2 tail*/
slouken@1330
    95
slouken@1330
    96
#ifdef __STDC__
slouken@1330
    97
	double __ieee754_pow(double x, double y)
slouken@1330
    98
#else
slouken@1330
    99
	double __ieee754_pow(x,y)
slouken@1330
   100
	double x, y;
slouken@1330
   101
#endif
slouken@1330
   102
{
slouken@1330
   103
	double z,ax,z_h,z_l,p_h,p_l;
slouken@1330
   104
	double y1,t1,t2,r,s,t,u,v,w;
slouken@1330
   105
	int32_t i,j,k,yisint,n;
slouken@1330
   106
	int32_t hx,hy,ix,iy;
slouken@1330
   107
	u_int32_t lx,ly;
slouken@1330
   108
slouken@1330
   109
	EXTRACT_WORDS(hx,lx,x);
slouken@1330
   110
	EXTRACT_WORDS(hy,ly,y);
slouken@1330
   111
	ix = hx&0x7fffffff;  iy = hy&0x7fffffff;
slouken@1330
   112
slouken@1330
   113
    /* y==zero: x**0 = 1 */
slouken@1330
   114
	if((iy|ly)==0) return one;
slouken@1330
   115
slouken@1330
   116
    /* +-NaN return x+y */
slouken@1330
   117
	if(ix > 0x7ff00000 || ((ix==0x7ff00000)&&(lx!=0)) ||
slouken@1330
   118
	   iy > 0x7ff00000 || ((iy==0x7ff00000)&&(ly!=0)))
slouken@1330
   119
		return x+y;
slouken@1330
   120
slouken@1330
   121
    /* determine if y is an odd int when x < 0
slouken@1330
   122
     * yisint = 0	... y is not an integer
slouken@1330
   123
     * yisint = 1	... y is an odd int
slouken@1330
   124
     * yisint = 2	... y is an even int
slouken@1330
   125
     */
slouken@1330
   126
	yisint  = 0;
slouken@1330
   127
	if(hx<0) {
slouken@1330
   128
	    if(iy>=0x43400000) yisint = 2; /* even integer y */
slouken@1330
   129
	    else if(iy>=0x3ff00000) {
slouken@1330
   130
		k = (iy>>20)-0x3ff;	   /* exponent */
slouken@1330
   131
		if(k>20) {
slouken@1330
   132
		    j = ly>>(52-k);
slouken@1345
   133
		    if((u_int32_t)(j<<(52-k))==ly) yisint = 2-(j&1);
slouken@1330
   134
		} else if(ly==0) {
slouken@1330
   135
		    j = iy>>(20-k);
slouken@1330
   136
		    if((j<<(20-k))==iy) yisint = 2-(j&1);
slouken@1330
   137
		}
slouken@1330
   138
	    }
slouken@1330
   139
	}
slouken@1330
   140
slouken@1330
   141
    /* special value of y */
slouken@1330
   142
	if(ly==0) {
slouken@1330
   143
	    if (iy==0x7ff00000) {	/* y is +-inf */
slouken@1330
   144
	        if(((ix-0x3ff00000)|lx)==0)
slouken@1330
   145
		    return  y - y;	/* inf**+-1 is NaN */
slouken@1330
   146
	        else if (ix >= 0x3ff00000)/* (|x|>1)**+-inf = inf,0 */
slouken@1330
   147
		    return (hy>=0)? y: zero;
slouken@1330
   148
	        else			/* (|x|<1)**-,+inf = inf,0 */
slouken@1330
   149
		    return (hy<0)?-y: zero;
slouken@1330
   150
	    }
slouken@1330
   151
	    if(iy==0x3ff00000) {	/* y is  +-1 */
slouken@1330
   152
		if(hy<0) return one/x; else return x;
slouken@1330
   153
	    }
slouken@1330
   154
	    if(hy==0x40000000) return x*x; /* y is  2 */
slouken@1330
   155
	    if(hy==0x3fe00000) {	/* y is  0.5 */
slouken@1330
   156
		if(hx>=0)	/* x >= +0 */
slouken@1330
   157
		return __ieee754_sqrt(x);
slouken@1330
   158
	    }
slouken@1330
   159
	}
slouken@1330
   160
slouken@1330
   161
	ax   = x < 0 ? -x : x; /*fabs(x);*/
slouken@1330
   162
    /* special value of x */
slouken@1330
   163
	if(lx==0) {
slouken@1330
   164
	    if(ix==0x7ff00000||ix==0||ix==0x3ff00000){
slouken@1330
   165
		z = ax;			/*x is +-0,+-inf,+-1*/
slouken@1330
   166
		if(hy<0) z = one/z;	/* z = (1/|x|) */
slouken@1330
   167
		if(hx<0) {
slouken@1330
   168
		    if(((ix-0x3ff00000)|yisint)==0) {
slouken@1330
   169
			z = (z-z)/(z-z); /* (-1)**non-int is NaN */
slouken@1330
   170
		    } else if(yisint==1)
slouken@1330
   171
			z = -z;		/* (x<0)**odd = -(|x|**odd) */
slouken@1330
   172
		}
slouken@1330
   173
		return z;
slouken@1330
   174
	    }
slouken@1330
   175
	}
slouken@1330
   176
slouken@1330
   177
    /* (x<0)**(non-int) is NaN */
slouken@1330
   178
	if(((((u_int32_t)hx>>31)-1)|yisint)==0) return (x-x)/(x-x);
slouken@1330
   179
slouken@1330
   180
    /* |y| is huge */
slouken@1330
   181
	if(iy>0x41e00000) { /* if |y| > 2**31 */
slouken@1330
   182
	    if(iy>0x43f00000){	/* if |y| > 2**64, must o/uflow */
slouken@1330
   183
		if(ix<=0x3fefffff) return (hy<0)? huge*huge:tiny*tiny;
slouken@1330
   184
		if(ix>=0x3ff00000) return (hy>0)? huge*huge:tiny*tiny;
slouken@1330
   185
	    }
slouken@1330
   186
	/* over/underflow if x is not close to one */
slouken@1330
   187
	    if(ix<0x3fefffff) return (hy<0)? huge*huge:tiny*tiny;
slouken@1330
   188
	    if(ix>0x3ff00000) return (hy>0)? huge*huge:tiny*tiny;
slouken@1330
   189
	/* now |1-x| is tiny <= 2**-20, suffice to compute
slouken@1330
   190
	   log(x) by x-x^2/2+x^3/3-x^4/4 */
slouken@1330
   191
	    t = x-1;		/* t has 20 trailing zeros */
slouken@1330
   192
	    w = (t*t)*(0.5-t*(0.3333333333333333333333-t*0.25));
slouken@1330
   193
	    u = ivln2_h*t;	/* ivln2_h has 21 sig. bits */
slouken@1330
   194
	    v = t*ivln2_l-w*ivln2;
slouken@1330
   195
	    t1 = u+v;
slouken@1330
   196
	    SET_LOW_WORD(t1,0);
slouken@1330
   197
	    t2 = v-(t1-u);
slouken@1330
   198
	} else {
slouken@1330
   199
	    double s2,s_h,s_l,t_h,t_l;
slouken@1330
   200
	    n = 0;
slouken@1330
   201
	/* take care subnormal number */
slouken@1330
   202
	    if(ix<0x00100000)
slouken@1330
   203
		{ax *= two53; n -= 53; GET_HIGH_WORD(ix,ax); }
slouken@1330
   204
	    n  += ((ix)>>20)-0x3ff;
slouken@1330
   205
	    j  = ix&0x000fffff;
slouken@1330
   206
	/* determine interval */
slouken@1330
   207
	    ix = j|0x3ff00000;		/* normalize ix */
slouken@1330
   208
	    if(j<=0x3988E) k=0;		/* |x|<sqrt(3/2) */
slouken@1330
   209
	    else if(j<0xBB67A) k=1;	/* |x|<sqrt(3)   */
slouken@1330
   210
	    else {k=0;n+=1;ix -= 0x00100000;}
slouken@1330
   211
	    SET_HIGH_WORD(ax,ix);
slouken@1330
   212
slouken@1330
   213
	/* compute s = s_h+s_l = (x-1)/(x+1) or (x-1.5)/(x+1.5) */
slouken@1330
   214
	    u = ax-bp[k];		/* bp[0]=1.0, bp[1]=1.5 */
slouken@1330
   215
	    v = one/(ax+bp[k]);
slouken@1330
   216
	    s = u*v;
slouken@1330
   217
	    s_h = s;
slouken@1330
   218
	    SET_LOW_WORD(s_h,0);
slouken@1330
   219
	/* t_h=ax+bp[k] High */
slouken@1330
   220
	    t_h = zero;
slouken@1330
   221
	    SET_HIGH_WORD(t_h,((ix>>1)|0x20000000)+0x00080000+(k<<18));
slouken@1330
   222
	    t_l = ax - (t_h-bp[k]);
slouken@1330
   223
	    s_l = v*((u-s_h*t_h)-s_h*t_l);
slouken@1330
   224
	/* compute log(ax) */
slouken@1330
   225
	    s2 = s*s;
slouken@1330
   226
	    r = s2*s2*(L1+s2*(L2+s2*(L3+s2*(L4+s2*(L5+s2*L6)))));
slouken@1330
   227
	    r += s_l*(s_h+s);
slouken@1330
   228
	    s2  = s_h*s_h;
slouken@1330
   229
	    t_h = 3.0+s2+r;
slouken@1330
   230
	    SET_LOW_WORD(t_h,0);
slouken@1330
   231
	    t_l = r-((t_h-3.0)-s2);
slouken@1330
   232
	/* u+v = s*(1+...) */
slouken@1330
   233
	    u = s_h*t_h;
slouken@1330
   234
	    v = s_l*t_h+t_l*s;
slouken@1330
   235
	/* 2/(3log2)*(s+...) */
slouken@1330
   236
	    p_h = u+v;
slouken@1330
   237
	    SET_LOW_WORD(p_h,0);
slouken@1330
   238
	    p_l = v-(p_h-u);
slouken@1330
   239
	    z_h = cp_h*p_h;		/* cp_h+cp_l = 2/(3*log2) */
slouken@1330
   240
	    z_l = cp_l*p_h+p_l*cp+dp_l[k];
slouken@1330
   241
	/* log2(ax) = (s+..)*2/(3*log2) = n + dp_h + z_h + z_l */
slouken@1330
   242
	    t = (double)n;
slouken@1330
   243
	    t1 = (((z_h+z_l)+dp_h[k])+t);
slouken@1330
   244
	    SET_LOW_WORD(t1,0);
slouken@1330
   245
	    t2 = z_l-(((t1-t)-dp_h[k])-z_h);
slouken@1330
   246
	}
slouken@1330
   247
slouken@1330
   248
	s = one; /* s (sign of result -ve**odd) = -1 else = 1 */
slouken@1330
   249
	if(((((u_int32_t)hx>>31)-1)|(yisint-1))==0)
slouken@1330
   250
	    s = -one;/* (-ve)**(odd int) */
slouken@1330
   251
slouken@1330
   252
    /* split up y into y1+y2 and compute (y1+y2)*(t1+t2) */
slouken@1330
   253
	y1  = y;
slouken@1330
   254
	SET_LOW_WORD(y1,0);
slouken@1330
   255
	p_l = (y-y1)*t1+y*t2;
slouken@1330
   256
	p_h = y1*t1;
slouken@1330
   257
	z = p_l+p_h;
slouken@1330
   258
	EXTRACT_WORDS(j,i,z);
slouken@1330
   259
	if (j>=0x40900000) {				/* z >= 1024 */
slouken@1330
   260
	    if(((j-0x40900000)|i)!=0)			/* if z > 1024 */
slouken@1330
   261
		return s*huge*huge;			/* overflow */
slouken@1330
   262
	    else {
slouken@1330
   263
		if(p_l+ovt>z-p_h) return s*huge*huge;	/* overflow */
slouken@1330
   264
	    }
slouken@1330
   265
	} else if((j&0x7fffffff)>=0x4090cc00 ) {	/* z <= -1075 */
slouken@1330
   266
	    if(((j-0xc090cc00)|i)!=0) 		/* z < -1075 */
slouken@1330
   267
		return s*tiny*tiny;		/* underflow */
slouken@1330
   268
	    else {
slouken@1330
   269
		if(p_l<=z-p_h) return s*tiny*tiny;	/* underflow */
slouken@1330
   270
	    }
slouken@1330
   271
	}
slouken@1330
   272
    /*
slouken@1330
   273
     * compute 2**(p_h+p_l)
slouken@1330
   274
     */
slouken@1330
   275
	i = j&0x7fffffff;
slouken@1330
   276
	k = (i>>20)-0x3ff;
slouken@1330
   277
	n = 0;
slouken@1330
   278
	if(i>0x3fe00000) {		/* if |z| > 0.5, set n = [z+0.5] */
slouken@1330
   279
	    n = j+(0x00100000>>(k+1));
slouken@1330
   280
	    k = ((n&0x7fffffff)>>20)-0x3ff;	/* new k for n */
slouken@1330
   281
	    t = zero;
slouken@1330
   282
	    SET_HIGH_WORD(t,n&~(0x000fffff>>k));
slouken@1330
   283
	    n = ((n&0x000fffff)|0x00100000)>>(20-k);
slouken@1330
   284
	    if(j<0) n = -n;
slouken@1330
   285
	    p_h -= t;
slouken@1330
   286
	}
slouken@1330
   287
	t = p_l+p_h;
slouken@1330
   288
	SET_LOW_WORD(t,0);
slouken@1330
   289
	u = t*lg2_h;
slouken@1330
   290
	v = (p_l-(t-p_h))*lg2+t*lg2_l;
slouken@1330
   291
	z = u+v;
slouken@1330
   292
	w = v-(z-u);
slouken@1330
   293
	t  = z*z;
slouken@1330
   294
	t1  = z - t*(P1+t*(P2+t*(P3+t*(P4+t*P5))));
slouken@1330
   295
	r  = (z*t1)/(t1-two)-(w+z*w);
slouken@1330
   296
	z  = one-(r-z);
slouken@1330
   297
	GET_HIGH_WORD(j,z);
slouken@1330
   298
	j += (n<<20);
slouken@1424
   299
	if((j>>20)<=0) z = SDL_NAME(scalbn)(z,n);	/* subnormal output */
slouken@1330
   300
	else SET_HIGH_WORD(z,j);
slouken@1330
   301
	return s*z;
slouken@1330
   302
}