src/video/e_pow.h
author Edgar Simo <bobbens@gmail.com>
Sun, 06 Jul 2008 17:06:37 +0000
branchgsoc2008_force_feedback
changeset 2498 ab567bd667bf
parent 1895 c121d94672cb
child 2752 edd2839b36f7
permissions -rw-r--r--
Fixed various mistakes in the doxygen.
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@1895
    70
  bp[] = { 1.0, 1.5, }, dp_h[] = {
slouken@1895
    71
0.0, 5.84962487220764160156e-01,},      /* 0x3FE2B803, 0x40000000 */
slouken@1895
    72
slouken@1895
    73
    dp_l[] = {
slouken@1895
    74
0.0, 1.35003920212974897128e-08,},      /* 0x3E4CFDEB, 0x43CFD006 */
slouken@1895
    75
slouken@1895
    76
    /* poly coefs for (3/2)*(log(x)-2s-2/3*s**3 */
slouken@1895
    77
    L1 = 5.99999999999994648725e-01,    /* 0x3FE33333, 0x33333303 */
slouken@1895
    78
    L2 = 4.28571428578550184252e-01,    /* 0x3FDB6DB6, 0xDB6FABFF */
slouken@1895
    79
    L3 = 3.33333329818377432918e-01,    /* 0x3FD55555, 0x518F264D */
slouken@1895
    80
    L4 = 2.72728123808534006489e-01,    /* 0x3FD17460, 0xA91D4101 */
slouken@1895
    81
    L5 = 2.30660745775561754067e-01,    /* 0x3FCD864A, 0x93C9DB65 */
slouken@1895
    82
    L6 = 2.06975017800338417784e-01,    /* 0x3FCA7E28, 0x4A454EEF */
slouken@1895
    83
    P1 = 1.66666666666666019037e-01,    /* 0x3FC55555, 0x5555553E */
slouken@1895
    84
    P2 = -2.77777777770155933842e-03,   /* 0xBF66C16C, 0x16BEBD93 */
slouken@1895
    85
    P3 = 6.61375632143793436117e-05,    /* 0x3F11566A, 0xAF25DE2C */
slouken@1895
    86
    P4 = -1.65339022054652515390e-06,   /* 0xBEBBBD41, 0xC5D26BF1 */
slouken@1895
    87
    P5 = 4.13813679705723846039e-08,    /* 0x3E663769, 0x72BEA4D0 */
slouken@1895
    88
    lg2 = 6.93147180559945286227e-01,   /* 0x3FE62E42, 0xFEFA39EF */
slouken@1895
    89
    lg2_h = 6.93147182464599609375e-01, /* 0x3FE62E43, 0x00000000 */
slouken@1895
    90
    lg2_l = -1.90465429995776804525e-09,        /* 0xBE205C61, 0x0CA86C39 */
slouken@1895
    91
    ovt = 8.0085662595372944372e-0017,  /* -(1024-log2(ovfl+.5ulp)) */
slouken@1895
    92
    cp = 9.61796693925975554329e-01,    /* 0x3FEEC709, 0xDC3A03FD =2/(3ln2) */
slouken@1895
    93
    cp_h = 9.61796700954437255859e-01,  /* 0x3FEEC709, 0xE0000000 =(float)cp */
slouken@1895
    94
    cp_l = -7.02846165095275826516e-09, /* 0xBE3E2FE0, 0x145B01F5 =tail of cp_h */
slouken@1895
    95
    ivln2 = 1.44269504088896338700e+00, /* 0x3FF71547, 0x652B82FE =1/ln2 */
slouken@1895
    96
    ivln2_h = 1.44269502162933349609e+00,       /* 0x3FF71547, 0x60000000 =24b 1/ln2 */
slouken@1895
    97
    ivln2_l = 1.92596299112661746887e-08;       /* 0x3E54AE0B, 0xF85DDF44 =1/ln2 tail */
slouken@1330
    98
slouken@1330
    99
#ifdef __STDC__
slouken@1895
   100
double
slouken@1895
   101
__ieee754_pow(double x, double y)
slouken@1330
   102
#else
slouken@1895
   103
double
slouken@1895
   104
__ieee754_pow(x, y)
slouken@1895
   105
     double x, y;
slouken@1330
   106
#endif
slouken@1330
   107
{
slouken@1895
   108
    double z, ax, z_h, z_l, p_h, p_l;
slouken@1895
   109
    double y1, t1, t2, r, s, t, u, v, w;
slouken@1895
   110
    int32_t i, j, k, yisint, n;
slouken@1895
   111
    int32_t hx, hy, ix, iy;
slouken@1895
   112
    u_int32_t lx, ly;
slouken@1330
   113
slouken@1895
   114
    EXTRACT_WORDS(hx, lx, x);
slouken@1895
   115
    EXTRACT_WORDS(hy, ly, y);
slouken@1895
   116
    ix = hx & 0x7fffffff;
slouken@1895
   117
    iy = hy & 0x7fffffff;
slouken@1330
   118
slouken@1330
   119
    /* y==zero: x**0 = 1 */
slouken@1895
   120
    if ((iy | ly) == 0)
slouken@1895
   121
        return one;
slouken@1330
   122
slouken@1330
   123
    /* +-NaN return x+y */
slouken@1895
   124
    if (ix > 0x7ff00000 || ((ix == 0x7ff00000) && (lx != 0)) ||
slouken@1895
   125
        iy > 0x7ff00000 || ((iy == 0x7ff00000) && (ly != 0)))
slouken@1895
   126
        return x + y;
slouken@1330
   127
slouken@1330
   128
    /* determine if y is an odd int when x < 0
slouken@1895
   129
     * yisint = 0       ... y is not an integer
slouken@1895
   130
     * yisint = 1       ... y is an odd int
slouken@1895
   131
     * yisint = 2       ... y is an even int
slouken@1330
   132
     */
slouken@1895
   133
    yisint = 0;
slouken@1895
   134
    if (hx < 0) {
slouken@1895
   135
        if (iy >= 0x43400000)
slouken@1895
   136
            yisint = 2;         /* even integer y */
slouken@1895
   137
        else if (iy >= 0x3ff00000) {
slouken@1895
   138
            k = (iy >> 20) - 0x3ff;     /* exponent */
slouken@1895
   139
            if (k > 20) {
slouken@1895
   140
                j = ly >> (52 - k);
slouken@1895
   141
                if ((u_int32_t) (j << (52 - k)) == ly)
slouken@1895
   142
                    yisint = 2 - (j & 1);
slouken@1895
   143
            } else if (ly == 0) {
slouken@1895
   144
                j = iy >> (20 - k);
slouken@1895
   145
                if ((j << (20 - k)) == iy)
slouken@1895
   146
                    yisint = 2 - (j & 1);
slouken@1895
   147
            }
slouken@1895
   148
        }
slouken@1895
   149
    }
slouken@1330
   150
slouken@1330
   151
    /* special value of y */
slouken@1895
   152
    if (ly == 0) {
slouken@1895
   153
        if (iy == 0x7ff00000) { /* y is +-inf */
slouken@1895
   154
            if (((ix - 0x3ff00000) | lx) == 0)
slouken@1895
   155
                return y - y;   /* inf**+-1 is NaN */
slouken@1895
   156
            else if (ix >= 0x3ff00000)  /* (|x|>1)**+-inf = inf,0 */
slouken@1895
   157
                return (hy >= 0) ? y : zero;
slouken@1895
   158
            else                /* (|x|<1)**-,+inf = inf,0 */
slouken@1895
   159
                return (hy < 0) ? -y : zero;
slouken@1895
   160
        }
slouken@1895
   161
        if (iy == 0x3ff00000) { /* y is  +-1 */
slouken@1895
   162
            if (hy < 0)
slouken@1895
   163
                return one / x;
slouken@1895
   164
            else
slouken@1895
   165
                return x;
slouken@1895
   166
        }
slouken@1895
   167
        if (hy == 0x40000000)
slouken@1895
   168
            return x * x;       /* y is  2 */
slouken@1895
   169
        if (hy == 0x3fe00000) { /* y is  0.5 */
slouken@1895
   170
            if (hx >= 0)        /* x >= +0 */
slouken@1895
   171
                return __ieee754_sqrt(x);
slouken@1895
   172
        }
slouken@1895
   173
    }
slouken@1330
   174
slouken@1895
   175
    ax = x < 0 ? -x : x;        /*fabs(x); */
slouken@1330
   176
    /* special value of x */
slouken@1895
   177
    if (lx == 0) {
slouken@1895
   178
        if (ix == 0x7ff00000 || ix == 0 || ix == 0x3ff00000) {
slouken@1895
   179
            z = ax;             /*x is +-0,+-inf,+-1 */
slouken@1895
   180
            if (hy < 0)
slouken@1895
   181
                z = one / z;    /* z = (1/|x|) */
slouken@1895
   182
            if (hx < 0) {
slouken@1895
   183
                if (((ix - 0x3ff00000) | yisint) == 0) {
slouken@1895
   184
                    z = (z - z) / (z - z);      /* (-1)**non-int is NaN */
slouken@1895
   185
                } else if (yisint == 1)
slouken@1895
   186
                    z = -z;     /* (x<0)**odd = -(|x|**odd) */
slouken@1895
   187
            }
slouken@1895
   188
            return z;
slouken@1895
   189
        }
slouken@1895
   190
    }
slouken@1330
   191
slouken@1330
   192
    /* (x<0)**(non-int) is NaN */
slouken@1895
   193
    if (((((u_int32_t) hx >> 31) - 1) | yisint) == 0)
slouken@1895
   194
        return (x - x) / (x - x);
slouken@1330
   195
slouken@1330
   196
    /* |y| is huge */
slouken@1895
   197
    if (iy > 0x41e00000) {      /* if |y| > 2**31 */
slouken@1895
   198
        if (iy > 0x43f00000) {  /* if |y| > 2**64, must o/uflow */
slouken@1895
   199
            if (ix <= 0x3fefffff)
slouken@1895
   200
                return (hy < 0) ? huge * huge : tiny * tiny;
slouken@1895
   201
            if (ix >= 0x3ff00000)
slouken@1895
   202
                return (hy > 0) ? huge * huge : tiny * tiny;
slouken@1895
   203
        }
slouken@1895
   204
        /* over/underflow if x is not close to one */
slouken@1895
   205
        if (ix < 0x3fefffff)
slouken@1895
   206
            return (hy < 0) ? huge * huge : tiny * tiny;
slouken@1895
   207
        if (ix > 0x3ff00000)
slouken@1895
   208
            return (hy > 0) ? huge * huge : tiny * tiny;
slouken@1895
   209
        /* now |1-x| is tiny <= 2**-20, suffice to compute
slouken@1895
   210
           log(x) by x-x^2/2+x^3/3-x^4/4 */
slouken@1895
   211
        t = x - 1;              /* t has 20 trailing zeros */
slouken@1895
   212
        w = (t * t) * (0.5 - t * (0.3333333333333333333333 - t * 0.25));
slouken@1895
   213
        u = ivln2_h * t;        /* ivln2_h has 21 sig. bits */
slouken@1895
   214
        v = t * ivln2_l - w * ivln2;
slouken@1895
   215
        t1 = u + v;
slouken@1895
   216
        SET_LOW_WORD(t1, 0);
slouken@1895
   217
        t2 = v - (t1 - u);
slouken@1895
   218
    } else {
slouken@1895
   219
        double s2, s_h, s_l, t_h, t_l;
slouken@1895
   220
        n = 0;
slouken@1895
   221
        /* take care subnormal number */
slouken@1895
   222
        if (ix < 0x00100000) {
slouken@1895
   223
            ax *= two53;
slouken@1895
   224
            n -= 53;
slouken@1895
   225
            GET_HIGH_WORD(ix, ax);
slouken@1895
   226
        }
slouken@1895
   227
        n += ((ix) >> 20) - 0x3ff;
slouken@1895
   228
        j = ix & 0x000fffff;
slouken@1895
   229
        /* determine interval */
slouken@1895
   230
        ix = j | 0x3ff00000;    /* normalize ix */
slouken@1895
   231
        if (j <= 0x3988E)
slouken@1895
   232
            k = 0;              /* |x|<sqrt(3/2) */
slouken@1895
   233
        else if (j < 0xBB67A)
slouken@1895
   234
            k = 1;              /* |x|<sqrt(3)   */
slouken@1895
   235
        else {
slouken@1895
   236
            k = 0;
slouken@1895
   237
            n += 1;
slouken@1895
   238
            ix -= 0x00100000;
slouken@1895
   239
        }
slouken@1895
   240
        SET_HIGH_WORD(ax, ix);
slouken@1330
   241
slouken@1895
   242
        /* compute s = s_h+s_l = (x-1)/(x+1) or (x-1.5)/(x+1.5) */
slouken@1895
   243
        u = ax - bp[k];         /* bp[0]=1.0, bp[1]=1.5 */
slouken@1895
   244
        v = one / (ax + bp[k]);
slouken@1895
   245
        s = u * v;
slouken@1895
   246
        s_h = s;
slouken@1895
   247
        SET_LOW_WORD(s_h, 0);
slouken@1895
   248
        /* t_h=ax+bp[k] High */
slouken@1895
   249
        t_h = zero;
slouken@1895
   250
        SET_HIGH_WORD(t_h, ((ix >> 1) | 0x20000000) + 0x00080000 + (k << 18));
slouken@1895
   251
        t_l = ax - (t_h - bp[k]);
slouken@1895
   252
        s_l = v * ((u - s_h * t_h) - s_h * t_l);
slouken@1895
   253
        /* compute log(ax) */
slouken@1895
   254
        s2 = s * s;
slouken@1895
   255
        r = s2 * s2 * (L1 +
slouken@1895
   256
                       s2 * (L2 +
slouken@1895
   257
                             s2 * (L3 + s2 * (L4 + s2 * (L5 + s2 * L6)))));
slouken@1895
   258
        r += s_l * (s_h + s);
slouken@1895
   259
        s2 = s_h * s_h;
slouken@1895
   260
        t_h = 3.0 + s2 + r;
slouken@1895
   261
        SET_LOW_WORD(t_h, 0);
slouken@1895
   262
        t_l = r - ((t_h - 3.0) - s2);
slouken@1895
   263
        /* u+v = s*(1+...) */
slouken@1895
   264
        u = s_h * t_h;
slouken@1895
   265
        v = s_l * t_h + t_l * s;
slouken@1895
   266
        /* 2/(3log2)*(s+...) */
slouken@1895
   267
        p_h = u + v;
slouken@1895
   268
        SET_LOW_WORD(p_h, 0);
slouken@1895
   269
        p_l = v - (p_h - u);
slouken@1895
   270
        z_h = cp_h * p_h;       /* cp_h+cp_l = 2/(3*log2) */
slouken@1895
   271
        z_l = cp_l * p_h + p_l * cp + dp_l[k];
slouken@1895
   272
        /* log2(ax) = (s+..)*2/(3*log2) = n + dp_h + z_h + z_l */
slouken@1895
   273
        t = (double) n;
slouken@1895
   274
        t1 = (((z_h + z_l) + dp_h[k]) + t);
slouken@1895
   275
        SET_LOW_WORD(t1, 0);
slouken@1895
   276
        t2 = z_l - (((t1 - t) - dp_h[k]) - z_h);
slouken@1895
   277
    }
slouken@1330
   278
slouken@1895
   279
    s = one;                    /* s (sign of result -ve**odd) = -1 else = 1 */
slouken@1895
   280
    if (((((u_int32_t) hx >> 31) - 1) | (yisint - 1)) == 0)
slouken@1895
   281
        s = -one;               /* (-ve)**(odd int) */
slouken@1330
   282
slouken@1330
   283
    /* split up y into y1+y2 and compute (y1+y2)*(t1+t2) */
slouken@1895
   284
    y1 = y;
slouken@1895
   285
    SET_LOW_WORD(y1, 0);
slouken@1895
   286
    p_l = (y - y1) * t1 + y * t2;
slouken@1895
   287
    p_h = y1 * t1;
slouken@1895
   288
    z = p_l + p_h;
slouken@1895
   289
    EXTRACT_WORDS(j, i, z);
slouken@1895
   290
    if (j >= 0x40900000) {      /* z >= 1024 */
slouken@1895
   291
        if (((j - 0x40900000) | i) != 0)        /* if z > 1024 */
slouken@1895
   292
            return s * huge * huge;     /* overflow */
slouken@1895
   293
        else {
slouken@1895
   294
            if (p_l + ovt > z - p_h)
slouken@1895
   295
                return s * huge * huge; /* overflow */
slouken@1895
   296
        }
slouken@1895
   297
    } else if ((j & 0x7fffffff) >= 0x4090cc00) {        /* z <= -1075 */
slouken@1895
   298
        if (((j - 0xc090cc00) | i) != 0)        /* z < -1075 */
slouken@1895
   299
            return s * tiny * tiny;     /* underflow */
slouken@1895
   300
        else {
slouken@1895
   301
            if (p_l <= z - p_h)
slouken@1895
   302
                return s * tiny * tiny; /* underflow */
slouken@1895
   303
        }
slouken@1895
   304
    }
slouken@1330
   305
    /*
slouken@1330
   306
     * compute 2**(p_h+p_l)
slouken@1330
   307
     */
slouken@1895
   308
    i = j & 0x7fffffff;
slouken@1895
   309
    k = (i >> 20) - 0x3ff;
slouken@1895
   310
    n = 0;
slouken@1895
   311
    if (i > 0x3fe00000) {       /* if |z| > 0.5, set n = [z+0.5] */
slouken@1895
   312
        n = j + (0x00100000 >> (k + 1));
slouken@1895
   313
        k = ((n & 0x7fffffff) >> 20) - 0x3ff;   /* new k for n */
slouken@1895
   314
        t = zero;
slouken@1895
   315
        SET_HIGH_WORD(t, n & ~(0x000fffff >> k));
slouken@1895
   316
        n = ((n & 0x000fffff) | 0x00100000) >> (20 - k);
slouken@1895
   317
        if (j < 0)
slouken@1895
   318
            n = -n;
slouken@1895
   319
        p_h -= t;
slouken@1895
   320
    }
slouken@1895
   321
    t = p_l + p_h;
slouken@1895
   322
    SET_LOW_WORD(t, 0);
slouken@1895
   323
    u = t * lg2_h;
slouken@1895
   324
    v = (p_l - (t - p_h)) * lg2 + t * lg2_l;
slouken@1895
   325
    z = u + v;
slouken@1895
   326
    w = v - (z - u);
slouken@1895
   327
    t = z * z;
slouken@1895
   328
    t1 = z - t * (P1 + t * (P2 + t * (P3 + t * (P4 + t * P5))));
slouken@1895
   329
    r = (z * t1) / (t1 - two) - (w + z * w);
slouken@1895
   330
    z = one - (r - z);
slouken@1895
   331
    GET_HIGH_WORD(j, z);
slouken@1895
   332
    j += (n << 20);
slouken@1895
   333
    if ((j >> 20) <= 0)
slouken@1895
   334
        z = SDL_NAME(scalbn) (z, n);    /* subnormal output */
slouken@1895
   335
    else
slouken@1895
   336
        SET_HIGH_WORD(z, j);
slouken@1895
   337
    return s * z;
slouken@1330
   338
}
slouken@1895
   339
slouken@1895
   340
/* vi: set ts=4 sw=4 expandtab: */