POK
/home/jaouen/pok_official/pok/trunk/libpok/libm/e_powf.c
00001 /*
00002  *                               POK header
00003  * 
00004  * The following file is a part of the POK project. Any modification should
00005  * made according to the POK licence. You CANNOT use this file or a part of
00006  * this file is this part of a file for your own project
00007  *
00008  * For more information on the POK licence, please see our LICENCE FILE
00009  *
00010  * Please follow the coding guidelines described in doc/CODING_GUIDELINES
00011  *
00012  *                                      Copyright (c) 2007-2009 POK team 
00013  *
00014  * Created by julien on Fri Jan 30 14:41:34 2009 
00015  */
00016 
00017 /* e_powf.c -- float version of e_pow.c.
00018  * Conversion to float by Ian Lance Taylor, Cygnus Support, ian@cygnus.com.
00019  */
00020 
00021 /*
00022  * ====================================================
00023  * Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved.
00024  *
00025  * Developed at SunPro, a Sun Microsystems, Inc. business.
00026  * Permission to use, copy, modify, and distribute this
00027  * software is freely granted, provided that this notice
00028  * is preserved.
00029  * ====================================================
00030  */
00031 
00032 #ifdef POK_NEEDS_LIBMATH
00033 
00034 #include <libm.h>
00035 #include "math_private.h"
00036 
00037 static const float huge = 1.0e+30, tiny = 1.0e-30;
00038 
00039 static const float
00040 bp[] = {1.0, 1.5,},
00041 dp_h[] = { 0.0, 5.84960938e-01,}, /* 0x3f15c000 */
00042 dp_l[] = { 0.0, 1.56322085e-06,}, /* 0x35d1cfdc */
00043 zero    =  0.0,
00044 one     =  1.0,
00045 two     =  2.0,
00046 two24   =  16777216.0,  /* 0x4b800000 */
00047         /* poly coefs for (3/2)*(log(x)-2s-2/3*s**3 */
00048 L1  =  6.0000002384e-01, /* 0x3f19999a */
00049 L2  =  4.2857143283e-01, /* 0x3edb6db7 */
00050 L3  =  3.3333334327e-01, /* 0x3eaaaaab */
00051 L4  =  2.7272811532e-01, /* 0x3e8ba305 */
00052 L5  =  2.3066075146e-01, /* 0x3e6c3255 */
00053 L6  =  2.0697501302e-01, /* 0x3e53f142 */
00054 P1   =  1.6666667163e-01, /* 0x3e2aaaab */
00055 P2   = -2.7777778450e-03, /* 0xbb360b61 */
00056 P3   =  6.6137559770e-05, /* 0x388ab355 */
00057 P4   = -1.6533901999e-06, /* 0xb5ddea0e */
00058 P5   =  4.1381369442e-08, /* 0x3331bb4c */
00059 lg2  =  6.9314718246e-01, /* 0x3f317218 */
00060 lg2_h  =  6.93145752e-01, /* 0x3f317200 */
00061 lg2_l  =  1.42860654e-06, /* 0x35bfbe8c */
00062 ovt =  4.2995665694e-08, /* -(128-log2(ovfl+.5ulp)) */
00063 cp    =  9.6179670095e-01, /* 0x3f76384f =2/(3ln2) */
00064 cp_h  =  9.6179199219e-01, /* 0x3f763800 =head of cp */
00065 cp_l  =  4.7017383622e-06, /* 0x369dc3a0 =tail of cp_h */
00066 ivln2    =  1.4426950216e+00, /* 0x3fb8aa3b =1/ln2 */
00067 ivln2_h  =  1.4426879883e+00, /* 0x3fb8aa00 =16b 1/ln2*/
00068 ivln2_l  =  7.0526075433e-06; /* 0x36eca570 =1/ln2 tail*/
00069 
00070 float
00071 __ieee754_powf(float x, float y)
00072 {
00073         float z,ax,z_h,z_l,p_h,p_l;
00074         float yy1,t1,t2,r,s,t,u,v,w;
00075         int32_t i,j,k,yisint,n;
00076         int32_t hx,hy,ix,iy,is;
00077 
00078         GET_FLOAT_WORD(hx,x);
00079         GET_FLOAT_WORD(hy,y);
00080         ix = hx&0x7fffffff;  iy = hy&0x7fffffff;
00081 
00082     /* y==zero: x**0 = 1 */
00083         if(iy==0) return one;
00084 
00085     /* +-NaN return x+y */
00086         if(ix > 0x7f800000 ||
00087            iy > 0x7f800000)
00088                 return x+y;
00089 
00090     /* determine if y is an odd int when x < 0
00091      * yisint = 0       ... y is not an integer
00092      * yisint = 1       ... y is an odd int
00093      * yisint = 2       ... y is an even int
00094      */
00095         yisint  = 0;
00096         if(hx<0) {
00097             if(iy>=0x4b800000) yisint = 2; /* even integer y */
00098             else if(iy>=0x3f800000) {
00099                 k = (iy>>23)-0x7f;         /* exponent */
00100                 j = iy>>(23-k);
00101                 if((j<<(23-k))==iy) yisint = 2-(j&1);
00102             }
00103         }
00104 
00105     /* special value of y */
00106         if (iy==0x7f800000) {   /* y is +-inf */
00107             if (ix==0x3f800000)
00108                 return  y - y;  /* inf**+-1 is NaN */
00109             else if (ix > 0x3f800000)/* (|x|>1)**+-inf = inf,0 */
00110                 return (hy>=0)? y: zero;
00111             else                        /* (|x|<1)**-,+inf = inf,0 */
00112                 return (hy<0)?-y: zero;
00113         }
00114         if(iy==0x3f800000) {    /* y is  +-1 */
00115             if(hy<0) return one/x; else return x;
00116         }
00117         if(hy==0x40000000) return x*x; /* y is  2 */
00118         if(hy==0x3f000000) {    /* y is  0.5 */
00119             if(hx>=0)   /* x >= +0 */
00120             return __ieee754_sqrtf(x);
00121         }
00122 
00123         ax   = fabsf(x);
00124     /* special value of x */
00125         if(ix==0x7f800000||ix==0||ix==0x3f800000){
00126             z = ax;                     /*x is +-0,+-inf,+-1*/
00127             if(hy<0) z = one/z; /* z = (1/|x|) */
00128             if(hx<0) {
00129                 if(((ix-0x3f800000)|yisint)==0) {
00130                     z = (z-z)/(z-z); /* (-1)**non-int is NaN */
00131                 } else if(yisint==1)
00132                     z = -z;             /* (x<0)**odd = -(|x|**odd) */
00133             }
00134             return z;
00135         }
00136 
00137     /* (x<0)**(non-int) is NaN */
00138         if(((((uint32_t)hx>>31)-1)|yisint)==0) return (x-x)/(x-x);
00139 
00140     /* |y| is huge */
00141         if(iy>0x4d000000) { /* if |y| > 2**27 */
00142         /* over/underflow if x is not close to one */
00143             if(ix<0x3f7ffff8) return (hy<0)? huge*huge:tiny*tiny;
00144             if(ix>0x3f800007) return (hy>0)? huge*huge:tiny*tiny;
00145         /* now |1-x| is tiny <= 2**-20, suffice to compute
00146            log(x) by x-x^2/2+x^3/3-x^4/4 */
00147             t = ax-one;         /* t has 20 trailing zeros */
00148             w = (t*t)*((float)0.5-t*((float)0.333333333333-t*(float)0.25));
00149             u = ivln2_h*t;      /* ivln2_h has 16 sig. bits */
00150             v = t*ivln2_l-w*ivln2;
00151             t1 = u+v;
00152             GET_FLOAT_WORD(is,t1);
00153             SET_FLOAT_WORD(t1,is&0xfffff000);
00154             t2 = v-(t1-u);
00155         } else {
00156             float s2,s_h,s_l,t_h,t_l;
00157             n = 0;
00158         /* take care subnormal number */
00159             if(ix<0x00800000)
00160                 {ax *= two24; n -= 24; GET_FLOAT_WORD(ix,ax); }
00161             n  += ((ix)>>23)-0x7f;
00162             j  = ix&0x007fffff;
00163         /* determine interval */
00164             ix = j|0x3f800000;          /* normalize ix */
00165             if(j<=0x1cc471) k=0;        /* |x|<sqrt(3/2) */
00166             else if(j<0x5db3d7) k=1;    /* |x|<sqrt(3)   */
00167             else {k=0;n+=1;ix -= 0x00800000;}
00168             SET_FLOAT_WORD(ax,ix);
00169 
00170         /* compute s = s_h+s_l = (x-1)/(x+1) or (x-1.5)/(x+1.5) */
00171             u = ax-bp[k];               /* bp[0]=1.0, bp[1]=1.5 */
00172             v = one/(ax+bp[k]);
00173             s = u*v;
00174             s_h = s;
00175             GET_FLOAT_WORD(is,s_h);
00176             SET_FLOAT_WORD(s_h,is&0xfffff000);
00177         /* t_h=ax+bp[k] High */
00178             SET_FLOAT_WORD(t_h,((ix>>1)|0x20000000)+0x0040000+(k<<21));
00179             t_l = ax - (t_h-bp[k]);
00180             s_l = v*((u-s_h*t_h)-s_h*t_l);
00181         /* compute log(ax) */
00182             s2 = s*s;
00183             r = s2*s2*(L1+s2*(L2+s2*(L3+s2*(L4+s2*(L5+s2*L6)))));
00184             r += s_l*(s_h+s);
00185             s2  = s_h*s_h;
00186             t_h = (float)3.0+s2+r;
00187             GET_FLOAT_WORD(is,t_h);
00188             SET_FLOAT_WORD(t_h,is&0xfffff000);
00189             t_l = r-((t_h-(float)3.0)-s2);
00190         /* u+v = s*(1+...) */
00191             u = s_h*t_h;
00192             v = s_l*t_h+t_l*s;
00193         /* 2/(3log2)*(s+...) */
00194             p_h = u+v;
00195             GET_FLOAT_WORD(is,p_h);
00196             SET_FLOAT_WORD(p_h,is&0xfffff000);
00197             p_l = v-(p_h-u);
00198             z_h = cp_h*p_h;             /* cp_h+cp_l = 2/(3*log2) */
00199             z_l = cp_l*p_h+p_l*cp+dp_l[k];
00200         /* log2(ax) = (s+..)*2/(3*log2) = n + dp_h + z_h + z_l */
00201             t = (float)n;
00202             t1 = (((z_h+z_l)+dp_h[k])+t);
00203             GET_FLOAT_WORD(is,t1);
00204             SET_FLOAT_WORD(t1,is&0xfffff000);
00205             t2 = z_l-(((t1-t)-dp_h[k])-z_h);
00206         }
00207 
00208         s = one; /* s (sign of result -ve**odd) = -1 else = 1 */
00209         if(((((uint32_t)hx>>31)-1)|(yisint-1))==0)
00210             s = -one;   /* (-ve)**(odd int) */
00211 
00212     /* split up y into yy1+y2 and compute (yy1+y2)*(t1+t2) */
00213         GET_FLOAT_WORD(is,y);
00214         SET_FLOAT_WORD(yy1,is&0xfffff000);
00215         p_l = (y-yy1)*t1+y*t2;
00216         p_h = yy1*t1;
00217         z = p_l+p_h;
00218         GET_FLOAT_WORD(j,z);
00219         if (j>0x43000000)                               /* if z > 128 */
00220             return s*huge*huge;                         /* overflow */
00221         else if (j==0x43000000) {                       /* if z == 128 */
00222             if(p_l+ovt>z-p_h) return s*huge*huge;       /* overflow */
00223         }
00224         else if ((uint32_t)j==0xc3160000){              /* z == -150 */
00225             if(p_l<=z-p_h) return s*tiny*tiny;          /* underflow */
00226         }
00227         else if ((j&0x7fffffff)>0x43160000)             /* z <= -150 */
00228             return s*tiny*tiny;                         /* underflow */
00229     /*
00230      * compute 2**(p_h+p_l)
00231      */
00232         i = j&0x7fffffff;
00233         k = (i>>23)-0x7f;
00234         n = 0;
00235         if(i>0x3f000000) {              /* if |z| > 0.5, set n = [z+0.5] */
00236             n = j+(0x00800000>>(k+1));
00237             k = ((n&0x7fffffff)>>23)-0x7f;      /* new k for n */
00238             SET_FLOAT_WORD(t,n&~(0x007fffff>>k));
00239             n = ((n&0x007fffff)|0x00800000)>>(23-k);
00240             if(j<0) n = -n;
00241             p_h -= t;
00242         }
00243         t = p_l+p_h;
00244         GET_FLOAT_WORD(is,t);
00245         SET_FLOAT_WORD(t,is&0xfffff000);
00246         u = t*lg2_h;
00247         v = (p_l-(t-p_h))*lg2+t*lg2_l;
00248         z = u+v;
00249         w = v-(z-u);
00250         t  = z*z;
00251         t1  = z - t*(P1+t*(P2+t*(P3+t*(P4+t*P5))));
00252         r  = (z*t1)/(t1-two)-(w+z*w);
00253         z  = one-(r-z);
00254         GET_FLOAT_WORD(j,z);
00255         j += (n<<23);
00256         if((j>>23)<=0) z = scalbnf(z,n);        /* subnormal output */
00257         else SET_FLOAT_WORD(z,j);
00258         return s*z;
00259 }
00260 
00261 #endif