12116Sjkh/* s_erff.c -- float version of s_erf.c.
22116Sjkh * Conversion to float by Ian Lance Taylor, Cygnus Support, ian@cygnus.com.
32116Sjkh */
42116Sjkh
52116Sjkh/*
62116Sjkh * ====================================================
72116Sjkh * Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved.
82116Sjkh *
92116Sjkh * Developed at SunPro, a Sun Microsystems, Inc. business.
102116Sjkh * Permission to use, copy, modify, and distribute this
118870Srgrimes * software is freely granted, provided that this notice
122116Sjkh * is preserved.
132116Sjkh * ====================================================
142116Sjkh */
152116Sjkh
16176451Sdas#include <sys/cdefs.h>
17176451Sdas__FBSDID("$FreeBSD$");
182116Sjkh
192116Sjkh#include "math.h"
202116Sjkh#include "math_private.h"
212116Sjkh
222116Sjkhstatic const float
232116Sjkhtiny	    = 1e-30,
242116Sjkhhalf=  5.0000000000e-01, /* 0x3F000000 */
252116Sjkhone =  1.0000000000e+00, /* 0x3F800000 */
262116Sjkhtwo =  2.0000000000e+00, /* 0x40000000 */
272116Sjkh/*
28254969Skargl * Coefficients for approximation to erf on [0,0.84375]
292116Sjkh */
302116Sjkhefx =  1.2837916613e-01, /* 0x3e0375d4 */
312116Sjkhefx8=  1.0270333290e+00, /* 0x3f8375d4 */
322116Sjkh/*
33254969Skargl *  Domain [0, 0.84375], range ~[-5.4446e-10,5.5197e-10]:
34254969Skargl *  |(erf(x) - x)/x - p(x)/q(x)| < 2**-31.
352116Sjkh */
36254969Skarglpp0  =  1.28379166e-01F, /*  0x1.06eba8p-3 */
37254969Skarglpp1  = -3.36030394e-01F, /* -0x1.58185ap-2 */
38254969Skarglpp2  = -1.86260219e-03F, /* -0x1.e8451ep-10 */
39254969Skarglqq1  =  3.12324286e-01F, /*  0x1.3fd1f0p-2 */
40254969Skarglqq2  =  2.16070302e-02F, /*  0x1.620274p-6 */
41254969Skarglqq3  = -1.98859419e-03F, /* -0x1.04a626p-9 */
422116Sjkh/*
43254969Skargl * Domain [0.84375, 1.25], range ~[-1.953e-11,1.940e-11]:
44254969Skargl * |(erf(x) - erx) - p(x)/q(x)| < 2**-36.
452116Sjkh */
46254969Skarglerx  =  8.42697144e-01F, /*  0x1.af7600p-1.  erf(1) rounded to 16 bits. */
47254969Skarglpa0  =  3.64939137e-06F, /*  0x1.e9d022p-19 */
48254969Skarglpa1  =  4.15109694e-01F, /*  0x1.a91284p-2 */
49254969Skarglpa2  = -1.65179938e-01F, /* -0x1.5249dcp-3 */
50254969Skarglpa3  =  1.10914491e-01F, /*  0x1.c64e46p-4 */
51254969Skarglqa1  =  6.02074385e-01F, /*  0x1.344318p-1 */
52254969Skarglqa2  =  5.35934687e-01F, /*  0x1.126608p-1 */
53254969Skarglqa3  =  1.68576106e-01F, /*  0x1.593e6ep-3 */
54254969Skarglqa4  =  5.62181212e-02F, /*  0x1.cc89f2p-5 */
552116Sjkh/*
56254969Skargl * Domain [1.25,1/0.35], range ~[-7.043e-10,7.457e-10]:
57254969Skargl * |log(x*erfc(x)) + x**2 + 0.5625 - r(x)/s(x)| < 2**-30
582116Sjkh */
59254969Skarglra0  = -9.87132732e-03F, /* -0x1.4376b2p-7 */
60254969Skarglra1  = -5.53605914e-01F, /* -0x1.1b723cp-1 */
61254969Skarglra2  = -2.17589188e+00F, /* -0x1.1683a0p+1 */
62254969Skarglra3  = -1.43268085e+00F, /* -0x1.6ec42cp+0 */
63254969Skarglsa1  =  5.45995426e+00F, /*  0x1.5d6fe4p+2 */
64254969Skarglsa2  =  6.69798088e+00F, /*  0x1.acabb8p+2 */
65254969Skarglsa3  =  1.43113089e+00F, /*  0x1.6e5e98p+0 */
66254969Skarglsa4  = -5.77397496e-02F, /* -0x1.d90108p-5 */
67254969Skargl/*
68254969Skargl * Domain [1/0.35, 11], range ~[-2.264e-13,2.336e-13]:
69254969Skargl * |log(x*erfc(x)) + x**2 + 0.5625 - r(x)/s(x)| < 2**-42
70254969Skargl */
71254969Skarglrb0  = -9.86494310e-03F, /* -0x1.434124p-7 */
72254969Skarglrb1  = -6.25171244e-01F, /* -0x1.401672p-1 */
73254969Skarglrb2  = -6.16498327e+00F, /* -0x1.8a8f16p+2 */
74254969Skarglrb3  = -1.66696873e+01F, /* -0x1.0ab70ap+4 */
75254969Skarglrb4  = -9.53764343e+00F, /* -0x1.313460p+3 */
76254969Skarglsb1  =  1.26884899e+01F, /*  0x1.96081cp+3 */
77254969Skarglsb2  =  4.51839523e+01F, /*  0x1.6978bcp+5 */
78254969Skarglsb3  =  4.72810211e+01F, /*  0x1.7a3f88p+5 */
79254969Skarglsb4  =  8.93033314e+00F; /*  0x1.1dc54ap+3 */
802116Sjkh
8197413Salfredfloat
8297413Salfrederff(float x)
832116Sjkh{
842116Sjkh	int32_t hx,ix,i;
852116Sjkh	float R,S,P,Q,s,y,z,r;
862116Sjkh	GET_FLOAT_WORD(hx,x);
872116Sjkh	ix = hx&0x7fffffff;
882116Sjkh	if(ix>=0x7f800000) {		/* erf(nan)=nan */
892116Sjkh	    i = ((u_int32_t)hx>>31)<<1;
902116Sjkh	    return (float)(1-i)+one/x;	/* erf(+-inf)=+-1 */
912116Sjkh	}
922116Sjkh
932116Sjkh	if(ix < 0x3f580000) {		/* |x|<0.84375 */
94254969Skargl	    if(ix < 0x38800000) { 	/* |x|<2**-14 */
95254969Skargl	        if (ix < 0x04000000)	/* |x|<0x1p-119 */
96254969Skargl		    return (8*x+efx8*x)/8;	/* avoid spurious underflow */
972116Sjkh		return x + efx*x;
982116Sjkh	    }
992116Sjkh	    z = x*x;
100254969Skargl	    r = pp0+z*(pp1+z*pp2);
101254969Skargl	    s = one+z*(qq1+z*(qq2+z*qq3));
1022116Sjkh	    y = r/s;
1032116Sjkh	    return x + x*y;
1042116Sjkh	}
1052116Sjkh	if(ix < 0x3fa00000) {		/* 0.84375 <= |x| < 1.25 */
1062116Sjkh	    s = fabsf(x)-one;
107254969Skargl	    P = pa0+s*(pa1+s*(pa2+s*pa3));
108254969Skargl	    Q = one+s*(qa1+s*(qa2+s*(qa3+s*qa4)));
1092116Sjkh	    if(hx>=0) return erx + P/Q; else return -erx - P/Q;
1102116Sjkh	}
111254969Skargl	if (ix >= 0x40800000) {		/* inf>|x|>=4 */
1122116Sjkh	    if(hx>=0) return one-tiny; else return tiny-one;
1132116Sjkh	}
1142116Sjkh	x = fabsf(x);
1152116Sjkh 	s = one/(x*x);
1162116Sjkh	if(ix< 0x4036DB6E) {	/* |x| < 1/0.35 */
117254969Skargl	    R=ra0+s*(ra1+s*(ra2+s*ra3));
118254969Skargl	    S=one+s*(sa1+s*(sa2+s*(sa3+s*sa4)));
1192116Sjkh	} else {	/* |x| >= 1/0.35 */
120254969Skargl	    R=rb0+s*(rb1+s*(rb2+s*(rb3+s*rb4)));
121254969Skargl	    S=one+s*(sb1+s*(sb2+s*(sb3+s*sb4)));
1222116Sjkh	}
123254969Skargl	SET_FLOAT_WORD(z,hx&0xffffe000);
124254969Skargl	r  = expf(-z*z-0.5625F)*expf((z-x)*(z+x)+R/S);
1252116Sjkh	if(hx>=0) return one-r/x; else return  r/x-one;
1262116Sjkh}
1272116Sjkh
12897413Salfredfloat
12997413Salfrederfcf(float x)
1302116Sjkh{
1312116Sjkh	int32_t hx,ix;
1322116Sjkh	float R,S,P,Q,s,y,z,r;
1332116Sjkh	GET_FLOAT_WORD(hx,x);
1342116Sjkh	ix = hx&0x7fffffff;
1352116Sjkh	if(ix>=0x7f800000) {			/* erfc(nan)=nan */
1362116Sjkh						/* erfc(+-inf)=0,2 */
1372116Sjkh	    return (float)(((u_int32_t)hx>>31)<<1)+one/x;
1382116Sjkh	}
1392116Sjkh
1402116Sjkh	if(ix < 0x3f580000) {		/* |x|<0.84375 */
141254969Skargl	    if(ix < 0x33800000)  	/* |x|<2**-24 */
1422116Sjkh		return one-x;
1432116Sjkh	    z = x*x;
144254969Skargl	    r = pp0+z*(pp1+z*pp2);
145254969Skargl	    s = one+z*(qq1+z*(qq2+z*qq3));
1462116Sjkh	    y = r/s;
1472116Sjkh	    if(hx < 0x3e800000) {  	/* x<1/4 */
1482116Sjkh		return one-(x+x*y);
1492116Sjkh	    } else {
1502116Sjkh		r = x*y;
1512116Sjkh		r += (x-half);
1522116Sjkh	        return half - r ;
1532116Sjkh	    }
1542116Sjkh	}
1552116Sjkh	if(ix < 0x3fa00000) {		/* 0.84375 <= |x| < 1.25 */
1562116Sjkh	    s = fabsf(x)-one;
157254969Skargl	    P = pa0+s*(pa1+s*(pa2+s*pa3));
158254969Skargl	    Q = one+s*(qa1+s*(qa2+s*(qa3+s*qa4)));
1592116Sjkh	    if(hx>=0) {
1608870Srgrimes	        z  = one-erx; return z - P/Q;
1612116Sjkh	    } else {
1622116Sjkh		z = erx+P/Q; return one+z;
1632116Sjkh	    }
1642116Sjkh	}
165254969Skargl	if (ix < 0x41300000) {		/* |x|<11 */
1662116Sjkh	    x = fabsf(x);
1672116Sjkh 	    s = one/(x*x);
1682116Sjkh	    if(ix< 0x4036DB6D) {	/* |x| < 1/.35 ~ 2.857143*/
169254969Skargl	        R=ra0+s*(ra1+s*(ra2+s*ra3));
170254969Skargl	        S=one+s*(sa1+s*(sa2+s*(sa3+s*sa4)));
1712116Sjkh	    } else {			/* |x| >= 1/.35 ~ 2.857143 */
172254969Skargl		if(hx<0&&ix>=0x40a00000) return two-tiny;/* x < -5 */
173254969Skargl	        R=rb0+s*(rb1+s*(rb2+s*(rb3+s*rb4)));
174254969Skargl		S=one+s*(sb1+s*(sb2+s*(sb3+s*sb4)));
1752116Sjkh	    }
176254969Skargl	    SET_FLOAT_WORD(z,hx&0xffffe000);
177254969Skargl	    r  = expf(-z*z-0.5625F)*expf((z-x)*(z+x)+R/S);
1782116Sjkh	    if(hx>0) return r/x; else return two-r/x;
1792116Sjkh	} else {
1802116Sjkh	    if(hx>0) return tiny*tiny; else return two-tiny;
1812116Sjkh	}
1822116Sjkh}
183