xref: /freebsd/lib/msun/src/s_erf.c (revision 8aac90f18aef7c9eea906c3ff9a001ca7b94f375)
1 /*
2  * ====================================================
3  * Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved.
4  *
5  * Developed at SunPro, a Sun Microsystems, Inc. business.
6  * Permission to use, copy, modify, and distribute this
7  * software is freely granted, provided that this notice
8  * is preserved.
9  * ====================================================
10  */
11 
12 /* double erf(double x)
13  * double erfc(double x)
14  *			     x
15  *		      2      |\
16  *     erf(x)  =  ---------  | exp(-t*t)dt
17  *	 	   sqrt(pi) \|
18  *			     0
19  *
20  *     erfc(x) =  1-erf(x)
21  *  Note that
22  *		erf(-x) = -erf(x)
23  *		erfc(-x) = 2 - erfc(x)
24  *
25  * Method:
26  *	1. For |x| in [0, 0.84375]
27  *	    erf(x)  = x + x*R(x^2)
28  *          erfc(x) = 1 - erf(x)           if x in [-.84375,0.25]
29  *                  = 0.5 + ((0.5-x)-x*R)  if x in [0.25,0.84375]
30  *	   where R = P/Q where P is an odd poly of degree 8 and
31  *	   Q is an odd poly of degree 10.
32  *						 -57.90
33  *			| R - (erf(x)-x)/x | <= 2
34  *
35  *
36  *	   Remark. The formula is derived by noting
37  *          erf(x) = (2/sqrt(pi))*(x - x^3/3 + x^5/10 - x^7/42 + ....)
38  *	   and that
39  *          2/sqrt(pi) = 1.128379167095512573896158903121545171688
40  *	   is close to one. The interval is chosen because the fix
41  *	   point of erf(x) is near 0.6174 (i.e., erf(x)=x when x is
42  *	   near 0.6174), and by some experiment, 0.84375 is chosen to
43  * 	   guarantee the error is less than one ulp for erf.
44  *
45  *      2. For |x| in [0.84375,1.25], let s = |x| - 1, and
46  *         c = 0.84506291151 rounded to single (24 bits)
47  *         	erf(x)  = sign(x) * (c  + P1(s)/Q1(s))
48  *         	erfc(x) = (1-c)  - P1(s)/Q1(s) if x > 0
49  *			  1+(c+P1(s)/Q1(s))    if x < 0
50  *         	|P1/Q1 - (erf(|x|)-c)| <= 2**-59.06
51  *	   Remark: here we use the taylor series expansion at x=1.
52  *		erf(1+s) = erf(1) + s*Poly(s)
53  *			 = 0.845.. + P1(s)/Q1(s)
54  *	   That is, we use rational approximation to approximate
55  *			erf(1+s) - (c = (single)0.84506291151)
56  *	   Note that |P1/Q1|< 0.078 for x in [0.84375,1.25]
57  *	   where
58  *		P1(s) = degree 6 poly in s
59  *		Q1(s) = degree 6 poly in s
60  *
61  *      3. For x in [1.25,1/0.35(~2.857143)],
62  *         	erfc(x) = (1/x)*exp(-x*x-0.5625+R1/S1)
63  *         	erf(x)  = 1 - erfc(x)
64  *	   where
65  *		R1(z) = degree 7 poly in z, (z=1/x^2)
66  *		S1(z) = degree 8 poly in z
67  *
68  *      4. For x in [1/0.35,28]
69  *         	erfc(x) = (1/x)*exp(-x*x-0.5625+R2/S2) if x > 0
70  *			= 2.0 - (1/x)*exp(-x*x-0.5625+R2/S2) if -6<x<0
71  *			= 2.0 - tiny		(if x <= -6)
72  *         	erf(x)  = sign(x)*(1.0 - erfc(x)) if x < 6, else
73  *         	erf(x)  = sign(x)*(1.0 - tiny)
74  *	   where
75  *		R2(z) = degree 6 poly in z, (z=1/x^2)
76  *		S2(z) = degree 7 poly in z
77  *
78  *      Note1:
79  *	   To compute exp(-x*x-0.5625+R/S), let s be a single
80  *	   precision number and s := x; then
81  *		-x*x = -s*s + (s-x)*(s+x)
82  *	        exp(-x*x-0.5626+R/S) =
83  *			exp(-s*s-0.5625)*exp((s-x)*(s+x)+R/S);
84  *      Note2:
85  *	   Here 4 and 5 make use of the asymptotic series
86  *			  exp(-x*x)
87  *		erfc(x) ~ ---------- * ( 1 + Poly(1/x^2) )
88  *			  x*sqrt(pi)
89  *	   We use rational approximation to approximate
90  *      	g(s)=f(1/x^2) = log(erfc(x)*x) - x*x + 0.5625
91  *	   Here is the error bound for R1/S1 and R2/S2
92  *      	|R1/S1 - f(x)|  < 2**(-62.57)
93  *      	|R2/S2 - f(x)|  < 2**(-61.52)
94  *
95  *      5. For inf > x >= 28
96  *         	erf(x)  = sign(x) *(1 - tiny)  (raise inexact)
97  *         	erfc(x) = tiny*tiny (raise underflow) if x > 0
98  *			= 2 - tiny if x<0
99  *
100  *      7. Special case:
101  *         	erf(0)  = 0, erf(inf)  = 1, erf(-inf) = -1,
102  *         	erfc(0) = 1, erfc(inf) = 0, erfc(-inf) = 2,
103  *	   	erfc/erf(NaN) is NaN
104  */
105 
106 #include <float.h>
107 #include "math.h"
108 #include "math_private.h"
109 
110 /* XXX Prevent compilers from erroneously constant folding: */
111 static const volatile double tiny= 1e-300;
112 
113 static const double
114 half= 0.5,
115 one = 1,
116 two = 2,
117 /* c = (float)0.84506291151 */
118 erx =  8.45062911510467529297e-01, /* 0x3FEB0AC1, 0x60000000 */
119 /*
120  * In the domain [0, 2**-28], only the first term in the power series
121  * expansion of erf(x) is used.  The magnitude of the first neglected
122  * terms is less than 2**-84.
123  */
124 efx =  1.28379167095512586316e-01, /* 0x3FC06EBA, 0x8214DB69 */
125 efx8=  1.02703333676410069053e+00, /* 0x3FF06EBA, 0x8214DB69 */
126 /*
127  * Coefficients for approximation to erf on [0,0.84375]
128  */
129 pp0  =  1.28379167095512558561e-01, /* 0x3FC06EBA, 0x8214DB68 */
130 pp1  = -3.25042107247001499370e-01, /* 0xBFD4CD7D, 0x691CB913 */
131 pp2  = -2.84817495755985104766e-02, /* 0xBF9D2A51, 0xDBD7194F */
132 pp3  = -5.77027029648944159157e-03, /* 0xBF77A291, 0x236668E4 */
133 pp4  = -2.37630166566501626084e-05, /* 0xBEF8EAD6, 0x120016AC */
134 qq1  =  3.97917223959155352819e-01, /* 0x3FD97779, 0xCDDADC09 */
135 qq2  =  6.50222499887672944485e-02, /* 0x3FB0A54C, 0x5536CEBA */
136 qq3  =  5.08130628187576562776e-03, /* 0x3F74D022, 0xC4D36B0F */
137 qq4  =  1.32494738004321644526e-04, /* 0x3F215DC9, 0x221C1A10 */
138 qq5  = -3.96022827877536812320e-06, /* 0xBED09C43, 0x42A26120 */
139 /*
140  * Coefficients for approximation to erf in [0.84375,1.25]
141  */
142 pa0  = -2.36211856075265944077e-03, /* 0xBF6359B8, 0xBEF77538 */
143 pa1  =  4.14856118683748331666e-01, /* 0x3FDA8D00, 0xAD92B34D */
144 pa2  = -3.72207876035701323847e-01, /* 0xBFD7D240, 0xFBB8C3F1 */
145 pa3  =  3.18346619901161753674e-01, /* 0x3FD45FCA, 0x805120E4 */
146 pa4  = -1.10894694282396677476e-01, /* 0xBFBC6398, 0x3D3E28EC */
147 pa5  =  3.54783043256182359371e-02, /* 0x3FA22A36, 0x599795EB */
148 pa6  = -2.16637559486879084300e-03, /* 0xBF61BF38, 0x0A96073F */
149 qa1  =  1.06420880400844228286e-01, /* 0x3FBB3E66, 0x18EEE323 */
150 qa2  =  5.40397917702171048937e-01, /* 0x3FE14AF0, 0x92EB6F33 */
151 qa3  =  7.18286544141962662868e-02, /* 0x3FB2635C, 0xD99FE9A7 */
152 qa4  =  1.26171219808761642112e-01, /* 0x3FC02660, 0xE763351F */
153 qa5  =  1.36370839120290507362e-02, /* 0x3F8BEDC2, 0x6B51DD1C */
154 qa6  =  1.19844998467991074170e-02, /* 0x3F888B54, 0x5735151D */
155 /*
156  * Coefficients for approximation to erfc in [1.25,1/0.35]
157  */
158 ra0  = -9.86494403484714822705e-03, /* 0xBF843412, 0x600D6435 */
159 ra1  = -6.93858572707181764372e-01, /* 0xBFE63416, 0xE4BA7360 */
160 ra2  = -1.05586262253232909814e+01, /* 0xC0251E04, 0x41B0E726 */
161 ra3  = -6.23753324503260060396e+01, /* 0xC04F300A, 0xE4CBA38D */
162 ra4  = -1.62396669462573470355e+02, /* 0xC0644CB1, 0x84282266 */
163 ra5  = -1.84605092906711035994e+02, /* 0xC067135C, 0xEBCCABB2 */
164 ra6  = -8.12874355063065934246e+01, /* 0xC0545265, 0x57E4D2F2 */
165 ra7  = -9.81432934416914548592e+00, /* 0xC023A0EF, 0xC69AC25C */
166 sa1  =  1.96512716674392571292e+01, /* 0x4033A6B9, 0xBD707687 */
167 sa2  =  1.37657754143519042600e+02, /* 0x4061350C, 0x526AE721 */
168 sa3  =  4.34565877475229228821e+02, /* 0x407B290D, 0xD58A1A71 */
169 sa4  =  6.45387271733267880336e+02, /* 0x40842B19, 0x21EC2868 */
170 sa5  =  4.29008140027567833386e+02, /* 0x407AD021, 0x57700314 */
171 sa6  =  1.08635005541779435134e+02, /* 0x405B28A3, 0xEE48AE2C */
172 sa7  =  6.57024977031928170135e+00, /* 0x401A47EF, 0x8E484A93 */
173 sa8  = -6.04244152148580987438e-02, /* 0xBFAEEFF2, 0xEE749A62 */
174 /*
175  * Coefficients for approximation to erfc in [1/.35,28]
176  */
177 rb0  = -9.86494292470009928597e-03, /* 0xBF843412, 0x39E86F4A */
178 rb1  = -7.99283237680523006574e-01, /* 0xBFE993BA, 0x70C285DE */
179 rb2  = -1.77579549177547519889e+01, /* 0xC031C209, 0x555F995A */
180 rb3  = -1.60636384855821916062e+02, /* 0xC064145D, 0x43C5ED98 */
181 rb4  = -6.37566443368389627722e+02, /* 0xC083EC88, 0x1375F228 */
182 rb5  = -1.02509513161107724954e+03, /* 0xC0900461, 0x6A2E5992 */
183 rb6  = -4.83519191608651397019e+02, /* 0xC07E384E, 0x9BDC383F */
184 sb1  =  3.03380607434824582924e+01, /* 0x403E568B, 0x261D5190 */
185 sb2  =  3.25792512996573918826e+02, /* 0x40745CAE, 0x221B9F0A */
186 sb3  =  1.53672958608443695994e+03, /* 0x409802EB, 0x189D5118 */
187 sb4  =  3.19985821950859553908e+03, /* 0x40A8FFB7, 0x688C246A */
188 sb5  =  2.55305040643316442583e+03, /* 0x40A3F219, 0xCEDF3BE6 */
189 sb6  =  4.74528541206955367215e+02, /* 0x407DA874, 0xE79FE763 */
190 sb7  = -2.24409524465858183362e+01; /* 0xC03670E2, 0x42712D62 */
191 
192 double
193 erf(double x)
194 {
195 	int32_t hx,ix,i;
196 	double R,S,P,Q,s,y,z,r;
197 	GET_HIGH_WORD(hx,x);
198 	ix = hx&0x7fffffff;
199 	if(ix>=0x7ff00000) {		/* erf(nan)=nan */
200 	    i = ((u_int32_t)hx>>31)<<1;
201 	    return (double)(1-i)+one/x;	/* erf(+-inf)=+-1 */
202 	}
203 
204 	if(ix < 0x3feb0000) {		/* |x|<0.84375 */
205 	    if(ix < 0x3e300000) { 	/* |x|<2**-28 */
206 	        if (ix < 0x00800000)
207 		    return (8*x+efx8*x)/8;	/* avoid spurious underflow */
208 		return x + efx*x;
209 	    }
210 	    z = x*x;
211 	    r = pp0+z*(pp1+z*(pp2+z*(pp3+z*pp4)));
212 	    s = one+z*(qq1+z*(qq2+z*(qq3+z*(qq4+z*qq5))));
213 	    y = r/s;
214 	    return x + x*y;
215 	}
216 	if(ix < 0x3ff40000) {		/* 0.84375 <= |x| < 1.25 */
217 	    s = fabs(x)-one;
218 	    P = pa0+s*(pa1+s*(pa2+s*(pa3+s*(pa4+s*(pa5+s*pa6)))));
219 	    Q = one+s*(qa1+s*(qa2+s*(qa3+s*(qa4+s*(qa5+s*qa6)))));
220 	    if(hx>=0) return erx + P/Q; else return -erx - P/Q;
221 	}
222 	if (ix >= 0x40180000) {		/* inf>|x|>=6 */
223 	    if(hx>=0) return one-tiny; else return tiny-one;
224 	}
225 	x = fabs(x);
226  	s = one/(x*x);
227 	if(ix< 0x4006DB6E) {	/* |x| < 1/0.35 */
228 	    R=ra0+s*(ra1+s*(ra2+s*(ra3+s*(ra4+s*(ra5+s*(ra6+s*ra7))))));
229 	    S=one+s*(sa1+s*(sa2+s*(sa3+s*(sa4+s*(sa5+s*(sa6+s*(sa7+
230 		s*sa8)))))));
231 	} else {	/* |x| >= 1/0.35 */
232 	    R=rb0+s*(rb1+s*(rb2+s*(rb3+s*(rb4+s*(rb5+s*rb6)))));
233 	    S=one+s*(sb1+s*(sb2+s*(sb3+s*(sb4+s*(sb5+s*(sb6+s*sb7))))));
234 	}
235 	z  = x;
236 	SET_LOW_WORD(z,0);
237 	r  =  exp(-z*z-0.5625)*exp((z-x)*(z+x)+R/S);
238 	if(hx>=0) return one-r/x; else return  r/x-one;
239 }
240 
241 #if (LDBL_MANT_DIG == 53)
242 __weak_reference(erf, erfl);
243 #endif
244 
245 double
246 erfc(double x)
247 {
248 	int32_t hx,ix;
249 	double R,S,P,Q,s,y,z,r;
250 	GET_HIGH_WORD(hx,x);
251 	ix = hx&0x7fffffff;
252 	if(ix>=0x7ff00000) {			/* erfc(nan)=nan */
253 						/* erfc(+-inf)=0,2 */
254 	    return (double)(((u_int32_t)hx>>31)<<1)+one/x;
255 	}
256 
257 	if(ix < 0x3feb0000) {		/* |x|<0.84375 */
258 	    if(ix < 0x3c700000)  	/* |x|<2**-56 */
259 		return one-x;
260 	    z = x*x;
261 	    r = pp0+z*(pp1+z*(pp2+z*(pp3+z*pp4)));
262 	    s = one+z*(qq1+z*(qq2+z*(qq3+z*(qq4+z*qq5))));
263 	    y = r/s;
264 	    if(hx < 0x3fd00000) {  	/* x<1/4 */
265 		return one-(x+x*y);
266 	    } else {
267 		r = x*y;
268 		r += (x-half);
269 	        return half - r ;
270 	    }
271 	}
272 	if(ix < 0x3ff40000) {		/* 0.84375 <= |x| < 1.25 */
273 	    s = fabs(x)-one;
274 	    P = pa0+s*(pa1+s*(pa2+s*(pa3+s*(pa4+s*(pa5+s*pa6)))));
275 	    Q = one+s*(qa1+s*(qa2+s*(qa3+s*(qa4+s*(qa5+s*qa6)))));
276 	    if(hx>=0) {
277 	        z  = one-erx; return z - P/Q;
278 	    } else {
279 		z = erx+P/Q; return one+z;
280 	    }
281 	}
282 	if (ix < 0x403c0000) {		/* |x|<28 */
283 	    x = fabs(x);
284  	    s = one/(x*x);
285 	    if(ix< 0x4006DB6D) {	/* |x| < 1/.35 ~ 2.857143*/
286 		R=ra0+s*(ra1+s*(ra2+s*(ra3+s*(ra4+s*(ra5+s*(ra6+s*ra7))))));
287 		S=one+s*(sa1+s*(sa2+s*(sa3+s*(sa4+s*(sa5+s*(sa6+s*(sa7+
288 		    s*sa8)))))));
289 	    } else {			/* |x| >= 1/.35 ~ 2.857143 */
290 		if(hx<0&&ix>=0x40180000) return two-tiny;/* x < -6 */
291 		R=rb0+s*(rb1+s*(rb2+s*(rb3+s*(rb4+s*(rb5+s*rb6)))));
292 		S=one+s*(sb1+s*(sb2+s*(sb3+s*(sb4+s*(sb5+s*(sb6+s*sb7))))));
293 	    }
294 	    z  = x;
295 	    SET_LOW_WORD(z,0);
296 	    r  =  exp(-z*z-0.5625)*exp((z-x)*(z+x)+R/S);
297 	    if(hx>0) return r/x; else return two-r/x;
298 	} else {
299 	    if(hx>0) return tiny*tiny; else return two-tiny;
300 	}
301 }
302 
303 #if (LDBL_MANT_DIG == 53)
304 __weak_reference(erfc, erfcl);
305 #endif
306