189
|
1 |
/*
|
|
2 |
* Portions Copyright (c) 2006, 2009 Nokia Corporation and/or its subsidiary(-ies).
|
|
3 |
* All rights reserved.
|
|
4 |
* This component and the accompanying materials are made available
|
|
5 |
* under the terms of "Eclipse Public License v1.0"
|
|
6 |
* which accompanies this distribution, and is available
|
|
7 |
* at the URL "http://www.eclipse.org/legal/epl-v10.html".
|
|
8 |
*
|
|
9 |
* Initial Contributors:
|
|
10 |
* Nokia Corporation - initial contribution.
|
|
11 |
*
|
|
12 |
* Contributors:
|
|
13 |
*
|
|
14 |
* Description:
|
|
15 |
*/
|
|
16 |
|
|
17 |
/* @(#)s_erf.c 5.1 93/09/24 */
|
|
18 |
/*
|
|
19 |
* ====================================================
|
|
20 |
* Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved.
|
|
21 |
*
|
|
22 |
* Developed at SunPro, a Sun Microsystems, Inc. business.
|
|
23 |
* Permission to use, copy, modify, and distribute this
|
|
24 |
* software is freely granted, provided that this notice
|
|
25 |
* is preserved.
|
|
26 |
* ====================================================
|
|
27 |
*/
|
|
28 |
#ifndef __SYMBIAN32__
|
|
29 |
#ifndef lint
|
|
30 |
static char rcsid[] = "$FreeBSD: src/lib/msun/src/s_erf.c,v 1.7 2002/05/28 18:15:04 alfred Exp $";
|
|
31 |
#endif
|
|
32 |
#endif //__SYMBIAN32__
|
|
33 |
|
|
34 |
#include <e32std.h>
|
|
35 |
|
|
36 |
#include "openc.h"
|
|
37 |
|
|
38 |
/* double erf(double x)
|
|
39 |
* double erfc(double x)
|
|
40 |
* x
|
|
41 |
* 2 |\
|
|
42 |
* erf(x) = --------- | exp(-t*t)dt
|
|
43 |
* sqrt(pi) \|
|
|
44 |
* 0
|
|
45 |
*
|
|
46 |
* erfc(x) = 1-erf(x)
|
|
47 |
* Note that
|
|
48 |
* erf(-x) = -erf(x)
|
|
49 |
* erfc(-x) = 2 - erfc(x)
|
|
50 |
*
|
|
51 |
* Method:
|
|
52 |
* 1. For |x| in [0, 0.84375]
|
|
53 |
* erf(x) = x + x*R(x^2)
|
|
54 |
* erfc(x) = 1 - erf(x) if x in [-.84375,0.25]
|
|
55 |
* = 0.5 + ((0.5-x)-x*R) if x in [0.25,0.84375]
|
|
56 |
* where R = P/Q where P is an odd poly of degree 8 and
|
|
57 |
* Q is an odd poly of degree 10.
|
|
58 |
* -57.90
|
|
59 |
* | R - (erf(x)-x)/x | <= 2
|
|
60 |
*
|
|
61 |
*
|
|
62 |
* Remark. The formula is derived by noting
|
|
63 |
* erf(x) = (2/sqrt(pi))*(x - x^3/3 + x^5/10 - x^7/42 + ....)
|
|
64 |
* and that
|
|
65 |
* 2/sqrt(pi) = 1.128379167095512573896158903121545171688
|
|
66 |
* is close to one. The interval is chosen because the fix
|
|
67 |
* point of erf(x) is near 0.6174 (i.e., erf(x)=x when x is
|
|
68 |
* near 0.6174), and by some experiment, 0.84375 is chosen to
|
|
69 |
* guarantee the error is less than one ulp for erf.
|
|
70 |
*
|
|
71 |
* 2. For |x| in [0.84375,1.25], let s = |x| - 1, and
|
|
72 |
* c = 0.84506291151 rounded to single (24 bits)
|
|
73 |
* erf(x) = sign(x) * (c + P1(s)/Q1(s))
|
|
74 |
* erfc(x) = (1-c) - P1(s)/Q1(s) if x > 0
|
|
75 |
* 1+(c+P1(s)/Q1(s)) if x < 0
|
|
76 |
* |P1/Q1 - (erf(|x|)-c)| <= 2**-59.06
|
|
77 |
* Remark: here we use the taylor series expansion at x=1.
|
|
78 |
* erf(1+s) = erf(1) + s*Poly(s)
|
|
79 |
* = 0.845.. + P1(s)/Q1(s)
|
|
80 |
* That is, we use rational approximation to approximate
|
|
81 |
* erf(1+s) - (c = (single)0.84506291151)
|
|
82 |
* Note that |P1/Q1|< 0.078 for x in [0.84375,1.25]
|
|
83 |
* where
|
|
84 |
* P1(s) = degree 6 poly in s
|
|
85 |
* Q1(s) = degree 6 poly in s
|
|
86 |
*
|
|
87 |
* 3. For x in [1.25,1/0.35(~2.857143)],
|
|
88 |
* erfc(x) = (1/x)*exp(-x*x-0.5625+R1/S1)
|
|
89 |
* erf(x) = 1 - erfc(x)
|
|
90 |
* where
|
|
91 |
* R1(z) = degree 7 poly in z, (z=1/x^2)
|
|
92 |
* S1(z) = degree 8 poly in z
|
|
93 |
*
|
|
94 |
* 4. For x in [1/0.35,28]
|
|
95 |
* erfc(x) = (1/x)*exp(-x*x-0.5625+R2/S2) if x > 0
|
|
96 |
* = 2.0 - (1/x)*exp(-x*x-0.5625+R2/S2) if -6<x<0
|
|
97 |
* = 2.0 - tiny (if x <= -6)
|
|
98 |
* erf(x) = sign(x)*(1.0 - erfc(x)) if x < 6, else
|
|
99 |
* erf(x) = sign(x)*(1.0 - tiny)
|
|
100 |
* where
|
|
101 |
* R2(z) = degree 6 poly in z, (z=1/x^2)
|
|
102 |
* S2(z) = degree 7 poly in z
|
|
103 |
*
|
|
104 |
* Note1:
|
|
105 |
* To compute exp(-x*x-0.5625+R/S), let s be a single
|
|
106 |
* precision number and s := x; then
|
|
107 |
* -x*x = -s*s + (s-x)*(s+x)
|
|
108 |
* exp(-x*x-0.5626+R/S) =
|
|
109 |
* exp(-s*s-0.5625)*exp((s-x)*(s+x)+R/S);
|
|
110 |
* Note2:
|
|
111 |
* Here 4 and 5 make use of the asymptotic series
|
|
112 |
* exp(-x*x)
|
|
113 |
* erfc(x) ~ ---------- * ( 1 + Poly(1/x^2) )
|
|
114 |
* x*sqrt(pi)
|
|
115 |
* We use rational approximation to approximate
|
|
116 |
* g(s)=f(1/x^2) = log(erfc(x)*x) - x*x + 0.5625
|
|
117 |
* Here is the error bound for R1/S1 and R2/S2
|
|
118 |
* |R1/S1 - f(x)| < 2**(-62.57)
|
|
119 |
* |R2/S2 - f(x)| < 2**(-61.52)
|
|
120 |
*
|
|
121 |
* 5. For inf > x >= 28
|
|
122 |
* erf(x) = sign(x) *(1 - tiny) (raise inexact)
|
|
123 |
* erfc(x) = tiny*tiny (raise underflow) if x > 0
|
|
124 |
* = 2 - tiny if x<0
|
|
125 |
*
|
|
126 |
* 7. Special case:
|
|
127 |
* erf(0) = 0, erf(inf) = 1, erf(-inf) = -1,
|
|
128 |
* erfc(0) = 1, erfc(inf) = 0, erfc(-inf) = 2,
|
|
129 |
* erfc/erf(NaN) is NaN
|
|
130 |
*/
|
|
131 |
|
|
132 |
|
|
133 |
|
|
134 |
|
|
135 |
|
|
136 |
////--------------------------------------------------
|
|
137 |
|
|
138 |
#define __ieee754_exp exp
|
|
139 |
typedef TUint32 u_int32_t;
|
|
140 |
typedef TInt32 int32_t;
|
|
141 |
|
|
142 |
typedef union
|
|
143 |
{
|
|
144 |
double value;
|
|
145 |
struct
|
|
146 |
{
|
|
147 |
u_int32_t lsw;
|
|
148 |
u_int32_t msw;
|
|
149 |
} parts;
|
|
150 |
} ieee_double_shape_type;
|
|
151 |
|
|
152 |
|
|
153 |
inline void GET_HIGH_WORD(int32_t& aHighWord, double aValue)
|
|
154 |
{
|
|
155 |
ieee_double_shape_type gh_u;
|
|
156 |
gh_u.value = aValue;
|
|
157 |
aHighWord = gh_u.parts.msw;
|
|
158 |
}
|
|
159 |
|
|
160 |
inline void SET_LOW_WORD(double& aValue, int32_t aLowWord)
|
|
161 |
{
|
|
162 |
ieee_double_shape_type sl_u;
|
|
163 |
sl_u.value = aValue;
|
|
164 |
sl_u.parts.lsw = aLowWord;
|
|
165 |
aValue = sl_u.value;
|
|
166 |
}
|
|
167 |
|
|
168 |
//----------------------------------------------------------------math_private.h
|
|
169 |
|
|
170 |
static const double tiny = 1e-300;
|
|
171 |
static const double tinySquare = 0.00; // tiny * tiny
|
|
172 |
static const double half = 5.00000000000000000000e-01; /* 0x3FE00000, 0x00000000 */
|
|
173 |
static const double one = 1.00000000000000000000e+00; /* 0x3FF00000, 0x00000000 */
|
|
174 |
static const double two = 2.00000000000000000000e+00; /* 0x40000000, 0x00000000 */
|
|
175 |
/* c = (float)0.84506291151 */
|
|
176 |
static const double erx = 8.45062911510467529297e-01; /* 0x3FEB0AC1, 0x60000000 */
|
|
177 |
/*
|
|
178 |
* Coefficients for approximation to erf on [0,0.84375]
|
|
179 |
*/
|
|
180 |
static const double efx = 1.28379167095512586316e-01; /* 0x3FC06EBA, 0x8214DB69 */
|
|
181 |
static const double efx8 = 1.02703333676410069053e+00; /* 0x3FF06EBA, 0x8214DB69 */
|
|
182 |
static const double pp0 = 1.28379167095512558561e-01; /* 0x3FC06EBA, 0x8214DB68 */
|
|
183 |
static const double pp1 = -3.25042107247001499370e-01; /* 0xBFD4CD7D, 0x691CB913 */
|
|
184 |
static const double pp2 = -2.84817495755985104766e-02; /* 0xBF9D2A51, 0xDBD7194F */
|
|
185 |
static const double pp3 = -5.77027029648944159157e-03; /* 0xBF77A291, 0x236668E4 */
|
|
186 |
static const double pp4 = -2.37630166566501626084e-05; /* 0xBEF8EAD6, 0x120016AC */
|
|
187 |
static const double qq1 = 3.97917223959155352819e-01; /* 0x3FD97779, 0xCDDADC09 */
|
|
188 |
static const double qq2 = 6.50222499887672944485e-02; /* 0x3FB0A54C, 0x5536CEBA */
|
|
189 |
static const double qq3 = 5.08130628187576562776e-03; /* 0x3F74D022, 0xC4D36B0F */
|
|
190 |
static const double qq4 = 1.32494738004321644526e-04; /* 0x3F215DC9, 0x221C1A10 */
|
|
191 |
static const double qq5 = -3.96022827877536812320e-06; /* 0xBED09C43, 0x42A26120 */
|
|
192 |
/*
|
|
193 |
* Coefficients for approximation to erf in [0.84375,1.25]
|
|
194 |
*/
|
|
195 |
static const double pa0 = -2.36211856075265944077e-03; /* 0xBF6359B8, 0xBEF77538 */
|
|
196 |
static const double pa1 = 4.14856118683748331666e-01; /* 0x3FDA8D00, 0xAD92B34D */
|
|
197 |
static const double pa2 = -3.72207876035701323847e-01; /* 0xBFD7D240, 0xFBB8C3F1 */
|
|
198 |
static const double pa3 = 3.18346619901161753674e-01; /* 0x3FD45FCA, 0x805120E4 */
|
|
199 |
static const double pa4 = -1.10894694282396677476e-01; /* 0xBFBC6398, 0x3D3E28EC */
|
|
200 |
static const double pa5 = 3.54783043256182359371e-02; /* 0x3FA22A36, 0x599795EB */
|
|
201 |
static const double pa6 = -2.16637559486879084300e-03; /* 0xBF61BF38, 0x0A96073F */
|
|
202 |
static const double qa1 = 1.06420880400844228286e-01; /* 0x3FBB3E66, 0x18EEE323 */
|
|
203 |
static const double qa2 = 5.40397917702171048937e-01; /* 0x3FE14AF0, 0x92EB6F33 */
|
|
204 |
static const double qa3 = 7.18286544141962662868e-02; /* 0x3FB2635C, 0xD99FE9A7 */
|
|
205 |
static const double qa4 = 1.26171219808761642112e-01; /* 0x3FC02660, 0xE763351F */
|
|
206 |
static const double qa5 = 1.36370839120290507362e-02; /* 0x3F8BEDC2, 0x6B51DD1C */
|
|
207 |
static const double qa6 = 1.19844998467991074170e-02; /* 0x3F888B54, 0x5735151D */
|
|
208 |
/*
|
|
209 |
* Coefficients for approximation to erfc in [1.25,1/0.35]
|
|
210 |
*/
|
|
211 |
static const double ra0 = -9.86494403484714822705e-03; /* 0xBF843412, 0x600D6435 */
|
|
212 |
static const double ra1 = -6.93858572707181764372e-01; /* 0xBFE63416, 0xE4BA7360 */
|
|
213 |
static const double ra2 = -1.05586262253232909814e+01; /* 0xC0251E04, 0x41B0E726 */
|
|
214 |
static const double ra3 = -6.23753324503260060396e+01; /* 0xC04F300A, 0xE4CBA38D */
|
|
215 |
static const double ra4 = -1.62396669462573470355e+02; /* 0xC0644CB1, 0x84282266 */
|
|
216 |
static const double ra5 = -1.84605092906711035994e+02; /* 0xC067135C, 0xEBCCABB2 */
|
|
217 |
static const double ra6 = -8.12874355063065934246e+01; /* 0xC0545265, 0x57E4D2F2 */
|
|
218 |
static const double ra7 = -9.81432934416914548592e+00; /* 0xC023A0EF, 0xC69AC25C */
|
|
219 |
static const double sa1 = 1.96512716674392571292e+01; /* 0x4033A6B9, 0xBD707687 */
|
|
220 |
static const double sa2 = 1.37657754143519042600e+02; /* 0x4061350C, 0x526AE721 */
|
|
221 |
static const double sa3 = 4.34565877475229228821e+02; /* 0x407B290D, 0xD58A1A71 */
|
|
222 |
static const double sa4 = 6.45387271733267880336e+02; /* 0x40842B19, 0x21EC2868 */
|
|
223 |
static const double sa5 = 4.29008140027567833386e+02; /* 0x407AD021, 0x57700314 */
|
|
224 |
static const double sa6 = 1.08635005541779435134e+02; /* 0x405B28A3, 0xEE48AE2C */
|
|
225 |
static const double sa7 = 6.57024977031928170135e+00; /* 0x401A47EF, 0x8E484A93 */
|
|
226 |
static const double sa8 = -6.04244152148580987438e-02; /* 0xBFAEEFF2, 0xEE749A62 */
|
|
227 |
/*
|
|
228 |
* Coefficients for approximation to erfc in [1/.35,28]
|
|
229 |
*/
|
|
230 |
static const double rb0 = -9.86494292470009928597e-03; /* 0xBF843412, 0x39E86F4A */
|
|
231 |
static const double rb1 = -7.99283237680523006574e-01; /* 0xBFE993BA, 0x70C285DE */
|
|
232 |
static const double rb2 = -1.77579549177547519889e+01; /* 0xC031C209, 0x555F995A */
|
|
233 |
static const double rb3 = -1.60636384855821916062e+02; /* 0xC064145D, 0x43C5ED98 */
|
|
234 |
static const double rb4 = -6.37566443368389627722e+02; /* 0xC083EC88, 0x1375F228 */
|
|
235 |
static const double rb5 = -1.02509513161107724954e+03; /* 0xC0900461, 0x6A2E5992 */
|
|
236 |
static const double rb6 = -4.83519191608651397019e+02; /* 0xC07E384E, 0x9BDC383F */
|
|
237 |
static const double sb1 = 3.03380607434824582924e+01; /* 0x403E568B, 0x261D5190 */
|
|
238 |
static const double sb2 = 3.25792512996573918826e+02; /* 0x40745CAE, 0x221B9F0A */
|
|
239 |
static const double sb3 = 1.53672958608443695994e+03; /* 0x409802EB, 0x189D5118 */
|
|
240 |
static const double sb4 = 3.19985821950859553908e+03; /* 0x40A8FFB7, 0x688C246A */
|
|
241 |
static const double sb5 = 2.55305040643316442583e+03; /* 0x40A3F219, 0xCEDF3BE6 */
|
|
242 |
static const double sb6 = 4.74528541206955367215e+02; /* 0x407DA874, 0xE79FE763 */
|
|
243 |
static const double sb7 = -2.24409524465858183362e+01; /* 0xC03670E2, 0x42712D62 */
|
|
244 |
|
|
245 |
double erf(double x)
|
|
246 |
{
|
|
247 |
TInt32 hx,ix,i;
|
|
248 |
double R,S,P,Q,s,y,z,r;
|
|
249 |
GET_HIGH_WORD(hx,x);
|
|
250 |
ix = hx&0x7fffffff;
|
|
251 |
if(ix>=0x7ff00000) { /* erf(nan)=nan */
|
|
252 |
i = ((TUint32)hx>>31)<<1;
|
|
253 |
return (double)(1-i)+one/x; /* erf(+-inf)=+-1 */
|
|
254 |
}
|
|
255 |
|
|
256 |
if(ix < 0x3feb0000) { /* |x|<0.84375 */
|
|
257 |
if(ix < 0x3e300000) { /* |x|<2**-28 */
|
|
258 |
if (ix < 0x00800000)
|
|
259 |
return 0.125*(8.0*x+efx8*x); /*avoid underflow */
|
|
260 |
return x + efx*x;
|
|
261 |
}
|
|
262 |
z = x*x;
|
|
263 |
r = pp0+z*(pp1+z*(pp2+z*(pp3+z*pp4)));
|
|
264 |
s = one+z*(qq1+z*(qq2+z*(qq3+z*(qq4+z*qq5))));
|
|
265 |
y = r/s;
|
|
266 |
return x + x*y;
|
|
267 |
}
|
|
268 |
if(ix < 0x3ff40000) { /* 0.84375 <= |x| < 1.25 */
|
|
269 |
s = fabs(x)-one;
|
|
270 |
P = pa0+s*(pa1+s*(pa2+s*(pa3+s*(pa4+s*(pa5+s*pa6)))));
|
|
271 |
Q = one+s*(qa1+s*(qa2+s*(qa3+s*(qa4+s*(qa5+s*qa6)))));
|
|
272 |
if(hx>=0) return erx + P/Q; else return -erx - P/Q;
|
|
273 |
}
|
|
274 |
if (ix >= 0x40180000) { /* inf>|x|>=6 */
|
|
275 |
if(hx>=0) return one-tiny; else return tiny-one;
|
|
276 |
}
|
|
277 |
x = fabs(x);
|
|
278 |
s = one/(x*x);
|
|
279 |
if(ix< 0x4006DB6E) { /* |x| < 1/0.35 */
|
|
280 |
R=ra0+s*(ra1+s*(ra2+s*(ra3+s*(ra4+s*(
|
|
281 |
ra5+s*(ra6+s*ra7))))));
|
|
282 |
S=one+s*(sa1+s*(sa2+s*(sa3+s*(sa4+s*(
|
|
283 |
sa5+s*(sa6+s*(sa7+s*sa8)))))));
|
|
284 |
} else { /* |x| >= 1/0.35 */
|
|
285 |
R=rb0+s*(rb1+s*(rb2+s*(rb3+s*(rb4+s*(
|
|
286 |
rb5+s*rb6)))));
|
|
287 |
S=one+s*(sb1+s*(sb2+s*(sb3+s*(sb4+s*(
|
|
288 |
sb5+s*(sb6+s*sb7))))));
|
|
289 |
}
|
|
290 |
z = x;
|
|
291 |
SET_LOW_WORD(z,0);
|
|
292 |
r = __ieee754_exp(-z*z-0.5625)*__ieee754_exp((z-x)*(z+x)+R/S);
|
|
293 |
if(hx>=0) return one-r/x; else return r/x-one;
|
|
294 |
}
|
|
295 |
|
|
296 |
double erfc(double x)
|
|
297 |
{
|
|
298 |
int32_t hx,ix;
|
|
299 |
double R,S,P,Q,s,y,z,r;
|
|
300 |
GET_HIGH_WORD(hx,x);
|
|
301 |
ix = hx&0x7fffffff;
|
|
302 |
if(ix>=0x7ff00000) { /* erfc(nan)=nan */
|
|
303 |
/* erfc(+-inf)=0,2 */
|
|
304 |
return (double)(((u_int32_t)hx>>31)<<1)+one/x;
|
|
305 |
}
|
|
306 |
|
|
307 |
if(ix < 0x3feb0000) { /* |x|<0.84375 */
|
|
308 |
if(ix < 0x3c700000) /* |x|<2**-56 */
|
|
309 |
return one-x;
|
|
310 |
z = x*x;
|
|
311 |
r = pp0+z*(pp1+z*(pp2+z*(pp3+z*pp4)));
|
|
312 |
s = one+z*(qq1+z*(qq2+z*(qq3+z*(qq4+z*qq5))));
|
|
313 |
y = r/s;
|
|
314 |
if(hx < 0x3fd00000) { /* x<1/4 */
|
|
315 |
return one-(x+x*y);
|
|
316 |
} else {
|
|
317 |
r = x*y;
|
|
318 |
r += (x-half);
|
|
319 |
return half - r ;
|
|
320 |
}
|
|
321 |
}
|
|
322 |
if(ix < 0x3ff40000) { /* 0.84375 <= |x| < 1.25 */
|
|
323 |
s = fabs(x)-one;
|
|
324 |
P = pa0+s*(pa1+s*(pa2+s*(pa3+s*(pa4+s*(pa5+s*pa6)))));
|
|
325 |
Q = one+s*(qa1+s*(qa2+s*(qa3+s*(qa4+s*(qa5+s*qa6)))));
|
|
326 |
if(hx>=0) {
|
|
327 |
z = one-erx; return z - P/Q;
|
|
328 |
} else {
|
|
329 |
z = erx+P/Q; return one+z;
|
|
330 |
}
|
|
331 |
}
|
|
332 |
if (ix < 0x403c0000) { /* |x|<28 */
|
|
333 |
x = fabs(x);
|
|
334 |
s = one/(x*x);
|
|
335 |
if(ix< 0x4006DB6D) { /* |x| < 1/.35 ~ 2.857143*/
|
|
336 |
R=ra0+s*(ra1+s*(ra2+s*(ra3+s*(ra4+s*(
|
|
337 |
ra5+s*(ra6+s*ra7))))));
|
|
338 |
S=one+s*(sa1+s*(sa2+s*(sa3+s*(sa4+s*(
|
|
339 |
sa5+s*(sa6+s*(sa7+s*sa8)))))));
|
|
340 |
} else { /* |x| >= 1/.35 ~ 2.857143 */
|
|
341 |
if(hx<0&&ix>=0x40180000) return two-tiny;/* x < -6 */
|
|
342 |
R=rb0+s*(rb1+s*(rb2+s*(rb3+s*(rb4+s*(
|
|
343 |
rb5+s*rb6)))));
|
|
344 |
S=one+s*(sb1+s*(sb2+s*(sb3+s*(sb4+s*(
|
|
345 |
sb5+s*(sb6+s*sb7))))));
|
|
346 |
}
|
|
347 |
z = x;
|
|
348 |
SET_LOW_WORD(z,0);
|
|
349 |
r = __ieee754_exp(-z*z-0.5625)*
|
|
350 |
__ieee754_exp((z-x)*(z+x)+R/S);
|
|
351 |
if(hx>0) return r/x; else return two-r/x;
|
|
352 |
} else {
|
|
353 |
if(hx>0) return tinySquare; else return two-tiny;
|
|
354 |
}
|
|
355 |
}
|