1 |
|
|
/* s_erff.c -- float version of s_erf.c. |
2 |
|
|
* Conversion to float by Ian Lance Taylor, Cygnus Support, ian@cygnus.com. |
3 |
|
|
*/ |
4 |
|
|
|
5 |
|
|
/* |
6 |
|
|
* ==================================================== |
7 |
|
|
* Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved. |
8 |
|
|
* |
9 |
|
|
* Developed at SunPro, a Sun Microsystems, Inc. business. |
10 |
|
|
* Permission to use, copy, modify, and distribute this |
11 |
|
|
* software is freely granted, provided that this notice |
12 |
|
|
* is preserved. |
13 |
|
|
* ==================================================== |
14 |
|
|
*/ |
15 |
|
|
|
16 |
|
|
#include "math.h" |
17 |
|
|
#include "math_private.h" |
18 |
|
|
|
19 |
|
|
static const float |
20 |
|
|
tiny = 1e-30, |
21 |
|
|
half= 5.0000000000e-01, /* 0x3F000000 */ |
22 |
|
|
one = 1.0000000000e+00, /* 0x3F800000 */ |
23 |
|
|
two = 2.0000000000e+00, /* 0x40000000 */ |
24 |
|
|
/* c = (subfloat)0.84506291151 */ |
25 |
|
|
erx = 8.4506291151e-01, /* 0x3f58560b */ |
26 |
|
|
/* |
27 |
|
|
* Coefficients for approximation to erf on [0,0.84375] |
28 |
|
|
*/ |
29 |
|
|
efx = 1.2837916613e-01, /* 0x3e0375d4 */ |
30 |
|
|
efx8= 1.0270333290e+00, /* 0x3f8375d4 */ |
31 |
|
|
pp0 = 1.2837916613e-01, /* 0x3e0375d4 */ |
32 |
|
|
pp1 = -3.2504209876e-01, /* 0xbea66beb */ |
33 |
|
|
pp2 = -2.8481749818e-02, /* 0xbce9528f */ |
34 |
|
|
pp3 = -5.7702702470e-03, /* 0xbbbd1489 */ |
35 |
|
|
pp4 = -2.3763017452e-05, /* 0xb7c756b1 */ |
36 |
|
|
qq1 = 3.9791721106e-01, /* 0x3ecbbbce */ |
37 |
|
|
qq2 = 6.5022252500e-02, /* 0x3d852a63 */ |
38 |
|
|
qq3 = 5.0813062117e-03, /* 0x3ba68116 */ |
39 |
|
|
qq4 = 1.3249473704e-04, /* 0x390aee49 */ |
40 |
|
|
qq5 = -3.9602282413e-06, /* 0xb684e21a */ |
41 |
|
|
/* |
42 |
|
|
* Coefficients for approximation to erf in [0.84375,1.25] |
43 |
|
|
*/ |
44 |
|
|
pa0 = -2.3621185683e-03, /* 0xbb1acdc6 */ |
45 |
|
|
pa1 = 4.1485610604e-01, /* 0x3ed46805 */ |
46 |
|
|
pa2 = -3.7220788002e-01, /* 0xbebe9208 */ |
47 |
|
|
pa3 = 3.1834661961e-01, /* 0x3ea2fe54 */ |
48 |
|
|
pa4 = -1.1089469492e-01, /* 0xbde31cc2 */ |
49 |
|
|
pa5 = 3.5478305072e-02, /* 0x3d1151b3 */ |
50 |
|
|
pa6 = -2.1663755178e-03, /* 0xbb0df9c0 */ |
51 |
|
|
qa1 = 1.0642088205e-01, /* 0x3dd9f331 */ |
52 |
|
|
qa2 = 5.4039794207e-01, /* 0x3f0a5785 */ |
53 |
|
|
qa3 = 7.1828655899e-02, /* 0x3d931ae7 */ |
54 |
|
|
qa4 = 1.2617121637e-01, /* 0x3e013307 */ |
55 |
|
|
qa5 = 1.3637083583e-02, /* 0x3c5f6e13 */ |
56 |
|
|
qa6 = 1.1984500103e-02, /* 0x3c445aa3 */ |
57 |
|
|
/* |
58 |
|
|
* Coefficients for approximation to erfc in [1.25,1/0.35] |
59 |
|
|
*/ |
60 |
|
|
ra0 = -9.8649440333e-03, /* 0xbc21a093 */ |
61 |
|
|
ra1 = -6.9385856390e-01, /* 0xbf31a0b7 */ |
62 |
|
|
ra2 = -1.0558626175e+01, /* 0xc128f022 */ |
63 |
|
|
ra3 = -6.2375331879e+01, /* 0xc2798057 */ |
64 |
|
|
ra4 = -1.6239666748e+02, /* 0xc322658c */ |
65 |
|
|
ra5 = -1.8460508728e+02, /* 0xc3389ae7 */ |
66 |
|
|
ra6 = -8.1287437439e+01, /* 0xc2a2932b */ |
67 |
|
|
ra7 = -9.8143291473e+00, /* 0xc11d077e */ |
68 |
|
|
sa1 = 1.9651271820e+01, /* 0x419d35ce */ |
69 |
|
|
sa2 = 1.3765776062e+02, /* 0x4309a863 */ |
70 |
|
|
sa3 = 4.3456588745e+02, /* 0x43d9486f */ |
71 |
|
|
sa4 = 6.4538726807e+02, /* 0x442158c9 */ |
72 |
|
|
sa5 = 4.2900814819e+02, /* 0x43d6810b */ |
73 |
|
|
sa6 = 1.0863500214e+02, /* 0x42d9451f */ |
74 |
|
|
sa7 = 6.5702495575e+00, /* 0x40d23f7c */ |
75 |
|
|
sa8 = -6.0424413532e-02, /* 0xbd777f97 */ |
76 |
|
|
/* |
77 |
|
|
* Coefficients for approximation to erfc in [1/.35,28] |
78 |
|
|
*/ |
79 |
|
|
rb0 = -9.8649431020e-03, /* 0xbc21a092 */ |
80 |
|
|
rb1 = -7.9928326607e-01, /* 0xbf4c9dd4 */ |
81 |
|
|
rb2 = -1.7757955551e+01, /* 0xc18e104b */ |
82 |
|
|
rb3 = -1.6063638306e+02, /* 0xc320a2ea */ |
83 |
|
|
rb4 = -6.3756646729e+02, /* 0xc41f6441 */ |
84 |
|
|
rb5 = -1.0250950928e+03, /* 0xc480230b */ |
85 |
|
|
rb6 = -4.8351919556e+02, /* 0xc3f1c275 */ |
86 |
|
|
sb1 = 3.0338060379e+01, /* 0x41f2b459 */ |
87 |
|
|
sb2 = 3.2579251099e+02, /* 0x43a2e571 */ |
88 |
|
|
sb3 = 1.5367296143e+03, /* 0x44c01759 */ |
89 |
|
|
sb4 = 3.1998581543e+03, /* 0x4547fdbb */ |
90 |
|
|
sb5 = 2.5530502930e+03, /* 0x451f90ce */ |
91 |
|
|
sb6 = 4.7452853394e+02, /* 0x43ed43a7 */ |
92 |
|
|
sb7 = -2.2440952301e+01; /* 0xc1b38712 */ |
93 |
|
|
|
94 |
|
|
float |
95 |
|
|
erff(float x) |
96 |
|
|
{ |
97 |
|
|
int32_t hx,ix,i; |
98 |
|
|
float R,S,P,Q,s,y,z,r; |
99 |
|
|
GET_FLOAT_WORD(hx,x); |
100 |
|
|
ix = hx&0x7fffffff; |
101 |
|
|
if(ix>=0x7f800000) { /* erf(nan)=nan */ |
102 |
|
|
i = ((u_int32_t)hx>>31)<<1; |
103 |
|
|
return (float)(1-i)+one/x; /* erf(+-inf)=+-1 */ |
104 |
|
|
} |
105 |
|
|
|
106 |
|
|
if(ix < 0x3f580000) { /* |x|<0.84375 */ |
107 |
|
|
if(ix < 0x31800000) { /* |x|<2**-28 */ |
108 |
|
|
if (ix < 0x04000000) |
109 |
|
|
/*avoid underflow */ |
110 |
|
|
return (float)0.125*((float)8.0*x+efx8*x); |
111 |
|
|
return x + efx*x; |
112 |
|
|
} |
113 |
|
|
z = x*x; |
114 |
|
|
r = pp0+z*(pp1+z*(pp2+z*(pp3+z*pp4))); |
115 |
|
|
s = one+z*(qq1+z*(qq2+z*(qq3+z*(qq4+z*qq5)))); |
116 |
|
|
y = r/s; |
117 |
|
|
return x + x*y; |
118 |
|
|
} |
119 |
|
|
if(ix < 0x3fa00000) { /* 0.84375 <= |x| < 1.25 */ |
120 |
|
|
s = fabsf(x)-one; |
121 |
|
|
P = pa0+s*(pa1+s*(pa2+s*(pa3+s*(pa4+s*(pa5+s*pa6))))); |
122 |
|
|
Q = one+s*(qa1+s*(qa2+s*(qa3+s*(qa4+s*(qa5+s*qa6))))); |
123 |
|
|
if(hx>=0) return erx + P/Q; else return -erx - P/Q; |
124 |
|
|
} |
125 |
|
|
if (ix >= 0x40c00000) { /* inf>|x|>=6 */ |
126 |
|
|
if(hx>=0) return one-tiny; else return tiny-one; |
127 |
|
|
} |
128 |
|
|
x = fabsf(x); |
129 |
|
|
s = one/(x*x); |
130 |
|
|
if(ix< 0x4036DB6E) { /* |x| < 1/0.35 */ |
131 |
|
|
R=ra0+s*(ra1+s*(ra2+s*(ra3+s*(ra4+s*( |
132 |
|
|
ra5+s*(ra6+s*ra7)))))); |
133 |
|
|
S=one+s*(sa1+s*(sa2+s*(sa3+s*(sa4+s*( |
134 |
|
|
sa5+s*(sa6+s*(sa7+s*sa8))))))); |
135 |
|
|
} else { /* |x| >= 1/0.35 */ |
136 |
|
|
R=rb0+s*(rb1+s*(rb2+s*(rb3+s*(rb4+s*( |
137 |
|
|
rb5+s*rb6))))); |
138 |
|
|
S=one+s*(sb1+s*(sb2+s*(sb3+s*(sb4+s*( |
139 |
|
|
sb5+s*(sb6+s*sb7)))))); |
140 |
|
|
} |
141 |
|
|
GET_FLOAT_WORD(ix,x); |
142 |
|
|
SET_FLOAT_WORD(z,ix&0xfffff000); |
143 |
|
|
r = expf(-z*z-(float)0.5625)*expf((z-x)*(z+x)+R/S); |
144 |
|
|
if(hx>=0) return one-r/x; else return r/x-one; |
145 |
|
|
} |
146 |
|
|
|
147 |
|
|
float |
148 |
|
|
erfcf(float x) |
149 |
|
|
{ |
150 |
|
|
int32_t hx,ix; |
151 |
|
|
float R,S,P,Q,s,y,z,r; |
152 |
|
|
GET_FLOAT_WORD(hx,x); |
153 |
|
|
ix = hx&0x7fffffff; |
154 |
|
|
if(ix>=0x7f800000) { /* erfc(nan)=nan */ |
155 |
|
|
/* erfc(+-inf)=0,2 */ |
156 |
|
|
return (float)(((u_int32_t)hx>>31)<<1)+one/x; |
157 |
|
|
} |
158 |
|
|
|
159 |
|
|
if(ix < 0x3f580000) { /* |x|<0.84375 */ |
160 |
|
|
if(ix < 0x23800000) /* |x|<2**-56 */ |
161 |
|
|
return one-x; |
162 |
|
|
z = x*x; |
163 |
|
|
r = pp0+z*(pp1+z*(pp2+z*(pp3+z*pp4))); |
164 |
|
|
s = one+z*(qq1+z*(qq2+z*(qq3+z*(qq4+z*qq5)))); |
165 |
|
|
y = r/s; |
166 |
|
|
if(hx < 0x3e800000) { /* x<1/4 */ |
167 |
|
|
return one-(x+x*y); |
168 |
|
|
} else { |
169 |
|
|
r = x*y; |
170 |
|
|
r += (x-half); |
171 |
|
|
return half - r ; |
172 |
|
|
} |
173 |
|
|
} |
174 |
|
|
if(ix < 0x3fa00000) { /* 0.84375 <= |x| < 1.25 */ |
175 |
|
|
s = fabsf(x)-one; |
176 |
|
|
P = pa0+s*(pa1+s*(pa2+s*(pa3+s*(pa4+s*(pa5+s*pa6))))); |
177 |
|
|
Q = one+s*(qa1+s*(qa2+s*(qa3+s*(qa4+s*(qa5+s*qa6))))); |
178 |
|
|
if(hx>=0) { |
179 |
|
|
z = one-erx; return z - P/Q; |
180 |
|
|
} else { |
181 |
|
|
z = erx+P/Q; return one+z; |
182 |
|
|
} |
183 |
|
|
} |
184 |
|
|
if (ix < 0x41e00000) { /* |x|<28 */ |
185 |
|
|
x = fabsf(x); |
186 |
|
|
s = one/(x*x); |
187 |
|
|
if(ix< 0x4036DB6D) { /* |x| < 1/.35 ~ 2.857143*/ |
188 |
|
|
R=ra0+s*(ra1+s*(ra2+s*(ra3+s*(ra4+s*( |
189 |
|
|
ra5+s*(ra6+s*ra7)))))); |
190 |
|
|
S=one+s*(sa1+s*(sa2+s*(sa3+s*(sa4+s*( |
191 |
|
|
sa5+s*(sa6+s*(sa7+s*sa8))))))); |
192 |
|
|
} else { /* |x| >= 1/.35 ~ 2.857143 */ |
193 |
|
|
if(hx<0&&ix>=0x40c00000) return two-tiny;/* x < -6 */ |
194 |
|
|
R=rb0+s*(rb1+s*(rb2+s*(rb3+s*(rb4+s*( |
195 |
|
|
rb5+s*rb6))))); |
196 |
|
|
S=one+s*(sb1+s*(sb2+s*(sb3+s*(sb4+s*( |
197 |
|
|
sb5+s*(sb6+s*sb7)))))); |
198 |
|
|
} |
199 |
|
|
GET_FLOAT_WORD(ix,x); |
200 |
|
|
SET_FLOAT_WORD(z,ix&0xfffff000); |
201 |
|
|
r = expf(-z*z-(float)0.5625) * expf((z-x)*(z+x)+R/S); |
202 |
|
|
if(hx>0) return r/x; else return two-r/x; |
203 |
|
|
} else { |
204 |
|
|
if(hx>0) return tiny*tiny; else return two-tiny; |
205 |
|
|
} |
206 |
|
|
} |