1 | /* sf_erf.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 "fdlibm.h" |
---|
17 | |
---|
18 | #ifdef __v810__ |
---|
19 | #define const |
---|
20 | #endif |
---|
21 | |
---|
22 | #ifdef __STDC__ |
---|
23 | static const float |
---|
24 | #else |
---|
25 | static float |
---|
26 | #endif |
---|
27 | tiny = 1e-30, |
---|
28 | half= 5.0000000000e-01, /* 0x3F000000 */ |
---|
29 | one = 1.0000000000e+00, /* 0x3F800000 */ |
---|
30 | two = 2.0000000000e+00, /* 0x40000000 */ |
---|
31 | /* c = (subfloat)0.84506291151 */ |
---|
32 | erx = 8.4506291151e-01, /* 0x3f58560b */ |
---|
33 | /* |
---|
34 | * Coefficients for approximation to erf on [0,0.84375] |
---|
35 | */ |
---|
36 | efx = 1.2837916613e-01, /* 0x3e0375d4 */ |
---|
37 | efx8= 1.0270333290e+00, /* 0x3f8375d4 */ |
---|
38 | pp0 = 1.2837916613e-01, /* 0x3e0375d4 */ |
---|
39 | pp1 = -3.2504209876e-01, /* 0xbea66beb */ |
---|
40 | pp2 = -2.8481749818e-02, /* 0xbce9528f */ |
---|
41 | pp3 = -5.7702702470e-03, /* 0xbbbd1489 */ |
---|
42 | pp4 = -2.3763017452e-05, /* 0xb7c756b1 */ |
---|
43 | qq1 = 3.9791721106e-01, /* 0x3ecbbbce */ |
---|
44 | qq2 = 6.5022252500e-02, /* 0x3d852a63 */ |
---|
45 | qq3 = 5.0813062117e-03, /* 0x3ba68116 */ |
---|
46 | qq4 = 1.3249473704e-04, /* 0x390aee49 */ |
---|
47 | qq5 = -3.9602282413e-06, /* 0xb684e21a */ |
---|
48 | /* |
---|
49 | * Coefficients for approximation to erf in [0.84375,1.25] |
---|
50 | */ |
---|
51 | pa0 = -2.3621185683e-03, /* 0xbb1acdc6 */ |
---|
52 | pa1 = 4.1485610604e-01, /* 0x3ed46805 */ |
---|
53 | pa2 = -3.7220788002e-01, /* 0xbebe9208 */ |
---|
54 | pa3 = 3.1834661961e-01, /* 0x3ea2fe54 */ |
---|
55 | pa4 = -1.1089469492e-01, /* 0xbde31cc2 */ |
---|
56 | pa5 = 3.5478305072e-02, /* 0x3d1151b3 */ |
---|
57 | pa6 = -2.1663755178e-03, /* 0xbb0df9c0 */ |
---|
58 | qa1 = 1.0642088205e-01, /* 0x3dd9f331 */ |
---|
59 | qa2 = 5.4039794207e-01, /* 0x3f0a5785 */ |
---|
60 | qa3 = 7.1828655899e-02, /* 0x3d931ae7 */ |
---|
61 | qa4 = 1.2617121637e-01, /* 0x3e013307 */ |
---|
62 | qa5 = 1.3637083583e-02, /* 0x3c5f6e13 */ |
---|
63 | qa6 = 1.1984500103e-02, /* 0x3c445aa3 */ |
---|
64 | /* |
---|
65 | * Coefficients for approximation to erfc in [1.25,1/0.35] |
---|
66 | */ |
---|
67 | ra0 = -9.8649440333e-03, /* 0xbc21a093 */ |
---|
68 | ra1 = -6.9385856390e-01, /* 0xbf31a0b7 */ |
---|
69 | ra2 = -1.0558626175e+01, /* 0xc128f022 */ |
---|
70 | ra3 = -6.2375331879e+01, /* 0xc2798057 */ |
---|
71 | ra4 = -1.6239666748e+02, /* 0xc322658c */ |
---|
72 | ra5 = -1.8460508728e+02, /* 0xc3389ae7 */ |
---|
73 | ra6 = -8.1287437439e+01, /* 0xc2a2932b */ |
---|
74 | ra7 = -9.8143291473e+00, /* 0xc11d077e */ |
---|
75 | sa1 = 1.9651271820e+01, /* 0x419d35ce */ |
---|
76 | sa2 = 1.3765776062e+02, /* 0x4309a863 */ |
---|
77 | sa3 = 4.3456588745e+02, /* 0x43d9486f */ |
---|
78 | sa4 = 6.4538726807e+02, /* 0x442158c9 */ |
---|
79 | sa5 = 4.2900814819e+02, /* 0x43d6810b */ |
---|
80 | sa6 = 1.0863500214e+02, /* 0x42d9451f */ |
---|
81 | sa7 = 6.5702495575e+00, /* 0x40d23f7c */ |
---|
82 | sa8 = -6.0424413532e-02, /* 0xbd777f97 */ |
---|
83 | /* |
---|
84 | * Coefficients for approximation to erfc in [1/.35,28] |
---|
85 | */ |
---|
86 | rb0 = -9.8649431020e-03, /* 0xbc21a092 */ |
---|
87 | rb1 = -7.9928326607e-01, /* 0xbf4c9dd4 */ |
---|
88 | rb2 = -1.7757955551e+01, /* 0xc18e104b */ |
---|
89 | rb3 = -1.6063638306e+02, /* 0xc320a2ea */ |
---|
90 | rb4 = -6.3756646729e+02, /* 0xc41f6441 */ |
---|
91 | rb5 = -1.0250950928e+03, /* 0xc480230b */ |
---|
92 | rb6 = -4.8351919556e+02, /* 0xc3f1c275 */ |
---|
93 | sb1 = 3.0338060379e+01, /* 0x41f2b459 */ |
---|
94 | sb2 = 3.2579251099e+02, /* 0x43a2e571 */ |
---|
95 | sb3 = 1.5367296143e+03, /* 0x44c01759 */ |
---|
96 | sb4 = 3.1998581543e+03, /* 0x4547fdbb */ |
---|
97 | sb5 = 2.5530502930e+03, /* 0x451f90ce */ |
---|
98 | sb6 = 4.7452853394e+02, /* 0x43ed43a7 */ |
---|
99 | sb7 = -2.2440952301e+01; /* 0xc1b38712 */ |
---|
100 | |
---|
101 | #ifdef __STDC__ |
---|
102 | float erff(float x) |
---|
103 | #else |
---|
104 | float erff(x) |
---|
105 | float x; |
---|
106 | #endif |
---|
107 | { |
---|
108 | __int32_t hx,ix,i; |
---|
109 | float R,S,P,Q,s,y,z,r; |
---|
110 | GET_FLOAT_WORD(hx,x); |
---|
111 | ix = hx&0x7fffffff; |
---|
112 | if(!FLT_UWORD_IS_FINITE(ix)) { /* erf(nan)=nan */ |
---|
113 | i = ((__uint32_t)hx>>31)<<1; |
---|
114 | return (float)(1-i)+one/x; /* erf(+-inf)=+-1 */ |
---|
115 | } |
---|
116 | |
---|
117 | if(ix < 0x3f580000) { /* |x|<0.84375 */ |
---|
118 | if(ix < 0x31800000) { /* |x|<2**-28 */ |
---|
119 | if (ix < 0x04000000) |
---|
120 | /*avoid underflow */ |
---|
121 | return (float)0.125*((float)8.0*x+efx8*x); |
---|
122 | return x + efx*x; |
---|
123 | } |
---|
124 | z = x*x; |
---|
125 | r = pp0+z*(pp1+z*(pp2+z*(pp3+z*pp4))); |
---|
126 | s = one+z*(qq1+z*(qq2+z*(qq3+z*(qq4+z*qq5)))); |
---|
127 | y = r/s; |
---|
128 | return x + x*y; |
---|
129 | } |
---|
130 | if(ix < 0x3fa00000) { /* 0.84375 <= |x| < 1.25 */ |
---|
131 | s = fabsf(x)-one; |
---|
132 | P = pa0+s*(pa1+s*(pa2+s*(pa3+s*(pa4+s*(pa5+s*pa6))))); |
---|
133 | Q = one+s*(qa1+s*(qa2+s*(qa3+s*(qa4+s*(qa5+s*qa6))))); |
---|
134 | if(hx>=0) return erx + P/Q; else return -erx - P/Q; |
---|
135 | } |
---|
136 | if (ix >= 0x40c00000) { /* inf>|x|>=6 */ |
---|
137 | if(hx>=0) return one-tiny; else return tiny-one; |
---|
138 | } |
---|
139 | x = fabsf(x); |
---|
140 | s = one/(x*x); |
---|
141 | if(ix< 0x4036DB6E) { /* |x| < 1/0.35 */ |
---|
142 | R=ra0+s*(ra1+s*(ra2+s*(ra3+s*(ra4+s*( |
---|
143 | ra5+s*(ra6+s*ra7)))))); |
---|
144 | S=one+s*(sa1+s*(sa2+s*(sa3+s*(sa4+s*( |
---|
145 | sa5+s*(sa6+s*(sa7+s*sa8))))))); |
---|
146 | } else { /* |x| >= 1/0.35 */ |
---|
147 | R=rb0+s*(rb1+s*(rb2+s*(rb3+s*(rb4+s*( |
---|
148 | rb5+s*rb6))))); |
---|
149 | S=one+s*(sb1+s*(sb2+s*(sb3+s*(sb4+s*( |
---|
150 | sb5+s*(sb6+s*sb7)))))); |
---|
151 | } |
---|
152 | GET_FLOAT_WORD(ix,x); |
---|
153 | SET_FLOAT_WORD(z,ix&0xfffff000); |
---|
154 | r = __ieee754_expf(-z*z-(float)0.5625)*__ieee754_expf((z-x)*(z+x)+R/S); |
---|
155 | if(hx>=0) return one-r/x; else return r/x-one; |
---|
156 | } |
---|
157 | |
---|
158 | #ifdef __STDC__ |
---|
159 | float erfcf(float x) |
---|
160 | #else |
---|
161 | float erfcf(x) |
---|
162 | float x; |
---|
163 | #endif |
---|
164 | { |
---|
165 | __int32_t hx,ix; |
---|
166 | float R,S,P,Q,s,y,z,r; |
---|
167 | GET_FLOAT_WORD(hx,x); |
---|
168 | ix = hx&0x7fffffff; |
---|
169 | if(!FLT_UWORD_IS_FINITE(ix)) { /* erfc(nan)=nan */ |
---|
170 | /* erfc(+-inf)=0,2 */ |
---|
171 | return (float)(((__uint32_t)hx>>31)<<1)+one/x; |
---|
172 | } |
---|
173 | |
---|
174 | if(ix < 0x3f580000) { /* |x|<0.84375 */ |
---|
175 | if(ix < 0x23800000) /* |x|<2**-56 */ |
---|
176 | return one-x; |
---|
177 | z = x*x; |
---|
178 | r = pp0+z*(pp1+z*(pp2+z*(pp3+z*pp4))); |
---|
179 | s = one+z*(qq1+z*(qq2+z*(qq3+z*(qq4+z*qq5)))); |
---|
180 | y = r/s; |
---|
181 | if(hx < 0x3e800000) { /* x<1/4 */ |
---|
182 | return one-(x+x*y); |
---|
183 | } else { |
---|
184 | r = x*y; |
---|
185 | r += (x-half); |
---|
186 | return half - r ; |
---|
187 | } |
---|
188 | } |
---|
189 | if(ix < 0x3fa00000) { /* 0.84375 <= |x| < 1.25 */ |
---|
190 | s = fabsf(x)-one; |
---|
191 | P = pa0+s*(pa1+s*(pa2+s*(pa3+s*(pa4+s*(pa5+s*pa6))))); |
---|
192 | Q = one+s*(qa1+s*(qa2+s*(qa3+s*(qa4+s*(qa5+s*qa6))))); |
---|
193 | if(hx>=0) { |
---|
194 | z = one-erx; return z - P/Q; |
---|
195 | } else { |
---|
196 | z = erx+P/Q; return one+z; |
---|
197 | } |
---|
198 | } |
---|
199 | if (ix < 0x41e00000) { /* |x|<28 */ |
---|
200 | x = fabsf(x); |
---|
201 | s = one/(x*x); |
---|
202 | if(ix< 0x4036DB6D) { /* |x| < 1/.35 ~ 2.857143*/ |
---|
203 | R=ra0+s*(ra1+s*(ra2+s*(ra3+s*(ra4+s*( |
---|
204 | ra5+s*(ra6+s*ra7)))))); |
---|
205 | S=one+s*(sa1+s*(sa2+s*(sa3+s*(sa4+s*( |
---|
206 | sa5+s*(sa6+s*(sa7+s*sa8))))))); |
---|
207 | } else { /* |x| >= 1/.35 ~ 2.857143 */ |
---|
208 | if(hx<0&&ix>=0x40c00000) return two-tiny;/* x < -6 */ |
---|
209 | R=rb0+s*(rb1+s*(rb2+s*(rb3+s*(rb4+s*( |
---|
210 | rb5+s*rb6))))); |
---|
211 | S=one+s*(sb1+s*(sb2+s*(sb3+s*(sb4+s*( |
---|
212 | sb5+s*(sb6+s*sb7)))))); |
---|
213 | } |
---|
214 | GET_FLOAT_WORD(ix,x); |
---|
215 | SET_FLOAT_WORD(z,ix&0xfffff000); |
---|
216 | r = __ieee754_expf(-z*z-(float)0.5625)* |
---|
217 | __ieee754_expf((z-x)*(z+x)+R/S); |
---|
218 | if(hx>0) return r/x; else return two-r/x; |
---|
219 | } else { |
---|
220 | if(hx>0) return tiny*tiny; else return two-tiny; |
---|
221 | } |
---|
222 | } |
---|
223 | |
---|
224 | #ifdef _DOUBLE_IS_32BITS |
---|
225 | |
---|
226 | #ifdef __STDC__ |
---|
227 | double erf(double x) |
---|
228 | #else |
---|
229 | double erf(x) |
---|
230 | double x; |
---|
231 | #endif |
---|
232 | { |
---|
233 | return (double) erff((float) x); |
---|
234 | } |
---|
235 | |
---|
236 | #ifdef __STDC__ |
---|
237 | double erfc(double x) |
---|
238 | #else |
---|
239 | double erfc(x) |
---|
240 | double x; |
---|
241 | #endif |
---|
242 | { |
---|
243 | return (double) erfcf((float) x); |
---|
244 | } |
---|
245 | |
---|
246 | #endif /* defined(_DOUBLE_IS_32BITS) */ |
---|