1 /* erf_lgamma.c -- float version of er_lgamma.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 
17 #define _ADD_UNDER_R_TO_FUNCS
18 
19 #include "fdlibm.h"
20 
21 static const float two23 = 8.3886080000e+06, /* 0x4b000000 */
22     half = 5.0000000000e-01, /* 0x3f000000 */
23     one = 1.0000000000e+00, /* 0x3f800000 */
24     pi = 3.1415927410e+00, /* 0x40490fdb */
25     a0 = 7.7215664089e-02, /* 0x3d9e233f */
26     a1 = 3.2246702909e-01, /* 0x3ea51a66 */
27     a2 = 6.7352302372e-02, /* 0x3d89f001 */
28     a3 = 2.0580807701e-02, /* 0x3ca89915 */
29     a4 = 7.3855509982e-03, /* 0x3bf2027e */
30     a5 = 2.8905137442e-03, /* 0x3b3d6ec6 */
31     a6 = 1.1927076848e-03, /* 0x3a9c54a1 */
32     a7 = 5.1006977446e-04, /* 0x3a05b634 */
33     a8 = 2.2086278477e-04, /* 0x39679767 */
34     a9 = 1.0801156895e-04, /* 0x38e28445 */
35     a10 = 2.5214456400e-05, /* 0x37d383a2 */
36     a11 = 4.4864096708e-05, /* 0x383c2c75 */
37     tc = 1.4616321325e+00, /* 0x3fbb16c3 */
38     tf = -1.2148628384e-01, /* 0xbdf8cdcd */
39     /* tt = -(tail of tf) */
40     tt = 6.6971006518e-09, /* 0x31e61c52 */
41     t0 = 4.8383611441e-01, /* 0x3ef7b95e */
42     t1 = -1.4758771658e-01, /* 0xbe17213c */
43     t2 = 6.4624942839e-02, /* 0x3d845a15 */
44     t3 = -3.2788541168e-02, /* 0xbd064d47 */
45     t4 = 1.7970675603e-02, /* 0x3c93373d */
46     t5 = -1.0314224288e-02, /* 0xbc28fcfe */
47     t6 = 6.1005386524e-03, /* 0x3bc7e707 */
48     t7 = -3.6845202558e-03, /* 0xbb7177fe */
49     t8 = 2.2596477065e-03, /* 0x3b141699 */
50     t9 = -1.4034647029e-03, /* 0xbab7f476 */
51     t10 = 8.8108185446e-04, /* 0x3a66f867 */
52     t11 = -5.3859531181e-04, /* 0xba0d3085 */
53     t12 = 3.1563205994e-04, /* 0x39a57b6b */
54     t13 = -3.1275415677e-04, /* 0xb9a3f927 */
55     t14 = 3.3552918467e-04, /* 0x39afe9f7 */
56     u0 = -7.7215664089e-02, /* 0xbd9e233f */
57     u1 = 6.3282704353e-01, /* 0x3f2200f4 */
58     u2 = 1.4549225569e+00, /* 0x3fba3ae7 */
59     u3 = 9.7771751881e-01, /* 0x3f7a4bb2 */
60     u4 = 2.2896373272e-01, /* 0x3e6a7578 */
61     u5 = 1.3381091878e-02, /* 0x3c5b3c5e */
62     v1 = 2.4559779167e+00, /* 0x401d2ebe */
63     v2 = 2.1284897327e+00, /* 0x4008392d */
64     v3 = 7.6928514242e-01, /* 0x3f44efdf */
65     v4 = 1.0422264785e-01, /* 0x3dd572af */
66     v5 = 3.2170924824e-03, /* 0x3b52d5db */
67     s0 = -7.7215664089e-02, /* 0xbd9e233f */
68     s1 = 2.1498242021e-01, /* 0x3e5c245a */
69     s2 = 3.2577878237e-01, /* 0x3ea6cc7a */
70     s3 = 1.4635047317e-01, /* 0x3e15dce6 */
71     s4 = 2.6642270386e-02, /* 0x3cda40e4 */
72     s5 = 1.8402845599e-03, /* 0x3af135b4 */
73     s6 = 3.1947532989e-05, /* 0x3805ff67 */
74     r1 = 1.3920053244e+00, /* 0x3fb22d3b */
75     r2 = 7.2193557024e-01, /* 0x3f38d0c5 */
76     r3 = 1.7193385959e-01, /* 0x3e300f6e */
77     r4 = 1.8645919859e-02, /* 0x3c98bf54 */
78     r5 = 7.7794247773e-04, /* 0x3a4beed6 */
79     r6 = 7.3266842264e-06, /* 0x36f5d7bd */
80     w0 = 4.1893854737e-01, /* 0x3ed67f1d */
81     w1 = 8.3333335817e-02, /* 0x3daaaaab */
82     w2 = -2.7777778450e-03, /* 0xbb360b61 */
83     w3 = 7.9365057172e-04, /* 0x3a500cfd */
84     w4 = -5.9518753551e-04, /* 0xba1c065c */
85     w5 = 8.3633989561e-04, /* 0x3a5b3dd2 */
86     w6 = -1.6309292987e-03; /* 0xbad5c4e8 */
87 
88 static const float zero = 0.0000000000e+00;
89 
90 static float
sin_pif(float x)91 sin_pif(float x)
92 {
93     float y, z;
94     __int32_t n, ix;
95 
96     GET_FLOAT_WORD(ix, x);
97     ix &= 0x7fffffff;
98 
99     if (ix < 0x3e800000)
100         return __kernel_sinf(pi * x, zero, 0);
101     y = -x; /* x is assume negative */
102 
103     /*
104      * argument reduction, make sure inexact flag not raised if input
105      * is an integer
106      */
107     z = floorf(y);
108     if (z != y) { /* inexact anyway */
109         y *= (float)0.5;
110         y = (float)2.0 * (y - floorf(y)); /* y = |x| mod 2.0 */
111         n = (__int32_t)(y * (float)4.0);
112     } else {
113         if (ix >= 0x4b800000) {
114             y = zero;
115             n = 0; /* y must be even */
116         } else {
117             if (ix < 0x4b000000)
118                 z = y + two23; /* exact */
119             GET_FLOAT_WORD(n, z);
120             n &= 1;
121             y = n;
122             n <<= 2;
123         }
124     }
125     switch (n) {
126     case 0:
127         y = __kernel_sinf(pi * y, zero, 0);
128         break;
129     case 1:
130     case 2:
131         y = __kernel_cosf(pi * ((float)0.5 - y), zero);
132         break;
133     case 3:
134     case 4:
135         y = __kernel_sinf(pi * (one - y), zero, 0);
136         break;
137     case 5:
138     case 6:
139         y = -__kernel_cosf(pi * (y - (float)1.5), zero);
140         break;
141     default:
142         y = __kernel_sinf(pi * (y - (float)2.0), zero, 0);
143         break;
144     }
145     return -y;
146 }
147 
148 float
__math_lgammaf_r(float x,int * signgamp,int * divzero)149 __math_lgammaf_r(float x, int *signgamp, int *divzero)
150 {
151     float t, y, z, nadj = 0.0, p, p1, p2, p3, q, r, w;
152     __int32_t i, hx, ix;
153 
154     GET_FLOAT_WORD(hx, x);
155 
156     /* purge off +-inf, NaN, +-0, and negative arguments */
157     *signgamp = 1;
158     ix = hx & 0x7fffffff;
159     if (ix >= 0x7f800000)
160         return fabsf(x+x);
161     if (ix == 0) {
162         if (hx < 0)
163             *signgamp = -1;
164         *divzero = 1;
165         return __math_divzerof(0);
166     }
167     if (ix < 0x1c800000) { /* |x|<2**-70, return -log(|x|) */
168         if (hx < 0) {
169             *signgamp = -1;
170             return -logf(-x);
171         } else
172             return -logf(x);
173     }
174     if (hx < 0) {
175         if (ix >= 0x4b000000) { /* |x|>=2**23, must be -integer */
176             *divzero = 1;
177             return __math_divzerof(0);
178         }
179         t = sin_pif(x);
180         if (t == zero) {
181             *divzero = 1;
182             return __math_divzerof(0);
183         }
184         nadj = logf(pi / fabsf(t * x));
185         if (t < zero)
186             *signgamp = -1;
187         x = -x;
188     }
189 
190     /* purge off 1 and 2 */
191     if (ix == 0x3f800000 || ix == 0x40000000)
192         r = 0;
193     /* for x < 2.0 */
194     else if (ix < 0x40000000) {
195         if (ix <= 0x3f666666) { /* lgamma(x) = lgamma(x+1)-log(x) */
196             r = -logf(x);
197             if (ix >= 0x3f3b4a20) {
198                 y = one - x;
199                 i = 0;
200             } else if (ix >= 0x3e6d3308) {
201                 y = x - (tc - one);
202                 i = 1;
203             } else {
204                 y = x;
205                 i = 2;
206             }
207         } else {
208             r = zero;
209             if (ix >= 0x3fdda618) {
210                 y = (float)2.0 - x;
211                 i = 0;
212             } /* [1.7316,2] */
213             else if (ix >= 0x3F9da620) {
214                 y = x - tc;
215                 i = 1;
216             } /* [1.23,1.73] */
217             else {
218                 y = x - one;
219                 i = 2;
220             }
221         }
222         switch (i) {
223         case 0:
224             z = y * y;
225             p1 = a0 + z * (a2 + z * (a4 + z * (a6 + z * (a8 + z * a10))));
226             p2 = z * (a1 + z * (a3 + z * (a5 + z * (a7 + z * (a9 + z * a11)))));
227             p = y * p1 + p2;
228             r += (p - (float)0.5 * y);
229             break;
230         case 1:
231             z = y * y;
232             w = z * y;
233             p1 = t0 +
234                  w * (t3 + w * (t6 + w * (t9 + w * t12))); /* parallel comp */
235             p2 = t1 + w * (t4 + w * (t7 + w * (t10 + w * t13)));
236             p3 = t2 + w * (t5 + w * (t8 + w * (t11 + w * t14)));
237             p = z * p1 - (tt - w * (p2 + y * p3));
238             r += (tf + p);
239             break;
240         case 2:
241             p1 = y * (u0 + y * (u1 + y * (u2 + y * (u3 + y * (u4 + y * u5)))));
242             p2 = one + y * (v1 + y * (v2 + y * (v3 + y * (v4 + y * v5))));
243             r += (-(float)0.5 * y + p1 / p2);
244         }
245     } else if (ix < 0x41000000) { /* x < 8.0 */
246         i = (__int32_t)x;
247         y = x - (float)i;
248         p = y * (s0 +
249                  y * (s1 + y * (s2 + y * (s3 + y * (s4 + y * (s5 + y * s6))))));
250         q = one + y * (r1 + y * (r2 + y * (r3 + y * (r4 + y * (r5 + y * r6)))));
251         r = half * y + p / q;
252         z = one; /* lgamma(1+s) = log(s) + lgamma(s) */
253         switch (i) {
254         case 7:
255             z *= (y + (float)6.0); /* FALLTHRU */
256         case 6:
257             z *= (y + (float)5.0); /* FALLTHRU */
258         case 5:
259             z *= (y + (float)4.0); /* FALLTHRU */
260         case 4:
261             z *= (y + (float)3.0); /* FALLTHRU */
262         case 3:
263             z *= (y + (float)2.0); /* FALLTHRU */
264             r += logf(z);
265             break;
266         }
267         /* 8.0 <= x < 2**58 */
268     } else if (ix < 0x5c800000) {
269         t = logf(x);
270         z = one / x;
271         y = z * z;
272         w = w0 + z * (w1 + y * (w2 + y * (w3 + y * (w4 + y * (w5 + y * w6)))));
273         r = (x - half) * (t - one) + w;
274     } else
275         /* 2**58 <= x <= inf */
276         r = x * (logf(x) - one);
277     if (hx < 0)
278         r = nadj - r;
279     return check_oflowf(r);
280 }
281 
282 float
lgammaf_r(float x,int * signgamp)283 lgammaf_r(float x, int *signgamp)
284 {
285     int divzero = 0;
286     return __math_lgammaf_r(x, signgamp, &divzero);
287 }
288 
289 _MATH_ALIAS_f_fI(lgamma)
290