• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /* origin: FreeBSD /usr/src/lib/msun/src/e_lgammaf_r.c */
2 /*
3  * Conversion to float by Ian Lance Taylor, Cygnus Support, ian@cygnus.com.
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 use super::{floorf, k_cosf, k_sinf, logf};
17 
18 const PI: f32 = 3.1415927410e+00; /* 0x40490fdb */
19 const A0: f32 = 7.7215664089e-02; /* 0x3d9e233f */
20 const A1: f32 = 3.2246702909e-01; /* 0x3ea51a66 */
21 const A2: f32 = 6.7352302372e-02; /* 0x3d89f001 */
22 const A3: f32 = 2.0580807701e-02; /* 0x3ca89915 */
23 const A4: f32 = 7.3855509982e-03; /* 0x3bf2027e */
24 const A5: f32 = 2.8905137442e-03; /* 0x3b3d6ec6 */
25 const A6: f32 = 1.1927076848e-03; /* 0x3a9c54a1 */
26 const A7: f32 = 5.1006977446e-04; /* 0x3a05b634 */
27 const A8: f32 = 2.2086278477e-04; /* 0x39679767 */
28 const A9: f32 = 1.0801156895e-04; /* 0x38e28445 */
29 const A10: f32 = 2.5214456400e-05; /* 0x37d383a2 */
30 const A11: f32 = 4.4864096708e-05; /* 0x383c2c75 */
31 const TC: f32 = 1.4616321325e+00; /* 0x3fbb16c3 */
32 const TF: f32 = -1.2148628384e-01; /* 0xbdf8cdcd */
33 /* TT = -(tail of TF) */
34 const TT: f32 = 6.6971006518e-09; /* 0x31e61c52 */
35 const T0: f32 = 4.8383611441e-01; /* 0x3ef7b95e */
36 const T1: f32 = -1.4758771658e-01; /* 0xbe17213c */
37 const T2: f32 = 6.4624942839e-02; /* 0x3d845a15 */
38 const T3: f32 = -3.2788541168e-02; /* 0xbd064d47 */
39 const T4: f32 = 1.7970675603e-02; /* 0x3c93373d */
40 const T5: f32 = -1.0314224288e-02; /* 0xbc28fcfe */
41 const T6: f32 = 6.1005386524e-03; /* 0x3bc7e707 */
42 const T7: f32 = -3.6845202558e-03; /* 0xbb7177fe */
43 const T8: f32 = 2.2596477065e-03; /* 0x3b141699 */
44 const T9: f32 = -1.4034647029e-03; /* 0xbab7f476 */
45 const T10: f32 = 8.8108185446e-04; /* 0x3a66f867 */
46 const T11: f32 = -5.3859531181e-04; /* 0xba0d3085 */
47 const T12: f32 = 3.1563205994e-04; /* 0x39a57b6b */
48 const T13: f32 = -3.1275415677e-04; /* 0xb9a3f927 */
49 const T14: f32 = 3.3552918467e-04; /* 0x39afe9f7 */
50 const U0: f32 = -7.7215664089e-02; /* 0xbd9e233f */
51 const U1: f32 = 6.3282704353e-01; /* 0x3f2200f4 */
52 const U2: f32 = 1.4549225569e+00; /* 0x3fba3ae7 */
53 const U3: f32 = 9.7771751881e-01; /* 0x3f7a4bb2 */
54 const U4: f32 = 2.2896373272e-01; /* 0x3e6a7578 */
55 const U5: f32 = 1.3381091878e-02; /* 0x3c5b3c5e */
56 const V1: f32 = 2.4559779167e+00; /* 0x401d2ebe */
57 const V2: f32 = 2.1284897327e+00; /* 0x4008392d */
58 const V3: f32 = 7.6928514242e-01; /* 0x3f44efdf */
59 const V4: f32 = 1.0422264785e-01; /* 0x3dd572af */
60 const V5: f32 = 3.2170924824e-03; /* 0x3b52d5db */
61 const S0: f32 = -7.7215664089e-02; /* 0xbd9e233f */
62 const S1: f32 = 2.1498242021e-01; /* 0x3e5c245a */
63 const S2: f32 = 3.2577878237e-01; /* 0x3ea6cc7a */
64 const S3: f32 = 1.4635047317e-01; /* 0x3e15dce6 */
65 const S4: f32 = 2.6642270386e-02; /* 0x3cda40e4 */
66 const S5: f32 = 1.8402845599e-03; /* 0x3af135b4 */
67 const S6: f32 = 3.1947532989e-05; /* 0x3805ff67 */
68 const R1: f32 = 1.3920053244e+00; /* 0x3fb22d3b */
69 const R2: f32 = 7.2193557024e-01; /* 0x3f38d0c5 */
70 const R3: f32 = 1.7193385959e-01; /* 0x3e300f6e */
71 const R4: f32 = 1.8645919859e-02; /* 0x3c98bf54 */
72 const R5: f32 = 7.7794247773e-04; /* 0x3a4beed6 */
73 const R6: f32 = 7.3266842264e-06; /* 0x36f5d7bd */
74 const W0: f32 = 4.1893854737e-01; /* 0x3ed67f1d */
75 const W1: f32 = 8.3333335817e-02; /* 0x3daaaaab */
76 const W2: f32 = -2.7777778450e-03; /* 0xbb360b61 */
77 const W3: f32 = 7.9365057172e-04; /* 0x3a500cfd */
78 const W4: f32 = -5.9518753551e-04; /* 0xba1c065c */
79 const W5: f32 = 8.3633989561e-04; /* 0x3a5b3dd2 */
80 const W6: f32 = -1.6309292987e-03; /* 0xbad5c4e8 */
81 
82 /* sin(PI*x) assuming x > 2^-100, if sin(PI*x)==0 the sign is arbitrary */
sin_pi(mut x: f32) -> f3283 fn sin_pi(mut x: f32) -> f32 {
84     let mut y: f64;
85     let mut n: isize;
86 
87     /* spurious inexact if odd int */
88     x = 2.0 * (x * 0.5 - floorf(x * 0.5)); /* x mod 2.0 */
89 
90     n = (x * 4.0) as isize;
91     n = (n + 1) / 2;
92     y = (x as f64) - (n as f64) * 0.5;
93     y *= 3.14159265358979323846;
94     match n {
95         1 => k_cosf(y),
96         2 => k_sinf(-y),
97         3 => -k_cosf(y),
98         0 | _ => k_sinf(y),
99     }
100 }
101 
lgammaf_r(mut x: f32) -> (f32, i32)102 pub fn lgammaf_r(mut x: f32) -> (f32, i32) {
103     let u = x.to_bits();
104     let mut t: f32;
105     let y: f32;
106     let mut z: f32;
107     let nadj: f32;
108     let p: f32;
109     let p1: f32;
110     let p2: f32;
111     let p3: f32;
112     let q: f32;
113     let mut r: f32;
114     let w: f32;
115     let ix: u32;
116     let i: i32;
117     let sign: bool;
118     let mut signgam: i32;
119 
120     /* purge off +-inf, NaN, +-0, tiny and negative arguments */
121     signgam = 1;
122     sign = (u >> 31) != 0;
123     ix = u & 0x7fffffff;
124     if ix >= 0x7f800000 {
125         return (x * x, signgam);
126     }
127     if ix < 0x35000000 {
128         /* |x| < 2**-21, return -log(|x|) */
129         if sign {
130             signgam = -1;
131             x = -x;
132         }
133         return (-logf(x), signgam);
134     }
135     if sign {
136         x = -x;
137         t = sin_pi(x);
138         if t == 0.0 {
139             /* -integer */
140             return (1.0 / (x - x), signgam);
141         }
142         if t > 0.0 {
143             signgam = -1;
144         } else {
145             t = -t;
146         }
147         nadj = logf(PI / (t * x));
148     } else {
149         nadj = 0.0;
150     }
151 
152     /* purge off 1 and 2 */
153     if ix == 0x3f800000 || ix == 0x40000000 {
154         r = 0.0;
155     }
156     /* for x < 2.0 */
157     else if ix < 0x40000000 {
158         if ix <= 0x3f666666 {
159             /* lgamma(x) = lgamma(x+1)-log(x) */
160             r = -logf(x);
161             if ix >= 0x3f3b4a20 {
162                 y = 1.0 - x;
163                 i = 0;
164             } else if ix >= 0x3e6d3308 {
165                 y = x - (TC - 1.0);
166                 i = 1;
167             } else {
168                 y = x;
169                 i = 2;
170             }
171         } else {
172             r = 0.0;
173             if ix >= 0x3fdda618 {
174                 /* [1.7316,2] */
175                 y = 2.0 - x;
176                 i = 0;
177             } else if ix >= 0x3F9da620 {
178                 /* [1.23,1.73] */
179                 y = x - TC;
180                 i = 1;
181             } else {
182                 y = x - 1.0;
183                 i = 2;
184             }
185         }
186         match i {
187             0 => {
188                 z = y * y;
189                 p1 = A0 + z * (A2 + z * (A4 + z * (A6 + z * (A8 + z * A10))));
190                 p2 = z * (A1 + z * (A3 + z * (A5 + z * (A7 + z * (A9 + z * A11)))));
191                 p = y * p1 + p2;
192                 r += p - 0.5 * y;
193             }
194             1 => {
195                 z = y * y;
196                 w = z * y;
197                 p1 = T0 + w * (T3 + w * (T6 + w * (T9 + w * T12))); /* parallel comp */
198                 p2 = T1 + w * (T4 + w * (T7 + w * (T10 + w * T13)));
199                 p3 = T2 + w * (T5 + w * (T8 + w * (T11 + w * T14)));
200                 p = z * p1 - (TT - w * (p2 + y * p3));
201                 r += TF + p;
202             }
203             2 => {
204                 p1 = y * (U0 + y * (U1 + y * (U2 + y * (U3 + y * (U4 + y * U5)))));
205                 p2 = 1.0 + y * (V1 + y * (V2 + y * (V3 + y * (V4 + y * V5))));
206                 r += -0.5 * y + p1 / p2;
207             }
208             #[cfg(debug_assertions)]
209             _ => unreachable!(),
210             #[cfg(not(debug_assertions))]
211             _ => {}
212         }
213     } else if ix < 0x41000000 {
214         /* x < 8.0 */
215         i = x as i32;
216         y = x - (i as f32);
217         p = y * (S0 + y * (S1 + y * (S2 + y * (S3 + y * (S4 + y * (S5 + y * S6))))));
218         q = 1.0 + y * (R1 + y * (R2 + y * (R3 + y * (R4 + y * (R5 + y * R6)))));
219         r = 0.5 * y + p / q;
220         z = 1.0; /* lgamma(1+s) = log(s) + lgamma(s) */
221         // TODO: In C, this was implemented using switch jumps with fallthrough.
222         // Does this implementation have performance problems?
223         if i >= 7 {
224             z *= y + 6.0;
225         }
226         if i >= 6 {
227             z *= y + 5.0;
228         }
229         if i >= 5 {
230             z *= y + 4.0;
231         }
232         if i >= 4 {
233             z *= y + 3.0;
234         }
235         if i >= 3 {
236             z *= y + 2.0;
237             r += logf(z);
238         }
239     } else if ix < 0x5c800000 {
240         /* 8.0 <= x < 2**58 */
241         t = logf(x);
242         z = 1.0 / x;
243         y = z * z;
244         w = W0 + z * (W1 + y * (W2 + y * (W3 + y * (W4 + y * (W5 + y * W6)))));
245         r = (x - 0.5) * (t - 1.0) + w;
246     } else {
247         /* 2**58 <= x <= inf */
248         r = x * (logf(x) - 1.0);
249     }
250     if sign {
251         r = nadj - r;
252     }
253     return (r, signgam);
254 }
255