assembly optimizations for fmod/remainder functions
[musl] / src / math / log1pf.c
1 /* origin: FreeBSD /usr/src/lib/msun/src/s_log1pf.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 #include "libm.h"
17
18 static const float
19 ln2_hi = 6.9313812256e-01, /* 0x3f317180 */
20 ln2_lo = 9.0580006145e-06, /* 0x3717f7d1 */
21 two25  = 3.355443200e+07,  /* 0x4c000000 */
22 Lp1 = 6.6666668653e-01, /* 3F2AAAAB */
23 Lp2 = 4.0000000596e-01, /* 3ECCCCCD */
24 Lp3 = 2.8571429849e-01, /* 3E924925 */
25 Lp4 = 2.2222198546e-01, /* 3E638E29 */
26 Lp5 = 1.8183572590e-01, /* 3E3A3325 */
27 Lp6 = 1.5313838422e-01, /* 3E1CD04F */
28 Lp7 = 1.4798198640e-01; /* 3E178897 */
29
30 static const float zero = 0.0;
31
32 float log1pf(float x)
33 {
34         float hfsq,f,c,s,z,R,u;
35         int32_t k,hx,hu,ax;
36
37         GET_FLOAT_WORD(hx, x);
38         ax = hx & 0x7fffffff;
39
40         k = 1;
41         if (hx < 0x3ed413d0) {  /* 1+x < sqrt(2)+  */
42                 if (ax >= 0x3f800000) {  /* x <= -1.0 */
43                         if (x == -1.0f)
44                                 return -two25/zero; /* log1p(-1)=+inf */
45                         return (x-x)/(x-x);         /* log1p(x<-1)=NaN */
46                 }
47                 if (ax < 0x38000000) {   /* |x| < 2**-15 */
48                         /* raise inexact */
49                         if (two25 + x > zero && ax < 0x33800000)  /* |x| < 2**-24 */
50                                 return x;
51                         return x - x*x*0.5f;
52                 }
53                 if (hx > 0 || hx <= (int32_t)0xbe95f619) { /* sqrt(2)/2- <= 1+x < sqrt(2)+ */
54                         k = 0;
55                         f = x;
56                         hu = 1;
57                 }
58         }
59         if (hx >= 0x7f800000)
60                 return x+x;
61         if (k != 0) {
62                 if (hx < 0x5a000000) {
63                         STRICT_ASSIGN(float, u, 1.0f + x);
64                         GET_FLOAT_WORD(hu, u);
65                         k = (hu>>23) - 127;
66                         /* correction term */
67                         c = k > 0 ? 1.0f-(u-x) : x-(u-1.0f);
68                         c /= u;
69                 } else {
70                         u = x;
71                         GET_FLOAT_WORD(hu,u);
72                         k = (hu>>23) - 127;
73                         c = 0;
74                 }
75                 hu &= 0x007fffff;
76                 /*
77                  * The approximation to sqrt(2) used in thresholds is not
78                  * critical.  However, the ones used above must give less
79                  * strict bounds than the one here so that the k==0 case is
80                  * never reached from here, since here we have committed to
81                  * using the correction term but don't use it if k==0.
82                  */
83                 if (hu < 0x3504f4) {  /* u < sqrt(2) */
84                         SET_FLOAT_WORD(u, hu|0x3f800000);  /* normalize u */
85                 } else {
86                         k += 1;
87                         SET_FLOAT_WORD(u, hu|0x3f000000);  /* normalize u/2 */
88                         hu = (0x00800000-hu)>>2;
89                 }
90                 f = u - 1.0f;
91         }
92         hfsq = 0.5f * f * f;
93         if (hu == 0) {  /* |f| < 2**-20 */
94                 if (f == zero) {
95                         if (k == 0)
96                                 return zero;
97                         c += k*ln2_lo;
98                         return k*ln2_hi+c;
99                 }
100                 R = hfsq*(1.0f - 0.66666666666666666f * f);
101                 if (k == 0)
102                         return f - R;
103                 return k*ln2_hi - ((R-(k*ln2_lo+c))-f);
104         }
105         s = f/(2.0f + f);
106         z = s*s;
107         R = z*(Lp1+z*(Lp2+z*(Lp3+z*(Lp4+z*(Lp5+z*(Lp6+z*Lp7))))));
108         if (k == 0)
109                 return f - (hfsq-s*(hfsq+R));
110         return k*ln2_hi - ((hfsq-(s*(hfsq+R)+(k*ln2_lo+c)))-f);
111 }