• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /* origin: FreeBSD /usr/src/lib/msun/ld80/e_rem_pio2.c */
2 /*
3  * ====================================================
4  * Copyright (C) 1993 by Sun Microsystems, Inc. All rights reserved.
5  * Copyright (c) 2008 Steven G. Kargl, David Schultz, Bruce D. Evans.
6  *
7  * Developed at SunSoft, a Sun Microsystems, Inc. business.
8  * Permission to use, copy, modify, and distribute this
9  * software is freely granted, provided that this notice
10  * is preserved.
11  * ====================================================
12  *
13  * Optimized by Bruce D. Evans.
14  */
15 #include "libm.h"
16 #if (LDBL_MANT_DIG == 64 || LDBL_MANT_DIG == 113) && LDBL_MAX_EXP == 16384
17 /* ld80 and ld128 version of __rem_pio2(x,y)
18  *
19  * return the remainder of x rem pi/2 in y[0]+y[1]
20  * use __rem_pio2_large() for large x
21  */
22 
23 static const long double toint = 1.5/LDBL_EPSILON;
24 
25 #if LDBL_MANT_DIG == 64
26 /* u ~< 0x1p25*pi/2 */
27 #define SMALL(u) (((u.i.se & 0x7fffU)<<16 | u.i.m>>48) < ((0x3fff + 25)<<16 | 0x921f>>1 | 0x8000))
28 #define QUOBITS(x) ((uint32_t)(int32_t)x & 0x7fffffff)
29 #define ROUND1 22
30 #define ROUND2 61
31 #define NX 3
32 #define NY 2
33 /*
34  * invpio2:  64 bits of 2/pi
35  * pio2_1:   first  39 bits of pi/2
36  * pio2_1t:  pi/2 - pio2_1
37  * pio2_2:   second 39 bits of pi/2
38  * pio2_2t:  pi/2 - (pio2_1+pio2_2)
39  * pio2_3:   third  39 bits of pi/2
40  * pio2_3t:  pi/2 - (pio2_1+pio2_2+pio2_3)
41  */
42 static const double
43 pio2_1 =  1.57079632679597125389e+00, /* 0x3FF921FB, 0x54444000 */
44 pio2_2 = -1.07463465549783099519e-12, /* -0x12e7b967674000.0p-92 */
45 pio2_3 =  6.36831716351370313614e-25; /*  0x18a2e037074000.0p-133 */
46 static const long double
47 pio4    =  0x1.921fb54442d1846ap-1L,
48 invpio2 =  6.36619772367581343076e-01L, /*  0xa2f9836e4e44152a.0p-64 */
49 pio2_1t = -1.07463465549719416346e-12L, /* -0x973dcb3b399d747f.0p-103 */
50 pio2_2t =  6.36831716351095013979e-25L, /*  0xc51701b839a25205.0p-144 */
51 pio2_3t = -2.75299651904407171810e-37L; /* -0xbb5bf6c7ddd660ce.0p-185 */
52 #elif LDBL_MANT_DIG == 113
53 /* u ~< 0x1p45*pi/2 */
54 #define SMALL(u) (((u.i.se & 0x7fffU)<<16 | u.i.top) < ((0x3fff + 45)<<16 | 0x921f))
55 #define QUOBITS(x) ((uint32_t)(int64_t)x & 0x7fffffff)
56 #define ROUND1 51
57 #define ROUND2 119
58 #define NX 5
59 #define NY 3
60 static const long double
61 pio4    =  0x1.921fb54442d18469898cc51701b8p-1L,
62 invpio2 =  6.3661977236758134307553505349005747e-01L,	/*  0x145f306dc9c882a53f84eafa3ea6a.0p-113 */
63 pio2_1  =  1.5707963267948966192292994253909555e+00L,	/*  0x1921fb54442d18469800000000000.0p-112 */
64 pio2_1t =  2.0222662487959507323996846200947577e-21L,	/*  0x13198a2e03707344a4093822299f3.0p-181 */
65 pio2_2  =  2.0222662487959507323994779168837751e-21L,	/*  0x13198a2e03707344a400000000000.0p-181 */
66 pio2_2t =  2.0670321098263988236496903051604844e-43L,	/*  0x127044533e63a0105df531d89cd91.0p-254 */
67 pio2_3  =  2.0670321098263988236499468110329591e-43L,	/*  0x127044533e63a0105e00000000000.0p-254 */
68 pio2_3t = -2.5650587247459238361625433492959285e-65L;	/* -0x159c4ec64ddaeb5f78671cbfb2210.0p-327 */
69 #endif
70 
__rem_pio2l(long double x,long double * y)71 int __rem_pio2l(long double x, long double *y)
72 {
73 	union ldshape u,uz;
74 	long double z,w,t,r,fn;
75 	double tx[NX],ty[NY];
76 	int ex,ey,n,i;
77 
78 	u.f = x;
79 	ex = u.i.se & 0x7fff;
80 	if (SMALL(u)) {
81 		/* rint(x/(pi/2)) */
82 		fn = x*invpio2 + toint - toint;
83 		n = QUOBITS(fn);
84 		r = x-fn*pio2_1;
85 		w = fn*pio2_1t;  /* 1st round good to 102/180 bits (ld80/ld128) */
86 		/* Matters with directed rounding. */
87 		if (predict_false(r - w < -pio4)) {
88 			n--;
89 			fn--;
90 			r = x - fn*pio2_1;
91 			w = fn*pio2_1t;
92 		} else if (predict_false(r - w > pio4)) {
93 			n++;
94 			fn++;
95 			r = x - fn*pio2_1;
96 			w = fn*pio2_1t;
97 		}
98 		y[0] = r-w;
99 		u.f = y[0];
100 		ey = u.i.se & 0x7fff;
101 		if (ex - ey > ROUND1) {  /* 2nd iteration needed, good to 141/248 (ld80/ld128) */
102 			t = r;
103 			w = fn*pio2_2;
104 			r = t-w;
105 			w = fn*pio2_2t-((t-r)-w);
106 			y[0] = r-w;
107 			u.f = y[0];
108 			ey = u.i.se & 0x7fff;
109 			if (ex - ey > ROUND2) {  /* 3rd iteration, good to 180/316 bits */
110 				t = r; /* will cover all possible cases (not verified for ld128) */
111 				w = fn*pio2_3;
112 				r = t-w;
113 				w = fn*pio2_3t-((t-r)-w);
114 				y[0] = r-w;
115 			}
116 		}
117 		y[1] = (r - y[0]) - w;
118 		return n;
119 	}
120 	/*
121 	 * all other (large) arguments
122 	 */
123 	if (ex == 0x7fff) {                /* x is inf or NaN */
124 		y[0] = y[1] = x - x;
125 		return 0;
126 	}
127 	/* set z = scalbn(|x|,-ilogb(x)+23) */
128 	uz.f = x;
129 	uz.i.se = 0x3fff + 23;
130 	z = uz.f;
131 	for (i=0; i < NX - 1; i++) {
132 		tx[i] = (double)(int32_t)z;
133 		z     = (z-tx[i])*0x1p24;
134 	}
135 	tx[i] = z;
136 	while (tx[i] == 0)
137 		i--;
138 	n = __rem_pio2_large(tx, ty, ex-0x3fff-23, i+1, NY);
139 	w = ty[1];
140 	if (NY == 3)
141 		w += ty[2];
142 	r = ty[0] + w;
143 	/* TODO: for ld128 this does not follow the recommendation of the
144 	comments of __rem_pio2_large which seem wrong if |ty[0]| > |ty[1]+ty[2]| */
145 	w -= r - ty[0];
146 	if (u.i.se >> 15) {
147 		y[0] = -r;
148 		y[1] = -w;
149 		return -n;
150 	}
151 	y[0] = r;
152 	y[1] = w;
153 	return n;
154 }
155 #endif
156