1 /* Quad-precision floating point sine and cosine on <-pi/4,pi/4>.
2    Copyright (C) 1999-2022 Free Software Foundation, Inc.
3    This file is part of the GNU C Library.
4 
5    The GNU C Library is free software; you can redistribute it and/or
6    modify it under the terms of the GNU Lesser General Public
7    License as published by the Free Software Foundation; either
8    version 2.1 of the License, or (at your option) any later version.
9 
10    The GNU C Library is distributed in the hope that it will be useful,
11    but WITHOUT ANY WARRANTY; without even the implied warranty of
12    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
13    Lesser General Public License for more details.
14 
15    You should have received a copy of the GNU Lesser General Public
16    License along with the GNU C Library; if not, see
17    <https://www.gnu.org/licenses/>.  */
18 
19 #include <float.h>
20 #include <math.h>
21 #include <math_private.h>
22 #include <math-underflow.h>
23 
24 static const long double c[] = {
25 #define ONE c[0]
26  1.00000000000000000000000000000000000E+00L, /* 3fff0000000000000000000000000000 */
27 
28 /* cos x ~ ONE + x^2 ( SCOS1 + SCOS2 * x^2 + ... + SCOS4 * x^6 + SCOS5 * x^8 )
29    x in <0,1/256>  */
30 #define SCOS1 c[1]
31 #define SCOS2 c[2]
32 #define SCOS3 c[3]
33 #define SCOS4 c[4]
34 #define SCOS5 c[5]
35 -5.00000000000000000000000000000000000E-01L, /* bffe0000000000000000000000000000 */
36  4.16666666666666666666666666556146073E-02L, /* 3ffa5555555555555555555555395023 */
37 -1.38888888888888888888309442601939728E-03L, /* bff56c16c16c16c16c16a566e42c0375 */
38  2.48015873015862382987049502531095061E-05L, /* 3fefa01a01a019ee02dcf7da2d6d5444 */
39 -2.75573112601362126593516899592158083E-07L, /* bfe927e4f5dce637cb0b54908754bde0 */
40 
41 /* cos x ~ ONE + x^2 ( COS1 + COS2 * x^2 + ... + COS7 * x^12 + COS8 * x^14 )
42    x in <0,0.1484375>  */
43 #define COS1 c[6]
44 #define COS2 c[7]
45 #define COS3 c[8]
46 #define COS4 c[9]
47 #define COS5 c[10]
48 #define COS6 c[11]
49 #define COS7 c[12]
50 #define COS8 c[13]
51 -4.99999999999999999999999999999999759E-01L, /* bffdfffffffffffffffffffffffffffb */
52  4.16666666666666666666666666651287795E-02L, /* 3ffa5555555555555555555555516f30 */
53 -1.38888888888888888888888742314300284E-03L, /* bff56c16c16c16c16c16c16a463dfd0d */
54  2.48015873015873015867694002851118210E-05L, /* 3fefa01a01a01a01a0195cebe6f3d3a5 */
55 -2.75573192239858811636614709689300351E-07L, /* bfe927e4fb7789f5aa8142a22044b51f */
56  2.08767569877762248667431926878073669E-09L, /* 3fe21eed8eff881d1e9262d7adff4373 */
57 -1.14707451049343817400420280514614892E-11L, /* bfda9397496922a9601ed3d4ca48944b */
58  4.77810092804389587579843296923533297E-14L, /* 3fd2ae5f8197cbcdcaf7c3fb4523414c */
59 
60 /* sin x ~ ONE * x + x^3 ( SSIN1 + SSIN2 * x^2 + ... + SSIN4 * x^6 + SSIN5 * x^8 )
61    x in <0,1/256>  */
62 #define SSIN1 c[14]
63 #define SSIN2 c[15]
64 #define SSIN3 c[16]
65 #define SSIN4 c[17]
66 #define SSIN5 c[18]
67 -1.66666666666666666666666666666666659E-01L, /* bffc5555555555555555555555555555 */
68  8.33333333333333333333333333146298442E-03L, /* 3ff81111111111111111111110fe195d */
69 -1.98412698412698412697726277416810661E-04L, /* bff2a01a01a01a01a019e7121e080d88 */
70  2.75573192239848624174178393552189149E-06L, /* 3fec71de3a556c640c6aaa51aa02ab41 */
71 -2.50521016467996193495359189395805639E-08L, /* bfe5ae644ee90c47dc71839de75b2787 */
72 
73 /* sin x ~ ONE * x + x^3 ( SIN1 + SIN2 * x^2 + ... + SIN7 * x^12 + SIN8 * x^14 )
74    x in <0,0.1484375>  */
75 #define SIN1 c[19]
76 #define SIN2 c[20]
77 #define SIN3 c[21]
78 #define SIN4 c[22]
79 #define SIN5 c[23]
80 #define SIN6 c[24]
81 #define SIN7 c[25]
82 #define SIN8 c[26]
83 -1.66666666666666666666666666666666538e-01L, /* bffc5555555555555555555555555550 */
84  8.33333333333333333333333333307532934e-03L, /* 3ff811111111111111111111110e7340 */
85 -1.98412698412698412698412534478712057e-04L, /* bff2a01a01a01a01a01a019e7a626296 */
86  2.75573192239858906520896496653095890e-06L, /* 3fec71de3a556c7338fa38527474b8f5 */
87 -2.50521083854417116999224301266655662e-08L, /* bfe5ae64567f544e16c7de65c2ea551f */
88  1.60590438367608957516841576404938118e-10L, /* 3fde6124613a811480538a9a41957115 */
89 -7.64716343504264506714019494041582610e-13L, /* bfd6ae7f3d5aef30c7bc660b060ef365 */
90  2.81068754939739570236322404393398135e-15L, /* 3fce9510115aabf87aceb2022a9a9180 */
91 };
92 
93 #define SINCOSL_COS_HI 0
94 #define SINCOSL_COS_LO 1
95 #define SINCOSL_SIN_HI 2
96 #define SINCOSL_SIN_LO 3
97 extern const long double __sincosl_table[];
98 
99 void
__kernel_sincosl(long double x,long double y,long double * sinx,long double * cosx,int iy)100 __kernel_sincosl(long double x, long double y, long double *sinx, long double *cosx, int iy)
101 {
102   long double h, l, z, sin_l, cos_l_m1;
103   int64_t ix;
104   uint32_t tix, hix, index;
105   double xhi, hhi;
106 
107   xhi = ldbl_high (x);
108   EXTRACT_WORDS64 (ix, xhi);
109   tix = ((uint64_t)ix) >> 32;
110   tix &= ~0x80000000;			/* tix = |x|'s high 32 bits */
111   if (tix < 0x3fc30000)			/* |x| < 0.1484375 */
112     {
113       /* Argument is small enough to approximate it by a Chebyshev
114 	 polynomial of degree 16(17).  */
115       if (tix < 0x3c600000)		/* |x| < 2^-57 */
116 	{
117 	  math_check_force_underflow (x);
118 	  if (!((int)x))			/* generate inexact */
119 	    {
120 	      *sinx = x;
121 	      *cosx = ONE;
122 	      return;
123 	    }
124 	}
125       z = x * x;
126       *sinx = x + (x * (z*(SIN1+z*(SIN2+z*(SIN3+z*(SIN4+
127 			z*(SIN5+z*(SIN6+z*(SIN7+z*SIN8)))))))));
128       *cosx = ONE + (z*(COS1+z*(COS2+z*(COS3+z*(COS4+
129 		     z*(COS5+z*(COS6+z*(COS7+z*COS8))))))));
130     }
131   else
132     {
133       /* So that we don't have to use too large polynomial,  we find
134 	 l and h such that x = l + h,  where fabsl(l) <= 1.0/256 with 83
135 	 possible values for h.  We look up cosl(h) and sinl(h) in
136 	 pre-computed tables,  compute cosl(l) and sinl(l) using a
137 	 Chebyshev polynomial of degree 10(11) and compute
138 	 sinl(h+l) = sinl(h)cosl(l) + cosl(h)sinl(l) and
139 	 cosl(h+l) = cosl(h)cosl(l) - sinl(h)sinl(l).  */
140       int six = tix;
141       tix = ((six - 0x3ff00000) >> 4) + 0x3fff0000;
142       index = 0x3ffe - (tix >> 16);
143       hix = (tix + (0x200 << index)) & (0xfffffc00 << index);
144       x = fabsl (x);
145       switch (index)
146 	{
147 	case 0: index = ((45 << 10) + hix - 0x3ffe0000) >> 8; break;
148 	case 1: index = ((13 << 11) + hix - 0x3ffd0000) >> 9; break;
149 	default:
150 	case 2: index = (hix - 0x3ffc3000) >> 10; break;
151 	}
152       hix = (hix << 4) & 0x3fffffff;
153 /*
154     The following should work for double but generates the wrong index.
155     For now the code above converts double to ieee extended to compute
156     the index back to double for the h value.
157 
158 
159       index = 0x3fe - (tix >> 20);
160       hix = (tix + (0x2000 << index)) & (0xffffc000 << index);
161       if (signbit (x))
162 	{
163 	  x = -x;
164 	  y = -y;
165 	}
166       switch (index)
167 	{
168 	case 0: index = ((45 << 14) + hix - 0x3fe00000) >> 12; break;
169 	case 1: index = ((13 << 15) + hix - 0x3fd00000) >> 13; break;
170 	default:
171 	case 2: index = (hix - 0x3fc30000) >> 14; break;
172 	}
173 */
174       INSERT_WORDS64 (hhi, ((uint64_t)hix) << 32);
175       h = hhi;
176       if (iy)
177 	l = y - (h - x);
178       else
179 	l = x - h;
180       z = l * l;
181       sin_l = l*(ONE+z*(SSIN1+z*(SSIN2+z*(SSIN3+z*(SSIN4+z*SSIN5)))));
182       cos_l_m1 = z*(SCOS1+z*(SCOS2+z*(SCOS3+z*(SCOS4+z*SCOS5))));
183       z = __sincosl_table [index + SINCOSL_SIN_HI]
184 	  + (__sincosl_table [index + SINCOSL_SIN_LO]
185 	     + (__sincosl_table [index + SINCOSL_SIN_HI] * cos_l_m1)
186 	     + (__sincosl_table [index + SINCOSL_COS_HI] * sin_l));
187       *sinx = (ix < 0) ? -z : z;
188       *cosx = __sincosl_table [index + SINCOSL_COS_HI]
189 	      + (__sincosl_table [index + SINCOSL_COS_LO]
190 		 - (__sincosl_table [index + SINCOSL_SIN_HI] * sin_l
191 		    - __sincosl_table [index + SINCOSL_COS_HI] * cos_l_m1));
192     }
193 }
194