xref: /netbsd-src/external/gpl3/gcc.old/dist/libquadmath/math/cosq_kernel.c (revision 627f7eb200a4419d89b531d55fccd2ee3ffdcde0)
1*627f7eb2Smrg /* Quad-precision floating point cosine on <-pi/4,pi/4>.
2*627f7eb2Smrg    Copyright (C) 1999-2018 Free Software Foundation, Inc.
3*627f7eb2Smrg    This file is part of the GNU C Library.
4*627f7eb2Smrg    Contributed by Jakub Jelinek <jj@ultra.linux.cz>
5*627f7eb2Smrg 
6*627f7eb2Smrg    The GNU C Library is free software; you can redistribute it and/or
7*627f7eb2Smrg    modify it under the terms of the GNU Lesser General Public
8*627f7eb2Smrg    License as published by the Free Software Foundation; either
9*627f7eb2Smrg    version 2.1 of the License, or (at your option) any later version.
10*627f7eb2Smrg 
11*627f7eb2Smrg    The GNU C Library is distributed in the hope that it will be useful,
12*627f7eb2Smrg    but WITHOUT ANY WARRANTY; without even the implied warranty of
13*627f7eb2Smrg    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
14*627f7eb2Smrg    Lesser General Public License for more details.
15*627f7eb2Smrg 
16*627f7eb2Smrg    You should have received a copy of the GNU Lesser General Public
17*627f7eb2Smrg    License along with the GNU C Library; if not, see
18*627f7eb2Smrg    <http://www.gnu.org/licenses/>.  */
19*627f7eb2Smrg 
20*627f7eb2Smrg #include "quadmath-imp.h"
21*627f7eb2Smrg 
22*627f7eb2Smrg static const __float128 c[] = {
23*627f7eb2Smrg #define ONE c[0]
24*627f7eb2Smrg  1.00000000000000000000000000000000000E+00Q, /* 3fff0000000000000000000000000000 */
25*627f7eb2Smrg 
26*627f7eb2Smrg /* cos x ~ ONE + x^2 ( SCOS1 + SCOS2 * x^2 + ... + SCOS4 * x^6 + SCOS5 * x^8 )
27*627f7eb2Smrg    x in <0,1/256>  */
28*627f7eb2Smrg #define SCOS1 c[1]
29*627f7eb2Smrg #define SCOS2 c[2]
30*627f7eb2Smrg #define SCOS3 c[3]
31*627f7eb2Smrg #define SCOS4 c[4]
32*627f7eb2Smrg #define SCOS5 c[5]
33*627f7eb2Smrg -5.00000000000000000000000000000000000E-01Q, /* bffe0000000000000000000000000000 */
34*627f7eb2Smrg  4.16666666666666666666666666556146073E-02Q, /* 3ffa5555555555555555555555395023 */
35*627f7eb2Smrg -1.38888888888888888888309442601939728E-03Q, /* bff56c16c16c16c16c16a566e42c0375 */
36*627f7eb2Smrg  2.48015873015862382987049502531095061E-05Q, /* 3fefa01a01a019ee02dcf7da2d6d5444 */
37*627f7eb2Smrg -2.75573112601362126593516899592158083E-07Q, /* bfe927e4f5dce637cb0b54908754bde0 */
38*627f7eb2Smrg 
39*627f7eb2Smrg /* cos x ~ ONE + x^2 ( COS1 + COS2 * x^2 + ... + COS7 * x^12 + COS8 * x^14 )
40*627f7eb2Smrg    x in <0,0.1484375>  */
41*627f7eb2Smrg #define COS1 c[6]
42*627f7eb2Smrg #define COS2 c[7]
43*627f7eb2Smrg #define COS3 c[8]
44*627f7eb2Smrg #define COS4 c[9]
45*627f7eb2Smrg #define COS5 c[10]
46*627f7eb2Smrg #define COS6 c[11]
47*627f7eb2Smrg #define COS7 c[12]
48*627f7eb2Smrg #define COS8 c[13]
49*627f7eb2Smrg -4.99999999999999999999999999999999759E-01Q, /* bffdfffffffffffffffffffffffffffb */
50*627f7eb2Smrg  4.16666666666666666666666666651287795E-02Q, /* 3ffa5555555555555555555555516f30 */
51*627f7eb2Smrg -1.38888888888888888888888742314300284E-03Q, /* bff56c16c16c16c16c16c16a463dfd0d */
52*627f7eb2Smrg  2.48015873015873015867694002851118210E-05Q, /* 3fefa01a01a01a01a0195cebe6f3d3a5 */
53*627f7eb2Smrg -2.75573192239858811636614709689300351E-07Q, /* bfe927e4fb7789f5aa8142a22044b51f */
54*627f7eb2Smrg  2.08767569877762248667431926878073669E-09Q, /* 3fe21eed8eff881d1e9262d7adff4373 */
55*627f7eb2Smrg -1.14707451049343817400420280514614892E-11Q, /* bfda9397496922a9601ed3d4ca48944b */
56*627f7eb2Smrg  4.77810092804389587579843296923533297E-14Q, /* 3fd2ae5f8197cbcdcaf7c3fb4523414c */
57*627f7eb2Smrg 
58*627f7eb2Smrg /* sin x ~ ONE * x + x^3 ( SSIN1 + SSIN2 * x^2 + ... + SSIN4 * x^6 + SSIN5 * x^8 )
59*627f7eb2Smrg    x in <0,1/256>  */
60*627f7eb2Smrg #define SSIN1 c[14]
61*627f7eb2Smrg #define SSIN2 c[15]
62*627f7eb2Smrg #define SSIN3 c[16]
63*627f7eb2Smrg #define SSIN4 c[17]
64*627f7eb2Smrg #define SSIN5 c[18]
65*627f7eb2Smrg -1.66666666666666666666666666666666659E-01Q, /* bffc5555555555555555555555555555 */
66*627f7eb2Smrg  8.33333333333333333333333333146298442E-03Q, /* 3ff81111111111111111111110fe195d */
67*627f7eb2Smrg -1.98412698412698412697726277416810661E-04Q, /* bff2a01a01a01a01a019e7121e080d88 */
68*627f7eb2Smrg  2.75573192239848624174178393552189149E-06Q, /* 3fec71de3a556c640c6aaa51aa02ab41 */
69*627f7eb2Smrg -2.50521016467996193495359189395805639E-08Q, /* bfe5ae644ee90c47dc71839de75b2787 */
70*627f7eb2Smrg };
71*627f7eb2Smrg 
72*627f7eb2Smrg #define SINCOSL_COS_HI 0
73*627f7eb2Smrg #define SINCOSL_COS_LO 1
74*627f7eb2Smrg #define SINCOSL_SIN_HI 2
75*627f7eb2Smrg #define SINCOSL_SIN_LO 3
76*627f7eb2Smrg extern const __float128 __sincosq_table[];
77*627f7eb2Smrg 
78*627f7eb2Smrg __float128
__quadmath_kernel_cosq(__float128 x,__float128 y)79*627f7eb2Smrg __quadmath_kernel_cosq(__float128 x, __float128 y)
80*627f7eb2Smrg {
81*627f7eb2Smrg   __float128 h, l, z, sin_l, cos_l_m1;
82*627f7eb2Smrg   int64_t ix;
83*627f7eb2Smrg   uint32_t tix, hix, index;
84*627f7eb2Smrg   GET_FLT128_MSW64 (ix, x);
85*627f7eb2Smrg   tix = ((uint64_t)ix) >> 32;
86*627f7eb2Smrg   tix &= ~0x80000000;			/* tix = |x|'s high 32 bits */
87*627f7eb2Smrg   if (tix < 0x3ffc3000)			/* |x| < 0.1484375 */
88*627f7eb2Smrg     {
89*627f7eb2Smrg       /* Argument is small enough to approximate it by a Chebyshev
90*627f7eb2Smrg 	 polynomial of degree 16.  */
91*627f7eb2Smrg       if (tix < 0x3fc60000)		/* |x| < 2^-57 */
92*627f7eb2Smrg 	if (!((int)x)) return ONE;	/* generate inexact */
93*627f7eb2Smrg       z = x * x;
94*627f7eb2Smrg       return ONE + (z*(COS1+z*(COS2+z*(COS3+z*(COS4+
95*627f7eb2Smrg 		    z*(COS5+z*(COS6+z*(COS7+z*COS8))))))));
96*627f7eb2Smrg     }
97*627f7eb2Smrg   else
98*627f7eb2Smrg     {
99*627f7eb2Smrg       /* So that we don't have to use too large polynomial,  we find
100*627f7eb2Smrg 	 l and h such that x = l + h,  where fabsq(l) <= 1.0/256 with 83
101*627f7eb2Smrg 	 possible values for h.  We look up cosq(h) and sinq(h) in
102*627f7eb2Smrg 	 pre-computed tables,  compute cosq(l) and sinq(l) using a
103*627f7eb2Smrg 	 Chebyshev polynomial of degree 10(11) and compute
104*627f7eb2Smrg 	 cosq(h+l) = cosq(h)cosq(l) - sinq(h)sinq(l).  */
105*627f7eb2Smrg       index = 0x3ffe - (tix >> 16);
106*627f7eb2Smrg       hix = (tix + (0x200 << index)) & (0xfffffc00 << index);
107*627f7eb2Smrg       if (signbitq (x))
108*627f7eb2Smrg 	{
109*627f7eb2Smrg 	  x = -x;
110*627f7eb2Smrg 	  y = -y;
111*627f7eb2Smrg 	}
112*627f7eb2Smrg       switch (index)
113*627f7eb2Smrg 	{
114*627f7eb2Smrg 	case 0: index = ((45 << 10) + hix - 0x3ffe0000) >> 8; break;
115*627f7eb2Smrg 	case 1: index = ((13 << 11) + hix - 0x3ffd0000) >> 9; break;
116*627f7eb2Smrg 	default:
117*627f7eb2Smrg 	case 2: index = (hix - 0x3ffc3000) >> 10; break;
118*627f7eb2Smrg 	}
119*627f7eb2Smrg 
120*627f7eb2Smrg       SET_FLT128_WORDS64(h, ((uint64_t)hix) << 32, 0);
121*627f7eb2Smrg       l = y - (h - x);
122*627f7eb2Smrg       z = l * l;
123*627f7eb2Smrg       sin_l = l*(ONE+z*(SSIN1+z*(SSIN2+z*(SSIN3+z*(SSIN4+z*SSIN5)))));
124*627f7eb2Smrg       cos_l_m1 = z*(SCOS1+z*(SCOS2+z*(SCOS3+z*(SCOS4+z*SCOS5))));
125*627f7eb2Smrg       return __sincosq_table [index + SINCOSL_COS_HI]
126*627f7eb2Smrg 	     + (__sincosq_table [index + SINCOSL_COS_LO]
127*627f7eb2Smrg 		- (__sincosq_table [index + SINCOSL_SIN_HI] * sin_l
128*627f7eb2Smrg 		   - __sincosq_table [index + SINCOSL_COS_HI] * cos_l_m1));
129*627f7eb2Smrg     }
130*627f7eb2Smrg }
131