1 1.1 mrg /* Quad-precision floating point cosine on <-pi/4,pi/4>. 2 1.1 mrg Copyright (C) 1999-2018 Free Software Foundation, Inc. 3 1.1 mrg This file is part of the GNU C Library. 4 1.1 mrg Contributed by Jakub Jelinek <jj (at) ultra.linux.cz> 5 1.1 mrg 6 1.1 mrg The GNU C Library is free software; you can redistribute it and/or 7 1.1 mrg modify it under the terms of the GNU Lesser General Public 8 1.1 mrg License as published by the Free Software Foundation; either 9 1.1 mrg version 2.1 of the License, or (at your option) any later version. 10 1.1 mrg 11 1.1 mrg The GNU C Library is distributed in the hope that it will be useful, 12 1.1 mrg but WITHOUT ANY WARRANTY; without even the implied warranty of 13 1.1 mrg MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 14 1.1 mrg Lesser General Public License for more details. 15 1.1 mrg 16 1.1 mrg You should have received a copy of the GNU Lesser General Public 17 1.1 mrg License along with the GNU C Library; if not, see 18 1.1 mrg <http://www.gnu.org/licenses/>. */ 19 1.1 mrg 20 1.1 mrg #include "quadmath-imp.h" 21 1.1 mrg 22 1.1 mrg static const __float128 c[] = { 23 1.1 mrg #define ONE c[0] 24 1.1 mrg 1.00000000000000000000000000000000000E+00Q, /* 3fff0000000000000000000000000000 */ 25 1.1 mrg 26 1.1 mrg /* cos x ~ ONE + x^2 ( SCOS1 + SCOS2 * x^2 + ... + SCOS4 * x^6 + SCOS5 * x^8 ) 27 1.1 mrg x in <0,1/256> */ 28 1.1 mrg #define SCOS1 c[1] 29 1.1 mrg #define SCOS2 c[2] 30 1.1 mrg #define SCOS3 c[3] 31 1.1 mrg #define SCOS4 c[4] 32 1.1 mrg #define SCOS5 c[5] 33 1.1 mrg -5.00000000000000000000000000000000000E-01Q, /* bffe0000000000000000000000000000 */ 34 1.1 mrg 4.16666666666666666666666666556146073E-02Q, /* 3ffa5555555555555555555555395023 */ 35 1.1 mrg -1.38888888888888888888309442601939728E-03Q, /* bff56c16c16c16c16c16a566e42c0375 */ 36 1.1 mrg 2.48015873015862382987049502531095061E-05Q, /* 3fefa01a01a019ee02dcf7da2d6d5444 */ 37 1.1 mrg -2.75573112601362126593516899592158083E-07Q, /* bfe927e4f5dce637cb0b54908754bde0 */ 38 1.1 mrg 39 1.1 mrg /* cos x ~ ONE + x^2 ( COS1 + COS2 * x^2 + ... + COS7 * x^12 + COS8 * x^14 ) 40 1.1 mrg x in <0,0.1484375> */ 41 1.1 mrg #define COS1 c[6] 42 1.1 mrg #define COS2 c[7] 43 1.1 mrg #define COS3 c[8] 44 1.1 mrg #define COS4 c[9] 45 1.1 mrg #define COS5 c[10] 46 1.1 mrg #define COS6 c[11] 47 1.1 mrg #define COS7 c[12] 48 1.1 mrg #define COS8 c[13] 49 1.1 mrg -4.99999999999999999999999999999999759E-01Q, /* bffdfffffffffffffffffffffffffffb */ 50 1.1 mrg 4.16666666666666666666666666651287795E-02Q, /* 3ffa5555555555555555555555516f30 */ 51 1.1 mrg -1.38888888888888888888888742314300284E-03Q, /* bff56c16c16c16c16c16c16a463dfd0d */ 52 1.1 mrg 2.48015873015873015867694002851118210E-05Q, /* 3fefa01a01a01a01a0195cebe6f3d3a5 */ 53 1.1 mrg -2.75573192239858811636614709689300351E-07Q, /* bfe927e4fb7789f5aa8142a22044b51f */ 54 1.1 mrg 2.08767569877762248667431926878073669E-09Q, /* 3fe21eed8eff881d1e9262d7adff4373 */ 55 1.1 mrg -1.14707451049343817400420280514614892E-11Q, /* bfda9397496922a9601ed3d4ca48944b */ 56 1.1 mrg 4.77810092804389587579843296923533297E-14Q, /* 3fd2ae5f8197cbcdcaf7c3fb4523414c */ 57 1.1 mrg 58 1.1 mrg /* sin x ~ ONE * x + x^3 ( SSIN1 + SSIN2 * x^2 + ... + SSIN4 * x^6 + SSIN5 * x^8 ) 59 1.1 mrg x in <0,1/256> */ 60 1.1 mrg #define SSIN1 c[14] 61 1.1 mrg #define SSIN2 c[15] 62 1.1 mrg #define SSIN3 c[16] 63 1.1 mrg #define SSIN4 c[17] 64 1.1 mrg #define SSIN5 c[18] 65 1.1 mrg -1.66666666666666666666666666666666659E-01Q, /* bffc5555555555555555555555555555 */ 66 1.1 mrg 8.33333333333333333333333333146298442E-03Q, /* 3ff81111111111111111111110fe195d */ 67 1.1 mrg -1.98412698412698412697726277416810661E-04Q, /* bff2a01a01a01a01a019e7121e080d88 */ 68 1.1 mrg 2.75573192239848624174178393552189149E-06Q, /* 3fec71de3a556c640c6aaa51aa02ab41 */ 69 1.1 mrg -2.50521016467996193495359189395805639E-08Q, /* bfe5ae644ee90c47dc71839de75b2787 */ 70 1.1 mrg }; 71 1.1 mrg 72 1.1 mrg #define SINCOSL_COS_HI 0 73 1.1 mrg #define SINCOSL_COS_LO 1 74 1.1 mrg #define SINCOSL_SIN_HI 2 75 1.1 mrg #define SINCOSL_SIN_LO 3 76 1.1 mrg extern const __float128 __sincosq_table[]; 77 1.1 mrg 78 1.1 mrg __float128 79 1.1 mrg __quadmath_kernel_cosq(__float128 x, __float128 y) 80 1.1 mrg { 81 1.1 mrg __float128 h, l, z, sin_l, cos_l_m1; 82 1.1 mrg int64_t ix; 83 1.1 mrg uint32_t tix, hix, index; 84 1.1 mrg GET_FLT128_MSW64 (ix, x); 85 1.1 mrg tix = ((uint64_t)ix) >> 32; 86 1.1 mrg tix &= ~0x80000000; /* tix = |x|'s high 32 bits */ 87 1.1 mrg if (tix < 0x3ffc3000) /* |x| < 0.1484375 */ 88 1.1 mrg { 89 1.1 mrg /* Argument is small enough to approximate it by a Chebyshev 90 1.1 mrg polynomial of degree 16. */ 91 1.1 mrg if (tix < 0x3fc60000) /* |x| < 2^-57 */ 92 1.1 mrg if (!((int)x)) return ONE; /* generate inexact */ 93 1.1 mrg z = x * x; 94 1.1 mrg return ONE + (z*(COS1+z*(COS2+z*(COS3+z*(COS4+ 95 1.1 mrg z*(COS5+z*(COS6+z*(COS7+z*COS8)))))))); 96 1.1 mrg } 97 1.1 mrg else 98 1.1 mrg { 99 1.1 mrg /* So that we don't have to use too large polynomial, we find 100 1.1 mrg l and h such that x = l + h, where fabsq(l) <= 1.0/256 with 83 101 1.1 mrg possible values for h. We look up cosq(h) and sinq(h) in 102 1.1 mrg pre-computed tables, compute cosq(l) and sinq(l) using a 103 1.1 mrg Chebyshev polynomial of degree 10(11) and compute 104 1.1 mrg cosq(h+l) = cosq(h)cosq(l) - sinq(h)sinq(l). */ 105 1.1 mrg index = 0x3ffe - (tix >> 16); 106 1.1 mrg hix = (tix + (0x200 << index)) & (0xfffffc00 << index); 107 1.1 mrg if (signbitq (x)) 108 1.1 mrg { 109 1.1 mrg x = -x; 110 1.1 mrg y = -y; 111 1.1 mrg } 112 1.1 mrg switch (index) 113 1.1 mrg { 114 1.1 mrg case 0: index = ((45 << 10) + hix - 0x3ffe0000) >> 8; break; 115 1.1 mrg case 1: index = ((13 << 11) + hix - 0x3ffd0000) >> 9; break; 116 1.1 mrg default: 117 1.1 mrg case 2: index = (hix - 0x3ffc3000) >> 10; break; 118 1.1 mrg } 119 1.1 mrg 120 1.1 mrg SET_FLT128_WORDS64(h, ((uint64_t)hix) << 32, 0); 121 1.1 mrg l = y - (h - x); 122 1.1 mrg z = l * l; 123 1.1 mrg sin_l = l*(ONE+z*(SSIN1+z*(SSIN2+z*(SSIN3+z*(SSIN4+z*SSIN5))))); 124 1.1 mrg cos_l_m1 = z*(SCOS1+z*(SCOS2+z*(SCOS3+z*(SCOS4+z*SCOS5)))); 125 1.1 mrg return __sincosq_table [index + SINCOSL_COS_HI] 126 1.1 mrg + (__sincosq_table [index + SINCOSL_COS_LO] 127 1.1 mrg - (__sincosq_table [index + SINCOSL_SIN_HI] * sin_l 128 1.1 mrg - __sincosq_table [index + SINCOSL_COS_HI] * cos_l_m1)); 129 1.1 mrg } 130 1.1 mrg } 131