| 1 |
/* |
| 2 |
* // Copyright (c) Radzivon Bartoshyk 7/2025. All rights reserved. |
| 3 |
* // |
| 4 |
* // Redistribution and use in source and binary forms, with or without modification, |
| 5 |
* // are permitted provided that the following conditions are met: |
| 6 |
* // |
| 7 |
* // 1. Redistributions of source code must retain the above copyright notice, this |
| 8 |
* // list of conditions and the following disclaimer. |
| 9 |
* // |
| 10 |
* // 2. Redistributions in binary form must reproduce the above copyright notice, |
| 11 |
* // this list of conditions and the following disclaimer in the documentation |
| 12 |
* // and/or other materials provided with the distribution. |
| 13 |
* // |
| 14 |
* // 3. Neither the name of the copyright holder nor the names of its |
| 15 |
* // contributors may be used to endorse or promote products derived from |
| 16 |
* // this software without specific prior written permission. |
| 17 |
* // |
| 18 |
* // THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" |
| 19 |
* // AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE |
| 20 |
* // IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE |
| 21 |
* // DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE |
| 22 |
* // FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL |
| 23 |
* // DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR |
| 24 |
* // SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER |
| 25 |
* // CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, |
| 26 |
* // OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE |
| 27 |
* // OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. |
| 28 |
*/ |
| 29 |
use crate::double_double::DoubleDouble; |
| 30 |
use crate::dyadic_float::DyadicFloat128; |
| 31 |
use crate::rounding::CpuRound; |
| 32 |
use crate::sin::{SinCos, get_sin_k_rational, sincos_eval}; |
| 33 |
use crate::sin_table::SIN_K_PI_OVER_128; |
| 34 |
use crate::sincos_dyadic::{range_reduction_small_f128_f128, sincos_eval_dyadic}; |
| 35 |
|
| 36 |
#[inline] |
| 37 |
fn sin_eval_dd(z: DoubleDouble) -> DoubleDouble { |
| 38 |
const SIN_C: [(u64, u64); 5] = [ |
| 39 |
(0x0000000000000000, 0x3ff0000000000000), |
| 40 |
(0xbc65555555555555, 0xbfc5555555555555), |
| 41 |
(0x3c01111111111111, 0x3f81111111111111), |
| 42 |
(0xbb6a01a01a01a01a, 0xbf2a01a01a01a01a), |
| 43 |
(0xbb6c154f8ddc6c00, 0x3ec71de3a556c734), |
| 44 |
]; |
| 45 |
let x2 = DoubleDouble::quick_mult(z, z); |
| 46 |
let mut p = DoubleDouble::mul_add( |
| 47 |
x2, |
| 48 |
DoubleDouble::from_bit_pair(SIN_C[4]), |
| 49 |
DoubleDouble::from_bit_pair(SIN_C[3]), |
| 50 |
); |
| 51 |
|
| 52 |
p = DoubleDouble::mul_add(x2, p, DoubleDouble::from_bit_pair(SIN_C[2])); |
| 53 |
p = DoubleDouble::mul_add(x2, p, DoubleDouble::from_bit_pair(SIN_C[1])); |
| 54 |
p = DoubleDouble::mul_add(x2, p, DoubleDouble::from_bit_pair(SIN_C[0])); |
| 55 |
DoubleDouble::quick_mult(p, z) |
| 56 |
} |
| 57 |
|
| 58 |
pub(crate) fn sin_dd_small(z: DoubleDouble) -> DoubleDouble { |
| 59 |
let x_e = (z.hi.to_bits() >> 52) & 0x7ff; |
| 60 |
const E_BIAS: u64 = (1u64 << (11 - 1u64)) - 1u64; |
| 61 |
|
| 62 |
if x_e < E_BIAS - 8 { |
| 63 |
return sin_eval_dd(z); |
| 64 |
} |
| 65 |
|
| 66 |
let (u_f128, k) = range_reduction_small_dd(z); |
| 67 |
|
| 68 |
let sin_cos = sincos_eval_dd(u_f128); |
| 69 |
|
| 70 |
// Fast look up version, but needs 256-entry table. |
| 71 |
// cos(k * pi/128) = sin(k * pi/128 + pi/2) = sin((k + 64) * pi/128). |
| 72 |
let sk = SIN_K_PI_OVER_128[(k & 255) as usize]; |
| 73 |
let ck = SIN_K_PI_OVER_128[((k.wrapping_add(64)) & 255) as usize]; |
| 74 |
|
| 75 |
let sin_k = DoubleDouble::from_bit_pair(sk); |
| 76 |
let cos_k = DoubleDouble::from_bit_pair(ck); |
| 77 |
|
| 78 |
let sin_k_cos_y = DoubleDouble::quick_mult(sin_cos.v_cos, sin_k); |
| 79 |
let cos_k_sin_y = DoubleDouble::quick_mult(sin_cos.v_sin, cos_k); |
| 80 |
|
| 81 |
// sin_k_cos_y is always >> cos_k_sin_y |
| 82 |
let mut rr = DoubleDouble::from_exact_add(sin_k_cos_y.hi, cos_k_sin_y.hi); |
| 83 |
rr.lo += sin_k_cos_y.lo + cos_k_sin_y.lo; |
| 84 |
rr |
| 85 |
} |
| 86 |
|
| 87 |
pub(crate) fn sin_dd_small_fast(z: DoubleDouble) -> DoubleDouble { |
| 88 |
let x_e = (z.hi.to_bits() >> 52) & 0x7ff; |
| 89 |
const E_BIAS: u64 = (1u64 << (11 - 1u64)) - 1u64; |
| 90 |
|
| 91 |
if x_e < E_BIAS - 8 { |
| 92 |
return sin_eval_dd(z); |
| 93 |
} |
| 94 |
|
| 95 |
let (u_f128, k) = range_reduction_small_dd(z); |
| 96 |
|
| 97 |
let sin_cos = sincos_eval(u_f128); |
| 98 |
|
| 99 |
// Fast look up version, but needs 256-entry table. |
| 100 |
// cos(k * pi/128) = sin(k * pi/128 + pi/2) = sin((k + 64) * pi/128). |
| 101 |
let sk = SIN_K_PI_OVER_128[(k & 255) as usize]; |
| 102 |
let ck = SIN_K_PI_OVER_128[((k.wrapping_add(64)) & 255) as usize]; |
| 103 |
|
| 104 |
let sin_k = DoubleDouble::from_bit_pair(sk); |
| 105 |
let cos_k = DoubleDouble::from_bit_pair(ck); |
| 106 |
|
| 107 |
let sin_k_cos_y = DoubleDouble::quick_mult(sin_cos.v_cos, sin_k); |
| 108 |
let cos_k_sin_y = DoubleDouble::quick_mult(sin_cos.v_sin, cos_k); |
| 109 |
|
| 110 |
// sin_k_cos_y is always >> cos_k_sin_y |
| 111 |
let mut rr = DoubleDouble::from_exact_add(sin_k_cos_y.hi, cos_k_sin_y.hi); |
| 112 |
rr.lo += sin_k_cos_y.lo + cos_k_sin_y.lo; |
| 113 |
rr |
| 114 |
} |
| 115 |
|
| 116 |
#[inline] |
| 117 |
fn cos_eval_dd(z: DoubleDouble) -> DoubleDouble { |
| 118 |
let x2 = DoubleDouble::quick_mult(z, z); |
| 119 |
const COS_C: [(u64, u64); 5] = [ |
| 120 |
(0x0000000000000000, 0x3ff0000000000000), |
| 121 |
(0x0000000000000000, 0xbfe0000000000000), |
| 122 |
(0x3c45555555555555, 0x3fa5555555555555), |
| 123 |
(0x3bef49f49f49f49f, 0xbf56c16c16c16c17), |
| 124 |
(0x3b3a01a01a01a01a, 0x3efa01a01a01a01a), |
| 125 |
]; |
| 126 |
|
| 127 |
let mut p = DoubleDouble::mul_add( |
| 128 |
x2, |
| 129 |
DoubleDouble::from_bit_pair(COS_C[4]), |
| 130 |
DoubleDouble::from_bit_pair(COS_C[3]), |
| 131 |
); |
| 132 |
|
| 133 |
p = DoubleDouble::mul_add(x2, p, DoubleDouble::from_bit_pair(COS_C[2])); |
| 134 |
p = DoubleDouble::mul_add(x2, p, DoubleDouble::from_bit_pair(COS_C[1])); |
| 135 |
p = DoubleDouble::mul_add(x2, p, DoubleDouble::from_bit_pair(COS_C[0])); |
| 136 |
|
| 137 |
p |
| 138 |
} |
| 139 |
|
| 140 |
pub(crate) fn cos_dd_small(z: DoubleDouble) -> DoubleDouble { |
| 141 |
let x_e = (z.hi.to_bits() >> 52) & 0x7ff; |
| 142 |
const E_BIAS: u64 = (1u64 << (11 - 1u64)) - 1u64; |
| 143 |
|
| 144 |
if x_e < E_BIAS - 8 { |
| 145 |
return cos_eval_dd(z); |
| 146 |
} |
| 147 |
|
| 148 |
let (u_f128, k) = range_reduction_small_dd(z); |
| 149 |
|
| 150 |
let sin_cos = sincos_eval_dd(u_f128); |
| 151 |
|
| 152 |
// cos(k * pi/128) = sin(k * pi/128 + pi/2) = sin((k + 64) * pi/128). |
| 153 |
let sk = SIN_K_PI_OVER_128[(k.wrapping_add(128) & 255) as usize]; |
| 154 |
let ck = SIN_K_PI_OVER_128[((k.wrapping_add(64)) & 255) as usize]; |
| 155 |
let msin_k = DoubleDouble::from_bit_pair(sk); |
| 156 |
let cos_k = DoubleDouble::from_bit_pair(ck); |
| 157 |
|
| 158 |
let cos_k_cos_y = DoubleDouble::quick_mult(sin_cos.v_cos, cos_k); |
| 159 |
let cos_k_msin_y = DoubleDouble::quick_mult(sin_cos.v_sin, msin_k); |
| 160 |
|
| 161 |
// cos_k_cos_y is always >> cos_k_msin_y |
| 162 |
let mut rr = DoubleDouble::from_exact_add(cos_k_cos_y.hi, cos_k_msin_y.hi); |
| 163 |
rr.lo += cos_k_cos_y.lo + cos_k_msin_y.lo; |
| 164 |
|
| 165 |
rr |
| 166 |
} |
| 167 |
|
| 168 |
pub(crate) fn cos_dd_small_fast(z: DoubleDouble) -> DoubleDouble { |
| 169 |
let x_e = (z.hi.to_bits() >> 52) & 0x7ff; |
| 170 |
const E_BIAS: u64 = (1u64 << (11 - 1u64)) - 1u64; |
| 171 |
|
| 172 |
if x_e < E_BIAS - 8 { |
| 173 |
return cos_eval_dd(z); |
| 174 |
} |
| 175 |
|
| 176 |
let (u_f128, k) = range_reduction_small_dd(z); |
| 177 |
|
| 178 |
let sin_cos = sincos_eval(u_f128); |
| 179 |
|
| 180 |
// cos(k * pi/128) = sin(k * pi/128 + pi/2) = sin((k + 64) * pi/128). |
| 181 |
let sk = SIN_K_PI_OVER_128[(k.wrapping_add(128) & 255) as usize]; |
| 182 |
let ck = SIN_K_PI_OVER_128[((k.wrapping_add(64)) & 255) as usize]; |
| 183 |
let msin_k = DoubleDouble::from_bit_pair(sk); |
| 184 |
let cos_k = DoubleDouble::from_bit_pair(ck); |
| 185 |
|
| 186 |
let cos_k_cos_y = DoubleDouble::quick_mult(sin_cos.v_cos, cos_k); |
| 187 |
let cos_k_msin_y = DoubleDouble::quick_mult(sin_cos.v_sin, msin_k); |
| 188 |
|
| 189 |
// cos_k_cos_y is always >> cos_k_msin_y |
| 190 |
let mut rr = DoubleDouble::from_exact_add(cos_k_cos_y.hi, cos_k_msin_y.hi); |
| 191 |
rr.lo += cos_k_cos_y.lo + cos_k_msin_y.lo; |
| 192 |
|
| 193 |
rr |
| 194 |
} |
| 195 |
|
| 196 |
pub(crate) fn sin_f128_small(z: DyadicFloat128) -> DyadicFloat128 { |
| 197 |
let (u_f128, k) = range_reduction_small_f128_f128(z); |
| 198 |
|
| 199 |
let sin_cos = sincos_eval_dyadic(&u_f128); |
| 200 |
// cos(k * pi/128) = sin(k * pi/128 + pi/2) = sin((k + 64) * pi/128). |
| 201 |
let sin_k_f128 = get_sin_k_rational(k as u64); |
| 202 |
let cos_k_f128 = get_sin_k_rational((k as u64).wrapping_add(64)); |
| 203 |
|
| 204 |
// sin(x) = sin(k * pi/128 + u) |
| 205 |
// = sin(u) * cos(k*pi/128) + cos(u) * sin(k*pi/128) |
| 206 |
(sin_k_f128 * sin_cos.v_cos) + (cos_k_f128 * sin_cos.v_sin) |
| 207 |
} |
| 208 |
|
| 209 |
pub(crate) fn cos_f128_small(z: DyadicFloat128) -> DyadicFloat128 { |
| 210 |
let (u_f128, k) = range_reduction_small_f128_f128(z); |
| 211 |
|
| 212 |
let sin_cos = sincos_eval_dyadic(&u_f128); |
| 213 |
// -sin(k * pi/128) = sin((k + 128) * pi/128) |
| 214 |
// cos(k * pi/128) = sin(k * pi/128 + pi/2) = sin((k + 64) * pi/128). |
| 215 |
let msin_k_f128 = get_sin_k_rational((k as u64).wrapping_add(128)); |
| 216 |
let cos_k_f128 = get_sin_k_rational((k as u64).wrapping_add(64)); |
| 217 |
|
| 218 |
// cos(x) = cos((k * pi/128 + u) |
| 219 |
// = cos(u) * cos(k*pi/128) - sin(u) * sin(k*pi/128) |
| 220 |
(cos_k_f128 * sin_cos.v_cos) + (msin_k_f128 * sin_cos.v_sin) |
| 221 |
} |
| 222 |
|
| 223 |
#[inline] |
| 224 |
pub(crate) fn sincos_eval_dd(u: DoubleDouble) -> SinCos { |
| 225 |
const SIN_C: [(u64, u64); 5] = [ |
| 226 |
(0x0000000000000000, 0x3ff0000000000000), |
| 227 |
(0xbc65555555555555, 0xbfc5555555555555), |
| 228 |
(0x3c01111111111111, 0x3f81111111111111), |
| 229 |
(0xbb6a01a01a01a01a, 0xbf2a01a01a01a01a), |
| 230 |
(0xbb6c154f8ddc6c00, 0x3ec71de3a556c734), |
| 231 |
]; |
| 232 |
let x2 = DoubleDouble::quick_mult(u, u); |
| 233 |
let mut p = DoubleDouble::quick_mul_add( |
| 234 |
x2, |
| 235 |
DoubleDouble::from_bit_pair(SIN_C[4]), |
| 236 |
DoubleDouble::from_bit_pair(SIN_C[3]), |
| 237 |
); |
| 238 |
|
| 239 |
p = DoubleDouble::quick_mul_add(x2, p, DoubleDouble::from_bit_pair(SIN_C[2])); |
| 240 |
p = DoubleDouble::quick_mul_add(x2, p, DoubleDouble::from_bit_pair(SIN_C[1])); |
| 241 |
p = DoubleDouble::quick_mul_add(x2, p, DoubleDouble::from_bit_pair(SIN_C[0])); |
| 242 |
let sin_u = DoubleDouble::quick_mult(p, u); |
| 243 |
|
| 244 |
const COS_C: [(u64, u64); 5] = [ |
| 245 |
(0x0000000000000000, 0x3ff0000000000000), |
| 246 |
(0x0000000000000000, 0xbfe0000000000000), |
| 247 |
(0x3c45555555555555, 0x3fa5555555555555), |
| 248 |
(0x3bef49f49f49f49f, 0xbf56c16c16c16c17), |
| 249 |
(0x3b3a01a01a01a01a, 0x3efa01a01a01a01a), |
| 250 |
]; |
| 251 |
|
| 252 |
let mut p = DoubleDouble::quick_mul_add( |
| 253 |
x2, |
| 254 |
DoubleDouble::from_bit_pair(COS_C[4]), |
| 255 |
DoubleDouble::from_bit_pair(COS_C[3]), |
| 256 |
); |
| 257 |
|
| 258 |
p = DoubleDouble::quick_mul_add(x2, p, DoubleDouble::from_bit_pair(COS_C[2])); |
| 259 |
p = DoubleDouble::quick_mul_add(x2, p, DoubleDouble::from_bit_pair(COS_C[1])); |
| 260 |
p = DoubleDouble::quick_mul_add(x2, p, DoubleDouble::from_bit_pair(COS_C[0])); |
| 261 |
|
| 262 |
let cos_u = p; |
| 263 |
SinCos { |
| 264 |
v_sin: sin_u, |
| 265 |
v_cos: cos_u, |
| 266 |
err: 0., |
| 267 |
} |
| 268 |
} |
| 269 |
|
| 270 |
#[inline] |
| 271 |
pub(crate) fn range_reduction_small_dd(x: DoubleDouble) -> (DoubleDouble, i64) { |
| 272 |
const MPI_OVER_128: [u64; 5] = [ |
| 273 |
0xbf9921fb54400000, |
| 274 |
0xbd70b4611a600000, |
| 275 |
0xbb43198a2e000000, |
| 276 |
0xb91b839a25200000, |
| 277 |
0xb6b2704453400000, |
| 278 |
]; |
| 279 |
const ONE_TWENTY_EIGHT_OVER_PI_D: f64 = f64::from_bits(0x40445f306dc9c883); |
| 280 |
let prod_hi = DoubleDouble::quick_mult_f64(x, ONE_TWENTY_EIGHT_OVER_PI_D); |
| 281 |
let kd = prod_hi.to_f64().cpu_round(); |
| 282 |
|
| 283 |
let p_hi = f64::from_bits(MPI_OVER_128[0]); // hi |
| 284 |
let p_mid = f64::from_bits(MPI_OVER_128[1]); // mid |
| 285 |
let p_lo = f64::from_bits(MPI_OVER_128[2]); // lo |
| 286 |
let p_lo_lo = f64::from_bits(MPI_OVER_128[3]); // lo_lo |
| 287 |
|
| 288 |
let mut q = DoubleDouble::f64_mul_f64_add(kd, p_hi, x); |
| 289 |
q = DoubleDouble::f64_mul_f64_add(kd, p_mid, q); |
| 290 |
q = DoubleDouble::f64_mul_f64_add(kd, p_lo, q); |
| 291 |
q = DoubleDouble::f64_mul_f64_add(kd, p_lo_lo, q); |
| 292 |
|
| 293 |
(q, unsafe { kd.to_int_unchecked::<i64>() }) |
| 294 |
} |