| 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::bessel::alpha0::{ |
| 30 |
bessel_0_asympt_alpha, bessel_0_asympt_alpha_fast, bessel_0_asympt_alpha_hard, |
| 31 |
}; |
| 32 |
use crate::bessel::beta0::{ |
| 33 |
bessel_0_asympt_beta, bessel_0_asympt_beta_fast, bessel_0_asympt_beta_hard, |
| 34 |
}; |
| 35 |
use crate::bessel::i0::bessel_rsqrt_hard; |
| 36 |
use crate::bessel::j0::j0_maclaurin_series; |
| 37 |
use crate::bessel::y0_coeffs::Y0_COEFFS; |
| 38 |
use crate::bessel::y0_coeffs_taylor::Y0_COEFFS_TAYLOR; |
| 39 |
use crate::bessel::y0f_coeffs::{Y0_ZEROS, Y0_ZEROS_VALUES}; |
| 40 |
use crate::common::f_fmla; |
| 41 |
use crate::double_double::DoubleDouble; |
| 42 |
use crate::dyadic_float::{DyadicFloat128, DyadicSign}; |
| 43 |
use crate::logs::log_dd_fast; |
| 44 |
use crate::polyeval::{f_polyeval12, f_polyeval13, f_polyeval15, f_polyeval22, f_polyeval24}; |
| 45 |
use crate::rounding::CpuCeil; |
| 46 |
use crate::sin_helper::{sin_dd_small, sin_dd_small_fast, sin_f128_small}; |
| 47 |
use crate::sincos_reduce::{AngleReduced, rem2pi_any, rem2pi_f128}; |
| 48 |
|
| 49 |
/// Bessel of the second kind of order 0 (Y0) |
| 50 |
pub fn f_y0(x: f64) -> f64 { |
| 51 |
let ix = x.to_bits(); |
| 52 |
|
| 53 |
if ix >= 0x7ffu64 << 52 || ix == 0 { |
| 54 |
// |x| == NaN, x == inf, |x| == 0, x < 0 |
| 55 |
if ix.wrapping_shl(1) == 0 { |
| 56 |
// |x| == 0 |
| 57 |
return f64::NEG_INFINITY; |
| 58 |
} |
| 59 |
|
| 60 |
if x.is_infinite() { |
| 61 |
if x.is_sign_negative() { |
| 62 |
return f64::NAN; |
| 63 |
} |
| 64 |
return 0.; |
| 65 |
} |
| 66 |
return x + f64::NAN; // x == NaN |
| 67 |
} |
| 68 |
|
| 69 |
let xb = x.to_bits(); |
| 70 |
|
| 71 |
if xb <= 0x4052d9999999999au64 { |
| 72 |
// x <= 75.4 |
| 73 |
if xb <= 0x4000000000000000u64 { |
| 74 |
// x <= 2 |
| 75 |
if xb <= 0x3ff599999999999au64 { |
| 76 |
// x < 1.35 |
| 77 |
return y0_near_zero_fast(x); |
| 78 |
} |
| 79 |
// transient zone from 1.46 to 2 have bad behavior for log poly already, |
| 80 |
// and not yet good to be easily covered, thus it use its own poly |
| 81 |
return y0_transient_area_fast(x); |
| 82 |
} |
| 83 |
return y0_small_argument_fast(x); |
| 84 |
} |
| 85 |
|
| 86 |
y0_asympt_fast(x) |
| 87 |
} |
| 88 |
|
| 89 |
/** |
| 90 |
Generated by SageMath: |
| 91 |
Evaluates: |
| 92 |
Y0(x) = 2/pi*(euler_gamma + log(x/2))*J0(x) - sum((-1)^m*(x/2)^(2*m)/(m!)^2*sum(1+1/2 + ... 1/m)) |
| 93 |
expressed as: |
| 94 |
Y0(x)=log(x)*W0(x) - Z0(x) |
| 95 |
```python |
| 96 |
from sage.all import * |
| 97 |
|
| 98 |
R = LaurentSeriesRing(RealField(300), 'x',default_prec=300) |
| 99 |
x = R.gen() |
| 100 |
N = 10 # Number of terms (adjust as needed) |
| 101 |
gamma = RealField(300)(euler_gamma) |
| 102 |
d2 = RealField(300)(2) |
| 103 |
pi = RealField(300).pi() |
| 104 |
|
| 105 |
# Define J0(x) Taylor expansion at x = 0 |
| 106 |
def j_series(n, x): |
| 107 |
return sum([(-1)**m * (x/2)**(ZZ(n) + ZZ(2)*ZZ(m)) / (ZZ(m).factorial() * (ZZ(m) + ZZ(n)).factorial()) for m in range(N)]) |
| 108 |
|
| 109 |
J0_series = j_series(0, x) |
| 110 |
|
| 111 |
def z_series(x): |
| 112 |
return sum([(-1)**m * (x/2)**(ZZ(2)*ZZ(m)) / ZZ(m).factorial()**ZZ(2) * sum(RealField(300)(1)/RealField(300)(k) for k in range(1, m+1)) for m in range(1, N)]) |
| 113 |
|
| 114 |
W0 = (d2/pi) * J0_series |
| 115 |
Z0 = -gamma * (d2/pi) * J0_series + RealField(300)(2).log() * (d2/pi) * J0_series + (d2/pi) * z_series(x) |
| 116 |
|
| 117 |
# see the series |
| 118 |
print(W0) |
| 119 |
print(Z0) |
| 120 |
``` |
| 121 |
**/ |
| 122 |
#[inline] |
| 123 |
fn y0_near_zero_fast(x: f64) -> f64 { |
| 124 |
const W: [(u64, u64); 15] = [ |
| 125 |
(0xbc86b01ec5417056, 0x3fe45f306dc9c883), |
| 126 |
(0x3c66b01ec5417056, 0xbfc45f306dc9c883), |
| 127 |
(0xbc26b01ec5417056, 0x3f845f306dc9c883), |
| 128 |
(0xbbd67fe4a5feb897, 0xbf321bb945252402), |
| 129 |
(0x3b767fe4a5feb897, 0x3ed21bb945252402), |
| 130 |
(0xbaf5c2495706f745, 0xbe672db9f21b0f5f), |
| 131 |
(0x3a90c8209874dfad, 0x3df49a6c656d62ff), |
| 132 |
(0x3a12921e91b07dd0, 0xbd7ae90af76a4d0f), |
| 133 |
(0xb992921e91b07dd0, 0x3cfae90af76a4d0f), |
| 134 |
(0x39089b0d8a9228ca, 0xbc754331c053fdad), |
| 135 |
(0x3878d321ddfd3c6e, 0x3beb3749ebf0a0dd), |
| 136 |
(0x37e77548130d809b, 0xbb5cca5ae46eae67), |
| 137 |
(0xb73a848e7ca1c943, 0x3ac9976d3cd4293f), |
| 138 |
(0xb6c884706195a054, 0xba336206ff1ce731), |
| 139 |
(0xb6387a7d2389630d, 0x39995103e9f1818f), |
| 140 |
]; |
| 141 |
let x2 = DoubleDouble::from_exact_mult(x, x); |
| 142 |
|
| 143 |
let w0 = f_polyeval12( |
| 144 |
x2.hi, |
| 145 |
f64::from_bits(W[3].1), |
| 146 |
f64::from_bits(W[4].1), |
| 147 |
f64::from_bits(W[5].1), |
| 148 |
f64::from_bits(W[6].1), |
| 149 |
f64::from_bits(W[7].1), |
| 150 |
f64::from_bits(W[8].1), |
| 151 |
f64::from_bits(W[9].1), |
| 152 |
f64::from_bits(W[10].1), |
| 153 |
f64::from_bits(W[11].1), |
| 154 |
f64::from_bits(W[12].1), |
| 155 |
f64::from_bits(W[13].1), |
| 156 |
f64::from_bits(W[14].1), |
| 157 |
); |
| 158 |
|
| 159 |
let mut w = DoubleDouble::mul_f64_add(x2, w0, DoubleDouble::from_bit_pair(W[2])); |
| 160 |
w = DoubleDouble::mul_add(x2, w, DoubleDouble::from_bit_pair(W[1])); |
| 161 |
w = DoubleDouble::mul_add(x2, w, DoubleDouble::from_bit_pair(W[0])); |
| 162 |
|
| 163 |
const Z: [(u64, u64); 15] = [ |
| 164 |
(0xbc5ddfd831a70821, 0x3fb2e4d699cbd01f), |
| 165 |
(0xbc6d93e63489aea6, 0xbfc6bbcb41034286), |
| 166 |
(0xbc1b88525c2e130b, 0x3f9075b1bbf41364), |
| 167 |
(0x3be097334e26e578, 0xbf41a6206b7b973d), |
| 168 |
(0x3b51c64a34c78cda, 0x3ee3e99794203bbd), |
| 169 |
(0xbb1c407b0f5b2805, 0xbe7bce4a600d3ea4), |
| 170 |
(0xbaa57d1e1e88c9ca, 0x3e0a6ee796b871b6), |
| 171 |
(0x3a3b6e7030a77899, 0xbd92393d82c6b2e4), |
| 172 |
(0x397fcfedacb03781, 0x3d131085da82054c), |
| 173 |
(0xb8e45f51f6118e46, 0xbc8f4ed4b492ebcc), |
| 174 |
(0xb89bd46046c3c8de, 0x3c04b7ac8a1b15d0), |
| 175 |
(0x37d1a206fb205e32, 0xbb769201941d0d49), |
| 176 |
(0x3782f38acbf23993, 0x3ae4987e587ab039), |
| 177 |
(0x36b691bdabf5672b, 0xba4ff1953e0a7c5b), |
| 178 |
(0x3636e1c8cd260e18, 0x39b55031dc5e1967), |
| 179 |
]; |
| 180 |
let z0 = f_polyeval12( |
| 181 |
x2.hi, |
| 182 |
f64::from_bits(Z[3].1), |
| 183 |
f64::from_bits(Z[4].1), |
| 184 |
f64::from_bits(Z[5].1), |
| 185 |
f64::from_bits(Z[6].1), |
| 186 |
f64::from_bits(Z[7].1), |
| 187 |
f64::from_bits(Z[8].1), |
| 188 |
f64::from_bits(Z[9].1), |
| 189 |
f64::from_bits(Z[10].1), |
| 190 |
f64::from_bits(Z[11].1), |
| 191 |
f64::from_bits(Z[12].1), |
| 192 |
f64::from_bits(Z[13].1), |
| 193 |
f64::from_bits(Z[14].1), |
| 194 |
); |
| 195 |
|
| 196 |
let mut z = DoubleDouble::mul_f64_add(x2, z0, DoubleDouble::from_bit_pair(Z[2])); |
| 197 |
z = DoubleDouble::mul_add(x2, z, DoubleDouble::from_bit_pair(Z[1])); |
| 198 |
z = DoubleDouble::mul_add(x2, z, DoubleDouble::from_bit_pair(Z[0])); |
| 199 |
let w_log = log_dd_fast(x); |
| 200 |
let p = DoubleDouble::mul_add(w, w_log, -z); |
| 201 |
let err = f_fmla( |
| 202 |
p.hi, |
| 203 |
f64::from_bits(0x3c50000000000000), // 2^-58 |
| 204 |
f64::from_bits(0x3c30000000000000), // 2^-60 |
| 205 |
); |
| 206 |
let ub = p.hi + (p.lo + err); |
| 207 |
let lb = p.hi + (p.lo - err); |
| 208 |
if ub == lb { |
| 209 |
return p.to_f64(); |
| 210 |
} |
| 211 |
y0_near_zero(x, w_log) |
| 212 |
} |
| 213 |
|
| 214 |
/** |
| 215 |
Generated by SageMath: |
| 216 |
Evaluates: |
| 217 |
Y0(x) = 2/pi*(euler_gamma + log(x/2))*J0(x) - sum((-1)^m*(x/2)^(2*m)/(m!)^2*sum(1+1/2 + ... 1/m)) |
| 218 |
expressed as: |
| 219 |
Y0(x)=log(x)*W0(x) - Z0(x) |
| 220 |
```python |
| 221 |
from sage.all import * |
| 222 |
|
| 223 |
R = LaurentSeriesRing(RealField(300), 'x',default_prec=300) |
| 224 |
x = R.gen() |
| 225 |
N = 10 # Number of terms (adjust as needed) |
| 226 |
gamma = RealField(300)(euler_gamma) |
| 227 |
d2 = RealField(300)(2) |
| 228 |
pi = RealField(300).pi() |
| 229 |
|
| 230 |
# Define J0(x) Taylor expansion at x = 0 |
| 231 |
def j_series(n, x): |
| 232 |
return sum([(-1)**m * (x/2)**(ZZ(n) + ZZ(2)*ZZ(m)) / (ZZ(m).factorial() * (ZZ(m) + ZZ(n)).factorial()) for m in range(N)]) |
| 233 |
|
| 234 |
J0_series = j_series(0, x) |
| 235 |
|
| 236 |
def z_series(x): |
| 237 |
return sum([(-1)**m * (x/2)**(ZZ(2)*ZZ(m)) / ZZ(m).factorial()**ZZ(2) * sum(RealField(300)(1)/RealField(300)(k) for k in range(1, m+1)) for m in range(1, N)]) |
| 238 |
|
| 239 |
W0 = (d2/pi) * J0_series |
| 240 |
Z0 = -gamma * (d2/pi) * J0_series + RealField(300)(2).log() * (d2/pi) * J0_series + (d2/pi) * z_series(x) |
| 241 |
|
| 242 |
# see the series |
| 243 |
print(W0) |
| 244 |
print(Z0) |
| 245 |
``` |
| 246 |
**/ |
| 247 |
#[cold] |
| 248 |
#[inline(never)] |
| 249 |
fn y0_near_zero(x: f64, w_log: DoubleDouble) -> f64 { |
| 250 |
const W: [(u64, u64); 15] = [ |
| 251 |
(0xbc86b01ec5417056, 0x3fe45f306dc9c883), |
| 252 |
(0x3c66b01ec5417056, 0xbfc45f306dc9c883), |
| 253 |
(0xbc26b01ec5417056, 0x3f845f306dc9c883), |
| 254 |
(0xbbd67fe4a5feb897, 0xbf321bb945252402), |
| 255 |
(0x3b767fe4a5feb897, 0x3ed21bb945252402), |
| 256 |
(0xbaf5c2495706f745, 0xbe672db9f21b0f5f), |
| 257 |
(0x3a90c8209874dfad, 0x3df49a6c656d62ff), |
| 258 |
(0x3a12921e91b07dd0, 0xbd7ae90af76a4d0f), |
| 259 |
(0xb992921e91b07dd0, 0x3cfae90af76a4d0f), |
| 260 |
(0x39089b0d8a9228ca, 0xbc754331c053fdad), |
| 261 |
(0x3878d321ddfd3c6e, 0x3beb3749ebf0a0dd), |
| 262 |
(0x37e77548130d809b, 0xbb5cca5ae46eae67), |
| 263 |
(0xb73a848e7ca1c943, 0x3ac9976d3cd4293f), |
| 264 |
(0xb6c884706195a054, 0xba336206ff1ce731), |
| 265 |
(0xb6387a7d2389630d, 0x39995103e9f1818f), |
| 266 |
]; |
| 267 |
let x2 = DoubleDouble::from_exact_mult(x, x); |
| 268 |
let w = f_polyeval15( |
| 269 |
x2, |
| 270 |
DoubleDouble::from_bit_pair(W[0]), |
| 271 |
DoubleDouble::from_bit_pair(W[1]), |
| 272 |
DoubleDouble::from_bit_pair(W[2]), |
| 273 |
DoubleDouble::from_bit_pair(W[3]), |
| 274 |
DoubleDouble::from_bit_pair(W[4]), |
| 275 |
DoubleDouble::from_bit_pair(W[5]), |
| 276 |
DoubleDouble::from_bit_pair(W[6]), |
| 277 |
DoubleDouble::from_bit_pair(W[7]), |
| 278 |
DoubleDouble::from_bit_pair(W[8]), |
| 279 |
DoubleDouble::from_bit_pair(W[9]), |
| 280 |
DoubleDouble::from_bit_pair(W[10]), |
| 281 |
DoubleDouble::from_bit_pair(W[11]), |
| 282 |
DoubleDouble::from_bit_pair(W[12]), |
| 283 |
DoubleDouble::from_bit_pair(W[13]), |
| 284 |
DoubleDouble::from_bit_pair(W[14]), |
| 285 |
); |
| 286 |
|
| 287 |
const Z: [(u64, u64); 15] = [ |
| 288 |
(0xbc5ddfd831a70821, 0x3fb2e4d699cbd01f), |
| 289 |
(0xbc6d93e63489aea6, 0xbfc6bbcb41034286), |
| 290 |
(0xbc1b88525c2e130b, 0x3f9075b1bbf41364), |
| 291 |
(0x3be097334e26e578, 0xbf41a6206b7b973d), |
| 292 |
(0x3b51c64a34c78cda, 0x3ee3e99794203bbd), |
| 293 |
(0xbb1c407b0f5b2805, 0xbe7bce4a600d3ea4), |
| 294 |
(0xbaa57d1e1e88c9ca, 0x3e0a6ee796b871b6), |
| 295 |
(0x3a3b6e7030a77899, 0xbd92393d82c6b2e4), |
| 296 |
(0x397fcfedacb03781, 0x3d131085da82054c), |
| 297 |
(0xb8e45f51f6118e46, 0xbc8f4ed4b492ebcc), |
| 298 |
(0xb89bd46046c3c8de, 0x3c04b7ac8a1b15d0), |
| 299 |
(0x37d1a206fb205e32, 0xbb769201941d0d49), |
| 300 |
(0x3782f38acbf23993, 0x3ae4987e587ab039), |
| 301 |
(0x36b691bdabf5672b, 0xba4ff1953e0a7c5b), |
| 302 |
(0x3636e1c8cd260e18, 0x39b55031dc5e1967), |
| 303 |
]; |
| 304 |
let z = f_polyeval15( |
| 305 |
x2, |
| 306 |
DoubleDouble::from_bit_pair(Z[0]), |
| 307 |
DoubleDouble::from_bit_pair(Z[1]), |
| 308 |
DoubleDouble::from_bit_pair(Z[2]), |
| 309 |
DoubleDouble::from_bit_pair(Z[3]), |
| 310 |
DoubleDouble::from_bit_pair(Z[4]), |
| 311 |
DoubleDouble::from_bit_pair(Z[5]), |
| 312 |
DoubleDouble::from_bit_pair(Z[6]), |
| 313 |
DoubleDouble::from_bit_pair(Z[7]), |
| 314 |
DoubleDouble::from_bit_pair(Z[8]), |
| 315 |
DoubleDouble::from_bit_pair(Z[9]), |
| 316 |
DoubleDouble::from_bit_pair(Z[10]), |
| 317 |
DoubleDouble::from_bit_pair(Z[11]), |
| 318 |
DoubleDouble::from_bit_pair(Z[12]), |
| 319 |
DoubleDouble::from_bit_pair(Z[13]), |
| 320 |
DoubleDouble::from_bit_pair(Z[14]), |
| 321 |
); |
| 322 |
DoubleDouble::mul_add(w, w_log, -z).to_f64() |
| 323 |
} |
| 324 |
|
| 325 |
/** |
| 326 |
Path for transient area between 1.35 to 2. |
| 327 |
**/ |
| 328 |
#[inline] |
| 329 |
pub(crate) fn y0_transient_area_fast(x: f64) -> f64 { |
| 330 |
/** |
| 331 |
Polynomial generated by Wolfram: |
| 332 |
```text |
| 333 |
<<FunctionApproximations` |
| 334 |
ClearAll["Global`*"] |
| 335 |
f[x_]:= BesselY[0,x + 2.1971413260310170351490335626990] |
| 336 |
{approx,error} = MiniMaxApproximation[f[x],{x,{1.35 - 2.1971413260310170351490335626990, 2 - 2.1971413260310170351490335626990 },27,0},WorkingPrecision->120] |
| 337 |
poly=error[[1]]; |
| 338 |
coeffs=CoefficientList[poly,x]; |
| 339 |
TableForm[Table[Row[{"'",NumberForm[coeffs[[i+1]],{50,50}, ExponentFunction->(Null&)],"',"}],{i,0,Length[coeffs]-1}]] |
| 340 |
``` |
| 341 |
**/ |
| 342 |
const C: [(u64, u64); 28] = [ |
| 343 |
(0xbc689e111675434b, 0x3fe0aa48442f014b), |
| 344 |
(0x396ffb11562e8c70, 0x3cc1806f07aceb3c), |
| 345 |
(0xbc6156edff56513d, 0xbfd0aa48442f0030), |
| 346 |
(0xbc278dbff5ee7db4, 0x3fa439fac165db2d), |
| 347 |
(0x3c1f9463c2023663, 0x3f80d2af4ebc8fc4), |
| 348 |
(0xbbee09e5733a1236, 0x3f4f716488aebd9c), |
| 349 |
(0xbbf261a1f255ddf4, 0xbf5444bd2a7e6254), |
| 350 |
(0x3bd4f543544f1fe7, 0x3f384c300e8674d8), |
| 351 |
(0xbb96ef8d95fe049b, 0xbf217a0fc83af41e), |
| 352 |
(0x3ba2be82573ae98d, 0x3f0dbc664048e495), |
| 353 |
(0xbb942b15646c85f2, 0xbef8522f83e4a3e3), |
| 354 |
(0x3b7a127725ba4606, 0x3ee775c010ce4146), |
| 355 |
(0x3ae4a02f0b2a18e2, 0x3e7d1d7cf40f9697), |
| 356 |
(0x3b8fcf1a3d27236b, 0x3eea9c226c21712d), |
| 357 |
(0xbb70b3aa0a1e9ffb, 0x3ef8f237831ec74b), |
| 358 |
(0x3baa6c24261245f3, 0x3f08ebfea3ea469e), |
| 359 |
(0x3bb5fa1b8c4587c4, 0x3f1474022b90cbda), |
| 360 |
(0x3b69545db8d098d1, 0x3f1d153dc04c51c0), |
| 361 |
(0x3bc68eab6520d21b, 0x3f2198a4578cb6ca), |
| 362 |
(0xbbc255734bc49c8b, 0x3f2212febf7cecdd), |
| 363 |
(0x3bb8dd02722339f5, 0x3f1f314deec17049), |
| 364 |
(0x3bbb6ef8f04b26a2, 0x3f1657d051699088), |
| 365 |
(0x3b878233fbf501dc, 0x3f0a1a422dafcef6), |
| 366 |
(0xbb73730138d1dbc2, 0x3ef8423cd021f1dd), |
| 367 |
(0x3b7e2a0d7009d709, 0x3ee145cae37afe1b), |
| 368 |
(0x3b6af21aeaba4e57, 0x3ec1bc74380f6d7b), |
| 369 |
(0xbb3607fb9242657f, 0x3e977341fc10cdc8), |
| 370 |
(0xbac747923880f651, 0x3e5e30218bc1fee3), |
| 371 |
]; |
| 372 |
|
| 373 |
const ZERO: DoubleDouble = |
| 374 |
DoubleDouble::from_bit_pair((0xbc8bd1e50d219bfd, 0x400193bed4dff243)); |
| 375 |
|
| 376 |
let r = DoubleDouble::full_add_f64(-ZERO, x); |
| 377 |
|
| 378 |
let p0 = f_polyeval24( |
| 379 |
r.to_f64(), |
| 380 |
f64::from_bits(C[4].1), |
| 381 |
f64::from_bits(C[5].1), |
| 382 |
f64::from_bits(C[6].1), |
| 383 |
f64::from_bits(C[7].1), |
| 384 |
f64::from_bits(C[8].1), |
| 385 |
f64::from_bits(C[9].1), |
| 386 |
f64::from_bits(C[10].1), |
| 387 |
f64::from_bits(C[11].1), |
| 388 |
f64::from_bits(C[12].1), |
| 389 |
f64::from_bits(C[13].1), |
| 390 |
f64::from_bits(C[14].1), |
| 391 |
f64::from_bits(C[15].1), |
| 392 |
f64::from_bits(C[16].1), |
| 393 |
f64::from_bits(C[17].1), |
| 394 |
f64::from_bits(C[18].1), |
| 395 |
f64::from_bits(C[19].1), |
| 396 |
f64::from_bits(C[20].1), |
| 397 |
f64::from_bits(C[21].1), |
| 398 |
f64::from_bits(C[22].1), |
| 399 |
f64::from_bits(C[23].1), |
| 400 |
f64::from_bits(C[24].1), |
| 401 |
f64::from_bits(C[25].1), |
| 402 |
f64::from_bits(C[26].1), |
| 403 |
f64::from_bits(C[27].1), |
| 404 |
); |
| 405 |
|
| 406 |
let mut p = DoubleDouble::mul_f64_add(r, p0, DoubleDouble::from_bit_pair(C[3])); |
| 407 |
p = DoubleDouble::mul_add(p, r, DoubleDouble::from_bit_pair(C[2])); |
| 408 |
p = DoubleDouble::mul_add(p, r, DoubleDouble::from_bit_pair(C[1])); |
| 409 |
p = DoubleDouble::mul_add(p, r, DoubleDouble::from_bit_pair(C[0])); |
| 410 |
|
| 411 |
let err = f_fmla( |
| 412 |
p.hi, |
| 413 |
f64::from_bits(0x3c50000000000000), // 2^-58 |
| 414 |
f64::from_bits(0x3b10000000000000), // 2^-78 |
| 415 |
); |
| 416 |
let ub = p.hi + (p.lo + err); |
| 417 |
let lb = p.hi + (p.lo - err); |
| 418 |
if ub != lb { |
| 419 |
return y0_transient_area_moderate(x); |
| 420 |
} |
| 421 |
p.to_f64() |
| 422 |
} |
| 423 |
|
| 424 |
/** |
| 425 |
Path for transient area between 1.35 to 2. |
| 426 |
**/ |
| 427 |
fn y0_transient_area_moderate(x: f64) -> f64 { |
| 428 |
/** |
| 429 |
Polynomial generated by Wolfram: |
| 430 |
```text |
| 431 |
<<FunctionApproximations` |
| 432 |
ClearAll["Global`*"] |
| 433 |
f[x_]:= BesselY[0,x + 2.1971413260310170351490335626990] |
| 434 |
{approx,error} = MiniMaxApproximation[f[x],{x,{1.35 - 2.1971413260310170351490335626990, 2- 2.1971413260310170351490335626990 },27,0},WorkingPrecision->120] |
| 435 |
poly=error[[1]]; |
| 436 |
coeffs=CoefficientList[poly,x]; |
| 437 |
TableForm[Table[Row[{"'",NumberForm[coeffs[[i+1]],{50,50}, ExponentFunction->(Null&)],"',"}],{i,0,Length[coeffs]-1}]] |
| 438 |
``` |
| 439 |
**/ |
| 440 |
const C: [(u64, u64); 28] = [ |
| 441 |
(0xbc689e111675434b, 0x3fe0aa48442f014b), |
| 442 |
(0x396ffb11562e8c70, 0x3cc1806f07aceb3c), |
| 443 |
(0xbc6156edff56513d, 0xbfd0aa48442f0030), |
| 444 |
(0xbc278dbff5ee7db4, 0x3fa439fac165db2d), |
| 445 |
(0x3c1f9463c2023663, 0x3f80d2af4ebc8fc4), |
| 446 |
(0xbbee09e5733a1236, 0x3f4f716488aebd9c), |
| 447 |
(0xbbf261a1f255ddf4, 0xbf5444bd2a7e6254), |
| 448 |
(0x3bd4f543544f1fe7, 0x3f384c300e8674d8), |
| 449 |
(0xbb96ef8d95fe049b, 0xbf217a0fc83af41e), |
| 450 |
(0x3ba2be82573ae98d, 0x3f0dbc664048e495), |
| 451 |
(0xbb942b15646c85f2, 0xbef8522f83e4a3e3), |
| 452 |
(0x3b7a127725ba4606, 0x3ee775c010ce4146), |
| 453 |
(0x3ae4a02f0b2a18e2, 0x3e7d1d7cf40f9697), |
| 454 |
(0x3b8fcf1a3d27236b, 0x3eea9c226c21712d), |
| 455 |
(0xbb70b3aa0a1e9ffb, 0x3ef8f237831ec74b), |
| 456 |
(0x3baa6c24261245f3, 0x3f08ebfea3ea469e), |
| 457 |
(0x3bb5fa1b8c4587c4, 0x3f1474022b90cbda), |
| 458 |
(0x3b69545db8d098d1, 0x3f1d153dc04c51c0), |
| 459 |
(0x3bc68eab6520d21b, 0x3f2198a4578cb6ca), |
| 460 |
(0xbbc255734bc49c8b, 0x3f2212febf7cecdd), |
| 461 |
(0x3bb8dd02722339f5, 0x3f1f314deec17049), |
| 462 |
(0x3bbb6ef8f04b26a2, 0x3f1657d051699088), |
| 463 |
(0x3b878233fbf501dc, 0x3f0a1a422dafcef6), |
| 464 |
(0xbb73730138d1dbc2, 0x3ef8423cd021f1dd), |
| 465 |
(0x3b7e2a0d7009d709, 0x3ee145cae37afe1b), |
| 466 |
(0x3b6af21aeaba4e57, 0x3ec1bc74380f6d7b), |
| 467 |
(0xbb3607fb9242657f, 0x3e977341fc10cdc8), |
| 468 |
(0xbac747923880f651, 0x3e5e30218bc1fee3), |
| 469 |
]; |
| 470 |
|
| 471 |
const ZERO: DoubleDouble = |
| 472 |
DoubleDouble::from_bit_pair((0xbc8bd1e50d219bfd, 0x400193bed4dff243)); |
| 473 |
|
| 474 |
let mut r = DoubleDouble::full_add_f64(-ZERO, x); |
| 475 |
r = DoubleDouble::from_exact_add(r.hi, r.lo); |
| 476 |
|
| 477 |
let p0 = f_polyeval13( |
| 478 |
r.to_f64(), |
| 479 |
f64::from_bits(C[15].1), |
| 480 |
f64::from_bits(C[16].1), |
| 481 |
f64::from_bits(C[17].1), |
| 482 |
f64::from_bits(C[18].1), |
| 483 |
f64::from_bits(C[19].1), |
| 484 |
f64::from_bits(C[20].1), |
| 485 |
f64::from_bits(C[21].1), |
| 486 |
f64::from_bits(C[22].1), |
| 487 |
f64::from_bits(C[23].1), |
| 488 |
f64::from_bits(C[24].1), |
| 489 |
f64::from_bits(C[25].1), |
| 490 |
f64::from_bits(C[26].1), |
| 491 |
f64::from_bits(C[27].1), |
| 492 |
); |
| 493 |
|
| 494 |
let mut p_e = DoubleDouble::mul_f64_add(r, p0, DoubleDouble::from_bit_pair(C[14])); |
| 495 |
p_e = DoubleDouble::mul_add(p_e, r, DoubleDouble::from_bit_pair(C[13])); |
| 496 |
p_e = DoubleDouble::mul_add(p_e, r, DoubleDouble::from_bit_pair(C[12])); |
| 497 |
p_e = DoubleDouble::mul_add(p_e, r, DoubleDouble::from_bit_pair(C[11])); |
| 498 |
p_e = DoubleDouble::mul_add(p_e, r, DoubleDouble::from_bit_pair(C[10])); |
| 499 |
p_e = DoubleDouble::mul_add(p_e, r, DoubleDouble::from_bit_pair(C[9])); |
| 500 |
p_e = DoubleDouble::mul_add(p_e, r, DoubleDouble::from_bit_pair(C[8])); |
| 501 |
p_e = DoubleDouble::mul_add(p_e, r, DoubleDouble::from_bit_pair(C[7])); |
| 502 |
p_e = DoubleDouble::mul_add(p_e, r, DoubleDouble::from_bit_pair(C[6])); |
| 503 |
p_e = DoubleDouble::mul_add(p_e, r, DoubleDouble::from_bit_pair(C[5])); |
| 504 |
p_e = DoubleDouble::mul_add(p_e, r, DoubleDouble::from_bit_pair(C[4])); |
| 505 |
p_e = DoubleDouble::mul_add(p_e, r, DoubleDouble::from_bit_pair(C[3])); |
| 506 |
p_e = DoubleDouble::mul_add(p_e, r, DoubleDouble::from_bit_pair(C[2])); |
| 507 |
p_e = DoubleDouble::mul_add(p_e, r, DoubleDouble::from_bit_pair(C[1])); |
| 508 |
p_e = DoubleDouble::mul_add(p_e, r, DoubleDouble::from_bit_pair(C[0])); |
| 509 |
|
| 510 |
let p = DoubleDouble::from_exact_add(p_e.hi, p_e.lo); |
| 511 |
let err = f_fmla( |
| 512 |
p.hi, |
| 513 |
f64::from_bits(0x3c10000000000000), // 2^-62 |
| 514 |
f64::from_bits(0x3a30000000000000), // 2^-91 |
| 515 |
); |
| 516 |
let ub = p.hi + (p.lo + err); |
| 517 |
let lb = p.hi + (p.lo - err); |
| 518 |
if ub != lb { |
| 519 |
return y0_transient_area_hard(x); |
| 520 |
} |
| 521 |
p.to_f64() |
| 522 |
} |
| 523 |
|
| 524 |
#[cold] |
| 525 |
#[inline(never)] |
| 526 |
fn y0_transient_area_hard(x: f64) -> f64 { |
| 527 |
const ZERO: DyadicFloat128 = DyadicFloat128 { |
| 528 |
sign: DyadicSign::Pos, |
| 529 |
exponent: -126, |
| 530 |
mantissa: 0x8c9df6a6_ff921721_70d796f3_2017e155_u128, |
| 531 |
}; |
| 532 |
let r = DyadicFloat128::new_from_f64(x) - ZERO; |
| 533 |
/* |
| 534 |
Polynomial generated by Wolfram: |
| 535 |
```text |
| 536 |
<<FunctionApproximations` |
| 537 |
ClearAll["Global`*"] |
| 538 |
f[x_]:= BesselY[0,x + 2.1971413260310170351490335626990] |
| 539 |
{approx,error} = MiniMaxApproximation[f[x],{x,{1.35 - 2.1971413260310170351490335626990, 2- 2.1971413260310170351490335626990 },27,0},WorkingPrecision->120] |
| 540 |
poly=error[[1]]; |
| 541 |
coeffs=CoefficientList[poly,x]; |
| 542 |
TableForm[Table[Row[{"'",NumberForm[coeffs[[i+1]],{50,50}, ExponentFunction->(Null&)],"',"}],{i,0,Length[coeffs]-1}]] |
| 543 |
``` |
| 544 |
*/ |
| 545 |
static C: [DyadicFloat128; 28] = [ |
| 546 |
DyadicFloat128 { |
| 547 |
sign: DyadicSign::Pos, |
| 548 |
exponent: -128, |
| 549 |
mantissa: 0x85524221_780a573b_0f774c55_e5a946dc_u128, |
| 550 |
}, |
| 551 |
DyadicFloat128 { |
| 552 |
sign: DyadicSign::Pos, |
| 553 |
exponent: -178, |
| 554 |
mantissa: 0x8c03783d_6759e3ff_622ac5d1_8df27811_u128, |
| 555 |
}, |
| 556 |
DyadicFloat128 { |
| 557 |
sign: DyadicSign::Neg, |
| 558 |
exponent: -129, |
| 559 |
mantissa: 0x85524221_78018115_6edff565_13cf55ab_u128, |
| 560 |
}, |
| 561 |
DyadicFloat128 { |
| 562 |
sign: DyadicSign::Pos, |
| 563 |
exponent: -132, |
| 564 |
mantissa: 0xa1cfd60b_2ed96743_9200508c_125cf3d8_u128, |
| 565 |
}, |
| 566 |
DyadicFloat128 { |
| 567 |
sign: DyadicSign::Pos, |
| 568 |
exponent: -134, |
| 569 |
mantissa: 0x86957a75_e47e21f9_463c2023_663178df_u128, |
| 570 |
}, |
| 571 |
DyadicFloat128 { |
| 572 |
sign: DyadicSign::Pos, |
| 573 |
exponent: -138, |
| 574 |
mantissa: 0xfb8b2445_75ecdc3e_c35198bd_b9330775_u128, |
| 575 |
}, |
| 576 |
DyadicFloat128 { |
| 577 |
sign: DyadicSign::Neg, |
| 578 |
exponent: -137, |
| 579 |
mantissa: 0xa225e953_f312a24c_343e4abb_be73338f_u128, |
| 580 |
}, |
| 581 |
DyadicFloat128 { |
| 582 |
sign: DyadicSign::Pos, |
| 583 |
exponent: -139, |
| 584 |
mantissa: 0xc2618074_33a6c29e_a86a89e3_fcde0075_u128, |
| 585 |
}, |
| 586 |
DyadicFloat128 { |
| 587 |
sign: DyadicSign::Neg, |
| 588 |
exponent: -140, |
| 589 |
mantissa: 0x8bd07e41_d7a0f05b_be3657f8_126b9715_u128, |
| 590 |
}, |
| 591 |
DyadicFloat128 { |
| 592 |
sign: DyadicSign::Pos, |
| 593 |
exponent: -142, |
| 594 |
mantissa: 0xede33202_4724aa57_d04ae75d_31ad69cd_u128, |
| 595 |
}, |
| 596 |
DyadicFloat128 { |
| 597 |
sign: DyadicSign::Neg, |
| 598 |
exponent: -143, |
| 599 |
mantissa: 0xc2917c1f_251f1a85_62ac8d90_be4550ff_u128, |
| 600 |
}, |
| 601 |
DyadicFloat128 { |
| 602 |
sign: DyadicSign::Pos, |
| 603 |
exponent: -144, |
| 604 |
mantissa: 0xbbae0086_720a31a1_27725ba4_605e0479_u128, |
| 605 |
}, |
| 606 |
DyadicFloat128 { |
| 607 |
sign: DyadicSign::Pos, |
| 608 |
exponent: -151, |
| 609 |
mantissa: 0xe8ebe7a0_7cb4b852_80bc2ca8_63864ee0_u128, |
| 610 |
}, |
| 611 |
DyadicFloat128 { |
| 612 |
sign: DyadicSign::Pos, |
| 613 |
exponent: -144, |
| 614 |
mantissa: 0xd4e11361_0b896bf9_e347a4e4_6d642f8e_u128, |
| 615 |
}, |
| 616 |
DyadicFloat128 { |
| 617 |
sign: DyadicSign::Pos, |
| 618 |
exponent: -143, |
| 619 |
mantissa: 0xc791bc18_f63a577a_62afaf0b_002753e2_u128, |
| 620 |
}, |
| 621 |
DyadicFloat128 { |
| 622 |
sign: DyadicSign::Pos, |
| 623 |
exponent: -142, |
| 624 |
mantissa: 0xc75ff51f_5234f34d_8484c248_be6beef2_u128, |
| 625 |
}, |
| 626 |
DyadicFloat128 { |
| 627 |
sign: DyadicSign::Pos, |
| 628 |
exponent: -141, |
| 629 |
mantissa: 0xa3a0115c_865ed2bf_437188b0_f87883dc_u128, |
| 630 |
}, |
| 631 |
DyadicFloat128 { |
| 632 |
sign: DyadicSign::Pos, |
| 633 |
exponent: -141, |
| 634 |
mantissa: 0xe8a9ee02_628e0019_545db8d0_98d12eff_u128, |
| 635 |
}, |
| 636 |
DyadicFloat128 { |
| 637 |
sign: DyadicSign::Pos, |
| 638 |
exponent: -140, |
| 639 |
mantissa: 0x8cc522bc_65b652d1_d56ca41a_43537f6a_u128, |
| 640 |
}, |
| 641 |
DyadicFloat128 { |
| 642 |
sign: DyadicSign::Pos, |
| 643 |
exponent: -140, |
| 644 |
mantissa: 0x9097f5fb_e766e5b5_5196876c_6ea798ce_u128, |
| 645 |
}, |
| 646 |
DyadicFloat128 { |
| 647 |
sign: DyadicSign::Pos, |
| 648 |
exponent: -141, |
| 649 |
mantissa: 0xf98a6f76_0b824b1b_a04e4467_3ea487e1_u128, |
| 650 |
}, |
| 651 |
DyadicFloat128 { |
| 652 |
sign: DyadicSign::Pos, |
| 653 |
exponent: -141, |
| 654 |
mantissa: 0xb2be828b_4c84436d_df1e0964_d44f420f_u128, |
| 655 |
}, |
| 656 |
DyadicFloat128 { |
| 657 |
sign: DyadicSign::Pos, |
| 658 |
exponent: -142, |
| 659 |
mantissa: 0xd0d2116d_7e77b0bc_119fdfa8_0ee2dfee_u128, |
| 660 |
}, |
| 661 |
DyadicFloat128 { |
| 662 |
sign: DyadicSign::Pos, |
| 663 |
exponent: -143, |
| 664 |
mantissa: 0xc211e681_0f8ee764_67f63971_21f1a199_u128, |
| 665 |
}, |
| 666 |
DyadicFloat128 { |
| 667 |
sign: DyadicSign::Pos, |
| 668 |
exponent: -144, |
| 669 |
mantissa: 0x8a2e571b_d7f0d9e2_a0d7009d_70969fa2_u128, |
| 670 |
}, |
| 671 |
DyadicFloat128 { |
| 672 |
sign: DyadicSign::Pos, |
| 673 |
exponent: -146, |
| 674 |
mantissa: 0x8de3a1c0_7b6bdb5e_435d5749_cadf3edd_u128, |
| 675 |
}, |
| 676 |
DyadicFloat128 { |
| 677 |
sign: DyadicSign::Pos, |
| 678 |
exponent: -149, |
| 679 |
mantissa: 0xbb9a0fe0_866e3d3f_008db7b3_5029fe59_u128, |
| 680 |
}, |
| 681 |
DyadicFloat128 { |
| 682 |
sign: DyadicSign::Pos, |
| 683 |
exponent: -153, |
| 684 |
mantissa: 0xf1810c5e_0ff717a2_e1b71dfc_26babb9f_u128, |
| 685 |
}, |
| 686 |
]; |
| 687 |
let mut z = C[27]; |
| 688 |
for i in (0..27).rev() { |
| 689 |
z = r * z + C[i]; |
| 690 |
} |
| 691 |
z.fast_as_f64() |
| 692 |
} |
| 693 |
|
| 694 |
/// This method on small range searches for nearest zero or extremum. |
| 695 |
/// Then picks stored series expansion at the point end evaluates the poly at the point. |
| 696 |
#[inline] |
| 697 |
pub(crate) fn y0_small_argument_fast(x: f64) -> f64 { |
| 698 |
let x_abs = x; |
| 699 |
|
| 700 |
// let avg_step = 74.607799 / 47.0; |
| 701 |
// let inv_step = 1.0 / avg_step; |
| 702 |
|
| 703 |
const INV_STEP: f64 = 0.6299609508652038; |
| 704 |
|
| 705 |
let fx = x_abs * INV_STEP; |
| 706 |
const Y0_ZEROS_COUNT: f64 = (Y0_ZEROS.len() - 1) as f64; |
| 707 |
let idx0 = unsafe { fx.min(Y0_ZEROS_COUNT).to_int_unchecked::<usize>() }; |
| 708 |
let idx1 = unsafe { |
| 709 |
fx.cpu_ceil() |
| 710 |
.min(Y0_ZEROS_COUNT) |
| 711 |
.to_int_unchecked::<usize>() |
| 712 |
}; |
| 713 |
|
| 714 |
let found_zero0 = DoubleDouble::from_bit_pair(Y0_ZEROS[idx0]); |
| 715 |
let found_zero1 = DoubleDouble::from_bit_pair(Y0_ZEROS[idx1]); |
| 716 |
|
| 717 |
let dist0 = (found_zero0.hi - x_abs).abs(); |
| 718 |
let dist1 = (found_zero1.hi - x_abs).abs(); |
| 719 |
|
| 720 |
let (found_zero, idx, dist) = if dist0 < dist1 { |
| 721 |
(found_zero0, idx0, dist0) |
| 722 |
} else { |
| 723 |
(found_zero1, idx1, dist1) |
| 724 |
}; |
| 725 |
|
| 726 |
if idx == 0 { |
| 727 |
return j0_maclaurin_series(x); |
| 728 |
} |
| 729 |
|
| 730 |
let is_too_close_to_zero = dist.abs() < 1e-3; |
| 731 |
|
| 732 |
let c = if is_too_close_to_zero { |
| 733 |
&Y0_COEFFS_TAYLOR[idx - 1] |
| 734 |
} else { |
| 735 |
&Y0_COEFFS[idx - 1] |
| 736 |
}; |
| 737 |
|
| 738 |
let r = DoubleDouble::full_add_f64(-found_zero, x); |
| 739 |
|
| 740 |
// We hit exact zero, value, better to return it directly |
| 741 |
if dist == 0. { |
| 742 |
return f64::from_bits(Y0_ZEROS_VALUES[idx]); |
| 743 |
} |
| 744 |
|
| 745 |
let p = f_polyeval22( |
| 746 |
r.hi, |
| 747 |
f64::from_bits(c[6].1), |
| 748 |
f64::from_bits(c[7].1), |
| 749 |
f64::from_bits(c[8].1), |
| 750 |
f64::from_bits(c[9].1), |
| 751 |
f64::from_bits(c[10].1), |
| 752 |
f64::from_bits(c[11].1), |
| 753 |
f64::from_bits(c[12].1), |
| 754 |
f64::from_bits(c[13].1), |
| 755 |
f64::from_bits(c[14].1), |
| 756 |
f64::from_bits(c[15].1), |
| 757 |
f64::from_bits(c[16].1), |
| 758 |
f64::from_bits(c[17].1), |
| 759 |
f64::from_bits(c[18].1), |
| 760 |
f64::from_bits(c[19].1), |
| 761 |
f64::from_bits(c[20].1), |
| 762 |
f64::from_bits(c[21].1), |
| 763 |
f64::from_bits(c[22].1), |
| 764 |
f64::from_bits(c[23].1), |
| 765 |
f64::from_bits(c[24].1), |
| 766 |
f64::from_bits(c[25].1), |
| 767 |
f64::from_bits(c[26].1), |
| 768 |
f64::from_bits(c[27].1), |
| 769 |
); |
| 770 |
|
| 771 |
let mut z = DoubleDouble::mul_f64_add(r, p, DoubleDouble::from_bit_pair(c[5])); |
| 772 |
z = DoubleDouble::mul_add(z, r, DoubleDouble::from_bit_pair(c[4])); |
| 773 |
z = DoubleDouble::mul_add(z, r, DoubleDouble::from_bit_pair(c[3])); |
| 774 |
z = DoubleDouble::mul_add(z, r, DoubleDouble::from_bit_pair(c[2])); |
| 775 |
z = DoubleDouble::mul_add(z, r, DoubleDouble::from_bit_pair(c[1])); |
| 776 |
z = DoubleDouble::mul_add(z, r, DoubleDouble::from_bit_pair(c[0])); |
| 777 |
let p = z; |
| 778 |
let err = f_fmla( |
| 779 |
p.hi, |
| 780 |
f64::from_bits(0x3c60000000000000), // 2^-57 |
| 781 |
f64::from_bits(0x3c20000000000000), // 2^-61 |
| 782 |
); |
| 783 |
let ub = p.hi + (p.lo + err); |
| 784 |
let lb = p.hi + (p.lo - err); |
| 785 |
if ub != lb { |
| 786 |
return y0_small_argument_moderate(r, c); |
| 787 |
} |
| 788 |
z.to_f64() |
| 789 |
} |
| 790 |
|
| 791 |
/// This method on small range searches for nearest zero or extremum. |
| 792 |
/// Then picks stored series expansion at the point end evaluates the poly at the point. |
| 793 |
fn y0_small_argument_moderate(r: DoubleDouble, c: &[(u64, u64); 28]) -> f64 { |
| 794 |
let c0 = &c[15..]; |
| 795 |
|
| 796 |
let p0 = f_polyeval13( |
| 797 |
r.to_f64(), |
| 798 |
f64::from_bits(c0[0].1), |
| 799 |
f64::from_bits(c0[1].1), |
| 800 |
f64::from_bits(c0[2].1), |
| 801 |
f64::from_bits(c0[3].1), |
| 802 |
f64::from_bits(c0[4].1), |
| 803 |
f64::from_bits(c0[5].1), |
| 804 |
f64::from_bits(c0[6].1), |
| 805 |
f64::from_bits(c0[7].1), |
| 806 |
f64::from_bits(c0[8].1), |
| 807 |
f64::from_bits(c0[9].1), |
| 808 |
f64::from_bits(c0[10].1), |
| 809 |
f64::from_bits(c0[11].1), |
| 810 |
f64::from_bits(c0[12].1), |
| 811 |
); |
| 812 |
|
| 813 |
let mut p_e = DoubleDouble::mul_f64_add(r, p0, DoubleDouble::from_bit_pair(c[14])); |
| 814 |
p_e = DoubleDouble::mul_add(p_e, r, DoubleDouble::from_bit_pair(c[13])); |
| 815 |
p_e = DoubleDouble::mul_add(p_e, r, DoubleDouble::from_bit_pair(c[12])); |
| 816 |
p_e = DoubleDouble::mul_add(p_e, r, DoubleDouble::from_bit_pair(c[11])); |
| 817 |
p_e = DoubleDouble::mul_add(p_e, r, DoubleDouble::from_bit_pair(c[10])); |
| 818 |
p_e = DoubleDouble::mul_add(p_e, r, DoubleDouble::from_bit_pair(c[9])); |
| 819 |
p_e = DoubleDouble::mul_add(p_e, r, DoubleDouble::from_bit_pair(c[8])); |
| 820 |
p_e = DoubleDouble::mul_add(p_e, r, DoubleDouble::from_bit_pair(c[7])); |
| 821 |
p_e = DoubleDouble::mul_add(p_e, r, DoubleDouble::from_bit_pair(c[6])); |
| 822 |
p_e = DoubleDouble::mul_add(p_e, r, DoubleDouble::from_bit_pair(c[5])); |
| 823 |
p_e = DoubleDouble::mul_add(p_e, r, DoubleDouble::from_bit_pair(c[4])); |
| 824 |
p_e = DoubleDouble::mul_add(p_e, r, DoubleDouble::from_bit_pair(c[3])); |
| 825 |
p_e = DoubleDouble::mul_add(p_e, r, DoubleDouble::from_bit_pair(c[2])); |
| 826 |
p_e = DoubleDouble::mul_add(p_e, r, DoubleDouble::from_bit_pair(c[1])); |
| 827 |
p_e = DoubleDouble::mul_add(p_e, r, DoubleDouble::from_bit_pair(c[0])); |
| 828 |
|
| 829 |
let p = DoubleDouble::from_exact_add(p_e.hi, p_e.lo); |
| 830 |
let err = f_fmla( |
| 831 |
p.hi, |
| 832 |
f64::from_bits(0x3c30000000000000), // 2^-60 |
| 833 |
f64::from_bits(0x3a30000000000000), // 2^-91 |
| 834 |
); |
| 835 |
let ub = p.hi + (p.lo + err); |
| 836 |
let lb = p.hi + (p.lo - err); |
| 837 |
if ub != lb { |
| 838 |
return y0_small_argument_hard(r, c); |
| 839 |
} |
| 840 |
p.to_f64() |
| 841 |
} |
| 842 |
|
| 843 |
#[cold] |
| 844 |
#[inline(never)] |
| 845 |
fn y0_small_argument_hard(r: DoubleDouble, c: &[(u64, u64); 28]) -> f64 { |
| 846 |
let mut p = DoubleDouble::from_bit_pair(c[27]); |
| 847 |
for i in (0..27).rev() { |
| 848 |
p = DoubleDouble::mul_add(r, p, DoubleDouble::from_bit_pair(c[i])); |
| 849 |
p = DoubleDouble::from_exact_add(p.hi, p.lo); |
| 850 |
} |
| 851 |
p.to_f64() |
| 852 |
} |
| 853 |
|
| 854 |
/* |
| 855 |
Evaluates: |
| 856 |
Y0 = sqrt(2/(PI*x)) * beta(x) * sin(x - PI/4 - alpha(x)) |
| 857 |
*/ |
| 858 |
#[inline] |
| 859 |
pub(crate) fn y0_asympt_fast(x: f64) -> f64 { |
| 860 |
const SQRT_2_OVER_PI: DoubleDouble = DoubleDouble::new( |
| 861 |
f64::from_bits(0xbc8cbc0d30ebfd15), |
| 862 |
f64::from_bits(0x3fe9884533d43651), |
| 863 |
); |
| 864 |
const MPI_OVER_4: DoubleDouble = DoubleDouble::new( |
| 865 |
f64::from_bits(0xbc81a62633145c07), |
| 866 |
f64::from_bits(0xbfe921fb54442d18), |
| 867 |
); |
| 868 |
|
| 869 |
let recip = if x.to_bits() > 0x7fd000000000000u64 { |
| 870 |
DoubleDouble::quick_mult_f64(DoubleDouble::from_exact_div_fma(4.0, x), 0.25) |
| 871 |
} else { |
| 872 |
DoubleDouble::from_recip(x) |
| 873 |
}; |
| 874 |
|
| 875 |
let alpha = bessel_0_asympt_alpha_fast(recip); |
| 876 |
let beta = bessel_0_asympt_beta_fast(recip); |
| 877 |
|
| 878 |
let AngleReduced { angle } = rem2pi_any(x); |
| 879 |
|
| 880 |
// Without full subtraction cancellation happens sometimes |
| 881 |
let x0pi34 = DoubleDouble::full_dd_sub(MPI_OVER_4, alpha); |
| 882 |
let r0 = DoubleDouble::full_dd_add(angle, x0pi34); |
| 883 |
|
| 884 |
let m_cos = sin_dd_small_fast(r0); |
| 885 |
let z0 = DoubleDouble::quick_mult(beta, m_cos); |
| 886 |
let r_sqrt = DoubleDouble::from_rsqrt_fast(x); |
| 887 |
let scale = DoubleDouble::quick_mult(SQRT_2_OVER_PI, r_sqrt); |
| 888 |
let r = DoubleDouble::quick_mult(scale, z0); |
| 889 |
let p = DoubleDouble::from_exact_add(r.hi, r.lo); |
| 890 |
let err = f_fmla( |
| 891 |
p.hi, |
| 892 |
f64::from_bits(0x3c40000000000000), // 2^-59 |
| 893 |
f64::from_bits(0x3c10000000000000), // 2^-62 |
| 894 |
); |
| 895 |
let ub = p.hi + (p.lo + err); |
| 896 |
let lb = p.hi + (p.lo - err); |
| 897 |
|
| 898 |
if ub == lb { |
| 899 |
return p.to_f64(); |
| 900 |
} |
| 901 |
y0_asympt(x, recip, r_sqrt, angle) |
| 902 |
} |
| 903 |
|
| 904 |
/* |
| 905 |
Evaluates: |
| 906 |
Y0 = sqrt(2/(PI*x)) * beta(x) * sin(x - PI/4 - alpha(x)) |
| 907 |
*/ |
| 908 |
fn y0_asympt(x: f64, recip: DoubleDouble, r_sqrt: DoubleDouble, angle: DoubleDouble) -> f64 { |
| 909 |
const SQRT_2_OVER_PI: DoubleDouble = DoubleDouble::new( |
| 910 |
f64::from_bits(0xbc8cbc0d30ebfd15), |
| 911 |
f64::from_bits(0x3fe9884533d43651), |
| 912 |
); |
| 913 |
const MPI_OVER_4: DoubleDouble = DoubleDouble::new( |
| 914 |
f64::from_bits(0xbc81a62633145c07), |
| 915 |
f64::from_bits(0xbfe921fb54442d18), |
| 916 |
); |
| 917 |
|
| 918 |
let alpha = bessel_0_asympt_alpha(recip); |
| 919 |
let beta = bessel_0_asympt_beta(recip); |
| 920 |
|
| 921 |
// Without full subtraction cancellation happens sometimes |
| 922 |
let x0pi34 = DoubleDouble::full_dd_sub(MPI_OVER_4, alpha); |
| 923 |
let r0 = DoubleDouble::full_dd_add(angle, x0pi34); |
| 924 |
|
| 925 |
let m_cos = sin_dd_small(r0); |
| 926 |
let z0 = DoubleDouble::quick_mult(beta, m_cos); |
| 927 |
let scale = DoubleDouble::quick_mult(SQRT_2_OVER_PI, r_sqrt); |
| 928 |
let r = DoubleDouble::quick_mult(scale, z0); |
| 929 |
let p = DoubleDouble::from_exact_add(r.hi, r.lo); |
| 930 |
let err = f_fmla( |
| 931 |
p.hi, |
| 932 |
f64::from_bits(0x3c30000000000000), // 2^-60 |
| 933 |
f64::from_bits(0x3a20000000000000), // 2^-93 |
| 934 |
); |
| 935 |
let ub = p.hi + (p.lo + err); |
| 936 |
let lb = p.hi + (p.lo - err); |
| 937 |
|
| 938 |
if ub == lb { |
| 939 |
return p.to_f64(); |
| 940 |
} |
| 941 |
y0_asympt_hard(x) |
| 942 |
} |
| 943 |
|
| 944 |
/* |
| 945 |
Evaluates: |
| 946 |
Y0 = sqrt(2/(PI*x)) * beta(x) * sin(x - PI/4 - alpha(x)) |
| 947 |
*/ |
| 948 |
#[cold] |
| 949 |
#[inline(never)] |
| 950 |
fn y0_asympt_hard(x: f64) -> f64 { |
| 951 |
const SQRT_2_OVER_PI: DyadicFloat128 = DyadicFloat128 { |
| 952 |
sign: DyadicSign::Pos, |
| 953 |
exponent: -128, |
| 954 |
mantissa: 0xcc42299e_a1b28468_7e59e280_5d5c7180_u128, |
| 955 |
}; |
| 956 |
|
| 957 |
const MPI_OVER_4: DyadicFloat128 = DyadicFloat128 { |
| 958 |
sign: DyadicSign::Neg, |
| 959 |
exponent: -128, |
| 960 |
mantissa: 0xc90fdaa2_2168c234_c4c6628b_80dc1cd1_u128, |
| 961 |
}; |
| 962 |
|
| 963 |
let x_dyadic = DyadicFloat128::new_from_f64(x); |
| 964 |
let recip = DyadicFloat128::accurate_reciprocal(x); |
| 965 |
|
| 966 |
let alpha = bessel_0_asympt_alpha_hard(recip); |
| 967 |
let beta = bessel_0_asympt_beta_hard(recip); |
| 968 |
|
| 969 |
let angle = rem2pi_f128(x_dyadic); |
| 970 |
|
| 971 |
let x0pi34 = MPI_OVER_4 - alpha; |
| 972 |
let r0 = angle + x0pi34; |
| 973 |
|
| 974 |
let m_sin = sin_f128_small(r0); |
| 975 |
|
| 976 |
let z0 = beta * m_sin; |
| 977 |
let r_sqrt = bessel_rsqrt_hard(x, recip); |
| 978 |
let scale = SQRT_2_OVER_PI * r_sqrt; |
| 979 |
let p = scale * z0; |
| 980 |
p.fast_as_f64() |
| 981 |
} |
| 982 |
|
| 983 |
#[cfg(test)] |
| 984 |
mod tests { |
| 985 |
use super::*; |
| 986 |
|
| 987 |
#[test] |
| 988 |
fn test_y0() { |
| 989 |
//ULP should be less than 0.5, but it was 1017.1969361449036, on 3 result 0.37685001001284685, using f_y0 and MPFR 0.3768500100127904 |
| 990 |
assert_eq!(f_y0(3.), 0.3768500100127904); |
| 991 |
assert_eq!(f_y0(0.906009703874588), 0.01085796448629276); |
| 992 |
assert_eq!(f_y0(80.), -0.05562033908977); |
| 993 |
assert_eq!(f_y0(5.), -0.30851762524903376); |
| 994 |
assert_eq!( |
| 995 |
f_y0(f64::from_bits(0x3fec982eb8d417ea)), |
| 996 |
-0.000000000000000023389279284062102 |
| 997 |
); |
| 998 |
assert!(f_y0(f64::NAN).is_nan()); |
| 999 |
assert_eq!(f_y0(f64::INFINITY), 0.); |
| 1000 |
assert!(f_y0(f64::NEG_INFINITY).is_nan()); |
| 1001 |
} |
| 1002 |
|
| 1003 |
#[test] |
| 1004 |
fn test_y0_edge_values() { |
| 1005 |
assert_eq!(f_y0(0.8900000000138676), -0.0031519646708080126); |
| 1006 |
assert_eq!(f_y0(0.8900000000409116), -0.0031519646469294936); |
| 1007 |
assert_eq!(f_y0(98.1760435789366), 0.0000000000000056889416242533015); |
| 1008 |
assert_eq!( |
| 1009 |
f_y0(91.8929453121571802176), |
| 1010 |
-0.00000000000000007281665706677893 |
| 1011 |
); |
| 1012 |
assert_eq!( |
| 1013 |
f_y0(f64::from_bits(0x6e7c1d741dc52512u64)), |
| 1014 |
f64::from_bits(0x2696f860815bc669) |
| 1015 |
); |
| 1016 |
assert_eq!(f_y0(f64::from_bits(0x3e04cdee58a47edd)), -13.58605001628649); |
| 1017 |
assert_eq!( |
| 1018 |
f_y0(0.89357696627916749), |
| 1019 |
-0.000000000000000023389279284062102 |
| 1020 |
); |
| 1021 |
} |
| 1022 |
} |