| 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::i0::bessel_rsqrt_hard; |
| 30 |
use crate::bessel::i0_exp; |
| 31 |
use crate::common::f_fmla; |
| 32 |
use crate::double_double::DoubleDouble; |
| 33 |
use crate::dyadic_float::{DyadicFloat128, DyadicSign}; |
| 34 |
use crate::exponents::rational128_exp; |
| 35 |
use crate::polyeval::{f_estrin_polyeval5, f_polyeval6}; |
| 36 |
|
| 37 |
/// Modified Bessel of the first kind of order 1 |
| 38 |
/// |
| 39 |
/// Max found ULP 0.5 |
| 40 |
pub fn f_i1(x: f64) -> f64 { |
| 41 |
let ux = x.to_bits().wrapping_shl(1); |
| 42 |
|
| 43 |
if ux >= 0x7ffu64 << 53 || ux <= 0x7960000000000000u64 { |
| 44 |
// |x| <= f64::EPSILON, |x| == inf, x == NaN |
| 45 |
if ux <= 0x760af31dc4611874u64 { |
| 46 |
// Power series of I1(x) ~ x/2 + O(x^3) |
| 47 |
// |x| <= 2.2204460492503131e-24 |
| 48 |
return x * 0.5; |
| 49 |
} |
| 50 |
if ux <= 0x7960000000000000u64 { |
| 51 |
// |x| <= f64::EPSILON |
| 52 |
// Power series of I1(x) ~ x/2 + x^3/16 + O(x^4) |
| 53 |
const A0: f64 = 1. / 2.; |
| 54 |
const A1: f64 = 1. / 16.; |
| 55 |
let r0 = f_fmla(x, x * A1, A0); |
| 56 |
return r0 * x; |
| 57 |
} |
| 58 |
if x.is_infinite() { |
| 59 |
return if x.is_sign_positive() { |
| 60 |
f64::INFINITY |
| 61 |
} else { |
| 62 |
f64::NEG_INFINITY |
| 63 |
}; |
| 64 |
} |
| 65 |
return x + f64::NAN; // x == NaN |
| 66 |
} |
| 67 |
|
| 68 |
let xb = x.to_bits() & 0x7fff_ffff_ffff_ffff; |
| 69 |
|
| 70 |
if xb >= 0x40864fe69ff9fec8u64 { |
| 71 |
// |x| >= 713.9876098185423 |
| 72 |
return if x.is_sign_negative() { |
| 73 |
f64::NEG_INFINITY |
| 74 |
} else { |
| 75 |
f64::INFINITY |
| 76 |
}; |
| 77 |
} |
| 78 |
|
| 79 |
static SIGN: [f64; 2] = [1., -1.]; |
| 80 |
|
| 81 |
let sign_scale = SIGN[x.is_sign_negative() as usize]; |
| 82 |
|
| 83 |
if xb < 0x401f000000000000u64 { |
| 84 |
// |x| <= 7.75 |
| 85 |
return f64::copysign(i1_0_to_7p75(f64::from_bits(xb)).to_f64(), sign_scale); |
| 86 |
} |
| 87 |
|
| 88 |
i1_asympt(f64::from_bits(xb), sign_scale) |
| 89 |
} |
| 90 |
|
| 91 |
/** |
| 92 |
Computes |
| 93 |
I1(x) = x/2 * (1 + 1 * (x/2)^2 + (x/2)^4 * P((x/2)^2)) |
| 94 |
|
| 95 |
Polynomial generated by Wolfram Mathematica: |
| 96 |
```text |
| 97 |
<<FunctionApproximations` |
| 98 |
ClearAll["Global`*"] |
| 99 |
f[x_]:=(BesselI[1,x]*2/x-1-1/2(x/2)^2)/(x/2)^4 |
| 100 |
g[z_]:=f[2 Sqrt[z]] |
| 101 |
{err,approx}=MiniMaxApproximation[g[z],{z,{0.000000001,7.5},9,9},WorkingPrecision->60] |
| 102 |
poly=Numerator[approx][[1]]; |
| 103 |
coeffs=CoefficientList[poly,z]; |
| 104 |
TableForm[Table[Row[{"'",NumberForm[coeffs[[i+1]],{50,50},ExponentFunction->(Null&)],"',"}],{i,0,Length[coeffs]-1}]] |
| 105 |
poly=Denominator[approx][[1]]; |
| 106 |
coeffs=CoefficientList[poly,z]; |
| 107 |
TableForm[Table[Row[{"'",NumberForm[coeffs[[i+1]],{50,50},ExponentFunction->(Null&)],"',"}],{i,0,Length[coeffs]-1}]] |
| 108 |
``` |
| 109 |
**/ |
| 110 |
#[inline] |
| 111 |
pub(crate) fn i1_0_to_7p75(x: f64) -> DoubleDouble { |
| 112 |
let half_x = x * 0.5; // this is exact |
| 113 |
let eval_x = DoubleDouble::from_exact_mult(half_x, half_x); |
| 114 |
|
| 115 |
const P: [(u64, u64); 5] = [ |
| 116 |
(0x3c55555555555555, 0x3fb5555555555555), |
| 117 |
(0x3c1253e1df138479, 0x3f7304597c4fbd4c), |
| 118 |
(0x3bcec398b7059ee9, 0x3f287b5b01f6b9c0), |
| 119 |
(0xbb7354e7c92c4f77, 0x3ed21de117470d10), |
| 120 |
(0xbb1d35ac0d7923cc, 0x3e717f3714dddc04), |
| 121 |
]; |
| 122 |
|
| 123 |
let ps_num = f_estrin_polyeval5( |
| 124 |
eval_x.hi, |
| 125 |
f64::from_bits(0x3e063684ca1944a4), |
| 126 |
f64::from_bits(0x3d92ac4a0e48a9bb), |
| 127 |
f64::from_bits(0x3d1425988b0b0aea), |
| 128 |
f64::from_bits(0x3c899839e74ddefc), |
| 129 |
f64::from_bits(0x3bed8747bcdd1e61), |
| 130 |
); |
| 131 |
|
| 132 |
let mut p_num = DoubleDouble::mul_f64_add(eval_x, ps_num, DoubleDouble::from_bit_pair(P[4])); |
| 133 |
p_num = DoubleDouble::mul_add(eval_x, p_num, DoubleDouble::from_bit_pair(P[3])); |
| 134 |
p_num = DoubleDouble::mul_add(eval_x, p_num, DoubleDouble::from_bit_pair(P[2])); |
| 135 |
p_num = DoubleDouble::mul_add(eval_x, p_num, DoubleDouble::from_bit_pair(P[1])); |
| 136 |
p_num = DoubleDouble::mul_add(eval_x, p_num, DoubleDouble::from_bit_pair(P[0])); |
| 137 |
|
| 138 |
const Q: [(u64, u64); 4] = [ |
| 139 |
(0x0000000000000000, 0x3ff0000000000000), |
| 140 |
(0xbc3e59afb81ac7ea, 0xbf9c4848e0661d70), |
| 141 |
(0x3bd62fa3dbc1a19c, 0x3f38a9eafcd7e674), |
| 142 |
(0x3b6f4688b9eab7d0, 0xbecbfdec51454533), |
| 143 |
]; |
| 144 |
|
| 145 |
let ps_den = f_polyeval6( |
| 146 |
eval_x.hi, |
| 147 |
f64::from_bits(0x3e56e7cde9266a32), |
| 148 |
f64::from_bits(0xbddc316dff4a672f), |
| 149 |
f64::from_bits(0x3d5a43aaee30ebb5), |
| 150 |
f64::from_bits(0xbcd1fb023f4f1fa0), |
| 151 |
f64::from_bits(0x3c4089ede324209f), |
| 152 |
f64::from_bits(0xbb9f64f47ba69604), |
| 153 |
); |
| 154 |
|
| 155 |
let mut p_den = DoubleDouble::mul_f64_add(eval_x, ps_den, DoubleDouble::from_bit_pair(Q[3])); |
| 156 |
p_den = DoubleDouble::mul_add(eval_x, p_den, DoubleDouble::from_bit_pair(Q[2])); |
| 157 |
p_den = DoubleDouble::mul_add(eval_x, p_den, DoubleDouble::from_bit_pair(Q[1])); |
| 158 |
p_den = DoubleDouble::mul_add_f64(eval_x, p_den, f64::from_bits(0x3ff0000000000000)); |
| 159 |
|
| 160 |
let p = DoubleDouble::div(p_num, p_den); |
| 161 |
|
| 162 |
let eval_sqr = DoubleDouble::quick_mult(eval_x, eval_x); |
| 163 |
|
| 164 |
let mut z = DoubleDouble::mul_f64_add_f64(eval_x, 0.5, 1.); |
| 165 |
z = DoubleDouble::mul_add(p, eval_sqr, z); |
| 166 |
|
| 167 |
let x_over_05 = DoubleDouble::from_exact_mult(x, 0.5); |
| 168 |
|
| 169 |
let r = DoubleDouble::quick_mult(z, x_over_05); |
| 170 |
|
| 171 |
let err = f_fmla( |
| 172 |
r.hi, |
| 173 |
f64::from_bits(0x3c40000000000000), // 2^-59 |
| 174 |
f64::from_bits(0x3be0000000000000), // 2^-65 |
| 175 |
); |
| 176 |
let ub = r.hi + (r.lo + err); |
| 177 |
let lb = r.hi + (r.lo - err); |
| 178 |
if ub == lb { |
| 179 |
return r; |
| 180 |
} |
| 181 |
i1_0_to_7p5_hard(x) |
| 182 |
} |
| 183 |
|
| 184 |
// Polynomial generated by Wolfram Mathematica: |
| 185 |
// I1(x) = x/2 * (1 + 1 * (x/2)^2 + (x/2)^4 * P((x/2)^2)) |
| 186 |
// <<FunctionApproximations` |
| 187 |
// ClearAll["Global`*"] |
| 188 |
// f[x_]:=(BesselI[1,x]*2/x-1-1/2(x/2)^2)/(x/2)^4 |
| 189 |
// g[z_]:=f[2 Sqrt[z]] |
| 190 |
// {err,approx}=MiniMaxApproximation[g[z],{z,{0.000000001,7.5},9,9},WorkingPrecision->60] |
| 191 |
// poly=Numerator[approx][[1]]; |
| 192 |
// coeffs=CoefficientList[poly,z]; |
| 193 |
// TableForm[Table[Row[{"'",NumberForm[coeffs[[i+1]],{50,50},ExponentFunction->(Null&)],"',"}],{i,0,Length[coeffs]-1}]] |
| 194 |
// poly=Denominator[approx][[1]]; |
| 195 |
// coeffs=CoefficientList[poly,z]; |
| 196 |
// TableForm[Table[Row[{"'",NumberForm[coeffs[[i+1]],{50,50},ExponentFunction->(Null&)],"',"}],{i,0,Length[coeffs]-1}]] |
| 197 |
#[cold] |
| 198 |
#[inline(never)] |
| 199 |
pub(crate) fn i1_0_to_7p5_hard(x: f64) -> DoubleDouble { |
| 200 |
const ONE_OVER_4: f64 = 1. / 4.; |
| 201 |
let eval_x = DoubleDouble::quick_mult_f64(DoubleDouble::from_exact_mult(x, x), ONE_OVER_4); |
| 202 |
|
| 203 |
const P: [(u64, u64); 10] = [ |
| 204 |
(0x3c55555555555555, 0x3fb5555555555555), |
| 205 |
(0x3c1253e1df138479, 0x3f7304597c4fbd4c), |
| 206 |
(0x3bcec398b7059ee9, 0x3f287b5b01f6b9c0), |
| 207 |
(0xbb7354e7c92c4f77, 0x3ed21de117470d10), |
| 208 |
(0xbb1d35ac0d7923cc, 0x3e717f3714dddc04), |
| 209 |
(0xba928dee24678e32, 0x3e063684ca1944a4), |
| 210 |
(0xba36aa59912fcbed, 0x3d92ac4a0e48a9bb), |
| 211 |
(0x39bad76f18b5ad37, 0x3d1425988b0b0aea), |
| 212 |
(0xb923a6bab6928df4, 0x3c899839e74ddefc), |
| 213 |
(0x3864356cdfa7b321, 0x3bed8747bcdd1e61), |
| 214 |
]; |
| 215 |
|
| 216 |
let mut p_num = DoubleDouble::mul_add( |
| 217 |
eval_x, |
| 218 |
DoubleDouble::from_bit_pair(P[9]), |
| 219 |
DoubleDouble::from_bit_pair(P[8]), |
| 220 |
); |
| 221 |
p_num = DoubleDouble::mul_add(eval_x, p_num, DoubleDouble::from_bit_pair(P[7])); |
| 222 |
p_num = DoubleDouble::mul_add(eval_x, p_num, DoubleDouble::from_bit_pair(P[6])); |
| 223 |
p_num = DoubleDouble::mul_add(eval_x, p_num, DoubleDouble::from_bit_pair(P[5])); |
| 224 |
p_num = DoubleDouble::mul_add(eval_x, p_num, DoubleDouble::from_bit_pair(P[4])); |
| 225 |
p_num = DoubleDouble::mul_add(eval_x, p_num, DoubleDouble::from_bit_pair(P[3])); |
| 226 |
p_num = DoubleDouble::mul_add(eval_x, p_num, DoubleDouble::from_bit_pair(P[2])); |
| 227 |
p_num = DoubleDouble::mul_add(eval_x, p_num, DoubleDouble::from_bit_pair(P[1])); |
| 228 |
p_num = DoubleDouble::mul_add(eval_x, p_num, DoubleDouble::from_bit_pair(P[0])); |
| 229 |
|
| 230 |
const Q: [(u64, u64); 10] = [ |
| 231 |
(0x0000000000000000, 0x3ff0000000000000), |
| 232 |
(0xbc3e59afb81ac7ea, 0xbf9c4848e0661d70), |
| 233 |
(0x3bd62fa3dbc1a19c, 0x3f38a9eafcd7e674), |
| 234 |
(0x3b6f4688b9eab7d0, 0xbecbfdec51454533), |
| 235 |
(0x3af0fb4a17103ef8, 0x3e56e7cde9266a32), |
| 236 |
(0xba71755779c6d4bd, 0xbddc316dff4a672f), |
| 237 |
(0x39cf8ed8d449e2c6, 0x3d5a43aaee30ebb5), |
| 238 |
(0x39704e900a373874, 0xbcd1fb023f4f1fa0), |
| 239 |
(0xb8e33e87e4bffbb1, 0x3c4089ede324209f), |
| 240 |
(0x380fb09b3fd49d5c, 0xbb9f64f47ba69604), |
| 241 |
]; |
| 242 |
|
| 243 |
let mut p_den = DoubleDouble::mul_add( |
| 244 |
eval_x, |
| 245 |
DoubleDouble::from_bit_pair(Q[9]), |
| 246 |
DoubleDouble::from_bit_pair(Q[8]), |
| 247 |
); |
| 248 |
p_den = DoubleDouble::mul_add(eval_x, p_den, DoubleDouble::from_bit_pair(Q[7])); |
| 249 |
p_den = DoubleDouble::mul_add(eval_x, p_den, DoubleDouble::from_bit_pair(Q[6])); |
| 250 |
p_den = DoubleDouble::mul_add(eval_x, p_den, DoubleDouble::from_bit_pair(Q[5])); |
| 251 |
p_den = DoubleDouble::mul_add(eval_x, p_den, DoubleDouble::from_bit_pair(Q[4])); |
| 252 |
p_den = DoubleDouble::mul_add(eval_x, p_den, DoubleDouble::from_bit_pair(Q[3])); |
| 253 |
p_den = DoubleDouble::mul_add(eval_x, p_den, DoubleDouble::from_bit_pair(Q[2])); |
| 254 |
p_den = DoubleDouble::mul_add(eval_x, p_den, DoubleDouble::from_bit_pair(Q[1])); |
| 255 |
p_den = DoubleDouble::mul_add(eval_x, p_den, DoubleDouble::from_bit_pair(Q[0])); |
| 256 |
|
| 257 |
let p = DoubleDouble::div(p_num, p_den); |
| 258 |
|
| 259 |
let eval_sqr = DoubleDouble::quick_mult(eval_x, eval_x); |
| 260 |
|
| 261 |
let mut z = DoubleDouble::mul_f64_add_f64(eval_x, 0.5, 1.); |
| 262 |
z = DoubleDouble::mul_add(p, eval_sqr, z); |
| 263 |
|
| 264 |
let x_over_05 = DoubleDouble::from_exact_mult(x, 0.5); |
| 265 |
|
| 266 |
DoubleDouble::quick_mult(z, x_over_05) |
| 267 |
} |
| 268 |
|
| 269 |
/** |
| 270 |
Asymptotic expansion for I1. |
| 271 |
|
| 272 |
Computes: |
| 273 |
sqrt(x) * exp(-x) * I1(x) = Pn(1/x)/Qn(1/x) |
| 274 |
hence: |
| 275 |
I1(x) = Pn(1/x)/Qm(1/x)*exp(x)/sqrt(x) |
| 276 |
|
| 277 |
Generated by Wolfram Mathematica: |
| 278 |
```text |
| 279 |
<<FunctionApproximations` |
| 280 |
ClearAll["Global`*"] |
| 281 |
f[x_]:=Sqrt[x] Exp[-x] BesselI[1,x] |
| 282 |
g[z_]:=f[1/z] |
| 283 |
{err,approx}=MiniMaxApproximation[g[z],{z,{1/713.98,1/7.75},11,11},WorkingPrecision->120] |
| 284 |
poly=Numerator[approx][[1]]; |
| 285 |
coeffs=CoefficientList[poly,z]; |
| 286 |
TableForm[Table[Row[{"'",NumberForm[coeffs[[i+1]],{50,50},ExponentFunction->(Null&)],"',"}],{i,0,Length[coeffs]-1}]] |
| 287 |
poly=Denominator[approx][[1]]; |
| 288 |
coeffs=CoefficientList[poly,z]; |
| 289 |
TableForm[Table[Row[{"'",NumberForm[coeffs[[i+1]],{50,50},ExponentFunction->(Null&)],"',"}],{i,0,Length[coeffs]-1}]] |
| 290 |
``` |
| 291 |
**/ |
| 292 |
#[inline] |
| 293 |
fn i1_asympt(x: f64, sign_scale: f64) -> f64 { |
| 294 |
let dx = x; |
| 295 |
let recip = DoubleDouble::from_quick_recip(x); |
| 296 |
const P: [(u64, u64); 12] = [ |
| 297 |
(0xbc73a823f28a2f5e, 0x3fd9884533d43651), |
| 298 |
(0x3cc0d5bb78e674b3, 0xc0354325c8029263), |
| 299 |
(0x3d20e1155aaaa283, 0x4080c09b027c46a4), |
| 300 |
(0xbd5b90dcf81b99c1, 0xc0bfc1311090c839), |
| 301 |
(0xbd98f2fda9e8fa1b, 0x40f3bb81bb190ae2), |
| 302 |
(0xbdcec960752b60da, 0xc1207c0bbbc31cd9), |
| 303 |
(0x3dd3c9a299c9c41f, 0x414253e25c4584af), |
| 304 |
(0xbde82e7b9a3e1acc, 0xc159a656aece377c), |
| 305 |
(0x3e0d3d30d701a8ab, 0x416398df24c74ef2), |
| 306 |
(0xbdf57b85ab7006e2, 0xc151fd119be1702b), |
| 307 |
(0x3dd760928f4515fd, 0xc1508327e42639bc), |
| 308 |
(0x3dc09e71bc648589, 0x4143e4933afa621c), |
| 309 |
]; |
| 310 |
|
| 311 |
let x2 = DoubleDouble::quick_mult(recip, recip); |
| 312 |
let x4 = DoubleDouble::quick_mult(x2, x2); |
| 313 |
let x8 = DoubleDouble::quick_mult(x4, x4); |
| 314 |
|
| 315 |
let e0 = DoubleDouble::mul_add( |
| 316 |
recip, |
| 317 |
DoubleDouble::from_bit_pair(P[1]), |
| 318 |
DoubleDouble::from_bit_pair(P[0]), |
| 319 |
); |
| 320 |
let e1 = DoubleDouble::mul_add( |
| 321 |
recip, |
| 322 |
DoubleDouble::from_bit_pair(P[3]), |
| 323 |
DoubleDouble::from_bit_pair(P[2]), |
| 324 |
); |
| 325 |
let e2 = DoubleDouble::mul_add( |
| 326 |
recip, |
| 327 |
DoubleDouble::from_bit_pair(P[5]), |
| 328 |
DoubleDouble::from_bit_pair(P[4]), |
| 329 |
); |
| 330 |
let e3 = DoubleDouble::mul_add( |
| 331 |
recip, |
| 332 |
DoubleDouble::from_bit_pair(P[7]), |
| 333 |
DoubleDouble::from_bit_pair(P[6]), |
| 334 |
); |
| 335 |
let e4 = DoubleDouble::mul_add( |
| 336 |
recip, |
| 337 |
DoubleDouble::from_bit_pair(P[9]), |
| 338 |
DoubleDouble::from_bit_pair(P[8]), |
| 339 |
); |
| 340 |
let e5 = DoubleDouble::mul_add( |
| 341 |
recip, |
| 342 |
DoubleDouble::from_bit_pair(P[11]), |
| 343 |
DoubleDouble::from_bit_pair(P[10]), |
| 344 |
); |
| 345 |
|
| 346 |
let f0 = DoubleDouble::mul_add(x2, e1, e0); |
| 347 |
let f1 = DoubleDouble::mul_add(x2, e3, e2); |
| 348 |
let f2 = DoubleDouble::mul_add(x2, e5, e4); |
| 349 |
|
| 350 |
let g0 = DoubleDouble::mul_add(x4, f1, f0); |
| 351 |
|
| 352 |
let p_num = DoubleDouble::mul_add(x8, f2, g0); |
| 353 |
|
| 354 |
const Q: [(u64, u64); 12] = [ |
| 355 |
(0x0000000000000000, 0x3ff0000000000000), |
| 356 |
(0xbcb334d5a476d9ad, 0xc04a75f94c1a0c1a), |
| 357 |
(0xbd324d58ed98bfae, 0x4094b00e60301c42), |
| 358 |
(0x3d7c8725666c4360, 0xc0d36b9d28d45928), |
| 359 |
(0x3d7f8457c2945822, 0x4107d6c398a174ed), |
| 360 |
(0x3dbc655ea216594b, 0xc1339393e6776e38), |
| 361 |
(0xbdebb5dffef78272, 0x415537198d23f6a1), |
| 362 |
(0xbdb577f8abad883e, 0xc16c6c399dcd6949), |
| 363 |
(0x3e14261c5362f109, 0x4173c02446576949), |
| 364 |
(0x3dc382ededad42c5, 0xc1547dff5770f4ec), |
| 365 |
(0xbe05c0f74d4c7956, 0xc165c88046952562), |
| 366 |
(0xbdbf9145927aa2c7, 0x414395e46bc45d5b), |
| 367 |
]; |
| 368 |
|
| 369 |
let e0 = DoubleDouble::mul_add_f64( |
| 370 |
recip, |
| 371 |
DoubleDouble::from_bit_pair(Q[1]), |
| 372 |
f64::from_bits(0x3ff0000000000000), |
| 373 |
); |
| 374 |
let e1 = DoubleDouble::mul_add( |
| 375 |
recip, |
| 376 |
DoubleDouble::from_bit_pair(Q[3]), |
| 377 |
DoubleDouble::from_bit_pair(Q[2]), |
| 378 |
); |
| 379 |
let e2 = DoubleDouble::mul_add( |
| 380 |
recip, |
| 381 |
DoubleDouble::from_bit_pair(Q[5]), |
| 382 |
DoubleDouble::from_bit_pair(Q[4]), |
| 383 |
); |
| 384 |
let e3 = DoubleDouble::mul_add( |
| 385 |
recip, |
| 386 |
DoubleDouble::from_bit_pair(Q[7]), |
| 387 |
DoubleDouble::from_bit_pair(Q[6]), |
| 388 |
); |
| 389 |
let e4 = DoubleDouble::mul_add( |
| 390 |
recip, |
| 391 |
DoubleDouble::from_bit_pair(Q[9]), |
| 392 |
DoubleDouble::from_bit_pair(Q[8]), |
| 393 |
); |
| 394 |
let e5 = DoubleDouble::mul_add( |
| 395 |
recip, |
| 396 |
DoubleDouble::from_bit_pair(Q[11]), |
| 397 |
DoubleDouble::from_bit_pair(Q[10]), |
| 398 |
); |
| 399 |
|
| 400 |
let f0 = DoubleDouble::mul_add(x2, e1, e0); |
| 401 |
let f1 = DoubleDouble::mul_add(x2, e3, e2); |
| 402 |
let f2 = DoubleDouble::mul_add(x2, e5, e4); |
| 403 |
|
| 404 |
let g0 = DoubleDouble::mul_add(x4, f1, f0); |
| 405 |
|
| 406 |
let p_den = DoubleDouble::mul_add(x8, f2, g0); |
| 407 |
|
| 408 |
let z = DoubleDouble::div(p_num, p_den); |
| 409 |
|
| 410 |
let e = i0_exp(dx * 0.5); |
| 411 |
let r_sqrt = DoubleDouble::from_rsqrt_fast(dx); |
| 412 |
|
| 413 |
let r = DoubleDouble::quick_mult(z * r_sqrt * e, e); |
| 414 |
|
| 415 |
let err = f_fmla( |
| 416 |
r.hi, |
| 417 |
f64::from_bits(0x3c40000000000000), // 2^-59 |
| 418 |
f64::from_bits(0x3ba0000000000000), // 2^-69 |
| 419 |
); |
| 420 |
let up = r.hi + (r.lo + err); |
| 421 |
let lb = r.hi + (r.lo - err); |
| 422 |
if up == lb { |
| 423 |
return f64::copysign(r.to_f64(), sign_scale); |
| 424 |
} |
| 425 |
i1_asympt_hard(x, sign_scale) |
| 426 |
} |
| 427 |
|
| 428 |
/** |
| 429 |
Asymptotic expansion for I1. |
| 430 |
|
| 431 |
Computes: |
| 432 |
sqrt(x) * exp(-x) * I1(x) = Pn(1/x)/Qn(1/x) |
| 433 |
hence: |
| 434 |
I1(x) = Pn(1/x)/Qm(1/x)*exp(x)/sqrt(x) |
| 435 |
|
| 436 |
Generated by Wolfram Mathematica: |
| 437 |
```text |
| 438 |
<<FunctionApproximations` |
| 439 |
ClearAll["Global`*"] |
| 440 |
f[x_]:=Sqrt[x] Exp[-x] BesselI[1,x] |
| 441 |
g[z_]:=f[1/z] |
| 442 |
{err,approx}=MiniMaxApproximation[g[z],{z,{1/713.98,1/7.75},15,15},WorkingPrecision->120] |
| 443 |
poly=Numerator[approx][[1]]; |
| 444 |
coeffs=CoefficientList[poly,z]; |
| 445 |
TableForm[Table[Row[{"'",NumberForm[coeffs[[i+1]],{50,50},ExponentFunction->(Null&)],"',"}],{i,0,Length[coeffs]-1}]] |
| 446 |
poly=Denominator[approx][[1]]; |
| 447 |
coeffs=CoefficientList[poly,z]; |
| 448 |
TableForm[Table[Row[{"'",NumberForm[coeffs[[i+1]],{50,50},ExponentFunction->(Null&)],"',"}],{i,0,Length[coeffs]-1}]] |
| 449 |
``` |
| 450 |
**/ |
| 451 |
#[cold] |
| 452 |
#[inline(never)] |
| 453 |
fn i1_asympt_hard(x: f64, sign_scale: f64) -> f64 { |
| 454 |
static P: [DyadicFloat128; 16] = [ |
| 455 |
DyadicFloat128 { |
| 456 |
sign: DyadicSign::Pos, |
| 457 |
exponent: -129, |
| 458 |
mantissa: 0xcc42299e_a1b28468_bea7da47_28f13acc_u128, |
| 459 |
}, |
| 460 |
DyadicFloat128 { |
| 461 |
sign: DyadicSign::Neg, |
| 462 |
exponent: -124, |
| 463 |
mantissa: 0xda979406_3df6e66f_cf31c3f5_f194b48c_u128, |
| 464 |
}, |
| 465 |
DyadicFloat128 { |
| 466 |
sign: DyadicSign::Neg, |
| 467 |
exponent: -120, |
| 468 |
mantissa: 0xd60b7b96_c958929b_cabe1d8c_3d874767_u128, |
| 469 |
}, |
| 470 |
DyadicFloat128 { |
| 471 |
sign: DyadicSign::Pos, |
| 472 |
exponent: -113, |
| 473 |
mantissa: 0xd27aad9a_8fb38d56_46ab4510_8479306e_u128, |
| 474 |
}, |
| 475 |
DyadicFloat128 { |
| 476 |
sign: DyadicSign::Neg, |
| 477 |
exponent: -108, |
| 478 |
mantissa: 0xe0167305_c451bd1f_d2f17b68_5c62e2ff_u128, |
| 479 |
}, |
| 480 |
DyadicFloat128 { |
| 481 |
sign: DyadicSign::Pos, |
| 482 |
exponent: -103, |
| 483 |
mantissa: 0x8f6d238f_c80d8e4a_08c130f6_24e1c925_u128, |
| 484 |
}, |
| 485 |
DyadicFloat128 { |
| 486 |
sign: DyadicSign::Neg, |
| 487 |
exponent: -100, |
| 488 |
mantissa: 0xfe32280f_2ea99024_d9924472_92d7ac8f_u128, |
| 489 |
}, |
| 490 |
DyadicFloat128 { |
| 491 |
sign: DyadicSign::Pos, |
| 492 |
exponent: -96, |
| 493 |
mantissa: 0xa48815ac_d265609f_da4ace94_811390b2_u128, |
| 494 |
}, |
| 495 |
DyadicFloat128 { |
| 496 |
sign: DyadicSign::Neg, |
| 497 |
exponent: -93, |
| 498 |
mantissa: 0x9ededfe5_833b4cc1_731efd5c_f8729c6c_u128, |
| 499 |
}, |
| 500 |
DyadicFloat128 { |
| 501 |
sign: DyadicSign::Pos, |
| 502 |
exponent: -91, |
| 503 |
mantissa: 0xe5b43203_2784ae6a_f7458556_0a8308ea_u128, |
| 504 |
}, |
| 505 |
DyadicFloat128 { |
| 506 |
sign: DyadicSign::Neg, |
| 507 |
exponent: -89, |
| 508 |
mantissa: 0xf5df279a_3fb4ef60_8d10adee_7dd2f47b_u128, |
| 509 |
}, |
| 510 |
DyadicFloat128 { |
| 511 |
sign: DyadicSign::Pos, |
| 512 |
exponent: -87, |
| 513 |
mantissa: 0xbdb59963_7a757ed1_87280e0e_7f93ca2b_u128, |
| 514 |
}, |
| 515 |
DyadicFloat128 { |
| 516 |
sign: DyadicSign::Neg, |
| 517 |
exponent: -86, |
| 518 |
mantissa: 0xc87fdea5_53177ca8_c91de5fb_3f8f78d3_u128, |
| 519 |
}, |
| 520 |
DyadicFloat128 { |
| 521 |
sign: DyadicSign::Pos, |
| 522 |
exponent: -85, |
| 523 |
mantissa: 0x846d16a7_4663ef5d_ad42d599_5bc726b8_u128, |
| 524 |
}, |
| 525 |
DyadicFloat128 { |
| 526 |
sign: DyadicSign::Neg, |
| 527 |
exponent: -86, |
| 528 |
mantissa: 0xb3ed2055_74262d95_389f33e4_2ac3774a_u128, |
| 529 |
}, |
| 530 |
DyadicFloat128 { |
| 531 |
sign: DyadicSign::Pos, |
| 532 |
exponent: -88, |
| 533 |
mantissa: 0xa511aa32_c18c34e4_3d029a90_a71b7a55_u128, |
| 534 |
}, |
| 535 |
]; |
| 536 |
static Q: [DyadicFloat128; 16] = [ |
| 537 |
DyadicFloat128 { |
| 538 |
sign: DyadicSign::Pos, |
| 539 |
exponent: -127, |
| 540 |
mantissa: 0x80000000_00000000_00000000_00000000_u128, |
| 541 |
}, |
| 542 |
DyadicFloat128 { |
| 543 |
sign: DyadicSign::Neg, |
| 544 |
exponent: -122, |
| 545 |
mantissa: 0x877b771a_ad8f5fd3_5aacf5f9_f04ee9de_u128, |
| 546 |
}, |
| 547 |
DyadicFloat128 { |
| 548 |
sign: DyadicSign::Neg, |
| 549 |
exponent: -118, |
| 550 |
mantissa: 0x89475ecd_9c84361e_800c8a3a_c8af23bf_u128, |
| 551 |
}, |
| 552 |
DyadicFloat128 { |
| 553 |
sign: DyadicSign::Pos, |
| 554 |
exponent: -111, |
| 555 |
mantissa: 0x837d1196_cf2723f1_23b54da8_225efe05_u128, |
| 556 |
}, |
| 557 |
DyadicFloat128 { |
| 558 |
sign: DyadicSign::Neg, |
| 559 |
exponent: -106, |
| 560 |
mantissa: 0x8ae3aecb_15355751_a9ee12e5_a4dd9dde_u128, |
| 561 |
}, |
| 562 |
DyadicFloat128 { |
| 563 |
sign: DyadicSign::Pos, |
| 564 |
exponent: -102, |
| 565 |
mantissa: 0xb0886afa_bc13f996_ab45d252_75c8f587_u128, |
| 566 |
}, |
| 567 |
DyadicFloat128 { |
| 568 |
sign: DyadicSign::Neg, |
| 569 |
exponent: -98, |
| 570 |
mantissa: 0x9b37d7cd_b114b86b_7d14a389_26599aa1_u128, |
| 571 |
}, |
| 572 |
DyadicFloat128 { |
| 573 |
sign: DyadicSign::Pos, |
| 574 |
exponent: -95, |
| 575 |
mantissa: 0xc716bf54_09d5dd9f_bc16679b_93aaeca4_u128, |
| 576 |
}, |
| 577 |
DyadicFloat128 { |
| 578 |
sign: DyadicSign::Neg, |
| 579 |
exponent: -92, |
| 580 |
mantissa: 0xbe0cd82e_c8af8371_ab028ed9_c7902dd2_u128, |
| 581 |
}, |
| 582 |
DyadicFloat128 { |
| 583 |
sign: DyadicSign::Pos, |
| 584 |
exponent: -89, |
| 585 |
mantissa: 0x875f8d91_8ef5d434_a39d00f9_2aed3d2a_u128, |
| 586 |
}, |
| 587 |
DyadicFloat128 { |
| 588 |
sign: DyadicSign::Neg, |
| 589 |
exponent: -87, |
| 590 |
mantissa: 0x8e030781_5aa4ce7f_70156b82_8b216b7c_u128, |
| 591 |
}, |
| 592 |
DyadicFloat128 { |
| 593 |
sign: DyadicSign::Pos, |
| 594 |
exponent: -86, |
| 595 |
mantissa: 0xd4dd2687_92646fbd_5ea2d422_da64fc0b_u128, |
| 596 |
}, |
| 597 |
DyadicFloat128 { |
| 598 |
sign: DyadicSign::Neg, |
| 599 |
exponent: -85, |
| 600 |
mantissa: 0xd6d72ab3_64b4a827_0499af0f_13a51a80_u128, |
| 601 |
}, |
| 602 |
DyadicFloat128 { |
| 603 |
sign: DyadicSign::Pos, |
| 604 |
exponent: -84, |
| 605 |
mantissa: 0x828f4e8b_728747a9_2cebe54a_810e2681_u128, |
| 606 |
}, |
| 607 |
DyadicFloat128 { |
| 608 |
sign: DyadicSign::Neg, |
| 609 |
exponent: -85, |
| 610 |
mantissa: 0x91570096_36a3fcfb_6b936d44_68dda1be_u128, |
| 611 |
}, |
| 612 |
DyadicFloat128 { |
| 613 |
sign: DyadicSign::Pos, |
| 614 |
exponent: -89, |
| 615 |
mantissa: 0xf082ad00_86024ed4_dd31613b_ec41e3f8_u128, |
| 616 |
}, |
| 617 |
]; |
| 618 |
|
| 619 |
let recip = DyadicFloat128::accurate_reciprocal(x); |
| 620 |
|
| 621 |
let mut p_num = P[15]; |
| 622 |
for i in (0..15).rev() { |
| 623 |
p_num = recip * p_num + P[i]; |
| 624 |
} |
| 625 |
let mut p_den = Q[15]; |
| 626 |
for i in (0..15).rev() { |
| 627 |
p_den = recip * p_den + Q[i]; |
| 628 |
} |
| 629 |
let z = p_num * p_den.reciprocal(); |
| 630 |
let r_sqrt = bessel_rsqrt_hard(x, recip); |
| 631 |
let f_exp = rational128_exp(x); |
| 632 |
(z * r_sqrt * f_exp).fast_as_f64() * sign_scale |
| 633 |
} |
| 634 |
|
| 635 |
#[cfg(test)] |
| 636 |
mod tests { |
| 637 |
use super::*; |
| 638 |
#[test] |
| 639 |
fn test_fi1() { |
| 640 |
assert_eq!( |
| 641 |
f_i1(0.0000000000000000000000000000000006423424234121), |
| 642 |
3.2117121170605e-34 |
| 643 |
); |
| 644 |
assert_eq!(f_i1(7.750000000757874), 315.8524811496668); |
| 645 |
assert_eq!(f_i1(7.482812501363189), 245.58002285881892); |
| 646 |
assert_eq!(f_i1(-7.750000000757874), -315.8524811496668); |
| 647 |
assert_eq!(f_i1(-7.482812501363189), -245.58002285881892); |
| 648 |
assert!(f_i1(f64::NAN).is_nan()); |
| 649 |
assert_eq!(f_i1(f64::INFINITY), f64::INFINITY); |
| 650 |
assert_eq!(f_i1(f64::NEG_INFINITY), f64::NEG_INFINITY); |
| 651 |
assert_eq!(f_i1(0.01), 0.005000062500260418); |
| 652 |
assert_eq!(f_i1(-0.01), -0.005000062500260418); |
| 653 |
assert_eq!(f_i1(-9.01), -1040.752038018038); |
| 654 |
assert_eq!(f_i1(9.01), 1040.752038018038); |
| 655 |
} |
| 656 |
} |