| 1 |
/* |
| 2 |
* // Copyright (c) Radzivon Bartoshyk 8/2025. 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::logs::{fast_log_d_to_dd, log_dd}; |
| 36 |
use crate::polyeval::f_polyeval3; |
| 37 |
|
| 38 |
/// Modified Bessel of the second kind of order 1 |
| 39 |
/// |
| 40 |
/// Max ULP 0.5 |
| 41 |
pub fn f_k1(x: f64) -> f64 { |
| 42 |
let ix = x.to_bits(); |
| 43 |
|
| 44 |
if ix >= 0x7ffu64 << 52 || ix == 0 { |
| 45 |
// |x| == NaN, x == inf, |x| == 0, x < 0 |
| 46 |
if ix.wrapping_shl(1) == 0 { |
| 47 |
// |x| == 0 |
| 48 |
return f64::INFINITY; |
| 49 |
} |
| 50 |
if x.is_infinite() { |
| 51 |
return if x.is_sign_positive() { 0. } else { f64::NAN }; |
| 52 |
} |
| 53 |
return x + f64::NAN; // x == NaN |
| 54 |
} |
| 55 |
|
| 56 |
let xb = x.to_bits(); |
| 57 |
|
| 58 |
if xb >= 0x4086140538aa7d38u64 { |
| 59 |
// 706.5025494880165 |
| 60 |
return 0.; |
| 61 |
} |
| 62 |
|
| 63 |
if xb <= 0x3ff0000000000000 { |
| 64 |
// x <= 1 |
| 65 |
return k1_small(x).to_f64(); |
| 66 |
} |
| 67 |
|
| 68 |
k1_asympt(x) |
| 69 |
} |
| 70 |
|
| 71 |
// Generated by Wolfram Mathematica: |
| 72 |
// <<FunctionApproximations` |
| 73 |
// ClearAll["Global`*"] |
| 74 |
// f[x_]:=(BesselI[1,x]*2/x-1-1/2(x/2)^2)/(x/2)^4 |
| 75 |
// g[z_]:=f[2 Sqrt[z]] |
| 76 |
// {err,approx}=MiniMaxApproximation[g[z],{z,{0.000000001,1},5,5},WorkingPrecision->60] |
| 77 |
// poly=Numerator[approx][[1]]; |
| 78 |
// coeffs=CoefficientList[poly,z]; |
| 79 |
// TableForm[Table[Row[{"'",NumberForm[coeffs[[i+1]],{50,50},ExponentFunction->(Null&)],"',"}],{i,0,Length[coeffs]-1}]] |
| 80 |
// poly=Denominator[approx][[1]]; |
| 81 |
// coeffs=CoefficientList[poly,z]; |
| 82 |
// TableForm[Table[Row[{"'",NumberForm[coeffs[[i+1]],{50,50},ExponentFunction->(Null&)],"',"}],{i,0,Length[coeffs]-1}]] |
| 83 |
#[inline] |
| 84 |
fn i1_fast(x: f64) -> DoubleDouble { |
| 85 |
let half_x = x * 0.5; // this is exact |
| 86 |
let eval_x = DoubleDouble::from_exact_mult(half_x, half_x); |
| 87 |
|
| 88 |
const P: [(u64, u64); 3] = [ |
| 89 |
(0x3c5555555553c008, 0x3fb5555555555555), |
| 90 |
(0x3c06f1014b703de8, 0x3f6dfda17d0a2cef), |
| 91 |
(0xbbc2594d655d84db, 0x3f21b2c299108f7b), |
| 92 |
]; |
| 93 |
|
| 94 |
let ps_num = f_polyeval3( |
| 95 |
eval_x.hi, |
| 96 |
f64::from_bits(0x3ec37625c178f5e2), |
| 97 |
f64::from_bits(0x3e5843215f0d5088), |
| 98 |
f64::from_bits(0x3dd97f1f45f47244), |
| 99 |
); |
| 100 |
let mut p_num = DoubleDouble::mul_f64_add(eval_x, ps_num, DoubleDouble::from_bit_pair(P[2])); |
| 101 |
p_num = DoubleDouble::mul_add(eval_x, p_num, DoubleDouble::from_bit_pair(P[1])); |
| 102 |
p_num = DoubleDouble::mul_add(eval_x, p_num, DoubleDouble::from_bit_pair(P[0])); |
| 103 |
|
| 104 |
const Q: [(u64, u64); 3] = [ |
| 105 |
(0x0000000000000000, 0x3ff0000000000000), |
| 106 |
(0xbc32ebd3ac0e6253, 0xbfa42c718ce308f7), |
| 107 |
(0xbbe1626e81e3c1bc, 0x3f482772320eab0e), |
| 108 |
]; |
| 109 |
|
| 110 |
let ps_den = f_polyeval3( |
| 111 |
eval_x.hi, |
| 112 |
f64::from_bits(0xbee169811ef4f4a1), |
| 113 |
f64::from_bits(0x3e6ebdab5dbe02a5), |
| 114 |
f64::from_bits(0xbdeb1dbb29fec52a), |
| 115 |
); |
| 116 |
let mut p_den = DoubleDouble::mul_f64_add(eval_x, ps_den, DoubleDouble::from_bit_pair(Q[2])); |
| 117 |
p_den = DoubleDouble::mul_add(eval_x, p_den, DoubleDouble::from_bit_pair(Q[1])); |
| 118 |
p_den = DoubleDouble::mul_add(eval_x, p_den, DoubleDouble::from_bit_pair(Q[0])); |
| 119 |
let p = DoubleDouble::div(p_num, p_den); |
| 120 |
|
| 121 |
let eval_sqr = DoubleDouble::quick_mult(eval_x, eval_x); |
| 122 |
|
| 123 |
let mut z = DoubleDouble::mul_f64_add_f64(eval_x, 0.5, 1.); |
| 124 |
z = DoubleDouble::mul_add(p, eval_sqr, z); |
| 125 |
|
| 126 |
let x_over_05 = DoubleDouble::from_exact_mult(x, 0.5); |
| 127 |
|
| 128 |
DoubleDouble::quick_mult(z, x_over_05) |
| 129 |
} |
| 130 |
|
| 131 |
/** |
| 132 |
Rational approximant for |
| 133 |
f(x) := BesselK(1, x) - Log(x)*BesselI(1, x) - 1/x |
| 134 |
|
| 135 |
Generated by Wolfram Mathematica: |
| 136 |
```text |
| 137 |
<<FunctionApproximations` |
| 138 |
ClearAll["Global`*"] |
| 139 |
f[x_]:=(BesselK[1,x]-Log[x]BesselI[1,x]-1/x)/x |
| 140 |
g[z_]:=f[Sqrt[z]] |
| 141 |
{err,approx}=MiniMaxApproximation[g[z],{z,{0.0000000000001,1},5,4},WorkingPrecision->60] |
| 142 |
poly=Numerator[approx][[1]]; |
| 143 |
coeffs=CoefficientList[poly,z]; |
| 144 |
TableForm[Table[Row[{"'",NumberForm[coeffs[[i+1]],{50,50},ExponentFunction->(Null&)],"',"}],{i,0,Length[coeffs]-1}]] |
| 145 |
poly=Denominator[approx][[1]]; |
| 146 |
coeffs=CoefficientList[poly,z]; |
| 147 |
TableForm[Table[Row[{"'",NumberForm[coeffs[[i+1]],{50,50},ExponentFunction->(Null&)],"',"}],{i,0,Length[coeffs]-1}]] |
| 148 |
``` |
| 149 |
**/ |
| 150 |
#[inline] |
| 151 |
pub(crate) fn k1_small(x: f64) -> DoubleDouble { |
| 152 |
let rcp = DoubleDouble::from_quick_recip(x); |
| 153 |
let x2 = DoubleDouble::from_exact_mult(x, x); |
| 154 |
|
| 155 |
const P: [(u64, u64); 6] = [ |
| 156 |
(0xbc7037c12b888927, 0xbfd3b5b6028a83d6), |
| 157 |
(0x3c39dba459d023e5, 0xbfb4bac288cfe0cd), |
| 158 |
(0x3be0575395050120, 0xbf6c4a1abe9061df), |
| 159 |
(0x3b755df8e375b3d4, 0xbf0c850679678599), |
| 160 |
(0xbb097e0ec926785f, 0xbe98c4a9b608ae1f), |
| 161 |
(0xbaa029f31c786e81, 0xbe104efe2246ee51), |
| 162 |
]; |
| 163 |
|
| 164 |
let ps_num = f_polyeval3( |
| 165 |
x2.hi, |
| 166 |
f64::from_bits(0xbf0c850679678599), |
| 167 |
f64::from_bits(0xbe98c4a9b608ae1f), |
| 168 |
f64::from_bits(0xbe104efe2246ee51), |
| 169 |
); |
| 170 |
|
| 171 |
let mut p_num = DoubleDouble::mul_f64_add(x2, ps_num, DoubleDouble::from_bit_pair(P[2])); |
| 172 |
p_num = DoubleDouble::mul_add(x2, p_num, DoubleDouble::from_bit_pair(P[1])); |
| 173 |
p_num = DoubleDouble::mul_add(x2, p_num, DoubleDouble::from_bit_pair(P[0])); |
| 174 |
|
| 175 |
const Q: [(u64, u64); 5] = [ |
| 176 |
(0x0000000000000000, 0x3ff0000000000000), |
| 177 |
(0x3c19f62e592f3e71, 0xbf8d3bd595449ca9), |
| 178 |
(0xbba8472b975a12d7, 0x3f194de71babe24a), |
| 179 |
(0xbb2eec4b611c19b5, 0xbe994a5dbec84e4d), |
| 180 |
(0x3a9bae2028402903, 0x3e0981ded64a954b), |
| 181 |
]; |
| 182 |
|
| 183 |
let ps_den = f_fmla( |
| 184 |
x2.hi, |
| 185 |
f64::from_bits(0x3e0981ded64a954b), |
| 186 |
f64::from_bits(0xbe994a5dbec84e4d), |
| 187 |
); |
| 188 |
|
| 189 |
let mut p_den = DoubleDouble::mul_f64_add(x2, ps_den, DoubleDouble::from_bit_pair(Q[2])); |
| 190 |
p_den = DoubleDouble::mul_add(x2, p_den, DoubleDouble::from_bit_pair(Q[1])); |
| 191 |
p_den = DoubleDouble::mul_add_f64(x2, p_den, f64::from_bits(0x3ff0000000000000)); |
| 192 |
|
| 193 |
let p = DoubleDouble::div(p_num, p_den); |
| 194 |
|
| 195 |
let lg = fast_log_d_to_dd(x); |
| 196 |
let v_i = i1_fast(x); |
| 197 |
let z = DoubleDouble::mul_add(v_i, lg, rcp); |
| 198 |
let r = DoubleDouble::mul_f64_add(p, x, z); |
| 199 |
let err = f_fmla( |
| 200 |
r.hi, |
| 201 |
f64::from_bits(0x3c20000000000000), // 2^-61 |
| 202 |
f64::from_bits(0x3a80000000000000), // 2^-87 |
| 203 |
); |
| 204 |
let ub = r.hi + (r.lo + err); |
| 205 |
let lb = r.hi + (r.lo - err); |
| 206 |
if ub == lb { |
| 207 |
return r; |
| 208 |
} |
| 209 |
k1_small_hard(x) |
| 210 |
} |
| 211 |
|
| 212 |
/** |
| 213 |
Rational approximant for |
| 214 |
f(x) := BesselK(1, x) - Log(x)*BesselI(1, x) - 1/x |
| 215 |
|
| 216 |
Generated by Wolfram Mathematica: |
| 217 |
```text |
| 218 |
<<FunctionApproximations` |
| 219 |
ClearAll["Global`*"] |
| 220 |
f[x_]:=(BesselK[1,x]-Log[x]BesselI[1,x]-1/x)/x |
| 221 |
g[z_]:=f[Sqrt[z]] |
| 222 |
{err,approx}=MiniMaxApproximation[g[z],{z,{0.0000000000001,1},5,4},WorkingPrecision->60] |
| 223 |
poly=Numerator[approx][[1]]; |
| 224 |
coeffs=CoefficientList[poly,z]; |
| 225 |
TableForm[Table[Row[{"'",NumberForm[coeffs[[i+1]],{50,50},ExponentFunction->(Null&)],"',"}],{i,0,Length[coeffs]-1}]] |
| 226 |
poly=Denominator[approx][[1]]; |
| 227 |
coeffs=CoefficientList[poly,z]; |
| 228 |
TableForm[Table[Row[{"'",NumberForm[coeffs[[i+1]],{50,50},ExponentFunction->(Null&)],"',"}],{i,0,Length[coeffs]-1}]] |
| 229 |
``` |
| 230 |
**/ |
| 231 |
#[cold] |
| 232 |
#[inline(never)] |
| 233 |
fn k1_small_hard(x: f64) -> DoubleDouble { |
| 234 |
let rcp = DoubleDouble::from_quick_recip(x); |
| 235 |
let x2 = DoubleDouble::from_exact_mult(x, x); |
| 236 |
|
| 237 |
const P: [(u64, u64); 6] = [ |
| 238 |
(0xbc7037c12b888927, 0xbfd3b5b6028a83d6), |
| 239 |
(0x3c39dba459d023e5, 0xbfb4bac288cfe0cd), |
| 240 |
(0x3be0575395050120, 0xbf6c4a1abe9061df), |
| 241 |
(0x3b755df8e375b3d4, 0xbf0c850679678599), |
| 242 |
(0xbb097e0ec926785f, 0xbe98c4a9b608ae1f), |
| 243 |
(0xbaa029f31c786e81, 0xbe104efe2246ee51), |
| 244 |
]; |
| 245 |
|
| 246 |
let mut p_num = DoubleDouble::mul_add( |
| 247 |
x2, |
| 248 |
DoubleDouble::from_bit_pair(P[5]), |
| 249 |
DoubleDouble::from_bit_pair(P[4]), |
| 250 |
); |
| 251 |
p_num = DoubleDouble::mul_add(x2, p_num, DoubleDouble::from_bit_pair(P[3])); |
| 252 |
p_num = DoubleDouble::mul_add(x2, p_num, DoubleDouble::from_bit_pair(P[2])); |
| 253 |
p_num = DoubleDouble::mul_add(x2, p_num, DoubleDouble::from_bit_pair(P[1])); |
| 254 |
p_num = DoubleDouble::mul_add(x2, p_num, DoubleDouble::from_bit_pair(P[0])); |
| 255 |
|
| 256 |
const Q: [(u64, u64); 5] = [ |
| 257 |
(0x0000000000000000, 0x3ff0000000000000), |
| 258 |
(0x3c19f62e592f3e71, 0xbf8d3bd595449ca9), |
| 259 |
(0xbba8472b975a12d7, 0x3f194de71babe24a), |
| 260 |
(0xbb2eec4b611c19b5, 0xbe994a5dbec84e4d), |
| 261 |
(0x3a9bae2028402903, 0x3e0981ded64a954b), |
| 262 |
]; |
| 263 |
|
| 264 |
let mut p_den = DoubleDouble::mul_add( |
| 265 |
x2, |
| 266 |
DoubleDouble::from_bit_pair(Q[4]), |
| 267 |
DoubleDouble::from_bit_pair(Q[3]), |
| 268 |
); |
| 269 |
p_den = DoubleDouble::mul_add(x2, p_den, DoubleDouble::from_bit_pair(Q[2])); |
| 270 |
p_den = DoubleDouble::mul_add(x2, p_den, DoubleDouble::from_bit_pair(Q[1])); |
| 271 |
p_den = DoubleDouble::mul_add(x2, p_den, DoubleDouble::from_bit_pair(Q[0])); |
| 272 |
|
| 273 |
let p = DoubleDouble::div(p_num, p_den); |
| 274 |
|
| 275 |
let lg = log_dd(x); |
| 276 |
let v_i = i1_fast(x); |
| 277 |
let z = DoubleDouble::mul_add(v_i, lg, rcp); |
| 278 |
DoubleDouble::mul_f64_add(p, x, z) |
| 279 |
} |
| 280 |
|
| 281 |
/** |
| 282 |
Generated by Wolfram Mathematica: |
| 283 |
```text |
| 284 |
<<FunctionApproximations` |
| 285 |
ClearAll["Global`*"] |
| 286 |
f[x_]:=Sqrt[x] Exp[x] BesselK[1,x] |
| 287 |
g[z_]:=f[1/z] |
| 288 |
{err,approx}=MiniMaxApproximation[g[z],{z,{0.0000000000001,1},11,11},WorkingPrecision->60] |
| 289 |
poly=Numerator[approx][[1]]; |
| 290 |
coeffs=CoefficientList[poly,z]; |
| 291 |
TableForm[Table[Row[{"'",NumberForm[coeffs[[i+1]],{50,50},ExponentFunction->(Null&)],"',"}],{i,0,Length[coeffs]-1}]] |
| 292 |
poly=Denominator[approx][[1]]; |
| 293 |
coeffs=CoefficientList[poly,z]; |
| 294 |
TableForm[Table[Row[{"'",NumberForm[coeffs[[i+1]],{50,50},ExponentFunction->(Null&)],"',"}],{i,0,Length[coeffs]-1}]] |
| 295 |
``` |
| 296 |
**/ |
| 297 |
#[inline] |
| 298 |
fn k1_asympt(x: f64) -> f64 { |
| 299 |
let recip = DoubleDouble::from_quick_recip(x); |
| 300 |
let e = i0_exp(x * 0.5); |
| 301 |
let r_sqrt = DoubleDouble::from_sqrt(x); |
| 302 |
|
| 303 |
const P: [(u64, u64); 12] = [ |
| 304 |
(0xbc9a6a0690becb3b, 0x3ff40d931ff62706), |
| 305 |
(0xbce573e1bbf2f0b7, 0x40402cebfab5721d), |
| 306 |
(0x3d11a739b7c11e7b, 0x4074f58abc0cfbf1), |
| 307 |
(0xbd2682a09ded0116, 0x409c8315f8facef2), |
| 308 |
(0xbd3a19e91a120168, 0x40b65f7a4caed8b9), |
| 309 |
(0x3d449c3d2b834543, 0x40c4fe41fdb4e7b8), |
| 310 |
(0xbd6bdd415ac7f7e1, 0x40c7aa402d035d03), |
| 311 |
(0x3d528412ff0d6b24, 0x40bf68faddd7d850), |
| 312 |
(0xbd48f4bb3f61dac6, 0x40a75f5650249952), |
| 313 |
(0xbd1dc534b275e309, 0x4081bddd259c0582), |
| 314 |
(0xbcce5103350bd226, 0x4046c7a049014484), |
| 315 |
(0x3c8935f8acd6c1d0, 0x3fef7524082b1859), |
| 316 |
]; |
| 317 |
|
| 318 |
let x2 = DoubleDouble::quick_mult(recip, recip); |
| 319 |
let x4 = DoubleDouble::quick_mult(x2, x2); |
| 320 |
let x8 = DoubleDouble::quick_mult(x4, x4); |
| 321 |
|
| 322 |
let e0 = DoubleDouble::mul_add( |
| 323 |
recip, |
| 324 |
DoubleDouble::from_bit_pair(P[1]), |
| 325 |
DoubleDouble::from_bit_pair(P[0]), |
| 326 |
); |
| 327 |
let e1 = DoubleDouble::mul_add( |
| 328 |
recip, |
| 329 |
DoubleDouble::from_bit_pair(P[3]), |
| 330 |
DoubleDouble::from_bit_pair(P[2]), |
| 331 |
); |
| 332 |
let e2 = DoubleDouble::mul_add( |
| 333 |
recip, |
| 334 |
DoubleDouble::from_bit_pair(P[5]), |
| 335 |
DoubleDouble::from_bit_pair(P[4]), |
| 336 |
); |
| 337 |
let e3 = DoubleDouble::mul_add( |
| 338 |
recip, |
| 339 |
DoubleDouble::from_bit_pair(P[7]), |
| 340 |
DoubleDouble::from_bit_pair(P[6]), |
| 341 |
); |
| 342 |
let e4 = DoubleDouble::mul_add( |
| 343 |
recip, |
| 344 |
DoubleDouble::from_bit_pair(P[9]), |
| 345 |
DoubleDouble::from_bit_pair(P[8]), |
| 346 |
); |
| 347 |
let e5 = DoubleDouble::mul_add( |
| 348 |
recip, |
| 349 |
DoubleDouble::from_bit_pair(P[11]), |
| 350 |
DoubleDouble::from_bit_pair(P[10]), |
| 351 |
); |
| 352 |
|
| 353 |
let f0 = DoubleDouble::mul_add(x2, e1, e0); |
| 354 |
let f1 = DoubleDouble::mul_add(x2, e3, e2); |
| 355 |
let f2 = DoubleDouble::mul_add(x2, e5, e4); |
| 356 |
|
| 357 |
let g0 = DoubleDouble::mul_add(x4, f1, f0); |
| 358 |
|
| 359 |
let p_num = DoubleDouble::mul_add(x8, f2, g0); |
| 360 |
|
| 361 |
const Q: [(u64, u64); 12] = [ |
| 362 |
(0x0000000000000000, 0x3ff0000000000000), |
| 363 |
(0x3cc0d2508437b3f4, 0x40396ff483adec14), |
| 364 |
(0xbd130a9c9f8a5338, 0x4070225588d8c15d), |
| 365 |
(0xbceceba8fa0e65a2, 0x4095481f6684e3bb), |
| 366 |
(0x3d4099f3c178fd2a, 0x40afedc8a778bf42), |
| 367 |
(0xbd3a7e6a6276a3e7, 0x40bc0c060112692e), |
| 368 |
(0x3d11538c155b16d8, 0x40bcb12bd1101782), |
| 369 |
(0xbd5f7b04cdea2c61, 0x40b07fa363202e10), |
| 370 |
(0xbce444ed035b66c6, 0x4093d6fe8f44f838), |
| 371 |
(0xbcf6f88fb942b610, 0x4065c99fa44030c3), |
| 372 |
(0xbcbd1d2aedee5bc9, 0x40207ffabeb00eea), |
| 373 |
(0xbc39a0c8091102c9, 0x3facff3d892cd57a), |
| 374 |
]; |
| 375 |
|
| 376 |
let e0 = DoubleDouble::mul_add_f64( |
| 377 |
recip, |
| 378 |
DoubleDouble::from_bit_pair(Q[1]), |
| 379 |
f64::from_bits(0x3ff0000000000000), |
| 380 |
); |
| 381 |
let e1 = DoubleDouble::mul_add( |
| 382 |
recip, |
| 383 |
DoubleDouble::from_bit_pair(Q[3]), |
| 384 |
DoubleDouble::from_bit_pair(Q[2]), |
| 385 |
); |
| 386 |
let e2 = DoubleDouble::mul_add( |
| 387 |
recip, |
| 388 |
DoubleDouble::from_bit_pair(Q[5]), |
| 389 |
DoubleDouble::from_bit_pair(Q[4]), |
| 390 |
); |
| 391 |
let e3 = DoubleDouble::mul_add( |
| 392 |
recip, |
| 393 |
DoubleDouble::from_bit_pair(Q[7]), |
| 394 |
DoubleDouble::from_bit_pair(Q[6]), |
| 395 |
); |
| 396 |
let e4 = DoubleDouble::mul_add( |
| 397 |
recip, |
| 398 |
DoubleDouble::from_bit_pair(Q[9]), |
| 399 |
DoubleDouble::from_bit_pair(Q[8]), |
| 400 |
); |
| 401 |
let e5 = DoubleDouble::mul_add( |
| 402 |
recip, |
| 403 |
DoubleDouble::from_bit_pair(Q[11]), |
| 404 |
DoubleDouble::from_bit_pair(Q[10]), |
| 405 |
); |
| 406 |
|
| 407 |
let f0 = DoubleDouble::mul_add(x2, e1, e0); |
| 408 |
let f1 = DoubleDouble::mul_add(x2, e3, e2); |
| 409 |
let f2 = DoubleDouble::mul_add(x2, e5, e4); |
| 410 |
|
| 411 |
let g0 = DoubleDouble::mul_add(x4, f1, f0); |
| 412 |
|
| 413 |
let p_den = DoubleDouble::mul_add(x8, f2, g0); |
| 414 |
|
| 415 |
let z = DoubleDouble::div(p_num, p_den); |
| 416 |
|
| 417 |
let mut r_e = DoubleDouble::quick_mult(e, r_sqrt); |
| 418 |
r_e = DoubleDouble::from_exact_add(r_e.hi, r_e.lo); |
| 419 |
r_e = DoubleDouble::quick_mult(r_e, e); |
| 420 |
r_e = DoubleDouble::from_exact_add(r_e.hi, r_e.lo); |
| 421 |
|
| 422 |
let r = DoubleDouble::div(z, r_e); |
| 423 |
|
| 424 |
let err = r.hi * f64::from_bits(0x3c10000000000000); // 2^-61 |
| 425 |
let ub = r.hi + (r.lo + err); |
| 426 |
let lb = r.hi + (r.lo - err); |
| 427 |
if ub != lb { |
| 428 |
return k1_asympt_hard(x); |
| 429 |
} |
| 430 |
r.to_f64() |
| 431 |
} |
| 432 |
|
| 433 |
/** |
| 434 |
Generated by Wolfram Mathematica: |
| 435 |
```text |
| 436 |
<<FunctionApproximations` |
| 437 |
ClearAll["Global`*"] |
| 438 |
f[x_]:=Sqrt[x] Exp[x] BesselK[1,x] |
| 439 |
g[z_]:=f[1/z] |
| 440 |
{err,approx}=MiniMaxApproximation[g[z],{z,{0.0000000000001,1},14,14},WorkingPrecision->70] |
| 441 |
poly=Numerator[approx][[1]]; |
| 442 |
coeffs=CoefficientList[poly,z]; |
| 443 |
TableForm[Table[Row[{"'",NumberForm[coeffs[[i+1]],{50,50},ExponentFunction->(Null&)],"',"}],{i,0,Length[coeffs]-1}]] |
| 444 |
poly=Denominator[approx][[1]]; |
| 445 |
coeffs=CoefficientList[poly,z]; |
| 446 |
TableForm[Table[Row[{"'",NumberForm[coeffs[[i+1]],{50,50},ExponentFunction->(Null&)],"',"}],{i,0,Length[coeffs]-1}]] |
| 447 |
``` |
| 448 |
**/ |
| 449 |
#[cold] |
| 450 |
#[inline(never)] |
| 451 |
fn k1_asympt_hard(x: f64) -> f64 { |
| 452 |
static P: [DyadicFloat128; 15] = [ |
| 453 |
DyadicFloat128 { |
| 454 |
sign: DyadicSign::Pos, |
| 455 |
exponent: -127, |
| 456 |
mantissa: 0xa06c98ff_b1382cb2_be5210ac_f26f25d1_u128, |
| 457 |
}, |
| 458 |
DyadicFloat128 { |
| 459 |
sign: DyadicSign::Pos, |
| 460 |
exponent: -122, |
| 461 |
mantissa: 0xc5f546cb_659a39d0_fafbd188_36ca05b9_u128, |
| 462 |
}, |
| 463 |
DyadicFloat128 { |
| 464 |
sign: DyadicSign::Pos, |
| 465 |
exponent: -118, |
| 466 |
mantissa: 0xcd0b7cfa_de158d26_7084bbe9_f1bdb66d_u128, |
| 467 |
}, |
| 468 |
DyadicFloat128 { |
| 469 |
sign: DyadicSign::Pos, |
| 470 |
exponent: -115, |
| 471 |
mantissa: 0xeac7be2f_957d1260_8849508a_2a5a8972_u128, |
| 472 |
}, |
| 473 |
DyadicFloat128 { |
| 474 |
sign: DyadicSign::Pos, |
| 475 |
exponent: -112, |
| 476 |
mantissa: 0xa4d14fec_fecc6444_4c7b0287_dad71a86_u128, |
| 477 |
}, |
| 478 |
DyadicFloat128 { |
| 479 |
sign: DyadicSign::Pos, |
| 480 |
exponent: -110, |
| 481 |
mantissa: 0x94e3180c_01df9932_ad2acd8b_bab59c05_u128, |
| 482 |
}, |
| 483 |
DyadicFloat128 { |
| 484 |
sign: DyadicSign::Pos, |
| 485 |
exponent: -109, |
| 486 |
mantissa: 0xb0de10f8_74918442_94a96368_8eaa4d0d_u128, |
| 487 |
}, |
| 488 |
DyadicFloat128 { |
| 489 |
sign: DyadicSign::Pos, |
| 490 |
exponent: -108, |
| 491 |
mantissa: 0x8adfea76_d6dbe5d9_46bfaf83_9341f4b5_u128, |
| 492 |
}, |
| 493 |
DyadicFloat128 { |
| 494 |
sign: DyadicSign::Pos, |
| 495 |
exponent: -108, |
| 496 |
mantissa: 0x8f0a4337_b69b602c_cf187222_f3a3379f_u128, |
| 497 |
}, |
| 498 |
DyadicFloat128 { |
| 499 |
sign: DyadicSign::Pos, |
| 500 |
exponent: -109, |
| 501 |
mantissa: 0xbd4c3ebf_c2db0fad_1b425641_cc470043_u128, |
| 502 |
}, |
| 503 |
DyadicFloat128 { |
| 504 |
sign: DyadicSign::Pos, |
| 505 |
exponent: -110, |
| 506 |
mantissa: 0x9b14d29f_9b97e3c8_c1a7b9d0_787f0ddb_u128, |
| 507 |
}, |
| 508 |
DyadicFloat128 { |
| 509 |
sign: DyadicSign::Pos, |
| 510 |
exponent: -112, |
| 511 |
mantissa: 0x93e670d2_07a553ef_a90d4895_cf1b5011_u128, |
| 512 |
}, |
| 513 |
DyadicFloat128 { |
| 514 |
sign: DyadicSign::Pos, |
| 515 |
exponent: -115, |
| 516 |
mantissa: 0x93e0ee0a_cb4d8910_6b4d3e37_f4f9df49_u128, |
| 517 |
}, |
| 518 |
DyadicFloat128 { |
| 519 |
sign: DyadicSign::Pos, |
| 520 |
exponent: -120, |
| 521 |
mantissa: 0xff0ce10d_5585abd1_e8a53a12_65131ad4_u128, |
| 522 |
}, |
| 523 |
DyadicFloat128 { |
| 524 |
sign: DyadicSign::Pos, |
| 525 |
exponent: -126, |
| 526 |
mantissa: 0xf020536d_822cbe51_c8de095a_03367c83_u128, |
| 527 |
}, |
| 528 |
]; |
| 529 |
|
| 530 |
static Q: [DyadicFloat128; 15] = [ |
| 531 |
DyadicFloat128 { |
| 532 |
sign: DyadicSign::Pos, |
| 533 |
exponent: -127, |
| 534 |
mantissa: 0x80000000_00000000_00000000_00000000_u128, |
| 535 |
}, |
| 536 |
DyadicFloat128 { |
| 537 |
sign: DyadicSign::Pos, |
| 538 |
exponent: -122, |
| 539 |
mantissa: 0x9c729dd5_4828a918_42807f58_d485a511_u128, |
| 540 |
}, |
| 541 |
DyadicFloat128 { |
| 542 |
sign: DyadicSign::Pos, |
| 543 |
exponent: -118, |
| 544 |
mantissa: 0x9ff6f631_0794001d_433ab0c5_d4c682a9_u128, |
| 545 |
}, |
| 546 |
DyadicFloat128 { |
| 547 |
sign: DyadicSign::Pos, |
| 548 |
exponent: -115, |
| 549 |
mantissa: 0xb3f81e8b_1e0e85a6_3928342e_c83088a1_u128, |
| 550 |
}, |
| 551 |
DyadicFloat128 { |
| 552 |
sign: DyadicSign::Pos, |
| 553 |
exponent: -113, |
| 554 |
mantissa: 0xf6b1c203_a60d4294_239ad045_2c67c224_u128, |
| 555 |
}, |
| 556 |
DyadicFloat128 { |
| 557 |
sign: DyadicSign::Pos, |
| 558 |
exponent: -111, |
| 559 |
mantissa: 0xd7a98b14_7a499762_abde5c38_3a5b40e4_u128, |
| 560 |
}, |
| 561 |
DyadicFloat128 { |
| 562 |
sign: DyadicSign::Pos, |
| 563 |
exponent: -110, |
| 564 |
mantissa: 0xf4eb8b77_a2cdc686_afd1273f_d464c8b7_u128, |
| 565 |
}, |
| 566 |
DyadicFloat128 { |
| 567 |
sign: DyadicSign::Pos, |
| 568 |
exponent: -109, |
| 569 |
mantissa: 0xb4c1e12a_93ee86fc_930c6f94_cfa6ac3a_u128, |
| 570 |
}, |
| 571 |
DyadicFloat128 { |
| 572 |
sign: DyadicSign::Pos, |
| 573 |
exponent: -109, |
| 574 |
mantissa: 0xaaeaab88_32b776b7_fdd76b0f_24349f41_u128, |
| 575 |
}, |
| 576 |
DyadicFloat128 { |
| 577 |
sign: DyadicSign::Pos, |
| 578 |
exponent: -110, |
| 579 |
mantissa: 0xc8ec9d61_5bf2ee9b_878b4962_4a5cee85_u128, |
| 580 |
}, |
| 581 |
DyadicFloat128 { |
| 582 |
sign: DyadicSign::Pos, |
| 583 |
exponent: -111, |
| 584 |
mantissa: 0x8b97bab0_3351673f_22f10d40_fd1c9ff3_u128, |
| 585 |
}, |
| 586 |
DyadicFloat128 { |
| 587 |
sign: DyadicSign::Pos, |
| 588 |
exponent: -114, |
| 589 |
mantissa: 0xd31cb80a_bf8cbedc_b0dcf7e7_c599f79e_u128, |
| 590 |
}, |
| 591 |
DyadicFloat128 { |
| 592 |
sign: DyadicSign::Pos, |
| 593 |
exponent: -117, |
| 594 |
mantissa: 0x96b354c8_69197193_ea4f608f_81943988_u128, |
| 595 |
}, |
| 596 |
DyadicFloat128 { |
| 597 |
sign: DyadicSign::Pos, |
| 598 |
exponent: -122, |
| 599 |
mantissa: 0x989af1bb_e48b5c44_7cd09746_f15e935a_u128, |
| 600 |
}, |
| 601 |
DyadicFloat128 { |
| 602 |
sign: DyadicSign::Pos, |
| 603 |
exponent: -130, |
| 604 |
mantissa: 0xb7b51326_23c29ed5_8d3dcf5a_79bd9a4f_u128, |
| 605 |
}, |
| 606 |
]; |
| 607 |
|
| 608 |
let recip = DyadicFloat128::accurate_reciprocal(x); |
| 609 |
let e = rational128_exp(x); |
| 610 |
let r_sqrt = bessel_rsqrt_hard(x, recip); |
| 611 |
|
| 612 |
let mut p0 = P[14]; |
| 613 |
for i in (0..14).rev() { |
| 614 |
p0 = recip * p0 + P[i]; |
| 615 |
} |
| 616 |
|
| 617 |
let mut q0 = Q[14]; |
| 618 |
for i in (0..14).rev() { |
| 619 |
q0 = recip * q0 + Q[i]; |
| 620 |
} |
| 621 |
|
| 622 |
let v = p0 * q0.reciprocal(); |
| 623 |
let r = v * (e.reciprocal() * r_sqrt); |
| 624 |
r.fast_as_f64() |
| 625 |
} |
| 626 |
|
| 627 |
#[cfg(test)] |
| 628 |
mod tests { |
| 629 |
use super::*; |
| 630 |
#[test] |
| 631 |
fn test_k1() { |
| 632 |
assert_eq!(f_k1(0.643), 1.184534109892725); |
| 633 |
assert_eq!(f_k1(0.964), 0.6402280656771248); |
| 634 |
assert_eq!(f_k1(2.964), 0.04192888446074039); |
| 635 |
assert_eq!(f_k1(8.43), 9.824733212831289e-5); |
| 636 |
assert_eq!(f_k1(16.43), 2.3142404075259965e-8); |
| 637 |
assert_eq!(f_k1(423.43), 7.793648638470207e-186); |
| 638 |
assert_eq!(f_k1(0.), f64::INFINITY); |
| 639 |
assert_eq!(f_k1(-0.), f64::INFINITY); |
| 640 |
assert!(f_k1(-0.5).is_nan()); |
| 641 |
assert!(f_k1(f64::NEG_INFINITY).is_nan()); |
| 642 |
assert_eq!(f_k1(f64::INFINITY), 0.); |
| 643 |
} |
| 644 |
} |