| 1 |
/* |
| 2 |
* // Copyright (c) Radzivon Bartoshyk 9/2025. 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::common::f_fmla; |
| 30 |
use crate::exponents::core_expdf; |
| 31 |
use crate::polyeval::{f_estrin_polyeval8, f_polyeval6}; |
| 32 |
|
| 33 |
#[inline] |
| 34 |
fn core_erfcx(x: f32) -> f64 { |
| 35 |
// x here is already always > 1 |
| 36 |
let dx = x as f64; |
| 37 |
if x < 8. { |
| 38 |
// Rational approximant generated by Wolfram Mathematica: |
| 39 |
// <<FunctionApproximations` |
| 40 |
// ClearAll["Global`*"] |
| 41 |
// f[x_]:=Exp[x^2]Erfc[x] |
| 42 |
// {err0,approx,err1}=MiniMaxApproximation[f[z],{z,{1,8},7,7},WorkingPrecision->75,MaxIterations->100] |
| 43 |
// num=Numerator[approx]; |
| 44 |
// den=Denominator[approx]; |
| 45 |
// coeffs=CoefficientList[num,z]; |
| 46 |
// TableForm[Table[Row[{"'",NumberForm[coeffs[[i+1]],{50,50},ExponentFunction->(Null&)],"',"}],{i,0,Length[coeffs]-1}]] |
| 47 |
// coeffs=CoefficientList[den,z]; |
| 48 |
// TableForm[Table[Row[{"'",NumberForm[coeffs[[i+1]],{50,50},ExponentFunction->(Null&)],"',"}],{i,0,Length[coeffs]-1}]] |
| 49 |
let p_num = f_estrin_polyeval8( |
| 50 |
dx, |
| 51 |
f64::from_bits(0x3ff00000804c8f8f), |
| 52 |
f64::from_bits(0x3ffb7307ea8fdbeb), |
| 53 |
f64::from_bits(0x3ff7081ba7bc735c), |
| 54 |
f64::from_bits(0x3fe767338b33532a), |
| 55 |
f64::from_bits(0x3fce3c8288507fd6), |
| 56 |
f64::from_bits(0x3fa7ca2cb4ae697f), |
| 57 |
f64::from_bits(0x3f72b11b0dfb2348), |
| 58 |
f64::from_bits(0xbd9f64f0c15c479b), |
| 59 |
); |
| 60 |
let p_den = f_estrin_polyeval8( |
| 61 |
dx, |
| 62 |
f64::from_bits(0x3ff0000000000000), |
| 63 |
f64::from_bits(0x4006c071e850132e), |
| 64 |
f64::from_bits(0x400d30326bc347ee), |
| 65 |
f64::from_bits(0x40060d8d56bada75), |
| 66 |
f64::from_bits(0x3ff56643fc4580eb), |
| 67 |
f64::from_bits(0x3fdb0e194e72a513), |
| 68 |
f64::from_bits(0x3fb5154759b61be3), |
| 69 |
f64::from_bits(0x3f8090b063cce524), |
| 70 |
); |
| 71 |
return p_num / p_den; |
| 72 |
} |
| 73 |
// for large x erfcx(x) ~ 1/sqrt(pi) / x * R(1/x) |
| 74 |
const ONE_OVER_SQRT_PI: f64 = f64::from_bits(0x3fe20dd750429b6d); |
| 75 |
let r = 1. / dx; |
| 76 |
// Rational approximant generated by Wolfram Mathematica: |
| 77 |
// <<FunctionApproximations` |
| 78 |
// ClearAll["Global`*"] |
| 79 |
// f[x_]:=Exp[1/x^2]Erfc[1/x]/x*Sqrt[Pi] |
| 80 |
// {err0,approx}=MiniMaxApproximation[f[z],{z,{2^-12,1/8},5,5},WorkingPrecision->75,MaxIterations->100] |
| 81 |
// num=Numerator[approx][[1]]; |
| 82 |
// den=Denominator[approx][[1]]; |
| 83 |
// coeffs=CoefficientList[num,z]; |
| 84 |
// TableForm[Table[Row[{"'",NumberForm[coeffs[[i+1]],{50,50},ExponentFunction->(Null&)],"',"}],{i,0,Length[coeffs]-1}]] |
| 85 |
// coeffs=CoefficientList[den,z]; |
| 86 |
// TableForm[Table[Row[{"'",NumberForm[coeffs[[i+1]],{50,50},ExponentFunction->(Null&)],"',"}],{i,0,Length[coeffs]-1}]] |
| 87 |
let p_num = f_polyeval6( |
| 88 |
r, |
| 89 |
f64::from_bits(0x3ff0000000000002), |
| 90 |
f64::from_bits(0xbfd09caf2bb541c3), |
| 91 |
f64::from_bits(0x40132238367ae454), |
| 92 |
f64::from_bits(0xc0060bc62c3711b1), |
| 93 |
f64::from_bits(0x40024a90d229158d), |
| 94 |
f64::from_bits(0xc0013665d8ff3813), |
| 95 |
); |
| 96 |
let p_den = f_polyeval6( |
| 97 |
r, |
| 98 |
f64::from_bits(0x3ff0000000000000), |
| 99 |
f64::from_bits(0xbfd09caf2bb5101d), |
| 100 |
f64::from_bits(0x4015223836772f2c), |
| 101 |
f64::from_bits(0xc00715911b5f5f5c), |
| 102 |
f64::from_bits(0x4010b66411ec4e1f), |
| 103 |
f64::from_bits(0xc00b325c767ed436), |
| 104 |
); |
| 105 |
(r * ONE_OVER_SQRT_PI) * (p_num / p_den) |
| 106 |
} |
| 107 |
|
| 108 |
/// Scaled complementary error function (exp(x^2)*erfc(x)) |
| 109 |
/// |
| 110 |
/// ulp 0.5 |
| 111 |
pub fn f_erfcxf(x: f32) -> f32 { |
| 112 |
let ux = x.to_bits().wrapping_shl(1); |
| 113 |
if ux >= 0xffu32 << 24 || ux <= 0x6499_999au32 { |
| 114 |
// |x| == 0, |x| == inf, |x| == NaN, |x| <= 1.19209290e-08 |
| 115 |
if ux <= 0x6499_999au32 { |
| 116 |
// |x| == 0, |x| <= 1.19209290e-08 |
| 117 |
return 1.; |
| 118 |
} |
| 119 |
if x.is_infinite() { |
| 120 |
return if x.is_sign_positive() { |
| 121 |
0. |
| 122 |
} else { |
| 123 |
f32::INFINITY |
| 124 |
}; |
| 125 |
} |
| 126 |
return f32::NAN; // x == NaN |
| 127 |
} |
| 128 |
let ax = x.to_bits() & 0x7fff_ffff; |
| 129 |
if x <= -9.382415 { |
| 130 |
// x <= -9.382415 |
| 131 |
return f32::INFINITY; |
| 132 |
} |
| 133 |
if ax <= 0x34000000u32 { |
| 134 |
// |x| < ulp(1) we use taylor series at 0 |
| 135 |
// erfcx(x) ~ 1-(2 x)/Sqrt[\[Pi]]+x^2-(4 x^3)/(3 Sqrt[\[Pi]])+x^4/2-(8 x^5)/(15 Sqrt[\[Pi]])+O[x]^6 |
| 136 |
#[cfg(any( |
| 137 |
all( |
| 138 |
any(target_arch = "x86", target_arch = "x86_64"), |
| 139 |
target_feature = "fma" |
| 140 |
), |
| 141 |
target_arch = "aarch64" |
| 142 |
))] |
| 143 |
{ |
| 144 |
use crate::common::f_fmlaf; |
| 145 |
const M_TWO_OVER_SQRT_PI: f32 = f32::from_bits(0xbf906ebb); |
| 146 |
return f_fmlaf(x, M_TWO_OVER_SQRT_PI, 1.); |
| 147 |
} |
| 148 |
#[cfg(not(any( |
| 149 |
all( |
| 150 |
any(target_arch = "x86", target_arch = "x86_64"), |
| 151 |
target_feature = "fma" |
| 152 |
), |
| 153 |
target_arch = "aarch64" |
| 154 |
)))] |
| 155 |
{ |
| 156 |
use crate::common::f_fmla; |
| 157 |
const M_TWO_OVER_SQRT_PI: f64 = f64::from_bits(0xbff20dd750429b6d); |
| 158 |
let dx = x as f64; |
| 159 |
return f_fmla(dx, M_TWO_OVER_SQRT_PI, 1.) as f32; |
| 160 |
} |
| 161 |
} |
| 162 |
|
| 163 |
if ax <= 0x3f800000u32 { |
| 164 |
// |x| <= 1 |
| 165 |
let dx = x as f64; |
| 166 |
// Generated by Wolfram Mathematica: |
| 167 |
// <<FunctionApproximations` |
| 168 |
// ClearAll["Global`*"] |
| 169 |
// f[x_]:=Exp[x^2]Erfc[x] |
| 170 |
// {err0,approx}=MiniMaxApproximation[f[z],{z,{-1,1},7,7},WorkingPrecision->75,MaxIterations->100] |
| 171 |
// num=Numerator[approx][[1]]; |
| 172 |
// den=Denominator[approx][[1]]; |
| 173 |
// coeffs=CoefficientList[num,z]; |
| 174 |
// TableForm[Table[Row[{"'",NumberForm[coeffs[[i+1]],{50,50},ExponentFunction->(Null&)],"',"}],{i,0,Length[coeffs]-1}]] |
| 175 |
// coeffs=CoefficientList[den,z]; |
| 176 |
// TableForm[Table[Row[{"'",NumberForm[coeffs[[i+1]],{50,50},ExponentFunction->(Null&)],"',"}],{i,0,Length[coeffs]-1}]] |
| 177 |
let p_num = f_estrin_polyeval8( |
| 178 |
dx, |
| 179 |
f64::from_bits(0x3feffffffffffff8), |
| 180 |
f64::from_bits(0x3ff26c328bd2dc5f), |
| 181 |
f64::from_bits(0x3fe6f91b9fa5f58c), |
| 182 |
f64::from_bits(0x3fd09edf3fcf5ee1), |
| 183 |
f64::from_bits(0x3faddb3bcedbff91), |
| 184 |
f64::from_bits(0x3f7e43b5dd4b7587), |
| 185 |
f64::from_bits(0x3f3baab6b3e61d7b), |
| 186 |
f64::from_bits(0xbe83e7d629825321), |
| 187 |
); |
| 188 |
let p_den = f_estrin_polyeval8( |
| 189 |
dx, |
| 190 |
f64::from_bits(0x3ff0000000000000), |
| 191 |
f64::from_bits(0x40023d04ee0abc28), |
| 192 |
f64::from_bits(0x400252b377263d61), |
| 193 |
f64::from_bits(0x3ff510af7f826479), |
| 194 |
f64::from_bits(0x3fddfc089c4731ed), |
| 195 |
f64::from_bits(0x3fba79b040e28b0a), |
| 196 |
f64::from_bits(0x3f8aea2f3579235a), |
| 197 |
f64::from_bits(0x3f485d2875b4f88c), |
| 198 |
); |
| 199 |
return (p_num / p_den) as f32; |
| 200 |
} |
| 201 |
|
| 202 |
let erfcx_abs_x = core_erfcx(f32::from_bits(ax)); |
| 203 |
if x < 0. { |
| 204 |
// exp(x^2)erfc(-x) = 2*exp(x^2) - erfcx(|x|) |
| 205 |
let dx = x as f64; |
| 206 |
return f_fmla(2., core_expdf(dx * dx), -erfcx_abs_x) as f32; |
| 207 |
} |
| 208 |
erfcx_abs_x as f32 |
| 209 |
} |
| 210 |
|
| 211 |
#[cfg(test)] |
| 212 |
mod tests { |
| 213 |
use super::*; |
| 214 |
#[test] |
| 215 |
fn test_erfcx() { |
| 216 |
assert_eq!(f_erfcxf(5.19209290e-09), 1.0); |
| 217 |
assert_eq!(f_erfcxf(1.19209290e-08), 1.0); |
| 218 |
assert_eq!(f_erfcxf(f32::EPSILON), 0.9999999); |
| 219 |
assert_eq!(f_erfcxf(12.1), 0.046469606); |
| 220 |
assert_eq!(f_erfcxf(7.1), 0.07869752); |
| 221 |
assert_eq!(f_erfcxf(1.1), 0.40173045); |
| 222 |
assert_eq!(f_erfcxf(-0.23), 1.3232007); |
| 223 |
assert_eq!(f_erfcxf(-1.4325), 15.234794); |
| 224 |
assert_eq!(f_erfcxf(-10.), f32::INFINITY); |
| 225 |
assert_eq!(f_erfcxf(f32::INFINITY), 0.); |
| 226 |
assert_eq!(f_erfcxf(f32::NEG_INFINITY), f32::INFINITY); |
| 227 |
assert!(f_erfcxf(f32::NAN).is_nan()); |
| 228 |
} |
| 229 |
} |