| 1 |
/* |
| 2 |
* // Copyright (c) Radzivon Bartoshyk 4/2025. 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, f_fmlaf}; |
| 30 |
|
| 31 |
#[inline(always)] |
| 32 |
fn atanf_gen_impl<Q: Fn(f64, f64, f64) -> f64>(x: f32, fma: Q) -> f32 { |
| 33 |
const PI2: f64 = f64::from_bits(0x3ff921fb54442d18); |
| 34 |
let t = x.to_bits(); |
| 35 |
let e = (t >> 23) & 0xff; |
| 36 |
let gt = e >= 127; |
| 37 |
let ta = t & 0x7fffffff; |
| 38 |
if ta >= 0x4c700518u32 { |
| 39 |
// |x| >= 6.29198e+07 |
| 40 |
if ta > 0x7f800000u32 { |
| 41 |
return x + x; |
| 42 |
} // nan |
| 43 |
return f32::copysign(PI2 as f32, x); // inf or |x| >= 6.29198e+07 |
| 44 |
} |
| 45 |
if e < 127 - 13 { |
| 46 |
// |x| < 2^-13 |
| 47 |
if e < 127 - 25 { |
| 48 |
// |x| < 2^-25 |
| 49 |
if t << 1 == 0 { |
| 50 |
return x; |
| 51 |
} |
| 52 |
let res = f_fmlaf(-x, x.abs(), x); |
| 53 |
return res; |
| 54 |
} |
| 55 |
return f_fmlaf(-f64::from_bits(0x3fd5555560000000) as f32 * x, x * x, x); |
| 56 |
} |
| 57 |
/* now |x| >= 0.00012207 */ |
| 58 |
let mut z = x as f64; |
| 59 |
if gt { |
| 60 |
z = 1.0 / z; |
| 61 |
} /* gt is non-zero for |x| >= 1 */ |
| 62 |
let z2 = z * z; |
| 63 |
let z4 = z2 * z2; |
| 64 |
let z8 = z4 * z4; |
| 65 |
/* polynomials generated using rminimax |
| 66 |
(https://gitlab.inria.fr/sfilip/rminimax) with the following command: |
| 67 |
./ratapprox --function="atan(x)" --dom=[0.000122070,1] --num=[x,x^3,x^5,x^7,x^9,x^11,x^13] --den=[1,x^2,x^4,x^6,x^8,x^10,x^12] --output=atanf.sollya --log |
| 68 |
(see output atanf.sollya) |
| 69 |
The coefficient cd[0] was slightly reduced from the original value |
| 70 |
0.330005 to avoid an exceptional case for |x| = 0.069052 |
| 71 |
and rounding to nearest. |
| 72 |
*/ |
| 73 |
const CN: [u64; 7] = [ |
| 74 |
0x3fd51eccde075d67, |
| 75 |
0x3fea76bb5637f2f2, |
| 76 |
0x3fe81e0eed20de88, |
| 77 |
0x3fd376c8ca67d11d, |
| 78 |
0x3faaec7b69202ac6, |
| 79 |
0x3f69561899acc73e, |
| 80 |
0x3efbf9fa5b67e600, |
| 81 |
]; |
| 82 |
const CD: [u64; 7] = [ |
| 83 |
0x3fd51eccde075d66, |
| 84 |
0x3fedfbdd7b392d28, |
| 85 |
0x3ff0000000000000, |
| 86 |
0x3fdfd22bf0e89b54, |
| 87 |
0x3fbd91ff8b576282, |
| 88 |
0x3f8653ea99fc9bb0, |
| 89 |
0x3f31e7fcc202340a, |
| 90 |
]; |
| 91 |
let mut cn0 = fma(z2, f64::from_bits(CN[1]), f64::from_bits(CN[0])); |
| 92 |
let cn2 = fma(z2, f64::from_bits(CN[3]), f64::from_bits(CN[2])); |
| 93 |
let mut cn4 = fma(z2, f64::from_bits(CN[5]), f64::from_bits(CN[4])); |
| 94 |
let cn6 = f64::from_bits(CN[6]); |
| 95 |
cn0 = fma(z4, cn2, cn0); |
| 96 |
cn4 = fma(z4, cn6, cn4); |
| 97 |
cn0 = fma(z8, cn4, cn0); |
| 98 |
cn0 *= z; |
| 99 |
let mut cd0 = fma(z2, f64::from_bits(CD[1]), f64::from_bits(CD[0])); |
| 100 |
let cd2 = fma(z2, f64::from_bits(CD[3]), f64::from_bits(CD[2])); |
| 101 |
let mut cd4 = fma(z2, f64::from_bits(CD[5]), f64::from_bits(CD[4])); |
| 102 |
let cd6 = f64::from_bits(CD[6]); |
| 103 |
cd0 = fma(z4, cd2, cd0); |
| 104 |
cd4 = fma(z4, cd6, cd4); |
| 105 |
cd0 = fma(z8, cd4, cd0); |
| 106 |
let r = cn0 / cd0; |
| 107 |
if !gt { |
| 108 |
return r as f32; |
| 109 |
} /* for |x| < 1, (float) r is correctly rounded */ |
| 110 |
|
| 111 |
const PI_OVER2_H: f64 = f64::from_bits(0x3ff9000000000000); |
| 112 |
const PI_OVER2_L: f64 = f64::from_bits(0x3f80fdaa22168c23); |
| 113 |
/* now r approximates atan(1/x), we use atan(x) + atan(1/x) = sign(x)*pi/2, |
| 114 |
where PI_OVER2_H + PI_OVER2_L approximates pi/2. |
| 115 |
With sign(z)*L + (-r + sign(z)*H), it fails for x=0x1.98c252p+12 and |
| 116 |
rounding upward. |
| 117 |
With sign(z)*PI - r, where PI is a double approximation of pi to nearest, |
| 118 |
it fails for x=0x1.ddf9f6p+0 and rounding upward. */ |
| 119 |
((f64::copysign(PI_OVER2_L, z) - r) + f64::copysign(PI_OVER2_H, z)) as f32 |
| 120 |
} |
| 121 |
|
| 122 |
#[cfg(any(target_arch = "x86", target_arch = "x86_64"))] |
| 123 |
#[target_feature(enable = "avx", enable = "fma")] |
| 124 |
unsafe fn atanf_fma_impl(x: f32) -> f32 { |
| 125 |
atanf_gen_impl(x, f64::mul_add) |
| 126 |
} |
| 127 |
|
| 128 |
/// Computes atan |
| 129 |
/// |
| 130 |
/// Max found ULP 0.49999973 |
| 131 |
#[inline] |
| 132 |
pub fn f_atanf(x: f32) -> f32 { |
| 133 |
#[cfg(not(any(target_arch = "x86", target_arch = "x86_64")))] |
| 134 |
{ |
| 135 |
atanf_gen_impl(x, f_fmla) |
| 136 |
} |
| 137 |
#[cfg(any(target_arch = "x86", target_arch = "x86_64"))] |
| 138 |
{ |
| 139 |
use std::sync::OnceLock; |
| 140 |
static EXECUTOR: OnceLock<unsafe fn(f32) -> f32> = OnceLock::new(); |
| 141 |
let q = EXECUTOR.get_or_init(|| { |
| 142 |
if std::arch::is_x86_feature_detected!("avx") |
| 143 |
&& std::arch::is_x86_feature_detected!("fma") |
| 144 |
{ |
| 145 |
atanf_fma_impl |
| 146 |
} else { |
| 147 |
fn def_atanf(x: f32) -> f32 { |
| 148 |
atanf_gen_impl(x, f_fmla) |
| 149 |
} |
| 150 |
def_atanf |
| 151 |
} |
| 152 |
}); |
| 153 |
unsafe { q(x) } |
| 154 |
} |
| 155 |
} |
| 156 |
|
| 157 |
#[cfg(test)] |
| 158 |
mod tests { |
| 159 |
use super::*; |
| 160 |
|
| 161 |
#[test] |
| 162 |
fn f_atan_test() { |
| 163 |
assert!( |
| 164 |
(f_atanf(1.0) - std::f32::consts::PI / 4f32).abs() < 1e-6, |
| 165 |
"Invalid result {}", |
| 166 |
f_atanf(1f32) |
| 167 |
); |
| 168 |
assert!( |
| 169 |
(f_atanf(2f32) - 1.107148717794090503017065f32).abs() < 1e-6, |
| 170 |
"Invalid result {}", |
| 171 |
f_atanf(2f32) |
| 172 |
); |
| 173 |
assert!( |
| 174 |
(f_atanf(5f32) - 1.3734007669450158608612719264f32).abs() < 1e-6, |
| 175 |
"Invalid result {}", |
| 176 |
f_atanf(5f32) |
| 177 |
); |
| 178 |
} |
| 179 |
} |