| 1 |
/* |
| 2 |
* // Copyright (c) Radzivon Bartoshyk 6/2025. 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 |
|
| 31 |
static OFF: [f32; 8] = [0.0, 0.5, 1.0, 0.5, -0.0, -0.5, -1.0, -0.5]; |
| 32 |
static SGNF: [f32; 2] = [1., -1.]; |
| 33 |
static SGN: [f64; 2] = [1., -1.]; |
| 34 |
|
| 35 |
#[inline(always)] |
| 36 |
fn atan2pif_gen_impl<Q: Fn(f64, f64, f64) -> f64>(y: f32, x: f32, fma: Q) -> f32 { |
| 37 |
let tx = x.to_bits(); |
| 38 |
let ty: u32 = y.to_bits(); |
| 39 |
let ux: u32 = tx; |
| 40 |
let uy: u32 = ty; |
| 41 |
let ax: u32 = ux & 0x7fff_ffff; |
| 42 |
let ay = uy & 0x7fff_ffff; |
| 43 |
if ay >= (0xff << 23) || ax >= (0xff << 23) { |
| 44 |
if ay > (0xff << 23) { |
| 45 |
return x + y; |
| 46 |
} // nan |
| 47 |
if ax > (0xff << 23) { |
| 48 |
return x + y; |
| 49 |
} // nan |
| 50 |
let yinf = ay == (0xff << 23); |
| 51 |
let xinf = ax == (0xff << 23); |
| 52 |
if yinf & xinf { |
| 53 |
return if (ux >> 31) != 0 { |
| 54 |
0.75 * SGNF[(uy >> 31) as usize] |
| 55 |
} else { |
| 56 |
0.25 * SGNF[(uy >> 31) as usize] |
| 57 |
}; |
| 58 |
} |
| 59 |
if xinf { |
| 60 |
return if (ux >> 31) != 0 { |
| 61 |
SGNF[(uy >> 31) as usize] |
| 62 |
} else { |
| 63 |
0.0 * SGNF[(uy >> 31) as usize] |
| 64 |
}; |
| 65 |
} |
| 66 |
if yinf { |
| 67 |
return 0.5 * SGNF[(uy >> 31) as usize]; |
| 68 |
} |
| 69 |
} |
| 70 |
if ay == 0 { |
| 71 |
if (ay | ax) == 0 { |
| 72 |
let i: u32 = (uy >> 31) * 4 + (ux >> 31) * 2; |
| 73 |
return OFF[i as usize]; |
| 74 |
} |
| 75 |
if (ux >> 31) == 0 { |
| 76 |
return 0.0 * SGNF[(uy >> 31) as usize]; |
| 77 |
} |
| 78 |
} |
| 79 |
if ax == ay { |
| 80 |
static S: [f32; 4] = [0.25, 0.75, -0.25, -0.75]; |
| 81 |
let i = (uy >> 31) * 2 + (ux >> 31); |
| 82 |
return S[i as usize]; |
| 83 |
} |
| 84 |
let gt: usize = (ay > ax) as usize; |
| 85 |
let i: u32 = (uy >> 31) * 4 + (ux >> 31) * 2 + gt as u32; |
| 86 |
|
| 87 |
let zx = x as f64; |
| 88 |
let zy = y as f64; |
| 89 |
static M: [f64; 2] = [0., 1.]; |
| 90 |
|
| 91 |
let mut z = fma(M[gt], zx, M[1 - gt] * zy) / fma(M[gt], zy, M[1 - gt] * zx); |
| 92 |
|
| 93 |
const CN: [u64; 7] = [ |
| 94 |
0x3fd45f306dc9c883, |
| 95 |
0x3fe988d83a142ada, |
| 96 |
0x3fe747bebf492057, |
| 97 |
0x3fd2cc5645094ff3, |
| 98 |
0x3faa0521c711ab66, |
| 99 |
0x3f6881b8058b9a0d, |
| 100 |
0x3efb16ff514a0af0, |
| 101 |
]; |
| 102 |
|
| 103 |
let mut r = f64::from_bits(CN[0]); |
| 104 |
let z2 = z * z; |
| 105 |
z *= SGN[gt]; |
| 106 |
// avoid spurious underflow in the polynomial evaluation excluding tiny arguments |
| 107 |
if z2 > f64::from_bits(0x3c90000000000000) { |
| 108 |
let z4 = z2 * z2; |
| 109 |
let z8 = z4 * z4; |
| 110 |
let mut cn0 = fma(z2, f64::from_bits(CN[1]), r); |
| 111 |
let cn2 = fma(z2, f64::from_bits(CN[3]), f64::from_bits(CN[2])); |
| 112 |
let mut cn4 = fma(z2, f64::from_bits(CN[5]), f64::from_bits(CN[4])); |
| 113 |
let cn6 = f64::from_bits(CN[6]); |
| 114 |
cn0 += z4 * cn2; |
| 115 |
cn4 += z4 * cn6; |
| 116 |
cn0 += z8 * cn4; |
| 117 |
|
| 118 |
const CD: [u64; 7] = [ |
| 119 |
0x3ff0000000000000, |
| 120 |
0x4006b8b143a3f6da, |
| 121 |
0x4008421201d18ed5, |
| 122 |
0x3ff8221d086914eb, |
| 123 |
0x3fd670657e3a07ba, |
| 124 |
0x3fa0f4951fd1e72d, |
| 125 |
0x3f4b3874b8798286, |
| 126 |
]; |
| 127 |
|
| 128 |
let mut cd0 = fma(z2, f64::from_bits(CD[1]), f64::from_bits(CD[0])); |
| 129 |
let cd2 = fma(z2, f64::from_bits(CD[3]), f64::from_bits(CD[2])); |
| 130 |
let mut cd4 = fma(z2, f64::from_bits(CD[5]), f64::from_bits(CD[4])); |
| 131 |
let cd6 = f64::from_bits(CD[6]); |
| 132 |
cd0 = fma(z4, cd2, cd0); |
| 133 |
cd4 = fma(z4, cd6, cd4); |
| 134 |
cd0 = fma(z8, cd4, cd0); |
| 135 |
|
| 136 |
r = cn0 / cd0; |
| 137 |
} |
| 138 |
fma(z, r, OFF[i as usize] as f64) as f32 |
| 139 |
} |
| 140 |
|
| 141 |
#[cfg(any(target_arch = "x86", target_arch = "x86_64"))] |
| 142 |
#[target_feature(enable = "avx", enable = "fma")] |
| 143 |
unsafe fn atan2pif_fma_impl(y: f32, x: f32) -> f32 { |
| 144 |
atan2pif_gen_impl(y, x, f64::mul_add) |
| 145 |
} |
| 146 |
|
| 147 |
/// Computes atan(x/y) / PI |
| 148 |
/// |
| 149 |
/// Max found ULP 0.5 |
| 150 |
#[inline] |
| 151 |
pub fn f_atan2pif(y: f32, x: f32) -> f32 { |
| 152 |
#[cfg(not(any(target_arch = "x86", target_arch = "x86_64")))] |
| 153 |
{ |
| 154 |
atan2pif_gen_impl(y, x, f_fmla) |
| 155 |
} |
| 156 |
#[cfg(any(target_arch = "x86", target_arch = "x86_64"))] |
| 157 |
{ |
| 158 |
use std::sync::OnceLock; |
| 159 |
static EXECUTOR: OnceLock<unsafe fn(f32, f32) -> f32> = OnceLock::new(); |
| 160 |
let q = EXECUTOR.get_or_init(|| { |
| 161 |
if std::arch::is_x86_feature_detected!("avx") |
| 162 |
&& std::arch::is_x86_feature_detected!("fma") |
| 163 |
{ |
| 164 |
atan2pif_fma_impl |
| 165 |
} else { |
| 166 |
fn def_atan2pif(y: f32, x: f32) -> f32 { |
| 167 |
atan2pif_gen_impl(y, x, f_fmla) |
| 168 |
} |
| 169 |
def_atan2pif |
| 170 |
} |
| 171 |
}); |
| 172 |
unsafe { q(y, x) } |
| 173 |
} |
| 174 |
} |
| 175 |
|
| 176 |
#[cfg(test)] |
| 177 |
mod tests { |
| 178 |
use super::*; |
| 179 |
#[test] |
| 180 |
fn test_atan2pif() { |
| 181 |
assert_eq!(f_atan2pif(0.32131, 0.987565), 0.10012555); |
| 182 |
assert_eq!(f_atan2pif(532.32131, 12.987565), 0.49223542); |
| 183 |
assert_eq!(f_atan2pif(-754.32131, 12.987565), -0.494520042); |
| 184 |
} |
| 185 |
} |