| 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::EXP_MASK_F32; |
| 30 |
|
| 31 |
/// Hypot function |
| 32 |
/// |
| 33 |
/// Max ULP 0.5 |
| 34 |
#[inline] |
| 35 |
pub fn f_hypotf(x: f32, y: f32) -> f32 { |
| 36 |
let x_abs = f32::from_bits(x.to_bits() & 0x7fff_ffffu32); |
| 37 |
let y_abs = f32::from_bits(y.to_bits() & 0x7fff_ffffu32); |
| 38 |
|
| 39 |
let a_bits = x_abs.to_bits().max(y_abs.to_bits()); |
| 40 |
let b_bits = x_abs.to_bits().min(y_abs.to_bits()); |
| 41 |
|
| 42 |
let a_u = a_bits; |
| 43 |
let b_u = b_bits; |
| 44 |
|
| 45 |
if a_u >= EXP_MASK_F32 { |
| 46 |
// x or y is inf or nan |
| 47 |
if f32::from_bits(a_bits).is_nan() || f32::from_bits(b_bits).is_nan() { |
| 48 |
return f32::NAN; |
| 49 |
} |
| 50 |
if f32::from_bits(a_bits).is_infinite() || f32::from_bits(b_bits).is_infinite() { |
| 51 |
return f32::INFINITY; |
| 52 |
} |
| 53 |
return f32::from_bits(a_bits); |
| 54 |
} |
| 55 |
|
| 56 |
if a_u.wrapping_sub(b_u) >= ((23u32 + 2) << 23) { |
| 57 |
return x_abs + y_abs; |
| 58 |
} |
| 59 |
|
| 60 |
#[cfg(any( |
| 61 |
all( |
| 62 |
any(target_arch = "x86", target_arch = "x86_64"), |
| 63 |
target_feature = "fma" |
| 64 |
), |
| 65 |
target_arch = "aarch64" |
| 66 |
))] |
| 67 |
{ |
| 68 |
let ad = x as f64; |
| 69 |
let bd = y as f64; |
| 70 |
use crate::common::f_fmla; |
| 71 |
// for FMA environment we're using Kahan style summation which is short and reliable. |
| 72 |
let w = bd * bd; // RN(bc) |
| 73 |
let e = f_fmla(-bd, bd, w); // RN(w − bc) |
| 74 |
let f = f_fmla(ad, ad, w); // RN(ad + w) |
| 75 |
let r = e + f; // RN(f + e) |
| 76 |
let hyp = r.sqrt(); // sqrt(x^2 + y^2) |
| 77 |
hyp as f32 |
| 78 |
} |
| 79 |
#[cfg(not(any( |
| 80 |
all( |
| 81 |
any(target_arch = "x86", target_arch = "x86_64"), |
| 82 |
target_feature = "fma" |
| 83 |
), |
| 84 |
target_arch = "aarch64" |
| 85 |
)))] |
| 86 |
{ |
| 87 |
let ad = f32::from_bits(a_bits) as f64; |
| 88 |
let bd = f32::from_bits(b_bits) as f64; |
| 89 |
use crate::double_double::DoubleDouble; |
| 90 |
let dy2 = DoubleDouble::from_exact_mult(bd, bd); |
| 91 |
let fdx = DoubleDouble::from_exact_mult(ad, ad); |
| 92 |
// elements are always sorted thus fdx.hi > dy2.hi, thus fasttwosum requirements is fulfilled |
| 93 |
let f = DoubleDouble::add_f64(fdx, dy2.hi).to_f64(); |
| 94 |
let r = dy2.lo + f; |
| 95 |
let cath = r.sqrt(); |
| 96 |
cath as f32 |
| 97 |
} |
| 98 |
} |
| 99 |
|
| 100 |
#[cfg(test)] |
| 101 |
mod tests { |
| 102 |
use super::*; |
| 103 |
|
| 104 |
#[test] |
| 105 |
fn test_hypotf() { |
| 106 |
assert_eq!( |
| 107 |
f_hypotf( |
| 108 |
0.000000000000000000000000000000000000000091771, |
| 109 |
0.000000000000000000000000000000000000011754585 |
| 110 |
), |
| 111 |
0.000000000000000000000000000000000000011754944 |
| 112 |
); |
| 113 |
assert_eq!( |
| 114 |
f_hypotf(9.177e-41, 1.1754585e-38), |
| 115 |
0.000000000000000000000000000000000000011754944 |
| 116 |
); |
| 117 |
let dx = (f_hypotf(1f32, 1f32) - (1f32 * 1f32 + 1f32 * 1f32).sqrt()).abs(); |
| 118 |
assert!(dx < 1e-5); |
| 119 |
let dx = (f_hypotf(5f32, 5f32) - (5f32 * 5f32 + 5f32 * 5f32).sqrt()).abs(); |
| 120 |
assert!(dx < 1e-5); |
| 121 |
} |
| 122 |
|
| 123 |
#[test] |
| 124 |
fn test_hypotf_edge_cases() { |
| 125 |
assert_eq!(f_hypotf(-1.0, -3.0), 3.1622777); |
| 126 |
assert_eq!(f_hypotf(0.0, 0.0), 0.0); |
| 127 |
assert_eq!(f_hypotf(f32::INFINITY, 0.0), f32::INFINITY); |
| 128 |
assert_eq!(f_hypotf(0.0, f32::INFINITY), f32::INFINITY); |
| 129 |
assert_eq!(f_hypotf(f32::INFINITY, f32::INFINITY), f32::INFINITY); |
| 130 |
assert_eq!(f_hypotf(f32::NEG_INFINITY, 0.0), f32::INFINITY); |
| 131 |
assert_eq!(f_hypotf(0.0, f32::NEG_INFINITY), f32::INFINITY); |
| 132 |
assert_eq!( |
| 133 |
f_hypotf(f32::NEG_INFINITY, f32::NEG_INFINITY), |
| 134 |
f32::INFINITY |
| 135 |
); |
| 136 |
assert!(f_hypotf(f32::NAN, 1.0).is_nan()); |
| 137 |
assert!(f_hypotf(1.0, f32::NAN).is_nan()); |
| 138 |
} |
| 139 |
} |