| 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; |
| 30 |
use crate::cube_roots::cbrt::{CbrtBackend, GenericCbrtBackend}; |
| 31 |
|
| 32 |
#[inline(always)] |
| 33 |
pub(crate) fn halley_refine_d(x: f64, a: f64) -> f64 { |
| 34 |
let tx = x * x * x; |
| 35 |
x * f_fmla(2., a, tx) / f_fmla(2., tx, a) |
| 36 |
} |
| 37 |
|
| 38 |
#[inline(always)] |
| 39 |
#[allow(unused)] |
| 40 |
pub(crate) fn halley_refine_d_fma(x: f64, a: f64) -> f64 { |
| 41 |
let tx = x * x * x; |
| 42 |
x * f64::mul_add(2., a, tx) / f64::mul_add(2., tx, a) |
| 43 |
} |
| 44 |
|
| 45 |
#[inline(always)] |
| 46 |
const fn halley_refine(x: f32, a: f32) -> f32 { |
| 47 |
let tx = x * x * x; |
| 48 |
x * (tx + 2f32 * a) / (2f32 * tx + a) |
| 49 |
} |
| 50 |
|
| 51 |
/// Cbrt for given value for const context. |
| 52 |
/// This is simplified version just to make a good approximation on const context. |
| 53 |
#[inline] |
| 54 |
pub const fn cbrtf(x: f32) -> f32 { |
| 55 |
let u = x.to_bits(); |
| 56 |
let au = u.wrapping_shl(1); |
| 57 |
if au < (1u32 << 24) || au >= (0xffu32 << 24) { |
| 58 |
if au >= (0xffu32 << 24) { |
| 59 |
return x + x; /* inf, nan */ |
| 60 |
} |
| 61 |
if au == 0 { |
| 62 |
return x; /* +-0 */ |
| 63 |
} |
| 64 |
} |
| 65 |
|
| 66 |
const B1: u32 = 709958130; |
| 67 |
let mut t: f32; |
| 68 |
let mut ui: u32 = x.to_bits(); |
| 69 |
let mut hx: u32 = ui & 0x7fffffff; |
| 70 |
|
| 71 |
hx = (hx / 3).wrapping_add(B1); |
| 72 |
ui &= 0x80000000; |
| 73 |
ui |= hx; |
| 74 |
|
| 75 |
t = f32::from_bits(ui); |
| 76 |
t = halley_refine(t, x); |
| 77 |
halley_refine(t, x) |
| 78 |
} |
| 79 |
|
| 80 |
#[inline(always)] |
| 81 |
fn cbrtf_gen_impl<B: CbrtBackend>(x: f32, backend: B) -> f32 { |
| 82 |
let u = x.to_bits(); |
| 83 |
let au = u.wrapping_shl(1); |
| 84 |
if au < (1u32 << 24) || au >= (0xffu32 << 24) { |
| 85 |
if au >= (0xffu32 << 24) { |
| 86 |
return x + x; /* inf, nan */ |
| 87 |
} |
| 88 |
if au == 0 { |
| 89 |
return x; /* +-0 */ |
| 90 |
} |
| 91 |
} |
| 92 |
|
| 93 |
let mut ui: u32 = x.to_bits(); |
| 94 |
let mut hx: u32 = ui & 0x7fffffff; |
| 95 |
|
| 96 |
if hx < 0x00800000 { |
| 97 |
/* zero or subnormal? */ |
| 98 |
if hx == 0 { |
| 99 |
return x; /* cbrt(+-0) is itself */ |
| 100 |
} |
| 101 |
const TWO_EXP_24: f32 = f32::from_bits(0x4b800000); |
| 102 |
ui = (x * TWO_EXP_24).to_bits(); |
| 103 |
hx = ui & 0x7fffffff; |
| 104 |
const B2: u32 = 642849266; |
| 105 |
hx = (hx / 3).wrapping_add(B2); |
| 106 |
} else { |
| 107 |
const B1: u32 = 709958130; |
| 108 |
hx = (hx / 3).wrapping_add(B1); |
| 109 |
} |
| 110 |
ui &= 0x80000000; |
| 111 |
ui |= hx; |
| 112 |
|
| 113 |
let mut t = f32::from_bits(ui) as f64; |
| 114 |
let dx = x as f64; |
| 115 |
t = backend.halley(t, dx); |
| 116 |
backend.halley(t, dx) as f32 |
| 117 |
} |
| 118 |
|
| 119 |
#[cfg(any(target_arch = "x86", target_arch = "x86_64"))] |
| 120 |
#[target_feature(enable = "avx", enable = "fma")] |
| 121 |
unsafe fn cbrtf_fma_impl(x: f32) -> f32 { |
| 122 |
use crate::cube_roots::cbrt::FmaCbrtBackend; |
| 123 |
cbrtf_gen_impl(x, FmaCbrtBackend {}) |
| 124 |
} |
| 125 |
|
| 126 |
/// Computes cube root |
| 127 |
/// |
| 128 |
/// Peak ULP on 64 bit = 0.49999577 |
| 129 |
#[inline] |
| 130 |
pub fn f_cbrtf(x: f32) -> f32 { |
| 131 |
#[cfg(not(any(target_arch = "x86", target_arch = "x86_64")))] |
| 132 |
{ |
| 133 |
cbrtf_gen_impl(x, GenericCbrtBackend {}) |
| 134 |
} |
| 135 |
#[cfg(any(target_arch = "x86", target_arch = "x86_64"))] |
| 136 |
{ |
| 137 |
use std::sync::OnceLock; |
| 138 |
static EXECUTOR: OnceLock<unsafe fn(f32) -> f32> = OnceLock::new(); |
| 139 |
let q = EXECUTOR.get_or_init(|| { |
| 140 |
if std::arch::is_x86_feature_detected!("avx") |
| 141 |
&& std::arch::is_x86_feature_detected!("fma") |
| 142 |
{ |
| 143 |
cbrtf_fma_impl |
| 144 |
} else { |
| 145 |
fn def_cbrt(x: f32) -> f32 { |
| 146 |
cbrtf_gen_impl(x, GenericCbrtBackend {}) |
| 147 |
} |
| 148 |
def_cbrt |
| 149 |
} |
| 150 |
}); |
| 151 |
unsafe { q(x) } |
| 152 |
} |
| 153 |
} |
| 154 |
|
| 155 |
#[cfg(test)] |
| 156 |
mod tests { |
| 157 |
use super::*; |
| 158 |
|
| 159 |
#[test] |
| 160 |
fn test_fcbrtf() { |
| 161 |
assert_eq!(f_cbrtf(0.0), 0.0); |
| 162 |
assert_eq!(f_cbrtf(-27.0), -3.0); |
| 163 |
assert_eq!(f_cbrtf(27.0), 3.0); |
| 164 |
assert_eq!(f_cbrtf(64.0), 4.0); |
| 165 |
assert_eq!(f_cbrtf(-64.0), -4.0); |
| 166 |
assert_eq!(f_cbrtf(f32::NEG_INFINITY), f32::NEG_INFINITY); |
| 167 |
assert_eq!(f_cbrtf(f32::INFINITY), f32::INFINITY); |
| 168 |
assert!(f_cbrtf(f32::NAN).is_nan()); |
| 169 |
} |
| 170 |
} |