| 1 |
/* |
| 2 |
* // Copyright (c) Radzivon Bartoshyk 8/2025. 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::polyeval::f_polyeval5; |
| 31 |
use crate::sin_cosf::sincosf_eval::sincosf_eval; |
| 32 |
|
| 33 |
#[inline(always)] |
| 34 |
fn sinf_gen_impl(x: f32) -> f32 { |
| 35 |
let x_abs = x.to_bits() & 0x7fff_ffffu32; |
| 36 |
let xd = x as f64; |
| 37 |
|
| 38 |
// |x| <= pi/16 |
| 39 |
if x_abs <= 0x3e49_0fdbu32 { |
| 40 |
// |x| < 0.000443633 |
| 41 |
if x_abs < 0x39e8_9769u32 { |
| 42 |
if x_abs == 0u32 { |
| 43 |
// For signed zeros. |
| 44 |
return x; |
| 45 |
} |
| 46 |
#[cfg(any( |
| 47 |
all( |
| 48 |
any(target_arch = "x86", target_arch = "x86_64"), |
| 49 |
target_feature = "fma" |
| 50 |
), |
| 51 |
target_arch = "aarch64" |
| 52 |
))] |
| 53 |
{ |
| 54 |
use crate::common::f_fmlaf; |
| 55 |
return f_fmlaf(x, f32::from_bits(0xb3000000), x); |
| 56 |
} |
| 57 |
#[cfg(not(any( |
| 58 |
all( |
| 59 |
any(target_arch = "x86", target_arch = "x86_64"), |
| 60 |
target_feature = "fma" |
| 61 |
), |
| 62 |
target_arch = "aarch64" |
| 63 |
)))] |
| 64 |
{ |
| 65 |
return f_fmla(xd, f64::from_bits(0xbe60000000000000), xd) as f32; |
| 66 |
} |
| 67 |
} |
| 68 |
|
| 69 |
let xsqr = xd * xd; |
| 70 |
|
| 71 |
/* |
| 72 |
Generated by Sollya: |
| 73 |
f_sinpi_16 = sin(x)/x; |
| 74 |
Q = fpminimax(f_sinpi_16, [|0, 2, 4, 6, 8|], [|1, D...|], [0, pi/16]); |
| 75 |
|
| 76 |
See ./notes/sinf.sollya |
| 77 |
*/ |
| 78 |
let p = f_polyeval5( |
| 79 |
xsqr, |
| 80 |
f64::from_bits(0x3ff0000000000000), |
| 81 |
f64::from_bits(0xbfc55555555554c6), |
| 82 |
f64::from_bits(0x3f81111111085e65), |
| 83 |
f64::from_bits(0xbf2a019f70fb4d4f), |
| 84 |
f64::from_bits(0x3ec718d179815e74), |
| 85 |
); |
| 86 |
return (xd * p) as f32; |
| 87 |
} |
| 88 |
|
| 89 |
if x_abs >= 0x7f80_0000u32 { |
| 90 |
return x + f32::NAN; |
| 91 |
} |
| 92 |
|
| 93 |
// Formula: |
| 94 |
// sin(x) = sin((k + y)*pi/32) |
| 95 |
// = sin(y*pi/32) * cos(k*pi/32) + cos(y*pi/32) * sin(k*pi/32) |
| 96 |
// The values of sin(k*pi/32) and cos(k*pi/32) for k = 0..31 are precomputed |
| 97 |
// and stored using a vector of 32 doubles. Sin(y*pi/32) and cos(y*pi/32) are |
| 98 |
// computed using degree-7 and degree-6 minimax polynomials generated by |
| 99 |
// Sollya respectively. |
| 100 |
|
| 101 |
let rs = sincosf_eval(xd, x_abs); |
| 102 |
f_fmla(rs.sin_y, rs.cos_k, f_fmla(rs.cosm1_y, rs.sin_k, rs.sin_k)) as f32 |
| 103 |
} |
| 104 |
|
| 105 |
#[cfg(any(target_arch = "x86", target_arch = "x86_64"))] |
| 106 |
#[target_feature(enable = "avx", enable = "fma")] |
| 107 |
unsafe fn sinf_fma_impl(x: f32) -> f32 { |
| 108 |
let x_abs = x.to_bits() & 0x7fff_ffffu32; |
| 109 |
let xd = x as f64; |
| 110 |
|
| 111 |
// |x| <= pi/16 |
| 112 |
if x_abs <= 0x3e49_0fdbu32 { |
| 113 |
// |x| < 0.000443633 |
| 114 |
if x_abs < 0x39e8_9769u32 { |
| 115 |
if x_abs == 0u32 { |
| 116 |
// For signed zeros. |
| 117 |
return x; |
| 118 |
} |
| 119 |
return f32::mul_add(x, f32::from_bits(0xb3000000), x); |
| 120 |
} |
| 121 |
|
| 122 |
let xsqr = xd * xd; |
| 123 |
|
| 124 |
/* |
| 125 |
Generated by Sollya: |
| 126 |
f_sinpi_16 = sin(x)/x; |
| 127 |
Q = fpminimax(f_sinpi_16, [|0, 2, 4, 6, 8|], [|1, D...|], [0, pi/16]); |
| 128 |
|
| 129 |
See ./notes/sinf.sollya |
| 130 |
*/ |
| 131 |
use crate::polyeval::d_polyeval5; |
| 132 |
let p = d_polyeval5( |
| 133 |
xsqr, |
| 134 |
f64::from_bits(0x3ff0000000000000), |
| 135 |
f64::from_bits(0xbfc55555555554c6), |
| 136 |
f64::from_bits(0x3f81111111085e65), |
| 137 |
f64::from_bits(0xbf2a019f70fb4d4f), |
| 138 |
f64::from_bits(0x3ec718d179815e74), |
| 139 |
); |
| 140 |
return (xd * p) as f32; |
| 141 |
} |
| 142 |
|
| 143 |
if x_abs >= 0x7f80_0000u32 { |
| 144 |
return x + f32::NAN; |
| 145 |
} |
| 146 |
|
| 147 |
// Formula: |
| 148 |
// sin(x) = sin((k + y)*pi/32) |
| 149 |
// = sin(y*pi/32) * cos(k*pi/32) + cos(y*pi/32) * sin(k*pi/32) |
| 150 |
// The values of sin(k*pi/32) and cos(k*pi/32) for k = 0..31 are precomputed |
| 151 |
// and stored using a vector of 32 doubles. Sin(y*pi/32) and cos(y*pi/32) are |
| 152 |
// computed using degree-7 and degree-6 minimax polynomials generated by |
| 153 |
// Sollya respectively. |
| 154 |
use crate::sin_cosf::sincosf_eval::sincosf_eval_fma; |
| 155 |
let rs = sincosf_eval_fma(xd, x_abs); |
| 156 |
f64::mul_add( |
| 157 |
rs.sin_y, |
| 158 |
rs.cos_k, |
| 159 |
f64::mul_add(rs.cosm1_y, rs.sin_k, rs.sin_k), |
| 160 |
) as f32 |
| 161 |
} |
| 162 |
|
| 163 |
/// Sine function |
| 164 |
/// |
| 165 |
/// Max found ULP 0.5 |
| 166 |
#[inline] |
| 167 |
pub fn f_sinf(x: f32) -> f32 { |
| 168 |
#[cfg(not(any(target_arch = "x86", target_arch = "x86_64")))] |
| 169 |
{ |
| 170 |
sinf_gen_impl(x) |
| 171 |
} |
| 172 |
#[cfg(any(target_arch = "x86", target_arch = "x86_64"))] |
| 173 |
{ |
| 174 |
use std::sync::OnceLock; |
| 175 |
static EXECUTOR: OnceLock<unsafe fn(f32) -> f32> = OnceLock::new(); |
| 176 |
let q = EXECUTOR.get_or_init(|| { |
| 177 |
if std::arch::is_x86_feature_detected!("avx") |
| 178 |
&& std::arch::is_x86_feature_detected!("fma") |
| 179 |
{ |
| 180 |
sinf_fma_impl |
| 181 |
} else { |
| 182 |
sinf_gen_impl |
| 183 |
} |
| 184 |
}); |
| 185 |
unsafe { q(x) } |
| 186 |
} |
| 187 |
} |
| 188 |
|
| 189 |
#[cfg(test)] |
| 190 |
mod tests { |
| 191 |
use super::*; |
| 192 |
|
| 193 |
#[test] |
| 194 |
fn f_sinf_test() { |
| 195 |
assert_eq!(f_sinf(0.0), 0.0); |
| 196 |
assert_eq!(f_sinf(1.0), 0.84147096); |
| 197 |
assert_eq!(f_sinf(0.3), 0.29552022); |
| 198 |
assert_eq!(f_sinf(-1.0), -0.84147096); |
| 199 |
assert_eq!(f_sinf(-0.3), -0.29552022); |
| 200 |
assert_eq!(f_sinf(std::f32::consts::PI / 2.), 1.); |
| 201 |
assert!(f_sinf(f32::INFINITY).is_nan()); |
| 202 |
assert!(f_sinf(f32::NEG_INFINITY).is_nan()); |
| 203 |
assert!((f_sinf(std::f32::consts::PI) - 0f32).abs() < 1e-6); |
| 204 |
assert!((f_sinf(std::f32::consts::FRAC_PI_2) - 1f32).abs() < 1e-6); |
| 205 |
} |
| 206 |
} |