| 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::acosf::poly12; |
| 30 |
use crate::common::{dd_fmlaf, f_fmla}; |
| 31 |
|
| 32 |
#[cold] |
| 33 |
fn as_special(x: f32) -> f32 { |
| 34 |
let t = x.to_bits(); |
| 35 |
let ax = t.wrapping_shl(1); |
| 36 |
if ax > (0xffu32 << 24) { |
| 37 |
return x + x; |
| 38 |
} // nan |
| 39 |
f32::NAN |
| 40 |
} |
| 41 |
|
| 42 |
#[inline(always)] |
| 43 |
/// fma - fma |
| 44 |
/// dd_fma - mandatory fma fallback |
| 45 |
fn asinf_gen_impl<Q: Fn(f64, f64, f64) -> f64, F: Fn(f32, f32, f32) -> f32>( |
| 46 |
x: f32, |
| 47 |
fma: Q, |
| 48 |
dd_fma: F, |
| 49 |
) -> f32 { |
| 50 |
const PI2: f64 = f64::from_bits(0x3ff921fb54442d18); |
| 51 |
let xs = x as f64; |
| 52 |
let mut r; |
| 53 |
let t = x.to_bits(); |
| 54 |
let ax = t.wrapping_shl(1); |
| 55 |
if ax > 0x7f << 24 { |
| 56 |
return as_special(x); |
| 57 |
} |
| 58 |
if ax < 0x7ec29000u32 { |
| 59 |
// |x| < 1.49029 |
| 60 |
if ax < 115 << 24 { |
| 61 |
// |x| < 0.000244141 |
| 62 |
return dd_fma(x, f64::from_bits(0x3e60000000000000) as f32, x); |
| 63 |
} |
| 64 |
const B: [u64; 16] = [ |
| 65 |
0x3ff0000000000005, |
| 66 |
0x3fc55557aeca105d, |
| 67 |
0x3fb3314ec3db7d12, |
| 68 |
0x3fa775738a5a6f92, |
| 69 |
0x3f75d5f7ce1c8538, |
| 70 |
0x3fd605c6d58740f0, |
| 71 |
0xc005728b732d73c6, |
| 72 |
0x402f152170f151eb, |
| 73 |
0xc04f962ea3ca992e, |
| 74 |
0x40671971e17375a0, |
| 75 |
0xc07860512b4ba230, |
| 76 |
0x40826a3b8d4bdb14, |
| 77 |
0xc0836f2ea5698b51, |
| 78 |
0x407b3d722aebfa2e, |
| 79 |
0xc066cf89703b1289, |
| 80 |
0x4041518af6a65e2d, |
| 81 |
]; |
| 82 |
let z = xs; |
| 83 |
let z2 = z * z; |
| 84 |
let w0 = fma(z2, f64::from_bits(B[1]), f64::from_bits(B[0])); |
| 85 |
let w1 = fma(z2, f64::from_bits(B[3]), f64::from_bits(B[2])); |
| 86 |
let w2 = fma(z2, f64::from_bits(B[5]), f64::from_bits(B[4])); |
| 87 |
let w3 = fma(z2, f64::from_bits(B[7]), f64::from_bits(B[6])); |
| 88 |
let w4 = fma(z2, f64::from_bits(B[9]), f64::from_bits(B[8])); |
| 89 |
let w5 = fma(z2, f64::from_bits(B[11]), f64::from_bits(B[10])); |
| 90 |
let w6 = fma(z2, f64::from_bits(B[13]), f64::from_bits(B[12])); |
| 91 |
let w7 = fma(z2, f64::from_bits(B[15]), f64::from_bits(B[14])); |
| 92 |
|
| 93 |
let z4 = z2 * z2; |
| 94 |
let z8 = z4 * z4; |
| 95 |
let z16 = z8 * z8; |
| 96 |
|
| 97 |
r = z |
| 98 |
* ((fma(z4, w1, w0) + z8 * fma(z4, w3, w2)) |
| 99 |
+ z16 * (fma(z4, w5, w4) + z8 * fma(z4, w7, w6))); |
| 100 |
let ub = r; |
| 101 |
let lb = r - z * f64::from_bits(0x3e0efa8eb0000000); |
| 102 |
// Ziv's accuracy test |
| 103 |
if ub == lb { |
| 104 |
return ub as f32; |
| 105 |
} |
| 106 |
} |
| 107 |
if ax < (0x7eu32 << 24) { |
| 108 |
const C: [u64; 12] = [ |
| 109 |
0x3fc555555555529c, |
| 110 |
0x3fb333333337e0dd, |
| 111 |
0x3fa6db6db3b4465e, |
| 112 |
0x3f9f1c72e13ac306, |
| 113 |
0x3f96e89cebe06bc4, |
| 114 |
0x3f91c6dcf5289094, |
| 115 |
0x3f8c6dbbcc7c6315, |
| 116 |
0x3f88f8dc2615e996, |
| 117 |
0x3f7a5833b7bf15e8, |
| 118 |
0x3f943f44ace1665c, |
| 119 |
0xbf90fb17df881c73, |
| 120 |
0x3fa07520c026b2d6, |
| 121 |
]; |
| 122 |
let z = xs; |
| 123 |
let z2 = z * z; |
| 124 |
let c0 = poly12(z2, C, &fma); |
| 125 |
r = z + (z * z2) * c0; |
| 126 |
} else { |
| 127 |
if ax == 0x7e55688au32 { |
| 128 |
return f32::copysign(f64::from_bits(0x3fe75b8a20000000) as f32, x) |
| 129 |
+ f32::copysign(f64::from_bits(0x3e50000000000000) as f32, x); |
| 130 |
} |
| 131 |
if ax == 0x7e107434u32 { |
| 132 |
return f32::copysign(f64::from_bits(0x3fe1f4b640000000) as f32, x) |
| 133 |
+ f32::copysign(f64::from_bits(0x3e50000000000000) as f32, x); |
| 134 |
} |
| 135 |
let bx = xs.abs(); |
| 136 |
let z = 1.0 - bx; |
| 137 |
let s = z.sqrt(); |
| 138 |
const C: [u64; 12] = [ |
| 139 |
0x3ff6a09e667f3bcb, |
| 140 |
0x3fbe2b7dddff2db9, |
| 141 |
0x3f9b27247ab42dbc, |
| 142 |
0x3f802995cc4e0744, |
| 143 |
0x3f65ffb0276ec8ea, |
| 144 |
0x3f5033885a928dec, |
| 145 |
0x3f3911f2be23f8c7, |
| 146 |
0x3f24c3c55d2437fd, |
| 147 |
0x3f0af477e1d7b461, |
| 148 |
0x3f0abd6bdff67dcb, |
| 149 |
0xbef1717e86d0fa28, |
| 150 |
0x3ef6ff526de46023, |
| 151 |
]; |
| 152 |
r = PI2 - s * poly12(z, C, &fma); |
| 153 |
r = f64::copysign(r, xs); |
| 154 |
} |
| 155 |
r as f32 |
| 156 |
} |
| 157 |
|
| 158 |
#[cfg(any(target_arch = "x86", target_arch = "x86_64"))] |
| 159 |
#[target_feature(enable = "avx", enable = "fma")] |
| 160 |
unsafe fn asinf_fma_impl(x: f32) -> f32 { |
| 161 |
asinf_gen_impl(x, f64::mul_add, f32::mul_add) |
| 162 |
} |
| 163 |
|
| 164 |
/// Computes asin |
| 165 |
/// |
| 166 |
/// Max found ULP 0.49999928 |
| 167 |
#[inline] |
| 168 |
pub fn f_asinf(x: f32) -> f32 { |
| 169 |
#[cfg(not(any(target_arch = "x86", target_arch = "x86_64")))] |
| 170 |
{ |
| 171 |
asinf_gen_impl(x, f_fmla, dd_fmlaf) |
| 172 |
} |
| 173 |
#[cfg(any(target_arch = "x86", target_arch = "x86_64"))] |
| 174 |
{ |
| 175 |
use std::sync::OnceLock; |
| 176 |
static EXECUTOR: OnceLock<unsafe fn(f32) -> f32> = OnceLock::new(); |
| 177 |
let q = EXECUTOR.get_or_init(|| { |
| 178 |
if std::arch::is_x86_feature_detected!("avx") |
| 179 |
&& std::arch::is_x86_feature_detected!("fma") |
| 180 |
{ |
| 181 |
asinf_fma_impl |
| 182 |
} else { |
| 183 |
fn def_asinf(x: f32) -> f32 { |
| 184 |
asinf_gen_impl(x, f_fmla, dd_fmlaf) |
| 185 |
} |
| 186 |
def_asinf |
| 187 |
} |
| 188 |
}); |
| 189 |
unsafe { q(x) } |
| 190 |
} |
| 191 |
} |
| 192 |
|
| 193 |
#[cfg(test)] |
| 194 |
mod tests { |
| 195 |
use super::*; |
| 196 |
|
| 197 |
#[test] |
| 198 |
fn test_asinf() { |
| 199 |
assert_eq!(f_asinf(-0.5), -std::f32::consts::FRAC_PI_6); |
| 200 |
assert_eq!(f_asinf(0.5), std::f32::consts::FRAC_PI_6); |
| 201 |
assert!(f_asinf(7.).is_nan()); |
| 202 |
} |
| 203 |
} |