| 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 |
use crate::sin_cosf::ArgumentReducerPi; |
| 31 |
use crate::tangent::evalf::tanpif_eval; |
| 32 |
|
| 33 |
#[inline(always)] |
| 34 |
fn tanpif_gen_impl(x: f32) -> f32 { |
| 35 |
let ix = x.to_bits(); |
| 36 |
let e = ix & (0xff << 23); |
| 37 |
if e > (150 << 23) { |
| 38 |
// |x| > 2^23 |
| 39 |
if e == (0xff << 23) { |
| 40 |
// x = nan or inf |
| 41 |
if (ix.wrapping_shl(9)) == 0 { |
| 42 |
// x = inf |
| 43 |
return f32::NAN; |
| 44 |
} |
| 45 |
return x + x; // x = nan |
| 46 |
} |
| 47 |
return f32::copysign(0.0, x); |
| 48 |
} |
| 49 |
|
| 50 |
let argument_reduction = ArgumentReducerPi { x: x as f64 }; |
| 51 |
|
| 52 |
let (y, k) = argument_reduction.reduce(); |
| 53 |
|
| 54 |
if y == 0.0 { |
| 55 |
let km = (k.abs() & 31) as i32; // k mod 32 |
| 56 |
|
| 57 |
match km { |
| 58 |
0 => return 0.0f32.copysign(x), // tanpi(n) = 0 |
| 59 |
16 => return f32::copysign(f32::INFINITY, x), // tanpi(n+0.5) = ±∞ |
| 60 |
8 => return f32::copysign(1.0, x), // tanpi(n+0.25) = ±1 |
| 61 |
24 => return -f32::copysign(1.0, x), // tanpi(n+0.75) = ∓1 |
| 62 |
_ => {} |
| 63 |
} |
| 64 |
} |
| 65 |
|
| 66 |
let ax = ix & 0x7fff_ffff; |
| 67 |
if ax < 0x38d1b717u32 { |
| 68 |
// taylor series for tan(PI*x) where |x| < 0.0001 |
| 69 |
let dx = x as f64; |
| 70 |
let dx_sqr = dx * dx; |
| 71 |
// tan(PI*x) ~ PI*x + PI^3*x^3/3 + O(x^5) |
| 72 |
let r = f_fmla( |
| 73 |
dx_sqr, |
| 74 |
f64::from_bits(0x4024abbce625be53), |
| 75 |
f64::from_bits(0x400921fb54442d18), |
| 76 |
); |
| 77 |
return (r * dx) as f32; |
| 78 |
} |
| 79 |
|
| 80 |
// tanpif_eval returns: |
| 81 |
// - rs.tan_y = tan(pi/32 * y) -> tangent of the remainder |
| 82 |
// - rs.tan_k = tan(pi/32 * k) -> tan of the main angle multiple |
| 83 |
let rs = tanpif_eval(y, k); |
| 84 |
|
| 85 |
// Then computing tan through identities |
| 86 |
// num = tan(k*pi/32) + tan(y*pi/32) |
| 87 |
let num = rs.tan_y + rs.tan_k; |
| 88 |
// den = 1 - tan(k*pi/32) * tan(y*pi/32) |
| 89 |
let den = f_fmla(rs.tan_y, -rs.tan_k, 1.); |
| 90 |
(num / den) as f32 |
| 91 |
} |
| 92 |
|
| 93 |
#[cfg(any(target_arch = "x86", target_arch = "x86_64"))] |
| 94 |
#[target_feature(enable = "avx", enable = "fma")] |
| 95 |
unsafe fn tanpif_fma_impl(x: f32) -> f32 { |
| 96 |
let ix = x.to_bits(); |
| 97 |
let e = ix & (0xff << 23); |
| 98 |
if e > (150 << 23) { |
| 99 |
// |x| > 2^23 |
| 100 |
if e == (0xff << 23) { |
| 101 |
// x = nan or inf |
| 102 |
if (ix.wrapping_shl(9)) == 0 { |
| 103 |
// x = inf |
| 104 |
return f32::NAN; |
| 105 |
} |
| 106 |
return x + x; // x = nan |
| 107 |
} |
| 108 |
return f32::copysign(0.0, x); |
| 109 |
} |
| 110 |
|
| 111 |
let argument_reduction = ArgumentReducerPi { x: x as f64 }; |
| 112 |
|
| 113 |
let (y, k) = argument_reduction.reduce_fma(); |
| 114 |
|
| 115 |
if y == 0.0 { |
| 116 |
let km = (k.abs() & 31) as i32; // k mod 32 |
| 117 |
|
| 118 |
match km { |
| 119 |
0 => return 0.0f32.copysign(x), // tanpi(n) = 0 |
| 120 |
16 => return f32::copysign(f32::INFINITY, x), // tanpi(n+0.5) = ±∞ |
| 121 |
8 => return f32::copysign(1.0, x), // tanpi(n+0.25) = ±1 |
| 122 |
24 => return -f32::copysign(1.0, x), // tanpi(n+0.75) = ∓1 |
| 123 |
_ => {} |
| 124 |
} |
| 125 |
} |
| 126 |
|
| 127 |
let ax = ix & 0x7fff_ffff; |
| 128 |
if ax < 0x38d1b717u32 { |
| 129 |
// taylor series for tan(PI*x) where |x| < 0.0001 |
| 130 |
let dx = x as f64; |
| 131 |
let dx_sqr = dx * dx; |
| 132 |
// tan(PI*x) ~ PI*x + PI^3*x^3/3 + O(x^5) |
| 133 |
let r = f64::mul_add( |
| 134 |
dx_sqr, |
| 135 |
f64::from_bits(0x4024abbce625be53), |
| 136 |
f64::from_bits(0x400921fb54442d18), |
| 137 |
); |
| 138 |
return (r * dx) as f32; |
| 139 |
} |
| 140 |
|
| 141 |
// tanpif_eval returns: |
| 142 |
// - rs.tan_y = tan(pi/32 * y) -> tangent of the remainder |
| 143 |
// - rs.tan_k = tan(pi/32 * k) -> tan of the main angle multiple |
| 144 |
use crate::tangent::evalf::tanpif_eval_fma; |
| 145 |
let rs = tanpif_eval_fma(y, k); |
| 146 |
|
| 147 |
// Then computing tan through identities |
| 148 |
// num = tan(k*pi/32) + tan(y*pi/32) |
| 149 |
let num = rs.tan_y + rs.tan_k; |
| 150 |
// den = 1 - tan(k*pi/32) * tan(y*pi/32) |
| 151 |
let den = f64::mul_add(rs.tan_y, -rs.tan_k, 1.); |
| 152 |
(num / den) as f32 |
| 153 |
} |
| 154 |
|
| 155 |
/// Computes tan(PI*x) |
| 156 |
/// |
| 157 |
/// Max found ULP 0.5 |
| 158 |
#[inline] |
| 159 |
pub fn f_tanpif(x: f32) -> f32 { |
| 160 |
#[cfg(not(any(target_arch = "x86", target_arch = "x86_64")))] |
| 161 |
{ |
| 162 |
tanpif_gen_impl(x) |
| 163 |
} |
| 164 |
#[cfg(any(target_arch = "x86", target_arch = "x86_64"))] |
| 165 |
{ |
| 166 |
use std::sync::OnceLock; |
| 167 |
static EXECUTOR: OnceLock<unsafe fn(f32) -> f32> = OnceLock::new(); |
| 168 |
let q = EXECUTOR.get_or_init(|| { |
| 169 |
if std::arch::is_x86_feature_detected!("avx") |
| 170 |
&& std::arch::is_x86_feature_detected!("fma") |
| 171 |
{ |
| 172 |
tanpif_fma_impl |
| 173 |
} else { |
| 174 |
tanpif_gen_impl |
| 175 |
} |
| 176 |
}); |
| 177 |
unsafe { q(x) } |
| 178 |
} |
| 179 |
} |
| 180 |
|
| 181 |
#[cfg(test)] |
| 182 |
mod tests { |
| 183 |
use super::*; |
| 184 |
|
| 185 |
#[test] |
| 186 |
fn test_tanpif() { |
| 187 |
assert_eq!(f_tanpif(3.666738e-5), 0.00011519398); |
| 188 |
assert_eq!(f_tanpif(1.0355987e-25), 3.2534293e-25); |
| 189 |
assert_eq!(f_tanpif(5.5625), -5.0273395); |
| 190 |
assert_eq!(f_tanpif(-29.75), 1.0); |
| 191 |
assert_eq!(f_tanpif(-21.5625), 5.0273395); |
| 192 |
assert_eq!(f_tanpif(-15.611655), 2.7329326); |
| 193 |
assert_eq!(f_tanpif(115.30706), 1.4426143); |
| 194 |
assert!(f_tanpif(f32::INFINITY).is_nan()); |
| 195 |
assert!(f_tanpif(f32::NAN).is_nan()); |
| 196 |
} |
| 197 |
} |