| 1 |
/* |
| 2 |
* // Copyright (c) Radzivon Bartoshyk 7/2025. 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::double_double::DoubleDouble; |
| 31 |
use crate::sin::{get_sin_k_rational, range_reduction_small, sincos_eval}; |
| 32 |
use crate::sin_table::SIN_K_PI_OVER_128; |
| 33 |
use crate::sincos_dyadic::{range_reduction_small_f128, sincos_eval_dyadic}; |
| 34 |
use crate::sincos_reduce::LargeArgumentReduction; |
| 35 |
|
| 36 |
#[cold] |
| 37 |
fn sec_accurate(x: f64, argument_reduction: &mut LargeArgumentReduction, x_e: u64, k: u64) -> f64 { |
| 38 |
const EXP_BIAS: u64 = (1u64 << (11 - 1u64)) - 1u64; |
| 39 |
let u_f128 = if x_e < EXP_BIAS + 16 { |
| 40 |
range_reduction_small_f128(x) |
| 41 |
} else { |
| 42 |
argument_reduction.accurate() |
| 43 |
}; |
| 44 |
|
| 45 |
let sin_cos = sincos_eval_dyadic(&u_f128); |
| 46 |
|
| 47 |
// -sin(k * pi/128) = sin((k + 128) * pi/128) |
| 48 |
// cos(k * pi/128) = sin(k * pi/128 + pi/2) = sin((k + 64) * pi/128). |
| 49 |
let msin_k_f128 = get_sin_k_rational(k.wrapping_add(128)); |
| 50 |
let cos_k_f128 = get_sin_k_rational(k.wrapping_add(64)); |
| 51 |
|
| 52 |
// cos(x) = cos((k * pi/128 + u) |
| 53 |
// = cos(u) * cos(k*pi/128) - sin(u) * sin(k*pi/128) |
| 54 |
let r = (cos_k_f128 * sin_cos.v_cos) + (msin_k_f128 * sin_cos.v_sin); |
| 55 |
r.reciprocal().fast_as_f64() |
| 56 |
} |
| 57 |
|
| 58 |
/// Secant for double precision |
| 59 |
/// |
| 60 |
/// ULP 0.5 |
| 61 |
pub fn f_sec(x: f64) -> f64 { |
| 62 |
let x_e = (x.to_bits() >> 52) & 0x7ff; |
| 63 |
const E_BIAS: u64 = (1u64 << (11 - 1u64)) - 1u64; |
| 64 |
|
| 65 |
let y: DoubleDouble; |
| 66 |
let k; |
| 67 |
|
| 68 |
let mut argument_reduction = LargeArgumentReduction::default(); |
| 69 |
|
| 70 |
if x_e < E_BIAS + 16 { |
| 71 |
// |x| < 2^32 (with FMA) or |x| < 2^23 (w/o FMA) |
| 72 |
if x_e < E_BIAS - 7 { |
| 73 |
// |x| < 2^-7 |
| 74 |
if x_e < E_BIAS - 27 { |
| 75 |
// |x| < 2^-27 |
| 76 |
if x == 0.0 { |
| 77 |
// Signed zeros. |
| 78 |
return 1.0; |
| 79 |
} |
| 80 |
// taylor series for sec(x) ~ 1 + x^2/2 + O(x^4) |
| 81 |
return f_fmla(x, x * 0.5, 1.); |
| 82 |
} |
| 83 |
k = 0; |
| 84 |
y = DoubleDouble::new(0.0, x); |
| 85 |
} else { |
| 86 |
// Small range reduction. |
| 87 |
(y, k) = range_reduction_small(x); |
| 88 |
} |
| 89 |
} else { |
| 90 |
// Inf or NaN |
| 91 |
if x_e > 2 * E_BIAS { |
| 92 |
// sec(+-Inf) = NaN |
| 93 |
return x + f64::NAN; |
| 94 |
} |
| 95 |
|
| 96 |
// Large range reduction. |
| 97 |
(k, y) = argument_reduction.reduce(x); |
| 98 |
} |
| 99 |
let r_sincos = sincos_eval(y); |
| 100 |
|
| 101 |
// Fast look up version, but needs 256-entry table. |
| 102 |
// cos(k * pi/128) = sin(k * pi/128 + pi/2) = sin((k + 64) * pi/128). |
| 103 |
let sk = SIN_K_PI_OVER_128[(k.wrapping_add(128) & 255) as usize]; |
| 104 |
let ck = SIN_K_PI_OVER_128[((k.wrapping_add(64)) & 255) as usize]; |
| 105 |
let msin_k = DoubleDouble::from_bit_pair(sk); |
| 106 |
let cos_k = DoubleDouble::from_bit_pair(ck); |
| 107 |
|
| 108 |
let cos_k_cos_y = DoubleDouble::quick_mult(r_sincos.v_cos, cos_k); |
| 109 |
let cos_k_msin_y = DoubleDouble::quick_mult(r_sincos.v_sin, msin_k); |
| 110 |
|
| 111 |
// cos_k_cos_y is always >> cos_k_msin_y |
| 112 |
let mut rr = DoubleDouble::from_exact_add(cos_k_cos_y.hi, cos_k_msin_y.hi); |
| 113 |
rr.lo += cos_k_cos_y.lo + cos_k_msin_y.lo; |
| 114 |
|
| 115 |
rr = DoubleDouble::from_exact_add(rr.hi, rr.lo); |
| 116 |
rr = rr.recip(); |
| 117 |
|
| 118 |
let rlp = rr.lo + r_sincos.err; |
| 119 |
let rlm = rr.lo - r_sincos.err; |
| 120 |
|
| 121 |
let r_upper = rr.hi + rlp; // (rr.lo + ERR); |
| 122 |
let r_lower = rr.hi + rlm; // (rr.lo - ERR); |
| 123 |
|
| 124 |
// Ziv's accuracy test |
| 125 |
if r_upper == r_lower { |
| 126 |
return rr.to_f64(); |
| 127 |
} |
| 128 |
|
| 129 |
sec_accurate(x, &mut argument_reduction, x_e, k) |
| 130 |
} |
| 131 |
|
| 132 |
#[cfg(test)] |
| 133 |
mod tests { |
| 134 |
use super::*; |
| 135 |
|
| 136 |
#[test] |
| 137 |
fn test_sec() { |
| 138 |
assert_eq!(f_sec(-175432.), 1.461049620895326); |
| 139 |
assert_eq!(f_sec(175432.), 1.461049620895326); |
| 140 |
assert_eq!(f_sec(-10.), -1.1917935066878957); |
| 141 |
assert_eq!(f_sec(10.), -1.1917935066878957); |
| 142 |
assert_eq!(f_sec(5.), 3.5253200858160882); |
| 143 |
assert_eq!(f_sec(-5.), 3.5253200858160882); |
| 144 |
assert_eq!(f_sec(0.), 1.0); |
| 145 |
assert!(f_sec(f64::NAN).is_nan()); |
| 146 |
assert!(f_sec(f64::INFINITY).is_nan()); |
| 147 |
assert!(f_sec(f64::NEG_INFINITY).is_nan()); |
| 148 |
} |
| 149 |
} |