| 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::double_double::DoubleDouble; |
| 30 |
use crate::exponents::auxiliary::fast_ldexp; |
| 31 |
use crate::exponents::exp::{EXP_REDUCE_T0, EXP_REDUCE_T1, to_denormal}; |
| 32 |
use crate::exponents::expf::{ExpfBackend, GenericExpfBackend}; |
| 33 |
use crate::rounding::CpuRoundTiesEven; |
| 34 |
use std::hint::black_box; |
| 35 |
|
| 36 |
#[inline] |
| 37 |
fn exp10_poly_dd(z: DoubleDouble) -> DoubleDouble { |
| 38 |
const C: [(u64, u64); 6] = [ |
| 39 |
(0xbcaf48ad494ea102, 0x40026bb1bbb55516), |
| 40 |
(0xbcae2bfab318d399, 0x40053524c73cea69), |
| 41 |
(0x3ca81f50779e162b, 0x4000470591de2ca4), |
| 42 |
(0x3c931a5cc5d3d313, 0x3ff2bd7609fd98c4), |
| 43 |
(0x3c8910de8c68a0c2, 0x3fe1429ffd336aa3), |
| 44 |
(0xbc605e703d496537, 0x3fca7ed7086882b4), |
| 45 |
]; |
| 46 |
|
| 47 |
let mut r = DoubleDouble::quick_mul_add( |
| 48 |
DoubleDouble::from_bit_pair(C[5]), |
| 49 |
z, |
| 50 |
DoubleDouble::from_bit_pair(C[4]), |
| 51 |
); |
| 52 |
r = DoubleDouble::quick_mul_add(r, z, DoubleDouble::from_bit_pair(C[3])); |
| 53 |
r = DoubleDouble::quick_mul_add(r, z, DoubleDouble::from_bit_pair(C[2])); |
| 54 |
r = DoubleDouble::quick_mul_add(r, z, DoubleDouble::from_bit_pair(C[1])); |
| 55 |
DoubleDouble::quick_mul_add(r, z, DoubleDouble::from_bit_pair(C[0])) |
| 56 |
} |
| 57 |
|
| 58 |
#[cold] |
| 59 |
fn as_exp10_accurate(x: f64) -> f64 { |
| 60 |
let mut ix = x.to_bits(); |
| 61 |
let t = (f64::from_bits(0x40ca934f0979a371) * x).cpu_round_ties_even(); |
| 62 |
let jt: i64 = unsafe { |
| 63 |
t.to_int_unchecked::<i64>() // t is already integer, this is just a conversion |
| 64 |
}; |
| 65 |
let i1 = jt & 0x3f; |
| 66 |
let i0 = (jt >> 6) & 0x3f; |
| 67 |
let ie = jt >> 12; |
| 68 |
let t0 = DoubleDouble::from_bit_pair(EXP_REDUCE_T0[i0 as usize]); |
| 69 |
let t1 = DoubleDouble::from_bit_pair(EXP_REDUCE_T1[i1 as usize]); |
| 70 |
let dt = DoubleDouble::quick_mult(t0, t1); |
| 71 |
|
| 72 |
const L0: f64 = f64::from_bits(0x3f13441350800000); |
| 73 |
const L1: f64 = f64::from_bits(0xbd1f79fef311f12b); |
| 74 |
const L2: f64 = f64::from_bits(0xb9aac0b7c917826b); |
| 75 |
|
| 76 |
let dx = x - L0 * t; |
| 77 |
let dx_dd = DoubleDouble::quick_mult_f64(DoubleDouble::new(L2, L1), t); |
| 78 |
let dz = DoubleDouble::full_add_f64(dx_dd, dx); |
| 79 |
let mut f = exp10_poly_dd(dz); |
| 80 |
f = DoubleDouble::quick_mult(dz, f); |
| 81 |
|
| 82 |
let mut zfh: f64; |
| 83 |
|
| 84 |
if ix < 0xc0733a7146f72a42u64 { |
| 85 |
if (jt & 0xfff) == 0 { |
| 86 |
f = DoubleDouble::from_exact_add(f.hi, f.lo); |
| 87 |
let zt = DoubleDouble::from_exact_add(dt.hi, f.hi); |
| 88 |
f.hi = zt.lo; |
| 89 |
f = DoubleDouble::from_exact_add(f.hi, f.lo); |
| 90 |
ix = f.hi.to_bits(); |
| 91 |
if (ix.wrapping_shl(12)) == 0 { |
| 92 |
let l = f.lo.to_bits(); |
| 93 |
let sfh: i64 = ((ix as i64) >> 63) ^ ((l as i64) >> 63); |
| 94 |
ix = ix.wrapping_add(((1i64 << 51) ^ sfh) as u64); |
| 95 |
} |
| 96 |
zfh = zt.hi + f64::from_bits(ix); |
| 97 |
} else { |
| 98 |
f = DoubleDouble::quick_mult(f, dt); |
| 99 |
f = DoubleDouble::add(dt, f); |
| 100 |
f = DoubleDouble::from_exact_add(f.hi, f.lo); |
| 101 |
zfh = f.hi; |
| 102 |
} |
| 103 |
zfh = fast_ldexp(zfh, ie as i32); |
| 104 |
} else { |
| 105 |
ix = (1u64.wrapping_sub(ie as u64)) << 52; |
| 106 |
f = DoubleDouble::quick_mult(f, dt); |
| 107 |
f = DoubleDouble::add(dt, f); |
| 108 |
|
| 109 |
let zt = DoubleDouble::from_exact_add(f64::from_bits(ix), f.hi); |
| 110 |
f.hi = zt.hi; |
| 111 |
f.lo += zt.lo; |
| 112 |
|
| 113 |
zfh = to_denormal(f.to_f64()); |
| 114 |
} |
| 115 |
zfh |
| 116 |
} |
| 117 |
|
| 118 |
#[inline(always)] |
| 119 |
fn exp10_gen<B: ExpfBackend>(x: f64, backend: B) -> f64 { |
| 120 |
let mut ix = x.to_bits(); |
| 121 |
let aix = ix & 0x7fff_ffff_ffff_ffff; |
| 122 |
if aix > 0x40734413509f79feu64 { |
| 123 |
// |x| > 0x40734413509f79fe |
| 124 |
if aix > 0x7ff0000000000000u64 { |
| 125 |
return x + x; |
| 126 |
} // nan |
| 127 |
if aix == 0x7ff0000000000000u64 { |
| 128 |
return if (ix >> 63) != 0 { 0.0 } else { x }; |
| 129 |
} |
| 130 |
if (ix >> 63) == 0 { |
| 131 |
return f64::from_bits(0x7fe0000000000000) * 2.0; // x > 308.255 |
| 132 |
} |
| 133 |
if aix > 0x407439b746e36b52u64 { |
| 134 |
// x < -323.607 |
| 135 |
return black_box(f64::from_bits(0x0018000000000000)) |
| 136 |
* black_box(f64::from_bits(0x3c80000000000000)); |
| 137 |
} |
| 138 |
} |
| 139 |
|
| 140 |
// check x integer to avoid a spurious inexact exception |
| 141 |
if ix.wrapping_shl(16) == 0 && (aix >> 48) <= 0x4036 { |
| 142 |
let kx = backend.round_ties_even(x); |
| 143 |
if kx == x { |
| 144 |
let k = kx as i64; |
| 145 |
if k >= 0 { |
| 146 |
let mut r = 1.0; |
| 147 |
for _ in 0..k { |
| 148 |
r *= 10.0; |
| 149 |
} |
| 150 |
return r; |
| 151 |
} |
| 152 |
} |
| 153 |
} |
| 154 |
/* avoid spurious underflow: for |x| <= 2.41082e-17 |
| 155 |
exp10(x) rounds to 1 to nearest */ |
| 156 |
if aix <= 0x3c7bcb7b1526e50eu64 { |
| 157 |
return 1.0 + x; // |x| <= 2.41082e-17 |
| 158 |
} |
| 159 |
let t = backend.round_ties_even(f64::from_bits(0x40ca934f0979a371) * x); |
| 160 |
let jt: i64 = unsafe { t.to_int_unchecked::<i64>() }; // t is already integer this is just a conversion |
| 161 |
let i1 = jt & 0x3f; |
| 162 |
let i0 = (jt >> 6) & 0x3f; |
| 163 |
let ie = jt >> 12; |
| 164 |
let t00 = EXP_REDUCE_T0[i0 as usize]; |
| 165 |
let t01 = EXP_REDUCE_T1[i1 as usize]; |
| 166 |
let t0 = DoubleDouble::from_bit_pair(t00); |
| 167 |
let t1 = DoubleDouble::from_bit_pair(t01); |
| 168 |
let mut tz = backend.quick_mult(t0, t1); |
| 169 |
const L0: f64 = f64::from_bits(0x3f13441350800000); |
| 170 |
const L1: f64 = f64::from_bits(0x3d1f79fef311f12b); |
| 171 |
let dx = backend.fma(-L1, t, backend.fma(-L0, t, x)); |
| 172 |
let dx2 = dx * dx; |
| 173 |
|
| 174 |
const CH: [u64; 4] = [ |
| 175 |
0x40026bb1bbb55516, |
| 176 |
0x40053524c73cea69, |
| 177 |
0x4000470591fd74e1, |
| 178 |
0x3ff2bd760a1f32a5, |
| 179 |
]; |
| 180 |
|
| 181 |
let p0 = backend.fma(dx, f64::from_bits(CH[1]), f64::from_bits(CH[0])); |
| 182 |
let p1 = backend.fma(dx, f64::from_bits(CH[3]), f64::from_bits(CH[2])); |
| 183 |
|
| 184 |
let p = backend.fma(dx2, p1, p0); |
| 185 |
|
| 186 |
let mut fh = tz.hi; |
| 187 |
let fx = tz.hi * dx; |
| 188 |
let mut fl = backend.fma(fx, p, tz.lo); |
| 189 |
const EPS: f64 = 1.63e-19; |
| 190 |
if ix < 0xc0733a7146f72a42u64 { |
| 191 |
// x > -307.653 |
| 192 |
// x > -0x1.33a7146f72a42p+8 |
| 193 |
let ub = fh + (fl + EPS); |
| 194 |
let lb = fh + (fl - EPS); |
| 195 |
|
| 196 |
if lb != ub { |
| 197 |
return as_exp10_accurate(x); |
| 198 |
} |
| 199 |
fh = fast_ldexp(fh + fl, ie as i32); |
| 200 |
} else { |
| 201 |
// x <= -307.653: exp10(x) < 2^-1022 |
| 202 |
ix = 1u64.wrapping_sub(ie as u64).wrapping_shl(52); |
| 203 |
tz = DoubleDouble::from_exact_add(f64::from_bits(ix), fh); |
| 204 |
fl += tz.lo; |
| 205 |
|
| 206 |
let ub = fh + (fl + EPS); |
| 207 |
let lb = fh + (fl - EPS); |
| 208 |
|
| 209 |
if lb != ub { |
| 210 |
return as_exp10_accurate(x); |
| 211 |
} |
| 212 |
|
| 213 |
fh = to_denormal(fh + fl); |
| 214 |
} |
| 215 |
fh |
| 216 |
} |
| 217 |
|
| 218 |
#[cfg(any(target_arch = "x86", target_arch = "x86_64"))] |
| 219 |
#[target_feature(enable = "avx", enable = "fma")] |
| 220 |
unsafe fn exp10_fma_impl(x: f64) -> f64 { |
| 221 |
use crate::exponents::expf::FmaBackend; |
| 222 |
exp10_gen(x, FmaBackend {}) |
| 223 |
} |
| 224 |
|
| 225 |
/// Computes exp10 |
| 226 |
/// |
| 227 |
/// Max found ULP 0.5 |
| 228 |
pub fn f_exp10(x: f64) -> f64 { |
| 229 |
#[cfg(not(any(target_arch = "x86", target_arch = "x86_64")))] |
| 230 |
{ |
| 231 |
exp10_gen(x, GenericExpfBackend {}) |
| 232 |
} |
| 233 |
#[cfg(any(target_arch = "x86", target_arch = "x86_64"))] |
| 234 |
{ |
| 235 |
use std::sync::OnceLock; |
| 236 |
static EXECUTOR: OnceLock<unsafe fn(f64) -> f64> = OnceLock::new(); |
| 237 |
let q = EXECUTOR.get_or_init(|| { |
| 238 |
if std::arch::is_x86_feature_detected!("avx") |
| 239 |
&& std::arch::is_x86_feature_detected!("fma") |
| 240 |
{ |
| 241 |
exp10_fma_impl |
| 242 |
} else { |
| 243 |
fn def_exp10(x: f64) -> f64 { |
| 244 |
exp10_gen(x, GenericExpfBackend {}) |
| 245 |
} |
| 246 |
def_exp10 |
| 247 |
} |
| 248 |
}); |
| 249 |
unsafe { q(x) } |
| 250 |
} |
| 251 |
} |
| 252 |
|
| 253 |
#[cfg(test)] |
| 254 |
mod tests { |
| 255 |
use super::*; |
| 256 |
|
| 257 |
#[test] |
| 258 |
fn test_exp10f() { |
| 259 |
assert_eq!(f_exp10(-3.370739843267434), 0.00042585343701025656); |
| 260 |
assert_eq!(f_exp10(1.), 10.0); |
| 261 |
assert_eq!(f_exp10(2.), 100.0); |
| 262 |
assert_eq!(f_exp10(3.), 1000.0); |
| 263 |
assert_eq!(f_exp10(4.), 10000.0); |
| 264 |
assert_eq!(f_exp10(5.), 100000.0); |
| 265 |
assert_eq!(f_exp10(6.), 1000000.0); |
| 266 |
assert_eq!(f_exp10(7.), 10000000.0); |
| 267 |
assert_eq!(f_exp10(f64::INFINITY), f64::INFINITY); |
| 268 |
assert_eq!(f_exp10(f64::NEG_INFINITY), 0.); |
| 269 |
assert!(f_exp10(f64::NAN).is_nan()); |
| 270 |
} |
| 271 |
} |