1 | //! Exp approximation for a single-precision float. |
2 | //! |
3 | //! Method described at: <https://stackoverflow.com/a/6985769/2036035> |
4 | |
5 | use super::{EXPONENT_BIAS, F32}; |
6 | use core::f32::consts; |
7 | |
8 | impl F32 { |
9 | /// Returns `e^(self)`, (the exponential function). |
10 | #[inline ] |
11 | pub fn exp(self) -> Self { |
12 | self.exp_ln2_approx(4) |
13 | } |
14 | |
15 | /// Exp approximation for `f32`. |
16 | pub(crate) fn exp_ln2_approx(self, partial_iter: u32) -> Self { |
17 | if self == Self::ZERO { |
18 | return Self::ONE; |
19 | } |
20 | |
21 | if (self - Self::ONE).abs() < f32::EPSILON { |
22 | return consts::E.into(); |
23 | } |
24 | |
25 | if (self - (-Self::ONE)).abs() < f32::EPSILON { |
26 | return Self::ONE / consts::E; |
27 | } |
28 | |
29 | // log base 2(E) == 1/ln(2) |
30 | // x_fract + x_whole = x/ln2_recip |
31 | // ln2*(x_fract + x_whole) = x |
32 | let x_ln2recip = self * consts::LOG2_E; |
33 | let x_fract = x_ln2recip.fract(); |
34 | let x_trunc = x_ln2recip.trunc(); |
35 | |
36 | //guaranteed to be 0 < x < 1.0 |
37 | let x_fract = x_fract * consts::LN_2; |
38 | let fract_exp = x_fract.exp_smallx(partial_iter); |
39 | |
40 | //need the 2^n portion, we can just extract that from the whole number exp portion |
41 | let fract_exponent: i32 = fract_exp |
42 | .extract_exponent_value() |
43 | .saturating_add(x_trunc.0 as i32); |
44 | |
45 | if fract_exponent < -(EXPONENT_BIAS as i32) { |
46 | return Self::ZERO; |
47 | } |
48 | |
49 | if fract_exponent > ((EXPONENT_BIAS + 1) as i32) { |
50 | return Self::INFINITY; |
51 | } |
52 | |
53 | fract_exp.set_exponent(fract_exponent) |
54 | } |
55 | |
56 | /// if x is between 0.0 and 1.0, we can approximate it with the a series |
57 | /// |
58 | /// Series from here: |
59 | /// <https://stackoverflow.com/a/6984495> |
60 | /// |
61 | /// e^x ~= 1 + x(1 + x/2(1 + (x? |
62 | #[inline ] |
63 | pub(crate) fn exp_smallx(self, iter: u32) -> Self { |
64 | let mut total = 1.0; |
65 | |
66 | for i in (1..=iter).rev() { |
67 | total = 1.0 + ((self.0 / (i as f32)) * total); |
68 | } |
69 | |
70 | Self(total) |
71 | } |
72 | } |
73 | |
74 | #[cfg (test)] |
75 | mod tests { |
76 | use super::F32; |
77 | |
78 | pub(crate) const MAX_ERROR: f32 = 0.001; |
79 | |
80 | /// exp test vectors - `(input, output)` |
81 | pub(crate) const TEST_VECTORS: &[(f32, f32)] = &[ |
82 | (1e-07, 1.0000001), |
83 | (1e-06, 1.000001), |
84 | (1e-05, 1.00001), |
85 | (1e-04, 1.0001), |
86 | (0.001, 1.0010005), |
87 | (0.01, 1.0100502), |
88 | (0.1, 1.105171), |
89 | (1.0, 2.7182817), |
90 | (10.0, 22026.465), |
91 | (-1e-08, 1.0), |
92 | (-1e-07, 0.9999999), |
93 | (-1e-06, 0.999999), |
94 | (-1e-05, 0.99999), |
95 | (-1e-04, 0.9999), |
96 | (-0.001, 0.9990005), |
97 | (-0.01, 0.99004984), |
98 | (-0.1, 0.9048374), |
99 | (-1.0, 0.36787945), |
100 | (-10.0, 4.539_993e-5), |
101 | ]; |
102 | |
103 | #[test ] |
104 | fn sanity_check() { |
105 | assert_eq!(F32(-1000000.0).exp(), F32::ZERO); |
106 | |
107 | for &(x, expected) in TEST_VECTORS { |
108 | let exp_x = F32(x).exp(); |
109 | let relative_error = (exp_x - expected).abs() / expected; |
110 | |
111 | assert!( |
112 | relative_error <= MAX_ERROR, |
113 | "relative_error {} too large for input {} : {} vs {}" , |
114 | relative_error, |
115 | x, |
116 | exp_x, |
117 | expected |
118 | ); |
119 | } |
120 | } |
121 | } |
122 | |