1 | // Copyright 2014-2020 Optimal Computing (NZ) Ltd. |
2 | // Licensed under the MIT license. See LICENSE for details. |
3 | |
4 | use core::{f32, f64}; |
5 | #[cfg (feature = "num-traits" )] |
6 | #[allow (unused_imports)] |
7 | use num_traits::float::FloatCore; |
8 | use super::Ulps; |
9 | |
10 | /// A trait for approximate equality comparisons. |
11 | pub trait ApproxEq: Sized { |
12 | /// This type type defines a margin within which two values are to be |
13 | /// considered approximately equal. It must implement `Default` so that |
14 | /// `approx_eq()` can be called on unknown types. |
15 | type Margin: Copy + Default; |
16 | |
17 | /// This method tests that the `self` and `other` values are equal within `margin` |
18 | /// of each other. |
19 | fn approx_eq<M: Into<Self::Margin>>(self, other: Self, margin: M) -> bool; |
20 | |
21 | /// This method tests that the `self` and `other` values are not within `margin` |
22 | /// of each other. |
23 | fn approx_ne<M: Into<Self::Margin>>(self, other: Self, margin: M) -> bool { |
24 | !self.approx_eq(other, margin) |
25 | } |
26 | } |
27 | |
28 | /// This type defines a margin within two `f32` values might be considered equal, |
29 | /// and is intended as the associated type for the `ApproxEq` trait. |
30 | /// |
31 | /// Two tests are used to determine approximate equality. |
32 | /// |
33 | /// The first test considers two values approximately equal if they differ by <= |
34 | /// `epsilon`. This will only succeed for very small numbers. Note that it may |
35 | /// succeed even if the parameters are of differing signs, straddling zero. |
36 | /// |
37 | /// The second test considers how many ULPs (units of least precision, units in |
38 | /// the last place, which is the integer number of floating-point representations |
39 | /// that the parameters are separated by) different the parameters are and considers |
40 | /// them approximately equal if this is <= `ulps`. For large floating-point numbers, |
41 | /// an ULP can be a rather large gap, but this kind of comparison is necessary |
42 | /// because floating-point operations must round to the nearest representable value |
43 | /// and so larger floating-point values accumulate larger errors. |
44 | #[repr (C)] |
45 | #[derive(Debug, Clone, Copy)] |
46 | pub struct F32Margin { |
47 | pub epsilon: f32, |
48 | pub ulps: i32 |
49 | } |
50 | impl Default for F32Margin { |
51 | #[inline ] |
52 | fn default() -> F32Margin { |
53 | F32Margin { |
54 | epsilon: f32::EPSILON, |
55 | ulps: 4 |
56 | } |
57 | } |
58 | } |
59 | impl F32Margin { |
60 | #[inline ] |
61 | pub fn zero() -> F32Margin { |
62 | F32Margin { |
63 | epsilon: 0.0, |
64 | ulps: 0 |
65 | } |
66 | } |
67 | pub fn epsilon(self, epsilon: f32) -> Self { |
68 | F32Margin { |
69 | epsilon, |
70 | ..self |
71 | } |
72 | } |
73 | pub fn ulps(self, ulps: i32) -> Self { |
74 | F32Margin { |
75 | ulps, |
76 | ..self |
77 | } |
78 | } |
79 | } |
80 | impl From<(f32, i32)> for F32Margin { |
81 | fn from(m: (f32, i32)) -> F32Margin { |
82 | F32Margin { |
83 | epsilon: m.0, |
84 | ulps: m.1 |
85 | } |
86 | } |
87 | } |
88 | |
89 | impl ApproxEq for f32 { |
90 | type Margin = F32Margin; |
91 | |
92 | fn approx_eq<M: Into<Self::Margin>>(self, other: f32, margin: M) -> bool { |
93 | let margin = margin.into(); |
94 | |
95 | // Check for exact equality first. This is often true, and so we get the |
96 | // performance benefit of only doing one compare in most cases. |
97 | self==other || |
98 | |
99 | // Perform epsilon comparison next |
100 | ((self - other).abs() <= margin.epsilon) || |
101 | |
102 | { |
103 | // Perform ulps comparion last |
104 | let diff: i32 = self.ulps(&other); |
105 | saturating_abs_i32!(diff) <= margin.ulps |
106 | } |
107 | } |
108 | } |
109 | |
110 | #[test] |
111 | fn f32_approx_eq_test1() { |
112 | let f: f32 = 0.0_f32; |
113 | let g: f32 = -0.0000000000000005551115123125783_f32; |
114 | assert!(f != g); // Should not be directly equal |
115 | assert!(f.approx_eq(g, (f32::EPSILON, 0)) == true); |
116 | } |
117 | #[test] |
118 | fn f32_approx_eq_test2() { |
119 | let f: f32 = 0.0_f32; |
120 | let g: f32 = -0.0_f32; |
121 | assert!(f.approx_eq(g, (f32::EPSILON, 0)) == true); |
122 | } |
123 | #[test] |
124 | fn f32_approx_eq_test3() { |
125 | let f: f32 = 0.0_f32; |
126 | let g: f32 = 0.00000000000000001_f32; |
127 | assert!(f.approx_eq(g, (f32::EPSILON, 0)) == true); |
128 | } |
129 | #[test] |
130 | fn f32_approx_eq_test4() { |
131 | let f: f32 = 0.00001_f32; |
132 | let g: f32 = 0.00000000000000001_f32; |
133 | assert!(f.approx_eq(g, (f32::EPSILON, 0)) == false); |
134 | } |
135 | #[test] |
136 | fn f32_approx_eq_test5() { |
137 | let f: f32 = 0.1_f32; |
138 | let mut sum: f32 = 0.0_f32; |
139 | for _ in 0_isize..10_isize { sum += f; } |
140 | let product: f32 = f * 10.0_f32; |
141 | assert!(sum != product); // Should not be directly equal: |
142 | assert!(sum.approx_eq(product, (f32::EPSILON, 1)) == true); |
143 | assert!(sum.approx_eq(product, F32Margin::zero()) == false); |
144 | } |
145 | #[test] |
146 | fn f32_approx_eq_test6() { |
147 | let x: f32 = 1000000_f32; |
148 | let y: f32 = 1000000.1_f32; |
149 | assert!(x != y); // Should not be directly equal |
150 | assert!(x.approx_eq(y, (0.0, 2)) == true); // 2 ulps does it |
151 | // epsilon method no good here: |
152 | assert!(x.approx_eq(y, (1000.0 * f32::EPSILON, 0)) == false); |
153 | } |
154 | |
155 | /// This type defines a margin within two `f64` values might be considered equal, |
156 | /// and is intended as the associated type for the `ApproxEq` trait. |
157 | /// |
158 | /// Two tests are used to determine approximate equality. |
159 | /// |
160 | /// The first test considers two values approximately equal if they differ by <= |
161 | /// `epsilon`. This will only succeed for very small numbers. Note that it may |
162 | /// succeed even if the parameters are of differing signs, straddling zero. |
163 | /// |
164 | /// The second test considers how many ULPs (units of least precision, units in |
165 | /// the last place, which is the integer number of floating-point representations |
166 | /// that the parameters are separated by) different the parameters are and considers |
167 | /// them approximately equal if this is <= `ulps`. For large floating-point numbers, |
168 | /// an ULP can be a rather large gap, but this kind of comparison is necessary |
169 | /// because floating-point operations must round to the nearest representable value |
170 | /// and so larger floating-point values accumulate larger errors. |
171 | #[derive(Debug, Clone, Copy)] |
172 | pub struct F64Margin { |
173 | pub epsilon: f64, |
174 | pub ulps: i64 |
175 | } |
176 | impl Default for F64Margin { |
177 | #[inline ] |
178 | fn default() -> F64Margin { |
179 | F64Margin { |
180 | epsilon: f64::EPSILON, |
181 | ulps: 4 |
182 | } |
183 | } |
184 | } |
185 | impl F64Margin { |
186 | #[inline ] |
187 | pub fn zero() -> F64Margin { |
188 | F64Margin { |
189 | epsilon: 0.0, |
190 | ulps: 0 |
191 | } |
192 | } |
193 | pub fn epsilon(self, epsilon: f64) -> Self { |
194 | F64Margin { |
195 | epsilon, |
196 | ..self |
197 | } |
198 | } |
199 | pub fn ulps(self, ulps: i64) -> Self { |
200 | F64Margin { |
201 | ulps, |
202 | ..self |
203 | } |
204 | } |
205 | } |
206 | impl From<(f64, i64)> for F64Margin { |
207 | fn from(m: (f64, i64)) -> F64Margin { |
208 | F64Margin { |
209 | epsilon: m.0, |
210 | ulps: m.1 |
211 | } |
212 | } |
213 | } |
214 | |
215 | impl ApproxEq for f64 { |
216 | type Margin = F64Margin; |
217 | |
218 | fn approx_eq<M: Into<Self::Margin>>(self, other: f64, margin: M) -> bool { |
219 | let margin = margin.into(); |
220 | |
221 | // Check for exact equality first. This is often true, and so we get the |
222 | // performance benefit of only doing one compare in most cases. |
223 | self == other || |
224 | |
225 | // Perform epsilon comparison next |
226 | ((self - other).abs() <= margin.epsilon) || |
227 | |
228 | { |
229 | // Perform ulps comparion last |
230 | let diff: i64 = self.ulps(&other); |
231 | saturating_abs_i64!(diff) <= margin.ulps |
232 | } |
233 | } |
234 | } |
235 | |
236 | #[test] |
237 | fn f64_approx_eq_test1() { |
238 | let f: f64 = 0.0_f64; |
239 | let g: f64 = -0.0000000000000005551115123125783_f64; |
240 | assert!(f != g); // Should not be precisely equal. |
241 | assert!(f.approx_eq(g, (3.0 * f64::EPSILON, 0)) == true); // 3e is enough. |
242 | // ULPs test won't ever call these equal. |
243 | } |
244 | #[test] |
245 | fn f64_approx_eq_test2() { |
246 | let f: f64 = 0.0_f64; |
247 | let g: f64 = -0.0_f64; |
248 | assert!(f.approx_eq(g, (f64::EPSILON, 0)) == true); |
249 | } |
250 | #[test] |
251 | fn f64_approx_eq_test3() { |
252 | let f: f64 = 0.0_f64; |
253 | let g: f64 = 1e-17_f64; |
254 | assert!(f.approx_eq(g, (f64::EPSILON, 0)) == true); |
255 | } |
256 | #[test] |
257 | fn f64_approx_eq_test4() { |
258 | let f: f64 = 0.00001_f64; |
259 | let g: f64 = 0.00000000000000001_f64; |
260 | assert!(f.approx_eq(g, (f64::EPSILON, 0)) == false); |
261 | } |
262 | #[test] |
263 | fn f64_approx_eq_test5() { |
264 | let f: f64 = 0.1_f64; |
265 | let mut sum: f64 = 0.0_f64; |
266 | for _ in 0_isize..10_isize { sum += f; } |
267 | let product: f64 = f * 10.0_f64; |
268 | assert!(sum != product); // Should not be precisely equaly. |
269 | assert!(sum.approx_eq(product, (f64::EPSILON, 0)) == true); |
270 | assert!(sum.approx_eq(product, (0.0, 1)) == true); |
271 | } |
272 | #[test] |
273 | fn f64_approx_eq_test6() { |
274 | let x: f64 = 1000000_f64; |
275 | let y: f64 = 1000000.0000000003_f64; |
276 | assert!(x != y); // Should not be precisely equal. |
277 | assert!(x.approx_eq(y, (0.0, 3)) == true); |
278 | } |
279 | #[test] |
280 | fn f64_code_triggering_issue_20() { |
281 | assert_eq!((-25.0f64).approx_eq(25.0, (0.00390625, 1)), false); |
282 | } |
283 | |
284 | impl<T> ApproxEq for &[T] |
285 | where T: Copy + ApproxEq { |
286 | type Margin = <T as ApproxEq>::Margin; |
287 | |
288 | fn approx_eq<M: Into<Self::Margin>>(self, other: Self, margin: M) -> bool { |
289 | let margin = margin.into(); |
290 | if self.len() != other.len() { return false; } |
291 | self.iter().zip(other.iter()).all(|(a,b)| { |
292 | a.approx_eq(*b, margin) |
293 | }) |
294 | } |
295 | } |
296 | |
297 | #[test] |
298 | fn test_slices() { |
299 | assert!( [1.33, 2.4, 2.5].approx_eq(&[1.33, 2.4, 2.5], (0.0, 0_i64)) ); |
300 | assert!( ! [1.33, 2.4, 2.6].approx_eq(&[1.33, 2.4, 2.5], (0.0, 0_i64)) ); |
301 | assert!( ! [1.33, 2.4].approx_eq(&[1.33, 2.4, 2.5], (0.0, 0_i64)) ); |
302 | assert!( ! [1.33, 2.4, 2.5].approx_eq(&[1.33, 2.4], (0.0, 0_i64)) ); |
303 | } |
304 | |
305 | |