Commit 7f9b67443ed79dac2fc20b86fab675821298bfc8

Authored by Georg Hopp
1 parent 93df7ada

Add tangens and improve fractional rep

Showing 1 changed file with 77 additions and 56 deletions
1 1 //
2 2 // Some trigonometic functions with Fractions results.
3   -// Currently only sin and cos are implemented. As I was unable
4   -// to find a really good integral approximation for them I
5   -// implement them as tables which are predefined using the
6   -// floating point function f64::sin and then transformed into
7   -// a fraction of a given PRECISION.
  3 +// Currently only sin, cos and tan are implemented.
  4 +// As I was unable to find a really good integral approximation for them I
  5 +// implement them as a table which is predefined using the floating point
  6 +// function f64::sin and then transformed into a fraction of a given
  7 +// PRECISION.
  8 +// These approximations are quite good and for a few edge cases
  9 +// even better than the floating point implementations.
8 10 //
9 11 // Georg Hopp <georg@steffers.org>
10 12 //
... ... @@ -23,75 +25,94 @@
23 25 // You should have received a copy of the GNU General Public License
24 26 // along with this program. If not, see <http://www.gnu.org/licenses/>.
25 27 //
  28 +use std::cmp::Ordering;
26 29 use crate::{Fractional};
27 30
28 31 pub const PI :Fractional = Fractional(355, 113); // This is a really close
29 32 // fractional approximation
30 33 // for pi
31 34
32   -const PRECISION :i64 = 100000;
  35 +// Try to keep precision as high as possible while having a denominator
  36 +// as small as possible.
  37 +// The values are taken by triel and error.
  38 +const PRECISION :i64 = 1000000;
  39 +const MAX_DENOMINATOR :i64 = 7000;
33 40
34   -#[inline]
35   -pub fn rad(d: u32) -> f64 {
36   - use std::f64::consts::PI;
37   - d as f64 * PI / 180.0
38   -}
39   -
40   -pub fn sin(d: i32) -> Fractional {
41   - // hold sin Fractionals from 0 to 89 ...
42   - lazy_static::lazy_static! {
43   - static ref SINTAB :Vec<Fractional> =
44   - (0..90).map(|x| _sin(x)).collect();
  41 +pub fn sin(d :i32) -> Fractional {
  42 + match d {
  43 + 0 ..=90 => SINTAB[d as usize],
  44 + 91 ..=180 => SINTAB[180 - d as usize],
  45 + 181..=270 => -SINTAB[d as usize - 180],
  46 + 271..=359 => -SINTAB[360 - d as usize],
  47 + _ => sin(d % 360),
45 48 }
  49 +}
46 50
47   - // fractional sin from f64 sin. (From 1° to 89°)
48   - fn _sin(d: u32) -> Fractional {
49   - match d {
50   - 0 => Fractional(0, 1),
51   - _ => {
52   - // This is undefined behaviour for very large f64, but our f64
53   - // is always between 0.0 and 10000.0 which should be fine.
54   - let s = (f64::sin(rad(d)) * PRECISION as f64).round() as i64;
55   - Fractional(s, PRECISION).reduce()
56   - }
57   - }
  51 +pub fn cos(d :i32) -> Fractional {
  52 + match d {
  53 + 0 ..=90 => SINTAB[90 - d as usize],
  54 + 91 ..=180 => -SINTAB[90 - (180 - d as usize)],
  55 + 181..=270 => -SINTAB[90 - (d as usize - 180)],
  56 + 271..=359 => SINTAB[90 - (360 - d as usize)],
  57 + _ => cos(d % 360),
58 58 }
  59 +}
59 60
  61 +pub fn tan(d :i32) -> Fractional {
60 62 match d {
61   - 90 => Fractional(1, 1),
62   - 180 => SINTAB[0],
63   - 270 => -Fractional(1, 1),
64   - 1..=89 => SINTAB[d as usize],
65   - 91..=179 => SINTAB[180 - d as usize],
66   - 181..=269 => -SINTAB[d as usize - 180],
67   - 271..=359 => -SINTAB[360 - d as usize],
68   - _ => sin(d % 360),
  63 + 0 ..=179 => TANTAB[d as usize],
  64 + 180..=359 => TANTAB[d as usize - 180],
  65 + _ => tan(d % 360),
69 66 }
70 67 }
71 68
72   -pub fn cos(d: i32) -> Fractional {
73   - lazy_static::lazy_static! {
74   - static ref COSTAB :Vec<Fractional> =
75   - (0..90).map(|x| _cos(x)).collect();
76   - }
  69 +// hold sin Fractionals from 0 to 89 ...
  70 +// luckily with a bit of index tweeking this can also be used for
  71 +// cosine values.
  72 +lazy_static::lazy_static! {
  73 + static ref SINTAB :Vec<Fractional> =
  74 + (0..=90).map(|x| _sin(x)).collect();
  75 +}
77 76
78   - fn _cos(d: u32) -> Fractional {
79   - match d {
80   - 0 => Fractional(1, 1),
81   - _ => {
82   - let s = (f64::cos(rad(d)) * PRECISION as f64).round() as i64;
83   - Fractional(s, PRECISION).reduce()
84   - }
85   - }
  77 +// This table exists only because the sin(α) / cos(α) method
  78 +// yields very large unreducable denominators in a lot of cases.
  79 +lazy_static::lazy_static! {
  80 + static ref TANTAB :Vec<Fractional> =
  81 + (0..180).map(|x| _tan(x)).collect();
  82 +}
  83 +
  84 +// fractional sin from f64 sin. (From 0° to 90°)
  85 +fn _sin(d: u32) -> Fractional {
  86 + match d {
  87 + 0 => Fractional(0, 1),
  88 + 90 => Fractional(1, 1),
  89 + _ => reduce(d, PRECISION, &f64::sin),
86 90 }
  91 +}
87 92
  93 +// fractional tan from f64 tan. (From 0° to 179°)
  94 +fn _tan(d: u32) -> Fractional {
88 95 match d {
89   - 90 | 270 => Fractional(0, 1),
90   - 180 => -COSTAB[0],
91   - 1..=89 => COSTAB[d as usize],
92   - 91..=179 => -COSTAB[180 - d as usize],
93   - 181..=269 => -COSTAB[d as usize - 180],
94   - 271..=359 => COSTAB[360 - d as usize],
95   - _ => cos(d % 360),
  96 + 0 => Fractional(0, 1),
  97 + 45 => Fractional(1, 1),
  98 + 90 => Fractional(1, 0), // although they are both inf and -inf.
  99 + 135 => -Fractional(1, 1),
  100 + _ => reduce(d, PRECISION, &f64::tan),
  101 + }
  102 +}
  103 +
  104 +// search for a fraction with a denominator less than 10000 that
  105 +// provides the minimal precision criteria.
  106 +// !! With f = &f64::tan and d close to the inf boundarys of tan
  107 +// we get very large numerators because the numerator becomes a
  108 +// multiple of the denominator.
  109 +fn reduce(d :u32, p :i64, f :&dyn Fn(f64) -> f64) -> Fractional {
  110 + // This is undefined behaviour for very large f64, but our f64
  111 + // is always between 0.0 and 10000000.0 which should be fine.
  112 + let s = (f(f64::to_radians(d as f64)) * p as f64).round() as i64;
  113 + let Fractional(n, dn) = Fractional(s, p).reduce();
  114 + match dn.abs().cmp(&MAX_DENOMINATOR) {
  115 + Ordering::Less => Fractional(n, dn),
  116 + _ => reduce(d, p + 1, f),
96 117 }
97 118 }
... ...
Please register or login to post a comment