| 1 | // Copyright 2012 Google Inc. |
| 2 | // Copyright 2020 Yevhenii Reizner |
| 3 | // |
| 4 | // Use of this source code is governed by a BSD-style license that can be |
| 5 | // found in the LICENSE file. |
| 6 | |
| 7 | use tiny_skia_path::{Scalar, SCALAR_MAX}; |
| 8 | |
| 9 | #[cfg (all(not(feature = "std" ), feature = "no-std-float" ))] |
| 10 | use tiny_skia_path::NoStdFloat; |
| 11 | |
| 12 | // Must be first, because of macro scope rules. |
| 13 | #[macro_use ] |
| 14 | pub mod point64; |
| 15 | |
| 16 | pub mod cubic64; |
| 17 | pub mod line_cubic_intersections; |
| 18 | mod quad64; |
| 19 | |
| 20 | // The code below is from SkPathOpsTypes. |
| 21 | |
| 22 | const DBL_EPSILON_ERR: f64 = f64::EPSILON * 4.0; |
| 23 | const FLT_EPSILON_HALF: f64 = (f32::EPSILON / 2.0) as f64; |
| 24 | const FLT_EPSILON_CUBED: f64 = (f32::EPSILON * f32::EPSILON * f32::EPSILON) as f64; |
| 25 | const FLT_EPSILON_INVERSE: f64 = 1.0 / f32::EPSILON as f64; |
| 26 | |
| 27 | pub trait Scalar64 { |
| 28 | fn bound(self, min: Self, max: Self) -> Self; |
| 29 | fn between(self, a: f64, b: f64) -> bool; |
| 30 | fn precisely_zero(self) -> bool; |
| 31 | fn approximately_zero_or_more(self) -> bool; |
| 32 | fn approximately_one_or_less(self) -> bool; |
| 33 | fn approximately_zero(self) -> bool; |
| 34 | fn approximately_zero_inverse(self) -> bool; |
| 35 | fn approximately_zero_cubed(self) -> bool; |
| 36 | fn approximately_zero_half(self) -> bool; |
| 37 | fn approximately_zero_when_compared_to(self, other: Self) -> bool; |
| 38 | fn approximately_equal(self, other: Self) -> bool; |
| 39 | fn approximately_equal_half(self, other: Self) -> bool; |
| 40 | fn almost_dequal_ulps(self, other: Self) -> bool; |
| 41 | } |
| 42 | |
| 43 | impl Scalar64 for f64 { |
| 44 | // Works just like SkTPin, returning `max` for NaN/inf |
| 45 | fn bound(self, min: Self, max: Self) -> Self { |
| 46 | max.min(self).max(min) |
| 47 | } |
| 48 | |
| 49 | /// Returns true if (a <= self <= b) || (a >= self >= b). |
| 50 | fn between(self, a: f64, b: f64) -> bool { |
| 51 | debug_assert!( |
| 52 | ((a <= self && self <= b) || (a >= self && self >= b)) |
| 53 | == ((a - self) * (b - self) <= 0.0) |
| 54 | || (a.precisely_zero() && self.precisely_zero() && b.precisely_zero()) |
| 55 | ); |
| 56 | |
| 57 | (a - self) * (b - self) <= 0.0 |
| 58 | } |
| 59 | |
| 60 | fn precisely_zero(self) -> bool { |
| 61 | self.abs() < DBL_EPSILON_ERR |
| 62 | } |
| 63 | |
| 64 | fn approximately_zero_or_more(self) -> bool { |
| 65 | self > -f64::EPSILON |
| 66 | } |
| 67 | |
| 68 | fn approximately_one_or_less(self) -> bool { |
| 69 | self < 1.0 + f64::EPSILON |
| 70 | } |
| 71 | |
| 72 | fn approximately_zero(self) -> bool { |
| 73 | self.abs() < f64::EPSILON |
| 74 | } |
| 75 | |
| 76 | fn approximately_zero_inverse(self) -> bool { |
| 77 | self.abs() > FLT_EPSILON_INVERSE |
| 78 | } |
| 79 | |
| 80 | fn approximately_zero_cubed(self) -> bool { |
| 81 | self.abs() < FLT_EPSILON_CUBED |
| 82 | } |
| 83 | |
| 84 | fn approximately_zero_half(self) -> bool { |
| 85 | self < FLT_EPSILON_HALF |
| 86 | } |
| 87 | |
| 88 | fn approximately_zero_when_compared_to(self, other: Self) -> bool { |
| 89 | self == 0.0 || self.abs() < (other * (f32::EPSILON as f64)).abs() |
| 90 | } |
| 91 | |
| 92 | // Use this for comparing Ts in the range of 0 to 1. For general numbers (larger and smaller) use |
| 93 | // AlmostEqualUlps instead. |
| 94 | fn approximately_equal(self, other: Self) -> bool { |
| 95 | (self - other).approximately_zero() |
| 96 | } |
| 97 | |
| 98 | fn approximately_equal_half(self, other: Self) -> bool { |
| 99 | (self - other).approximately_zero_half() |
| 100 | } |
| 101 | |
| 102 | fn almost_dequal_ulps(self, other: Self) -> bool { |
| 103 | if self.abs() < SCALAR_MAX as f64 && other.abs() < SCALAR_MAX as f64 { |
| 104 | (self as f32).almost_dequal_ulps(other as f32) |
| 105 | } else { |
| 106 | (self - other).abs() / self.abs().max(other.abs()) < (f32::EPSILON * 16.0) as f64 |
| 107 | } |
| 108 | } |
| 109 | } |
| 110 | |
| 111 | pub fn cube_root(x: f64) -> f64 { |
| 112 | if x.approximately_zero_cubed() { |
| 113 | return 0.0; |
| 114 | } |
| 115 | |
| 116 | let result: f64 = halley_cbrt3d(x.abs()); |
| 117 | if x < 0.0 { |
| 118 | -result |
| 119 | } else { |
| 120 | result |
| 121 | } |
| 122 | } |
| 123 | |
| 124 | // cube root approximation using 3 iterations of Halley's method (double) |
| 125 | fn halley_cbrt3d(d: f64) -> f64 { |
| 126 | let mut a: f64 = cbrt_5d(d); |
| 127 | a = cbrta_halleyd(a, r:d); |
| 128 | a = cbrta_halleyd(a, r:d); |
| 129 | cbrta_halleyd(a, r:d) |
| 130 | } |
| 131 | |
| 132 | // cube root approximation using bit hack for 64-bit float |
| 133 | // adapted from Kahan's cbrt |
| 134 | fn cbrt_5d(d: f64) -> f64 { |
| 135 | let b1: u32 = 715094163; |
| 136 | let mut t: f64 = 0.0; |
| 137 | let pt: &mut [u32; 2] = bytemuck::cast_mut(&mut t); |
| 138 | let px: [u32; 2] = bytemuck::cast(d); |
| 139 | pt[1] = px[1] / 3 + b1; |
| 140 | t |
| 141 | } |
| 142 | |
| 143 | // iterative cube root approximation using Halley's method (double) |
| 144 | fn cbrta_halleyd(a: f64, r: f64) -> f64 { |
| 145 | let a3: f64 = a * a * a; |
| 146 | a * (a3 + r + r) / (a3 + a3 + r) |
| 147 | } |
| 148 | |
| 149 | fn interp(a: f64, b: f64, t: f64) -> f64 { |
| 150 | a + (b - a) * t |
| 151 | } |
| 152 | |