| 1 |
use super::plumbing::*; |
| 2 |
use super::*; |
| 3 |
|
| 4 |
use std::fmt::{self, Debug}; |
| 5 |
|
| 6 |
/// The `split` function takes arbitrary data and a closure that knows how to |
| 7 |
/// split it, and turns this into a `ParallelIterator`. |
| 8 |
/// |
| 9 |
/// # Examples |
| 10 |
/// |
| 11 |
/// As a simple example, Rayon can recursively split ranges of indices |
| 12 |
/// |
| 13 |
/// ``` |
| 14 |
/// use rayon::iter; |
| 15 |
/// use rayon::prelude::*; |
| 16 |
/// use std::ops::Range; |
| 17 |
/// |
| 18 |
/// |
| 19 |
/// // We define a range of indices as follows |
| 20 |
/// type Range1D = Range<usize>; |
| 21 |
/// |
| 22 |
/// // Splitting it in two can be done like this |
| 23 |
/// fn split_range1(r: Range1D) -> (Range1D, Option<Range1D>) { |
| 24 |
/// // We are mathematically unable to split the range if there is only |
| 25 |
/// // one point inside of it, but we could stop splitting before that. |
| 26 |
/// if r.end - r.start <= 1 { return (r, None); } |
| 27 |
/// |
| 28 |
/// // Here, our range is considered large enough to be splittable |
| 29 |
/// let midpoint = r.start + (r.end - r.start) / 2; |
| 30 |
/// (r.start..midpoint, Some(midpoint..r.end)) |
| 31 |
/// } |
| 32 |
/// |
| 33 |
/// // By using iter::split, Rayon will split the range until it has enough work |
| 34 |
/// // to feed the CPU cores, then give us the resulting sub-ranges |
| 35 |
/// iter::split(0..4096, split_range1).for_each(|sub_range| { |
| 36 |
/// // As our initial range had a power-of-two size, the final sub-ranges |
| 37 |
/// // should have power-of-two sizes too |
| 38 |
/// assert!((sub_range.end - sub_range.start).is_power_of_two()); |
| 39 |
/// }); |
| 40 |
/// ``` |
| 41 |
/// |
| 42 |
/// This recursive splitting can be extended to two or three dimensions, |
| 43 |
/// to reproduce a classic "block-wise" parallelization scheme of graphics and |
| 44 |
/// numerical simulations: |
| 45 |
/// |
| 46 |
/// ``` |
| 47 |
/// # use rayon::iter; |
| 48 |
/// # use rayon::prelude::*; |
| 49 |
/// # use std::ops::Range; |
| 50 |
/// # type Range1D = Range<usize>; |
| 51 |
/// # fn split_range1(r: Range1D) -> (Range1D, Option<Range1D>) { |
| 52 |
/// # if r.end - r.start <= 1 { return (r, None); } |
| 53 |
/// # let midpoint = r.start + (r.end - r.start) / 2; |
| 54 |
/// # (r.start..midpoint, Some(midpoint..r.end)) |
| 55 |
/// # } |
| 56 |
/// # |
| 57 |
/// // A two-dimensional range of indices can be built out of two 1D ones |
| 58 |
/// struct Range2D { |
| 59 |
/// // Range of horizontal indices |
| 60 |
/// pub rx: Range1D, |
| 61 |
/// |
| 62 |
/// // Range of vertical indices |
| 63 |
/// pub ry: Range1D, |
| 64 |
/// } |
| 65 |
/// |
| 66 |
/// // We want to recursively split them by the largest dimension until we have |
| 67 |
/// // enough sub-ranges to feed our mighty multi-core CPU. This function |
| 68 |
/// // carries out one such split. |
| 69 |
/// fn split_range2(r2: Range2D) -> (Range2D, Option<Range2D>) { |
| 70 |
/// // Decide on which axis (horizontal/vertical) the range should be split |
| 71 |
/// let width = r2.rx.end - r2.rx.start; |
| 72 |
/// let height = r2.ry.end - r2.ry.start; |
| 73 |
/// if width >= height { |
| 74 |
/// // This is a wide range, split it on the horizontal axis |
| 75 |
/// let (split_rx, ry) = (split_range1(r2.rx), r2.ry); |
| 76 |
/// let out1 = Range2D { |
| 77 |
/// rx: split_rx.0, |
| 78 |
/// ry: ry.clone(), |
| 79 |
/// }; |
| 80 |
/// let out2 = split_rx.1.map(|rx| Range2D { rx, ry }); |
| 81 |
/// (out1, out2) |
| 82 |
/// } else { |
| 83 |
/// // This is a tall range, split it on the vertical axis |
| 84 |
/// let (rx, split_ry) = (r2.rx, split_range1(r2.ry)); |
| 85 |
/// let out1 = Range2D { |
| 86 |
/// rx: rx.clone(), |
| 87 |
/// ry: split_ry.0, |
| 88 |
/// }; |
| 89 |
/// let out2 = split_ry.1.map(|ry| Range2D { rx, ry, }); |
| 90 |
/// (out1, out2) |
| 91 |
/// } |
| 92 |
/// } |
| 93 |
/// |
| 94 |
/// // Again, rayon can handle the recursive splitting for us |
| 95 |
/// let range = Range2D { rx: 0..800, ry: 0..600 }; |
| 96 |
/// iter::split(range, split_range2).for_each(|sub_range| { |
| 97 |
/// // If the sub-ranges were indeed split by the largest dimension, then |
| 98 |
/// // if no dimension was twice larger than the other initially, this |
| 99 |
/// // property will remain true in the final sub-ranges. |
| 100 |
/// let width = sub_range.rx.end - sub_range.rx.start; |
| 101 |
/// let height = sub_range.ry.end - sub_range.ry.start; |
| 102 |
/// assert!((width / 2 <= height) && (height / 2 <= width)); |
| 103 |
/// }); |
| 104 |
/// ``` |
| 105 |
/// |
| 106 |
pub fn split<D, S>(data: D, splitter: S) -> Split<D, S> |
| 107 |
where |
| 108 |
D: Send, |
| 109 |
S: Fn(D) -> (D, Option<D>) + Sync, |
| 110 |
{ |
| 111 |
Split { data, splitter } |
| 112 |
} |
| 113 |
|
| 114 |
/// `Split` is a parallel iterator using arbitrary data and a splitting function. |
| 115 |
/// This struct is created by the [`split()`] function. |
| 116 |
/// |
| 117 |
/// [`split()`]: fn.split.html |
| 118 |
#[derive(Clone)] |
| 119 |
pub struct Split<D, S> { |
| 120 |
data: D, |
| 121 |
splitter: S, |
| 122 |
} |
| 123 |
|
| 124 |
impl<D: Debug, S> Debug for Split<D, S> { |
| 125 |
fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { |
| 126 |
f.debug_struct("Split").field("data", &self.data).finish() |
| 127 |
} |
| 128 |
} |
| 129 |
|
| 130 |
impl<D, S> ParallelIterator for Split<D, S> |
| 131 |
where |
| 132 |
D: Send, |
| 133 |
S: Fn(D) -> (D, Option<D>) + Sync + Send, |
| 134 |
{ |
| 135 |
type Item = D; |
| 136 |
|
| 137 |
fn drive_unindexed<C>(self, consumer: C) -> C::Result |
| 138 |
where |
| 139 |
C: UnindexedConsumer<Self::Item>, |
| 140 |
{ |
| 141 |
let producer = SplitProducer { |
| 142 |
data: self.data, |
| 143 |
splitter: &self.splitter, |
| 144 |
}; |
| 145 |
bridge_unindexed(producer, consumer) |
| 146 |
} |
| 147 |
} |
| 148 |
|
| 149 |
struct SplitProducer<'a, D, S> { |
| 150 |
data: D, |
| 151 |
splitter: &'a S, |
| 152 |
} |
| 153 |
|
| 154 |
impl<'a, D, S> UnindexedProducer for SplitProducer<'a, D, S> |
| 155 |
where |
| 156 |
D: Send, |
| 157 |
S: Fn(D) -> (D, Option<D>) + Sync, |
| 158 |
{ |
| 159 |
type Item = D; |
| 160 |
|
| 161 |
fn split(mut self) -> (Self, Option<Self>) { |
| 162 |
let splitter = self.splitter; |
| 163 |
let (left, right) = splitter(self.data); |
| 164 |
self.data = left; |
| 165 |
(self, right.map(|data| SplitProducer { data, splitter })) |
| 166 |
} |
| 167 |
|
| 168 |
fn fold_with<F>(self, folder: F) -> F |
| 169 |
where |
| 170 |
F: Folder<Self::Item>, |
| 171 |
{ |
| 172 |
folder.consume(self.data) |
| 173 |
} |
| 174 |
} |