Provided Methods

fn by_exponential_blocks(self: Self) -> ExponentialBlocks<Self>

Divides an iterator into sequential blocks of exponentially-increasing size.

Normally, parallel iterators are recursively divided into tasks in parallel. This adaptor changes the default behavior by splitting the iterator into a sequence of parallel iterators of increasing sizes. Sizes grow exponentially in order to avoid creating too many blocks. This also allows to balance the current block with all previous ones.

This can have many applications but the most notable ones are:

• better performance with [find_first()][ParallelIterator::find_first]
• more predictable performance with [find_any()][ParallelIterator::find_any] or any interruptible computation

Examples

use rayon::prelude::*;
assert_eq!((0..10_000).into_par_iter()
                      .by_exponential_blocks()
                      .find_first(|&e| e==4_999), Some(4_999))

In this example, without blocks, rayon will split the initial range into two but all work on the right hand side (from 5,000 onwards) is useless since the sequential algorithm never goes there. This means that if two threads are used there will be no speedup at all.

by_exponential_blocks on the other hand will start with the leftmost range from 0 to p (threads number), continue with p to 3p, the 3p to 7p...

Each subrange is treated in parallel, while all subranges are treated sequentially. We therefore ensure a logarithmic number of blocks (and overhead) while guaranteeing we stop at the first block containing the searched data.

fn by_uniform_blocks(self: Self, block_size: usize) -> UniformBlocks<Self>

Divides an iterator into sequential blocks of the given size.

Normally, parallel iterators are recursively divided into tasks in parallel. This adaptor changes the default behavior by splitting the iterator into a sequence of parallel iterators of given block_size. The main application is to obtain better memory locality (especially if the reduce operation re-use folded data).

Panics if block_size is 0.

Example

use rayon::prelude::*;
// during most reductions v1 and v2 fit the cache
let v = (0u32..10_000_000)
    .into_par_iter()
    .by_uniform_blocks(1_000_000)
    .fold(Vec::new, |mut v, e| { v.push(e); v})
    .reduce(Vec::new, |mut v1, mut v2| { v1.append(&mut v2); v1});
assert_eq!(v, (0u32..10_000_000).collect::<Vec<u32>>());

fn collect_into_vec(self: Self, target: &mut Vec<<Self as >::Item>)

Collects the results of the iterator into the specified vector. The vector is always cleared before execution begins. If possible, reusing the vector across calls can lead to better performance since it reuses the same backing buffer.

Examples

use rayon::prelude::*;

// any prior data will be cleared
let mut vec = vec![-1, -2, -3];

(0..5).into_par_iter()
    .collect_into_vec(&mut vec);

assert_eq!(vec, [0, 1, 2, 3, 4]);

fn unzip_into_vecs<A, B>(self: Self, left: &mut Vec<A>, right: &mut Vec<B>)
where
    Self: IndexedParallelIterator<Item = (A, B)>,
    A: Send,
    B: Send

Unzips the results of the iterator into the specified vectors. The vectors are always cleared before execution begins. If possible, reusing the vectors across calls can lead to better performance since they reuse the same backing buffer.

Examples

use rayon::prelude::*;

// any prior data will be cleared
let mut left = vec![42; 10];
let mut right = vec![-1; 10];

(10..15).into_par_iter()
    .enumerate()
    .unzip_into_vecs(&mut left, &mut right);

assert_eq!(left, [0, 1, 2, 3, 4]);
assert_eq!(right, [10, 11, 12, 13, 14]);

fn zip<Z>(self: Self, zip_op: Z) -> Zip<Self, <Z as >::Iter>
where
    Z: IntoParallelIterator,
    <Z as >::Iter: IndexedParallelIterator

Iterates over tuples (A, B), where the items A are from this iterator and B are from the iterator given as argument. Like the zip method on ordinary iterators, if the two iterators are of unequal length, you only get the items they have in common.

Examples

use rayon::prelude::*;

let result: Vec<_> = (1..4)
    .into_par_iter()
    .zip(vec!['a', 'b', 'c'])
    .collect();

assert_eq!(result, [(1, 'a'), (2, 'b'), (3, 'c')]);

fn zip_eq<Z>(self: Self, zip_op: Z) -> ZipEq<Self, <Z as >::Iter>
where
    Z: IntoParallelIterator,
    <Z as >::Iter: IndexedParallelIterator

The same as Zip, but requires that both iterators have the same length.

Panics

Will panic if self and zip_op are not the same length.

use rayon::prelude::*;

let one = [1u8];
let two = [2u8, 2];
let one_iter = one.par_iter();
let two_iter = two.par_iter();

// this will panic
let zipped: Vec<(&u8, &u8)> = one_iter.zip_eq(two_iter).collect();

// we should never get here
assert_eq!(1, zipped.len());

fn interleave<I>(self: Self, other: I) -> Interleave<Self, <I as >::Iter>
where
    I: IntoParallelIterator<Item = <Self as >::Item>,
    <I as >::Iter: IndexedParallelIterator<Item = <Self as >::Item>

Interleaves elements of this iterator and the other given iterator. Alternately yields elements from this iterator and the given iterator, until both are exhausted. If one iterator is exhausted before the other, the last elements are provided from the other.

Examples

use rayon::prelude::*;
let (x, y) = (vec![1, 2], vec![3, 4, 5, 6]);
let r: Vec<i32> = x.into_par_iter().interleave(y).collect();
assert_eq!(r, vec![1, 3, 2, 4, 5, 6]);

fn interleave_shortest<I>(self: Self, other: I) -> InterleaveShortest<Self, <I as >::Iter>
where
    I: IntoParallelIterator<Item = <Self as >::Item>,
    <I as >::Iter: IndexedParallelIterator<Item = <Self as >::Item>

Interleaves elements of this iterator and the other given iterator, until one is exhausted.

Examples

use rayon::prelude::*;
let (x, y) = (vec![1, 2, 3, 4], vec![5, 6]);
let r: Vec<i32> = x.into_par_iter().interleave_shortest(y).collect();
assert_eq!(r, vec![1, 5, 2, 6, 3]);

fn chunks(self: Self, chunk_size: usize) -> Chunks<Self>

Splits an iterator up into fixed-size chunks.

Returns an iterator that returns Vecs of the given number of elements. If the number of elements in the iterator is not divisible by chunk_size, the last chunk may be shorter than chunk_size.

See also par_chunks() and par_chunks_mut() for similar behavior on slices, without having to allocate intermediate Vecs for the chunks.

Panics if chunk_size is 0.

Examples

use rayon::prelude::*;
let a = vec![1, 2, 3, 4, 5, 6, 7, 8, 9, 10];
let r: Vec<Vec<i32>> = a.into_par_iter().chunks(3).collect();
assert_eq!(r, vec![vec![1,2,3], vec![4,5,6], vec![7,8,9], vec![10]]);

fn fold_chunks<T, ID, F>(self: Self, chunk_size: usize, identity: ID, fold_op: F) -> FoldChunks<Self, ID, F>
where
    ID: Fn() -> T + Send + Sync,
    F: Fn(T, <Self as >::Item) -> T + Send + Sync,
    T: Send

Splits an iterator into fixed-size chunks, performing a sequential fold() on each chunk.

Returns an iterator that produces a folded result for each chunk of items produced by this iterator.

This works essentially like:

iter.chunks(chunk_size)
    .map(|chunk|
        chunk.into_iter()
            .fold(identity, fold_op)
    )

except there is no per-chunk allocation overhead.

Panics if chunk_size is 0.

Examples

use rayon::prelude::*;
let nums = vec![1, 2, 3, 4, 5, 6, 7, 8, 9, 10];
let chunk_sums = nums.into_par_iter().fold_chunks(2, || 0, |a, n| a + n).collect::<Vec<_>>();
assert_eq!(chunk_sums, vec![3, 7, 11, 15, 19]);

fn fold_chunks_with<T, F>(self: Self, chunk_size: usize, init: T, fold_op: F) -> FoldChunksWith<Self, T, F>
where
    T: Send + Clone,
    F: Fn(T, <Self as >::Item) -> T + Send + Sync

Splits an iterator into fixed-size chunks, performing a sequential fold() on each chunk.

Returns an iterator that produces a folded result for each chunk of items produced by this iterator.

This works essentially like fold_chunks(chunk_size, || init.clone(), fold_op), except it doesn't require the init type to be Sync, nor any other form of added synchronization.

Panics if chunk_size is 0.

Examples

use rayon::prelude::*;
let nums = vec![1, 2, 3, 4, 5, 6, 7, 8, 9, 10];
let chunk_sums = nums.into_par_iter().fold_chunks_with(2, 0, |a, n| a + n).collect::<Vec<_>>();
assert_eq!(chunk_sums, vec![3, 7, 11, 15, 19]);

fn cmp<I>(self: Self, other: I) -> Ordering
where
    I: IntoParallelIterator<Item = <Self as >::Item>,
    <I as >::Iter: IndexedParallelIterator,
    <Self as >::Item: Ord

Lexicographically compares the elements of this ParallelIterator with those of another.

Examples

use rayon::prelude::*;
use std::cmp::Ordering::*;

let x = vec![1, 2, 3];
assert_eq!(x.par_iter().cmp(&vec![1, 3, 0]), Less);
assert_eq!(x.par_iter().cmp(&vec![1, 2, 3]), Equal);
assert_eq!(x.par_iter().cmp(&vec![1, 2]), Greater);

fn partial_cmp<I>(self: Self, other: I) -> Option<Ordering>
where
    I: IntoParallelIterator,
    <I as >::Iter: IndexedParallelIterator,
    <Self as >::Item: PartialOrd<<I as >::Item>

Lexicographically compares the elements of this ParallelIterator with those of another.

Examples

use rayon::prelude::*;
use std::cmp::Ordering::*;
use std::f64::NAN;

let x = vec![1.0, 2.0, 3.0];
assert_eq!(x.par_iter().partial_cmp(&vec![1.0, 3.0, 0.0]), Some(Less));
assert_eq!(x.par_iter().partial_cmp(&vec![1.0, 2.0, 3.0]), Some(Equal));
assert_eq!(x.par_iter().partial_cmp(&vec![1.0, 2.0]), Some(Greater));
assert_eq!(x.par_iter().partial_cmp(&vec![1.0, NAN]), None);

fn eq<I>(self: Self, other: I) -> bool
where
    I: IntoParallelIterator,
    <I as >::Iter: IndexedParallelIterator,
    <Self as >::Item: PartialEq<<I as >::Item>

Determines if the elements of this ParallelIterator are equal to those of another

fn ne<I>(self: Self, other: I) -> bool
where
    I: IntoParallelIterator,
    <I as >::Iter: IndexedParallelIterator,
    <Self as >::Item: PartialEq<<I as >::Item>

Determines if the elements of this ParallelIterator are unequal to those of another

fn lt<I>(self: Self, other: I) -> bool
where
    I: IntoParallelIterator,
    <I as >::Iter: IndexedParallelIterator,
    <Self as >::Item: PartialOrd<<I as >::Item>

Determines if the elements of this ParallelIterator are lexicographically less than those of another.

fn le<I>(self: Self, other: I) -> bool
where
    I: IntoParallelIterator,
    <I as >::Iter: IndexedParallelIterator,
    <Self as >::Item: PartialOrd<<I as >::Item>

Determines if the elements of this ParallelIterator are less or equal to those of another.

fn gt<I>(self: Self, other: I) -> bool
where
    I: IntoParallelIterator,
    <I as >::Iter: IndexedParallelIterator,
    <Self as >::Item: PartialOrd<<I as >::Item>

Determines if the elements of this ParallelIterator are lexicographically greater than those of another.

fn ge<I>(self: Self, other: I) -> bool
where
    I: IntoParallelIterator,
    <I as >::Iter: IndexedParallelIterator,
    <Self as >::Item: PartialOrd<<I as >::Item>

Determines if the elements of this ParallelIterator are less or equal to those of another.

fn enumerate(self: Self) -> Enumerate<Self>

Yields an index along with each item.

Examples

use rayon::prelude::*;

let chars = vec!['a', 'b', 'c'];
let result: Vec<_> = chars
    .into_par_iter()
    .enumerate()
    .collect();

assert_eq!(result, [(0, 'a'), (1, 'b'), (2, 'c')]);

fn step_by(self: Self, step: usize) -> StepBy<Self>

Creates an iterator that steps by the given amount

Examples

use rayon::prelude::*;

let range = (3..10);
let result: Vec<i32> = range
   .into_par_iter()
   .step_by(3)
   .collect();

assert_eq!(result, [3, 6, 9])

fn skip(self: Self, n: usize) -> Skip<Self>

Creates an iterator that skips the first n elements.

Examples

use rayon::prelude::*;

let result: Vec<_> = (0..100)
    .into_par_iter()
    .skip(95)
    .collect();

assert_eq!(result, [95, 96, 97, 98, 99]);

fn take(self: Self, n: usize) -> Take<Self>

Creates an iterator that yields the first n elements.

Examples

use rayon::prelude::*;

let result: Vec<_> = (0..100)
    .into_par_iter()
    .take(5)
    .collect();

assert_eq!(result, [0, 1, 2, 3, 4]);

fn position_any<P>(self: Self, predicate: P) -> Option<usize>
where
    P: Fn(<Self as >::Item) -> bool + Sync + Send

Searches for some item in the parallel iterator that matches the given predicate, and returns its index. Like ParallelIterator::find_any, the parallel search will not necessarily find the first match, and once a match is found we'll attempt to stop processing any more.

Examples

use rayon::prelude::*;

let a = [1, 2, 3, 3];

let i = a.par_iter().position_any(|&x| x == 3).expect("found");
assert!(i == 2 || i == 3);

assert_eq!(a.par_iter().position_any(|&x| x == 100), None);

fn position_first<P>(self: Self, predicate: P) -> Option<usize>
where
    P: Fn(<Self as >::Item) -> bool + Sync + Send

Searches for the sequentially first item in the parallel iterator that matches the given predicate, and returns its index.

Like ParallelIterator::find_first, once a match is found, all attempts to the right of the match will be stopped, while attempts to the left must continue in case an earlier match is found.

Note that not all parallel iterators have a useful order, much like sequential HashMap iteration, so "first" may be nebulous. If you just want the first match that discovered anywhere in the iterator, position_any is a better choice.

Examples

use rayon::prelude::*;

let a = [1, 2, 3, 3];

assert_eq!(a.par_iter().position_first(|&x| x == 3), Some(2));

assert_eq!(a.par_iter().position_first(|&x| x == 100), None);

fn position_last<P>(self: Self, predicate: P) -> Option<usize>
where
    P: Fn(<Self as >::Item) -> bool + Sync + Send

Searches for the sequentially last item in the parallel iterator that matches the given predicate, and returns its index.

Like ParallelIterator::find_last, once a match is found, all attempts to the left of the match will be stopped, while attempts to the right must continue in case a later match is found.

Note that not all parallel iterators have a useful order, much like sequential HashMap iteration, so "last" may be nebulous. When the order doesn't actually matter to you, position_any is a better choice.

Examples

use rayon::prelude::*;

let a = [1, 2, 3, 3];

assert_eq!(a.par_iter().position_last(|&x| x == 3), Some(3));

assert_eq!(a.par_iter().position_last(|&x| x == 100), None);

fn positions<P>(self: Self, predicate: P) -> Positions<Self, P>
where
    P: Fn(<Self as >::Item) -> bool + Sync + Send

Searches for items in the parallel iterator that match the given predicate, and returns their indices.

Examples

use rayon::prelude::*;

let primes = vec![2, 3, 5, 7, 11, 13, 17, 19, 23, 29];

// Find the positions of primes congruent to 1 modulo 6
let p1mod6: Vec<_> = primes.par_iter().positions(|&p| p % 6 == 1).collect();
assert_eq!(p1mod6, [3, 5, 7]); // primes 7, 13, and 19

// Find the positions of primes congruent to 5 modulo 6
let p5mod6: Vec<_> = primes.par_iter().positions(|&p| p % 6 == 5).collect();
assert_eq!(p5mod6, [2, 4, 6, 8, 9]); // primes 5, 11, 17, 23, and 29

fn rev(self: Self) -> Rev<Self>

Produces a new iterator with the elements of this iterator in reverse order.

Examples

use rayon::prelude::*;

let result: Vec<_> = (0..5)
    .into_par_iter()
    .rev()
    .collect();

assert_eq!(result, [4, 3, 2, 1, 0]);

fn with_min_len(self: Self, min: usize) -> MinLen<Self>

Sets the minimum length of iterators desired to process in each rayon job. Rayon will not split any smaller than this length, but of course an iterator could already be smaller to begin with.

Producers like zip and interleave will use greater of the two minimums. Chained iterators and iterators inside flat_map may each use their own minimum length.

Examples

use rayon::prelude::*;

let min = (0..1_000_000)
    .into_par_iter()
    .with_min_len(1234)
    .fold(|| 0, |acc, _| acc + 1) // count how many are in this segment
    .min().unwrap();

assert!(min >= 1234);

fn with_max_len(self: Self, max: usize) -> MaxLen<Self>

Sets the maximum length of iterators desired to process in each rayon job. Rayon will try to split at least below this length, unless that would put it below the length from with_min_len(). For example, given min=10 and max=15, a length of 16 will not be split any further.

Producers like zip and interleave will use lesser of the two maximums. Chained iterators and iterators inside flat_map may each use their own maximum length.

Examples

use rayon::prelude::*;

let max = (0..1_000_000)
    .into_par_iter()
    .with_max_len(1234)
    .fold(|| 0, |acc, _| acc + 1) // count how many are in this segment
    .max().unwrap();

assert!(max <= 1234);