Provided Methods

fn for_each<OP>(self: Self, op: OP)
where
    OP: Fn(<Self as >::Item) + Sync + Send

Executes OP on each item produced by the iterator, in parallel.

Examples

use rayon::prelude::*;

(0..100).into_par_iter().for_each(|x| println!("{:?}", x));

fn for_each_with<OP, T>(self: Self, init: T, op: OP)
where
    OP: Fn(&mut T, <Self as >::Item) + Sync + Send,
    T: Send + Clone

Executes OP on the given init value with each item produced by the iterator, in parallel.

The init value will be cloned only as needed to be paired with the group of items in each rayon job. It does not require the type to be Sync.

Examples

use std::sync::mpsc::channel;
use rayon::prelude::*;

let (sender, receiver) = channel();

(0..5).into_par_iter().for_each_with(sender, |s, x| s.send(x).unwrap());

let mut res: Vec<_> = receiver.iter().collect();

res.sort();

assert_eq!(&res[..], &[0, 1, 2, 3, 4])

fn for_each_init<OP, INIT, T>(self: Self, init: INIT, op: OP)
where
    OP: Fn(&mut T, <Self as >::Item) + Sync + Send,
    INIT: Fn() -> T + Sync + Send

Executes OP on a value returned by init with each item produced by the iterator, in parallel.

The init function will be called only as needed for a value to be paired with the group of items in each rayon job. There is no constraint on that returned type at all!

Examples

use rand::Rng;
use rayon::prelude::*;

let mut v = vec![0u8; 1_000_000];

v.par_chunks_mut(1000)
    .for_each_init(
        || rand::thread_rng(),
        |rng, chunk| rng.fill(chunk),
    );

// There's a remote chance that this will fail...
for i in 0u8..=255 {
    assert!(v.contains(&i));
}

fn try_for_each<OP, R>(self: Self, op: OP) -> R
where
    OP: Fn(<Self as >::Item) -> R + Sync + Send,
    R: Try<Output = ()> + Send

Executes a fallible OP on each item produced by the iterator, in parallel.

If the OP returns Result::Err or Option::None, we will attempt to stop processing the rest of the items in the iterator as soon as possible, and we will return that terminating value. Otherwise, we will return an empty Result::Ok(()) or Option::Some(()). If there are multiple errors in parallel, it is not specified which will be returned.

Examples

use rayon::prelude::*;
use std::io::{self, Write};

// This will stop iteration early if there's any write error, like
// having piped output get closed on the other end.
(0..100).into_par_iter()
    .try_for_each(|x| writeln!(io::stdout(), "{:?}", x))
    .expect("expected no write errors");

fn try_for_each_with<OP, T, R>(self: Self, init: T, op: OP) -> R
where
    OP: Fn(&mut T, <Self as >::Item) -> R + Sync + Send,
    T: Send + Clone,
    R: Try<Output = ()> + Send

Executes a fallible OP on the given init value with each item produced by the iterator, in parallel.

This combines the init semantics of for_each_with() and the failure semantics of try_for_each().

Examples

use std::sync::mpsc::channel;
use rayon::prelude::*;

let (sender, receiver) = channel();

(0..5).into_par_iter()
    .try_for_each_with(sender, |s, x| s.send(x))
    .expect("expected no send errors");

let mut res: Vec<_> = receiver.iter().collect();

res.sort();

assert_eq!(&res[..], &[0, 1, 2, 3, 4])

fn try_for_each_init<OP, INIT, T, R>(self: Self, init: INIT, op: OP) -> R
where
    OP: Fn(&mut T, <Self as >::Item) -> R + Sync + Send,
    INIT: Fn() -> T + Sync + Send,
    R: Try<Output = ()> + Send

Executes a fallible OP on a value returned by init with each item produced by the iterator, in parallel.

This combines the init semantics of for_each_init() and the failure semantics of try_for_each().

Examples

use rand::Rng;
use rayon::prelude::*;

let mut v = vec![0u8; 1_000_000];

v.par_chunks_mut(1000)
    .try_for_each_init(
        || rand::thread_rng(),
        |rng, chunk| rng.try_fill(chunk),
    )
    .expect("expected no rand errors");

// There's a remote chance that this will fail...
for i in 0u8..=255 {
    assert!(v.contains(&i));
}

fn count(self: Self) -> usize

Counts the number of items in this parallel iterator.

Examples

use rayon::prelude::*;

let count = (0..100).into_par_iter().count();

assert_eq!(count, 100);

fn map<F, R>(self: Self, map_op: F) -> Map<Self, F>
where
    F: Fn(<Self as >::Item) -> R + Sync + Send,
    R: Send

Applies map_op to each item of this iterator, producing a new iterator with the results.

Examples

use rayon::prelude::*;

let mut par_iter = (0..5).into_par_iter().map(|x| x * 2);

let doubles: Vec<_> = par_iter.collect();

assert_eq!(&doubles[..], &[0, 2, 4, 6, 8]);

fn map_with<F, T, R>(self: Self, init: T, map_op: F) -> MapWith<Self, T, F>
where
    F: Fn(&mut T, <Self as >::Item) -> R + Sync + Send,
    T: Send + Clone,
    R: Send

Applies map_op to the given init value with each item of this iterator, producing a new iterator with the results.

The init value will be cloned only as needed to be paired with the group of items in each rayon job. It does not require the type to be Sync.

Examples

use std::sync::mpsc::channel;
use rayon::prelude::*;

let (sender, receiver) = channel();

let a: Vec<_> = (0..5)
                .into_par_iter()            // iterating over i32
                .map_with(sender, |s, x| {
                    s.send(x).unwrap();     // sending i32 values through the channel
                    x                       // returning i32
                })
                .collect();                 // collecting the returned values into a vector

let mut b: Vec<_> = receiver.iter()         // iterating over the values in the channel
                            .collect();     // and collecting them
b.sort();

assert_eq!(a, b);

fn map_init<F, INIT, T, R>(self: Self, init: INIT, map_op: F) -> MapInit<Self, INIT, F>
where
    F: Fn(&mut T, <Self as >::Item) -> R + Sync + Send,
    INIT: Fn() -> T + Sync + Send,
    R: Send

Applies map_op to a value returned by init with each item of this iterator, producing a new iterator with the results.

The init function will be called only as needed for a value to be paired with the group of items in each rayon job. There is no constraint on that returned type at all!

Examples

use rand::Rng;
use rayon::prelude::*;

let a: Vec<_> = (1i32..1_000_000)
    .into_par_iter()
    .map_init(
        || rand::thread_rng(),  // get the thread-local RNG
        |rng, x| if rng.gen() { // randomly negate items
            -x
        } else {
            x
        },
    ).collect();

// There's a remote chance that this will fail...
assert!(a.iter().any(|&x| x < 0));
assert!(a.iter().any(|&x| x > 0));

fn cloned<'a, T>(self: Self) -> Cloned<Self>
where
    T: 'a + Clone + Send,
    Self: ParallelIterator<Item = &'a T>

Creates an iterator which clones all of its elements. This may be useful when you have an iterator over &T, but you need T, and that type implements Clone. See also copied().

Examples

use rayon::prelude::*;

let a = [1, 2, 3];

let v_cloned: Vec<_> = a.par_iter().cloned().collect();

// cloned is the same as .map(|&x| x), for integers
let v_map: Vec<_> = a.par_iter().map(|&x| x).collect();

assert_eq!(v_cloned, vec![1, 2, 3]);
assert_eq!(v_map, vec![1, 2, 3]);

fn copied<'a, T>(self: Self) -> Copied<Self>
where
    T: 'a + Copy + Send,
    Self: ParallelIterator<Item = &'a T>

Creates an iterator which copies all of its elements. This may be useful when you have an iterator over &T, but you need T, and that type implements Copy. See also cloned().

Examples

use rayon::prelude::*;

let a = [1, 2, 3];

let v_copied: Vec<_> = a.par_iter().copied().collect();

// copied is the same as .map(|&x| x), for integers
let v_map: Vec<_> = a.par_iter().map(|&x| x).collect();

assert_eq!(v_copied, vec![1, 2, 3]);
assert_eq!(v_map, vec![1, 2, 3]);

fn inspect<OP>(self: Self, inspect_op: OP) -> Inspect<Self, OP>
where
    OP: Fn(&<Self as >::Item) + Sync + Send

Applies inspect_op to a reference to each item of this iterator, producing a new iterator passing through the original items. This is often useful for debugging to see what's happening in iterator stages.

Examples

use rayon::prelude::*;

let a = [1, 4, 2, 3];

// this iterator sequence is complex.
let sum = a.par_iter()
            .cloned()
            .filter(|&x| x % 2 == 0)
            .reduce(|| 0, |sum, i| sum + i);

println!("{}", sum);

// let's add some inspect() calls to investigate what's happening
let sum = a.par_iter()
            .cloned()
            .inspect(|x| println!("about to filter: {}", x))
            .filter(|&x| x % 2 == 0)
            .inspect(|x| println!("made it through filter: {}", x))
            .reduce(|| 0, |sum, i| sum + i);

println!("{}", sum);

fn update<F>(self: Self, update_op: F) -> Update<Self, F>
where
    F: Fn(&mut <Self as >::Item) + Sync + Send

Mutates each item of this iterator before yielding it.

Examples

use rayon::prelude::*;

let par_iter = (0..5).into_par_iter().update(|x| {*x *= 2;});

let doubles: Vec<_> = par_iter.collect();

assert_eq!(&doubles[..], &[0, 2, 4, 6, 8]);

fn filter<P>(self: Self, filter_op: P) -> Filter<Self, P>
where
    P: Fn(&<Self as >::Item) -> bool + Sync + Send

Applies filter_op to each item of this iterator, producing a new iterator with only the items that gave true results.

Examples

use rayon::prelude::*;

let mut par_iter = (0..10).into_par_iter().filter(|x| x % 2 == 0);

let even_numbers: Vec<_> = par_iter.collect();

assert_eq!(&even_numbers[..], &[0, 2, 4, 6, 8]);

fn filter_map<P, R>(self: Self, filter_op: P) -> FilterMap<Self, P>
where
    P: Fn(<Self as >::Item) -> Option<R> + Sync + Send,
    R: Send

Applies filter_op to each item of this iterator to get an Option, producing a new iterator with only the items from Some results.

Examples

use rayon::prelude::*;

let mut par_iter = (0..10).into_par_iter()
                        .filter_map(|x| {
                            if x % 2 == 0 { Some(x * 3) }
                            else { None }
                        });

let even_numbers: Vec<_> = par_iter.collect();

assert_eq!(&even_numbers[..], &[0, 6, 12, 18, 24]);

fn flat_map<F, PI>(self: Self, map_op: F) -> FlatMap<Self, F>
where
    F: Fn(<Self as >::Item) -> PI + Sync + Send,
    PI: IntoParallelIterator

Applies map_op to each item of this iterator to get nested parallel iterators, producing a new parallel iterator that flattens these back into one.

See also flat_map_iter.

Examples

use rayon::prelude::*;

let a = [[1, 2], [3, 4], [5, 6], [7, 8]];

let par_iter = a.par_iter().cloned().flat_map(|a| a.to_vec());

let vec: Vec<_> = par_iter.collect();

assert_eq!(&vec[..], &[1, 2, 3, 4, 5, 6, 7, 8]);

fn flat_map_iter<F, SI>(self: Self, map_op: F) -> FlatMapIter<Self, F>
where
    F: Fn(<Self as >::Item) -> SI + Sync + Send,
    SI: IntoIterator,
    <SI as >::Item: Send

Applies map_op to each item of this iterator to get nested serial iterators, producing a new parallel iterator that flattens these back into one.

flat_map_iter versus flat_map

These two methods are similar but behave slightly differently. With flat_map, each of the nested iterators must be a parallel iterator, and they will be further split up with nested parallelism. With flat_map_iter, each nested iterator is a sequential Iterator, and we only parallelize between them, while the items produced by each nested iterator are processed sequentially.

When choosing between these methods, consider whether nested parallelism suits the potential iterators at hand. If there's little computation involved, or its length is much less than the outer parallel iterator, then it may perform better to avoid the overhead of parallelism, just flattening sequentially with flat_map_iter. If there is a lot of computation, potentially outweighing the outer parallel iterator, then the nested parallelism of flat_map may be worthwhile.

Examples

use rayon::prelude::*;
use std::cell::RefCell;

let a = [[1, 2], [3, 4], [5, 6], [7, 8]];

let par_iter = a.par_iter().flat_map_iter(|a| {
    // The serial iterator doesn't have to be thread-safe, just its items.
    let cell_iter = RefCell::new(a.iter().cloned());
    std::iter::from_fn(move || cell_iter.borrow_mut().next())
});

let vec: Vec<_> = par_iter.collect();

assert_eq!(&vec[..], &[1, 2, 3, 4, 5, 6, 7, 8]);

fn flatten(self: Self) -> Flatten<Self>
where
    <Self as >::Item: IntoParallelIterator

An adaptor that flattens parallel-iterable Items into one large iterator.

See also flatten_iter.

Examples

use rayon::prelude::*;

let x: Vec<Vec<_>> = vec![vec![1, 2], vec![3, 4]];
let y: Vec<_> = x.into_par_iter().flatten().collect();

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

fn flatten_iter(self: Self) -> FlattenIter<Self>
where
    <Self as >::Item: IntoIterator,
    <<Self as >::Item as IntoIterator>::Item: Send

An adaptor that flattens serial-iterable Items into one large iterator.

See also flatten and the analogous comparison of flat_map_iter versus flat_map.

Examples

use rayon::prelude::*;

let x: Vec<Vec<_>> = vec![vec![1, 2], vec![3, 4]];
let iters: Vec<_> = x.into_iter().map(Vec::into_iter).collect();
let y: Vec<_> = iters.into_par_iter().flatten_iter().collect();

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

fn reduce<OP, ID>(self: Self, identity: ID, op: OP) -> <Self as >::Item
where
    OP: Fn(<Self as >::Item, <Self as >::Item) -> <Self as >::Item + Sync + Send,
    ID: Fn() -> <Self as >::Item + Sync + Send

Reduces the items in the iterator into one item using op. The argument identity should be a closure that can produce "identity" value which may be inserted into the sequence as needed to create opportunities for parallel execution. So, for example, if you are doing a summation, then identity() ought to produce something that represents the zero for your type (but consider just calling sum() in that case).

Examples

// Iterate over a sequence of pairs `(x0, y0), ..., (xN, yN)`
// and use reduce to compute one pair `(x0 + ... + xN, y0 + ... + yN)`
// where the first/second elements are summed separately.
use rayon::prelude::*;
let sums = [(0, 1), (5, 6), (16, 2), (8, 9)]
           .par_iter()        // iterating over &(i32, i32)
           .cloned()          // iterating over (i32, i32)
           .reduce(|| (0, 0), // the "identity" is 0 in both columns
                   |a, b| (a.0 + b.0, a.1 + b.1));
assert_eq!(sums, (0 + 5 + 16 + 8, 1 + 6 + 2 + 9));

Note: unlike a sequential fold operation, the order in which op will be applied to reduce the result is not fully specified. So op should be associative or else the results will be non-deterministic. And of course identity() should produce a true identity.

fn reduce_with<OP>(self: Self, op: OP) -> Option<<Self as >::Item>
where
    OP: Fn(<Self as >::Item, <Self as >::Item) -> <Self as >::Item + Sync + Send

Reduces the items in the iterator into one item using op. If the iterator is empty, None is returned; otherwise, Some is returned.

This version of reduce is simple but somewhat less efficient. If possible, it is better to call reduce(), which requires an identity element.

Examples

use rayon::prelude::*;
let sums = [(0, 1), (5, 6), (16, 2), (8, 9)]
           .par_iter()        // iterating over &(i32, i32)
           .cloned()          // iterating over (i32, i32)
           .reduce_with(|a, b| (a.0 + b.0, a.1 + b.1))
           .unwrap();
assert_eq!(sums, (0 + 5 + 16 + 8, 1 + 6 + 2 + 9));

Note: unlike a sequential fold operation, the order in which op will be applied to reduce the result is not fully specified. So op should be associative or else the results will be non-deterministic.

fn try_reduce<T, OP, ID>(self: Self, identity: ID, op: OP) -> <Self as >::Item
where
    OP: Fn(T, T) -> <Self as >::Item + Sync + Send,
    ID: Fn() -> T + Sync + Send,
    <Self as >::Item: Try<Output = T>

Reduces the items in the iterator into one item using a fallible op. The identity argument is used the same way as in reduce().

If a Result::Err or Option::None item is found, or if op reduces to one, we will attempt to stop processing the rest of the items in the iterator as soon as possible, and we will return that terminating value. Otherwise, we will return the final reduced Result::Ok(T) or Option::Some(T). If there are multiple errors in parallel, it is not specified which will be returned.

Examples

use rayon::prelude::*;

// Compute the sum of squares, being careful about overflow.
fn sum_squares<I: IntoParallelIterator<Item = i32>>(iter: I) -> Option<i32> {
    iter.into_par_iter()
        .map(|i| i.checked_mul(i))            // square each item,
        .try_reduce(|| 0, i32::checked_add)   // and add them up!
}
assert_eq!(sum_squares(0..5), Some(0 + 1 + 4 + 9 + 16));

// The sum might overflow
assert_eq!(sum_squares(0..10_000), None);

// Or the squares might overflow before it even reaches `try_reduce`
assert_eq!(sum_squares(1_000_000..1_000_001), None);

fn try_reduce_with<T, OP>(self: Self, op: OP) -> Option<<Self as >::Item>
where
    OP: Fn(T, T) -> <Self as >::Item + Sync + Send,
    <Self as >::Item: Try<Output = T>

Reduces the items in the iterator into one item using a fallible op.

Like reduce_with(), if the iterator is empty, None is returned; otherwise, Some is returned. Beyond that, it behaves like try_reduce() for handling Err/None.

For instance, with Option items, the return value may be:

• None, the iterator was empty
• Some(None), we stopped after encountering None.
• Some(Some(x)), the entire iterator reduced to x.

With Result items, the nesting is more obvious:

• None, the iterator was empty
• Some(Err(e)), we stopped after encountering an error e.
• Some(Ok(x)), the entire iterator reduced to x.

Examples

use rayon::prelude::*;

let files = ["/dev/null", "/does/not/exist"];

// Find the biggest file
files.into_par_iter()
    .map(|path| std::fs::metadata(path).map(|m| (path, m.len())))
    .try_reduce_with(|a, b| {
        Ok(if a.1 >= b.1 { a } else { b })
    })
    .expect("Some value, since the iterator is not empty")
    .expect_err("not found");

fn fold<T, ID, F>(self: Self, identity: ID, fold_op: F) -> Fold<Self, ID, F>
where
    F: Fn(T, <Self as >::Item) -> T + Sync + Send,
    ID: Fn() -> T + Sync + Send,
    T: Send

Parallel fold is similar to sequential fold except that the sequence of items may be subdivided before it is folded. Consider a list of numbers like 22 3 77 89 46. If you used sequential fold to add them (fold(0, |a,b| a+b), you would wind up first adding 0 + 22, then 22 + 3, then 25 + 77, and so forth. The parallel fold works similarly except that it first breaks up your list into sublists, and hence instead of yielding up a single sum at the end, it yields up multiple sums. The number of results is nondeterministic, as is the point where the breaks occur.

So if we did the same parallel fold (fold(0, |a,b| a+b)) on our example list, we might wind up with a sequence of two numbers, like so:

22 3 77 89 46
      |     |
    102   135

Or perhaps these three numbers:

22 3 77 89 46
      |  |  |
    102 89 46

In general, Rayon will attempt to find good breaking points that keep all of your cores busy.

Fold versus reduce

The fold() and reduce() methods each take an identity element and a combining function, but they operate rather differently.

reduce() requires that the identity function has the same type as the things you are iterating over, and it fully reduces the list of items into a single item. So, for example, imagine we are iterating over a list of bytes bytes: [128_u8, 64_u8, 64_u8]. If we used bytes.reduce(|| 0_u8, |a: u8, b: u8| a + b), we would get an overflow. This is because 0, a, and b here are all bytes, just like the numbers in the list (I wrote the types explicitly above, but those are the only types you can use). To avoid the overflow, we would need to do something like bytes.map(|b| b as u32).reduce(|| 0, |a, b| a + b), in which case our result would be 256.

In contrast, with fold(), the identity function does not have to have the same type as the things you are iterating over, and you potentially get back many results. So, if we continue with the bytes example from the previous paragraph, we could do bytes.fold(|| 0_u32, |a, b| a + (b as u32)) to convert our bytes into u32. And of course we might not get back a single sum.

There is a more subtle distinction as well, though it's actually implied by the above points. When you use reduce(), your reduction function is sometimes called with values that were never part of your original parallel iterator (for example, both the left and right might be a partial sum). With fold(), in contrast, the left value in the fold function is always the accumulator, and the right value is always from your original sequence.

Fold vs Map/Reduce

Fold makes sense if you have some operation where it is cheaper to create groups of elements at a time. For example, imagine collecting characters into a string. If you were going to use map/reduce, you might try this:

use rayon::prelude::*;

let s =
    ['a', 'b', 'c', 'd', 'e']
    .par_iter()
    .map(|c: &char| format!("{}", c))
    .reduce(|| String::new(),
            |mut a: String, b: String| { a.push_str(&b); a });

assert_eq!(s, "abcde");

Because reduce produces the same type of element as its input, you have to first map each character into a string, and then you can reduce them. This means we create one string per element in our iterator -- not so great. Using fold, we can do this instead:

use rayon::prelude::*;

let s =
    ['a', 'b', 'c', 'd', 'e']
    .par_iter()
    .fold(|| String::new(),
            |mut s: String, c: &char| { s.push(*c); s })
    .reduce(|| String::new(),
            |mut a: String, b: String| { a.push_str(&b); a });

assert_eq!(s, "abcde");

Now fold will process groups of our characters at a time, and we only make one string per group. We should wind up with some small-ish number of strings roughly proportional to the number of CPUs you have (it will ultimately depend on how busy your processors are). Note that we still need to do a reduce afterwards to combine those groups of strings into a single string.

You could use a similar trick to save partial results (e.g., a cache) or something similar.

Combining fold with other operations

You can combine fold with reduce if you want to produce a single value. This is then roughly equivalent to a map/reduce combination in effect:

use rayon::prelude::*;

let bytes = 0..22_u8;
let sum = bytes.into_par_iter()
               .fold(|| 0_u32, |a: u32, b: u8| a + (b as u32))
               .sum::<u32>();

assert_eq!(sum, (0..22).sum()); // compare to sequential

fn fold_with<F, T>(self: Self, init: T, fold_op: F) -> FoldWith<Self, T, F>
where
    F: Fn(T, <Self as >::Item) -> T + Sync + Send,
    T: Send + Clone

Applies fold_op to the given init value with each item of this iterator, finally producing the value for further use.

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

Examples

use rayon::prelude::*;

let bytes = 0..22_u8;
let sum = bytes.into_par_iter()
               .fold_with(0_u32, |a: u32, b: u8| a + (b as u32))
               .sum::<u32>();

assert_eq!(sum, (0..22).sum()); // compare to sequential

fn try_fold<T, R, ID, F>(self: Self, identity: ID, fold_op: F) -> TryFold<Self, R, ID, F>
where
    F: Fn(T, <Self as >::Item) -> R + Sync + Send,
    ID: Fn() -> T + Sync + Send,
    R: Try<Output = T> + Send

Performs a fallible parallel fold.

This is a variation of fold() for operations which can fail with Option::None or Result::Err. The first such failure stops processing the local set of items, without affecting other folds in the iterator's subdivisions.

Often, try_fold() will be followed by try_reduce() for a final reduction and global short-circuiting effect.

Examples

use rayon::prelude::*;

let bytes = 0..22_u8;
let sum = bytes.into_par_iter()
               .try_fold(|| 0_u32, |a: u32, b: u8| a.checked_add(b as u32))
               .try_reduce(|| 0, u32::checked_add);

assert_eq!(sum, Some((0..22).sum())); // compare to sequential

fn try_fold_with<F, T, R>(self: Self, init: T, fold_op: F) -> TryFoldWith<Self, R, F>
where
    F: Fn(T, <Self as >::Item) -> R + Sync + Send,
    R: Try<Output = T> + Send,
    T: Clone + Send

Performs a fallible parallel fold with a cloneable init value.

This combines the init semantics of fold_with() and the failure semantics of try_fold().

use rayon::prelude::*;

let bytes = 0..22_u8;
let sum = bytes.into_par_iter()
               .try_fold_with(0_u32, |a: u32, b: u8| a.checked_add(b as u32))
               .try_reduce(|| 0, u32::checked_add);

assert_eq!(sum, Some((0..22).sum())); // compare to sequential

fn sum<S>(self: Self) -> S
where
    S: Send + Sum<<Self as >::Item> + Sum<S>

Sums up the items in the iterator.

Note that the order in items will be reduced is not specified, so if the + operator is not truly associative (as is the case for floating point numbers), then the results are not fully deterministic.

Basically equivalent to self.reduce(|| 0, |a, b| a + b), except that the type of 0 and the + operation may vary depending on the type of value being produced.

Examples

use rayon::prelude::*;

let a = [1, 5, 7];

let sum: i32 = a.par_iter().sum();

assert_eq!(sum, 13);

fn product<P>(self: Self) -> P
where
    P: Send + Product<<Self as >::Item> + Product<P>

Multiplies all the items in the iterator.

Note that the order in items will be reduced is not specified, so if the * operator is not truly associative (as is the case for floating point numbers), then the results are not fully deterministic.

Basically equivalent to self.reduce(|| 1, |a, b| a * b), except that the type of 1 and the * operation may vary depending on the type of value being produced.

Examples

use rayon::prelude::*;

fn factorial(n: u32) -> u32 {
   (1..n+1).into_par_iter().product()
}

assert_eq!(factorial(0), 1);
assert_eq!(factorial(1), 1);
assert_eq!(factorial(5), 120);

fn min(self: Self) -> Option<<Self as >::Item>
where
    <Self as >::Item: Ord

Computes the minimum of all the items in the iterator. If the iterator is empty, None is returned; otherwise, Some(min) is returned.

Note that the order in which the items will be reduced is not specified, so if the Ord impl is not truly associative, then the results are not deterministic.

Basically equivalent to self.reduce_with(|a, b| Ord::min(a, b)).

Examples

use rayon::prelude::*;

let a = [45, 74, 32];

assert_eq!(a.par_iter().min(), Some(&32));

let b: [i32; 0] = [];

assert_eq!(b.par_iter().min(), None);

fn min_by<F>(self: Self, f: F) -> Option<<Self as >::Item>
where
    F: Sync + Send + Fn(&<Self as >::Item, &<Self as >::Item) -> Ordering

Computes the minimum of all the items in the iterator with respect to the given comparison function. If the iterator is empty, None is returned; otherwise, Some(min) is returned.

Note that the order in which the items will be reduced is not specified, so if the comparison function is not associative, then the results are not deterministic.

Examples

use rayon::prelude::*;

let a = [-3_i32, 77, 53, 240, -1];

assert_eq!(a.par_iter().min_by(|x, y| x.cmp(y)), Some(&-3));

fn min_by_key<K, F>(self: Self, f: F) -> Option<<Self as >::Item>
where
    K: Ord + Send,
    F: Sync + Send + Fn(&<Self as >::Item) -> K

Computes the item that yields the minimum value for the given function. If the iterator is empty, None is returned; otherwise, Some(item) is returned.

Note that the order in which the items will be reduced is not specified, so if the Ord impl is not truly associative, then the results are not deterministic.

Examples

use rayon::prelude::*;

let a = [-3_i32, 34, 2, 5, -10, -3, -23];

assert_eq!(a.par_iter().min_by_key(|x| x.abs()), Some(&2));

fn max(self: Self) -> Option<<Self as >::Item>
where
    <Self as >::Item: Ord

Computes the maximum of all the items in the iterator. If the iterator is empty, None is returned; otherwise, Some(max) is returned.

Note that the order in which the items will be reduced is not specified, so if the Ord impl is not truly associative, then the results are not deterministic.

Basically equivalent to self.reduce_with(|a, b| Ord::max(a, b)).

Examples

use rayon::prelude::*;

let a = [45, 74, 32];

assert_eq!(a.par_iter().max(), Some(&74));

let b: [i32; 0] = [];

assert_eq!(b.par_iter().max(), None);

fn max_by<F>(self: Self, f: F) -> Option<<Self as >::Item>
where
    F: Sync + Send + Fn(&<Self as >::Item, &<Self as >::Item) -> Ordering

Computes the maximum of all the items in the iterator with respect to the given comparison function. If the iterator is empty, None is returned; otherwise, Some(max) is returned.

Note that the order in which the items will be reduced is not specified, so if the comparison function is not associative, then the results are not deterministic.

Examples

use rayon::prelude::*;

let a = [-3_i32, 77, 53, 240, -1];

assert_eq!(a.par_iter().max_by(|x, y| x.abs().cmp(&y.abs())), Some(&240));

fn max_by_key<K, F>(self: Self, f: F) -> Option<<Self as >::Item>
where
    K: Ord + Send,
    F: Sync + Send + Fn(&<Self as >::Item) -> K

Computes the item that yields the maximum value for the given function. If the iterator is empty, None is returned; otherwise, Some(item) is returned.

Note that the order in which the items will be reduced is not specified, so if the Ord impl is not truly associative, then the results are not deterministic.

Examples

use rayon::prelude::*;

let a = [-3_i32, 34, 2, 5, -10, -3, -23];

assert_eq!(a.par_iter().max_by_key(|x| x.abs()), Some(&34));

fn chain<C>(self: Self, chain: C) -> Chain<Self, <C as >::Iter>
where
    C: IntoParallelIterator<Item = <Self as >::Item>

Takes two iterators and creates a new iterator over both.

Examples

use rayon::prelude::*;

let a = [0, 1, 2];
let b = [9, 8, 7];

let par_iter = a.par_iter().chain(b.par_iter());

let chained: Vec<_> = par_iter.cloned().collect();

assert_eq!(&chained[..], &[0, 1, 2, 9, 8, 7]);

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

Searches for some item in the parallel iterator that matches the given predicate and returns it. This operation is similar to find on sequential iterators but the item returned may not be the first one in the parallel sequence which matches, since we search the entire sequence in parallel.

Once a match is found, we will attempt to stop processing the rest of the items in the iterator as soon as possible (just as find stops iterating once a match is found).

Examples

use rayon::prelude::*;

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

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

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

fn find_first<P>(self: Self, predicate: P) -> Option<<Self as >::Item>
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 it.

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.

For added performance, you might consider using find_first in conjunction with [by_exponential_blocks()][IndexedParallelIterator::by_exponential_blocks].

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, find_any is a better choice.

Examples

use rayon::prelude::*;

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

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

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

fn find_last<P>(self: Self, predicate: P) -> Option<<Self as >::Item>
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 it.

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, find_any is a better choice.

Examples

use rayon::prelude::*;

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

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

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

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

Applies the given predicate to the items in the parallel iterator and returns any non-None result of the map operation.

Once a non-None value is produced from the map operation, we will attempt to stop processing the rest of the items in the iterator as soon as possible.

Note that this method only returns some item in the parallel iterator that is not None from the map predicate. The item returned may not be the first non-None value produced in the parallel sequence, since the entire sequence is mapped over in parallel.

Examples

use rayon::prelude::*;

let c = ["lol", "NaN", "5", "5"];

let found_number = c.par_iter().find_map_any(|s| s.parse().ok());

assert_eq!(found_number, Some(5));

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

Applies the given predicate to the items in the parallel iterator and returns the sequentially first non-None result of the map operation.

Once a non-None value is produced from the map operation, 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 non-None value discovered anywhere in the iterator, find_map_any is a better choice.

Examples

use rayon::prelude::*;

let c = ["lol", "NaN", "2", "5"];

let first_number = c.par_iter().find_map_first(|s| s.parse().ok());

assert_eq!(first_number, Some(2));

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

Applies the given predicate to the items in the parallel iterator and returns the sequentially last non-None result of the map operation.

Once a non-None value is produced from the map operation, 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 "first" may be nebulous. If you just want the first non-None value discovered anywhere in the iterator, find_map_any is a better choice.

Examples

use rayon::prelude::*;

let c = ["lol", "NaN", "2", "5"];

let last_number = c.par_iter().find_map_last(|s| s.parse().ok());

assert_eq!(last_number, Some(5));

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

Searches for some item in the parallel iterator that matches the given predicate, and if so returns true. Once a match is found, we'll attempt to stop process the rest of the items. Proving that there's no match, returning false, does require visiting every item.

Examples

use rayon::prelude::*;

let a = [0, 12, 3, 4, 0, 23, 0];

let is_valid = a.par_iter().any(|&x| x > 10);

assert!(is_valid);

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

Tests that every item in the parallel iterator matches the given predicate, and if so returns true. If a counter-example is found, we'll attempt to stop processing more items, then return false.

Examples

use rayon::prelude::*;

let a = [0, 12, 3, 4, 0, 23, 0];

let is_valid = a.par_iter().all(|&x| x > 10);

assert!(!is_valid);

fn while_some<T>(self: Self) -> WhileSome<Self>
where
    Self: ParallelIterator<Item = Option<T>>,
    T: Send

Creates an iterator over the Some items of this iterator, halting as soon as any None is found.

Examples

use rayon::prelude::*;
use std::sync::atomic::{AtomicUsize, Ordering};

let counter = AtomicUsize::new(0);
let value = (0_i32..2048)
    .into_par_iter()
    .map(|x| {
             counter.fetch_add(1, Ordering::SeqCst);
             if x < 1024 { Some(x) } else { None }
         })
    .while_some()
    .max();

assert!(value < Some(1024));
assert!(counter.load(Ordering::SeqCst) < 2048); // should not have visited every single one

fn panic_fuse(self: Self) -> PanicFuse<Self>

Wraps an iterator with a fuse in case of panics, to halt all threads as soon as possible.

Panics within parallel iterators are always propagated to the caller, but they don't always halt the rest of the iterator right away, due to the internal semantics of join. This adaptor makes a greater effort to stop processing other items sooner, with the cost of additional synchronization overhead, which may also inhibit some optimizations.

Examples

If this code didn't use panic_fuse(), it would continue processing many more items in other threads (with long sleep delays) before the panic is finally propagated.

use rayon::prelude::*;
use std::{thread, time};

(0..1_000_000)
    .into_par_iter()
    .panic_fuse()
    .for_each(|i| {
        // simulate some work
        thread::sleep(time::Duration::from_secs(1));
        assert!(i > 0); // oops!
    });

fn collect<C>(self: Self) -> C
where
    C: FromParallelIterator<<Self as >::Item>

Creates a fresh collection containing all the elements produced by this parallel iterator.

You may prefer collect_into_vec() implemented on IndexedParallelIterator, if your underlying iterator also implements it. collect_into_vec() allocates efficiently with precise knowledge of how many elements the iterator contains, and even allows you to reuse an existing vector's backing store rather than allocating a fresh vector.

See also [collect_vec_list()][Self::collect_vec_list] for collecting into a LinkedList<Vec<T>>.

Examples

use rayon::prelude::*;

let sync_vec: Vec<_> = (0..100).into_iter().collect();

let async_vec: Vec<_> = (0..100).into_par_iter().collect();

assert_eq!(sync_vec, async_vec);

You can collect a pair of collections like unzip for paired items:

use rayon::prelude::*;

let a = [(0, 1), (1, 2), (2, 3), (3, 4)];
let (first, second): (Vec<_>, Vec<_>) = a.into_par_iter().collect();

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

Or like partition_map for Either items:

use rayon::prelude::*;
use rayon::iter::Either;

let (left, right): (Vec<_>, Vec<_>) = (0..8).into_par_iter().map(|x| {
    if x % 2 == 0 {
        Either::Left(x * 4)
    } else {
        Either::Right(x * 3)
    }
}).collect();

assert_eq!(left, [0, 8, 16, 24]);
assert_eq!(right, [3, 9, 15, 21]);

You can even collect an arbitrarily-nested combination of pairs and Either:

use rayon::prelude::*;
use rayon::iter::Either;

let (first, (left, right)): (Vec<_>, (Vec<_>, Vec<_>))
    = (0..8).into_par_iter().map(|x| {
        if x % 2 == 0 {
            (x, Either::Left(x * 4))
        } else {
            (-x, Either::Right(x * 3))
        }
    }).collect();

assert_eq!(first, [0, -1, 2, -3, 4, -5, 6, -7]);
assert_eq!(left, [0, 8, 16, 24]);
assert_eq!(right, [3, 9, 15, 21]);

All of that can also be combined with short-circuiting collection of Result or Option types:

use rayon::prelude::*;
use rayon::iter::Either;

let result: Result<(Vec<_>, (Vec<_>, Vec<_>)), _>
    = (0..8).into_par_iter().map(|x| {
        if x > 5 {
            Err(x)
        } else if x % 2 == 0 {
            Ok((x, Either::Left(x * 4)))
        } else {
            Ok((-x, Either::Right(x * 3)))
        }
    }).collect();

let error = result.unwrap_err();
assert!(error == 6 || error == 7);

fn unzip<A, B, FromA, FromB>(self: Self) -> (FromA, FromB)
where
    Self: ParallelIterator<Item = (A, B)>,
    FromA: Default + Send + ParallelExtend<A>,
    FromB: Default + Send + ParallelExtend<B>,
    A: Send,
    B: Send

Unzips the items of a parallel iterator into a pair of arbitrary ParallelExtend containers.

You may prefer to use unzip_into_vecs(), which allocates more efficiently with precise knowledge of how many elements the iterator contains, and even allows you to reuse existing vectors' backing stores rather than allocating fresh vectors.

Examples

use rayon::prelude::*;

let a = [(0, 1), (1, 2), (2, 3), (3, 4)];

let (left, right): (Vec<_>, Vec<_>) = a.par_iter().cloned().unzip();

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

Nested pairs can be unzipped too.

use rayon::prelude::*;

let (values, (squares, cubes)): (Vec<_>, (Vec<_>, Vec<_>)) = (0..4).into_par_iter()
    .map(|i| (i, (i * i, i * i * i)))
    .unzip();

assert_eq!(values, [0, 1, 2, 3]);
assert_eq!(squares, [0, 1, 4, 9]);
assert_eq!(cubes, [0, 1, 8, 27]);

fn partition<A, B, P>(self: Self, predicate: P) -> (A, B)
where
    A: Default + Send + ParallelExtend<<Self as >::Item>,
    B: Default + Send + ParallelExtend<<Self as >::Item>,
    P: Fn(&<Self as >::Item) -> bool + Sync + Send

Partitions the items of a parallel iterator into a pair of arbitrary ParallelExtend containers. Items for which the predicate returns true go into the first container, and the rest go into the second.

Note: unlike the standard Iterator::partition, this allows distinct collection types for the left and right items. This is more flexible, but may require new type annotations when converting sequential code that used type inference assuming the two were the same.

Examples

use rayon::prelude::*;

let (left, right): (Vec<_>, Vec<_>) = (0..8).into_par_iter().partition(|x| x % 2 == 0);

assert_eq!(left, [0, 2, 4, 6]);
assert_eq!(right, [1, 3, 5, 7]);

fn partition_map<A, B, P, L, R>(self: Self, predicate: P) -> (A, B)
where
    A: Default + Send + ParallelExtend<L>,
    B: Default + Send + ParallelExtend<R>,
    P: Fn(<Self as >::Item) -> Either<L, R> + Sync + Send,
    L: Send,
    R: Send

Partitions and maps the items of a parallel iterator into a pair of arbitrary ParallelExtend containers. Either::Left items go into the first container, and Either::Right items go into the second.

Examples

use rayon::prelude::*;
use rayon::iter::Either;

let (left, right): (Vec<_>, Vec<_>) = (0..8).into_par_iter()
    .partition_map(|x| {
        if x % 2 == 0 {
            Either::Left(x * 4)
        } else {
            Either::Right(x * 3)
        }
    });

assert_eq!(left, [0, 8, 16, 24]);
assert_eq!(right, [3, 9, 15, 21]);

Nested Either enums can be split as well.

use rayon::prelude::*;
use rayon::iter::Either::*;

let ((fizzbuzz, fizz), (buzz, other)): ((Vec<_>, Vec<_>), (Vec<_>, Vec<_>)) = (1..20)
    .into_par_iter()
    .partition_map(|x| match (x % 3, x % 5) {
        (0, 0) => Left(Left(x)),
        (0, _) => Left(Right(x)),
        (_, 0) => Right(Left(x)),
        (_, _) => Right(Right(x)),
    });

assert_eq!(fizzbuzz, [15]);
assert_eq!(fizz, [3, 6, 9, 12, 18]);
assert_eq!(buzz, [5, 10]);
assert_eq!(other, [1, 2, 4, 7, 8, 11, 13, 14, 16, 17, 19]);

fn intersperse(self: Self, element: <Self as >::Item) -> Intersperse<Self>
where
    <Self as >::Item: Clone

Intersperses clones of an element between items of this iterator.

Examples

use rayon::prelude::*;

let x = vec![1, 2, 3];
let r: Vec<_> = x.into_par_iter().intersperse(-1).collect();

assert_eq!(r, vec![1, -1, 2, -1, 3]);

fn take_any(self: Self, n: usize) -> TakeAny<Self>

Creates an iterator that yields n elements from anywhere in the original iterator.

This is similar to IndexedParallelIterator::take without being constrained to the "first" n of the original iterator order. The taken items will still maintain their relative order where that is visible in collect, reduce, and similar outputs.

Examples

use rayon::prelude::*;

let result: Vec<_> = (0..100)
    .into_par_iter()
    .filter(|&x| x % 2 == 0)
    .take_any(5)
    .collect();

assert_eq!(result.len(), 5);
assert!(result.windows(2).all(|w| w[0] < w[1]));

fn skip_any(self: Self, n: usize) -> SkipAny<Self>

Creates an iterator that skips n elements from anywhere in the original iterator.

This is similar to IndexedParallelIterator::skip without being constrained to the "first" n of the original iterator order. The remaining items will still maintain their relative order where that is visible in collect, reduce, and similar outputs.

Examples

use rayon::prelude::*;

let result: Vec<_> = (0..100)
    .into_par_iter()
    .filter(|&x| x % 2 == 0)
    .skip_any(5)
    .collect();

assert_eq!(result.len(), 45);
assert!(result.windows(2).all(|w| w[0] < w[1]));

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

Creates an iterator that takes elements from anywhere in the original iterator until the given predicate returns false.

The predicate may be anything -- e.g. it could be checking a fact about the item, a global condition unrelated to the item itself, or some combination thereof.

If parallel calls to the predicate race and give different results, then the true results will still take those particular items, while respecting the false result from elsewhere to skip any further items.

This is similar to Iterator::take_while without being constrained to the original iterator order. The taken items will still maintain their relative order where that is visible in collect, reduce, and similar outputs.

Examples

use rayon::prelude::*;

let result: Vec<_> = (0..100)
    .into_par_iter()
    .take_any_while(|x| *x < 50)
    .collect();

assert!(result.len() <= 50);
assert!(result.windows(2).all(|w| w[0] < w[1]));

use rayon::prelude::*;
use std::sync::atomic::AtomicUsize;
use std::sync::atomic::Ordering::Relaxed;

// Collect any group of items that sum <= 1000
let quota = AtomicUsize::new(1000);
let result: Vec<_> = (0_usize..100)
    .into_par_iter()
    .take_any_while(|&x| {
        quota.fetch_update(Relaxed, Relaxed, |q| q.checked_sub(x))
            .is_ok()
    })
    .collect();

let sum = result.iter().sum::<usize>();
assert!(matches!(sum, 902..=1000));

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

Creates an iterator that skips elements from anywhere in the original iterator until the given predicate returns false.

The predicate may be anything -- e.g. it could be checking a fact about the item, a global condition unrelated to the item itself, or some combination thereof.

If parallel calls to the predicate race and give different results, then the true results will still skip those particular items, while respecting the false result from elsewhere to skip any further items.

This is similar to Iterator::skip_while without being constrained to the original iterator order. The remaining items will still maintain their relative order where that is visible in collect, reduce, and similar outputs.

Examples

use rayon::prelude::*;

let result: Vec<_> = (0..100)
    .into_par_iter()
    .skip_any_while(|x| *x < 50)
    .collect();

assert!(result.len() >= 50);
assert!(result.windows(2).all(|w| w[0] < w[1]));

fn collect_vec_list(self: Self) -> LinkedList<Vec<<Self as >::Item>>

Collects this iterator into a linked list of vectors.

This is useful when you need to condense a parallel iterator into a collection, but have no specific requirements for what that collection should be. If you plan to store the collection longer-term, Vec<T> is, as always, likely the best default choice, despite the overhead that comes from concatenating each vector. Or, if this is an IndexedParallelIterator, you should also prefer to just collect to a Vec<T>.

Internally, most FromParallelIterator/ParallelExtend implementations use this strategy; each job collecting their chunk of the iterator to a Vec<T> and those chunks getting merged into a LinkedList, before then extending the collection with each vector. This is a very efficient way to collect an unindexed parallel iterator, without much intermediate data movement.

Examples

# use std::collections::LinkedList;
use rayon::prelude::*;

let result: LinkedList<Vec<_>> = (0..=100)
    .into_par_iter()
    .filter(|x| x % 2 == 0)
    .flat_map(|x| 0..x)
    .collect_vec_list();

// `par_iter.collect_vec_list().into_iter().flatten()` turns
// a parallel iterator into a serial one
let total_len = result.into_iter().flatten().count();
assert_eq!(total_len, 2550);

fn opt_len(self: &Self) -> Option<usize>

Internal method used to define the behavior of this parallel iterator. You should not need to call this directly.

Returns the number of items produced by this iterator, if known statically. This can be used by consumers to trigger special fast paths. Therefore, if Some(_) is returned, this iterator must only use the (indexed) Consumer methods when driving a consumer, such as split_at(). Calling UnindexedConsumer::split_off_left() or other UnindexedConsumer methods -- or returning an inaccurate value -- may result in panics.

This method is currently used to optimize collect for want of true Rust specialization; it may be removed when specialization is stable.