diff options
Diffstat (limited to 'third_party/rust/rayon/src/iter')
58 files changed, 17514 insertions, 0 deletions
diff --git a/third_party/rust/rayon/src/iter/chain.rs b/third_party/rust/rayon/src/iter/chain.rs new file mode 100644 index 0000000000..48fce07ed8 --- /dev/null +++ b/third_party/rust/rayon/src/iter/chain.rs @@ -0,0 +1,268 @@ +use super::plumbing::*; +use super::*; +use rayon_core::join; +use std::cmp; +use std::iter; + +/// `Chain` is an iterator that joins `b` after `a` in one continuous iterator. +/// This struct is created by the [`chain()`] method on [`ParallelIterator`] +/// +/// [`chain()`]: trait.ParallelIterator.html#method.chain +/// [`ParallelIterator`]: trait.ParallelIterator.html +#[must_use = "iterator adaptors are lazy and do nothing unless consumed"] +#[derive(Debug, Clone)] +pub struct Chain<A, B> +where + A: ParallelIterator, + B: ParallelIterator<Item = A::Item>, +{ + a: A, + b: B, +} + +impl<A, B> Chain<A, B> +where + A: ParallelIterator, + B: ParallelIterator<Item = A::Item>, +{ + /// Creates a new `Chain` iterator. + pub(super) fn new(a: A, b: B) -> Self { + Chain { a, b } + } +} + +impl<A, B> ParallelIterator for Chain<A, B> +where + A: ParallelIterator, + B: ParallelIterator<Item = A::Item>, +{ + type Item = A::Item; + + fn drive_unindexed<C>(self, consumer: C) -> C::Result + where + C: UnindexedConsumer<Self::Item>, + { + let Chain { a, b } = self; + + // If we returned a value from our own `opt_len`, then the collect consumer in particular + // will balk at being treated like an actual `UnindexedConsumer`. But when we do know the + // length, we can use `Consumer::split_at` instead, and this is still harmless for other + // truly-unindexed consumers too. + let (left, right, reducer) = if let Some(len) = a.opt_len() { + consumer.split_at(len) + } else { + let reducer = consumer.to_reducer(); + (consumer.split_off_left(), consumer, reducer) + }; + + let (a, b) = join(|| a.drive_unindexed(left), || b.drive_unindexed(right)); + reducer.reduce(a, b) + } + + fn opt_len(&self) -> Option<usize> { + self.a.opt_len()?.checked_add(self.b.opt_len()?) + } +} + +impl<A, B> IndexedParallelIterator for Chain<A, B> +where + A: IndexedParallelIterator, + B: IndexedParallelIterator<Item = A::Item>, +{ + fn drive<C>(self, consumer: C) -> C::Result + where + C: Consumer<Self::Item>, + { + let Chain { a, b } = self; + let (left, right, reducer) = consumer.split_at(a.len()); + let (a, b) = join(|| a.drive(left), || b.drive(right)); + reducer.reduce(a, b) + } + + fn len(&self) -> usize { + self.a.len().checked_add(self.b.len()).expect("overflow") + } + + fn with_producer<CB>(self, callback: CB) -> CB::Output + where + CB: ProducerCallback<Self::Item>, + { + let a_len = self.a.len(); + return self.a.with_producer(CallbackA { + callback, + a_len, + b: self.b, + }); + + struct CallbackA<CB, B> { + callback: CB, + a_len: usize, + b: B, + } + + impl<CB, B> ProducerCallback<B::Item> for CallbackA<CB, B> + where + B: IndexedParallelIterator, + CB: ProducerCallback<B::Item>, + { + type Output = CB::Output; + + fn callback<A>(self, a_producer: A) -> Self::Output + where + A: Producer<Item = B::Item>, + { + self.b.with_producer(CallbackB { + callback: self.callback, + a_len: self.a_len, + a_producer, + }) + } + } + + struct CallbackB<CB, A> { + callback: CB, + a_len: usize, + a_producer: A, + } + + impl<CB, A> ProducerCallback<A::Item> for CallbackB<CB, A> + where + A: Producer, + CB: ProducerCallback<A::Item>, + { + type Output = CB::Output; + + fn callback<B>(self, b_producer: B) -> Self::Output + where + B: Producer<Item = A::Item>, + { + let producer = ChainProducer::new(self.a_len, self.a_producer, b_producer); + self.callback.callback(producer) + } + } + } +} + +/// //////////////////////////////////////////////////////////////////////// + +struct ChainProducer<A, B> +where + A: Producer, + B: Producer<Item = A::Item>, +{ + a_len: usize, + a: A, + b: B, +} + +impl<A, B> ChainProducer<A, B> +where + A: Producer, + B: Producer<Item = A::Item>, +{ + fn new(a_len: usize, a: A, b: B) -> Self { + ChainProducer { a_len, a, b } + } +} + +impl<A, B> Producer for ChainProducer<A, B> +where + A: Producer, + B: Producer<Item = A::Item>, +{ + type Item = A::Item; + type IntoIter = ChainSeq<A::IntoIter, B::IntoIter>; + + fn into_iter(self) -> Self::IntoIter { + ChainSeq::new(self.a.into_iter(), self.b.into_iter()) + } + + fn min_len(&self) -> usize { + cmp::max(self.a.min_len(), self.b.min_len()) + } + + fn max_len(&self) -> usize { + cmp::min(self.a.max_len(), self.b.max_len()) + } + + fn split_at(self, index: usize) -> (Self, Self) { + if index <= self.a_len { + let a_rem = self.a_len - index; + let (a_left, a_right) = self.a.split_at(index); + let (b_left, b_right) = self.b.split_at(0); + ( + ChainProducer::new(index, a_left, b_left), + ChainProducer::new(a_rem, a_right, b_right), + ) + } else { + let (a_left, a_right) = self.a.split_at(self.a_len); + let (b_left, b_right) = self.b.split_at(index - self.a_len); + ( + ChainProducer::new(self.a_len, a_left, b_left), + ChainProducer::new(0, a_right, b_right), + ) + } + } + + fn fold_with<F>(self, mut folder: F) -> F + where + F: Folder<A::Item>, + { + folder = self.a.fold_with(folder); + if folder.full() { + folder + } else { + self.b.fold_with(folder) + } + } +} + +/// //////////////////////////////////////////////////////////////////////// +/// Wrapper for Chain to implement ExactSizeIterator + +struct ChainSeq<A, B> { + chain: iter::Chain<A, B>, +} + +impl<A, B> ChainSeq<A, B> { + fn new(a: A, b: B) -> ChainSeq<A, B> + where + A: ExactSizeIterator, + B: ExactSizeIterator<Item = A::Item>, + { + ChainSeq { chain: a.chain(b) } + } +} + +impl<A, B> Iterator for ChainSeq<A, B> +where + A: Iterator, + B: Iterator<Item = A::Item>, +{ + type Item = A::Item; + + fn next(&mut self) -> Option<Self::Item> { + self.chain.next() + } + + fn size_hint(&self) -> (usize, Option<usize>) { + self.chain.size_hint() + } +} + +impl<A, B> ExactSizeIterator for ChainSeq<A, B> +where + A: ExactSizeIterator, + B: ExactSizeIterator<Item = A::Item>, +{ +} + +impl<A, B> DoubleEndedIterator for ChainSeq<A, B> +where + A: DoubleEndedIterator, + B: DoubleEndedIterator<Item = A::Item>, +{ + fn next_back(&mut self) -> Option<Self::Item> { + self.chain.next_back() + } +} diff --git a/third_party/rust/rayon/src/iter/chunks.rs b/third_party/rust/rayon/src/iter/chunks.rs new file mode 100644 index 0000000000..ec48278d02 --- /dev/null +++ b/third_party/rust/rayon/src/iter/chunks.rs @@ -0,0 +1,226 @@ +use std::cmp::min; + +use super::plumbing::*; +use super::*; +use crate::math::div_round_up; + +/// `Chunks` is an iterator that groups elements of an underlying iterator. +/// +/// This struct is created by the [`chunks()`] method on [`IndexedParallelIterator`] +/// +/// [`chunks()`]: trait.IndexedParallelIterator.html#method.chunks +/// [`IndexedParallelIterator`]: trait.IndexedParallelIterator.html +#[must_use = "iterator adaptors are lazy and do nothing unless consumed"] +#[derive(Debug, Clone)] +pub struct Chunks<I> +where + I: IndexedParallelIterator, +{ + size: usize, + i: I, +} + +impl<I> Chunks<I> +where + I: IndexedParallelIterator, +{ + /// Creates a new `Chunks` iterator + pub(super) fn new(i: I, size: usize) -> Self { + Chunks { i, size } + } +} + +impl<I> ParallelIterator for Chunks<I> +where + I: IndexedParallelIterator, +{ + type Item = Vec<I::Item>; + + fn drive_unindexed<C>(self, consumer: C) -> C::Result + where + C: Consumer<Vec<I::Item>>, + { + bridge(self, consumer) + } + + fn opt_len(&self) -> Option<usize> { + Some(self.len()) + } +} + +impl<I> IndexedParallelIterator for Chunks<I> +where + I: IndexedParallelIterator, +{ + fn drive<C>(self, consumer: C) -> C::Result + where + C: Consumer<Self::Item>, + { + bridge(self, consumer) + } + + fn len(&self) -> usize { + div_round_up(self.i.len(), self.size) + } + + fn with_producer<CB>(self, callback: CB) -> CB::Output + where + CB: ProducerCallback<Self::Item>, + { + let len = self.i.len(); + return self.i.with_producer(Callback { + size: self.size, + len, + callback, + }); + + struct Callback<CB> { + size: usize, + len: usize, + callback: CB, + } + + impl<T, CB> ProducerCallback<T> for Callback<CB> + where + CB: ProducerCallback<Vec<T>>, + { + type Output = CB::Output; + + fn callback<P>(self, base: P) -> CB::Output + where + P: Producer<Item = T>, + { + let producer = ChunkProducer::new(self.size, self.len, base, Vec::from_iter); + self.callback.callback(producer) + } + } + } +} + +pub(super) struct ChunkProducer<P, F> { + chunk_size: usize, + len: usize, + base: P, + map: F, +} + +impl<P, F> ChunkProducer<P, F> { + pub(super) fn new(chunk_size: usize, len: usize, base: P, map: F) -> Self { + Self { + chunk_size, + len, + base, + map, + } + } +} + +impl<P, F, T> Producer for ChunkProducer<P, F> +where + P: Producer, + F: Fn(P::IntoIter) -> T + Send + Clone, +{ + type Item = T; + type IntoIter = std::iter::Map<ChunkSeq<P>, F>; + + fn into_iter(self) -> Self::IntoIter { + let chunks = ChunkSeq { + chunk_size: self.chunk_size, + len: self.len, + inner: if self.len > 0 { Some(self.base) } else { None }, + }; + chunks.map(self.map) + } + + fn split_at(self, index: usize) -> (Self, Self) { + let elem_index = min(index * self.chunk_size, self.len); + let (left, right) = self.base.split_at(elem_index); + ( + ChunkProducer { + chunk_size: self.chunk_size, + len: elem_index, + base: left, + map: self.map.clone(), + }, + ChunkProducer { + chunk_size: self.chunk_size, + len: self.len - elem_index, + base: right, + map: self.map, + }, + ) + } + + fn min_len(&self) -> usize { + div_round_up(self.base.min_len(), self.chunk_size) + } + + fn max_len(&self) -> usize { + self.base.max_len() / self.chunk_size + } +} + +pub(super) struct ChunkSeq<P> { + chunk_size: usize, + len: usize, + inner: Option<P>, +} + +impl<P> Iterator for ChunkSeq<P> +where + P: Producer, +{ + type Item = P::IntoIter; + + fn next(&mut self) -> Option<Self::Item> { + let producer = self.inner.take()?; + if self.len > self.chunk_size { + let (left, right) = producer.split_at(self.chunk_size); + self.inner = Some(right); + self.len -= self.chunk_size; + Some(left.into_iter()) + } else { + debug_assert!(self.len > 0); + self.len = 0; + Some(producer.into_iter()) + } + } + + fn size_hint(&self) -> (usize, Option<usize>) { + let len = self.len(); + (len, Some(len)) + } +} + +impl<P> ExactSizeIterator for ChunkSeq<P> +where + P: Producer, +{ + #[inline] + fn len(&self) -> usize { + div_round_up(self.len, self.chunk_size) + } +} + +impl<P> DoubleEndedIterator for ChunkSeq<P> +where + P: Producer, +{ + fn next_back(&mut self) -> Option<Self::Item> { + let producer = self.inner.take()?; + if self.len > self.chunk_size { + let mut size = self.len % self.chunk_size; + if size == 0 { + size = self.chunk_size; + } + let (left, right) = producer.split_at(self.len - size); + self.inner = Some(left); + self.len -= size; + Some(right.into_iter()) + } else { + debug_assert!(self.len > 0); + self.len = 0; + Some(producer.into_iter()) + } + } +} diff --git a/third_party/rust/rayon/src/iter/cloned.rs b/third_party/rust/rayon/src/iter/cloned.rs new file mode 100644 index 0000000000..8d5f420efa --- /dev/null +++ b/third_party/rust/rayon/src/iter/cloned.rs @@ -0,0 +1,223 @@ +use super::plumbing::*; +use super::*; + +use std::iter; + +/// `Cloned` is an iterator that clones the elements of an underlying iterator. +/// +/// This struct is created by the [`cloned()`] method on [`ParallelIterator`] +/// +/// [`cloned()`]: trait.ParallelIterator.html#method.cloned +/// [`ParallelIterator`]: trait.ParallelIterator.html +#[must_use = "iterator adaptors are lazy and do nothing unless consumed"] +#[derive(Debug, Clone)] +pub struct Cloned<I: ParallelIterator> { + base: I, +} + +impl<I> Cloned<I> +where + I: ParallelIterator, +{ + /// Creates a new `Cloned` iterator. + pub(super) fn new(base: I) -> Self { + Cloned { base } + } +} + +impl<'a, T, I> ParallelIterator for Cloned<I> +where + I: ParallelIterator<Item = &'a T>, + T: 'a + Clone + Send + Sync, +{ + type Item = T; + + fn drive_unindexed<C>(self, consumer: C) -> C::Result + where + C: UnindexedConsumer<Self::Item>, + { + let consumer1 = ClonedConsumer::new(consumer); + self.base.drive_unindexed(consumer1) + } + + fn opt_len(&self) -> Option<usize> { + self.base.opt_len() + } +} + +impl<'a, T, I> IndexedParallelIterator for Cloned<I> +where + I: IndexedParallelIterator<Item = &'a T>, + T: 'a + Clone + Send + Sync, +{ + fn drive<C>(self, consumer: C) -> C::Result + where + C: Consumer<Self::Item>, + { + let consumer1 = ClonedConsumer::new(consumer); + self.base.drive(consumer1) + } + + fn len(&self) -> usize { + self.base.len() + } + + fn with_producer<CB>(self, callback: CB) -> CB::Output + where + CB: ProducerCallback<Self::Item>, + { + return self.base.with_producer(Callback { callback }); + + struct Callback<CB> { + callback: CB, + } + + impl<'a, T, CB> ProducerCallback<&'a T> for Callback<CB> + where + CB: ProducerCallback<T>, + T: 'a + Clone + Send, + { + type Output = CB::Output; + + fn callback<P>(self, base: P) -> CB::Output + where + P: Producer<Item = &'a T>, + { + let producer = ClonedProducer { base }; + self.callback.callback(producer) + } + } + } +} + +/// //////////////////////////////////////////////////////////////////////// + +struct ClonedProducer<P> { + base: P, +} + +impl<'a, T, P> Producer for ClonedProducer<P> +where + P: Producer<Item = &'a T>, + T: 'a + Clone, +{ + type Item = T; + type IntoIter = iter::Cloned<P::IntoIter>; + + fn into_iter(self) -> Self::IntoIter { + self.base.into_iter().cloned() + } + + fn min_len(&self) -> usize { + self.base.min_len() + } + + fn max_len(&self) -> usize { + self.base.max_len() + } + + fn split_at(self, index: usize) -> (Self, Self) { + let (left, right) = self.base.split_at(index); + ( + ClonedProducer { base: left }, + ClonedProducer { base: right }, + ) + } + + fn fold_with<F>(self, folder: F) -> F + where + F: Folder<Self::Item>, + { + self.base.fold_with(ClonedFolder { base: folder }).base + } +} + +/// //////////////////////////////////////////////////////////////////////// +/// Consumer implementation + +struct ClonedConsumer<C> { + base: C, +} + +impl<C> ClonedConsumer<C> { + fn new(base: C) -> Self { + ClonedConsumer { base } + } +} + +impl<'a, T, C> Consumer<&'a T> for ClonedConsumer<C> +where + C: Consumer<T>, + T: 'a + Clone, +{ + type Folder = ClonedFolder<C::Folder>; + type Reducer = C::Reducer; + type Result = C::Result; + + fn split_at(self, index: usize) -> (Self, Self, Self::Reducer) { + let (left, right, reducer) = self.base.split_at(index); + ( + ClonedConsumer::new(left), + ClonedConsumer::new(right), + reducer, + ) + } + + fn into_folder(self) -> Self::Folder { + ClonedFolder { + base: self.base.into_folder(), + } + } + + fn full(&self) -> bool { + self.base.full() + } +} + +impl<'a, T, C> UnindexedConsumer<&'a T> for ClonedConsumer<C> +where + C: UnindexedConsumer<T>, + T: 'a + Clone, +{ + fn split_off_left(&self) -> Self { + ClonedConsumer::new(self.base.split_off_left()) + } + + fn to_reducer(&self) -> Self::Reducer { + self.base.to_reducer() + } +} + +struct ClonedFolder<F> { + base: F, +} + +impl<'a, T, F> Folder<&'a T> for ClonedFolder<F> +where + F: Folder<T>, + T: 'a + Clone, +{ + type Result = F::Result; + + fn consume(self, item: &'a T) -> Self { + ClonedFolder { + base: self.base.consume(item.clone()), + } + } + + fn consume_iter<I>(mut self, iter: I) -> Self + where + I: IntoIterator<Item = &'a T>, + { + self.base = self.base.consume_iter(iter.into_iter().cloned()); + self + } + + fn complete(self) -> F::Result { + self.base.complete() + } + + fn full(&self) -> bool { + self.base.full() + } +} diff --git a/third_party/rust/rayon/src/iter/collect/consumer.rs b/third_party/rust/rayon/src/iter/collect/consumer.rs new file mode 100644 index 0000000000..acd67dffa7 --- /dev/null +++ b/third_party/rust/rayon/src/iter/collect/consumer.rs @@ -0,0 +1,186 @@ +use super::super::plumbing::*; +use crate::SendPtr; +use std::marker::PhantomData; +use std::ptr; +use std::slice; + +pub(super) struct CollectConsumer<'c, T: Send> { + /// See `CollectResult` for explanation of why this is not a slice + start: SendPtr<T>, + len: usize, + marker: PhantomData<&'c mut T>, +} + +impl<T: Send> CollectConsumer<'_, T> { + /// Create a collector for `len` items in the unused capacity of the vector. + pub(super) fn appender(vec: &mut Vec<T>, len: usize) -> CollectConsumer<'_, T> { + let start = vec.len(); + assert!(vec.capacity() - start >= len); + + // SAFETY: We already made sure to have the additional space allocated. + // The pointer is derived from `Vec` directly, not through a `Deref`, + // so it has provenance over the whole allocation. + unsafe { CollectConsumer::new(vec.as_mut_ptr().add(start), len) } + } +} + +impl<'c, T: Send + 'c> CollectConsumer<'c, T> { + /// The target memory is considered uninitialized, and will be + /// overwritten without reading or dropping existing values. + unsafe fn new(start: *mut T, len: usize) -> Self { + CollectConsumer { + start: SendPtr(start), + len, + marker: PhantomData, + } + } +} + +/// CollectResult represents an initialized part of the target slice. +/// +/// This is a proxy owner of the elements in the slice; when it drops, +/// the elements will be dropped, unless its ownership is released before then. +#[must_use] +pub(super) struct CollectResult<'c, T> { + /// This pointer and length has the same representation as a slice, + /// but retains the provenance of the entire array so that we can merge + /// these regions together in `CollectReducer`. + start: SendPtr<T>, + total_len: usize, + /// The current initialized length after `start` + initialized_len: usize, + /// Lifetime invariance guarantees that the data flows from consumer to result, + /// especially for the `scope_fn` callback in `Collect::with_consumer`. + invariant_lifetime: PhantomData<&'c mut &'c mut [T]>, +} + +unsafe impl<'c, T> Send for CollectResult<'c, T> where T: Send {} + +impl<'c, T> CollectResult<'c, T> { + /// The current length of the collect result + pub(super) fn len(&self) -> usize { + self.initialized_len + } + + /// Release ownership of the slice of elements, and return the length + pub(super) fn release_ownership(mut self) -> usize { + let ret = self.initialized_len; + self.initialized_len = 0; + ret + } +} + +impl<'c, T> Drop for CollectResult<'c, T> { + fn drop(&mut self) { + // Drop the first `self.initialized_len` elements, which have been recorded + // to be initialized by the folder. + unsafe { + ptr::drop_in_place(slice::from_raw_parts_mut( + self.start.0, + self.initialized_len, + )); + } + } +} + +impl<'c, T: Send + 'c> Consumer<T> for CollectConsumer<'c, T> { + type Folder = CollectResult<'c, T>; + type Reducer = CollectReducer; + type Result = CollectResult<'c, T>; + + fn split_at(self, index: usize) -> (Self, Self, CollectReducer) { + let CollectConsumer { start, len, .. } = self; + + // Produce new consumers. + // SAFETY: This assert checks that `index` is a valid offset for `start` + unsafe { + assert!(index <= len); + ( + CollectConsumer::new(start.0, index), + CollectConsumer::new(start.0.add(index), len - index), + CollectReducer, + ) + } + } + + fn into_folder(self) -> Self::Folder { + // Create a result/folder that consumes values and writes them + // into the region after start. The initial result has length 0. + CollectResult { + start: self.start, + total_len: self.len, + initialized_len: 0, + invariant_lifetime: PhantomData, + } + } + + fn full(&self) -> bool { + false + } +} + +impl<'c, T: Send + 'c> Folder<T> for CollectResult<'c, T> { + type Result = Self; + + fn consume(mut self, item: T) -> Self { + assert!( + self.initialized_len < self.total_len, + "too many values pushed to consumer" + ); + + // SAFETY: The assert above is a bounds check for this write, and we + // avoid assignment here so we do not drop an uninitialized T. + unsafe { + // Write item and increase the initialized length + self.start.0.add(self.initialized_len).write(item); + self.initialized_len += 1; + } + + self + } + + fn complete(self) -> Self::Result { + // NB: We don't explicitly check that the local writes were complete, + // but Collect will assert the total result length in the end. + self + } + + fn full(&self) -> bool { + false + } +} + +/// Pretend to be unindexed for `special_collect_into_vec`, +/// but we should never actually get used that way... +impl<'c, T: Send + 'c> UnindexedConsumer<T> for CollectConsumer<'c, T> { + fn split_off_left(&self) -> Self { + unreachable!("CollectConsumer must be indexed!") + } + fn to_reducer(&self) -> Self::Reducer { + CollectReducer + } +} + +/// CollectReducer combines adjacent chunks; the result must always +/// be contiguous so that it is one combined slice. +pub(super) struct CollectReducer; + +impl<'c, T> Reducer<CollectResult<'c, T>> for CollectReducer { + fn reduce( + self, + mut left: CollectResult<'c, T>, + right: CollectResult<'c, T>, + ) -> CollectResult<'c, T> { + // Merge if the CollectResults are adjacent and in left to right order + // else: drop the right piece now and total length will end up short in the end, + // when the correctness of the collected result is asserted. + unsafe { + let left_end = left.start.0.add(left.initialized_len); + if left_end == right.start.0 { + left.total_len += right.total_len; + left.initialized_len += right.release_ownership(); + } + left + } + } +} diff --git a/third_party/rust/rayon/src/iter/collect/mod.rs b/third_party/rust/rayon/src/iter/collect/mod.rs new file mode 100644 index 0000000000..4044a685b8 --- /dev/null +++ b/third_party/rust/rayon/src/iter/collect/mod.rs @@ -0,0 +1,116 @@ +use super::{IndexedParallelIterator, ParallelIterator}; + +mod consumer; +use self::consumer::CollectConsumer; +use self::consumer::CollectResult; +use super::unzip::unzip_indexed; + +mod test; + +/// Collects the results of the exact iterator into the specified vector. +/// +/// This is called by `IndexedParallelIterator::collect_into_vec`. +pub(super) fn collect_into_vec<I, T>(pi: I, v: &mut Vec<T>) +where + I: IndexedParallelIterator<Item = T>, + T: Send, +{ + v.truncate(0); // clear any old data + let len = pi.len(); + collect_with_consumer(v, len, |consumer| pi.drive(consumer)); +} + +/// Collects the results of the iterator into the specified vector. +/// +/// Technically, this only works for `IndexedParallelIterator`, but we're faking a +/// bit of specialization here until Rust can do that natively. Callers are +/// using `opt_len` to find the length before calling this, and only exact +/// iterators will return anything but `None` there. +/// +/// Since the type system doesn't understand that contract, we have to allow +/// *any* `ParallelIterator` here, and `CollectConsumer` has to also implement +/// `UnindexedConsumer`. That implementation panics `unreachable!` in case +/// there's a bug where we actually do try to use this unindexed. +pub(super) fn special_extend<I, T>(pi: I, len: usize, v: &mut Vec<T>) +where + I: ParallelIterator<Item = T>, + T: Send, +{ + collect_with_consumer(v, len, |consumer| pi.drive_unindexed(consumer)); +} + +/// Unzips the results of the exact iterator into the specified vectors. +/// +/// This is called by `IndexedParallelIterator::unzip_into_vecs`. +pub(super) fn unzip_into_vecs<I, A, B>(pi: I, left: &mut Vec<A>, right: &mut Vec<B>) +where + I: IndexedParallelIterator<Item = (A, B)>, + A: Send, + B: Send, +{ + // clear any old data + left.truncate(0); + right.truncate(0); + + let len = pi.len(); + collect_with_consumer(right, len, |right_consumer| { + let mut right_result = None; + collect_with_consumer(left, len, |left_consumer| { + let (left_r, right_r) = unzip_indexed(pi, left_consumer, right_consumer); + right_result = Some(right_r); + left_r + }); + right_result.unwrap() + }); +} + +/// Create a consumer on the slice of memory we are collecting into. +/// +/// The consumer needs to be used inside the scope function, and the +/// complete collect result passed back. +/// +/// This method will verify the collect result, and panic if the slice +/// was not fully written into. Otherwise, in the successful case, +/// the vector is complete with the collected result. +fn collect_with_consumer<T, F>(vec: &mut Vec<T>, len: usize, scope_fn: F) +where + T: Send, + F: FnOnce(CollectConsumer<'_, T>) -> CollectResult<'_, T>, +{ + // Reserve space for `len` more elements in the vector, + vec.reserve(len); + + // Create the consumer and run the callback for collection. + let result = scope_fn(CollectConsumer::appender(vec, len)); + + // The `CollectResult` represents a contiguous part of the slice, that has + // been written to. On unwind here, the `CollectResult` will be dropped. If + // some producers on the way did not produce enough elements, partial + // `CollectResult`s may have been dropped without being reduced to the final + // result, and we will see that as the length coming up short. + // + // Here, we assert that added length is fully initialized. This is checked + // by the following assert, which verifies if a complete `CollectResult` + // was produced; if the length is correct, it is necessarily covering the + // target slice. Since we know that the consumer cannot have escaped from + // `drive` (by parametricity, essentially), we know that any stores that + // will happen, have happened. Unless some code is buggy, that means we + // should have seen `len` total writes. + let actual_writes = result.len(); + assert!( + actual_writes == len, + "expected {} total writes, but got {}", + len, + actual_writes + ); + + // Release the result's mutable borrow and "proxy ownership" + // of the elements, before the vector takes it over. + result.release_ownership(); + + let new_len = vec.len() + len; + + unsafe { + vec.set_len(new_len); + } +} diff --git a/third_party/rust/rayon/src/iter/collect/test.rs b/third_party/rust/rayon/src/iter/collect/test.rs new file mode 100644 index 0000000000..b5f676f5de --- /dev/null +++ b/third_party/rust/rayon/src/iter/collect/test.rs @@ -0,0 +1,371 @@ +#![cfg(test)] +#![allow(unused_assignments)] + +// These tests are primarily targeting "abusive" producers that will +// try to drive the "collect consumer" incorrectly. These should +// result in panics. + +use super::collect_with_consumer; +use crate::iter::plumbing::*; +use rayon_core::join; + +use std::fmt; +use std::panic; +use std::sync::atomic::{AtomicUsize, Ordering}; +use std::thread::Result as ThreadResult; + +/// Promises to produce 2 items, but then produces 3. Does not do any +/// splits at all. +#[test] +#[should_panic(expected = "too many values")] +fn produce_too_many_items() { + let mut v = vec![]; + collect_with_consumer(&mut v, 2, |consumer| { + let mut folder = consumer.into_folder(); + folder = folder.consume(22); + folder = folder.consume(23); + folder = folder.consume(24); + unreachable!("folder does not complete") + }); +} + +/// Produces fewer items than promised. Does not do any +/// splits at all. +#[test] +#[should_panic(expected = "expected 5 total writes, but got 2")] +fn produce_fewer_items() { + let mut v = vec![]; + collect_with_consumer(&mut v, 5, |consumer| { + let mut folder = consumer.into_folder(); + folder = folder.consume(22); + folder = folder.consume(23); + folder.complete() + }); +} + +// Complete is not called by the consumer. Hence,the collection vector is not fully initialized. +#[test] +#[should_panic(expected = "expected 4 total writes, but got 2")] +fn left_produces_items_with_no_complete() { + let mut v = vec![]; + collect_with_consumer(&mut v, 4, |consumer| { + let (left_consumer, right_consumer, _) = consumer.split_at(2); + let mut left_folder = left_consumer.into_folder(); + let mut right_folder = right_consumer.into_folder(); + left_folder = left_folder.consume(0).consume(1); + right_folder = right_folder.consume(2).consume(3); + right_folder.complete() + }); +} + +// Complete is not called by the right consumer. Hence,the +// collection vector is not fully initialized. +#[test] +#[should_panic(expected = "expected 4 total writes, but got 2")] +fn right_produces_items_with_no_complete() { + let mut v = vec![]; + collect_with_consumer(&mut v, 4, |consumer| { + let (left_consumer, right_consumer, _) = consumer.split_at(2); + let mut left_folder = left_consumer.into_folder(); + let mut right_folder = right_consumer.into_folder(); + left_folder = left_folder.consume(0).consume(1); + right_folder = right_folder.consume(2).consume(3); + left_folder.complete() + }); +} + +// Complete is not called by the consumer. Hence,the collection vector is not fully initialized. +#[test] +fn produces_items_with_no_complete() { + let counter = DropCounter::default(); + let mut v = vec![]; + let panic_result = panic::catch_unwind(panic::AssertUnwindSafe(|| { + collect_with_consumer(&mut v, 2, |consumer| { + let mut folder = consumer.into_folder(); + folder = folder.consume(counter.element()); + folder = folder.consume(counter.element()); + panic!("folder does not complete"); + }); + })); + assert!(v.is_empty()); + assert_is_panic_with_message(&panic_result, "folder does not complete"); + counter.assert_drop_count(); +} + +// The left consumer produces too many items while the right +// consumer produces correct number. +#[test] +#[should_panic(expected = "too many values")] +fn left_produces_too_many_items() { + let mut v = vec![]; + collect_with_consumer(&mut v, 4, |consumer| { + let (left_consumer, right_consumer, _) = consumer.split_at(2); + let mut left_folder = left_consumer.into_folder(); + let mut right_folder = right_consumer.into_folder(); + left_folder = left_folder.consume(0).consume(1).consume(2); + right_folder = right_folder.consume(2).consume(3); + let _ = right_folder.complete(); + unreachable!("folder does not complete"); + }); +} + +// The right consumer produces too many items while the left +// consumer produces correct number. +#[test] +#[should_panic(expected = "too many values")] +fn right_produces_too_many_items() { + let mut v = vec![]; + collect_with_consumer(&mut v, 4, |consumer| { + let (left_consumer, right_consumer, _) = consumer.split_at(2); + let mut left_folder = left_consumer.into_folder(); + let mut right_folder = right_consumer.into_folder(); + left_folder = left_folder.consume(0).consume(1); + right_folder = right_folder.consume(2).consume(3).consume(4); + let _ = left_folder.complete(); + unreachable!("folder does not complete"); + }); +} + +// The left consumer produces fewer items while the right +// consumer produces correct number. +#[test] +#[should_panic(expected = "expected 4 total writes, but got 1")] +fn left_produces_fewer_items() { + let mut v = vec![]; + collect_with_consumer(&mut v, 4, |consumer| { + let reducer = consumer.to_reducer(); + let (left_consumer, right_consumer, _) = consumer.split_at(2); + let mut left_folder = left_consumer.into_folder(); + let mut right_folder = right_consumer.into_folder(); + left_folder = left_folder.consume(0); + right_folder = right_folder.consume(2).consume(3); + let left_result = left_folder.complete(); + let right_result = right_folder.complete(); + reducer.reduce(left_result, right_result) + }); +} + +// The left and right consumer produce the correct number but +// only left result is returned +#[test] +#[should_panic(expected = "expected 4 total writes, but got 2")] +fn only_left_result() { + let mut v = vec![]; + collect_with_consumer(&mut v, 4, |consumer| { + let (left_consumer, right_consumer, _) = consumer.split_at(2); + let mut left_folder = left_consumer.into_folder(); + let mut right_folder = right_consumer.into_folder(); + left_folder = left_folder.consume(0).consume(1); + right_folder = right_folder.consume(2).consume(3); + let left_result = left_folder.complete(); + let _ = right_folder.complete(); + left_result + }); +} + +// The left and right consumer produce the correct number but +// only right result is returned +#[test] +#[should_panic(expected = "expected 4 total writes, but got 2")] +fn only_right_result() { + let mut v = vec![]; + collect_with_consumer(&mut v, 4, |consumer| { + let (left_consumer, right_consumer, _) = consumer.split_at(2); + let mut left_folder = left_consumer.into_folder(); + let mut right_folder = right_consumer.into_folder(); + left_folder = left_folder.consume(0).consume(1); + right_folder = right_folder.consume(2).consume(3); + let _ = left_folder.complete(); + right_folder.complete() + }); +} + +// The left and right consumer produce the correct number but reduce +// in the wrong order. +#[test] +#[should_panic(expected = "expected 4 total writes, but got 2")] +fn reducer_does_not_preserve_order() { + let mut v = vec![]; + collect_with_consumer(&mut v, 4, |consumer| { + let reducer = consumer.to_reducer(); + let (left_consumer, right_consumer, _) = consumer.split_at(2); + let mut left_folder = left_consumer.into_folder(); + let mut right_folder = right_consumer.into_folder(); + left_folder = left_folder.consume(0).consume(1); + right_folder = right_folder.consume(2).consume(3); + let left_result = left_folder.complete(); + let right_result = right_folder.complete(); + reducer.reduce(right_result, left_result) + }); +} + +// The right consumer produces fewer items while the left +// consumer produces correct number. +#[test] +#[should_panic(expected = "expected 4 total writes, but got 3")] +fn right_produces_fewer_items() { + let mut v = vec![]; + collect_with_consumer(&mut v, 4, |consumer| { + let reducer = consumer.to_reducer(); + let (left_consumer, right_consumer, _) = consumer.split_at(2); + let mut left_folder = left_consumer.into_folder(); + let mut right_folder = right_consumer.into_folder(); + left_folder = left_folder.consume(0).consume(1); + right_folder = right_folder.consume(2); + let left_result = left_folder.complete(); + let right_result = right_folder.complete(); + reducer.reduce(left_result, right_result) + }); +} + +// The left consumer panics and the right stops short, like `panic_fuse()`. +// We should get the left panic without finishing `collect_with_consumer`. +#[test] +#[should_panic(expected = "left consumer panic")] +fn left_panics() { + let mut v = vec![]; + collect_with_consumer(&mut v, 4, |consumer| { + let reducer = consumer.to_reducer(); + let (left_consumer, right_consumer, _) = consumer.split_at(2); + let (left_result, right_result) = join( + || { + let mut left_folder = left_consumer.into_folder(); + left_folder = left_folder.consume(0); + panic!("left consumer panic"); + }, + || { + let mut right_folder = right_consumer.into_folder(); + right_folder = right_folder.consume(2); + right_folder.complete() // early return + }, + ); + reducer.reduce(left_result, right_result) + }); + unreachable!(); +} + +// The right consumer panics and the left stops short, like `panic_fuse()`. +// We should get the right panic without finishing `collect_with_consumer`. +#[test] +#[should_panic(expected = "right consumer panic")] +fn right_panics() { + let mut v = vec![]; + collect_with_consumer(&mut v, 4, |consumer| { + let reducer = consumer.to_reducer(); + let (left_consumer, right_consumer, _) = consumer.split_at(2); + let (left_result, right_result) = join( + || { + let mut left_folder = left_consumer.into_folder(); + left_folder = left_folder.consume(0); + left_folder.complete() // early return + }, + || { + let mut right_folder = right_consumer.into_folder(); + right_folder = right_folder.consume(2); + panic!("right consumer panic"); + }, + ); + reducer.reduce(left_result, right_result) + }); + unreachable!(); +} + +// The left consumer produces fewer items while the right +// consumer produces correct number; check that created elements are dropped +#[test] +fn left_produces_fewer_items_drops() { + let counter = DropCounter::default(); + let mut v = vec![]; + let panic_result = panic::catch_unwind(panic::AssertUnwindSafe(|| { + collect_with_consumer(&mut v, 4, |consumer| { + let reducer = consumer.to_reducer(); + let (left_consumer, right_consumer, _) = consumer.split_at(2); + let mut left_folder = left_consumer.into_folder(); + let mut right_folder = right_consumer.into_folder(); + left_folder = left_folder.consume(counter.element()); + right_folder = right_folder + .consume(counter.element()) + .consume(counter.element()); + let left_result = left_folder.complete(); + let right_result = right_folder.complete(); + reducer.reduce(left_result, right_result) + }); + })); + assert!(v.is_empty()); + assert_is_panic_with_message(&panic_result, "expected 4 total writes, but got 1"); + counter.assert_drop_count(); +} + +/// This counter can create elements, and then count and verify +/// the number of which have actually been dropped again. +#[derive(Default)] +struct DropCounter { + created: AtomicUsize, + dropped: AtomicUsize, +} + +struct Element<'a>(&'a AtomicUsize); + +impl DropCounter { + fn created(&self) -> usize { + self.created.load(Ordering::SeqCst) + } + + fn dropped(&self) -> usize { + self.dropped.load(Ordering::SeqCst) + } + + fn element(&self) -> Element<'_> { + self.created.fetch_add(1, Ordering::SeqCst); + Element(&self.dropped) + } + + fn assert_drop_count(&self) { + assert_eq!( + self.created(), + self.dropped(), + "Expected {} dropped elements, but found {}", + self.created(), + self.dropped() + ); + } +} + +impl<'a> Drop for Element<'a> { + fn drop(&mut self) { + self.0.fetch_add(1, Ordering::SeqCst); + } +} + +/// Assert that the result from catch_unwind is a panic that contains expected message +fn assert_is_panic_with_message<T>(result: &ThreadResult<T>, expected: &str) +where + T: fmt::Debug, +{ + match result { + Ok(value) => { + panic!( + "assertion failure: Expected panic, got successful {:?}", + value + ); + } + Err(error) => { + let message_str = error.downcast_ref::<&'static str>().cloned(); + let message_string = error.downcast_ref::<String>().map(String::as_str); + if let Some(message) = message_str.or(message_string) { + if !message.contains(expected) { + panic!( + "assertion failure: Expected {:?}, but found panic with {:?}", + expected, message + ); + } + // assertion passes + } else { + panic!( + "assertion failure: Expected {:?}, but found panic with unknown value", + expected + ); + } + } + } +} diff --git a/third_party/rust/rayon/src/iter/copied.rs b/third_party/rust/rayon/src/iter/copied.rs new file mode 100644 index 0000000000..12c9c5b5bf --- /dev/null +++ b/third_party/rust/rayon/src/iter/copied.rs @@ -0,0 +1,223 @@ +use super::plumbing::*; +use super::*; + +use std::iter; + +/// `Copied` is an iterator that copies the elements of an underlying iterator. +/// +/// This struct is created by the [`copied()`] method on [`ParallelIterator`] +/// +/// [`copied()`]: trait.ParallelIterator.html#method.copied +/// [`ParallelIterator`]: trait.ParallelIterator.html +#[must_use = "iterator adaptors are lazy and do nothing unless consumed"] +#[derive(Debug, Clone)] +pub struct Copied<I: ParallelIterator> { + base: I, +} + +impl<I> Copied<I> +where + I: ParallelIterator, +{ + /// Creates a new `Copied` iterator. + pub(super) fn new(base: I) -> Self { + Copied { base } + } +} + +impl<'a, T, I> ParallelIterator for Copied<I> +where + I: ParallelIterator<Item = &'a T>, + T: 'a + Copy + Send + Sync, +{ + type Item = T; + + fn drive_unindexed<C>(self, consumer: C) -> C::Result + where + C: UnindexedConsumer<Self::Item>, + { + let consumer1 = CopiedConsumer::new(consumer); + self.base.drive_unindexed(consumer1) + } + + fn opt_len(&self) -> Option<usize> { + self.base.opt_len() + } +} + +impl<'a, T, I> IndexedParallelIterator for Copied<I> +where + I: IndexedParallelIterator<Item = &'a T>, + T: 'a + Copy + Send + Sync, +{ + fn drive<C>(self, consumer: C) -> C::Result + where + C: Consumer<Self::Item>, + { + let consumer1 = CopiedConsumer::new(consumer); + self.base.drive(consumer1) + } + + fn len(&self) -> usize { + self.base.len() + } + + fn with_producer<CB>(self, callback: CB) -> CB::Output + where + CB: ProducerCallback<Self::Item>, + { + return self.base.with_producer(Callback { callback }); + + struct Callback<CB> { + callback: CB, + } + + impl<'a, T, CB> ProducerCallback<&'a T> for Callback<CB> + where + CB: ProducerCallback<T>, + T: 'a + Copy + Send, + { + type Output = CB::Output; + + fn callback<P>(self, base: P) -> CB::Output + where + P: Producer<Item = &'a T>, + { + let producer = CopiedProducer { base }; + self.callback.callback(producer) + } + } + } +} + +/// //////////////////////////////////////////////////////////////////////// + +struct CopiedProducer<P> { + base: P, +} + +impl<'a, T, P> Producer for CopiedProducer<P> +where + P: Producer<Item = &'a T>, + T: 'a + Copy, +{ + type Item = T; + type IntoIter = iter::Copied<P::IntoIter>; + + fn into_iter(self) -> Self::IntoIter { + self.base.into_iter().copied() + } + + fn min_len(&self) -> usize { + self.base.min_len() + } + + fn max_len(&self) -> usize { + self.base.max_len() + } + + fn split_at(self, index: usize) -> (Self, Self) { + let (left, right) = self.base.split_at(index); + ( + CopiedProducer { base: left }, + CopiedProducer { base: right }, + ) + } + + fn fold_with<F>(self, folder: F) -> F + where + F: Folder<Self::Item>, + { + self.base.fold_with(CopiedFolder { base: folder }).base + } +} + +/// //////////////////////////////////////////////////////////////////////// +/// Consumer implementation + +struct CopiedConsumer<C> { + base: C, +} + +impl<C> CopiedConsumer<C> { + fn new(base: C) -> Self { + CopiedConsumer { base } + } +} + +impl<'a, T, C> Consumer<&'a T> for CopiedConsumer<C> +where + C: Consumer<T>, + T: 'a + Copy, +{ + type Folder = CopiedFolder<C::Folder>; + type Reducer = C::Reducer; + type Result = C::Result; + + fn split_at(self, index: usize) -> (Self, Self, Self::Reducer) { + let (left, right, reducer) = self.base.split_at(index); + ( + CopiedConsumer::new(left), + CopiedConsumer::new(right), + reducer, + ) + } + + fn into_folder(self) -> Self::Folder { + CopiedFolder { + base: self.base.into_folder(), + } + } + + fn full(&self) -> bool { + self.base.full() + } +} + +impl<'a, T, C> UnindexedConsumer<&'a T> for CopiedConsumer<C> +where + C: UnindexedConsumer<T>, + T: 'a + Copy, +{ + fn split_off_left(&self) -> Self { + CopiedConsumer::new(self.base.split_off_left()) + } + + fn to_reducer(&self) -> Self::Reducer { + self.base.to_reducer() + } +} + +struct CopiedFolder<F> { + base: F, +} + +impl<'a, T, F> Folder<&'a T> for CopiedFolder<F> +where + F: Folder<T>, + T: 'a + Copy, +{ + type Result = F::Result; + + fn consume(self, &item: &'a T) -> Self { + CopiedFolder { + base: self.base.consume(item), + } + } + + fn consume_iter<I>(mut self, iter: I) -> Self + where + I: IntoIterator<Item = &'a T>, + { + self.base = self.base.consume_iter(iter.into_iter().copied()); + self + } + + fn complete(self) -> F::Result { + self.base.complete() + } + + fn full(&self) -> bool { + self.base.full() + } +} diff --git a/third_party/rust/rayon/src/iter/empty.rs b/third_party/rust/rayon/src/iter/empty.rs new file mode 100644 index 0000000000..85a2e5fd93 --- /dev/null +++ b/third_party/rust/rayon/src/iter/empty.rs @@ -0,0 +1,104 @@ +use crate::iter::plumbing::*; +use crate::iter::*; + +use std::fmt; +use std::marker::PhantomData; + +/// Creates a parallel iterator that produces nothing. +/// +/// This admits no parallelism on its own, but it could be used for code that +/// deals with generic parallel iterators. +/// +/// # Examples +/// +/// ``` +/// use rayon::prelude::*; +/// use rayon::iter::empty; +/// +/// let pi = (0..1234).into_par_iter() +/// .chain(empty()) +/// .chain(1234..10_000); +/// +/// assert_eq!(pi.count(), 10_000); +/// ``` +pub fn empty<T: Send>() -> Empty<T> { + Empty { + marker: PhantomData, + } +} + +/// Iterator adaptor for [the `empty()` function](fn.empty.html). +pub struct Empty<T: Send> { + marker: PhantomData<T>, +} + +impl<T: Send> Clone for Empty<T> { + fn clone(&self) -> Self { + empty() + } +} + +impl<T: Send> fmt::Debug for Empty<T> { + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + f.pad("Empty") + } +} + +impl<T: Send> ParallelIterator for Empty<T> { + type Item = T; + + fn drive_unindexed<C>(self, consumer: C) -> C::Result + where + C: UnindexedConsumer<Self::Item>, + { + self.drive(consumer) + } + + fn opt_len(&self) -> Option<usize> { + Some(0) + } +} + +impl<T: Send> IndexedParallelIterator for Empty<T> { + fn drive<C>(self, consumer: C) -> C::Result + where + C: Consumer<Self::Item>, + { + consumer.into_folder().complete() + } + + fn len(&self) -> usize { + 0 + } + + fn with_producer<CB>(self, callback: CB) -> CB::Output + where + CB: ProducerCallback<Self::Item>, + { + callback.callback(EmptyProducer(PhantomData)) + } +} + +/// Private empty producer +struct EmptyProducer<T: Send>(PhantomData<T>); + +impl<T: Send> Producer for EmptyProducer<T> { + type Item = T; + type IntoIter = std::iter::Empty<T>; + + fn into_iter(self) -> Self::IntoIter { + std::iter::empty() + } + + fn split_at(self, index: usize) -> (Self, Self) { + debug_assert_eq!(index, 0); + (self, EmptyProducer(PhantomData)) + } + + fn fold_with<F>(self, folder: F) -> F + where + F: Folder<Self::Item>, + { + folder + } +} diff --git a/third_party/rust/rayon/src/iter/enumerate.rs b/third_party/rust/rayon/src/iter/enumerate.rs new file mode 100644 index 0000000000..980ee7cc7d --- /dev/null +++ b/third_party/rust/rayon/src/iter/enumerate.rs @@ -0,0 +1,133 @@ +use super::plumbing::*; +use super::*; +use std::iter; +use std::ops::Range; +use std::usize; + +/// `Enumerate` is an iterator that returns the current count along with the element. +/// This struct is created by the [`enumerate()`] method on [`IndexedParallelIterator`] +/// +/// [`enumerate()`]: trait.IndexedParallelIterator.html#method.enumerate +/// [`IndexedParallelIterator`]: trait.IndexedParallelIterator.html +#[must_use = "iterator adaptors are lazy and do nothing unless consumed"] +#[derive(Debug, Clone)] +pub struct Enumerate<I: IndexedParallelIterator> { + base: I, +} + +impl<I> Enumerate<I> +where + I: IndexedParallelIterator, +{ + /// Creates a new `Enumerate` iterator. + pub(super) fn new(base: I) -> Self { + Enumerate { base } + } +} + +impl<I> ParallelIterator for Enumerate<I> +where + I: IndexedParallelIterator, +{ + type Item = (usize, I::Item); + + fn drive_unindexed<C>(self, consumer: C) -> C::Result + where + C: UnindexedConsumer<Self::Item>, + { + bridge(self, consumer) + } + + fn opt_len(&self) -> Option<usize> { + Some(self.len()) + } +} + +impl<I> IndexedParallelIterator for Enumerate<I> +where + I: IndexedParallelIterator, +{ + fn drive<C: Consumer<Self::Item>>(self, consumer: C) -> C::Result { + bridge(self, consumer) + } + + fn len(&self) -> usize { + self.base.len() + } + + fn with_producer<CB>(self, callback: CB) -> CB::Output + where + CB: ProducerCallback<Self::Item>, + { + return self.base.with_producer(Callback { callback }); + + struct Callback<CB> { + callback: CB, + } + + impl<I, CB> ProducerCallback<I> for Callback<CB> + where + CB: ProducerCallback<(usize, I)>, + { + type Output = CB::Output; + fn callback<P>(self, base: P) -> CB::Output + where + P: Producer<Item = I>, + { + let producer = EnumerateProducer { base, offset: 0 }; + self.callback.callback(producer) + } + } + } +} + +/// //////////////////////////////////////////////////////////////////////// +/// Producer implementation + +struct EnumerateProducer<P> { + base: P, + offset: usize, +} + +impl<P> Producer for EnumerateProducer<P> +where + P: Producer, +{ + type Item = (usize, P::Item); + type IntoIter = iter::Zip<Range<usize>, P::IntoIter>; + + fn into_iter(self) -> Self::IntoIter { + // Enumerate only works for IndexedParallelIterators. Since those + // have a max length of usize::MAX, their max index is + // usize::MAX - 1, so the range 0..usize::MAX includes all + // possible indices. + // + // However, we should to use a precise end to the range, otherwise + // reversing the iterator may have to walk back a long ways before + // `Zip::next_back` can produce anything. + let base = self.base.into_iter(); + let end = self.offset + base.len(); + (self.offset..end).zip(base) + } + + fn min_len(&self) -> usize { + self.base.min_len() + } + fn max_len(&self) -> usize { + self.base.max_len() + } + + fn split_at(self, index: usize) -> (Self, Self) { + let (left, right) = self.base.split_at(index); + ( + EnumerateProducer { + base: left, + offset: self.offset, + }, + EnumerateProducer { + base: right, + offset: self.offset + index, + }, + ) + } +} diff --git a/third_party/rust/rayon/src/iter/extend.rs b/third_party/rust/rayon/src/iter/extend.rs new file mode 100644 index 0000000000..1769d476bd --- /dev/null +++ b/third_party/rust/rayon/src/iter/extend.rs @@ -0,0 +1,604 @@ +use super::noop::NoopConsumer; +use super::plumbing::{Consumer, Folder, Reducer, UnindexedConsumer}; +use super::{IntoParallelIterator, ParallelExtend, ParallelIterator}; + +use std::borrow::Cow; +use std::collections::LinkedList; +use std::collections::{BTreeMap, BTreeSet, HashMap, HashSet}; +use std::collections::{BinaryHeap, VecDeque}; +use std::hash::{BuildHasher, Hash}; + +/// Performs a generic `par_extend` by collecting to a `LinkedList<Vec<_>>` in +/// parallel, then extending the collection sequentially. +macro_rules! extend { + ($self:ident, $par_iter:ident, $extend:ident) => { + $extend( + $self, + $par_iter.into_par_iter().drive_unindexed(ListVecConsumer), + ); + }; +} + +/// Computes the total length of a `LinkedList<Vec<_>>`. +fn len<T>(list: &LinkedList<Vec<T>>) -> usize { + list.iter().map(Vec::len).sum() +} + +struct ListVecConsumer; + +struct ListVecFolder<T> { + vec: Vec<T>, +} + +impl<T: Send> Consumer<T> for ListVecConsumer { + type Folder = ListVecFolder<T>; + type Reducer = ListReducer; + type Result = LinkedList<Vec<T>>; + + fn split_at(self, _index: usize) -> (Self, Self, Self::Reducer) { + (Self, Self, ListReducer) + } + + fn into_folder(self) -> Self::Folder { + ListVecFolder { vec: Vec::new() } + } + + fn full(&self) -> bool { + false + } +} + +impl<T: Send> UnindexedConsumer<T> for ListVecConsumer { + fn split_off_left(&self) -> Self { + Self + } + + fn to_reducer(&self) -> Self::Reducer { + ListReducer + } +} + +impl<T> Folder<T> for ListVecFolder<T> { + type Result = LinkedList<Vec<T>>; + + fn consume(mut self, item: T) -> Self { + self.vec.push(item); + self + } + + fn consume_iter<I>(mut self, iter: I) -> Self + where + I: IntoIterator<Item = T>, + { + self.vec.extend(iter); + self + } + + fn complete(self) -> Self::Result { + let mut list = LinkedList::new(); + if !self.vec.is_empty() { + list.push_back(self.vec); + } + list + } + + fn full(&self) -> bool { + false + } +} + +fn heap_extend<T, Item>(heap: &mut BinaryHeap<T>, list: LinkedList<Vec<Item>>) +where + BinaryHeap<T>: Extend<Item>, +{ + heap.reserve(len(&list)); + for vec in list { + heap.extend(vec); + } +} + +/// Extends a binary heap with items from a parallel iterator. +impl<T> ParallelExtend<T> for BinaryHeap<T> +where + T: Ord + Send, +{ + fn par_extend<I>(&mut self, par_iter: I) + where + I: IntoParallelIterator<Item = T>, + { + extend!(self, par_iter, heap_extend); + } +} + +/// Extends a binary heap with copied items from a parallel iterator. +impl<'a, T> ParallelExtend<&'a T> for BinaryHeap<T> +where + T: 'a + Copy + Ord + Send + Sync, +{ + fn par_extend<I>(&mut self, par_iter: I) + where + I: IntoParallelIterator<Item = &'a T>, + { + extend!(self, par_iter, heap_extend); + } +} + +fn btree_map_extend<K, V, Item>(map: &mut BTreeMap<K, V>, list: LinkedList<Vec<Item>>) +where + BTreeMap<K, V>: Extend<Item>, +{ + for vec in list { + map.extend(vec); + } +} + +/// Extends a B-tree map with items from a parallel iterator. +impl<K, V> ParallelExtend<(K, V)> for BTreeMap<K, V> +where + K: Ord + Send, + V: Send, +{ + fn par_extend<I>(&mut self, par_iter: I) + where + I: IntoParallelIterator<Item = (K, V)>, + { + extend!(self, par_iter, btree_map_extend); + } +} + +/// Extends a B-tree map with copied items from a parallel iterator. +impl<'a, K: 'a, V: 'a> ParallelExtend<(&'a K, &'a V)> for BTreeMap<K, V> +where + K: Copy + Ord + Send + Sync, + V: Copy + Send + Sync, +{ + fn par_extend<I>(&mut self, par_iter: I) + where + I: IntoParallelIterator<Item = (&'a K, &'a V)>, + { + extend!(self, par_iter, btree_map_extend); + } +} + +fn btree_set_extend<T, Item>(set: &mut BTreeSet<T>, list: LinkedList<Vec<Item>>) +where + BTreeSet<T>: Extend<Item>, +{ + for vec in list { + set.extend(vec); + } +} + +/// Extends a B-tree set with items from a parallel iterator. +impl<T> ParallelExtend<T> for BTreeSet<T> +where + T: Ord + Send, +{ + fn par_extend<I>(&mut self, par_iter: I) + where + I: IntoParallelIterator<Item = T>, + { + extend!(self, par_iter, btree_set_extend); + } +} + +/// Extends a B-tree set with copied items from a parallel iterator. +impl<'a, T> ParallelExtend<&'a T> for BTreeSet<T> +where + T: 'a + Copy + Ord + Send + Sync, +{ + fn par_extend<I>(&mut self, par_iter: I) + where + I: IntoParallelIterator<Item = &'a T>, + { + extend!(self, par_iter, btree_set_extend); + } +} + +fn hash_map_extend<K, V, S, Item>(map: &mut HashMap<K, V, S>, list: LinkedList<Vec<Item>>) +where + HashMap<K, V, S>: Extend<Item>, + K: Eq + Hash, + S: BuildHasher, +{ + map.reserve(len(&list)); + for vec in list { + map.extend(vec); + } +} + +/// Extends a hash map with items from a parallel iterator. +impl<K, V, S> ParallelExtend<(K, V)> for HashMap<K, V, S> +where + K: Eq + Hash + Send, + V: Send, + S: BuildHasher + Send, +{ + fn par_extend<I>(&mut self, par_iter: I) + where + I: IntoParallelIterator<Item = (K, V)>, + { + // See the map_collect benchmarks in rayon-demo for different strategies. + extend!(self, par_iter, hash_map_extend); + } +} + +/// Extends a hash map with copied items from a parallel iterator. +impl<'a, K: 'a, V: 'a, S> ParallelExtend<(&'a K, &'a V)> for HashMap<K, V, S> +where + K: Copy + Eq + Hash + Send + Sync, + V: Copy + Send + Sync, + S: BuildHasher + Send, +{ + fn par_extend<I>(&mut self, par_iter: I) + where + I: IntoParallelIterator<Item = (&'a K, &'a V)>, + { + extend!(self, par_iter, hash_map_extend); + } +} + +fn hash_set_extend<T, S, Item>(set: &mut HashSet<T, S>, list: LinkedList<Vec<Item>>) +where + HashSet<T, S>: Extend<Item>, + T: Eq + Hash, + S: BuildHasher, +{ + set.reserve(len(&list)); + for vec in list { + set.extend(vec); + } +} + +/// Extends a hash set with items from a parallel iterator. +impl<T, S> ParallelExtend<T> for HashSet<T, S> +where + T: Eq + Hash + Send, + S: BuildHasher + Send, +{ + fn par_extend<I>(&mut self, par_iter: I) + where + I: IntoParallelIterator<Item = T>, + { + extend!(self, par_iter, hash_set_extend); + } +} + +/// Extends a hash set with copied items from a parallel iterator. +impl<'a, T, S> ParallelExtend<&'a T> for HashSet<T, S> +where + T: 'a + Copy + Eq + Hash + Send + Sync, + S: BuildHasher + Send, +{ + fn par_extend<I>(&mut self, par_iter: I) + where + I: IntoParallelIterator<Item = &'a T>, + { + extend!(self, par_iter, hash_set_extend); + } +} + +/// Extends a linked list with items from a parallel iterator. +impl<T> ParallelExtend<T> for LinkedList<T> +where + T: Send, +{ + fn par_extend<I>(&mut self, par_iter: I) + where + I: IntoParallelIterator<Item = T>, + { + let mut list = par_iter.into_par_iter().drive_unindexed(ListConsumer); + self.append(&mut list); + } +} + +/// Extends a linked list with copied items from a parallel iterator. +impl<'a, T> ParallelExtend<&'a T> for LinkedList<T> +where + T: 'a + Copy + Send + Sync, +{ + fn par_extend<I>(&mut self, par_iter: I) + where + I: IntoParallelIterator<Item = &'a T>, + { + self.par_extend(par_iter.into_par_iter().copied()) + } +} + +struct ListConsumer; + +struct ListFolder<T> { + list: LinkedList<T>, +} + +struct ListReducer; + +impl<T: Send> Consumer<T> for ListConsumer { + type Folder = ListFolder<T>; + type Reducer = ListReducer; + type Result = LinkedList<T>; + + fn split_at(self, _index: usize) -> (Self, Self, Self::Reducer) { + (Self, Self, ListReducer) + } + + fn into_folder(self) -> Self::Folder { + ListFolder { + list: LinkedList::new(), + } + } + + fn full(&self) -> bool { + false + } +} + +impl<T: Send> UnindexedConsumer<T> for ListConsumer { + fn split_off_left(&self) -> Self { + Self + } + + fn to_reducer(&self) -> Self::Reducer { + ListReducer + } +} + +impl<T> Folder<T> for ListFolder<T> { + type Result = LinkedList<T>; + + fn consume(mut self, item: T) -> Self { + self.list.push_back(item); + self + } + + fn consume_iter<I>(mut self, iter: I) -> Self + where + I: IntoIterator<Item = T>, + { + self.list.extend(iter); + self + } + + fn complete(self) -> Self::Result { + self.list + } + + fn full(&self) -> bool { + false + } +} + +impl<T> Reducer<LinkedList<T>> for ListReducer { + fn reduce(self, mut left: LinkedList<T>, mut right: LinkedList<T>) -> LinkedList<T> { + left.append(&mut right); + left + } +} + +fn flat_string_extend(string: &mut String, list: LinkedList<String>) { + string.reserve(list.iter().map(String::len).sum()); + string.extend(list); +} + +/// Extends a string with characters from a parallel iterator. +impl ParallelExtend<char> for String { + fn par_extend<I>(&mut self, par_iter: I) + where + I: IntoParallelIterator<Item = char>, + { + // This is like `extend`, but `Vec<char>` is less efficient to deal + // with than `String`, so instead collect to `LinkedList<String>`. + let list = par_iter.into_par_iter().drive_unindexed(ListStringConsumer); + flat_string_extend(self, list); + } +} + +/// Extends a string with copied characters from a parallel iterator. +impl<'a> ParallelExtend<&'a char> for String { + fn par_extend<I>(&mut self, par_iter: I) + where + I: IntoParallelIterator<Item = &'a char>, + { + self.par_extend(par_iter.into_par_iter().copied()) + } +} + +struct ListStringConsumer; + +struct ListStringFolder { + string: String, +} + +impl Consumer<char> for ListStringConsumer { + type Folder = ListStringFolder; + type Reducer = ListReducer; + type Result = LinkedList<String>; + + fn split_at(self, _index: usize) -> (Self, Self, Self::Reducer) { + (Self, Self, ListReducer) + } + + fn into_folder(self) -> Self::Folder { + ListStringFolder { + string: String::new(), + } + } + + fn full(&self) -> bool { + false + } +} + +impl UnindexedConsumer<char> for ListStringConsumer { + fn split_off_left(&self) -> Self { + Self + } + + fn to_reducer(&self) -> Self::Reducer { + ListReducer + } +} + +impl Folder<char> for ListStringFolder { + type Result = LinkedList<String>; + + fn consume(mut self, item: char) -> Self { + self.string.push(item); + self + } + + fn consume_iter<I>(mut self, iter: I) -> Self + where + I: IntoIterator<Item = char>, + { + self.string.extend(iter); + self + } + + fn complete(self) -> Self::Result { + let mut list = LinkedList::new(); + if !self.string.is_empty() { + list.push_back(self.string); + } + list + } + + fn full(&self) -> bool { + false + } +} + +fn string_extend<Item>(string: &mut String, list: LinkedList<Vec<Item>>) +where + String: Extend<Item>, + Item: AsRef<str>, +{ + let len = list.iter().flatten().map(Item::as_ref).map(str::len).sum(); + string.reserve(len); + for vec in list { + string.extend(vec); + } +} + +/// Extends a string with string slices from a parallel iterator. +impl<'a> ParallelExtend<&'a str> for String { + fn par_extend<I>(&mut self, par_iter: I) + where + I: IntoParallelIterator<Item = &'a str>, + { + extend!(self, par_iter, string_extend); + } +} + +/// Extends a string with strings from a parallel iterator. +impl ParallelExtend<String> for String { + fn par_extend<I>(&mut self, par_iter: I) + where + I: IntoParallelIterator<Item = String>, + { + extend!(self, par_iter, string_extend); + } +} + +/// Extends a string with string slices from a parallel iterator. +impl<'a> ParallelExtend<Cow<'a, str>> for String { + fn par_extend<I>(&mut self, par_iter: I) + where + I: IntoParallelIterator<Item = Cow<'a, str>>, + { + extend!(self, par_iter, string_extend); + } +} + +fn deque_extend<T, Item>(deque: &mut VecDeque<T>, list: LinkedList<Vec<Item>>) +where + VecDeque<T>: Extend<Item>, +{ + deque.reserve(len(&list)); + for vec in list { + deque.extend(vec); + } +} + +/// Extends a deque with items from a parallel iterator. +impl<T> ParallelExtend<T> for VecDeque<T> +where + T: Send, +{ + fn par_extend<I>(&mut self, par_iter: I) + where + I: IntoParallelIterator<Item = T>, + { + extend!(self, par_iter, deque_extend); + } +} + +/// Extends a deque with copied items from a parallel iterator. +impl<'a, T> ParallelExtend<&'a T> for VecDeque<T> +where + T: 'a + Copy + Send + Sync, +{ + fn par_extend<I>(&mut self, par_iter: I) + where + I: IntoParallelIterator<Item = &'a T>, + { + extend!(self, par_iter, deque_extend); + } +} + +fn vec_append<T>(vec: &mut Vec<T>, list: LinkedList<Vec<T>>) { + vec.reserve(len(&list)); + for mut other in list { + vec.append(&mut other); + } +} + +/// Extends a vector with items from a parallel iterator. +impl<T> ParallelExtend<T> for Vec<T> +where + T: Send, +{ + fn par_extend<I>(&mut self, par_iter: I) + where + I: IntoParallelIterator<Item = T>, + { + // See the vec_collect benchmarks in rayon-demo for different strategies. + let par_iter = par_iter.into_par_iter(); + match par_iter.opt_len() { + Some(len) => { + // When Rust gets specialization, we can get here for indexed iterators + // without relying on `opt_len`. Until then, `special_extend()` fakes + // an unindexed mode on the promise that `opt_len()` is accurate. + super::collect::special_extend(par_iter, len, self); + } + None => { + // This works like `extend`, but `Vec::append` is more efficient. + let list = par_iter.drive_unindexed(ListVecConsumer); + vec_append(self, list); + } + } + } +} + +/// Extends a vector with copied items from a parallel iterator. +impl<'a, T> ParallelExtend<&'a T> for Vec<T> +where + T: 'a + Copy + Send + Sync, +{ + fn par_extend<I>(&mut self, par_iter: I) + where + I: IntoParallelIterator<Item = &'a T>, + { + self.par_extend(par_iter.into_par_iter().copied()) + } +} + +/// Collapses all unit items from a parallel iterator into one. +impl ParallelExtend<()> for () { + fn par_extend<I>(&mut self, par_iter: I) + where + I: IntoParallelIterator<Item = ()>, + { + par_iter.into_par_iter().drive_unindexed(NoopConsumer) + } +} diff --git a/third_party/rust/rayon/src/iter/filter.rs b/third_party/rust/rayon/src/iter/filter.rs new file mode 100644 index 0000000000..e1b74ba520 --- /dev/null +++ b/third_party/rust/rayon/src/iter/filter.rs @@ -0,0 +1,141 @@ +use super::plumbing::*; +use super::*; + +use std::fmt::{self, Debug}; + +/// `Filter` takes a predicate `filter_op` and filters out elements that match. +/// This struct is created by the [`filter()`] method on [`ParallelIterator`] +/// +/// [`filter()`]: trait.ParallelIterator.html#method.filter +/// [`ParallelIterator`]: trait.ParallelIterator.html +#[must_use = "iterator adaptors are lazy and do nothing unless consumed"] +#[derive(Clone)] +pub struct Filter<I: ParallelIterator, P> { + base: I, + filter_op: P, +} + +impl<I: ParallelIterator + Debug, P> Debug for Filter<I, P> { + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + f.debug_struct("Filter").field("base", &self.base).finish() + } +} + +impl<I, P> Filter<I, P> +where + I: ParallelIterator, +{ + /// Creates a new `Filter` iterator. + pub(super) fn new(base: I, filter_op: P) -> Self { + Filter { base, filter_op } + } +} + +impl<I, P> ParallelIterator for Filter<I, P> +where + I: ParallelIterator, + P: Fn(&I::Item) -> bool + Sync + Send, +{ + type Item = I::Item; + + fn drive_unindexed<C>(self, consumer: C) -> C::Result + where + C: UnindexedConsumer<Self::Item>, + { + let consumer1 = FilterConsumer::new(consumer, &self.filter_op); + self.base.drive_unindexed(consumer1) + } +} + +/// //////////////////////////////////////////////////////////////////////// +/// Consumer implementation + +struct FilterConsumer<'p, C, P> { + base: C, + filter_op: &'p P, +} + +impl<'p, C, P> FilterConsumer<'p, C, P> { + fn new(base: C, filter_op: &'p P) -> Self { + FilterConsumer { base, filter_op } + } +} + +impl<'p, T, C, P: 'p> Consumer<T> for FilterConsumer<'p, C, P> +where + C: Consumer<T>, + P: Fn(&T) -> bool + Sync, +{ + type Folder = FilterFolder<'p, C::Folder, P>; + type Reducer = C::Reducer; + type Result = C::Result; + + fn split_at(self, index: usize) -> (Self, Self, C::Reducer) { + let (left, right, reducer) = self.base.split_at(index); + ( + FilterConsumer::new(left, self.filter_op), + FilterConsumer::new(right, self.filter_op), + reducer, + ) + } + + fn into_folder(self) -> Self::Folder { + FilterFolder { + base: self.base.into_folder(), + filter_op: self.filter_op, + } + } + + fn full(&self) -> bool { + self.base.full() + } +} + +impl<'p, T, C, P: 'p> UnindexedConsumer<T> for FilterConsumer<'p, C, P> +where + C: UnindexedConsumer<T>, + P: Fn(&T) -> bool + Sync, +{ + fn split_off_left(&self) -> Self { + FilterConsumer::new(self.base.split_off_left(), self.filter_op) + } + + fn to_reducer(&self) -> Self::Reducer { + self.base.to_reducer() + } +} + +struct FilterFolder<'p, C, P> { + base: C, + filter_op: &'p P, +} + +impl<'p, C, P, T> Folder<T> for FilterFolder<'p, C, P> +where + C: Folder<T>, + P: Fn(&T) -> bool + 'p, +{ + type Result = C::Result; + + fn consume(self, item: T) -> Self { + let filter_op = self.filter_op; + if filter_op(&item) { + let base = self.base.consume(item); + FilterFolder { base, filter_op } + } else { + self + } + } + + // This cannot easily specialize `consume_iter` to be better than + // the default, because that requires checking `self.base.full()` + // during a call to `self.base.consume_iter()`. (#632) + + fn complete(self) -> Self::Result { + self.base.complete() + } + + fn full(&self) -> bool { + self.base.full() + } +} diff --git a/third_party/rust/rayon/src/iter/filter_map.rs b/third_party/rust/rayon/src/iter/filter_map.rs new file mode 100644 index 0000000000..db1c7e3f02 --- /dev/null +++ b/third_party/rust/rayon/src/iter/filter_map.rs @@ -0,0 +1,142 @@ +use super::plumbing::*; +use super::*; + +use std::fmt::{self, Debug}; + +/// `FilterMap` creates an iterator that uses `filter_op` to both filter and map elements. +/// This struct is created by the [`filter_map()`] method on [`ParallelIterator`]. +/// +/// [`filter_map()`]: trait.ParallelIterator.html#method.filter_map +/// [`ParallelIterator`]: trait.ParallelIterator.html +#[must_use = "iterator adaptors are lazy and do nothing unless consumed"] +#[derive(Clone)] +pub struct FilterMap<I: ParallelIterator, P> { + base: I, + filter_op: P, +} + +impl<I: ParallelIterator + Debug, P> Debug for FilterMap<I, P> { + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + f.debug_struct("FilterMap") + .field("base", &self.base) + .finish() + } +} + +impl<I: ParallelIterator, P> FilterMap<I, P> { + /// Creates a new `FilterMap` iterator. + pub(super) fn new(base: I, filter_op: P) -> Self { + FilterMap { base, filter_op } + } +} + +impl<I, P, R> ParallelIterator for FilterMap<I, P> +where + I: ParallelIterator, + P: Fn(I::Item) -> Option<R> + Sync + Send, + R: Send, +{ + type Item = R; + + fn drive_unindexed<C>(self, consumer: C) -> C::Result + where + C: UnindexedConsumer<Self::Item>, + { + let consumer = FilterMapConsumer::new(consumer, &self.filter_op); + self.base.drive_unindexed(consumer) + } +} + +/// //////////////////////////////////////////////////////////////////////// +/// Consumer implementation + +struct FilterMapConsumer<'p, C, P> { + base: C, + filter_op: &'p P, +} + +impl<'p, C, P: 'p> FilterMapConsumer<'p, C, P> { + fn new(base: C, filter_op: &'p P) -> Self { + FilterMapConsumer { base, filter_op } + } +} + +impl<'p, T, U, C, P> Consumer<T> for FilterMapConsumer<'p, C, P> +where + C: Consumer<U>, + P: Fn(T) -> Option<U> + Sync + 'p, +{ + type Folder = FilterMapFolder<'p, C::Folder, P>; + type Reducer = C::Reducer; + type Result = C::Result; + + fn split_at(self, index: usize) -> (Self, Self, Self::Reducer) { + let (left, right, reducer) = self.base.split_at(index); + ( + FilterMapConsumer::new(left, self.filter_op), + FilterMapConsumer::new(right, self.filter_op), + reducer, + ) + } + + fn into_folder(self) -> Self::Folder { + let base = self.base.into_folder(); + FilterMapFolder { + base, + filter_op: self.filter_op, + } + } + + fn full(&self) -> bool { + self.base.full() + } +} + +impl<'p, T, U, C, P> UnindexedConsumer<T> for FilterMapConsumer<'p, C, P> +where + C: UnindexedConsumer<U>, + P: Fn(T) -> Option<U> + Sync + 'p, +{ + fn split_off_left(&self) -> Self { + FilterMapConsumer::new(self.base.split_off_left(), self.filter_op) + } + + fn to_reducer(&self) -> Self::Reducer { + self.base.to_reducer() + } +} + +struct FilterMapFolder<'p, C, P> { + base: C, + filter_op: &'p P, +} + +impl<'p, T, U, C, P> Folder<T> for FilterMapFolder<'p, C, P> +where + C: Folder<U>, + P: Fn(T) -> Option<U> + Sync + 'p, +{ + type Result = C::Result; + + fn consume(self, item: T) -> Self { + let filter_op = self.filter_op; + if let Some(mapped_item) = filter_op(item) { + let base = self.base.consume(mapped_item); + FilterMapFolder { base, filter_op } + } else { + self + } + } + + // This cannot easily specialize `consume_iter` to be better than + // the default, because that requires checking `self.base.full()` + // during a call to `self.base.consume_iter()`. (#632) + + fn complete(self) -> C::Result { + self.base.complete() + } + + fn full(&self) -> bool { + self.base.full() + } +} diff --git a/third_party/rust/rayon/src/iter/find.rs b/third_party/rust/rayon/src/iter/find.rs new file mode 100644 index 0000000000..b16ee84466 --- /dev/null +++ b/third_party/rust/rayon/src/iter/find.rs @@ -0,0 +1,120 @@ +use super::plumbing::*; +use super::*; +use std::sync::atomic::{AtomicBool, Ordering}; + +pub(super) fn find<I, P>(pi: I, find_op: P) -> Option<I::Item> +where + I: ParallelIterator, + P: Fn(&I::Item) -> bool + Sync, +{ + let found = AtomicBool::new(false); + let consumer = FindConsumer::new(&find_op, &found); + pi.drive_unindexed(consumer) +} + +struct FindConsumer<'p, P> { + find_op: &'p P, + found: &'p AtomicBool, +} + +impl<'p, P> FindConsumer<'p, P> { + fn new(find_op: &'p P, found: &'p AtomicBool) -> Self { + FindConsumer { find_op, found } + } +} + +impl<'p, T, P: 'p> Consumer<T> for FindConsumer<'p, P> +where + T: Send, + P: Fn(&T) -> bool + Sync, +{ + type Folder = FindFolder<'p, T, P>; + type Reducer = FindReducer; + type Result = Option<T>; + + fn split_at(self, _index: usize) -> (Self, Self, Self::Reducer) { + (self.split_off_left(), self, FindReducer) + } + + fn into_folder(self) -> Self::Folder { + FindFolder { + find_op: self.find_op, + found: self.found, + item: None, + } + } + + fn full(&self) -> bool { + self.found.load(Ordering::Relaxed) + } +} + +impl<'p, T, P: 'p> UnindexedConsumer<T> for FindConsumer<'p, P> +where + T: Send, + P: Fn(&T) -> bool + Sync, +{ + fn split_off_left(&self) -> Self { + FindConsumer::new(self.find_op, self.found) + } + + fn to_reducer(&self) -> Self::Reducer { + FindReducer + } +} + +struct FindFolder<'p, T, P> { + find_op: &'p P, + found: &'p AtomicBool, + item: Option<T>, +} + +impl<'p, T, P> Folder<T> for FindFolder<'p, T, P> +where + P: Fn(&T) -> bool + 'p, +{ + type Result = Option<T>; + + fn consume(mut self, item: T) -> Self { + if (self.find_op)(&item) { + self.found.store(true, Ordering::Relaxed); + self.item = Some(item); + } + self + } + + fn consume_iter<I>(mut self, iter: I) -> Self + where + I: IntoIterator<Item = T>, + { + fn not_full<T>(found: &AtomicBool) -> impl Fn(&T) -> bool + '_ { + move |_| !found.load(Ordering::Relaxed) + } + + self.item = iter + .into_iter() + // stop iterating if another thread has found something + .take_while(not_full(self.found)) + .find(self.find_op); + if self.item.is_some() { + self.found.store(true, Ordering::Relaxed) + } + self + } + + fn complete(self) -> Self::Result { + self.item + } + + fn full(&self) -> bool { + self.found.load(Ordering::Relaxed) + } +} + +struct FindReducer; + +impl<T> Reducer<Option<T>> for FindReducer { + fn reduce(self, left: Option<T>, right: Option<T>) -> Option<T> { + left.or(right) + } +} diff --git a/third_party/rust/rayon/src/iter/find_first_last/mod.rs b/third_party/rust/rayon/src/iter/find_first_last/mod.rs new file mode 100644 index 0000000000..e5da8f0dd8 --- /dev/null +++ b/third_party/rust/rayon/src/iter/find_first_last/mod.rs @@ -0,0 +1,238 @@ +use super::plumbing::*; +use super::*; +use std::cell::Cell; +use std::sync::atomic::{AtomicUsize, Ordering}; + +#[cfg(test)] +mod test; + +// The key optimization for find_first is that a consumer can stop its search if +// some consumer to its left already found a match (and similarly for consumers +// to the right for find_last). To make this work, all consumers need some +// notion of their position in the data relative to other consumers, including +// unindexed consumers that have no built-in notion of position. +// +// To solve this, we assign each consumer a lower and upper bound for an +// imaginary "range" of data that it consumes. The initial consumer starts with +// the range 0..usize::max_value(). The split divides this range in half so that +// one resulting consumer has the range 0..(usize::max_value() / 2), and the +// other has (usize::max_value() / 2)..usize::max_value(). Every subsequent +// split divides the range in half again until it cannot be split anymore +// (i.e. its length is 1), in which case the split returns two consumers with +// the same range. In that case both consumers will continue to consume all +// their data regardless of whether a better match is found, but the reducer +// will still return the correct answer. + +#[derive(Copy, Clone)] +enum MatchPosition { + Leftmost, + Rightmost, +} + +/// Returns true if pos1 is a better match than pos2 according to MatchPosition +#[inline] +fn better_position(pos1: usize, pos2: usize, mp: MatchPosition) -> bool { + match mp { + MatchPosition::Leftmost => pos1 < pos2, + MatchPosition::Rightmost => pos1 > pos2, + } +} + +pub(super) fn find_first<I, P>(pi: I, find_op: P) -> Option<I::Item> +where + I: ParallelIterator, + P: Fn(&I::Item) -> bool + Sync, +{ + let best_found = AtomicUsize::new(usize::max_value()); + let consumer = FindConsumer::new(&find_op, MatchPosition::Leftmost, &best_found); + pi.drive_unindexed(consumer) +} + +pub(super) fn find_last<I, P>(pi: I, find_op: P) -> Option<I::Item> +where + I: ParallelIterator, + P: Fn(&I::Item) -> bool + Sync, +{ + let best_found = AtomicUsize::new(0); + let consumer = FindConsumer::new(&find_op, MatchPosition::Rightmost, &best_found); + pi.drive_unindexed(consumer) +} + +struct FindConsumer<'p, P> { + find_op: &'p P, + lower_bound: Cell<usize>, + upper_bound: usize, + match_position: MatchPosition, + best_found: &'p AtomicUsize, +} + +impl<'p, P> FindConsumer<'p, P> { + fn new(find_op: &'p P, match_position: MatchPosition, best_found: &'p AtomicUsize) -> Self { + FindConsumer { + find_op, + lower_bound: Cell::new(0), + upper_bound: usize::max_value(), + match_position, + best_found, + } + } + + fn current_index(&self) -> usize { + match self.match_position { + MatchPosition::Leftmost => self.lower_bound.get(), + MatchPosition::Rightmost => self.upper_bound, + } + } +} + +impl<'p, T, P> Consumer<T> for FindConsumer<'p, P> +where + T: Send, + P: Fn(&T) -> bool + Sync, +{ + type Folder = FindFolder<'p, T, P>; + type Reducer = FindReducer; + type Result = Option<T>; + + fn split_at(self, _index: usize) -> (Self, Self, Self::Reducer) { + let dir = self.match_position; + ( + self.split_off_left(), + self, + FindReducer { + match_position: dir, + }, + ) + } + + fn into_folder(self) -> Self::Folder { + FindFolder { + find_op: self.find_op, + boundary: self.current_index(), + match_position: self.match_position, + best_found: self.best_found, + item: None, + } + } + + fn full(&self) -> bool { + // can stop consuming if the best found index so far is *strictly* + // better than anything this consumer will find + better_position( + self.best_found.load(Ordering::Relaxed), + self.current_index(), + self.match_position, + ) + } +} + +impl<'p, T, P> UnindexedConsumer<T> for FindConsumer<'p, P> +where + T: Send, + P: Fn(&T) -> bool + Sync, +{ + fn split_off_left(&self) -> Self { + // Upper bound for one consumer will be lower bound for the other. This + // overlap is okay, because only one of the bounds will be used for + // comparing against best_found; the other is kept only to be able to + // divide the range in half. + // + // When the resolution of usize has been exhausted (i.e. when + // upper_bound = lower_bound), both results of this split will have the + // same range. When that happens, we lose the ability to tell one + // consumer to stop working when the other finds a better match, but the + // reducer ensures that the best answer is still returned (see the test + // above). + let old_lower_bound = self.lower_bound.get(); + let median = old_lower_bound + ((self.upper_bound - old_lower_bound) / 2); + self.lower_bound.set(median); + + FindConsumer { + find_op: self.find_op, + lower_bound: Cell::new(old_lower_bound), + upper_bound: median, + match_position: self.match_position, + best_found: self.best_found, + } + } + + fn to_reducer(&self) -> Self::Reducer { + FindReducer { + match_position: self.match_position, + } + } +} + +struct FindFolder<'p, T, P> { + find_op: &'p P, + boundary: usize, + match_position: MatchPosition, + best_found: &'p AtomicUsize, + item: Option<T>, +} + +impl<'p, P: 'p + Fn(&T) -> bool, T> Folder<T> for FindFolder<'p, T, P> { + type Result = Option<T>; + + fn consume(mut self, item: T) -> Self { + let found_best_in_range = match self.match_position { + MatchPosition::Leftmost => self.item.is_some(), + MatchPosition::Rightmost => false, + }; + + if !found_best_in_range && (self.find_op)(&item) { + // Continuously try to set best_found until we succeed or we + // discover a better match was already found. + let mut current = self.best_found.load(Ordering::Relaxed); + loop { + if better_position(current, self.boundary, self.match_position) { + break; + } + match self.best_found.compare_exchange_weak( + current, + self.boundary, + Ordering::Relaxed, + Ordering::Relaxed, + ) { + Ok(_) => { + self.item = Some(item); + break; + } + Err(v) => current = v, + } + } + } + self + } + + fn complete(self) -> Self::Result { + self.item + } + + fn full(&self) -> bool { + let found_best_in_range = match self.match_position { + MatchPosition::Leftmost => self.item.is_some(), + MatchPosition::Rightmost => false, + }; + + found_best_in_range + || better_position( + self.best_found.load(Ordering::Relaxed), + self.boundary, + self.match_position, + ) + } +} + +struct FindReducer { + match_position: MatchPosition, +} + +impl<T> Reducer<Option<T>> for FindReducer { + fn reduce(self, left: Option<T>, right: Option<T>) -> Option<T> { + match self.match_position { + MatchPosition::Leftmost => left.or(right), + MatchPosition::Rightmost => right.or(left), + } + } +} diff --git a/third_party/rust/rayon/src/iter/find_first_last/test.rs b/third_party/rust/rayon/src/iter/find_first_last/test.rs new file mode 100644 index 0000000000..05271bc5b3 --- /dev/null +++ b/third_party/rust/rayon/src/iter/find_first_last/test.rs @@ -0,0 +1,106 @@ +use super::*; +use std::sync::atomic::AtomicUsize; + +#[test] +fn same_range_first_consumers_return_correct_answer() { + let find_op = |x: &i32| x % 2 == 0; + let first_found = AtomicUsize::new(usize::max_value()); + let far_right_consumer = FindConsumer::new(&find_op, MatchPosition::Leftmost, &first_found); + + // We save a consumer that will be far to the right of the main consumer (and therefore not + // sharing an index range with that consumer) for fullness testing + let consumer = far_right_consumer.split_off_left(); + + // split until we have an indivisible range + let bits_in_usize = usize::min_value().count_zeros(); + + for _ in 0..bits_in_usize { + consumer.split_off_left(); + } + + let reducer = consumer.to_reducer(); + // the left and right folders should now have the same range, having + // exhausted the resolution of usize + let left_folder = consumer.split_off_left().into_folder(); + let right_folder = consumer.into_folder(); + + let left_folder = left_folder.consume(0).consume(1); + assert_eq!(left_folder.boundary, right_folder.boundary); + // expect not full even though a better match has been found because the + // ranges are the same + assert!(!right_folder.full()); + assert!(far_right_consumer.full()); + let right_folder = right_folder.consume(2).consume(3); + assert_eq!( + reducer.reduce(left_folder.complete(), right_folder.complete()), + Some(0) + ); +} + +#[test] +fn same_range_last_consumers_return_correct_answer() { + let find_op = |x: &i32| x % 2 == 0; + let last_found = AtomicUsize::new(0); + let consumer = FindConsumer::new(&find_op, MatchPosition::Rightmost, &last_found); + + // We save a consumer that will be far to the left of the main consumer (and therefore not + // sharing an index range with that consumer) for fullness testing + let far_left_consumer = consumer.split_off_left(); + + // split until we have an indivisible range + let bits_in_usize = usize::min_value().count_zeros(); + for _ in 0..bits_in_usize { + consumer.split_off_left(); + } + + let reducer = consumer.to_reducer(); + // due to the exact calculation in split_off_left, the very last consumer has a + // range of width 2, so we use the second-to-last consumer instead to get + // the same boundary on both folders + let consumer = consumer.split_off_left(); + let left_folder = consumer.split_off_left().into_folder(); + let right_folder = consumer.into_folder(); + let right_folder = right_folder.consume(2).consume(3); + assert_eq!(left_folder.boundary, right_folder.boundary); + // expect not full even though a better match has been found because the + // ranges are the same + assert!(!left_folder.full()); + assert!(far_left_consumer.full()); + let left_folder = left_folder.consume(0).consume(1); + assert_eq!( + reducer.reduce(left_folder.complete(), right_folder.complete()), + Some(2) + ); +} + +// These tests requires that a folder be assigned to an iterator with more than +// one element. We can't necessarily determine when that will happen for a given +// input to find_first/find_last, so we test the folder directly here instead. +#[test] +fn find_first_folder_does_not_clobber_first_found() { + let best_found = AtomicUsize::new(usize::max_value()); + let f = FindFolder { + find_op: &(|&_: &i32| -> bool { true }), + boundary: 0, + match_position: MatchPosition::Leftmost, + best_found: &best_found, + item: None, + }; + let f = f.consume(0_i32).consume(1_i32).consume(2_i32); + assert!(f.full()); + assert_eq!(f.complete(), Some(0_i32)); +} + +#[test] +fn find_last_folder_yields_last_match() { + let best_found = AtomicUsize::new(0); + let f = FindFolder { + find_op: &(|&_: &i32| -> bool { true }), + boundary: 0, + match_position: MatchPosition::Rightmost, + best_found: &best_found, + item: None, + }; + let f = f.consume(0_i32).consume(1_i32).consume(2_i32); + assert_eq!(f.complete(), Some(2_i32)); +} diff --git a/third_party/rust/rayon/src/iter/flat_map.rs b/third_party/rust/rayon/src/iter/flat_map.rs new file mode 100644 index 0000000000..f264e1e144 --- /dev/null +++ b/third_party/rust/rayon/src/iter/flat_map.rs @@ -0,0 +1,154 @@ +use super::plumbing::*; +use super::*; + +use std::fmt::{self, Debug}; + +/// `FlatMap` maps each element to a parallel iterator, then flattens these iterators together. +/// This struct is created by the [`flat_map()`] method on [`ParallelIterator`] +/// +/// [`flat_map()`]: trait.ParallelIterator.html#method.flat_map +/// [`ParallelIterator`]: trait.ParallelIterator.html +#[must_use = "iterator adaptors are lazy and do nothing unless consumed"] +#[derive(Clone)] +pub struct FlatMap<I: ParallelIterator, F> { + base: I, + map_op: F, +} + +impl<I: ParallelIterator + Debug, F> Debug for FlatMap<I, F> { + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + f.debug_struct("FlatMap").field("base", &self.base).finish() + } +} + +impl<I: ParallelIterator, F> FlatMap<I, F> { + /// Creates a new `FlatMap` iterator. + pub(super) fn new(base: I, map_op: F) -> Self { + FlatMap { base, map_op } + } +} + +impl<I, F, PI> ParallelIterator for FlatMap<I, F> +where + I: ParallelIterator, + F: Fn(I::Item) -> PI + Sync + Send, + PI: IntoParallelIterator, +{ + type Item = PI::Item; + + fn drive_unindexed<C>(self, consumer: C) -> C::Result + where + C: UnindexedConsumer<Self::Item>, + { + let consumer = FlatMapConsumer::new(consumer, &self.map_op); + self.base.drive_unindexed(consumer) + } +} + +/// //////////////////////////////////////////////////////////////////////// +/// Consumer implementation + +struct FlatMapConsumer<'f, C, F> { + base: C, + map_op: &'f F, +} + +impl<'f, C, F> FlatMapConsumer<'f, C, F> { + fn new(base: C, map_op: &'f F) -> Self { + FlatMapConsumer { base, map_op } + } +} + +impl<'f, T, U, C, F> Consumer<T> for FlatMapConsumer<'f, C, F> +where + C: UnindexedConsumer<U::Item>, + F: Fn(T) -> U + Sync, + U: IntoParallelIterator, +{ + type Folder = FlatMapFolder<'f, C, F, C::Result>; + type Reducer = C::Reducer; + type Result = C::Result; + + fn split_at(self, index: usize) -> (Self, Self, C::Reducer) { + let (left, right, reducer) = self.base.split_at(index); + ( + FlatMapConsumer::new(left, self.map_op), + FlatMapConsumer::new(right, self.map_op), + reducer, + ) + } + + fn into_folder(self) -> Self::Folder { + FlatMapFolder { + base: self.base, + map_op: self.map_op, + previous: None, + } + } + + fn full(&self) -> bool { + self.base.full() + } +} + +impl<'f, T, U, C, F> UnindexedConsumer<T> for FlatMapConsumer<'f, C, F> +where + C: UnindexedConsumer<U::Item>, + F: Fn(T) -> U + Sync, + U: IntoParallelIterator, +{ + fn split_off_left(&self) -> Self { + FlatMapConsumer::new(self.base.split_off_left(), self.map_op) + } + + fn to_reducer(&self) -> Self::Reducer { + self.base.to_reducer() + } +} + +struct FlatMapFolder<'f, C, F, R> { + base: C, + map_op: &'f F, + previous: Option<R>, +} + +impl<'f, T, U, C, F> Folder<T> for FlatMapFolder<'f, C, F, C::Result> +where + C: UnindexedConsumer<U::Item>, + F: Fn(T) -> U + Sync, + U: IntoParallelIterator, +{ + type Result = C::Result; + + fn consume(self, item: T) -> Self { + let map_op = self.map_op; + let par_iter = map_op(item).into_par_iter(); + let consumer = self.base.split_off_left(); + let result = par_iter.drive_unindexed(consumer); + + let previous = match self.previous { + None => Some(result), + Some(previous) => { + let reducer = self.base.to_reducer(); + Some(reducer.reduce(previous, result)) + } + }; + + FlatMapFolder { + base: self.base, + map_op, + previous, + } + } + + fn complete(self) -> Self::Result { + match self.previous { + Some(previous) => previous, + None => self.base.into_folder().complete(), + } + } + + fn full(&self) -> bool { + self.base.full() + } +} diff --git a/third_party/rust/rayon/src/iter/flat_map_iter.rs b/third_party/rust/rayon/src/iter/flat_map_iter.rs new file mode 100644 index 0000000000..c76cf68381 --- /dev/null +++ b/third_party/rust/rayon/src/iter/flat_map_iter.rs @@ -0,0 +1,147 @@ +use super::plumbing::*; +use super::*; + +use std::fmt::{self, Debug}; + +/// `FlatMapIter` maps each element to a serial iterator, then flattens these iterators together. +/// This struct is created by the [`flat_map_iter()`] method on [`ParallelIterator`] +/// +/// [`flat_map_iter()`]: trait.ParallelIterator.html#method.flat_map_iter +/// [`ParallelIterator`]: trait.ParallelIterator.html +#[must_use = "iterator adaptors are lazy and do nothing unless consumed"] +#[derive(Clone)] +pub struct FlatMapIter<I: ParallelIterator, F> { + base: I, + map_op: F, +} + +impl<I: ParallelIterator + Debug, F> Debug for FlatMapIter<I, F> { + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + f.debug_struct("FlatMapIter") + .field("base", &self.base) + .finish() + } +} + +impl<I: ParallelIterator, F> FlatMapIter<I, F> { + /// Creates a new `FlatMapIter` iterator. + pub(super) fn new(base: I, map_op: F) -> Self { + FlatMapIter { base, map_op } + } +} + +impl<I, F, SI> ParallelIterator for FlatMapIter<I, F> +where + I: ParallelIterator, + F: Fn(I::Item) -> SI + Sync + Send, + SI: IntoIterator, + SI::Item: Send, +{ + type Item = SI::Item; + + fn drive_unindexed<C>(self, consumer: C) -> C::Result + where + C: UnindexedConsumer<Self::Item>, + { + let consumer = FlatMapIterConsumer::new(consumer, &self.map_op); + self.base.drive_unindexed(consumer) + } +} + +/// //////////////////////////////////////////////////////////////////////// +/// Consumer implementation + +struct FlatMapIterConsumer<'f, C, F> { + base: C, + map_op: &'f F, +} + +impl<'f, C, F> FlatMapIterConsumer<'f, C, F> { + fn new(base: C, map_op: &'f F) -> Self { + FlatMapIterConsumer { base, map_op } + } +} + +impl<'f, T, U, C, F> Consumer<T> for FlatMapIterConsumer<'f, C, F> +where + C: UnindexedConsumer<U::Item>, + F: Fn(T) -> U + Sync, + U: IntoIterator, +{ + type Folder = FlatMapIterFolder<'f, C::Folder, F>; + type Reducer = C::Reducer; + type Result = C::Result; + + fn split_at(self, index: usize) -> (Self, Self, C::Reducer) { + let (left, right, reducer) = self.base.split_at(index); + ( + FlatMapIterConsumer::new(left, self.map_op), + FlatMapIterConsumer::new(right, self.map_op), + reducer, + ) + } + + fn into_folder(self) -> Self::Folder { + FlatMapIterFolder { + base: self.base.into_folder(), + map_op: self.map_op, + } + } + + fn full(&self) -> bool { + self.base.full() + } +} + +impl<'f, T, U, C, F> UnindexedConsumer<T> for FlatMapIterConsumer<'f, C, F> +where + C: UnindexedConsumer<U::Item>, + F: Fn(T) -> U + Sync, + U: IntoIterator, +{ + fn split_off_left(&self) -> Self { + FlatMapIterConsumer::new(self.base.split_off_left(), self.map_op) + } + + fn to_reducer(&self) -> Self::Reducer { + self.base.to_reducer() + } +} + +struct FlatMapIterFolder<'f, C, F> { + base: C, + map_op: &'f F, +} + +impl<'f, T, U, C, F> Folder<T> for FlatMapIterFolder<'f, C, F> +where + C: Folder<U::Item>, + F: Fn(T) -> U, + U: IntoIterator, +{ + type Result = C::Result; + + fn consume(self, item: T) -> Self { + let map_op = self.map_op; + let base = self.base.consume_iter(map_op(item)); + FlatMapIterFolder { base, map_op } + } + + fn consume_iter<I>(self, iter: I) -> Self + where + I: IntoIterator<Item = T>, + { + let map_op = self.map_op; + let iter = iter.into_iter().flat_map(map_op); + let base = self.base.consume_iter(iter); + FlatMapIterFolder { base, map_op } + } + + fn complete(self) -> Self::Result { + self.base.complete() + } + + fn full(&self) -> bool { + self.base.full() + } +} diff --git a/third_party/rust/rayon/src/iter/flatten.rs b/third_party/rust/rayon/src/iter/flatten.rs new file mode 100644 index 0000000000..29d88f99ed --- /dev/null +++ b/third_party/rust/rayon/src/iter/flatten.rs @@ -0,0 +1,140 @@ +use super::plumbing::*; +use super::*; + +/// `Flatten` turns each element to a parallel iterator, then flattens these iterators +/// together. This struct is created by the [`flatten()`] method on [`ParallelIterator`]. +/// +/// [`flatten()`]: trait.ParallelIterator.html#method.flatten +/// [`ParallelIterator`]: trait.ParallelIterator.html +#[must_use = "iterator adaptors are lazy and do nothing unless consumed"] +#[derive(Debug, Clone)] +pub struct Flatten<I: ParallelIterator> { + base: I, +} + +impl<I> Flatten<I> +where + I: ParallelIterator, + I::Item: IntoParallelIterator, +{ + /// Creates a new `Flatten` iterator. + pub(super) fn new(base: I) -> Self { + Flatten { base } + } +} + +impl<I> ParallelIterator for Flatten<I> +where + I: ParallelIterator, + I::Item: IntoParallelIterator, +{ + type Item = <I::Item as IntoParallelIterator>::Item; + + fn drive_unindexed<C>(self, consumer: C) -> C::Result + where + C: UnindexedConsumer<Self::Item>, + { + let consumer = FlattenConsumer::new(consumer); + self.base.drive_unindexed(consumer) + } +} + +/// //////////////////////////////////////////////////////////////////////// +/// Consumer implementation + +struct FlattenConsumer<C> { + base: C, +} + +impl<C> FlattenConsumer<C> { + fn new(base: C) -> Self { + FlattenConsumer { base } + } +} + +impl<T, C> Consumer<T> for FlattenConsumer<C> +where + C: UnindexedConsumer<T::Item>, + T: IntoParallelIterator, +{ + type Folder = FlattenFolder<C, C::Result>; + type Reducer = C::Reducer; + type Result = C::Result; + + fn split_at(self, index: usize) -> (Self, Self, C::Reducer) { + let (left, right, reducer) = self.base.split_at(index); + ( + FlattenConsumer::new(left), + FlattenConsumer::new(right), + reducer, + ) + } + + fn into_folder(self) -> Self::Folder { + FlattenFolder { + base: self.base, + previous: None, + } + } + + fn full(&self) -> bool { + self.base.full() + } +} + +impl<T, C> UnindexedConsumer<T> for FlattenConsumer<C> +where + C: UnindexedConsumer<T::Item>, + T: IntoParallelIterator, +{ + fn split_off_left(&self) -> Self { + FlattenConsumer::new(self.base.split_off_left()) + } + + fn to_reducer(&self) -> Self::Reducer { + self.base.to_reducer() + } +} + +struct FlattenFolder<C, R> { + base: C, + previous: Option<R>, +} + +impl<T, C> Folder<T> for FlattenFolder<C, C::Result> +where + C: UnindexedConsumer<T::Item>, + T: IntoParallelIterator, +{ + type Result = C::Result; + + fn consume(self, item: T) -> Self { + let par_iter = item.into_par_iter(); + let consumer = self.base.split_off_left(); + let result = par_iter.drive_unindexed(consumer); + + let previous = match self.previous { + None => Some(result), + Some(previous) => { + let reducer = self.base.to_reducer(); + Some(reducer.reduce(previous, result)) + } + }; + + FlattenFolder { + base: self.base, + previous, + } + } + + fn complete(self) -> Self::Result { + match self.previous { + Some(previous) => previous, + None => self.base.into_folder().complete(), + } + } + + fn full(&self) -> bool { + self.base.full() + } +} diff --git a/third_party/rust/rayon/src/iter/flatten_iter.rs b/third_party/rust/rayon/src/iter/flatten_iter.rs new file mode 100644 index 0000000000..3ce0a3ce6d --- /dev/null +++ b/third_party/rust/rayon/src/iter/flatten_iter.rs @@ -0,0 +1,132 @@ +use super::plumbing::*; +use super::*; + +/// `FlattenIter` turns each element to a serial iterator, then flattens these iterators +/// together. This struct is created by the [`flatten_iter()`] method on [`ParallelIterator`]. +/// +/// [`flatten_iter()`]: trait.ParallelIterator.html#method.flatten_iter +/// [`ParallelIterator`]: trait.ParallelIterator.html +#[must_use = "iterator adaptors are lazy and do nothing unless consumed"] +#[derive(Debug, Clone)] +pub struct FlattenIter<I: ParallelIterator> { + base: I, +} + +impl<I> FlattenIter<I> +where + I: ParallelIterator, + I::Item: IntoIterator, + <I::Item as IntoIterator>::Item: Send, +{ + /// Creates a new `FlattenIter` iterator. + pub(super) fn new(base: I) -> Self { + FlattenIter { base } + } +} + +impl<I> ParallelIterator for FlattenIter<I> +where + I: ParallelIterator, + I::Item: IntoIterator, + <I::Item as IntoIterator>::Item: Send, +{ + type Item = <I::Item as IntoIterator>::Item; + + fn drive_unindexed<C>(self, consumer: C) -> C::Result + where + C: UnindexedConsumer<Self::Item>, + { + let consumer = FlattenIterConsumer::new(consumer); + self.base.drive_unindexed(consumer) + } +} + +/// //////////////////////////////////////////////////////////////////////// +/// Consumer implementation + +struct FlattenIterConsumer<C> { + base: C, +} + +impl<C> FlattenIterConsumer<C> { + fn new(base: C) -> Self { + FlattenIterConsumer { base } + } +} + +impl<T, C> Consumer<T> for FlattenIterConsumer<C> +where + C: UnindexedConsumer<T::Item>, + T: IntoIterator, +{ + type Folder = FlattenIterFolder<C::Folder>; + type Reducer = C::Reducer; + type Result = C::Result; + + fn split_at(self, index: usize) -> (Self, Self, C::Reducer) { + let (left, right, reducer) = self.base.split_at(index); + ( + FlattenIterConsumer::new(left), + FlattenIterConsumer::new(right), + reducer, + ) + } + + fn into_folder(self) -> Self::Folder { + FlattenIterFolder { + base: self.base.into_folder(), + } + } + + fn full(&self) -> bool { + self.base.full() + } +} + +impl<T, C> UnindexedConsumer<T> for FlattenIterConsumer<C> +where + C: UnindexedConsumer<T::Item>, + T: IntoIterator, +{ + fn split_off_left(&self) -> Self { + FlattenIterConsumer::new(self.base.split_off_left()) + } + + fn to_reducer(&self) -> Self::Reducer { + self.base.to_reducer() + } +} + +struct FlattenIterFolder<C> { + base: C, +} + +impl<T, C> Folder<T> for FlattenIterFolder<C> +where + C: Folder<T::Item>, + T: IntoIterator, +{ + type Result = C::Result; + + fn consume(self, item: T) -> Self { + let base = self.base.consume_iter(item); + FlattenIterFolder { base } + } + + fn consume_iter<I>(self, iter: I) -> Self + where + I: IntoIterator<Item = T>, + { + let iter = iter.into_iter().flatten(); + let base = self.base.consume_iter(iter); + FlattenIterFolder { base } + } + + fn complete(self) -> Self::Result { + self.base.complete() + } + + fn full(&self) -> bool { + self.base.full() + } +} diff --git a/third_party/rust/rayon/src/iter/fold.rs b/third_party/rust/rayon/src/iter/fold.rs new file mode 100644 index 0000000000..345afbd501 --- /dev/null +++ b/third_party/rust/rayon/src/iter/fold.rs @@ -0,0 +1,302 @@ +use super::plumbing::*; +use super::*; + +use std::fmt::{self, Debug}; + +impl<U, I, ID, F> Fold<I, ID, F> +where + I: ParallelIterator, + F: Fn(U, I::Item) -> U + Sync + Send, + ID: Fn() -> U + Sync + Send, + U: Send, +{ + pub(super) fn new(base: I, identity: ID, fold_op: F) -> Self { + Fold { + base, + identity, + fold_op, + } + } +} + +/// `Fold` is an iterator that applies a function over an iterator producing a single value. +/// This struct is created by the [`fold()`] method on [`ParallelIterator`] +/// +/// [`fold()`]: trait.ParallelIterator.html#method.fold +/// [`ParallelIterator`]: trait.ParallelIterator.html +#[must_use = "iterator adaptors are lazy and do nothing unless consumed"] +#[derive(Clone)] +pub struct Fold<I, ID, F> { + base: I, + identity: ID, + fold_op: F, +} + +impl<I: ParallelIterator + Debug, ID, F> Debug for Fold<I, ID, F> { + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + f.debug_struct("Fold").field("base", &self.base).finish() + } +} + +impl<U, I, ID, F> ParallelIterator for Fold<I, ID, F> +where + I: ParallelIterator, + F: Fn(U, I::Item) -> U + Sync + Send, + ID: Fn() -> U + Sync + Send, + U: Send, +{ + type Item = U; + + fn drive_unindexed<C>(self, consumer: C) -> C::Result + where + C: UnindexedConsumer<Self::Item>, + { + let consumer1 = FoldConsumer { + base: consumer, + fold_op: &self.fold_op, + identity: &self.identity, + }; + self.base.drive_unindexed(consumer1) + } +} + +struct FoldConsumer<'c, C, ID, F> { + base: C, + fold_op: &'c F, + identity: &'c ID, +} + +impl<'r, U, T, C, ID, F> Consumer<T> for FoldConsumer<'r, C, ID, F> +where + C: Consumer<U>, + F: Fn(U, T) -> U + Sync, + ID: Fn() -> U + Sync, + U: Send, +{ + type Folder = FoldFolder<'r, C::Folder, U, F>; + type Reducer = C::Reducer; + type Result = C::Result; + + fn split_at(self, index: usize) -> (Self, Self, Self::Reducer) { + let (left, right, reducer) = self.base.split_at(index); + ( + FoldConsumer { base: left, ..self }, + FoldConsumer { + base: right, + ..self + }, + reducer, + ) + } + + fn into_folder(self) -> Self::Folder { + FoldFolder { + base: self.base.into_folder(), + item: (self.identity)(), + fold_op: self.fold_op, + } + } + + fn full(&self) -> bool { + self.base.full() + } +} + +impl<'r, U, T, C, ID, F> UnindexedConsumer<T> for FoldConsumer<'r, C, ID, F> +where + C: UnindexedConsumer<U>, + F: Fn(U, T) -> U + Sync, + ID: Fn() -> U + Sync, + U: Send, +{ + fn split_off_left(&self) -> Self { + FoldConsumer { + base: self.base.split_off_left(), + ..*self + } + } + + fn to_reducer(&self) -> Self::Reducer { + self.base.to_reducer() + } +} + +struct FoldFolder<'r, C, ID, F> { + base: C, + fold_op: &'r F, + item: ID, +} + +impl<'r, C, ID, F, T> Folder<T> for FoldFolder<'r, C, ID, F> +where + C: Folder<ID>, + F: Fn(ID, T) -> ID + Sync, +{ + type Result = C::Result; + + fn consume(self, item: T) -> Self { + let item = (self.fold_op)(self.item, item); + FoldFolder { + base: self.base, + fold_op: self.fold_op, + item, + } + } + + fn consume_iter<I>(self, iter: I) -> Self + where + I: IntoIterator<Item = T>, + { + fn not_full<C, ID, T>(base: &C) -> impl Fn(&T) -> bool + '_ + where + C: Folder<ID>, + { + move |_| !base.full() + } + + let base = self.base; + let item = iter + .into_iter() + // stop iterating if another thread has finished + .take_while(not_full(&base)) + .fold(self.item, self.fold_op); + + FoldFolder { + base, + item, + fold_op: self.fold_op, + } + } + + fn complete(self) -> C::Result { + self.base.consume(self.item).complete() + } + + fn full(&self) -> bool { + self.base.full() + } +} + +// /////////////////////////////////////////////////////////////////////////// + +impl<U, I, F> FoldWith<I, U, F> +where + I: ParallelIterator, + F: Fn(U, I::Item) -> U + Sync + Send, + U: Send + Clone, +{ + pub(super) fn new(base: I, item: U, fold_op: F) -> Self { + FoldWith { + base, + item, + fold_op, + } + } +} + +/// `FoldWith` is an iterator that applies a function over an iterator producing a single value. +/// This struct is created by the [`fold_with()`] method on [`ParallelIterator`] +/// +/// [`fold_with()`]: trait.ParallelIterator.html#method.fold_with +/// [`ParallelIterator`]: trait.ParallelIterator.html +#[must_use = "iterator adaptors are lazy and do nothing unless consumed"] +#[derive(Clone)] +pub struct FoldWith<I, U, F> { + base: I, + item: U, + fold_op: F, +} + +impl<I: ParallelIterator + Debug, U: Debug, F> Debug for FoldWith<I, U, F> { + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + f.debug_struct("FoldWith") + .field("base", &self.base) + .field("item", &self.item) + .finish() + } +} + +impl<U, I, F> ParallelIterator for FoldWith<I, U, F> +where + I: ParallelIterator, + F: Fn(U, I::Item) -> U + Sync + Send, + U: Send + Clone, +{ + type Item = U; + + fn drive_unindexed<C>(self, consumer: C) -> C::Result + where + C: UnindexedConsumer<Self::Item>, + { + let consumer1 = FoldWithConsumer { + base: consumer, + item: self.item, + fold_op: &self.fold_op, + }; + self.base.drive_unindexed(consumer1) + } +} + +struct FoldWithConsumer<'c, C, U, F> { + base: C, + item: U, + fold_op: &'c F, +} + +impl<'r, U, T, C, F> Consumer<T> for FoldWithConsumer<'r, C, U, F> +where + C: Consumer<U>, + F: Fn(U, T) -> U + Sync, + U: Send + Clone, +{ + type Folder = FoldFolder<'r, C::Folder, U, F>; + type Reducer = C::Reducer; + type Result = C::Result; + + fn split_at(self, index: usize) -> (Self, Self, Self::Reducer) { + let (left, right, reducer) = self.base.split_at(index); + ( + FoldWithConsumer { + base: left, + item: self.item.clone(), + ..self + }, + FoldWithConsumer { + base: right, + ..self + }, + reducer, + ) + } + + fn into_folder(self) -> Self::Folder { + FoldFolder { + base: self.base.into_folder(), + item: self.item, + fold_op: self.fold_op, + } + } + + fn full(&self) -> bool { + self.base.full() + } +} + +impl<'r, U, T, C, F> UnindexedConsumer<T> for FoldWithConsumer<'r, C, U, F> +where + C: UnindexedConsumer<U>, + F: Fn(U, T) -> U + Sync, + U: Send + Clone, +{ + fn split_off_left(&self) -> Self { + FoldWithConsumer { + base: self.base.split_off_left(), + item: self.item.clone(), + ..*self + } + } + + fn to_reducer(&self) -> Self::Reducer { + self.base.to_reducer() + } +} diff --git a/third_party/rust/rayon/src/iter/fold_chunks.rs b/third_party/rust/rayon/src/iter/fold_chunks.rs new file mode 100644 index 0000000000..185fb1a65d --- /dev/null +++ b/third_party/rust/rayon/src/iter/fold_chunks.rs @@ -0,0 +1,236 @@ +use std::fmt::{self, Debug}; + +use super::chunks::ChunkProducer; +use super::plumbing::*; +use super::*; +use crate::math::div_round_up; + +/// `FoldChunks` is an iterator that groups elements of an underlying iterator and applies a +/// function over them, producing a single value for each group. +/// +/// This struct is created by the [`fold_chunks()`] method on [`IndexedParallelIterator`] +/// +/// [`fold_chunks()`]: trait.IndexedParallelIterator.html#method.fold_chunks +/// [`IndexedParallelIterator`]: trait.IndexedParallelIterator.html +#[must_use = "iterator adaptors are lazy and do nothing unless consumed"] +#[derive(Clone)] +pub struct FoldChunks<I, ID, F> +where + I: IndexedParallelIterator, +{ + base: I, + chunk_size: usize, + fold_op: F, + identity: ID, +} + +impl<I: IndexedParallelIterator + Debug, ID, F> Debug for FoldChunks<I, ID, F> { + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + f.debug_struct("Fold") + .field("base", &self.base) + .field("chunk_size", &self.chunk_size) + .finish() + } +} + +impl<I, ID, U, F> FoldChunks<I, ID, F> +where + I: IndexedParallelIterator, + ID: Fn() -> U + Send + Sync, + F: Fn(U, I::Item) -> U + Send + Sync, + U: Send, +{ + /// Creates a new `FoldChunks` iterator + pub(super) fn new(base: I, chunk_size: usize, identity: ID, fold_op: F) -> Self { + FoldChunks { + base, + chunk_size, + identity, + fold_op, + } + } +} + +impl<I, ID, U, F> ParallelIterator for FoldChunks<I, ID, F> +where + I: IndexedParallelIterator, + ID: Fn() -> U + Send + Sync, + F: Fn(U, I::Item) -> U + Send + Sync, + U: Send, +{ + type Item = U; + + fn drive_unindexed<C>(self, consumer: C) -> C::Result + where + C: Consumer<U>, + { + bridge(self, consumer) + } + + fn opt_len(&self) -> Option<usize> { + Some(self.len()) + } +} + +impl<I, ID, U, F> IndexedParallelIterator for FoldChunks<I, ID, F> +where + I: IndexedParallelIterator, + ID: Fn() -> U + Send + Sync, + F: Fn(U, I::Item) -> U + Send + Sync, + U: Send, +{ + fn len(&self) -> usize { + div_round_up(self.base.len(), self.chunk_size) + } + + fn drive<C>(self, consumer: C) -> C::Result + where + C: Consumer<Self::Item>, + { + bridge(self, consumer) + } + + fn with_producer<CB>(self, callback: CB) -> CB::Output + where + CB: ProducerCallback<Self::Item>, + { + let len = self.base.len(); + return self.base.with_producer(Callback { + chunk_size: self.chunk_size, + len, + identity: self.identity, + fold_op: self.fold_op, + callback, + }); + + struct Callback<CB, ID, F> { + chunk_size: usize, + len: usize, + identity: ID, + fold_op: F, + callback: CB, + } + + impl<T, CB, ID, U, F> ProducerCallback<T> for Callback<CB, ID, F> + where + CB: ProducerCallback<U>, + ID: Fn() -> U + Send + Sync, + F: Fn(U, T) -> U + Send + Sync, + { + type Output = CB::Output; + + fn callback<P>(self, base: P) -> CB::Output + where + P: Producer<Item = T>, + { + let identity = &self.identity; + let fold_op = &self.fold_op; + let fold_iter = move |iter: P::IntoIter| iter.fold(identity(), fold_op); + let producer = ChunkProducer::new(self.chunk_size, self.len, base, fold_iter); + self.callback.callback(producer) + } + } + } +} + +#[cfg(test)] +mod test { + use super::*; + use std::ops::Add; + + #[test] + fn check_fold_chunks() { + let words = "bishbashbosh!" + .chars() + .collect::<Vec<_>>() + .into_par_iter() + .fold_chunks(4, String::new, |mut s, c| { + s.push(c); + s + }) + .collect::<Vec<_>>(); + + assert_eq!(words, vec!["bish", "bash", "bosh", "!"]); + } + + // 'closure' values for tests below + fn id() -> i32 { + 0 + } + fn sum<T, U>(x: T, y: U) -> T + where + T: Add<U, Output = T>, + { + x + y + } + + #[test] + #[should_panic(expected = "chunk_size must not be zero")] + fn check_fold_chunks_zero_size() { + let _: Vec<i32> = vec![1, 2, 3] + .into_par_iter() + .fold_chunks(0, id, sum) + .collect(); + } + + #[test] + fn check_fold_chunks_even_size() { + assert_eq!( + vec![1 + 2 + 3, 4 + 5 + 6, 7 + 8 + 9], + (1..10) + .into_par_iter() + .fold_chunks(3, id, sum) + .collect::<Vec<i32>>() + ); + } + + #[test] + fn check_fold_chunks_empty() { + let v: Vec<i32> = vec![]; + let expected: Vec<i32> = vec![]; + assert_eq!( + expected, + v.into_par_iter() + .fold_chunks(2, id, sum) + .collect::<Vec<i32>>() + ); + } + + #[test] + fn check_fold_chunks_len() { + assert_eq!(4, (0..8).into_par_iter().fold_chunks(2, id, sum).len()); + assert_eq!(3, (0..9).into_par_iter().fold_chunks(3, id, sum).len()); + assert_eq!(3, (0..8).into_par_iter().fold_chunks(3, id, sum).len()); + assert_eq!(1, (&[1]).par_iter().fold_chunks(3, id, sum).len()); + assert_eq!(0, (0..0).into_par_iter().fold_chunks(3, id, sum).len()); + } + + #[test] + fn check_fold_chunks_uneven() { + let cases: Vec<(Vec<u32>, usize, Vec<u32>)> = vec![ + ((0..5).collect(), 3, vec![0 + 1 + 2, 3 + 4]), + (vec![1], 5, vec![1]), + ((0..4).collect(), 3, vec![0 + 1 + 2, 3]), + ]; + + for (i, (v, n, expected)) in cases.into_iter().enumerate() { + let mut res: Vec<u32> = vec![]; + v.par_iter() + .fold_chunks(n, || 0, sum) + .collect_into_vec(&mut res); + assert_eq!(expected, res, "Case {} failed", i); + + res.truncate(0); + v.into_par_iter() + .fold_chunks(n, || 0, sum) + .rev() + .collect_into_vec(&mut res); + assert_eq!( + expected.into_iter().rev().collect::<Vec<u32>>(), + res, + "Case {} reversed failed", + i + ); + } + } +} diff --git a/third_party/rust/rayon/src/iter/fold_chunks_with.rs b/third_party/rust/rayon/src/iter/fold_chunks_with.rs new file mode 100644 index 0000000000..af120aec47 --- /dev/null +++ b/third_party/rust/rayon/src/iter/fold_chunks_with.rs @@ -0,0 +1,231 @@ +use std::fmt::{self, Debug}; + +use super::chunks::ChunkProducer; +use super::plumbing::*; +use super::*; +use crate::math::div_round_up; + +/// `FoldChunksWith` is an iterator that groups elements of an underlying iterator and applies a +/// function over them, producing a single value for each group. +/// +/// This struct is created by the [`fold_chunks_with()`] method on [`IndexedParallelIterator`] +/// +/// [`fold_chunks_with()`]: trait.IndexedParallelIterator.html#method.fold_chunks +/// [`IndexedParallelIterator`]: trait.IndexedParallelIterator.html +#[must_use = "iterator adaptors are lazy and do nothing unless consumed"] +#[derive(Clone)] +pub struct FoldChunksWith<I, U, F> +where + I: IndexedParallelIterator, +{ + base: I, + chunk_size: usize, + item: U, + fold_op: F, +} + +impl<I: IndexedParallelIterator + Debug, U: Debug, F> Debug for FoldChunksWith<I, U, F> { + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + f.debug_struct("Fold") + .field("base", &self.base) + .field("chunk_size", &self.chunk_size) + .field("item", &self.item) + .finish() + } +} + +impl<I, U, F> FoldChunksWith<I, U, F> +where + I: IndexedParallelIterator, + U: Send + Clone, + F: Fn(U, I::Item) -> U + Send + Sync, +{ + /// Creates a new `FoldChunksWith` iterator + pub(super) fn new(base: I, chunk_size: usize, item: U, fold_op: F) -> Self { + FoldChunksWith { + base, + chunk_size, + item, + fold_op, + } + } +} + +impl<I, U, F> ParallelIterator for FoldChunksWith<I, U, F> +where + I: IndexedParallelIterator, + U: Send + Clone, + F: Fn(U, I::Item) -> U + Send + Sync, +{ + type Item = U; + + fn drive_unindexed<C>(self, consumer: C) -> C::Result + where + C: Consumer<U>, + { + bridge(self, consumer) + } + + fn opt_len(&self) -> Option<usize> { + Some(self.len()) + } +} + +impl<I, U, F> IndexedParallelIterator for FoldChunksWith<I, U, F> +where + I: IndexedParallelIterator, + U: Send + Clone, + F: Fn(U, I::Item) -> U + Send + Sync, +{ + fn len(&self) -> usize { + div_round_up(self.base.len(), self.chunk_size) + } + + fn drive<C>(self, consumer: C) -> C::Result + where + C: Consumer<Self::Item>, + { + bridge(self, consumer) + } + + fn with_producer<CB>(self, callback: CB) -> CB::Output + where + CB: ProducerCallback<Self::Item>, + { + let len = self.base.len(); + return self.base.with_producer(Callback { + chunk_size: self.chunk_size, + len, + item: self.item, + fold_op: self.fold_op, + callback, + }); + + struct Callback<CB, T, F> { + chunk_size: usize, + len: usize, + item: T, + fold_op: F, + callback: CB, + } + + impl<T, U, F, CB> ProducerCallback<T> for Callback<CB, U, F> + where + CB: ProducerCallback<U>, + U: Send + Clone, + F: Fn(U, T) -> U + Send + Sync, + { + type Output = CB::Output; + + fn callback<P>(self, base: P) -> CB::Output + where + P: Producer<Item = T>, + { + let item = self.item; + let fold_op = &self.fold_op; + let fold_iter = move |iter: P::IntoIter| iter.fold(item.clone(), fold_op); + let producer = ChunkProducer::new(self.chunk_size, self.len, base, fold_iter); + self.callback.callback(producer) + } + } + } +} + +#[cfg(test)] +mod test { + use super::*; + use std::ops::Add; + + #[test] + fn check_fold_chunks_with() { + let words = "bishbashbosh!" + .chars() + .collect::<Vec<_>>() + .into_par_iter() + .fold_chunks_with(4, String::new(), |mut s, c| { + s.push(c); + s + }) + .collect::<Vec<_>>(); + + assert_eq!(words, vec!["bish", "bash", "bosh", "!"]); + } + + // 'closure' value for tests below + fn sum<T, U>(x: T, y: U) -> T + where + T: Add<U, Output = T>, + { + x + y + } + + #[test] + #[should_panic(expected = "chunk_size must not be zero")] + fn check_fold_chunks_zero_size() { + let _: Vec<i32> = vec![1, 2, 3] + .into_par_iter() + .fold_chunks_with(0, 0, sum) + .collect(); + } + + #[test] + fn check_fold_chunks_even_size() { + assert_eq!( + vec![1 + 2 + 3, 4 + 5 + 6, 7 + 8 + 9], + (1..10) + .into_par_iter() + .fold_chunks_with(3, 0, sum) + .collect::<Vec<i32>>() + ); + } + + #[test] + fn check_fold_chunks_with_empty() { + let v: Vec<i32> = vec![]; + let expected: Vec<i32> = vec![]; + assert_eq!( + expected, + v.into_par_iter() + .fold_chunks_with(2, 0, sum) + .collect::<Vec<i32>>() + ); + } + + #[test] + fn check_fold_chunks_len() { + assert_eq!(4, (0..8).into_par_iter().fold_chunks_with(2, 0, sum).len()); + assert_eq!(3, (0..9).into_par_iter().fold_chunks_with(3, 0, sum).len()); + assert_eq!(3, (0..8).into_par_iter().fold_chunks_with(3, 0, sum).len()); + assert_eq!(1, (&[1]).par_iter().fold_chunks_with(3, 0, sum).len()); + assert_eq!(0, (0..0).into_par_iter().fold_chunks_with(3, 0, sum).len()); + } + + #[test] + fn check_fold_chunks_uneven() { + let cases: Vec<(Vec<u32>, usize, Vec<u32>)> = vec![ + ((0..5).collect(), 3, vec![0 + 1 + 2, 3 + 4]), + (vec![1], 5, vec![1]), + ((0..4).collect(), 3, vec![0 + 1 + 2, 3]), + ]; + + for (i, (v, n, expected)) in cases.into_iter().enumerate() { + let mut res: Vec<u32> = vec![]; + v.par_iter() + .fold_chunks_with(n, 0, sum) + .collect_into_vec(&mut res); + assert_eq!(expected, res, "Case {} failed", i); + + res.truncate(0); + v.into_par_iter() + .fold_chunks_with(n, 0, sum) + .rev() + .collect_into_vec(&mut res); + assert_eq!( + expected.into_iter().rev().collect::<Vec<u32>>(), + res, + "Case {} reversed failed", + i + ); + } + } +} diff --git a/third_party/rust/rayon/src/iter/for_each.rs b/third_party/rust/rayon/src/iter/for_each.rs new file mode 100644 index 0000000000..3b77bebe1b --- /dev/null +++ b/third_party/rust/rayon/src/iter/for_each.rs @@ -0,0 +1,77 @@ +use super::noop::*; +use super::plumbing::*; +use super::ParallelIterator; + +pub(super) fn for_each<I, F, T>(pi: I, op: &F) +where + I: ParallelIterator<Item = T>, + F: Fn(T) + Sync, + T: Send, +{ + let consumer = ForEachConsumer { op }; + pi.drive_unindexed(consumer) +} + +struct ForEachConsumer<'f, F> { + op: &'f F, +} + +impl<'f, F, T> Consumer<T> for ForEachConsumer<'f, F> +where + F: Fn(T) + Sync, +{ + type Folder = ForEachConsumer<'f, F>; + type Reducer = NoopReducer; + type Result = (); + + fn split_at(self, _index: usize) -> (Self, Self, NoopReducer) { + (self.split_off_left(), self, NoopReducer) + } + + fn into_folder(self) -> Self { + self + } + + fn full(&self) -> bool { + false + } +} + +impl<'f, F, T> Folder<T> for ForEachConsumer<'f, F> +where + F: Fn(T) + Sync, +{ + type Result = (); + + fn consume(self, item: T) -> Self { + (self.op)(item); + self + } + + fn consume_iter<I>(self, iter: I) -> Self + where + I: IntoIterator<Item = T>, + { + iter.into_iter().for_each(self.op); + self + } + + fn complete(self) {} + + fn full(&self) -> bool { + false + } +} + +impl<'f, F, T> UnindexedConsumer<T> for ForEachConsumer<'f, F> +where + F: Fn(T) + Sync, +{ + fn split_off_left(&self) -> Self { + ForEachConsumer { op: self.op } + } + + fn to_reducer(&self) -> NoopReducer { + NoopReducer + } +} diff --git a/third_party/rust/rayon/src/iter/from_par_iter.rs b/third_party/rust/rayon/src/iter/from_par_iter.rs new file mode 100644 index 0000000000..3240f32e21 --- /dev/null +++ b/third_party/rust/rayon/src/iter/from_par_iter.rs @@ -0,0 +1,228 @@ +use super::noop::NoopConsumer; +use super::{FromParallelIterator, IntoParallelIterator, ParallelExtend, ParallelIterator}; + +use std::borrow::Cow; +use std::collections::LinkedList; +use std::collections::{BTreeMap, BTreeSet, HashMap, HashSet}; +use std::collections::{BinaryHeap, VecDeque}; +use std::hash::{BuildHasher, Hash}; + +/// Creates an empty default collection and extends it. +fn collect_extended<C, I>(par_iter: I) -> C +where + I: IntoParallelIterator, + C: ParallelExtend<I::Item> + Default, +{ + let mut collection = C::default(); + collection.par_extend(par_iter); + collection +} + +/// Collects items from a parallel iterator into a vector. +impl<T> FromParallelIterator<T> for Vec<T> +where + T: Send, +{ + fn from_par_iter<I>(par_iter: I) -> Self + where + I: IntoParallelIterator<Item = T>, + { + collect_extended(par_iter) + } +} + +/// Collects items from a parallel iterator into a vecdeque. +impl<T> FromParallelIterator<T> for VecDeque<T> +where + T: Send, +{ + fn from_par_iter<I>(par_iter: I) -> Self + where + I: IntoParallelIterator<Item = T>, + { + Vec::from_par_iter(par_iter).into() + } +} + +/// Collects items from a parallel iterator into a binaryheap. +/// The heap-ordering is calculated serially after all items are collected. +impl<T> FromParallelIterator<T> for BinaryHeap<T> +where + T: Ord + Send, +{ + fn from_par_iter<I>(par_iter: I) -> Self + where + I: IntoParallelIterator<Item = T>, + { + Vec::from_par_iter(par_iter).into() + } +} + +/// Collects items from a parallel iterator into a freshly allocated +/// linked list. +impl<T> FromParallelIterator<T> for LinkedList<T> +where + T: Send, +{ + fn from_par_iter<I>(par_iter: I) -> Self + where + I: IntoParallelIterator<Item = T>, + { + collect_extended(par_iter) + } +} + +/// Collects (key, value) pairs from a parallel iterator into a +/// hashmap. If multiple pairs correspond to the same key, then the +/// ones produced earlier in the parallel iterator will be +/// overwritten, just as with a sequential iterator. +impl<K, V, S> FromParallelIterator<(K, V)> for HashMap<K, V, S> +where + K: Eq + Hash + Send, + V: Send, + S: BuildHasher + Default + Send, +{ + fn from_par_iter<I>(par_iter: I) -> Self + where + I: IntoParallelIterator<Item = (K, V)>, + { + collect_extended(par_iter) + } +} + +/// Collects (key, value) pairs from a parallel iterator into a +/// btreemap. If multiple pairs correspond to the same key, then the +/// ones produced earlier in the parallel iterator will be +/// overwritten, just as with a sequential iterator. +impl<K, V> FromParallelIterator<(K, V)> for BTreeMap<K, V> +where + K: Ord + Send, + V: Send, +{ + fn from_par_iter<I>(par_iter: I) -> Self + where + I: IntoParallelIterator<Item = (K, V)>, + { + collect_extended(par_iter) + } +} + +/// Collects values from a parallel iterator into a hashset. +impl<V, S> FromParallelIterator<V> for HashSet<V, S> +where + V: Eq + Hash + Send, + S: BuildHasher + Default + Send, +{ + fn from_par_iter<I>(par_iter: I) -> Self + where + I: IntoParallelIterator<Item = V>, + { + collect_extended(par_iter) + } +} + +/// Collects values from a parallel iterator into a btreeset. +impl<V> FromParallelIterator<V> for BTreeSet<V> +where + V: Send + Ord, +{ + fn from_par_iter<I>(par_iter: I) -> Self + where + I: IntoParallelIterator<Item = V>, + { + collect_extended(par_iter) + } +} + +/// Collects characters from a parallel iterator into a string. +impl FromParallelIterator<char> for String { + fn from_par_iter<I>(par_iter: I) -> Self + where + I: IntoParallelIterator<Item = char>, + { + collect_extended(par_iter) + } +} + +/// Collects characters from a parallel iterator into a string. +impl<'a> FromParallelIterator<&'a char> for String { + fn from_par_iter<I>(par_iter: I) -> Self + where + I: IntoParallelIterator<Item = &'a char>, + { + collect_extended(par_iter) + } +} + +/// Collects string slices from a parallel iterator into a string. +impl<'a> FromParallelIterator<&'a str> for String { + fn from_par_iter<I>(par_iter: I) -> Self + where + I: IntoParallelIterator<Item = &'a str>, + { + collect_extended(par_iter) + } +} + +/// Collects strings from a parallel iterator into one large string. +impl FromParallelIterator<String> for String { + fn from_par_iter<I>(par_iter: I) -> Self + where + I: IntoParallelIterator<Item = String>, + { + collect_extended(par_iter) + } +} + +/// Collects string slices from a parallel iterator into a string. +impl<'a> FromParallelIterator<Cow<'a, str>> for String { + fn from_par_iter<I>(par_iter: I) -> Self + where + I: IntoParallelIterator<Item = Cow<'a, str>>, + { + collect_extended(par_iter) + } +} + +/// Collects an arbitrary `Cow` collection. +/// +/// Note, the standard library only has `FromIterator` for `Cow<'a, str>` and +/// `Cow<'a, [T]>`, because no one thought to add a blanket implementation +/// before it was stabilized. +impl<'a, C: ?Sized, T> FromParallelIterator<T> for Cow<'a, C> +where + C: ToOwned, + C::Owned: FromParallelIterator<T>, + T: Send, +{ + fn from_par_iter<I>(par_iter: I) -> Self + where + I: IntoParallelIterator<Item = T>, + { + Cow::Owned(C::Owned::from_par_iter(par_iter)) + } +} + +/// Collapses all unit items from a parallel iterator into one. +/// +/// This is more useful when combined with higher-level abstractions, like +/// collecting to a `Result<(), E>` where you only care about errors: +/// +/// ``` +/// use std::io::*; +/// use rayon::prelude::*; +/// +/// let data = vec![1, 2, 3, 4, 5]; +/// let res: Result<()> = data.par_iter() +/// .map(|x| writeln!(stdout(), "{}", x)) +/// .collect(); +/// assert!(res.is_ok()); +/// ``` +impl FromParallelIterator<()> for () { + fn from_par_iter<I>(par_iter: I) -> Self + where + I: IntoParallelIterator<Item = ()>, + { + par_iter.into_par_iter().drive_unindexed(NoopConsumer) + } +} diff --git a/third_party/rust/rayon/src/iter/inspect.rs b/third_party/rust/rayon/src/iter/inspect.rs new file mode 100644 index 0000000000..c50ca022da --- /dev/null +++ b/third_party/rust/rayon/src/iter/inspect.rs @@ -0,0 +1,257 @@ +use super::plumbing::*; +use super::*; + +use std::fmt::{self, Debug}; +use std::iter; + +/// `Inspect` is an iterator that calls a function with a reference to each +/// element before yielding it. +/// +/// This struct is created by the [`inspect()`] method on [`ParallelIterator`] +/// +/// [`inspect()`]: trait.ParallelIterator.html#method.inspect +/// [`ParallelIterator`]: trait.ParallelIterator.html +#[must_use = "iterator adaptors are lazy and do nothing unless consumed"] +#[derive(Clone)] +pub struct Inspect<I: ParallelIterator, F> { + base: I, + inspect_op: F, +} + +impl<I: ParallelIterator + Debug, F> Debug for Inspect<I, F> { + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + f.debug_struct("Inspect").field("base", &self.base).finish() + } +} + +impl<I, F> Inspect<I, F> +where + I: ParallelIterator, +{ + /// Creates a new `Inspect` iterator. + pub(super) fn new(base: I, inspect_op: F) -> Self { + Inspect { base, inspect_op } + } +} + +impl<I, F> ParallelIterator for Inspect<I, F> +where + I: ParallelIterator, + F: Fn(&I::Item) + Sync + Send, +{ + type Item = I::Item; + + fn drive_unindexed<C>(self, consumer: C) -> C::Result + where + C: UnindexedConsumer<Self::Item>, + { + let consumer1 = InspectConsumer::new(consumer, &self.inspect_op); + self.base.drive_unindexed(consumer1) + } + + fn opt_len(&self) -> Option<usize> { + self.base.opt_len() + } +} + +impl<I, F> IndexedParallelIterator for Inspect<I, F> +where + I: IndexedParallelIterator, + F: Fn(&I::Item) + Sync + Send, +{ + fn drive<C>(self, consumer: C) -> C::Result + where + C: Consumer<Self::Item>, + { + let consumer1 = InspectConsumer::new(consumer, &self.inspect_op); + self.base.drive(consumer1) + } + + fn len(&self) -> usize { + self.base.len() + } + + fn with_producer<CB>(self, callback: CB) -> CB::Output + where + CB: ProducerCallback<Self::Item>, + { + return self.base.with_producer(Callback { + callback, + inspect_op: self.inspect_op, + }); + + struct Callback<CB, F> { + callback: CB, + inspect_op: F, + } + + impl<T, F, CB> ProducerCallback<T> for Callback<CB, F> + where + CB: ProducerCallback<T>, + F: Fn(&T) + Sync, + { + type Output = CB::Output; + + fn callback<P>(self, base: P) -> CB::Output + where + P: Producer<Item = T>, + { + let producer = InspectProducer { + base, + inspect_op: &self.inspect_op, + }; + self.callback.callback(producer) + } + } + } +} + +/// //////////////////////////////////////////////////////////////////////// + +struct InspectProducer<'f, P, F> { + base: P, + inspect_op: &'f F, +} + +impl<'f, P, F> Producer for InspectProducer<'f, P, F> +where + P: Producer, + F: Fn(&P::Item) + Sync, +{ + type Item = P::Item; + type IntoIter = iter::Inspect<P::IntoIter, &'f F>; + + fn into_iter(self) -> Self::IntoIter { + self.base.into_iter().inspect(self.inspect_op) + } + + fn min_len(&self) -> usize { + self.base.min_len() + } + + fn max_len(&self) -> usize { + self.base.max_len() + } + + fn split_at(self, index: usize) -> (Self, Self) { + let (left, right) = self.base.split_at(index); + ( + InspectProducer { + base: left, + inspect_op: self.inspect_op, + }, + InspectProducer { + base: right, + inspect_op: self.inspect_op, + }, + ) + } + + fn fold_with<G>(self, folder: G) -> G + where + G: Folder<Self::Item>, + { + let folder1 = InspectFolder { + base: folder, + inspect_op: self.inspect_op, + }; + self.base.fold_with(folder1).base + } +} + +/// //////////////////////////////////////////////////////////////////////// +/// Consumer implementation + +struct InspectConsumer<'f, C, F> { + base: C, + inspect_op: &'f F, +} + +impl<'f, C, F> InspectConsumer<'f, C, F> { + fn new(base: C, inspect_op: &'f F) -> Self { + InspectConsumer { base, inspect_op } + } +} + +impl<'f, T, C, F> Consumer<T> for InspectConsumer<'f, C, F> +where + C: Consumer<T>, + F: Fn(&T) + Sync, +{ + type Folder = InspectFolder<'f, C::Folder, F>; + type Reducer = C::Reducer; + type Result = C::Result; + + fn split_at(self, index: usize) -> (Self, Self, Self::Reducer) { + let (left, right, reducer) = self.base.split_at(index); + ( + InspectConsumer::new(left, self.inspect_op), + InspectConsumer::new(right, self.inspect_op), + reducer, + ) + } + + fn into_folder(self) -> Self::Folder { + InspectFolder { + base: self.base.into_folder(), + inspect_op: self.inspect_op, + } + } + + fn full(&self) -> bool { + self.base.full() + } +} + +impl<'f, T, C, F> UnindexedConsumer<T> for InspectConsumer<'f, C, F> +where + C: UnindexedConsumer<T>, + F: Fn(&T) + Sync, +{ + fn split_off_left(&self) -> Self { + InspectConsumer::new(self.base.split_off_left(), self.inspect_op) + } + + fn to_reducer(&self) -> Self::Reducer { + self.base.to_reducer() + } +} + +struct InspectFolder<'f, C, F> { + base: C, + inspect_op: &'f F, +} + +impl<'f, T, C, F> Folder<T> for InspectFolder<'f, C, F> +where + C: Folder<T>, + F: Fn(&T), +{ + type Result = C::Result; + + fn consume(self, item: T) -> Self { + (self.inspect_op)(&item); + InspectFolder { + base: self.base.consume(item), + inspect_op: self.inspect_op, + } + } + + fn consume_iter<I>(mut self, iter: I) -> Self + where + I: IntoIterator<Item = T>, + { + self.base = self + .base + .consume_iter(iter.into_iter().inspect(self.inspect_op)); + self + } + + fn complete(self) -> C::Result { + self.base.complete() + } + + fn full(&self) -> bool { + self.base.full() + } +} diff --git a/third_party/rust/rayon/src/iter/interleave.rs b/third_party/rust/rayon/src/iter/interleave.rs new file mode 100644 index 0000000000..3cacc49f95 --- /dev/null +++ b/third_party/rust/rayon/src/iter/interleave.rs @@ -0,0 +1,336 @@ +use super::plumbing::*; +use super::*; +use std::cmp; +use std::iter::Fuse; + +/// `Interleave` is an iterator that interleaves elements of iterators +/// `i` and `j` in one continuous iterator. This struct is created by +/// the [`interleave()`] method on [`IndexedParallelIterator`] +/// +/// [`interleave()`]: trait.IndexedParallelIterator.html#method.interleave +/// [`IndexedParallelIterator`]: trait.IndexedParallelIterator.html +#[must_use = "iterator adaptors are lazy and do nothing unless consumed"] +#[derive(Debug, Clone)] +pub struct Interleave<I, J> +where + I: IndexedParallelIterator, + J: IndexedParallelIterator<Item = I::Item>, +{ + i: I, + j: J, +} + +impl<I, J> Interleave<I, J> +where + I: IndexedParallelIterator, + J: IndexedParallelIterator<Item = I::Item>, +{ + /// Creates a new `Interleave` iterator + pub(super) fn new(i: I, j: J) -> Self { + Interleave { i, j } + } +} + +impl<I, J> ParallelIterator for Interleave<I, J> +where + I: IndexedParallelIterator, + J: IndexedParallelIterator<Item = I::Item>, +{ + type Item = I::Item; + + fn drive_unindexed<C>(self, consumer: C) -> C::Result + where + C: Consumer<I::Item>, + { + bridge(self, consumer) + } + + fn opt_len(&self) -> Option<usize> { + Some(self.len()) + } +} + +impl<I, J> IndexedParallelIterator for Interleave<I, J> +where + I: IndexedParallelIterator, + J: IndexedParallelIterator<Item = I::Item>, +{ + fn drive<C>(self, consumer: C) -> C::Result + where + C: Consumer<Self::Item>, + { + bridge(self, consumer) + } + + fn len(&self) -> usize { + self.i.len().checked_add(self.j.len()).expect("overflow") + } + + fn with_producer<CB>(self, callback: CB) -> CB::Output + where + CB: ProducerCallback<Self::Item>, + { + let (i_len, j_len) = (self.i.len(), self.j.len()); + return self.i.with_producer(CallbackI { + callback, + i_len, + j_len, + i_next: false, + j: self.j, + }); + + struct CallbackI<CB, J> { + callback: CB, + i_len: usize, + j_len: usize, + i_next: bool, + j: J, + } + + impl<CB, J> ProducerCallback<J::Item> for CallbackI<CB, J> + where + J: IndexedParallelIterator, + CB: ProducerCallback<J::Item>, + { + type Output = CB::Output; + + fn callback<I>(self, i_producer: I) -> Self::Output + where + I: Producer<Item = J::Item>, + { + self.j.with_producer(CallbackJ { + i_producer, + i_len: self.i_len, + j_len: self.j_len, + i_next: self.i_next, + callback: self.callback, + }) + } + } + + struct CallbackJ<CB, I> { + callback: CB, + i_len: usize, + j_len: usize, + i_next: bool, + i_producer: I, + } + + impl<CB, I> ProducerCallback<I::Item> for CallbackJ<CB, I> + where + I: Producer, + CB: ProducerCallback<I::Item>, + { + type Output = CB::Output; + + fn callback<J>(self, j_producer: J) -> Self::Output + where + J: Producer<Item = I::Item>, + { + let producer = InterleaveProducer::new( + self.i_producer, + j_producer, + self.i_len, + self.j_len, + self.i_next, + ); + self.callback.callback(producer) + } + } + } +} + +struct InterleaveProducer<I, J> +where + I: Producer, + J: Producer<Item = I::Item>, +{ + i: I, + j: J, + i_len: usize, + j_len: usize, + i_next: bool, +} + +impl<I, J> InterleaveProducer<I, J> +where + I: Producer, + J: Producer<Item = I::Item>, +{ + fn new(i: I, j: J, i_len: usize, j_len: usize, i_next: bool) -> InterleaveProducer<I, J> { + InterleaveProducer { + i, + j, + i_len, + j_len, + i_next, + } + } +} + +impl<I, J> Producer for InterleaveProducer<I, J> +where + I: Producer, + J: Producer<Item = I::Item>, +{ + type Item = I::Item; + type IntoIter = InterleaveSeq<I::IntoIter, J::IntoIter>; + + fn into_iter(self) -> Self::IntoIter { + InterleaveSeq { + i: self.i.into_iter().fuse(), + j: self.j.into_iter().fuse(), + i_next: self.i_next, + } + } + + fn min_len(&self) -> usize { + cmp::max(self.i.min_len(), self.j.min_len()) + } + + fn max_len(&self) -> usize { + cmp::min(self.i.max_len(), self.j.max_len()) + } + + /// We know 0 < index <= self.i_len + self.j_len + /// + /// Find a, b satisfying: + /// + /// (1) 0 < a <= self.i_len + /// (2) 0 < b <= self.j_len + /// (3) a + b == index + /// + /// For even splits, set a = b = index/2. + /// For odd splits, set a = (index/2)+1, b = index/2, if `i` + /// should yield the next element, otherwise, if `j` should yield + /// the next element, set a = index/2 and b = (index/2)+1 + fn split_at(self, index: usize) -> (Self, Self) { + #[inline] + fn odd_offset(flag: bool) -> usize { + (!flag) as usize + } + + let even = index % 2 == 0; + let idx = index >> 1; + + // desired split + let (i_idx, j_idx) = ( + idx + odd_offset(even || self.i_next), + idx + odd_offset(even || !self.i_next), + ); + + let (i_split, j_split) = if self.i_len >= i_idx && self.j_len >= j_idx { + (i_idx, j_idx) + } else if self.i_len >= i_idx { + // j too short + (index - self.j_len, self.j_len) + } else { + // i too short + (self.i_len, index - self.i_len) + }; + + let trailing_i_next = even == self.i_next; + let (i_left, i_right) = self.i.split_at(i_split); + let (j_left, j_right) = self.j.split_at(j_split); + + ( + InterleaveProducer::new(i_left, j_left, i_split, j_split, self.i_next), + InterleaveProducer::new( + i_right, + j_right, + self.i_len - i_split, + self.j_len - j_split, + trailing_i_next, + ), + ) + } +} + +/// Wrapper for Interleave to implement DoubleEndedIterator and +/// ExactSizeIterator. +/// +/// This iterator is fused. +struct InterleaveSeq<I, J> { + i: Fuse<I>, + j: Fuse<J>, + + /// Flag to control which iterator should provide the next element. When + /// `false` then `i` produces the next element, otherwise `j` produces the + /// next element. + i_next: bool, +} + +/// Iterator implementation for InterleaveSeq. This implementation is +/// taken more or less verbatim from itertools. It is replicated here +/// (instead of calling itertools directly), because we also need to +/// implement `DoubledEndedIterator` and `ExactSizeIterator`. +impl<I, J> Iterator for InterleaveSeq<I, J> +where + I: Iterator, + J: Iterator<Item = I::Item>, +{ + type Item = I::Item; + + #[inline] + fn next(&mut self) -> Option<Self::Item> { + self.i_next = !self.i_next; + if self.i_next { + match self.i.next() { + None => self.j.next(), + r => r, + } + } else { + match self.j.next() { + None => self.i.next(), + r => r, + } + } + } + + fn size_hint(&self) -> (usize, Option<usize>) { + let (ih, jh) = (self.i.size_hint(), self.j.size_hint()); + let min = ih.0.saturating_add(jh.0); + let max = match (ih.1, jh.1) { + (Some(x), Some(y)) => x.checked_add(y), + _ => None, + }; + (min, max) + } +} + +// The implementation for DoubleEndedIterator requires +// ExactSizeIterator to provide `next_back()`. The last element will +// come from the iterator that runs out last (ie has the most elements +// in it). If the iterators have the same number of elements, then the +// last iterator will provide the last element. +impl<I, J> DoubleEndedIterator for InterleaveSeq<I, J> +where + I: DoubleEndedIterator + ExactSizeIterator, + J: DoubleEndedIterator<Item = I::Item> + ExactSizeIterator<Item = I::Item>, +{ + #[inline] + fn next_back(&mut self) -> Option<I::Item> { + match self.i.len().cmp(&self.j.len()) { + Ordering::Less => self.j.next_back(), + Ordering::Equal => { + if self.i_next { + self.i.next_back() + } else { + self.j.next_back() + } + } + Ordering::Greater => self.i.next_back(), + } + } +} + +impl<I, J> ExactSizeIterator for InterleaveSeq<I, J> +where + I: ExactSizeIterator, + J: ExactSizeIterator<Item = I::Item>, +{ + #[inline] + fn len(&self) -> usize { + self.i.len() + self.j.len() + } +} diff --git a/third_party/rust/rayon/src/iter/interleave_shortest.rs b/third_party/rust/rayon/src/iter/interleave_shortest.rs new file mode 100644 index 0000000000..7d81369807 --- /dev/null +++ b/third_party/rust/rayon/src/iter/interleave_shortest.rs @@ -0,0 +1,85 @@ +use super::plumbing::*; +use super::*; + +/// `InterleaveShortest` is an iterator that works similarly to +/// `Interleave`, but this version stops returning elements once one +/// of the iterators run out. +/// +/// This struct is created by the [`interleave_shortest()`] method on +/// [`IndexedParallelIterator`]. +/// +/// [`interleave_shortest()`]: trait.IndexedParallelIterator.html#method.interleave_shortest +/// [`IndexedParallelIterator`]: trait.IndexedParallelIterator.html +#[must_use = "iterator adaptors are lazy and do nothing unless consumed"] +#[derive(Debug, Clone)] +pub struct InterleaveShortest<I, J> +where + I: IndexedParallelIterator, + J: IndexedParallelIterator<Item = I::Item>, +{ + interleave: Interleave<Take<I>, Take<J>>, +} + +impl<I, J> InterleaveShortest<I, J> +where + I: IndexedParallelIterator, + J: IndexedParallelIterator<Item = I::Item>, +{ + /// Creates a new `InterleaveShortest` iterator + pub(super) fn new(i: I, j: J) -> Self { + InterleaveShortest { + interleave: if i.len() <= j.len() { + // take equal lengths from both iterators + let n = i.len(); + i.take(n).interleave(j.take(n)) + } else { + // take one extra item from the first iterator + let n = j.len(); + i.take(n + 1).interleave(j.take(n)) + }, + } + } +} + +impl<I, J> ParallelIterator for InterleaveShortest<I, J> +where + I: IndexedParallelIterator, + J: IndexedParallelIterator<Item = I::Item>, +{ + type Item = I::Item; + + fn drive_unindexed<C>(self, consumer: C) -> C::Result + where + C: Consumer<I::Item>, + { + bridge(self, consumer) + } + + fn opt_len(&self) -> Option<usize> { + Some(self.len()) + } +} + +impl<I, J> IndexedParallelIterator for InterleaveShortest<I, J> +where + I: IndexedParallelIterator, + J: IndexedParallelIterator<Item = I::Item>, +{ + fn drive<C>(self, consumer: C) -> C::Result + where + C: Consumer<Self::Item>, + { + bridge(self, consumer) + } + + fn len(&self) -> usize { + self.interleave.len() + } + + fn with_producer<CB>(self, callback: CB) -> CB::Output + where + CB: ProducerCallback<Self::Item>, + { + self.interleave.with_producer(callback) + } +} diff --git a/third_party/rust/rayon/src/iter/intersperse.rs b/third_party/rust/rayon/src/iter/intersperse.rs new file mode 100644 index 0000000000..798bdc1d67 --- /dev/null +++ b/third_party/rust/rayon/src/iter/intersperse.rs @@ -0,0 +1,410 @@ +use super::plumbing::*; +use super::*; +use std::cell::Cell; +use std::iter::{self, Fuse}; + +/// `Intersperse` is an iterator that inserts a particular item between each +/// item of the adapted iterator. This struct is created by the +/// [`intersperse()`] method on [`ParallelIterator`] +/// +/// [`intersperse()`]: trait.ParallelIterator.html#method.intersperse +/// [`ParallelIterator`]: trait.ParallelIterator.html +#[must_use = "iterator adaptors are lazy and do nothing unless consumed"] +#[derive(Clone, Debug)] +pub struct Intersperse<I> +where + I: ParallelIterator, + I::Item: Clone, +{ + base: I, + item: I::Item, +} + +impl<I> Intersperse<I> +where + I: ParallelIterator, + I::Item: Clone, +{ + /// Creates a new `Intersperse` iterator + pub(super) fn new(base: I, item: I::Item) -> Self { + Intersperse { base, item } + } +} + +impl<I> ParallelIterator for Intersperse<I> +where + I: ParallelIterator, + I::Item: Clone + Send, +{ + type Item = I::Item; + + fn drive_unindexed<C>(self, consumer: C) -> C::Result + where + C: UnindexedConsumer<I::Item>, + { + let consumer1 = IntersperseConsumer::new(consumer, self.item); + self.base.drive_unindexed(consumer1) + } + + fn opt_len(&self) -> Option<usize> { + match self.base.opt_len()? { + 0 => Some(0), + len => len.checked_add(len - 1), + } + } +} + +impl<I> IndexedParallelIterator for Intersperse<I> +where + I: IndexedParallelIterator, + I::Item: Clone + Send, +{ + fn drive<C>(self, consumer: C) -> C::Result + where + C: Consumer<Self::Item>, + { + let consumer1 = IntersperseConsumer::new(consumer, self.item); + self.base.drive(consumer1) + } + + fn len(&self) -> usize { + let len = self.base.len(); + if len > 0 { + len.checked_add(len - 1).expect("overflow") + } else { + 0 + } + } + + fn with_producer<CB>(self, callback: CB) -> CB::Output + where + CB: ProducerCallback<Self::Item>, + { + let len = self.len(); + return self.base.with_producer(Callback { + callback, + item: self.item, + len, + }); + + struct Callback<CB, T> { + callback: CB, + item: T, + len: usize, + } + + impl<T, CB> ProducerCallback<T> for Callback<CB, T> + where + CB: ProducerCallback<T>, + T: Clone + Send, + { + type Output = CB::Output; + + fn callback<P>(self, base: P) -> CB::Output + where + P: Producer<Item = T>, + { + let producer = IntersperseProducer::new(base, self.item, self.len); + self.callback.callback(producer) + } + } + } +} + +struct IntersperseProducer<P> +where + P: Producer, +{ + base: P, + item: P::Item, + len: usize, + clone_first: bool, +} + +impl<P> IntersperseProducer<P> +where + P: Producer, +{ + fn new(base: P, item: P::Item, len: usize) -> Self { + IntersperseProducer { + base, + item, + len, + clone_first: false, + } + } +} + +impl<P> Producer for IntersperseProducer<P> +where + P: Producer, + P::Item: Clone + Send, +{ + type Item = P::Item; + type IntoIter = IntersperseIter<P::IntoIter>; + + fn into_iter(self) -> Self::IntoIter { + IntersperseIter { + base: self.base.into_iter().fuse(), + item: self.item, + clone_first: self.len > 0 && self.clone_first, + + // If there's more than one item, then even lengths end the opposite + // of how they started with respect to interspersed clones. + clone_last: self.len > 1 && ((self.len & 1 == 0) ^ self.clone_first), + } + } + + fn min_len(&self) -> usize { + self.base.min_len() + } + fn max_len(&self) -> usize { + self.base.max_len() + } + + fn split_at(self, index: usize) -> (Self, Self) { + debug_assert!(index <= self.len); + + // The left needs half of the items from the base producer, and the + // other half will be our interspersed item. If we're not leading with + // a cloned item, then we need to round up the base number of items, + // otherwise round down. + let base_index = (index + !self.clone_first as usize) / 2; + let (left_base, right_base) = self.base.split_at(base_index); + + let left = IntersperseProducer { + base: left_base, + item: self.item.clone(), + len: index, + clone_first: self.clone_first, + }; + + let right = IntersperseProducer { + base: right_base, + item: self.item, + len: self.len - index, + + // If the index is odd, the right side toggles `clone_first`. + clone_first: (index & 1 == 1) ^ self.clone_first, + }; + + (left, right) + } + + fn fold_with<F>(self, folder: F) -> F + where + F: Folder<Self::Item>, + { + let folder1 = IntersperseFolder { + base: folder, + item: self.item, + clone_first: self.clone_first, + }; + self.base.fold_with(folder1).base + } +} + +struct IntersperseIter<I> +where + I: Iterator, +{ + base: Fuse<I>, + item: I::Item, + clone_first: bool, + clone_last: bool, +} + +impl<I> Iterator for IntersperseIter<I> +where + I: DoubleEndedIterator + ExactSizeIterator, + I::Item: Clone, +{ + type Item = I::Item; + + fn next(&mut self) -> Option<Self::Item> { + if self.clone_first { + self.clone_first = false; + Some(self.item.clone()) + } else if let next @ Some(_) = self.base.next() { + // If there are any items left, we'll need another clone in front. + self.clone_first = self.base.len() != 0; + next + } else if self.clone_last { + self.clone_last = false; + Some(self.item.clone()) + } else { + None + } + } + + fn size_hint(&self) -> (usize, Option<usize>) { + let len = self.len(); + (len, Some(len)) + } +} + +impl<I> DoubleEndedIterator for IntersperseIter<I> +where + I: DoubleEndedIterator + ExactSizeIterator, + I::Item: Clone, +{ + fn next_back(&mut self) -> Option<Self::Item> { + if self.clone_last { + self.clone_last = false; + Some(self.item.clone()) + } else if let next_back @ Some(_) = self.base.next_back() { + // If there are any items left, we'll need another clone in back. + self.clone_last = self.base.len() != 0; + next_back + } else if self.clone_first { + self.clone_first = false; + Some(self.item.clone()) + } else { + None + } + } +} + +impl<I> ExactSizeIterator for IntersperseIter<I> +where + I: DoubleEndedIterator + ExactSizeIterator, + I::Item: Clone, +{ + fn len(&self) -> usize { + let len = self.base.len(); + len + len.saturating_sub(1) + self.clone_first as usize + self.clone_last as usize + } +} + +struct IntersperseConsumer<C, T> { + base: C, + item: T, + clone_first: Cell<bool>, +} + +impl<C, T> IntersperseConsumer<C, T> +where + C: Consumer<T>, +{ + fn new(base: C, item: T) -> Self { + IntersperseConsumer { + base, + item, + clone_first: false.into(), + } + } +} + +impl<C, T> Consumer<T> for IntersperseConsumer<C, T> +where + C: Consumer<T>, + T: Clone + Send, +{ + type Folder = IntersperseFolder<C::Folder, T>; + type Reducer = C::Reducer; + type Result = C::Result; + + fn split_at(mut self, index: usize) -> (Self, Self, Self::Reducer) { + // We'll feed twice as many items to the base consumer, except if we're + // not currently leading with a cloned item, then it's one less. + let base_index = index + index.saturating_sub(!self.clone_first.get() as usize); + let (left, right, reducer) = self.base.split_at(base_index); + + let right = IntersperseConsumer { + base: right, + item: self.item.clone(), + clone_first: true.into(), + }; + self.base = left; + (self, right, reducer) + } + + fn into_folder(self) -> Self::Folder { + IntersperseFolder { + base: self.base.into_folder(), + item: self.item, + clone_first: self.clone_first.get(), + } + } + + fn full(&self) -> bool { + self.base.full() + } +} + +impl<C, T> UnindexedConsumer<T> for IntersperseConsumer<C, T> +where + C: UnindexedConsumer<T>, + T: Clone + Send, +{ + fn split_off_left(&self) -> Self { + let left = IntersperseConsumer { + base: self.base.split_off_left(), + item: self.item.clone(), + clone_first: self.clone_first.clone(), + }; + self.clone_first.set(true); + left + } + + fn to_reducer(&self) -> Self::Reducer { + self.base.to_reducer() + } +} + +struct IntersperseFolder<C, T> { + base: C, + item: T, + clone_first: bool, +} + +impl<C, T> Folder<T> for IntersperseFolder<C, T> +where + C: Folder<T>, + T: Clone, +{ + type Result = C::Result; + + fn consume(mut self, item: T) -> Self { + if self.clone_first { + self.base = self.base.consume(self.item.clone()); + if self.base.full() { + return self; + } + } else { + self.clone_first = true; + } + self.base = self.base.consume(item); + self + } + + fn consume_iter<I>(self, iter: I) -> Self + where + I: IntoIterator<Item = T>, + { + let mut clone_first = self.clone_first; + let between_item = self.item; + let base = self.base.consume_iter(iter.into_iter().flat_map(|item| { + let first = if clone_first { + Some(between_item.clone()) + } else { + clone_first = true; + None + }; + first.into_iter().chain(iter::once(item)) + })); + IntersperseFolder { + base, + item: between_item, + clone_first, + } + } + + fn complete(self) -> C::Result { + self.base.complete() + } + + fn full(&self) -> bool { + self.base.full() + } +} diff --git a/third_party/rust/rayon/src/iter/len.rs b/third_party/rust/rayon/src/iter/len.rs new file mode 100644 index 0000000000..8ec7f33a15 --- /dev/null +++ b/third_party/rust/rayon/src/iter/len.rs @@ -0,0 +1,271 @@ +use super::plumbing::*; +use super::*; +use std::cmp; + +/// `MinLen` is an iterator that imposes a minimum length on iterator splits. +/// This struct is created by the [`with_min_len()`] method on [`IndexedParallelIterator`] +/// +/// [`with_min_len()`]: trait.IndexedParallelIterator.html#method.with_min_len +/// [`IndexedParallelIterator`]: trait.IndexedParallelIterator.html +#[must_use = "iterator adaptors are lazy and do nothing unless consumed"] +#[derive(Debug, Clone)] +pub struct MinLen<I: IndexedParallelIterator> { + base: I, + min: usize, +} + +impl<I> MinLen<I> +where + I: IndexedParallelIterator, +{ + /// Creates a new `MinLen` iterator. + pub(super) fn new(base: I, min: usize) -> Self { + MinLen { base, min } + } +} + +impl<I> ParallelIterator for MinLen<I> +where + I: IndexedParallelIterator, +{ + type Item = I::Item; + + fn drive_unindexed<C>(self, consumer: C) -> C::Result + where + C: UnindexedConsumer<Self::Item>, + { + bridge(self, consumer) + } + + fn opt_len(&self) -> Option<usize> { + Some(self.len()) + } +} + +impl<I> IndexedParallelIterator for MinLen<I> +where + I: IndexedParallelIterator, +{ + fn drive<C: Consumer<Self::Item>>(self, consumer: C) -> C::Result { + bridge(self, consumer) + } + + fn len(&self) -> usize { + self.base.len() + } + + fn with_producer<CB>(self, callback: CB) -> CB::Output + where + CB: ProducerCallback<Self::Item>, + { + return self.base.with_producer(Callback { + callback, + min: self.min, + }); + + struct Callback<CB> { + callback: CB, + min: usize, + } + + impl<T, CB> ProducerCallback<T> for Callback<CB> + where + CB: ProducerCallback<T>, + { + type Output = CB::Output; + fn callback<P>(self, base: P) -> CB::Output + where + P: Producer<Item = T>, + { + let producer = MinLenProducer { + base, + min: self.min, + }; + self.callback.callback(producer) + } + } + } +} + +/// //////////////////////////////////////////////////////////////////////// +/// `MinLenProducer` implementation + +struct MinLenProducer<P> { + base: P, + min: usize, +} + +impl<P> Producer for MinLenProducer<P> +where + P: Producer, +{ + type Item = P::Item; + type IntoIter = P::IntoIter; + + fn into_iter(self) -> Self::IntoIter { + self.base.into_iter() + } + + fn min_len(&self) -> usize { + cmp::max(self.min, self.base.min_len()) + } + + fn max_len(&self) -> usize { + self.base.max_len() + } + + fn split_at(self, index: usize) -> (Self, Self) { + let (left, right) = self.base.split_at(index); + ( + MinLenProducer { + base: left, + min: self.min, + }, + MinLenProducer { + base: right, + min: self.min, + }, + ) + } + + fn fold_with<F>(self, folder: F) -> F + where + F: Folder<Self::Item>, + { + self.base.fold_with(folder) + } +} + +/// `MaxLen` is an iterator that imposes a maximum length on iterator splits. +/// This struct is created by the [`with_max_len()`] method on [`IndexedParallelIterator`] +/// +/// [`with_max_len()`]: trait.IndexedParallelIterator.html#method.with_max_len +/// [`IndexedParallelIterator`]: trait.IndexedParallelIterator.html +#[must_use = "iterator adaptors are lazy and do nothing unless consumed"] +#[derive(Debug, Clone)] +pub struct MaxLen<I: IndexedParallelIterator> { + base: I, + max: usize, +} + +impl<I> MaxLen<I> +where + I: IndexedParallelIterator, +{ + /// Creates a new `MaxLen` iterator. + pub(super) fn new(base: I, max: usize) -> Self { + MaxLen { base, max } + } +} + +impl<I> ParallelIterator for MaxLen<I> +where + I: IndexedParallelIterator, +{ + type Item = I::Item; + + fn drive_unindexed<C>(self, consumer: C) -> C::Result + where + C: UnindexedConsumer<Self::Item>, + { + bridge(self, consumer) + } + + fn opt_len(&self) -> Option<usize> { + Some(self.len()) + } +} + +impl<I> IndexedParallelIterator for MaxLen<I> +where + I: IndexedParallelIterator, +{ + fn drive<C: Consumer<Self::Item>>(self, consumer: C) -> C::Result { + bridge(self, consumer) + } + + fn len(&self) -> usize { + self.base.len() + } + + fn with_producer<CB>(self, callback: CB) -> CB::Output + where + CB: ProducerCallback<Self::Item>, + { + return self.base.with_producer(Callback { + callback, + max: self.max, + }); + + struct Callback<CB> { + callback: CB, + max: usize, + } + + impl<T, CB> ProducerCallback<T> for Callback<CB> + where + CB: ProducerCallback<T>, + { + type Output = CB::Output; + fn callback<P>(self, base: P) -> CB::Output + where + P: Producer<Item = T>, + { + let producer = MaxLenProducer { + base, + max: self.max, + }; + self.callback.callback(producer) + } + } + } +} + +/// //////////////////////////////////////////////////////////////////////// +/// `MaxLenProducer` implementation + +struct MaxLenProducer<P> { + base: P, + max: usize, +} + +impl<P> Producer for MaxLenProducer<P> +where + P: Producer, +{ + type Item = P::Item; + type IntoIter = P::IntoIter; + + fn into_iter(self) -> Self::IntoIter { + self.base.into_iter() + } + + fn min_len(&self) -> usize { + self.base.min_len() + } + + fn max_len(&self) -> usize { + cmp::min(self.max, self.base.max_len()) + } + + fn split_at(self, index: usize) -> (Self, Self) { + let (left, right) = self.base.split_at(index); + ( + MaxLenProducer { + base: left, + max: self.max, + }, + MaxLenProducer { + base: right, + max: self.max, + }, + ) + } + + fn fold_with<F>(self, folder: F) -> F + where + F: Folder<Self::Item>, + { + self.base.fold_with(folder) + } +} diff --git a/third_party/rust/rayon/src/iter/map.rs b/third_party/rust/rayon/src/iter/map.rs new file mode 100644 index 0000000000..da14d4082c --- /dev/null +++ b/third_party/rust/rayon/src/iter/map.rs @@ -0,0 +1,259 @@ +use super::plumbing::*; +use super::*; + +use std::fmt::{self, Debug}; +use std::iter; + +/// `Map` is an iterator that transforms the elements of an underlying iterator. +/// +/// This struct is created by the [`map()`] method on [`ParallelIterator`] +/// +/// [`map()`]: trait.ParallelIterator.html#method.map +/// [`ParallelIterator`]: trait.ParallelIterator.html +#[must_use = "iterator adaptors are lazy and do nothing unless consumed"] +#[derive(Clone)] +pub struct Map<I: ParallelIterator, F> { + base: I, + map_op: F, +} + +impl<I: ParallelIterator + Debug, F> Debug for Map<I, F> { + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + f.debug_struct("Map").field("base", &self.base).finish() + } +} + +impl<I, F> Map<I, F> +where + I: ParallelIterator, +{ + /// Creates a new `Map` iterator. + pub(super) fn new(base: I, map_op: F) -> Self { + Map { base, map_op } + } +} + +impl<I, F, R> ParallelIterator for Map<I, F> +where + I: ParallelIterator, + F: Fn(I::Item) -> R + Sync + Send, + R: Send, +{ + type Item = F::Output; + + fn drive_unindexed<C>(self, consumer: C) -> C::Result + where + C: UnindexedConsumer<Self::Item>, + { + let consumer1 = MapConsumer::new(consumer, &self.map_op); + self.base.drive_unindexed(consumer1) + } + + fn opt_len(&self) -> Option<usize> { + self.base.opt_len() + } +} + +impl<I, F, R> IndexedParallelIterator for Map<I, F> +where + I: IndexedParallelIterator, + F: Fn(I::Item) -> R + Sync + Send, + R: Send, +{ + fn drive<C>(self, consumer: C) -> C::Result + where + C: Consumer<Self::Item>, + { + let consumer1 = MapConsumer::new(consumer, &self.map_op); + self.base.drive(consumer1) + } + + fn len(&self) -> usize { + self.base.len() + } + + fn with_producer<CB>(self, callback: CB) -> CB::Output + where + CB: ProducerCallback<Self::Item>, + { + return self.base.with_producer(Callback { + callback, + map_op: self.map_op, + }); + + struct Callback<CB, F> { + callback: CB, + map_op: F, + } + + impl<T, F, R, CB> ProducerCallback<T> for Callback<CB, F> + where + CB: ProducerCallback<R>, + F: Fn(T) -> R + Sync, + R: Send, + { + type Output = CB::Output; + + fn callback<P>(self, base: P) -> CB::Output + where + P: Producer<Item = T>, + { + let producer = MapProducer { + base, + map_op: &self.map_op, + }; + self.callback.callback(producer) + } + } + } +} + +/// //////////////////////////////////////////////////////////////////////// + +struct MapProducer<'f, P, F> { + base: P, + map_op: &'f F, +} + +impl<'f, P, F, R> Producer for MapProducer<'f, P, F> +where + P: Producer, + F: Fn(P::Item) -> R + Sync, + R: Send, +{ + type Item = F::Output; + type IntoIter = iter::Map<P::IntoIter, &'f F>; + + fn into_iter(self) -> Self::IntoIter { + self.base.into_iter().map(self.map_op) + } + + fn min_len(&self) -> usize { + self.base.min_len() + } + fn max_len(&self) -> usize { + self.base.max_len() + } + + fn split_at(self, index: usize) -> (Self, Self) { + let (left, right) = self.base.split_at(index); + ( + MapProducer { + base: left, + map_op: self.map_op, + }, + MapProducer { + base: right, + map_op: self.map_op, + }, + ) + } + + fn fold_with<G>(self, folder: G) -> G + where + G: Folder<Self::Item>, + { + let folder1 = MapFolder { + base: folder, + map_op: self.map_op, + }; + self.base.fold_with(folder1).base + } +} + +/// //////////////////////////////////////////////////////////////////////// +/// Consumer implementation + +struct MapConsumer<'f, C, F> { + base: C, + map_op: &'f F, +} + +impl<'f, C, F> MapConsumer<'f, C, F> { + fn new(base: C, map_op: &'f F) -> Self { + MapConsumer { base, map_op } + } +} + +impl<'f, T, R, C, F> Consumer<T> for MapConsumer<'f, C, F> +where + C: Consumer<F::Output>, + F: Fn(T) -> R + Sync, + R: Send, +{ + type Folder = MapFolder<'f, C::Folder, F>; + type Reducer = C::Reducer; + type Result = C::Result; + + fn split_at(self, index: usize) -> (Self, Self, Self::Reducer) { + let (left, right, reducer) = self.base.split_at(index); + ( + MapConsumer::new(left, self.map_op), + MapConsumer::new(right, self.map_op), + reducer, + ) + } + + fn into_folder(self) -> Self::Folder { + MapFolder { + base: self.base.into_folder(), + map_op: self.map_op, + } + } + + fn full(&self) -> bool { + self.base.full() + } +} + +impl<'f, T, R, C, F> UnindexedConsumer<T> for MapConsumer<'f, C, F> +where + C: UnindexedConsumer<F::Output>, + F: Fn(T) -> R + Sync, + R: Send, +{ + fn split_off_left(&self) -> Self { + MapConsumer::new(self.base.split_off_left(), self.map_op) + } + + fn to_reducer(&self) -> Self::Reducer { + self.base.to_reducer() + } +} + +struct MapFolder<'f, C, F> { + base: C, + map_op: &'f F, +} + +impl<'f, T, R, C, F> Folder<T> for MapFolder<'f, C, F> +where + C: Folder<F::Output>, + F: Fn(T) -> R, +{ + type Result = C::Result; + + fn consume(self, item: T) -> Self { + let mapped_item = (self.map_op)(item); + MapFolder { + base: self.base.consume(mapped_item), + map_op: self.map_op, + } + } + + fn consume_iter<I>(mut self, iter: I) -> Self + where + I: IntoIterator<Item = T>, + { + self.base = self.base.consume_iter(iter.into_iter().map(self.map_op)); + self + } + + fn complete(self) -> C::Result { + self.base.complete() + } + + fn full(&self) -> bool { + self.base.full() + } +} diff --git a/third_party/rust/rayon/src/iter/map_with.rs b/third_party/rust/rayon/src/iter/map_with.rs new file mode 100644 index 0000000000..10b1b4caca --- /dev/null +++ b/third_party/rust/rayon/src/iter/map_with.rs @@ -0,0 +1,573 @@ +use super::plumbing::*; +use super::*; + +use std::fmt::{self, Debug}; + +/// `MapWith` is an iterator that transforms the elements of an underlying iterator. +/// +/// This struct is created by the [`map_with()`] method on [`ParallelIterator`] +/// +/// [`map_with()`]: trait.ParallelIterator.html#method.map_with +/// [`ParallelIterator`]: trait.ParallelIterator.html +#[must_use = "iterator adaptors are lazy and do nothing unless consumed"] +#[derive(Clone)] +pub struct MapWith<I: ParallelIterator, T, F> { + base: I, + item: T, + map_op: F, +} + +impl<I: ParallelIterator + Debug, T: Debug, F> Debug for MapWith<I, T, F> { + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + f.debug_struct("MapWith") + .field("base", &self.base) + .field("item", &self.item) + .finish() + } +} + +impl<I, T, F> MapWith<I, T, F> +where + I: ParallelIterator, +{ + /// Creates a new `MapWith` iterator. + pub(super) fn new(base: I, item: T, map_op: F) -> Self { + MapWith { base, item, map_op } + } +} + +impl<I, T, F, R> ParallelIterator for MapWith<I, T, F> +where + I: ParallelIterator, + T: Send + Clone, + F: Fn(&mut T, I::Item) -> R + Sync + Send, + R: Send, +{ + type Item = R; + + fn drive_unindexed<C>(self, consumer: C) -> C::Result + where + C: UnindexedConsumer<Self::Item>, + { + let consumer1 = MapWithConsumer::new(consumer, self.item, &self.map_op); + self.base.drive_unindexed(consumer1) + } + + fn opt_len(&self) -> Option<usize> { + self.base.opt_len() + } +} + +impl<I, T, F, R> IndexedParallelIterator for MapWith<I, T, F> +where + I: IndexedParallelIterator, + T: Send + Clone, + F: Fn(&mut T, I::Item) -> R + Sync + Send, + R: Send, +{ + fn drive<C>(self, consumer: C) -> C::Result + where + C: Consumer<Self::Item>, + { + let consumer1 = MapWithConsumer::new(consumer, self.item, &self.map_op); + self.base.drive(consumer1) + } + + fn len(&self) -> usize { + self.base.len() + } + + fn with_producer<CB>(self, callback: CB) -> CB::Output + where + CB: ProducerCallback<Self::Item>, + { + return self.base.with_producer(Callback { + callback, + item: self.item, + map_op: self.map_op, + }); + + struct Callback<CB, U, F> { + callback: CB, + item: U, + map_op: F, + } + + impl<T, U, F, R, CB> ProducerCallback<T> for Callback<CB, U, F> + where + CB: ProducerCallback<R>, + U: Send + Clone, + F: Fn(&mut U, T) -> R + Sync, + R: Send, + { + type Output = CB::Output; + + fn callback<P>(self, base: P) -> CB::Output + where + P: Producer<Item = T>, + { + let producer = MapWithProducer { + base, + item: self.item, + map_op: &self.map_op, + }; + self.callback.callback(producer) + } + } + } +} + +/// //////////////////////////////////////////////////////////////////////// + +struct MapWithProducer<'f, P, U, F> { + base: P, + item: U, + map_op: &'f F, +} + +impl<'f, P, U, F, R> Producer for MapWithProducer<'f, P, U, F> +where + P: Producer, + U: Send + Clone, + F: Fn(&mut U, P::Item) -> R + Sync, + R: Send, +{ + type Item = R; + type IntoIter = MapWithIter<'f, P::IntoIter, U, F>; + + fn into_iter(self) -> Self::IntoIter { + MapWithIter { + base: self.base.into_iter(), + item: self.item, + map_op: self.map_op, + } + } + + fn min_len(&self) -> usize { + self.base.min_len() + } + fn max_len(&self) -> usize { + self.base.max_len() + } + + fn split_at(self, index: usize) -> (Self, Self) { + let (left, right) = self.base.split_at(index); + ( + MapWithProducer { + base: left, + item: self.item.clone(), + map_op: self.map_op, + }, + MapWithProducer { + base: right, + item: self.item, + map_op: self.map_op, + }, + ) + } + + fn fold_with<G>(self, folder: G) -> G + where + G: Folder<Self::Item>, + { + let folder1 = MapWithFolder { + base: folder, + item: self.item, + map_op: self.map_op, + }; + self.base.fold_with(folder1).base + } +} + +struct MapWithIter<'f, I, U, F> { + base: I, + item: U, + map_op: &'f F, +} + +impl<'f, I, U, F, R> Iterator for MapWithIter<'f, I, U, F> +where + I: Iterator, + F: Fn(&mut U, I::Item) -> R + Sync, + R: Send, +{ + type Item = R; + + fn next(&mut self) -> Option<R> { + let item = self.base.next()?; + Some((self.map_op)(&mut self.item, item)) + } + + fn size_hint(&self) -> (usize, Option<usize>) { + self.base.size_hint() + } +} + +impl<'f, I, U, F, R> DoubleEndedIterator for MapWithIter<'f, I, U, F> +where + I: DoubleEndedIterator, + F: Fn(&mut U, I::Item) -> R + Sync, + R: Send, +{ + fn next_back(&mut self) -> Option<R> { + let item = self.base.next_back()?; + Some((self.map_op)(&mut self.item, item)) + } +} + +impl<'f, I, U, F, R> ExactSizeIterator for MapWithIter<'f, I, U, F> +where + I: ExactSizeIterator, + F: Fn(&mut U, I::Item) -> R + Sync, + R: Send, +{ +} + +/// //////////////////////////////////////////////////////////////////////// +/// Consumer implementation + +struct MapWithConsumer<'f, C, U, F> { + base: C, + item: U, + map_op: &'f F, +} + +impl<'f, C, U, F> MapWithConsumer<'f, C, U, F> { + fn new(base: C, item: U, map_op: &'f F) -> Self { + MapWithConsumer { base, item, map_op } + } +} + +impl<'f, T, U, R, C, F> Consumer<T> for MapWithConsumer<'f, C, U, F> +where + C: Consumer<R>, + U: Send + Clone, + F: Fn(&mut U, T) -> R + Sync, + R: Send, +{ + type Folder = MapWithFolder<'f, C::Folder, U, F>; + type Reducer = C::Reducer; + type Result = C::Result; + + fn split_at(self, index: usize) -> (Self, Self, Self::Reducer) { + let (left, right, reducer) = self.base.split_at(index); + ( + MapWithConsumer::new(left, self.item.clone(), self.map_op), + MapWithConsumer::new(right, self.item, self.map_op), + reducer, + ) + } + + fn into_folder(self) -> Self::Folder { + MapWithFolder { + base: self.base.into_folder(), + item: self.item, + map_op: self.map_op, + } + } + + fn full(&self) -> bool { + self.base.full() + } +} + +impl<'f, T, U, R, C, F> UnindexedConsumer<T> for MapWithConsumer<'f, C, U, F> +where + C: UnindexedConsumer<R>, + U: Send + Clone, + F: Fn(&mut U, T) -> R + Sync, + R: Send, +{ + fn split_off_left(&self) -> Self { + MapWithConsumer::new(self.base.split_off_left(), self.item.clone(), self.map_op) + } + + fn to_reducer(&self) -> Self::Reducer { + self.base.to_reducer() + } +} + +struct MapWithFolder<'f, C, U, F> { + base: C, + item: U, + map_op: &'f F, +} + +impl<'f, T, U, R, C, F> Folder<T> for MapWithFolder<'f, C, U, F> +where + C: Folder<R>, + F: Fn(&mut U, T) -> R, +{ + type Result = C::Result; + + fn consume(mut self, item: T) -> Self { + let mapped_item = (self.map_op)(&mut self.item, item); + self.base = self.base.consume(mapped_item); + self + } + + fn consume_iter<I>(mut self, iter: I) -> Self + where + I: IntoIterator<Item = T>, + { + fn with<'f, T, U, R>( + item: &'f mut U, + map_op: impl Fn(&mut U, T) -> R + 'f, + ) -> impl FnMut(T) -> R + 'f { + move |x| map_op(item, x) + } + + { + let mapped_iter = iter.into_iter().map(with(&mut self.item, self.map_op)); + self.base = self.base.consume_iter(mapped_iter); + } + self + } + + fn complete(self) -> C::Result { + self.base.complete() + } + + fn full(&self) -> bool { + self.base.full() + } +} + +// ------------------------------------------------------------------------------------------------ + +/// `MapInit` is an iterator that transforms the elements of an underlying iterator. +/// +/// This struct is created by the [`map_init()`] method on [`ParallelIterator`] +/// +/// [`map_init()`]: trait.ParallelIterator.html#method.map_init +/// [`ParallelIterator`]: trait.ParallelIterator.html +#[must_use = "iterator adaptors are lazy and do nothing unless consumed"] +#[derive(Clone)] +pub struct MapInit<I: ParallelIterator, INIT, F> { + base: I, + init: INIT, + map_op: F, +} + +impl<I: ParallelIterator + Debug, INIT, F> Debug for MapInit<I, INIT, F> { + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + f.debug_struct("MapInit").field("base", &self.base).finish() + } +} + +impl<I, INIT, F> MapInit<I, INIT, F> +where + I: ParallelIterator, +{ + /// Creates a new `MapInit` iterator. + pub(super) fn new(base: I, init: INIT, map_op: F) -> Self { + MapInit { base, init, map_op } + } +} + +impl<I, INIT, T, F, R> ParallelIterator for MapInit<I, INIT, F> +where + I: ParallelIterator, + INIT: Fn() -> T + Sync + Send, + F: Fn(&mut T, I::Item) -> R + Sync + Send, + R: Send, +{ + type Item = R; + + fn drive_unindexed<C>(self, consumer: C) -> C::Result + where + C: UnindexedConsumer<Self::Item>, + { + let consumer1 = MapInitConsumer::new(consumer, &self.init, &self.map_op); + self.base.drive_unindexed(consumer1) + } + + fn opt_len(&self) -> Option<usize> { + self.base.opt_len() + } +} + +impl<I, INIT, T, F, R> IndexedParallelIterator for MapInit<I, INIT, F> +where + I: IndexedParallelIterator, + INIT: Fn() -> T + Sync + Send, + F: Fn(&mut T, I::Item) -> R + Sync + Send, + R: Send, +{ + fn drive<C>(self, consumer: C) -> C::Result + where + C: Consumer<Self::Item>, + { + let consumer1 = MapInitConsumer::new(consumer, &self.init, &self.map_op); + self.base.drive(consumer1) + } + + fn len(&self) -> usize { + self.base.len() + } + + fn with_producer<CB>(self, callback: CB) -> CB::Output + where + CB: ProducerCallback<Self::Item>, + { + return self.base.with_producer(Callback { + callback, + init: self.init, + map_op: self.map_op, + }); + + struct Callback<CB, INIT, F> { + callback: CB, + init: INIT, + map_op: F, + } + + impl<T, INIT, U, F, R, CB> ProducerCallback<T> for Callback<CB, INIT, F> + where + CB: ProducerCallback<R>, + INIT: Fn() -> U + Sync, + F: Fn(&mut U, T) -> R + Sync, + R: Send, + { + type Output = CB::Output; + + fn callback<P>(self, base: P) -> CB::Output + where + P: Producer<Item = T>, + { + let producer = MapInitProducer { + base, + init: &self.init, + map_op: &self.map_op, + }; + self.callback.callback(producer) + } + } + } +} + +/// //////////////////////////////////////////////////////////////////////// + +struct MapInitProducer<'f, P, INIT, F> { + base: P, + init: &'f INIT, + map_op: &'f F, +} + +impl<'f, P, INIT, U, F, R> Producer for MapInitProducer<'f, P, INIT, F> +where + P: Producer, + INIT: Fn() -> U + Sync, + F: Fn(&mut U, P::Item) -> R + Sync, + R: Send, +{ + type Item = R; + type IntoIter = MapWithIter<'f, P::IntoIter, U, F>; + + fn into_iter(self) -> Self::IntoIter { + MapWithIter { + base: self.base.into_iter(), + item: (self.init)(), + map_op: self.map_op, + } + } + + fn min_len(&self) -> usize { + self.base.min_len() + } + fn max_len(&self) -> usize { + self.base.max_len() + } + + fn split_at(self, index: usize) -> (Self, Self) { + let (left, right) = self.base.split_at(index); + ( + MapInitProducer { + base: left, + init: self.init, + map_op: self.map_op, + }, + MapInitProducer { + base: right, + init: self.init, + map_op: self.map_op, + }, + ) + } + + fn fold_with<G>(self, folder: G) -> G + where + G: Folder<Self::Item>, + { + let folder1 = MapWithFolder { + base: folder, + item: (self.init)(), + map_op: self.map_op, + }; + self.base.fold_with(folder1).base + } +} + +/// //////////////////////////////////////////////////////////////////////// +/// Consumer implementation + +struct MapInitConsumer<'f, C, INIT, F> { + base: C, + init: &'f INIT, + map_op: &'f F, +} + +impl<'f, C, INIT, F> MapInitConsumer<'f, C, INIT, F> { + fn new(base: C, init: &'f INIT, map_op: &'f F) -> Self { + MapInitConsumer { base, init, map_op } + } +} + +impl<'f, T, INIT, U, R, C, F> Consumer<T> for MapInitConsumer<'f, C, INIT, F> +where + C: Consumer<R>, + INIT: Fn() -> U + Sync, + F: Fn(&mut U, T) -> R + Sync, + R: Send, +{ + type Folder = MapWithFolder<'f, C::Folder, U, F>; + type Reducer = C::Reducer; + type Result = C::Result; + + fn split_at(self, index: usize) -> (Self, Self, Self::Reducer) { + let (left, right, reducer) = self.base.split_at(index); + ( + MapInitConsumer::new(left, self.init, self.map_op), + MapInitConsumer::new(right, self.init, self.map_op), + reducer, + ) + } + + fn into_folder(self) -> Self::Folder { + MapWithFolder { + base: self.base.into_folder(), + item: (self.init)(), + map_op: self.map_op, + } + } + + fn full(&self) -> bool { + self.base.full() + } +} + +impl<'f, T, INIT, U, R, C, F> UnindexedConsumer<T> for MapInitConsumer<'f, C, INIT, F> +where + C: UnindexedConsumer<R>, + INIT: Fn() -> U + Sync, + F: Fn(&mut U, T) -> R + Sync, + R: Send, +{ + fn split_off_left(&self) -> Self { + MapInitConsumer::new(self.base.split_off_left(), self.init, self.map_op) + } + + fn to_reducer(&self) -> Self::Reducer { + self.base.to_reducer() + } +} diff --git a/third_party/rust/rayon/src/iter/mod.rs b/third_party/rust/rayon/src/iter/mod.rs new file mode 100644 index 0000000000..98c93267f4 --- /dev/null +++ b/third_party/rust/rayon/src/iter/mod.rs @@ -0,0 +1,3385 @@ +//! Traits for writing parallel programs using an iterator-style interface +//! +//! You will rarely need to interact with this module directly unless you have +//! need to name one of the iterator types. +//! +//! Parallel iterators make it easy to write iterator-like chains that +//! execute in parallel: typically all you have to do is convert the +//! first `.iter()` (or `iter_mut()`, `into_iter()`, etc) method into +//! `par_iter()` (or `par_iter_mut()`, `into_par_iter()`, etc). For +//! example, to compute the sum of the squares of a sequence of +//! integers, one might write: +//! +//! ```rust +//! use rayon::prelude::*; +//! fn sum_of_squares(input: &[i32]) -> i32 { +//! input.par_iter() +//! .map(|i| i * i) +//! .sum() +//! } +//! ``` +//! +//! Or, to increment all the integers in a slice, you could write: +//! +//! ```rust +//! use rayon::prelude::*; +//! fn increment_all(input: &mut [i32]) { +//! input.par_iter_mut() +//! .for_each(|p| *p += 1); +//! } +//! ``` +//! +//! To use parallel iterators, first import the traits by adding +//! something like `use rayon::prelude::*` to your module. You can +//! then call `par_iter`, `par_iter_mut`, or `into_par_iter` to get a +//! parallel iterator. Like a [regular iterator][], parallel +//! iterators work by first constructing a computation and then +//! executing it. +//! +//! In addition to `par_iter()` and friends, some types offer other +//! ways to create (or consume) parallel iterators: +//! +//! - Slices (`&[T]`, `&mut [T]`) offer methods like `par_split` and +//! `par_windows`, as well as various parallel sorting +//! operations. See [the `ParallelSlice` trait] for the full list. +//! - Strings (`&str`) offer methods like `par_split` and `par_lines`. +//! See [the `ParallelString` trait] for the full list. +//! - Various collections offer [`par_extend`], which grows a +//! collection given a parallel iterator. (If you don't have a +//! collection to extend, you can use [`collect()`] to create a new +//! one from scratch.) +//! +//! [the `ParallelSlice` trait]: ../slice/trait.ParallelSlice.html +//! [the `ParallelString` trait]: ../str/trait.ParallelString.html +//! [`par_extend`]: trait.ParallelExtend.html +//! [`collect()`]: trait.ParallelIterator.html#method.collect +//! +//! To see the full range of methods available on parallel iterators, +//! check out the [`ParallelIterator`] and [`IndexedParallelIterator`] +//! traits. +//! +//! If you'd like to build a custom parallel iterator, or to write your own +//! combinator, then check out the [split] function and the [plumbing] module. +//! +//! [regular iterator]: https://doc.rust-lang.org/std/iter/trait.Iterator.html +//! [`ParallelIterator`]: trait.ParallelIterator.html +//! [`IndexedParallelIterator`]: trait.IndexedParallelIterator.html +//! [split]: fn.split.html +//! [plumbing]: plumbing/index.html +//! +//! Note: Several of the `ParallelIterator` methods rely on a `Try` trait which +//! has been deliberately obscured from the public API. This trait is intended +//! to mirror the unstable `std::ops::Try` with implementations for `Option` and +//! `Result`, where `Some`/`Ok` values will let those iterators continue, but +//! `None`/`Err` values will exit early. +//! +//! A note about object safety: It is currently _not_ possible to wrap +//! a `ParallelIterator` (or any trait that depends on it) using a +//! `Box<dyn ParallelIterator>` or other kind of dynamic allocation, +//! because `ParallelIterator` is **not object-safe**. +//! (This keeps the implementation simpler and allows extra optimizations.) + +use self::plumbing::*; +use self::private::Try; +pub use either::Either; +use std::cmp::{self, Ordering}; +use std::iter::{Product, Sum}; +use std::ops::{Fn, RangeBounds}; + +pub mod plumbing; + +#[cfg(test)] +mod test; + +// There is a method to the madness here: +// +// - These modules are private but expose certain types to the end-user +// (e.g., `enumerate::Enumerate`) -- specifically, the types that appear in the +// public API surface of the `ParallelIterator` traits. +// - In **this** module, those public types are always used unprefixed, which forces +// us to add a `pub use` and helps identify if we missed anything. +// - In contrast, items that appear **only** in the body of a method, +// e.g. `find::find()`, are always used **prefixed**, so that they +// can be readily distinguished. + +mod chain; +mod chunks; +mod cloned; +mod collect; +mod copied; +mod empty; +mod enumerate; +mod extend; +mod filter; +mod filter_map; +mod find; +mod find_first_last; +mod flat_map; +mod flat_map_iter; +mod flatten; +mod flatten_iter; +mod fold; +mod fold_chunks; +mod fold_chunks_with; +mod for_each; +mod from_par_iter; +mod inspect; +mod interleave; +mod interleave_shortest; +mod intersperse; +mod len; +mod map; +mod map_with; +mod multizip; +mod noop; +mod once; +mod panic_fuse; +mod par_bridge; +mod positions; +mod product; +mod reduce; +mod repeat; +mod rev; +mod skip; +mod splitter; +mod step_by; +mod sum; +mod take; +mod try_fold; +mod try_reduce; +mod try_reduce_with; +mod unzip; +mod update; +mod while_some; +mod zip; +mod zip_eq; + +pub use self::{ + chain::Chain, + chunks::Chunks, + cloned::Cloned, + copied::Copied, + empty::{empty, Empty}, + enumerate::Enumerate, + filter::Filter, + filter_map::FilterMap, + flat_map::FlatMap, + flat_map_iter::FlatMapIter, + flatten::Flatten, + flatten_iter::FlattenIter, + fold::{Fold, FoldWith}, + fold_chunks::FoldChunks, + fold_chunks_with::FoldChunksWith, + inspect::Inspect, + interleave::Interleave, + interleave_shortest::InterleaveShortest, + intersperse::Intersperse, + len::{MaxLen, MinLen}, + map::Map, + map_with::{MapInit, MapWith}, + multizip::MultiZip, + once::{once, Once}, + panic_fuse::PanicFuse, + par_bridge::{IterBridge, ParallelBridge}, + positions::Positions, + repeat::{repeat, repeatn, Repeat, RepeatN}, + rev::Rev, + skip::Skip, + splitter::{split, Split}, + step_by::StepBy, + take::Take, + try_fold::{TryFold, TryFoldWith}, + update::Update, + while_some::WhileSome, + zip::Zip, + zip_eq::ZipEq, +}; + +/// `IntoParallelIterator` implements the conversion to a [`ParallelIterator`]. +/// +/// By implementing `IntoParallelIterator` for a type, you define how it will +/// transformed into an iterator. This is a parallel version of the standard +/// library's [`std::iter::IntoIterator`] trait. +/// +/// [`ParallelIterator`]: trait.ParallelIterator.html +/// [`std::iter::IntoIterator`]: https://doc.rust-lang.org/std/iter/trait.IntoIterator.html +pub trait IntoParallelIterator { + /// The parallel iterator type that will be created. + type Iter: ParallelIterator<Item = Self::Item>; + + /// The type of item that the parallel iterator will produce. + type Item: Send; + + /// Converts `self` into a parallel iterator. + /// + /// # Examples + /// + /// ``` + /// use rayon::prelude::*; + /// + /// println!("counting in parallel:"); + /// (0..100).into_par_iter() + /// .for_each(|i| println!("{}", i)); + /// ``` + /// + /// This conversion is often implicit for arguments to methods like [`zip`]. + /// + /// ``` + /// use rayon::prelude::*; + /// + /// let v: Vec<_> = (0..5).into_par_iter().zip(5..10).collect(); + /// assert_eq!(v, [(0, 5), (1, 6), (2, 7), (3, 8), (4, 9)]); + /// ``` + /// + /// [`zip`]: trait.IndexedParallelIterator.html#method.zip + fn into_par_iter(self) -> Self::Iter; +} + +/// `IntoParallelRefIterator` implements the conversion to a +/// [`ParallelIterator`], providing shared references to the data. +/// +/// This is a parallel version of the `iter()` method +/// defined by various collections. +/// +/// This trait is automatically implemented +/// `for I where &I: IntoParallelIterator`. In most cases, users +/// will want to implement [`IntoParallelIterator`] rather than implement +/// this trait directly. +/// +/// [`ParallelIterator`]: trait.ParallelIterator.html +/// [`IntoParallelIterator`]: trait.IntoParallelIterator.html +pub trait IntoParallelRefIterator<'data> { + /// The type of the parallel iterator that will be returned. + type Iter: ParallelIterator<Item = Self::Item>; + + /// The type of item that the parallel iterator will produce. + /// This will typically be an `&'data T` reference type. + type Item: Send + 'data; + + /// Converts `self` into a parallel iterator. + /// + /// # Examples + /// + /// ``` + /// use rayon::prelude::*; + /// + /// let v: Vec<_> = (0..100).collect(); + /// assert_eq!(v.par_iter().sum::<i32>(), 100 * 99 / 2); + /// + /// // `v.par_iter()` is shorthand for `(&v).into_par_iter()`, + /// // producing the exact same references. + /// assert!(v.par_iter().zip(&v) + /// .all(|(a, b)| std::ptr::eq(a, b))); + /// ``` + fn par_iter(&'data self) -> Self::Iter; +} + +impl<'data, I: 'data + ?Sized> IntoParallelRefIterator<'data> for I +where + &'data I: IntoParallelIterator, +{ + type Iter = <&'data I as IntoParallelIterator>::Iter; + type Item = <&'data I as IntoParallelIterator>::Item; + + fn par_iter(&'data self) -> Self::Iter { + self.into_par_iter() + } +} + +/// `IntoParallelRefMutIterator` implements the conversion to a +/// [`ParallelIterator`], providing mutable references to the data. +/// +/// This is a parallel version of the `iter_mut()` method +/// defined by various collections. +/// +/// This trait is automatically implemented +/// `for I where &mut I: IntoParallelIterator`. In most cases, users +/// will want to implement [`IntoParallelIterator`] rather than implement +/// this trait directly. +/// +/// [`ParallelIterator`]: trait.ParallelIterator.html +/// [`IntoParallelIterator`]: trait.IntoParallelIterator.html +pub trait IntoParallelRefMutIterator<'data> { + /// The type of iterator that will be created. + type Iter: ParallelIterator<Item = Self::Item>; + + /// The type of item that will be produced; this is typically an + /// `&'data mut T` reference. + type Item: Send + 'data; + + /// Creates the parallel iterator from `self`. + /// + /// # Examples + /// + /// ``` + /// use rayon::prelude::*; + /// + /// let mut v = vec![0usize; 5]; + /// v.par_iter_mut().enumerate().for_each(|(i, x)| *x = i); + /// assert_eq!(v, [0, 1, 2, 3, 4]); + /// ``` + fn par_iter_mut(&'data mut self) -> Self::Iter; +} + +impl<'data, I: 'data + ?Sized> IntoParallelRefMutIterator<'data> for I +where + &'data mut I: IntoParallelIterator, +{ + type Iter = <&'data mut I as IntoParallelIterator>::Iter; + type Item = <&'data mut I as IntoParallelIterator>::Item; + + fn par_iter_mut(&'data mut self) -> Self::Iter { + self.into_par_iter() + } +} + +/// Parallel version of the standard iterator trait. +/// +/// The combinators on this trait are available on **all** parallel +/// iterators. Additional methods can be found on the +/// [`IndexedParallelIterator`] trait: those methods are only +/// available for parallel iterators where the number of items is +/// known in advance (so, e.g., after invoking `filter`, those methods +/// become unavailable). +/// +/// For examples of using parallel iterators, see [the docs on the +/// `iter` module][iter]. +/// +/// [iter]: index.html +/// [`IndexedParallelIterator`]: trait.IndexedParallelIterator.html +pub trait ParallelIterator: Sized + Send { + /// The type of item that this parallel iterator produces. + /// For example, if you use the [`for_each`] method, this is the type of + /// item that your closure will be invoked with. + /// + /// [`for_each`]: #method.for_each + type Item: 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<OP>(self, op: OP) + where + OP: Fn(Self::Item) + Sync + Send, + { + for_each::for_each(self, &op) + } + + /// 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_with<OP, T>(self, init: T, op: OP) + where + OP: Fn(&mut T, Self::Item) + Sync + Send, + T: Send + Clone, + { + self.map_with(init, op).collect() + } + + /// 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 for_each_init<OP, INIT, T>(self, init: INIT, op: OP) + where + OP: Fn(&mut T, Self::Item) + Sync + Send, + INIT: Fn() -> T + Sync + Send, + { + self.map_init(init, op).collect() + } + + /// 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<OP, R>(self, op: OP) -> R + where + OP: Fn(Self::Item) -> R + Sync + Send, + R: Try<Output = ()> + Send, + { + fn ok<R: Try<Output = ()>>(_: (), _: ()) -> R { + R::from_output(()) + } + + self.map(op).try_reduce(<()>::default, ok) + } + + /// 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()`]. + /// + /// [`for_each_with()`]: #method.for_each_with + /// [`try_for_each()`]: #method.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_with<OP, T, R>(self, init: T, op: OP) -> R + where + OP: Fn(&mut T, Self::Item) -> R + Sync + Send, + T: Send + Clone, + R: Try<Output = ()> + Send, + { + fn ok<R: Try<Output = ()>>(_: (), _: ()) -> R { + R::from_output(()) + } + + self.map_with(init, op).try_reduce(<()>::default, ok) + } + + /// 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()`]. + /// + /// [`for_each_init()`]: #method.for_each_init + /// [`try_for_each()`]: #method.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 try_for_each_init<OP, INIT, T, R>(self, init: INIT, op: OP) -> R + where + OP: Fn(&mut T, Self::Item) -> R + Sync + Send, + INIT: Fn() -> T + Sync + Send, + R: Try<Output = ()> + Send, + { + fn ok<R: Try<Output = ()>>(_: (), _: ()) -> R { + R::from_output(()) + } + + self.map_init(init, op).try_reduce(<()>::default, ok) + } + + /// 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 count(self) -> usize { + fn one<T>(_: T) -> usize { + 1 + } + + self.map(one).sum() + } + + /// 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<F, R>(self, map_op: F) -> Map<Self, F> + where + F: Fn(Self::Item) -> R + Sync + Send, + R: Send, + { + Map::new(self, map_op) + } + + /// 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_with<F, T, R>(self, init: T, map_op: F) -> MapWith<Self, T, F> + where + F: Fn(&mut T, Self::Item) -> R + Sync + Send, + T: Send + Clone, + R: Send, + { + MapWith::new(self, init, map_op) + } + + /// 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 map_init<F, INIT, T, R>(self, init: INIT, map_op: F) -> MapInit<Self, INIT, F> + where + F: Fn(&mut T, Self::Item) -> R + Sync + Send, + INIT: Fn() -> T + Sync + Send, + R: Send, + { + MapInit::new(self, init, map_op) + } + + /// 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()`]. + /// + /// [`copied()`]: #method.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 cloned<'a, T>(self) -> Cloned<Self> + where + T: 'a + Clone + Send, + Self: ParallelIterator<Item = &'a T>, + { + Cloned::new(self) + } + + /// 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()`]. + /// + /// [`cloned()`]: #method.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 copied<'a, T>(self) -> Copied<Self> + where + T: 'a + Copy + Send, + Self: ParallelIterator<Item = &'a T>, + { + Copied::new(self) + } + + /// 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 inspect<OP>(self, inspect_op: OP) -> Inspect<Self, OP> + where + OP: Fn(&Self::Item) + Sync + Send, + { + Inspect::new(self, inspect_op) + } + + /// 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 update<F>(self, update_op: F) -> Update<Self, F> + where + F: Fn(&mut Self::Item) + Sync + Send, + { + Update::new(self, update_op) + } + + /// 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<P>(self, filter_op: P) -> Filter<Self, P> + where + P: Fn(&Self::Item) -> bool + Sync + Send, + { + Filter::new(self, filter_op) + } + + /// 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 filter_map<P, R>(self, filter_op: P) -> FilterMap<Self, P> + where + P: Fn(Self::Item) -> Option<R> + Sync + Send, + R: Send, + { + FilterMap::new(self, filter_op) + } + + /// 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`](#method.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<F, PI>(self, map_op: F) -> FlatMap<Self, F> + where + F: Fn(Self::Item) -> PI + Sync + Send, + PI: IntoParallelIterator, + { + FlatMap::new(self, map_op) + } + + /// 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. + /// + /// [`flat_map`]: #method.flat_map + /// + /// # 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 flat_map_iter<F, SI>(self, map_op: F) -> FlatMapIter<Self, F> + where + F: Fn(Self::Item) -> SI + Sync + Send, + SI: IntoIterator, + SI::Item: Send, + { + FlatMapIter::new(self, map_op) + } + + /// An adaptor that flattens parallel-iterable `Item`s into one large iterator. + /// + /// See also [`flatten_iter`](#method.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(self) -> Flatten<Self> + where + Self::Item: IntoParallelIterator, + { + Flatten::new(self) + } + + /// An adaptor that flattens serial-iterable `Item`s into one large iterator. + /// + /// See also [`flatten`](#method.flatten) and the analogous comparison of + /// [`flat_map_iter` versus `flat_map`](#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 flatten_iter(self) -> FlattenIter<Self> + where + Self::Item: IntoIterator, + <Self::Item as IntoIterator>::Item: Send, + { + FlattenIter::new(self) + } + + /// 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. + /// + /// [associative]: https://en.wikipedia.org/wiki/Associative_property + fn reduce<OP, ID>(self, identity: ID, op: OP) -> Self::Item + where + OP: Fn(Self::Item, Self::Item) -> Self::Item + Sync + Send, + ID: Fn() -> Self::Item + Sync + Send, + { + reduce::reduce(self, identity, op) + } + + /// 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. + /// + /// [associative]: https://en.wikipedia.org/wiki/Associative_property + fn reduce_with<OP>(self, op: OP) -> Option<Self::Item> + where + OP: Fn(Self::Item, Self::Item) -> Self::Item + Sync + Send, + { + fn opt_fold<T>(op: impl Fn(T, T) -> T) -> impl Fn(Option<T>, T) -> Option<T> { + move |opt_a, b| match opt_a { + Some(a) => Some(op(a, b)), + None => Some(b), + } + } + + fn opt_reduce<T>(op: impl Fn(T, T) -> T) -> impl Fn(Option<T>, Option<T>) -> Option<T> { + move |opt_a, opt_b| match (opt_a, opt_b) { + (Some(a), Some(b)) => Some(op(a, b)), + (Some(v), None) | (None, Some(v)) => Some(v), + (None, None) => None, + } + } + + self.fold(<_>::default, opt_fold(&op)) + .reduce(<_>::default, opt_reduce(&op)) + } + + /// Reduces the items in the iterator into one item using a fallible `op`. + /// The `identity` argument is used the same way as in [`reduce()`]. + /// + /// [`reduce()`]: #method.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<T, OP, ID>(self, identity: ID, op: OP) -> Self::Item + where + OP: Fn(T, T) -> Self::Item + Sync + Send, + ID: Fn() -> T + Sync + Send, + Self::Item: Try<Output = T>, + { + try_reduce::try_reduce(self, identity, op) + } + + /// 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`. + /// + /// [`reduce_with()`]: #method.reduce_with + /// [`try_reduce()`]: #method.try_reduce + /// + /// 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 try_reduce_with<T, OP>(self, op: OP) -> Option<Self::Item> + where + OP: Fn(T, T) -> Self::Item + Sync + Send, + Self::Item: Try<Output = T>, + { + try_reduce_with::try_reduce_with(self, op) + } + + /// 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: + /// + /// ```notrust + /// 22 3 77 89 46 + /// | | + /// 102 135 + /// ``` + /// + /// Or perhaps these three numbers: + /// + /// ```notrust + /// 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<T, ID, F>(self, identity: ID, fold_op: F) -> Fold<Self, ID, F> + where + F: Fn(T, Self::Item) -> T + Sync + Send, + ID: Fn() -> T + Sync + Send, + T: Send, + { + Fold::new(self, identity, fold_op) + } + + /// 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 fold_with<F, T>(self, init: T, fold_op: F) -> FoldWith<Self, T, F> + where + F: Fn(T, Self::Item) -> T + Sync + Send, + T: Send + Clone, + { + FoldWith::new(self, init, fold_op) + } + + /// 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. + /// + /// [`fold()`]: #method.fold + /// [`try_reduce()`]: #method.try_reduce + /// + /// # 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<T, R, ID, F>(self, identity: ID, fold_op: F) -> TryFold<Self, R, ID, F> + where + F: Fn(T, Self::Item) -> R + Sync + Send, + ID: Fn() -> T + Sync + Send, + R: Try<Output = T> + Send, + { + TryFold::new(self, identity, fold_op) + } + + /// 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()`]. + /// + /// [`fold_with()`]: #method.fold_with + /// [`try_fold()`]: #method.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 try_fold_with<F, T, R>(self, init: T, fold_op: F) -> TryFoldWith<Self, R, F> + where + F: Fn(T, Self::Item) -> R + Sync + Send, + R: Try<Output = T> + Send, + T: Clone + Send, + { + TryFoldWith::new(self, init, fold_op) + } + + /// 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. + /// + /// [associative]: https://en.wikipedia.org/wiki/Associative_property + /// + /// 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 sum<S>(self) -> S + where + S: Send + Sum<Self::Item> + Sum<S>, + { + sum::sum(self) + } + + /// 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. + /// + /// [associative]: https://en.wikipedia.org/wiki/Associative_property + /// + /// 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 product<P>(self) -> P + where + P: Send + Product<Self::Item> + Product<P>, + { + product::product(self) + } + + /// 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| cmp::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(self) -> Option<Self::Item> + where + Self::Item: Ord, + { + self.reduce_with(cmp::min) + } + + /// 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<F>(self, f: F) -> Option<Self::Item> + where + F: Sync + Send + Fn(&Self::Item, &Self::Item) -> Ordering, + { + fn min<T>(f: impl Fn(&T, &T) -> Ordering) -> impl Fn(T, T) -> T { + move |a, b| match f(&a, &b) { + Ordering::Greater => b, + _ => a, + } + } + + self.reduce_with(min(f)) + } + + /// 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 min_by_key<K, F>(self, f: F) -> Option<Self::Item> + where + K: Ord + Send, + F: Sync + Send + Fn(&Self::Item) -> K, + { + fn key<T, K>(f: impl Fn(&T) -> K) -> impl Fn(T) -> (K, T) { + move |x| (f(&x), x) + } + + fn min_key<T, K: Ord>(a: (K, T), b: (K, T)) -> (K, T) { + match (a.0).cmp(&b.0) { + Ordering::Greater => b, + _ => a, + } + } + + let (_, x) = self.map(key(f)).reduce_with(min_key)?; + Some(x) + } + + /// 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| cmp::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(self) -> Option<Self::Item> + where + Self::Item: Ord, + { + self.reduce_with(cmp::max) + } + + /// 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(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().max_by(|x, y| x.abs().cmp(&y.abs())), Some(&240)); + /// ``` + fn max_by<F>(self, f: F) -> Option<Self::Item> + where + F: Sync + Send + Fn(&Self::Item, &Self::Item) -> Ordering, + { + fn max<T>(f: impl Fn(&T, &T) -> Ordering) -> impl Fn(T, T) -> T { + move |a, b| match f(&a, &b) { + Ordering::Greater => a, + _ => b, + } + } + + self.reduce_with(max(f)) + } + + /// 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 max_by_key<K, F>(self, f: F) -> Option<Self::Item> + where + K: Ord + Send, + F: Sync + Send + Fn(&Self::Item) -> K, + { + fn key<T, K>(f: impl Fn(&T) -> K) -> impl Fn(T) -> (K, T) { + move |x| (f(&x), x) + } + + fn max_key<T, K: Ord>(a: (K, T), b: (K, T)) -> (K, T) { + match (a.0).cmp(&b.0) { + Ordering::Greater => a, + _ => b, + } + } + + let (_, x) = self.map(key(f)).reduce_with(max_key)?; + Some(x) + } + + /// 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 chain<C>(self, chain: C) -> Chain<Self, C::Iter> + where + C: IntoParallelIterator<Item = Self::Item>, + { + Chain::new(self, chain.into_par_iter()) + } + + /// 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][find] 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). + /// + /// [find]: https://doc.rust-lang.org/std/iter/trait.Iterator.html#method.find + /// + /// # 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_any<P>(self, predicate: P) -> Option<Self::Item> + where + P: Fn(&Self::Item) -> bool + Sync + Send, + { + find::find(self, predicate) + } + + /// 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. + /// + /// 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_first<P>(self, predicate: P) -> Option<Self::Item> + where + P: Fn(&Self::Item) -> bool + Sync + Send, + { + find_first_last::find_first(self, predicate) + } + + /// 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_last<P>(self, predicate: P) -> Option<Self::Item> + where + P: Fn(&Self::Item) -> bool + Sync + Send, + { + find_first_last::find_last(self, predicate) + } + + /// 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_any<P, R>(self, predicate: P) -> Option<R> + where + P: Fn(Self::Item) -> Option<R> + Sync + Send, + R: Send, + { + fn yes<T>(_: &T) -> bool { + true + } + self.filter_map(predicate).find_any(yes) + } + + /// 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_first<P, R>(self, predicate: P) -> Option<R> + where + P: Fn(Self::Item) -> Option<R> + Sync + Send, + R: Send, + { + fn yes<T>(_: &T) -> bool { + true + } + self.filter_map(predicate).find_first(yes) + } + + /// 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 find_map_last<P, R>(self, predicate: P) -> Option<R> + where + P: Fn(Self::Item) -> Option<R> + Sync + Send, + R: Send, + { + fn yes<T>(_: &T) -> bool { + true + } + self.filter_map(predicate).find_last(yes) + } + + #[doc(hidden)] + #[deprecated(note = "parallel `find` does not search in order -- use `find_any`, \\ + `find_first`, or `find_last`")] + fn find<P>(self, predicate: P) -> Option<Self::Item> + where + P: Fn(&Self::Item) -> bool + Sync + Send, + { + self.find_any(predicate) + } + + /// 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 any<P>(self, predicate: P) -> bool + where + P: Fn(Self::Item) -> bool + Sync + Send, + { + self.map(predicate).find_any(bool::clone).is_some() + } + + /// 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 all<P>(self, predicate: P) -> bool + where + P: Fn(Self::Item) -> bool + Sync + Send, + { + #[inline] + fn is_false(x: &bool) -> bool { + !x + } + + self.map(predicate).find_any(is_false).is_none() + } + + /// 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 while_some<T>(self) -> WhileSome<Self> + where + Self: ParallelIterator<Item = Option<T>>, + T: Send, + { + WhileSome::new(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. + /// + /// [`join`]: ../fn.join.html#panics + /// + /// # 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. + /// + /// ```should_panic + /// 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 panic_fuse(self) -> PanicFuse<Self> { + PanicFuse::new(self) + } + + /// 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. + /// + /// [`IndexedParallelIterator`]: trait.IndexedParallelIterator.html + /// [`collect_into_vec()`]: + /// trait.IndexedParallelIterator.html#method.collect_into_vec + /// + /// # 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`](#method.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`](#method.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 collect<C>(self) -> C + where + C: FromParallelIterator<Self::Item>, + { + C::from_par_iter(self) + } + + /// 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 unzip<A, B, FromA, FromB>(self) -> (FromA, FromB) + where + Self: ParallelIterator<Item = (A, B)>, + FromA: Default + Send + ParallelExtend<A>, + FromB: Default + Send + ParallelExtend<B>, + A: Send, + B: Send, + { + unzip::unzip(self) + } + + /// 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<A, B, P>(self, predicate: P) -> (A, B) + where + A: Default + Send + ParallelExtend<Self::Item>, + B: Default + Send + ParallelExtend<Self::Item>, + P: Fn(&Self::Item) -> bool + Sync + Send, + { + unzip::partition(self, predicate) + } + + /// 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 partition_map<A, B, P, L, R>(self, predicate: P) -> (A, B) + where + A: Default + Send + ParallelExtend<L>, + B: Default + Send + ParallelExtend<R>, + P: Fn(Self::Item) -> Either<L, R> + Sync + Send, + L: Send, + R: Send, + { + unzip::partition_map(self, predicate) + } + + /// 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 intersperse(self, element: Self::Item) -> Intersperse<Self> + where + Self::Item: Clone, + { + Intersperse::new(self, element) + } + + /// Internal method used to define the behavior of this parallel + /// iterator. You should not need to call this directly. + /// + /// This method causes the iterator `self` to start producing + /// items and to feed them to the consumer `consumer` one by one. + /// It may split the consumer before doing so to create the + /// opportunity to produce in parallel. + /// + /// See the [README] for more details on the internals of parallel + /// iterators. + /// + /// [README]: https://github.com/rayon-rs/rayon/blob/master/src/iter/plumbing/README.md + fn drive_unindexed<C>(self, consumer: C) -> C::Result + where + C: UnindexedConsumer<Self::Item>; + + /// 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. + fn opt_len(&self) -> Option<usize> { + None + } +} + +impl<T: ParallelIterator> IntoParallelIterator for T { + type Iter = T; + type Item = T::Item; + + fn into_par_iter(self) -> T { + self + } +} + +/// An iterator that supports "random access" to its data, meaning +/// that you can split it at arbitrary indices and draw data from +/// those points. +/// +/// **Note:** Not implemented for `u64`, `i64`, `u128`, or `i128` ranges +// Waiting for `ExactSizeIterator::is_empty` to be stabilized. See rust-lang/rust#35428 +#[allow(clippy::len_without_is_empty)] +pub trait IndexedParallelIterator: ParallelIterator { + /// Collects the results of the iterator into the specified + /// vector. The vector is always truncated 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 truncated + /// 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 collect_into_vec(self, target: &mut Vec<Self::Item>) { + collect::collect_into_vec(self, target); + } + + /// Unzips the results of the iterator into the specified + /// vectors. The vectors are always truncated 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 truncated + /// 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 unzip_into_vecs<A, B>(self, left: &mut Vec<A>, right: &mut Vec<B>) + where + Self: IndexedParallelIterator<Item = (A, B)>, + A: Send, + B: Send, + { + collect::unzip_into_vecs(self, left, right); + } + + /// 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<Z>(self, zip_op: Z) -> Zip<Self, Z::Iter> + where + Z: IntoParallelIterator, + Z::Iter: IndexedParallelIterator, + { + Zip::new(self, zip_op.into_par_iter()) + } + + /// 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. + /// + /// ```should_panic + /// 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()); + /// ``` + #[track_caller] + fn zip_eq<Z>(self, zip_op: Z) -> ZipEq<Self, Z::Iter> + where + Z: IntoParallelIterator, + Z::Iter: IndexedParallelIterator, + { + let zip_op_iter = zip_op.into_par_iter(); + assert_eq!( + self.len(), + zip_op_iter.len(), + "iterators must have the same length" + ); + ZipEq::new(self, zip_op_iter) + } + + /// 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<I>(self, other: I) -> Interleave<Self, I::Iter> + where + I: IntoParallelIterator<Item = Self::Item>, + I::Iter: IndexedParallelIterator<Item = Self::Item>, + { + Interleave::new(self, other.into_par_iter()) + } + + /// 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 interleave_shortest<I>(self, other: I) -> InterleaveShortest<Self, I::Iter> + where + I: IntoParallelIterator<Item = Self::Item>, + I::Iter: IndexedParallelIterator<Item = Self::Item>, + { + InterleaveShortest::new(self, other.into_par_iter()) + } + + /// Splits an iterator up into fixed-size chunks. + /// + /// Returns an iterator that returns `Vec`s 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 `Vec`s for the chunks. + /// + /// [`par_chunks()`]: ../slice/trait.ParallelSlice.html#method.par_chunks + /// [`par_chunks_mut()`]: ../slice/trait.ParallelSliceMut.html#method.par_chunks_mut + /// + /// # 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 chunks(self, chunk_size: usize) -> Chunks<Self> { + assert!(chunk_size != 0, "chunk_size must not be zero"); + Chunks::new(self, chunk_size) + } + + /// 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: + /// + /// ```text + /// iter.chunks(chunk_size) + /// .map(|chunk| + /// chunk.into_iter() + /// .fold(identity, fold_op) + /// ) + /// ``` + /// + /// except there is no per-chunk allocation overhead. + /// + /// [`fold()`]: std::iter::Iterator#method.fold + /// + /// **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]); + /// ``` + #[track_caller] + fn fold_chunks<T, ID, F>( + self, + chunk_size: usize, + identity: ID, + fold_op: F, + ) -> FoldChunks<Self, ID, F> + where + ID: Fn() -> T + Send + Sync, + F: Fn(T, Self::Item) -> T + Send + Sync, + T: Send, + { + assert!(chunk_size != 0, "chunk_size must not be zero"); + FoldChunks::new(self, chunk_size, identity, fold_op) + } + + /// 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. + /// + /// [`fold()`]: std::iter::Iterator#method.fold + /// + /// **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]); + /// ``` + #[track_caller] + fn fold_chunks_with<T, F>( + self, + chunk_size: usize, + init: T, + fold_op: F, + ) -> FoldChunksWith<Self, T, F> + where + T: Send + Clone, + F: Fn(T, Self::Item) -> T + Send + Sync, + { + assert!(chunk_size != 0, "chunk_size must not be zero"); + FoldChunksWith::new(self, chunk_size, init, fold_op) + } + + /// 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 cmp<I>(self, other: I) -> Ordering + where + I: IntoParallelIterator<Item = Self::Item>, + I::Iter: IndexedParallelIterator, + Self::Item: Ord, + { + #[inline] + fn ordering<T: Ord>((x, y): (T, T)) -> Ordering { + Ord::cmp(&x, &y) + } + + #[inline] + fn inequal(&ord: &Ordering) -> bool { + ord != Ordering::Equal + } + + let other = other.into_par_iter(); + let ord_len = self.len().cmp(&other.len()); + self.zip(other) + .map(ordering) + .find_first(inequal) + .unwrap_or(ord_len) + } + + /// 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 partial_cmp<I>(self, other: I) -> Option<Ordering> + where + I: IntoParallelIterator, + I::Iter: IndexedParallelIterator, + Self::Item: PartialOrd<I::Item>, + { + #[inline] + fn ordering<T: PartialOrd<U>, U>((x, y): (T, U)) -> Option<Ordering> { + PartialOrd::partial_cmp(&x, &y) + } + + #[inline] + fn inequal(&ord: &Option<Ordering>) -> bool { + ord != Some(Ordering::Equal) + } + + let other = other.into_par_iter(); + let ord_len = self.len().cmp(&other.len()); + self.zip(other) + .map(ordering) + .find_first(inequal) + .unwrap_or(Some(ord_len)) + } + + /// Determines if the elements of this `ParallelIterator` + /// are equal to those of another + fn eq<I>(self, other: I) -> bool + where + I: IntoParallelIterator, + I::Iter: IndexedParallelIterator, + Self::Item: PartialEq<I::Item>, + { + #[inline] + fn eq<T: PartialEq<U>, U>((x, y): (T, U)) -> bool { + PartialEq::eq(&x, &y) + } + + let other = other.into_par_iter(); + self.len() == other.len() && self.zip(other).all(eq) + } + + /// Determines if the elements of this `ParallelIterator` + /// are unequal to those of another + fn ne<I>(self, other: I) -> bool + where + I: IntoParallelIterator, + I::Iter: IndexedParallelIterator, + Self::Item: PartialEq<I::Item>, + { + !self.eq(other) + } + + /// Determines if the elements of this `ParallelIterator` + /// are lexicographically less than those of another. + fn lt<I>(self, other: I) -> bool + where + I: IntoParallelIterator, + I::Iter: IndexedParallelIterator, + Self::Item: PartialOrd<I::Item>, + { + self.partial_cmp(other) == Some(Ordering::Less) + } + + /// Determines if the elements of this `ParallelIterator` + /// are less or equal to those of another. + fn le<I>(self, other: I) -> bool + where + I: IntoParallelIterator, + I::Iter: IndexedParallelIterator, + Self::Item: PartialOrd<I::Item>, + { + let ord = self.partial_cmp(other); + ord == Some(Ordering::Equal) || ord == Some(Ordering::Less) + } + + /// Determines if the elements of this `ParallelIterator` + /// are lexicographically greater than those of another. + fn gt<I>(self, other: I) -> bool + where + I: IntoParallelIterator, + I::Iter: IndexedParallelIterator, + Self::Item: PartialOrd<I::Item>, + { + self.partial_cmp(other) == Some(Ordering::Greater) + } + + /// Determines if the elements of this `ParallelIterator` + /// are less or equal to those of another. + fn ge<I>(self, other: I) -> bool + where + I: IntoParallelIterator, + I::Iter: IndexedParallelIterator, + Self::Item: PartialOrd<I::Item>, + { + let ord = self.partial_cmp(other); + ord == Some(Ordering::Equal) || ord == Some(Ordering::Greater) + } + + /// 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 enumerate(self) -> Enumerate<Self> { + Enumerate::new(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 step_by(self, step: usize) -> StepBy<Self> { + StepBy::new(self, step) + } + + /// 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 skip(self, n: usize) -> Skip<Self> { + Skip::new(self, n) + } + + /// 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 take(self, n: usize) -> Take<Self> { + Take::new(self, n) + } + + /// 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_any<P>(self, predicate: P) -> Option<usize> + where + P: Fn(Self::Item) -> bool + Sync + Send, + { + #[inline] + fn check(&(_, p): &(usize, bool)) -> bool { + p + } + + let (i, _) = self.map(predicate).enumerate().find_any(check)?; + Some(i) + } + + /// 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_first<P>(self, predicate: P) -> Option<usize> + where + P: Fn(Self::Item) -> bool + Sync + Send, + { + #[inline] + fn check(&(_, p): &(usize, bool)) -> bool { + p + } + + let (i, _) = self.map(predicate).enumerate().find_first(check)?; + Some(i) + } + + /// 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 position_last<P>(self, predicate: P) -> Option<usize> + where + P: Fn(Self::Item) -> bool + Sync + Send, + { + #[inline] + fn check(&(_, p): &(usize, bool)) -> bool { + p + } + + let (i, _) = self.map(predicate).enumerate().find_last(check)?; + Some(i) + } + + #[doc(hidden)] + #[deprecated( + note = "parallel `position` does not search in order -- use `position_any`, \\ + `position_first`, or `position_last`" + )] + fn position<P>(self, predicate: P) -> Option<usize> + where + P: Fn(Self::Item) -> bool + Sync + Send, + { + self.position_any(predicate) + } + + /// 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 positions<P>(self, predicate: P) -> Positions<Self, P> + where + P: Fn(Self::Item) -> bool + Sync + Send, + { + Positions::new(self, predicate) + } + + /// 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 rev(self) -> Rev<Self> { + Rev::new(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_min_len(self, min: usize) -> MinLen<Self> { + MinLen::new(self, min) + } + + /// 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); + /// ``` + fn with_max_len(self, max: usize) -> MaxLen<Self> { + MaxLen::new(self, max) + } + + /// Produces an exact count of how many items this iterator will + /// produce, presuming no panic occurs. + /// + /// # Examples + /// + /// ``` + /// use rayon::prelude::*; + /// + /// let par_iter = (0..100).into_par_iter().zip(vec![0; 10]); + /// assert_eq!(par_iter.len(), 10); + /// + /// let vec: Vec<_> = par_iter.collect(); + /// assert_eq!(vec.len(), 10); + /// ``` + fn len(&self) -> usize; + + /// Internal method used to define the behavior of this parallel + /// iterator. You should not need to call this directly. + /// + /// This method causes the iterator `self` to start producing + /// items and to feed them to the consumer `consumer` one by one. + /// It may split the consumer before doing so to create the + /// opportunity to produce in parallel. If a split does happen, it + /// will inform the consumer of the index where the split should + /// occur (unlike `ParallelIterator::drive_unindexed()`). + /// + /// See the [README] for more details on the internals of parallel + /// iterators. + /// + /// [README]: https://github.com/rayon-rs/rayon/blob/master/src/iter/plumbing/README.md + fn drive<C: Consumer<Self::Item>>(self, consumer: C) -> C::Result; + + /// Internal method used to define the behavior of this parallel + /// iterator. You should not need to call this directly. + /// + /// This method converts the iterator into a producer P and then + /// invokes `callback.callback()` with P. Note that the type of + /// this producer is not defined as part of the API, since + /// `callback` must be defined generically for all producers. This + /// allows the producer type to contain references; it also means + /// that parallel iterators can adjust that type without causing a + /// breaking change. + /// + /// See the [README] for more details on the internals of parallel + /// iterators. + /// + /// [README]: https://github.com/rayon-rs/rayon/blob/master/src/iter/plumbing/README.md + fn with_producer<CB: ProducerCallback<Self::Item>>(self, callback: CB) -> CB::Output; +} + +/// `FromParallelIterator` implements the creation of a collection +/// from a [`ParallelIterator`]. By implementing +/// `FromParallelIterator` for a given type, you define how it will be +/// created from an iterator. +/// +/// `FromParallelIterator` is used through [`ParallelIterator`]'s [`collect()`] method. +/// +/// [`ParallelIterator`]: trait.ParallelIterator.html +/// [`collect()`]: trait.ParallelIterator.html#method.collect +/// +/// # Examples +/// +/// Implementing `FromParallelIterator` for your type: +/// +/// ``` +/// use rayon::prelude::*; +/// use std::mem; +/// +/// struct BlackHole { +/// mass: usize, +/// } +/// +/// impl<T: Send> FromParallelIterator<T> for BlackHole { +/// fn from_par_iter<I>(par_iter: I) -> Self +/// where I: IntoParallelIterator<Item = T> +/// { +/// let par_iter = par_iter.into_par_iter(); +/// BlackHole { +/// mass: par_iter.count() * mem::size_of::<T>(), +/// } +/// } +/// } +/// +/// let bh: BlackHole = (0i32..1000).into_par_iter().collect(); +/// assert_eq!(bh.mass, 4000); +/// ``` +pub trait FromParallelIterator<T> +where + T: Send, +{ + /// Creates an instance of the collection from the parallel iterator `par_iter`. + /// + /// If your collection is not naturally parallel, the easiest (and + /// fastest) way to do this is often to collect `par_iter` into a + /// [`LinkedList`] or other intermediate data structure and then + /// sequentially extend your collection. However, a more 'native' + /// technique is to use the [`par_iter.fold`] or + /// [`par_iter.fold_with`] methods to create the collection. + /// Alternatively, if your collection is 'natively' parallel, you + /// can use `par_iter.for_each` to process each element in turn. + /// + /// [`LinkedList`]: https://doc.rust-lang.org/std/collections/struct.LinkedList.html + /// [`par_iter.fold`]: trait.ParallelIterator.html#method.fold + /// [`par_iter.fold_with`]: trait.ParallelIterator.html#method.fold_with + /// [`par_iter.for_each`]: trait.ParallelIterator.html#method.for_each + fn from_par_iter<I>(par_iter: I) -> Self + where + I: IntoParallelIterator<Item = T>; +} + +/// `ParallelExtend` extends an existing collection with items from a [`ParallelIterator`]. +/// +/// [`ParallelIterator`]: trait.ParallelIterator.html +/// +/// # Examples +/// +/// Implementing `ParallelExtend` for your type: +/// +/// ``` +/// use rayon::prelude::*; +/// use std::mem; +/// +/// struct BlackHole { +/// mass: usize, +/// } +/// +/// impl<T: Send> ParallelExtend<T> for BlackHole { +/// fn par_extend<I>(&mut self, par_iter: I) +/// where I: IntoParallelIterator<Item = T> +/// { +/// let par_iter = par_iter.into_par_iter(); +/// self.mass += par_iter.count() * mem::size_of::<T>(); +/// } +/// } +/// +/// let mut bh = BlackHole { mass: 0 }; +/// bh.par_extend(0i32..1000); +/// assert_eq!(bh.mass, 4000); +/// bh.par_extend(0i64..10); +/// assert_eq!(bh.mass, 4080); +/// ``` +pub trait ParallelExtend<T> +where + T: Send, +{ + /// Extends an instance of the collection with the elements drawn + /// from the parallel iterator `par_iter`. + /// + /// # Examples + /// + /// ``` + /// use rayon::prelude::*; + /// + /// let mut vec = vec![]; + /// vec.par_extend(0..5); + /// vec.par_extend((0..5).into_par_iter().map(|i| i * i)); + /// assert_eq!(vec, [0, 1, 2, 3, 4, 0, 1, 4, 9, 16]); + /// ``` + fn par_extend<I>(&mut self, par_iter: I) + where + I: IntoParallelIterator<Item = T>; +} + +/// `ParallelDrainFull` creates a parallel iterator that moves all items +/// from a collection while retaining the original capacity. +/// +/// Types which are indexable typically implement [`ParallelDrainRange`] +/// instead, where you can drain fully with `par_drain(..)`. +/// +/// [`ParallelDrainRange`]: trait.ParallelDrainRange.html +pub trait ParallelDrainFull { + /// The draining parallel iterator type that will be created. + type Iter: ParallelIterator<Item = Self::Item>; + + /// The type of item that the parallel iterator will produce. + /// This is usually the same as `IntoParallelIterator::Item`. + type Item: Send; + + /// Returns a draining parallel iterator over an entire collection. + /// + /// When the iterator is dropped, all items are removed, even if the + /// iterator was not fully consumed. If the iterator is leaked, for example + /// using `std::mem::forget`, it is unspecified how many items are removed. + /// + /// # Examples + /// + /// ``` + /// use rayon::prelude::*; + /// use std::collections::{BinaryHeap, HashSet}; + /// + /// let squares: HashSet<i32> = (0..10).map(|x| x * x).collect(); + /// + /// let mut heap: BinaryHeap<_> = squares.iter().copied().collect(); + /// assert_eq!( + /// // heaps are drained in arbitrary order + /// heap.par_drain() + /// .inspect(|x| assert!(squares.contains(x))) + /// .count(), + /// squares.len(), + /// ); + /// assert!(heap.is_empty()); + /// assert!(heap.capacity() >= squares.len()); + /// ``` + fn par_drain(self) -> Self::Iter; +} + +/// `ParallelDrainRange` creates a parallel iterator that moves a range of items +/// from a collection while retaining the original capacity. +/// +/// Types which are not indexable may implement [`ParallelDrainFull`] instead. +/// +/// [`ParallelDrainFull`]: trait.ParallelDrainFull.html +pub trait ParallelDrainRange<Idx = usize> { + /// The draining parallel iterator type that will be created. + type Iter: ParallelIterator<Item = Self::Item>; + + /// The type of item that the parallel iterator will produce. + /// This is usually the same as `IntoParallelIterator::Item`. + type Item: Send; + + /// Returns a draining parallel iterator over a range of the collection. + /// + /// When the iterator is dropped, all items in the range are removed, even + /// if the iterator was not fully consumed. If the iterator is leaked, for + /// example using `std::mem::forget`, it is unspecified how many items are + /// removed. + /// + /// # Examples + /// + /// ``` + /// use rayon::prelude::*; + /// + /// let squares: Vec<i32> = (0..10).map(|x| x * x).collect(); + /// + /// println!("RangeFull"); + /// let mut vec = squares.clone(); + /// assert!(vec.par_drain(..) + /// .eq(squares.par_iter().copied())); + /// assert!(vec.is_empty()); + /// assert!(vec.capacity() >= squares.len()); + /// + /// println!("RangeFrom"); + /// let mut vec = squares.clone(); + /// assert!(vec.par_drain(5..) + /// .eq(squares[5..].par_iter().copied())); + /// assert_eq!(&vec[..], &squares[..5]); + /// assert!(vec.capacity() >= squares.len()); + /// + /// println!("RangeTo"); + /// let mut vec = squares.clone(); + /// assert!(vec.par_drain(..5) + /// .eq(squares[..5].par_iter().copied())); + /// assert_eq!(&vec[..], &squares[5..]); + /// assert!(vec.capacity() >= squares.len()); + /// + /// println!("RangeToInclusive"); + /// let mut vec = squares.clone(); + /// assert!(vec.par_drain(..=5) + /// .eq(squares[..=5].par_iter().copied())); + /// assert_eq!(&vec[..], &squares[6..]); + /// assert!(vec.capacity() >= squares.len()); + /// + /// println!("Range"); + /// let mut vec = squares.clone(); + /// assert!(vec.par_drain(3..7) + /// .eq(squares[3..7].par_iter().copied())); + /// assert_eq!(&vec[..3], &squares[..3]); + /// assert_eq!(&vec[3..], &squares[7..]); + /// assert!(vec.capacity() >= squares.len()); + /// + /// println!("RangeInclusive"); + /// let mut vec = squares.clone(); + /// assert!(vec.par_drain(3..=7) + /// .eq(squares[3..=7].par_iter().copied())); + /// assert_eq!(&vec[..3], &squares[..3]); + /// assert_eq!(&vec[3..], &squares[8..]); + /// assert!(vec.capacity() >= squares.len()); + /// ``` + fn par_drain<R: RangeBounds<Idx>>(self, range: R) -> Self::Iter; +} + +/// We hide the `Try` trait in a private module, as it's only meant to be a +/// stable clone of the standard library's `Try` trait, as yet unstable. +mod private { + use std::convert::Infallible; + use std::ops::ControlFlow::{self, Break, Continue}; + use std::task::Poll; + + /// Clone of `std::ops::Try`. + /// + /// Implementing this trait is not permitted outside of `rayon`. + pub trait Try { + private_decl! {} + + type Output; + type Residual; + + fn from_output(output: Self::Output) -> Self; + + fn from_residual(residual: Self::Residual) -> Self; + + fn branch(self) -> ControlFlow<Self::Residual, Self::Output>; + } + + impl<B, C> Try for ControlFlow<B, C> { + private_impl! {} + + type Output = C; + type Residual = ControlFlow<B, Infallible>; + + fn from_output(output: Self::Output) -> Self { + Continue(output) + } + + fn from_residual(residual: Self::Residual) -> Self { + match residual { + Break(b) => Break(b), + Continue(_) => unreachable!(), + } + } + + fn branch(self) -> ControlFlow<Self::Residual, Self::Output> { + match self { + Continue(c) => Continue(c), + Break(b) => Break(Break(b)), + } + } + } + + impl<T> Try for Option<T> { + private_impl! {} + + type Output = T; + type Residual = Option<Infallible>; + + fn from_output(output: Self::Output) -> Self { + Some(output) + } + + fn from_residual(residual: Self::Residual) -> Self { + match residual { + None => None, + Some(_) => unreachable!(), + } + } + + fn branch(self) -> ControlFlow<Self::Residual, Self::Output> { + match self { + Some(c) => Continue(c), + None => Break(None), + } + } + } + + impl<T, E> Try for Result<T, E> { + private_impl! {} + + type Output = T; + type Residual = Result<Infallible, E>; + + fn from_output(output: Self::Output) -> Self { + Ok(output) + } + + fn from_residual(residual: Self::Residual) -> Self { + match residual { + Err(e) => Err(e), + Ok(_) => unreachable!(), + } + } + + fn branch(self) -> ControlFlow<Self::Residual, Self::Output> { + match self { + Ok(c) => Continue(c), + Err(e) => Break(Err(e)), + } + } + } + + impl<T, E> Try for Poll<Result<T, E>> { + private_impl! {} + + type Output = Poll<T>; + type Residual = Result<Infallible, E>; + + fn from_output(output: Self::Output) -> Self { + output.map(Ok) + } + + fn from_residual(residual: Self::Residual) -> Self { + match residual { + Err(e) => Poll::Ready(Err(e)), + Ok(_) => unreachable!(), + } + } + + fn branch(self) -> ControlFlow<Self::Residual, Self::Output> { + match self { + Poll::Pending => Continue(Poll::Pending), + Poll::Ready(Ok(c)) => Continue(Poll::Ready(c)), + Poll::Ready(Err(e)) => Break(Err(e)), + } + } + } + + impl<T, E> Try for Poll<Option<Result<T, E>>> { + private_impl! {} + + type Output = Poll<Option<T>>; + type Residual = Result<Infallible, E>; + + fn from_output(output: Self::Output) -> Self { + match output { + Poll::Ready(o) => Poll::Ready(o.map(Ok)), + Poll::Pending => Poll::Pending, + } + } + + fn from_residual(residual: Self::Residual) -> Self { + match residual { + Err(e) => Poll::Ready(Some(Err(e))), + Ok(_) => unreachable!(), + } + } + + fn branch(self) -> ControlFlow<Self::Residual, Self::Output> { + match self { + Poll::Pending => Continue(Poll::Pending), + Poll::Ready(None) => Continue(Poll::Ready(None)), + Poll::Ready(Some(Ok(c))) => Continue(Poll::Ready(Some(c))), + Poll::Ready(Some(Err(e))) => Break(Err(e)), + } + } + } +} diff --git a/third_party/rust/rayon/src/iter/multizip.rs b/third_party/rust/rayon/src/iter/multizip.rs new file mode 100644 index 0000000000..8e36d08a29 --- /dev/null +++ b/third_party/rust/rayon/src/iter/multizip.rs @@ -0,0 +1,338 @@ +use super::plumbing::*; +use super::*; + +/// `MultiZip` is an iterator that zips up a tuple of parallel iterators to +/// produce tuples of their items. +/// +/// It is created by calling `into_par_iter()` on a tuple of types that +/// implement `IntoParallelIterator`, or `par_iter()`/`par_iter_mut()` with +/// types that are iterable by reference. +/// +/// The implementation currently support tuples up to length 12. +/// +/// # Examples +/// +/// ``` +/// use rayon::prelude::*; +/// +/// // This will iterate `r` by mutable reference, like `par_iter_mut()`, while +/// // ranges are all iterated by value like `into_par_iter()`. +/// // Note that the zipped iterator is only as long as the shortest input. +/// let mut r = vec![0; 3]; +/// (&mut r, 1..10, 10..100, 100..1000).into_par_iter() +/// .for_each(|(r, x, y, z)| *r = x * y + z); +/// +/// assert_eq!(&r, &[1 * 10 + 100, 2 * 11 + 101, 3 * 12 + 102]); +/// ``` +/// +/// For a group that should all be iterated by reference, you can use a tuple reference. +/// +/// ``` +/// use rayon::prelude::*; +/// +/// let xs: Vec<_> = (1..10).collect(); +/// let ys: Vec<_> = (10..100).collect(); +/// let zs: Vec<_> = (100..1000).collect(); +/// +/// // Reference each input separately with `IntoParallelIterator`: +/// let r1: Vec<_> = (&xs, &ys, &zs).into_par_iter() +/// .map(|(x, y, z)| x * y + z) +/// .collect(); +/// +/// // Reference them all together with `IntoParallelRefIterator`: +/// let r2: Vec<_> = (xs, ys, zs).par_iter() +/// .map(|(x, y, z)| x * y + z) +/// .collect(); +/// +/// assert_eq!(r1, r2); +/// ``` +/// +/// Mutable references to a tuple will work similarly. +/// +/// ``` +/// use rayon::prelude::*; +/// +/// let mut xs: Vec<_> = (1..4).collect(); +/// let mut ys: Vec<_> = (-4..-1).collect(); +/// let mut zs = vec![0; 3]; +/// +/// // Mutably reference each input separately with `IntoParallelIterator`: +/// (&mut xs, &mut ys, &mut zs).into_par_iter().for_each(|(x, y, z)| { +/// *z += *x + *y; +/// std::mem::swap(x, y); +/// }); +/// +/// assert_eq!(xs, (vec![-4, -3, -2])); +/// assert_eq!(ys, (vec![1, 2, 3])); +/// assert_eq!(zs, (vec![-3, -1, 1])); +/// +/// // Mutably reference them all together with `IntoParallelRefMutIterator`: +/// let mut tuple = (xs, ys, zs); +/// tuple.par_iter_mut().for_each(|(x, y, z)| { +/// *z += *x + *y; +/// std::mem::swap(x, y); +/// }); +/// +/// assert_eq!(tuple, (vec![1, 2, 3], vec![-4, -3, -2], vec![-6, -2, 2])); +/// ``` +#[derive(Debug, Clone)] +pub struct MultiZip<T> { + tuple: T, +} + +// These macros greedily consume 4 or 2 items first to achieve log2 nesting depth. +// For example, 5 => 4,1 => (2,2),1. +// +// The tuples go up to 12, so we might want to greedily consume 8 too, but +// the depth works out the same if we let that expand on the right: +// 9 => 4,5 => (2,2),(4,1) => (2,2),((2,2),1) +// 12 => 4,8 => (2,2),(4,4) => (2,2),((2,2),(2,2)) +// +// But if we ever increase to 13, we would want to split 8,5 rather than 4,9. + +macro_rules! reduce { + ($a:expr, $b:expr, $c:expr, $d:expr, $( $x:expr ),+ => $fn:path) => { + reduce!(reduce!($a, $b, $c, $d => $fn), + reduce!($( $x ),+ => $fn) + => $fn) + }; + ($a:expr, $b:expr, $( $x:expr ),+ => $fn:path) => { + reduce!(reduce!($a, $b => $fn), + reduce!($( $x ),+ => $fn) + => $fn) + }; + ($a:expr, $b:expr => $fn:path) => { $fn($a, $b) }; + ($a:expr => $fn:path) => { $a }; +} + +macro_rules! nest { + ($A:tt, $B:tt, $C:tt, $D:tt, $( $X:tt ),+) => { + (nest!($A, $B, $C, $D), nest!($( $X ),+)) + }; + ($A:tt, $B:tt, $( $X:tt ),+) => { + (($A, $B), nest!($( $X ),+)) + }; + ($A:tt, $B:tt) => { ($A, $B) }; + ($A:tt) => { $A }; +} + +macro_rules! flatten { + ($( $T:ident ),+) => {{ + #[allow(non_snake_case)] + fn flatten<$( $T ),+>(nest!($( $T ),+) : nest!($( $T ),+)) -> ($( $T, )+) { + ($( $T, )+) + } + flatten + }}; +} + +macro_rules! multizip_impls { + ($( + $Tuple:ident { + $(($idx:tt) -> $T:ident)+ + } + )+) => { + $( + impl<$( $T, )+> IntoParallelIterator for ($( $T, )+) + where + $( + $T: IntoParallelIterator, + $T::Iter: IndexedParallelIterator, + )+ + { + type Item = ($( $T::Item, )+); + type Iter = MultiZip<($( $T::Iter, )+)>; + + fn into_par_iter(self) -> Self::Iter { + MultiZip { + tuple: ( $( self.$idx.into_par_iter(), )+ ), + } + } + } + + impl<'a, $( $T, )+> IntoParallelIterator for &'a ($( $T, )+) + where + $( + $T: IntoParallelRefIterator<'a>, + $T::Iter: IndexedParallelIterator, + )+ + { + type Item = ($( $T::Item, )+); + type Iter = MultiZip<($( $T::Iter, )+)>; + + fn into_par_iter(self) -> Self::Iter { + MultiZip { + tuple: ( $( self.$idx.par_iter(), )+ ), + } + } + } + + impl<'a, $( $T, )+> IntoParallelIterator for &'a mut ($( $T, )+) + where + $( + $T: IntoParallelRefMutIterator<'a>, + $T::Iter: IndexedParallelIterator, + )+ + { + type Item = ($( $T::Item, )+); + type Iter = MultiZip<($( $T::Iter, )+)>; + + fn into_par_iter(self) -> Self::Iter { + MultiZip { + tuple: ( $( self.$idx.par_iter_mut(), )+ ), + } + } + } + + impl<$( $T, )+> ParallelIterator for MultiZip<($( $T, )+)> + where + $( $T: IndexedParallelIterator, )+ + { + type Item = ($( $T::Item, )+); + + fn drive_unindexed<CONSUMER>(self, consumer: CONSUMER) -> CONSUMER::Result + where + CONSUMER: UnindexedConsumer<Self::Item>, + { + self.drive(consumer) + } + + fn opt_len(&self) -> Option<usize> { + Some(self.len()) + } + } + + impl<$( $T, )+> IndexedParallelIterator for MultiZip<($( $T, )+)> + where + $( $T: IndexedParallelIterator, )+ + { + fn drive<CONSUMER>(self, consumer: CONSUMER) -> CONSUMER::Result + where + CONSUMER: Consumer<Self::Item>, + { + reduce!($( self.tuple.$idx ),+ => IndexedParallelIterator::zip) + .map(flatten!($( $T ),+)) + .drive(consumer) + } + + fn len(&self) -> usize { + reduce!($( self.tuple.$idx.len() ),+ => Ord::min) + } + + fn with_producer<CB>(self, callback: CB) -> CB::Output + where + CB: ProducerCallback<Self::Item>, + { + reduce!($( self.tuple.$idx ),+ => IndexedParallelIterator::zip) + .map(flatten!($( $T ),+)) + .with_producer(callback) + } + } + )+ + } +} + +multizip_impls! { + Tuple1 { + (0) -> A + } + Tuple2 { + (0) -> A + (1) -> B + } + Tuple3 { + (0) -> A + (1) -> B + (2) -> C + } + Tuple4 { + (0) -> A + (1) -> B + (2) -> C + (3) -> D + } + Tuple5 { + (0) -> A + (1) -> B + (2) -> C + (3) -> D + (4) -> E + } + Tuple6 { + (0) -> A + (1) -> B + (2) -> C + (3) -> D + (4) -> E + (5) -> F + } + Tuple7 { + (0) -> A + (1) -> B + (2) -> C + (3) -> D + (4) -> E + (5) -> F + (6) -> G + } + Tuple8 { + (0) -> A + (1) -> B + (2) -> C + (3) -> D + (4) -> E + (5) -> F + (6) -> G + (7) -> H + } + Tuple9 { + (0) -> A + (1) -> B + (2) -> C + (3) -> D + (4) -> E + (5) -> F + (6) -> G + (7) -> H + (8) -> I + } + Tuple10 { + (0) -> A + (1) -> B + (2) -> C + (3) -> D + (4) -> E + (5) -> F + (6) -> G + (7) -> H + (8) -> I + (9) -> J + } + Tuple11 { + (0) -> A + (1) -> B + (2) -> C + (3) -> D + (4) -> E + (5) -> F + (6) -> G + (7) -> H + (8) -> I + (9) -> J + (10) -> K + } + Tuple12 { + (0) -> A + (1) -> B + (2) -> C + (3) -> D + (4) -> E + (5) -> F + (6) -> G + (7) -> H + (8) -> I + (9) -> J + (10) -> K + (11) -> L + } +} diff --git a/third_party/rust/rayon/src/iter/noop.rs b/third_party/rust/rayon/src/iter/noop.rs new file mode 100644 index 0000000000..1e55ecb20e --- /dev/null +++ b/third_party/rust/rayon/src/iter/noop.rs @@ -0,0 +1,59 @@ +use super::plumbing::*; + +pub(super) struct NoopConsumer; + +impl<T> Consumer<T> for NoopConsumer { + type Folder = NoopConsumer; + type Reducer = NoopReducer; + type Result = (); + + fn split_at(self, _index: usize) -> (Self, Self, NoopReducer) { + (NoopConsumer, NoopConsumer, NoopReducer) + } + + fn into_folder(self) -> Self { + self + } + + fn full(&self) -> bool { + false + } +} + +impl<T> Folder<T> for NoopConsumer { + type Result = (); + + fn consume(self, _item: T) -> Self { + self + } + + fn consume_iter<I>(self, iter: I) -> Self + where + I: IntoIterator<Item = T>, + { + iter.into_iter().for_each(drop); + self + } + + fn complete(self) {} + + fn full(&self) -> bool { + false + } +} + +impl<T> UnindexedConsumer<T> for NoopConsumer { + fn split_off_left(&self) -> Self { + NoopConsumer + } + + fn to_reducer(&self) -> NoopReducer { + NoopReducer + } +} + +pub(super) struct NoopReducer; + +impl Reducer<()> for NoopReducer { + fn reduce(self, _left: (), _right: ()) {} +} diff --git a/third_party/rust/rayon/src/iter/once.rs b/third_party/rust/rayon/src/iter/once.rs new file mode 100644 index 0000000000..5140b6b9f8 --- /dev/null +++ b/third_party/rust/rayon/src/iter/once.rs @@ -0,0 +1,68 @@ +use crate::iter::plumbing::*; +use crate::iter::*; + +/// Creates a parallel iterator that produces an element exactly once. +/// +/// This admits no parallelism on its own, but it could be chained to existing +/// parallel iterators to extend their contents, or otherwise used for any code +/// that deals with generic parallel iterators. +/// +/// # Examples +/// +/// ``` +/// use rayon::prelude::*; +/// use rayon::iter::once; +/// +/// let pi = (0..1234).into_par_iter() +/// .chain(once(-1)) +/// .chain(1234..10_000); +/// +/// assert_eq!(pi.clone().count(), 10_001); +/// assert_eq!(pi.clone().filter(|&x| x < 0).count(), 1); +/// assert_eq!(pi.position_any(|x| x < 0), Some(1234)); +/// ``` +pub fn once<T: Send>(item: T) -> Once<T> { + Once { item } +} + +/// Iterator adaptor for [the `once()` function](fn.once.html). +#[derive(Clone, Debug)] +pub struct Once<T: Send> { + item: T, +} + +impl<T: Send> ParallelIterator for Once<T> { + type Item = T; + + fn drive_unindexed<C>(self, consumer: C) -> C::Result + where + C: UnindexedConsumer<Self::Item>, + { + self.drive(consumer) + } + + fn opt_len(&self) -> Option<usize> { + Some(1) + } +} + +impl<T: Send> IndexedParallelIterator for Once<T> { + fn drive<C>(self, consumer: C) -> C::Result + where + C: Consumer<Self::Item>, + { + consumer.into_folder().consume(self.item).complete() + } + + fn len(&self) -> usize { + 1 + } + + fn with_producer<CB>(self, callback: CB) -> CB::Output + where + CB: ProducerCallback<Self::Item>, + { + // Let `OptionProducer` handle it. + Some(self.item).into_par_iter().with_producer(callback) + } +} diff --git a/third_party/rust/rayon/src/iter/panic_fuse.rs b/third_party/rust/rayon/src/iter/panic_fuse.rs new file mode 100644 index 0000000000..7487230e57 --- /dev/null +++ b/third_party/rust/rayon/src/iter/panic_fuse.rs @@ -0,0 +1,342 @@ +use super::plumbing::*; +use super::*; +use std::sync::atomic::{AtomicBool, Ordering}; +use std::thread; + +/// `PanicFuse` is an adaptor that wraps an iterator with a fuse in case +/// of panics, to halt all threads as soon as possible. +/// +/// This struct is created by the [`panic_fuse()`] method on [`ParallelIterator`] +/// +/// [`panic_fuse()`]: trait.ParallelIterator.html#method.panic_fuse +/// [`ParallelIterator`]: trait.ParallelIterator.html +#[must_use = "iterator adaptors are lazy and do nothing unless consumed"] +#[derive(Debug, Clone)] +pub struct PanicFuse<I: ParallelIterator> { + base: I, +} + +/// Helper that sets a bool to `true` if dropped while unwinding. +#[derive(Clone)] +struct Fuse<'a>(&'a AtomicBool); + +impl<'a> Drop for Fuse<'a> { + #[inline] + fn drop(&mut self) { + if thread::panicking() { + self.0.store(true, Ordering::Relaxed); + } + } +} + +impl<'a> Fuse<'a> { + #[inline] + fn panicked(&self) -> bool { + self.0.load(Ordering::Relaxed) + } +} + +impl<I> PanicFuse<I> +where + I: ParallelIterator, +{ + /// Creates a new `PanicFuse` iterator. + pub(super) fn new(base: I) -> PanicFuse<I> { + PanicFuse { base } + } +} + +impl<I> ParallelIterator for PanicFuse<I> +where + I: ParallelIterator, +{ + type Item = I::Item; + + fn drive_unindexed<C>(self, consumer: C) -> C::Result + where + C: UnindexedConsumer<Self::Item>, + { + let panicked = AtomicBool::new(false); + let consumer1 = PanicFuseConsumer { + base: consumer, + fuse: Fuse(&panicked), + }; + self.base.drive_unindexed(consumer1) + } + + fn opt_len(&self) -> Option<usize> { + self.base.opt_len() + } +} + +impl<I> IndexedParallelIterator for PanicFuse<I> +where + I: IndexedParallelIterator, +{ + fn drive<C>(self, consumer: C) -> C::Result + where + C: Consumer<Self::Item>, + { + let panicked = AtomicBool::new(false); + let consumer1 = PanicFuseConsumer { + base: consumer, + fuse: Fuse(&panicked), + }; + self.base.drive(consumer1) + } + + fn len(&self) -> usize { + self.base.len() + } + + fn with_producer<CB>(self, callback: CB) -> CB::Output + where + CB: ProducerCallback<Self::Item>, + { + return self.base.with_producer(Callback { callback }); + + struct Callback<CB> { + callback: CB, + } + + impl<T, CB> ProducerCallback<T> for Callback<CB> + where + CB: ProducerCallback<T>, + { + type Output = CB::Output; + + fn callback<P>(self, base: P) -> CB::Output + where + P: Producer<Item = T>, + { + let panicked = AtomicBool::new(false); + let producer = PanicFuseProducer { + base, + fuse: Fuse(&panicked), + }; + self.callback.callback(producer) + } + } + } +} + +/// //////////////////////////////////////////////////////////////////////// +/// Producer implementation + +struct PanicFuseProducer<'a, P> { + base: P, + fuse: Fuse<'a>, +} + +impl<'a, P> Producer for PanicFuseProducer<'a, P> +where + P: Producer, +{ + type Item = P::Item; + type IntoIter = PanicFuseIter<'a, P::IntoIter>; + + fn into_iter(self) -> Self::IntoIter { + PanicFuseIter { + base: self.base.into_iter(), + fuse: self.fuse, + } + } + + fn min_len(&self) -> usize { + self.base.min_len() + } + fn max_len(&self) -> usize { + self.base.max_len() + } + + fn split_at(self, index: usize) -> (Self, Self) { + let (left, right) = self.base.split_at(index); + ( + PanicFuseProducer { + base: left, + fuse: self.fuse.clone(), + }, + PanicFuseProducer { + base: right, + fuse: self.fuse, + }, + ) + } + + fn fold_with<G>(self, folder: G) -> G + where + G: Folder<Self::Item>, + { + let folder1 = PanicFuseFolder { + base: folder, + fuse: self.fuse, + }; + self.base.fold_with(folder1).base + } +} + +struct PanicFuseIter<'a, I> { + base: I, + fuse: Fuse<'a>, +} + +impl<'a, I> Iterator for PanicFuseIter<'a, I> +where + I: Iterator, +{ + type Item = I::Item; + + fn next(&mut self) -> Option<Self::Item> { + if self.fuse.panicked() { + None + } else { + self.base.next() + } + } + + fn size_hint(&self) -> (usize, Option<usize>) { + self.base.size_hint() + } +} + +impl<'a, I> DoubleEndedIterator for PanicFuseIter<'a, I> +where + I: DoubleEndedIterator, +{ + fn next_back(&mut self) -> Option<Self::Item> { + if self.fuse.panicked() { + None + } else { + self.base.next_back() + } + } +} + +impl<'a, I> ExactSizeIterator for PanicFuseIter<'a, I> +where + I: ExactSizeIterator, +{ + fn len(&self) -> usize { + self.base.len() + } +} + +/// //////////////////////////////////////////////////////////////////////// +/// Consumer implementation + +struct PanicFuseConsumer<'a, C> { + base: C, + fuse: Fuse<'a>, +} + +impl<'a, T, C> Consumer<T> for PanicFuseConsumer<'a, C> +where + C: Consumer<T>, +{ + type Folder = PanicFuseFolder<'a, C::Folder>; + type Reducer = PanicFuseReducer<'a, C::Reducer>; + type Result = C::Result; + + fn split_at(self, index: usize) -> (Self, Self, Self::Reducer) { + let (left, right, reducer) = self.base.split_at(index); + ( + PanicFuseConsumer { + base: left, + fuse: self.fuse.clone(), + }, + PanicFuseConsumer { + base: right, + fuse: self.fuse.clone(), + }, + PanicFuseReducer { + base: reducer, + _fuse: self.fuse, + }, + ) + } + + fn into_folder(self) -> Self::Folder { + PanicFuseFolder { + base: self.base.into_folder(), + fuse: self.fuse, + } + } + + fn full(&self) -> bool { + self.fuse.panicked() || self.base.full() + } +} + +impl<'a, T, C> UnindexedConsumer<T> for PanicFuseConsumer<'a, C> +where + C: UnindexedConsumer<T>, +{ + fn split_off_left(&self) -> Self { + PanicFuseConsumer { + base: self.base.split_off_left(), + fuse: self.fuse.clone(), + } + } + + fn to_reducer(&self) -> Self::Reducer { + PanicFuseReducer { + base: self.base.to_reducer(), + _fuse: self.fuse.clone(), + } + } +} + +struct PanicFuseFolder<'a, C> { + base: C, + fuse: Fuse<'a>, +} + +impl<'a, T, C> Folder<T> for PanicFuseFolder<'a, C> +where + C: Folder<T>, +{ + type Result = C::Result; + + fn consume(mut self, item: T) -> Self { + self.base = self.base.consume(item); + self + } + + fn consume_iter<I>(mut self, iter: I) -> Self + where + I: IntoIterator<Item = T>, + { + fn cool<'a, T>(fuse: &'a Fuse<'_>) -> impl Fn(&T) -> bool + 'a { + move |_| !fuse.panicked() + } + + self.base = { + let fuse = &self.fuse; + let iter = iter.into_iter().take_while(cool(fuse)); + self.base.consume_iter(iter) + }; + self + } + + fn complete(self) -> C::Result { + self.base.complete() + } + + fn full(&self) -> bool { + self.fuse.panicked() || self.base.full() + } +} + +struct PanicFuseReducer<'a, C> { + base: C, + _fuse: Fuse<'a>, +} + +impl<'a, T, C> Reducer<T> for PanicFuseReducer<'a, C> +where + C: Reducer<T>, +{ + fn reduce(self, left: T, right: T) -> T { + self.base.reduce(left, right) + } +} diff --git a/third_party/rust/rayon/src/iter/par_bridge.rs b/third_party/rust/rayon/src/iter/par_bridge.rs new file mode 100644 index 0000000000..8398274b35 --- /dev/null +++ b/third_party/rust/rayon/src/iter/par_bridge.rs @@ -0,0 +1,162 @@ +use std::sync::atomic::{AtomicBool, AtomicUsize, Ordering}; +use std::sync::Mutex; + +use crate::iter::plumbing::{bridge_unindexed, Folder, UnindexedConsumer, UnindexedProducer}; +use crate::iter::ParallelIterator; +use crate::{current_num_threads, current_thread_index}; + +/// Conversion trait to convert an `Iterator` to a `ParallelIterator`. +/// +/// This creates a "bridge" from a sequential iterator to a parallel one, by distributing its items +/// across the Rayon thread pool. This has the advantage of being able to parallelize just about +/// anything, but the resulting `ParallelIterator` can be less efficient than if you started with +/// `par_iter` instead. However, it can still be useful for iterators that are difficult to +/// parallelize by other means, like channels or file or network I/O. +/// +/// The resulting iterator is not guaranteed to keep the order of the original iterator. +/// +/// # Examples +/// +/// To use this trait, take an existing `Iterator` and call `par_bridge` on it. After that, you can +/// use any of the `ParallelIterator` methods: +/// +/// ``` +/// use rayon::iter::ParallelBridge; +/// use rayon::prelude::ParallelIterator; +/// use std::sync::mpsc::channel; +/// +/// let rx = { +/// let (tx, rx) = channel(); +/// +/// tx.send("one!"); +/// tx.send("two!"); +/// tx.send("three!"); +/// +/// rx +/// }; +/// +/// let mut output: Vec<&'static str> = rx.into_iter().par_bridge().collect(); +/// output.sort_unstable(); +/// +/// assert_eq!(&*output, &["one!", "three!", "two!"]); +/// ``` +pub trait ParallelBridge: Sized { + /// Creates a bridge from this type to a `ParallelIterator`. + fn par_bridge(self) -> IterBridge<Self>; +} + +impl<T: Iterator + Send> ParallelBridge for T +where + T::Item: Send, +{ + fn par_bridge(self) -> IterBridge<Self> { + IterBridge { iter: self } + } +} + +/// `IterBridge` is a parallel iterator that wraps a sequential iterator. +/// +/// This type is created when using the `par_bridge` method on `ParallelBridge`. See the +/// [`ParallelBridge`] documentation for details. +/// +/// [`ParallelBridge`]: trait.ParallelBridge.html +#[derive(Debug, Clone)] +pub struct IterBridge<Iter> { + iter: Iter, +} + +impl<Iter: Iterator + Send> ParallelIterator for IterBridge<Iter> +where + Iter::Item: Send, +{ + type Item = Iter::Item; + + fn drive_unindexed<C>(self, consumer: C) -> C::Result + where + C: UnindexedConsumer<Self::Item>, + { + let num_threads = current_num_threads(); + let threads_started: Vec<_> = (0..num_threads).map(|_| AtomicBool::new(false)).collect(); + + bridge_unindexed( + &IterParallelProducer { + split_count: AtomicUsize::new(num_threads), + iter: Mutex::new(self.iter.fuse()), + threads_started: &threads_started, + }, + consumer, + ) + } +} + +struct IterParallelProducer<'a, Iter> { + split_count: AtomicUsize, + iter: Mutex<std::iter::Fuse<Iter>>, + threads_started: &'a [AtomicBool], +} + +impl<Iter: Iterator + Send> UnindexedProducer for &IterParallelProducer<'_, Iter> { + type Item = Iter::Item; + + fn split(self) -> (Self, Option<Self>) { + let mut count = self.split_count.load(Ordering::SeqCst); + + loop { + // Check if the iterator is exhausted + if let Some(new_count) = count.checked_sub(1) { + match self.split_count.compare_exchange_weak( + count, + new_count, + Ordering::SeqCst, + Ordering::SeqCst, + ) { + Ok(_) => return (self, Some(self)), + Err(last_count) => count = last_count, + } + } else { + return (self, None); + } + } + } + + fn fold_with<F>(self, mut folder: F) -> F + where + F: Folder<Self::Item>, + { + // Guard against work-stealing-induced recursion, in case `Iter::next()` + // calls rayon internally, so we don't deadlock our mutex. We might also + // be recursing via `folder` methods, which doesn't present a mutex hazard, + // but it's lower overhead for us to just check this once, rather than + // updating additional shared state on every mutex lock/unlock. + // (If this isn't a rayon thread, then there's no work-stealing anyway...) + if let Some(i) = current_thread_index() { + // Note: If the number of threads in the pool ever grows dynamically, then + // we'll end up sharing flags and may falsely detect recursion -- that's + // still fine for overall correctness, just not optimal for parallelism. + let thread_started = &self.threads_started[i % self.threads_started.len()]; + if thread_started.swap(true, Ordering::Relaxed) { + // We can't make progress with a nested mutex, so just return and let + // the outermost loop continue with the rest of the iterator items. + return folder; + } + } + + loop { + if let Ok(mut iter) = self.iter.lock() { + if let Some(it) = iter.next() { + drop(iter); + folder = folder.consume(it); + if folder.full() { + return folder; + } + } else { + return folder; + } + } else { + // any panics from other threads will have been caught by the pool, + // and will be re-thrown when joined - just exit + return folder; + } + } + } +} diff --git a/third_party/rust/rayon/src/iter/plumbing/README.md b/third_party/rust/rayon/src/iter/plumbing/README.md new file mode 100644 index 0000000000..dbee36ba00 --- /dev/null +++ b/third_party/rust/rayon/src/iter/plumbing/README.md @@ -0,0 +1,315 @@ +# Parallel Iterators + +These are some notes on the design of the parallel iterator traits. +This file does not describe how to **use** parallel iterators. + +## The challenge + +Parallel iterators are more complicated than sequential iterators. +The reason is that they have to be able to split themselves up and +operate in parallel across the two halves. + +The current design for parallel iterators has two distinct modes in +which they can be used; as we will see, not all iterators support both +modes (which is why there are two): + +- **Pull mode** (the `Producer` and `UnindexedProducer` traits): in this mode, + the iterator is asked to produce the next item using a call to `next`. This + is basically like a normal iterator, but with a twist: you can split the + iterator in half to produce disjoint items in separate threads. + - in the `Producer` trait, splitting is done with `split_at`, which accepts + an index where the split should be performed. Only indexed iterators can + work in this mode, as they know exactly how much data they will produce, + and how to locate the requested index. + - in the `UnindexedProducer` trait, splitting is done with `split`, which + simply requests that the producer divide itself *approximately* in half. + This is useful when the exact length and/or layout is unknown, as with + `String` characters, or when the length might exceed `usize`, as with + `Range<u64>` on 32-bit platforms. + - In theory, any `Producer` could act unindexed, but we don't currently + use that possibility. When you know the exact length, a `split` can + simply be implemented as `split_at(length/2)`. +- **Push mode** (the `Consumer` and `UnindexedConsumer` traits): in + this mode, the iterator instead is *given* each item in turn, which + is then processed. This is the opposite of a normal iterator. It's + more like a `for_each` call: each time a new item is produced, the + `consume` method is called with that item. (The traits themselves are + a bit more complex, as they support state that can be threaded + through and ultimately reduced.) Unlike producers, there are two + variants of consumers. The difference is how the split is performed: + - in the `Consumer` trait, splitting is done with `split_at`, which + accepts an index where the split should be performed. All + iterators can work in this mode. The resulting halves thus have an + idea about how much data they expect to consume. + - in the `UnindexedConsumer` trait, splitting is done with + `split_off_left`. There is no index: the resulting halves must be + prepared to process any amount of data, and they don't know where that + data falls in the overall stream. + - Not all consumers can operate in this mode. It works for + `for_each` and `reduce`, for example, but it does not work for + `collect_into_vec`, since in that case the position of each item is + important for knowing where it ends up in the target collection. + +## How iterator execution proceeds + +We'll walk through this example iterator chain to start. This chain +demonstrates more-or-less the full complexity of what can happen. + +```rust +vec1.par_iter() + .zip(vec2.par_iter()) + .flat_map(some_function) + .for_each(some_other_function) +``` + +To handle an iterator chain, we start by creating consumers. This +works from the end. So in this case, the call to `for_each` is the +final step, so it will create a `ForEachConsumer` that, given an item, +just calls `some_other_function` with that item. (`ForEachConsumer` is +a very simple consumer because it doesn't need to thread any state +between items at all.) + +Now, the `for_each` call will pass this consumer to the base iterator, +which is the `flat_map`. It will do this by calling the `drive_unindexed` +method on the `ParallelIterator` trait. `drive_unindexed` basically +says "produce items for this iterator and feed them to this consumer"; +it only works for unindexed consumers. + +(As an aside, it is interesting that only some consumers can work in +unindexed mode, but all producers can *drive* an unindexed consumer. +In contrast, only some producers can drive an *indexed* consumer, but +all consumers can be supplied indexes. Isn't variance neat.) + +As it happens, `FlatMap` only works with unindexed consumers anyway. +This is because flat-map basically has no idea how many items it will +produce. If you ask flat-map to produce the 22nd item, it can't do it, +at least not without some intermediate state. It doesn't know whether +processing the first item will create 1 item, 3 items, or 100; +therefore, to produce an arbitrary item, it would basically just have +to start at the beginning and execute sequentially, which is not what +we want. But for unindexed consumers, this doesn't matter, since they +don't need to know how much data they will get. + +Therefore, `FlatMap` can wrap the `ForEachConsumer` with a +`FlatMapConsumer` that feeds to it. This `FlatMapConsumer` will be +given one item. It will then invoke `some_function` to get a parallel +iterator out. It will then ask this new parallel iterator to drive the +`ForEachConsumer`. The `drive_unindexed` method on `flat_map` can then +pass the `FlatMapConsumer` up the chain to the previous item, which is +`zip`. At this point, something interesting happens. + +## Switching from push to pull mode + +If you think about `zip`, it can't really be implemented as a +consumer, at least not without an intermediate thread and some +channels or something (or maybe coroutines). The problem is that it +has to walk two iterators *in lockstep*. Basically, it can't call two +`drive` methods simultaneously, it can only call one at a time. So at +this point, the `zip` iterator needs to switch from *push mode* into +*pull mode*. + +You'll note that `Zip` is only usable if its inputs implement +`IndexedParallelIterator`, meaning that they can produce data starting +at random points in the stream. This need to switch to push mode is +exactly why. If we want to split a zip iterator at position 22, we +need to be able to start zipping items from index 22 right away, +without having to start from index 0. + +Anyway, so at this point, the `drive_unindexed` method for `Zip` stops +creating consumers. Instead, it creates a *producer*, a `ZipProducer`, +to be exact, and calls the `bridge` function in the `internals` +module. Creating a `ZipProducer` will in turn create producers for +the two iterators being zipped. This is possible because they both +implement `IndexedParallelIterator`. + +The `bridge` function will then connect the consumer, which is +handling the `flat_map` and `for_each`, with the producer, which is +handling the `zip` and its predecessors. It will split down until the +chunks seem reasonably small, then pull items from the producer and +feed them to the consumer. + +## The base case + +The other time that `bridge` gets used is when we bottom out in an +indexed producer, such as a slice or range. There is also a +`bridge_unindexed` equivalent for - you guessed it - unindexed producers, +such as string characters. + +<a name="producer-callback"> + +## What on earth is `ProducerCallback`? + +We saw that when you call a parallel action method like +`par_iter.reduce()`, that will create a "reducing" consumer and then +invoke `par_iter.drive_unindexed()` (or `par_iter.drive()`) as +appropriate. This may create yet more consumers as we proceed up the +parallel iterator chain. But at some point we're going to get to the +start of the chain, or to a parallel iterator (like `zip()`) that has +to coordinate multiple inputs. At that point, we need to start +converting parallel iterators into producers. + +The way we do this is by invoking the method `with_producer()`, defined on +`IndexedParallelIterator`. This is a callback scheme. In an ideal world, +it would work like this: + +```rust +base_iter.with_producer(|base_producer| { + // here, `base_producer` is the producer for `base_iter` +}); +``` + +In that case, we could implement a combinator like `map()` by getting +the producer for the base iterator, wrapping it to make our own +`MapProducer`, and then passing that to the callback. Something like +this: + +```rust +struct MapProducer<'f, P, F: 'f> { + base: P, + map_op: &'f F, +} + +impl<I, F> IndexedParallelIterator for Map<I, F> + where I: IndexedParallelIterator, + F: MapOp<I::Item>, +{ + fn with_producer<CB>(self, callback: CB) -> CB::Output { + let map_op = &self.map_op; + self.base_iter.with_producer(|base_producer| { + // Here `producer` is the producer for `self.base_iter`. + // Wrap that to make a `MapProducer` + let map_producer = MapProducer { + base: base_producer, + map_op: map_op + }; + + // invoke the callback with the wrapped version + callback(map_producer) + }); + } +}); +``` + +This example demonstrates some of the power of the callback scheme. +It winds up being a very flexible setup. For one thing, it means we +can take ownership of `par_iter`; we can then in turn give ownership +away of its bits and pieces into the producer (this is very useful if +the iterator owns an `&mut` slice, for example), or create shared +references and put *those* in the producer. In the case of map, for +example, the parallel iterator owns the `map_op`, and we borrow +references to it which we then put into the `MapProducer` (this means +the `MapProducer` can easily split itself and share those references). +The `with_producer` method can also create resources that are needed +during the parallel execution, since the producer does not have to be +returned. + +Unfortunately there is a catch. We can't actually use closures the way +I showed you. To see why, think about the type that `map_producer` +would have to have. If we were going to write the `with_producer` +method using a closure, it would have to look something like this: + +```rust +pub trait IndexedParallelIterator: ParallelIterator { + type Producer; + fn with_producer<CB, R>(self, callback: CB) -> R + where CB: FnOnce(Self::Producer) -> R; + ... +} +``` + +Note that we had to add this associated type `Producer` so that +we could specify the argument of the callback to be `Self::Producer`. +Now, imagine trying to write that `MapProducer` impl using this style: + +```rust +impl<I, F> IndexedParallelIterator for Map<I, F> + where I: IndexedParallelIterator, + F: MapOp<I::Item>, +{ + type MapProducer = MapProducer<'f, P::Producer, F>; + // ^^ wait, what is this `'f`? + + fn with_producer<CB, R>(self, callback: CB) -> R + where CB: FnOnce(Self::Producer) -> R + { + let map_op = &self.map_op; + // ^^^^^^ `'f` is (conceptually) the lifetime of this reference, + // so it will be different for each call to `with_producer`! + } +} +``` + +This may look familiar to you: it's the same problem that we have +trying to define an `Iterable` trait. Basically, the producer type +needs to include a lifetime (here, `'f`) that refers to the body of +`with_producer` and hence is not in scope at the impl level. + +If we had [associated type constructors][1598], we could solve this +problem that way. But there is another solution. We can use a +dedicated callback trait like `ProducerCallback`, instead of `FnOnce`: + +[1598]: https://github.com/rust-lang/rfcs/pull/1598 + +```rust +pub trait ProducerCallback<T> { + type Output; + fn callback<P>(self, producer: P) -> Self::Output + where P: Producer<Item=T>; +} +``` + +Using this trait, the signature of `with_producer()` looks like this: + +```rust +fn with_producer<CB: ProducerCallback<Self::Item>>(self, callback: CB) -> CB::Output; +``` + +Notice that this signature **never has to name the producer type** -- +there is no associated type `Producer` anymore. This is because the +`callback()` method is generically over **all** producers `P`. + +The problem is that now the `||` sugar doesn't work anymore. So we +have to manually create the callback struct, which is a mite tedious. +So our `MapProducer` code looks like this: + +```rust +impl<I, F> IndexedParallelIterator for Map<I, F> + where I: IndexedParallelIterator, + F: MapOp<I::Item>, +{ + fn with_producer<CB>(self, callback: CB) -> CB::Output + where CB: ProducerCallback<Self::Item> + { + return self.base.with_producer(Callback { callback: callback, map_op: self.map_op }); + // ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ + // Manual version of the closure sugar: create an instance + // of a struct that implements `ProducerCallback`. + + // The struct declaration. Each field is something that need to capture from the + // creating scope. + struct Callback<CB, F> { + callback: CB, + map_op: F, + } + + // Implement the `ProducerCallback` trait. This is pure boilerplate. + impl<T, F, CB> ProducerCallback<T> for Callback<CB, F> + where F: MapOp<T>, + CB: ProducerCallback<F::Output> + { + type Output = CB::Output; + + fn callback<P>(self, base: P) -> CB::Output + where P: Producer<Item=T> + { + // The body of the closure is here: + let producer = MapProducer { base: base, + map_op: &self.map_op }; + self.callback.callback(producer) + } + } + } +} +``` + +OK, a bit tedious, but it works! diff --git a/third_party/rust/rayon/src/iter/plumbing/mod.rs b/third_party/rust/rayon/src/iter/plumbing/mod.rs new file mode 100644 index 0000000000..71d4fb4c3d --- /dev/null +++ b/third_party/rust/rayon/src/iter/plumbing/mod.rs @@ -0,0 +1,484 @@ +//! Traits and functions used to implement parallel iteration. These are +//! low-level details -- users of parallel iterators should not need to +//! interact with them directly. See [the `plumbing` README][r] for a general overview. +//! +//! [r]: https://github.com/rayon-rs/rayon/blob/master/src/iter/plumbing/README.md + +use crate::join_context; + +use super::IndexedParallelIterator; + +use std::cmp; +use std::usize; + +/// The `ProducerCallback` trait is a kind of generic closure, +/// [analogous to `FnOnce`][FnOnce]. See [the corresponding section in +/// the plumbing README][r] for more details. +/// +/// [r]: https://github.com/rayon-rs/rayon/blob/master/src/iter/plumbing/README.md#producer-callback +/// [FnOnce]: https://doc.rust-lang.org/std/ops/trait.FnOnce.html +pub trait ProducerCallback<T> { + /// The type of value returned by this callback. Analogous to + /// [`Output` from the `FnOnce` trait][Output]. + /// + /// [Output]: https://doc.rust-lang.org/std/ops/trait.FnOnce.html#associatedtype.Output + type Output; + + /// Invokes the callback with the given producer as argument. The + /// key point of this trait is that this method is generic over + /// `P`, and hence implementors must be defined for any producer. + fn callback<P>(self, producer: P) -> Self::Output + where + P: Producer<Item = T>; +} + +/// A `Producer` is effectively a "splittable `IntoIterator`". That +/// is, a producer is a value which can be converted into an iterator +/// at any time: at that point, it simply produces items on demand, +/// like any iterator. But what makes a `Producer` special is that, +/// *before* we convert to an iterator, we can also **split** it at a +/// particular point using the `split_at` method. This will yield up +/// two producers, one producing the items before that point, and one +/// producing the items after that point (these two producers can then +/// independently be split further, or be converted into iterators). +/// In Rayon, this splitting is used to divide between threads. +/// See [the `plumbing` README][r] for further details. +/// +/// Note that each producer will always produce a fixed number of +/// items N. However, this number N is not queryable through the API; +/// the consumer is expected to track it. +/// +/// NB. You might expect `Producer` to extend the `IntoIterator` +/// trait. However, [rust-lang/rust#20671][20671] prevents us from +/// declaring the DoubleEndedIterator and ExactSizeIterator +/// constraints on a required IntoIterator trait, so we inline +/// IntoIterator here until that issue is fixed. +/// +/// [r]: https://github.com/rayon-rs/rayon/blob/master/src/iter/plumbing/README.md +/// [20671]: https://github.com/rust-lang/rust/issues/20671 +pub trait Producer: Send + Sized { + /// The type of item that will be produced by this producer once + /// it is converted into an iterator. + type Item; + + /// The type of iterator we will become. + type IntoIter: Iterator<Item = Self::Item> + DoubleEndedIterator + ExactSizeIterator; + + /// Convert `self` into an iterator; at this point, no more parallel splits + /// are possible. + fn into_iter(self) -> Self::IntoIter; + + /// The minimum number of items that we will process + /// sequentially. Defaults to 1, which means that we will split + /// all the way down to a single item. This can be raised higher + /// using the [`with_min_len`] method, which will force us to + /// create sequential tasks at a larger granularity. Note that + /// Rayon automatically normally attempts to adjust the size of + /// parallel splits to reduce overhead, so this should not be + /// needed. + /// + /// [`with_min_len`]: ../trait.IndexedParallelIterator.html#method.with_min_len + fn min_len(&self) -> usize { + 1 + } + + /// The maximum number of items that we will process + /// sequentially. Defaults to MAX, which means that we can choose + /// not to split at all. This can be lowered using the + /// [`with_max_len`] method, which will force us to create more + /// parallel tasks. Note that Rayon automatically normally + /// attempts to adjust the size of parallel splits to reduce + /// overhead, so this should not be needed. + /// + /// [`with_max_len`]: ../trait.IndexedParallelIterator.html#method.with_max_len + fn max_len(&self) -> usize { + usize::MAX + } + + /// Split into two producers; one produces items `0..index`, the + /// other `index..N`. Index must be less than or equal to `N`. + fn split_at(self, index: usize) -> (Self, Self); + + /// Iterate the producer, feeding each element to `folder`, and + /// stop when the folder is full (or all elements have been consumed). + /// + /// The provided implementation is sufficient for most iterables. + fn fold_with<F>(self, folder: F) -> F + where + F: Folder<Self::Item>, + { + folder.consume_iter(self.into_iter()) + } +} + +/// A consumer is effectively a [generalized "fold" operation][fold], +/// and in fact each consumer will eventually be converted into a +/// [`Folder`]. What makes a consumer special is that, like a +/// [`Producer`], it can be **split** into multiple consumers using +/// the `split_at` method. When a consumer is split, it produces two +/// consumers, as well as a **reducer**. The two consumers can be fed +/// items independently, and when they are done the reducer is used to +/// combine their two results into one. See [the `plumbing` +/// README][r] for further details. +/// +/// [r]: https://github.com/rayon-rs/rayon/blob/master/src/iter/plumbing/README.md +/// [fold]: https://doc.rust-lang.org/std/iter/trait.Iterator.html#method.fold +/// [`Folder`]: trait.Folder.html +/// [`Producer`]: trait.Producer.html +pub trait Consumer<Item>: Send + Sized { + /// The type of folder that this consumer can be converted into. + type Folder: Folder<Item, Result = Self::Result>; + + /// The type of reducer that is produced if this consumer is split. + type Reducer: Reducer<Self::Result>; + + /// The type of result that this consumer will ultimately produce. + type Result: Send; + + /// Divide the consumer into two consumers, one processing items + /// `0..index` and one processing items from `index..`. Also + /// produces a reducer that can be used to reduce the results at + /// the end. + fn split_at(self, index: usize) -> (Self, Self, Self::Reducer); + + /// Convert the consumer into a folder that can consume items + /// sequentially, eventually producing a final result. + fn into_folder(self) -> Self::Folder; + + /// Hint whether this `Consumer` would like to stop processing + /// further items, e.g. if a search has been completed. + fn full(&self) -> bool; +} + +/// The `Folder` trait encapsulates [the standard fold +/// operation][fold]. It can be fed many items using the `consume` +/// method. At the end, once all items have been consumed, it can then +/// be converted (using `complete`) into a final value. +/// +/// [fold]: https://doc.rust-lang.org/std/iter/trait.Iterator.html#method.fold +pub trait Folder<Item>: Sized { + /// The type of result that will ultimately be produced by the folder. + type Result; + + /// Consume next item and return new sequential state. + fn consume(self, item: Item) -> Self; + + /// Consume items from the iterator until full, and return new sequential state. + /// + /// This method is **optional**. The default simply iterates over + /// `iter`, invoking `consume` and checking after each iteration + /// whether `full` returns false. + /// + /// The main reason to override it is if you can provide a more + /// specialized, efficient implementation. + fn consume_iter<I>(mut self, iter: I) -> Self + where + I: IntoIterator<Item = Item>, + { + for item in iter { + self = self.consume(item); + if self.full() { + break; + } + } + self + } + + /// Finish consuming items, produce final result. + fn complete(self) -> Self::Result; + + /// Hint whether this `Folder` would like to stop processing + /// further items, e.g. if a search has been completed. + fn full(&self) -> bool; +} + +/// The reducer is the final step of a `Consumer` -- after a consumer +/// has been split into two parts, and each of those parts has been +/// fully processed, we are left with two results. The reducer is then +/// used to combine those two results into one. See [the `plumbing` +/// README][r] for further details. +/// +/// [r]: https://github.com/rayon-rs/rayon/blob/master/src/iter/plumbing/README.md +pub trait Reducer<Result> { + /// Reduce two final results into one; this is executed after a + /// split. + fn reduce(self, left: Result, right: Result) -> Result; +} + +/// A stateless consumer can be freely copied. These consumers can be +/// used like regular consumers, but they also support a +/// `split_off_left` method that does not take an index to split, but +/// simply splits at some arbitrary point (`for_each`, for example, +/// produces an unindexed consumer). +pub trait UnindexedConsumer<I>: Consumer<I> { + /// Splits off a "left" consumer and returns it. The `self` + /// consumer should then be used to consume the "right" portion of + /// the data. (The ordering matters for methods like find_first -- + /// values produced by the returned value are given precedence + /// over values produced by `self`.) Once the left and right + /// halves have been fully consumed, you should reduce the results + /// with the result of `to_reducer`. + fn split_off_left(&self) -> Self; + + /// Creates a reducer that can be used to combine the results from + /// a split consumer. + fn to_reducer(&self) -> Self::Reducer; +} + +/// A variant on `Producer` which does not know its exact length or +/// cannot represent it in a `usize`. These producers act like +/// ordinary producers except that they cannot be told to split at a +/// particular point. Instead, you just ask them to split 'somewhere'. +/// +/// (In principle, `Producer` could extend this trait; however, it +/// does not because to do so would require producers to carry their +/// own length with them.) +pub trait UnindexedProducer: Send + Sized { + /// The type of item returned by this producer. + type Item; + + /// Split midway into a new producer if possible, otherwise return `None`. + fn split(self) -> (Self, Option<Self>); + + /// Iterate the producer, feeding each element to `folder`, and + /// stop when the folder is full (or all elements have been consumed). + fn fold_with<F>(self, folder: F) -> F + where + F: Folder<Self::Item>; +} + +/// A splitter controls the policy for splitting into smaller work items. +/// +/// Thief-splitting is an adaptive policy that starts by splitting into +/// enough jobs for every worker thread, and then resets itself whenever a +/// job is actually stolen into a different thread. +#[derive(Clone, Copy)] +struct Splitter { + /// The `splits` tell us approximately how many remaining times we'd + /// like to split this job. We always just divide it by two though, so + /// the effective number of pieces will be `next_power_of_two()`. + splits: usize, +} + +impl Splitter { + #[inline] + fn new() -> Splitter { + Splitter { + splits: crate::current_num_threads(), + } + } + + #[inline] + fn try_split(&mut self, stolen: bool) -> bool { + let Splitter { splits } = *self; + + if stolen { + // This job was stolen! Reset the number of desired splits to the + // thread count, if that's more than we had remaining anyway. + self.splits = cmp::max(crate::current_num_threads(), self.splits / 2); + true + } else if splits > 0 { + // We have splits remaining, make it so. + self.splits /= 2; + true + } else { + // Not stolen, and no more splits -- we're done! + false + } + } +} + +/// The length splitter is built on thief-splitting, but additionally takes +/// into account the remaining length of the iterator. +#[derive(Clone, Copy)] +struct LengthSplitter { + inner: Splitter, + + /// The smallest we're willing to divide into. Usually this is just 1, + /// but you can choose a larger working size with `with_min_len()`. + min: usize, +} + +impl LengthSplitter { + /// Creates a new splitter based on lengths. + /// + /// The `min` is a hard lower bound. We'll never split below that, but + /// of course an iterator might start out smaller already. + /// + /// The `max` is an upper bound on the working size, used to determine + /// the minimum number of times we need to split to get under that limit. + /// The adaptive algorithm may very well split even further, but never + /// smaller than the `min`. + #[inline] + fn new(min: usize, max: usize, len: usize) -> LengthSplitter { + let mut splitter = LengthSplitter { + inner: Splitter::new(), + min: cmp::max(min, 1), + }; + + // Divide the given length by the max working length to get the minimum + // number of splits we need to get under that max. This rounds down, + // but the splitter actually gives `next_power_of_two()` pieces anyway. + // e.g. len 12345 / max 100 = 123 min_splits -> 128 pieces. + let min_splits = len / cmp::max(max, 1); + + // Only update the value if it's not splitting enough already. + if min_splits > splitter.inner.splits { + splitter.inner.splits = min_splits; + } + + splitter + } + + #[inline] + fn try_split(&mut self, len: usize, stolen: bool) -> bool { + // If splitting wouldn't make us too small, try the inner splitter. + len / 2 >= self.min && self.inner.try_split(stolen) + } +} + +/// This helper function is used to "connect" a parallel iterator to a +/// consumer. It will convert the `par_iter` into a producer P and +/// then pull items from P and feed them to `consumer`, splitting and +/// creating parallel threads as needed. +/// +/// This is useful when you are implementing your own parallel +/// iterators: it is often used as the definition of the +/// [`drive_unindexed`] or [`drive`] methods. +/// +/// [`drive_unindexed`]: ../trait.ParallelIterator.html#tymethod.drive_unindexed +/// [`drive`]: ../trait.IndexedParallelIterator.html#tymethod.drive +pub fn bridge<I, C>(par_iter: I, consumer: C) -> C::Result +where + I: IndexedParallelIterator, + C: Consumer<I::Item>, +{ + let len = par_iter.len(); + return par_iter.with_producer(Callback { len, consumer }); + + struct Callback<C> { + len: usize, + consumer: C, + } + + impl<C, I> ProducerCallback<I> for Callback<C> + where + C: Consumer<I>, + { + type Output = C::Result; + fn callback<P>(self, producer: P) -> C::Result + where + P: Producer<Item = I>, + { + bridge_producer_consumer(self.len, producer, self.consumer) + } + } +} + +/// This helper function is used to "connect" a producer and a +/// consumer. You may prefer to call [`bridge`], which wraps this +/// function. This function will draw items from `producer` and feed +/// them to `consumer`, splitting and creating parallel tasks when +/// needed. +/// +/// This is useful when you are implementing your own parallel +/// iterators: it is often used as the definition of the +/// [`drive_unindexed`] or [`drive`] methods. +/// +/// [`bridge`]: fn.bridge.html +/// [`drive_unindexed`]: ../trait.ParallelIterator.html#tymethod.drive_unindexed +/// [`drive`]: ../trait.IndexedParallelIterator.html#tymethod.drive +pub fn bridge_producer_consumer<P, C>(len: usize, producer: P, consumer: C) -> C::Result +where + P: Producer, + C: Consumer<P::Item>, +{ + let splitter = LengthSplitter::new(producer.min_len(), producer.max_len(), len); + return helper(len, false, splitter, producer, consumer); + + fn helper<P, C>( + len: usize, + migrated: bool, + mut splitter: LengthSplitter, + producer: P, + consumer: C, + ) -> C::Result + where + P: Producer, + C: Consumer<P::Item>, + { + if consumer.full() { + consumer.into_folder().complete() + } else if splitter.try_split(len, migrated) { + let mid = len / 2; + let (left_producer, right_producer) = producer.split_at(mid); + let (left_consumer, right_consumer, reducer) = consumer.split_at(mid); + let (left_result, right_result) = join_context( + |context| { + helper( + mid, + context.migrated(), + splitter, + left_producer, + left_consumer, + ) + }, + |context| { + helper( + len - mid, + context.migrated(), + splitter, + right_producer, + right_consumer, + ) + }, + ); + reducer.reduce(left_result, right_result) + } else { + producer.fold_with(consumer.into_folder()).complete() + } + } +} + +/// A variant of [`bridge_producer_consumer`] where the producer is an unindexed producer. +/// +/// [`bridge_producer_consumer`]: fn.bridge_producer_consumer.html +pub fn bridge_unindexed<P, C>(producer: P, consumer: C) -> C::Result +where + P: UnindexedProducer, + C: UnindexedConsumer<P::Item>, +{ + let splitter = Splitter::new(); + bridge_unindexed_producer_consumer(false, splitter, producer, consumer) +} + +fn bridge_unindexed_producer_consumer<P, C>( + migrated: bool, + mut splitter: Splitter, + producer: P, + consumer: C, +) -> C::Result +where + P: UnindexedProducer, + C: UnindexedConsumer<P::Item>, +{ + if consumer.full() { + consumer.into_folder().complete() + } else if splitter.try_split(migrated) { + match producer.split() { + (left_producer, Some(right_producer)) => { + let (reducer, left_consumer, right_consumer) = + (consumer.to_reducer(), consumer.split_off_left(), consumer); + let bridge = bridge_unindexed_producer_consumer; + let (left_result, right_result) = join_context( + |context| bridge(context.migrated(), splitter, left_producer, left_consumer), + |context| bridge(context.migrated(), splitter, right_producer, right_consumer), + ); + reducer.reduce(left_result, right_result) + } + (producer, None) => producer.fold_with(consumer.into_folder()).complete(), + } + } else { + producer.fold_with(consumer.into_folder()).complete() + } +} diff --git a/third_party/rust/rayon/src/iter/positions.rs b/third_party/rust/rayon/src/iter/positions.rs new file mode 100644 index 0000000000..f584bb28e2 --- /dev/null +++ b/third_party/rust/rayon/src/iter/positions.rs @@ -0,0 +1,137 @@ +use super::plumbing::*; +use super::*; + +use std::fmt::{self, Debug}; + +/// `Positions` takes a predicate `predicate` and filters out elements that match, +/// yielding their indices. +/// +/// This struct is created by the [`positions()`] method on [`IndexedParallelIterator`] +/// +/// [`positions()`]: trait.IndexedParallelIterator.html#method.positions +/// [`IndexedParallelIterator`]: trait.IndexedParallelIterator.html +#[must_use = "iterator adaptors are lazy and do nothing unless consumed"] +#[derive(Clone)] +pub struct Positions<I: IndexedParallelIterator, P> { + base: I, + predicate: P, +} + +impl<I: IndexedParallelIterator + Debug, P> Debug for Positions<I, P> { + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + f.debug_struct("Positions") + .field("base", &self.base) + .finish() + } +} + +impl<I, P> Positions<I, P> +where + I: IndexedParallelIterator, +{ + /// Create a new `Positions` iterator. + pub(super) fn new(base: I, predicate: P) -> Self { + Positions { base, predicate } + } +} + +impl<I, P> ParallelIterator for Positions<I, P> +where + I: IndexedParallelIterator, + P: Fn(I::Item) -> bool + Sync + Send, +{ + type Item = usize; + + fn drive_unindexed<C>(self, consumer: C) -> C::Result + where + C: UnindexedConsumer<Self::Item>, + { + let consumer1 = PositionsConsumer::new(consumer, &self.predicate, 0); + self.base.drive(consumer1) + } +} + +/// //////////////////////////////////////////////////////////////////////// +/// Consumer implementation + +struct PositionsConsumer<'p, C, P> { + base: C, + predicate: &'p P, + offset: usize, +} + +impl<'p, C, P> PositionsConsumer<'p, C, P> { + fn new(base: C, predicate: &'p P, offset: usize) -> Self { + PositionsConsumer { + base, + predicate, + offset, + } + } +} + +impl<'p, T, C, P> Consumer<T> for PositionsConsumer<'p, C, P> +where + C: Consumer<usize>, + P: Fn(T) -> bool + Sync, +{ + type Folder = PositionsFolder<'p, C::Folder, P>; + type Reducer = C::Reducer; + type Result = C::Result; + + fn split_at(self, index: usize) -> (Self, Self, C::Reducer) { + let (left, right, reducer) = self.base.split_at(index); + ( + PositionsConsumer::new(left, self.predicate, self.offset), + PositionsConsumer::new(right, self.predicate, self.offset + index), + reducer, + ) + } + + fn into_folder(self) -> Self::Folder { + PositionsFolder { + base: self.base.into_folder(), + predicate: self.predicate, + offset: self.offset, + } + } + + fn full(&self) -> bool { + self.base.full() + } +} + +struct PositionsFolder<'p, F, P> { + base: F, + predicate: &'p P, + offset: usize, +} + +impl<F, P, T> Folder<T> for PositionsFolder<'_, F, P> +where + F: Folder<usize>, + P: Fn(T) -> bool, +{ + type Result = F::Result; + + fn consume(mut self, item: T) -> Self { + let index = self.offset; + self.offset += 1; + if (self.predicate)(item) { + self.base = self.base.consume(index); + } + self + } + + // This cannot easily specialize `consume_iter` to be better than + // the default, because that requires checking `self.base.full()` + // during a call to `self.base.consume_iter()`. (#632) + + fn complete(self) -> Self::Result { + self.base.complete() + } + + fn full(&self) -> bool { + self.base.full() + } +} diff --git a/third_party/rust/rayon/src/iter/product.rs b/third_party/rust/rayon/src/iter/product.rs new file mode 100644 index 0000000000..a3d0727f93 --- /dev/null +++ b/third_party/rust/rayon/src/iter/product.rs @@ -0,0 +1,114 @@ +use super::plumbing::*; +use super::ParallelIterator; + +use std::iter::{self, Product}; +use std::marker::PhantomData; + +pub(super) fn product<PI, P>(pi: PI) -> P +where + PI: ParallelIterator, + P: Send + Product<PI::Item> + Product, +{ + pi.drive_unindexed(ProductConsumer::new()) +} + +fn mul<T: Product>(left: T, right: T) -> T { + iter::once(left).chain(iter::once(right)).product() +} + +struct ProductConsumer<P: Send> { + _marker: PhantomData<*const P>, +} + +unsafe impl<P: Send> Send for ProductConsumer<P> {} + +impl<P: Send> ProductConsumer<P> { + fn new() -> ProductConsumer<P> { + ProductConsumer { + _marker: PhantomData, + } + } +} + +impl<P, T> Consumer<T> for ProductConsumer<P> +where + P: Send + Product<T> + Product, +{ + type Folder = ProductFolder<P>; + type Reducer = Self; + type Result = P; + + fn split_at(self, _index: usize) -> (Self, Self, Self) { + ( + ProductConsumer::new(), + ProductConsumer::new(), + ProductConsumer::new(), + ) + } + + fn into_folder(self) -> Self::Folder { + ProductFolder { + product: iter::empty::<T>().product(), + } + } + + fn full(&self) -> bool { + false + } +} + +impl<P, T> UnindexedConsumer<T> for ProductConsumer<P> +where + P: Send + Product<T> + Product, +{ + fn split_off_left(&self) -> Self { + ProductConsumer::new() + } + + fn to_reducer(&self) -> Self::Reducer { + ProductConsumer::new() + } +} + +impl<P> Reducer<P> for ProductConsumer<P> +where + P: Send + Product, +{ + fn reduce(self, left: P, right: P) -> P { + mul(left, right) + } +} + +struct ProductFolder<P> { + product: P, +} + +impl<P, T> Folder<T> for ProductFolder<P> +where + P: Product<T> + Product, +{ + type Result = P; + + fn consume(self, item: T) -> Self { + ProductFolder { + product: mul(self.product, iter::once(item).product()), + } + } + + fn consume_iter<I>(self, iter: I) -> Self + where + I: IntoIterator<Item = T>, + { + ProductFolder { + product: mul(self.product, iter.into_iter().product()), + } + } + + fn complete(self) -> P { + self.product + } + + fn full(&self) -> bool { + false + } +} diff --git a/third_party/rust/rayon/src/iter/reduce.rs b/third_party/rust/rayon/src/iter/reduce.rs new file mode 100644 index 0000000000..321b5dd864 --- /dev/null +++ b/third_party/rust/rayon/src/iter/reduce.rs @@ -0,0 +1,116 @@ +use super::plumbing::*; +use super::ParallelIterator; + +pub(super) fn reduce<PI, R, ID, T>(pi: PI, identity: ID, reduce_op: R) -> T +where + PI: ParallelIterator<Item = T>, + R: Fn(T, T) -> T + Sync, + ID: Fn() -> T + Sync, + T: Send, +{ + let consumer = ReduceConsumer { + identity: &identity, + reduce_op: &reduce_op, + }; + pi.drive_unindexed(consumer) +} + +struct ReduceConsumer<'r, R, ID> { + identity: &'r ID, + reduce_op: &'r R, +} + +impl<'r, R, ID> Copy for ReduceConsumer<'r, R, ID> {} + +impl<'r, R, ID> Clone for ReduceConsumer<'r, R, ID> { + fn clone(&self) -> Self { + *self + } +} + +impl<'r, R, ID, T> Consumer<T> for ReduceConsumer<'r, R, ID> +where + R: Fn(T, T) -> T + Sync, + ID: Fn() -> T + Sync, + T: Send, +{ + type Folder = ReduceFolder<'r, R, T>; + type Reducer = Self; + type Result = T; + + fn split_at(self, _index: usize) -> (Self, Self, Self) { + (self, self, self) + } + + fn into_folder(self) -> Self::Folder { + ReduceFolder { + reduce_op: self.reduce_op, + item: (self.identity)(), + } + } + + fn full(&self) -> bool { + false + } +} + +impl<'r, R, ID, T> UnindexedConsumer<T> for ReduceConsumer<'r, R, ID> +where + R: Fn(T, T) -> T + Sync, + ID: Fn() -> T + Sync, + T: Send, +{ + fn split_off_left(&self) -> Self { + *self + } + + fn to_reducer(&self) -> Self::Reducer { + *self + } +} + +impl<'r, R, ID, T> Reducer<T> for ReduceConsumer<'r, R, ID> +where + R: Fn(T, T) -> T + Sync, +{ + fn reduce(self, left: T, right: T) -> T { + (self.reduce_op)(left, right) + } +} + +struct ReduceFolder<'r, R, T> { + reduce_op: &'r R, + item: T, +} + +impl<'r, R, T> Folder<T> for ReduceFolder<'r, R, T> +where + R: Fn(T, T) -> T, +{ + type Result = T; + + fn consume(self, item: T) -> Self { + ReduceFolder { + reduce_op: self.reduce_op, + item: (self.reduce_op)(self.item, item), + } + } + + fn consume_iter<I>(self, iter: I) -> Self + where + I: IntoIterator<Item = T>, + { + ReduceFolder { + reduce_op: self.reduce_op, + item: iter.into_iter().fold(self.item, self.reduce_op), + } + } + + fn complete(self) -> T { + self.item + } + + fn full(&self) -> bool { + false + } +} diff --git a/third_party/rust/rayon/src/iter/repeat.rs b/third_party/rust/rayon/src/iter/repeat.rs new file mode 100644 index 0000000000..f84a6fe779 --- /dev/null +++ b/third_party/rust/rayon/src/iter/repeat.rs @@ -0,0 +1,241 @@ +use super::plumbing::*; +use super::*; +use std::iter; +use std::usize; + +/// Iterator adaptor for [the `repeat()` function](fn.repeat.html). +#[derive(Debug, Clone)] +pub struct Repeat<T: Clone + Send> { + element: T, +} + +/// Creates a parallel iterator that endlessly repeats `elt` (by +/// cloning it). Note that this iterator has "infinite" length, so +/// typically you would want to use `zip` or `take` or some other +/// means to shorten it, or consider using +/// [the `repeatn()` function](fn.repeatn.html) instead. +/// +/// # Examples +/// +/// ``` +/// use rayon::prelude::*; +/// use rayon::iter::repeat; +/// let x: Vec<(i32, i32)> = repeat(22).zip(0..3).collect(); +/// assert_eq!(x, vec![(22, 0), (22, 1), (22, 2)]); +/// ``` +pub fn repeat<T: Clone + Send>(elt: T) -> Repeat<T> { + Repeat { element: elt } +} + +impl<T> Repeat<T> +where + T: Clone + Send, +{ + /// Takes only `n` repeats of the element, similar to the general + /// [`take()`](trait.IndexedParallelIterator.html#method.take). + /// + /// The resulting `RepeatN` is an `IndexedParallelIterator`, allowing + /// more functionality than `Repeat` alone. + pub fn take(self, n: usize) -> RepeatN<T> { + repeatn(self.element, n) + } + + /// Iterates tuples, repeating the element with items from another + /// iterator, similar to the general + /// [`zip()`](trait.IndexedParallelIterator.html#method.zip). + pub fn zip<Z>(self, zip_op: Z) -> Zip<RepeatN<T>, Z::Iter> + where + Z: IntoParallelIterator, + Z::Iter: IndexedParallelIterator, + { + let z = zip_op.into_par_iter(); + let n = z.len(); + self.take(n).zip(z) + } +} + +impl<T> ParallelIterator for Repeat<T> +where + T: Clone + Send, +{ + type Item = T; + + fn drive_unindexed<C>(self, consumer: C) -> C::Result + where + C: UnindexedConsumer<Self::Item>, + { + let producer = RepeatProducer { + element: self.element, + }; + bridge_unindexed(producer, consumer) + } +} + +/// Unindexed producer for `Repeat`. +struct RepeatProducer<T: Clone + Send> { + element: T, +} + +impl<T: Clone + Send> UnindexedProducer for RepeatProducer<T> { + type Item = T; + + fn split(self) -> (Self, Option<Self>) { + ( + RepeatProducer { + element: self.element.clone(), + }, + Some(RepeatProducer { + element: self.element, + }), + ) + } + + fn fold_with<F>(self, folder: F) -> F + where + F: Folder<T>, + { + folder.consume_iter(iter::repeat(self.element)) + } +} + +/// Iterator adaptor for [the `repeatn()` function](fn.repeatn.html). +#[derive(Debug, Clone)] +pub struct RepeatN<T: Clone + Send> { + element: T, + count: usize, +} + +/// Creates a parallel iterator that produces `n` repeats of `elt` +/// (by cloning it). +/// +/// # Examples +/// +/// ``` +/// use rayon::prelude::*; +/// use rayon::iter::repeatn; +/// let x: Vec<(i32, i32)> = repeatn(22, 3).zip(0..3).collect(); +/// assert_eq!(x, vec![(22, 0), (22, 1), (22, 2)]); +/// ``` +pub fn repeatn<T: Clone + Send>(elt: T, n: usize) -> RepeatN<T> { + RepeatN { + element: elt, + count: n, + } +} + +impl<T> ParallelIterator for RepeatN<T> +where + T: Clone + Send, +{ + type Item = T; + + fn drive_unindexed<C>(self, consumer: C) -> C::Result + where + C: UnindexedConsumer<Self::Item>, + { + bridge(self, consumer) + } + + fn opt_len(&self) -> Option<usize> { + Some(self.count) + } +} + +impl<T> IndexedParallelIterator for RepeatN<T> +where + T: Clone + Send, +{ + fn drive<C>(self, consumer: C) -> C::Result + where + C: Consumer<Self::Item>, + { + bridge(self, consumer) + } + + fn with_producer<CB>(self, callback: CB) -> CB::Output + where + CB: ProducerCallback<Self::Item>, + { + callback.callback(RepeatNProducer { + element: self.element, + count: self.count, + }) + } + + fn len(&self) -> usize { + self.count + } +} + +/// Producer for `RepeatN`. +struct RepeatNProducer<T: Clone + Send> { + element: T, + count: usize, +} + +impl<T: Clone + Send> Producer for RepeatNProducer<T> { + type Item = T; + type IntoIter = Iter<T>; + + fn into_iter(self) -> Self::IntoIter { + Iter { + element: self.element, + count: self.count, + } + } + + fn split_at(self, index: usize) -> (Self, Self) { + ( + RepeatNProducer { + element: self.element.clone(), + count: index, + }, + RepeatNProducer { + element: self.element, + count: self.count - index, + }, + ) + } +} + +/// Iterator for `RepeatN`. +/// +/// This is conceptually like `std::iter::Take<std::iter::Repeat<T>>`, but +/// we need `DoubleEndedIterator` and unconditional `ExactSizeIterator`. +struct Iter<T: Clone> { + element: T, + count: usize, +} + +impl<T: Clone> Iterator for Iter<T> { + type Item = T; + + #[inline] + fn next(&mut self) -> Option<T> { + if self.count > 0 { + self.count -= 1; + Some(self.element.clone()) + } else { + None + } + } + + #[inline] + fn size_hint(&self) -> (usize, Option<usize>) { + (self.count, Some(self.count)) + } +} + +impl<T: Clone> DoubleEndedIterator for Iter<T> { + #[inline] + fn next_back(&mut self) -> Option<T> { + self.next() + } +} + +impl<T: Clone> ExactSizeIterator for Iter<T> { + #[inline] + fn len(&self) -> usize { + self.count + } +} diff --git a/third_party/rust/rayon/src/iter/rev.rs b/third_party/rust/rayon/src/iter/rev.rs new file mode 100644 index 0000000000..a4c3b7c2e1 --- /dev/null +++ b/third_party/rust/rayon/src/iter/rev.rs @@ -0,0 +1,123 @@ +use super::plumbing::*; +use super::*; +use std::iter; + +/// `Rev` is an iterator that produces elements in reverse order. This struct +/// is created by the [`rev()`] method on [`IndexedParallelIterator`] +/// +/// [`rev()`]: trait.IndexedParallelIterator.html#method.rev +/// [`IndexedParallelIterator`]: trait.IndexedParallelIterator.html +#[must_use = "iterator adaptors are lazy and do nothing unless consumed"] +#[derive(Debug, Clone)] +pub struct Rev<I: IndexedParallelIterator> { + base: I, +} + +impl<I> Rev<I> +where + I: IndexedParallelIterator, +{ + /// Creates a new `Rev` iterator. + pub(super) fn new(base: I) -> Self { + Rev { base } + } +} + +impl<I> ParallelIterator for Rev<I> +where + I: IndexedParallelIterator, +{ + type Item = I::Item; + + fn drive_unindexed<C>(self, consumer: C) -> C::Result + where + C: UnindexedConsumer<Self::Item>, + { + bridge(self, consumer) + } + + fn opt_len(&self) -> Option<usize> { + Some(self.len()) + } +} + +impl<I> IndexedParallelIterator for Rev<I> +where + I: IndexedParallelIterator, +{ + fn drive<C: Consumer<Self::Item>>(self, consumer: C) -> C::Result { + bridge(self, consumer) + } + + fn len(&self) -> usize { + self.base.len() + } + + fn with_producer<CB>(self, callback: CB) -> CB::Output + where + CB: ProducerCallback<Self::Item>, + { + let len = self.base.len(); + return self.base.with_producer(Callback { callback, len }); + + struct Callback<CB> { + callback: CB, + len: usize, + } + + impl<T, CB> ProducerCallback<T> for Callback<CB> + where + CB: ProducerCallback<T>, + { + type Output = CB::Output; + fn callback<P>(self, base: P) -> CB::Output + where + P: Producer<Item = T>, + { + let producer = RevProducer { + base, + len: self.len, + }; + self.callback.callback(producer) + } + } + } +} + +struct RevProducer<P> { + base: P, + len: usize, +} + +impl<P> Producer for RevProducer<P> +where + P: Producer, +{ + type Item = P::Item; + type IntoIter = iter::Rev<P::IntoIter>; + + fn into_iter(self) -> Self::IntoIter { + self.base.into_iter().rev() + } + + fn min_len(&self) -> usize { + self.base.min_len() + } + fn max_len(&self) -> usize { + self.base.max_len() + } + + fn split_at(self, index: usize) -> (Self, Self) { + let (left, right) = self.base.split_at(self.len - index); + ( + RevProducer { + base: right, + len: index, + }, + RevProducer { + base: left, + len: self.len - index, + }, + ) + } +} diff --git a/third_party/rust/rayon/src/iter/skip.rs b/third_party/rust/rayon/src/iter/skip.rs new file mode 100644 index 0000000000..2d0f9477e2 --- /dev/null +++ b/third_party/rust/rayon/src/iter/skip.rs @@ -0,0 +1,95 @@ +use super::noop::NoopConsumer; +use super::plumbing::*; +use super::*; +use std::cmp::min; + +/// `Skip` is an iterator that skips over the first `n` elements. +/// This struct is created by the [`skip()`] method on [`IndexedParallelIterator`] +/// +/// [`skip()`]: trait.IndexedParallelIterator.html#method.skip +/// [`IndexedParallelIterator`]: trait.IndexedParallelIterator.html +#[must_use = "iterator adaptors are lazy and do nothing unless consumed"] +#[derive(Debug, Clone)] +pub struct Skip<I> { + base: I, + n: usize, +} + +impl<I> Skip<I> +where + I: IndexedParallelIterator, +{ + /// Creates a new `Skip` iterator. + pub(super) fn new(base: I, n: usize) -> Self { + let n = min(base.len(), n); + Skip { base, n } + } +} + +impl<I> ParallelIterator for Skip<I> +where + I: IndexedParallelIterator, +{ + type Item = I::Item; + + fn drive_unindexed<C>(self, consumer: C) -> C::Result + where + C: UnindexedConsumer<Self::Item>, + { + bridge(self, consumer) + } + + fn opt_len(&self) -> Option<usize> { + Some(self.len()) + } +} + +impl<I> IndexedParallelIterator for Skip<I> +where + I: IndexedParallelIterator, +{ + fn len(&self) -> usize { + self.base.len() - self.n + } + + fn drive<C: Consumer<Self::Item>>(self, consumer: C) -> C::Result { + bridge(self, consumer) + } + + fn with_producer<CB>(self, callback: CB) -> CB::Output + where + CB: ProducerCallback<Self::Item>, + { + return self.base.with_producer(Callback { + callback, + n: self.n, + }); + + struct Callback<CB> { + callback: CB, + n: usize, + } + + impl<T, CB> ProducerCallback<T> for Callback<CB> + where + CB: ProducerCallback<T>, + { + type Output = CB::Output; + fn callback<P>(self, base: P) -> CB::Output + where + P: Producer<Item = T>, + { + crate::in_place_scope(|scope| { + let Self { callback, n } = self; + let (before_skip, after_skip) = base.split_at(n); + + // Run the skipped part separately for side effects. + // We'll still get any panics propagated back by the scope. + scope.spawn(move |_| bridge_producer_consumer(n, before_skip, NoopConsumer)); + + callback.callback(after_skip) + }) + } + } + } +} diff --git a/third_party/rust/rayon/src/iter/splitter.rs b/third_party/rust/rayon/src/iter/splitter.rs new file mode 100644 index 0000000000..40935ac2b0 --- /dev/null +++ b/third_party/rust/rayon/src/iter/splitter.rs @@ -0,0 +1,174 @@ +use super::plumbing::*; +use super::*; + +use std::fmt::{self, Debug}; + +/// The `split` function takes arbitrary data and a closure that knows how to +/// split it, and turns this into a `ParallelIterator`. +/// +/// # Examples +/// +/// As a simple example, Rayon can recursively split ranges of indices +/// +/// ``` +/// use rayon::iter; +/// use rayon::prelude::*; +/// use std::ops::Range; +/// +/// +/// // We define a range of indices as follows +/// type Range1D = Range<usize>; +/// +/// // Splitting it in two can be done like this +/// fn split_range1(r: Range1D) -> (Range1D, Option<Range1D>) { +/// // We are mathematically unable to split the range if there is only +/// // one point inside of it, but we could stop splitting before that. +/// if r.end - r.start <= 1 { return (r, None); } +/// +/// // Here, our range is considered large enough to be splittable +/// let midpoint = r.start + (r.end - r.start) / 2; +/// (r.start..midpoint, Some(midpoint..r.end)) +/// } +/// +/// // By using iter::split, Rayon will split the range until it has enough work +/// // to feed the CPU cores, then give us the resulting sub-ranges +/// iter::split(0..4096, split_range1).for_each(|sub_range| { +/// // As our initial range had a power-of-two size, the final sub-ranges +/// // should have power-of-two sizes too +/// assert!((sub_range.end - sub_range.start).is_power_of_two()); +/// }); +/// ``` +/// +/// This recursive splitting can be extended to two or three dimensions, +/// to reproduce a classic "block-wise" parallelization scheme of graphics and +/// numerical simulations: +/// +/// ``` +/// # use rayon::iter; +/// # use rayon::prelude::*; +/// # use std::ops::Range; +/// # type Range1D = Range<usize>; +/// # fn split_range1(r: Range1D) -> (Range1D, Option<Range1D>) { +/// # if r.end - r.start <= 1 { return (r, None); } +/// # let midpoint = r.start + (r.end - r.start) / 2; +/// # (r.start..midpoint, Some(midpoint..r.end)) +/// # } +/// # +/// // A two-dimensional range of indices can be built out of two 1D ones +/// struct Range2D { +/// // Range of horizontal indices +/// pub rx: Range1D, +/// +/// // Range of vertical indices +/// pub ry: Range1D, +/// } +/// +/// // We want to recursively split them by the largest dimension until we have +/// // enough sub-ranges to feed our mighty multi-core CPU. This function +/// // carries out one such split. +/// fn split_range2(r2: Range2D) -> (Range2D, Option<Range2D>) { +/// // Decide on which axis (horizontal/vertical) the range should be split +/// let width = r2.rx.end - r2.rx.start; +/// let height = r2.ry.end - r2.ry.start; +/// if width >= height { +/// // This is a wide range, split it on the horizontal axis +/// let (split_rx, ry) = (split_range1(r2.rx), r2.ry); +/// let out1 = Range2D { +/// rx: split_rx.0, +/// ry: ry.clone(), +/// }; +/// let out2 = split_rx.1.map(|rx| Range2D { rx, ry }); +/// (out1, out2) +/// } else { +/// // This is a tall range, split it on the vertical axis +/// let (rx, split_ry) = (r2.rx, split_range1(r2.ry)); +/// let out1 = Range2D { +/// rx: rx.clone(), +/// ry: split_ry.0, +/// }; +/// let out2 = split_ry.1.map(|ry| Range2D { rx, ry, }); +/// (out1, out2) +/// } +/// } +/// +/// // Again, rayon can handle the recursive splitting for us +/// let range = Range2D { rx: 0..800, ry: 0..600 }; +/// iter::split(range, split_range2).for_each(|sub_range| { +/// // If the sub-ranges were indeed split by the largest dimension, then +/// // if no dimension was twice larger than the other initially, this +/// // property will remain true in the final sub-ranges. +/// let width = sub_range.rx.end - sub_range.rx.start; +/// let height = sub_range.ry.end - sub_range.ry.start; +/// assert!((width / 2 <= height) && (height / 2 <= width)); +/// }); +/// ``` +/// +pub fn split<D, S>(data: D, splitter: S) -> Split<D, S> +where + D: Send, + S: Fn(D) -> (D, Option<D>) + Sync, +{ + Split { data, splitter } +} + +/// `Split` is a parallel iterator using arbitrary data and a splitting function. +/// This struct is created by the [`split()`] function. +/// +/// [`split()`]: fn.split.html +#[derive(Clone)] +pub struct Split<D, S> { + data: D, + splitter: S, +} + +impl<D: Debug, S> Debug for Split<D, S> { + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + f.debug_struct("Split").field("data", &self.data).finish() + } +} + +impl<D, S> ParallelIterator for Split<D, S> +where + D: Send, + S: Fn(D) -> (D, Option<D>) + Sync + Send, +{ + type Item = D; + + fn drive_unindexed<C>(self, consumer: C) -> C::Result + where + C: UnindexedConsumer<Self::Item>, + { + let producer = SplitProducer { + data: self.data, + splitter: &self.splitter, + }; + bridge_unindexed(producer, consumer) + } +} + +struct SplitProducer<'a, D, S> { + data: D, + splitter: &'a S, +} + +impl<'a, D, S> UnindexedProducer for SplitProducer<'a, D, S> +where + D: Send, + S: Fn(D) -> (D, Option<D>) + Sync, +{ + type Item = D; + + fn split(mut self) -> (Self, Option<Self>) { + let splitter = self.splitter; + let (left, right) = splitter(self.data); + self.data = left; + (self, right.map(|data| SplitProducer { data, splitter })) + } + + fn fold_with<F>(self, folder: F) -> F + where + F: Folder<Self::Item>, + { + folder.consume(self.data) + } +} diff --git a/third_party/rust/rayon/src/iter/step_by.rs b/third_party/rust/rayon/src/iter/step_by.rs new file mode 100644 index 0000000000..94b8334e86 --- /dev/null +++ b/third_party/rust/rayon/src/iter/step_by.rs @@ -0,0 +1,143 @@ +use std::cmp::min; + +use super::plumbing::*; +use super::*; +use crate::math::div_round_up; +use std::iter; +use std::usize; + +/// `StepBy` is an iterator that skips `n` elements between each yield, where `n` is the given step. +/// This struct is created by the [`step_by()`] method on [`IndexedParallelIterator`] +/// +/// [`step_by()`]: trait.IndexedParallelIterator.html#method.step_by +/// [`IndexedParallelIterator`]: trait.IndexedParallelIterator.html +#[must_use = "iterator adaptors are lazy and do nothing unless consumed"] +#[derive(Debug, Clone)] +pub struct StepBy<I: IndexedParallelIterator> { + base: I, + step: usize, +} + +impl<I> StepBy<I> +where + I: IndexedParallelIterator, +{ + /// Creates a new `StepBy` iterator. + pub(super) fn new(base: I, step: usize) -> Self { + StepBy { base, step } + } +} + +impl<I> ParallelIterator for StepBy<I> +where + I: IndexedParallelIterator, +{ + type Item = I::Item; + + fn drive_unindexed<C>(self, consumer: C) -> C::Result + where + C: UnindexedConsumer<Self::Item>, + { + bridge(self, consumer) + } + + fn opt_len(&self) -> Option<usize> { + Some(self.len()) + } +} + +impl<I> IndexedParallelIterator for StepBy<I> +where + I: IndexedParallelIterator, +{ + fn drive<C: Consumer<Self::Item>>(self, consumer: C) -> C::Result { + bridge(self, consumer) + } + + fn len(&self) -> usize { + div_round_up(self.base.len(), self.step) + } + + fn with_producer<CB>(self, callback: CB) -> CB::Output + where + CB: ProducerCallback<Self::Item>, + { + let len = self.base.len(); + return self.base.with_producer(Callback { + callback, + step: self.step, + len, + }); + + struct Callback<CB> { + callback: CB, + step: usize, + len: usize, + } + + impl<T, CB> ProducerCallback<T> for Callback<CB> + where + CB: ProducerCallback<T>, + { + type Output = CB::Output; + fn callback<P>(self, base: P) -> CB::Output + where + P: Producer<Item = T>, + { + let producer = StepByProducer { + base, + step: self.step, + len: self.len, + }; + self.callback.callback(producer) + } + } + } +} + +/// //////////////////////////////////////////////////////////////////////// +/// Producer implementation + +struct StepByProducer<P> { + base: P, + step: usize, + len: usize, +} + +impl<P> Producer for StepByProducer<P> +where + P: Producer, +{ + type Item = P::Item; + type IntoIter = iter::StepBy<P::IntoIter>; + + fn into_iter(self) -> Self::IntoIter { + self.base.into_iter().step_by(self.step) + } + + fn split_at(self, index: usize) -> (Self, Self) { + let elem_index = min(index * self.step, self.len); + + let (left, right) = self.base.split_at(elem_index); + ( + StepByProducer { + base: left, + step: self.step, + len: elem_index, + }, + StepByProducer { + base: right, + step: self.step, + len: self.len - elem_index, + }, + ) + } + + fn min_len(&self) -> usize { + div_round_up(self.base.min_len(), self.step) + } + + fn max_len(&self) -> usize { + self.base.max_len() / self.step + } +} diff --git a/third_party/rust/rayon/src/iter/sum.rs b/third_party/rust/rayon/src/iter/sum.rs new file mode 100644 index 0000000000..a73e0bf3d7 --- /dev/null +++ b/third_party/rust/rayon/src/iter/sum.rs @@ -0,0 +1,110 @@ +use super::plumbing::*; +use super::ParallelIterator; + +use std::iter::{self, Sum}; +use std::marker::PhantomData; + +pub(super) fn sum<PI, S>(pi: PI) -> S +where + PI: ParallelIterator, + S: Send + Sum<PI::Item> + Sum, +{ + pi.drive_unindexed(SumConsumer::new()) +} + +fn add<T: Sum>(left: T, right: T) -> T { + iter::once(left).chain(iter::once(right)).sum() +} + +struct SumConsumer<S: Send> { + _marker: PhantomData<*const S>, +} + +unsafe impl<S: Send> Send for SumConsumer<S> {} + +impl<S: Send> SumConsumer<S> { + fn new() -> SumConsumer<S> { + SumConsumer { + _marker: PhantomData, + } + } +} + +impl<S, T> Consumer<T> for SumConsumer<S> +where + S: Send + Sum<T> + Sum, +{ + type Folder = SumFolder<S>; + type Reducer = Self; + type Result = S; + + fn split_at(self, _index: usize) -> (Self, Self, Self) { + (SumConsumer::new(), SumConsumer::new(), SumConsumer::new()) + } + + fn into_folder(self) -> Self::Folder { + SumFolder { + sum: iter::empty::<T>().sum(), + } + } + + fn full(&self) -> bool { + false + } +} + +impl<S, T> UnindexedConsumer<T> for SumConsumer<S> +where + S: Send + Sum<T> + Sum, +{ + fn split_off_left(&self) -> Self { + SumConsumer::new() + } + + fn to_reducer(&self) -> Self::Reducer { + SumConsumer::new() + } +} + +impl<S> Reducer<S> for SumConsumer<S> +where + S: Send + Sum, +{ + fn reduce(self, left: S, right: S) -> S { + add(left, right) + } +} + +struct SumFolder<S> { + sum: S, +} + +impl<S, T> Folder<T> for SumFolder<S> +where + S: Sum<T> + Sum, +{ + type Result = S; + + fn consume(self, item: T) -> Self { + SumFolder { + sum: add(self.sum, iter::once(item).sum()), + } + } + + fn consume_iter<I>(self, iter: I) -> Self + where + I: IntoIterator<Item = T>, + { + SumFolder { + sum: add(self.sum, iter.into_iter().sum()), + } + } + + fn complete(self) -> S { + self.sum + } + + fn full(&self) -> bool { + false + } +} diff --git a/third_party/rust/rayon/src/iter/take.rs b/third_party/rust/rayon/src/iter/take.rs new file mode 100644 index 0000000000..52d15d805c --- /dev/null +++ b/third_party/rust/rayon/src/iter/take.rs @@ -0,0 +1,86 @@ +use super::plumbing::*; +use super::*; +use std::cmp::min; + +/// `Take` is an iterator that iterates over the first `n` elements. +/// This struct is created by the [`take()`] method on [`IndexedParallelIterator`] +/// +/// [`take()`]: trait.IndexedParallelIterator.html#method.take +/// [`IndexedParallelIterator`]: trait.IndexedParallelIterator.html +#[must_use = "iterator adaptors are lazy and do nothing unless consumed"] +#[derive(Debug, Clone)] +pub struct Take<I> { + base: I, + n: usize, +} + +impl<I> Take<I> +where + I: IndexedParallelIterator, +{ + /// Creates a new `Take` iterator. + pub(super) fn new(base: I, n: usize) -> Self { + let n = min(base.len(), n); + Take { base, n } + } +} + +impl<I> ParallelIterator for Take<I> +where + I: IndexedParallelIterator, +{ + type Item = I::Item; + + fn drive_unindexed<C>(self, consumer: C) -> C::Result + where + C: UnindexedConsumer<Self::Item>, + { + bridge(self, consumer) + } + + fn opt_len(&self) -> Option<usize> { + Some(self.len()) + } +} + +impl<I> IndexedParallelIterator for Take<I> +where + I: IndexedParallelIterator, +{ + fn len(&self) -> usize { + self.n + } + + fn drive<C: Consumer<Self::Item>>(self, consumer: C) -> C::Result { + bridge(self, consumer) + } + + fn with_producer<CB>(self, callback: CB) -> CB::Output + where + CB: ProducerCallback<Self::Item>, + { + return self.base.with_producer(Callback { + callback, + n: self.n, + }); + + struct Callback<CB> { + callback: CB, + n: usize, + } + + impl<T, CB> ProducerCallback<T> for Callback<CB> + where + CB: ProducerCallback<T>, + { + type Output = CB::Output; + fn callback<P>(self, base: P) -> CB::Output + where + P: Producer<Item = T>, + { + let (producer, _) = base.split_at(self.n); + self.callback.callback(producer) + } + } + } +} diff --git a/third_party/rust/rayon/src/iter/test.rs b/third_party/rust/rayon/src/iter/test.rs new file mode 100644 index 0000000000..94323d79df --- /dev/null +++ b/third_party/rust/rayon/src/iter/test.rs @@ -0,0 +1,2185 @@ +use std::sync::atomic::{AtomicUsize, Ordering}; + +use super::*; +use crate::prelude::*; +use rayon_core::*; + +use rand::distributions::Standard; +use rand::{Rng, SeedableRng}; +use rand_xorshift::XorShiftRng; +use std::collections::LinkedList; +use std::collections::{BTreeMap, BTreeSet, HashMap, HashSet}; +use std::collections::{BinaryHeap, VecDeque}; +use std::f64; +use std::fmt::Debug; +use std::sync::mpsc; +use std::usize; + +fn is_indexed<T: IndexedParallelIterator>(_: T) {} + +fn seeded_rng() -> XorShiftRng { + let mut seed = <XorShiftRng as SeedableRng>::Seed::default(); + (0..).zip(seed.as_mut()).for_each(|(i, x)| *x = i); + XorShiftRng::from_seed(seed) +} + +#[test] +fn execute() { + let a: Vec<i32> = (0..1024).collect(); + let mut b = vec![]; + a.par_iter().map(|&i| i + 1).collect_into_vec(&mut b); + let c: Vec<i32> = (0..1024).map(|i| i + 1).collect(); + assert_eq!(b, c); +} + +#[test] +fn execute_cloned() { + let a: Vec<i32> = (0..1024).collect(); + let mut b: Vec<i32> = vec![]; + a.par_iter().cloned().collect_into_vec(&mut b); + let c: Vec<i32> = (0..1024).collect(); + assert_eq!(b, c); +} + +#[test] +fn execute_range() { + let a = 0i32..1024; + let mut b = vec![]; + a.into_par_iter().map(|i| i + 1).collect_into_vec(&mut b); + let c: Vec<i32> = (0..1024).map(|i| i + 1).collect(); + assert_eq!(b, c); +} + +#[test] +fn execute_unindexed_range() { + let a = 0i64..1024; + let b: LinkedList<i64> = a.into_par_iter().map(|i| i + 1).collect(); + let c: LinkedList<i64> = (0..1024).map(|i| i + 1).collect(); + assert_eq!(b, c); +} + +#[test] +fn execute_pseudo_indexed_range() { + use std::i128::MAX; + let range = MAX - 1024..MAX; + + // Given `Some` length, collecting `Vec` will try to act indexed. + let a = range.clone().into_par_iter(); + assert_eq!(a.opt_len(), Some(1024)); + + let b: Vec<i128> = a.map(|i| i + 1).collect(); + let c: Vec<i128> = range.map(|i| i + 1).collect(); + assert_eq!(b, c); +} + +#[test] +fn check_map_indexed() { + let a = [1, 2, 3]; + is_indexed(a.par_iter().map(|x| x)); +} + +#[test] +fn map_sum() { + let a: Vec<i32> = (0..1024).collect(); + let r1: i32 = a.par_iter().map(|&i| i + 1).sum(); + let r2 = a.iter().map(|&i| i + 1).sum(); + assert_eq!(r1, r2); +} + +#[test] +fn map_reduce() { + let a: Vec<i32> = (0..1024).collect(); + let r1 = a.par_iter().map(|&i| i + 1).reduce(|| 0, |i, j| i + j); + let r2 = a.iter().map(|&i| i + 1).sum(); + assert_eq!(r1, r2); +} + +#[test] +fn map_reduce_with() { + let a: Vec<i32> = (0..1024).collect(); + let r1 = a.par_iter().map(|&i| i + 1).reduce_with(|i, j| i + j); + let r2 = a.iter().map(|&i| i + 1).sum(); + assert_eq!(r1, Some(r2)); +} + +#[test] +fn fold_map_reduce() { + // Kind of a weird test, but it demonstrates various + // transformations that are taking place. Relies on + // `with_max_len(1).fold()` being equivalent to `map()`. + // + // Take each number from 0 to 32 and fold them by appending to a + // vector. Because of `with_max_len(1)`, this will produce 32 vectors, + // each with one item. We then collect all of these into an + // individual vector by mapping each into their own vector (so we + // have Vec<Vec<i32>>) and then reducing those into a single + // vector. + let r1 = (0_i32..32) + .into_par_iter() + .with_max_len(1) + .fold(Vec::new, |mut v, e| { + v.push(e); + v + }) + .map(|v| vec![v]) + .reduce_with(|mut v_a, v_b| { + v_a.extend(v_b); + v_a + }); + assert_eq!( + r1, + Some(vec![ + vec![0], + vec![1], + vec![2], + vec![3], + vec![4], + vec![5], + vec![6], + vec![7], + vec![8], + vec![9], + vec![10], + vec![11], + vec![12], + vec![13], + vec![14], + vec![15], + vec![16], + vec![17], + vec![18], + vec![19], + vec![20], + vec![21], + vec![22], + vec![23], + vec![24], + vec![25], + vec![26], + vec![27], + vec![28], + vec![29], + vec![30], + vec![31] + ]) + ); +} + +#[test] +fn fold_is_full() { + let counter = AtomicUsize::new(0); + let a = (0_i32..2048) + .into_par_iter() + .inspect(|_| { + counter.fetch_add(1, Ordering::SeqCst); + }) + .fold(|| 0, |a, b| a + b) + .find_any(|_| true); + assert!(a.is_some()); + assert!(counter.load(Ordering::SeqCst) < 2048); // should not have visited every single one +} + +#[test] +fn check_step_by() { + let a: Vec<i32> = (0..1024).step_by(2).collect(); + let b: Vec<i32> = (0..1024).into_par_iter().step_by(2).collect(); + + assert_eq!(a, b); +} + +#[test] +fn check_step_by_unaligned() { + let a: Vec<i32> = (0..1029).step_by(10).collect(); + let b: Vec<i32> = (0..1029).into_par_iter().step_by(10).collect(); + + assert_eq!(a, b) +} + +#[test] +fn check_step_by_rev() { + let a: Vec<i32> = (0..1024).step_by(2).rev().collect(); + let b: Vec<i32> = (0..1024).into_par_iter().step_by(2).rev().collect(); + + assert_eq!(a, b); +} + +#[test] +fn check_enumerate() { + let a: Vec<usize> = (0..1024).rev().collect(); + + let mut b = vec![]; + a.par_iter() + .enumerate() + .map(|(i, &x)| i + x) + .collect_into_vec(&mut b); + assert!(b.iter().all(|&x| x == a.len() - 1)); +} + +#[test] +fn check_enumerate_rev() { + let a: Vec<usize> = (0..1024).rev().collect(); + + let mut b = vec![]; + a.par_iter() + .enumerate() + .rev() + .map(|(i, &x)| i + x) + .collect_into_vec(&mut b); + assert!(b.iter().all(|&x| x == a.len() - 1)); +} + +#[test] +fn check_indices_after_enumerate_split() { + let a: Vec<i32> = (0..1024).collect(); + a.par_iter().enumerate().with_producer(WithProducer); + + struct WithProducer; + impl<'a> ProducerCallback<(usize, &'a i32)> for WithProducer { + type Output = (); + fn callback<P>(self, producer: P) + where + P: Producer<Item = (usize, &'a i32)>, + { + let (a, b) = producer.split_at(512); + for ((index, value), trusted_index) in a.into_iter().zip(0..) { + assert_eq!(index, trusted_index); + assert_eq!(index, *value as usize); + } + for ((index, value), trusted_index) in b.into_iter().zip(512..) { + assert_eq!(index, trusted_index); + assert_eq!(index, *value as usize); + } + } + } +} + +#[test] +fn check_increment() { + let mut a: Vec<usize> = (0..1024).rev().collect(); + + a.par_iter_mut().enumerate().for_each(|(i, v)| *v += i); + + assert!(a.iter().all(|&x| x == a.len() - 1)); +} + +#[test] +fn check_skip() { + let a: Vec<usize> = (0..1024).collect(); + + let mut v1 = Vec::new(); + a.par_iter().skip(16).collect_into_vec(&mut v1); + let v2 = a.iter().skip(16).collect::<Vec<_>>(); + assert_eq!(v1, v2); + + let mut v1 = Vec::new(); + a.par_iter().skip(2048).collect_into_vec(&mut v1); + let v2 = a.iter().skip(2048).collect::<Vec<_>>(); + assert_eq!(v1, v2); + + let mut v1 = Vec::new(); + a.par_iter().skip(0).collect_into_vec(&mut v1); + let v2 = a.iter().skip(0).collect::<Vec<_>>(); + assert_eq!(v1, v2); + + // Check that the skipped elements side effects are executed + use std::sync::atomic::{AtomicUsize, Ordering}; + let num = AtomicUsize::new(0); + a.par_iter() + .map(|&n| num.fetch_add(n, Ordering::Relaxed)) + .skip(512) + .count(); + assert_eq!(num.load(Ordering::Relaxed), a.iter().sum::<usize>()); +} + +#[test] +fn check_take() { + let a: Vec<usize> = (0..1024).collect(); + + let mut v1 = Vec::new(); + a.par_iter().take(16).collect_into_vec(&mut v1); + let v2 = a.iter().take(16).collect::<Vec<_>>(); + assert_eq!(v1, v2); + + let mut v1 = Vec::new(); + a.par_iter().take(2048).collect_into_vec(&mut v1); + let v2 = a.iter().take(2048).collect::<Vec<_>>(); + assert_eq!(v1, v2); + + let mut v1 = Vec::new(); + a.par_iter().take(0).collect_into_vec(&mut v1); + let v2 = a.iter().take(0).collect::<Vec<_>>(); + assert_eq!(v1, v2); +} + +#[test] +fn check_inspect() { + use std::sync::atomic::{AtomicUsize, Ordering}; + + let a = AtomicUsize::new(0); + let b: usize = (0_usize..1024) + .into_par_iter() + .inspect(|&i| { + a.fetch_add(i, Ordering::Relaxed); + }) + .sum(); + + assert_eq!(a.load(Ordering::Relaxed), b); +} + +#[test] +fn check_move() { + let a = vec![vec![1, 2, 3]]; + let ptr = a[0].as_ptr(); + + let mut b = vec![]; + a.into_par_iter().collect_into_vec(&mut b); + + // a simple move means the inner vec will be completely unchanged + assert_eq!(ptr, b[0].as_ptr()); +} + +#[test] +fn check_drops() { + use std::sync::atomic::{AtomicUsize, Ordering}; + + let c = AtomicUsize::new(0); + let a = vec![DropCounter(&c); 10]; + + let mut b = vec![]; + a.clone().into_par_iter().collect_into_vec(&mut b); + assert_eq!(c.load(Ordering::Relaxed), 0); + + b.into_par_iter(); + assert_eq!(c.load(Ordering::Relaxed), 10); + + a.into_par_iter().with_producer(Partial); + assert_eq!(c.load(Ordering::Relaxed), 20); + + #[derive(Clone)] + struct DropCounter<'a>(&'a AtomicUsize); + impl<'a> Drop for DropCounter<'a> { + fn drop(&mut self) { + self.0.fetch_add(1, Ordering::Relaxed); + } + } + + struct Partial; + impl<'a> ProducerCallback<DropCounter<'a>> for Partial { + type Output = (); + fn callback<P>(self, producer: P) + where + P: Producer<Item = DropCounter<'a>>, + { + let (a, _) = producer.split_at(5); + a.into_iter().next(); + } + } +} + +#[test] +fn check_slice_indexed() { + let a = vec![1, 2, 3]; + is_indexed(a.par_iter()); +} + +#[test] +fn check_slice_mut_indexed() { + let mut a = vec![1, 2, 3]; + is_indexed(a.par_iter_mut()); +} + +#[test] +fn check_vec_indexed() { + let a = vec![1, 2, 3]; + is_indexed(a.into_par_iter()); +} + +#[test] +fn check_range_indexed() { + is_indexed((1..5).into_par_iter()); +} + +#[test] +fn check_cmp_direct() { + let a = (0..1024).into_par_iter(); + let b = (0..1024).into_par_iter(); + + let result = a.cmp(b); + + assert!(result == ::std::cmp::Ordering::Equal); +} + +#[test] +fn check_cmp_to_seq() { + assert_eq!( + (0..1024).into_par_iter().cmp(0..1024), + (0..1024).cmp(0..1024) + ); +} + +#[test] +fn check_cmp_rng_to_seq() { + let mut rng = seeded_rng(); + let rng = &mut rng; + let a: Vec<i32> = rng.sample_iter(&Standard).take(1024).collect(); + let b: Vec<i32> = rng.sample_iter(&Standard).take(1024).collect(); + for i in 0..a.len() { + let par_result = a[i..].par_iter().cmp(b[i..].par_iter()); + let seq_result = a[i..].iter().cmp(b[i..].iter()); + + assert_eq!(par_result, seq_result); + } +} + +#[test] +fn check_cmp_lt_direct() { + let a = (0..1024).into_par_iter(); + let b = (1..1024).into_par_iter(); + + let result = a.cmp(b); + + assert!(result == ::std::cmp::Ordering::Less); +} + +#[test] +fn check_cmp_lt_to_seq() { + assert_eq!( + (0..1024).into_par_iter().cmp(1..1024), + (0..1024).cmp(1..1024) + ) +} + +#[test] +fn check_cmp_gt_direct() { + let a = (1..1024).into_par_iter(); + let b = (0..1024).into_par_iter(); + + let result = a.cmp(b); + + assert!(result == ::std::cmp::Ordering::Greater); +} + +#[test] +fn check_cmp_gt_to_seq() { + assert_eq!( + (1..1024).into_par_iter().cmp(0..1024), + (1..1024).cmp(0..1024) + ) +} + +#[test] +fn check_cmp_short_circuit() { + // We only use a single thread in order to make the short-circuit behavior deterministic. + let pool = ThreadPoolBuilder::new().num_threads(1).build().unwrap(); + + let a = vec![0; 1024]; + let mut b = a.clone(); + b[42] = 1; + + pool.install(|| { + let expected = ::std::cmp::Ordering::Less; + assert_eq!(a.par_iter().cmp(&b), expected); + + for len in 1..10 { + let counter = AtomicUsize::new(0); + let result = a + .par_iter() + .with_max_len(len) + .inspect(|_| { + counter.fetch_add(1, Ordering::SeqCst); + }) + .cmp(&b); + assert_eq!(result, expected); + // should not have visited every single one + assert!(counter.into_inner() < a.len()); + } + }); +} + +#[test] +fn check_partial_cmp_short_circuit() { + // We only use a single thread to make the short-circuit behavior deterministic. + let pool = ThreadPoolBuilder::new().num_threads(1).build().unwrap(); + + let a = vec![0; 1024]; + let mut b = a.clone(); + b[42] = 1; + + pool.install(|| { + let expected = Some(::std::cmp::Ordering::Less); + assert_eq!(a.par_iter().partial_cmp(&b), expected); + + for len in 1..10 { + let counter = AtomicUsize::new(0); + let result = a + .par_iter() + .with_max_len(len) + .inspect(|_| { + counter.fetch_add(1, Ordering::SeqCst); + }) + .partial_cmp(&b); + assert_eq!(result, expected); + // should not have visited every single one + assert!(counter.into_inner() < a.len()); + } + }); +} + +#[test] +fn check_partial_cmp_nan_short_circuit() { + // We only use a single thread to make the short-circuit behavior deterministic. + let pool = ThreadPoolBuilder::new().num_threads(1).build().unwrap(); + + let a = vec![0.0; 1024]; + let mut b = a.clone(); + b[42] = f64::NAN; + + pool.install(|| { + let expected = None; + assert_eq!(a.par_iter().partial_cmp(&b), expected); + + for len in 1..10 { + let counter = AtomicUsize::new(0); + let result = a + .par_iter() + .with_max_len(len) + .inspect(|_| { + counter.fetch_add(1, Ordering::SeqCst); + }) + .partial_cmp(&b); + assert_eq!(result, expected); + // should not have visited every single one + assert!(counter.into_inner() < a.len()); + } + }); +} + +#[test] +fn check_partial_cmp_direct() { + let a = (0..1024).into_par_iter(); + let b = (0..1024).into_par_iter(); + + let result = a.partial_cmp(b); + + assert!(result == Some(::std::cmp::Ordering::Equal)); +} + +#[test] +fn check_partial_cmp_to_seq() { + let par_result = (0..1024).into_par_iter().partial_cmp(0..1024); + let seq_result = (0..1024).partial_cmp(0..1024); + assert_eq!(par_result, seq_result); +} + +#[test] +fn check_partial_cmp_rng_to_seq() { + let mut rng = seeded_rng(); + let rng = &mut rng; + let a: Vec<i32> = rng.sample_iter(&Standard).take(1024).collect(); + let b: Vec<i32> = rng.sample_iter(&Standard).take(1024).collect(); + for i in 0..a.len() { + let par_result = a[i..].par_iter().partial_cmp(b[i..].par_iter()); + let seq_result = a[i..].iter().partial_cmp(b[i..].iter()); + + assert_eq!(par_result, seq_result); + } +} + +#[test] +fn check_partial_cmp_lt_direct() { + let a = (0..1024).into_par_iter(); + let b = (1..1024).into_par_iter(); + + let result = a.partial_cmp(b); + + assert!(result == Some(::std::cmp::Ordering::Less)); +} + +#[test] +fn check_partial_cmp_lt_to_seq() { + let par_result = (0..1024).into_par_iter().partial_cmp(1..1024); + let seq_result = (0..1024).partial_cmp(1..1024); + assert_eq!(par_result, seq_result); +} + +#[test] +fn check_partial_cmp_gt_direct() { + let a = (1..1024).into_par_iter(); + let b = (0..1024).into_par_iter(); + + let result = a.partial_cmp(b); + + assert!(result == Some(::std::cmp::Ordering::Greater)); +} + +#[test] +fn check_partial_cmp_gt_to_seq() { + let par_result = (1..1024).into_par_iter().partial_cmp(0..1024); + let seq_result = (1..1024).partial_cmp(0..1024); + assert_eq!(par_result, seq_result); +} + +#[test] +fn check_partial_cmp_none_direct() { + let a = vec![f64::NAN, 0.0]; + let b = vec![0.0, 1.0]; + + let result = a.par_iter().partial_cmp(b.par_iter()); + + assert!(result == None); +} + +#[test] +fn check_partial_cmp_none_to_seq() { + let a = vec![f64::NAN, 0.0]; + let b = vec![0.0, 1.0]; + + let par_result = a.par_iter().partial_cmp(b.par_iter()); + let seq_result = a.iter().partial_cmp(b.iter()); + + assert_eq!(par_result, seq_result); +} + +#[test] +fn check_partial_cmp_late_nan_direct() { + let a = vec![0.0, f64::NAN]; + let b = vec![1.0, 1.0]; + + let result = a.par_iter().partial_cmp(b.par_iter()); + + assert!(result == Some(::std::cmp::Ordering::Less)); +} + +#[test] +fn check_partial_cmp_late_nane_to_seq() { + let a = vec![0.0, f64::NAN]; + let b = vec![1.0, 1.0]; + + let par_result = a.par_iter().partial_cmp(b.par_iter()); + let seq_result = a.iter().partial_cmp(b.iter()); + + assert_eq!(par_result, seq_result); +} + +#[test] +fn check_cmp_lengths() { + // comparisons should consider length if they are otherwise equal + let a = vec![0; 1024]; + let b = vec![0; 1025]; + + assert_eq!(a.par_iter().cmp(&b), a.iter().cmp(&b)); + assert_eq!(a.par_iter().partial_cmp(&b), a.iter().partial_cmp(&b)); +} + +#[test] +fn check_eq_direct() { + let a = (0..1024).into_par_iter(); + let b = (0..1024).into_par_iter(); + + let result = a.eq(b); + + assert!(result); +} + +#[test] +fn check_eq_to_seq() { + let par_result = (0..1024).into_par_iter().eq((0..1024).into_par_iter()); + let seq_result = (0..1024).eq(0..1024); + + assert_eq!(par_result, seq_result); +} + +#[test] +fn check_ne_direct() { + let a = (0..1024).into_par_iter(); + let b = (1..1024).into_par_iter(); + + let result = a.ne(b); + + assert!(result); +} + +#[test] +fn check_ne_to_seq() { + let par_result = (0..1024).into_par_iter().ne((1..1025).into_par_iter()); + let seq_result = (0..1024).ne(1..1025); + + assert_eq!(par_result, seq_result); +} + +#[test] +fn check_ne_lengths() { + // equality should consider length too + let a = vec![0; 1024]; + let b = vec![0; 1025]; + + assert_eq!(a.par_iter().eq(&b), a.iter().eq(&b)); + assert_eq!(a.par_iter().ne(&b), a.iter().ne(&b)); +} + +#[test] +fn check_lt_direct() { + assert!((0..1024).into_par_iter().lt(1..1024)); + assert!(!(1..1024).into_par_iter().lt(0..1024)); +} + +#[test] +fn check_lt_to_seq() { + let par_result = (0..1024).into_par_iter().lt((1..1024).into_par_iter()); + let seq_result = (0..1024).lt(1..1024); + + assert_eq!(par_result, seq_result); +} + +#[test] +fn check_le_equal_direct() { + assert!((0..1024).into_par_iter().le((0..1024).into_par_iter())); +} + +#[test] +fn check_le_equal_to_seq() { + let par_result = (0..1024).into_par_iter().le((0..1024).into_par_iter()); + let seq_result = (0..1024).le(0..1024); + + assert_eq!(par_result, seq_result); +} + +#[test] +fn check_le_less_direct() { + assert!((0..1024).into_par_iter().le((1..1024).into_par_iter())); +} + +#[test] +fn check_le_less_to_seq() { + let par_result = (0..1024).into_par_iter().le((1..1024).into_par_iter()); + let seq_result = (0..1024).le(1..1024); + + assert_eq!(par_result, seq_result); +} + +#[test] +fn check_gt_direct() { + assert!((1..1024).into_par_iter().gt((0..1024).into_par_iter())); +} + +#[test] +fn check_gt_to_seq() { + let par_result = (1..1024).into_par_iter().gt((0..1024).into_par_iter()); + let seq_result = (1..1024).gt(0..1024); + + assert_eq!(par_result, seq_result); +} + +#[test] +fn check_ge_equal_direct() { + assert!((0..1024).into_par_iter().ge((0..1024).into_par_iter())); +} + +#[test] +fn check_ge_equal_to_seq() { + let par_result = (0..1024).into_par_iter().ge((0..1024).into_par_iter()); + let seq_result = (0..1024).ge(0..1024); + + assert_eq!(par_result, seq_result); +} + +#[test] +fn check_ge_greater_direct() { + assert!((1..1024).into_par_iter().ge((0..1024).into_par_iter())); +} + +#[test] +fn check_ge_greater_to_seq() { + let par_result = (1..1024).into_par_iter().ge((0..1024).into_par_iter()); + let seq_result = (1..1024).ge(0..1024); + + assert_eq!(par_result, seq_result); +} + +#[test] +fn check_zip() { + let mut a: Vec<usize> = (0..1024).rev().collect(); + let b: Vec<usize> = (0..1024).collect(); + + a.par_iter_mut().zip(&b[..]).for_each(|(a, &b)| *a += b); + + assert!(a.iter().all(|&x| x == a.len() - 1)); +} + +#[test] +fn check_zip_into_par_iter() { + let mut a: Vec<usize> = (0..1024).rev().collect(); + let b: Vec<usize> = (0..1024).collect(); + + a.par_iter_mut() + .zip(&b) // here we rely on &b iterating over &usize + .for_each(|(a, &b)| *a += b); + + assert!(a.iter().all(|&x| x == a.len() - 1)); +} + +#[test] +fn check_zip_into_mut_par_iter() { + let a: Vec<usize> = (0..1024).rev().collect(); + let mut b: Vec<usize> = (0..1024).collect(); + + a.par_iter().zip(&mut b).for_each(|(&a, b)| *b += a); + + assert!(b.iter().all(|&x| x == b.len() - 1)); +} + +#[test] +fn check_zip_range() { + let mut a: Vec<usize> = (0..1024).rev().collect(); + + a.par_iter_mut() + .zip(0usize..1024) + .for_each(|(a, b)| *a += b); + + assert!(a.iter().all(|&x| x == a.len() - 1)); +} + +#[test] +fn check_zip_eq() { + let mut a: Vec<usize> = (0..1024).rev().collect(); + let b: Vec<usize> = (0..1024).collect(); + + a.par_iter_mut().zip_eq(&b[..]).for_each(|(a, &b)| *a += b); + + assert!(a.iter().all(|&x| x == a.len() - 1)); +} + +#[test] +fn check_zip_eq_into_par_iter() { + let mut a: Vec<usize> = (0..1024).rev().collect(); + let b: Vec<usize> = (0..1024).collect(); + + a.par_iter_mut() + .zip_eq(&b) // here we rely on &b iterating over &usize + .for_each(|(a, &b)| *a += b); + + assert!(a.iter().all(|&x| x == a.len() - 1)); +} + +#[test] +fn check_zip_eq_into_mut_par_iter() { + let a: Vec<usize> = (0..1024).rev().collect(); + let mut b: Vec<usize> = (0..1024).collect(); + + a.par_iter().zip_eq(&mut b).for_each(|(&a, b)| *b += a); + + assert!(b.iter().all(|&x| x == b.len() - 1)); +} + +#[test] +fn check_zip_eq_range() { + let mut a: Vec<usize> = (0..1024).rev().collect(); + + a.par_iter_mut() + .zip_eq(0usize..1024) + .for_each(|(a, b)| *a += b); + + assert!(a.iter().all(|&x| x == a.len() - 1)); +} + +#[test] +fn check_sum_filtered_ints() { + let a: Vec<i32> = vec![1, 2, 3, 4, 5, 6, 7, 8, 9, 10]; + let par_sum_evens: i32 = a.par_iter().filter(|&x| (x & 1) == 0).sum(); + let seq_sum_evens = a.iter().filter(|&x| (x & 1) == 0).sum(); + assert_eq!(par_sum_evens, seq_sum_evens); +} + +#[test] +fn check_sum_filtermap_ints() { + let a: Vec<i32> = vec![1, 2, 3, 4, 5, 6, 7, 8, 9, 10]; + let par_sum_evens: u32 = a + .par_iter() + .filter_map(|&x| if (x & 1) == 0 { Some(x as u32) } else { None }) + .sum(); + let seq_sum_evens = a + .iter() + .filter_map(|&x| if (x & 1) == 0 { Some(x as u32) } else { None }) + .sum(); + assert_eq!(par_sum_evens, seq_sum_evens); +} + +#[test] +fn check_flat_map_nested_ranges() { + // FIXME -- why are precise type hints required on the integers here? + + let v: i32 = (0_i32..10) + .into_par_iter() + .flat_map(|i| (0_i32..10).into_par_iter().map(move |j| (i, j))) + .map(|(i, j)| i * j) + .sum(); + + let w = (0_i32..10) + .flat_map(|i| (0_i32..10).map(move |j| (i, j))) + .map(|(i, j)| i * j) + .sum(); + + assert_eq!(v, w); +} + +#[test] +fn check_empty_flat_map_sum() { + let a: Vec<i32> = (0..1024).collect(); + let empty = &a[..0]; + + // empty on the inside + let b: i32 = a.par_iter().flat_map(|_| empty).sum(); + assert_eq!(b, 0); + + // empty on the outside + let c: i32 = empty.par_iter().flat_map(|_| a.par_iter()).sum(); + assert_eq!(c, 0); +} + +#[test] +fn check_flatten_vec() { + let a: Vec<i32> = (0..1024).collect(); + let b: Vec<Vec<i32>> = vec![a.clone(), a.clone(), a.clone(), a.clone()]; + let c: Vec<i32> = b.par_iter().flatten().cloned().collect(); + let mut d = a.clone(); + d.extend(&a); + d.extend(&a); + d.extend(&a); + + assert_eq!(d, c); +} + +#[test] +fn check_flatten_vec_empty() { + let a: Vec<Vec<i32>> = vec![vec![]]; + let b: Vec<i32> = a.par_iter().flatten().cloned().collect(); + + assert_eq!(vec![] as Vec<i32>, b); +} + +#[test] +fn check_slice_split() { + let v: Vec<_> = (0..1000).collect(); + for m in 1..100 { + let a: Vec<_> = v.split(|x| x % m == 0).collect(); + let b: Vec<_> = v.par_split(|x| x % m == 0).collect(); + assert_eq!(a, b); + } + + // same as std::slice::split() examples + let slice = [10, 40, 33, 20]; + let v: Vec<_> = slice.par_split(|num| num % 3 == 0).collect(); + assert_eq!(v, &[&slice[..2], &slice[3..]]); + + let slice = [10, 40, 33]; + let v: Vec<_> = slice.par_split(|num| num % 3 == 0).collect(); + assert_eq!(v, &[&slice[..2], &slice[..0]]); + + let slice = [10, 6, 33, 20]; + let v: Vec<_> = slice.par_split(|num| num % 3 == 0).collect(); + assert_eq!(v, &[&slice[..1], &slice[..0], &slice[3..]]); +} + +#[test] +fn check_slice_split_mut() { + let mut v1: Vec<_> = (0..1000).collect(); + let mut v2 = v1.clone(); + for m in 1..100 { + let a: Vec<_> = v1.split_mut(|x| x % m == 0).collect(); + let b: Vec<_> = v2.par_split_mut(|x| x % m == 0).collect(); + assert_eq!(a, b); + } + + // same as std::slice::split_mut() example + let mut v = [10, 40, 30, 20, 60, 50]; + v.par_split_mut(|num| num % 3 == 0).for_each(|group| { + group[0] = 1; + }); + assert_eq!(v, [1, 40, 30, 1, 60, 1]); +} + +#[test] +fn check_chunks() { + let a: Vec<i32> = vec![1, 5, 10, 4, 100, 3, 1000, 2, 10000, 1]; + let par_sum_product_pairs: i32 = a.par_chunks(2).map(|c| c.iter().product::<i32>()).sum(); + let seq_sum_product_pairs = a.chunks(2).map(|c| c.iter().product::<i32>()).sum(); + assert_eq!(par_sum_product_pairs, 12345); + assert_eq!(par_sum_product_pairs, seq_sum_product_pairs); + + let par_sum_product_triples: i32 = a.par_chunks(3).map(|c| c.iter().product::<i32>()).sum(); + let seq_sum_product_triples = a.chunks(3).map(|c| c.iter().product::<i32>()).sum(); + assert_eq!(par_sum_product_triples, 5_0 + 12_00 + 20_000_000 + 1); + assert_eq!(par_sum_product_triples, seq_sum_product_triples); +} + +#[test] +fn check_chunks_mut() { + let mut a: Vec<i32> = vec![1, 2, 3, 4, 5, 6, 7, 8, 9, 10]; + let mut b: Vec<i32> = a.clone(); + a.par_chunks_mut(2).for_each(|c| c[0] = c.iter().sum()); + b.chunks_mut(2).for_each(|c| c[0] = c.iter().sum()); + assert_eq!(a, &[3, 2, 7, 4, 11, 6, 15, 8, 19, 10]); + assert_eq!(a, b); + + let mut a: Vec<i32> = vec![1, 2, 3, 4, 5, 6, 7, 8, 9, 10]; + let mut b: Vec<i32> = a.clone(); + a.par_chunks_mut(3).for_each(|c| c[0] = c.iter().sum()); + b.chunks_mut(3).for_each(|c| c[0] = c.iter().sum()); + assert_eq!(a, &[6, 2, 3, 15, 5, 6, 24, 8, 9, 10]); + assert_eq!(a, b); +} + +#[test] +fn check_windows() { + let a: Vec<i32> = (0..1024).collect(); + let par: Vec<_> = a.par_windows(2).collect(); + let seq: Vec<_> = a.windows(2).collect(); + assert_eq!(par, seq); + + let par: Vec<_> = a.par_windows(100).collect(); + let seq: Vec<_> = a.windows(100).collect(); + assert_eq!(par, seq); + + let par: Vec<_> = a.par_windows(1_000_000).collect(); + let seq: Vec<_> = a.windows(1_000_000).collect(); + assert_eq!(par, seq); + + let par: Vec<_> = a + .par_windows(2) + .chain(a.par_windows(1_000_000)) + .zip(a.par_windows(2)) + .collect(); + let seq: Vec<_> = a + .windows(2) + .chain(a.windows(1_000_000)) + .zip(a.windows(2)) + .collect(); + assert_eq!(par, seq); +} + +#[test] +fn check_options() { + let mut a = vec![None, Some(1), None, None, Some(2), Some(4)]; + + assert_eq!(7, a.par_iter().flat_map(|opt| opt).sum::<i32>()); + assert_eq!(7, a.par_iter().flat_map(|opt| opt).sum::<i32>()); + + a.par_iter_mut() + .flat_map(|opt| opt) + .for_each(|x| *x = *x * *x); + + assert_eq!(21, a.into_par_iter().flat_map(|opt| opt).sum::<i32>()); +} + +#[test] +fn check_results() { + let mut a = vec![Err(()), Ok(1i32), Err(()), Err(()), Ok(2), Ok(4)]; + + assert_eq!(7, a.par_iter().flat_map(|res| res).sum::<i32>()); + + assert_eq!(Err::<i32, ()>(()), a.par_iter().cloned().sum()); + assert_eq!(Ok(7), a.par_iter().cloned().filter(Result::is_ok).sum()); + + assert_eq!(Err::<i32, ()>(()), a.par_iter().cloned().product()); + assert_eq!(Ok(8), a.par_iter().cloned().filter(Result::is_ok).product()); + + a.par_iter_mut() + .flat_map(|res| res) + .for_each(|x| *x = *x * *x); + + assert_eq!(21, a.into_par_iter().flat_map(|res| res).sum::<i32>()); +} + +#[test] +fn check_binary_heap() { + use std::collections::BinaryHeap; + + let a: BinaryHeap<i32> = (0..10).collect(); + + assert_eq!(45, a.par_iter().sum::<i32>()); + assert_eq!(45, a.into_par_iter().sum::<i32>()); +} + +#[test] +fn check_btree_map() { + use std::collections::BTreeMap; + + let mut a: BTreeMap<i32, i32> = (0..10).map(|i| (i, -i)).collect(); + + assert_eq!(45, a.par_iter().map(|(&k, _)| k).sum::<i32>()); + assert_eq!(-45, a.par_iter().map(|(_, &v)| v).sum::<i32>()); + + a.par_iter_mut().for_each(|(k, v)| *v += *k); + + assert_eq!(0, a.into_par_iter().map(|(_, v)| v).sum::<i32>()); +} + +#[test] +fn check_btree_set() { + use std::collections::BTreeSet; + + let a: BTreeSet<i32> = (0..10).collect(); + + assert_eq!(45, a.par_iter().sum::<i32>()); + assert_eq!(45, a.into_par_iter().sum::<i32>()); +} + +#[test] +fn check_hash_map() { + use std::collections::HashMap; + + let mut a: HashMap<i32, i32> = (0..10).map(|i| (i, -i)).collect(); + + assert_eq!(45, a.par_iter().map(|(&k, _)| k).sum::<i32>()); + assert_eq!(-45, a.par_iter().map(|(_, &v)| v).sum::<i32>()); + + a.par_iter_mut().for_each(|(k, v)| *v += *k); + + assert_eq!(0, a.into_par_iter().map(|(_, v)| v).sum::<i32>()); +} + +#[test] +fn check_hash_set() { + use std::collections::HashSet; + + let a: HashSet<i32> = (0..10).collect(); + + assert_eq!(45, a.par_iter().sum::<i32>()); + assert_eq!(45, a.into_par_iter().sum::<i32>()); +} + +#[test] +fn check_linked_list() { + use std::collections::LinkedList; + + let mut a: LinkedList<i32> = (0..10).collect(); + + assert_eq!(45, a.par_iter().sum::<i32>()); + + a.par_iter_mut().for_each(|x| *x = -*x); + + assert_eq!(-45, a.into_par_iter().sum::<i32>()); +} + +#[test] +fn check_vec_deque() { + use std::collections::VecDeque; + + let mut a: VecDeque<i32> = (0..10).collect(); + + // try to get it to wrap around + a.drain(..5); + a.extend(0..5); + + assert_eq!(45, a.par_iter().sum::<i32>()); + + a.par_iter_mut().for_each(|x| *x = -*x); + + assert_eq!(-45, a.into_par_iter().sum::<i32>()); +} + +#[test] +fn check_chain() { + let mut res = vec![]; + + // stays indexed in the face of madness + Some(0) + .into_par_iter() + .chain(Ok::<_, ()>(1)) + .chain(1..4) + .chain(Err("huh?")) + .chain(None) + .chain(vec![5, 8, 13]) + .map(|x| (x as u8 + b'a') as char) + .chain(vec!['x', 'y', 'z']) + .zip((0i32..1000).into_par_iter().map(|x| -x)) + .enumerate() + .map(|(a, (b, c))| (a, b, c)) + .chain(None) + .collect_into_vec(&mut res); + + assert_eq!( + res, + vec![ + (0, 'a', 0), + (1, 'b', -1), + (2, 'b', -2), + (3, 'c', -3), + (4, 'd', -4), + (5, 'f', -5), + (6, 'i', -6), + (7, 'n', -7), + (8, 'x', -8), + (9, 'y', -9), + (10, 'z', -10) + ] + ); + + // unindexed is ok too + let res: Vec<i32> = Some(1i32) + .into_par_iter() + .chain( + (2i32..4) + .into_par_iter() + .chain(vec![5, 6, 7, 8, 9]) + .chain(Some((10, 100)).into_par_iter().flat_map(|(a, b)| a..b)) + .filter(|x| x & 1 == 1), + ) + .collect(); + let other: Vec<i32> = (0..100).filter(|x| x & 1 == 1).collect(); + assert_eq!(res, other); + + // chain collect is ok with the "fake" specialization + let res: Vec<i32> = Some(1i32).into_par_iter().chain(None).collect(); + assert_eq!(res, &[1]); +} + +#[test] +fn check_count() { + let c0 = (0_u32..24 * 1024).filter(|i| i % 2 == 0).count(); + let c1 = (0_u32..24 * 1024) + .into_par_iter() + .filter(|i| i % 2 == 0) + .count(); + assert_eq!(c0, c1); +} + +#[test] +fn find_any() { + let a: Vec<i32> = (0..1024).collect(); + + assert!(a.par_iter().find_any(|&&x| x % 42 == 41).is_some()); + assert_eq!( + a.par_iter().find_any(|&&x| x % 19 == 1 && x % 53 == 0), + Some(&742_i32) + ); + assert_eq!(a.par_iter().find_any(|&&x| x < 0), None); + + assert!(a.par_iter().position_any(|&x| x % 42 == 41).is_some()); + assert_eq!( + a.par_iter().position_any(|&x| x % 19 == 1 && x % 53 == 0), + Some(742_usize) + ); + assert_eq!(a.par_iter().position_any(|&x| x < 0), None); + + assert!(a.par_iter().any(|&x| x > 1000)); + assert!(!a.par_iter().any(|&x| x < 0)); + + assert!(!a.par_iter().all(|&x| x > 1000)); + assert!(a.par_iter().all(|&x| x >= 0)); +} + +#[test] +fn find_first_or_last() { + let a: Vec<i32> = (0..1024).collect(); + + assert_eq!(a.par_iter().find_first(|&&x| x % 42 == 41), Some(&41_i32)); + assert_eq!( + a.par_iter().find_first(|&&x| x % 19 == 1 && x % 53 == 0), + Some(&742_i32) + ); + assert_eq!(a.par_iter().find_first(|&&x| x < 0), None); + + assert_eq!( + a.par_iter().position_first(|&x| x % 42 == 41), + Some(41_usize) + ); + assert_eq!( + a.par_iter().position_first(|&x| x % 19 == 1 && x % 53 == 0), + Some(742_usize) + ); + assert_eq!(a.par_iter().position_first(|&x| x < 0), None); + + assert_eq!(a.par_iter().find_last(|&&x| x % 42 == 41), Some(&1007_i32)); + assert_eq!( + a.par_iter().find_last(|&&x| x % 19 == 1 && x % 53 == 0), + Some(&742_i32) + ); + assert_eq!(a.par_iter().find_last(|&&x| x < 0), None); + + assert_eq!( + a.par_iter().position_last(|&x| x % 42 == 41), + Some(1007_usize) + ); + assert_eq!( + a.par_iter().position_last(|&x| x % 19 == 1 && x % 53 == 0), + Some(742_usize) + ); + assert_eq!(a.par_iter().position_last(|&x| x < 0), None); +} + +#[test] +fn find_map_first_or_last_or_any() { + let mut a: Vec<i32> = vec![]; + + assert!(a.par_iter().find_map_any(half_if_positive).is_none()); + assert!(a.par_iter().find_map_first(half_if_positive).is_none()); + assert!(a.par_iter().find_map_last(half_if_positive).is_none()); + + a = (-1024..-3).collect(); + + assert!(a.par_iter().find_map_any(half_if_positive).is_none()); + assert!(a.par_iter().find_map_first(half_if_positive).is_none()); + assert!(a.par_iter().find_map_last(half_if_positive).is_none()); + + assert!(a.par_iter().find_map_any(half_if_negative).is_some()); + assert_eq!( + a.par_iter().find_map_first(half_if_negative), + Some(-512_i32) + ); + assert_eq!(a.par_iter().find_map_last(half_if_negative), Some(-2_i32)); + + a.append(&mut (2..1025).collect()); + + assert!(a.par_iter().find_map_any(half_if_positive).is_some()); + assert_eq!(a.par_iter().find_map_first(half_if_positive), Some(1_i32)); + assert_eq!(a.par_iter().find_map_last(half_if_positive), Some(512_i32)); + + fn half_if_positive(x: &i32) -> Option<i32> { + if *x > 0 { + Some(x / 2) + } else { + None + } + } + + fn half_if_negative(x: &i32) -> Option<i32> { + if *x < 0 { + Some(x / 2) + } else { + None + } + } +} + +#[test] +fn check_find_not_present() { + let counter = AtomicUsize::new(0); + let value: Option<i32> = (0_i32..2048).into_par_iter().find_any(|&p| { + counter.fetch_add(1, Ordering::SeqCst); + p >= 2048 + }); + assert!(value.is_none()); + assert!(counter.load(Ordering::SeqCst) == 2048); // should have visited every single one +} + +#[test] +fn check_find_is_present() { + let counter = AtomicUsize::new(0); + let value: Option<i32> = (0_i32..2048).into_par_iter().find_any(|&p| { + counter.fetch_add(1, Ordering::SeqCst); + (1024..1096).contains(&p) + }); + let q = value.unwrap(); + assert!((1024..1096).contains(&q)); + assert!(counter.load(Ordering::SeqCst) < 2048); // should not have visited every single one +} + +#[test] +fn check_while_some() { + let value = (0_i32..2048).into_par_iter().map(Some).while_some().max(); + assert_eq!(value, Some(2047)); + + 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 +} + +#[test] +fn par_iter_collect_option() { + let a: Option<Vec<_>> = (0_i32..2048).map(Some).collect(); + let b: Option<Vec<_>> = (0_i32..2048).into_par_iter().map(Some).collect(); + assert_eq!(a, b); + + let c: Option<Vec<_>> = (0_i32..2048) + .into_par_iter() + .map(|x| if x == 1234 { None } else { Some(x) }) + .collect(); + assert_eq!(c, None); +} + +#[test] +fn par_iter_collect_result() { + let a: Result<Vec<_>, ()> = (0_i32..2048).map(Ok).collect(); + let b: Result<Vec<_>, ()> = (0_i32..2048).into_par_iter().map(Ok).collect(); + assert_eq!(a, b); + + let c: Result<Vec<_>, _> = (0_i32..2048) + .into_par_iter() + .map(|x| if x == 1234 { Err(x) } else { Ok(x) }) + .collect(); + assert_eq!(c, Err(1234)); + + let d: Result<Vec<_>, _> = (0_i32..2048) + .into_par_iter() + .map(|x| if x % 100 == 99 { Err(x) } else { Ok(x) }) + .collect(); + assert_eq!(d.map_err(|x| x % 100), Err(99)); +} + +#[test] +fn par_iter_collect() { + let a: Vec<i32> = (0..1024).collect(); + let b: Vec<i32> = a.par_iter().map(|&i| i + 1).collect(); + let c: Vec<i32> = (0..1024).map(|i| i + 1).collect(); + assert_eq!(b, c); +} + +#[test] +fn par_iter_collect_vecdeque() { + let a: Vec<i32> = (0..1024).collect(); + let b: VecDeque<i32> = a.par_iter().cloned().collect(); + let c: VecDeque<i32> = a.iter().cloned().collect(); + assert_eq!(b, c); +} + +#[test] +fn par_iter_collect_binaryheap() { + let a: Vec<i32> = (0..1024).collect(); + let mut b: BinaryHeap<i32> = a.par_iter().cloned().collect(); + assert_eq!(b.peek(), Some(&1023)); + assert_eq!(b.len(), 1024); + for n in (0..1024).rev() { + assert_eq!(b.pop(), Some(n)); + assert_eq!(b.len() as i32, n); + } +} + +#[test] +fn par_iter_collect_hashmap() { + let a: Vec<i32> = (0..1024).collect(); + let b: HashMap<i32, String> = a.par_iter().map(|&i| (i, format!("{}", i))).collect(); + assert_eq!(&b[&3], "3"); + assert_eq!(b.len(), 1024); +} + +#[test] +fn par_iter_collect_hashset() { + let a: Vec<i32> = (0..1024).collect(); + let b: HashSet<i32> = a.par_iter().cloned().collect(); + assert_eq!(b.len(), 1024); +} + +#[test] +fn par_iter_collect_btreemap() { + let a: Vec<i32> = (0..1024).collect(); + let b: BTreeMap<i32, String> = a.par_iter().map(|&i| (i, format!("{}", i))).collect(); + assert_eq!(&b[&3], "3"); + assert_eq!(b.len(), 1024); +} + +#[test] +fn par_iter_collect_btreeset() { + let a: Vec<i32> = (0..1024).collect(); + let b: BTreeSet<i32> = a.par_iter().cloned().collect(); + assert_eq!(b.len(), 1024); +} + +#[test] +fn par_iter_collect_linked_list() { + let a: Vec<i32> = (0..1024).collect(); + let b: LinkedList<_> = a.par_iter().map(|&i| (i, format!("{}", i))).collect(); + let c: LinkedList<_> = a.iter().map(|&i| (i, format!("{}", i))).collect(); + assert_eq!(b, c); +} + +#[test] +fn par_iter_collect_linked_list_flat_map_filter() { + let b: LinkedList<i32> = (0_i32..1024) + .into_par_iter() + .flat_map(|i| (0..i)) + .filter(|&i| i % 2 == 0) + .collect(); + let c: LinkedList<i32> = (0_i32..1024) + .flat_map(|i| (0..i)) + .filter(|&i| i % 2 == 0) + .collect(); + assert_eq!(b, c); +} + +#[test] +fn par_iter_collect_cows() { + use std::borrow::Cow; + + let s = "Fearless Concurrency with Rust"; + + // Collects `i32` into a `Vec` + let a: Cow<'_, [i32]> = (0..1024).collect(); + let b: Cow<'_, [i32]> = a.par_iter().cloned().collect(); + assert_eq!(a, b); + + // Collects `char` into a `String` + let a: Cow<'_, str> = s.chars().collect(); + let b: Cow<'_, str> = s.par_chars().collect(); + assert_eq!(a, b); + + // Collects `str` into a `String` + let a: Cow<'_, str> = s.split_whitespace().collect(); + let b: Cow<'_, str> = s.par_split_whitespace().collect(); + assert_eq!(a, b); + + // Collects `String` into a `String` + let a: Cow<'_, str> = s.split_whitespace().map(str::to_owned).collect(); + let b: Cow<'_, str> = s.par_split_whitespace().map(str::to_owned).collect(); + assert_eq!(a, b); +} + +#[test] +fn par_iter_unindexed_flat_map() { + let b: Vec<i64> = (0_i64..1024).into_par_iter().flat_map(Some).collect(); + let c: Vec<i64> = (0_i64..1024).flat_map(Some).collect(); + assert_eq!(b, c); +} + +#[test] +fn min_max() { + let rng = seeded_rng(); + let a: Vec<i32> = rng.sample_iter(&Standard).take(1024).collect(); + for i in 0..=a.len() { + let slice = &a[..i]; + assert_eq!(slice.par_iter().min(), slice.iter().min()); + assert_eq!(slice.par_iter().max(), slice.iter().max()); + } +} + +#[test] +fn min_max_by() { + let rng = seeded_rng(); + // Make sure there are duplicate keys, for testing sort stability + let r: Vec<i32> = rng.sample_iter(&Standard).take(512).collect(); + let a: Vec<(i32, u16)> = r.iter().chain(&r).cloned().zip(0..).collect(); + for i in 0..=a.len() { + let slice = &a[..i]; + assert_eq!( + slice.par_iter().min_by(|x, y| x.0.cmp(&y.0)), + slice.iter().min_by(|x, y| x.0.cmp(&y.0)) + ); + assert_eq!( + slice.par_iter().max_by(|x, y| x.0.cmp(&y.0)), + slice.iter().max_by(|x, y| x.0.cmp(&y.0)) + ); + } +} + +#[test] +fn min_max_by_key() { + let rng = seeded_rng(); + // Make sure there are duplicate keys, for testing sort stability + let r: Vec<i32> = rng.sample_iter(&Standard).take(512).collect(); + let a: Vec<(i32, u16)> = r.iter().chain(&r).cloned().zip(0..).collect(); + for i in 0..=a.len() { + let slice = &a[..i]; + assert_eq!( + slice.par_iter().min_by_key(|x| x.0), + slice.iter().min_by_key(|x| x.0) + ); + assert_eq!( + slice.par_iter().max_by_key(|x| x.0), + slice.iter().max_by_key(|x| x.0) + ); + } +} + +#[test] +fn check_rev() { + let a: Vec<usize> = (0..1024).rev().collect(); + let b: Vec<usize> = (0..1024).collect(); + + assert!(a.par_iter().rev().zip(b).all(|(&a, b)| a == b)); +} + +#[test] +fn scope_mix() { + let counter_p = &AtomicUsize::new(0); + scope(|s| { + s.spawn(move |s| { + divide_and_conquer(s, counter_p, 1024); + }); + s.spawn(move |_| { + let a: Vec<i32> = (0..1024).collect(); + let r1 = a.par_iter().map(|&i| i + 1).reduce_with(|i, j| i + j); + let r2 = a.iter().map(|&i| i + 1).sum(); + assert_eq!(r1.unwrap(), r2); + }); + }); +} + +fn divide_and_conquer<'scope>(scope: &Scope<'scope>, counter: &'scope AtomicUsize, size: usize) { + if size > 1 { + scope.spawn(move |scope| divide_and_conquer(scope, counter, size / 2)); + scope.spawn(move |scope| divide_and_conquer(scope, counter, size / 2)); + } else { + // count the leaves + counter.fetch_add(1, Ordering::SeqCst); + } +} + +#[test] +fn check_split() { + use std::ops::Range; + + let a = (0..1024).into_par_iter(); + + let b = split(0..1024, |Range { start, end }| { + let mid = (end - start) / 2; + if mid > start { + (start..mid, Some(mid..end)) + } else { + (start..end, None) + } + }) + .flat_map(|range| range); + + assert_eq!(a.collect::<Vec<_>>(), b.collect::<Vec<_>>()); +} + +#[test] +fn check_lengths() { + fn check(min: usize, max: usize) { + let range = 0..1024 * 1024; + + // Check against normalized values. + let min_check = cmp::min(cmp::max(min, 1), range.len()); + let max_check = cmp::max(max, min_check.saturating_add(min_check - 1)); + + assert!( + range + .into_par_iter() + .with_min_len(min) + .with_max_len(max) + .fold(|| 0, |count, _| count + 1) + .all(|c| c >= min_check && c <= max_check), + "check_lengths failed {:?} -> {:?} ", + (min, max), + (min_check, max_check) + ); + } + + let lengths = [0, 1, 10, 100, 1_000, 10_000, 100_000, 1_000_000, usize::MAX]; + for &min in &lengths { + for &max in &lengths { + check(min, max); + } + } +} + +#[test] +fn check_map_with() { + let (sender, receiver) = mpsc::channel(); + let a: HashSet<_> = (0..1024).collect(); + + a.par_iter() + .cloned() + .map_with(sender, |s, i| s.send(i).unwrap()) + .count(); + + let b: HashSet<_> = receiver.iter().collect(); + assert_eq!(a, b); +} + +#[test] +fn check_fold_with() { + let (sender, receiver) = mpsc::channel(); + let a: HashSet<_> = (0..1024).collect(); + + a.par_iter() + .cloned() + .fold_with(sender, |s, i| { + s.send(i).unwrap(); + s + }) + .count(); + + let b: HashSet<_> = receiver.iter().collect(); + assert_eq!(a, b); +} + +#[test] +fn check_for_each_with() { + let (sender, receiver) = mpsc::channel(); + let a: HashSet<_> = (0..1024).collect(); + + a.par_iter() + .cloned() + .for_each_with(sender, |s, i| s.send(i).unwrap()); + + let b: HashSet<_> = receiver.iter().collect(); + assert_eq!(a, b); +} + +#[test] +fn check_extend_items() { + fn check<C>() + where + C: Default + + Eq + + Debug + + Extend<i32> + + for<'a> Extend<&'a i32> + + ParallelExtend<i32> + + for<'a> ParallelExtend<&'a i32>, + { + let mut serial = C::default(); + let mut parallel = C::default(); + + // extend with references + let v: Vec<_> = (0..128).collect(); + serial.extend(&v); + parallel.par_extend(&v); + assert_eq!(serial, parallel); + + // extend with values + serial.extend(-128..0); + parallel.par_extend(-128..0); + assert_eq!(serial, parallel); + } + + check::<BTreeSet<_>>(); + check::<HashSet<_>>(); + check::<LinkedList<_>>(); + check::<Vec<_>>(); + check::<VecDeque<_>>(); +} + +#[test] +fn check_extend_heap() { + let mut serial: BinaryHeap<_> = Default::default(); + let mut parallel: BinaryHeap<_> = Default::default(); + + // extend with references + let v: Vec<_> = (0..128).collect(); + serial.extend(&v); + parallel.par_extend(&v); + assert_eq!( + serial.clone().into_sorted_vec(), + parallel.clone().into_sorted_vec() + ); + + // extend with values + serial.extend(-128..0); + parallel.par_extend(-128..0); + assert_eq!(serial.into_sorted_vec(), parallel.into_sorted_vec()); +} + +#[test] +fn check_extend_pairs() { + fn check<C>() + where + C: Default + + Eq + + Debug + + Extend<(usize, i32)> + + for<'a> Extend<(&'a usize, &'a i32)> + + ParallelExtend<(usize, i32)> + + for<'a> ParallelExtend<(&'a usize, &'a i32)>, + { + let mut serial = C::default(); + let mut parallel = C::default(); + + // extend with references + let m: HashMap<_, _> = (0..128).enumerate().collect(); + serial.extend(&m); + parallel.par_extend(&m); + assert_eq!(serial, parallel); + + // extend with values + let v: Vec<(_, _)> = (-128..0).enumerate().collect(); + serial.extend(v.clone()); + parallel.par_extend(v); + assert_eq!(serial, parallel); + } + + check::<BTreeMap<usize, i32>>(); + check::<HashMap<usize, i32>>(); +} + +#[test] +fn check_unzip_into_vecs() { + let mut a = vec![]; + let mut b = vec![]; + (0..1024) + .into_par_iter() + .map(|i| i * i) + .enumerate() + .unzip_into_vecs(&mut a, &mut b); + + let (c, d): (Vec<_>, Vec<_>) = (0..1024).map(|i| i * i).enumerate().unzip(); + assert_eq!(a, c); + assert_eq!(b, d); +} + +#[test] +fn check_unzip() { + // indexed, unindexed + let (a, b): (Vec<_>, HashSet<_>) = (0..1024).into_par_iter().map(|i| i * i).enumerate().unzip(); + let (c, d): (Vec<_>, HashSet<_>) = (0..1024).map(|i| i * i).enumerate().unzip(); + assert_eq!(a, c); + assert_eq!(b, d); + + // unindexed, indexed + let (a, b): (HashSet<_>, Vec<_>) = (0..1024).into_par_iter().map(|i| i * i).enumerate().unzip(); + let (c, d): (HashSet<_>, Vec<_>) = (0..1024).map(|i| i * i).enumerate().unzip(); + assert_eq!(a, c); + assert_eq!(b, d); + + // indexed, indexed + let (a, b): (Vec<_>, Vec<_>) = (0..1024).into_par_iter().map(|i| i * i).enumerate().unzip(); + let (c, d): (Vec<_>, Vec<_>) = (0..1024).map(|i| i * i).enumerate().unzip(); + assert_eq!(a, c); + assert_eq!(b, d); + + // unindexed producer + let (a, b): (Vec<_>, Vec<_>) = (0..1024) + .into_par_iter() + .filter_map(|i| Some((i, i * i))) + .unzip(); + let (c, d): (Vec<_>, Vec<_>) = (0..1024).map(|i| (i, i * i)).unzip(); + assert_eq!(a, c); + assert_eq!(b, d); +} + +#[test] +fn check_partition() { + let (a, b): (Vec<_>, Vec<_>) = (0..1024).into_par_iter().partition(|&i| i % 3 == 0); + let (c, d): (Vec<_>, Vec<_>) = (0..1024).partition(|&i| i % 3 == 0); + assert_eq!(a, c); + assert_eq!(b, d); +} + +#[test] +fn check_partition_map() { + let input = "a b c 1 2 3 x y z"; + let (a, b): (Vec<_>, String) = + input + .par_split_whitespace() + .partition_map(|s| match s.parse::<i32>() { + Ok(n) => Either::Left(n), + Err(_) => Either::Right(s), + }); + assert_eq!(a, vec![1, 2, 3]); + assert_eq!(b, "abcxyz"); +} + +#[test] +fn check_either() { + type I = crate::vec::IntoIter<i32>; + type E = Either<I, I>; + + let v: Vec<i32> = (0..1024).collect(); + + // try iterating the left side + let left: E = Either::Left(v.clone().into_par_iter()); + assert!(left.eq(v.clone())); + + // try iterating the right side + let right: E = Either::Right(v.clone().into_par_iter()); + assert!(right.eq(v.clone())); + + // try an indexed iterator + let left: E = Either::Left(v.clone().into_par_iter()); + assert!(left.enumerate().eq(v.into_par_iter().enumerate())); +} + +#[test] +fn check_either_extend() { + type E = Either<Vec<i32>, HashSet<i32>>; + + let v: Vec<i32> = (0..1024).collect(); + + // try extending the left side + let mut left: E = Either::Left(vec![]); + left.par_extend(v.clone()); + assert_eq!(left.as_ref(), Either::Left(&v)); + + // try extending the right side + let mut right: E = Either::Right(HashSet::default()); + right.par_extend(v.clone()); + assert_eq!(right, Either::Right(v.iter().cloned().collect())); +} + +#[test] +fn check_interleave_eq() { + let xs: Vec<usize> = (0..10).collect(); + let ys: Vec<usize> = (10..20).collect(); + + let mut actual = vec![]; + xs.par_iter() + .interleave(&ys) + .map(|&i| i) + .collect_into_vec(&mut actual); + + let expected: Vec<usize> = (0..10) + .zip(10..20) + .flat_map(|(i, j)| vec![i, j].into_iter()) + .collect(); + assert_eq!(expected, actual); +} + +#[test] +fn check_interleave_uneven() { + let cases: Vec<(Vec<usize>, Vec<usize>, Vec<usize>)> = vec![ + ( + (0..9).collect(), + vec![10], + vec![0, 10, 1, 2, 3, 4, 5, 6, 7, 8], + ), + ( + vec![10], + (0..9).collect(), + vec![10, 0, 1, 2, 3, 4, 5, 6, 7, 8], + ), + ( + (0..5).collect(), + (5..10).collect(), + (0..5) + .zip(5..10) + .flat_map(|(i, j)| vec![i, j].into_iter()) + .collect(), + ), + (vec![], (0..9).collect(), (0..9).collect()), + ((0..9).collect(), vec![], (0..9).collect()), + ( + (0..50).collect(), + (50..100).collect(), + (0..50) + .zip(50..100) + .flat_map(|(i, j)| vec![i, j].into_iter()) + .collect(), + ), + ]; + + for (i, (xs, ys, expected)) in cases.into_iter().enumerate() { + let mut res = vec![]; + xs.par_iter() + .interleave(&ys) + .map(|&i| i) + .collect_into_vec(&mut res); + assert_eq!(expected, res, "Case {} failed", i); + + res.truncate(0); + xs.par_iter() + .interleave(&ys) + .rev() + .map(|&i| i) + .collect_into_vec(&mut res); + assert_eq!( + expected.into_iter().rev().collect::<Vec<usize>>(), + res, + "Case {} reversed failed", + i + ); + } +} + +#[test] +fn check_interleave_shortest() { + let cases: Vec<(Vec<usize>, Vec<usize>, Vec<usize>)> = vec![ + ((0..9).collect(), vec![10], vec![0, 10, 1]), + (vec![10], (0..9).collect(), vec![10, 0]), + ( + (0..5).collect(), + (5..10).collect(), + (0..5) + .zip(5..10) + .flat_map(|(i, j)| vec![i, j].into_iter()) + .collect(), + ), + (vec![], (0..9).collect(), vec![]), + ((0..9).collect(), vec![], vec![0]), + ( + (0..50).collect(), + (50..100).collect(), + (0..50) + .zip(50..100) + .flat_map(|(i, j)| vec![i, j].into_iter()) + .collect(), + ), + ]; + + for (i, (xs, ys, expected)) in cases.into_iter().enumerate() { + let mut res = vec![]; + xs.par_iter() + .interleave_shortest(&ys) + .map(|&i| i) + .collect_into_vec(&mut res); + assert_eq!(expected, res, "Case {} failed", i); + + res.truncate(0); + xs.par_iter() + .interleave_shortest(&ys) + .rev() + .map(|&i| i) + .collect_into_vec(&mut res); + assert_eq!( + expected.into_iter().rev().collect::<Vec<usize>>(), + res, + "Case {} reversed failed", + i + ); + } +} + +#[test] +#[should_panic(expected = "chunk_size must not be zero")] +fn check_chunks_zero_size() { + let _: Vec<Vec<i32>> = vec![1, 2, 3].into_par_iter().chunks(0).collect(); +} + +#[test] +fn check_chunks_even_size() { + assert_eq!( + vec![vec![1, 2, 3], vec![4, 5, 6], vec![7, 8, 9]], + (1..10).into_par_iter().chunks(3).collect::<Vec<Vec<i32>>>() + ); +} + +#[test] +fn check_chunks_empty() { + let v: Vec<i32> = vec![]; + let expected: Vec<Vec<i32>> = vec![]; + assert_eq!( + expected, + v.into_par_iter().chunks(2).collect::<Vec<Vec<i32>>>() + ); +} + +#[test] +fn check_chunks_len() { + assert_eq!(4, (0..8).into_par_iter().chunks(2).len()); + assert_eq!(3, (0..9).into_par_iter().chunks(3).len()); + assert_eq!(3, (0..8).into_par_iter().chunks(3).len()); + assert_eq!(1, [1].par_iter().chunks(3).len()); + assert_eq!(0, (0..0).into_par_iter().chunks(3).len()); +} + +#[test] +fn check_chunks_uneven() { + let cases: Vec<(Vec<u32>, usize, Vec<Vec<u32>>)> = vec![ + ((0..5).collect(), 3, vec![vec![0, 1, 2], vec![3, 4]]), + (vec![1], 5, vec![vec![1]]), + ((0..4).collect(), 3, vec![vec![0, 1, 2], vec![3]]), + ]; + + for (i, (v, n, expected)) in cases.into_iter().enumerate() { + let mut res: Vec<Vec<u32>> = vec![]; + v.par_iter() + .chunks(n) + .map(|v| v.into_iter().cloned().collect()) + .collect_into_vec(&mut res); + assert_eq!(expected, res, "Case {} failed", i); + + res.truncate(0); + v.into_par_iter().chunks(n).rev().collect_into_vec(&mut res); + assert_eq!( + expected.into_iter().rev().collect::<Vec<Vec<u32>>>(), + res, + "Case {} reversed failed", + i + ); + } +} + +#[test] +#[ignore] // it's quick enough on optimized 32-bit platforms, but otherwise... ... ... +#[should_panic(expected = "overflow")] +#[cfg(debug_assertions)] +fn check_repeat_unbounded() { + // use just one thread, so we don't get infinite adaptive splitting + // (forever stealing and re-splitting jobs that will panic on overflow) + let pool = ThreadPoolBuilder::new().num_threads(1).build().unwrap(); + pool.install(|| { + println!("counted {} repeats", repeat(()).count()); + }); +} + +#[test] +fn check_repeat_find_any() { + let even = repeat(4).find_any(|&x| x % 2 == 0); + assert_eq!(even, Some(4)); +} + +#[test] +fn check_repeat_take() { + let v: Vec<_> = repeat(4).take(4).collect(); + assert_eq!(v, [4, 4, 4, 4]); +} + +#[test] +fn check_repeat_zip() { + let v = vec![4, 4, 4, 4]; + let mut fours: Vec<_> = repeat(4).zip(v).collect(); + assert_eq!(fours.len(), 4); + while let Some(item) = fours.pop() { + assert_eq!(item, (4, 4)); + } +} + +#[test] +fn check_repeatn_zip_left() { + let v = vec![4, 4, 4, 4]; + let mut fours: Vec<_> = repeatn(4, usize::MAX).zip(v).collect(); + assert_eq!(fours.len(), 4); + while let Some(item) = fours.pop() { + assert_eq!(item, (4, 4)); + } +} + +#[test] +fn check_repeatn_zip_right() { + let v = vec![4, 4, 4, 4]; + let mut fours: Vec<_> = v.into_par_iter().zip(repeatn(4, usize::MAX)).collect(); + assert_eq!(fours.len(), 4); + while let Some(item) = fours.pop() { + assert_eq!(item, (4, 4)); + } +} + +#[test] +fn check_empty() { + // drive_unindexed + let mut v: Vec<i32> = empty().filter(|_| unreachable!()).collect(); + assert!(v.is_empty()); + + // drive (indexed) + empty().collect_into_vec(&mut v); + assert!(v.is_empty()); + + // with_producer + let v: Vec<(i32, i32)> = empty().zip(1..10).collect(); + assert!(v.is_empty()); +} + +#[test] +fn check_once() { + // drive_unindexed + let mut v: Vec<i32> = once(42).filter(|_| true).collect(); + assert_eq!(v, &[42]); + + // drive (indexed) + once(42).collect_into_vec(&mut v); + assert_eq!(v, &[42]); + + // with_producer + let v: Vec<(i32, i32)> = once(42).zip(1..10).collect(); + assert_eq!(v, &[(42, 1)]); +} + +#[test] +fn check_update() { + let mut v: Vec<Vec<_>> = vec![vec![1], vec![3, 2, 1]]; + v.par_iter_mut().update(|v| v.push(0)).for_each(|_| ()); + + assert_eq!(v, vec![vec![1, 0], vec![3, 2, 1, 0]]); +} diff --git a/third_party/rust/rayon/src/iter/try_fold.rs b/third_party/rust/rayon/src/iter/try_fold.rs new file mode 100644 index 0000000000..6d1048d75d --- /dev/null +++ b/third_party/rust/rayon/src/iter/try_fold.rs @@ -0,0 +1,298 @@ +use super::plumbing::*; +use super::ParallelIterator; +use super::Try; + +use std::fmt::{self, Debug}; +use std::marker::PhantomData; +use std::ops::ControlFlow::{self, Break, Continue}; + +impl<U, I, ID, F> TryFold<I, U, ID, F> +where + I: ParallelIterator, + F: Fn(U::Output, I::Item) -> U + Sync + Send, + ID: Fn() -> U::Output + Sync + Send, + U: Try + Send, +{ + pub(super) fn new(base: I, identity: ID, fold_op: F) -> Self { + TryFold { + base, + identity, + fold_op, + marker: PhantomData, + } + } +} + +/// `TryFold` is an iterator that applies a function over an iterator producing a single value. +/// This struct is created by the [`try_fold()`] method on [`ParallelIterator`] +/// +/// [`try_fold()`]: trait.ParallelIterator.html#method.try_fold +/// [`ParallelIterator`]: trait.ParallelIterator.html +#[must_use = "iterator adaptors are lazy and do nothing unless consumed"] +#[derive(Clone)] +pub struct TryFold<I, U, ID, F> { + base: I, + identity: ID, + fold_op: F, + marker: PhantomData<U>, +} + +impl<U, I: ParallelIterator + Debug, ID, F> Debug for TryFold<I, U, ID, F> { + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + f.debug_struct("TryFold").field("base", &self.base).finish() + } +} + +impl<U, I, ID, F> ParallelIterator for TryFold<I, U, ID, F> +where + I: ParallelIterator, + F: Fn(U::Output, I::Item) -> U + Sync + Send, + ID: Fn() -> U::Output + Sync + Send, + U: Try + Send, +{ + type Item = U; + + fn drive_unindexed<C>(self, consumer: C) -> C::Result + where + C: UnindexedConsumer<Self::Item>, + { + let consumer1 = TryFoldConsumer { + base: consumer, + identity: &self.identity, + fold_op: &self.fold_op, + marker: PhantomData, + }; + self.base.drive_unindexed(consumer1) + } +} + +struct TryFoldConsumer<'c, U, C, ID, F> { + base: C, + identity: &'c ID, + fold_op: &'c F, + marker: PhantomData<U>, +} + +impl<'r, U, T, C, ID, F> Consumer<T> for TryFoldConsumer<'r, U, C, ID, F> +where + C: Consumer<U>, + F: Fn(U::Output, T) -> U + Sync, + ID: Fn() -> U::Output + Sync, + U: Try + Send, +{ + type Folder = TryFoldFolder<'r, C::Folder, U, F>; + type Reducer = C::Reducer; + type Result = C::Result; + + fn split_at(self, index: usize) -> (Self, Self, Self::Reducer) { + let (left, right, reducer) = self.base.split_at(index); + ( + TryFoldConsumer { base: left, ..self }, + TryFoldConsumer { + base: right, + ..self + }, + reducer, + ) + } + + fn into_folder(self) -> Self::Folder { + TryFoldFolder { + base: self.base.into_folder(), + control: Continue((self.identity)()), + fold_op: self.fold_op, + } + } + + fn full(&self) -> bool { + self.base.full() + } +} + +impl<'r, U, T, C, ID, F> UnindexedConsumer<T> for TryFoldConsumer<'r, U, C, ID, F> +where + C: UnindexedConsumer<U>, + F: Fn(U::Output, T) -> U + Sync, + ID: Fn() -> U::Output + Sync, + U: Try + Send, +{ + fn split_off_left(&self) -> Self { + TryFoldConsumer { + base: self.base.split_off_left(), + ..*self + } + } + + fn to_reducer(&self) -> Self::Reducer { + self.base.to_reducer() + } +} + +struct TryFoldFolder<'r, C, U: Try, F> { + base: C, + fold_op: &'r F, + control: ControlFlow<U::Residual, U::Output>, +} + +impl<'r, C, U, F, T> Folder<T> for TryFoldFolder<'r, C, U, F> +where + C: Folder<U>, + F: Fn(U::Output, T) -> U + Sync, + U: Try, +{ + type Result = C::Result; + + fn consume(mut self, item: T) -> Self { + let fold_op = self.fold_op; + if let Continue(acc) = self.control { + self.control = fold_op(acc, item).branch(); + } + self + } + + fn complete(self) -> C::Result { + let item = match self.control { + Continue(c) => U::from_output(c), + Break(r) => U::from_residual(r), + }; + self.base.consume(item).complete() + } + + fn full(&self) -> bool { + match self.control { + Break(_) => true, + _ => self.base.full(), + } + } +} + +// /////////////////////////////////////////////////////////////////////////// + +impl<U, I, F> TryFoldWith<I, U, F> +where + I: ParallelIterator, + F: Fn(U::Output, I::Item) -> U + Sync, + U: Try + Send, + U::Output: Clone + Send, +{ + pub(super) fn new(base: I, item: U::Output, fold_op: F) -> Self { + TryFoldWith { + base, + item, + fold_op, + } + } +} + +/// `TryFoldWith` is an iterator that applies a function over an iterator producing a single value. +/// This struct is created by the [`try_fold_with()`] method on [`ParallelIterator`] +/// +/// [`try_fold_with()`]: trait.ParallelIterator.html#method.try_fold_with +/// [`ParallelIterator`]: trait.ParallelIterator.html +#[must_use = "iterator adaptors are lazy and do nothing unless consumed"] +#[derive(Clone)] +pub struct TryFoldWith<I, U: Try, F> { + base: I, + item: U::Output, + fold_op: F, +} + +impl<I: ParallelIterator + Debug, U: Try, F> Debug for TryFoldWith<I, U, F> +where + U::Output: Debug, +{ + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + f.debug_struct("TryFoldWith") + .field("base", &self.base) + .field("item", &self.item) + .finish() + } +} + +impl<U, I, F> ParallelIterator for TryFoldWith<I, U, F> +where + I: ParallelIterator, + F: Fn(U::Output, I::Item) -> U + Sync + Send, + U: Try + Send, + U::Output: Clone + Send, +{ + type Item = U; + + fn drive_unindexed<C>(self, consumer: C) -> C::Result + where + C: UnindexedConsumer<Self::Item>, + { + let consumer1 = TryFoldWithConsumer { + base: consumer, + item: self.item, + fold_op: &self.fold_op, + }; + self.base.drive_unindexed(consumer1) + } +} + +struct TryFoldWithConsumer<'c, C, U: Try, F> { + base: C, + item: U::Output, + fold_op: &'c F, +} + +impl<'r, U, T, C, F> Consumer<T> for TryFoldWithConsumer<'r, C, U, F> +where + C: Consumer<U>, + F: Fn(U::Output, T) -> U + Sync, + U: Try + Send, + U::Output: Clone + Send, +{ + type Folder = TryFoldFolder<'r, C::Folder, U, F>; + type Reducer = C::Reducer; + type Result = C::Result; + + fn split_at(self, index: usize) -> (Self, Self, Self::Reducer) { + let (left, right, reducer) = self.base.split_at(index); + ( + TryFoldWithConsumer { + base: left, + item: self.item.clone(), + ..self + }, + TryFoldWithConsumer { + base: right, + ..self + }, + reducer, + ) + } + + fn into_folder(self) -> Self::Folder { + TryFoldFolder { + base: self.base.into_folder(), + control: Continue(self.item), + fold_op: self.fold_op, + } + } + + fn full(&self) -> bool { + self.base.full() + } +} + +impl<'r, U, T, C, F> UnindexedConsumer<T> for TryFoldWithConsumer<'r, C, U, F> +where + C: UnindexedConsumer<U>, + F: Fn(U::Output, T) -> U + Sync, + U: Try + Send, + U::Output: Clone + Send, +{ + fn split_off_left(&self) -> Self { + TryFoldWithConsumer { + base: self.base.split_off_left(), + item: self.item.clone(), + ..*self + } + } + + fn to_reducer(&self) -> Self::Reducer { + self.base.to_reducer() + } +} diff --git a/third_party/rust/rayon/src/iter/try_reduce.rs b/third_party/rust/rayon/src/iter/try_reduce.rs new file mode 100644 index 0000000000..35a724c94f --- /dev/null +++ b/third_party/rust/rayon/src/iter/try_reduce.rs @@ -0,0 +1,131 @@ +use super::plumbing::*; +use super::ParallelIterator; +use super::Try; + +use std::ops::ControlFlow::{self, Break, Continue}; +use std::sync::atomic::{AtomicBool, Ordering}; + +pub(super) fn try_reduce<PI, R, ID, T>(pi: PI, identity: ID, reduce_op: R) -> T +where + PI: ParallelIterator<Item = T>, + R: Fn(T::Output, T::Output) -> T + Sync, + ID: Fn() -> T::Output + Sync, + T: Try + Send, +{ + let full = AtomicBool::new(false); + let consumer = TryReduceConsumer { + identity: &identity, + reduce_op: &reduce_op, + full: &full, + }; + pi.drive_unindexed(consumer) +} + +struct TryReduceConsumer<'r, R, ID> { + identity: &'r ID, + reduce_op: &'r R, + full: &'r AtomicBool, +} + +impl<'r, R, ID> Copy for TryReduceConsumer<'r, R, ID> {} + +impl<'r, R, ID> Clone for TryReduceConsumer<'r, R, ID> { + fn clone(&self) -> Self { + *self + } +} + +impl<'r, R, ID, T> Consumer<T> for TryReduceConsumer<'r, R, ID> +where + R: Fn(T::Output, T::Output) -> T + Sync, + ID: Fn() -> T::Output + Sync, + T: Try + Send, +{ + type Folder = TryReduceFolder<'r, R, T>; + type Reducer = Self; + type Result = T; + + fn split_at(self, _index: usize) -> (Self, Self, Self) { + (self, self, self) + } + + fn into_folder(self) -> Self::Folder { + TryReduceFolder { + reduce_op: self.reduce_op, + control: Continue((self.identity)()), + full: self.full, + } + } + + fn full(&self) -> bool { + self.full.load(Ordering::Relaxed) + } +} + +impl<'r, R, ID, T> UnindexedConsumer<T> for TryReduceConsumer<'r, R, ID> +where + R: Fn(T::Output, T::Output) -> T + Sync, + ID: Fn() -> T::Output + Sync, + T: Try + Send, +{ + fn split_off_left(&self) -> Self { + *self + } + + fn to_reducer(&self) -> Self::Reducer { + *self + } +} + +impl<'r, R, ID, T> Reducer<T> for TryReduceConsumer<'r, R, ID> +where + R: Fn(T::Output, T::Output) -> T + Sync, + T: Try, +{ + fn reduce(self, left: T, right: T) -> T { + match (left.branch(), right.branch()) { + (Continue(left), Continue(right)) => (self.reduce_op)(left, right), + (Break(r), _) | (_, Break(r)) => T::from_residual(r), + } + } +} + +struct TryReduceFolder<'r, R, T: Try> { + reduce_op: &'r R, + control: ControlFlow<T::Residual, T::Output>, + full: &'r AtomicBool, +} + +impl<'r, R, T> Folder<T> for TryReduceFolder<'r, R, T> +where + R: Fn(T::Output, T::Output) -> T, + T: Try, +{ + type Result = T; + + fn consume(mut self, item: T) -> Self { + let reduce_op = self.reduce_op; + self.control = match (self.control, item.branch()) { + (Continue(left), Continue(right)) => reduce_op(left, right).branch(), + (control @ Break(_), _) | (_, control @ Break(_)) => control, + }; + if let Break(_) = self.control { + self.full.store(true, Ordering::Relaxed); + } + self + } + + fn complete(self) -> T { + match self.control { + Continue(c) => T::from_output(c), + Break(r) => T::from_residual(r), + } + } + + fn full(&self) -> bool { + match self.control { + Break(_) => true, + _ => self.full.load(Ordering::Relaxed), + } + } +} diff --git a/third_party/rust/rayon/src/iter/try_reduce_with.rs b/third_party/rust/rayon/src/iter/try_reduce_with.rs new file mode 100644 index 0000000000..cd7c83e272 --- /dev/null +++ b/third_party/rust/rayon/src/iter/try_reduce_with.rs @@ -0,0 +1,132 @@ +use super::plumbing::*; +use super::ParallelIterator; +use super::Try; + +use std::ops::ControlFlow::{self, Break, Continue}; +use std::sync::atomic::{AtomicBool, Ordering}; + +pub(super) fn try_reduce_with<PI, R, T>(pi: PI, reduce_op: R) -> Option<T> +where + PI: ParallelIterator<Item = T>, + R: Fn(T::Output, T::Output) -> T + Sync, + T: Try + Send, +{ + let full = AtomicBool::new(false); + let consumer = TryReduceWithConsumer { + reduce_op: &reduce_op, + full: &full, + }; + pi.drive_unindexed(consumer) +} + +struct TryReduceWithConsumer<'r, R> { + reduce_op: &'r R, + full: &'r AtomicBool, +} + +impl<'r, R> Copy for TryReduceWithConsumer<'r, R> {} + +impl<'r, R> Clone for TryReduceWithConsumer<'r, R> { + fn clone(&self) -> Self { + *self + } +} + +impl<'r, R, T> Consumer<T> for TryReduceWithConsumer<'r, R> +where + R: Fn(T::Output, T::Output) -> T + Sync, + T: Try + Send, +{ + type Folder = TryReduceWithFolder<'r, R, T>; + type Reducer = Self; + type Result = Option<T>; + + fn split_at(self, _index: usize) -> (Self, Self, Self) { + (self, self, self) + } + + fn into_folder(self) -> Self::Folder { + TryReduceWithFolder { + reduce_op: self.reduce_op, + opt_control: None, + full: self.full, + } + } + + fn full(&self) -> bool { + self.full.load(Ordering::Relaxed) + } +} + +impl<'r, R, T> UnindexedConsumer<T> for TryReduceWithConsumer<'r, R> +where + R: Fn(T::Output, T::Output) -> T + Sync, + T: Try + Send, +{ + fn split_off_left(&self) -> Self { + *self + } + + fn to_reducer(&self) -> Self::Reducer { + *self + } +} + +impl<'r, R, T> Reducer<Option<T>> for TryReduceWithConsumer<'r, R> +where + R: Fn(T::Output, T::Output) -> T + Sync, + T: Try, +{ + fn reduce(self, left: Option<T>, right: Option<T>) -> Option<T> { + let reduce_op = self.reduce_op; + match (left, right) { + (Some(left), Some(right)) => match (left.branch(), right.branch()) { + (Continue(left), Continue(right)) => Some(reduce_op(left, right)), + (Break(r), _) | (_, Break(r)) => Some(T::from_residual(r)), + }, + (None, x) | (x, None) => x, + } + } +} + +struct TryReduceWithFolder<'r, R, T: Try> { + reduce_op: &'r R, + opt_control: Option<ControlFlow<T::Residual, T::Output>>, + full: &'r AtomicBool, +} + +impl<'r, R, T> Folder<T> for TryReduceWithFolder<'r, R, T> +where + R: Fn(T::Output, T::Output) -> T, + T: Try, +{ + type Result = Option<T>; + + fn consume(mut self, item: T) -> Self { + let reduce_op = self.reduce_op; + let control = match (self.opt_control, item.branch()) { + (Some(Continue(left)), Continue(right)) => reduce_op(left, right).branch(), + (Some(control @ Break(_)), _) | (_, control) => control, + }; + if let Break(_) = control { + self.full.store(true, Ordering::Relaxed) + } + self.opt_control = Some(control); + self + } + + fn complete(self) -> Option<T> { + match self.opt_control { + Some(Continue(c)) => Some(T::from_output(c)), + Some(Break(r)) => Some(T::from_residual(r)), + None => None, + } + } + + fn full(&self) -> bool { + match self.opt_control { + Some(Break(_)) => true, + _ => self.full.load(Ordering::Relaxed), + } + } +} diff --git a/third_party/rust/rayon/src/iter/unzip.rs b/third_party/rust/rayon/src/iter/unzip.rs new file mode 100644 index 0000000000..0b7074e929 --- /dev/null +++ b/third_party/rust/rayon/src/iter/unzip.rs @@ -0,0 +1,525 @@ +use super::plumbing::*; +use super::*; + +/// This trait abstracts the different ways we can "unzip" one parallel +/// iterator into two distinct consumers, which we can handle almost +/// identically apart from how to process the individual items. +trait UnzipOp<T>: Sync + Send { + /// The type of item expected by the left consumer. + type Left: Send; + + /// The type of item expected by the right consumer. + type Right: Send; + + /// Consumes one item and feeds it to one or both of the underlying folders. + fn consume<FA, FB>(&self, item: T, left: FA, right: FB) -> (FA, FB) + where + FA: Folder<Self::Left>, + FB: Folder<Self::Right>; + + /// Reports whether this op may support indexed consumers. + /// - e.g. true for `unzip` where the item count passed through directly. + /// - e.g. false for `partition` where the sorting is not yet known. + fn indexable() -> bool { + false + } +} + +/// Runs an unzip-like operation into default `ParallelExtend` collections. +fn execute<I, OP, FromA, FromB>(pi: I, op: OP) -> (FromA, FromB) +where + I: ParallelIterator, + OP: UnzipOp<I::Item>, + FromA: Default + Send + ParallelExtend<OP::Left>, + FromB: Default + Send + ParallelExtend<OP::Right>, +{ + let mut a = FromA::default(); + let mut b = FromB::default(); + execute_into(&mut a, &mut b, pi, op); + (a, b) +} + +/// Runs an unzip-like operation into `ParallelExtend` collections. +fn execute_into<I, OP, FromA, FromB>(a: &mut FromA, b: &mut FromB, pi: I, op: OP) +where + I: ParallelIterator, + OP: UnzipOp<I::Item>, + FromA: Send + ParallelExtend<OP::Left>, + FromB: Send + ParallelExtend<OP::Right>, +{ + // We have no idea what the consumers will look like for these + // collections' `par_extend`, but we can intercept them in our own + // `drive_unindexed`. Start with the left side, type `A`: + let iter = UnzipA { base: pi, op, b }; + a.par_extend(iter); +} + +/// Unzips the items of a parallel iterator into a pair of arbitrary +/// `ParallelExtend` containers. +/// +/// This is called by `ParallelIterator::unzip`. +pub(super) fn unzip<I, A, B, FromA, FromB>(pi: I) -> (FromA, FromB) +where + I: ParallelIterator<Item = (A, B)>, + FromA: Default + Send + ParallelExtend<A>, + FromB: Default + Send + ParallelExtend<B>, + A: Send, + B: Send, +{ + execute(pi, Unzip) +} + +/// Unzips an `IndexedParallelIterator` into two arbitrary `Consumer`s. +/// +/// This is called by `super::collect::unzip_into_vecs`. +pub(super) fn unzip_indexed<I, A, B, CA, CB>(pi: I, left: CA, right: CB) -> (CA::Result, CB::Result) +where + I: IndexedParallelIterator<Item = (A, B)>, + CA: Consumer<A>, + CB: Consumer<B>, + A: Send, + B: Send, +{ + let consumer = UnzipConsumer { + op: &Unzip, + left, + right, + }; + pi.drive(consumer) +} + +/// An `UnzipOp` that splits a tuple directly into the two consumers. +struct Unzip; + +impl<A: Send, B: Send> UnzipOp<(A, B)> for Unzip { + type Left = A; + type Right = B; + + fn consume<FA, FB>(&self, item: (A, B), left: FA, right: FB) -> (FA, FB) + where + FA: Folder<A>, + FB: Folder<B>, + { + (left.consume(item.0), right.consume(item.1)) + } + + fn indexable() -> bool { + true + } +} + +/// Partitions the items of a parallel iterator into a pair of arbitrary +/// `ParallelExtend` containers. +/// +/// This is called by `ParallelIterator::partition`. +pub(super) fn partition<I, A, B, P>(pi: I, predicate: P) -> (A, B) +where + I: ParallelIterator, + A: Default + Send + ParallelExtend<I::Item>, + B: Default + Send + ParallelExtend<I::Item>, + P: Fn(&I::Item) -> bool + Sync + Send, +{ + execute(pi, Partition { predicate }) +} + +/// An `UnzipOp` that routes items depending on a predicate function. +struct Partition<P> { + predicate: P, +} + +impl<P, T> UnzipOp<T> for Partition<P> +where + P: Fn(&T) -> bool + Sync + Send, + T: Send, +{ + type Left = T; + type Right = T; + + fn consume<FA, FB>(&self, item: T, left: FA, right: FB) -> (FA, FB) + where + FA: Folder<T>, + FB: Folder<T>, + { + if (self.predicate)(&item) { + (left.consume(item), right) + } else { + (left, right.consume(item)) + } + } +} + +/// Partitions and maps the items of a parallel iterator into a pair of +/// arbitrary `ParallelExtend` containers. +/// +/// This called by `ParallelIterator::partition_map`. +pub(super) fn partition_map<I, A, B, P, L, R>(pi: I, predicate: P) -> (A, B) +where + I: ParallelIterator, + A: Default + Send + ParallelExtend<L>, + B: Default + Send + ParallelExtend<R>, + P: Fn(I::Item) -> Either<L, R> + Sync + Send, + L: Send, + R: Send, +{ + execute(pi, PartitionMap { predicate }) +} + +/// An `UnzipOp` that routes items depending on how they are mapped `Either`. +struct PartitionMap<P> { + predicate: P, +} + +impl<P, L, R, T> UnzipOp<T> for PartitionMap<P> +where + P: Fn(T) -> Either<L, R> + Sync + Send, + L: Send, + R: Send, +{ + type Left = L; + type Right = R; + + fn consume<FA, FB>(&self, item: T, left: FA, right: FB) -> (FA, FB) + where + FA: Folder<L>, + FB: Folder<R>, + { + match (self.predicate)(item) { + Either::Left(item) => (left.consume(item), right), + Either::Right(item) => (left, right.consume(item)), + } + } +} + +/// A fake iterator to intercept the `Consumer` for type `A`. +struct UnzipA<'b, I, OP, FromB> { + base: I, + op: OP, + b: &'b mut FromB, +} + +impl<'b, I, OP, FromB> ParallelIterator for UnzipA<'b, I, OP, FromB> +where + I: ParallelIterator, + OP: UnzipOp<I::Item>, + FromB: Send + ParallelExtend<OP::Right>, +{ + type Item = OP::Left; + + fn drive_unindexed<C>(self, consumer: C) -> C::Result + where + C: UnindexedConsumer<Self::Item>, + { + let mut result = None; + { + // Now it's time to find the consumer for type `B` + let iter = UnzipB { + base: self.base, + op: self.op, + left_consumer: consumer, + left_result: &mut result, + }; + self.b.par_extend(iter); + } + // NB: If for some reason `b.par_extend` doesn't actually drive the + // iterator, then we won't have a result for the left side to return + // at all. We can't fake an arbitrary consumer's result, so panic. + result.expect("unzip consumers didn't execute!") + } + + fn opt_len(&self) -> Option<usize> { + if OP::indexable() { + self.base.opt_len() + } else { + None + } + } +} + +/// A fake iterator to intercept the `Consumer` for type `B`. +struct UnzipB<'r, I, OP, CA> +where + I: ParallelIterator, + OP: UnzipOp<I::Item>, + CA: UnindexedConsumer<OP::Left>, + CA::Result: 'r, +{ + base: I, + op: OP, + left_consumer: CA, + left_result: &'r mut Option<CA::Result>, +} + +impl<'r, I, OP, CA> ParallelIterator for UnzipB<'r, I, OP, CA> +where + I: ParallelIterator, + OP: UnzipOp<I::Item>, + CA: UnindexedConsumer<OP::Left>, +{ + type Item = OP::Right; + + fn drive_unindexed<C>(self, consumer: C) -> C::Result + where + C: UnindexedConsumer<Self::Item>, + { + // Now that we have two consumers, we can unzip the real iterator. + let consumer = UnzipConsumer { + op: &self.op, + left: self.left_consumer, + right: consumer, + }; + + let result = self.base.drive_unindexed(consumer); + *self.left_result = Some(result.0); + result.1 + } + + fn opt_len(&self) -> Option<usize> { + if OP::indexable() { + self.base.opt_len() + } else { + None + } + } +} + +/// `Consumer` that unzips into two other `Consumer`s +struct UnzipConsumer<'a, OP, CA, CB> { + op: &'a OP, + left: CA, + right: CB, +} + +impl<'a, T, OP, CA, CB> Consumer<T> for UnzipConsumer<'a, OP, CA, CB> +where + OP: UnzipOp<T>, + CA: Consumer<OP::Left>, + CB: Consumer<OP::Right>, +{ + type Folder = UnzipFolder<'a, OP, CA::Folder, CB::Folder>; + type Reducer = UnzipReducer<CA::Reducer, CB::Reducer>; + type Result = (CA::Result, CB::Result); + + fn split_at(self, index: usize) -> (Self, Self, Self::Reducer) { + let (left1, left2, left_reducer) = self.left.split_at(index); + let (right1, right2, right_reducer) = self.right.split_at(index); + + ( + UnzipConsumer { + op: self.op, + left: left1, + right: right1, + }, + UnzipConsumer { + op: self.op, + left: left2, + right: right2, + }, + UnzipReducer { + left: left_reducer, + right: right_reducer, + }, + ) + } + + fn into_folder(self) -> Self::Folder { + UnzipFolder { + op: self.op, + left: self.left.into_folder(), + right: self.right.into_folder(), + } + } + + fn full(&self) -> bool { + // don't stop until everyone is full + self.left.full() && self.right.full() + } +} + +impl<'a, T, OP, CA, CB> UnindexedConsumer<T> for UnzipConsumer<'a, OP, CA, CB> +where + OP: UnzipOp<T>, + CA: UnindexedConsumer<OP::Left>, + CB: UnindexedConsumer<OP::Right>, +{ + fn split_off_left(&self) -> Self { + UnzipConsumer { + op: self.op, + left: self.left.split_off_left(), + right: self.right.split_off_left(), + } + } + + fn to_reducer(&self) -> Self::Reducer { + UnzipReducer { + left: self.left.to_reducer(), + right: self.right.to_reducer(), + } + } +} + +/// `Folder` that unzips into two other `Folder`s +struct UnzipFolder<'a, OP, FA, FB> { + op: &'a OP, + left: FA, + right: FB, +} + +impl<'a, T, OP, FA, FB> Folder<T> for UnzipFolder<'a, OP, FA, FB> +where + OP: UnzipOp<T>, + FA: Folder<OP::Left>, + FB: Folder<OP::Right>, +{ + type Result = (FA::Result, FB::Result); + + fn consume(self, item: T) -> Self { + let (left, right) = self.op.consume(item, self.left, self.right); + UnzipFolder { + op: self.op, + left, + right, + } + } + + fn complete(self) -> Self::Result { + (self.left.complete(), self.right.complete()) + } + + fn full(&self) -> bool { + // don't stop until everyone is full + self.left.full() && self.right.full() + } +} + +/// `Reducer` that unzips into two other `Reducer`s +struct UnzipReducer<RA, RB> { + left: RA, + right: RB, +} + +impl<A, B, RA, RB> Reducer<(A, B)> for UnzipReducer<RA, RB> +where + RA: Reducer<A>, + RB: Reducer<B>, +{ + fn reduce(self, left: (A, B), right: (A, B)) -> (A, B) { + ( + self.left.reduce(left.0, right.0), + self.right.reduce(left.1, right.1), + ) + } +} + +impl<A, B, FromA, FromB> ParallelExtend<(A, B)> for (FromA, FromB) +where + A: Send, + B: Send, + FromA: Send + ParallelExtend<A>, + FromB: Send + ParallelExtend<B>, +{ + fn par_extend<I>(&mut self, pi: I) + where + I: IntoParallelIterator<Item = (A, B)>, + { + execute_into(&mut self.0, &mut self.1, pi.into_par_iter(), Unzip); + } +} + +impl<L, R, A, B> ParallelExtend<Either<L, R>> for (A, B) +where + L: Send, + R: Send, + A: Send + ParallelExtend<L>, + B: Send + ParallelExtend<R>, +{ + fn par_extend<I>(&mut self, pi: I) + where + I: IntoParallelIterator<Item = Either<L, R>>, + { + execute_into(&mut self.0, &mut self.1, pi.into_par_iter(), UnEither); + } +} + +/// An `UnzipOp` that routes items depending on their `Either` variant. +struct UnEither; + +impl<L, R> UnzipOp<Either<L, R>> for UnEither +where + L: Send, + R: Send, +{ + type Left = L; + type Right = R; + + fn consume<FL, FR>(&self, item: Either<L, R>, left: FL, right: FR) -> (FL, FR) + where + FL: Folder<L>, + FR: Folder<R>, + { + match item { + Either::Left(item) => (left.consume(item), right), + Either::Right(item) => (left, right.consume(item)), + } + } +} + +impl<A, B, FromA, FromB> FromParallelIterator<(A, B)> for (FromA, FromB) +where + A: Send, + B: Send, + FromA: Send + FromParallelIterator<A>, + FromB: Send + FromParallelIterator<B>, +{ + fn from_par_iter<I>(pi: I) -> Self + where + I: IntoParallelIterator<Item = (A, B)>, + { + let (a, b): (Collector<FromA>, Collector<FromB>) = pi.into_par_iter().unzip(); + (a.result.unwrap(), b.result.unwrap()) + } +} + +impl<L, R, A, B> FromParallelIterator<Either<L, R>> for (A, B) +where + L: Send, + R: Send, + A: Send + FromParallelIterator<L>, + B: Send + FromParallelIterator<R>, +{ + fn from_par_iter<I>(pi: I) -> Self + where + I: IntoParallelIterator<Item = Either<L, R>>, + { + fn identity<T>(x: T) -> T { + x + } + + let (a, b): (Collector<A>, Collector<B>) = pi.into_par_iter().partition_map(identity); + (a.result.unwrap(), b.result.unwrap()) + } +} + +/// Shim to implement a one-time `ParallelExtend` using `FromParallelIterator`. +struct Collector<FromT> { + result: Option<FromT>, +} + +impl<FromT> Default for Collector<FromT> { + fn default() -> Self { + Collector { result: None } + } +} + +impl<T, FromT> ParallelExtend<T> for Collector<FromT> +where + T: Send, + FromT: Send + FromParallelIterator<T>, +{ + fn par_extend<I>(&mut self, pi: I) + where + I: IntoParallelIterator<Item = T>, + { + debug_assert!(self.result.is_none()); + self.result = Some(pi.into_par_iter().collect()); + } +} diff --git a/third_party/rust/rayon/src/iter/update.rs b/third_party/rust/rayon/src/iter/update.rs new file mode 100644 index 0000000000..c693ac8d60 --- /dev/null +++ b/third_party/rust/rayon/src/iter/update.rs @@ -0,0 +1,327 @@ +use super::plumbing::*; +use super::*; + +use std::fmt::{self, Debug}; + +/// `Update` is an iterator that mutates the elements of an +/// underlying iterator before they are yielded. +/// +/// This struct is created by the [`update()`] method on [`ParallelIterator`] +/// +/// [`update()`]: trait.ParallelIterator.html#method.update +/// [`ParallelIterator`]: trait.ParallelIterator.html +#[must_use = "iterator adaptors are lazy and do nothing unless consumed"] +#[derive(Clone)] +pub struct Update<I: ParallelIterator, F> { + base: I, + update_op: F, +} + +impl<I: ParallelIterator + Debug, F> Debug for Update<I, F> { + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + f.debug_struct("Update").field("base", &self.base).finish() + } +} + +impl<I, F> Update<I, F> +where + I: ParallelIterator, +{ + /// Creates a new `Update` iterator. + pub(super) fn new(base: I, update_op: F) -> Self { + Update { base, update_op } + } +} + +impl<I, F> ParallelIterator for Update<I, F> +where + I: ParallelIterator, + F: Fn(&mut I::Item) + Send + Sync, +{ + type Item = I::Item; + + fn drive_unindexed<C>(self, consumer: C) -> C::Result + where + C: UnindexedConsumer<Self::Item>, + { + let consumer1 = UpdateConsumer::new(consumer, &self.update_op); + self.base.drive_unindexed(consumer1) + } + + fn opt_len(&self) -> Option<usize> { + self.base.opt_len() + } +} + +impl<I, F> IndexedParallelIterator for Update<I, F> +where + I: IndexedParallelIterator, + F: Fn(&mut I::Item) + Send + Sync, +{ + fn drive<C>(self, consumer: C) -> C::Result + where + C: Consumer<Self::Item>, + { + let consumer1 = UpdateConsumer::new(consumer, &self.update_op); + self.base.drive(consumer1) + } + + fn len(&self) -> usize { + self.base.len() + } + + fn with_producer<CB>(self, callback: CB) -> CB::Output + where + CB: ProducerCallback<Self::Item>, + { + return self.base.with_producer(Callback { + callback, + update_op: self.update_op, + }); + + struct Callback<CB, F> { + callback: CB, + update_op: F, + } + + impl<T, F, CB> ProducerCallback<T> for Callback<CB, F> + where + CB: ProducerCallback<T>, + F: Fn(&mut T) + Send + Sync, + { + type Output = CB::Output; + + fn callback<P>(self, base: P) -> CB::Output + where + P: Producer<Item = T>, + { + let producer = UpdateProducer { + base, + update_op: &self.update_op, + }; + self.callback.callback(producer) + } + } + } +} + +/// //////////////////////////////////////////////////////////////////////// + +struct UpdateProducer<'f, P, F> { + base: P, + update_op: &'f F, +} + +impl<'f, P, F> Producer for UpdateProducer<'f, P, F> +where + P: Producer, + F: Fn(&mut P::Item) + Send + Sync, +{ + type Item = P::Item; + type IntoIter = UpdateSeq<P::IntoIter, &'f F>; + + fn into_iter(self) -> Self::IntoIter { + UpdateSeq { + base: self.base.into_iter(), + update_op: self.update_op, + } + } + + fn min_len(&self) -> usize { + self.base.min_len() + } + fn max_len(&self) -> usize { + self.base.max_len() + } + + fn split_at(self, index: usize) -> (Self, Self) { + let (left, right) = self.base.split_at(index); + ( + UpdateProducer { + base: left, + update_op: self.update_op, + }, + UpdateProducer { + base: right, + update_op: self.update_op, + }, + ) + } + + fn fold_with<G>(self, folder: G) -> G + where + G: Folder<Self::Item>, + { + let folder1 = UpdateFolder { + base: folder, + update_op: self.update_op, + }; + self.base.fold_with(folder1).base + } +} + +/// //////////////////////////////////////////////////////////////////////// +/// Consumer implementation + +struct UpdateConsumer<'f, C, F> { + base: C, + update_op: &'f F, +} + +impl<'f, C, F> UpdateConsumer<'f, C, F> { + fn new(base: C, update_op: &'f F) -> Self { + UpdateConsumer { base, update_op } + } +} + +impl<'f, T, C, F> Consumer<T> for UpdateConsumer<'f, C, F> +where + C: Consumer<T>, + F: Fn(&mut T) + Send + Sync, +{ + type Folder = UpdateFolder<'f, C::Folder, F>; + type Reducer = C::Reducer; + type Result = C::Result; + + fn split_at(self, index: usize) -> (Self, Self, Self::Reducer) { + let (left, right, reducer) = self.base.split_at(index); + ( + UpdateConsumer::new(left, self.update_op), + UpdateConsumer::new(right, self.update_op), + reducer, + ) + } + + fn into_folder(self) -> Self::Folder { + UpdateFolder { + base: self.base.into_folder(), + update_op: self.update_op, + } + } + + fn full(&self) -> bool { + self.base.full() + } +} + +impl<'f, T, C, F> UnindexedConsumer<T> for UpdateConsumer<'f, C, F> +where + C: UnindexedConsumer<T>, + F: Fn(&mut T) + Send + Sync, +{ + fn split_off_left(&self) -> Self { + UpdateConsumer::new(self.base.split_off_left(), self.update_op) + } + + fn to_reducer(&self) -> Self::Reducer { + self.base.to_reducer() + } +} + +struct UpdateFolder<'f, C, F> { + base: C, + update_op: &'f F, +} + +fn apply<T>(update_op: impl Fn(&mut T)) -> impl Fn(T) -> T { + move |mut item| { + update_op(&mut item); + item + } +} + +impl<'f, T, C, F> Folder<T> for UpdateFolder<'f, C, F> +where + C: Folder<T>, + F: Fn(&mut T), +{ + type Result = C::Result; + + fn consume(self, mut item: T) -> Self { + (self.update_op)(&mut item); + + UpdateFolder { + base: self.base.consume(item), + update_op: self.update_op, + } + } + + fn consume_iter<I>(mut self, iter: I) -> Self + where + I: IntoIterator<Item = T>, + { + let update_op = self.update_op; + self.base = self + .base + .consume_iter(iter.into_iter().map(apply(update_op))); + self + } + + fn complete(self) -> C::Result { + self.base.complete() + } + + fn full(&self) -> bool { + self.base.full() + } +} + +/// Standard Update adaptor, based on `itertools::adaptors::Update` +#[must_use = "iterator adaptors are lazy and do nothing unless consumed"] +#[derive(Debug, Clone)] +struct UpdateSeq<I, F> { + base: I, + update_op: F, +} + +impl<I, F> Iterator for UpdateSeq<I, F> +where + I: Iterator, + F: Fn(&mut I::Item), +{ + type Item = I::Item; + + fn next(&mut self) -> Option<Self::Item> { + let mut v = self.base.next()?; + (self.update_op)(&mut v); + Some(v) + } + + fn size_hint(&self) -> (usize, Option<usize>) { + self.base.size_hint() + } + + fn fold<Acc, G>(self, init: Acc, g: G) -> Acc + where + G: FnMut(Acc, Self::Item) -> Acc, + { + self.base.map(apply(self.update_op)).fold(init, g) + } + + // if possible, re-use inner iterator specializations in collect + fn collect<C>(self) -> C + where + C: ::std::iter::FromIterator<Self::Item>, + { + self.base.map(apply(self.update_op)).collect() + } +} + +impl<I, F> ExactSizeIterator for UpdateSeq<I, F> +where + I: ExactSizeIterator, + F: Fn(&mut I::Item), +{ +} + +impl<I, F> DoubleEndedIterator for UpdateSeq<I, F> +where + I: DoubleEndedIterator, + F: Fn(&mut I::Item), +{ + fn next_back(&mut self) -> Option<Self::Item> { + let mut v = self.base.next_back()?; + (self.update_op)(&mut v); + Some(v) + } +} diff --git a/third_party/rust/rayon/src/iter/while_some.rs b/third_party/rust/rayon/src/iter/while_some.rs new file mode 100644 index 0000000000..215047bc6e --- /dev/null +++ b/third_party/rust/rayon/src/iter/while_some.rs @@ -0,0 +1,154 @@ +use super::plumbing::*; +use super::*; +use std::sync::atomic::{AtomicBool, Ordering}; + +/// `WhileSome` is an iterator that yields the `Some` elements of an iterator, +/// halting as soon as any `None` is produced. +/// +/// This struct is created by the [`while_some()`] method on [`ParallelIterator`] +/// +/// [`while_some()`]: trait.ParallelIterator.html#method.while_some +/// [`ParallelIterator`]: trait.ParallelIterator.html +#[must_use = "iterator adaptors are lazy and do nothing unless consumed"] +#[derive(Debug, Clone)] +pub struct WhileSome<I: ParallelIterator> { + base: I, +} + +impl<I> WhileSome<I> +where + I: ParallelIterator, +{ + /// Creates a new `WhileSome` iterator. + pub(super) fn new(base: I) -> Self { + WhileSome { base } + } +} + +impl<I, T> ParallelIterator for WhileSome<I> +where + I: ParallelIterator<Item = Option<T>>, + T: Send, +{ + type Item = T; + + fn drive_unindexed<C>(self, consumer: C) -> C::Result + where + C: UnindexedConsumer<Self::Item>, + { + let full = AtomicBool::new(false); + let consumer1 = WhileSomeConsumer { + base: consumer, + full: &full, + }; + self.base.drive_unindexed(consumer1) + } +} + +/// //////////////////////////////////////////////////////////////////////// +/// Consumer implementation + +struct WhileSomeConsumer<'f, C> { + base: C, + full: &'f AtomicBool, +} + +impl<'f, T, C> Consumer<Option<T>> for WhileSomeConsumer<'f, C> +where + C: Consumer<T>, + T: Send, +{ + type Folder = WhileSomeFolder<'f, C::Folder>; + type Reducer = C::Reducer; + type Result = C::Result; + + fn split_at(self, index: usize) -> (Self, Self, Self::Reducer) { + let (left, right, reducer) = self.base.split_at(index); + ( + WhileSomeConsumer { base: left, ..self }, + WhileSomeConsumer { + base: right, + ..self + }, + reducer, + ) + } + + fn into_folder(self) -> Self::Folder { + WhileSomeFolder { + base: self.base.into_folder(), + full: self.full, + } + } + + fn full(&self) -> bool { + self.full.load(Ordering::Relaxed) || self.base.full() + } +} + +impl<'f, T, C> UnindexedConsumer<Option<T>> for WhileSomeConsumer<'f, C> +where + C: UnindexedConsumer<T>, + T: Send, +{ + fn split_off_left(&self) -> Self { + WhileSomeConsumer { + base: self.base.split_off_left(), + ..*self + } + } + + fn to_reducer(&self) -> Self::Reducer { + self.base.to_reducer() + } +} + +struct WhileSomeFolder<'f, C> { + base: C, + full: &'f AtomicBool, +} + +impl<'f, T, C> Folder<Option<T>> for WhileSomeFolder<'f, C> +where + C: Folder<T>, +{ + type Result = C::Result; + + fn consume(mut self, item: Option<T>) -> Self { + match item { + Some(item) => self.base = self.base.consume(item), + None => self.full.store(true, Ordering::Relaxed), + } + self + } + + fn consume_iter<I>(mut self, iter: I) -> Self + where + I: IntoIterator<Item = Option<T>>, + { + fn some<T>(full: &AtomicBool) -> impl Fn(&Option<T>) -> bool + '_ { + move |x| match *x { + Some(_) => !full.load(Ordering::Relaxed), + None => { + full.store(true, Ordering::Relaxed); + false + } + } + } + + self.base = self.base.consume_iter( + iter.into_iter() + .take_while(some(self.full)) + .map(Option::unwrap), + ); + self + } + + fn complete(self) -> C::Result { + self.base.complete() + } + + fn full(&self) -> bool { + self.full.load(Ordering::Relaxed) || self.base.full() + } +} diff --git a/third_party/rust/rayon/src/iter/zip.rs b/third_party/rust/rayon/src/iter/zip.rs new file mode 100644 index 0000000000..33823db9e4 --- /dev/null +++ b/third_party/rust/rayon/src/iter/zip.rs @@ -0,0 +1,159 @@ +use super::plumbing::*; +use super::*; +use std::cmp; +use std::iter; + +/// `Zip` is an iterator that zips up `a` and `b` into a single iterator +/// of pairs. This struct is created by the [`zip()`] method on +/// [`IndexedParallelIterator`] +/// +/// [`zip()`]: trait.IndexedParallelIterator.html#method.zip +/// [`IndexedParallelIterator`]: trait.IndexedParallelIterator.html +#[must_use = "iterator adaptors are lazy and do nothing unless consumed"] +#[derive(Debug, Clone)] +pub struct Zip<A: IndexedParallelIterator, B: IndexedParallelIterator> { + a: A, + b: B, +} + +impl<A, B> Zip<A, B> +where + A: IndexedParallelIterator, + B: IndexedParallelIterator, +{ + /// Creates a new `Zip` iterator. + pub(super) fn new(a: A, b: B) -> Self { + Zip { a, b } + } +} + +impl<A, B> ParallelIterator for Zip<A, B> +where + A: IndexedParallelIterator, + B: IndexedParallelIterator, +{ + type Item = (A::Item, B::Item); + + fn drive_unindexed<C>(self, consumer: C) -> C::Result + where + C: UnindexedConsumer<Self::Item>, + { + bridge(self, consumer) + } + + fn opt_len(&self) -> Option<usize> { + Some(self.len()) + } +} + +impl<A, B> IndexedParallelIterator for Zip<A, B> +where + A: IndexedParallelIterator, + B: IndexedParallelIterator, +{ + fn drive<C>(self, consumer: C) -> C::Result + where + C: Consumer<Self::Item>, + { + bridge(self, consumer) + } + + fn len(&self) -> usize { + cmp::min(self.a.len(), self.b.len()) + } + + fn with_producer<CB>(self, callback: CB) -> CB::Output + where + CB: ProducerCallback<Self::Item>, + { + return self.a.with_producer(CallbackA { + callback, + b: self.b, + }); + + struct CallbackA<CB, B> { + callback: CB, + b: B, + } + + impl<CB, ITEM, B> ProducerCallback<ITEM> for CallbackA<CB, B> + where + B: IndexedParallelIterator, + CB: ProducerCallback<(ITEM, B::Item)>, + { + type Output = CB::Output; + + fn callback<A>(self, a_producer: A) -> Self::Output + where + A: Producer<Item = ITEM>, + { + self.b.with_producer(CallbackB { + a_producer, + callback: self.callback, + }) + } + } + + struct CallbackB<CB, A> { + a_producer: A, + callback: CB, + } + + impl<CB, A, ITEM> ProducerCallback<ITEM> for CallbackB<CB, A> + where + A: Producer, + CB: ProducerCallback<(A::Item, ITEM)>, + { + type Output = CB::Output; + + fn callback<B>(self, b_producer: B) -> Self::Output + where + B: Producer<Item = ITEM>, + { + self.callback.callback(ZipProducer { + a: self.a_producer, + b: b_producer, + }) + } + } + } +} + +/// //////////////////////////////////////////////////////////////////////// + +struct ZipProducer<A: Producer, B: Producer> { + a: A, + b: B, +} + +impl<A: Producer, B: Producer> Producer for ZipProducer<A, B> { + type Item = (A::Item, B::Item); + type IntoIter = iter::Zip<A::IntoIter, B::IntoIter>; + + fn into_iter(self) -> Self::IntoIter { + self.a.into_iter().zip(self.b.into_iter()) + } + + fn min_len(&self) -> usize { + cmp::max(self.a.min_len(), self.b.min_len()) + } + + fn max_len(&self) -> usize { + cmp::min(self.a.max_len(), self.b.max_len()) + } + + fn split_at(self, index: usize) -> (Self, Self) { + let (a_left, a_right) = self.a.split_at(index); + let (b_left, b_right) = self.b.split_at(index); + ( + ZipProducer { + a: a_left, + b: b_left, + }, + ZipProducer { + a: a_right, + b: b_right, + }, + ) + } +} diff --git a/third_party/rust/rayon/src/iter/zip_eq.rs b/third_party/rust/rayon/src/iter/zip_eq.rs new file mode 100644 index 0000000000..4e64397d91 --- /dev/null +++ b/third_party/rust/rayon/src/iter/zip_eq.rs @@ -0,0 +1,72 @@ +use super::plumbing::*; +use super::*; + +/// An [`IndexedParallelIterator`] that iterates over two parallel iterators of equal +/// length simultaneously. +/// +/// This struct is created by the [`zip_eq`] method on [`IndexedParallelIterator`], +/// see its documentation for more information. +/// +/// [`zip_eq`]: trait.IndexedParallelIterator.html#method.zip_eq +/// [`IndexedParallelIterator`]: trait.IndexedParallelIterator.html +#[must_use = "iterator adaptors are lazy and do nothing unless consumed"] +#[derive(Debug, Clone)] +pub struct ZipEq<A: IndexedParallelIterator, B: IndexedParallelIterator> { + zip: Zip<A, B>, +} + +impl<A, B> ZipEq<A, B> +where + A: IndexedParallelIterator, + B: IndexedParallelIterator, +{ + /// Creates a new `ZipEq` iterator. + pub(super) fn new(a: A, b: B) -> Self { + ZipEq { + zip: super::Zip::new(a, b), + } + } +} + +impl<A, B> ParallelIterator for ZipEq<A, B> +where + A: IndexedParallelIterator, + B: IndexedParallelIterator, +{ + type Item = (A::Item, B::Item); + + fn drive_unindexed<C>(self, consumer: C) -> C::Result + where + C: UnindexedConsumer<Self::Item>, + { + bridge(self.zip, consumer) + } + + fn opt_len(&self) -> Option<usize> { + Some(self.zip.len()) + } +} + +impl<A, B> IndexedParallelIterator for ZipEq<A, B> +where + A: IndexedParallelIterator, + B: IndexedParallelIterator, +{ + fn drive<C>(self, consumer: C) -> C::Result + where + C: Consumer<Self::Item>, + { + bridge(self.zip, consumer) + } + + fn len(&self) -> usize { + self.zip.len() + } + + fn with_producer<CB>(self, callback: CB) -> CB::Output + where + CB: ProducerCallback<Self::Item>, + { + self.zip.with_producer(callback) + } +} |