Mirrors/bevy
2
0
Fork 0
mirror of https://github.com/bevyengine/bevy synced 2024-11-22 04:33:37 +00:00
Code Issues Projects Releases Packages Wiki Activity

Opportunistically use dense iter for archetypal iteration in Par_iter (#14673)

Browse source 
# Objective

- follow of #14049 ,we could use it on our Parallel Iterator,this pr
also unified the used function in both regular iter and parallel
iterations.


## Performance 


![image](https://github.com/user-attachments/assets/cba700bc-169c-4b58-b504-823bdca8ec05)

no performance regression for regular itertaion

3.5X faster in hybrid parallel iteraion,this number is far greater than
the benefits obtained in regular iteration(~1.81) because mutable
iterations on continuous memory can effectively reduce the cost of
mataining core cache coherence
This commit is contained in:
re0312 2024-09-04 07:41:10 +08:00 committed by GitHub
parent ec728c31c1
commit 739007f148
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
5 changed files with 121 additions and 76 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

2
benches/benches/bevy_ecs/iteration/iter_simple_foreach_hybrid.rs
View file

@ -20,7 +20,7 @@ impl<'w> Benchmark<'w> {
let mut v = vec![];
for _ in 0..10000 {
world.spawn((TableData(0.0), SparseData(0.0))).id();
world.spawn((TableData(0.0), SparseData(0.0)));
v.push(world.spawn(TableData(0.)).id());
}

5
benches/benches/bevy_ecs/iteration/mod.rs
View file

@ -20,6 +20,7 @@ mod iter_simple_system;
mod iter_simple_wide;
mod iter_simple_wide_sparse_set;
mod par_iter_simple;
mod par_iter_simple_foreach_hybrid;
use heavy_compute::*;
@ -135,4 +136,8 @@ fn par_iter_simple(c: &mut Criterion) {
b.iter(move || bench.run());
});
}
group.bench_function(format!("hybrid"), |b| {
let mut bench = par_iter_simple_foreach_hybrid::Benchmark::new();
b.iter(move || bench.run());
});
}

45
benches/benches/bevy_ecs/iteration/par_iter_simple_foreach_hybrid.rs Normal file
View file

@ -0,0 +1,45 @@
use bevy_ecs::prelude::*;
use bevy_tasks::{ComputeTaskPool, TaskPool};
use rand::{prelude::SliceRandom, SeedableRng};
use rand_chacha::ChaCha8Rng;
#[derive(Component, Copy, Clone)]
struct TableData(f32);
#[derive(Component, Copy, Clone)]
#[component(storage = "SparseSet")]
struct SparseData(f32);
fn deterministic_rand() -> ChaCha8Rng {
ChaCha8Rng::seed_from_u64(42)
}
pub struct Benchmark<'w>(World, QueryState<(&'w mut TableData, &'w SparseData)>);
impl<'w> Benchmark<'w> {
pub fn new() -> Self {
let mut world = World::new();
ComputeTaskPool::get_or_init(TaskPool::default);
let mut v = vec![];
for _ in 0..100000 {
world.spawn((TableData(0.0), SparseData(0.0)));
v.push(world.spawn(TableData(0.)).id());
}
// by shuffling ,randomize the archetype iteration order to significantly deviate from the table order. This maximizes the loss of cache locality during archetype-based iteration.
v.shuffle(&mut deterministic_rand());
for e in v.into_iter() {
world.entity_mut(e).despawn();
}
let query = world.query::<(&mut TableData, &SparseData)>();
Self(world, query)
}
#[inline(never)]
pub fn run(&mut self) {
self.1
.par_iter_mut(&mut self.0)
.for_each(|(mut v1, v2)| v1.0 += v2.0)
}
}

112
crates/bevy_ecs/src/query/iter.rs
View file

@ -122,6 +122,67 @@ impl<'w, 's, D: QueryData, F: QueryFilter> QueryIter<'w, 's, D, F> {
}
}
/// Executes the equivalent of [`Iterator::fold`] over a contiguous segment
/// from an storage.
///
/// # Safety
/// - `range` must be in `[0, storage::entity_count)` or None.
#[inline]
pub(super) unsafe fn fold_over_storage_range<B, Func>(
&mut self,
mut accum: B,
func: &mut Func,
storage: StorageId,
range: Option<Range<usize>>,
) -> B
where
Func: FnMut(B, D::Item<'w>) -> B,
{
if self.cursor.is_dense {
// SAFETY: `self.cursor.is_dense` is true, so storage ids are guaranteed to be table ids.
let table_id = unsafe { storage.table_id };
// SAFETY: Matched table IDs are guaranteed to still exist.
let table = unsafe { self.tables.get(table_id).debug_checked_unwrap() };
let range = range.unwrap_or(0..table.entity_count());
accum =
// SAFETY:
// - The fetched table matches both D and F
// - caller ensures `range` is within `[0, table.entity_count)`
// - The if block ensures that the query iteration is dense
unsafe { self.fold_over_table_range(accum, func, table, range) };
} else {
// SAFETY: `self.cursor.is_dense` is false, so storage ids are guaranteed to be archetype ids.
let archetype_id = unsafe { storage.archetype_id };
// SAFETY: Matched archetype IDs are guaranteed to still exist.
let archetype = unsafe { self.archetypes.get(archetype_id).debug_checked_unwrap() };
// SAFETY: Matched table IDs are guaranteed to still exist.
let table = unsafe { self.tables.get(archetype.table_id()).debug_checked_unwrap() };
let range = range.unwrap_or(0..archetype.len());
// When an archetype and its table have equal entity counts, dense iteration can be safely used.
// this leverages cache locality to optimize performance.
if table.entity_count() == archetype.len() {
accum =
// SAFETY:
// - The fetched archetype matches both D and F
// - The provided archetype and its' table have the same length.
// - caller ensures `range` is within `[0, archetype.len)`
// - The if block ensures that the query iteration is not dense.
unsafe { self.fold_over_dense_archetype_range(accum, func, archetype,range) };
} else {
accum =
// SAFETY:
// - The fetched archetype matches both D and F
// - caller ensures `range` is within `[0, archetype.len)`
// - The if block ensures that the query iteration is not dense.
unsafe { self.fold_over_archetype_range(accum, func, archetype,range) };
}
}
accum
}
/// Executes the equivalent of [`Iterator::fold`] over a contiguous segment
/// from an table.
///
@ -143,7 +204,7 @@ impl<'w, 's, D: QueryData, F: QueryFilter> QueryIter<'w, 's, D, F> {
if table.is_empty() {
return accum;
}
assert!(
debug_assert!(
rows.end <= u32::MAX as usize,
"TableRow is only valid up to u32::MAX"
);
@ -267,12 +328,11 @@ impl<'w, 's, D: QueryData, F: QueryFilter> QueryIter<'w, 's, D, F> {
if archetype.is_empty() {
return accum;
}
assert!(
debug_assert!(
rows.end <= u32::MAX as usize,
"TableRow is only valid up to u32::MAX"
);
let table = self.tables.get(archetype.table_id()).debug_checked_unwrap();
debug_assert!(
archetype.len() == table.entity_count(),
"archetype and it's table must have the same length. "
@ -1032,48 +1092,10 @@ impl<'w, 's, D: QueryData, F: QueryFilter> Iterator for QueryIter<'w, 's, D, F>
accum = func(accum, item);
}
if self.cursor.is_dense {
for id in self.cursor.storage_id_iter.clone() {
// SAFETY: `self.cursor.is_dense` is true, so storage ids are guaranteed to be table ids.
let table_id = unsafe { id.table_id };
// SAFETY: Matched table IDs are guaranteed to still exist.
let table = unsafe { self.tables.get(table_id).debug_checked_unwrap() };
accum =
// SAFETY:
// - The fetched table matches both D and F
// - The provided range is equivalent to [0, table.entity_count)
// - The if block ensures that the query iteration is dense
unsafe { self.fold_over_table_range(accum, &mut func, table, 0..table.entity_count()) };
}
} else {
for id in self.cursor.storage_id_iter.clone() {
// SAFETY: `self.cursor.is_dense` is false, so storage ids are guaranteed to be archetype ids.
let archetype_id = unsafe { id.archetype_id };
// SAFETY: Matched archetype IDs are guaranteed to still exist.
let archetype = unsafe { self.archetypes.get(archetype_id).debug_checked_unwrap() };
// SAFETY: Matched table IDs are guaranteed to still exist.
let table = unsafe { self.tables.get(archetype.table_id()).debug_checked_unwrap() };
// When an archetype and its table have equal entity counts, dense iteration can be safely used.
// this leverages cache locality to optimize performance.
if table.entity_count() == archetype.len() {
accum =
// SAFETY:
// - The fetched archetype matches both D and F
// - The provided archetype and its' table have the same length.
// - The provided range is equivalent to [0, archetype.len)
// - The if block ensures that the query iteration is not dense.
unsafe { self.fold_over_dense_archetype_range(accum, &mut func, archetype, 0..archetype.len()) };
} else {
accum =
// SAFETY:
// - The fetched archetype matches both D and F
// - The provided range is equivalent to [0, archetype.len)
// - The if block ensures that the query iteration is not dense.
unsafe { self.fold_over_archetype_range(accum, &mut func, archetype, 0..archetype.len()) };
}
}
for id in self.cursor.storage_id_iter.clone().copied() {
// SAFETY:
// - The range(None) is equivalent to [0, storage.entity_count)
accum = unsafe { self.fold_over_storage_range(accum, &mut func, id, None) };
}
accum
}

33
crates/bevy_ecs/src/query/state.rs
View file

@ -1505,25 +1505,7 @@ impl<D: QueryData, F: QueryFilter> QueryState<D, F> {
let mut iter = self.iter_unchecked_manual(world, last_run, this_run);
let mut accum = init_accum();
for storage_id in queue {
if self.is_dense {
let id = storage_id.table_id;
let table = &world.storages().tables.get(id).debug_checked_unwrap();
accum = iter.fold_over_table_range(
accum,
&mut func,
table,
0..table.entity_count(),
);
} else {
let id = storage_id.archetype_id;
let archetype = world.archetypes().get(id).debug_checked_unwrap();
accum = iter.fold_over_archetype_range(
accum,
&mut func,
archetype,
0..archetype.len(),
);
}
accum = iter.fold_over_storage_range(accum, &mut func, storage_id, None);
}
});
};
@ -1539,17 +1521,8 @@ impl<D: QueryData, F: QueryFilter> QueryState<D, F> {
#[cfg(feature = "trace")]
let _span = self.par_iter_span.enter();
let accum = init_accum();
if self.is_dense {
let id = storage_id.table_id;
let table = world.storages().tables.get(id).debug_checked_unwrap();
self.iter_unchecked_manual(world, last_run, this_run)
.fold_over_table_range(accum, &mut func, table, batch);
} else {
let id = storage_id.archetype_id;
let archetype = world.archetypes().get(id).debug_checked_unwrap();
self.iter_unchecked_manual(world, last_run, this_run)
.fold_over_archetype_range(accum, &mut func, archetype, batch);
}
self.iter_unchecked_manual(world, last_run, this_run)
.fold_over_storage_range(accum, &mut func, storage_id, Some(batch));
});
}
};