style: Make style parallel traversal more tunable at runtime

This adds two prefs to configure the parallel traversal work item size
and kick-off threshold, but otherwise shouldn't change behavior.

I switched from iterator generics to just a slice while at it, mostly
for simplicity, but there is a trade-off:

  * When switching from sequential to parallel traversal, we potentially
    pay the price of memmoving the VecDeque around once to make them a
    contiguous slice.

  * However we win in the common case of the smaller-than-work-unit size
    in which case we no longer need to copy stuff to a WorkItem. So I
    think overall this should be an improvement.

Differential Revision: https://phabricator.services.mozilla.com/D178656
This commit is contained in:
Emilio Cobos Álvarez 2023-05-24 06:36:08 +00:00 committed by Martin Robinson
parent ea3fcce25f
commit d49b014c78
2 changed files with 88 additions and 87 deletions

View file

@ -11,7 +11,7 @@ use crate::context::{PerThreadTraversalStatistics, StyleContext};
use crate::context::{ThreadLocalStyleContext, TraversalStatistics};
use crate::dom::{SendNode, TElement, TNode};
use crate::parallel;
use crate::parallel::{DispatchMode, WORK_UNIT_MAX};
use crate::parallel::{work_unit_max, DispatchMode};
use crate::scoped_tls::ScopedTLS;
use crate::traversal::{DomTraversal, PerLevelTraversalData, PreTraverseToken};
use rayon;
@ -48,17 +48,19 @@ fn report_statistics(stats: &PerThreadTraversalStatistics) {
gecko_stats.mStylesReused += stats.styles_reused;
}
/// Do a DOM traversal for top-down and (optionally) bottom-up processing,
/// generic over `D`.
fn parallelism_threshold() -> usize {
static_prefs::pref!("layout.css.stylo-parallelism-threshold") as usize
}
/// Do a DOM traversal for top-down and (optionally) bottom-up processing, generic over `D`.
///
/// We use an adaptive traversal strategy. We start out with simple sequential
/// processing, until we arrive at a wide enough level in the DOM that the
/// parallel traversal would parallelize it. If a thread pool is provided, we
/// then transfer control over to the parallel traversal.
/// We use an adaptive traversal strategy. We start out with simple sequential processing, until we
/// arrive at a wide enough level in the DOM that the parallel traversal would parallelize it.
/// If a thread pool is provided, we then transfer control over to the parallel traversal.
///
/// Returns true if the traversal was parallel, and also returns the statistics
/// object containing information on nodes traversed (on nightly only). Not
/// all of its fields will be initialized since we don't call finish().
/// Returns true if the traversal was parallel, and also returns the statistics object containing
/// information on nodes traversed (on nightly only). Not all of its fields will be initialized
/// since we don't call finish().
pub fn traverse_dom<E, D>(
traversal: &D,
token: PreTraverseToken<E>,
@ -100,7 +102,9 @@ where
// Process the nodes breadth-first, just like the parallel traversal does.
// This helps keep similar traversal characteristics for the style sharing
// cache.
let mut discovered = VecDeque::<SendNode<E::ConcreteNode>>::with_capacity(WORK_UNIT_MAX * 2);
let work_unit_max = work_unit_max();
let parallelism_threshold = parallelism_threshold();
let mut discovered = VecDeque::<SendNode<E::ConcreteNode>>::with_capacity(work_unit_max * 2);
let mut depth = root.depth();
let mut nodes_remaining_at_current_depth = 1;
discovered.push_back(unsafe { SendNode::new(root.as_node()) });
@ -122,17 +126,21 @@ where
);
nodes_remaining_at_current_depth -= 1;
if nodes_remaining_at_current_depth == 0 {
// If there is enough work to parallelize over, and the caller allows parallelism, switch
// to the parallel driver. We do this only when moving to the next level in the dom so that
// we can pass the same depth for all the children.
if nodes_remaining_at_current_depth != 0 {
continue;
}
depth += 1;
// If there is enough work to parallelize over, and the caller allows
// parallelism, switch to the parallel driver. We do this only when
// moving to the next level in the dom so that we can pass the same
// depth for all the children.
if pool.is_some() && discovered.len() > WORK_UNIT_MAX {
if pool.is_some() &&
discovered.len() > parallelism_threshold &&
parallelism_threshold > 0
{
let pool = pool.unwrap();
let tls = ScopedTLS::<ThreadLocalStyleContext<E>>::new(pool);
let root_opaque = root.as_node().opaque();
let drain = discovered.drain(..);
pool.scope_fifo(|scope| {
// Enable a breadth-first rayon traversal. This causes the work
// queue to be always FIFO, rather than FIFO for stealers and
@ -142,7 +150,7 @@ where
#[cfg(feature = "gecko")]
gecko_profiler_label!(Layout, StyleComputation);
parallel::traverse_nodes(
drain,
discovered.make_contiguous(),
DispatchMode::TailCall,
/* recursion_ok = */ true,
root_opaque,
@ -161,7 +169,6 @@ where
}
nodes_remaining_at_current_depth = discovered.len();
}
}
// Collect statistics from thread-locals if requested.
if dump_stats || report_stats {

View file

@ -26,8 +26,6 @@ use crate::context::{StyleContext, ThreadLocalStyleContext};
use crate::dom::{OpaqueNode, SendNode, TElement};
use crate::scoped_tls::ScopedTLS;
use crate::traversal::{DomTraversal, PerLevelTraversalData};
use arrayvec::ArrayVec;
use itertools::Itertools;
use rayon;
use smallvec::SmallVec;
@ -59,23 +57,10 @@ pub const STYLE_THREAD_STACK_SIZE_KB: usize = 512;
///
pub const STACK_SAFETY_MARGIN_KB: usize = 168;
/// The maximum number of child nodes that we will process as a single unit.
///
/// Larger values will increase style sharing cache hits and general DOM
/// locality at the expense of decreased opportunities for parallelism. There
/// are some measurements in
/// https://bugzilla.mozilla.org/show_bug.cgi?id=1385982#c11 and comments 12
/// and 13 that investigate some slightly different values for the work unit
/// size. If the size is significantly increased, make sure to adjust the
/// condition for kicking off a new work unit in top_down_dom, because
/// otherwise we're likely to end up doing too much work serially. For
/// example, the condition there could become some fraction of WORK_UNIT_MAX
/// instead of WORK_UNIT_MAX.
pub const WORK_UNIT_MAX: usize = 16;
/// A set of nodes, sized to the work unit. This gets copied when sent to other
/// threads, so we keep it compact.
type WorkUnit<N> = ArrayVec<SendNode<N>, WORK_UNIT_MAX>;
/// See documentation of the pref for performance characteristics.
pub fn work_unit_max() -> usize {
static_prefs::pref!("layout.css.stylo-work-unit-size") as usize
}
/// A callback to create our thread local context. This needs to be
/// out of line so we don't allocate stack space for the entire struct
@ -115,14 +100,15 @@ fn top_down_dom<'a, 'scope, E, D>(
E: TElement + 'scope,
D: DomTraversal<E>,
{
debug_assert!(nodes.len() <= WORK_UNIT_MAX);
let work_unit_max = work_unit_max();
debug_assert!(nodes.len() <= work_unit_max);
// We set this below, when we have a borrow of the thread-local-context
// available.
let recursion_ok;
// Collect all the children of the elements in our work unit. This will
// contain the combined children of up to WORK_UNIT_MAX nodes, which may
// contain the combined children of up to work_unit_max nodes, which may
// be numerous. As such, we store it in a large SmallVec to minimize heap-
// spilling, and never move it.
let mut discovered_child_nodes = SmallVec::<[SendNode<E::ConcreteNode>; 128]>::new();
@ -171,19 +157,19 @@ fn top_down_dom<'a, 'scope, E, D>(
// following.
//
// The worst case behavior for waiting until we have a full work
// item is a deep tree which has WORK_UNIT_MAX "linear" branches,
// hence WORK_UNIT_MAX elements at each level. Such a tree would
// item is a deep tree which has work_unit_max "linear" branches,
// hence work_unit_max elements at each level. Such a tree would
// end up getting processed entirely sequentially, because we would
// process each level one at a time as a single work unit, whether
// via our end-of-loop tail call or not. If we kicked off a
// traversal as soon as we discovered kids, we would instead
// process such a tree more or less with a thread-per-branch,
// multiplexed across our actual threadpool.
if discovered_child_nodes.len() >= WORK_UNIT_MAX {
if discovered_child_nodes.len() >= work_unit_max {
let mut traversal_data_copy = traversal_data.clone();
traversal_data_copy.current_dom_depth += 1;
traverse_nodes(
discovered_child_nodes.drain(..),
&discovered_child_nodes,
DispatchMode::NotTailCall,
recursion_ok,
root,
@ -193,6 +179,7 @@ fn top_down_dom<'a, 'scope, E, D>(
traversal,
tls,
);
discovered_child_nodes.clear();
}
let node = **n;
@ -213,7 +200,7 @@ fn top_down_dom<'a, 'scope, E, D>(
if !discovered_child_nodes.is_empty() {
traversal_data.current_dom_depth += 1;
traverse_nodes(
discovered_child_nodes.drain(..),
&discovered_child_nodes,
DispatchMode::TailCall,
recursion_ok,
root,
@ -245,8 +232,8 @@ impl DispatchMode {
/// Enqueues |nodes| for processing, possibly on this thread if the tail call
/// conditions are met.
#[inline]
pub fn traverse_nodes<'a, 'scope, E, D, I>(
nodes: I,
pub fn traverse_nodes<'a, 'scope, E, D>(
nodes: &[SendNode<E::ConcreteNode>],
mode: DispatchMode,
recursion_ok: bool,
root: OpaqueNode,
@ -258,7 +245,6 @@ pub fn traverse_nodes<'a, 'scope, E, D, I>(
) where
E: TElement + 'scope,
D: DomTraversal<E>,
I: ExactSizeIterator<Item = SendNode<E::ConcreteNode>>,
{
debug_assert_ne!(nodes.len(), 0);
@ -272,29 +258,37 @@ pub fn traverse_nodes<'a, 'scope, E, D, I>(
let may_dispatch_tail =
mode.is_tail_call() && recursion_ok && !pool.current_thread_has_pending_tasks().unwrap();
// In the common case, our children fit within a single work unit, in which
// case we can pass the SmallVec directly and avoid extra allocation.
if nodes.len() <= WORK_UNIT_MAX {
let work: WorkUnit<E::ConcreteNode> = nodes.collect();
let work_unit_max = work_unit_max();
// In the common case, our children fit within a single work unit, in which case we can pass
// the nodes directly and avoid extra allocation.
if nodes.len() <= work_unit_max {
if may_dispatch_tail {
top_down_dom(&work, root, traversal_data, scope, pool, traversal, tls);
top_down_dom(&nodes, root, traversal_data, scope, pool, traversal, tls);
} else {
let work = nodes.to_vec();
scope.spawn_fifo(move |scope| {
#[cfg(feature = "gecko")]
gecko_profiler_label!(Layout, StyleComputation);
let work = work;
top_down_dom(&work, root, traversal_data, scope, pool, traversal, tls);
});
}
} else {
for chunk in nodes.chunks(WORK_UNIT_MAX).into_iter() {
let nodes: WorkUnit<E::ConcreteNode> = chunk.collect();
for chunk in nodes.chunks(work_unit_max) {
let work = chunk.to_vec();
let traversal_data_copy = traversal_data.clone();
scope.spawn_fifo(move |scope| {
#[cfg(feature = "gecko")]
gecko_profiler_label!(Layout, StyleComputation);
let n = nodes;
top_down_dom(&*n, root, traversal_data_copy, scope, pool, traversal, tls)
let work = work;
top_down_dom(
&work,
root,
traversal_data_copy,
scope,
pool,
traversal,
tls,
)
});
}
}