Rewrite the style sharing candidate cache.

The style candidate cache had regressed a few times (see #12534), and my
intuition is that being able to disable all style sharing with a single rule in
the page is really unfortunate.

This commit redesigns the style sharing cache in order to be a optimistic cache,
but then reject candidates if they match different sibling-affecting selectors
in the page, for example.

So far the numbers have improved, but not so much as I'd wanted (~10%/20% of
non-incremental restyling time in general). The current implementation is really
dumb though (we recompute and re-match a lot of stuff), so we should be able to
optimise it quite a bit.

I have different ideas for improving it (that may or may not work), apart of the
low-hanging fruit like don't re-matching candidates all the time but I have to
measure the real impact.

Also, I need to verify it against try.
This commit is contained in:
Emilio Cobos Álvarez 2016-07-29 17:24:12 -07:00
parent ec53136863
commit 3af774bd75
No known key found for this signature in database
GPG key ID: 056B727BB9C1027C
23 changed files with 467 additions and 236 deletions

View file

@ -12,6 +12,8 @@ use dom::{OpaqueNode, TNode, UnsafeNode};
use std::mem;
use std::sync::atomic::Ordering;
use traversal::{RestyleResult, DomTraversalContext};
use traversal::{STYLE_SHARING_CACHE_HITS, STYLE_SHARING_CACHE_MISSES};
use util::opts;
use workqueue::{WorkQueue, WorkUnit, WorkerProxy};
#[allow(dead_code)]
@ -42,13 +44,28 @@ pub fn run_queue_with_custom_work_data_type<To, F, SharedContext: Sync>(
pub fn traverse_dom<N, C>(root: N,
queue_data: &C::SharedContext,
queue: &mut WorkQueue<C::SharedContext, WorkQueueData>)
where N: TNode, C: DomTraversalContext<N> {
where N: TNode,
C: DomTraversalContext<N>
{
if opts::get().style_sharing_stats {
STYLE_SHARING_CACHE_HITS.store(0, Ordering::SeqCst);
STYLE_SHARING_CACHE_MISSES.store(0, Ordering::SeqCst);
}
run_queue_with_custom_work_data_type(queue, |queue| {
queue.push(WorkUnit {
fun: top_down_dom::<N, C>,
data: (Box::new(vec![root.to_unsafe()]), root.opaque()),
});
}, queue_data);
if opts::get().style_sharing_stats {
let hits = STYLE_SHARING_CACHE_HITS.load(Ordering::SeqCst);
let misses = STYLE_SHARING_CACHE_MISSES.load(Ordering::SeqCst);
println!("Style sharing stats:");
println!(" * Hits: {}", hits);
println!(" * Misses: {}", misses);
}
}
/// A parallel top-down DOM traversal.
@ -102,6 +119,10 @@ fn top_down_dom<N, C>(unsafe_nodes: UnsafeNodeList,
}
}
// NB: In parallel traversal mode we have to purge the LRU cache in order to
// be able to access it without races.
context.local_context().style_sharing_candidate_cache.borrow_mut().clear();
for chunk in discovered_child_nodes.chunks(CHUNK_SIZE) {
proxy.push(WorkUnit {
fun: top_down_dom::<N, C>,