layout: Port parallel layout over to a generic "work queue"

infrastructure.

The work queue accepts abstract generic "work units", which in this case
are layout operations. The same speedups have been observed.
This commit is contained in:
Patrick Walton 2014-01-23 13:43:03 -08:00
parent 86c29d253a
commit 18a2050a64
12 changed files with 518 additions and 351 deletions

View file

@ -198,6 +198,9 @@ impl IOCompositor {
// Drain compositor port, sometimes messages contain channels that are blocking
// another task from finishing (i.e. SetIds)
while self.port.try_recv().is_some() {}
// Tell the profiler to shut down.
self.profiler_chan.send(time::ExitMsg);
}
fn handle_message(&mut self) {

View file

@ -329,7 +329,7 @@ impl BlockFlow {
// top or bottom borders nor top or bottom padding, and it has a 'height' of either 0 or 'auto',
// and it does not contain a line box, and all of its in-flow children's margins (if any) collapse.
let screen_height = ctx.shared.screen_size.height;
let screen_height = ctx.screen_size.height;
let mut height = if self.is_root {
// FIXME(pcwalton): The max is taken here so that you can scroll the page, but this is
@ -627,7 +627,7 @@ impl Flow for BlockFlow {
if self.is_root {
debug!("Setting root position");
self.base.position.origin = Au::zero_point();
self.base.position.size.width = ctx.shared.screen_size.width;
self.base.position.size.width = ctx.screen_size.width;
self.base.floats_in = FloatContext::new(self.base.num_floats);
self.base.flags_info.flags.set_inorder(false);
}
@ -672,7 +672,7 @@ impl Flow for BlockFlow {
margin_bottom,
margin_left));
let screen_size = ctx.shared.screen_size;
let screen_size = ctx.screen_size;
let (x, w) = box_.get_x_coord_and_new_width_if_fixed(screen_size.width,
screen_size.height,
width,

View file

@ -1338,7 +1338,7 @@ impl Box {
iframe_box.pipeline_id,
iframe_box.subpage_id);
let msg = FrameRectMsg(iframe_box.pipeline_id, iframe_box.subpage_id, rect);
layout_context.shared.constellation_chan.send(msg)
layout_context.constellation_chan.send(msg)
}
}

View file

@ -32,6 +32,7 @@ use layout::text::TextRunScanner;
use layout::util::LayoutDataAccess;
use layout::wrapper::{LayoutNode, PostorderNodeMutTraversal};
use gfx::font_context::FontContext;
use script::dom::element::{HTMLIframeElementTypeId, HTMLImageElementTypeId};
use script::dom::node::{CommentNodeTypeId, DoctypeNodeTypeId, DocumentFragmentNodeTypeId};
use script::dom::node::{DocumentNodeTypeId, ElementNodeTypeId, TextNodeTypeId};
@ -209,14 +210,19 @@ pub struct FlowConstructor<'a> {
///
/// FIXME(pcwalton): This is going to have to be atomic; can't we do something better?
next_flow_id: RefCell<int>,
/// The font context.
font_context: ~FontContext,
}
impl<'fc> FlowConstructor<'fc> {
/// Creates a new flow constructor.
pub fn init<'a>(layout_context: &'a mut LayoutContext) -> FlowConstructor<'a> {
let font_context = ~FontContext::new(layout_context.font_context_info.clone());
FlowConstructor {
layout_context: layout_context,
next_flow_id: RefCell::new(0),
font_context: font_context,
}
}
@ -235,7 +241,7 @@ impl<'fc> FlowConstructor<'fc> {
Some(url) => {
// FIXME(pcwalton): The fact that image boxes store the cache within them makes
// little sense to me.
Some(ImageBoxInfo::new(&node, url, self.layout_context.shared.image_cache.clone()))
Some(ImageBoxInfo::new(&node, url, self.layout_context.image_cache.clone()))
}
}
}
@ -262,20 +268,19 @@ impl<'fc> FlowConstructor<'fc> {
/// otherwise.
#[inline(always)]
fn flush_inline_boxes_to_flow(&mut self, boxes: ~[Box], flow: &mut ~Flow, node: LayoutNode) {
if boxes.len() > 0 {
let inline_base = BaseFlow::new(self.next_flow_id(), node);
let mut inline_flow = ~InlineFlow::from_boxes(inline_base, boxes) as ~Flow;
self.layout_context.shared.leaf_set.access(|leaf_set| leaf_set.insert(&inline_flow));
TextRunScanner::new().scan_for_runs(self.layout_context, inline_flow);
let mut inline_flow = Some(inline_flow);
self.layout_context.shared.leaf_set.access(|leaf_set| {
flow.add_new_child(inline_flow.take_unwrap(), leaf_set)
})
if boxes.len() == 0 {
return
}
let inline_base = BaseFlow::new(self.next_flow_id(), node);
let mut inline_flow = ~InlineFlow::from_boxes(inline_base, boxes) as ~Flow;
self.layout_context.leaf_set.access(|leaf_set| leaf_set.insert(&inline_flow));
TextRunScanner::new().scan_for_runs(self.font_context, inline_flow);
let mut inline_flow = Some(inline_flow);
self.layout_context.leaf_set.access(|leaf_set| {
flow.add_new_child(inline_flow.take_unwrap(), leaf_set)
})
}
/// Creates an inline flow from a set of inline boxes, if present, and adds it as a child of
@ -319,7 +324,7 @@ impl<'fc> FlowConstructor<'fc> {
flow,
node);
let mut kid_flow = Some(kid_flow);
self.layout_context.shared.leaf_set.access(|leaf_set| {
self.layout_context.leaf_set.access(|leaf_set| {
flow.add_new_child(kid_flow.take_unwrap(), leaf_set)
})
}
@ -362,7 +367,7 @@ impl<'fc> FlowConstructor<'fc> {
// Push the flow generated by the {ib} split onto our list of
// flows.
let mut kid_flow = Some(kid_flow);
self.layout_context.shared.leaf_set.access(|leaf_set| {
self.layout_context.leaf_set.access(|leaf_set| {
flow.add_new_child(kid_flow.take_unwrap(), leaf_set)
})
}
@ -391,7 +396,7 @@ impl<'fc> FlowConstructor<'fc> {
let box_ = self.build_box_for_node(node);
let mut flow = ~BlockFlow::from_box(base, box_, is_fixed) as ~Flow;
self.layout_context.shared.leaf_set.access(|leaf_set| leaf_set.insert(&flow));
self.layout_context.leaf_set.access(|leaf_set| leaf_set.insert(&flow));
self.build_children_of_block_flow(&mut flow, node);
flow
@ -406,7 +411,7 @@ impl<'fc> FlowConstructor<'fc> {
let mut flow = ~BlockFlow::float_from_box(base, float_type, box_) as ~Flow;
self.layout_context.shared.leaf_set.access(|leaf_set| leaf_set.insert(&flow));
self.layout_context.leaf_set.access(|leaf_set| leaf_set.insert(&flow));
self.build_children_of_block_flow(&mut flow, node);
flow
@ -484,7 +489,7 @@ impl<'fc> FlowConstructor<'fc> {
fn set_inline_info_for_inline_child(&mut self, boxes: &mut ~[Box], parent_node: LayoutNode) {
let parent_box = self.build_box_for_node(parent_node);
let font_style = parent_box.font_style();
let font_group = self.layout_context.font_ctx.get_resolved_font_for_style(&font_style);
let font_group = self.font_context.get_resolved_font_for_style(&font_style);
let (font_ascent,font_descent) = font_group.borrow().with_mut( |fg| {
fg.fonts[0].borrow().with_mut( |font| {
(font.metrics.ascent,font.metrics.descent)
@ -569,7 +574,7 @@ impl<'a> PostorderNodeMutTraversal for FlowConstructor<'a> {
// `display: none` contributes no flow construction result. Nuke the flow construction
// results of children.
(display::none, _, _) => {
self.layout_context.shared.leaf_set.access(|leaf_set| {
self.layout_context.leaf_set.access(|leaf_set| {
for child in node.children() {
let mut old_result = child.swap_out_construction_result();
old_result.destroy(leaf_set)

View file

@ -5,17 +5,26 @@
//! Data needed by the layout task.
use extra::arc::MutexArc;
use green::task::GreenTask;
use layout::flow::LeafSet;
use std::cast;
use std::ptr;
use std::rt::Runtime;
use std::rt::local::Local;
use std::rt::task::Task;
use geom::size::Size2D;
use gfx::font_context::FontContext;
use gfx::font_context::{FontContext, FontContextInfo};
use servo_msg::constellation_msg::ConstellationChan;
use servo_net::local_image_cache::LocalImageCache;
use servo_util::geometry::Au;
#[thread_local]
static mut FONT_CONTEXT: *mut FontContext = 0 as *mut FontContext;
/// Data shared by all layout workers.
#[deriving(Clone)]
pub struct SharedLayoutInfo {
pub struct LayoutContext {
/// The local image cache.
image_cache: MutexArc<LocalImageCache>,
@ -27,14 +36,30 @@ pub struct SharedLayoutInfo {
/// The set of leaf flows.
leaf_set: MutexArc<LeafSet>,
/// Information needed to construct a font context.
font_context_info: FontContextInfo,
}
/// Data specific to a layout worker.
pub struct LayoutContext {
/// Shared layout info.
shared: SharedLayoutInfo,
impl LayoutContext {
pub fn font_context<'a>(&'a mut self) -> &'a mut FontContext {
// Sanity check.
let mut task = Local::borrow(None::<Task>);
match task.get().maybe_take_runtime::<GreenTask>() {
Some(green) => {
task.get().put_runtime(green as ~Runtime);
fail!("can't call this on a green task!")
}
None => {}
}
/// The current font context.
font_ctx: ~FontContext,
unsafe {
if FONT_CONTEXT == ptr::mut_null() {
let context = ~FontContext::new(self.font_context_info.clone());
FONT_CONTEXT = cast::transmute(context)
}
cast::transmute(FONT_CONTEXT)
}
}
}

View file

@ -9,7 +9,7 @@ use css::matching::MatchMethods;
use css::select::new_stylist;
use css::node_style::StyledNode;
use layout::construct::{FlowConstructionResult, FlowConstructor, NoConstructionResult};
use layout::context::{LayoutContext, SharedLayoutInfo};
use layout::context::LayoutContext;
use layout::display_list_builder::{DisplayListBuilder, ToGfxColor};
use layout::extra::LayoutAuxMethods;
use layout::flow::{Flow, ImmutableFlowUtils, LeafSet, MutableFlowUtils, MutableOwnedFlowUtils};
@ -17,7 +17,8 @@ use layout::flow::{PreorderFlowTraversal, PostorderFlowTraversal};
use layout::flow;
use layout::incremental::RestyleDamage;
use layout::parallel::{AssignHeightsAndStoreOverflowTraversalKind, BubbleWidthsTraversalKind};
use layout::parallel::{ParallelPostorderFlowTraversal};
use layout::parallel::{UnsafeFlow};
use layout::parallel;
use layout::util::{LayoutDataAccess, OpaqueNode, LayoutDataWrapper};
use layout::wrapper::LayoutNode;
@ -25,7 +26,7 @@ use extra::arc::{Arc, MutexArc, RWArc};
use geom::rect::Rect;
use geom::size::Size2D;
use gfx::display_list::{ClipDisplayItemClass, DisplayItem, DisplayItemIterator, DisplayList};
use gfx::font_context::{FontContext, FontContextInfo};
use gfx::font_context::FontContextInfo;
use gfx::opts::Opts;
use gfx::render_task::{RenderMsg, RenderChan, RenderLayer};
use gfx::{render_task, color};
@ -46,10 +47,12 @@ use servo_util::geometry::Au;
use servo_util::time::{ProfilerChan, profile};
use servo_util::time;
use servo_util::task::spawn_named;
use servo_util::workqueue::WorkQueue;
use std::cast::transmute;
use std::cast;
use std::cell::RefCell;
use std::comm::Port;
use std::ptr;
use std::util;
use style::{AuthorOrigin, Stylesheet, Stylist};
@ -91,7 +94,7 @@ pub struct LayoutTask {
stylist: RWArc<Stylist>,
/// The workers that we use for parallel operation.
parallel_traversal: Option<ParallelPostorderFlowTraversal>,
parallel_traversal: Option<WorkQueue<*mut LayoutContext,UnsafeFlow>>,
/// The channel on which messages can be sent to the profiler.
profiler_chan: ProfilerChan,
@ -142,12 +145,14 @@ impl PreorderFlowTraversal for PropagateDamageTraversal {
/// The bubble-widths traversal, the first part of layout computation. This computes preferred
/// and intrinsic widths and bubbles them up the tree.
pub struct BubbleWidthsTraversal<'a>(&'a mut LayoutContext);
pub struct BubbleWidthsTraversal<'a> {
layout_context: &'a mut LayoutContext,
}
impl<'a> PostorderFlowTraversal for BubbleWidthsTraversal<'a> {
#[inline]
fn process(&mut self, flow: &mut Flow) -> bool {
flow.bubble_widths(**self);
flow.bubble_widths(self.layout_context);
true
}
@ -174,13 +179,15 @@ impl<'a> PreorderFlowTraversal for AssignWidthsTraversal<'a> {
/// The assign-heights-and-store-overflow traversal, the last (and most expensive) part of layout
/// computation. Determines the final heights for all layout objects, computes positions, and
/// computes overflow regions. In Gecko this corresponds to `FinishAndStoreOverflow`.
pub struct AssignHeightsAndStoreOverflowTraversal<'a>(&'a mut LayoutContext);
pub struct AssignHeightsAndStoreOverflowTraversal<'a> {
layout_context: &'a mut LayoutContext,
}
impl<'a> PostorderFlowTraversal for AssignHeightsAndStoreOverflowTraversal<'a> {
#[inline]
fn process(&mut self, flow: &mut Flow) -> bool {
flow.assign_height(**self);
flow.store_overflow(**self);
flow.assign_height(self.layout_context);
flow.store_overflow(self.layout_context);
true
}
@ -248,13 +255,8 @@ impl LayoutTask {
-> LayoutTask {
let local_image_cache = MutexArc::new(LocalImageCache(image_cache_task.clone()));
let screen_size = Size2D(Au(0), Au(0));
let font_context_info = FontContextInfo {
backend: opts.render_backend,
needs_font_list: true,
profiler_chan: profiler_chan.clone(),
};
let parallel_traversal = if opts.layout_threads != 1 {
Some(ParallelPostorderFlowTraversal::new(font_context_info, opts.layout_threads))
Some(WorkQueue::new(opts.layout_threads, ptr::mut_null()))
} else {
None
};
@ -288,20 +290,18 @@ impl LayoutTask {
// Create a layout context for use in building display lists, hit testing, &c.
fn build_layout_context(&self) -> LayoutContext {
let font_ctx = ~FontContext::new(FontContextInfo {
let font_context_info = FontContextInfo {
backend: self.opts.render_backend,
needs_font_list: true,
profiler_chan: self.profiler_chan.clone(),
});
};
LayoutContext {
shared: SharedLayoutInfo {
image_cache: self.local_image_cache.clone(),
screen_size: self.screen_size.clone(),
constellation_chan: self.constellation_chan.clone(),
leaf_set: self.leaf_set.clone(),
},
font_ctx: font_ctx,
image_cache: self.local_image_cache.clone(),
screen_size: self.screen_size.clone(),
constellation_chan: self.constellation_chan.clone(),
leaf_set: self.leaf_set.clone(),
font_context_info: font_context_info,
}
}
@ -418,7 +418,12 @@ impl LayoutTask {
fn solve_constraints(&mut self,
layout_root: &mut Flow,
layout_context: &mut LayoutContext) {
layout_root.traverse_postorder(&mut BubbleWidthsTraversal(layout_context));
{
let mut traversal = BubbleWidthsTraversal {
layout_context: layout_context,
};
layout_root.traverse_postorder(&mut traversal);
}
// FIXME(kmc): We want to prune nodes without the Reflow restyle damage
// bit, but FloatContext values can't be reused, so we need to
@ -428,8 +433,12 @@ impl LayoutTask {
layout_root.traverse_preorder(&mut AssignWidthsTraversal(layout_context));
// FIXME(pcwalton): Prune this pass as well.
layout_root.traverse_postorder(&mut AssignHeightsAndStoreOverflowTraversal(
layout_context));
{
let mut traversal = AssignHeightsAndStoreOverflowTraversal {
layout_context: layout_context,
};
layout_root.traverse_postorder(&mut traversal);
}
}
/// Performs layout constraint solving in parallel.
@ -443,9 +452,11 @@ impl LayoutTask {
match self.parallel_traversal {
None => fail!("solve_contraints_parallel() called with no parallel traversal ready"),
Some(ref mut traversal) => {
traversal.start(BubbleWidthsTraversalKind,
layout_context,
self.profiler_chan.clone());
parallel::traverse_flow_tree(BubbleWidthsTraversalKind,
&self.leaf_set,
self.profiler_chan.clone(),
layout_context,
traversal);
// NOTE: this currently computes borders, so any pruning should separate that
// operation out.
@ -453,9 +464,11 @@ impl LayoutTask {
// because this is a top-down traversal, unlike the others.
layout_root.traverse_preorder(&mut AssignWidthsTraversal(layout_context));
traversal.start(AssignHeightsAndStoreOverflowTraversalKind,
layout_context,
self.profiler_chan.clone());
parallel::traverse_flow_tree(AssignHeightsAndStoreOverflowTraversalKind,
&self.leaf_set,
self.profiler_chan.clone(),
layout_context,
traversal);
}
}
}

View file

@ -4,44 +4,29 @@
//! Implements parallel traversals over the flow tree.
use layout::context::{LayoutContext, SharedLayoutInfo};
use layout::flow::{Flow, PostorderFlowTraversal};
use layout::context::LayoutContext;
use layout::flow::{Flow, LeafSet, PostorderFlowTraversal};
use layout::flow;
use layout::layout_task::{AssignHeightsAndStoreOverflowTraversal, BubbleWidthsTraversal};
use gfx::font_context::{FontContext, FontContextInfo};
use native;
use extra::arc::MutexArc;
use servo_util::time::{ProfilerChan, profile};
use servo_util::time;
use servo_util::workqueue::{WorkQueue, WorkUnit, WorkerProxy};
use std::cast;
use std::comm::SharedChan;
use std::libc::c_void;
use std::ptr;
use std::sync::atomics::{AtomicInt, Relaxed, SeqCst};
use std::sync::deque::{Abort, BufferPool, Data, Empty, Stealer, Worker};
use std::util;
enum WorkerMsg {
/// Tells the worker to start a traversal.
StartMsg(Worker<UnsafeFlow>, TraversalKind, SharedLayoutInfo),
/// Tells the worker to stop. It can be restarted again with a `StartMsg`.
StopMsg,
/// Tells the worker thread to terminate.
ExitMsg,
pub enum TraversalKind {
BubbleWidthsTraversalKind,
AssignHeightsAndStoreOverflowTraversalKind,
}
enum SupervisorMsg {
/// Sent once the last flow is processed.
FinishedMsg,
pub type UnsafeFlow = (uint, uint);
/// Returns the deque to the supervisor.
ReturnDequeMsg(uint, Worker<UnsafeFlow>),
fn null_unsafe_flow() -> UnsafeFlow {
(0, 0)
}
pub type UnsafeFlow = (*c_void, *c_void);
pub fn owned_flow_to_unsafe_flow(flow: *~Flow) -> UnsafeFlow {
unsafe {
cast::transmute_copy(&*flow)
@ -54,10 +39,6 @@ pub fn mut_owned_flow_to_unsafe_flow(flow: *mut ~Flow) -> UnsafeFlow {
}
}
fn null_unsafe_flow() -> UnsafeFlow {
(ptr::null(), ptr::null())
}
/// Information that we need stored in each flow.
pub struct FlowParallelInfo {
/// The number of children that still need work done.
@ -75,270 +56,97 @@ impl FlowParallelInfo {
}
}
/// Information that the supervisor thread keeps about the worker threads.
struct WorkerInfo {
/// The communication channel to the workers.
chan: Chan<WorkerMsg>,
/// The buffer pool for this deque.
pool: BufferPool<UnsafeFlow>,
/// The worker end of the deque, if we have it.
deque: Option<Worker<UnsafeFlow>>,
/// The thief end of the work-stealing deque.
thief: Stealer<UnsafeFlow>,
}
/// Information that each worker needs to do its job.
struct PostorderWorker {
/// The font context.
font_context: Option<~FontContext>,
/// Communications for the worker.
comm: PostorderWorkerComm,
}
/// Communication channels for postorder workers.
struct PostorderWorkerComm {
/// The index of this worker.
index: uint,
/// The communication port from the supervisor.
port: Port<WorkerMsg>,
/// The communication channel to the supervisor.
chan: SharedChan<SupervisorMsg>,
/// The thief end of the work-stealing deque for all other workers.
other_deques: ~[Stealer<UnsafeFlow>],
}
/// The type of traversal we're performing.
pub enum TraversalKind {
BubbleWidthsTraversalKind,
AssignHeightsAndStoreOverflowTraversalKind,
}
impl PostorderWorker {
/// Starts up the worker and listens for messages.
pub fn start(&mut self) {
/// A parallel bottom-up flow traversal.
trait ParallelPostorderFlowTraversal : PostorderFlowTraversal {
fn run_parallel(&mut self, mut unsafe_flow: UnsafeFlow) {
loop {
// Wait for a start message.
let (mut deque, kind, shared_layout_info) = match self.comm.port.recv() {
StopMsg => fail!("unexpected stop message"),
StartMsg(deque, kind, shared_layout_info) => (deque, kind, shared_layout_info),
ExitMsg => return,
};
unsafe {
// Get a real flow.
let flow: &mut ~Flow = cast::transmute(&unsafe_flow);
// Set up our traversal context.
let mut traversal = LayoutContext {
shared: shared_layout_info,
font_ctx: self.font_context.take_unwrap(),
};
// And we're off!
'outer: loop {
let unsafe_flow;
match deque.pop() {
Some(the_flow) => unsafe_flow = the_flow,
None => {
// Become a thief.
let mut i = 0;
loop {
if self.comm.other_deques.len() != 0 {
match self.comm.other_deques[i].steal() {
Empty => {
// Try the next one.
i += 1;
if i >= self.comm.other_deques.len() {
i = 0
}
}
Abort => {
// Continue.
}
Data(the_flow) => {
unsafe_flow = the_flow;
break
}
}
}
if i == 0 {
match self.comm.port.try_recv() {
Some(StopMsg) => break 'outer,
Some(ExitMsg) => return,
Some(_) => fail!("unexpected message!"),
None => {}
}
}
}
}
// Perform the appropriate traversal.
if self.should_process(*flow) {
self.process(*flow);
}
// OK, we've got some data. The rest of this is unsafe code.
unsafe {
// Get a real flow.
let flow: &mut ~Flow = cast::transmute(&unsafe_flow);
let base = flow::mut_base(*flow);
// Perform the appropriate traversal.
match kind {
BubbleWidthsTraversalKind => {
let mut traversal = BubbleWidthsTraversal(&mut traversal);
if traversal.should_process(*flow) {
traversal.process(*flow);
}
}
AssignHeightsAndStoreOverflowTraversalKind => {
let mut traversal =
AssignHeightsAndStoreOverflowTraversal(&mut traversal);
if traversal.should_process(*flow) {
traversal.process(*flow);
}
}
}
// Reset the count of children for the next layout traversal.
base.parallel.children_count.store(base.children.len() as int, Relaxed);
let base = flow::mut_base(*flow);
// Possibly enqueue the parent.
let unsafe_parent = base.parallel.parent;
if unsafe_parent == null_unsafe_flow() {
// We're done!
break
}
// Reset the count of children for the next layout traversal.
base.parallel.children_count.store(base.children.len() as int, Relaxed);
// Possibly enqueue the parent.
let unsafe_parent = base.parallel.parent;
if unsafe_parent == null_unsafe_flow() {
// We're done!
self.comm.chan.send(FinishedMsg);
} else {
// No, we're not at the root yet. Then are we the last sibling of our
// parent? If so, we can enqueue our parent; otherwise, we've gotta wait.
let parent: &mut ~Flow = cast::transmute(&unsafe_parent);
let parent_base = flow::mut_base(*parent);
if parent_base.parallel.children_count.fetch_sub(1, SeqCst) == 1 {
// We were the last child of our parent. Enqueue the parent.
deque.push(unsafe_parent)
}
}
// No, we're not at the root yet. Then are we the last sibling of our parent? If
// so, we can continue on with our parent; otherwise, we've gotta wait.
let parent: &mut ~Flow = cast::transmute(&unsafe_parent);
let parent_base = flow::mut_base(*parent);
if parent_base.parallel.children_count.fetch_sub(1, SeqCst) == 1 {
// We were the last child of our parent. Reflow our parent.
unsafe_flow = unsafe_parent
} else {
// Stop.
break
}
}
// Destroy the traversal and save the font context.
let LayoutContext {
font_ctx: font_context,
..
} = traversal;
self.font_context = Some(font_context);
// Give the deque back to the supervisor.
self.comm.chan.send(ReturnDequeMsg(self.comm.index, deque))
}
}
}
/// A parallel bottom-up traversal.
pub struct ParallelPostorderFlowTraversal {
/// Information about each of the workers.
workers: ~[WorkerInfo],
/// A port on which information can be received from the workers.
port: Port<SupervisorMsg>,
impl<'a> ParallelPostorderFlowTraversal for BubbleWidthsTraversal<'a> {}
impl<'a> ParallelPostorderFlowTraversal for AssignHeightsAndStoreOverflowTraversal<'a> {}
fn bubble_widths(unsafe_flow: UnsafeFlow, proxy: &mut WorkerProxy<*mut LayoutContext,UnsafeFlow>) {
let layout_context: &mut LayoutContext = unsafe {
cast::transmute(*proxy.user_data())
};
let mut bubble_widths_traversal = BubbleWidthsTraversal {
layout_context: layout_context,
};
bubble_widths_traversal.run_parallel(unsafe_flow)
}
impl ParallelPostorderFlowTraversal {
pub fn new(font_context_info: FontContextInfo, thread_count: uint)
-> ParallelPostorderFlowTraversal {
let (supervisor_port, supervisor_chan) = SharedChan::new();
let (mut infos, mut comms) = (~[], ~[]);
for i in range(0, thread_count) {
let (worker_port, worker_chan) = Chan::new();
let mut pool = BufferPool::new();
let (worker, thief) = pool.deque();
infos.push(WorkerInfo {
chan: worker_chan,
pool: pool,
deque: Some(worker),
thief: thief,
});
comms.push(PostorderWorkerComm {
index: i,
port: worker_port,
chan: supervisor_chan.clone(),
other_deques: ~[],
});
}
fn assign_heights_and_store_overflow(unsafe_flow: UnsafeFlow,
proxy: &mut WorkerProxy<*mut LayoutContext,UnsafeFlow>) {
let layout_context: &mut LayoutContext = unsafe {
cast::transmute(*proxy.user_data())
};
let mut assign_heights_traversal = AssignHeightsAndStoreOverflowTraversal {
layout_context: layout_context,
};
assign_heights_traversal.run_parallel(unsafe_flow)
}
for i in range(0, thread_count) {
for j in range(0, thread_count) {
if i != j {
comms[i].other_deques.push(infos[j].thief.clone())
}
}
assert!(comms[i].other_deques.len() == thread_count - 1)
}
for comm in comms.move_iter() {
let font_context_info = font_context_info.clone();
native::task::spawn(proc() {
let mut worker = PostorderWorker {
font_context: Some(~FontContext::new(font_context_info)),
comm: comm,
};
worker.start()
})
}
ParallelPostorderFlowTraversal {
workers: infos,
port: supervisor_port,
}
pub fn traverse_flow_tree(kind: TraversalKind,
leaf_set: &MutexArc<LeafSet>,
profiler_chan: ProfilerChan,
layout_context: &mut LayoutContext,
queue: &mut WorkQueue<*mut LayoutContext,UnsafeFlow>) {
unsafe {
queue.data = cast::transmute(layout_context)
}
/// TODO(pcwalton): This could be parallelized.
fn warmup(&mut self, layout_context: &mut LayoutContext) {
layout_context.shared.leaf_set.access(|leaf_set| {
let fun = match kind {
BubbleWidthsTraversalKind => bubble_widths,
AssignHeightsAndStoreOverflowTraversalKind => assign_heights_and_store_overflow,
};
profile(time::LayoutParallelWarmupCategory, profiler_chan, || {
leaf_set.access(|leaf_set| {
for &flow in leaf_set.iter() {
match self.workers[0].deque {
None => fail!("no deque!"),
Some(ref mut deque) => {
deque.push(flow);
}
}
queue.push(WorkUnit {
fun: fun,
data: flow,
})
}
});
}
})
});
/// Traverses the given flow tree in parallel.
pub fn start(&mut self,
kind: TraversalKind,
layout_context: &mut LayoutContext,
profiler_chan: ProfilerChan) {
profile(time::LayoutParallelWarmupCategory, profiler_chan, || self.warmup(layout_context));
for worker in self.workers.mut_iter() {
worker.chan.send(StartMsg(util::replace(&mut worker.deque, None).unwrap(),
kind,
layout_context.shared.clone()))
}
// Wait for them to finish.
let _ = self.port.recv();
// Tell everyone to stop.
for worker in self.workers.iter() {
worker.chan.send(StopMsg);
}
// Get our deques back.
//
// TODO(pcwalton): Might be able to get a little parallelism over multiple traversals by
// doing this lazily.
for _ in range(0, self.workers.len()) {
match self.port.recv() {
ReturnDequeMsg(returned_deque_index, returned_deque) => {
self.workers[returned_deque_index].deque = Some(returned_deque)
}
_ => fail!("unexpected message received during return queue phase"),
}
}
}
/// Shuts down all the worker threads.
pub fn shutdown(&mut self) {
for worker in self.workers.iter() {
worker.chan.send(ExitMsg)
}
}
queue.run()
}

View file

@ -5,10 +5,10 @@
//! Text layout.
use layout::box_::{Box, ScannedTextBox, ScannedTextBoxInfo, UnscannedTextBox};
use layout::context::LayoutContext;
use layout::flow::Flow;
use extra::arc::Arc;
use gfx::font_context::FontContext;
use gfx::text::text_run::TextRun;
use gfx::text::util::{CompressWhitespaceNewline, transform_text, CompressNone};
use servo_util::range::Range;
@ -27,7 +27,7 @@ impl TextRunScanner {
}
}
pub fn scan_for_runs(&mut self, ctx: &mut LayoutContext, flow: &mut Flow) {
pub fn scan_for_runs(&mut self, font_context: &mut FontContext, flow: &mut Flow) {
{
let inline = flow.as_immutable_inline();
debug!("TextRunScanner: scanning {:u} boxes for text runs...", inline.boxes.len());
@ -40,13 +40,16 @@ impl TextRunScanner {
if box_i > 0 && !can_coalesce_text_nodes(flow.as_immutable_inline().boxes,
box_i - 1,
box_i) {
last_whitespace = self.flush_clump_to_list(ctx, flow, last_whitespace, &mut out_boxes);
last_whitespace = self.flush_clump_to_list(font_context,
flow,
last_whitespace,
&mut out_boxes);
}
self.clump.extend_by(1);
}
// handle remaining clumps
if self.clump.length() > 0 {
self.flush_clump_to_list(ctx, flow, last_whitespace, &mut out_boxes);
self.flush_clump_to_list(font_context, flow, last_whitespace, &mut out_boxes);
}
debug!("TextRunScanner: swapping out boxes.");
@ -73,7 +76,7 @@ impl TextRunScanner {
/// FIXME(pcwalton): Stop cloning boxes. Instead we will need to consume the `in_box`es as we
/// iterate over them.
pub fn flush_clump_to_list(&mut self,
ctx: &mut LayoutContext,
font_context: &mut FontContext,
flow: &mut Flow,
last_whitespace: bool,
out_boxes: &mut ~[Box])
@ -130,7 +133,7 @@ impl TextRunScanner {
// TODO(#177): Text run creation must account for the renderability of text by
// font group fonts. This is probably achieved by creating the font group above
// and then letting `FontGroup` decide which `Font` to stick into the text run.
let fontgroup = ctx.font_ctx.get_resolved_font_for_style(&font_style);
let fontgroup = font_context.get_resolved_font_for_style(&font_style);
let run = ~fontgroup.borrow().with(|fg| fg.create_textrun(transformed_text.clone(), decoration));
debug!("TextRunScanner: pushing single text box in range: {} ({})",
@ -151,7 +154,7 @@ impl TextRunScanner {
// and then letting `FontGroup` decide which `Font` to stick into the text run.
let in_box = &in_boxes[self.clump.begin()];
let font_style = in_box.font_style();
let fontgroup = ctx.font_ctx.get_resolved_font_for_style(&font_style);
let fontgroup = font_context.get_resolved_font_for_style(&font_style);
let decoration = in_box.text_decoration();
// TODO(#115): Use the actual CSS `white-space` property of the relevant style.

View file

@ -6,7 +6,7 @@
#[comment = "The Servo Parallel Browser Project"];
#[license = "MPL"];
#[feature(globs, macro_rules, managed_boxes)];
#[feature(globs, macro_rules, managed_boxes, thread_local)];
extern mod alert;
extern mod azure;

View file

@ -37,10 +37,12 @@ impl ProfilerChan {
}
pub enum ProfilerMsg {
// Normal message used for reporting time
/// Normal message used for reporting time
TimeMsg(ProfilerCategory, f64),
// Message used to force print the profiling metrics
/// Message used to force print the profiling metrics
PrintMsg,
/// Tells the profiler to shut down.
ExitMsg,
}
#[deriving(Eq, Clone, TotalEq, TotalOrd)]
@ -136,7 +138,12 @@ impl Profiler {
None => {
// no-op to handle profiler messages when the profiler is inactive
spawn_named("Profiler", proc() {
while port.recv_opt().is_some() {}
loop {
match port.recv_opt() {
None | Some(ExitMsg) => break,
_ => {}
}
}
});
}
}
@ -156,13 +163,17 @@ impl Profiler {
loop {
let msg = self.port.recv_opt();
match msg {
Some (msg) => self.handle_msg(msg),
Some(msg) => {
if !self.handle_msg(msg) {
break
}
}
None => break
}
}
}
fn handle_msg(&mut self, msg: ProfilerMsg) {
fn handle_msg(&mut self, msg: ProfilerMsg) -> bool {
match msg {
TimeMsg(category, t) => self.buckets.find_mut(&category).unwrap().push(t),
PrintMsg => match self.last_msg {
@ -170,8 +181,10 @@ impl Profiler {
Some(TimeMsg(..)) => self.print_buckets(),
_ => ()
},
ExitMsg => return false,
};
self.last_msg = Some(msg);
true
}
fn print_buckets(&mut self) {

View file

@ -9,6 +9,7 @@
extern mod extra;
extern mod geom;
extern mod native;
pub mod cache;
pub mod geometry;
@ -19,4 +20,5 @@ pub mod vec;
pub mod debug;
pub mod io;
pub mod task;
pub mod workqueue;

View file

@ -0,0 +1,295 @@
/* This Source Code Form is subject to the terms of the Mozilla Public
* License, v. 2.0. If a copy of the MPL was not distributed with this
* file, You can obtain one at http://mozilla.org/MPL/2.0/. */
//! A work queue for scheduling units of work across threads in a fork-join fashion.
//!
//! Data associated with queues is simply a pair of unsigned integers. It is expected that a
//! higher-level API on top of this could allow safe fork-join parallelism.
use native;
use std::cast;
use std::rand::{Rng, XorShiftRng};
use std::rand;
use std::sync::atomics::{AtomicUint, SeqCst};
use std::sync::deque::{Abort, BufferPool, Data, Empty, Stealer, Worker};
use std::unstable::intrinsics;
/// A unit of work.
///
/// The type parameter `QUD` stands for "queue user data" and represents global custom data for the
/// entire work queue, and the type parameter `WUD` stands for "work user data" and represents
/// custom data specific to each unit of work.
pub struct WorkUnit<QUD,WUD> {
/// The function to execute.
fun: extern "Rust" fn(WUD, &mut WorkerProxy<QUD,WUD>),
/// Arbitrary data.
data: WUD,
}
/// Messages from the supervisor to the worker.
enum WorkerMsg<QUD,WUD> {
/// Tells the worker to start work.
StartMsg(Worker<WorkUnit<QUD,WUD>>, *mut AtomicUint, *QUD),
/// Tells the worker to stop. It can be restarted again with a `StartMsg`.
StopMsg,
/// Tells the worker thread to terminate.
ExitMsg,
}
/// Messages to the supervisor.
enum SupervisorMsg<QUD,WUD> {
FinishedMsg,
ReturnDequeMsg(uint, Worker<WorkUnit<QUD,WUD>>),
}
/// Information that the supervisor thread keeps about the worker threads.
struct WorkerInfo<QUD,WUD> {
/// The communication channel to the workers.
chan: Chan<WorkerMsg<QUD,WUD>>,
/// The buffer pool for this deque.
pool: BufferPool<WorkUnit<QUD,WUD>>,
/// The worker end of the deque, if we have it.
deque: Option<Worker<WorkUnit<QUD,WUD>>>,
/// The thief end of the work-stealing deque.
thief: Stealer<WorkUnit<QUD,WUD>>,
}
/// Information specific to each worker thread that the thread keeps.
struct WorkerThread<QUD,WUD> {
/// The index of this worker.
index: uint,
/// The communication port from the supervisor.
port: Port<WorkerMsg<QUD,WUD>>,
/// The communication channel on which messages are sent to the supervisor.
chan: SharedChan<SupervisorMsg<QUD,WUD>>,
/// The thief end of the work-stealing deque for all other workers.
other_deques: ~[Stealer<WorkUnit<QUD,WUD>>],
/// The random number generator for this worker.
rng: XorShiftRng,
}
static SPIN_COUNT: uint = 1000;
impl<QUD:Send,WUD:Send> WorkerThread<QUD,WUD> {
/// The main logic. This function starts up the worker and listens for
/// messages.
pub fn start(&mut self) {
loop {
// Wait for a start message.
let (mut deque, ref_count, queue_data) = match self.port.recv() {
StartMsg(deque, ref_count, queue_data) => (deque, ref_count, queue_data),
StopMsg => fail!("unexpected stop message"),
ExitMsg => return,
};
// We're off!
//
// FIXME(pcwalton): Can't use labeled break or continue cross-crate due to a Rust bug.
loop {
// FIXME(pcwalton): Nasty workaround for the lack of labeled break/continue
// cross-crate.
let mut work_unit = unsafe {
intrinsics::uninit()
};
match deque.pop() {
Some(work) => work_unit = work,
None => {
// Become a thief.
let mut i = 0;
let mut should_continue = true;
loop {
let victim = (self.rng.next_u32() as uint) % self.other_deques.len();
match self.other_deques[victim].steal() {
Empty | Abort => {
// Continue.
}
Data(work) => {
work_unit = work;
break
}
}
if i == SPIN_COUNT {
match self.port.try_recv() {
Some(StopMsg) => {
should_continue = false;
break
}
Some(ExitMsg) => return,
Some(_) => fail!("unexpected message"),
None => {}
}
i = 0
} else {
i += 1
}
}
if !should_continue {
break
}
}
}
// At this point, we have some work. Perform it.
let mut proxy = WorkerProxy {
worker: &mut deque,
ref_count: ref_count,
queue_data: queue_data,
};
(work_unit.fun)(work_unit.data, &mut proxy);
// The work is done. Now decrement the count of outstanding work items. If this was
// the last work unit in the queue, then send a message on the channel.
unsafe {
if (*ref_count).fetch_sub(1, SeqCst) == 1 {
self.chan.send(FinishedMsg)
}
}
}
// Give the deque back to the supervisor.
self.chan.send(ReturnDequeMsg(self.index, deque))
}
}
}
/// A handle to the work queue that individual work units have.
pub struct WorkerProxy<'a,QUD,WUD> {
priv worker: &'a mut Worker<WorkUnit<QUD,WUD>>,
priv ref_count: *mut AtomicUint,
priv queue_data: *QUD,
}
impl<'a,QUD,WUD:Send> WorkerProxy<'a,QUD,WUD> {
/// Enqueues a block into the work queue.
#[inline]
pub fn push(&mut self, work_unit: WorkUnit<QUD,WUD>) {
unsafe {
drop((*self.ref_count).fetch_add(1, SeqCst));
}
self.worker.push(work_unit);
}
/// Retrieves the queue user data.
#[inline]
pub fn user_data<'a>(&'a self) -> &'a QUD {
unsafe {
cast::transmute(self.queue_data)
}
}
}
/// A work queue on which units of work can be submitted.
pub struct WorkQueue<QUD,WUD> {
/// Information about each of the workers.
priv workers: ~[WorkerInfo<QUD,WUD>],
/// A port on which deques can be received from the workers.
priv port: Port<SupervisorMsg<QUD,WUD>>,
/// The amount of work that has been enqueued.
priv work_count: uint,
/// Arbitrary user data.
data: QUD,
}
impl<QUD:Send,WUD:Send> WorkQueue<QUD,WUD> {
/// Creates a new work queue and spawns all the threads associated with
/// it.
pub fn new(thread_count: uint, user_data: QUD) -> WorkQueue<QUD,WUD> {
// Set up data structures.
let (supervisor_port, supervisor_chan) = SharedChan::new();
let (mut infos, mut threads) = (~[], ~[]);
for i in range(0, thread_count) {
let (worker_port, worker_chan) = Chan::new();
let mut pool = BufferPool::new();
let (worker, thief) = pool.deque();
infos.push(WorkerInfo {
chan: worker_chan,
pool: pool,
deque: Some(worker),
thief: thief,
});
threads.push(WorkerThread {
index: i,
port: worker_port,
chan: supervisor_chan.clone(),
other_deques: ~[],
rng: rand::weak_rng(),
});
}
// Connect workers to one another.
for i in range(0, thread_count) {
for j in range(0, thread_count) {
if i != j {
threads[i].other_deques.push(infos[j].thief.clone())
}
}
assert!(threads[i].other_deques.len() == thread_count - 1)
}
// Spawn threads.
for thread in threads.move_iter() {
native::task::spawn(proc() {
let mut thread = thread;
thread.start()
})
}
WorkQueue {
workers: infos,
port: supervisor_port,
work_count: 0,
data: user_data,
}
}
/// Enqueues a block into the work queue.
#[inline]
pub fn push(&mut self, work_unit: WorkUnit<QUD,WUD>) {
match self.workers[0].deque {
None => {
fail!("tried to push a block but we don't have the deque?!")
}
Some(ref mut deque) => deque.push(work_unit),
}
self.work_count += 1
}
/// Synchronously runs all the enqueued tasks and waits for them to complete.
pub fn run(&mut self) {
// Tell the workers to start.
let mut work_count = AtomicUint::new(self.work_count);
for worker in self.workers.mut_iter() {
worker.chan.send(StartMsg(worker.deque.take_unwrap(), &mut work_count, &self.data))
}
// Wait for the work to finish.
drop(self.port.recv());
self.work_count = 0;
// Tell everyone to stop.
for worker in self.workers.iter() {
worker.chan.send(StopMsg)
}
// Get our deques back.
for _ in range(0, self.workers.len()) {
match self.port.recv() {
ReturnDequeMsg(index, deque) => self.workers[index].deque = Some(deque),
FinishedMsg => fail!("unexpected finished message!"),
}
}
}
pub fn shutdown(&mut self) {
for worker in self.workers.iter() {
worker.chan.send(ExitMsg)
}
}
}