Files
zed/crates/gpui/src/queue.rs
T
Anthony Eid 6d9ddb96b2 Bench app context phase 2 (#58202)
This PR builds out GPUI's benchmark harness so render benchmarks measure
realistic frame costs using GPUI-owned, runtime-gated instrumentation.
It adds measurements for frame draw time, dirty-to-draw latency,
invalidation coalescing, and frame-budget overruns, and runs benchmark
workloads with production-like concurrency.

## How it works

**Frame timings flow through the GPUI profiler.** `Window::draw` emits a
`FrameTiming { window_id, dirty_at, invalidations, draw_start, draw_end
}` event into a global ring buffer in `gpui::profiler`, mirroring the
existing task-timing channel. Collection is runtime-gated by
`profiler::set_frame_trace_enabled` (one relaxed atomic load when
disabled — no `Instant::now` calls in production). `BenchReport` is a
pure listener: it drains events through a cursor-based
`FrameTimingCollector` and builds histograms in the bench layer.
`Window` carries no bench-only cfg fields, and the same event channel
can later feed the miniprofiler UI or an in-app frame-time HUD.

**`BenchDispatcher`** is a multithreaded `PlatformDispatcher` for
benchmarks: background tasks run on a worker pool (same priority queue
as `LinuxDispatcher`, with task-profiler hooks), timers fire in real
time on a dedicated thread, and foreground tasks queue until the bench
thread drains them with a blocking `run_until_idle()`. Unlike
`TestDispatcher`, work executes in parallel in real time, so wall-clock
measurements reflect production concurrency. In-flight accounting is
panic-safe via drop guards.

**`gpui::bench_platform()`** returns a per-process `TestPlatform` backed
by the `BenchDispatcher`, cached in a thread-local so worker threads
persist across Criterion calibration passes. This replaces the earlier
approach of constructing a real platform per invocation, which had
process-global singleton issues, never ran foreground tasks (no run loop
pumped the main queue), and couldn't open windows on headless CI.

**Text shaping** uses `NoopTextSystem`: deterministic across
machines/font installations and CI-portable. Measured cost of this
trade: ~10% of editor draw time vs `MacTextSystem` (Noop still emits one
glyph per character at fixed advances, so downstream layout/paint
structure is preserved).

**GPU coverage (macOS only for now).** `PlatformHeadlessRenderer` gained
`render_scene`, which encodes and submits the scene to Metal against a
cached offscreen target without blocking on completion or reading pixels
back — matching production `present()` CPU cost (`render_scene_to_image`
would overstate it: it waits for the GPU and copies pixels back).
`TestWindow::draw` forwards scenes to the renderer, the real
`MetalAtlas` means glyph/SVG rasterization happens during paint, and
`bench_renderer` presents after each measured update. Platforms without
a headless renderer degrade to discarding the scene.

## Example

```rust
#[gpui::bench]
fn editor_render(cx: &mut BenchAppContext) {
    init_context(cx);

    let buffer = cx.update(|cx| { /* build a MultiBuffer */ });

    let mut window = cx.add_empty_window();
    let editor = window.update(|window, cx| {
        let editor = window.replace_root(cx, |window, cx| {
            let mut editor = Editor::new(EditorMode::full(), buffer, None, window, cx);
            editor.set_style(editor::EditorStyle::default(), window, cx);
            editor
        });
        window.focus(&editor.focus_handle(cx), cx);
        editor
    });

    let mut move_down = true;
    cx.bench_renderer(editor, move |editor, window, cx| {
        if move_down {
            editor.move_down(&MoveDown, window, cx);
        } else {
            editor.move_up(&MoveUp, window, cx);
        }
        move_down = !move_down;
    });
}
```

## Example output (release, M-series)

```
editor_render           time:   [329.75 µs 330.17 µs 330.69 µs]

GPUI bench report (all observed iterations): editor_render
  note: includes Criterion warmup/calibration
  window dirty-to-draw:
    samples: 31533
    mean: 0.321ms
    p50: 0.322ms
    p90: 0.336ms
    p95: 0.342ms
    p99: 0.360ms
    max: 0.504ms
    frame budget overruns total: 0
    frame budget overruns max: 0
  window draw:
    samples: 31533
    mean: 0.295ms
    p50: 0.295ms
    p90: 0.307ms
    p95: 0.313ms
    p99: 0.330ms
    max: 0.455ms
    frame budget overruns total: 0
    frame budget overruns max: 0
  invalidations per frame: mean 5.00, max 5
```

(`invalidations per frame: mean 5.00` is real signal: each `move_down`
notifies the window five times before the draw.)

## Known limitations

- **Draw-per-flush**: the harness draws synchronously when effects flush
rather than coalescing invalidations to a vsync tick, so `dirty-to-draw`
excludes queueing delay, and `frame budget overruns` is a draw-time
budget proxy rather than actual missed presents. A frame-paced mode is
natural follow-up work.
- **GPU submission is measured on macOS only**; other platforms have no
headless renderer yet.
- The GPUI report includes Criterion warmup/calibration samples (noted
in the output); Criterion's `time` is the regression-gating number.
- `run_until_idle` waits for queued, running, and already-due work, but
not for timers that haven't reached their due time — the dispatcher runs
in real time and can't skip ahead like `TestDispatcher`'s virtual clock.

## Future work

- A vsync-like frame-pacing mode (suppress draw-on-flush; tick-driven
draw + present) so dirty-to-draw captures queueing delay
- Record present duration in `FrameTiming` so the report can split draw
vs present
- Benches that scroll through novel content (cold layout caches) and an
agent-panel render bench
- Headless renderers for Windows/Linux
- Move benches into a dedicated crate

Self-Review Checklist:
- [x] I've reviewed my own diff for quality, security, and reliability
- [x] Unsafe blocks (if any) have justifying comments
- [x] The content is consistent with the UI/UX checklist
- [x] Tests cover the new/changed behavior
- [x] Performance impact has been considered and is acceptable

Release Notes:

- N/A
2026-06-10 09:26:50 +00:00

430 lines
12 KiB
Rust

use std::{
collections::VecDeque,
fmt,
iter::FusedIterator,
sync::{Arc, atomic::AtomicUsize},
};
use rand::{Rng, SeedableRng, rngs::SmallRng};
use crate::Priority;
struct PriorityQueues<T> {
high_priority: VecDeque<T>,
medium_priority: VecDeque<T>,
low_priority: VecDeque<T>,
}
impl<T> PriorityQueues<T> {
fn is_empty(&self) -> bool {
self.high_priority.is_empty()
&& self.medium_priority.is_empty()
&& self.low_priority.is_empty()
}
}
struct PriorityQueueState<T> {
queues: parking_lot::Mutex<PriorityQueues<T>>,
condvar: parking_lot::Condvar,
receiver_count: AtomicUsize,
sender_count: AtomicUsize,
}
impl<T> PriorityQueueState<T> {
fn send(&self, priority: Priority, item: T) -> Result<(), SendError<T>> {
if self
.receiver_count
.load(std::sync::atomic::Ordering::Relaxed)
== 0
{
return Err(SendError(item));
}
let mut queues = self.queues.lock();
Self::push(&mut queues, priority, item);
self.condvar.notify_one();
Ok(())
}
fn spin_send(&self, priority: Priority, item: T) -> Result<(), SendError<T>> {
if self
.receiver_count
.load(std::sync::atomic::Ordering::Relaxed)
== 0
{
return Err(SendError(item));
}
let mut queues = loop {
if let Some(guard) = self.queues.try_lock() {
break guard;
}
std::hint::spin_loop();
};
Self::push(&mut queues, priority, item);
self.condvar.notify_one();
Ok(())
}
fn push(queues: &mut PriorityQueues<T>, priority: Priority, item: T) {
match priority {
Priority::RealtimeAudio => unreachable!(
"Realtime audio priority runs on a dedicated thread and is never queued"
),
Priority::High => queues.high_priority.push_back(item),
Priority::Medium => queues.medium_priority.push_back(item),
Priority::Low => queues.low_priority.push_back(item),
};
}
fn recv<'a>(&'a self) -> Result<parking_lot::MutexGuard<'a, PriorityQueues<T>>, RecvError> {
let mut queues = self.queues.lock();
let sender_count = self.sender_count.load(std::sync::atomic::Ordering::Relaxed);
if queues.is_empty() && sender_count == 0 {
return Err(crate::queue::RecvError);
}
while queues.is_empty() {
self.condvar.wait(&mut queues);
}
Ok(queues)
}
fn try_recv<'a>(
&'a self,
) -> Result<Option<parking_lot::MutexGuard<'a, PriorityQueues<T>>>, RecvError> {
let mut queues = self.queues.lock();
let sender_count = self.sender_count.load(std::sync::atomic::Ordering::Relaxed);
if queues.is_empty() && sender_count == 0 {
return Err(crate::queue::RecvError);
}
if queues.is_empty() {
Ok(None)
} else {
Ok(Some(queues))
}
}
fn spin_try_recv<'a>(
&'a self,
) -> Result<Option<parking_lot::MutexGuard<'a, PriorityQueues<T>>>, RecvError> {
let queues = loop {
if let Some(guard) = self.queues.try_lock() {
break guard;
}
std::hint::spin_loop();
};
let sender_count = self.sender_count.load(std::sync::atomic::Ordering::Relaxed);
if queues.is_empty() && sender_count == 0 {
return Err(crate::queue::RecvError);
}
if queues.is_empty() {
Ok(None)
} else {
Ok(Some(queues))
}
}
}
#[doc(hidden)]
pub struct PriorityQueueSender<T> {
state: Arc<PriorityQueueState<T>>,
}
impl<T> PriorityQueueSender<T> {
fn new(state: Arc<PriorityQueueState<T>>) -> Self {
Self { state }
}
pub fn send(&self, priority: Priority, item: T) -> Result<(), SendError<T>> {
self.state.send(priority, item)?;
Ok(())
}
pub fn spin_send(&self, priority: Priority, item: T) -> Result<(), SendError<T>> {
self.state.spin_send(priority, item)?;
Ok(())
}
}
impl<T> Drop for PriorityQueueSender<T> {
fn drop(&mut self) {
self.state
.sender_count
.fetch_sub(1, std::sync::atomic::Ordering::AcqRel);
}
}
#[doc(hidden)]
pub struct PriorityQueueReceiver<T> {
state: Arc<PriorityQueueState<T>>,
rand: SmallRng,
disconnected: bool,
}
impl<T> Clone for PriorityQueueReceiver<T> {
fn clone(&self) -> Self {
self.state
.receiver_count
.fetch_add(1, std::sync::atomic::Ordering::AcqRel);
Self {
state: Arc::clone(&self.state),
rand: SmallRng::seed_from_u64(0),
disconnected: self.disconnected,
}
}
}
#[doc(hidden)]
pub struct SendError<T>(pub T);
impl<T: fmt::Debug> fmt::Debug for SendError<T> {
fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
f.debug_tuple("SendError").field(&self.0).finish()
}
}
#[derive(Debug)]
#[doc(hidden)]
pub struct RecvError;
#[allow(dead_code)]
impl<T> PriorityQueueReceiver<T> {
pub fn new() -> (PriorityQueueSender<T>, Self) {
let state = PriorityQueueState {
queues: parking_lot::Mutex::new(PriorityQueues {
high_priority: VecDeque::new(),
medium_priority: VecDeque::new(),
low_priority: VecDeque::new(),
}),
condvar: parking_lot::Condvar::new(),
receiver_count: AtomicUsize::new(1),
sender_count: AtomicUsize::new(1),
};
let state = Arc::new(state);
let sender = PriorityQueueSender::new(Arc::clone(&state));
let receiver = PriorityQueueReceiver {
state,
rand: SmallRng::seed_from_u64(0),
disconnected: false,
};
(sender, receiver)
}
/// Returns whether the queue currently contains no elements.
pub fn is_empty(&self) -> bool {
self.state.queues.lock().is_empty()
}
/// Tries to pop one element from the priority queue without blocking.
///
/// This will early return if there are no elements in the queue.
///
/// This method is best suited if you only intend to pop one element, for better performance
/// on large queues see [`Self::try_iter`]
///
/// # Errors
///
/// If the sender was dropped
pub fn try_pop(&mut self) -> Result<Option<T>, RecvError> {
self.pop_inner(false)
}
pub fn spin_try_pop(&mut self) -> Result<Option<T>, RecvError> {
use Priority as P;
let Some(mut queues) = self.state.spin_try_recv()? else {
return Ok(None);
};
let high = P::High.weight() * !queues.high_priority.is_empty() as u32;
let medium = P::Medium.weight() * !queues.medium_priority.is_empty() as u32;
let low = P::Low.weight() * !queues.low_priority.is_empty() as u32;
let mut mass = high + medium + low;
if !queues.high_priority.is_empty() {
let flip = self.rand.random_ratio(P::High.weight(), mass);
if flip {
return Ok(queues.high_priority.pop_front());
}
mass -= P::High.weight();
}
if !queues.medium_priority.is_empty() {
let flip = self.rand.random_ratio(P::Medium.weight(), mass);
if flip {
return Ok(queues.medium_priority.pop_front());
}
mass -= P::Medium.weight();
}
if !queues.low_priority.is_empty() {
let flip = self.rand.random_ratio(P::Low.weight(), mass);
if flip {
return Ok(queues.low_priority.pop_front());
}
}
Ok(None)
}
/// Pops an element from the priority queue blocking if necessary.
///
/// This method is best suited if you only intend to pop one element, for better performance
/// on large queues see [`Self::iter``]
///
/// # Errors
///
/// If the sender was dropped
pub fn pop(&mut self) -> Result<T, RecvError> {
self.pop_inner(true).map(|e| e.unwrap())
}
/// Returns an iterator over the elements of the queue
/// this iterator will end when all elements have been consumed and will not wait for new ones.
pub fn try_iter(self) -> TryIter<T> {
TryIter {
receiver: self,
ended: false,
}
}
/// Returns an iterator over the elements of the queue
/// this iterator will wait for new elements if the queue is empty.
pub fn iter(self) -> Iter<T> {
Iter(self)
}
#[inline(always)]
// algorithm is the loaded die from biased coin from
// https://www.keithschwarz.com/darts-dice-coins/
fn pop_inner(&mut self, block: bool) -> Result<Option<T>, RecvError> {
use Priority as P;
let mut queues = if !block {
let Some(queues) = self.state.try_recv()? else {
return Ok(None);
};
queues
} else {
self.state.recv()?
};
let high = P::High.weight() * !queues.high_priority.is_empty() as u32;
let medium = P::Medium.weight() * !queues.medium_priority.is_empty() as u32;
let low = P::Low.weight() * !queues.low_priority.is_empty() as u32;
let mut mass = high + medium + low; //%
if !queues.high_priority.is_empty() {
let flip = self.rand.random_ratio(P::High.weight(), mass);
if flip {
return Ok(queues.high_priority.pop_front());
}
mass -= P::High.weight();
}
if !queues.medium_priority.is_empty() {
let flip = self.rand.random_ratio(P::Medium.weight(), mass);
if flip {
return Ok(queues.medium_priority.pop_front());
}
mass -= P::Medium.weight();
}
if !queues.low_priority.is_empty() {
let flip = self.rand.random_ratio(P::Low.weight(), mass);
if flip {
return Ok(queues.low_priority.pop_front());
}
}
Ok(None)
}
}
impl<T> Drop for PriorityQueueReceiver<T> {
fn drop(&mut self) {
self.state
.receiver_count
.fetch_sub(1, std::sync::atomic::Ordering::AcqRel);
}
}
#[doc(hidden)]
pub struct Iter<T>(PriorityQueueReceiver<T>);
impl<T> Iterator for Iter<T> {
type Item = T;
fn next(&mut self) -> Option<Self::Item> {
self.0.pop().ok()
}
}
impl<T> FusedIterator for Iter<T> {}
#[doc(hidden)]
pub struct TryIter<T> {
receiver: PriorityQueueReceiver<T>,
ended: bool,
}
impl<T> Iterator for TryIter<T> {
type Item = Result<T, RecvError>;
fn next(&mut self) -> Option<Self::Item> {
if self.ended {
return None;
}
let res = self.receiver.try_pop();
self.ended = res.is_err();
res.transpose()
}
}
impl<T> FusedIterator for TryIter<T> {}
#[cfg(test)]
mod tests {
use collections::HashSet;
use super::*;
#[test]
fn all_tasks_get_yielded() {
let (tx, mut rx) = PriorityQueueReceiver::new();
tx.send(Priority::Medium, 20).unwrap();
tx.send(Priority::High, 30).unwrap();
tx.send(Priority::Low, 10).unwrap();
tx.send(Priority::Medium, 21).unwrap();
tx.send(Priority::High, 31).unwrap();
drop(tx);
assert_eq!(
rx.iter().collect::<HashSet<_>>(),
[30, 31, 20, 21, 10].into_iter().collect::<HashSet<_>>()
)
}
#[test]
fn new_high_prio_task_get_scheduled_quickly() {
let (tx, mut rx) = PriorityQueueReceiver::new();
for _ in 0..100 {
tx.send(Priority::Low, 1).unwrap();
}
assert_eq!(rx.pop().unwrap(), 1);
tx.send(Priority::High, 3).unwrap();
assert_eq!(rx.pop().unwrap(), 3);
assert_eq!(rx.pop().unwrap(), 1);
}
}