Skip to content

Open source CapCut alternative.

TypeScript

test(site): the perf eval holds the picture to its sound on a low-end machine

David Le3443bc92ff5d
diff --git a/evals/cut-perf.latest-report.json b/evals/cut-perf.latest-report.json
index ce920b02..6be31b60 100644
--- a/evals/cut-perf.latest-report.json
+++ b/evals/cut-perf.latest-report.json
@@ -5,6 +5,7 @@
"cpu": 1,
"slots": 0,
"softwareMs": 0,
+ "outputLatencyS": 0,
"lagP95S": 0.05,
"stallShare": 0.002,
"lateShare": 0.02,
@@ -19,7 +20,6 @@
"longTaskMs": 16,
"filmstripSlackS": 0.2,
"lagP95S": 0.05,
- "lagMaxS": 0.25,
"decayRatio": 1.5,
"warmMb": 400,
"stallShare": 0.002,
@@ -37,10 +37,10 @@
"pass": true,
"notes": [],
"scrub": {
- "p50": 7.4,
- "p95": 15.6,
- "mean": 8.45,
- "max": 24.1,
+ "p50": 6.9,
+ "p95": 15.4,
+ "mean": 8.06,
+ "max": 38.6,
"n": 41
},
"longTasks": {
@@ -59,10 +59,10 @@
"pass": true,
"notes": [],
"scrub": {
- "p50": 7.4,
- "p95": 15.7,
- "mean": 8.2,
- "max": 16.4,
+ "p50": 7,
+ "p95": 16.1,
+ "mean": 8.22,
+ "max": 18.1,
"n": 41
},
"longTasks": {
@@ -72,7 +72,7 @@
"max": 0,
"n": 0
},
- "liveSamples": 77,
+ "liveSamples": 72,
"machine": "desktop"
},
{
@@ -81,10 +81,10 @@
"pass": true,
"notes": [],
"scrub": {
- "p50": 7.4,
- "p95": 15.7,
- "mean": 8.34,
- "max": 17.9,
+ "p50": 7,
+ "p95": 14.6,
+ "mean": 7.96,
+ "max": 15.7,
"n": 41
},
"longTasks": {
@@ -94,7 +94,7 @@
"max": 0,
"n": 0
},
- "liveSamples": 64,
+ "liveSamples": 46,
"machine": "desktop"
},
{
@@ -103,10 +103,10 @@
"pass": true,
"notes": [],
"scrub": {
- "p50": 15.5,
- "p95": 18.2,
- "mean": 14.96,
- "max": 23.7,
+ "p50": 15,
+ "p95": 22.9,
+ "mean": 14.73,
+ "max": 30.3,
"n": 41
},
"longTasks": {
@@ -116,7 +116,7 @@
"max": 0,
"n": 0
},
- "liveSamples": 71,
+ "liveSamples": 88,
"machine": "desktop"
},
{
@@ -143,7 +143,7 @@
"pass": true,
"notes": [],
"drops": 0,
- "presented": 2074,
+ "presented": 2073,
"boundaryDrops": 0,
"longTasks": {
"p50": 0,
@@ -161,19 +161,23 @@
"pass": true,
"notes": [],
"drops": 0,
- "presented": 1617,
+ "presented": 1621,
"lag": {
"p50": 0.02,
"p95": 0.03,
"mean": 0.02,
"max": 0.03,
- "n": 1617
+ "n": 1621
},
"stallShare": 0,
"decay": {
"first": 0,
"last": 0
},
+ "clock": {
+ "lead": 0.22,
+ "reported": 0.22
+ },
"longTasks": {
"p50": 0,
"p95": 0,
@@ -193,19 +197,59 @@
"pass": true,
"notes": [],
"drops": 0,
- "presented": 6481,
+ "presented": 6478,
+ "lag": {
+ "p50": 0.02,
+ "p95": 0.03,
+ "mean": 0.02,
+ "max": 0.03,
+ "n": 6478
+ },
+ "stallShare": 0,
+ "decay": {
+ "first": 0,
+ "last": 0
+ },
+ "clock": {
+ "lead": 0.22,
+ "reported": 0.22
+ },
+ "longTasks": {
+ "p50": 0,
+ "p95": 0,
+ "mean": 0,
+ "max": 0,
+ "n": 0
+ },
+ "liveSamples": 120,
+ "liveSources": 12,
+ "keptSources": 20,
+ "warmMb": 354,
+ "machine": "desktop"
+ },
+ {
+ "name": "play-fast-cuts-graded",
+ "bucket": "playback",
+ "pass": true,
+ "notes": [],
+ "drops": 0,
+ "presented": 3236,
"lag": {
"p50": 0.02,
"p95": 0.03,
"mean": 0.02,
"max": 0.03,
- "n": 6481
+ "n": 3236
},
"stallShare": 0,
"decay": {
"first": 0,
"last": 0
},
+ "clock": {
+ "lead": 0.22,
+ "reported": 0.22
+ },
"longTasks": {
"p50": 0,
"p95": 0,
diff --git a/site/scripts/eval-cut-perf.ts b/site/scripts/eval-cut-perf.ts
index 51db812e..bfc9f498 100644
--- a/site/scripts/eval-cut-perf.ts
+++ b/site/scripts/eval-cut-perf.ts
@@ -110,27 +110,37 @@ interface Machine {
* frame the decoder failed to deliver on time is a frame the viewer sees
* repeated or missing. */
lateShare: number;
+ /**
+ * What this machine's audio output holds before a rendered sample is heard.
+ *
+ * The developer machine is the quiet one here too: CoreAudio reports a few
+ * tens of milliseconds and a picture drawn against the wrong audio clock is
+ * off by less than a frame, which nobody sees and no run catches. Shared-mode
+ * WASAPI — every Windows browser, by default — reports several times that,
+ * and a Bluetooth output more again. Zero leaves the device alone.
+ */
+ outputLatencyS: number;
}
const MACHINES: Machine[] = [
// The machine this is being written on: nothing binds, and the numbers are
// held to the tightest budget in the file.
- { name: "desktop", cpu: 1, slots: 0, softwareMs: 0, lagP95S: 0.05, stallShare: 0.002, lateShare: 0.02, decayFloor: 0.02 },
+ { name: "desktop", cpu: 1, slots: 0, softwareMs: 0, outputLatencyS: 0, lagP95S: 0.05, stallShare: 0.002, lateShare: 0.02, decayFloor: 0.02 },
// A mainstream laptop with an integrated GPU.
- { name: "laptop", cpu: 4, slots: 6, softwareMs: 6, lagP95S: 0.06, stallShare: 0.04, lateShare: 0.1, decayFloor: 0.1 },
+ { name: "laptop", cpu: 4, slots: 6, softwareMs: 6, outputLatencyS: 0.12, lagP95S: 0.06, stallShare: 0.04, lateShare: 0.1, decayFloor: 0.1 },
// The one people report from: a 15W six-core with Vega graphics, running a
// browser that is also drawing the editor.
- { name: "ryzen-5500u", cpu: 10, slots: 4, softwareMs: 8, lagP95S: 0.06, stallShare: 0.03, lateShare: 0.05, decayFloor: 0.08 },
+ { name: "ryzen-5500u", cpu: 10, slots: 4, softwareMs: 8, outputLatencyS: 0.12, lagP95S: 0.06, stallShare: 0.03, lateShare: 0.05, decayFloor: 0.08 },
// Worse than anything anyone has reported, which is the point: what holds
// here holds on the machines that have not written in yet.
- { name: "potato", cpu: 20, slots: 2, softwareMs: 12, lagP95S: 0.1, stallShare: 0.06, lateShare: 0.2, decayFloor: 0.15 },
+ { name: "potato", cpu: 20, slots: 2, softwareMs: 12, outputLatencyS: 0.2, lagP95S: 0.1, stallShare: 0.06, lateShare: 0.2, decayFloor: 0.15 },
// No hardware video decode at all. This is not an exotic machine — a
// blocklisted driver, a codec the GPU does not carry, a browser started with
// acceleration off, and every stream is on the CPU. A software decoder is
// slower than a hardware one; it is not broken, and neither is an editor
// running on top of one. This profile is held to the same standard a laptop
// is, because that is the claim.
- { name: "software-decode", cpu: 4, slots: 0, softwareMs: 8, lagP95S: 0.06, stallShare: 0.04, lateShare: 0.1, decayFloor: 0.1 },
+ { name: "software-decode", cpu: 4, slots: 0, softwareMs: 8, outputLatencyS: 0.12, lagP95S: 0.06, stallShare: 0.04, lateShare: 0.1, decayFloor: 0.1 },
];
const MACHINE_NAME = arg("--machine") ?? "desktop";
@@ -208,6 +218,13 @@ const GATE = {
/** How far behind the sound the picture has to be for that frame to read as a
* hitch rather than a beat of lag. */
const STALL_S = 0.25;
+/** How far the preview's hold-back may sit from what the output device reports
+ * before the picture and the sound are telling different stories. One frame at
+ * sixty hertz. */
+const CLOCK_SLACK_S = 0.02;
+/** Where the preview stops believing a device's claim, mirrored here so a
+ * profile past it does not fail a run the preview is handling correctly. */
+const CLOCK_CAP_S = 0.4;
// ── Types shared with perfTrace.ts ──────────────────────────────────────────
interface PresentRecord {
@@ -233,6 +250,8 @@ interface Trace {
liveSources: number[];
keptSources: number[];
warmMb: number[];
+ audioLead: number[];
+ audioLatency: number[];
startedAt: number;
}
@@ -666,6 +685,11 @@ interface CaseResult {
stallShare?: number;
/** Late-frame share over the first and last third of a long play. */
decay?: { first: number; last: number };
+ /** Seconds the preview held the picture back for the output device, against
+ * the seconds the device reported holding. These agree when the picture is
+ * drawn against the sound that is audible rather than the sound that has
+ * been rendered. */
+ clock?: { lead: number; reported: number };
/** Seconds between a strip tile's own moment and the moment its picture
* was captured at. */
tileErr?: Agg;
@@ -976,6 +1000,20 @@ const montageCase = (name: string, passes: number, dressed = false): EvalCase =>
}
if (decoders && decoders.peak === 0) notes.push("nothing decoded at all");
if (lag.p95 > GATE.lagP95S) notes.push(`picture ${lag.p95}s behind the clock at p95`);
+ // Which clock the picture is drawn against, which the lag above cannot ask:
+ // a picture drawn against the rendered clock is exactly on time against it
+ // and a tenth of a second early against the speakers. The device's own
+ // answer is the reference, capped where the preview caps it so a driver
+ // claiming something absurd cannot fail a run on its own.
+ const lead = agg(trace.audioLead ?? []);
+ const reported = agg(trace.audioLatency ?? []);
+ const owed = Math.min(CLOCK_CAP_S, reported.p50);
+ if (reported.n > 0 && Math.abs(lead.p50 - owed) > CLOCK_SLACK_S) {
+ notes.push(
+ `picture drawn ${Math.round((owed - lead.p50) * 1000)}ms ahead of the sound: ` +
+ `the output holds ${Math.round(owed * 1000)}ms and the clock gives back ${Math.round(lead.p50 * 1000)}ms`
+ );
+ }
const behind = played.filter((p) => p.srcTs !== null && p.wantSrc! - p.srcTs! > STALL_S);
const stallShare = +(behind.length / played.length).toFixed(4);
if (stallShare > GATE.stallShare) {
@@ -999,6 +1037,7 @@ const montageCase = (name: string, passes: number, dressed = false): EvalCase =>
lag,
stallShare,
decay,
+ clock: { lead: lead.p50, reported: +reported.p50.toFixed(3) },
longTasks: agg(trace.longTasks.map((l) => l.ms)),
liveSamples: Math.max(0, ...trace.liveSamples),
liveSources: Math.max(0, ...trace.liveSources),
@@ -1464,6 +1503,37 @@ const decoderShim = ({ slots, softwareMs }: Machine) => `
})();
`;
+/**
+ * Make the page's audio output hold as long as this machine's does.
+ *
+ * `outputLatency` is the device's answer, and on the machine this is written on
+ * the answer is small enough that a preview ignoring it is wrong by less than a
+ * frame. Reporting a Windows-sized buffer instead is what makes the gate below
+ * mean something: a picture drawn against the rendered clock rather than the
+ * audible one now sits a tenth of a second ahead of its own sound, which is
+ * what the reports describe.
+ *
+ * Only the reading changes. The samples still leave the machine when they
+ * leave it; what is under test is whether the preview reads the device at all.
+ */
+const audioShim = (seconds: number) => `
+(() => {
+ const patch = (Klass) => {
+ if (!Klass) return;
+ const d = Object.getOwnPropertyDescriptor(Klass.prototype, "outputLatency");
+ Object.defineProperty(Klass.prototype, "outputLatency", {
+ configurable: true,
+ get() {
+ const own = d && d.get ? d.get.call(this) : 0;
+ return (own || 0) + ${seconds};
+ },
+ });
+ };
+ patch(window.AudioContext);
+ patch(window.webkitAudioContext);
+})();
+`;
+
async function launch(): Promise<{ browser: Browser; page: Page }> {
const browser = await chromium.launch({
channel: "chrome",
@@ -1477,6 +1547,8 @@ async function launch(): Promise<{ browser: Browser; page: Page }> {
viewport: { width: 1600, height: 1000 },
});
if (MACHINE.softwareMs > 0) await context.addInitScript({ content: decoderShim(MACHINE) });
+ if (MACHINE.outputLatencyS > 0)
+ await context.addInitScript({ content: audioShim(MACHINE.outputLatencyS) });
// The editor's session gate is a client-side cookie read, and the only sign-in
// this build offers is Google's. Answering that one request for the dev-bypass
// user is the whole of the fake: every other request the page makes is real,
@@ -1703,7 +1775,7 @@ const detailOf = (r: CaseResult) =>
: r.idleTicks !== undefined
? `ticks=${r.idleTicks}`
: r.lag
- ? `late=${r.drops}/${r.presented} lagP95=${r.lag.p95}s hitch=${((r.stallShare ?? 0) * 100).toFixed(1)}% lagMax=${r.lag.max}s sources=${r.liveSources} sw=${r.decoders?.softwarePeak ?? "-"}/${r.decoders?.software ?? "-"} busy=${r.decoders?.busyPeak ?? "-"} warm=${r.warmMb}MB decay=${r.decay?.first}→${r.decay?.last}`
+ ? `late=${r.drops}/${r.presented} lagP95=${r.lag.p95}s hitch=${((r.stallShare ?? 0) * 100).toFixed(1)}% lagMax=${r.lag.max}s sources=${r.liveSources} sw=${r.decoders?.softwarePeak ?? "-"}/${r.decoders?.software ?? "-"} busy=${r.decoders?.busyPeak ?? "-"} clock=${Math.round((r.clock?.lead ?? 0) * 1000)}/${Math.round((r.clock?.reported ?? 0) * 1000)}ms warm=${r.warmMb}MB decay=${r.decay?.first}→${r.decay?.last}`
: `late=${r.drops}/${r.presented} atCut=${r.boundaryDrops}`;
async function main(): Promise<void> {