mirror of
https://github.com/arcodange-org/mediabunny.git
synced 2026-09-27 02:43:48 +02:00
Fix NodeAv memory leak (fixes #392)
This commit is contained in:
@@ -12,10 +12,28 @@ import { CODEC_TO_CODEC_ID, getChannelLayout } from './misc';
|
||||
import { assert, toUint8Array } from '../../../src/misc';
|
||||
import { AvFrameAudioSampleResource } from './audio-sample';
|
||||
|
||||
type NodeAvState = {
|
||||
frame: NodeAv.Frame;
|
||||
packet: NodeAv.Packet;
|
||||
codecContext: NodeAv.CodecContext | null;
|
||||
};
|
||||
|
||||
const freeState = (state: NodeAvState) => {
|
||||
state.codecContext?.freeContext();
|
||||
state.frame.free();
|
||||
state.packet.free();
|
||||
};
|
||||
|
||||
// Needed for proper freeing if close isn't called
|
||||
let finalizationRegistry: FinalizationRegistry<NodeAvState> | null = null;
|
||||
if (typeof FinalizationRegistry !== 'undefined') {
|
||||
finalizationRegistry = new FinalizationRegistry<NodeAvState>((state) => {
|
||||
freeState(state);
|
||||
});
|
||||
}
|
||||
|
||||
export class NodeAvAudioDecoder extends CustomAudioDecoder {
|
||||
frame!: NodeAv.Frame;
|
||||
packet!: NodeAv.Packet;
|
||||
codecContext!: NodeAv.CodecContext;
|
||||
state!: NodeAvState;
|
||||
|
||||
// eslint-disable-next-line @typescript-eslint/no-unused-vars
|
||||
static override supports(codec: AudioCodec, config: AudioDecoderConfig): boolean {
|
||||
@@ -29,10 +47,13 @@ export class NodeAvAudioDecoder extends CustomAudioDecoder {
|
||||
}
|
||||
|
||||
async init(): Promise<void> {
|
||||
this.frame = new NodeAv.Frame();
|
||||
this.frame.alloc();
|
||||
this.packet = new NodeAv.Packet();
|
||||
this.packet.alloc();
|
||||
const frame = new NodeAv.Frame();
|
||||
frame.alloc();
|
||||
const packet = new NodeAv.Packet();
|
||||
packet.alloc();
|
||||
this.state = { frame, packet, codecContext: null };
|
||||
|
||||
finalizationRegistry?.register(this, this.state, this);
|
||||
|
||||
const codecId = CODEC_TO_CODEC_ID[this.codec];
|
||||
assert(codecId !== undefined);
|
||||
@@ -57,22 +78,26 @@ export class NodeAvAudioDecoder extends CustomAudioDecoder {
|
||||
const ret = await codecContext.open2();
|
||||
NodeAv.FFmpegError.throwIfError(ret, 'Open codec context');
|
||||
|
||||
this.codecContext = codecContext;
|
||||
this.state.codecContext = codecContext;
|
||||
}
|
||||
|
||||
async decode(packet: EncodedPacket): Promise<void> {
|
||||
this.packet.isKeyframe = packet.type === 'key';
|
||||
this.packet.data = Buffer.from(packet.data);
|
||||
this.packet.timeBase = { num: 1, den: this.config.sampleRate };
|
||||
this.packet.pts = BigInt(Math.round(packet.timestamp * this.config.sampleRate));
|
||||
this.packet.dts = NodeAv.AV_NOPTS_VALUE;
|
||||
this.packet.duration = BigInt(Math.round(packet.duration * this.config.sampleRate));
|
||||
assert(this.state.codecContext);
|
||||
|
||||
const ret = await this.codecContext.sendPacket(this.packet);
|
||||
this.state.packet.isKeyframe = packet.type === 'key';
|
||||
this.state.packet.data = Buffer.from(packet.data);
|
||||
this.state.packet.timeBase = { num: 1, den: this.config.sampleRate };
|
||||
this.state.packet.pts = BigInt(Math.round(packet.timestamp * this.config.sampleRate));
|
||||
this.state.packet.dts = NodeAv.AV_NOPTS_VALUE;
|
||||
this.state.packet.duration = BigInt(Math.round(packet.duration * this.config.sampleRate));
|
||||
|
||||
const ret = await this.state.codecContext.sendPacket(this.state.packet);
|
||||
NodeAv.FFmpegError.throwIfError(ret, 'Send packet');
|
||||
|
||||
this.state.packet.unref(); // Don't need the data again, so just unref it
|
||||
|
||||
while (true) {
|
||||
const receiveRet = await this.codecContext.receiveFrame(this.frame);
|
||||
const receiveRet = await this.state.codecContext.receiveFrame(this.state.frame);
|
||||
if (receiveRet === NodeAv.AVERROR_EAGAIN || receiveRet === NodeAv.AVERROR_EOF) {
|
||||
break;
|
||||
}
|
||||
@@ -84,7 +109,7 @@ export class NodeAvAudioDecoder extends CustomAudioDecoder {
|
||||
receiveFrame(ret: number) {
|
||||
NodeAv.FFmpegError.throwIfError(ret, 'Receive frame');
|
||||
|
||||
const clone = this.frame.clone();
|
||||
const clone = this.state.frame.clone();
|
||||
if (!clone) {
|
||||
throw new Error('Allocation failure during frame clone.');
|
||||
}
|
||||
@@ -94,13 +119,15 @@ export class NodeAvAudioDecoder extends CustomAudioDecoder {
|
||||
}
|
||||
|
||||
async flush(): Promise<void> {
|
||||
assert(this.state.codecContext);
|
||||
|
||||
// Send null packet to signal flush
|
||||
const ret = await this.codecContext.sendPacket(null);
|
||||
const ret = await this.state.codecContext.sendPacket(null);
|
||||
NodeAv.FFmpegError.throwIfError(ret, 'Flush decoder');
|
||||
|
||||
// Keep receiving frames until no more are available
|
||||
while (true) {
|
||||
const receiveRet = await this.codecContext.receiveFrame(this.frame);
|
||||
const receiveRet = await this.state.codecContext.receiveFrame(this.state.frame);
|
||||
if (receiveRet === NodeAv.AVERROR_EAGAIN || receiveRet === NodeAv.AVERROR_EOF) {
|
||||
// No more frames available
|
||||
break;
|
||||
@@ -109,12 +136,11 @@ export class NodeAvAudioDecoder extends CustomAudioDecoder {
|
||||
this.receiveFrame(receiveRet);
|
||||
}
|
||||
|
||||
this.codecContext.flushBuffers();
|
||||
this.state.codecContext.flushBuffers();
|
||||
}
|
||||
|
||||
close(): MaybePromise<void> {
|
||||
this.codecContext.freeContext();
|
||||
this.frame.free();
|
||||
this.packet.free();
|
||||
finalizationRegistry?.unregister(this);
|
||||
freeState(this.state);
|
||||
}
|
||||
}
|
||||
|
||||
@@ -32,19 +32,39 @@ const AC3_SAMPLE_RATES = [32000, 44100, 48000];
|
||||
|
||||
const FRAME_SIZE_FALLBACK = 1024; // Just 'cause
|
||||
|
||||
type NodeAvState = {
|
||||
frame: NodeAv.Frame;
|
||||
packet: NodeAv.Packet;
|
||||
codecContext: NodeAv.CodecContext | null;
|
||||
resampler: NodeAv.SoftwareResampleContext | null;
|
||||
dstFrame: NodeAv.Frame | null;
|
||||
};
|
||||
|
||||
const freeState = (state: NodeAvState) => {
|
||||
state.codecContext?.freeContext();
|
||||
state.frame.free();
|
||||
state.packet.free();
|
||||
state.dstFrame?.free();
|
||||
state.resampler?.free();
|
||||
};
|
||||
|
||||
// Needed for proper freeing if close isn't called
|
||||
let finalizationRegistry: FinalizationRegistry<NodeAvState> | null = null;
|
||||
if (typeof FinalizationRegistry !== 'undefined') {
|
||||
finalizationRegistry = new FinalizationRegistry<NodeAvState>((state) => {
|
||||
freeState(state);
|
||||
});
|
||||
}
|
||||
|
||||
export class NodeAvAudioEncoder extends CustomAudioEncoder {
|
||||
frame!: NodeAv.Frame;
|
||||
packet!: NodeAv.Packet;
|
||||
state!: NodeAvState;
|
||||
avCodec!: NodeAv.Codec;
|
||||
codecContext: NodeAv.CodecContext | null = null;
|
||||
firstExpectedTimestamp: number | null = null;
|
||||
outputTimestampOffset = 0;
|
||||
|
||||
resampler: NodeAv.SoftwareResampleContext | null = null;
|
||||
inputParametersKey: string | null = null;
|
||||
resamplerInputSampleRate: number | null = null;
|
||||
nextResamplerPts: bigint | null = null;
|
||||
dstFrame: NodeAv.Frame | null = null;
|
||||
packetEmitted = false;
|
||||
adtsHeaderTemplate: AdtsHeaderTemplate | null = null;
|
||||
|
||||
@@ -76,10 +96,13 @@ export class NodeAvAudioEncoder extends CustomAudioEncoder {
|
||||
}
|
||||
|
||||
async init(): Promise<void> {
|
||||
this.frame = new NodeAv.Frame();
|
||||
this.frame.alloc();
|
||||
this.packet = new NodeAv.Packet();
|
||||
this.packet.alloc();
|
||||
const frame = new NodeAv.Frame();
|
||||
frame.alloc();
|
||||
const packet = new NodeAv.Packet();
|
||||
packet.alloc();
|
||||
this.state = { frame, packet, codecContext: null, resampler: null, dstFrame: null };
|
||||
|
||||
finalizationRegistry?.register(this, this.state, this);
|
||||
|
||||
const codecId = CODEC_TO_CODEC_ID[this.codec];
|
||||
assert(codecId !== undefined);
|
||||
@@ -95,7 +118,7 @@ export class NodeAvAudioEncoder extends CustomAudioEncoder {
|
||||
}
|
||||
|
||||
async createCodecContext() {
|
||||
assert(this.codecContext === null);
|
||||
assert(this.state.codecContext === null);
|
||||
|
||||
const codecContext = new NodeAv.CodecContext();
|
||||
codecContext.allocContext3(this.avCodec);
|
||||
@@ -122,28 +145,32 @@ export class NodeAvAudioEncoder extends CustomAudioEncoder {
|
||||
const ret = await codecContext.open2();
|
||||
NodeAv.FFmpegError.throwIfError(ret, 'Open codec context');
|
||||
|
||||
this.codecContext = codecContext;
|
||||
this.state.codecContext = codecContext;
|
||||
}
|
||||
|
||||
async encode(audioSample: AudioSample): Promise<void> {
|
||||
if (this.codecContext === null) {
|
||||
if (this.state.codecContext === null) {
|
||||
await this.createCodecContext();
|
||||
assert(this.codecContext);
|
||||
assert(this.state.codecContext);
|
||||
}
|
||||
|
||||
this.firstExpectedTimestamp ??= audioSample.timestamp;
|
||||
|
||||
if (audioSample._data instanceof AvFrameAudioSampleResource) {
|
||||
this.frame.ref(audioSample._data.frame);
|
||||
// Release any buffers still referenced from the previous encode before reffing the new frame, otherwise
|
||||
// av_frame_ref leaks them
|
||||
// https://github.com/Vanilagy/mediabunny/issues/392
|
||||
this.state.frame.unref();
|
||||
this.state.frame.ref(audioSample._data.frame);
|
||||
} else {
|
||||
copyAudioSampleToAvFrame(audioSample, this.frame);
|
||||
copyAudioSampleToAvFrame(audioSample, this.state.frame);
|
||||
}
|
||||
|
||||
this.frame.pts = BigInt(Math.round(audioSample.timestamp * this.config.sampleRate));
|
||||
this.frame.duration = BigInt(Math.round(audioSample.duration * this.config.sampleRate));
|
||||
this.frame.timeBase = new NodeAv.Rational(1, this.config.sampleRate);
|
||||
this.state.frame.pts = BigInt(Math.round(audioSample.timestamp * this.config.sampleRate));
|
||||
this.state.frame.duration = BigInt(Math.round(audioSample.duration * this.config.sampleRate));
|
||||
this.state.frame.timeBase = new NodeAv.Rational(1, this.config.sampleRate);
|
||||
|
||||
const key = `${this.frame.sampleRate}:${this.frame.channels}:${this.frame.format}`;
|
||||
const key = `${this.state.frame.sampleRate}:${this.state.frame.channels}:${this.state.frame.format}`;
|
||||
if (this.inputParametersKey !== null && this.inputParametersKey !== key) {
|
||||
throw new Error(
|
||||
'Input audio parameters changed. For this audio encoder, you cannot change the input audio'
|
||||
@@ -156,83 +183,83 @@ export class NodeAvAudioEncoder extends CustomAudioEncoder {
|
||||
// 1. Format conversion is needed (sample format, sample rate, or channel count differs)
|
||||
// 2. The codec requires fixed frame sizes
|
||||
const requiresResampler
|
||||
= this.codecContext.frameSize > 0
|
||||
|| this.codecContext.sampleFormat !== this.frame.format
|
||||
|| this.codecContext.sampleRate !== this.frame.sampleRate
|
||||
|| this.codecContext.channels !== this.frame.channels;
|
||||
= this.state.codecContext.frameSize > 0
|
||||
|| this.state.codecContext.sampleFormat !== this.state.frame.format
|
||||
|| this.state.codecContext.sampleRate !== this.state.frame.sampleRate
|
||||
|| this.state.codecContext.channels !== this.state.frame.channels;
|
||||
|
||||
if (requiresResampler) {
|
||||
if (!this.resampler) {
|
||||
this.resampler = new NodeAv.SoftwareResampleContext();
|
||||
this.resamplerInputSampleRate = this.frame.sampleRate;
|
||||
if (!this.state.resampler) {
|
||||
this.state.resampler = new NodeAv.SoftwareResampleContext();
|
||||
this.resamplerInputSampleRate = this.state.frame.sampleRate;
|
||||
|
||||
const outLayout = getChannelLayout(this.codecContext.channels);
|
||||
const inLayout = getChannelLayout(this.frame.channels);
|
||||
const outLayout = getChannelLayout(this.state.codecContext.channels);
|
||||
const inLayout = getChannelLayout(this.state.frame.channels);
|
||||
|
||||
const ret = this.resampler.allocSetOpts2(
|
||||
outLayout, this.codecContext.sampleFormat, this.codecContext.sampleRate,
|
||||
inLayout, this.frame.format as NodeAv.AVSampleFormat, this.frame.sampleRate,
|
||||
const ret = this.state.resampler.allocSetOpts2(
|
||||
outLayout, this.state.codecContext.sampleFormat, this.state.codecContext.sampleRate,
|
||||
inLayout, this.state.frame.format as NodeAv.AVSampleFormat, this.state.frame.sampleRate,
|
||||
);
|
||||
NodeAv.FFmpegError.throwIfError(ret, 'allocSetOpts2');
|
||||
|
||||
const ret2 = this.resampler.init();
|
||||
const ret2 = this.state.resampler.init();
|
||||
NodeAv.FFmpegError.throwIfError(ret2, 'init');
|
||||
|
||||
this.dstFrame = new NodeAv.Frame();
|
||||
this.dstFrame.alloc();
|
||||
this.dstFrame.channelLayout = outLayout;
|
||||
this.dstFrame.sampleRate = this.codecContext.sampleRate;
|
||||
this.dstFrame.format = this.codecContext.sampleFormat;
|
||||
this.dstFrame.nbSamples = this.codecContext.frameSize || FRAME_SIZE_FALLBACK;
|
||||
this.dstFrame.duration = BigInt(this.dstFrame.nbSamples);
|
||||
this.dstFrame.allocBuffer();
|
||||
this.state.dstFrame = new NodeAv.Frame();
|
||||
this.state.dstFrame.alloc();
|
||||
this.state.dstFrame.channelLayout = outLayout;
|
||||
this.state.dstFrame.sampleRate = this.state.codecContext.sampleRate;
|
||||
this.state.dstFrame.format = this.state.codecContext.sampleFormat;
|
||||
this.state.dstFrame.nbSamples = this.state.codecContext.frameSize || FRAME_SIZE_FALLBACK;
|
||||
this.state.dstFrame.duration = BigInt(this.state.dstFrame.nbSamples);
|
||||
this.state.dstFrame.allocBuffer();
|
||||
|
||||
this.nextResamplerPts = this.frame.pts;
|
||||
this.nextResamplerPts = this.state.frame.pts;
|
||||
}
|
||||
|
||||
const inputBuffers = this.frame.data;
|
||||
const inputBuffers = this.state.frame.data;
|
||||
if (!inputBuffers) {
|
||||
throw new DOMException('Frame has no data', 'EncodingError');
|
||||
}
|
||||
await this.resampler.convert(null, 0, inputBuffers, this.frame.nbSamples);
|
||||
await this.state.resampler.convert(null, 0, inputBuffers, this.state.frame.nbSamples);
|
||||
|
||||
await this.pullResampledFrames();
|
||||
} else {
|
||||
await this.sendFrameAndReceivePackets(this.frame);
|
||||
await this.sendFrameAndReceivePackets(this.state.frame);
|
||||
}
|
||||
}
|
||||
|
||||
async pullResampledFrames() {
|
||||
assert(this.codecContext);
|
||||
assert(this.resampler);
|
||||
assert(this.dstFrame);
|
||||
assert(this.state.codecContext);
|
||||
assert(this.state.resampler);
|
||||
assert(this.state.dstFrame);
|
||||
assert(this.nextResamplerPts !== null);
|
||||
|
||||
const frameSize = this.codecContext.frameSize || FRAME_SIZE_FALLBACK;
|
||||
const frameSize = this.state.codecContext.frameSize || FRAME_SIZE_FALLBACK;
|
||||
|
||||
while (true) {
|
||||
const available = this.resampler.getOutSamples(0);
|
||||
const available = this.state.resampler.getOutSamples(0);
|
||||
if (available < frameSize) {
|
||||
break;
|
||||
}
|
||||
|
||||
await this.resampler.convert(this.dstFrame.data, frameSize, null, 0);
|
||||
await this.state.resampler.convert(this.state.dstFrame.data, frameSize, null, 0);
|
||||
|
||||
this.dstFrame.pts = this.nextResamplerPts;
|
||||
this.state.dstFrame.pts = this.nextResamplerPts;
|
||||
|
||||
await this.sendFrameAndReceivePackets(this.dstFrame);
|
||||
await this.sendFrameAndReceivePackets(this.state.dstFrame);
|
||||
this.nextResamplerPts += BigInt(frameSize);
|
||||
}
|
||||
}
|
||||
|
||||
async sendFrameAndReceivePackets(frame: NodeAv.Frame | null) {
|
||||
assert(this.codecContext);
|
||||
assert(this.state.codecContext);
|
||||
|
||||
const ret = await this.codecContext.sendFrame(frame);
|
||||
const ret = await this.state.codecContext.sendFrame(frame);
|
||||
NodeAv.FFmpegError.throwIfError(ret, 'Send frame');
|
||||
|
||||
while (true) {
|
||||
const receiveRet = await this.codecContext.receivePacket(this.packet);
|
||||
const receiveRet = await this.state.codecContext.receivePacket(this.state.packet);
|
||||
if (receiveRet === NodeAv.AVERROR_EAGAIN || receiveRet === NodeAv.AVERROR_EOF) {
|
||||
break;
|
||||
}
|
||||
@@ -242,18 +269,18 @@ export class NodeAvAudioEncoder extends CustomAudioEncoder {
|
||||
}
|
||||
|
||||
receivePacket(ret: number) {
|
||||
assert(this.codecContext);
|
||||
assert(this.state.codecContext);
|
||||
assert(this.firstExpectedTimestamp !== null);
|
||||
NodeAv.FFmpegError.throwIfError(ret, 'Receive packet');
|
||||
|
||||
if (!this.packet.data) {
|
||||
if (!this.state.packet.data) {
|
||||
return;
|
||||
}
|
||||
|
||||
let timestamp = Number(this.packet.pts) / this.codecContext.sampleRate;
|
||||
const duration = Number(this.packet.duration) / this.codecContext.sampleRate;
|
||||
let timestamp = Number(this.state.packet.pts) / this.state.codecContext.sampleRate;
|
||||
const duration = Number(this.state.packet.duration) / this.state.codecContext.sampleRate;
|
||||
|
||||
let data: Uint8Array = this.packet.data;
|
||||
let data: Uint8Array = this.state.packet.data;
|
||||
|
||||
let metadata: EncodedAudioChunkMetadata | undefined;
|
||||
if (this.packetEmitted) {
|
||||
@@ -265,8 +292,8 @@ export class NodeAvAudioEncoder extends CustomAudioEncoder {
|
||||
this.outputTimestampOffset = Math.max(this.firstExpectedTimestamp - timestamp, 0);
|
||||
|
||||
const codecString = this.config.codec;
|
||||
let description = this.codecContext.extraData
|
||||
? toUint8Array(this.codecContext.extraData)
|
||||
let description = this.state.codecContext.extraData
|
||||
? toUint8Array(this.state.codecContext.extraData)
|
||||
: undefined;
|
||||
|
||||
if (this.codec === 'aac') {
|
||||
@@ -311,8 +338,8 @@ export class NodeAvAudioEncoder extends CustomAudioEncoder {
|
||||
metadata = {
|
||||
decoderConfig: {
|
||||
codec: codecString,
|
||||
sampleRate: this.codecContext.sampleRate,
|
||||
numberOfChannels: this.codecContext.channels,
|
||||
sampleRate: this.state.codecContext.sampleRate,
|
||||
numberOfChannels: this.state.codecContext.channels,
|
||||
description,
|
||||
},
|
||||
};
|
||||
@@ -344,53 +371,50 @@ export class NodeAvAudioEncoder extends CustomAudioEncoder {
|
||||
}
|
||||
|
||||
async flush(): Promise<void> {
|
||||
if (!this.codecContext) {
|
||||
if (!this.state.codecContext) {
|
||||
return;
|
||||
}
|
||||
|
||||
outer:
|
||||
if (this.resampler) {
|
||||
if (this.state.resampler) {
|
||||
assert(this.resamplerInputSampleRate !== null);
|
||||
|
||||
const currentOutSamples = this.resampler.getOutSamples(0);
|
||||
const currentOutSamples = this.state.resampler.getOutSamples(0);
|
||||
if (currentOutSamples === 0) {
|
||||
break outer; // Clean cut-off point
|
||||
}
|
||||
|
||||
const frameSize = this.codecContext.frameSize || FRAME_SIZE_FALLBACK;
|
||||
const frameSize = this.state.codecContext.frameSize || FRAME_SIZE_FALLBACK;
|
||||
assert(currentOutSamples < frameSize); // Because if it's more, it would've already been retrieved
|
||||
|
||||
const inputSamplesNeeded = Math.ceil(
|
||||
((frameSize - currentOutSamples) / this.codecContext.sampleRate) * this.resamplerInputSampleRate,
|
||||
((frameSize - currentOutSamples) / this.state.codecContext.sampleRate) * this.resamplerInputSampleRate,
|
||||
);
|
||||
this.resampler.injectSilence(inputSamplesNeeded);
|
||||
this.state.resampler.injectSilence(inputSamplesNeeded);
|
||||
|
||||
await this.pullResampledFrames();
|
||||
}
|
||||
|
||||
await this.sendFrameAndReceivePackets(null);
|
||||
|
||||
this.codecContext.freeContext();
|
||||
this.codecContext = null;
|
||||
this.state.codecContext.freeContext();
|
||||
this.state.codecContext = null;
|
||||
this.packetEmitted = false;
|
||||
this.firstExpectedTimestamp = null;
|
||||
this.outputTimestampOffset = 0;
|
||||
this.adtsHeaderTemplate = null;
|
||||
|
||||
this.resampler?.free();
|
||||
this.resampler = null;
|
||||
this.state.resampler?.free();
|
||||
this.state.resampler = null;
|
||||
this.inputParametersKey = null;
|
||||
this.resamplerInputSampleRate = null;
|
||||
this.nextResamplerPts = null;
|
||||
this.dstFrame?.free();
|
||||
this.dstFrame = null;
|
||||
this.state.dstFrame?.free();
|
||||
this.state.dstFrame = null;
|
||||
}
|
||||
|
||||
close(): MaybePromise<void> {
|
||||
this.codecContext?.freeContext();
|
||||
this.frame.free();
|
||||
this.packet.free();
|
||||
this.dstFrame?.free();
|
||||
this.resampler?.free();
|
||||
finalizationRegistry?.unregister(this);
|
||||
freeState(this.state);
|
||||
}
|
||||
}
|
||||
|
||||
@@ -78,6 +78,8 @@ export { AvFrameAudioSampleResource } from './audio-sample';
|
||||
export const toAvFrame = async (sample: VideoSample | AudioSample, frame: NodeAv.Frame) => {
|
||||
if (sample instanceof VideoSample) {
|
||||
if (sample._data instanceof AvFrameVideoSampleResource) {
|
||||
// We're overriding the frame, so release whatever it referenced before, otherwise av_frame_ref leaks it
|
||||
frame.unref();
|
||||
frame.ref(sample._data.frame);
|
||||
} else {
|
||||
if (sample.format === null) {
|
||||
@@ -92,6 +94,8 @@ export const toAvFrame = async (sample: VideoSample | AudioSample, frame: NodeAv
|
||||
frame.timeBase = new NodeAv.Rational(1, 1e6);
|
||||
} else {
|
||||
if (sample._data instanceof AvFrameAudioSampleResource) {
|
||||
// We're overriding the frame, so release whatever it referenced before, otherwise av_frame_ref leaks it
|
||||
frame.unref();
|
||||
frame.ref(sample._data.frame);
|
||||
} else {
|
||||
copyAudioSampleToAvFrame(sample, frame);
|
||||
|
||||
@@ -12,10 +12,28 @@ import { CODEC_TO_CODEC_ID, getHardwareDecoderCodec } from './misc';
|
||||
import { assert, binarySearchLessOrEqual, simplifyRational, toUint8Array } from '../../../src/misc';
|
||||
import { AvFrameVideoSampleResource } from './video-sample';
|
||||
|
||||
type NodeAvState = {
|
||||
frame: NodeAv.Frame;
|
||||
packet: NodeAv.Packet;
|
||||
codecContext: NodeAv.CodecContext | null;
|
||||
};
|
||||
|
||||
const freeState = (state: NodeAvState) => {
|
||||
state.codecContext?.freeContext();
|
||||
state.frame.free();
|
||||
state.packet.free();
|
||||
};
|
||||
|
||||
// Needed for proper freeing if close isn't called
|
||||
let finalizationRegistry: FinalizationRegistry<NodeAvState> | null = null;
|
||||
if (typeof FinalizationRegistry !== 'undefined') {
|
||||
finalizationRegistry = new FinalizationRegistry<NodeAvState>((state) => {
|
||||
freeState(state);
|
||||
});
|
||||
}
|
||||
|
||||
export class NodeAvVideoDecoder extends CustomVideoDecoder {
|
||||
frame!: NodeAv.Frame;
|
||||
packet!: NodeAv.Packet;
|
||||
codecContext: NodeAv.CodecContext | null = null;
|
||||
state!: NodeAvState;
|
||||
pixelAspectRatio!: Rational;
|
||||
|
||||
// Bookkeeping to restore the original timing information
|
||||
@@ -33,14 +51,17 @@ export class NodeAvVideoDecoder extends CustomVideoDecoder {
|
||||
}
|
||||
|
||||
async init() {
|
||||
this.frame = new NodeAv.Frame();
|
||||
this.frame.alloc();
|
||||
this.packet = new NodeAv.Packet();
|
||||
this.packet.alloc();
|
||||
const frame = new NodeAv.Frame();
|
||||
frame.alloc();
|
||||
const packet = new NodeAv.Packet();
|
||||
packet.alloc();
|
||||
this.state = { frame, packet, codecContext: null };
|
||||
|
||||
finalizationRegistry?.register(this, this.state, this);
|
||||
}
|
||||
|
||||
async initCodecContext(packet: EncodedPacket) {
|
||||
assert(this.codecContext === null);
|
||||
assert(this.state.codecContext === null);
|
||||
|
||||
const codecId = CODEC_TO_CODEC_ID[this.codec];
|
||||
assert(codecId !== undefined);
|
||||
@@ -85,28 +106,29 @@ export class NodeAvVideoDecoder extends CustomVideoDecoder {
|
||||
const ret = await codecContext.open2();
|
||||
NodeAv.FFmpegError.throwIfError(ret, 'Open codec context');
|
||||
|
||||
this.codecContext = codecContext;
|
||||
this.state.codecContext = codecContext;
|
||||
}
|
||||
|
||||
async decode(packet: EncodedPacket) {
|
||||
if (this.codecContext === null) {
|
||||
if (this.state.codecContext === null) {
|
||||
await this.initCodecContext(packet);
|
||||
assert(this.codecContext);
|
||||
}
|
||||
|
||||
this.packet.isKeyframe = packet.type === 'key';
|
||||
this.packet.data = Buffer.from(packet.data);
|
||||
this.packet.timeBase = { num: 1, den: 1e6 };
|
||||
this.packet.pts = BigInt(packet.microsecondTimestamp);
|
||||
this.packet.dts = NodeAv.AV_NOPTS_VALUE;
|
||||
this.packet.duration = BigInt(packet.microsecondDuration);
|
||||
assert(this.state.codecContext);
|
||||
|
||||
this.state.packet.isKeyframe = packet.type === 'key';
|
||||
this.state.packet.data = Buffer.from(packet.data);
|
||||
this.state.packet.timeBase = { num: 1, den: 1e6 };
|
||||
this.state.packet.pts = BigInt(packet.microsecondTimestamp);
|
||||
this.state.packet.dts = NodeAv.AV_NOPTS_VALUE;
|
||||
this.state.packet.duration = BigInt(packet.microsecondDuration);
|
||||
|
||||
if (packet.sideData.alpha) {
|
||||
const matroskaBlockAdditional = Buffer.alloc(8 + packet.sideData.alpha.byteLength);
|
||||
matroskaBlockAdditional[7] = 1; // BlockAddId
|
||||
matroskaBlockAdditional.set(packet.sideData.alpha, 8);
|
||||
|
||||
this.packet.addSideData(NodeAv.AV_PKT_DATA_MATROSKA_BLOCKADDITIONAL, matroskaBlockAdditional);
|
||||
this.state.packet.addSideData(NodeAv.AV_PKT_DATA_MATROSKA_BLOCKADDITIONAL, matroskaBlockAdditional);
|
||||
}
|
||||
|
||||
const preciseTimingIndex = binarySearchLessOrEqual(
|
||||
@@ -141,11 +163,13 @@ export class NodeAvVideoDecoder extends CustomVideoDecoder {
|
||||
}
|
||||
}
|
||||
|
||||
const ret = await this.codecContext.sendPacket(this.packet);
|
||||
const ret = await this.state.codecContext.sendPacket(this.state.packet);
|
||||
NodeAv.FFmpegError.throwIfError(ret, 'Send packet');
|
||||
|
||||
this.state.packet.unref(); // Don't need the data again, so just unref it
|
||||
|
||||
while (true) {
|
||||
const receiveRet = await this.codecContext.receiveFrame(this.frame);
|
||||
const receiveRet = await this.state.codecContext.receiveFrame(this.state.frame);
|
||||
if (receiveRet === NodeAv.AVERROR_EAGAIN || receiveRet === NodeAv.AVERROR_EOF) {
|
||||
break;
|
||||
}
|
||||
@@ -157,14 +181,14 @@ export class NodeAvVideoDecoder extends CustomVideoDecoder {
|
||||
receiveFrame(ret: number) {
|
||||
NodeAv.FFmpegError.throwIfError(ret, 'Receive frame');
|
||||
|
||||
this.frame.sampleAspectRatio = new NodeAv.Rational(this.pixelAspectRatio.num, this.pixelAspectRatio.den);
|
||||
this.state.frame.sampleAspectRatio = new NodeAv.Rational(this.pixelAspectRatio.num, this.pixelAspectRatio.den);
|
||||
|
||||
let timestamp = Number(this.frame.pts) / 1e6;
|
||||
let duration = Number(this.frame.duration) / 1e6;
|
||||
let timestamp = Number(this.state.frame.pts) / 1e6;
|
||||
let duration = Number(this.state.frame.duration) / 1e6;
|
||||
|
||||
const preciseTimingIndex = binarySearchLessOrEqual(
|
||||
this.preciseTimings,
|
||||
Number(this.frame.pts),
|
||||
Number(this.state.frame.pts),
|
||||
x => x.microsecondTimestamp,
|
||||
);
|
||||
const entry = preciseTimingIndex !== -1
|
||||
@@ -173,7 +197,7 @@ export class NodeAvVideoDecoder extends CustomVideoDecoder {
|
||||
|
||||
// If there's a relevant timing entry, refine the frame's timing data to get better accuracy than
|
||||
// microseconds
|
||||
if (entry && entry.microsecondTimestamp === Number(this.frame.pts)) {
|
||||
if (entry && entry.microsecondTimestamp === Number(this.state.frame.pts)) {
|
||||
if (entry.timestampIsValid) {
|
||||
timestamp = entry.timestamp;
|
||||
}
|
||||
@@ -182,7 +206,7 @@ export class NodeAvVideoDecoder extends CustomVideoDecoder {
|
||||
}
|
||||
}
|
||||
|
||||
const clone = this.frame.clone();
|
||||
const clone = this.state.frame.clone();
|
||||
if (!clone) {
|
||||
throw new Error('Frame clone allocation failed.');
|
||||
}
|
||||
@@ -194,17 +218,17 @@ export class NodeAvVideoDecoder extends CustomVideoDecoder {
|
||||
}
|
||||
|
||||
async flush() {
|
||||
if (!this.codecContext) {
|
||||
if (!this.state.codecContext) {
|
||||
return;
|
||||
}
|
||||
|
||||
// Send null packet to signal flush
|
||||
const ret = await this.codecContext.sendPacket(null);
|
||||
const ret = await this.state.codecContext.sendPacket(null);
|
||||
NodeAv.FFmpegError.throwIfError(ret, 'Flush decoder');
|
||||
|
||||
// Keep receiving frames until no more are available
|
||||
while (true) {
|
||||
const receiveRet = await this.codecContext.receiveFrame(this.frame);
|
||||
const receiveRet = await this.state.codecContext.receiveFrame(this.state.frame);
|
||||
if (receiveRet === NodeAv.AVERROR_EAGAIN || receiveRet === NodeAv.AVERROR_EOF) {
|
||||
// No more frames available
|
||||
break;
|
||||
@@ -213,12 +237,11 @@ export class NodeAvVideoDecoder extends CustomVideoDecoder {
|
||||
this.receiveFrame(receiveRet);
|
||||
}
|
||||
|
||||
this.codecContext.flushBuffers();
|
||||
this.state.codecContext.flushBuffers();
|
||||
}
|
||||
|
||||
close(): MaybePromise<void> {
|
||||
this.codecContext?.freeContext();
|
||||
this.frame.free();
|
||||
this.packet.free();
|
||||
finalizationRegistry?.unregister(this);
|
||||
freeState(this.state);
|
||||
}
|
||||
}
|
||||
|
||||
@@ -41,17 +41,36 @@ import {
|
||||
import { extractVideoCodecString } from '../../../src/codec';
|
||||
import { assert, binarySearchLessOrEqual, simplifyRational, toUint8Array } from '../../../src/misc';
|
||||
|
||||
type NodeAvState = {
|
||||
frame: NodeAv.Frame;
|
||||
packet: NodeAv.Packet;
|
||||
codecContext: NodeAv.CodecContext | null;
|
||||
scaler: NodeAv.SoftwareScaleContext | null;
|
||||
dstFrame: NodeAv.Frame | null;
|
||||
};
|
||||
|
||||
const freeState = (state: NodeAvState) => {
|
||||
state.codecContext?.freeContext();
|
||||
state.frame.free();
|
||||
state.packet.free();
|
||||
state.scaler?.freeContext();
|
||||
state.dstFrame?.free();
|
||||
};
|
||||
|
||||
// Needed for proper freeing if close isn't called
|
||||
let finalizationRegistry: FinalizationRegistry<NodeAvState> | null = null;
|
||||
if (typeof FinalizationRegistry !== 'undefined') {
|
||||
finalizationRegistry = new FinalizationRegistry<NodeAvState>((state) => {
|
||||
freeState(state);
|
||||
});
|
||||
}
|
||||
|
||||
export class NodeAvVideoEncoder extends CustomVideoEncoder {
|
||||
frame!: NodeAv.Frame;
|
||||
packet!: NodeAv.Packet;
|
||||
state!: NodeAvState;
|
||||
avCodec!: NodeAv.Codec;
|
||||
codecContext: NodeAv.CodecContext | null = null;
|
||||
lastBuffer: Buffer | null = null;
|
||||
packetEmitted = false;
|
||||
|
||||
scaler: NodeAv.SoftwareScaleContext | null = null;
|
||||
lastScalerKey: string | null = null;
|
||||
dstFrame: NodeAv.Frame | null = null;
|
||||
|
||||
// Bookkeeping to restore the original timing information
|
||||
preciseTimings: {
|
||||
@@ -68,12 +87,16 @@ export class NodeAvVideoEncoder extends CustomVideoEncoder {
|
||||
}
|
||||
|
||||
async init(): Promise<void> {
|
||||
this.frame = new NodeAv.Frame();
|
||||
this.frame.alloc();
|
||||
this.frame.timeBase = new NodeAv.Rational(1, 1e6);
|
||||
const frame = new NodeAv.Frame();
|
||||
frame.alloc();
|
||||
frame.timeBase = new NodeAv.Rational(1, 1e6);
|
||||
|
||||
this.packet = new NodeAv.Packet();
|
||||
this.packet.alloc();
|
||||
const packet = new NodeAv.Packet();
|
||||
packet.alloc();
|
||||
|
||||
this.state = { frame, packet, codecContext: null, scaler: null, dstFrame: null };
|
||||
|
||||
finalizationRegistry?.register(this, this.state, this);
|
||||
|
||||
const codecId = CODEC_TO_CODEC_ID[this.codec];
|
||||
assert(codecId !== undefined);
|
||||
@@ -97,7 +120,7 @@ export class NodeAvVideoEncoder extends CustomVideoEncoder {
|
||||
}
|
||||
|
||||
async createCodecContext() {
|
||||
assert(this.codecContext === null);
|
||||
assert(this.state.codecContext === null);
|
||||
|
||||
const codecContext = new NodeAv.CodecContext();
|
||||
codecContext.allocContext3(this.avCodec);
|
||||
@@ -177,65 +200,69 @@ export class NodeAvVideoEncoder extends CustomVideoEncoder {
|
||||
const ret = await codecContext.open2();
|
||||
NodeAv.FFmpegError.throwIfError(ret, 'Open codec context');
|
||||
|
||||
this.codecContext = codecContext;
|
||||
this.state.codecContext = codecContext;
|
||||
}
|
||||
|
||||
async encode(videoSample: VideoSample, options: VideoEncoderEncodeOptions): Promise<void> {
|
||||
if (this.codecContext === null) {
|
||||
if (this.state.codecContext === null) {
|
||||
await this.createCodecContext();
|
||||
assert(this.codecContext);
|
||||
}
|
||||
assert(this.state.codecContext);
|
||||
|
||||
if (videoSample._data instanceof AvFrameVideoSampleResource) {
|
||||
this.frame.ref(videoSample._data.frame);
|
||||
// Release any buffers still referenced from the previous encode before reffing the new frame, otherwise
|
||||
// av_frame_ref leaks them
|
||||
// https://github.com/Vanilagy/mediabunny/issues/392
|
||||
this.state.frame.unref();
|
||||
this.state.frame.ref(videoSample._data.frame);
|
||||
} else {
|
||||
if (videoSample.format === null) {
|
||||
throw new Error('Cannot encode foreign VideoSample with unknown (null) format.');
|
||||
}
|
||||
|
||||
this.lastBuffer = await copyVideoSampleToAvFrame(videoSample, this.frame, this.lastBuffer);
|
||||
this.lastBuffer = await copyVideoSampleToAvFrame(videoSample, this.state.frame, this.lastBuffer);
|
||||
}
|
||||
|
||||
let frameToEncode = this.frame;
|
||||
let frameToEncode = this.state.frame;
|
||||
|
||||
const requiresScaler
|
||||
= this.codecContext.pixelFormat !== this.frame.format
|
||||
|| this.codecContext.width !== this.frame.width
|
||||
|| this.codecContext.height !== this.frame.height;
|
||||
= this.state.codecContext.pixelFormat !== this.state.frame.format
|
||||
|| this.state.codecContext.width !== this.state.frame.width
|
||||
|| this.state.codecContext.height !== this.state.frame.height;
|
||||
|
||||
if (requiresScaler) {
|
||||
if (!this.scaler) {
|
||||
this.scaler = new NodeAv.SoftwareScaleContext();
|
||||
if (!this.state.scaler) {
|
||||
this.state.scaler = new NodeAv.SoftwareScaleContext();
|
||||
}
|
||||
|
||||
const key = `${this.frame.width}x${this.frame.height}:${this.frame.format}`;
|
||||
const key = `${this.state.frame.width}x${this.state.frame.height}:${this.state.frame.format}`;
|
||||
const needsConfigure = key !== this.lastScalerKey;
|
||||
|
||||
if (needsConfigure) {
|
||||
this.scaler.getContext(
|
||||
this.frame.width, this.frame.height, this.frame.format as NodeAv.AVPixelFormat,
|
||||
this.codecContext.width, this.codecContext.height, this.codecContext.pixelFormat,
|
||||
this.state.scaler.getContext(
|
||||
this.state.frame.width, this.state.frame.height, this.state.frame.format as NodeAv.AVPixelFormat,
|
||||
this.state.codecContext.width, this.state.codecContext.height, this.state.codecContext.pixelFormat,
|
||||
NodeAv.SWS_FAST_BILINEAR,
|
||||
);
|
||||
|
||||
this.lastScalerKey = key;
|
||||
|
||||
const ret = this.scaler.initContext();
|
||||
const ret = this.state.scaler.initContext();
|
||||
NodeAv.FFmpegError.throwIfError(ret, 'initContext');
|
||||
}
|
||||
|
||||
if (!this.dstFrame) {
|
||||
this.dstFrame = new NodeAv.Frame();
|
||||
this.dstFrame.alloc();
|
||||
this.dstFrame.width = this.codecContext.width;
|
||||
this.dstFrame.height = this.codecContext.height;
|
||||
this.dstFrame.format = this.codecContext.pixelFormat;
|
||||
this.dstFrame.allocBuffer();
|
||||
if (!this.state.dstFrame) {
|
||||
this.state.dstFrame = new NodeAv.Frame();
|
||||
this.state.dstFrame.alloc();
|
||||
this.state.dstFrame.width = this.state.codecContext.width;
|
||||
this.state.dstFrame.height = this.state.codecContext.height;
|
||||
this.state.dstFrame.format = this.state.codecContext.pixelFormat;
|
||||
this.state.dstFrame.allocBuffer();
|
||||
}
|
||||
|
||||
await this.scaler.scaleFrame(this.dstFrame, this.frame);
|
||||
this.dstFrame.copyProps(this.frame);
|
||||
frameToEncode = this.dstFrame;
|
||||
await this.state.scaler.scaleFrame(this.state.dstFrame, this.state.frame);
|
||||
this.state.dstFrame.copyProps(this.state.frame);
|
||||
frameToEncode = this.state.dstFrame;
|
||||
}
|
||||
|
||||
frameToEncode.pts = BigInt(videoSample.microsecondTimestamp);
|
||||
@@ -282,12 +309,12 @@ export class NodeAvVideoEncoder extends CustomVideoEncoder {
|
||||
}
|
||||
}
|
||||
|
||||
const ret = await this.codecContext.sendFrame(frameToEncode);
|
||||
const ret = await this.state.codecContext.sendFrame(frameToEncode);
|
||||
NodeAv.FFmpegError.throwIfError(ret, 'Send frame');
|
||||
|
||||
// Keep receiving packets until no more are available for this frame
|
||||
while (true) {
|
||||
const receiveRet = await this.codecContext.receivePacket(this.packet);
|
||||
const receiveRet = await this.state.codecContext.receivePacket(this.state.packet);
|
||||
if (receiveRet === NodeAv.AVERROR_EAGAIN || receiveRet === NodeAv.AVERROR_EOF) {
|
||||
break;
|
||||
}
|
||||
@@ -297,20 +324,20 @@ export class NodeAvVideoEncoder extends CustomVideoEncoder {
|
||||
}
|
||||
|
||||
receivePacket(ret: number) {
|
||||
assert(this.codecContext);
|
||||
assert(this.state.codecContext);
|
||||
NodeAv.FFmpegError.throwIfError(ret, 'Receive packet');
|
||||
|
||||
if (!this.packet.data) {
|
||||
if (!this.state.packet.data) {
|
||||
return;
|
||||
}
|
||||
let packetData = toUint8Array(this.packet.data);
|
||||
let packetData = toUint8Array(this.state.packet.data);
|
||||
|
||||
let timestamp = Number(this.packet.pts) / 1e6;
|
||||
let duration = Number(this.packet.duration) / 1e6;
|
||||
let timestamp = Number(this.state.packet.pts) / 1e6;
|
||||
let duration = Number(this.state.packet.duration) / 1e6;
|
||||
|
||||
const preciseTimingIndex = binarySearchLessOrEqual(
|
||||
this.preciseTimings,
|
||||
Number(this.packet.pts),
|
||||
Number(this.state.packet.pts),
|
||||
x => x.microsecondTimestamp,
|
||||
);
|
||||
const entry = preciseTimingIndex !== -1
|
||||
@@ -319,7 +346,7 @@ export class NodeAvVideoEncoder extends CustomVideoEncoder {
|
||||
|
||||
// If there's a relevant timing entry, refine the packet's timing data to get better accuracy than
|
||||
// microseconds
|
||||
if (entry && entry.microsecondTimestamp === Number(this.packet.pts)) {
|
||||
if (entry && entry.microsecondTimestamp === Number(this.state.packet.pts)) {
|
||||
if (entry.timestampIsValid) {
|
||||
timestamp = entry.timestamp;
|
||||
}
|
||||
@@ -346,14 +373,14 @@ export class NodeAvVideoEncoder extends CustomVideoEncoder {
|
||||
let serializedRecord: Uint8Array;
|
||||
|
||||
if (this.codec === 'avc') {
|
||||
const record = extractAvcDecoderConfigurationRecord(this.packet.data);
|
||||
const record = extractAvcDecoderConfigurationRecord(this.state.packet.data);
|
||||
if (!record) {
|
||||
throw new Error('Invalid AVC data, could not extract decoder configuration record.');
|
||||
}
|
||||
|
||||
serializedRecord = serializeAvcDecoderConfigurationRecord(record);
|
||||
} else {
|
||||
const record = extractHevcDecoderConfigurationRecord(this.packet.data);
|
||||
const record = extractHevcDecoderConfigurationRecord(this.state.packet.data);
|
||||
if (!record) {
|
||||
throw new Error('Invalid HEVC data, could not extract decoder configuration record.');
|
||||
}
|
||||
@@ -488,14 +515,14 @@ export class NodeAvVideoEncoder extends CustomVideoEncoder {
|
||||
}
|
||||
|
||||
const sideData: EncodedPacketSideData = {};
|
||||
const matroskaBlockAdditional = this.packet.getSideData(NodeAv.AV_PKT_DATA_MATROSKA_BLOCKADDITIONAL);
|
||||
const matroskaBlockAdditional = this.state.packet.getSideData(NodeAv.AV_PKT_DATA_MATROSKA_BLOCKADDITIONAL);
|
||||
if (matroskaBlockAdditional) {
|
||||
sideData.alpha = toUint8Array(matroskaBlockAdditional).subarray(8); // Skip the BlockAddId
|
||||
}
|
||||
|
||||
const packet = new EncodedPacket(
|
||||
packetData,
|
||||
this.packet.isKeyframe ? 'key' : 'delta',
|
||||
this.state.packet.isKeyframe ? 'key' : 'delta',
|
||||
timestamp,
|
||||
duration,
|
||||
undefined,
|
||||
@@ -507,21 +534,19 @@ export class NodeAvVideoEncoder extends CustomVideoEncoder {
|
||||
// Create the decoder config
|
||||
metadata.decoderConfig = {
|
||||
codec: decoderConfigCodecString,
|
||||
codedWidth: this.codecContext.width,
|
||||
codedHeight: this.codecContext.height,
|
||||
displayAspectWidth: this.config.displayWidth ?? this.codecContext.width,
|
||||
displayAspectHeight: this.config.displayHeight ?? this.codecContext.height,
|
||||
codedWidth: this.state.codecContext.width,
|
||||
codedHeight: this.state.codecContext.height,
|
||||
displayAspectWidth: this.config.displayWidth ?? this.state.codecContext.width,
|
||||
displayAspectHeight: this.config.displayHeight ?? this.state.codecContext.height,
|
||||
description: decoderConfigDescription ?? undefined,
|
||||
colorSpace: {
|
||||
primaries:
|
||||
unmapColorPrimaries(this.codecContext.colorPrimaries) as VideoColorPrimaries,
|
||||
matrix:
|
||||
unmapMatrixCoefficients(this.codecContext.colorSpace) as VideoMatrixCoefficients,
|
||||
primaries: unmapColorPrimaries(this.state.codecContext.colorPrimaries) as VideoColorPrimaries,
|
||||
matrix: unmapMatrixCoefficients(this.state.codecContext.colorSpace) as VideoMatrixCoefficients,
|
||||
transfer:
|
||||
unmapTransferCharacteristics(this.codecContext.colorTrc) as VideoTransferCharacteristics,
|
||||
fullRange: this.codecContext.colorRange === NodeAv.AVCOL_RANGE_JPEG
|
||||
unmapTransferCharacteristics(this.state.codecContext.colorTrc) as VideoTransferCharacteristics,
|
||||
fullRange: this.state.codecContext.colorRange === NodeAv.AVCOL_RANGE_JPEG
|
||||
? true
|
||||
: this.codecContext.colorRange === NodeAv.AVCOL_RANGE_MPEG
|
||||
: this.state.codecContext.colorRange === NodeAv.AVCOL_RANGE_MPEG
|
||||
? false
|
||||
: undefined,
|
||||
},
|
||||
@@ -533,14 +558,14 @@ export class NodeAvVideoEncoder extends CustomVideoEncoder {
|
||||
}
|
||||
|
||||
async flush(): Promise<void> {
|
||||
if (this.codecContext) {
|
||||
if (this.state.codecContext) {
|
||||
// Send null frame to signal flush
|
||||
const ret = await this.codecContext.sendFrame(null);
|
||||
const ret = await this.state.codecContext.sendFrame(null);
|
||||
NodeAv.FFmpegError.throwIfError(ret, 'Send frame');
|
||||
|
||||
// Keep receiving packets until no more are available
|
||||
while (true) {
|
||||
const receiveRet = await this.codecContext.receivePacket(this.packet);
|
||||
const receiveRet = await this.state.codecContext.receivePacket(this.state.packet);
|
||||
if (receiveRet === NodeAv.AVERROR_EAGAIN || receiveRet === NodeAv.AVERROR_EOF) {
|
||||
break;
|
||||
}
|
||||
@@ -548,8 +573,8 @@ export class NodeAvVideoEncoder extends CustomVideoEncoder {
|
||||
this.receivePacket(receiveRet);
|
||||
}
|
||||
|
||||
this.codecContext.freeContext();
|
||||
this.codecContext = null;
|
||||
this.state.codecContext.freeContext();
|
||||
this.state.codecContext = null;
|
||||
// The codec is done now and can't be reused. Any subsequent encode call will first need to recreate a
|
||||
// codec context.
|
||||
}
|
||||
@@ -558,10 +583,7 @@ export class NodeAvVideoEncoder extends CustomVideoEncoder {
|
||||
}
|
||||
|
||||
close(): MaybePromise<void> {
|
||||
this.codecContext?.freeContext();
|
||||
this.frame.free();
|
||||
this.packet.free();
|
||||
this.scaler?.freeContext();
|
||||
this.dstFrame?.free();
|
||||
finalizationRegistry?.unregister(this);
|
||||
freeState(this.state);
|
||||
}
|
||||
}
|
||||
|
||||
@@ -871,6 +871,88 @@ describe('Video', async () => {
|
||||
expect(plane[3]).toBe(0xff);
|
||||
});
|
||||
|
||||
// https://github.com/Vanilagy/mediabunny/issues/392
|
||||
test('Memory doesn\'t leak', async () => {
|
||||
const encoder = new NodeAvVideoEncoder();
|
||||
// @ts-expect-error Readonly
|
||||
encoder.codec = 'avc';
|
||||
// @ts-expect-error Readonly
|
||||
encoder.config = {
|
||||
codec: buildVideoCodecString('avc', 1920, 1080, 1e6),
|
||||
width: 1920,
|
||||
height: 1080,
|
||||
bitrate: 1e6,
|
||||
} satisfies VideoEncoderConfig;
|
||||
|
||||
const pendingPackets: EncodedPacket[] = [];
|
||||
const decodedFrames: VideoSample[] = [];
|
||||
let decoderConfig: VideoDecoderConfig | null = null;
|
||||
|
||||
// @ts-expect-error Readonly
|
||||
encoder.onPacket = (packet: EncodedPacket, meta: EncodedVideoChunkMetadata) => {
|
||||
if (meta.decoderConfig) {
|
||||
decoderConfig = meta.decoderConfig;
|
||||
}
|
||||
pendingPackets.push(packet);
|
||||
};
|
||||
|
||||
await encoder.init();
|
||||
|
||||
// Encode a single white frame to get one packet (and the decoder config) to drive the loop with
|
||||
const data = new Uint8Array(1920 * 1080 * 4).fill(0xff); // White
|
||||
using bootstrapSample = new VideoSample(data, {
|
||||
format: 'RGBX',
|
||||
codedWidth: 1920,
|
||||
codedHeight: 1080,
|
||||
timestamp: 0,
|
||||
duration: 1 / 30,
|
||||
});
|
||||
await encoder.encode(bootstrapSample, {});
|
||||
await encoder.flush();
|
||||
assert(decoderConfig);
|
||||
assert(pendingPackets.length > 0);
|
||||
|
||||
const sourcePacket = pendingPackets[0]!;
|
||||
pendingPackets.length = 0;
|
||||
|
||||
const decoder = new NodeAvVideoDecoder();
|
||||
// @ts-expect-error Readonly
|
||||
decoder.codec = 'avc';
|
||||
// @ts-expect-error Readonly
|
||||
decoder.config = decoderConfig;
|
||||
// @ts-expect-error Readonly
|
||||
decoder.onSample = (sample: VideoSample) => {
|
||||
decodedFrames.push(sample);
|
||||
};
|
||||
await decoder.init();
|
||||
|
||||
const iterations = 200;
|
||||
let baselineRss = 0;
|
||||
|
||||
for (let i = 1; i <= iterations; i++) {
|
||||
await decoder.decode(sourcePacket.clone({ timestamp: i / 30 }));
|
||||
|
||||
for (const frame of decodedFrames) {
|
||||
await encoder.encode(frame, {});
|
||||
frame.close();
|
||||
}
|
||||
decodedFrames.length = 0;
|
||||
pendingPackets.length = 0;
|
||||
|
||||
if (i === 1) {
|
||||
baselineRss = process.memoryUsage().rss;
|
||||
}
|
||||
}
|
||||
|
||||
const finalRss = process.memoryUsage().rss;
|
||||
const growth = finalRss - baselineRss;
|
||||
|
||||
expect(growth).toBeLessThan(100 * 1024 * 1024);
|
||||
|
||||
await encoder.close();
|
||||
await decoder.close();
|
||||
});
|
||||
|
||||
describe('VideoSample transformation', () => {
|
||||
// 400x400 image: red everywhere, with a 200x200 blue square filling the bottom-left quadrant.
|
||||
const TEST_IMAGE = (() => {
|
||||
|
||||
Reference in New Issue
Block a user