mirror of
https://github.com/mozilla/gecko-dev.git
synced 2025-02-26 20:30:41 +00:00

Without this patch, if the video source changes resolution, this would be reflected in a HTMLVideoElement rendering the track, even though the track is disabled. Such resizes would be observable both through resize events on the video element, through the videoWidth and videoHeight attributes, and by drawing the video element to a CanvasRenderingContext2D. Differential Revision: https://phabricator.services.mozilla.com/D174088
308 lines
11 KiB
C++
308 lines
11 KiB
C++
/* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*-*/
|
|
/* 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/. */
|
|
|
|
#ifndef VideoOutput_h
|
|
#define VideoOutput_h
|
|
|
|
#include "MediaTrackListener.h"
|
|
#include "VideoFrameContainer.h"
|
|
|
|
namespace mozilla {
|
|
|
|
static bool SetImageToBlackPixel(layers::PlanarYCbCrImage* aImage) {
|
|
uint8_t blackPixel[] = {0x10, 0x80, 0x80};
|
|
|
|
layers::PlanarYCbCrData data;
|
|
data.mYChannel = blackPixel;
|
|
data.mCbChannel = blackPixel + 1;
|
|
data.mCrChannel = blackPixel + 2;
|
|
data.mYStride = data.mCbCrStride = 1;
|
|
data.mPictureRect = gfx::IntRect(0, 0, 1, 1);
|
|
data.mYUVColorSpace = gfx::YUVColorSpace::BT601;
|
|
// This could be made FULL once bug 1568745 is complete. A black pixel being
|
|
// 0x00, 0x80, 0x80
|
|
data.mColorRange = gfx::ColorRange::LIMITED;
|
|
|
|
return aImage->CopyData(data);
|
|
}
|
|
|
|
class VideoOutput : public DirectMediaTrackListener {
|
|
protected:
|
|
typedef layers::Image Image;
|
|
typedef layers::ImageContainer ImageContainer;
|
|
typedef layers::ImageContainer::FrameID FrameID;
|
|
typedef layers::ImageContainer::ProducerID ProducerID;
|
|
|
|
virtual ~VideoOutput() = default;
|
|
|
|
void DropPastFrames() {
|
|
TimeStamp now = TimeStamp::Now();
|
|
size_t nrChunksInPast = 0;
|
|
for (const auto& idChunkPair : mFrames) {
|
|
const VideoChunk& chunk = idChunkPair.second;
|
|
if (chunk.mTimeStamp > now) {
|
|
break;
|
|
}
|
|
++nrChunksInPast;
|
|
}
|
|
if (nrChunksInPast > 1) {
|
|
// We need to keep one frame that starts in the past, because it only ends
|
|
// when the next frame starts (which also needs to be in the past for it
|
|
// to drop).
|
|
mFrames.RemoveElementsAt(0, nrChunksInPast - 1);
|
|
}
|
|
}
|
|
|
|
void SendFramesEnsureLocked() {
|
|
mMutex.AssertCurrentThreadOwns();
|
|
SendFrames();
|
|
}
|
|
|
|
void SendFrames() {
|
|
DropPastFrames();
|
|
|
|
if (mFrames.IsEmpty()) {
|
|
return;
|
|
}
|
|
|
|
if (!mEnabled && mDisabledBlackImageSent) {
|
|
return;
|
|
}
|
|
|
|
// Collect any new frames produced in this iteration.
|
|
AutoTArray<ImageContainer::NonOwningImage, 16> images;
|
|
PrincipalHandle lastPrincipalHandle = PRINCIPAL_HANDLE_NONE;
|
|
|
|
for (const auto& idChunkPair : mFrames) {
|
|
ImageContainer::FrameID frameId = idChunkPair.first;
|
|
const VideoChunk& chunk = idChunkPair.second;
|
|
const VideoFrame& frame = chunk.mFrame;
|
|
Image* image = frame.GetImage();
|
|
if (frame.GetForceBlack() || !mEnabled) {
|
|
if (!mBlackImage) {
|
|
RefPtr<Image> blackImage = mVideoFrameContainer->GetImageContainer()
|
|
->CreatePlanarYCbCrImage();
|
|
if (blackImage) {
|
|
// Sets the image to a single black pixel, which will be scaled to
|
|
// fill the rendered size.
|
|
if (SetImageToBlackPixel(blackImage->AsPlanarYCbCrImage())) {
|
|
mBlackImage = blackImage;
|
|
}
|
|
}
|
|
}
|
|
if (mBlackImage) {
|
|
image = mBlackImage;
|
|
}
|
|
}
|
|
if (!image) {
|
|
// We ignore null images.
|
|
continue;
|
|
}
|
|
images.AppendElement(ImageContainer::NonOwningImage(
|
|
image, chunk.mTimeStamp, frameId, mProducerID));
|
|
|
|
lastPrincipalHandle = chunk.GetPrincipalHandle();
|
|
|
|
if (!mEnabled && mBlackImage) {
|
|
MOZ_ASSERT(images.Length() == 1);
|
|
mDisabledBlackImageSent = true;
|
|
break;
|
|
}
|
|
}
|
|
|
|
if (images.IsEmpty()) {
|
|
// This could happen if the only images in mFrames are null. We leave the
|
|
// container at the current frame in this case.
|
|
mVideoFrameContainer->ClearFutureFrames();
|
|
return;
|
|
}
|
|
|
|
bool principalHandleChanged =
|
|
lastPrincipalHandle != PRINCIPAL_HANDLE_NONE &&
|
|
lastPrincipalHandle != mVideoFrameContainer->GetLastPrincipalHandle();
|
|
|
|
if (principalHandleChanged) {
|
|
mVideoFrameContainer->UpdatePrincipalHandleForFrameID(
|
|
lastPrincipalHandle, images.LastElement().mFrameID);
|
|
}
|
|
|
|
mVideoFrameContainer->SetCurrentFrames(
|
|
mFrames[0].second.mFrame.GetIntrinsicSize(), images);
|
|
mMainThread->Dispatch(NewRunnableMethod("VideoFrameContainer::Invalidate",
|
|
mVideoFrameContainer,
|
|
&VideoFrameContainer::Invalidate));
|
|
}
|
|
|
|
FrameID NewFrameID() {
|
|
mMutex.AssertCurrentThreadOwns();
|
|
return ++mFrameID;
|
|
}
|
|
|
|
public:
|
|
VideoOutput(VideoFrameContainer* aContainer, AbstractThread* aMainThread)
|
|
: mMutex("VideoOutput::mMutex"),
|
|
mVideoFrameContainer(aContainer),
|
|
mMainThread(aMainThread) {}
|
|
void NotifyRealtimeTrackData(MediaTrackGraph* aGraph, TrackTime aTrackOffset,
|
|
const MediaSegment& aMedia) override {
|
|
MOZ_ASSERT(aMedia.GetType() == MediaSegment::VIDEO);
|
|
const VideoSegment& video = static_cast<const VideoSegment&>(aMedia);
|
|
MutexAutoLock lock(mMutex);
|
|
for (VideoSegment::ConstChunkIterator i(video); !i.IsEnded(); i.Next()) {
|
|
if (!mLastFrameTime.IsNull() && i->mTimeStamp < mLastFrameTime) {
|
|
// Time can go backwards if the source is a captured MediaDecoder and
|
|
// it seeks, as the previously buffered frames would stretch into the
|
|
// future. If this happens, we clear the buffered frames and start over.
|
|
mFrames.ClearAndRetainStorage();
|
|
}
|
|
mFrames.AppendElement(std::make_pair(NewFrameID(), *i));
|
|
mLastFrameTime = i->mTimeStamp;
|
|
}
|
|
|
|
SendFramesEnsureLocked();
|
|
}
|
|
void NotifyRemoved(MediaTrackGraph* aGraph) override {
|
|
// Doesn't need locking by mMutex, since the direct listener is removed from
|
|
// the track before we get notified.
|
|
if (mFrames.Length() <= 1) {
|
|
// The compositor has already received the last frame.
|
|
mFrames.ClearAndRetainStorage();
|
|
mVideoFrameContainer->ClearFutureFrames();
|
|
return;
|
|
}
|
|
|
|
// The compositor has multiple frames. ClearFutureFrames() would only retain
|
|
// the first as that's normally the current one. We however stop doing
|
|
// SetCurrentFrames() once we've received the last frame in a track, so
|
|
// there might be old frames lingering. We'll find the current one and
|
|
// re-send that.
|
|
DropPastFrames();
|
|
mFrames.RemoveLastElements(mFrames.Length() - 1);
|
|
SendFrames();
|
|
mFrames.ClearAndRetainStorage();
|
|
}
|
|
void NotifyEnded(MediaTrackGraph* aGraph) override {
|
|
// Doesn't need locking by mMutex, since for the track to end, it must have
|
|
// been ended by the source, meaning that the source won't append more data.
|
|
if (mFrames.IsEmpty()) {
|
|
return;
|
|
}
|
|
|
|
// Re-send only the last one to the compositor.
|
|
mFrames.RemoveElementsAt(0, mFrames.Length() - 1);
|
|
SendFrames();
|
|
mFrames.ClearAndRetainStorage();
|
|
}
|
|
void NotifyEnabledStateChanged(MediaTrackGraph* aGraph,
|
|
bool aEnabled) override {
|
|
MutexAutoLock lock(mMutex);
|
|
mEnabled = aEnabled;
|
|
DropPastFrames();
|
|
if (mEnabled) {
|
|
mDisabledBlackImageSent = false;
|
|
}
|
|
if (!mEnabled || mFrames.Length() > 1) {
|
|
// Re-send frames when disabling, as new frames may not arrive. When
|
|
// enabling we keep them black until new frames arrive, or re-send if we
|
|
// already have frames in the future. If we're disabling and there are no
|
|
// frames available yet, we invent one. Unfortunately with a hardcoded
|
|
// size.
|
|
//
|
|
// Since mEnabled will affect whether
|
|
// frames are real, or black, we assign new FrameIDs whenever we re-send
|
|
// frames after an mEnabled change.
|
|
for (auto& idChunkPair : mFrames) {
|
|
idChunkPair.first = NewFrameID();
|
|
}
|
|
if (mFrames.IsEmpty()) {
|
|
VideoSegment v;
|
|
v.AppendFrame(nullptr, gfx::IntSize(640, 480), PRINCIPAL_HANDLE_NONE,
|
|
true, TimeStamp::Now());
|
|
mFrames.AppendElement(std::make_pair(NewFrameID(), *v.GetLastChunk()));
|
|
}
|
|
SendFramesEnsureLocked();
|
|
}
|
|
}
|
|
|
|
Mutex mMutex MOZ_UNANNOTATED;
|
|
TimeStamp mLastFrameTime;
|
|
// Once the frame is forced to black, we initialize mBlackImage for use in any
|
|
// following forced-black frames.
|
|
RefPtr<Image> mBlackImage;
|
|
// True once mBlackImage has been sent due to mEnabled being false.
|
|
bool mDisabledBlackImageSent = false;
|
|
bool mEnabled = true;
|
|
// This array is accessed from both the direct video thread, and the graph
|
|
// thread. Protected by mMutex.
|
|
nsTArray<std::pair<ImageContainer::FrameID, VideoChunk>> mFrames;
|
|
// Accessed from both the direct video thread, and the graph thread. Protected
|
|
// by mMutex.
|
|
FrameID mFrameID = 0;
|
|
const RefPtr<VideoFrameContainer> mVideoFrameContainer;
|
|
const RefPtr<AbstractThread> mMainThread;
|
|
const ProducerID mProducerID = ImageContainer::AllocateProducerID();
|
|
};
|
|
|
|
/**
|
|
* This listener observes the first video frame to arrive with a non-empty size,
|
|
* and renders it to its VideoFrameContainer.
|
|
*/
|
|
class FirstFrameVideoOutput : public VideoOutput {
|
|
public:
|
|
FirstFrameVideoOutput(VideoFrameContainer* aContainer,
|
|
AbstractThread* aMainThread)
|
|
: VideoOutput(aContainer, aMainThread) {
|
|
MOZ_ASSERT(NS_IsMainThread());
|
|
}
|
|
|
|
// NB that this overrides VideoOutput::NotifyRealtimeTrackData, so we can
|
|
// filter out all frames but the first one with a real size. This allows us to
|
|
// later re-use the logic in VideoOutput for rendering that frame.
|
|
void NotifyRealtimeTrackData(MediaTrackGraph* aGraph, TrackTime aTrackOffset,
|
|
const MediaSegment& aMedia) override {
|
|
MOZ_ASSERT(aMedia.GetType() == MediaSegment::VIDEO);
|
|
|
|
if (mInitialSizeFound) {
|
|
return;
|
|
}
|
|
|
|
const VideoSegment& video = static_cast<const VideoSegment&>(aMedia);
|
|
for (VideoSegment::ConstChunkIterator c(video); !c.IsEnded(); c.Next()) {
|
|
if (c->mFrame.GetIntrinsicSize() != gfx::IntSize(0, 0)) {
|
|
mInitialSizeFound = true;
|
|
|
|
mMainThread->Dispatch(NS_NewRunnableFunction(
|
|
"FirstFrameVideoOutput::FirstFrameRenderedSetter",
|
|
[self = RefPtr<FirstFrameVideoOutput>(this)] {
|
|
self->mFirstFrameRendered = true;
|
|
}));
|
|
|
|
// Pick the first frame and run it through the rendering code.
|
|
VideoSegment segment;
|
|
segment.AppendFrame(do_AddRef(c->mFrame.GetImage()),
|
|
c->mFrame.GetIntrinsicSize(),
|
|
c->mFrame.GetPrincipalHandle(),
|
|
c->mFrame.GetForceBlack(), c->mTimeStamp);
|
|
VideoOutput::NotifyRealtimeTrackData(aGraph, aTrackOffset, segment);
|
|
return;
|
|
}
|
|
}
|
|
}
|
|
|
|
// Main thread only.
|
|
Watchable<bool> mFirstFrameRendered = {
|
|
false, "FirstFrameVideoOutput::mFirstFrameRendered"};
|
|
|
|
private:
|
|
// Whether a frame with a concrete size has been received. May only be
|
|
// accessed on the MTG's appending thread. (this is a direct listener so we
|
|
// get called by whoever is producing this track's data)
|
|
bool mInitialSizeFound = false;
|
|
};
|
|
|
|
} // namespace mozilla
|
|
|
|
#endif // VideoOutput_h
|