shaka-packager/packager/media/formats/webm/segmenter.cc

444 lines
16 KiB
C++
Raw Normal View History

// Copyright 2015 Google LLC. All rights reserved.
//
// Use of this source code is governed by a BSD-style
// license that can be found in the LICENSE file or at
// https://developers.google.com/open-source/licenses/bsd
#include <packager/media/formats/webm/segmenter.h>
#include <mkvmuxer/mkvmuxerutil.h>
#include <packager/media/base/audio_stream_info.h>
#include <packager/media/base/media_handler.h>
#include <packager/media/base/muxer_options.h>
#include <packager/media/base/video_stream_info.h>
#include <packager/media/codecs/vp_codec_configuration_record.h>
#include <packager/media/event/muxer_listener.h>
#include <packager/media/event/progress_listener.h>
#include <packager/media/formats/webm/encryptor.h>
#include <packager/media/formats/webm/webm_constants.h>
#include <packager/version/version.h>
using mkvmuxer::AudioTrack;
using mkvmuxer::VideoTrack;
namespace shaka {
namespace media {
namespace webm {
namespace {
const int64_t kTimecodeScale = 1000000;
const int64_t kSecondsToNs = 1000000000L;
// Round to closest integer.
uint64_t Round(double value) {
return static_cast<uint64_t>(value + 0.5);
}
// There are three different kinds of timestamp here:
// (1) ISO-BMFF timestamp (seconds scaled by ISO-BMFF timescale)
// This is used in our MediaSample and StreamInfo structures.
// (2) WebM timecode (seconds scaled by kSecondsToNs / WebM timecode scale)
// This is used in most WebM structures.
// (3) Nanoseconds (seconds scaled by kSecondsToNs)
// This is used in some WebM structures, e.g. Frame.
// We use Nanoseconds as intermediate format here for conversion, in
// uint64_t/int64_t, which is sufficient to represent a time as large as 292
// years.
int64_t BmffTimestampToNs(int64_t timestamp, int64_t time_scale) {
// Casting to double is needed otherwise kSecondsToNs * timestamp may overflow
// uint64_t/int64_t.
return Round(static_cast<double>(timestamp) / time_scale * kSecondsToNs);
}
int64_t NsToBmffTimestamp(int64_t ns, int64_t time_scale) {
// Casting to double is needed otherwise ns * time_scale may overflow
// uint64_t/int64_t.
return Round(static_cast<double>(ns) / kSecondsToNs * time_scale);
}
int64_t NsToWebMTimecode(int64_t ns, int64_t timecode_scale) {
return ns / timecode_scale;
}
int64_t WebMTimecodeToNs(int64_t timecode, int64_t timecode_scale) {
return timecode * timecode_scale;
}
} // namespace
Segmenter::Segmenter(const MuxerOptions& options) : options_(options) {}
Segmenter::~Segmenter() {}
Status Segmenter::Initialize(const StreamInfo& info,
ProgressListener* progress_listener,
MuxerListener* muxer_listener) {
is_encrypted_ = info.is_encrypted();
duration_ = info.duration();
time_scale_ = info.time_scale();
muxer_listener_ = muxer_listener;
// Use media duration as progress target.
progress_target_ = info.duration();
progress_listener_ = progress_listener;
segment_info_.Init();
segment_info_.set_timecode_scale(kTimecodeScale);
const std::string version = GetPackagerVersion();
if (!version.empty()) {
segment_info_.set_writing_app(
(GetPackagerProjectUrl() + " version " + version).c_str());
}
if (options().segment_template.empty()) {
// Set an initial duration so the duration element is written; will be
// overwritten at the end. This works because this is a float and floats
// are always the same size.
segment_info_.set_duration(1);
}
// Create the track info.
// The seed is only used to create a UID which we overwrite later.
unsigned int seed = 0;
std::unique_ptr<mkvmuxer::Track> track;
Status status;
switch (info.stream_type()) {
case kStreamVideo: {
std::unique_ptr<VideoTrack> video_track(new VideoTrack(&seed));
status = InitializeVideoTrack(static_cast<const VideoStreamInfo&>(info),
video_track.get());
track = std::move(video_track);
break;
}
case kStreamAudio: {
std::unique_ptr<AudioTrack> audio_track(new AudioTrack(&seed));
status = InitializeAudioTrack(static_cast<const AudioStreamInfo&>(info),
audio_track.get());
track = std::move(audio_track);
break;
}
default:
NOTIMPLEMENTED() << "Not implemented for stream type: "
<< info.stream_type();
status = Status(error::UNIMPLEMENTED, "Not implemented for stream type");
}
if (!status.ok())
return status;
if (info.is_encrypted()) {
if (info.encryption_config().per_sample_iv_size != kWebMIvSize)
return Status(error::MUXER_FAILURE, "Incorrect size WebM encryption IV.");
status = UpdateTrackForEncryption(info.encryption_config().key_id,
track.get());
if (!status.ok())
return status;
}
tracks_.AddTrack(track.get(), info.track_id());
// number() is only available after the above instruction.
track_id_ = track->number();
// |tracks_| owns |track|.
track.release();
return DoInitialize();
}
Status Segmenter::Finalize() {
if (prev_sample_ && !prev_sample_->end_of_stream()) {
int64_t duration =
prev_sample_->pts() - first_timestamp_ + prev_sample_->duration();
segment_info_.set_duration(FromBmffTimestamp(duration));
}
return DoFinalize();
}
Status Segmenter::AddSample(const MediaSample& source_sample) {
std::shared_ptr<MediaSample> sample(source_sample.Clone());
if (sample_duration_ == 0) {
first_timestamp_ = sample->pts();
sample_duration_ = sample->duration();
if (muxer_listener_)
muxer_listener_->OnSampleDurationReady(sample_duration_);
}
UpdateProgress(sample->duration());
// This writes frames in a delay. Meaning that the previous frame is written
// on this call to AddSample. The current frame is stored until the next
// call. This is done to determine which frame is the last in a Cluster.
// This first block determines if this is a new Cluster and writes the
// previous frame first before creating the new Cluster.
Status status;
if (new_segment_ || new_subsegment_) {
status = NewSegment(sample->pts(), new_subsegment_);
} else {
status = WriteFrame(false /* write_duration */);
}
if (!status.ok())
return status;
if (is_encrypted_)
UpdateFrameForEncryption(sample.get());
new_subsegment_ = false;
new_segment_ = false;
prev_sample_ = sample;
return Status::OK;
}
Status Segmenter::FinalizeSegment(int64_t /*start_timestamp*/,
int64_t /*duration_timestamp*/,
bool is_subsegment) {
if (is_subsegment)
new_subsegment_ = true;
else
new_segment_ = true;
return WriteFrame(true /* write duration */);
}
float Segmenter::GetDurationInSeconds() const {
return WebMTimecodeToNs(segment_info_.duration(),
segment_info_.timecode_scale()) /
static_cast<double>(kSecondsToNs);
}
int64_t Segmenter::FromBmffTimestamp(int64_t bmff_timestamp) {
return NsToWebMTimecode(
BmffTimestampToNs(bmff_timestamp, time_scale_),
segment_info_.timecode_scale());
}
int64_t Segmenter::FromWebMTimecode(int64_t webm_timecode) {
return NsToBmffTimestamp(
WebMTimecodeToNs(webm_timecode, segment_info_.timecode_scale()),
time_scale_);
}
Status Segmenter::WriteSegmentHeader(uint64_t file_size, MkvWriter* writer) {
Status error_status(error::FILE_FAILURE, "Error writing segment header.");
if (!WriteEbmlHeader(writer))
return error_status;
if (WriteID(writer, libwebm::kMkvSegment) != 0)
return error_status;
const uint64_t segment_size_size = 8;
segment_payload_pos_ = writer->Position() + segment_size_size;
if (file_size > 0) {
// We want the size of the segment element, so subtract the header.
if (WriteUIntSize(writer, file_size - segment_payload_pos_,
segment_size_size) != 0)
return error_status;
if (!seek_head_.Write(writer))
return error_status;
} else {
if (SerializeInt(writer, mkvmuxer::kEbmlUnknownValue, segment_size_size) !=
0)
return error_status;
// We don't know the header size, so write a placeholder.
if (!seek_head_.WriteVoid(writer))
return error_status;
}
seek_head_.set_info_pos(writer->Position() - segment_payload_pos_);
if (!segment_info_.Write(writer))
return error_status;
seek_head_.set_tracks_pos(writer->Position() - segment_payload_pos_);
if (!tracks_.Write(writer))
return error_status;
return Status::OK;
}
Status Segmenter::SetCluster(int64_t start_webm_timecode,
uint64_t position,
MkvWriter* writer) {
const int64_t scale = segment_info_.timecode_scale();
cluster_.reset(new mkvmuxer::Cluster(start_webm_timecode, position, scale));
cluster_->Init(writer);
return Status::OK;
}
void Segmenter::UpdateProgress(uint64_t progress) {
accumulated_progress_ += progress;
if (!progress_listener_ || progress_target_ == 0)
return;
// It might happen that accumulated progress exceeds progress_target due to
// computation errors, e.g. rounding error. Cap it so it never reports > 100%
// progress.
if (accumulated_progress_ >= progress_target_) {
progress_listener_->OnProgress(1.0);
} else {
progress_listener_->OnProgress(static_cast<double>(accumulated_progress_) /
progress_target_);
}
}
Status Segmenter::InitializeVideoTrack(const VideoStreamInfo& info,
VideoTrack* track) {
if (info.codec() == kCodecAV1) {
track->set_codec_id("V_AV1");
if (!track->SetCodecPrivate(info.codec_config().data(),
info.codec_config().size())) {
return Status(error::INTERNAL_ERROR,
"Private codec data required for AV1 streams");
}
} else if (info.codec() == kCodecVP8) {
track->set_codec_id("V_VP8");
} else if (info.codec() == kCodecVP9) {
track->set_codec_id("V_VP9");
// The |StreamInfo::codec_config| field is stored using the MP4 format; we
// need to convert it to the WebM format.
VPCodecConfigurationRecord vp_config;
if (!vp_config.ParseMP4(info.codec_config())) {
return Status(error::INTERNAL_ERROR,
"Unable to parse VP9 codec configuration");
}
mkvmuxer::Colour colour;
if (vp_config.matrix_coefficients() != AVCOL_SPC_UNSPECIFIED) {
colour.set_matrix_coefficients(vp_config.matrix_coefficients());
}
if (vp_config.transfer_characteristics() != AVCOL_TRC_UNSPECIFIED) {
colour.set_transfer_characteristics(vp_config.transfer_characteristics());
}
if (vp_config.color_primaries() != AVCOL_PRI_UNSPECIFIED) {
colour.set_primaries(vp_config.color_primaries());
}
if (!track->SetColour(colour)) {
return Status(error::INTERNAL_ERROR,
"Failed to setup color element for VPx streams");
}
std::vector<uint8_t> codec_config;
vp_config.WriteWebM(&codec_config);
if (!track->SetCodecPrivate(codec_config.data(), codec_config.size())) {
return Status(error::INTERNAL_ERROR,
"Private codec data required for VPx streams");
}
} else {
LOG(ERROR) << "Only VP8, VP9 and AV1 video codecs are supported in WebM.";
return Status(error::UNIMPLEMENTED,
"Only VP8, VP9 and AV1 video codecs are supported in WebM.");
}
track->set_uid(info.track_id());
if (!info.language().empty())
track->set_language(info.language().c_str());
track->set_type(mkvmuxer::Tracks::kVideo);
track->set_width(info.width());
track->set_height(info.height());
track->set_display_height(info.height());
track->set_display_width(info.width() * info.pixel_width() /
info.pixel_height());
return Status::OK;
}
Status Segmenter::InitializeAudioTrack(const AudioStreamInfo& info,
AudioTrack* track) {
if (info.codec() == kCodecOpus) {
track->set_codec_id(mkvmuxer::Tracks::kOpusCodecId);
} else if (info.codec() == kCodecVorbis) {
track->set_codec_id(mkvmuxer::Tracks::kVorbisCodecId);
} else {
LOG(ERROR) << "Only Vorbis and Opus audio codec are supported in WebM.";
return Status(error::UNIMPLEMENTED,
"Only Vorbis and Opus audio codecs are supported in WebM.");
}
if (!track->SetCodecPrivate(info.codec_config().data(),
info.codec_config().size())) {
return Status(error::INTERNAL_ERROR,
"Private codec data required for audio streams");
}
track->set_uid(info.track_id());
if (!info.language().empty())
track->set_language(info.language().c_str());
track->set_type(mkvmuxer::Tracks::kAudio);
track->set_sample_rate(info.sampling_frequency());
track->set_channels(info.num_channels());
track->set_seek_pre_roll(info.seek_preroll_ns());
track->set_codec_delay(info.codec_delay_ns());
return Status::OK;
}
Status Segmenter::WriteFrame(bool write_duration) {
// Create a frame manually so we can create non-SimpleBlock frames. This
// is required to allow the frame duration to be added. If the duration
// is not set, then a SimpleBlock will still be written.
mkvmuxer::Frame frame;
if (!frame.Init(prev_sample_->data(), prev_sample_->data_size())) {
return Status(error::MUXER_FAILURE,
"Error adding sample to segment: Frame::Init failed");
}
if (write_duration) {
frame.set_duration(
BmffTimestampToNs(prev_sample_->duration(), time_scale_));
}
frame.set_is_key(prev_sample_->is_key_frame());
frame.set_timestamp(
BmffTimestampToNs(prev_sample_->pts(), time_scale_));
frame.set_track_number(track_id_);
if (prev_sample_->side_data_size() > 0) {
uint64_t block_add_id;
// First 8 bytes of side_data is the BlockAddID element's value, which is
// done to mimic ffmpeg behavior. See webm_cluster_parser.cc for details.
CHECK_GT(prev_sample_->side_data_size(), sizeof(block_add_id));
memcpy(&block_add_id, prev_sample_->side_data(), sizeof(block_add_id));
if (!frame.AddAdditionalData(
prev_sample_->side_data() + sizeof(block_add_id),
prev_sample_->side_data_size() - sizeof(block_add_id),
block_add_id)) {
return Status(
error::MUXER_FAILURE,
"Error adding sample to segment: Frame::AddAditionalData Failed");
}
}
if (!prev_sample_->is_key_frame() && !frame.CanBeSimpleBlock()) {
frame.set_reference_block_timestamp(
BmffTimestampToNs(reference_frame_timestamp_, time_scale_));
}
// GetRelativeTimecode will return -1 if the relative timecode is too large
// to fit in the frame.
if (cluster_->GetRelativeTimecode(NsToWebMTimecode(
frame.timestamp(), cluster_->timecode_scale())) < 0) {
const double segment_duration =
static_cast<double>(frame.timestamp() -
WebMTimecodeToNs(cluster_->timecode(),
cluster_->timecode_scale())) /
kSecondsToNs;
LOG(ERROR) << "Error adding sample to segment: segment too large, "
<< segment_duration
<< " seconds. Please check your GOP size and segment duration.";
return Status(error::MUXER_FAILURE,
"Error adding sample to segment: segment too large");
}
if (!cluster_->AddFrame(&frame)) {
return Status(error::MUXER_FAILURE,
"Error adding sample to segment: Cluster::AddFrame failed");
}
// A reference frame is needed for non-keyframes. Having a reference to the
// previous block is good enough.
// See libwebm Segment::AddGenericFrame
reference_frame_timestamp_ = prev_sample_->pts();
return Status::OK;
}
} // namespace webm
} // namespace media
} // namespace shaka