1 // Copyright 2014 The Chromium Authors
2 // Use of this source code is governed by a BSD-style license that can be
3 // found in the LICENSE file.
5 #ifndef MEDIA_FILTERS_FRAME_PROCESSOR_H_
6 #define MEDIA_FILTERS_FRAME_PROCESSOR_H_
11 #include "base/callback_forward.h"
12 #include "base/memory/raw_ptr.h"
13 #include "base/time/time.h"
14 #include "media/base/media_export.h"
15 #include "media/base/media_log.h"
16 #include "media/base/stream_parser.h"
17 #include "media/filters/chunk_demuxer.h"
18 #include "media/filters/source_buffer_parse_warnings.h"
24 // Helper class that implements Media Source Extension's coded frame processing
26 class MEDIA_EXPORT FrameProcessor {
28 using UpdateDurationCB = base::RepeatingCallback<void(base::TimeDelta)>;
30 FrameProcessor(UpdateDurationCB update_duration_cb, MediaLog* media_log);
32 FrameProcessor(const FrameProcessor&) = delete;
33 FrameProcessor& operator=(const FrameProcessor&) = delete;
37 // This must be called exactly once, before doing any track buffer creation or
39 void SetParseWarningCallback(SourceBufferParseWarningCB parse_warning_cb);
41 // Get/set the current append mode, which if true means "sequence" and if
42 // false means "segments".
43 // See http://www.w3.org/TR/media-source/#widl-SourceBuffer-mode.
44 bool sequence_mode() { return sequence_mode_; }
45 void SetSequenceMode(bool sequence_mode);
47 // Processes buffers in |buffer_queue_map|.
48 // Returns true on success or false on failure which indicates decode error.
49 // |append_window_start| and |append_window_end| correspond to the MSE spec's
50 // similarly named source buffer attributes that are used in coded frame
52 // Uses |*timestamp_offset| according to the coded frame processing algorithm,
53 // including updating it as required in 'sequence' mode frame processing.
54 bool ProcessFrames(const StreamParser::BufferQueueMap& buffer_queue_map,
55 base::TimeDelta append_window_start,
56 base::TimeDelta append_window_end,
57 base::TimeDelta* timestamp_offset);
59 // Signals the frame processor to update its group start timestamp to be
60 // |timestamp_offset| if it is in sequence append mode.
61 void SetGroupStartTimestampIfInSequenceMode(base::TimeDelta timestamp_offset);
63 // Adds a new track with unique track ID |id|.
64 // If |id| has previously been added, returns false to indicate error.
65 // Otherwise, returns true, indicating future ProcessFrames() will emit
66 // frames for the track |id| to |stream|.
67 bool AddTrack(StreamParser::TrackId id, ChunkDemuxerStream* stream);
69 // A map that describes how track ids changed between init segment. Maps the
70 // old track id for a new track id for the same track.
71 using TrackIdChanges = std::map<StreamParser::TrackId, StreamParser::TrackId>;
73 // Updates the internal mapping of TrackIds to track buffers. The input
74 // parameter |track_id_changes| maps old track ids to new ones. The track ids
75 // not present in the map must be assumed unchanged. Returns false if
77 bool UpdateTrackIds(const TrackIdChanges& track_id_changes);
79 // Sets the need random access point flag on all track buffers to true.
80 void SetAllTrackBuffersNeedRandomAccessPoint();
82 // Resets state for the coded frame processing algorithm as described in steps
83 // 2-5 of the MSE Reset Parser State algorithm described at
84 // http://www.w3.org/TR/media-source/#sourcebuffer-reset-parser-state
87 // Must be called when the audio config is updated. Used to manage when
88 // the preroll buffer is cleared and the allowed "fudge" factor between
90 void OnPossibleAudioConfigUpdate(const AudioDecoderConfig& config);
93 friend class FrameProcessorTest;
95 // If |track_buffers_| contains |id|, returns a pointer to the associated
96 // MseTrackBuffer. Otherwise, returns NULL.
97 MseTrackBuffer* FindTrack(StreamParser::TrackId id);
99 // Signals all track buffers' streams that a coded frame group is starting
100 // with |start_dts| and |start_pts|.
101 void NotifyStartOfCodedFrameGroup(DecodeTimestamp start_dts,
102 base::TimeDelta start_pts);
104 // Helper that signals each track buffer to append any processed, but not yet
105 // appended, frames to its stream. Returns true on success, or false if one or
106 // more of the appends failed.
107 bool FlushProcessedFrames();
109 // Handles partial append window trimming of |buffer|. Returns true if the
110 // given |buffer| can be partially trimmed or have preroll added; otherwise,
113 // If |buffer| overlaps |append_window_start|, the portion of |buffer| before
114 // |append_window_start| will be marked for post-decode discard. Further, if
115 // |audio_preroll_buffer_| exists and abuts |buffer|, it will be set as
116 // preroll on |buffer| and |audio_preroll_buffer_| will be cleared. If the
117 // preroll buffer does not abut |buffer|, it will be discarded unused.
119 // Likewise, if |buffer| overlaps |append_window_end|, the portion of |buffer|
120 // after |append_window_end| will be marked for post-decode discard.
122 // If |buffer| lies entirely before |append_window_start|, and thus would
123 // normally be discarded, |audio_preroll_buffer_| will be updated and the
124 // method will return false. In this case, the updated preroll will be
125 // |buffer| iff |buffer| is a keyframe, otherwise the preroll will be cleared.
126 bool HandlePartialAppendWindowTrimming(
127 base::TimeDelta append_window_start,
128 base::TimeDelta append_window_end,
129 scoped_refptr<StreamParserBuffer> buffer);
131 // Enables rejection of audio frame streams with nonkeyframe timestamps that
132 // do not monotonically increase since the last keyframe. Returns true if
133 // |frame| appears to be in order, false if |frame|'s order is not supported.
134 // |track_needs_random_access_point| should be the corresponding value for the
135 // frame's track buffer. This helper should only be called when
136 // |has_dependent_audio_frames_| is true, and only for an audio |frame|. This
137 // method also uses and updates
138 // |last_audio_pts_for_nonkeyframe_monotonicity_check_|.
139 bool CheckAudioPresentationOrder(const StreamParserBuffer& frame,
140 bool track_needs_random_access_point);
142 // Helper that processes one frame with the coded frame processing algorithm.
143 // Returns false on error or true on success.
144 bool ProcessFrame(scoped_refptr<StreamParserBuffer> frame,
145 base::TimeDelta append_window_start,
146 base::TimeDelta append_window_end,
147 base::TimeDelta* timestamp_offset);
149 // TrackId-indexed map of each track's stream.
150 using TrackBuffersMap =
151 std::map<StreamParser::TrackId, std::unique_ptr<MseTrackBuffer>>;
152 TrackBuffersMap track_buffers_;
154 // The last audio buffer seen by the frame processor that was removed because
155 // it was entirely before the start of the append window.
156 scoped_refptr<StreamParserBuffer> audio_preroll_buffer_;
158 // The AudioDecoderConfig associated with buffers handed to ProcessFrames().
159 // TODO(wolenetz): Associate current audio config and the derived
160 // |has_dependent_audio_frames_|, |sample_duration_| and
161 // |last_audio_pts_for_nonkeyframe_monotonicity_check_| with MseTrackBuffer
162 // instead to enable handling more than 1 audio track in a SourceBuffer
163 // simultaneously. See https://crbug.com/1081952.
164 AudioDecoderConfig current_audio_config_;
165 bool has_dependent_audio_frames_ = false;
166 base::TimeDelta sample_duration_;
168 // When |has_dependent_audio_frames_| is true, holds the PTS of the last
169 // successfully processed audio frame. If the next audio frame is not a
170 // keyframe and has lower PTS, the stream is invalid. Currently, the only
171 // supported audio streams that could contain nonkeyframes are in-order (PTS
172 // increases monotonically since last keyframe), e.g. xHE-AAC.
173 base::TimeDelta last_audio_pts_for_nonkeyframe_monotonicity_check_ =
176 // The AppendMode of the associated SourceBuffer.
177 // See SetSequenceMode() for interpretation of |sequence_mode_|.
178 // Per http://www.w3.org/TR/media-source/#widl-SourceBuffer-mode:
179 // Controls how a sequence of media segments are handled. This is initially
180 // set to false ("segments").
181 bool sequence_mode_ = false;
183 // Tracks whether or not we need to notify all track buffers of a new coded
184 // frame group (see https://w3c.github.io/media-source/#coded-frame-group)
185 // upon the next successfully processed frame. Set true initially and upon
186 // detection of DTS discontinuity, parser reset during 'segments' mode, or
187 // switching from 'sequence' to 'segments' mode. Individual track buffers can
188 // also be notified of an updated coded frame group start in edge cases. See
189 // further comments in ProcessFrame().
190 bool pending_notify_all_group_start_ = true;
192 // Tracks the MSE coded frame processing variable of same name.
193 // Initially kNoTimestamp, meaning "unset".
194 base::TimeDelta group_start_timestamp_;
196 // Tracks the MSE coded frame processing variable of same name. It stores the
197 // highest coded frame end timestamp across all coded frames in the current
198 // coded frame group. It is set to 0 when the SourceBuffer object is created
199 // and gets updated by ProcessFrames().
200 base::TimeDelta group_end_timestamp_;
202 const UpdateDurationCB update_duration_cb_;
204 // MediaLog for reporting messages and properties to debug content and engine.
205 raw_ptr<MediaLog> media_log_;
207 // Callback for reporting problematic conditions that are not necessarily
209 SourceBufferParseWarningCB parse_warning_cb_;
211 // Counters that limit spam to |media_log_| for frame processor warnings.
212 int num_dropped_preroll_warnings_ = 0;
213 int num_audio_non_keyframe_warnings_ = 0;
214 int num_muxed_sequence_mode_warnings_ = 0;
215 int num_skipped_empty_frame_warnings_ = 0;
216 int num_partial_discard_warnings_ = 0;
217 int num_dropped_frame_warnings_ = 0;
222 #endif // MEDIA_FILTERS_FRAME_PROCESSOR_H_