media/filters/ffmpeg_audio_decoder.cc

   1 // Copyright (c) 2012 The Chromium Authors. All rights reserved.
   2 // Use of this source code is governed by a BSD-style license that can be
   3 // found in the LICENSE file.
   4
   5 #include "media/filters/ffmpeg_audio_decoder.h"
   6
   7 #include "base/callback_helpers.h"
   8 #include "base/single_thread_task_runner.h"
   9 #include "media/base/audio_buffer.h"
  10 #include "media/base/audio_bus.h"
  11 #include "media/base/audio_decoder_config.h"
  12 #include "media/base/audio_discard_helper.h"
  13 #include "media/base/bind_to_current_loop.h"
  14 #include "media/base/decoder_buffer.h"
  15 #include "media/base/limits.h"
  16 #include "media/ffmpeg/ffmpeg_common.h"
  17 #include "media/filters/ffmpeg_glue.h"
  18
  19 namespace media {
  20
  21 // Returns true if the decode result was end of stream.
  22 static inline bool IsEndOfStream(int result,
  23                                  int decoded_size,
  24                                  const scoped_refptr<DecoderBuffer>& input) {
  25   // Three conditions to meet to declare end of stream for this decoder:
  26   // 1. FFmpeg didn't read anything.
  27   // 2. FFmpeg didn't output anything.
  28   // 3. An end of stream buffer is received.
  29   return result == 0 && decoded_size == 0 && input->end_of_stream();
  30 }
  31
  32 // Return the number of channels from the data in |frame|.
  33 static inline int DetermineChannels(AVFrame* frame) {
  34 #if defined(CHROMIUM_NO_AVFRAME_CHANNELS)
  35   // When use_system_ffmpeg==1, libav's AVFrame doesn't have channels field.
  36   return av_get_channel_layout_nb_channels(frame->channel_layout);
  37 #else
  38   return frame->channels;
  39 #endif
  40 }
  41
  42 // Called by FFmpeg's allocation routine to free a buffer. |opaque| is the
  43 // AudioBuffer allocated, so unref it.
  44 static void ReleaseAudioBufferImpl(void* opaque, uint8* data) {
  45   scoped_refptr<AudioBuffer> buffer;
  46   buffer.swap(reinterpret_cast<AudioBuffer**>(&opaque));
  47 }
  48
  49 // Called by FFmpeg's allocation routine to allocate a buffer. Uses
  50 // AVCodecContext.opaque to get the object reference in order to call
  51 // GetAudioBuffer() to do the actual allocation.
  52 static int GetAudioBuffer(struct AVCodecContext* s, AVFrame* frame, int flags) {
  53   DCHECK(s->codec->capabilities & CODEC_CAP_DR1);
  54   DCHECK_EQ(s->codec_type, AVMEDIA_TYPE_AUDIO);
  55
  56   // Since this routine is called by FFmpeg when a buffer is required for audio
  57   // data, use the values supplied by FFmpeg (ignoring the current settings).
  58   // FFmpegDecode() gets to determine if the buffer is useable or not.
  59   AVSampleFormat format = static_cast<AVSampleFormat>(frame->format);
  60   SampleFormat sample_format = AVSampleFormatToSampleFormat(format);
  61   int channels = DetermineChannels(frame);
  62   if (channels <= 0 || channels >= limits::kMaxChannels) {
  63     DLOG(ERROR) << "Requested number of channels (" << channels
  64                 << ") exceeds limit.";
  65     return AVERROR(EINVAL);
  66   }
  67
  68   int bytes_per_channel = SampleFormatToBytesPerChannel(sample_format);
  69   if (frame->nb_samples <= 0)
  70     return AVERROR(EINVAL);
  71
  72   if (s->channels != channels) {
  73     DLOG(ERROR) << "AVCodecContext and AVFrame disagree on channel count.";
  74     return AVERROR(EINVAL);
  75   }
  76
  77   // Determine how big the buffer should be and allocate it. FFmpeg may adjust
  78   // how big each channel data is in order to meet the alignment policy, so
  79   // we need to take this into consideration.
  80   int buffer_size_in_bytes =
  81       av_samples_get_buffer_size(&frame->linesize[0],
  82                                  channels,
  83                                  frame->nb_samples,
  84                                  format,
  85                                  AudioBuffer::kChannelAlignment);
  86   // Check for errors from av_samples_get_buffer_size().
  87   if (buffer_size_in_bytes < 0)
  88     return buffer_size_in_bytes;
  89   int frames_required = buffer_size_in_bytes / bytes_per_channel / channels;
  90   DCHECK_GE(frames_required, frame->nb_samples);
  91   scoped_refptr<AudioBuffer> buffer = AudioBuffer::CreateBuffer(
  92       sample_format,
  93       ChannelLayoutToChromeChannelLayout(s->channel_layout, s->channels),
  94       channels,
  95       s->sample_rate,
  96       frames_required);
  97
  98   // Initialize the data[] and extended_data[] fields to point into the memory
  99   // allocated for AudioBuffer. |number_of_planes| will be 1 for interleaved
 100   // audio and equal to |channels| for planar audio.
 101   int number_of_planes = buffer->channel_data().size();
 102   if (number_of_planes <= AV_NUM_DATA_POINTERS) {
 103     DCHECK_EQ(frame->extended_data, frame->data);
 104     for (int i = 0; i < number_of_planes; ++i)
 105       frame->data[i] = buffer->channel_data()[i];
 106   } else {
 107     // There are more channels than can fit into data[], so allocate
 108     // extended_data[] and fill appropriately.
 109     frame->extended_data = static_cast<uint8**>(
 110         av_malloc(number_of_planes * sizeof(*frame->extended_data)));
 111     int i = 0;
 112     for (; i < AV_NUM_DATA_POINTERS; ++i)
 113       frame->extended_data[i] = frame->data[i] = buffer->channel_data()[i];
 114     for (; i < number_of_planes; ++i)
 115       frame->extended_data[i] = buffer->channel_data()[i];
 116   }
 117
 118   // Now create an AVBufferRef for the data just allocated. It will own the
 119   // reference to the AudioBuffer object.
 120   void* opaque = NULL;
 121   buffer.swap(reinterpret_cast<AudioBuffer**>(&opaque));
 122   frame->buf[0] = av_buffer_create(
 123       frame->data[0], buffer_size_in_bytes, ReleaseAudioBufferImpl, opaque, 0);
 124   return 0;
 125 }
 126
 127 FFmpegAudioDecoder::FFmpegAudioDecoder(
 128     const scoped_refptr<base::SingleThreadTaskRunner>& task_runner,
 129     const scoped_refptr<MediaLog>& media_log)
 130     : task_runner_(task_runner),
 131       state_(kUninitialized),
 132       av_sample_format_(0),
 133       media_log_(media_log) {
 134 }
 135
 136 FFmpegAudioDecoder::~FFmpegAudioDecoder() {
 137   DCHECK(task_runner_->BelongsToCurrentThread());
 138
 139   if (state_ != kUninitialized)
 140     ReleaseFFmpegResources();
 141 }
 142
 143 std::string FFmpegAudioDecoder::GetDisplayName() const {
 144   return "FFmpegAudioDecoder";
 145 }
 146
 147 void FFmpegAudioDecoder::Initialize(const AudioDecoderConfig& config,
 148                                     const InitCB& init_cb,
 149                                     const OutputCB& output_cb) {
 150   DCHECK(task_runner_->BelongsToCurrentThread());
 151   DCHECK(!config.is_encrypted());
 152
 153   FFmpegGlue::InitializeFFmpeg();
 154
 155   config_ = config;
 156   InitCB bound_init_cb = BindToCurrentLoop(init_cb);
 157
 158   if (!config.IsValidConfig() || !ConfigureDecoder()) {
 159     bound_init_cb.Run(false);
 160     return;
 161   }
 162
 163   // Success!
 164   output_cb_ = BindToCurrentLoop(output_cb);
 165   state_ = kNormal;
 166   bound_init_cb.Run(true);
 167 }
 168
 169 void FFmpegAudioDecoder::Decode(const scoped_refptr<DecoderBuffer>& buffer,
 170                                 const DecodeCB& decode_cb) {
 171   DCHECK(task_runner_->BelongsToCurrentThread());
 172   DCHECK(!decode_cb.is_null());
 173   CHECK_NE(state_, kUninitialized);
 174   DecodeCB decode_cb_bound = BindToCurrentLoop(decode_cb);
 175
 176   if (state_ == kError) {
 177     decode_cb_bound.Run(kDecodeError);
 178     return;
 179   }
 180
 181   // Do nothing if decoding has finished.
 182   if (state_ == kDecodeFinished) {
 183     decode_cb_bound.Run(kOk);
 184     return;
 185   }
 186
 187   DecodeBuffer(buffer, decode_cb_bound);
 188 }
 189
 190 void FFmpegAudioDecoder::Reset(const base::Closure& closure) {
 191   DCHECK(task_runner_->BelongsToCurrentThread());
 192
 193   avcodec_flush_buffers(codec_context_.get());
 194   state_ = kNormal;
 195   ResetTimestampState();
 196   task_runner_->PostTask(FROM_HERE, closure);
 197 }
 198
 199 void FFmpegAudioDecoder::DecodeBuffer(
 200     const scoped_refptr<DecoderBuffer>& buffer,
 201     const DecodeCB& decode_cb) {
 202   DCHECK(task_runner_->BelongsToCurrentThread());
 203   DCHECK_NE(state_, kUninitialized);
 204   DCHECK_NE(state_, kDecodeFinished);
 205   DCHECK_NE(state_, kError);
 206   DCHECK(buffer.get());
 207
 208   // Make sure we are notified if http://crbug.com/49709 returns.  Issue also
 209   // occurs with some damaged files.
 210   if (!buffer->end_of_stream() && buffer->timestamp() == kNoTimestamp()) {
 211     DVLOG(1) << "Received a buffer without timestamps!";
 212     decode_cb.Run(kDecodeError);
 213     return;
 214   }
 215
 216   bool has_produced_frame;
 217   do {
 218     has_produced_frame = false;
 219     if (!FFmpegDecode(buffer, &has_produced_frame)) {
 220       state_ = kError;
 221       decode_cb.Run(kDecodeError);
 222       return;
 223     }
 224     // Repeat to flush the decoder after receiving EOS buffer.
 225   } while (buffer->end_of_stream() && has_produced_frame);
 226
 227   if (buffer->end_of_stream())
 228     state_ = kDecodeFinished;
 229
 230   decode_cb.Run(kOk);
 231 }
 232
 233 bool FFmpegAudioDecoder::FFmpegDecode(
 234     const scoped_refptr<DecoderBuffer>& buffer,
 235     bool* has_produced_frame) {
 236   DCHECK(!*has_produced_frame);
 237
 238   AVPacket packet;
 239   av_init_packet(&packet);
 240   if (buffer->end_of_stream()) {
 241     packet.data = NULL;
 242     packet.size = 0;
 243   } else {
 244     packet.data = const_cast<uint8*>(buffer->data());
 245     packet.size = buffer->data_size();
 246   }
 247
 248   // Each audio packet may contain several frames, so we must call the decoder
 249   // until we've exhausted the packet.  Regardless of the packet size we always
 250   // want to hand it to the decoder at least once, otherwise we would end up
 251   // skipping end of stream packets since they have a size of zero.
 252   do {
 253     int frame_decoded = 0;
 254     const int result = avcodec_decode_audio4(
 255         codec_context_.get(), av_frame_.get(), &frame_decoded, &packet);
 256
 257     if (result < 0) {
 258       DCHECK(!buffer->end_of_stream())
 259           << "End of stream buffer produced an error! "
 260           << "This is quite possibly a bug in the audio decoder not handling "
 261           << "end of stream AVPackets correctly.";
 262
 263       MEDIA_LOG(DEBUG, media_log_)
 264           << "Dropping audio frame which failed decode with timestamp: "
 265           << buffer->timestamp().InMicroseconds()
 266           << " us, duration: " << buffer->duration().InMicroseconds()
 267           << " us, packet size: " << buffer->data_size() << " bytes";
 268
 269       break;
 270     }
 271
 272     // Update packet size and data pointer in case we need to call the decoder
 273     // with the remaining bytes from this packet.
 274     packet.size -= result;
 275     packet.data += result;
 276
 277     scoped_refptr<AudioBuffer> output;
 278     const int channels = DetermineChannels(av_frame_.get());
 279     if (frame_decoded) {
 280       if (av_frame_->sample_rate != config_.samples_per_second() ||
 281           channels != ChannelLayoutToChannelCount(config_.channel_layout()) ||
 282           av_frame_->format != av_sample_format_) {
 283         DLOG(ERROR) << "Unsupported midstream configuration change!"
 284                     << " Sample Rate: " << av_frame_->sample_rate << " vs "
 285                     << config_.samples_per_second()
 286                     << ", Channels: " << channels << " vs "
 287                     << ChannelLayoutToChannelCount(config_.channel_layout())
 288                     << ", Sample Format: " << av_frame_->format << " vs "
 289                     << av_sample_format_;
 290
 291         if (config_.codec() == kCodecAAC &&
 292             av_frame_->sample_rate == 2 * config_.samples_per_second()) {
 293           MEDIA_LOG(DEBUG, media_log_)
 294               << "Implicit HE-AAC signalling is being"
 295               << " used. Please use mp4a.40.5 instead of"
 296               << " mp4a.40.2 in the mimetype.";
 297         }
 298         // This is an unrecoverable error, so bail out.
 299         av_frame_unref(av_frame_.get());
 300         return false;
 301       }
 302
 303       // Get the AudioBuffer that the data was decoded into. Adjust the number
 304       // of frames, in case fewer than requested were actually decoded.
 305       output = reinterpret_cast<AudioBuffer*>(
 306           av_buffer_get_opaque(av_frame_->buf[0]));
 307
 308       DCHECK_EQ(ChannelLayoutToChannelCount(config_.channel_layout()),
 309                 output->channel_count());
 310       const int unread_frames = output->frame_count() - av_frame_->nb_samples;
 311       DCHECK_GE(unread_frames, 0);
 312       if (unread_frames > 0)
 313         output->TrimEnd(unread_frames);
 314       av_frame_unref(av_frame_.get());
 315     }
 316
 317     // WARNING: |av_frame_| no longer has valid data at this point.
 318     const int decoded_frames = frame_decoded ? output->frame_count() : 0;
 319     if (IsEndOfStream(result, decoded_frames, buffer)) {
 320       DCHECK_EQ(packet.size, 0);
 321     } else if (discard_helper_->ProcessBuffers(buffer, output)) {
 322       *has_produced_frame = true;
 323       output_cb_.Run(output);
 324     }
 325   } while (packet.size > 0);
 326
 327   return true;
 328 }
 329
 330 void FFmpegAudioDecoder::ReleaseFFmpegResources() {
 331   codec_context_.reset();
 332   av_frame_.reset();
 333 }
 334
 335 bool FFmpegAudioDecoder::ConfigureDecoder() {
 336   if (!config_.IsValidConfig()) {
 337     DLOG(ERROR) << "Invalid audio stream -"
 338                 << " codec: " << config_.codec()
 339                 << " channel layout: " << config_.channel_layout()
 340                 << " bits per channel: " << config_.bits_per_channel()
 341                 << " samples per second: " << config_.samples_per_second();
 342     return false;
 343   }
 344
 345   if (config_.is_encrypted()) {
 346     DLOG(ERROR) << "Encrypted audio stream not supported";
 347     return false;
 348   }
 349
 350   // Release existing decoder resources if necessary.
 351   ReleaseFFmpegResources();
 352
 353   // Initialize AVCodecContext structure.
 354   codec_context_.reset(avcodec_alloc_context3(NULL));
 355   AudioDecoderConfigToAVCodecContext(config_, codec_context_.get());
 356
 357   codec_context_->opaque = this;
 358   codec_context_->get_buffer2 = GetAudioBuffer;
 359   codec_context_->refcounted_frames = 1;
 360
 361   AVCodec* codec = avcodec_find_decoder(codec_context_->codec_id);
 362   if (!codec || avcodec_open2(codec_context_.get(), codec, NULL) < 0) {
 363     DLOG(ERROR) << "Could not initialize audio decoder: "
 364                 << codec_context_->codec_id;
 365     ReleaseFFmpegResources();
 366     state_ = kUninitialized;
 367     return false;
 368   }
 369
 370   // Success!
 371   av_frame_.reset(av_frame_alloc());
 372   av_sample_format_ = codec_context_->sample_fmt;
 373
 374   if (codec_context_->channels !=
 375       ChannelLayoutToChannelCount(config_.channel_layout())) {
 376     DLOG(ERROR) << "Audio configuration specified "
 377                 << ChannelLayoutToChannelCount(config_.channel_layout())
 378                 << " channels, but FFmpeg thinks the file contains "
 379                 << codec_context_->channels << " channels";
 380     ReleaseFFmpegResources();
 381     state_ = kUninitialized;
 382     return false;
 383   }
 384
 385   ResetTimestampState();
 386   return true;
 387 }
 388
 389 void FFmpegAudioDecoder::ResetTimestampState() {
 390   discard_helper_.reset(new AudioDiscardHelper(config_.samples_per_second(),
 391                                                config_.codec_delay()));
 392   discard_helper_->Reset(config_.codec_delay());
 393 }
 394
 395 }  // namespace media