src/media/playback/mediaplayer/ffmpeg/ffmpeg_audio_decoder.cc

// Copyright 2016 The Fuchsia Authors. All rights reserved.
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file.

#include "src/media/playback/mediaplayer/ffmpeg/ffmpeg_audio_decoder.h"

#include "lib/media/cpp/timeline_rate.h"
#include "src/lib/syslog/cpp/logger.h"

namespace media_player {

// static
std::shared_ptr<Processor> FfmpegAudioDecoder::Create(AvCodecContextPtr av_codec_context) {
  return std::make_shared<FfmpegAudioDecoder>(std::move(av_codec_context));
}

FfmpegAudioDecoder::FfmpegAudioDecoder(AvCodecContextPtr av_codec_context)
    : FfmpegDecoderBase(std::move(av_codec_context)) {
  FX_DCHECK(context());
  FX_DCHECK(context()->channels > 0);

  std::unique_ptr<StreamType> stream_type = output_stream_type();
  FX_DCHECK(stream_type);
  FX_DCHECK(stream_type->audio());
  set_pts_rate(media::TimelineRate(stream_type->audio()->frames_per_second(), 1));

  stream_type_ = std::move(stream_type);

  if (av_sample_fmt_is_planar(context()->sample_fmt)) {
    // Prepare for interleaving.
    lpcm_util_ = LpcmUtil::Create(*stream_type_->audio());
    copy_or_interleave_ = true;
  }

  // Codec-specific code goes here.
  switch (context()->codec_id) {
    case AV_CODEC_ID_OPUS:
      // The opus decoder allocates buffers six times as large as the resulting payload. We need
      // to copy the payloads, so we don't use up all the buffer space available to the renderer.
      copy_or_interleave_ = true;
      break;
    default:
      break;
  }
}

FfmpegAudioDecoder::~FfmpegAudioDecoder() {}

void FfmpegAudioDecoder::ConfigureConnectors() {
  ConfigureInputToUseLocalMemory(0, 2);
  // TODO(dalesat): Real numbers here. How big are packets?
  // We're OK for now, because the audio renderer asks for a single VMO that's
  // big enough to handle any packet we want to produce.
  ConfigureOutputToUseLocalMemory(0, 1, 1);
}

void FfmpegAudioDecoder::OnNewInputPacket(const PacketPtr& packet) {
  if (next_pts() == Packet::kNoPts) {
    set_next_pts(packet->GetPts(pts_rate()));
  }

  context()->reordered_opaque = packet->discontinuity() ? 1 : 0;
  context()->pkt_timebase.num = pts_rate().reference_delta();
  context()->pkt_timebase.den = pts_rate().subject_delta();
}

int FfmpegAudioDecoder::BuildAVFrame(const AVCodecContext& av_codec_context, AVFrame* av_frame) {
  FX_DCHECK(av_frame);

  AVSampleFormat av_sample_format = static_cast<AVSampleFormat>(av_frame->format);

  int buffer_size = av_samples_get_buffer_size(&av_frame->linesize[0], av_codec_context.channels,
                                               av_frame->nb_samples, av_sample_format,
                                               FfmpegAudioDecoder::kChannelAlign);
  if (buffer_size < 0) {
    FX_LOGS(WARNING) << "av_samples_get_buffer_size failed";
    return buffer_size;
  }

  // Get the right payload buffer. If we need to copy or interleave later, we just get
  // a buffer allocated using malloc. If not, we ask the stage for a buffer.
  fbl::RefPtr<PayloadBuffer> buffer = copy_or_interleave_
                                          ? PayloadBuffer::CreateWithMalloc(buffer_size)
                                          : AllocatePayloadBuffer(buffer_size);

  if (!buffer) {
    // TODO(dalesat): Record/report packet drop.
    return -1;
  }

  // Check that the allocator has met the common alignment requirements and
  // that those requirements are good enough for the decoder.
  FX_DCHECK(PayloadBuffer::IsAligned(buffer->data()));
  FX_DCHECK(PayloadBuffer::kByteAlignment >= kChannelAlign);

  if (!av_sample_fmt_is_planar(av_sample_format)) {
    // Samples are interleaved. There's just one buffer.
    av_frame->data[0] = reinterpret_cast<uint8_t*>(buffer->data());
  } else {
    // Samples are not interleaved. There's one buffer per channel.
    int channels = av_codec_context.channels;
    int bytes_per_channel = buffer_size / channels;
    uint8_t* channel_buffer = reinterpret_cast<uint8_t*>(buffer->data());

    FX_DCHECK(buffer || bytes_per_channel == 0);

    if (channels <= AV_NUM_DATA_POINTERS) {
      // The buffer pointers will fit in av_frame->data.
      FX_DCHECK(av_frame->extended_data == av_frame->data);
      for (int channel = 0; channel < channels; ++channel) {
        av_frame->data[channel] = channel_buffer;
        channel_buffer += bytes_per_channel;
      }
    } else {
      // Too many channels for av_frame->data. We have to use
      // av_frame->extended_data
      av_frame->extended_data =
          static_cast<uint8_t**>(av_malloc(channels * sizeof(*av_frame->extended_data)));

      // The first AV_NUM_DATA_POINTERS go in both data and extended_data.
      int channel = 0;
      for (; channel < AV_NUM_DATA_POINTERS; ++channel) {
        av_frame->extended_data[channel] = av_frame->data[channel] = channel_buffer;
        channel_buffer += bytes_per_channel;
      }

      // The rest go only in extended_data.
      for (; channel < channels; ++channel) {
        av_frame->extended_data[channel] = channel_buffer;
        channel_buffer += bytes_per_channel;
      }
    }
  }

  av_frame->buf[0] = CreateAVBuffer(std::move(buffer));
  av_frame->reordered_opaque = av_codec_context.reordered_opaque;

  return 0;
}

PacketPtr FfmpegAudioDecoder::CreateOutputPacket(const AVFrame& av_frame,
                                                 fbl::RefPtr<PayloadBuffer> payload_buffer) {
  FX_DCHECK(av_frame.buf[0]);
  FX_DCHECK(payload_buffer);

  // We infer the PTS for a packet based on the assumption that the decoder
  // produces an uninterrupted stream of frames. The PTS value in av_frame is
  // often bogus, and we get bad results if we try to use it. This approach is
  // consistent with the way Chromium deals with the ffmpeg audio decoders.
  int64_t pts = next_pts();

  if (pts != Packet::kNoPts) {
    set_next_pts(pts + av_frame.nb_samples);
  }

  FX_DCHECK(stream_type_);
  FX_DCHECK(stream_type_->audio());

  uint64_t payload_size = stream_type_->audio()->min_buffer_size(av_frame.nb_samples);

  if (copy_or_interleave_) {
    // We need to copy or interleave. The original frames are in
    // |payload_buffer|, which was allocated from system memory. That buffer
    // will get released later in ReleaseBufferForAvFrame. We need a new
    // buffer for the output payload, which we get from the stage.
    auto new_payload_buffer = AllocatePayloadBuffer(payload_size);
    if (!new_payload_buffer) {
      // TODO(dalesat): Record/report packet drop.
      return nullptr;
    }

    if (lpcm_util_) {
      lpcm_util_->Interleave(payload_buffer->data(),
                             av_frame.linesize[0] * stream_type_->audio()->channels(),
                             new_payload_buffer->data(), av_frame.nb_samples);
    } else {
      memcpy(new_payload_buffer->data(), payload_buffer->data(), payload_size);
    }

    // |new_payload_buffer| is the buffer we want to attach to the |Packet|.
    // This assignment drops the reference to the original |payload_buffer|, so
    // it will be recycled once the |AVBuffer| is released.
    payload_buffer = std::move(new_payload_buffer);
  }

  // Create the output packet. We set the discontinuity bit on the packet if
  // the corresponding input packet had one.
  return Packet::Create(pts, pts_rate(),
                        false,                           // Not a keyframe
                        av_frame.reordered_opaque != 0,  // discontinuity
                        false,  // Not end-of-stream. The base class handles end-of-stream.
                        payload_size, std::move(payload_buffer));
}

const char* FfmpegAudioDecoder::label() const { return "audio_decoder"; }

}  // namespace media_player