Rewriting FFmpegAudioDecoder and eliminating DecoderBase.

media/filters/ffmpeg_audio_decoder.cc

Index: media/filters/ffmpeg_audio_decoder.cc
diff --git a/media/filters/ffmpeg_audio_decoder.cc b/media/filters/ffmpeg_audio_decoder.cc
index 50335d3829b77adf5d1cf852bb76fedbbae9d0e5..cb3f846078329514071f4791569be5aa59b7b128 100644
--- a/media/filters/ffmpeg_audio_decoder.cc
+++ b/media/filters/ffmpeg_audio_decoder.cc
@@ -4,50 +4,111 @@
 
 #include "media/filters/ffmpeg_audio_decoder.h"
 
+#include "base/bind.h"
 #include "media/base/data_buffer.h"
+#include "media/base/demuxer.h"
+#include "media/base/filter_host.h"
 #include "media/base/limits.h"
 #include "media/ffmpeg/ffmpeg_common.h"
-#include "media/filters/ffmpeg_demuxer.h"
-
-#if !defined(USE_SSE)
-#if defined(__SSE__) || defined(ARCH_CPU_X86_64) || _M_IX86_FP==1
-#define USE_SSE 1
-#else
-#define USE_SSE 0
-#endif
-#endif
-#if USE_SSE
-#include <xmmintrin.h>
-#endif
 
 namespace media {
 
-// Size of the decoded audio buffer.
-const size_t FFmpegAudioDecoder::kOutputBufferSize =
-    AVCODEC_MAX_AUDIO_FRAME_SIZE;
+// Returns true if the decode result was an error.
+static bool IsErrorResult(int result, int decoded_size) {
+  return result < 0 ||
+      decoded_size < 0 ||
+      decoded_size > AVCODEC_MAX_AUDIO_FRAME_SIZE;
+}
+
+// Returns true if the decode result produced audio samples.
+static bool ProducedAudioSamples(int decoded_size) {
+  return decoded_size > 0;
+}
+
+// Returns true if the decode result was a timestamp packet and not actual audio
+// data.
+static bool IsTimestampMarkerPacket(int result, Buffer* input) {
+  // We can get a positive result but no decoded data.  This is ok because this
+  // this can be a marker packet that only contains timestamp.
+  return result > 0 && !input->IsEndOfStream() &&
+      input->GetTimestamp() != kNoTimestamp &&
+      input->GetDuration() != kNoTimestamp;
+}
+
+// Returns true if the decode result was end of stream.
+static bool IsEndOfStream(int result, int decoded_size, Buffer* input) {
+  // Three conditions to meet to declare end of stream for this decoder:
+  // 1. FFmpeg didn't read anything.
+  // 2. FFmpeg didn't output anything.
+  // 3. An end of stream buffer is received.
+  return result == 0 && decoded_size == 0 && input->IsEndOfStream();
+}
+
 
 FFmpegAudioDecoder::FFmpegAudioDecoder(MessageLoop* message_loop)
-    : DecoderBase<AudioDecoder, Buffer>(message_loop),
+    : message_loop_(message_loop),
       codec_context_(NULL),
       bits_per_channel_(0),
       channel_layout_(CHANNEL_LAYOUT_NONE),
       sample_rate_(0),
-      estimated_next_timestamp_(kNoTimestamp) {
+      decoded_audio_size_(AVCODEC_MAX_AUDIO_FRAME_SIZE),
+      decoded_audio_(static_cast<uint8*>(av_malloc(decoded_audio_size_))),
+      pending_reads_(0) {
 }
 
 FFmpegAudioDecoder::~FFmpegAudioDecoder() {
+  av_free(decoded_audio_);
 }
 
-void FFmpegAudioDecoder::DoInitialize(DemuxerStream* demuxer_stream,
-                                      bool* success,
-                                      Task* done_cb) {
-  base::ScopedTaskRunner done_runner(done_cb);
-  *success = false;
+void FFmpegAudioDecoder::Flush(FilterCallback* callback) {
+  message_loop_->PostTask(
+      FROM_HERE,
+      NewRunnableMethod(this, &FFmpegAudioDecoder::DoFlush, callback));
+}
 
-  AVStream* av_stream = demuxer_stream->GetAVStream();
-  if (!av_stream) {
-    return;
-  }
+void FFmpegAudioDecoder::Initialize(
+    DemuxerStream* stream,
+    FilterCallback* callback,
+    StatisticsCallback* stats_callback) {
+  // TODO(scherkus): change Initialize() signature to pass |stream| as a
+  // scoped_refptr<>.
+  scoped_refptr<DemuxerStream> ref_stream(stream);
+  message_loop_->PostTask(
+      FROM_HERE,
+      NewRunnableMethod(this, &FFmpegAudioDecoder::DoInitialize,
+                        ref_stream, callback, stats_callback));
+}
+
+void FFmpegAudioDecoder::ProduceAudioSamples(scoped_refptr<Buffer> buffer) {
+  message_loop_->PostTask(
+      FROM_HERE,
+      NewRunnableMethod(this, &FFmpegAudioDecoder::DoProduceAudioSamples,
+                        buffer));
+}
+
+int FFmpegAudioDecoder::bits_per_channel() {
+  return bits_per_channel_;
+}
+
+ChannelLayout FFmpegAudioDecoder::channel_layout() {
+  return channel_layout_;
+}
+
+int FFmpegAudioDecoder::sample_rate() {
+  return sample_rate_;
+}
+
+void FFmpegAudioDecoder::DoInitialize(
+    const scoped_refptr<DemuxerStream>& stream,
+    FilterCallback* callback,
+    StatisticsCallback* stats_callback) {
+  scoped_ptr<FilterCallback> c(callback);
+
+  demuxer_stream_ = stream;
+  AVStream* av_stream = demuxer_stream_->GetAVStream();
+  CHECK(av_stream);
+
+  stats_callback_.reset(stats_callback);
 
   // Grab the AVStream's codec context and make sure we have sensible values.
   codec_context_ = av_stream->codec;
@@ -58,213 +119,177 @@ void FFmpegAudioDecoder::DoInitialize(DemuxerStream* demuxer_stream,
       bps <= 0 || bps > Limits::kMaxBitsPerSample ||
       codec_context_->sample_rate <= 0 ||
       codec_context_->sample_rate > Limits::kMaxSampleRate) {
-    DLOG(WARNING) << "Invalid audio stream -"
-        << " channels: " << codec_context_->channels
-        << " channel layout:" << codec_context_->channel_layout
-        << " bps: " << bps
-        << " sample rate: " << codec_context_->sample_rate;
+    DLOG(ERROR) << "Invalid audio stream -"
+                << " channels: " << codec_context_->channels
+                << " channel layout:" << codec_context_->channel_layout
+                << " bps: " << bps
+                << " sample rate: " << codec_context_->sample_rate;
+
+    host()->SetError(PIPELINE_ERROR_DECODE);
+    callback->Run();
     return;
   }
 
   // Serialize calls to avcodec_open().
   AVCodec* codec = avcodec_find_decoder(codec_context_->codec_id);
   if (!codec || avcodec_open(codec_context_, codec) < 0) {
+    DLOG(ERROR) << "Could not initialize audio decoder: "
+                << codec_context_->codec_id;
+
+    host()->SetError(PIPELINE_ERROR_DECODE);
+    callback->Run();
     return;
   }
 
+  // Success!
   bits_per_channel_ = av_get_bits_per_sample_fmt(codec_context_->sample_fmt);
   channel_layout_ =
       ChannelLayoutToChromeChannelLayout(codec_context_->channel_layout,
                                          codec_context_->channels);
   sample_rate_ = codec_context_->sample_rate;
 
-  // Prepare the output buffer.
-  output_buffer_.reset(static_cast<uint8*>(av_malloc(kOutputBufferSize)));
-  if (!output_buffer_.get()) {
-    host()->SetError(PIPELINE_ERROR_OUT_OF_MEMORY);
-    return;
-  }
-  *success = true;
-}
-
-void FFmpegAudioDecoder::ProduceAudioSamples(scoped_refptr<Buffer> output) {
-  DecoderBase<AudioDecoder, Buffer>::PostReadTaskHack(output);
+  callback->Run();
 }
 
-int FFmpegAudioDecoder::bits_per_channel() {
-  return bits_per_channel_;
-}
-
-ChannelLayout FFmpegAudioDecoder::channel_layout() {
-  return channel_layout_;
-}
-
-int FFmpegAudioDecoder::sample_rate() {
-  return sample_rate_;
-}
-
-void FFmpegAudioDecoder::DoSeek(base::TimeDelta time, Task* done_cb) {
+void FFmpegAudioDecoder::DoFlush(FilterCallback* callback) {
   avcodec_flush_buffers(codec_context_);
   estimated_next_timestamp_ = kNoTimestamp;
-  done_cb->Run();
-  delete done_cb;
-}
 
-// ConvertAudioF32ToS32() converts float audio (F32) to int (S32) in place.
-// This is a temporary solution.
-// The purpose of this short term fix is to enable WMApro, which decodes to
-// float.
-// The audio driver has been tested by passing the float audio thru.
-// FFmpeg for ChromeOS only exposes U8, S16 and F32.
-// To properly expose new audio sample types at the audio driver layer, a enum
-// should be created to represent all suppported types, including types
-// for Pepper.  FFmpeg should be queried for type and passed along.
-
-// TODO(fbarchard): Remove this function.  Expose all FFmpeg types to driver.
-// TODO(fbarchard): If this function is kept, move it to audio_util.cc
-
-#if USE_SSE
-const __m128 kFloatScaler = _mm_set1_ps( 2147483648.0f );
-static void FloatToIntSaturate(float* p) {
-  __m128 a = _mm_set1_ps(*p);
-  a = _mm_mul_ss(a, kFloatScaler);
-  *reinterpret_cast<int32*>(p) = _mm_cvtss_si32(a);
+  callback->Run();
+  delete callback;
 }
-#else
-const float kFloatScaler = 2147483648.0f;
-const int kMinSample = std::numeric_limits<int32>::min();
-const int kMaxSample = std::numeric_limits<int32>::max();
-const float kMinSampleFloat =
-    static_cast<float>(std::numeric_limits<int32>::min());
-const float kMaxSampleFloat =
-    static_cast<float>(std::numeric_limits<int32>::max());
-static void FloatToIntSaturate(float* p) {
-  float f = *p * kFloatScaler + 0.5f;
-  int sample;
-  if (f <= kMinSampleFloat) {
-    sample = kMinSample;
-  } else if (f >= kMaxSampleFloat) {
-    sample = kMaxSample;
-  } else {
-    sample = static_cast<int32>(f);
-  }
-  *reinterpret_cast<int32*>(p) = sample;
-}
-#endif
-static void ConvertAudioF32ToS32(void* buffer, int buffer_size) {
-  for (int i = 0; i < buffer_size / 4; ++i) {
-    FloatToIntSaturate(reinterpret_cast<float*>(buffer) + i);
-  }
+
+void FFmpegAudioDecoder::DoProduceAudioSamples(
+    const scoped_refptr<Buffer>& output) {
+  output_buffers_.push_back(output);
+  ReadFromDemuxerStream();
 }
 
-void FFmpegAudioDecoder::DoDecode(Buffer* input) {
-  PipelineStatistics statistics;
+void FFmpegAudioDecoder::DoDecodeBuffer(const scoped_refptr<Buffer>& input) {
+  DCHECK(!output_buffers_.empty());
+  DCHECK_GT(pending_reads_, 0);
+  pending_reads_--;
 
   // FFmpeg tends to seek Ogg audio streams in the middle of nowhere, giving us
   // a whole bunch of AV_NOPTS_VALUE packets.  Discard them until we find
   // something valid.  Refer to http://crbug.com/49709
-  // TODO(hclam): remove this once fixing the issue in FFmpeg.
   if (input->GetTimestamp() == kNoTimestamp &&
       estimated_next_timestamp_ == kNoTimestamp &&
       !input->IsEndOfStream()) {
-    DecoderBase<AudioDecoder, Buffer>::OnDecodeComplete(statistics);
+    ReadFromDemuxerStream();
     return;
   }
 
-  // Due to FFmpeg API changes we no longer have const read-only pointers.
   AVPacket packet;
   av_init_packet(&packet);
-  packet.data = const_cast<uint8*>(input->GetData());
-  packet.size = input->GetDataSize();
+  if (input->IsEndOfStream()) {
+    packet.data = NULL;
+    packet.size = 0;
+  } else {
+    packet.data = const_cast<uint8*>(input->GetData());
+    packet.size = input->GetDataSize();
+  }
 
+  PipelineStatistics statistics;
   statistics.audio_bytes_decoded = input->GetDataSize();
 
-  int16_t* output_buffer = reinterpret_cast<int16_t*>(output_buffer_.get());
-  int output_buffer_size = kOutputBufferSize;
-  int result = avcodec_decode_audio3(codec_context_,
-                                     output_buffer,
-                                     &output_buffer_size,
-                                     &packet);
+  int decoded_audio_size = decoded_audio_size_;
+  int result = avcodec_decode_audio3(
+      codec_context_, reinterpret_cast<int16_t*>(decoded_audio_),
+      &decoded_audio_size, &packet);
 
-  if (codec_context_->sample_fmt == SAMPLE_FMT_FLT) {
-    ConvertAudioF32ToS32(output_buffer, output_buffer_size);
-  }
+  if (IsErrorResult(result, decoded_audio_size)) {
+    DCHECK(!input->IsEndOfStream())
+        << "End of stream buffer produced an error! "
+        << "This is quite possibly a bug in the audio decoder not handling "
+        << "end of stream AVPackets correctly.";
 
-  // TODO(ajwong): Consider if kOutputBufferSize should just be an int instead
-  // of a size_t.
-  if (result < 0 ||
-      output_buffer_size < 0 ||
-      static_cast<size_t>(output_buffer_size) > kOutputBufferSize) {
-    VLOG(1) << "Error decoding an audio frame with timestamp: "
-            << input->GetTimestamp().InMicroseconds() << " us, duration: "
-            << input->GetDuration().InMicroseconds() << " us, packet size: "
-            << input->GetDataSize() << " bytes";
-    DecoderBase<AudioDecoder, Buffer>::OnDecodeComplete(statistics);
-    return;
-  }
+    DLOG(ERROR) << "Error decoding an audio frame with timestamp: "
+                << input->GetTimestamp().InMicroseconds() << " us, duration: "
+                << input->GetDuration().InMicroseconds() << " us, packet size: "
+                << input->GetDataSize() << " bytes";
 
-  // If we have decoded something, enqueue the result.
-  if (output_buffer_size) {
-    DataBuffer* result_buffer = new DataBuffer(output_buffer_size);
-    result_buffer->SetDataSize(output_buffer_size);
-    uint8* data = result_buffer->GetWritableData();
-    memcpy(data, output_buffer, output_buffer_size);
-
-    // We don't trust the demuxer, so always calculate the duration based on
-    // the actual number of samples decoded.
-    base::TimeDelta duration = CalculateDuration(output_buffer_size);
-    result_buffer->SetDuration(duration);
-
-    // Use an estimated timestamp unless the incoming buffer has a valid one.
-    if (input->GetTimestamp() == kNoTimestamp) {
-      result_buffer->SetTimestamp(estimated_next_timestamp_);
-
-      // Keep the estimated timestamp invalid until we get an incoming buffer
-      // with a valid timestamp.  This can happen during seeks, etc...
-      if (estimated_next_timestamp_ != kNoTimestamp) {
-        estimated_next_timestamp_ += duration;
-      }
-    } else {
-      result_buffer->SetTimestamp(input->GetTimestamp());
-      estimated_next_timestamp_ = input->GetTimestamp() + duration;
-    }
-
-    EnqueueResult(result_buffer);
-    DecoderBase<AudioDecoder, Buffer>::OnDecodeComplete(statistics);
+    ReadFromDemuxerStream();
     return;
   }
 
-  // We can get a positive result but no decoded data.  This is ok because this
-  // this can be a marker packet that only contains timestamp.  In this case we
-  // save the timestamp for later use.
-  if (result && !input->IsEndOfStream() &&
-      input->GetTimestamp() != kNoTimestamp &&
-      input->GetDuration() != kNoTimestamp) {
+  scoped_refptr<DataBuffer> output;
+
+  if (ProducedAudioSamples(decoded_audio_size)) {
+    // Copy the audio samples into an output buffer.
+    output = new DataBuffer(decoded_audio_size);
+    output->SetDataSize(decoded_audio_size);
+    uint8* data = output->GetWritableData();
+    memcpy(data, decoded_audio_, decoded_audio_size);
+
+    UpdateDurationAndTimestamp(input, output);
+  } else if (IsTimestampMarkerPacket(result, input)) {
+    // Nothing else to do here but update our estimation.
     estimated_next_timestamp_ = input->GetTimestamp() + input->GetDuration();
-    DecoderBase<AudioDecoder, Buffer>::OnDecodeComplete(statistics);
+  } else if (IsEndOfStream(result, decoded_audio_size, input)) {
+    // Create an end of stream output buffer.
+    output = new DataBuffer(0);
+    output->SetTimestamp(input->GetTimestamp());
+    output->SetDuration(input->GetDuration());
+  }
+
+  // Decoding finished successfully, update stats and execute callback.
+  stats_callback_->Run(statistics);
+  if (output) {
+    DCHECK_GT(output_buffers_.size(), 0u);
+    output_buffers_.pop_front();
+
+    ConsumeAudioSamples(output);
+  } else {
+    ReadFromDemuxerStream();
+  }
+}
+
+void FFmpegAudioDecoder::ReadFromDemuxerStream() {
+  DCHECK(!output_buffers_.empty())
+      << "Reads should only occur if there are output buffers.";
+
+  pending_reads_++;
+  demuxer_stream_->Read(base::Bind(&FFmpegAudioDecoder::DecodeBuffer, this));
+}
+
+void FFmpegAudioDecoder::DecodeBuffer(Buffer* buffer) {
+  // TODO(scherkus): change DemuxerStream::Read() to use scoped_refptr<> for
+  // callback.
+  scoped_refptr<Buffer> ref_buffer(buffer);
+  message_loop_->PostTask(
+      FROM_HERE,
+      NewRunnableMethod(this, &FFmpegAudioDecoder::DoDecodeBuffer, ref_buffer));
+}
+
+void FFmpegAudioDecoder::UpdateDurationAndTimestamp(
+    const Buffer* input,
+    DataBuffer* output) {
+  // Always calculate duration based on the actual number of samples decoded.
+  base::TimeDelta duration = CalculateDuration(output->GetDataSize());
+  output->SetDuration(duration);
+
+  // Use the incoming timestamp if it's valid.
+  if (input->GetTimestamp() != kNoTimestamp) {
+    output->SetTimestamp(input->GetTimestamp());
+    estimated_next_timestamp_ = input->GetTimestamp() + duration;
     return;
   }
 
-  // Three conditions to meet to declare end of stream for this decoder:
-  // 1. FFmpeg didn't read anything.
-  // 2. FFmpeg didn't output anything.
-  // 3. An end of stream buffer is received.
-  if (result == 0 && output_buffer_size == 0 && input->IsEndOfStream()) {
-    DataBuffer* result_buffer = new DataBuffer(0);
-    result_buffer->SetTimestamp(input->GetTimestamp());
-    result_buffer->SetDuration(input->GetDuration());
-    EnqueueResult(result_buffer);
+  // Otherwise use an estimated timestamp and attempt to update the estimation
+  // as long as it's valid.
+  output->SetTimestamp(estimated_next_timestamp_);
+  if (estimated_next_timestamp_ != kNoTimestamp) {
+    estimated_next_timestamp_ += duration;
   }
-  DecoderBase<AudioDecoder, Buffer>::OnDecodeComplete(statistics);
 }
 
-base::TimeDelta FFmpegAudioDecoder::CalculateDuration(size_t size) {
-  int64 denominator = codec_context_->channels *
-      av_get_bits_per_sample_fmt(codec_context_->sample_fmt) / 8 *
-      codec_context_->sample_rate;
+base::TimeDelta FFmpegAudioDecoder::CalculateDuration(int size) {
+  int64 denominator = ChannelLayoutToChannelCount(channel_layout_) *
+      bits_per_channel_ / 8 * sample_rate_;
   double microseconds = size /
       (denominator / static_cast<double>(base::Time::kMicrosecondsPerSecond));
   return base::TimeDelta::FromMicroseconds(static_cast<int64>(microseconds));
 }
 
-}  // namespace
+}  // namespace media