Switch to pcm_low_latency and enable resampling.

media/base/audio_renderer_mixer.cc

Index: media/base/audio_renderer_mixer.cc
diff --git a/media/base/audio_renderer_mixer.cc b/media/base/audio_renderer_mixer.cc
index 5b89aa41daebe06c3e5fcd40b1105db64a6f3fb8..960c23577814822bf3118630cf31995f695c186f 100644
--- a/media/base/audio_renderer_mixer.cc
+++ b/media/base/audio_renderer_mixer.cc
@@ -4,17 +4,51 @@
 
 #include "media/base/audio_renderer_mixer.h"
 
+#include "base/bind.h"
+#include "base/bind_helpers.h"
 #include "base/logging.h"
+#include "media/audio/audio_util.h"
+#include "media/base/limits.h"
 
 namespace media {
 
 AudioRendererMixer::AudioRendererMixer(
+    const AudioParameters& input_params, const AudioParameters& output_params,
+    const scoped_refptr<AudioRendererSink>& sink)
+    : audio_sink_(sink),
+      current_audio_delay_milliseconds_(0) {
+  // Sanity check sample rates.
+  DCHECK_LE(input_params.sample_rate(), limits::kMaxSampleRate);
+  DCHECK_GE(input_params.sample_rate(), limits::kMinSampleRate);
+  DCHECK_LE(output_params.sample_rate(), limits::kMaxSampleRate);
+  DCHECK_GE(output_params.sample_rate(), limits::kMinSampleRate);
+
+  Initialize(input_params, output_params);
+}
+
+AudioRendererMixer::AudioRendererMixer(
     const AudioParameters& params, const scoped_refptr<AudioRendererSink>& sink)
-    : audio_parameters_(params),
-      audio_sink_(sink) {
-  // TODO(dalecurtis): Once we have resampling we'll need to pass on a different
-  // set of AudioParameters than the ones we're given.
-  audio_sink_->Initialize(audio_parameters_, this);
+    : audio_sink_(sink),
+      current_audio_delay_milliseconds_(0) {
+  // Create output parameters for passing on to the output sink.
+  // TODO(dalecurtis): Set format and buffer size in AudioRendererImpl.
+  Initialize(params, AudioParameters(

[scherkus (not reviewing), 2012/07/12 23:02:52]

should the test-only ARM(in, out, sink) ctor be "the only" ctor?

What's the rationale for baking in the output params for ARM clients? Should ARM
clients be able to choose their output params?

[DaleCurtis, 2012/07/12 23:16:29]

We could bake it into AudioRenderMixerManager (next CL) instead.  Then
media/base doesn't need to care about the audio hardware stuff, only
content/renderer/media.

We have to bake them in at least here because AudioRendererImpl needs to keep
percolating up the AudioParameters of the source sampling rate etc.  As such ARM
clients can't set their own output parameters.

Having them live in ARMM sounds fine with me. WDYT?

[DaleCurtis, 2012/07/13 00:01:25]

Next patch set does this.

+      AudioParameters::AUDIO_PCM_LOW_LATENCY, params.channel_layout(),
+      GetAudioHardwareSampleRate(), 16, GetAudioHardwareBufferSize()));
+}
+
+void AudioRendererMixer::Initialize(const AudioParameters& input_params,
+                                    const AudioParameters& output_params) {
+  // Only resample if necessary since it's expensive.
+  if (input_params.sample_rate() != output_params.sample_rate()) {
+    resampler_.reset(new MultiChannelResampler(
+        output_params.channels(),
+        input_params.sample_rate() / static_cast<double>(
+            output_params.sample_rate()),
+        base::Bind(&AudioRendererMixer::ProvideInput, base::Unretained(this))));
+  }
+
+  audio_sink_->Initialize(output_params, this);
   audio_sink_->Start();
 }
 
@@ -22,6 +56,11 @@ AudioRendererMixer::~AudioRendererMixer() {
   // AudioRendererSinks must be stopped before being destructed.
   audio_sink_->Stop();
 
+  // Clean up |mixer_input_audio_data_|.
+  for (size_t i = 0; i < mixer_input_audio_data_.size(); ++i)
+    delete [] mixer_input_audio_data_[i];
+  mixer_input_audio_data_.clear();
+
   // Ensures that all mixer inputs have stopped themselves prior to destruction
   // and have called RemoveMixerInput().
   DCHECK_EQ(mixer_inputs_.size(), 0U);
@@ -42,11 +81,40 @@ void AudioRendererMixer::RemoveMixerInput(
 int AudioRendererMixer::Render(const std::vector<float*>& audio_data,
                                int number_of_frames,
                                int audio_delay_milliseconds) {
+  current_audio_delay_milliseconds_ = audio_delay_milliseconds;

[scherkus (not reviewing), 2012/07/12 23:02:52]

quick check: are we still planning on removing delay at some point?

[DaleCurtis, 2012/07/12 23:16:29]

Yes, as far as I know. We can't do that until the rest of the pipeline is in
place though.

[scherkus (not reviewing), 2012/07/14 01:45:02]

SGTM

+
+  if (resampler_ != NULL)

[scherkus (not reviewing), 2012/07/12 23:02:52]

s/resampler_ != NULL/resampler_.get()/

[DaleCurtis, 2012/07/13 00:01:25]

Done.

+    resampler_->Resample(audio_data, number_of_frames);
+  else
+    ProvideInput(audio_data, number_of_frames);
+
+  // Always return the full number of frames requested, ProvideInput() will pad
+  // with silence if it wasn't able to acquire enough data.
+  return number_of_frames;
+}
+
+void AudioRendererMixer::ProvideInput(const std::vector<float*>& audio_data,
+                                      int number_of_frames) {
   base::AutoLock auto_lock(mixer_inputs_lock_);
 
+  // Allocate staging area for each mixer input's audio data on first call.  We
+  // won't know how much to allocate until here because of resampling.
+  if (mixer_input_audio_data_.size() == 0) {
+    // TODO(dalecurtis): If we switch to AVX/SSE optimization, we'll need to
+    // allocate these on 32-byte boundaries and ensure they're sized % 32 bytes.
+    mixer_input_audio_data_.reserve(audio_data.size());
+    for (size_t i = 0; i < audio_data.size(); ++i)
+      mixer_input_audio_data_.push_back(new float[number_of_frames]);
+    mixer_input_audio_data_size_ = number_of_frames;
+  }
+
+  // Sanity check our inputs.
+  DCHECK_LE(number_of_frames, mixer_input_audio_data_size_);
+  DCHECK_EQ(audio_data.size(), mixer_input_audio_data_.size());
+
   // Zero |audio_data| so we're mixing into a clean buffer and return silence if
   // we couldn't get enough data from our inputs.
-  for (int i = 0; i < audio_parameters_.channels(); ++i)
+  for (size_t i = 0; i < audio_data.size(); ++i)
     memset(audio_data[i], 0, number_of_frames * sizeof(*audio_data[i]));
 
   // Have each mixer render its data into an output buffer then mix the result.
@@ -57,24 +125,21 @@ int AudioRendererMixer::Render(const std::vector<float*>& audio_data,
     double volume;
     input->GetVolume(&volume);
 
-    // Nothing to do if the input isn't playing or the volume is zero.
-    if (!input->playing() || volume == 0.0f)
+    // Nothing to do if the input isn't playing.
+    if (!input->playing())
       continue;
 
-    const std::vector<float*>& mixer_input_audio_data = input->audio_data();
-
     int frames_filled = input->callback()->Render(
-        mixer_input_audio_data, number_of_frames, audio_delay_milliseconds);
+        mixer_input_audio_data_, number_of_frames,
+        current_audio_delay_milliseconds_);
     if (frames_filled == 0)
       continue;
 
-    // TODO(dalecurtis): Resample audio data.
-
     // Volume adjust and mix each mixer input into |audio_data| after rendering.
     // TODO(dalecurtis): Optimize with NEON/SSE/AVX vector_fmac from FFmpeg.
-    for (int j = 0; j < audio_parameters_.channels(); ++j) {
+    for (size_t j = 0; j < audio_data.size(); ++j) {
       float* dest = audio_data[j];
-      float* source = mixer_input_audio_data[j];
+      float* source = mixer_input_audio_data_[j];
       for (int k = 0; k < frames_filled; ++k)
         dest[k] += source[k] * static_cast<float>(volume);
     }
@@ -82,10 +147,6 @@ int AudioRendererMixer::Render(const std::vector<float*>& audio_data,
     // No need to clamp values as InterleaveFloatToInt() will take care of this
     // for us later when data is transferred to the browser process.
   }
-
-  // Always return the full number of frames requested, padded with silence if
-  // we couldn't get enough data.
-  return number_of_frames;
 }
 
 void AudioRendererMixer::OnRenderError() {