[Chromecast] Process streams with different post-processing.

chromecast/media/cma/backend/alsa/stream_mixer_alsa.cc

Index: chromecast/media/cma/backend/alsa/stream_mixer_alsa.cc
diff --git a/chromecast/media/cma/backend/alsa/stream_mixer_alsa.cc b/chromecast/media/cma/backend/alsa/stream_mixer_alsa.cc
index 463a300aba2498a5aca63e6438cf046cc7c816c3..6b508585a18f817204ea6d2396fa8aa2d7800de0 100644
--- a/chromecast/media/cma/backend/alsa/stream_mixer_alsa.cc
+++ b/chromecast/media/cma/backend/alsa/stream_mixer_alsa.cc
@@ -7,20 +7,25 @@
 #include <algorithm>
 #include <cmath>
 #include <limits>
+#include <unordered_set>
 #include <utility>
 
 #include "base/bind_helpers.h"
 #include "base/command_line.h"
 #include "base/lazy_instance.h"
+#include "base/memory/ptr_util.h"
 #include "base/memory/weak_ptr.h"
 #include "base/single_thread_task_runner.h"
 #include "base/strings/string_number_conversions.h"
 #include "base/threading/platform_thread.h"
 #include "base/threading/thread_task_runner_handle.h"
 #include "chromecast/base/chromecast_switches.h"
+#include "chromecast/media/base/audio_device_ids.h"
 #include "chromecast/media/cma/backend/alsa/alsa_wrapper.h"
 #include "chromecast/media/cma/backend/alsa/audio_filter_factory.h"
+#include "chromecast/media/cma/backend/alsa/filter_group.h"
 #include "chromecast/media/cma/backend/alsa/stream_mixer_alsa_input_impl.h"
+#include "media/audio/audio_device_description.h"
 #include "media/base/audio_bus.h"
 #include "media/base/media_switches.h"
 
@@ -112,9 +117,6 @@ static int* kAlsaDirDontCare = nullptr;
 const snd_pcm_format_t kPreferredSampleFormats[] = {SND_PCM_FORMAT_S32,
                                                     SND_PCM_FORMAT_S16};
 
-// How many seconds of silence should be passed to the filters to flush them.
-const float kSilenceSecondsToFilter = 1.0f;
-
 const int64_t kNoTimestamp = std::numeric_limits<int64_t>::min();
 
 int64_t TimespecToMicroseconds(struct timespec time) {
@@ -167,6 +169,23 @@ bool GetSwitchValueAsNonNegativeInt(const std::string& switch_name,
   return true;
 }
 
+// Computes |A| +|B| with clipping
+int32_t SaturateSum(int32_t a, int32_t b) {

[kmackay, 2017/02/23 20:52:56]

base/numerics/saturated_arithmetic.h

[bshaya, 2017/02/24 00:17:08]

Done.

+  if (a > 0 && std::numeric_limits<int32_t>::max() - a < b) {
+    return std::numeric_limits<int32_t>::max();
+  }
+  if (a < 0 && std::numeric_limits<int32_t>::min() - a > b) {
+    return std::numeric_limits<int32_t>::min();
+  }
+  return a + b;
+}
+
+void VectorAccumulate(const int32_t* source, size_t size, int32_t* dest) {
+  for (size_t i = 0; i < size; ++i) {
+    dest[i] = SaturateSum(source[i], dest[i]);
+  }
+}
+
 class StreamMixerAlsaInstance : public StreamMixerAlsa {
  public:
   StreamMixerAlsaInstance() {}
@@ -245,11 +264,22 @@ StreamMixerAlsa::StreamMixerAlsa()
           ? kLowSampleRateCutoff
           : 0;
 
-  // Create filters
-  pre_loopback_filter_ = AudioFilterFactory::MakeAudioFilter(
-      AudioFilterFactory::PRE_LOOPBACK_FILTER);
-  post_loopback_filter_ = AudioFilterFactory::MakeAudioFilter(
-      AudioFilterFactory::POST_LOOPBACK_FILTER);
+  // Create filter groups.
+  filter_groups_.push_back(base::MakeUnique<FilterGroup>(
+      std::unordered_set<std::string>(
+          {::media::AudioDeviceDescription::kCommunicationsDeviceId}),
+      AudioFilterFactory::COMMUNICATION_AUDIO_FILTER));
+  filter_groups_.push_back(base::MakeUnique<FilterGroup>(
+      std::unordered_set<std::string>({kAlarmAudioDeviceId}),
+      AudioFilterFactory::ALARM_AUDIO_FILTER));
+  filter_groups_.push_back(base::MakeUnique<FilterGroup>(
+      std::unordered_set<std::string>({kTtsAudioDeviceId}),
+      AudioFilterFactory::TTS_AUDIO_FILTER));
+  filter_groups_.push_back(base::MakeUnique<FilterGroup>(
+      std::unordered_set<std::string>(
+          {::media::AudioDeviceDescription::kDefaultDeviceId,
+           kEarconAudioDeviceId}),
+      AudioFilterFactory::MEDIA_AUDIO_FILTER));

[gfhuang, 2017/02/23 21:24:56]

I don't think this is the right thing to do.

These code means name (or device_id) is no longer an opaque id through the
public API boundary.

Can you use AudioContentType in Ken's CL?
https://eureka-internal-review.git.corp.google.com/#/c/65141/

[kmackay, 2017/02/23 21:30:36]

I think this is OK. It's only temporary (as I understand it); and, this is our
code anyway (so I figure it's OK to use the name field if we want to).

In the short term we want to have streams with the same volume control behaviour
(eg content_type), but different filters. This won't be necessary in the long
term, I think (at least for this use case).

[bshaya, 2017/02/24 00:17:08]

The long term probably still has this. For instance, a podcast app should use
the TTS filter and the media volume.

[kmackay, 2017/02/24 17:24:57]

We haven't planned any way to distinguish content from apps. I don't think it is
really feasible to require different filter behaviour for streams with the same
volume control behaviour; the complexity of implementation goes way up. For
example if we were using ALSA plugins we would need <num volume control
types>*<num filter types> different pipelines to handle all possible cases. So
I'd prefer not to make this a long-term thing.

[bshaya, 2017/02/27 17:00:14]

Done.

 
   DefineAlsaParameters();
 }
@@ -529,14 +559,9 @@ void StreamMixerAlsa::Start() {
   }
 
   // Initialize filters
-  if (pre_loopback_filter_) {
-    pre_loopback_filter_->SetSampleRateAndFormat(
-        output_samples_per_second_, ::media::SampleFormat::kSampleFormatS32);
-  }
-
-  if (post_loopback_filter_) {
-    post_loopback_filter_->SetSampleRateAndFormat(
-        output_samples_per_second_, ::media::SampleFormat::kSampleFormatS32);
+  for (auto&& filter_group : filter_groups_) {
+    filter_group->Initialize(output_samples_per_second_,
+                             ::media::SampleFormat::kSampleFormatS32);
   }
 
   RETURN_REPORT_ERROR(PcmPrepare, pcm_);
@@ -762,6 +787,8 @@ void StreamMixerAlsa::WriteFrames() {
 
 bool StreamMixerAlsa::TryWriteFrames() {
   DCHECK(mixer_task_runner_->BelongsToCurrentThread());
+  DCHECK_GE(filter_groups_.size(), 1u);
+
   if (state_ != kStateNormalPlayback) {
     return false;
   }
@@ -769,12 +796,21 @@ bool StreamMixerAlsa::TryWriteFrames() {
   const int min_frames_in_buffer =
       output_samples_per_second_ * kMinBufferedDataMs / 1000;
   int chunk_size = output_samples_per_second_ * kMaxWriteSizeMs / 1000;
-  std::vector<InputQueue*> active_inputs;
+  bool is_silence = true;
   for (auto&& input : inputs_) {
     int read_size = input->MaxReadSize();
     if (read_size > 0) {
-      active_inputs.push_back(input.get());
+      bool group_match = false;
+      for (auto&& filter_group : filter_groups_) {
+        if (filter_group->TryAddActiveInput(input.get())) {
+          group_match = true;
+          break;
+        }
+      }
+      DCHECK(group_match) << "Could not find a filter group for "
+                          << input->name();
       chunk_size = std::min(chunk_size, read_size);
+      is_silence = false;
     } else if (input->primary()) {
       if (alsa_->PcmStatus(pcm_, pcm_status_) != 0) {
         LOG(ERROR) << "Failed to get status";
@@ -801,57 +837,52 @@ bool StreamMixerAlsa::TryWriteFrames() {
     }
   }
 
-  if (active_inputs.empty()) {
-    // No inputs have any data to provide. Fill with silence to avoid underrun.
+  if (is_silence) {
+    // No inputs have any data to provide. Push silence to prevent underrun.
     chunk_size = kPreventUnderrunChunkSize;
-    if (!mixed_ || mixed_->frames() < chunk_size) {
-      mixed_ = ::media::AudioBus::Create(kNumOutputChannels, chunk_size);
-    }
-
-    mixed_->Zero();
-    WriteMixedPcm(*mixed_, chunk_size, true /* is_silence */);
-    return true;
   }
 
-  // If |mixed_| has not been allocated, or it is too small, allocate a buffer.
-  if (!mixed_ || mixed_->frames() < chunk_size) {
-    mixed_ = ::media::AudioBus::Create(kNumOutputChannels, chunk_size);
-  }
-
-  // If |temp_| has not been allocated, or is too small, allocate a buffer.
-  if (!temp_ || temp_->frames() < chunk_size) {
-    temp_ = ::media::AudioBus::Create(kNumOutputChannels, chunk_size);
+  // Mix and filter each group.
+  std::vector<uint8_t>* interleaved = nullptr;
+  for (auto&& filter_group : filter_groups_) {
+    if (filter_group->MixAndFilter(chunk_size)) {
+      if (!interleaved) {
+        interleaved = filter_group->GetInterleaved();
+      } else {
+        DCHECK_EQ(4, BytesPerOutputFormatSample());
+        VectorAccumulate(
+            reinterpret_cast<int32_t*>(filter_group->GetInterleaved()->data()),
+            chunk_size * kNumOutputChannels,
+            reinterpret_cast<int32_t*>(interleaved->data()));
+      }
+    }
   }
 
-  mixed_->ZeroFramesPartial(0, chunk_size);
-
-  // Loop through active inputs, polling them for data, and mixing them.
-  for (InputQueue* input : active_inputs) {
-    input->GetResampledData(temp_.get(), chunk_size);
-    for (int c = 0; c < kNumOutputChannels; ++c) {
-      input->VolumeScaleAccumulate(c, temp_->channel(c), chunk_size,
-                                   mixed_->channel(c));
-    }
+  if (!interleaved) {
+    // No group has any data, write empty buffer.
+    filter_groups_[0]->ClearInterleaved(chunk_size);
+    interleaved = filter_groups_[0]->GetInterleaved();
   }
 
-  WriteMixedPcm(*mixed_, chunk_size, false /* is_silence */);
+  WriteMixedPcm(interleaved, chunk_size);
   return true;
 }
 
+size_t StreamMixerAlsa::InterleavedSize(int frames) {
+  return BytesPerOutputFormatSample() *
+         static_cast<size_t>(frames * kNumOutputChannels);
+}
+
 ssize_t StreamMixerAlsa::BytesPerOutputFormatSample() {
   return alsa_->PcmFormatSize(pcm_format_, 1);
 }
 
-void StreamMixerAlsa::WriteMixedPcm(const ::media::AudioBus& mixed,
-                                    int frames, bool is_silence) {
+void StreamMixerAlsa::WriteMixedPcm(std::vector<uint8_t>* interleaved,
+                                    int frames) {
   DCHECK(mixer_task_runner_->BelongsToCurrentThread());
   CHECK_PCM_INITIALIZED();
-
-  size_t interleaved_size = static_cast<size_t>(frames * kNumOutputChannels) *
-                            BytesPerOutputFormatSample();
-  if (interleaved_.size() < interleaved_size) {
-    interleaved_.resize(interleaved_size);
-  }
+  DCHECK(interleaved);
+  DCHECK_GE(interleaved->size(), InterleavedSize(frames));
 
   int64_t expected_playback_time;
   if (rendering_delay_.timestamp_microseconds == kNoTimestamp) {
@@ -861,37 +892,10 @@ void StreamMixerAlsa::WriteMixedPcm(const ::media::AudioBus& mixed,
                              rendering_delay_.delay_microseconds;
   }
 
-  mixed.ToInterleaved(frames, BytesPerOutputFormatSample(),
-                      interleaved_.data());
-
-  // Ensure that, on onset of silence, at least |kSilenceSecondsToFilter|
-  // second of audio get pushed through the filters to clear any memory.
-  bool filter_frames = true;
-  if (is_silence) {
-    int silence_frames_to_filter =
-        output_samples_per_second_ * kSilenceSecondsToFilter;
-    if (silence_frames_filtered_ < silence_frames_to_filter) {
-      silence_frames_filtered_ += frames;
-    } else {
-      filter_frames = false;
-    }
-  } else {
-    silence_frames_filtered_ = 0;
-  }
-
-  // Filter, send to observers, and post filter
-  if (pre_loopback_filter_ && filter_frames) {
-    pre_loopback_filter_->ProcessInterleaved(interleaved_.data(), frames);
-  }
-
   for (CastMediaShlib::LoopbackAudioObserver* observer : loopback_observers_) {
     observer->OnLoopbackAudio(expected_playback_time, kSampleFormatS32,
                               output_samples_per_second_, kNumOutputChannels,
-                              interleaved_.data(), interleaved_size);
-  }
-
-  if (post_loopback_filter_ && filter_frames) {
-    post_loopback_filter_->ProcessInterleaved(interleaved_.data(), frames);
+                              interleaved->data(), InterleavedSize(frames));
   }
 
   // If the PCM has been drained it will be in SND_PCM_STATE_SETUP and need
@@ -901,7 +905,7 @@ void StreamMixerAlsa::WriteMixedPcm(const ::media::AudioBus& mixed,
   }
 
   int frames_left = frames;
-  uint8_t* data = &interleaved_[0];
+  uint8_t* data = interleaved->data();
   while (frames_left) {
     int frames_or_error;
     while ((frames_or_error = alsa_->PcmWritei(pcm_, data, frames_left)) < 0) {