Bug 1887774 delay mPacketizerInput initialization until required r=pehrsons

mPacketizerInput won't need to be re-created for every channel count change if
no processing is done for some channel counts.

More significantly, this will facilitate delaying, in a subsequent patch,
AudioProcessing construction while its parameters may still be changing.

Differential Revision: https://phabricator.services.mozilla.com/D206867

dom/media/gtest/TestAudioInputProcessing.cpp

@@ -112,16 +112,12 @@ TEST(TestAudioInputProcessing, Buffering)
     EXPECT_EQ(aip->NumBufferedFrames(graph), 0);
   }
 
-  // aip has been started and set to processing mode so it will insert 80 frames
-  // into aip's internal buffer as pre-buffering.
+  // aip has been set to processing mode and is started.
   aip->Start(graph);
   {
     // Need (nextTime - processedTime) = 256 - 256 = 0 frames this round.
-    // The Process() aip will take 0 frames from input, packetize and process
-    // these frames into 0 80-frame packet(0 frames left in packetizer), insert
-    // packets into aip's internal buffer, then move 0 frames the internal
-    // buffer to output, leaving 80 + 0 - 0 = 80 frames in aip's internal
-    // buffer.
+    // Process() will return early on 0 frames of input.
+    // Pre-buffering is not triggered.
     processedTime = nextTime;
     nextTime = MediaTrackGraphImpl::RoundUpToEndOfAudioBlock(3 * frames);
 
@@ -131,12 +127,14 @@ TEST(TestAudioInputProcessing, Buffering)
     aip->Process(graph, processedTime, nextTime, &input, &output);
     EXPECT_EQ(input.GetDuration(), nextTime - processedTime);
     EXPECT_EQ(output.GetDuration(), nextTime);
-    EXPECT_EQ(aip->NumBufferedFrames(graph), 80);
+    EXPECT_EQ(aip->NumBufferedFrames(graph), 0);
   }
 
   {
     // Need (nextTime - processedTime) = 384 - 256 = 128 frames this round.
-    // The Process() aip will take 128 frames from input, packetize and process
+    // On receipt of the these first frames, aip will insert 80 frames
+    // into its internal buffer as pre-buffering.
+    // Process() will take 128 frames from input, packetize and process
     // these frames into floor(128/80) = 1 80-frame packet (48 frames left in
     // packetizer), insert packets into aip's internal buffer, then move 128
     // frames the internal buffer to output, leaving 80 + 80 - 128 = 32 frames
@@ -274,23 +272,22 @@ TEST(TestAudioInputProcessing, ProcessDataWithDifferentPrincipals)
   EXPECT_EQ(aip->PassThrough(graph), false);
   aip->Start(graph);
   {
-    EXPECT_EQ(aip->NumBufferedFrames(graph), 480);
     AudioSegment output;
     {
-      // Trim the prebuffering silence.
-
       AudioSegment data;
       aip->Process(graph, 0, 4800, &input, &data);
       EXPECT_EQ(input.GetDuration(), 4800);
       EXPECT_EQ(data.GetDuration(), 4800);
 
+      // Extract another 480 frames to account for delay from pre-buffering.
+      EXPECT_EQ(aip->NumBufferedFrames(graph), 480);
       AudioSegment dummy;
       dummy.AppendNullData(480);
       aip->Process(graph, 0, 480, &dummy, &data);
       EXPECT_EQ(dummy.GetDuration(), 480);
       EXPECT_EQ(data.GetDuration(), 480 + 4800);
 
-      // Ignore the pre-buffering data
+      // Ignore the pre-buffering silence.
       output.AppendSlice(data, 480, 480 + 4800);
     }
 

dom/media/webrtc/MediaEngineWebRTCAudio.cpp

@@ -431,12 +431,13 @@ void AudioInputProcessing::SetPassThrough(MediaTrackGraph* aGraph,
   }
 
   if (aPassThrough) {
-    // Turn on pass-through
+    // Switching to pass-through.  Clear state so that it doesn't affect any
+    // future processing, if re-enabled.
     ResetAudioProcessing(aGraph);
   } else {
-    // Turn off pass-through
+    // No longer pass-through.  Processing will not use old state.
+    // Packetizer setup is deferred until needed.
     MOZ_ASSERT(!mPacketizerInput);
-    EnsureAudioProcessing(aGraph, mRequestedInputChannelCount);
   }
 }
 
@@ -465,7 +466,6 @@ void AudioInputProcessing::Start(MediaTrackGraph* aGraph) {
   }
 
   MOZ_ASSERT(!mPacketizerInput);
-  EnsureAudioProcessing(aGraph, mRequestedInputChannelCount);
 }
 
 void AudioInputProcessing::Stop(MediaTrackGraph* aGraph) {
@@ -627,13 +627,11 @@ void AudioInputProcessing::Process(MediaTrackGraph* aGraph, GraphTime aFrom,
     return;
   }
 
-  // SetPassThrough(false) must be called before reaching here.
-  MOZ_ASSERT(mPacketizerInput);
   // If mRequestedInputChannelCount is updated, create a new packetizer. No
   // need to change the pre-buffering since the rate is always the same. The
   // frames left in the packetizer would be replaced by null data and then
   // transferred to mSegment.
-  EnsureAudioProcessing(aGraph, mRequestedInputChannelCount);
+  EnsurePacketizer(aGraph, mRequestedInputChannelCount);
 
   // Preconditions of the audio-processing logic.
   MOZ_ASSERT(static_cast<uint32_t>(mSegment.GetDuration()) +
@@ -993,7 +991,7 @@ TrackTime AudioInputProcessing::NumBufferedFrames(
   return mSegment.GetDuration();
 }
 
-void AudioInputProcessing::EnsureAudioProcessing(MediaTrackGraph* aGraph,
+void AudioInputProcessing::EnsurePacketizer(MediaTrackGraph* aGraph,
                                             uint32_t aChannels) {
   aGraph->AssertOnGraphThread();
   MOZ_ASSERT(aChannels > 0);

dom/media/webrtc/MediaEngineWebRTCAudio.h

@@ -164,7 +164,7 @@ class AudioInputProcessing : public AudioDataListener {
   ~AudioInputProcessing() = default;
   webrtc::AudioProcessing::Config ConfigForPrefs(
       const MediaEnginePrefs& aPrefs);
-  void EnsureAudioProcessing(MediaTrackGraph* aGraph, uint32_t aChannels);
+  void EnsurePacketizer(MediaTrackGraph* aGraph, uint32_t aChannels);
   void ResetAudioProcessing(MediaTrackGraph* aGraph);
   PrincipalHandle GetCheckedPrincipal(const AudioSegment& aSegment);
   // This implements the processing algoritm to apply to the input (e.g. a