Bug 1423923 - Properly feed reverse stream to the AudioProcessingModule. r=pehrsons

We need to feed deinterleaved data, not interleaved data.

MozReview-Commit-ID: 99z8HA7tJgT

--HG--
extra : rebase_source : ca58203084bfd7018036c2d7299d2011dc27270f
extra : amend_source : d88a8d760f22026add0639c75e680435eafa8588
extra : source : 3ba7fe1cddec0a3dcaaf526a85b7f34072c3e199

dom/media/webrtc/MediaEngineWebRTCAudio.cpp

@@ -741,7 +741,7 @@ MediaEngineWebRTCMicrophoneSource::PacketizeAndProcess(MediaStreamGraph* aGraph,
       continue;
     }
     AudioDataValue* packetDataPointer = buffer->mData;
-    AutoTArray<AudioDataValue*, MAX_CHANNELS> deinterleavedPacketDataChannelPointers;
+    AutoTArray<float*, MAX_CHANNELS> deinterleavedPacketDataChannelPointers;
     AudioDataValue* interleavedFarend = nullptr;
     uint32_t channelCountFarend = 0;
     uint32_t framesPerPacketFarend = 0;
@@ -771,61 +771,39 @@ MediaEngineWebRTCMicrophoneSource::PacketizeAndProcess(MediaStreamGraph* aGraph,
                (channelCountFarend == 1 || channelCountFarend == 2) &&
                framesPerPacketFarend);
 
+    if (mInputBuffer.Length() < framesPerPacketFarend * channelCountFarend) {
+      mInputBuffer.SetLength(framesPerPacketFarend * channelCountFarend);
+    }
+
     offset = 0;
     for (size_t i = 0; i < deinterleavedPacketDataChannelPointers.Length(); ++i) {
-      deinterleavedPacketDataChannelPointers[i] = packetDataPointer + offset;
+      deinterleavedPacketDataChannelPointers[i] = mInputBuffer.Data() + offset;
       offset += framesPerPacketFarend;
     }
 
-    // deinterleave back into the FarEndAudioChunk buffer to save an alloc.
-    // There is enough room because either there is the same number of
-    // channels/frames or we've just downmixed.
-    Deinterleave(interleavedFarend,
+    // Deinterleave, prepare a channel pointers array, with enough storage for
+    // the frames.
+    //
+    // If this is a platform that uses s16 for audio input and output,
+    // convert to floats, the APM API we use only accepts floats.
+    DeinterleaveAndConvertBuffer(interleavedFarend,
                                  framesPerPacketFarend,
                                  channelCountFarend,
                                  deinterleavedPacketDataChannelPointers.Elements());
 
     // Having the same config for input and output means we potentially save
-    // some CPU. We won't need the output here, the API forces us to set a
-    // valid pointer with enough space.
+    // some CPU.
     StreamConfig inputConfig(mAudioOutputObserver->PlayoutFrequency(),
                              channelCountFarend,
                              false /* we don't use typing detection*/);
     StreamConfig outputConfig = inputConfig;
 
-    // Prepare a channel pointers array, with enough storage for the
-    // frames.
-    //
-    // If this is a platform that uses s16 for audio input and output,
-    // convert to floats, the APM API we use only accepts floats.
-
-    float* inputData = nullptr;
-#ifdef MOZ_SAMPLE_TYPE_S16
-    // Convert to floats, use mInputBuffer for this.
-    size_t sampleCount = framesPerPacketFarend * channelCountFarend;
-    if (mInputBuffer.Length() < sampleCount) {
-      mInputBuffer.SetLength(sampleCount);
-    }
-    ConvertAudioSamples(buffer->mData, mInputBuffer.Data(), sampleCount);
-    inputData = mInputBuffer.Data();
-#else // MOZ_SAMPLE_TYPE_F32
-    inputData = buffer->mData;
-#endif
-
-    AutoTArray<float*, MAX_CHANNELS> channelsPointers;
-    channelsPointers.SetLength(channelCountFarend);
-    offset = 0;
-    for (size_t i = 0; i < channelsPointers.Length(); ++i) {
-      channelsPointers[i]  = inputData + offset;
-      offset += framesPerPacketFarend;
-    }
-
     // Passing the same pointers here saves a copy inside this function.
     int err =
-      mAudioProcessing->ProcessReverseStream(channelsPointers.Elements(),
+      mAudioProcessing->ProcessReverseStream(deinterleavedPacketDataChannelPointers.Elements(),
                                              inputConfig,
                                              outputConfig,
-                                             channelsPointers.Elements());
+                                             deinterleavedPacketDataChannelPointers.Elements());
 
     if (err) {
       MOZ_LOG(GetMediaManagerLog(), LogLevel::Error,
@@ -843,6 +821,8 @@ MediaEngineWebRTCMicrophoneSource::PacketizeAndProcess(MediaStreamGraph* aGraph,
       mPacketizer->Channels();
     if (mInputBuffer.Length() < samplesPerPacket) {
       mInputBuffer.SetLength(samplesPerPacket);
+    }
+    if (mDeinterleavedBuffer.Length() < samplesPerPacket) {
       mDeinterleavedBuffer.SetLength(samplesPerPacket);
     }
     float* packet = mInputBuffer.Data();