Merge pull request #5311 from ligfx/mixerdpl2

AudioCommon: Move DPL2 decoding into Mixer

10 files changed, 169 insertions, 124 deletions

diff --git a/Source/Core/AudioCommon/AudioCommon.vcxproj b/Source/Core/AudioCommon/AudioCommon.vcxproj
index 6ab07e55f4..dad6e891ed 100644
--- a/Source/Core/AudioCommon/AudioCommon.vcxproj
+++ b/Source/Core/AudioCommon/AudioCommon.vcxproj
@@ -38,6 +38,7 @@
   <ItemGroup>
     <ClCompile Include="aldlist.cpp" />
     <ClCompile Include="AudioCommon.cpp" />
+    <ClCompile Include="AudioStretcher.cpp" />
     <ClCompile Include="CubebStream.cpp" />
     <ClCompile Include="CubebUtils.cpp" />
     <ClCompile Include="DPL2Decoder.cpp" />
@@ -54,6 +55,7 @@
     <ClInclude Include="aldlist.h" />
     <ClInclude Include="AlsaSoundStream.h" />
     <ClInclude Include="AudioCommon.h" />
+    <ClInclude Include="AudioStretcher.h" />
     <ClInclude Include="CoreAudioSoundStream.h" />
     <ClInclude Include="CubebStream.h" />
     <ClInclude Include="CubebUtils.h" />
diff --git a/Source/Core/AudioCommon/AudioCommon.vcxproj.filters b/Source/Core/AudioCommon/AudioCommon.vcxproj.filters
index 35ba0a6c7d..661c6ce0ea 100644
--- a/Source/Core/AudioCommon/AudioCommon.vcxproj.filters
+++ b/Source/Core/AudioCommon/AudioCommon.vcxproj.filters
@@ -8,6 +8,7 @@
   <ItemGroup>
     <ClCompile Include="aldlist.cpp" />
     <ClCompile Include="AudioCommon.cpp" />
+    <ClCompile Include="AudioStretcher.cpp" />
     <ClCompile Include="CubebUtils.cpp" />
     <ClCompile Include="DPL2Decoder.cpp" />
     <ClCompile Include="Mixer.cpp" />
@@ -31,6 +32,7 @@
   <ItemGroup>
     <ClInclude Include="aldlist.h" />
     <ClInclude Include="AudioCommon.h" />
+    <ClInclude Include="AudioStretcher.h" />
     <ClInclude Include="CubebUtils.h" />
     <ClInclude Include="DPL2Decoder.h" />
     <ClInclude Include="Mixer.h" />
diff --git a/Source/Core/AudioCommon/AudioStretcher.cpp b/Source/Core/AudioCommon/AudioStretcher.cpp
new file mode 100644
index 0000000000..88721b0e35
--- /dev/null
+++ b/Source/Core/AudioCommon/AudioStretcher.cpp
@@ -0,0 +1,89 @@
+// Copyright 2017 Dolphin Emulator Project
+// Licensed under GPLv2+
+// Refer to the license.txt file included.
+
+#include <algorithm>
+#include <cmath>
+#include <cstddef>
+
+#include "AudioCommon/AudioStretcher.h"
+#include "Common/Logging/Log.h"
+#include "Core/ConfigManager.h"
+
+namespace AudioCommon
+{
+AudioStretcher::AudioStretcher(unsigned int sample_rate) : m_sample_rate(sample_rate)
+{
+  m_sound_touch.setChannels(2);
+  m_sound_touch.setSampleRate(sample_rate);
+  m_sound_touch.setPitch(1.0);
+  m_sound_touch.setTempo(1.0);
+  m_sound_touch.setSetting(SETTING_USE_QUICKSEEK, 0);
+  m_sound_touch.setSetting(SETTING_SEQUENCE_MS, 62);
+  m_sound_touch.setSetting(SETTING_SEEKWINDOW_MS, 28);
+  m_sound_touch.setSetting(SETTING_OVERLAP_MS, 8);
+}
+
+void AudioStretcher::Clear()
+{
+  m_sound_touch.clear();
+}
+
+void AudioStretcher::ProcessSamples(const short* in, unsigned int num_in, unsigned int num_out)
+{
+  const double time_delta = static_cast<double>(num_out) / m_sample_rate;  // seconds
+
+  // We were given actual_samples number of samples, and num_samples were requested from us.
+  double current_ratio = static_cast<double>(num_in) / static_cast<double>(num_out);
+
+  const double max_latency = SConfig::GetInstance().m_audio_stretch_max_latency;
+  const double max_backlog = m_sample_rate * max_latency / 1000.0 / m_stretch_ratio;
+  const double backlog_fullness = m_sound_touch.numSamples() / max_backlog;
+  if (backlog_fullness > 5.0)
+  {
+    // Too many samples in backlog: Don't push anymore on
+    num_in = 0;
+  }
+
+  // We ideally want the backlog to be about 50% full.
+  // This gives some headroom both ways to prevent underflow and overflow.
+  // We tweak current_ratio to encourage this.
+  constexpr double tweak_time_scale = 0.5;  // seconds
+  current_ratio *= 1.0 + 2.0 * (backlog_fullness - 0.5) * (time_delta / tweak_time_scale);
+
+  // This low-pass filter smoothes out variance in the calculated stretch ratio.
+  // The time-scale determines how responsive this filter is.
+  constexpr double lpf_time_scale = 1.0;  // seconds
+  const double lpf_gain = 1.0 - std::exp(-time_delta / lpf_time_scale);
+  m_stretch_ratio += lpf_gain * (current_ratio - m_stretch_ratio);
+
+  // Place a lower limit of 10% speed.  When a game boots up, there will be
+  // many silence samples.  These do not need to be timestretched.
+  m_stretch_ratio = std::max(m_stretch_ratio, 0.1);
+  m_sound_touch.setTempo(m_stretch_ratio);
+
+  DEBUG_LOG(AUDIO, "Audio stretching: samples:%u/%u ratio:%f backlog:%f gain: %f", num_in, num_out,
+            m_stretch_ratio, backlog_fullness, lpf_gain);
+
+  m_sound_touch.putSamples(in, num_in);
+}
+
+void AudioStretcher::GetStretchedSamples(short* out, unsigned int num_out)
+{
+  const size_t samples_received = m_sound_touch.receiveSamples(out, num_out);
+
+  if (samples_received != 0)
+  {
+    m_last_stretched_sample[0] = out[samples_received * 2 - 2];
+    m_last_stretched_sample[1] = out[samples_received * 2 - 1];
+  }
+
+  // Perform padding if we've run out of samples.
+  for (size_t i = samples_received; i < num_out; i++)
+  {
+    out[i * 2 + 0] = m_last_stretched_sample[0];
+    out[i * 2 + 1] = m_last_stretched_sample[1];
+  }
+}
+
+}  // namespace AudioCommon
diff --git a/Source/Core/AudioCommon/AudioStretcher.h b/Source/Core/AudioCommon/AudioStretcher.h
new file mode 100644
index 0000000000..b914fb924b
--- /dev/null
+++ b/Source/Core/AudioCommon/AudioStretcher.h
@@ -0,0 +1,28 @@
+// Copyright 2017 Dolphin Emulator Project
+// Licensed under GPLv2+
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <array>
+
+#include <soundtouch/SoundTouch.h>
+
+namespace AudioCommon
+{
+class AudioStretcher
+{
+public:
+  explicit AudioStretcher(unsigned int sample_rate);
+  void ProcessSamples(const short* in, unsigned int num_in, unsigned int num_out);
+  void GetStretchedSamples(short* out, unsigned int num_out);
+  void Clear();
+
+private:
+  unsigned int m_sample_rate;
+  std::array<short, 2> m_last_stretched_sample = {};
+  soundtouch::SoundTouch m_sound_touch;
+  double m_stretch_ratio = 1.0;
+};
+
+}  // AudioCommon
diff --git a/Source/Core/AudioCommon/CMakeLists.txt b/Source/Core/AudioCommon/CMakeLists.txt
index 94eb45c9ed..7ede29eb41 100644
--- a/Source/Core/AudioCommon/CMakeLists.txt
+++ b/Source/Core/AudioCommon/CMakeLists.txt
@@ -1,5 +1,6 @@
 set(SRCS
   AudioCommon.cpp
+  AudioStretcher.cpp
   CubebStream.cpp
   CubebUtils.cpp
   DPL2Decoder.cpp
diff --git a/Source/Core/AudioCommon/CubebStream.cpp b/Source/Core/AudioCommon/CubebStream.cpp
index c9052e29c9..272060cd74 100644
--- a/Source/Core/AudioCommon/CubebStream.cpp
+++ b/Source/Core/AudioCommon/CubebStream.cpp
@@ -21,29 +21,9 @@ long CubebStream::DataCallback(cubeb_stream* stream, void* user_data, const void
   auto* self = static_cast<CubebStream*>(user_data);
 
   if (self->m_stereo)
-  {
     self->m_mixer->Mix(static_cast<short*>(output_buffer), num_frames);
-  }
   else
-  {
-    size_t required_capacity = num_frames * 2;
-    if (required_capacity > self->m_short_buffer.capacity() ||
-        required_capacity > self->m_floatstereo_buffer.capacity())
-    {
-      INFO_LOG(AUDIO, "Expanding conversion buffers size: %li frames", num_frames);
-      self->m_short_buffer.reserve(required_capacity);
-      self->m_floatstereo_buffer.reserve(required_capacity);
-    }
-
-    self->m_mixer->Mix(self->m_short_buffer.data(), num_frames);
-
-    // s16 to float
-    for (size_t i = 0; i < static_cast<size_t>(num_frames) * 2; ++i)
-      self->m_floatstereo_buffer[i] = self->m_short_buffer[i] / static_cast<float>(1 << 15);
-
-    // DPL2Decode output: LEFTFRONT, RIGHTFRONT, CENTREFRONT, (sub), LEFTREAR, RIGHTREAR
-    DPL2Decode(self->m_floatstereo_buffer.data(), num_frames, static_cast<float*>(output_buffer));
-  }
+    self->m_mixer->MixSurround(static_cast<float*>(output_buffer), num_frames);
 
   return num_frames;
 }
diff --git a/Source/Core/AudioCommon/Mixer.cpp b/Source/Core/AudioCommon/Mixer.cpp
index 8149da088c..659deb819f 100644
--- a/Source/Core/AudioCommon/Mixer.cpp
+++ b/Source/Core/AudioCommon/Mixer.cpp
@@ -7,24 +7,18 @@
 #include <cmath>
 #include <cstring>
 
+#include "AudioCommon/DPL2Decoder.h"
 #include "Common/CommonTypes.h"
 #include "Common/Logging/Log.h"
 #include "Common/MathUtil.h"
 #include "Common/Swap.h"
 #include "Core/ConfigManager.h"
 
-CMixer::CMixer(unsigned int BackendSampleRate) : m_sampleRate(BackendSampleRate)
+CMixer::CMixer(unsigned int BackendSampleRate)
+    : m_sampleRate(BackendSampleRate), m_stretcher(BackendSampleRate)
 {
   INFO_LOG(AUDIO_INTERFACE, "Mixer is initialized");
-
-  m_sound_touch.setChannels(2);
-  m_sound_touch.setSampleRate(BackendSampleRate);
-  m_sound_touch.setPitch(1.0);
-  m_sound_touch.setTempo(1.0);
-  m_sound_touch.setSetting(SETTING_USE_QUICKSEEK, 0);
-  m_sound_touch.setSetting(SETTING_SEQUENCE_MS, 62);
-  m_sound_touch.setSetting(SETTING_SEEKWINDOW_MS, 28);
-  m_sound_touch.setSetting(SETTING_OVERLAP_MS, 8);
+  DPL2Reset();
 }
 
 CMixer::~CMixer()
@@ -135,18 +129,19 @@ unsigned int CMixer::Mix(short* samples, unsigned int num_samples)
     unsigned int available_samples =
         std::min(m_dma_mixer.AvailableSamples(), m_streaming_mixer.AvailableSamples());
 
-    m_stretch_buffer.fill(0);
+    m_scratch_buffer.fill(0);
 
-    m_dma_mixer.Mix(m_stretch_buffer.data(), available_samples, false);
-    m_streaming_mixer.Mix(m_stretch_buffer.data(), available_samples, false);
-    m_wiimote_speaker_mixer.Mix(m_stretch_buffer.data(), available_samples, false);
+    m_dma_mixer.Mix(m_scratch_buffer.data(), available_samples, false);
+    m_streaming_mixer.Mix(m_scratch_buffer.data(), available_samples, false);
+    m_wiimote_speaker_mixer.Mix(m_scratch_buffer.data(), available_samples, false);
 
     if (!m_is_stretching)
     {
-      m_sound_touch.clear();
+      m_stretcher.Clear();
       m_is_stretching = true;
     }
-    StretchAudio(m_stretch_buffer.data(), available_samples, samples, num_samples);
+    m_stretcher.ProcessSamples(m_scratch_buffer.data(), available_samples, num_samples);
+    m_stretcher.GetStretchedSamples(samples, num_samples);
   }
   else
   {
@@ -159,58 +154,25 @@ unsigned int CMixer::Mix(short* samples, unsigned int num_samples)
   return num_samples;
 }
 
-void CMixer::StretchAudio(const short* in, unsigned int num_in, short* out, unsigned int num_out)
+unsigned int CMixer::MixSurround(float* samples, unsigned int num_samples)
 {
-  const double time_delta = static_cast<double>(num_out) / m_sampleRate;  // seconds
+  if (!num_samples)
+    return 0;
 
-  // We were given actual_samples number of samples, and num_samples were requested from us.
-  double current_ratio = static_cast<double>(num_in) / static_cast<double>(num_out);
+  memset(samples, 0, num_samples * 6 * sizeof(float));
 
-  const double max_latency = SConfig::GetInstance().m_audio_stretch_max_latency;
-  const double max_backlog = m_sampleRate * max_latency / 1000.0 / m_stretch_ratio;
-  const double backlog_fullness = m_sound_touch.numSamples() / max_backlog;
-  if (backlog_fullness > 5.0)
+  // Mix() may also use m_scratch_buffer internally, but is safe because it alternates reads and
+  // writes.
+  unsigned int available_samples = Mix(m_scratch_buffer.data(), num_samples);
+  for (size_t i = 0; i < static_cast<size_t>(available_samples) * 2; ++i)
   {
-    // Too many samples in backlog: Don't push anymore on
-    num_in = 0;
+    m_float_conversion_buffer[i] =
+        m_scratch_buffer[i] / static_cast<float>(std::numeric_limits<short>::max());
   }
 
-  // We ideally want the backlog to be about 50% full.
-  // This gives some headroom both ways to prevent underflow and overflow.
-  // We tweak current_ratio to encourage this.
-  constexpr double tweak_time_scale = 0.5;  // seconds
-  current_ratio *= 1.0 + 2.0 * (backlog_fullness - 0.5) * (time_delta / tweak_time_scale);
-
-  // This low-pass filter smoothes out variance in the calculated stretch ratio.
-  // The time-scale determines how responsive this filter is.
-  constexpr double lpf_time_scale = 1.0;  // seconds
-  const double m_lpf_gain = 1.0 - std::exp(-time_delta / lpf_time_scale);
-  m_stretch_ratio += m_lpf_gain * (current_ratio - m_stretch_ratio);
-
-  // Place a lower limit of 10% speed.  When a game boots up, there will be
-  // many silence samples.  These do not need to be timestretched.
-  m_stretch_ratio = std::max(m_stretch_ratio, 0.1);
-  m_sound_touch.setTempo(m_stretch_ratio);
-
-  DEBUG_LOG(AUDIO, "Audio stretching: samples:%u/%u ratio:%f backlog:%f gain: %f", num_in, num_out,
-            m_stretch_ratio, backlog_fullness, m_lpf_gain);
-
-  m_sound_touch.putSamples(in, num_in);
+  DPL2Decode(m_float_conversion_buffer.data(), available_samples, samples);
 
-  const size_t samples_received = m_sound_touch.receiveSamples(out, num_out);
-
-  if (samples_received != 0)
-  {
-    m_last_stretched_sample[0] = out[samples_received * 2 - 2];
-    m_last_stretched_sample[1] = out[samples_received * 2 - 1];
-  }
-
-  // Preform padding if we've run out of samples.
-  for (size_t i = samples_received; i < num_out; i++)
-  {
-    out[i * 2 + 0] = m_last_stretched_sample[0];
-    out[i * 2 + 1] = m_last_stretched_sample[1];
-  }
+  return available_samples;
 }
 
 void CMixer::MixerFifo::PushSamples(const short* samples, unsigned int num_samples)
diff --git a/Source/Core/AudioCommon/Mixer.h b/Source/Core/AudioCommon/Mixer.h
index f2127cceba..2707df3383 100644
--- a/Source/Core/AudioCommon/Mixer.h
+++ b/Source/Core/AudioCommon/Mixer.h
@@ -7,12 +7,10 @@
 #include <array>
 #include <atomic>
 
+#include "AudioCommon/AudioStretcher.h"
 #include "AudioCommon/WaveFile.h"
 #include "Common/CommonTypes.h"
 
-#include <soundtouch/STTypes.h>
-#include <soundtouch/SoundTouch.h>
-
 class CMixer final
 {
 public:
@@ -21,6 +19,7 @@ public:
 
   // Called from audio threads
   unsigned int Mix(short* samples, unsigned int numSamples);
+  unsigned int MixSurround(float* samples, unsigned int num_samples);
 
   // Called from main thread
   void PushSamples(const short* samples, unsigned int num_samples);
@@ -75,18 +74,15 @@ private:
     u32 m_frac = 0;
   };
 
-  void StretchAudio(const short* in, unsigned int num_in, short* out, unsigned int num_out);
-
   MixerFifo m_dma_mixer{this, 32000};
   MixerFifo m_streaming_mixer{this, 48000};
   MixerFifo m_wiimote_speaker_mixer{this, 3000};
   unsigned int m_sampleRate;
 
   bool m_is_stretching = false;
-  soundtouch::SoundTouch m_sound_touch;
-  double m_stretch_ratio = 1.0;
-  std::array<short, 2> m_last_stretched_sample = {};
-  std::array<short, MAX_SAMPLES * 2> m_stretch_buffer;
+  AudioCommon::AudioStretcher m_stretcher;
+  std::array<short, MAX_SAMPLES * 2> m_scratch_buffer;
+  std::array<float, MAX_SAMPLES * 2> m_float_conversion_buffer;
 
   WaveFileWriter m_wave_writer_dtk;
   WaveFileWriter m_wave_writer_dsp;
diff --git a/Source/Core/AudioCommon/OpenALStream.cpp b/Source/Core/AudioCommon/OpenALStream.cpp
index 3d4bc90e1d..ca343be005 100644
--- a/Source/Core/AudioCommon/OpenALStream.cpp
+++ b/Source/Core/AudioCommon/OpenALStream.cpp
@@ -6,7 +6,6 @@
 #include <cstring>
 #include <thread>
 
-#include "AudioCommon/DPL2Decoder.h"
 #include "AudioCommon/OpenALStream.h"
 #include "AudioCommon/aldlist.h"
 #include "Common/Logging/Log.h"
@@ -66,9 +65,6 @@ bool OpenALStream::Start()
     PanicAlertT("OpenAL: can't find sound devices");
   }
 
-  // Initialize DPL2 parameters
-  DPL2Reset();
-
   return bReturn;
 }
 
@@ -228,23 +224,18 @@ void OpenALStream::SoundLoop()
       numBuffersQueued -= numBuffersProcessed;
     }
 
-    // DPL2 accepts 240 samples minimum (FWRDURATION)
-    unsigned int minSamples = surround_capable ? 240 : 0;
-
     unsigned int numSamples = OAL_MAX_SAMPLES;
-    numSamples = m_mixer->Mix(realtimeBuffer, numSamples);
-
-    // Convert the samples from short to float
-    for (u32 i = 0; i < numSamples * STEREO_CHANNELS; ++i)
-      sampleBuffer[i] = static_cast<float>(realtimeBuffer[i]) / (1 << 15);
-
-    if (numSamples <= minSamples)
-      continue;
 
     if (surround_capable)
     {
+      // DPL2 accepts 240 samples minimum (FWRDURATION)
+      unsigned int minSamples = 240;
+
       float dpl2[OAL_MAX_SAMPLES * OAL_MAX_BUFFERS * SURROUND_CHANNELS];
-      DPL2Decode(sampleBuffer, numSamples, dpl2);
+      numSamples = m_mixer->MixSurround(dpl2, numSamples);
+
+      if (numSamples < minSamples)
+        continue;
 
       // zero-out the subwoofer channel - DPL2Decode generates a pretty
       // good 5.0 but not a good 5.1 output.  Sadly there is not a 5.0
@@ -311,6 +302,15 @@ void OpenALStream::SoundLoop()
     }
     else
     {
+      numSamples = m_mixer->Mix(realtimeBuffer, numSamples);
+
+      // Convert the samples from short to float
+      for (u32 i = 0; i < numSamples * STEREO_CHANNELS; ++i)
+        sampleBuffer[i] = static_cast<float>(realtimeBuffer[i]) / (1 << 15);
+
+      if (!numSamples)
+        continue;
+
       if (float32_capable)
       {
         alBufferData(uiBuffers[nextBuffer], AL_FORMAT_STEREO_FLOAT32, sampleBuffer,
diff --git a/Source/Core/AudioCommon/PulseAudioStream.cpp b/Source/Core/AudioCommon/PulseAudioStream.cpp
index e6e6a45ce2..827816d4e8 100644
--- a/Source/Core/AudioCommon/PulseAudioStream.cpp
+++ b/Source/Core/AudioCommon/PulseAudioStream.cpp
@@ -4,7 +4,6 @@
 
 #include <cstring>
 
-#include "AudioCommon/DPL2Decoder.h"
 #include "AudioCommon/PulseAudioStream.h"
 #include "Common/CommonTypes.h"
 #include "Common/Logging/Log.h"
@@ -30,9 +29,6 @@ bool PulseAudio::Start()
   m_run_thread.Set();
   m_thread = std::thread(&PulseAudio::SoundLoop, this);
 
-  // Initialize DPL2 parameters
-  DPL2Reset();
-
   return true;
 }
 
@@ -194,23 +190,12 @@ void PulseAudio::WriteCallback(pa_stream* s, size_t length)
   }
   else
   {
-    // get a floating point mix
-    s16 s16buffer_stereo[frames * 2];
-    m_mixer->Mix(s16buffer_stereo, frames);  // implicitly mixes to 16-bit stereo
-
-    float floatbuffer_stereo[frames * 2];
-    // s16 to float
-    for (int i = 0; i < frames * 2; ++i)
-    {
-      floatbuffer_stereo[i] = s16buffer_stereo[i] / float(1 << 15);
-    }
-
     if (m_channels == 5)  // Extract dpl2/5.0 Surround
     {
       float floatbuffer_6chan[frames * 6];
-      // DPL2Decode output: LEFTFRONT, RIGHTFRONT, CENTREFRONT, (sub), LEFTREAR, RIGHTREAR
-      DPL2Decode(floatbuffer_stereo, frames, floatbuffer_6chan);
+      m_mixer->MixSurround(floatbuffer_6chan, frames);
 
+      // DPL2Decode output: LEFTFRONT, RIGHTFRONT, CENTREFRONT, (sub), LEFTREAR, RIGHTREAR
       // Discard the subwoofer channel - DPL2Decode generates a pretty
       // good 5.0 but not a good 5.1 output.
       const int dpl2_to_5chan[] = {0, 1, 2, 4, 5};