Add a stand-alone upsampler for higher-order ambisonic oputput

5 files changed, 118 insertions, 10 deletions

diff --git a/Alc/ALc.c b/Alc/ALc.c
index 13f881dd..8c6381ad 100644
--- a/Alc/ALc.c
+++ b/Alc/ALc.c
@@ -2040,7 +2040,7 @@ static ALCenum UpdateDeviceParams(ALCdevice *device, const ALCint *attrList)
         size += (ChannelsFromDevFmt(device->FmtChans)+4) * sizeof(device->Dry.Buffer[0]);
     else if(device->Hrtf || device->Uhj_Encoder || device->AmbiDecoder)
         size += ChannelsFromDevFmt(device->FmtChans) * sizeof(device->Dry.Buffer[0]);
-    else if(device->FmtChans == DevFmtAmbi2 || device->FmtChans == DevFmtAmbi3)
+    else if(device->FmtChans > DevFmtAmbi1 && device->FmtChans <= DevFmtAmbi3)
         size += 4 * sizeof(device->Dry.Buffer[0]);
     device->Dry.Buffer = al_calloc(16, size);
     if(!device->Dry.Buffer)
@@ -2061,7 +2061,7 @@ static ALCenum UpdateDeviceParams(ALCdevice *device, const ALCint *attrList)
     }
 
     if((device->AmbiDecoder && bformatdec_getOrder(device->AmbiDecoder) >= 2) ||
-       device->FmtChans == DevFmtAmbi2 || device->FmtChans == DevFmtAmbi3)
+       (device->FmtChans > DevFmtAmbi1 && device->FmtChans <= DevFmtAmbi3))
     {
         /* Higher-order rendering requires upsampling first-order content, so
          * make sure to mix it separately.
@@ -2211,6 +2211,9 @@ static ALCvoid FreeDevice(ALCdevice *device)
     bformatdec_free(device->AmbiDecoder);
     device->AmbiDecoder = NULL;
 
+    ambiup_free(device->AmbiUp);
+    device->AmbiUp = NULL;
+
     AL_STRING_DEINIT(device->DeviceName);
 
     al_free(device->Dry.Buffer);
diff --git a/Alc/ALu.c b/Alc/ALu.c
index c1478b50..d1a12672 100644
--- a/Alc/ALu.c
+++ b/Alc/ALu.c
@@ -1534,6 +1534,13 @@ ALvoid aluMixData(ALCdevice *device, ALvoid *buffer, ALsizei size)
                 device->Dry.Buffer, SamplesToDo
             );
         }
+        else if(device->AmbiUp)
+        {
+            ambiup_process(device->AmbiUp,
+                device->RealOut.Buffer, device->RealOut.NumChannels,
+                device->FOAOut.Buffer, SamplesToDo
+            );
+        }
         else if(device->Uhj_Encoder)
         {
             int lidx = GetChannelIdxByName(device->RealOut, FrontLeft);
diff --git a/Alc/bformatdec.c b/Alc/bformatdec.c
index d6cf25e2..3c175b33 100644
--- a/Alc/bformatdec.c
+++ b/Alc/bformatdec.c
@@ -130,6 +130,16 @@ static const ALfloat SquareMatrix[4][FB_Max][MAX_AMBI_COEFFS] = {
 };
 static ALfloat SquareEncoder[4][MAX_AMBI_COEFFS];
 
+static const ALfloat CubePoints[8][3] = {
+    { -0.577350269f,  0.577350269f, -0.577350269f },
+    {  0.577350269f,  0.577350269f, -0.577350269f },
+    { -0.577350269f,  0.577350269f,  0.577350269f },
+    {  0.577350269f,  0.577350269f,  0.577350269f },
+    { -0.577350269f, -0.577350269f, -0.577350269f },
+    {  0.577350269f, -0.577350269f, -0.577350269f },
+    { -0.577350269f, -0.577350269f,  0.577350269f },
+    {  0.577350269f, -0.577350269f,  0.577350269f },
+};
 static const ALfloat CubeMatrix[8][FB_Max][MAX_AMBI_COEFFS] = {
     { { 0.25f,  0.14425f,  0.14425f,  0.14425f }, { 0.125f,  0.125f,  0.125f,  0.125f } },
     { { 0.25f, -0.14425f,  0.14425f,  0.14425f }, { 0.125f, -0.125f,  0.125f,  0.125f } },
@@ -167,14 +177,8 @@ static void init_bformatdec(void)
 
     MixMatrixRow = SelectMixer();
 
-    CalcXYZCoeffs(-0.577350269f,  0.577350269f, -0.577350269f, 0.0f, CubeEncoder[0]);
-    CalcXYZCoeffs( 0.577350269f,  0.577350269f, -0.577350269f, 0.0f, CubeEncoder[1]);
-    CalcXYZCoeffs(-0.577350269f,  0.577350269f,  0.577350269f, 0.0f, CubeEncoder[2]);
-    CalcXYZCoeffs( 0.577350269f,  0.577350269f,  0.577350269f, 0.0f, CubeEncoder[3]);
-    CalcXYZCoeffs(-0.577350269f, -0.577350269f, -0.577350269f, 0.0f, CubeEncoder[4]);
-    CalcXYZCoeffs( 0.577350269f, -0.577350269f, -0.577350269f, 0.0f, CubeEncoder[5]);
-    CalcXYZCoeffs(-0.577350269f, -0.577350269f,  0.577350269f, 0.0f, CubeEncoder[6]);
-    CalcXYZCoeffs( 0.577350269f, -0.577350269f,  0.577350269f, 0.0f, CubeEncoder[7]);
+    for(i = 0;i < COUNTOF(CubePoints);i++)
+        CalcDirectionCoeffs(CubePoints[i], 0.0f, CubeEncoder[i]);
 
     CalcXYZCoeffs(-0.707106781f,  0.0f, -0.707106781f, 0.0f, SquareEncoder[0]);
     CalcXYZCoeffs( 0.707106781f,  0.0f, -0.707106781f, 0.0f, SquareEncoder[1]);
@@ -588,3 +592,68 @@ void bformatdec_upSample(struct BFormatDec *dec, ALfloat (*restrict OutBuffer)[B
         }
     }
 }
+
+
+typedef struct AmbiUpsampler {
+    alignas(16) ALfloat SamplesHF[4][BUFFERSIZE];
+    alignas(16) ALfloat SamplesLF[4][BUFFERSIZE];
+
+    alignas(16) ALfloat ChannelMix[BUFFERSIZE];
+
+    BandSplitter XOver[4];
+
+    ALfloat Gains[8][MAX_OUTPUT_CHANNELS];
+    ALuint NumChannels;
+} AmbiUpsampler;
+
+AmbiUpsampler *ambiup_alloc()
+{
+    alcall_once(&bformatdec_inited, init_bformatdec);
+    return al_calloc(16, sizeof(AmbiUpsampler));
+}
+
+void ambiup_free(struct AmbiUpsampler *ambiup)
+{
+    al_free(ambiup);
+}
+
+void ambiup_reset(struct AmbiUpsampler *ambiup, const ALCdevice *device)
+{
+    ALfloat ratio;
+    ALuint i;
+
+    ratio = 400.0f / (ALfloat)device->Frequency;
+    for(i = 0;i < 4;i++)
+        bandsplit_init(&ambiup->XOver[i], ratio);
+
+    ambiup->NumChannels = COUNTOF(CubePoints);
+    for(i = 0;i < ambiup->NumChannels;i++)
+        ComputePanningGains(device->Dry, CubeEncoder[i], 1.0f, ambiup->Gains[i]);
+}
+
+void ambiup_process(struct AmbiUpsampler *ambiup, ALfloat (*restrict OutBuffer)[BUFFERSIZE], ALuint OutChannels, ALfloat (*restrict InSamples)[BUFFERSIZE], ALuint SamplesToDo)
+{
+    ALuint i, j, k;
+
+    for(i = 0;i < 4;i++)
+        bandsplit_process(&ambiup->XOver[i], ambiup->SamplesHF[i], ambiup->SamplesLF[i],
+                          InSamples[i], SamplesToDo);
+
+    for(k = 0;k < ambiup->NumChannels;k++)
+    {
+        memset(ambiup->ChannelMix, 0, SamplesToDo*sizeof(ALfloat));
+        MixMatrixRow(ambiup->ChannelMix, CubeMatrix[k][FB_HighFreq],
+                     ambiup->SamplesHF, 4, SamplesToDo);
+        MixMatrixRow(ambiup->ChannelMix, CubeMatrix[k][FB_LowFreq],
+                     ambiup->SamplesLF, 4, SamplesToDo);
+
+        for(j = 0;j < OutChannels;j++)
+        {
+            ALfloat gain = ambiup->Gains[k][j];
+            if(!(fabsf(gain) > GAIN_SILENCE_THRESHOLD))
+                continue;
+            for(i = 0;i < SamplesToDo;i++)
+                OutBuffer[j][i] += ambiup->ChannelMix[i] * gain;
+        }
+    }
+}
diff --git a/Alc/bformatdec.h b/Alc/bformatdec.h
index 3ad06373..be5a69f6 100644
--- a/Alc/bformatdec.h
+++ b/Alc/bformatdec.h
@@ -5,6 +5,7 @@
 
 struct AmbDecConf;
 struct BFormatDec;
+struct AmbiUpsampler;
 
 enum BFormatDecFlags {
     BFDF_DistanceComp = 1<<0
@@ -21,4 +22,13 @@ void bformatdec_process(struct BFormatDec *dec, ALfloat (*restrict OutBuffer)[BU
 /* Up-samples a first-order input to the decoder's configuration. */
 void bformatdec_upSample(struct BFormatDec *dec, ALfloat (*restrict OutBuffer)[BUFFERSIZE], ALfloat (*restrict InSamples)[BUFFERSIZE], ALuint InChannels, ALuint SamplesToDo);
 
+/* Stand-alone first-order upsampler. Kept here because it shares some stuff
+ * with bformatdec.
+ */
+struct AmbiUpsampler *ambiup_alloc();
+void ambiup_free(struct AmbiUpsampler *ambiup);
+void ambiup_reset(struct AmbiUpsampler *ambiup, const ALCdevice *device);
+
+void ambiup_process(struct AmbiUpsampler *ambiup, ALfloat (*restrict OutBuffer)[BUFFERSIZE], ALuint OutChannels, ALfloat (*restrict InSamples)[BUFFERSIZE], ALuint SamplesToDo);
+
 #endif /* BFORMATDEC_H */
diff --git a/Alc/panning.c b/Alc/panning.c
index 0c1bf06f..80909564 100644
--- a/Alc/panning.c
+++ b/Alc/panning.c
@@ -622,6 +622,9 @@ static void InitPanning(ALCdevice *device)
         device->Dry.CoeffCount = 0;
         device->Dry.NumChannels = count;
 
+        /* FOA output is always ACN+N3D for higher-order ambisonic output. The
+         * upsampler expects this and will convert it for output.
+         */
         memset(&device->FOAOut.Ambi, 0, sizeof(device->FOAOut.Ambi));
         for(i = 0;i < 4;i++)
         {
@@ -629,6 +632,8 @@ static void InitPanning(ALCdevice *device)
             device->FOAOut.Ambi.Map[i].Index = i;
         }
         device->FOAOut.CoeffCount = 0;
+
+        ambiup_reset(device->AmbiUp, device);
     }
     else
     {
@@ -903,6 +908,8 @@ void aluInitRenderer(ALCdevice *device, ALint hrtf_id, enum HrtfRequestMode hrtf
 
         if(pconf && GetConfigValueBool(devname, "decoder", "hq-mode", 0))
         {
+            ambiup_free(device->AmbiUp);
+            device->AmbiUp = NULL;
             if(!device->AmbiDecoder)
                 device->AmbiDecoder = bformatdec_alloc();
         }
@@ -910,6 +917,16 @@ void aluInitRenderer(ALCdevice *device, ALint hrtf_id, enum HrtfRequestMode hrtf
         {
             bformatdec_free(device->AmbiDecoder);
             device->AmbiDecoder = NULL;
+            if(device->FmtChans > DevFmtAmbi1 && device->FmtChans <= DevFmtAmbi3)
+            {
+                if(!device->AmbiUp)
+                    device->AmbiUp = ambiup_alloc();
+            }
+            else
+            {
+                ambiup_free(device->AmbiUp);
+                device->AmbiUp = NULL;
+            }
         }
 
         if(!pconf)
@@ -923,6 +940,8 @@ void aluInitRenderer(ALCdevice *device, ALint hrtf_id, enum HrtfRequestMode hrtf
         return;
     }
 
+    ambiup_free(device->AmbiUp);
+    device->AmbiUp = NULL;
     bformatdec_free(device->AmbiDecoder);
     device->AmbiDecoder = NULL;