Up-sample first-order content when using a higher order HQ decoder

4 files changed, 101 insertions, 2 deletions

diff --git a/Alc/ALu.c b/Alc/ALu.c
index b5b7de9c..206aceda 100644
--- a/Alc/ALu.c
+++ b/Alc/ALu.c
@@ -1367,6 +1367,9 @@ ALvoid aluMixData(ALCdevice *device, ALvoid *buffer, ALsizei size)
             memset(device->VirtOut.Buffer[c], 0, SamplesToDo*sizeof(ALfloat));
         for(c = 0;c < device->RealOut.NumChannels;c++)
             memset(device->RealOut.Buffer[c], 0, SamplesToDo*sizeof(ALfloat));
+        if(device->Dry.Buffer != device->FOAOut.Buffer)
+            for(c = 0;c < device->FOAOut.NumChannels;c++)
+                memset(device->FOAOut.Buffer[c], 0, SamplesToDo*sizeof(ALfloat));
 
         V0(device->Backend,lock)();
 
@@ -1467,6 +1470,11 @@ ALvoid aluMixData(ALCdevice *device, ALvoid *buffer, ALsizei size)
         }
         else if(device->AmbiDecoder)
         {
+            if(device->VirtOut.Buffer != device->FOAOut.Buffer)
+                bformatdec_upSample(device->AmbiDecoder,
+                    device->VirtOut.Buffer, device->FOAOut.Buffer,
+                    device->FOAOut.NumChannels, SamplesToDo
+                );
             bformatdec_process(device->AmbiDecoder,
                 device->RealOut.Buffer, device->RealOut.NumChannels,
                 device->VirtOut.Buffer, SamplesToDo
diff --git a/Alc/bformatdec.c b/Alc/bformatdec.c
index 37be67e4..ab148d1b 100644
--- a/Alc/bformatdec.c
+++ b/Alc/bformatdec.c
@@ -5,6 +5,8 @@
 #include "ambdec.h"
 #include "alu.h"
 
+#include "threads.h"
+
 
 typedef struct BandSplitter {
     ALfloat coeff;
@@ -105,6 +107,46 @@ static const ALfloat FuMa2N3DScale[MAX_AMBI_COEFFS] = {
 };
 
 
+static const ALfloat SquareMatrixHF[4][MAX_AMBI_COEFFS] = {
+    { 0.353553f,  0.204094f,  0.0f,  0.204094f },
+    { 0.353553f, -0.204094f,  0.0f,  0.204094f },
+    { 0.353553f,  0.204094f,  0.0f, -0.204094f },
+    { 0.353553f, -0.204094f,  0.0f, -0.204094f },
+};
+static ALfloat SquareEncoder[4][MAX_AMBI_COEFFS];
+
+static const ALfloat CubeMatrixHF[8][MAX_AMBI_COEFFS] = {
+    { 0.25f,  0.14425f,  0.14425f,  0.14425f },
+    { 0.25f, -0.14425f,  0.14425f,  0.14425f },
+    { 0.25f,  0.14425f,  0.14425f, -0.14425f },
+    { 0.25f, -0.14425f,  0.14425f, -0.14425f },
+    { 0.25f,  0.14425f, -0.14425f,  0.14425f },
+    { 0.25f, -0.14425f, -0.14425f,  0.14425f },
+    { 0.25f,  0.14425f, -0.14425f, -0.14425f },
+    { 0.25f, -0.14425f, -0.14425f, -0.14425f },
+};
+static ALfloat CubeEncoder[8][MAX_AMBI_COEFFS];
+
+static alonce_flag encoder_inited = AL_ONCE_FLAG_INIT;
+
+static void init_encoder(void)
+{
+    CalcXYZCoeffs(-0.577350269f,  0.577350269f, -0.577350269f, CubeEncoder[0]);
+    CalcXYZCoeffs( 0.577350269f,  0.577350269f, -0.577350269f, CubeEncoder[1]);
+    CalcXYZCoeffs(-0.577350269f,  0.577350269f,  0.577350269f, CubeEncoder[2]);
+    CalcXYZCoeffs( 0.577350269f,  0.577350269f,  0.577350269f, CubeEncoder[3]);
+    CalcXYZCoeffs(-0.577350269f, -0.577350269f, -0.577350269f, CubeEncoder[4]);
+    CalcXYZCoeffs( 0.577350269f, -0.577350269f, -0.577350269f, CubeEncoder[5]);
+    CalcXYZCoeffs(-0.577350269f, -0.577350269f,  0.577350269f, CubeEncoder[6]);
+    CalcXYZCoeffs( 0.577350269f, -0.577350269f,  0.577350269f, CubeEncoder[7]);
+
+    CalcXYZCoeffs(-0.707106781f,  0.0f, -0.707106781f, SquareEncoder[0]);
+    CalcXYZCoeffs( 0.707106781f,  0.0f, -0.707106781f, SquareEncoder[1]);
+    CalcXYZCoeffs(-0.707106781f,  0.0f,  0.707106781f, SquareEncoder[2]);
+    CalcXYZCoeffs( 0.707106781f,  0.0f,  0.707106781f, SquareEncoder[3]);
+}
+
+
 /* NOTE: Low-frequency (LF) fields and BandSplitter filters are unused with
  * single-band decoding
  */
@@ -119,12 +161,19 @@ typedef struct BFormatDec {
     ALfloat (*SamplesHF)[BUFFERSIZE];
     ALfloat (*SamplesLF)[BUFFERSIZE];
 
+    struct {
+        const ALfloat (*restrict MatrixHF)[MAX_AMBI_COEFFS];
+        const ALfloat (*restrict Encoder)[MAX_AMBI_COEFFS];
+        ALuint NumChannels;
+    } UpSampler;
+
     ALuint NumChannels;
     ALboolean DualBand;
 } BFormatDec;
 
 BFormatDec *bformatdec_alloc()
 {
+    alcall_once(&encoder_inited, init_encoder);
     return al_calloc(16, sizeof(BFormatDec));
 }
 
@@ -172,6 +221,19 @@ void bformatdec_reset(BFormatDec *dec, const AmbDecConf *conf, ALuint chancount,
     else if(conf->CoeffScale == ADS_FuMa)
         coeff_scale = FuMa2N3DScale;
 
+    if((conf->ChanMask & ~0x831b))
+    {
+        dec->UpSampler.MatrixHF = CubeMatrixHF;
+        dec->UpSampler.Encoder = (const ALfloat(*)[MAX_AMBI_COEFFS])CubeEncoder;
+        dec->UpSampler.NumChannels = 8;
+    }
+    else
+    {
+        dec->UpSampler.MatrixHF = SquareMatrixHF;
+        dec->UpSampler.Encoder = (const ALfloat(*)[MAX_AMBI_COEFFS])SquareEncoder;
+        dec->UpSampler.NumChannels = 4;
+    }
+
     if(conf->FreqBands == 1)
     {
         dec->DualBand = AL_FALSE;
@@ -264,3 +326,29 @@ void bformatdec_process(struct BFormatDec *dec, ALfloat (*restrict OutBuffer)[BU
                       dec->NumChannels, SamplesToDo);
     }
 }
+
+
+void bformatdec_upSample(struct BFormatDec *dec, ALfloat (*restrict OutBuffer)[BUFFERSIZE], ALfloat (*restrict InSamples)[BUFFERSIZE], ALuint InChannels, ALuint SamplesToDo)
+{
+    ALuint i, j, k;
+
+    /* This up-sampler is very simplistic. It essentially decodes the first-
+     * order content to a square channel array (or cube if height is desired),
+     * then encodes those points onto the higher order soundfield.
+     */
+    for(k = 0;k < dec->UpSampler.NumChannels;k++)
+    {
+        memset(dec->Samples[0], 0, SamplesToDo*sizeof(ALfloat));
+        apply_row(dec->Samples[0], dec->UpSampler.MatrixHF[k], InSamples,
+                  InChannels, SamplesToDo);
+
+        for(j = 0;j < dec->NumChannels;j++)
+        {
+            ALfloat gain = dec->UpSampler.Encoder[k][j];
+            if(!(fabsf(gain) > GAIN_SILENCE_THRESHOLD))
+                continue;
+            for(i = 0;i < SamplesToDo;i++)
+                OutBuffer[j][i] += dec->Samples[0][i] * gain;
+        }
+    }
+}
diff --git a/Alc/bformatdec.h b/Alc/bformatdec.h
index 5f17d711..623e7b09 100644
--- a/Alc/bformatdec.h
+++ b/Alc/bformatdec.h
@@ -10,6 +10,11 @@ struct BFormatDec *bformatdec_alloc();
 void bformatdec_free(struct BFormatDec *dec);
 int bformatdec_getOrder(const struct BFormatDec *dec);
 void bformatdec_reset(struct BFormatDec *dec, const struct AmbDecConf *conf, ALuint chancount, ALuint srate, const ALuint chanmap[MAX_OUTPUT_CHANNELS]);
+
+/* Decodes the ambisonic input to the given output channels. */
 void bformatdec_process(struct BFormatDec *dec, ALfloat (*restrict OutBuffer)[BUFFERSIZE], ALuint OutChannels, ALfloat (*restrict InSamples)[BUFFERSIZE], ALuint SamplesToDo);
 
+/* 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);
+
 #endif /* BFORMATDEC_H */
diff --git a/Alc/panning.c b/Alc/panning.c
index 031b518d..9c504b04 100644
--- a/Alc/panning.c
+++ b/Alc/panning.c
@@ -641,8 +641,6 @@ ALvoid aluInitPanning(ALCdevice *device)
         if(!MakeSpeakerMap(device, &conf, speakermap))
             goto ambi_fail;
 
-        if(conf.ChanMask > 0xf)
-            ERR("Second-order HQ decoding does not currently handle first-order sources\n");
         count = (conf.ChanMask > 0xf) ? COUNTOF(Ambi3D) : 4;
         chanmap = Ambi3D;
         ambiscale = 1.0f;