Use second-order ambisonics for basic HRTF rendering

This should improve positional quality for relatively low cost. Full HRTF
rendering still only uses first-order since the only use of the dry buffer
there is for first-order content (B-Format buffers, effects).

6 files changed, 95 insertions, 25 deletions

diff --git a/Alc/ALc.c b/Alc/ALc.c
index c9507564..0b558ff7 100644
--- a/Alc/ALc.c
+++ b/Alc/ALc.c
@@ -2055,9 +2055,13 @@ static ALCenum UpdateDeviceParams(ALCdevice *device, const ALCint *attrList)
     if((device->AmbiDecoder && bformatdec_getOrder(device->AmbiDecoder) >= 2))
         size += (ChannelsFromDevFmt(device->FmtChans)+4) * sizeof(device->Dry.Buffer[0]);
     else if(device->Hrtf.Handle || device->Uhj_Encoder || device->AmbiDecoder)
+    {
         size += ChannelsFromDevFmt(device->FmtChans) * sizeof(device->Dry.Buffer[0]);
-    else if(device->FmtChans > DevFmtAmbi1 && device->FmtChans <= DevFmtAmbi3)
+        if(device->AmbiUp) size += 4 * sizeof(device->Dry.Buffer[0]);
+    }
+    else if(device->AmbiUp)
         size += 4 * sizeof(device->Dry.Buffer[0]);
+    TRACE("Allocating "SZFMT" channels, "SZFMT" bytes\n", size/sizeof(device->Dry.Buffer[0]), size);
     device->Dry.Buffer = al_calloc(16, size);
     if(!device->Dry.Buffer)
     {
@@ -2076,8 +2080,7 @@ static ALCenum UpdateDeviceParams(ALCdevice *device, const ALCint *attrList)
         device->RealOut.NumChannels = device->Dry.NumChannels;
     }
 
-    if((device->AmbiDecoder && bformatdec_getOrder(device->AmbiDecoder) >= 2) ||
-       (device->FmtChans > DevFmtAmbi1 && device->FmtChans <= DevFmtAmbi3))
+    if((device->AmbiDecoder && bformatdec_getOrder(device->AmbiDecoder) >= 2) || device->AmbiUp)
     {
         /* Higher-order rendering requires upsampling first-order content, so
          * make sure to mix it separately.
@@ -2090,6 +2093,8 @@ static ALCenum UpdateDeviceParams(ALCdevice *device, const ALCint *attrList)
         device->FOAOut.Buffer = device->Dry.Buffer;
         device->FOAOut.NumChannels = device->Dry.NumChannels;
     }
+    TRACE("Channel config, Dry: %u, FOA: %u, Real: %u\n", device->Dry.NumChannels,
+          device->FOAOut.NumChannels, device->RealOut.NumChannels);
 
     SetMixerFPUMode(&oldMode);
     if(device->DefaultSlot)
diff --git a/Alc/ALu.c b/Alc/ALu.c
index dc979522..192f2a35 100644
--- a/Alc/ALu.c
+++ b/Alc/ALu.c
@@ -1488,6 +1488,13 @@ ALvoid aluMixData(ALCdevice *device, ALvoid *buffer, ALsizei size)
         {
             int lidx = GetChannelIdxByName(device->RealOut, FrontLeft);
             int ridx = GetChannelIdxByName(device->RealOut, FrontRight);
+
+            if(device->AmbiUp)
+                ambiup_process(device->AmbiUp,
+                    device->Dry.Buffer, device->Dry.NumChannels,
+                    SAFE_CONST(ALfloatBUFFERSIZE*,device->FOAOut.Buffer), SamplesToDo
+                );
+
             if(lidx != -1 && ridx != -1)
             {
                 HrtfDirectMixerFunc HrtfMix = SelectHrtfMixer();
diff --git a/Alc/hrtf.c b/Alc/hrtf.c
index 38cd0618..a6efa34b 100644
--- a/Alc/hrtf.c
+++ b/Alc/hrtf.c
@@ -136,12 +136,14 @@ void GetHrtfCoeffs(const struct Hrtf *Hrtf, ALfloat elevation, ALfloat azimuth,
 }
 
 
-ALuint BuildBFormatHrtf(const struct Hrtf *Hrtf, ALfloat (*coeffs)[HRIR_LENGTH][2], ALuint NumChannels)
+ALuint BuildBFormatHrtf(const struct Hrtf *Hrtf, ALfloat (*coeffs)[HRIR_LENGTH][2], ALuint NumChannels, const ALuint *AmbiMap)
 {
+#define HRTF_AMBI_CHAN_COUNT 14
+    /* NOTE: azimuth goes clockwise. */
     static const struct {
         ALfloat elevation;
         ALfloat azimuth;
-    } Ambi3DPoints[14] = {
+    } Ambi3DPoints[HRTF_AMBI_CHAN_COUNT] = {
         { DEG2RAD( 90.0f), DEG2RAD(   0.0f) },
         { DEG2RAD( 35.0f), DEG2RAD( -45.0f) },
         { DEG2RAD( 35.0f), DEG2RAD(  45.0f) },
@@ -157,7 +159,7 @@ ALuint BuildBFormatHrtf(const struct Hrtf *Hrtf, ALfloat (*coeffs)[HRIR_LENGTH][
         { DEG2RAD(-35.0f), DEG2RAD(-135.0f) },
         { DEG2RAD(-90.0f), DEG2RAD(   0.0f) },
     };
-    static const ALfloat Ambi3DMatrix[14][2][MAX_AMBI_COEFFS] = {
+    static const ALfloat Ambi3DMatrixFOA[HRTF_AMBI_CHAN_COUNT][2][MAX_AMBI_COEFFS] = {
         { { 1.88982237e-001f,  0.00000000e+000f,  1.90399923e-001f,  0.00000000e+000f }, { 7.14285714e-002f,  0.00000000e+000f,  1.24646009e-001f,  0.00000000e+000f } },
         { { 1.88982237e-001f,  1.09057783e-001f,  1.09208910e-001f,  1.09057783e-001f }, { 7.14285714e-002f,  7.13950780e-002f,  7.14940135e-002f,  7.13950780e-002f } },
         { { 1.88982237e-001f, -1.09057783e-001f,  1.09208910e-001f,  1.09057783e-001f }, { 7.14285714e-002f, -7.13950780e-002f,  7.14940135e-002f,  7.13950780e-002f } },
@@ -172,9 +174,25 @@ ALuint BuildBFormatHrtf(const struct Hrtf *Hrtf, ALfloat (*coeffs)[HRIR_LENGTH][
         { { 1.88982237e-001f, -1.09057783e-001f, -1.09208910e-001f, -1.09057783e-001f }, { 7.14285714e-002f, -7.13950780e-002f, -7.14940135e-002f, -7.13950780e-002f } },
         { { 1.88982237e-001f,  1.09057783e-001f, -1.09208910e-001f, -1.09057783e-001f }, { 7.14285714e-002f,  7.13950780e-002f, -7.14940135e-002f, -7.13950780e-002f } },
         { { 1.88982237e-001f,  0.00000000e+000f, -1.90399923e-001f,  0.00000000e+000f }, { 7.14285714e-002f,  0.00000000e+000f, -1.24646009e-001f,  0.00000000e+000f } }
+    }, Ambi3DMatrixHOA[HRTF_AMBI_CHAN_COUNT][2][MAX_AMBI_COEFFS] = {
+        { { 1.43315266e-001f,  0.00000000e+000f,  1.90399923e-001f,  0.00000000e+000f,  0.00000000e+000f,  0.00000000e+000f,  1.18020996e-001f,  0.00000000e+000f,  0.00000000e+000f }, { 7.26741039e-002f,  0.00000000e+000f,  1.24646009e-001f,  0.00000000e+000f,  0.00000000e+000f,  0.00000000e+000f,  1.49618920e-001f,  0.00000000e+000f,  0.00000000e+000f } },
+        { { 1.40852210e-001f,  1.09057783e-001f,  1.09208910e-001f,  1.09057783e-001f,  7.58818830e-002f,  7.66295578e-002f, -3.28314629e-004f,  7.66295578e-002f,  0.00000000e+000f }, { 7.14251066e-002f,  7.13950780e-002f,  7.14940135e-002f,  7.13950780e-002f,  9.61978444e-002f,  9.71456952e-002f, -4.16214759e-004f,  9.71456952e-002f,  0.00000000e+000f } },
+        { { 1.40852210e-001f, -1.09057783e-001f,  1.09208910e-001f,  1.09057783e-001f, -7.58818830e-002f, -7.66295578e-002f, -3.28314629e-004f,  7.66295578e-002f,  0.00000000e+000f }, { 7.14251066e-002f, -7.13950780e-002f,  7.14940135e-002f,  7.13950780e-002f, -9.61978444e-002f, -9.71456952e-002f, -4.16214759e-004f,  9.71456952e-002f,  0.00000000e+000f } },
+        { { 1.40852210e-001f, -1.09057783e-001f,  1.09208910e-001f, -1.09057783e-001f,  7.58818830e-002f, -7.66295578e-002f, -3.28314629e-004f, -7.66295578e-002f,  0.00000000e+000f }, { 7.14251066e-002f, -7.13950780e-002f,  7.14940135e-002f, -7.13950780e-002f,  9.61978444e-002f, -9.71456952e-002f, -4.16214759e-004f, -9.71456952e-002f,  0.00000000e+000f } },
+        { { 1.40852210e-001f,  1.09057783e-001f,  1.09208910e-001f, -1.09057783e-001f, -7.58818830e-002f,  7.66295578e-002f, -3.28314629e-004f, -7.66295578e-002f,  0.00000000e+000f }, { 7.14251066e-002f,  7.13950780e-002f,  7.14940135e-002f, -7.13950780e-002f, -9.61978444e-002f,  9.71456952e-002f, -4.16214759e-004f, -9.71456952e-002f,  0.00000000e+000f } },
+        { { 1.39644596e-001f,  0.00000000e+000f,  0.00000000e+000f,  1.88281281e-001f,  0.00000000e+000f,  0.00000000e+000f, -5.83538687e-002f,  0.00000000e+000f,  1.01835015e-001f }, { 7.08127349e-002f,  0.00000000e+000f,  0.00000000e+000f,  1.23259031e-001f,  0.00000000e+000f,  0.00000000e+000f, -7.39770307e-002f,  0.00000000e+000f,  1.29099445e-001f } },
+        { { 1.39644596e-001f, -1.88281281e-001f,  0.00000000e+000f,  0.00000000e+000f,  0.00000000e+000f,  0.00000000e+000f, -5.83538687e-002f,  0.00000000e+000f, -1.01835015e-001f }, { 7.08127349e-002f, -1.23259031e-001f,  0.00000000e+000f,  0.00000000e+000f,  0.00000000e+000f,  0.00000000e+000f, -7.39770307e-002f,  0.00000000e+000f, -1.29099445e-001f } },
+        { { 1.39644596e-001f,  0.00000000e+000f,  0.00000000e+000f, -1.88281281e-001f,  0.00000000e+000f,  0.00000000e+000f, -5.83538687e-002f,  0.00000000e+000f,  1.01835015e-001f }, { 7.08127349e-002f,  0.00000000e+000f,  0.00000000e+000f, -1.23259031e-001f,  0.00000000e+000f,  0.00000000e+000f, -7.39770307e-002f,  0.00000000e+000f,  1.29099445e-001f } },
+        { { 1.39644596e-001f,  1.88281281e-001f,  0.00000000e+000f,  0.00000000e+000f,  0.00000000e+000f,  0.00000000e+000f, -5.83538687e-002f,  0.00000000e+000f, -1.01835015e-001f }, { 7.08127349e-002f,  1.23259031e-001f,  0.00000000e+000f,  0.00000000e+000f,  0.00000000e+000f,  0.00000000e+000f, -7.39770307e-002f,  0.00000000e+000f, -1.29099445e-001f } },
+        { { 1.40852210e-001f,  1.09057783e-001f, -1.09208910e-001f,  1.09057783e-001f,  7.58818830e-002f, -7.66295578e-002f, -3.28314629e-004f, -7.66295578e-002f,  0.00000000e+000f }, { 7.14251066e-002f,  7.13950780e-002f, -7.14940135e-002f,  7.13950780e-002f,  9.61978444e-002f, -9.71456952e-002f, -4.16214759e-004f, -9.71456952e-002f,  0.00000000e+000f } },
+        { { 1.40852210e-001f, -1.09057783e-001f, -1.09208910e-001f,  1.09057783e-001f, -7.58818830e-002f,  7.66295578e-002f, -3.28314629e-004f, -7.66295578e-002f,  0.00000000e+000f }, { 7.14251066e-002f, -7.13950780e-002f, -7.14940135e-002f,  7.13950780e-002f, -9.61978444e-002f,  9.71456952e-002f, -4.16214759e-004f, -9.71456952e-002f,  0.00000000e+000f } },
+        { { 1.40852210e-001f, -1.09057783e-001f, -1.09208910e-001f, -1.09057783e-001f,  7.58818830e-002f,  7.66295578e-002f, -3.28314629e-004f,  7.66295578e-002f,  0.00000000e+000f }, { 7.14251066e-002f, -7.13950780e-002f, -7.14940135e-002f, -7.13950780e-002f,  9.61978444e-002f,  9.71456952e-002f, -4.16214759e-004f,  9.71456952e-002f,  0.00000000e+000f } },
+        { { 1.40852210e-001f,  1.09057783e-001f, -1.09208910e-001f, -1.09057783e-001f, -7.58818830e-002f, -7.66295578e-002f, -3.28314629e-004f,  7.66295578e-002f,  0.00000000e+000f }, { 7.14251066e-002f,  7.13950780e-002f, -7.14940135e-002f, -7.13950780e-002f, -9.61978444e-002f, -9.71456952e-002f, -4.16214759e-004f,  9.71456952e-002f,  0.00000000e+000f } },
+        { { 1.43315266e-001f,  0.00000000e+000f, -1.90399923e-001f,  0.00000000e+000f,  0.00000000e+000f,  0.00000000e+000f,  1.18020996e-001f,  0.00000000e+000f,  0.00000000e+000f }, { 7.26741039e-002f,  0.00000000e+000f, -1.24646009e-001f,  0.00000000e+000f,  0.00000000e+000f,  0.00000000e+000f,  1.49618920e-001f,  0.00000000e+000f,  0.00000000e+000f } },
     };
+    const ALfloat (*Ambi3DMatrix)[2][MAX_AMBI_COEFFS] = AmbiMap ? Ambi3DMatrixHOA : Ambi3DMatrixFOA;
 
-/* Change this to 2 for dual-band HRTF processing. May require a higher quality
+/* Set this to 2 for dual-band HRTF processing. May require a higher quality
  * band-splitter, or better calculation of the new IR length to deal with the
  * tail generated by the filter.
  */
@@ -186,9 +204,7 @@ ALuint BuildBFormatHrtf(const struct Hrtf *Hrtf, ALfloat (*coeffs)[HRIR_LENGTH][
     ALuint max_length = 0;
     ALuint i, j, c, b;
 
-    assert(NumChannels == 4);
-
-    for(c = 0;c < COUNTOF(Ambi3DPoints);c++)
+    for(c = 0;c < HRTF_AMBI_CHAN_COUNT;c++)
     {
         ALuint evidx, azidx;
         ALuint evoffset;
@@ -215,7 +231,7 @@ ALuint BuildBFormatHrtf(const struct Hrtf *Hrtf, ALfloat (*coeffs)[HRIR_LENGTH][
 
     memset(temps, 0, sizeof(temps));
     bandsplit_init(&splitter, 400.0f / (ALfloat)Hrtf->sampleRate);
-    for(c = 0;c < COUNTOF(Ambi3DMatrix);c++)
+    for(c = 0;c < HRTF_AMBI_CHAN_COUNT;c++)
     {
         const ALshort *fir;
         ALuint delay;
@@ -240,11 +256,12 @@ ALuint BuildBFormatHrtf(const struct Hrtf *Hrtf, ALfloat (*coeffs)[HRIR_LENGTH][
         delay = Hrtf->delays[lidx[c]] - min_delay;
         for(i = 0;i < NumChannels;++i)
         {
+            const ALsizei a = AmbiMap ? AmbiMap[i] : i;
             for(b = 0;b < NUM_BANDS;b++)
             {
                 ALuint k = 0;
                 for(j = delay;j < HRIR_LENGTH;++j)
-                    coeffs[i][j][0] += temps[b][k++] * Ambi3DMatrix[c][b][i];
+                    coeffs[i][j][0] += temps[b][k++] * Ambi3DMatrix[c][b][a];
             }
         }
         max_length = maxu(max_length, minu(delay + Hrtf->irSize, HRIR_LENGTH));
@@ -269,11 +286,12 @@ ALuint BuildBFormatHrtf(const struct Hrtf *Hrtf, ALfloat (*coeffs)[HRIR_LENGTH][
         delay = Hrtf->delays[ridx[c]] - min_delay;
         for(i = 0;i < NumChannels;++i)
         {
+            const ALsizei a = AmbiMap ? AmbiMap[i] : i;
             for(b = 0;b < NUM_BANDS;b++)
             {
                 ALuint k = 0;
                 for(j = delay;j < HRIR_LENGTH;++j)
-                    coeffs[i][j][1] += temps[b][k++] * Ambi3DMatrix[c][b][i];
+                    coeffs[i][j][1] += temps[b][k++] * Ambi3DMatrix[c][b][a];
             }
         }
         max_length = maxu(max_length, minu(delay + Hrtf->irSize, HRIR_LENGTH));
@@ -282,6 +300,7 @@ ALuint BuildBFormatHrtf(const struct Hrtf *Hrtf, ALfloat (*coeffs)[HRIR_LENGTH][
 #undef NUM_BANDS
 
     return max_length;
+#undef HRTF_AMBI_CHAN_COUNT
 }
 
 
diff --git a/Alc/hrtf.h b/Alc/hrtf.h
index ebba0d50..7620a3d2 100644
--- a/Alc/hrtf.h
+++ b/Alc/hrtf.h
@@ -47,6 +47,6 @@ void GetHrtfCoeffs(const struct Hrtf *Hrtf, ALfloat elevation, ALfloat azimuth,
  * for first-order. Returns the maximum impulse-response length of the
  * generated coefficients.
  */
-ALuint BuildBFormatHrtf(const struct Hrtf *Hrtf, ALfloat (*coeffs)[HRIR_LENGTH][2], ALuint NumChannels);
+ALuint BuildBFormatHrtf(const struct Hrtf *Hrtf, ALfloat (*coeffs)[HRIR_LENGTH][2], ALuint NumChannels, const ALuint *AmbiMap);
 
 #endif /* ALC_HRTF_H */
diff --git a/Alc/panning.c b/Alc/panning.c
index 4e4caf46..de9b8cb1 100644
--- a/Alc/panning.c
+++ b/Alc/panning.c
@@ -756,25 +756,45 @@ static void InitHQPanning(ALCdevice *device, const AmbDecConf *conf, const ALuin
     }
 }
 
-static void InitHrtfPanning(ALCdevice *device)
+static void InitHrtfPanning(ALCdevice *device, bool hoa_mode)
 {
-    size_t count = 4;
+    static const ALuint map_foa[] = { 0, 1, 2, 3 };
+    static const ALuint map_hoa[] = { 0, 1, 2, 3, 4, 5, 6, 7, 8 };
+    const ALuint *ambi_map = hoa_mode ? map_hoa : map_foa;
+    size_t count = hoa_mode ? COUNTOF(map_hoa) : COUNTOF(map_foa);
     ALuint i;
 
+    static_assert(COUNTOF(map_hoa) <= COUNTOF(device->Hrtf.Coeffs), "ALCdevice::Hrtf.Values/Coeffs size is too small");
+
     for(i = 0;i < count;i++)
     {
         device->Dry.Ambi.Map[i].Scale = 1.0f;
-        device->Dry.Ambi.Map[i].Index = i;
+        device->Dry.Ambi.Map[i].Index = ambi_map[i];
     }
     device->Dry.CoeffCount = 0;
     device->Dry.NumChannels = count;
 
-    device->FOAOut.Ambi = device->Dry.Ambi;
-    device->FOAOut.CoeffCount = device->Dry.CoeffCount;
+    if(!hoa_mode)
+    {
+        device->FOAOut.Ambi = device->Dry.Ambi;
+        device->FOAOut.CoeffCount = device->Dry.CoeffCount;
+    }
+    else
+    {
+        memset(&device->FOAOut.Ambi, 0, sizeof(device->FOAOut.Ambi));
+        for(i = 0;i < 4;i++)
+        {
+            device->FOAOut.Ambi.Map[i].Scale = 1.0f;
+            device->FOAOut.Ambi.Map[i].Index = i;
+        }
+        device->FOAOut.CoeffCount = 0;
+
+        ambiup_reset(device->AmbiUp, device);
+    }
 
     memset(device->Hrtf.Coeffs, 0, sizeof(device->Hrtf.Coeffs));
     device->Hrtf.IrSize = BuildBFormatHrtf(device->Hrtf.Handle,
-        device->Hrtf.Coeffs, device->Dry.NumChannels
+        device->Hrtf.Coeffs, device->Dry.NumChannels, hoa_mode ? ambi_map : NULL
     );
 
     /* Round up to the nearest multiple of 8 */
@@ -895,8 +915,6 @@ 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;
 
@@ -964,6 +982,8 @@ void aluInitRenderer(ALCdevice *device, ALint hrtf_id, enum HrtfRequestMode hrtf
 
     if(device->Hrtf.Handle)
     {
+        bool hoa_mode;
+
         device->Render_Mode = HrtfRender;
         if(ConfigValueStr(al_string_get_cstr(device->DeviceName), NULL, "hrtf-mode", &mode))
         {
@@ -975,11 +995,27 @@ void aluInitRenderer(ALCdevice *device, ALint hrtf_id, enum HrtfRequestMode hrtf
                 ERR("Unexpected hrtf-mode: %s\n", mode);
         }
 
+        if(device->Render_Mode == HrtfRender)
+        {
+            /* Don't bother with HOA when using full HRTF rendering. Nothing
+             * needs it, and it eases the CPU/memory load.
+             */
+            ambiup_free(device->AmbiUp);
+            device->AmbiUp = NULL;
+            hoa_mode = false;
+        }
+        else
+        {
+            if(!device->AmbiUp)
+                device->AmbiUp = ambiup_alloc();
+            hoa_mode = true;
+        }
+
         TRACE("%s HRTF rendering enabled, using \"%s\"\n",
             ((device->Render_Mode == HrtfRender) ? "Full" : "Basic"),
             al_string_get_cstr(device->Hrtf.Name)
         );
-        InitHrtfPanning(device);
+        InitHrtfPanning(device, hoa_mode);
         return;
     }
     device->Hrtf.Status = ALC_HRTF_UNSUPPORTED_FORMAT_SOFT;
@@ -987,6 +1023,9 @@ void aluInitRenderer(ALCdevice *device, ALint hrtf_id, enum HrtfRequestMode hrtf
 no_hrtf:
     TRACE("HRTF disabled\n");
 
+    ambiup_free(device->AmbiUp);
+    device->AmbiUp = NULL;
+
     bs2blevel = ((headphones && hrtf_appreq != Hrtf_Disable) ||
                  (hrtf_appreq == Hrtf_Enable)) ? 5 : 0;
     if(device->Type != Loopback)
diff --git a/OpenAL32/Include/alMain.h b/OpenAL32/Include/alMain.h
index 5a5c3923..975dd11e 100644
--- a/OpenAL32/Include/alMain.h
+++ b/OpenAL32/Include/alMain.h
@@ -643,8 +643,8 @@ struct ALCdevice_struct
         const struct Hrtf *Handle;
 
         /* HRTF filter state for dry buffer content */
-        alignas(16) ALfloat Values[4][HRIR_LENGTH][2];
-        alignas(16) ALfloat Coeffs[4][HRIR_LENGTH][2];
+        alignas(16) ALfloat Values[9][HRIR_LENGTH][2];
+        alignas(16) ALfloat Coeffs[9][HRIR_LENGTH][2];
         ALuint Offset;
         ALuint IrSize;
     } Hrtf;