Modify bs2b_cross_feed to do multiple samples at once

3 files changed, 78 insertions, 59 deletions

diff --git a/Alc/ALu.c b/Alc/ALu.c
index 2a94d940..c1478b50 100644
--- a/Alc/ALu.c
+++ b/Alc/ALu.c
@@ -1534,33 +1534,28 @@ ALvoid aluMixData(ALCdevice *device, ALvoid *buffer, ALsizei size)
                 device->Dry.Buffer, SamplesToDo
             );
         }
-        else
+        else if(device->Uhj_Encoder)
         {
-            if(device->Uhj_Encoder)
+            int lidx = GetChannelIdxByName(device->RealOut, FrontLeft);
+            int ridx = GetChannelIdxByName(device->RealOut, FrontRight);
+            if(lidx != -1 && ridx != -1)
             {
-                int lidx = GetChannelIdxByName(device->RealOut, FrontLeft);
-                int ridx = GetChannelIdxByName(device->RealOut, FrontRight);
-                if(lidx != -1 && ridx != -1)
-                {
-                    /* Encode to stereo-compatible 2-channel UHJ output. */
-                    EncodeUhj2(device->Uhj_Encoder,
-                        device->RealOut.Buffer[lidx], device->RealOut.Buffer[ridx],
-                        device->Dry.Buffer, SamplesToDo
-                    );
-                }
+                /* Encode to stereo-compatible 2-channel UHJ output. */
+                EncodeUhj2(device->Uhj_Encoder,
+                    device->RealOut.Buffer[lidx], device->RealOut.Buffer[ridx],
+                    device->Dry.Buffer, SamplesToDo
+                );
             }
-            if(device->Bs2b)
+        }
+        else if(device->Bs2b)
+        {
+            int lidx = GetChannelIdxByName(device->RealOut, FrontLeft);
+            int ridx = GetChannelIdxByName(device->RealOut, FrontRight);
+            if(lidx != -1 && ridx != -1)
             {
                 /* Apply binaural/crossfeed filter */
-                for(i = 0;i < SamplesToDo;i++)
-                {
-                    float samples[2];
-                    samples[0] = device->RealOut.Buffer[0][i];
-                    samples[1] = device->RealOut.Buffer[1][i];
-                    bs2b_cross_feed(device->Bs2b, samples);
-                    device->RealOut.Buffer[0][i] = samples[0];
-                    device->RealOut.Buffer[1][i] = samples[1];
-                }
+                bs2b_cross_feed(device->Bs2b, device->RealOut.Buffer[lidx],
+                                device->RealOut.Buffer[ridx], SamplesToDo);
             }
         }
 
diff --git a/Alc/bs2b.c b/Alc/bs2b.c
index 6c3f052b..ddc2e2f2 100644
--- a/Alc/bs2b.c
+++ b/Alc/bs2b.c
@@ -129,4 +129,59 @@ void bs2b_clear(struct bs2b *bs2b)
     memset(&bs2b->last_sample, 0, sizeof(bs2b->last_sample));
 } /* bs2b_clear */
 
-extern inline void bs2b_cross_feed(struct bs2b *bs2b, float *restrict samples);
+void bs2b_cross_feed(struct bs2b *bs2b, float *restrict Left, float *restrict Right, unsigned int SamplesToDo)
+{
+    float lsamples[128][2];
+    float rsamples[128][2];
+    unsigned int base;
+
+    for(base = 0;base < SamplesToDo;)
+    {
+        unsigned int todo = minu(128, SamplesToDo-base);
+        unsigned int i;
+
+        /* Process left input */
+        lsamples[0][0] = bs2b->a0_lo*Left[0] +
+                         bs2b->b1_lo*bs2b->last_sample[0].lo;
+        lsamples[0][1] = bs2b->a0_hi*Left[0] +
+                         bs2b->a1_hi*bs2b->last_sample[0].asis +
+                         bs2b->b1_hi*bs2b->last_sample[0].hi;
+        for(i = 1;i < todo;i++)
+        {
+            lsamples[i][0] = bs2b->a0_lo*Left[i] +
+                             bs2b->b1_lo*lsamples[i-1][0];
+            lsamples[i][1] = bs2b->a0_hi*Left[i] +
+                             bs2b->a1_hi*Left[i-1] +
+                             bs2b->b1_hi*lsamples[i-1][1];
+        }
+        bs2b->last_sample[0].asis = Left[i-1];
+        bs2b->last_sample[0].lo = lsamples[i-1][0];
+        bs2b->last_sample[0].hi = lsamples[i-1][1];
+
+        /* Process right input */
+        rsamples[0][0] = bs2b->a0_lo*Right[0] +
+                         bs2b->b1_lo*bs2b->last_sample[1].lo;
+        rsamples[0][1] = bs2b->a0_hi*Right[0] +
+                         bs2b->a1_hi*bs2b->last_sample[1].asis +
+                         bs2b->b1_hi*bs2b->last_sample[1].hi;
+        for(i = 1;i < todo;i++)
+        {
+            rsamples[i][0] = bs2b->a0_lo*Right[i] +
+                             bs2b->b1_lo*rsamples[i-1][0];
+            rsamples[i][1] = bs2b->a0_hi*Right[i] +
+                             bs2b->a1_hi*Right[i-1] +
+                             bs2b->b1_hi*rsamples[i-1][1];
+        }
+        bs2b->last_sample[1].asis = Right[i-1];
+        bs2b->last_sample[1].lo = rsamples[i-1][0];
+        bs2b->last_sample[1].hi = rsamples[i-1][1];
+
+        /* Crossfeed */
+        for(i = 0;i < todo;i++)
+            *(Left++) = lsamples[i][1] + rsamples[i][0];
+        for(i = 0;i < todo;i++)
+            *(Right++) = rsamples[i][1] + lsamples[i][0];
+
+        base += todo;
+    }
+} /* bs2b_cross_feed */
diff --git a/OpenAL32/Include/bs2b.h b/OpenAL32/Include/bs2b.h
index 903c6bc5..bfe5c274 100644
--- a/OpenAL32/Include/bs2b.h
+++ b/OpenAL32/Include/bs2b.h
@@ -63,10 +63,10 @@ struct bs2b {
      * [0] - first channel, [1] - second channel
      */
     struct t_last_sample {
-        float asis[2];
-        float lo[2];
-        float hi[2];
-    } last_sample;
+        float asis;
+        float lo;
+        float hi;
+    } last_sample[2];
 };
 
 /* Clear buffers and set new coefficients with new crossfeed level and sample
@@ -85,38 +85,7 @@ int bs2b_get_srate(struct bs2b *bs2b);
 /* Clear buffer */
 void bs2b_clear(struct bs2b *bs2b);
 
-/* Crossfeeds one stereo sample that are pointed by sample.
- * [0] - first channel, [1] - second channel.
- * Returns crossfided sample by sample pointer.
- */
-inline void bs2b_cross_feed(struct bs2b *bs2b, float *restrict sample)
-{
-/* Single pole IIR filter.
- * O[n] = a0*I[n] + a1*I[n-1] + b1*O[n-1]
- */
-
-/* Lowpass filter */
-#define lo_filter(in, out_1) (bs2b->a0_lo*(in) + bs2b->b1_lo*(out_1))
-
-/* Highboost filter */
-#define hi_filter(in, in_1, out_1) (bs2b->a0_hi*(in) + bs2b->a1_hi*(in_1) + bs2b->b1_hi*(out_1))
-
-    /* Lowpass filter */
-    bs2b->last_sample.lo[0] = lo_filter(sample[0], bs2b->last_sample.lo[0]);
-    bs2b->last_sample.lo[1] = lo_filter(sample[1], bs2b->last_sample.lo[1]);
-
-    /* Highboost filter */
-    bs2b->last_sample.hi[0] = hi_filter(sample[0], bs2b->last_sample.asis[0], bs2b->last_sample.hi[0]);
-    bs2b->last_sample.hi[1] = hi_filter(sample[1], bs2b->last_sample.asis[1], bs2b->last_sample.hi[1]);
-    bs2b->last_sample.asis[0] = sample[0];
-    bs2b->last_sample.asis[1] = sample[1];
-
-    /* Crossfeed */
-    sample[0] = bs2b->last_sample.hi[0] + bs2b->last_sample.lo[1];
-    sample[1] = bs2b->last_sample.hi[1] + bs2b->last_sample.lo[0];
-#undef hi_filter
-#undef lo_filter
-} /* bs2b_cross_feed */
+void bs2b_cross_feed(struct bs2b *bs2b, float *restrict Left, float *restrict Right, unsigned int SamplesToDo);
 
 #ifdef __cplusplus
 }    /* extern "C" */