1-channel SIMD code can coerce alignment in one special case, too.

icculus · Apr 25, 2018 · a3bb60f · a3bb60f
1 parent d110ee9
commit a3bb60f
Showing 1 changed file with 22 additions and 10 deletions.
diff --git a/mojoal.c b/mojoal.c
@@ -734,7 +734,16 @@ static void mix_float32_c1_sse(const ALfloat * restrict panning, const float * r
     const int leftover = mixframes % 8;
     ALsizei i;
 
-    if ( (((size_t)stream) % 16) || (((size_t)data) % 16) ) {
+    /* We can align this to 16 in one special case. */
+    if ( ((((size_t)data) % 16) == 8) && ((((size_t)stream) % 16) == 0) && (mixframes >= 2) ) {
+        stream[0] += data[0] * left;
+        stream[1] += data[0] * right;
+        stream[2] += data[1] * left;
+        stream[3] += data[1] * right;
+        stream += 4;
+        data += 2;
+        mix_float32_c1_sse(panning, data + 2, stream + 2, mixframes - 2);
+    } else if ( (((size_t)stream) % 16) || (((size_t)data) % 16) ) {
         /* unaligned, do scalar version. */
         mix_float32_c1_scalar(panning, data, stream, mixframes);
     } else if ((left == 1.0f) && (right == 1.0f)) {
@@ -796,10 +805,7 @@ static void mix_float32_c2_sse(const ALfloat * restrict panning, const float * r
         stream += 2;
         data += 2;
         mix_float32_c2_sse(panning, data + 2, stream + 2, mixframes - 1);
-        return;
-    }
-
-    if ( (((size_t)stream) % 16) || (((size_t)data) % 16) ) {
+    } else if ( (((size_t)stream) % 16) || (((size_t)data) % 16) ) {
         /* unaligned, do scalar version. */
         mix_float32_c2_scalar(panning, data, stream, mixframes);
     } else if ((left == 1.0f) && (right == 1.0f)) {
@@ -843,7 +849,16 @@ static void mix_float32_c1_neon(const ALfloat * restrict panning, const float *
     const int leftover = mixframes % 8;
     ALsizei i;
 
-    if ( (((size_t)stream) % 16) || (((size_t)data) % 16) ) {
+    /* We can align this to 16 in one special case. */
+    if ( ((((size_t)data) % 16) == 8) && ((((size_t)stream) % 16) == 0) && (mixframes >= 2) ) {
+        stream[0] += data[0] * left;
+        stream[1] += data[0] * right;
+        stream[2] += data[1] * left;
+        stream[3] += data[1] * right;
+        stream += 4;
+        data += 2;
+        mix_float32_c1_neon(panning, data + 2, stream + 2, mixframes - 2);
+    } else if ( (((size_t)stream) % 16) || (((size_t)data) % 16) ) {
         /* unaligned, do scalar version. */
         mix_float32_c1_scalar(panning, data, stream, mixframes);
     } else if ((left == 1.0f) && (right == 1.0f)) {
@@ -905,10 +920,7 @@ static void mix_float32_c2_neon(const ALfloat * restrict panning, const float *
         stream += 2;
         data += 2;
         mix_float32_c2_neon(panning, data + 2, stream + 2, mixframes - 1);
-        return;
-    }
-
-    if ( (((size_t)stream) % 16) || (((size_t)data) % 16) ) {
+    } else if ( (((size_t)stream) % 16) || (((size_t)data) % 16) ) {
         /* unaligned, do scalar version. */
         mix_float32_c2_scalar(panning, data, stream, mixframes);
     } else if ((left == 1.0f) && (right == 1.0f)) {