Ducking fix #2: Removed the aec state as an input parameter to the FilterFar function.

webrtc/modules/audio_processing/aec/aec_core_neon.c

Index: webrtc/modules/audio_processing/aec/aec_core_neon.c
diff --git a/webrtc/modules/audio_processing/aec/aec_core_neon.c b/webrtc/modules/audio_processing/aec/aec_core_neon.c
index c8c5c7e3040e46f2407687f80c3a3f7a10bfc864..c911a7fa2006bf88af3e692f64c76842b711bbad 100644
--- a/webrtc/modules/audio_processing/aec/aec_core_neon.c
+++ b/webrtc/modules/audio_processing/aec/aec_core_neon.c
@@ -34,24 +34,28 @@ __inline static float MulIm(float aRe, float aIm, float bRe, float bIm) {
   return aRe * bIm + aIm * bRe;
 }
 
-static void FilterFarNEON(AecCore* aec, float yf[2][PART_LEN1]) {
+static void FilterFarNEON(int num_partitions,
+                          int xfBufBlockPos,
+                          float xfBuf[2][kExtendedNumPartitions * PART_LEN1],
+                          float wfBuf[2][kExtendedNumPartitions * PART_LEN1],
+                          float yf[2][PART_LEN1]) {
   int i;
-  const int num_partitions = aec->num_partitions;
-  for (i = 0; i < num_partitions; i++) {
+  const int num_partitions_local = num_partitions;

[hlundin-webrtc, 2015/11/20 10:59:40]

You don't need the local variant, do you?

[peah-webrtc, 2015/11/24 06:48:56]

I'm not 100% sure. It is const-declared, and this is code that is supposed to be
optimized for Neon, so in theory it could have an impact.

I agree with you, however, and don't think that it will matter, so I'll remove
the local variable. But please let me know if you think (in light of the above)
that is not the correct thing to do!

+  for (i = 0; i < num_partitions_local; i++) {
     int j;
-    int xPos = (i + aec->xfBufBlockPos) * PART_LEN1;
+    int xPos = (i + xfBufBlockPos) * PART_LEN1;
     int pos = i * PART_LEN1;
     // Check for wrap
-    if (i + aec->xfBufBlockPos >= num_partitions) {
-      xPos -= num_partitions * PART_LEN1;
+    if (i + xfBufBlockPos >= num_partitions_local) {
+      xPos -= num_partitions_local * PART_LEN1;
     }
 
     // vectorized code (four at once)
     for (j = 0; j + 3 < PART_LEN1; j += 4) {
-      const float32x4_t xfBuf_re = vld1q_f32(&aec->xfBuf[0][xPos + j]);
-      const float32x4_t xfBuf_im = vld1q_f32(&aec->xfBuf[1][xPos + j]);
-      const float32x4_t wfBuf_re = vld1q_f32(&aec->wfBuf[0][pos + j]);
-      const float32x4_t wfBuf_im = vld1q_f32(&aec->wfBuf[1][pos + j]);
+      const float32x4_t xfBuf_re = vld1q_f32(&xfBuf[0][xPos + j]);
+      const float32x4_t xfBuf_im = vld1q_f32(&xfBuf[1][xPos + j]);
+      const float32x4_t wfBuf_re = vld1q_f32(&wfBuf[0][pos + j]);
+      const float32x4_t wfBuf_im = vld1q_f32(&wfBuf[1][pos + j]);
       const float32x4_t yf_re = vld1q_f32(&yf[0][j]);
       const float32x4_t yf_im = vld1q_f32(&yf[1][j]);
       const float32x4_t a = vmulq_f32(xfBuf_re, wfBuf_re);
@@ -65,14 +69,14 @@ static void FilterFarNEON(AecCore* aec, float yf[2][PART_LEN1]) {
     }
     // scalar code for the remaining items.
     for (; j < PART_LEN1; j++) {
-      yf[0][j] += MulRe(aec->xfBuf[0][xPos + j],
-                        aec->xfBuf[1][xPos + j],
-                        aec->wfBuf[0][pos + j],
-                        aec->wfBuf[1][pos + j]);
-      yf[1][j] += MulIm(aec->xfBuf[0][xPos + j],
-                        aec->xfBuf[1][xPos + j],
-                        aec->wfBuf[0][pos + j],
-                        aec->wfBuf[1][pos + j]);
+      yf[0][j] += MulRe(xfBuf[0][xPos + j],
+                        xfBuf[1][xPos + j],
+                        wfBuf[0][pos + j],
+                        wfBuf[1][pos + j]);
+      yf[1][j] += MulIm(xfBuf[0][xPos + j],
+                        xfBuf[1][xPos + j],
+                        wfBuf[0][pos + j],
+                        wfBuf[1][pos + j]);
     }
   }
 }