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

webrtc/modules/audio_processing/aec/aec_core_neon.c

 /*
  *  Copyright (c) 2014 The WebRTC project authors. All Rights Reserved.
  *
  *  Use of this source code is governed by a BSD-style license
  *  that can be found in the LICENSE file in the root of the source
  *  tree. An additional intellectual property rights grant can be found
  *  in the file PATENTS.  All contributing project authors may
  *  be found in the AUTHORS file in the root of the source tree.
  */
 
@@ skipping 16 matching lines @@
 enum { kFloatExponentShift = 23 };
 
 __inline static float MulRe(float aRe, float aIm, float bRe, float bIm) {
   return aRe * bRe - aIm * bIm;
 }
 
 __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);
       const float32x4_t e = vmlsq_f32(a, xfBuf_im, wfBuf_im);
       const float32x4_t c = vmulq_f32(xfBuf_re, wfBuf_im);
       const float32x4_t f = vmlaq_f32(c, xfBuf_im, wfBuf_re);
       const float32x4_t g = vaddq_f32(yf_re, e);
       const float32x4_t h = vaddq_f32(yf_im, f);
       vst1q_f32(&yf[0][j], g);
       vst1q_f32(&yf[1][j], h);
     }
     // 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]);
     }
   }
 }
 
 // ARM64's arm_neon.h has already defined vdivq_f32 vsqrtq_f32.
 #if !defined (WEBRTC_ARCH_ARM64)
 static float32x4_t vdivq_f32(float32x4_t a, float32x4_t b) {
   int i;
   float32x4_t x = vrecpeq_f32(b);
   // from arm documentation
@@ skipping 644 matching lines @@
   }
 }
 
 void WebRtcAec_InitAec_neon(void) {
   WebRtcAec_FilterFar = FilterFarNEON;
   WebRtcAec_ScaleErrorSignal = ScaleErrorSignalNEON;
   WebRtcAec_FilterAdaptation = FilterAdaptationNEON;
   WebRtcAec_OverdriveAndSuppress = OverdriveAndSuppressNEON;
   WebRtcAec_SubbandCoherence = SubbandCoherenceNEON;
 }