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 c911a7fa2006bf88af3e692f64c76842b711bbad..78af18cf833b2d1809026ea80701b64ba9359444 100644 |
--- a/webrtc/modules/audio_processing/aec/aec_core_neon.c |
+++ b/webrtc/modules/audio_processing/aec/aec_core_neon.c |
@@ -129,7 +129,7 @@ static float32x4_t vsqrtq_f32(float32x4_t s) { |
static void ScaleErrorSignalNEON(int extended_filter_enabled, |
float normal_mu, |
float normal_error_threshold, |
- float *xPow, |
+ float xPow[PART_LEN1], |
float ef[2][PART_LEN1]) { |
const float mu = extended_filter_enabled ? kExtendedMu : normal_mu; |
const float error_threshold = extended_filter_enabled ? |
@@ -186,25 +186,29 @@ static void ScaleErrorSignalNEON(int extended_filter_enabled, |
} |
} |
-static void FilterAdaptationNEON(AecCore* aec, |
- float* fft, |
- float ef[2][PART_LEN1]) { |
+static void FilterAdaptationNEON( |
+ int num_partitions, |
+ int xfBufBlockPos, |
+ float xfBuf[2][kExtendedNumPartitions * PART_LEN1], |
+ float ef[2][PART_LEN1], |
+ float wfBuf[2][kExtendedNumPartitions * PART_LEN1]) { |
+ float fft[PART_LEN2]; |
int i; |
- const int num_partitions = aec->num_partitions; |
+ const int num_partitions = num_partitions; |
for (i = 0; i < num_partitions; i++) { |
- int xPos = (i + aec->xfBufBlockPos) * PART_LEN1; |
+ int xPos = (i + xfBufBlockPos) * PART_LEN1; |
int pos = i * PART_LEN1; |
int j; |
// Check for wrap |
- if (i + aec->xfBufBlockPos >= num_partitions) { |
+ if (i + xfBufBlockPos >= num_partitions) { |
xPos -= num_partitions * PART_LEN1; |
} |
// Process the whole array... |
for (j = 0; j < PART_LEN; j += 4) { |
// Load xfBuf and ef. |
- 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 xfBuf_re = vld1q_f32(&xfBuf[0][xPos + j]); |
+ const float32x4_t xfBuf_im = vld1q_f32(&xfBuf[1][xPos + j]); |
const float32x4_t ef_re = vld1q_f32(&ef[0][j]); |
const float32x4_t ef_im = vld1q_f32(&ef[1][j]); |
// Calculate the product of conjugate(xfBuf) by ef. |
@@ -221,8 +225,8 @@ static void FilterAdaptationNEON(AecCore* aec, |
vst1q_f32(&fft[2 * j + 4], g_n_h.val[1]); |
} |
// ... and fixup the first imaginary entry. |
- fft[1] = MulRe(aec->xfBuf[0][xPos + PART_LEN], |
- -aec->xfBuf[1][xPos + PART_LEN], |
+ fft[1] = MulRe(xfBuf[0][xPos + PART_LEN], |
+ -xfBuf[1][xPos + PART_LEN], |
ef[0][PART_LEN], |
ef[1][PART_LEN]); |
@@ -242,21 +246,21 @@ static void FilterAdaptationNEON(AecCore* aec, |
aec_rdft_forward_128(fft); |
{ |
- const float wt1 = aec->wfBuf[1][pos]; |
- aec->wfBuf[0][pos + PART_LEN] += fft[1]; |
+ const float wt1 = wfBuf[1][pos]; |
+ wfBuf[0][pos + PART_LEN] += fft[1]; |
for (j = 0; j < PART_LEN; j += 4) { |
- float32x4_t wtBuf_re = vld1q_f32(&aec->wfBuf[0][pos + j]); |
- float32x4_t wtBuf_im = vld1q_f32(&aec->wfBuf[1][pos + j]); |
+ float32x4_t wtBuf_re = vld1q_f32(&wfBuf[0][pos + j]); |
+ float32x4_t wtBuf_im = vld1q_f32(&wfBuf[1][pos + j]); |
const float32x4_t fft0 = vld1q_f32(&fft[2 * j + 0]); |
const float32x4_t fft4 = vld1q_f32(&fft[2 * j + 4]); |
const float32x4x2_t fft_re_im = vuzpq_f32(fft0, fft4); |
wtBuf_re = vaddq_f32(wtBuf_re, fft_re_im.val[0]); |
wtBuf_im = vaddq_f32(wtBuf_im, fft_re_im.val[1]); |
- vst1q_f32(&aec->wfBuf[0][pos + j], wtBuf_re); |
- vst1q_f32(&aec->wfBuf[1][pos + j], wtBuf_im); |
+ vst1q_f32(&wfBuf[0][pos + j], wtBuf_re); |
+ vst1q_f32(&wfBuf[1][pos + j], wtBuf_im); |
} |
- aec->wfBuf[1][pos] = wt1; |
+ wfBuf[1][pos] = wt1; |
} |
} |
} |