admin/webrtc_m130
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Removed the aec state as an input parameter to the FilterFar function.

Browse Source 
BUG=webrtc:5201

Review URL: https://codereview.webrtc.org/1454983006

Cr-Commit-Position: refs/heads/master@{#10787}
This commit is contained in:
peah 2015-11-25 07:43:12 -08:00 committed by Commit bot
parent 880896ab09
commit 54eb5e2e9a
5 changed files with 84 additions and 59 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

38
webrtc/modules/audio_processing/aec/aec_core.c
View File

@ -151,26 +151,30 @@ static int CmpFloat(const void* a, const void* b) {
return (*da > *db) - (*da < *db);
}
static void FilterFar(AecCore* aec, float yf[2][PART_LEN1]) {
static void FilterFar(int num_partitions,
int x_fft_buffer_block_pos,
float x_fft_buffer[2][kExtendedNumPartitions * PART_LEN1],
float h_fft_buffer[2][kExtendedNumPartitions * PART_LEN1],
float y_fft[2][PART_LEN1]) {
int i;
for (i = 0; i < aec->num_partitions; i++) {
for (i = 0; i < num_partitions; i++) {
int j;
int xPos = (i + aec->xfBufBlockPos) * PART_LEN1;
int x_pos = (i + x_fft_buffer_block_pos) * PART_LEN1;
int pos = i * PART_LEN1;
// Check for wrap
if (i + aec->xfBufBlockPos >= aec->num_partitions) {
xPos -= aec->num_partitions * (PART_LEN1);
// Check for wrapped buffer.
if (i + x_fft_buffer_block_pos >= num_partitions) {
x_pos -= num_partitions * (PART_LEN1);
}
for (j = 0; 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]);
y_fft[0][j] += MulRe(x_fft_buffer[0][x_pos + j],
x_fft_buffer[1][x_pos + j],
h_fft_buffer[0][pos + j],
h_fft_buffer[1][pos + j]);
y_fft[1][j] += MulIm(x_fft_buffer[0][x_pos + j],
x_fft_buffer[1][x_pos + j],
h_fft_buffer[0][pos + j],
h_fft_buffer[1][pos + j]);
}
}
}
@ -971,7 +975,11 @@ static void EchoSubtraction(AecCore* aec,
memset(yf, 0, sizeof(yf));
// Produce frequency domain echo estimate.
WebRtcAec_FilterFar(aec, yf);
WebRtcAec_FilterFar(aec->num_partitions,
aec->xfBufBlockPos,
aec->xfBuf,
aec->wfBuf,
yf);
// Inverse fft to obtain echo estimate and error.
FrequencyToTime(yf, fft);

7
webrtc/modules/audio_processing/aec/aec_core_internal.h
View File

@ -170,7 +170,12 @@ struct AecCore {
#endif
};
typedef void (*WebRtcAecFilterFar)(AecCore* aec, float yf[2][PART_LEN1]);
typedef void (*WebRtcAecFilterFar)(
int num_partitions,
int x_fft_buffer_block_pos,
float x_fft_buffer[2][kExtendedNumPartitions * PART_LEN1],
float h_fft_buffer[2][kExtendedNumPartitions * PART_LEN1],
float y_fft[2][PART_LEN1]);
extern WebRtcAecFilterFar WebRtcAec_FilterFar;
typedef void (*WebRtcAecScaleErrorSignal)(int extended_filter_enabled,
float normal_mu,

27
webrtc/modules/audio_processing/aec/aec_core_mips.c
View File

@ -320,21 +320,26 @@ void WebRtcAec_ComfortNoise_mips(AecCore* aec,
}
}
void WebRtcAec_FilterFar_mips(AecCore* aec, float yf[2][PART_LEN1]) {
void WebRtcAec_FilterFar_mips(
int num_partitions,
int xfBufBlockPos,
float xfBuf[2][kExtendedNumPartitions * PART_LEN1],
float wfBuf[2][kExtendedNumPartitions * PART_LEN1],
float yf[2][PART_LEN1]) {
int i;
for (i = 0; i < aec->num_partitions; i++) {
int xPos = (i + aec->xfBufBlockPos) * PART_LEN1;
for (i = 0; i < num_partitions; i++) {
int xPos = (i + xfBufBlockPos) * PART_LEN1;
int pos = i * PART_LEN1;
// Check for wrap
if (i + aec->xfBufBlockPos >= aec->num_partitions) {
xPos -= aec->num_partitions * (PART_LEN1);
if (i + xfBufBlockPos >= num_partitions) {
xPos -= num_partitions * (PART_LEN1);
}
float* yf0 = yf[0];
float* yf1 = yf[1];
float* aRe = aec->xfBuf[0] + xPos;
float* aIm = aec->xfBuf[1] + xPos;
float* bRe = aec->wfBuf[0] + pos;
float* bIm = aec->wfBuf[1] + pos;
float* aRe = xfBuf[0] + xPos;
float* aIm = xfBuf[1] + xPos;
float* bRe = wfBuf[0] + pos;
float* bIm = wfBuf[1] + pos;
float f0, f1, f2, f3, f4, f5, f6, f7, f8, f9, f10, f11, f12, f13;
int len = PART_LEN1 >> 1;
@ -722,7 +727,7 @@ void WebRtcAec_ScaleErrorSignal_mips(int extended_filter_enabled,
".set push \n\t"
".set noreorder \n\t"
"1: \n\t"
"lwc1 %[f0], 0(%[x_pow]) \n\t"
"lwc1 %[f0], 0(%[x_pow]) \n\t"
"lwc1 %[f1], 0(%[ef0]) \n\t"
"lwc1 %[f2], 0(%[ef1]) \n\t"
"add.s %[f0], %[f0], %[fac1] \n\t"
@ -750,7 +755,7 @@ void WebRtcAec_ScaleErrorSignal_mips(int extended_filter_enabled,
"swc1 %[f1], 0(%[ef0]) \n\t"
"swc1 %[f2], 0(%[ef1]) \n\t"
"addiu %[len], %[len], -1 \n\t"
"addiu %[x_pow], %[x_pow], 4 \n\t"
"addiu %[x_pow], %[x_pow], 4 \n\t"
"addiu %[ef0], %[ef0], 4 \n\t"
"bgtz %[len], 1b \n\t"
" addiu %[ef1], %[ef1], 4 \n\t"

35
webrtc/modules/audio_processing/aec/aec_core_neon.c
View File

@ -34,24 +34,27 @@ __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++) {
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) {
if (i + xfBufBlockPos >= num_partitions) {
xPos -= num_partitions * 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 +68,14 @@ 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]);
}
}
}

36
webrtc/modules/audio_processing/aec/aec_core_sse2.c
View File

@ -29,24 +29,28 @@ __inline static float MulIm(float aRe, float aIm, float bRe, float bIm) {
return aRe * bIm + aIm * bRe;
}
static void FilterFarSSE2(AecCore* aec, float yf[2][PART_LEN1]) {
static void FilterFarSSE2(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++) {
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) {
if (i + xfBufBlockPos >= num_partitions) {
xPos -= num_partitions * (PART_LEN1);
}
// vectorized code (four at once)
for (j = 0; j + 3 < PART_LEN1; j += 4) {
const __m128 xfBuf_re = _mm_loadu_ps(&aec->xfBuf[0][xPos + j]);
const __m128 xfBuf_im = _mm_loadu_ps(&aec->xfBuf[1][xPos + j]);
const __m128 wfBuf_re = _mm_loadu_ps(&aec->wfBuf[0][pos + j]);
const __m128 wfBuf_im = _mm_loadu_ps(&aec->wfBuf[1][pos + j]);
const __m128 xfBuf_re = _mm_loadu_ps(&xfBuf[0][xPos + j]);
const __m128 xfBuf_im = _mm_loadu_ps(&xfBuf[1][xPos + j]);
const __m128 wfBuf_re = _mm_loadu_ps(&wfBuf[0][pos + j]);
const __m128 wfBuf_im = _mm_loadu_ps(&wfBuf[1][pos + j]);
const __m128 yf_re = _mm_loadu_ps(&yf[0][j]);
const __m128 yf_im = _mm_loadu_ps(&yf[1][j]);
const __m128 a = _mm_mul_ps(xfBuf_re, wfBuf_re);
@ -62,14 +66,14 @@ static void FilterFarSSE2(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]);
}
}
}