aec_rdft_128: one entry point for each sign.

Review URL: http://webrtc-codereview.appspot.com/61007

git-svn-id: http://webrtc.googlecode.com/svn/trunk@153 4adac7df-926f-26a2-2b94-8c16560cd09d
This commit is contained in:
cduvivier@google.com 2011-07-06 23:35:37 +00:00
parent fae3b31707
commit d0159d8eb0
4 changed files with 35 additions and 29 deletions

View File

@ -290,7 +290,7 @@ static void FilterAdaptation(aec_t *aec, float *fft, float ef[2][PART_LEN1]) {
-aec->xfBuf[1][xPos + PART_LEN], -aec->xfBuf[1][xPos + PART_LEN],
ef[0][PART_LEN], ef[1][PART_LEN]); ef[0][PART_LEN], ef[1][PART_LEN]);
aec_rdft_128(-1, fft); aec_rdft_inverse_128(fft);
memset(fft + PART_LEN, 0, sizeof(float) * PART_LEN); memset(fft + PART_LEN, 0, sizeof(float) * PART_LEN);
// fft scaling // fft scaling
@ -300,7 +300,7 @@ static void FilterAdaptation(aec_t *aec, float *fft, float ef[2][PART_LEN1]) {
fft[j] *= scale; fft[j] *= scale;
} }
} }
aec_rdft_128(1, fft); aec_rdft_forward_128(fft);
aec->wfBuf[0][pos] += fft[0]; aec->wfBuf[0][pos] += fft[0];
aec->wfBuf[0][pos + PART_LEN] += fft[1]; aec->wfBuf[0][pos + PART_LEN] += fft[1];
@ -609,7 +609,7 @@ static void ProcessBlock(aec_t *aec, const short *farend,
memcpy(aec->dBufH + PART_LEN, dH, sizeof(float) * PART_LEN); memcpy(aec->dBufH + PART_LEN, dH, sizeof(float) * PART_LEN);
} }
aec_rdft_128(1, fft); aec_rdft_forward_128(fft);
// Far fft // Far fft
xf[1][0] = 0; xf[1][0] = 0;
@ -624,7 +624,7 @@ static void ProcessBlock(aec_t *aec, const short *farend,
// Near fft // Near fft
memcpy(fft, aec->dBuf, sizeof(float) * PART_LEN2); memcpy(fft, aec->dBuf, sizeof(float) * PART_LEN2);
aec_rdft_128(1, fft); aec_rdft_forward_128(fft);
df[0][1] = 0; df[0][1] = 0;
df[PART_LEN][1] = 0; df[PART_LEN][1] = 0;
df[0][0] = fft[0]; df[0][0] = fft[0];
@ -700,7 +700,7 @@ static void ProcessBlock(aec_t *aec, const short *farend,
fft[2 * i] = yf[0][i]; fft[2 * i] = yf[0][i];
fft[2 * i + 1] = yf[1][i]; fft[2 * i + 1] = yf[1][i];
} }
aec_rdft_128(-1, fft); aec_rdft_inverse_128(fft);
scale = 2.0f / PART_LEN2; scale = 2.0f / PART_LEN2;
for (i = 0; i < PART_LEN; i++) { for (i = 0; i < PART_LEN; i++) {
@ -715,7 +715,7 @@ static void ProcessBlock(aec_t *aec, const short *farend,
memcpy(aec->eBuf + PART_LEN, e, sizeof(float) * PART_LEN); memcpy(aec->eBuf + PART_LEN, e, sizeof(float) * PART_LEN);
memset(fft, 0, sizeof(float) * PART_LEN); memset(fft, 0, sizeof(float) * PART_LEN);
memcpy(fft + PART_LEN, e, sizeof(float) * PART_LEN); memcpy(fft + PART_LEN, e, sizeof(float) * PART_LEN);
aec_rdft_128(1, fft); aec_rdft_forward_128(fft);
ef[1][0] = 0; ef[1][0] = 0;
ef[1][PART_LEN] = 0; ef[1][PART_LEN] = 0;
@ -838,7 +838,7 @@ static void NonLinearProcessing(aec_t *aec, short *output, short *outputH)
fft[i] = aec->xBuf[i] * sqrtHanning[i]; fft[i] = aec->xBuf[i] * sqrtHanning[i];
fft[PART_LEN + i] = aec->xBuf[PART_LEN + i] * sqrtHanning[PART_LEN - i]; fft[PART_LEN + i] = aec->xBuf[PART_LEN + i] * sqrtHanning[PART_LEN - i];
} }
aec_rdft_128(1, fft); aec_rdft_forward_128(fft);
xfw[0][1] = 0; xfw[0][1] = 0;
xfw[PART_LEN][1] = 0; xfw[PART_LEN][1] = 0;
@ -860,7 +860,7 @@ static void NonLinearProcessing(aec_t *aec, short *output, short *outputH)
fft[i] = aec->dBuf[i] * sqrtHanning[i]; fft[i] = aec->dBuf[i] * sqrtHanning[i];
fft[PART_LEN + i] = aec->dBuf[PART_LEN + i] * sqrtHanning[PART_LEN - i]; fft[PART_LEN + i] = aec->dBuf[PART_LEN + i] * sqrtHanning[PART_LEN - i];
} }
aec_rdft_128(1, fft); aec_rdft_forward_128(fft);
dfw[1][0] = 0; dfw[1][0] = 0;
dfw[1][PART_LEN] = 0; dfw[1][PART_LEN] = 0;
@ -876,7 +876,7 @@ static void NonLinearProcessing(aec_t *aec, short *output, short *outputH)
fft[i] = aec->eBuf[i] * sqrtHanning[i]; fft[i] = aec->eBuf[i] * sqrtHanning[i];
fft[PART_LEN + i] = aec->eBuf[PART_LEN + i] * sqrtHanning[PART_LEN - i]; fft[PART_LEN + i] = aec->eBuf[PART_LEN + i] * sqrtHanning[PART_LEN - i];
} }
aec_rdft_128(1, fft); aec_rdft_forward_128(fft);
efw[1][0] = 0; efw[1][0] = 0;
efw[1][PART_LEN] = 0; efw[1][PART_LEN] = 0;
efw[0][0] = fft[0]; efw[0][0] = fft[0];
@ -1053,7 +1053,7 @@ static void NonLinearProcessing(aec_t *aec, short *output, short *outputH)
// Sign change required by Ooura fft. // Sign change required by Ooura fft.
fft[2*i + 1] = -efw[1][i]; fft[2*i + 1] = -efw[1][i];
} }
aec_rdft_128(-1, fft); aec_rdft_inverse_128(fft);
// Overlap and add to obtain output. // Overlap and add to obtain output.
scale = 2.0f / PART_LEN2; scale = 2.0f / PART_LEN2;
@ -1085,7 +1085,7 @@ static void NonLinearProcessing(aec_t *aec, short *output, short *outputH)
fft[2*i] = comfortNoiseHband[i][0]; fft[2*i] = comfortNoiseHband[i][0];
fft[2*i + 1] = comfortNoiseHband[i][1]; fft[2*i + 1] = comfortNoiseHband[i][1];
} }
aec_rdft_128(-1, fft); aec_rdft_inverse_128(fft);
scale = 2.0f / PART_LEN2; scale = 2.0f / PART_LEN2;
} }

View File

@ -174,7 +174,7 @@ static void FilterAdaptationSSE2(aec_t *aec, float *fft, float ef[2][PART_LEN1])
-aec->xfBuf[1][xPos + PART_LEN], -aec->xfBuf[1][xPos + PART_LEN],
ef[0][PART_LEN], ef[1][PART_LEN]); ef[0][PART_LEN], ef[1][PART_LEN]);
aec_rdft_128(-1, fft); aec_rdft_inverse_128(fft);
memset(fft + PART_LEN, 0, sizeof(float)*PART_LEN); memset(fft + PART_LEN, 0, sizeof(float)*PART_LEN);
// fft scaling // fft scaling
@ -187,7 +187,7 @@ static void FilterAdaptationSSE2(aec_t *aec, float *fft, float ef[2][PART_LEN1])
_mm_storeu_ps(&fft[j], fft_scale); _mm_storeu_ps(&fft[j], fft_scale);
} }
} }
aec_rdft_128(1, fft); aec_rdft_forward_128(fft);
{ {
float wt1 = aec->wfBuf[1][pos]; float wt1 = aec->wfBuf[1][pos];

View File

@ -477,26 +477,31 @@ static void rftbsub_128_C(float *a) {
a[65] = -a[65]; a[65] = -a[65];
} }
void aec_rdft_128(int isgn, float *a) { void aec_rdft_forward_128(float *a) {
const int n = 128; const int n = 128;
int nw; int nw;
float xi; float xi;
nw = ip[0]; nw = ip[0];
if (isgn >= 0) { bitrv2_32or128(n, ip + 2, a);
bitrv2_32or128(n, ip + 2, a); cftfsub_128(a);
cftfsub_128(a); rftfsub_128(a);
rftfsub_128(a); xi = a[0] - a[1];
xi = a[0] - a[1]; a[0] += a[1];
a[0] += a[1]; a[1] = xi;
a[1] = xi; }
} else {
a[1] = 0.5f * (a[0] - a[1]); void aec_rdft_inverse_128(float *a) {
a[0] -= a[1]; const int n = 128;
rftbsub_128(a); int nw;
bitrv2_32or128(n, ip + 2, a); float xi;
cftbsub_128(a);
} nw = ip[0];
a[1] = 0.5f * (a[0] - a[1]);
a[0] -= a[1];
rftbsub_128(a);
bitrv2_32or128(n, ip + 2, a);
cftbsub_128(a);
} }
// code path selection // code path selection

View File

@ -19,4 +19,5 @@ extern rft_sub_128_t rftbsub_128;
// entry points // entry points
void aec_rdft_init(void); void aec_rdft_init(void);
void aec_rdft_init_sse2(void); void aec_rdft_init_sse2(void);
void aec_rdft_128(int isgn, float *a); void aec_rdft_forward_128(float *a);
void aec_rdft_inverse_128(float *a);