ref: a1ae821c964960ed659134e8f834ba31a3db1a6b
parent: 1d7dea17d56be2a4a79725dd3ab7c906e2793c2b
author: Linfeng Zhang <[email protected]>
date: Wed Sep 7 11:29:03 EDT 2016
Replace call of celt_inner_prod_c() (step 1) Should call celt_inner_prod(). This change is bit exact as original, except for x86 floating-point. In x86 floating-point, it calls celt_inner_prod_sse() which may have different output with the change of floating-point operations' orders. Change-Id: Ia2381e2e198a84296ac28305183f15be842b3454 Signed-off-by: Jean-Marc Valin <[email protected]>
--- a/celt/bands.c
+++ b/celt/bands.c
@@ -92,10 +92,11 @@
#ifdef FIXED_POINT
/* Compute the amplitude (sqrt energy) in each of the bands */
-void compute_band_energies(const CELTMode *m, const celt_sig *X, celt_ener *bandE, int end, int C, int LM)
+void compute_band_energies(const CELTMode *m, const celt_sig *X, celt_ener *bandE, int end, int C, int LM, int arch)
{
int i, c, N;
const opus_int16 *eBands = m->eBands;
+ (void)arch;
N = m->shortMdctSize<<LM;
c=0; do {
for (i=0;i<end;i++)
@@ -155,7 +156,7 @@
#else /* FIXED_POINT */
/* Compute the amplitude (sqrt energy) in each of the bands */
-void compute_band_energies(const CELTMode *m, const celt_sig *X, celt_ener *bandE, int end, int C, int LM)
+void compute_band_energies(const CELTMode *m, const celt_sig *X, celt_ener *bandE, int end, int C, int LM, int arch)
{
int i, c, N;
const opus_int16 *eBands = m->eBands;
@@ -164,7 +165,7 @@
for (i=0;i<end;i++)
{
opus_val32 sum;
- sum = 1e-27f + celt_inner_prod_c(&X[c*N+(eBands[i]<<LM)], &X[c*N+(eBands[i]<<LM)], (eBands[i+1]-eBands[i])<<LM);
+ sum = 1e-27f + celt_inner_prod(&X[c*N+(eBands[i]<<LM)], &X[c*N+(eBands[i]<<LM)], (eBands[i+1]-eBands[i])<<LM, arch);
bandE[i+c*m->nbEBands] = celt_sqrt(sum);
/*printf ("%f ", bandE[i+c*m->nbEBands]);*/
}
--- a/celt/bands.h
+++ b/celt/bands.h
@@ -44,7 +44,7 @@
* @param X Spectrum
* @param bandE Square root of the energy for each band (returned)
*/
-void compute_band_energies(const CELTMode *m, const celt_sig *X, celt_ener *bandE, int end, int C, int LM);
+void compute_band_energies(const CELTMode *m, const celt_sig *X, celt_ener *bandE, int end, int C, int LM, int arch);
/*void compute_noise_energies(const CELTMode *m, const celt_sig *X, const opus_val16 *tonality, celt_ener *bandE);*/
--- a/celt/celt_encoder.c
+++ b/celt/celt_encoder.c
@@ -1655,7 +1655,7 @@
if (secondMdct)
{
compute_mdcts(mode, 0, in, freq, C, CC, LM, st->upsample, st->arch);
- compute_band_energies(mode, freq, bandE, effEnd, C, LM);
+ compute_band_energies(mode, freq, bandE, effEnd, C, LM, st->arch);
amp2Log2(mode, effEnd, end, bandE, bandLogE2, C);
for (i=0;i<C*nbEBands;i++)
bandLogE2[i] += HALF16(SHL16(LM, DB_SHIFT));
@@ -1664,7 +1664,7 @@
compute_mdcts(mode, shortBlocks, in, freq, C, CC, LM, st->upsample, st->arch);
if (CC==2&&C==1)
tf_chan = 0;
- compute_band_energies(mode, freq, bandE, effEnd, C, LM);
+ compute_band_energies(mode, freq, bandE, effEnd, C, LM, st->arch);
if (st->lfe)
{
@@ -1788,7 +1788,7 @@
isTransient = 1;
shortBlocks = M;
compute_mdcts(mode, shortBlocks, in, freq, C, CC, LM, st->upsample, st->arch);
- compute_band_energies(mode, freq, bandE, effEnd, C, LM);
+ compute_band_energies(mode, freq, bandE, effEnd, C, LM, st->arch);
amp2Log2(mode, effEnd, end, bandE, bandLogE, C);
/* Compensate for the scaling of short vs long mdcts */
for (i=0;i<C*nbEBands;i++)
--- a/celt/pitch.c
+++ b/celt/pitch.c
@@ -378,7 +378,7 @@
for (j=0;j<len>>1;j++)
sum += SHR32(MULT16_16(x_lp[j],y[i+j]), shift);
#else
- sum = celt_inner_prod_c(x_lp, y+i, len>>1);
+ sum = celt_inner_prod(x_lp, y+i, len>>1, arch);
#endif
xcorr[i] = MAX32(-1, sum);
#ifdef FIXED_POINT
--- a/src/opus_multistream_encoder.c
+++ b/src/opus_multistream_encoder.c
@@ -326,7 +326,7 @@
freq[i] = 0;
}
- compute_band_energies(celt_mode, freq, tmpE, 21, 1, LM);
+ compute_band_energies(celt_mode, freq, tmpE, 21, 1, LM, arch);
/* If we have multiple frames, take the max energy. */
for (i=0;i<21;i++)
bandE[i] = MAX32(bandE[i], tmpE[i]);