diff --git a/celt/celt_lpc.c b/celt/celt_lpc.c
index 1861b890310b34b560219cb71213fd1a22b71c41..523c36aae12c671f1fbc0b03ae1f97d2f68a5a1b 100644
--- a/celt/celt_lpc.c
+++ b/celt/celt_lpc.c
@@ -221,7 +221,7 @@ void celt_iir(const opus_val32 *_x,
#endif
}
-void _celt_autocorr(
+int _celt_autocorr(
const opus_val16 *x, /* in: [0...n-1] samples x */
opus_val32 *ac, /* out: [0...lag-1] ac values */
const opus_val16 *window,
@@ -231,49 +231,79 @@ void _celt_autocorr(
)
{
opus_val32 d;
- int i;
+ int i, k;
int fastN=n-lag;
+ int shift;
+ const opus_val16 *xptr;
VARDECL(opus_val16, xx);
SAVE_STACK;
ALLOC(xx, n, opus_val16);
celt_assert(n>0);
celt_assert(overlap>=0);
- for (i=0;i<n;i++)
- xx[i] = x[i];
- for (i=0;i<overlap;i++)
+ if (overlap == 0)
{
- xx[i] = MULT16_16_Q15(x[i],window[i]);
- xx[n-i-1] = MULT16_16_Q15(x[n-i-1],window[i]);
+ xptr = x;
+ } else {
+ for (i=0;i<n;i++)
+ xx[i] = x[i];
+ for (i=0;i<overlap;i++)
+ {
+ xx[i] = MULT16_16_Q15(x[i],window[i]);
+ xx[n-i-1] = MULT16_16_Q15(x[n-i-1],window[i]);
+ }
+ xptr = xx;
}
+ shift=0;
#ifdef FIXED_POINT
{
opus_val32 ac0;
- int shift;
- ac0 = 1+n;
- if (n&1) ac0 += SHR32(MULT16_16(xx[0],xx[0]),9);
+ ac0 = 1+(n<<7);
+ if (n&1) ac0 += SHR32(MULT16_16(xptr[0],xptr[0]),9);
for(i=(n&1);i<n;i+=2)
{
- ac0 += SHR32(MULT16_16(xx[i],xx[i]),9);
- ac0 += SHR32(MULT16_16(xx[i+1],xx[i+1]),9);
+ ac0 += SHR32(MULT16_16(xptr[i],xptr[i]),9);
+ ac0 += SHR32(MULT16_16(xptr[i+1],xptr[i+1]),9);
}
shift = celt_ilog2(ac0)-30+10;
- shift = (shift+1)/2;
- for(i=0;i<n;i++)
- xx[i] = VSHR32(xx[i], shift);
+ shift = (shift)/2;
+ if (shift>0)
+ {
+ for(i=0;i<n;i++)
+ xx[i] = PSHR32(xptr[i], shift);
+ xptr = xx;
+ } else
+ shift = 0;
}
#endif
- celt_pitch_xcorr(xx, xx, ac, fastN, lag+1);
- while (lag>=0)
+ celt_pitch_xcorr(xptr, xptr, ac, fastN, lag+1);
+ for (k=0;k<=lag;k++)
+ {
+ for (i = k+fastN, d = 0; i < n; i++)
+ d = MAC16_16(d, xptr[i], xptr[i-k]);
+ ac[k] += d;
+ }
+#ifdef FIXED_POINT
+ shift = 2*shift;
+ if (shift<=0)
+ ac[0] += SHL32((opus_int32)1, -shift);
+ if (ac[0] < 268435456)
{
- for (i = lag+fastN, d = 0; i < n; i++)
- d = MAC16_16(d, xx[i], xx[i-lag]);
- ac[lag] += d;
- /*printf ("%f ", ac[lag]);*/
- lag--;
+ int shift2 = 29 - EC_ILOG(ac[0]);
+ for (i=0;i<=lag;i++)
+ ac[i] = SHL32(ac[i], shift2);
+ shift -= shift2;
+ } else if (ac[0] >= 536870912)
+ {
+ int shift2=1;
+ if (ac[0] >= 1073741824)
+ shift2++;
+ for (i=0;i<=lag;i++)
+ ac[i] = SHR32(ac[i], shift2);
+ shift += shift2;
}
- /*printf ("\n");*/
- ac[0] += 10;
+#endif
RESTORE_STACK;
+ return shift;
}
diff --git a/celt/celt_lpc.h b/celt/celt_lpc.h
index 2baa77edf8b8226634d726365988a32315f1b55c..19279a0ed66f2787b5968ef96373636b5e8f3120 100644
--- a/celt/celt_lpc.h
+++ b/celt/celt_lpc.h
@@ -48,6 +48,6 @@ void celt_iir(const opus_val32 *x,
int ord,
opus_val16 *mem);
-void _celt_autocorr(const opus_val16 *x, opus_val32 *ac, const opus_val16 *window, int overlap, int lag, int n);
+int _celt_autocorr(const opus_val16 *x, opus_val32 *ac, const opus_val16 *window, int overlap, int lag, int n);
#endif /* PLC_H */
diff --git a/silk/fixed/autocorr_FIX.c b/silk/fixed/autocorr_FIX.c
index 3ac6650be7e4b6e4652977bc18ec8b0b5d899e13..4cd8f6fc05916810a9cc5f345a5c80f9e588a21b 100644
--- a/silk/fixed/autocorr_FIX.c
+++ b/silk/fixed/autocorr_FIX.c
@@ -30,6 +30,7 @@ POSSIBILITY OF SUCH DAMAGE.
#endif
#include "SigProc_FIX.h"
+#include "celt_lpc.h"
/* Compute autocorrelation */
void silk_autocorr(
@@ -40,37 +41,7 @@ void silk_autocorr(
const opus_int correlationCount /* I Number of correlation taps to compute */
)
{
- opus_int i, lz, nRightShifts, corrCount;
- opus_int64 corr64;
-
+ opus_int corrCount;
corrCount = silk_min_int( inputDataSize, correlationCount );
-
- /* compute energy (zero-lag correlation) */
- corr64 = silk_inner_prod16_aligned_64( inputData, inputData, inputDataSize );
-
- /* deal with all-zero input data */
- corr64 += 1;
-
- /* number of leading zeros */
- lz = silk_CLZ64( corr64 );
-
- /* scaling: number of right shifts applied to correlations */
- nRightShifts = 35 - lz;
- *scale = nRightShifts;
-
- if( nRightShifts <= 0 ) {
- results[ 0 ] = silk_LSHIFT( (opus_int32)silk_CHECK_FIT32( corr64 ), -nRightShifts );
-
- /* compute remaining correlations based on int32 inner product */
- for( i = 1; i < corrCount; i++ ) {
- results[ i ] = silk_LSHIFT( silk_inner_prod_aligned( inputData, inputData + i, inputDataSize - i ), -nRightShifts );
- }
- } else {
- results[ 0 ] = (opus_int32)silk_CHECK_FIT32( silk_RSHIFT64( corr64, nRightShifts ) );
-
- /* compute remaining correlations based on int64 inner product */
- for( i = 1; i < corrCount; i++ ) {
- results[ i ] = (opus_int32)silk_CHECK_FIT32( silk_RSHIFT64( silk_inner_prod16_aligned_64( inputData, inputData + i, inputDataSize - i ), nRightShifts ) );
- }
- }
+ *scale = _celt_autocorr(inputData, results, NULL, 0, corrCount-1, inputDataSize);
}