search for: silk_smulwb

...t > the code where speech probability is calculated based on which SNR is > estimated. Below is the part of the code I am talking about. > > /*********************************/ > /* Speech Probability Estimation */ > /*********************************/ > SA_Q15 = silk_sigm_Q15( silk_SMULWB( VAD_SNR_FACTOR_Q16, pSNR_dB_Q7 ) - > VAD_NEGATIVE_OFFSET_Q5 ); // step1: Calculate speech probability : comment > by me > > /* Power scaling */ > if( speech_nrg <= 0 ) { // step2: update speech probability based on > speech energy : comment by me > SA_Q15 = silk_RSHIFT( SA_...

...used in opus. We are looking at the code where speech probability is calculated based on which SNR is estimated. Below is the part of the code I am talking about. /*********************************/ /* Speech Probability Estimation */ /*********************************/ SA_Q15 = silk_sigm_Q15( silk_SMULWB( VAD_SNR_FACTOR_Q16, pSNR_dB_Q7 ) - VAD_NEGATIVE_OFFSET_Q5 ); // step1: Calculate speech probability : comment by me /* Power scaling */ if( speech_nrg <= 0 ) { // step2: update speech probability based on speech energy : comment by me SA_Q15 = silk_RSHIFT( SA_Q15, 1 ); } else if( speech_nr...

...t sign-extension > e.g. 0x00008b3b), but that doesn't seem to be the case here. > The problem is even worse when opus_int is defined as 16-bit in the > platform - SA_Q15 overflows to negative right here with a similar effect > speech_nrg = silk_SQRT_APPROX(speech_nrg); > SA_Q15 = silk_SMULWB(32768 + speech_nrg, SA_Q15); > > I can't speak for the logic where the speech energy gets doubled. It > obviously seems intentional but I don't know why. Maybe so that > smoothing is performed at a constant rate regardless of whether > framesize is 10 or 20ms? > > On...

...+), 5 deletions(-) diff --git a/silk/macros.h b/silk/macros.h index 1ba614a..e1e05b9 100644 --- a/silk/macros.h +++ b/silk/macros.h @@ -48,14 +48,14 @@ POSSIBILITY OF SUCH DAMAGE. /* (a32 * (opus_int32)((opus_int16)(b32))) >> 16 output have to be 32bit int */ #if OPUS_FAST_INT64 -#define silk_SMULWB(a32, b32) (((a32) * (opus_int64)((opus_int16)(b32))) >> 16) +#define silk_SMULWB(a32, b32) ((opus_int32)(((a32) * (opus_int64)((opus_int16)(b32))) >> 16)) #else #define silk_SMULWB(a32, b32) ((((a32) >> 16) * (opus_int32)((opus_int16)(b32))) + ((...

...used in opus. We are looking at the code where speech probability is calculated based on which SNR is estimated. Below is the part of the code I am talking about. /*********************************/ /* Speech Probability Estimation */ /*********************************/ SA_Q15 = silk_sigm_Q15( silk_SMULWB( VAD_SNR_FACTOR_Q16, pSNR_dB_Q7 ) - VAD_NEGATIVE_OFFSET_Q5 ); // step1: Calculate speech probability : comment by me /* Power scaling */ if( speech_nrg <= 0 ) { // step2: update speech probability based on speech energy : comment by me SA_Q15 = silk_RSHIFT( SA_Q15, 1 ); } else if( speech_nr...

...ed in a 32-bit int without sign-extension e.g. 0x00008b3b), but that doesn't seem to be the case here. The problem is even worse when opus_int is defined as 16-bit in the platform - SA_Q15 overflows to negative right here with a similar effect speech_nrg = silk_SQRT_APPROX(speech_nrg); SA_Q15 = silk_SMULWB(32768 + speech_nrg, SA_Q15); I can't speak for the logic where the speech energy gets doubled. It obviously seems intentional but I don't know why. Maybe so that smoothing is performed at a constant rate regardless of whether framesize is 10 or 20ms? On Sun, Nov 26, 2017 at 8:07 PM, Chand...

...eech probability is calculated based on which SNR is >> estimated. Below is the part of the code I am talking about. >> >> /*********************************/ >> /* Speech Probability Estimation */ >> /*********************************/ >> SA_Q15 = silk_sigm_Q15( silk_SMULWB( VAD_SNR_FACTOR_Q16, pSNR_dB_Q7 ) - >> VAD_NEGATIVE_OFFSET_Q5 ); // step1: Calculate speech probability : comment >> by me >> >> /* Power scaling */ >> if( speech_nrg <= 0 ) { // step2: update speech probability based on >> speech energy : comment by me >&g...

...86_64__) || defined(__LP64__) || defined(_WIN64)) + /* This is an OPUS_INLINE header file for general platform. */ /* (a32 * (opus_int32)((opus_int16)(b32))) >> 16 output have to be 32bit int */ -#if defined(__x86_64__) || defined(__LP64__) || defined(_WIN64) +#if OPUS_FAST_INT64 #define silk_SMULWB(a32, b32) (((a32) * (opus_int64)((opus_int16)(b32))) >> 16) #else #define silk_SMULWB(a32, b32) ((((a32) >> 16) * (opus_int32)((opus_int16)(b32))) + ((((a32) & 0x0000FFFF) * (opus_int32)((opus_int16)(b32))) >> 16)) #endif /* a32 + (b32 * (opus_int32...

--- celt/arch.h | 5 +++++ silk/macros.h | 4 +--- 2 files changed, 6 insertions(+), 3 deletions(-) diff --git a/celt/arch.h b/celt/arch.h index 9f74ddd..670527b 100644 --- a/celt/arch.h +++ b/celt/arch.h @@ -78,6 +78,11 @@ static OPUS_INLINE void _celt_fatal(const char *str, const char *file, int line) #define UADD32(a,b) ((a)+(b)) #define USUB32(a,b) ((a)-(b)) +/* Set this if opus_int64

--- configure.ac | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/configure.ac b/configure.ac index f52d2c2..e1a6e9b 100644 --- a/configure.ac +++ b/configure.ac @@ -190,7 +190,7 @@ AC_ARG_ENABLE([rtcd], [enable_rtcd=yes]) AC_ARG_ENABLE([intrinsics], - [AS_HELP_STRING([--disable-intrinsics], [Disable intrinsics optimizations for ARM(float) X86(fixed)])],, +

...0x00003FFF; /* lower part */ A0_U_Q28 = silk_RSHIFT( -A_Q28[ 0 ], 14 ); /* upper part */ A1_L_Q28 = ( -A_Q28[ 1 ] ) & 0x00003FFF; /* lower part */ A1_U_Q28 = silk_RSHIFT( -A_Q28[ 1 ], 14 ); /* upper part */ ... S[ 0 ] = S[1] + silk_RSHIFT_ROUND( silk_SMULWB( out32_Q14, A0_L_Q28 ), 14 ); S[ 0 ] = silk_SMLAWB( S[ 0 ], out32_Q14, A0_U_Q28 ); S[ 0 ] = silk_SMLAWB( S[ 0 ], B_Q28[ 1 ], inval); S[ 1 ] = silk_RSHIFT_ROUND( silk_SMULWB( out32_Q14, A1_L_Q28 ), 14 ); S[ 1 ] = silk_SMLAWB( S[ 1 ], out32_Q14, A1_U_Q28 ); S[...

Hi Jean-Marc, Tested on my chromebook, when stride (channel) == 1, the optimization has no gain compared with C function. When stride (channel) == 2, the optimization is 1.2%-1.8% faster (1.6% at Complexity 8) compared with C function. Please let me know and I can remove the optimization of stride 1 case. If it's allowed to skip the split of A_Q28 and replace by 32-bit multiplication

...SHIFT_ROUND( (opus_int64)out32_Q14[ {0,1,0,1} ] * (-A_Q28[ {0,0,1,1} ]), 30 ) */ *S_s32x4 = vaddq_s32(*S_s32x4, t_s32x4); /* S[ {0,1,2,3} ] = {S[ {2,3} ],0,0} + silk_RSHIFT_ROUND( ); */ t_s32x4 = vqdmulhq_s32(inval_s32x4, B_Q28_s32x4); /* silk_SMULWB(B_Q28[ {1,1,2,2} ], in[ k * 2 + {0,1,0,1} ] ) */ *S_s32x4 = vaddq_s32(*S_s32x4, t_s32x4); /* S[ {0,1,2,3} ] = silk_SMLAWB( S[ {0,1,2,3} ], B_Q28[ {1,1,2,2} ], in[ k * 2 + {0,1,0,1} ] ); */ } Thanks, Linfeng -------------- next part -------------- An HTML attach...

...gt; *S_s32x4 = vaddq_s32(*S_s32x4, t_s32x4); > /* S[ {0,1,2,3} ] = {S[ {2,3} ],0,0} + silk_RSHIFT_ROUND( ); > */ > t_s32x4 = vqdmulhq_s32(inval_s32x4, B_Q28_s32x4); > /* silk_SMULWB(B_Q28[ {1,1,2,2} ], in[ k * 2 + {0,1,0,1} ] ) > */ > *S_s32x4 = vaddq_s32(*S_s32x4, t_s32x4); > /* S[ {0,1,2,3} ] = silk_SMLAWB( S[ {0,1,2,3} ], B_Q28[ > {1,1,2,2} ], in[ k * 2 + {0,1,0,1} ] ); */ &gt...

...VEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifndef SILK_MACROS_ARM5E_H +#define SILK_MACROS_ARM5E_H + +/* (a32 * (opus_int32)((opus_int16)(b32))) >> 16 output have to be 32bit int */ +static inline opus_int32 silk_SMULWB(opus_int32 a, opus_int16 b) +{ + int res; + __asm__( + "smulwb %0, %1, %2;\n" + : "=&r"(res) + : "r"(a), "r"(b) + ); + return res; +} + +/* a32 + (b32 * (opus_int32)((opus_int16)(c32))) >> 16 output have to be 32bit int */ +st...

...arch /* I Architecture */ ) { opus_int n, is10msFrame, denom_Q16, delta0_Q13, delta1_Q13; @@ -98,8 +99,8 @@ void silk_stereo_LR_to_MS( SILK_FIX_CONST( STEREO_RATIO_SMOOTH_COEF, 16 ); smooth_coef_Q16 = silk_SMULWB( silk_SMULBB( prev_speech_act_Q8, prev_speech_act_Q8 ), smooth_coef_Q16 ); - pred_Q13[ 0 ] = silk_stereo_find_predictor( &LP_ratio_Q14, LP_mid, LP_side, &state->mid_side_amp_Q0[ 0 ], frame_length, smooth_coef_Q16 ); - pred_Q13[ 1 ] = silk_stereo_find_predictor( &HP_ratio_Q14,...

..._Q14[ {0,1,0,1} ] * (-A_Q28[ > {0,0,1,1} ]), 30 ) */ > *S_s32x4 = vaddq_s32(*S_s32x4, t_s32x4); > /* S[ {0,1,2,3} ] = {S[ {2,3} ],0,0} + silk_RSHIFT_ROUND( ); > */ > t_s32x4 = vqdmulhq_s32(inval_s32x4, B_Q28_s32x4); > /* silk_SMULWB(B_Q28[ {1,1,2,2} ], in[ k * 2 + {0,1,0,1} ] ) > */ > *S_s32x4 = vaddq_s32(*S_s32x4, t_s32x4); > /* S[ {0,1,2,3} ] = silk_SMLAWB( S[ {0,1,2,3} ], B_Q28[ {1,1,2,2} ], in[ > k * 2 + {0,1,0,1} ] ); */ > } > > Thanks, > Linfeng > ----------...

...> *S_s32x4 = vaddq_s32(*S_s32x4, t_s32x4); > > /* S[ {0,1,2,3} ] = {S[ {2,3} ],0,0} + silk_RSHIFT_ROUND( ); > > */ > > t_s32x4 = vqdmulhq_s32(inval_s32x4, B_Q28_s32x4); > > /* silk_SMULWB(B_Q28[ {1,1,2,2} ], in[ k * 2 + {0,1,0,1} ] ) > > */ > > *S_s32x4 = vaddq_s32(*S_s32x4, t_s32x4); > > /* S[ {0,1,2,3} ] = silk_SMLAWB( S[ {0,1,2,3} ], B_Q28[ > > {1,1,2,2} ], in[ k * 2 + {0,1,0,1} ] ); */...

...IFT( LTP_pred_Q14, 1 ); /* Q13 -> Q14 */ + pred_lag_ptr++; + } else { + LTP_pred_Q14 = 0; + } + + /* Long-term shaping */ + if( lag > 0 ) { + /* Symmetric, packed FIR coefficients */ + n_LTP_Q14 = silk_SMULWB( silk_ADD32( shp_lag_ptr[ 0 ], shp_lag_ptr[ -2 ] ), HarmShapeFIRPacked_Q14 ); + n_LTP_Q14 = silk_SMLAWT( n_LTP_Q14, shp_lag_ptr[ -1 ], HarmShapeFIRPacked_Q14 ); + n_LTP_Q14 = silk_SUB_LSHIFT32( LTP_pred_Q14, n_LTP_Q14, 2 ); /* Q12 -> Q14 */ +...

NSQ_LPC_BUF_LENGTH is independent of DECISION_DELAY. --- silk/define.h | 4 ---- 1 file changed, 4 deletions(-) diff --git a/silk/define.h b/silk/define.h index 781cfdc..1286048 100644 --- a/silk/define.h +++ b/silk/define.h @@ -173,11 +173,7 @@ extern "C" #define MAX_MATRIX_SIZE MAX_LPC_ORDER /* Max of LPC Order and LTP order */ -#if( MAX_LPC_ORDER >