diff options
Diffstat (limited to 'thirdparty/opus/silk')
-rw-r--r-- | thirdparty/opus/silk/CNG.c | 61 | ||||
-rw-r--r-- | thirdparty/opus/silk/NLSF_del_dec_quant.c | 14 | ||||
-rw-r--r-- | thirdparty/opus/silk/NLSF_encode.c | 5 | ||||
-rw-r--r-- | thirdparty/opus/silk/NSQ.c | 58 | ||||
-rw-r--r-- | thirdparty/opus/silk/NSQ_del_dec.c | 40 | ||||
-rw-r--r-- | thirdparty/opus/silk/PLC.c | 3 | ||||
-rw-r--r-- | thirdparty/opus/silk/decode_core.c | 2 | ||||
-rw-r--r-- | thirdparty/opus/silk/fixed/burg_modified_FIX.c | 21 | ||||
-rw-r--r-- | thirdparty/opus/silk/fixed/x86/burg_modified_FIX_sse.c | 14 | ||||
-rw-r--r-- | thirdparty/opus/silk/macros.h | 22 | ||||
-rw-r--r-- | thirdparty/opus/silk/mips/NSQ_del_dec_mipsr1.h | 6 | ||||
-rw-r--r-- | thirdparty/opus/silk/process_NLSFs.c | 6 | ||||
-rw-r--r-- | thirdparty/opus/silk/sort.c | 2 | ||||
-rw-r--r-- | thirdparty/opus/silk/stereo_LR_to_MS.c | 8 | ||||
-rw-r--r-- | thirdparty/opus/silk/x86/NSQ_sse.c | 2 | ||||
-rw-r--r-- | thirdparty/opus/silk/x86/main_sse.h | 3 |
16 files changed, 130 insertions, 137 deletions
diff --git a/thirdparty/opus/silk/CNG.c b/thirdparty/opus/silk/CNG.c index 61787c2302..8443ad63bb 100644 --- a/thirdparty/opus/silk/CNG.c +++ b/thirdparty/opus/silk/CNG.c @@ -34,9 +34,8 @@ POSSIBILITY OF SUCH DAMAGE. /* Generates excitation for CNG LPC synthesis */ static OPUS_INLINE void silk_CNG_exc( - opus_int32 exc_Q10[], /* O CNG excitation signal Q10 */ + opus_int32 exc_Q14[], /* O CNG excitation signal Q10 */ opus_int32 exc_buf_Q14[], /* I Random samples buffer Q10 */ - opus_int32 Gain_Q16, /* I Gain to apply */ opus_int length, /* I Length */ opus_int32 *rand_seed /* I/O Seed to random index generator */ ) @@ -55,7 +54,7 @@ static OPUS_INLINE void silk_CNG_exc( idx = (opus_int)( silk_RSHIFT( seed, 24 ) & exc_mask ); silk_assert( idx >= 0 ); silk_assert( idx <= CNG_BUF_MASK_MAX ); - exc_Q10[ i ] = (opus_int16)silk_SAT16( silk_SMULWW( exc_buf_Q14[ idx ], Gain_Q16 >> 4 ) ); + exc_Q14[ i ] = exc_buf_Q14[ idx ]; } *rand_seed = seed; } @@ -85,7 +84,7 @@ void silk_CNG( ) { opus_int i, subfr; - opus_int32 sum_Q6, max_Gain_Q16, gain_Q16; + opus_int32 LPC_pred_Q10, max_Gain_Q16, gain_Q16, gain_Q10; opus_int16 A_Q12[ MAX_LPC_ORDER ]; silk_CNG_struct *psCNG = &psDec->sCNG; SAVE_STACK; @@ -124,8 +123,8 @@ void silk_CNG( /* Add CNG when packet is lost or during DTX */ if( psDec->lossCnt ) { - VARDECL( opus_int32, CNG_sig_Q10 ); - ALLOC( CNG_sig_Q10, length + MAX_LPC_ORDER, opus_int32 ); + VARDECL( opus_int32, CNG_sig_Q14 ); + ALLOC( CNG_sig_Q14, length + MAX_LPC_ORDER, opus_int32 ); /* Generate CNG excitation */ gain_Q16 = silk_SMULWW( psDec->sPLC.randScale_Q14, psDec->sPLC.prevGain_Q16[1] ); @@ -138,42 +137,46 @@ void silk_CNG( gain_Q16 = silk_SUB_LSHIFT32(silk_SMULWW( psCNG->CNG_smth_Gain_Q16, psCNG->CNG_smth_Gain_Q16 ), gain_Q16, 5 ); gain_Q16 = silk_LSHIFT32( silk_SQRT_APPROX( gain_Q16 ), 8 ); } - silk_CNG_exc( CNG_sig_Q10 + MAX_LPC_ORDER, psCNG->CNG_exc_buf_Q14, gain_Q16, length, &psCNG->rand_seed ); + gain_Q10 = silk_RSHIFT( gain_Q16, 6 ); + + silk_CNG_exc( CNG_sig_Q14 + MAX_LPC_ORDER, psCNG->CNG_exc_buf_Q14, length, &psCNG->rand_seed ); /* Convert CNG NLSF to filter representation */ silk_NLSF2A( A_Q12, psCNG->CNG_smth_NLSF_Q15, psDec->LPC_order ); /* Generate CNG signal, by synthesis filtering */ - silk_memcpy( CNG_sig_Q10, psCNG->CNG_synth_state, MAX_LPC_ORDER * sizeof( opus_int32 ) ); + silk_memcpy( CNG_sig_Q14, psCNG->CNG_synth_state, MAX_LPC_ORDER * sizeof( opus_int32 ) ); for( i = 0; i < length; i++ ) { silk_assert( psDec->LPC_order == 10 || psDec->LPC_order == 16 ); /* Avoids introducing a bias because silk_SMLAWB() always rounds to -inf */ - sum_Q6 = silk_RSHIFT( psDec->LPC_order, 1 ); - sum_Q6 = silk_SMLAWB( sum_Q6, CNG_sig_Q10[ MAX_LPC_ORDER + i - 1 ], A_Q12[ 0 ] ); - sum_Q6 = silk_SMLAWB( sum_Q6, CNG_sig_Q10[ MAX_LPC_ORDER + i - 2 ], A_Q12[ 1 ] ); - sum_Q6 = silk_SMLAWB( sum_Q6, CNG_sig_Q10[ MAX_LPC_ORDER + i - 3 ], A_Q12[ 2 ] ); - sum_Q6 = silk_SMLAWB( sum_Q6, CNG_sig_Q10[ MAX_LPC_ORDER + i - 4 ], A_Q12[ 3 ] ); - sum_Q6 = silk_SMLAWB( sum_Q6, CNG_sig_Q10[ MAX_LPC_ORDER + i - 5 ], A_Q12[ 4 ] ); - sum_Q6 = silk_SMLAWB( sum_Q6, CNG_sig_Q10[ MAX_LPC_ORDER + i - 6 ], A_Q12[ 5 ] ); - sum_Q6 = silk_SMLAWB( sum_Q6, CNG_sig_Q10[ MAX_LPC_ORDER + i - 7 ], A_Q12[ 6 ] ); - sum_Q6 = silk_SMLAWB( sum_Q6, CNG_sig_Q10[ MAX_LPC_ORDER + i - 8 ], A_Q12[ 7 ] ); - sum_Q6 = silk_SMLAWB( sum_Q6, CNG_sig_Q10[ MAX_LPC_ORDER + i - 9 ], A_Q12[ 8 ] ); - sum_Q6 = silk_SMLAWB( sum_Q6, CNG_sig_Q10[ MAX_LPC_ORDER + i - 10 ], A_Q12[ 9 ] ); + LPC_pred_Q10 = silk_RSHIFT( psDec->LPC_order, 1 ); + LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, CNG_sig_Q14[ MAX_LPC_ORDER + i - 1 ], A_Q12[ 0 ] ); + LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, CNG_sig_Q14[ MAX_LPC_ORDER + i - 2 ], A_Q12[ 1 ] ); + LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, CNG_sig_Q14[ MAX_LPC_ORDER + i - 3 ], A_Q12[ 2 ] ); + LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, CNG_sig_Q14[ MAX_LPC_ORDER + i - 4 ], A_Q12[ 3 ] ); + LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, CNG_sig_Q14[ MAX_LPC_ORDER + i - 5 ], A_Q12[ 4 ] ); + LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, CNG_sig_Q14[ MAX_LPC_ORDER + i - 6 ], A_Q12[ 5 ] ); + LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, CNG_sig_Q14[ MAX_LPC_ORDER + i - 7 ], A_Q12[ 6 ] ); + LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, CNG_sig_Q14[ MAX_LPC_ORDER + i - 8 ], A_Q12[ 7 ] ); + LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, CNG_sig_Q14[ MAX_LPC_ORDER + i - 9 ], A_Q12[ 8 ] ); + LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, CNG_sig_Q14[ MAX_LPC_ORDER + i - 10 ], A_Q12[ 9 ] ); if( psDec->LPC_order == 16 ) { - sum_Q6 = silk_SMLAWB( sum_Q6, CNG_sig_Q10[ MAX_LPC_ORDER + i - 11 ], A_Q12[ 10 ] ); - sum_Q6 = silk_SMLAWB( sum_Q6, CNG_sig_Q10[ MAX_LPC_ORDER + i - 12 ], A_Q12[ 11 ] ); - sum_Q6 = silk_SMLAWB( sum_Q6, CNG_sig_Q10[ MAX_LPC_ORDER + i - 13 ], A_Q12[ 12 ] ); - sum_Q6 = silk_SMLAWB( sum_Q6, CNG_sig_Q10[ MAX_LPC_ORDER + i - 14 ], A_Q12[ 13 ] ); - sum_Q6 = silk_SMLAWB( sum_Q6, CNG_sig_Q10[ MAX_LPC_ORDER + i - 15 ], A_Q12[ 14 ] ); - sum_Q6 = silk_SMLAWB( sum_Q6, CNG_sig_Q10[ MAX_LPC_ORDER + i - 16 ], A_Q12[ 15 ] ); + LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, CNG_sig_Q14[ MAX_LPC_ORDER + i - 11 ], A_Q12[ 10 ] ); + LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, CNG_sig_Q14[ MAX_LPC_ORDER + i - 12 ], A_Q12[ 11 ] ); + LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, CNG_sig_Q14[ MAX_LPC_ORDER + i - 13 ], A_Q12[ 12 ] ); + LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, CNG_sig_Q14[ MAX_LPC_ORDER + i - 14 ], A_Q12[ 13 ] ); + LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, CNG_sig_Q14[ MAX_LPC_ORDER + i - 15 ], A_Q12[ 14 ] ); + LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, CNG_sig_Q14[ MAX_LPC_ORDER + i - 16 ], A_Q12[ 15 ] ); } /* Update states */ - CNG_sig_Q10[ MAX_LPC_ORDER + i ] = silk_ADD_LSHIFT( CNG_sig_Q10[ MAX_LPC_ORDER + i ], sum_Q6, 4 ); - - frame[ i ] = silk_ADD_SAT16( frame[ i ], silk_RSHIFT_ROUND( CNG_sig_Q10[ MAX_LPC_ORDER + i ], 10 ) ); + CNG_sig_Q14[ MAX_LPC_ORDER + i ] = silk_ADD_LSHIFT( CNG_sig_Q14[ MAX_LPC_ORDER + i ], LPC_pred_Q10, 4 ); + + /* Scale with Gain and add to input signal */ + frame[ i ] = (opus_int16)silk_ADD_SAT16( frame[ i ], silk_SAT16( silk_RSHIFT_ROUND( silk_SMULWW( CNG_sig_Q14[ MAX_LPC_ORDER + i ], gain_Q10 ), 8 ) ) ); + } - silk_memcpy( psCNG->CNG_synth_state, &CNG_sig_Q10[ length ], MAX_LPC_ORDER * sizeof( opus_int32 ) ); + silk_memcpy( psCNG->CNG_synth_state, &CNG_sig_Q14[ length ], MAX_LPC_ORDER * sizeof( opus_int32 ) ); } else { silk_memset( psCNG->CNG_synth_state, 0, psDec->LPC_order * sizeof( opus_int32 ) ); } diff --git a/thirdparty/opus/silk/NLSF_del_dec_quant.c b/thirdparty/opus/silk/NLSF_del_dec_quant.c index c3b9efccfa..de88fee060 100644 --- a/thirdparty/opus/silk/NLSF_del_dec_quant.c +++ b/thirdparty/opus/silk/NLSF_del_dec_quant.c @@ -46,8 +46,9 @@ opus_int32 silk_NLSF_del_dec_quant( /* O Returns ) { opus_int i, j, nStates, ind_tmp, ind_min_max, ind_max_min, in_Q10, res_Q10; - opus_int pred_Q10, diff_Q10, out0_Q10, out1_Q10, rate0_Q5, rate1_Q5; - opus_int32 RD_tmp_Q25, min_Q25, min_max_Q25, max_min_Q25, pred_coef_Q16; + opus_int pred_Q10, diff_Q10, rate0_Q5, rate1_Q5; + opus_int16 out0_Q10, out1_Q10; + opus_int32 RD_tmp_Q25, min_Q25, min_max_Q25, max_min_Q25; opus_int ind_sort[ NLSF_QUANT_DEL_DEC_STATES ]; opus_int8 ind[ NLSF_QUANT_DEL_DEC_STATES ][ MAX_LPC_ORDER ]; opus_int16 prev_out_Q10[ 2 * NLSF_QUANT_DEL_DEC_STATES ]; @@ -74,8 +75,8 @@ opus_int32 silk_NLSF_del_dec_quant( /* O Returns out0_Q10 = silk_ADD16( out0_Q10, SILK_FIX_CONST( NLSF_QUANT_LEVEL_ADJ, 10 ) ); out1_Q10 = silk_ADD16( out1_Q10, SILK_FIX_CONST( NLSF_QUANT_LEVEL_ADJ, 10 ) ); } - out0_Q10_table[ i + NLSF_QUANT_MAX_AMPLITUDE_EXT ] = silk_SMULWB( (opus_int32)out0_Q10, quant_step_size_Q16 ); - out1_Q10_table[ i + NLSF_QUANT_MAX_AMPLITUDE_EXT ] = silk_SMULWB( (opus_int32)out1_Q10, quant_step_size_Q16 ); + out0_Q10_table[ i + NLSF_QUANT_MAX_AMPLITUDE_EXT ] = silk_RSHIFT( silk_SMULBB( out0_Q10, quant_step_size_Q16 ), 16 ); + out1_Q10_table[ i + NLSF_QUANT_MAX_AMPLITUDE_EXT ] = silk_RSHIFT( silk_SMULBB( out1_Q10, quant_step_size_Q16 ), 16 ); } silk_assert( (NLSF_QUANT_DEL_DEC_STATES & (NLSF_QUANT_DEL_DEC_STATES-1)) == 0 ); /* must be power of two */ @@ -85,12 +86,11 @@ opus_int32 silk_NLSF_del_dec_quant( /* O Returns prev_out_Q10[ 0 ] = 0; for( i = order - 1; ; i-- ) { rates_Q5 = &ec_rates_Q5[ ec_ix[ i ] ]; - pred_coef_Q16 = silk_LSHIFT( (opus_int32)pred_coef_Q8[ i ], 8 ); in_Q10 = x_Q10[ i ]; for( j = 0; j < nStates; j++ ) { - pred_Q10 = silk_SMULWB( pred_coef_Q16, prev_out_Q10[ j ] ); + pred_Q10 = silk_RSHIFT( silk_SMULBB( (opus_int16)pred_coef_Q8[ i ], prev_out_Q10[ j ] ), 8 ); res_Q10 = silk_SUB16( in_Q10, pred_Q10 ); - ind_tmp = silk_SMULWB( (opus_int32)inv_quant_step_size_Q6, res_Q10 ); + ind_tmp = silk_RSHIFT( silk_SMULBB( inv_quant_step_size_Q6, res_Q10 ), 16 ); ind_tmp = silk_LIMIT( ind_tmp, -NLSF_QUANT_MAX_AMPLITUDE_EXT, NLSF_QUANT_MAX_AMPLITUDE_EXT-1 ); ind[ j ][ i ] = (opus_int8)ind_tmp; diff --git a/thirdparty/opus/silk/NLSF_encode.c b/thirdparty/opus/silk/NLSF_encode.c index 03a036fda2..f03c3f1c35 100644 --- a/thirdparty/opus/silk/NLSF_encode.c +++ b/thirdparty/opus/silk/NLSF_encode.c @@ -46,7 +46,7 @@ opus_int32 silk_NLSF_encode( /* O Returns ) { opus_int i, s, ind1, bestIndex, prob_Q8, bits_q7; - opus_int32 W_tmp_Q9; + opus_int32 W_tmp_Q9, ret; VARDECL( opus_int32, err_Q26 ); VARDECL( opus_int32, RD_Q25 ); VARDECL( opus_int, tempIndices1 ); @@ -131,6 +131,7 @@ opus_int32 silk_NLSF_encode( /* O Returns /* Decode */ silk_NLSF_decode( pNLSF_Q15, NLSFIndices, psNLSF_CB ); + ret = RD_Q25[ 0 ]; RESTORE_STACK; - return RD_Q25[ 0 ]; + return ret; } diff --git a/thirdparty/opus/silk/NSQ.c b/thirdparty/opus/silk/NSQ.c index a065884070..43e3fee7e0 100644 --- a/thirdparty/opus/silk/NSQ.c +++ b/thirdparty/opus/silk/NSQ.c @@ -31,6 +31,8 @@ POSSIBILITY OF SUCH DAMAGE. #include "main.h" #include "stack_alloc.h" +#include "NSQ.h" + static OPUS_INLINE void silk_nsq_scale_states( const silk_encoder_state *psEncC, /* I Encoder State */ @@ -66,7 +68,8 @@ static OPUS_INLINE void silk_noise_shape_quantizer( opus_int offset_Q10, /* I */ opus_int length, /* I Input length */ opus_int shapingLPCOrder, /* I Noise shaping AR filter order */ - opus_int predictLPCOrder /* I Prediction filter order */ + opus_int predictLPCOrder, /* I Prediction filter order */ + int arch /* I Architecture */ ); #endif @@ -155,7 +158,7 @@ void silk_NSQ_c silk_noise_shape_quantizer( NSQ, psIndices->signalType, x_sc_Q10, pulses, pxq, sLTP_Q15, A_Q12, B_Q14, AR_shp_Q13, lag, HarmShapeFIRPacked_Q14, Tilt_Q14[ k ], LF_shp_Q14[ k ], Gains_Q16[ k ], Lambda_Q10, - offset_Q10, psEncC->subfr_length, psEncC->shapingLPCOrder, psEncC->predictLPCOrder ); + offset_Q10, psEncC->subfr_length, psEncC->shapingLPCOrder, psEncC->predictLPCOrder, psEncC->arch ); x_Q3 += psEncC->subfr_length; pulses += psEncC->subfr_length; @@ -198,15 +201,19 @@ void silk_noise_shape_quantizer( opus_int offset_Q10, /* I */ opus_int length, /* I Input length */ opus_int shapingLPCOrder, /* I Noise shaping AR filter order */ - opus_int predictLPCOrder /* I Prediction filter order */ + opus_int predictLPCOrder, /* I Prediction filter order */ + int arch /* I Architecture */ ) { - opus_int i, j; + opus_int i; opus_int32 LTP_pred_Q13, LPC_pred_Q10, n_AR_Q12, n_LTP_Q13; opus_int32 n_LF_Q12, r_Q10, rr_Q10, q1_Q0, q1_Q10, q2_Q10, rd1_Q20, rd2_Q20; opus_int32 exc_Q14, LPC_exc_Q14, xq_Q14, Gain_Q10; opus_int32 tmp1, tmp2, sLF_AR_shp_Q14; opus_int32 *psLPC_Q14, *shp_lag_ptr, *pred_lag_ptr; +#ifdef silk_short_prediction_create_arch_coef + opus_int32 a_Q12_arch[MAX_LPC_ORDER]; +#endif shp_lag_ptr = &NSQ->sLTP_shp_Q14[ NSQ->sLTP_shp_buf_idx - lag + HARM_SHAPE_FIR_TAPS / 2 ]; pred_lag_ptr = &sLTP_Q15[ NSQ->sLTP_buf_idx - lag + LTP_ORDER / 2 ]; @@ -215,32 +222,16 @@ void silk_noise_shape_quantizer( /* Set up short term AR state */ psLPC_Q14 = &NSQ->sLPC_Q14[ NSQ_LPC_BUF_LENGTH - 1 ]; +#ifdef silk_short_prediction_create_arch_coef + silk_short_prediction_create_arch_coef(a_Q12_arch, a_Q12, predictLPCOrder); +#endif + for( i = 0; i < length; i++ ) { /* Generate dither */ NSQ->rand_seed = silk_RAND( NSQ->rand_seed ); /* Short-term prediction */ - silk_assert( predictLPCOrder == 10 || predictLPCOrder == 16 ); - /* Avoids introducing a bias because silk_SMLAWB() always rounds to -inf */ - LPC_pred_Q10 = silk_RSHIFT( predictLPCOrder, 1 ); - LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, psLPC_Q14[ 0 ], a_Q12[ 0 ] ); - LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, psLPC_Q14[ -1 ], a_Q12[ 1 ] ); - LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, psLPC_Q14[ -2 ], a_Q12[ 2 ] ); - LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, psLPC_Q14[ -3 ], a_Q12[ 3 ] ); - LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, psLPC_Q14[ -4 ], a_Q12[ 4 ] ); - LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, psLPC_Q14[ -5 ], a_Q12[ 5 ] ); - LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, psLPC_Q14[ -6 ], a_Q12[ 6 ] ); - LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, psLPC_Q14[ -7 ], a_Q12[ 7 ] ); - LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, psLPC_Q14[ -8 ], a_Q12[ 8 ] ); - LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, psLPC_Q14[ -9 ], a_Q12[ 9 ] ); - if( predictLPCOrder == 16 ) { - LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, psLPC_Q14[ -10 ], a_Q12[ 10 ] ); - LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, psLPC_Q14[ -11 ], a_Q12[ 11 ] ); - LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, psLPC_Q14[ -12 ], a_Q12[ 12 ] ); - LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, psLPC_Q14[ -13 ], a_Q12[ 13 ] ); - LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, psLPC_Q14[ -14 ], a_Q12[ 14 ] ); - LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, psLPC_Q14[ -15 ], a_Q12[ 15 ] ); - } + LPC_pred_Q10 = silk_noise_shape_quantizer_short_prediction(psLPC_Q14, a_Q12, a_Q12_arch, predictLPCOrder, arch); /* Long-term prediction */ if( signalType == TYPE_VOICED ) { @@ -259,23 +250,8 @@ void silk_noise_shape_quantizer( /* Noise shape feedback */ silk_assert( ( shapingLPCOrder & 1 ) == 0 ); /* check that order is even */ - tmp2 = psLPC_Q14[ 0 ]; - tmp1 = NSQ->sAR2_Q14[ 0 ]; - NSQ->sAR2_Q14[ 0 ] = tmp2; - n_AR_Q12 = silk_RSHIFT( shapingLPCOrder, 1 ); - n_AR_Q12 = silk_SMLAWB( n_AR_Q12, tmp2, AR_shp_Q13[ 0 ] ); - for( j = 2; j < shapingLPCOrder; j += 2 ) { - tmp2 = NSQ->sAR2_Q14[ j - 1 ]; - NSQ->sAR2_Q14[ j - 1 ] = tmp1; - n_AR_Q12 = silk_SMLAWB( n_AR_Q12, tmp1, AR_shp_Q13[ j - 1 ] ); - tmp1 = NSQ->sAR2_Q14[ j + 0 ]; - NSQ->sAR2_Q14[ j + 0 ] = tmp2; - n_AR_Q12 = silk_SMLAWB( n_AR_Q12, tmp2, AR_shp_Q13[ j ] ); - } - NSQ->sAR2_Q14[ shapingLPCOrder - 1 ] = tmp1; - n_AR_Q12 = silk_SMLAWB( n_AR_Q12, tmp1, AR_shp_Q13[ shapingLPCOrder - 1 ] ); + n_AR_Q12 = silk_NSQ_noise_shape_feedback_loop(psLPC_Q14, NSQ->sAR2_Q14, AR_shp_Q13, shapingLPCOrder, arch); - n_AR_Q12 = silk_LSHIFT32( n_AR_Q12, 1 ); /* Q11 -> Q12 */ n_AR_Q12 = silk_SMLAWB( n_AR_Q12, NSQ->sLF_AR_shp_Q14, Tilt_Q14 ); n_LF_Q12 = silk_SMULWB( NSQ->sLTP_shp_Q14[ NSQ->sLTP_shp_buf_idx - 1 ], LF_shp_Q14 ); diff --git a/thirdparty/opus/silk/NSQ_del_dec.c b/thirdparty/opus/silk/NSQ_del_dec.c index aff560c221..ab6feeac98 100644 --- a/thirdparty/opus/silk/NSQ_del_dec.c +++ b/thirdparty/opus/silk/NSQ_del_dec.c @@ -31,6 +31,8 @@ POSSIBILITY OF SUCH DAMAGE. #include "main.h" #include "stack_alloc.h" +#include "NSQ.h" + typedef struct { opus_int32 sLPC_Q14[ MAX_SUB_FRAME_LENGTH + NSQ_LPC_BUF_LENGTH ]; @@ -106,7 +108,8 @@ static OPUS_INLINE void silk_noise_shape_quantizer_del_dec( opus_int warping_Q16, /* I */ opus_int nStatesDelayedDecision, /* I Number of states in decision tree */ opus_int *smpl_buf_idx, /* I Index to newest samples in buffers */ - opus_int decisionDelay /* I */ + opus_int decisionDelay, /* I */ + int arch /* I */ ); void silk_NSQ_del_dec_c( @@ -260,7 +263,7 @@ void silk_NSQ_del_dec_c( silk_noise_shape_quantizer_del_dec( NSQ, psDelDec, psIndices->signalType, x_sc_Q10, pulses, pxq, sLTP_Q15, delayedGain_Q10, A_Q12, B_Q14, AR_shp_Q13, lag, HarmShapeFIRPacked_Q14, Tilt_Q14[ k ], LF_shp_Q14[ k ], Gains_Q16[ k ], Lambda_Q10, offset_Q10, psEncC->subfr_length, subfr++, psEncC->shapingLPCOrder, - psEncC->predictLPCOrder, psEncC->warping_Q16, psEncC->nStatesDelayedDecision, &smpl_buf_idx, decisionDelay ); + psEncC->predictLPCOrder, psEncC->warping_Q16, psEncC->nStatesDelayedDecision, &smpl_buf_idx, decisionDelay, psEncC->arch ); x_Q3 += psEncC->subfr_length; pulses += psEncC->subfr_length; @@ -333,7 +336,8 @@ static OPUS_INLINE void silk_noise_shape_quantizer_del_dec( opus_int warping_Q16, /* I */ opus_int nStatesDelayedDecision, /* I Number of states in decision tree */ opus_int *smpl_buf_idx, /* I Index to newest samples in buffers */ - opus_int decisionDelay /* I */ + opus_int decisionDelay, /* I */ + int arch /* I */ ) { opus_int i, j, k, Winner_ind, RDmin_ind, RDmax_ind, last_smple_idx; @@ -343,6 +347,10 @@ static OPUS_INLINE void silk_noise_shape_quantizer_del_dec( opus_int32 q1_Q0, q1_Q10, q2_Q10, exc_Q14, LPC_exc_Q14, xq_Q14, Gain_Q10; opus_int32 tmp1, tmp2, sLF_AR_shp_Q14; opus_int32 *pred_lag_ptr, *shp_lag_ptr, *psLPC_Q14; +#ifdef silk_short_prediction_create_arch_coef + opus_int32 a_Q12_arch[MAX_LPC_ORDER]; +#endif + VARDECL( NSQ_sample_pair, psSampleState ); NSQ_del_dec_struct *psDD; NSQ_sample_struct *psSS; @@ -355,6 +363,10 @@ static OPUS_INLINE void silk_noise_shape_quantizer_del_dec( pred_lag_ptr = &sLTP_Q15[ NSQ->sLTP_buf_idx - lag + LTP_ORDER / 2 ]; Gain_Q10 = silk_RSHIFT( Gain_Q16, 6 ); +#ifdef silk_short_prediction_create_arch_coef + silk_short_prediction_create_arch_coef(a_Q12_arch, a_Q12, predictLPCOrder); +#endif + for( i = 0; i < length; i++ ) { /* Perform common calculations used in all states */ @@ -398,27 +410,7 @@ static OPUS_INLINE void silk_noise_shape_quantizer_del_dec( /* Pointer used in short term prediction and shaping */ psLPC_Q14 = &psDD->sLPC_Q14[ NSQ_LPC_BUF_LENGTH - 1 + i ]; /* Short-term prediction */ - silk_assert( predictLPCOrder == 10 || predictLPCOrder == 16 ); - /* Avoids introducing a bias because silk_SMLAWB() always rounds to -inf */ - LPC_pred_Q14 = silk_RSHIFT( predictLPCOrder, 1 ); - LPC_pred_Q14 = silk_SMLAWB( LPC_pred_Q14, psLPC_Q14[ 0 ], a_Q12[ 0 ] ); - LPC_pred_Q14 = silk_SMLAWB( LPC_pred_Q14, psLPC_Q14[ -1 ], a_Q12[ 1 ] ); - LPC_pred_Q14 = silk_SMLAWB( LPC_pred_Q14, psLPC_Q14[ -2 ], a_Q12[ 2 ] ); - LPC_pred_Q14 = silk_SMLAWB( LPC_pred_Q14, psLPC_Q14[ -3 ], a_Q12[ 3 ] ); - LPC_pred_Q14 = silk_SMLAWB( LPC_pred_Q14, psLPC_Q14[ -4 ], a_Q12[ 4 ] ); - LPC_pred_Q14 = silk_SMLAWB( LPC_pred_Q14, psLPC_Q14[ -5 ], a_Q12[ 5 ] ); - LPC_pred_Q14 = silk_SMLAWB( LPC_pred_Q14, psLPC_Q14[ -6 ], a_Q12[ 6 ] ); - LPC_pred_Q14 = silk_SMLAWB( LPC_pred_Q14, psLPC_Q14[ -7 ], a_Q12[ 7 ] ); - LPC_pred_Q14 = silk_SMLAWB( LPC_pred_Q14, psLPC_Q14[ -8 ], a_Q12[ 8 ] ); - LPC_pred_Q14 = silk_SMLAWB( LPC_pred_Q14, psLPC_Q14[ -9 ], a_Q12[ 9 ] ); - if( predictLPCOrder == 16 ) { - LPC_pred_Q14 = silk_SMLAWB( LPC_pred_Q14, psLPC_Q14[ -10 ], a_Q12[ 10 ] ); - LPC_pred_Q14 = silk_SMLAWB( LPC_pred_Q14, psLPC_Q14[ -11 ], a_Q12[ 11 ] ); - LPC_pred_Q14 = silk_SMLAWB( LPC_pred_Q14, psLPC_Q14[ -12 ], a_Q12[ 12 ] ); - LPC_pred_Q14 = silk_SMLAWB( LPC_pred_Q14, psLPC_Q14[ -13 ], a_Q12[ 13 ] ); - LPC_pred_Q14 = silk_SMLAWB( LPC_pred_Q14, psLPC_Q14[ -14 ], a_Q12[ 14 ] ); - LPC_pred_Q14 = silk_SMLAWB( LPC_pred_Q14, psLPC_Q14[ -15 ], a_Q12[ 15 ] ); - } + LPC_pred_Q14 = silk_noise_shape_quantizer_short_prediction(psLPC_Q14, a_Q12, a_Q12_arch, predictLPCOrder, arch); LPC_pred_Q14 = silk_LSHIFT( LPC_pred_Q14, 4 ); /* Q10 -> Q14 */ /* Noise shape feedback */ diff --git a/thirdparty/opus/silk/PLC.c b/thirdparty/opus/silk/PLC.c index 34a94bc313..fb6ea887b7 100644 --- a/thirdparty/opus/silk/PLC.c +++ b/thirdparty/opus/silk/PLC.c @@ -365,7 +365,8 @@ static OPUS_INLINE void silk_PLC_conceal( } /* Add prediction to LPC excitation */ - sLPC_Q14_ptr[ MAX_LPC_ORDER + i ] = silk_ADD_LSHIFT32( sLPC_Q14_ptr[ MAX_LPC_ORDER + i ], LPC_pred_Q10, 4 ); + sLPC_Q14_ptr[ MAX_LPC_ORDER + i ] = silk_ADD_SAT32( sLPC_Q14_ptr[ MAX_LPC_ORDER + i ], + silk_LSHIFT_SAT32( LPC_pred_Q10, 4 )); /* Scale with Gain */ frame[ i ] = (opus_int16)silk_SAT16( silk_SAT16( silk_RSHIFT_ROUND( silk_SMULWW( sLPC_Q14_ptr[ MAX_LPC_ORDER + i ], prevGain_Q10[ 1 ] ), 8 ) ) ); diff --git a/thirdparty/opus/silk/decode_core.c b/thirdparty/opus/silk/decode_core.c index b88991e349..e569c0e72b 100644 --- a/thirdparty/opus/silk/decode_core.c +++ b/thirdparty/opus/silk/decode_core.c @@ -219,7 +219,7 @@ void silk_decode_core( } /* Add prediction to LPC excitation */ - sLPC_Q14[ MAX_LPC_ORDER + i ] = silk_ADD_LSHIFT32( pres_Q14[ i ], LPC_pred_Q10, 4 ); + sLPC_Q14[ MAX_LPC_ORDER + i ] = silk_ADD_SAT32( pres_Q14[ i ], silk_LSHIFT_SAT32( LPC_pred_Q10, 4 ) ); /* Scale with gain */ pxq[ i ] = (opus_int16)silk_SAT16( silk_RSHIFT_ROUND( silk_SMULWW( sLPC_Q14[ MAX_LPC_ORDER + i ], Gain_Q10 ), 8 ) ); diff --git a/thirdparty/opus/silk/fixed/burg_modified_FIX.c b/thirdparty/opus/silk/fixed/burg_modified_FIX.c index 4878553b65..17d0e0993c 100644 --- a/thirdparty/opus/silk/fixed/burg_modified_FIX.c +++ b/thirdparty/opus/silk/fixed/burg_modified_FIX.c @@ -150,8 +150,11 @@ void silk_burg_modified_c( C_first_row[ k ] = silk_MLA( C_first_row[ k ], x1, x_ptr[ n - k - 1 ] ); /* Q( -rshifts ) */ C_last_row[ k ] = silk_MLA( C_last_row[ k ], x2, x_ptr[ subfr_length - n + k ] ); /* Q( -rshifts ) */ Atmp1 = silk_RSHIFT_ROUND( Af_QA[ k ], QA - 17 ); /* Q17 */ - tmp1 = silk_MLA( tmp1, x_ptr[ n - k - 1 ], Atmp1 ); /* Q17 */ - tmp2 = silk_MLA( tmp2, x_ptr[ subfr_length - n + k ], Atmp1 ); /* Q17 */ + /* We sometimes have get overflows in the multiplications (even beyond +/- 2^32), + but they cancel each other and the real result seems to always fit in a 32-bit + signed integer. This was determined experimentally, not theoretically (unfortunately). */ + tmp1 = silk_MLA_ovflw( tmp1, x_ptr[ n - k - 1 ], Atmp1 ); /* Q17 */ + tmp2 = silk_MLA_ovflw( tmp2, x_ptr[ subfr_length - n + k ], Atmp1 ); /* Q17 */ } tmp1 = -tmp1; /* Q17 */ tmp2 = -tmp2; /* Q17 */ @@ -200,12 +203,14 @@ void silk_burg_modified_c( /* Max prediction gain exceeded; set reflection coefficient such that max prediction gain is exactly hit */ tmp2 = ( (opus_int32)1 << 30 ) - silk_DIV32_varQ( minInvGain_Q30, invGain_Q30, 30 ); /* Q30 */ rc_Q31 = silk_SQRT_APPROX( tmp2 ); /* Q15 */ - /* Newton-Raphson iteration */ - rc_Q31 = silk_RSHIFT32( rc_Q31 + silk_DIV32( tmp2, rc_Q31 ), 1 ); /* Q15 */ - rc_Q31 = silk_LSHIFT32( rc_Q31, 16 ); /* Q31 */ - if( num < 0 ) { - /* Ensure adjusted reflection coefficients has the original sign */ - rc_Q31 = -rc_Q31; + if( rc_Q31 > 0 ) { + /* Newton-Raphson iteration */ + rc_Q31 = silk_RSHIFT32( rc_Q31 + silk_DIV32( tmp2, rc_Q31 ), 1 ); /* Q15 */ + rc_Q31 = silk_LSHIFT32( rc_Q31, 16 ); /* Q31 */ + if( num < 0 ) { + /* Ensure adjusted reflection coefficients has the original sign */ + rc_Q31 = -rc_Q31; + } } invGain_Q30 = minInvGain_Q30; reached_max_gain = 1; diff --git a/thirdparty/opus/silk/fixed/x86/burg_modified_FIX_sse.c b/thirdparty/opus/silk/fixed/x86/burg_modified_FIX_sse.c index 3756095fbe..3c3583c5fc 100644 --- a/thirdparty/opus/silk/fixed/x86/burg_modified_FIX_sse.c +++ b/thirdparty/opus/silk/fixed/x86/burg_modified_FIX_sse.c @@ -300,12 +300,14 @@ void silk_burg_modified_sse4_1( /* Max prediction gain exceeded; set reflection coefficient such that max prediction gain is exactly hit */ tmp2 = ( (opus_int32)1 << 30 ) - silk_DIV32_varQ( minInvGain_Q30, invGain_Q30, 30 ); /* Q30 */ rc_Q31 = silk_SQRT_APPROX( tmp2 ); /* Q15 */ - /* Newton-Raphson iteration */ - rc_Q31 = silk_RSHIFT32( rc_Q31 + silk_DIV32( tmp2, rc_Q31 ), 1 ); /* Q15 */ - rc_Q31 = silk_LSHIFT32( rc_Q31, 16 ); /* Q31 */ - if( num < 0 ) { - /* Ensure adjusted reflection coefficients has the original sign */ - rc_Q31 = -rc_Q31; + if( rc_Q31 > 0 ) { + /* Newton-Raphson iteration */ + rc_Q31 = silk_RSHIFT32( rc_Q31 + silk_DIV32( tmp2, rc_Q31 ), 1 ); /* Q15 */ + rc_Q31 = silk_LSHIFT32( rc_Q31, 16 ); /* Q31 */ + if( num < 0 ) { + /* Ensure adjusted reflection coefficients has the original sign */ + rc_Q31 = -rc_Q31; + } } invGain_Q30 = minInvGain_Q30; reached_max_gain = 1; diff --git a/thirdparty/opus/silk/macros.h b/thirdparty/opus/silk/macros.h index bc30303466..d3ca347520 100644 --- a/thirdparty/opus/silk/macros.h +++ b/thirdparty/opus/silk/macros.h @@ -34,6 +34,7 @@ POSSIBILITY OF SUCH DAMAGE. #include "opus_types.h" #include "opus_defines.h" +#include "arch.h" #if OPUS_GNUC_PREREQ(3, 0) #define opus_likely(x) (__builtin_expect(!!(x), 1)) @@ -43,31 +44,32 @@ POSSIBILITY OF SUCH DAMAGE. #define opus_unlikely(x) (!!(x)) #endif -/* Set this if opus_int64 is a native type of the CPU. */ -#define OPUS_FAST_INT64 (defined(__x86_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 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))) + ((((a32) & 0x0000FFFF) * (opus_int32)((opus_int16)(b32))) >> 16)) #endif /* a32 + (b32 * (opus_int32)((opus_int16)(c32))) >> 16 output have to be 32bit int */ #if OPUS_FAST_INT64 -#define silk_SMLAWB(a32, b32, c32) ((a32) + (((b32) * (opus_int64)((opus_int16)(c32))) >> 16)) +#define silk_SMLAWB(a32, b32, c32) ((opus_int32)((a32) + (((b32) * (opus_int64)((opus_int16)(c32))) >> 16))) #else #define silk_SMLAWB(a32, b32, c32) ((a32) + ((((b32) >> 16) * (opus_int32)((opus_int16)(c32))) + ((((b32) & 0x0000FFFF) * (opus_int32)((opus_int16)(c32))) >> 16))) #endif /* (a32 * (b32 >> 16)) >> 16 */ +#if OPUS_FAST_INT64 +#define silk_SMULWT(a32, b32) ((opus_int32)(((a32) * (opus_int64)((b32) >> 16)) >> 16)) +#else #define silk_SMULWT(a32, b32) (((a32) >> 16) * ((b32) >> 16) + ((((a32) & 0x0000FFFF) * ((b32) >> 16)) >> 16)) +#endif /* a32 + (b32 * (c32 >> 16)) >> 16 */ #if OPUS_FAST_INT64 -#define silk_SMLAWT(a32, b32, c32) ((a32) + (((b32) * ((opus_int64)(c32) >> 16)) >> 16)) +#define silk_SMLAWT(a32, b32, c32) ((opus_int32)((a32) + (((b32) * ((opus_int64)(c32) >> 16)) >> 16))) #else #define silk_SMLAWT(a32, b32, c32) ((a32) + (((b32) >> 16) * ((c32) >> 16)) + ((((b32) & 0x0000FFFF) * ((c32) >> 16)) >> 16)) #endif @@ -89,14 +91,14 @@ POSSIBILITY OF SUCH DAMAGE. /* (a32 * b32) >> 16 */ #if OPUS_FAST_INT64 -#define silk_SMULWW(a32, b32) (((opus_int64)(a32) * (b32)) >> 16) +#define silk_SMULWW(a32, b32) ((opus_int32)(((opus_int64)(a32) * (b32)) >> 16)) #else #define silk_SMULWW(a32, b32) silk_MLA(silk_SMULWB((a32), (b32)), (a32), silk_RSHIFT_ROUND((b32), 16)) #endif /* a32 + ((b32 * c32) >> 16) */ #if OPUS_FAST_INT64 -#define silk_SMLAWW(a32, b32, c32) ((a32) + (((opus_int64)(b32) * (c32)) >> 16)) +#define silk_SMLAWW(a32, b32, c32) ((opus_int32)((a32) + (((opus_int64)(b32) * (c32)) >> 16))) #else #define silk_SMLAWW(a32, b32, c32) silk_MLA(silk_SMLAWB((a32), (b32), (c32)), (b32), silk_RSHIFT_ROUND((c32), 16)) #endif @@ -149,5 +151,9 @@ static OPUS_INLINE opus_int32 silk_CLZ32(opus_int32 in32) #include "arm/macros_armv5e.h" #endif +#ifdef OPUS_ARM_PRESUME_AARCH64_NEON_INTR +#include "arm/macros_arm64.h" +#endif + #endif /* SILK_MACROS_H */ diff --git a/thirdparty/opus/silk/mips/NSQ_del_dec_mipsr1.h b/thirdparty/opus/silk/mips/NSQ_del_dec_mipsr1.h index f6afd923e8..ad1cfe2a9b 100644 --- a/thirdparty/opus/silk/mips/NSQ_del_dec_mipsr1.h +++ b/thirdparty/opus/silk/mips/NSQ_del_dec_mipsr1.h @@ -62,7 +62,8 @@ static inline void silk_noise_shape_quantizer_del_dec( opus_int warping_Q16, /* I */ opus_int nStatesDelayedDecision, /* I Number of states in decision tree */ opus_int *smpl_buf_idx, /* I Index to newest samples in buffers */ - opus_int decisionDelay /* I */ + opus_int decisionDelay, /* I */ + int arch /* I */ ) { opus_int i, j, k, Winner_ind, RDmin_ind, RDmax_ind, last_smple_idx; @@ -82,6 +83,9 @@ static inline void silk_noise_shape_quantizer_del_dec( opus_int32 cur, prev, next; + /*Unused.*/ + (void)arch; + //Intialize b_Q14 variables b_Q14_0 = b_Q14[ 0 ]; b_Q14_1 = b_Q14[ 1 ]; diff --git a/thirdparty/opus/silk/process_NLSFs.c b/thirdparty/opus/silk/process_NLSFs.c index c27cf03046..0ab71f0163 100644 --- a/thirdparty/opus/silk/process_NLSFs.c +++ b/thirdparty/opus/silk/process_NLSFs.c @@ -41,7 +41,7 @@ void silk_process_NLSFs( { opus_int i, doInterpolate; opus_int NLSF_mu_Q20; - opus_int32 i_sqr_Q15; + opus_int16 i_sqr_Q15; opus_int16 pNLSF0_temp_Q15[ MAX_LPC_ORDER ]; opus_int16 pNLSFW_QW[ MAX_LPC_ORDER ]; opus_int16 pNLSFW0_temp_QW[ MAX_LPC_ORDER ]; @@ -79,7 +79,8 @@ void silk_process_NLSFs( /* Update NLSF weights with contribution from first half */ i_sqr_Q15 = silk_LSHIFT( silk_SMULBB( psEncC->indices.NLSFInterpCoef_Q2, psEncC->indices.NLSFInterpCoef_Q2 ), 11 ); for( i = 0; i < psEncC->predictLPCOrder; i++ ) { - pNLSFW_QW[ i ] = silk_SMLAWB( silk_RSHIFT( pNLSFW_QW[ i ], 1 ), (opus_int32)pNLSFW0_temp_QW[ i ], i_sqr_Q15 ); + pNLSFW_QW[ i ] = silk_ADD16( silk_RSHIFT( pNLSFW_QW[ i ], 1 ), silk_RSHIFT( + silk_SMULBB( pNLSFW0_temp_QW[ i ], i_sqr_Q15 ), 16) ); silk_assert( pNLSFW_QW[ i ] >= 1 ); } } @@ -100,6 +101,7 @@ void silk_process_NLSFs( } else { /* Copy LPC coefficients for first half from second half */ + silk_assert( psEncC->predictLPCOrder <= MAX_LPC_ORDER ); silk_memcpy( PredCoef_Q12[ 0 ], PredCoef_Q12[ 1 ], psEncC->predictLPCOrder * sizeof( opus_int16 ) ); } } diff --git a/thirdparty/opus/silk/sort.c b/thirdparty/opus/silk/sort.c index 8670dbdd02..7187c9efb1 100644 --- a/thirdparty/opus/silk/sort.c +++ b/thirdparty/opus/silk/sort.c @@ -33,7 +33,7 @@ POSSIBILITY OF SUCH DAMAGE. /* Best case: O(n) for an already sorted array */ /* Worst case: O(n^2) for an inversely sorted array */ /* */ -/* Shell short: http://en.wikipedia.org/wiki/Shell_sort */ +/* Shell short: https://en.wikipedia.org/wiki/Shell_sort */ #include "SigProc_FIX.h" diff --git a/thirdparty/opus/silk/stereo_LR_to_MS.c b/thirdparty/opus/silk/stereo_LR_to_MS.c index 42906e6f67..dda0298de2 100644 --- a/thirdparty/opus/silk/stereo_LR_to_MS.c +++ b/thirdparty/opus/silk/stereo_LR_to_MS.c @@ -77,7 +77,7 @@ void silk_stereo_LR_to_MS( ALLOC( LP_mid, frame_length, opus_int16 ); ALLOC( HP_mid, frame_length, opus_int16 ); for( n = 0; n < frame_length; n++ ) { - sum = silk_RSHIFT_ROUND( silk_ADD_LSHIFT( mid[ n ] + mid[ n + 2 ], mid[ n + 1 ], 1 ), 2 ); + sum = silk_RSHIFT_ROUND( silk_ADD_LSHIFT( mid[ n ] + (opus_int32)mid[ n + 2 ], mid[ n + 1 ], 1 ), 2 ); LP_mid[ n ] = sum; HP_mid[ n ] = mid[ n + 1 ] - sum; } @@ -86,7 +86,7 @@ void silk_stereo_LR_to_MS( ALLOC( LP_side, frame_length, opus_int16 ); ALLOC( HP_side, frame_length, opus_int16 ); for( n = 0; n < frame_length; n++ ) { - sum = silk_RSHIFT_ROUND( silk_ADD_LSHIFT( side[ n ] + side[ n + 2 ], side[ n + 1 ], 1 ), 2 ); + sum = silk_RSHIFT_ROUND( silk_ADD_LSHIFT( side[ n ] + (opus_int32)side[ n + 2 ], side[ n + 1 ], 1 ), 2 ); LP_side[ n ] = sum; HP_side[ n ] = side[ n + 1 ] - sum; } @@ -207,7 +207,7 @@ void silk_stereo_LR_to_MS( pred0_Q13 += delta0_Q13; pred1_Q13 += delta1_Q13; w_Q24 += deltaw_Q24; - sum = silk_LSHIFT( silk_ADD_LSHIFT( mid[ n ] + mid[ n + 2 ], mid[ n + 1 ], 1 ), 9 ); /* Q11 */ + sum = silk_LSHIFT( silk_ADD_LSHIFT( mid[ n ] + (opus_int32)mid[ n + 2 ], mid[ n + 1 ], 1 ), 9 ); /* Q11 */ sum = silk_SMLAWB( silk_SMULWB( w_Q24, side[ n + 1 ] ), sum, pred0_Q13 ); /* Q8 */ sum = silk_SMLAWB( sum, silk_LSHIFT( (opus_int32)mid[ n + 1 ], 11 ), pred1_Q13 ); /* Q8 */ x2[ n - 1 ] = (opus_int16)silk_SAT16( silk_RSHIFT_ROUND( sum, 8 ) ); @@ -217,7 +217,7 @@ void silk_stereo_LR_to_MS( pred1_Q13 = -pred_Q13[ 1 ]; w_Q24 = silk_LSHIFT( width_Q14, 10 ); for( n = STEREO_INTERP_LEN_MS * fs_kHz; n < frame_length; n++ ) { - sum = silk_LSHIFT( silk_ADD_LSHIFT( mid[ n ] + mid[ n + 2 ], mid[ n + 1 ], 1 ), 9 ); /* Q11 */ + sum = silk_LSHIFT( silk_ADD_LSHIFT( mid[ n ] + (opus_int32)mid[ n + 2 ], mid[ n + 1 ], 1 ), 9 ); /* Q11 */ sum = silk_SMLAWB( silk_SMULWB( w_Q24, side[ n + 1 ] ), sum, pred0_Q13 ); /* Q8 */ sum = silk_SMLAWB( sum, silk_LSHIFT( (opus_int32)mid[ n + 1 ], 11 ), pred1_Q13 ); /* Q8 */ x2[ n - 1 ] = (opus_int16)silk_SAT16( silk_RSHIFT_ROUND( sum, 8 ) ); diff --git a/thirdparty/opus/silk/x86/NSQ_sse.c b/thirdparty/opus/silk/x86/NSQ_sse.c index 72f34fd6fc..bb3c5f1955 100644 --- a/thirdparty/opus/silk/x86/NSQ_sse.c +++ b/thirdparty/opus/silk/x86/NSQ_sse.c @@ -221,7 +221,7 @@ void silk_NSQ_sse4_1( { silk_noise_shape_quantizer( NSQ, psIndices->signalType, x_sc_Q10, pulses, pxq, sLTP_Q15, A_Q12, B_Q14, AR_shp_Q13, lag, HarmShapeFIRPacked_Q14, Tilt_Q14[ k ], LF_shp_Q14[ k ], Gains_Q16[ k ], Lambda_Q10, - offset_Q10, psEncC->subfr_length, psEncC->shapingLPCOrder, psEncC->predictLPCOrder ); + offset_Q10, psEncC->subfr_length, psEncC->shapingLPCOrder, psEncC->predictLPCOrder, psEncC->arch ); } x_Q3 += psEncC->subfr_length; diff --git a/thirdparty/opus/silk/x86/main_sse.h b/thirdparty/opus/silk/x86/main_sse.h index afd5ec26e1..d8d61310ed 100644 --- a/thirdparty/opus/silk/x86/main_sse.h +++ b/thirdparty/opus/silk/x86/main_sse.h @@ -207,7 +207,8 @@ void silk_noise_shape_quantizer( opus_int offset_Q10, /* I */ opus_int length, /* I Input length */ opus_int shapingLPCOrder, /* I Noise shaping AR filter order */ - opus_int predictLPCOrder /* I Prediction filter order */ + opus_int predictLPCOrder, /* I Prediction filter order */ + int arch /* I Architecture */ ); /**************************/ |