Merge branch 'acelp_prec_imp' into 'main' (a584f6ee) · Commits · SA4 / Audio / IVAS BASOP

lib_com/ivas_prot_fx.h

+3 −2

Original line number	Diff line number	Diff line
		@@ -2998,10 +2998,11 @@ void acelp_fast_fx(
		BSTR_ENC_HANDLE hBstr, /* i/o: encoder bitstream handle */
		const Word16 cdk_index, /* i : codebook index */
		const Word16 dn_orig[L_SUBFR],
		/* i : corr. between target and h[]. */ // Q_new + 1
		/* i : corr. between target and h[]. */ // Q_dn
		Word16 Q_dn,
		const Word16 cn[L_SUBFR],
		/* i : residual after long term prediction */ // Q_new + 1
		/* i : residual after long term prediction */ // q_cn
		const Word16 q_cn,
		const Word16 H[L_SUBFR],
		/* i : impulse response of weighted synthesis filter */ // e(norm_s(H[0])+1)
		Word16 code[L_SUBFR], /* o : algebraic (fixed) codebook excitation */

lib_com/ivas_spar_com_fx.c

+6 −4

Original line number	Diff line number	Diff line
		@@ -2658,14 +2658,16 @@ static void ivas_calc_p_coeffs_per_band_enc_fx(
		factor = L_max( factor, tmp ); // q_factor
		}

		tmp = L_shl_sat( IVAS_FIX_EPS_Q40, sub( q_factor, 40 ) );
		tmp = L_shl_sat( 189 /* 1e-20 in Q74 */, sub( q_factor, 74 ) );

		Word16 factor_exp = 0;
		move16();
		IF( LE_32( factor, tmp ) )
		{
		factor = 1250000000;
		factor_exp = Q31 - ( -4 );
		factor = 22204; // (1 / 1e-20) in Q(-52)
		factor_exp = Q15 - ( -52 );
		move32();
		move16();
		}
		ELSE
		{
		@@ -2688,7 +2690,7 @@ static void ivas_calc_p_coeffs_per_band_enc_fx(
		{
		q_tmp = W_norm( W_tmp );
		}
		cov_uu_re[i - num_dmx][j - num_dmx] = W_extract_h( W_shl( W_mult0_32_32( cov_uu_re[i - num_dmx][j - num_dmx], factor ), q_tmp ) ); /q_cov_uu_re+15-factor_exp+q_tmp-32/
		cov_uu_re[i - num_dmx][j - num_dmx] = W_extract_h( W_shl( W_tmp, q_tmp ) ); /q_cov_uu_re+15-factor_exp+q_tmp-32/
		move32();
		q_cov_uu_re_per_value[i - num_dmx][j - num_dmx] = sub( add( add( q_cov_uu_re, sub( 15, factor_exp ) ), q_tmp ), 32 );
		move16();

lib_com/prot_fx.h

+1 −8

Original line number	Diff line number	Diff line
		@@ -5077,14 +5077,7 @@ void fine_gain_quant_fx(
		Word16 fg_pred, / i/o: Predicted gains / Corrected gains Q12 */
		const Word16 gopt / i : Optimal gains Q12 */
		);
		void fine_gain_quant_ivas_fx(
		BSTR_ENC_HANDLE hBstr, /* i/o: encoder bitstream handle */
		const Word16 ord, / i : Indices for energy order Q0 */
		const Word16 num_sfm, /* i : Number of bands Q0 */
		const Word16 gain_bits, / i : Gain adjustment bits per sub band Q0 */
		Word16 fg_pred, / i/o: Predicted gains / Corrected gains i:Q12 / o:Q11 */
		const Word16 gopt / i : Optimal gains Q12 */
		);

		void get_max_pulses_fx(
		const Word16 band_start, / i : Sub band start indices */
		const Word16 band_end, / i : Sub band end indices */

lib_com/pvq_com_fx.c

+0 −67

Original line number	Diff line number	Diff line
		@@ -509,73 +509,6 @@ void apply_gain_fx(
		*
		* Fine gain quantization
		--------------------------------------------------------------------------/
		void fine_gain_quant_ivas_fx(
		BSTR_ENC_HANDLE hBstr, /* i/o: encoder bitstream handle */
		const Word16 ord, / i : Indices for energy order Q0 */
		const Word16 num_sfm, /* i : Number of bands Q0 */
		const Word16 gain_bits, / i : Gain adjustment bits per sub band Q0 */
		Word16 fg_pred, / i/o: Predicted gains / Corrected gains i:Q12 / o:Q11 */
		const Word16 gopt / i : Optimal gains Q12 */
		)
		{
		Word16 band;
		Word16 gbits;
		Word16 idx;
		Word16 gain_db, gain_dbq;
		Word16 err;

		Word16 tmp1, tmp2, exp1, exp2;
		Word32 L_tmp;
		UWord16 lsb;
		#ifdef BASOP_NOGLOB_DECLARE_LOCAL
		Flag Overflow = 0;
		move32();
		#endif

		FOR( band = 0; band < num_sfm; band++ )
		{
		gbits = gain_bits[ord[band]]; /* Q0 */
		move16();
		test();
		IF( fg_pred[band] != 0 && gbits > 0 )
		{
		exp1 = norm_s( gopt[band] );
		exp1 = sub( exp1, 1 );
		tmp1 = shl( gopt[band], exp1 );
		exp2 = norm_s( fg_pred[band] );
		tmp2 = shl( fg_pred[band], exp2 ); /* Q12 + exp2 */
		exp1 = add( 15, sub( exp1, exp2 ) );
		err = div_s( tmp1, tmp2 ); /* Q15 */
		tmp1 = norm_s( err );
		exp2 = Log2_norm_lc( L_deposit_h( shl( err, tmp1 ) ) );
		tmp1 = sub( 14, tmp1 );
		tmp1 = sub( tmp1, exp1 );
		L_tmp = L_Comp( tmp1, exp2 );
		Mpy_32_16_ss( L_tmp, 24660, &L_tmp, &lsb ); /* 24660 = 20log10(2) in Q12 / /16+12-15=13 /
		gain_db = round_fx_sat( L_shl_o( L_tmp, 17, &Overflow ) ); /* Q14 */

		idx = squant_fx( gain_db, &gain_dbq, finegain_fx[gbits - 1], gain_cb_size[gbits - 1] ); /* Q0 */
		push_indice( hBstr, IND_PVQ_FINE_GAIN, idx, gbits );

		L_tmp = L_mult0( gain_dbq, 21771 ); /* 21771=0.05log2(10) / /* 14+17=31 */
		L_tmp = L_shr( L_tmp, 15 ); /* Q16 */
		tmp1 = L_Extract_lc( L_tmp, &exp1 );
		tmp1 = abs_s( tmp1 );
		tmp1 = extract_l( Pow2( 14, tmp1 ) );
		exp1 = sub( 14, exp1 );

		L_tmp = L_mult0( fg_pred[band], tmp1 ); /12+exp1 /
		fg_pred[band] = round_fx( L_shl( L_tmp, sub( 15, exp1 ) ) ); /12+exp1+16-exp1-16=12 - 1-> Q11/
		move16();
		}
		ELSE
		{
		fg_pred[band] = shr( fg_pred[band], 1 ); // Q12 -> Q11 To align all the fg_pred indices in same Q.
		}
		}

		return;
		}
		void fine_gain_quant_fx(
		BSTR_ENC_HANDLE hBstr, /* i/o: encoder bitstream handle */
		const Word16 ord, / i : Indices for energy order Q0 */

lib_enc/cod4t64_fast.c

+121 −45

Original line number	Diff line number	Diff line
		@@ -123,10 +123,11 @@ void acelp_fast_fx(
		BSTR_ENC_HANDLE hBstr, /* i/o: encoder bitstream handle */
		const Word16 cdk_index, /* i : codebook index */
		const Word16 dn_orig[L_SUBFR],
		/* i : corr. between target and h[]. */ // Q_new + 1
		/* i : corr. between target and h[]. */ // Q_dn
		Word16 Q_dn,
		const Word16 cn[L_SUBFR],
		/* i : residual after long term prediction */ // Q_new + 1
		/* i : residual after long term prediction */ // q_cn
		const Word16 q_cn,
		const Word16 H[L_SUBFR],
		/* i : impulse response of weighted synthesis filter */ // e(norm_s(H[0])+1)
		Word16 code[L_SUBFR],
		@@ -161,7 +162,9 @@ void acelp_fast_fx(
		Word64 s64;
		Word16 flag = 0;
		move16();

		Word32 temp1, temp2, temp3, temp4, temp5, temp6;
		Word16 q_temp1, q_temp2;
		Word16 scale_temp1, scale_temp2;
		/-----------------------------------------------------------------
		* Initialization
		-----------------------------------------------------------------/
		@@ -294,16 +297,23 @@ void acelp_fast_fx(

		exp = sub( Q31, shl( Q_dn, 1 ) );

		s64 = 0;
		move64();
		FOR( i = 0; i < L_subfr; i++ )
		{
		s64 = W_mac_16_16( s64, dn_orig[i], dn_orig[i] ); // 2 * Q_dn + 1
		}
		dndn_fx = 21474836 /0.01f in Q31 /;
		move32();
		dndn_e = 0;
		move16();
		FOR( i = 0; i < L_subfr; i++ )
		IF( s64 )
		{
		dndn_fx = BASOP_Util_Add_Mant32Exp( dndn_fx, dndn_e, L_mult0( dn_orig[i], dn_orig[i] ), exp, &dndn_e ); // Q(dndn_e)
		Word16 new_exp1 = W_norm( s64 );
		dndn_fx = W_extract_h( W_shl( s64, new_exp1 ) ); // 2 * Q_dyn + exp1 - 31
		dndn_e = sub( 31, sub( add( add( shl( Q_dn, 1 ), 1 ), new_exp1 ), 32 ) );
		}

		exp1 = sub( Q31, shl( sub( Q_dn, 1 ), 1 ) );

		cncn_fx = 214748365 /* 0.1f in Q31 */;
		move32();
		@@ -312,31 +322,42 @@ void acelp_fast_fx(

		FOR( q = 0; q < nb_tracks; q++ )
		{
		s64 = 0;
		move64();
		FOR( i = 0; i < L_subfr; i += nb_tracks )
		{
		s64 = W_mac_16_16( s64, cn[i + q], cn[i + q] ); // 2 * q_cn + 1
		}

		cncn_track[q] = 214748365 /* 0.1f in Q31 */;
		move32();
		cncn_track_e[q] = 0;
		move16();

		FOR( i = 0; i < L_subfr; i += nb_tracks )
		IF( s64 )
		{
		Word32 L_tmp = L_mult0( cn[i + q], cn[i + q] );
		shift = norm_l( L_tmp );
		L_tmp = L_shl( L_tmp, shift );
		cncn_track[q] = BASOP_Util_Add_Mant32Exp( cncn_track[q], cncn_track_e[q], L_tmp, sub( exp1, shift ), &cncn_track_e[q] ); // Q(cncn_track_e[q])
		move32();
		Word16 new_exp1 = W_norm( s64 );
		cncn_track[q] = W_extract_h( W_shl( s64, new_exp1 ) ); // 2 * Q_dyn + exp1 - 31
		cncn_track_e[q] = sub( 31, sub( add( add( shl( q_cn, 1 ), 1 ), new_exp1 ), 32 ) );
		}
		cncn_fx = BASOP_Util_Add_Mant32Exp( cncn_fx, cncn_e, cncn_track[q], cncn_track_e[q], &cncn_e ); // Q(cncn_e)
		}

		Word16 tmp;
		Word16 tmp = 0;
		move16();
		s_coef_fx = BASOP_Util_Divide3232_Scale( dndn_fx, cncn_fx, &tmp );
		tmp = add( tmp, sub( dndn_e, cncn_e ) );
		s_coef_fx = Sqrt16( s_coef_fx, &tmp ); // Q(15 - tmp)


		q_temp1 = add( add( sub( Q15, tmp ), q_cn ), Q1 );
		scale_temp1 = sub( q_temp1, Q_dn );
		FOR( i = 0; i < L_subfr; i++ )
		{
		bn_orig_fx[i] = L_mac0( L_shr( L_mult( s_coef_fx, cn[i] ), sub( 15, tmp ) ), beta1_fx, dn_orig[i] ); // Q_dn
		temp1 = L_mult( s_coef_fx, cn[i] ); // Q(15 - tmp)+q_cn+1
		temp2 = L_mult( beta1_fx, dn_orig[i] ); // 1+Q_dn+1
		/* bn_orig_fx[i] is being used in Q_dn */
		temp2 = L_shr( temp2, 1 );
		temp1 = L_shr( temp1, scale_temp1 );
		bn_orig_fx[i] = L_add( temp1, temp2 ); // Q_dn
		move32();

		IF( bn_orig_fx[i] >= 0 )
		@@ -576,12 +597,12 @@ void acelp_fast_fx(
		}
		ELSE
		{
		Gn = i_mult( s[0], shr( dn_orig[m[0]], 1 ) ); // Q_dn - 1
		Gn = i_mult( s[0], dn_orig[m[0]] ); // Q_dn
		Gd = alp[0]; // Q6
		move16();
		G = Gn; // Q_dn - 1
		G = Gn; // Q_dn
		move16();
		G = i_mult( G, s[0] ); // Q_dn - 1
		G = i_mult( G, s[0] ); // Q_dn

		track = track_order[q * nb_tracks + 1]; // Q0
		move16();
		@@ -596,14 +617,17 @@ void acelp_fast_fx(
		move64();
		FOR( i = track; i < L_subfr; i += nb_tracks )
		{
		dn[i] = L_shr( L_msu( L_mult0( Gd, dn_orig[i] ), G, *alp_pos0 ), 6 ); // Q_dn
		temp1 = L_mult0( Gd, dn_orig[i] );
		temp2 = L_mult0( G, *alp_pos0 );
		temp3 = L_sub( temp1, temp2 );
		dn[i] = L_shr( temp3, 6 );
		move32();
		alp_pos0 += nb_tracks;
		s64 = W_mac_32_32( s64, dn[i], dn[i] ); // 2 * Q_dn + 1
		}
		exp1 = W_norm( s64 );
		dndn_fx = W_extract_h( W_shl( s64, exp1 ) ); // 2 * Q_dyn + exp1 - 31
		dndn_e = sub( 31, sub( add( shl( Q_dn, 1 ), exp ), 31 ) );
		dndn_e = sub( 31, sub( add( add( shl( Q_dn, 1 ), 1 ), exp1 ), 32 ) );

		IF( dndn_fx == 0 )
		{
		@@ -612,17 +636,28 @@ void acelp_fast_fx(
		dndn_e = 0;
		move16();
		}
		exp1 = 0;
		move16();
		s_coef_fx = BASOP_Util_Divide3232_Scale( dndn_fx, cncn_track[track], &exp1 );
		exp1 = add( exp1, sub( dndn_e, cncn_track_e[track] ) );
		s_coef_fx = Sqrt16( s_coef_fx, &exp1 );
		exp1 = sub( exp1, shift );
		max_val_fx = EPSILLON_FX;
		move16();
		m[1] = track; // Q0
		move16();
		q_temp1 = add( add( sub( Q15, exp1 ), q_cn ), 1 );
		q_temp2 = add( Q_dn, Q2 );
		scale_temp1 = sub( q_temp1, Q_dn );
		scale_temp2 = sub( q_temp2, Q_dn );
		FOR( i = track; i < L_subfr; i += nb_tracks )
		{
		dn[i] = L_add( L_shr( L_mult( s_coef_fx, cn[i] ), sub( 15, exp1 ) ), L_shr( imult3216( dn[i], beta2_fx ), 2 ) ); // Q_dn
		temp1 = L_mult( s_coef_fx, cn[i] ); // Q(15 - tmp)+q_cn+1
		temp2 = imult3216( dn[i], beta2_fx ); // Q_dn + 2

		/* bn_orig_fx[i] is being used in Q_dn */
		temp2 = L_shr( temp2, scale_temp2 );
		temp1 = L_shr( temp1, scale_temp1 );
		dn[i] = L_add( temp1, temp2 ); // Q_dn
		move32();
		temp_fx = imult3216( dn[i], sign_fx[i] ); // Q_dn

		@@ -645,14 +680,14 @@ void acelp_fast_fx(

		IF( GE_16( nb_pulse, 3 ) )
		{
		Gn = add( Gn, i_mult( s[1], shr( dn_orig[m[1]], 1 ) ) ); // Q_dn -1
		Gn = add( Gn, i_mult( s[1], dn_orig[m[1]] ) ); // Q_dn
		Gd32 = Gd;
		move16();
		Gd32 = L_add( Gd32, L_add( alp[0], L_mult0( i_mult( shl( s[0], 1 ), s[1] ), alp[m[0] - m[1]] ) ) ); // Q6
		G = Gn; // Q_dn - 1
		G = Gn; // Q_dn
		move16();
		G1 = i_mult( G, s[1] ); // Q_dn-1
		G = i_mult( G, s[0] ); // Q_dn-1
		G1 = i_mult( G, s[1] ); // Q_dn
		G = i_mult( G, s[0] ); // Q_dn

		track = track_order[q * nb_tracks + 2]; // Q0
		move16();
		@@ -661,7 +696,12 @@ void acelp_fast_fx(

		FOR( i = track; i < L_subfr; i += nb_tracks )
		{
		dn[i] = L_shr( L_msu( L_msu0( imult3216( Gd32, dn_orig[i] ), G, alp_pos0 ), G1, alp_pos1 ), 6 ); // Q_dn
		temp1 = imult3216( Gd32, dn_orig[i] );
		temp2 = L_mult0( G, *alp_pos0 );
		temp3 = L_mult0( G1, *alp_pos1 );
		temp4 = L_sub( temp1, temp2 );
		temp4 = L_sub( temp4, temp3 );
		dn[i] = L_shr( temp4, 6 );
		move32();
		alp_pos0 += nb_tracks;
		alp_pos1 += nb_tracks;
		@@ -677,15 +717,19 @@ void acelp_fast_fx(

		IF( GE_16( nb_pulse, 4 ) )
		{
		Gn = add( Gn, i_mult( s[2], shr( dn_orig[m[2]], 1 ) ) ); // Q_dn-1
		Gn = add( Gn, i_mult( s[2], dn_orig[m[2]] ) ); // Q_dn
		Gd32 = Gd;
		move16();
		Gd32 = L_add( Gd32, L_add( L_add( alp[0], L_mult0( i_mult( shl( s[0], 1 ), s[2] ), alp[m[0] - m[2]] ) ), L_mult0( i_mult( shl( s[1], 1 ), s[2] ), alp[m[1] - m[2]] ) ) ); // Q6
		G = Gn; // Q_dn-1
		temp1 = alp[0];
		move32();
		temp2 = L_mult0( i_mult( shl( s[0], 1 ), s[2] ), alp[m[0] - m[2]] );
		temp3 = L_mult0( i_mult( shl( s[1], 1 ), s[2] ), alp[m[1] - m[2]] );
		Gd32 = L_add( Gd32, L_add( L_add( temp1, temp2 ), temp3 ) ); // Q6
		G = Gn; // Q_dn
		move16();
		G1 = i_mult( G, s[1] ); // Q_dn-1
		G2 = i_mult( G, s[2] ); // Q_dn-1
		G = i_mult( G, s[0] ); // Q_dn-1
		G1 = i_mult( G, s[1] ); // Q_dn
		G2 = i_mult( G, s[2] ); // Q_dn
		G = i_mult( G, s[0] ); // Q_dn

		track = track_order[q * nb_tracks + 3];
		move16();
		@@ -695,7 +739,15 @@ void acelp_fast_fx(

		FOR( i = track; i < L_subfr; i += nb_tracks )
		{
		dn[i] = L_shr( L_msu( L_msu( L_msu( imult3216( Gd32, dn_orig[i] ), G, alp_pos0 ), G1, alp_pos1 ), G2, *alp_pos2 ), 6 ); // Q_dn

		temp1 = imult3216( Gd32, dn_orig[i] );
		temp2 = L_mult0( G, *alp_pos0 );
		temp3 = L_mult0( G1, *alp_pos1 );
		temp4 = L_mult0( G2, *alp_pos2 );
		temp5 = L_sub( temp1, temp2 );
		temp5 = L_sub( temp5, temp3 );
		temp5 = L_sub( temp5, temp4 );
		dn[i] = L_shr( temp5, 6 );
		move32();
		alp_pos0 += nb_tracks;
		alp_pos1 += nb_tracks;
		@@ -717,16 +769,22 @@ void acelp_fast_fx(

		IF( GE_16( nb_pulse, 5 ) )
		{
		Gn = add( Gn, i_mult( s[3], shr( dn_orig[m[3]], 1 ) ) ); // Q_dn-1
		Gn = add( Gn, i_mult( s[3], dn_orig[m[3]] ) ); // Q_dn
		Gd32 = Gd;
		move16();
		Gd32 = L_add( Gd32, L_add( L_add( L_add( alp[0], L_mult0( i_mult( shl( s[0], 1 ), s[3] ), alp[m[0] - m[3]] ) ), L_mult0( i_mult( shl( s[1], 1 ), s[3] ), alp[m[1] - m[3]] ) ), L_mult0( i_mult( shl( s[2], 1 ), s[3] ), alp[m[2] - m[3]] ) ) ); // Q6
		temp1 = alp[0];
		move32();
		temp2 = L_mult0( i_mult( shl( s[0], 1 ), s[3] ), alp[m[0] - m[3]] );
		temp3 = L_mult0( i_mult( shl( s[1], 1 ), s[3] ), alp[m[1] - m[3]] );
		temp4 = L_mult0( i_mult( shl( s[2], 1 ), s[3] ), alp[m[2] - m[3]] );

		Gd32 = L_add( Gd32, L_add( L_add( L_add( temp1, temp2 ), temp3 ), temp4 ) ); // Q6
		G = Gn;
		move16(); // Q_dn-1
		G1 = i_mult( G, s[1] ); // Q_dn-1
		G2 = i_mult( G, s[2] ); // Q_dn-1
		G3 = i_mult( G, s[3] ); // Q_dn-1
		G = i_mult( G, s[0] ); // Q_dn-1
		move16(); // Q_dn
		G1 = i_mult( G, s[1] ); // Q_dn
		G2 = i_mult( G, s[2] ); // Q_dn
		G3 = i_mult( G, s[3] ); // Q_dn
		G = i_mult( G, s[0] ); // Q_dn

		IF( EQ_16( cdk_index, 6 ) )
		{
		@@ -740,7 +798,16 @@ void acelp_fast_fx(

		FOR( i = track; i < L_subfr; i += nb_tracks )
		{
		dn[i] = L_shr( L_msu( L_msu( L_msu( L_msu( imult3216( Gd32, dn_orig[i] ), G, alp_pos0 ), G1, alp_pos1 ), G2, alp_pos2 ), G3, alp_pos3 ), 6 ); // Q_dn
		temp1 = imult3216( Gd32, dn_orig[i] );
		temp2 = L_mult0( G, *alp_pos0 );
		temp3 = L_mult0( G1, *alp_pos1 );
		temp4 = L_mult0( G2, *alp_pos2 );
		temp5 = L_mult0( G3, *alp_pos3 );
		temp6 = L_sub( temp1, temp2 );
		temp6 = L_sub( temp6, temp3 );
		temp6 = L_sub( temp6, temp4 );
		temp6 = L_sub( temp6, temp5 );
		dn[i] = L_shr( temp6, 6 );
		move32();
		alp_pos0 += nb_tracks;
		alp_pos1 += nb_tracks;
		@@ -760,7 +827,16 @@ void acelp_fast_fx(

		FOR( i = 0; i < L_subfr; i++ )
		{
		dn[i] = L_shr( L_msu( L_msu( L_msu( L_msu( imult3216( Gd32, dn_orig[i] ), G, alp_pos0 ), G1, alp_pos1 ), G2, alp_pos2 ), G3, alp_pos3 ), 6 ); /Q_dn/
		temp1 = imult3216( Gd32, dn_orig[i] );
		temp2 = L_mult0( G, *alp_pos0 );
		temp3 = L_mult0( G1, *alp_pos1 );
		temp4 = L_mult0( G2, *alp_pos2 );
		temp5 = L_mult0( G3, *alp_pos3 );
		temp6 = L_sub( temp1, temp2 );
		temp6 = L_sub( temp6, temp3 );
		temp6 = L_sub( temp6, temp4 );
		temp6 = L_sub( temp6, temp5 );
		dn[i] = L_shr( temp6, 6 );
		move16();
		alp_pos0++;
		alp_pos1++;