Merge branch 'main' into 3gpp_issue_1396_fix (69468392) · Commits · SA4 / Audio / IVAS BASOP

lib_enc/ivas_core_pre_proc_front_fx.c

+32 −19

Original line number	Diff line number	Diff line
		@@ -56,7 +56,7 @@
		#define SCE_SMC_THR 16000


		static void calculate_energy_buffer_ivas_fx( CPE_ENC_HANDLE hCPE, Word32 enerBuffer_dft[], const Word16 no_channels, const Word32 input_Fs, Word16 *enerBuffer_dft_e );
		static void calculate_energy_buffer_ivas_fx( CPE_ENC_HANDLE hCPE, Word32 enerBuffer_dft[], const Word16 no_channels, const Word32 input_Fs, Word16 enerBuffer_dft_e[] );

		#ifdef IVAS_FIXED_ENC
		static void calculate_energy_buffer_fx( CPE_ENC_HANDLE hCPE, Word64 enerBuffer_dft_fx[], Word16 *enerBuffer_dft_q_fx, const Word16 no_channels, const Word32 input_Fs );
		@@ -222,6 +222,7 @@ ivas_error pre_proc_front_ivas_fx(
		Word16 imagBuffer16[CLDFB_NO_COL_MAX][CLDFB_NO_CHANNELS_MAX];
		Word16 res_cod_SNR_M_fx_e[STEREO_DFT_BAND_MAX];
		Word16 Qfact_PS, q_lf_E_fx;
		Word16 enerBuffer_fx_exp_buf[CLDFB_NO_CHANNELS_MAX];
		#ifdef DEBUG_MODE_INFO
		Word32 *in_buff_temp;
		Word16 in_q_temp;
		@@ -482,7 +483,7 @@ ivas_error pre_proc_front_ivas_fx(
		}
		ELSE IF( EQ_16( element_mode, IVAS_CPE_DFT ) )
		{
		calculate_energy_buffer_ivas_fx( hCPE, enerBuffer_fx, st->cldfbAnaEnc->no_channels, input_Fs, enerBuffer_fx_exp );
		calculate_energy_buffer_ivas_fx( hCPE, enerBuffer_fx, st->cldfbAnaEnc->no_channels, input_Fs, enerBuffer_fx_exp_buf );
		}
		ELSE
		{
		@@ -849,11 +850,19 @@ ivas_error pre_proc_front_ivas_fx(
		/----------------------------------------------------------------
		* NB/WB/SWB/FB bandwidth detector
		----------------------------------------------------------------/

		IF( EQ_16( element_mode, IVAS_CPE_DFT ) )
		{
		FOR( i = 0; i < CLDFB_NO_CHANNELS_MAX; i++ )
		{
		sf_energySum[i] = *enerBuffer_fx_exp;
		sf_energySum[i] = enerBuffer_fx_exp_buf[i];
		move16();
		}
		}
		ELSE
		{
		set16_fx( sf_energySum, *enerBuffer_fx_exp, CLDFB_NO_CHANNELS_MAX );
		}

		test();
		IF( st->idchan == 0 && NE_16( element_mode, IVAS_CPE_MDCT ) )
		@@ -864,6 +873,15 @@ ivas_error pre_proc_front_ivas_fx(
		bw_detect_fx( st, input_fx_tmp, NULL, enerBuffer_fx, sf_energySum, ivas_format, 0 );
		}

		IF( EQ_16( element_mode, IVAS_CPE_DFT ) )
		{
		maximum_fx( sf_energySum, CLDFB_NO_CHANNELS_MAX, enerBuffer_fx_exp );
		FOR( i = 0; i < CLDFB_NO_CHANNELS_MAX; i++ )
		{
		enerBuffer_fx[i] = L_shl( enerBuffer_fx[i], sub( sf_energySum[i], *enerBuffer_fx_exp ) );
		move32();
		}
		}

		IF( NE_16( element_mode, IVAS_CPE_MDCT ) ) /* in MDCT stereo, set_bw_stereo() is used instead */
		{
		@@ -1848,7 +1866,7 @@ static void calculate_energy_buffer_ivas_fx(
		Word32 enerBuffer_dft_fx[], /* o : energy buffer() exp(enerBuffer_dft_e)*/
		const Word16 no_channels, /* i : no. of used CLDFB channels Q0*/
		const Word32 input_Fs, /* i : input sampling rate Q0*/
		Word16 *enerBuffer_dft_e )
		Word16 enerBuffer_dft_e[] )
		{
		Word16 i, j, guard_bits;
		Word64 nrg_DMX_fx[CLDFB_NO_CHANNELS_MAX];
		@@ -1856,11 +1874,8 @@ static void calculate_energy_buffer_ivas_fx(
		Word32 pDFT_DMX_fx; /q_DFT_DMX_fx*/
		Word32 chan_width_bins_fx;
		Word16 band_res_dft_fx, chan_width_f_fx, start, stop;
		Word16 norm_nrg_DMX_fx;
		Word64 max_abs_nrg_DMX_fx;
		Word16 norm_nrg_DMX_fx[CLDFB_NO_CHANNELS_MAX];

		max_abs_nrg_DMX_fx = 0;
		move64();
		Word16 temp_q1 = sub( norm_l( input_Fs ), 1 );
		Word16 temp_q2 = norm_s( hCPE->hStereoDft->NFFT );

		@@ -1892,30 +1907,28 @@ static void calculate_energy_buffer_ivas_fx(
		p_nrg_DMX_fx = W_add( p_nrg_DMX_fx, W_shr( W_mult_32_32( pDFT_DMX_fx[2 * j + 1], pDFT_DMX_fx[2 * j + 1] ), guard_bits ) ); /* 2 * q_DFT_DMX_fx + 1 - guard_bits */
		move64();
		}
		IF( GT_64( W_abs( *p_nrg_DMX_fx ), max_abs_nrg_DMX_fx ) )
		{
		max_abs_nrg_DMX_fx = W_abs( *p_nrg_DMX_fx );
		}
		norm_nrg_DMX_fx[i] = W_norm( W_abs( *p_nrg_DMX_fx ) );
		move16();
		start = stop;
		move16();
		p_nrg_DMX_fx++;
		}

		norm_nrg_DMX_fx = W_norm( max_abs_nrg_DMX_fx );
		FOR( i = 0; i < no_channels; i++ ) /* Consider only used channels, dependent on Fs */
		{
		enerBuffer_dft_fx[i] = Mpy_32_16_1( W_extract_h( W_shl( nrg_DMX_fx[i], norm_nrg_DMX_fx ) ), 10923 /* 1/3 in Q15 / ); / 2 * q_DFT_DMX_fx + 1 - guard_bits + norm_nrg_DMX_fx - 32 */
		enerBuffer_dft_fx[i] = Mpy_32_16_1( W_extract_h( W_shl( nrg_DMX_fx[i], norm_nrg_DMX_fx[i] ) ), 10923 /* 1/3 in Q15 / ); / 2 * q_DFT_DMX_fx + 1 - guard_bits + norm_nrg_DMX_fx[i] - 32 */
		move32();
		}

		*enerBuffer_dft_e = sub( add( shl( hCPE->hStereoDft->DFT_fx_e[0], 1 ), guard_bits ), norm_nrg_DMX_fx );
		enerBuffer_dft_e[i] = sub( add( shl( hCPE->hStereoDft->DFT_fx_e[0], 1 ), guard_bits ), norm_nrg_DMX_fx[i] );
		move16();
		}

		/* Set remaining entries of enerBuffer to zero */
		FOR( ; i < CLDFB_NO_CHANNELS_MAX; i++ )
		{
		enerBuffer_dft_fx[i] = 0;
		move32();
		enerBuffer_dft_e[i] = 0;
		move16();
		}
		return;
		}

lib_enc/ivas_cpe_enc_fx.c

+25 −28

Original line number	Diff line number	Diff line
		@@ -127,6 +127,7 @@ ivas_error ivas_cpe_enc_fx(
		Word16 e_old_wsp[CPE_CHANNELS], q_old_wsp;
		move16(); // Q_new
		move16(); // Q_new
		Word16 q_com;

		error = IVAS_ERR_OK;
		move32();
		@@ -649,35 +650,31 @@ ivas_error ivas_cpe_enc_fx(
		hCPE->hStereoDft->Spd_R_smooth_fx_e = sub( hCPE->hStereoDft->Spd_R_smooth_fx_e, temp );
		move16();

		shift = getScaleFactor16( sts[1]->old_input_signal_fx, input_frame );
		Scale_sig( sts[1]->old_input_signal_fx, input_frame, shift ); /* sts[1]->q_old_inp + shift */
		sts[1]->q_old_inp = add( sts[1]->q_old_inp, shift );
		move16();
		shift = getScaleFactor16( sts[1]->input_fx, input_frame );
		Copy_Scale_sig_32_16( sts[1]->input32_fx, sts[1]->input_fx, input_frame, sub( add( sts[1]->q_inp, shift ), sts[1]->q_inp32 ) ); /* sts[1]->q_inp + shift */
		sts[1]->q_inp = add( sts[1]->q_inp, shift );
		move16();
		Scale_sig( sts[1]->input_fx, input_frame, sub( s_min( sts[1]->q_inp, sts[1]->q_old_inp ), sts[1]->q_inp ) ); /* min( sts[1]->q_inp, sts[1]->q_old_inp) */
		Scale_sig( sts[1]->old_input_signal_fx, input_frame, sub( s_min( sts[1]->q_old_inp, sts[1]->q_inp ), sts[1]->q_old_inp ) ); /* min( sts[1]->q_old_inp, sts[1]->q_inp) */
		sts[1]->q_inp = s_min( sts[1]->q_inp, sts[1]->q_old_inp );
		move16();
		sts[1]->q_old_inp = sts[1]->q_inp;
		move16();

		shift = getScaleFactor16( sts[0]->old_input_signal_fx, input_frame );
		Scale_sig( sts[0]->old_input_signal_fx, input_frame, shift ); /* sts[0]->q_old_inp + shift */
		sts[0]->q_old_inp = add( sts[0]->q_old_inp, shift );
		FOR( n = 0; n < CPE_CHANNELS; n++ )
		{
		q_com = sub( add( L_norm_arr( sts[n]->input32_fx, input_frame ), sts[n]->q_inp32 ), 16 );
		q_com = s_min( q_com, add( norm_arr( sts[n]->old_input_signal_fx, input_frame ), sts[n]->q_old_inp ) );
		q_com = s_min( q_com, add( norm_arr( hCPE->input_mem_fx[n], hCPE->hStereoDft->dft_ovl ), hCPE->q_input_mem[n] ) );

		if ( EQ_16( q_com, Q15 ) )
		{
		q_com = 0;
		move16();
		shift = getScaleFactor16( sts[0]->input_fx, input_frame );
		Copy_Scale_sig_32_16( sts[0]->input32_fx, sts[0]->input_fx, input_frame, sub( add( sts[0]->q_inp, shift ), sts[0]->q_inp32 ) ); /* sts[0]->q_inp + shift */
		sts[0]->q_inp = add( sts[0]->q_inp, shift );
		}

		Copy_Scale_sig_32_16( sts[n]->input32_fx, sts[n]->input_fx, input_frame, sub( q_com, sts[n]->q_inp32 ) );
		sts[n]->q_inp = q_com;
		move16();
		Scale_sig( sts[0]->input_fx, input_frame, sub( s_min( sts[0]->q_inp, sts[0]->q_old_inp ), sts[0]->q_inp ) ); /* min( sts[1]->q_inp, sts[1]->q_old_inp) */
		Scale_sig( sts[0]->old_input_signal_fx, input_frame, sub( s_min( sts[0]->q_old_inp, sts[0]->q_inp ), sts[0]->q_old_inp ) ); /* min( sts[1]->q_old_inp, sts[1]->q_inp) */
		sts[0]->q_inp = s_min( sts[0]->q_inp, sts[0]->q_old_inp );

		scale_sig( sts[n]->old_input_signal_fx, input_frame, sub( q_com, sts[n]->q_old_inp ) );
		sts[n]->q_old_inp = q_com;
		move16();
		sts[0]->q_old_inp = sts[0]->q_inp;

		scale_sig( hCPE->input_mem_fx[n], hCPE->hStereoDft->dft_ovl, sub( q_com, hCPE->q_input_mem[n] ) );
		hCPE->q_input_mem[n] = q_com;
		move16();
		}

		stereo_dft_hybrid_ITD_flag_fx( hCPE->hStereoDft->hConfig, input_Fs, hCPE->hStereoDft->hItd->hybrid_itd_max );

lib_enc/ivas_masa_enc_fx.c

+1 −3

Original line number	Diff line number	Diff line
		@@ -3979,9 +3979,7 @@ static void quantize_ratio_ism_vector_ivas_fx(

		FOR( i = 0; i < no_ism_loc; i++ )
		{
		idx[i] = mult( ratio_ism_loc[i], max_sum_idx ); // exponent : ratio_ism_e + 15
		move16();
		idx[i] = shl( idx[i], sub( add( ratio_ism_e, 15 ), 15 ) ); // Q0
		idx[i] = extract_l( L_shr( L_mult0( ratio_ism_loc[i], max_sum_idx ), sub( 15, ratio_ism_e ) ) ); // Q0
		move16();
		part_idx_sum = add( part_idx_sum, idx[i] );

lib_enc/ivas_stereo_dft_enc_fx.c

+4 −13

Original line number	Diff line number	Diff line
		@@ -1064,23 +1064,14 @@ void stereo_dft_enc_analyze_fx(

		Word16 q_shift, guarded_bits;
		guarded_bits = find_guarded_bits_fx( NFFT );
		q_shift = sub( getScaleFactor32( pDFT_L, NFFT ), guarded_bits );
		q_shift = s_min( q_shift, sub( getScaleFactor32( pDFT_R, NFFT ), guarded_bits ) );
		FOR( Word16 j = 0; j < NFFT; j++ )
		{
		pDFT_L[j] = L_shl( pDFT_L[j], q_shift );
		move32();
		}

		q_shift = sub( L_norm_arr( pDFT_L, NFFT ), guarded_bits );
		scale_sig32( pDFT_L, NFFT, q_shift );
		DFT_e[0] = sub( DFT_e[0], q_shift );
		move16();

		FOR( Word16 j = 0; j < NFFT; j++ )
		{
		pDFT_R[j] = L_shl( pDFT_R[j], q_shift );
		move32();
		}

		q_shift = sub( L_norm_arr( pDFT_R, NFFT ), guarded_bits );
		scale_sig32( pDFT_R, NFFT, q_shift );
		DFT_e[1] = sub( DFT_e[1], q_shift );
		move16();

lib_enc/ivas_stereo_dft_td_itd_fx.c

+26 −10

Original line number	Diff line number	Diff line
		@@ -380,15 +380,6 @@ void stereo_td_itd_fx(
		move16();
		set16_fx( shift_input, 0, input_frame );

		FOR( n = 0; n < CPE_CHANNELS; n++ )
		{

		Scale_sig( &input_mem[n][0], dft_ovl, sts[n]->q_inp - q_input_mem[n] ); // Q(sts[n]->q_inp)
		// Scale_sig(&sts[n]->old_input_signal_fx, 1965, sts[n]->q_inp - sts[n]->q_old_inp);
		// sts[n]->q_old_inp = sts[n]->q_inp;
		q_input_mem[n] = sts[n]->q_inp;
		move16();
		}
		IF( EQ_16( sts[0]->element_mode, IVAS_CPE_MDCT ) )
		{
		FOR( n = 0; n < CPE_CHANNELS; n++ )
		@@ -540,7 +531,7 @@ void stereo_td_itd_mdct_stereo_fx(
		const Word16 input_frame /* i : frame length */
		)
		{
		Word16 i;
		Word16 i, n, q_com;
		Word32 bin_nrgL_fx[STEREO_DFT_N_32k_ENC];
		Word16 bin_nrgL_e[STEREO_DFT_N_32k_ENC];
		Word32 bin_nrgR_fx[STEREO_DFT_N_32k_ENC];
		@@ -584,6 +575,31 @@ void stereo_td_itd_mdct_stereo_fx(
		/call ITD function/
		stereo_dft_enc_compute_itd_fx( hCPE, DFT_fx[0], DFT_tmp_e[0], DFT_fx[1], DFT_tmp_e[1], STEREO_DFT_OFFSET, input_frame, vad_flag_dtx, vad_hover_flag, bin_nrgL_fx, bin_nrgL_e, bin_nrgR_fx, bin_nrgR_e );

		FOR( n = 0; n < CPE_CHANNELS; n++ )
		{
		q_com = add( norm_arr( hCPE->hCoreCoder[n]->input_fx, input_frame ), hCPE->hCoreCoder[n]->q_inp );
		q_com = s_min( q_com, add( norm_arr( hCPE->hCoreCoder[n]->old_input_signal_fx, input_frame ), hCPE->hCoreCoder[n]->q_old_inp ) );
		q_com = s_min( q_com, add( norm_arr( hCPE->input_mem_fx[n], hStereoMdct->hDft_ana->dft_ovl ), hCPE->q_input_mem[n] ) );

		if ( EQ_16( q_com, Q15 ) )
		{
		q_com = 0;
		move16();
		}

		scale_sig( hCPE->hCoreCoder[n]->input_fx, input_frame, sub( q_com, hCPE->hCoreCoder[n]->q_inp ) );
		hCPE->hCoreCoder[n]->q_inp = q_com;
		move16();

		scale_sig( hCPE->hCoreCoder[n]->old_input_signal_fx, input_frame, sub( q_com, hCPE->hCoreCoder[n]->q_old_inp ) );
		hCPE->hCoreCoder[n]->q_old_inp = q_com;
		move16();

		scale_sig( hCPE->input_mem_fx[n], hStereoMdct->hDft_ana->dft_ovl, sub( q_com, hCPE->q_input_mem[n] ) );
		hCPE->q_input_mem[n] = q_com;
		move16();
		}

		/* Time Domain ITD compensation using extrapolation */
		stereo_td_itd_fx( hStereoMdct->hItd, NULL, NULL, 1, hStereoMdct->hDft_ana->dft_ovl, hCPE->hCoreCoder, input_frame, hCPE->input_mem_fx, hCPE->q_input_mem );
		}