deindex_sph_idx_fx and ivas_dirac_dec_output_synthesis_process_subframe_psd_ls_fx updates (ec0aad54) · Commits · SA4 / Audio / IVAS BASOP

lib_com/ivas_masa_com.c

+254 −11

Original line number	Diff line number	Diff line
		@@ -66,6 +66,258 @@ static int16_t quantize_phi_masa( float phi, const int16_t flag_delta, float *ph
		static Word16 quantize_theta_masa_fx( const Word32 x_fx, const Word16 no_cb, Word32 *xhat_fx );

		static Word16 quantize_phi_masa_fx( const Word32 phi, const Word16 flag_delta, Word32 *phi_hat, const Word16 n );

		static Word32 estim_round[MASA_NO_CIRCLES + 1] = {
		/* Q0 */
		-423,
		0,
		422,
		845,
		1267,
		1689,
		2111,
		2532,
		2953,
		3373,
		3793,
		4212,
		4630,
		5047,
		5464,
		5880,
		6294,
		6708,
		7120,
		7532,
		7941,
		8350,
		8757,
		9163,
		9567,
		9969,
		10370,
		10769,
		11166,
		11561,
		11955,
		12346,
		12735,
		13122,
		13507,
		13890,
		14270,
		14648,
		15023,
		15396,
		15766,
		16134,
		16499,
		16861,
		17220,
		17576,
		17930,
		18280,
		18627,
		18971,
		19312,
		19650,
		19984,
		20315,
		20643,
		20967,
		21288,
		21605,
		21918,
		22228,
		22534,
		22836,
		23135,
		23429,
		23720,
		24007,
		24289,
		24568,
		24842,
		25112,
		25378,
		25640,
		25898,
		26151,
		26400,
		26644,
		26884,
		27119,
		27350,
		27576,
		27798,
		28015,
		28227,
		28435,
		28637,
		28835,
		29029,
		29217,
		29400,
		29579,
		29752,
		29921,
		30084,
		30243,
		30396,
		30544,
		30688,
		30826,
		30959,
		31086,
		31209,
		31326,
		31438,
		31545,
		31647,
		31743,
		31834,
		31919,
		32000,
		32075,
		32144,
		32208,
		32267,
		32320,
		32368,
		32411,
		32448,
		32480,
		32506,
		32527,
		32542,
		32552
		};

		static Word32 estim_ceil[MASA_NO_CIRCLES + 1] = {
		/* Q0 */
		-422,
		0,
		423,
		845,
		1268,
		1690,
		2111,
		2532,
		2953,
		3374,
		3793,
		4212,
		4630,
		5048,
		5465,
		5880,
		6295,
		6708,
		7121,
		7532,
		7942,
		8350,
		8758,
		9163,
		9567,
		9970,
		10371,
		10770,
		11167,
		11562,
		11955,
		12347,
		12736,
		13123,
		13508,
		13890,
		14270,
		14648,
		15024,
		15396,
		15767,
		16134,
		16499,
		16861,
		17220,
		17577,
		17930,
		18280,
		18628,
		18972,
		19313,
		19650,
		19985,
		20316,
		20644,
		20968,
		21288,
		21605,
		21919,
		22229,
		22535,
		22837,
		23135,
		23430,
		23720,
		24007,
		24290,
		24568,
		24843,
		25113,
		25379,
		25641,
		25898,
		26151,
		26400,
		26644,
		26884,
		27120,
		27350,
		27577,
		27798,
		28015,
		28228,
		28435,
		28638,
		28836,
		29029,
		29217,
		29401,
		29579,
		29753,
		29921,
		30085,
		30243,
		30397,
		30545,
		30688,
		30826,
		30959,
		31087,
		31210,
		31327,
		31439,
		31546,
		31647,
		31743,
		31834,
		31920,
		32000,
		32075,
		32145,
		32209,
		32268,
		32321,
		32369,
		32411,
		32449,
		32480,
		32506,
		32527,
		32543,
		32552
		};
		#endif


		@@ -1564,26 +1816,17 @@ void deindex_sph_idx_fx(
		}
		ELSE
		{
		estim_fx = Mpy_32_32( MASA_ANGLE_AT_EQUATOR_Q31, L_sub( L_shl( id_th, Q22 ), 2097152 ) /* 0.5f in Q22 / ); / Q22 */
		base_low = n[0];
		move32();
		IF( GE_16( id_th, 2 ) )
		{
		tmp32 = estim_fx % 26353590; /* 2 * PI in Q22 */
		tmp32 = Mpy_32_32( tmp32, 341782638 /* 2147483647 / 2 * PI / ); / Q7 */
		tmp16 = extract_l( L_shr( tmp32, Q7 ) );

		tmp16 = getSineWord16R2( tmp16 );
		tmp16 = sub( tmp16, MASA_ASIN_OFFSET_Q15 );
		tmp32 = Mpy_32_16_1( MASA_NTOT2_FAC_Q15, tmp16 ); /* Q15 */

		IF( EQ_16( id_th, 2 ) )
		{
		base_low = L_add( base_low, L_shl( extract_l( L_shr( ceil_fixed( tmp32, Q15 ), Q15 ) ), 1 ) ); /* Q0 */
		base_low = L_add( base_low, L_shl( estim_ceil[id_th], 1 ) ); /* Q0 */
		}
		ELSE
		{
		base_low = L_add( base_low, L_shl( extract_l( L_shr_r( tmp32, Q15 ) ), 1 ) ); /* Q0 */
		base_low = L_add( base_low, L_shl( estim_round[id_th], 1 ) ); /* Q0 */
		}
		}
		base_up = L_add( base_low, L_shl( n[id_th], 1 ) );

lib_dec/ivas_stereo_icbwe_dec.c

+1 −1

Original line number	Diff line number	Diff line
		@@ -1844,7 +1844,7 @@ void stereo_icBWE_decproc_fx(
		FOR( i = 0; i < output_frame; i++ )
		{
		#ifdef FIX_826_PRECISION_LOST_AND_COMPL
		outputHB[0][i] = W_shr( W_mac_32_16( W_mult_32_16( outputHB[0][i], 16384 ), outputHB[1][i], 16384 ), Q16 );
		outputHB[0][i] = W_extract_l( W_shr( W_mac_32_16( W_mult_32_16( outputHB[0][i], 16384 ), outputHB[1][i], 16384 ), Q16 ) );
		#else
		outputHB[0][i] = L_shr( ( outputHB[0][i] + outputHB[1][i] ), 1 );
		#endif

lib_rend/ivas_dirac_output_synthesis_dec.c

+123 −38

Original line number	Diff line number	Diff line
		@@ -2947,6 +2947,7 @@ void ivas_dirac_dec_output_synthesis_process_subframe_psd_ls_fx(
		Word32 p_gains_dir, p_gains_diff;
		Word32 g, g1, g2;
		Word32 p_cy_auto_dir_smooth, p_cy_auto_dir_smooth_prev;
		Word16 q_cy_auto_dir_smooth_local[MAX_OUTPUT_CHANNELS], q_cy_auto_dir_smooth_prev_local[MAX_OUTPUT_CHANNELS];
		Word32 p_cy_cross_dir_smooth, p_cy_cross_dir_smooth_prev;
		Word32 p_cy_auto_diff_smooth, p_cy_auto_diff_smooth_prev;
		Word32 gains_dir[CLDFB_NO_CHANNELS_MAX * MAX_OUTPUT_CHANNELS];
		@@ -2972,6 +2973,10 @@ void ivas_dirac_dec_output_synthesis_process_subframe_psd_ls_fx(
		Word32 tmp32;
		move16();

		Word64 Cldfb_RealBuffer64_fx[MAX_OUTPUT_CHANNELS][MAX_PARAM_SPATIAL_SUBFRAMES][CLDFB_NO_CHANNELS_MAX];
		Word64 Cldfb_ImagBuffer64_fx[MAX_OUTPUT_CHANNELS][MAX_PARAM_SPATIAL_SUBFRAMES][CLDFB_NO_CHANNELS_MAX];
		Word64 W_temp = 0;
		move64();
		push_wmops( "dirac_out_synth_sfr" );

		h_dirac_output_synthesis_params = &( hDirACRend->h_output_synthesis_psd_params );
		@@ -2986,6 +2991,7 @@ void ivas_dirac_dec_output_synthesis_process_subframe_psd_ls_fx(
		move16();
		num_freq_bands = hSpatParamRendCom->num_freq_bands;
		move16();
		set16_fx( q_cy_auto_dir_smooth_local, h_dirac_output_synthesis_state->q_cy_auto_dir_smooth, nchan_out_woLFE );

		/-----------------------------------------------------------------
		* compute target PSDs
		@@ -3022,7 +3028,7 @@ void ivas_dirac_dec_output_synthesis_process_subframe_psd_ls_fx(
		h_dirac_output_synthesis_state->direct_responses_fx,
		h_dirac_output_synthesis_state->direct_responses_square_fx,
		h_dirac_output_synthesis_state->cy_auto_dir_smooth_fx,
		&h_dirac_output_synthesis_state->q_cy_auto_dir_smooth,
		q_cy_auto_dir_smooth_local,
		h_dirac_output_synthesis_state->cy_cross_dir_smooth_fx,
		&h_dirac_output_synthesis_state->q_cy_cross_dir_smooth );

		@@ -3063,7 +3069,7 @@ void ivas_dirac_dec_output_synthesis_process_subframe_psd_ls_fx(

		p_cy_auto_dir_smooth = h_dirac_output_synthesis_state->cy_auto_dir_smooth_fx; // q_cy_auto_dir_smooth
		p_cy_auto_diff_smooth = h_dirac_output_synthesis_state->cy_auto_diff_smooth_fx; // q_cy_auto_diff_smooth
		q_com = s_min( h_dirac_output_synthesis_state->q_cy_auto_dir_smooth, h_dirac_output_synthesis_state->q_cy_auto_diff_smooth );
		q_com = s_min( q_cy_auto_dir_smooth_local[1], h_dirac_output_synthesis_state->q_cy_auto_diff_smooth );

		IF( EQ_32( hDirACRend->synthesisConf, DIRAC_SYNTHESIS_MONO ) )
		{
		@@ -3071,7 +3077,7 @@ void ivas_dirac_dec_output_synthesis_process_subframe_psd_ls_fx(
		move16();
		exp1 = 0;
		move16();
		tmp32 = BASOP_Util_Divide3232_Scale_cadence( L_shl( p_cy_auto_dir_smooth[num_freq_bands], sub( q_com, h_dirac_output_synthesis_state->q_cy_auto_dir_smooth ) ),
		tmp32 = BASOP_Util_Divide3232_Scale_cadence( L_shl( p_cy_auto_dir_smooth[num_freq_bands], sub( q_com, q_cy_auto_dir_smooth_local[1] ) ),
		( L_add( Sqrt32( h_dirac_output_synthesis_state->direct_power_factor_fx[0], &exp ), EPSILON_FX ) ), // (Q31 - exp)
		&exp1 );
		target_power_y = L_shr( tmp32, 1 ); // Q31 + (q_com - (31 - exp))
		@@ -3095,7 +3101,7 @@ void ivas_dirac_dec_output_synthesis_process_subframe_psd_ls_fx(
		ELSE
		{
		target_power_y = L_add(
		L_shl( p_cy_auto_dir_smooth[num_freq_bands], sub( q_com, h_dirac_output_synthesis_state->q_cy_auto_dir_smooth ) ),
		L_shl( p_cy_auto_dir_smooth[num_freq_bands], sub( q_com, q_cy_auto_dir_smooth_local[1] ) ),
		L_shl( p_cy_auto_diff_smooth[num_freq_bands], sub( q_com, h_dirac_output_synthesis_state->q_cy_auto_diff_smooth ) ) ); // q_com
		exp = q_com;
		move16();
		@@ -3249,7 +3255,7 @@ void ivas_dirac_dec_output_synthesis_process_subframe_psd_ls_fx(
		}

		// Move proto_power_smooth_fx to common Q-factor
		min_exp = 0;
		min_exp = MIN_16;
		move16();
		q_tmp = Q31;
		move16();
		@@ -3295,12 +3301,21 @@ void ivas_dirac_dec_output_synthesis_process_subframe_psd_ls_fx(

		p_cy_auto_dir_smooth = h_dirac_output_synthesis_state->cy_auto_dir_smooth_fx;
		p_cy_auto_dir_smooth_prev = h_dirac_output_synthesis_state->cy_auto_dir_smooth_prev_fx;
		q_com = s_min( h_dirac_output_synthesis_state->q_cy_auto_dir_smooth, h_dirac_output_synthesis_state->q_cy_auto_dir_smooth_prev );
		scale_sig32( p_cy_auto_dir_smooth, imult1616( nchan_out_woLFE, num_freq_bands ), sub( q_com, h_dirac_output_synthesis_state->q_cy_auto_dir_smooth ) );
		scale_sig32( p_cy_auto_dir_smooth_prev, imult1616( nchan_out_woLFE, num_freq_bands ), sub( q_com, h_dirac_output_synthesis_state->q_cy_auto_dir_smooth_prev ) );
		h_dirac_output_synthesis_state->q_cy_auto_dir_smooth = h_dirac_output_synthesis_state->q_cy_auto_dir_smooth_prev = q_com;
		FOR( k = 0; k < nchan_out_woLFE; k++ )
		{
		q_cy_auto_dir_smooth_prev_local[k] = getScaleFactor32( p_cy_auto_dir_smooth_prev + imult1616( k, num_freq_bands ), num_freq_bands );
		move16();
		scale_sig32( p_cy_auto_dir_smooth_prev + imult1616( k, num_freq_bands ), num_freq_bands, q_cy_auto_dir_smooth_prev_local[k] );
		q_cy_auto_dir_smooth_prev_local[k] = add( q_cy_auto_dir_smooth_prev_local[k], h_dirac_output_synthesis_state->q_cy_auto_dir_smooth_prev );
		move16();
		q_com = s_min( q_cy_auto_dir_smooth_local[k], q_cy_auto_dir_smooth_prev_local[k] );
		scale_sig32( p_cy_auto_dir_smooth + imult1616( k, num_freq_bands ), num_freq_bands, sub( q_com, q_cy_auto_dir_smooth_local[k] ) );
		scale_sig32( p_cy_auto_dir_smooth_prev + imult1616( k, num_freq_bands ), num_freq_bands, sub( q_com, q_cy_auto_dir_smooth_prev_local[k] ) );
		q_cy_auto_dir_smooth_local[k] = q_cy_auto_dir_smooth_prev_local[k] = q_com;
		move16();
		move16();
		}


		p_cy_cross_dir_smooth = h_dirac_output_synthesis_state->cy_cross_dir_smooth_fx;
		p_cy_cross_dir_smooth_prev = h_dirac_output_synthesis_state->cy_cross_dir_smooth_prev_fx;
		@@ -3331,16 +3346,16 @@ void ivas_dirac_dec_output_synthesis_process_subframe_psd_ls_fx(
		move32();
		g2 = L_sub( ONE_IN_Q31, g1 ); // Q31
		( p_cy_auto_dir_smooth_prev ) = L_add( Mpy_32_32( g1, ( ( p_cy_auto_dir_smooth++ ) ) ),
		Mpy_32_32( g2, ( *( p_cy_auto_dir_smooth_prev ) ) ) ); // (Q31, q_cy_auto_dir_smooth_prev) -> q_cy_auto_dir_smooth_prev
		Mpy_32_32( g2, ( *( p_cy_auto_dir_smooth_prev ) ) ) ); // (Q31, q_cy_auto_dir_smooth_prev_local) -> q_cy_auto_dir_smooth_prev_local
		move32();
		( p_cy_cross_dir_smooth_prev ) = L_add( Mpy_32_32( g1, ( ( p_cy_cross_dir_smooth++ ) ) ),
		Mpy_32_32( g2, ( *( p_cy_cross_dir_smooth_prev ) ) ) ); // (Q31, q_cy_cross_dir_smooth_prev) -> q_cy_cross_dir_smooth_prev
		move32();

		power_smooth_temp = L_shl( p_power_smooth, norm_l( p_power_smooth ) );
		L_tmp = Mpy_32_32( power_smooth_temp, ( ( p_cy_auto_dir_smooth_prev++ ) ) ); // proto_power_smooth_q + norm_l( p_power_smooth ) ) + q_cy_auto_dir_smooth_prev - 31
		L_tmp = Mpy_32_32( power_smooth_temp, ( ( p_cy_auto_dir_smooth_prev++ ) ) ); // proto_power_smooth_q + norm_l( p_power_smooth ) ) + q_cy_auto_dir_smooth_prev_local - 31
		exp = sub( Q31, sub( add( add( h_dirac_output_synthesis_state->proto_power_smooth_q, norm_l( *p_power_smooth ) ),
		h_dirac_output_synthesis_state->q_cy_auto_dir_smooth_prev ),
		q_cy_auto_dir_smooth_prev_local[k] ),
		Q31 ) );
		p_power_smooth++;

		@@ -3406,17 +3421,21 @@ void ivas_dirac_dec_output_synthesis_process_subframe_psd_ls_fx(
		g1 = alpha[l]; // Q31
		move32();
		g2 = L_sub( ONE_IN_Q31, g1 ); // Q31
		( p_cy_auto_dir_smooth_prev ) = L_add( Mpy_32_32( g1, ( ( p_cy_auto_dir_smooth++ ) ) ),
		Mpy_32_32( g2, ( *( p_cy_auto_dir_smooth_prev ) ) ) ); // (Q31, q_cy_auto_dir_smooth_prev) -> q_cy_auto_dir_smooth_prev
		W_temp = W_add( W_mult_32_32( g1, ( *( p_cy_auto_dir_smooth++ ) ) ),
		W_mult_32_32( g2, ( *( p_cy_auto_dir_smooth_prev ) ) ) );
		q_tmp = W_norm( W_temp );
		L_tmp = W_extract_h( W_shl( W_temp, q_tmp ) ); // q_cy_auto_dir_smooth_prev_local + q_tmp
		*( p_cy_auto_dir_smooth_prev++ ) = L_shr_r( L_tmp, q_tmp );

		move32();
		( p_cy_cross_dir_smooth_prev ) = L_add( Mpy_32_32( g1, ( ( p_cy_cross_dir_smooth++ ) ) ),
		Mpy_32_32( g2, ( *( p_cy_cross_dir_smooth_prev ) ) ) ); // (Q31, q_cy_cross_dir_smooth_prev) -> q_cy_cross_dir_smooth_prev
		move32();

		power_smooth_temp = L_shl( p_power_smooth, norm_l( p_power_smooth ) );
		L_tmp = Mpy_32_32( power_smooth_temp, ( ( p_cy_auto_dir_smooth_prev++ ) ) ); // proto_power_smooth_q + norm_l( p_power_smooth ) ) + q_cy_auto_dir_smooth_prev - 31
		L_tmp = Mpy_32_32( power_smooth_temp, L_tmp ); // proto_power_smooth_q + norm_l( *p_power_smooth ) ) + q_cy_auto_dir_smooth_prev_local - 31
		exp = sub( Q31, sub( add( add( h_dirac_output_synthesis_state->proto_power_smooth_q, norm_l( *p_power_smooth ) ),
		h_dirac_output_synthesis_state->q_cy_auto_dir_smooth_prev ),
		add( q_cy_auto_dir_smooth_prev_local[k], q_tmp ) ),
		Q31 ) );

		*( p_gains_dir ) = Sqrt32( L_tmp, &exp ); // (Q31 - exp)
		@@ -3498,10 +3517,12 @@ void ivas_dirac_dec_output_synthesis_process_subframe_psd_ls_fx(
		FOR( l = 0; l < num_freq_bands; l++ )
		{
		g = L_add( Mpy_32_32( g1, ( p_gain_1++ ) ), Mpy_32_32( g2, ( p_gain_2++ ) ) ); // (Q31, gains_dir_prev_q) -> gains_dir_prev_q
		RealBuffer[k][buf_idx][l] = Mpy_32_32( g, ( *( p_power_smooth++ ) ) ); // (gains_dir_prev_q, q_proto_direct_buffer) -> gains_dir_prev_q + q_proto_direct_buffer - 31
		move32();
		ImagBuffer[k][buf_idx][l] = Mpy_32_32( g, ( *( p_power_smooth++ ) ) ); // (gains_dir_prev_q, q_proto_direct_buffer) -> gains_dir_prev_q + q_proto_direct_buffer - 31
		move32();

		Cldfb_RealBuffer64_fx[k][buf_idx][l] = W_mult0_32_32( g, ( *( p_power_smooth++ ) ) ); // (gains_dir_prev_q, q_proto_direct_buffer) -> gains_dir_prev_q + q_proto_direct_buffer
		move64();

		Cldfb_ImagBuffer64_fx[k][buf_idx][l] = W_mult0_32_32( g, ( *( p_power_smooth++ ) ) ); // (gains_dir_prev_q, q_proto_direct_buffer) -> gains_dir_prev_q + q_proto_direct_buffer
		move64();
		}
		}

		@@ -3518,12 +3539,23 @@ void ivas_dirac_dec_output_synthesis_process_subframe_psd_ls_fx(
		FOR( l = 0; l < h_dirac_output_synthesis_params->max_band_decorr; l++ )
		{
		g = L_add( Mpy_32_32( g1, ( p_gain_1++ ) ), Mpy_32_32( g2, ( p_gain_2++ ) ) ); // (Q31, gains_diff_prev_q) -> gains_diff_prev_q
		RealBuffer[k][buf_idx][l] = L_add( L_shr( RealBuffer[k][buf_idx][l], Q1 ),
		L_shr( Mpy_32_32( g, ( *( p_power_smooth_diff++ ) ) ), sub( Q1, q_align ) ) ); // (gains_diff_prev_q, q_proto_direct_buffer) >> Q1 -> gains_diff_prev_q + q_proto_direct_buffer - 32
		move32();
		ImagBuffer[k][buf_idx][l] = L_add( L_shr( ImagBuffer[k][buf_idx][l], Q1 ),
		L_shr( Mpy_32_32( g, ( *( p_power_smooth_diff++ ) ) ), sub( Q1, q_align ) ) ); // (gains_diff_prev_q, q_proto_direct_buffer) >> Q1 -> gains_diff_prev_q + q_proto_direct_buffer - 32
		move32();
		Cldfb_RealBuffer64_fx[k][buf_idx][l] = W_add( Cldfb_RealBuffer64_fx[k][buf_idx][l],
		W_shr( W_mult0_32_32( g, ( *( p_power_smooth_diff++ ) ) ), negate( q_align ) ) ); // (gains_diff_prev_q, q_proto_direct_buffer) -> gains_diff_prev_q + q_proto_direct_buffer
		move64();

		if ( LT_64( W_temp, W_abs( Cldfb_RealBuffer64_fx[k][buf_idx][l] ) ) )
		{
		W_temp = W_abs( Cldfb_RealBuffer64_fx[k][buf_idx][l] );
		}

		Cldfb_ImagBuffer64_fx[k][buf_idx][l] = W_add( Cldfb_ImagBuffer64_fx[k][buf_idx][l],
		W_shr( W_mult0_32_32( g, ( *( p_power_smooth_diff++ ) ) ), negate( q_align ) ) ); // (gains_diff_prev_q, q_proto_direct_buffer) -> gains_diff_prev_q + q_proto_direct_buffer
		move64();

		if ( LT_64( W_temp, W_abs( Cldfb_ImagBuffer64_fx[k][buf_idx][l] ) ) )
		{
		W_temp = W_abs( Cldfb_ImagBuffer64_fx[k][buf_idx][l] );
		}
		}

		/Direct proto/
		@@ -3534,16 +3566,42 @@ void ivas_dirac_dec_output_synthesis_process_subframe_psd_ls_fx(
		FOR( ; l < num_freq_bands; l++ )
		{
		g = L_add( Mpy_32_32( g1, ( p_gain_1++ ) ), Mpy_32_32( g2, ( p_gain_2++ ) ) ); // (Q31, gains_diff_prev_q) -> gains_diff_prev_q
		RealBuffer[k][buf_idx][l] = L_add( L_shr( RealBuffer[k][buf_idx][l], Q1 ),
		L_shr( Mpy_32_32( g, ( *( p_power_smooth++ ) ) ), Q1 ) ); // (gains_diff_prev_q, q_proto_direct_buffer) >> Q1 -> gains_diff_prev_q + q_proto_direct_buffer - 32
		Cldfb_RealBuffer64_fx[k][buf_idx][l] = W_add( Cldfb_RealBuffer64_fx[k][buf_idx][l],
		W_mult0_32_32( g, ( *( p_power_smooth++ ) ) ) ); // (gains_diff_prev_q, q_proto_direct_buffer) >> Q1 -> gains_diff_prev_q + q_proto_direct_buffer - 32
		move64();

		if ( LT_64( W_temp, W_abs( Cldfb_RealBuffer64_fx[k][buf_idx][l] ) ) )
		{
		W_temp = W_abs( Cldfb_RealBuffer64_fx[k][buf_idx][l] );
		}

		Cldfb_ImagBuffer64_fx[k][buf_idx][l] = W_add( Cldfb_ImagBuffer64_fx[k][buf_idx][l],
		W_mult0_32_32( g, ( *( p_power_smooth++ ) ) ) ); // (gains_diff_prev_q, q_proto_direct_buffer) >> Q1 -> gains_diff_prev_q + q_proto_direct_buffer - 32
		move64();

		if ( LT_64( W_temp, W_abs( Cldfb_ImagBuffer64_fx[k][buf_idx][l] ) ) )
		{
		W_temp = W_abs( Cldfb_ImagBuffer64_fx[k][buf_idx][l] );
		}
		}
		}
		}
		q_align = W_norm( W_temp );
		FOR( buf_idx = 0; buf_idx < nbslots; ++buf_idx )
		{
		FOR( k = 0; k < nchan_out_woLFE; k++ )
		{
		FOR( l = 0; l < num_freq_bands; l++ )
		{
		RealBuffer[k][buf_idx][l] = W_extract_h( W_shl( Cldfb_RealBuffer64_fx[k][buf_idx][l], q_align ) );
		move32();
		ImagBuffer[k][buf_idx][l] = L_add( L_shr( ImagBuffer[k][buf_idx][l], Q1 ),
		L_shr( Mpy_32_32( g, ( *( p_power_smooth++ ) ) ), Q1 ) ); // (gains_diff_prev_q, q_proto_direct_buffer) >> Q1 -> gains_diff_prev_q + q_proto_direct_buffer - 32
		ImagBuffer[k][buf_idx][l] = W_extract_h( W_shl( Cldfb_ImagBuffer64_fx[k][buf_idx][l], q_align ) );
		move32();
		}
		}
		}
		*q_Cldfb = sub( sub( add( h_dirac_output_synthesis_state->proto_direct_buffer_f_q, h_dirac_output_synthesis_state->gains_dir_prev_q ), Q31 ), Q1 );

		*q_Cldfb = sub( add( add( h_dirac_output_synthesis_state->proto_direct_buffer_f_q, h_dirac_output_synthesis_state->gains_dir_prev_q ), q_align ), 32 );
		move16();

		/-----------------------------------------------------------------
		@@ -3561,9 +3619,21 @@ void ivas_dirac_dec_output_synthesis_process_subframe_psd_ls_fx(
		set_zero_fx( h_dirac_output_synthesis_state->proto_power_diff_smooth_fx, h_dirac_output_synthesis_params->max_band_decorr * nchan_out_woLFE );
		}

		minimum_fx( q_cy_auto_dir_smooth_prev_local, nchan_out_woLFE, &h_dirac_output_synthesis_state->q_cy_auto_dir_smooth_prev );
		FOR( k = 0; k < nchan_out_woLFE; k++ )
		{
		scale_sig32( h_dirac_output_synthesis_state->cy_auto_dir_smooth_prev_fx + ( k * num_freq_bands ), num_freq_bands, sub( h_dirac_output_synthesis_state->q_cy_auto_dir_smooth_prev, q_cy_auto_dir_smooth_prev_local[k] ) );
		}

		set_zero_fx( h_dirac_output_synthesis_state->cy_auto_dir_smooth_fx, imult1616( num_freq_bands, nchan_out_woLFE ) );
		h_dirac_output_synthesis_state->q_cy_auto_dir_smooth = 0;
		move16();
		set_zero_fx( h_dirac_output_synthesis_state->cy_cross_dir_smooth_fx, imult1616( num_freq_bands, nchan_out_woLFE ) );
		h_dirac_output_synthesis_state->q_cy_cross_dir_smooth = 0;
		move16();
		set_zero_fx( h_dirac_output_synthesis_state->cy_auto_diff_smooth_fx, imult1616( num_freq_bands, nchan_out_woLFE ) );
		h_dirac_output_synthesis_state->q_cy_auto_diff_smooth = 0;
		move16();

		pop_wmops();

		@@ -5938,7 +6008,9 @@ static void computeTargetPSDs_direct_subframe_fx(
		Word32 *cy_cross_dir_smooth,
		Word16 *q_cy_cross_dir_smooth )
		{
		Word16 ch_idx, cur_idx;
		Word16 ch_idx, cur_idx, i, q_tmp;
		Word64 W_tmp[CLDFB_NO_CHANNELS_MAX], W_max;
		set64_fx( W_tmp, 0, CLDFB_NO_CHANNELS_MAX );

		/* segment auxiliary buffer */
		Word32 direct_power[CLDFB_NO_CHANNELS_MAX]; /* size: num_freq_bands. */
		@@ -5951,12 +6023,25 @@ static void computeTargetPSDs_direct_subframe_fx(
		{
		cur_idx = imult1616( ch_idx, num_freq_bands );

		v_mult_fixed( direct_power, &direct_responses_square[cur_idx], &cy_auto_dir_smooth[cur_idx], num_freq_bands ); // (q_reference_power, Q31) -> q_reference_power
		W_max = 0;
		move64();
		FOR( i = 0; i < num_freq_bands; i++ )
		{
		W_tmp[i] = W_mult_32_32( direct_power[i], direct_responses_square[cur_idx + i] ); // (q_reference_power, Q31) + 1
		move64();
		W_max = W_max < W_abs( W_tmp[i] ) ? W_abs( W_tmp[i] ) : W_max;
		}
		q_tmp = W_norm( W_max );
		FOR( i = 0; i < num_freq_bands; i++ )
		{
		cy_auto_dir_smooth[cur_idx + i] = W_extract_h( W_shl( W_tmp[i], q_tmp ) );
		move32();
		}
		q_cy_auto_dir_smooth[ch_idx] = add( *q_reference_power, q_tmp );
		move16();
		v_mult_fixed( direct_power, &direct_responses[cur_idx], &cy_cross_dir_smooth[cur_idx], num_freq_bands ); // (q_reference_power, Q31) -> q_reference_power
		}

		q_cy_auto_dir_smooth = q_reference_power;
		move16();
		q_cy_cross_dir_smooth = q_reference_power;
		move16();