added/changed FIX1072_SPEEDUP_ivas_dirac_dec_output_synthesis_process_slot &... (1bb7dd5c) · Commits · SA4 / Audio / IVAS BASOP

lib_com/ivas_prot_fx.h

+8 −0

Original line number	Diff line number	Diff line
		@@ -4994,6 +4994,14 @@ void ivas_dirac_dec_get_response_fx(
		const Word16 ambisonics_order,
		Word16 Q_out );

		#ifdef FIX_1072_SPEEDUP_ivas_dirac_dec_get_response_fx
		void ivas_dirac_dec_get_response_fx_29(
		const Word16 azimuth,
		const Word16 elevation,
		Word32 response_fx, /Q_out*/
		const Word16 ambisonics_order);
		#endif

		void calculate_hodirac_sector_parameters_fx(
		DIRAC_ENC_HANDLE hDirAC, /* i : DirAC handle */
		Word32 RealBuffer_fx[DIRAC_MAX_ANA_CHANS][DIRAC_NO_FB_BANDS_MAX], /* i : signal vector (L+1)^2 x N_bins, real part */

lib_com/ivas_spar_com.c

+135 −0

Original line number	Diff line number	Diff line
		@@ -7173,6 +7173,141 @@ void ivas_dirac_dec_get_response_fx(
		return;
		}

		#ifdef FIX_1072_SPEEDUP_ivas_dirac_dec_get_response_fx
		void ivas_dirac_dec_get_response_fx_29(
		const Word16 azimuth,
		const Word16 elevation,
		Word32 response_fx, /Q_out*/
		const Word16 ambisonics_order)
		{
		Word16 index_azimuth, index_elevation;
		Word16 el, az;
		Word32 cos_1_fx, cos_2_fx, sin_1_fx, cos_az_fx[3];
		Word32 sin_az_fx[3];
		Word32 f_fx;
		Word32 c_fx_better;
		Word16 l, m;
		Word16 b, b1, b_2, b1_2;
		//Word16 Q_out = 29;

		push_wmops( "ivas_dirac_dec_get_response_fx_29" );
		index_azimuth = add( azimuth, 180 ) % 360;
		move16();
		index_elevation = add( elevation, 90 );

		Word32 e_fac = L_add(0x7FFFFFFF, 0);

		if ( GT_16( index_elevation, 90 ) )
		{
		e_fac = L_add(0x80000000, 0);
		}


		el = index_elevation;
		move16();

		if ( GT_16( index_elevation, 90 ) )
		{
		el = sub( 180, index_elevation );
		}

		az = index_azimuth;
		move16();

		if ( GT_16( index_azimuth, 180 ) )
		{
		az = sub( 360, index_azimuth );
		}

		f_fx = 1;
		move16();

		if ( GT_16( index_azimuth, 180 ) )
		{
		f_fx = -1;
		}

		cos_1_fx = L_shr( dirac_gains_trg_term_fx[az][0], 1 ); // q30
		cos_2_fx = L_shl( Mpy_32_32( cos_1_fx, cos_1_fx ), 1 ); // q30
		sin_1_fx = L_shr( dirac_gains_trg_term_fx[az][1], 1 ); // q30

		if ( EQ_32( f_fx, -1 ) )
		{
		sin_1_fx = L_negate( sin_1_fx ); // q30
		}
		cos_az_fx[0] = cos_1_fx; // q30
		move32();
		cos_az_fx[1] = L_shl( L_sub( cos_2_fx, ONE_IN_Q29 /0.5 q30/ ), 1 ); /q30/
		move32();
		cos_az_fx[2] = L_sub( L_shl( Mpy_32_32( cos_1_fx, cos_az_fx[1] ), 2 ), cos_az_fx[0] /* cos_az_fx[0] q30/ ); /q30*/
		move32();
		sin_az_fx[0] = sin_1_fx; /q30/
		move32();
		sin_az_fx[1] = L_shl( Mpy_32_32( sin_1_fx, cos_1_fx ), 2 ); /q30/
		move32();
		sin_az_fx[2] = L_shl( Mpy_32_32( sin_1_fx, L_sub( cos_2_fx, ONE_IN_Q28 /1/4 q30/ ) ), 3 ); /q30/
		move32();

		//response_fx[0] = L_shl_sat( 1, Q_out ); // Q_out
		response_fx[0] = 0x20000000;
		move32();

		//q_diff = sub( Q_out, 29 );

		push_wmops( "ivas_dirac_dec_get_response_fx_29_LOOPS" );

		FOR( l = 1; l <= ambisonics_order; l++ )
		{
		Word16 a;
		b_2 = imult1616( l, l );
		b1_2 = add( b_2, shl( l, 1 ) );
		FOR( m = 0; m < l; m += 2 )
		{
		b = b_2 + m;
		a = dirac_gains_P_idx[b];

		c_fx_better = local_result_table[el][a]; // q30
		move32();
		response_fx[b] = Mpy_32_32( c_fx_better, sin_az_fx[l - m - 1] ); // Q_out
		move32();

		b1 = b1_2 - m;
		response_fx[b1] = Mpy_32_32( c_fx_better, cos_az_fx[l - m - 1] ); // Q_out
		move32();
		}

		FOR( m = 1; m < l; m += 2 )
		{
		b = b_2 + m;
		a = dirac_gains_P_idx[b];
		c_fx_better = local_result_table[el][a]; // q30
		move32();
		c_fx_better = Mpy_32_32( c_fx_better, e_fac ); // q30
		response_fx[b] = Mpy_32_32( c_fx_better, sin_az_fx[l - m - 1] ); // Q_out
		move32();

		b1 = b1_2 - m;
		response_fx[b1] = Mpy_32_32( c_fx_better, cos_az_fx[l - m - 1] ); // Q_out
		move32();
		}

		b = add( b_2, l );
		a = dirac_gains_P_idx[b];
		c_fx_better = local_result_table_2[el][a]; // q30
		move32();
		if ( s_and( l, 0x01 ) )
		{
		c_fx_better = Mpy_32_32( c_fx_better, e_fac ); // q30
		}
		response_fx[b] = L_shl( c_fx_better, -1 ); // Q_out
		move32();
		}

		pop_wmops(); /push_wmops( "ivas_dirac_dec_get_response_fx_29_LOOPS" );/
		pop_wmops(); /push_wmops( "ivas_dirac_dec_get_response_fx_29" );/
		return;
		}
		#endif /FIX_1072_SPEEDUP_ivas_dirac_dec_get_response_fx/
		/-----------------------------------------------------------------------------------------
		* Function ivas_get_bits_to_encode
		*

lib_com/options.h

+4 −3

Original line number	Diff line number	Diff line
		@@ -164,11 +164,12 @@
		#define FIX_ISSUE_1247
		#define NONBE_FIX_1087_OOB_SBA_DTX_RS /* VA: issue 1087: Extend the length of the buffer for MCT decoding to avoid out-of-bound writing in SBA SID bitrate switching decoding */
		#define FIX_1285_DECODER_CRASH

		#define FIX_MINOR_SVD_WMOPS_MR1010X /* FhG: Minor WMOPS tuning, bit-exact to previous version, saves about 8.2 WMOPS for MR1010 */
		#define FIX_1072_SPEEDUP_gainpanning /* FhG: Minor WMOPS tuning, in development*/
		#define FIX_1072_SPEEDUP_COMPUTEDIFUSENESS /* "-" */


		#define FIX_1072_SPEEDUP_gainpanning /* FhG: WMOPS tuning, in development*/
		#define FIX_1072_SPEEDUP_COMPUTEDIFUSENESS /* "-" */
		#define FIX_1072_SPEEDUP_ivas_dirac_dec_get_response_fx /FhG: WMOPS tuning, in development/
		#define FIX1072_SPEEDUP_ivas_dirac_dec_output_synthesis_process_slot /FhG: WMOPS tuning, in development/

		#endif
		No newline at end of file

lib_rend/ivas_dirac_output_synthesis_dec.c

+90 −27

Original line number	Diff line number	Diff line
		@@ -821,7 +821,7 @@ void ivas_dirac_dec_output_synthesis_process_slot_fx(
		Word16 k;
		IF( ch_idx != 0 )
		{
		push_wmops( "(IDR) LOOP1 PSDs PATH3 B3.3 bigloop IF <<<<<-\|" );
		push_wmops( "(IDR) LOOP1 PSDs PATH3 B3.3 bigloop A <<<<<-\|" );
		;
		Word32 a, c;
		Word16 b, b_exp, sqr_exp, q_diff_aab, q_diff_c;
		@@ -912,8 +912,50 @@ void ivas_dirac_dec_output_synthesis_process_slot_fx(
		}
		}
		c = Madd_32_16( ONE_IN_Q27 /1 Q27/, L_sub( h_dirac_output_synthesis_params->diffuse_compensation_factor_fx, ONE_IN_Q27 /1 Q27/ ), 5461 ); /Diffuseness modellling nrg compensation/ /* 1.0 / 6.0 = 5461 in Q15/ /Q27*/
		pop_wmops(); /push_wmops( "(IDR) LOOP1 PSDs PATH3 B3.3 bigloop A <<<<<-\|" );/
		push_wmops( "(IDR) LOOP1 PSDs PATH3 B3.3 bigloop B <<<<<-\|" );
		#ifdef FIX1072_SPEEDUP_ivas_dirac_dec_output_synthesis_process_slot
		FOR( ; k < num_freq_bands; k++ )
		{
		a = h_dirac_output_synthesis_state->direct_responses_fx[ch_idx * num_freq_bands + k]; // Q = h_dirac_output_synthesis_state->q_direct_responses
		move32();
		IF( reference_power[k + num_freq_bands] == 0 )
		{
		sqr_inp = Mpy_32_32( diffuseness[k], c );
		sqr_exp = sub( 31 - 4, q_diffuseness );
		}
		ELSE
		{
		IF( reference_power[k + ( ch_idx + 1 ) * num_freq_bands] == 0 )
		{
		mpy_a_a_b = Mpy_32_32( a, a ); // Q = (h_dirac_output_synthesis_state->q_direct_responses + (15 - b_exp) - 15) + (h_dirac_output_synthesis_state->q_direct_responses) - 31
		mpy_diff_aab = Mpy_32_32( L_sub( L_shl( 1, q_diffuseness ), diffuseness[k] ), mpy_a_a_b ); // Q = 2*(h_dirac_output_synthesis_state->q_direct_responses) - b_exp - 31 + q_diffuseness -31
		mpy_diff_c = Mpy_32_32( diffuseness[k], c ); // Q = q_diffuseness - 4
		//q_diff_aab = add( h_dirac_output_synthesis_state->direct_responses_q, add( sub( h_dirac_output_synthesis_state->direct_responses_q, 31 ), sub( q_diffuseness, 31 ) ) );
		q_diff_aab = sub( add( h_dirac_output_synthesis_state->direct_responses_q, add( h_dirac_output_synthesis_state->direct_responses_q, q_diffuseness ) ), 62 );
		q_diff_c = sub( q_diffuseness, 4 );

		push_wmops( "(IDR) LOOP1 PSDs PATH3 B3.3 bigloop IF nfreqbds <<<<<<-\|" );
		sqr_inp = BASOP_Util_Add_Mant32Exp( mpy_diff_c, sub( 31, q_diff_c ), mpy_diff_aab, sub( 31, q_diff_aab ), &sqr_exp ); /q(31-sqr_exp)/
		}
		ELSE
		{
		b = BASOP_Util_Divide3232_Scale( reference_power[k + num_freq_bands], reference_power[k + ( ch_idx + 1 ) * num_freq_bands], &b_exp ); /q(15-b_exp)/

		mpy_a_a_b = Mpy_32_32( a, Mpy_32_16_1( a, b ) ); // Q = (h_dirac_output_synthesis_state->q_direct_responses + (15 - b_exp) - 15) + (h_dirac_output_synthesis_state->q_direct_responses) - 31
		mpy_diff_aab = Mpy_32_32( L_sub( L_shl( 1, q_diffuseness ), diffuseness[k] ), mpy_a_a_b ); // Q = 2*(h_dirac_output_synthesis_state->q_direct_responses) - b_exp - 31 + q_diffuseness -31
		mpy_diff_c = Mpy_32_32( diffuseness[k], c ); // Q = q_diffuseness - 4
		//q_diff_aab = add( add(h_dirac_output_synthesis_state->direct_responses_q , sub( sub( 15, b_exp ), 15 )), add( sub( h_dirac_output_synthesis_state->direct_responses_q, 31 ), sub( q_diffuseness, 31 ) ) );
		q_diff_aab = add( sub( h_dirac_output_synthesis_state->direct_responses_q, b_exp ), ( sub( add( h_dirac_output_synthesis_state->direct_responses_q, q_diffuseness ), 62 ) ) );
		q_diff_c = sub( q_diffuseness, 4 );

		sqr_inp = BASOP_Util_Add_Mant32Exp( mpy_diff_c, sub( 31, q_diff_c ), mpy_diff_aab, sub( 31, q_diff_aab ), &sqr_exp ); /q(31-sqr_exp)/

		}
		}
		sqr = Sqrt32( sqr_inp, &sqr_exp ); /Q(31-sqr_exp)/
		sqr = L_shr( sqr, 2 ); /Q(31-sqr_exp)/
		}
		#else
		FOR( ; k < num_freq_bands; k++ )
		{
		a = h_dirac_output_synthesis_state->direct_responses_fx[ch_idx * num_freq_bands + k]; // Q = h_dirac_output_synthesis_state->q_direct_responses
		@@ -942,7 +984,7 @@ void ivas_dirac_dec_output_synthesis_process_slot_fx(
		mpy_a_a_b = Mpy_32_32( a, Mpy_32_16_1( a, b ) ); // Q = (h_dirac_output_synthesis_state->q_direct_responses + (15 - b_exp) - 15) + (h_dirac_output_synthesis_state->q_direct_responses) - 31
		mpy_diff_aab = Mpy_32_32( L_sub( L_shl( 1, q_diffuseness ), diffuseness[k] ), mpy_a_a_b ); // Q = 2*(h_dirac_output_synthesis_state->q_direct_responses) - b_exp - 31 + q_diffuseness -31
		mpy_diff_c = Mpy_32_32( diffuseness[k], c ); // Q = q_diffuseness - 4
		q_diff_aab = add( h_dirac_output_synthesis_state->direct_responses_q + sub( sub( 15, b_exp ), 15 ), add( sub( h_dirac_output_synthesis_state->direct_responses_q, 31 ), sub( q_diffuseness, 31 ) ) );
		q_diff_aab = add( add( h_dirac_output_synthesis_state->direct_responses_q, sub( sub( 15, b_exp ), 15 ) ), add( sub( h_dirac_output_synthesis_state->direct_responses_q, 31 ), sub( q_diffuseness, 31 ) ) );
		q_diff_c = sub( q_diffuseness, 4 );

		test();
		@@ -968,6 +1010,9 @@ void ivas_dirac_dec_output_synthesis_process_slot_fx(

		sqr = Sqrt32( sqr_inp, &sqr_exp ); /Q(31-sqr_exp)/
		sqr = L_shr( sqr, 2 ); /Q(31-sqr_exp)/
		}
		#endif

		IF( h_dirac_output_synthesis_state->cy_cross_dir_smooth_fx[ch_idx * num_freq_bands + k] != 0 )
		{
		IF( LT_16( sub( 31, sqr_exp ), h_dirac_output_synthesis_state->q_cy_cross_dir_smooth ) )
		@@ -993,13 +1038,12 @@ void ivas_dirac_dec_output_synthesis_process_slot_fx(
		Q_temp_cy_cross_dir_smooth_fx[ch_idx * num_freq_bands + k] = sub( 31, sqr_exp );
		move16();
		}
		}
		pop_wmops(); /push_wmops( "(IDR) LOOP1 PSDs PATH3 B3.3 bigloop IF nfreqbds <<<<<<-\|" );/
		pop_wmops(); /push_wmops( "(IDR) LOOP1 PSDs PATH3 B3.3 bigloop IF <<<<<-\|" );/

		pop_wmops(); /push_wmops( "(IDR) LOOP1 PSDs PATH3 B3.3 bigloop B <<<<<-\|" );/
		}
		ELSE
		{
		push_wmops( "(IDR) LOOP1 PSDs PATH3 B3.3 bigloop ELSE <<<<<-\|" );
		push_wmops( "(IDR) LOOP1 PSDs PATH3 B3.3 bigloop C <<<<<-\|" );
		Word32 sqr_inp, mpy_diff, sqr;
		Word16 sqr_exp;
		/Diffuseness modellling nrg compensation/
		@@ -1037,7 +1081,8 @@ void ivas_dirac_dec_output_synthesis_process_slot_fx(
		move16();
		}
		}
		push_wmops( "(IDR) LOOP1 PSDs PATH3 B3.3 bigloop ELSE nfreqbds <<<<<<-\|" );
		pop_wmops(); /push_wmops( "(IDR) LOOP1 PSDs PATH3 B3.3 bigloop C <<<<<<-\|" );/
		push_wmops( "(IDR) LOOP1 PSDs PATH3 B3.3 bigloop D <<<<<<-\|" );
		FOR( ; k < num_freq_bands; k++ )
		{
		mpy_diff = Mpy_32_32( diffuseness[k], L_sub( h_dirac_output_synthesis_params->diffuse_compensation_factor_decorr_fx, ONE_IN_Q29 /1 Q29/ ) ); // Q = q_diffuseness - 1
		@@ -1071,8 +1116,8 @@ void ivas_dirac_dec_output_synthesis_process_slot_fx(
		move16();
		}
		}
		pop_wmops(); /push_wmops( "(IDR) LOOP1 PSDs PATH3 B3.3 bigloop ELSE nfreqbds <<<<<<-\|" );/
		pop_wmops();/push_wmops( "(IDR) LOOP1 PSDs PATH3 B3.3 bigloop ELSE <<<<<-\|" );//

		pop_wmops();/push_wmops( "(IDR) LOOP1 PSDs PATH3 B3.3 bigloop D <<<<<-\|" );//
		}
		}
		pop_wmops(); /push_wmops( "(IDR) LOOP1 PSDs PATH3 B3.3 bigloop <<<<-\|" );//
		@@ -3066,24 +3111,38 @@ void ivas_dirac_dec_compute_directional_responses_fx(
		exp_direct_response_dir2 = 0;
		move16();

		push_wmops( "(IDR PATH3 B3.1) HOA3 PANNING IF -- getResponse" );
		IF( p_Rmat != 0 )
		{
		push_wmops( "(IDR PATH3 B3.1) HOA3 PANNING IF -- getResponse_SPLIT_FX" );
		ivas_dirac_dec_get_response_split_order_fx( azimuth[k], elevation[k], direct_response_hoa_fx, shd_rot_max_order, p_Rmat, &Q_direct_response_hoa );

		IF( hodirac_flag )
		{
		ivas_dirac_dec_get_response_split_order_fx( azimuth2[k], elevation2[k], direct_response_dir2_fx, shd_rot_max_order, p_Rmat, &Q_direct_response_dir2 );
		}
		pop_wmops(); /push_wmops( "(IDR PATH3 B3.1) HOA3 PANNING IF -- getResponse_SPLIT_FX" );/
		}
		ELSE
		{
		push_wmops( "(IDR PATH3 B3.1) HOA3 PANNING IF -- getResponse__FX" );
		#ifdef FIX_1072_SPEEDUP_ivas_dirac_dec_get_response_fx
		ivas_dirac_dec_get_response_fx_29( azimuth[k], elevation[k], direct_response_hoa_fx, hDirACRend->hOutSetup.ambisonics_order);
		#else
		ivas_dirac_dec_get_response_fx( azimuth[k], elevation[k], direct_response_hoa_fx, hDirACRend->hOutSetup.ambisonics_order, Q_direct_response_hoa );
		#endif

		IF( hodirac_flag )
		{
		#ifdef FIX_1072_SPEEDUP_ivas_dirac_dec_get_response_fx
		ivas_dirac_dec_get_response_fx_29( azimuth2[k], elevation2[k], direct_response_dir2_fx, hDirACRend->hOutSetup.ambisonics_order);
		#else
		ivas_dirac_dec_get_response_fx( azimuth2[k], elevation2[k], direct_response_dir2_fx, hDirACRend->hOutSetup.ambisonics_order, Q_direct_response_dir2 );
		#endif
		}
		pop_wmops(); /push_wmops( "(IDR PATH3 B3.1) HOA3 PANNING IF -- getResponse__FX" );/
		}
		pop_wmops(); /push_wmops( "(IDR PATH3 B3.1) HOA3 PANNING IF -- getResponse" );/

		test();
		test();
		@@ -3091,16 +3150,19 @@ void ivas_dirac_dec_compute_directional_responses_fx(
		test();
		IF( masa_band_mapping == NULL && EQ_16( hDirACRend->synthesisConf, DIRAC_SYNTHESIS_GAIN_SHD ) )
		{
		push_wmops( "(IDR PATH3 B3.1) HOA3 PANNING IF -- PATH1" );
		mvr2r_inc_fixed( direct_response_hoa_fx, 1, &hDirACRend->h_output_synthesis_psd_state.direct_responses_fx[k], hSpatParamRendCom->num_freq_bands, num_channels_dir ); /Q_direct_response_hoa/

		IF( hodirac_flag )
		{
		mvr2r_inc_fixed( direct_response_dir2_fx, 1, &hDirACRend->h_output_synthesis_psd_state.direct_responses_fx[k + hSpatParamRendCom->num_freq_bands * num_channels_dir], hSpatParamRendCom->num_freq_bands, num_channels_dir ); /Q_direct_response_dir2/
		}
		pop_wmops(); /push_wmops( "(IDR PATH3 B3.1) HOA3 PANNING IF -- PATH1" );/
		}
		ELSE IF( ( ( EQ_16( hDirACRend->synthesisConf, DIRAC_SYNTHESIS_GAIN_SHD ) ) && ( masa_band_mapping != NULL ) ) \|\|
		EQ_16( hDirACRend->synthesisConf, DIRAC_SYNTHESIS_PSD_SHD ) \|\| EQ_16( hDirACRend->synthesisConf, DIRAC_SYNTHESIS_MONO ) )
		{
		push_wmops( "(IDR PATH3 B3.1) HOA3 PANNING IF -- PATH2" );
		/* Synthesize the first direction */
		IF( GT_16( Q_direct_response_hoa, Q29 ) )
		{
		@@ -3347,6 +3409,7 @@ void ivas_dirac_dec_compute_directional_responses_fx(
		}

		mvr2r_inc_fixed( direct_response_fx, 1, &hDirACRend->h_output_synthesis_psd_state.direct_responses_fx[k], hSpatParamRendCom->num_freq_bands, num_channels_dir ); /q29/
		pop_wmops(); /push_wmops( "(IDR PATH3 B3.1) HOA3 PANNING IF -- PATH2" );/
		}
		ELSE
		{