Merge branch 'optimizations_renderer_sba_enc' into 'main' (0f6d8180) · Commits · SA4 / Audio / IVAS BASOP

lib_com/ivas_fb_mixer_fx.c

+36 −0

Original line number	Diff line number	Diff line
		@@ -759,6 +759,13 @@ void ivas_fb_mixer_process_fx(
		pMdft_out_fx[0] = Mdft_out_0_fx;
		pMdft_out_fx[1] = Mdft_out_1_fx;

		#ifdef OPT_SBA_ENC_V1_BE
		Word16 total_guard = find_guarded_bits_fx( num_bands );
		Word16 total_guard_2 = find_guarded_bits_fx( shl( frame_len, 1 ) );
		Word16 tmp_q = sub( sub( *q_mixer_mat_fx, total_guard ), 32 ); // Q30 + hFbMixer->q_prior_mixer_fx - 31 - total_guard - 31
		Word16 len = shl( frame_len, 1 );
		Word16 res_q, q_check;
		#endif
		FOR( ch = ( hFbMixer->fb_cfg->active_w_mixing == 0 ); ch < hFbMixer->fb_cfg->num_out_chans; ch++ )
		{
		/* Run a loop of 2 to calculate current frame's filterbank output and prev frame's output */
		@@ -773,6 +780,12 @@ void ivas_fb_mixer_process_fx(
		IF( in_out_mixer_map[ch][j] != 0 )
		{

		#ifdef OPT_SBA_ENC_V1_BE
		res_q = add( tmp_q, hFbMixer->q_ppFilterbank_inFR[j] );
		q_check = s_min( q_pOut_fr_fx, res_q );
		scale_sig32( pOut_fr_re_fx, frame_len, sub( q_check, q_pOut_fr_fx ) );
		scale_sig32( pOut_fr_im_fx, frame_len, sub( q_check, q_pOut_fr_fx ) );
		#endif
		Word32 filterbank_mixer_bins_re_fx[L_FRAME48k];
		Word32 filterbank_mixer_bins_im_fx[L_FRAME48k];
		Word32 *pFb_inFR_re_fx = hFbMixer->ppFilterbank_inFR_re_fx[j]; // Q(hFbMixer->q_ppFilterbank_inFR_re_fx)
		@@ -781,8 +794,10 @@ void ivas_fb_mixer_process_fx(
		set_zero_fx( filterbank_mixer_bins_re_fx, frame_len );
		set_zero_fx( filterbank_mixer_bins_im_fx, frame_len );

		#ifndef OPT_SBA_ENC_V1_BE
		Word16 total_guard = find_guarded_bits_fx( num_bands );
		move16();
		#endif
		FOR( i = 0; i < num_bands; i++ )
		{
		Word16 start_offset = pFb->fb_consts.pFilterbank_bins_start_offset[i];
		@@ -808,15 +823,19 @@ void ivas_fb_mixer_process_fx(
		hFbMixer->prior_mixer_fx[ch][j][i] = mixer_mat_fx[ch][j][i]; // Q(q_mixer_mat_fx)
		move32();
		}
		#ifndef OPT_SBA_ENC_V1_BE
		Word16 res_q = 0;
		move16();
		#endif
		FOR( k = 0; k < frame_len; k++ )
		{
		Word32 temp_out_re_fx, temp_out_im_fx;

		ivas_cmult_fix( filterbank_mixer_bins_re_fx[k], filterbank_mixer_bins_im_fx[k], pFb_inFR_re_fx[k],
		pFb_inFR_im_fx[k], &temp_out_re_fx, &temp_out_im_fx );
		#ifndef OPT_SBA_ENC_V1_BE
		res_q = sub( add( sub( sub( add( 30, *q_mixer_mat_fx ), 31 ), total_guard ), hFbMixer->q_ppFilterbank_inFR[j] ), 31 );

		Word16 q_check = s_min( q_pOut_fr_fx, res_q );
		IF( NE_16( q_check, q_pOut_fr_fx ) )
		{
		@@ -825,31 +844,48 @@ void ivas_fb_mixer_process_fx(
		pOut_fr_im_fx[k] = L_shr( pOut_fr_im_fx[k], sub( q_pOut_fr_fx, q_check ) ); // q_pOut_fr_fx -> q_check
		move32();
		}
		#endif
		IF( NE_16( q_check, res_q ) )
		{
		temp_out_re_fx = L_shr( temp_out_re_fx, sub( res_q, q_check ) ); // res_q -> q_check
		temp_out_im_fx = L_shr( temp_out_im_fx, sub( res_q, q_check ) ); // res_q -> q_check
		}
		#ifndef OPT_SBA_ENC_V1_BE
		res_q = q_check;
		move16();
		#endif

		pOut_fr_re_fx[k] = L_add_sat( pOut_fr_re_fx[k], temp_out_re_fx ); // res_q
		move32();
		pOut_fr_im_fx[k] = L_add_sat( pOut_fr_im_fx[k], temp_out_im_fx ); // res_q
		move32();
		}
		#ifdef OPT_SBA_ENC_V1_BE
		q_pOut_fr_fx = q_check;
		#else
		q_pOut_fr_fx = res_q;
		move16();
		#endif
		}
		}
		#ifndef OPT_SBA_ENC_V1_BE
		Word16 scale = sub( s_min( L_norm_arr( pOut_fr_re_fx, frame_len ), L_norm_arr( pOut_fr_im_fx, frame_len ) ), find_guarded_bits_fx( shl( frame_len, 1 ) ) );
		#else
		Word16 scale = sub( s_min( L_norm_arr( pOut_fr_re_fx, frame_len ), L_norm_arr( pOut_fr_im_fx, frame_len ) ), total_guard_2 );
		#endif
		scale_sig32( pOut_fr_re_fx, frame_len, scale );
		scale_sig32( pOut_fr_im_fx, frame_len, scale );
		ivas_imdft_fx( pOut_fr_re_fx, pOut_fr_im_fx, pMdft_out_fx[hist], frame_len );
		q_pMdft_out_fx[hist] = add( q_pOut_fr_fx, scale );
		move16();
		}
		#ifdef OPT_SBA_ENC_V1_BE
		scale_sig32( pMdft_out_fx[0], len, sub( s_min( q_pMdft_out_fx[0], q_pMdft_out_fx[1] ), q_pMdft_out_fx[0] ) );
		scale_sig32( pMdft_out_fx[1], len, sub( s_min( q_pMdft_out_fx[0], q_pMdft_out_fx[1] ), q_pMdft_out_fx[1] ) );
		#else
		scale_sig32( pMdft_out_fx[0], shl( frame_len, 1 ), sub( s_min( q_pMdft_out_fx[0], q_pMdft_out_fx[1] ), q_pMdft_out_fx[0] ) );
		scale_sig32( pMdft_out_fx[1], shl( frame_len, 1 ), sub( s_min( q_pMdft_out_fx[0], q_pMdft_out_fx[1] ), q_pMdft_out_fx[1] ) );
		#endif
		ivas_fb_mixer_cross_fading_fx( hFbMixer, ppOut_pcm_fx, pMdft_out_fx[0], pMdft_out_fx[1], ch, frame_len, frame_len );
		q_ppOut_pcm_fx[ch] = s_min( q_pMdft_out_fx[0], q_pMdft_out_fx[1] );
		move16();

lib_com/ivas_spar_com_fx.c

+75 −3

Original line number	Diff line number	Diff line
		@@ -1995,15 +1995,23 @@ static void ivas_calc_post_pred_per_band_enc_fx(
		Word16 *q_postpred_cov_re )
		{
		Word16 i, j, k;
		#ifndef OPT_SBA_ENC_V1_BE
		Word32 dmx_mat_conj[IVAS_SPAR_MAX_CH][IVAS_SPAR_MAX_CH];
		Word32 temp_mat[IVAS_SPAR_MAX_CH][IVAS_SPAR_MAX_CH];
		Word16 temp_mat_e[IVAS_SPAR_MAX_CH][IVAS_SPAR_MAX_CH];
		Word16 q_postpred_cov_re_buf[IVAS_SPAR_MAX_CH][IVAS_SPAR_MAX_CH];
		Word16 min_val;
		#else
		Word16 max_val;
		Word16 temp;
		Word32 temp_add;
		Word16 e_postpred_cov_re_buf[IVAS_SPAR_MAX_CH][IVAS_SPAR_MAX_CH];
		#endif
		Word32 temp_mat[IVAS_SPAR_MAX_CH][IVAS_SPAR_MAX_CH];
		Word16 temp_mat_e[IVAS_SPAR_MAX_CH][IVAS_SPAR_MAX_CH];
		Word32 tmp_re;
		Word16 tmp_q;
		Word16 tmp_e;

		#ifndef OPT_SBA_ENC_V1_BE
		FOR( i = 0; i < num_ch; i++ )
		{
		FOR( j = 0; j < num_ch; j++ )
		@@ -2013,25 +2021,40 @@ static void ivas_calc_post_pred_per_band_enc_fx(
		}
		}


		FOR( i = 0; i < num_ch; i++ )
		{
		set32_fx( temp_mat[i], 0, num_ch );
		set32_fx( postpred_cov_re[i], 0, num_ch );
		}
		#endif

		#ifdef OPT_SBA_ENC_V1_BE
		max_val = MIN16B;
		move16();
		temp = sub( 62, q_mixer_mat );
		#else
		min_val = MAX16B;
		move16();
		#endif

		/* num_ch x num_ch mult */
		FOR( i = 0; i < num_ch; i++ )
		{
		FOR( j = 0; j < num_ch; j++ )
		{
		#ifndef OPT_SBA_ENC_V1_BE
		temp_mat[i][j] = 0;
		#else
		temp_add = 0;
		#endif
		move32();
		temp_mat_e[i][j] = 0;
		move16();

		FOR( k = 0; k < num_ch; k++ )
		{
		#ifndef OPT_SBA_ENC_V1_BE
		tmp_re = Mpy_32_32( cov_real[i][k][band_idx], dmx_mat_conj[k][j] );
		tmp_e = sub( 62, add( q_cov_real[i][k][band_idx], q_mixer_mat ) );
		IF( tmp_re )
		@@ -2039,34 +2062,71 @@ static void ivas_calc_post_pred_per_band_enc_fx(
		temp_mat[i][j] = BASOP_Util_Add_Mant32Exp( temp_mat[i][j], temp_mat_e[i][j], tmp_re, tmp_e, &temp_mat_e[i][j] );
		move32();
		}
		#else
		tmp_re = Mpy_32_32( cov_real[i][k][band_idx], mixer_mat[j][k][band_idx] );
		tmp_e = sub( temp, q_cov_real[i][k][band_idx] );
		IF( tmp_re )
		{
		temp_add = BASOP_Util_Add_Mant32Exp( temp_add, temp_mat_e[i][j], tmp_re, tmp_e, &temp_mat_e[i][j] );
		}
		#endif
		}
		#ifdef OPT_SBA_ENC_V1_BE
		temp_mat[i][j] = temp_add;
		move32();
		#endif
		}
		}

		tmp_e = sub( Q31, q_mixer_mat );
		/* num_ch x num_ch mult */
		FOR( i = 0; i < num_ch; i++ )
		{
		FOR( j = i; j < num_ch; j++ )
		{
		#ifdef OPT_SBA_ENC_V1_BE
		temp_add = 0;
		move32();
		e_postpred_cov_re_buf[i][j] = sub( 31, *q_postpred_cov_re );
		#else
		q_postpred_cov_re_buf[i][j] = *q_postpred_cov_re;
		#endif
		move16();
		FOR( k = 0; k < num_ch; k++ )
		{
		tmp_re = Mpy_32_32( mixer_mat[i][k][band_idx], temp_mat[k][j] );
		#ifdef OPT_SBA_ENC_V1_BE
		tmp_q = add( tmp_e, temp_mat_e[k][j] );
		#else
		tmp_q = sub( q_mixer_mat, temp_mat_e[k][j] );
		#endif
		IF( tmp_re )
		{
		#ifndef OPT_SBA_ENC_V1_BE
		tmp_e = sub( 31, q_postpred_cov_re_buf[i][j] );
		postpred_cov_re[i][j] = BASOP_Util_Add_Mant32Exp( postpred_cov_re[i][j], tmp_e, tmp_re, sub( Q31, tmp_q ), &tmp_e );
		move32();
		q_postpred_cov_re_buf[i][j] = sub( 31, tmp_e );
		move16();
		#else
		temp_add = BASOP_Util_Add_Mant32Exp( temp_add, e_postpred_cov_re_buf[i][j], tmp_re, tmp_q, &e_postpred_cov_re_buf[i][j] );
		#endif
		}
		}
		#ifndef OPT_SBA_ENC_V1_BE
		IF( postpred_cov_re[i][j] )
		{
		min_val = s_min( min_val, q_postpred_cov_re_buf[i][j] );
		#else
		if ( temp_add )
		{
		max_val = s_max( max_val, e_postpred_cov_re_buf[i][j] );
		#endif
		}
		#ifdef OPT_SBA_ENC_V1_BE
		postpred_cov_re[i][j] = temp_add;
		move32();
		#endif
		}
		}

		@@ -2077,7 +2137,11 @@ static void ivas_calc_post_pred_per_band_enc_fx(
		{
		IF( postpred_cov_re[i][j] )
		{
		#ifdef OPT_SBA_ENC_V1_BE
		postpred_cov_re[i][j] = L_shr( postpred_cov_re[i][j], sub( max_val, e_postpred_cov_re_buf[i][j] ) );
		#else
		postpred_cov_re[i][j] = L_shl( postpred_cov_re[i][j], sub( min_val, q_postpred_cov_re_buf[i][j] ) );
		#endif
		move32();
		}
		}
		@@ -2092,13 +2156,21 @@ static void ivas_calc_post_pred_per_band_enc_fx(
		move32();
		}
		}

		#ifdef OPT_SBA_ENC_V1_BE
		if ( EQ_16( max_val, MIN16B ) )
		{
		max_val = 0;
		move16();
		}
		*q_postpred_cov_re = sub( Q31, max_val );
		#else
		if ( EQ_16( min_val, MAX16B ) )
		{
		min_val = Q31;
		move16();
		}
		*q_postpred_cov_re = min_val;
		#endif
		move16();

		return;

lib_com/options.h

+3 −0

Original line number	Diff line number	Diff line
		@@ -75,6 +75,9 @@
		#define FIX_1379_MASA_ANGLE_ROUND

		/* Note: each compile switch (FIX_1101_...) is independent from the other ones */
		#define OPT_SBA_ENC_V1_BE
		#define OPT_BIN_RENDERER_V1
		#define OPT_BIN_RENDERER_V2
		#define OPT_STEREO_32KBPS_V1 /* Optimization made in stereo decoding path for 32kbps decoding */
		#define OPT_AVOID_STATE_BUF_RESCALE /* Optimization made to avoid rescale of synth state buffer */
		#define FIX_1310_SPEEDUP_ivas_dirac_dec_get_response_fx /FhG: WMOPS tuning, nonbe/

lib_dec/ivas_binRenderer_internal_fx.c

+139 −0

Original line number	Diff line number	Diff line
		@@ -54,7 +54,137 @@
		*
		*
		-------------------------------------------------------------------------/
		#ifdef OPT_BIN_RENDERER_V2
		static void ivas_binRenderer_filterModule_fx(
		Word32 out_Conv_CLDFB_real[BINAURAL_CHANNELS][MAX_PARAM_SPATIAL_SUBFRAMES][CLDFB_NO_CHANNELS_MAX], /* o : real part of Binaural signals Q6 */
		Word32 out_Conv_CLDFB_imag[BINAURAL_CHANNELS][MAX_PARAM_SPATIAL_SUBFRAMES][CLDFB_NO_CHANNELS_MAX], /* o : imag part of Binaural signals Q6 */
		Word32 CLDFB_real[][MAX_PARAM_SPATIAL_SUBFRAMES][CLDFB_NO_CHANNELS_MAX], /* i : real part of LS signals Q_curr*/
		Word32 CLDFB_imag[][MAX_PARAM_SPATIAL_SUBFRAMES][CLDFB_NO_CHANNELS_MAX], /* i : imag part of LS signals Q_curr*/
		const Word16 numTimeSlots, /* i : number of time slots to process */
		BINAURAL_RENDERER_HANDLE hBinRenderer, /* i/o: fastconv binaural renderer handle */
		Word16 Q_curr )
		{
		Word16 bandIdx, k, chIdx, tapIdx;
		Word32 filterStatesLeftRealPtr_fx, filterStatesLeftImagPtr_fx;
		Word16 Q_filterStates;
		const Word32 filterTapsLeftRealPtr_fx, filterTapsLeftImagPtr_fx, filterTapsRightRealPtr_fx, filterTapsRightImagPtr_fx;
		Word16 shift_q;
		Word16 shift_q6 = sub( -29 + 6, Q_curr );
		Q_filterStates = hBinRenderer->hBinRenConvModule->Q_filterStatesLeft;
		move16();

		// to be checked: feasibility with 32 bit buffers
		Word64 Cldfb_RealBuffer_64fx[BINAURAL_CHANNELS][MAX_PARAM_SPATIAL_SUBFRAMES];
		Word64 Cldfb_ImagBuffer_64fx[BINAURAL_CHANNELS][MAX_PARAM_SPATIAL_SUBFRAMES];

		FOR( bandIdx = 0; bandIdx < hBinRenderer->conv_band; bandIdx++ )
		{
		set64_fx( &Cldfb_RealBuffer_64fx[0][0], 0, BINAURAL_CHANNELS * MAX_PARAM_SPATIAL_SUBFRAMES );
		set64_fx( &Cldfb_ImagBuffer_64fx[0][0], 0, BINAURAL_CHANNELS * MAX_PARAM_SPATIAL_SUBFRAMES );

		FOR( chIdx = 0; chIdx < hBinRenderer->nInChannels; chIdx++ )
		{
		filterStatesLeftRealPtr_fx = (Word32 *) &( hBinRenderer->hBinRenConvModule->filterStatesLeftReal_fx[bandIdx][chIdx][0] );
		filterStatesLeftImagPtr_fx = (Word32 *) &( hBinRenderer->hBinRenConvModule->filterStatesLeftImag_fx[bandIdx][chIdx][0] );

		filterTapsLeftRealPtr_fx = hBinRenderer->hBinRenConvModule->filterTapsLeftReal_fx[bandIdx][chIdx]; // Q29
		filterTapsLeftImagPtr_fx = hBinRenderer->hBinRenConvModule->filterTapsLeftImag_fx[bandIdx][chIdx]; // Q29
		filterTapsRightRealPtr_fx = hBinRenderer->hBinRenConvModule->filterTapsRightReal_fx[bandIdx][chIdx]; // Q29
		filterTapsRightImagPtr_fx = hBinRenderer->hBinRenConvModule->filterTapsRightImag_fx[bandIdx][chIdx]; // Q29

		FOR( k = 0; k < numTimeSlots; k++ )
		{
		Word64 outRealLeft_fx = 0, outRealRight_fx = 0, outImagLeft_fx = 0, outImagRight_fx = 0;
		move64();
		move64();
		move64();
		move64();

		FOR( tapIdx = hBinRenderer->hBinRenConvModule->numTapsArray[bandIdx] - 1; tapIdx > 0; tapIdx-- )
		{
		filterStatesLeftRealPtr_fx[tapIdx] = filterStatesLeftRealPtr_fx[tapIdx - 1];
		move32();
		filterStatesLeftImagPtr_fx[tapIdx] = filterStatesLeftImagPtr_fx[tapIdx - 1];
		move32();
		Word32 neg_filterStatesLeftImagPtr_fx = L_negate( filterStatesLeftImagPtr_fx[tapIdx] );


		outRealLeft_fx = W_mac_32_32( outRealLeft_fx, filterStatesLeftRealPtr_fx[tapIdx], filterTapsLeftRealPtr_fx[tapIdx] ); // Q30 + Q_filterStates
		outRealLeft_fx = W_mac_32_32( outRealLeft_fx, neg_filterStatesLeftImagPtr_fx, filterTapsLeftImagPtr_fx[tapIdx] ); // Q30 + Q_filterStates

		outImagLeft_fx = W_mac_32_32( outImagLeft_fx, filterStatesLeftRealPtr_fx[tapIdx], filterTapsLeftImagPtr_fx[tapIdx] ); // Q30 + Q_filterStates
		outImagLeft_fx = W_mac_32_32( outImagLeft_fx, filterStatesLeftImagPtr_fx[tapIdx], filterTapsLeftRealPtr_fx[tapIdx] ); // Q30 + Q_filterStates

		outRealRight_fx = W_mac_32_32( outRealRight_fx, filterStatesLeftRealPtr_fx[tapIdx], filterTapsRightRealPtr_fx[tapIdx] ); // Q30 + Q_filterStates
		outRealRight_fx = W_mac_32_32( outRealRight_fx, neg_filterStatesLeftImagPtr_fx, filterTapsRightImagPtr_fx[tapIdx] ); // Q30 + Q_filterStates

		outImagRight_fx = W_mac_32_32( outImagRight_fx, filterStatesLeftRealPtr_fx[tapIdx], filterTapsRightImagPtr_fx[tapIdx] ); // Q30 + Q_filterStates
		outImagRight_fx = W_mac_32_32( outImagRight_fx, filterStatesLeftImagPtr_fx[tapIdx], filterTapsRightRealPtr_fx[tapIdx] ); // Q30 + Q_filterStates
		}

		shift_q = add( sub( Q_filterStates, Q_curr ), 1 );

		IF( shift_q != 0 )
		{
		outRealLeft_fx = W_shr( outRealLeft_fx, shift_q ); // Q_curr + Q29
		outImagLeft_fx = W_shr( outImagLeft_fx, shift_q ); // Q_curr + Q29
		outRealRight_fx = W_shr( outRealRight_fx, shift_q ); // Q_curr + Q29
		outImagRight_fx = W_shr( outImagRight_fx, shift_q ); // Q_curr + Q29
		hBinRenderer->hBinRenConvModule->Q_filterStatesLeft = Q_curr;
		move16();
		}

		filterStatesLeftRealPtr_fx[0] = CLDFB_real[chIdx][k][bandIdx];
		move32();
		filterStatesLeftImagPtr_fx[0] = CLDFB_imag[chIdx][k][bandIdx];
		move32();


		/* Left Real and Imag */

		Word32 temp1 = L_shr( filterStatesLeftRealPtr_fx[0], 1 ); // Q_curr -1
		Word32 temp2 = L_shr( filterStatesLeftImagPtr_fx[0], 1 ); // Q_curr -1
		Word32 neg_temp2 = L_negate( temp2 ); // Q_curr -1


		outRealLeft_fx = W_mac_32_32( outRealLeft_fx, temp1, filterTapsLeftRealPtr_fx[0] );
		outRealLeft_fx = W_mac_32_32( outRealLeft_fx, neg_temp2, filterTapsLeftImagPtr_fx[0] );
		Cldfb_RealBuffer_64fx[0][k] = W_add( Cldfb_RealBuffer_64fx[0][k], outRealLeft_fx ); // Q29 + Q_curr
		move64();

		outImagLeft_fx = W_mac_32_32( outImagLeft_fx, temp1, filterTapsLeftImagPtr_fx[0] );
		outImagLeft_fx = W_mac_32_32( outImagLeft_fx, temp2, filterTapsLeftRealPtr_fx[0] );
		Cldfb_ImagBuffer_64fx[0][k] = W_add( Cldfb_ImagBuffer_64fx[0][k], outImagLeft_fx ); // Q29 + Q_curr
		move64();

		/* Right Real and Imag */
		outRealRight_fx = W_mac_32_32( outRealRight_fx, temp1, filterTapsRightRealPtr_fx[0] );
		outRealRight_fx = W_mac_32_32( outRealRight_fx, neg_temp2, filterTapsRightImagPtr_fx[0] );
		Cldfb_RealBuffer_64fx[1][k] = W_add( Cldfb_RealBuffer_64fx[1][k], outRealRight_fx ); // Q29 + Q_curr
		move64();

		outImagRight_fx = W_mac_32_32( outImagRight_fx, temp1, filterTapsRightImagPtr_fx[0] );
		outImagRight_fx = W_mac_32_32( outImagRight_fx, temp2, filterTapsRightRealPtr_fx[0] );
		Cldfb_ImagBuffer_64fx[1][k] = W_add( Cldfb_ImagBuffer_64fx[1][k], outImagRight_fx ); // Q29 + Q_curr
		move64();
		}
		}
		FOR( chIdx = 0; chIdx < BINAURAL_CHANNELS; chIdx++ )
		{
		FOR( k = 0; k < MAX_PARAM_SPATIAL_SUBFRAMES; k++ )
		{
		out_Conv_CLDFB_real[chIdx][k][bandIdx] = W_shl_sat_l( Cldfb_RealBuffer_64fx[chIdx][k], shift_q6 ); // Q6
		move32();
		out_Conv_CLDFB_imag[chIdx][k][bandIdx] = W_shl_sat_l( Cldfb_ImagBuffer_64fx[chIdx][k], shift_q6 ); // Q6
		move32();
		}
		}
		}


		return;
		}
		#else /* OPT_BIN_RENDERER_V2 */
		static void ivas_binRenderer_filterModule_fx(
		Word64 out_Conv_CLDFB_real[BINAURAL_CHANNELS][MAX_PARAM_SPATIAL_SUBFRAMES][CLDFB_NO_CHANNELS_MAX], /* o : real part of Binaural signals Q29 */
		Word64 out_Conv_CLDFB_imag[BINAURAL_CHANNELS][MAX_PARAM_SPATIAL_SUBFRAMES][CLDFB_NO_CHANNELS_MAX], /* o : imag part of Binaural signals Q29 */
		@@ -164,6 +294,7 @@ static void ivas_binRenderer_filterModule_fx(

		return;
		}
		#endif /* OPT_BIN_RENDERER_V2 */

		/*-------------------------------------------------------------------------
		* ivas_binRenderer_convModuleOpen()
		@@ -1482,9 +1613,11 @@ void ivas_binRenderer_fx(
		)
		{
		Word16 chIdx, i, j, k;
		#ifndef OPT_BIN_RENDERER_V2
		// to be checked: feasibility with 32 bit buffers
		Word64 Cldfb_RealBuffer_Binaural_64fx[BINAURAL_CHANNELS][MAX_PARAM_SPATIAL_SUBFRAMES][CLDFB_NO_CHANNELS_MAX];
		Word64 Cldfb_ImagBuffer_Binaural_64fx[BINAURAL_CHANNELS][MAX_PARAM_SPATIAL_SUBFRAMES][CLDFB_NO_CHANNELS_MAX];
		#endif /* OPT_BIN_RENDERER_V2 */
		push_wmops( "fastconv_binaural_rendering" );

		/* Compute Convolution */
		@@ -1495,8 +1628,10 @@ void ivas_binRenderer_fx(
		{
		set32_fx( Cldfb_RealBuffer_Binaural_fx[chIdx][k], 0, CLDFB_NO_CHANNELS_MAX );
		set32_fx( Cldfb_ImagBuffer_Binaural_fx[chIdx][k], 0, CLDFB_NO_CHANNELS_MAX );
		#ifndef OPT_BIN_RENDERER_V2
		set64_fx( Cldfb_RealBuffer_Binaural_64fx[chIdx][k], 0, CLDFB_NO_CHANNELS_MAX );
		set64_fx( Cldfb_ImagBuffer_Binaural_64fx[chIdx][k], 0, CLDFB_NO_CHANNELS_MAX );
		#endif /* OPT_BIN_RENDERER_V2 */
		}
		}

		@@ -1536,6 +1671,9 @@ void ivas_binRenderer_fx(
		ivas_sba2mc_cldfb_fixed( *( hBinRenderer->hInputSetup ), RealBuffer_fx, ImagBuffer_fx,
		hBinRenderer->nInChannels, hBinRenderer->conv_band, numTimeSlots, hBinRenderer->hoa_dec_mtx );
		}
		#ifdef OPT_BIN_RENDERER_V2
		ivas_binRenderer_filterModule_fx( Cldfb_RealBuffer_Binaural_fx, Cldfb_ImagBuffer_Binaural_fx, RealBuffer_fx, ImagBuffer_fx, numTimeSlots, hBinRenderer, *Q_in );
		#else /* OPT_BIN_RENDERER_V2 */
		ivas_binRenderer_filterModule_fx( Cldfb_RealBuffer_Binaural_64fx, Cldfb_ImagBuffer_Binaural_64fx, RealBuffer_fx, ImagBuffer_fx, numTimeSlots, hBinRenderer, *Q_in );

		FOR( i = 0; i < BINAURAL_CHANNELS; i++ )
		@@ -1551,6 +1689,7 @@ void ivas_binRenderer_fx(
		}
		}
		}
		#endif /* OPT_BIN_RENDERER_V2 */
		/* Obtain the binaural dmx and compute the reverb */
		IF( hBinRenderer->hReverb != NULL )
		{

lib_dec/ivas_dirac_dec_fx.c

+17 −2

Original line number	Diff line number	Diff line
		@@ -2227,6 +2227,7 @@ void ivas_dirac_dec_render_sf_fx(
		}
		}

		#ifndef OPT_BIN_RENDERER_V2
		FOR( i = 0; i < BINAURAL_CHANNELS; i++ )
		{
		FOR( Word16 j = 0; j < MAX_PARAM_SPATIAL_SUBFRAMES; j++ )
		@@ -2235,6 +2236,7 @@ void ivas_dirac_dec_render_sf_fx(
		set32_fx( Cldfb_RealBuffer_Binaural_fx[i][j], 0, CLDFB_NO_CHANNELS_MAX );
		}
		}
		#endif /* OPT_BIN_RENDERER_V2 */

		/* local copies of azi, ele, diffuseness */
		Word16 azimuth[CLDFB_NO_CHANNELS_MAX];
		@@ -3973,9 +3975,19 @@ void ivas_dirac_dec_render_sf_fx(
		{
		Word32 g_fx;
		w2_fx = sub( MAX16B, *w1_fx );
		#ifdef OPT_BIN_RENDERER_V2
		g_fx = Madd_32_16( Mpy_32_16_1( gain_fx, *w1_fx ), prev_gain_fx, w2_fx ); // Q15
		#else /* OPT_BIN_RENDERER_V2 */
		g_fx = L_add( Mpy_32_16_1( gain_fx, *w1_fx ), Mpy_32_16_1( prev_gain_fx, w2_fx ) ); // Q15
		#endif /* OPT_BIN_RENDERER_V2 */
		FOR( l = 0; l < hSpatParamRendCom->num_freq_bands; l++ )
		{
		#ifdef OPT_BIN_RENDERER_V2
		Cldfb_RealBuffer_fx[j2][0][k * hSpatParamRendCom->num_freq_bands + l] = Madd_32_32( Cldfb_RealBuffer_fx[j2][0][k * hSpatParamRendCom->num_freq_bands + l], g_fx, *tc_re_fx );
		move32();
		tc_re_fx++;
		Cldfb_ImagBuffer_fx[j2][0][k * hSpatParamRendCom->num_freq_bands + l] = Madd_32_32( Cldfb_ImagBuffer_fx[j2][0][k * hSpatParamRendCom->num_freq_bands + l], g_fx, *tc_im_fx );
		#else /* OPT_BIN_RENDERER_V2 */
		Word32 tmp;
		tmp = Mpy_32_32( g_fx, *tc_re_fx );
		Cldfb_RealBuffer_fx[j2][0][k * hSpatParamRendCom->num_freq_bands + l] = L_add( Cldfb_RealBuffer_fx[j2][0][k * hSpatParamRendCom->num_freq_bands + l], tmp );
		@@ -3983,6 +3995,7 @@ void ivas_dirac_dec_render_sf_fx(
		tc_re_fx++;
		tmp = Mpy_32_32( g_fx, *tc_im_fx );
		Cldfb_ImagBuffer_fx[j2][0][k * hSpatParamRendCom->num_freq_bands + l] = L_add( Cldfb_ImagBuffer_fx[j2][0][k * hSpatParamRendCom->num_freq_bands + l], tmp );
		#endif /* OPT_BIN_RENDERER_V2 */
		move32();
		tc_re_fx++;
		}
		@@ -4012,8 +4025,10 @@ void ivas_dirac_dec_render_sf_fx(
		Cldfb_RealBuffer_Binaural_fx, Cldfb_ImagBuffer_Binaural_fx,
		Cldfb_RealBuffer_fx, Cldfb_ImagBuffer_fx, &input_q );

		#ifndef OPT_BIN_RENDERER_V2
		Scale_sig32( Cldfb_RealBuffer_Binaural_fx[0][0], i_mult( BINAURAL_CHANNELS, i_mult( MAX_PARAM_SPATIAL_SUBFRAMES, CLDFB_NO_CHANNELS_MAX ) ), sub( Q6, input_q ) ); // Q6
		Scale_sig32( Cldfb_ImagBuffer_Binaural_fx[0][0], i_mult( BINAURAL_CHANNELS, i_mult( MAX_PARAM_SPATIAL_SUBFRAMES, CLDFB_NO_CHANNELS_MAX ) ), sub( Q6, input_q ) ); // Q6
		#endif /* OPT_BIN_RENDERER_V2 */

		/* Inverse CLDFB*/
		FOR( ch = 0; ch < st_ivas->hDecoderConfig->nchan_out; ch++ )