LTV test for encoder and renderer optimizations (b9581d08) · Commits · SA4 / Audio / IVAS BASOP

lib_com/cldfb.c

+176 −1

Original line number	Diff line number	Diff line
		@@ -60,6 +60,19 @@ static void cldfb_init_proto_and_twiddles( HANDLE_CLDFB_FILTER_BANK hs );

		static void cldfb_init_proto_and_twiddles_enc_fx( HANDLE_CLDFB_FILTER_BANK hs );

		static void GetEnergyCldfb_ivas_fx( Word32 energyLookahead, /!< o: Q(sf_energyLookahead) \| pointer to the result in the core look-ahead slot /
		Word16 sf_energyLookahead, /!< o: pointer to the scalefactor of the result in the core look-ahead slot */
		const Word16 numLookahead, /!< i: Q0 the number of look-ahead time-slots /
		Word16 *realValues, /!< i: Q(sf_Values) \| the real part of the CLDFB subsamples */
		Word16 *imagValues, /!< i: Q(sf_Values) \| the imaginary part of the CLDFB subsamples */
		Word16 sf_Values, /!< i: scalefactor of the CLDFB subcamples - apply as a negated Exponent /
		Word16 numberBands, /!< i: Q0 \| number of CLDFB bands /
		Word16 numberCols, /!< i: Q0 \| number of CLDFB subsamples /
		Word32 energyHF, /!< o: Q31 \| pointer to HF energy */
		Word16 energyHF_Exp, /!< o: pointer to exponent of HF energy */
		Word32 energyValuesSum, /!< o: Q(2sf_Values-4) \| pointer to sum array of energy values, not initialized/
		Word16 energyValuesSum_Exp, /!< o: pointer to exponents of energyValuesSum, not initialized */
		TEC_ENC_HANDLE hTecEnc );

		/-------------------------------------------------------------------
		* cplxMult()
		@@ -1552,6 +1565,168 @@ void resampleCldfb_ivas_fx(
		return;
		}

		static void GetEnergyCldfb_ivas_fx( Word32 energyLookahead, /!< o: Q(sf_energyLookahead) \| pointer to the result in the core look-ahead slot /
		Word16 sf_energyLookahead, /!< o: pointer to the scalefactor of the result in the core look-ahead slot */
		const Word16 numLookahead, /!< i: Q0 the number of look-ahead time-slots /
		Word16 *realValues, /!< i: Q(sf_Values) \| the real part of the CLDFB subsamples */
		Word16 *imagValues, /!< i: Q(sf_Values) \| the imaginary part of the CLDFB subsamples */
		Word16 sf_Values, /!< i: scalefactor of the CLDFB subcamples - apply as a negated Exponent /
		Word16 numberBands, /!< i: Q0 \| number of CLDFB bands /
		Word16 numberCols, /!< i: Q0 \| number of CLDFB subsamples /
		Word32 energyHF, /!< o: Q31 \| pointer to HF energy */
		Word16 energyHF_Exp, /!< o: pointer to exponent of HF energy */
		Word32 energyValuesSum, /!< o: Q(2sf_Values-4) \| pointer to sum array of energy values, not initialized/
		Word16 energyValuesSum_Exp, /!< o: pointer to exponents of energyValuesSum, not initialized */
		TEC_ENC_HANDLE hTecEnc )
		{
		Word16 j;
		Word16 k;
		Word16 s;
		Word16 sm;
		Word32 nrg;
		Word16 numberColsL;
		Word16 numberBandsM;
		Word16 numberBandsM20;
		Word32 energyValues[CLDFB_NO_COL_MAX][CLDFB_NO_CHANNELS_MAX];
		Word16 energyValuesSumE[CLDFB_NO_CHANNELS_MAX];
		// Word16 freqTable[2] = {20, 40};
		#ifdef BASOP_NOGLOB_DECLARE_LOCAL
		Flag Overflow = 0;
		move32();
		#endif

		FOR( k = 0; k < numberCols; k++ )
		{
		FOR( j = 0; j < numberBands; j++ )
		{
		nrg = L_mult0( realValues[k][j], realValues[k][j] ); // Q(2*sf_Values)
		nrg = L_mac0( nrg, imagValues[k][j], imagValues[k][j] ); // Q(2*sf_Values)

		energyValues[k][j] = nrg;
		move32();
		}
		}

		IF( GE_16( numberBands, freqTable[1] ) && hTecEnc != NULL )
		{
		Word32 *tempEnergyValuesArry[CLDFB_NO_COL_MAX];
		Word16 ScaleX2;
		assert( numberCols == CLDFB_NO_COL_MAX );
		FOR( j = 0; j < numberCols; j++ )
		{
		tempEnergyValuesArry[j] = &energyValues[j][0];
		}

		ScaleX2 = shl( sf_Values, 1 );
		calcHiEnvLoBuff_Fix(
		numberCols,
		freqTable,
		1,
		tempEnergyValuesArry,
		hTecEnc->loBuffer,
		hTecEnc->hiTempEnv,
		ScaleX2 );
		}

		FOR( j = 0; j < numberBands; j++ )
		{
		energyValuesSum[j] = 0;
		move32();
		energyValuesSumE[j] = 31;
		move16();
		FOR( k = 0; k < CLDFB_NO_COL_MAX; k++ )
		{
		nrg = L_shr_r( energyValues[k][j], sub( energyValuesSumE[j], 31 ) ); // Q(2*sf_Values - (energyValuesSumE[j]-31))
		IF( LT_32( L_sub( maxWord32, nrg ), energyValuesSum[j] ) )
		{
		energyValuesSumE[j] = add( energyValuesSumE[j], 1 );
		move16();
		energyValuesSum[j] = L_shr_r( energyValuesSum[j], 1 );
		move32();
		nrg = L_shr_r( nrg, 1 );
		}
		energyValuesSum[j] = L_add( energyValuesSum[j], nrg );
		move32();
		}
		test();
		if ( j == 0 \|\| GT_16( energyValuesSumE[j], *energyValuesSum_Exp ) )
		{
		*energyValuesSum_Exp = energyValuesSumE[j];
		move16();
		}
		}
		FOR( j = 0; j < numberBands; j++ )
		{
		energyValuesSum[j] = L_shr_r( energyValuesSum[j], sub( energyValuesSum_Exp, energyValuesSumE[j] ) ); // Q(energyValuesSum_Exp - (2sf_Values))
		move32();
		}
		energyValuesSum_Exp = sub( energyValuesSum_Exp, shl( sf_Values, 1 ) );
		move16();

		IF( GT_16( numberBands, 20 ) )
		{
		numberBandsM = s_min( numberBands, 40 );
		numberBandsM20 = sub( numberBandsM, 20 );

		numberColsL = sub( numberCols, numLookahead );

		/* sum up CLDFB energy above 8 kHz */
		s = BASOP_util_norm_s_bands2shift( i_mult( numberColsL, numberBandsM20 ) );
		s = sub( s, 4 );
		nrg = 0;
		move32();
		FOR( k = 0; k < numberColsL; k++ )
		{
		FOR( j = 20; j < numberBandsM; j++ )
		{
		nrg = L_add_o( nrg, L_shr_o( energyValues[k][j], s, &Overflow ), &Overflow );
		}
		}

		s = sub( sub( shl( sf_Values, 1 ), 1 ), s );
		sm = sub( s_min( s, *sf_energyLookahead ), 1 );

		*energyHF = L_add( L_shr( nrg, limitScale32( sub( s, sm ) ) ),
		L_shr( energyLookahead, sub( sf_energyLookahead, sm ) ) ); // Q(31-(-nm))
		move32();

		*energyHF_Exp = negate( sm );
		move16();

		/* process look-ahead region */
		s = BASOP_util_norm_s_bands2shift( i_mult( numLookahead, numberBandsM20 ) );
		s = sub( s, 2 );
		nrg = 0;
		move32();
		FOR( k = numberColsL; k < numberCols; k++ )
		{
		FOR( j = 20; j < numberBandsM; j++ )
		{
		nrg = L_add_o( nrg, L_shr_o( energyValues[k][j], s, &Overflow ), &Overflow );
		}
		}

		s = sub( shl( sf_Values, 1 ), s );
		sm = sub( s_min( s, 44 ), 1 );
		BASOP_SATURATE_WARNING_OFF_EVS
		/* nrg + 6.1e-5f => value 0x40000000, scale 44 */
		*energyLookahead = L_add_sat( L_shr_sat( nrg, sub( s, sm ) ),
		L_shr_sat( 0x40000000, s_max( -31, s_min( 31, sub( 44, sm ) ) ) ) );
		move32();
		BASOP_SATURATE_WARNING_ON_EVS
		*sf_energyLookahead = sm;
		move16();

		return;
		}


		*energyHF = 0x40000000;
		move32();
		*energyHF_Exp = 17;
		move16();
		}


		void analysisCldfbEncoder_ivas_fx(
		Encoder_State st, / i/o: encoder state structure */
		@@ -1613,7 +1788,7 @@ void analysisCldfbEncoder_ivas_fx(

		AnalysisPostSpectrumScaling_Fx( st->cldfbAnaEnc, ppBuf_Real, ppBuf_Imag, ppBuf_Real16, ppBuf_Imag16, &enerScale.lb_scale16 );

		GetEnergyCldfb( &st->energyCoreLookahead_Fx, &st->sf_energyCoreLookahead_Fx, 1, ppBuf_Real16, ppBuf_Imag16, enerScale.lb_scale16, st->cldfbAnaEnc->no_channels, st->cldfbAnaEnc->no_col, &st->currEnergyHF_fx, &st->currEnergyHF_e_fx, ppBuf_Ener, enerBuffSum_exp, st->hTECEnc );
		GetEnergyCldfb_ivas_fx( &st->energyCoreLookahead_Fx, &st->sf_energyCoreLookahead_Fx, 1, ppBuf_Real16, ppBuf_Imag16, enerScale.lb_scale16, st->cldfbAnaEnc->no_channels, st->cldfbAnaEnc->no_col, &st->currEnergyHF_fx, &st->currEnergyHF_e_fx, ppBuf_Ener, enerBuffSum_exp, st->hTECEnc );

		return;
		}

lib_com/ivas_fb_mixer_fx.c

+48 −2

Original line number	Diff line number	Diff line
		@@ -759,6 +759,12 @@ 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
		Word16 total_guard_1 = find_guarded_bits_fx( num_bands );
		Word16 total_guard_2 = find_guarded_bits_fx( shl( frame_len, 1 ) );
		Word16 tmp_q = sub( sub( sub( add( 30, *q_mixer_mat_fx ), 31 ), total_guard_1 ), 31 );
		Word16 len = shl( frame_len, 1 );
		#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 */
		@@ -772,7 +778,12 @@ void ivas_fb_mixer_process_fx(
		{
		IF( in_out_mixer_map[ch][j] != 0 )
		{

		#ifdef OPT_SBA_ENC_V1
		Word16 res_q = add(tmp_q, hFbMixer->q_ppFilterbank_inFR[j]);
		Word16 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)
		@@ -781,25 +792,39 @@ 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
		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];
		move16();
		Word16 num_bins = pFb->fb_consts.pFilterbank_bins_per_band[i];
		move16();
		#ifndef OPT_SBA_ENC_V1
		Word32 mixer_const_fx = hFbMixer->prior_mixer_fx[ch][j][i]; // Q(hFbMixer->q_prior_mixer_fx)
		move32();
		#else
		Word32 mixer_const_fx = L_shr( hFbMixer->prior_mixer_fx[ch][j][i], total_guard_1 ); // Q(hFbMixer->q_prior_mixer_fx) - total_guard
		move32();
		#endif
		pFilterbank_bin_to_band_re_fx = pFb->fb_consts.ppFilterbank_FRs_fx[0][i]; // Q30
		pFilterbank_bin_to_band_im_fx = pFb->fb_consts.ppFilterbank_FRs_fx[1][i]; // Q30

		FOR( k = start_offset; k < num_bins + start_offset; k++ )
		{
		#ifndef OPT_SBA_ENC_V1
		filterbank_mixer_bins_re_fx[k] = L_add_sat( filterbank_mixer_bins_re_fx[k], L_shr( Mpy_32_32( *pFilterbank_bin_to_band_re_fx, mixer_const_fx ), total_guard ) ); // Q30 + hFbMixer->q_prior_mixer_fx - 31 - total_guard
		move32();
		filterbank_mixer_bins_im_fx[k] = L_add_sat( filterbank_mixer_bins_im_fx[k], L_shr( Mpy_32_32( *pFilterbank_bin_to_band_im_fx, mixer_const_fx ), total_guard ) ); // Q30 + hFbMixer->q_prior_mixer_fx - 31 - total_guard
		move32();
		#else
		filterbank_mixer_bins_re_fx[k] = Madd_32_32( filterbank_mixer_bins_re_fx[k], *pFilterbank_bin_to_band_re_fx, mixer_const_fx ); // Q30 + hFbMixer->q_prior_mixer_fx - 31 - total_guard
		move32();
		filterbank_mixer_bins_im_fx[k] = Madd_32_32( filterbank_mixer_bins_im_fx[k], *pFilterbank_bin_to_band_im_fx, mixer_const_fx ); // Q30 + hFbMixer->q_prior_mixer_fx - 31 - total_guard
		move32();
		#endif
		/filterbank_mixer_bins_im_fx q 30 /
		/mixer_const_fx q q_ppOut_pcm_fx /
		pFilterbank_bin_to_band_re_fx++;
		@@ -808,15 +833,19 @@ 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
		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 );
		res_q = sub( add( sub( sub( add( 30, *q_mixer_mat_fx ), 31 ), total_guard ), hFbMixer->q_ppFilterbank_inFR[j] ), 31 );
		#ifndef OPT_SBA_ENC_V1
		res_q = sub( add( sub( sub( add( 30, *q_mixer_mat_fx ), 31 ), total_guard_1 ), 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 +854,48 @@ 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
		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
		q_pOut_fr_fx = q_check;
		#else
		q_pOut_fr_fx = res_q;
		move16();
		#endif
		}
		}
		#ifndef OPT_SBA_ENC_V1
		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
		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_prot_fx.h

+1 −1

Original line number	Diff line number	Diff line
		@@ -3341,7 +3341,7 @@ void mctStereoIGF_enc_fx(
		Word32 powerSpec_fx[MCT_MAX_CHANNELS][L_FRAME48k], /* i/o: MDCT^2 + MDST^2 spectrum,or estimate */
		Word16 q_powerSpec[MCT_MAX_CHANNELS], /* i : Q for powSpec_fx */
		Word32 powerSpecMsInv_fx[MCT_MAX_CHANNELS][NB_DIV], / i : same as powerSpec_fx but for inverse spect.*/
		Word16 q_powerSpecMsInv[MCT_MAX_CHANNELS], /* i : Q for powSpecMsInv_fx */
		Word16 q_powerSpecMsInv[MCT_MAX_CHANNELS][NB_DIV], / i : Q for powSpecMsInv_fx */
		Word32 inv_spectrum_fx[MCT_MAX_CHANNELS][NB_DIV], / i : inverse spectrum */
		const Word16 sp_aud_decision0[MCT_MAX_CHANNELS] /* i : speech audio decision */
		);

lib_com/ivas_spar_com_fx.c

+61 −14

Original line number	Diff line number	Diff line
		@@ -1995,15 +1995,22 @@ static void ivas_calc_post_pred_per_band_enc_fx(
		Word16 *q_postpred_cov_re )
		{
		Word16 i, j, k;
		#ifndef OPT_SBA_ENC_V1
		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_q[IVAS_SPAR_MAX_CH][IVAS_SPAR_MAX_CH];
		#else
		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 temp;
		Word32 temp_add;
		#endif
		Word32 temp_mat[IVAS_SPAR_MAX_CH][IVAS_SPAR_MAX_CH];
		Word16 e_postpred_cov_re_buf[IVAS_SPAR_MAX_CH][IVAS_SPAR_MAX_CH];
		Word16 min_val;
		Word32 tmp_re;
		Word16 tmp_q;
		Word16 tmp_e;

		#ifndef OPT_SBA_ENC_V1
		FOR( i = 0; i < num_ch; i++ )
		{
		FOR( j = 0; j < num_ch; j++ )
		@@ -2013,25 +2020,40 @@ 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
		min_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
		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
		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 +2061,56 @@ 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
		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++ )
		{
		q_postpred_cov_re_buf[i][j] = *q_postpred_cov_re;
		#ifdef OPT_SBA_ENC_V1
		temp_add = 0;
		move32();
		#endif
		e_postpred_cov_re_buf[i][j] = sub( 31, *q_postpred_cov_re );
		move16();
		FOR( k = 0; k < num_ch; k++ )
		{
		tmp_re = Mpy_32_32( mixer_mat[i][k][band_idx], temp_mat[k][j] );
		tmp_q = sub( q_mixer_mat, temp_mat_e[k][j] );
		tmp_q = add( tmp_e, temp_mat_e[k][j] );
		IF( tmp_re )
		{
		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 );
		#ifndef OPT_SBA_ENC_V1
		postpred_cov_re[i][j] = BASOP_Util_Add_Mant32Exp( postpred_cov_re[i][j], e_postpred_cov_re_buf[i][j], tmp_re, tmp_q, &e_postpred_cov_re_buf[i][j] );
		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
		}
		}
		IF( postpred_cov_re[i][j] )
		if ( temp_add )
		{
		min_val = s_min( min_val, q_postpred_cov_re_buf[i][j] );
		min_val = s_max( min_val, e_postpred_cov_re_buf[i][j] );
		}
		#ifdef OPT_SBA_ENC_V1
		postpred_cov_re[i][j] = temp_add;
		move32();
		#endif
		}
		}

		@@ -2077,7 +2121,7 @@ static void ivas_calc_post_pred_per_band_enc_fx(
		{
		IF( postpred_cov_re[i][j] )
		{
		postpred_cov_re[i][j] = L_shl( postpred_cov_re[i][j], sub( min_val, q_postpred_cov_re_buf[i][j] ) );
		postpred_cov_re[i][j] = L_shr( postpred_cov_re[i][j], sub( min_val, e_postpred_cov_re_buf[i][j] ) );
		move32();
		}
		}
		@@ -2093,13 +2137,16 @@ static void ivas_calc_post_pred_per_band_enc_fx(
		}
		}

		if ( EQ_16( min_val, MAX16B ) )
		IF( EQ_16( min_val, MIN16B ) )
		{
		min_val = Q31;
		move16();
		}
		*q_postpred_cov_re = min_val;
		ELSE
		{
		*q_postpred_cov_re = sub( Q31, min_val );
		move16();
		}

		return;
		}

lib_com/options.h

+3 −0

Original line number	Diff line number	Diff line
		@@ -74,6 +74,9 @@
		#define FIX_1379_MASA_ANGLE_ROUND

		/* Note: each compile switch (FIX_1101_...) is independent from the other ones */
		#define OPT_SBA_ENC_V1
		#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/