LTV for optimizations of hp20 (291c363a) · Commits · SA4 / Audio / IVAS BASOP

lib_com/hp50_fx.c

+30 −64

Original line number	Diff line number	Diff line
		@@ -458,14 +458,8 @@ void hp20_fx_32(
		{
		Word16 i;
		Word32 a1_fx, a2_fx, b1_fx, b2_fx;
		#ifdef OPT_STEREO_32KBPS_V1
		Word16 Qy1, Qy2, Qmin;
		Word64 y0_fx64, y1_fx64, y2_fx64;
		Word32 x0, x1, x2;
		#else /* OPT_STEREO_32KBPS_V1 */
		Word16 Qx0, Qx1, Qx2, Qy1, Qprev_y1, Qy2, Qprev_y2, Qmin;
		Word64 x0_fx64, x1_fx64, x2_fx64, y0_fx64, y1_fx64, y2_fx64, R1, R2, R3, R4, R5;
		#endif /* OPT_STEREO_32KBPS_V1 */

		IF( EQ_32( Fs, 8000 ) )
		{
		@@ -516,64 +510,16 @@ void hp20_fx_32(
		move32();
		move32();

		#ifdef OPT_STEREO_32KBPS_V1
		y1_fx64 = W_add( W_deposit32_l( mem_fx[0] ), W_deposit32_h( mem_fx[1] ) );
		y2_fx64 = W_add( W_deposit32_l( mem_fx[2] ), W_deposit32_h( mem_fx[3] ) );

		x0 = mem_fx[4];
		move32();
		x1 = mem_fx[5];
		move32();
		#else /* OPT_STEREO_32KBPS_V1 */
		Qprev_y1 = extract_l( mem_fx[4] );
		Qprev_y2 = extract_l( mem_fx[5] );
		y1_fx64 = W_deposit32_l( mem_fx[0] );
		y2_fx64 = W_deposit32_l( mem_fx[1] );
		x0_fx64 = W_deposit32_l( mem_fx[2] );
		x1_fx64 = W_deposit32_l( mem_fx[3] );
		#endif /* OPT_STEREO_32KBPS_V1 */

		FOR( i = 0; i < lg; i++ )
		{
		#ifdef OPT_STEREO_32KBPS_V1
		x2 = x1;
		move32();
		x1 = x0;
		move32();
		x0 = signal_fx[i];
		move32();

		Qy1 = W_norm( y1_fx64 );
		if ( y1_fx64 == 0 )
		{
		Qy1 = 62;
		move16();
		}

		Qy2 = W_norm( y2_fx64 );
		if ( y2_fx64 == 0 )
		{
		Qy2 = 62;
		move16();
		}

		Qmin = s_min( Qy1, Qy2 );

		Qmin = sub( Qmin, 34 );

		y0_fx64 = W_mac_32_32( W_mult_32_32( W_shl_sat_l( y1_fx64, Qmin ), a1_fx ), W_shl_sat_l( y2_fx64, Qmin ), a2_fx ); // Qmin + Q29 + Q30 + 1

		Word64 temp = W_mac_32_32( W_mac_32_32( W_mult_32_32( x2, b2_fx ), x1, b1_fx ), x0, b2_fx ); // Q30
		Word64 y0_fx = W_shr( y0_fx64, add( Qmin, Q30 ) ); // Q30
		y0_fx64 = W_add( temp, y0_fx ); // Q30
		signal_fx[i] = W_shl_sat_l( y0_fx64, -Q30 );
		move32();

		y2_fx64 = y1_fx64;
		move64();
		y1_fx64 = y0_fx64;
		move64();
		#else /* OPT_STEREO_32KBPS_V1 */
		x2_fx64 = x1_fx64;
		move64();
		x1_fx64 = x0_fx64;
		@@ -587,7 +533,11 @@ void hp20_fx_32(
		move16();
		}
		Qy1 = sub( Qy1, 34 );
		#ifdef OPT_STEREO_32KBPS_V1
		R1 = W_mult0_32_32( W_shl_sat_l( y1_fx64, Qy1 ), a1_fx );
		#else /* OPT_STEREO_32KBPS_V1 */
		R1 = W_mult0_32_32( W_extract_l( W_shl( y1_fx64, Qy1 ) ), a1_fx );
		#endif /* OPT_STEREO_32KBPS_V1 */
		Qy1 = add( Qy1, Qprev_y1 );

		Qy2 = W_norm( y2_fx64 );
		@@ -597,7 +547,11 @@ void hp20_fx_32(
		move16();
		}
		Qy2 = sub( Qy2, 34 );
		#ifdef OPT_STEREO_32KBPS_V1
		R2 = W_mult0_32_32( W_shl_sat_l( y2_fx64, Qy2 ), a2_fx );
		#else /* OPT_STEREO_32KBPS_V1 */
		R2 = W_mult0_32_32( W_extract_l( W_shl( y2_fx64, Qy2 ) ), a2_fx );
		#endif /* OPT_STEREO_32KBPS_V1 */
		Qy2 = add( Qy2, Qprev_y2 );

		Qx0 = W_norm( x0_fx64 );
		@@ -607,7 +561,9 @@ void hp20_fx_32(
		move16();
		}
		Qx0 = sub( Qx0, 34 );
		#ifndef OPT_STEREO_32KBPS_V1
		R3 = W_mult0_32_32( W_extract_l( W_shl( x0_fx64, Qx0 ) ), b2_fx );
		#endif /* OPT_STEREO_32KBPS_V1 */

		Qx1 = W_norm( x1_fx64 );
		if ( x1_fx64 == 0 )
		@@ -616,7 +572,9 @@ void hp20_fx_32(
		move16();
		}
		Qx1 = sub( Qx1, 34 );
		#ifndef OPT_STEREO_32KBPS_V1
		R4 = W_mult0_32_32( W_extract_l( W_shl( x1_fx64, Qx1 ) ), b1_fx );
		#endif /* OPT_STEREO_32KBPS_V1 */

		Qx2 = W_norm( x2_fx64 );
		if ( x2_fx64 == 0 )
		@@ -625,6 +583,7 @@ void hp20_fx_32(
		move16();
		}
		Qx2 = sub( Qx2, 34 );
		#ifndef OPT_STEREO_32KBPS_V1
		R5 = W_mult0_32_32( W_extract_l( W_shl( x2_fx64, Qx2 ) ), b2_fx );

		Qmin = s_min( Qy1, Qy2 );
		@@ -632,20 +591,37 @@ void hp20_fx_32(
		y0_fx64 = W_add( W_shr( R1, sub( Qy1, Qmin ) ), W_shr( R2, sub( Qy2, Qmin ) ) );

		Qmin = s_min( Qmin, Qx0 );
		#else /* OPT_STEREO_32KBPS_V1 */

		Word16 Qy = s_min(Qy1, Qy2);

		y0_fx64 = W_add(W_shr(R1, sub(Qy1, Qy)), W_shr(R2, sub(Qy2, Qy)));

		Qmin = s_min( Qy, Qx0 );
		#endif /* OPT_STEREO_32KBPS_V1 */
		Qmin = s_min( Qmin, Qx1 );
		Qmin = s_min( Qmin, Qx2 );

		#ifdef OPT_STEREO_32KBPS_V1
		R3 = W_mult0_32_32( W_shl_sat_l( x0_fx64, Qmin ), b2_fx );
		R4 = W_mult0_32_32( W_shl_sat_l( x1_fx64, Qmin ), b1_fx );
		R5 = W_mult0_32_32( W_shl_sat_l( x2_fx64, Qmin ), b2_fx );
		y0_fx64 = W_add( W_shr( y0_fx64, sub( Qy, Qmin ) ), W_add(R3, W_add( R4, R5 ) ) );
		#else /* OPT_STEREO_32KBPS_V1 */
		y0_fx64 = W_add( W_shr( y0_fx64, sub( s_min( Qy1, Qy2 ), Qmin ) ), W_add( W_shr( R3, sub( Qx0, Qmin ) ), W_add( W_shr( R4, sub( Qx1, Qmin ) ), W_shr( R5, sub( Qx2, Qmin ) ) ) ) );
		#endif /* OPT_STEREO_32KBPS_V1 */

		y0_fx64 = W_shr( y0_fx64, 29 );

		signal_fx[i] = W_extract_l( W_shr( y0_fx64, Qmin ) );
		move32();
		#ifndef OPT_STEREO_32KBPS_V1
		IF( signal_fx[i] < 0 )
		{
		signal_fx[i] = L_add( signal_fx[i], 1 );
		move32();
		}
		#endif /* OPT_STEREO_32KBPS_V1 */

		y2_fx64 = y1_fx64;
		y1_fx64 = y0_fx64;
		@@ -655,17 +631,8 @@ void hp20_fx_32(
		move64();
		move16();
		move16();
		#endif /* OPT_STEREO_32KBPS_V1 */
		}

		#ifdef OPT_STEREO_32KBPS_V1
		mem_fx[0] = W_extract_l( y1_fx64 );
		mem_fx[1] = W_extract_h( y1_fx64 );
		mem_fx[2] = W_extract_l( y2_fx64 );
		mem_fx[3] = W_extract_h( y2_fx64 );
		mem_fx[4] = x0;
		mem_fx[5] = x1;
		#else /* OPT_STEREO_32KBPS_V1 */
		Qy1 = W_norm( y1_fx64 );
		test();
		IF( y1_fx64 != 0 && LT_16( Qy1, 32 ) )
		@@ -688,7 +655,6 @@ void hp20_fx_32(
		mem_fx[3] = W_extract_l( x1_fx64 );
		mem_fx[4] = Qprev_y1;
		mem_fx[5] = Qprev_y2;
		#endif /* OPT_STEREO_32KBPS_V1 */

		move32();
		move32();

lib_com/mslvq_com_fx.c

+23 −8

Original line number	Diff line number	Diff line
		@@ -128,18 +128,26 @@ void init_lvq_fx(
		FOR( i = 0; i < MAX_NO_MODES; i++ )
		{
		#ifdef OPT_STEREO_32KBPS_V1
		FOR( ( j = 0, k = 0 ); j < MAX_NO_SCALES; ( j++, k++ ) )
		FOR( ( j = 0, k = 0 ); j < MAX_NO_SCALES; j++ )
		{
		if ( ( no_lead_fx[i][j] <= 0 ) )
		if ( no_lead_fx[i][j] > 0 )
		{
		j = MAX_NO_SCALES;
		k++;
		}
		if ( no_lead_fx[i][j] <= 0 )
		{
		j = MAX_NO_SCALES - 1;
		}
		}
		no_scales[i][0] = k;
		move16();

		FOR( k = 0; j < MAX_NO_SCALES << 1; ( j++, k++ ) )
		FOR( k = 0; j < MAX_NO_SCALES << 1; j++ )
		{
		if ( no_lead_fx[i][j] > 0 )
		{
		k++;
		}
		if ( no_lead_fx[i][j] <= 0 )
		{
		j = MAX_NO_SCALES << 1;
		@@ -172,19 +180,26 @@ void init_lvq_fx(
		FOR( i = 0; i < MAX_NO_MODES_p; i++ )
		{
		#ifdef OPT_STEREO_32KBPS_V1
		FOR( ( j = 0, k = 0 ); j < MAX_NO_SCALES; ( j++, k++ ) )
		FOR( ( j = 0, k = 0 ); j < MAX_NO_SCALES; j++ )
		{

		if ( no_lead_p_fx[i][j] > 0 )
		{
		k++;
		}
		if ( ( no_lead_p_fx[i][j] <= 0 ) )
		{
		j = MAX_NO_SCALES;
		j = MAX_NO_SCALES - 1;
		}
		}
		no_scales_p[i][0] = k;
		move16();

		FOR( k = 0; j < MAX_NO_SCALES << 1; ( j++, k++ ) )
		FOR( k = 0; j < MAX_NO_SCALES << 1; j++ )
		{
		if ( no_lead_p_fx[i][j] > 0 )
		{
		k++;
		}

		if ( ( no_lead_p_fx[i][j] <= 0 ) )
		{

lib_com/options.h

+1 −1

Original line number	Diff line number	Diff line
		@@ -68,7 +68,7 @@
		#endif

		/* Note: each compile switch (FIX_1101_...) is independent from the other ones */
		//#define OPT_STEREO_32KBPS_V1 /* Optimization made in stereo decoding path for 32kbps decoding */
		#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/
		#define FIX_1310_SPEEDUP_ivas_dirac_dec_output_synthesis_process_slot /FhG: WMOPS tuning, nonbe/

lib_enc/ivas_front_vad_fx.c

+35 −4

Original line number	Diff line number	Diff line
		@@ -666,14 +666,45 @@ ivas_error front_vad_spar_fx(

		old_pitch = st->pitch[1];
		move16();
		Scale_sig( wsp_fx, 368, sub( Q8, Q_inp_12k8 ) ); // Q8
		pitch_ol_ivas_fx( st->pitch, st->voicing_fx, &st->old_pitch, &st->old_corr_fx, corr_shift_fx, &st->old_thres_fx, &st->delta_pit, st->old_wsp2_fx, wsp_fx, st->mem_decim2_fx, relE_fx, st->clas, st->input_bwidth, st->Opt_SC_VBR, Q8 );
		Word16 shift, Q_wsp;
		Word16 shift1 = norm_arr( old_wsp_fx, L_WSP_MEM );
		Word16 shift2 = norm_arr( wsp_fx, L_WSP - L_WSP_MEM );
		maximum_abs_16_fx( old_wsp_fx, L_WSP_MEM, &shift );
		if ( !shift )
		{
		shift1 = Q15;
		move16();
		}
		maximum_abs_16_fx( wsp_fx, L_WSP - L_WSP_MEM, &shift );
		if ( !shift )
		{
		shift2 = Q15;
		move16();
		}

		shift = s_min( Q15, s_min( add( Q_inp_12k8, shift1 ), add( Q_inp_12k8, shift2 ) ) );
		shift = s_min( shift, add( norm_arr( st->mem_decim2_fx, 3 ), st->Q_old_wsp2 ) );
		shift = s_min( shift, add( norm_arr( st->old_wsp2_fx, ( L_WSP_MEM - L_INTERPOL ) / OPL_DECIM ), st->Q_old_wsp2 ) );

		scale_sig( old_wsp_fx, L_WSP_MEM, sub( shift, Q_inp_12k8 ) );
		scale_sig( wsp_fx, L_WSP - L_WSP_MEM, sub( shift, Q_inp_12k8 ) );

		Q_wsp = shift;
		move16();

		scale_sig( st->mem_decim2_fx, 3, sub( Q_wsp, st->Q_old_wsp2 ) ); // Q( mem_decim ) = Q( old_wsp2 )
		scale_sig( st->old_wsp2_fx, ( L_WSP_MEM - L_INTERPOL ) / OPL_DECIM, sub( Q_wsp, st->Q_old_wsp2 ) ); // Q_wsp

		st->Q_old_wsp2 = Q_wsp;
		move16();

		pitch_ol_ivas_fx( st->pitch, st->voicing_fx, &st->old_pitch, &st->old_corr_fx, corr_shift_fx, &st->old_thres_fx, &st->delta_pit, st->old_wsp2_fx, wsp_fx, st->mem_decim2_fx, relE_fx, st->clas, st->input_bwidth, st->Opt_SC_VBR, Q_wsp );

		/* Updates for adaptive lag window memory */
		st->old_pitch_la = st->pitch[2]; /* Q0 */
		move16();
		Scale_sig( wsp_fx, 368, Q9 - Q8 ); /* Q9 */
		StableHighPitchDetect_ivas_fx( &flag_spitch, st->pitch, st->voicing_fx, wsp_fx, st->localVAD, &st->voicing_sm_fx, &st->voicing0_sm_fx, &st->LF_EnergyRatio_sm_fx, &st->predecision_flag, &st->diff_sm_fx, &st->energy_sm_fx, Q12, st->lgBin_E_fx );

		StableHighPitchDetect_ivas_fx( &flag_spitch, st->pitch, st->voicing_fx, wsp_fx, st->localVAD, &st->voicing_sm_fx, &st->voicing0_sm_fx, &st->LF_EnergyRatio_sm_fx, &st->predecision_flag, &st->diff_sm_fx, &st->energy_sm_fx, Q_wsp, st->lgBin_E_fx );
		IF( st->hSpMusClas != NULL )
		{
		Word16 dummy_int;

lib_rend/ivas_reverb_fx.c

+15 −16

Original line number	Diff line number	Diff line
		@@ -230,6 +230,7 @@ static void ivas_binaural_reverb_setReverbTimes_fx(
		Word16 tmp, tmp_exp, scale, tmpVal_exp, attenuationFactorPerSample_exp, attenuationFactorPerSampleSq_exp, energyBuildup_exp, currentEnergy_exp, intendedEnergy_exp, actualizedEnergy_exp;
		Word16 sine_inp, norm, div_exp1, div1, sine, binCenterFreq_exp;
		Word16 reverb_exp = 0;
		Word32 tmp32;
		move16();

		hReverb->binRend_RandNext = (UWord16) BIN_REND_RANDOM_SEED;
		@@ -376,14 +377,13 @@ static void ivas_binaural_reverb_setReverbTimes_fx(

		UWord16 ret_binRend = binRend_rand( hReverb );

		tmp = BASOP_Util_Divide3232_Scale( ret_binRend, PCM16_TO_FLT_FAC_FX, &tmp_exp );
		L_tmp = BASOP_Util_Add_Mant32Exp( L_deposit_h( tmp ), tmp_exp, L_negate( 1073741824 ), 0, &exp );
		L_tmp = Mpy_32_32( L_tmp, 214748364 ); // exp + 0
		tmp32 = BASOP_Util_Divide3232_Scale_cadence( ret_binRend, PCM16_TO_FLT_FAC_FX, &tmp_exp );
		L_tmp = BASOP_Util_Add_Mant32Exp( tmp32, tmp_exp, L_negate( 1073741824 ), 0, &exp );
		L_tmp = Mpy_32_32( L_tmp, 214748365 ); // exp + 0
		L_tmp = BASOP_Util_Add_Mant32Exp( L_tmp, exp, currentEnergy_fx, currentEnergy_exp, &exp );
		energyBuildup_fx = BASOP_Util_Add_Mant32Exp( energyBuildup_fx, energyBuildup_exp, L_tmp, exp, &energyBuildup_exp );
		IF( energyBuildup_fx >= 0 ) /* A new filter tap is added at this condition */
		{
		IF( ( BASOP_Util_Cmp_Mant32Exp( energyBuildup_fx, energyBuildup_exp, 1, 31 ) > 0 ) )

		IF( ( BASOP_Util_Cmp_Mant32Exp( energyBuildup_fx, energyBuildup_exp, 1, 31 ) >= 0 ) )
		{
		/* Four efficient phase operations: npi/2, n=0,1,2,3 /
		hReverb->tapPhaseShiftType[bin][ch][tap] = (Word16) ( binRend_rand( hReverb ) % 4 );
		@@ -399,7 +399,6 @@ static void ivas_binaural_reverb_setReverbTimes_fx(

		actualizedEnergy_fx = BASOP_Util_Add_Mant32Exp( actualizedEnergy_fx, actualizedEnergy_exp, 1073741824, 1, &actualizedEnergy_exp );
		}
		}

		currentEnergy_fx = BASOP_Util_Add_Mant32Exp( currentEnergy_fx, currentEnergy_exp, 0, 0, &currentEnergy_exp );
		currentEnergy_fx = Mpy_32_32( currentEnergy_fx, attenuationFactorPerSampleSq_fx );