Merge remote-tracking branch 'origin/main' into... (8a752bdd) · Commits · SA4 / Audio / IVAS BASOP

.gitlab-ci.yml

+2 −2

Original line number	Diff line number	Diff line
		@@ -23,7 +23,7 @@ variables:
		PRM_FILES: "scripts/config/self_test.prm scripts/config/self_test_ltv.prm"
		TESTCASE_TIMEOUT_STV: 900
		TESTCASE_TIMEOUT_LTV: 2400
		TESTCASE_TIMEOUT_STV_SANITIZERS: 1800
		TESTCASE_TIMEOUT_LTV_SANITIZERS: 6000
		CI_REGRESSION_THRESH_MLD: "0.1"
		CI_REGRESSION_THRESH_MAX_ABS_DIFF: "50"
		CI_REGRESSION_THRESH_SSNR: "-1"
		@@ -511,7 +511,7 @@ stages:
		- make clean
		- make -j CLANG=$CLANG_NUM
		- if [[ $CLANG_NUM == 3 ]]; then export UBSAN_OPTIONS="suppressions=scripts/ubsan.supp,report_error_type=1"; fi
		- testcase_timeout=$TESTCASE_TIMEOUT_STV_SANITIZERS
		- testcase_timeout=$TESTCASE_TIMEOUT_LTV_SANITIZERS
		- python3 -m pytest $TEST_SUITE -v --tb=no --update_ref 1 --html=report.html --self-contained-html --junit-xml=report-junit.xml --testcase_timeout $testcase_timeout --ref_encoder_path $DUT_ENCODER_PATH --ref_decoder_path $DUT_DECODER_PATH
		artifacts:
		name: "$CI_JOB_NAME--sha-$CI_COMMIT_SHORT_SHA--results"

lib_com/options.h

+7 −0

Original line number	Diff line number	Diff line
		@@ -108,4 +108,11 @@
		#define FIX_1132_STACK_CORRUPTION /* Stack corruption issue due of extending index access*/
		#define FIX_ISSUE_1092 /* Ittiam: Fix for Issue 1092: BASOP asserts in stereo fx encoder for selection test inputs*/
		#define FIX_ISSUE_1148
		#define FIX_ISSUE_1147 /Ittiam: Fix for issue 1147: Added saturation in DetectTnsFilt_fx./
		#define FIX_ISSUE_1150 /* Ittiam: Fix for Issue 1150: Assertion error observed in evs_enc_fx (with option stereo_dmx_evs) from tcx_ltp_find_gain function*/
		#define FIX_ISSUE_1151 /* Ittiam: Fix for Issue 1151: Assertion error observed in evs_enc_fx (with option stereo_dmx_evs) from sp_mus_classif_gmm_fx function*/
		#define FIX_ISSUE_1153 /* Ittiam: Fix for Issue 1153: Assertion error observed in stereo_dmx_evs_enc_fx from calc_poc_fx function*/
		#define FIX_ISSUE_1154 /* Ittiam: Fix for Issue 1154: Encoder crash for ParamMC 7.1 at 96kbps in ivas_param_mc_param_est_enc_fx() */
		#define FIX_ISSUE_1157 /* Ittiam: Fix for Issue 1157: Encoder crash for Stereo at 48/64kbps DTX on/off in kernel_switch_trafo_fx() */
		#define FIX_ISSUE_1152 /* Ittiam: Fix for issue 1152: Assertion error observed in evs_enc_fx (with option stereo_dmx_evs) from find_tilt_fx function*/
		#endif

lib_com/swb_tbe_com_fx.c

+1 −6

Original line number	Diff line number	Diff line
		@@ -7847,13 +7847,8 @@ void prep_tbe_exc_ivas_fx(
		}
		ELSE
		{
		Word16 shift = 8;
		Word16 shift = 4;
		move16();
		if ( element_mode == EVS_MONO )
		{
		shift = 4;
		move16();
		}
		IF( gain_preQ_fx != 0 )
		{
		FOR( i = 0; i < L_subfr; i++ )

lib_enc/cod4t64_fast.c

+24 −1

Original line number	Diff line number	Diff line
		@@ -770,7 +770,30 @@ void acelp_fast_fx(
		alp_pos3++;
		}

		i = emaximum_32fx( Q_dn, dn, L_subfr, &L_tmp1 );
		Word16 k;
		Word64 W_tmp, W_tmp1;
		Word64 emax;

		emax = W_mult0_32_32( dn[0], dn[0] );
		i = 0;
		move16();

		FOR( k = 1; k < L_subfr; k++ )
		{
		W_tmp = W_mult0_32_32( dn[k], dn[k] );
		W_tmp1 = W_sub( W_tmp, emax );
		if ( W_tmp1 > 0 )
		{
		i = k;
		move16();
		}
		if ( LE_64( emax, W_tmp ) )
		{
		emax = W_tmp;
		move64();
		}
		}

		track = i % nb_tracks;
		move16();

lib_enc/find_tilt.c

+100 −95

Original line number	Diff line number	Diff line
		@@ -56,22 +56,22 @@
		* Find LF/HF energy ratio
		-------------------------------------------------------------------/
		void find_tilt_ivas_fx(
		const Word32 fr_bands[], /* i : energy in frequency bands Q_new*/
		const Word32 bckr[], /* i : per band background noise energy estimate Q_new*/
		const Word32 fr_bands[], /* i : energy in frequency bands q_fr_bands*/
		const Word16 q_fr_bands, /* i : Q of fr_bands Q0*/
		const Word32 bckr[], /* i : per band background noise energy estimate q_bckr*/
		const Word16 q_bckr, /* i : Q of bckr Q0*/
		Word32 ee[2], /* o : lf/hf E ration for present frame Q6*/
		const Word16 pitch[3], /* i : open loop pitch values for 3 half-frames Q0*/
		const Word16 voicing[3], /* i : normalized correlation for 3 half-frames Q15*/
		const Word32 lf_E, / i : per bin energy for low frequencies Q_new - 2*/
		const Word32 lf_E, / i : per bin energy for low frequencies q_lf_E*/
		const Word16 q_lf_E, /* i : Q of lf_E */
		const Word16 corr_shift, /* i : normalized correlation correction Q15*/
		const Word16 bwidth, /* i : input signal bandwidth Q0*/
		const Word16 max_band, /* i : maximum critical band Q0*/
		const Word16 bwidth, /* i : input signal bandwidth */
		const Word16 max_band, /* i : maximum critical band */
		Word32 hp_E[], /* o : energy in HF Q_new*/
		const Word16 codec_mode, /* i : MODE1 or MODE2 Q0*/
		const Word16 Q_new, /* i : scaling factor */
		Word32 bckr_tilt_lt / i/o: lf/hf E ratio of background noise Q16*/
		,
		Word16 Opt_vbr_mode /* Q0 */
		)
		const Word16 codec_mode, /* i : MODE1 or MODE2 */
		Word32 bckr_tilt_lt, / i/o: lf/hf E ratio of background noise Q16 */
		Word16 Opt_vbr_mode )
		{
		Word32 lp_bckr = 0, hp_bckr = 0, lp_E, Ltmp;
		const Word32 pt_E, pt_bands, pt_bckr, hf_bands, *tmp_E;
		@@ -81,6 +81,9 @@ void find_tilt_ivas_fx(
		Word16 m_Fs, e_Fs;
		Word16 m_cnt, e_cnt;
		Word16 m_hpE, e_hpE;
		Word64 sum;
		Word16 inv_bands, q_lp_E;
		Word32 Le_min_scaled, Ltmp2;
		#ifdef BASOP_NOGLOB_DECLARE_LOCAL
		Flag Overflow = 0;
		move32();
		@@ -91,7 +94,7 @@ void find_tilt_ivas_fx(
		* Initializations
		-----------------------------------------------------------------/

		IF( NE_16( bwidth, NB ) )
		IF( ( bwidth != NB ) )
		{
		/* WB processing */
		bin = BIN4_FX;
		@@ -100,16 +103,20 @@ void find_tilt_ivas_fx(
		tmp_E = lf_E; /* Q_new - 2 */
		pt_bckr = bckr; /* Q_new */
		nb_bands = 10;
		inv_bands = 3277 /* 1/10 in Q15 */;
		move16();
		move16();
		}
		ELSE
		{
		/* NB processing */
		bin = add( shl( BIN4_FX, 1 ), BIN4_FX ); /* First useful frequency bin ~ 150 Hz */
		bin = 3 * BIN4_FX; /* First useful frequency bin ~ 150 Hz */
		pt_bands = fr_bands + 1; /* Exlcude 1st critical band */
		tmp_E = lf_E + 2; /* Start at the 3rd bin (150 Hz) */
		pt_bckr = bckr + 1; /* Exlcude 1st critical band */
		nb_bands = 9;
		inv_bands = 3641; /* 1/9 in Q15 */
		move16();
		move16(); /* Nb. of "low" frequency bands taken into account in NB processing */
		}

		@@ -122,21 +129,23 @@ void find_tilt_ivas_fx(

		/* bckr + voicing */
		/lp_bckr = mean( pt_bckr, nb_bands );/ /* estimated noise E in first critical bands, up to 1270 Hz */
		lp_bckr = Mean32( pt_bckr, nb_bands );
		sum = 0;
		move64();
		FOR( i = 0; i < nb_bands; i++ )
		{
		sum = W_mac_32_16( sum, pt_bckr[i], inv_bands ); // q_bckr+16
		}
		lp_bckr = W_shl_sat_l( sum, -16 ); // q_bckr
		/hp_bckr = 0.5f (bckr[max_band-1] + bckr[max_band]);/ / estimated noise E in last 2 critical bands */
		#ifdef BASOP_NOGLOB
		hp_bckr = L_shr( L_add_sat( bckr[max_band - 1], bckr[max_band] ), 1 ); /* Q_new - 1 */
		#else
		hp_bckr = L_shr( L_add( bckr[max_band - 1], bckr[max_band] ), 1 );
		#endif
		hp_bckr = W_extract_h( W_mac_32_32( W_mult_32_32( bckr[max_band - 1], ONE_IN_Q30 ), bckr[max_band], ONE_IN_Q30 ) ); // q_bckr
		if ( hp_bckr == 0 ) /* Avoid division by zero. */
		{
		hp_bckr = L_deposit_l( 1 );
		}
		Ltmp = BASOP_Util_Divide3232_Scale_cadence( lp_bckr, hp_bckr, &e_tmp ); /* exp(e_tmp) */
		Ltmp = Mpy_32_16_r( Ltmp, 3277 /* 0.1 in Q15 */ );
		Ltmp = BASOP_Util_Divide3232_Scale_cadence( lp_bckr, hp_bckr, &e_tmp );
		Ltmp = Mpy_32_16_r( Ltmp, 3277 /* 0.1f in Q15 */ );
		Ltmp = L_shr_sat( Ltmp, sub( 15, e_tmp ) );
		bckr_tilt_lt = L_add_sat( Mpy_32_16_r( bckr_tilt_lt, 29491 ), Ltmp ); /* Q16 */
		bckr_tilt_lt = L_add_sat( Mpy_32_16_r( bckr_tilt_lt, 29491 /* 0.9f in Q15 */ ), Ltmp ); // Q16
		move32();

		test();
		@@ -171,34 +180,31 @@ void find_tilt_ivas_fx(
		e_Fs = sub( 15, e_tmp );
		f0 = shr( m_Fs, sub( e_Fs, 4 ) ); /* Q4 */

		Le_min_scaled = L_shl( E_MIN_FXQ31, sub( q_fr_bands, Q31 ) ); // q_fr_bands=q_bckr

		FOR( i = 0; i < 2; i++ )
		{
		/hp_E[i] = 0.5f (hf_bands[max_band-1] + hf_bands[max_band]) - hp_bckr; / / averaged E in last 2 critical bands */
		Ltmp = L_add( L_shr( hf_bands[max_band - 1], 1 ), L_shr( hf_bands[max_band], 1 ) ); /* Q_new */
		hp_E[i] = L_sub( Ltmp, hp_bckr ); /* Q_new */
		move32();
		IF( Opt_vbr_mode == 0 )
		{
		IF( EQ_16( BASOP_Util_Cmp_Mant32Exp( hp_E[i], sub( 31, Q_new ), E_MIN_IVAS_FX, 31 - Q19 ), -1 ) )
		Ltmp = W_extract_h( W_mac_32_32( W_mult_32_32( hf_bands[max_band - 1], ONE_IN_Q30 ), hf_bands[max_band], ONE_IN_Q30 ) ); // q_fr_bands
		Ltmp = L_sub( Ltmp, hp_bckr ); // q_fr_bands

		Ltmp2 = L_max( Ltmp, L_shl_sat( 1, q_fr_bands ) ); // q_fr_bands, saturation is added because q_fr_bands is limited to 31
		if ( Opt_vbr_mode == 0 )
		{
		hp_E[i] = L_shl( E_MIN_IVAS_FX, sub( Q_new, Q19 ) );
		move32();
		Ltmp2 = L_max( Ltmp, Le_min_scaled ); // q_fr_bands
		}
		}
		ELSE
		{
		hp_E[i] = L_max( hp_E[i], L_shl( 1, Q_new ) );
		hp_E[i] = L_max( Ltmp2, 1 ); // to prevent division by zero
		move32();
		}

		test();
		IF( GT_16( mean_voi, TH_COR_FX ) && LT_16( pitch[2], TH_PIT_FX ) ) /* High-pitched voiced frames */
		{
		freq = bin;
		freq = bin; // Q4
		move16(); /* 1st useful frequency bin */
		m_cnt = 0;
		move16();
		lp_E = L_deposit_l( 0 );
		sum = 0;
		move64();

		f1 = add( shr( f0, 1 ), f0 ); /* Middle between 2 harmonics */
		f2 = f0;
		@@ -208,14 +214,10 @@ void find_tilt_ivas_fx(
		FOR( ; freq <= f1; freq += BIN4_FX )
		{
		/* include only bins sufficiently close to harmonics */
		tmp = sub( freq, f2 ); /* Q4 */
		IF( L_mac0( -(Word32) TH_D_FX * TH_D_FX, tmp, tmp ) < 0 )
		tmp = sub( freq, f2 );
		IF( L_mac0( -TH_D_FX * TH_D_FX, tmp, tmp ) < 0 )
		{
		#ifdef BASOP_NOGLOB
		lp_E = L_add_o( pt_E, lp_E, &Overflow ); / Q_new - 1 */
		#else
		lp_E = L_add( *pt_E, lp_E );
		#endif
		sum = W_mac_32_16( sum, *pt_E, 1 ); // q_lf_E+1
		m_cnt = add( m_cnt, 1 );
		}
		pt_E++;
		@@ -229,52 +231,65 @@ void find_tilt_ivas_fx(
		#endif
		}
		/lp_E = lp_E / (float)cnt - lp_bckr;/
		e_tmp = sub( norm_l( lp_E ), 1 );
		m_tmp = extract_h( L_shl( lp_E, e_tmp ) );

		e_tmp = sub( e_tmp, 2 ); /* lf_e divided by 4 in anal_sp */
		e_tmp = sub( W_norm( sum ), 1 );
		m_tmp = extract_h( W_extract_h( W_shl( sum, e_tmp ) ) ); // q_lf_E+1+e_tmp-32-16
		e_tmp = sub( add( q_lf_E, e_tmp ), 47 );

		e_cnt = norm_s( m_cnt );
		m_cnt = shl( m_cnt, e_cnt );
		m_cnt = shl( m_cnt, e_cnt ); // e_cnt

		m_tmp = div_s( m_tmp, m_cnt );
		e_tmp = sub( e_tmp, e_cnt );

		#ifdef BASOP_NOGLOB
		lp_E = L_sub_o( L_shr_o( m_tmp, sub( e_tmp, 1 ), &Overflow ), lp_bckr, &Overflow );
		#else
		lp_E = L_sub( L_shr( m_tmp, sub( e_tmp, 1 ) ), lp_bckr );
		#endif
		m_tmp = div_s( m_tmp, m_cnt ); // Q15+e_tmp-e_cnt
		e_tmp = add( Q15, sub( e_tmp, e_cnt ) );
		sum = W_shl( m_tmp, sub( add( q_bckr, 1 ), e_tmp ) ); // q_bckr+1
		sum = W_msu_32_16( sum, lp_bckr, 1 ); // q_bckr+1
		q_lp_E = W_norm( sum );
		lp_E = W_extract_h( W_shl( sum, q_lp_E ) ); // q_bckr+1+q_lp_E-32
		q_lp_E = sub( add( q_lp_E, q_bckr ), Q31 );

		pt_E = tmp_E + VOIC_BINS; /* Update for next half-frame Q_new - 1 */
		}
		ELSE /* Other than high-pitched voiced frames */
		{
		/lp_E = mean( pt_bands, nb_bands ) - lp_bckr;/ /* averaged E in first critical bands, up to 1270 Hz */
		lp_E = L_sub( Mean32( pt_bands, nb_bands ), lp_bckr );
		}
		IF( Opt_vbr_mode == 0 )
		sum = 0;
		move64();
		FOR( Word16 j = 0; j < nb_bands; j++ )
		{
		IF( EQ_16( BASOP_Util_Cmp_Mant32Exp( lp_E, sub( 31, Q_new ), E_MIN_IVAS_FX, 31 - Q19 ), -1 ) )
		{
		lp_E = L_shl( E_MIN_IVAS_FX, sub( Q_new, Q19 ) );
		sum = W_mac_32_16( sum, pt_bands[j], inv_bands ); // q_fr_bands+16
		}
		sum = W_mac_32_16( sum, lp_bckr, -ONE_IN_Q15 ); // q_fr_bands+16
		lp_E = W_round48_L( sum ); // q_fr_bands=q_bckr
		q_lp_E = q_bckr; // q_fr_bands=q_bckr
		move16();
		}
		ELSE
		test();
		IF( Opt_vbr_mode == 0 && GT_32( L_shl_sat( Le_min_scaled, sub( q_lp_E, q_bckr ) ), lp_E ) )
		{
		lp_E = E_MIN_FXQ31;
		q_lp_E = Q31;
		move32();
		move16();
		}
		if ( Opt_vbr_mode != 0 )
		{
		lp_E = L_max( lp_E, 0 );
		lp_E = L_max( lp_E, 0 ); // q_lp_E
		}
		/ee[i] = lp_E / hp_E[i];/ /* LF/HF ratio */

		if ( lp_E == 0 )
		{
		ee[i] = 0;
		move32();
		}
		test();
		IF( lp_E != 0 && hp_E[i] != 0 )
		IF( lp_E != 0 )
		{
		e_tmp = sub( norm_l( lp_E ), 1 );
		m_tmp = extract_h( L_shl( lp_E, e_tmp ) );
		m_tmp = extract_h( L_shl( lp_E, e_tmp ) ); // e_tmp+q_lp_E-16
		e_hpE = norm_l( hp_E[i] );
		m_hpE = extract_h( L_shl( hp_E[i], e_hpE ) );
		m_tmp = div_s( m_tmp, m_hpE ); /* e_tmp + e_hpE */
		e_tmp = sub( e_tmp, e_hpE );

		m_hpE = extract_h( L_shl( hp_E[i], e_hpE ) ); // e_hpE+q_bckr-16
		m_tmp = div_s( m_tmp, m_hpE ); // Q15+(e_tmp+q_lp_E)-(e_hpE+q_bckr)
		e_tmp = sub( add( e_tmp, q_lp_E ), add( e_hpE, q_bckr ) );
		#ifdef BASOP_NOGLOB
		ee[i] = L_shr_o( m_tmp, add( e_tmp, 15 - 6 ), &Overflow ); /* ee in Q6 */
		#else /* BASOP_NOGLOB */
		@@ -282,18 +297,8 @@ void find_tilt_ivas_fx(
		#endif
		move32();
		}
		ELSE IF( lp_E == 0 )
		{
		ee[i] = L_deposit_l( 0 );
		move32();
		}
		ELSE
		{
		ee[i] = MAX_32;
		move32();
		}

		IF( EQ_16( bwidth, NB ) ) /* For NB input, compensate for the missing bands */
		IF( bwidth == NB ) /* For NB input, compensate for the missing bands */
		{
		#ifdef BASOP_NOGLOB
		Ltmp = L_shl_o( ee[i], 3, &Overflow ); /* Q6 */