Bug fixes, high mld fixes and LTV crash fixes for MASA and OMASA cases (513b0c16) · Commits · SA4 / Audio / IVAS BASOP

lib_com/prot_fx.h

+4 −4

Original line number	Diff line number	Diff line
		@@ -7694,10 +7694,10 @@ void fd_bwe_dec_init(
		void hq_core_enc_ivas_fx(
		Encoder_State st, / i/o: encoder state structure */
		const Word16 audio_fx, / i : input audio signal Q0 */
		const Word16 input_frame, /* i : frame length */
		const Word16 hq_core_type, /* i : HQ core type */
		const Word16 Voicing_flag, /* i : Voicing flag for FER method selection */
		const Word16 vad_hover_flag /* i : VAD hangover flag */
		const Word16 input_frame, /* i : frame length Q0*/
		const Word16 hq_core_type, /* i : HQ core type Q0*/
		const Word16 Voicing_flag, /* i : Voicing flag for FER method selection Q0*/
		const Word16 vad_hover_flag /* i : VAD hangover flag Q0*/
		);

		// hq_core_dec_fx.c

lib_enc/gain_enc_fx.c

+10 −0

Original line number	Diff line number	Diff line
		@@ -622,6 +622,16 @@ void gain_enc_mless_ivas_fx(
		/* gain_inov = 1.0f / sqrt((dot_product(code, code, L_SUBFR) + 0.01) / L_SUBFR) */
		L_tmp = Dot_product12( code, code, L_SUBFR, &exp_code );
		exp_inov = sub( exp_code, 18 + 6 );

		// To avoid crash in case code value is 0
		IF( EQ_16( BASOP_Util_Cmp_Mant32Exp( L_tmp, exp_inov, 21474836, 0 ), -1 ) )
		{
		L_tmp = 21474836;
		exp_inov = 0;
		move32();
		move16();
		}

		exp_code = sub( exp_code, 30 );

		/Ei = 10 log10((dot_product(code, code, L_SUBFR) + 0.01) / L_SUBFR) */

lib_enc/hq_classifier_enc_fx.c

+177 −139

File changed.

Preview size limit exceeded, changes collapsed.

lib_enc/hq_core_enc.c

+37 −33

Original line number	Diff line number	Diff line
		@@ -54,10 +54,10 @@
		void hq_core_enc_ivas_fx(
		Encoder_State st, / i/o: encoder state structure */
		const Word16 audio_fx, / i : input audio signal Q0 */
		const Word16 input_frame_orig, /* i : frame length */
		const Word16 hq_core_type, /* i : HQ core type */
		const Word16 Voicing_flag, /* i : Voicing flag for FER method selection */
		const Word16 vad_hover_flag /* i : VAD hangover flag */
		const Word16 input_frame_orig, /* i : frame length Q0*/
		const Word16 hq_core_type, /* i : HQ core type Q0*/
		const Word16 Voicing_flag, /* i : Voicing flag for FER method selection Q0*/
		const Word16 vad_hover_flag /* i : VAD hangover flag Q0*/
		)
		{
		Word16 i, is_transient, num_bits, extra_unused;
		@@ -89,18 +89,19 @@ void hq_core_enc_ivas_fx(
		move16();
		set_zero_fx( t_audio_fx, L_FRAME48k_EXT );
		/* set input_frame length */
		input_frame = input_frame_orig;
		input_frame = input_frame_orig; /* Q0 */

		/* Sanity check, it should never happen at the encoder side (no BFI) */
		IF( st->hTcxCfg->tcx_curr_overlap_mode == FULL_OVERLAP )
		{
		st->hTcxCfg->tcx_last_overlap_mode = ALDO_WINDOW;
		move16();
		}
		ELSE
		{
		st->hTcxCfg->tcx_last_overlap_mode = st->hTcxCfg->tcx_curr_overlap_mode;
		}
		move16();
		}
		st->hTcxCfg->tcx_curr_overlap_mode = ALDO_WINDOW;
		move16();

		@@ -110,8 +111,8 @@ void hq_core_enc_ivas_fx(
		* Write signaling information
		--------------------------------------------------------------------------/

		num_bits = BASOP_Util_Divide3232_Scale( st->total_brate, FRAMES_PER_SEC, &exp );
		num_bits = shr( num_bits, sub( 15, exp ) );
		num_bits = BASOP_Util_Divide3232_Scale( st->total_brate, FRAMES_PER_SEC, &exp ); /* Q15-exp */
		num_bits = shr( num_bits, sub( 15, exp ) ); /* Q0 */
		extra_unused = 0;
		move16();

		@@ -119,7 +120,7 @@ void hq_core_enc_ivas_fx(
		* Detect signal transition
		--------------------------------------------------------------------------/

		is_transient = detect_transient_fx( audio_fx, input_frame, 0, st );
		is_transient = detect_transient_fx( audio_fx, input_frame, 0, st ); /* Q0 */
		move16();

		test();
		@@ -144,10 +145,10 @@ void hq_core_enc_ivas_fx(
		Q_audio = sub( Q16, q );
		TCX_MDCT( wtda_audio_fx16, t_audio_fx, &Q_audio, left_overlap, sub( L_spec, shr( add( left_overlap, right_overlap ), 1 ) ), right_overlap, st->element_mode );
		Q_audio = sub( Q31, Q_audio );
		Copy_Scale_sig_16_32( wtda_audio_fx16, wtda_audio_fx32, 2 * L_FRAME48k, sub( Q_audio, q ) );
		Copy_Scale_sig_16_32( wtda_audio_fx16, wtda_audio_fx32, 2 * L_FRAME48k, sub( Q_audio, q ) ); /* Q_audio */

		inner_frame = inner_frame_tbl[st->bwidth];
		L_spec = l_spec_ext_tbl[st->bwidth];
		inner_frame = inner_frame_tbl[st->bwidth]; /* Q0 */
		L_spec = l_spec_ext_tbl[st->bwidth]; /* Q0 */
		is_transient = 0;
		move16();
		move16();
		@@ -161,14 +162,14 @@ void hq_core_enc_ivas_fx(
		--------------------------------------------------------------------------/
		Q_audio = 0;
		move16();
		Scale_sig( st->old_input_signal_fx, input_frame, negate( st->q_old_inp ) );
		Scale_sig( st->input_fx, add( input_frame, NS2SA( 48000, DELAY_FIR_RESAMPL_NS ) ), negate( st->q_inp ) );
		Scale_sig( st->old_input_signal_fx, input_frame, negate( st->q_old_inp ) ); /* Q0 */
		Scale_sig( st->input_fx, add( input_frame, NS2SA( 48000, DELAY_FIR_RESAMPL_NS ) ), negate( st->q_inp ) ); /* Q0 */
		st->q_old_inp = 0;
		move16();
		st->q_inp = 0;
		move16();
		Copy( st->old_input_signal_fx, two_frames_buffer, input_frame );
		Copy( audio_fx, two_frames_buffer + input_frame, input_frame );
		Copy( st->old_input_signal_fx, two_frames_buffer, input_frame ); /* Q0 */
		Copy( audio_fx, two_frames_buffer + input_frame, input_frame ); /* Q0 */

		wtda_fx( two_frames_buffer + input_frame, &Q_audio, wtda_audio_fx32, NULL, 0,
		st->hTcxCfg->tcx_last_overlap_mode, st->hTcxCfg->tcx_curr_overlap_mode, input_frame );
		@@ -191,7 +192,7 @@ void hq_core_enc_ivas_fx(
		Word16 tmp_q = Q_audio;
		move16();
		direct_transform_fx( wtda_audio_fx32, t_audio_fx, is_transient, input_frame, &Q_audio, st->element_mode );
		scale_sig32( wtda_audio_fx32, L_FRAME48k_EXT, sub( Q_audio, tmp_q ) );
		scale_sig32( wtda_audio_fx32, L_FRAME48k_EXT, sub( Q_audio, tmp_q ) ); /* Q_audio */

		/* scale coefficients to their nominal level (8kHz) */
		IF( NE_16( input_frame, NORM_MDCT_FACTOR ) )
		@@ -199,17 +200,18 @@ void hq_core_enc_ivas_fx(
		UWord16 lsb;
		tmp = mult_r( input_frame, 410 / 2 ); /* 1/8000 in Q15 */
		Word16 ener_match_fx = hq_nominal_scaling[tmp];
		move16();
		FOR( i = 0; i < input_frame; i++ )
		{
		/t_audio_q[i] = ener_match; */
		Mpy_32_16_ss( t_audio_fx[i], ener_match_fx, &t_audio_fx[i], &lsb );
		move16(); /* Q12 */
		Mpy_32_16_ss( t_audio_fx[i], ener_match_fx, &t_audio_fx[i], &lsb ); /* Q12 */
		move16();
		}
		}

		/* limit encoded band-width according to the command-line OR BWD limitation */
		inner_frame = inner_frame_tbl[st->bwidth];
		L_spec = l_spec_tbl[st->bwidth];
		inner_frame = inner_frame_tbl[st->bwidth]; /* Q0 */
		L_spec = l_spec_tbl[st->bwidth]; /* Q0 */

		move16();
		move16();
		@@ -222,7 +224,7 @@ void hq_core_enc_ivas_fx(
		FOR( i = 1; i < NUM_TIME_SWITCHING_BLOCKS; i++ )
		{
		tmp = shr( inner_frame, 2 );
		Copy32( t_audio_fx + i_mult2( i, shr( input_frame, 2 ) ), t_audio_fx + i_mult2( i, tmp ), tmp );
		Copy32( t_audio_fx + i_mult2( i, shr( input_frame, 2 ) ), t_audio_fx + i_mult2( i, tmp ), tmp ); /* Q_audio */
		}
		}

		@@ -232,7 +234,7 @@ void hq_core_enc_ivas_fx(


		/* subtract signaling bits */
		num_bits = sub( num_bits, hBstr->nb_bits_tot );
		num_bits = sub( num_bits, hBstr->nb_bits_tot ); /* Q0 */

		/*--------------------------------------------------------------------------
		* High-band gain control in case of BWS
		@@ -241,9 +243,9 @@ void hq_core_enc_ivas_fx(
		IF( st->bwidth_sw_cnt > 0 )
		{
		Word32 L_tmp;
		tmp = BASOP_Util_Divide1616_Scale( 3, BWS_TRAN_PERIOD, &exp );
		tmp = BASOP_Util_Divide1616_Scale( 3, BWS_TRAN_PERIOD, &exp ); /* Q15-exp */
		shr( tmp, exp );
		L_tmp = L_deposit_h( tmp );
		L_tmp = L_deposit_h( tmp ); /* Q31-exp */
		IF( is_transient )
		{
		FOR( i = 0; i < NUM_TIME_SWITCHING_BLOCKS; i++ )
		@@ -315,7 +317,7 @@ void hq_core_enc_ivas_fx(
		test();
		IF( st->element_mode > EVS_MONO && ( st->last_core == ACELP_CORE \|\| EQ_16( st->last_core, AMR_WB_CORE ) ) )
		{
		overlap = st->hTcxCfg->tcx_mdct_window_length;
		overlap = st->hTcxCfg->tcx_mdct_window_length; /* Q0 */
		move16();

		nz = NS2SA( st->sr_core, N_ZERO_MDCT_NS );
		@@ -349,7 +351,7 @@ void hq_core_enc_ivas_fx(


		/Compute windowed synthesis in case of switching to ALDO windows in next frame/
		Copy( wtda_audio_fx16 + sub( L_frame, nz ), st->hTcxEnc->old_out_fx, nz + overlap );
		Copy( wtda_audio_fx16 + sub( L_frame, nz ), st->hTcxEnc->old_out_fx, nz + overlap ); /* Q0 */
		set16_fx( st->hTcxEnc->old_out_fx + add( nz, overlap ), 0, nz );

		tcx_windowing_synthesis_past_frame( st->hTcxEnc->old_out_fx + nz, st->hTcxCfg->tcx_aldo_window_1_trunc, st->hTcxCfg->tcx_mdct_window_half, st->hTcxCfg->tcx_mdct_window_minimum, overlap, st->hTcxCfg->tcx_mdct_window_half_length, st->hTcxCfg->tcx_mdct_window_min_length, FULL_OVERLAP );
		@@ -359,9 +361,11 @@ void hq_core_enc_ivas_fx(
		// This implementation is diffrent in float code need to revisit to check its correctness.
		st->hTcxEnc->old_out_fx[nz + overlap + i] = mult( wtda_audio_fx16[L_frame - 1 - i], st->hTcxCfg->tcx_aldo_window_1_trunc[i].v.re ); // Q0 + Q15 - Q15 -> Q0;
		st->hTcxEnc->old_out_fx[nz / 2 + ( nz + overlap + i )] = mult( wtda_audio_fx16[nz / 2 + ( L_frame - 1 - i )], st->hTcxCfg->tcx_aldo_window_1_trunc[nz / 2 - 1 - i].v.im ); // Q0 + Q15 - Q15 -> Q0;
		move16();
		move16();
		}

		Copy( wtda_audio_fx16 + sub( shr( overlap, 1 ), tcx_offset ), output_fx, st->L_frame );
		Copy( wtda_audio_fx16 + sub( shr( overlap, 1 ), tcx_offset ), output_fx, st->L_frame ); /* Q0 */
		}
		ELSE
		{
		@@ -371,12 +375,12 @@ void hq_core_enc_ivas_fx(
		move32();
		FOR( i = 0; i < input_frame; i++ )
		{
		t_audio_fx[i] = Mpy_32_32( t_audio_fx[i], ener_match_fx );
		move32(); /* Q12 - 1 -> Q11 */
		t_audio_fx[i] = Mpy_32_32( t_audio_fx[i], ener_match_fx ); /* Q12 - 1 -> Q11 */
		move32();
		}

		Q_audio = sub( Q_audio, 1 );
		scale_sig32( wtda_audio_fx32, 2 * L_FRAME48k, sub( Q_audio, tmp_q ) );
		scale_sig32( wtda_audio_fx32, 2 * L_FRAME48k, sub( Q_audio, tmp_q ) ); /* Q_audio */
		Inverse_Transform( t_audio_fx, &Q_audio, wtda_audio_fx32, is_transient, L_FRAME16k, inner_frame, st->element_mode );


		@@ -384,8 +388,8 @@ void hq_core_enc_ivas_fx(
		move16();
		window_ola_fx( wtda_audio_fx32, output_fx, &out_q, st->hTcxEnc->old_out_fx, &old_q, L_FRAME16k, st->hTcxCfg->tcx_last_overlap_mode, st->hTcxCfg->tcx_curr_overlap_mode, 0, 0, NULL );

		Scale_sig( output_fx, L_FRAME16k, negate( out_q ) );
		Scale_sig( st->hTcxEnc->old_out_fx, L_FRAME32k, negate( old_q ) );
		Scale_sig( output_fx, L_FRAME16k, negate( out_q ) ); /* Q0 */
		Scale_sig( st->hTcxEnc->old_out_fx, L_FRAME32k, negate( old_q ) ); /* Q0 */
		st->hTcxEnc->Q_old_out = 0;
		move16();
		}

lib_enc/hq_core_enc_fx.c

+24 −18

Original line number	Diff line number	Diff line
		@@ -19,10 +19,10 @@
		void hq_core_enc_fx(
		Encoder_State *st_fx,
		const Word16 audio, / i : input audio signal Q0 */
		const Word16 input_frame_orig, /* i : frame length */
		const Word16 hq_core_type, /* i : HQ core type */
		const Word16 Voicing_flag, /* i : Voicing flag for FER method selection */
		const Word16 vad_hover_flag /* i : VAD hangover flag */
		const Word16 input_frame_orig, /* i : frame length Q0*/
		const Word16 hq_core_type, /* i : HQ core type Q0*/
		const Word16 Voicing_flag, /* i : Voicing flag for FER method selection Q0*/
		const Word16 vad_hover_flag /* i : VAD hangover flag Q0*/
		)
		{
		Word16 i, is_transient, num_bits, extra_unused;
		@@ -31,6 +31,7 @@ void hq_core_enc_fx(
		Word16 Q_audio = 0;
		Word16 inner_frame, input_frame;
		Word16 ener_match; /* Q13/Q15 */
		move16();

		Word16 tmp;
		Word32 L_tmp;
		@@ -52,7 +53,7 @@ void hq_core_enc_fx(
		move16();

		/* set input_frame length */
		input_frame = input_frame_orig;
		input_frame = input_frame_orig; /* Q0 */
		move16();
		/* Sanity check, it should never happen at the encoder side (no BFI) */
		IF( EQ_16( st_fx->hTcxCfg->tcx_curr_overlap_mode, FULL_OVERLAP ) )
		@@ -85,13 +86,13 @@ void hq_core_enc_fx(
		* Detect signal transition
		--------------------------------------------------------------------------/

		is_transient = detect_transient_fx( audio, input_frame, 0, st_fx );
		is_transient = detect_transient_fx( audio, input_frame, 0, st_fx ); /* Q0 */

		#ifdef ADD_IVAS_HQ_CODE
		test();
		test();
		test();
		IF( GT_16( st_fx->element_mode, EVS_MONO ) && ( EQ_16( st_fx->last_core, ACELP_CORE ) \|\| EQ_16( st_fx->last_core, AMR_WB_CORE ) ) )
		IF( ( st_fx->element_mode > EVS_MONO ) && ( ( st_fx->last_core == ACELP_CORE ) \|\| EQ_16( st_fx->last_core, AMR_WB_CORE ) ) )
		{
		/*--------------------------------------------------------------------------
		* IVAS switching frame
		@@ -124,8 +125,8 @@ void hq_core_enc_fx(
		* DCT transform
		--------------------------------------------------------------------------/

		Copy( st_fx->old_input_signal_fx, two_frames_buffer, input_frame );
		Copy( audio, two_frames_buffer + input_frame, input_frame );
		Copy( st_fx->old_input_signal_fx, two_frames_buffer, input_frame ); /* st_fx->q_old_inp */
		Copy( audio, two_frames_buffer + input_frame, input_frame ); /* Q0 */

		wtda_fx( two_frames_buffer + input_frame, &Q_audio, wtda_audio, NULL, 0,
		st_fx->hTcxCfg->tcx_last_overlap_mode, st_fx->hTcxCfg->tcx_curr_overlap_mode, input_frame );
		@@ -146,7 +147,7 @@ void hq_core_enc_fx(
		}
		#ifndef ADD_IVAS_HQ_CODE
		/* subtract signalling bits */
		num_bits = sub( num_bits, hBstr->nb_bits_tot );
		num_bits = sub( num_bits, hBstr->nb_bits_tot ); /* Q0 */
		#endif
		direct_transform_fx( wtda_audio, t_audio, is_transient, input_frame, &Q_audio, st_fx->element_mode );

		@@ -161,17 +162,18 @@ void hq_core_enc_fx(
		{
		tmp = mult_r( input_frame, 410 / 2 ); /* 1/8000 in Q15 */
		ener_match = hq_nominal_scaling[tmp];
		move16();
		FOR( i = 0; i < input_frame; i++ )
		{
		/t_audio_q[i] = ener_match; */
		Mpy_32_16_ss( t_audio[i], ener_match, &t_audio[i], &lsb );
		move16(); /* Q12 */
		Mpy_32_16_ss( t_audio[i], ener_match, &t_audio[i], &lsb ); /* Q12 */
		move16();
		}
		}
		}

		/* limit encoded band-width according to the command-line OR BWD limitation */
		inner_frame = inner_frame_tbl[st_fx->bwidth];
		inner_frame = inner_frame_tbl[st_fx->bwidth]; /* Q0 */
		move16();
		#ifdef ADD_IVAS_HQ_CODE_L_SPEC
		L_spec = l_spec_tbl[st_fx->bwidth];
		@@ -185,7 +187,7 @@ void hq_core_enc_fx(
		{
		tmp = i_mult2( i, shr( input_frame, 2 ) );
		tmp = shr( inner_frame, 2 );
		Copy32( t_audio + i_mult2( i, shr( input_frame, 2 ) ), t_audio + i_mult2( i, tmp ), tmp );
		Copy32( t_audio + i_mult2( i, shr( input_frame, 2 ) ), t_audio + i_mult2( i, tmp ), tmp ); /* Q12 */
		}
		}

		@@ -221,6 +223,8 @@ void hq_core_enc_fx(
		* Classify whether to put extra bits for FER mitigation
		--------------------------------------------------------------------------/

		test();
		test();
		test();
		IF( ( EQ_16( st_fx->last_core, TCX_20_CORE ) \|\| EQ_16( st_fx->last_core, TCX_10_CORE ) \|\| EQ_16( st_fx->last_core, HQ_CORE ) ) && GT_32( st_fx->core_brate, MINIMUM_RATE_TO_ENCODE_VOICING_FLAG ) )
		{
		@@ -246,6 +250,7 @@ void hq_core_enc_fx(
		FOR( i = 0; i < inner_frame; i++ )
		{
		t_audio[i] = L_shr( t_audio[i], sub( Q_audio, 12 ) ); /* Q12 */
		move32();
		}
		hq_lr_enc_fx( st_fx, t_audio, inner_frame, &num_bits, is_transient );
		Q_audio = 12;
		@@ -257,6 +262,7 @@ void hq_core_enc_fx(
		FOR( i = 0; i < inner_frame; i++ )
		{
		t_audio[i] = L_shr( t_audio[i], sub( Q_audio, 12 ) ); /* Q12 */
		move32();
		}

		#ifdef ADD_IVAS_HQ_CODE_L_SPEC
		@@ -360,7 +366,7 @@ void HQ_core_enc_init_fx(
		move16();
		hHQ_core->prev_stab_hfe2 = 0;
		move16();
		hHQ_core->prev_ni_ratio_fx = 16384;
		hHQ_core->prev_ni_ratio_fx = 16384; /* Q15 */
		move16(); /* 0.5 */
		set16_fx( hHQ_core->prev_En_sb_fx, 0, NB_SWB_SUBBANDS );
		set16_fx( hHQ_core->last_bitalloc_max_band, 0, 2 );
		@@ -370,11 +376,11 @@ void HQ_core_enc_init_fx(
		move16();

		#ifdef MSAN_FIX
		hHQ_core->crest_lp_fx = HQ_CREST_THRESHOLD_FX;
		hHQ_core->crest_lp_fx = HQ_CREST_THRESHOLD_FX; /* Q28 */
		move32();
		hHQ_core->crest_lp_q = Q28;
		move16();
		hHQ_core->crest_mod_lp_fx = HQ_CREST_MOD_THRESHOLD_FX;
		hHQ_core->crest_mod_lp_fx = HQ_CREST_MOD_THRESHOLD_FX; /* Q29 */
		move32();
		hHQ_core->crest_mod_lp_q = Q29;
		move16();