port CR (2024077c) · Commits · SA4 / Audio / IVAS BASOP

lib_com/cnst.h

+3 −0

Original line number	Diff line number	Diff line
		@@ -668,6 +668,9 @@ enum
		#define ACELP_TCX_TRANS_NS 1250000 /* Duration of the ACELP->TCX overlap - 1.25 ms */
		#define L_FRAME_MAX L_FRAME48k /* Max 20ms frame size @48kHz */
		#define L_FRAME_PLUS 1200 /* Max frame size (long TCX frame) */
		#ifdef FIX_1320_STACK_CPE_DECODER
		#define L_FRAME_PLUS_INTERNAL 800 /* Max frame size (long TCX frame) at maximum internal sampling rate */
		#endif
		#define L_MDCT_OVLP_MAX NS2SA( 48000, ACELP_LOOK_NS ) /* = Max mdct overlap */
		#define N_TCX10_MAX 480 /* Max size of TCX10 MDCT spectrum */
		#define BITS_TEC 1 /* number of bits for TEC */

lib_com/options.h

+1 −0

Original line number	Diff line number	Diff line
		@@ -168,6 +168,7 @@
		#define NONBE_1293_SR_HRTF /* VA: issue 1293: add support of external HRTFs in split rendering */
		#define NONBE_1220_OMASA_JBM_BRATE_SW_FLUSH /* VA: issue 1220: fix bug in renderer flush in OMASA 1ISM JBM bitrate switching */
		#define FIX_1139_REV_COLORATION_SHORT_T60 /* Nokia,FhG: Fix issue 1139, prevent sound coloration artefacts at very low reverberation times */
		#define FIX_1320_STACK_CPE_DECODER /* VA: issue 1320: Optimize the stack memory consumption in the CPE decoder */

		/* #################### End BASOP porting switches ############################ */

lib_dec/dec_LPD_fx.c

+8 −0

Original line number	Diff line number	Diff line
		@@ -36,7 +36,11 @@ void decoder_LPD_fx(
		)
		{
		Word16 *param_lpc;
		#ifdef FIX_1320_STACK_CPE_DECODER
		Word16 synth_buf[OLD_SYNTH_INTERNAL_DEC + L_FRAME_PLUS_INTERNAL + M];
		#else
		Word16 synth_buf[OLD_SYNTH_SIZE_DEC + L_FRAME_PLUS + M];
		#endif
		Word16 *synth;
		Word16 synth_bufFB[OLD_SYNTH_SIZE_DEC + L_FRAME_PLUS + M];
		Word16 *synthFB;
		@@ -114,7 +118,11 @@ void decoder_LPD_fx(
		synth = synth_buf + hTcxDec->old_synth_len;
		/st->old_synth: Q_0/
		Copy( hTcxDec->old_synth, synth_buf, hTcxDec->old_synth_len );
		#ifdef FIX_1320_STACK_CPE_DECODER
		set16_fx( synth, 0, L_FRAME_PLUS_INTERNAL + M );
		#else
		set16_fx( synth, 0, L_FRAME_PLUS + M );
		#endif

		synthFB = synth_bufFB + hTcxDec->old_synth_lenFB;
		Copy( hTcxDec->old_synthFB_fx, synth_bufFB, hTcxDec->old_synth_lenFB );

lib_dec/ivas_core_dec_fx.c

+33 −0

Original line number	Diff line number	Diff line
		@@ -76,7 +76,12 @@ ivas_error ivas_core_dec_fx(
		set16_fx( tmp_buffer_fx, 0, L_FRAME48k );

		Word16 tmps, incr;
		#ifdef FIX_1320_STACK_CPE_DECODER
		Word32 *bwe_exc_extended_fx[CPE_CHANNELS] = { NULL, NULL };
		Word16 flag_bwe_bws;
		#else
		Word32 bwe_exc_extended_fx[CPE_CHANNELS][L_FRAME32k + NL_BUFF_OFFSET];
		#endif

		Word16 voice_factors_fx[CPE_CHANNELS][NB_SUBFR16k]; // Q15
		Word16 core_switching_flag[CPE_CHANNELS];
		@@ -298,6 +303,9 @@ ivas_error ivas_core_dec_fx(
		set16_fx( voice_factors_fx[n], 0, NB_SUBFR16k );
		set32_fx( hb_synth_32_fx[n], 0, L_FRAME48k );
		set16_fx( hb_synth_16_fx[n], 0, L_FRAME48k );
		#ifdef FIX_1320_STACK_CPE_DECODER
		bwe_exc_extended_fx[n] = hb_synth_32_fx[n]; /* note: reuse the buffer */
		#endif

		/------------------------------------------------------------------
		* Decision matrix (selection of technologies)
		@@ -1044,9 +1052,18 @@ ivas_error ivas_core_dec_fx(
		* SWB(FB) BWE decoding
		---------------------------------------------------------------------/

		#ifdef FIX_1320_STACK_CPE_DECODER
		test();
		test();
		test();
		flag_bwe_bws = ( GE_32( output_Fs, 32000 ) && st->core == ACELP_CORE && st->bwidth > NB && st->bws_cnt > 0 && st->bfi == 0 );
		move16();
		#endif

		Q_white_exc = 0;
		move16();

		#ifndef FIX_1320_STACK_CPE_DECODER
		test();
		test();
		test();
		@@ -1054,6 +1071,7 @@ ivas_error ivas_core_dec_fx(
		test();
		test();
		test();
		#endif
		test();
		test();
		test();
		@@ -1085,7 +1103,11 @@ ivas_error ivas_core_dec_fx(
		fb_tbe_dec_ivas_fx( st, tmp_buffer_fx /fb_exc/, Q_white_exc, hb_synth_32_fx[n], 0, tmp_buffer_fx /fb_synth_ref/, Q_white_exc, output_frame );
		}
		}
		#ifdef FIX_1320_STACK_CPE_DECODER
		ELSE IF( st->extl == SWB_BWE \|\| st->extl == FB_BWE \|\| flag_bwe_bws )
		#else
		ELSE IF( EQ_16( st->extl, SWB_BWE ) \|\| EQ_16( st->extl, FB_BWE ) \|\| ( GE_32( output_Fs, 32000 ) && st->core == ACELP_CORE && st->bwidth > NB && st->bws_cnt > 0 && !st->ppp_mode_dec && !( EQ_16( st->nelp_mode_dec, 1 ) && EQ_16( st->bfi, 1 ) ) ) )
		#endif
		{
		/* SWB BWE decoder */
		Q_syn_hb = swb_bwe_dec_fx32( st, output_32_fx[n], synth_32_fx[n], hb_synth_32_fx[n], use_cldfb_for_dft, output_frame );
		@@ -1095,6 +1117,17 @@ ivas_error ivas_core_dec_fx(
		Copy_Scale_sig_32_16( hBWE_FD->L_old_wtda_swb_fx32, hBWE_FD->L_old_wtda_swb_fx, output_frame, sub( hBWE_FD->old_wtda_swb_fx_exp, Q11 ) ); // old_wtda_swb_fx_exp
		}

		#ifdef FIX_1320_STACK_CPE_DECODER
		test();
		test();
		test();
		IF( ( st->core == ACELP_CORE && ( st->extl == -1 \|\| st->extl == SWB_CNG ) ) && flag_bwe_bws == 0 )
		{
		set32_fx( hb_synth_32_fx[n], 0, L_FRAME48k );
		}

		#endif

		/---------------------------------------------------------------------
		* FEC - recovery after lost HQ core (smoothing of the BWE component)
		---------------------------------------------------------------------/

lib_dec/ivas_cpe_dec_fx.c

+217 −2

Original line number	Diff line number	Diff line
		@@ -51,6 +51,10 @@ static void read_stereo_mode_and_bwidth_fx( CPE_DEC_HANDLE hCPE, const Decoder_S

		static void stereo_mode_combined_format_dec_fx( const Decoder_Struct *st_ivas, CPE_DEC_HANDLE hCPE );

		#ifdef FIX_1320_STACK_CPE_DECODER
		static ivas_error stereo_dft_dec_main( CPE_DEC_HANDLE hCPE, const Word32 ivas_total_brate, const Word16 n_channels, Word32 p_res_buf_fx, Word32 output[], Word32 outputHB[][L_FRAME48k], const Word16 output_frame, const Word32 output_Fs );
		#endif


		/--------------------------------------------------------------------------
		* ivas_cpe_dec_fx()
		@@ -71,8 +75,12 @@ ivas_error ivas_cpe_dec_fx(
		Word16 last_bwidth;
		Word16 tdm_ratio_idx;
		Word32 outputHB_fx[CPE_CHANNELS][L_FRAME48k]; /* buffer for output HB synthesis, both channels / / Q11 */
		#ifdef FIX_1320_STACK_CPE_DECODER
		Word32 res_buf_fx = NULL; / Q8 */
		#else
		Word16 q_res_buf;
		Word32 res_buf_fx[STEREO_DFT_N_8k]; /* Q(q_res_buf) */
		#endif
		CPE_DEC_HANDLE hCPE;
		Decoder_State **sts;
		Word32 ivas_total_brate;
		@@ -83,8 +91,10 @@ ivas_error ivas_cpe_dec_fx(

		error = IVAS_ERR_OK;
		move32();
		#ifndef FIX_1320_STACK_CPE_DECODER
		q_res_buf = Q8;
		move16();
		#endif

		push_wmops( "ivas_cpe_dec" );

		@@ -347,6 +357,10 @@ ivas_error ivas_cpe_dec_fx(
		}
		ELSE
		{
		#ifdef FIX_1320_STACK_CPE_DECODER
		res_buf_fx = outputHB_fx[0]; /* note: temporarily reused buffer */

		#endif
		test();
		IF( EQ_32( st_ivas->ivas_format, MASA_FORMAT ) \|\| EQ_32( st_ivas->ivas_format, MASA_ISM_FORMAT ) )
		{
		@@ -616,10 +630,17 @@ ivas_error ivas_cpe_dec_fx(
		/----------------------------------------------------------------
		* Stereo decoder & upmixing
		----------------------------------------------------------------/

		test();
		test();
		IF( EQ_16( hCPE->element_mode, IVAS_CPE_DFT ) && !( EQ_16( hCPE->nchan_out, 1 ) && EQ_16( hCPE->hStereoDft->hConfig->res_cod_mode, STEREO_DFT_RES_COD_OFF ) ) )
		{
		#ifdef FIX_1320_STACK_CPE_DECODER
		IF( NE_32( ( error = stereo_dft_dec_main( hCPE, ivas_total_brate, n_channels, res_buf_fx, output, outputHB_fx, output_frame, st_ivas->hDecoderConfig->output_Fs ) ), IVAS_ERR_OK ) )
		{
		return error;
		}
		#else
		Word32 DFT_fx[CPE_CHANNELS][STEREO_DFT_BUF_MAX]; // q_dft
		set32_fx( DFT_fx[0], 0, STEREO_DFT_BUF_MAX );
		set32_fx( DFT_fx[1], 0, STEREO_DFT_BUF_MAX );
		@@ -784,6 +805,7 @@ ivas_error ivas_cpe_dec_fx(
		hCPE->q_output_mem_fx[n] = Q11;
		move16();
		}
		#endif
		}
		ELSE IF( EQ_16( hCPE->element_mode, IVAS_CPE_TD ) )
		{
		@@ -823,7 +845,6 @@ ivas_error ivas_cpe_dec_fx(
		* Update parameters for stereo CNA
		----------------------------------------------------------------/

		// stereo_cna_update_params( hCPE, output_flt, output_frame, tdm_ratio_idx );
		stereo_cna_update_params_fx( hCPE, output, output_frame, tdm_ratio_idx );

		/----------------------------------------------------------------
		@@ -857,7 +878,6 @@ ivas_error ivas_cpe_dec_fx(

		stereo_tca_dec_fx( hCPE, output, output_frame );


		/----------------------------------------------------------------
		* Common Stereo updates
		----------------------------------------------------------------/
		@@ -889,6 +909,201 @@ ivas_error ivas_cpe_dec_fx(
		return error;
		}

		#ifdef FIX_1320_STACK_CPE_DECODER

		/*-------------------------------------------------------------------------
		* stereo_dft_dec_main()
		*
		* DFT decoder main function
		-------------------------------------------------------------------------/

		static ivas_error stereo_dft_dec_main(
		CPE_DEC_HANDLE hCPE, /* i/o: CPE decoder structure */
		const Word32 ivas_total_brate, /* i : IVAS total bitrate */
		const Word16 n_channels, /* i : number of channels to be decoded */
		Word32 p_res_buf_fx, / i : DFT stereo residual S signal */
		Word32 output[], / o : output synthesis signal */
		Word32 outputHB_fx[][L_FRAME48k], /* o : output HB synthesis signal */
		const Word16 output_frame, /* i : output frame length per channel */
		const Word32 output_Fs /* i : output sampling rate */
		)
		{
		Word32 DFT_fx[CPE_CHANNELS][STEREO_DFT_BUF_MAX];
		Word16 n;
		Word16 q_res_buf, q_dft;
		Decoder_State *st0;
		ivas_error error;
		Word16 shift;
		Word32 tmp1, tmp2;
		Word16 shift1, shift2;

		st0 = hCPE->hCoreCoder[0];

		q_res_buf = Q8;
		move16();

		/* copy from temporary buffer */
		test();
		IF( hCPE->hStereoDft->res_cod_band_max > 0 && !st0->bfi )
		{
		Copy32( p_res_buf_fx, DFT_fx[1], STEREO_DFT_N_8k );
		}

		/* core decoder */
		IF( NE_32( ( error = ivas_core_dec_fx( NULL, NULL, hCPE, NULL, n_channels, output, outputHB_fx, DFT_fx, 0 ) ), IVAS_ERR_OK ) )
		{
		return error;
		}

		/* Scaling of DFT's */
		maximum_abs_32_fx( DFT_fx[0], STEREO_DFT_BUF_MAX, &tmp1 );
		maximum_abs_32_fx( DFT_fx[1], STEREO_DFT_BUF_MAX, &tmp2 );

		IF( tmp1 == 0 )
		{
		shift1 = Q31;
		move16();
		}
		ELSE
		{
		shift1 = norm_l( tmp1 );
		}
		IF( tmp2 == 0 )
		{
		shift2 = Q31;
		move16();
		}
		ELSE
		{
		shift2 = norm_l( tmp2 );
		}
		shift = s_min( shift1, shift2 );

		IF( NE_16( shift, 31 ) )
		{
		shift = sub( add( hCPE->hStereoDft->q_dft, shift ), Q16 ); /* Q16 for guard bits */

		IF( GT_16( shift, hCPE->hStereoDft->q_dft ) )
		{
		Scale_sig32( DFT_fx[0], STEREO_DFT_BUF_MAX, sub( shift, hCPE->hStereoDft->q_dft ) ); // shift
		Scale_sig32( DFT_fx[1], STEREO_DFT_BUF_MAX, sub( shift, hCPE->hStereoDft->q_dft ) ); // shift
		hCPE->hStereoDft->q_dft = shift;
		move16();
		}
		}
		ELSE
		{
		hCPE->hStereoDft->q_dft = Q8;
		move16();
		}

		/* DFT Stereo residual decoding */
		test();
		IF( hCPE->hStereoDft->res_cod_band_max > 0 && !st0->bfi )
		{
		Word16 q;
		Word16 q_out_DFT[2];

		q = Q11;
		move16();

		Copy_Scale_sig_32_16( hCPE->hCoreCoder[0]->old_pitch_buf_fx, hCPE->hCoreCoder[0]->old_pitch_buf_16_fx, ( ( 2 * NB_SUBFR16k ) + 2 ), -( Q10 ) ); // Q16->Q6

		stereo_dft_dec_res_fx( hCPE, p_res_buf_fx, q_res_buf, output[1] );

		Copy_Scale_sig_16_32_DEPREC( hCPE->hCoreCoder[0]->old_pitch_buf_16_fx, hCPE->hCoreCoder[0]->old_pitch_buf_fx, ( ( 2 * NB_SUBFR16k ) + 2 ), 10 ); // Q6->Q16

		Scale_sig32( output[1], L_FRAME8k, ( Q11 - Q15 ) ); // Q15 -> Q11

		q_out_DFT[0] = q_out_DFT[1] = hCPE->hStereoDft->q_dft;
		move16();
		move16();

		stereo_dft_dec_analyze_fx( hCPE, output[1], DFT_fx, 1, L_FRAME8k, output_frame, DFT_STEREO_DEC_ANA_LB, 0, 0, &q, q_out_DFT );

		Scale_sig32( DFT_fx[1], STEREO_DFT_BUF_MAX, sub( hCPE->hStereoDft->q_dft, q_out_DFT[1] ) ); // q_dft
		}

		/* DFT stereo CNG */
		q_dft = hCPE->hStereoDft->q_dft;
		move16();
		stereo_dtf_cng_fx( hCPE, ivas_total_brate, DFT_fx, output_frame, q_dft );

		/* decoding */
		IF( EQ_16( hCPE->nchan_out, 1 ) )
		{
		IF( EQ_16( hCPE->hStereoDft->first_frame, 1 ) )
		{
		hCPE->hStereoDft->q_smoothed_nrg = hCPE->hStereoDft->q_dft;
		move16();
		FOR( Word16 ii = 0; ii < (Word16) ( sizeof( hCPE->hStereoDft->q_DFT_past_DMX_fx ) / sizeof( hCPE->hStereoDft->q_DFT_past_DMX_fx[0] ) ); ii++ )
		{
		hCPE->hStereoDft->q_DFT_past_DMX_fx[ii] = hCPE->hStereoDft->q_dft;
		move16();
		}
		hCPE->hStereoDft->first_frame = 0;
		move16();
		}

		scale_sig32( hCPE->hStereoDft->res_cod_mem_fx, STEREO_DFT_OVL_8k, sub( hCPE->hStereoDft->q_dft, hCPE->hStereoDft->q_res_cod_mem_fx ) ); // q_dft
		hCPE->hStereoDft->q_res_cod_mem_fx = hCPE->hStereoDft->q_dft;
		move16();
		stereo_dft_unify_dmx_fx( hCPE->hStereoDft, st0, DFT_fx, hCPE->input_mem_fx[1], hCPE->hStereoCng->prev_sid_nodata );
		scale_sig32( hCPE->hStereoDft->res_cod_mem_fx, STEREO_DFT_OVL_8k, sub( Q15, hCPE->hStereoDft->q_res_cod_mem_fx ) ); // Q15
		hCPE->hStereoDft->q_res_cod_mem_fx = Q15;
		move16();
		}
		ELSE
		{
		IF( EQ_16( hCPE->hStereoDft->first_frame, 1 ) )
		{
		hCPE->hStereoDft->q_smoothed_nrg = hCPE->hStereoDft->q_dft;
		move16();
		FOR( Word16 ii = 0; ii < (Word16) ( sizeof( hCPE->hStereoDft->q_DFT_past_DMX_fx ) / sizeof( hCPE->hStereoDft->q_DFT_past_DMX_fx[0] ) ); ii++ )
		{
		hCPE->hStereoDft->q_DFT_past_DMX_fx[ii] = hCPE->hStereoDft->q_dft;
		move16();
		}
		hCPE->hStereoDft->first_frame = 0;
		move16();
		}

		scale_sig32( hCPE->hStereoDft->res_cod_mem_fx, STEREO_DFT_OVL_8k, sub( hCPE->hStereoDft->q_dft, hCPE->hStereoDft->q_res_cod_mem_fx ) ); // q_dft
		hCPE->hStereoDft->q_res_cod_mem_fx = hCPE->hStereoDft->q_dft;
		move16();

		stereo_dft_dec_fx( hCPE->hStereoDft, st0, DFT_fx, hCPE->input_mem_fx[1], hCPE->hStereoCng, 0, 0, 0, 0, 0, 0, MAX_PARAM_SPATIAL_SUBFRAMES );
		scale_sig32( hCPE->hStereoDft->res_cod_mem_fx, STEREO_DFT_OVL_8k, sub( Q15, hCPE->hStereoDft->q_res_cod_mem_fx ) ); // Q15
		hCPE->hStereoDft->q_res_cod_mem_fx = Q15;
		move16();
		}

		FOR( n = 0; n < hCPE->nchan_out; n++ )
		{
		Scale_sig32( output[n], L_FRAME48k, sub( hCPE->hStereoDft->q_dft, Q11 ) ); // q_dft
		scale_sig32( hCPE->output_mem_fx[n], NS2SA_FX2( output_Fs, STEREO_DFT32MS_OVL_NS ), sub( hCPE->hStereoDft->q_dft, Q11 ) ); // q_dft
		hCPE->q_output_mem_fx[n] = hCPE->hStereoDft->q_dft;
		move16();
		}

		/* synthesis iFFT */
		FOR( n = 0; n < hCPE->nchan_out; n++ )
		{
		stereo_dft_dec_synthesize_fx( hCPE, DFT_fx, n, output[n], output_frame );
		}

		FOR( n = 0; n < hCPE->nchan_out; n++ )
		{
		Scale_sig32( output[n], L_FRAME48k, sub( Q11, hCPE->hStereoDft->q_dft ) ); // Q11
		scale_sig32( hCPE->output_mem_fx[n], NS2SA_FX2( output_Fs, STEREO_DFT32MS_OVL_NS ), sub( Q11, hCPE->hStereoDft->q_dft ) ); // Q11
		hCPE->q_output_mem_fx[n] = Q11;
		move16();
		}

		return IVAS_ERR_OK;
		}

		#endif

		/*-------------------------------------------------------------------------
		* create_cpe_dec_fx()