Merge branch 'basop-2280-back-and-forth-scaling-in-core-coder-nonBE' into 'main' (b3c0bbf8) · Commits · SA4 / Audio / IVAS BASOP

lib_com/ivas_prot_fx.h

+12 −0

Original line number	Diff line number	Diff line
		@@ -1587,8 +1587,12 @@ void ivas_hq_core_dec_fx(
		const Word16 output_frame, /* i : output frame length */
		const Word16 hq_core_type, /* i : HQ core type */
		const Word16 core_switching_flag, /* i : ACELP->HQ switching frame flag */
		#ifdef FIX_2280_REDUCTION_UNNECESSARY_SCALING
		Word32 output_32_fx / o : synthesis @internal_Fs, Q11 */
		#else
		Word16 output[],
		Word16 *Q_output
		#endif
		);

		#ifndef HARMONIZE_TBE
		@@ -1890,7 +1894,11 @@ void GenShapedWBExcitation_ivas_fx(
		/* o : Q_syn_hb*/
		Word16 ivas_wb_bwe_dec_fx(
		Decoder_State st_fx, / i/o: decoder state structure */
		#ifdef FIX_2280_REDUCTION_UNNECESSARY_SCALING
		Word32 output_fx32[], /* o : synthesis @internal Fs Q11*/
		#else
		const Word16 output[], /* i : suntehsis @ internal Fs Q_input */
		#endif
		Word16 synth_fx, / i/o: ACELP core synthesis/final synthesis Q0/Qpost */
		Word16 hb_synth_fx, / o : SHB synthesis/final synthesis Q_syn_hb */
		const Word16 use_cldfb_for_dft, /* i : flag to use of CLDFB for DFT Stereo */
		@@ -2088,7 +2096,11 @@ void ivas_buffer_deinterleaved_to_interleaved_fx(
		void stereo_tcx_core_dec_fx(
		Decoder_State st, / i/o: decoder state structure */
		const FRAME_MODE frameMode, /* i : Decoder frame mode */
		#ifdef FIX_2280_REDUCTION_UNNECESSARY_SCALING
		Word32 signal_out_32_fx, / o : synthesis @internal_Fs, Q11 */
		#else
		Word16 signal_out, / o : synthesis @internal_Fs, Q0 */
		#endif
		Word16 signal_outFB, / o : synthesis @output_Fs, Q0 */
		Word16 pitch_buf[], /* o : floating pitch for each subframe, Q6 */
		const Word16 sba_dirac_stereo_flag, /* i : signal stereo output for SBA DirAC */

lib_com/options.h

+3 −0

Original line number	Diff line number	Diff line
		@@ -87,6 +87,9 @@
		#define FIX_2397_COPY_AQ_MDCT_CORE_BFI /* FhG: prevent copying of uninit memory in MDCT stereo core if bfi is set */
		#define HARMONIZE_TBE /* VA: harmonize core-coder TBE function duplications */
		#define FIX_2405_HARM_SMC_INIT /* VA, basop 2405, harmonisation of SMC init */
		#define FIX_2280_REDUCTION_UNNECESSARY_SCALING /* VA: reduction of unnecessary scaling */
		#define FIX_2280_REDUCTION_UNNECESSARY_SCALING_NONBE /* VA: reduction of unnecessary scaling, non-BE part */

		/* #################### End BE switches ################################## */

		/* #################### Start NON-BE switches ############################ */

lib_com/prot_fx.h

+21 −8

Original line number	Diff line number	Diff line
		@@ -7574,8 +7574,13 @@ ivas_error core_switching_pre_dec_fx(
		ivas_error core_switching_post_dec_fx(
		Decoder_State st_fx, / i/o: decoder state structure */
		Word16 synth, / i/o: output synthesis Qsynth*/
		#ifdef FIX_2280_REDUCTION_UNNECESSARY_SCALING
		Word32 output_fx, / i/o: LB synth/upsampled LB synth Q11*/
		Word32 output_mem_fx32[], /* i : OLA memory from last TCX/HQ frame Qx*/
		#else
		Word32 output_fx, / i/o: LB synth/upsampled LB synth Q4*/
		Word16 output_mem_fx[], /* i : OLA memory from last TCX/HQ frame Qx*/
		#endif
		const Word16 use_cldfb_for_dft, /* i : flag to use of CLDFB for DFT Stereo Q0*/
		const Word16 output_frame, /* i : frame length Q0*/
		const Word16 core_switching_flag, /* i : ACELP->HQ switching flag Q0*/
		@@ -10110,9 +10115,17 @@ Word16 swb_bwe_dec_fx32(

		ivas_error acelp_core_dec_fx(
		Decoder_State st, / i/o: decoder state structure */
		#ifdef FIX_2280_REDUCTION_UNNECESSARY_SCALING
		Word32 output_fx32[], /* o : synthesis @internal Fs Q11*/
		#else
		Word16 output_fx[], /* o : synthesis @internal Fs */
		#endif
		Word16 synth_fx16[], /* o : synthesis */
		#ifdef FIX_2280_REDUCTION_UNNECESSARY_SCALING
		Word32 save_hb_synth_fx32[], /* o : HB synthesis Q0*/
		#else
		Word16 save_hb_synth_fx16[], /* o : HB synthesis */
		#endif
		Word32 bwe_exc_extended_fx[], /* i/o: bandwidth extended excitation */
		Word16 voice_factors_fx, / o : voicing factors */
		Word16 old_syn_12k8_16k_fx[], /* o : intermediate ACELP synthesis at 12.8kHz or 16kHz to be used by SWB BWE */

lib_dec/acelp_core_dec_fx.c

+29 −5

Original line number	Diff line number	Diff line
		@@ -53,9 +53,17 @@

		ivas_error acelp_core_dec_fx(
		Decoder_State st, / i/o: decoder state structure */
		#ifdef FIX_2280_REDUCTION_UNNECESSARY_SCALING
		Word32 output_fx32[], /* o : synthesis @internal Fs Q11*/
		#else
		Word16 output_fx[], /* o : synthesis @internal Fs Q_syn*/
		#endif
		Word16 synth_fx16[], /* o : synthesis Q_syn2*/
		#ifdef FIX_2280_REDUCTION_UNNECESSARY_SCALING
		Word32 save_hb_synth_fx32[], /* o : HB synthesis Q0*/
		#else
		Word16 save_hb_synth_fx16[], /* o : HB synthesis Q0*/
		#endif
		Word32 bwe_exc_extended_fx[], /* i/o: bandwidth extended excitation 2Q_exc/
		Word16 voice_factors_fx, / o : voicing factors Q15*/
		Word16 old_syn_12k8_16k_fx[], /* o : intermediate ACELP synthesis at 12.8kHz or 16kHz to be used by SWB BWE Q_syn2-1*/
		@@ -205,7 +213,11 @@ ivas_error acelp_core_dec_fx(
		Copy( lsp_new_fx, st->lsp_old_fx, M ); /Q15/
		}

		#ifdef FIX_2280_REDUCTION_UNNECESSARY_SCALING
		set32_fx( output_fx32, 0, st->L_frame ); /* output and synth are not used in DFT domain CNG generation and the decoder output is unaffected if they are left uninitalized */
		#else
		set16_fx( output_fx, 0, output_frame ); /* output and synth are not used in DFT domain CNG generation and the decoder output is unaffected if they are left uninitalized */
		#endif
		set16_fx( synth_fx16, 0, output_frame ); /* They are however read in a few places which causes errors in the valgrind tests. Simplest solution from a code perspective was to set them to zero. */

		/* CN generation done in DFT domain */
		@@ -2008,7 +2020,11 @@ ivas_error acelp_core_dec_fx(
		}
		}

		#ifdef FIX_2280_REDUCTION_UNNECESSARY_SCALING
		IF( save_hb_synth_fx32 != NULL )
		#else
		IF( save_hb_synth_fx16 != NULL )
		#endif
		{
		/* save and then zero-out lowband */
		max_real = 0;
		@@ -2049,9 +2065,13 @@ ivas_error acelp_core_dec_fx(
		}
		}

		#ifdef FIX_2280_REDUCTION_UNNECESSARY_SCALING
		cldfbSynthesis_ivas_fx( realBuffer_fx, imagBuffer_fx, save_hb_synth_fx32, -1, sub( Q11, Q_real ), -10, st->cldfbSynHB );
		#else
		cldfbSynthesis_ivas_fx( realBuffer_fx, imagBuffer_fx, synth_fx /tmp buffer for save_hb_synth_fx[] /, -1, sub( Q11, Q_real ), -10, st->cldfbSynHB );

		Copy_Scale_sig_32_16( synth_fx, save_hb_synth_fx16, output_frame, 0 ); // Q0
		#endif

		/* restore lowband */
		FOR( j = 0; j < CLDFB_NO_CHANNELS_MAX; j++ )
		@@ -2194,10 +2214,14 @@ ivas_error acelp_core_dec_fx(

		/* Copy output signal */
		Scale_sig( syn_tmp_fx, add( st->L_frame, L_SUBFR ), negate( st->Q_syn ) ); // Q0
		#ifdef FIX_2280_REDUCTION_UNNECESSARY_SCALING
		Copy_Scale_sig_16_32_no_sat( psyn_fx, output_fx32, st->L_frame, sub( Q11, st->Q_syn2 ) ); // Q_syn2->Q11
		#else
		IF( st->element_mode > EVS_MONO )
		{
		Copy( psyn_fx, output_fx, st->L_frame ); /Q_syn/
		}
		#endif

		Copy_Scale_sig_32_16( synth_fx, synth_fx16, output_frame, 0 ); // Q_syn2
		st->Q_syn2 = 0;

lib_dec/core_switching_dec_fx.c

+68 −8

Original line number	Diff line number	Diff line
		@@ -172,7 +172,6 @@ void bw_switching_pre_proc_fx(
		}
		}


		test();
		test();
		test();
		@@ -207,11 +206,25 @@ void bw_switching_pre_proc_fx(
		}


		/---------------------------------------------------------------------
		* core_switching_post_dec()
		*
		* Postprocessing for ACELP/HQ core switching
		---------------------------------------------------------------------/

		ivas_error core_switching_post_dec_fx(
		Decoder_State st_fx, / i/o: decoder state structure */
		Word16 synth, / i/o: output synthesis Qsynth*/
		#ifdef FIX_2280_REDUCTION_UNNECESSARY_SCALING
		Word32 output_fx, / i/o: LB synth/upsampled LB synth Q11*/
		#else
		Word32 output_fx, / i/o: LB synth/upsampled LB synth Q4*/
		#endif
		#ifdef FIX_2280_REDUCTION_UNNECESSARY_SCALING
		Word32 output_mem_fx32[], /* i : OLA memory from last TCX/HQ frame Qx*/
		#else
		Word16 output_mem_fx[], /* i : OLA memory from last TCX/HQ frame Qx*/
		#endif
		const Word16 use_cldfb_for_dft, /* i : flag to use of CLDFB for DFT Stereo Q0*/
		const Word16 output_frame, /* i : frame length Q0*/
		const Word16 core_switching_flag, /* i : ACELP->HQ switching flag Q0*/
		@@ -433,7 +446,11 @@ ivas_error core_switching_post_dec_fx(
		test();
		IF( ( ( NE_32( st_fx->last_core_brate, SID_2k40 ) && NE_32( st_fx->last_core_brate, FRAME_NO_DATA ) ) \|\| ( NE_16( st_fx->element_mode, IVAS_CPE_DFT ) && NE_16( st_fx->element_mode, IVAS_CPE_TD ) ) \|\| EQ_16( nchan_out, 1 ) ) && !( EQ_16( st_fx->element_mode, IVAS_CPE_MDCT ) && EQ_16( st_fx->idchan, 1 ) && ( EQ_16( nchan_out, 1 ) \|\| EQ_16( last_element_mode, IVAS_CPE_DFT ) ) ) )
		{
		#ifdef FIX_2280_REDUCTION_UNNECESSARY_SCALING
		Scale_sig32( output_fx, L_FRAME48k, Q10 - Q11 ); /* Q10 */
		#else
		Scale_sig32( output_fx, L_FRAME48k, Q10 - Q4 ); /* Q10 */
		#endif
		core_switch_lb_upsamp_fx( st_fx, output_fx );
		}

		@@ -491,7 +508,11 @@ ivas_error core_switching_post_dec_fx(
		FOR( i = 0; i < delta; i++ )
		{
		st_fx->cldfbAna->cldfb_state_fx[offset - delta + i] =
		#ifdef FIX_2280_REDUCTION_UNNECESSARY_SCALING
		Mpy_32_16_1( L_shl( output_fx[st_fx->L_frame - delta + i], Q10 - Q11 ), alpha ); /* Q10 */
		#else
		Mpy_32_16_1( L_shl( output_fx[st_fx->L_frame - delta + i], Q10 - Q4 ), alpha ); /* Q10 */
		#endif
		move32();
		IF( LT_16( alpha, sub( 32767, tmp ) ) )
		{
		@@ -540,7 +561,11 @@ ivas_error core_switching_post_dec_fx(
		FOR( i = 0; i < delta; i++ )
		{
		st_fx->cldfbAna->cldfb_state_fx[offset - delta + i] =
		#ifdef FIX_2280_REDUCTION_UNNECESSARY_SCALING
		Mpy_32_16_1( L_shl( output_fx[st_fx->L_frame - delta + i], Q10 - Q11 ), alpha ); /* Q10 */
		#else
		Mpy_32_16_1( L_shl( output_fx[st_fx->L_frame - delta + i], Q10 - Q4 ), alpha ); /* Q10 */
		#endif
		move32();
		IF( LT_16( alpha, sub( 32767, tmp ) ) )
		{
		@@ -602,10 +627,18 @@ ivas_error core_switching_post_dec_fx(
		hHQ_core->Q_old_wtda = Qtmp;
		move16();
		}
		#ifdef FIX_2280_REDUCTION_UNNECESSARY_SCALING
		Word16 output_mem_fx[NS2SA( 48000, STEREO_DFT32MS_OVL_NS )];
		IF( output_mem_fx32 != NULL )
		{
		Copy_Scale_sig_32_16( output_mem_fx32, output_mem_fx, NS2SA_FX2( st_fx->output_Fs, STEREO_DFT32MS_OVL_NS ), sub( Qtmp, Q11 ) ); // Qtmp
		}
		#else
		IF( output_mem_fx != NULL )
		{
		Scale_sig( output_mem_fx, NS2SA_FX2( st_fx->output_Fs, STEREO_DFT32MS_OVL_NS ), Qtmp ); /* Qtmp */
		}
		#endif
		*Qsynth = Qtmp;
		move16();

		@@ -795,7 +828,6 @@ ivas_error core_switching_post_dec_fx(
		test();
		test();
		test();

		IF( ( ( EQ_16( st_fx->extl, SWB_TBE ) \|\| EQ_16( st_fx->extl, FB_TBE ) \|\| EQ_16( st_fx->extl, SWB_CNG ) ) &&
		( NE_16( st_fx->L_frame, st_fx->last_L_frame ) \|\| ( NE_16( st_fx->last_extl, SWB_TBE ) && NE_16( st_fx->last_extl, FB_TBE ) && NE_16( st_fx->last_core, TCX_20_CORE ) && NE_16( st_fx->last_core, TCX_10_CORE ) ) \|\| EQ_16( st_fx->last_core, HQ_CORE ) ) ) \|\|
		( LT_16( st_fx->bwidth, st_fx->last_bwidth ) && NE_16( st_fx->last_extl, SWB_TBE ) ) \|\| st_fx->old_ppp_mode \|\| ( ( EQ_16( st_fx->prev_coder_type, AUDIO ) \|\| EQ_16( st_fx->prev_coder_type, INACTIVE ) ) && st_fx->bws_cnt > 0 ) \|\| ( st_fx->bws_cnt == 0 && EQ_16( st_fx->prev_bws_cnt, N_WS2N_FRAMES ) ) )
		@@ -871,6 +903,7 @@ ivas_error core_switching_post_dec_fx(
		return error;
		}


		/---------------------------------------------------------------------
		* core_switching_hq_prepare_dec()
		*
		@@ -954,9 +987,17 @@ void core_switching_hq_prepare_dec_fx(

		/* reset state of old_out if switching */
		set16_fx( st_fx->hHQ_core->old_out_fx, 0, output_frame );

		return;
		}


		/---------------------------------------------------------------------
		* core_switch_lb_upsamp()
		*
		* Resample HQ/TCX-LB to the output sampling rate (8/16/32/48 kHz)
		---------------------------------------------------------------------/

		static void core_switch_lb_upsamp_fx(
		Decoder_State st, / i/o: Decoder state */
		Word32 output / i/o: LB synth/upsampled LB synth Q10*/
		@@ -1032,17 +1073,36 @@ static void core_switch_lb_upsamp_fx(
		no_col = s_min( no_col, temp );
		}

		#ifdef FIX_2280_REDUCTION_UNNECESSARY_SCALING
		Scale_sig32( output + imult1616( no_col, st->cldfbSyn->no_channels ), sub( st->hTcxDec->L_frameTCX, imult1616( no_col, st->cldfbSyn->no_channels ) ), Q11 - Q10 ); /* Q11 */
		#else
		Scale_sig32( output + imult1616( no_col, st->cldfbSyn->no_channels ), sub( L_FRAME48k, imult1616( no_col, st->cldfbSyn->no_channels ) ), 4 - 10 ); /* Q4 */
		#endif

		/* save synthesis - needed in case of core switching */
		IF( st->hTcxDec != NULL )
		{
		Copy_Scale_sig_32_16( output, st->previoussynth_fx, st->hTcxDec->L_frameTCX, -4 ); /* Q0 */
		#ifdef FIX_2280_REDUCTION_UNNECESSARY_SCALING
		Copy_Scale_sig_32_16( output, st->previoussynth_fx, st->hTcxDec->L_frameTCX, -Q4 ); /* Q0 */
		#else
		Copy_Scale_sig_32_16( output, st->previoussynth_fx, st->hTcxDec->L_frameTCX, -Q4 ); /* Q0 */
		#endif
		}

		#ifdef FIX_2280_REDUCTION_UNNECESSARY_SCALING
		Scale_sig32( output, imult1616( no_col, st->cldfbSyn->no_channels ), Q11 - Q4 ); /* Q11 */
		#endif

		return;
		}


		/---------------------------------------------------------------------
		* smoothTransitionDtxToTcx()
		*
		* apply smoothing to the transition part for inactive to active transitions in DTX
		---------------------------------------------------------------------/

		#define TRANSITION_SMOOTHING_LEN_16k 15
		#define TRANSITION_SMOOTHING_LEN_32k 31
		#define TRANSITION_SMOOTHING_LEN_48k 47