Encoder LTV crash fixes and float code clean up (051c900b) · Commits · SA4 / Audio / IVAS BASOP

lib_com/cnst.h

+1 −0

Original line number	Diff line number	Diff line
		@@ -751,6 +751,7 @@ enum
		#define E_MIN_Q11_FX 7 /* minimum allowable energy in Q11*/
		#define E_MIN 0.0035f /* minimum allowable energy */
		#define E_MIN_Q11_FX 7 /* minimum allowable energy in Q11*/
		#define E_MIN_Q16_FX 229 /* minimum allowable energy in Q16*/
		#define STEP_DELTA 0.0625f /* quantization step for tilt compensation of gaussian cb. excitation */
		#define GAMMA_EV 0.92f /* weighting factor for core synthesis error weighting */
		#define FORMANT_SHARPENING_NOISE_THRESHOLD 21.0f /* lp_noise level above which formant sharpening is deactivated */

lib_com/core_com_config.c

+4 −18

Original line number	Diff line number	Diff line
		@@ -1257,6 +1257,7 @@ Word16 getTcxLpcShapedAri(
		}


		#ifndef IVAS_FLOAT_FIXED
		/-----------------------------------------------------------------------
		* init_tcx_window_cfg()
		*
		@@ -1362,6 +1363,7 @@ void init_tcx_window_cfg(

		return;
		}
		#endif

		/-----------------------------------------------------------------------
		* init_tcx()
		@@ -1369,6 +1371,7 @@ void init_tcx_window_cfg(
		* Initialization of TCX
		-----------------------------------------------------------------------/

		#ifndef IVAS_FLOAT_FIXED
		void init_tcx_cfg(
		TCX_CONFIG_HANDLE hTcxCfg,
		const int32_t total_brate,
		@@ -1405,22 +1408,14 @@ void init_tcx_cfg(
		init_tcx_window_cfg( hTcxCfg, sr_core, input_Fs, L_frame, L_frameTCX, encoderLookahead_enc, encoderLookahead_FB, mdctWindowLength, mdctWindowLengthFB, element_mode );

		/* SQ deadzone & memory quantization*/
		#ifndef IVAS_FLOAT_FIXED
		hTcxCfg->sq_rounding_flt = 0.375f; /deadzone of 1.25->rounding=1-1.25/2 (No deadzone=0.5)/
		#else
		hTcxCfg->sq_rounding = 12288; /deadzone of 1.25->rounding=1-1.25/2 (No deadzone=0.5)/
		#endif

		/* TCX rate loop */
		hTcxCfg->tcxRateLoopOpt = ( tcxonly ) ? 2 : 0;
		hTcxCfg->tcxRateLoopOpt = ( element_mode == IVAS_CPE_MDCT ) ? 3 : hTcxCfg->tcxRateLoopOpt;

		/* TCX bandwidth */
		#ifndef IVAS_FLOAT_FIXED
		hTcxCfg->bandwidth_flt = getTcxBandwidth_flt( bwidth );
		#else
		hTcxCfg->bandwidth = getTcxBandwidth( bwidth );
		#endif

		/* set number of coded lines */
		hTcxCfg->tcx_coded_lines = getNumTcxCodedLines( bwidth );
		@@ -1464,11 +1459,7 @@ void init_tcx_cfg(
		}

		/* Scale TCX for non-active frames to adjust loudness with ACELP*/
		#ifndef IVAS_FLOAT_FIXED
		hTcxCfg->na_scale_flt = 1.f;
		#else
		hTcxCfg->na_scale = 32767;
		#endif

		if ( na_scale_bwidth < SWB && !tcxonly )
		{
		@@ -1482,11 +1473,7 @@ void init_tcx_cfg(
		{
		i--;
		}
		#ifndef IVAS_FLOAT_FIXED
		hTcxCfg->na_scale_flt = scaleTcxTable[i].scale_flt;
		#else
		hTcxCfg->na_scale = scaleTcxTable[i].scale;
		#endif
		break;
		}
		}
		@@ -1503,8 +1490,7 @@ void init_tcx_cfg(

		return;
		}

		#ifdef IVAS_FLOAT_FIXED
		#else
		void init_tcx_cfg_ivas_fx(
		TCX_CONFIG_HANDLE hTcxCfg,
		const Word32 total_brate, /Q0/

lib_com/ivas_prot.h

+23 −12

Original line number	Diff line number	Diff line
		@@ -236,7 +236,7 @@ ivas_error pre_proc_front_ivas_fx(
		Word16 old_inp_12k8_fx[], /* o : (Q-1) buffer of old input signal */
		float old_inp_12k8[], /* o : buffer of old input signal */
		float old_inp_16k[], /* o : buffer of old input signal @16kHz */
		float ener, / o : residual energy from Levinson-Durbin */
		Word32 ener_fx, / o : residual energy from Levinson-Durbin Q_r */
		Word16 relE_fx, / o : frame relative energy Q8 */
		Word16 A_fx[NB_SUBFR16k * ( M + 1 )], /* o : A(z) unquantized for the 4 subframes */
		Word16 Aw_fx[NB_SUBFR16k * ( M + 1 )], /* o : weighted A(z) unquantized for subframes */
		@@ -254,9 +254,10 @@ ivas_error pre_proc_front_ivas_fx(
		Word16 loc_harm, / o : harmonicity flag */
		Word16 cor_map_sum_fx, / o : speech/music clasif. parameter Q8 */
		Word16 vad_flag_dtx, / o : HE-SAD flag with additional DTX HO */
		float enerBuffer[CLDFB_NO_CHANNELS_MAX], /* o : energy buffer */
		float fft_buff[2 * L_FFT], /* o : FFT buffer */
		Word32 enerBuffer_fx[CLDFB_NO_CHANNELS_MAX], /* o : energy buffer */
		Word16 enerBuffer_fx_exp, / o : energy buffer */
		Word16 fft_buff_fx[2 * L_FFT], /* o : FFT buffer */
		Word16 fft_buff_fx_q, / o : FFT buffer */
		const Word16 tdm_A_PCh_fx[M + 1], /* i : unq. LP coeff. of primary channel Q12 */
		const Word16 tdm_lsp_new_PCh_fx[M], /* i : unq. LSPs of primary channel Q15 */
		const Word16 currFlatness_fx, /* i : flatness parameter Q7 */
		@@ -285,13 +286,13 @@ ivas_error pre_proc_ivas(
		Word16 old_inp_12k8_fx[], /* i/o: buffer of old input signal */
		float old_inp_16k[], /* i/o: buffer of old input signal @ 16kHz */
		float *inp, / o : ptr. to inp. signal in the current frame*/
		float ener, / o : residual energy from Levinson-Durbin */
		Word32 ener_fx, / o : residual energy from Levinson-Durbin Q6 */
		Word16 A[NB_SUBFR16k * ( M + 1 )], /* i/o: A(z) unquantized for the 4 subframes */
		Word16 Aw[NB_SUBFR16k * ( M + 1 )], /* i/o: weighted A(z) unquantized for subframes */
		float epsP[M + 1], /* i/o: LP prediction errors */
		Word16 lsp_new[M], /* i/o: LSPs at the end of the frame */
		Word16 lsp_mid[M], /* i/o: LSPs in the middle of the frame */
		float new_inp_resamp16k, / o : new input signal @16kHz, non pre-emphasised, used by the WB TBE/BWE */
		Word16 new_inp_resamp16k_fx, / o : new input signal @16kHz, non pre-emphasised, used by the WB TBE/BWE */
		int16_t Voicing_flag, / o : voicing flag for HQ FEC */
		float old_wsp[], /* i : weighted input signal buffer */
		const int16_t loc_harm, /* i : harmonicity flag */
		@@ -525,7 +526,7 @@ ivas_error ivas_core_enc(
		const int16_t n_CoreChannels, /* i : number of core channels to be coded */
		float old_inp_12k8[][L_INP_12k8], /* i : buffer of old input signal */
		float old_inp_16k[][L_INP], /* i : buffer of old input signal */
		float ener[], /* i : residual energy from Levinson-Durbin */
		Word32 ener_fx[], /* i : residual energy from Levinson-Durbin Q6*/
		Word16 A_fx[][NB_SUBFR16k * ( M + 1 )], /* i : A(z) unquantized for the 4 subframes */
		Word16 Aw_fx[][NB_SUBFR16k * ( M + 1 )], /* i : weighted A(z) unquantized for subframes*/
		float epsP[][M + 1], /* i : LP prediction errors */
		@@ -533,14 +534,16 @@ ivas_error ivas_core_enc(
		Word16 lsp_mid[][M], /* i : LSPs in the middle of the frame Q15 */
		const int16_t vad_hover_flag[], /* i : VAD hanglover flag */
		int16_t attack_flag[], /* i : attack flag (GSC or TC) */
		float realBuffer[][CLDFB_NO_COL_MAX][CLDFB_NO_CHANNELS_MAX], /* i/o: real buffer */
		float imagBuffer[][CLDFB_NO_COL_MAX][CLDFB_NO_CHANNELS_MAX], /* i/o: imag buffer */
		Word32 realBuffer_fx[][CLDFB_NO_COL_MAX][CLDFB_NO_CHANNELS_MAX], /* i/o: real buffer */
		Word32 imagBuffer_fx[][CLDFB_NO_COL_MAX][CLDFB_NO_CHANNELS_MAX], /* i/o: imag buffer */
		Word16 q_re_im_buf,
		float old_wsp[][L_WSP], /* i : weighted input signal buffer */
		const int16_t loc_harm[], /* i : harmonicity flag */
		const Word16 cor_map_sum[], /* i : speech/music clasif. parameter Q8 */
		const int16_t vad_flag_dtx[], /* i : HE-SAD flag with additional DTX HO */
		float enerBuffer[][CLDFB_NO_CHANNELS_MAX], /* i : energy buffer */
		float fft_buff[][2 * L_FFT], /* i : FFT buffer */
		Word32 enerBuffer_fx[][CLDFB_NO_CHANNELS_MAX], /* o : energy buffer */
		Word16 enerBuffer_fx_exp[], /* o : energy buffer */
		Word16 fft_buff_fx[][2 * L_FFT], /* i : FFT buffer */
		const int16_t tdm_SM_flag, /* i : channel combination scheme flag */
		const int16_t ivas_format, /* i : IVAS format */
		const int16_t flag_16k_smc /* i : flag to indicate if the OL SMC is run at 16 kHz */
		@@ -1390,6 +1393,7 @@ void ivas_ism_dtx_limit_noise_energy_for_near_silence(
		* DFT Stereo prototypes
		----------------------------------------------------------------------------------/

		#ifndef IVAS_FLOAT_FIXED
		ivas_error stereo_dft_enc_create(
		STEREO_DFT_ENC_DATA_HANDLE hStereoDft, / o : encoder DFT stereo handle */
		const int32_t input_Fs, /* i : input sampling rate */
		@@ -1404,16 +1408,15 @@ void stereo_enc_itd_init(
		ITD_DATA_HANDLE hItd /* i/o: encoder ITD handle */
		);

		#ifndef IVAS_FLOAT_FIXED
		void stereo_dft_enc_update(
		STEREO_DFT_ENC_DATA_HANDLE hStereoDft, /* i/o: encoder DFT stereo handle */
		const int16_t max_bwidth /* i : maximum encoded bandwidth */
		);
		#endif

		void stereo_dft_enc_destroy(
		STEREO_DFT_ENC_DATA_HANDLE hStereoDft / i/o: encoder DFT stereo handle */
		);
		#endif

		void stereo_dft_enc_analyze(
		Encoder_State *sts, / i/o: encoder state structure */
		@@ -1816,6 +1819,12 @@ int16_t read_BS_adapt_GR_sg(
		int16_t GR_ord, / o : GR order to be used */
		const int16_t map0 / i : mapping array for side gains */
		);

		void stereo_dft_hybrid_ITD_flag(
		STEREO_DFT_CONFIG_DATA_HANDLE hConfig, /* o : DFT stereo configuration */
		const int32_t input_Fs, /* i : CPE element sampling rate */
		const int16_t hybrid_itd_max /* i : flag for hybrid ITD for very large ITDs */
		);
		#endif

		void stereo_dft_enc_compute_itd(
		@@ -4643,11 +4652,13 @@ void ivas_spar_dec_upmixer(
		);

		/* MD module */
		#ifndef IVAS_FLOAT_FIXED
		ivas_error ivas_spar_md_enc_open(
		ivas_spar_md_enc_state_t *hMdEnc, / i/o: SPAR MD encoder handle */
		const ENCODER_CONFIG_HANDLE hEncoderConfig, /* i : configuration structure */
		const int16_t sba_order /* i : Ambisonic (SBA) order */
		);
		#endif

		void ivas_spar_md_enc_close(
		ivas_spar_md_enc_state_t *hMdEnc / i/o: SPAR MD encoder handle */

lib_com/ivas_prot_fx.h

+15 −2

Original line number	Diff line number	Diff line
		@@ -1395,6 +1395,15 @@ void stereo_tdm_downmix_ivas_fx(
		const Word16 tdm_ratio_idx_SM /* i : TDM ratio index for SM mode */
		);

		ivas_error initMdctItdHandling_fx(
		STEREO_MDCT_ENC_DATA hStereoMdct, / i/o: mdct stereo parameters structure */
		const Word32 input_Fs /* i : input sampling rate */
		);

		void stereo_mdct_enc_destroy_fx(
		STEREO_MDCT_ENC_DATA_HANDLE hStereoMdct / i/o: encoder MDCT stereo handle */
		);

		void stereo_mdct_init_bands_fx(
		const Word16 L_frame, /* i : frame length */
		const Word16 tmp_tcx_mode, /* i : tcx mode (TCX10, TCX 20), -1 if transition frame */
		@@ -4244,6 +4253,10 @@ void stereo_dft_enc_update_fx(
		const Word16 max_bwidth /* i : maximum encoded bandwidth */
		);

		void stereo_dft_enc_destroy_fx(
		STEREO_DFT_ENC_DATA_HANDLE hStereoDft / i/o: encoder DFT stereo handle */
		);

		void stereo_dft_quantize_res_gains_fx(
		const Word32 *g, // Q31
		const Word32 *r, // Q31
		@@ -4337,7 +4350,7 @@ Word16 write_bitstream_GR(
		const Word16 GR_ord /* i : GR order */
		);

		void stereo_dft_hybrid_ITD_flag(
		void stereo_dft_hybrid_ITD_flag_fx(
		STEREO_DFT_CONFIG_DATA_HANDLE hConfig, /* o : DFT stereo configuration */
		const Word32 input_Fs, /* i : CPE element sampling rate */
		const Word16 hybrid_itd_max /* i : flag for hybrid ITD for very large ITDs */
		@@ -5838,7 +5851,7 @@ ivas_error ivas_compute_core_buffers_fx(
		const Word16 input_frame, /* i : frame length */
		const Word16 last_element_mode, /* i : last element mode */
		const Word32 sr_core, /* i : core-coder sampling rate */
		Word32 ener, / o : residual energy from Levinson-Durbin */
		Word32 ener_fx, / o : residual energy from Levinson-Durbin Q_r */
		Word16 A_fx[NB_SUBFR16k * ( M + 1 )], /* i/o: A(z) unquantized for the 4 subframes */
		Word16 Aw_fx[NB_SUBFR16k * ( M + 1 )], /* i/o: weighted A(z) unquantized for subframes */
		Word32 epsP[M + 1], /* i/o: LP prediction errors */

lib_com/lerp.c

+1 −1

Original line number	Diff line number	Diff line
		@@ -496,7 +496,7 @@ void L_lerp_fx( Word32 f /q/, Word32 f_out /q/, Word16 bufferNewSize /Q0
		IF( LT_16( getScaleFactor32( f, bufferOldSize ), guard_bits ) )
		{
		q = sub( q, guard_bits );
		FOR( Word16 ind = 0; ind < bufferNewSize; ind++ )
		FOR( Word16 ind = 0; ind < bufferOldSize; ind++ )
		{
		f[ind] = L_shr( f[ind], guard_bits ); /Q(guard_bits)/
		move32();