Merge branch 'hq_core_enc_fdcng_enc_fxd' into 'main' (92a13f5d) · Commits · SA4 / Audio / IVAS BASOP

lib_com/prot_fx.h

+9 −0

Original line number	Diff line number	Diff line
		@@ -7454,6 +7454,15 @@ void fd_bwe_dec_init(
		FD_BWE_DEC_HANDLE hBWE_FD /* i/o: FD BWE data handle */
		);

		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 */
		);

		// hq_core_dec_fx.c
		void hq_core_dec_fx(
		Decoder_State st_fx, / i/o: decoder state structure fx */

lib_enc/core_enc_init.c

+25 −2

Original line number	Diff line number	Diff line
		@@ -44,6 +44,10 @@
		#include "prot_fx.h"
		#include "prot_fx_enc.h"

		#ifdef IVAS_FLOAT_FIXED
		#include "prot_fx.h"
		#endif

		/-----------------------------------------------------------------------
		* Local function prototypes
		-----------------------------------------------------------------------/
		@@ -440,6 +444,11 @@ void init_coder_ace_plus(
		set_zero( hTcxEnc->Txnq_flt, L_FRAME32k / 2 + 64 );
		hTcxEnc->acelp_zir_flt = hTcxEnc->Txnq_flt + L_FRAME / 2;
		hTcxEnc->tcx_target_bits_fac_flt = 1.0f;

		#ifdef IVAS_FLOAT_FIXED
		set16_fx( hTcxEnc->Txnq, 0, L_FRAME32k / 2 + 64 );
		hTcxEnc->acelp_zir = hTcxEnc->Txnq + L_FRAME / 2;
		#endif
		}
		}

		@@ -1061,8 +1070,13 @@ static void init_sig_buffers_ivas_fx( Encoder_State *st, const Word16 L_frame_ol

		if ( st->hTcxEnc != NULL )
		{
		hTcxEnc->new_speech_TCX = st->input_buff_fx + L_FRAME48k + NS2SA( 48000, DELAY_FIR_RESAMPL_NS ) - NS2SA( st->input_Fs, DELAY_FIR_RESAMPL_NS );
		hTcxEnc->speech_TCX = hTcxEnc->new_speech_TCX - st->encoderLookahead_FB;
		st->hTcxEnc->new_speech_TCX_flt = st->input_buff + st->input_Fs / FRAMES_PER_SEC; /* note: in EVS st->new_speech_TCX == st->input - 0.9375ms; in IVAS st->new_speech_TCX == st->input */
		st->hTcxEnc->speech_TCX_flt = st->hTcxEnc->new_speech_TCX_flt - st->encoderLookahead_FB;

		#ifdef IVAS_FLOAT_FIXED
		st->hTcxEnc->new_speech_TCX = st->input_buff_fx + Mpy_32_32( st->input_Fs, ONE_BY_FRAMES_PER_SEC_Q31 );
		st->hTcxEnc->speech_TCX = st->hTcxEnc->new_speech_TCX - st->encoderLookahead_FB;
		#endif
		}
		st->speech_enc = st->buf_speech_enc + st->encoderPastSamples_enc;
		st->speech_enc_pe = st->buf_speech_enc_pe + st->encoderPastSamples_enc;
		@@ -1227,6 +1241,11 @@ static void init_acelp(
		{
		set_zero( st->hTcxEnc->Txnq_flt, L_FRAME32k / 2 + 64 );
		st->hTcxEnc->acelp_zir_flt = st->hTcxEnc->Txnq_flt + ( st->L_frame / 2 );

		#ifdef IVAS_FLOAT_FIXED
		set16_fx( st->hTcxEnc->Txnq, 0, L_FRAME32k / 2 + 64 );
		st->hTcxEnc->acelp_zir = st->hTcxEnc->Txnq + ( st->L_frame / 2 );
		#endif
		}
		}
		else /Rate switching/
		@@ -1242,6 +1261,10 @@ static void init_acelp(
		lerp_flt( st->hTcxEnc->Txnq_flt, st->hTcxEnc->Txnq_flt, st->hTcxCfg->tcx_mdct_window_length, st->hTcxCfg->tcx_mdct_window_length_old );
		}
		st->hTcxEnc->acelp_zir_flt = st->hTcxEnc->Txnq_flt + ( st->L_frame / 2 );

		#ifdef IVAS_FLOAT_FIXED
		st->hTcxEnc->acelp_zir = st->hTcxEnc->Txnq + ( st->L_frame / 2 );
		#endif
		}

		/* Rate switching */

lib_enc/evs_enc.c

+175 −0

Original line number	Diff line number	Diff line
		@@ -42,6 +42,8 @@
		#include "wmc_auto.h"
		#ifdef IVAS_FLOAT_FIXED
		#include "ivas_prot_fx.h"
		#include "prot_fx.h"
		#include "prot_fx_enc.h"
		#endif

		/-------------------------------------------------------------------
		@@ -248,7 +250,180 @@ ivas_error evs_enc(

		if ( st->core == HQ_CORE )
		{
		#ifdef IVAS_FLOAT_FIXED
		#ifdef IVAS_FLOAT_FIXED_CONVERSIONS
		PWord16 *p;
		Word16 q = 0;
		Word16 overlap, tmp;
		Word16 left_overlap, right_overlap;
		IF( st->element_mode > EVS_MONO && ( st->last_core == ACELP_CORE \|\| st->last_core == AMR_WB_CORE ) )
		{
		left_overlap = st->hTcxCfg->tcx_mdct_window_lengthFB;
		right_overlap = st->hTcxCfg->tcx_mdct_window_delayFB;
		Word16 L_frame1 = 640;

		p = st->hTcxCfg->tcx_mdct_window_minimumFB;
		tmp = shr( st->hTcxCfg->tcx_mdct_window_min_lengthFB, 1 );

		Word16 q1 = Q_factor_arr( st->hTcxCfg->tcx_mdct_window_minimumFB_flt, tmp );
		FOR( i = 0; i < tmp; i++ )
		{
		p[i].v.im = float_to_fix16( st->hTcxCfg->tcx_mdct_window_minimumFB_flt[i], q1 );
		}
		FOR( i = 0; i < tmp; i++ )
		{
		p[tmp - 1 - i].v.re = float_to_fix16( st->hTcxCfg->tcx_mdct_window_minimumFB_flt[i + tmp], q1 );
		}

		p = st->hTcxCfg->tcx_mdct_window_transFB;
		tmp = shr( st->hTcxCfg->tcx_mdct_window_min_lengthFB, 1 );

		FOR( i = 0; i < tmp; i++ )
		{
		p[i].v.im = float_to_fix16( st->hTcxCfg->tcx_mdct_window_transFB_flt[i], q1 );
		}
		FOR( i = 0; i < tmp; i++ )
		{
		p[tmp - 1 - i].v.re = float_to_fix16( st->hTcxCfg->tcx_mdct_window_transFB_flt[i + tmp], q1 );
		}

		p = st->hTcxCfg->tcx_aldo_window_2_FB;
		tmp = shr( st->hTcxCfg->tcx_mdct_window_delayFB, 1 );

		FOR( i = 0; i < tmp; i++ )
		{
		p[i].v.im = float_to_fix16( st->hTcxCfg->tcx_aldo_window_2_FB_flt[i], q1 );
		}
		FOR( i = 0; i < tmp; i++ )
		{
		p[tmp - 1 - i].v.re = float_to_fix16( st->hTcxCfg->tcx_aldo_window_2_FB_flt[i + tmp], q1 );
		}

		tmp = st->hTcxCfg->tcx_mdct_window_min_length / 2;
		p = st->hTcxCfg->tcx_mdct_window_minimum;
		FOR( i = 0; i < tmp; i++ )
		{
		p[i].v.im = float_to_fix16( st->hTcxCfg->tcx_mdct_window_minimum_flt[i], q1 );
		}
		FOR( i = 0; i < tmp; i++ )
		{
		p[tmp - 1 - i].v.re = float_to_fix16( st->hTcxCfg->tcx_mdct_window_minimum_flt[i + tmp], q1 );
		}


		Word16 l1 = ( shr( left_overlap, 1 ) * 2 ) + sub( L_frame1, shr( add( left_overlap, right_overlap ), 1 ) ) + ( shr( right_overlap, 1 ) * 2 ) - shr( left_overlap, 1 ) + st->hTcxCfg->tcx_offsetFB;
		q = Q_factor_arr( st->hTcxEnc->speech_TCX_flt, l1 ) - 1;
		st->hTcxEnc->q_speech_TCX = q;

		Word16 offset, overlap1;
		offset = negate( shr( st->hTcxCfg->tcx_mdct_window_trans_lengthFB, 1 ) );
		overlap1 = st->hTcxCfg->tcx_mdct_window_trans_lengthFB;

		floatToFixed_arr( st->hTcxEnc->speech_TCX_flt - overlap1 / 2 + offset, st->hTcxEnc->speech_TCX - overlap1 / 2 + offset, q, l1 );
		floatToFixed_arr( st->hTcxEnc->speech_TCX_flt - overlap1 / 2 + offset, st->hTcxEnc->speech_TCX - overlap1 / 2 + offset, q, 2 * L_FRAME48k + 960 );

		q1 = 15;
		tmp = st->hTcxCfg->tcx_mdct_window_half_length / 2;
		p = st->hTcxCfg->tcx_mdct_window_half;
		FOR( i = 0; i < tmp; i++ )
		{
		p[i].v.im = float_to_fix16( st->hTcxCfg->tcx_mdct_window_half_flt[i], q1 );
		}
		FOR( i = 0; i < tmp; i++ )
		{
		p[tmp - 1 - i].v.re = float_to_fix16( st->hTcxCfg->tcx_mdct_window_half_flt[i + tmp], q1 );
		}

		tmp = st->hTcxCfg->tcx_mdct_window_min_length / 2;
		p = st->hTcxCfg->tcx_mdct_window_trans;
		FOR( i = 0; i < tmp; i++ )
		{
		p[i].v.im = float_to_fix16( st->hTcxCfg->tcx_mdct_window_trans_flt[i], q1 );
		}
		FOR( i = 0; i < tmp; i++ )
		{
		p[tmp - 1 - i].v.re = float_to_fix16( st->hTcxCfg->tcx_mdct_window_trans_flt[i + tmp], q1 );
		}

		overlap = st->hTcxCfg->tcx_mdct_window_length;
		tmp = overlap / 2;
		p = st->hTcxCfg->tcx_aldo_window_2;
		FOR( i = 0; i < tmp; i++ )
		{
		p[i].v.im = float_to_fix16( st->hTcxCfg->tcx_aldo_window_2_flt[i], q1 );
		}
		FOR( i = 0; i < tmp; i++ )
		{
		p[tmp - 1 - i].v.re = float_to_fix16( st->hTcxCfg->tcx_aldo_window_2_flt[i + tmp], q1 );
		}

		tmp = overlap / 2;
		p = st->hTcxCfg->tcx_aldo_window_1_trunc;
		FOR( i = -NS2SA( st->sr_core, N_ZERO_MDCT_NS ); i < tmp; i++ )
		{
		p[i].v.im = float_to_fix16( st->hTcxCfg->tcx_aldo_window_1_trunc_flt[i], q1 );
		}
		FOR( i = -NS2SA( st->sr_core, N_ZERO_MDCT_NS ); i < tmp; i++ )
		{
		p[tmp - 1 - i].v.re = float_to_fix16( st->hTcxCfg->tcx_aldo_window_1_trunc_flt[i + tmp], q1 );
		}

		FOR( i = 0; i < L_FRAME32k / 2 + 64; i++ )
		{
		st->hTcxEnc->Txnq[i] = float_to_fix16( st->hTcxEnc->Txnq_flt[i], 0 );
		}
		}
		ELSE
		{
		FOR( i = 0; i < L_FRAME32k; i++ )
		{
		st->hTcxEnc->old_out_fx[i] = float_to_fix16( st->hTcxEnc->old_out[i], 0 );
		}
		}

		IF( NE_16( hq_core_type, LOW_RATE_HQ_CORE ) )
		{
		IF( NE_16( st->element_mode, EVS_MONO ) )
		{
		Word16 temp_e;

		f2me( st->hHQ_core->crest_lp, &st->hHQ_core->crest_lp_fx, &temp_e );
		st->hHQ_core->crest_lp_q = sub( Q31, temp_e );
		f2me( st->hHQ_core->crest_mod_lp, &st->hHQ_core->crest_mod_lp_fx, &temp_e );
		st->hHQ_core->crest_mod_lp_q = sub( Q31, temp_e );
		}
		}

		floatToFixed_arr( st->input - delay, st->input_fx - delay, 0, 960 );
		#endif
		hq_core_enc_ivas_fx( st, st->input_fx - delay, input_frame, hq_core_type, Voicing_flag, vad_hover_flag );

		#ifdef IVAS_FLOAT_FIXED_CONVERSIONS
		IF( EQ_16( hq_core_type, LOW_RATE_HQ_CORE ) )
		{
		fixedToFloat_arrL( st->hHQ_core->last_ni_gain_fx, st->hHQ_core->last_ni_gain, Q17, BANDS_MAX );
		fixedToFloat_arr( st->hHQ_core->last_env_fx, st->hHQ_core->last_env, Q1, BANDS_MAX );
		}
		ELSE
		{
		IF( NE_16( st->element_mode, EVS_MONO ) )
		{

		st->hHQ_core->crest_lp = fix_to_float( st->hHQ_core->crest_lp_fx, st->hHQ_core->crest_lp_q );
		st->hHQ_core->crest_mod_lp = fix_to_float( st->hHQ_core->crest_mod_lp_fx, st->hHQ_core->crest_mod_lp_q );
		}
		}
		IF( st->element_mode > EVS_MONO )
		{
		FOR( i = 0; i < L_FRAME32k; i++ )
		{
		st->hLPDmem->old_exc_flt[i] = fix16_to_float( st->hLPDmem->old_exc[i], 0 );
		}
		}
		#endif
		#else
		hq_core_enc( st, st->input - delay, input_frame, hq_core_type, Voicing_flag, vad_hover_flag );
		#endif
		}

		/---------------------------------------------------------------------

lib_enc/fd_cng_enc.c

+131 −23

Original line number	Diff line number	Diff line
		@@ -1180,10 +1180,19 @@ void FdCngEncodeMDCTStereoSID(
		int16_t gain_idx[CPE_CHANNELS];
		int16_t N, stages, ch, p, coh_idx;
		float side_energy;
		#ifdef IVAS_FLOAT_FIXED
		Word32 *lr_out_ptr_fx[CPE_CHANNELS];
		Word32 gain_fx[CPE_CHANNELS];
		Word32 side_energy_fx;
		Word32 *invTrfMatrix_fx;
		Word32 tmpRAM_fx[FDCNG_VQ_MAX_LEN][FDCNG_VQ_DCT_MAXTRUNC]; /2418*/
		invTrfMatrix_fx = (Word32 ) tmpRAM_fx; / dynamically filled */
		#endif
		int16_t no_side_flag;
		int16_t is_inp_ms;

		float tot_sig_ext[FDCNG_VQ_MAX_LEN], dct_target[CPE_CHANNELS][FDCNG_VQ_DCT_MAXTRUNC]; /* 24 +218/
		float tot_sig_ext[FDCNG_VQ_MAX_LEN] /, dct_target[CPE_CHANNELS][FDCNG_VQ_DCT_MAXTRUNC]/; /* 24 +218/
		Word32 tot_sig_ext_fx[FDCNG_VQ_MAX_LEN], dct_target_fx[CPE_CHANNELS][FDCNG_VQ_DCT_MAXTRUNC]; /* 24 +218/
		float *invTrfMatrix;
		float tmpRAM[FDCNG_VQ_MAX_LEN][FDCNG_VQ_DCT_MAXTRUNC]; /2418*/
		invTrfMatrix = (float ) tmpRAM; / dynamically filled */
		@@ -1202,6 +1211,9 @@ void FdCngEncodeMDCTStereoSID(
		lr_in_ptr[ch] = &sts[ch]->hFdCngEnc->msNoiseEst[0];
		ms_ptr[ch] = &logNoiseEst[ch][0];
		lr_out_ptr[ch] = &sts[ch]->hFdCngEnc->hFdCngCom->sidNoiseEst_flt[0];
		#ifdef IVAS_FLOAT_FIXED
		lr_out_ptr_fx[ch] = &sts[ch]->hFdCngEnc->hFdCngCom->sidNoiseEst[0];
		#endif
		}
		N = sts[0]->hFdCngEnc->npartDec;
		set_f( weights, 1.f, NPART );
		@@ -1216,17 +1228,19 @@ void FdCngEncodeMDCTStereoSID(
		E[ch] += ms_ptr[ch][p];
		}
		}

		#ifdef IVAS_FLOAT_FIXED_CONVERSIONS
		Word32 ms_ptr_fx[2][NPART];
		Word16 q = s_min( Q_factor_arrL( ms_ptr[0], N ), Q_factor_arrL( ms_ptr[1], N ) ) - 1;
		floatToFixed_arrL( ms_ptr[0], ms_ptr_fx[0], q, N );
		floatToFixed_arrL( ms_ptr[1], ms_ptr_fx[1], q, N );
		#endif
		/* M/S transform on log envelopes */
		if ( is_inp_ms == 0 )
		{
		#ifndef IVAS_FLOAT_FIXED
		convertToMS( N, ms_ptr[0], ms_ptr[1], 0.5f );
		#else
		Word32 ms_ptr_fx[2][NPART];
		Word16 q = s_min( Q_factor_arrL( ms_ptr[0], N ), Q_factor_arrL( ms_ptr[1], N ) ) - 1;
		floatToFixed_arrL( ms_ptr[0], ms_ptr_fx[0], q, N );
		floatToFixed_arrL( ms_ptr[1], ms_ptr_fx[1], q, N );


		convertToMS_fx( N, ms_ptr_fx[0], ms_ptr_fx[1], ONE_IN_Q30 );

		@@ -1234,8 +1248,15 @@ void FdCngEncodeMDCTStereoSID(
		fixedToFloat_arrL( ms_ptr_fx[1], ms_ptr[1], q, N );
		#endif
		}

		#ifndef IVAS_FLOAT_FIXED
		side_energy = sum2_f( ms_ptr[1], N );
		#else
		Word16 gb = find_guarded_bits_fx( N );
		side_energy_fx = sum2_f_32_fx( ms_ptr_fx[1], N, gb );
		#ifdef IVAS_FLOAT_FIXED_CONVERSIONS
		side_energy = fixedToFloat( side_energy_fx, 2 * q - 31 - gb );
		#endif
		#endif

		/* do not transmit side shape if initial noise shapes are very similar */
		if ( side_energy <= 0.1f )
		@@ -1269,27 +1290,68 @@ void FdCngEncodeMDCTStereoSID(
		/* High quality cosine smooth basis extension used to not introduce noise in stage#1 DCT24 analysis and subsequent VQ-steps */
		if ( N == FDCNG_VQ_MAX_LEN_WB )
		{
		create_IDCT_N_Matrix( invTrfMatrix, N, FDCNG_VQ_DCT_MAXTRUNC, sizeof( tmpRAM ) / ( sizeof( float ) ) ); /WB: create truncated IDCT21 matrix /
		#ifdef IVAS_FLOAT_FIXED
		Word16 size_value, temp_e;
		size_value = BASOP_Util_Divide1616_Scale( sizeof( tmpRAM_fx ), ( sizeof( Word32 ) ), &temp_e ); /Q15/
		size_value = shr( size_value, sub( 15, temp_e ) );
		create_IDCT_N_Matrix_fx( invTrfMatrix_fx, N, FDCNG_VQ_DCT_MAXTRUNC, size_value ); // Q31 /WB: create truncated IDCT21 matrix /
		#ifdef IVAS_FLOAT_FIXED_CONVERSIONS
		for ( int i = 0; i < FDCNG_VQ_MAX_LEN; i++ )
		{
		fixedToFloat_arrL( tmpRAM_fx[i], tmpRAM[i], Q31, FDCNG_VQ_DCT_MAXTRUNC );
		}
		q = s_min( Q_factor_arrL( ms_ptr[0], N ), Q_factor_arrL( ms_ptr[1], N ) ) - 1;
		floatToFixed_arrL( ms_ptr[0], ms_ptr_fx[0], q, N );
		floatToFixed_arrL( ms_ptr[1], ms_ptr_fx[1], q, N );

		#endif
		#else
		create_IDCT_N_Matrix( invTrfMatrix, N, FDCNG_VQ_DCT_MAXTRUNC, sizeof( tmpRAM ) / ( sizeof( float ) ) ); /WB: create truncated IDCT21 matrix /
		#endif
		for ( ch = 0; ch < CPE_CHANNELS; ch++ )
		{
		/* run DCT_N N==21 , truncated at 18/21 ~= 86% , i.e use a bit better better quality in extrapolation , than subsequent DCT24 analysis which is truncated at 75%*/
		#ifdef IVAS_FLOAT_FIXED
		/* truncated DCT 21 analysis */
		dctT2_N_apply_matrix_fx( (const Word32 *) ms_ptr_fx[ch], dct_target_fx[ch], FDCNG_VQ_DCT_MAXTRUNC, N, invTrfMatrix_fx, FDCNG_VQ_DCT_MAXTRUNC, DCT_T2_21_XX );
		/* extrapolate extend fdcng envelope signal in the fdncg ienvelope/"time" domain using DCT21 basis vectors,
		estimated DCT21 coeffs scaling extended basis vectors are used to create extrapolated length 24 input target envelope signal */
		/* this DCT21 extension does not introduce DCT24 coefficient noise for the subsequent dct24 target analysis, and later in IDCT24 synthesis */

		/* truncated IDCT 21 extension synthesis */
		extend_dctN_input_fx( ms_ptr_fx[ch], dct_target_fx[ch], N, tot_sig_ext_fx, FDCNG_VQ_MAX_LEN, invTrfMatrix_fx /* DCT_N basis vectors /, FDCNG_VQ_DCT_MAXTRUNC, IDCT_T2_XX_21 ); / use 18 basis vectors*/
		#ifdef IVAS_FLOAT_FIXED_CONVERSIONS
		fixedToFloat_arrL( tot_sig_ext_fx, tot_sig_ext, q, FDCNG_VQ_MAX_LEN );
		#endif
		#else
		/* truncated DCT 21 analysis */
		dctT2_N_apply_matrix( (const float *) ms_ptr[ch], dct_target[ch], FDCNG_VQ_DCT_MAXTRUNC, N, invTrfMatrix, FDCNG_VQ_DCT_MAXTRUNC, DCT_T2_21_XX );

		/* extrapolate extend fdcng envelope signal in the fdncg ienvelope/"time" domain using DCT21 basis vectors,
		estimated DCT21 coeffs scaling extended basis vectors are used to create extrapolated length 24 input target envelope signal */
		/* this DCT21 extension does not introduce DCT24 coefficient noise for the subsequent dct24 target analysis, and later in IDCT24 synthesis */

		/* truncated IDCT 21 extension synthesis */
		extend_dctN_input( ms_ptr[ch], dct_target[ch], N, tot_sig_ext, FDCNG_VQ_MAX_LEN, invTrfMatrix /* DCT_N basis vectors /, FDCNG_VQ_DCT_MAXTRUNC, IDCT_T2_XX_21 ); / use 18 basis vectors*/
		#endif


		mvr2r( tot_sig_ext, ms_ptr[ch], FDCNG_VQ_MAX_LEN ); /* write extended result as input to VQ */
		}
		}
		#ifdef IVAS_FLOAT_FIXED
		Word16 size_value, temp_e;
		size_value = BASOP_Util_Divide1616_Scale( sizeof( tmpRAM_fx ), ( sizeof( Word32 ) ), &temp_e ); /Q15/
		size_value = shr( size_value, sub( 15, temp_e ) );
		create_IDCT_N_Matrix_fx( invTrfMatrix_fx, FDCNG_VQ_MAX_LEN, FDCNG_VQ_DCT_MAXTRUNC, size_value ); /always create/set up IDCT24 matrix in RAM /
		#ifdef IVAS_FLOAT_FIXED_CONVERSIONS
		for ( int i = 0; i < FDCNG_VQ_MAX_LEN; i++ )
		{
		fixedToFloat_arrL( tmpRAM_fx[i], tmpRAM[i], Q31, FDCNG_VQ_DCT_MAXTRUNC );
		}
		#endif
		#else
		create_IDCT_N_Matrix( invTrfMatrix, FDCNG_VQ_MAX_LEN, FDCNG_VQ_DCT_MAXTRUNC, sizeof( tmpRAM ) / ( sizeof( float ) ) ); /always create/set up IDCT24 matrix in RAM /

		#endif
		/* end split */
		for ( ch = 0; ch < CPE_CHANNELS; ch++ )
		{
		@@ -1310,19 +1372,17 @@ void FdCngEncodeMDCTStereoSID(
		{
		#ifdef IVAS_FLOAT_FIXED
		#ifdef IVAS_FLOAT_FIXED_CONVERSIONS
		Word32 ms_ptr_fx[24];

		Word16 ms_ptr_e;
		Word32 *invTrfMatrix_fx;
		Word32 tmpRAM_fx[FDCNG_VQ_MAX_LEN][FDCNG_VQ_DCT_MAXTRUNC];
		Word16 weights_fx[24];
		invTrfMatrix_fx = (Word32 *) tmpRAM_fx;

		f2me_buf( ms_ptr[ch], ms_ptr_fx, &ms_ptr_e, N );
		f2me_buf( ms_ptr[ch], ms_ptr_fx[ch], &ms_ptr_e, N );
		floatToFixed_arrL( invTrfMatrix, invTrfMatrix_fx, Q31, FDCNG_VQ_MAX_LEN * FDCNG_VQ_DCT_MAXTRUNC );
		floatToFixed_arr( weights, weights_fx, Q8, 24 );

		#endif
		msvq_enc_ivas_fx( ivas_cdk_37bits_fx, Q7, NULL, NULL, ms_ptr_fx, ms_ptr_e, levels_37bits, FD_CNG_maxC_37bits, stages, weights_fx, N, FD_CNG_maxN_37bits, 1, invTrfMatrix_fx, indices[ch] );
		msvq_enc_ivas_fx( ivas_cdk_37bits_fx, Q7, NULL, NULL, ms_ptr_fx[ch], ms_ptr_e, levels_37bits, FD_CNG_maxC_37bits, stages, weights_fx, N, FD_CNG_maxN_37bits, 1, invTrfMatrix_fx, indices[ch] );
		#ifdef IVAS_FLOAT_FIXED_CONVERSIONS
		fixedToFloat_arrL( invTrfMatrix_fx, invTrfMatrix, Q31, FDCNG_VQ_MAX_LEN * FDCNG_VQ_DCT_MAXTRUNC );
		#endif
		@@ -1344,15 +1404,17 @@ void FdCngEncodeMDCTStereoSID(
		#ifndef IVAS_FLOAT_FIXED
		convertToMS( N, ms_ptr[0], ms_ptr[1], 1.0f );
		#else
		Word32 ms_ptr_fx[2][NPART];
		Word16 q = s_min( Q_factor_arrL( ms_ptr[0], N ), Q_factor_arrL( ms_ptr[1], N ) ) - 1;
		#ifdef IVAS_FLOAT_FIXED_CONVERSIONS
		// Word32 ms_ptr_fx[2][NPART];
		q = s_min( Q_factor_arrL( ms_ptr[0], N ), Q_factor_arrL( ms_ptr[1], N ) ) - 1;
		floatToFixed_arrL( ms_ptr[0], ms_ptr_fx[0], q, N );
		floatToFixed_arrL( ms_ptr[1], ms_ptr_fx[1], q, N );

		#endif
		convertToMS_fx( N, ms_ptr_fx[0], ms_ptr_fx[1], ONE_IN_Q31 );

		#ifdef IVAS_FLOAT_FIXED_CONVERSIONS
		fixedToFloat_arrL( ms_ptr_fx[0], ms_ptr[0], q, N );
		fixedToFloat_arrL( ms_ptr_fx[1], ms_ptr[1], q, N );
		#endif
		#endif
		}

		@@ -1365,8 +1427,18 @@ void FdCngEncodeMDCTStereoSID(
		gain[ch] += ms_ptr[ch][p];
		}
		gain[ch] = ( E[ch] - gain[ch] ) / (float) N;
		#ifdef IVAS_FLOAT_FIXED
		#ifdef IVAS_FLOAT_FIXED_CONVERSIONS
		gain_fx[ch] = floatToFixed_32( gain[ch], Q23 );
		#endif
		apply_scale( &gain_fx[ch], sts[ch]->hFdCngEnc->hFdCngCom->CngBandwidth, sts[ch]->element_brate, scaleTableStereo, SIZE_SCALE_TABLE_STEREO );
		#ifdef IVAS_FLOAT_FIXED_CONVERSIONS
		gain[ch] = fixedToFloat_32( gain_fx[ch], Q23 );
		#endif
		#else
		apply_scale_flt( &gain[ch], sts[ch]->hFdCngEnc->hFdCngCom->CngBandwidth, sts[ch]->element_brate, scaleTableStereo, SIZE_SCALE_TABLE_STEREO );

		#endif
		/* quantize gain */
		gain_idx[ch] = (int16_t) floor( gain[ch] * 1.5f + GAIN_Q_OFFSET_IVAS + .5f );
		gain_idx[ch] = max( 0, min( 127, gain_idx[ch] ) );
		@@ -1385,11 +1457,27 @@ void FdCngEncodeMDCTStereoSID(
		lr_out_ptr[ch][p] = powf( 10.f, ( ms_ptr[ch][p] + gain[ch] ) / 10.f );
		}

		#ifdef IVAS_FLOAT_FIXED
		#ifdef IVAS_FLOAT_FIXED_CONVERSIONS
		q = Q_factor_arrL( lr_out_ptr[ch], N );
		floatToFixed_arrL( lr_out_ptr[ch], lr_out_ptr_fx[ch], q, N );
		#endif
		/* scale bands and get scalefactors */
		scalebands( lr_out_ptr_fx[ch], hFdCngEnc->partDec, N, hFdCngEnc->midbandDec, hFdCngEnc->nFFTpartDec, sub( hFdCngEnc->stopBandDec, hFdCngEnc->startBandDec ), hFdCngCom->cngNoiseLevel, 1 );
		#ifdef IVAS_FLOAT_FIXED_CONVERSIONS
		fixedToFloat_arrL( hFdCngCom->cngNoiseLevel, hFdCngCom->cngNoiseLevel_flt, q, FFTCLDFBLEN );
		#endif
		lpc_from_spectrum( hFdCngCom, hFdCngEnc->startBandDec, hFdCngEnc->stopFFTbinDec, sts[ch]->preemph_fac );
		#ifdef IVAS_FLOAT_FIXED_CONVERSIONS
		sts[ch]->preemph_fac_flt = fixedToFloat( sts[ch]->preemph_fac, Q15 );
		#endif
		#else
		/* scale bands and get scalefactors */
		scalebands_flt( lr_out_ptr[ch], hFdCngEnc->partDec, N, hFdCngEnc->midbandDec, hFdCngEnc->nFFTpartDec, hFdCngEnc->stopBandDec - hFdCngEnc->startBandDec, hFdCngCom->cngNoiseLevel_flt, 1 );

		lpc_from_spectrum_flt( hFdCngCom, hFdCngEnc->startBandDec, hFdCngEnc->stopFFTbinDec, sts[ch]->preemph_fac_flt );

		#endif

		sts[ch]->hDtxEnc->last_CNG_L_frame = sts[ch]->L_frame;
		}

		@@ -1451,6 +1539,9 @@ void FdCngEncodeDiracMDCTStereoSID(
		float *lr_in_ptr[CPE_CHANNELS];
		float *ms_ptr[CPE_CHANNELS];
		float *lr_out_ptr[CPE_CHANNELS];
		#ifdef IVAS_FLOAT_FIXED
		Word32 *lr_out_ptr_fx[CPE_CHANNELS];
		#endif
		float logNoiseEst[CPE_CHANNELS][NPART];
		float E[CPE_CHANNELS];
		float gain[CPE_CHANNELS];
		@@ -1473,6 +1564,9 @@ void FdCngEncodeDiracMDCTStereoSID(
		lr_in_ptr[ch] = &sts[ch]->hFdCngEnc->msNoiseEst[0];
		ms_ptr[ch] = &logNoiseEst[ch][0];
		lr_out_ptr[ch] = &sts[ch]->hFdCngEnc->hFdCngCom->sidNoiseEst_flt[0];
		#ifdef IVAS_FLOAT_FIXED
		lr_out_ptr_fx[ch] = &sts[ch]->hFdCngEnc->hFdCngCom->sidNoiseEst[0];
		#endif
		}
		set_f( weights, 1.f, NPART );

		@@ -1602,12 +1696,26 @@ void FdCngEncodeDiracMDCTStereoSID(
		{
		lr_out_ptr[ch][N[ch] - 1] *= SWB_13k2_LAST_BAND_SCALE_FLT;
		}

		#ifdef IVAS_FLOAT_FIXED
		#ifdef IVAS_FLOAT_FIXED_CONVERSIONS
		q = Q_factor_arrL( lr_out_ptr[ch], N[ch] );
		floatToFixed_arrL( lr_out_ptr[ch], lr_out_ptr_fx[ch], q, N[ch] );
		#endif
		/* scale bands and get scalefactors */
		scalebands( lr_out_ptr_fx[ch], hFdCngEnc->partDec, N[ch], hFdCngEnc->midbandDec, hFdCngEnc->nFFTpartDec, sub( hFdCngEnc->stopBandDec, hFdCngEnc->startBandDec ), hFdCngCom->cngNoiseLevel, 1 );
		#ifdef IVAS_FLOAT_FIXED_CONVERSIONS
		fixedToFloat_arrL( hFdCngCom->cngNoiseLevel, hFdCngCom->cngNoiseLevel_flt, q, FFTCLDFBLEN );
		#endif
		lpc_from_spectrum( hFdCngCom, hFdCngEnc->startBandDec, hFdCngEnc->stopFFTbinDec, sts[ch]->preemph_fac );
		#ifdef IVAS_FLOAT_FIXED_CONVERSIONS
		sts[ch]->preemph_fac_flt = fixedToFloat( sts[ch]->preemph_fac, Q15 );
		#endif
		#else
		/* scale bands and get scalefactors */
		scalebands_flt( lr_out_ptr[ch], hFdCngEnc->partDec, N[ch], hFdCngEnc->midbandDec, hFdCngEnc->nFFTpartDec, hFdCngEnc->stopBandDec - hFdCngEnc->startBandDec, hFdCngCom->cngNoiseLevel_flt, 1 );

		lpc_from_spectrum_flt( hFdCngCom, hFdCngEnc->startBandDec, hFdCngEnc->stopFFTbinDec, sts[ch]->preemph_fac_flt );

		#endif
		sts[ch]->hDtxEnc->last_CNG_L_frame = sts[ch]->L_frame;
		}
		sts[0]->hFdCngEnc->hFdCngCom->coherence_flt = 0.0f;

lib_enc/hq_core_enc.c

+364 −50

File changed.

Preview size limit exceeded, changes collapsed.