Float structure elements cleanup for encoder (5732a17d) · Commits · SA4 / Audio / IVAS BASOP

lib_enc/acelp_core_enc.c

+0 −2

Original line number	Diff line number	Diff line
		@@ -556,8 +556,6 @@ ivas_error acelp_core_enc(
		{
		st->mem_deemp_preQ_fx = 0;
		move16();

		st->mem_preemp_preQ = 0.0f;
		st->last_code_preq = 0;
		move16();
		st->last_nq_preQ = 0;

lib_enc/core_enc_init.c

+47 −22

Original line number	Diff line number	Diff line
		@@ -823,9 +823,14 @@ static void init_sig_buffers(
		/* Initialize Signal Buffers and Pointers at encoder-sampling-rate */
		if ( st->ini_frame == 0 )
		{
		#ifdef IVAS_FLOAT_FIXED
		set16_fx( st->buf_speech_enc_pe, 0, L_PAST_MAX_32k + L_FRAME32k + L_NEXT_MAX_32k );
		st->exp_buf_speech_enc_pe = 0;
		set16_fx( st->buf_speech_enc, 0, L_PAST_MAX_32k + L_FRAME32k + L_NEXT_MAX_32k );
		st->exp_buf_speech_enc = 0;
		#else
		set_zero( st->buf_speech_enc_flt, L_PAST_MAX_32k + L_FRAME32k + L_NEXT_MAX_32k );
		set_zero( st->buf_speech_enc_pe_flt, L_PAST_MAX_32k + L_FRAME32k + L_NEXT_MAX_32k );
		#ifndef IVAS_FLOAT_FIXED
		set_zero( st->buf_wspeech_enc_flt, L_FRAME16k + L_SUBFR + L_FRAME16k + L_NEXT_MAX_16k + 320 );
		#endif
		if ( hTcxEnc != NULL )
		@@ -836,33 +841,50 @@ static void init_sig_buffers(
		else if ( st->L_frame != L_frame_old && !( ( total_brate >= ACELP_16k40 && total_brate <= ACELP_24k40 ) &&
		( total_brate == last_total_brate ) && ( st->last_bwidth == st->bwidth ) ) )
		{
		#ifndef IVAS_FLOAT_FIXED
		lerp_flt( st->buf_speech_enc_flt, st->buf_speech_enc_flt, st->L_frame, L_frame_old );

		if ( ( st->last_core != TCX_20_CORE ) && ( st->last_core != TCX_10_CORE ) )
		{
		#ifdef IVAS_FLOAT_FIXED
		floatToFixed_arr( st->buf_speech_enc_flt, st->buf_speech_enc, 0, st->L_frame );
		#endif
		}

		#ifndef IVAS_FLOAT_FIXED
		mvr2r( st->old_wsp, st->buf_wspeech_enc_flt + st->L_frame + L_SUBFR - L_WSP_MEM, L_WSP_MEM );
		#endif
		/Resamp buffers needed only for ACELP/
		if ( st->L_frame == L_FRAME && !st->tcxonly )
		{
		#ifndef IVAS_FLOAT_FIXED // Need to remove after cleanup of old_inp_12k8
		mvr2r( st->old_inp_12k8, st->buf_speech_enc_pe_flt + st->L_frame - L_INP_MEM, L_INP_MEM );
		#else
		Word16 tmp;
		f2me_buf_16( st->old_inp_12k8, st->buf_speech_enc_pe + st->L_frame - L_INP_MEM, &tmp, L_INP_MEM );
		/* SCaling to common exponent*/
		Scale_sig( st->buf_speech_enc_pe + st->L_frame - L_INP_MEM, L_INP_MEM, sub( tmp, s_max( tmp, st->exp_buf_speech_enc_pe ) ) );
		Scale_sig( st->buf_speech_enc_pe, st->L_frame - L_INP_MEM, sub( st->exp_buf_speech_enc_pe, s_max( tmp, st->exp_buf_speech_enc_pe ) ) );
		Scale_sig( st->buf_speech_enc_pe + st->L_frame, sub( L_PAST_MAX_32k + L_FRAME32k + L_NEXT_MAX_32k, st->L_frame ), sub( st->exp_buf_speech_enc_pe, s_max( tmp, st->exp_buf_speech_enc_pe ) ) );
		st->exp_buf_speech_enc_pe = s_max( tmp, st->exp_buf_speech_enc_pe );
		#endif
		}
		else if ( st->L_frame == L_FRAME16k && !st->tcxonly )
		{
		#ifndef IVAS_FLOAT_FIXED
		lerp_flt( st->buf_wspeech_enc_flt + st->L_frame + L_SUBFR - L_WSP_MEM, st->buf_wspeech_enc_flt + st->L_frame + L_SUBFR - 310, 310, L_WSP_MEM );
		#endif
		#ifndef IVAS_FLOAT_FIXED // Need to remove after cleanup of old_inp_12k8
		mvr2r( st->old_inp_16k, st->buf_speech_enc_pe_flt + st->L_frame - L_INP_MEM, L_INP_MEM );
		#else
		Word16 tmp;
		f2me_buf_16( st->old_inp_12k8, st->buf_speech_enc_pe + st->L_frame - L_INP_MEM, &tmp, L_INP_MEM );
		/* SCaling to common exponent*/
		Scale_sig( st->buf_speech_enc_pe + st->L_frame - L_INP_MEM, L_INP_MEM, sub( tmp, s_max( tmp, st->exp_buf_speech_enc_pe ) ) );
		Scale_sig( st->buf_speech_enc_pe, st->L_frame - L_INP_MEM, sub( st->exp_buf_speech_enc_pe, s_max( tmp, st->exp_buf_speech_enc_pe ) ) );
		Scale_sig( st->buf_speech_enc_pe + st->L_frame, sub( L_PAST_MAX_32k + L_FRAME32k + L_NEXT_MAX_32k, st->L_frame ), sub( st->exp_buf_speech_enc_pe, s_max( tmp, st->exp_buf_speech_enc_pe ) ) );
		st->exp_buf_speech_enc_pe = s_max( tmp, st->exp_buf_speech_enc_pe );
		#endif
		}

		st->mem_preemph_enc_flt = st->buf_speech_enc_flt[st->L_frame - 1];
		#ifndef IVAS_FLOAT_FIXED
		st->mem_preemph_enc_flt = st->buf_speech_enc_flt[st->L_frame - 1];
		st->mem_wsp_enc_flt = st->buf_wspeech_enc_flt[st->L_frame + L_SUBFR - 1];
		#endif
		}
		@@ -881,23 +903,11 @@ static void init_sig_buffers(
		#endif
		}

		#ifndef IVAS_FLOAT_FIXED
		st->new_speech_enc_flt = st->buf_speech_enc_flt + st->encoderPastSamples_enc + st->encoderLookahead_enc;
		st->new_speech_enc_pe_flt = st->buf_speech_enc_pe_flt + st->encoderPastSamples_enc + st->encoderLookahead_enc;

		if ( hTcxEnc != NULL )
		{
		// hTcxEnc->new_speech_ltp_flt = hTcxEnc->buf_speech_ltp_flt + st->encoderPastSamples_enc + st->encoderLookahead_enc;
		}

		st->speech_enc_flt = st->buf_speech_enc_flt + st->encoderPastSamples_enc;
		st->speech_enc_pe_flt = st->buf_speech_enc_pe_flt + st->encoderPastSamples_enc;

		if ( hTcxEnc != NULL )
		{
		// hTcxEnc->speech_ltp_flt = hTcxEnc->buf_speech_ltp_flt + st->encoderPastSamples_enc;
		}

		#ifndef IVAS_FLOAT_FIXED
		if ( st->element_mode > EVS_MONO )
		{
		st->wspeech_enc_flt = st->buf_wspeech_enc_flt + st->L_frame + L_SUBFR;
		@@ -932,7 +942,11 @@ static void init_sig_buffers_ivas_fx( Encoder_State *st, const Word16 L_frame_ol
		IF( st->ini_frame == 0 )
		{
		set16_fx( st->buf_speech_enc, 0, L_PAST_MAX_32k + L_FRAME32k + L_NEXT_MAX_32k );
		st->exp_buf_speech_enc = 0;
		move16();
		set16_fx( st->buf_speech_enc_pe, 0, L_PAST_MAX_32k + L_FRAME32k + L_NEXT_MAX_32k );
		st->exp_buf_speech_enc_pe = 0;
		move16();
		if ( hTcxEnc != NULL )
		{
		set16_fx( hTcxEnc->buf_speech_ltp, 0, L_PAST_MAX_32k + L_FRAME32k + L_NEXT_MAX_32k );
		@@ -955,7 +969,9 @@ static void init_sig_buffers_ivas_fx( Encoder_State *st, const Word16 L_frame_ol
		IF( NE_16( st->last_core, TCX_20_CORE ) && NE_16( st->last_core, TCX_10_CORE ) ) /* condition should be checked again */
		{
		Copy( st->buf_speech_enc, hTcxEnc->buf_speech_ltp, st->L_frame );
		hTcxEnc->exp_buf_speech_ltp = st->exp_buf_speech_enc;
		Scale_sig( hTcxEnc->buf_speech_ltp, st->L_frame, sub( st->exp_buf_speech_enc, s_max( st->exp_buf_speech_enc, hTcxEnc->exp_buf_speech_ltp ) ) );
		Scale_sig( hTcxEnc->buf_speech_ltp + st->L_frame, sub( L_PAST_MAX_32k + L_FRAME32k + L_NEXT_MAX_32k, st->L_frame ), sub( hTcxEnc->exp_buf_speech_ltp, s_max( st->exp_buf_speech_enc, hTcxEnc->exp_buf_speech_ltp ) ) );
		hTcxEnc->exp_buf_speech_ltp = s_max( hTcxEnc->exp_buf_speech_ltp, st->exp_buf_speech_enc );
		move16();
		}

		@@ -971,19 +987,28 @@ static void init_sig_buffers_ivas_fx( Encoder_State *st, const Word16 L_frame_ol
		{
		// Copy_Scale_sig( st->old_inp_12k8_fx, st->buf_speech_enc_pe + st->L_frame - L_INP_MEM, L_INP_MEM, sub( st->prev_Q_new, st->prev_Q_old ) );
		Copy( st->old_inp_12k8_fx, st->buf_speech_enc_pe + st->L_frame - L_INP_MEM, L_INP_MEM );
		st->exp_buf_speech_enc_pe = st->exp_old_inp_12k8;
		/* SCaling to common exponent*/
		Scale_sig( st->buf_speech_enc_pe + sub( st->L_frame, L_INP_MEM ), L_INP_MEM, sub( st->exp_old_inp_12k8, s_max( st->exp_old_inp_12k8, st->exp_buf_speech_enc_pe ) ) );
		Scale_sig( st->buf_speech_enc_pe, sub( st->L_frame, L_INP_MEM ), sub( st->exp_buf_speech_enc_pe, s_max( st->exp_old_inp_12k8, st->exp_buf_speech_enc_pe ) ) );
		Scale_sig( st->buf_speech_enc_pe + st->L_frame, sub( L_PAST_MAX_32k + L_FRAME32k + L_NEXT_MAX_32k, st->L_frame ), sub( st->exp_buf_speech_enc_pe, s_max( st->exp_old_inp_12k8, st->exp_buf_speech_enc_pe ) ) );
		st->exp_buf_speech_enc_pe = s_max( st->exp_old_inp_12k8, st->exp_buf_speech_enc_pe );
		move16();
		}
		ELSE IF( EQ_16( st->L_frame, L_FRAME16k ) && !st->tcxonly )
		{
		lerp( st->buf_wspeech_enc + st->L_frame + L_SUBFR - L_WSP_MEM, st->buf_wspeech_enc + st->L_frame + L_SUBFR - 310, 310, L_WSP_MEM );
		Copy( st->old_inp_16k_fx, st->buf_speech_enc_pe + st->L_frame - L_INP_MEM, L_INP_MEM );
		st->exp_buf_speech_enc_pe = st->exp_old_inp_16k;
		/* SCaling to common exponent*/
		Scale_sig( st->buf_speech_enc_pe + sub( st->L_frame, L_INP_MEM ), L_INP_MEM, sub( st->exp_old_inp_12k8, s_max( st->exp_old_inp_12k8, st->exp_buf_speech_enc_pe ) ) );
		Scale_sig( st->buf_speech_enc_pe, sub( st->L_frame, L_INP_MEM ), sub( st->exp_buf_speech_enc_pe, s_max( st->exp_old_inp_12k8, st->exp_buf_speech_enc_pe ) ) );
		Scale_sig( st->buf_speech_enc_pe + st->L_frame, sub( L_PAST_MAX_32k + L_FRAME32k + L_NEXT_MAX_32k, st->L_frame ), sub( st->exp_buf_speech_enc_pe, s_max( st->exp_old_inp_12k8, st->exp_buf_speech_enc_pe ) ) );
		st->exp_buf_speech_enc_pe = s_max( st->exp_old_inp_12k8, st->exp_buf_speech_enc_pe );
		move16();
		}

		st->mem_preemph_enc = st->buf_speech_enc[st->encoderPastSamples_enc + st->encoderLookahead_enc - 1];
		move16();
		st->exp_mem_preemph_enc = st->exp_buf_speech_enc;
		st->mem_wsp_enc = shr( st->buf_wspeech_enc[st->L_frame + L_SUBFR - 1], sub( Q16, st->exp_buf_wspeech_enc ) ); // Q-1
		move16();
		}

lib_enc/core_enc_init_fx.c

+4 −0

Original line number	Diff line number	Diff line
		@@ -462,7 +462,11 @@ void init_sig_buffers_fx( Encoder_State *st, const Word16 L_frame_old, const Wor
		IF( st->ini_frame == 0 )
		{
		set16_fx( st->buf_speech_enc, 0, L_PAST_MAX_32k + L_FRAME32k + L_NEXT_MAX_32k );
		st->exp_buf_speech_enc = 0;
		move16();
		set16_fx( st->buf_speech_enc_pe, 0, L_PAST_MAX_32k + L_FRAME32k + L_NEXT_MAX_32k );
		st->exp_buf_speech_enc_pe = 0;
		move16();
		set16_fx( hTcxEnc->buf_speech_ltp, 0, L_PAST_MAX_32k + L_FRAME32k + L_NEXT_MAX_32k );
		set16_fx( st->buf_wspeech_enc, 0, L_FRAME16k + L_SUBFR + L_FRAME16k + L_NEXT_MAX_16k );
		}

lib_enc/core_enc_switch.c

+0 −1

Original line number	Diff line number	Diff line
		@@ -207,7 +207,6 @@ void core_coder_mode_switch_ivas_fx(
		move32();

		/* Initialize Signal Buffers */
		f2me_buf_16( st->buf_speech_enc_flt, st->buf_speech_enc, &st->exp_buf_speech_enc, st->L_frame );
		f2me_buf_16( st->old_wsp, st->old_wsp_fx, &st->exp_old_wsp, L_WSP_MEM );
		f2me_buf_16( st->old_inp_12k8, st->old_inp_12k8_fx, &st->exp_old_inp_12k8, L_INP_MEM );
		f2me_buf_16( st->old_inp_16k, st->old_inp_16k_fx, &st->exp_old_inp_16k, L_INP_MEM );

lib_enc/init_enc.c

+14 −24

Original line number	Diff line number	Diff line
		@@ -177,19 +177,21 @@ ivas_error init_encoder(
		#ifdef IVAS_FLOAT_FIXED
		set16_fx( st->old_wsp_fx, 0, L_WSP_MEM );
		#endif
		#ifndef IVAS_FLOAT_FIXED
		set_f( st->old_wsp2, 0, ( L_WSP_MEM - L_INTERPOL ) / OPL_DECIM );

		st->mem_preemph = 0.0f;
		st->mem_preemph16k = 0.0f;
		st->mem_preemph_enc_flt = 0.0;
		#endif

		st->mem_preemph = 0.0f;
		#ifdef IVAS_FLOAT_FIXED
		st->mem_preemph_fx = 0;
		st->mem_preemph16k_fx = 0;
		st->mem_preemph_enc = 0;
		st->exp_mem_preemph_enc = 0;
		st->mem_deemp_preQ_fx = 0;
		#endif
		/* AVQ pre-quantizer memory */
		st->mem_preemp_preQ = 0.0f;
		st->last_nq_preQ = 0;
		st->last_code_preq = 0;
		st->use_acelp_preq = 0;
		@@ -197,11 +199,12 @@ ivas_error init_encoder(

		/* (Decimated) Weighted Speech Memory */
		#ifndef IVAS_FLOAT_FIXED
		st->mem_preemp_preQ = 0.0f;
		st->mem_wsp_enc_flt = 0.0;
		#endif
		st->mem_wsp = 0.0f;

		set_f( st->mem_decim2, 0, 3 );
		#endif

		#ifndef IVAS_FLOAT_FIXED
		set_f( st->Bin_E, 0, L_FFT );
		@@ -427,12 +430,9 @@ ivas_error init_encoder(
		st->old_inp_12k8_fx = st->hSignalBuf->old_inp_12k8_fx;
		st->old_inp_16k_fx = st->hSignalBuf->old_inp_16k_fx;
		#endif
		st->buf_speech_enc_pe_flt = st->hSignalBuf->buf_speech_enc_pe_flt;
		#ifndef IVAS_FLOAT_FIXED
		st->buf_synth_flt = st->hSignalBuf->buf_synth_flt;
		#endif
		st->buf_speech_enc_flt = st->hSignalBuf->buf_speech_enc_flt;
		#ifndef IVAS_FLOAT_FIXED
		st->buf_wspeech_enc_flt = st->hSignalBuf->buf_wspeech_enc_flt;
		#endif

		@@ -502,12 +502,9 @@ ivas_error init_encoder(
		st->mem_decim16k = NULL;
		st->old_inp_12k8 = NULL;
		st->old_inp_16k = NULL;
		st->buf_speech_enc_pe_flt = NULL;
		#ifndef IVAS_FLOAT_FIXED
		st->buf_synth_flt = NULL;
		#endif
		st->buf_speech_enc_flt = NULL;
		#ifdef IVAS_FLOAT_FIXED
		st->buf_speech_enc = NULL;
		#endif
		#ifndef IVAS_FLOAT_FIXED
		@@ -1363,6 +1360,8 @@ ivas_error init_encoder_ivas_fx(
		pitch_ol_init_fx( &st->old_thres_fx, &st->old_pitch, &st->delta_pit, &st->old_corr_fx );
		set16_fx( st->old_wsp_fx, 0, L_WSP_MEM );
		set16_fx( st->old_wsp2_fx, 0, ( L_WSP_MEM - L_INTERPOL ) / OPL_DECIM ); // Needs to change depending on usage.
		st->Q_old_wsp2 = 0;
		move16();

		st->mem_preemph_fx = 0;
		move16();
		@@ -1370,16 +1369,12 @@ ivas_error init_encoder_ivas_fx(
		move16();
		st->mem_preemph_enc = 0;
		move16();
		st->exp_mem_preemph_enc = 0;
		move16();

		#if 1 // TODO: Float Initializations. To be removed later
		set_f( st->old_wsp, 0, L_WSP_MEM );
		set_f( st->old_wsp2, 0, ( L_WSP_MEM - L_INTERPOL ) / OPL_DECIM ); // Needs to change depending on usage.
		st->mem_preemph = 0.0f;
		st->mem_preemph16k = 0.0f;
		st->mem_preemph_enc_flt = 0.0;
		st->mem_preemp_preQ = 0.0f;
		st->mem_wsp = 0.0f;
		set_f( st->mem_decim2, 0, 3 );
		st->mem_preemph_DFT = 0.0f;
		set_f( st->inp_12k8_mem_stereo_sw, 0, STEREO_DFT_OVL_12k8 - L_MEM_RECALC_12K8 - L_FILT );
		st->mem_preemph16k_DFT = 0.0f;
		@@ -1583,6 +1578,8 @@ ivas_error init_encoder_ivas_fx(
		st->buf_wspeech_enc = st->hSignalBuf->buf_wspeech_enc;

		set16_fx( st->hSignalBuf->buf_speech_enc, 0, L_PAST_MAX_32k + L_FRAME32k + L_NEXT_MAX_32k );
		st->exp_buf_speech_enc = 0;
		move16();
		set16_fx( st->hSignalBuf->buf_wspeech_enc, 0, L_FRAME16k + L_SUBFR + L_FRAME16k + L_NEXT_MAX_16k + 320 );
		st->exp_buf_wspeech_enc = 0;
		move16();
		@@ -1624,20 +1621,17 @@ ivas_error init_encoder_ivas_fx(
		st->mem_decim16k = st->hSignalBuf->mem_decim16k_flt;
		st->old_inp_12k8 = st->hSignalBuf->old_inp_12k8_flt;
		st->old_inp_16k = st->hSignalBuf->old_inp_16k_flt;
		st->buf_speech_enc_pe_flt = st->hSignalBuf->buf_speech_enc_pe_flt;
		#ifndef IVAS_FLOAT_FIXED
		st->buf_synth_flt = st->hSignalBuf->buf_synth_flt;
		#endif
		st->buf_speech_enc_flt = st->hSignalBuf->buf_speech_enc_flt;
		#ifndef IVAS_FLOAT_FIXED
		st->buf_wspeech_enc_flt = st->hSignalBuf->buf_wspeech_enc_flt;
		#endif
		set_f( st->mem_decim, 0, 2 * L_FILT_MAX );
		set_f( st->mem_decim16k, 0, 2 * L_FILT_MAX );
		set_f( st->old_inp_12k8, 0, L_INP_MEM );
		set_f( st->old_inp_16k, 0, L_INP_MEM );
		set_f( st->hSignalBuf->buf_speech_enc_flt, 0, L_PAST_MAX_32k + L_FRAME32k + L_NEXT_MAX_32k );
		#ifndef IVAS_FLOAT_FIXED
		set_f( st->hSignalBuf->buf_speech_enc_flt, 0, L_PAST_MAX_32k + L_FRAME32k + L_NEXT_MAX_32k );
		set_f( st->hSignalBuf->buf_wspeech_enc_flt, 0, L_FRAME16k + L_SUBFR + L_FRAME16k + L_NEXT_MAX_16k + 320 );
		#endif
		#endif
		@@ -1660,12 +1654,9 @@ ivas_error init_encoder_ivas_fx(
		st->mem_decim16k = NULL;
		st->old_inp_12k8 = NULL;
		st->old_inp_16k = NULL;
		st->buf_speech_enc_pe_flt = NULL;
		#ifndef IVAS_FLOAT_FIXED
		st->buf_synth_flt = NULL;
		#endif
		st->buf_speech_enc_flt = NULL;
		#ifndef IVAS_FLOAT_FIXED
		st->buf_wspeech_enc_flt = NULL;
		#endif
		#endif
		@@ -2299,7 +2290,6 @@ ivas_error init_encoder_ivas_fx(
		#endif

		/* Initialize Signal Buffers */
		f2me_buf_16( st->buf_speech_enc_flt, st->buf_speech_enc, &st->exp_buf_speech_enc, st->L_frame );
		f2me_buf_16( st->old_wsp, st->old_wsp_fx, &st->exp_old_wsp, L_WSP_MEM );
		f2me_buf_16( st->old_inp_12k8, st->old_inp_12k8_fx, &st->exp_old_inp_12k8, L_INP_MEM );
		f2me_buf_16( st->old_inp_16k, st->old_inp_16k_fx, &st->exp_old_inp_16k, L_INP_MEM );