Correct scale of st->mem_wsp_enc (f25ca718) · Commits · SA4 / Audio / IVAS BASOP

lib_enc/core_enc_init_fx.c

+12 −0

Original line number	Diff line number	Diff line
		@@ -1355,6 +1355,10 @@ static void init_sig_buffers_ivas_fx( Encoder_State *st, const Word16 L_frame_ol
		Copy( st->old_wsp_fx, st->buf_wspeech_enc + st->L_frame + L_SUBFR - L_WSP_MEM, L_WSP_MEM );
		st->exp_buf_wspeech_enc = st->exp_old_wsp;
		move16();
		#ifdef NONBE_FIX_ISSUE_2206
		st->mem_wsp_enc = st->buf_wspeech_enc[st->L_frame + L_SUBFR - 1];
		move16();
		#endif

		/Resamp buffers needed only for ACELP/
		test();
		@@ -1397,8 +1401,10 @@ static void init_sig_buffers_ivas_fx( Encoder_State *st, const Word16 L_frame_ol
		move16();
		st->exp_mem_preemph_enc = st->exp_buf_speech_enc;
		move16();
		#ifndef NONBE_FIX_ISSUE_2206
		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();
		#endif
		}
		/coming from TCXonly modes/
		ELSE IF( !st->tcxonly && GE_32( last_total_brate, ACELP_32k ) )
		@@ -1407,6 +1413,10 @@ static void init_sig_buffers_ivas_fx( Encoder_State *st, const Word16 L_frame_ol
		Scale_sig( st->buf_wspeech_enc, L_FRAME16k + L_SUBFR + L_FRAME16k + L_NEXT_MAX_16k + 320, sub( 0, sub( st->exp_old_wsp, st->exp_buf_wspeech_enc ) ) );
		Copy( st->old_wsp_fx, st->buf_wspeech_enc + st->L_frame + L_SUBFR - L_WSP_MEM, L_WSP_MEM );
		// Copy_Scale_sig( st->old_wsp_fx, st->buf_wspeech_enc + st->L_frame + L_SUBFR - L_WSP_MEM, L_WSP_MEM, sub( st->prev_Q_new, st->prev_Q_old ) );
		#ifdef NONBE_FIX_ISSUE_2206
		st->mem_wsp_enc = st->buf_wspeech_enc[st->L_frame + L_SUBFR - 1];
		move16();
		#endif
		st->exp_buf_wspeech_enc = st->exp_old_wsp;
		move16(); /Resamp buffers needed only for ACELP/
		IF( EQ_16( st->L_frame, L_FRAME16k ) )
		@@ -1415,8 +1425,10 @@ static void init_sig_buffers_ivas_fx( Encoder_State *st, const Word16 L_frame_ol
		}
		hLPDmem->mem_w0 = 0;
		move16();
		#ifndef NONBE_FIX_ISSUE_2206
		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();
		#endif
		}
		}

lib_enc/ivas_core_pre_proc_front_fx.c

+1 −4

Original line number	Diff line number	Diff line
		@@ -1581,10 +1581,7 @@ ivas_error pre_proc_front_ivas_fx(
		*epsP_fx_q = add( Q_r[0], 1 );
		move16();

		#ifdef NONBE_FIX_ISSUE_2206
		// st->mem_wsp_enc = shl( st->mem_wsp_enc, sub( 0, st->mem_wsp_q ) ); // Q0
		// move16();
		#else
		#ifndef NONBE_FIX_ISSUE_2206
		st->mem_wsp_enc = shl( st->mem_wsp_enc, sub( 0, sub( Q15, st->exp_buf_wspeech_enc ) ) ); // Q0
		move16();
		#endif

lib_enc/ivas_core_pre_proc_fx.c

+13 −5

Original line number	Diff line number	Diff line
		@@ -704,6 +704,7 @@ ivas_error pre_proc_ivas_fx(
		tmp = s_min( tmp, norm_arr( st->buf_speech_enc_pe, st->L_frame ) );
		tmp = s_min( tmp, norm_arr( st->buf_speech_enc_pe + st->L_frame + L_FRAME, L_PAST_MAX_32k + L_FRAME32k + L_NEXT_MAX_32k - st->L_frame - L_FRAME ) );
		tmp = add( s_min( sub( *Q_new, 1 ), sub( Q15, st->exp_buf_speech_enc_pe ) ), tmp );
		tmp = s_min( tmp, sub( Q15, st->exp_buf_wspeech_enc ) ); // because of find_wsp()
		scale_sig( st->buf_speech_enc_pe, st->L_frame, sub( tmp, sub( Q15, st->exp_buf_speech_enc_pe ) ) );
		scale_sig( st->buf_speech_enc_pe + st->L_frame + L_FRAME, L_PAST_MAX_32k + L_FRAME32k + L_NEXT_MAX_32k - st->L_frame - L_FRAME, sub( tmp, sub( Q15, st->exp_buf_speech_enc_pe ) ) );
		Copy_Scale_sig_nosat( new_inp_12k8_fx, st->buf_speech_enc_pe + st->L_frame, L_FRAME, sub( tmp, sub( *Q_new, 1 ) ) );
		@@ -731,14 +732,14 @@ ivas_error pre_proc_ivas_fx(
		{
		#ifdef NONBE_FIX_ISSUE_2206
		// L_FRAME16k + L_SUBFR + L_FRAME16k + L_NEXT_MAX_16k + 320 );
		tmp = norm_arr( wsp_fx, L_FRAME + L_LOOK_12k8 );
		tmp = s_min( tmp, norm_arr( st->buf_wspeech_enc, (Word16) ( st->wspeech_enc - st->buf_wspeech_enc ) ) );
		tmp = s_min( tmp, norm_arr( st->wspeech_enc + L_FRAME + L_LOOK_12k8, /* L_NEXT_MAX_16k + */ 320 ) );
		tmp = add( s_min( sub( Q15, e_old_wsp ), sub( Q15, st->exp_buf_wspeech_enc ) ), tmp );
		// scale_sig( st->buf_wspeech_enc, L_FRAME16k + L_SUBFR + L_FRAME16k + L_NEXT_MAX_16k + 320, tmp );
		tmp = add( norm_arr( wsp_fx, L_FRAME + L_LOOK_12k8 ), sub( Q15, e_old_wsp ) );
		tmp = s_min( tmp, add( norm_s( st->mem_wsp_enc ), sub( Q15, st->exp_buf_wspeech_enc ) ) );
		tmp = s_min( tmp, add( norm_arr( st->buf_wspeech_enc, (Word16) ( st->wspeech_enc - st->buf_wspeech_enc ) ), sub( Q15, st->exp_buf_wspeech_enc ) ) );
		tmp = s_min( tmp, add( norm_arr( st->wspeech_enc + L_FRAME + L_LOOK_12k8, /* L_NEXT_MAX_16k + */ 320 ), sub( Q15, st->exp_buf_wspeech_enc ) ) );
		scale_sig( st->buf_wspeech_enc, (Word16) ( st->wspeech_enc - st->buf_wspeech_enc ), sub( tmp, sub( Q15, st->exp_buf_wspeech_enc ) ) );
		scale_sig( st->wspeech_enc + L_FRAME + L_LOOK_12k8, /* L_NEXT_MAX_16k + */ 320, sub( tmp, sub( Q15, st->exp_buf_wspeech_enc ) ) );
		Copy_Scale_sig_nosat( wsp_fx, st->wspeech_enc, L_FRAME + L_LOOK_12k8, sub( tmp, sub( Q15, e_old_wsp ) ) );
		st->mem_wsp_enc = shl( st->mem_wsp_enc, sub( st->exp_buf_wspeech_enc, sub( Q15, tmp ) ) );
		st->exp_buf_wspeech_enc = sub( Q15, tmp );
		move16();
		#else
		@@ -1467,6 +1468,7 @@ ivas_error ivas_compute_core_buffers_fx(
		tmp = s_min( tmp, norm_arr( st->buf_speech_enc_pe, L_FRAME16k - STEREO_DFT_OVL_16k ) );
		tmp = s_min( tmp, norm_arr( st->buf_speech_enc_pe + L_FRAME16k + L_FRAME16k, L_PAST_MAX_32k + L_NEXT_MAX_32k ) );
		tmp = add( s_min( sub( *Q_new, 1 ), sub( Q15, st->exp_buf_speech_enc_pe ) ), tmp );
		tmp = s_min( tmp, sub( Q15, st->exp_buf_wspeech_enc ) ); // because of find_wsp()
		scale_sig( st->buf_speech_enc_pe, L_FRAME16k - STEREO_DFT_OVL_16k, sub( tmp, sub( Q15, st->exp_buf_speech_enc_pe ) ) );
		scale_sig( st->buf_speech_enc_pe + L_FRAME16k + L_FRAME16k, L_PAST_MAX_32k + L_NEXT_MAX_32k, sub( tmp, sub( Q15, st->exp_buf_speech_enc_pe ) ) );
		Copy_Scale_sig_nosat( new_inp_16k_fx - STEREO_DFT_OVL_16k, st->buf_speech_enc_pe + L_FRAME16k - STEREO_DFT_OVL_16k, L_FRAME16k + STEREO_DFT_OVL_16k, sub( tmp, sub( *Q_new, 1 ) ) );
		@@ -1480,6 +1482,7 @@ ivas_error ivas_compute_core_buffers_fx(
		tmp = s_min( tmp, norm_arr( st->buf_speech_enc_pe, L_FRAME16k - lMemRecalc_16k - L_FILT16k ) );
		tmp = s_min( tmp, norm_arr( st->buf_speech_enc_pe + L_FRAME16k + L_FRAME16k, L_PAST_MAX_32k + L_NEXT_MAX_32k ) );
		tmp = add( s_min( sub( *Q_new, 1 ), sub( Q15, st->exp_buf_speech_enc_pe ) ), tmp );
		tmp = s_min( tmp, sub( Q15, st->exp_buf_wspeech_enc ) ); // because of find_wsp()
		scale_sig( st->buf_speech_enc_pe, L_FRAME16k - lMemRecalc_16k - L_FILT16k, sub( tmp, sub( Q15, st->exp_buf_speech_enc_pe ) ) );
		scale_sig( st->buf_speech_enc_pe + L_FRAME16k + L_FRAME16k, L_PAST_MAX_32k + L_NEXT_MAX_32k, sub( tmp, sub( Q15, st->exp_buf_speech_enc_pe ) ) );
		Copy_Scale_sig_nosat( new_inp_16k_fx - lMemRecalc_16k, st->buf_speech_enc_pe + L_FRAME16k - lMemRecalc_16k - L_FILT16k, L_FRAME16k + lMemRecalc_16k + L_FILT16k, sub( tmp, sub( *Q_new, 1 ) ) );
		@@ -1501,6 +1504,7 @@ ivas_error ivas_compute_core_buffers_fx(
		tmp = norm_arr( new_inp_16k_fx, L_FRAME16k );
		tmp = s_min( tmp, norm_arr( st->buf_speech_enc_pe, L_FRAME16k ) );
		tmp = add( s_min( sub( *Q_new, 1 ), sub( Q15, st->exp_buf_speech_enc_pe ) ), tmp );
		tmp = s_min( tmp, sub( Q15, st->exp_buf_wspeech_enc ) ); // because of find_wsp()
		scale_sig( st->buf_speech_enc_pe, L_FRAME16k, sub( tmp, sub( Q15, st->exp_buf_speech_enc_pe ) ) );
		Copy_Scale_sig_nosat( new_inp_16k_fx, st->buf_speech_enc_pe + L_FRAME16k, L_FRAME16k, sub( tmp, sub( *Q_new, 1 ) ) );
		st->exp_buf_speech_enc_pe = sub( Q15, tmp );
		@@ -1557,6 +1561,9 @@ ivas_error ivas_compute_core_buffers_fx(
		/--------------------------------------------------------------
		* Compute Weighted Input
		---------------------------------------------------------------/
		#ifdef NONBE_FIX_ISSUE_2206
		st->mem_wsp_enc = shl( st->mem_wsp_enc, sub( st->exp_buf_wspeech_enc, st->exp_buf_speech_enc_pe ) );
		#endif

		ivas_find_wsp_fx( L_FRAME16k, L_SUBFR, NB_SUBFR16k, A_fx, Aw_fx, st->speech_enc_pe, PREEMPH_FAC_16k, st->wspeech_enc, &st->mem_wsp_enc, st->gamma, L_LOOK_16k );
		#ifdef NONBE_FIX_ISSUE_2206
		@@ -1572,6 +1579,7 @@ ivas_error ivas_compute_core_buffers_fx(
		scale_sig( st->wspeech_enc, L_FRAME16k + L_LOOK_16k, sub( tmp, sub( Q15, st->exp_buf_speech_enc_pe ) ) );
		scale_sig( st->wspeech_enc + L_FRAME16k + L_LOOK_16k, L_FRAME16k + L_SUBFR + L_FRAME16k + L_NEXT_MAX_16k + 320 - ( (Word16) ( st->wspeech_enc - st->buf_wspeech_enc ) + L_FRAME16k + L_LOOK_16k ), sub( tmp, sub( Q15, st->exp_buf_wspeech_enc ) ) );
		st->exp_buf_wspeech_enc = sub( Q15, tmp );
		st->mem_wsp_enc = shl( st->mem_wsp_enc, sub( st->exp_buf_speech_enc_pe, st->exp_buf_wspeech_enc ) );
		move16();
		#endif
		}

lib_enc/ivas_tcx_core_enc_fx.c

+4 −7

Original line number	Diff line number	Diff line
		@@ -559,10 +559,8 @@ void stereo_tcx_core_enc(
		IF( st->tcxonly == 0 )
		{
		#ifdef NONBE_FIX_ISSUE_2206
		// Is this a mixup with st->mem_wsp_enc ?
		s = s_min( s, add( norm_arr( st->buf_wspeech_enc, L_FRAME16k + L_SUBFR + L_FRAME16k + L_NEXT_MAX_16k + 320 ), sub( sub( Q15, st->exp_buf_wspeech_enc ), Q_new ) ) );
		Scale_sig( st->buf_wspeech_enc, L_FRAME16k + L_SUBFR + L_FRAME16k + L_NEXT_MAX_16k + 320, sub( add( Q_new, s ), sub( Q15, st->exp_buf_wspeech_enc ) ) );
		st->exp_buf_wspeech_enc = sub( Q15, add( Q_new, s ) );
		s = s_min( s, add( norm_s( st->buf_wspeech_enc[st->L_frame - 1] ), sub( sub( Q15, st->exp_buf_wspeech_enc ), Q_new ) ) );
		st->wspeech_enc[st->L_frame - 1] = shl( st->wspeech_enc[st->L_frame - 1], sub( add( Q_new, s ), sub( Q15, st->exp_buf_wspeech_enc ) ) );
		#else
		s = s_min( s, norm_s( st->wspeech_enc[st->L_frame - 1] ) );
		st->wspeech_enc[st->L_frame - 1] = shl( st->wspeech_enc[st->L_frame - 1], s );
		@@ -720,9 +718,8 @@ void stereo_tcx_core_enc(
		IF( st->tcxonly == 0 )
		{
		#ifdef NONBE_FIX_ISSUE_2206
		// Is this a mixup with st->mem_wsp_enc ?
		Scale_sig( st->buf_wspeech_enc, L_FRAME16k + L_SUBFR + L_FRAME16k + L_NEXT_MAX_16k + 320, sub( 0, sub( Q15, st->exp_buf_wspeech_enc ) ) );
		st->exp_buf_wspeech_enc = sub( Q15, 0 );
		/* Revert scale change from above. */
		st->wspeech_enc[st->L_frame - 1] = shl( st->wspeech_enc[st->L_frame - 1], sub( sub( Q15, st->exp_buf_wspeech_enc ), Q_new ) );
		#else
		st->wspeech_enc[st->L_frame - 1] = shl( st->wspeech_enc[st->L_frame - 1], -st->Q_syn );
		move16();