optimizations 14-04-25 (6a68711b) · Commits · SA4 / Audio / IVAS BASOP

lib_com/cldfb.c

+446 −44

File changed.

Preview size limit exceeded, changes collapsed.

lib_com/ivas_dirac_com_fx.c

+105 −2

Original line number	Diff line number	Diff line
		@@ -917,15 +917,16 @@ void computeDiffuseness_fixed(
		move32();
		Word32 *p_tmp;
		const Word32 *p_tmp_c;
		#ifndef OPT_SBA_DEC_PATH
		Word16 min_q_shift1, min_q_shift2, exp1, exp2, q_tmp;
		Word16 q_ene, q_intensity, q_intensity_slow;

		#endif /* OPT_SBA_DEC_PATH */
		/* Compute Intensity slow and energy slow buffer_intensity and buffer_energy */

		set_zero_fx( intensity_slow, i_mult( DIRAC_NUM_DIMS, CLDFB_NO_CHANNELS_MAX ) );
		set_zero_fx( intensity_slow_abs, CLDFB_NO_CHANNELS_MAX );
		set_zero_fx( energy_slow, CLDFB_NO_CHANNELS_MAX );

		#ifndef OPT_SBA_DEC_PATH
		/* Calculate max possible shift for the buffer buffer_energy and buffer_intensity */
		min_q_shift1 = Q31;
		move16();
		@@ -1064,6 +1065,108 @@ void computeDiffuseness_fixed(
		move32();
		}
		}
		#else /* OPT_SBA_DEC_PATH */
		Word16 gaurd_bits = find_guarded_bits_fx( DIRAC_NO_COL_AVG_DIFF );
		Word16 norm_arr = getScaleFactor32( buffer_energy, i_mult( DIRAC_NO_COL_AVG_DIFF, num_freq_bands ) );
		Word16 shift_ene = sub( norm_arr, gaurd_bits );
		Word16 q_ene = q_factor_energy[0]; // = add(*q_factor_energy, shift_ene);
		move16();
		norm_arr = 31;
		move16();

		FOR( i = 0; i < DIRAC_NUM_DIMS; i++ )
		{
		FOR( j = 0; j < DIRAC_NO_COL_AVG_DIFF; j++ )
		{
		norm_arr = s_min( norm_arr, getScaleFactor32( buffer_intensity[i][j], num_freq_bands ) );
		}
		}
		Word16 shift_inten = sub( norm_arr, gaurd_bits );
		Word16 q_inten = q_factor_intensity[0]; // = add(*q_factor_intensity, shift_inten);
		move16();

		FOR( i = 1; i < DIRAC_NO_COL_AVG_DIFF; ++i )
		{
		q_ene = s_min( q_ene, q_factor_energy[i] );
		q_inten = s_min( q_inten, q_factor_intensity[i] );
		}
		q_ene = add( q_ene, shift_ene );
		q_inten = add( q_inten, shift_inten );

		FOR( i = 0; i < DIRAC_NO_COL_AVG_DIFF; ++i )
		{
		/* Energy slow */
		p_tmp_c = buffer_energy + i * num_freq_bands;
		shift_ene = sub( q_ene, q_factor_energy[i] );
		shift_inten = sub( q_inten, q_factor_intensity[i] );
		FOR( k = 0; k < num_freq_bands; k++ )
		{
		energy_slow[k] = L_add( energy_slow[k], L_shl( *p_tmp_c, shift_ene ) );
		move32();
		p_tmp_c++;
		}

		/* Intensity slow */
		FOR( j = 0; j < DIRAC_NUM_DIMS; ++j )
		{
		p_tmp = buffer_intensity[j][i];

		FOR( k = 0; k < num_freq_bands; k++ )
		{
		intensity_slow[j * num_freq_bands + k] = L_add( intensity_slow[j * num_freq_bands + k], L_shl( *p_tmp, shift_inten ) );
		move32();
		p_tmp++;
		}
		}
		}

		gaurd_bits = shr( add( find_guarded_bits_fx( DIRAC_NUM_DIMS ), 1 ), 1 );
		norm_arr = getScaleFactor32( intensity_slow, i_mult( DIRAC_NUM_DIMS, num_freq_bands ) );

		Word16 shift = sub( norm_arr, gaurd_bits );
		scale_sig32( intensity_slow, i_mult( DIRAC_NUM_DIMS, num_freq_bands ), shift );
		q_inten = add( q_inten, shift );

		/* intensity_slow.^2 + intensity_slow_abs*/
		FOR( j = 0; j < DIRAC_NUM_DIMS; ++j )
		{
		p_tmp = intensity_slow + j * num_freq_bands;

		FOR( k = 0; k < num_freq_bands; k++ )
		{
		intensity_slow_abs[k] = Madd_32_32( intensity_slow_abs[k], p_tmp[k], p_tmp[k] );
		move32();
		}
		}

		Word16 q_inten_slow = sub( add( q_inten, q_inten ), 31 );


		/* Compute Diffuseness */
		Word16 exp1, exp2, q_tmp;
		Word16 diff = sub( 62, q_ene );
		p_tmp = intensity_slow_abs;
		FOR( i = 0; i < num_freq_bands; ++i )
		{

		exp1 = sub( 31, q_inten_slow );
		tmp = Sqrt32( p_tmp[i], &exp1 );

		tmp = BASOP_Util_Divide3232_Scale_cadence( tmp, L_add( energy_slow[i], EPSILLON_FX ), &exp2 );
		q_tmp = sub( diff, add( exp1, exp2 ) );
		// bring to Q30
		tmp = L_shl_sat( tmp, sub( Q30, q_tmp ) );
		diffuseness[i] = L_sub( ONE_IN_Q30, tmp );
		move32();

		if ( LT_32( diffuseness[i], 0 ) )
		{
		diffuseness[i] = 0;
		move32();
		}
		}

		#endif /* OPT_SBA_DEC_PATH */
		*q_diffuseness = Q30;
		move16();

lib_com/ivas_filters_fx.c

+14 −3

Original line number	Diff line number	Diff line
		@@ -223,7 +223,6 @@ void ivas_filter_process_exp_fx(
		return;
		}


		/-----------------------------------------------------------------------------------------
		* Function ivas_iir_2_filter()
		*
		@@ -263,13 +262,25 @@ static void ivas_iir_2_filter_fx(
		{
		L_tmp_prod = Mpy_32_32( filter_state->num_fx[stage][j], tmp_pIn_buf_i_fx ); // Q31-L_prod_e
		L_prod_e = add( filter_state->num_e[stage][j], tmp_pIn_buf_i_e );

		#if 1//ndef OPT_SBA_DEC_PATH
		L_tmp = BASOP_Util_Add_Mant32Exp( filter_state->state_fx[stage][j], filter_state->state_e[stage][j], L_tmp_prod, L_prod_e, &L_tmp_e ); // Q31 - L_tmp_e
		#else /* OPT_SBA_DEC_PATH */
		L_tmp_e = s_max( filter_state->state_e[stage][j], L_prod_e );
		L_tmp_e = add( L_tmp_e, 1 );
		L_tmp = L_add( L_shr( filter_state->state_fx[stage][j], sub( L_tmp_e, filter_state->state_e[stage][j] ) ), L_shr( L_tmp_prod, sub( L_tmp_e, L_prod_e ) ) );

		#endif /* OPT_SBA_DEC_PATH */

		L_tmp_prod = Mpy_32_32( filter_state->den_fx[stage][j], pOut_fx[i] ); // Q31 - ( pIn_Out_e[i]+filter_state->den_e[stage][j] )
		L_prod_e = add( pIn_Out_e[i], filter_state->den_e[stage][j] );

		#if 1//ndef OPT_SBA_DEC_PATH
		filter_state->state_fx[stage][j - 1] = BASOP_Util_Add_Mant32Exp( L_tmp, L_tmp_e, L_negate( L_tmp_prod ), L_prod_e, &filter_state->state_e[stage][j - 1] ); // Q31 - filter_state->state_e[stage][j - 1]
		#else /* OPT_SBA_DEC_PATH */
		Word16 com_e = s_max( L_tmp_e, L_prod_e );
		filter_state->state_e[stage][j - 1] = add( com_e, 1 );
		filter_state->state_fx[stage][j - 1] = L_sub( L_shr( L_tmp, sub( filter_state->state_e[stage][j - 1], L_tmp_e ) ), L_shr( L_tmp_prod, sub( filter_state->state_e[stage][j - 1], L_prod_e ) ) );
		move16();
		#endif /* OPT_SBA_DEC_PATH */
		move32();
		/In case when exponent is less than -31 the value is very small and negligible hence resetting it to zero to avoid exponent overflow/
		IF( LT_16( filter_state->state_e[stage][j - 1], -31 ) )

lib_com/ivas_prot_fx.h

+11 −5

Original line number	Diff line number	Diff line
		@@ -1094,8 +1094,11 @@ void ivas_mct_dec_mct_fx(
		void apply_MCT_dec_fx(
		MCT_DEC_HANDLE hMCT, /* i/o: MCT decoder structure */
		Decoder_State *sts, / i/o: decoder state structure */
		Word32 x[MCT_MAX_CHANNELS][NB_DIV], / i/o: decoded and dequan. spect. input to MCT */
		Word16 q_x[MCT_MAX_CHANNELS] );
		Word32 x[MCT_MAX_CHANNELS][NB_DIV] / i/o: decoded and dequan. spect. input to MCT */
		#ifndef OPT_SBA_DEC_PATH
		, Word16 q_x[MCT_MAX_CHANNELS]
		#endif /* OPT_SBA_DEC_PATH */
		);

		void mctStereoIGF_dec_fx(
		MCT_DEC_HANDLE hMCT, /* i/o: MCT decoder structure */
		@@ -1129,9 +1132,12 @@ void stereo_decoder_tcx_fx(
		const Word16 mct_on, /* i : flag mct block (1) or stereo (0) */
		const Word16 last_core_l, /* i : last core for left channel */
		const Word16 last_core_r, /* i : last core for right channel */
		const Word16 tmp_plc_upmix, /* i : indicates temp upmix for PLC decision */
		Word16 *q_x_ch2,
		Word16 *q_x_ch1 );
		const Word16 tmp_plc_upmix /* i : indicates temp upmix for PLC decision */
		#ifndef OPT_SBA_DEC_PATH
		,Word16 *q_x_ch2,
		Word16 *q_x_ch1
		#endif /* OPT_SBA_DEC_PATH */
		);

		void v_multc_acc_32_16(
		const Word32 x[], /* i : Input vector */

lib_com/ivas_sns_com_fx.c

+10 −3

Original line number	Diff line number	Diff line
		@@ -202,7 +202,14 @@ void sns_compute_scf_fx(
		-Q6 is for division with FDNS_NPTS and -Q1 is to reduce Q by one */
		mean = W_shl_sat_l( sum, -Q7 ); // q_out
		nf = Mpy_32_32( mean, 214748 /* powf( 10.0f, -4.0f ) in Q31 */ ); // q_out
		nf = L_max( nf, L_shl( 256, sub( q_out, 40 ) ) /* powf( 2.0f, -32.0f ) in Q40 */ ); // q_out

		IF( LE_32( nf, L_shl_sat( 256, sub( q_out, 40 ) ) ) ) /* powf( 2.0f, -32.0f ) in Q40 */
		{
		nf = 256;
		move32();
		q_out = 40;
		move16();
		}

		FOR( i = 0; i < FDNS_NPTS; i++ )
		{