Few bug fixes, performance optimizations (f9c4d5e1) · Commits · SA4 / Audio / IVAS BASOP

lib_com/ivas_dirac_com.c

+34 −17

Original line number	Diff line number	Diff line
		@@ -966,38 +966,57 @@ void computeDiffuseness_fixed(
		p_tmp_c = buffer_energy + i * num_freq_bands;

		q_tmp = add( q_factor_energy[i], min_q_shift1 );

		IF( LT_16( q_tmp, q_ene ) )
		{
		Word16 shift_q = sub( q_ene, q_tmp );
		FOR( k = 0; k < num_freq_bands; k++ )
		{
		tmp = L_shl( p_tmp_c[k], min_q_shift1 );
		IF( LT_16( q_tmp, q_ene ) )
		{
		energy_slow[k] = L_add( L_shr( energy_slow[k], sub( q_ene, q_tmp ) ), tmp );
		energy_slow[k] = L_add( L_shr( energy_slow[k], shift_q ), tmp );
		move32();
		}
		}
		ELSE
		{
		energy_slow[k] = L_add( energy_slow[k], L_shr( tmp, sub( q_tmp, q_ene ) ) );
		Word16 shift_q = sub( q_tmp, q_ene );
		FOR( k = 0; k < num_freq_bands; k++ )
		{
		tmp = L_shl( p_tmp_c[k], min_q_shift1 );
		energy_slow[k] = L_add( energy_slow[k], L_shr( tmp, shift_q ) );
		move32();
		}
		}

		q_ene = s_min( q_ene, q_tmp );

		/* Intensity slow */
		q_tmp = add( q_factor_intensity[i], min_q_shift2 );

		IF( LT_16( q_intensity, q_tmp ) )
		{
		Word16 shift_q = sub( q_tmp, q_intensity );
		FOR( j = 0; j < DIRAC_NUM_DIMS; ++j )
		{
		p_tmp = buffer_intensity[j][i];
		FOR( k = 0; k < num_freq_bands; k++ )
		{
		tmp = L_shl( p_tmp[k], min_q_shift2 );
		IF( LT_16( q_intensity, q_tmp ) )
		{
		intensity_slow[j * num_freq_bands + k] = L_add( intensity_slow[j * num_freq_bands + k], L_shr( tmp, sub( q_tmp, q_intensity ) ) );
		intensity_slow[j * num_freq_bands + k] = L_add( intensity_slow[j * num_freq_bands + k], L_shr( tmp, shift_q ) );
		move32();
		}
		}
		}
		ELSE
		{
		intensity_slow[j * num_freq_bands + k] = L_add( L_shr( intensity_slow[j * num_freq_bands + k], sub( q_intensity, q_tmp ) ), tmp );
		Word16 shift_q = sub( q_intensity, q_tmp );
		FOR( j = 0; j < DIRAC_NUM_DIMS; ++j )
		{
		p_tmp = buffer_intensity[j][i];
		FOR( k = 0; k < num_freq_bands; k++ )
		{
		tmp = L_shl( p_tmp[k], min_q_shift2 );
		intensity_slow[j * num_freq_bands + k] = L_add( L_shr( intensity_slow[j * num_freq_bands + k], shift_q ), tmp );
		move32();
		}
		}
		@@ -1017,9 +1036,7 @@ void computeDiffuseness_fixed(

		FOR( k = 0; k < num_freq_bands; k++ )
		{
		p_tmp[k] = Mpy_32_32( p_tmp[k], p_tmp[k] );
		move32();
		intensity_slow_abs[k] = L_add( intensity_slow_abs[k], p_tmp[k] );
		intensity_slow_abs[k] = Madd_32_32( intensity_slow_abs[k], p_tmp[k], p_tmp[k] );
		move32();
		}
		}

lib_enc/ivas_cpe_enc.c

+15 −4

Original line number	Diff line number	Diff line
		@@ -209,16 +209,27 @@ ivas_error ivas_cpe_enc_fx(
		Copy32( data_fx_ch0, sts[0]->input32_fx, input_frame ); // Q(q_data_fx)
		sts[0]->q_inp32 = q_data_fx;
		move16();
		Copy_Scale_sig32_16( sts[0]->input32_fx, sts[0]->input_fx, input_frame, sub( Q16, q_data_fx ) ); // Q(q_data_fx) -> Q0
		sts[0]->q_inp = 0;
		Word16 norm = L_norm_arr( sts[0]->input32_fx, input_frame );
		scale_sig32( sts[0]->input32_fx, input_frame, norm );
		sts[0]->q_inp32 = add( sts[0]->q_inp32, norm );
		move16();

		Copy_Scale_sig32_16( sts[0]->input32_fx, sts[0]->input_fx, input_frame, 0 );
		sts[0]->q_inp = sub( sts[0]->q_inp32, Q16 );
		move16();
		IF( data_fx_ch1 != NULL ) /this may happen for cases with odd number of channels/
		{
		Copy32( data_fx_ch1, sts[1]->input32_fx, input_frame ); // Q(q_data_fx)
		sts[1]->q_inp32 = q_data_fx;
		move16();
		Copy_Scale_sig32_16( sts[1]->input32_fx, sts[1]->input_fx, input_frame, sub( Q16, q_data_fx ) ); // Q(q_data_fx) -> Q0
		sts[1]->q_inp = 0;

		norm = L_norm_arr( sts[1]->input32_fx, input_frame );
		scale_sig32( sts[1]->input32_fx, input_frame, norm );
		sts[1]->q_inp32 = add( sts[1]->q_inp32, norm );
		move16();

		Copy_Scale_sig32_16( sts[1]->input32_fx, sts[1]->input_fx, input_frame, 0 );
		sts[1]->q_inp = sub( sts[1]->q_inp32, Q16 );
		move16();
		}

lib_enc/ivas_stereo_ica_enc.c

+10 −19

Original line number	Diff line number	Diff line
		@@ -67,7 +67,7 @@ static void unclr_calc_corr_features_fx(
		#define XH_BIAS_FX_Q15 13107
		#define XL_WIDTH_FX_Q15 3932
		#define XH_WIDTH_FX_Q15 4915
		#define SMOOTH_DIST_FACTOR_FX_Q15 14418
		#define SMOOTH_DIST_FACTOR_FX_Q15 13107
		#define A_BIAS_FX_Q15 3277
		#define B_BIAS_FX_Q15 -16384
		#define A_WIDTH_FX_Q31 32212264
		@@ -840,23 +840,13 @@ static void corrStatsEst_fx(
		scale_sig32( corrEst_fx, 2 * L_NCSHIFT_DS + 1, temp ); /* Q31-corrEst_exp */
		corrEst_exp = sub( corrEst_exp, temp );

		IF( GT_16( corrEst_exp, hStereoTCA->corrEstPrev_exp ) )
		{
		scale_sig32( hStereoTCA->corrEstPrev_fx[0], 2 * L_NCSHIFT_DS + 1, sub( hStereoTCA->corrEstPrev_exp, corrEst_exp ) ); /* Q31-hStereoTCA->corrEstPrev_exp */
		scale_sig32( hStereoTCA->corrEstPrev_fx[1], 2 * L_NCSHIFT_DS + 1, sub( hStereoTCA->corrEstPrev_exp, corrEst_exp ) ); /* Q31-hStereoTCA->corrEstPrev_exp */
		Copy32( corrEst_fx, hStereoTCA->corrEstPrev_fx[2], tempLen ); /* Q31-corrEst_exp */
		hStereoTCA->corrEstPrev_exp = corrEst_exp;
		move16();
		}
		ELSE
		{
		Copy32( corrEst_fx, hStereoTCA->corrEstPrev_fx[2], tempLen ); /* Q31-corrEst_exp */
		scale_sig32( hStereoTCA->corrEstPrev_fx[2], 2 * L_NCSHIFT_DS + 1, sub( corrEst_exp, hStereoTCA->corrEstPrev_exp ) ); /* Q31-corrEst_exp */
		}
		scale_sig32( hStereoTCA->corrEstPrev_fx[0], 2 * L_NCSHIFT_DS + 1, -1 ); /* Q31-hStereoTCA->corrEstPrev_exp-1 */
		scale_sig32( hStereoTCA->corrEstPrev_fx[1], 2 * L_NCSHIFT_DS + 1, -1 ); /* Q31-hStereoTCA->corrEstPrev_exp-1 */
		scale_sig32( hStereoTCA->corrEstPrev_fx[2], 2 * L_NCSHIFT_DS + 1, -1 ); /* Q31-hStereoTCA->corrEstPrev_exp-1 */
		hStereoTCA->corrEstPrev_exp = add( hStereoTCA->corrEstPrev_exp, 1 );
		Word16 gb = find_guarded_bits_fx( 2 * L_NCSHIFT_DS + 1 );

		scale_sig32( hStereoTCA->corrEstPrev_fx[2], 2 * L_NCSHIFT_DS + 1, -gb ); /* Q31-hStereoTCA->corrEstPrev_exp-1 */
		hStereoTCA->corrEstPrev_exp = add( hStereoTCA->corrEstPrev_exp, gb );
		move16();
		Word32 buf1_fx_temp[L_FRAME_DS];
		Word32 buf2_fx_temp[L_FRAME_DS];
		@@ -1032,7 +1022,8 @@ static void corrStatsEst_fx(
		loc_weight_win_fx[i] = win_bias_fx; // Q15
		move16();
		}

		reg_prv_corr_fx = L_shr( reg_prv_corr_fx, 1 );
		reg_prv_corr_exp = add( reg_prv_corr_exp, 1 );
		Word16 x = TRUNC_FX( reg_prv_corr_fx, reg_prv_corr_exp ); /* Q0 */
		move16();
		for ( i = 0, j = ( L_NCSHIFT_DS - x ); i < 2 * L_NCSHIFT_DS + 1; i++, j++ )

lib_rend/ivas_efap.c

+6 −56

Original line number	Diff line number	Diff line
		@@ -2516,64 +2516,14 @@ static Word16 in_tri_fx(
		move32();
		matInv[1][1] = Mpy_32_32( tmpDot1[0], invFactor ); // q=22+invFactor_exp
		move32();

		/* Computing S (Q13 + matInv_exp_final[i] + P_minus_A_exp_final + invFactor_exp - 1 ) =
		matInv (Q22 + matInv_exp_final[i] + invFactor_exp) (P-A) (Q22 + P_minus_A_exp_final) /
		Word16 matInv_exp[2][2], P_minus_A_exp[2];
		Word16 matInv_exp_final[2], P_minus_A_exp_final;
		FOR( Word32 i = 0; i < 2; i++ )
		{
		FOR( Word32 j = 0; j < 2; j++ )
		{
		matInv_exp[i][j] = 31;
		move16();
		IF( matInv[i][j] != 0 )
		{
		matInv_exp[i][j] = norm_l( matInv[i][j] );
		move16();
		}
		}
		matInv_exp_final[i] = s_min( matInv_exp[i][0], matInv_exp[i][1] );
		move16();
		P_minus_A_exp[i] = 31;
		move16();
		IF( P_minus_A[i] != 0 )
		{
		P_minus_A_exp[i] = norm_l( P_minus_A[i] );
		move16();
		}
		}
		P_minus_A_exp_final = s_min( P_minus_A_exp[0], P_minus_A_exp[1] );

		S[0] = L_add( L_shr( Mpy_32_32( L_shl( matInv[0][0], matInv_exp_final[0] ), L_shl( P_minus_A[0], P_minus_A_exp_final ) ), Q1 ),
		L_shr( Mpy_32_32( L_shl( matInv[0][1], matInv_exp_final[0] ), L_shl( P_minus_A[1], P_minus_A_exp_final ) ), Q1 ) ); //(22+invFactor_exp+matInv_exp_final[0]+22+P_minus_A_exp_final-1)-31=>12+invFactor_exp+matInv_exp_final[0]+P_minus_A_exp_final
		move64();
		S[1] = L_add( L_shr( Mpy_32_32( L_shl( matInv[1][0], matInv_exp_final[1] ), L_shl( P_minus_A[0], P_minus_A_exp_final ) ), Q1 ),
		L_shr( Mpy_32_32( L_shl( matInv[1][1], matInv_exp_final[1] ), L_shl( P_minus_A[1], P_minus_A_exp_final ) ), Q1 ) ); //(22+invFactor_exp+matInv_exp_final[1]+22+P_minus_A_exp_final-1)-31=>12+invFactor_exp+matInv_exp_final[0]+P_minus_A_exp_final
		S[0] = W_add( W_mult_32_32( matInv[0][0], P_minus_A[0] ), W_mult_32_32( matInv[0][1], P_minus_A[1] ) ); // Q22+invFactor_exp +Q22
		move64();

		/* Checking if we are in the triangle; For the theory, check Christian Borss article, section 3.2 */
		// Q32 S
		IF( sub( sub( sub( Q20, matInv_exp_final[0] ), P_minus_A_exp_final ), invFactor_exp ) < 0 )
		{
		S[0] = W_shr( S[0], sub( add( add( matInv_exp_final[0], P_minus_A_exp_final ), invFactor_exp ), Q20 ) ); // q32
		move64();
		}
		ELSE
		{
		S[0] = W_shl( S[0], sub( sub( sub( Q20, matInv_exp_final[0] ), P_minus_A_exp_final ), invFactor_exp ) ); // q32
		S[0] = W_shr( S[0], add( 13, invFactor_exp ) ); // q32
		move64();
		}
		IF( sub( sub( sub( Q20, matInv_exp_final[1] ), P_minus_A_exp_final ), invFactor_exp ) < 0 )
		{
		S[1] = W_shr( S[1], sub( add( add( matInv_exp_final[1], P_minus_A_exp_final ), invFactor_exp ), Q20 ) ); // q32
		S[1] = W_add( W_mult_32_32( matInv[1][0], P_minus_A[0] ), W_mult_32_32( matInv[1][1], P_minus_A[1] ) );
		move64();
		}
		ELSE
		{
		S[1] = W_shl( S[1], sub( sub( sub( Q20, matInv_exp_final[1] ), P_minus_A_exp_final ), invFactor_exp ) ); // q32
		S[1] = W_shr( S[1], add( 13, invFactor_exp ) ); // q32
		move64();
		}

		test();
		test();