Commit bd041b6c authored by Sandesh Venkatesh's avatar Sandesh Venkatesh
Browse files

Merge branch 'MLD_improvements_8' into 'main' 

MLD improvements through precision changes, exponent usage correction and saturation fixes

See merge request !560
parents ce1ed6e6 844239a6

lib_dec/core_switching_dec_fx.c

+1 −3
Original line number Diff line number Diff line
@@ -2193,9 +2193,7 @@ ivas_error core_switching_post_dec_ivas_fx( 
            set16_fx( hBWE_TD->int_3_over_2_tbemem_dec_fx, 0, INTERP_3_2_MEM_LEN );

            set32_fx( hBWE_TD->int_3_over_2_tbemem_dec_fx_32, 0, INTERP_3_2_MEM_LEN );

        }

        ELSE IF( ( EQ_16( st_fx->extl, SWB_TBE ) || EQ_16( st_fx->extl, FB_TBE ) ) &&

                 ( NE_32( st_fx->last_total_brate, st_fx->total_brate ) || NE_16( st_fx->last_bwidth, st_fx->bwidth ) ||

                   NE_16( st_fx->last_codec_mode, MODE1 ) || NE_16( st_fx->rf_flag, st_fx->rf_flag_last ) ) )

        ELSE IF( ( EQ_16( st_fx->extl, SWB_TBE ) || EQ_16( st_fx->extl, FB_TBE ) ) && ( ( EQ_16( st_fx->element_mode, IVAS_CPE_TD ) && NE_16( st_fx->last_extl, SWB_TBE ) && NE_16( st_fx->last_extl, FB_TBE ) ) || ( NE_16( st_fx->element_mode, IVAS_CPE_TD ) && NE_32( st_fx->last_total_brate, st_fx->total_brate ) ) || NE_16( st_fx->last_bwidth, st_fx->bwidth ) || NE_16( st_fx->last_codec_mode, MODE1 ) || NE_16( st_fx->rf_flag, st_fx->rf_flag_last ) ) )

        {

            set16_fx( hBWE_TD->state_lpc_syn_fx, 0, LPC_SHB_ORDER );

            set16_fx( hBWE_TD->state_syn_shbexc_fx, 0, L_SHB_LAHEAD );

lib_dec/ivas_mc_param_dec.c

+2 −2
Original line number Diff line number Diff line
@@ -1578,7 +1578,7 @@ ivas_error ivas_param_mc_dec_reconfig_fx( 
                    v_add_fixed_me( tmp_buf_fx, cov_state_old.cy_old_e[parameter_band_mapping.source_band_idx[new_param_band_idx][source_param_idx]], hParamMC->h_output_synthesis_cov_state.cy_old_fx[new_param_band_idx], hParamMC->h_output_synthesis_cov_state.cy_old_e[new_param_band_idx], hParamMC->h_output_synthesis_cov_state.cy_old_fx[new_param_band_idx], &hParamMC->h_output_synthesis_cov_state.cy_old_e[new_param_band_idx], imult1616( nchan_out_cov, nchan_out_cov ), 0 );

                    /* mixing matrix*/

                    v_multc_fixed_16( cov_state_old.mixing_matrix_old_fx[parameter_band_mapping.source_band_idx[new_param_band_idx][source_param_idx]], parameter_band_mapping.source_band_factor_fx[new_param_band_idx][source_param_idx], tmp_buf_fx, imult1616( nchan_transport_old, nchan_out_cov ) );

                    v_add_fx( tmp_buf_fx, hParamMC->h_output_synthesis_cov_state.mixing_matrix_old_fx[new_param_band_idx], hParamMC->h_output_synthesis_cov_state.mixing_matrix_old_fx[new_param_band_idx], imult1616( nchan_transport_old, nchan_out_cov ) );

                    v_add_fixed_me( tmp_buf_fx, cov_state_old.mixing_matrix_old_exp[parameter_band_mapping.source_band_idx[new_param_band_idx][source_param_idx]], hParamMC->h_output_synthesis_cov_state.mixing_matrix_old_fx[new_param_band_idx], hParamMC->h_output_synthesis_cov_state.mixing_matrix_old_exp[new_param_band_idx], hParamMC->h_output_synthesis_cov_state.mixing_matrix_old_fx[new_param_band_idx], &hParamMC->h_output_synthesis_cov_state.mixing_matrix_old_exp[new_param_band_idx], imult1616( nchan_transport_old, nchan_out_cov ), 0 );

                }

            }

            FOR( new_param_band_idx = 0; new_param_band_idx < max_param_band_residual; new_param_band_idx++ )
@@ -1587,7 +1587,7 @@ ivas_error ivas_param_mc_dec_reconfig_fx( 
                {

                    /* residual mixing matrix*/

                    v_multc_fixed_16( cov_state_old.mixing_matrix_res_old_fx[parameter_band_mapping.source_band_idx[new_param_band_idx][source_param_idx]], parameter_band_mapping.source_band_factor_fx[new_param_band_idx][source_param_idx], tmp_buf_fx, imult1616( nchan_out_cov, nchan_out_cov ) );

                    v_add_fx( tmp_buf_fx, hParamMC->h_output_synthesis_cov_state.mixing_matrix_res_old_fx[new_param_band_idx], hParamMC->h_output_synthesis_cov_state.mixing_matrix_res_old_fx[new_param_band_idx], imult1616( nchan_out_cov, nchan_out_cov ) );

                    v_add_fixed_me( tmp_buf_fx, cov_state_old.mixing_matrix_res_old_exp[parameter_band_mapping.source_band_idx[new_param_band_idx][source_param_idx]], hParamMC->h_output_synthesis_cov_state.mixing_matrix_res_old_fx[new_param_band_idx], hParamMC->h_output_synthesis_cov_state.mixing_matrix_res_old_exp[new_param_band_idx], hParamMC->h_output_synthesis_cov_state.mixing_matrix_res_old_fx[new_param_band_idx], &hParamMC->h_output_synthesis_cov_state.mixing_matrix_res_old_exp[new_param_band_idx], imult1616( nchan_out_cov, nchan_out_cov ), 0 );

                }

            }

        }

lib_dec/ivas_tcx_core_dec.c

+1 −1
Original line number Diff line number Diff line
@@ -733,7 +733,7 @@ void stereo_tcx_core_dec_fx( 
            TonalMDCTConceal_SaveTimeSignal( st->hTonalMDCTConc, synthFB_fx, hTcxDec->L_frameTCX );

        }



        decoder_tcx_post_fx( st, synth_fx, synthFB_fx, Aq_fx, bfi );

        decoder_tcx_post_ivas_fx( st, synth_fx, synthFB_fx, Aq_fx, bfi, 0 );



        IF( EQ_16( st->core, TCX_20_CORE ) )

        {

lib_rend/ivas_dirac_decorr_dec.c

+37 −8
Original line number Diff line number Diff line
@@ -1425,16 +1425,45 @@ void ivas_dirac_dec_decorr_process_fx( 
         *-----------------------------------------------------------------*/



        q_shift = sub( q_input_frame, q_frame_f );

        Word16 q_if_local = 0; // shift adjust for input_frame_fx values

        IF( q_shift > 0 )

        {

            Word16 sf = MAX_16;

            FOR( ch_idx = 0; ch_idx < num_channels; ++ch_idx )

            {

                sf = s_min( sf, getScaleFactor32( &frame_dec_fx[shl( imult1616( ch_idx, num_freq_bands ), 1 )], shl( max_band_decorr, 1 ) ) );

            }

            sf = s_min( sub( sf, 1 ), q_shift );

            q_if_local = sub( q_shift, sf );

            q_shift = sf;

            move16();

            // scaling it to sf

#ifdef MSAN_FIX

            FOR( ch_idx = 0; ch_idx < num_channels; ++ch_idx )

            {

                scale_sig32( &frame_dec_fx[shl( imult1616( ch_idx, num_freq_bands ), 1 )], shl( max_band_decorr, 1 ), q_shift );

            }

#else

            Scale_sig32( frame_dec_fx, ( 2 * max_band_decorr + incr_aux ) * num_channels, q_shift ); // scaling it to input q

#endif

            q_frame_f = add( q_frame_f, sf );

        }

        ELSE IF( q_shift < 0 )

        {

            // scaling it to input q

#ifdef MSAN_FIX

            FOR( ch_idx = 0; ch_idx < num_channels; ++ch_idx )

            {

            Scale_sig32( &frame_dec_fx[shl( imult1616( ch_idx, num_freq_bands ), 1 )], shl( max_band_decorr, 1 ), q_shift );

                scale_sig32( &frame_dec_fx[shl( imult1616( ch_idx, num_freq_bands ), 1 )], shl( max_band_decorr, 1 ), q_shift );

            }

#else

            Scale_sig32( frame_dec_fx, ( 2 * max_band_decorr + incr_aux ) * num_channels, q_shift ); // scaling it to input q

#endif

            q_frame_f = q_input_frame;

            q_if_local = 0;

            move16();

            move16();

        }

        move16();



        IF( EQ_16( h_freq_domain_decorr_ap_params->add_back_onsets_on, 1 ) )
@@ -1445,8 +1474,8 @@ void ivas_dirac_dec_decorr_process_fx( 


                FOR( k = 0; k < max_band_decorr; ++k )

                {

                    aux_buffer_fx[2 * k] = Mpy_32_32( input_frame_fx[add( shl( offset, 1 ), shl( k, 1 ) )], L_sub( ONE_IN_Q31, onset_filter_fx[add( offset, k )] ) );

                    aux_buffer_fx[add( shl( k, 1 ), 1 )] = Mpy_32_32( input_frame_fx[add( add( shl( offset, 1 ), shl( k, 1 ) ), 1 )], L_sub( ONE_IN_Q31, onset_filter_fx[add( offset, k )] ) ); // q_input_f

                    aux_buffer_fx[2 * k] = Mpy_32_32( L_shr_r( input_frame_fx[add( shl( offset, 1 ), shl( k, 1 ) )], q_if_local ), L_sub( ONE_IN_Q31, onset_filter_fx[add( offset, k )] ) );

                    aux_buffer_fx[add( shl( k, 1 ), 1 )] = Mpy_32_32( L_shr_r( input_frame_fx[add( add( shl( offset, 1 ), shl( k, 1 ) ), 1 )], q_if_local ), L_sub( ONE_IN_Q31, onset_filter_fx[add( offset, k )] ) ); // q_frame_f

                    move32();

                    move32();

                }

lib_rend/ivas_dirac_rend.c

+48 −17
Original line number Diff line number Diff line
@@ -2857,6 +2857,9 @@ void protoSignalComputation2_fx( 
    Word32 tempSpaced_fx, tempDmx_fx;

    Word16 q_shift, min_q_shift, exp, q_temp, temp_q_shift, q_temp2;

    Word32 temp;

    Word64 W_tmp1, W_tmp2;

    Word64 reference_power_64fx[CLDFB_NO_CHANNELS_MAX];

    Word16 q_reference_power_64fx;

    /* Calculate maximum possible shift for the buffers RealBuffer_fx and ImagBuffer_fx */

    min_q_shift = Q31;

    move16();
@@ -2919,11 +2922,15 @@ void protoSignalComputation2_fx( 
            Real_aux_fx = L_add( re1, re2 );

            Imag_aux_fx = L_add( im1, im2 );



            Left_power_fx = Madd_32_32( Mpy_32_32( re1, re1 ), im1, im1 );

            Right_power_fx = Madd_32_32( Mpy_32_32( re2, re2 ), im2, im2 );

            // Left_power_fx = Madd_32_32( Mpy_32_32( re1, re1 ), im1, im1 );

            W_tmp1 = W_add( W_mult0_32_32( re1, re1 ), W_mult0_32_32( im1, im1 ) ); // 2*(q_cldfb+min_q_shift)

            // Right_power_fx = Madd_32_32( Mpy_32_32( re2, re2 ), im2, im2 );

            W_tmp2 = W_add( W_mult0_32_32( re2, re2 ), W_mult0_32_32( im2, im2 ) ); // 2*(q_cldfb+min_q_shift)



            reference_power_fx[l] = L_add( Left_power_fx, Right_power_fx );

            move32();



            // reference_power_fx[l] = L_add( Left_power_fx, Right_power_fx );

            reference_power_64fx[l] = W_add( W_tmp1, W_tmp2 ); // 2*(q_cldfb+min_q_shift)

            move64();



            temp = Madd_32_32( Mpy_32_32( Real_aux_fx, Real_aux_fx ), Imag_aux_fx, Imag_aux_fx );
@@ -3055,21 +3062,28 @@ void protoSignalComputation2_fx( 
            Imag_aux_fx = L_add( im1, im2 );



            /* Compute reference power */

            Left_power_fx = Madd_32_32( Mpy_32_32( re1, re1 ), im1, im1 );

            Right_power_fx = Madd_32_32( Mpy_32_32( re2, re2 ), im2, im2 );

            // Left_power_fx = Madd_32_32( Mpy_32_32( re1, re1 ), im1, im1 );

            W_tmp1 = W_add( W_mult0_32_32( re1, re1 ), W_mult0_32_32( im1, im1 ) ); // 2*(q_cldfb+min_q_shift)

            Left_power_fx = W_extract_l( W_shr( W_tmp1, 31 ) );

            // Right_power_fx = Madd_32_32( Mpy_32_32( re2, re2 ), im2, im2 );

            W_tmp2 = W_add( W_mult0_32_32( re2, re2 ), W_mult0_32_32( im2, im2 ) ); // 2*(q_cldfb+min_q_shift)

            Right_power_fx = W_extract_l( W_shr( W_tmp2, 31 ) );



            reference_power_fx[l] = L_add( Left_power_fx, Right_power_fx );

            move32();

            // reference_power_fx[l] = L_add( Left_power_fx, Right_power_fx );

            reference_power_64fx[l] = W_add( W_tmp1, W_tmp2 ); // 2*(q_cldfb+min_q_shift)

            move64();



            left_bb_power_fx = L_add( left_bb_power_fx, Left_power_fx );

            right_bb_power_fx = L_add( right_bb_power_fx, Right_power_fx );

            total_bb_power_fx = L_add( total_bb_power_fx, reference_power_fx[l] );

            // total_bb_power_fx = L_add( total_bb_power_fx, reference_power_fx[l] );

            total_bb_power_fx = L_add( total_bb_power_fx, W_extract_l( W_shr( reference_power_64fx[l], 31 ) ) );



            IF( GT_16( l, MASA_HI_FREQ_START_BIN ) )

            {

                left_hi_power_fx = L_add( left_hi_power_fx, Left_power_fx );

                right_hi_power_fx = L_add( right_hi_power_fx, Right_power_fx );

                total_hi_power_fx = L_add( total_hi_power_fx, reference_power_fx[l] );

                // total_hi_power_fx = L_add( total_hi_power_fx, reference_power_fx[l] );

                total_hi_power_fx = L_add( total_hi_power_fx, W_extract_l( W_shr( reference_power_64fx[l], 31 ) ) );

            }



            IF( LT_16( l, s_min( num_freq_bands, MASA_SUM_FREQ_RANGE_BINS ) ) )
@@ -3091,7 +3105,7 @@ void protoSignalComputation2_fx( 
                    move32();

                }



                temp = Mpy_32_32( a_fx, reference_power_fx[l] );

                temp = Mpy_32_32( a_fx, W_extract_l( W_shr( reference_power_64fx[l], 31 ) ) );

                IF( LT_16( q_temp, stereo_type_detect->q_total_power ) )

                {

                    stereo_type_detect->total_power_fx[l] = L_add( temp, L_shr( Mpy_32_32( b_fx, stereo_type_detect->total_power_fx[l] ), sub( stereo_type_detect->q_total_power, q_temp ) ) );
@@ -3552,17 +3566,18 @@ void protoSignalComputation2_fx( 
            Real_aux_fx = L_add( re1, re2 );

            Imag_aux_fx = L_add( im1, im2 );



            reference_power_fx[l] = Madd_32_32( Mpy_32_32( Real_aux_fx, Real_aux_fx ), Imag_aux_fx, Imag_aux_fx );

            move32();

            // reference_power_fx[l] = Madd_32_32( Mpy_32_32( Real_aux_fx, Real_aux_fx ), Imag_aux_fx, Imag_aux_fx );

            reference_power_64fx[l] = W_add( W_mult0_32_32( Real_aux_fx, Real_aux_fx ), W_mult0_32_32( Imag_aux_fx, Imag_aux_fx ) ); // 2*(q_cldfb+min_q_shift)

            move64();



            IF( LT_16( q_temp, *q_proto_power_smooth ) )

            {

                proto_power_smooth_fx[l] = L_add( L_shr( proto_power_smooth_fx[l], sub( *q_proto_power_smooth, q_temp ) ), reference_power_fx[l] );

                proto_power_smooth_fx[l] = L_add( L_shr( proto_power_smooth_fx[l], sub( *q_proto_power_smooth, q_temp ) ), W_extract_l( W_shr( reference_power_64fx[l], 31 ) ) );

                move32();

            }

            ELSE

            {

                proto_power_smooth_fx[l] = L_add( proto_power_smooth_fx[l], L_shr( reference_power_fx[l], sub( q_temp, *q_proto_power_smooth ) ) );

                proto_power_smooth_fx[l] = L_add( proto_power_smooth_fx[l], L_shr( W_extract_l( W_shr( reference_power_64fx[l], 31 ) ), sub( q_temp, *q_proto_power_smooth ) ) );

                move32();

            }
@@ -3603,14 +3618,30 @@ void protoSignalComputation2_fx( 
            move32();

        }

    }

    q_reference_power_64fx = shl( add( q_cldfb, min_q_shift ), 1 );

    Word16 norm_shift = 63;

    move16();

    FOR( l = 0; l < num_freq_bands; l++ )

    {

        IF( reference_power_64fx[l] )

        {

            norm_shift = s_min( norm_shift, W_norm( reference_power_64fx[l] ) );

        }

    }

    FOR( l = 0; l < num_freq_bands; l++ )

    {

        reference_power_fx[l] = W_extract_h( W_shl( reference_power_64fx[l], norm_shift ) );

        move32();

    }

    *q_reference_power = sub( add( q_reference_power_64fx, norm_shift ), 32 );

    move16();



    *q_proto_frame_f = add( q_cldfb, min_q_shift );

    move16();

    *q_proto_direct_buffer_f = add( q_cldfb, min_q_shift );

    move16();

    *q_proto_power_smooth = s_min( *q_proto_power_smooth, q_temp );

    move16();

    *q_reference_power = q_temp;

    move16();



    return;

}