Commit ac9660db authored by Sandesh Venkatesh's avatar Sandesh Venkatesh
Browse files

High MLD fixes for LTV, HP20 filter optimization, float code cleanup 

[x] HP20 filter recoding for WMOPS optimization
[x] Fix for LTV High MLD due to old_out in ivas_core_dec, decoder_tcx_ivas_fx
[x] Apa-Exec float code cleanup
[x] Fix for high mld issue for the case - [ltv-stereo at 13.2 kbps, 32kHz in, 32kHz out, DTX on]
parent 94e7768c

lib_com/hp50.c

+176 −0
Original line number Diff line number Diff line
@@ -436,6 +436,181 @@ void hp20( Word16 signal[], /* i/o: signal to filter any * 
}





#ifdef HP20_FIX32_RECODING

void hp20_fix32(

    Word32 signal_fx[],

    const Word16 lg,

    Word32 mem_fx[],

    const Word32 Fs )

{

    Word32 i;

    Word32 a1_fx, a2_fx, b1_fx, b2_fx;

    Word32 diff_pos, diff_neg;

#ifdef BASOP_NOGLOB

    Flag Overflow = 0;

#endif

    Word16 prescale, prescaleOld, prescale_current_frame, diff;



    prescale = getScaleFactor32( signal_fx, lg );

    prescale_current_frame = s_min( 3, sub( 1 + HP20_COEFF_SCALE, prescale ) );





    prescaleOld = extract_l( mem_fx[4] );



#ifdef BASOP_NOGLOB

    diff_pos = norm_l( L_shl_o( L_max( mem_fx[2], mem_fx[3] ), prescaleOld, &Overflow ) );

    diff_neg = norm_l( L_shl_o( L_min( mem_fx[2], mem_fx[3] ), prescaleOld, &Overflow ) );

#else

    diff_pos = norm_l( L_shl( L_max( mem_fx[2], mem_fx[3] ), prescaleOld ) );

    diff_neg = norm_l( L_shl( L_min( mem_fx[2], mem_fx[3] ), prescaleOld ) );

#endif



    diff = L_max( diff_pos, diff_neg );



    IF( NE_16( diff, 0 ) )

    {

        prescale = s_min( prescale, diff );

    }



    prescale = s_min( 3, sub( 1 + HP20_COEFF_SCALE, prescale ) );



    diff = sub( prescale, prescaleOld );

#ifdef BASOP_NOGLOB

    mem_fx[0] = L_shr_o( mem_fx[0], diff, &Overflow );

    move32();

    mem_fx[1] = L_shr_o( mem_fx[1], diff, &Overflow );

    move32();

    mem_fx[2] = L_shr_o( mem_fx[2], diff, &Overflow );

    move32();

    mem_fx[3] = L_shr_o( mem_fx[3], diff, &Overflow );

    move32();

#else

    mem_fx[0] = L_shr( mem_fx[0], diff );

    move32();

    mem_fx[1] = L_shr( mem_fx[1], diff );

    move32();

    mem_fx[2] = L_shr( mem_fx[2], diff );

    move32();

    mem_fx[3] = L_shr( mem_fx[3], diff );

    move32();

#endif

    mem_fx[4] = L_deposit_l( prescale_current_frame );

    move32();



    IF( EQ_32( Fs, 8000 ) )

    {

        /* hp filter 20Hz at 3dB for 8000KHz input sampling rate

           [b,a] = butter(2, 20.0/4000.0, 'high');

           b = [0.988954248067140  -1.977908496134280   0.988954248067140]

           a =[1.000000000000000  -1.977786483776764   0.978030508491796]*/

        a1_fx = 1061816033l /* 1.977786483776764 Q29*/;

        a2_fx = -525076131l /*-0.978030508491796 Q29*/;

        b1_fx = -1061881538l /*-1.977908496134280 Q29*/;

        b2_fx = 530940769l /* 0.988954248067140 Q29*/;

    }

    ELSE IF( EQ_32( Fs, 16000 ) )

    {

        /* hp filter 20Hz at 3dB for 16000KHz sampling rate

           [b,a] = butter(2, 20.0/8000.0, 'high');

           b =[ 0.994461788958195  -1.988923577916390   0.994461788958195]

           a =[1.000000000000000  -1.988892905899653   0.988954249933127] */

        a1_fx = 1067778748l /* 1.988892905899653 Q29*/;

        a2_fx = -530940770l /*-0.988954249933127 Q29*/;

        b1_fx = -1067795215l /*-1.988923577916390 Q29*/;

        b2_fx = 533897608l /* 0.994461788958195 Q29*/;

    }

    ELSE IF( EQ_32( Fs, 32000 ) )

    {

        /* hp filter 20Hz at 3dB for 32000KHz sampling rate

           [b,a] = butter(2, 20.0/16000.0, 'high');

           b =[0.997227049904470  -1.994454099808940   0.997227049904470]

           a =[1.000000000000000  -1.994446410541927   0.994461789075954]*/

        a1_fx = 1070760263l /* 1.994446410541927 Q29*/;

        a2_fx = -533897608l /*-0.994461789075954 Q29*/;

        b1_fx = -1070764392l /*-1.994454099808940 Q29*/;

        b2_fx = 535382196l /* 0.997227049904470 Q29*/;

    }

    ELSE

    {

        /* hp filter 20Hz at 3dB for 48000KHz sampling rate

           [b,a] = butter(2, 20.0/24000.0, 'high');

           b =[ 0.998150511190452  -1.996301022380904   0.998150511190452]

           a =[1.000000000000000  -1.996297601769122   0.996304442992686]*/

        a1_fx = 1071754114l /* 1.996297601769122 Q29*/;

        a2_fx = -534886875l /*-0.996304442992686 Q29*/;

        b1_fx = -1071755951l /*-1.996301022380904 Q29*/;

        b2_fx = 535877975l /* 0.998150511190452 Q29*/;

    }

    move32();

    move32();

    move32();

    move32();

    Word64 W_sum, W_y0, W_y1, W_y2;

    Word32 x0, x1, x2;



    W_sum = W_mult_32_32( b2_fx, mem_fx[2] );       /* b2*x2 */

    W_sum = W_mac_32_32( W_sum, b1_fx, mem_fx[3] ); /* b1*x1 */

    x2 = L_shr( signal_fx[0], prescale );

    W_sum = W_mac_32_32( W_sum, b2_fx, x2 );        /* b2*x0 */

    W_sum = W_mac_32_32( W_sum, mem_fx[0], a2_fx ); /* y2*a2 */

    W_sum = W_mac_32_32( W_sum, mem_fx[1], a1_fx ); /* y1*a1 */

    W_y2 = W_shl( W_sum, HP20_COEFF_SCALE );

    signal_fx[0] = W_extract_h( W_shl( W_y2, prescale ) );

    move32();



    W_sum = W_mult_32_32( b2_fx, mem_fx[3] ); /* b2*x2 */

    W_sum = W_mac_32_32( W_sum, b1_fx, x2 );  /* b1*x1 */

    x1 = L_shr( signal_fx[1], prescale );

    W_sum = W_mac_32_32( W_sum, b2_fx, x1 );                  /* b2*x0 */

    W_sum = W_mac_32_32( W_sum, mem_fx[1], a2_fx );           /* y2*a2 */

    W_sum = W_mac_32_32( W_sum, W_extract_h( W_y2 ), a1_fx ); /* y1*a1 */

    W_y1 = W_shl( W_sum, HP20_COEFF_SCALE );

    signal_fx[1] = W_extract_h( W_shl( W_y1, prescale ) );

    move32();



    diff = sub( prescale_current_frame, prescale );

    W_y1 = W_shr( W_y1, diff );

    W_y2 = W_shr( W_y2, diff );

    x2 = L_shr( x2, diff );

    x1 = L_shr( x1, diff );



    FOR( i = 2; i < lg; i++ )

    {

        W_sum = W_mult_32_32( b2_fx, x2 );       /* b2*x2 */

        W_sum = W_mac_32_32( W_sum, b1_fx, x1 ); /* b1*x1 */

        x0 = L_shr( signal_fx[i], prescale_current_frame );

        W_sum = W_mac_32_32( W_sum, b2_fx, x0 );                  /* b2*x0 */

        W_sum = W_mac_32_32( W_sum, W_extract_h( W_y2 ), a2_fx ); /* y2*a2 */

        W_sum = W_mac_32_32( W_sum, W_extract_h( W_y1 ), a1_fx ); /* y1*a1 */

        W_y0 = W_shl( W_sum, HP20_COEFF_SCALE );



        signal_fx[i] = W_extract_h( W_shl( W_y0, prescale_current_frame ) );

        move32();



        x2 = x1;

        x1 = x0;

        W_y2 = W_y1;

        W_y1 = W_y0;



        move32();

        move32();

        move64();

        move64();

    }



    mem_fx[0] = W_extract_h( W_y2 );

    mem_fx[1] = W_extract_h( W_y1 );

    mem_fx[2] = x2;

    mem_fx[3] = x1;



    move32();

    move32();

    move32();

    move32();



    return;

}

#else

void hp20_fix32(

    Word32 signal_fx[],

    const Word16 lg,
@@ -620,3 +795,4 @@ void hp20_fix32( 
    return;

}

#endif

#endif

lib_com/ivas_prot.h

+7 −0
Original line number Diff line number Diff line
@@ -1085,6 +1085,13 @@ void ivas_jbm_dec_feed_tc_to_renderer_fx( 
    Word32 *data                                                 /* i/o: transport channels/output synthesis signal                  */

);



void ivas_jbm_dec_feed_tc_to_renderer_fx(

    Decoder_Struct *st_ivas,                                    /* i/o: IVAS decoder structure                                      */

    const int16_t nSamplesForRendering,                         /* i  : number of TC samples available for rendering                */

    int16_t *nSamplesResidual,                                  /* o  : number of samples not fitting into the renderer grid and buffer for the next call*/

    Word32 *data                                                 /* i/o: transport channels/output synthesis signal                  */

);



ivas_error ivas_jbm_dec_set_discard_samples(

    Decoder_Struct *st_ivas                                     /* i/o: main IVAS decoder structre                                  */

);

lib_com/options.h

+1 −0
Original line number Diff line number Diff line
@@ -111,6 +111,7 @@ 
#define FIX_818_COMPLEXITY_IMPROV /* Fix for issue 818 */

#define FIX_805_VERIFICATION /* Fix for issue 805 */

#define FIX_827_HIGH_MLD /* Fix for issue 827 */

#define HP20_FIX32_RECODING /* Improved hp20 filtering code */

#endif

#define ISM_DISABLE

#define FIX_TMP_714

lib_dec/core_dec_init_fx.c

+0 −8
Original line number Diff line number Diff line
@@ -1454,12 +1454,6 @@ void open_decoder_LPD_ivas_fx( 
    IF( st->hTECDec != NULL )

    {

        /*constraint for adaptive bpf, otherwise parameter estimation and post-processing not time aligned*/

        if ( ( st->tcxonly == 0 ) )

        {

            // To be added later

            // assert(0 == (st->hTcxCfg->lfacnext > 0 ? st->hTcxCfg->lfacnext : 0));

        }



        resetTecDec_Fx( st->hTECDec );

    }
@@ -1624,8 +1618,6 @@ void open_decoder_LPD_ivas_fx( 


            IF( st->hTcxCfg != NULL )

            {

                hHQ_core->Q_old_wtda = -1; /*To be removed later when hHQ_core->Q_old_wtda vaue is updated*/

                move16();

                Copy_Scale_sig( hHQ_core->old_out_fx + NS2SA_FX2( st->output_Fs, N_ZERO_MDCT_NS ), hTcxDec->syn_OverlFB, st->hTcxCfg->tcx_mdct_window_lengthFB, negate( add( hHQ_core->Q_old_wtda, TCX_IMDCT_HEADROOM ) ) );

                *Q_syn_OverlFB = *Q_old_out;

                move16();

lib_dec/dec_tcx_fx.c

+8 −5
Original line number Diff line number Diff line
@@ -3826,9 +3826,12 @@ void decoder_tcx_ivas_fx( 
    /* Scaling down buffers for decoder_tcx_imdct_fx*/

    Scale_sig( st->hTcxDec->syn_Overl_TDACFB, L_FRAME_MAX / 2, 1 );                              // Scaling to Q_syn

    Scale_sig( st->hTcxDec->syn_Overl_TDAC, L_FRAME32k / 2, 1 );                                 // Scaling to Q_syn

    Scale_sig( st->hHQ_core->old_out_fx, L_FRAME48k, negate( st->hHQ_core->Q_old_out ) );    // Scaling to Q_syn

    Scale_sig( st->hHQ_core->old_out_LB_fx, L_FRAME32k, negate( st->hHQ_core->Q_old_out ) ); // Scaling to Q_syn

    st->hHQ_core->Q_old_out = 0;

    Scale_sig( st->hHQ_core->old_out_fx, L_FRAME48k, negate( st->hHQ_core->Q_old_wtda ) );       // Scaling to Q_syn

    Scale_sig( st->hHQ_core->old_out_LB_fx, L_FRAME32k, negate( st->hHQ_core->Q_old_wtda_LB ) ); // Scaling to Q_syn

    st->hHQ_core->Q_old_wtda = 0;

    move16();

    st->hHQ_core->Q_old_wtda_LB = 0;

    move16();



    Scale_sig( st->hTcxDec->old_syn_Overl, 320, 1 ); // Scaling to Q_syn

    Copy_Scale_sig_16_32_no_sat( st->old_Aq_12_8_fx, st->old_Aq_12_8_fx_32, M + 1, ( sub( 28, ( sub( 15, norm_s( sub( st->old_Aq_12_8_fx[0], 1 ) ) ) ) ) ) );