Commit de4b0bd4 authored by Sandesh Venkatesh's avatar Sandesh Venkatesh
Browse files

Renderer updates, high MLD fixes 

[x] Renderer fixes for dirac dec path
[x] Fix for reverb signal scaling
[x] Stab_fac_fx calculation fix in updt_dec_common
[x] Updates in pca_tools for SBA high MLD cases.
parent 2fb72211

lib_com/ivas_pca_tools.c

+78 −42
Original line number Diff line number Diff line
@@ -560,6 +560,22 @@ void dquat2mat( 
    return;

}



static Word16 check_bound( Word32 tmp )

{

    IF( GT_32( tmp, MAX16B ) )

    {

        return MAX16B;

    }

    ELSE IF( LT_32( tmp, MIN16B ) )

    {

        return MIN16B;

    }

    ELSE

    {

        return extract_l( tmp );

    }

}



#ifdef IVAS_FLOAT_FIXED

void dquat2mat_fx(

    const Word16 *ql,
@@ -605,37 +621,37 @@ void dquat2mat_fx( 
    dx = mult( d, x );

    dy = mult( d, y );

    dz = mult( d, z );

    m[0] = sub( sub( aw, bx ), add( cy, dz ) );

    m[0] = check_bound( L_sub( L_sub( aw, bx ), L_add( cy, dz ) ) );

    move16();

    m[1] = sub( sub( cz, dy ), add( ax, bw ) );

    m[1] = check_bound( L_sub( L_sub( cz, dy ), L_add( ax, bw ) ) );

    move16();

    m[2] = add( sub( sub( negate( ay ), bz ), cw ), dx );

    m[2] = check_bound( L_add( L_sub( L_sub( negate( ay ), bz ), cw ), dx ) );

    move16();

    m[3] = sub( sub( by, az ), add( cx, dw ) );

    m[3] = check_bound( L_sub( L_sub( by, az ), L_add( cx, dw ) ) );

    move16();

    m[4] = add( sub( add( bw, ax ), dy ), cz );

    m[4] = check_bound( L_add( L_sub( L_add( bw, ax ), dy ), cz ) );

    move16();

    m[5] = add( add( add( negate( bx ), aw ), dz ), cy );

    m[5] = check_bound( L_add( L_add( L_add( negate( bx ), aw ), dz ), cy ) );

    move16();

    m[6] = sub( sub( add( negate( by ), az ), dw ), cx );

    m[6] = check_bound( L_sub( L_sub( L_add( negate( by ), az ), dw ), cx ) );

    move16();

    m[7] = add( sub( sub( negate( bz ), ay ), dx ), cw );

    m[7] = check_bound( L_add( L_sub( L_sub( negate( bz ), ay ), dx ), cw ) );

    move16();

    m[8] = sub( add( add( cw, dx ), ay ), bz );

    m[8] = check_bound( L_sub( L_add( L_add( cw, dx ), ay ), bz ) );

    move16();

    m[9] = sub( sub( add( negate( cx ), dw ), az ), by );

    m[9] = check_bound( L_sub( L_sub( L_add( negate( cx ), dw ), az ), by ) );

    move16();

    m[10] = add( add( add( negate( cy ), dz ), aw ), bx );

    m[10] = check_bound( L_add( L_add( L_add( negate( cy ), dz ), aw ), bx ) );

    move16();

    m[11] = sub( add( sub( negate( cz ), dy ), ax ), bw );

    m[11] = check_bound( L_sub( L_add( L_sub( negate( cz ), dy ), ax ), bw ) );

    move16();

    m[12] = add( add( sub( dw, cx ), by ), az );

    m[12] = check_bound( L_add( L_add( L_sub( dw, cx ), by ), az ) );

    move16();

    m[13] = add( sub( sub( negate( dx ), cw ), bz ), ay );

    m[13] = check_bound( L_add( L_sub( L_sub( negate( dx ), cw ), bz ), ay ) );

    move16();

    m[14] = sub( add( sub( negate( dy ), cz ), bw ), ax );

    m[14] = check_bound( L_sub( L_add( L_sub( negate( dy ), cz ), bw ), ax ) );

    move16();

    m[15] = add( add( add( negate( dz ), cy ), bx ), aw );

    m[15] = check_bound( L_add( L_add( L_add( negate( dz ), cy ), bx ), aw ) );

    move16();



    return;
@@ -828,11 +844,11 @@ static void norm_quat_fx( 
    move16();

    norm_q = dotp_fx( q, q, IVAS_PCA_INTERP, &exp1 );

    exp1 = sub( 31, add( exp1, 2 ) );

    norm_q = ISqrt32( norm_q, &exp1 ); /*q(15 - exp)*/

    norm_q = ISqrt32( norm_q, &exp1 ); /*Q(31 - exp)*/



    FOR( i = 0; i < IVAS_PCA_INTERP; i++ )

    {

        q[i] = mult( q[i], extract_l( L_shr( norm_q, 15 ) ) ); // todo : recheck

        q[i] = round_fx( L_shl( Mpy_32_16_1( norm_q, q[i] ), exp1 ) ); /* Q(15) */

        move16();

    }
@@ -922,7 +938,7 @@ void pca_interp_preproc_fx( 
{

    Word16 alpha;

    Word16 j;

    Word16 tmp, tmp2, tmp3;

    Word16 tmp, tmp2, tmp3, tmp_e;

    FOR( j = 0; j < len; j++ )

    {

        tmp = sub( len, 1 );
@@ -933,11 +949,12 @@ void pca_interp_preproc_fx( 
        }

        ELSE

        {

            alpha = idiv1616( j, tmp ); // the increment can be updated by simple delta //q15

            alpha = BASOP_Util_Divide1616_Scale( j, tmp, &tmp_e ); // the increment can be updated by simple delta

            alpha = shl_sat( alpha, tmp_e );                       /* Q15 */

        }

        tmp2 = mult( EVS_PI_FX, alpha );           // q13

        tmp3 = getCosWord16( tmp2 );               // q14

        alpha = shr( sub( ONE_IN_Q14, tmp3 ), 1 ); // q15

        tmp2 = mult( EVS_PI_FX, alpha );     /* Q13 */

        tmp3 = getCosWord16( tmp2 );         /* Q14 */

        alpha = sub_sat( ONE_IN_Q14, tmp3 ); /* Q15 */

        alpha = sub( MAX_16, alpha );

        quat_nlerp_preproc_fx( prev_ql, ql, alpha, &ql_interp[j * IVAS_PCA_INTERP] );

        quat_nlerp_preproc_fx( prev_qr, qr, alpha, &qr_interp[j * IVAS_PCA_INTERP] );
@@ -986,16 +1003,30 @@ static void sp2cart_fx( 
    Word16 *q )

{

    Word16 s1, s2, s1s2;

    s1 = getSinWord16( ph1 );                 // q15

    s2 = getSinWord16( ph2 );                 // q15

    s1s2 = mult( s1, s2 );                    // q15

    q[3] = mult( getSinWord16( ph3 ), s1s2 ); // q15

    Word16 sin_ph3, cos_ph3;



    sin_ph3 = cos_ph3 = ph3;

    move16();

    q[2] = mult( getCosWord16( ph3 ), s1s2 ); // q15

    move16();

    q[1] = mult( getCosWord16( ph2 ), s1 ); // q15



    IF( GT_16( ph3, 12868 /* PI in Q12 */ ) )

    {

        sin_ph3 = sub( 12868, ph3 ); /* sin(x) = sin(PI - x) */

        cos_ph3 = sub( 25736, ph3 ); /* cos(x) = cos(2*PI - x) */

    }

    sin_ph3 = shl( sin_ph3, 1 ); /* Q12 -> Q13 */

    cos_ph3 = shl( cos_ph3, 1 ); /* Q12 -> Q13 */



    s1 = getSinWord16( ph1 );                     /* Q15 */

    s2 = getSinWord16( ph2 );                     /* Q15 */

    s1s2 = mult( s1, s2 );                        /* Q15 */

    q[3] = mult( getSinWord16( sin_ph3 ), s1s2 ); /* Q15 */

    move16();

    q[2] = shl_sat( mult( getCosWord16( cos_ph3 ), s1s2 ), 1 ); /* Q15 */

    move16();

    q[0] = shl_sat( getCosWord16( ph1 ), 1 ); // q15

    q[1] = shl_sat( mult( getCosWord16( ph2 ), s1 ), 1 ); /* Q15 */

    move16();

    q[0] = shl_sat( getCosWord16( ph1 ), 1 ); /* Q15 */

    move16();



    return;
@@ -1021,9 +1052,9 @@ static Word16 calc_n2_fx( 
{

    Word16 n2;

    Word32 temp;

    temp = L_mult( 23040, getSinWord16( ph1 ) ); // q8

    n2 = round_fx( temp );

    n2 = shr( n2, 7 );

    temp = L_mult( 23040, getSinWord16( ph1 ) ); /* Q8 + Q15 + Q1 = Q24 */

    n2 = round_fx( temp );                       /* Q24 -> Q8 */

    n2 = shr( n2, 8 );

    IF( s_and( n2, 1 ) == 0 )

    {

        n2 = add( n2, 1 );
@@ -1061,10 +1092,10 @@ static Word16 calc_n3_fx( 
    const Word16 ph2 )

{

    Word16 n3;

    Word16 temp1 = mult( getSinWord16( ph2 ), getSinWord16( ph1 ) ); // q7 + q15 - q15

    n3 = round_fx( L_mult( temp1, getSinWord16( ph2 ) ) );           // q7 + q15 - q15

    Word16 temp1 = mult( getSinWord16( ph2 ), getSinWord16( ph1 ) ); /* Q15 */

    n3 = round_fx( L_mult( temp1, 23040 /* 180.0f in Q7 */ ) );      /* Q15 + Q7 + Q1 - Q16 -> Q7*/



    n3 = shr( n3, 8 );

    n3 = shr( n3, 7 );



    IF( n3 == 0 )

    {
@@ -1489,9 +1520,11 @@ void pca_dec_s3_fx( 
    num_fx = 12868;

    move16();



    d_fx = idiv1616( num_fx, n1 );     // Q12

    d_fx = idiv1616( num_fx, sub( n1, 1 ) ); // Q12

    ph1_q_fx = i_mult( index1, d_fx );       // Q12



    ph1_q_fx = shl( ph1_q_fx, 1 ); /* Q12 -> Q13 */



    n2 = calc_n2_fx( ph1_q_fx );



    j = L_sub( j, ivas_pca_offset_index1[index1] );
@@ -1521,6 +1554,7 @@ void pca_dec_s3_fx( 
        move16();

        d_fx = idiv1616( num_fx, sub( n2, 1 ) ); // Q12

        ph2_q_fx = i_mult( index2, d_fx );       // Q12

        ph2_q_fx = shl( ph2_q_fx, 1 );           /* Q12 -> Q13 */

    }



    j = L_sub( j, ivas_pca_offset_index2[add( index2, get_pca_offset_n2_fx( index1 ) )] );
@@ -1536,11 +1570,13 @@ void pca_dec_s3_fx( 
    }

    ELSE

    {

        num_fx = 6434;

        Word16 exp;

        num_fx = 25736; /* PI2 in Q12 */

        move16();

        d_fx = idiv1616( num_fx, n3 );                 // Q10

        ph3_q_fx = round_fx( L_mult( index3, d_fx ) ); // Q10

        // ph3_q_fx = shl(ph3_q_fx, 2);

        d_fx = BASOP_Util_Divide1616_Scale( num_fx, n3, &exp ); /* Q12 */

        exp = add( 3 - 15, exp );

        d_fx = shl( d_fx, sub( exp, 3 ) ); /* Q12 */

        ph3_q_fx = i_mult( index3, d_fx ); /* Q12 */

    }

    sp2cart_fx( ph1_q_fx, ph2_q_fx, ph3_q_fx, q_fx );

lib_dec/ivas_stereo_switching_dec.c

+2 −0
Original line number Diff line number Diff line
@@ -151,6 +151,8 @@ static ivas_error allocate_CoreCoder_TCX_fx( 
        // st->hTcxDec->CngLevelBackgroundTrace_bfi = 0; //initializing to avoid garbage overflow;

        set16_fx( st->hTcxDec->syn_OverlFB, 0, L_FRAME48k / 2 );

        set16_fx( st->hTcxDec->old_synth, 0, OLD_SYNTH_INTERNAL_DEC );

        st->hTcxDec->q_old_synth = 0;

        move16();

        set16_fx( st->hTcxDec->synth_history_fx, 0, L_PROT48k + L_FRAME48k );

    }

lib_dec/updt_dec_fx.c

+1 −1
Original line number Diff line number Diff line
@@ -1007,7 +1007,7 @@ void ivas_updt_dec_common_fx( 


        IF( EQ_16( st_fx->core, HQ_CORE ) )

        {

            st_fx->stab_fac_fx = extract_l( L_min( MAX_16, L_add( L_shr( STAB_FAC_EST1_FX, Q15 ), Madd_32_32( L_shr( Mpy_32_16_1( STAB_FAC_EST2_FX, st_fx->hHQ_core->mem_env_delta ), Q15 ), STAB_FAC_EST3_FX, st_fx->log_energy_diff_lt_fx ) ) ) );

            st_fx->stab_fac_fx = extract_l( L_min( MAX_16, L_add( L_shr( STAB_FAC_EST1_FX, Q15 ), Madd_32_32( Mpy_32_16_1( STAB_FAC_EST2_FX, st_fx->hHQ_core->mem_env_delta ), STAB_FAC_EST3_FX, st_fx->log_energy_diff_lt_fx ) ) ) );

            st_fx->stab_fac_fx = s_max( 0, st_fx->stab_fac_fx );

        }

        st_fx->stab_fac_smooth_lt_fx = extract_h( L_add( L_mult( ENV_SMOOTH_FAC_FX, st_fx->stab_fac_fx ),

lib_rend/ivas_dirac_dec_binaural_functions.c

+4 −3
Original line number Diff line number Diff line
@@ -7805,7 +7805,6 @@ static void ivas_masa_ext_rend_parambin_internal_fx( 


    /* CLDFB Analysis of input */



    Word16 q_cldfb = Q11;

    FOR( slot = 0; slot < hSpatParamRendCom->subframe_nbslots[subframe]; slot++ )

    {

        FOR( ch = 0; ch < numInChannels; ch++ )
@@ -7813,6 +7812,8 @@ static void ivas_masa_ext_rend_parambin_internal_fx( 
            test();

            IF( ch == 0 || EQ_16( nchan_transport, 2 ) )

            {

                Word16 q_cldfb = Q11;

                move16();

                cldfbAnalysis_ts_fx_fixed_q(

                    &( output_fx[ch][nBins * slot] ),

                    Cldfb_RealBuffer_in_fx[ch][slot],
@@ -7823,8 +7824,8 @@ static void ivas_masa_ext_rend_parambin_internal_fx( 
            {

                /* At mono input duplicate the channel to dual-mono, and apply gain

                correction to ensure same overall level as in stereo mode  */

                v_multc_fixed( Cldfb_RealBuffer_in_fx[0][slot], 151800249 /* INV_SQRT_2 in Q31 */, Cldfb_RealBuffer_in_fx[0][slot], nBins );

                v_multc_fixed( Cldfb_ImagBuffer_in_fx[0][slot], 151800249 /* INV_SQRT_2 in Q31 */, Cldfb_ImagBuffer_in_fx[0][slot], nBins );

                v_multc_fixed( Cldfb_RealBuffer_in_fx[0][slot], 1518500224 /* INV_SQRT_2 in Q31 */, Cldfb_RealBuffer_in_fx[0][slot], nBins );

                v_multc_fixed( Cldfb_ImagBuffer_in_fx[0][slot], 1518500224 /* INV_SQRT_2 in Q31 */, Cldfb_ImagBuffer_in_fx[0][slot], nBins );



                Copy32( Cldfb_RealBuffer_in_fx[0][slot], Cldfb_RealBuffer_in_fx[1][slot], nBins );

                Copy32( Cldfb_ImagBuffer_in_fx[0][slot], Cldfb_ImagBuffer_in_fx[1][slot], nBins );

lib_rend/ivas_dirac_output_synthesis_dec.c

+7 −2
Original line number Diff line number Diff line
@@ -3483,6 +3483,11 @@ void ivas_dirac_dec_output_synthesis_process_subframe_psd_ls_fx( 
    /*-----------------------------------------------------------------*

     * gain interpolation and output streams

     *-----------------------------------------------------------------*/

    Word16 q_align = sub( h_dirac_output_synthesis_state->proto_direct_buffer_f_q, h_dirac_output_synthesis_state->proto_power_diff_smooth_q );

    if ( h_dirac_output_synthesis_params->max_band_decorr != 0 )

    {

        q_align = sub( h_dirac_output_synthesis_state->proto_direct_buffer_f_q, h_dirac_output_synthesis_state->proto_diffuse_buffer_f_q );

    }



    FOR( buf_idx = 0; buf_idx < nbslots; ++buf_idx )

    {
@@ -3521,10 +3526,10 @@ void ivas_dirac_dec_output_synthesis_process_subframe_psd_ls_fx( 
            {

                g = L_add( Mpy_32_32( g1, *( p_gain_1++ ) ), Mpy_32_32( g2, *( p_gain_2++ ) ) ); // (Q31, gains_diff_prev_q) -> gains_diff_prev_q

                RealBuffer[k][buf_idx][l] = L_add( L_shr( RealBuffer[k][buf_idx][l], Q1 ),

                                                   L_shr( Mpy_32_32( g, ( *( p_power_smooth_diff++ ) ) ), Q1 ) ); // (gains_diff_prev_q, q_proto_direct_buffer) >> Q1 -> gains_diff_prev_q + q_proto_direct_buffer - 32

                                                   L_shr( Mpy_32_32( g, ( *( p_power_smooth_diff++ ) ) ), sub( Q1, q_align ) ) ); // (gains_diff_prev_q, q_proto_direct_buffer) >> Q1 -> gains_diff_prev_q + q_proto_direct_buffer - 32

                move32();

                ImagBuffer[k][buf_idx][l] = L_add( L_shr( ImagBuffer[k][buf_idx][l], Q1 ),

                                                   L_shr( Mpy_32_32( g, ( *( p_power_smooth_diff++ ) ) ), Q1 ) ); // (gains_diff_prev_q, q_proto_direct_buffer) >> Q1 -> gains_diff_prev_q + q_proto_direct_buffer - 32

                                                   L_shr( Mpy_32_32( g, ( *( p_power_smooth_diff++ ) ) ), sub( Q1, q_align ) ) ); // (gains_diff_prev_q, q_proto_direct_buffer) >> Q1 -> gains_diff_prev_q + q_proto_direct_buffer - 32

                move32();

            }