Commit 167f8cf4 authored by Nicolas Roussin's avatar Nicolas Roussin
Browse files

Address Tommy's comments.

parent 3339c98f

lib_rend/ivas_dirac_dec_binaural_functions_fx.c

+54 −58
Original line number Diff line number Diff line
@@ -3519,11 +3519,12 @@ static void eig2x2_fx( 
    Word16 *q_D )

{

#ifdef NONBE_OPT_2193_EIG2X2

    Word32 s_fx[BINAURAL_CHANNELS];

    Word32 pm_fx, add_fx;

    Word32 tmp1, tmp2, e1, e2, c_re, c_im, c0_im, c1_im;

    Word32 condition, s0_fx, s1_fx, nval0_fx, nval1_fx;

    Word64 crossSquare_fx, tmp3, tmp4;

    Word16 q_crossSquare, q_min, q_diff, q_tmp1, q_tmp2, exp, q_e, q_c;

    Word16 nval0_q, nval1_q;

    Word32 i01, i00, i11, i10;

    Word64 eps_fx = ( (Word64) EPSILON_MANT ) << 32;

    Word16 eps_q = 63 - EPSILON_EXP;
@@ -3634,8 +3635,8 @@ static void eig2x2_fx( 


    // Numeric case, when input is practically zeros

    // if ( D_fx[0] < EPSILON_FX )



    if ( LT_32( L_shl_sat( D_fx[0], sub( 31 - EPSILON_EXP, *q_D ) ), EPSILON_MANT ) )

    condition = LT_32( L_shl_sat( D_fx[0], sub( 31 - EPSILON_EXP, *q_D ) ), EPSILON_MANT );

    if ( condition != 0 )

    {

        return;

    }
@@ -3646,8 +3647,8 @@ static void eig2x2_fx( 
    {

        tmp1 = L_shr( tmp1, q_diff );

    }



    if ( LT_32( tmp2, tmp1 ) )

    condition = LT_32( tmp2, tmp1 );

    if ( condition != 0 )

    {

        return;

    }
@@ -3681,84 +3682,79 @@ static void eig2x2_fx( 
        e2 = L_shl( e2, q_diff );

    }



    s_fx[0] = L_sub( tmp1, e1 ); // D_fx[0] - e1

    s0_fx = L_sub( tmp1, e1 ); // D_fx[0] - e1

    tmp1 = L_sub( tmp1, e2 );  // D_fx[0] - e2

    s_fx[1] = L_sub( tmp2, e1 ); // D_fx[1] - e1

    s1_fx = L_sub( tmp2, e1 ); // D_fx[1] - e1

    tmp2 = L_sub( tmp2, e2 );  // D_fx[1] - e2



    i01 = GT_32( L_abs( tmp1 ), L_abs( s_fx[0] ) ); // fabsf( D_fx[0] - e2 ) > fabsf( D_fx[0] - e1 )

    i11 = GT_32( L_abs( tmp2 ), L_abs( s_fx[1] ) ); // fabsf( D_fx[1] - e2 ) > fabsf( D_fx[1] - e1 )

    i01 = GT_32( L_abs( tmp1 ), L_abs( s0_fx ) ); // fabsf( D_fx[0] - e2 ) > fabsf( D_fx[0] - e1 )

    i11 = GT_32( L_abs( tmp2 ), L_abs( s1_fx ) ); // fabsf( D_fx[1] - e2 ) > fabsf( D_fx[1] - e1 )



    if ( i01 )

    {

        s_fx[0] = tmp1;

        s0_fx = tmp1;

        move32();

    }



    if ( i11 )

    {

        s_fx[1] = tmp2;

        s1_fx = tmp2;

        move32();

    }



    // normVal = sqrtf( 1.0f / ( 1e-12f + crossSquare + s * s ) );



    Word32 nvalm[BINAURAL_CHANNELS];

    Word16 nvalq[BINAURAL_CHANNELS];



    q_tmp2 = shl( q_tmp1, 1 );

    q_min = s_min( q_tmp2, q_crossSquare );

    q_min = s_min( q_min, eps_q );



    Word64 nval64m[BINAURAL_CHANNELS];



    q_diff = sub( q_tmp2, q_min );

    nval64m[0] = W_shr( W_mult0_32_32( s_fx[0], s_fx[0] ), q_diff );

    nval64m[1] = W_shr( W_mult0_32_32( s_fx[1], s_fx[1] ), q_diff );

    tmp3 = W_shr( W_mult0_32_32( s0_fx, s0_fx ), q_diff );

    tmp4 = W_shr( W_mult0_32_32( s1_fx, s1_fx ), q_diff );



    q_diff = sub( q_crossSquare, q_min );

    crossSquare_fx = W_shr( crossSquare_fx, q_diff );

    nval64m[0] = W_add( nval64m[0], crossSquare_fx );

    nval64m[1] = W_add( nval64m[1], crossSquare_fx );

    tmp3 = W_add( tmp3, crossSquare_fx );

    tmp4 = W_add( tmp4, crossSquare_fx );



    q_diff = sub( eps_q, q_min );

    eps_fx = W_shr( eps_fx, q_diff );

    nval64m[0] = W_add( nval64m[0], eps_fx );

    nval64m[1] = W_add( nval64m[1], eps_fx );

    tmp3 = W_add( tmp3, eps_fx );

    tmp4 = W_add( tmp4, eps_fx );



    q_diff = W_norm( nval64m[0] );

    nval64m[0] = W_shl( nval64m[0], q_diff );

    nvalq[0] = add( q_min, q_diff );

    q_diff = W_norm( tmp3 );

    tmp3 = W_shl( tmp3, q_diff );

    nval0_q = add( q_min, q_diff );



    q_diff = W_norm( nval64m[1] );

    nval64m[1] = W_shl( nval64m[1], q_diff );

    nvalq[1] = add( q_min, q_diff );

    q_diff = W_norm( tmp4 );

    tmp4 = W_shl( tmp4, q_diff );

    nval1_q = add( q_min, q_diff );



    // nvalm[0] = BASOP_Util_Divide3232_Scale_newton( ONE_IN_Q30, W_extract_h( nval64m[0] ), &exp );

    // exp = sub( exp, sub( 62, nvalq[0] ) );

    // nval0_fx = BASOP_Util_Divide3232_Scale_newton( ONE_IN_Q30, W_extract_h( tmp3 ), &exp );

    // exp = sub( exp, sub( 62, nval0_q ) );

    //

    // is equivalent to:

    //

    // nvalm[0] = div_w_newton( ONE_IN_Q30, W_extract_h( nval64m[0] ) );

    // exp = sub( nvalq[0], 61 );

    // nval0_fx = div_w_newton( ONE_IN_Q30, W_extract_h( tmp3 ) );

    // exp = sub( nval0_q, 61 );



    nvalm[0] = eig2x2_div_fx( ONE_IN_Q30, W_extract_h( nval64m[0] ) );

    exp = sub( nvalq[0], 61 );

    nvalm[0] = Sqrt32( nvalm[0], &exp );

    nvalq[0] = sub( 31, exp );

    nval0_fx = eig2x2_div_fx( ONE_IN_Q30, W_extract_h( tmp3 ) );

    exp = sub( nval0_q, 61 );

    nval0_fx = Sqrt32( nval0_fx, &exp );

    nval0_q = sub( 31, exp );



    // nvalm[1] = BASOP_Util_Divide3232_Scale_newton( ONE_IN_Q30, W_extract_h( nval64m[1] ), &exp );

    // exp = sub( exp, sub( 62, nvalq[1] ) );

    // nval1_fx = BASOP_Util_Divide3232_Scale_newton( ONE_IN_Q30, W_extract_h( tmp4 ), &exp );

    // exp = sub( exp, sub( 62, nval1_q ) );

    //

    // is equivalent to:

    //

    // nvalm[1] = div_w_newton( ONE_IN_Q30, W_extract_h( nval64m[1] ) );

    // exp = sub( nvalq[1], 61 );

    // nval1_fx = div_w_newton( ONE_IN_Q30, W_extract_h( tmp4 ) );

    // exp = sub( nval1_q, 61 );



    nvalm[1] = eig2x2_div_fx( ONE_IN_Q30, W_extract_h( nval64m[1] ) );

    exp = sub( nvalq[1], 61 );

    nvalm[1] = Sqrt32( nvalm[1], &exp );

    nvalq[1] = sub( 31, exp );

    nval1_fx = eig2x2_div_fx( ONE_IN_Q30, W_extract_h( tmp4 ) );

    exp = sub( nval1_q, 61 );

    nval1_fx = Sqrt32( nval1_fx, &exp );

    nval1_q = sub( 31, exp );



    q_diff = sub( q_c, q_tmp1 );

    q_tmp1 = s_min( q_tmp1, q_c );
@@ -3775,25 +3771,25 @@ static void eig2x2_fx( 


    if ( q_diff < 0 )

    {

        s_fx[0] = L_shl( s_fx[0], q_diff );

        s0_fx = L_shl( s0_fx, q_diff );

    }



    if ( q_diff < 0 )

    {

        s_fx[1] = L_shl( s_fx[1], q_diff );

        s1_fx = L_shl( s1_fx, q_diff );

    }



    q_diff = sub( nvalq[0], nvalq[1] );

    q_tmp2 = s_min( nvalq[0], nvalq[1] );

    q_diff = sub( nval0_q, nval1_q );

    q_tmp2 = s_min( nval0_q, nval1_q );



    if ( q_diff > 0 )

    {

        nvalm[0] = L_shr( nvalm[0], q_diff );

        nval0_fx = L_shr( nval0_fx, q_diff );

    }



    if ( q_diff < 0 )

    {

        nvalm[1] = L_shl( nvalm[1], q_diff );

        nval1_fx = L_shl( nval1_fx, q_diff );

    }



    *q_U = sub( add( q_tmp1, q_tmp2 ), 31 );
@@ -3815,18 +3811,18 @@ static void eig2x2_fx( 
        c1_im = L_negate( c1_im );

    }



    Ure_fx[i00][0] = Mpy_32_32( s_fx[0], nvalm[0] );

    Ure_fx[i00][0] = Mpy_32_32( s0_fx, nval0_fx );

    move32();

    Ure_fx[i01][0] = Mpy_32_32( c_re, nvalm[0] );

    Ure_fx[i01][0] = Mpy_32_32( c_re, nval0_fx );

    move32();

    Uim_fx[i01][0] = Mpy_32_32( c0_im, nvalm[0] );

    Uim_fx[i01][0] = Mpy_32_32( c0_im, nval0_fx );

    move32();



    Ure_fx[i10][1] = Mpy_32_32( s_fx[1], nvalm[1] );

    Ure_fx[i10][1] = Mpy_32_32( s1_fx, nval1_fx );

    move32();

    Ure_fx[i11][1] = Mpy_32_32( c_re, nvalm[1] );

    Ure_fx[i11][1] = Mpy_32_32( c_re, nval1_fx );

    move32();

    Uim_fx[i11][1] = Mpy_32_32( c1_im, nvalm[1] );

    Uim_fx[i11][1] = Mpy_32_32( c1_im, nval1_fx );

    move32();

#else

    Word16 chA, chB, ch;