Commit bfde9510 authored by ber's avatar ber
Browse files

Some more fixes: FIX_1072_SPEEDUP_matrixTransp2Mul_fx_scaleSwitch,... 

Some more fixes: FIX_1072_SPEEDUP_matrixTransp2Mul_fx_scaleSwitch, FIX_1072_SPEEDUP_matrixMul_fx, FIX_1072_SPEEDUP_matrixTransp2Mul_fx_reducetest, FIX_1072_SPEEDUP_matrixTransp2Mul_fx_reducetest_reduceLabs - all inactive, need to be tested more
parent 11307a29

lib_rend/ivas_dirac_dec_binaural_functions.c

+518 −22
Original line number Diff line number Diff line
@@ -47,7 +47,11 @@ 


#include "wmc_auto.h"



#define FIX_1072_REPLACE_DIV_SQRT_BY_ISQRT

//#define FIX_1072_SPEEDUP_matrixTransp2Mul_fx_scaleSwitch

//#define FIX_1072_SPEEDUP_matrixMul_fx

//#define FIX_1072_SPEEDUP_matrixTransp2Mul_fx_reducetest

//#define FIX_1072_SPEEDUP_matrixTransp2Mul_fx_reducetest_reduceLabs

//#define FIX_1072_REPLACE_DIV_SQRT_BY_ISQRT



#ifndef FIX_1072_REPLACE_DIV_SQRT_BY_ISQRT

//#define FIX_1072_SET_DIV3232_RETURN_VAL /*only meaningful if FIX_1072_REPLACE_DIV_SQRT_BY_ISQRT is not defined*/
@@ -134,7 +138,20 @@ static void formulate2x2MixingMatrix_fx( Word32 Ein1_fx /*q_Ein*/, Word32 Ein2_f 


static void matrixMul_fx( Word32 Are[BINAURAL_CHANNELS][BINAURAL_CHANNELS] /*q_A*/, Word32 Aim[BINAURAL_CHANNELS][BINAURAL_CHANNELS] /*q_A*/, Word16 *q_A, Word32 Bre[BINAURAL_CHANNELS][BINAURAL_CHANNELS] /*q_B*/, Word32 Bim[BINAURAL_CHANNELS][BINAURAL_CHANNELS] /*q_B*/, Word16 *q_B, Word32 outRe[BINAURAL_CHANNELS][BINAURAL_CHANNELS] /*q_out*/, Word32 outIm[BINAURAL_CHANNELS][BINAURAL_CHANNELS] /*q_out*/, Word16 *q_out );



static void matrixTransp2Mul_fx( Word32 Are[BINAURAL_CHANNELS][BINAURAL_CHANNELS] /*q_A*/, Word32 Aim[BINAURAL_CHANNELS][BINAURAL_CHANNELS] /*q_A*/, Word16 *q_A, Word32 Bre[BINAURAL_CHANNELS][BINAURAL_CHANNELS] /*q_B*/, Word32 Bim[BINAURAL_CHANNELS][BINAURAL_CHANNELS] /*q_B*/, Word16 *q_B, Word32 outRe[BINAURAL_CHANNELS][BINAURAL_CHANNELS] /*q_out*/, Word32 outIm[BINAURAL_CHANNELS][BINAURAL_CHANNELS] /*q_out*/, Word16 *q_out );

static void matrixTransp2Mul_fx( 

  Word32 Are[BINAURAL_CHANNELS][BINAURAL_CHANNELS] /*q_A*/, 

  Word32 Aim[BINAURAL_CHANNELS][BINAURAL_CHANNELS] /*q_A*/, 

  Word16 *q_A, Word32 Bre[BINAURAL_CHANNELS][BINAURAL_CHANNELS] /*q_B*/, 

  Word32 Bim[BINAURAL_CHANNELS][BINAURAL_CHANNELS] /*q_B*/, 

  Word16 *q_B, 

#ifdef FIX_1072_SPEEDUP_matrixTransp2Mul_fx_scaleSwitch

    int Ascale,

    int Bscale,

#endif

  Word32 outRe[BINAURAL_CHANNELS][BINAURAL_CHANNELS] /*q_out*/, 

  Word32 outIm[BINAURAL_CHANNELS][BINAURAL_CHANNELS] /*q_out*/, 

  Word16 *q_out 

);



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

 * ivas_dirac_dec_init_binaural_data()
@@ -1961,7 +1978,13 @@ static void ivas_dirac_dec_binaural_determine_processing_matrices_fx( 


        /* Make matrix multiplication M*Cx*M' to determine resulting covariance matrix of processing input with M */

        matrixMul_fx( Mre_fx, Mim_fx, &q_M, CxRe_fx, CxIm_fx, &q_Cx, tmpMtxRe_fx, tmpMtxIm_fx, &q_tmp );

        matrixTransp2Mul_fx( tmpMtxRe_fx, tmpMtxIm_fx, &q_tmp, Mre_fx, Mim_fx, &q_M, resultMtxRe_fx, resultMtxIm_fx, &q_res );

        matrixTransp2Mul_fx( 

          tmpMtxRe_fx, tmpMtxIm_fx, &q_tmp, Mre_fx, Mim_fx, &q_M, 

#ifdef FIX_1072_SPEEDUP_matrixTransp2Mul_fx_scaleSwitch

            1 /*int Ascale*/,

            0 /*int Bscale*/,

#endif

            resultMtxRe_fx, resultMtxIm_fx, &q_res );



        /* When below the frequency limit where decorrelation is applied, we inject the decorrelated

         * residual (or missing) signal component. The procedure is active when there are not enough independent
@@ -2071,7 +2094,7 @@ static void ivas_dirac_dec_binaural_determine_processing_matrices_fx( 
                                         Q31, CrEneL_fx, CrEneR_fx, q_CrEne,

                                         CrCrossRe_fx, CrCrossIm_fx, q_CrCross,

                                         prototypeMtx_fx, MdecRe_fx, MdecIm_fx, &q_Mdec, 3277 ); // 3277 = 0.2 in Q14

            pop_wmops();

            pop_wmops(); /*IDDB_detProcMat_bigLoop1_bigBranch1*/

        }

        ELSE

        {
@@ -2298,10 +2321,10 @@ static void ivas_dirac_dec_binaural_determine_processing_matrices_fx( 
                    move16();

                }

            }

            pop_wmops();

            pop_wmops(); /*IDDB_detProcMat_bigLoop1_bigBranch2*/

        }

    }

    pop_wmops();

    pop_wmops(); /*IDDB_detProcMat_bigLoop1*/



    /* Aligning Q-factors of all bins in the processing matrices to a common Q-factor */

    minimum_s( q_processMtx, nBins, &hDiracDecBin->q_processMtx );
@@ -2322,7 +2345,6 @@ static void ivas_dirac_dec_binaural_determine_processing_matrices_fx( 
    minimum_s( q_processMtxDec, nBins, &hDiracDecBin->q_processMtxDec );

    minimum_s( q_processMtxDecPrev, nBins, &hDiracDecBin->q_processMtxDecPrev );



    push_wmops( "IDDB_detProcMat_bigLoop2" );

    FOR( bin = 0; bin < nBins; bin++ )

    {

        FOR( chA = 0; chA < BINAURAL_CHANNELS; chA++ )
@@ -2362,7 +2384,6 @@ static void ivas_dirac_dec_binaural_determine_processing_matrices_fx( 
            }

        }

    }

    pop_wmops();



    return;

}
@@ -3522,6 +3543,243 @@ static void matrixMul_fx( 
            move32();

            outIm_fx[chA][chB] = L_add( outIm_fx[chA][chB], W_extract_h( W_mac_32_32( W_mult_32_32( Are_fx[chA][0], Bim_fx[0][chB] ), Are_fx[chA][1], Bim_fx[1][chB] ) ) );

            move32();

#else

#ifdef FIX_1072_SPEEDUP_matrixMul_fx

            /* Create testVariables*/

            Word32 Are_fx_chA_0_GE_0 = GE_32( Are_fx[chA][0], 0 );

            Word32 Bre_fx_0_chB_GE_0 = GE_32( Bre_fx[0][chB], 0 );

            Word32 Are_fx_chA_0_LT_0 = LT_32( Are_fx[chA][0], 0 );

            Word32 Bre_fx_0_chB_LT_0 = LT_32( Bre_fx[0][chB], 0 );

            Word32 Are_fx_chA_1_GE_0 = GE_32( Are_fx[chA][1], 0 );

            Word32 Bre_fx_1_chB_GE_0 = GE_32( Bre_fx[1][chB], 0 );

            Word32 Are_fx_chA_1_LT_0 = LT_32( Are_fx[chA][1], 0 );

            Word32 Bre_fx_1_chB_LT_0 = LT_32( Bre_fx[1][chB], 0 );

            Word32 Aim_fx_chA_0_GE_0 = GE_32( Aim_fx[chA][0], 0 );

            Word32 Bim_fx_0_chB_GE_0 = GE_32( Bim_fx[0][chB], 0 );

            Word32 Aim_fx_chA_0_LT_0 = LT_32( Aim_fx[chA][0], 0 );

            Word32 Bim_fx_0_chB_LT_0 = LT_32( Bim_fx[0][chB], 0 );

            Word32 Aim_fx_chA_1_GE_0 = GE_32( Aim_fx[chA][1], 0 );

            Word32 Bim_fx_1_chB_GE_0 = GE_32( Bim_fx[1][chB], 0 );

            Word32 Aim_fx_chA_1_LT_0 = LT_32( Aim_fx[chA][1], 0 );

            Word32 Bim_fx_1_chB_LT_0 = LT_32( Bim_fx[1][chB], 0 );

            Word32 cond1, cond2, cond3;



#if 1

            tmp1 = Mpy_32_32( Are_fx[chA][0], Bre_fx[0][chB]) ;

            cond2 = Are_fx_chA_0_GE_0 || Bre_fx_0_chB_GE_0;

            move32();

            cond1 = Are_fx_chA_0_LT_0 || Bre_fx_0_chB_LT_0;

            move32();

            cond3 = cond1 && cond2;

            move32();



            if ( cond3 )

                tmp1 = ( L_abs( tmp1 ) );



            if ( cond3 )

                tmp1 = L_negate( tmp1 );

#else

            IF( ( Are_fx_chA_0_GE_0 && Bre_fx_0_chB_GE_0 ) || ( Are_fx_chA_0_LT_0 && Bre_fx_0_chB_LT_0 ) )

            {

                tmp1 = Mpy_32_32( Are_fx[chA][0], Bre_fx[0][chB] );

            }

            ELSE

            {

                tmp1 = L_negate( Mpy_32_32( L_abs( Are_fx[chA][0] ), L_abs( Bre_fx[0][chB] ) ) );

            }

#endif



#if 1



            tmp2 = Mpy_32_32( Are_fx[chA][1], Bre_fx[1][chB] );

            cond2 = Are_fx_chA_1_GE_0 || Bre_fx_1_chB_GE_0;

            move32();

            cond1 = Are_fx_chA_1_LT_0 || Bre_fx_1_chB_LT_0;

            move32();

            cond3 = cond1 && cond2;

            move32();



            if ( cond3 )

                tmp2 = ( L_abs( tmp2 ) );



            if ( cond3 )

                tmp2 = L_negate( tmp2 );

#else

            IF( ( Are_fx_chA_1_GE_0 && Bre_fx_1_chB_GE_0 ) || ( Are_fx_chA_1_LT_0 && Bre_fx_1_chB_LT_0 ) )

            {

                tmp2 = Mpy_32_32( Are_fx[chA][1], Bre_fx[1][chB] );

            }

            ELSE

            {

                tmp2 = L_negate( Mpy_32_32( L_abs( Are_fx[chA][1] ), L_abs( Bre_fx[1][chB] ) ) );

            }

#endif

            outRe_fx[chA][chB] = L_add( tmp1, tmp2 );

            move32();



#if 1

            tmp1 = Mpy_32_32( Aim_fx[chA][0], Bim_fx[0][chB] );

            cond2 = Aim_fx_chA_0_GE_0 || Bim_fx_0_chB_GE_0;

            move32();

            cond1 = Aim_fx_chA_0_LT_0 || Bim_fx_0_chB_LT_0;

            move32();

            cond3 = cond1 && cond2;

            move32();



            if ( cond3 )

                tmp1 = ( L_abs( tmp1 ) );



            if ( cond3 )

                tmp1 = L_negate( tmp1 );

#else

            IF( ( Aim_fx_chA_0_GE_0 && Bim_fx_0_chB_GE_0 ) || ( Aim_fx_chA_0_LT_0 && Bim_fx_0_chB_LT_0 ) )

            {

                tmp1 = Mpy_32_32( Aim_fx[chA][0], Bim_fx[0][chB] );

            }

            ELSE

            {

                tmp1 = L_negate( Mpy_32_32( L_abs( Aim_fx[chA][0] ), L_abs( Bim_fx[0][chB] ) ) );

            }

#endif



#if 1



            tmp2 = Mpy_32_32( Aim_fx[chA][1], Bim_fx[1][chB] );

            cond2 = Aim_fx_chA_1_GE_0 || Bim_fx_1_chB_GE_0;

            move32();

            cond1 = Aim_fx_chA_1_LT_0 || Bim_fx_1_chB_LT_0;

            move32();

            cond3 = cond1 && cond2;

            move32();



            if ( cond3 )

                tmp2 = ( L_abs( tmp2 ) );



            if ( cond3 )

                tmp2 = L_negate( tmp2 );

#else

            IF( ( Aim_fx_chA_1_GE_0 && Bim_fx_1_chB_GE_0 ) || ( Aim_fx_chA_1_LT_0 && Bim_fx_1_chB_LT_0 ) )

            {

                tmp2 = Mpy_32_32( Aim_fx[chA][1], Bim_fx[1][chB] );

            }

            ELSE

            {

                tmp2 = L_negate( Mpy_32_32( L_abs( Aim_fx[chA][1] ), L_abs( Bim_fx[1][chB] ) ) );

            }

#endif



            outRe_fx[chA][chB] = L_sub( outRe_fx[chA][chB], L_add( tmp1, tmp2 ) );

            move32();



#if 1

            tmp1 = Mpy_32_32( Aim_fx[chA][0], Bre_fx[0][chB] );

            cond2 = Aim_fx_chA_0_GE_0 || Bre_fx_0_chB_GE_0;

            move32();

            cond1 = Aim_fx_chA_0_LT_0 || Bre_fx_0_chB_LT_0;

            move32();

            cond3 = cond1 && cond2;

            move32();



            if ( cond3 )

                tmp1 = ( L_abs( tmp1 ) );



            if ( cond3 )

                tmp1 = L_negate( tmp1 );

#else



            IF( ( Aim_fx_chA_0_GE_0 && Bre_fx_0_chB_GE_0 ) || ( Aim_fx_chA_0_LT_0 && Bre_fx_0_chB_LT_0 ) )

            {

                tmp1 = Mpy_32_32( Aim_fx[chA][0], Bre_fx[0][chB] );

            }

            ELSE

            {

                tmp1 = L_negate( Mpy_32_32( L_abs( Aim_fx[chA][0] ), L_abs( Bre_fx[0][chB] ) ) );

            }

#endif



#if 1



            tmp2 = Mpy_32_32( Aim_fx[chA][1], Bre_fx[1][chB] );

            cond2 = Aim_fx_chA_1_GE_0 || Bre_fx_1_chB_GE_0;

            move32();

            cond1 = Aim_fx_chA_1_LT_0 || Bre_fx_1_chB_LT_0;

            move32();

            cond3 = cond1 && cond2;

            move32();



            if ( cond3 )

                tmp2 = ( L_abs( tmp2 ) );



            if ( cond3 )

                tmp2 = L_negate( tmp2 );

#else

            IF( ( Aim_fx_chA_1_GE_0 && Bre_fx_1_chB_GE_0 ) || ( Aim_fx_chA_1_LT_0 && Bre_fx_1_chB_LT_0 ) )

            {

                tmp2 = Mpy_32_32( Aim_fx[chA][1], Bre_fx[1][chB] );

            }

            ELSE

            {

                tmp2 = L_negate( Mpy_32_32( L_abs( Aim_fx[chA][1] ), L_abs( Bre_fx[1][chB] ) ) );

            }

#endif





            outIm_fx[chA][chB] = L_add( tmp1, tmp2 );

            move32();



#if 1

            tmp1 = Mpy_32_32( Are_fx[chA][0], Bim_fx[0][chB] );

            cond2 = Are_fx_chA_0_GE_0 || Bim_fx_0_chB_GE_0;

            move32();

            cond1 = Are_fx_chA_0_LT_0 || Bim_fx_0_chB_LT_0;

            move32();

            cond3 = cond1 && cond2;

            move32();



            if ( cond3 )

                tmp1 = ( L_abs( tmp1 ) );



            if ( cond3 )

                tmp1 = L_negate( tmp1 );

#else

            IF( ( Are_fx_chA_0_GE_0 && Bim_fx_0_chB_GE_0 ) || ( Are_fx_chA_0_LT_0 && Bim_fx_0_chB_LT_0 ) )

            {

                tmp1 = Mpy_32_32( Are_fx[chA][0], Bim_fx[0][chB] );

            }

            ELSE

            {

                tmp1 = L_negate( Mpy_32_32( L_abs( Are_fx[chA][0] ), L_abs( Bim_fx[0][chB] ) ) );

            }

#endif



#if 1

            tmp2 = Mpy_32_32( Are_fx[chA][1], Bim_fx[1][chB] );

            cond2 = Are_fx_chA_1_GE_0 || Bim_fx_1_chB_GE_0;

            move32();

            cond1 = Are_fx_chA_1_LT_0 || Bim_fx_1_chB_LT_0;

            move32();

            cond3 = cond1 && cond2;

            move32();



            if ( cond3 )

                tmp2 = ( L_abs( tmp2 ) );



            if ( cond3 )

                tmp2 = L_negate( tmp2 );

#else

            IF( ( Are_fx_chA_1_GE_0 && Bim_fx_1_chB_GE_0 ) || ( Are_fx_chA_1_LT_0 && Bim_fx_1_chB_LT_0 ) )

            {

                tmp2 = Mpy_32_32( Are_fx[chA][1], Bim_fx[1][chB] );

            }

            ELSE

            {

                tmp2 = L_negate( Mpy_32_32( L_abs( Are_fx[chA][1] ), L_abs( Bim_fx[1][chB] ) ) );

            }

#endif



            outIm_fx[chA][chB] = L_add( outIm_fx[chA][chB], L_add( tmp1, tmp2 ) );

            move32();



#else

            test();

            test();
@@ -3621,6 +3879,9 @@ static void matrixMul_fx( 
            }

            outIm_fx[chA][chB] = L_add( outIm_fx[chA][chB], L_add( tmp1, tmp2 ) );

            move32();





#endif

#endif /* #ifdef IVAS_ENH64_CADENCE_CHANGES */

        }

    }
@@ -3742,6 +4003,10 @@ static void matrixTransp2Mul_fx( 
    Word32 Bre_fx[BINAURAL_CHANNELS][BINAURAL_CHANNELS], /*q_B*/

    Word32 Bim_fx[BINAURAL_CHANNELS][BINAURAL_CHANNELS], /*q_B*/

    Word16 *q_B,

#ifdef FIX_1072_SPEEDUP_matrixTransp2Mul_fx_scaleSwitch

    int Ascale,

    int Bscale,

#endif

    Word32 outRe_fx[BINAURAL_CHANNELS][BINAURAL_CHANNELS], /*q_out*/

    Word32 outIm_fx[BINAURAL_CHANNELS][BINAURAL_CHANNELS], /*q_out*/

    Word16 *q_out )
@@ -3753,16 +4018,27 @@ static void matrixTransp2Mul_fx( 
    Word32 tmp1, tmp2;

#endif



#ifdef FIX_1072_SPEEDUP_matrixTransp2Mul_fx_scaleSwitch

    IF (Ascale == 1)

#endif

    {

        min_q_shift = sub( s_min( L_norm_arr( Are_fx[0], size ), L_norm_arr( Aim_fx[0], size ) ), 1 );

        scale_sig32( Are_fx[0], size, min_q_shift );

        scale_sig32( Aim_fx[0], size, min_q_shift );

        *q_A = add( *q_A, min_q_shift );

        move16();

    }



#ifdef FIX_1072_SPEEDUP_matrixTransp2Mul_fx_scaleSwitch

    IF (Bscale == 1)

#endif

    {

        min_q_shift = sub( s_min( L_norm_arr( Bre_fx[0], size ), L_norm_arr( Bim_fx[0], size ) ), 1 );

        scale_sig32( Bre_fx[0], size, min_q_shift );

        scale_sig32( Bim_fx[0], size, min_q_shift );

        *q_B = add( *q_B, min_q_shift );

        move16();

    }



    FOR( chA = 0; chA < BINAURAL_CHANNELS; chA++ )

    {
@@ -3778,6 +4054,216 @@ static void matrixTransp2Mul_fx( 
            outIm_fx[chA][chB] = L_add( outIm_fx[chA][chB], W_extract_h( W_mac_32_32( W_mult_32_32( Are_fx[chA][0], L_negate( Bim_fx[chB][0] ) ), Are_fx[chA][1], L_negate( Bim_fx[chB][1] ) ) ) );

            move32();

#else

#ifdef FIX_1072_SPEEDUP_matrixTransp2Mul_fx_reducetest

            /* Create testVariables*/

            Word32 Are_fx_chA_0_GE_0 = GE_32( Are_fx[chA][0], 0 );

            Word32 Bre_fx_chB_0_GE_0 = GE_32( Bre_fx[chB][0], 0 );

            Word32 Are_fx_chA_0_LT_0 = LT_32( Are_fx[chA][0], 0 );

            Word32 Bre_fx_chB_0_LT_0 = LT_32( Bre_fx[chB][0], 0 );

            Word32 Are_fx_chA_1_GE_0 = GE_32( Are_fx[chA][1], 0 );

            Word32 Bre_fx_chB_1_GE_0 = GE_32( Bre_fx[chB][1], 0 );

            Word32 Are_fx_chA_1_LT_0 = LT_32( Are_fx[chA][1], 0 );

            Word32 Bre_fx_chB_1_LT_0 = LT_32( Bre_fx[chB][1], 0 );

            Word32 Aim_fx_chA_0_GE_0 = GE_32( Aim_fx[chA][0], 0 );

            Word32 Lneg_Bim_fx_chB_0_GE_0 = GE_32( L_negate( Bim_fx[chB][0] ), 0 );

            Word32 Aim_fx_chA_0_LT_0 = LT_32( Aim_fx[chA][0], 0 );

            Word32 Lneg_Bim_fx_chB_0_LT_0 = LT_32( L_negate( Bim_fx[chB][0] ), 0 );

            Word32 Aim_fx_chA_1_GE_0 = GE_32( Aim_fx[chA][1], 0 );

            Word32 Lneg_Bim_fx_chB_1_GE_0 = GE_32( L_negate( Bim_fx[chB][1] ), 0 );

            Word32 Aim_fx_chA_1_LT_0 = LT_32( Aim_fx[chA][1], 0 );

            Word32 Lneg_Bim_fx_chB_1_LT_0 = LT_32( L_negate( Bim_fx[chB][1] ), 0 );

            Word32 cond1, cond2, cond3;

            

#ifdef FIX_1072_SPEEDUP_matrixTransp2Mul_fx_reducetest_reduceLabs

            tmp1 = Mpy_32_32( Are_fx[chA][0], Bre_fx[chB][0] );

            cond2 = Are_fx_chA_0_GE_0 || Bre_fx_chB_0_GE_0;

            move32();

            cond1 = Are_fx_chA_0_LT_0 || Bre_fx_chB_0_LT_0;

            move32();

            cond3 = cond1 && cond2;

            move32();



            if ( cond3 )

                tmp1 = ( L_abs( tmp1 ) );



            if ( cond3 )

                tmp1 = L_negate( tmp1 );

#else

            IF( ( Are_fx_chA_0_GE_0 && Bre_fx_chB_0_GE_0 ) || ( Are_fx_chA_0_LT_0 && Bre_fx_chB_0_LT_0 ) )

            tmp1 = Mpy_32_32( Are_fx[chA][0], Bre_fx[chB][0] );

            ELSE

                tmp1 = L_negate( Mpy_32_32( L_abs( Are_fx[chA][0] ), L_abs( Bre_fx[chB][0] ) ) );

#endif /*FIX_1072_SPEEDUP_matrixTransp2Mul_fx_reducetest_reduceLabs*/



#ifdef FIX_1072_SPEEDUP_matrixTransp2Mul_fx_reducetest_reduceLabs



            tmp2 = Mpy_32_32( Are_fx[chA][1], Bre_fx[chB][1] );

            cond2 = Are_fx_chA_1_GE_0 || Bre_fx_chB_1_GE_0;

            move32();

            cond1 = Are_fx_chA_1_LT_0 || Bre_fx_chB_1_LT_0;

            move32();

            cond3 = cond1 && cond2;

            move32();



            if ( cond3 )

                tmp2 = ( L_abs( tmp2 ) );



            if ( cond3 )

                tmp2 = L_negate( tmp2 );

#else

            IF( ( Are_fx_chA_1_GE_0 && Bre_fx_chB_1_GE_0 ) || ( Are_fx_chA_1_LT_0 && Bre_fx_chB_1_LT_0 ) )

            tmp2 = Mpy_32_32( Are_fx[chA][1], Bre_fx[chB][1] );

            ELSE

                tmp2 = L_negate( Mpy_32_32( L_abs( Are_fx[chA][1] ), L_abs( Bre_fx[chB][1] ) ) );

#endif /*FIX_1072_SPEEDUP_matrixTransp2Mul_fx_reducetest_reduceLabs*/





            outRe_fx[chA][chB] = L_add( tmp1, tmp2 );

            move32();



#ifdef FIX_1072_SPEEDUP_matrixTransp2Mul_fx_reducetest_reduceLabs

            tmp1 = Mpy_32_32( Aim_fx[chA][0], L_negate( Bim_fx[chB][0] ) );

            cond2 = Aim_fx_chA_0_GE_0 || Lneg_Bim_fx_chB_0_GE_0;

            move32();

            cond1 = Aim_fx_chA_0_LT_0 || Lneg_Bim_fx_chB_0_LT_0;

            move32();

            cond3 = cond1 && cond2;

            move32();



            if ( cond3 )

                tmp1 = ( L_abs( tmp1 ) );



            if ( cond3 )

                tmp1 = L_negate( tmp1 );

#else



            IF( ( Aim_fx_chA_0_GE_0 && Lneg_Bim_fx_chB_0_GE_0 ) || ( Aim_fx_chA_0_LT_0 && Lneg_Bim_fx_chB_0_LT_0 ) )

            tmp1 = Mpy_32_32( Aim_fx[chA][0], -Bim_fx[chB][0] );

            ELSE

                tmp1 = L_negate( Mpy_32_32( L_abs( Aim_fx[chA][0] ), L_abs( Bim_fx[chB][0] ) ) );

#endif /*FIX_1072_SPEEDUP_matrixTransp2Mul_fx_reducetest_reduceLabs*/





#ifdef FIX_1072_SPEEDUP_matrixTransp2Mul_fx_reducetest_reduceLabs

            tmp2 = Mpy_32_32( Aim_fx[chA][1], L_negate( Bim_fx[chB][1] ) );

            cond2 = Aim_fx_chA_1_GE_0 || Lneg_Bim_fx_chB_1_GE_0;

            move32();

            cond1 = Aim_fx_chA_1_LT_0 || Lneg_Bim_fx_chB_1_LT_0;

            move32();

            cond3 = cond1 && cond2;

            move32();



            if ( cond3 )

                tmp2 = ( L_abs( tmp2 ) );



            if ( cond3 )

                tmp2 = L_negate( tmp2 );



#else



            IF( ( Aim_fx_chA_1_GE_0 && Lneg_Bim_fx_chB_1_GE_0 ) || ( Aim_fx_chA_1_LT_0 && Lneg_Bim_fx_chB_1_LT_0 ) )

            tmp2 = Mpy_32_32( Aim_fx[chA][1], -Bim_fx[chB][1] );

            ELSE

                tmp2 = L_negate( Mpy_32_32( L_abs( Aim_fx[chA][1] ), L_abs( -Bim_fx[chB][1] ) ) );

#endif /*FIX_1072_SPEEDUP_matrixTransp2Mul_fx_reducetest_reduceLabs*/



            outRe_fx[chA][chB] = L_sub( outRe_fx[chA][chB], L_add( tmp1, tmp2 ) );

            move32();



#ifdef FIX_1072_SPEEDUP_matrixTransp2Mul_fx_reducetest_reduceLabs

            tmp1 = Mpy_32_32( Aim_fx[chA][0], Bre_fx[chB][0] );

            cond2 = Aim_fx_chA_0_GE_0 || Bre_fx_chB_0_GE_0;

            move32();

            cond1 = Aim_fx_chA_0_LT_0 || Bre_fx_chB_0_LT_0;

            move32();

            cond3 = cond1 && cond2;

            move32();



            if ( cond3 )

                tmp1 = ( L_abs( tmp1 ) );



            if ( cond3 )

                tmp1 = L_negate( tmp1 );

#else



            IF( ( Aim_fx_chA_0_GE_0 && Bre_fx_chB_0_GE_0 ) || ( Aim_fx_chA_0_LT_0 && Bre_fx_chB_0_LT_0 ) )

            tmp1 = Mpy_32_32( Aim_fx[chA][0], Bre_fx[chB][0] );

            ELSE

                tmp1 = L_negate( Mpy_32_32( L_abs( Aim_fx[chA][0] ), L_abs( Bre_fx[chB][0] ) ) );

#endif /*FIX_1072_SPEEDUP_matrixTransp2Mul_fx_reducetest_reduceLabs*/



#ifdef FIX_1072_SPEEDUP_matrixTransp2Mul_fx_reducetest_reduceLabs

            tmp2 = Mpy_32_32( Aim_fx[chA][1], Bre_fx[chB][1] );

            cond2 = Aim_fx_chA_1_GE_0 || Bre_fx_chB_1_GE_0;

            move32();

            cond1 = Aim_fx_chA_1_LT_0 || Bre_fx_chB_1_LT_0;

            move32();

            cond3 = cond1 && cond2;

            move32();



            if ( cond3 )

                tmp2 = ( L_abs( tmp2 ) );



            if ( cond3 )

                tmp2 = L_negate( tmp2 );

#else

            IF( ( Aim_fx_chA_1_GE_0 && Bre_fx_chB_1_GE_0 ) || ( Aim_fx_chA_1_LT_0 && Bre_fx_chB_1_LT_0 ) )

            tmp2 = Mpy_32_32( Aim_fx[chA][1], Bre_fx[chB][1] );

            ELSE

                tmp2 = L_negate( Mpy_32_32( L_abs( Aim_fx[chA][1] ), L_abs( Bre_fx[chB][1] ) ) );

#endif /*FIX_1072_SPEEDUP_matrixTransp2Mul_fx_reducetest_reduceLabs*/



            outIm_fx[chA][chB] = L_add( tmp1, tmp2 );

            move32();



#ifdef FIX_1072_SPEEDUP_matrixTransp2Mul_fx_reducetest_reduceLabs

            tmp1 = Mpy_32_32( Are_fx[chA][0], L_negate( Bim_fx[chB][0] ) );

            cond2 = Are_fx_chA_0_GE_0 || Lneg_Bim_fx_chB_0_GE_0;

            move32();

            cond1 = Are_fx_chA_0_LT_0 || Lneg_Bim_fx_chB_0_LT_0;

            move32();

            cond3 = cond1 && cond2;

            move32();



            if ( cond3 )

                tmp1 = ( L_abs( tmp1 ) );



            if ( cond3 )

                tmp1 = L_negate( tmp1 );

#else



            IF( ( Are_fx_chA_0_GE_0 && Lneg_Bim_fx_chB_0_GE_0 ) || ( Are_fx_chA_0_LT_0 && Lneg_Bim_fx_chB_0_LT_0 ) )

            tmp1 = Mpy_32_32( Are_fx[chA][0], -Bim_fx[chB][0] );

            ELSE

                tmp1 = L_negate( Mpy_32_32( L_abs( Are_fx[chA][0] ), L_abs( -Bim_fx[chB][0] ) ) );

#endif /*FIX_1072_SPEEDUP_matrixTransp2Mul_fx_reducetest_reduceLabs*/



#ifdef FIX_1072_SPEEDUP_matrixTransp2Mul_fx_reducetest_reduceLabs

            tmp2 = Mpy_32_32( Are_fx[chA][1], L_negate( Bim_fx[chB][1] ) );

            cond2 = Are_fx_chA_1_GE_0 || Lneg_Bim_fx_chB_1_GE_0;

            move32();

            cond1 = Are_fx_chA_1_LT_0 || Lneg_Bim_fx_chB_1_LT_0;

            move32();

            cond3 = cond1 && cond2;

            move32();



            if ( cond3 )

                tmp2 = ( L_abs( tmp2 ) );



            if ( cond3 )

                tmp2 = L_negate( tmp2 );

#else

            IF( ( Are_fx_chA_1_GE_0 && Lneg_Bim_fx_chB_1_GE_0 ) || ( Are_fx_chA_1_LT_0 && Lneg_Bim_fx_chB_1_LT_0 ) )

            tmp2 = Mpy_32_32( Are_fx[chA][1], -Bim_fx[chB][1] );

            ELSE

                tmp2 = L_negate( Mpy_32_32( L_abs( Are_fx[chA][1] ), L_abs( -Bim_fx[chB][1] ) ) );

#endif /*FIX_1072_SPEEDUP_matrixTransp2Mul_fx_reducetest_reduceLabs*/



            outIm_fx[chA][chB] = L_add( outIm_fx[chA][chB], L_add( tmp1, tmp2 ) );

            move32();

#else /*FIX_1072_SPEEDUP_matrixTransp2Mul_fx_reducetest*/





            test();

            test();

            test();
@@ -3845,6 +4331,8 @@ static void matrixTransp2Mul_fx( 
                tmp2 = L_negate( Mpy_32_32( L_abs( Are_fx[chA][1] ), L_abs( -Bim_fx[chB][1] ) ) );

            outIm_fx[chA][chB] = L_add( outIm_fx[chA][chB], L_add( tmp1, tmp2 ) );

            move32();

#endif /*FIX_1072_SPEEDUP_matrixTransp2Mul_fx_reducetest*/



#endif /* #ifdef IVAS_ENH64_CADENCE_CHANGES */

        }

    }
@@ -3994,7 +4482,6 @@ static void chol2x2_fx( 
        // 4611686 = Q62

        IF( outRe[1][1] == 0 )

        {

            push_wmops( "DIV3232_0" );

#if !defined( FIX_1072_REPLACE_DIV_SQRT_BY_ISQRT )

            outRe[0][1] = BASOP_Util_Divide3232_Scale_cadence( c_re, 4611686, &exp );

            move32();
@@ -4018,7 +4505,6 @@ static void chol2x2_fx( 
            move32();

            q_im = add( sub( 31, exp ), sub( q_c, 62 ) );

#endif

            pop_wmops();

        }

        ELSE

        {
@@ -4575,7 +5061,12 @@ static void formulate2x2MixingMatrix_fx( 
        }

    }



    matrixTransp2Mul_fx( tmpRe_fx, tmpIm_fx, &q_temp, Ure_fx, Uim_fx, &q_U, Pre_fx, Pim_fx, &q_P ); /* Nearest orthonormal matrix P to matrix A formulated */

    matrixTransp2Mul_fx( tmpRe_fx, tmpIm_fx, &q_temp, Ure_fx, Uim_fx, &q_U, 

#ifdef FIX_1072_SPEEDUP_matrixTransp2Mul_fx_scaleSwitch

                         1 /*int Ascale*/,

                         0 /*int Bscale*/,

#endif

      Pre_fx, Pim_fx, &q_P ); /* Nearest orthonormal matrix P to matrix A formulated */



    /* These are the final formulas of the JAES publication M = Ky P Kx^(-1) */

    FOR( chA = 0; chA < BINAURAL_CHANNELS; chA++ )
@@ -4622,7 +5113,12 @@ static void formulate2x2MixingMatrix_fx( 


    matrixMul_fx( KyRe_fx, KyIm_fx, &q_ky, Pre_fx, Pim_fx, &q_P, tmpRe_fx, tmpIm_fx, &q_temp );



    matrixTransp2Mul_fx( tmpRe_fx, tmpIm_fx, &q_temp, Uxre_fx, Uxim_fx, &q_Ux, Mre_fx, Mim_fx, q_M );

    matrixTransp2Mul_fx( tmpRe_fx, tmpIm_fx, &q_temp, Uxre_fx, Uxim_fx, &q_Ux, 

#ifdef FIX_1072_SPEEDUP_matrixTransp2Mul_fx_scaleSwitch

                         1 /*int Ascale*/,

                         0 /*int Bscale*/,

#endif

      Mre_fx, Mim_fx, q_M );



    pop_wmops();

    return;