Commit 9028c880 authored by Fabian Bauer's avatar Fabian Bauer Committed by Sandesh Venkatesh
Browse files

activated speedup 01 02 03 04

parent 12809f5b

lib_rend/ivas_dirac_dec_binaural_functions_fx.c

+32 −12
Original line number Diff line number Diff line
@@ -46,14 +46,14 @@ 
#include "wmc_auto.h"



//#define FIX_1326_SPEEDUP_00 //make sqrt(1) a const - catch bitstreams //no occurence in current bitstream

//#define FIX_1326_SPEEDUP_01 // optimize matrixTransp1Mul_fx -> eig2x2_fx :  .4 WMOPS

//#define FIX_1326_SPEEDUP_02 // speedup eig2x2_fx //  .3 WMOPS

//#define FIX_1326_SPEEDUP_03 // speedup eig2x2_fx //  .1 WMOPS

//#define FIX_1326_SPEEDUP_04 // speedup eig2x2_fx //  .2 WMOPS

//#define FIX_1326_SPEEDUP_05 // div->sqrt =>isqrt // 3.5 WMOPS //Quite bad diffs

//#define FIX_1326_SPEEDUP_06 // div->sqrt =>isqrt // 3.0 WMOPS  //Quite bad diffs

#define FIX_1326_SPEEDUP_07 // div->sqrt =>isqrt // 2.8 WMOPS //Big DIffs , no replacement of divSqrt

#define FIX_1326_SPEEDUP_08 // "-"               // 3.0 WMOPS //small diffs, PIPELINE GREEN!

#define FIX_1326_SPEEDUP_01 // optimize matrixTransp1Mul_fx -> eig2x2_fx :  .4 WMOPS

#define FIX_1326_SPEEDUP_02 // speedup eig2x2_fx //  .3 WMOPS

#define FIX_1326_SPEEDUP_03 // speedup eig2x2_fx //  .1 WMOPS

#define FIX_1326_SPEEDUP_04 // speedup eig2x2_fx //  .2 WMOPS

//#define FIX_1326_SPEEDUP_05 // div->sqrt =>isqrt // 3.5 WMOPS //Quite bad diffs --> DONT USE

//#define FIX_1326_SPEEDUP_06 // div->sqrt =>isqrt // 3.0 WMOPS  //Quite bad diffs -- > DONT USE

#define FIX_1326_SPEEDUP_07 // div->sqrt =>isqrt //  ?  WMOPS //Big DIffs , no replacement of divSqrt , PIPELINE GREEN, --> USE

#define FIX_1326_SPEEDUP_08 // "-"               // 3.0 WMOPS //small diffs, PIPELINE GREEN! -- > USE

Word16 slot_fx[4] = { 32767, 16384, 10922, 8192 };



/*-------------------------------------------------------------------------
@@ -4521,8 +4521,12 @@ static void formulate2x2MixingMatrix_fx( 
#ifdef FIX_1326_SPEEDUP_05

    {

        Word16 shift = norm_l( temp );

#if 1 // oldcode

        temp = BASOP_Util_Add_Mant32Exp( temp, sub( 31, q_ein ), EPSILON_MANT, EPSILON_EXP, &exp_temp );

#else



        temp = L_add( L_shl( temp, sub( shift, 1 ) ), L_shl_sat( EPSILON_MANT, sub( sub( EPSILON_EXP, shift ), 1 ) ) );

        exp_temp = sub( 31, q_ein );

        exp_temp = sub( 30, q_ein );

        if ( temp == 0 )

        {

            exp_temp = EPSILON_EXP;
@@ -4533,6 +4537,7 @@ static void formulate2x2MixingMatrix_fx( 
            temp = EPSILON_MANT;

            move32();

        }

#endif

        temp = ISqrt32( temp, &exp_temp );

        shift = sub( 31, q_eout );

        Ghat_fx[0] = Mpy_32_32( Sqrt32( E_out1, &shift ), temp );
@@ -4582,8 +4587,11 @@ static void formulate2x2MixingMatrix_fx( 
#ifdef FIX_1326_SPEEDUP_06

    {

        Word16 shift = norm_l( temp );

#if 0 //oldcode

        temp = BASOP_Util_Add_Mant32Exp( temp, sub( 31, q_ein ), EPSILON_MANT, EPSILON_EXP, &exp_temp );

#else

        temp = L_add( L_shl( temp, sub( shift, 1 ) ), L_shl_sat( EPSILON_MANT, sub( sub( EPSILON_EXP, shift ), 1 ) ) );

        exp_temp = sub( 31, q_ein );

        exp_temp = sub( 31 - 1, q_ein );

        if ( temp == 0 )

        {

            exp_temp = add( 0, EPSILON_EXP );
@@ -4592,10 +4600,22 @@ static void formulate2x2MixingMatrix_fx( 
        {

            temp = L_add( 0, EPSILON_MANT );

        }

#endif

#if 1 //oldcode - new code introduces too much noise

        push_wmops( "formulate2x2MixingMatrix Division" );

        temp = BASOP_Util_Divide3232_Scale_cadence( E_out2, temp, &exp1 );

        pop_wmops(); /*push_wmops( "formulate2x2MixingMatrix Division" )*/

        exp1 = sub( exp1, sub( q_eout, sub( 31, exp_temp ) ) );

#ifdef FIX_1009_REPLACE_DIV_SQRT_BY_ISQRT_LC

        Ghat_fx[1] = Sqrt32( temp, &exp1 ); // Q = 31 - exp1

#endif

#else

        temp = ISqrt32( temp, &exp_temp );

        shift = sub( 31, q_eout );

        Ghat_fx[1] = Mpy_32_32( temp, ISqrt32( E_out2, &shift ) );

        exp_temp = add( shift, exp_temp );

        Ghat_fx[1] = Mpy_32_32( temp, Sqrt32( E_out2, &shift ) );

        exp1 = add( shift, exp_temp );

#endif 



    }

#else

    {