Commit 3ad19ded authored by Fabian Bauer's avatar Fabian Bauer
Browse files

introduced STAGE2, but still not functional and inactive - implemented speedup... 

introduced STAGE2, but still not functional and inactive - implemented speedup for STAGE0 and activated
parent 834a5383

lib_com/options.h

+3 −2
Original line number Diff line number Diff line
@@ -83,8 +83,9 @@ 
#define FIX_1439_SPEEDUP_Copy_Scale_sig_16_32_no_sat            /*FhG: reduces WMOPS - bit-exact*/

#define FIX_1439_SPEEDUP_stereo_icBWE_dec_fx                    /*FhG: reduces WMOPS - bit-exact*/

#define FIX_1439_SPEEDUP_ivas_swb_tbe_dec_fx                    /*FhG: reduces WMOPS - bit-exact*/

#define FIX_1439_SPEEDUP_SIMPLIFY_elliptic_bpf_48k_generic              /*FhG: reduces maintenance complexity & reduces WMOPS & prepares another speedup patch*/



#define FIX_1439_SPEEDUP_synthesise_fb_high_band_fx             /*FhG: reduces WMOPS - bit-exact*/



#define FIX_1439_SPEEDUP_SIMPLIFY_elliptic_bpf_48k_generic              /*FhG: reduces maintenance complexity & reduces WMOPS & prepares STAGE2 patch*/

//#define FIX_1439_SPEEDUP_SIMPLIFY_elliptic_bpf_48k_generic_STAGE2       /*FhG: reduces WMOPS* /



#endif

lib_com/prot_fx.h

+4 −0
Original line number Diff line number Diff line
@@ -3258,7 +3258,11 @@ void interp_code_4over2_fx( 
void wb_tbe_extras_reset_synth_fx( Word16 state_lsyn_filt_shb[], Word16 state_lsyn_filt_dwn_shb[], Word16 state_32and48k_WB_upsample[], Word16 state_resamp_HB[] );















void elliptic_bpf_48k_generic_fx(













#ifdef FIX_1439_SPEEDUP_SIMPLIFY_elliptic_bpf_48k_generic_STAGE2













    int isIVAS;













#endif















#ifdef FIX_1439_SPEEDUP_SIMPLIFY_elliptic_bpf_48k_generic













    Word16 IsUpsampled3,















    Word16 input_fx[], /* i  : input signal                             Q_input_fx*/















#else















    const Word16 input_fx[], /* i  : input signal                             Q_input_fx*/

































lib_com/swb_tbe_com_fx.c



+116
−11




























Original line number
Diff line number
Diff line








@@ -6695,7 +6695,11 @@ void wb_tbe_extras_reset_synth_fx(













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































void elliptic_bpf_48k_generic_fx(













#ifdef FIX_1439_SPEEDUP_SIMPLIFY_elliptic_bpf_48k_generic_STAGE2













    int isIVAS;













#endif















#ifdef FIX_1439_SPEEDUP_SIMPLIFY_elliptic_bpf_48k_generic













    Word16 IsUpsampled3,















    Word16 input_fx[], /* i  : input signal                             Q_input_fx*/















#else















    const Word16 input_fx[], /* i  : input signal                             Q_input_fx*/










@@ -6742,6 +6746,26 @@ void elliptic_bpf_48k_generic_fx(













    }































#ifdef FIX_1439_SPEEDUP_SIMPLIFY_elliptic_bpf_48k_generic













#ifdef FIX_1439_SPEEDUP_SIMPLIFY_elliptic_bpf_48k_generic_STAGE2













    FOR( i = 0; i < L_FRAME48k; i++ )













    {













        Word64 W_tmpX;













        Word64 W_tmpY;













        W_tmpX = W_mac_16_16( 0, input_fx[i - 4], full_band_bpf_fx[0][4] );













        W_tmpX = W_mac_16_16( W_tmpX, input_fx[i - 3], full_band_bpf_fx[0][3] );













        W_tmpY = W_msu_32_16( 0, L_tmp[i - 1], full_band_bpf_fx[3][1] );













        W_tmpX = W_mac_16_16( W_tmpX, input_fx[i - 2], full_band_bpf_fx[0][2] );













        W_tmpY = W_msu_32_16( W_tmpY, L_tmp[i - 2], full_band_bpf_fx[3][2] );













        W_tmpX = W_mac_16_16( W_tmpX, input_fx[i - 1], full_band_bpf_fx[0][1] );













        W_tmpY = W_msu_32_16( W_tmpY, L_tmp[i - 3], full_band_bpf_fx[3][3] );













        W_tmpX = W_mac_16_16( W_tmpX, input_fx[i], full_band_bpf_fx[0][0] );













        W_tmpY = W_msu_32_16( W_tmpY, L_tmp[i - 4], full_band_bpf_fx[3][4] );













        L_tmp[i] = W_sat_l( W_shr( W_add( W_tmpX, W_shl( W_tmpY, 2 - 16 + 3 ) ), 3 ) ); /*Q_input_fx + 11*/













        move32();













    }













#else













    IF( !IsUpsampled3 )













    {















        FOR( i = 0; i < L_FRAME48k; i++ )















        {















            L_tmpX = L_shr( L_mult( input_fx[i - 4], full_band_bpf_fx[0][4] ), 3 );                             /*Q_input_fx + 13 + 1 - 3*/









@@ -6755,6 +6779,24 @@ void elliptic_bpf_48k_generic_fx(













            L_tmp[i] = L_sub_sat( L_tmpX, L_shl_sat( Mult_32_16( L_tmp[i - 4], full_band_bpf_fx[3][4] ), 2 ) ); /*Q_input_fx + 11 + 13  -15 +2*/















            move32();















        }













    }













    ELSE













    {













        FOR( i = 0; i < L_FRAME48k; i++ )













        {













            L_tmpX = L_shr( L_mult( input_fx[i - 4], full_band_bpf_fx[0][4] ), 3 );                             /*Q_input_fx + 13 + 1 - 3*/













            //L_tmpX = L_add_sat( L_shr( L_mult( input_fx[i - 3], full_band_bpf_fx[0][3] ), 3 ), L_tmpX );        /*Q_input_fx + 13 + 1 - 3*/













            L_tmpX = L_sub_sat( L_tmpX, L_shl_sat( Mult_32_16( L_tmp[i - 1], full_band_bpf_fx[3][1] ), 2 ) );   /*Q_input_fx + 11 + 13  -15 +2*/













            //L_tmpX = L_add_sat( L_shr( L_mult( input_fx[i - 2], full_band_bpf_fx[0][2] ), 3 ), L_tmpX );        /*Q_input_fx + 13 + 1 - 3*/













            L_tmpX = L_sub_sat( L_tmpX, L_shl_sat( Mult_32_16( L_tmp[i - 2], full_band_bpf_fx[3][2] ), 2 ) );   /*Q_input_fx + 11 + 13  -15 +2*/













            L_tmpX = L_add_sat( L_shr( L_mult( input_fx[i - 1], full_band_bpf_fx[0][1] ), 3 ), L_tmpX );        /*Q_input_fx + 13 + 1 - 3*/













            L_tmpX = L_sub_sat( L_tmpX, L_shl_sat( Mult_32_16( L_tmp[i - 3], full_band_bpf_fx[3][3] ), 2 ) );   /*Q_input_fx + 11 + 13  -15 +2*/













            //L_tmpX = L_add_sat( L_shr( L_mult( input_fx[i], full_band_bpf_fx[0][0] ), 3 ), L_tmpX );            /*Q_input_fx + 13 + 1 - 3*/













            L_tmp[i] = L_sub_sat( L_tmpX, L_shl_sat( Mult_32_16( L_tmp[i - 4], full_band_bpf_fx[3][4] ), 2 ) ); /*Q_input_fx + 11 + 13  -15 +2*/













            move32();













        }













    }













#endif















#else















    L_tmpX = L_shr( L_mult( memory_fx[0][0], full_band_bpf_fx[0][4] ), 3 );                             /*Q_input_fx + 13 + 1 - 3*/















    L_tmpX = L_add( L_shr( L_mult( memory_fx[0][1], full_band_bpf_fx[0][3] ), 3 ), L_tmpX );            /*Q_input_fx + 13 + 1 - 3*/










@@ -6827,6 +6869,26 @@ void elliptic_bpf_48k_generic_fx(













    move32();































#ifdef FIX_1439_SPEEDUP_SIMPLIFY_elliptic_bpf_48k_generic













#ifdef FIX_1439_SPEEDUP_SIMPLIFY_elliptic_bpf_48k_generic_STAGE2













    FOR( i = 0; i < L_FRAME48k; i++ )













    {













        Word64 W_tmpX;













        Word64 W_tmpY;













        W_tmpX = W_mac_32_16( 0, L_tmp[i - 4], full_band_bpf_fx[1][4] );













        W_tmpX = W_mac_32_16( W_tmpX, L_tmp[i - 3], full_band_bpf_fx[1][3] );













        W_tmpY = W_msu_32_16( 0, L_tmp2[i - 1], full_band_bpf_fx[4][1] );













        W_tmpX = W_mac_32_16( W_tmpX, L_tmp[i - 2], full_band_bpf_fx[1][2] );













        W_tmpY = W_msu_32_16( W_tmpY, L_tmp2[i - 2], full_band_bpf_fx[4][2] );













        W_tmpX = W_mac_32_16( W_tmpX, L_tmp[i - 1], full_band_bpf_fx[1][1] );













        W_tmpY = W_msu_32_16( W_tmpY, L_tmp2[i - 3], full_band_bpf_fx[4][3] );













        W_tmpX = W_mac_32_16( W_tmpX, L_tmp[i], full_band_bpf_fx[1][0] );













        W_tmpY = W_msu_32_16( W_tmpY, L_tmp2[i - 4], full_band_bpf_fx[4][4] );













        // L_tmp2[i] = W_sat_l( W_add( W_shr( W_tmpX, 3 + 16 ), W_shl( W_tmpY, 2 - 16 ) ) );













        L_tmp2[i] = W_sat_l( W_shr( W_add( W_tmpX, W_shl( W_tmpY, 2 - 16 + 3 + 16 ) ), 3 + 16 ) );













        move32();













        L_tmpMax = L_max( L_tmpMax, L_abs( L_tmp2[i] ) );













    }













#else















    FOR( i = 0; i < L_FRAME48k; i++ )















    {















        L_tmpX = L_shr( Mult_32_16( L_tmp[i - 4], full_band_bpf_fx[1][4] ), 3 );                              /*Q_input_fx + 11 + 13 - 15 -3*/









@@ -6841,6 +6903,7 @@ void elliptic_bpf_48k_generic_fx(













        move32();















        L_tmpMax = L_max( L_tmpMax, L_abs( L_tmp2[i] ) );















    }













#endif















#else















    L_tmpX = L_shr( Mult_32_16( memory2_fx[1][0], full_band_bpf_fx[1][4] ), 3 );                             /*Q_input_fx + 11 + 13 - 15 -3*/















    L_tmpX = L_add( L_shr( Mult_32_16( memory2_fx[1][1], full_band_bpf_fx[1][3] ), 3 ), L_tmpX );            /*Q_input_fx + 11 + 13 - 15 -3*/










@@ -6925,6 +6988,30 @@ void elliptic_bpf_48k_generic_fx(













        move32();















    }















#ifdef FIX_1439_SPEEDUP_SIMPLIFY_elliptic_bpf_48k_generic













#ifdef FIX_1439_SPEEDUP_SIMPLIFY_elliptic_bpf_48k_generic_STAGE2













    FOR( 0 = 4; i < L_FRAME48k; i++ )













    {













        Word64 W_tmpX;













        Word64 W_tmpY;



























        W_tmpX = W_mac_32_16( 0, L_tmp2[i - 4], full_band_bpf_fx[2][4] );













        W_tmpX = W_mac_32_16( W_tmpX, L_tmp2[i - 3], full_band_bpf_fx[2][3] );













        W_tmpY = W_msu_32_16( 0, L_output[i - 1], full_band_bpf_fx[5][1] );



























        W_tmpX = W_mac_32_16( W_tmpX, L_tmp2[i - 2], full_band_bpf_fx[2][2] );













        W_tmpY = W_msu_32_16( W_tmpY, L_output[i - 2], full_band_bpf_fx[5][2] );



























        W_tmpX = W_mac_32_16( W_tmpX, L_tmp2[i - 1], full_band_bpf_fx[2][1] );













        W_tmpY = W_msu_32_16( W_tmpY, L_output[i - 3], full_band_bpf_fx[5][3] );



























        W_tmpX = W_mac_32_16( W_tmpX, L_tmp2[i], full_band_bpf_fx[2][0] );













        W_tmpY = W_msu_32_16( W_tmpY, L_output[i - 4], full_band_bpf_fx[5][4] );













        // L_output[i] = W_sat_l( W_add( W_shr( W_tmpX, 3 + 16 ), W_shl( W_tmpY, 2 - 16 ) ) );













        L_output[i] = W_sat_l( W_shr( W_add( W_tmpX, W_shl( W_tmpY, 2 - 16 + 3 + 16 ) ), 3 + 16 ) );













        move32();













        L_tmpMax = L_max( L_tmpMax, L_abs( L_output[i] ) );













    }













#else















    FOR( i = 0; i < L_FRAME48k; i++ )















    {















        L_tmpX = L_shr( Mult_32_16( L_tmp2[i - 4], full_band_bpf_fx[2][4] ), 3 );                            /*Q_input_fx + 6 +Q_temp+13 -15 -3 */









@@ -6942,6 +7029,7 @@ void elliptic_bpf_48k_generic_fx(













        move32();















        L_tmpMax = L_max( L_tmpMax, L_abs( L_output[i] ) );















    }













#endif















#else















    L_tmpX = L_shr( Mult_32_16( memory2_fx_2[0], full_band_bpf_fx[2][4] ), 3 );                               /* *Q_input_fx+6 +Q_temp +13 -15 -3 */















    L_tmpX = L_add_sat( L_shr( Mult_32_16( memory2_fx_2[1], full_band_bpf_fx[2][3] ), 3 ), L_tmpX );          /*Q_input_fx + 6 +Q_temp+13 -15 -3*/










@@ -7108,12 +7196,29 @@ void synthesise_fb_high_band_fx(













    IF( EQ_16( L_frame, L_FRAME16k ) )















    {















        /* for 16kHz ACELP core */













        elliptic_bpf_48k_generic_fx( excitation_in_interp3, &exp_tmp, tmp, bpf_memory, bpf_memory_Q, full_band_bpf_3_fx );













        elliptic_bpf_48k_generic_fx(













#ifdef FIX_1439_SPEEDUP_SIMPLIFY_elliptic_bpf_48k_generic_STAGE2













            1 // isIVAS













#endif













#ifdef FIX_1439_SPEEDUP_SIMPLIFY_elliptic_bpf_48k_generic













            1, // IsUpsampled3













#endif













            excitation_in_interp3, &exp_tmp, tmp, bpf_memory, bpf_memory_Q, full_band_bpf_3_fx



























        );















    }















    ELSE















    {















        /* for 12.8kHz ACELP core */













        elliptic_bpf_48k_generic_fx( excitation_in_interp3, &exp_tmp, tmp, bpf_memory, bpf_memory_Q, full_band_bpf_1_fx );













        elliptic_bpf_48k_generic_fx(













#ifdef FIX_1439_SPEEDUP_SIMPLIFY_elliptic_bpf_48k_generic_STAGE2













            1 // isIVAS













#endif













#ifdef FIX_1439_SPEEDUP_SIMPLIFY_elliptic_bpf_48k_generic













            1, // IsUpsampled3













#endif













            excitation_in_interp3, &exp_tmp, tmp, bpf_memory, bpf_memory_Q, full_band_bpf_1_fx













        );















    }















    temp1 = sum2_fx_mod( tmp, L_FRAME48k );

















































lib_enc/swb_tbe_enc_fx.c



+22
−3




























Original line number
Diff line number
Diff line








@@ -7330,7 +7330,14 @@ void fb_tbe_enc_fx(





























    Copy_Scale_sig( new_input, input_fhb, L_FRAME48k, exp_temp );





























    elliptic_bpf_48k_generic_fx( input_fhb, &exp_temp, tmp_vec, hBWE_TD->elliptic_bpf_2_48k_mem_fx, hBWE_TD->elliptic_bpf_2_48k_mem_fx_Q, full_band_bpf_2_fx );













    elliptic_bpf_48k_generic_fx(













#ifdef FIX_1439_SPEEDUP_SIMPLIFY_elliptic_bpf_48k_generic_STAGE2













        1 // isIVAS













#endif













#ifdef FIX_1439_SPEEDUP_SIMPLIFY_elliptic_bpf_48k_generic













        0, // IsUpsampled3













#endif













        input_fhb, &exp_temp, tmp_vec, hBWE_TD->elliptic_bpf_2_48k_mem_fx, hBWE_TD->elliptic_bpf_2_48k_mem_fx_Q, full_band_bpf_2_fx );















    Sample_Delay_HP = NS2SA( 48000, ACELP_LOOK_NS + DELAY_FD_BWE_ENC_12k8_NS + DELAY_FIR_RESAMPL_NS ) - L_FRAME48k / 2;































    IF( NE_16( st->last_extl, FB_TBE ) )










@@ -7459,11 +7466,23 @@ void fb_tbe_enc_ivas_fx(





























    IF( EQ_16( st->element_mode, IVAS_CPE_DFT ) )















    {













        elliptic_bpf_48k_generic_fx( input_fhb_new, &exp_temp, tmp_vec, hBWE_TD->elliptic_bpf_2_48k_mem_fx, hBWE_TD->elliptic_bpf_2_48k_mem_fx_Q, full_band_bpf_2_fx );













        elliptic_bpf_48k_generic_fx( 













#ifdef FIX_1439_SPEEDUP_SIMPLIFY_elliptic_bpf_48k_generic_STAGE2













        1 // isIVAS













#endif













#ifdef FIX_1439_SPEEDUP_SIMPLIFY_elliptic_bpf_48k_generic













        0, // IsUpsampled3













#endif













          input_fhb_new, &exp_temp, tmp_vec, hBWE_TD->elliptic_bpf_2_48k_mem_fx, hBWE_TD->elliptic_bpf_2_48k_mem_fx_Q, full_band_bpf_2_fx );















    }















    ELSE















    {













        elliptic_bpf_48k_generic_fx( input_fhb_new + NS2SA( 48000, DELAY_FIR_RESAMPL_NS ), &exp_temp, tmp_vec, hBWE_TD->elliptic_bpf_2_48k_mem_fx, hBWE_TD->elliptic_bpf_2_48k_mem_fx_Q, full_band_bpf_2_fx );













        elliptic_bpf_48k_generic_fx( input_fhb_new + NS2SA( 48000, DELAY_FIR_RESAMPL_NS ), &exp_temp, tmp_vec, hBWE_TD->elliptic_bpf_2_48k_mem_fx, hBWE_TD->elliptic_bpf_2_48k_mem_fx_Q, full_band_bpf_2_fx 













#ifdef FIX_1439_SPEEDUP_SIMPLIFY_elliptic_bpf_48k_generic_STAGE2













    , 1 // IsUpsampled3













    , 1 // isIVAS













#endif 













        );















    }































    test();