Commit cdc6a0c2 authored by Fabian Bauer's avatar Fabian Bauer
Browse files

move stage2 code into loops

parent 21a7f761

lib_com/swb_tbe_com_fx.c

+70 −45
Original line number Diff line number Diff line
@@ -6785,6 +6785,46 @@ void elliptic_bpf_48k_generic_fx( 
    {

        FOR( i = 0; i < L_FRAME48k; )

        {

#ifdef FIX_1439_SPEEDUP_SIMPLIFY_elliptic_bpf_48k_generic_STAGE2

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

            W_tmpX = W_mac_16_16( 0, 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();

            i++;



            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();

            i++;



            // 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( 0, 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();

            i++;

#else

            // 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 ), 0 );           /*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*/
@@ -6820,6 +6860,7 @@ 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();

            i++;

#endif /*#ifdef FIX_1439_SPEEDUP_SIMPLIFY_elliptic_bpf_48k_generic_STAGE2*/

        }

    }

#else  /*FIX_1439_SPEEDUP_SIMPLIFY_elliptic_bpf_48k_generic*/
@@ -6894,11 +6935,9 @@ 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;

#ifdef FIX_1439_SPEEDUP_SIMPLIFY_elliptic_bpf_48k_generic_STAGE2

        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] );
@@ -6908,14 +6947,10 @@ void elliptic_bpf_48k_generic_fx( 
        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++ )

    {

#else  /*FIX_1439_SPEEDUP_SIMPLIFY_elliptic_bpf_48k_generic_STAGE2*/

        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*/

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

        L_tmpX = L_sub_sat( L_tmpX, L_shl_sat( Mult_32_16( L_tmp2[i - 1], full_band_bpf_fx[4][1] ), 2 ) );    /*Q_input_fx + 6 +13 -15 +2 */
@@ -6927,8 +6962,8 @@ void elliptic_bpf_48k_generic_fx( 
        L_tmp2[i] = L_sub_sat( L_tmpX, L_shl_sat( Mult_32_16( L_tmp2[i - 4], full_band_bpf_fx[4][4] ), 2 ) ); /*Q_input_fx + 6 +13 -15 +2 */

        move32();

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

#endif /*FIX_1439_SPEEDUP_SIMPLIFY_elliptic_bpf_48k_generic_STAGE2*/

    }

#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*/
@@ -7013,32 +7048,22 @@ 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++ )

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

    {

        Word64 W_tmpX;

        Word64 W_tmpY;



#ifdef FIX_1439_SPEEDUP_SIMPLIFY_elliptic_bpf_48k_generic_STAGE2

        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++ )

    {

#else  /*FIX_1439_SPEEDUP_SIMPLIFY_elliptic_bpf_48k_generic_STAGE2*/

        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 */

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

        L_tmpX = L_sub_sat( L_tmpX, L_shl_sat( Mult_32_16( L_output[i - 1], full_band_bpf_fx[5][1] ), 2 ) ); /*Q_input_fx + 1 +Q_temp+13 -15 + 2*/
@@ -7053,8 +7078,8 @@ void elliptic_bpf_48k_generic_fx( 
        L_output[i] = L_sub_sat( L_tmpX, L_shl_sat( Mult_32_16( L_output[i - 4], full_band_bpf_fx[5][4] ), 2 ) ); /*Q_input_fx + 1 +Q_temp+13 -15 + 2*/

        move32();

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

#endif /*FIX_1439_SPEEDUP_SIMPLIFY_elliptic_bpf_48k_generic_STAGE2*/

    }

#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*/