Commit f1e32ecf authored by Fabian Bauer's avatar Fabian Bauer
Browse files

add FIX_1439_SPEEDUP_elliptic_bpf_48k_generic_upsampledsig

parent 932c3f55

lib_com/options.h

+1 −0
Original line number Diff line number Diff line
@@ -85,5 +85,6 @@ 
#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_elliptic_bpf_48k_generic_upsampledsig  /*FhG: reduces WMOPS - bit-exact*/ //   |



#endif

lib_com/swb_tbe_com_fx.c

+83 −10
Original line number Diff line number Diff line
@@ -6763,6 +6763,78 @@ void elliptic_bpf_48k_generic_fx( 
    L_tmpX = L_sub_sat( L_tmpX, L_shl_sat( Mult_32_16( L_tmp[0], full_band_bpf_fx[3][3] ), 2 ) );           /*Q_input_fx + 11 + 13  -15 +2*/

    L_tmp[3] = L_sub_sat( L_tmpX, L_shl_sat( Mult_32_16( memory2_fx[1][3], full_band_bpf_fx[3][4] ), 2 ) ); /*Q_input_fx + 11 + 13  -15 +2*/

    move32();



#ifdef FIX_1439_SPEEDUP_elliptic_bpf_48k_generic_upsampledsig

    test();

    IF( full_band_bpf_fx == full_band_bpf_3_fx || full_band_bpf_fx == full_band_bpf_1_fx )

    {

        i = 4;

        L_tmpX = L_shr( L_mult( input_fx[i - 4], full_band_bpf_fx[0][4] ), 3 ); // 0 /*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 );        // 1 /*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 );        // 2 /*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 );      // 3 /*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 );            // 4 /*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();

        i++;



        // L_tmpX = L_shr( L_mult( input_fx[i - 4], full_band_bpf_fx[0][4] ), 3 );                             // 1 /*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 );        // 2 /*Q_input_fx + 13 + 1 - 3*/

        L_tmpX = L_sub_sat( 0, 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 );      // 3 /*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 );        // 4 /*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 );            // 5 /*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();

        i++;



        FOR( ; i < L_FRAME48k; )

        {

            // L_tmpX = L_shr( L_mult( input_fx[i - 4], full_band_bpf_fx[0][4] ), 3 );                             // 2 /*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 );           // 3 /*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 );        // 4 /*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 );        // 5 /*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 );            // 6 /*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();

            i++;



            L_tmpX = L_shr( L_mult( input_fx[i - 4], full_band_bpf_fx[0][4] ), 3 ); // 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 );        // 4 /*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 );        // 5 /*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 );      // 6 /*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 );            // 7 /*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();

            i++;



            // L_tmpX = L_shr( L_mult( input_fx[i - 4], full_band_bpf_fx[0][4] ), 3 );                             // 4 /*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 );        // 5 /*Q_input_fx + 13 + 1 - 3*/

            L_tmpX = L_sub_sat( 0, 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 );      // 6 /*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 );        // 7 /*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 );            // 8 /*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();

            i++;

        }

    }

    ELSE

#endif /*FIX_1439_SPEEDUP_elliptic_bpf_48k_generic_upsampledsig*/

    {

        FOR( i = 4; 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*/
@@ -6776,6 +6848,7 @@ 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();

        }

    }



    memory_fx2[0][0] = input_fx[L_FRAME48k - 4];

    memory_fx2[0][1] = input_fx[L_FRAME48k - 3];