Commit 9fa09725 authored by Arthur Tritthart's avatar Arthur Tritthart
Browse files

added changes according to review comments, results should be same as in last commit

parent a77a208f

lib_com/hp50_fx.c

+18 −14
Original line number Diff line number Diff line
@@ -353,19 +353,26 @@ void hp20_fx_32_opt( 
    Word32 tmp_mem[4];



    prescale = L_norm_arr( signal_fx, lg );

    prescale_current_frame = sub( add( 1, HP20_FX_COEFF_SCALE ), prescale );

    prescale_current_frame = sub( 1 + HP20_FX_COEFF_SCALE, prescale );



    prescaleOld = extract_l( mem_fx[4] );



    Copy32( mem_fx, tmp_mem, 4 );

    Scale_sig32( tmp_mem, 4, prescaleOld );

    tmp_mem[0] = L_shl_sat( mem_fx[0], prescaleOld );

    tmp_mem[1] = L_shl_sat( mem_fx[1], prescaleOld );

    tmp_mem[2] = L_shl_sat( mem_fx[2], prescaleOld );

    tmp_mem[3] = L_shl_sat( mem_fx[3], prescaleOld );

    move32();

    move32();

    move32();

    move32();



    diff = L_norm_arr( tmp_mem, 4 );



    prescale = s_min( prescale, diff );



    prescale = sub( add( 1, HP20_FX_COEFF_SCALE ), prescale );

    prescale = sub( 1 + HP20_FX_COEFF_SCALE, prescale );



    IF( EQ_16( prescale_current_frame, -29 ) ) // signal_fx buffer contains only zeros, so use the mem_fx scale_factor instead

    if( EQ_16( prescale_current_frame, 1 + HP20_FX_COEFF_SCALE - 31 ) ) // signal_fx buffer contains only zeros, so use the mem_fx scale_factor instead

    {

        prescale_current_frame = prescale;

    }
@@ -373,21 +380,18 @@ void hp20_fx_32_opt( 
    diff = sub( prescale, prescaleOld );

#ifdef ISSUE_1836_replace_overflow_libcom

    mem_fx[0] = L_shr_sat( mem_fx[0], diff );

    move32();

    mem_fx[1] = L_shr_sat( mem_fx[1], diff );

    move32();

    mem_fx[2] = L_shr_sat( mem_fx[2], diff );

    move32();

    mem_fx[3] = L_shr_sat( mem_fx[3], diff );

#else

    mem_fx[0] = L_shr_o( mem_fx[0], diff, &Overflow );

    move32();

    mem_fx[1] = L_shr_o( mem_fx[1], diff, &Overflow );

    move32();

    mem_fx[2] = L_shr_o( mem_fx[2], diff, &Overflow );

    move32();

    mem_fx[3] = L_shr_o( mem_fx[3], diff, &Overflow );

#endif

    move32();

    move32();

    move32();

    move32();

    mem_fx[4] = L_deposit_l( prescale_current_frame );

    move32();
@@ -449,7 +453,7 @@ void hp20_fx_32_opt( 
    W_sum = W_add( W_sum, W_mult0_32_32( b2_fx, x2 ) );        /* b2*x0 */

    W_sum = W_add( W_sum, W_mult0_32_32( mem_fx[0], a2_fx ) ); /* y2*a2 */

    W_sum = W_add( W_sum, W_mult0_32_32( mem_fx[1], a1_fx ) ); /* y1*a1 */

    W_y2 = W_shl( W_sum, add( 1, HP20_FX_COEFF_SCALE ) );

    W_y2 = W_shl( W_sum, 1 + HP20_FX_COEFF_SCALE );

    signal_fx[0] = W_round64_L( W_shl( W_y2, prescale ) );

    move32();
@@ -459,7 +463,7 @@ void hp20_fx_32_opt( 
    W_sum = W_add( W_sum, W_mult0_32_32( b2_fx, x1 ) );                  /* b2*x0 */

    W_sum = W_add( W_sum, W_mult0_32_32( mem_fx[1], a2_fx ) );           /* y2*a2 */

    W_sum = W_add( W_sum, W_mult0_32_32( W_round64_L( W_y2 ), a1_fx ) ); /* y1*a1 */

    W_y1 = W_shl( W_sum, add( 1, HP20_FX_COEFF_SCALE ) );

    W_y1 = W_shl( W_sum, 1 + HP20_FX_COEFF_SCALE );

    signal_fx[1] = W_round64_L( W_shl( W_y1, prescale ) );

    move32();
@@ -477,7 +481,7 @@ void hp20_fx_32_opt( 
        W_sum = W_add( W_sum, W_mult0_32_32( b2_fx, x0 ) );                  /* b2*x0 */

        W_sum = W_add( W_sum, W_mult0_32_32( W_round64_L( W_y2 ), a2_fx ) ); /* y2*a2 */

        W_sum = W_add( W_sum, W_mult0_32_32( W_round64_L( W_y1 ), a1_fx ) ); /* y1*a1 */

        W_y0 = W_shl( W_sum, add( 1, HP20_FX_COEFF_SCALE ) );

        W_y0 = W_shl( W_sum, 1 + HP20_FX_COEFF_SCALE );



        signal_fx[i] = W_round64_L( W_shl( W_y0, prescale_current_frame ) );

        move32();