Commit ad64b267 authored by Arthur Tritthart's avatar Arthur Tritthart
Browse files

Author: Arthur Tritthart, FhG, 06-DEC-2024 

Changes for BASOP tuning (ticket 1009):

File lib_com/basop_util.c:
--------------------------
modified imult1616 to directly use i_mult, WMOPS weights reduced: 2 -> 1

File lib_com/ivas_tools.c:
--------------------------
Added an IF-conditionned branch for interleaved to linear format. This is the
way, the function is currently used, WMOPS weights reduced: 5 -> 2

File lib_com/tools_fx.c:
------------------------
Simplified set32_fx function, stripped use of L_deposit_l, WMOPS weights reduced 2 -> 1

File lib_rend/ivas_dirac_decorr_dec.c:
--------------------------------------
- use of is_zero_arr, stripped constant find_guarded_bits(2)
- strip offset computation for interleaved real/imag buffer
- tune AR filter loop for WMOPS
- fix and simplify 64-bit power computation loop
- tuned energy smoothing loops for WMOPS
- skip energy scaling, if q_shift equals zero
- strip offset computation for interleaved real/imag buffer

File lib_dec/ivas_mc_param_dec.c, ivas_mct_dec_mct_fx.c:
--------------------------------------------------------
- simplify zero checks for output synthesis
- replace div(x / 1) or div(x / 2) by shift ops
- simplified shifting output

Total WMOPS saving for bitstream stv714MC48c_128kbps.192/7_1_4: 164 WMops
parent e86c049a

lib_com/basop_util.c

+1 −1
Original line number Diff line number Diff line
@@ -1626,7 +1626,7 @@ Word16 findIndexOfMinWord32( Word32 *x, const Word16 len ) 
Word16 imult1616( Word16 x, Word16 y )

{

    assert( (int) x * (int) y < 32768 && (int) x * (int) y >= -32768 );

    return extract_l( L_mult0( x, y ) );

    return i_mult(x, y);

}



Word32 imult3216( Word32 x, Word16 y )

lib_com/ivas_tools.c

+17 −0
Original line number Diff line number Diff line
@@ -461,6 +461,23 @@ void v_add_inc_fx( 
)

{

    Word16 i;



    /* The use of this function is currently always for the interleaved input format, */

    /* that means, the following conditions are always true and thus obsolete.        */

    test();

    test();

    test();

    test();

    IF ((sub(x_inc, 2) == 0) && (sub(x2_inc, 2) == 0) && (sub(y_inc, 1) == 0) && (&x1[1] == &x2[0]) )

    {

        /* Interleaved input case, linear output */

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

        {

            y[i] = L_add( x1[2*i+0], x1[2*i+1] ); /*Qx*/

            move32();

        }

        return;

    }

    Word16 ix1 = 0;

    Word16 ix2 = 0;

    Word16 iy = 0;

lib_com/tools_fx.c

+5 −17
Original line number Diff line number Diff line
@@ -648,25 +648,13 @@ void set32_fx( 
    const Word16 N  /* i  : Lenght of the vector                */

)

{

    Word16 i, tmp;

    tmp = extract_l( a );

    IF( EQ_32( L_deposit_l( tmp ), a ) )

    {

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

        {

            y[i] = L_deposit_l( tmp );

            move32();

        }

    }

    ELSE

    {

    Word16 i;



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

    {

        y[i] = a;

        move32();

    }

    }



    return;

}

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

lib_dec/ivas_mc_param_dec.c

+5 −23
Original line number Diff line number Diff line
@@ -3786,38 +3786,20 @@ void ivas_param_mc_dec_render_fx( 
    slot_idx_start_cldfb_synth = 0;

    move16();



    Flag is_zero = 1;

    move32();

    FOR( j = 0; j < st_ivas->hParamMC->hMetadataPMC->nbands_coded; j++ )

    {

        is_zero = 1;

        move16();

        FOR( i = 0; i < hParamMC->h_output_synthesis_cov_state.mixing_matrix_len; i++ )

        Flag is_zero = is_zero_arr( hParamMC->h_output_synthesis_cov_state.mixing_matrix_fx[j], hParamMC->h_output_synthesis_cov_state.mixing_matrix_len );

        {

            IF( hParamMC->h_output_synthesis_cov_state.mixing_matrix_fx[j][i] != 0 )

            if ( is_zero != 0 )

            {

                is_zero = 0;

                hParamMC->h_output_synthesis_cov_state.mixing_matrix_res_exp[j] = 0;

                move16();

            }

        }

        IF( is_zero )

        {

            hParamMC->h_output_synthesis_cov_state.mixing_matrix_exp[j] = 0;

            move16();

        }

        is_zero = 1;

        move16();

        IF( LT_16( st_ivas->hParamMC->band_grouping[j], st_ivas->hParamMC->h_output_synthesis_params.max_band_decorr ) )

        {

            FOR( i = 0; i < hParamMC->h_output_synthesis_cov_state.mixing_matrix_res_len; i++ )

            {

                IF( NE_32( hParamMC->h_output_synthesis_cov_state.mixing_matrix_res_fx[j][i], 0 ) )

                {

                    is_zero = 0;

                    move16();

                }

            }

            IF( is_zero )

            is_zero = is_zero_arr( hParamMC->h_output_synthesis_cov_state.mixing_matrix_res_fx[j], hParamMC->h_output_synthesis_cov_state.mixing_matrix_res_len );

            if( is_zero != 0)

            {

                hParamMC->h_output_synthesis_cov_state.mixing_matrix_res_exp[j] = 0;

                move16();

lib_dec/ivas_mct_dec_mct_fx.c

+14 −2
Original line number Diff line number Diff line
@@ -316,6 +316,7 @@ void mctStereoIGF_dec_fx( 
            test();

            IF( NE_16( hMCT->hBlockData[b]->hStereoMdct->IGFStereoMode[k], SMDCT_DUAL_MONO ) || NE_16( hMCT->hBlockData[b]->hStereoMdct->mdct_stereo_mode[k], SMDCT_DUAL_MONO ) )

            {

#if 0

                tmp = BASOP_Util_Divide1616_Scale( sts[0]->hTcxCfg->tcx_coded_lines, nSubframes, &tmp_e );

                L_spec[0] = shr( tmp, add( 15, negate( tmp_e ) ) );

                move16();
@@ -325,6 +326,15 @@ void mctStereoIGF_dec_fx( 


                tmp = BASOP_Util_Divide1616_Scale( sts[0]->hTcxDec->L_frameTCX, nSubframes, &tmp_e );

                L_frameTCX_nSubframe = shr( tmp, add( 15, negate( tmp_e ) ) );

#else

                assert( nSubframes == 1 || nSubframes == 2 );

                /* Note: nSubframes is in limited range [1, 2] for this function */

                Word16 shr_div = sub( nSubframes, 1 ); /* 2 -> 1, 1 -> 0 */

                L_spec[0] = shr(sts[0]->hTcxCfg->tcx_coded_lines, shr_div);

                move16();

                L_frame_nSubframe = shr(sts[0]->L_frame, shr_div);

                L_frameTCX_nSubframe = shr( sts[0]->hTcxDec->L_frameTCX , shr_div);

#endif



                init_tcx_info_fx( sts[0], L_frame_nSubframe, L_frameTCX_nSubframe, k, bfi, &tcx_offset[0], &tcx_offsetFB[0], &L_frame[0], &L_frameTCX[0], &left_rect[0], &L_spec[0] );
@@ -334,14 +344,16 @@ void mctStereoIGF_dec_fx( 
                decoder_tcx_IGF_stereo_fx( sts, hMCT->hBlockData[b]->hStereoMdct, hMCT->hBlockData[b]->mask, p_x, p_x_e, p_x_len, L_frame[0], left_rect[0], k, bfi, 1 /* MCT_flag */ );



                // Shifting output with variable exponent back to Q12

                Word16 shr_k = sub( 31 - Q12, p_x_e[0][k] );

                FOR( Word16 i = 0; i < p_x_len[0][k]; i++ )

                {

                    p_x[0][k][i] = L_shr( p_x[0][k][i], sub( 31 - Q12, p_x_e[0][k] ) );

                    p_x[0][k][i] = L_shr( p_x[0][k][i], shr_k );

                    move32();

                }

                shr_k = sub( 31 - Q12, p_x_e[1][k] );

                FOR( Word16 i = 0; i < p_x_len[1][k]; i++ )

                {

                    p_x[1][k][i] = L_shr( p_x[1][k][i], sub( 31 - Q12, p_x_e[1][k] ) );

                    p_x[1][k][i] = L_shr( p_x[1][k][i], shr_k );

                    move32();

                }

            }