Commit ec04ed46 authored by vaclav's avatar vaclav
Browse files

Merge branch 'basop-2308-optimisation-of-find-target' into 'main' 

basop-2308-optimisation-of-find-target [non-be][rend-non-be][split-non-be][allow-regression]

See merge request !2689
parents 34de2668 29fde41c

lib_com/options.h

+1 −0
Original line number Diff line number Diff line
@@ -159,6 +159,7 @@ 
#define NONBE_OPT_2239_IVAS_FILTER_PROCESS  /* Dolby: Issue 2239, optimize ivas_filter_process_fx. */

#define NONBE_OPT_2193_EIG2X2               /* Dolby: Issue 2193, optimize eig2x2_fx. */

#define BE_FIX_2240_COMPUTE_COV_MTC_FX_FAST /* FhG: Speeds up covariance calculation e.g. 60 WMOPS for encoding -mc 7_1_4 24400 48 */

#define OPT_2308_FIND_TARGET                /* VA: Issue 2308, Speeds up computation and improve accuracy of the impulse response */



/* #################### End BASOP optimization switches ############################ */

lib_enc/find_tar_fx.c

+46 −0
Original line number Diff line number Diff line
@@ -173,7 +173,12 @@ void find_targets_ivas_fx( 
{

    Word16 i;

    Word16 temp[M + 6 * L_SUBFR]; /* error of quantization */

#ifndef OPT_2308_FIND_TARGET

    Word16 scale, scaleq, j, d, s, s2, tmp;

#else

    Word16 scale, scaleq, j, d, s, tmp;

    Word32 s2;

#endif

    Word16 Aqs[M + 1];

    Word32 h1_32[6 * L_SUBFR];

    Word16 sf;
@@ -233,17 +238,27 @@ void find_targets_ivas_fx( 
    {

        Copy( p_Aq, Aqs, M + 1 ); /* Q12 */

        s = add( scaleq, 1 );

#ifndef OPT_2308_FIND_TARGET

        s2 = shr( 16384, d );

#else

        s2 = L_shr( ( 16384 << 16 ), d );

#endif

    }

    ELSE

    {

        Copy_Scale_sig( p_Aq, Aqs, M + 1, d ); /* Q12 */

        s = add( scale, 1 );

#ifndef OPT_2308_FIND_TARGET

        s2 = 16384;

        move16();

#else

        s2 = ( 16384 << 16 );

        move32();

#endif

    }



    set32_fx( h1_32, 0, L_subfr );

#ifndef OPT_2308_FIND_TARGET

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

    {

        Ltmp64 = W_mult_16_16( Ap[i], s2 ); /* Q27 */
@@ -273,7 +288,38 @@ void find_targets_ivas_fx( 
        h1_32[i] = W_extract_l( Ltmp64 ); /* Q27 */

        move32();

    }

#else

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

    {

        Ltmp64 = W_mult_32_16( s2, Ap[i] ); /* Q27 */

        FOR( j = 1; j <= i; j++ )

        {

            Ltmp64 = W_msu_32_16( Ltmp64, L_shl_sat( h1_32[i - j], s ), Aqs[j] ); /* Q27 */

        }

        h1_32[i] = W_sat_m( Ltmp64 ); /* Q27 */

        move32();

    }



    Ltmp64 = W_mult_32_16( s2, Ap[i] ); /* Q27 */

    FOR( j = 1; j <= M; j++ )

    {

        Ltmp64 = W_msu_32_16( Ltmp64, L_shl_sat( h1_32[i - j], s ), Aqs[j] ); /* Q27 */

    }

    h1_32[M] = W_sat_m( Ltmp64 ); /* Q27 */

    move32();



    FOR( i = M + 1; i < L_subfr; i++ )

    {

        Ltmp64 = W_msu_32_16( 0, L_shl_sat( h1_32[i - 1], s ), Aqs[1] ); /* Q27 */

        FOR( j = 2; j <= M; j++ )

        {

            Ltmp64 = W_msu_32_16( Ltmp64, L_shl_sat( h1_32[i - j], s ), Aqs[j] ); /* Q27 */

        }

        h1_32[i] = W_sat_m( Ltmp64 ); /* Q27 */

        move32();

    }



#endif

    sf = sub( L_norm_arr( h1_32, L_subfr ), 1 );

    Copy_Scale_sig32_16( h1_32, h1, L_subfr, sf ); // Q11 + sf