Commit 4e41d8e1 authored by Sandesh Venkatesh's avatar Sandesh Venkatesh
Browse files

Multi channel bit-exact optimizations - 3 

Residu3_fx, dotp_fixed_32, SQ_gain_estimate_stereo_fx, dotp_fixed_o,
IGF_getCrest_new_fx updated. Around 4 WMOPs reduction observed
parent bf9ba511

lib_com/options.h

+1 −0
Original line number Diff line number Diff line
@@ -79,6 +79,7 @@ 




/* Note: each compile switch (FIX_1101_...) is independent from the other ones */

#define OPT_MCT_ENC_V2_BE

#define OPT_MCH_DEC_V1_NBE

#define OPT_MCT_ENC_48KB_NBE

#define OPT_MCH_DEC_V1_BE

lib_com/residu_fx.c

+1 −1
Original line number Diff line number Diff line
@@ -156,7 +156,7 @@ void Residu3_fx( 
    move16();

#endif

    q = add( norm_s( a[0] ), 1 );

    IF( shift != 0 )

    if ( shift != 0 )

    {

        q = add( q, shift );

    }

lib_com/tools.c

+8 −0
Original line number Diff line number Diff line
@@ -688,7 +688,11 @@ Word32 dotp_fixed_o( 
    move16();

    test();

    test();

#ifdef OPT_MCT_ENC_V2_BE

    FOR( ; ( suma > MAX_32 ) || ( suma < MIN_32 ) || ( *res_q > 31 ); )

#else

    WHILE( GT_64( suma, MAX_32 ) || LT_64( suma, MIN_32 ) || GT_16( *res_q, 31 ) )

#endif

    {

        suma = W_shr( suma, 1 );

        *res_q = sub( *res_q, 1 );
@@ -718,7 +722,11 @@ Word32 dotp_fixed_32( 
    move16();

    test();

    test();

#ifdef OPT_MCT_ENC_V2_BE

    FOR( ; ( suma > MAX_32 ) || ( suma < MIN_32 ) || ( *res_q > 31 ); )

#else

    WHILE( GT_64( suma, MAX_32 ) || LT_64( suma, MIN_32 ) || GT_16( *res_q, 31 ) )

#endif

    {

        suma = W_shr( suma, 1 );

        *res_q = sub( *res_q, 1 );

lib_enc/igf_enc.c

+1 −1
Original line number Diff line number Diff line
@@ -137,7 +137,7 @@ static Word16 IGF_getCrest_new_fx( 
        move16();

        x_eff = L_add( x_eff, L_mult0( x, x ) );



        IF( GT_16( x, x_max ) )

        if ( GT_16( x, x_max ) )

        {

            x_max = x;

            move16();

lib_enc/ivas_stereo_mdct_stereo_enc_fx.c

+20 −2
Original line number Diff line number Diff line
@@ -882,7 +882,7 @@ static Word32 SQ_gain_estimate_stereo_fx( // e_res 
        move32();

        en_fx[q] = Mpy_32_16_1( L_add( L_shl( e_ener, Q25 ), en_fx[q] ), 9864 /* log10(2) in Q15 */ ); // Q(25)

        move32();

        i = add( i, 4 );

        i += 4;

    }

    i = 0;

    move16();
@@ -901,7 +901,7 @@ static Word32 SQ_gain_estimate_stereo_fx( // e_res 
        move32();

        en_fx[q] = Mpy_32_16_1( L_add( L_shl( e_ener, Q25 ), en_fx[q] ), 9864 /* log10(2) in Q15 */ ); // Q(25)

        move32();

        i = add( i, 4 );

        i += 4;

    }



    /* SQ scale: 4 bits / 6 dB per quadruple */
@@ -911,6 +911,9 @@ static Word32 SQ_gain_estimate_stereo_fx( // e_res 
    offset_fx = fac_fx; /* Q25 */

    move32();



#ifdef OPT_MCT_ENC_V2_BE

    Word64 target_64_fx = W_shl( target_fx, 7 ); // Q25

#endif

    /* find offset (0 to 128 dB with step of 0.125dB) */

    FOR( iter = 0; iter < 10; iter++ )

    {
@@ -923,6 +926,20 @@ static Word32 SQ_gain_estimate_stereo_fx( // e_res 
        {

            tmp_32 = L_sub( en_fx[i], offset_fx ); /* Q25 */



#ifdef OPT_MCT_ENC_V2_BE

            /* avoid SV with 1 bin of amp < 0.5f */

            if ( GT_32( tmp_32, 10066329 /*0.3 Q25*/ ) )

            {

                W_tmp = W_add( W_tmp, tmp_32 ); /* Q25 */

            }

            /* if ener is above target -> break and increase offset */

            IF( GT_64( W_tmp, target_64_fx ) )

            {

                offset_fx = L_add( offset_fx, fac_fx ); /* Q25 */

                BREAK;

            }

#else



            /* avoid SV with 1 bin of amp < 0.5f */

            IF( GT_32( tmp_32, 10066329 /*0.3 Q25*/ ) )

            {
@@ -935,6 +952,7 @@ static Word32 SQ_gain_estimate_stereo_fx( // e_res 
                    BREAK;

                }

            }

#endif

        }

    }