Commit ac454fc2 authored by Sandesh Venkatesh's avatar Sandesh Venkatesh
Browse files

Merge branch 'stereo_fmt_optimizations' into 'main' 

Optimization of stereo decode path 32kHz @32kbps

See merge request !1205
parents fc98a680 defa4937

lib_com/fft_fx.c

+334 −0
Original line number Diff line number Diff line
@@ -4727,6 +4727,16 @@ static void fft_len16( 
    cmplx t[4];

    cmplx y[16];



#ifdef OPT_STEREO_32KBPS_V1

    s[0] = x[0]; // Qx

    move64();

    s[1] = x[4]; // Qx

    move64();

    s[2] = x[8]; // Qx

    move64();

    s[3] = x[12]; // Qx

    move64();

#else  /* OPT_STEREO_32KBPS_V1 */

    s[0] = CL_shr( x[0], SCALEFACTOR16 ); // Qx

    move64();

    s[1] = CL_shr( x[4], SCALEFACTOR16 ); // Qx
@@ -4735,6 +4745,7 @@ static void fft_len16( 
    move64();

    s[3] = CL_shr( x[12], SCALEFACTOR16 ); // Qx

    move64();

#endif /* OPT_STEREO_32KBPS_V1 */



    t[0] = CL_add( s[0], s[2] );

    move64();
@@ -4754,6 +4765,16 @@ static void fft_len16( 
    y[3] = CL_add( t[1], t[3] );

    move64();



#ifdef OPT_STEREO_32KBPS_V1

    s[0] = x[1]; // Qx

    move64();

    s[1] = x[5]; // Qx

    move64();

    s[2] = x[9]; // Qx

    move64();

    s[3] = x[13]; // Qx

    move64();

#else  /* OPT_STEREO_32KBPS_V1 */

    s[0] = CL_shr( x[1], SCALEFACTOR16 ); // Qx

    move64();

    s[1] = CL_shr( x[5], SCALEFACTOR16 ); // Qx
@@ -4762,6 +4783,7 @@ static void fft_len16( 
    move64();

    s[3] = CL_shr( x[13], SCALEFACTOR16 ); // Qx

    move64();

#endif /* OPT_STEREO_32KBPS_V1 */



    t[0] = CL_add( s[0], s[2] );

    move64();
@@ -4781,6 +4803,16 @@ static void fft_len16( 
    y[7] = CL_add( t[1], t[3] );

    move64();



#ifdef OPT_STEREO_32KBPS_V1

    s[0] = x[2]; // Qx

    move64();

    s[1] = x[6]; // Qx

    move64();

    s[2] = x[10]; // Qx

    move64();

    s[3] = x[14]; // Qx

    move64();

#else  /* OPT_STEREO_32KBPS_V1 */

    s[0] = CL_shr( x[2], SCALEFACTOR16 ); // Qx

    move64();

    s[1] = CL_shr( x[6], SCALEFACTOR16 ); // Qx
@@ -4789,6 +4821,7 @@ static void fft_len16( 
    move64();

    s[3] = CL_shr( x[14], SCALEFACTOR16 ); // Qx

    move64();

#endif /* OPT_STEREO_32KBPS_V1 */



    t[0] = CL_add( s[0], s[2] );

    move64();
@@ -4810,6 +4843,16 @@ static void fft_len16( 
    y[11] = CL_add( t[1], t[3] );

    move64();



#ifdef OPT_STEREO_32KBPS_V1

    s[0] = x[3]; // Qx

    move64();

    s[1] = x[7]; // Qx

    move64();

    s[2] = x[11]; // Qx

    move64();

    s[3] = x[15]; // Qx

    move64();

#else  /* OPT_STEREO_32KBPS_V1 */

    s[0] = CL_shr( x[3], SCALEFACTOR16 ); // Qx

    move64();

    s[1] = CL_shr( x[7], SCALEFACTOR16 ); // Qx
@@ -4818,6 +4861,7 @@ static void fft_len16( 
    move64();

    s[3] = CL_shr( x[15], SCALEFACTOR16 ); // Qx

    move64();

#endif /* OPT_STEREO_32KBPS_V1 */



    t[0] = CL_add( s[0], s[2] );

    move64();
@@ -4978,6 +5022,18 @@ static void fft_len20_fx( 
    cmplx tt[4];

    cmplx y[20];



#ifdef OPT_STEREO_32KBPS_V1

    xx[0] = x[0]; // Qx

    move64();

    xx[1] = x[16]; // Qx

    move64();

    xx[2] = x[12]; // Qx

    move64();

    xx[3] = x[8]; // Qx

    move64();

    xx[4] = x[4]; // Qx

    move64();

#else  /* OPT_STEREO_32KBPS_V1 */

    xx[0] = CL_shr( x[0], SCALEFACTOR20 ); // Qx

    move64();

    xx[1] = CL_shr( x[16], SCALEFACTOR20 ); // Qx
@@ -4988,6 +5044,7 @@ static void fft_len20_fx( 
    move64();

    xx[4] = CL_shr( x[4], SCALEFACTOR20 ); // Qx

    move64();

#endif /* OPT_STEREO_32KBPS_V1 */



    s[0] = CL_add( xx[1], xx[4] );

    move64();
@@ -5023,6 +5080,18 @@ static void fft_len20_fx( 
    y[12] = CL_msu_j( s[2], s[3] );

    move64();



#ifdef OPT_STEREO_32KBPS_V1

    xx[0] = x[5];

    move64();

    xx[1] = x[1];

    move64();

    xx[2] = x[17];

    move64();

    xx[3] = x[13];

    move64();

    xx[4] = x[9];

    move64();

#else  /* OPT_STEREO_32KBPS_V1 */

    xx[0] = CL_shr( x[5], SCALEFACTOR20 );

    move64();

    xx[1] = CL_shr( x[1], SCALEFACTOR20 );
@@ -5033,6 +5102,7 @@ static void fft_len20_fx( 
    move64();

    xx[4] = CL_shr( x[9], SCALEFACTOR20 );

    move64();

#endif /* OPT_STEREO_32KBPS_V1 */



    s[0] = CL_add( xx[1], xx[4] );

    move64();
@@ -5068,6 +5138,18 @@ static void fft_len20_fx( 
    y[13] = CL_msu_j( s[2], s[3] );

    move64();



#ifdef OPT_STEREO_32KBPS_V1

    xx[0] = x[10];

    move64();

    xx[1] = x[6];

    move64();

    xx[2] = x[2];

    move64();

    xx[3] = x[18];

    move64();

    xx[4] = x[14];

    move64();

#else  /* OPT_STEREO_32KBPS_V1 */

    xx[0] = CL_shr( x[10], SCALEFACTOR20 );

    move64();

    xx[1] = CL_shr( x[6], SCALEFACTOR20 );
@@ -5078,6 +5160,7 @@ static void fft_len20_fx( 
    move64();

    xx[4] = CL_shr( x[14], SCALEFACTOR20 );

    move64();

#endif /* OPT_STEREO_32KBPS_V1 */



    s[0] = CL_add( xx[1], xx[4] );

    move64();
@@ -5113,6 +5196,18 @@ static void fft_len20_fx( 
    y[14] = CL_msu_j( s[2], s[3] );

    move64();



#ifdef OPT_STEREO_32KBPS_V1

    xx[0] = x[15];

    move64();

    xx[1] = x[11];

    move64();

    xx[2] = x[7];

    move64();

    xx[3] = x[3];

    move64();

    xx[4] = x[19];

    move64();

#else  /* OPT_STEREO_32KBPS_V1 */

    xx[0] = CL_shr( x[15], SCALEFACTOR20 );

    move64();

    xx[1] = CL_shr( x[11], SCALEFACTOR20 );
@@ -5123,6 +5218,7 @@ static void fft_len20_fx( 
    move64();

    xx[4] = CL_shr( x[19], SCALEFACTOR20 );

    move64();

#endif /* OPT_STEREO_32KBPS_V1 */



    s[0] = CL_add( xx[1], xx[4] );

    move64();
@@ -6501,6 +6597,239 @@ static void fft_lenN( 
            cmplx s[8];

            cmplx y[8];



#ifdef OPT_STEREO_32KBPS_V1

            y[1] = xx[1 * dim1];

            move64();

            y[2] = xx[2 * dim1];

            move64();

            y[3] = xx[3 * dim1];

            move64();

            y[4] = xx[4 * dim1];

            move64();

            y[5] = xx[5 * dim1];

            move64();

            y[6] = xx[6 * dim1];

            move64();

            y[7] = xx[7 * dim1];

            move64();



            test();

            test();

            IF( EQ_16( dim1, 8 ) || EQ_16( dim1, 16 ) || EQ_16( dim1, 32 ) )

            {

                FOR( i = 0; i < dim1; i++ )

                {

                    {

                        y[0] = xx[i];

                        move64();

                    };

                    IF( i > 0 )

                    {

                        {

                            y[1] = CL_mac_j( CL_scale( xx[( i + ( 1 * dim1 ) )], W[( ( ( sc * i ) + ( ( sc * 1 ) * dim1 ) ) - Woff )] ),

                                             CL_scale( xx[( i + ( 1 * dim1 ) )], W[( ( ( ( sc * i ) + ( ( sc * 1 ) * dim1 ) ) + 1 ) - Woff )] ) ); // Qx

                            move64();

                        };

                        {

                            y[2] = CL_mac_j( CL_scale( xx[( i + ( 2 * dim1 ) )], W[( ( ( sc * i ) + ( ( sc * 2 ) * dim1 ) ) - Woff )] ),

                                             CL_scale( xx[( i + ( 2 * dim1 ) )], W[( ( ( ( sc * i ) + ( ( sc * 2 ) * dim1 ) ) + 1 ) - Woff )] ) ); // Qx

                            move64();

                        };

                        {

                            y[3] = CL_mac_j( CL_scale( xx[( i + ( 3 * dim1 ) )], W[( ( ( sc * i ) + ( ( sc * 3 ) * dim1 ) ) - Woff )] ),

                                             CL_scale( xx[( i + ( 3 * dim1 ) )], W[( ( ( ( sc * i ) + ( ( sc * 3 ) * dim1 ) ) + 1 ) - Woff )] ) ); // Qx

                            move64();

                        };

                        {

                            y[4] = CL_mac_j( CL_scale( xx[( i + ( 4 * dim1 ) )], W[( ( ( sc * i ) + ( ( sc * 4 ) * dim1 ) ) - Woff )] ),

                                             CL_scale( xx[( i + ( 4 * dim1 ) )], W[( ( ( ( sc * i ) + ( ( sc * 4 ) * dim1 ) ) + 1 ) - Woff )] ) ); // Qx

                            move64();

                        };

                        {

                            y[5] = CL_mac_j( CL_scale( xx[( i + ( 5 * dim1 ) )], W[( ( ( sc * i ) + ( ( sc * 5 ) * dim1 ) ) - Woff )] ),

                                             CL_scale( xx[( i + ( 5 * dim1 ) )], W[( ( ( ( sc * i ) + ( ( sc * 5 ) * dim1 ) ) + 1 ) - Woff )] ) ); // Qx

                            move64();

                        };

                        {

                            y[6] = CL_mac_j( CL_scale( xx[( i + ( 6 * dim1 ) )], W[( ( ( sc * i ) + ( ( sc * 6 ) * dim1 ) ) - Woff )] ),

                                             CL_scale( xx[( i + ( 6 * dim1 ) )], W[( ( ( ( sc * i ) + ( ( sc * 6 ) * dim1 ) ) + 1 ) - Woff )] ) ); // Qx

                            move64();

                        };

                        {

                            y[7] = CL_mac_j( CL_scale( xx[( i + ( 7 * dim1 ) )], W[( ( ( sc * i ) + ( ( sc * 7 ) * dim1 ) ) - Woff )] ),

                                             CL_scale( xx[( i + ( 7 * dim1 ) )], W[( ( ( ( sc * i ) + ( ( sc * 7 ) * dim1 ) ) + 1 ) - Woff )] ) ); // Qx

                            move64();

                        };

                    }



                    t[0] = CL_add( y[0], y[4] );

                    move64();

                    t[1] = CL_sub( y[0], y[4] );

                    move64();

                    t[2] = CL_add( y[1], y[5] );

                    move64();

                    t[3] = CL_sub( y[1], y[5] );

                    move64();

                    t[4] = CL_add( y[2], y[6] );

                    move64();

                    t[5] = CL_sub( y[2], y[6] );

                    move64();

                    t[6] = CL_add( y[3], y[7] );

                    move64();

                    t[7] = CL_sub( y[3], y[7] );

                    move64();



                    s[0] = CL_add( t[0], t[4] );

                    move64();

                    s[2] = CL_sub( t[0], t[4] );

                    move64();

                    s[4] = CL_mac_j( t[1], t[5] );

                    move64();

                    s[5] = CL_msu_j( t[1], t[5] );

                    move64();

                    s[1] = CL_add( t[2], t[6] );

                    move64();

                    s[3] = CL_swap_real_imag( CL_sub( CL_conjugate( t[2] ), CL_conjugate( t[6] ) ) );

                    move64();



                    t[0] = CL_swap_real_imag( CL_add( t[3], t[7] ) );

                    move64();

                    t[1] = CL_sub( t[3], t[7] );

                    move64();



                    s[6] = CL_scale( CL_add( CL_conjugate( t[0] ), t[1] ), FFT_C81 ); // Qx

                    move64();

                    s[7] = CL_scale( CL_sub( t[0], CL_conjugate( t[1] ) ), FFT_C81 ); // Qx

                    move64();

                    s[7] = CL_conjugate( s[7] );

                    move64();



                    x[i] = CL_add( s[0], s[1] );

                    move64();

                    x[( i + ( 1 * dim1 ) )] = CL_add( s[5], s[6] );

                    move64();

                    x[( i + ( 2 * dim1 ) )] = CL_sub( s[2], s[3] );

                    move64();

                    x[( i + ( 3 * dim1 ) )] = CL_add( s[4], s[7] );

                    move64();

                    x[( i + ( 4 * dim1 ) )] = CL_sub( s[0], s[1] );

                    move64();

                    x[( i + ( 5 * dim1 ) )] = CL_sub( s[5], s[6] );

                    move64();

                    x[( i + ( 6 * dim1 ) )] = CL_add( s[2], s[3] );

                    move64();

                    x[( i + ( 7 * dim1 ) )] = CL_sub( s[4], s[7] );

                    move64();

                }

            }

            ELSE

            {

                FOR( i = 0; i < dim1; i++ )

                {

                    {

                        y[0] = xx[i];

                        move64();

                    };

                    IF( i > 0 )

                    {

                        {

                            y[1] = CL_mac_j( CL_scale( xx[( i + ( 1 * dim1 ) )], W[( ( ( sc * i ) + ( ( ( sc * 1 ) * dim1 ) << 1 ) ) - Woff )] ),

                                             CL_scale( xx[( i + ( 1 * dim1 ) )], W[( ( ( ( sc * i ) + ( ( ( sc * 1 ) * dim1 ) << 1 ) ) + 1 ) - Woff )] ) ); // Qx

                            move64();

                        };

                        {

                            y[2] = CL_mac_j( CL_scale( xx[( i + ( 2 * dim1 ) )], W[( ( ( sc * i ) + ( ( ( sc * 2 ) * dim1 ) << 1 ) ) - Woff )] ),

                                             CL_scale( xx[( i + ( 2 * dim1 ) )], W[( ( ( ( sc * i ) + ( ( ( sc * 2 ) * dim1 ) << 1 ) ) + 1 ) - Woff )] ) ); // Qx

                            move64();

                        };

                        {

                            y[3] = CL_mac_j( CL_scale( xx[( i + ( 3 * dim1 ) )], W[( ( ( sc * i ) + ( ( ( sc * 3 ) * dim1 ) << 1 ) ) - Woff )] ),

                                             CL_scale( xx[( i + ( 3 * dim1 ) )], W[( ( ( ( sc * i ) + ( ( ( sc * 3 ) * dim1 ) << 1 ) ) + 1 ) - Woff )] ) ); // Qx

                            move64();

                        };

                        {

                            y[4] = CL_mac_j( CL_scale( xx[( i + ( 4 * dim1 ) )], W[( ( ( sc * i ) + ( ( ( sc * 4 ) * dim1 ) << 1 ) ) - Woff )] ),

                                             CL_scale( xx[( i + ( 4 * dim1 ) )], W[( ( ( ( sc * i ) + ( ( ( sc * 4 ) * dim1 ) << 1 ) ) + 1 ) - Woff )] ) ); // Qx

                            move64();

                        };

                        {

                            y[5] = CL_mac_j( CL_scale( xx[( i + ( 5 * dim1 ) )], W[( ( ( sc * i ) + ( ( ( sc * 5 ) * dim1 ) << 1 ) ) - Woff )] ),

                                             CL_scale( xx[( i + ( 5 * dim1 ) )], W[( ( ( ( sc * i ) + ( ( ( sc * 5 ) * dim1 ) << 1 ) ) + 1 ) - Woff )] ) ); // Qx

                            move64();

                        };

                        {

                            y[6] = CL_mac_j( CL_scale( xx[( i + ( 6 * dim1 ) )], W[( ( ( sc * i ) + ( ( ( sc * 6 ) * dim1 ) << 1 ) ) - Woff )] ),

                                             CL_scale( xx[( i + ( 6 * dim1 ) )], W[( ( ( ( sc * i ) + ( ( ( sc * 6 ) * dim1 ) << 1 ) ) + 1 ) - Woff )] ) ); // Qx

                            move64();

                        };

                        {

                            y[7] = CL_mac_j( CL_scale( xx[( i + ( 7 * dim1 ) )], W[( ( ( sc * i ) + ( ( ( sc * 7 ) * dim1 ) << 1 ) ) - Woff )] ),

                                             CL_scale( xx[( i + ( 7 * dim1 ) )], W[( ( ( ( sc * i ) + ( ( ( sc * 7 ) * dim1 ) << 1 ) ) + 1 ) - Woff )] ) ); // Qx

                            move64();

                        };

                    }



                    t[0] = CL_add( y[0], y[4] );

                    move64();

                    t[1] = CL_sub( y[0], y[4] );

                    move64();

                    t[2] = CL_add( y[1], y[5] );

                    move64();

                    t[3] = CL_sub( y[1], y[5] );

                    move64();

                    t[4] = CL_add( y[2], y[6] );

                    move64();

                    t[5] = CL_sub( y[2], y[6] );

                    move64();

                    t[6] = CL_add( y[3], y[7] );

                    move64();

                    t[7] = CL_sub( y[3], y[7] );

                    move64();



                    s[0] = CL_add( t[0], t[4] );

                    move64();

                    s[2] = CL_sub( t[0], t[4] );

                    move64();

                    s[4] = CL_mac_j( t[1], t[5] );

                    move64();

                    s[5] = CL_msu_j( t[1], t[5] );

                    move64();

                    s[1] = CL_add( t[2], t[6] );

                    move64();

                    s[3] = CL_swap_real_imag( CL_sub( CL_conjugate( t[2] ), CL_conjugate( t[6] ) ) );

                    move64();



                    t[0] = CL_swap_real_imag( CL_add( t[3], t[7] ) );

                    move64();

                    t[1] = CL_sub( t[3], t[7] );

                    move64();



                    s[6] = CL_scale( CL_add( CL_conjugate( t[0] ), t[1] ), FFT_C81 ); // Qx

                    move64();

                    s[7] = CL_scale( CL_sub( t[0], CL_conjugate( t[1] ) ), FFT_C81 ); // Qx

                    move64();

                    s[7] = CL_conjugate( s[7] );

                    move64();



                    x[i] = CL_add( s[0], s[1] );

                    move64();

                    x[( i + ( 1 * dim1 ) )] = CL_add( s[5], s[6] );

                    move64();

                    x[( i + ( 2 * dim1 ) )] = CL_sub( s[2], s[3] );

                    move64();

                    x[( i + ( 3 * dim1 ) )] = CL_add( s[4], s[7] );

                    move64();

                    x[( i + ( 4 * dim1 ) )] = CL_sub( s[0], s[1] );

                    move64();

                    x[( i + ( 5 * dim1 ) )] = CL_sub( s[5], s[6] );

                    move64();

                    x[( i + ( 6 * dim1 ) )] = CL_add( s[2], s[3] );

                    move64();

                    x[( i + ( 7 * dim1 ) )] = CL_sub( s[4], s[7] );

                    move64();

                }

            }

#else  /* OPT_STEREO_32KBPS_V1 */

            test();

            test();

            test();
@@ -6781,6 +7110,7 @@ static void fft_lenN( 
                    move64();

                }

            }

#endif /* OPT_STEREO_32KBPS_V1 */

            BREAK;

        }
@@ -7173,7 +7503,11 @@ void rfft_fx( 
                move32();

                x[( length - ( i << 1 ) )] = Mpy_32_16_1( L_add( t1, t3 ), 16384 /*0.5.Q15*/ );

                move32();

#ifdef OPT_STEREO_32KBPS_V1

                x[( ( length - ( i << 1 ) ) + 1 )] = Mpy_32_16_1( L_add( t2, t4 ), -16384 /*0.5.Q15*/ );

#else  /* OPT_STEREO_32KBPS_V1 */

                x[( ( length - ( i << 1 ) ) + 1 )] = Mpy_32_16_1( L_negate( L_add( t2, t4 ) ), 16384 /*0.5.Q15*/ );

#endif /* OPT_STEREO_32KBPS_V1 */

                move32();

            }

lib_com/hp50_fx.c

+66 −0
Original line number Diff line number Diff line
@@ -469,8 +469,14 @@ void hp20_fx_32( 
{

    Word16 i;

    Word32 a1_fx, a2_fx, b1_fx, b2_fx;

#ifdef OPT_STEREO_32KBPS_V1

    Word16 Qy1, Qy2, Qmin;

    Word64 y0_fx64, y1_fx64, y2_fx64;

    Word32 x0, x1, x2;

#else  /* OPT_STEREO_32KBPS_V1 */

    Word16 Qx0, Qx1, Qx2, Qy1, Qprev_y1, Qy2, Qprev_y2, Qmin;

    Word64 x0_fx64, x1_fx64, x2_fx64, y0_fx64, y1_fx64, y2_fx64, R1, R2, R3, R4, R5;

#endif /* OPT_STEREO_32KBPS_V1 */



    IF( EQ_32( Fs, 8000 ) )

    {
@@ -521,15 +527,64 @@ void hp20_fx_32( 
    move32();

    move32();



#ifdef OPT_STEREO_32KBPS_V1

    y1_fx64 = W_add( W_deposit32_l( mem_fx[0] ), W_deposit32_h( mem_fx[1] ) );

    y2_fx64 = W_add( W_deposit32_l( mem_fx[2] ), W_deposit32_h( mem_fx[3] ) );



    x0 = mem_fx[4];

    move32();

    x1 = mem_fx[5];

    move32();

#else  /* OPT_STEREO_32KBPS_V1 */

    Qprev_y1 = extract_l( mem_fx[4] );

    Qprev_y2 = extract_l( mem_fx[5] );

    y1_fx64 = W_deposit32_l( mem_fx[0] );

    y2_fx64 = W_deposit32_l( mem_fx[1] );

    x0_fx64 = W_deposit32_l( mem_fx[2] );

    x1_fx64 = W_deposit32_l( mem_fx[3] );

#endif /* OPT_STEREO_32KBPS_V1 */



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

    {

#ifdef OPT_STEREO_32KBPS_V1

        x2 = x1;

        move32();

        x1 = x0;

        move32();

        x0 = signal_fx[i];

        move32();



        Qy1 = W_norm( y1_fx64 );

        if ( y1_fx64 == 0 )

        {

            Qy1 = 62;

            move16();

        }



        Qy2 = W_norm( y2_fx64 );

        if ( y2_fx64 == 0 )

        {

            Qy2 = 62;

            move16();

        }



        Qmin = s_min( Qy1, Qy2 );



        Qmin = sub( Qmin, 34 );



        y0_fx64 = W_mac_32_32( W_mult_32_32( W_shl_sat_l( y1_fx64, Qmin ), a1_fx ), W_shl_sat_l( y2_fx64, Qmin ), a2_fx ); // Qmin + Q29 + Q30 + 1



        Word64 temp = W_mac_32_32( W_mac_32_32( W_mult_32_32( x2, b2_fx ), x1, b1_fx ), x0, b2_fx ); // Q30

        Word64 y0_fx = W_shr( y0_fx64, add( Qmin, Q30 ) );                                           // Q30

        y0_fx64 = W_add( temp, y0_fx );                                                              // Q30

        signal_fx[i] = W_shl_sat_l( y0_fx64, -Q30 );

        move32();



        y2_fx64 = y1_fx64;

        move64();

        y1_fx64 = y0_fx64;

        move64();

#else  /* OPT_STEREO_32KBPS_V1 */

        x2_fx64 = x1_fx64;

        move64();

        x1_fx64 = x0_fx64;
@@ -611,8 +666,17 @@ void hp20_fx_32( 
        move64();

        move16();

        move16();

#endif /* OPT_STEREO_32KBPS_V1 */

    }



#ifdef OPT_STEREO_32KBPS_V1

    mem_fx[0] = W_extract_l( y1_fx64 );

    mem_fx[1] = W_extract_h( y1_fx64 );

    mem_fx[2] = W_extract_l( y2_fx64 );

    mem_fx[3] = W_extract_h( y2_fx64 );

    mem_fx[4] = x0;

    mem_fx[5] = x1;

#else  /* OPT_STEREO_32KBPS_V1 */

    Qy1 = W_norm( y1_fx64 );

    test();

    IF( y1_fx64 != 0 && LT_16( Qy1, 32 ) )
@@ -635,6 +699,8 @@ void hp20_fx_32( 
    mem_fx[3] = W_extract_l( x1_fx64 );

    mem_fx[4] = Qprev_y1;

    mem_fx[5] = Qprev_y2;

#endif /* OPT_STEREO_32KBPS_V1 */



    move32();

    move32();

    move32();

lib_com/modif_fs.c

+94 −0

File changed.

Preview size limit exceeded, changes collapsed.

lib_com/mslvq_com_fx.c

+50 −0
Original line number Diff line number Diff line
@@ -122,9 +122,33 @@ void init_lvq_fx( 
)

{

    Word16 i, j;

#ifdef OPT_STEREO_32KBPS_V1

    Word16 k;

#endif /* OPT_STEREO_32KBPS_V1 */

    /* safety-net mode */

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

    {

#ifdef OPT_STEREO_32KBPS_V1

        FOR( ( j = 0, k = 0 ); j < MAX_NO_SCALES; ( j++, k++ ) )

        {

            if ( ( no_lead_fx[i][j] <= 0 ) )

            {

                j = MAX_NO_SCALES;

            }

        }

        no_scales[i][0] = k;

        move16();



        FOR( k = 0; j < MAX_NO_SCALES << 1; ( j++, k++ ) )

        {

            if ( no_lead_fx[i][j] <= 0 )

            {

                j = MAX_NO_SCALES << 1;

            }

        }

        no_scales[i][1] = k;

        move16();

#else  /* OPT_STEREO_32KBPS_V1 */

        j = 0;

        move16();

        test();
@@ -143,10 +167,35 @@ void init_lvq_fx( 
        }

        no_scales[i][1] = sub( j, MAX_NO_SCALES );

        move16();

#endif /* OPT_STEREO_32KBPS_V1 */

    }

    /* predictive mode */

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

    {

#ifdef OPT_STEREO_32KBPS_V1

        FOR( ( j = 0, k = 0 ); j < MAX_NO_SCALES; ( j++, k++ ) )

        {



            if ( ( no_lead_p_fx[i][j] <= 0 ) )

            {

                j = MAX_NO_SCALES;

            }

        }

        no_scales_p[i][0] = k;

        move16();



        FOR( k = 0; j < MAX_NO_SCALES << 1; ( j++, k++ ) )

        {



            if ( ( no_lead_p_fx[i][j] <= 0 ) )

            {

                j = MAX_NO_SCALES << 1;

            }

        }



        no_scales_p[i][1] = k;

        move16();

#else  /* OPT_STEREO_32KBPS_V1 */

        j = 0;

        move16();

        WHILE( ( LT_16( j, MAX_NO_SCALES ) ) && ( no_lead_p_fx[i][j] > 0 ) )
@@ -165,6 +214,7 @@ void init_lvq_fx( 
        }

        no_scales_p[i][1] = sub( j, MAX_NO_SCALES );

        move16();

#endif /* OPT_STEREO_32KBPS_V1 */

    }

    /* index offsets for each truncation */

    init_offset_fx( offset_scale1, offset_scale2, offset_scale1_p, offset_scale2_p, no_scales, no_scales_p );

lib_com/options.h

+1 −0
Original line number Diff line number Diff line
@@ -172,4 +172,5 @@ 
#define FIX_1301_CORRECT_TD_CNST                /* VA: Fix 1301, correct wrong constant in TD stereo */

#define NONBE_FIX_1277_EVS_DTX_HIGH_RATE_THRESHOLD      /* VA/Eri: FLP issue 1277: Fix Mismatch in DTX high-rate threshold between EVS float and BASOP */

#define NONBE_FIX_708_OSBA_BR_SWITCHING_CRASH   /* FhG: issue 708: fix crash in OSBA BR switching with long test vectors */

//#define OPT_STEREO_32KBPS_V1                    /* Optimization made in stereo decoding path for 32kbps decoding */

#endif