analyse.c

H.264 source codes

    else
    {
        CHROMA4x4MC( 2,4, a->l0.me4x8[i8x8][0], 0,0 );
        CHROMA4x4MC( 2,4, a->l0.me4x8[i8x8][1], 2,0 );
    }

    return h->pixf.satd[PIXEL_4x4]( &h->mb.pic.p_fenc[1][off], i_stride, pix1, 8 )
         + h->pixf.satd[PIXEL_4x4]( &h->mb.pic.p_fenc[2][off], i_stride, pix2, 8 );
}

static void x264_mb_analyse_inter_p4x4( x264_t *h, x264_mb_analysis_t *a, int i8x8 )
{
    uint8_t  **p_fref = h->mb.pic.p_fref[0][a->l0.i_ref];
    uint8_t  **p_fenc = h->mb.pic.p_fenc;

    int i4x4;

    /* XXX Needed for x264_mb_predict_mv */
    h->mb.i_partition = D_8x8;

    for( i4x4 = 0; i4x4 < 4; i4x4++ )
    {
        const int idx = 4*i8x8 + i4x4;
        const int x4 = block_idx_x[idx];
        const int y4 = block_idx_y[idx];
        const int i_mvc = (i4x4 == 0);

        x264_me_t *m = &a->l0.me4x4[i8x8][i4x4];

        m->i_pixel = PIXEL_4x4;
        m->p_cost_mv = a->p_cost_mv;

        LOAD_FENC( m, p_fenc, 4*x4, 4*y4 );
        LOAD_HPELS( m, p_fref, 4*x4, 4*y4 );

        x264_mb_predict_mv( h, 0, idx, 1, m->mvp );
        x264_me_search( h, m, &a->l0.me8x8[i8x8].mv, i_mvc );

        x264_macroblock_cache_mv( h, x4, y4, 1, 1, 0, m->mv[0], m->mv[1] );
    }

    a->l0.i_cost4x4[i8x8] = a->l0.me4x4[i8x8][0].cost +
                         a->l0.me4x4[i8x8][1].cost +
                         a->l0.me4x4[i8x8][2].cost +
                         a->l0.me4x4[i8x8][3].cost +
                         a->i_lambda * i_sub_mb_p_cost_table[D_L0_4x4];
    if( h->mb.b_chroma_me )
        a->l0.i_cost4x4[i8x8] += x264_mb_analyse_inter_p4x4_chroma( h, a, p_fref, i8x8, PIXEL_4x4 );
}

static void x264_mb_analyse_inter_p8x4( x264_t *h, x264_mb_analysis_t *a, int i8x8 )
{
    uint8_t  **p_fref = h->mb.pic.p_fref[0][a->l0.i_ref];
    uint8_t  **p_fenc = h->mb.pic.p_fenc;

    int i8x4;

    /* XXX Needed for x264_mb_predict_mv */
    h->mb.i_partition = D_8x8;

    for( i8x4 = 0; i8x4 < 2; i8x4++ )
    {
        const int idx = 4*i8x8 + 2*i8x4;
        const int x4 = block_idx_x[idx];
        const int y4 = block_idx_y[idx];
        const int i_mvc = (i8x4 == 0);

        x264_me_t *m = &a->l0.me8x4[i8x8][i8x4];

        m->i_pixel = PIXEL_8x4;
        m->p_cost_mv = a->p_cost_mv;

        LOAD_FENC( m, p_fenc, 4*x4, 4*y4 );
        LOAD_HPELS( m, p_fref, 4*x4, 4*y4 );

        x264_mb_predict_mv( h, 0, idx, 2, m->mvp );
        x264_me_search( h, m, &a->l0.me4x4[i8x8][0].mv, i_mvc );

        x264_macroblock_cache_mv( h, x4, y4, 2, 1, 0, m->mv[0], m->mv[1] );
    }

    a->l0.i_cost8x4[i8x8] = a->l0.me8x4[i8x8][0].cost + a->l0.me8x4[i8x8][1].cost +
                            a->i_lambda * i_sub_mb_p_cost_table[D_L0_8x4];
    if( h->mb.b_chroma_me )
        a->l0.i_cost8x4[i8x8] += x264_mb_analyse_inter_p4x4_chroma( h, a, p_fref, i8x8, PIXEL_8x4 );
}

static void x264_mb_analyse_inter_p4x8( x264_t *h, x264_mb_analysis_t *a, int i8x8 )
{
    uint8_t  **p_fref = h->mb.pic.p_fref[0][a->l0.i_ref];
    uint8_t  **p_fenc = h->mb.pic.p_fenc;

    int i4x8;

    /* XXX Needed for x264_mb_predict_mv */
    h->mb.i_partition = D_8x8;

    for( i4x8 = 0; i4x8 < 2; i4x8++ )
    {
        const int idx = 4*i8x8 + i4x8;
        const int x4 = block_idx_x[idx];
        const int y4 = block_idx_y[idx];
        const int i_mvc = (i4x8 == 0);

        x264_me_t *m = &a->l0.me4x8[i8x8][i4x8];

        m->i_pixel = PIXEL_4x8;
        m->p_cost_mv = a->p_cost_mv;

        LOAD_FENC( m, p_fenc, 4*x4, 4*y4 );
        LOAD_HPELS( m, p_fref, 4*x4, 4*y4 );

        x264_mb_predict_mv( h, 0, idx, 1, m->mvp );
        x264_me_search( h, m, &a->l0.me4x4[i8x8][0].mv, i_mvc );

        x264_macroblock_cache_mv( h, x4, y4, 1, 2, 0, m->mv[0], m->mv[1] );
    }

    a->l0.i_cost4x8[i8x8] = a->l0.me4x8[i8x8][0].cost + a->l0.me4x8[i8x8][1].cost +
                            a->i_lambda * i_sub_mb_p_cost_table[D_L0_4x8];
    if( h->mb.b_chroma_me )
        a->l0.i_cost4x8[i8x8] += x264_mb_analyse_inter_p4x4_chroma( h, a, p_fref, i8x8, PIXEL_4x8 );
}

static void x264_mb_analyse_inter_direct( x264_t *h, x264_mb_analysis_t *a )
{
    /* Assumes that fdec still contains the results of
     * x264_mb_predict_mv_direct16x16 and x264_mb_mc */

    uint8_t **p_fenc = h->mb.pic.p_fenc;
    uint8_t **p_fdec = h->mb.pic.p_fdec;
    int i_stride= h->mb.pic.i_stride[0];
    int i;

    a->i_cost16x16direct = 0;
    for( i = 0; i < 4; i++ )
    {
        const int x8 = i%2;
        const int y8 = i/2;
        const int off = 8 * x8 + 8 * i_stride * y8;
        a->i_cost16x16direct +=
        a->i_cost8x8direct[i] =
            h->pixf.satd[PIXEL_8x8]( &p_fenc[0][off], i_stride, &p_fdec[0][off], i_stride );

        /* mb type cost */
        a->i_cost8x8direct[i] += a->i_lambda * i_sub_mb_b_cost_table[D_DIRECT_8x8];
    }

    a->i_cost16x16direct += a->i_lambda * i_mb_b_cost_table[B_DIRECT];
}

#define WEIGHTED_AVG( size, pix1, stride1, src2, stride2 ) \
    { \
        if( h->param.analyse.b_weighted_bipred ) \
            h->pixf.avg_weight[size]( pix1, stride1, src2, stride2, \
                    h->mb.bipred_weight[a->l0.i_ref][a->l1.i_ref] ); \
        else \
            h->pixf.avg[size]( pix1, stride1, src2, stride2 ); \
    }

static void x264_mb_analyse_inter_b16x16( x264_t *h, x264_mb_analysis_t *a )
{
    uint8_t pix1[16*16], pix2[16*16];
    uint8_t *src2;
    int stride2 = 16;
    int src2_ref, pix1_ref;

    x264_me_t m;
    int i_ref;
    int mvc[5][2], i_mvc;
    int i_fullpel_thresh = INT_MAX;
    int *p_fullpel_thresh = h->i_ref0>1 ? &i_fullpel_thresh : NULL;

    /* 16x16 Search on all ref frame */
    m.i_pixel = PIXEL_16x16;
    m.p_cost_mv = a->p_cost_mv;
    LOAD_FENC( &m, h->mb.pic.p_fenc, 0, 0 );

    /* ME for List 0 */
    a->l0.me16x16.cost = INT_MAX;
    for( i_ref = 0; i_ref < h->i_ref0; i_ref++ )
    {
        /* search with ref */
        LOAD_HPELS( &m, h->mb.pic.p_fref[0][i_ref], 0, 0 );
        x264_mb_predict_mv_16x16( h, 0, i_ref, m.mvp );
        x264_mb_predict_mv_ref16x16( h, 0, i_ref, mvc, &i_mvc );
        x264_me_search_ref( h, &m, mvc, i_mvc, p_fullpel_thresh );

        /* add ref cost */
        m.cost += a->i_lambda * bs_size_te( h->sh.i_num_ref_idx_l0_active - 1, i_ref );

        if( m.cost < a->l0.me16x16.cost )
        {
            a->l0.i_ref = i_ref;
            a->l0.me16x16 = m;
        }

        /* save mv for predicting neighbors */
        h->mb.mvr[0][i_ref][h->mb.i_mb_xy][0] = m.mv[0];
        h->mb.mvr[0][i_ref][h->mb.i_mb_xy][1] = m.mv[1];
    }
    /* subtract ref cost, so we don't have to add it for the other MB types */
    a->l0.me16x16.cost -= a->i_lambda * bs_size_te( h->sh.i_num_ref_idx_l0_active - 1, a->l0.i_ref );

    /* ME for list 1 */
    i_fullpel_thresh = INT_MAX;
    p_fullpel_thresh = h->i_ref1>1 ? &i_fullpel_thresh : NULL;
    a->l1.me16x16.cost = INT_MAX;
    for( i_ref = 0; i_ref < h->i_ref1; i_ref++ )
    {
        /* search with ref */
        LOAD_HPELS( &m, h->mb.pic.p_fref[1][i_ref], 0, 0 );
        x264_mb_predict_mv_16x16( h, 1, i_ref, m.mvp );
        x264_mb_predict_mv_ref16x16( h, 1, i_ref, mvc, &i_mvc );
        x264_me_search_ref( h, &m, mvc, i_mvc, p_fullpel_thresh );

        /* add ref cost */
        m.cost += a->i_lambda * bs_size_te( h->sh.i_num_ref_idx_l1_active - 1, i_ref );

        if( m.cost < a->l1.me16x16.cost )
        {
            a->l1.i_ref = i_ref;
            a->l1.me16x16 = m;
        }

        /* save mv for predicting neighbors */
        h->mb.mvr[1][i_ref][h->mb.i_mb_xy][0] = m.mv[0];
        h->mb.mvr[1][i_ref][h->mb.i_mb_xy][1] = m.mv[1];
    }
    /* subtract ref cost, so we don't have to add it for the other MB types */
    a->l1.me16x16.cost -= a->i_lambda * bs_size_te( h->sh.i_num_ref_idx_l1_active - 1, a->l1.i_ref );

    /* Set global ref, needed for other modes? */
    x264_macroblock_cache_ref( h, 0, 0, 4, 4, 0, a->l0.i_ref );
    x264_macroblock_cache_ref( h, 0, 0, 4, 4, 1, a->l1.i_ref );

    /* get cost of BI mode */
    if ( ((a->l0.me16x16.mv[0] | a->l0.me16x16.mv[1]) & 1) == 0 )
    {
        /* l0 reference is halfpel, so get_ref on it will make it faster */
        src2 = h->mc.get_ref( h->mb.pic.p_fref[0][a->l0.i_ref], h->mb.pic.i_stride[0],
                        pix2, &stride2,
                        a->l0.me16x16.mv[0], a->l0.me16x16.mv[1],
                        16, 16 );
        h->mc.mc_luma( h->mb.pic.p_fref[1][a->l1.i_ref], h->mb.pic.i_stride[0],
                        pix1, 16,
                        a->l1.me16x16.mv[0], a->l1.me16x16.mv[1],
                        16, 16 );
        src2_ref = a->l0.i_ref;
        pix1_ref = a->l1.i_ref;
    } 
    else
    {
        /* if l0 was qpel, we'll use get_ref on l1 instead */
        h->mc.mc_luma( h->mb.pic.p_fref[0][a->l0.i_ref], h->mb.pic.i_stride[0],
                        pix1, 16,
                        a->l0.me16x16.mv[0], a->l0.me16x16.mv[1],
                        16, 16 );
        src2 = h->mc.get_ref( h->mb.pic.p_fref[1][a->l1.i_ref], h->mb.pic.i_stride[0],
                        pix2, &stride2,
                        a->l1.me16x16.mv[0], a->l1.me16x16.mv[1],
                        16, 16 );
        src2_ref = a->l1.i_ref;
        pix1_ref = a->l0.i_ref;
    }

    if( h->param.analyse.b_weighted_bipred )
        h->pixf.avg_weight[PIXEL_16x16]( pix1, 16, src2, stride2,
                h->mb.bipred_weight[pix1_ref][src2_ref] );
    else
        h->pixf.avg[PIXEL_16x16]( pix1, 16, src2, stride2 );

    a->i_cost16x16bi = h->pixf.satd[PIXEL_16x16]( h->mb.pic.p_fenc[0], h->mb.pic.i_stride[0], pix1, 16 )
                     + a->i_lambda * ( bs_size_te( h->sh.i_num_ref_idx_l0_active - 1, a->l0.i_ref )
                                     + bs_size_te( h->sh.i_num_ref_idx_l1_active - 1, a->l1.i_ref ) )
                     + a->l0.me16x16.cost_mv
                     + a->l1.me16x16.cost_mv;

    /* mb type cost */
    a->i_cost16x16bi   += a->i_lambda * i_mb_b_cost_table[B_BI_BI];
    a->l0.me16x16.cost += a->i_lambda * i_mb_b_cost_table[B_L0_L0];
    a->l1.me16x16.cost += a->i_lambda * i_mb_b_cost_table[B_L1_L1];
}

#define CACHE_MV_BI(x,y,dx,dy,me0,me1,part) \
    if( x264_mb_partition_listX_table[0][part] ) \
    { \
        x264_macroblock_cache_ref( h, x,y,dx,dy, 0, a->l0.i_ref ); \
        x264_macroblock_cache_mv(  h, x,y,dx,dy, 0, me0.mv[0], me0.mv[1] ); \
    } \
    else \
    { \
        x264_macroblock_cache_ref( h, x,y,dx,dy, 0, -1 ); \
        x264_macroblock_cache_mv(  h, x,y,dx,dy, 0, 0, 0 ); \
        if( b_mvd ) \
            x264_macroblock_cache_mvd( h, x,y,dx,dy, 0, 0, 0 ); \
    } \
    if( x264_mb_partition_listX_table[1][part] ) \
    { \
        x264_macroblock_cache_ref( h, x,y,dx,dy, 1, a->l1.i_ref ); \
        x264_macroblock_cache_mv(  h, x,y,dx,dy, 1, me1.mv[0], me1.mv[1] ); \
    } \
    else \
    { \
        x264_macroblock_cache_ref( h, x,y,dx,dy, 1, -1 ); \
        x264_macroblock_cache_mv(  h, x,y,dx,dy, 1, 0, 0 ); \
        if( b_mvd ) \
            x264_macroblock_cache_mvd( h, x,y,dx,dy, 1, 0, 0 ); \
    }

static inline void x264_mb_cache_mv_b8x8( x264_t *h, x264_mb_analysis_t *a, int i, int b_mvd )
{
    int x = (i%2)*2;
    int y = (i/2)*2;
    if( h->mb.i_sub_partition[i] == D_DIRECT_8x8 )
    {
        x264_mb_load_mv_direct8x8( h, i );
        if( b_mvd )
        {
            x264_macroblock_cache_mvd(  h, x, y, 2, 2, 0, 0, 0 );
            x264_macroblock_cache_mvd(  h, x, y, 2, 2, 1, 0, 0 );
            x264_macroblock_cache_skip( h, x, y, 2, 2, 1 );
        }
    }
    else
    {
        CACHE_MV_BI( x, y, 2, 2, a->l0.me8x8[i], a->l1.me8x8[i], h->mb.i_sub_partition[i] );
    }
}
static inline void x264_mb_cache_mv_b16x8( x264_t *h, x264_mb_analysis_t *a, int i, int b_mvd )
{
    CACHE_MV_BI( 0, 2*i, 4, 2, a->l0.me16x8[i], a->l1.me16x8[i], a->i_mb_partition16x8[i] );
}
static inline void x264_mb_cache_mv_b8x16( x264_t *h, x264_mb_analysis_t *a, int i, int b_mvd )
{
    CACHE_MV_BI( 2*i, 0, 2, 4, a->l0.me8x16[i], a->l1.me8x16[i], a->i_mb_partition8x16[i] );
}
#undef CACHE_MV_BI

static void x264_mb_analyse_inter_b8x8( x264_t *h, x264_mb_analysis_t *a )
{
    uint8_t **p_fref[2] =
        { h->mb.pic.p_fref[0][a->l0.i_ref],
          h->mb.pic.p_fref[1][a->l1.i_ref] };
    uint8_t pix[2][8*8];
    int i, l;

    /* XXX Needed for x264_mb_predict_mv */
    h->mb.i_partition = D_8x8;

    a->i_cost8x8bi = 0;

    for( i = 0; i < 4; i++ )
    {
        const int x8 = i%2;
        const int y8 = i/2;
        int i_part_cost;
        int i_part_cost_bi = 0;

        for( l = 0; l < 2; l++ )
        {
            x264_mb_analysis_list_t *lX = l ? &a->l1 : &a->l0;
            x264_me_t *m = &lX->me8x8[i];

            m->i_pixel = PIXEL_8x8;
            m->p_cost_mv = a->p_cost_mv;

            LOAD_FENC( m, h->mb.pic.p_fenc, 8*x8, 8*y8 );
            LOAD_HPELS( m, p_fref[l], 8*x8, 8*y8 );

            x264_mb_predict_mv( h, l, 4*i, 2, m->mvp );