analyse.c

linux下编译已经通过

                        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 );
    }

    if( h->param.analyse.b_weighted_bipred )
        h->mc.avg_weight[PIXEL_16x16]( pix1, 16, src2, stride2, weight );
    else
        h->mc.avg[PIXEL_16x16]( pix1, 16, src2, stride2 );

    a->i_cost16x16bi = h->pixf.mbcmp[PIXEL_16x16]( h->mb.pic.p_fenc[0], FENC_STRIDE, pix1, 16 )
                     + REF_COST( 0, a->l0.i_ref )
                     + REF_COST( 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];
}

static inline void x264_mb_cache_mv_p8x8( x264_t *h, x264_mb_analysis_t *a, int i )
{
    const int x = 2*(i%2);
    const int y = 2*(i/2);

    switch( h->mb.i_sub_partition[i] )
    {
        case D_L0_8x8:
            x264_macroblock_cache_mv( h, x, y, 2, 2, 0, a->l0.me8x8[i].mv[0], a->l0.me8x8[i].mv[1] );
            break;
        case D_L0_8x4:
            x264_macroblock_cache_mv( h, x, y+0, 2, 1, 0, a->l0.me8x4[i][0].mv[0], a->l0.me8x4[i][0].mv[1] );
            x264_macroblock_cache_mv( h, x, y+1, 2, 1, 0, a->l0.me8x4[i][1].mv[0], a->l0.me8x4[i][1].mv[1] );
            break;
        case D_L0_4x8:
            x264_macroblock_cache_mv( h, x+0, y, 1, 2, 0, a->l0.me4x8[i][0].mv[0], a->l0.me4x8[i][0].mv[1] );
            x264_macroblock_cache_mv( h, x+1, y, 1, 2, 0, a->l0.me4x8[i][1].mv[0], a->l0.me4x8[i][1].mv[1] );
            break;
        case D_L0_4x4:
            x264_macroblock_cache_mv( h, x+0, y+0, 1, 1, 0, a->l0.me4x4[i][0].mv[0], a->l0.me4x4[i][0].mv[1] );
            x264_macroblock_cache_mv( h, x+1, y+0, 1, 1, 0, a->l0.me4x4[i][1].mv[0], a->l0.me4x4[i][1].mv[1] );
            x264_macroblock_cache_mv( h, x+0, y+1, 1, 1, 0, a->l0.me4x4[i][2].mv[0], a->l0.me4x4[i][2].mv[1] );
            x264_macroblock_cache_mv( h, x+1, y+1, 1, 1, 0, a->l0.me4x4[i][3].mv[0], a->l0.me4x4[i][3].mv[1] );
            break;
        default:
            x264_log( h, X264_LOG_ERROR, "internal error\n" );
            break;
    }
}

#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], l, lX->i_ref, 8*x8, 8*y8 );

            x264_mb_predict_mv( h, l, 4*i, 2, m->mvp );
            x264_me_search( h, m, &lX->me16x16.mv, 1 );

            x264_macroblock_cache_mv( h, 2*x8, 2*y8, 2, 2, l, m->mv[0], m->mv[1] );

            /* BI mode */
            h->mc.mc_luma( m->p_fref, m->i_stride[0], pix[l], 8,
                            m->mv[0], m->mv[1], 8, 8 );
            i_part_cost_bi += m->cost_mv;
            /* FIXME: ref cost */
        }

        WEIGHTED_AVG( PIXEL_8x8, pix[0], 8, pix[1], 8 );
        i_part_cost_bi += h->pixf.mbcmp[PIXEL_8x8]( a->l0.me8x8[i].p_fenc[0], FENC_STRIDE, pix[0], 8 )
                        + a->i_lambda * i_sub_mb_b_cost_table[D_BI_8x8];
        a->l0.me8x8[i].cost += a->i_lambda * i_sub_mb_b_cost_table[D_L0_8x8];
        a->l1.me8x8[i].cost += a->i_lambda * i_sub_mb_b_cost_table[D_L1_8x8];

        i_part_cost = a->l0.me8x8[i].cost;
        h->mb.i_sub_partition[i] = D_L0_8x8;
        COPY2_IF_LT( i_part_cost, a->l1.me8x8[i].cost, h->mb.i_sub_partition[i], D_L1_8x8 );
        COPY2_IF_LT( i_part_cost, i_part_cost_bi, h->mb.i_sub_partition[i], D_BI_8x8 );
        COPY2_IF_LT( i_part_cost, a->i_cost8x8direct[i], h->mb.i_sub_partition[i], D_DIRECT_8x8 );
        a->i_cost8x8bi += i_part_cost;

        /* XXX Needed for x264_mb_predict_mv */
        x264_mb_cache_mv_b8x8( h, a, i, 0 );
    }

    /* mb type cost */
    a->i_cost8x8bi += a->i_lambda * i_mb_b_cost_table[B_8x8];
}

static void x264_mb_analyse_inter_b16x8( 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] };
    DECLARE_ALIGNED( uint8_t,  pix[2][16*8], 16 );
    int mvc[2][2];
    int i, l;

    h->mb.i_partition = D_16x8;
    a->i_cost16x8bi = 0;

    for( i = 0; i < 2; i++ )
    {
        int i_part_cost;
        int i_part_cost_bi = 0;

        /* TODO: check only the list(s) that were used in b8x8? */
        for( l = 0; l < 2; l++ )
        {
            x264_mb_analysis_list_t *lX = l ? &a->l1 : &a->l0;
            x264_me_t *m = &lX->me16x8[i];

            m->i_pixel = PIXEL_16x8;
            m->p_cost_mv = a->p_cost_mv;

            LOAD_FENC( m, h->mb.pic.p_fenc, 0, 8*i );
            LOAD_HPELS( m, p_fref[l], l, lX->i_ref, 0, 8*i );

            mvc[0][0] = lX->me8x8[2*i].mv[0];
            mvc[0][1] = lX->me8x8[2*i].mv[1];
            mvc[1][0] = lX->me8x8[2*i+1].mv[0];
            mvc[1][1] = lX->me8x8[2*i+1].mv[1];

            x264_mb_predict_mv( h, 0, 8*i, 2, m->mvp );
            x264_me_search( h, m, mvc, 2 );

            /* BI mode */
            h->mc.mc_luma( m->p_fref, m->i_stride[0], pix[l], 16,
                            m->mv[0], m->mv[1], 16, 8 );
            /* FIXME: ref cost */
            i_part_cost_bi += m->cost_mv;
        }

        WEIGHTED_AVG( PIXEL_16x8, pix[0], 16, pix[1], 16 );
        i_part_cost_bi += h->pixf.mbcmp[PIXEL_16x8]( a->l0.me16x8[i].p_fenc[0], FENC_STRIDE, pix[0], 16 );

        i_part_cost = a->l0.me16x8[i].cost;
        a->i_mb_partition16x8[i] = D_L0_8x8; /* not actually 8x8, only the L0 matters */
        if( a->l1.me16x8[i].cost < i_part_cost )
        {
            i_part_cost = a->l1.me16x8[i].cost;
            a->i_mb_partition16x8[i] = D_L1_8x8;
        }
        if( i_part_cost_bi + a->i_lambda * 1 < i_part_cost )
        {
            i_part_cost = i_part_cost_bi;
            a->i_mb_partition16x8[i] = D_BI_8x8;
        }
        a->i_cost16x8bi += i_part_cost;

        x264_mb_cache_mv_b16x8( h, a, i, 0 );
    }

    /* mb type cost */
    a->i_mb_type16x8 = B_L0_L0
        + (a->i_mb_partition16x8[0]>>2) * 3
        + (a->i_mb_partition16x8[1]>>2);
    a->i_cost16x8bi += a->i_lambda * i_mb_b16x8_cost_table[a->i_mb_type16x8];
}

static void x264_mb_analyse_inter_b8x16( 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*16];
    int mvc[2][2];
    int i, l;

    h->mb.i_partition = D_8x16;
    a->i_cost8x16bi = 0;

    for( i = 0; i < 2; i++ )
    {
        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->me8x16[i];

            m->i_pixel = PIXEL_8x16;
            m->p_cost_mv = a->p_cost_mv;

            LOAD_FENC( m, h->mb.pic.p_fenc, 8*i, 0 );
            LOAD_HPELS( m, p_fref[l], l, lX->i_ref, 8*i, 0 );

            mvc[0][0] = lX->me8x8[i].mv[0];
            mvc[0][1] = lX->me8x8[i].mv[1];
            mvc[1][0] = lX->me8x8[i+2].mv[0];
            mvc[1][1] = lX->me8x8[i+2].mv[1];

            x264_mb_predict_mv( h, 0, 4*i, 2, m->mvp );
            x264_me_search( h, m, mvc, 2 );

            /* BI mode */
            h->mc.mc_luma( m->p_fref, m->i_stride[0], pix[l], 8,
                            m->mv[0], m->mv[1], 8, 16 );
            /* FIXME: ref cost */
            i_part_cost_bi += m->cost_mv;
        }

        WEIGHTED_AVG( PIXEL_8x16, pix[0], 8, pix[1], 8 );
        i_part_cost_bi += h->pixf.mbcmp[PIXEL_8x16]( a->l0.me8x16[i].p_fenc[0], FENC_STRIDE, pix[0], 8 );

        i_part_cost = a->l0.me8x16[i].cost;
        a->i_mb_partition8x16[i] = D_L0_8x8;
        if( a->l1.me8x16[i].cost < i_part_cost )
        {
            i_part_cost = a->l1.me8x16[i].cost;
            a->i_mb_partition8x16[i] = D_L1_8x8;
        }
        if( i_part_cost_bi + a->i_lambda * 1 < i_part_cost )
        {
            i_part_cost = i_part_cost_bi;
            a->i_mb_partition8x16[i] = D_BI_8x8;
        }
        a->i_cost8x16bi += i_part_cost;

        x264_mb_cache_mv_b8x16( h, a, i, 0 );
    }

    /* mb type cost */
    a->i_mb_type8x16 = B_L0_L0
        + (a->i_mb_partition8x16[0]>>2) * 3
        + (a->i_mb_partition8x16[1]>>2);
    a->i_cost8x16bi += a->i_lambda * i_mb_b16x8_cost_table[a->i_mb_type8x16];
}

static void x264_mb_analyse_p_rd( x264_t *h, x264_mb_analysis_t *a, int i_satd )
{
    int thresh = i_satd * 5/4;

    h->mb.i_type = P_L0;
    if( a->l0.i_rd16x16 == COST_MAX && a->l0.me16x16.cost <= i_satd * 3/2 )
    {
        h->mb.i_partition = D_16x16;
        x264_analyse_update_cache( h, a );
        a->l0.i_rd16x16 = x264_rd_cost_mb( h, a->i_lambda2 );
    }
    a->l0.me16x16.cost = a->l0.i_rd16x16;

    if( a->l0.i_cost16x8 <= thresh )
    {
        h->mb.i_partition = D_16x8;
        x264_analyse_update_cache( h, a );
        a->l0.i_cost16x8 = x264_rd_cost_mb( h, a->i_lambda2 );
    }
    else
        a->l0.i_cost16x8 = COST_MAX;

    if( a->l0.i_cost8x16 <= thresh )
    {
        h->mb.i_partition = D_8x16;
        x264_analyse_update_cache( h, a );
        a->l0.i_cost8x16 = x264_rd_cost_mb( h, a->i_lambda2 );
    }
    else
        a->l0.i_cost8x16 = COST_MAX;

    if( a->l0.i_cost8x8 <= thresh )
    {
        h->mb.i_type = P_8x8;
        x264_analyse_update_cache( h, a );
        a->l0.i_cost8x8 = x264_rd_cost_mb( h, a->i_lambda2 );

        if( h->param.analyse.inter & X264_ANALYSE_PSUB8x8 )
        {
            /* FIXME: RD per subpartition */
            int part_bak[4];
            int i, i_cost;
            int b_sub8x8 = 0;
            for( i=0; i<4; i++ )
            {
                part_bak[i] = h->mb.i_sub_partition[i];
                b_sub8x8 |= (part_bak[i] != D_L0_8x8);
            }
            if( b_sub8x8 )
            {
                h->mb.i_sub_partition[0] = h->mb.i_sub_partition[1] =
                h->mb.i_sub_partition[2] = h->mb.i_sub_partition[3] = D_L0_8x8;
                i_cost = x264_rd_cost_mb( h, a->i_lambda2 );
                if( a->l0.i_cost8x8 < i_cost )
                {
                    for( i=0; i<4; i++ )
                        h->mb.i_sub_partitio