h264.c

64-bits H.264 from ffmpeg 2008 version Build in VC++ 2008 no error warning, Jesse Stone, Taiwan

    tprintf(h->s.avctx, "pred_pskip: (%d) (%d) at %2d %2d\n", top_ref, left_ref, h->s.mb_x, h->s.mb_y);

    if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
       || (top_ref == 0  && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0)
       || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){

        *mx = *my = 0;
        return;
    }

    pred_motion(h, 0, 4, 0, 0, mx, my);

    return;
}

static  void direct_dist_scale_factor(H264Context * const h){
    const int poc = h->s.current_picture_ptr->poc;
    const int poc1 = h->ref_list[1][0].poc;
    int i;
    for(i=0; i<(int)h->ref_count[0]; i++){
        int poc0 = h->ref_list[0][i].poc;
        int td = av_clip(poc1 - poc0, -128, 127);
        if(td == 0 /* FIXME || pic0 is a long-term ref */){
            h->dist_scale_factor[i] = 256;
        }else{
            int tb = av_clip(poc - poc0, -128, 127);
            int tx = (16384 + (FFABS(td) >> 1)) / td;
            h->dist_scale_factor[i] = av_clip((tb*tx + 32) >> 6, -1024, 1023);
        }
    }
    if(FRAME_MBAFF){
        for(i=0; i<(int)h->ref_count[0]; i++){
            h->dist_scale_factor_field[2*i] =
            h->dist_scale_factor_field[2*i+1] = h->dist_scale_factor[i];
        }
    }
}
static  void direct_ref_list_init(H264Context * const h){
    MpegEncContext * const s = &h->s;
    Picture * const ref1 = &h->ref_list[1][0];
    Picture * const cur = s->current_picture_ptr;
    int list, i, j;
    if(cur->pict_type == FF_I_TYPE)
        cur->ref_count[0] = 0;
    if(cur->pict_type != FF_B_TYPE)
        cur->ref_count[1] = 0;
    for(list=0; list<2; list++){
        cur->ref_count[list] = h->ref_count[list];
        for(j=0; j<(int)h->ref_count[list]; j++)
            cur->ref_poc[list][j] = h->ref_list[list][j].poc;
    }
    if(cur->pict_type != FF_B_TYPE || h->direct_spatial_mv_pred)
        return;
    for(list=0; list<2; list++){
        for(i=0; i<ref1->ref_count[list]; i++){
            const int poc = ref1->ref_poc[list][i];
            h->map_col_to_list0[list][i] = 0; /* bogus; fills in for missing frames */
            for(j=0; j<(int)h->ref_count[list]; j++)
                if(h->ref_list[list][j].poc == poc){
                    h->map_col_to_list0[list][i] = j;
                    break;
                }
        }
    }
    if(FRAME_MBAFF){
        for(list=0; list<2; list++){
            for(i=0; i<ref1->ref_count[list]; i++){
                j = h->map_col_to_list0[list][i];
                h->map_col_to_list0_field[list][2*i] = 2*j;
                h->map_col_to_list0_field[list][2*i+1] = 2*j+1;
            }
        }
    }
}

static  void pred_direct_motion(H264Context * const h, int *mb_type){
    MpegEncContext * const s = &h->s;
    const int mb_xy =   h->mb_xy;
    const int b8_xy = 2*s->mb_x + 2*s->mb_y*h->b8_stride;
    const int b4_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
    const int mb_type_col = h->ref_list[1][0].mb_type[mb_xy];
    const int16_t (*l1mv0)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[0][b4_xy];
    const int16_t (*l1mv1)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[1][b4_xy];
    const int8_t *l1ref0 = &h->ref_list[1][0].ref_index[0][b8_xy];
    const int8_t *l1ref1 = &h->ref_list[1][0].ref_index[1][b8_xy];
    const int is_b8x8 = IS_8X8(*mb_type);
    unsigned int sub_mb_type;
    int i8, i4;

#define MB_TYPE_16x16_OR_INTRA (MB_TYPE_16x16|MB_TYPE_INTRA4x4|MB_TYPE_INTRA16x16|MB_TYPE_INTRA_PCM)
    if(IS_8X8(mb_type_col) && !h->sps.direct_8x8_inference_flag){
        /* FIXME save sub mb types from previous frames (or derive from MVs)
         * so we know exactly what block size to use */
        sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */
        *mb_type =    MB_TYPE_8x8|MB_TYPE_L0L1;
    }else if(!is_b8x8 && (mb_type_col & MB_TYPE_16x16_OR_INTRA)){
        sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
        *mb_type =    MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */
    }else{
        sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
        *mb_type =    MB_TYPE_8x8|MB_TYPE_L0L1;
    }
    if(!is_b8x8)
        *mb_type |= MB_TYPE_DIRECT2;
    if(MB_FIELD)
        *mb_type |= MB_TYPE_INTERLACED;

    tprintf(s->avctx, "mb_type = %08x, sub_mb_type = %08x, is_b8x8 = %d, mb_type_col = %08x\n", *mb_type, sub_mb_type, is_b8x8, mb_type_col);

    if(h->direct_spatial_mv_pred){
        int ref[2];
        int mv[2][2];
        int list;

        /* FIXME interlacing + spatial direct uses wrong colocated block positions */

        /* ref = min(neighbors) */
        for(list=0; list<2; list++){
            int refa = h->ref_cache[list][scan8[0] - 1];
            int refb = h->ref_cache[list][scan8[0] - 8];
            int refc = h->ref_cache[list][scan8[0] - 8 + 4];
            if(refc == -2)
                refc = h->ref_cache[list][scan8[0] - 8 - 1];
            ref[list] = refa;
            if(ref[list] < 0 || (refb < ref[list] && refb >= 0))
                ref[list] = refb;
            if(ref[list] < 0 || (refc < ref[list] && refc >= 0))
                ref[list] = refc;
            if(ref[list] < 0)
                ref[list] = -1;
        }

        if(ref[0] < 0 && ref[1] < 0){
            ref[0] = ref[1] = 0;
            mv[0][0] = mv[0][1] =
            mv[1][0] = mv[1][1] = 0;
        }else{
            for(list=0; list<2; list++){
                if(ref[list] >= 0)
                    pred_motion(h, 0, 4, list, ref[list], &mv[list][0], &mv[list][1]);
                else
                    mv[list][0] = mv[list][1] = 0;
            }
        }

        if(ref[1] < 0){
            if(!is_b8x8)
                *mb_type &= ~MB_TYPE_L1;
            sub_mb_type &= ~MB_TYPE_L1;
        }else if(ref[0] < 0){
            if(!is_b8x8)
                *mb_type &= ~MB_TYPE_L0;
            sub_mb_type &= ~MB_TYPE_L0;
        }

        if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col)){
            int pair_xy = s->mb_x + (s->mb_y&~1)*s->mb_stride;
            int mb_types_col[2];
            int b8_stride = h->b8_stride;
            int b4_stride = h->b_stride;

            *mb_type = (*mb_type & ~MB_TYPE_16x16) | MB_TYPE_8x8;

            if(IS_INTERLACED(*mb_type)){
                mb_types_col[0] = h->ref_list[1][0].mb_type[pair_xy];
                mb_types_col[1] = h->ref_list[1][0].mb_type[pair_xy+s->mb_stride];
                if(s->mb_y&1){
                    l1ref0 -= 2*b8_stride;
                    l1ref1 -= 2*b8_stride;
                    l1mv0 -= 4*b4_stride;
                    l1mv1 -= 4*b4_stride;
                }
                b8_stride *= 3;
                b4_stride *= 6;
            }else{
                int cur_poc = s->current_picture_ptr->poc;
                int *col_poc = h->ref_list[1]->field_poc;
                int col_parity = FFABS(col_poc[0] - cur_poc) >= FFABS(col_poc[1] - cur_poc);
                int dy = 2*col_parity - (s->mb_y&1);
                mb_types_col[0] =
                mb_types_col[1] = h->ref_list[1][0].mb_type[pair_xy + col_parity*s->mb_stride];
                l1ref0 += dy*b8_stride;
                l1ref1 += dy*b8_stride;
                l1mv0 += 2*dy*b4_stride;
                l1mv1 += 2*dy*b4_stride;
                b8_stride = 0;
            }

            for(i8=0; i8<4; i8++){
                int x8 = i8&1;
                int y8 = i8>>1;
                int xy8 = x8+y8*b8_stride;
                int xy4 = 3*x8+y8*b4_stride;
                int a=0, b=0;

                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
                    continue;
                h->sub_mb_type[i8] = sub_mb_type;

                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);
                if(!IS_INTRA(mb_types_col[y8])
                   && (   (l1ref0[xy8] == 0 && FFABS(l1mv0[xy4][0]) <= 1 && FFABS(l1mv0[xy4][1]) <= 1)
                       || (l1ref0[xy8]  < 0 && l1ref1[xy8] == 0 && FFABS(l1mv1[xy4][0]) <= 1 && FFABS(l1mv1[xy4][1]) <= 1))){
                    if(ref[0] > 0)
                        a= pack16to32(mv[0][0],mv[0][1]);
                    if(ref[1] > 0)
                        b= pack16to32(mv[1][0],mv[1][1]);
                }else{
                    a= pack16to32(mv[0][0],mv[0][1]);
                    b= pack16to32(mv[1][0],mv[1][1]);
                }
                fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, a, 4);
                fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, b, 4);
            }
        }else if(IS_16X16(*mb_type)){
            int a=0, b=0;

            fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, (uint8_t)ref[0], 1);
            fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, (uint8_t)ref[1], 1);
            if(!IS_INTRA(mb_type_col)
               && (   (l1ref0[0] == 0 && FFABS(l1mv0[0][0]) <= 1 && FFABS(l1mv0[0][1]) <= 1)
                   || (l1ref0[0]  < 0 && l1ref1[0] == 0 && FFABS(l1mv1[0][0]) <= 1 && FFABS(l1mv1[0][1]) <= 1
                       && (h->x264_build>33 || !h->x264_build)))){
                if(ref[0] > 0)
                    a= pack16to32(mv[0][0],mv[0][1]);
                if(ref[1] > 0)
                    b= pack16to32(mv[1][0],mv[1][1]);
            }else{
                a= pack16to32(mv[0][0],mv[0][1]);
                b= pack16to32(mv[1][0],mv[1][1]);
            }
            fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, a, 4);
            fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, b, 4);
        }else{
            for(i8=0; i8<4; i8++){
                const int x8 = i8&1;
                const int y8 = i8>>1;

                if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
                    continue;
                h->sub_mb_type[i8] = sub_mb_type;

                fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mv[0][0],mv[0][1]), 4);
                fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mv[1][0],mv[1][1]), 4);
                fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);
                fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);

                /* col_zero_flag */
                if(!IS_INTRA(mb_type_col) && (   l1ref0[x8 + y8*h->b8_stride] == 0
                                              || (l1ref0[x8 + y8*h->b8_stride] < 0 && l1ref1[x8 + y8*h->b8_stride] == 0
                                                  && (h->x264_build>33 || !h->x264_build)))){
                    const int16_t (*l1mv)[2]= l1ref0[x8 + y8*h->b8_stride] == 0 ? l1mv0 : l1mv1;
                    if(IS_SUB_8X8(sub_mb_type)){
                        const int16_t *mv_col = l1mv[x8*3 + y8*3*h->b_stride];
                        if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
                            if(ref[0] == 0)
                                fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
                            if(ref[1] == 0)
                                fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
                        }
                    }else
                    for(i4=0; i4<4; i4++){
                        const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
                        if(FFABS(mv_col[0]) <= 1 && FFABS(mv_col[1]) <= 1){
                            if(ref[0] == 0)
                                *(uint32_t*)h->mv_cache[0][scan8[i8*4+i4]] = 0;
                            if(ref[1] == 0)
                                *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = 0;
                        }
                    }
                }
            }
        }
    }else{ /* direct temporal mv pred */
        const int *map_col_to_list0[2] = {h->map_col_to_list0[0], h->map_col_to_list0[1]};
        const int *dist_scale_factor = h->dist_scale_factor;

        if(FRAME_MBAFF){
            if(IS_INTERLACED(*mb_type)){
                map_col_to_list0[0] = h->map_col_to_list0_field[0];
                map_col_to_list0[1] = h->map_col_to_list0_field[1];
                dist_scale_factor = h->dist_scale_factor_field;
            }
            if(IS_INTERLACED(*mb_type) != IS_INTERLACED(mb_type_col)){
                /* FIXME assumes direct_8x8_inference == 1 */
                const int pair_xy = s->mb_x + (s->mb_y&~1)*s->mb_stride;
                int mb_types_col[2];
                int y_shift;

                *mb_type = MB_TYPE_8x8|MB_TYPE_L0L1
                         | (is_b8x8 ? 0 : MB_TYPE_DIRECT2)
                         | (*mb_type & MB_TYPE_INTERLACED);
                sub_mb_type = MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2|MB_TYPE_16x16;

                if(IS_INTERLACED(*mb_type)){
                    /* frame to field scaling */
                    mb_types_col[0] = h->ref_list[1][0].mb_type[pair_xy];
                    mb_types_col[1] = h->ref_list[1][0].mb_type[pair_xy+s->mb_stride];
                    if(s->mb_y&1){
                        l1ref0 -= 2*h->b8_stride;
                        l1ref1 -= 2*h->b8_stride;
                        l1mv0 -= 4*h->b_stride;
                        l1mv1 -= 4*h->b_stride;