vc1.c

        }
        break;
    case TT_8X4:
        for(j = 0; j < 2; j++) {
            last = subblkpat & (1 << (1 - j));
            i = 0;
            off = j * 32;
            while (!last) {
                vc1_decode_ac_coeff(v, &last, &skip, &value, v->codingset2);
                i += skip;
                if(i > 31)
                    break;
                if(v->profile < PROFILE_ADVANCED)
                    idx = ff_vc1_simple_progressive_8x4_zz[i++];
                else
                    idx = ff_vc1_adv_progressive_8x4_zz[i++];
                block[idx + off] = value * scale;
                if(!v->pquantizer)
                    block[idx + off] += (block[idx + off] < 0) ? -mquant : mquant;
            }
            if(!(subblkpat & (1 << (1 - j))))
                s->dsp.vc1_inv_trans_8x4(block, j);
        }
        break;
    case TT_4X8:
        for(j = 0; j < 2; j++) {
            last = subblkpat & (1 << (1 - j));
            i = 0;
            off = j * 4;
            while (!last) {
                vc1_decode_ac_coeff(v, &last, &skip, &value, v->codingset2);
                i += skip;
                if(i > 31)
                    break;
                if(v->profile < PROFILE_ADVANCED)
                    idx = ff_vc1_simple_progressive_4x8_zz[i++];
                else
                    idx = ff_vc1_adv_progressive_4x8_zz[i++];
                block[idx + off] = value * scale;
                if(!v->pquantizer)
                    block[idx + off] += (block[idx + off] < 0) ? -mquant : mquant;
            }
            if(!(subblkpat & (1 << (1 - j))))
                s->dsp.vc1_inv_trans_4x8(block, j);
        }
        break;
    }
    return 0;
}


/** Decode one P-frame MB (in Simple/Main profile)
 */
static int vc1_decode_p_mb(VC1Context *v)
{
    MpegEncContext *s = &v->s;
    GetBitContext *gb = &s->gb;
    int i, j;
    int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
    int cbp; /* cbp decoding stuff */
    int mqdiff, mquant; /* MB quantization */
    int ttmb = v->ttfrm; /* MB Transform type */
    int status;

    static const int size_table[6] = { 0, 2, 3, 4, 5, 8 },
      offset_table[6] = { 0, 1, 3, 7, 15, 31 };
    int mb_has_coeffs = 1; /* last_flag */
    int dmv_x, dmv_y; /* Differential MV components */
    int index, index1; /* LUT indices */
    int val, sign; /* temp values */
    int first_block = 1;
    int dst_idx, off;
    int skipped, fourmv;

    mquant = v->pq; /* Loosy initialization */

    if (v->mv_type_is_raw)
        fourmv = get_bits1(gb);
    else
        fourmv = v->mv_type_mb_plane[mb_pos];
    if (v->skip_is_raw)
        skipped = get_bits1(gb);
    else
        skipped = v->s.mbskip_table[mb_pos];

    s->dsp.clear_blocks(s->block[0]);

    if (!fourmv) /* 1MV mode */
    {
        if (!skipped)
        {
            GET_MVDATA(dmv_x, dmv_y);

            if (s->mb_intra) {
                s->current_picture.motion_val[1][s->block_index[0]][0] = 0;
                s->current_picture.motion_val[1][s->block_index[0]][1] = 0;
            }
            s->current_picture.mb_type[mb_pos] = s->mb_intra ? MB_TYPE_INTRA : MB_TYPE_16x16;
            vc1_pred_mv(s, 0, dmv_x, dmv_y, 1, v->range_x, v->range_y, v->mb_type[0]);

            /* FIXME Set DC val for inter block ? */
            if (s->mb_intra && !mb_has_coeffs)
            {
                GET_MQUANT();
                s->ac_pred = get_bits(gb, 1);
                cbp = 0;
            }
            else if (mb_has_coeffs)
            {
                if (s->mb_intra) s->ac_pred = get_bits(gb, 1);
                cbp = get_vlc2(&v->s.gb, v->cbpcy_vlc->table, VC1_CBPCY_P_VLC_BITS, 2);
                GET_MQUANT();
            }
            else
            {
                mquant = v->pq;
                cbp = 0;
            }
            s->current_picture.qscale_table[mb_pos] = mquant;

            if (!v->ttmbf && !s->mb_intra && mb_has_coeffs)
                ttmb = get_vlc2(gb, ff_vc1_ttmb_vlc[v->tt_index].table,
                                VC1_TTMB_VLC_BITS, 2);
            if(!s->mb_intra) vc1_mc_1mv(v, 0);
            dst_idx = 0;
            for (i=0; i<6; i++)
            {
                s->dc_val[0][s->block_index[i]] = 0;
                dst_idx += i >> 2;
                val = ((cbp >> (5 - i)) & 1);
                off = (i & 4) ? 0 : ((i & 1) * 8 + (i & 2) * 4 * s->linesize);
                v->mb_type[0][s->block_index[i]] = s->mb_intra;
                if(s->mb_intra) {
                    /* check if prediction blocks A and C are available */
                    v->a_avail = v->c_avail = 0;
                    if(i == 2 || i == 3 || !s->first_slice_line)
                        v->a_avail = v->mb_type[0][s->block_index[i] - s->block_wrap[i]];
                    if(i == 1 || i == 3 || s->mb_x)
                        v->c_avail = v->mb_type[0][s->block_index[i] - 1];

                    vc1_decode_intra_block(v, s->block[i], i, val, mquant, (i&4)?v->codingset2:v->codingset);
                    if((i>3) && (s->flags & CODEC_FLAG_GRAY)) continue;
                    s->dsp.vc1_inv_trans_8x8(s->block[i]);
                    if(v->rangeredfrm) for(j = 0; j < 64; j++) s->block[i][j] <<= 1;
                    for(j = 0; j < 64; j++) s->block[i][j] += 128;
                    if(!v->res_fasttx && v->res_x8) for(j = 0; j < 64; j++) s->block[i][j] += 16;
                    s->dsp.put_pixels_clamped(s->block[i], s->dest[dst_idx] + off, s->linesize >> ((i & 4) >> 2));
                    if(v->pq >= 9 && v->overlap) {
                        if(v->c_avail)
                            s->dsp.vc1_h_overlap(s->dest[dst_idx] + off, s->linesize >> ((i & 4) >> 2));
                        if(v->a_avail)
                            s->dsp.vc1_v_overlap(s->dest[dst_idx] + off, s->linesize >> ((i & 4) >> 2));
                    }
                } else if(val) {
                    vc1_decode_p_block(v, s->block[i], i, mquant, ttmb, first_block);
                    if(!v->ttmbf && ttmb < 8) ttmb = -1;
                    first_block = 0;
                    if((i<4) || !(s->flags & CODEC_FLAG_GRAY))
                        s->dsp.add_pixels_clamped(s->block[i], s->dest[dst_idx] + off, (i&4)?s->uvlinesize:s->linesize);
                }
            }
        }
        else //Skipped
        {
            s->mb_intra = 0;
            for(i = 0; i < 6; i++) {
                v->mb_type[0][s->block_index[i]] = 0;
                s->dc_val[0][s->block_index[i]] = 0;
            }
            s->current_picture.mb_type[mb_pos] = MB_TYPE_SKIP;
            s->current_picture.qscale_table[mb_pos] = 0;
            vc1_pred_mv(s, 0, 0, 0, 1, v->range_x, v->range_y, v->mb_type[0]);
            vc1_mc_1mv(v, 0);
            return 0;
        }
    } //1MV mode
    else //4MV mode
    {
        if (!skipped /* unskipped MB */)
        {
            int intra_count = 0, coded_inter = 0;
            int is_intra[6], is_coded[6];
            /* Get CBPCY */
            cbp = get_vlc2(&v->s.gb, v->cbpcy_vlc->table, VC1_CBPCY_P_VLC_BITS, 2);
            for (i=0; i<6; i++)
            {
                val = ((cbp >> (5 - i)) & 1);
                s->dc_val[0][s->block_index[i]] = 0;
                s->mb_intra = 0;
                if(i < 4) {
                    dmv_x = dmv_y = 0;
                    s->mb_intra = 0;
                    mb_has_coeffs = 0;
                    if(val) {
                        GET_MVDATA(dmv_x, dmv_y);
                    }
                    vc1_pred_mv(s, i, dmv_x, dmv_y, 0, v->range_x, v->range_y, v->mb_type[0]);
                    if(!s->mb_intra) vc1_mc_4mv_luma(v, i);
                    intra_count += s->mb_intra;
                    is_intra[i] = s->mb_intra;
                    is_coded[i] = mb_has_coeffs;
                }
                if(i&4){
                    is_intra[i] = (intra_count >= 3);
                    is_coded[i] = val;
                }
                if(i == 4) vc1_mc_4mv_chroma(v);
                v->mb_type[0][s->block_index[i]] = is_intra[i];
                if(!coded_inter) coded_inter = !is_intra[i] & is_coded[i];
            }
            // if there are no coded blocks then don't do anything more
            if(!intra_count && !coded_inter) return 0;
            dst_idx = 0;
            GET_MQUANT();
            s->current_picture.qscale_table[mb_pos] = mquant;
            /* test if block is intra and has pred */
            {
                int intrapred = 0;
                for(i=0; i<6; i++)
                    if(is_intra[i]) {
                        if(((!s->first_slice_line || (i==2 || i==3)) && v->mb_type[0][s->block_index[i] - s->block_wrap[i]])
                            || ((s->mb_x || (i==1 || i==3)) && v->mb_type[0][s->block_index[i] - 1])) {
                            intrapred = 1;
                            break;
                        }
                    }
                if(intrapred)s->ac_pred = get_bits(gb, 1);
                else s->ac_pred = 0;
            }
            if (!v->ttmbf && coded_inter)
                ttmb = get_vlc2(gb, ff_vc1_ttmb_vlc[v->tt_index].table, VC1_TTMB_VLC_BITS, 2);
            for (i=0; i<6; i++)
            {
                dst_idx += i >> 2;
                off = (i & 4) ? 0 : ((i & 1) * 8 + (i & 2) * 4 * s->linesize);
                s->mb_intra = is_intra[i];
                if (is_intra[i]) {
                    /* check if prediction blocks A and C are available */
                    v->a_avail = v->c_avail = 0;
                    if(i == 2 || i == 3 || !s->first_slice_line)
                        v->a_avail = v->mb_type[0][s->block_index[i] - s->block_wrap[i]];
                    if(i == 1 || i == 3 || s->mb_x)
                        v->c_avail = v->mb_type[0][s->block_index[i] - 1];

                    vc1_decode_intra_block(v, s->block[i], i, is_coded[i], mquant, (i&4)?v->codingset2:v->codingset);
                    if((i>3) && (s->flags & CODEC_FLAG_GRAY)) continue;
                    s->dsp.vc1_inv_trans_8x8(s->block[i]);
                    if(v->rangeredfrm) for(j = 0; j < 64; j++) s->block[i][j] <<= 1;
                    for(j = 0; j < 64; j++) s->block[i][j] += 128;
                    if(!v->res_fasttx && v->res_x8) for(j = 0; j < 64; j++) s->block[i][j] += 16;
                    s->dsp.put_pixels_clamped(s->block[i], s->dest[dst_idx] + off, (i&4)?s->uvlinesize:s->linesize);
                    if(v->pq >= 9 && v->overlap) {
                        if(v->c_avail)
                            s->dsp.vc1_h_overlap(s->dest[dst_idx] + off, s->linesize >> ((i & 4) >> 2));
                        if(v->a_avail)
                            s->dsp.vc1_v_overlap(s->dest[dst_idx] + off, s->linesize >> ((i & 4) >> 2));
                    }
                } else if(is_coded[i]) {
                    status = vc1_decode_p_block(v, s->block[i], i, mquant, ttmb, first_block);
                    if(!v->ttmbf && ttmb < 8) ttmb = -1;
                    first_block = 0;
                    if((i<4) || !(s->flags & CODEC_FLAG_GRAY))
                        s->dsp.add_pixels_clamped(s->block[i], s->dest[dst_idx] + off, (i&4)?s->uvlinesize:s->linesize);
                }
            }
            return status;
        }
        else //Skipped MB
        {
            s->mb_intra = 0;
            s->current_picture.qscale_table[mb_pos] = 0;
            for (i=0; i<6; i++) {
                v->mb_type[0][s->block_index[i]] = 0;
                s->dc_val[0][s->block_index[i]] = 0;
            }
            for (i=0; i<4; i++)
            {
                vc1_pred_mv(s, i, 0, 0, 0, v->range_x, v->range_y, v->mb_type[0]);
                vc1_mc_4mv_luma(v, i);
            }
            vc1_mc_4mv_chroma(v);
            s->current_picture.qscale_table[mb_pos] = 0;
            return 0;
        }
    }

    /* Should never happen */
    return -1;
}

/** Decode one B-frame MB (in Main profile)
 */
static void vc1_decode_b_mb(VC1Context *v)
{
    MpegEncContext *s = &v->s;
    GetBitContext *gb = &s->gb;
    int i, j;
    int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
    int cbp = 0; /* cbp decoding stuff */
    int mqdiff, mquant; /* MB quantization */
    int ttmb = v->ttfrm; /* MB Transform type */

    static const int size_table[6] = { 0, 2, 3, 4, 5, 8 },
      offset_table[6] = { 0, 1, 3, 7, 15, 31 };
    int mb_has_coeffs = 0; /* last_flag */
    int index, index1; /* LUT indices */
    int val, sign; /* temp values */
    int first_block = 1;
    int dst_idx, off;
    int skipped, direct;
    int dmv_x[2], dmv_y[2];
    int bmvtype = BMV_TYPE_BACKWARD;

    mquant = v->pq; /* Loosy initialization */
    s->mb_intra = 0;

    if (v->dmb_is_raw)
        direct = get_bits1(gb);
    else
        direct = v->direct_mb_plane[mb_pos];
    if (v->skip_is_raw)
        skipped = get_bits1(gb);
    else
        skipped = v->s.mbskip_table[mb_pos];

    s->dsp.clear_blocks(s->block[0]);
    dmv_x[0] = dmv_x[1] = dmv_y[0] = dmv_y[1] = 0;
    for(i = 0; i < 6; i++) {
        v->mb_type[0][s->block_index[i]] = 0;
        s->dc_val[0][s->block_index[i]] = 0;
    }
    s->current_picture.qscale_table[mb_pos] = 0;

    if (!direct) {
        if (!skipped) {
            GET_MVDATA(dmv_x[0], dmv_y[0]);
            dmv_x[1] = dmv_x[0];
            dmv_y[1] = dmv_y[0];
        }
        if(skipped || !s->mb_intra) {
            bmvtype = decode012(gb);
            switch(bmvtype) {
            case 0:
                bmvtype = (v->bfraction >= (B_FRACTION_DEN/2)) ? BMV_TYPE_BACKWARD : BMV_TYPE_FORWARD;
                break;
            case 1:
                bmvtype = (v->bfraction >= (B_FRACTION_DEN/2)) ? BMV_TYPE_FORWARD : BMV_TYPE_BACKWARD;
                break;
            case 2:
                bmvtype = BMV_TYPE_INTERPOLATED;
                dmv_x[0] = dmv_y[0] = 0;
            }
        }
    }
    for(i = 0; i < 6; i++)
        v->mb_type[0][s->block_index[i]] = s->mb_intra;

    if (skipped) {
        if(direct) bmvtype = BMV_TYPE_INTERPOLATED;
        vc1_pred_b_mv(v, dmv_x, dmv_y, direct, bmvtype);
        vc1_b_mc(v, dmv_x, dmv_y, direct, bmvtype);
        return;
    }
    if (direct) {
        cbp = get_vlc2(&v->s.gb, v->cbpcy_vlc->table, VC1_CBPCY_P_VLC_BITS, 2);
        GET_MQUANT();
        s->mb_intra = 0;
        mb_has_coeffs = 0;
        s->current_picture.qscale_table[mb_pos] = mquant;
        if(!v->ttmbf)
            ttmb = get_vlc2(gb, ff_vc1_ttmb_vlc[v->tt_index].table, VC1_TTMB_VLC_BITS, 2);
        dmv_x[0] = dmv_y[0] = dmv_x[1] = dmv_y[1] = 0;
        vc1_pred_b_mv(v, dmv_x, dmv_y, direct, bmvtype);
        vc1_b_mc(v, dmv_x, dmv_y, direct, bmvtype);
    } else {
        if(!mb_has_coeffs && !s->mb_intra) {
            /* no coded blocks - effectively skipped */
            vc1_pred_b_mv(v, dmv_x, dmv_y, direct, bmvtype);
            vc1_b_mc(v, dmv_x, dmv_y, direct, bmvtype);
            return;
        }
        if(s->mb_intra && !mb_has_coeffs) {
            GET_MQUANT();
            s->current_picture.qscale_table[mb_pos] = mquant;
            s->ac_pred = get_bits1(gb);
            cbp = 0;
            vc1_pred_b_mv(v, dmv_x, dmv_y, direct, bmvtype);
        } else {
            if(bmvtype == BMV_TYPE_INTERPOLATED) {
                GET_MVDATA(dmv_x[0], dmv_y[0]);
                if(!mb_has_coeffs) {
                    /* interpolated skipped block */
                    vc1_pred_b_mv(v, dmv_x, dmv_y, direct, bmvtype);
                    vc1_b_mc(v, dmv_x, dmv_y, direct, bmvtype);
                    return;
                }
            }
            vc1_pred_b_mv(v, dmv_x, dmv_y, direct, bmvtype);
            if(!s->mb_intra) {
                vc1_b_mc(v, dmv_x, dmv_y, direct, bmvtype);
            }
            if(s->mb_intra)
                s->ac_pred = get_bits1(gb);
            cbp = get_vlc2(&v->s.gb, v->cbpcy_vlc->table, VC1_CBPCY_P_VLC_BITS, 2);
            GET_MQUANT();
            s->current_picture.qscale_table[mb_pos] = mquant;
            if(!v->ttmbf && !s->mb_intra && mb_has_coeffs)
                ttmb = get_vlc2(gb, ff_vc1_ttmb_vlc[v->tt_index].table, VC1_TTMB_VLC_BITS, 2);
        }
    }
    dst_idx = 0;
    for (i=0; i<6; i++)
    {
        s->dc_val[0][s->block_index[i]] = 0;
        dst_idx += i >> 2;
        val = ((cbp >> (5 - i)) & 1);
        off = (i & 4) ? 0 : ((i & 1) * 8 + (i & 2) * 4 * s->linesize);
        v->mb_type[0][s->block_index[i]] = s->mb_intra;
        if(s->mb_intra) {
            /* check if prediction blocks A and C are available */
            v->a_avail = v->c_avail = 0;
            if(i == 2 || i == 3 || !s->first_slice_line)
                v->a_avail = v->mb_type[0][s->block_index[i] - s->block_wrap[i]];
            if(i == 1 || i == 3 || s->mb_x)
                v->c_avail = v->mb_type[0][s->block_index[i] - 1];

            vc1_decode_intra_block(v, s->block[i], i, val, mquant, (i&4)?v->codingset2:v->codingset);
            if((i>3) && (s->flags & CODEC_FLAG_GRAY)) continue;
            s->dsp.vc1_inv_trans_8x8(s->block[i]);
            if(v->rangeredfrm) for(j = 0; j < 64; j++) s->block[i][j] <<= 1;
            for(j = 0; j < 64; j++) s->block[i][j] += 128;
            s->dsp.put_pixels_clamped(s->block[i], s->dest[dst_idx] + off, s->linesize >> ((i & 4) >> 2));
        } else if(val) {
            vc1_decode_p_block(v, s->block[i], i, mquant, ttmb, first_block);
            if(!v->ttmbf && ttmb < 8) ttmb = -1;
            first_block = 0;
            if((i<4) || !(s->flags & CODEC_FLAG_GRAY))
                s->dsp.add_pixels_clamped(s->block[i], s->dest[dst_idx] + off, (i&4)?s->uvlinesize:s->linesize);
        }
    }
}

/** Decode blocks of I-frame
 */
static void vc1_decode_i_blocks(VC1Context *v)
{
    int k, j;
    MpegEncContext *s = &v->s;
    int cbp, val;
    uint8_t *coded_val;
    int mb_pos;

    /* select codingmode used for VLC tables selection */
    switch(v->y_ac_table_index){
    case 0:
        v->codingset = (v->pqindex <= 8) ? CS_HIGH_RATE_INTRA : CS_LOW_MOT_INTRA;
        break;
    case 1:
        v->codingset = CS_HIGH_MOT_INTRA;
        break;
    case 2:
        v->codingset = CS_MID_RATE_INTRA;
        break;
    }

    switch(v->c_ac_table_index){
    case 0:
        v->codingset2 = (v->pqindex <= 8) ? CS_HIGH_RATE_INTER : CS_LOW_MOT_INTER;
        break;
    case 1:
        v->codingset2 = CS_HIGH_MOT_INTER;
        break;
    case 2:
        v->codingset2 = CS_MID_RATE_INTER;
        break;
    }

    /* Set DC scale - y and c use the same */
    s->y_dc_scale = s->y_dc_scale_table[v->pq];
    s->c_dc_scale = s->c_dc_scale_table[v->pq];

    //do frame decode
    s->mb_x = s->mb_y = 0;
    s->mb_intra = 1;
    s->first_slice_line = 1;
    ff_er_add_slice(s, 0, 0, s->mb_width - 1, s->mb_height - 1, (AC_END|DC_END|MV_END));
    for(s->mb_y = 0; s->mb_y < s->mb_height; s->mb_y++) {
        for(s->mb_x = 0; s->mb_x < s->mb_width; s->mb_x++) {
            ff_init_block_index(s);
            ff_update_block_index(s);
            s->dsp.clear_blocks(s->block[0]);
            mb_pos = s->mb_x + s->mb_y * s->mb_width;
            s->current_picture.mb_type[mb_pos] = MB_TYPE_INTRA;
            s->current_picture.qscale_table[mb_pos] = v->pq;
            s->current_picture.motion_val[1][s->block_index[0]][0] = 0;
            s->current_picture.motion_val[1][s->block_index[0]][1] = 0;

            // do actual MB decoding and displaying
            cbp = get_vlc2(&v->s.gb, ff_msmp4_mb_i_vlc.table, MB_INTRA_VLC_BITS, 2);
            v->s.ac_pred = get_bits(&v->s.gb, 1);

            for(k = 0; k < 6; k++) {
                val = ((cbp >> (5 - k)) & 1);

                if (k < 4) {
                    int pred = vc1_coded_block_pred(&v->s, k, &coded_val);
                    val = val ^ pred;
                    *coded_val = val;
                }
                cbp |= val << (5 - k);

                vc1_decode_i_block(v, s->block[k], k, val, (k<4)? v->codingset : v->codingset2);

                s->dsp.vc1_inv_trans_8x8(s->block[k]);
                if(!v->res_fasttx && !v->res_x8) for(j = 0; j < 64; j++) s->block[k][j] -= 16;
                if(v->pq >= 9 && v->overlap) {
                    for(j = 0; j < 64; j++) s->block[k][j] += 128;
                }
            }

            vc1_put_block(v, s->block);
            if(v->pq >= 9 && v->overlap) {
                if(s->mb_x) {
                    s->dsp.vc1_h_overlap(s->dest[0], s->linesize);
                    s->dsp.vc1_h_overlap(s->dest[0] + 8 * s->linesize, s->linesize);
                    if(!(s->flags & CODEC_FLAG_GRAY)) {
                        s->dsp.vc1_h_overlap(s->dest[1], s->uvlinesize);
                        s->dsp.vc1_h_overlap(s->dest[2], s->uvlinesize);
                    }
                }
                s->dsp.vc1_h_overlap(s->dest[0] + 8, s->linesize);
                s->dsp.vc1_h_overlap(s->dest[0] + 8 * s->linesize + 8, s->linesize);
                if(!s->first_slice_line) {
                    s->dsp.vc1_v_overlap(s->dest[0], s->linesize);
                    s->dsp.vc1_v_overlap(s->dest[0] + 8, s->linesize);
                    if(!(s->flags & CODEC_FLAG_GRAY)) {
                        s->dsp.vc1_v_overlap(s->dest[1], s->uvlinesize);
                        s->dsp.vc1_v_overlap(s->dest[2], s->uvlinesize);
                    }
                }
                s->dsp.vc1_v_overlap(s->dest[0] + 8 * s->linesize, s->linesize);
                s->dsp.vc1_v_overlap(s->dest[0] + 8 * s->linesize + 8, s->linesize);
            }

            if(get_bits_count(&s->gb) > v->bits) {
                av_log(s->avctx, AV_LOG_ERROR, "Bits overconsumption: %i > %i\n", get_bits_count(&s->gb), v->bits);
                return;
            }
        }
        ff_draw_horiz_band(s, s->mb_y * 16, 16);
        s->first_slice_line = 0;
    }
}

/** Decode blocks of I-frame for advanced profile
 */
static void vc1_decode_i_blocks_adv(VC1Context *v)
{
    int k, j;
    MpegEncContext *s = &v->s;
    int cbp, val;
    uint8_t *coded_val;
    int mb_pos;
    int mquant = v->pq;
    int mqdiff;
    int overlap;
    GetBitContext *gb = &s->gb;

    /* select codingmode used for VLC tables selection */
    switch(v->y_ac_table_index){
    case 0:
        v->codingset = (v->pqindex <= 8) ? CS_HIGH_RATE_INTRA : CS_LOW_MOT_INTRA;
        break;
    case 1:
        v->codingset = CS_HIGH_MOT_INTRA;
        break;
    case 2:
        v->codingset = CS_MID_RATE_INTRA;
        break;
    }

    switch(v->c_ac_table_index){
    case 0:
        v->codingset2 = (v->pqindex <= 8) ? CS_HIGH_RATE_INTER : CS_LOW_MOT_INTER;
        break;
    case 1:
        v->codingset2 = CS_HIGH_MOT_INTER;
        break;
    case 2:
        v->codingset2 = CS_MID_RATE_INTER;
        break;
    }

    //do frame decode
    s->mb_x = s->mb_y = 0;
    s->mb_intra = 1;
    s->first_slice_line = 1;
    ff_er_add_slice(s, 0, 0, s->mb_width - 1, s->mb_height - 1, (AC_END|DC_END|MV_END));
    for(s->mb_y = 0; s->mb_y < s->mb_height; s->mb_y++) {
        for(s->mb_x = 0; s->mb_x < s->mb_width; s->mb_x++) {
            ff_init_block_index(s);
            ff_update_block_index(s);
            s->dsp.clear_blocks(s->block[0]);
            mb_pos = s->mb_x + s->mb_y * s->mb_stride;
            s->current_picture.mb_type[mb_pos] = MB_TYPE_INTRA;
            s->current_picture.motion_val[1][s->block_index[0]][0] = 0;
            s->current_picture.motion_val[1][s->block_index[0]][1] = 0;

            // do actual MB decoding and displaying
            cbp = get_vlc2(&v->s.gb, ff_msmp4_mb_i_vlc.table, MB_INTRA_VLC_BITS, 2);
            if(v->acpred_is_raw)
                v->s.ac_pred = get_bits(&v->s.gb, 1);
            else
                v->s.ac_pred = v->acpred_plane[mb_pos];

            if(v->condover == CONDOVER_SELECT) {
                if(v->overflg_is_raw)
                    overlap = get_bits(&v->s.gb, 1);
                else
                    overlap = v->over_flags_plane[mb_pos];
            } else
                overlap = (v->condover == CONDOVER_ALL);

            GET_MQUANT();

            s->current_picture.qscale_table[mb_pos] = mquant;
            /* Set DC scale - y and c use the same */
            s->y_dc_scale = s->y_dc_scale_table[mquant];
            s->c_dc_scale = s->c_dc_scale_table[mquant];

            for(k = 0; k < 6; k++) {
                val = ((cbp >> (5 - k)) & 1);

                if (k < 4) {
                    int pred = vc1_coded_block_pred(&v->s, k, &coded_val);
                    val = val ^ pred;
                    *coded_val = val;
                }
                cbp |= val << (5 - k);

                v->a_avail = !s->first_slice_line || (k==2 || k==3);
                v->c_avail = !!s->mb_x || (k==1 || k==3);

                vc1_decode_i_block_adv(v, s->block[k], k, val, (k<4)? v->codingset : v->codingset2, mquant);

                s->dsp.vc1_inv_trans_8x8(s->block[k]);
                for(j = 0; j < 64; j++) s->block[k][j] += 128;
            }

            vc1_put_block(v, s->block);
            if(overlap) {
                if(s->mb_x) {
                    s->dsp.vc1_h_overlap(s->dest[0], s->linesize);
                    s->dsp.vc1_h_overlap(s->dest[0] + 8 * s->linesize, s->linesize);
                    if(!(s->flags & CODEC_FLAG_GRAY)) {
                        s->dsp.vc1_h_overlap(s->dest[1], s->uvlinesize);
                        s->dsp.vc1_h_overlap(s->dest[2], s->uvlinesize);
                    }
                }
                s->dsp.vc1_h_overlap(s->dest[0] + 8, s->linesize);
                s->dsp.vc1_h_overlap(s->dest[0] + 8 * s->linesize + 8, s->linesize);
                if(!s->first_slice_line) {
                    s->dsp.vc1_v_overlap(s->dest[0], s->linesize);
                    s->dsp.vc1_v_overlap(s->dest[0] + 8, s->linesize);
                    if(!(s->flags & CODEC_FLAG_GRAY)) {
                        s->dsp.vc1_v_overlap(s->dest[1], s->uvlinesize);
                        s->dsp.vc1_v_overlap(s->dest[2], s->uvlinesize);
                    }
                }
                s->dsp.vc1_v_overlap(s->dest[0] + 8 * s->linesize, s->linesize);
                s->dsp.vc1_v_overlap(s->dest[0] + 8 * s->linesize + 8, s->linesize);
            }

            if(get_bits_count(&s->gb) > v->bits) {
                av_log(s->avctx, AV_LOG_ERROR, "Bits overconsumption: %i > %i\n", get_bits_count(&s->gb), v->bits);
                return;
            }
        }
        ff_draw_horiz_band(s, s->mb_y * 16, 16);
        s->first_slice_line = 0;
    }
}

static void vc1_decode_p_blocks(VC1Context *v)
{
    MpegEncContext *s = &v->s;

    /* select codingmode used for VLC tables selection */
    switch(v->c_ac_table_index){
    case 0:
        v->codingset = (v->pqindex <= 8) ? CS_HIGH_RATE_INTRA : CS_LOW_MOT_INTRA;
        break;
    case 1:
        v->codingset = CS_HIGH_MOT_INTRA;
        break;
    case 2:
        v->codingset = CS_MID_RATE_INTRA;
        break;
    }

    switch(v->c_ac_table_index){
    case 0:
        v->codingset2 = (v->pqindex <= 8) ? CS_HIGH_RATE_INTER : CS_LOW_MOT_INTER;
        break;
    case 1:
        v->codingset2 = CS_HIGH_MOT_INTER;
        break;
    case 2:
        v->codingset2 = CS_MID_RATE_INTER;
        break;
    }

    ff_er_add_slice(s, 0, 0, s->mb_width - 1, s->mb_height - 1, (AC_END|DC_END|MV_END));
    s->first_slice_line = 1;
    for(s->mb_y = 0; s->mb_y < s->mb_height; s->mb_y++) {
        for(s->mb_x = 0; s->mb_x < s->mb_width; s->mb_x++) {
            ff_init_block_index(s);
            ff_update_block_index(s);
            s->dsp.clear_blocks(s->block[0]);

            vc1_decode_p_mb(v);
            if(get_bits_count(&s->gb) > v->bits || get_bits_count(&s->gb) < 0) {
                av_log(s->avctx, AV_LOG_ERROR, "Bits overconsumption: %i > %i at %ix%i\n", get_bits_count(&s->gb), v->bits,s->mb_x,s->mb_y);
                return;
            }
        }
        ff_draw_horiz_band(s, s->mb_y * 16, 16);
        s->first_slice_line = 0;
    }
}

static void vc1_decode_b_blocks(VC1Context *v)
{
    MpegEncContext *s = &v->s;

    /* select codingmode used for VLC tables selection */
    switch(v->c_ac_table_index){
    case 0:
        v->codingset = (v->pqindex <= 8) ? CS_HIGH_RATE_INTRA : CS_LOW_MOT_INTRA;
        break;
    case 1:
        v->codingset = CS_HIGH_MOT_INTRA;
        break;
    case 2:
        v->codingset = CS_MID_RATE_INTRA;
        break;
    }

    switch(v->c_ac_table_index){
    case 0:
        v->codingset2 = (v->pqindex <= 8) ? CS_HIGH_RATE_INTER : CS_LOW_MOT_INTER;
        break;
    case 1:
        v->codingset2 = CS_HIGH_MOT_INTER;
        break;
    case 2:
        v->codingset2 = CS_MID_RATE_INTER;
        break;
    }

    ff_er_add_slice(s, 0, 0, s->mb_width - 1, s->mb_height - 1, (AC_END|DC_END|MV_END));
    s->first_slice_line = 1;
    for(s->mb_y = 0; s->mb_y < s->mb_height; s->mb_y++) {
        for(s->mb_x = 0; s->mb_x < s->mb_width; s->mb_x++) {
            ff_init_block_index(s);
            ff_update_block_index(s);
            s->dsp.clear_blocks(s->block[0]);

            vc1_decode_b_mb(v);
            if(get_bits_count(&s->gb) > v->bits || get_bits_count(&s->gb) < 0) {
                av_log(s->avctx, AV_LOG_ERROR, "Bits overconsumption: %i > %i at %ix%i\n", get_bits_count(&s->gb), v->bits,s->mb_x,s->mb_y);
                return;
            }
        }
        ff_draw_horiz_band(s, s->mb_y * 16, 16);
        s->first_slice_line = 0;
    }
}

static void vc1_decode_skip_blocks(VC1Context *v)
{
    MpegEncContext *s = &v->s;

    ff_er_add_slice(s, 0, 0, s->mb_width - 1, s->mb_height - 1, (AC_END|DC_END|MV_END));
    s->first_slice_line = 1;
    for(s->mb_y = 0; s->mb_y < s->mb_height; s->mb_y++) {
        s->mb_x = 0;
        ff_init_block_index(s);
        ff_update_block_index(s);
        memcpy(s->dest[0], s->last_picture.data[0] + s->mb_y * 16 * s->linesize, s->linesize * 16);
        memcpy(s->dest[1], s->last_picture.data[1] + s->mb_y * 8 * s->uvlinesize, s->uvlinesize * 8);
        memcpy(s->dest[2], s->last_picture.data[2] + s->mb_y * 8 * s->uvlinesize, s->uvlinesize * 8);
        ff_draw_horiz_band(s, s->mb_y * 16, 16);
        s->first_slice_line = 0;
    }
    s->pict_type = P_TYPE;
}

static void vc1_decode_blocks(VC1Context *v)
{

    v->s.esc3_level_length = 0;

    switch(v->s.pict_type) {
    case I_TYPE:
        if(v->profile == PROFILE_ADVANCED)
            vc1_decode_i_blocks_adv(v);
        else
            vc1_decode_i_blocks(v);
        break;
    case P_TYPE:
        if(v->p_frame_skipped)
            vc1_decode_skip_blocks(v);
        else
            vc1_decode_p_blocks(v);
        break;
    case B_TYPE:
        if(v->bi_type){
            if(v->profile == PROFILE_ADVANCED)
                vc1_decode_i_blocks_adv(v);
            else
                vc1_decode_i_blocks(v);
        }else
            vc1_decode_b_blocks(v);
        break;
    }
}

/** Find VC-1 marker in buffer
 * @return position where next marker starts or end of buffer if no marker found
 */
static av_always_inline uint8_t* find_next_marker(uint8_t *src, uint8_t *end)
{
    uint32_t mrk = 0xFFFFFFFF;

    if(end-src < 4) return end;
    while(src < end){
        mrk = (mrk << 8) | *src++;
        if(IS_MARKER(mrk))
            return src-4;
    }
    return end;
}

static av_always_inline int vc1_unescape_buffer(uint8_t *src, int size, uint8_t *dst)
{
    int dsize = 0, i;

    if(size < 4){
        for(dsize = 0; dsize < size; dsize++) *dst++ = *src++;
        return size;
    }
    for(i = 0; i < size; i++, src++) {
        if(src[0] == 3 && i >= 2 && !src[-1] && !src[-2] && i < size-1 && src[1] < 4) {
            dst[dsize++] = src[1];
            src++;
            i++;
        } else
            dst[dsize++] = *src;
    }
    return dsize;
}

/** Initialize a VC1/WMV3 decoder
 * @todo TODO: Handle VC-1 IDUs (Transport level?)
 * @todo TODO: Decypher remaining bits in extra_data
 */
static int vc1_decode_init(AVCodecContext *avctx)
{
    VC1Context *v = avctx->priv_data;
    MpegEncContext *s = &v->s;
    GetBitContext gb;

    if (!avctx->extradata_size || !avctx->extradata) return -1;
    if (!(avctx->flags & CODEC_FLAG_GRAY))
        avctx->pix_fmt = PIX_FMT_YUV420P;
    else
        avctx->pix_fmt = PIX_FMT_GRAY8;
    v->s.avctx = avctx;
    avctx->flags |= CODEC_FLAG_EMU_EDGE;
    v->s.flags |= CODEC_FLAG_EMU_EDGE;

    if(ff_h263_decode_init(avctx) < 0)
        return -1;
    if (vc1_init_common(v) < 0) return -1;

    avctx->coded_width = avctx->width;
    avctx->coded_height = avctx->height;
    if (avctx->codec_id == CODEC_ID_WMV3)
    {
        int count = 0;

        // looks like WMV3 has a sequence header stored in the extradata
        // advanced sequence header may be before the first frame
        // the last byte of the extradata is a version number, 1 for the
        // samples we can decode

        init_get_bits(&gb, avctx->extradata, avctx->extradata_size*8);

        if (decode_sequence_header(avctx, &gb) < 0)
          return -1;

        count = avctx->extradata_size*8 - get_bits_count(&gb);
        if (count>0)
        {
            av_log(avctx, AV_LOG_INFO, "Extra data: %i bits left, value: %X\n",
                   count, get_bits(&gb, count));
        }
        else if (count < 0)
        {
            av_log(avctx, AV_LOG_INFO, "Read %i bits in overflow\n", -count);
        }
    } else { // VC1/WVC1
        uint8_t *start = avctx->extradata, *end = avctx->extradata + avctx->extradata_size;
        uint8_t *next; int size, buf2_size;
        uint8_t *buf2 = NULL;
        int seq_inited = 0, ep_inited = 0;

        if(avctx->extradata_size < 16) {
            av_log(avctx, AV_LOG_ERROR, "Extradata size too small: %i\n", avctx->extradata_size);
            return -1;
        }

        buf2 = av_mallocz(avctx->extradata_size + FF_INPUT_BUFFER_PADDING_SIZE);
        if(start[0]) start++; // in WVC1 extradata first byte is its size
        next = start;
        for(; next < end; start = next){
            next = find_next_marker(start + 4, end);
            size = next - start - 4;
            if(size <= 0) continue;
            buf2_size = vc1_unescape_buffer(start + 4, size, buf2);
            init_get_bits(&gb, buf2, buf2_size * 8);
            switch(AV_RB32(start)){
            case VC1_CODE_SEQHDR:
                if(decode_sequence_header(avctx, &gb) < 0){
                    av_free(buf2);
                    return -1;
                }
                seq_inited = 1;
                break;
            case VC1_CODE_ENTRYPOINT:
                if(decode_entry_point(avctx, &gb) < 0){
                    av_free(buf2);
                    return -1;
                }
                ep_inited = 1;
                break;
            }
        }
        av_free(buf2);
        if(!seq_inited || !ep_inited){
            av_log(avctx, AV_LOG_ERROR, "Incomplete extradata\n");
            return -1;
        }
    }
    avctx->has_b_frames= !!(avctx->max_b_frames);
    s->low_delay = !avctx->has_b_frames;

    s->mb_width = (avctx->coded_width+15)>>4;
    s->mb_height = (avctx->coded_height+15)>>4;

    /* Allocate mb bitplanes */
    v->mv_type_mb_plane = av_malloc(s->mb_stride * s->mb_height);
    v->direct_mb_plane = av_malloc(s->mb_stride * s->mb_height);
    v->acpred_plane = av_malloc(s->mb_stride * s->mb_height);
    v->over_flags_plane = av_malloc(s->mb_stride * s->mb_height);

    /* allocate block type info in that way so it could be used with s->block_index[] */
    v->mb_type_base = av_malloc(s->b8_stride * (s->mb_height * 2 + 1) + s->mb_stride * (s->mb_height + 1) * 2);
    v->mb_type[0] = v->mb_type_base + s->b8_stride + 1;
    v->mb_type[1] = v->mb_type_base + s->b8_stride * (s->mb_height * 2 + 1) + s->mb_stride + 1;
    v->mb_type[2] = v->mb_type[1] + s->mb_stride * (s->mb_height + 1);

    /* Init coded blocks info */
    if (v->profile == PROFILE_ADVANCED)
    {
//        if (alloc_bitplane(&v->over_flags_plane, s->mb_width, s->mb_height) < 0)
//            return -1;
//        if (alloc_bitplane(&v->ac_pred_plane, s->mb_width, s->mb_height) < 0)
//            return -1;
    }

    return 0;
}


/** Decode a VC1/WMV3 frame
 * @todo TODO: Handle VC-1 IDUs (Transport level?)
 */
static int vc1_decode_frame(AVCodecContext *avctx,
                            void *data, int *data_size,
                            uint8_t *buf, int buf_size)
{
    VC1Context *v = avctx->priv_data;
    MpegEncContext *s = &v->s;
    AVFrame *pict = data;
    uint8_t *buf2 = NULL;

    /* no supplementary picture */
    if (buf_size == 0) {