vc1.c

                qx = (s->mb_x << 5);
                qy = (s->mb_y << 5);
                X = (s->mb_width << 5) - 4;
                Y = (s->mb_height << 5) - 4;
                if(qx + px < -28) px = -28 - qx;
                if(qy + py < -28) py = -28 - qy;
                if(qx + px > X) px = X - qx;
                if(qy + py > Y) py = Y - qy;
            } else {
                qx = (s->mb_x << 6);
                qy = (s->mb_y << 6);
                X = (s->mb_width << 6) - 4;
                Y = (s->mb_height << 6) - 4;
                if(qx + px < -60) px = -60 - qx;
                if(qy + py < -60) py = -60 - qy;
                if(qx + px > X) px = X - qx;
                if(qy + py > Y) py = Y - qy;
            }
        }
        /* Calculate hybrid prediction as specified in 8.3.5.3.5 */
        if(0 && !s->first_slice_line && s->mb_x) {
            if(is_intra[xy - wrap])
                sum = FFABS(px) + FFABS(py);
            else
                sum = FFABS(px - A[0]) + FFABS(py - A[1]);
            if(sum > 32) {
                if(get_bits1(&s->gb)) {
                    px = A[0];
                    py = A[1];
                } else {
                    px = C[0];
                    py = C[1];
                }
            } else {
                if(is_intra[xy - 2])
                    sum = FFABS(px) + FFABS(py);
                else
                    sum = FFABS(px - C[0]) + FFABS(py - C[1]);
                if(sum > 32) {
                    if(get_bits1(&s->gb)) {
                        px = A[0];
                        py = A[1];
                    } else {
                        px = C[0];
                        py = C[1];
                    }
                }
            }
        }
        /* store MV using signed modulus of MV range defined in 4.11 */
        s->mv[0][0][0] = ((px + dmv_x[0] + r_x) & ((r_x << 1) - 1)) - r_x;
        s->mv[0][0][1] = ((py + dmv_y[0] + r_y) & ((r_y << 1) - 1)) - r_y;
    }
    if((mvtype == BMV_TYPE_BACKWARD) || (mvtype == BMV_TYPE_INTERPOLATED)) {
        C = s->current_picture.motion_val[1][xy - 2];
        A = s->current_picture.motion_val[1][xy - wrap*2];
        off = (s->mb_x == (s->mb_width - 1)) ? -2 : 2;
        B = s->current_picture.motion_val[1][xy - wrap*2 + off];

        if(!s->mb_x) C[0] = C[1] = 0;
        if(!s->first_slice_line) { // predictor A is not out of bounds
            if(s->mb_width == 1) {
                px = A[0];
                py = A[1];
            } else {
                px = mid_pred(A[0], B[0], C[0]);
                py = mid_pred(A[1], B[1], C[1]);
            }
        } else if(s->mb_x) { // predictor C is not out of bounds
            px = C[0];
            py = C[1];
        } else {
            px = py = 0;
        }
        /* Pullback MV as specified in 8.3.5.3.4 */
        {
            int qx, qy, X, Y;
            if(v->profile < PROFILE_ADVANCED) {
                qx = (s->mb_x << 5);
                qy = (s->mb_y << 5);
                X = (s->mb_width << 5) - 4;
                Y = (s->mb_height << 5) - 4;
                if(qx + px < -28) px = -28 - qx;
                if(qy + py < -28) py = -28 - qy;
                if(qx + px > X) px = X - qx;
                if(qy + py > Y) py = Y - qy;
            } else {
                qx = (s->mb_x << 6);
                qy = (s->mb_y << 6);
                X = (s->mb_width << 6) - 4;
                Y = (s->mb_height << 6) - 4;
                if(qx + px < -60) px = -60 - qx;
                if(qy + py < -60) py = -60 - qy;
                if(qx + px > X) px = X - qx;
                if(qy + py > Y) py = Y - qy;
            }
        }
        /* Calculate hybrid prediction as specified in 8.3.5.3.5 */
        if(0 && !s->first_slice_line && s->mb_x) {
            if(is_intra[xy - wrap])
                sum = FFABS(px) + FFABS(py);
            else
                sum = FFABS(px - A[0]) + FFABS(py - A[1]);
            if(sum > 32) {
                if(get_bits1(&s->gb)) {
                    px = A[0];
                    py = A[1];
                } else {
                    px = C[0];
                    py = C[1];
                }
            } else {
                if(is_intra[xy - 2])
                    sum = FFABS(px) + FFABS(py);
                else
                    sum = FFABS(px - C[0]) + FFABS(py - C[1]);
                if(sum > 32) {
                    if(get_bits1(&s->gb)) {
                        px = A[0];
                        py = A[1];
                    } else {
                        px = C[0];
                        py = C[1];
                    }
                }
            }
        }
        /* store MV using signed modulus of MV range defined in 4.11 */

        s->mv[1][0][0] = ((px + dmv_x[1] + r_x) & ((r_x << 1) - 1)) - r_x;
        s->mv[1][0][1] = ((py + dmv_y[1] + r_y) & ((r_y << 1) - 1)) - r_y;
    }
    s->current_picture.motion_val[0][xy][0] = s->mv[0][0][0];
    s->current_picture.motion_val[0][xy][1] = s->mv[0][0][1];
    s->current_picture.motion_val[1][xy][0] = s->mv[1][0][0];
    s->current_picture.motion_val[1][xy][1] = s->mv[1][0][1];
}

/** Get predicted DC value for I-frames only
 * prediction dir: left=0, top=1
 * @param s MpegEncContext
 * @param[in] n block index in the current MB
 * @param dc_val_ptr Pointer to DC predictor
 * @param dir_ptr Prediction direction for use in AC prediction
 */
static inline int vc1_i_pred_dc(MpegEncContext *s, int overlap, int pq, int n,
                              int16_t **dc_val_ptr, int *dir_ptr)
{
    int a, b, c, wrap, pred, scale;
    int16_t *dc_val;
    static const uint16_t dcpred[32] = {
    -1, 1024,  512,  341,  256,  205,  171,  146,  128,
         114,  102,   93,   85,   79,   73,   68,   64,
          60,   57,   54,   51,   49,   47,   45,   43,
          41,   39,   38,   37,   35,   34,   33
    };

    /* find prediction - wmv3_dc_scale always used here in fact */
    if (n < 4)     scale = s->y_dc_scale;
    else           scale = s->c_dc_scale;

    wrap = s->block_wrap[n];
    dc_val= s->dc_val[0] + s->block_index[n];

    /* B A
     * C X
     */
    c = dc_val[ - 1];
    b = dc_val[ - 1 - wrap];
    a = dc_val[ - wrap];

    if (pq < 9 || !overlap)
    {
        /* Set outer values */
        if (s->first_slice_line && (n!=2 && n!=3)) b=a=dcpred[scale];
        if (s->mb_x == 0 && (n!=1 && n!=3)) b=c=dcpred[scale];
    }
    else
    {
        /* Set outer values */
        if (s->first_slice_line && (n!=2 && n!=3)) b=a=0;
        if (s->mb_x == 0 && (n!=1 && n!=3)) b=c=0;
    }

    if (abs(a - b) <= abs(b - c)) {
        pred = c;
        *dir_ptr = 1;//left
    } else {
        pred = a;
        *dir_ptr = 0;//top
    }

    /* update predictor */
    *dc_val_ptr = &dc_val[0];
    return pred;
}


/** Get predicted DC value
 * prediction dir: left=0, top=1
 * @param s MpegEncContext
 * @param[in] n block index in the current MB
 * @param dc_val_ptr Pointer to DC predictor
 * @param dir_ptr Prediction direction for use in AC prediction
 */
static inline int vc1_pred_dc(MpegEncContext *s, int overlap, int pq, int n,
                              int a_avail, int c_avail,
                              int16_t **dc_val_ptr, int *dir_ptr)
{
    int a, b, c, wrap, pred, scale;
    int16_t *dc_val;
    int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
    int q1, q2 = 0;

    /* find prediction - wmv3_dc_scale always used here in fact */
    if (n < 4)     scale = s->y_dc_scale;
    else           scale = s->c_dc_scale;

    wrap = s->block_wrap[n];
    dc_val= s->dc_val[0] + s->block_index[n];

    /* B A
     * C X
     */
    c = dc_val[ - 1];
    b = dc_val[ - 1 - wrap];
    a = dc_val[ - wrap];
    /* scale predictors if needed */
    q1 = s->current_picture.qscale_table[mb_pos];
    if(c_avail && (n!= 1 && n!=3)) {
        q2 = s->current_picture.qscale_table[mb_pos - 1];
        if(q2 && q2 != q1)
            c = (c * s->y_dc_scale_table[q2] * ff_vc1_dqscale[s->y_dc_scale_table[q1] - 1] + 0x20000) >> 18;
    }
    if(a_avail && (n!= 2 && n!=3)) {
        q2 = s->current_picture.qscale_table[mb_pos - s->mb_stride];
        if(q2 && q2 != q1)
            a = (a * s->y_dc_scale_table[q2] * ff_vc1_dqscale[s->y_dc_scale_table[q1] - 1] + 0x20000) >> 18;
    }
    if(a_avail && c_avail && (n!=3)) {
        int off = mb_pos;
        if(n != 1) off--;
        if(n != 2) off -= s->mb_stride;
        q2 = s->current_picture.qscale_table[off];
        if(q2 && q2 != q1)
            b = (b * s->y_dc_scale_table[q2] * ff_vc1_dqscale[s->y_dc_scale_table[q1] - 1] + 0x20000) >> 18;
    }

    if(a_avail && c_avail) {
        if(abs(a - b) <= abs(b - c)) {
            pred = c;
            *dir_ptr = 1;//left
        } else {
            pred = a;
            *dir_ptr = 0;//top
        }
    } else if(a_avail) {
        pred = a;
        *dir_ptr = 0;//top
    } else if(c_avail) {
        pred = c;
        *dir_ptr = 1;//left
    } else {
        pred = 0;
        *dir_ptr = 1;//left
    }

    /* update predictor */
    *dc_val_ptr = &dc_val[0];
    return pred;
}


/**
 * @defgroup std_mb VC1 Macroblock-level functions in Simple/Main Profiles
 * @see 7.1.4, p91 and 8.1.1.7, p(1)04
 * @{
 */

static inline int vc1_coded_block_pred(MpegEncContext * s, int n, uint8_t **coded_block_ptr)
{
    int xy, wrap, pred, a, b, c;

    xy = s->block_index[n];
    wrap = s->b8_stride;

    /* B C
     * A X
     */
    a = s->coded_block[xy - 1       ];
    b = s->coded_block[xy - 1 - wrap];
    c = s->coded_block[xy     - wrap];

    if (b == c) {
        pred = a;
    } else {
        pred = c;
    }

    /* store value */
    *coded_block_ptr = &s->coded_block[xy];

    return pred;
}

/**
 * Decode one AC coefficient
 * @param v The VC1 context
 * @param last Last coefficient
 * @param skip How much zero coefficients to skip
 * @param value Decoded AC coefficient value
 * @see 8.1.3.4
 */
static void vc1_decode_ac_coeff(VC1Context *v, int *last, int *skip, int *value, int codingset)
{
    GetBitContext *gb = &v->s.gb;
    int index, escape, run = 0, level = 0, lst = 0;

    index = get_vlc2(gb, ff_vc1_ac_coeff_table[codingset].table, AC_VLC_BITS, 3);
    if (index != vc1_ac_sizes[codingset] - 1) {
        run = vc1_index_decode_table[codingset][index][0];
        level = vc1_index_decode_table[codingset][index][1];
        lst = index >= vc1_last_decode_table[codingset];
        if(get_bits(gb, 1))
            level = -level;
    } else {
        escape = decode210(gb);
        if (escape != 2) {
            index = get_vlc2(gb, ff_vc1_ac_coeff_table[codingset].table, AC_VLC_BITS, 3);
            run = vc1_index_decode_table[codingset][index][0];
            level = vc1_index_decode_table[codingset][index][1];
            lst = index >= vc1_last_decode_table[codingset];
            if(escape == 0) {
                if(lst)
                    level += vc1_last_delta_level_table[codingset][run];
                else
                    level += vc1_delta_level_table[codingset][run];
            } else {
                if(lst)
                    run += vc1_last_delta_run_table[codingset][level] + 1;
                else
                    run += vc1_delta_run_table[codingset][level] + 1;
            }
            if(get_bits(gb, 1))
                level = -level;
        } else {
            int sign;
            lst = get_bits(gb, 1);
            if(v->s.esc3_level_length == 0) {
                if(v->pq < 8 || v->dquantfrm) { // table 59
                    v->s.esc3_level_length = get_bits(gb, 3);
                    if(!v->s.esc3_level_length)
                        v->s.esc3_level_length = get_bits(gb, 2) + 8;
                } else { //table 60
                    v->s.esc3_level_length = get_prefix(gb, 1, 6) + 2;
                }
                v->s.esc3_run_length = 3 + get_bits(gb, 2);
            }
            run = get_bits(gb, v->s.esc3_run_length);
            sign = get_bits(gb, 1);
            level = get_bits(gb, v->s.esc3_level_length);
            if(sign)
                level = -level;
        }
    }

    *last = lst;
    *skip = run;
    *value = level;
}

/** Decode intra block in intra frames - should be faster than decode_intra_block
 * @param v VC1Context
 * @param block block to decode
 * @param coded are AC coeffs present or not
 * @param codingset set of VLC to decode data
 */
static int vc1_decode_i_block(VC1Context *v, DCTELEM block[64], int n, int coded, int codingset)
{
    GetBitContext *gb = &v->s.gb;
    MpegEncContext *s = &v->s;
    int dc_pred_dir = 0; /* Direction of the DC prediction used */
    int run_diff, i;
    int16_t *dc_val;
    int16_t *ac_val, *ac_val2;
    int dcdiff;

    /* Get DC differential */
    if (n < 4) {
        dcdiff = get_vlc2(&s->gb, ff_msmp4_dc_luma_vlc[s->dc_table_index].table, DC_VLC_BITS, 3);
    } else {
        dcdiff = get_vlc2(&s->gb, ff_msmp4_dc_chroma_vlc[s->dc_table_index].table, DC_VLC_BITS, 3);
    }
    if (dcdiff < 0){
        av_log(s->avctx, AV_LOG_ERROR, "Illegal DC VLC\n");
        return -1;
    }
    if (dcdiff)
    {
        if (dcdiff == 119 /* ESC index value */)
        {
            /* TODO: Optimize */
            if (v->pq == 1) dcdiff = get_bits(gb, 10);
            else if (v->pq == 2) dcdiff = get_bits(gb, 9);
            else dcdiff = get_bits(gb, 8);
        }
        else
        {
            if (v->pq == 1)
                dcdiff = (dcdiff<<2) + get_bits(gb, 2) - 3;
            else if (v->pq == 2)
                dcdiff = (dcdiff<<1) + get_bits(gb, 1) - 1;
        }
        if (get_bits(gb, 1))
            dcdiff = -dcdiff;
    }

    /* Prediction */
    dcdiff += vc1_i_pred_dc(&v->s, v->overlap, v->pq, n, &dc_val, &dc_pred_dir);
    *dc_val = dcdiff;

    /* Store the quantized DC coeff, used for prediction */
    if (n < 4) {
        block[0] = dcdiff * s->y_dc_scale;
    } else {
        block[0] = dcdiff * s->c_dc_scale;
    }
    /* Skip ? */
    run_diff = 0;
    i = 0;
    if (!coded) {
        goto not_coded;
    }

    //AC Decoding
    i = 1;

    {
        int last = 0, skip, value;
        const int8_t *zz_table;
        int scale;
        int k;

        scale = v->pq * 2 + v->halfpq;

        if(v->s.ac_pred) {
            if(!dc_pred_dir)
                zz_table = ff_vc1_horizontal_zz;
            else
                zz_table = ff_vc1_vertical_zz;
        } else
            zz_table = ff_vc1_normal_zz;

        ac_val = s->ac_val[0][0] + s->block_index[n] * 16;
        ac_val2 = ac_val;
        if(dc_pred_dir) //left
            ac_val -= 16;
        else //top
            ac_val -= 16 * s->block_wrap[n];

        while (!last) {
            vc1_decode_ac_coeff(v, &last, &skip, &value, codingset);
            i += skip;
            if(i > 63)
                break;
            block[zz_table[i++]] = value;
        }

        /* apply AC prediction if needed */
        if(s->ac_pred) {
            if(dc_pred_dir) { //left
                for(k = 1; k < 8; k++)
                    block[k << 3] += ac_val[k];
            } else { //top
                for(k = 1; k < 8; k++)
                    block[k] += ac_val[k + 8];
            }
        }
        /* save AC coeffs for further prediction */
        for(k = 1; k < 8; k++) {
            ac_val2[k] = block[k << 3];
            ac_val2[k + 8] = block[k];
        }

        /* scale AC coeffs */
        for(k = 1; k < 64; k++)
            if(block[k]) {
                block[k] *= scale;
                if(!v->pquantizer)
                    block[k] += (block[k] < 0) ? -v->pq : v->pq;
            }

        if(s->ac_pred) i = 63;
    }

not_coded:
    if(!coded) {
        int k, scale;
        ac_val = s->ac_val[0][0] + s->block_index[n] * 16;
        ac_val2 = ac_val;

        scale = v->pq * 2 + v->halfpq;
        memset(ac_val2, 0, 16 * 2);
        if(dc_pred_dir) {//left
            ac_val -= 16;
            if(s->ac_pred)
                memcpy(ac_val2, ac_val, 8 * 2);
        } else {//top
            ac_val -= 16 * s->block_wrap[n];
            if(s->ac_pred)
                memcpy(ac_val2 + 8, ac_val + 8, 8 * 2);
        }

        /* apply AC prediction if needed */
        if(s->ac_pred) {
            if(dc_pred_dir) { //left
                for(k = 1; k < 8; k++) {
                    block[k << 3] = ac_val[k] * scale;
                    if(!v->pquantizer && block[k << 3])
                        block[k << 3] += (block[k << 3] < 0) ? -v->pq : v->pq;
                }
            } else { //top
                for(k = 1; k < 8; k++) {
                    block[k] = ac_val[k + 8] * scale;
                    if(!v->pquantizer && block[k])
                        block[k] += (block[k] < 0) ? -v->pq : v->pq;
                }
            }
            i = 63;
        }
    }
    s->block_last_index[n] = i;

    return 0;
}

/** Decode intra block in intra frames - should be faster than decode_intra_block
 * @param v VC1Context
 * @param block block to decode
 * @param coded are AC coeffs present or not
 * @param codingset set of VLC to decode data
 */
static int vc1_decode_i_block_adv(VC1Context *v, DCTELEM block[64], int n, int coded, int codingset, int mquant)
{
    GetBitContext *gb = &v->s.gb;
    MpegEncContext *s = &v->s;
    int dc_pred_dir = 0; /* Direction of the DC prediction used */
    int run_diff, i;
    int16_t *dc_val;
    int16_t *ac_val, *ac_val2;
    int dcdiff;
    int a_avail = v->a_avail, c_avail = v->c_avail;
    int use_pred = s->ac_pred;
    int scale;
    int q1, q2 = 0;
    int mb_pos = s->mb_x + s->mb_y * s->mb_stride;

    /* Get DC differential */
    if (n < 4) {
        dcdiff = get_vlc2(&s->gb, ff_msmp4_dc_luma_vlc[s->dc_table_index].table, DC_VLC_BITS, 3);
    } else {
        dcdiff = get_vlc2(&s->gb, ff_msmp4_dc_chroma_vlc[s->dc_table_index].table, DC_VLC_BITS, 3);
    }
    if (dcdiff < 0){
        av_log(s->avctx, AV_LOG_ERROR, "Illegal DC VLC\n");
        return -1;
    }
    if (dcdiff)
    {
        if (dcdiff == 119 /* ESC index value */)
        {
            /* TODO: Optimize */
            if (mquant == 1) dcdiff = get_bits(gb, 10);
            else if (mquant == 2) dcdiff = get_bits(gb, 9);
            else dcdiff = get_bits(gb, 8);
        }
        else
        {
            if (mquant == 1)
                dcdiff = (dcdiff<<2) + get_bits(gb, 2) - 3;
            else if (mquant == 2)
                dcdiff = (dcdiff<<1) + get_bits(gb, 1) - 1;
        }
        if (get_bits(gb, 1))
            dcdiff = -dcdiff;
    }

    /* Prediction */
    dcdiff += vc1_pred_dc(&v->s, v->overlap, mquant, n, v->a_avail, v->c_avail, &dc_val, &dc_pred_dir);
    *dc_val = dcdiff;

    /* Store the quantized DC coeff, used for prediction */
    if (n < 4) {
        block[0] = dcdiff * s->y_dc_scale;
    } else {
        block[0] = dcdiff * s->c_dc_scale;
    }
    /* Skip ? */
    run_diff = 0;
    i = 0;

    //AC Decoding
    i = 1;

    /* check if AC is needed at all */
    if(!a_avail && !c_avail) use_pred = 0;
    ac_val = s->ac_val[0][0] + s->block_index[n] * 16;
    ac_val2 = ac_val;

    scale = mquant * 2 + v->halfpq;

    if(dc_pred_dir) //left
        ac_val -= 16;
    else //top
        ac_val -= 16 * s->block_wrap[n];

    q1 = s->current_picture.qscale_table[mb_pos];
    if(dc_pred_dir && c_avail && mb_pos) q2 = s->current_picture.qscale_table[mb_pos - 1];
    if(!dc_pred_dir && a_avail && mb_pos >= s->mb_stride) q2 = s->current_picture.qscale_table[mb_pos - s->mb_stride];
    if(dc_pred_dir && n==1) q2 = q1;
    if(!dc_pred_dir && n==2) q2 = q1;
    if(n==3) q2 = q1;

    if(coded) {
        int last = 0, skip, value;
        const int8_t *zz_table;
        int k;

        if(v->s.ac_pred) {
            if(!dc_pred_dir)
                zz_table = ff_vc1_horizontal_zz;
            else
                zz_table = ff_vc1_vertical_zz;
        } else
            zz_table = ff_vc1_normal_zz;

        while (!last) {
            vc1_decode_ac_coeff(v, &last, &skip, &value, codingset);
            i += skip;
            if(i > 63)
                break;
            block[zz_table[i++]] = value;
        }

        /* apply AC prediction if needed */
        if(use_pred) {
            /* scale predictors if needed*/
            if(q2 && q1!=q2) {
                q1 = q1 * 2 + ((q1 == v->pq) ? v->halfpq : 0) - 1;
                q2 = q2 * 2 + ((q2 == v->pq) ? v->halfpq : 0) - 1;

                if(dc_pred_dir) { //left
                    for(k = 1; k < 8; k++)
                        block[k << 3] += (ac_val[k] * q2 * ff_vc1_dqscale[q1 - 1] + 0x20000) >> 18;
                } else { //top
                    for(k = 1; k < 8; k++)
                        block[k] += (ac_val[k + 8] * q2 * ff_vc1_dqscale[q1 - 1] + 0x20000) >> 18;
                }
            } else {
                if(dc_pred_dir) { //left
                    for(k = 1; k < 8; k++)
                        block[k << 3] += ac_val[k];
                } else { //top
                    for(k = 1; k < 8; k++)
                        block[k] += ac_val[k + 8];
                }
            }
        }
        /* save AC coeffs for further prediction */
        for(k = 1; k < 8; k++) {
            ac_val2[k] = block[k << 3];
            ac_val2[k + 8] = block[k];
        }

        /* scale AC coeffs */
        for(k = 1; k < 64; k++)
            if(block[k]) {
                block[k] *= scale;
                if(!v->pquantizer)
                    block[k] += (block[k] < 0) ? -mquant : mquant;
            }

        if(use_pred) i = 63;
    } else { // no AC coeffs
        int k;

        memset(ac_val2, 0, 16 * 2);
        if(dc_pred_dir) {//left
            if(use_pred) {
                memcpy(ac_val2, ac_val, 8 * 2);
                if(q2 && q1!=q2) {
                    q1 = q1 * 2 + ((q1 == v->pq) ? v->halfpq : 0) - 1;
                    q2 = q2 * 2 + ((q2 == v->pq) ? v->halfpq : 0) - 1;
                    for(k = 1; k < 8; k++)
                        ac_val2[k] = (ac_val2[k] * q2 * ff_vc1_dqscale[q1 - 1] + 0x20000) >> 18;
                }
            }
        } else {//top
            if(use_pred) {
                memcpy(ac_val2 + 8, ac_val + 8, 8 * 2);
                if(q2 && q1!=q2) {
                    q1 = q1 * 2 + ((q1 == v->pq) ? v->halfpq : 0) - 1;
                    q2 = q2 * 2 + ((q2 == v->pq) ? v->halfpq : 0) - 1;
                    for(k = 1; k < 8; k++)
                        ac_val2[k + 8] = (ac_val2[k + 8] * q2 * ff_vc1_dqscale[q1 - 1] + 0x20000) >> 18;
                }
            }
        }

        /* apply AC prediction if needed */
        if(use_pred) {
            if(dc_pred_dir) { //left
                for(k = 1; k < 8; k++) {
                    block[k << 3] = ac_val2[k] * scale;
                    if(!v->pquantizer && block[k << 3])
                        block[k << 3] += (block[k << 3] < 0) ? -mquant : mquant;
                }
            } else { //top
                for(k = 1; k < 8; k++) {
                    block[k] = ac_val2[k + 8] * scale;
                    if(!v->pquantizer && block[k])
                        block[k] += (block[k] < 0) ? -mquant : mquant;
                }
            }
            i = 63;
        }
    }
    s->block_last_index[n] = i;

    return 0;
}

/** Decode intra block in inter frames - more generic version than vc1_decode_i_block
 * @param v VC1Context
 * @param block block to decode
 * @param coded are AC coeffs present or not
 * @param mquant block quantizer
 * @param codingset set of VLC to decode data
 */
static int vc1_decode_intra_block(VC1Context *v, DCTELEM block[64], int n, int coded, int mquant, int codingset)
{
    GetBitContext *gb = &v->s.gb;
    MpegEncContext *s = &v->s;
    int dc_pred_dir = 0; /* Direction of the DC prediction used */
    int run_diff, i;
    int16_t *dc_val;
    int16_t *ac_val, *ac_val2;
    int dcdiff;
    int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
    int a_avail = v->a_avail, c_avail = v->c_avail;
    int use_pred = s->ac_pred;
    int scale;
    int q1, q2 = 0;

    /* XXX: Guard against dumb values of mquant */
    mquant = (mquant < 1) ? 0 : ( (mquant>31) ? 31 : 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];

    /* Get DC differential */
    if (n < 4) {
        dcdiff = get_vlc2(&s->gb, ff_msmp4_dc_luma_vlc[s->dc_table_index].table, DC_VLC_BITS, 3);
    } else {
        dcdiff = get_vlc2(&s->gb, ff_msmp4_dc_chroma_vlc[s->dc_table_index].table, DC_VLC_BITS, 3);
    }
    if (dcdiff < 0){
        av_log(s->avctx, AV_LOG_ERROR, "Illegal DC VLC\n");
        return -1;
    }
    if (dcdiff)
    {
        if (dcdiff == 119 /* ESC index value */)
        {
            /* TODO: Optimize */
            if (mquant == 1) dcdiff = get_bits(gb, 10);
            else if (mquant == 2) dcdiff = get_bits(gb, 9);
            else dcdiff = get_bits(gb, 8);
        }
        else
        {
            if (mquant == 1)
                dcdiff = (dcdiff<<2) + get_bits(gb, 2) - 3;
            else if (mquant == 2)
                dcdiff = (dcdiff<<1) + get_bits(gb, 1) - 1;
        }
        if (get_bits(gb, 1))
            dcdiff = -dcdiff;
    }

    /* Prediction */
    dcdiff += vc1_pred_dc(&v->s, v->overlap, mquant, n, a_avail, c_avail, &dc_val, &dc_pred_dir);
    *dc_val = dcdiff;

    /* Store the quantized DC coeff, used for prediction */

    if (n < 4) {
        block[0] = dcdiff * s->y_dc_scale;
    } else {
        block[0] = dcdiff * s->c_dc_scale;
    }
    /* Skip ? */
    run_diff = 0;
    i = 0;

    //AC Decoding
    i = 1;

    /* check if AC is needed at all and adjust direction if needed */
    if(!a_avail) dc_pred_dir = 1;
    if(!c_avail) dc_pred_dir = 0;
    if(!a_avail && !c_avail) use_pred = 0;
    ac_val = s->ac_val[0][0] + s->block_index[n] * 16;
    ac_val2 = ac_val;

    scale = mquant * 2 + v->halfpq;

    if(dc_pred_dir) //left
        ac_val -= 16;
    else //top
        ac_val -= 16 * s->block_wrap[n];

    q1 = s->current_picture.qscale_table[mb_pos];
    if(dc_pred_dir && c_avail && mb_pos) q2 = s->current_picture.qscale_table[mb_pos - 1];
    if(!dc_pred_dir && a_avail && mb_pos >= s->mb_stride) q2 = s->current_picture.qscale_table[mb_pos - s->mb_stride];
    if(dc_pred_dir && n==1) q2 = q1;
    if(!dc_pred_dir && n==2) q2 = q1;
    if(n==3) q2 = q1;

    if(coded) {
        int last = 0, skip, value;
        const int8_t *zz_table;
        int k;

        zz_table = ff_vc1_simple_progressive_8x8_zz;

        while (!last) {
            vc1_decode_ac_coeff(v, &last, &skip, &value, codingset);
            i += skip;
            if(i > 63)
                break;
            block[zz_table[i++]] = value;
        }

        /* apply AC prediction if needed */
        if(use_pred) {
            /* scale predictors if needed*/
            if(q2 && q1!=q2) {
                q1 = q1 * 2 + ((q1 == v->pq) ? v->halfpq : 0) - 1;
                q2 = q2 * 2 + ((q2 == v->pq) ? v->halfpq : 0) - 1;

                if(dc_pred_dir) { //left
                    for(k = 1; k < 8; k++)
                        block[k << 3] += (ac_val[k] * q2 * ff_vc1_dqscale[q1 - 1] + 0x20000) >> 18;
                } else { //top
                    for(k = 1; k < 8; k++)
                        block[k] += (ac_val[k + 8] * q2 * ff_vc1_dqscale[q1 - 1] + 0x20000) >> 18;
                }
            } else {
                if(dc_pred_dir) { //left
                    for(k = 1; k < 8; k++)
                        block[k << 3] += ac_val[k];
                } else { //top
                    for(k = 1; k < 8; k++)
                        block[k] += ac_val[k + 8];
                }
            }
        }
        /* save AC coeffs for further prediction */
        for(k = 1; k < 8; k++) {
            ac_val2[k] = block[k << 3];
            ac_val2[k + 8] = block[k];
        }

        /* scale AC coeffs */
        for(k = 1; k < 64; k++)
            if(block[k]) {
                block[k] *= scale;
                if(!v->pquantizer)
                    block[k] += (block[k] < 0) ? -mquant : mquant;
            }

        if(use_pred) i = 63;
    } else { // no AC coeffs
        int k;

        memset(ac_val2, 0, 16 * 2);
        if(dc_pred_dir) {//left
            if(use_pred) {
                memcpy(ac_val2, ac_val, 8 * 2);
                if(q2 && q1!=q2) {
                    q1 = q1 * 2 + ((q1 == v->pq) ? v->halfpq : 0) - 1;
                    q2 = q2 * 2 + ((q2 == v->pq) ? v->halfpq : 0) - 1;
                    for(k = 1; k < 8; k++)
                        ac_val2[k] = (ac_val2[k] * q2 * ff_vc1_dqscale[q1 - 1] + 0x20000) >> 18;
                }
            }
        } else {//top
            if(use_pred) {
                memcpy(ac_val2 + 8, ac_val + 8, 8 * 2);
                if(q2 && q1!=q2) {
                    q1 = q1 * 2 + ((q1 == v->pq) ? v->halfpq : 0) - 1;
                    q2 = q2 * 2 + ((q2 == v->pq) ? v->halfpq : 0) - 1;
                    for(k = 1; k < 8; k++)
                        ac_val2[k + 8] = (ac_val2[k + 8] * q2 * ff_vc1_dqscale[q1 - 1] + 0x20000) >> 18;
                }
            }
        }

        /* apply AC prediction if needed */
        if(use_pred) {
            if(dc_pred_dir) { //left
                for(k = 1; k < 8; k++) {
                    block[k << 3] = ac_val2[k] * scale;
                    if(!v->pquantizer && block[k << 3])
                        block[k << 3] += (block[k << 3] < 0) ? -mquant : mquant;
                }
            } else { //top
                for(k = 1; k < 8; k++) {
                    block[k] = ac_val2[k + 8] * scale;
                    if(!v->pquantizer && block[k])
                        block[k] += (block[k] < 0) ? -mquant : mquant;
                }
            }
            i = 63;
        }
    }
    s->block_last_index[n] = i;

    return 0;
}

/** Decode P block
 */
static int vc1_decode_p_block(VC1Context *v, DCTELEM block[64], int n, int mquant, int ttmb, int first_block)
{
    MpegEncContext *s = &v->s;
    GetBitContext *gb = &s->gb;
    int i, j;
    int subblkpat = 0;
    int scale, off, idx, last, skip, value;
    int ttblk = ttmb & 7;

    if(ttmb == -1) {
        ttblk = ff_vc1_ttblk_to_tt[v->tt_index][get_vlc2(gb, ff_vc1_ttblk_vlc[v->tt_index].table, VC1_TTBLK_VLC_BITS, 1)];
    }
    if(ttblk == TT_4X4) {
        subblkpat = ~(get_vlc2(gb, ff_vc1_subblkpat_vlc[v->tt_index].table, VC1_SUBBLKPAT_VLC_BITS, 1) + 1);
    }
    if((ttblk != TT_8X8 && ttblk != TT_4X4) && (v->ttmbf || (ttmb != -1 && (ttmb & 8) && !first_block))) {
        subblkpat = decode012(gb);
        if(subblkpat) subblkpat ^= 3; //swap decoded pattern bits
        if(ttblk == TT_8X4_TOP || ttblk == TT_8X4_BOTTOM) ttblk = TT_8X4;
        if(ttblk == TT_4X8_RIGHT || ttblk == TT_4X8_LEFT) ttblk = TT_4X8;
    }
    scale = 2 * mquant + ((v->pq == mquant) ? v->halfpq : 0);

    // convert transforms like 8X4_TOP to generic TT and SUBBLKPAT
    if(ttblk == TT_8X4_TOP || ttblk == TT_8X4_BOTTOM) {
        subblkpat = 2 - (ttblk == TT_8X4_TOP);
        ttblk = TT_8X4;
    }
    if(ttblk == TT_4X8_RIGHT || ttblk == TT_4X8_LEFT) {
        subblkpat = 2 - (ttblk == TT_4X8_LEFT);
        ttblk = TT_4X8;
    }
    switch(ttblk) {
    case TT_8X8:
        i = 0;
        last = 0;
        while (!last) {
            vc1_decode_ac_coeff(v, &last, &skip, &value, v->codingset2);
            i += skip;
            if(i > 63)
                break;
            idx = ff_vc1_simple_progressive_8x8_zz[i++];
            block[idx] = value * scale;
            if(!v->pquantizer)
                block[idx] += (block[idx] < 0) ? -mquant : mquant;
        }
        s->dsp.vc1_inv_trans_8x8(block);
        break;
    case TT_4X4:
        for(j = 0; j < 4; j++) {
            last = subblkpat & (1 << (3 - j));
            i = 0;
            off = (j & 1) * 4 + (j & 2) * 16;
            while (!last) {
                vc1_decode_ac_coeff(v, &last, &skip, &value, v->codingset2);
                i += skip;
                if(i > 15)
                    break;
                idx = ff_vc1_simple_progressive_4x4_zz[i++];
                block[idx + off] = value * scale;
                if(!v->pquantizer)
                    block[idx + off] += (block[idx + off] < 0) ? -mquant : mquant;
            }
            if(!(subblkpat & (1 << (3 - j))))
                s->dsp.vc1_inv_trans_4x4(block, j);