/lib/ffmpeg/libavcodec/vc1dec.c

Large files files are truncated, but you can click here to view the full file

/*
 * VC-1 and WMV3 decoder
 * Copyright (c) 2011 Mashiat Sarker Shakkhar
 * Copyright (c) 2006-2007 Konstantin Shishkov
 * Partly based on vc9.c (c) 2005 Anonymous, Alex Beregszaszi, Michael Niedermayer
 *
 * This file is part of FFmpeg.
 *
 * FFmpeg is free software; you can redistribute it and/or
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
 * version 2.1 of the License, or (at your option) any later version.
 *
 * FFmpeg is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
 *
 * You should have received a copy of the GNU Lesser General Public
 * License along with FFmpeg; if not, write to the Free Software
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 */

/**
 * @file
 * VC-1 and WMV3 decoder
 */

#include "internal.h"
#include "avcodec.h"
#include "mpegvideo.h"
#include "h263.h"
#include "h264chroma.h"
#include "vc1.h"
#include "vc1data.h"
#include "vc1acdata.h"
#include "msmpeg4data.h"
#include "unary.h"
#include "mathops.h"
#include "vdpau_internal.h"
#include "libavutil/avassert.h"

#undef NDEBUG
#include <assert.h>

#define MB_INTRA_VLC_BITS 9
#define DC_VLC_BITS 9


// offset tables for interlaced picture MVDATA decoding
static const int offset_table1[9] = {  0,  1,  2,  4,  8, 16, 32,  64, 128 };
static const int offset_table2[9] = {  0,  1,  3,  7, 15, 31, 63, 127, 255 };

/***********************************************************************/
/**
 * @name VC-1 Bitplane decoding
 * @see 8.7, p56
 * @{
 */

/**
 * Imode types
 * @{
 */
enum Imode {
    IMODE_RAW,
    IMODE_NORM2,
    IMODE_DIFF2,
    IMODE_NORM6,
    IMODE_DIFF6,
    IMODE_ROWSKIP,
    IMODE_COLSKIP
};
/** @} */ //imode defines

static void init_block_index(VC1Context *v)
{
    MpegEncContext *s = &v->s;
    ff_init_block_index(s);
    if (v->field_mode && v->second_field) {
        s->dest[0] += s->current_picture_ptr->f.linesize[0];
        s->dest[1] += s->current_picture_ptr->f.linesize[1];
        s->dest[2] += s->current_picture_ptr->f.linesize[2];
    }
}


/** @} */ //Bitplane group

static void vc1_put_signed_blocks_clamped(VC1Context *v)
{
    MpegEncContext *s = &v->s;
    int topleft_mb_pos, top_mb_pos;
    int stride_y, fieldtx = 0;
    int v_dist;

    /* The put pixels loop is always one MB row behind the decoding loop,
     * because we can only put pixels when overlap filtering is done, and
     * for filtering of the bottom edge of a MB, we need the next MB row
     * present as well.
     * Within the row, the put pixels loop is also one MB col behind the
     * decoding loop. The reason for this is again, because for filtering
     * of the right MB edge, we need the next MB present. */
    if (!s->first_slice_line) {
        if (s->mb_x) {
            topleft_mb_pos = (s->mb_y - 1) * s->mb_stride + s->mb_x - 1;
            if (v->fcm == ILACE_FRAME)
                fieldtx = v->fieldtx_plane[topleft_mb_pos];
            stride_y       = s->linesize << fieldtx;
            v_dist         = (16 - fieldtx) >> (fieldtx == 0);
            s->dsp.put_signed_pixels_clamped(v->block[v->topleft_blk_idx][0],
                                             s->dest[0] - 16 * s->linesize - 16,
                                             stride_y);
            s->dsp.put_signed_pixels_clamped(v->block[v->topleft_blk_idx][1],
                                             s->dest[0] - 16 * s->linesize - 8,
                                             stride_y);
            s->dsp.put_signed_pixels_clamped(v->block[v->topleft_blk_idx][2],
                                             s->dest[0] - v_dist * s->linesize - 16,
                                             stride_y);
            s->dsp.put_signed_pixels_clamped(v->block[v->topleft_blk_idx][3],
                                             s->dest[0] - v_dist * s->linesize - 8,
                                             stride_y);
            s->dsp.put_signed_pixels_clamped(v->block[v->topleft_blk_idx][4],
                                             s->dest[1] - 8 * s->uvlinesize - 8,
                                             s->uvlinesize);
            s->dsp.put_signed_pixels_clamped(v->block[v->topleft_blk_idx][5],
                                             s->dest[2] - 8 * s->uvlinesize - 8,
                                             s->uvlinesize);
        }
        if (s->mb_x == s->mb_width - 1) {
            top_mb_pos = (s->mb_y - 1) * s->mb_stride + s->mb_x;
            if (v->fcm == ILACE_FRAME)
                fieldtx = v->fieldtx_plane[top_mb_pos];
            stride_y   = s->linesize << fieldtx;
            v_dist     = fieldtx ? 15 : 8;
            s->dsp.put_signed_pixels_clamped(v->block[v->top_blk_idx][0],
                                             s->dest[0] - 16 * s->linesize,
                                             stride_y);
            s->dsp.put_signed_pixels_clamped(v->block[v->top_blk_idx][1],
                                             s->dest[0] - 16 * s->linesize + 8,
                                             stride_y);
            s->dsp.put_signed_pixels_clamped(v->block[v->top_blk_idx][2],
                                             s->dest[0] - v_dist * s->linesize,
                                             stride_y);
            s->dsp.put_signed_pixels_clamped(v->block[v->top_blk_idx][3],
                                             s->dest[0] - v_dist * s->linesize + 8,
                                             stride_y);
            s->dsp.put_signed_pixels_clamped(v->block[v->top_blk_idx][4],
                                             s->dest[1] - 8 * s->uvlinesize,
                                             s->uvlinesize);
            s->dsp.put_signed_pixels_clamped(v->block[v->top_blk_idx][5],
                                             s->dest[2] - 8 * s->uvlinesize,
                                             s->uvlinesize);
        }
    }

#define inc_blk_idx(idx) do { \
        idx++; \
        if (idx >= v->n_allocated_blks) \
            idx = 0; \
    } while (0)

    inc_blk_idx(v->topleft_blk_idx);
    inc_blk_idx(v->top_blk_idx);
    inc_blk_idx(v->left_blk_idx);
    inc_blk_idx(v->cur_blk_idx);
}

static void vc1_loop_filter_iblk(VC1Context *v, int pq)
{
    MpegEncContext *s = &v->s;
    int j;
    if (!s->first_slice_line) {
        v->vc1dsp.vc1_v_loop_filter16(s->dest[0], s->linesize, pq);
        if (s->mb_x)
            v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 16 * s->linesize, s->linesize, pq);
        v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 16 * s->linesize + 8, s->linesize, pq);
        for (j = 0; j < 2; j++) {
            v->vc1dsp.vc1_v_loop_filter8(s->dest[j + 1], s->uvlinesize, pq);
            if (s->mb_x)
                v->vc1dsp.vc1_h_loop_filter8(s->dest[j + 1] - 8 * s->uvlinesize, s->uvlinesize, pq);
        }
    }
    v->vc1dsp.vc1_v_loop_filter16(s->dest[0] + 8 * s->linesize, s->linesize, pq);

    if (s->mb_y == s->end_mb_y - 1) {
        if (s->mb_x) {
            v->vc1dsp.vc1_h_loop_filter16(s->dest[0], s->linesize, pq);
            v->vc1dsp.vc1_h_loop_filter8(s->dest[1], s->uvlinesize, pq);
            v->vc1dsp.vc1_h_loop_filter8(s->dest[2], s->uvlinesize, pq);
        }
        v->vc1dsp.vc1_h_loop_filter16(s->dest[0] + 8, s->linesize, pq);
    }
}

static void vc1_loop_filter_iblk_delayed(VC1Context *v, int pq)
{
    MpegEncContext *s = &v->s;
    int j;

    /* The loopfilter runs 1 row and 1 column behind the overlap filter, which
     * means it runs two rows/cols behind the decoding loop. */
    if (!s->first_slice_line) {
        if (s->mb_x) {
            if (s->mb_y >= s->start_mb_y + 2) {
                v->vc1dsp.vc1_v_loop_filter16(s->dest[0] - 16 * s->linesize - 16, s->linesize, pq);

                if (s->mb_x >= 2)
                    v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 32 * s->linesize - 16, s->linesize, pq);
                v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 32 * s->linesize - 8, s->linesize, pq);
                for (j = 0; j < 2; j++) {
                    v->vc1dsp.vc1_v_loop_filter8(s->dest[j + 1] - 8 * s->uvlinesize - 8, s->uvlinesize, pq);
                    if (s->mb_x >= 2) {
                        v->vc1dsp.vc1_h_loop_filter8(s->dest[j + 1] - 16 * s->uvlinesize - 8, s->uvlinesize, pq);
                    }
                }
            }
            v->vc1dsp.vc1_v_loop_filter16(s->dest[0] - 8 * s->linesize - 16, s->linesize, pq);
        }

        if (s->mb_x == s->mb_width - 1) {
            if (s->mb_y >= s->start_mb_y + 2) {
                v->vc1dsp.vc1_v_loop_filter16(s->dest[0] - 16 * s->linesize, s->linesize, pq);

                if (s->mb_x)
                    v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 32 * s->linesize, s->linesize, pq);
                v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 32 * s->linesize + 8, s->linesize, pq);
                for (j = 0; j < 2; j++) {
                    v->vc1dsp.vc1_v_loop_filter8(s->dest[j + 1] - 8 * s->uvlinesize, s->uvlinesize, pq);
                    if (s->mb_x >= 2) {
                        v->vc1dsp.vc1_h_loop_filter8(s->dest[j + 1] - 16 * s->uvlinesize, s->uvlinesize, pq);
                    }
                }
            }
            v->vc1dsp.vc1_v_loop_filter16(s->dest[0] - 8 * s->linesize, s->linesize, pq);
        }

        if (s->mb_y == s->end_mb_y) {
            if (s->mb_x) {
                if (s->mb_x >= 2)
                    v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 16 * s->linesize - 16, s->linesize, pq);
                v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 16 * s->linesize - 8, s->linesize, pq);
                if (s->mb_x >= 2) {
                    for (j = 0; j < 2; j++) {
                        v->vc1dsp.vc1_h_loop_filter8(s->dest[j + 1] - 8 * s->uvlinesize - 8, s->uvlinesize, pq);
                    }
                }
            }

            if (s->mb_x == s->mb_width - 1) {
                if (s->mb_x)
                    v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 16 * s->linesize, s->linesize, pq);
                v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 16 * s->linesize + 8, s->linesize, pq);
                if (s->mb_x) {
                    for (j = 0; j < 2; j++) {
                        v->vc1dsp.vc1_h_loop_filter8(s->dest[j + 1] - 8 * s->uvlinesize, s->uvlinesize, pq);
                    }
                }
            }
        }
    }
}

static void vc1_smooth_overlap_filter_iblk(VC1Context *v)
{
    MpegEncContext *s = &v->s;
    int mb_pos;

    if (v->condover == CONDOVER_NONE)
        return;

    mb_pos = s->mb_x + s->mb_y * s->mb_stride;

    /* Within a MB, the horizontal overlap always runs before the vertical.
     * To accomplish that, we run the H on left and internal borders of the
     * currently decoded MB. Then, we wait for the next overlap iteration
     * to do H overlap on the right edge of this MB, before moving over and
     * running the V overlap. Therefore, the V overlap makes us trail by one
     * MB col and the H overlap filter makes us trail by one MB row. This
     * is reflected in the time at which we run the put_pixels loop. */
    if (v->condover == CONDOVER_ALL || v->pq >= 9 || v->over_flags_plane[mb_pos]) {
        if (s->mb_x && (v->condover == CONDOVER_ALL || v->pq >= 9 ||
                        v->over_flags_plane[mb_pos - 1])) {
            v->vc1dsp.vc1_h_s_overlap(v->block[v->left_blk_idx][1],
                                      v->block[v->cur_blk_idx][0]);
            v->vc1dsp.vc1_h_s_overlap(v->block[v->left_blk_idx][3],
                                      v->block[v->cur_blk_idx][2]);
            if (!(s->flags & CODEC_FLAG_GRAY)) {
                v->vc1dsp.vc1_h_s_overlap(v->block[v->left_blk_idx][4],
                                          v->block[v->cur_blk_idx][4]);
                v->vc1dsp.vc1_h_s_overlap(v->block[v->left_blk_idx][5],
                                          v->block[v->cur_blk_idx][5]);
            }
        }
        v->vc1dsp.vc1_h_s_overlap(v->block[v->cur_blk_idx][0],
                                  v->block[v->cur_blk_idx][1]);
        v->vc1dsp.vc1_h_s_overlap(v->block[v->cur_blk_idx][2],
                                  v->block[v->cur_blk_idx][3]);

        if (s->mb_x == s->mb_width - 1) {
            if (!s->first_slice_line && (v->condover == CONDOVER_ALL || v->pq >= 9 ||
                                         v->over_flags_plane[mb_pos - s->mb_stride])) {
                v->vc1dsp.vc1_v_s_overlap(v->block[v->top_blk_idx][2],
                                          v->block[v->cur_blk_idx][0]);
                v->vc1dsp.vc1_v_s_overlap(v->block[v->top_blk_idx][3],
                                          v->block[v->cur_blk_idx][1]);
                if (!(s->flags & CODEC_FLAG_GRAY)) {
                    v->vc1dsp.vc1_v_s_overlap(v->block[v->top_blk_idx][4],
                                              v->block[v->cur_blk_idx][4]);
                    v->vc1dsp.vc1_v_s_overlap(v->block[v->top_blk_idx][5],
                                              v->block[v->cur_blk_idx][5]);
                }
            }
            v->vc1dsp.vc1_v_s_overlap(v->block[v->cur_blk_idx][0],
                                      v->block[v->cur_blk_idx][2]);
            v->vc1dsp.vc1_v_s_overlap(v->block[v->cur_blk_idx][1],
                                      v->block[v->cur_blk_idx][3]);
        }
    }
    if (s->mb_x && (v->condover == CONDOVER_ALL || v->over_flags_plane[mb_pos - 1])) {
        if (!s->first_slice_line && (v->condover == CONDOVER_ALL || v->pq >= 9 ||
                                     v->over_flags_plane[mb_pos - s->mb_stride - 1])) {
            v->vc1dsp.vc1_v_s_overlap(v->block[v->topleft_blk_idx][2],
                                      v->block[v->left_blk_idx][0]);
            v->vc1dsp.vc1_v_s_overlap(v->block[v->topleft_blk_idx][3],
                                      v->block[v->left_blk_idx][1]);
            if (!(s->flags & CODEC_FLAG_GRAY)) {
                v->vc1dsp.vc1_v_s_overlap(v->block[v->topleft_blk_idx][4],
                                          v->block[v->left_blk_idx][4]);
                v->vc1dsp.vc1_v_s_overlap(v->block[v->topleft_blk_idx][5],
                                          v->block[v->left_blk_idx][5]);
            }
        }
        v->vc1dsp.vc1_v_s_overlap(v->block[v->left_blk_idx][0],
                                  v->block[v->left_blk_idx][2]);
        v->vc1dsp.vc1_v_s_overlap(v->block[v->left_blk_idx][1],
                                  v->block[v->left_blk_idx][3]);
    }
}

/** Do motion compensation over 1 macroblock
 * Mostly adapted hpel_motion and qpel_motion from mpegvideo.c
 */
static void vc1_mc_1mv(VC1Context *v, int dir)
{
    MpegEncContext *s = &v->s;
    DSPContext *dsp   = &v->s.dsp;
    H264ChromaContext *h264chroma = &v->h264chroma;
    uint8_t *srcY, *srcU, *srcV;
    int dxy, mx, my, uvmx, uvmy, src_x, src_y, uvsrc_x, uvsrc_y;
    int off, off_uv;
    int v_edge_pos = s->v_edge_pos >> v->field_mode;

    if ((!v->field_mode ||
         (v->ref_field_type[dir] == 1 && v->cur_field_type == 1)) &&
        !v->s.last_picture.f.data[0])
        return;

    mx = s->mv[dir][0][0];
    my = s->mv[dir][0][1];

    // store motion vectors for further use in B frames
    if (s->pict_type == AV_PICTURE_TYPE_P) {
        s->current_picture.f.motion_val[1][s->block_index[0] + v->blocks_off][0] = mx;
        s->current_picture.f.motion_val[1][s->block_index[0] + v->blocks_off][1] = my;
    }

    uvmx = (mx + ((mx & 3) == 3)) >> 1;
    uvmy = (my + ((my & 3) == 3)) >> 1;
    v->luma_mv[s->mb_x][0] = uvmx;
    v->luma_mv[s->mb_x][1] = uvmy;

    if (v->field_mode &&
        v->cur_field_type != v->ref_field_type[dir]) {
        my   = my   - 2 + 4 * v->cur_field_type;
        uvmy = uvmy - 2 + 4 * v->cur_field_type;
    }

    // fastuvmc shall be ignored for interlaced frame picture
    if (v->fastuvmc && (v->fcm != ILACE_FRAME)) {
        uvmx = uvmx + ((uvmx < 0) ? (uvmx & 1) : -(uvmx & 1));
        uvmy = uvmy + ((uvmy < 0) ? (uvmy & 1) : -(uvmy & 1));
    }
    if (v->field_mode) { // interlaced field picture
        if (!dir) {
            if ((v->cur_field_type != v->ref_field_type[dir]) && v->second_field) {
                srcY = s->current_picture.f.data[0];
                srcU = s->current_picture.f.data[1];
                srcV = s->current_picture.f.data[2];
            } else {
                srcY = s->last_picture.f.data[0];
                srcU = s->last_picture.f.data[1];
                srcV = s->last_picture.f.data[2];
            }
        } else {
            srcY = s->next_picture.f.data[0];
            srcU = s->next_picture.f.data[1];
            srcV = s->next_picture.f.data[2];
        }
    } else {
        if (!dir) {
            srcY = s->last_picture.f.data[0];
            srcU = s->last_picture.f.data[1];
            srcV = s->last_picture.f.data[2];
        } else {
            srcY = s->next_picture.f.data[0];
            srcU = s->next_picture.f.data[1];
            srcV = s->next_picture.f.data[2];
        }
    }

    if(!srcY)
        return;

    src_x   = s->mb_x * 16 + (mx   >> 2);
    src_y   = s->mb_y * 16 + (my   >> 2);
    uvsrc_x = s->mb_x *  8 + (uvmx >> 2);
    uvsrc_y = s->mb_y *  8 + (uvmy >> 2);

    if (v->profile != PROFILE_ADVANCED) {
        src_x   = av_clip(  src_x, -16, s->mb_width  * 16);
        src_y   = av_clip(  src_y, -16, s->mb_height * 16);
        uvsrc_x = av_clip(uvsrc_x,  -8, s->mb_width  *  8);
        uvsrc_y = av_clip(uvsrc_y,  -8, s->mb_height *  8);
    } else {
        src_x   = av_clip(  src_x, -17, s->avctx->coded_width);
        src_y   = av_clip(  src_y, -18, s->avctx->coded_height + 1);
        uvsrc_x = av_clip(uvsrc_x,  -8, s->avctx->coded_width  >> 1);
        uvsrc_y = av_clip(uvsrc_y,  -8, s->avctx->coded_height >> 1);
    }

    srcY += src_y   * s->linesize   + src_x;
    srcU += uvsrc_y * s->uvlinesize + uvsrc_x;
    srcV += uvsrc_y * s->uvlinesize + uvsrc_x;

    if (v->field_mode && v->ref_field_type[dir]) {
        srcY += s->current_picture_ptr->f.linesize[0];
        srcU += s->current_picture_ptr->f.linesize[1];
        srcV += s->current_picture_ptr->f.linesize[2];
    }

    /* for grayscale we should not try to read from unknown area */
    if (s->flags & CODEC_FLAG_GRAY) {
        srcU = s->edge_emu_buffer + 18 * s->linesize;
        srcV = s->edge_emu_buffer + 18 * s->linesize;
    }

    if (v->rangeredfrm || (v->mv_mode == MV_PMODE_INTENSITY_COMP)
        || s->h_edge_pos < 22 || v_edge_pos < 22
        || (unsigned)(src_x - s->mspel) > s->h_edge_pos - (mx&3) - 16 - s->mspel * 3
        || (unsigned)(src_y - 1)        > v_edge_pos    - (my&3) - 16 - 3) {
        uint8_t *uvbuf = s->edge_emu_buffer + 19 * s->linesize;

        srcY -= s->mspel * (1 + s->linesize);
        s->vdsp.emulated_edge_mc(s->edge_emu_buffer, srcY, s->linesize,
                                 17 + s->mspel * 2, 17 + s->mspel * 2,
                                 src_x - s->mspel, src_y - s->mspel,
                                 s->h_edge_pos, v_edge_pos);
        srcY = s->edge_emu_buffer;
        s->vdsp.emulated_edge_mc(uvbuf     , srcU, s->uvlinesize, 8 + 1, 8 + 1,
                                 uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, v_edge_pos >> 1);
        s->vdsp.emulated_edge_mc(uvbuf + 16, srcV, s->uvlinesize, 8 + 1, 8 + 1,
                                 uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, v_edge_pos >> 1);
        srcU = uvbuf;
        srcV = uvbuf + 16;
        /* if we deal with range reduction we need to scale source blocks */
        if (v->rangeredfrm) {
            int i, j;
            uint8_t *src, *src2;

            src = srcY;
            for (j = 0; j < 17 + s->mspel * 2; j++) {
                for (i = 0; i < 17 + s->mspel * 2; i++)
                    src[i] = ((src[i] - 128) >> 1) + 128;
                src += s->linesize;
            }
            src  = srcU;
            src2 = srcV;
            for (j = 0; j < 9; j++) {
                for (i = 0; i < 9; i++) {
                    src[i]  = ((src[i]  - 128) >> 1) + 128;
                    src2[i] = ((src2[i] - 128) >> 1) + 128;
                }
                src  += s->uvlinesize;
                src2 += s->uvlinesize;
            }
        }
        /* if we deal with intensity compensation we need to scale source blocks */
        if (v->mv_mode == MV_PMODE_INTENSITY_COMP) {
            int i, j;
            uint8_t *src, *src2;

            src = srcY;
            for (j = 0; j < 17 + s->mspel * 2; j++) {
                for (i = 0; i < 17 + s->mspel * 2; i++)
                    src[i] = v->luty[src[i]];
                src += s->linesize;
            }
            src  = srcU;
            src2 = srcV;
            for (j = 0; j < 9; j++) {
                for (i = 0; i < 9; i++) {
                    src[i]  = v->lutuv[src[i]];
                    src2[i] = v->lutuv[src2[i]];
                }
                src  += s->uvlinesize;
                src2 += s->uvlinesize;
            }
        }
        srcY += s->mspel * (1 + s->linesize);
    }

        off    = 0;
        off_uv = 0;
    if (s->mspel) {
        dxy = ((my & 3) << 2) | (mx & 3);
        v->vc1dsp.put_vc1_mspel_pixels_tab[dxy](s->dest[0] + off    , srcY    , s->linesize, v->rnd);
        v->vc1dsp.put_vc1_mspel_pixels_tab[dxy](s->dest[0] + off + 8, srcY + 8, s->linesize, v->rnd);
        srcY += s->linesize * 8;
        v->vc1dsp.put_vc1_mspel_pixels_tab[dxy](s->dest[0] + off + 8 * s->linesize    , srcY    , s->linesize, v->rnd);
        v->vc1dsp.put_vc1_mspel_pixels_tab[dxy](s->dest[0] + off + 8 * s->linesize + 8, srcY + 8, s->linesize, v->rnd);
    } else { // hpel mc - always used for luma
        dxy = (my & 2) | ((mx & 2) >> 1);
        if (!v->rnd)
            dsp->put_pixels_tab[0][dxy](s->dest[0] + off, srcY, s->linesize, 16);
        else
            dsp->put_no_rnd_pixels_tab[0][dxy](s->dest[0] + off, srcY, s->linesize, 16);
    }

    if (s->flags & CODEC_FLAG_GRAY) return;
    /* Chroma MC always uses qpel bilinear */
    uvmx = (uvmx & 3) << 1;
    uvmy = (uvmy & 3) << 1;
    if (!v->rnd) {
        h264chroma->put_h264_chroma_pixels_tab[0](s->dest[1] + off_uv, srcU, s->uvlinesize, 8, uvmx, uvmy);
        h264chroma->put_h264_chroma_pixels_tab[0](s->dest[2] + off_uv, srcV, s->uvlinesize, 8, uvmx, uvmy);
    } else {
        v->vc1dsp.put_no_rnd_vc1_chroma_pixels_tab[0](s->dest[1] + off_uv, srcU, s->uvlinesize, 8, uvmx, uvmy);
        v->vc1dsp.put_no_rnd_vc1_chroma_pixels_tab[0](s->dest[2] + off_uv, srcV, s->uvlinesize, 8, uvmx, uvmy);
    }
}

static inline int median4(int a, int b, int c, int d)
{
    if (a < b) {
        if (c < d) return (FFMIN(b, d) + FFMAX(a, c)) / 2;
        else       return (FFMIN(b, c) + FFMAX(a, d)) / 2;
    } else {
        if (c < d) return (FFMIN(a, d) + FFMAX(b, c)) / 2;
        else       return (FFMIN(a, c) + FFMAX(b, d)) / 2;
    }
}

/** Do motion compensation for 4-MV macroblock - luminance block
 */
static void vc1_mc_4mv_luma(VC1Context *v, int n, int dir)
{
    MpegEncContext *s = &v->s;
    DSPContext *dsp = &v->s.dsp;
    uint8_t *srcY;
    int dxy, mx, my, src_x, src_y;
    int off;
    int fieldmv = (v->fcm == ILACE_FRAME) ? v->blk_mv_type[s->block_index[n]] : 0;
    int v_edge_pos = s->v_edge_pos >> v->field_mode;

    if ((!v->field_mode ||
         (v->ref_field_type[dir] == 1 && v->cur_field_type == 1)) &&
        !v->s.last_picture.f.data[0])
        return;

    mx = s->mv[dir][n][0];
    my = s->mv[dir][n][1];

    if (!dir) {
        if (v->field_mode) {
            if ((v->cur_field_type != v->ref_field_type[dir]) && v->second_field)
                srcY = s->current_picture.f.data[0];
            else
                srcY = s->last_picture.f.data[0];
        } else
            srcY = s->last_picture.f.data[0];
    } else
        srcY = s->next_picture.f.data[0];

    if(!srcY)
        return;

    if (v->field_mode) {
        if (v->cur_field_type != v->ref_field_type[dir])
            my = my - 2 + 4 * v->cur_field_type;
    }

    if (s->pict_type == AV_PICTURE_TYPE_P && n == 3 && v->field_mode) {
        int same_count = 0, opp_count = 0, k;
        int chosen_mv[2][4][2], f;
        int tx, ty;
        for (k = 0; k < 4; k++) {
            f = v->mv_f[0][s->block_index[k] + v->blocks_off];
            chosen_mv[f][f ? opp_count : same_count][0] = s->mv[0][k][0];
            chosen_mv[f][f ? opp_count : same_count][1] = s->mv[0][k][1];
            opp_count  += f;
            same_count += 1 - f;
        }
        f = opp_count > same_count;
        switch (f ? opp_count : same_count) {
        case 4:
            tx = median4(chosen_mv[f][0][0], chosen_mv[f][1][0],
                         chosen_mv[f][2][0], chosen_mv[f][3][0]);
            ty = median4(chosen_mv[f][0][1], chosen_mv[f][1][1],
                         chosen_mv[f][2][1], chosen_mv[f][3][1]);
            break;
        case 3:
            tx = mid_pred(chosen_mv[f][0][0], chosen_mv[f][1][0], chosen_mv[f][2][0]);
            ty = mid_pred(chosen_mv[f][0][1], chosen_mv[f][1][1], chosen_mv[f][2][1]);
            break;
        case 2:
            tx = (chosen_mv[f][0][0] + chosen_mv[f][1][0]) / 2;
            ty = (chosen_mv[f][0][1] + chosen_mv[f][1][1]) / 2;
            break;
        default:
            av_assert2(0);
        }
        s->current_picture.f.motion_val[1][s->block_index[0] + v->blocks_off][0] = tx;
        s->current_picture.f.motion_val[1][s->block_index[0] + v->blocks_off][1] = ty;
        for (k = 0; k < 4; k++)
            v->mv_f[1][s->block_index[k] + v->blocks_off] = f;
    }

    if (v->fcm == ILACE_FRAME) {  // not sure if needed for other types of picture
        int qx, qy;
        int width  = s->avctx->coded_width;
        int height = s->avctx->coded_height >> 1;
        qx = (s->mb_x * 16) + (mx >> 2);
        qy = (s->mb_y *  8) + (my >> 3);

        if (qx < -17)
            mx -= 4 * (qx + 17);
        else if (qx > width)
            mx -= 4 * (qx - width);
        if (qy < -18)
            my -= 8 * (qy + 18);
        else if (qy > height + 1)
            my -= 8 * (qy - height - 1);
    }

    if ((v->fcm == ILACE_FRAME) && fieldmv)
        off = ((n > 1) ? s->linesize : 0) + (n & 1) * 8;
    else
        off = s->linesize * 4 * (n & 2) + (n & 1) * 8;

    src_x = s->mb_x * 16 + (n & 1) * 8 + (mx >> 2);
    if (!fieldmv)
        src_y = s->mb_y * 16 + (n & 2) * 4 + (my >> 2);
    else
        src_y = s->mb_y * 16 + ((n > 1) ? 1 : 0) + (my >> 2);

    if (v->profile != PROFILE_ADVANCED) {
        src_x = av_clip(src_x, -16, s->mb_width  * 16);
        src_y = av_clip(src_y, -16, s->mb_height * 16);
    } else {
        src_x = av_clip(src_x, -17, s->avctx->coded_width);
        if (v->fcm == ILACE_FRAME) {
            if (src_y & 1)
                src_y = av_clip(src_y, -17, s->avctx->coded_height + 1);
            else
                src_y = av_clip(src_y, -18, s->avctx->coded_height);
        } else {
            src_y = av_clip(src_y, -18, s->avctx->coded_height + 1);
        }
    }

    srcY += src_y * s->linesize + src_x;
    if (v->field_mode && v->ref_field_type[dir])
        srcY += s->current_picture_ptr->f.linesize[0];

    if (fieldmv && !(src_y & 1))
        v_edge_pos--;
    if (fieldmv && (src_y & 1) && src_y < 4)
        src_y--;
    if (v->rangeredfrm || (v->mv_mode == MV_PMODE_INTENSITY_COMP)
        || s->h_edge_pos < 13 || v_edge_pos < 23
        || (unsigned)(src_x - s->mspel) > s->h_edge_pos - (mx & 3) - 8 - s->mspel * 2
        || (unsigned)(src_y - (s->mspel << fieldmv)) > v_edge_pos - (my & 3) - ((8 + s->mspel * 2) << fieldmv)) {
        srcY -= s->mspel * (1 + (s->linesize << fieldmv));
        /* check emulate edge stride and offset */
        s->vdsp.emulated_edge_mc(s->edge_emu_buffer, srcY, s->linesize,
                                 9 + s->mspel * 2, (9 + s->mspel * 2) << fieldmv,
                                 src_x - s->mspel, src_y - (s->mspel << fieldmv),
                                 s->h_edge_pos, v_edge_pos);
        srcY = s->edge_emu_buffer;
        /* if we deal with range reduction we need to scale source blocks */
        if (v->rangeredfrm) {
            int i, j;
            uint8_t *src;

            src = srcY;
            for (j = 0; j < 9 + s->mspel * 2; j++) {
                for (i = 0; i < 9 + s->mspel * 2; i++)
                    src[i] = ((src[i] - 128) >> 1) + 128;
                src += s->linesize << fieldmv;
            }
        }
        /* if we deal with intensity compensation we need to scale source blocks */
        if (v->mv_mode == MV_PMODE_INTENSITY_COMP) {
            int i, j;
            uint8_t *src;

            src = srcY;
            for (j = 0; j < 9 + s->mspel * 2; j++) {
                for (i = 0; i < 9 + s->mspel * 2; i++)
                    src[i] = v->luty[src[i]];
                src += s->linesize << fieldmv;
            }
        }
        srcY += s->mspel * (1 + (s->linesize << fieldmv));
    }

    if (s->mspel) {
        dxy = ((my & 3) << 2) | (mx & 3);
        v->vc1dsp.put_vc1_mspel_pixels_tab[dxy](s->dest[0] + off, srcY, s->linesize << fieldmv, v->rnd);
    } else { // hpel mc - always used for luma
        dxy = (my & 2) | ((mx & 2) >> 1);
        if (!v->rnd)
            dsp->put_pixels_tab[1][dxy](s->dest[0] + off, srcY, s->linesize, 8);
        else
            dsp->put_no_rnd_pixels_tab[1][dxy](s->dest[0] + off, srcY, s->linesize, 8);
    }
}

static av_always_inline int get_chroma_mv(int *mvx, int *mvy, int *a, int flag, int *tx, int *ty)
{
    int idx, i;
    static const int count[16] = { 0, 1, 1, 2, 1, 2, 2, 3, 1, 2, 2, 3, 2, 3, 3, 4};

    idx =  ((a[3] != flag) << 3)
         | ((a[2] != flag) << 2)
         | ((a[1] != flag) << 1)
         |  (a[0] != flag);
    if (!idx) {
        *tx = median4(mvx[0], mvx[1], mvx[2], mvx[3]);
        *ty = median4(mvy[0], mvy[1], mvy[2], mvy[3]);
        return 4;
    } else if (count[idx] == 1) {
        switch (idx) {
        case 0x1:
            *tx = mid_pred(mvx[1], mvx[2], mvx[3]);
            *ty = mid_pred(mvy[1], mvy[2], mvy[3]);
            return 3;
        case 0x2:
            *tx = mid_pred(mvx[0], mvx[2], mvx[3]);
            *ty = mid_pred(mvy[0], mvy[2], mvy[3]);
            return 3;
        case 0x4:
            *tx = mid_pred(mvx[0], mvx[1], mvx[3]);
            *ty = mid_pred(mvy[0], mvy[1], mvy[3]);
            return 3;
        case 0x8:
            *tx = mid_pred(mvx[0], mvx[1], mvx[2]);
            *ty = mid_pred(mvy[0], mvy[1], mvy[2]);
            return 3;
        }
    } else if (count[idx] == 2) {
        int t1 = 0, t2 = 0;
        for (i = 0; i < 3; i++)
            if (!a[i]) {
                t1 = i;
                break;
            }
        for (i = t1 + 1; i < 4; i++)
            if (!a[i]) {
                t2 = i;
                break;
            }
        *tx = (mvx[t1] + mvx[t2]) / 2;
        *ty = (mvy[t1] + mvy[t2]) / 2;
        return 2;
    } else {
        return 0;
    }
    return -1;
}

/** Do motion compensation for 4-MV macroblock - both chroma blocks
 */
static void vc1_mc_4mv_chroma(VC1Context *v, int dir)
{
    MpegEncContext *s = &v->s;
    H264ChromaContext *h264chroma = &v->h264chroma;
    uint8_t *srcU, *srcV;
    int uvmx, uvmy, uvsrc_x, uvsrc_y;
    int k, tx = 0, ty = 0;
    int mvx[4], mvy[4], intra[4], mv_f[4];
    int valid_count;
    int chroma_ref_type = v->cur_field_type, off = 0;
    int v_edge_pos = s->v_edge_pos >> v->field_mode;

    if (!v->field_mode && !v->s.last_picture.f.data[0])
        return;
    if (s->flags & CODEC_FLAG_GRAY)
        return;

    for (k = 0; k < 4; k++) {
        mvx[k] = s->mv[dir][k][0];
        mvy[k] = s->mv[dir][k][1];
        intra[k] = v->mb_type[0][s->block_index[k]];
        if (v->field_mode)
            mv_f[k] = v->mv_f[dir][s->block_index[k] + v->blocks_off];
    }

    /* calculate chroma MV vector from four luma MVs */
    if (!v->field_mode || (v->field_mode && !v->numref)) {
        valid_count = get_chroma_mv(mvx, mvy, intra, 0, &tx, &ty);
        chroma_ref_type = v->reffield;
        if (!valid_count) {
            s->current_picture.f.motion_val[1][s->block_index[0] + v->blocks_off][0] = 0;
            s->current_picture.f.motion_val[1][s->block_index[0] + v->blocks_off][1] = 0;
            v->luma_mv[s->mb_x][0] = v->luma_mv[s->mb_x][1] = 0;
            return; //no need to do MC for intra blocks
        }
    } else {
        int dominant = 0;
        if (mv_f[0] + mv_f[1] + mv_f[2] + mv_f[3] > 2)
            dominant = 1;
        valid_count = get_chroma_mv(mvx, mvy, mv_f, dominant, &tx, &ty);
        if (dominant)
            chroma_ref_type = !v->cur_field_type;
    }
    if (v->field_mode && chroma_ref_type == 1 && v->cur_field_type == 1 && !v->s.last_picture.f.data[0])
        return;
    s->current_picture.f.motion_val[1][s->block_index[0] + v->blocks_off][0] = tx;
    s->current_picture.f.motion_val[1][s->block_index[0] + v->blocks_off][1] = ty;
    uvmx = (tx + ((tx & 3) == 3)) >> 1;
    uvmy = (ty + ((ty & 3) == 3)) >> 1;

    v->luma_mv[s->mb_x][0] = uvmx;
    v->luma_mv[s->mb_x][1] = uvmy;

    if (v->fastuvmc) {
        uvmx = uvmx + ((uvmx < 0) ? (uvmx & 1) : -(uvmx & 1));
        uvmy = uvmy + ((uvmy < 0) ? (uvmy & 1) : -(uvmy & 1));
    }
    // Field conversion bias
    if (v->cur_field_type != chroma_ref_type)
        uvmy += 2 - 4 * chroma_ref_type;

    uvsrc_x = s->mb_x * 8 + (uvmx >> 2);
    uvsrc_y = s->mb_y * 8 + (uvmy >> 2);

    if (v->profile != PROFILE_ADVANCED) {
        uvsrc_x = av_clip(uvsrc_x, -8, s->mb_width  * 8);
        uvsrc_y = av_clip(uvsrc_y, -8, s->mb_height * 8);
    } else {
        uvsrc_x = av_clip(uvsrc_x, -8, s->avctx->coded_width  >> 1);
        uvsrc_y = av_clip(uvsrc_y, -8, s->avctx->coded_height >> 1);
    }

    if (!dir) {
        if (v->field_mode) {
            if ((v->cur_field_type != chroma_ref_type) && v->cur_field_type) {
                srcU = s->current_picture.f.data[1];
                srcV = s->current_picture.f.data[2];
            } else {
                srcU = s->last_picture.f.data[1];
                srcV = s->last_picture.f.data[2];
            }
        } else {
            srcU = s->last_picture.f.data[1];
            srcV = s->last_picture.f.data[2];
        }
    } else {
        srcU = s->next_picture.f.data[1];
        srcV = s->next_picture.f.data[2];
    }

    if(!srcU)
        return;

    srcU += uvsrc_y * s->uvlinesize + uvsrc_x;
    srcV += uvsrc_y * s->uvlinesize + uvsrc_x;

    if (v->field_mode) {
        if (chroma_ref_type) {
            srcU += s->current_picture_ptr->f.linesize[1];
            srcV += s->current_picture_ptr->f.linesize[2];
        }
        off = 0;
    }

    if (v->rangeredfrm || (v->mv_mode == MV_PMODE_INTENSITY_COMP)
        || s->h_edge_pos < 18 || v_edge_pos < 18
        || (unsigned)uvsrc_x > (s->h_edge_pos >> 1) - 9
        || (unsigned)uvsrc_y > (v_edge_pos    >> 1) - 9) {
        s->vdsp.emulated_edge_mc(s->edge_emu_buffer     , srcU, s->uvlinesize,
                                 8 + 1, 8 + 1, uvsrc_x, uvsrc_y,
                                 s->h_edge_pos >> 1, v_edge_pos >> 1);
        s->vdsp.emulated_edge_mc(s->edge_emu_buffer + 16, srcV, s->uvlinesize,
                                 8 + 1, 8 + 1, uvsrc_x, uvsrc_y,
                                 s->h_edge_pos >> 1, v_edge_pos >> 1);
        srcU = s->edge_emu_buffer;
        srcV = s->edge_emu_buffer + 16;

        /* if we deal with range reduction we need to scale source blocks */
        if (v->rangeredfrm) {
            int i, j;
            uint8_t *src, *src2;

            src  = srcU;
            src2 = srcV;
            for (j = 0; j < 9; j++) {
                for (i = 0; i < 9; i++) {
                    src[i]  = ((src[i]  - 128) >> 1) + 128;
                    src2[i] = ((src2[i] - 128) >> 1) + 128;
                }
                src  += s->uvlinesize;
                src2 += s->uvlinesize;
            }
        }
        /* if we deal with intensity compensation we need to scale source blocks */
        if (v->mv_mode == MV_PMODE_INTENSITY_COMP) {
            int i, j;
            uint8_t *src, *src2;

            src  = srcU;
            src2 = srcV;
            for (j = 0; j < 9; j++) {
                for (i = 0; i < 9; i++) {
                    src[i]  = v->lutuv[src[i]];
                    src2[i] = v->lutuv[src2[i]];
                }
                src  += s->uvlinesize;
                src2 += s->uvlinesize;
            }
        }
    }

    /* Chroma MC always uses qpel bilinear */
    uvmx = (uvmx & 3) << 1;
    uvmy = (uvmy & 3) << 1;
    if (!v->rnd) {
        h264chroma->put_h264_chroma_pixels_tab[0](s->dest[1] + off, srcU, s->uvlinesize, 8, uvmx, uvmy);
        h264chroma->put_h264_chroma_pixels_tab[0](s->dest[2] + off, srcV, s->uvlinesize, 8, uvmx, uvmy);
    } else {
        v->vc1dsp.put_no_rnd_vc1_chroma_pixels_tab[0](s->dest[1] + off, srcU, s->uvlinesize, 8, uvmx, uvmy);
        v->vc1dsp.put_no_rnd_vc1_chroma_pixels_tab[0](s->dest[2] + off, srcV, s->uvlinesize, 8, uvmx, uvmy);
    }
}

/** Do motion compensation for 4-MV field chroma macroblock (both U and V)
 */
static void vc1_mc_4mv_chroma4(VC1Context *v)
{
    MpegEncContext *s = &v->s;
    H264ChromaContext *h264chroma = &v->h264chroma;
    uint8_t *srcU, *srcV;
    int uvsrc_x, uvsrc_y;
    int uvmx_field[4], uvmy_field[4];
    int i, off, tx, ty;
    int fieldmv = v->blk_mv_type[s->block_index[0]];
    static const int s_rndtblfield[16] = { 0, 0, 1, 2, 4, 4, 5, 6, 2, 2, 3, 8, 6, 6, 7, 12 };
    int v_dist = fieldmv ? 1 : 4; // vertical offset for lower sub-blocks
    int v_edge_pos = s->v_edge_pos >> 1;

    if (!v->s.last_picture.f.data[0])
        return;
    if (s->flags & CODEC_FLAG_GRAY)
        return;

    for (i = 0; i < 4; i++) {
        tx = s->mv[0][i][0];
        uvmx_field[i] = (tx + ((tx & 3) == 3)) >> 1;
        ty = s->mv[0][i][1];
        if (fieldmv)
            uvmy_field[i] = (ty >> 4) * 8 + s_rndtblfield[ty & 0xF];
        else
            uvmy_field[i] = (ty + ((ty & 3) == 3)) >> 1;
    }

    for (i = 0; i < 4; i++) {
        off = (i & 1) * 4 + ((i & 2) ? v_dist * s->uvlinesize : 0);
        uvsrc_x = s->mb_x * 8 +  (i & 1) * 4           + (uvmx_field[i] >> 2);
        uvsrc_y = s->mb_y * 8 + ((i & 2) ? v_dist : 0) + (uvmy_field[i] >> 2);
        // FIXME: implement proper pull-back (see vc1cropmv.c, vc1CROPMV_ChromaPullBack())
        uvsrc_x = av_clip(uvsrc_x, -8, s->avctx->coded_width  >> 1);
        uvsrc_y = av_clip(uvsrc_y, -8, s->avctx->coded_height >> 1);
        srcU = s->last_picture.f.data[1] + uvsrc_y * s->uvlinesize + uvsrc_x;
        srcV = s->last_picture.f.data[2] + uvsrc_y * s->uvlinesize + uvsrc_x;
        uvmx_field[i] = (uvmx_field[i] & 3) << 1;
        uvmy_field[i] = (uvmy_field[i] & 3) << 1;

        if (fieldmv && !(uvsrc_y & 1))
            v_edge_pos = (s->v_edge_pos >> 1) - 1;

        if (fieldmv && (uvsrc_y & 1) && uvsrc_y < 2)
            uvsrc_y--;
        if ((v->mv_mode == MV_PMODE_INTENSITY_COMP)
            || s->h_edge_pos < 10 || v_edge_pos < (5 << fieldmv)
            || (unsigned)uvsrc_x > (s->h_edge_pos >> 1) - 5
            || (unsigned)uvsrc_y > v_edge_pos - (5 << fieldmv)) {
            s->vdsp.emulated_edge_mc(s->edge_emu_buffer, srcU, s->uvlinesize,
                                     5, (5 << fieldmv), uvsrc_x, uvsrc_y,
                                     s->h_edge_pos >> 1, v_edge_pos);
            s->vdsp.emulated_edge_mc(s->edge_emu_buffer + 16, srcV, s->uvlinesize,
                                     5, (5 << fieldmv), uvsrc_x, uvsrc_y,
                                     s->h_edge_pos >> 1, v_edge_pos);
            srcU = s->edge_emu_buffer;
            srcV = s->edge_emu_buffer + 16;

            /* if we deal with intensity compensation we need to scale source blocks */
            if (v->mv_mode == MV_PMODE_INTENSITY_COMP) {
                int i, j;
                uint8_t *src, *src2;

                src  = srcU;
                src2 = srcV;
                for (j = 0; j < 5; j++) {
                    for (i = 0; i < 5; i++) {
                        src[i]  = v->lutuv[src[i]];
                        src2[i] = v->lutuv[src2[i]];
                    }
                    src  += s->uvlinesize << 1;
                    src2 += s->uvlinesize << 1;
                }
            }
        }
        if (!v->rnd) {
            h264chroma->put_h264_chroma_pixels_tab[1](s->dest[1] + off, srcU, s->uvlinesize << fieldmv, 4, uvmx_field[i], uvmy_field[i]);
            h264chroma->put_h264_chroma_pixels_tab[1](s->dest[2] + off, srcV, s->uvlinesize << fieldmv, 4, uvmx_field[i], uvmy_field[i]);
        } else {
            v->vc1dsp.put_no_rnd_vc1_chroma_pixels_tab[1](s->dest[1] + off, srcU, s->uvlinesize << fieldmv, 4, uvmx_field[i], uvmy_field[i]);
            v->vc1dsp.put_no_rnd_vc1_chroma_pixels_tab[1](s->dest[2] + off, srcV, s->uvlinesize << fieldmv, 4, uvmx_field[i], uvmy_field[i]);
        }
    }
}

/***********************************************************************/
/**
 * @name VC-1 Block-level functions
 * @see 7.1.4, p91 and 8.1.1.7, p(1)04
 * @{
 */

/**
 * @def GET_MQUANT
 * @brief Get macroblock-level quantizer scale
 */
#define GET_MQUANT()                                           \
    if (v->dquantfrm) {                                        \
        int edges = 0;                                         \
        if (v->dqprofile == DQPROFILE_ALL_MBS) {               \
            if (v->dqbilevel) {                                \
                mquant = (get_bits1(gb)) ? v->altpq : v->pq;   \
            } else {                                           \
                mqdiff = get_bits(gb, 3);                      \
                if (mqdiff != 7)                               \
                    mquant = v->pq + mqdiff;                   \
                else                                           \
                    mquant = get_bits(gb, 5);                  \
            }                                                  \
        }                                                      \
        if (v->dqprofile == DQPROFILE_SINGLE_EDGE)             \
            edges = 1 << v->dqsbedge;                          \
        else if (v->dqprofile == DQPROFILE_DOUBLE_EDGES)       \
            edges = (3 << v->dqsbedge) % 15;                   \
        else if (v->dqprofile == DQPROFILE_FOUR_EDGES)         \
            edges = 15;                                        \
        if ((edges&1) && !s->mb_x)                             \
            mquant = v->altpq;                                 \
        if ((edges&2) && s->first_slice_line)                  \
            mquant = v->altpq;                                 \
        if ((edges&4) && s->mb_x == (s->mb_width - 1))         \
            mquant = v->altpq;                                 \
        if ((edges&8) && s->mb_y == (s->mb_height - 1))        \
            mquant = v->altpq;                                 \
        if (!mquant || mquant > 31) {                          \
            av_log(v->s.avctx, AV_LOG_ERROR,                   \
                   "Overriding invalid mquant %d\n", mquant);  \
            mquant = 1;                                        \
        }                                                      \
    }

/**
 * @def GET_MVDATA(_dmv_x, _dmv_y)
 * @brief Get MV differentials
 * @see MVDATA decoding from 8.3.5.2, p(1)20
 * @param _dmv_x Horizontal differential for decoded MV
 * @param _dmv_y Vertical differential for decoded MV
 */
#define GET_MVDATA(_dmv_x, _dmv_y)                                      \
    index = 1 + get_vlc2(gb, ff_vc1_mv_diff_vlc[s->mv_table_index].table, \
                         VC1_MV_DIFF_VLC_BITS, 2);                      \
    if (index > 36) {                                                   \
        mb_has_coeffs = 1;                                              \
        index -= 37;                                                    \
    } else                                                              \
        mb_has_coeffs = 0;                                              \
    s->mb_intra = 0;                                                    \
    if (!index) {                                                       \
        _dmv_x = _dmv_y = 0;                                            \
    } else if (index == 35) {                                           \
        _dmv_x = get_bits(gb, v->k_x - 1 + s->quarter_sample);          \
        _dmv_y = get_bits(gb, v->k_y - 1 + s->quarter_sample);          \
    } else if (index == 36) {                                           \
        _dmv_x = 0;                                                     \
        _dmv_y = 0;                                                     \
        s->mb_intra = 1;                                                \
    } else {                                                            \
        index1 = index % 6;                                             \
        if (!s->quarter_sample && index1 == 5) val = 1;                 \
        else                                   val = 0;                 \
        if (size_table[index1] - val > 0)                               \
            val = get_bits(gb, size_table[index1] - val);               \
        else                                   val = 0;                 \
        sign = 0 - (val&1);                                             \
        _dmv_x = (sign ^ ((val>>1) + offset_table[index1])) - sign;     \
                                                                        \
        index1 = index / 6;                                             \
        if (!s->quarter_sample && index1 == 5) val = 1;                 \
        else                                   val = 0;                 \
        if (size_table[index1] - val > 0)                               \
            val = get_bits(gb, size_table[index1] - val);               \
        else                                   val = 0;                 \
        sign = 0 - (val & 1);                                           \
        _dmv_y = (sign ^ ((val >> 1) + offset_table[index1])) - sign;   \
    }

static av_always_inline void get_mvdata_interlaced(VC1Context *v, int *dmv_x,
                                                   int *dmv_y, int *pred_flag)
{
    int index, index1;
    int extend_x = 0, extend_y = 0;
    GetBitContext *gb = &v->s.gb;
    int bits, esc;
    int val, sign;
    const int* offs_tab;

    if (v->numref) {
        bits = VC1_2REF_MVDATA_VLC_BITS;
        esc  = 125;
    } else {
        bits = VC1_1REF_MVDATA_VLC_BITS;
        esc  = 71;
    }
    switch (v->dmvrange) {
    case 1:
        extend_x = 1;
        break;
    case 2:
        extend_y = 1;
        break;
    case 3:
        extend_x = extend_y = 1;
        break;
    }
    index = get_vlc2(gb, v->imv_vlc->table, bits, 3);
    if (index == esc) {
        *dmv_x = get_bits(gb, v->k_x);
        *dmv_y = get_bits(gb, v->k_y);
        if (v->numref) {
            if (pred_flag) {
                *pred_flag = *dmv_y & 1;
                *dmv_y     = (*dmv_y + *pred_flag) >> 1;
            } else {
                *dmv_y     = (*dmv_y + (*dmv_y & 1)) >> 1;
            }
        }
    }
    else {
        av_assert0(index < esc);
        if (extend_x)
            offs_tab = offset_table2;
        else
            offs_tab = offset_table1;
        index1 = (index + 1) % 9;
        if (index1 != 0) {
            val    = get_bits(gb, index1 + extend_x);
            sign   = 0 -(val & 1);
            *dmv_x = (sign ^ ((val >> 1) + offs_tab[index1])) - sign;
        } else
            *dmv_x = 0;
        if (extend_y)
            offs_tab = offset_table2;
        else
            offs_tab = offset_table1;
        index1 = (index + 1) / 9;
        if (index1 > v->numref) {
            val    = get_bits(gb, (index1 + (extend_y << v->numref)) >> v->numref);
            sign   = 0 - (val & 1);
            *dmv_y = (sign ^ ((val >> 1) + offs_tab[index1 >> v->numref])) - sign;
        } else
            *dmv_y = 0;
        if (v->numref && pred_flag)
            *pred_flag = index1 & 1;
    }
}

static av_always_inline int scaleforsame_x(VC1Context *v, int n /* MV */, int dir)
{
    int scaledvalue, refdist;
    int scalesame1, scalesame2;
    int scalezone1_x, zone1offset_x;
    int table_index = dir ^ v->second_field;

    if (v->s.pict_type != AV_PICTURE_TYPE_B)
        refdist = v->refdist;
    else
        refdist = dir ? v->brfd : v->frfd;
    if (refdist > 3)
        refdist = 3;
    scalesame1    = ff_vc1_field_mvpred_scales[table_index][1][refdist];
    scalesame2    = ff_vc1_field_mvpred_scales[table_index][2][refdist];
    scalezone1_x  = ff_vc1_field_mvpred_scales[table_index][3][refdist];
    zone1offset_x = ff_vc1_field_mvpred_scales[table_index][5][refdist];

    if (FFABS(n) > 255)
        scaledvalue = n;
    else {
        if (FFABS(n) < scalezone1_x)
            scaledvalue = (n * scalesame1) >> 8;
        else {
            if (n < 0)
                scaledvalue = ((n * scalesame2) >> 8) - zone1offset_x;
            else
                scaledvalue = ((n * scalesame2) >> 8) + zone1offset_x;
        }
    }
    return av_clip(scaledvalue, -v->range_x, v->range_x - 1);
}

static av_always_inline int scaleforsame_y(VC1Context *v, int i, int n /* MV */, int dir)
{
    int scaledvalue, refdist;
    int scalesame1, scalesame2;
    int scalezone1_y,…