libavcodec/vc1dec.c

   1 /*
   2  * VC-1 and WMV3 decoder
   3  * Copyright (c) 2011 Mashiat Sarker Shakkhar
   4  * Copyright (c) 2006-2007 Konstantin Shishkov
   5  * Partly based on vc9.c (c) 2005 Anonymous, Alex Beregszaszi, Michael Niedermayer
   6  *
   7  * This file is part of Libav.
   8  *
   9  * Libav is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public
  11  * License as published by the Free Software Foundation; either
  12  * version 2.1 of the License, or (at your option) any later version.
  13  *
  14  * Libav is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with Libav; if not, write to the Free Software
  21  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  22  */
  23
  24 /**
  25  * @file
  26  * VC-1 and WMV3 decoder
  27  */
  28
  29 #include "internal.h"
  30 #include "dsputil.h"
  31 #include "avcodec.h"
  32 #include "mpegvideo.h"
  33 #include "h263.h"
  34 #include "vc1.h"
  35 #include "vc1data.h"
  36 #include "vc1acdata.h"
  37 #include "msmpeg4data.h"
  38 #include "unary.h"
  39 #include "mathops.h"
  40 #include "vdpau_internal.h"
  41
  42 #undef NDEBUG
  43 #include <assert.h>
  44
  45 #define MB_INTRA_VLC_BITS 9
  46 #define DC_VLC_BITS 9
  47
  48
  49 // offset tables for interlaced picture MVDATA decoding
  50 static const int offset_table1[9] = {  0,  1,  2,  4,  8, 16, 32,  64, 128 };
  51 static const int offset_table2[9] = {  0,  1,  3,  7, 15, 31, 63, 127, 255 };
  52
  53 /***********************************************************************/
  54 /**
  55  * @name VC-1 Bitplane decoding
  56  * @see 8.7, p56
  57  * @{
  58  */
  59
  60 /**
  61  * Imode types
  62  * @{
  63  */
  64 enum Imode {
  65     IMODE_RAW,
  66     IMODE_NORM2,
  67     IMODE_DIFF2,
  68     IMODE_NORM6,
  69     IMODE_DIFF6,
  70     IMODE_ROWSKIP,
  71     IMODE_COLSKIP
  72 };
  73 /** @} */ //imode defines
  74
  75
  76 /** @} */ //Bitplane group
  77
  78 static void vc1_put_signed_blocks_clamped(VC1Context *v)
  79 {
  80     MpegEncContext *s = &v->s;
  81     int topleft_mb_pos, top_mb_pos;
  82     int stride_y, fieldtx;
  83     int v_dist;
  84
  85     /* The put pixels loop is always one MB row behind the decoding loop,
  86      * because we can only put pixels when overlap filtering is done, and
  87      * for filtering of the bottom edge of a MB, we need the next MB row
  88      * present as well.
  89      * Within the row, the put pixels loop is also one MB col behind the
  90      * decoding loop. The reason for this is again, because for filtering
  91      * of the right MB edge, we need the next MB present. */
  92     if (!s->first_slice_line) {
  93         if (s->mb_x) {
  94             topleft_mb_pos = (s->mb_y - 1) * s->mb_stride + s->mb_x - 1;
  95             fieldtx        = v->fieldtx_plane[topleft_mb_pos];
  96             stride_y       = s->linesize << fieldtx;
  97             v_dist         = (16 - fieldtx) >> (fieldtx == 0);
  98             s->dsp.put_signed_pixels_clamped(v->block[v->topleft_blk_idx][0],
  99                                              s->dest[0] - 16 * s->linesize - 16,
 100                                              stride_y);
 101             s->dsp.put_signed_pixels_clamped(v->block[v->topleft_blk_idx][1],
 102                                              s->dest[0] - 16 * s->linesize - 8,
 103                                              stride_y);
 104             s->dsp.put_signed_pixels_clamped(v->block[v->topleft_blk_idx][2],
 105                                              s->dest[0] - v_dist * s->linesize - 16,
 106                                              stride_y);
 107             s->dsp.put_signed_pixels_clamped(v->block[v->topleft_blk_idx][3],
 108                                              s->dest[0] - v_dist * s->linesize - 8,
 109                                              stride_y);
 110             s->dsp.put_signed_pixels_clamped(v->block[v->topleft_blk_idx][4],
 111                                              s->dest[1] - 8 * s->uvlinesize - 8,
 112                                              s->uvlinesize);
 113             s->dsp.put_signed_pixels_clamped(v->block[v->topleft_blk_idx][5],
 114                                              s->dest[2] - 8 * s->uvlinesize - 8,
 115                                              s->uvlinesize);
 116         }
 117         if (s->mb_x == s->mb_width - 1) {
 118             top_mb_pos = (s->mb_y - 1) * s->mb_stride + s->mb_x;
 119             fieldtx    = v->fieldtx_plane[top_mb_pos];
 120             stride_y   = s->linesize << fieldtx;
 121             v_dist     = fieldtx ? 15 : 8;
 122             s->dsp.put_signed_pixels_clamped(v->block[v->top_blk_idx][0],
 123                                              s->dest[0] - 16 * s->linesize,
 124                                              stride_y);
 125             s->dsp.put_signed_pixels_clamped(v->block[v->top_blk_idx][1],
 126                                              s->dest[0] - 16 * s->linesize + 8,
 127                                              stride_y);
 128             s->dsp.put_signed_pixels_clamped(v->block[v->top_blk_idx][2],
 129                                              s->dest[0] - v_dist * s->linesize,
 130                                              stride_y);
 131             s->dsp.put_signed_pixels_clamped(v->block[v->top_blk_idx][3],
 132                                              s->dest[0] - v_dist * s->linesize + 8,
 133                                              stride_y);
 134             s->dsp.put_signed_pixels_clamped(v->block[v->top_blk_idx][4],
 135                                              s->dest[1] - 8 * s->uvlinesize,
 136                                              s->uvlinesize);
 137             s->dsp.put_signed_pixels_clamped(v->block[v->top_blk_idx][5],
 138                                              s->dest[2] - 8 * s->uvlinesize,
 139                                              s->uvlinesize);
 140         }
 141     }
 142
 143 #define inc_blk_idx(idx) do { \
 144         idx++; \
 145         if (idx >= v->n_allocated_blks) \
 146             idx = 0; \
 147     } while (0)
 148
 149     inc_blk_idx(v->topleft_blk_idx);
 150     inc_blk_idx(v->top_blk_idx);
 151     inc_blk_idx(v->left_blk_idx);
 152     inc_blk_idx(v->cur_blk_idx);
 153 }
 154
 155 static void vc1_loop_filter_iblk(VC1Context *v, int pq)
 156 {
 157     MpegEncContext *s = &v->s;
 158     int j;
 159     if (!s->first_slice_line) {
 160         v->vc1dsp.vc1_v_loop_filter16(s->dest[0], s->linesize, pq);
 161         if (s->mb_x)
 162             v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 16 * s->linesize, s->linesize, pq);
 163         v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 16 * s->linesize + 8, s->linesize, pq);
 164         for (j = 0; j < 2; j++) {
 165             v->vc1dsp.vc1_v_loop_filter8(s->dest[j + 1], s->uvlinesize, pq);
 166             if (s->mb_x)
 167                 v->vc1dsp.vc1_h_loop_filter8(s->dest[j + 1] - 8 * s->uvlinesize, s->uvlinesize, pq);
 168         }
 169     }
 170     v->vc1dsp.vc1_v_loop_filter16(s->dest[0] + 8 * s->linesize, s->linesize, pq);
 171
 172     if (s->mb_y == s->end_mb_y - 1) {
 173         if (s->mb_x) {
 174             v->vc1dsp.vc1_h_loop_filter16(s->dest[0], s->linesize, pq);
 175             v->vc1dsp.vc1_h_loop_filter8(s->dest[1], s->uvlinesize, pq);
 176             v->vc1dsp.vc1_h_loop_filter8(s->dest[2], s->uvlinesize, pq);
 177         }
 178         v->vc1dsp.vc1_h_loop_filter16(s->dest[0] + 8, s->linesize, pq);
 179     }
 180 }
 181
 182 static void vc1_loop_filter_iblk_delayed(VC1Context *v, int pq)
 183 {
 184     MpegEncContext *s = &v->s;
 185     int j;
 186
 187     /* The loopfilter runs 1 row and 1 column behind the overlap filter, which
 188      * means it runs two rows/cols behind the decoding loop. */
 189     if (!s->first_slice_line) {
 190         if (s->mb_x) {
 191             if (s->mb_y >= s->start_mb_y + 2) {
 192                 v->vc1dsp.vc1_v_loop_filter16(s->dest[0] - 16 * s->linesize - 16, s->linesize, pq);
 193
 194                 if (s->mb_x >= 2)
 195                     v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 32 * s->linesize - 16, s->linesize, pq);
 196                 v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 32 * s->linesize - 8, s->linesize, pq);
 197                 for (j = 0; j < 2; j++) {
 198                     v->vc1dsp.vc1_v_loop_filter8(s->dest[j + 1] - 8 * s->uvlinesize - 8, s->uvlinesize, pq);
 199                     if (s->mb_x >= 2) {
 200                         v->vc1dsp.vc1_h_loop_filter8(s->dest[j + 1] - 16 * s->uvlinesize - 8, s->uvlinesize, pq);
 201                     }
 202                 }
 203             }
 204             v->vc1dsp.vc1_v_loop_filter16(s->dest[0] - 8 * s->linesize - 16, s->linesize, pq);
 205         }
 206
 207         if (s->mb_x == s->mb_width - 1) {
 208             if (s->mb_y >= s->start_mb_y + 2) {
 209                 v->vc1dsp.vc1_v_loop_filter16(s->dest[0] - 16 * s->linesize, s->linesize, pq);
 210
 211                 if (s->mb_x)
 212                     v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 32 * s->linesize, s->linesize, pq);
 213                 v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 32 * s->linesize + 8, s->linesize, pq);
 214                 for (j = 0; j < 2; j++) {
 215                     v->vc1dsp.vc1_v_loop_filter8(s->dest[j + 1] - 8 * s->uvlinesize, s->uvlinesize, pq);
 216                     if (s->mb_x >= 2) {
 217                         v->vc1dsp.vc1_h_loop_filter8(s->dest[j + 1] - 16 * s->uvlinesize, s->uvlinesize, pq);
 218                     }
 219                 }
 220             }
 221             v->vc1dsp.vc1_v_loop_filter16(s->dest[0] - 8 * s->linesize, s->linesize, pq);
 222         }
 223
 224         if (s->mb_y == s->end_mb_y) {
 225             if (s->mb_x) {
 226                 if (s->mb_x >= 2)
 227                     v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 16 * s->linesize - 16, s->linesize, pq);
 228                 v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 16 * s->linesize - 8, s->linesize, pq);
 229                 if (s->mb_x >= 2) {
 230                     for (j = 0; j < 2; j++) {
 231                         v->vc1dsp.vc1_h_loop_filter8(s->dest[j + 1] - 8 * s->uvlinesize - 8, s->uvlinesize, pq);
 232                     }
 233                 }
 234             }
 235
 236             if (s->mb_x == s->mb_width - 1) {
 237                 if (s->mb_x)
 238                     v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 16 * s->linesize, s->linesize, pq);
 239                 v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 16 * s->linesize + 8, s->linesize, pq);
 240                 if (s->mb_x) {
 241                     for (j = 0; j < 2; j++) {
 242                         v->vc1dsp.vc1_h_loop_filter8(s->dest[j + 1] - 8 * s->uvlinesize, s->uvlinesize, pq);
 243                     }
 244                 }
 245             }
 246         }
 247     }
 248 }
 249
 250 static void vc1_smooth_overlap_filter_iblk(VC1Context *v)
 251 {
 252     MpegEncContext *s = &v->s;
 253     int mb_pos;
 254
 255     if (v->condover == CONDOVER_NONE)
 256         return;
 257
 258     mb_pos = s->mb_x + s->mb_y * s->mb_stride;
 259
 260     /* Within a MB, the horizontal overlap always runs before the vertical.
 261      * To accomplish that, we run the H on left and internal borders of the
 262      * currently decoded MB. Then, we wait for the next overlap iteration
 263      * to do H overlap on the right edge of this MB, before moving over and
 264      * running the V overlap. Therefore, the V overlap makes us trail by one
 265      * MB col and the H overlap filter makes us trail by one MB row. This
 266      * is reflected in the time at which we run the put_pixels loop. */
 267     if (v->condover == CONDOVER_ALL || v->pq >= 9 || v->over_flags_plane[mb_pos]) {
 268         if (s->mb_x && (v->condover == CONDOVER_ALL || v->pq >= 9 ||
 269                         v->over_flags_plane[mb_pos - 1])) {
 270             v->vc1dsp.vc1_h_s_overlap(v->block[v->left_blk_idx][1],
 271                                       v->block[v->cur_blk_idx][0]);
 272             v->vc1dsp.vc1_h_s_overlap(v->block[v->left_blk_idx][3],
 273                                       v->block[v->cur_blk_idx][2]);
 274             if (!(s->flags & CODEC_FLAG_GRAY)) {
 275                 v->vc1dsp.vc1_h_s_overlap(v->block[v->left_blk_idx][4],
 276                                           v->block[v->cur_blk_idx][4]);
 277                 v->vc1dsp.vc1_h_s_overlap(v->block[v->left_blk_idx][5],
 278                                           v->block[v->cur_blk_idx][5]);
 279             }
 280         }
 281         v->vc1dsp.vc1_h_s_overlap(v->block[v->cur_blk_idx][0],
 282                                   v->block[v->cur_blk_idx][1]);
 283         v->vc1dsp.vc1_h_s_overlap(v->block[v->cur_blk_idx][2],
 284                                   v->block[v->cur_blk_idx][3]);
 285
 286         if (s->mb_x == s->mb_width - 1) {
 287             if (!s->first_slice_line && (v->condover == CONDOVER_ALL || v->pq >= 9 ||
 288                                          v->over_flags_plane[mb_pos - s->mb_stride])) {
 289                 v->vc1dsp.vc1_v_s_overlap(v->block[v->top_blk_idx][2],
 290                                           v->block[v->cur_blk_idx][0]);
 291                 v->vc1dsp.vc1_v_s_overlap(v->block[v->top_blk_idx][3],
 292                                           v->block[v->cur_blk_idx][1]);
 293                 if (!(s->flags & CODEC_FLAG_GRAY)) {
 294                     v->vc1dsp.vc1_v_s_overlap(v->block[v->top_blk_idx][4],
 295                                               v->block[v->cur_blk_idx][4]);
 296                     v->vc1dsp.vc1_v_s_overlap(v->block[v->top_blk_idx][5],
 297                                               v->block[v->cur_blk_idx][5]);
 298                 }
 299             }
 300             v->vc1dsp.vc1_v_s_overlap(v->block[v->cur_blk_idx][0],
 301                                       v->block[v->cur_blk_idx][2]);
 302             v->vc1dsp.vc1_v_s_overlap(v->block[v->cur_blk_idx][1],
 303                                       v->block[v->cur_blk_idx][3]);
 304         }
 305     }
 306     if (s->mb_x && (v->condover == CONDOVER_ALL || v->over_flags_plane[mb_pos - 1])) {
 307         if (!s->first_slice_line && (v->condover == CONDOVER_ALL || v->pq >= 9 ||
 308                                      v->over_flags_plane[mb_pos - s->mb_stride - 1])) {
 309             v->vc1dsp.vc1_v_s_overlap(v->block[v->topleft_blk_idx][2],
 310                                       v->block[v->left_blk_idx][0]);
 311             v->vc1dsp.vc1_v_s_overlap(v->block[v->topleft_blk_idx][3],
 312                                       v->block[v->left_blk_idx][1]);
 313             if (!(s->flags & CODEC_FLAG_GRAY)) {
 314                 v->vc1dsp.vc1_v_s_overlap(v->block[v->topleft_blk_idx][4],
 315                                           v->block[v->left_blk_idx][4]);
 316                 v->vc1dsp.vc1_v_s_overlap(v->block[v->topleft_blk_idx][5],
 317                                           v->block[v->left_blk_idx][5]);
 318             }
 319         }
 320         v->vc1dsp.vc1_v_s_overlap(v->block[v->left_blk_idx][0],
 321                                   v->block[v->left_blk_idx][2]);
 322         v->vc1dsp.vc1_v_s_overlap(v->block[v->left_blk_idx][1],
 323                                   v->block[v->left_blk_idx][3]);
 324     }
 325 }
 326
 327 /** Do motion compensation over 1 macroblock
 328  * Mostly adapted hpel_motion and qpel_motion from mpegvideo.c
 329  */
 330 static void vc1_mc_1mv(VC1Context *v, int dir)
 331 {
 332     MpegEncContext *s = &v->s;
 333     DSPContext *dsp   = &v->s.dsp;
 334     uint8_t *srcY, *srcU, *srcV;
 335     int dxy, mx, my, uvmx, uvmy, src_x, src_y, uvsrc_x, uvsrc_y;
 336     int off, off_uv;
 337     int v_edge_pos = s->v_edge_pos >> v->field_mode;
 338
 339     if ((!v->field_mode ||
 340          (v->ref_field_type[dir] == 1 && v->cur_field_type == 1)) &&
 341         !v->s.last_picture.f.data[0])
 342         return;
 343
 344     mx = s->mv[dir][0][0];
 345     my = s->mv[dir][0][1];
 346
 347     // store motion vectors for further use in B frames
 348     if (s->pict_type == AV_PICTURE_TYPE_P) {
 349         s->current_picture.f.motion_val[1][s->block_index[0] + v->blocks_off][0] = mx;
 350         s->current_picture.f.motion_val[1][s->block_index[0] + v->blocks_off][1] = my;
 351     }
 352
 353     uvmx = (mx + ((mx & 3) == 3)) >> 1;
 354     uvmy = (my + ((my & 3) == 3)) >> 1;
 355     v->luma_mv[s->mb_x][0] = uvmx;
 356     v->luma_mv[s->mb_x][1] = uvmy;
 357
 358     if (v->field_mode &&
 359         v->cur_field_type != v->ref_field_type[dir]) {
 360         my   = my   - 2 + 4 * v->cur_field_type;
 361         uvmy = uvmy - 2 + 4 * v->cur_field_type;
 362     }
 363
 364     // fastuvmc shall be ignored for interlaced frame picture
 365     if (v->fastuvmc && (v->fcm != ILACE_FRAME)) {
 366         uvmx = uvmx + ((uvmx < 0) ? (uvmx & 1) : -(uvmx & 1));
 367         uvmy = uvmy + ((uvmy < 0) ? (uvmy & 1) : -(uvmy & 1));
 368     }
 369     if (v->field_mode) { // interlaced field picture
 370         if (!dir) {
 371             if ((v->cur_field_type != v->ref_field_type[dir]) && v->cur_field_type) {
 372                 srcY = s->current_picture.f.data[0];
 373                 srcU = s->current_picture.f.data[1];
 374                 srcV = s->current_picture.f.data[2];
 375             } else {
 376                 srcY = s->last_picture.f.data[0];
 377                 srcU = s->last_picture.f.data[1];
 378                 srcV = s->last_picture.f.data[2];
 379             }
 380         } else {
 381             srcY = s->next_picture.f.data[0];
 382             srcU = s->next_picture.f.data[1];
 383             srcV = s->next_picture.f.data[2];
 384         }
 385     } else {
 386         if (!dir) {
 387             srcY = s->last_picture.f.data[0];
 388             srcU = s->last_picture.f.data[1];
 389             srcV = s->last_picture.f.data[2];
 390         } else {
 391             srcY = s->next_picture.f.data[0];
 392             srcU = s->next_picture.f.data[1];
 393             srcV = s->next_picture.f.data[2];
 394         }
 395     }
 396
 397     src_x   = s->mb_x * 16 + (mx   >> 2);
 398     src_y   = s->mb_y * 16 + (my   >> 2);
 399     uvsrc_x = s->mb_x *  8 + (uvmx >> 2);
 400     uvsrc_y = s->mb_y *  8 + (uvmy >> 2);
 401
 402     if (v->profile != PROFILE_ADVANCED) {
 403         src_x   = av_clip(  src_x, -16, s->mb_width  * 16);
 404         src_y   = av_clip(  src_y, -16, s->mb_height * 16);
 405         uvsrc_x = av_clip(uvsrc_x,  -8, s->mb_width  *  8);
 406         uvsrc_y = av_clip(uvsrc_y,  -8, s->mb_height *  8);
 407     } else {
 408         src_x   = av_clip(  src_x, -17, s->avctx->coded_width);
 409         src_y   = av_clip(  src_y, -18, s->avctx->coded_height + 1);
 410         uvsrc_x = av_clip(uvsrc_x,  -8, s->avctx->coded_width  >> 1);
 411         uvsrc_y = av_clip(uvsrc_y,  -8, s->avctx->coded_height >> 1);
 412     }
 413
 414     srcY += src_y   * s->linesize   + src_x;
 415     srcU += uvsrc_y * s->uvlinesize + uvsrc_x;
 416     srcV += uvsrc_y * s->uvlinesize + uvsrc_x;
 417
 418     if (v->field_mode && v->ref_field_type[dir]) {
 419         srcY += s->current_picture_ptr->f.linesize[0];
 420         srcU += s->current_picture_ptr->f.linesize[1];
 421         srcV += s->current_picture_ptr->f.linesize[2];
 422     }
 423
 424     /* for grayscale we should not try to read from unknown area */
 425     if (s->flags & CODEC_FLAG_GRAY) {
 426         srcU = s->edge_emu_buffer + 18 * s->linesize;
 427         srcV = s->edge_emu_buffer + 18 * s->linesize;
 428     }
 429
 430     if (v->rangeredfrm || (v->mv_mode == MV_PMODE_INTENSITY_COMP)
 431         || s->h_edge_pos < 22 || v_edge_pos < 22
 432         || (unsigned)(src_x - s->mspel) > s->h_edge_pos - (mx&3) - 16 - s->mspel * 3
 433         || (unsigned)(src_y - 1)        > v_edge_pos    - (my&3) - 16 - 3) {
 434         uint8_t *uvbuf = s->edge_emu_buffer + 19 * s->linesize;
 435
 436         srcY -= s->mspel * (1 + s->linesize);
 437         s->dsp.emulated_edge_mc(s->edge_emu_buffer, srcY, s->linesize,
 438                                 17 + s->mspel * 2, 17 + s->mspel * 2,
 439                                 src_x - s->mspel, src_y - s->mspel,
 440                                 s->h_edge_pos, v_edge_pos);
 441         srcY = s->edge_emu_buffer;
 442         s->dsp.emulated_edge_mc(uvbuf     , srcU, s->uvlinesize, 8 + 1, 8 + 1,
 443                                 uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, v_edge_pos >> 1);
 444         s->dsp.emulated_edge_mc(uvbuf + 16, srcV, s->uvlinesize, 8 + 1, 8 + 1,
 445                                 uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, v_edge_pos >> 1);
 446         srcU = uvbuf;
 447         srcV = uvbuf + 16;
 448         /* if we deal with range reduction we need to scale source blocks */
 449         if (v->rangeredfrm) {
 450             int i, j;
 451             uint8_t *src, *src2;
 452
 453             src = srcY;
 454             for (j = 0; j < 17 + s->mspel * 2; j++) {
 455                 for (i = 0; i < 17 + s->mspel * 2; i++)
 456                     src[i] = ((src[i] - 128) >> 1) + 128;
 457                 src += s->linesize;
 458             }
 459             src  = srcU;
 460             src2 = srcV;
 461             for (j = 0; j < 9; j++) {
 462                 for (i = 0; i < 9; i++) {
 463                     src[i]  = ((src[i]  - 128) >> 1) + 128;
 464                     src2[i] = ((src2[i] - 128) >> 1) + 128;
 465                 }
 466                 src  += s->uvlinesize;
 467                 src2 += s->uvlinesize;
 468             }
 469         }
 470         /* if we deal with intensity compensation we need to scale source blocks */
 471         if (v->mv_mode == MV_PMODE_INTENSITY_COMP) {
 472             int i, j;
 473             uint8_t *src, *src2;
 474
 475             src = srcY;
 476             for (j = 0; j < 17 + s->mspel * 2; j++) {
 477                 for (i = 0; i < 17 + s->mspel * 2; i++)
 478                     src[i] = v->luty[src[i]];
 479                 src += s->linesize;
 480             }
 481             src  = srcU;
 482             src2 = srcV;
 483             for (j = 0; j < 9; j++) {
 484                 for (i = 0; i < 9; i++) {
 485                     src[i]  = v->lutuv[src[i]];
 486                     src2[i] = v->lutuv[src2[i]];
 487                 }
 488                 src  += s->uvlinesize;
 489                 src2 += s->uvlinesize;
 490             }
 491         }
 492         srcY += s->mspel * (1 + s->linesize);
 493     }
 494
 495     if (v->field_mode && v->cur_field_type) {
 496         off    = s->current_picture_ptr->f.linesize[0];
 497         off_uv = s->current_picture_ptr->f.linesize[1];
 498     } else {
 499         off    = 0;
 500         off_uv = 0;
 501     }
 502     if (s->mspel) {
 503         dxy = ((my & 3) << 2) | (mx & 3);
 504         v->vc1dsp.put_vc1_mspel_pixels_tab[dxy](s->dest[0] + off    , srcY    , s->linesize, v->rnd);
 505         v->vc1dsp.put_vc1_mspel_pixels_tab[dxy](s->dest[0] + off + 8, srcY + 8, s->linesize, v->rnd);
 506         srcY += s->linesize * 8;
 507         v->vc1dsp.put_vc1_mspel_pixels_tab[dxy](s->dest[0] + off + 8 * s->linesize    , srcY    , s->linesize, v->rnd);
 508         v->vc1dsp.put_vc1_mspel_pixels_tab[dxy](s->dest[0] + off + 8 * s->linesize + 8, srcY + 8, s->linesize, v->rnd);
 509     } else { // hpel mc - always used for luma
 510         dxy = (my & 2) | ((mx & 2) >> 1);
 511         if (!v->rnd)
 512             dsp->put_pixels_tab[0][dxy](s->dest[0] + off, srcY, s->linesize, 16);
 513         else
 514             dsp->put_no_rnd_pixels_tab[0][dxy](s->dest[0] + off, srcY, s->linesize, 16);
 515     }
 516
 517     if (s->flags & CODEC_FLAG_GRAY) return;
 518     /* Chroma MC always uses qpel bilinear */
 519     uvmx = (uvmx & 3) << 1;
 520     uvmy = (uvmy & 3) << 1;
 521     if (!v->rnd) {
 522         dsp->put_h264_chroma_pixels_tab[0](s->dest[1] + off_uv, srcU, s->uvlinesize, 8, uvmx, uvmy);
 523         dsp->put_h264_chroma_pixels_tab[0](s->dest[2] + off_uv, srcV, s->uvlinesize, 8, uvmx, uvmy);
 524     } else {
 525         v->vc1dsp.put_no_rnd_vc1_chroma_pixels_tab[0](s->dest[1] + off_uv, srcU, s->uvlinesize, 8, uvmx, uvmy);
 526         v->vc1dsp.put_no_rnd_vc1_chroma_pixels_tab[0](s->dest[2] + off_uv, srcV, s->uvlinesize, 8, uvmx, uvmy);
 527     }
 528 }
 529
 530 static inline int median4(int a, int b, int c, int d)
 531 {
 532     if (a < b) {
 533         if (c < d) return (FFMIN(b, d) + FFMAX(a, c)) / 2;
 534         else       return (FFMIN(b, c) + FFMAX(a, d)) / 2;
 535     } else {
 536         if (c < d) return (FFMIN(a, d) + FFMAX(b, c)) / 2;
 537         else       return (FFMIN(a, c) + FFMAX(b, d)) / 2;
 538     }
 539 }
 540
 541 /** Do motion compensation for 4-MV macroblock - luminance block
 542  */
 543 static void vc1_mc_4mv_luma(VC1Context *v, int n, int dir)
 544 {
 545     MpegEncContext *s = &v->s;
 546     DSPContext *dsp = &v->s.dsp;
 547     uint8_t *srcY;
 548     int dxy, mx, my, src_x, src_y;
 549     int off;
 550     int fieldmv = (v->fcm == ILACE_FRAME) ? v->blk_mv_type[s->block_index[n]] : 0;
 551     int v_edge_pos = s->v_edge_pos >> v->field_mode;
 552
 553     if ((!v->field_mode ||
 554          (v->ref_field_type[dir] == 1 && v->cur_field_type == 1)) &&
 555         !v->s.last_picture.f.data[0])
 556         return;
 557
 558     mx = s->mv[dir][n][0];
 559     my = s->mv[dir][n][1];
 560
 561     if (!dir) {
 562         if (v->field_mode) {
 563             if ((v->cur_field_type != v->ref_field_type[dir]) && v->cur_field_type)
 564                 srcY = s->current_picture.f.data[0];
 565             else
 566                 srcY = s->last_picture.f.data[0];
 567         } else
 568             srcY = s->last_picture.f.data[0];
 569     } else
 570         srcY = s->next_picture.f.data[0];
 571
 572     if (v->field_mode) {
 573         if (v->cur_field_type != v->ref_field_type[dir])
 574             my = my - 2 + 4 * v->cur_field_type;
 575     }
 576
 577     if (s->pict_type == AV_PICTURE_TYPE_P && n == 3 && v->field_mode) {
 578         int same_count = 0, opp_count = 0, k;
 579         int chosen_mv[2][4][2], f;
 580         int tx, ty;
 581         for (k = 0; k < 4; k++) {
 582             f = v->mv_f[0][s->block_index[k] + v->blocks_off];
 583             chosen_mv[f][f ? opp_count : same_count][0] = s->mv[0][k][0];
 584             chosen_mv[f][f ? opp_count : same_count][1] = s->mv[0][k][1];
 585             opp_count  += f;
 586             same_count += 1 - f;
 587         }
 588         f = opp_count > same_count;
 589         switch (f ? opp_count : same_count) {
 590         case 4:
 591             tx = median4(chosen_mv[f][0][0], chosen_mv[f][1][0],
 592                          chosen_mv[f][2][0], chosen_mv[f][3][0]);
 593             ty = median4(chosen_mv[f][0][1], chosen_mv[f][1][1],
 594                          chosen_mv[f][2][1], chosen_mv[f][3][1]);
 595             break;
 596         case 3:
 597             tx = mid_pred(chosen_mv[f][0][0], chosen_mv[f][1][0], chosen_mv[f][2][0]);
 598             ty = mid_pred(chosen_mv[f][0][1], chosen_mv[f][1][1], chosen_mv[f][2][1]);
 599             break;
 600         case 2:
 601             tx = (chosen_mv[f][0][0] + chosen_mv[f][1][0]) / 2;
 602             ty = (chosen_mv[f][0][1] + chosen_mv[f][1][1]) / 2;
 603             break;
 604         }
 605         s->current_picture.f.motion_val[1][s->block_index[0] + v->blocks_off][0] = tx;
 606         s->current_picture.f.motion_val[1][s->block_index[0] + v->blocks_off][1] = ty;
 607         for (k = 0; k < 4; k++)
 608             v->mv_f[1][s->block_index[k] + v->blocks_off] = f;
 609     }
 610
 611     if (v->fcm == ILACE_FRAME) {  // not sure if needed for other types of picture
 612         int qx, qy;
 613         int width  = s->avctx->coded_width;
 614         int height = s->avctx->coded_height >> 1;
 615         qx = (s->mb_x * 16) + (mx >> 2);
 616         qy = (s->mb_y *  8) + (my >> 3);
 617
 618         if (qx < -17)
 619             mx -= 4 * (qx + 17);
 620         else if (qx > width)
 621             mx -= 4 * (qx - width);
 622         if (qy < -18)
 623             my -= 8 * (qy + 18);
 624         else if (qy > height + 1)
 625             my -= 8 * (qy - height - 1);
 626     }
 627
 628     if ((v->fcm == ILACE_FRAME) && fieldmv)
 629         off = ((n > 1) ? s->linesize : 0) + (n & 1) * 8;
 630     else
 631         off = s->linesize * 4 * (n & 2) + (n & 1) * 8;
 632     if (v->field_mode && v->cur_field_type)
 633         off += s->current_picture_ptr->f.linesize[0];
 634
 635     src_x = s->mb_x * 16 + (n & 1) * 8 + (mx >> 2);
 636     if (!fieldmv)
 637         src_y = s->mb_y * 16 + (n & 2) * 4 + (my >> 2);
 638     else
 639         src_y = s->mb_y * 16 + ((n > 1) ? 1 : 0) + (my >> 2);
 640
 641     if (v->profile != PROFILE_ADVANCED) {
 642         src_x = av_clip(src_x, -16, s->mb_width  * 16);
 643         src_y = av_clip(src_y, -16, s->mb_height * 16);
 644     } else {
 645         src_x = av_clip(src_x, -17, s->avctx->coded_width);
 646         if (v->fcm == ILACE_FRAME) {
 647             if (src_y & 1)
 648                 src_y = av_clip(src_y, -17, s->avctx->coded_height + 1);
 649             else
 650                 src_y = av_clip(src_y, -18, s->avctx->coded_height);
 651         } else {
 652             src_y = av_clip(src_y, -18, s->avctx->coded_height + 1);
 653         }
 654     }
 655
 656     srcY += src_y * s->linesize + src_x;
 657     if (v->field_mode && v->ref_field_type[dir])
 658         srcY += s->current_picture_ptr->f.linesize[0];
 659
 660     if (fieldmv && !(src_y & 1))
 661         v_edge_pos--;
 662     if (fieldmv && (src_y & 1) && src_y < 4)
 663         src_y--;
 664     if (v->rangeredfrm || (v->mv_mode == MV_PMODE_INTENSITY_COMP)
 665         || s->h_edge_pos < 13 || v_edge_pos < 23
 666         || (unsigned)(src_x - s->mspel) > s->h_edge_pos - (mx & 3) - 8 - s->mspel * 2
 667         || (unsigned)(src_y - (s->mspel << fieldmv)) > v_edge_pos - (my & 3) - ((8 + s->mspel * 2) << fieldmv)) {
 668         srcY -= s->mspel * (1 + (s->linesize << fieldmv));
 669         /* check emulate edge stride and offset */
 670         s->dsp.emulated_edge_mc(s->edge_emu_buffer, srcY, s->linesize,
 671                                 9 + s->mspel * 2, (9 + s->mspel * 2) << fieldmv,
 672                                 src_x - s->mspel, src_y - (s->mspel << fieldmv),
 673                                 s->h_edge_pos, v_edge_pos);
 674         srcY = s->edge_emu_buffer;
 675         /* if we deal with range reduction we need to scale source blocks */
 676         if (v->rangeredfrm) {
 677             int i, j;
 678             uint8_t *src;
 679
 680             src = srcY;
 681             for (j = 0; j < 9 + s->mspel * 2; j++) {
 682                 for (i = 0; i < 9 + s->mspel * 2; i++)
 683                     src[i] = ((src[i] - 128) >> 1) + 128;
 684                 src += s->linesize << fieldmv;
 685             }
 686         }
 687         /* if we deal with intensity compensation we need to scale source blocks */
 688         if (v->mv_mode == MV_PMODE_INTENSITY_COMP) {
 689             int i, j;
 690             uint8_t *src;
 691
 692             src = srcY;
 693             for (j = 0; j < 9 + s->mspel * 2; j++) {
 694                 for (i = 0; i < 9 + s->mspel * 2; i++)
 695                     src[i] = v->luty[src[i]];
 696                 src += s->linesize << fieldmv;
 697             }
 698         }
 699         srcY += s->mspel * (1 + (s->linesize << fieldmv));
 700     }
 701
 702     if (s->mspel) {
 703         dxy = ((my & 3) << 2) | (mx & 3);
 704         v->vc1dsp.put_vc1_mspel_pixels_tab[dxy](s->dest[0] + off, srcY, s->linesize << fieldmv, v->rnd);
 705     } else { // hpel mc - always used for luma
 706         dxy = (my & 2) | ((mx & 2) >> 1);
 707         if (!v->rnd)
 708             dsp->put_pixels_tab[1][dxy](s->dest[0] + off, srcY, s->linesize, 8);
 709         else
 710             dsp->put_no_rnd_pixels_tab[1][dxy](s->dest[0] + off, srcY, s->linesize, 8);
 711     }
 712 }
 713
 714 static av_always_inline int get_chroma_mv(int *mvx, int *mvy, int *a, int flag, int *tx, int *ty)
 715 {
 716     int idx, i;
 717     static const int count[16] = { 0, 1, 1, 2, 1, 2, 2, 3, 1, 2, 2, 3, 2, 3, 3, 4};
 718
 719     idx =  ((a[3] != flag) << 3)
 720          | ((a[2] != flag) << 2)
 721          | ((a[1] != flag) << 1)
 722          |  (a[0] != flag);
 723     if (!idx) {
 724         *tx = median4(mvx[0], mvx[1], mvx[2], mvx[3]);
 725         *ty = median4(mvy[0], mvy[1], mvy[2], mvy[3]);
 726         return 4;
 727     } else if (count[idx] == 1) {
 728         switch (idx) {
 729         case 0x1:
 730             *tx = mid_pred(mvx[1], mvx[2], mvx[3]);
 731             *ty = mid_pred(mvy[1], mvy[2], mvy[3]);
 732             return 3;
 733         case 0x2:
 734             *tx = mid_pred(mvx[0], mvx[2], mvx[3]);
 735             *ty = mid_pred(mvy[0], mvy[2], mvy[3]);
 736             return 3;
 737         case 0x4:
 738             *tx = mid_pred(mvx[0], mvx[1], mvx[3]);
 739             *ty = mid_pred(mvy[0], mvy[1], mvy[3]);
 740             return 3;
 741         case 0x8:
 742             *tx = mid_pred(mvx[0], mvx[1], mvx[2]);
 743             *ty = mid_pred(mvy[0], mvy[1], mvy[2]);
 744             return 3;
 745         }
 746     } else if (count[idx] == 2) {
 747         int t1 = 0, t2 = 0;
 748         for (i = 0; i < 3; i++)
 749             if (!a[i]) {
 750                 t1 = i;
 751                 break;
 752             }
 753         for (i = t1 + 1; i < 4; i++)
 754             if (!a[i]) {
 755                 t2 = i;
 756                 break;
 757             }
 758         *tx = (mvx[t1] + mvx[t2]) / 2;
 759         *ty = (mvy[t1] + mvy[t2]) / 2;
 760         return 2;
 761     } else {
 762         return 0;
 763     }
 764     return -1;
 765 }
 766
 767 /** Do motion compensation for 4-MV macroblock - both chroma blocks
 768  */
 769 static void vc1_mc_4mv_chroma(VC1Context *v, int dir)
 770 {
 771     MpegEncContext *s = &v->s;
 772     DSPContext *dsp   = &v->s.dsp;
 773     uint8_t *srcU, *srcV;
 774     int uvmx, uvmy, uvsrc_x, uvsrc_y;
 775     int k, tx = 0, ty = 0;
 776     int mvx[4], mvy[4], intra[4], mv_f[4];
 777     int valid_count;
 778     int chroma_ref_type = v->cur_field_type, off = 0;
 779     int v_edge_pos = s->v_edge_pos >> v->field_mode;
 780
 781     if (!v->field_mode && !v->s.last_picture.f.data[0])
 782         return;
 783     if (s->flags & CODEC_FLAG_GRAY)
 784         return;
 785
 786     for (k = 0; k < 4; k++) {
 787         mvx[k] = s->mv[dir][k][0];
 788         mvy[k] = s->mv[dir][k][1];
 789         intra[k] = v->mb_type[0][s->block_index[k]];
 790         if (v->field_mode)
 791             mv_f[k] = v->mv_f[dir][s->block_index[k] + v->blocks_off];
 792     }
 793
 794     /* calculate chroma MV vector from four luma MVs */
 795     if (!v->field_mode || (v->field_mode && !v->numref)) {
 796         valid_count = get_chroma_mv(mvx, mvy, intra, 0, &tx, &ty);
 797         chroma_ref_type = v->reffield;
 798         if (!valid_count) {
 799             s->current_picture.f.motion_val[1][s->block_index[0] + v->blocks_off][0] = 0;
 800             s->current_picture.f.motion_val[1][s->block_index[0] + v->blocks_off][1] = 0;
 801             v->luma_mv[s->mb_x][0] = v->luma_mv[s->mb_x][1] = 0;
 802             return; //no need to do MC for intra blocks
 803         }
 804     } else {
 805         int dominant = 0;
 806         if (mv_f[0] + mv_f[1] + mv_f[2] + mv_f[3] > 2)
 807             dominant = 1;
 808         valid_count = get_chroma_mv(mvx, mvy, mv_f, dominant, &tx, &ty);
 809         if (dominant)
 810             chroma_ref_type = !v->cur_field_type;
 811     }
 812     if (v->field_mode && chroma_ref_type == 1 && v->cur_field_type == 1 && !v->s.last_picture.f.data[0])
 813         return;
 814     s->current_picture.f.motion_val[1][s->block_index[0] + v->blocks_off][0] = tx;
 815     s->current_picture.f.motion_val[1][s->block_index[0] + v->blocks_off][1] = ty;
 816     uvmx = (tx + ((tx & 3) == 3)) >> 1;
 817     uvmy = (ty + ((ty & 3) == 3)) >> 1;
 818
 819     v->luma_mv[s->mb_x][0] = uvmx;
 820     v->luma_mv[s->mb_x][1] = uvmy;
 821
 822     if (v->fastuvmc) {
 823         uvmx = uvmx + ((uvmx < 0) ? (uvmx & 1) : -(uvmx & 1));
 824         uvmy = uvmy + ((uvmy < 0) ? (uvmy & 1) : -(uvmy & 1));
 825     }
 826     // Field conversion bias
 827     if (v->cur_field_type != chroma_ref_type)
 828         uvmy += 2 - 4 * chroma_ref_type;
 829
 830     uvsrc_x = s->mb_x * 8 + (uvmx >> 2);
 831     uvsrc_y = s->mb_y * 8 + (uvmy >> 2);
 832
 833     if (v->profile != PROFILE_ADVANCED) {
 834         uvsrc_x = av_clip(uvsrc_x, -8, s->mb_width  * 8);
 835         uvsrc_y = av_clip(uvsrc_y, -8, s->mb_height * 8);
 836     } else {
 837         uvsrc_x = av_clip(uvsrc_x, -8, s->avctx->coded_width  >> 1);
 838         uvsrc_y = av_clip(uvsrc_y, -8, s->avctx->coded_height >> 1);
 839     }
 840
 841     if (!dir) {
 842         if (v->field_mode) {
 843             if ((v->cur_field_type != chroma_ref_type) && v->cur_field_type) {
 844                 srcU = s->current_picture.f.data[1] + uvsrc_y * s->uvlinesize + uvsrc_x;
 845                 srcV = s->current_picture.f.data[2] + uvsrc_y * s->uvlinesize + uvsrc_x;
 846             } else {
 847                 srcU = s->last_picture.f.data[1] + uvsrc_y * s->uvlinesize + uvsrc_x;
 848                 srcV = s->last_picture.f.data[2] + uvsrc_y * s->uvlinesize + uvsrc_x;
 849             }
 850         } else {
 851             srcU = s->last_picture.f.data[1] + uvsrc_y * s->uvlinesize + uvsrc_x;
 852             srcV = s->last_picture.f.data[2] + uvsrc_y * s->uvlinesize + uvsrc_x;
 853         }
 854     } else {
 855         srcU = s->next_picture.f.data[1] + uvsrc_y * s->uvlinesize + uvsrc_x;
 856         srcV = s->next_picture.f.data[2] + uvsrc_y * s->uvlinesize + uvsrc_x;
 857     }
 858
 859     if (v->field_mode) {
 860         if (chroma_ref_type) {
 861             srcU += s->current_picture_ptr->f.linesize[1];
 862             srcV += s->current_picture_ptr->f.linesize[2];
 863         }
 864         off = v->cur_field_type ? s->current_picture_ptr->f.linesize[1] : 0;
 865     }
 866
 867     if (v->rangeredfrm || (v->mv_mode == MV_PMODE_INTENSITY_COMP)
 868         || s->h_edge_pos < 18 || v_edge_pos < 18
 869         || (unsigned)uvsrc_x > (s->h_edge_pos >> 1) - 9
 870         || (unsigned)uvsrc_y > (v_edge_pos    >> 1) - 9) {
 871         s->dsp.emulated_edge_mc(s->edge_emu_buffer     , srcU, s->uvlinesize,
 872                                 8 + 1, 8 + 1, uvsrc_x, uvsrc_y,
 873                                 s->h_edge_pos >> 1, v_edge_pos >> 1);
 874         s->dsp.emulated_edge_mc(s->edge_emu_buffer + 16, srcV, s->uvlinesize,
 875                                 8 + 1, 8 + 1, uvsrc_x, uvsrc_y,
 876                                 s->h_edge_pos >> 1, v_edge_pos >> 1);
 877         srcU = s->edge_emu_buffer;
 878         srcV = s->edge_emu_buffer + 16;
 879
 880         /* if we deal with range reduction we need to scale source blocks */
 881         if (v->rangeredfrm) {
 882             int i, j;
 883             uint8_t *src, *src2;
 884
 885             src  = srcU;
 886             src2 = srcV;
 887             for (j = 0; j < 9; j++) {
 888                 for (i = 0; i < 9; i++) {
 889                     src[i]  = ((src[i]  - 128) >> 1) + 128;
 890                     src2[i] = ((src2[i] - 128) >> 1) + 128;
 891                 }
 892                 src  += s->uvlinesize;
 893                 src2 += s->uvlinesize;
 894             }
 895         }
 896         /* if we deal with intensity compensation we need to scale source blocks */
 897         if (v->mv_mode == MV_PMODE_INTENSITY_COMP) {
 898             int i, j;
 899             uint8_t *src, *src2;
 900
 901             src  = srcU;
 902             src2 = srcV;
 903             for (j = 0; j < 9; j++) {
 904                 for (i = 0; i < 9; i++) {
 905                     src[i]  = v->lutuv[src[i]];
 906                     src2[i] = v->lutuv[src2[i]];
 907                 }
 908                 src  += s->uvlinesize;
 909                 src2 += s->uvlinesize;
 910             }
 911         }
 912     }
 913
 914     /* Chroma MC always uses qpel bilinear */
 915     uvmx = (uvmx & 3) << 1;
 916     uvmy = (uvmy & 3) << 1;
 917     if (!v->rnd) {
 918         dsp->put_h264_chroma_pixels_tab[0](s->dest[1] + off, srcU, s->uvlinesize, 8, uvmx, uvmy);
 919         dsp->put_h264_chroma_pixels_tab[0](s->dest[2] + off, srcV, s->uvlinesize, 8, uvmx, uvmy);
 920     } else {
 921         v->vc1dsp.put_no_rnd_vc1_chroma_pixels_tab[0](s->dest[1] + off, srcU, s->uvlinesize, 8, uvmx, uvmy);
 922         v->vc1dsp.put_no_rnd_vc1_chroma_pixels_tab[0](s->dest[2] + off, srcV, s->uvlinesize, 8, uvmx, uvmy);
 923     }
 924 }
 925
 926 /** Do motion compensation for 4-MV field chroma macroblock (both U and V)
 927  */
 928 static void vc1_mc_4mv_chroma4(VC1Context *v)
 929 {
 930     MpegEncContext *s = &v->s;
 931     DSPContext *dsp = &v->s.dsp;
 932     uint8_t *srcU, *srcV;
 933     int uvsrc_x, uvsrc_y;
 934     int uvmx_field[4], uvmy_field[4];
 935     int i, off, tx, ty;
 936     int fieldmv = v->blk_mv_type[s->block_index[0]];
 937     static const int s_rndtblfield[16] = { 0, 0, 1, 2, 4, 4, 5, 6, 2, 2, 3, 8, 6, 6, 7, 12 };
 938     int v_dist = fieldmv ? 1 : 4; // vertical offset for lower sub-blocks
 939     int v_edge_pos = s->v_edge_pos >> 1;
 940
 941     if (!v->s.last_picture.f.data[0])
 942         return;
 943     if (s->flags & CODEC_FLAG_GRAY)
 944         return;
 945
 946     for (i = 0; i < 4; i++) {
 947         tx = s->mv[0][i][0];
 948         uvmx_field[i] = (tx + ((tx & 3) == 3)) >> 1;
 949         ty = s->mv[0][i][1];
 950         if (fieldmv)
 951             uvmy_field[i] = (ty >> 4) * 8 + s_rndtblfield[ty & 0xF];
 952         else
 953             uvmy_field[i] = (ty + ((ty & 3) == 3)) >> 1;
 954     }
 955
 956     for (i = 0; i < 4; i++) {
 957         off = (i & 1) * 4 + ((i & 2) ? v_dist * s->uvlinesize : 0);
 958         uvsrc_x = s->mb_x * 8 +  (i & 1) * 4           + (uvmx_field[i] >> 2);
 959         uvsrc_y = s->mb_y * 8 + ((i & 2) ? v_dist : 0) + (uvmy_field[i] >> 2);
 960         // FIXME: implement proper pull-back (see vc1cropmv.c, vc1CROPMV_ChromaPullBack())
 961         uvsrc_x = av_clip(uvsrc_x, -8, s->avctx->coded_width  >> 1);
 962         uvsrc_y = av_clip(uvsrc_y, -8, s->avctx->coded_height >> 1);
 963         srcU = s->last_picture.f.data[1] + uvsrc_y * s->uvlinesize + uvsrc_x;
 964         srcV = s->last_picture.f.data[2] + uvsrc_y * s->uvlinesize + uvsrc_x;
 965         uvmx_field[i] = (uvmx_field[i] & 3) << 1;
 966         uvmy_field[i] = (uvmy_field[i] & 3) << 1;
 967
 968         if (fieldmv && !(uvsrc_y & 1))
 969             v_edge_pos--;
 970         if (fieldmv && (uvsrc_y & 1) && uvsrc_y < 2)
 971             uvsrc_y--;
 972         if ((v->mv_mode == MV_PMODE_INTENSITY_COMP)
 973             || s->h_edge_pos < 10 || v_edge_pos < (5 << fieldmv)
 974             || (unsigned)uvsrc_x > (s->h_edge_pos >> 1) - 5
 975             || (unsigned)uvsrc_y > v_edge_pos - (5 << fieldmv)) {
 976             s->dsp.emulated_edge_mc(s->edge_emu_buffer, srcU, s->uvlinesize,
 977                                     5, (5 << fieldmv), uvsrc_x, uvsrc_y,
 978                                     s->h_edge_pos >> 1, v_edge_pos);
 979             s->dsp.emulated_edge_mc(s->edge_emu_buffer + 16, srcV, s->uvlinesize,
 980                                     5, (5 << fieldmv), uvsrc_x, uvsrc_y,
 981                                     s->h_edge_pos >> 1, v_edge_pos);
 982             srcU = s->edge_emu_buffer;
 983             srcV = s->edge_emu_buffer + 16;
 984
 985             /* if we deal with intensity compensation we need to scale source blocks */
 986             if (v->mv_mode == MV_PMODE_INTENSITY_COMP) {
 987                 int i, j;
 988                 uint8_t *src, *src2;
 989
 990                 src  = srcU;
 991                 src2 = srcV;
 992                 for (j = 0; j < 5; j++) {
 993                     for (i = 0; i < 5; i++) {
 994                         src[i]  = v->lutuv[src[i]];
 995                         src2[i] = v->lutuv[src2[i]];
 996                     }
 997                     src  += s->uvlinesize << 1;
 998                     src2 += s->uvlinesize << 1;
 999                 }
1000             }
1001         }
1002         if (!v->rnd) {
1003             dsp->put_h264_chroma_pixels_tab[1](s->dest[1] + off, srcU, s->uvlinesize << fieldmv, 4, uvmx_field[i], uvmy_field[i]);
1004             dsp->put_h264_chroma_pixels_tab[1](s->dest[2] + off, srcV, s->uvlinesize << fieldmv, 4, uvmx_field[i], uvmy_field[i]);
1005         } else {
1006             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]);
1007             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]);
1008         }
1009     }
1010 }
1011
1012 /***********************************************************************/
1013 /**
1014  * @name VC-1 Block-level functions
1015  * @see 7.1.4, p91 and 8.1.1.7, p(1)04
1016  * @{
1017  */
1018
1019 /**
1020  * @def GET_MQUANT
1021  * @brief Get macroblock-level quantizer scale
1022  */
1023 #define GET_MQUANT()                                           \
1024     if (v->dquantfrm) {                                        \
1025         int edges = 0;                                         \
1026         if (v->dqprofile == DQPROFILE_ALL_MBS) {               \
1027             if (v->dqbilevel) {                                \
1028                 mquant = (get_bits1(gb)) ? v->altpq : v->pq;   \
1029             } else {                                           \
1030                 mqdiff = get_bits(gb, 3);                      \
1031                 if (mqdiff != 7)                               \
1032                     mquant = v->pq + mqdiff;                   \
1033                 else                                           \
1034                     mquant = get_bits(gb, 5);                  \
1035             }                                                  \
1036         }                                                      \
1037         if (v->dqprofile == DQPROFILE_SINGLE_EDGE)             \
1038             edges = 1 << v->dqsbedge;                          \
1039         else if (v->dqprofile == DQPROFILE_DOUBLE_EDGES)       \
1040             edges = (3 << v->dqsbedge) % 15;                   \
1041         else if (v->dqprofile == DQPROFILE_FOUR_EDGES)         \
1042             edges = 15;                                        \
1043         if ((edges&1) && !s->mb_x)                             \
1044             mquant = v->altpq;                                 \
1045         if ((edges&2) && s->first_slice_line)                  \
1046             mquant = v->altpq;                                 \
1047         if ((edges&4) && s->mb_x == (s->mb_width - 1))         \
1048             mquant = v->altpq;                                 \
1049         if ((edges&8) && s->mb_y == (s->mb_height - 1))        \
1050             mquant = v->altpq;                                 \
1051         if (!mquant || mquant > 31) {                          \
1052             av_log(v->s.avctx, AV_LOG_ERROR,                   \
1053                    "Overriding invalid mquant %d\n", mquant);  \
1054             mquant = 1;                                        \
1055         }                                                      \
1056     }
1057
1058 /**
1059  * @def GET_MVDATA(_dmv_x, _dmv_y)
1060  * @brief Get MV differentials
1061  * @see MVDATA decoding from 8.3.5.2, p(1)20
1062  * @param _dmv_x Horizontal differential for decoded MV
1063  * @param _dmv_y Vertical differential for decoded MV
1064  */
1065 #define GET_MVDATA(_dmv_x, _dmv_y)                                      \
1066     index = 1 + get_vlc2(gb, ff_vc1_mv_diff_vlc[s->mv_table_index].table, \
1067                          VC1_MV_DIFF_VLC_BITS, 2);                      \
1068     if (index > 36) {                                                   \
1069         mb_has_coeffs = 1;                                              \
1070         index -= 37;                                                    \
1071     } else                                                              \
1072         mb_has_coeffs = 0;                                              \
1073     s->mb_intra = 0;                                                    \
1074     if (!index) {                                                       \
1075         _dmv_x = _dmv_y = 0;                                            \
1076     } else if (index == 35) {                                           \
1077         _dmv_x = get_bits(gb, v->k_x - 1 + s->quarter_sample);          \
1078         _dmv_y = get_bits(gb, v->k_y - 1 + s->quarter_sample);          \
1079     } else if (index == 36) {                                           \
1080         _dmv_x = 0;                                                     \
1081         _dmv_y = 0;                                                     \
1082         s->mb_intra = 1;                                                \
1083     } else {                                                            \
1084         index1 = index % 6;                                             \
1085         if (!s->quarter_sample && index1 == 5) val = 1;                 \
1086         else                                   val = 0;                 \
1087         if (size_table[index1] - val > 0)                               \
1088             val = get_bits(gb, size_table[index1] - val);               \
1089         else                                   val = 0;                 \
1090         sign = 0 - (val&1);                                             \
1091         _dmv_x = (sign ^ ((val>>1) + offset_table[index1])) - sign;     \
1092                                                                         \
1093         index1 = index / 6;                                             \
1094         if (!s->quarter_sample && index1 == 5) val = 1;                 \
1095         else                                   val = 0;                 \
1096         if (size_table[index1] - val > 0)                               \
1097             val = get_bits(gb, size_table[index1] - val);               \
1098         else                                   val = 0;                 \
1099         sign = 0 - (val & 1);                                           \
1100         _dmv_y = (sign ^ ((val >> 1) + offset_table[index1])) - sign;   \
1101     }
1102
1103 static av_always_inline void get_mvdata_interlaced(VC1Context *v, int *dmv_x,
1104                                                    int *dmv_y, int *pred_flag)
1105 {
1106     int index, index1;
1107     int extend_x = 0, extend_y = 0;
1108     GetBitContext *gb = &v->s.gb;
1109     int bits, esc;
1110     int val, sign;
1111     const int* offs_tab;
1112
1113     if (v->numref) {
1114         bits = VC1_2REF_MVDATA_VLC_BITS;
1115         esc  = 125;
1116     } else {
1117         bits = VC1_1REF_MVDATA_VLC_BITS;
1118         esc  = 71;
1119     }
1120     switch (v->dmvrange) {
1121     case 1:
1122         extend_x = 1;
1123         break;
1124     case 2:
1125         extend_y = 1;
1126         break;
1127     case 3:
1128         extend_x = extend_y = 1;
1129         break;
1130     }
1131     index = get_vlc2(gb, v->imv_vlc->table, bits, 3);
1132     if (index == esc) {
1133         *dmv_x = get_bits(gb, v->k_x);
1134         *dmv_y = get_bits(gb, v->k_y);
1135         if (v->numref) {
1136             *pred_flag = *dmv_y & 1;
1137             *dmv_y     = (*dmv_y + *pred_flag) >> 1;
1138         }
1139     }
1140     else {
1141         if (extend_x)
1142             offs_tab = offset_table2;
1143         else
1144             offs_tab = offset_table1;
1145         index1 = (index + 1) % 9;
1146         if (index1 != 0) {
1147             val    = get_bits(gb, index1 + extend_x);
1148             sign   = 0 -(val & 1);
1149             *dmv_x = (sign ^ ((val >> 1) + offs_tab[index1])) - sign;
1150         } else
1151             *dmv_x = 0;
1152         if (extend_y)
1153             offs_tab = offset_table2;
1154         else
1155             offs_tab = offset_table1;
1156         index1 = (index + 1) / 9;
1157         if (index1 > v->numref) {
1158             val    = get_bits(gb, (index1 + (extend_y << v->numref)) >> v->numref);
1159             sign   = 0 - (val & 1);
1160             *dmv_y = (sign ^ ((val >> 1) + offs_tab[index1 >> v->numref])) - sign;
1161         } else
1162             *dmv_y = 0;
1163         if (v->numref)
1164             *pred_flag = index1 & 1;
1165     }
1166 }
1167
1168 static av_always_inline int scaleforsame_x(VC1Context *v, int n /* MV */, int dir)
1169 {
1170     int scaledvalue, refdist;
1171     int scalesame1, scalesame2;
1172     int scalezone1_x, zone1offset_x;
1173     int table_index = dir ^ v->second_field;
1174
1175     if (v->s.pict_type != AV_PICTURE_TYPE_B)
1176         refdist = v->refdist;
1177     else
1178         refdist = dir ? v->brfd : v->frfd;
1179     if (refdist > 3)
1180         refdist = 3;
1181     scalesame1    = ff_vc1_field_mvpred_scales[table_index][1][refdist];
1182     scalesame2    = ff_vc1_field_mvpred_scales[table_index][2][refdist];
1183     scalezone1_x  = ff_vc1_field_mvpred_scales[table_index][3][refdist];
1184     zone1offset_x = ff_vc1_field_mvpred_scales[table_index][5][refdist];
1185
1186     if (FFABS(n) > 255)
1187         scaledvalue = n;
1188     else {
1189         if (FFABS(n) < scalezone1_x)
1190             scaledvalue = (n * scalesame1) >> 8;
1191         else {
1192             if (n < 0)
1193                 scaledvalue = ((n * scalesame2) >> 8) - zone1offset_x;
1194             else
1195                 scaledvalue = ((n * scalesame2) >> 8) + zone1offset_x;
1196         }
1197     }
1198     return av_clip(scaledvalue, -v->range_x, v->range_x - 1);
1199 }
1200
1201 static av_always_inline int scaleforsame_y(VC1Context *v, int i, int n /* MV */, int dir)
1202 {
1203     int scaledvalue, refdist;
1204     int scalesame1, scalesame2;
1205     int scalezone1_y, zone1offset_y;
1206     int table_index = dir ^ v->second_field;
1207
1208     if (v->s.pict_type != AV_PICTURE_TYPE_B)
1209         refdist = v->refdist;
1210     else
1211         refdist = dir ? v->brfd : v->frfd;
1212     if (refdist > 3)
1213         refdist = 3;
1214     scalesame1    = ff_vc1_field_mvpred_scales[table_index][1][refdist];
1215     scalesame2    = ff_vc1_field_mvpred_scales[table_index][2][refdist];
1216     scalezone1_y  = ff_vc1_field_mvpred_scales[table_index][4][refdist];
1217     zone1offset_y = ff_vc1_field_mvpred_scales[table_index][6][refdist];
1218
1219     if (FFABS(n) > 63)
1220         scaledvalue = n;
1221     else {
1222         if (FFABS(n) < scalezone1_y)
1223             scaledvalue = (n * scalesame1) >> 8;
1224         else {
1225             if (n < 0)
1226                 scaledvalue = ((n * scalesame2) >> 8) - zone1offset_y;
1227             else
1228                 scaledvalue = ((n * scalesame2) >> 8) + zone1offset_y;
1229         }
1230     }
1231
1232     if (v->cur_field_type && !v->ref_field_type[dir])
1233         return av_clip(scaledvalue, -v->range_y / 2 + 1, v->range_y / 2);
1234     else
1235         return av_clip(scaledvalue, -v->range_y / 2, v->range_y / 2 - 1);
1236 }
1237
1238 static av_always_inline int scaleforopp_x(VC1Context *v, int n /* MV */)
1239 {
1240     int scalezone1_x, zone1offset_x;
1241     int scaleopp1, scaleopp2, brfd;
1242     int scaledvalue;
1243
1244     brfd = FFMIN(v->brfd, 3);
1245     scalezone1_x  = ff_vc1_b_field_mvpred_scales[3][brfd];
1246     zone1offset_x = ff_vc1_b_field_mvpred_scales[5][brfd];
1247     scaleopp1     = ff_vc1_b_field_mvpred_scales[1][brfd];
1248     scaleopp2     = ff_vc1_b_field_mvpred_scales[2][brfd];
1249
1250     if (FFABS(n) > 255)
1251         scaledvalue = n;
1252     else {
1253         if (FFABS(n) < scalezone1_x)
1254             scaledvalue = (n * scaleopp1) >> 8;
1255         else {
1256             if (n < 0)
1257                 scaledvalue = ((n * scaleopp2) >> 8) - zone1offset_x;
1258             else
1259                 scaledvalue = ((n * scaleopp2) >> 8) + zone1offset_x;
1260         }
1261     }
1262     return av_clip(scaledvalue, -v->range_x, v->range_x - 1);
1263 }
1264
1265 static av_always_inline int scaleforopp_y(VC1Context *v, int n /* MV */, int dir)
1266 {
1267     int scalezone1_y, zone1offset_y;
1268     int scaleopp1, scaleopp2, brfd;
1269     int scaledvalue;
1270
1271     brfd = FFMIN(v->brfd, 3);
1272     scalezone1_y  = ff_vc1_b_field_mvpred_scales[4][brfd];
1273     zone1offset_y = ff_vc1_b_field_mvpred_scales[6][brfd];
1274     scaleopp1     = ff_vc1_b_field_mvpred_scales[1][brfd];
1275     scaleopp2     = ff_vc1_b_field_mvpred_scales[2][brfd];
1276
1277     if (FFABS(n) > 63)
1278         scaledvalue = n;
1279     else {
1280         if (FFABS(n) < scalezone1_y)
1281             scaledvalue = (n * scaleopp1) >> 8;
1282         else {
1283             if (n < 0)
1284                 scaledvalue = ((n * scaleopp2) >> 8) - zone1offset_y;
1285             else
1286                 scaledvalue = ((n * scaleopp2) >> 8) + zone1offset_y;
1287         }
1288     }
1289     if (v->cur_field_type && !v->ref_field_type[dir]) {
1290         return av_clip(scaledvalue, -v->range_y / 2 + 1, v->range_y / 2);
1291     } else {
1292         return av_clip(scaledvalue, -v->range_y / 2, v->range_y / 2 - 1);
1293     }
1294 }
1295
1296 static av_always_inline int scaleforsame(VC1Context *v, int i, int n /* MV */,
1297                                          int dim, int dir)
1298 {
1299     int brfd, scalesame;
1300     int hpel = 1 - v->s.quarter_sample;
1301
1302     n >>= hpel;
1303     if (v->s.pict_type != AV_PICTURE_TYPE_B || v->second_field || !dir) {
1304         if (dim)
1305             n = scaleforsame_y(v, i, n, dir) << hpel;
1306         else
1307             n = scaleforsame_x(v, n, dir) << hpel;
1308         return n;
1309     }
1310     brfd      = FFMIN(v->brfd, 3);
1311     scalesame = ff_vc1_b_field_mvpred_scales[0][brfd];
1312
1313     n = (n * scalesame >> 8) << hpel;
1314     return n;
1315 }
1316
1317 static av_always_inline int scaleforopp(VC1Context *v, int n /* MV */,
1318                                         int dim, int dir)
1319 {
1320     int refdist, scaleopp;
1321     int hpel = 1 - v->s.quarter_sample;
1322
1323     n >>= hpel;
1324     if (v->s.pict_type == AV_PICTURE_TYPE_B && !v->second_field && dir == 1) {
1325         if (dim)
1326             n = scaleforopp_y(v, n, dir) << hpel;
1327         else
1328             n = scaleforopp_x(v, n) << hpel;
1329         return n;
1330     }
1331     if (v->s.pict_type != AV_PICTURE_TYPE_B)
1332         refdist = FFMIN(v->refdist, 3);
1333     else
1334         refdist = dir ? v->brfd : v->frfd;
1335     scaleopp = ff_vc1_field_mvpred_scales[dir ^ v->second_field][0][refdist];
1336
1337     n = (n * scaleopp >> 8) << hpel;
1338     return n;
1339 }
1340
1341 /** Predict and set motion vector
1342  */
1343 static inline void vc1_pred_mv(VC1Context *v, int n, int dmv_x, int dmv_y,
1344                                int mv1, int r_x, int r_y, uint8_t* is_intra,
1345                                int pred_flag, int dir)
1346 {
1347     MpegEncContext *s = &v->s;
1348     int xy, wrap, off = 0;
1349     int16_t *A, *B, *C;
1350     int px, py;
1351     int sum;
1352     int mixedmv_pic, num_samefield = 0, num_oppfield = 0;
1353     int opposite, a_f, b_f, c_f;
1354     int16_t field_predA[2];
1355     int16_t field_predB[2];
1356     int16_t field_predC[2];
1357     int a_valid, b_valid, c_valid;
1358     int hybridmv_thresh, y_bias = 0;
1359
1360     if (v->mv_mode == MV_PMODE_MIXED_MV ||
1361         ((v->mv_mode == MV_PMODE_INTENSITY_COMP) && (v->mv_mode2 == MV_PMODE_MIXED_MV)))
1362         mixedmv_pic = 1;
1363     else
1364         mixedmv_pic = 0;
1365     /* scale MV difference to be quad-pel */
1366     dmv_x <<= 1 - s->quarter_sample;
1367     dmv_y <<= 1 - s->quarter_sample;
1368
1369     wrap = s->b8_stride;
1370     xy   = s->block_index[n];
1371
1372     if (s->mb_intra) {
1373         s->mv[0][n][0] = s->current_picture.f.motion_val[0][xy + v->blocks_off][0] = 0;
1374         s->mv[0][n][1] = s->current_picture.f.motion_val[0][xy + v->blocks_off][1] = 0;
1375         s->current_picture.f.motion_val[1][xy + v->blocks_off][0] = 0;
1376         s->current_picture.f.motion_val[1][xy + v->blocks_off][1] = 0;
1377         if (mv1) { /* duplicate motion data for 1-MV block */
1378             s->current_picture.f.motion_val[0][xy + 1 + v->blocks_off][0]        = 0;
1379             s->current_picture.f.motion_val[0][xy + 1 + v->blocks_off][1]        = 0;
1380             s->current_picture.f.motion_val[0][xy + wrap + v->blocks_off][0]     = 0;
1381             s->current_picture.f.motion_val[0][xy + wrap + v->blocks_off][1]     = 0;
1382             s->current_picture.f.motion_val[0][xy + wrap + 1 + v->blocks_off][0] = 0;
1383             s->current_picture.f.motion_val[0][xy + wrap + 1 + v->blocks_off][1] = 0;
1384             v->luma_mv[s->mb_x][0] = v->luma_mv[s->mb_x][1] = 0;
1385             s->current_picture.f.motion_val[1][xy + 1 + v->blocks_off][0]        = 0;
1386             s->current_picture.f.motion_val[1][xy + 1 + v->blocks_off][1]        = 0;
1387             s->current_picture.f.motion_val[1][xy + wrap][0]                     = 0;
1388             s->current_picture.f.motion_val[1][xy + wrap + v->blocks_off][1]     = 0;
1389             s->current_picture.f.motion_val[1][xy + wrap + 1 + v->blocks_off][0] = 0;
1390             s->current_picture.f.motion_val[1][xy + wrap + 1 + v->blocks_off][1] = 0;
1391         }
1392         return;
1393     }
1394
1395     C = s->current_picture.f.motion_val[dir][xy -    1 + v->blocks_off];
1396     A = s->current_picture.f.motion_val[dir][xy - wrap + v->blocks_off];
1397     if (mv1) {
1398         if (v->field_mode && mixedmv_pic)
1399             off = (s->mb_x == (s->mb_width - 1)) ? -2 : 2;
1400         else
1401             off = (s->mb_x == (s->mb_width - 1)) ? -1 : 2;
1402     } else {
1403         //in 4-MV mode different blocks have different B predictor position
1404         switch (n) {
1405         case 0:
1406             off = (s->mb_x > 0) ? -1 : 1;
1407             break;
1408         case 1:
1409             off = (s->mb_x == (s->mb_width - 1)) ? -1 : 1;
1410             break;
1411         case 2:
1412             off = 1;
1413             break;
1414         case 3:
1415             off = -1;
1416         }
1417     }
1418     B = s->current_picture.f.motion_val[dir][xy - wrap + off + v->blocks_off];
1419
1420     a_valid = !s->first_slice_line || (n == 2 || n == 3);
1421     b_valid = a_valid && (s->mb_width > 1);
1422     c_valid = s->mb_x || (n == 1 || n == 3);
1423     if (v->field_mode) {
1424         a_valid = a_valid && !is_intra[xy - wrap];
1425         b_valid = b_valid && !is_intra[xy - wrap + off];
1426         c_valid = c_valid && !is_intra[xy - 1];
1427     }
1428
1429     if (a_valid) {
1430         a_f = v->mv_f[dir][xy - wrap + v->blocks_off];
1431         num_oppfield  += a_f;
1432         num_samefield += 1 - a_f;
1433         field_predA[0] = A[0];
1434         field_predA[1] = A[1];
1435     } else {
1436         field_predA[0] = field_predA[1] = 0;
1437         a_f = 0;
1438     }
1439     if (b_valid) {
1440         b_f = v->mv_f[dir][xy - wrap + off + v->blocks_off];
1441         num_oppfield  += b_f;
1442         num_samefield += 1 - b_f;
1443         field_predB[0] = B[0];
1444         field_predB[1] = B[1];
1445     } else {
1446         field_predB[0] = field_predB[1] = 0;
1447         b_f = 0;
1448     }
1449     if (c_valid) {
1450         c_f = v->mv_f[dir][xy - 1 + v->blocks_off];
1451         num_oppfield  += c_f;
1452         num_samefield += 1 - c_f;
1453         field_predC[0] = C[0];
1454         field_predC[1] = C[1];
1455     } else {
1456         field_predC[0] = field_predC[1] = 0;
1457         c_f = 0;
1458     }
1459
1460     if (v->field_mode) {
1461         if (!v->numref)
1462             // REFFIELD determines if the last field or the second-last field is
1463             // to be used as reference
1464             opposite = 1 - v->reffield;
1465         else {
1466             if (num_samefield <= num_oppfield)
1467                 opposite = 1 - pred_flag;
1468             else
1469                 opposite = pred_flag;
1470         }
1471     } else
1472         opposite = 0;
1473     if (opposite) {
1474         if (a_valid && !a_f) {
1475             field_predA[0] = scaleforopp(v, field_predA[0], 0, dir);
1476             field_predA[1] = scaleforopp(v, field_predA[1], 1, dir);
1477         }
1478         if (b_valid && !b_f) {
1479             field_predB[0] = scaleforopp(v, field_predB[0], 0, dir);
1480             field_predB[1] = scaleforopp(v, field_predB[1], 1, dir);
1481         }
1482         if (c_valid && !c_f) {
1483             field_predC[0] = scaleforopp(v, field_predC[0], 0, dir);
1484             field_predC[1] = scaleforopp(v, field_predC[1], 1, dir);
1485         }
1486         v->mv_f[dir][xy + v->blocks_off] = 1;
1487         v->ref_field_type[dir] = !v->cur_field_type;
1488     } else {
1489         if (a_valid && a_f) {
1490             field_predA[0] = scaleforsame(v, n, field_predA[0], 0, dir);
1491             field_predA[1] = scaleforsame(v, n, field_predA[1], 1, dir);
1492         }
1493         if (b_valid && b_f) {
1494             field_predB[0] = scaleforsame(v, n, field_predB[0], 0, dir);
1495             field_predB[1] = scaleforsame(v, n, field_predB[1], 1, dir);
1496         }
1497         if (c_valid && c_f) {
1498             field_predC[0] = scaleforsame(v, n, field_predC[0], 0, dir);
1499             field_predC[1] = scaleforsame(v, n, field_predC[1], 1, dir);
1500         }
1501         v->mv_f[dir][xy + v->blocks_off] = 0;
1502         v->ref_field_type[dir] = v->cur_field_type;
1503     }
1504
1505     if (a_valid) {
1506         px = field_predA[0];
1507         py = field_predA[1];
1508     } else if (c_valid) {
1509         px = field_predC[0];
1510         py = field_predC[1];
1511     } else if (b_valid) {
1512         px = field_predB[0];
1513         py = field_predB[1];
1514     } else {
1515         px = 0;
1516         py = 0;
1517     }
1518
1519     if (num_samefield + num_oppfield > 1) {
1520         px = mid_pred(field_predA[0], field_predB[0], field_predC[0]);
1521         py = mid_pred(field_predA[1], field_predB[1], field_predC[1]);
1522     }
1523
1524     /* Pullback MV as specified in 8.3.5.3.4 */
1525     if (!v->field_mode) {
1526         int qx, qy, X, Y;
1527         qx = (s->mb_x << 6) + ((n == 1 || n == 3) ? 32 : 0);
1528         qy = (s->mb_y << 6) + ((n == 2 || n == 3) ? 32 : 0);
1529         X  = (s->mb_width  << 6) - 4;
1530         Y  = (s->mb_height << 6) - 4;
1531         if (mv1) {
1532             if (qx + px < -60) px = -60 - qx;
1533             if (qy + py < -60) py = -60 - qy;
1534         } else {
1535             if (qx + px < -28) px = -28 - qx;
1536             if (qy + py < -28) py = -28 - qy;
1537         }
1538         if (qx + px > X) px = X - qx;
1539         if (qy + py > Y) py = Y - qy;
1540     }
1541
1542     if (!v->field_mode || s->pict_type != AV_PICTURE_TYPE_B) {
1543         /* Calculate hybrid prediction as specified in 8.3.5.3.5 (also 10.3.5.4.3.5) */
1544         hybridmv_thresh = 32;
1545         if (a_valid && c_valid) {
1546             if (is_intra[xy - wrap])
1547                 sum = FFABS(px) + FFABS(py);
1548             else
1549                 sum = FFABS(px - field_predA[0]) + FFABS(py - field_predA[1]);
1550             if (sum > hybridmv_thresh) {
1551                 if (get_bits1(&s->gb)) {     // read HYBRIDPRED bit
1552                     px = field_predA[0];
1553                     py = field_predA[1];
1554                 } else {
1555                     px = field_predC[0];
1556                     py = field_predC[1];
1557                 }
1558             } else {
1559                 if (is_intra[xy - 1])
1560                     sum = FFABS(px) + FFABS(py);
1561                 else
1562                     sum = FFABS(px - field_predC[0]) + FFABS(py - field_predC[1]);
1563                 if (sum > hybridmv_thresh) {
1564                     if (get_bits1(&s->gb)) {
1565                         px = field_predA[0];
1566                         py = field_predA[1];
1567                     } else {
1568                         px = field_predC[0];
1569                         py = field_predC[1];
1570                     }
1571                 }
1572             }
1573         }
1574     }
1575
1576     if (v->field_mode && v->numref)
1577         r_y >>= 1;
1578     if (v->field_mode && v->cur_field_type && v->ref_field_type[dir] == 0)
1579         y_bias = 1;
1580     /* store MV using signed modulus of MV range defined in 4.11 */
1581     s->mv[dir][n][0] = s->current_picture.f.motion_val[dir][xy + v->blocks_off][0] = ((px + dmv_x + r_x) & ((r_x << 1) - 1)) - r_x;
1582     s->mv[dir][n][1] = s->current_picture.f.motion_val[dir][xy + v->blocks_off][1] = ((py + dmv_y + r_y - y_bias) & ((r_y << 1) - 1)) - r_y + y_bias;
1583     if (mv1) { /* duplicate motion data for 1-MV block */
1584         s->current_picture.f.motion_val[dir][xy +    1 +     v->blocks_off][0] = s->current_picture.f.motion_val[dir][xy + v->blocks_off][0];
1585         s->current_picture.f.motion_val[dir][xy +    1 +     v->blocks_off][1] = s->current_picture.f.motion_val[dir][xy + v->blocks_off][1];
1586         s->current_picture.f.motion_val[dir][xy + wrap +     v->blocks_off][0] = s->current_picture.f.motion_val[dir][xy + v->blocks_off][0];
1587         s->current_picture.f.motion_val[dir][xy + wrap +     v->blocks_off][1] = s->current_picture.f.motion_val[dir][xy + v->blocks_off][1];
1588         s->current_picture.f.motion_val[dir][xy + wrap + 1 + v->blocks_off][0] = s->current_picture.f.motion_val[dir][xy + v->blocks_off][0];
1589         s->current_picture.f.motion_val[dir][xy + wrap + 1 + v->blocks_off][1] = s->current_picture.f.motion_val[dir][xy + v->blocks_off][1];
1590         v->mv_f[dir][xy +    1 + v->blocks_off] = v->mv_f[dir][xy +            v->blocks_off];
1591         v->mv_f[dir][xy + wrap + v->blocks_off] = v->mv_f[dir][xy + wrap + 1 + v->blocks_off] = v->mv_f[dir][xy + v->blocks_off];
1592     }
1593 }
1594
1595 /** Predict and set motion vector for interlaced frame picture MBs
1596  */
1597 static inline void vc1_pred_mv_intfr(VC1Context *v, int n, int dmv_x, int dmv_y,
1598                                      int mvn, int r_x, int r_y, uint8_t* is_intra)
1599 {
1600     MpegEncContext *s = &v->s;
1601     int xy, wrap, off = 0;
1602     int A[2], B[2], C[2];
1603     int px, py;
1604     int a_valid = 0, b_valid = 0, c_valid = 0;
1605     int field_a, field_b, field_c; // 0: same, 1: opposit
1606     int total_valid, num_samefield, num_oppfield;
1607     int pos_c, pos_b, n_adj;
1608
1609     wrap = s->b8_stride;
1610     xy = s->block_index[n];
1611
1612     if (s->mb_intra) {
1613         s->mv[0][n][0] = s->current_picture.f.motion_val[0][xy][0] = 0;
1614         s->mv[0][n][1] = s->current_picture.f.motion_val[0][xy][1] = 0;
1615         s->current_picture.f.motion_val[1][xy][0] = 0;
1616         s->current_picture.f.motion_val[1][xy][1] = 0;
1617         if (mvn == 1) { /* duplicate motion data for 1-MV block */
1618             s->current_picture.f.motion_val[0][xy + 1][0]        = 0;
1619             s->current_picture.f.motion_val[0][xy + 1][1]        = 0;
1620             s->current_picture.f.motion_val[0][xy + wrap][0]     = 0;
1621             s->current_picture.f.motion_val[0][xy + wrap][1]     = 0;
1622             s->current_picture.f.motion_val[0][xy + wrap + 1][0] = 0;
1623             s->current_picture.f.motion_val[0][xy + wrap + 1][1] = 0;
1624             v->luma_mv[s->mb_x][0] = v->luma_mv[s->mb_x][1] = 0;
1625             s->current_picture.f.motion_val[1][xy + 1][0]        = 0;
1626             s->current_picture.f.motion_val[1][xy + 1][1]        = 0;
1627             s->current_picture.f.motion_val[1][xy + wrap][0]     = 0;
1628             s->current_picture.f.motion_val[1][xy + wrap][1]     = 0;
1629             s->current_picture.f.motion_val[1][xy + wrap + 1][0] = 0;
1630             s->current_picture.f.motion_val[1][xy + wrap + 1][1] = 0;
1631         }
1632         return;
1633     }
1634
1635     off = ((n == 0) || (n == 1)) ? 1 : -1;
1636     /* predict A */
1637     if (s->mb_x || (n == 1) || (n == 3)) {
1638         if ((v->blk_mv_type[xy]) // current block (MB) has a field MV
1639             || (!v->blk_mv_type[xy] && !v->blk_mv_type[xy - 1])) { // or both have frame MV
1640             A[0] = s->current_picture.f.motion_val[0][xy - 1][0];
1641             A[1] = s->current_picture.f.motion_val[0][xy - 1][1];
1642             a_valid = 1;
1643         } else { // current block has frame mv and cand. has field MV (so average)
1644             A[0] = (s->current_picture.f.motion_val[0][xy - 1][0]
1645                     + s->current_picture.f.motion_val[0][xy - 1 + off * wrap][0] + 1) >> 1;
1646             A[1] = (s->current_picture.f.motion_val[0][xy - 1][1]
1647                     + s->current_picture.f.motion_val[0][xy - 1 + off * wrap][1] + 1) >> 1;
1648             a_valid = 1;
1649         }
1650         if (!(n & 1) && v->is_intra[s->mb_x - 1]) {
1651             a_valid = 0;
1652             A[0] = A[1] = 0;
1653         }
1654     } else
1655         A[0] = A[1] = 0;
1656     /* Predict B and C */
1657     B[0] = B[1] = C[0] = C[1] = 0;
1658     if (n == 0 || n == 1 || v->blk_mv_type[xy]) {
1659         if (!s->first_slice_line) {
1660             if (!v->is_intra[s->mb_x - s->mb_stride]) {
1661                 b_valid = 1;
1662                 n_adj   = n | 2;
1663                 pos_b   = s->block_index[n_adj] - 2 * wrap;
1664                 if (v->blk_mv_type[pos_b] && v->blk_mv_type[xy]) {
1665                     n_adj = (n & 2) | (n & 1);
1666                 }
1667                 B[0] = s->current_picture.f.motion_val[0][s->block_index[n_adj] - 2 * wrap][0];
1668                 B[1] = s->current_picture.f.motion_val[0][s->block_index[n_adj] - 2 * wrap][1];
1669                 if (v->blk_mv_type[pos_b] && !v->blk_mv_type[xy]) {
1670                     B[0] = (B[0] + s->current_picture.f.motion_val[0][s->block_index[n_adj ^ 2] - 2 * wrap][0] + 1) >> 1;
1671                     B[1] = (B[1] + s->current_picture.f.motion_val[0][s->block_index[n_adj ^ 2] - 2 * wrap][1] + 1) >> 1;
1672                 }
1673             }
1674             if (s->mb_width > 1) {
1675                 if (!v->is_intra[s->mb_x - s->mb_stride + 1]) {
1676                     c_valid = 1;
1677                     n_adj   = 2;
1678                     pos_c   = s->block_index[2] - 2 * wrap + 2;
1679                     if (v->blk_mv_type[pos_c] && v->blk_mv_type[xy]) {
1680                         n_adj = n & 2;
1681                     }
1682                     C[0] = s->current_picture.f.motion_val[0][s->block_index[n_adj] - 2 * wrap + 2][0];
1683                     C[1] = s->current_picture.f.motion_val[0][s->block_index[n_adj] - 2 * wrap + 2][1];
1684                     if (v->blk_mv_type[pos_c] && !v->blk_mv_type[xy]) {
1685                         C[0] = (1 + C[0] + (s->current_picture.f.motion_val[0][s->block_index[n_adj ^ 2] - 2 * wrap + 2][0])) >> 1;
1686                         C[1] = (1 + C[1] + (s->current_picture.f.motion_val[0][s->block_index[n_adj ^ 2] - 2 * wrap + 2][1])) >> 1;
1687                     }
1688                     if (s->mb_x == s->mb_width - 1) {
1689                         if (!v->is_intra[s->mb_x - s->mb_stride - 1]) {
1690                             c_valid = 1;
1691                             n_adj   = 3;
1692                             pos_c   = s->block_index[3] - 2 * wrap - 2;
1693                             if (v->blk_mv_type[pos_c] && v->blk_mv_type[xy]) {
1694                                 n_adj = n | 1;
1695                             }
1696                             C[0] = s->current_picture.f.motion_val[0][s->block_index[n_adj] - 2 * wrap - 2][0];
1697                             C[1] = s->current_picture.f.motion_val[0][s->block_index[n_adj] - 2 * wrap - 2][1];
1698                             if (v->blk_mv_type[pos_c] && !v->blk_mv_type[xy]) {
1699                                 C[0] = (1 + C[0] + s->current_picture.f.motion_val[0][s->block_index[1] - 2 * wrap - 2][0]) >> 1;
1700                                 C[1] = (1 + C[1] + s->current_picture.f.motion_val[0][s->block_index[1] - 2 * wrap - 2][1]) >> 1;
1701                             }
1702                         } else
1703                             c_valid = 0;
1704                     }
1705                 }
1706             }
1707         }
1708     } else {
1709         pos_b   = s->block_index[1];
1710         b_valid = 1;
1711         B[0]    = s->current_picture.f.motion_val[0][pos_b][0];
1712         B[1]    = s->current_picture.f.motion_val[0][pos_b][1];
1713         pos_c   = s->block_index[0];
1714         c_valid = 1;
1715         C[0]    = s->current_picture.f.motion_val[0][pos_c][0];
1716         C[1]    = s->current_picture.f.motion_val[0][pos_c][1];
1717     }
1718
1719     total_valid = a_valid + b_valid + c_valid;
1720     // check if predictor A is out of bounds
1721     if (!s->mb_x && !(n == 1 || n == 3)) {
1722         A[0] = A[1] = 0;
1723     }
1724     // check if predictor B is out of bounds
1725     if ((s->first_slice_line && v->blk_mv_type[xy]) || (s->first_slice_line && !(n & 2))) {
1726         B[0] = B[1] = C[0] = C[1] = 0;
1727     }
1728     if (!v->blk_mv_type[xy]) {
1729         if (s->mb_width == 1) {
1730             px = B[0];
1731             py = B[1];
1732         } else {
1733             if (total_valid >= 2) {
1734                 px = mid_pred(A[0], B[0], C[0]);
1735                 py = mid_pred(A[1], B[1], C[1]);
1736             } else if (total_valid) {
1737                 if (a_valid) { px = A[0]; py = A[1]; }
1738                 if (b_valid) { px = B[0]; py = B[1]; }
1739                 if (c_valid) { px = C[0]; py = C[1]; }
1740             } else
1741                 px = py = 0;
1742         }
1743     } else {
1744         if (a_valid)
1745             field_a = (A[1] & 4) ? 1 : 0;
1746         else
1747             field_a = 0;
1748         if (b_valid)
1749             field_b = (B[1] & 4) ? 1 : 0;
1750         else
1751             field_b = 0;
1752         if (c_valid)
1753             field_c = (C[1] & 4) ? 1 : 0;
1754         else
1755             field_c = 0;
1756
1757         num_oppfield  = field_a + field_b + field_c;
1758         num_samefield = total_valid - num_oppfield;
1759         if (total_valid == 3) {
1760             if ((num_samefield == 3) || (num_oppfield == 3)) {
1761                 px = mid_pred(A[0], B[0], C[0]);
1762                 py = mid_pred(A[1], B[1], C[1]);
1763             } else if (num_samefield >= num_oppfield) {
1764                 /* take one MV from same field set depending on priority
1765                 the check for B may not be necessary */
1766                 px = !field_a ? A[0] : B[0];
1767                 py = !field_a ? A[1] : B[1];
1768             } else {
1769                 px =  field_a ? A[0] : B[0];
1770                 py =  field_a ? A[1] : B[1];
1771             }
1772         } else if (total_valid == 2) {
1773             if (num_samefield >= num_oppfield) {
1774                 if (!field_a && a_valid) {
1775                     px = A[0];
1776                     py = A[1];
1777                 } else if (!field_b && b_valid) {
1778                     px = B[0];
1779                     py = B[1];
1780                 } else if (c_valid) {
1781                     px = C[0];
1782                     py = C[1];
1783                 } else px = py = 0;
1784             } else {
1785                 if (field_a && a_valid) {
1786                     px = A[0];
1787                     py = A[1];
1788                 } else if (field_b && b_valid) {
1789                     px = B[0];
1790                     py = B[1];
1791                 } else if (c_valid) {
1792                     px = C[0];
1793                     py = C[1];
1794                 }
1795             }
1796         } else if (total_valid == 1) {
1797             px = (a_valid) ? A[0] : ((b_valid) ? B[0] : C[0]);
1798             py = (a_valid) ? A[1] : ((b_valid) ? B[1] : C[1]);
1799         } else
1800             px = py = 0;
1801     }
1802
1803     /* store MV using signed modulus of MV range defined in 4.11 */
1804     s->mv[0][n][0] = s->current_picture.f.motion_val[0][xy][0] = ((px + dmv_x + r_x) & ((r_x << 1) - 1)) - r_x;
1805     s->mv[0][n][1] = s->current_picture.f.motion_val[0][xy][1] = ((py + dmv_y + r_y) & ((r_y << 1) - 1)) - r_y;
1806     if (mvn == 1) { /* duplicate motion data for 1-MV block */
1807         s->current_picture.f.motion_val[0][xy +    1    ][0] = s->current_picture.f.motion_val[0][xy][0];
1808         s->current_picture.f.motion_val[0][xy +    1    ][1] = s->current_picture.f.motion_val[0][xy][1];
1809         s->current_picture.f.motion_val[0][xy + wrap    ][0] = s->current_picture.f.motion_val[0][xy][0];
1810         s->current_picture.f.motion_val[0][xy + wrap    ][1] = s->current_picture.f.motion_val[0][xy][1];
1811         s->current_picture.f.motion_val[0][xy + wrap + 1][0] = s->current_picture.f.motion_val[0][xy][0];
1812         s->current_picture.f.motion_val[0][xy + wrap + 1][1] = s->current_picture.f.motion_val[0][xy][1];
1813     } else if (mvn == 2) { /* duplicate motion data for 2-Field MV block */
1814         s->current_picture.f.motion_val[0][xy + 1][0] = s->current_picture.f.motion_val[0][xy][0];
1815         s->current_picture.f.motion_val[0][xy + 1][1] = s->current_picture.f.motion_val[0][xy][1];
1816         s->mv[0][n + 1][0] = s->mv[0][n][0];
1817         s->mv[0][n + 1][1] = s->mv[0][n][1];
1818     }
1819 }
1820
1821 /** Motion compensation for direct or interpolated blocks in B-frames
1822  */
1823 static void vc1_interp_mc(VC1Context *v)
1824 {
1825     MpegEncContext *s = &v->s;
1826     DSPContext *dsp = &v->s.dsp;
1827     uint8_t *srcY, *srcU, *srcV;
1828     int dxy, mx, my, uvmx, uvmy, src_x, src_y, uvsrc_x, uvsrc_y;
1829     int off, off_uv;
1830     int v_edge_pos = s->v_edge_pos >> v->field_mode;
1831
1832     if (!v->field_mode && !v->s.next_picture.f.data[0])
1833         return;
1834
1835     mx   = s->mv[1][0][0];
1836     my   = s->mv[1][0][1];
1837     uvmx = (mx + ((mx & 3) == 3)) >> 1;
1838     uvmy = (my + ((my & 3) == 3)) >> 1;
1839     if (v->field_mode) {
1840         if (v->cur_field_type != v->ref_field_type[1])
1841             my   = my   - 2 + 4 * v->cur_field_type;
1842             uvmy = uvmy - 2 + 4 * v->cur_field_type;
1843     }
1844     if (v->fastuvmc) {
1845         uvmx = uvmx + ((uvmx < 0) ? -(uvmx & 1) : (uvmx & 1));
1846         uvmy = uvmy + ((uvmy < 0) ? -(uvmy & 1) : (uvmy & 1));
1847     }
1848     srcY = s->next_picture.f.data[0];
1849     srcU = s->next_picture.f.data[1];
1850     srcV = s->next_picture.f.data[2];
1851
1852     src_x   = s->mb_x * 16 + (mx   >> 2);
1853     src_y   = s->mb_y * 16 + (my   >> 2);
1854     uvsrc_x = s->mb_x *  8 + (uvmx >> 2);
1855     uvsrc_y = s->mb_y *  8 + (uvmy >> 2);
1856
1857     if (v->profile != PROFILE_ADVANCED) {
1858         src_x   = av_clip(  src_x, -16, s->mb_width  * 16);
1859         src_y   = av_clip(  src_y, -16, s->mb_height * 16);
1860         uvsrc_x = av_clip(uvsrc_x,  -8, s->mb_width  *  8);
1861         uvsrc_y = av_clip(uvsrc_y,  -8, s->mb_height *  8);
1862     } else {
1863         src_x   = av_clip(  src_x, -17, s->avctx->coded_width);
1864         src_y   = av_clip(  src_y, -18, s->avctx->coded_height + 1);
1865         uvsrc_x = av_clip(uvsrc_x,  -8, s->avctx->coded_width  >> 1);
1866         uvsrc_y = av_clip(uvsrc_y,  -8, s->avctx->coded_height >> 1);
1867     }
1868
1869     srcY += src_y   * s->linesize   + src_x;
1870     srcU += uvsrc_y * s->uvlinesize + uvsrc_x;
1871     srcV += uvsrc_y * s->uvlinesize + uvsrc_x;
1872
1873     if (v->field_mode && v->ref_field_type[1]) {
1874         srcY += s->current_picture_ptr->f.linesize[0];
1875         srcU += s->current_picture_ptr->f.linesize[1];
1876         srcV += s->current_picture_ptr->f.linesize[2];
1877     }
1878
1879     /* for grayscale we should not try to read from unknown area */
1880     if (s->flags & CODEC_FLAG_GRAY) {
1881         srcU = s->edge_emu_buffer + 18 * s->linesize;
1882         srcV = s->edge_emu_buffer + 18 * s->linesize;
1883     }
1884
1885     if (v->rangeredfrm || s->h_edge_pos < 22 || v_edge_pos < 22
1886         || (unsigned)(src_x - 1) > s->h_edge_pos - (mx & 3) - 16 - 3
1887         || (unsigned)(src_y - 1) > v_edge_pos    - (my & 3) - 16 - 3) {
1888         uint8_t *uvbuf = s->edge_emu_buffer + 19 * s->linesize;
1889
1890         srcY -= s->mspel * (1 + s->linesize);
1891         s->dsp.emulated_edge_mc(s->edge_emu_buffer, srcY, s->linesize,
1892                                 17 + s->mspel * 2, 17 + s->mspel * 2,
1893                                 src_x - s->mspel, src_y - s->mspel,
1894                                 s->h_edge_pos, v_edge_pos);
1895         srcY = s->edge_emu_buffer;
1896         s->dsp.emulated_edge_mc(uvbuf     , srcU, s->uvlinesize, 8 + 1, 8 + 1,
1897                                 uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, v_edge_pos >> 1);
1898         s->dsp.emulated_edge_mc(uvbuf + 16, srcV, s->uvlinesize, 8 + 1, 8 + 1,
1899                                 uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, v_edge_pos >> 1);
1900         srcU = uvbuf;
1901         srcV = uvbuf + 16;
1902         /* if we deal with range reduction we need to scale source blocks */
1903         if (v->rangeredfrm) {
1904             int i, j;
1905             uint8_t *src, *src2;
1906
1907             src = srcY;
1908             for (j = 0; j < 17 + s->mspel * 2; j++) {
1909                 for (i = 0; i < 17 + s->mspel * 2; i++)
1910                     src[i] = ((src[i] - 128) >> 1) + 128;
1911                 src += s->linesize;
1912             }
1913             src = srcU;
1914             src2 = srcV;
1915             for (j = 0; j < 9; j++) {
1916                 for (i = 0; i < 9; i++) {
1917                     src[i]  = ((src[i]  - 128) >> 1) + 128;
1918                     src2[i] = ((src2[i] - 128) >> 1) + 128;
1919                 }
1920                 src  += s->uvlinesize;
1921                 src2 += s->uvlinesize;
1922             }
1923         }
1924         srcY += s->mspel * (1 + s->linesize);
1925     }
1926
1927     if (v->field_mode && v->cur_field_type) {
1928         off    = s->current_picture_ptr->f.linesize[0];
1929         off_uv = s->current_picture_ptr->f.linesize[1];
1930     } else {
1931         off    = 0;
1932         off_uv = 0;
1933     }
1934
1935     if (s->mspel) {
1936         dxy = ((my & 3) << 2) | (mx & 3);
1937         v->vc1dsp.avg_vc1_mspel_pixels_tab[dxy](s->dest[0] + off    , srcY    , s->linesize, v->rnd);
1938         v->vc1dsp.avg_vc1_mspel_pixels_tab[dxy](s->dest[0] + off + 8, srcY + 8, s->linesize, v->rnd);
1939         srcY += s->linesize * 8;
1940         v->vc1dsp.avg_vc1_mspel_pixels_tab[dxy](s->dest[0] + off + 8 * s->linesize    , srcY    , s->linesize, v->rnd);
1941         v->vc1dsp.avg_vc1_mspel_pixels_tab[dxy](s->dest[0] + off + 8 * s->linesize + 8, srcY + 8, s->linesize, v->rnd);
1942     } else { // hpel mc
1943         dxy = (my & 2) | ((mx & 2) >> 1);
1944
1945         if (!v->rnd)
1946             dsp->avg_pixels_tab[0][dxy](s->dest[0] + off, srcY, s->linesize, 16);
1947         else
1948             dsp->avg_no_rnd_pixels_tab[0][dxy](s->dest[0] + off, srcY, s->linesize, 16);
1949     }
1950
1951     if (s->flags & CODEC_FLAG_GRAY) return;
1952     /* Chroma MC always uses qpel blilinear */
1953     uvmx = (uvmx & 3) << 1;
1954     uvmy = (uvmy & 3) << 1;
1955     if (!v->rnd) {
1956         dsp->avg_h264_chroma_pixels_tab[0](s->dest[1] + off_uv, srcU, s->uvlinesize, 8, uvmx, uvmy);
1957         dsp->avg_h264_chroma_pixels_tab[0](s->dest[2] + off_uv, srcV, s->uvlinesize, 8, uvmx, uvmy);
1958     } else {
1959         v->vc1dsp.avg_no_rnd_vc1_chroma_pixels_tab[0](s->dest[1] + off_uv, srcU, s->uvlinesize, 8, uvmx, uvmy);
1960         v->vc1dsp.avg_no_rnd_vc1_chroma_pixels_tab[0](s->dest[2] + off_uv, srcV, s->uvlinesize, 8, uvmx, uvmy);
1961     }
1962 }
1963
1964 static av_always_inline int scale_mv(int value, int bfrac, int inv, int qs)
1965 {
1966     int n = bfrac;
1967
1968 #if B_FRACTION_DEN==256
1969     if (inv)
1970         n -= 256;
1971     if (!qs)
1972         return 2 * ((value * n + 255) >> 9);
1973     return (value * n + 128) >> 8;
1974 #else
1975     if (inv)
1976         n -= B_FRACTION_DEN;
1977     if (!qs)
1978         return 2 * ((value * n + B_FRACTION_DEN - 1) / (2 * B_FRACTION_DEN));
1979     return (value * n + B_FRACTION_DEN/2) / B_FRACTION_DEN;
1980 #endif
1981 }
1982
1983 /** Reconstruct motion vector for B-frame and do motion compensation
1984  */
1985 static inline void vc1_b_mc(VC1Context *v, int dmv_x[2], int dmv_y[2],
1986                             int direct, int mode)
1987 {
1988     if (v->use_ic) {
1989         v->mv_mode2 = v->mv_mode;
1990         v->mv_mode  = MV_PMODE_INTENSITY_COMP;
1991     }
1992     if (direct) {
1993         vc1_mc_1mv(v, 0);
1994         vc1_interp_mc(v);
1995         if (v->use_ic)
1996             v->mv_mode = v->mv_mode2;
1997         return;
1998     }
1999     if (mode == BMV_TYPE_INTERPOLATED) {
2000         vc1_mc_1mv(v, 0);
2001         vc1_interp_mc(v);
2002         if (v->use_ic)
2003             v->mv_mode = v->mv_mode2;
2004         return;
2005     }
2006
2007     if (v->use_ic && (mode == BMV_TYPE_BACKWARD))
2008         v->mv_mode = v->mv_mode2;
2009     vc1_mc_1mv(v, (mode == BMV_TYPE_BACKWARD));
2010     if (v->use_ic)
2011         v->mv_mode = v->mv_mode2;
2012 }
2013
2014 static inline void vc1_pred_b_mv(VC1Context *v, int dmv_x[2], int dmv_y[2],
2015                                  int direct, int mvtype)
2016 {
2017     MpegEncContext *s = &v->s;
2018     int xy, wrap, off = 0;
2019     int16_t *A, *B, *C;
2020     int px, py;
2021     int sum;
2022     int r_x, r_y;
2023     const uint8_t *is_intra = v->mb_type[0];
2024
2025     r_x = v->range_x;
2026     r_y = v->range_y;
2027     /* scale MV difference to be quad-pel */
2028     dmv_x[0] <<= 1 - s->quarter_sample;
2029     dmv_y[0] <<= 1 - s->quarter_sample;
2030     dmv_x[1] <<= 1 - s->quarter_sample;
2031     dmv_y[1] <<= 1 - s->quarter_sample;
2032
2033     wrap = s->b8_stride;
2034     xy = s->block_index[0];
2035
2036     if (s->mb_intra) {
2037         s->current_picture.f.motion_val[0][xy + v->blocks_off][0] =
2038         s->current_picture.f.motion_val[0][xy + v->blocks_off][1] =
2039         s->current_picture.f.motion_val[1][xy + v->blocks_off][0] =
2040         s->current_picture.f.motion_val[1][xy + v->blocks_off][1] = 0;
2041         return;
2042     }
2043     if (!v->field_mode) {
2044         s->mv[0][0][0] = scale_mv(s->next_picture.f.motion_val[1][xy][0], v->bfraction, 0, s->quarter_sample);
2045         s->mv[0][0][1] = scale_mv(s->next_picture.f.motion_val[1][xy][1], v->bfraction, 0, s->quarter_sample);
2046         s->mv[1][0][0] = scale_mv(s->next_picture.f.motion_val[1][xy][0], v->bfraction, 1, s->quarter_sample);
2047         s->mv[1][0][1] = scale_mv(s->next_picture.f.motion_val[1][xy][1], v->bfraction, 1, s->quarter_sample);
2048
2049         /* Pullback predicted motion vectors as specified in 8.4.5.4 */
2050         s->mv[0][0][0] = av_clip(s->mv[0][0][0], -60 - (s->mb_x << 6), (s->mb_width  << 6) - 4 - (s->mb_x << 6));
2051         s->mv[0][0][1] = av_clip(s->mv[0][0][1], -60 - (s->mb_y << 6), (s->mb_height << 6) - 4 - (s->mb_y << 6));
2052         s->mv[1][0][0] = av_clip(s->mv[1][0][0], -60 - (s->mb_x << 6), (s->mb_width  << 6) - 4 - (s->mb_x << 6));
2053         s->mv[1][0][1] = av_clip(s->mv[1][0][1], -60 - (s->mb_y << 6), (s->mb_height << 6) - 4 - (s->mb_y << 6));
2054     }
2055     if (direct) {
2056         s->current_picture.f.motion_val[0][xy + v->blocks_off][0] = s->mv[0][0][0];
2057         s->current_picture.f.motion_val[0][xy + v->blocks_off][1] = s->mv[0][0][1];
2058         s->current_picture.f.motion_val[1][xy + v->blocks_off][0] = s->mv[1][0][0];
2059         s->current_picture.f.motion_val[1][xy + v->blocks_off][1] = s->mv[1][0][1];
2060         return;
2061     }
2062
2063     if ((mvtype == BMV_TYPE_FORWARD) || (mvtype == BMV_TYPE_INTERPOLATED)) {
2064         C   = s->current_picture.f.motion_val[0][xy - 2];
2065         A   = s->current_picture.f.motion_val[0][xy - wrap * 2];
2066         off = (s->mb_x == (s->mb_width - 1)) ? -2 : 2;
2067         B   = s->current_picture.f.motion_val[0][xy - wrap * 2 + off];
2068
2069         if (!s->mb_x) C[0] = C[1] = 0;
2070         if (!s->first_slice_line) { // predictor A is not out of bounds
2071             if (s->mb_width == 1) {
2072                 px = A[0];
2073                 py = A[1];
2074             } else {
2075                 px = mid_pred(A[0], B[0], C[0]);
2076                 py = mid_pred(A[1], B[1], C[1]);
2077             }
2078         } else if (s->mb_x) { // predictor C is not out of bounds
2079             px = C[0];
2080             py = C[1];
2081         } else {
2082             px = py = 0;
2083         }
2084         /* Pullback MV as specified in 8.3.5.3.4 */
2085         {
2086             int qx, qy, X, Y;
2087             if (v->profile < PROFILE_ADVANCED) {
2088                 qx = (s->mb_x << 5);
2089                 qy = (s->mb_y << 5);
2090                 X  = (s->mb_width  << 5) - 4;
2091                 Y  = (s->mb_height << 5) - 4;
2092                 if (qx + px < -28) px = -28 - qx;
2093                 if (qy + py < -28) py = -28 - qy;
2094                 if (qx + px > X) px = X - qx;
2095                 if (qy + py > Y) py = Y - qy;
2096             } else {
2097                 qx = (s->mb_x << 6);
2098                 qy = (s->mb_y << 6);
2099                 X  = (s->mb_width  << 6) - 4;
2100                 Y  = (s->mb_height << 6) - 4;
2101                 if (qx + px < -60) px = -60 - qx;
2102                 if (qy + py < -60) py = -60 - qy;
2103                 if (qx + px > X) px = X - qx;
2104                 if (qy + py > Y) py = Y - qy;
2105             }
2106         }
2107         /* Calculate hybrid prediction as specified in 8.3.5.3.5 */
2108         if (0 && !s->first_slice_line && s->mb_x) {
2109             if (is_intra[xy - wrap])
2110                 sum = FFABS(px) + FFABS(py);
2111             else
2112                 sum = FFABS(px - A[0]) + FFABS(py - A[1]);
2113             if (sum > 32) {
2114                 if (get_bits1(&s->gb)) {
2115                     px = A[0];
2116                     py = A[1];
2117                 } else {
2118                     px = C[0];
2119                     py = C[1];
2120                 }
2121             } else {
2122                 if (is_intra[xy - 2])
2123                     sum = FFABS(px) + FFABS(py);
2124                 else
2125                     sum = FFABS(px - C[0]) + FFABS(py - C[1]);
2126                 if (sum > 32) {
2127                     if (get_bits1(&s->gb)) {
2128                         px = A[0];
2129                         py = A[1];
2130                     } else {
2131                         px = C[0];
2132                         py = C[1];
2133                     }
2134                 }
2135             }
2136         }
2137         /* store MV using signed modulus of MV range defined in 4.11 */
2138         s->mv[0][0][0] = ((px + dmv_x[0] + r_x) & ((r_x << 1) - 1)) - r_x;
2139         s->mv[0][0][1] = ((py + dmv_y[0] + r_y) & ((r_y << 1) - 1)) - r_y;
2140     }
2141     if ((mvtype == BMV_TYPE_BACKWARD) || (mvtype == BMV_TYPE_INTERPOLATED)) {
2142         C   = s->current_picture.f.motion_val[1][xy - 2];
2143         A   = s->current_picture.f.motion_val[1][xy - wrap * 2];
2144         off = (s->mb_x == (s->mb_width - 1)) ? -2 : 2;
2145         B   = s->current_picture.f.motion_val[1][xy - wrap * 2 + off];
2146
2147         if (!s->mb_x)
2148             C[0] = C[1] = 0;
2149         if (!s->first_slice_line) { // predictor A is not out of bounds
2150             if (s->mb_width == 1) {
2151                 px = A[0];
2152                 py = A[1];
2153             } else {
2154                 px = mid_pred(A[0], B[0], C[0]);
2155                 py = mid_pred(A[1], B[1], C[1]);
2156             }
2157         } else if (s->mb_x) { // predictor C is not out of bounds
2158             px = C[0];
2159             py = C[1];
2160         } else {
2161             px = py = 0;
2162         }
2163         /* Pullback MV as specified in 8.3.5.3.4 */
2164         {
2165             int qx, qy, X, Y;
2166             if (v->profile < PROFILE_ADVANCED) {
2167                 qx = (s->mb_x << 5);
2168                 qy = (s->mb_y << 5);
2169                 X  = (s->mb_width  << 5) - 4;
2170                 Y  = (s->mb_height << 5) - 4;
2171                 if (qx + px < -28) px = -28 - qx;
2172                 if (qy + py < -28) py = -28 - qy;
2173                 if (qx + px > X) px = X - qx;
2174                 if (qy + py > Y) py = Y - qy;
2175             } else {
2176                 qx = (s->mb_x << 6);
2177                 qy = (s->mb_y << 6);
2178                 X  = (s->mb_width  << 6) - 4;
2179                 Y  = (s->mb_height << 6) - 4;
2180                 if (qx + px < -60) px = -60 - qx;
2181                 if (qy + py < -60) py = -60 - qy;
2182                 if (qx + px > X) px = X - qx;
2183                 if (qy + py > Y) py = Y - qy;
2184             }
2185         }
2186         /* Calculate hybrid prediction as specified in 8.3.5.3.5 */
2187         if (0 && !s->first_slice_line && s->mb_x) {
2188             if (is_intra[xy - wrap])
2189                 sum = FFABS(px) + FFABS(py);
2190             else
2191                 sum = FFABS(px - A[0]) + FFABS(py - A[1]);
2192             if (sum > 32) {
2193                 if (get_bits1(&s->gb)) {
2194                     px = A[0];
2195                     py = A[1];
2196                 } else {
2197                     px = C[0];
2198                     py = C[1];
2199                 }
2200             } else {
2201                 if (is_intra[xy - 2])
2202                     sum = FFABS(px) + FFABS(py);
2203                 else
2204                     sum = FFABS(px - C[0]) + FFABS(py - C[1]);
2205                 if (sum > 32) {
2206                     if (get_bits1(&s->gb)) {
2207                         px = A[0];
2208                         py = A[1];
2209                     } else {
2210                         px = C[0];
2211                         py = C[1];
2212                     }
2213                 }
2214             }
2215         }
2216         /* store MV using signed modulus of MV range defined in 4.11 */
2217
2218         s->mv[1][0][0] = ((px + dmv_x[1] + r_x) & ((r_x << 1) - 1)) - r_x;
2219         s->mv[1][0][1] = ((py + dmv_y[1] + r_y) & ((r_y << 1) - 1)) - r_y;
2220     }
2221     s->current_picture.f.motion_val[0][xy][0] = s->mv[0][0][0];
2222     s->current_picture.f.motion_val[0][xy][1] = s->mv[0][0][1];
2223     s->current_picture.f.motion_val[1][xy][0] = s->mv[1][0][0];
2224     s->current_picture.f.motion_val[1][xy][1] = s->mv[1][0][1];
2225 }
2226
2227 static inline void vc1_pred_b_mv_intfi(VC1Context *v, int n, int *dmv_x, int *dmv_y, int mv1, int *pred_flag)
2228 {
2229     int dir = (v->bmvtype == BMV_TYPE_BACKWARD) ? 1 : 0;
2230     MpegEncContext *s = &v->s;
2231     int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
2232
2233     if (v->bmvtype == BMV_TYPE_DIRECT) {
2234         int total_opp, k, f;
2235         if (s->next_picture.f.mb_type[mb_pos + v->mb_off] != MB_TYPE_INTRA) {
2236             s->mv[0][0][0] = scale_mv(s->next_picture.f.motion_val[1][s->block_index[0] + v->blocks_off][0],
2237                                       v->bfraction, 0, s->quarter_sample);
2238             s->mv[0][0][1] = scale_mv(s->next_picture.f.motion_val[1][s->block_index[0] + v->blocks_off][1],
2239                                       v->bfraction, 0, s->quarter_sample);
2240             s->mv[1][0][0] = scale_mv(s->next_picture.f.motion_val[1][s->block_index[0] + v->blocks_off][0],
2241                                       v->bfraction, 1, s->quarter_sample);
2242             s->mv[1][0][1] = scale_mv(s->next_picture.f.motion_val[1][s->block_index[0] + v->blocks_off][1],
2243                                       v->bfraction, 1, s->quarter_sample);
2244
2245             total_opp = v->mv_f_next[0][s->block_index[0] + v->blocks_off]
2246                       + v->mv_f_next[0][s->block_index[1] + v->blocks_off]
2247                       + v->mv_f_next[0][s->block_index[2] + v->blocks_off]
2248                       + v->mv_f_next[0][s->block_index[3] + v->blocks_off];
2249             f = (total_opp > 2) ? 1 : 0;
2250         } else {
2251             s->mv[0][0][0] = s->mv[0][0][1] = 0;
2252             s->mv[1][0][0] = s->mv[1][0][1] = 0;
2253             f = 0;
2254         }
2255         v->ref_field_type[0] = v->ref_field_type[1] = v->cur_field_type ^ f;
2256         for (k = 0; k < 4; k++) {
2257             s->current_picture.f.motion_val[0][s->block_index[k] + v->blocks_off][0] = s->mv[0][0][0];
2258             s->current_picture.f.motion_val[0][s->block_index[k] + v->blocks_off][1] = s->mv[0][0][1];
2259             s->current_picture.f.motion_val[1][s->block_index[k] + v->blocks_off][0] = s->mv[1][0][0];
2260             s->current_picture.f.motion_val[1][s->block_index[k] + v->blocks_off][1] = s->mv[1][0][1];
2261             v->mv_f[0][s->block_index[k] + v->blocks_off] = f;
2262             v->mv_f[1][s->block_index[k] + v->blocks_off] = f;
2263         }
2264         return;
2265     }
2266     if (v->bmvtype == BMV_TYPE_INTERPOLATED) {
2267         vc1_pred_mv(v, 0, dmv_x[0], dmv_y[0],   1, v->range_x, v->range_y, v->mb_type[0], pred_flag[0], 0);
2268         vc1_pred_mv(v, 0, dmv_x[1], dmv_y[1],   1, v->range_x, v->range_y, v->mb_type[0], pred_flag[1], 1);
2269         return;
2270     }
2271     if (dir) { // backward
2272         vc1_pred_mv(v, n, dmv_x[1], dmv_y[1], mv1, v->range_x, v->range_y, v->mb_type[0], pred_flag[1], 1);
2273         if (n == 3 || mv1) {
2274             vc1_pred_mv(v, 0, dmv_x[0], dmv_y[0],   1, v->range_x, v->range_y, v->mb_type[0], 0, 0);
2275         }
2276     } else { // forward
2277         vc1_pred_mv(v, n, dmv_x[0], dmv_y[0], mv1, v->range_x, v->range_y, v->mb_type[0], pred_flag[0], 0);
2278         if (n == 3 || mv1) {
2279             vc1_pred_mv(v, 0, dmv_x[1], dmv_y[1],   1, v->range_x, v->range_y, v->mb_type[0], 0, 1);
2280         }
2281     }
2282 }
2283
2284 /** Get predicted DC value for I-frames only
2285  * prediction dir: left=0, top=1
2286  * @param s MpegEncContext
2287  * @param overlap flag indicating that overlap filtering is used
2288  * @param pq integer part of picture quantizer
2289  * @param[in] n block index in the current MB
2290  * @param dc_val_ptr Pointer to DC predictor
2291  * @param dir_ptr Prediction direction for use in AC prediction
2292  */
2293 static inline int vc1_i_pred_dc(MpegEncContext *s, int overlap, int pq, int n,
2294                                 int16_t **dc_val_ptr, int *dir_ptr)
2295 {
2296     int a, b, c, wrap, pred, scale;
2297     int16_t *dc_val;
2298     static const uint16_t dcpred[32] = {
2299         -1, 1024,  512,  341,  256,  205,  171,  146,  128,
2300              114,  102,   93,   85,   79,   73,   68,   64,
2301               60,   57,   54,   51,   49,   47,   45,   43,
2302               41,   39,   38,   37,   35,   34,   33
2303     };
2304
2305     /* find prediction - wmv3_dc_scale always used here in fact */
2306     if (n < 4) scale = s->y_dc_scale;
2307     else       scale = s->c_dc_scale;
2308
2309     wrap   = s->block_wrap[n];
2310     dc_val = s->dc_val[0] + s->block_index[n];
2311
2312     /* B A
2313      * C X
2314      */
2315     c = dc_val[ - 1];
2316     b = dc_val[ - 1 - wrap];
2317     a = dc_val[ - wrap];
2318
2319     if (pq < 9 || !overlap) {
2320         /* Set outer values */
2321         if (s->first_slice_line && (n != 2 && n != 3))
2322             b = a = dcpred[scale];
2323         if (s->mb_x == 0 && (n != 1 && n != 3))
2324             b = c = dcpred[scale];
2325     } else {
2326         /* Set outer values */
2327         if (s->first_slice_line && (n != 2 && n != 3))
2328             b = a = 0;
2329         if (s->mb_x == 0 && (n != 1 && n != 3))
2330             b = c = 0;
2331     }
2332
2333     if (abs(a - b) <= abs(b - c)) {
2334         pred     = c;
2335         *dir_ptr = 1; // left
2336     } else {
2337         pred     = a;
2338         *dir_ptr = 0; // top
2339     }
2340
2341     /* update predictor */
2342     *dc_val_ptr = &dc_val[0];
2343     return pred;
2344 }
2345
2346
2347 /** Get predicted DC value
2348  * prediction dir: left=0, top=1
2349  * @param s MpegEncContext
2350  * @param overlap flag indicating that overlap filtering is used
2351  * @param pq integer part of picture quantizer
2352  * @param[in] n block index in the current MB
2353  * @param a_avail flag indicating top block availability
2354  * @param c_avail flag indicating left block availability
2355  * @param dc_val_ptr Pointer to DC predictor
2356  * @param dir_ptr Prediction direction for use in AC prediction
2357  */
2358 static inline int vc1_pred_dc(MpegEncContext *s, int overlap, int pq, int n,
2359                               int a_avail, int c_avail,
2360                               int16_t **dc_val_ptr, int *dir_ptr)
2361 {
2362     int a, b, c, wrap, pred;
2363     int16_t *dc_val;
2364     int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
2365     int q1, q2 = 0;
2366     int dqscale_index;
2367
2368     wrap = s->block_wrap[n];
2369     dc_val = s->dc_val[0] + s->block_index[n];
2370
2371     /* B A
2372      * C X
2373      */
2374     c = dc_val[ - 1];
2375     b = dc_val[ - 1 - wrap];
2376     a = dc_val[ - wrap];
2377     /* scale predictors if needed */
2378     q1 = s->current_picture.f.qscale_table[mb_pos];
2379     dqscale_index = s->y_dc_scale_table[q1] - 1;
2380     if (dqscale_index < 0)
2381         return 0;
2382     if (c_avail && (n != 1 && n != 3)) {
2383         q2 = s->current_picture.f.qscale_table[mb_pos - 1];
2384         if (q2 && q2 != q1)
2385             c = (c * s->y_dc_scale_table[q2] * ff_vc1_dqscale[dqscale_index] + 0x20000) >> 18;
2386     }
2387     if (a_avail && (n != 2 && n != 3)) {
2388         q2 = s->current_picture.f.qscale_table[mb_pos - s->mb_stride];
2389         if (q2 && q2 != q1)
2390             a = (a * s->y_dc_scale_table[q2] * ff_vc1_dqscale[dqscale_index] + 0x20000) >> 18;
2391     }
2392     if (a_avail && c_avail && (n != 3)) {
2393         int off = mb_pos;
2394         if (n != 1)
2395             off--;
2396         if (n != 2)
2397             off -= s->mb_stride;
2398         q2 = s->current_picture.f.qscale_table[off];
2399         if (q2 && q2 != q1)
2400             b = (b * s->y_dc_scale_table[q2] * ff_vc1_dqscale[dqscale_index] + 0x20000) >> 18;
2401     }
2402
2403     if (a_avail && c_avail) {
2404         if (abs(a - b) <= abs(b - c)) {
2405             pred     = c;
2406             *dir_ptr = 1; // left
2407         } else {
2408             pred     = a;
2409             *dir_ptr = 0; // top
2410         }
2411     } else if (a_avail) {
2412         pred     = a;
2413         *dir_ptr = 0; // top
2414     } else if (c_avail) {
2415         pred     = c;
2416         *dir_ptr = 1; // left
2417     } else {
2418         pred     = 0;
2419         *dir_ptr = 1; // left
2420     }
2421
2422     /* update predictor */
2423     *dc_val_ptr = &dc_val[0];
2424     return pred;
2425 }
2426
2427 /** @} */ // Block group
2428
2429 /**
2430  * @name VC1 Macroblock-level functions in Simple/Main Profiles
2431  * @see 7.1.4, p91 and 8.1.1.7, p(1)04
2432  * @{
2433  */
2434
2435 static inline int vc1_coded_block_pred(MpegEncContext * s, int n,
2436                                        uint8_t **coded_block_ptr)
2437 {
2438     int xy, wrap, pred, a, b, c;
2439
2440     xy   = s->block_index[n];
2441     wrap = s->b8_stride;
2442
2443     /* B C
2444      * A X
2445      */
2446     a = s->coded_block[xy - 1       ];
2447     b = s->coded_block[xy - 1 - wrap];
2448     c = s->coded_block[xy     - wrap];
2449
2450     if (b == c) {
2451         pred = a;
2452     } else {
2453         pred = c;
2454     }
2455
2456     /* store value */
2457     *coded_block_ptr = &s->coded_block[xy];
2458
2459     return pred;
2460 }
2461
2462 /**
2463  * Decode one AC coefficient
2464  * @param v The VC1 context
2465  * @param last Last coefficient
2466  * @param skip How much zero coefficients to skip
2467  * @param value Decoded AC coefficient value
2468  * @param codingset set of VLC to decode data
2469  * @see 8.1.3.4
2470  */
2471 static void vc1_decode_ac_coeff(VC1Context *v, int *last, int *skip,
2472                                 int *value, int codingset)
2473 {
2474     GetBitContext *gb = &v->s.gb;
2475     int index, escape, run = 0, level = 0, lst = 0;
2476
2477     index = get_vlc2(gb, ff_vc1_ac_coeff_table[codingset].table, AC_VLC_BITS, 3);
2478     if (index != ff_vc1_ac_sizes[codingset] - 1) {
2479         run   = vc1_index_decode_table[codingset][index][0];
2480         level = vc1_index_decode_table[codingset][index][1];
2481         lst   = index >= vc1_last_decode_table[codingset] || get_bits_left(gb) < 0;
2482         if (get_bits1(gb))
2483             level = -level;
2484     } else {
2485         escape = decode210(gb);
2486         if (escape != 2) {
2487             index = get_vlc2(gb, ff_vc1_ac_coeff_table[codingset].table, AC_VLC_BITS, 3);
2488             run   = vc1_index_decode_table[codingset][index][0];
2489             level = vc1_index_decode_table[codingset][index][1];
2490             lst   = index >= vc1_last_decode_table[codingset];
2491             if (escape == 0) {
2492                 if (lst)
2493                     level += vc1_last_delta_level_table[codingset][run];
2494                 else
2495                     level += vc1_delta_level_table[codingset][run];
2496             } else {
2497                 if (lst)
2498                     run += vc1_last_delta_run_table[codingset][level] + 1;
2499                 else
2500                     run += vc1_delta_run_table[codingset][level] + 1;
2501             }
2502             if (get_bits1(gb))
2503                 level = -level;
2504         } else {
2505             int sign;
2506             lst = get_bits1(gb);
2507             if (v->s.esc3_level_length == 0) {
2508                 if (v->pq < 8 || v->dquantfrm) { // table 59
2509                     v->s.esc3_level_length = get_bits(gb, 3);
2510                     if (!v->s.esc3_level_length)
2511                         v->s.esc3_level_length = get_bits(gb, 2) + 8;
2512                 } else { // table 60
2513                     v->s.esc3_level_length = get_unary(gb, 1, 6) + 2;
2514                 }
2515                 v->s.esc3_run_length = 3 + get_bits(gb, 2);
2516             }
2517             run   = get_bits(gb, v->s.esc3_run_length);
2518             sign  = get_bits1(gb);
2519             level = get_bits(gb, v->s.esc3_level_length);
2520             if (sign)
2521                 level = -level;
2522         }
2523     }
2524
2525     *last  = lst;
2526     *skip  = run;
2527     *value = level;
2528 }
2529
2530 /** Decode intra block in intra frames - should be faster than decode_intra_block
2531  * @param v VC1Context
2532  * @param block block to decode
2533  * @param[in] n subblock index
2534  * @param coded are AC coeffs present or not
2535  * @param codingset set of VLC to decode data
2536  */
2537 static int vc1_decode_i_block(VC1Context *v, DCTELEM block[64], int n,
2538                               int coded, int codingset)
2539 {
2540     GetBitContext *gb = &v->s.gb;
2541     MpegEncContext *s = &v->s;
2542     int dc_pred_dir = 0; /* Direction of the DC prediction used */
2543     int i;
2544     int16_t *dc_val;
2545     int16_t *ac_val, *ac_val2;
2546     int dcdiff;
2547
2548     /* Get DC differential */
2549     if (n < 4) {
2550         dcdiff = get_vlc2(&s->gb, ff_msmp4_dc_luma_vlc[s->dc_table_index].table, DC_VLC_BITS, 3);
2551     } else {
2552         dcdiff = get_vlc2(&s->gb, ff_msmp4_dc_chroma_vlc[s->dc_table_index].table, DC_VLC_BITS, 3);
2553     }
2554     if (dcdiff < 0) {
2555         av_log(s->avctx, AV_LOG_ERROR, "Illegal DC VLC\n");
2556         return -1;
2557     }
2558     if (dcdiff) {
2559         if (dcdiff == 119 /* ESC index value */) {
2560             /* TODO: Optimize */
2561             if (v->pq == 1)      dcdiff = get_bits(gb, 10);
2562             else if (v->pq == 2) dcdiff = get_bits(gb, 9);
2563             else                 dcdiff = get_bits(gb, 8);
2564         } else {
2565             if (v->pq == 1)
2566                 dcdiff = (dcdiff << 2) + get_bits(gb, 2) - 3;
2567             else if (v->pq == 2)
2568                 dcdiff = (dcdiff << 1) + get_bits1(gb)   - 1;
2569         }
2570         if (get_bits1(gb))
2571             dcdiff = -dcdiff;
2572     }
2573
2574     /* Prediction */
2575     dcdiff += vc1_i_pred_dc(&v->s, v->overlap, v->pq, n, &dc_val, &dc_pred_dir);
2576     *dc_val = dcdiff;
2577
2578     /* Store the quantized DC coeff, used for prediction */
2579     if (n < 4) {
2580         block[0] = dcdiff * s->y_dc_scale;
2581     } else {
2582         block[0] = dcdiff * s->c_dc_scale;
2583     }
2584     /* Skip ? */
2585     if (!coded) {
2586         goto not_coded;
2587     }
2588
2589     // AC Decoding
2590     i = 1;
2591
2592     {
2593         int last = 0, skip, value;
2594         const uint8_t *zz_table;
2595         int scale;
2596         int k;
2597
2598         scale = v->pq * 2 + v->halfpq;
2599
2600         if (v->s.ac_pred) {
2601             if (!dc_pred_dir)
2602                 zz_table = v->zz_8x8[2];
2603             else
2604                 zz_table = v->zz_8x8[3];
2605         } else
2606             zz_table = v->zz_8x8[1];
2607
2608         ac_val  = s->ac_val[0][0] + s->block_index[n] * 16;
2609         ac_val2 = ac_val;
2610         if (dc_pred_dir) // left
2611             ac_val -= 16;
2612         else // top
2613             ac_val -= 16 * s->block_wrap[n];
2614
2615         while (!last) {
2616             vc1_decode_ac_coeff(v, &last, &skip, &value, codingset);
2617             i += skip;
2618             if (i > 63)
2619                 break;
2620             block[zz_table[i++]] = value;
2621         }
2622
2623         /* apply AC prediction if needed */
2624         if (s->ac_pred) {
2625             if (dc_pred_dir) { // left
2626                 for (k = 1; k < 8; k++)
2627                     block[k << v->left_blk_sh] += ac_val[k];
2628             } else { // top
2629                 for (k = 1; k < 8; k++)
2630                     block[k << v->top_blk_sh] += ac_val[k + 8];
2631             }
2632         }