git.sesse.net Git - ffmpeg/blob - libavcodec/mpegvideo.c

   1 /*
   2  * The simplest mpeg encoder (well, it was the simplest!)
   3  * Copyright (c) 2000,2001 Fabrice Bellard
   4  * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
   5  *
   6  * 4MV & hq & B-frame encoding stuff by Michael Niedermayer <michaelni@gmx.at>
   7  *
   8  * This file is part of FFmpeg.
   9  *
  10  * FFmpeg is free software; you can redistribute it and/or
  11  * modify it under the terms of the GNU Lesser General Public
  12  * License as published by the Free Software Foundation; either
  13  * version 2.1 of the License, or (at your option) any later version.
  14  *
  15  * FFmpeg is distributed in the hope that it will be useful,
  16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  18  * Lesser General Public License for more details.
  19  *
  20  * You should have received a copy of the GNU Lesser General Public
  21  * License along with FFmpeg; if not, write to the Free Software
  22  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  23  */
  24
  25 /**
  26  * @file
  27  * The simplest mpeg encoder (well, it was the simplest!).
  28  */
  29
  30 #include "libavutil/attributes.h"
  31 #include "libavutil/avassert.h"
  32 #include "libavutil/imgutils.h"
  33 #include "libavutil/internal.h"
  34 #include "libavutil/motion_vector.h"
  35 #include "avcodec.h"
  36 #include "blockdsp.h"
  37 #include "h264chroma.h"
  38 #include "idctdsp.h"
  39 #include "internal.h"
  40 #include "mathops.h"
  41 #include "mpeg_er.h"
  42 #include "mpegutils.h"
  43 #include "mpegvideo.h"
  44 #include "mpegvideodata.h"
  45 #include "mjpegenc.h"
  46 #include "msmpeg4.h"
  47 #include "qpeldsp.h"
  48 #include "thread.h"
  49 #include "wmv2.h"
  50 #include <limits.h>
  51
  52 static void dct_unquantize_mpeg1_intra_c(MpegEncContext *s,
  53                                    int16_t *block, int n, int qscale)
  54 {
  55     int i, level, nCoeffs;
  56     const uint16_t *quant_matrix;
  57
  58     nCoeffs= s->block_last_index[n];
  59
  60     block[0] *= n < 4 ? s->y_dc_scale : s->c_dc_scale;
  61     /* XXX: only MPEG-1 */
  62     quant_matrix = s->intra_matrix;
  63     for(i=1;i<=nCoeffs;i++) {
  64         int j= s->intra_scantable.permutated[i];
  65         level = block[j];
  66         if (level) {
  67             if (level < 0) {
  68                 level = -level;
  69                 level = (int)(level * qscale * quant_matrix[j]) >> 3;
  70                 level = (level - 1) | 1;
  71                 level = -level;
  72             } else {
  73                 level = (int)(level * qscale * quant_matrix[j]) >> 3;
  74                 level = (level - 1) | 1;
  75             }
  76             block[j] = level;
  77         }
  78     }
  79 }
  80
  81 static void dct_unquantize_mpeg1_inter_c(MpegEncContext *s,
  82                                    int16_t *block, int n, int qscale)
  83 {
  84     int i, level, nCoeffs;
  85     const uint16_t *quant_matrix;
  86
  87     nCoeffs= s->block_last_index[n];
  88
  89     quant_matrix = s->inter_matrix;
  90     for(i=0; i<=nCoeffs; i++) {
  91         int j= s->intra_scantable.permutated[i];
  92         level = block[j];
  93         if (level) {
  94             if (level < 0) {
  95                 level = -level;
  96                 level = (((level << 1) + 1) * qscale *
  97                          ((int) (quant_matrix[j]))) >> 4;
  98                 level = (level - 1) | 1;
  99                 level = -level;
 100             } else {
 101                 level = (((level << 1) + 1) * qscale *
 102                          ((int) (quant_matrix[j]))) >> 4;
 103                 level = (level - 1) | 1;
 104             }
 105             block[j] = level;
 106         }
 107     }
 108 }
 109
 110 static void dct_unquantize_mpeg2_intra_c(MpegEncContext *s,
 111                                    int16_t *block, int n, int qscale)
 112 {
 113     int i, level, nCoeffs;
 114     const uint16_t *quant_matrix;
 115
 116     if (s->q_scale_type) qscale = ff_mpeg2_non_linear_qscale[qscale];
 117     else                 qscale <<= 1;
 118
 119     if(s->alternate_scan) nCoeffs= 63;
 120     else nCoeffs= s->block_last_index[n];
 121
 122     block[0] *= n < 4 ? s->y_dc_scale : s->c_dc_scale;
 123     quant_matrix = s->intra_matrix;
 124     for(i=1;i<=nCoeffs;i++) {
 125         int j= s->intra_scantable.permutated[i];
 126         level = block[j];
 127         if (level) {
 128             if (level < 0) {
 129                 level = -level;
 130                 level = (int)(level * qscale * quant_matrix[j]) >> 4;
 131                 level = -level;
 132             } else {
 133                 level = (int)(level * qscale * quant_matrix[j]) >> 4;
 134             }
 135             block[j] = level;
 136         }
 137     }
 138 }
 139
 140 static void dct_unquantize_mpeg2_intra_bitexact(MpegEncContext *s,
 141                                    int16_t *block, int n, int qscale)
 142 {
 143     int i, level, nCoeffs;
 144     const uint16_t *quant_matrix;
 145     int sum=-1;
 146
 147     if (s->q_scale_type) qscale = ff_mpeg2_non_linear_qscale[qscale];
 148     else                 qscale <<= 1;
 149
 150     if(s->alternate_scan) nCoeffs= 63;
 151     else nCoeffs= s->block_last_index[n];
 152
 153     block[0] *= n < 4 ? s->y_dc_scale : s->c_dc_scale;
 154     sum += block[0];
 155     quant_matrix = s->intra_matrix;
 156     for(i=1;i<=nCoeffs;i++) {
 157         int j= s->intra_scantable.permutated[i];
 158         level = block[j];
 159         if (level) {
 160             if (level < 0) {
 161                 level = -level;
 162                 level = (int)(level * qscale * quant_matrix[j]) >> 4;
 163                 level = -level;
 164             } else {
 165                 level = (int)(level * qscale * quant_matrix[j]) >> 4;
 166             }
 167             block[j] = level;
 168             sum+=level;
 169         }
 170     }
 171     block[63]^=sum&1;
 172 }
 173
 174 static void dct_unquantize_mpeg2_inter_c(MpegEncContext *s,
 175                                    int16_t *block, int n, int qscale)
 176 {
 177     int i, level, nCoeffs;
 178     const uint16_t *quant_matrix;
 179     int sum=-1;
 180
 181     if (s->q_scale_type) qscale = ff_mpeg2_non_linear_qscale[qscale];
 182     else                 qscale <<= 1;
 183
 184     if(s->alternate_scan) nCoeffs= 63;
 185     else nCoeffs= s->block_last_index[n];
 186
 187     quant_matrix = s->inter_matrix;
 188     for(i=0; i<=nCoeffs; i++) {
 189         int j= s->intra_scantable.permutated[i];
 190         level = block[j];
 191         if (level) {
 192             if (level < 0) {
 193                 level = -level;
 194                 level = (((level << 1) + 1) * qscale *
 195                          ((int) (quant_matrix[j]))) >> 5;
 196                 level = -level;
 197             } else {
 198                 level = (((level << 1) + 1) * qscale *
 199                          ((int) (quant_matrix[j]))) >> 5;
 200             }
 201             block[j] = level;
 202             sum+=level;
 203         }
 204     }
 205     block[63]^=sum&1;
 206 }
 207
 208 static void dct_unquantize_h263_intra_c(MpegEncContext *s,
 209                                   int16_t *block, int n, int qscale)
 210 {
 211     int i, level, qmul, qadd;
 212     int nCoeffs;
 213
 214     av_assert2(s->block_last_index[n]>=0 || s->h263_aic);
 215
 216     qmul = qscale << 1;
 217
 218     if (!s->h263_aic) {
 219         block[0] *= n < 4 ? s->y_dc_scale : s->c_dc_scale;
 220         qadd = (qscale - 1) | 1;
 221     }else{
 222         qadd = 0;
 223     }
 224     if(s->ac_pred)
 225         nCoeffs=63;
 226     else
 227         nCoeffs= s->intra_scantable.raster_end[ s->block_last_index[n] ];
 228
 229     for(i=1; i<=nCoeffs; i++) {
 230         level = block[i];
 231         if (level) {
 232             if (level < 0) {
 233                 level = level * qmul - qadd;
 234             } else {
 235                 level = level * qmul + qadd;
 236             }
 237             block[i] = level;
 238         }
 239     }
 240 }
 241
 242 static void dct_unquantize_h263_inter_c(MpegEncContext *s,
 243                                   int16_t *block, int n, int qscale)
 244 {
 245     int i, level, qmul, qadd;
 246     int nCoeffs;
 247
 248     av_assert2(s->block_last_index[n]>=0);
 249
 250     qadd = (qscale - 1) | 1;
 251     qmul = qscale << 1;
 252
 253     nCoeffs= s->inter_scantable.raster_end[ s->block_last_index[n] ];
 254
 255     for(i=0; i<=nCoeffs; i++) {
 256         level = block[i];
 257         if (level) {
 258             if (level < 0) {
 259                 level = level * qmul - qadd;
 260             } else {
 261                 level = level * qmul + qadd;
 262             }
 263             block[i] = level;
 264         }
 265     }
 266 }
 267
 268
 269 static void gray16(uint8_t *dst, const uint8_t *src, ptrdiff_t linesize, int h)
 270 {
 271     while(h--)
 272         memset(dst + h*linesize, 128, 16);
 273 }
 274
 275 static void gray8(uint8_t *dst, const uint8_t *src, ptrdiff_t linesize, int h)
 276 {
 277     while(h--)
 278         memset(dst + h*linesize, 128, 8);
 279 }
 280
 281 /* init common dct for both encoder and decoder */
 282 static av_cold int dct_init(MpegEncContext *s)
 283 {
 284     ff_blockdsp_init(&s->bdsp, s->avctx);
 285     ff_h264chroma_init(&s->h264chroma, 8); //for lowres
 286     ff_hpeldsp_init(&s->hdsp, s->avctx->flags);
 287     ff_mpegvideodsp_init(&s->mdsp);
 288     ff_videodsp_init(&s->vdsp, s->avctx->bits_per_raw_sample);
 289
 290     if (s->avctx->debug & FF_DEBUG_NOMC) {
 291         int i;
 292         for (i=0; i<4; i++) {
 293             s->hdsp.avg_pixels_tab[0][i] = gray16;
 294             s->hdsp.put_pixels_tab[0][i] = gray16;
 295             s->hdsp.put_no_rnd_pixels_tab[0][i] = gray16;
 296
 297             s->hdsp.avg_pixels_tab[1][i] = gray8;
 298             s->hdsp.put_pixels_tab[1][i] = gray8;
 299             s->hdsp.put_no_rnd_pixels_tab[1][i] = gray8;
 300         }
 301     }
 302
 303     s->dct_unquantize_h263_intra = dct_unquantize_h263_intra_c;
 304     s->dct_unquantize_h263_inter = dct_unquantize_h263_inter_c;
 305     s->dct_unquantize_mpeg1_intra = dct_unquantize_mpeg1_intra_c;
 306     s->dct_unquantize_mpeg1_inter = dct_unquantize_mpeg1_inter_c;
 307     s->dct_unquantize_mpeg2_intra = dct_unquantize_mpeg2_intra_c;
 308     if (s->avctx->flags & AV_CODEC_FLAG_BITEXACT)
 309         s->dct_unquantize_mpeg2_intra = dct_unquantize_mpeg2_intra_bitexact;
 310     s->dct_unquantize_mpeg2_inter = dct_unquantize_mpeg2_inter_c;
 311
 312     if (HAVE_INTRINSICS_NEON)
 313         ff_mpv_common_init_neon(s);
 314
 315     if (ARCH_ALPHA)
 316         ff_mpv_common_init_axp(s);
 317     if (ARCH_ARM)
 318         ff_mpv_common_init_arm(s);
 319     if (ARCH_PPC)
 320         ff_mpv_common_init_ppc(s);
 321     if (ARCH_X86)
 322         ff_mpv_common_init_x86(s);
 323     if (ARCH_MIPS)
 324         ff_mpv_common_init_mips(s);
 325
 326     return 0;
 327 }
 328
 329 av_cold void ff_mpv_idct_init(MpegEncContext *s)
 330 {
 331     if (s->codec_id == AV_CODEC_ID_MPEG4)
 332         s->idsp.mpeg4_studio_profile = s->studio_profile;
 333     ff_idctdsp_init(&s->idsp, s->avctx);
 334
 335     /* load & permutate scantables
 336      * note: only wmv uses different ones
 337      */
 338     if (s->alternate_scan) {
 339         ff_init_scantable(s->idsp.idct_permutation, &s->inter_scantable, ff_alternate_vertical_scan);
 340         ff_init_scantable(s->idsp.idct_permutation, &s->intra_scantable, ff_alternate_vertical_scan);
 341     } else {
 342         ff_init_scantable(s->idsp.idct_permutation, &s->inter_scantable, ff_zigzag_direct);
 343         ff_init_scantable(s->idsp.idct_permutation, &s->intra_scantable, ff_zigzag_direct);
 344     }
 345     ff_init_scantable(s->idsp.idct_permutation, &s->intra_h_scantable, ff_alternate_horizontal_scan);
 346     ff_init_scantable(s->idsp.idct_permutation, &s->intra_v_scantable, ff_alternate_vertical_scan);
 347 }
 348
 349 static int alloc_picture(MpegEncContext *s, Picture *pic, int shared)
 350 {
 351     return ff_alloc_picture(s->avctx, pic, &s->me, &s->sc, shared, 0,
 352                             s->chroma_x_shift, s->chroma_y_shift, s->out_format,
 353                             s->mb_stride, s->mb_width, s->mb_height, s->b8_stride,
 354                             &s->linesize, &s->uvlinesize);
 355 }
 356
 357 static int init_duplicate_context(MpegEncContext *s)
 358 {
 359     int y_size = s->b8_stride * (2 * s->mb_height + 1);
 360     int c_size = s->mb_stride * (s->mb_height + 1);
 361     int yc_size = y_size + 2 * c_size;
 362     int i;
 363
 364     if (s->mb_height & 1)
 365         yc_size += 2*s->b8_stride + 2*s->mb_stride;
 366
 367     s->sc.edge_emu_buffer =
 368     s->me.scratchpad   =
 369     s->me.temp         =
 370     s->sc.rd_scratchpad   =
 371     s->sc.b_scratchpad    =
 372     s->sc.obmc_scratchpad = NULL;
 373
 374     if (s->encoding) {
 375         FF_ALLOCZ_OR_GOTO(s->avctx, s->me.map,
 376                           ME_MAP_SIZE * sizeof(uint32_t), fail)
 377         FF_ALLOCZ_OR_GOTO(s->avctx, s->me.score_map,
 378                           ME_MAP_SIZE * sizeof(uint32_t), fail)
 379         if (s->noise_reduction) {
 380             FF_ALLOCZ_OR_GOTO(s->avctx, s->dct_error_sum,
 381                               2 * 64 * sizeof(int), fail)
 382         }
 383     }
 384     FF_ALLOCZ_OR_GOTO(s->avctx, s->blocks, 64 * 12 * 2 * sizeof(int16_t), fail)
 385     s->block = s->blocks[0];
 386
 387     for (i = 0; i < 12; i++) {
 388         s->pblocks[i] = &s->block[i];
 389     }
 390
 391     FF_ALLOCZ_OR_GOTO(s->avctx, s->block32, sizeof(*s->block32), fail)
 392     s->dpcm_direction = 0;
 393     FF_ALLOCZ_OR_GOTO(s->avctx, s->dpcm_macroblock, sizeof(*s->dpcm_macroblock), fail)
 394
 395     if (s->avctx->codec_tag == AV_RL32("VCR2")) {
 396         // exchange uv
 397         FFSWAP(void *, s->pblocks[4], s->pblocks[5]);
 398     }
 399
 400     if (s->out_format == FMT_H263) {
 401         /* ac values */
 402         FF_ALLOCZ_OR_GOTO(s->avctx, s->ac_val_base,
 403                           yc_size * sizeof(int16_t) * 16, fail);
 404         s->ac_val[0] = s->ac_val_base + s->b8_stride + 1;
 405         s->ac_val[1] = s->ac_val_base + y_size + s->mb_stride + 1;
 406         s->ac_val[2] = s->ac_val[1] + c_size;
 407     }
 408
 409     return 0;
 410 fail:
 411     return -1; // free() through ff_mpv_common_end()
 412 }
 413
 414 static void free_duplicate_context(MpegEncContext *s)
 415 {
 416     if (!s)
 417         return;
 418
 419     av_freep(&s->sc.edge_emu_buffer);
 420     av_freep(&s->me.scratchpad);
 421     s->me.temp =
 422     s->sc.rd_scratchpad =
 423     s->sc.b_scratchpad =
 424     s->sc.obmc_scratchpad = NULL;
 425
 426     av_freep(&s->dct_error_sum);
 427     av_freep(&s->me.map);
 428     av_freep(&s->me.score_map);
 429     av_freep(&s->blocks);
 430     av_freep(&s->block32);
 431     av_freep(&s->dpcm_macroblock);
 432     av_freep(&s->ac_val_base);
 433     s->block = NULL;
 434 }
 435
 436 static void backup_duplicate_context(MpegEncContext *bak, MpegEncContext *src)
 437 {
 438 #define COPY(a) bak->a = src->a
 439     COPY(sc.edge_emu_buffer);
 440     COPY(me.scratchpad);
 441     COPY(me.temp);
 442     COPY(sc.rd_scratchpad);
 443     COPY(sc.b_scratchpad);
 444     COPY(sc.obmc_scratchpad);
 445     COPY(me.map);
 446     COPY(me.score_map);
 447     COPY(blocks);
 448     COPY(block);
 449     COPY(block32);
 450     COPY(dpcm_macroblock);
 451     COPY(dpcm_direction);
 452     COPY(start_mb_y);
 453     COPY(end_mb_y);
 454     COPY(me.map_generation);
 455     COPY(pb);
 456     COPY(dct_error_sum);
 457     COPY(dct_count[0]);
 458     COPY(dct_count[1]);
 459     COPY(ac_val_base);
 460     COPY(ac_val[0]);
 461     COPY(ac_val[1]);
 462     COPY(ac_val[2]);
 463 #undef COPY
 464 }
 465
 466 int ff_update_duplicate_context(MpegEncContext *dst, MpegEncContext *src)
 467 {
 468     MpegEncContext bak;
 469     int i, ret;
 470     // FIXME copy only needed parts
 471     backup_duplicate_context(&bak, dst);
 472     memcpy(dst, src, sizeof(MpegEncContext));
 473     backup_duplicate_context(dst, &bak);
 474     for (i = 0; i < 12; i++) {
 475         dst->pblocks[i] = &dst->block[i];
 476     }
 477     if (dst->avctx->codec_tag == AV_RL32("VCR2")) {
 478         // exchange uv
 479         FFSWAP(void *, dst->pblocks[4], dst->pblocks[5]);
 480     }
 481     if (!dst->sc.edge_emu_buffer &&
 482         (ret = ff_mpeg_framesize_alloc(dst->avctx, &dst->me,
 483                                        &dst->sc, dst->linesize)) < 0) {
 484         av_log(dst->avctx, AV_LOG_ERROR, "failed to allocate context "
 485                "scratch buffers.\n");
 486         return ret;
 487     }
 488     return 0;
 489 }
 490
 491 int ff_mpeg_update_thread_context(AVCodecContext *dst,
 492                                   const AVCodecContext *src)
 493 {
 494     int i, ret;
 495     MpegEncContext *s = dst->priv_data, *s1 = src->priv_data;
 496
 497     if (dst == src)
 498         return 0;
 499
 500     av_assert0(s != s1);
 501
 502     // FIXME can parameters change on I-frames?
 503     // in that case dst may need a reinit
 504     if (!s->context_initialized) {
 505         int err;
 506         memcpy(s, s1, sizeof(MpegEncContext));
 507
 508         s->avctx                 = dst;
 509         s->bitstream_buffer      = NULL;
 510         s->bitstream_buffer_size = s->allocated_bitstream_buffer_size = 0;
 511
 512         if (s1->context_initialized){
 513 //             s->picture_range_start  += MAX_PICTURE_COUNT;
 514 //             s->picture_range_end    += MAX_PICTURE_COUNT;
 515             ff_mpv_idct_init(s);
 516             if((err = ff_mpv_common_init(s)) < 0){
 517                 memset(s, 0, sizeof(MpegEncContext));
 518                 s->avctx = dst;
 519                 return err;
 520             }
 521         }
 522     }
 523
 524     if (s->height != s1->height || s->width != s1->width || s->context_reinit) {
 525         s->context_reinit = 0;
 526         s->height = s1->height;
 527         s->width  = s1->width;
 528         if ((ret = ff_mpv_common_frame_size_change(s)) < 0)
 529             return ret;
 530     }
 531
 532     s->avctx->coded_height  = s1->avctx->coded_height;
 533     s->avctx->coded_width   = s1->avctx->coded_width;
 534     s->avctx->width         = s1->avctx->width;
 535     s->avctx->height        = s1->avctx->height;
 536
 537     s->quarter_sample       = s1->quarter_sample;
 538
 539     s->coded_picture_number = s1->coded_picture_number;
 540     s->picture_number       = s1->picture_number;
 541
 542     av_assert0(!s->picture || s->picture != s1->picture);
 543     if(s->picture)
 544     for (i = 0; i < MAX_PICTURE_COUNT; i++) {
 545         ff_mpeg_unref_picture(s->avctx, &s->picture[i]);
 546         if (s1->picture && s1->picture[i].f->buf[0] &&
 547             (ret = ff_mpeg_ref_picture(s->avctx, &s->picture[i], &s1->picture[i])) < 0)
 548             return ret;
 549     }
 550
 551 #define UPDATE_PICTURE(pic)\
 552 do {\
 553     ff_mpeg_unref_picture(s->avctx, &s->pic);\
 554     if (s1->pic.f && s1->pic.f->buf[0])\
 555         ret = ff_mpeg_ref_picture(s->avctx, &s->pic, &s1->pic);\
 556     else\
 557         ret = ff_update_picture_tables(&s->pic, &s1->pic);\
 558     if (ret < 0)\
 559         return ret;\
 560 } while (0)
 561
 562     UPDATE_PICTURE(current_picture);
 563     UPDATE_PICTURE(last_picture);
 564     UPDATE_PICTURE(next_picture);
 565
 566 #define REBASE_PICTURE(pic, new_ctx, old_ctx)                                 \
 567     ((pic && pic >= old_ctx->picture &&                                       \
 568       pic < old_ctx->picture + MAX_PICTURE_COUNT) ?                           \
 569         &new_ctx->picture[pic - old_ctx->picture] : NULL)
 570
 571     s->last_picture_ptr    = REBASE_PICTURE(s1->last_picture_ptr,    s, s1);
 572     s->current_picture_ptr = REBASE_PICTURE(s1->current_picture_ptr, s, s1);
 573     s->next_picture_ptr    = REBASE_PICTURE(s1->next_picture_ptr,    s, s1);
 574
 575     // Error/bug resilience
 576     s->next_p_frame_damaged = s1->next_p_frame_damaged;
 577     s->workaround_bugs      = s1->workaround_bugs;
 578     s->padding_bug_score    = s1->padding_bug_score;
 579
 580     // MPEG-4 timing info
 581     memcpy(&s->last_time_base, &s1->last_time_base,
 582            (char *) &s1->pb_field_time + sizeof(s1->pb_field_time) -
 583            (char *) &s1->last_time_base);
 584
 585     // B-frame info
 586     s->max_b_frames = s1->max_b_frames;
 587     s->low_delay    = s1->low_delay;
 588     s->droppable    = s1->droppable;
 589
 590     // DivX handling (doesn't work)
 591     s->divx_packed  = s1->divx_packed;
 592
 593     if (s1->bitstream_buffer) {
 594         if (s1->bitstream_buffer_size +
 595             AV_INPUT_BUFFER_PADDING_SIZE > s->allocated_bitstream_buffer_size) {
 596             av_fast_malloc(&s->bitstream_buffer,
 597                            &s->allocated_bitstream_buffer_size,
 598                            s1->allocated_bitstream_buffer_size);
 599             if (!s->bitstream_buffer) {
 600                 s->bitstream_buffer_size = 0;
 601                 return AVERROR(ENOMEM);
 602             }
 603         }
 604         s->bitstream_buffer_size = s1->bitstream_buffer_size;
 605         memcpy(s->bitstream_buffer, s1->bitstream_buffer,
 606                s1->bitstream_buffer_size);
 607         memset(s->bitstream_buffer + s->bitstream_buffer_size, 0,
 608                AV_INPUT_BUFFER_PADDING_SIZE);
 609     }
 610
 611     // linesize-dependent scratch buffer allocation
 612     if (!s->sc.edge_emu_buffer)
 613         if (s1->linesize) {
 614             if (ff_mpeg_framesize_alloc(s->avctx, &s->me,
 615                                         &s->sc, s1->linesize) < 0) {
 616                 av_log(s->avctx, AV_LOG_ERROR, "Failed to allocate context "
 617                        "scratch buffers.\n");
 618                 return AVERROR(ENOMEM);
 619             }
 620         } else {
 621             av_log(s->avctx, AV_LOG_ERROR, "Context scratch buffers could not "
 622                    "be allocated due to unknown size.\n");
 623         }
 624
 625     // MPEG-2/interlacing info
 626     memcpy(&s->progressive_sequence, &s1->progressive_sequence,
 627            (char *) &s1->rtp_mode - (char *) &s1->progressive_sequence);
 628
 629     if (!s1->first_field) {
 630         s->last_pict_type = s1->pict_type;
 631         if (s1->current_picture_ptr)
 632             s->last_lambda_for[s1->pict_type] = s1->current_picture_ptr->f->quality;
 633     }
 634
 635     return 0;
 636 }
 637
 638 /**
 639  * Set the given MpegEncContext to common defaults
 640  * (same for encoding and decoding).
 641  * The changed fields will not depend upon the
 642  * prior state of the MpegEncContext.
 643  */
 644 void ff_mpv_common_defaults(MpegEncContext *s)
 645 {
 646     s->y_dc_scale_table      =
 647     s->c_dc_scale_table      = ff_mpeg1_dc_scale_table;
 648     s->chroma_qscale_table   = ff_default_chroma_qscale_table;
 649     s->progressive_frame     = 1;
 650     s->progressive_sequence  = 1;
 651     s->picture_structure     = PICT_FRAME;
 652
 653     s->coded_picture_number  = 0;
 654     s->picture_number        = 0;
 655
 656     s->f_code                = 1;
 657     s->b_code                = 1;
 658
 659     s->slice_context_count   = 1;
 660 }
 661
 662 /**
 663  * Set the given MpegEncContext to defaults for decoding.
 664  * the changed fields will not depend upon
 665  * the prior state of the MpegEncContext.
 666  */
 667 void ff_mpv_decode_defaults(MpegEncContext *s)
 668 {
 669     ff_mpv_common_defaults(s);
 670 }
 671
 672 void ff_mpv_decode_init(MpegEncContext *s, AVCodecContext *avctx)
 673 {
 674     s->avctx           = avctx;
 675     s->width           = avctx->coded_width;
 676     s->height          = avctx->coded_height;
 677     s->codec_id        = avctx->codec->id;
 678     s->workaround_bugs = avctx->workaround_bugs;
 679
 680     /* convert fourcc to upper case */
 681     s->codec_tag          = avpriv_toupper4(avctx->codec_tag);
 682 }
 683
 684 /**
 685  * Initialize and allocates MpegEncContext fields dependent on the resolution.
 686  */
 687 static int init_context_frame(MpegEncContext *s)
 688 {
 689     int y_size, c_size, yc_size, i, mb_array_size, mv_table_size, x, y;
 690
 691     s->mb_width   = (s->width + 15) / 16;
 692     s->mb_stride  = s->mb_width + 1;
 693     s->b8_stride  = s->mb_width * 2 + 1;
 694     mb_array_size = s->mb_height * s->mb_stride;
 695     mv_table_size = (s->mb_height + 2) * s->mb_stride + 1;
 696
 697     /* set default edge pos, will be overridden
 698      * in decode_header if needed */
 699     s->h_edge_pos = s->mb_width * 16;
 700     s->v_edge_pos = s->mb_height * 16;
 701
 702     s->mb_num     = s->mb_width * s->mb_height;
 703
 704     s->block_wrap[0] =
 705     s->block_wrap[1] =
 706     s->block_wrap[2] =
 707     s->block_wrap[3] = s->b8_stride;
 708     s->block_wrap[4] =
 709     s->block_wrap[5] = s->mb_stride;
 710
 711     y_size  = s->b8_stride * (2 * s->mb_height + 1);
 712     c_size  = s->mb_stride * (s->mb_height + 1);
 713     yc_size = y_size + 2   * c_size;
 714
 715     if (s->mb_height & 1)
 716         yc_size += 2*s->b8_stride + 2*s->mb_stride;
 717
 718     FF_ALLOCZ_OR_GOTO(s->avctx, s->mb_index2xy, (s->mb_num + 1) * sizeof(int),
 719                       fail); // error resilience code looks cleaner with this
 720     for (y = 0; y < s->mb_height; y++)
 721         for (x = 0; x < s->mb_width; x++)
 722             s->mb_index2xy[x + y * s->mb_width] = x + y * s->mb_stride;
 723
 724     s->mb_index2xy[s->mb_height * s->mb_width] = (s->mb_height - 1) * s->mb_stride + s->mb_width; // FIXME really needed?
 725
 726     if (s->encoding) {
 727         /* Allocate MV tables */
 728         FF_ALLOCZ_OR_GOTO(s->avctx, s->p_mv_table_base,                 mv_table_size * 2 * sizeof(int16_t), fail)
 729         FF_ALLOCZ_OR_GOTO(s->avctx, s->b_forw_mv_table_base,            mv_table_size * 2 * sizeof(int16_t), fail)
 730         FF_ALLOCZ_OR_GOTO(s->avctx, s->b_back_mv_table_base,            mv_table_size * 2 * sizeof(int16_t), fail)
 731         FF_ALLOCZ_OR_GOTO(s->avctx, s->b_bidir_forw_mv_table_base,      mv_table_size * 2 * sizeof(int16_t), fail)
 732         FF_ALLOCZ_OR_GOTO(s->avctx, s->b_bidir_back_mv_table_base,      mv_table_size * 2 * sizeof(int16_t), fail)
 733         FF_ALLOCZ_OR_GOTO(s->avctx, s->b_direct_mv_table_base,          mv_table_size * 2 * sizeof(int16_t), fail)
 734         s->p_mv_table            = s->p_mv_table_base + s->mb_stride + 1;
 735         s->b_forw_mv_table       = s->b_forw_mv_table_base + s->mb_stride + 1;
 736         s->b_back_mv_table       = s->b_back_mv_table_base + s->mb_stride + 1;
 737         s->b_bidir_forw_mv_table = s->b_bidir_forw_mv_table_base + s->mb_stride + 1;
 738         s->b_bidir_back_mv_table = s->b_bidir_back_mv_table_base + s->mb_stride + 1;
 739         s->b_direct_mv_table     = s->b_direct_mv_table_base + s->mb_stride + 1;
 740
 741         /* Allocate MB type table */
 742         FF_ALLOCZ_OR_GOTO(s->avctx, s->mb_type, mb_array_size * sizeof(uint16_t), fail) // needed for encoding
 743
 744         FF_ALLOCZ_OR_GOTO(s->avctx, s->lambda_table, mb_array_size * sizeof(int), fail)
 745
 746         FF_ALLOC_OR_GOTO(s->avctx, s->cplx_tab,
 747                          mb_array_size * sizeof(float), fail);
 748         FF_ALLOC_OR_GOTO(s->avctx, s->bits_tab,
 749                          mb_array_size * sizeof(float), fail);
 750
 751     }
 752
 753     if (s->codec_id == AV_CODEC_ID_MPEG4 ||
 754         (s->avctx->flags & AV_CODEC_FLAG_INTERLACED_ME)) {
 755         /* interlaced direct mode decoding tables */
 756         for (i = 0; i < 2; i++) {
 757             int j, k;
 758             for (j = 0; j < 2; j++) {
 759                 for (k = 0; k < 2; k++) {
 760                     FF_ALLOCZ_OR_GOTO(s->avctx,
 761                                       s->b_field_mv_table_base[i][j][k],
 762                                       mv_table_size * 2 * sizeof(int16_t),
 763                                       fail);
 764                     s->b_field_mv_table[i][j][k] = s->b_field_mv_table_base[i][j][k] +
 765                                                    s->mb_stride + 1;
 766                 }
 767                 FF_ALLOCZ_OR_GOTO(s->avctx, s->b_field_select_table [i][j], mb_array_size * 2 * sizeof(uint8_t), fail)
 768                 FF_ALLOCZ_OR_GOTO(s->avctx, s->p_field_mv_table_base[i][j], mv_table_size * 2 * sizeof(int16_t), fail)
 769                 s->p_field_mv_table[i][j] = s->p_field_mv_table_base[i][j] + s->mb_stride + 1;
 770             }
 771             FF_ALLOCZ_OR_GOTO(s->avctx, s->p_field_select_table[i], mb_array_size * 2 * sizeof(uint8_t), fail)
 772         }
 773     }
 774     if (s->out_format == FMT_H263) {
 775         /* cbp values */
 776         FF_ALLOCZ_OR_GOTO(s->avctx, s->coded_block_base, y_size + (s->mb_height&1)*2*s->b8_stride, fail);
 777         s->coded_block = s->coded_block_base + s->b8_stride + 1;
 778
 779         /* cbp, ac_pred, pred_dir */
 780         FF_ALLOCZ_OR_GOTO(s->avctx, s->cbp_table     , mb_array_size * sizeof(uint8_t), fail);
 781         FF_ALLOCZ_OR_GOTO(s->avctx, s->pred_dir_table, mb_array_size * sizeof(uint8_t), fail);
 782     }
 783
 784     if (s->h263_pred || s->h263_plus || !s->encoding) {
 785         /* dc values */
 786         // MN: we need these for error resilience of intra-frames
 787         FF_ALLOCZ_OR_GOTO(s->avctx, s->dc_val_base, yc_size * sizeof(int16_t), fail);
 788         s->dc_val[0] = s->dc_val_base + s->b8_stride + 1;
 789         s->dc_val[1] = s->dc_val_base + y_size + s->mb_stride + 1;
 790         s->dc_val[2] = s->dc_val[1] + c_size;
 791         for (i = 0; i < yc_size; i++)
 792             s->dc_val_base[i] = 1024;
 793     }
 794
 795     /* which mb is an intra block */
 796     FF_ALLOCZ_OR_GOTO(s->avctx, s->mbintra_table, mb_array_size, fail);
 797     memset(s->mbintra_table, 1, mb_array_size);
 798
 799     /* init macroblock skip table */
 800     FF_ALLOCZ_OR_GOTO(s->avctx, s->mbskip_table, mb_array_size + 2, fail);
 801     // Note the + 1 is for a quicker MPEG-4 slice_end detection
 802
 803     return ff_mpeg_er_init(s);
 804 fail:
 805     return AVERROR(ENOMEM);
 806 }
 807
 808 static void clear_context(MpegEncContext *s)
 809 {
 810     int i, j, k;
 811
 812     memset(&s->next_picture, 0, sizeof(s->next_picture));
 813     memset(&s->last_picture, 0, sizeof(s->last_picture));
 814     memset(&s->current_picture, 0, sizeof(s->current_picture));
 815     memset(&s->new_picture, 0, sizeof(s->new_picture));
 816
 817     memset(s->thread_context, 0, sizeof(s->thread_context));
 818
 819     s->me.map = NULL;
 820     s->me.score_map = NULL;
 821     s->dct_error_sum = NULL;
 822     s->block = NULL;
 823     s->blocks = NULL;
 824     s->block32 = NULL;
 825     memset(s->pblocks, 0, sizeof(s->pblocks));
 826     s->dpcm_direction = 0;
 827     s->dpcm_macroblock = NULL;
 828     s->ac_val_base = NULL;
 829     s->ac_val[0] =
 830     s->ac_val[1] =
 831     s->ac_val[2] =NULL;
 832     s->sc.edge_emu_buffer = NULL;
 833     s->me.scratchpad = NULL;
 834     s->me.temp =
 835     s->sc.rd_scratchpad =
 836     s->sc.b_scratchpad =
 837     s->sc.obmc_scratchpad = NULL;
 838
 839
 840     s->bitstream_buffer = NULL;
 841     s->allocated_bitstream_buffer_size = 0;
 842     s->picture          = NULL;
 843     s->mb_type          = NULL;
 844     s->p_mv_table_base  = NULL;
 845     s->b_forw_mv_table_base = NULL;
 846     s->b_back_mv_table_base = NULL;
 847     s->b_bidir_forw_mv_table_base = NULL;
 848     s->b_bidir_back_mv_table_base = NULL;
 849     s->b_direct_mv_table_base = NULL;
 850     s->p_mv_table            = NULL;
 851     s->b_forw_mv_table       = NULL;
 852     s->b_back_mv_table       = NULL;
 853     s->b_bidir_forw_mv_table = NULL;
 854     s->b_bidir_back_mv_table = NULL;
 855     s->b_direct_mv_table     = NULL;
 856     for (i = 0; i < 2; i++) {
 857         for (j = 0; j < 2; j++) {
 858             for (k = 0; k < 2; k++) {
 859                 s->b_field_mv_table_base[i][j][k] = NULL;
 860                 s->b_field_mv_table[i][j][k] = NULL;
 861             }
 862             s->b_field_select_table[i][j] = NULL;
 863             s->p_field_mv_table_base[i][j] = NULL;
 864             s->p_field_mv_table[i][j] = NULL;
 865         }
 866         s->p_field_select_table[i] = NULL;
 867     }
 868
 869     s->dc_val_base = NULL;
 870     s->coded_block_base = NULL;
 871     s->mbintra_table = NULL;
 872     s->cbp_table = NULL;
 873     s->pred_dir_table = NULL;
 874
 875     s->mbskip_table = NULL;
 876
 877     s->er.error_status_table = NULL;
 878     s->er.er_temp_buffer = NULL;
 879     s->mb_index2xy = NULL;
 880     s->lambda_table = NULL;
 881
 882     s->cplx_tab = NULL;
 883     s->bits_tab = NULL;
 884 }
 885
 886 /**
 887  * init common structure for both encoder and decoder.
 888  * this assumes that some variables like width/height are already set
 889  */
 890 av_cold int ff_mpv_common_init(MpegEncContext *s)
 891 {
 892     int i, ret;
 893     int nb_slices = (HAVE_THREADS &&
 894                      s->avctx->active_thread_type & FF_THREAD_SLICE) ?
 895                     s->avctx->thread_count : 1;
 896
 897     clear_context(s);
 898
 899     if (s->encoding && s->avctx->slices)
 900         nb_slices = s->avctx->slices;
 901
 902     if (s->codec_id == AV_CODEC_ID_MPEG2VIDEO && !s->progressive_sequence)
 903         s->mb_height = (s->height + 31) / 32 * 2;
 904     else
 905         s->mb_height = (s->height + 15) / 16;
 906
 907     if (s->avctx->pix_fmt == AV_PIX_FMT_NONE) {
 908         av_log(s->avctx, AV_LOG_ERROR,
 909                "decoding to AV_PIX_FMT_NONE is not supported.\n");
 910         return -1;
 911     }
 912
 913     if (nb_slices > MAX_THREADS || (nb_slices > s->mb_height && s->mb_height)) {
 914         int max_slices;
 915         if (s->mb_height)
 916             max_slices = FFMIN(MAX_THREADS, s->mb_height);
 917         else
 918             max_slices = MAX_THREADS;
 919         av_log(s->avctx, AV_LOG_WARNING, "too many threads/slices (%d),"
 920                " reducing to %d\n", nb_slices, max_slices);
 921         nb_slices = max_slices;
 922     }
 923
 924     if ((s->width || s->height) &&
 925         av_image_check_size(s->width, s->height, 0, s->avctx))
 926         return -1;
 927
 928     dct_init(s);
 929
 930     /* set chroma shifts */
 931     ret = av_pix_fmt_get_chroma_sub_sample(s->avctx->pix_fmt,
 932                                            &s->chroma_x_shift,
 933                                            &s->chroma_y_shift);
 934     if (ret)
 935         return ret;
 936
 937     FF_ALLOCZ_OR_GOTO(s->avctx, s->picture,
 938                       MAX_PICTURE_COUNT * sizeof(Picture), fail);
 939     for (i = 0; i < MAX_PICTURE_COUNT; i++) {
 940         s->picture[i].f = av_frame_alloc();
 941         if (!s->picture[i].f)
 942             goto fail;
 943     }
 944     s->next_picture.f = av_frame_alloc();
 945     if (!s->next_picture.f)
 946         goto fail;
 947     s->last_picture.f = av_frame_alloc();
 948     if (!s->last_picture.f)
 949         goto fail;
 950     s->current_picture.f = av_frame_alloc();
 951     if (!s->current_picture.f)
 952         goto fail;
 953     s->new_picture.f = av_frame_alloc();
 954     if (!s->new_picture.f)
 955         goto fail;
 956
 957     if (init_context_frame(s))
 958         goto fail;
 959
 960     s->parse_context.state = -1;
 961
 962     s->context_initialized = 1;
 963     memset(s->thread_context, 0, sizeof(s->thread_context));
 964     s->thread_context[0]   = s;
 965
 966 //     if (s->width && s->height) {
 967     if (nb_slices > 1) {
 968         for (i = 0; i < nb_slices; i++) {
 969             if (i) {
 970                 s->thread_context[i] = av_memdup(s, sizeof(MpegEncContext));
 971                 if (!s->thread_context[i])
 972                     goto fail;
 973             }
 974             if (init_duplicate_context(s->thread_context[i]) < 0)
 975                 goto fail;
 976             s->thread_context[i]->start_mb_y =
 977                 (s->mb_height * (i) + nb_slices / 2) / nb_slices;
 978             s->thread_context[i]->end_mb_y   =
 979                 (s->mb_height * (i + 1) + nb_slices / 2) / nb_slices;
 980         }
 981     } else {
 982         if (init_duplicate_context(s) < 0)
 983             goto fail;
 984         s->start_mb_y = 0;
 985         s->end_mb_y   = s->mb_height;
 986     }
 987     s->slice_context_count = nb_slices;
 988 //     }
 989
 990     return 0;
 991  fail:
 992     ff_mpv_common_end(s);
 993     return -1;
 994 }
 995
 996 /**
 997  * Frees and resets MpegEncContext fields depending on the resolution.
 998  * Is used during resolution changes to avoid a full reinitialization of the
 999  * codec.
1000  */
1001 static void free_context_frame(MpegEncContext *s)
1002 {
1003     int i, j, k;
1004
1005     av_freep(&s->mb_type);
1006     av_freep(&s->p_mv_table_base);
1007     av_freep(&s->b_forw_mv_table_base);
1008     av_freep(&s->b_back_mv_table_base);
1009     av_freep(&s->b_bidir_forw_mv_table_base);
1010     av_freep(&s->b_bidir_back_mv_table_base);
1011     av_freep(&s->b_direct_mv_table_base);
1012     s->p_mv_table            = NULL;
1013     s->b_forw_mv_table       = NULL;
1014     s->b_back_mv_table       = NULL;
1015     s->b_bidir_forw_mv_table = NULL;
1016     s->b_bidir_back_mv_table = NULL;
1017     s->b_direct_mv_table     = NULL;
1018     for (i = 0; i < 2; i++) {
1019         for (j = 0; j < 2; j++) {
1020             for (k = 0; k < 2; k++) {
1021                 av_freep(&s->b_field_mv_table_base[i][j][k]);
1022                 s->b_field_mv_table[i][j][k] = NULL;
1023             }
1024             av_freep(&s->b_field_select_table[i][j]);
1025             av_freep(&s->p_field_mv_table_base[i][j]);
1026             s->p_field_mv_table[i][j] = NULL;
1027         }
1028         av_freep(&s->p_field_select_table[i]);
1029     }
1030
1031     av_freep(&s->dc_val_base);
1032     av_freep(&s->coded_block_base);
1033     av_freep(&s->mbintra_table);
1034     av_freep(&s->cbp_table);
1035     av_freep(&s->pred_dir_table);
1036
1037     av_freep(&s->mbskip_table);
1038
1039     av_freep(&s->er.error_status_table);
1040     av_freep(&s->er.er_temp_buffer);
1041     av_freep(&s->mb_index2xy);
1042     av_freep(&s->lambda_table);
1043
1044     av_freep(&s->cplx_tab);
1045     av_freep(&s->bits_tab);
1046
1047     s->linesize = s->uvlinesize = 0;
1048 }
1049
1050 int ff_mpv_common_frame_size_change(MpegEncContext *s)
1051 {
1052     int i, err = 0;
1053
1054     if (!s->context_initialized)
1055         return AVERROR(EINVAL);
1056
1057     if (s->slice_context_count > 1) {
1058         for (i = 0; i < s->slice_context_count; i++) {
1059             free_duplicate_context(s->thread_context[i]);
1060         }
1061         for (i = 1; i < s->slice_context_count; i++) {
1062             av_freep(&s->thread_context[i]);
1063         }
1064     } else
1065         free_duplicate_context(s);
1066
1067     free_context_frame(s);
1068
1069     if (s->picture)
1070         for (i = 0; i < MAX_PICTURE_COUNT; i++) {
1071                 s->picture[i].needs_realloc = 1;
1072         }
1073
1074     s->last_picture_ptr         =
1075     s->next_picture_ptr         =
1076     s->current_picture_ptr      = NULL;
1077
1078     // init
1079     if (s->codec_id == AV_CODEC_ID_MPEG2VIDEO && !s->progressive_sequence)
1080         s->mb_height = (s->height + 31) / 32 * 2;
1081     else
1082         s->mb_height = (s->height + 15) / 16;
1083
1084     if ((s->width || s->height) &&
1085         (err = av_image_check_size(s->width, s->height, 0, s->avctx)) < 0)
1086         goto fail;
1087
1088     if ((err = init_context_frame(s)))
1089         goto fail;
1090
1091     memset(s->thread_context, 0, sizeof(s->thread_context));
1092     s->thread_context[0]   = s;
1093
1094     if (s->width && s->height) {
1095         int nb_slices = s->slice_context_count;
1096         if (nb_slices > 1) {
1097             for (i = 0; i < nb_slices; i++) {
1098                 if (i) {
1099                     s->thread_context[i] = av_memdup(s, sizeof(MpegEncContext));
1100                     if (!s->thread_context[i]) {
1101                         err = AVERROR(ENOMEM);
1102                         goto fail;
1103                     }
1104                 }
1105                 if ((err = init_duplicate_context(s->thread_context[i])) < 0)
1106                     goto fail;
1107                 s->thread_context[i]->start_mb_y =
1108                     (s->mb_height * (i) + nb_slices / 2) / nb_slices;
1109                 s->thread_context[i]->end_mb_y   =
1110                     (s->mb_height * (i + 1) + nb_slices / 2) / nb_slices;
1111             }
1112         } else {
1113             err = init_duplicate_context(s);
1114             if (err < 0)
1115                 goto fail;
1116             s->start_mb_y = 0;
1117             s->end_mb_y   = s->mb_height;
1118         }
1119         s->slice_context_count = nb_slices;
1120     }
1121
1122     return 0;
1123  fail:
1124     ff_mpv_common_end(s);
1125     return err;
1126 }
1127
1128 /* init common structure for both encoder and decoder */
1129 void ff_mpv_common_end(MpegEncContext *s)
1130 {
1131     int i;
1132
1133     if (!s)
1134         return ;
1135
1136     if (s->slice_context_count > 1) {
1137         for (i = 0; i < s->slice_context_count; i++) {
1138             free_duplicate_context(s->thread_context[i]);
1139         }
1140         for (i = 1; i < s->slice_context_count; i++) {
1141             av_freep(&s->thread_context[i]);
1142         }
1143         s->slice_context_count = 1;
1144     } else free_duplicate_context(s);
1145
1146     av_freep(&s->parse_context.buffer);
1147     s->parse_context.buffer_size = 0;
1148
1149     av_freep(&s->bitstream_buffer);
1150     s->allocated_bitstream_buffer_size = 0;
1151
1152     if (s->picture) {
1153         for (i = 0; i < MAX_PICTURE_COUNT; i++) {
1154             ff_free_picture_tables(&s->picture[i]);
1155             ff_mpeg_unref_picture(s->avctx, &s->picture[i]);
1156             av_frame_free(&s->picture[i].f);
1157         }
1158     }
1159     av_freep(&s->picture);
1160     ff_free_picture_tables(&s->last_picture);
1161     ff_mpeg_unref_picture(s->avctx, &s->last_picture);
1162     av_frame_free(&s->last_picture.f);
1163     ff_free_picture_tables(&s->current_picture);
1164     ff_mpeg_unref_picture(s->avctx, &s->current_picture);
1165     av_frame_free(&s->current_picture.f);
1166     ff_free_picture_tables(&s->next_picture);
1167     ff_mpeg_unref_picture(s->avctx, &s->next_picture);
1168     av_frame_free(&s->next_picture.f);
1169     ff_free_picture_tables(&s->new_picture);
1170     ff_mpeg_unref_picture(s->avctx, &s->new_picture);
1171     av_frame_free(&s->new_picture.f);
1172
1173     free_context_frame(s);
1174
1175     s->context_initialized      = 0;
1176     s->last_picture_ptr         =
1177     s->next_picture_ptr         =
1178     s->current_picture_ptr      = NULL;
1179     s->linesize = s->uvlinesize = 0;
1180 }
1181
1182
1183 static void gray_frame(AVFrame *frame)
1184 {
1185     int i, h_chroma_shift, v_chroma_shift;
1186
1187     av_pix_fmt_get_chroma_sub_sample(frame->format, &h_chroma_shift, &v_chroma_shift);
1188
1189     for(i=0; i<frame->height; i++)
1190         memset(frame->data[0] + frame->linesize[0]*i, 0x80, frame->width);
1191     for(i=0; i<AV_CEIL_RSHIFT(frame->height, v_chroma_shift); i++) {
1192         memset(frame->data[1] + frame->linesize[1]*i,
1193                0x80, AV_CEIL_RSHIFT(frame->width, h_chroma_shift));
1194         memset(frame->data[2] + frame->linesize[2]*i,
1195                0x80, AV_CEIL_RSHIFT(frame->width, h_chroma_shift));
1196     }
1197 }
1198
1199 /**
1200  * generic function called after decoding
1201  * the header and before a frame is decoded.
1202  */
1203 int ff_mpv_frame_start(MpegEncContext *s, AVCodecContext *avctx)
1204 {
1205     int i, ret;
1206     Picture *pic;
1207     s->mb_skipped = 0;
1208
1209     if (!ff_thread_can_start_frame(avctx)) {
1210         av_log(avctx, AV_LOG_ERROR, "Attempt to start a frame outside SETUP state\n");
1211         return -1;
1212     }
1213
1214     /* mark & release old frames */
1215     if (s->pict_type != AV_PICTURE_TYPE_B && s->last_picture_ptr &&
1216         s->last_picture_ptr != s->next_picture_ptr &&
1217         s->last_picture_ptr->f->buf[0]) {
1218         ff_mpeg_unref_picture(s->avctx, s->last_picture_ptr);
1219     }
1220
1221     /* release forgotten pictures */
1222     /* if (MPEG-124 / H.263) */
1223     for (i = 0; i < MAX_PICTURE_COUNT; i++) {
1224         if (&s->picture[i] != s->last_picture_ptr &&
1225             &s->picture[i] != s->next_picture_ptr &&
1226             s->picture[i].reference && !s->picture[i].needs_realloc) {
1227             ff_mpeg_unref_picture(s->avctx, &s->picture[i]);
1228         }
1229     }
1230
1231     ff_mpeg_unref_picture(s->avctx, &s->current_picture);
1232     ff_mpeg_unref_picture(s->avctx, &s->last_picture);
1233     ff_mpeg_unref_picture(s->avctx, &s->next_picture);
1234
1235     /* release non reference frames */
1236     for (i = 0; i < MAX_PICTURE_COUNT; i++) {
1237         if (!s->picture[i].reference)
1238             ff_mpeg_unref_picture(s->avctx, &s->picture[i]);
1239     }
1240
1241     if (s->current_picture_ptr && !s->current_picture_ptr->f->buf[0]) {
1242         // we already have an unused image
1243         // (maybe it was set before reading the header)
1244         pic = s->current_picture_ptr;
1245     } else {
1246         i   = ff_find_unused_picture(s->avctx, s->picture, 0);
1247         if (i < 0) {
1248             av_log(s->avctx, AV_LOG_ERROR, "no frame buffer available\n");
1249             return i;
1250         }
1251         pic = &s->picture[i];
1252     }
1253
1254     pic->reference = 0;
1255     if (!s->droppable) {
1256         if (s->pict_type != AV_PICTURE_TYPE_B)
1257             pic->reference = 3;
1258     }
1259
1260     pic->f->coded_picture_number = s->coded_picture_number++;
1261
1262     if (alloc_picture(s, pic, 0) < 0)
1263         return -1;
1264
1265     s->current_picture_ptr = pic;
1266     // FIXME use only the vars from current_pic
1267     s->current_picture_ptr->f->top_field_first = s->top_field_first;
1268     if (s->codec_id == AV_CODEC_ID_MPEG1VIDEO ||
1269         s->codec_id == AV_CODEC_ID_MPEG2VIDEO) {
1270         if (s->picture_structure != PICT_FRAME)
1271             s->current_picture_ptr->f->top_field_first =
1272                 (s->picture_structure == PICT_TOP_FIELD) == s->first_field;
1273     }
1274     s->current_picture_ptr->f->interlaced_frame = !s->progressive_frame &&
1275                                                  !s->progressive_sequence;
1276     s->current_picture_ptr->field_picture      =  s->picture_structure != PICT_FRAME;
1277
1278     s->current_picture_ptr->f->pict_type = s->pict_type;
1279     // if (s->avctx->flags && AV_CODEC_FLAG_QSCALE)
1280     //     s->current_picture_ptr->quality = s->new_picture_ptr->quality;
1281     s->current_picture_ptr->f->key_frame = s->pict_type == AV_PICTURE_TYPE_I;
1282
1283     if ((ret = ff_mpeg_ref_picture(s->avctx, &s->current_picture,
1284                                    s->current_picture_ptr)) < 0)
1285         return ret;
1286
1287     if (s->pict_type != AV_PICTURE_TYPE_B) {
1288         s->last_picture_ptr = s->next_picture_ptr;
1289         if (!s->droppable)
1290             s->next_picture_ptr = s->current_picture_ptr;
1291     }
1292     ff_dlog(s->avctx, "L%p N%p C%p L%p N%p C%p type:%d drop:%d\n",
1293             s->last_picture_ptr, s->next_picture_ptr,s->current_picture_ptr,
1294             s->last_picture_ptr    ? s->last_picture_ptr->f->data[0]    : NULL,
1295             s->next_picture_ptr    ? s->next_picture_ptr->f->data[0]    : NULL,
1296             s->current_picture_ptr ? s->current_picture_ptr->f->data[0] : NULL,
1297             s->pict_type, s->droppable);
1298
1299     if ((!s->last_picture_ptr || !s->last_picture_ptr->f->buf[0]) &&
1300         (s->pict_type != AV_PICTURE_TYPE_I)) {
1301         int h_chroma_shift, v_chroma_shift;
1302         av_pix_fmt_get_chroma_sub_sample(s->avctx->pix_fmt,
1303                                          &h_chroma_shift, &v_chroma_shift);
1304         if (s->pict_type == AV_PICTURE_TYPE_B && s->next_picture_ptr && s->next_picture_ptr->f->buf[0])
1305             av_log(avctx, AV_LOG_DEBUG,
1306                    "allocating dummy last picture for B frame\n");
1307         else if (s->pict_type != AV_PICTURE_TYPE_I)
1308             av_log(avctx, AV_LOG_ERROR,
1309                    "warning: first frame is no keyframe\n");
1310
1311         /* Allocate a dummy frame */
1312         i = ff_find_unused_picture(s->avctx, s->picture, 0);
1313         if (i < 0) {
1314             av_log(s->avctx, AV_LOG_ERROR, "no frame buffer available\n");
1315             return i;
1316         }
1317         s->last_picture_ptr = &s->picture[i];
1318
1319         s->last_picture_ptr->reference   = 3;
1320         s->last_picture_ptr->f->key_frame = 0;
1321         s->last_picture_ptr->f->pict_type = AV_PICTURE_TYPE_P;
1322
1323         if (alloc_picture(s, s->last_picture_ptr, 0) < 0) {
1324             s->last_picture_ptr = NULL;
1325             return -1;
1326         }
1327
1328         if (!avctx->hwaccel) {
1329             for(i=0; i<avctx->height; i++)
1330                 memset(s->last_picture_ptr->f->data[0] + s->last_picture_ptr->f->linesize[0]*i,
1331                        0x80, avctx->width);
1332             if (s->last_picture_ptr->f->data[2]) {
1333                 for(i=0; i<AV_CEIL_RSHIFT(avctx->height, v_chroma_shift); i++) {
1334                     memset(s->last_picture_ptr->f->data[1] + s->last_picture_ptr->f->linesize[1]*i,
1335                         0x80, AV_CEIL_RSHIFT(avctx->width, h_chroma_shift));
1336                     memset(s->last_picture_ptr->f->data[2] + s->last_picture_ptr->f->linesize[2]*i,
1337                         0x80, AV_CEIL_RSHIFT(avctx->width, h_chroma_shift));
1338                 }
1339             }
1340
1341             if(s->codec_id == AV_CODEC_ID_FLV1 || s->codec_id == AV_CODEC_ID_H263){
1342                 for(i=0; i<avctx->height; i++)
1343                 memset(s->last_picture_ptr->f->data[0] + s->last_picture_ptr->f->linesize[0]*i, 16, avctx->width);
1344             }
1345         }
1346
1347         ff_thread_report_progress(&s->last_picture_ptr->tf, INT_MAX, 0);
1348         ff_thread_report_progress(&s->last_picture_ptr->tf, INT_MAX, 1);
1349     }
1350     if ((!s->next_picture_ptr || !s->next_picture_ptr->f->buf[0]) &&
1351         s->pict_type == AV_PICTURE_TYPE_B) {
1352         /* Allocate a dummy frame */
1353         i = ff_find_unused_picture(s->avctx, s->picture, 0);
1354         if (i < 0) {
1355             av_log(s->avctx, AV_LOG_ERROR, "no frame buffer available\n");
1356             return i;
1357         }
1358         s->next_picture_ptr = &s->picture[i];
1359
1360         s->next_picture_ptr->reference   = 3;
1361         s->next_picture_ptr->f->key_frame = 0;
1362         s->next_picture_ptr->f->pict_type = AV_PICTURE_TYPE_P;
1363
1364         if (alloc_picture(s, s->next_picture_ptr, 0) < 0) {
1365             s->next_picture_ptr = NULL;
1366             return -1;
1367         }
1368         ff_thread_report_progress(&s->next_picture_ptr->tf, INT_MAX, 0);
1369         ff_thread_report_progress(&s->next_picture_ptr->tf, INT_MAX, 1);
1370     }
1371
1372 #if 0 // BUFREF-FIXME
1373     memset(s->last_picture.f->data, 0, sizeof(s->last_picture.f->data));
1374     memset(s->next_picture.f->data, 0, sizeof(s->next_picture.f->data));
1375 #endif
1376     if (s->last_picture_ptr) {
1377         if (s->last_picture_ptr->f->buf[0] &&
1378             (ret = ff_mpeg_ref_picture(s->avctx, &s->last_picture,
1379                                        s->last_picture_ptr)) < 0)
1380             return ret;
1381     }
1382     if (s->next_picture_ptr) {
1383         if (s->next_picture_ptr->f->buf[0] &&
1384             (ret = ff_mpeg_ref_picture(s->avctx, &s->next_picture,
1385                                        s->next_picture_ptr)) < 0)
1386             return ret;
1387     }
1388
1389     av_assert0(s->pict_type == AV_PICTURE_TYPE_I || (s->last_picture_ptr &&
1390                                                  s->last_picture_ptr->f->buf[0]));
1391
1392     if (s->picture_structure!= PICT_FRAME) {
1393         int i;
1394         for (i = 0; i < 4; i++) {
1395             if (s->picture_structure == PICT_BOTTOM_FIELD) {
1396                 s->current_picture.f->data[i] +=
1397                     s->current_picture.f->linesize[i];
1398             }
1399             s->current_picture.f->linesize[i] *= 2;
1400             s->last_picture.f->linesize[i]    *= 2;
1401             s->next_picture.f->linesize[i]    *= 2;
1402         }
1403     }
1404
1405     /* set dequantizer, we can't do it during init as
1406      * it might change for MPEG-4 and we can't do it in the header
1407      * decode as init is not called for MPEG-4 there yet */
1408     if (s->mpeg_quant || s->codec_id == AV_CODEC_ID_MPEG2VIDEO) {
1409         s->dct_unquantize_intra = s->dct_unquantize_mpeg2_intra;
1410         s->dct_unquantize_inter = s->dct_unquantize_mpeg2_inter;
1411     } else if (s->out_format == FMT_H263 || s->out_format == FMT_H261) {
1412         s->dct_unquantize_intra = s->dct_unquantize_h263_intra;
1413         s->dct_unquantize_inter = s->dct_unquantize_h263_inter;
1414     } else {
1415         s->dct_unquantize_intra = s->dct_unquantize_mpeg1_intra;
1416         s->dct_unquantize_inter = s->dct_unquantize_mpeg1_inter;
1417     }
1418
1419     if (s->avctx->debug & FF_DEBUG_NOMC) {
1420         gray_frame(s->current_picture_ptr->f);
1421     }
1422
1423     return 0;
1424 }
1425
1426 /* called after a frame has been decoded. */
1427 void ff_mpv_frame_end(MpegEncContext *s)
1428 {
1429     emms_c();
1430
1431     if (s->current_picture.reference)
1432         ff_thread_report_progress(&s->current_picture_ptr->tf, INT_MAX, 0);
1433 }
1434
1435 void ff_print_debug_info(MpegEncContext *s, Picture *p, AVFrame *pict)
1436 {
1437     ff_print_debug_info2(s->avctx, pict, s->mbskip_table, p->mb_type,
1438                          p->qscale_table, p->motion_val, &s->low_delay,
1439                          s->mb_width, s->mb_height, s->mb_stride, s->quarter_sample);
1440 }
1441
1442 int ff_mpv_export_qp_table(MpegEncContext *s, AVFrame *f, Picture *p, int qp_type)
1443 {
1444     AVBufferRef *ref = av_buffer_ref(p->qscale_table_buf);
1445     int offset = 2*s->mb_stride + 1;
1446     if(!ref)
1447         return AVERROR(ENOMEM);
1448     av_assert0(ref->size >= offset + s->mb_stride * ((f->height+15)/16));
1449     ref->size -= offset;
1450     ref->data += offset;
1451     return av_frame_set_qp_table(f, ref, s->mb_stride, qp_type);
1452 }
1453
1454 static inline int hpel_motion_lowres(MpegEncContext *s,
1455                                      uint8_t *dest, uint8_t *src,
1456                                      int field_based, int field_select,
1457                                      int src_x, int src_y,
1458                                      int width, int height, ptrdiff_t stride,
1459                                      int h_edge_pos, int v_edge_pos,
1460                                      int w, int h, h264_chroma_mc_func *pix_op,
1461                                      int motion_x, int motion_y)
1462 {
1463     const int lowres   = s->avctx->lowres;
1464     const int op_index = FFMIN(lowres, 3);
1465     const int s_mask   = (2 << lowres) - 1;
1466     int emu = 0;
1467     int sx, sy;
1468
1469     if (s->quarter_sample) {
1470         motion_x /= 2;
1471         motion_y /= 2;
1472     }
1473
1474     sx = motion_x & s_mask;
1475     sy = motion_y & s_mask;
1476     src_x += motion_x >> lowres + 1;
1477     src_y += motion_y >> lowres + 1;
1478
1479     src   += src_y * stride + src_x;
1480
1481     if ((unsigned)src_x > FFMAX( h_edge_pos - (!!sx) - w,                 0) ||
1482         (unsigned)src_y > FFMAX((v_edge_pos >> field_based) - (!!sy) - h, 0)) {
1483         s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, src,
1484                                  s->linesize, s->linesize,
1485                                  w + 1, (h + 1) << field_based,
1486                                  src_x, src_y   << field_based,
1487                                  h_edge_pos, v_edge_pos);
1488         src = s->sc.edge_emu_buffer;
1489         emu = 1;
1490     }
1491
1492     sx = (sx << 2) >> lowres;
1493     sy = (sy << 2) >> lowres;
1494     if (field_select)
1495         src += s->linesize;
1496     pix_op[op_index](dest, src, stride, h, sx, sy);
1497     return emu;
1498 }
1499
1500 /* apply one mpeg motion vector to the three components */
1501 static av_always_inline void mpeg_motion_lowres(MpegEncContext *s,
1502                                                 uint8_t *dest_y,
1503                                                 uint8_t *dest_cb,
1504                                                 uint8_t *dest_cr,
1505                                                 int field_based,
1506                                                 int bottom_field,
1507                                                 int field_select,
1508                                                 uint8_t **ref_picture,
1509                                                 h264_chroma_mc_func *pix_op,
1510                                                 int motion_x, int motion_y,
1511                                                 int h, int mb_y)
1512 {
1513     uint8_t *ptr_y, *ptr_cb, *ptr_cr;
1514     int mx, my, src_x, src_y, uvsrc_x, uvsrc_y, sx, sy, uvsx, uvsy;
1515     ptrdiff_t uvlinesize, linesize;
1516     const int lowres     = s->avctx->lowres;
1517     const int op_index   = FFMIN(lowres-1+s->chroma_x_shift, 3);
1518     const int block_s    = 8>>lowres;
1519     const int s_mask     = (2 << lowres) - 1;
1520     const int h_edge_pos = s->h_edge_pos >> lowres;
1521     const int v_edge_pos = s->v_edge_pos >> lowres;
1522     linesize   = s->current_picture.f->linesize[0] << field_based;
1523     uvlinesize = s->current_picture.f->linesize[1] << field_based;
1524
1525     // FIXME obviously not perfect but qpel will not work in lowres anyway
1526     if (s->quarter_sample) {
1527         motion_x /= 2;
1528         motion_y /= 2;
1529     }
1530
1531     if(field_based){
1532         motion_y += (bottom_field - field_select)*((1 << lowres)-1);
1533     }
1534
1535     sx = motion_x & s_mask;
1536     sy = motion_y & s_mask;
1537     src_x = s->mb_x * 2 * block_s + (motion_x >> lowres + 1);
1538     src_y = (mb_y * 2 * block_s >> field_based) + (motion_y >> lowres + 1);
1539
1540     if (s->out_format == FMT_H263) {
1541         uvsx    = ((motion_x >> 1) & s_mask) | (sx & 1);
1542         uvsy    = ((motion_y >> 1) & s_mask) | (sy & 1);
1543         uvsrc_x = src_x >> 1;
1544         uvsrc_y = src_y >> 1;
1545     } else if (s->out_format == FMT_H261) {
1546         // even chroma mv's are full pel in H261
1547         mx      = motion_x / 4;
1548         my      = motion_y / 4;
1549         uvsx    = (2 * mx) & s_mask;
1550         uvsy    = (2 * my) & s_mask;
1551         uvsrc_x = s->mb_x * block_s + (mx >> lowres);
1552         uvsrc_y =    mb_y * block_s + (my >> lowres);
1553     } else {
1554         if(s->chroma_y_shift){
1555             mx      = motion_x / 2;
1556             my      = motion_y / 2;
1557             uvsx    = mx & s_mask;
1558             uvsy    = my & s_mask;
1559             uvsrc_x = s->mb_x * block_s                 + (mx >> lowres + 1);
1560             uvsrc_y =   (mb_y * block_s >> field_based) + (my >> lowres + 1);
1561         } else {
1562             if(s->chroma_x_shift){
1563             //Chroma422
1564                 mx = motion_x / 2;
1565                 uvsx = mx & s_mask;
1566                 uvsy = motion_y & s_mask;
1567                 uvsrc_y = src_y;
1568                 uvsrc_x = s->mb_x*block_s               + (mx >> (lowres+1));
1569             } else {
1570             //Chroma444
1571                 uvsx = motion_x & s_mask;
1572                 uvsy = motion_y & s_mask;
1573                 uvsrc_x = src_x;
1574                 uvsrc_y = src_y;
1575             }
1576         }
1577     }
1578
1579     ptr_y  = ref_picture[0] + src_y   * linesize   + src_x;
1580     ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x;
1581     ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x;
1582
1583     if ((unsigned) src_x > FFMAX( h_edge_pos - (!!sx) - 2 * block_s,       0) || uvsrc_y<0 ||
1584         (unsigned) src_y > FFMAX((v_edge_pos >> field_based) - (!!sy) - h, 0)) {
1585         s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, ptr_y,
1586                                  linesize >> field_based, linesize >> field_based,
1587                                  17, 17 + field_based,
1588                                 src_x, src_y << field_based, h_edge_pos,
1589                                 v_edge_pos);
1590         ptr_y = s->sc.edge_emu_buffer;
1591         if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY)) {
1592             uint8_t *ubuf = s->sc.edge_emu_buffer + 18 * s->linesize;
1593             uint8_t *vbuf =ubuf + 10 * s->uvlinesize;
1594             if (s->workaround_bugs & FF_BUG_IEDGE)
1595                 vbuf -= s->uvlinesize;
1596             s->vdsp.emulated_edge_mc(ubuf,  ptr_cb,
1597                                      uvlinesize >> field_based, uvlinesize >> field_based,
1598                                      9, 9 + field_based,
1599                                     uvsrc_x, uvsrc_y << field_based,
1600                                     h_edge_pos >> 1, v_edge_pos >> 1);
1601             s->vdsp.emulated_edge_mc(vbuf,  ptr_cr,
1602                                      uvlinesize >> field_based,uvlinesize >> field_based,
1603                                      9, 9 + field_based,
1604                                     uvsrc_x, uvsrc_y << field_based,
1605                                     h_edge_pos >> 1, v_edge_pos >> 1);
1606             ptr_cb = ubuf;
1607             ptr_cr = vbuf;
1608         }
1609     }
1610
1611     // FIXME use this for field pix too instead of the obnoxious hack which changes picture.f->data
1612     if (bottom_field) {
1613         dest_y  += s->linesize;
1614         dest_cb += s->uvlinesize;
1615         dest_cr += s->uvlinesize;
1616     }
1617
1618     if (field_select) {
1619         ptr_y   += s->linesize;
1620         ptr_cb  += s->uvlinesize;
1621         ptr_cr  += s->uvlinesize;
1622     }
1623
1624     sx = (sx << 2) >> lowres;
1625     sy = (sy << 2) >> lowres;
1626     pix_op[lowres - 1](dest_y, ptr_y, linesize, h, sx, sy);
1627
1628     if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY)) {
1629         int hc = s->chroma_y_shift ? (h+1-bottom_field)>>1 : h;
1630         uvsx = (uvsx << 2) >> lowres;
1631         uvsy = (uvsy << 2) >> lowres;
1632         if (hc) {
1633             pix_op[op_index](dest_cb, ptr_cb, uvlinesize, hc, uvsx, uvsy);
1634             pix_op[op_index](dest_cr, ptr_cr, uvlinesize, hc, uvsx, uvsy);
1635         }
1636     }
1637     // FIXME h261 lowres loop filter
1638 }
1639
1640 static inline void chroma_4mv_motion_lowres(MpegEncContext *s,
1641                                             uint8_t *dest_cb, uint8_t *dest_cr,
1642                                             uint8_t **ref_picture,
1643                                             h264_chroma_mc_func * pix_op,
1644                                             int mx, int my)
1645 {
1646     const int lowres     = s->avctx->lowres;
1647     const int op_index   = FFMIN(lowres, 3);
1648     const int block_s    = 8 >> lowres;
1649     const int s_mask     = (2 << lowres) - 1;
1650     const int h_edge_pos = s->h_edge_pos >> lowres + 1;
1651     const int v_edge_pos = s->v_edge_pos >> lowres + 1;
1652     int emu = 0, src_x, src_y, sx, sy;
1653     ptrdiff_t offset;
1654     uint8_t *ptr;
1655
1656     if (s->quarter_sample) {
1657         mx /= 2;
1658         my /= 2;
1659     }
1660
1661     /* In case of 8X8, we construct a single chroma motion vector
1662        with a special rounding */
1663     mx = ff_h263_round_chroma(mx);
1664     my = ff_h263_round_chroma(my);
1665
1666     sx = mx & s_mask;
1667     sy = my & s_mask;
1668     src_x = s->mb_x * block_s + (mx >> lowres + 1);
1669     src_y = s->mb_y * block_s + (my >> lowres + 1);
1670
1671     offset = src_y * s->uvlinesize + src_x;
1672     ptr = ref_picture[1] + offset;
1673     if ((unsigned) src_x > FFMAX(h_edge_pos - (!!sx) - block_s, 0) ||
1674         (unsigned) src_y > FFMAX(v_edge_pos - (!!sy) - block_s, 0)) {
1675         s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, ptr,
1676                                  s->uvlinesize, s->uvlinesize,
1677                                  9, 9,
1678                                  src_x, src_y, h_edge_pos, v_edge_pos);
1679         ptr = s->sc.edge_emu_buffer;
1680         emu = 1;
1681     }
1682     sx = (sx << 2) >> lowres;
1683     sy = (sy << 2) >> lowres;
1684     pix_op[op_index](dest_cb, ptr, s->uvlinesize, block_s, sx, sy);
1685
1686     ptr = ref_picture[2] + offset;
1687     if (emu) {
1688         s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, ptr,
1689                                  s->uvlinesize, s->uvlinesize,
1690                                  9, 9,
1691                                  src_x, src_y, h_edge_pos, v_edge_pos);
1692         ptr = s->sc.edge_emu_buffer;
1693     }
1694     pix_op[op_index](dest_cr, ptr, s->uvlinesize, block_s, sx, sy);
1695 }
1696
1697 /**
1698  * motion compensation of a single macroblock
1699  * @param s context
1700  * @param dest_y luma destination pointer
1701  * @param dest_cb chroma cb/u destination pointer
1702  * @param dest_cr chroma cr/v destination pointer
1703  * @param dir direction (0->forward, 1->backward)
1704  * @param ref_picture array[3] of pointers to the 3 planes of the reference picture
1705  * @param pix_op halfpel motion compensation function (average or put normally)
1706  * the motion vectors are taken from s->mv and the MV type from s->mv_type
1707  */
1708 static inline void MPV_motion_lowres(MpegEncContext *s,
1709                                      uint8_t *dest_y, uint8_t *dest_cb,
1710                                      uint8_t *dest_cr,
1711                                      int dir, uint8_t **ref_picture,
1712                                      h264_chroma_mc_func *pix_op)
1713 {
1714     int mx, my;
1715     int mb_x, mb_y, i;
1716     const int lowres  = s->avctx->lowres;
1717     const int block_s = 8 >>lowres;
1718
1719     mb_x = s->mb_x;
1720     mb_y = s->mb_y;
1721
1722     switch (s->mv_type) {
1723     case MV_TYPE_16X16:
1724         mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
1725                            0, 0, 0,
1726                            ref_picture, pix_op,
1727                            s->mv[dir][0][0], s->mv[dir][0][1],
1728                            2 * block_s, mb_y);
1729         break;
1730     case MV_TYPE_8X8:
1731         mx = 0;
1732         my = 0;
1733         for (i = 0; i < 4; i++) {
1734             hpel_motion_lowres(s, dest_y + ((i & 1) + (i >> 1) *
1735                                s->linesize) * block_s,
1736                                ref_picture[0], 0, 0,
1737                                (2 * mb_x + (i & 1)) * block_s,
1738                                (2 * mb_y + (i >> 1)) * block_s,
1739                                s->width, s->height, s->linesize,
1740                                s->h_edge_pos >> lowres, s->v_edge_pos >> lowres,
1741                                block_s, block_s, pix_op,
1742                                s->mv[dir][i][0], s->mv[dir][i][1]);
1743
1744             mx += s->mv[dir][i][0];
1745             my += s->mv[dir][i][1];
1746         }
1747
1748         if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY))
1749             chroma_4mv_motion_lowres(s, dest_cb, dest_cr, ref_picture,
1750                                      pix_op, mx, my);
1751         break;
1752     case MV_TYPE_FIELD:
1753         if (s->picture_structure == PICT_FRAME) {
1754             /* top field */
1755             mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
1756                                1, 0, s->field_select[dir][0],
1757                                ref_picture, pix_op,
1758                                s->mv[dir][0][0], s->mv[dir][0][1],
1759                                block_s, mb_y);
1760             /* bottom field */
1761             mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
1762                                1, 1, s->field_select[dir][1],
1763                                ref_picture, pix_op,
1764                                s->mv[dir][1][0], s->mv[dir][1][1],
1765                                block_s, mb_y);
1766         } else {
1767             if (s->picture_structure != s->field_select[dir][0] + 1 &&
1768                 s->pict_type != AV_PICTURE_TYPE_B && !s->first_field) {
1769                 ref_picture = s->current_picture_ptr->f->data;
1770
1771             }
1772             mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
1773                                0, 0, s->field_select[dir][0],
1774                                ref_picture, pix_op,
1775                                s->mv[dir][0][0],
1776                                s->mv[dir][0][1], 2 * block_s, mb_y >> 1);
1777             }
1778         break;
1779     case MV_TYPE_16X8:
1780         for (i = 0; i < 2; i++) {
1781             uint8_t **ref2picture;
1782
1783             if (s->picture_structure == s->field_select[dir][i] + 1 ||
1784                 s->pict_type == AV_PICTURE_TYPE_B || s->first_field) {
1785                 ref2picture = ref_picture;
1786             } else {
1787                 ref2picture = s->current_picture_ptr->f->data;
1788             }
1789
1790             mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
1791                                0, 0, s->field_select[dir][i],
1792                                ref2picture, pix_op,
1793                                s->mv[dir][i][0], s->mv[dir][i][1] +
1794                                2 * block_s * i, block_s, mb_y >> 1);
1795
1796             dest_y  +=  2 * block_s *  s->linesize;
1797             dest_cb += (2 * block_s >> s->chroma_y_shift) * s->uvlinesize;
1798             dest_cr += (2 * block_s >> s->chroma_y_shift) * s->uvlinesize;
1799         }
1800         break;
1801     case MV_TYPE_DMV:
1802         if (s->picture_structure == PICT_FRAME) {
1803             for (i = 0; i < 2; i++) {
1804                 int j;
1805                 for (j = 0; j < 2; j++) {
1806                     mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
1807                                        1, j, j ^ i,
1808                                        ref_picture, pix_op,
1809                                        s->mv[dir][2 * i + j][0],
1810                                        s->mv[dir][2 * i + j][1],
1811                                        block_s, mb_y);
1812                 }
1813                 pix_op = s->h264chroma.avg_h264_chroma_pixels_tab;
1814             }
1815         } else {
1816             for (i = 0; i < 2; i++) {
1817                 mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
1818                                    0, 0, s->picture_structure != i + 1,
1819                                    ref_picture, pix_op,
1820                                    s->mv[dir][2 * i][0],s->mv[dir][2 * i][1],
1821                                    2 * block_s, mb_y >> 1);
1822
1823                 // after put we make avg of the same block
1824                 pix_op = s->h264chroma.avg_h264_chroma_pixels_tab;
1825
1826                 // opposite parity is always in the same
1827                 // frame if this is second field
1828                 if (!s->first_field) {
1829                     ref_picture = s->current_picture_ptr->f->data;
1830                 }
1831             }
1832         }
1833         break;
1834     default:
1835         av_assert2(0);
1836     }
1837 }
1838
1839 /**
1840  * find the lowest MB row referenced in the MVs
1841  */
1842 static int lowest_referenced_row(MpegEncContext *s, int dir)
1843 {
1844     int my_max = INT_MIN, my_min = INT_MAX, qpel_shift = !s->quarter_sample;
1845     int my, off, i, mvs;
1846
1847     if (s->picture_structure != PICT_FRAME || s->mcsel)
1848         goto unhandled;
1849
1850     switch (s->mv_type) {
1851         case MV_TYPE_16X16:
1852             mvs = 1;
1853             break;
1854         case MV_TYPE_16X8:
1855             mvs = 2;
1856             break;
1857         case MV_TYPE_8X8:
1858             mvs = 4;
1859             break;
1860         default:
1861             goto unhandled;
1862     }
1863
1864     for (i = 0; i < mvs; i++) {
1865         my = s->mv[dir][i][1];
1866         my_max = FFMAX(my_max, my);
1867         my_min = FFMIN(my_min, my);
1868     }
1869
1870     off = ((FFMAX(-my_min, my_max)<<qpel_shift) + 63) >> 6;
1871
1872     return av_clip(s->mb_y + off, 0, s->mb_height - 1);
1873 unhandled:
1874     return s->mb_height-1;
1875 }
1876
1877 /* put block[] to dest[] */
1878 static inline void put_dct(MpegEncContext *s,
1879                            int16_t *block, int i, uint8_t *dest, int line_size, int qscale)
1880 {
1881     s->dct_unquantize_intra(s, block, i, qscale);
1882     s->idsp.idct_put(dest, line_size, block);
1883 }
1884
1885 /* add block[] to dest[] */
1886 static inline void add_dct(MpegEncContext *s,
1887                            int16_t *block, int i, uint8_t *dest, int line_size)
1888 {
1889     if (s->block_last_index[i] >= 0) {
1890         s->idsp.idct_add(dest, line_size, block);
1891     }
1892 }
1893
1894 static inline void add_dequant_dct(MpegEncContext *s,
1895                            int16_t *block, int i, uint8_t *dest, int line_size, int qscale)
1896 {
1897     if (s->block_last_index[i] >= 0) {
1898         s->dct_unquantize_inter(s, block, i, qscale);
1899
1900         s->idsp.idct_add(dest, line_size, block);
1901     }
1902 }
1903
1904 /**
1905  * Clean dc, ac, coded_block for the current non-intra MB.
1906  */
1907 void ff_clean_intra_table_entries(MpegEncContext *s)
1908 {
1909     int wrap = s->b8_stride;
1910     int xy = s->block_index[0];
1911
1912     s->dc_val[0][xy           ] =
1913     s->dc_val[0][xy + 1       ] =
1914     s->dc_val[0][xy     + wrap] =
1915     s->dc_val[0][xy + 1 + wrap] = 1024;
1916     /* ac pred */
1917     memset(s->ac_val[0][xy       ], 0, 32 * sizeof(int16_t));
1918     memset(s->ac_val[0][xy + wrap], 0, 32 * sizeof(int16_t));
1919     if (s->msmpeg4_version>=3) {
1920         s->coded_block[xy           ] =
1921         s->coded_block[xy + 1       ] =
1922         s->coded_block[xy     + wrap] =
1923         s->coded_block[xy + 1 + wrap] = 0;
1924     }
1925     /* chroma */
1926     wrap = s->mb_stride;
1927     xy = s->mb_x + s->mb_y * wrap;
1928     s->dc_val[1][xy] =
1929     s->dc_val[2][xy] = 1024;
1930     /* ac pred */
1931     memset(s->ac_val[1][xy], 0, 16 * sizeof(int16_t));
1932     memset(s->ac_val[2][xy], 0, 16 * sizeof(int16_t));
1933
1934     s->mbintra_table[xy]= 0;
1935 }
1936
1937 /* generic function called after a macroblock has been parsed by the
1938    decoder or after it has been encoded by the encoder.
1939
1940    Important variables used:
1941    s->mb_intra : true if intra macroblock
1942    s->mv_dir   : motion vector direction
1943    s->mv_type  : motion vector type
1944    s->mv       : motion vector
1945    s->interlaced_dct : true if interlaced dct used (mpeg2)
1946  */
1947 static av_always_inline
1948 void mpv_reconstruct_mb_internal(MpegEncContext *s, int16_t block[12][64],
1949                             int lowres_flag, int is_mpeg12)
1950 {
1951     const int mb_xy = s->mb_y * s->mb_stride + s->mb_x;
1952
1953     if (CONFIG_XVMC &&
1954         s->avctx->hwaccel && s->avctx->hwaccel->decode_mb) {
1955         s->avctx->hwaccel->decode_mb(s);//xvmc uses pblocks
1956         return;
1957     }
1958
1959     if(s->avctx->debug&FF_DEBUG_DCT_COEFF) {
1960        /* print DCT coefficients */
1961        int i,j;
1962        av_log(s->avctx, AV_LOG_DEBUG, "DCT coeffs of MB at %dx%d:\n", s->mb_x, s->mb_y);
1963        for(i=0; i<6; i++){
1964            for(j=0; j<64; j++){
1965                av_log(s->avctx, AV_LOG_DEBUG, "%5d",
1966                       block[i][s->idsp.idct_permutation[j]]);
1967            }
1968            av_log(s->avctx, AV_LOG_DEBUG, "\n");
1969        }
1970     }
1971
1972     s->current_picture.qscale_table[mb_xy] = s->qscale;
1973
1974     /* update DC predictors for P macroblocks */
1975     if (!s->mb_intra) {
1976         if (!is_mpeg12 && (s->h263_pred || s->h263_aic)) {
1977             if(s->mbintra_table[mb_xy])
1978                 ff_clean_intra_table_entries(s);
1979         } else {
1980             s->last_dc[0] =
1981             s->last_dc[1] =
1982             s->last_dc[2] = 128 << s->intra_dc_precision;
1983         }
1984     }
1985     else if (!is_mpeg12 && (s->h263_pred || s->h263_aic))
1986         s->mbintra_table[mb_xy]=1;
1987
1988     if ((s->avctx->flags & AV_CODEC_FLAG_PSNR) || s->frame_skip_threshold || s->frame_skip_factor ||
1989         !(s->encoding && (s->intra_only || s->pict_type == AV_PICTURE_TYPE_B) &&
1990           s->avctx->mb_decision != FF_MB_DECISION_RD)) { // FIXME precalc
1991         uint8_t *dest_y, *dest_cb, *dest_cr;
1992         int dct_linesize, dct_offset;
1993         op_pixels_func (*op_pix)[4];
1994         qpel_mc_func (*op_qpix)[16];
1995         const int linesize   = s->current_picture.f->linesize[0]; //not s->linesize as this would be wrong for field pics
1996         const int uvlinesize = s->current_picture.f->linesize[1];
1997         const int readable= s->pict_type != AV_PICTURE_TYPE_B || s->encoding || s->avctx->draw_horiz_band || lowres_flag;
1998         const int block_size= lowres_flag ? 8>>s->avctx->lowres : 8;
1999
2000         /* avoid copy if macroblock skipped in last frame too */
2001         /* skip only during decoding as we might trash the buffers during encoding a bit */
2002         if(!s->encoding){
2003             uint8_t *mbskip_ptr = &s->mbskip_table[mb_xy];
2004
2005             if (s->mb_skipped) {
2006                 s->mb_skipped= 0;
2007                 av_assert2(s->pict_type!=AV_PICTURE_TYPE_I);
2008                 *mbskip_ptr = 1;
2009             } else if(!s->current_picture.reference) {
2010                 *mbskip_ptr = 1;
2011             } else{
2012                 *mbskip_ptr = 0; /* not skipped */
2013             }
2014         }
2015
2016         dct_linesize = linesize << s->interlaced_dct;
2017         dct_offset   = s->interlaced_dct ? linesize : linesize * block_size;
2018
2019         if(readable){
2020             dest_y=  s->dest[0];
2021             dest_cb= s->dest[1];
2022             dest_cr= s->dest[2];
2023         }else{
2024             dest_y = s->sc.b_scratchpad;
2025             dest_cb= s->sc.b_scratchpad+16*linesize;
2026             dest_cr= s->sc.b_scratchpad+32*linesize;
2027         }
2028
2029         if (!s->mb_intra) {
2030             /* motion handling */
2031             /* decoding or more than one mb_type (MC was already done otherwise) */
2032             if(!s->encoding){
2033
2034                 if(HAVE_THREADS && s->avctx->active_thread_type&FF_THREAD_FRAME) {
2035                     if (s->mv_dir & MV_DIR_FORWARD) {
2036                         ff_thread_await_progress(&s->last_picture_ptr->tf,
2037                                                  lowest_referenced_row(s, 0),
2038                                                  0);
2039                     }
2040                     if (s->mv_dir & MV_DIR_BACKWARD) {
2041                         ff_thread_await_progress(&s->next_picture_ptr->tf,
2042                                                  lowest_referenced_row(s, 1),
2043                                                  0);
2044                     }
2045                 }
2046
2047                 if(lowres_flag){
2048                     h264_chroma_mc_func *op_pix = s->h264chroma.put_h264_chroma_pixels_tab;
2049
2050                     if (s->mv_dir & MV_DIR_FORWARD) {
2051                         MPV_motion_lowres(s, dest_y, dest_cb, dest_cr, 0, s->last_picture.f->data, op_pix);
2052                         op_pix = s->h264chroma.avg_h264_chroma_pixels_tab;
2053                     }
2054                     if (s->mv_dir & MV_DIR_BACKWARD) {
2055                         MPV_motion_lowres(s, dest_y, dest_cb, dest_cr, 1, s->next_picture.f->data, op_pix);
2056                     }
2057                 }else{
2058                     op_qpix = s->me.qpel_put;
2059                     if ((!s->no_rounding) || s->pict_type==AV_PICTURE_TYPE_B){
2060                         op_pix = s->hdsp.put_pixels_tab;
2061                     }else{
2062                         op_pix = s->hdsp.put_no_rnd_pixels_tab;
2063                     }
2064                     if (s->mv_dir & MV_DIR_FORWARD) {
2065                         ff_mpv_motion(s, dest_y, dest_cb, dest_cr, 0, s->last_picture.f->data, op_pix, op_qpix);
2066                         op_pix = s->hdsp.avg_pixels_tab;
2067                         op_qpix= s->me.qpel_avg;
2068                     }
2069                     if (s->mv_dir & MV_DIR_BACKWARD) {
2070                         ff_mpv_motion(s, dest_y, dest_cb, dest_cr, 1, s->next_picture.f->data, op_pix, op_qpix);
2071                     }
2072                 }
2073             }
2074
2075             /* skip dequant / idct if we are really late ;) */
2076             if(s->avctx->skip_idct){
2077                 if(  (s->avctx->skip_idct >= AVDISCARD_NONREF && s->pict_type == AV_PICTURE_TYPE_B)
2078                    ||(s->avctx->skip_idct >= AVDISCARD_NONKEY && s->pict_type != AV_PICTURE_TYPE_I)
2079                    || s->avctx->skip_idct >= AVDISCARD_ALL)
2080                     goto skip_idct;
2081             }
2082
2083             /* add dct residue */
2084             if(s->encoding || !(   s->msmpeg4_version || s->codec_id==AV_CODEC_ID_MPEG1VIDEO || s->codec_id==AV_CODEC_ID_MPEG2VIDEO
2085                                 || (s->codec_id==AV_CODEC_ID_MPEG4 && !s->mpeg_quant))){
2086                 add_dequant_dct(s, block[0], 0, dest_y                          , dct_linesize, s->qscale);
2087                 add_dequant_dct(s, block[1], 1, dest_y              + block_size, dct_linesize, s->qscale);
2088                 add_dequant_dct(s, block[2], 2, dest_y + dct_offset             , dct_linesize, s->qscale);
2089                 add_dequant_dct(s, block[3], 3, dest_y + dct_offset + block_size, dct_linesize, s->qscale);
2090
2091                 if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY)) {
2092                     if (s->chroma_y_shift){
2093                         add_dequant_dct(s, block[4], 4, dest_cb, uvlinesize, s->chroma_qscale);
2094                         add_dequant_dct(s, block[5], 5, dest_cr, uvlinesize, s->chroma_qscale);
2095                     }else{
2096                         dct_linesize >>= 1;
2097                         dct_offset >>=1;
2098                         add_dequant_dct(s, block[4], 4, dest_cb,              dct_linesize, s->chroma_qscale);
2099                         add_dequant_dct(s, block[5], 5, dest_cr,              dct_linesize, s->chroma_qscale);
2100                         add_dequant_dct(s, block[6], 6, dest_cb + dct_offset, dct_linesize, s->chroma_qscale);
2101                         add_dequant_dct(s, block[7], 7, dest_cr + dct_offset, dct_linesize, s->chroma_qscale);
2102                     }
2103                 }
2104             } else if(is_mpeg12 || (s->codec_id != AV_CODEC_ID_WMV2)){
2105                 add_dct(s, block[0], 0, dest_y                          , dct_linesize);
2106                 add_dct(s, block[1], 1, dest_y              + block_size, dct_linesize);
2107                 add_dct(s, block[2], 2, dest_y + dct_offset             , dct_linesize);
2108                 add_dct(s, block[3], 3, dest_y + dct_offset + block_size, dct_linesize);
2109
2110                 if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY)) {
2111                     if(s->chroma_y_shift){//Chroma420
2112                         add_dct(s, block[4], 4, dest_cb, uvlinesize);
2113                         add_dct(s, block[5], 5, dest_cr, uvlinesize);
2114                     }else{
2115                         //chroma422
2116                         dct_linesize = uvlinesize << s->interlaced_dct;
2117                         dct_offset   = s->interlaced_dct ? uvlinesize : uvlinesize*block_size;
2118
2119                         add_dct(s, block[4], 4, dest_cb, dct_linesize);
2120                         add_dct(s, block[5], 5, dest_cr, dct_linesize);
2121                         add_dct(s, block[6], 6, dest_cb+dct_offset, dct_linesize);
2122                         add_dct(s, block[7], 7, dest_cr+dct_offset, dct_linesize);
2123                         if(!s->chroma_x_shift){//Chroma444
2124                             add_dct(s, block[8], 8, dest_cb+block_size, dct_linesize);
2125                             add_dct(s, block[9], 9, dest_cr+block_size, dct_linesize);
2126                             add_dct(s, block[10], 10, dest_cb+block_size+dct_offset, dct_linesize);
2127                             add_dct(s, block[11], 11, dest_cr+block_size+dct_offset, dct_linesize);
2128                         }
2129                     }
2130                 }//fi gray
2131             }
2132             else if (CONFIG_WMV2_DECODER || CONFIG_WMV2_ENCODER) {
2133                 ff_wmv2_add_mb(s, block, dest_y, dest_cb, dest_cr);
2134             }
2135         } else {
2136             /* Only MPEG-4 Simple Studio Profile is supported in > 8-bit mode.
2137                TODO: Integrate 10-bit properly into mpegvideo.c so that ER works properly */
2138             if (s->avctx->bits_per_raw_sample > 8){
2139                 const int act_block_size = block_size * 2;
2140
2141                 if(s->dpcm_direction == 0) {
2142                     s->idsp.idct_put(dest_y,                           dct_linesize, (int16_t*)(*s->block32)[0]);
2143                     s->idsp.idct_put(dest_y              + act_block_size, dct_linesize, (int16_t*)(*s->block32)[1]);
2144                     s->idsp.idct_put(dest_y + dct_offset,              dct_linesize, (int16_t*)(*s->block32)[2]);
2145                     s->idsp.idct_put(dest_y + dct_offset + act_block_size, dct_linesize, (int16_t*)(*s->block32)[3]);
2146
2147                     dct_linesize = uvlinesize << s->interlaced_dct;
2148                     dct_offset   = s->interlaced_dct ? uvlinesize : uvlinesize*block_size;
2149
2150                     s->idsp.idct_put(dest_cb,              dct_linesize, (int16_t*)(*s->block32)[4]);
2151                     s->idsp.idct_put(dest_cr,              dct_linesize, (int16_t*)(*s->block32)[5]);
2152                     s->idsp.idct_put(dest_cb + dct_offset, dct_linesize, (int16_t*)(*s->block32)[6]);
2153                     s->idsp.idct_put(dest_cr + dct_offset, dct_linesize, (int16_t*)(*s->block32)[7]);
2154                     if(!s->chroma_x_shift){//Chroma444
2155                         s->idsp.idct_put(dest_cb + act_block_size,              dct_linesize, (int16_t*)(*s->block32)[8]);
2156                         s->idsp.idct_put(dest_cr + act_block_size,              dct_linesize, (int16_t*)(*s->block32)[9]);
2157                         s->idsp.idct_put(dest_cb + act_block_size + dct_offset, dct_linesize, (int16_t*)(*s->block32)[10]);
2158                         s->idsp.idct_put(dest_cr + act_block_size + dct_offset, dct_linesize, (int16_t*)(*s->block32)[11]);
2159                     }
2160                 } else if(s->dpcm_direction == 1) {
2161                     int i, w, h;
2162                     uint16_t *dest_pcm[3] = {(uint16_t*)dest_y, (uint16_t*)dest_cb, (uint16_t*)dest_cr};
2163                     int linesize[3] = {dct_linesize, uvlinesize, uvlinesize};
2164                     for(i = 0; i < 3; i++) {
2165                         int idx = 0;
2166                         int vsub = i ? s->chroma_y_shift : 0;
2167                         int hsub = i ? s->chroma_x_shift : 0;
2168                         for(h = 0; h < (16 >> vsub); h++){
2169                             for(w = 0; w < (16 >> hsub); w++)
2170                                 dest_pcm[i][w] = (*s->dpcm_macroblock)[i][idx++];
2171                             dest_pcm[i] += linesize[i] / 2;
2172                         }
2173                     }
2174                 } else if(s->dpcm_direction == -1) {
2175                     int i, w, h;
2176                     uint16_t *dest_pcm[3] = {(uint16_t*)dest_y, (uint16_t*)dest_cb, (uint16_t*)dest_cr};
2177                     int linesize[3] = {dct_linesize, uvlinesize, uvlinesize};
2178                     for(i = 0; i < 3; i++) {
2179                         int idx = 0;
2180                         int vsub = i ? s->chroma_y_shift : 0;
2181                         int hsub = i ? s->chroma_x_shift : 0;
2182                         dest_pcm[i] += (linesize[i] / 2) * ((16 >> vsub) - 1);
2183                         for(h = (16 >> vsub)-1; h >= 1; h--){
2184                             for(w = (16 >> hsub)-1; w >= 1; w--)
2185                                 dest_pcm[i][w] = (*s->dpcm_macroblock)[i][idx++];
2186                             dest_pcm[i] -= linesize[i] / 2;
2187                         }
2188                     }
2189                 }
2190             }
2191             /* dct only in intra block */
2192             else if(s->encoding || !(s->codec_id==AV_CODEC_ID_MPEG1VIDEO || s->codec_id==AV_CODEC_ID_MPEG2VIDEO)){
2193                 put_dct(s, block[0], 0, dest_y                          , dct_linesize, s->qscale);
2194                 put_dct(s, block[1], 1, dest_y              + block_size, dct_linesize, s->qscale);
2195                 put_dct(s, block[2], 2, dest_y + dct_offset             , dct_linesize, s->qscale);
2196                 put_dct(s, block[3], 3, dest_y + dct_offset + block_size, dct_linesize, s->qscale);
2197
2198                 if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY)) {
2199                     if(s->chroma_y_shift){
2200                         put_dct(s, block[4], 4, dest_cb, uvlinesize, s->chroma_qscale);
2201                         put_dct(s, block[5], 5, dest_cr, uvlinesize, s->chroma_qscale);
2202                     }else{
2203                         dct_offset >>=1;
2204                         dct_linesize >>=1;
2205                         put_dct(s, block[4], 4, dest_cb,              dct_linesize, s->chroma_qscale);
2206                         put_dct(s, block[5], 5, dest_cr,              dct_linesize, s->chroma_qscale);
2207                         put_dct(s, block[6], 6, dest_cb + dct_offset, dct_linesize, s->chroma_qscale);
2208                         put_dct(s, block[7], 7, dest_cr + dct_offset, dct_linesize, s->chroma_qscale);
2209                     }
2210                 }
2211             }else{
2212                 s->idsp.idct_put(dest_y,                           dct_linesize, block[0]);
2213                 s->idsp.idct_put(dest_y              + block_size, dct_linesize, block[1]);
2214                 s->idsp.idct_put(dest_y + dct_offset,              dct_linesize, block[2]);
2215                 s->idsp.idct_put(dest_y + dct_offset + block_size, dct_linesize, block[3]);
2216
2217                 if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY)) {
2218                     if(s->chroma_y_shift){
2219                         s->idsp.idct_put(dest_cb, uvlinesize, block[4]);
2220                         s->idsp.idct_put(dest_cr, uvlinesize, block[5]);
2221                     }else{
2222
2223                         dct_linesize = uvlinesize << s->interlaced_dct;
2224                         dct_offset   = s->interlaced_dct ? uvlinesize : uvlinesize*block_size;
2225
2226                         s->idsp.idct_put(dest_cb,              dct_linesize, block[4]);
2227                         s->idsp.idct_put(dest_cr,              dct_linesize, block[5]);
2228                         s->idsp.idct_put(dest_cb + dct_offset, dct_linesize, block[6]);
2229                         s->idsp.idct_put(dest_cr + dct_offset, dct_linesize, block[7]);
2230                         if(!s->chroma_x_shift){//Chroma444
2231                             s->idsp.idct_put(dest_cb + block_size,              dct_linesize, block[8]);
2232                             s->idsp.idct_put(dest_cr + block_size,              dct_linesize, block[9]);
2233                             s->idsp.idct_put(dest_cb + block_size + dct_offset, dct_linesize, block[10]);
2234                             s->idsp.idct_put(dest_cr + block_size + dct_offset, dct_linesize, block[11]);
2235                         }
2236                     }
2237                 }//gray
2238             }
2239         }
2240 skip_idct:
2241         if(!readable){
2242             s->hdsp.put_pixels_tab[0][0](s->dest[0], dest_y ,   linesize,16);
2243             if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY)) {
2244                 s->hdsp.put_pixels_tab[s->chroma_x_shift][0](s->dest[1], dest_cb, uvlinesize,16 >> s->chroma_y_shift);
2245                 s->hdsp.put_pixels_tab[s->chroma_x_shift][0](s->dest[2], dest_cr, uvlinesize,16 >> s->chroma_y_shift);
2246             }
2247         }
2248     }
2249 }
2250
2251 void ff_mpv_reconstruct_mb(MpegEncContext *s, int16_t block[12][64])
2252 {
2253 #if !CONFIG_SMALL
2254     if(s->out_format == FMT_MPEG1) {
2255         if(s->avctx->lowres) mpv_reconstruct_mb_internal(s, block, 1, 1);
2256         else                 mpv_reconstruct_mb_internal(s, block, 0, 1);
2257     } else
2258 #endif
2259     if(s->avctx->lowres) mpv_reconstruct_mb_internal(s, block, 1, 0);
2260     else                  mpv_reconstruct_mb_internal(s, block, 0, 0);
2261 }
2262
2263 void ff_mpeg_draw_horiz_band(MpegEncContext *s, int y, int h)
2264 {
2265     ff_draw_horiz_band(s->avctx, s->current_picture_ptr->f,
2266                        s->last_picture_ptr ? s->last_picture_ptr->f : NULL, y, h, s->picture_structure,
2267                        s->first_field, s->low_delay);
2268 }
2269
2270 void ff_init_block_index(MpegEncContext *s){ //FIXME maybe rename
2271     const int linesize   = s->current_picture.f->linesize[0]; //not s->linesize as this would be wrong for field pics
2272     const int uvlinesize = s->current_picture.f->linesize[1];
2273     const int width_of_mb = (4 + (s->avctx->bits_per_raw_sample > 8)) - s->avctx->lowres;
2274     const int height_of_mb = 4 - s->avctx->lowres;
2275
2276     s->block_index[0]= s->b8_stride*(s->mb_y*2    ) - 2 + s->mb_x*2;
2277     s->block_index[1]= s->b8_stride*(s->mb_y*2    ) - 1 + s->mb_x*2;
2278     s->block_index[2]= s->b8_stride*(s->mb_y*2 + 1) - 2 + s->mb_x*2;
2279     s->block_index[3]= s->b8_stride*(s->mb_y*2 + 1) - 1 + s->mb_x*2;
2280     s->block_index[4]= s->mb_stride*(s->mb_y + 1)                + s->b8_stride*s->mb_height*2 + s->mb_x - 1;
2281     s->block_index[5]= s->mb_stride*(s->mb_y + s->mb_height + 2) + s->b8_stride*s->mb_height*2 + s->mb_x - 1;
2282     //block_index is not used by mpeg2, so it is not affected by chroma_format
2283
2284     s->dest[0] = s->current_picture.f->data[0] + (int)((s->mb_x - 1U) <<  width_of_mb);
2285     s->dest[1] = s->current_picture.f->data[1] + (int)((s->mb_x - 1U) << (width_of_mb - s->chroma_x_shift));
2286     s->dest[2] = s->current_picture.f->data[2] + (int)((s->mb_x - 1U) << (width_of_mb - s->chroma_x_shift));
2287
2288     if(!(s->pict_type==AV_PICTURE_TYPE_B && s->avctx->draw_horiz_band && s->picture_structure==PICT_FRAME))
2289     {
2290         if(s->picture_structure==PICT_FRAME){
2291         s->dest[0] += s->mb_y *   linesize << height_of_mb;
2292         s->dest[1] += s->mb_y * uvlinesize << (height_of_mb - s->chroma_y_shift);
2293         s->dest[2] += s->mb_y * uvlinesize << (height_of_mb - s->chroma_y_shift);
2294         }else{
2295             s->dest[0] += (s->mb_y>>1) *   linesize << height_of_mb;
2296             s->dest[1] += (s->mb_y>>1) * uvlinesize << (height_of_mb - s->chroma_y_shift);
2297             s->dest[2] += (s->mb_y>>1) * uvlinesize << (height_of_mb - s->chroma_y_shift);
2298             av_assert1((s->mb_y&1) == (s->picture_structure == PICT_BOTTOM_FIELD));
2299         }
2300     }
2301 }
2302
2303 void ff_mpeg_flush(AVCodecContext *avctx){
2304     int i;
2305     MpegEncContext *s = avctx->priv_data;
2306
2307     if (!s || !s->picture)
2308         return;
2309
2310     for (i = 0; i < MAX_PICTURE_COUNT; i++)
2311         ff_mpeg_unref_picture(s->avctx, &s->picture[i]);
2312     s->current_picture_ptr = s->last_picture_ptr = s->next_picture_ptr = NULL;
2313
2314     ff_mpeg_unref_picture(s->avctx, &s->current_picture);
2315     ff_mpeg_unref_picture(s->avctx, &s->last_picture);
2316     ff_mpeg_unref_picture(s->avctx, &s->next_picture);
2317
2318     s->mb_x= s->mb_y= 0;
2319     s->closed_gop= 0;
2320
2321     s->parse_context.state= -1;
2322     s->parse_context.frame_start_found= 0;
2323     s->parse_context.overread= 0;
2324     s->parse_context.overread_index= 0;
2325     s->parse_context.index= 0;
2326     s->parse_context.last_index= 0;
2327     s->bitstream_buffer_size=0;
2328     s->pp_time=0;
2329 }
2330
2331 /**
2332  * set qscale and update qscale dependent variables.
2333  */
2334 void ff_set_qscale(MpegEncContext * s, int qscale)
2335 {
2336     if (qscale < 1)
2337         qscale = 1;
2338     else if (qscale > 31)
2339         qscale = 31;
2340
2341     s->qscale = qscale;
2342     s->chroma_qscale= s->chroma_qscale_table[qscale];
2343
2344     s->y_dc_scale= s->y_dc_scale_table[ qscale ];
2345     s->c_dc_scale= s->c_dc_scale_table[ s->chroma_qscale ];
2346 }
2347
2348 void ff_mpv_report_decode_progress(MpegEncContext *s)
2349 {
2350     if (s->pict_type != AV_PICTURE_TYPE_B && !s->partitioned_frame && !s->er.error_occurred)
2351         ff_thread_report_progress(&s->current_picture_ptr->tf, s->mb_y, 0);
2352 }