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)
 350 {
 351     return ff_alloc_picture(s->avctx, pic, &s->me, &s->sc, 0, 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         if (!FF_ALLOCZ_TYPED_ARRAY(s->me.map,       ME_MAP_SIZE) ||
 376             !FF_ALLOCZ_TYPED_ARRAY(s->me.score_map, ME_MAP_SIZE))
 377             return AVERROR(ENOMEM);
 378
 379         if (s->noise_reduction) {
 380             if (!FF_ALLOCZ_TYPED_ARRAY(s->dct_error_sum,  2))
 381                 return AVERROR(ENOMEM);
 382         }
 383     }
 384     if (!FF_ALLOCZ_TYPED_ARRAY(s->blocks,  2))
 385         return AVERROR(ENOMEM);
 386     s->block = s->blocks[0];
 387
 388     for (i = 0; i < 12; i++) {
 389         s->pblocks[i] = &s->block[i];
 390     }
 391
 392     if (!(s->block32         = av_mallocz(sizeof(*s->block32))) ||
 393         !(s->dpcm_macroblock = av_mallocz(sizeof(*s->dpcm_macroblock))))
 394         return AVERROR(ENOMEM);
 395     s->dpcm_direction = 0;
 396
 397     if (s->avctx->codec_tag == AV_RL32("VCR2")) {
 398         // exchange uv
 399         FFSWAP(void *, s->pblocks[4], s->pblocks[5]);
 400     }
 401
 402     if (s->out_format == FMT_H263) {
 403         /* ac values */
 404         if (!FF_ALLOCZ_TYPED_ARRAY(s->ac_val_base,  yc_size))
 405             return AVERROR(ENOMEM);
 406         s->ac_val[0] = s->ac_val_base + s->b8_stride + 1;
 407         s->ac_val[1] = s->ac_val_base + y_size + s->mb_stride + 1;
 408         s->ac_val[2] = s->ac_val[1] + c_size;
 409     }
 410
 411     return 0;
 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     if (!FF_ALLOCZ_TYPED_ARRAY(s->mb_index2xy, s->mb_num + 1))
 719         return AVERROR(ENOMEM);
 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         if (!FF_ALLOCZ_TYPED_ARRAY(s->p_mv_table_base,            mv_table_size) ||
 729             !FF_ALLOCZ_TYPED_ARRAY(s->b_forw_mv_table_base,       mv_table_size) ||
 730             !FF_ALLOCZ_TYPED_ARRAY(s->b_back_mv_table_base,       mv_table_size) ||
 731             !FF_ALLOCZ_TYPED_ARRAY(s->b_bidir_forw_mv_table_base, mv_table_size) ||
 732             !FF_ALLOCZ_TYPED_ARRAY(s->b_bidir_back_mv_table_base, mv_table_size) ||
 733             !FF_ALLOCZ_TYPED_ARRAY(s->b_direct_mv_table_base,     mv_table_size))
 734             return AVERROR(ENOMEM);
 735         s->p_mv_table            = s->p_mv_table_base + s->mb_stride + 1;
 736         s->b_forw_mv_table       = s->b_forw_mv_table_base + s->mb_stride + 1;
 737         s->b_back_mv_table       = s->b_back_mv_table_base + s->mb_stride + 1;
 738         s->b_bidir_forw_mv_table = s->b_bidir_forw_mv_table_base + s->mb_stride + 1;
 739         s->b_bidir_back_mv_table = s->b_bidir_back_mv_table_base + s->mb_stride + 1;
 740         s->b_direct_mv_table     = s->b_direct_mv_table_base + s->mb_stride + 1;
 741
 742         /* Allocate MB type table */
 743         if (!FF_ALLOCZ_TYPED_ARRAY(s->mb_type,      mb_array_size) ||
 744             !FF_ALLOCZ_TYPED_ARRAY(s->lambda_table, mb_array_size) ||
 745             !FF_ALLOC_TYPED_ARRAY (s->cplx_tab,     mb_array_size) ||
 746             !FF_ALLOC_TYPED_ARRAY (s->bits_tab,     mb_array_size))
 747             return AVERROR(ENOMEM);
 748     }
 749
 750     if (s->codec_id == AV_CODEC_ID_MPEG4 ||
 751         (s->avctx->flags & AV_CODEC_FLAG_INTERLACED_ME)) {
 752         /* interlaced direct mode decoding tables */
 753         for (i = 0; i < 2; i++) {
 754             int j, k;
 755             for (j = 0; j < 2; j++) {
 756                 for (k = 0; k < 2; k++) {
 757                     if (!FF_ALLOCZ_TYPED_ARRAY(s->b_field_mv_table_base[i][j][k], mv_table_size))
 758                         return AVERROR(ENOMEM);
 759                     s->b_field_mv_table[i][j][k] = s->b_field_mv_table_base[i][j][k] +
 760                                                    s->mb_stride + 1;
 761                 }
 762                 if (!FF_ALLOCZ_TYPED_ARRAY(s->b_field_select_table [i][j], mv_table_size * 2) ||
 763                     !FF_ALLOCZ_TYPED_ARRAY(s->p_field_mv_table_base[i][j], mv_table_size))
 764                     return AVERROR(ENOMEM);
 765                 s->p_field_mv_table[i][j] = s->p_field_mv_table_base[i][j] + s->mb_stride + 1;
 766             }
 767             if (!FF_ALLOCZ_TYPED_ARRAY(s->p_field_select_table[i], mv_table_size * 2))
 768                 return AVERROR(ENOMEM);
 769         }
 770     }
 771     if (s->out_format == FMT_H263) {
 772         /* cbp values, cbp, ac_pred, pred_dir */
 773         if (!FF_ALLOCZ_TYPED_ARRAY(s->coded_block_base, y_size + (s->mb_height&1)*2*s->b8_stride) ||
 774             !FF_ALLOCZ_TYPED_ARRAY(s->cbp_table,        mb_array_size)                            ||
 775             !FF_ALLOCZ_TYPED_ARRAY(s->pred_dir_table,   mb_array_size))
 776             return AVERROR(ENOMEM);
 777         s->coded_block = s->coded_block_base + s->b8_stride + 1;
 778     }
 779
 780     if (s->h263_pred || s->h263_plus || !s->encoding) {
 781         /* dc values */
 782         // MN: we need these for error resilience of intra-frames
 783         if (!FF_ALLOCZ_TYPED_ARRAY(s->dc_val_base, yc_size))
 784             return AVERROR(ENOMEM);
 785         s->dc_val[0] = s->dc_val_base + s->b8_stride + 1;
 786         s->dc_val[1] = s->dc_val_base + y_size + s->mb_stride + 1;
 787         s->dc_val[2] = s->dc_val[1] + c_size;
 788         for (i = 0; i < yc_size; i++)
 789             s->dc_val_base[i] = 1024;
 790     }
 791
 792     /* which mb is an intra block,  init macroblock skip table */
 793     if (!FF_ALLOCZ_TYPED_ARRAY(s->mbintra_table, mb_array_size) ||
 794         // Note the + 1 is for a quicker MPEG-4 slice_end detection
 795         !FF_ALLOCZ_TYPED_ARRAY(s->mbskip_table,  mb_array_size + 2))
 796         return AVERROR(ENOMEM);
 797     memset(s->mbintra_table, 1, mb_array_size);
 798
 799     return ff_mpeg_er_init(s);
 800 }
 801
 802 static void clear_context(MpegEncContext *s)
 803 {
 804     int i, j, k;
 805
 806     memset(&s->next_picture, 0, sizeof(s->next_picture));
 807     memset(&s->last_picture, 0, sizeof(s->last_picture));
 808     memset(&s->current_picture, 0, sizeof(s->current_picture));
 809     memset(&s->new_picture, 0, sizeof(s->new_picture));
 810
 811     memset(s->thread_context, 0, sizeof(s->thread_context));
 812
 813     s->me.map = NULL;
 814     s->me.score_map = NULL;
 815     s->dct_error_sum = NULL;
 816     s->block = NULL;
 817     s->blocks = NULL;
 818     s->block32 = NULL;
 819     memset(s->pblocks, 0, sizeof(s->pblocks));
 820     s->dpcm_direction = 0;
 821     s->dpcm_macroblock = NULL;
 822     s->ac_val_base = NULL;
 823     s->ac_val[0] =
 824     s->ac_val[1] =
 825     s->ac_val[2] =NULL;
 826     s->sc.edge_emu_buffer = NULL;
 827     s->me.scratchpad = NULL;
 828     s->me.temp =
 829     s->sc.rd_scratchpad =
 830     s->sc.b_scratchpad =
 831     s->sc.obmc_scratchpad = NULL;
 832
 833
 834     s->bitstream_buffer = NULL;
 835     s->allocated_bitstream_buffer_size = 0;
 836     s->picture          = NULL;
 837     s->mb_type          = NULL;
 838     s->p_mv_table_base  = NULL;
 839     s->b_forw_mv_table_base = NULL;
 840     s->b_back_mv_table_base = NULL;
 841     s->b_bidir_forw_mv_table_base = NULL;
 842     s->b_bidir_back_mv_table_base = NULL;
 843     s->b_direct_mv_table_base = NULL;
 844     s->p_mv_table            = NULL;
 845     s->b_forw_mv_table       = NULL;
 846     s->b_back_mv_table       = NULL;
 847     s->b_bidir_forw_mv_table = NULL;
 848     s->b_bidir_back_mv_table = NULL;
 849     s->b_direct_mv_table     = NULL;
 850     for (i = 0; i < 2; i++) {
 851         for (j = 0; j < 2; j++) {
 852             for (k = 0; k < 2; k++) {
 853                 s->b_field_mv_table_base[i][j][k] = NULL;
 854                 s->b_field_mv_table[i][j][k] = NULL;
 855             }
 856             s->b_field_select_table[i][j] = NULL;
 857             s->p_field_mv_table_base[i][j] = NULL;
 858             s->p_field_mv_table[i][j] = NULL;
 859         }
 860         s->p_field_select_table[i] = NULL;
 861     }
 862
 863     s->dc_val_base = NULL;
 864     s->coded_block_base = NULL;
 865     s->mbintra_table = NULL;
 866     s->cbp_table = NULL;
 867     s->pred_dir_table = NULL;
 868
 869     s->mbskip_table = NULL;
 870
 871     s->er.error_status_table = NULL;
 872     s->er.er_temp_buffer = NULL;
 873     s->mb_index2xy = NULL;
 874     s->lambda_table = NULL;
 875
 876     s->cplx_tab = NULL;
 877     s->bits_tab = NULL;
 878 }
 879
 880 /**
 881  * init common structure for both encoder and decoder.
 882  * this assumes that some variables like width/height are already set
 883  */
 884 av_cold int ff_mpv_common_init(MpegEncContext *s)
 885 {
 886     int i, ret;
 887     int nb_slices = (HAVE_THREADS &&
 888                      s->avctx->active_thread_type & FF_THREAD_SLICE) ?
 889                     s->avctx->thread_count : 1;
 890
 891     clear_context(s);
 892
 893     if (s->encoding && s->avctx->slices)
 894         nb_slices = s->avctx->slices;
 895
 896     if (s->codec_id == AV_CODEC_ID_MPEG2VIDEO && !s->progressive_sequence)
 897         s->mb_height = (s->height + 31) / 32 * 2;
 898     else
 899         s->mb_height = (s->height + 15) / 16;
 900
 901     if (s->avctx->pix_fmt == AV_PIX_FMT_NONE) {
 902         av_log(s->avctx, AV_LOG_ERROR,
 903                "decoding to AV_PIX_FMT_NONE is not supported.\n");
 904         return AVERROR(EINVAL);
 905     }
 906
 907     if (nb_slices > MAX_THREADS || (nb_slices > s->mb_height && s->mb_height)) {
 908         int max_slices;
 909         if (s->mb_height)
 910             max_slices = FFMIN(MAX_THREADS, s->mb_height);
 911         else
 912             max_slices = MAX_THREADS;
 913         av_log(s->avctx, AV_LOG_WARNING, "too many threads/slices (%d),"
 914                " reducing to %d\n", nb_slices, max_slices);
 915         nb_slices = max_slices;
 916     }
 917
 918     if ((s->width || s->height) &&
 919         av_image_check_size(s->width, s->height, 0, s->avctx))
 920         return AVERROR(EINVAL);
 921
 922     dct_init(s);
 923
 924     /* set chroma shifts */
 925     ret = av_pix_fmt_get_chroma_sub_sample(s->avctx->pix_fmt,
 926                                            &s->chroma_x_shift,
 927                                            &s->chroma_y_shift);
 928     if (ret)
 929         return ret;
 930
 931     if (!FF_ALLOCZ_TYPED_ARRAY(s->picture, MAX_PICTURE_COUNT))
 932         return AVERROR(ENOMEM);
 933     for (i = 0; i < MAX_PICTURE_COUNT; i++) {
 934         s->picture[i].f = av_frame_alloc();
 935         if (!s->picture[i].f)
 936             return AVERROR(ENOMEM);
 937     }
 938
 939     if (!(s->next_picture.f    = av_frame_alloc()) ||
 940         !(s->last_picture.f    = av_frame_alloc()) ||
 941         !(s->current_picture.f = av_frame_alloc()) ||
 942         !(s->new_picture.f     = av_frame_alloc()))
 943         return AVERROR(ENOMEM);
 944
 945     if ((ret = init_context_frame(s)))
 946         return AVERROR(ENOMEM);
 947
 948     s->parse_context.state = -1;
 949
 950     s->context_initialized = 1;
 951     memset(s->thread_context, 0, sizeof(s->thread_context));
 952     s->thread_context[0]   = s;
 953
 954 //     if (s->width && s->height) {
 955     if (nb_slices > 1) {
 956         for (i = 0; i < nb_slices; i++) {
 957             if (i) {
 958                 s->thread_context[i] = av_memdup(s, sizeof(MpegEncContext));
 959                 if (!s->thread_context[i])
 960                     return AVERROR(ENOMEM);
 961             }
 962             if ((ret = init_duplicate_context(s->thread_context[i])) < 0)
 963                 return ret;
 964             s->thread_context[i]->start_mb_y =
 965                 (s->mb_height * (i) + nb_slices / 2) / nb_slices;
 966             s->thread_context[i]->end_mb_y   =
 967                 (s->mb_height * (i + 1) + nb_slices / 2) / nb_slices;
 968         }
 969     } else {
 970         if ((ret = init_duplicate_context(s)) < 0)
 971             return ret;
 972         s->start_mb_y = 0;
 973         s->end_mb_y   = s->mb_height;
 974     }
 975     s->slice_context_count = nb_slices;
 976 //     }
 977
 978     return 0;
 979 }
 980
 981 /**
 982  * Frees and resets MpegEncContext fields depending on the resolution.
 983  * Is used during resolution changes to avoid a full reinitialization of the
 984  * codec.
 985  */
 986 static void free_context_frame(MpegEncContext *s)
 987 {
 988     int i, j, k;
 989
 990     av_freep(&s->mb_type);
 991     av_freep(&s->p_mv_table_base);
 992     av_freep(&s->b_forw_mv_table_base);
 993     av_freep(&s->b_back_mv_table_base);
 994     av_freep(&s->b_bidir_forw_mv_table_base);
 995     av_freep(&s->b_bidir_back_mv_table_base);
 996     av_freep(&s->b_direct_mv_table_base);
 997     s->p_mv_table            = NULL;
 998     s->b_forw_mv_table       = NULL;
 999     s->b_back_mv_table       = NULL;
1000     s->b_bidir_forw_mv_table = NULL;
1001     s->b_bidir_back_mv_table = NULL;
1002     s->b_direct_mv_table     = NULL;
1003     for (i = 0; i < 2; i++) {
1004         for (j = 0; j < 2; j++) {
1005             for (k = 0; k < 2; k++) {
1006                 av_freep(&s->b_field_mv_table_base[i][j][k]);
1007                 s->b_field_mv_table[i][j][k] = NULL;
1008             }
1009             av_freep(&s->b_field_select_table[i][j]);
1010             av_freep(&s->p_field_mv_table_base[i][j]);
1011             s->p_field_mv_table[i][j] = NULL;
1012         }
1013         av_freep(&s->p_field_select_table[i]);
1014     }
1015
1016     av_freep(&s->dc_val_base);
1017     av_freep(&s->coded_block_base);
1018     av_freep(&s->mbintra_table);
1019     av_freep(&s->cbp_table);
1020     av_freep(&s->pred_dir_table);
1021
1022     av_freep(&s->mbskip_table);
1023
1024     av_freep(&s->er.error_status_table);
1025     av_freep(&s->er.er_temp_buffer);
1026     av_freep(&s->mb_index2xy);
1027     av_freep(&s->lambda_table);
1028
1029     av_freep(&s->cplx_tab);
1030     av_freep(&s->bits_tab);
1031
1032     s->linesize = s->uvlinesize = 0;
1033 }
1034
1035 int ff_mpv_common_frame_size_change(MpegEncContext *s)
1036 {
1037     int i, err = 0;
1038
1039     if (!s->context_initialized)
1040         return AVERROR(EINVAL);
1041
1042     if (s->slice_context_count > 1) {
1043         for (i = 0; i < s->slice_context_count; i++) {
1044             free_duplicate_context(s->thread_context[i]);
1045         }
1046         for (i = 1; i < s->slice_context_count; i++) {
1047             av_freep(&s->thread_context[i]);
1048         }
1049     } else
1050         free_duplicate_context(s);
1051
1052     free_context_frame(s);
1053
1054     if (s->picture)
1055         for (i = 0; i < MAX_PICTURE_COUNT; i++) {
1056                 s->picture[i].needs_realloc = 1;
1057         }
1058
1059     s->last_picture_ptr         =
1060     s->next_picture_ptr         =
1061     s->current_picture_ptr      = NULL;
1062
1063     // init
1064     if (s->codec_id == AV_CODEC_ID_MPEG2VIDEO && !s->progressive_sequence)
1065         s->mb_height = (s->height + 31) / 32 * 2;
1066     else
1067         s->mb_height = (s->height + 15) / 16;
1068
1069     if ((s->width || s->height) &&
1070         (err = av_image_check_size(s->width, s->height, 0, s->avctx)) < 0)
1071         return err;
1072
1073     if ((err = init_context_frame(s)))
1074         return err;
1075
1076     memset(s->thread_context, 0, sizeof(s->thread_context));
1077     s->thread_context[0]   = s;
1078
1079     if (s->width && s->height) {
1080         int nb_slices = s->slice_context_count;
1081         if (nb_slices > 1) {
1082             for (i = 0; i < nb_slices; i++) {
1083                 if (i) {
1084                     s->thread_context[i] = av_memdup(s, sizeof(MpegEncContext));
1085                     if (!s->thread_context[i]) {
1086                         return AVERROR(ENOMEM);
1087                     }
1088                 }
1089                 if ((err = init_duplicate_context(s->thread_context[i])) < 0)
1090                     return err;
1091                 s->thread_context[i]->start_mb_y =
1092                     (s->mb_height * (i) + nb_slices / 2) / nb_slices;
1093                 s->thread_context[i]->end_mb_y   =
1094                     (s->mb_height * (i + 1) + nb_slices / 2) / nb_slices;
1095             }
1096         } else {
1097             err = init_duplicate_context(s);
1098             if (err < 0)
1099                 return err;
1100             s->start_mb_y = 0;
1101             s->end_mb_y   = s->mb_height;
1102         }
1103         s->slice_context_count = nb_slices;
1104     }
1105
1106     return 0;
1107 }
1108
1109 /* init common structure for both encoder and decoder */
1110 void ff_mpv_common_end(MpegEncContext *s)
1111 {
1112     int i;
1113
1114     if (!s)
1115         return;
1116
1117     if (s->slice_context_count > 1) {
1118         for (i = 0; i < s->slice_context_count; i++) {
1119             free_duplicate_context(s->thread_context[i]);
1120         }
1121         for (i = 1; i < s->slice_context_count; i++) {
1122             av_freep(&s->thread_context[i]);
1123         }
1124         s->slice_context_count = 1;
1125     } else free_duplicate_context(s);
1126
1127     av_freep(&s->parse_context.buffer);
1128     s->parse_context.buffer_size = 0;
1129
1130     av_freep(&s->bitstream_buffer);
1131     s->allocated_bitstream_buffer_size = 0;
1132
1133     if (!s->avctx)
1134         return;
1135
1136     if (s->picture) {
1137         for (i = 0; i < MAX_PICTURE_COUNT; i++) {
1138             ff_free_picture_tables(&s->picture[i]);
1139             ff_mpeg_unref_picture(s->avctx, &s->picture[i]);
1140             av_frame_free(&s->picture[i].f);
1141         }
1142     }
1143     av_freep(&s->picture);
1144     ff_free_picture_tables(&s->last_picture);
1145     ff_mpeg_unref_picture(s->avctx, &s->last_picture);
1146     av_frame_free(&s->last_picture.f);
1147     ff_free_picture_tables(&s->current_picture);
1148     ff_mpeg_unref_picture(s->avctx, &s->current_picture);
1149     av_frame_free(&s->current_picture.f);
1150     ff_free_picture_tables(&s->next_picture);
1151     ff_mpeg_unref_picture(s->avctx, &s->next_picture);
1152     av_frame_free(&s->next_picture.f);
1153     ff_free_picture_tables(&s->new_picture);
1154     ff_mpeg_unref_picture(s->avctx, &s->new_picture);
1155     av_frame_free(&s->new_picture.f);
1156
1157     free_context_frame(s);
1158
1159     s->context_initialized      = 0;
1160     s->last_picture_ptr         =
1161     s->next_picture_ptr         =
1162     s->current_picture_ptr      = NULL;
1163     s->linesize = s->uvlinesize = 0;
1164 }
1165
1166
1167 static void gray_frame(AVFrame *frame)
1168 {
1169     int i, h_chroma_shift, v_chroma_shift;
1170
1171     av_pix_fmt_get_chroma_sub_sample(frame->format, &h_chroma_shift, &v_chroma_shift);
1172
1173     for(i=0; i<frame->height; i++)
1174         memset(frame->data[0] + frame->linesize[0]*i, 0x80, frame->width);
1175     for(i=0; i<AV_CEIL_RSHIFT(frame->height, v_chroma_shift); i++) {
1176         memset(frame->data[1] + frame->linesize[1]*i,
1177                0x80, AV_CEIL_RSHIFT(frame->width, h_chroma_shift));
1178         memset(frame->data[2] + frame->linesize[2]*i,
1179                0x80, AV_CEIL_RSHIFT(frame->width, h_chroma_shift));
1180     }
1181 }
1182
1183 /**
1184  * generic function called after decoding
1185  * the header and before a frame is decoded.
1186  */
1187 int ff_mpv_frame_start(MpegEncContext *s, AVCodecContext *avctx)
1188 {
1189     int i, ret;
1190     Picture *pic;
1191     s->mb_skipped = 0;
1192
1193     if (!ff_thread_can_start_frame(avctx)) {
1194         av_log(avctx, AV_LOG_ERROR, "Attempt to start a frame outside SETUP state\n");
1195         return -1;
1196     }
1197
1198     /* mark & release old frames */
1199     if (s->pict_type != AV_PICTURE_TYPE_B && s->last_picture_ptr &&
1200         s->last_picture_ptr != s->next_picture_ptr &&
1201         s->last_picture_ptr->f->buf[0]) {
1202         ff_mpeg_unref_picture(s->avctx, s->last_picture_ptr);
1203     }
1204
1205     /* release forgotten pictures */
1206     /* if (MPEG-124 / H.263) */
1207     for (i = 0; i < MAX_PICTURE_COUNT; i++) {
1208         if (&s->picture[i] != s->last_picture_ptr &&
1209             &s->picture[i] != s->next_picture_ptr &&
1210             s->picture[i].reference && !s->picture[i].needs_realloc) {
1211             ff_mpeg_unref_picture(s->avctx, &s->picture[i]);
1212         }
1213     }
1214
1215     ff_mpeg_unref_picture(s->avctx, &s->current_picture);
1216     ff_mpeg_unref_picture(s->avctx, &s->last_picture);
1217     ff_mpeg_unref_picture(s->avctx, &s->next_picture);
1218
1219     /* release non reference frames */
1220     for (i = 0; i < MAX_PICTURE_COUNT; i++) {
1221         if (!s->picture[i].reference)
1222             ff_mpeg_unref_picture(s->avctx, &s->picture[i]);
1223     }
1224
1225     if (s->current_picture_ptr && !s->current_picture_ptr->f->buf[0]) {
1226         // we already have an unused image
1227         // (maybe it was set before reading the header)
1228         pic = s->current_picture_ptr;
1229     } else {
1230         i   = ff_find_unused_picture(s->avctx, s->picture, 0);
1231         if (i < 0) {
1232             av_log(s->avctx, AV_LOG_ERROR, "no frame buffer available\n");
1233             return i;
1234         }
1235         pic = &s->picture[i];
1236     }
1237
1238     pic->reference = 0;
1239     if (!s->droppable) {
1240         if (s->pict_type != AV_PICTURE_TYPE_B)
1241             pic->reference = 3;
1242     }
1243
1244     pic->f->coded_picture_number = s->coded_picture_number++;
1245
1246     if (alloc_picture(s, pic) < 0)
1247         return -1;
1248
1249     s->current_picture_ptr = pic;
1250     // FIXME use only the vars from current_pic
1251     s->current_picture_ptr->f->top_field_first = s->top_field_first;
1252     if (s->codec_id == AV_CODEC_ID_MPEG1VIDEO ||
1253         s->codec_id == AV_CODEC_ID_MPEG2VIDEO) {
1254         if (s->picture_structure != PICT_FRAME)
1255             s->current_picture_ptr->f->top_field_first =
1256                 (s->picture_structure == PICT_TOP_FIELD) == s->first_field;
1257     }
1258     s->current_picture_ptr->f->interlaced_frame = !s->progressive_frame &&
1259                                                  !s->progressive_sequence;
1260     s->current_picture_ptr->field_picture      =  s->picture_structure != PICT_FRAME;
1261
1262     s->current_picture_ptr->f->pict_type = s->pict_type;
1263     // if (s->avctx->flags && AV_CODEC_FLAG_QSCALE)
1264     //     s->current_picture_ptr->quality = s->new_picture_ptr->quality;
1265     s->current_picture_ptr->f->key_frame = s->pict_type == AV_PICTURE_TYPE_I;
1266
1267     if ((ret = ff_mpeg_ref_picture(s->avctx, &s->current_picture,
1268                                    s->current_picture_ptr)) < 0)
1269         return ret;
1270
1271     if (s->pict_type != AV_PICTURE_TYPE_B) {
1272         s->last_picture_ptr = s->next_picture_ptr;
1273         if (!s->droppable)
1274             s->next_picture_ptr = s->current_picture_ptr;
1275     }
1276     ff_dlog(s->avctx, "L%p N%p C%p L%p N%p C%p type:%d drop:%d\n",
1277             s->last_picture_ptr, s->next_picture_ptr,s->current_picture_ptr,
1278             s->last_picture_ptr    ? s->last_picture_ptr->f->data[0]    : NULL,
1279             s->next_picture_ptr    ? s->next_picture_ptr->f->data[0]    : NULL,
1280             s->current_picture_ptr ? s->current_picture_ptr->f->data[0] : NULL,
1281             s->pict_type, s->droppable);
1282
1283     if ((!s->last_picture_ptr || !s->last_picture_ptr->f->buf[0]) &&
1284         (s->pict_type != AV_PICTURE_TYPE_I)) {
1285         int h_chroma_shift, v_chroma_shift;
1286         av_pix_fmt_get_chroma_sub_sample(s->avctx->pix_fmt,
1287                                          &h_chroma_shift, &v_chroma_shift);
1288         if (s->pict_type == AV_PICTURE_TYPE_B && s->next_picture_ptr && s->next_picture_ptr->f->buf[0])
1289             av_log(avctx, AV_LOG_DEBUG,
1290                    "allocating dummy last picture for B frame\n");
1291         else if (s->pict_type != AV_PICTURE_TYPE_I)
1292             av_log(avctx, AV_LOG_ERROR,
1293                    "warning: first frame is no keyframe\n");
1294
1295         /* Allocate a dummy frame */
1296         i = ff_find_unused_picture(s->avctx, s->picture, 0);
1297         if (i < 0) {
1298             av_log(s->avctx, AV_LOG_ERROR, "no frame buffer available\n");
1299             return i;
1300         }
1301         s->last_picture_ptr = &s->picture[i];
1302
1303         s->last_picture_ptr->reference   = 3;
1304         s->last_picture_ptr->f->key_frame = 0;
1305         s->last_picture_ptr->f->pict_type = AV_PICTURE_TYPE_P;
1306
1307         if (alloc_picture(s, s->last_picture_ptr) < 0) {
1308             s->last_picture_ptr = NULL;
1309             return -1;
1310         }
1311
1312         if (!avctx->hwaccel) {
1313             for(i=0; i<avctx->height; i++)
1314                 memset(s->last_picture_ptr->f->data[0] + s->last_picture_ptr->f->linesize[0]*i,
1315                        0x80, avctx->width);
1316             if (s->last_picture_ptr->f->data[2]) {
1317                 for(i=0; i<AV_CEIL_RSHIFT(avctx->height, v_chroma_shift); i++) {
1318                     memset(s->last_picture_ptr->f->data[1] + s->last_picture_ptr->f->linesize[1]*i,
1319                         0x80, AV_CEIL_RSHIFT(avctx->width, h_chroma_shift));
1320                     memset(s->last_picture_ptr->f->data[2] + s->last_picture_ptr->f->linesize[2]*i,
1321                         0x80, AV_CEIL_RSHIFT(avctx->width, h_chroma_shift));
1322                 }
1323             }
1324
1325             if(s->codec_id == AV_CODEC_ID_FLV1 || s->codec_id == AV_CODEC_ID_H263){
1326                 for(i=0; i<avctx->height; i++)
1327                 memset(s->last_picture_ptr->f->data[0] + s->last_picture_ptr->f->linesize[0]*i, 16, avctx->width);
1328             }
1329         }
1330
1331         ff_thread_report_progress(&s->last_picture_ptr->tf, INT_MAX, 0);
1332         ff_thread_report_progress(&s->last_picture_ptr->tf, INT_MAX, 1);
1333     }
1334     if ((!s->next_picture_ptr || !s->next_picture_ptr->f->buf[0]) &&
1335         s->pict_type == AV_PICTURE_TYPE_B) {
1336         /* Allocate a dummy frame */
1337         i = ff_find_unused_picture(s->avctx, s->picture, 0);
1338         if (i < 0) {
1339             av_log(s->avctx, AV_LOG_ERROR, "no frame buffer available\n");
1340             return i;
1341         }
1342         s->next_picture_ptr = &s->picture[i];
1343
1344         s->next_picture_ptr->reference   = 3;
1345         s->next_picture_ptr->f->key_frame = 0;
1346         s->next_picture_ptr->f->pict_type = AV_PICTURE_TYPE_P;
1347
1348         if (alloc_picture(s, s->next_picture_ptr) < 0) {
1349             s->next_picture_ptr = NULL;
1350             return -1;
1351         }
1352         ff_thread_report_progress(&s->next_picture_ptr->tf, INT_MAX, 0);
1353         ff_thread_report_progress(&s->next_picture_ptr->tf, INT_MAX, 1);
1354     }
1355
1356 #if 0 // BUFREF-FIXME
1357     memset(s->last_picture.f->data, 0, sizeof(s->last_picture.f->data));
1358     memset(s->next_picture.f->data, 0, sizeof(s->next_picture.f->data));
1359 #endif
1360     if (s->last_picture_ptr) {
1361         if (s->last_picture_ptr->f->buf[0] &&
1362             (ret = ff_mpeg_ref_picture(s->avctx, &s->last_picture,
1363                                        s->last_picture_ptr)) < 0)
1364             return ret;
1365     }
1366     if (s->next_picture_ptr) {
1367         if (s->next_picture_ptr->f->buf[0] &&
1368             (ret = ff_mpeg_ref_picture(s->avctx, &s->next_picture,
1369                                        s->next_picture_ptr)) < 0)
1370             return ret;
1371     }
1372
1373     av_assert0(s->pict_type == AV_PICTURE_TYPE_I || (s->last_picture_ptr &&
1374                                                  s->last_picture_ptr->f->buf[0]));
1375
1376     if (s->picture_structure!= PICT_FRAME) {
1377         int i;
1378         for (i = 0; i < 4; i++) {
1379             if (s->picture_structure == PICT_BOTTOM_FIELD) {
1380                 s->current_picture.f->data[i] +=
1381                     s->current_picture.f->linesize[i];
1382             }
1383             s->current_picture.f->linesize[i] *= 2;
1384             s->last_picture.f->linesize[i]    *= 2;
1385             s->next_picture.f->linesize[i]    *= 2;
1386         }
1387     }
1388
1389     /* set dequantizer, we can't do it during init as
1390      * it might change for MPEG-4 and we can't do it in the header
1391      * decode as init is not called for MPEG-4 there yet */
1392     if (s->mpeg_quant || s->codec_id == AV_CODEC_ID_MPEG2VIDEO) {
1393         s->dct_unquantize_intra = s->dct_unquantize_mpeg2_intra;
1394         s->dct_unquantize_inter = s->dct_unquantize_mpeg2_inter;
1395     } else if (s->out_format == FMT_H263 || s->out_format == FMT_H261) {
1396         s->dct_unquantize_intra = s->dct_unquantize_h263_intra;
1397         s->dct_unquantize_inter = s->dct_unquantize_h263_inter;
1398     } else {
1399         s->dct_unquantize_intra = s->dct_unquantize_mpeg1_intra;
1400         s->dct_unquantize_inter = s->dct_unquantize_mpeg1_inter;
1401     }
1402
1403     if (s->avctx->debug & FF_DEBUG_NOMC) {
1404         gray_frame(s->current_picture_ptr->f);
1405     }
1406
1407     return 0;
1408 }
1409
1410 /* called after a frame has been decoded. */
1411 void ff_mpv_frame_end(MpegEncContext *s)
1412 {
1413     emms_c();
1414
1415     if (s->current_picture.reference)
1416         ff_thread_report_progress(&s->current_picture_ptr->tf, INT_MAX, 0);
1417 }
1418
1419 void ff_print_debug_info(MpegEncContext *s, Picture *p, AVFrame *pict)
1420 {
1421     ff_print_debug_info2(s->avctx, pict, s->mbskip_table, p->mb_type,
1422                          p->qscale_table, p->motion_val, &s->low_delay,
1423                          s->mb_width, s->mb_height, s->mb_stride, s->quarter_sample);
1424 }
1425
1426 int ff_mpv_export_qp_table(MpegEncContext *s, AVFrame *f, Picture *p, int qp_type)
1427 {
1428     AVBufferRef *ref = av_buffer_ref(p->qscale_table_buf);
1429     int offset = 2*s->mb_stride + 1;
1430     if(!ref)
1431         return AVERROR(ENOMEM);
1432     av_assert0(ref->size >= offset + s->mb_stride * ((f->height+15)/16));
1433     ref->size -= offset;
1434     ref->data += offset;
1435     return av_frame_set_qp_table(f, ref, s->mb_stride, qp_type);
1436 }
1437
1438 static inline int hpel_motion_lowres(MpegEncContext *s,
1439                                      uint8_t *dest, uint8_t *src,
1440                                      int field_based, int field_select,
1441                                      int src_x, int src_y,
1442                                      int width, int height, ptrdiff_t stride,
1443                                      int h_edge_pos, int v_edge_pos,
1444                                      int w, int h, h264_chroma_mc_func *pix_op,
1445                                      int motion_x, int motion_y)
1446 {
1447     const int lowres   = s->avctx->lowres;
1448     const int op_index = FFMIN(lowres, 3);
1449     const int s_mask   = (2 << lowres) - 1;
1450     int emu = 0;
1451     int sx, sy;
1452
1453     if (s->quarter_sample) {
1454         motion_x /= 2;
1455         motion_y /= 2;
1456     }
1457
1458     sx = motion_x & s_mask;
1459     sy = motion_y & s_mask;
1460     src_x += motion_x >> lowres + 1;
1461     src_y += motion_y >> lowres + 1;
1462
1463     src   += src_y * stride + src_x;
1464
1465     if ((unsigned)src_x > FFMAX( h_edge_pos - (!!sx) - w,                 0) ||
1466         (unsigned)src_y > FFMAX((v_edge_pos >> field_based) - (!!sy) - h, 0)) {
1467         s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, src,
1468                                  s->linesize, s->linesize,
1469                                  w + 1, (h + 1) << field_based,
1470                                  src_x, src_y   << field_based,
1471                                  h_edge_pos, v_edge_pos);
1472         src = s->sc.edge_emu_buffer;
1473         emu = 1;
1474     }
1475
1476     sx = (sx << 2) >> lowres;
1477     sy = (sy << 2) >> lowres;
1478     if (field_select)
1479         src += s->linesize;
1480     pix_op[op_index](dest, src, stride, h, sx, sy);
1481     return emu;
1482 }
1483
1484 /* apply one mpeg motion vector to the three components */
1485 static av_always_inline void mpeg_motion_lowres(MpegEncContext *s,
1486                                                 uint8_t *dest_y,
1487                                                 uint8_t *dest_cb,
1488                                                 uint8_t *dest_cr,
1489                                                 int field_based,
1490                                                 int bottom_field,
1491                                                 int field_select,
1492                                                 uint8_t **ref_picture,
1493                                                 h264_chroma_mc_func *pix_op,
1494                                                 int motion_x, int motion_y,
1495                                                 int h, int mb_y)
1496 {
1497     uint8_t *ptr_y, *ptr_cb, *ptr_cr;
1498     int mx, my, src_x, src_y, uvsrc_x, uvsrc_y, sx, sy, uvsx, uvsy;
1499     ptrdiff_t uvlinesize, linesize;
1500     const int lowres     = s->avctx->lowres;
1501     const int op_index   = FFMIN(lowres-1+s->chroma_x_shift, 3);
1502     const int block_s    = 8>>lowres;
1503     const int s_mask     = (2 << lowres) - 1;
1504     const int h_edge_pos = s->h_edge_pos >> lowres;
1505     const int v_edge_pos = s->v_edge_pos >> lowres;
1506     linesize   = s->current_picture.f->linesize[0] << field_based;
1507     uvlinesize = s->current_picture.f->linesize[1] << field_based;
1508
1509     // FIXME obviously not perfect but qpel will not work in lowres anyway
1510     if (s->quarter_sample) {
1511         motion_x /= 2;
1512         motion_y /= 2;
1513     }
1514
1515     if(field_based){
1516         motion_y += (bottom_field - field_select)*((1 << lowres)-1);
1517     }
1518
1519     sx = motion_x & s_mask;
1520     sy = motion_y & s_mask;
1521     src_x = s->mb_x * 2 * block_s + (motion_x >> lowres + 1);
1522     src_y = (mb_y * 2 * block_s >> field_based) + (motion_y >> lowres + 1);
1523
1524     if (s->out_format == FMT_H263) {
1525         uvsx    = ((motion_x >> 1) & s_mask) | (sx & 1);
1526         uvsy    = ((motion_y >> 1) & s_mask) | (sy & 1);
1527         uvsrc_x = src_x >> 1;
1528         uvsrc_y = src_y >> 1;
1529     } else if (s->out_format == FMT_H261) {
1530         // even chroma mv's are full pel in H261
1531         mx      = motion_x / 4;
1532         my      = motion_y / 4;
1533         uvsx    = (2 * mx) & s_mask;
1534         uvsy    = (2 * my) & s_mask;
1535         uvsrc_x = s->mb_x * block_s + (mx >> lowres);
1536         uvsrc_y =    mb_y * block_s + (my >> lowres);
1537     } else {
1538         if(s->chroma_y_shift){
1539             mx      = motion_x / 2;
1540             my      = motion_y / 2;
1541             uvsx    = mx & s_mask;
1542             uvsy    = my & s_mask;
1543             uvsrc_x = s->mb_x * block_s                 + (mx >> lowres + 1);
1544             uvsrc_y =   (mb_y * block_s >> field_based) + (my >> lowres + 1);
1545         } else {
1546             if(s->chroma_x_shift){
1547             //Chroma422
1548                 mx = motion_x / 2;
1549                 uvsx = mx & s_mask;
1550                 uvsy = motion_y & s_mask;
1551                 uvsrc_y = src_y;
1552                 uvsrc_x = s->mb_x*block_s               + (mx >> (lowres+1));
1553             } else {
1554             //Chroma444
1555                 uvsx = motion_x & s_mask;
1556                 uvsy = motion_y & s_mask;
1557                 uvsrc_x = src_x;
1558                 uvsrc_y = src_y;
1559             }
1560         }
1561     }
1562
1563     ptr_y  = ref_picture[0] + src_y   * linesize   + src_x;
1564     ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x;
1565     ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x;
1566
1567     if ((unsigned) src_x > FFMAX( h_edge_pos - (!!sx) - 2 * block_s,       0) || uvsrc_y<0 ||
1568         (unsigned) src_y > FFMAX((v_edge_pos >> field_based) - (!!sy) - h, 0)) {
1569         s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, ptr_y,
1570                                  linesize >> field_based, linesize >> field_based,
1571                                  17, 17 + field_based,
1572                                 src_x, src_y << field_based, h_edge_pos,
1573                                 v_edge_pos);
1574         ptr_y = s->sc.edge_emu_buffer;
1575         if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY)) {
1576             uint8_t *ubuf = s->sc.edge_emu_buffer + 18 * s->linesize;
1577             uint8_t *vbuf =ubuf + 10 * s->uvlinesize;
1578             if (s->workaround_bugs & FF_BUG_IEDGE)
1579                 vbuf -= s->uvlinesize;
1580             s->vdsp.emulated_edge_mc(ubuf,  ptr_cb,
1581                                      uvlinesize >> field_based, uvlinesize >> field_based,
1582                                      9, 9 + field_based,
1583                                     uvsrc_x, uvsrc_y << field_based,
1584                                     h_edge_pos >> 1, v_edge_pos >> 1);
1585             s->vdsp.emulated_edge_mc(vbuf,  ptr_cr,
1586                                      uvlinesize >> field_based,uvlinesize >> field_based,
1587                                      9, 9 + field_based,
1588                                     uvsrc_x, uvsrc_y << field_based,
1589                                     h_edge_pos >> 1, v_edge_pos >> 1);
1590             ptr_cb = ubuf;
1591             ptr_cr = vbuf;
1592         }
1593     }
1594
1595     // FIXME use this for field pix too instead of the obnoxious hack which changes picture.f->data
1596     if (bottom_field) {
1597         dest_y  += s->linesize;
1598         dest_cb += s->uvlinesize;
1599         dest_cr += s->uvlinesize;
1600     }
1601
1602     if (field_select) {
1603         ptr_y   += s->linesize;
1604         ptr_cb  += s->uvlinesize;
1605         ptr_cr  += s->uvlinesize;
1606     }
1607
1608     sx = (sx << 2) >> lowres;
1609     sy = (sy << 2) >> lowres;
1610     pix_op[lowres - 1](dest_y, ptr_y, linesize, h, sx, sy);
1611
1612     if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY)) {
1613         int hc = s->chroma_y_shift ? (h+1-bottom_field)>>1 : h;
1614         uvsx = (uvsx << 2) >> lowres;
1615         uvsy = (uvsy << 2) >> lowres;
1616         if (hc) {
1617             pix_op[op_index](dest_cb, ptr_cb, uvlinesize, hc, uvsx, uvsy);
1618             pix_op[op_index](dest_cr, ptr_cr, uvlinesize, hc, uvsx, uvsy);
1619         }
1620     }
1621     // FIXME h261 lowres loop filter
1622 }
1623
1624 static inline void chroma_4mv_motion_lowres(MpegEncContext *s,
1625                                             uint8_t *dest_cb, uint8_t *dest_cr,
1626                                             uint8_t **ref_picture,
1627                                             h264_chroma_mc_func * pix_op,
1628                                             int mx, int my)
1629 {
1630     const int lowres     = s->avctx->lowres;
1631     const int op_index   = FFMIN(lowres, 3);
1632     const int block_s    = 8 >> lowres;
1633     const int s_mask     = (2 << lowres) - 1;
1634     const int h_edge_pos = s->h_edge_pos >> lowres + 1;
1635     const int v_edge_pos = s->v_edge_pos >> lowres + 1;
1636     int emu = 0, src_x, src_y, sx, sy;
1637     ptrdiff_t offset;
1638     uint8_t *ptr;
1639
1640     if (s->quarter_sample) {
1641         mx /= 2;
1642         my /= 2;
1643     }
1644
1645     /* In case of 8X8, we construct a single chroma motion vector
1646        with a special rounding */
1647     mx = ff_h263_round_chroma(mx);
1648     my = ff_h263_round_chroma(my);
1649
1650     sx = mx & s_mask;
1651     sy = my & s_mask;
1652     src_x = s->mb_x * block_s + (mx >> lowres + 1);
1653     src_y = s->mb_y * block_s + (my >> lowres + 1);
1654
1655     offset = src_y * s->uvlinesize + src_x;
1656     ptr = ref_picture[1] + offset;
1657     if ((unsigned) src_x > FFMAX(h_edge_pos - (!!sx) - block_s, 0) ||
1658         (unsigned) src_y > FFMAX(v_edge_pos - (!!sy) - block_s, 0)) {
1659         s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, ptr,
1660                                  s->uvlinesize, s->uvlinesize,
1661                                  9, 9,
1662                                  src_x, src_y, h_edge_pos, v_edge_pos);
1663         ptr = s->sc.edge_emu_buffer;
1664         emu = 1;
1665     }
1666     sx = (sx << 2) >> lowres;
1667     sy = (sy << 2) >> lowres;
1668     pix_op[op_index](dest_cb, ptr, s->uvlinesize, block_s, sx, sy);
1669
1670     ptr = ref_picture[2] + offset;
1671     if (emu) {
1672         s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, ptr,
1673                                  s->uvlinesize, s->uvlinesize,
1674                                  9, 9,
1675                                  src_x, src_y, h_edge_pos, v_edge_pos);
1676         ptr = s->sc.edge_emu_buffer;
1677     }
1678     pix_op[op_index](dest_cr, ptr, s->uvlinesize, block_s, sx, sy);
1679 }
1680
1681 /**
1682  * motion compensation of a single macroblock
1683  * @param s context
1684  * @param dest_y luma destination pointer
1685  * @param dest_cb chroma cb/u destination pointer
1686  * @param dest_cr chroma cr/v destination pointer
1687  * @param dir direction (0->forward, 1->backward)
1688  * @param ref_picture array[3] of pointers to the 3 planes of the reference picture
1689  * @param pix_op halfpel motion compensation function (average or put normally)
1690  * the motion vectors are taken from s->mv and the MV type from s->mv_type
1691  */
1692 static inline void MPV_motion_lowres(MpegEncContext *s,
1693                                      uint8_t *dest_y, uint8_t *dest_cb,
1694                                      uint8_t *dest_cr,
1695                                      int dir, uint8_t **ref_picture,
1696                                      h264_chroma_mc_func *pix_op)
1697 {
1698     int mx, my;
1699     int mb_x, mb_y, i;
1700     const int lowres  = s->avctx->lowres;
1701     const int block_s = 8 >>lowres;
1702
1703     mb_x = s->mb_x;
1704     mb_y = s->mb_y;
1705
1706     switch (s->mv_type) {
1707     case MV_TYPE_16X16:
1708         mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
1709                            0, 0, 0,
1710                            ref_picture, pix_op,
1711                            s->mv[dir][0][0], s->mv[dir][0][1],
1712                            2 * block_s, mb_y);
1713         break;
1714     case MV_TYPE_8X8:
1715         mx = 0;
1716         my = 0;
1717         for (i = 0; i < 4; i++) {
1718             hpel_motion_lowres(s, dest_y + ((i & 1) + (i >> 1) *
1719                                s->linesize) * block_s,
1720                                ref_picture[0], 0, 0,
1721                                (2 * mb_x + (i & 1)) * block_s,
1722                                (2 * mb_y + (i >> 1)) * block_s,
1723                                s->width, s->height, s->linesize,
1724                                s->h_edge_pos >> lowres, s->v_edge_pos >> lowres,
1725                                block_s, block_s, pix_op,
1726                                s->mv[dir][i][0], s->mv[dir][i][1]);
1727
1728             mx += s->mv[dir][i][0];
1729             my += s->mv[dir][i][1];
1730         }
1731
1732         if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY))
1733             chroma_4mv_motion_lowres(s, dest_cb, dest_cr, ref_picture,
1734                                      pix_op, mx, my);
1735         break;
1736     case MV_TYPE_FIELD:
1737         if (s->picture_structure == PICT_FRAME) {
1738             /* top field */
1739             mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
1740                                1, 0, s->field_select[dir][0],
1741                                ref_picture, pix_op,
1742                                s->mv[dir][0][0], s->mv[dir][0][1],
1743                                block_s, mb_y);
1744             /* bottom field */
1745             mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
1746                                1, 1, s->field_select[dir][1],
1747                                ref_picture, pix_op,
1748                                s->mv[dir][1][0], s->mv[dir][1][1],
1749                                block_s, mb_y);
1750         } else {
1751             if (s->picture_structure != s->field_select[dir][0] + 1 &&
1752                 s->pict_type != AV_PICTURE_TYPE_B && !s->first_field) {
1753                 ref_picture = s->current_picture_ptr->f->data;
1754
1755             }
1756             mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
1757                                0, 0, s->field_select[dir][0],
1758                                ref_picture, pix_op,
1759                                s->mv[dir][0][0],
1760                                s->mv[dir][0][1], 2 * block_s, mb_y >> 1);
1761             }
1762         break;
1763     case MV_TYPE_16X8:
1764         for (i = 0; i < 2; i++) {
1765             uint8_t **ref2picture;
1766
1767             if (s->picture_structure == s->field_select[dir][i] + 1 ||
1768                 s->pict_type == AV_PICTURE_TYPE_B || s->first_field) {
1769                 ref2picture = ref_picture;
1770             } else {
1771                 ref2picture = s->current_picture_ptr->f->data;
1772             }
1773
1774             mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
1775                                0, 0, s->field_select[dir][i],
1776                                ref2picture, pix_op,
1777                                s->mv[dir][i][0], s->mv[dir][i][1] +
1778                                2 * block_s * i, block_s, mb_y >> 1);
1779
1780             dest_y  +=  2 * block_s *  s->linesize;
1781             dest_cb += (2 * block_s >> s->chroma_y_shift) * s->uvlinesize;
1782             dest_cr += (2 * block_s >> s->chroma_y_shift) * s->uvlinesize;
1783         }
1784         break;
1785     case MV_TYPE_DMV:
1786         if (s->picture_structure == PICT_FRAME) {
1787             for (i = 0; i < 2; i++) {
1788                 int j;
1789                 for (j = 0; j < 2; j++) {
1790                     mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
1791                                        1, j, j ^ i,
1792                                        ref_picture, pix_op,
1793                                        s->mv[dir][2 * i + j][0],
1794                                        s->mv[dir][2 * i + j][1],
1795                                        block_s, mb_y);
1796                 }
1797                 pix_op = s->h264chroma.avg_h264_chroma_pixels_tab;
1798             }
1799         } else {
1800             for (i = 0; i < 2; i++) {
1801                 mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
1802                                    0, 0, s->picture_structure != i + 1,
1803                                    ref_picture, pix_op,
1804                                    s->mv[dir][2 * i][0],s->mv[dir][2 * i][1],
1805                                    2 * block_s, mb_y >> 1);
1806
1807                 // after put we make avg of the same block
1808                 pix_op = s->h264chroma.avg_h264_chroma_pixels_tab;
1809
1810                 // opposite parity is always in the same
1811                 // frame if this is second field
1812                 if (!s->first_field) {
1813                     ref_picture = s->current_picture_ptr->f->data;
1814                 }
1815             }
1816         }
1817         break;
1818     default:
1819         av_assert2(0);
1820     }
1821 }
1822
1823 /**
1824  * find the lowest MB row referenced in the MVs
1825  */
1826 static int lowest_referenced_row(MpegEncContext *s, int dir)
1827 {
1828     int my_max = INT_MIN, my_min = INT_MAX, qpel_shift = !s->quarter_sample;
1829     int my, off, i, mvs;
1830
1831     if (s->picture_structure != PICT_FRAME || s->mcsel)
1832         goto unhandled;
1833
1834     switch (s->mv_type) {
1835         case MV_TYPE_16X16:
1836             mvs = 1;
1837             break;
1838         case MV_TYPE_16X8:
1839             mvs = 2;
1840             break;
1841         case MV_TYPE_8X8:
1842             mvs = 4;
1843             break;
1844         default:
1845             goto unhandled;
1846     }
1847
1848     for (i = 0; i < mvs; i++) {
1849         my = s->mv[dir][i][1];
1850         my_max = FFMAX(my_max, my);
1851         my_min = FFMIN(my_min, my);
1852     }
1853
1854     off = ((FFMAX(-my_min, my_max)<<qpel_shift) + 63) >> 6;
1855
1856     return av_clip(s->mb_y + off, 0, s->mb_height - 1);
1857 unhandled:
1858     return s->mb_height-1;
1859 }
1860
1861 /* put block[] to dest[] */
1862 static inline void put_dct(MpegEncContext *s,
1863                            int16_t *block, int i, uint8_t *dest, int line_size, int qscale)
1864 {
1865     s->dct_unquantize_intra(s, block, i, qscale);
1866     s->idsp.idct_put(dest, line_size, block);
1867 }
1868
1869 /* add block[] to dest[] */
1870 static inline void add_dct(MpegEncContext *s,
1871                            int16_t *block, int i, uint8_t *dest, int line_size)
1872 {
1873     if (s->block_last_index[i] >= 0) {
1874         s->idsp.idct_add(dest, line_size, block);
1875     }
1876 }
1877
1878 static inline void add_dequant_dct(MpegEncContext *s,
1879                            int16_t *block, int i, uint8_t *dest, int line_size, int qscale)
1880 {
1881     if (s->block_last_index[i] >= 0) {
1882         s->dct_unquantize_inter(s, block, i, qscale);
1883
1884         s->idsp.idct_add(dest, line_size, block);
1885     }
1886 }
1887
1888 /**
1889  * Clean dc, ac, coded_block for the current non-intra MB.
1890  */
1891 void ff_clean_intra_table_entries(MpegEncContext *s)
1892 {
1893     int wrap = s->b8_stride;
1894     int xy = s->block_index[0];
1895
1896     s->dc_val[0][xy           ] =
1897     s->dc_val[0][xy + 1       ] =
1898     s->dc_val[0][xy     + wrap] =
1899     s->dc_val[0][xy + 1 + wrap] = 1024;
1900     /* ac pred */
1901     memset(s->ac_val[0][xy       ], 0, 32 * sizeof(int16_t));
1902     memset(s->ac_val[0][xy + wrap], 0, 32 * sizeof(int16_t));
1903     if (s->msmpeg4_version>=3) {
1904         s->coded_block[xy           ] =
1905         s->coded_block[xy + 1       ] =
1906         s->coded_block[xy     + wrap] =
1907         s->coded_block[xy + 1 + wrap] = 0;
1908     }
1909     /* chroma */
1910     wrap = s->mb_stride;
1911     xy = s->mb_x + s->mb_y * wrap;
1912     s->dc_val[1][xy] =
1913     s->dc_val[2][xy] = 1024;
1914     /* ac pred */
1915     memset(s->ac_val[1][xy], 0, 16 * sizeof(int16_t));
1916     memset(s->ac_val[2][xy], 0, 16 * sizeof(int16_t));
1917
1918     s->mbintra_table[xy]= 0;
1919 }
1920
1921 /* generic function called after a macroblock has been parsed by the
1922    decoder or after it has been encoded by the encoder.
1923
1924    Important variables used:
1925    s->mb_intra : true if intra macroblock
1926    s->mv_dir   : motion vector direction
1927    s->mv_type  : motion vector type
1928    s->mv       : motion vector
1929    s->interlaced_dct : true if interlaced dct used (mpeg2)
1930  */
1931 static av_always_inline
1932 void mpv_reconstruct_mb_internal(MpegEncContext *s, int16_t block[12][64],
1933                             int lowres_flag, int is_mpeg12)
1934 {
1935     const int mb_xy = s->mb_y * s->mb_stride + s->mb_x;
1936
1937     if (CONFIG_XVMC &&
1938         s->avctx->hwaccel && s->avctx->hwaccel->decode_mb) {
1939         s->avctx->hwaccel->decode_mb(s);//xvmc uses pblocks
1940         return;
1941     }
1942
1943     if(s->avctx->debug&FF_DEBUG_DCT_COEFF) {
1944        /* print DCT coefficients */
1945        int i,j;
1946        av_log(s->avctx, AV_LOG_DEBUG, "DCT coeffs of MB at %dx%d:\n", s->mb_x, s->mb_y);
1947        for(i=0; i<6; i++){
1948            for(j=0; j<64; j++){
1949                av_log(s->avctx, AV_LOG_DEBUG, "%5d",
1950                       block[i][s->idsp.idct_permutation[j]]);
1951            }
1952            av_log(s->avctx, AV_LOG_DEBUG, "\n");
1953        }
1954     }
1955
1956     s->current_picture.qscale_table[mb_xy] = s->qscale;
1957
1958     /* update DC predictors for P macroblocks */
1959     if (!s->mb_intra) {
1960         if (!is_mpeg12 && (s->h263_pred || s->h263_aic)) {
1961             if(s->mbintra_table[mb_xy])
1962                 ff_clean_intra_table_entries(s);
1963         } else {
1964             s->last_dc[0] =
1965             s->last_dc[1] =
1966             s->last_dc[2] = 128 << s->intra_dc_precision;
1967         }
1968     }
1969     else if (!is_mpeg12 && (s->h263_pred || s->h263_aic))
1970         s->mbintra_table[mb_xy]=1;
1971
1972     if ((s->avctx->flags & AV_CODEC_FLAG_PSNR) || s->frame_skip_threshold || s->frame_skip_factor ||
1973         !(s->encoding && (s->intra_only || s->pict_type == AV_PICTURE_TYPE_B) &&
1974           s->avctx->mb_decision != FF_MB_DECISION_RD)) { // FIXME precalc
1975         uint8_t *dest_y, *dest_cb, *dest_cr;
1976         int dct_linesize, dct_offset;
1977         op_pixels_func (*op_pix)[4];
1978         qpel_mc_func (*op_qpix)[16];
1979         const int linesize   = s->current_picture.f->linesize[0]; //not s->linesize as this would be wrong for field pics
1980         const int uvlinesize = s->current_picture.f->linesize[1];
1981         const int readable= s->pict_type != AV_PICTURE_TYPE_B || s->encoding || s->avctx->draw_horiz_band || lowres_flag;
1982         const int block_size= lowres_flag ? 8>>s->avctx->lowres : 8;
1983
1984         /* avoid copy if macroblock skipped in last frame too */
1985         /* skip only during decoding as we might trash the buffers during encoding a bit */
1986         if(!s->encoding){
1987             uint8_t *mbskip_ptr = &s->mbskip_table[mb_xy];
1988
1989             if (s->mb_skipped) {
1990                 s->mb_skipped= 0;
1991                 av_assert2(s->pict_type!=AV_PICTURE_TYPE_I);
1992                 *mbskip_ptr = 1;
1993             } else if(!s->current_picture.reference) {
1994                 *mbskip_ptr = 1;
1995             } else{
1996                 *mbskip_ptr = 0; /* not skipped */
1997             }
1998         }
1999
2000         dct_linesize = linesize << s->interlaced_dct;
2001         dct_offset   = s->interlaced_dct ? linesize : linesize * block_size;
2002
2003         if(readable){
2004             dest_y=  s->dest[0];
2005             dest_cb= s->dest[1];
2006             dest_cr= s->dest[2];
2007         }else{
2008             dest_y = s->sc.b_scratchpad;
2009             dest_cb= s->sc.b_scratchpad+16*linesize;
2010             dest_cr= s->sc.b_scratchpad+32*linesize;
2011         }
2012
2013         if (!s->mb_intra) {
2014             /* motion handling */
2015             /* decoding or more than one mb_type (MC was already done otherwise) */
2016             if(!s->encoding){
2017
2018                 if(HAVE_THREADS && s->avctx->active_thread_type&FF_THREAD_FRAME) {
2019                     if (s->mv_dir & MV_DIR_FORWARD) {
2020                         ff_thread_await_progress(&s->last_picture_ptr->tf,
2021                                                  lowest_referenced_row(s, 0),
2022                                                  0);
2023                     }
2024                     if (s->mv_dir & MV_DIR_BACKWARD) {
2025                         ff_thread_await_progress(&s->next_picture_ptr->tf,
2026                                                  lowest_referenced_row(s, 1),
2027                                                  0);
2028                     }
2029                 }
2030
2031                 if(lowres_flag){
2032                     h264_chroma_mc_func *op_pix = s->h264chroma.put_h264_chroma_pixels_tab;
2033
2034                     if (s->mv_dir & MV_DIR_FORWARD) {
2035                         MPV_motion_lowres(s, dest_y, dest_cb, dest_cr, 0, s->last_picture.f->data, op_pix);
2036                         op_pix = s->h264chroma.avg_h264_chroma_pixels_tab;
2037                     }
2038                     if (s->mv_dir & MV_DIR_BACKWARD) {
2039                         MPV_motion_lowres(s, dest_y, dest_cb, dest_cr, 1, s->next_picture.f->data, op_pix);
2040                     }
2041                 }else{
2042                     op_qpix = s->me.qpel_put;
2043                     if ((!s->no_rounding) || s->pict_type==AV_PICTURE_TYPE_B){
2044                         op_pix = s->hdsp.put_pixels_tab;
2045                     }else{
2046                         op_pix = s->hdsp.put_no_rnd_pixels_tab;
2047                     }
2048                     if (s->mv_dir & MV_DIR_FORWARD) {
2049                         ff_mpv_motion(s, dest_y, dest_cb, dest_cr, 0, s->last_picture.f->data, op_pix, op_qpix);
2050                         op_pix = s->hdsp.avg_pixels_tab;
2051                         op_qpix= s->me.qpel_avg;
2052                     }
2053                     if (s->mv_dir & MV_DIR_BACKWARD) {
2054                         ff_mpv_motion(s, dest_y, dest_cb, dest_cr, 1, s->next_picture.f->data, op_pix, op_qpix);
2055                     }
2056                 }
2057             }
2058
2059             /* skip dequant / idct if we are really late ;) */
2060             if(s->avctx->skip_idct){
2061                 if(  (s->avctx->skip_idct >= AVDISCARD_NONREF && s->pict_type == AV_PICTURE_TYPE_B)
2062                    ||(s->avctx->skip_idct >= AVDISCARD_NONKEY && s->pict_type != AV_PICTURE_TYPE_I)
2063                    || s->avctx->skip_idct >= AVDISCARD_ALL)
2064                     goto skip_idct;
2065             }
2066
2067             /* add dct residue */
2068             if(s->encoding || !(   s->msmpeg4_version || s->codec_id==AV_CODEC_ID_MPEG1VIDEO || s->codec_id==AV_CODEC_ID_MPEG2VIDEO
2069                                 || (s->codec_id==AV_CODEC_ID_MPEG4 && !s->mpeg_quant))){
2070                 add_dequant_dct(s, block[0], 0, dest_y                          , dct_linesize, s->qscale);
2071                 add_dequant_dct(s, block[1], 1, dest_y              + block_size, dct_linesize, s->qscale);
2072                 add_dequant_dct(s, block[2], 2, dest_y + dct_offset             , dct_linesize, s->qscale);
2073                 add_dequant_dct(s, block[3], 3, dest_y + dct_offset + block_size, dct_linesize, s->qscale);
2074
2075                 if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY)) {
2076                     if (s->chroma_y_shift){
2077                         add_dequant_dct(s, block[4], 4, dest_cb, uvlinesize, s->chroma_qscale);
2078                         add_dequant_dct(s, block[5], 5, dest_cr, uvlinesize, s->chroma_qscale);
2079                     }else{
2080                         dct_linesize >>= 1;
2081                         dct_offset >>=1;
2082                         add_dequant_dct(s, block[4], 4, dest_cb,              dct_linesize, s->chroma_qscale);
2083                         add_dequant_dct(s, block[5], 5, dest_cr,              dct_linesize, s->chroma_qscale);
2084                         add_dequant_dct(s, block[6], 6, dest_cb + dct_offset, dct_linesize, s->chroma_qscale);
2085                         add_dequant_dct(s, block[7], 7, dest_cr + dct_offset, dct_linesize, s->chroma_qscale);
2086                     }
2087                 }
2088             } else if(is_mpeg12 || (s->codec_id != AV_CODEC_ID_WMV2)){
2089                 add_dct(s, block[0], 0, dest_y                          , dct_linesize);
2090                 add_dct(s, block[1], 1, dest_y              + block_size, dct_linesize);
2091                 add_dct(s, block[2], 2, dest_y + dct_offset             , dct_linesize);
2092                 add_dct(s, block[3], 3, dest_y + dct_offset + block_size, dct_linesize);
2093
2094                 if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY)) {
2095                     if(s->chroma_y_shift){//Chroma420
2096                         add_dct(s, block[4], 4, dest_cb, uvlinesize);
2097                         add_dct(s, block[5], 5, dest_cr, uvlinesize);
2098                     }else{
2099                         //chroma422
2100                         dct_linesize = uvlinesize << s->interlaced_dct;
2101                         dct_offset   = s->interlaced_dct ? uvlinesize : uvlinesize*block_size;
2102
2103                         add_dct(s, block[4], 4, dest_cb, dct_linesize);
2104                         add_dct(s, block[5], 5, dest_cr, dct_linesize);
2105                         add_dct(s, block[6], 6, dest_cb+dct_offset, dct_linesize);
2106                         add_dct(s, block[7], 7, dest_cr+dct_offset, dct_linesize);
2107                         if(!s->chroma_x_shift){//Chroma444
2108                             add_dct(s, block[8], 8, dest_cb+block_size, dct_linesize);
2109                             add_dct(s, block[9], 9, dest_cr+block_size, dct_linesize);
2110                             add_dct(s, block[10], 10, dest_cb+block_size+dct_offset, dct_linesize);
2111                             add_dct(s, block[11], 11, dest_cr+block_size+dct_offset, dct_linesize);
2112                         }
2113                     }
2114                 }//fi gray
2115             }
2116             else if (CONFIG_WMV2_DECODER || CONFIG_WMV2_ENCODER) {
2117                 ff_wmv2_add_mb(s, block, dest_y, dest_cb, dest_cr);
2118             }
2119         } else {
2120             /* Only MPEG-4 Simple Studio Profile is supported in > 8-bit mode.
2121                TODO: Integrate 10-bit properly into mpegvideo.c so that ER works properly */
2122             if (s->avctx->bits_per_raw_sample > 8){
2123                 const int act_block_size = block_size * 2;
2124
2125                 if(s->dpcm_direction == 0) {
2126                     s->idsp.idct_put(dest_y,                           dct_linesize, (int16_t*)(*s->block32)[0]);
2127                     s->idsp.idct_put(dest_y              + act_block_size, dct_linesize, (int16_t*)(*s->block32)[1]);
2128                     s->idsp.idct_put(dest_y + dct_offset,              dct_linesize, (int16_t*)(*s->block32)[2]);
2129                     s->idsp.idct_put(dest_y + dct_offset + act_block_size, dct_linesize, (int16_t*)(*s->block32)[3]);
2130
2131                     dct_linesize = uvlinesize << s->interlaced_dct;
2132                     dct_offset   = s->interlaced_dct ? uvlinesize : uvlinesize*block_size;
2133
2134                     s->idsp.idct_put(dest_cb,              dct_linesize, (int16_t*)(*s->block32)[4]);
2135                     s->idsp.idct_put(dest_cr,              dct_linesize, (int16_t*)(*s->block32)[5]);
2136                     s->idsp.idct_put(dest_cb + dct_offset, dct_linesize, (int16_t*)(*s->block32)[6]);
2137                     s->idsp.idct_put(dest_cr + dct_offset, dct_linesize, (int16_t*)(*s->block32)[7]);
2138                     if(!s->chroma_x_shift){//Chroma444
2139                         s->idsp.idct_put(dest_cb + act_block_size,              dct_linesize, (int16_t*)(*s->block32)[8]);
2140                         s->idsp.idct_put(dest_cr + act_block_size,              dct_linesize, (int16_t*)(*s->block32)[9]);
2141                         s->idsp.idct_put(dest_cb + act_block_size + dct_offset, dct_linesize, (int16_t*)(*s->block32)[10]);
2142                         s->idsp.idct_put(dest_cr + act_block_size + dct_offset, dct_linesize, (int16_t*)(*s->block32)[11]);
2143                     }
2144                 } else if(s->dpcm_direction == 1) {
2145                     int i, w, h;
2146                     uint16_t *dest_pcm[3] = {(uint16_t*)dest_y, (uint16_t*)dest_cb, (uint16_t*)dest_cr};
2147                     int linesize[3] = {dct_linesize, uvlinesize, uvlinesize};
2148                     for(i = 0; i < 3; i++) {
2149                         int idx = 0;
2150                         int vsub = i ? s->chroma_y_shift : 0;
2151                         int hsub = i ? s->chroma_x_shift : 0;
2152                         for(h = 0; h < (16 >> vsub); h++){
2153                             for(w = 0; w < (16 >> hsub); w++)
2154                                 dest_pcm[i][w] = (*s->dpcm_macroblock)[i][idx++];
2155                             dest_pcm[i] += linesize[i] / 2;
2156                         }
2157                     }
2158                 } else if(s->dpcm_direction == -1) {
2159                     int i, w, h;
2160                     uint16_t *dest_pcm[3] = {(uint16_t*)dest_y, (uint16_t*)dest_cb, (uint16_t*)dest_cr};
2161                     int linesize[3] = {dct_linesize, uvlinesize, uvlinesize};
2162                     for(i = 0; i < 3; i++) {
2163                         int idx = 0;
2164                         int vsub = i ? s->chroma_y_shift : 0;
2165                         int hsub = i ? s->chroma_x_shift : 0;
2166                         dest_pcm[i] += (linesize[i] / 2) * ((16 >> vsub) - 1);
2167                         for(h = (16 >> vsub)-1; h >= 1; h--){
2168                             for(w = (16 >> hsub)-1; w >= 1; w--)
2169                                 dest_pcm[i][w] = (*s->dpcm_macroblock)[i][idx++];
2170                             dest_pcm[i] -= linesize[i] / 2;
2171                         }
2172                     }
2173                 }
2174             }
2175             /* dct only in intra block */
2176             else if(s->encoding || !(s->codec_id==AV_CODEC_ID_MPEG1VIDEO || s->codec_id==AV_CODEC_ID_MPEG2VIDEO)){
2177                 put_dct(s, block[0], 0, dest_y                          , dct_linesize, s->qscale);
2178                 put_dct(s, block[1], 1, dest_y              + block_size, dct_linesize, s->qscale);
2179                 put_dct(s, block[2], 2, dest_y + dct_offset             , dct_linesize, s->qscale);
2180                 put_dct(s, block[3], 3, dest_y + dct_offset + block_size, dct_linesize, s->qscale);
2181
2182                 if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY)) {
2183                     if(s->chroma_y_shift){
2184                         put_dct(s, block[4], 4, dest_cb, uvlinesize, s->chroma_qscale);
2185                         put_dct(s, block[5], 5, dest_cr, uvlinesize, s->chroma_qscale);
2186                     }else{
2187                         dct_offset >>=1;
2188                         dct_linesize >>=1;
2189                         put_dct(s, block[4], 4, dest_cb,              dct_linesize, s->chroma_qscale);
2190                         put_dct(s, block[5], 5, dest_cr,              dct_linesize, s->chroma_qscale);
2191                         put_dct(s, block[6], 6, dest_cb + dct_offset, dct_linesize, s->chroma_qscale);
2192                         put_dct(s, block[7], 7, dest_cr + dct_offset, dct_linesize, s->chroma_qscale);
2193                     }
2194                 }
2195             }else{
2196                 s->idsp.idct_put(dest_y,                           dct_linesize, block[0]);
2197                 s->idsp.idct_put(dest_y              + block_size, dct_linesize, block[1]);
2198                 s->idsp.idct_put(dest_y + dct_offset,              dct_linesize, block[2]);
2199                 s->idsp.idct_put(dest_y + dct_offset + block_size, dct_linesize, block[3]);
2200
2201                 if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY)) {
2202                     if(s->chroma_y_shift){
2203                         s->idsp.idct_put(dest_cb, uvlinesize, block[4]);
2204                         s->idsp.idct_put(dest_cr, uvlinesize, block[5]);
2205                     }else{
2206
2207                         dct_linesize = uvlinesize << s->interlaced_dct;
2208                         dct_offset   = s->interlaced_dct ? uvlinesize : uvlinesize*block_size;
2209
2210                         s->idsp.idct_put(dest_cb,              dct_linesize, block[4]);
2211                         s->idsp.idct_put(dest_cr,              dct_linesize, block[5]);
2212                         s->idsp.idct_put(dest_cb + dct_offset, dct_linesize, block[6]);
2213                         s->idsp.idct_put(dest_cr + dct_offset, dct_linesize, block[7]);
2214                         if(!s->chroma_x_shift){//Chroma444
2215                             s->idsp.idct_put(dest_cb + block_size,              dct_linesize, block[8]);
2216                             s->idsp.idct_put(dest_cr + block_size,              dct_linesize, block[9]);
2217                             s->idsp.idct_put(dest_cb + block_size + dct_offset, dct_linesize, block[10]);
2218                             s->idsp.idct_put(dest_cr + block_size + dct_offset, dct_linesize, block[11]);
2219                         }
2220                     }
2221                 }//gray
2222             }
2223         }
2224 skip_idct:
2225         if(!readable){
2226             s->hdsp.put_pixels_tab[0][0](s->dest[0], dest_y ,   linesize,16);
2227             if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY)) {
2228                 s->hdsp.put_pixels_tab[s->chroma_x_shift][0](s->dest[1], dest_cb, uvlinesize,16 >> s->chroma_y_shift);
2229                 s->hdsp.put_pixels_tab[s->chroma_x_shift][0](s->dest[2], dest_cr, uvlinesize,16 >> s->chroma_y_shift);
2230             }
2231         }
2232     }
2233 }
2234
2235 void ff_mpv_reconstruct_mb(MpegEncContext *s, int16_t block[12][64])
2236 {
2237 #if !CONFIG_SMALL
2238     if(s->out_format == FMT_MPEG1) {
2239         if(s->avctx->lowres) mpv_reconstruct_mb_internal(s, block, 1, 1);
2240         else                 mpv_reconstruct_mb_internal(s, block, 0, 1);
2241     } else
2242 #endif
2243     if(s->avctx->lowres) mpv_reconstruct_mb_internal(s, block, 1, 0);
2244     else                  mpv_reconstruct_mb_internal(s, block, 0, 0);
2245 }
2246
2247 void ff_mpeg_draw_horiz_band(MpegEncContext *s, int y, int h)
2248 {
2249     ff_draw_horiz_band(s->avctx, s->current_picture_ptr->f,
2250                        s->last_picture_ptr ? s->last_picture_ptr->f : NULL, y, h, s->picture_structure,
2251                        s->first_field, s->low_delay);
2252 }
2253
2254 void ff_init_block_index(MpegEncContext *s){ //FIXME maybe rename
2255     const int linesize   = s->current_picture.f->linesize[0]; //not s->linesize as this would be wrong for field pics
2256     const int uvlinesize = s->current_picture.f->linesize[1];
2257     const int width_of_mb = (4 + (s->avctx->bits_per_raw_sample > 8)) - s->avctx->lowres;
2258     const int height_of_mb = 4 - s->avctx->lowres;
2259
2260     s->block_index[0]= s->b8_stride*(s->mb_y*2    ) - 2 + s->mb_x*2;
2261     s->block_index[1]= s->b8_stride*(s->mb_y*2    ) - 1 + s->mb_x*2;
2262     s->block_index[2]= s->b8_stride*(s->mb_y*2 + 1) - 2 + s->mb_x*2;
2263     s->block_index[3]= s->b8_stride*(s->mb_y*2 + 1) - 1 + s->mb_x*2;
2264     s->block_index[4]= s->mb_stride*(s->mb_y + 1)                + s->b8_stride*s->mb_height*2 + s->mb_x - 1;
2265     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;
2266     //block_index is not used by mpeg2, so it is not affected by chroma_format
2267
2268     s->dest[0] = s->current_picture.f->data[0] + (int)((s->mb_x - 1U) <<  width_of_mb);
2269     s->dest[1] = s->current_picture.f->data[1] + (int)((s->mb_x - 1U) << (width_of_mb - s->chroma_x_shift));
2270     s->dest[2] = s->current_picture.f->data[2] + (int)((s->mb_x - 1U) << (width_of_mb - s->chroma_x_shift));
2271
2272     if(!(s->pict_type==AV_PICTURE_TYPE_B && s->avctx->draw_horiz_band && s->picture_structure==PICT_FRAME))
2273     {
2274         if(s->picture_structure==PICT_FRAME){
2275         s->dest[0] += s->mb_y *   linesize << height_of_mb;
2276         s->dest[1] += s->mb_y * uvlinesize << (height_of_mb - s->chroma_y_shift);
2277         s->dest[2] += s->mb_y * uvlinesize << (height_of_mb - s->chroma_y_shift);
2278         }else{
2279             s->dest[0] += (s->mb_y>>1) *   linesize << height_of_mb;
2280             s->dest[1] += (s->mb_y>>1) * uvlinesize << (height_of_mb - s->chroma_y_shift);
2281             s->dest[2] += (s->mb_y>>1) * uvlinesize << (height_of_mb - s->chroma_y_shift);
2282             av_assert1((s->mb_y&1) == (s->picture_structure == PICT_BOTTOM_FIELD));
2283         }
2284     }
2285 }
2286
2287 void ff_mpeg_flush(AVCodecContext *avctx){
2288     int i;
2289     MpegEncContext *s = avctx->priv_data;
2290
2291     if (!s || !s->picture)
2292         return;
2293
2294     for (i = 0; i < MAX_PICTURE_COUNT; i++)
2295         ff_mpeg_unref_picture(s->avctx, &s->picture[i]);
2296     s->current_picture_ptr = s->last_picture_ptr = s->next_picture_ptr = NULL;
2297
2298     ff_mpeg_unref_picture(s->avctx, &s->current_picture);
2299     ff_mpeg_unref_picture(s->avctx, &s->last_picture);
2300     ff_mpeg_unref_picture(s->avctx, &s->next_picture);
2301
2302     s->mb_x= s->mb_y= 0;
2303     s->closed_gop= 0;
2304
2305     s->parse_context.state= -1;
2306     s->parse_context.frame_start_found= 0;
2307     s->parse_context.overread= 0;
2308     s->parse_context.overread_index= 0;
2309     s->parse_context.index= 0;
2310     s->parse_context.last_index= 0;
2311     s->bitstream_buffer_size=0;
2312     s->pp_time=0;
2313 }
2314
2315 /**
2316  * set qscale and update qscale dependent variables.
2317  */
2318 void ff_set_qscale(MpegEncContext * s, int qscale)
2319 {
2320     if (qscale < 1)
2321         qscale = 1;
2322     else if (qscale > 31)
2323         qscale = 31;
2324
2325     s->qscale = qscale;
2326     s->chroma_qscale= s->chroma_qscale_table[qscale];
2327
2328     s->y_dc_scale= s->y_dc_scale_table[ qscale ];
2329     s->c_dc_scale= s->c_dc_scale_table[ s->chroma_qscale ];
2330 }
2331
2332 void ff_mpv_report_decode_progress(MpegEncContext *s)
2333 {
2334     if (s->pict_type != AV_PICTURE_TYPE_B && !s->partitioned_frame && !s->er.error_occurred)
2335         ff_thread_report_progress(&s->current_picture_ptr->tf, s->mb_y, 0);
2336 }