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 "libavutil/video_enc_params.h"
  36
  37 #include "avcodec.h"
  38 #include "blockdsp.h"
  39 #include "h264chroma.h"
  40 #include "idctdsp.h"
  41 #include "internal.h"
  42 #include "mathops.h"
  43 #include "mpeg_er.h"
  44 #include "mpegutils.h"
  45 #include "mpegvideo.h"
  46 #include "mpegvideodata.h"
  47 #include "mjpegenc.h"
  48 #include "msmpeg4.h"
  49 #include "qpeldsp.h"
  50 #include "thread.h"
  51 #include "wmv2.h"
  52 #include <limits.h>
  53
  54 static void dct_unquantize_mpeg1_intra_c(MpegEncContext *s,
  55                                    int16_t *block, int n, int qscale)
  56 {
  57     int i, level, nCoeffs;
  58     const uint16_t *quant_matrix;
  59
  60     nCoeffs= s->block_last_index[n];
  61
  62     block[0] *= n < 4 ? s->y_dc_scale : s->c_dc_scale;
  63     /* XXX: only MPEG-1 */
  64     quant_matrix = s->intra_matrix;
  65     for(i=1;i<=nCoeffs;i++) {
  66         int j= s->intra_scantable.permutated[i];
  67         level = block[j];
  68         if (level) {
  69             if (level < 0) {
  70                 level = -level;
  71                 level = (int)(level * qscale * quant_matrix[j]) >> 3;
  72                 level = (level - 1) | 1;
  73                 level = -level;
  74             } else {
  75                 level = (int)(level * qscale * quant_matrix[j]) >> 3;
  76                 level = (level - 1) | 1;
  77             }
  78             block[j] = level;
  79         }
  80     }
  81 }
  82
  83 static void dct_unquantize_mpeg1_inter_c(MpegEncContext *s,
  84                                    int16_t *block, int n, int qscale)
  85 {
  86     int i, level, nCoeffs;
  87     const uint16_t *quant_matrix;
  88
  89     nCoeffs= s->block_last_index[n];
  90
  91     quant_matrix = s->inter_matrix;
  92     for(i=0; i<=nCoeffs; i++) {
  93         int j= s->intra_scantable.permutated[i];
  94         level = block[j];
  95         if (level) {
  96             if (level < 0) {
  97                 level = -level;
  98                 level = (((level << 1) + 1) * qscale *
  99                          ((int) (quant_matrix[j]))) >> 4;
 100                 level = (level - 1) | 1;
 101                 level = -level;
 102             } else {
 103                 level = (((level << 1) + 1) * qscale *
 104                          ((int) (quant_matrix[j]))) >> 4;
 105                 level = (level - 1) | 1;
 106             }
 107             block[j] = level;
 108         }
 109     }
 110 }
 111
 112 static void dct_unquantize_mpeg2_intra_c(MpegEncContext *s,
 113                                    int16_t *block, int n, int qscale)
 114 {
 115     int i, level, nCoeffs;
 116     const uint16_t *quant_matrix;
 117
 118     if (s->q_scale_type) qscale = ff_mpeg2_non_linear_qscale[qscale];
 119     else                 qscale <<= 1;
 120
 121     if(s->alternate_scan) nCoeffs= 63;
 122     else nCoeffs= s->block_last_index[n];
 123
 124     block[0] *= n < 4 ? s->y_dc_scale : s->c_dc_scale;
 125     quant_matrix = s->intra_matrix;
 126     for(i=1;i<=nCoeffs;i++) {
 127         int j= s->intra_scantable.permutated[i];
 128         level = block[j];
 129         if (level) {
 130             if (level < 0) {
 131                 level = -level;
 132                 level = (int)(level * qscale * quant_matrix[j]) >> 4;
 133                 level = -level;
 134             } else {
 135                 level = (int)(level * qscale * quant_matrix[j]) >> 4;
 136             }
 137             block[j] = level;
 138         }
 139     }
 140 }
 141
 142 static void dct_unquantize_mpeg2_intra_bitexact(MpegEncContext *s,
 143                                    int16_t *block, int n, int qscale)
 144 {
 145     int i, level, nCoeffs;
 146     const uint16_t *quant_matrix;
 147     int sum=-1;
 148
 149     if (s->q_scale_type) qscale = ff_mpeg2_non_linear_qscale[qscale];
 150     else                 qscale <<= 1;
 151
 152     if(s->alternate_scan) nCoeffs= 63;
 153     else nCoeffs= s->block_last_index[n];
 154
 155     block[0] *= n < 4 ? s->y_dc_scale : s->c_dc_scale;
 156     sum += block[0];
 157     quant_matrix = s->intra_matrix;
 158     for(i=1;i<=nCoeffs;i++) {
 159         int j= s->intra_scantable.permutated[i];
 160         level = block[j];
 161         if (level) {
 162             if (level < 0) {
 163                 level = -level;
 164                 level = (int)(level * qscale * quant_matrix[j]) >> 4;
 165                 level = -level;
 166             } else {
 167                 level = (int)(level * qscale * quant_matrix[j]) >> 4;
 168             }
 169             block[j] = level;
 170             sum+=level;
 171         }
 172     }
 173     block[63]^=sum&1;
 174 }
 175
 176 static void dct_unquantize_mpeg2_inter_c(MpegEncContext *s,
 177                                    int16_t *block, int n, int qscale)
 178 {
 179     int i, level, nCoeffs;
 180     const uint16_t *quant_matrix;
 181     int sum=-1;
 182
 183     if (s->q_scale_type) qscale = ff_mpeg2_non_linear_qscale[qscale];
 184     else                 qscale <<= 1;
 185
 186     if(s->alternate_scan) nCoeffs= 63;
 187     else nCoeffs= s->block_last_index[n];
 188
 189     quant_matrix = s->inter_matrix;
 190     for(i=0; i<=nCoeffs; i++) {
 191         int j= s->intra_scantable.permutated[i];
 192         level = block[j];
 193         if (level) {
 194             if (level < 0) {
 195                 level = -level;
 196                 level = (((level << 1) + 1) * qscale *
 197                          ((int) (quant_matrix[j]))) >> 5;
 198                 level = -level;
 199             } else {
 200                 level = (((level << 1) + 1) * qscale *
 201                          ((int) (quant_matrix[j]))) >> 5;
 202             }
 203             block[j] = level;
 204             sum+=level;
 205         }
 206     }
 207     block[63]^=sum&1;
 208 }
 209
 210 static void dct_unquantize_h263_intra_c(MpegEncContext *s,
 211                                   int16_t *block, int n, int qscale)
 212 {
 213     int i, level, qmul, qadd;
 214     int nCoeffs;
 215
 216     av_assert2(s->block_last_index[n]>=0 || s->h263_aic);
 217
 218     qmul = qscale << 1;
 219
 220     if (!s->h263_aic) {
 221         block[0] *= n < 4 ? s->y_dc_scale : s->c_dc_scale;
 222         qadd = (qscale - 1) | 1;
 223     }else{
 224         qadd = 0;
 225     }
 226     if(s->ac_pred)
 227         nCoeffs=63;
 228     else
 229         nCoeffs= s->intra_scantable.raster_end[ s->block_last_index[n] ];
 230
 231     for(i=1; i<=nCoeffs; i++) {
 232         level = block[i];
 233         if (level) {
 234             if (level < 0) {
 235                 level = level * qmul - qadd;
 236             } else {
 237                 level = level * qmul + qadd;
 238             }
 239             block[i] = level;
 240         }
 241     }
 242 }
 243
 244 static void dct_unquantize_h263_inter_c(MpegEncContext *s,
 245                                   int16_t *block, int n, int qscale)
 246 {
 247     int i, level, qmul, qadd;
 248     int nCoeffs;
 249
 250     av_assert2(s->block_last_index[n]>=0);
 251
 252     qadd = (qscale - 1) | 1;
 253     qmul = qscale << 1;
 254
 255     nCoeffs= s->inter_scantable.raster_end[ s->block_last_index[n] ];
 256
 257     for(i=0; i<=nCoeffs; i++) {
 258         level = block[i];
 259         if (level) {
 260             if (level < 0) {
 261                 level = level * qmul - qadd;
 262             } else {
 263                 level = level * qmul + qadd;
 264             }
 265             block[i] = level;
 266         }
 267     }
 268 }
 269
 270
 271 static void gray16(uint8_t *dst, const uint8_t *src, ptrdiff_t linesize, int h)
 272 {
 273     while(h--)
 274         memset(dst + h*linesize, 128, 16);
 275 }
 276
 277 static void gray8(uint8_t *dst, const uint8_t *src, ptrdiff_t linesize, int h)
 278 {
 279     while(h--)
 280         memset(dst + h*linesize, 128, 8);
 281 }
 282
 283 /* init common dct for both encoder and decoder */
 284 static av_cold int dct_init(MpegEncContext *s)
 285 {
 286     ff_blockdsp_init(&s->bdsp, s->avctx);
 287     ff_h264chroma_init(&s->h264chroma, 8); //for lowres
 288     ff_hpeldsp_init(&s->hdsp, s->avctx->flags);
 289     ff_mpegvideodsp_init(&s->mdsp);
 290     ff_videodsp_init(&s->vdsp, s->avctx->bits_per_raw_sample);
 291
 292     if (s->avctx->debug & FF_DEBUG_NOMC) {
 293         int i;
 294         for (i=0; i<4; i++) {
 295             s->hdsp.avg_pixels_tab[0][i] = gray16;
 296             s->hdsp.put_pixels_tab[0][i] = gray16;
 297             s->hdsp.put_no_rnd_pixels_tab[0][i] = gray16;
 298
 299             s->hdsp.avg_pixels_tab[1][i] = gray8;
 300             s->hdsp.put_pixels_tab[1][i] = gray8;
 301             s->hdsp.put_no_rnd_pixels_tab[1][i] = gray8;
 302         }
 303     }
 304
 305     s->dct_unquantize_h263_intra = dct_unquantize_h263_intra_c;
 306     s->dct_unquantize_h263_inter = dct_unquantize_h263_inter_c;
 307     s->dct_unquantize_mpeg1_intra = dct_unquantize_mpeg1_intra_c;
 308     s->dct_unquantize_mpeg1_inter = dct_unquantize_mpeg1_inter_c;
 309     s->dct_unquantize_mpeg2_intra = dct_unquantize_mpeg2_intra_c;
 310     if (s->avctx->flags & AV_CODEC_FLAG_BITEXACT)
 311         s->dct_unquantize_mpeg2_intra = dct_unquantize_mpeg2_intra_bitexact;
 312     s->dct_unquantize_mpeg2_inter = dct_unquantize_mpeg2_inter_c;
 313
 314     if (HAVE_INTRINSICS_NEON)
 315         ff_mpv_common_init_neon(s);
 316
 317     if (ARCH_ALPHA)
 318         ff_mpv_common_init_axp(s);
 319     if (ARCH_ARM)
 320         ff_mpv_common_init_arm(s);
 321     if (ARCH_PPC)
 322         ff_mpv_common_init_ppc(s);
 323     if (ARCH_X86)
 324         ff_mpv_common_init_x86(s);
 325     if (ARCH_MIPS)
 326         ff_mpv_common_init_mips(s);
 327
 328     return 0;
 329 }
 330
 331 av_cold void ff_mpv_idct_init(MpegEncContext *s)
 332 {
 333     if (s->codec_id == AV_CODEC_ID_MPEG4)
 334         s->idsp.mpeg4_studio_profile = s->studio_profile;
 335     ff_idctdsp_init(&s->idsp, s->avctx);
 336
 337     /* load & permutate scantables
 338      * note: only wmv uses different ones
 339      */
 340     if (s->alternate_scan) {
 341         ff_init_scantable(s->idsp.idct_permutation, &s->inter_scantable, ff_alternate_vertical_scan);
 342         ff_init_scantable(s->idsp.idct_permutation, &s->intra_scantable, ff_alternate_vertical_scan);
 343     } else {
 344         ff_init_scantable(s->idsp.idct_permutation, &s->inter_scantable, ff_zigzag_direct);
 345         ff_init_scantable(s->idsp.idct_permutation, &s->intra_scantable, ff_zigzag_direct);
 346     }
 347     ff_init_scantable(s->idsp.idct_permutation, &s->intra_h_scantable, ff_alternate_horizontal_scan);
 348     ff_init_scantable(s->idsp.idct_permutation, &s->intra_v_scantable, ff_alternate_vertical_scan);
 349 }
 350
 351 static int alloc_picture(MpegEncContext *s, Picture *pic)
 352 {
 353     return ff_alloc_picture(s->avctx, pic, &s->me, &s->sc, 0, 0,
 354                             s->chroma_x_shift, s->chroma_y_shift, s->out_format,
 355                             s->mb_stride, s->mb_width, s->mb_height, s->b8_stride,
 356                             &s->linesize, &s->uvlinesize);
 357 }
 358
 359 static int init_duplicate_context(MpegEncContext *s)
 360 {
 361     int y_size = s->b8_stride * (2 * s->mb_height + 1);
 362     int c_size = s->mb_stride * (s->mb_height + 1);
 363     int yc_size = y_size + 2 * c_size;
 364     int i;
 365
 366     if (s->mb_height & 1)
 367         yc_size += 2*s->b8_stride + 2*s->mb_stride;
 368
 369     if (s->encoding) {
 370         if (!FF_ALLOCZ_TYPED_ARRAY(s->me.map,       ME_MAP_SIZE) ||
 371             !FF_ALLOCZ_TYPED_ARRAY(s->me.score_map, ME_MAP_SIZE))
 372             return AVERROR(ENOMEM);
 373
 374         if (s->noise_reduction) {
 375             if (!FF_ALLOCZ_TYPED_ARRAY(s->dct_error_sum,  2))
 376                 return AVERROR(ENOMEM);
 377         }
 378     }
 379     if (!FF_ALLOCZ_TYPED_ARRAY(s->blocks,  2))
 380         return AVERROR(ENOMEM);
 381     s->block = s->blocks[0];
 382
 383     for (i = 0; i < 12; i++) {
 384         s->pblocks[i] = &s->block[i];
 385     }
 386
 387     if (!(s->block32         = av_mallocz(sizeof(*s->block32))) ||
 388         !(s->dpcm_macroblock = av_mallocz(sizeof(*s->dpcm_macroblock))))
 389         return AVERROR(ENOMEM);
 390     s->dpcm_direction = 0;
 391
 392     if (s->avctx->codec_tag == AV_RL32("VCR2")) {
 393         // exchange uv
 394         FFSWAP(void *, s->pblocks[4], s->pblocks[5]);
 395     }
 396
 397     if (s->out_format == FMT_H263) {
 398         /* ac values */
 399         if (!FF_ALLOCZ_TYPED_ARRAY(s->ac_val_base,  yc_size))
 400             return AVERROR(ENOMEM);
 401         s->ac_val[0] = s->ac_val_base + s->b8_stride + 1;
 402         s->ac_val[1] = s->ac_val_base + y_size + s->mb_stride + 1;
 403         s->ac_val[2] = s->ac_val[1] + c_size;
 404     }
 405
 406     return 0;
 407 }
 408
 409 /**
 410  * Initialize an MpegEncContext's thread contexts. Presumes that
 411  * slice_context_count is already set and that all the fields
 412  * that are freed/reset in free_duplicate_context() are NULL.
 413  */
 414 static int init_duplicate_contexts(MpegEncContext *s)
 415 {
 416     int nb_slices = s->slice_context_count, ret;
 417
 418     /* We initialize the copies before the original so that
 419      * fields allocated in init_duplicate_context are NULL after
 420      * copying. This prevents double-frees upon allocation error. */
 421     for (int i = 1; i < nb_slices; i++) {
 422         s->thread_context[i] = av_memdup(s, sizeof(MpegEncContext));
 423         if (!s->thread_context[i])
 424             return AVERROR(ENOMEM);
 425         if ((ret = init_duplicate_context(s->thread_context[i])) < 0)
 426             return ret;
 427         s->thread_context[i]->start_mb_y =
 428             (s->mb_height * (i    ) + nb_slices / 2) / nb_slices;
 429         s->thread_context[i]->end_mb_y   =
 430             (s->mb_height * (i + 1) + nb_slices / 2) / nb_slices;
 431     }
 432     s->start_mb_y = 0;
 433     s->end_mb_y   = nb_slices > 1 ? (s->mb_height + nb_slices / 2) / nb_slices
 434                                   : s->mb_height;
 435     return init_duplicate_context(s);
 436 }
 437
 438 static void free_duplicate_context(MpegEncContext *s)
 439 {
 440     if (!s)
 441         return;
 442
 443     av_freep(&s->sc.edge_emu_buffer);
 444     av_freep(&s->me.scratchpad);
 445     s->me.temp =
 446     s->sc.rd_scratchpad =
 447     s->sc.b_scratchpad =
 448     s->sc.obmc_scratchpad = NULL;
 449
 450     av_freep(&s->dct_error_sum);
 451     av_freep(&s->me.map);
 452     av_freep(&s->me.score_map);
 453     av_freep(&s->blocks);
 454     av_freep(&s->block32);
 455     av_freep(&s->dpcm_macroblock);
 456     av_freep(&s->ac_val_base);
 457     s->block = NULL;
 458 }
 459
 460 static void backup_duplicate_context(MpegEncContext *bak, MpegEncContext *src)
 461 {
 462 #define COPY(a) bak->a = src->a
 463     COPY(sc.edge_emu_buffer);
 464     COPY(me.scratchpad);
 465     COPY(me.temp);
 466     COPY(sc.rd_scratchpad);
 467     COPY(sc.b_scratchpad);
 468     COPY(sc.obmc_scratchpad);
 469     COPY(me.map);
 470     COPY(me.score_map);
 471     COPY(blocks);
 472     COPY(block);
 473     COPY(block32);
 474     COPY(dpcm_macroblock);
 475     COPY(dpcm_direction);
 476     COPY(start_mb_y);
 477     COPY(end_mb_y);
 478     COPY(me.map_generation);
 479     COPY(pb);
 480     COPY(dct_error_sum);
 481     COPY(dct_count[0]);
 482     COPY(dct_count[1]);
 483     COPY(ac_val_base);
 484     COPY(ac_val[0]);
 485     COPY(ac_val[1]);
 486     COPY(ac_val[2]);
 487 #undef COPY
 488 }
 489
 490 int ff_update_duplicate_context(MpegEncContext *dst, MpegEncContext *src)
 491 {
 492     MpegEncContext bak;
 493     int i, ret;
 494     // FIXME copy only needed parts
 495     backup_duplicate_context(&bak, dst);
 496     memcpy(dst, src, sizeof(MpegEncContext));
 497     backup_duplicate_context(dst, &bak);
 498     for (i = 0; i < 12; i++) {
 499         dst->pblocks[i] = &dst->block[i];
 500     }
 501     if (dst->avctx->codec_tag == AV_RL32("VCR2")) {
 502         // exchange uv
 503         FFSWAP(void *, dst->pblocks[4], dst->pblocks[5]);
 504     }
 505     if (!dst->sc.edge_emu_buffer &&
 506         (ret = ff_mpeg_framesize_alloc(dst->avctx, &dst->me,
 507                                        &dst->sc, dst->linesize)) < 0) {
 508         av_log(dst->avctx, AV_LOG_ERROR, "failed to allocate context "
 509                "scratch buffers.\n");
 510         return ret;
 511     }
 512     return 0;
 513 }
 514
 515 int ff_mpeg_update_thread_context(AVCodecContext *dst,
 516                                   const AVCodecContext *src)
 517 {
 518     int i, ret;
 519     MpegEncContext *s = dst->priv_data, *s1 = src->priv_data;
 520
 521     if (dst == src)
 522         return 0;
 523
 524     av_assert0(s != s1);
 525
 526     // FIXME can parameters change on I-frames?
 527     // in that case dst may need a reinit
 528     if (!s->context_initialized) {
 529         int err;
 530         memcpy(s, s1, sizeof(MpegEncContext));
 531
 532         s->avctx                 = dst;
 533         s->bitstream_buffer      = NULL;
 534         s->bitstream_buffer_size = s->allocated_bitstream_buffer_size = 0;
 535
 536         if (s1->context_initialized){
 537 //             s->picture_range_start  += MAX_PICTURE_COUNT;
 538 //             s->picture_range_end    += MAX_PICTURE_COUNT;
 539             ff_mpv_idct_init(s);
 540             if((err = ff_mpv_common_init(s)) < 0){
 541                 memset(s, 0, sizeof(MpegEncContext));
 542                 s->avctx = dst;
 543                 return err;
 544             }
 545         }
 546     }
 547
 548     if (s->height != s1->height || s->width != s1->width || s->context_reinit) {
 549         s->context_reinit = 0;
 550         s->height = s1->height;
 551         s->width  = s1->width;
 552         if ((ret = ff_mpv_common_frame_size_change(s)) < 0)
 553             return ret;
 554     }
 555
 556     s->avctx->coded_height  = s1->avctx->coded_height;
 557     s->avctx->coded_width   = s1->avctx->coded_width;
 558     s->avctx->width         = s1->avctx->width;
 559     s->avctx->height        = s1->avctx->height;
 560
 561     s->quarter_sample       = s1->quarter_sample;
 562
 563     s->coded_picture_number = s1->coded_picture_number;
 564     s->picture_number       = s1->picture_number;
 565
 566     av_assert0(!s->picture || s->picture != s1->picture);
 567     if(s->picture)
 568     for (i = 0; i < MAX_PICTURE_COUNT; i++) {
 569         ff_mpeg_unref_picture(s->avctx, &s->picture[i]);
 570         if (s1->picture && s1->picture[i].f->buf[0] &&
 571             (ret = ff_mpeg_ref_picture(s->avctx, &s->picture[i], &s1->picture[i])) < 0)
 572             return ret;
 573     }
 574
 575 #define UPDATE_PICTURE(pic)\
 576 do {\
 577     ff_mpeg_unref_picture(s->avctx, &s->pic);\
 578     if (s1->pic.f && s1->pic.f->buf[0])\
 579         ret = ff_mpeg_ref_picture(s->avctx, &s->pic, &s1->pic);\
 580     else\
 581         ret = ff_update_picture_tables(&s->pic, &s1->pic);\
 582     if (ret < 0)\
 583         return ret;\
 584 } while (0)
 585
 586     UPDATE_PICTURE(current_picture);
 587     UPDATE_PICTURE(last_picture);
 588     UPDATE_PICTURE(next_picture);
 589
 590 #define REBASE_PICTURE(pic, new_ctx, old_ctx)                                 \
 591     ((pic && pic >= old_ctx->picture &&                                       \
 592       pic < old_ctx->picture + MAX_PICTURE_COUNT) ?                           \
 593         &new_ctx->picture[pic - old_ctx->picture] : NULL)
 594
 595     s->last_picture_ptr    = REBASE_PICTURE(s1->last_picture_ptr,    s, s1);
 596     s->current_picture_ptr = REBASE_PICTURE(s1->current_picture_ptr, s, s1);
 597     s->next_picture_ptr    = REBASE_PICTURE(s1->next_picture_ptr,    s, s1);
 598
 599     // Error/bug resilience
 600     s->next_p_frame_damaged = s1->next_p_frame_damaged;
 601     s->workaround_bugs      = s1->workaround_bugs;
 602     s->padding_bug_score    = s1->padding_bug_score;
 603
 604     // MPEG-4 timing info
 605     memcpy(&s->last_time_base, &s1->last_time_base,
 606            (char *) &s1->pb_field_time + sizeof(s1->pb_field_time) -
 607            (char *) &s1->last_time_base);
 608
 609     // B-frame info
 610     s->max_b_frames = s1->max_b_frames;
 611     s->low_delay    = s1->low_delay;
 612     s->droppable    = s1->droppable;
 613
 614     // DivX handling (doesn't work)
 615     s->divx_packed  = s1->divx_packed;
 616
 617     if (s1->bitstream_buffer) {
 618         if (s1->bitstream_buffer_size +
 619             AV_INPUT_BUFFER_PADDING_SIZE > s->allocated_bitstream_buffer_size) {
 620             av_fast_malloc(&s->bitstream_buffer,
 621                            &s->allocated_bitstream_buffer_size,
 622                            s1->allocated_bitstream_buffer_size);
 623             if (!s->bitstream_buffer) {
 624                 s->bitstream_buffer_size = 0;
 625                 return AVERROR(ENOMEM);
 626             }
 627         }
 628         s->bitstream_buffer_size = s1->bitstream_buffer_size;
 629         memcpy(s->bitstream_buffer, s1->bitstream_buffer,
 630                s1->bitstream_buffer_size);
 631         memset(s->bitstream_buffer + s->bitstream_buffer_size, 0,
 632                AV_INPUT_BUFFER_PADDING_SIZE);
 633     }
 634
 635     // linesize-dependent scratch buffer allocation
 636     if (!s->sc.edge_emu_buffer)
 637         if (s1->linesize) {
 638             if (ff_mpeg_framesize_alloc(s->avctx, &s->me,
 639                                         &s->sc, s1->linesize) < 0) {
 640                 av_log(s->avctx, AV_LOG_ERROR, "Failed to allocate context "
 641                        "scratch buffers.\n");
 642                 return AVERROR(ENOMEM);
 643             }
 644         } else {
 645             av_log(s->avctx, AV_LOG_ERROR, "Context scratch buffers could not "
 646                    "be allocated due to unknown size.\n");
 647         }
 648
 649     // MPEG-2/interlacing info
 650     memcpy(&s->progressive_sequence, &s1->progressive_sequence,
 651            (char *) &s1->rtp_mode - (char *) &s1->progressive_sequence);
 652
 653     if (!s1->first_field) {
 654         s->last_pict_type = s1->pict_type;
 655         if (s1->current_picture_ptr)
 656             s->last_lambda_for[s1->pict_type] = s1->current_picture_ptr->f->quality;
 657     }
 658
 659     return 0;
 660 }
 661
 662 /**
 663  * Set the given MpegEncContext to common defaults
 664  * (same for encoding and decoding).
 665  * The changed fields will not depend upon the
 666  * prior state of the MpegEncContext.
 667  */
 668 void ff_mpv_common_defaults(MpegEncContext *s)
 669 {
 670     s->y_dc_scale_table      =
 671     s->c_dc_scale_table      = ff_mpeg1_dc_scale_table;
 672     s->chroma_qscale_table   = ff_default_chroma_qscale_table;
 673     s->progressive_frame     = 1;
 674     s->progressive_sequence  = 1;
 675     s->picture_structure     = PICT_FRAME;
 676
 677     s->coded_picture_number  = 0;
 678     s->picture_number        = 0;
 679
 680     s->f_code                = 1;
 681     s->b_code                = 1;
 682
 683     s->slice_context_count   = 1;
 684 }
 685
 686 /**
 687  * Initialize the given MpegEncContext for decoding.
 688  * the changed fields will not depend upon
 689  * the prior state of the MpegEncContext.
 690  */
 691 void ff_mpv_decode_init(MpegEncContext *s, AVCodecContext *avctx)
 692 {
 693     ff_mpv_common_defaults(s);
 694
 695     s->avctx           = avctx;
 696     s->width           = avctx->coded_width;
 697     s->height          = avctx->coded_height;
 698     s->codec_id        = avctx->codec->id;
 699     s->workaround_bugs = avctx->workaround_bugs;
 700
 701     /* convert fourcc to upper case */
 702     s->codec_tag          = avpriv_toupper4(avctx->codec_tag);
 703 }
 704
 705 /**
 706  * Initialize and allocates MpegEncContext fields dependent on the resolution.
 707  */
 708 static int init_context_frame(MpegEncContext *s)
 709 {
 710     int y_size, c_size, yc_size, i, mb_array_size, mv_table_size, x, y;
 711
 712     s->mb_width   = (s->width + 15) / 16;
 713     s->mb_stride  = s->mb_width + 1;
 714     s->b8_stride  = s->mb_width * 2 + 1;
 715     mb_array_size = s->mb_height * s->mb_stride;
 716     mv_table_size = (s->mb_height + 2) * s->mb_stride + 1;
 717
 718     /* set default edge pos, will be overridden
 719      * in decode_header if needed */
 720     s->h_edge_pos = s->mb_width * 16;
 721     s->v_edge_pos = s->mb_height * 16;
 722
 723     s->mb_num     = s->mb_width * s->mb_height;
 724
 725     s->block_wrap[0] =
 726     s->block_wrap[1] =
 727     s->block_wrap[2] =
 728     s->block_wrap[3] = s->b8_stride;
 729     s->block_wrap[4] =
 730     s->block_wrap[5] = s->mb_stride;
 731
 732     y_size  = s->b8_stride * (2 * s->mb_height + 1);
 733     c_size  = s->mb_stride * (s->mb_height + 1);
 734     yc_size = y_size + 2   * c_size;
 735
 736     if (s->mb_height & 1)
 737         yc_size += 2*s->b8_stride + 2*s->mb_stride;
 738
 739     if (!FF_ALLOCZ_TYPED_ARRAY(s->mb_index2xy, s->mb_num + 1))
 740         return AVERROR(ENOMEM);
 741     for (y = 0; y < s->mb_height; y++)
 742         for (x = 0; x < s->mb_width; x++)
 743             s->mb_index2xy[x + y * s->mb_width] = x + y * s->mb_stride;
 744
 745     s->mb_index2xy[s->mb_height * s->mb_width] = (s->mb_height - 1) * s->mb_stride + s->mb_width; // FIXME really needed?
 746
 747     if (s->encoding) {
 748         /* Allocate MV tables */
 749         if (!FF_ALLOCZ_TYPED_ARRAY(s->p_mv_table_base,            mv_table_size) ||
 750             !FF_ALLOCZ_TYPED_ARRAY(s->b_forw_mv_table_base,       mv_table_size) ||
 751             !FF_ALLOCZ_TYPED_ARRAY(s->b_back_mv_table_base,       mv_table_size) ||
 752             !FF_ALLOCZ_TYPED_ARRAY(s->b_bidir_forw_mv_table_base, mv_table_size) ||
 753             !FF_ALLOCZ_TYPED_ARRAY(s->b_bidir_back_mv_table_base, mv_table_size) ||
 754             !FF_ALLOCZ_TYPED_ARRAY(s->b_direct_mv_table_base,     mv_table_size))
 755             return AVERROR(ENOMEM);
 756         s->p_mv_table            = s->p_mv_table_base + s->mb_stride + 1;
 757         s->b_forw_mv_table       = s->b_forw_mv_table_base + s->mb_stride + 1;
 758         s->b_back_mv_table       = s->b_back_mv_table_base + s->mb_stride + 1;
 759         s->b_bidir_forw_mv_table = s->b_bidir_forw_mv_table_base + s->mb_stride + 1;
 760         s->b_bidir_back_mv_table = s->b_bidir_back_mv_table_base + s->mb_stride + 1;
 761         s->b_direct_mv_table     = s->b_direct_mv_table_base + s->mb_stride + 1;
 762
 763         /* Allocate MB type table */
 764         if (!FF_ALLOCZ_TYPED_ARRAY(s->mb_type,      mb_array_size) ||
 765             !FF_ALLOCZ_TYPED_ARRAY(s->lambda_table, mb_array_size) ||
 766             !FF_ALLOC_TYPED_ARRAY (s->cplx_tab,     mb_array_size) ||
 767             !FF_ALLOC_TYPED_ARRAY (s->bits_tab,     mb_array_size))
 768             return AVERROR(ENOMEM);
 769     }
 770
 771     if (s->codec_id == AV_CODEC_ID_MPEG4 ||
 772         (s->avctx->flags & AV_CODEC_FLAG_INTERLACED_ME)) {
 773         /* interlaced direct mode decoding tables */
 774         for (i = 0; i < 2; i++) {
 775             int j, k;
 776             for (j = 0; j < 2; j++) {
 777                 for (k = 0; k < 2; k++) {
 778                     if (!FF_ALLOCZ_TYPED_ARRAY(s->b_field_mv_table_base[i][j][k], mv_table_size))
 779                         return AVERROR(ENOMEM);
 780                     s->b_field_mv_table[i][j][k] = s->b_field_mv_table_base[i][j][k] +
 781                                                    s->mb_stride + 1;
 782                 }
 783                 if (!FF_ALLOCZ_TYPED_ARRAY(s->b_field_select_table [i][j], mv_table_size * 2) ||
 784                     !FF_ALLOCZ_TYPED_ARRAY(s->p_field_mv_table_base[i][j], mv_table_size))
 785                     return AVERROR(ENOMEM);
 786                 s->p_field_mv_table[i][j] = s->p_field_mv_table_base[i][j] + s->mb_stride + 1;
 787             }
 788             if (!FF_ALLOCZ_TYPED_ARRAY(s->p_field_select_table[i], mv_table_size * 2))
 789                 return AVERROR(ENOMEM);
 790         }
 791     }
 792     if (s->out_format == FMT_H263) {
 793         /* cbp values, cbp, ac_pred, pred_dir */
 794         if (!FF_ALLOCZ_TYPED_ARRAY(s->coded_block_base, y_size + (s->mb_height&1)*2*s->b8_stride) ||
 795             !FF_ALLOCZ_TYPED_ARRAY(s->cbp_table,        mb_array_size)                            ||
 796             !FF_ALLOCZ_TYPED_ARRAY(s->pred_dir_table,   mb_array_size))
 797             return AVERROR(ENOMEM);
 798         s->coded_block = s->coded_block_base + s->b8_stride + 1;
 799     }
 800
 801     if (s->h263_pred || s->h263_plus || !s->encoding) {
 802         /* dc values */
 803         // MN: we need these for error resilience of intra-frames
 804         if (!FF_ALLOCZ_TYPED_ARRAY(s->dc_val_base, yc_size))
 805             return AVERROR(ENOMEM);
 806         s->dc_val[0] = s->dc_val_base + s->b8_stride + 1;
 807         s->dc_val[1] = s->dc_val_base + y_size + s->mb_stride + 1;
 808         s->dc_val[2] = s->dc_val[1] + c_size;
 809         for (i = 0; i < yc_size; i++)
 810             s->dc_val_base[i] = 1024;
 811     }
 812
 813     /* which mb is an intra block,  init macroblock skip table */
 814     if (!FF_ALLOC_TYPED_ARRAY(s->mbintra_table, mb_array_size) ||
 815         // Note the + 1 is for a quicker MPEG-4 slice_end detection
 816         !FF_ALLOCZ_TYPED_ARRAY(s->mbskip_table,  mb_array_size + 2))
 817         return AVERROR(ENOMEM);
 818     memset(s->mbintra_table, 1, mb_array_size);
 819
 820     return ff_mpeg_er_init(s);
 821 }
 822
 823 static void clear_context(MpegEncContext *s)
 824 {
 825     int i, j, k;
 826
 827     memset(&s->next_picture, 0, sizeof(s->next_picture));
 828     memset(&s->last_picture, 0, sizeof(s->last_picture));
 829     memset(&s->current_picture, 0, sizeof(s->current_picture));
 830     memset(&s->new_picture, 0, sizeof(s->new_picture));
 831
 832     memset(s->thread_context, 0, sizeof(s->thread_context));
 833
 834     s->me.map = NULL;
 835     s->me.score_map = NULL;
 836     s->dct_error_sum = NULL;
 837     s->block = NULL;
 838     s->blocks = NULL;
 839     s->block32 = NULL;
 840     memset(s->pblocks, 0, sizeof(s->pblocks));
 841     s->dpcm_direction = 0;
 842     s->dpcm_macroblock = NULL;
 843     s->ac_val_base = NULL;
 844     s->ac_val[0] =
 845     s->ac_val[1] =
 846     s->ac_val[2] =NULL;
 847     s->sc.edge_emu_buffer = NULL;
 848     s->me.scratchpad = NULL;
 849     s->me.temp =
 850     s->sc.rd_scratchpad =
 851     s->sc.b_scratchpad =
 852     s->sc.obmc_scratchpad = NULL;
 853
 854
 855     s->bitstream_buffer = NULL;
 856     s->allocated_bitstream_buffer_size = 0;
 857     s->picture          = NULL;
 858     s->mb_type          = NULL;
 859     s->p_mv_table_base  = NULL;
 860     s->b_forw_mv_table_base = NULL;
 861     s->b_back_mv_table_base = NULL;
 862     s->b_bidir_forw_mv_table_base = NULL;
 863     s->b_bidir_back_mv_table_base = NULL;
 864     s->b_direct_mv_table_base = NULL;
 865     s->p_mv_table            = NULL;
 866     s->b_forw_mv_table       = NULL;
 867     s->b_back_mv_table       = NULL;
 868     s->b_bidir_forw_mv_table = NULL;
 869     s->b_bidir_back_mv_table = NULL;
 870     s->b_direct_mv_table     = NULL;
 871     for (i = 0; i < 2; i++) {
 872         for (j = 0; j < 2; j++) {
 873             for (k = 0; k < 2; k++) {
 874                 s->b_field_mv_table_base[i][j][k] = NULL;
 875                 s->b_field_mv_table[i][j][k] = NULL;
 876             }
 877             s->b_field_select_table[i][j] = NULL;
 878             s->p_field_mv_table_base[i][j] = NULL;
 879             s->p_field_mv_table[i][j] = NULL;
 880         }
 881         s->p_field_select_table[i] = NULL;
 882     }
 883
 884     s->dc_val_base = NULL;
 885     s->coded_block_base = NULL;
 886     s->mbintra_table = NULL;
 887     s->cbp_table = NULL;
 888     s->pred_dir_table = NULL;
 889
 890     s->mbskip_table = NULL;
 891
 892     s->er.error_status_table = NULL;
 893     s->er.er_temp_buffer = NULL;
 894     s->mb_index2xy = NULL;
 895     s->lambda_table = NULL;
 896
 897     s->cplx_tab = NULL;
 898     s->bits_tab = NULL;
 899 }
 900
 901 /**
 902  * init common structure for both encoder and decoder.
 903  * this assumes that some variables like width/height are already set
 904  */
 905 av_cold int ff_mpv_common_init(MpegEncContext *s)
 906 {
 907     int i, ret;
 908     int nb_slices = (HAVE_THREADS &&
 909                      s->avctx->active_thread_type & FF_THREAD_SLICE) ?
 910                     s->avctx->thread_count : 1;
 911
 912     clear_context(s);
 913
 914     if (s->encoding && s->avctx->slices)
 915         nb_slices = s->avctx->slices;
 916
 917     if (s->codec_id == AV_CODEC_ID_MPEG2VIDEO && !s->progressive_sequence)
 918         s->mb_height = (s->height + 31) / 32 * 2;
 919     else
 920         s->mb_height = (s->height + 15) / 16;
 921
 922     if (s->avctx->pix_fmt == AV_PIX_FMT_NONE) {
 923         av_log(s->avctx, AV_LOG_ERROR,
 924                "decoding to AV_PIX_FMT_NONE is not supported.\n");
 925         return AVERROR(EINVAL);
 926     }
 927
 928     if (nb_slices > MAX_THREADS || (nb_slices > s->mb_height && s->mb_height)) {
 929         int max_slices;
 930         if (s->mb_height)
 931             max_slices = FFMIN(MAX_THREADS, s->mb_height);
 932         else
 933             max_slices = MAX_THREADS;
 934         av_log(s->avctx, AV_LOG_WARNING, "too many threads/slices (%d),"
 935                " reducing to %d\n", nb_slices, max_slices);
 936         nb_slices = max_slices;
 937     }
 938
 939     if ((s->width || s->height) &&
 940         av_image_check_size(s->width, s->height, 0, s->avctx))
 941         return AVERROR(EINVAL);
 942
 943     dct_init(s);
 944
 945     /* set chroma shifts */
 946     ret = av_pix_fmt_get_chroma_sub_sample(s->avctx->pix_fmt,
 947                                            &s->chroma_x_shift,
 948                                            &s->chroma_y_shift);
 949     if (ret)
 950         return ret;
 951
 952     if (!FF_ALLOCZ_TYPED_ARRAY(s->picture, MAX_PICTURE_COUNT))
 953         return AVERROR(ENOMEM);
 954     for (i = 0; i < MAX_PICTURE_COUNT; i++) {
 955         s->picture[i].f = av_frame_alloc();
 956         if (!s->picture[i].f)
 957             goto fail_nomem;
 958     }
 959
 960     if (!(s->next_picture.f    = av_frame_alloc()) ||
 961         !(s->last_picture.f    = av_frame_alloc()) ||
 962         !(s->current_picture.f = av_frame_alloc()) ||
 963         !(s->new_picture.f     = av_frame_alloc()))
 964         goto fail_nomem;
 965
 966     if ((ret = init_context_frame(s)))
 967         goto fail;
 968
 969     s->parse_context.state = -1;
 970
 971     s->context_initialized = 1;
 972     memset(s->thread_context, 0, sizeof(s->thread_context));
 973     s->thread_context[0]   = s;
 974     s->slice_context_count = nb_slices;
 975
 976 //     if (s->width && s->height) {
 977     ret = init_duplicate_contexts(s);
 978     if (ret < 0)
 979         goto fail;
 980 //     }
 981
 982     return 0;
 983  fail_nomem:
 984     ret = AVERROR(ENOMEM);
 985  fail:
 986     ff_mpv_common_end(s);
 987     return ret;
 988 }
 989
 990 /**
 991  * Frees and resets MpegEncContext fields depending on the resolution.
 992  * Is used during resolution changes to avoid a full reinitialization of the
 993  * codec.
 994  */
 995 static void free_context_frame(MpegEncContext *s)
 996 {
 997     int i, j, k;
 998
 999     av_freep(&s->mb_type);
1000     av_freep(&s->p_mv_table_base);
1001     av_freep(&s->b_forw_mv_table_base);
1002     av_freep(&s->b_back_mv_table_base);
1003     av_freep(&s->b_bidir_forw_mv_table_base);
1004     av_freep(&s->b_bidir_back_mv_table_base);
1005     av_freep(&s->b_direct_mv_table_base);
1006     s->p_mv_table            = NULL;
1007     s->b_forw_mv_table       = NULL;
1008     s->b_back_mv_table       = NULL;
1009     s->b_bidir_forw_mv_table = NULL;
1010     s->b_bidir_back_mv_table = NULL;
1011     s->b_direct_mv_table     = NULL;
1012     for (i = 0; i < 2; i++) {
1013         for (j = 0; j < 2; j++) {
1014             for (k = 0; k < 2; k++) {
1015                 av_freep(&s->b_field_mv_table_base[i][j][k]);
1016                 s->b_field_mv_table[i][j][k] = NULL;
1017             }
1018             av_freep(&s->b_field_select_table[i][j]);
1019             av_freep(&s->p_field_mv_table_base[i][j]);
1020             s->p_field_mv_table[i][j] = NULL;
1021         }
1022         av_freep(&s->p_field_select_table[i]);
1023     }
1024
1025     av_freep(&s->dc_val_base);
1026     av_freep(&s->coded_block_base);
1027     av_freep(&s->mbintra_table);
1028     av_freep(&s->cbp_table);
1029     av_freep(&s->pred_dir_table);
1030
1031     av_freep(&s->mbskip_table);
1032
1033     av_freep(&s->er.error_status_table);
1034     av_freep(&s->er.er_temp_buffer);
1035     av_freep(&s->mb_index2xy);
1036     av_freep(&s->lambda_table);
1037
1038     av_freep(&s->cplx_tab);
1039     av_freep(&s->bits_tab);
1040
1041     s->linesize = s->uvlinesize = 0;
1042 }
1043
1044 int ff_mpv_common_frame_size_change(MpegEncContext *s)
1045 {
1046     int i, err = 0;
1047
1048     if (!s->context_initialized)
1049         return AVERROR(EINVAL);
1050
1051     if (s->slice_context_count > 1) {
1052         for (i = 0; i < s->slice_context_count; i++) {
1053             free_duplicate_context(s->thread_context[i]);
1054         }
1055         for (i = 1; i < s->slice_context_count; i++) {
1056             av_freep(&s->thread_context[i]);
1057         }
1058     } else
1059         free_duplicate_context(s);
1060
1061     free_context_frame(s);
1062
1063     if (s->picture)
1064         for (i = 0; i < MAX_PICTURE_COUNT; i++) {
1065                 s->picture[i].needs_realloc = 1;
1066         }
1067
1068     s->last_picture_ptr         =
1069     s->next_picture_ptr         =
1070     s->current_picture_ptr      = NULL;
1071
1072     // init
1073     if (s->codec_id == AV_CODEC_ID_MPEG2VIDEO && !s->progressive_sequence)
1074         s->mb_height = (s->height + 31) / 32 * 2;
1075     else
1076         s->mb_height = (s->height + 15) / 16;
1077
1078     if ((s->width || s->height) &&
1079         (err = av_image_check_size(s->width, s->height, 0, s->avctx)) < 0)
1080         goto fail;
1081
1082     /* set chroma shifts */
1083     err = av_pix_fmt_get_chroma_sub_sample(s->avctx->pix_fmt,
1084                                            &s->chroma_x_shift,
1085                                            &s->chroma_y_shift);
1086     if (err < 0)
1087         goto fail;
1088
1089     if ((err = init_context_frame(s)))
1090         goto fail;
1091
1092     memset(s->thread_context, 0, sizeof(s->thread_context));
1093     s->thread_context[0]   = s;
1094
1095     if (s->width && s->height) {
1096         err = init_duplicate_contexts(s);
1097         if (err < 0)
1098             goto fail;
1099     }
1100
1101     return 0;
1102  fail:
1103     ff_mpv_common_end(s);
1104     return err;
1105 }
1106
1107 /* init common structure for both encoder and decoder */
1108 void ff_mpv_common_end(MpegEncContext *s)
1109 {
1110     int i;
1111
1112     if (!s)
1113         return;
1114
1115     if (s->slice_context_count > 1) {
1116         for (i = 0; i < s->slice_context_count; i++) {
1117             free_duplicate_context(s->thread_context[i]);
1118         }
1119         for (i = 1; i < s->slice_context_count; i++) {
1120             av_freep(&s->thread_context[i]);
1121         }
1122         s->slice_context_count = 1;
1123     } else free_duplicate_context(s);
1124
1125     av_freep(&s->parse_context.buffer);
1126     s->parse_context.buffer_size = 0;
1127
1128     av_freep(&s->bitstream_buffer);
1129     s->allocated_bitstream_buffer_size = 0;
1130
1131     if (!s->avctx)
1132         return;
1133
1134     if (s->picture) {
1135         for (i = 0; i < MAX_PICTURE_COUNT; i++) {
1136             ff_free_picture_tables(&s->picture[i]);
1137             ff_mpeg_unref_picture(s->avctx, &s->picture[i]);
1138             av_frame_free(&s->picture[i].f);
1139         }
1140     }
1141     av_freep(&s->picture);
1142     ff_free_picture_tables(&s->last_picture);
1143     ff_mpeg_unref_picture(s->avctx, &s->last_picture);
1144     av_frame_free(&s->last_picture.f);
1145     ff_free_picture_tables(&s->current_picture);
1146     ff_mpeg_unref_picture(s->avctx, &s->current_picture);
1147     av_frame_free(&s->current_picture.f);
1148     ff_free_picture_tables(&s->next_picture);
1149     ff_mpeg_unref_picture(s->avctx, &s->next_picture);
1150     av_frame_free(&s->next_picture.f);
1151     ff_free_picture_tables(&s->new_picture);
1152     ff_mpeg_unref_picture(s->avctx, &s->new_picture);
1153     av_frame_free(&s->new_picture.f);
1154
1155     free_context_frame(s);
1156
1157     s->context_initialized      = 0;
1158     s->last_picture_ptr         =
1159     s->next_picture_ptr         =
1160     s->current_picture_ptr      = NULL;
1161     s->linesize = s->uvlinesize = 0;
1162 }
1163
1164
1165 static void gray_frame(AVFrame *frame)
1166 {
1167     int i, h_chroma_shift, v_chroma_shift;
1168
1169     av_pix_fmt_get_chroma_sub_sample(frame->format, &h_chroma_shift, &v_chroma_shift);
1170
1171     for(i=0; i<frame->height; i++)
1172         memset(frame->data[0] + frame->linesize[0]*i, 0x80, frame->width);
1173     for(i=0; i<AV_CEIL_RSHIFT(frame->height, v_chroma_shift); i++) {
1174         memset(frame->data[1] + frame->linesize[1]*i,
1175                0x80, AV_CEIL_RSHIFT(frame->width, h_chroma_shift));
1176         memset(frame->data[2] + frame->linesize[2]*i,
1177                0x80, AV_CEIL_RSHIFT(frame->width, h_chroma_shift));
1178     }
1179 }
1180
1181 /**
1182  * generic function called after decoding
1183  * the header and before a frame is decoded.
1184  */
1185 int ff_mpv_frame_start(MpegEncContext *s, AVCodecContext *avctx)
1186 {
1187     int i, ret;
1188     Picture *pic;
1189     s->mb_skipped = 0;
1190
1191     if (!ff_thread_can_start_frame(avctx)) {
1192         av_log(avctx, AV_LOG_ERROR, "Attempt to start a frame outside SETUP state\n");
1193         return -1;
1194     }
1195
1196     /* mark & release old frames */
1197     if (s->pict_type != AV_PICTURE_TYPE_B && s->last_picture_ptr &&
1198         s->last_picture_ptr != s->next_picture_ptr &&
1199         s->last_picture_ptr->f->buf[0]) {
1200         ff_mpeg_unref_picture(s->avctx, s->last_picture_ptr);
1201     }
1202
1203     /* release forgotten pictures */
1204     /* if (MPEG-124 / H.263) */
1205     for (i = 0; i < MAX_PICTURE_COUNT; i++) {
1206         if (&s->picture[i] != s->last_picture_ptr &&
1207             &s->picture[i] != s->next_picture_ptr &&
1208             s->picture[i].reference && !s->picture[i].needs_realloc) {
1209             ff_mpeg_unref_picture(s->avctx, &s->picture[i]);
1210         }
1211     }
1212
1213     ff_mpeg_unref_picture(s->avctx, &s->current_picture);
1214     ff_mpeg_unref_picture(s->avctx, &s->last_picture);
1215     ff_mpeg_unref_picture(s->avctx, &s->next_picture);
1216
1217     /* release non reference frames */
1218     for (i = 0; i < MAX_PICTURE_COUNT; i++) {
1219         if (!s->picture[i].reference)
1220             ff_mpeg_unref_picture(s->avctx, &s->picture[i]);
1221     }
1222
1223     if (s->current_picture_ptr && !s->current_picture_ptr->f->buf[0]) {
1224         // we already have an unused image
1225         // (maybe it was set before reading the header)
1226         pic = s->current_picture_ptr;
1227     } else {
1228         i   = ff_find_unused_picture(s->avctx, s->picture, 0);
1229         if (i < 0) {
1230             av_log(s->avctx, AV_LOG_ERROR, "no frame buffer available\n");
1231             return i;
1232         }
1233         pic = &s->picture[i];
1234     }
1235
1236     pic->reference = 0;
1237     if (!s->droppable) {
1238         if (s->pict_type != AV_PICTURE_TYPE_B)
1239             pic->reference = 3;
1240     }
1241
1242     pic->f->coded_picture_number = s->coded_picture_number++;
1243
1244     if (alloc_picture(s, pic) < 0)
1245         return -1;
1246
1247     s->current_picture_ptr = pic;
1248     // FIXME use only the vars from current_pic
1249     s->current_picture_ptr->f->top_field_first = s->top_field_first;
1250     if (s->codec_id == AV_CODEC_ID_MPEG1VIDEO ||
1251         s->codec_id == AV_CODEC_ID_MPEG2VIDEO) {
1252         if (s->picture_structure != PICT_FRAME)
1253             s->current_picture_ptr->f->top_field_first =
1254                 (s->picture_structure == PICT_TOP_FIELD) == s->first_field;
1255     }
1256     s->current_picture_ptr->f->interlaced_frame = !s->progressive_frame &&
1257                                                  !s->progressive_sequence;
1258     s->current_picture_ptr->field_picture      =  s->picture_structure != PICT_FRAME;
1259
1260     s->current_picture_ptr->f->pict_type = s->pict_type;
1261     // if (s->avctx->flags && AV_CODEC_FLAG_QSCALE)
1262     //     s->current_picture_ptr->quality = s->new_picture_ptr->quality;
1263     s->current_picture_ptr->f->key_frame = s->pict_type == AV_PICTURE_TYPE_I;
1264
1265     if ((ret = ff_mpeg_ref_picture(s->avctx, &s->current_picture,
1266                                    s->current_picture_ptr)) < 0)
1267         return ret;
1268
1269     if (s->pict_type != AV_PICTURE_TYPE_B) {
1270         s->last_picture_ptr = s->next_picture_ptr;
1271         if (!s->droppable)
1272             s->next_picture_ptr = s->current_picture_ptr;
1273     }
1274     ff_dlog(s->avctx, "L%p N%p C%p L%p N%p C%p type:%d drop:%d\n",
1275             s->last_picture_ptr, s->next_picture_ptr,s->current_picture_ptr,
1276             s->last_picture_ptr    ? s->last_picture_ptr->f->data[0]    : NULL,
1277             s->next_picture_ptr    ? s->next_picture_ptr->f->data[0]    : NULL,
1278             s->current_picture_ptr ? s->current_picture_ptr->f->data[0] : NULL,
1279             s->pict_type, s->droppable);
1280
1281     if ((!s->last_picture_ptr || !s->last_picture_ptr->f->buf[0]) &&
1282         (s->pict_type != AV_PICTURE_TYPE_I)) {
1283         int h_chroma_shift, v_chroma_shift;
1284         av_pix_fmt_get_chroma_sub_sample(s->avctx->pix_fmt,
1285                                          &h_chroma_shift, &v_chroma_shift);
1286         if (s->pict_type == AV_PICTURE_TYPE_B && s->next_picture_ptr && s->next_picture_ptr->f->buf[0])
1287             av_log(avctx, AV_LOG_DEBUG,
1288                    "allocating dummy last picture for B frame\n");
1289         else if (s->pict_type != AV_PICTURE_TYPE_I)
1290             av_log(avctx, AV_LOG_ERROR,
1291                    "warning: first frame is no keyframe\n");
1292
1293         /* Allocate a dummy frame */
1294         i = ff_find_unused_picture(s->avctx, s->picture, 0);
1295         if (i < 0) {
1296             av_log(s->avctx, AV_LOG_ERROR, "no frame buffer available\n");
1297             return i;
1298         }
1299         s->last_picture_ptr = &s->picture[i];
1300
1301         s->last_picture_ptr->reference   = 3;
1302         s->last_picture_ptr->f->key_frame = 0;
1303         s->last_picture_ptr->f->pict_type = AV_PICTURE_TYPE_P;
1304
1305         if (alloc_picture(s, s->last_picture_ptr) < 0) {
1306             s->last_picture_ptr = NULL;
1307             return -1;
1308         }
1309
1310         if (!avctx->hwaccel) {
1311             for(i=0; i<avctx->height; i++)
1312                 memset(s->last_picture_ptr->f->data[0] + s->last_picture_ptr->f->linesize[0]*i,
1313                        0x80, avctx->width);
1314             if (s->last_picture_ptr->f->data[2]) {
1315                 for(i=0; i<AV_CEIL_RSHIFT(avctx->height, v_chroma_shift); i++) {
1316                     memset(s->last_picture_ptr->f->data[1] + s->last_picture_ptr->f->linesize[1]*i,
1317                         0x80, AV_CEIL_RSHIFT(avctx->width, h_chroma_shift));
1318                     memset(s->last_picture_ptr->f->data[2] + s->last_picture_ptr->f->linesize[2]*i,
1319                         0x80, AV_CEIL_RSHIFT(avctx->width, h_chroma_shift));
1320                 }
1321             }
1322
1323             if(s->codec_id == AV_CODEC_ID_FLV1 || s->codec_id == AV_CODEC_ID_H263){
1324                 for(i=0; i<avctx->height; i++)
1325                 memset(s->last_picture_ptr->f->data[0] + s->last_picture_ptr->f->linesize[0]*i, 16, avctx->width);
1326             }
1327         }
1328
1329         ff_thread_report_progress(&s->last_picture_ptr->tf, INT_MAX, 0);
1330         ff_thread_report_progress(&s->last_picture_ptr->tf, INT_MAX, 1);
1331     }
1332     if ((!s->next_picture_ptr || !s->next_picture_ptr->f->buf[0]) &&
1333         s->pict_type == AV_PICTURE_TYPE_B) {
1334         /* Allocate a dummy frame */
1335         i = ff_find_unused_picture(s->avctx, s->picture, 0);
1336         if (i < 0) {
1337             av_log(s->avctx, AV_LOG_ERROR, "no frame buffer available\n");
1338             return i;
1339         }
1340         s->next_picture_ptr = &s->picture[i];
1341
1342         s->next_picture_ptr->reference   = 3;
1343         s->next_picture_ptr->f->key_frame = 0;
1344         s->next_picture_ptr->f->pict_type = AV_PICTURE_TYPE_P;
1345
1346         if (alloc_picture(s, s->next_picture_ptr) < 0) {
1347             s->next_picture_ptr = NULL;
1348             return -1;
1349         }
1350         ff_thread_report_progress(&s->next_picture_ptr->tf, INT_MAX, 0);
1351         ff_thread_report_progress(&s->next_picture_ptr->tf, INT_MAX, 1);
1352     }
1353
1354 #if 0 // BUFREF-FIXME
1355     memset(s->last_picture.f->data, 0, sizeof(s->last_picture.f->data));
1356     memset(s->next_picture.f->data, 0, sizeof(s->next_picture.f->data));
1357 #endif
1358     if (s->last_picture_ptr) {
1359         if (s->last_picture_ptr->f->buf[0] &&
1360             (ret = ff_mpeg_ref_picture(s->avctx, &s->last_picture,
1361                                        s->last_picture_ptr)) < 0)
1362             return ret;
1363     }
1364     if (s->next_picture_ptr) {
1365         if (s->next_picture_ptr->f->buf[0] &&
1366             (ret = ff_mpeg_ref_picture(s->avctx, &s->next_picture,
1367                                        s->next_picture_ptr)) < 0)
1368             return ret;
1369     }
1370
1371     av_assert0(s->pict_type == AV_PICTURE_TYPE_I || (s->last_picture_ptr &&
1372                                                  s->last_picture_ptr->f->buf[0]));
1373
1374     if (s->picture_structure!= PICT_FRAME) {
1375         int i;
1376         for (i = 0; i < 4; i++) {
1377             if (s->picture_structure == PICT_BOTTOM_FIELD) {
1378                 s->current_picture.f->data[i] +=
1379                     s->current_picture.f->linesize[i];
1380             }
1381             s->current_picture.f->linesize[i] *= 2;
1382             s->last_picture.f->linesize[i]    *= 2;
1383             s->next_picture.f->linesize[i]    *= 2;
1384         }
1385     }
1386
1387     /* set dequantizer, we can't do it during init as
1388      * it might change for MPEG-4 and we can't do it in the header
1389      * decode as init is not called for MPEG-4 there yet */
1390     if (s->mpeg_quant || s->codec_id == AV_CODEC_ID_MPEG2VIDEO) {
1391         s->dct_unquantize_intra = s->dct_unquantize_mpeg2_intra;
1392         s->dct_unquantize_inter = s->dct_unquantize_mpeg2_inter;
1393     } else if (s->out_format == FMT_H263 || s->out_format == FMT_H261) {
1394         s->dct_unquantize_intra = s->dct_unquantize_h263_intra;
1395         s->dct_unquantize_inter = s->dct_unquantize_h263_inter;
1396     } else {
1397         s->dct_unquantize_intra = s->dct_unquantize_mpeg1_intra;
1398         s->dct_unquantize_inter = s->dct_unquantize_mpeg1_inter;
1399     }
1400
1401     if (s->avctx->debug & FF_DEBUG_NOMC) {
1402         gray_frame(s->current_picture_ptr->f);
1403     }
1404
1405     return 0;
1406 }
1407
1408 /* called after a frame has been decoded. */
1409 void ff_mpv_frame_end(MpegEncContext *s)
1410 {
1411     emms_c();
1412
1413     if (s->current_picture.reference)
1414         ff_thread_report_progress(&s->current_picture_ptr->tf, INT_MAX, 0);
1415 }
1416
1417 void ff_print_debug_info(MpegEncContext *s, Picture *p, AVFrame *pict)
1418 {
1419     ff_print_debug_info2(s->avctx, pict, s->mbskip_table, p->mb_type,
1420                          p->qscale_table, p->motion_val, &s->low_delay,
1421                          s->mb_width, s->mb_height, s->mb_stride, s->quarter_sample);
1422 }
1423
1424 int ff_mpv_export_qp_table(MpegEncContext *s, AVFrame *f, Picture *p, int qp_type)
1425 {
1426     AVVideoEncParams *par;
1427     int mult = (qp_type == FF_QSCALE_TYPE_MPEG1) ? 2 : 1;
1428     unsigned int nb_mb = p->alloc_mb_height * p->alloc_mb_width;
1429     unsigned int x, y;
1430
1431     if (!(s->avctx->export_side_data & AV_CODEC_EXPORT_DATA_VIDEO_ENC_PARAMS))
1432         return 0;
1433
1434     par = av_video_enc_params_create_side_data(f, AV_VIDEO_ENC_PARAMS_MPEG2, nb_mb);
1435     if (!par)
1436         return AVERROR(ENOMEM);
1437
1438     for (y = 0; y < p->alloc_mb_height; y++)
1439         for (x = 0; x < p->alloc_mb_width; x++) {
1440             const unsigned int block_idx = y * p->alloc_mb_width + x;
1441             const unsigned int     mb_xy = y * p->alloc_mb_stride + x;
1442             AVVideoBlockParams *b = av_video_enc_params_block(par, block_idx);
1443
1444             b->src_x = x * 16;
1445             b->src_y = y * 16;
1446             b->w     = 16;
1447             b->h     = 16;
1448
1449             b->delta_qp = p->qscale_table[mb_xy] * mult;
1450         }
1451
1452     return 0;
1453 }
1454
1455 static inline int hpel_motion_lowres(MpegEncContext *s,
1456                                      uint8_t *dest, uint8_t *src,
1457                                      int field_based, int field_select,
1458                                      int src_x, int src_y,
1459                                      int width, int height, ptrdiff_t stride,
1460                                      int h_edge_pos, int v_edge_pos,
1461                                      int w, int h, h264_chroma_mc_func *pix_op,
1462                                      int motion_x, int motion_y)
1463 {
1464     const int lowres   = s->avctx->lowres;
1465     const int op_index = FFMIN(lowres, 3);
1466     const int s_mask   = (2 << lowres) - 1;
1467     int emu = 0;
1468     int sx, sy;
1469
1470     if (s->quarter_sample) {
1471         motion_x /= 2;
1472         motion_y /= 2;
1473     }
1474
1475     sx = motion_x & s_mask;
1476     sy = motion_y & s_mask;
1477     src_x += motion_x >> lowres + 1;
1478     src_y += motion_y >> lowres + 1;
1479
1480     src   += src_y * stride + src_x;
1481
1482     if ((unsigned)src_x > FFMAX( h_edge_pos - (!!sx) - w,                 0) ||
1483         (unsigned)src_y > FFMAX((v_edge_pos >> field_based) - (!!sy) - h, 0)) {
1484         s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, src,
1485                                  s->linesize, s->linesize,
1486                                  w + 1, (h + 1) << field_based,
1487                                  src_x, src_y   << field_based,
1488                                  h_edge_pos, v_edge_pos);
1489         src = s->sc.edge_emu_buffer;
1490         emu = 1;
1491     }
1492
1493     sx = (sx << 2) >> lowres;
1494     sy = (sy << 2) >> lowres;
1495     if (field_select)
1496         src += s->linesize;
1497     pix_op[op_index](dest, src, stride, h, sx, sy);
1498     return emu;
1499 }
1500
1501 /* apply one mpeg motion vector to the three components */
1502 static av_always_inline void mpeg_motion_lowres(MpegEncContext *s,
1503                                                 uint8_t *dest_y,
1504                                                 uint8_t *dest_cb,
1505                                                 uint8_t *dest_cr,
1506                                                 int field_based,
1507                                                 int bottom_field,
1508                                                 int field_select,
1509                                                 uint8_t **ref_picture,
1510                                                 h264_chroma_mc_func *pix_op,
1511                                                 int motion_x, int motion_y,
1512                                                 int h, int mb_y)
1513 {
1514     uint8_t *ptr_y, *ptr_cb, *ptr_cr;
1515     int mx, my, src_x, src_y, uvsrc_x, uvsrc_y, sx, sy, uvsx, uvsy;
1516     ptrdiff_t uvlinesize, linesize;
1517     const int lowres     = s->avctx->lowres;
1518     const int op_index   = FFMIN(lowres-1+s->chroma_x_shift, 3);
1519     const int block_s    = 8>>lowres;
1520     const int s_mask     = (2 << lowres) - 1;
1521     const int h_edge_pos = s->h_edge_pos >> lowres;
1522     const int v_edge_pos = s->v_edge_pos >> lowres;
1523     linesize   = s->current_picture.f->linesize[0] << field_based;
1524     uvlinesize = s->current_picture.f->linesize[1] << field_based;
1525
1526     // FIXME obviously not perfect but qpel will not work in lowres anyway
1527     if (s->quarter_sample) {
1528         motion_x /= 2;
1529         motion_y /= 2;
1530     }
1531
1532     if(field_based){
1533         motion_y += (bottom_field - field_select)*((1 << lowres)-1);
1534     }
1535
1536     sx = motion_x & s_mask;
1537     sy = motion_y & s_mask;
1538     src_x = s->mb_x * 2 * block_s + (motion_x >> lowres + 1);
1539     src_y = (mb_y * 2 * block_s >> field_based) + (motion_y >> lowres + 1);
1540
1541     if (s->out_format == FMT_H263) {
1542         uvsx    = ((motion_x >> 1) & s_mask) | (sx & 1);
1543         uvsy    = ((motion_y >> 1) & s_mask) | (sy & 1);
1544         uvsrc_x = src_x >> 1;
1545         uvsrc_y = src_y >> 1;
1546     } else if (s->out_format == FMT_H261) {
1547         // even chroma mv's are full pel in H261
1548         mx      = motion_x / 4;
1549         my      = motion_y / 4;
1550         uvsx    = (2 * mx) & s_mask;
1551         uvsy    = (2 * my) & s_mask;
1552         uvsrc_x = s->mb_x * block_s + (mx >> lowres);
1553         uvsrc_y =    mb_y * block_s + (my >> lowres);
1554     } else {
1555         if(s->chroma_y_shift){
1556             mx      = motion_x / 2;
1557             my      = motion_y / 2;
1558             uvsx    = mx & s_mask;
1559             uvsy    = my & s_mask;
1560             uvsrc_x = s->mb_x * block_s                 + (mx >> lowres + 1);
1561             uvsrc_y =   (mb_y * block_s >> field_based) + (my >> lowres + 1);
1562         } else {
1563             if(s->chroma_x_shift){
1564             //Chroma422
1565                 mx = motion_x / 2;
1566                 uvsx = mx & s_mask;
1567                 uvsy = motion_y & s_mask;
1568                 uvsrc_y = src_y;
1569                 uvsrc_x = s->mb_x*block_s               + (mx >> (lowres+1));
1570             } else {
1571             //Chroma444
1572                 uvsx = motion_x & s_mask;
1573                 uvsy = motion_y & s_mask;
1574                 uvsrc_x = src_x;
1575                 uvsrc_y = src_y;
1576             }
1577         }
1578     }
1579
1580     ptr_y  = ref_picture[0] + src_y   * linesize   + src_x;
1581     ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x;
1582     ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x;
1583
1584     if ((unsigned) src_x > FFMAX( h_edge_pos - (!!sx) - 2 * block_s,       0) || uvsrc_y<0 ||
1585         (unsigned) src_y > FFMAX((v_edge_pos >> field_based) - (!!sy) - h, 0)) {
1586         s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, ptr_y,
1587                                  linesize >> field_based, linesize >> field_based,
1588                                  17, 17 + field_based,
1589                                 src_x, src_y << field_based, h_edge_pos,
1590                                 v_edge_pos);
1591         ptr_y = s->sc.edge_emu_buffer;
1592         if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY)) {
1593             uint8_t *ubuf = s->sc.edge_emu_buffer + 18 * s->linesize;
1594             uint8_t *vbuf =ubuf + 10 * s->uvlinesize;
1595             if (s->workaround_bugs & FF_BUG_IEDGE)
1596                 vbuf -= s->uvlinesize;
1597             s->vdsp.emulated_edge_mc(ubuf,  ptr_cb,
1598                                      uvlinesize >> field_based, uvlinesize >> field_based,
1599                                      9, 9 + field_based,
1600                                     uvsrc_x, uvsrc_y << field_based,
1601                                     h_edge_pos >> 1, v_edge_pos >> 1);
1602             s->vdsp.emulated_edge_mc(vbuf,  ptr_cr,
1603                                      uvlinesize >> field_based,uvlinesize >> field_based,
1604                                      9, 9 + field_based,
1605                                     uvsrc_x, uvsrc_y << field_based,
1606                                     h_edge_pos >> 1, v_edge_pos >> 1);
1607             ptr_cb = ubuf;
1608             ptr_cr = vbuf;
1609         }
1610     }
1611
1612     // FIXME use this for field pix too instead of the obnoxious hack which changes picture.f->data
1613     if (bottom_field) {
1614         dest_y  += s->linesize;
1615         dest_cb += s->uvlinesize;
1616         dest_cr += s->uvlinesize;
1617     }
1618
1619     if (field_select) {
1620         ptr_y   += s->linesize;
1621         ptr_cb  += s->uvlinesize;
1622         ptr_cr  += s->uvlinesize;
1623     }
1624
1625     sx = (sx << 2) >> lowres;
1626     sy = (sy << 2) >> lowres;
1627     pix_op[lowres - 1](dest_y, ptr_y, linesize, h, sx, sy);
1628
1629     if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY)) {
1630         int hc = s->chroma_y_shift ? (h+1-bottom_field)>>1 : h;
1631         uvsx = (uvsx << 2) >> lowres;
1632         uvsy = (uvsy << 2) >> lowres;
1633         if (hc) {
1634             pix_op[op_index](dest_cb, ptr_cb, uvlinesize, hc, uvsx, uvsy);
1635             pix_op[op_index](dest_cr, ptr_cr, uvlinesize, hc, uvsx, uvsy);
1636         }
1637     }
1638     // FIXME h261 lowres loop filter
1639 }
1640
1641 static inline void chroma_4mv_motion_lowres(MpegEncContext *s,
1642                                             uint8_t *dest_cb, uint8_t *dest_cr,
1643                                             uint8_t **ref_picture,
1644                                             h264_chroma_mc_func * pix_op,
1645                                             int mx, int my)
1646 {
1647     const int lowres     = s->avctx->lowres;
1648     const int op_index   = FFMIN(lowres, 3);
1649     const int block_s    = 8 >> lowres;
1650     const int s_mask     = (2 << lowres) - 1;
1651     const int h_edge_pos = s->h_edge_pos >> lowres + 1;
1652     const int v_edge_pos = s->v_edge_pos >> lowres + 1;
1653     int emu = 0, src_x, src_y, sx, sy;
1654     ptrdiff_t offset;
1655     uint8_t *ptr;
1656
1657     if (s->quarter_sample) {
1658         mx /= 2;
1659         my /= 2;
1660     }
1661
1662     /* In case of 8X8, we construct a single chroma motion vector
1663        with a special rounding */
1664     mx = ff_h263_round_chroma(mx);
1665     my = ff_h263_round_chroma(my);
1666
1667     sx = mx & s_mask;
1668     sy = my & s_mask;
1669     src_x = s->mb_x * block_s + (mx >> lowres + 1);
1670     src_y = s->mb_y * block_s + (my >> lowres + 1);
1671
1672     offset = src_y * s->uvlinesize + src_x;
1673     ptr = ref_picture[1] + offset;
1674     if ((unsigned) src_x > FFMAX(h_edge_pos - (!!sx) - block_s, 0) ||
1675         (unsigned) src_y > FFMAX(v_edge_pos - (!!sy) - block_s, 0)) {
1676         s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, ptr,
1677                                  s->uvlinesize, s->uvlinesize,
1678                                  9, 9,
1679                                  src_x, src_y, h_edge_pos, v_edge_pos);
1680         ptr = s->sc.edge_emu_buffer;
1681         emu = 1;
1682     }
1683     sx = (sx << 2) >> lowres;
1684     sy = (sy << 2) >> lowres;
1685     pix_op[op_index](dest_cb, ptr, s->uvlinesize, block_s, sx, sy);
1686
1687     ptr = ref_picture[2] + offset;
1688     if (emu) {
1689         s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, ptr,
1690                                  s->uvlinesize, s->uvlinesize,
1691                                  9, 9,
1692                                  src_x, src_y, h_edge_pos, v_edge_pos);
1693         ptr = s->sc.edge_emu_buffer;
1694     }
1695     pix_op[op_index](dest_cr, ptr, s->uvlinesize, block_s, sx, sy);
1696 }
1697
1698 /**
1699  * motion compensation of a single macroblock
1700  * @param s context
1701  * @param dest_y luma destination pointer
1702  * @param dest_cb chroma cb/u destination pointer
1703  * @param dest_cr chroma cr/v destination pointer
1704  * @param dir direction (0->forward, 1->backward)
1705  * @param ref_picture array[3] of pointers to the 3 planes of the reference picture
1706  * @param pix_op halfpel motion compensation function (average or put normally)
1707  * the motion vectors are taken from s->mv and the MV type from s->mv_type
1708  */
1709 static inline void MPV_motion_lowres(MpegEncContext *s,
1710                                      uint8_t *dest_y, uint8_t *dest_cb,
1711                                      uint8_t *dest_cr,
1712                                      int dir, uint8_t **ref_picture,
1713                                      h264_chroma_mc_func *pix_op)
1714 {
1715     int mx, my;
1716     int mb_x, mb_y, i;
1717     const int lowres  = s->avctx->lowres;
1718     const int block_s = 8 >>lowres;
1719
1720     mb_x = s->mb_x;
1721     mb_y = s->mb_y;
1722
1723     switch (s->mv_type) {
1724     case MV_TYPE_16X16:
1725         mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
1726                            0, 0, 0,
1727                            ref_picture, pix_op,
1728                            s->mv[dir][0][0], s->mv[dir][0][1],
1729                            2 * block_s, mb_y);
1730         break;
1731     case MV_TYPE_8X8:
1732         mx = 0;
1733         my = 0;
1734         for (i = 0; i < 4; i++) {
1735             hpel_motion_lowres(s, dest_y + ((i & 1) + (i >> 1) *
1736                                s->linesize) * block_s,
1737                                ref_picture[0], 0, 0,
1738                                (2 * mb_x + (i & 1)) * block_s,
1739                                (2 * mb_y + (i >> 1)) * block_s,
1740                                s->width, s->height, s->linesize,
1741                                s->h_edge_pos >> lowres, s->v_edge_pos >> lowres,
1742                                block_s, block_s, pix_op,
1743                                s->mv[dir][i][0], s->mv[dir][i][1]);
1744
1745             mx += s->mv[dir][i][0];
1746             my += s->mv[dir][i][1];
1747         }
1748
1749         if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY))
1750             chroma_4mv_motion_lowres(s, dest_cb, dest_cr, ref_picture,
1751                                      pix_op, mx, my);
1752         break;
1753     case MV_TYPE_FIELD:
1754         if (s->picture_structure == PICT_FRAME) {
1755             /* top field */
1756             mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
1757                                1, 0, s->field_select[dir][0],
1758                                ref_picture, pix_op,
1759                                s->mv[dir][0][0], s->mv[dir][0][1],
1760                                block_s, mb_y);
1761             /* bottom field */
1762             mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
1763                                1, 1, s->field_select[dir][1],
1764                                ref_picture, pix_op,
1765                                s->mv[dir][1][0], s->mv[dir][1][1],
1766                                block_s, mb_y);
1767         } else {
1768             if (s->picture_structure != s->field_select[dir][0] + 1 &&
1769                 s->pict_type != AV_PICTURE_TYPE_B && !s->first_field) {
1770                 ref_picture = s->current_picture_ptr->f->data;
1771
1772             }
1773             mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
1774                                0, 0, s->field_select[dir][0],
1775                                ref_picture, pix_op,
1776                                s->mv[dir][0][0],
1777                                s->mv[dir][0][1], 2 * block_s, mb_y >> 1);
1778             }
1779         break;
1780     case MV_TYPE_16X8:
1781         for (i = 0; i < 2; i++) {
1782             uint8_t **ref2picture;
1783
1784             if (s->picture_structure == s->field_select[dir][i] + 1 ||
1785                 s->pict_type == AV_PICTURE_TYPE_B || s->first_field) {
1786                 ref2picture = ref_picture;
1787             } else {
1788                 ref2picture = s->current_picture_ptr->f->data;
1789             }
1790
1791             mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
1792                                0, 0, s->field_select[dir][i],
1793                                ref2picture, pix_op,
1794                                s->mv[dir][i][0], s->mv[dir][i][1] +
1795                                2 * block_s * i, block_s, mb_y >> 1);
1796
1797             dest_y  +=  2 * block_s *  s->linesize;
1798             dest_cb += (2 * block_s >> s->chroma_y_shift) * s->uvlinesize;
1799             dest_cr += (2 * block_s >> s->chroma_y_shift) * s->uvlinesize;
1800         }
1801         break;
1802     case MV_TYPE_DMV:
1803         if (s->picture_structure == PICT_FRAME) {
1804             for (i = 0; i < 2; i++) {
1805                 int j;
1806                 for (j = 0; j < 2; j++) {
1807                     mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
1808                                        1, j, j ^ i,
1809                                        ref_picture, pix_op,
1810                                        s->mv[dir][2 * i + j][0],
1811                                        s->mv[dir][2 * i + j][1],
1812                                        block_s, mb_y);
1813                 }
1814                 pix_op = s->h264chroma.avg_h264_chroma_pixels_tab;
1815             }
1816         } else {
1817             for (i = 0; i < 2; i++) {
1818                 mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
1819                                    0, 0, s->picture_structure != i + 1,
1820                                    ref_picture, pix_op,
1821                                    s->mv[dir][2 * i][0],s->mv[dir][2 * i][1],
1822                                    2 * block_s, mb_y >> 1);
1823
1824                 // after put we make avg of the same block
1825                 pix_op = s->h264chroma.avg_h264_chroma_pixels_tab;
1826
1827                 // opposite parity is always in the same
1828                 // frame if this is second field
1829                 if (!s->first_field) {
1830                     ref_picture = s->current_picture_ptr->f->data;
1831                 }
1832             }
1833         }
1834         break;
1835     default:
1836         av_assert2(0);
1837     }
1838 }
1839
1840 /**
1841  * find the lowest MB row referenced in the MVs
1842  */
1843 static int lowest_referenced_row(MpegEncContext *s, int dir)
1844 {
1845     int my_max = INT_MIN, my_min = INT_MAX, qpel_shift = !s->quarter_sample;
1846     int my, off, i, mvs;
1847
1848     if (s->picture_structure != PICT_FRAME || s->mcsel)
1849         goto unhandled;
1850
1851     switch (s->mv_type) {
1852         case MV_TYPE_16X16:
1853             mvs = 1;
1854             break;
1855         case MV_TYPE_16X8:
1856             mvs = 2;
1857             break;
1858         case MV_TYPE_8X8:
1859             mvs = 4;
1860             break;
1861         default:
1862             goto unhandled;
1863     }
1864
1865     for (i = 0; i < mvs; i++) {
1866         my = s->mv[dir][i][1];
1867         my_max = FFMAX(my_max, my);
1868         my_min = FFMIN(my_min, my);
1869     }
1870
1871     off = ((FFMAX(-my_min, my_max)<<qpel_shift) + 63) >> 6;
1872
1873     return av_clip(s->mb_y + off, 0, s->mb_height - 1);
1874 unhandled:
1875     return s->mb_height-1;
1876 }
1877
1878 /* put block[] to dest[] */
1879 static inline void put_dct(MpegEncContext *s,
1880                            int16_t *block, int i, uint8_t *dest, int line_size, int qscale)
1881 {
1882     s->dct_unquantize_intra(s, block, i, qscale);
1883     s->idsp.idct_put(dest, line_size, block);
1884 }
1885
1886 /* add block[] to dest[] */
1887 static inline void add_dct(MpegEncContext *s,
1888                            int16_t *block, int i, uint8_t *dest, int line_size)
1889 {
1890     if (s->block_last_index[i] >= 0) {
1891         s->idsp.idct_add(dest, line_size, block);
1892     }
1893 }
1894
1895 static inline void add_dequant_dct(MpegEncContext *s,
1896                            int16_t *block, int i, uint8_t *dest, int line_size, int qscale)
1897 {
1898     if (s->block_last_index[i] >= 0) {
1899         s->dct_unquantize_inter(s, block, i, qscale);
1900
1901         s->idsp.idct_add(dest, line_size, block);
1902     }
1903 }
1904
1905 /**
1906  * Clean dc, ac, coded_block for the current non-intra MB.
1907  */
1908 void ff_clean_intra_table_entries(MpegEncContext *s)
1909 {
1910     int wrap = s->b8_stride;
1911     int xy = s->block_index[0];
1912
1913     s->dc_val[0][xy           ] =
1914     s->dc_val[0][xy + 1       ] =
1915     s->dc_val[0][xy     + wrap] =
1916     s->dc_val[0][xy + 1 + wrap] = 1024;
1917     /* ac pred */
1918     memset(s->ac_val[0][xy       ], 0, 32 * sizeof(int16_t));
1919     memset(s->ac_val[0][xy + wrap], 0, 32 * sizeof(int16_t));
1920     if (s->msmpeg4_version>=3) {
1921         s->coded_block[xy           ] =
1922         s->coded_block[xy + 1       ] =
1923         s->coded_block[xy     + wrap] =
1924         s->coded_block[xy + 1 + wrap] = 0;
1925     }
1926     /* chroma */
1927     wrap = s->mb_stride;
1928     xy = s->mb_x + s->mb_y * wrap;
1929     s->dc_val[1][xy] =
1930     s->dc_val[2][xy] = 1024;
1931     /* ac pred */
1932     memset(s->ac_val[1][xy], 0, 16 * sizeof(int16_t));
1933     memset(s->ac_val[2][xy], 0, 16 * sizeof(int16_t));
1934
1935     s->mbintra_table[xy]= 0;
1936 }
1937
1938 /* generic function called after a macroblock has been parsed by the
1939    decoder or after it has been encoded by the encoder.
1940
1941    Important variables used:
1942    s->mb_intra : true if intra macroblock
1943    s->mv_dir   : motion vector direction
1944    s->mv_type  : motion vector type
1945    s->mv       : motion vector
1946    s->interlaced_dct : true if interlaced dct used (mpeg2)
1947  */
1948 static av_always_inline
1949 void mpv_reconstruct_mb_internal(MpegEncContext *s, int16_t block[12][64],
1950                             int lowres_flag, int is_mpeg12)
1951 {
1952     const int mb_xy = s->mb_y * s->mb_stride + s->mb_x;
1953
1954     if (CONFIG_XVMC &&
1955         s->avctx->hwaccel && s->avctx->hwaccel->decode_mb) {
1956         s->avctx->hwaccel->decode_mb(s);//xvmc uses pblocks
1957         return;
1958     }
1959
1960     if(s->avctx->debug&FF_DEBUG_DCT_COEFF) {
1961        /* print DCT coefficients */
1962        int i,j;
1963        av_log(s->avctx, AV_LOG_DEBUG, "DCT coeffs of MB at %dx%d:\n", s->mb_x, s->mb_y);
1964        for(i=0; i<6; i++){
1965            for(j=0; j<64; j++){
1966                av_log(s->avctx, AV_LOG_DEBUG, "%5d",
1967                       block[i][s->idsp.idct_permutation[j]]);
1968            }
1969            av_log(s->avctx, AV_LOG_DEBUG, "\n");
1970        }
1971     }
1972
1973     s->current_picture.qscale_table[mb_xy] = s->qscale;
1974
1975     /* update DC predictors for P macroblocks */
1976     if (!s->mb_intra) {
1977         if (!is_mpeg12 && (s->h263_pred || s->h263_aic)) {
1978             if(s->mbintra_table[mb_xy])
1979                 ff_clean_intra_table_entries(s);
1980         } else {
1981             s->last_dc[0] =
1982             s->last_dc[1] =
1983             s->last_dc[2] = 128 << s->intra_dc_precision;
1984         }
1985     }
1986     else if (!is_mpeg12 && (s->h263_pred || s->h263_aic))
1987         s->mbintra_table[mb_xy]=1;
1988
1989     if ((s->avctx->flags & AV_CODEC_FLAG_PSNR) || s->frame_skip_threshold || s->frame_skip_factor ||
1990         !(s->encoding && (s->intra_only || s->pict_type == AV_PICTURE_TYPE_B) &&
1991           s->avctx->mb_decision != FF_MB_DECISION_RD)) { // FIXME precalc
1992         uint8_t *dest_y, *dest_cb, *dest_cr;
1993         int dct_linesize, dct_offset;
1994         op_pixels_func (*op_pix)[4];
1995         qpel_mc_func (*op_qpix)[16];
1996         const int linesize   = s->current_picture.f->linesize[0]; //not s->linesize as this would be wrong for field pics
1997         const int uvlinesize = s->current_picture.f->linesize[1];
1998         const int readable= s->pict_type != AV_PICTURE_TYPE_B || s->encoding || s->avctx->draw_horiz_band || lowres_flag;
1999         const int block_size= lowres_flag ? 8>>s->avctx->lowres : 8;
2000
2001         /* avoid copy if macroblock skipped in last frame too */
2002         /* skip only during decoding as we might trash the buffers during encoding a bit */
2003         if(!s->encoding){
2004             uint8_t *mbskip_ptr = &s->mbskip_table[mb_xy];
2005
2006             if (s->mb_skipped) {
2007                 s->mb_skipped= 0;
2008                 av_assert2(s->pict_type!=AV_PICTURE_TYPE_I);
2009                 *mbskip_ptr = 1;
2010             } else if(!s->current_picture.reference) {
2011                 *mbskip_ptr = 1;
2012             } else{
2013                 *mbskip_ptr = 0; /* not skipped */
2014             }
2015         }
2016
2017         dct_linesize = linesize << s->interlaced_dct;
2018         dct_offset   = s->interlaced_dct ? linesize : linesize * block_size;
2019
2020         if(readable){
2021             dest_y=  s->dest[0];
2022             dest_cb= s->dest[1];
2023             dest_cr= s->dest[2];
2024         }else{
2025             dest_y = s->sc.b_scratchpad;
2026             dest_cb= s->sc.b_scratchpad+16*linesize;
2027             dest_cr= s->sc.b_scratchpad+32*linesize;
2028         }
2029
2030         if (!s->mb_intra) {
2031             /* motion handling */
2032             /* decoding or more than one mb_type (MC was already done otherwise) */
2033             if(!s->encoding){
2034
2035                 if(HAVE_THREADS && s->avctx->active_thread_type&FF_THREAD_FRAME) {
2036                     if (s->mv_dir & MV_DIR_FORWARD) {
2037                         ff_thread_await_progress(&s->last_picture_ptr->tf,
2038                                                  lowest_referenced_row(s, 0),
2039                                                  0);
2040                     }
2041                     if (s->mv_dir & MV_DIR_BACKWARD) {
2042                         ff_thread_await_progress(&s->next_picture_ptr->tf,
2043                                                  lowest_referenced_row(s, 1),
2044                                                  0);
2045                     }
2046                 }
2047
2048                 if(lowres_flag){
2049                     h264_chroma_mc_func *op_pix = s->h264chroma.put_h264_chroma_pixels_tab;
2050
2051                     if (s->mv_dir & MV_DIR_FORWARD) {
2052                         MPV_motion_lowres(s, dest_y, dest_cb, dest_cr, 0, s->last_picture.f->data, op_pix);
2053                         op_pix = s->h264chroma.avg_h264_chroma_pixels_tab;
2054                     }
2055                     if (s->mv_dir & MV_DIR_BACKWARD) {
2056                         MPV_motion_lowres(s, dest_y, dest_cb, dest_cr, 1, s->next_picture.f->data, op_pix);
2057                     }
2058                 }else{
2059                     op_qpix = s->me.qpel_put;
2060                     if ((!s->no_rounding) || s->pict_type==AV_PICTURE_TYPE_B){
2061                         op_pix = s->hdsp.put_pixels_tab;
2062                     }else{
2063                         op_pix = s->hdsp.put_no_rnd_pixels_tab;
2064                     }
2065                     if (s->mv_dir & MV_DIR_FORWARD) {
2066                         ff_mpv_motion(s, dest_y, dest_cb, dest_cr, 0, s->last_picture.f->data, op_pix, op_qpix);
2067                         op_pix = s->hdsp.avg_pixels_tab;
2068                         op_qpix= s->me.qpel_avg;
2069                     }
2070                     if (s->mv_dir & MV_DIR_BACKWARD) {
2071                         ff_mpv_motion(s, dest_y, dest_cb, dest_cr, 1, s->next_picture.f->data, op_pix, op_qpix);
2072                     }
2073                 }
2074             }
2075
2076             /* skip dequant / idct if we are really late ;) */
2077             if(s->avctx->skip_idct){
2078                 if(  (s->avctx->skip_idct >= AVDISCARD_NONREF && s->pict_type == AV_PICTURE_TYPE_B)
2079                    ||(s->avctx->skip_idct >= AVDISCARD_NONKEY && s->pict_type != AV_PICTURE_TYPE_I)
2080                    || s->avctx->skip_idct >= AVDISCARD_ALL)
2081                     goto skip_idct;
2082             }
2083
2084             /* add dct residue */
2085             if(s->encoding || !(   s->msmpeg4_version || s->codec_id==AV_CODEC_ID_MPEG1VIDEO || s->codec_id==AV_CODEC_ID_MPEG2VIDEO
2086                                 || (s->codec_id==AV_CODEC_ID_MPEG4 && !s->mpeg_quant))){
2087                 add_dequant_dct(s, block[0], 0, dest_y                          , dct_linesize, s->qscale);
2088                 add_dequant_dct(s, block[1], 1, dest_y              + block_size, dct_linesize, s->qscale);
2089                 add_dequant_dct(s, block[2], 2, dest_y + dct_offset             , dct_linesize, s->qscale);
2090                 add_dequant_dct(s, block[3], 3, dest_y + dct_offset + block_size, dct_linesize, s->qscale);
2091
2092                 if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY)) {
2093                     if (s->chroma_y_shift){
2094                         add_dequant_dct(s, block[4], 4, dest_cb, uvlinesize, s->chroma_qscale);
2095                         add_dequant_dct(s, block[5], 5, dest_cr, uvlinesize, s->chroma_qscale);
2096                     }else{
2097                         dct_linesize >>= 1;
2098                         dct_offset >>=1;
2099                         add_dequant_dct(s, block[4], 4, dest_cb,              dct_linesize, s->chroma_qscale);
2100                         add_dequant_dct(s, block[5], 5, dest_cr,              dct_linesize, s->chroma_qscale);
2101                         add_dequant_dct(s, block[6], 6, dest_cb + dct_offset, dct_linesize, s->chroma_qscale);
2102                         add_dequant_dct(s, block[7], 7, dest_cr + dct_offset, dct_linesize, s->chroma_qscale);
2103                     }
2104                 }
2105             } else if(is_mpeg12 || (s->codec_id != AV_CODEC_ID_WMV2)){
2106                 add_dct(s, block[0], 0, dest_y                          , dct_linesize);
2107                 add_dct(s, block[1], 1, dest_y              + block_size, dct_linesize);
2108                 add_dct(s, block[2], 2, dest_y + dct_offset             , dct_linesize);
2109                 add_dct(s, block[3], 3, dest_y + dct_offset + block_size, dct_linesize);
2110
2111                 if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY)) {
2112                     if(s->chroma_y_shift){//Chroma420
2113                         add_dct(s, block[4], 4, dest_cb, uvlinesize);
2114                         add_dct(s, block[5], 5, dest_cr, uvlinesize);
2115                     }else{
2116                         //chroma422
2117                         dct_linesize = uvlinesize << s->interlaced_dct;
2118                         dct_offset   = s->interlaced_dct ? uvlinesize : uvlinesize*block_size;
2119
2120                         add_dct(s, block[4], 4, dest_cb, dct_linesize);
2121                         add_dct(s, block[5], 5, dest_cr, dct_linesize);
2122                         add_dct(s, block[6], 6, dest_cb+dct_offset, dct_linesize);
2123                         add_dct(s, block[7], 7, dest_cr+dct_offset, dct_linesize);
2124                         if(!s->chroma_x_shift){//Chroma444
2125                             add_dct(s, block[8], 8, dest_cb+block_size, dct_linesize);
2126                             add_dct(s, block[9], 9, dest_cr+block_size, dct_linesize);
2127                             add_dct(s, block[10], 10, dest_cb+block_size+dct_offset, dct_linesize);
2128                             add_dct(s, block[11], 11, dest_cr+block_size+dct_offset, dct_linesize);
2129                         }
2130                     }
2131                 }//fi gray
2132             }
2133             else if (CONFIG_WMV2_DECODER || CONFIG_WMV2_ENCODER) {
2134                 ff_wmv2_add_mb(s, block, dest_y, dest_cb, dest_cr);
2135             }
2136         } else {
2137             /* Only MPEG-4 Simple Studio Profile is supported in > 8-bit mode.
2138                TODO: Integrate 10-bit properly into mpegvideo.c so that ER works properly */
2139             if (s->avctx->bits_per_raw_sample > 8){
2140                 const int act_block_size = block_size * 2;
2141
2142                 if(s->dpcm_direction == 0) {
2143                     s->idsp.idct_put(dest_y,                           dct_linesize, (int16_t*)(*s->block32)[0]);
2144                     s->idsp.idct_put(dest_y              + act_block_size, dct_linesize, (int16_t*)(*s->block32)[1]);
2145                     s->idsp.idct_put(dest_y + dct_offset,              dct_linesize, (int16_t*)(*s->block32)[2]);
2146                     s->idsp.idct_put(dest_y + dct_offset + act_block_size, dct_linesize, (int16_t*)(*s->block32)[3]);
2147
2148                     dct_linesize = uvlinesize << s->interlaced_dct;
2149                     dct_offset   = s->interlaced_dct ? uvlinesize : uvlinesize*block_size;
2150
2151                     s->idsp.idct_put(dest_cb,              dct_linesize, (int16_t*)(*s->block32)[4]);
2152                     s->idsp.idct_put(dest_cr,              dct_linesize, (int16_t*)(*s->block32)[5]);
2153                     s->idsp.idct_put(dest_cb + dct_offset, dct_linesize, (int16_t*)(*s->block32)[6]);
2154                     s->idsp.idct_put(dest_cr + dct_offset, dct_linesize, (int16_t*)(*s->block32)[7]);
2155                     if(!s->chroma_x_shift){//Chroma444
2156                         s->idsp.idct_put(dest_cb + act_block_size,              dct_linesize, (int16_t*)(*s->block32)[8]);
2157                         s->idsp.idct_put(dest_cr + act_block_size,              dct_linesize, (int16_t*)(*s->block32)[9]);
2158                         s->idsp.idct_put(dest_cb + act_block_size + dct_offset, dct_linesize, (int16_t*)(*s->block32)[10]);
2159                         s->idsp.idct_put(dest_cr + act_block_size + dct_offset, dct_linesize, (int16_t*)(*s->block32)[11]);
2160                     }
2161                 } else if(s->dpcm_direction == 1) {
2162                     int i, w, h;
2163                     uint16_t *dest_pcm[3] = {(uint16_t*)dest_y, (uint16_t*)dest_cb, (uint16_t*)dest_cr};
2164                     int linesize[3] = {dct_linesize, uvlinesize, uvlinesize};
2165                     for(i = 0; i < 3; i++) {
2166                         int idx = 0;
2167                         int vsub = i ? s->chroma_y_shift : 0;
2168                         int hsub = i ? s->chroma_x_shift : 0;
2169                         for(h = 0; h < (16 >> vsub); h++){
2170                             for(w = 0; w < (16 >> hsub); w++)
2171                                 dest_pcm[i][w] = (*s->dpcm_macroblock)[i][idx++];
2172                             dest_pcm[i] += linesize[i] / 2;
2173                         }
2174                     }
2175                 } else if(s->dpcm_direction == -1) {
2176                     int i, w, h;
2177                     uint16_t *dest_pcm[3] = {(uint16_t*)dest_y, (uint16_t*)dest_cb, (uint16_t*)dest_cr};
2178                     int linesize[3] = {dct_linesize, uvlinesize, uvlinesize};
2179                     for(i = 0; i < 3; i++) {
2180                         int idx = 0;
2181                         int vsub = i ? s->chroma_y_shift : 0;
2182                         int hsub = i ? s->chroma_x_shift : 0;
2183                         dest_pcm[i] += (linesize[i] / 2) * ((16 >> vsub) - 1);
2184                         for(h = (16 >> vsub)-1; h >= 1; h--){
2185                             for(w = (16 >> hsub)-1; w >= 1; w--)
2186                                 dest_pcm[i][w] = (*s->dpcm_macroblock)[i][idx++];
2187                             dest_pcm[i] -= linesize[i] / 2;
2188                         }
2189                     }
2190                 }
2191             }
2192             /* dct only in intra block */
2193             else if(s->encoding || !(s->codec_id==AV_CODEC_ID_MPEG1VIDEO || s->codec_id==AV_CODEC_ID_MPEG2VIDEO)){
2194                 put_dct(s, block[0], 0, dest_y                          , dct_linesize, s->qscale);
2195                 put_dct(s, block[1], 1, dest_y              + block_size, dct_linesize, s->qscale);
2196                 put_dct(s, block[2], 2, dest_y + dct_offset             , dct_linesize, s->qscale);
2197                 put_dct(s, block[3], 3, dest_y + dct_offset + block_size, dct_linesize, s->qscale);
2198
2199                 if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY)) {
2200                     if(s->chroma_y_shift){
2201                         put_dct(s, block[4], 4, dest_cb, uvlinesize, s->chroma_qscale);
2202                         put_dct(s, block[5], 5, dest_cr, uvlinesize, s->chroma_qscale);
2203                     }else{
2204                         dct_offset >>=1;
2205                         dct_linesize >>=1;
2206                         put_dct(s, block[4], 4, dest_cb,              dct_linesize, s->chroma_qscale);
2207                         put_dct(s, block[5], 5, dest_cr,              dct_linesize, s->chroma_qscale);
2208                         put_dct(s, block[6], 6, dest_cb + dct_offset, dct_linesize, s->chroma_qscale);
2209                         put_dct(s, block[7], 7, dest_cr + dct_offset, dct_linesize, s->chroma_qscale);
2210                     }
2211                 }
2212             }else{
2213                 s->idsp.idct_put(dest_y,                           dct_linesize, block[0]);
2214                 s->idsp.idct_put(dest_y              + block_size, dct_linesize, block[1]);
2215                 s->idsp.idct_put(dest_y + dct_offset,              dct_linesize, block[2]);
2216                 s->idsp.idct_put(dest_y + dct_offset + block_size, dct_linesize, block[3]);
2217
2218                 if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY)) {
2219                     if(s->chroma_y_shift){
2220                         s->idsp.idct_put(dest_cb, uvlinesize, block[4]);
2221                         s->idsp.idct_put(dest_cr, uvlinesize, block[5]);
2222                     }else{
2223
2224                         dct_linesize = uvlinesize << s->interlaced_dct;
2225                         dct_offset   = s->interlaced_dct ? uvlinesize : uvlinesize*block_size;
2226
2227                         s->idsp.idct_put(dest_cb,              dct_linesize, block[4]);
2228                         s->idsp.idct_put(dest_cr,              dct_linesize, block[5]);
2229                         s->idsp.idct_put(dest_cb + dct_offset, dct_linesize, block[6]);
2230                         s->idsp.idct_put(dest_cr + dct_offset, dct_linesize, block[7]);
2231                         if(!s->chroma_x_shift){//Chroma444
2232                             s->idsp.idct_put(dest_cb + block_size,              dct_linesize, block[8]);
2233                             s->idsp.idct_put(dest_cr + block_size,              dct_linesize, block[9]);
2234                             s->idsp.idct_put(dest_cb + block_size + dct_offset, dct_linesize, block[10]);
2235                             s->idsp.idct_put(dest_cr + block_size + dct_offset, dct_linesize, block[11]);
2236                         }
2237                     }
2238                 }//gray
2239             }
2240         }
2241 skip_idct:
2242         if(!readable){
2243             s->hdsp.put_pixels_tab[0][0](s->dest[0], dest_y ,   linesize,16);
2244             if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY)) {
2245                 s->hdsp.put_pixels_tab[s->chroma_x_shift][0](s->dest[1], dest_cb, uvlinesize,16 >> s->chroma_y_shift);
2246                 s->hdsp.put_pixels_tab[s->chroma_x_shift][0](s->dest[2], dest_cr, uvlinesize,16 >> s->chroma_y_shift);
2247             }
2248         }
2249     }
2250 }
2251
2252 void ff_mpv_reconstruct_mb(MpegEncContext *s, int16_t block[12][64])
2253 {
2254 #if !CONFIG_SMALL
2255     if(s->out_format == FMT_MPEG1) {
2256         if(s->avctx->lowres) mpv_reconstruct_mb_internal(s, block, 1, 1);
2257         else                 mpv_reconstruct_mb_internal(s, block, 0, 1);
2258     } else
2259 #endif
2260     if(s->avctx->lowres) mpv_reconstruct_mb_internal(s, block, 1, 0);
2261     else                  mpv_reconstruct_mb_internal(s, block, 0, 0);
2262 }
2263
2264 void ff_mpeg_draw_horiz_band(MpegEncContext *s, int y, int h)
2265 {
2266     ff_draw_horiz_band(s->avctx, s->current_picture_ptr->f,
2267                        s->last_picture_ptr ? s->last_picture_ptr->f : NULL, y, h, s->picture_structure,
2268                        s->first_field, s->low_delay);
2269 }
2270
2271 void ff_init_block_index(MpegEncContext *s){ //FIXME maybe rename
2272     const int linesize   = s->current_picture.f->linesize[0]; //not s->linesize as this would be wrong for field pics
2273     const int uvlinesize = s->current_picture.f->linesize[1];
2274     const int width_of_mb = (4 + (s->avctx->bits_per_raw_sample > 8)) - s->avctx->lowres;
2275     const int height_of_mb = 4 - s->avctx->lowres;
2276
2277     s->block_index[0]= s->b8_stride*(s->mb_y*2    ) - 2 + s->mb_x*2;
2278     s->block_index[1]= s->b8_stride*(s->mb_y*2    ) - 1 + s->mb_x*2;
2279     s->block_index[2]= s->b8_stride*(s->mb_y*2 + 1) - 2 + s->mb_x*2;
2280     s->block_index[3]= s->b8_stride*(s->mb_y*2 + 1) - 1 + s->mb_x*2;
2281     s->block_index[4]= s->mb_stride*(s->mb_y + 1)                + s->b8_stride*s->mb_height*2 + s->mb_x - 1;
2282     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;
2283     //block_index is not used by mpeg2, so it is not affected by chroma_format
2284
2285     s->dest[0] = s->current_picture.f->data[0] + (int)((s->mb_x - 1U) <<  width_of_mb);
2286     s->dest[1] = s->current_picture.f->data[1] + (int)((s->mb_x - 1U) << (width_of_mb - s->chroma_x_shift));
2287     s->dest[2] = s->current_picture.f->data[2] + (int)((s->mb_x - 1U) << (width_of_mb - s->chroma_x_shift));
2288
2289     if(!(s->pict_type==AV_PICTURE_TYPE_B && s->avctx->draw_horiz_band && s->picture_structure==PICT_FRAME))
2290     {
2291         if(s->picture_structure==PICT_FRAME){
2292         s->dest[0] += s->mb_y *   linesize << height_of_mb;
2293         s->dest[1] += s->mb_y * uvlinesize << (height_of_mb - s->chroma_y_shift);
2294         s->dest[2] += s->mb_y * uvlinesize << (height_of_mb - s->chroma_y_shift);
2295         }else{
2296             s->dest[0] += (s->mb_y>>1) *   linesize << height_of_mb;
2297             s->dest[1] += (s->mb_y>>1) * uvlinesize << (height_of_mb - s->chroma_y_shift);
2298             s->dest[2] += (s->mb_y>>1) * uvlinesize << (height_of_mb - s->chroma_y_shift);
2299             av_assert1((s->mb_y&1) == (s->picture_structure == PICT_BOTTOM_FIELD));
2300         }
2301     }
2302 }
2303
2304 void ff_mpeg_flush(AVCodecContext *avctx){
2305     int i;
2306     MpegEncContext *s = avctx->priv_data;
2307
2308     if (!s || !s->picture)
2309         return;
2310
2311     for (i = 0; i < MAX_PICTURE_COUNT; i++)
2312         ff_mpeg_unref_picture(s->avctx, &s->picture[i]);
2313     s->current_picture_ptr = s->last_picture_ptr = s->next_picture_ptr = NULL;
2314
2315     ff_mpeg_unref_picture(s->avctx, &s->current_picture);
2316     ff_mpeg_unref_picture(s->avctx, &s->last_picture);
2317     ff_mpeg_unref_picture(s->avctx, &s->next_picture);
2318
2319     s->mb_x= s->mb_y= 0;
2320     s->closed_gop= 0;
2321
2322     s->parse_context.state= -1;
2323     s->parse_context.frame_start_found= 0;
2324     s->parse_context.overread= 0;
2325     s->parse_context.overread_index= 0;
2326     s->parse_context.index= 0;
2327     s->parse_context.last_index= 0;
2328     s->bitstream_buffer_size=0;
2329     s->pp_time=0;
2330 }
2331
2332 /**
2333  * set qscale and update qscale dependent variables.
2334  */
2335 void ff_set_qscale(MpegEncContext * s, int qscale)
2336 {
2337     if (qscale < 1)
2338         qscale = 1;
2339     else if (qscale > 31)
2340         qscale = 31;
2341
2342     s->qscale = qscale;
2343     s->chroma_qscale= s->chroma_qscale_table[qscale];
2344
2345     s->y_dc_scale= s->y_dc_scale_table[ qscale ];
2346     s->c_dc_scale= s->c_dc_scale_table[ s->chroma_qscale ];
2347 }
2348
2349 void ff_mpv_report_decode_progress(MpegEncContext *s)
2350 {
2351     if (s->pict_type != AV_PICTURE_TYPE_B && !s->partitioned_frame && !s->er.error_occurred)
2352         ff_thread_report_progress(&s->current_picture_ptr->tf, s->mb_y, 0);
2353 }