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