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/timer.h"
  36 #include "avcodec.h"
  37 #include "blockdsp.h"
  38 #include "h264chroma.h"
  39 #include "idctdsp.h"
  40 #include "internal.h"
  41 #include "mathops.h"
  42 #include "mpeg_er.h"
  43 #include "mpegutils.h"
  44 #include "mpegvideo.h"
  45 #include "mpegvideodata.h"
  46 #include "mjpegenc.h"
  47 #include "msmpeg4.h"
  48 #include "qpeldsp.h"
  49 #include "thread.h"
  50 #include "wmv2.h"
  51 #include <limits.h>
  52
  53 static void dct_unquantize_mpeg1_intra_c(MpegEncContext *s,
  54                                    int16_t *block, int n, int qscale)
  55 {
  56     int i, level, nCoeffs;
  57     const uint16_t *quant_matrix;
  58
  59     nCoeffs= s->block_last_index[n];
  60
  61     block[0] *= n < 4 ? s->y_dc_scale : s->c_dc_scale;
  62     /* XXX: only MPEG-1 */
  63     quant_matrix = s->intra_matrix;
  64     for(i=1;i<=nCoeffs;i++) {
  65         int j= s->intra_scantable.permutated[i];
  66         level = block[j];
  67         if (level) {
  68             if (level < 0) {
  69                 level = -level;
  70                 level = (int)(level * qscale * quant_matrix[j]) >> 3;
  71                 level = (level - 1) | 1;
  72                 level = -level;
  73             } else {
  74                 level = (int)(level * qscale * quant_matrix[j]) >> 3;
  75                 level = (level - 1) | 1;
  76             }
  77             block[j] = level;
  78         }
  79     }
  80 }
  81
  82 static void dct_unquantize_mpeg1_inter_c(MpegEncContext *s,
  83                                    int16_t *block, int n, int qscale)
  84 {
  85     int i, level, nCoeffs;
  86     const uint16_t *quant_matrix;
  87
  88     nCoeffs= s->block_last_index[n];
  89
  90     quant_matrix = s->inter_matrix;
  91     for(i=0; i<=nCoeffs; i++) {
  92         int j= s->intra_scantable.permutated[i];
  93         level = block[j];
  94         if (level) {
  95             if (level < 0) {
  96                 level = -level;
  97                 level = (((level << 1) + 1) * qscale *
  98                          ((int) (quant_matrix[j]))) >> 4;
  99                 level = (level - 1) | 1;
 100                 level = -level;
 101             } else {
 102                 level = (((level << 1) + 1) * qscale *
 103                          ((int) (quant_matrix[j]))) >> 4;
 104                 level = (level - 1) | 1;
 105             }
 106             block[j] = level;
 107         }
 108     }
 109 }
 110
 111 static void dct_unquantize_mpeg2_intra_c(MpegEncContext *s,
 112                                    int16_t *block, int n, int qscale)
 113 {
 114     int i, level, nCoeffs;
 115     const uint16_t *quant_matrix;
 116
 117     if (s->q_scale_type) qscale = ff_mpeg2_non_linear_qscale[qscale];
 118     else                 qscale <<= 1;
 119
 120     if(s->alternate_scan) nCoeffs= 63;
 121     else nCoeffs= s->block_last_index[n];
 122
 123     block[0] *= n < 4 ? s->y_dc_scale : s->c_dc_scale;
 124     quant_matrix = s->intra_matrix;
 125     for(i=1;i<=nCoeffs;i++) {
 126         int j= s->intra_scantable.permutated[i];
 127         level = block[j];
 128         if (level) {
 129             if (level < 0) {
 130                 level = -level;
 131                 level = (int)(level * qscale * quant_matrix[j]) >> 4;
 132                 level = -level;
 133             } else {
 134                 level = (int)(level * qscale * quant_matrix[j]) >> 4;
 135             }
 136             block[j] = level;
 137         }
 138     }
 139 }
 140
 141 static void dct_unquantize_mpeg2_intra_bitexact(MpegEncContext *s,
 142                                    int16_t *block, int n, int qscale)
 143 {
 144     int i, level, nCoeffs;
 145     const uint16_t *quant_matrix;
 146     int sum=-1;
 147
 148     if (s->q_scale_type) qscale = ff_mpeg2_non_linear_qscale[qscale];
 149     else                 qscale <<= 1;
 150
 151     if(s->alternate_scan) nCoeffs= 63;
 152     else nCoeffs= s->block_last_index[n];
 153
 154     block[0] *= n < 4 ? s->y_dc_scale : s->c_dc_scale;
 155     sum += block[0];
 156     quant_matrix = s->intra_matrix;
 157     for(i=1;i<=nCoeffs;i++) {
 158         int j= s->intra_scantable.permutated[i];
 159         level = block[j];
 160         if (level) {
 161             if (level < 0) {
 162                 level = -level;
 163                 level = (int)(level * qscale * quant_matrix[j]) >> 4;
 164                 level = -level;
 165             } else {
 166                 level = (int)(level * qscale * quant_matrix[j]) >> 4;
 167             }
 168             block[j] = level;
 169             sum+=level;
 170         }
 171     }
 172     block[63]^=sum&1;
 173 }
 174
 175 static void dct_unquantize_mpeg2_inter_c(MpegEncContext *s,
 176                                    int16_t *block, int n, int qscale)
 177 {
 178     int i, level, nCoeffs;
 179     const uint16_t *quant_matrix;
 180     int sum=-1;
 181
 182     if (s->q_scale_type) qscale = ff_mpeg2_non_linear_qscale[qscale];
 183     else                 qscale <<= 1;
 184
 185     if(s->alternate_scan) nCoeffs= 63;
 186     else nCoeffs= s->block_last_index[n];
 187
 188     quant_matrix = s->inter_matrix;
 189     for(i=0; i<=nCoeffs; i++) {
 190         int j= s->intra_scantable.permutated[i];
 191         level = block[j];
 192         if (level) {
 193             if (level < 0) {
 194                 level = -level;
 195                 level = (((level << 1) + 1) * qscale *
 196                          ((int) (quant_matrix[j]))) >> 5;
 197                 level = -level;
 198             } else {
 199                 level = (((level << 1) + 1) * qscale *
 200                          ((int) (quant_matrix[j]))) >> 5;
 201             }
 202             block[j] = level;
 203             sum+=level;
 204         }
 205     }
 206     block[63]^=sum&1;
 207 }
 208
 209 static void dct_unquantize_h263_intra_c(MpegEncContext *s,
 210                                   int16_t *block, int n, int qscale)
 211 {
 212     int i, level, qmul, qadd;
 213     int nCoeffs;
 214
 215     av_assert2(s->block_last_index[n]>=0 || s->h263_aic);
 216
 217     qmul = qscale << 1;
 218
 219     if (!s->h263_aic) {
 220         block[0] *= n < 4 ? s->y_dc_scale : s->c_dc_scale;
 221         qadd = (qscale - 1) | 1;
 222     }else{
 223         qadd = 0;
 224     }
 225     if(s->ac_pred)
 226         nCoeffs=63;
 227     else
 228         nCoeffs= s->intra_scantable.raster_end[ s->block_last_index[n] ];
 229
 230     for(i=1; i<=nCoeffs; i++) {
 231         level = block[i];
 232         if (level) {
 233             if (level < 0) {
 234                 level = level * qmul - qadd;
 235             } else {
 236                 level = level * qmul + qadd;
 237             }
 238             block[i] = level;
 239         }
 240     }
 241 }
 242
 243 static void dct_unquantize_h263_inter_c(MpegEncContext *s,
 244                                   int16_t *block, int n, int qscale)
 245 {
 246     int i, level, qmul, qadd;
 247     int nCoeffs;
 248
 249     av_assert2(s->block_last_index[n]>=0);
 250
 251     qadd = (qscale - 1) | 1;
 252     qmul = qscale << 1;
 253
 254     nCoeffs= s->inter_scantable.raster_end[ s->block_last_index[n] ];
 255
 256     for(i=0; i<=nCoeffs; i++) {
 257         level = block[i];
 258         if (level) {
 259             if (level < 0) {
 260                 level = level * qmul - qadd;
 261             } else {
 262                 level = level * qmul + qadd;
 263             }
 264             block[i] = level;
 265         }
 266     }
 267 }
 268
 269
 270 static void gray16(uint8_t *dst, const uint8_t *src, ptrdiff_t linesize, int h)
 271 {
 272     while(h--)
 273         memset(dst + h*linesize, 128, 16);
 274 }
 275
 276 static void gray8(uint8_t *dst, const uint8_t *src, ptrdiff_t linesize, int h)
 277 {
 278     while(h--)
 279         memset(dst + h*linesize, 128, 8);
 280 }
 281
 282 /* init common dct for both encoder and decoder */
 283 static av_cold int dct_init(MpegEncContext *s)
 284 {
 285     ff_blockdsp_init(&s->bdsp, s->avctx);
 286     ff_h264chroma_init(&s->h264chroma, 8); //for lowres
 287     ff_hpeldsp_init(&s->hdsp, s->avctx->flags);
 288     ff_mpegvideodsp_init(&s->mdsp);
 289     ff_videodsp_init(&s->vdsp, s->avctx->bits_per_raw_sample);
 290
 291     if (s->avctx->debug & FF_DEBUG_NOMC) {
 292         int i;
 293         for (i=0; i<4; i++) {
 294             s->hdsp.avg_pixels_tab[0][i] = gray16;
 295             s->hdsp.put_pixels_tab[0][i] = gray16;
 296             s->hdsp.put_no_rnd_pixels_tab[0][i] = gray16;
 297
 298             s->hdsp.avg_pixels_tab[1][i] = gray8;
 299             s->hdsp.put_pixels_tab[1][i] = gray8;
 300             s->hdsp.put_no_rnd_pixels_tab[1][i] = gray8;
 301         }
 302     }
 303
 304     s->dct_unquantize_h263_intra = dct_unquantize_h263_intra_c;
 305     s->dct_unquantize_h263_inter = dct_unquantize_h263_inter_c;
 306     s->dct_unquantize_mpeg1_intra = dct_unquantize_mpeg1_intra_c;
 307     s->dct_unquantize_mpeg1_inter = dct_unquantize_mpeg1_inter_c;
 308     s->dct_unquantize_mpeg2_intra = dct_unquantize_mpeg2_intra_c;
 309     if (s->avctx->flags & AV_CODEC_FLAG_BITEXACT)
 310         s->dct_unquantize_mpeg2_intra = dct_unquantize_mpeg2_intra_bitexact;
 311     s->dct_unquantize_mpeg2_inter = dct_unquantize_mpeg2_inter_c;
 312
 313     if (HAVE_INTRINSICS_NEON)
 314         ff_mpv_common_init_neon(s);
 315
 316     if (ARCH_ALPHA)
 317         ff_mpv_common_init_axp(s);
 318     if (ARCH_ARM)
 319         ff_mpv_common_init_arm(s);
 320     if (ARCH_PPC)
 321         ff_mpv_common_init_ppc(s);
 322     if (ARCH_X86)
 323         ff_mpv_common_init_x86(s);
 324     if (ARCH_MIPS)
 325         ff_mpv_common_init_mips(s);
 326
 327     return 0;
 328 }
 329
 330 av_cold void ff_mpv_idct_init(MpegEncContext *s)
 331 {
 332     ff_idctdsp_init(&s->idsp, s->avctx);
 333
 334     /* load & permutate scantables
 335      * note: only wmv uses different ones
 336      */
 337     if (s->alternate_scan) {
 338         ff_init_scantable(s->idsp.idct_permutation, &s->inter_scantable, ff_alternate_vertical_scan);
 339         ff_init_scantable(s->idsp.idct_permutation, &s->intra_scantable, ff_alternate_vertical_scan);
 340     } else {
 341         ff_init_scantable(s->idsp.idct_permutation, &s->inter_scantable, ff_zigzag_direct);
 342         ff_init_scantable(s->idsp.idct_permutation, &s->intra_scantable, ff_zigzag_direct);
 343     }
 344     ff_init_scantable(s->idsp.idct_permutation, &s->intra_h_scantable, ff_alternate_horizontal_scan);
 345     ff_init_scantable(s->idsp.idct_permutation, &s->intra_v_scantable, ff_alternate_vertical_scan);
 346 }
 347
 348 static int alloc_picture(MpegEncContext *s, Picture *pic, int shared)
 349 {
 350     return ff_alloc_picture(s->avctx, pic, &s->me, &s->sc, shared, 0,
 351                             s->chroma_x_shift, s->chroma_y_shift, s->out_format,
 352                             s->mb_stride, s->mb_width, s->mb_height, s->b8_stride,
 353                             &s->linesize, &s->uvlinesize);
 354 }
 355
 356 static int init_duplicate_context(MpegEncContext *s)
 357 {
 358     int y_size = s->b8_stride * (2 * s->mb_height + 1);
 359     int c_size = s->mb_stride * (s->mb_height + 1);
 360     int yc_size = y_size + 2 * c_size;
 361     int i;
 362
 363     if (s->mb_height & 1)
 364         yc_size += 2*s->b8_stride + 2*s->mb_stride;
 365
 366     s->sc.edge_emu_buffer =
 367     s->me.scratchpad   =
 368     s->me.temp         =
 369     s->sc.rd_scratchpad   =
 370     s->sc.b_scratchpad    =
 371     s->sc.obmc_scratchpad = NULL;
 372
 373     if (s->encoding) {
 374         FF_ALLOCZ_OR_GOTO(s->avctx, s->me.map,
 375                           ME_MAP_SIZE * sizeof(uint32_t), fail)
 376         FF_ALLOCZ_OR_GOTO(s->avctx, s->me.score_map,
 377                           ME_MAP_SIZE * sizeof(uint32_t), fail)
 378         if (s->noise_reduction) {
 379             FF_ALLOCZ_OR_GOTO(s->avctx, s->dct_error_sum,
 380                               2 * 64 * sizeof(int), fail)
 381         }
 382     }
 383     FF_ALLOCZ_OR_GOTO(s->avctx, s->blocks, 64 * 12 * 2 * sizeof(int16_t), fail)
 384     s->block = s->blocks[0];
 385
 386     for (i = 0; i < 12; i++) {
 387         s->pblocks[i] = &s->block[i];
 388     }
 389
 390     FF_ALLOCZ_OR_GOTO(s->avctx, s->block32, sizeof(*s->block32), fail)
 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         FF_ALLOCZ_OR_GOTO(s->avctx, s->ac_val_base,
 400                           yc_size * sizeof(int16_t) * 16, fail);
 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 fail:
 408     return -1; // free() through ff_mpv_common_end()
 409 }
 410
 411 static void free_duplicate_context(MpegEncContext *s)
 412 {
 413     if (!s)
 414         return;
 415
 416     av_freep(&s->sc.edge_emu_buffer);
 417     av_freep(&s->me.scratchpad);
 418     s->me.temp =
 419     s->sc.rd_scratchpad =
 420     s->sc.b_scratchpad =
 421     s->sc.obmc_scratchpad = NULL;
 422
 423     av_freep(&s->dct_error_sum);
 424     av_freep(&s->me.map);
 425     av_freep(&s->me.score_map);
 426     av_freep(&s->blocks);
 427     av_freep(&s->block32);
 428     av_freep(&s->ac_val_base);
 429     s->block = NULL;
 430 }
 431
 432 static void backup_duplicate_context(MpegEncContext *bak, MpegEncContext *src)
 433 {
 434 #define COPY(a) bak->a = src->a
 435     COPY(sc.edge_emu_buffer);
 436     COPY(me.scratchpad);
 437     COPY(me.temp);
 438     COPY(sc.rd_scratchpad);
 439     COPY(sc.b_scratchpad);
 440     COPY(sc.obmc_scratchpad);
 441     COPY(me.map);
 442     COPY(me.score_map);
 443     COPY(blocks);
 444     COPY(block);
 445     COPY(block32);
 446     COPY(start_mb_y);
 447     COPY(end_mb_y);
 448     COPY(me.map_generation);
 449     COPY(pb);
 450     COPY(dct_error_sum);
 451     COPY(dct_count[0]);
 452     COPY(dct_count[1]);
 453     COPY(ac_val_base);
 454     COPY(ac_val[0]);
 455     COPY(ac_val[1]);
 456     COPY(ac_val[2]);
 457 #undef COPY
 458 }
 459
 460 int ff_update_duplicate_context(MpegEncContext *dst, MpegEncContext *src)
 461 {
 462     MpegEncContext bak;
 463     int i, ret;
 464     // FIXME copy only needed parts
 465     // START_TIMER
 466     backup_duplicate_context(&bak, dst);
 467     memcpy(dst, src, sizeof(MpegEncContext));
 468     backup_duplicate_context(dst, &bak);
 469     for (i = 0; i < 12; i++) {
 470         dst->pblocks[i] = &dst->block[i];
 471     }
 472     if (dst->avctx->codec_tag == AV_RL32("VCR2")) {
 473         // exchange uv
 474         FFSWAP(void *, dst->pblocks[4], dst->pblocks[5]);
 475     }
 476     if (!dst->sc.edge_emu_buffer &&
 477         (ret = ff_mpeg_framesize_alloc(dst->avctx, &dst->me,
 478                                        &dst->sc, dst->linesize)) < 0) {
 479         av_log(dst->avctx, AV_LOG_ERROR, "failed to allocate context "
 480                "scratch buffers.\n");
 481         return ret;
 482     }
 483     // STOP_TIMER("update_duplicate_context")
 484     // about 10k cycles / 0.01 sec for  1000frames on 1ghz with 2 threads
 485     return 0;
 486 }
 487
 488 int ff_mpeg_update_thread_context(AVCodecContext *dst,
 489                                   const AVCodecContext *src)
 490 {
 491     int i, ret;
 492     MpegEncContext *s = dst->priv_data, *s1 = src->priv_data;
 493
 494     if (dst == src)
 495         return 0;
 496
 497     av_assert0(s != s1);
 498
 499     // FIXME can parameters change on I-frames?
 500     // in that case dst may need a reinit
 501     if (!s->context_initialized) {
 502         int err;
 503         memcpy(s, s1, sizeof(MpegEncContext));
 504
 505         s->avctx                 = dst;
 506         s->bitstream_buffer      = NULL;
 507         s->bitstream_buffer_size = s->allocated_bitstream_buffer_size = 0;
 508
 509         if (s1->context_initialized){
 510 //             s->picture_range_start  += MAX_PICTURE_COUNT;
 511 //             s->picture_range_end    += MAX_PICTURE_COUNT;
 512             ff_mpv_idct_init(s);
 513             if((err = ff_mpv_common_init(s)) < 0){
 514                 memset(s, 0, sizeof(MpegEncContext));
 515                 s->avctx = dst;
 516                 return err;
 517             }
 518         }
 519     }
 520
 521     if (s->height != s1->height || s->width != s1->width || s->context_reinit) {
 522         s->context_reinit = 0;
 523         s->height = s1->height;
 524         s->width  = s1->width;
 525         if ((ret = ff_mpv_common_frame_size_change(s)) < 0)
 526             return ret;
 527     }
 528
 529     s->avctx->coded_height  = s1->avctx->coded_height;
 530     s->avctx->coded_width   = s1->avctx->coded_width;
 531     s->avctx->width         = s1->avctx->width;
 532     s->avctx->height        = s1->avctx->height;
 533
 534     s->coded_picture_number = s1->coded_picture_number;
 535     s->picture_number       = s1->picture_number;
 536
 537     av_assert0(!s->picture || s->picture != s1->picture);
 538     if(s->picture)
 539     for (i = 0; i < MAX_PICTURE_COUNT; i++) {
 540         ff_mpeg_unref_picture(s->avctx, &s->picture[i]);
 541         if (s1->picture && s1->picture[i].f->buf[0] &&
 542             (ret = ff_mpeg_ref_picture(s->avctx, &s->picture[i], &s1->picture[i])) < 0)
 543             return ret;
 544     }
 545
 546 #define UPDATE_PICTURE(pic)\
 547 do {\
 548     ff_mpeg_unref_picture(s->avctx, &s->pic);\
 549     if (s1->pic.f && s1->pic.f->buf[0])\
 550         ret = ff_mpeg_ref_picture(s->avctx, &s->pic, &s1->pic);\
 551     else\
 552         ret = ff_update_picture_tables(&s->pic, &s1->pic);\
 553     if (ret < 0)\
 554         return ret;\
 555 } while (0)
 556
 557     UPDATE_PICTURE(current_picture);
 558     UPDATE_PICTURE(last_picture);
 559     UPDATE_PICTURE(next_picture);
 560
 561 #define REBASE_PICTURE(pic, new_ctx, old_ctx)                                 \
 562     ((pic && pic >= old_ctx->picture &&                                       \
 563       pic < old_ctx->picture + MAX_PICTURE_COUNT) ?                           \
 564         &new_ctx->picture[pic - old_ctx->picture] : NULL)
 565
 566     s->last_picture_ptr    = REBASE_PICTURE(s1->last_picture_ptr,    s, s1);
 567     s->current_picture_ptr = REBASE_PICTURE(s1->current_picture_ptr, s, s1);
 568     s->next_picture_ptr    = REBASE_PICTURE(s1->next_picture_ptr,    s, s1);
 569
 570     // Error/bug resilience
 571     s->next_p_frame_damaged = s1->next_p_frame_damaged;
 572     s->workaround_bugs      = s1->workaround_bugs;
 573     s->padding_bug_score    = s1->padding_bug_score;
 574
 575     // MPEG-4 timing info
 576     memcpy(&s->last_time_base, &s1->last_time_base,
 577            (char *) &s1->pb_field_time + sizeof(s1->pb_field_time) -
 578            (char *) &s1->last_time_base);
 579
 580     // B-frame info
 581     s->max_b_frames = s1->max_b_frames;
 582     s->low_delay    = s1->low_delay;
 583     s->droppable    = s1->droppable;
 584
 585     // DivX handling (doesn't work)
 586     s->divx_packed  = s1->divx_packed;
 587
 588     if (s1->bitstream_buffer) {
 589         if (s1->bitstream_buffer_size +
 590             AV_INPUT_BUFFER_PADDING_SIZE > s->allocated_bitstream_buffer_size) {
 591             av_fast_malloc(&s->bitstream_buffer,
 592                            &s->allocated_bitstream_buffer_size,
 593                            s1->allocated_bitstream_buffer_size);
 594             if (!s->bitstream_buffer) {
 595                 s->bitstream_buffer_size = 0;
 596                 return AVERROR(ENOMEM);
 597             }
 598         }
 599         s->bitstream_buffer_size = s1->bitstream_buffer_size;
 600         memcpy(s->bitstream_buffer, s1->bitstream_buffer,
 601                s1->bitstream_buffer_size);
 602         memset(s->bitstream_buffer + s->bitstream_buffer_size, 0,
 603                AV_INPUT_BUFFER_PADDING_SIZE);
 604     }
 605
 606     // linesize-dependent scratch buffer allocation
 607     if (!s->sc.edge_emu_buffer)
 608         if (s1->linesize) {
 609             if (ff_mpeg_framesize_alloc(s->avctx, &s->me,
 610                                         &s->sc, s1->linesize) < 0) {
 611                 av_log(s->avctx, AV_LOG_ERROR, "Failed to allocate context "
 612                        "scratch buffers.\n");
 613                 return AVERROR(ENOMEM);
 614             }
 615         } else {
 616             av_log(s->avctx, AV_LOG_ERROR, "Context scratch buffers could not "
 617                    "be allocated due to unknown size.\n");
 618         }
 619
 620     // MPEG-2/interlacing info
 621     memcpy(&s->progressive_sequence, &s1->progressive_sequence,
 622            (char *) &s1->rtp_mode - (char *) &s1->progressive_sequence);
 623
 624     if (!s1->first_field) {
 625         s->last_pict_type = s1->pict_type;
 626         if (s1->current_picture_ptr)
 627             s->last_lambda_for[s1->pict_type] = s1->current_picture_ptr->f->quality;
 628     }
 629
 630     return 0;
 631 }
 632
 633 /**
 634  * Set the given MpegEncContext to common defaults
 635  * (same for encoding and decoding).
 636  * The changed fields will not depend upon the
 637  * prior state of the MpegEncContext.
 638  */
 639 void ff_mpv_common_defaults(MpegEncContext *s)
 640 {
 641     s->y_dc_scale_table      =
 642     s->c_dc_scale_table      = ff_mpeg1_dc_scale_table;
 643     s->chroma_qscale_table   = ff_default_chroma_qscale_table;
 644     s->progressive_frame     = 1;
 645     s->progressive_sequence  = 1;
 646     s->picture_structure     = PICT_FRAME;
 647
 648     s->coded_picture_number  = 0;
 649     s->picture_number        = 0;
 650
 651     s->f_code                = 1;
 652     s->b_code                = 1;
 653
 654     s->slice_context_count   = 1;
 655 }
 656
 657 /**
 658  * Set the given MpegEncContext to defaults for decoding.
 659  * the changed fields will not depend upon
 660  * the prior state of the MpegEncContext.
 661  */
 662 void ff_mpv_decode_defaults(MpegEncContext *s)
 663 {
 664     ff_mpv_common_defaults(s);
 665 }
 666
 667 void ff_mpv_decode_init(MpegEncContext *s, AVCodecContext *avctx)
 668 {
 669     s->avctx           = avctx;
 670     s->width           = avctx->coded_width;
 671     s->height          = avctx->coded_height;
 672     s->codec_id        = avctx->codec->id;
 673     s->workaround_bugs = avctx->workaround_bugs;
 674
 675     /* convert fourcc to upper case */
 676     s->codec_tag          = avpriv_toupper4(avctx->codec_tag);
 677 }
 678
 679 /**
 680  * Initialize and allocates MpegEncContext fields dependent on the resolution.
 681  */
 682 static int init_context_frame(MpegEncContext *s)
 683 {
 684     int y_size, c_size, yc_size, i, mb_array_size, mv_table_size, x, y;
 685
 686     s->mb_width   = (s->width + 15) / 16;
 687     s->mb_stride  = s->mb_width + 1;
 688     s->b8_stride  = s->mb_width * 2 + 1;
 689     mb_array_size = s->mb_height * s->mb_stride;
 690     mv_table_size = (s->mb_height + 2) * s->mb_stride + 1;
 691
 692     /* set default edge pos, will be overridden
 693      * in decode_header if needed */
 694     s->h_edge_pos = s->mb_width * 16;
 695     s->v_edge_pos = s->mb_height * 16;
 696
 697     s->mb_num     = s->mb_width * s->mb_height;
 698
 699     s->block_wrap[0] =
 700     s->block_wrap[1] =
 701     s->block_wrap[2] =
 702     s->block_wrap[3] = s->b8_stride;
 703     s->block_wrap[4] =
 704     s->block_wrap[5] = s->mb_stride;
 705
 706     y_size  = s->b8_stride * (2 * s->mb_height + 1);
 707     c_size  = s->mb_stride * (s->mb_height + 1);
 708     yc_size = y_size + 2   * c_size;
 709
 710     if (s->mb_height & 1)
 711         yc_size += 2*s->b8_stride + 2*s->mb_stride;
 712
 713     FF_ALLOCZ_OR_GOTO(s->avctx, s->mb_index2xy, (s->mb_num + 1) * sizeof(int),
 714                       fail); // error resilience code looks cleaner with this
 715     for (y = 0; y < s->mb_height; y++)
 716         for (x = 0; x < s->mb_width; x++)
 717             s->mb_index2xy[x + y * s->mb_width] = x + y * s->mb_stride;
 718
 719     s->mb_index2xy[s->mb_height * s->mb_width] = (s->mb_height - 1) * s->mb_stride + s->mb_width; // FIXME really needed?
 720
 721     if (s->encoding) {
 722         /* Allocate MV tables */
 723         FF_ALLOCZ_OR_GOTO(s->avctx, s->p_mv_table_base,                 mv_table_size * 2 * sizeof(int16_t), fail)
 724         FF_ALLOCZ_OR_GOTO(s->avctx, s->b_forw_mv_table_base,            mv_table_size * 2 * sizeof(int16_t), fail)
 725         FF_ALLOCZ_OR_GOTO(s->avctx, s->b_back_mv_table_base,            mv_table_size * 2 * sizeof(int16_t), fail)
 726         FF_ALLOCZ_OR_GOTO(s->avctx, s->b_bidir_forw_mv_table_base,      mv_table_size * 2 * sizeof(int16_t), fail)
 727         FF_ALLOCZ_OR_GOTO(s->avctx, s->b_bidir_back_mv_table_base,      mv_table_size * 2 * sizeof(int16_t), fail)
 728         FF_ALLOCZ_OR_GOTO(s->avctx, s->b_direct_mv_table_base,          mv_table_size * 2 * sizeof(int16_t), fail)
 729         s->p_mv_table            = s->p_mv_table_base + s->mb_stride + 1;
 730         s->b_forw_mv_table       = s->b_forw_mv_table_base + s->mb_stride + 1;
 731         s->b_back_mv_table       = s->b_back_mv_table_base + s->mb_stride + 1;
 732         s->b_bidir_forw_mv_table = s->b_bidir_forw_mv_table_base + s->mb_stride + 1;
 733         s->b_bidir_back_mv_table = s->b_bidir_back_mv_table_base + s->mb_stride + 1;
 734         s->b_direct_mv_table     = s->b_direct_mv_table_base + s->mb_stride + 1;
 735
 736         /* Allocate MB type table */
 737         FF_ALLOCZ_OR_GOTO(s->avctx, s->mb_type, mb_array_size * sizeof(uint16_t), fail) // needed for encoding
 738
 739         FF_ALLOCZ_OR_GOTO(s->avctx, s->lambda_table, mb_array_size * sizeof(int), fail)
 740
 741         FF_ALLOC_OR_GOTO(s->avctx, s->cplx_tab,
 742                          mb_array_size * sizeof(float), fail);
 743         FF_ALLOC_OR_GOTO(s->avctx, s->bits_tab,
 744                          mb_array_size * sizeof(float), fail);
 745
 746     }
 747
 748     if (s->codec_id == AV_CODEC_ID_MPEG4 ||
 749         (s->avctx->flags & AV_CODEC_FLAG_INTERLACED_ME)) {
 750         /* interlaced direct mode decoding tables */
 751         for (i = 0; i < 2; i++) {
 752             int j, k;
 753             for (j = 0; j < 2; j++) {
 754                 for (k = 0; k < 2; k++) {
 755                     FF_ALLOCZ_OR_GOTO(s->avctx,
 756                                       s->b_field_mv_table_base[i][j][k],
 757                                       mv_table_size * 2 * sizeof(int16_t),
 758                                       fail);
 759                     s->b_field_mv_table[i][j][k] = s->b_field_mv_table_base[i][j][k] +
 760                                                    s->mb_stride + 1;
 761                 }
 762                 FF_ALLOCZ_OR_GOTO(s->avctx, s->b_field_select_table [i][j], mb_array_size * 2 * sizeof(uint8_t), fail)
 763                 FF_ALLOCZ_OR_GOTO(s->avctx, s->p_field_mv_table_base[i][j], mv_table_size * 2 * sizeof(int16_t), fail)
 764                 s->p_field_mv_table[i][j] = s->p_field_mv_table_base[i][j] + s->mb_stride + 1;
 765             }
 766             FF_ALLOCZ_OR_GOTO(s->avctx, s->p_field_select_table[i], mb_array_size * 2 * sizeof(uint8_t), fail)
 767         }
 768     }
 769     if (s->out_format == FMT_H263) {
 770         /* cbp values */
 771         FF_ALLOCZ_OR_GOTO(s->avctx, s->coded_block_base, y_size + (s->mb_height&1)*2*s->b8_stride, fail);
 772         s->coded_block = s->coded_block_base + s->b8_stride + 1;
 773
 774         /* cbp, ac_pred, pred_dir */
 775         FF_ALLOCZ_OR_GOTO(s->avctx, s->cbp_table     , mb_array_size * sizeof(uint8_t), fail);
 776         FF_ALLOCZ_OR_GOTO(s->avctx, s->pred_dir_table, mb_array_size * sizeof(uint8_t), fail);
 777     }
 778
 779     if (s->h263_pred || s->h263_plus || !s->encoding) {
 780         /* dc values */
 781         // MN: we need these for error resilience of intra-frames
 782         FF_ALLOCZ_OR_GOTO(s->avctx, s->dc_val_base, yc_size * sizeof(int16_t), fail);
 783         s->dc_val[0] = s->dc_val_base + s->b8_stride + 1;
 784         s->dc_val[1] = s->dc_val_base + y_size + s->mb_stride + 1;
 785         s->dc_val[2] = s->dc_val[1] + c_size;
 786         for (i = 0; i < yc_size; i++)
 787             s->dc_val_base[i] = 1024;
 788     }
 789
 790     /* which mb is an intra block */
 791     FF_ALLOCZ_OR_GOTO(s->avctx, s->mbintra_table, mb_array_size, fail);
 792     memset(s->mbintra_table, 1, mb_array_size);
 793
 794     /* init macroblock skip table */
 795     FF_ALLOCZ_OR_GOTO(s->avctx, s->mbskip_table, mb_array_size + 2, fail);
 796     // Note the + 1 is for a quicker MPEG-4 slice_end detection
 797
 798     return ff_mpeg_er_init(s);
 799 fail:
 800     return AVERROR(ENOMEM);
 801 }
 802
 803 static void clear_context(MpegEncContext *s)
 804 {
 805     int i, j, k;
 806
 807     memset(&s->next_picture, 0, sizeof(s->next_picture));
 808     memset(&s->last_picture, 0, sizeof(s->last_picture));
 809     memset(&s->current_picture, 0, sizeof(s->current_picture));
 810     memset(&s->new_picture, 0, sizeof(s->new_picture));
 811
 812     memset(s->thread_context, 0, sizeof(s->thread_context));
 813
 814     s->me.map = NULL;
 815     s->me.score_map = NULL;
 816     s->dct_error_sum = NULL;
 817     s->block = NULL;
 818     s->blocks = NULL;
 819     s->block32 = NULL;
 820     memset(s->pblocks, 0, sizeof(s->pblocks));
 821     s->ac_val_base = NULL;
 822     s->ac_val[0] =
 823     s->ac_val[1] =
 824     s->ac_val[2] =NULL;
 825     s->sc.edge_emu_buffer = NULL;
 826     s->me.scratchpad = NULL;
 827     s->me.temp =
 828     s->sc.rd_scratchpad =
 829     s->sc.b_scratchpad =
 830     s->sc.obmc_scratchpad = NULL;
 831
 832
 833     s->bitstream_buffer = NULL;
 834     s->allocated_bitstream_buffer_size = 0;
 835     s->picture          = NULL;
 836     s->mb_type          = NULL;
 837     s->p_mv_table_base  = NULL;
 838     s->b_forw_mv_table_base = NULL;
 839     s->b_back_mv_table_base = NULL;
 840     s->b_bidir_forw_mv_table_base = NULL;
 841     s->b_bidir_back_mv_table_base = NULL;
 842     s->b_direct_mv_table_base = NULL;
 843     s->p_mv_table            = NULL;
 844     s->b_forw_mv_table       = NULL;
 845     s->b_back_mv_table       = NULL;
 846     s->b_bidir_forw_mv_table = NULL;
 847     s->b_bidir_back_mv_table = NULL;
 848     s->b_direct_mv_table     = NULL;
 849     for (i = 0; i < 2; i++) {
 850         for (j = 0; j < 2; j++) {
 851             for (k = 0; k < 2; k++) {
 852                 s->b_field_mv_table_base[i][j][k] = NULL;
 853                 s->b_field_mv_table[i][j][k] = NULL;
 854             }
 855             s->b_field_select_table[i][j] = NULL;
 856             s->p_field_mv_table_base[i][j] = NULL;
 857             s->p_field_mv_table[i][j] = NULL;
 858         }
 859         s->p_field_select_table[i] = NULL;
 860     }
 861
 862     s->dc_val_base = NULL;
 863     s->coded_block_base = NULL;
 864     s->mbintra_table = NULL;
 865     s->cbp_table = NULL;
 866     s->pred_dir_table = NULL;
 867
 868     s->mbskip_table = NULL;
 869
 870     s->er.error_status_table = NULL;
 871     s->er.er_temp_buffer = NULL;
 872     s->mb_index2xy = NULL;
 873     s->lambda_table = NULL;
 874
 875     s->cplx_tab = NULL;
 876     s->bits_tab = NULL;
 877 }
 878
 879 /**
 880  * init common structure for both encoder and decoder.
 881  * this assumes that some variables like width/height are already set
 882  */
 883 av_cold int ff_mpv_common_init(MpegEncContext *s)
 884 {
 885     int i, ret;
 886     int nb_slices = (HAVE_THREADS &&
 887                      s->avctx->active_thread_type & FF_THREAD_SLICE) ?
 888                     s->avctx->thread_count : 1;
 889
 890     clear_context(s);
 891
 892     if (s->encoding && s->avctx->slices)
 893         nb_slices = s->avctx->slices;
 894
 895     if (s->codec_id == AV_CODEC_ID_MPEG2VIDEO && !s->progressive_sequence)
 896         s->mb_height = (s->height + 31) / 32 * 2;
 897     else
 898         s->mb_height = (s->height + 15) / 16;
 899
 900     if (s->avctx->pix_fmt == AV_PIX_FMT_NONE) {
 901         av_log(s->avctx, AV_LOG_ERROR,
 902                "decoding to AV_PIX_FMT_NONE is not supported.\n");
 903         return -1;
 904     }
 905
 906     if (nb_slices > MAX_THREADS || (nb_slices > s->mb_height && s->mb_height)) {
 907         int max_slices;
 908         if (s->mb_height)
 909             max_slices = FFMIN(MAX_THREADS, s->mb_height);
 910         else
 911             max_slices = MAX_THREADS;
 912         av_log(s->avctx, AV_LOG_WARNING, "too many threads/slices (%d),"
 913                " reducing to %d\n", nb_slices, max_slices);
 914         nb_slices = max_slices;
 915     }
 916
 917     if ((s->width || s->height) &&
 918         av_image_check_size(s->width, s->height, 0, s->avctx))
 919         return -1;
 920
 921     dct_init(s);
 922
 923     /* set chroma shifts */
 924     ret = av_pix_fmt_get_chroma_sub_sample(s->avctx->pix_fmt,
 925                                            &s->chroma_x_shift,
 926                                            &s->chroma_y_shift);
 927     if (ret)
 928         return ret;
 929
 930     FF_ALLOCZ_OR_GOTO(s->avctx, s->picture,
 931                       MAX_PICTURE_COUNT * sizeof(Picture), fail);
 932     for (i = 0; i < MAX_PICTURE_COUNT; i++) {
 933         s->picture[i].f = av_frame_alloc();
 934         if (!s->picture[i].f)
 935             goto fail;
 936     }
 937     s->next_picture.f = av_frame_alloc();
 938     if (!s->next_picture.f)
 939         goto fail;
 940     s->last_picture.f = av_frame_alloc();
 941     if (!s->last_picture.f)
 942         goto fail;
 943     s->current_picture.f = av_frame_alloc();
 944     if (!s->current_picture.f)
 945         goto fail;
 946     s->new_picture.f = av_frame_alloc();
 947     if (!s->new_picture.f)
 948         goto fail;
 949
 950     if (init_context_frame(s))
 951         goto fail;
 952
 953     s->parse_context.state = -1;
 954
 955     s->context_initialized = 1;
 956     memset(s->thread_context, 0, sizeof(s->thread_context));
 957     s->thread_context[0]   = s;
 958
 959 //     if (s->width && s->height) {
 960     if (nb_slices > 1) {
 961         for (i = 0; i < nb_slices; i++) {
 962             if (i) {
 963                 s->thread_context[i] = av_memdup(s, sizeof(MpegEncContext));
 964                 if (!s->thread_context[i])
 965                     goto fail;
 966             }
 967             if (init_duplicate_context(s->thread_context[i]) < 0)
 968                 goto fail;
 969             s->thread_context[i]->start_mb_y =
 970                 (s->mb_height * (i) + nb_slices / 2) / nb_slices;
 971             s->thread_context[i]->end_mb_y   =
 972                 (s->mb_height * (i + 1) + nb_slices / 2) / nb_slices;
 973         }
 974     } else {
 975         if (init_duplicate_context(s) < 0)
 976             goto fail;
 977         s->start_mb_y = 0;
 978         s->end_mb_y   = s->mb_height;
 979     }
 980     s->slice_context_count = nb_slices;
 981 //     }
 982
 983     return 0;
 984  fail:
 985     ff_mpv_common_end(s);
 986     return -1;
 987 }
 988
 989 /**
 990  * Frees and resets MpegEncContext fields depending on the resolution.
 991  * Is used during resolution changes to avoid a full reinitialization of the
 992  * codec.
 993  */
 994 static void free_context_frame(MpegEncContext *s)
 995 {
 996     int i, j, k;
 997
 998     av_freep(&s->mb_type);
 999     av_freep(&s->p_mv_table_base);
1000     av_freep(&s->b_forw_mv_table_base);
1001     av_freep(&s->b_back_mv_table_base);
1002     av_freep(&s->b_bidir_forw_mv_table_base);
1003     av_freep(&s->b_bidir_back_mv_table_base);
1004     av_freep(&s->b_direct_mv_table_base);
1005     s->p_mv_table            = NULL;
1006     s->b_forw_mv_table       = NULL;
1007     s->b_back_mv_table       = NULL;
1008     s->b_bidir_forw_mv_table = NULL;
1009     s->b_bidir_back_mv_table = NULL;
1010     s->b_direct_mv_table     = NULL;
1011     for (i = 0; i < 2; i++) {
1012         for (j = 0; j < 2; j++) {
1013             for (k = 0; k < 2; k++) {
1014                 av_freep(&s->b_field_mv_table_base[i][j][k]);
1015                 s->b_field_mv_table[i][j][k] = NULL;
1016             }
1017             av_freep(&s->b_field_select_table[i][j]);
1018             av_freep(&s->p_field_mv_table_base[i][j]);
1019             s->p_field_mv_table[i][j] = NULL;
1020         }
1021         av_freep(&s->p_field_select_table[i]);
1022     }
1023
1024     av_freep(&s->dc_val_base);
1025     av_freep(&s->coded_block_base);
1026     av_freep(&s->mbintra_table);
1027     av_freep(&s->cbp_table);
1028     av_freep(&s->pred_dir_table);
1029
1030     av_freep(&s->mbskip_table);
1031
1032     av_freep(&s->er.error_status_table);
1033     av_freep(&s->er.er_temp_buffer);
1034     av_freep(&s->mb_index2xy);
1035     av_freep(&s->lambda_table);
1036
1037     av_freep(&s->cplx_tab);
1038     av_freep(&s->bits_tab);
1039
1040     s->linesize = s->uvlinesize = 0;
1041 }
1042
1043 int ff_mpv_common_frame_size_change(MpegEncContext *s)
1044 {
1045     int i, err = 0;
1046
1047     if (!s->context_initialized)
1048         return AVERROR(EINVAL);
1049
1050     if (s->slice_context_count > 1) {
1051         for (i = 0; i < s->slice_context_count; i++) {
1052             free_duplicate_context(s->thread_context[i]);
1053         }
1054         for (i = 1; i < s->slice_context_count; i++) {
1055             av_freep(&s->thread_context[i]);
1056         }
1057     } else
1058         free_duplicate_context(s);
1059
1060     free_context_frame(s);
1061
1062     if (s->picture)
1063         for (i = 0; i < MAX_PICTURE_COUNT; i++) {
1064                 s->picture[i].needs_realloc = 1;
1065         }
1066
1067     s->last_picture_ptr         =
1068     s->next_picture_ptr         =
1069     s->current_picture_ptr      = NULL;
1070
1071     // init
1072     if (s->codec_id == AV_CODEC_ID_MPEG2VIDEO && !s->progressive_sequence)
1073         s->mb_height = (s->height + 31) / 32 * 2;
1074     else
1075         s->mb_height = (s->height + 15) / 16;
1076
1077     if ((s->width || s->height) &&
1078         (err = av_image_check_size(s->width, s->height, 0, s->avctx)) < 0)
1079         goto fail;
1080
1081     if ((err = init_context_frame(s)))
1082         goto fail;
1083
1084     memset(s->thread_context, 0, sizeof(s->thread_context));
1085     s->thread_context[0]   = s;
1086
1087     if (s->width && s->height) {
1088         int nb_slices = s->slice_context_count;
1089         if (nb_slices > 1) {
1090             for (i = 0; i < nb_slices; i++) {
1091                 if (i) {
1092                     s->thread_context[i] = av_memdup(s, sizeof(MpegEncContext));
1093                     if (!s->thread_context[i]) {
1094                         err = AVERROR(ENOMEM);
1095                         goto fail;
1096                     }
1097                 }
1098                 if ((err = init_duplicate_context(s->thread_context[i])) < 0)
1099                     goto fail;
1100                 s->thread_context[i]->start_mb_y =
1101                     (s->mb_height * (i) + nb_slices / 2) / nb_slices;
1102                 s->thread_context[i]->end_mb_y   =
1103                     (s->mb_height * (i + 1) + nb_slices / 2) / nb_slices;
1104             }
1105         } else {
1106             err = init_duplicate_context(s);
1107             if (err < 0)
1108                 goto fail;
1109             s->start_mb_y = 0;
1110             s->end_mb_y   = s->mb_height;
1111         }
1112         s->slice_context_count = nb_slices;
1113     }
1114
1115     return 0;
1116  fail:
1117     ff_mpv_common_end(s);
1118     return err;
1119 }
1120
1121 /* init common structure for both encoder and decoder */
1122 void ff_mpv_common_end(MpegEncContext *s)
1123 {
1124     int i;
1125
1126     if (!s)
1127         return ;
1128
1129     if (s->slice_context_count > 1) {
1130         for (i = 0; i < s->slice_context_count; i++) {
1131             free_duplicate_context(s->thread_context[i]);
1132         }
1133         for (i = 1; i < s->slice_context_count; i++) {
1134             av_freep(&s->thread_context[i]);
1135         }
1136         s->slice_context_count = 1;
1137     } else free_duplicate_context(s);
1138
1139     av_freep(&s->parse_context.buffer);
1140     s->parse_context.buffer_size = 0;
1141
1142     av_freep(&s->bitstream_buffer);
1143     s->allocated_bitstream_buffer_size = 0;
1144
1145     if (s->picture) {
1146         for (i = 0; i < MAX_PICTURE_COUNT; i++) {
1147             ff_free_picture_tables(&s->picture[i]);
1148             ff_mpeg_unref_picture(s->avctx, &s->picture[i]);
1149             av_frame_free(&s->picture[i].f);
1150         }
1151     }
1152     av_freep(&s->picture);
1153     ff_free_picture_tables(&s->last_picture);
1154     ff_mpeg_unref_picture(s->avctx, &s->last_picture);
1155     av_frame_free(&s->last_picture.f);
1156     ff_free_picture_tables(&s->current_picture);
1157     ff_mpeg_unref_picture(s->avctx, &s->current_picture);
1158     av_frame_free(&s->current_picture.f);
1159     ff_free_picture_tables(&s->next_picture);
1160     ff_mpeg_unref_picture(s->avctx, &s->next_picture);
1161     av_frame_free(&s->next_picture.f);
1162     ff_free_picture_tables(&s->new_picture);
1163     ff_mpeg_unref_picture(s->avctx, &s->new_picture);
1164     av_frame_free(&s->new_picture.f);
1165
1166     free_context_frame(s);
1167
1168     s->context_initialized      = 0;
1169     s->last_picture_ptr         =
1170     s->next_picture_ptr         =
1171     s->current_picture_ptr      = NULL;
1172     s->linesize = s->uvlinesize = 0;
1173 }
1174
1175
1176 static void gray_frame(AVFrame *frame)
1177 {
1178     int i, h_chroma_shift, v_chroma_shift;
1179
1180     av_pix_fmt_get_chroma_sub_sample(frame->format, &h_chroma_shift, &v_chroma_shift);
1181
1182     for(i=0; i<frame->height; i++)
1183         memset(frame->data[0] + frame->linesize[0]*i, 0x80, frame->width);
1184     for(i=0; i<AV_CEIL_RSHIFT(frame->height, v_chroma_shift); i++) {
1185         memset(frame->data[1] + frame->linesize[1]*i,
1186                0x80, AV_CEIL_RSHIFT(frame->width, h_chroma_shift));
1187         memset(frame->data[2] + frame->linesize[2]*i,
1188                0x80, AV_CEIL_RSHIFT(frame->width, h_chroma_shift));
1189     }
1190 }
1191
1192 /**
1193  * generic function called after decoding
1194  * the header and before a frame is decoded.
1195  */
1196 int ff_mpv_frame_start(MpegEncContext *s, AVCodecContext *avctx)
1197 {
1198     int i, ret;
1199     Picture *pic;
1200     s->mb_skipped = 0;
1201
1202     if (!ff_thread_can_start_frame(avctx)) {
1203         av_log(avctx, AV_LOG_ERROR, "Attempt to start a frame outside SETUP state\n");
1204         return -1;
1205     }
1206
1207     /* mark & release old frames */
1208     if (s->pict_type != AV_PICTURE_TYPE_B && s->last_picture_ptr &&
1209         s->last_picture_ptr != s->next_picture_ptr &&
1210         s->last_picture_ptr->f->buf[0]) {
1211         ff_mpeg_unref_picture(s->avctx, s->last_picture_ptr);
1212     }
1213
1214     /* release forgotten pictures */
1215     /* if (MPEG-124 / H.263) */
1216     for (i = 0; i < MAX_PICTURE_COUNT; i++) {
1217         if (&s->picture[i] != s->last_picture_ptr &&
1218             &s->picture[i] != s->next_picture_ptr &&
1219             s->picture[i].reference && !s->picture[i].needs_realloc) {
1220             ff_mpeg_unref_picture(s->avctx, &s->picture[i]);
1221         }
1222     }
1223
1224     ff_mpeg_unref_picture(s->avctx, &s->current_picture);
1225     ff_mpeg_unref_picture(s->avctx, &s->last_picture);
1226     ff_mpeg_unref_picture(s->avctx, &s->next_picture);
1227
1228     /* release non reference frames */
1229     for (i = 0; i < MAX_PICTURE_COUNT; i++) {
1230         if (!s->picture[i].reference)
1231             ff_mpeg_unref_picture(s->avctx, &s->picture[i]);
1232     }
1233
1234     if (s->current_picture_ptr && !s->current_picture_ptr->f->buf[0]) {
1235         // we already have an unused image
1236         // (maybe it was set before reading the header)
1237         pic = s->current_picture_ptr;
1238     } else {
1239         i   = ff_find_unused_picture(s->avctx, s->picture, 0);
1240         if (i < 0) {
1241             av_log(s->avctx, AV_LOG_ERROR, "no frame buffer available\n");
1242             return i;
1243         }
1244         pic = &s->picture[i];
1245     }
1246
1247     pic->reference = 0;
1248     if (!s->droppable) {
1249         if (s->pict_type != AV_PICTURE_TYPE_B)
1250             pic->reference = 3;
1251     }
1252
1253     pic->f->coded_picture_number = s->coded_picture_number++;
1254
1255     if (alloc_picture(s, pic, 0) < 0)
1256         return -1;
1257
1258     s->current_picture_ptr = pic;
1259     // FIXME use only the vars from current_pic
1260     s->current_picture_ptr->f->top_field_first = s->top_field_first;
1261     if (s->codec_id == AV_CODEC_ID_MPEG1VIDEO ||
1262         s->codec_id == AV_CODEC_ID_MPEG2VIDEO) {
1263         if (s->picture_structure != PICT_FRAME)
1264             s->current_picture_ptr->f->top_field_first =
1265                 (s->picture_structure == PICT_TOP_FIELD) == s->first_field;
1266     }
1267     s->current_picture_ptr->f->interlaced_frame = !s->progressive_frame &&
1268                                                  !s->progressive_sequence;
1269     s->current_picture_ptr->field_picture      =  s->picture_structure != PICT_FRAME;
1270
1271     s->current_picture_ptr->f->pict_type = s->pict_type;
1272     // if (s->avctx->flags && AV_CODEC_FLAG_QSCALE)
1273     //     s->current_picture_ptr->quality = s->new_picture_ptr->quality;
1274     s->current_picture_ptr->f->key_frame = s->pict_type == AV_PICTURE_TYPE_I;
1275
1276     if ((ret = ff_mpeg_ref_picture(s->avctx, &s->current_picture,
1277                                    s->current_picture_ptr)) < 0)
1278         return ret;
1279
1280     if (s->pict_type != AV_PICTURE_TYPE_B) {
1281         s->last_picture_ptr = s->next_picture_ptr;
1282         if (!s->droppable)
1283             s->next_picture_ptr = s->current_picture_ptr;
1284     }
1285     ff_dlog(s->avctx, "L%p N%p C%p L%p N%p C%p type:%d drop:%d\n",
1286             s->last_picture_ptr, s->next_picture_ptr,s->current_picture_ptr,
1287             s->last_picture_ptr    ? s->last_picture_ptr->f->data[0]    : NULL,
1288             s->next_picture_ptr    ? s->next_picture_ptr->f->data[0]    : NULL,
1289             s->current_picture_ptr ? s->current_picture_ptr->f->data[0] : NULL,
1290             s->pict_type, s->droppable);
1291
1292     if ((!s->last_picture_ptr || !s->last_picture_ptr->f->buf[0]) &&
1293         (s->pict_type != AV_PICTURE_TYPE_I)) {
1294         int h_chroma_shift, v_chroma_shift;
1295         av_pix_fmt_get_chroma_sub_sample(s->avctx->pix_fmt,
1296                                          &h_chroma_shift, &v_chroma_shift);
1297         if (s->pict_type == AV_PICTURE_TYPE_B && s->next_picture_ptr && s->next_picture_ptr->f->buf[0])
1298             av_log(avctx, AV_LOG_DEBUG,
1299                    "allocating dummy last picture for B frame\n");
1300         else if (s->pict_type != AV_PICTURE_TYPE_I)
1301             av_log(avctx, AV_LOG_ERROR,
1302                    "warning: first frame is no keyframe\n");
1303
1304         /* Allocate a dummy frame */
1305         i = ff_find_unused_picture(s->avctx, s->picture, 0);
1306         if (i < 0) {
1307             av_log(s->avctx, AV_LOG_ERROR, "no frame buffer available\n");
1308             return i;
1309         }
1310         s->last_picture_ptr = &s->picture[i];
1311
1312         s->last_picture_ptr->reference   = 3;
1313         s->last_picture_ptr->f->key_frame = 0;
1314         s->last_picture_ptr->f->pict_type = AV_PICTURE_TYPE_P;
1315
1316         if (alloc_picture(s, s->last_picture_ptr, 0) < 0) {
1317             s->last_picture_ptr = NULL;
1318             return -1;
1319         }
1320
1321         if (!avctx->hwaccel) {
1322             for(i=0; i<avctx->height; i++)
1323                 memset(s->last_picture_ptr->f->data[0] + s->last_picture_ptr->f->linesize[0]*i,
1324                        0x80, avctx->width);
1325             if (s->last_picture_ptr->f->data[2]) {
1326                 for(i=0; i<AV_CEIL_RSHIFT(avctx->height, v_chroma_shift); i++) {
1327                     memset(s->last_picture_ptr->f->data[1] + s->last_picture_ptr->f->linesize[1]*i,
1328                         0x80, AV_CEIL_RSHIFT(avctx->width, h_chroma_shift));
1329                     memset(s->last_picture_ptr->f->data[2] + s->last_picture_ptr->f->linesize[2]*i,
1330                         0x80, AV_CEIL_RSHIFT(avctx->width, h_chroma_shift));
1331                 }
1332             }
1333
1334             if(s->codec_id == AV_CODEC_ID_FLV1 || s->codec_id == AV_CODEC_ID_H263){
1335                 for(i=0; i<avctx->height; i++)
1336                 memset(s->last_picture_ptr->f->data[0] + s->last_picture_ptr->f->linesize[0]*i, 16, avctx->width);
1337             }
1338         }
1339
1340         ff_thread_report_progress(&s->last_picture_ptr->tf, INT_MAX, 0);
1341         ff_thread_report_progress(&s->last_picture_ptr->tf, INT_MAX, 1);
1342     }
1343     if ((!s->next_picture_ptr || !s->next_picture_ptr->f->buf[0]) &&
1344         s->pict_type == AV_PICTURE_TYPE_B) {
1345         /* Allocate a dummy frame */
1346         i = ff_find_unused_picture(s->avctx, s->picture, 0);
1347         if (i < 0) {
1348             av_log(s->avctx, AV_LOG_ERROR, "no frame buffer available\n");
1349             return i;
1350         }
1351         s->next_picture_ptr = &s->picture[i];
1352
1353         s->next_picture_ptr->reference   = 3;
1354         s->next_picture_ptr->f->key_frame = 0;
1355         s->next_picture_ptr->f->pict_type = AV_PICTURE_TYPE_P;
1356
1357         if (alloc_picture(s, s->next_picture_ptr, 0) < 0) {
1358             s->next_picture_ptr = NULL;
1359             return -1;
1360         }
1361         ff_thread_report_progress(&s->next_picture_ptr->tf, INT_MAX, 0);
1362         ff_thread_report_progress(&s->next_picture_ptr->tf, INT_MAX, 1);
1363     }
1364
1365 #if 0 // BUFREF-FIXME
1366     memset(s->last_picture.f->data, 0, sizeof(s->last_picture.f->data));
1367     memset(s->next_picture.f->data, 0, sizeof(s->next_picture.f->data));
1368 #endif
1369     if (s->last_picture_ptr) {
1370         if (s->last_picture_ptr->f->buf[0] &&
1371             (ret = ff_mpeg_ref_picture(s->avctx, &s->last_picture,
1372                                        s->last_picture_ptr)) < 0)
1373             return ret;
1374     }
1375     if (s->next_picture_ptr) {
1376         if (s->next_picture_ptr->f->buf[0] &&
1377             (ret = ff_mpeg_ref_picture(s->avctx, &s->next_picture,
1378                                        s->next_picture_ptr)) < 0)
1379             return ret;
1380     }
1381
1382     av_assert0(s->pict_type == AV_PICTURE_TYPE_I || (s->last_picture_ptr &&
1383                                                  s->last_picture_ptr->f->buf[0]));
1384
1385     if (s->picture_structure!= PICT_FRAME) {
1386         int i;
1387         for (i = 0; i < 4; i++) {
1388             if (s->picture_structure == PICT_BOTTOM_FIELD) {
1389                 s->current_picture.f->data[i] +=
1390                     s->current_picture.f->linesize[i];
1391             }
1392             s->current_picture.f->linesize[i] *= 2;
1393             s->last_picture.f->linesize[i]    *= 2;
1394             s->next_picture.f->linesize[i]    *= 2;
1395         }
1396     }
1397
1398     /* set dequantizer, we can't do it during init as
1399      * it might change for MPEG-4 and we can't do it in the header
1400      * decode as init is not called for MPEG-4 there yet */
1401     if (s->mpeg_quant || s->codec_id == AV_CODEC_ID_MPEG2VIDEO) {
1402         s->dct_unquantize_intra = s->dct_unquantize_mpeg2_intra;
1403         s->dct_unquantize_inter = s->dct_unquantize_mpeg2_inter;
1404     } else if (s->out_format == FMT_H263 || s->out_format == FMT_H261) {
1405         s->dct_unquantize_intra = s->dct_unquantize_h263_intra;
1406         s->dct_unquantize_inter = s->dct_unquantize_h263_inter;
1407     } else {
1408         s->dct_unquantize_intra = s->dct_unquantize_mpeg1_intra;
1409         s->dct_unquantize_inter = s->dct_unquantize_mpeg1_inter;
1410     }
1411
1412     if (s->avctx->debug & FF_DEBUG_NOMC) {
1413         gray_frame(s->current_picture_ptr->f);
1414     }
1415
1416     return 0;
1417 }
1418
1419 /* called after a frame has been decoded. */
1420 void ff_mpv_frame_end(MpegEncContext *s)
1421 {
1422     emms_c();
1423
1424     if (s->current_picture.reference)
1425         ff_thread_report_progress(&s->current_picture_ptr->tf, INT_MAX, 0);
1426 }
1427
1428 void ff_print_debug_info(MpegEncContext *s, Picture *p, AVFrame *pict)
1429 {
1430     ff_print_debug_info2(s->avctx, pict, s->mbskip_table, p->mb_type,
1431                          p->qscale_table, p->motion_val, &s->low_delay,
1432                          s->mb_width, s->mb_height, s->mb_stride, s->quarter_sample);
1433 }
1434
1435 int ff_mpv_export_qp_table(MpegEncContext *s, AVFrame *f, Picture *p, int qp_type)
1436 {
1437     AVBufferRef *ref = av_buffer_ref(p->qscale_table_buf);
1438     int offset = 2*s->mb_stride + 1;
1439     if(!ref)
1440         return AVERROR(ENOMEM);
1441     av_assert0(ref->size >= offset + s->mb_stride * ((f->height+15)/16));
1442     ref->size -= offset;
1443     ref->data += offset;
1444     return av_frame_set_qp_table(f, ref, s->mb_stride, qp_type);
1445 }
1446
1447 static inline int hpel_motion_lowres(MpegEncContext *s,
1448                                      uint8_t *dest, uint8_t *src,
1449                                      int field_based, int field_select,
1450                                      int src_x, int src_y,
1451                                      int width, int height, ptrdiff_t stride,
1452                                      int h_edge_pos, int v_edge_pos,
1453                                      int w, int h, h264_chroma_mc_func *pix_op,
1454                                      int motion_x, int motion_y)
1455 {
1456     const int lowres   = s->avctx->lowres;
1457     const int op_index = FFMIN(lowres, 3);
1458     const int s_mask   = (2 << lowres) - 1;
1459     int emu = 0;
1460     int sx, sy;
1461
1462     if (s->quarter_sample) {
1463         motion_x /= 2;
1464         motion_y /= 2;
1465     }
1466
1467     sx = motion_x & s_mask;
1468     sy = motion_y & s_mask;
1469     src_x += motion_x >> lowres + 1;
1470     src_y += motion_y >> lowres + 1;
1471
1472     src   += src_y * stride + src_x;
1473
1474     if ((unsigned)src_x > FFMAX( h_edge_pos - (!!sx) - w,                 0) ||
1475         (unsigned)src_y > FFMAX((v_edge_pos >> field_based) - (!!sy) - h, 0)) {
1476         s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, src,
1477                                  s->linesize, s->linesize,
1478                                  w + 1, (h + 1) << field_based,
1479                                  src_x, src_y   << field_based,
1480                                  h_edge_pos, v_edge_pos);
1481         src = s->sc.edge_emu_buffer;
1482         emu = 1;
1483     }
1484
1485     sx = (sx << 2) >> lowres;
1486     sy = (sy << 2) >> lowres;
1487     if (field_select)
1488         src += s->linesize;
1489     pix_op[op_index](dest, src, stride, h, sx, sy);
1490     return emu;
1491 }
1492
1493 /* apply one mpeg motion vector to the three components */
1494 static av_always_inline void mpeg_motion_lowres(MpegEncContext *s,
1495                                                 uint8_t *dest_y,
1496                                                 uint8_t *dest_cb,
1497                                                 uint8_t *dest_cr,
1498                                                 int field_based,
1499                                                 int bottom_field,
1500                                                 int field_select,
1501                                                 uint8_t **ref_picture,
1502                                                 h264_chroma_mc_func *pix_op,
1503                                                 int motion_x, int motion_y,
1504                                                 int h, int mb_y)
1505 {
1506     uint8_t *ptr_y, *ptr_cb, *ptr_cr;
1507     int mx, my, src_x, src_y, uvsrc_x, uvsrc_y, sx, sy, uvsx, uvsy;
1508     ptrdiff_t uvlinesize, linesize;
1509     const int lowres     = s->avctx->lowres;
1510     const int op_index   = FFMIN(lowres-1+s->chroma_x_shift, 3);
1511     const int block_s    = 8>>lowres;
1512     const int s_mask     = (2 << lowres) - 1;
1513     const int h_edge_pos = s->h_edge_pos >> lowres;
1514     const int v_edge_pos = s->v_edge_pos >> lowres;
1515     linesize   = s->current_picture.f->linesize[0] << field_based;
1516     uvlinesize = s->current_picture.f->linesize[1] << field_based;
1517
1518     // FIXME obviously not perfect but qpel will not work in lowres anyway
1519     if (s->quarter_sample) {
1520         motion_x /= 2;
1521         motion_y /= 2;
1522     }
1523
1524     if(field_based){
1525         motion_y += (bottom_field - field_select)*((1 << lowres)-1);
1526     }
1527
1528     sx = motion_x & s_mask;
1529     sy = motion_y & s_mask;
1530     src_x = s->mb_x * 2 * block_s + (motion_x >> lowres + 1);
1531     src_y = (mb_y * 2 * block_s >> field_based) + (motion_y >> lowres + 1);
1532
1533     if (s->out_format == FMT_H263) {
1534         uvsx    = ((motion_x >> 1) & s_mask) | (sx & 1);
1535         uvsy    = ((motion_y >> 1) & s_mask) | (sy & 1);
1536         uvsrc_x = src_x >> 1;
1537         uvsrc_y = src_y >> 1;
1538     } else if (s->out_format == FMT_H261) {
1539         // even chroma mv's are full pel in H261
1540         mx      = motion_x / 4;
1541         my      = motion_y / 4;
1542         uvsx    = (2 * mx) & s_mask;
1543         uvsy    = (2 * my) & s_mask;
1544         uvsrc_x = s->mb_x * block_s + (mx >> lowres);
1545         uvsrc_y =    mb_y * block_s + (my >> lowres);
1546     } else {
1547         if(s->chroma_y_shift){
1548             mx      = motion_x / 2;
1549             my      = motion_y / 2;
1550             uvsx    = mx & s_mask;
1551             uvsy    = my & s_mask;
1552             uvsrc_x = s->mb_x * block_s                 + (mx >> lowres + 1);
1553             uvsrc_y =   (mb_y * block_s >> field_based) + (my >> lowres + 1);
1554         } else {
1555             if(s->chroma_x_shift){
1556             //Chroma422
1557                 mx = motion_x / 2;
1558                 uvsx = mx & s_mask;
1559                 uvsy = motion_y & s_mask;
1560                 uvsrc_y = src_y;
1561                 uvsrc_x = s->mb_x*block_s               + (mx >> (lowres+1));
1562             } else {
1563             //Chroma444
1564                 uvsx = motion_x & s_mask;
1565                 uvsy = motion_y & s_mask;
1566                 uvsrc_x = src_x;
1567                 uvsrc_y = src_y;
1568             }
1569         }
1570     }
1571
1572     ptr_y  = ref_picture[0] + src_y   * linesize   + src_x;
1573     ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x;
1574     ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x;
1575
1576     if ((unsigned) src_x > FFMAX( h_edge_pos - (!!sx) - 2 * block_s,       0) || uvsrc_y<0 ||
1577         (unsigned) src_y > FFMAX((v_edge_pos >> field_based) - (!!sy) - h, 0)) {
1578         s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, ptr_y,
1579                                  linesize >> field_based, linesize >> field_based,
1580                                  17, 17 + field_based,
1581                                 src_x, src_y << field_based, h_edge_pos,
1582                                 v_edge_pos);
1583         ptr_y = s->sc.edge_emu_buffer;
1584         if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY)) {
1585             uint8_t *ubuf = s->sc.edge_emu_buffer + 18 * s->linesize;
1586             uint8_t *vbuf =ubuf + 10 * s->uvlinesize;
1587             if (s->workaround_bugs & FF_BUG_IEDGE)
1588                 vbuf -= s->uvlinesize;
1589             s->vdsp.emulated_edge_mc(ubuf,  ptr_cb,
1590                                      uvlinesize >> field_based, uvlinesize >> field_based,
1591                                      9, 9 + field_based,
1592                                     uvsrc_x, uvsrc_y << field_based,
1593                                     h_edge_pos >> 1, v_edge_pos >> 1);
1594             s->vdsp.emulated_edge_mc(vbuf,  ptr_cr,
1595                                      uvlinesize >> field_based,uvlinesize >> field_based,
1596                                      9, 9 + field_based,
1597                                     uvsrc_x, uvsrc_y << field_based,
1598                                     h_edge_pos >> 1, v_edge_pos >> 1);
1599             ptr_cb = ubuf;
1600             ptr_cr = vbuf;
1601         }
1602     }
1603
1604     // FIXME use this for field pix too instead of the obnoxious hack which changes picture.f->data
1605     if (bottom_field) {
1606         dest_y  += s->linesize;
1607         dest_cb += s->uvlinesize;
1608         dest_cr += s->uvlinesize;
1609     }
1610
1611     if (field_select) {
1612         ptr_y   += s->linesize;
1613         ptr_cb  += s->uvlinesize;
1614         ptr_cr  += s->uvlinesize;
1615     }
1616
1617     sx = (sx << 2) >> lowres;
1618     sy = (sy << 2) >> lowres;
1619     pix_op[lowres - 1](dest_y, ptr_y, linesize, h, sx, sy);
1620
1621     if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY)) {
1622         int hc = s->chroma_y_shift ? (h+1-bottom_field)>>1 : h;
1623         uvsx = (uvsx << 2) >> lowres;
1624         uvsy = (uvsy << 2) >> lowres;
1625         if (hc) {
1626             pix_op[op_index](dest_cb, ptr_cb, uvlinesize, hc, uvsx, uvsy);
1627             pix_op[op_index](dest_cr, ptr_cr, uvlinesize, hc, uvsx, uvsy);
1628         }
1629     }
1630     // FIXME h261 lowres loop filter
1631 }
1632
1633 static inline void chroma_4mv_motion_lowres(MpegEncContext *s,
1634                                             uint8_t *dest_cb, uint8_t *dest_cr,
1635                                             uint8_t **ref_picture,
1636                                             h264_chroma_mc_func * pix_op,
1637                                             int mx, int my)
1638 {
1639     const int lowres     = s->avctx->lowres;
1640     const int op_index   = FFMIN(lowres, 3);
1641     const int block_s    = 8 >> lowres;
1642     const int s_mask     = (2 << lowres) - 1;
1643     const int h_edge_pos = s->h_edge_pos >> lowres + 1;
1644     const int v_edge_pos = s->v_edge_pos >> lowres + 1;
1645     int emu = 0, src_x, src_y, sx, sy;
1646     ptrdiff_t offset;
1647     uint8_t *ptr;
1648
1649     if (s->quarter_sample) {
1650         mx /= 2;
1651         my /= 2;
1652     }
1653
1654     /* In case of 8X8, we construct a single chroma motion vector
1655        with a special rounding */
1656     mx = ff_h263_round_chroma(mx);
1657     my = ff_h263_round_chroma(my);
1658
1659     sx = mx & s_mask;
1660     sy = my & s_mask;
1661     src_x = s->mb_x * block_s + (mx >> lowres + 1);
1662     src_y = s->mb_y * block_s + (my >> lowres + 1);
1663
1664     offset = src_y * s->uvlinesize + src_x;
1665     ptr = ref_picture[1] + offset;
1666     if ((unsigned) src_x > FFMAX(h_edge_pos - (!!sx) - block_s, 0) ||
1667         (unsigned) src_y > FFMAX(v_edge_pos - (!!sy) - block_s, 0)) {
1668         s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, ptr,
1669                                  s->uvlinesize, s->uvlinesize,
1670                                  9, 9,
1671                                  src_x, src_y, h_edge_pos, v_edge_pos);
1672         ptr = s->sc.edge_emu_buffer;
1673         emu = 1;
1674     }
1675     sx = (sx << 2) >> lowres;
1676     sy = (sy << 2) >> lowres;
1677     pix_op[op_index](dest_cb, ptr, s->uvlinesize, block_s, sx, sy);
1678
1679     ptr = ref_picture[2] + offset;
1680     if (emu) {
1681         s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, ptr,
1682                                  s->uvlinesize, s->uvlinesize,
1683                                  9, 9,
1684                                  src_x, src_y, h_edge_pos, v_edge_pos);
1685         ptr = s->sc.edge_emu_buffer;
1686     }
1687     pix_op[op_index](dest_cr, ptr, s->uvlinesize, block_s, sx, sy);
1688 }
1689
1690 /**
1691  * motion compensation of a single macroblock
1692  * @param s context
1693  * @param dest_y luma destination pointer
1694  * @param dest_cb chroma cb/u destination pointer
1695  * @param dest_cr chroma cr/v destination pointer
1696  * @param dir direction (0->forward, 1->backward)
1697  * @param ref_picture array[3] of pointers to the 3 planes of the reference picture
1698  * @param pix_op halfpel motion compensation function (average or put normally)
1699  * the motion vectors are taken from s->mv and the MV type from s->mv_type
1700  */
1701 static inline void MPV_motion_lowres(MpegEncContext *s,
1702                                      uint8_t *dest_y, uint8_t *dest_cb,
1703                                      uint8_t *dest_cr,
1704                                      int dir, uint8_t **ref_picture,
1705                                      h264_chroma_mc_func *pix_op)
1706 {
1707     int mx, my;
1708     int mb_x, mb_y, i;
1709     const int lowres  = s->avctx->lowres;
1710     const int block_s = 8 >>lowres;
1711
1712     mb_x = s->mb_x;
1713     mb_y = s->mb_y;
1714
1715     switch (s->mv_type) {
1716     case MV_TYPE_16X16:
1717         mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
1718                            0, 0, 0,
1719                            ref_picture, pix_op,
1720                            s->mv[dir][0][0], s->mv[dir][0][1],
1721                            2 * block_s, mb_y);
1722         break;
1723     case MV_TYPE_8X8:
1724         mx = 0;
1725         my = 0;
1726         for (i = 0; i < 4; i++) {
1727             hpel_motion_lowres(s, dest_y + ((i & 1) + (i >> 1) *
1728                                s->linesize) * block_s,
1729                                ref_picture[0], 0, 0,
1730                                (2 * mb_x + (i & 1)) * block_s,
1731                                (2 * mb_y + (i >> 1)) * block_s,
1732                                s->width, s->height, s->linesize,
1733                                s->h_edge_pos >> lowres, s->v_edge_pos >> lowres,
1734                                block_s, block_s, pix_op,
1735                                s->mv[dir][i][0], s->mv[dir][i][1]);
1736
1737             mx += s->mv[dir][i][0];
1738             my += s->mv[dir][i][1];
1739         }
1740
1741         if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY))
1742             chroma_4mv_motion_lowres(s, dest_cb, dest_cr, ref_picture,
1743                                      pix_op, mx, my);
1744         break;
1745     case MV_TYPE_FIELD:
1746         if (s->picture_structure == PICT_FRAME) {
1747             /* top field */
1748             mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
1749                                1, 0, s->field_select[dir][0],
1750                                ref_picture, pix_op,
1751                                s->mv[dir][0][0], s->mv[dir][0][1],
1752                                block_s, mb_y);
1753             /* bottom field */
1754             mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
1755                                1, 1, s->field_select[dir][1],
1756                                ref_picture, pix_op,
1757                                s->mv[dir][1][0], s->mv[dir][1][1],
1758                                block_s, mb_y);
1759         } else {
1760             if (s->picture_structure != s->field_select[dir][0] + 1 &&
1761                 s->pict_type != AV_PICTURE_TYPE_B && !s->first_field) {
1762                 ref_picture = s->current_picture_ptr->f->data;
1763
1764             }
1765             mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
1766                                0, 0, s->field_select[dir][0],
1767                                ref_picture, pix_op,
1768                                s->mv[dir][0][0],
1769                                s->mv[dir][0][1], 2 * block_s, mb_y >> 1);
1770             }
1771         break;
1772     case MV_TYPE_16X8:
1773         for (i = 0; i < 2; i++) {
1774             uint8_t **ref2picture;
1775
1776             if (s->picture_structure == s->field_select[dir][i] + 1 ||
1777                 s->pict_type == AV_PICTURE_TYPE_B || s->first_field) {
1778                 ref2picture = ref_picture;
1779             } else {
1780                 ref2picture = s->current_picture_ptr->f->data;
1781             }
1782
1783             mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
1784                                0, 0, s->field_select[dir][i],
1785                                ref2picture, pix_op,
1786                                s->mv[dir][i][0], s->mv[dir][i][1] +
1787                                2 * block_s * i, block_s, mb_y >> 1);
1788
1789             dest_y  +=  2 * block_s *  s->linesize;
1790             dest_cb += (2 * block_s >> s->chroma_y_shift) * s->uvlinesize;
1791             dest_cr += (2 * block_s >> s->chroma_y_shift) * s->uvlinesize;
1792         }
1793         break;
1794     case MV_TYPE_DMV:
1795         if (s->picture_structure == PICT_FRAME) {
1796             for (i = 0; i < 2; i++) {
1797                 int j;
1798                 for (j = 0; j < 2; j++) {
1799                     mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
1800                                        1, j, j ^ i,
1801                                        ref_picture, pix_op,
1802                                        s->mv[dir][2 * i + j][0],
1803                                        s->mv[dir][2 * i + j][1],
1804                                        block_s, mb_y);
1805                 }
1806                 pix_op = s->h264chroma.avg_h264_chroma_pixels_tab;
1807             }
1808         } else {
1809             for (i = 0; i < 2; i++) {
1810                 mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
1811                                    0, 0, s->picture_structure != i + 1,
1812                                    ref_picture, pix_op,
1813                                    s->mv[dir][2 * i][0],s->mv[dir][2 * i][1],
1814                                    2 * block_s, mb_y >> 1);
1815
1816                 // after put we make avg of the same block
1817                 pix_op = s->h264chroma.avg_h264_chroma_pixels_tab;
1818
1819                 // opposite parity is always in the same
1820                 // frame if this is second field
1821                 if (!s->first_field) {
1822                     ref_picture = s->current_picture_ptr->f->data;
1823                 }
1824             }
1825         }
1826         break;
1827     default:
1828         av_assert2(0);
1829     }
1830 }
1831
1832 /**
1833  * find the lowest MB row referenced in the MVs
1834  */
1835 static int lowest_referenced_row(MpegEncContext *s, int dir)
1836 {
1837     int my_max = INT_MIN, my_min = INT_MAX, qpel_shift = !s->quarter_sample;
1838     int my, off, i, mvs;
1839
1840     if (s->picture_structure != PICT_FRAME || s->mcsel)
1841         goto unhandled;
1842
1843     switch (s->mv_type) {
1844         case MV_TYPE_16X16:
1845             mvs = 1;
1846             break;
1847         case MV_TYPE_16X8:
1848             mvs = 2;
1849             break;
1850         case MV_TYPE_8X8:
1851             mvs = 4;
1852             break;
1853         default:
1854             goto unhandled;
1855     }
1856
1857     for (i = 0; i < mvs; i++) {
1858         my = s->mv[dir][i][1];
1859         my_max = FFMAX(my_max, my);
1860         my_min = FFMIN(my_min, my);
1861     }
1862
1863     off = ((FFMAX(-my_min, my_max)<<qpel_shift) + 63) >> 6;
1864
1865     return av_clip(s->mb_y + off, 0, s->mb_height - 1);
1866 unhandled:
1867     return s->mb_height-1;
1868 }
1869
1870 /* put block[] to dest[] */
1871 static inline void put_dct(MpegEncContext *s,
1872                            int16_t *block, int i, uint8_t *dest, int line_size, int qscale)
1873 {
1874     s->dct_unquantize_intra(s, block, i, qscale);
1875     s->idsp.idct_put(dest, line_size, block);
1876 }
1877
1878 /* add block[] to dest[] */
1879 static inline void add_dct(MpegEncContext *s,
1880                            int16_t *block, int i, uint8_t *dest, int line_size)
1881 {
1882     if (s->block_last_index[i] >= 0) {
1883         s->idsp.idct_add(dest, line_size, block);
1884     }
1885 }
1886
1887 static inline void add_dequant_dct(MpegEncContext *s,
1888                            int16_t *block, int i, uint8_t *dest, int line_size, int qscale)
1889 {
1890     if (s->block_last_index[i] >= 0) {
1891         s->dct_unquantize_inter(s, block, i, qscale);
1892
1893         s->idsp.idct_add(dest, line_size, block);
1894     }
1895 }
1896
1897 /**
1898  * Clean dc, ac, coded_block for the current non-intra MB.
1899  */
1900 void ff_clean_intra_table_entries(MpegEncContext *s)
1901 {
1902     int wrap = s->b8_stride;
1903     int xy = s->block_index[0];
1904
1905     s->dc_val[0][xy           ] =
1906     s->dc_val[0][xy + 1       ] =
1907     s->dc_val[0][xy     + wrap] =
1908     s->dc_val[0][xy + 1 + wrap] = 1024;
1909     /* ac pred */
1910     memset(s->ac_val[0][xy       ], 0, 32 * sizeof(int16_t));
1911     memset(s->ac_val[0][xy + wrap], 0, 32 * sizeof(int16_t));
1912     if (s->msmpeg4_version>=3) {
1913         s->coded_block[xy           ] =
1914         s->coded_block[xy + 1       ] =
1915         s->coded_block[xy     + wrap] =
1916         s->coded_block[xy + 1 + wrap] = 0;
1917     }
1918     /* chroma */
1919     wrap = s->mb_stride;
1920     xy = s->mb_x + s->mb_y * wrap;
1921     s->dc_val[1][xy] =
1922     s->dc_val[2][xy] = 1024;
1923     /* ac pred */
1924     memset(s->ac_val[1][xy], 0, 16 * sizeof(int16_t));
1925     memset(s->ac_val[2][xy], 0, 16 * sizeof(int16_t));
1926
1927     s->mbintra_table[xy]= 0;
1928 }
1929
1930 /* generic function called after a macroblock has been parsed by the
1931    decoder or after it has been encoded by the encoder.
1932
1933    Important variables used:
1934    s->mb_intra : true if intra macroblock
1935    s->mv_dir   : motion vector direction
1936    s->mv_type  : motion vector type
1937    s->mv       : motion vector
1938    s->interlaced_dct : true if interlaced dct used (mpeg2)
1939  */
1940 static av_always_inline
1941 void mpv_reconstruct_mb_internal(MpegEncContext *s, int16_t block[12][64],
1942                             int lowres_flag, int is_mpeg12)
1943 {
1944     const int mb_xy = s->mb_y * s->mb_stride + s->mb_x;
1945
1946     if (CONFIG_XVMC &&
1947         s->avctx->hwaccel && s->avctx->hwaccel->decode_mb) {
1948         s->avctx->hwaccel->decode_mb(s);//xvmc uses pblocks
1949         return;
1950     }
1951
1952     if(s->avctx->debug&FF_DEBUG_DCT_COEFF) {
1953        /* print DCT coefficients */
1954        int i,j;
1955        av_log(s->avctx, AV_LOG_DEBUG, "DCT coeffs of MB at %dx%d:\n", s->mb_x, s->mb_y);
1956        for(i=0; i<6; i++){
1957            for(j=0; j<64; j++){
1958                av_log(s->avctx, AV_LOG_DEBUG, "%5d",
1959                       block[i][s->idsp.idct_permutation[j]]);
1960            }
1961            av_log(s->avctx, AV_LOG_DEBUG, "\n");
1962        }
1963     }
1964
1965     s->current_picture.qscale_table[mb_xy] = s->qscale;
1966
1967     /* update DC predictors for P macroblocks */
1968     if (!s->mb_intra) {
1969         if (!is_mpeg12 && (s->h263_pred || s->h263_aic)) {
1970             if(s->mbintra_table[mb_xy])
1971                 ff_clean_intra_table_entries(s);
1972         } else {
1973             s->last_dc[0] =
1974             s->last_dc[1] =
1975             s->last_dc[2] = 128 << s->intra_dc_precision;
1976         }
1977     }
1978     else if (!is_mpeg12 && (s->h263_pred || s->h263_aic))
1979         s->mbintra_table[mb_xy]=1;
1980
1981     if ((s->avctx->flags & AV_CODEC_FLAG_PSNR) || s->frame_skip_threshold || s->frame_skip_factor ||
1982         !(s->encoding && (s->intra_only || s->pict_type == AV_PICTURE_TYPE_B) &&
1983           s->avctx->mb_decision != FF_MB_DECISION_RD)) { // FIXME precalc
1984         uint8_t *dest_y, *dest_cb, *dest_cr;
1985         int dct_linesize, dct_offset;
1986         op_pixels_func (*op_pix)[4];
1987         qpel_mc_func (*op_qpix)[16];
1988         const int linesize   = s->current_picture.f->linesize[0]; //not s->linesize as this would be wrong for field pics
1989         const int uvlinesize = s->current_picture.f->linesize[1];
1990         const int readable= s->pict_type != AV_PICTURE_TYPE_B || s->encoding || s->avctx->draw_horiz_band || lowres_flag;
1991         const int block_size= lowres_flag ? 8>>s->avctx->lowres : 8;
1992
1993         /* avoid copy if macroblock skipped in last frame too */
1994         /* skip only during decoding as we might trash the buffers during encoding a bit */
1995         if(!s->encoding){
1996             uint8_t *mbskip_ptr = &s->mbskip_table[mb_xy];
1997
1998             if (s->mb_skipped) {
1999                 s->mb_skipped= 0;
2000                 av_assert2(s->pict_type!=AV_PICTURE_TYPE_I);
2001                 *mbskip_ptr = 1;
2002             } else if(!s->current_picture.reference) {
2003                 *mbskip_ptr = 1;
2004             } else{
2005                 *mbskip_ptr = 0; /* not skipped */
2006             }
2007         }
2008
2009         dct_linesize = linesize << s->interlaced_dct;
2010         dct_offset   = s->interlaced_dct ? linesize : linesize * block_size;
2011
2012         if(readable){
2013             dest_y=  s->dest[0];
2014             dest_cb= s->dest[1];
2015             dest_cr= s->dest[2];
2016         }else{
2017             dest_y = s->sc.b_scratchpad;
2018             dest_cb= s->sc.b_scratchpad+16*linesize;
2019             dest_cr= s->sc.b_scratchpad+32*linesize;
2020         }
2021
2022         if (!s->mb_intra) {
2023             /* motion handling */
2024             /* decoding or more than one mb_type (MC was already done otherwise) */
2025             if(!s->encoding){
2026
2027                 if(HAVE_THREADS && s->avctx->active_thread_type&FF_THREAD_FRAME) {
2028                     if (s->mv_dir & MV_DIR_FORWARD) {
2029                         ff_thread_await_progress(&s->last_picture_ptr->tf,
2030                                                  lowest_referenced_row(s, 0),
2031                                                  0);
2032                     }
2033                     if (s->mv_dir & MV_DIR_BACKWARD) {
2034                         ff_thread_await_progress(&s->next_picture_ptr->tf,
2035                                                  lowest_referenced_row(s, 1),
2036                                                  0);
2037                     }
2038                 }
2039
2040                 if(lowres_flag){
2041                     h264_chroma_mc_func *op_pix = s->h264chroma.put_h264_chroma_pixels_tab;
2042
2043                     if (s->mv_dir & MV_DIR_FORWARD) {
2044                         MPV_motion_lowres(s, dest_y, dest_cb, dest_cr, 0, s->last_picture.f->data, op_pix);
2045                         op_pix = s->h264chroma.avg_h264_chroma_pixels_tab;
2046                     }
2047                     if (s->mv_dir & MV_DIR_BACKWARD) {
2048                         MPV_motion_lowres(s, dest_y, dest_cb, dest_cr, 1, s->next_picture.f->data, op_pix);
2049                     }
2050                 }else{
2051                     op_qpix = s->me.qpel_put;
2052                     if ((!s->no_rounding) || s->pict_type==AV_PICTURE_TYPE_B){
2053                         op_pix = s->hdsp.put_pixels_tab;
2054                     }else{
2055                         op_pix = s->hdsp.put_no_rnd_pixels_tab;
2056                     }
2057                     if (s->mv_dir & MV_DIR_FORWARD) {
2058                         ff_mpv_motion(s, dest_y, dest_cb, dest_cr, 0, s->last_picture.f->data, op_pix, op_qpix);
2059                         op_pix = s->hdsp.avg_pixels_tab;
2060                         op_qpix= s->me.qpel_avg;
2061                     }
2062                     if (s->mv_dir & MV_DIR_BACKWARD) {
2063                         ff_mpv_motion(s, dest_y, dest_cb, dest_cr, 1, s->next_picture.f->data, op_pix, op_qpix);
2064                     }
2065                 }
2066             }
2067
2068             /* skip dequant / idct if we are really late ;) */
2069             if(s->avctx->skip_idct){
2070                 if(  (s->avctx->skip_idct >= AVDISCARD_NONREF && s->pict_type == AV_PICTURE_TYPE_B)
2071                    ||(s->avctx->skip_idct >= AVDISCARD_NONKEY && s->pict_type != AV_PICTURE_TYPE_I)
2072                    || s->avctx->skip_idct >= AVDISCARD_ALL)
2073                     goto skip_idct;
2074             }
2075
2076             /* add dct residue */
2077             if(s->encoding || !(   s->msmpeg4_version || s->codec_id==AV_CODEC_ID_MPEG1VIDEO || s->codec_id==AV_CODEC_ID_MPEG2VIDEO
2078                                 || (s->codec_id==AV_CODEC_ID_MPEG4 && !s->mpeg_quant))){
2079                 add_dequant_dct(s, block[0], 0, dest_y                          , dct_linesize, s->qscale);
2080                 add_dequant_dct(s, block[1], 1, dest_y              + block_size, dct_linesize, s->qscale);
2081                 add_dequant_dct(s, block[2], 2, dest_y + dct_offset             , dct_linesize, s->qscale);
2082                 add_dequant_dct(s, block[3], 3, dest_y + dct_offset + block_size, dct_linesize, s->qscale);
2083
2084                 if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY)) {
2085                     if (s->chroma_y_shift){
2086                         add_dequant_dct(s, block[4], 4, dest_cb, uvlinesize, s->chroma_qscale);
2087                         add_dequant_dct(s, block[5], 5, dest_cr, uvlinesize, s->chroma_qscale);
2088                     }else{
2089                         dct_linesize >>= 1;
2090                         dct_offset >>=1;
2091                         add_dequant_dct(s, block[4], 4, dest_cb,              dct_linesize, s->chroma_qscale);
2092                         add_dequant_dct(s, block[5], 5, dest_cr,              dct_linesize, s->chroma_qscale);
2093                         add_dequant_dct(s, block[6], 6, dest_cb + dct_offset, dct_linesize, s->chroma_qscale);
2094                         add_dequant_dct(s, block[7], 7, dest_cr + dct_offset, dct_linesize, s->chroma_qscale);
2095                     }
2096                 }
2097             } else if(is_mpeg12 || (s->codec_id != AV_CODEC_ID_WMV2)){
2098                 add_dct(s, block[0], 0, dest_y                          , dct_linesize);
2099                 add_dct(s, block[1], 1, dest_y              + block_size, dct_linesize);
2100                 add_dct(s, block[2], 2, dest_y + dct_offset             , dct_linesize);
2101                 add_dct(s, block[3], 3, dest_y + dct_offset + block_size, dct_linesize);
2102
2103                 if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY)) {
2104                     if(s->chroma_y_shift){//Chroma420
2105                         add_dct(s, block[4], 4, dest_cb, uvlinesize);
2106                         add_dct(s, block[5], 5, dest_cr, uvlinesize);
2107                     }else{
2108                         //chroma422
2109                         dct_linesize = uvlinesize << s->interlaced_dct;
2110                         dct_offset   = s->interlaced_dct ? uvlinesize : uvlinesize*block_size;
2111
2112                         add_dct(s, block[4], 4, dest_cb, dct_linesize);
2113                         add_dct(s, block[5], 5, dest_cr, dct_linesize);
2114                         add_dct(s, block[6], 6, dest_cb+dct_offset, dct_linesize);
2115                         add_dct(s, block[7], 7, dest_cr+dct_offset, dct_linesize);
2116                         if(!s->chroma_x_shift){//Chroma444
2117                             add_dct(s, block[8], 8, dest_cb+block_size, dct_linesize);
2118                             add_dct(s, block[9], 9, dest_cr+block_size, dct_linesize);
2119                             add_dct(s, block[10], 10, dest_cb+block_size+dct_offset, dct_linesize);
2120                             add_dct(s, block[11], 11, dest_cr+block_size+dct_offset, dct_linesize);
2121                         }
2122                     }
2123                 }//fi gray
2124             }
2125             else if (CONFIG_WMV2_DECODER || CONFIG_WMV2_ENCODER) {
2126                 ff_wmv2_add_mb(s, block, dest_y, dest_cb, dest_cr);
2127             }
2128         } else {
2129             /* Only MPEG-4 Simple Studio Profile is supported in > 8-bit mode.
2130                TODO: Integrate 10-bit properly into mpegvideo.c so that ER works properly */
2131             if (s->avctx->bits_per_raw_sample > 8){
2132                 const int act_block_size = block_size * 2;
2133                 s->idsp.idct_put(dest_y,                           dct_linesize, (int16_t*)(*s->block32)[0]);
2134                 s->idsp.idct_put(dest_y              + act_block_size, dct_linesize, (int16_t*)(*s->block32)[1]);
2135                 s->idsp.idct_put(dest_y + dct_offset,              dct_linesize, (int16_t*)(*s->block32)[2]);
2136                 s->idsp.idct_put(dest_y + dct_offset + act_block_size, dct_linesize, (int16_t*)(*s->block32)[3]);
2137
2138                 dct_linesize = uvlinesize << s->interlaced_dct;
2139                 dct_offset   = s->interlaced_dct ? uvlinesize : uvlinesize*block_size;
2140
2141                 s->idsp.idct_put(dest_cb,              dct_linesize, (int16_t*)(*s->block32)[4]);
2142                 s->idsp.idct_put(dest_cr,              dct_linesize, (int16_t*)(*s->block32)[5]);
2143                 s->idsp.idct_put(dest_cb + dct_offset, dct_linesize, (int16_t*)(*s->block32)[6]);
2144                 s->idsp.idct_put(dest_cr + dct_offset, dct_linesize, (int16_t*)(*s->block32)[7]);
2145                 if(!s->chroma_x_shift){//Chroma444
2146                     s->idsp.idct_put(dest_cb + act_block_size,              dct_linesize, (int16_t*)(*s->block32)[8]);
2147                     s->idsp.idct_put(dest_cr + act_block_size,              dct_linesize, (int16_t*)(*s->block32)[9]);
2148                     s->idsp.idct_put(dest_cb + act_block_size + dct_offset, dct_linesize, (int16_t*)(*s->block32)[10]);
2149                     s->idsp.idct_put(dest_cr + act_block_size + dct_offset, dct_linesize, (int16_t*)(*s->block32)[11]);
2150                 }
2151             }
2152             /* dct only in intra block */
2153             else if(s->encoding || !(s->codec_id==AV_CODEC_ID_MPEG1VIDEO || s->codec_id==AV_CODEC_ID_MPEG2VIDEO)){
2154                 put_dct(s, block[0], 0, dest_y                          , dct_linesize, s->qscale);
2155                 put_dct(s, block[1], 1, dest_y              + block_size, dct_linesize, s->qscale);
2156                 put_dct(s, block[2], 2, dest_y + dct_offset             , dct_linesize, s->qscale);
2157                 put_dct(s, block[3], 3, dest_y + dct_offset + block_size, dct_linesize, s->qscale);
2158
2159                 if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY)) {
2160                     if(s->chroma_y_shift){
2161                         put_dct(s, block[4], 4, dest_cb, uvlinesize, s->chroma_qscale);
2162                         put_dct(s, block[5], 5, dest_cr, uvlinesize, s->chroma_qscale);
2163                     }else{
2164                         dct_offset >>=1;
2165                         dct_linesize >>=1;
2166                         put_dct(s, block[4], 4, dest_cb,              dct_linesize, s->chroma_qscale);
2167                         put_dct(s, block[5], 5, dest_cr,              dct_linesize, s->chroma_qscale);
2168                         put_dct(s, block[6], 6, dest_cb + dct_offset, dct_linesize, s->chroma_qscale);
2169                         put_dct(s, block[7], 7, dest_cr + dct_offset, dct_linesize, s->chroma_qscale);
2170                     }
2171                 }
2172             }else{
2173                 s->idsp.idct_put(dest_y,                           dct_linesize, block[0]);
2174                 s->idsp.idct_put(dest_y              + block_size, dct_linesize, block[1]);
2175                 s->idsp.idct_put(dest_y + dct_offset,              dct_linesize, block[2]);
2176                 s->idsp.idct_put(dest_y + dct_offset + block_size, dct_linesize, block[3]);
2177
2178                 if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY)) {
2179                     if(s->chroma_y_shift){
2180                         s->idsp.idct_put(dest_cb, uvlinesize, block[4]);
2181                         s->idsp.idct_put(dest_cr, uvlinesize, block[5]);
2182                     }else{
2183
2184                         dct_linesize = uvlinesize << s->interlaced_dct;
2185                         dct_offset   = s->interlaced_dct ? uvlinesize : uvlinesize*block_size;
2186
2187                         s->idsp.idct_put(dest_cb,              dct_linesize, block[4]);
2188                         s->idsp.idct_put(dest_cr,              dct_linesize, block[5]);
2189                         s->idsp.idct_put(dest_cb + dct_offset, dct_linesize, block[6]);
2190                         s->idsp.idct_put(dest_cr + dct_offset, dct_linesize, block[7]);
2191                         if(!s->chroma_x_shift){//Chroma444
2192                             s->idsp.idct_put(dest_cb + block_size,              dct_linesize, block[8]);
2193                             s->idsp.idct_put(dest_cr + block_size,              dct_linesize, block[9]);
2194                             s->idsp.idct_put(dest_cb + block_size + dct_offset, dct_linesize, block[10]);
2195                             s->idsp.idct_put(dest_cr + block_size + dct_offset, dct_linesize, block[11]);
2196                         }
2197                     }
2198                 }//gray
2199             }
2200         }
2201 skip_idct:
2202         if(!readable){
2203             s->hdsp.put_pixels_tab[0][0](s->dest[0], dest_y ,   linesize,16);
2204             if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY)) {
2205                 s->hdsp.put_pixels_tab[s->chroma_x_shift][0](s->dest[1], dest_cb, uvlinesize,16 >> s->chroma_y_shift);
2206                 s->hdsp.put_pixels_tab[s->chroma_x_shift][0](s->dest[2], dest_cr, uvlinesize,16 >> s->chroma_y_shift);
2207             }
2208         }
2209     }
2210 }
2211
2212 void ff_mpv_reconstruct_mb(MpegEncContext *s, int16_t block[12][64])
2213 {
2214 #if !CONFIG_SMALL
2215     if(s->out_format == FMT_MPEG1) {
2216         if(s->avctx->lowres) mpv_reconstruct_mb_internal(s, block, 1, 1);
2217         else                 mpv_reconstruct_mb_internal(s, block, 0, 1);
2218     } else
2219 #endif
2220     if(s->avctx->lowres) mpv_reconstruct_mb_internal(s, block, 1, 0);
2221     else                  mpv_reconstruct_mb_internal(s, block, 0, 0);
2222 }
2223
2224 void ff_mpeg_draw_horiz_band(MpegEncContext *s, int y, int h)
2225 {
2226     ff_draw_horiz_band(s->avctx, s->current_picture_ptr->f,
2227                        s->last_picture_ptr ? s->last_picture_ptr->f : NULL, y, h, s->picture_structure,
2228                        s->first_field, s->low_delay);
2229 }
2230
2231 void ff_init_block_index(MpegEncContext *s){ //FIXME maybe rename
2232     const int linesize   = s->current_picture.f->linesize[0]; //not s->linesize as this would be wrong for field pics
2233     const int uvlinesize = s->current_picture.f->linesize[1];
2234     const int width_of_mb = (4 + (s->avctx->bits_per_raw_sample > 8)) - s->avctx->lowres;
2235     const int height_of_mb = 4 - s->avctx->lowres;
2236
2237     s->block_index[0]= s->b8_stride*(s->mb_y*2    ) - 2 + s->mb_x*2;
2238     s->block_index[1]= s->b8_stride*(s->mb_y*2    ) - 1 + s->mb_x*2;
2239     s->block_index[2]= s->b8_stride*(s->mb_y*2 + 1) - 2 + s->mb_x*2;
2240     s->block_index[3]= s->b8_stride*(s->mb_y*2 + 1) - 1 + s->mb_x*2;
2241     s->block_index[4]= s->mb_stride*(s->mb_y + 1)                + s->b8_stride*s->mb_height*2 + s->mb_x - 1;
2242     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;
2243     //block_index is not used by mpeg2, so it is not affected by chroma_format
2244
2245     s->dest[0] = s->current_picture.f->data[0] + (int)((s->mb_x - 1U) <<  width_of_mb);
2246     s->dest[1] = s->current_picture.f->data[1] + (int)((s->mb_x - 1U) << (width_of_mb - s->chroma_x_shift));
2247     s->dest[2] = s->current_picture.f->data[2] + (int)((s->mb_x - 1U) << (width_of_mb - s->chroma_x_shift));
2248
2249     if(!(s->pict_type==AV_PICTURE_TYPE_B && s->avctx->draw_horiz_band && s->picture_structure==PICT_FRAME))
2250     {
2251         if(s->picture_structure==PICT_FRAME){
2252         s->dest[0] += s->mb_y *   linesize << height_of_mb;
2253         s->dest[1] += s->mb_y * uvlinesize << (height_of_mb - s->chroma_y_shift);
2254         s->dest[2] += s->mb_y * uvlinesize << (height_of_mb - s->chroma_y_shift);
2255         }else{
2256             s->dest[0] += (s->mb_y>>1) *   linesize << height_of_mb;
2257             s->dest[1] += (s->mb_y>>1) * uvlinesize << (height_of_mb - s->chroma_y_shift);
2258             s->dest[2] += (s->mb_y>>1) * uvlinesize << (height_of_mb - s->chroma_y_shift);
2259             av_assert1((s->mb_y&1) == (s->picture_structure == PICT_BOTTOM_FIELD));
2260         }
2261     }
2262 }
2263
2264 void ff_mpeg_flush(AVCodecContext *avctx){
2265     int i;
2266     MpegEncContext *s = avctx->priv_data;
2267
2268     if (!s || !s->picture)
2269         return;
2270
2271     for (i = 0; i < MAX_PICTURE_COUNT; i++)
2272         ff_mpeg_unref_picture(s->avctx, &s->picture[i]);
2273     s->current_picture_ptr = s->last_picture_ptr = s->next_picture_ptr = NULL;
2274
2275     ff_mpeg_unref_picture(s->avctx, &s->current_picture);
2276     ff_mpeg_unref_picture(s->avctx, &s->last_picture);
2277     ff_mpeg_unref_picture(s->avctx, &s->next_picture);
2278
2279     s->mb_x= s->mb_y= 0;
2280     s->closed_gop= 0;
2281
2282     s->parse_context.state= -1;
2283     s->parse_context.frame_start_found= 0;
2284     s->parse_context.overread= 0;
2285     s->parse_context.overread_index= 0;
2286     s->parse_context.index= 0;
2287     s->parse_context.last_index= 0;
2288     s->bitstream_buffer_size=0;
2289     s->pp_time=0;
2290 }
2291
2292 /**
2293  * set qscale and update qscale dependent variables.
2294  */
2295 void ff_set_qscale(MpegEncContext * s, int qscale)
2296 {
2297     if (qscale < 1)
2298         qscale = 1;
2299     else if (qscale > 31)
2300         qscale = 31;
2301
2302     s->qscale = qscale;
2303     s->chroma_qscale= s->chroma_qscale_table[qscale];
2304
2305     s->y_dc_scale= s->y_dc_scale_table[ qscale ];
2306     s->c_dc_scale= s->c_dc_scale_table[ s->chroma_qscale ];
2307 }
2308
2309 void ff_mpv_report_decode_progress(MpegEncContext *s)
2310 {
2311     if (s->pict_type != AV_PICTURE_TYPE_B && !s->partitioned_frame && !s->er.error_occurred)
2312         ff_thread_report_progress(&s->current_picture_ptr->tf, s->mb_y, 0);
2313 }