git.sesse.net Git - ffmpeg/blob - libavcodec/h264.c

   1 /*
   2  * H.26L/H.264/AVC/JVT/14496-10/... decoder
   3  * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
   4  *
   5  * This file is part of FFmpeg.
   6  *
   7  * FFmpeg is free software; you can redistribute it and/or
   8  * modify it under the terms of the GNU Lesser General Public
   9  * License as published by the Free Software Foundation; either
  10  * version 2.1 of the License, or (at your option) any later version.
  11  *
  12  * FFmpeg is distributed in the hope that it will be useful,
  13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * Lesser General Public License for more details.
  16  *
  17  * You should have received a copy of the GNU Lesser General Public
  18  * License along with FFmpeg; if not, write to the Free Software
  19  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  20  */
  21
  22 /**
  23  * @file
  24  * H.264 / AVC / MPEG4 part10 codec.
  25  * @author Michael Niedermayer <michaelni@gmx.at>
  26  */
  27
  28 #define UNCHECKED_BITSTREAM_READER 1
  29
  30 #include "libavutil/imgutils.h"
  31 #include "libavutil/opt.h"
  32 #include "internal.h"
  33 #include "cabac.h"
  34 #include "cabac_functions.h"
  35 #include "dsputil.h"
  36 #include "avcodec.h"
  37 #include "mpegvideo.h"
  38 #include "h264.h"
  39 #include "h264data.h"
  40 #include "h264_mvpred.h"
  41 #include "golomb.h"
  42 #include "mathops.h"
  43 #include "rectangle.h"
  44 #include "thread.h"
  45 #include "vdpau_internal.h"
  46 #include "libavutil/avassert.h"
  47
  48 // #undef NDEBUG
  49 #include <assert.h>
  50
  51 const uint16_t ff_h264_mb_sizes[4] = { 256, 384, 512, 768 };
  52
  53 static const uint8_t rem6[QP_MAX_NUM + 1] = {
  54     0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2,
  55     3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5,
  56     0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2,
  57     3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5,
  58     0, 1, 2, 3,
  59 };
  60
  61 static const uint8_t div6[QP_MAX_NUM + 1] = {
  62     0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 3,  3,  3,
  63     3, 3, 3, 4, 4, 4, 4, 4, 4, 5, 5, 5, 5, 5, 5, 6, 6, 6, 6,  6,  6,
  64     7, 7, 7, 7, 7, 7, 8, 8, 8, 8, 8, 8, 9, 9, 9, 9, 9, 9, 10, 10, 10,
  65    10,10,10,11,11,11,11,11,11,12,12,12,12,12,12,13,13,13, 13, 13, 13,
  66    14,14,14,14,
  67 };
  68
  69 static const enum AVPixelFormat hwaccel_pixfmt_list_h264_jpeg_420[] = {
  70     AV_PIX_FMT_DXVA2_VLD,
  71     AV_PIX_FMT_VAAPI_VLD,
  72     AV_PIX_FMT_VDA_VLD,
  73     AV_PIX_FMT_YUVJ420P,
  74     AV_PIX_FMT_NONE
  75 };
  76
  77 int avpriv_h264_has_num_reorder_frames(AVCodecContext *avctx)
  78 {
  79     H264Context *h = avctx->priv_data;
  80     return h ? h->sps.num_reorder_frames : 0;
  81 }
  82
  83 /**
  84  * Check if the top & left blocks are available if needed and
  85  * change the dc mode so it only uses the available blocks.
  86  */
  87 int ff_h264_check_intra4x4_pred_mode(H264Context *h)
  88 {
  89     MpegEncContext *const s     = &h->s;
  90     static const int8_t top[12] = {
  91         -1, 0, LEFT_DC_PRED, -1, -1, -1, -1, -1, 0
  92     };
  93     static const int8_t left[12] = {
  94         0, -1, TOP_DC_PRED, 0, -1, -1, -1, 0, -1, DC_128_PRED
  95     };
  96     int i;
  97
  98     if (!(h->top_samples_available & 0x8000)) {
  99         for (i = 0; i < 4; i++) {
 100             int status = top[h->intra4x4_pred_mode_cache[scan8[0] + i]];
 101             if (status < 0) {
 102                 av_log(h->s.avctx, AV_LOG_ERROR,
 103                        "top block unavailable for requested intra4x4 mode %d at %d %d\n",
 104                        status, s->mb_x, s->mb_y);
 105                 return -1;
 106             } else if (status) {
 107                 h->intra4x4_pred_mode_cache[scan8[0] + i] = status;
 108             }
 109         }
 110     }
 111
 112     if ((h->left_samples_available & 0x8888) != 0x8888) {
 113         static const int mask[4] = { 0x8000, 0x2000, 0x80, 0x20 };
 114         for (i = 0; i < 4; i++)
 115             if (!(h->left_samples_available & mask[i])) {
 116                 int status = left[h->intra4x4_pred_mode_cache[scan8[0] + 8 * i]];
 117                 if (status < 0) {
 118                     av_log(h->s.avctx, AV_LOG_ERROR,
 119                            "left block unavailable for requested intra4x4 mode %d at %d %d\n",
 120                            status, s->mb_x, s->mb_y);
 121                     return -1;
 122                 } else if (status) {
 123                     h->intra4x4_pred_mode_cache[scan8[0] + 8 * i] = status;
 124                 }
 125             }
 126     }
 127
 128     return 0;
 129 } // FIXME cleanup like ff_h264_check_intra_pred_mode
 130
 131 /**
 132  * Check if the top & left blocks are available if needed and
 133  * change the dc mode so it only uses the available blocks.
 134  */
 135 int ff_h264_check_intra_pred_mode(H264Context *h, int mode, int is_chroma)
 136 {
 137     MpegEncContext *const s     = &h->s;
 138     static const int8_t top[7]  = { LEFT_DC_PRED8x8, 1, -1, -1 };
 139     static const int8_t left[7] = { TOP_DC_PRED8x8, -1, 2, -1, DC_128_PRED8x8 };
 140
 141     if (mode > 6U) {
 142         av_log(h->s.avctx, AV_LOG_ERROR,
 143                "out of range intra chroma pred mode at %d %d\n",
 144                s->mb_x, s->mb_y);
 145         return -1;
 146     }
 147
 148     if (!(h->top_samples_available & 0x8000)) {
 149         mode = top[mode];
 150         if (mode < 0) {
 151             av_log(h->s.avctx, AV_LOG_ERROR,
 152                    "top block unavailable for requested intra mode at %d %d\n",
 153                    s->mb_x, s->mb_y);
 154             return -1;
 155         }
 156     }
 157
 158     if ((h->left_samples_available & 0x8080) != 0x8080) {
 159         mode = left[mode];
 160         if (is_chroma && (h->left_samples_available & 0x8080)) {
 161             // mad cow disease mode, aka MBAFF + constrained_intra_pred
 162             mode = ALZHEIMER_DC_L0T_PRED8x8 +
 163                    (!(h->left_samples_available & 0x8000)) +
 164                    2 * (mode == DC_128_PRED8x8);
 165         }
 166         if (mode < 0) {
 167             av_log(h->s.avctx, AV_LOG_ERROR,
 168                    "left block unavailable for requested intra mode at %d %d\n",
 169                    s->mb_x, s->mb_y);
 170             return -1;
 171         }
 172     }
 173
 174     return mode;
 175 }
 176
 177 const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src,
 178                                   int *dst_length, int *consumed, int length)
 179 {
 180     int i, si, di;
 181     uint8_t *dst;
 182     int bufidx;
 183
 184     // src[0]&0x80; // forbidden bit
 185     h->nal_ref_idc   = src[0] >> 5;
 186     h->nal_unit_type = src[0] & 0x1F;
 187
 188     src++;
 189     length--;
 190
 191 #define STARTCODE_TEST                                                  \
 192         if (i + 2 < length && src[i + 1] == 0 && src[i + 2] <= 3) {     \
 193             if (src[i + 2] != 3) {                                      \
 194                 /* startcode, so we must be past the end */             \
 195                 length = i;                                             \
 196             }                                                           \
 197             break;                                                      \
 198         }
 199 #if HAVE_FAST_UNALIGNED
 200 #define FIND_FIRST_ZERO                                                 \
 201         if (i > 0 && !src[i])                                           \
 202             i--;                                                        \
 203         while (src[i])                                                  \
 204             i++
 205 #if HAVE_FAST_64BIT
 206     for (i = 0; i + 1 < length; i += 9) {
 207         if (!((~AV_RN64A(src + i) &
 208                (AV_RN64A(src + i) - 0x0100010001000101ULL)) &
 209               0x8000800080008080ULL))
 210             continue;
 211         FIND_FIRST_ZERO;
 212         STARTCODE_TEST;
 213         i -= 7;
 214     }
 215 #else
 216     for (i = 0; i + 1 < length; i += 5) {
 217         if (!((~AV_RN32A(src + i) &
 218                (AV_RN32A(src + i) - 0x01000101U)) &
 219               0x80008080U))
 220             continue;
 221         FIND_FIRST_ZERO;
 222         STARTCODE_TEST;
 223         i -= 3;
 224     }
 225 #endif
 226 #else
 227     for (i = 0; i + 1 < length; i += 2) {
 228         if (src[i])
 229             continue;
 230         if (i > 0 && src[i - 1] == 0)
 231             i--;
 232         STARTCODE_TEST;
 233     }
 234 #endif
 235
 236     // use second escape buffer for inter data
 237     bufidx = h->nal_unit_type == NAL_DPC ? 1 : 0;
 238
 239     si = h->rbsp_buffer_size[bufidx];
 240     av_fast_padded_malloc(&h->rbsp_buffer[bufidx], &h->rbsp_buffer_size[bufidx], length+MAX_MBPAIR_SIZE);
 241     dst = h->rbsp_buffer[bufidx];
 242
 243     if (dst == NULL)
 244         return NULL;
 245
 246     if(i>=length-1){ //no escaped 0
 247         *dst_length= length;
 248         *consumed= length+1; //+1 for the header
 249         if(h->s.avctx->flags2 & CODEC_FLAG2_FAST){
 250             return src;
 251         }else{
 252             memcpy(dst, src, length);
 253             return dst;
 254         }
 255     }
 256
 257     memcpy(dst, src, i);
 258     si = di = i;
 259     while (si + 2 < length) {
 260         // remove escapes (very rare 1:2^22)
 261         if (src[si + 2] > 3) {
 262             dst[di++] = src[si++];
 263             dst[di++] = src[si++];
 264         } else if (src[si] == 0 && src[si + 1] == 0) {
 265             if (src[si + 2] == 3) { // escape
 266                 dst[di++]  = 0;
 267                 dst[di++]  = 0;
 268                 si        += 3;
 269                 continue;
 270             } else // next start code
 271                 goto nsc;
 272         }
 273
 274         dst[di++] = src[si++];
 275     }
 276     while (si < length)
 277         dst[di++] = src[si++];
 278 nsc:
 279
 280     memset(dst + di, 0, FF_INPUT_BUFFER_PADDING_SIZE);
 281
 282     *dst_length = di;
 283     *consumed   = si + 1; // +1 for the header
 284     /* FIXME store exact number of bits in the getbitcontext
 285      * (it is needed for decoding) */
 286     return dst;
 287 }
 288
 289 /**
 290  * Identify the exact end of the bitstream
 291  * @return the length of the trailing, or 0 if damaged
 292  */
 293 static int decode_rbsp_trailing(H264Context *h, const uint8_t *src)
 294 {
 295     int v = *src;
 296     int r;
 297
 298     tprintf(h->s.avctx, "rbsp trailing %X\n", v);
 299
 300     for (r = 1; r < 9; r++) {
 301         if (v & 1)
 302             return r;
 303         v >>= 1;
 304     }
 305     return 0;
 306 }
 307
 308 static inline int get_lowest_part_list_y(H264Context *h, Picture *pic, int n,
 309                                          int height, int y_offset, int list)
 310 {
 311     int raw_my        = h->mv_cache[list][scan8[n]][1];
 312     int filter_height = (raw_my & 3) ? 2 : 0;
 313     int full_my       = (raw_my >> 2) + y_offset;
 314     int top           = full_my - filter_height;
 315     int bottom        = full_my + filter_height + height;
 316
 317     return FFMAX(abs(top), bottom);
 318 }
 319
 320 static inline void get_lowest_part_y(H264Context *h, int refs[2][48], int n,
 321                                      int height, int y_offset, int list0,
 322                                      int list1, int *nrefs)
 323 {
 324     MpegEncContext *const s = &h->s;
 325     int my;
 326
 327     y_offset += 16 * (s->mb_y >> MB_FIELD);
 328
 329     if (list0) {
 330         int ref_n    = h->ref_cache[0][scan8[n]];
 331         Picture *ref = &h->ref_list[0][ref_n];
 332
 333         // Error resilience puts the current picture in the ref list.
 334         // Don't try to wait on these as it will cause a deadlock.
 335         // Fields can wait on each other, though.
 336         if (ref->f.thread_opaque   != s->current_picture.f.thread_opaque ||
 337             (ref->f.reference & 3) != s->picture_structure) {
 338             my = get_lowest_part_list_y(h, ref, n, height, y_offset, 0);
 339             if (refs[0][ref_n] < 0)
 340                 nrefs[0] += 1;
 341             refs[0][ref_n] = FFMAX(refs[0][ref_n], my);
 342         }
 343     }
 344
 345     if (list1) {
 346         int ref_n    = h->ref_cache[1][scan8[n]];
 347         Picture *ref = &h->ref_list[1][ref_n];
 348
 349         if (ref->f.thread_opaque   != s->current_picture.f.thread_opaque ||
 350             (ref->f.reference & 3) != s->picture_structure) {
 351             my = get_lowest_part_list_y(h, ref, n, height, y_offset, 1);
 352             if (refs[1][ref_n] < 0)
 353                 nrefs[1] += 1;
 354             refs[1][ref_n] = FFMAX(refs[1][ref_n], my);
 355         }
 356     }
 357 }
 358
 359 /**
 360  * Wait until all reference frames are available for MC operations.
 361  *
 362  * @param h the H264 context
 363  */
 364 static void await_references(H264Context *h)
 365 {
 366     MpegEncContext *const s = &h->s;
 367     const int mb_xy   = h->mb_xy;
 368     const int mb_type = s->current_picture.f.mb_type[mb_xy];
 369     int refs[2][48];
 370     int nrefs[2] = { 0 };
 371     int ref, list;
 372
 373     memset(refs, -1, sizeof(refs));
 374
 375     if (IS_16X16(mb_type)) {
 376         get_lowest_part_y(h, refs, 0, 16, 0,
 377                           IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs);
 378     } else if (IS_16X8(mb_type)) {
 379         get_lowest_part_y(h, refs, 0, 8, 0,
 380                           IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs);
 381         get_lowest_part_y(h, refs, 8, 8, 8,
 382                           IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1), nrefs);
 383     } else if (IS_8X16(mb_type)) {
 384         get_lowest_part_y(h, refs, 0, 16, 0,
 385                           IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1), nrefs);
 386         get_lowest_part_y(h, refs, 4, 16, 0,
 387                           IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1), nrefs);
 388     } else {
 389         int i;
 390
 391         av_assert2(IS_8X8(mb_type));
 392
 393         for (i = 0; i < 4; i++) {
 394             const int sub_mb_type = h->sub_mb_type[i];
 395             const int n           = 4 * i;
 396             int y_offset          = (i & 2) << 2;
 397
 398             if (IS_SUB_8X8(sub_mb_type)) {
 399                 get_lowest_part_y(h, refs, n, 8, y_offset,
 400                                   IS_DIR(sub_mb_type, 0, 0),
 401                                   IS_DIR(sub_mb_type, 0, 1),
 402                                   nrefs);
 403             } else if (IS_SUB_8X4(sub_mb_type)) {
 404                 get_lowest_part_y(h, refs, n, 4, y_offset,
 405                                   IS_DIR(sub_mb_type, 0, 0),
 406                                   IS_DIR(sub_mb_type, 0, 1),
 407                                   nrefs);
 408                 get_lowest_part_y(h, refs, n + 2, 4, y_offset + 4,
 409                                   IS_DIR(sub_mb_type, 0, 0),
 410                                   IS_DIR(sub_mb_type, 0, 1),
 411                                   nrefs);
 412             } else if (IS_SUB_4X8(sub_mb_type)) {
 413                 get_lowest_part_y(h, refs, n, 8, y_offset,
 414                                   IS_DIR(sub_mb_type, 0, 0),
 415                                   IS_DIR(sub_mb_type, 0, 1),
 416                                   nrefs);
 417                 get_lowest_part_y(h, refs, n + 1, 8, y_offset,
 418                                   IS_DIR(sub_mb_type, 0, 0),
 419                                   IS_DIR(sub_mb_type, 0, 1),
 420                                   nrefs);
 421             } else {
 422                 int j;
 423                 av_assert2(IS_SUB_4X4(sub_mb_type));
 424                 for (j = 0; j < 4; j++) {
 425                     int sub_y_offset = y_offset + 2 * (j & 2);
 426                     get_lowest_part_y(h, refs, n + j, 4, sub_y_offset,
 427                                       IS_DIR(sub_mb_type, 0, 0),
 428                                       IS_DIR(sub_mb_type, 0, 1),
 429                                       nrefs);
 430                 }
 431             }
 432         }
 433     }
 434
 435     for (list = h->list_count - 1; list >= 0; list--)
 436         for (ref = 0; ref < 48 && nrefs[list]; ref++) {
 437             int row = refs[list][ref];
 438             if (row >= 0) {
 439                 Picture *ref_pic      = &h->ref_list[list][ref];
 440                 int ref_field         = ref_pic->f.reference - 1;
 441                 int ref_field_picture = ref_pic->field_picture;
 442                 int pic_height        = 16 * s->mb_height >> ref_field_picture;
 443
 444                 row <<= MB_MBAFF;
 445                 nrefs[list]--;
 446
 447                 if (!FIELD_PICTURE && ref_field_picture) { // frame referencing two fields
 448                     ff_thread_await_progress(&ref_pic->f,
 449                                              FFMIN((row >> 1) - !(row & 1),
 450                                                    pic_height - 1),
 451                                              1);
 452                     ff_thread_await_progress(&ref_pic->f,
 453                                              FFMIN((row >> 1), pic_height - 1),
 454                                              0);
 455                 } else if (FIELD_PICTURE && !ref_field_picture) { // field referencing one field of a frame
 456                     ff_thread_await_progress(&ref_pic->f,
 457                                              FFMIN(row * 2 + ref_field,
 458                                                    pic_height - 1),
 459                                              0);
 460                 } else if (FIELD_PICTURE) {
 461                     ff_thread_await_progress(&ref_pic->f,
 462                                              FFMIN(row, pic_height - 1),
 463                                              ref_field);
 464                 } else {
 465                     ff_thread_await_progress(&ref_pic->f,
 466                                              FFMIN(row, pic_height - 1),
 467                                              0);
 468                 }
 469             }
 470         }
 471 }
 472
 473 static av_always_inline void mc_dir_part(H264Context *h, Picture *pic,
 474                                          int n, int square, int height,
 475                                          int delta, int list,
 476                                          uint8_t *dest_y, uint8_t *dest_cb,
 477                                          uint8_t *dest_cr,
 478                                          int src_x_offset, int src_y_offset,
 479                                          qpel_mc_func *qpix_op,
 480                                          h264_chroma_mc_func chroma_op,
 481                                          int pixel_shift, int chroma_idc)
 482 {
 483     MpegEncContext *const s = &h->s;
 484     const int mx      = h->mv_cache[list][scan8[n]][0] + src_x_offset * 8;
 485     int my            = h->mv_cache[list][scan8[n]][1] + src_y_offset * 8;
 486     const int luma_xy = (mx & 3) + ((my & 3) << 2);
 487     int offset        = ((mx >> 2) << pixel_shift) + (my >> 2) * h->mb_linesize;
 488     uint8_t *src_y    = pic->f.data[0] + offset;
 489     uint8_t *src_cb, *src_cr;
 490     int extra_width  = h->emu_edge_width;
 491     int extra_height = h->emu_edge_height;
 492     int emu = 0;
 493     const int full_mx    = mx >> 2;
 494     const int full_my    = my >> 2;
 495     const int pic_width  = 16 * s->mb_width;
 496     const int pic_height = 16 * s->mb_height >> MB_FIELD;
 497     int ysh;
 498
 499     if (mx & 7)
 500         extra_width -= 3;
 501     if (my & 7)
 502         extra_height -= 3;
 503
 504     if (full_mx                <          0 - extra_width  ||
 505         full_my                <          0 - extra_height ||
 506         full_mx + 16 /*FIXME*/ > pic_width  + extra_width  ||
 507         full_my + 16 /*FIXME*/ > pic_height + extra_height) {
 508         s->dsp.emulated_edge_mc(s->edge_emu_buffer,
 509                                 src_y - (2 << pixel_shift) - 2 * h->mb_linesize,
 510                                 h->mb_linesize,
 511                                 16 + 5, 16 + 5 /*FIXME*/, full_mx - 2,
 512                                 full_my - 2, pic_width, pic_height);
 513         src_y = s->edge_emu_buffer + (2 << pixel_shift) + 2 * h->mb_linesize;
 514         emu   = 1;
 515     }
 516
 517     qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); // FIXME try variable height perhaps?
 518     if (!square)
 519         qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
 520
 521     if (CONFIG_GRAY && s->flags & CODEC_FLAG_GRAY)
 522         return;
 523
 524     if (chroma_idc == 3 /* yuv444 */) {
 525         src_cb = pic->f.data[1] + offset;
 526         if (emu) {
 527             s->dsp.emulated_edge_mc(s->edge_emu_buffer,
 528                                     src_cb - (2 << pixel_shift) - 2 * h->mb_linesize,
 529                                     h->mb_linesize,
 530                                     16 + 5, 16 + 5 /*FIXME*/,
 531                                     full_mx - 2, full_my - 2,
 532                                     pic_width, pic_height);
 533             src_cb = s->edge_emu_buffer + (2 << pixel_shift) + 2 * h->mb_linesize;
 534         }
 535         qpix_op[luma_xy](dest_cb, src_cb, h->mb_linesize); // FIXME try variable height perhaps?
 536         if (!square)
 537             qpix_op[luma_xy](dest_cb + delta, src_cb + delta, h->mb_linesize);
 538
 539         src_cr = pic->f.data[2] + offset;
 540         if (emu) {
 541             s->dsp.emulated_edge_mc(s->edge_emu_buffer,
 542                                     src_cr - (2 << pixel_shift) - 2 * h->mb_linesize,
 543                                     h->mb_linesize,
 544                                     16 + 5, 16 + 5 /*FIXME*/,
 545                                     full_mx - 2, full_my - 2,
 546                                     pic_width, pic_height);
 547             src_cr = s->edge_emu_buffer + (2 << pixel_shift) + 2 * h->mb_linesize;
 548         }
 549         qpix_op[luma_xy](dest_cr, src_cr, h->mb_linesize); // FIXME try variable height perhaps?
 550         if (!square)
 551             qpix_op[luma_xy](dest_cr + delta, src_cr + delta, h->mb_linesize);
 552         return;
 553     }
 554
 555     ysh = 3 - (chroma_idc == 2 /* yuv422 */);
 556     if (chroma_idc == 1 /* yuv420 */ && MB_FIELD) {
 557         // chroma offset when predicting from a field of opposite parity
 558         my  += 2 * ((s->mb_y & 1) - (pic->f.reference - 1));
 559         emu |= (my >> 3) < 0 || (my >> 3) + 8 >= (pic_height >> 1);
 560     }
 561
 562     src_cb = pic->f.data[1] + ((mx >> 3) << pixel_shift) +
 563              (my >> ysh) * h->mb_uvlinesize;
 564     src_cr = pic->f.data[2] + ((mx >> 3) << pixel_shift) +
 565              (my >> ysh) * h->mb_uvlinesize;
 566
 567     if (emu) {
 568         s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cb, h->mb_uvlinesize,
 569                                 9, 8 * chroma_idc + 1, (mx >> 3), (my >> ysh),
 570                                 pic_width >> 1, pic_height >> (chroma_idc == 1 /* yuv420 */));
 571         src_cb = s->edge_emu_buffer;
 572     }
 573     chroma_op(dest_cb, src_cb, h->mb_uvlinesize,
 574               height >> (chroma_idc == 1 /* yuv420 */),
 575               mx & 7, (my << (chroma_idc == 2 /* yuv422 */)) & 7);
 576
 577     if (emu) {
 578         s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cr, h->mb_uvlinesize,
 579                                 9, 8 * chroma_idc + 1, (mx >> 3), (my >> ysh),
 580                                 pic_width >> 1, pic_height >> (chroma_idc == 1 /* yuv420 */));
 581         src_cr = s->edge_emu_buffer;
 582     }
 583     chroma_op(dest_cr, src_cr, h->mb_uvlinesize, height >> (chroma_idc == 1 /* yuv420 */),
 584               mx & 7, (my << (chroma_idc == 2 /* yuv422 */)) & 7);
 585 }
 586
 587 static av_always_inline void mc_part_std(H264Context *h, int n, int square,
 588                                          int height, int delta,
 589                                          uint8_t *dest_y, uint8_t *dest_cb,
 590                                          uint8_t *dest_cr,
 591                                          int x_offset, int y_offset,
 592                                          qpel_mc_func *qpix_put,
 593                                          h264_chroma_mc_func chroma_put,
 594                                          qpel_mc_func *qpix_avg,
 595                                          h264_chroma_mc_func chroma_avg,
 596                                          int list0, int list1,
 597                                          int pixel_shift, int chroma_idc)
 598 {
 599     MpegEncContext *const s       = &h->s;
 600     qpel_mc_func *qpix_op         = qpix_put;
 601     h264_chroma_mc_func chroma_op = chroma_put;
 602
 603     dest_y += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize;
 604     if (chroma_idc == 3 /* yuv444 */) {
 605         dest_cb += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize;
 606         dest_cr += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize;
 607     } else if (chroma_idc == 2 /* yuv422 */) {
 608         dest_cb += (x_offset << pixel_shift) + 2 * y_offset * h->mb_uvlinesize;
 609         dest_cr += (x_offset << pixel_shift) + 2 * y_offset * h->mb_uvlinesize;
 610     } else { /* yuv420 */
 611         dest_cb += (x_offset << pixel_shift) + y_offset * h->mb_uvlinesize;
 612         dest_cr += (x_offset << pixel_shift) + y_offset * h->mb_uvlinesize;
 613     }
 614     x_offset += 8 * s->mb_x;
 615     y_offset += 8 * (s->mb_y >> MB_FIELD);
 616
 617     if (list0) {
 618         Picture *ref = &h->ref_list[0][h->ref_cache[0][scan8[n]]];
 619         mc_dir_part(h, ref, n, square, height, delta, 0,
 620                     dest_y, dest_cb, dest_cr, x_offset, y_offset,
 621                     qpix_op, chroma_op, pixel_shift, chroma_idc);
 622
 623         qpix_op   = qpix_avg;
 624         chroma_op = chroma_avg;
 625     }
 626
 627     if (list1) {
 628         Picture *ref = &h->ref_list[1][h->ref_cache[1][scan8[n]]];
 629         mc_dir_part(h, ref, n, square, height, delta, 1,
 630                     dest_y, dest_cb, dest_cr, x_offset, y_offset,
 631                     qpix_op, chroma_op, pixel_shift, chroma_idc);
 632     }
 633 }
 634
 635 static av_always_inline void mc_part_weighted(H264Context *h, int n, int square,
 636                                               int height, int delta,
 637                                               uint8_t *dest_y, uint8_t *dest_cb,
 638                                               uint8_t *dest_cr,
 639                                               int x_offset, int y_offset,
 640                                               qpel_mc_func *qpix_put,
 641                                               h264_chroma_mc_func chroma_put,
 642                                               h264_weight_func luma_weight_op,
 643                                               h264_weight_func chroma_weight_op,
 644                                               h264_biweight_func luma_weight_avg,
 645                                               h264_biweight_func chroma_weight_avg,
 646                                               int list0, int list1,
 647                                               int pixel_shift, int chroma_idc)
 648 {
 649     MpegEncContext *const s = &h->s;
 650     int chroma_height;
 651
 652     dest_y += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize;
 653     if (chroma_idc == 3 /* yuv444 */) {
 654         chroma_height     = height;
 655         chroma_weight_avg = luma_weight_avg;
 656         chroma_weight_op  = luma_weight_op;
 657         dest_cb += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize;
 658         dest_cr += (2 * x_offset << pixel_shift) + 2 * y_offset * h->mb_linesize;
 659     } else if (chroma_idc == 2 /* yuv422 */) {
 660         chroma_height = height;
 661         dest_cb      += (x_offset << pixel_shift) + 2 * y_offset * h->mb_uvlinesize;
 662         dest_cr      += (x_offset << pixel_shift) + 2 * y_offset * h->mb_uvlinesize;
 663     } else { /* yuv420 */
 664         chroma_height = height >> 1;
 665         dest_cb      += (x_offset << pixel_shift) + y_offset * h->mb_uvlinesize;
 666         dest_cr      += (x_offset << pixel_shift) + y_offset * h->mb_uvlinesize;
 667     }
 668     x_offset += 8 * s->mb_x;
 669     y_offset += 8 * (s->mb_y >> MB_FIELD);
 670
 671     if (list0 && list1) {
 672         /* don't optimize for luma-only case, since B-frames usually
 673          * use implicit weights => chroma too. */
 674         uint8_t *tmp_cb = h->bipred_scratchpad;
 675         uint8_t *tmp_cr = h->bipred_scratchpad + (16 << pixel_shift);
 676         uint8_t *tmp_y  = h->bipred_scratchpad + 16 * h->mb_uvlinesize;
 677         int refn0       = h->ref_cache[0][scan8[n]];
 678         int refn1       = h->ref_cache[1][scan8[n]];
 679
 680         mc_dir_part(h, &h->ref_list[0][refn0], n, square, height, delta, 0,
 681                     dest_y, dest_cb, dest_cr,
 682                     x_offset, y_offset, qpix_put, chroma_put,
 683                     pixel_shift, chroma_idc);
 684         mc_dir_part(h, &h->ref_list[1][refn1], n, square, height, delta, 1,
 685                     tmp_y, tmp_cb, tmp_cr,
 686                     x_offset, y_offset, qpix_put, chroma_put,
 687                     pixel_shift, chroma_idc);
 688
 689         if (h->use_weight == 2) {
 690             int weight0 = h->implicit_weight[refn0][refn1][s->mb_y & 1];
 691             int weight1 = 64 - weight0;
 692             luma_weight_avg(dest_y, tmp_y, h->mb_linesize,
 693                             height, 5, weight0, weight1, 0);
 694             chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize,
 695                               chroma_height, 5, weight0, weight1, 0);
 696             chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize,
 697                               chroma_height, 5, weight0, weight1, 0);
 698         } else {
 699             luma_weight_avg(dest_y, tmp_y, h->mb_linesize, height,
 700                             h->luma_log2_weight_denom,
 701                             h->luma_weight[refn0][0][0],
 702                             h->luma_weight[refn1][1][0],
 703                             h->luma_weight[refn0][0][1] +
 704                             h->luma_weight[refn1][1][1]);
 705             chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, chroma_height,
 706                               h->chroma_log2_weight_denom,
 707                               h->chroma_weight[refn0][0][0][0],
 708                               h->chroma_weight[refn1][1][0][0],
 709                               h->chroma_weight[refn0][0][0][1] +
 710                               h->chroma_weight[refn1][1][0][1]);
 711             chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, chroma_height,
 712                               h->chroma_log2_weight_denom,
 713                               h->chroma_weight[refn0][0][1][0],
 714                               h->chroma_weight[refn1][1][1][0],
 715                               h->chroma_weight[refn0][0][1][1] +
 716                               h->chroma_weight[refn1][1][1][1]);
 717         }
 718     } else {
 719         int list     = list1 ? 1 : 0;
 720         int refn     = h->ref_cache[list][scan8[n]];
 721         Picture *ref = &h->ref_list[list][refn];
 722         mc_dir_part(h, ref, n, square, height, delta, list,
 723                     dest_y, dest_cb, dest_cr, x_offset, y_offset,
 724                     qpix_put, chroma_put, pixel_shift, chroma_idc);
 725
 726         luma_weight_op(dest_y, h->mb_linesize, height,
 727                        h->luma_log2_weight_denom,
 728                        h->luma_weight[refn][list][0],
 729                        h->luma_weight[refn][list][1]);
 730         if (h->use_weight_chroma) {
 731             chroma_weight_op(dest_cb, h->mb_uvlinesize, chroma_height,
 732                              h->chroma_log2_weight_denom,
 733                              h->chroma_weight[refn][list][0][0],
 734                              h->chroma_weight[refn][list][0][1]);
 735             chroma_weight_op(dest_cr, h->mb_uvlinesize, chroma_height,
 736                              h->chroma_log2_weight_denom,
 737                              h->chroma_weight[refn][list][1][0],
 738                              h->chroma_weight[refn][list][1][1]);
 739         }
 740     }
 741 }
 742
 743 static av_always_inline void prefetch_motion(H264Context *h, int list,
 744                                              int pixel_shift, int chroma_idc)
 745 {
 746     /* fetch pixels for estimated mv 4 macroblocks ahead
 747      * optimized for 64byte cache lines */
 748     MpegEncContext *const s = &h->s;
 749     const int refn = h->ref_cache[list][scan8[0]];
 750     if (refn >= 0) {
 751         const int mx  = (h->mv_cache[list][scan8[0]][0] >> 2) + 16 * s->mb_x + 8;
 752         const int my  = (h->mv_cache[list][scan8[0]][1] >> 2) + 16 * s->mb_y;
 753         uint8_t **src = h->ref_list[list][refn].f.data;
 754         int off       = (mx << pixel_shift) +
 755                         (my + (s->mb_x & 3) * 4) * h->mb_linesize +
 756                         (64 << pixel_shift);
 757         s->dsp.prefetch(src[0] + off, s->linesize, 4);
 758         if (chroma_idc == 3 /* yuv444 */) {
 759             s->dsp.prefetch(src[1] + off, s->linesize, 4);
 760             s->dsp.prefetch(src[2] + off, s->linesize, 4);
 761         } else {
 762             off= (((mx>>1)+64)<<pixel_shift) + ((my>>1) + (s->mb_x&7))*s->uvlinesize;
 763             s->dsp.prefetch(src[1] + off, src[2] - src[1], 2);
 764         }
 765     }
 766 }
 767
 768 static void free_tables(H264Context *h, int free_rbsp)
 769 {
 770     int i;
 771     H264Context *hx;
 772
 773     av_freep(&h->intra4x4_pred_mode);
 774     av_freep(&h->chroma_pred_mode_table);
 775     av_freep(&h->cbp_table);
 776     av_freep(&h->mvd_table[0]);
 777     av_freep(&h->mvd_table[1]);
 778     av_freep(&h->direct_table);
 779     av_freep(&h->non_zero_count);
 780     av_freep(&h->slice_table_base);
 781     h->slice_table = NULL;
 782     av_freep(&h->list_counts);
 783
 784     av_freep(&h->mb2b_xy);
 785     av_freep(&h->mb2br_xy);
 786
 787     for (i = 0; i < MAX_THREADS; i++) {
 788         hx = h->thread_context[i];
 789         if (!hx)
 790             continue;
 791         av_freep(&hx->top_borders[1]);
 792         av_freep(&hx->top_borders[0]);
 793         av_freep(&hx->bipred_scratchpad);
 794         if (free_rbsp) {
 795             av_freep(&hx->rbsp_buffer[1]);
 796             av_freep(&hx->rbsp_buffer[0]);
 797             hx->rbsp_buffer_size[0] = 0;
 798             hx->rbsp_buffer_size[1] = 0;
 799         }
 800         if (i)
 801             av_freep(&h->thread_context[i]);
 802     }
 803 }
 804
 805 static void init_dequant8_coeff_table(H264Context *h)
 806 {
 807     int i, j, q, x;
 808     const int max_qp = 51 + 6 * (h->sps.bit_depth_luma - 8);
 809
 810     for (i = 0; i < 6; i++) {
 811         h->dequant8_coeff[i] = h->dequant8_buffer[i];
 812         for (j = 0; j < i; j++)
 813             if (!memcmp(h->pps.scaling_matrix8[j], h->pps.scaling_matrix8[i],
 814                         64 * sizeof(uint8_t))) {
 815                 h->dequant8_coeff[i] = h->dequant8_buffer[j];
 816                 break;
 817             }
 818         if (j < i)
 819             continue;
 820
 821         for (q = 0; q < max_qp + 1; q++) {
 822             int shift = div6[q];
 823             int idx   = rem6[q];
 824             for (x = 0; x < 64; x++)
 825                 h->dequant8_coeff[i][q][(x >> 3) | ((x & 7) << 3)] =
 826                     ((uint32_t)dequant8_coeff_init[idx][dequant8_coeff_init_scan[((x >> 1) & 12) | (x & 3)]] *
 827                      h->pps.scaling_matrix8[i][x]) << shift;
 828         }
 829     }
 830 }
 831
 832 static void init_dequant4_coeff_table(H264Context *h)
 833 {
 834     int i, j, q, x;
 835     const int max_qp = 51 + 6 * (h->sps.bit_depth_luma - 8);
 836     for (i = 0; i < 6; i++) {
 837         h->dequant4_coeff[i] = h->dequant4_buffer[i];
 838         for (j = 0; j < i; j++)
 839             if (!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i],
 840                         16 * sizeof(uint8_t))) {
 841                 h->dequant4_coeff[i] = h->dequant4_buffer[j];
 842                 break;
 843             }
 844         if (j < i)
 845             continue;
 846
 847         for (q = 0; q < max_qp + 1; q++) {
 848             int shift = div6[q] + 2;
 849             int idx   = rem6[q];
 850             for (x = 0; x < 16; x++)
 851                 h->dequant4_coeff[i][q][(x >> 2) | ((x << 2) & 0xF)] =
 852                     ((uint32_t)dequant4_coeff_init[idx][(x & 1) + ((x >> 2) & 1)] *
 853                      h->pps.scaling_matrix4[i][x]) << shift;
 854         }
 855     }
 856 }
 857
 858 static void init_dequant_tables(H264Context *h)
 859 {
 860     int i, x;
 861     init_dequant4_coeff_table(h);
 862     if (h->pps.transform_8x8_mode)
 863         init_dequant8_coeff_table(h);
 864     if (h->sps.transform_bypass) {
 865         for (i = 0; i < 6; i++)
 866             for (x = 0; x < 16; x++)
 867                 h->dequant4_coeff[i][0][x] = 1 << 6;
 868         if (h->pps.transform_8x8_mode)
 869             for (i = 0; i < 6; i++)
 870                 for (x = 0; x < 64; x++)
 871                     h->dequant8_coeff[i][0][x] = 1 << 6;
 872     }
 873 }
 874
 875 int ff_h264_alloc_tables(H264Context *h)
 876 {
 877     MpegEncContext *const s = &h->s;
 878     const int big_mb_num    = s->mb_stride * (s->mb_height + 1);
 879     const int row_mb_num    = 2*s->mb_stride*FFMAX(s->avctx->thread_count, 1);
 880     int x, y;
 881
 882     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->intra4x4_pred_mode,
 883                       row_mb_num * 8 * sizeof(uint8_t), fail)
 884     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->non_zero_count,
 885                       big_mb_num * 48 * sizeof(uint8_t), fail)
 886     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->slice_table_base,
 887                       (big_mb_num + s->mb_stride) * sizeof(*h->slice_table_base), fail)
 888     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->cbp_table,
 889                       big_mb_num * sizeof(uint16_t), fail)
 890     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->chroma_pred_mode_table,
 891                       big_mb_num * sizeof(uint8_t), fail)
 892     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[0],
 893                       16 * row_mb_num * sizeof(uint8_t), fail);
 894     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mvd_table[1],
 895                       16 * row_mb_num * sizeof(uint8_t), fail);
 896     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->direct_table,
 897                       4 * big_mb_num * sizeof(uint8_t), fail);
 898     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->list_counts,
 899                       big_mb_num * sizeof(uint8_t), fail)
 900
 901     memset(h->slice_table_base, -1,
 902            (big_mb_num + s->mb_stride) * sizeof(*h->slice_table_base));
 903     h->slice_table = h->slice_table_base + s->mb_stride * 2 + 1;
 904
 905     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2b_xy,
 906                       big_mb_num * sizeof(uint32_t), fail);
 907     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->mb2br_xy,
 908                       big_mb_num * sizeof(uint32_t), fail);
 909     for (y = 0; y < s->mb_height; y++)
 910         for (x = 0; x < s->mb_width; x++) {
 911             const int mb_xy = x + y * s->mb_stride;
 912             const int b_xy  = 4 * x + 4 * y * h->b_stride;
 913
 914             h->mb2b_xy[mb_xy]  = b_xy;
 915             h->mb2br_xy[mb_xy] = 8 * (FMO ? mb_xy : (mb_xy % (2 * s->mb_stride)));
 916         }
 917
 918     if (!h->dequant4_coeff[0])
 919         init_dequant_tables(h);
 920
 921     return 0;
 922
 923 fail:
 924     free_tables(h, 1);
 925     return -1;
 926 }
 927
 928 /**
 929  * Mimic alloc_tables(), but for every context thread.
 930  */
 931 static void clone_tables(H264Context *dst, H264Context *src, int i)
 932 {
 933     MpegEncContext *const s     = &src->s;
 934     dst->intra4x4_pred_mode     = src->intra4x4_pred_mode + i * 8 * 2 * s->mb_stride;
 935     dst->non_zero_count         = src->non_zero_count;
 936     dst->slice_table            = src->slice_table;
 937     dst->cbp_table              = src->cbp_table;
 938     dst->mb2b_xy                = src->mb2b_xy;
 939     dst->mb2br_xy               = src->mb2br_xy;
 940     dst->chroma_pred_mode_table = src->chroma_pred_mode_table;
 941     dst->mvd_table[0]           = src->mvd_table[0] + i * 8 * 2 * s->mb_stride;
 942     dst->mvd_table[1]           = src->mvd_table[1] + i * 8 * 2 * s->mb_stride;
 943     dst->direct_table           = src->direct_table;
 944     dst->list_counts            = src->list_counts;
 945     dst->bipred_scratchpad      = NULL;
 946     ff_h264_pred_init(&dst->hpc, src->s.codec_id, src->sps.bit_depth_luma,
 947                       src->sps.chroma_format_idc);
 948 }
 949
 950 /**
 951  * Init context
 952  * Allocate buffers which are not shared amongst multiple threads.
 953  */
 954 static int context_init(H264Context *h)
 955 {
 956     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[0],
 957                       h->s.mb_width * 16 * 3 * sizeof(uint8_t) * 2, fail)
 958     FF_ALLOCZ_OR_GOTO(h->s.avctx, h->top_borders[1],
 959                       h->s.mb_width * 16 * 3 * sizeof(uint8_t) * 2, fail)
 960
 961     h->ref_cache[0][scan8[5]  + 1] =
 962     h->ref_cache[0][scan8[7]  + 1] =
 963     h->ref_cache[0][scan8[13] + 1] =
 964     h->ref_cache[1][scan8[5]  + 1] =
 965     h->ref_cache[1][scan8[7]  + 1] =
 966     h->ref_cache[1][scan8[13] + 1] = PART_NOT_AVAILABLE;
 967
 968     return 0;
 969
 970 fail:
 971     return -1; // free_tables will clean up for us
 972 }
 973
 974 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size);
 975
 976 static av_cold void common_init(H264Context *h)
 977 {
 978     MpegEncContext *const s = &h->s;
 979
 980     s->width    = s->avctx->width;
 981     s->height   = s->avctx->height;
 982     s->codec_id = s->avctx->codec->id;
 983
 984     s->avctx->bits_per_raw_sample = 8;
 985     h->cur_chroma_format_idc = 1;
 986
 987     ff_h264dsp_init(&h->h264dsp,
 988                     s->avctx->bits_per_raw_sample, h->cur_chroma_format_idc);
 989     ff_h264_pred_init(&h->hpc, s->codec_id,
 990                       s->avctx->bits_per_raw_sample, h->cur_chroma_format_idc);
 991
 992     h->dequant_coeff_pps = -1;
 993     s->unrestricted_mv   = 1;
 994
 995     s->dsp.dct_bits = 16;
 996     /* needed so that IDCT permutation is known early */
 997     ff_dsputil_init(&s->dsp, s->avctx);
 998
 999     memset(h->pps.scaling_matrix4, 16, 6 * 16 * sizeof(uint8_t));
1000     memset(h->pps.scaling_matrix8, 16, 2 * 64 * sizeof(uint8_t));
1001 }
1002
1003 static int ff_h264_decode_extradata_internal(H264Context *h, const uint8_t *buf, int size)
1004 {
1005     AVCodecContext *avctx = h->s.avctx;
1006
1007     if (!buf || size <= 0)
1008         return -1;
1009
1010     if (buf[0] == 1) {
1011         int i, cnt, nalsize;
1012         const unsigned char *p = buf;
1013
1014         h->is_avc = 1;
1015
1016         if (size < 7) {
1017             av_log(avctx, AV_LOG_ERROR, "avcC too short\n");
1018             return -1;
1019         }
1020         /* sps and pps in the avcC always have length coded with 2 bytes,
1021          * so put a fake nal_length_size = 2 while parsing them */
1022         h->nal_length_size = 2;
1023         // Decode sps from avcC
1024         cnt = *(p + 5) & 0x1f; // Number of sps
1025         p  += 6;
1026         for (i = 0; i < cnt; i++) {
1027             nalsize = AV_RB16(p) + 2;
1028             if(nalsize > size - (p-buf))
1029                 return -1;
1030             if (decode_nal_units(h, p, nalsize) < 0) {
1031                 av_log(avctx, AV_LOG_ERROR,
1032                        "Decoding sps %d from avcC failed\n", i);
1033                 return -1;
1034             }
1035             p += nalsize;
1036         }
1037         // Decode pps from avcC
1038         cnt = *(p++); // Number of pps
1039         for (i = 0; i < cnt; i++) {
1040             nalsize = AV_RB16(p) + 2;
1041             if(nalsize > size - (p-buf))
1042                 return -1;
1043             if (decode_nal_units(h, p, nalsize) < 0) {
1044                 av_log(avctx, AV_LOG_ERROR,
1045                        "Decoding pps %d from avcC failed\n", i);
1046                 return -1;
1047             }
1048             p += nalsize;
1049         }
1050         // Now store right nal length size, that will be used to parse all other nals
1051         h->nal_length_size = (buf[4] & 0x03) + 1;
1052     } else {
1053         h->is_avc = 0;
1054         if (decode_nal_units(h, buf, size) < 0)
1055             return -1;
1056     }
1057     return size;
1058 }
1059
1060 int ff_h264_decode_extradata(H264Context *h, const uint8_t *buf, int size)
1061 {
1062     int ret;
1063     h->decoding_extradata = 1;
1064     ret = ff_h264_decode_extradata_internal(h, buf, size);
1065     h->decoding_extradata = 0;
1066     return ret;
1067 }
1068
1069 av_cold int ff_h264_decode_init(AVCodecContext *avctx)
1070 {
1071     H264Context *h = avctx->priv_data;
1072     MpegEncContext *const s = &h->s;
1073     int i;
1074
1075     ff_MPV_decode_defaults(s);
1076
1077     s->avctx = avctx;
1078     common_init(h);
1079
1080     s->out_format      = FMT_H264;
1081     s->workaround_bugs = avctx->workaround_bugs;
1082
1083     /* set defaults */
1084     // s->decode_mb = ff_h263_decode_mb;
1085     s->quarter_sample = 1;
1086     if (!avctx->has_b_frames)
1087         s->low_delay = 1;
1088
1089     avctx->chroma_sample_location = AVCHROMA_LOC_LEFT;
1090
1091     ff_h264_decode_init_vlc();
1092
1093     h->pixel_shift = 0;
1094     h->sps.bit_depth_luma = avctx->bits_per_raw_sample = 8;
1095
1096     h->thread_context[0] = h;
1097     h->outputed_poc      = h->next_outputed_poc = INT_MIN;
1098     for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
1099         h->last_pocs[i] = INT_MIN;
1100     h->prev_poc_msb = 1 << 16;
1101     h->prev_frame_num = -1;
1102     h->x264_build   = -1;
1103     ff_h264_reset_sei(h);
1104     if (avctx->codec_id == AV_CODEC_ID_H264) {
1105         if (avctx->ticks_per_frame == 1) {
1106             if(s->avctx->time_base.den < INT_MAX/2) {
1107                 s->avctx->time_base.den *= 2;
1108             } else
1109                 s->avctx->time_base.num /= 2;
1110         }
1111         avctx->ticks_per_frame = 2;
1112     }
1113
1114     if (avctx->extradata_size > 0 && avctx->extradata &&
1115         ff_h264_decode_extradata(h, avctx->extradata, avctx->extradata_size) < 0) {
1116         ff_h264_free_context(h);
1117         return -1;
1118     }
1119
1120     if (h->sps.bitstream_restriction_flag &&
1121         s->avctx->has_b_frames < h->sps.num_reorder_frames) {
1122         s->avctx->has_b_frames = h->sps.num_reorder_frames;
1123         s->low_delay           = 0;
1124     }
1125
1126     ff_init_cabac_states();
1127
1128     return 0;
1129 }
1130
1131 #define IN_RANGE(a, b, size) (((a) >= (b)) && ((a) < ((b) + (size))))
1132
1133 static void copy_picture_range(Picture **to, Picture **from, int count,
1134                                MpegEncContext *new_base,
1135                                MpegEncContext *old_base)
1136 {
1137     int i;
1138
1139     for (i = 0; i < count; i++) {
1140         assert((IN_RANGE(from[i], old_base, sizeof(*old_base)) ||
1141                 IN_RANGE(from[i], old_base->picture,
1142                          sizeof(Picture) * old_base->picture_count) ||
1143                 !from[i]));
1144         to[i] = REBASE_PICTURE(from[i], new_base, old_base);
1145     }
1146 }
1147
1148 static void copy_parameter_set(void **to, void **from, int count, int size)
1149 {
1150     int i;
1151
1152     for (i = 0; i < count; i++) {
1153         if (to[i] && !from[i])
1154             av_freep(&to[i]);
1155         else if (from[i] && !to[i])
1156             to[i] = av_malloc(size);
1157
1158         if (from[i])
1159             memcpy(to[i], from[i], size);
1160     }
1161 }
1162
1163 static int decode_init_thread_copy(AVCodecContext *avctx)
1164 {
1165     H264Context *h = avctx->priv_data;
1166
1167     if (!avctx->internal->is_copy)
1168         return 0;
1169     memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
1170     memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
1171
1172     return 0;
1173 }
1174
1175 #define copy_fields(to, from, start_field, end_field)                   \
1176     memcpy(&to->start_field, &from->start_field,                        \
1177            (char *)&to->end_field - (char *)&to->start_field)
1178
1179 static int decode_update_thread_context(AVCodecContext *dst,
1180                                         const AVCodecContext *src)
1181 {
1182     H264Context *h = dst->priv_data, *h1 = src->priv_data;
1183     MpegEncContext *const s = &h->s, *const s1 = &h1->s;
1184     int inited = s->context_initialized, err;
1185     int i;
1186
1187     if (dst == src)
1188         return 0;
1189
1190     err = ff_mpeg_update_thread_context(dst, src);
1191     if (err)
1192         return err;
1193
1194     // FIXME handle width/height changing
1195     if (!inited) {
1196         for (i = 0; i < MAX_SPS_COUNT; i++)
1197             av_freep(h->sps_buffers + i);
1198
1199         for (i = 0; i < MAX_PPS_COUNT; i++)
1200             av_freep(h->pps_buffers + i);
1201
1202         // copy all fields after MpegEnc
1203         memcpy(&h->s + 1, &h1->s + 1,
1204                sizeof(H264Context) - sizeof(MpegEncContext));
1205         memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
1206         memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
1207
1208         if (s1->context_initialized) {
1209         if (ff_h264_alloc_tables(h) < 0) {
1210             av_log(dst, AV_LOG_ERROR, "Could not allocate memory for h264\n");
1211             return AVERROR(ENOMEM);
1212         }
1213         context_init(h);
1214
1215         }
1216
1217         for (i = 0; i < 2; i++) {
1218             h->rbsp_buffer[i]      = NULL;
1219             h->rbsp_buffer_size[i] = 0;
1220         }
1221         h->bipred_scratchpad = NULL;
1222
1223         h->thread_context[0] = h;
1224
1225         s->dsp.clear_blocks(h->mb);
1226         s->dsp.clear_blocks(h->mb + (24 * 16 << h->pixel_shift));
1227     }
1228
1229     /* frame_start may not be called for the next thread (if it's decoding
1230      * a bottom field) so this has to be allocated here */
1231     if (!h->bipred_scratchpad && s->linesize)
1232         h->bipred_scratchpad = av_malloc(16 * 6 * s->linesize);
1233
1234     // extradata/NAL handling
1235     h->is_avc = h1->is_avc;
1236
1237     // SPS/PPS
1238     copy_parameter_set((void **)h->sps_buffers, (void **)h1->sps_buffers,
1239                        MAX_SPS_COUNT, sizeof(SPS));
1240     h->sps = h1->sps;
1241     copy_parameter_set((void **)h->pps_buffers, (void **)h1->pps_buffers,
1242                        MAX_PPS_COUNT, sizeof(PPS));
1243     h->pps = h1->pps;
1244
1245     // Dequantization matrices
1246     // FIXME these are big - can they be only copied when PPS changes?
1247     copy_fields(h, h1, dequant4_buffer, dequant4_coeff);
1248
1249     for (i = 0; i < 6; i++)
1250         h->dequant4_coeff[i] = h->dequant4_buffer[0] +
1251                                (h1->dequant4_coeff[i] - h1->dequant4_buffer[0]);
1252
1253     for (i = 0; i < 6; i++)
1254         h->dequant8_coeff[i] = h->dequant8_buffer[0] +
1255                                (h1->dequant8_coeff[i] - h1->dequant8_buffer[0]);
1256
1257     h->dequant_coeff_pps = h1->dequant_coeff_pps;
1258
1259     // POC timing
1260     copy_fields(h, h1, poc_lsb, redundant_pic_count);
1261
1262     // reference lists
1263     copy_fields(h, h1, ref_count, list_count);
1264     copy_fields(h, h1, ref_list, intra_gb);
1265     copy_fields(h, h1, short_ref, cabac_init_idc);
1266
1267     copy_picture_range(h->short_ref, h1->short_ref, 32, s, s1);
1268     copy_picture_range(h->long_ref, h1->long_ref, 32, s, s1);
1269     copy_picture_range(h->delayed_pic, h1->delayed_pic,
1270                        MAX_DELAYED_PIC_COUNT + 2, s, s1);
1271
1272     h->last_slice_type = h1->last_slice_type;
1273     h->sync            = h1->sync;
1274
1275     if (!s->current_picture_ptr)
1276         return 0;
1277
1278     if (!s->droppable) {
1279         err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1280         h->prev_poc_msb = h->poc_msb;
1281         h->prev_poc_lsb = h->poc_lsb;
1282     }
1283     h->prev_frame_num_offset = h->frame_num_offset;
1284     h->prev_frame_num        = h->frame_num;
1285     h->outputed_poc          = h->next_outputed_poc;
1286
1287     return err;
1288 }
1289
1290 int ff_h264_frame_start(H264Context *h)
1291 {
1292     MpegEncContext *const s = &h->s;
1293     int i;
1294     const int pixel_shift = h->pixel_shift;
1295
1296     if (ff_MPV_frame_start(s, s->avctx) < 0)
1297         return -1;
1298     ff_er_frame_start(s);
1299     /*
1300      * ff_MPV_frame_start uses pict_type to derive key_frame.
1301      * This is incorrect for H.264; IDR markings must be used.
1302      * Zero here; IDR markings per slice in frame or fields are ORed in later.
1303      * See decode_nal_units().
1304      */
1305     s->current_picture_ptr->f.key_frame = 0;
1306     s->current_picture_ptr->sync        = 0;
1307     s->current_picture_ptr->mmco_reset  = 0;
1308
1309     assert(s->linesize && s->uvlinesize);
1310
1311     for (i = 0; i < 16; i++) {
1312         h->block_offset[i]           = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * s->linesize * ((scan8[i] - scan8[0]) >> 3);
1313         h->block_offset[48 + i]      = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * s->linesize * ((scan8[i] - scan8[0]) >> 3);
1314     }
1315     for (i = 0; i < 16; i++) {
1316         h->block_offset[16 + i]      =
1317         h->block_offset[32 + i]      = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * s->uvlinesize * ((scan8[i] - scan8[0]) >> 3);
1318         h->block_offset[48 + 16 + i] =
1319         h->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * s->uvlinesize * ((scan8[i] - scan8[0]) >> 3);
1320     }
1321
1322     /* can't be in alloc_tables because linesize isn't known there.
1323      * FIXME: redo bipred weight to not require extra buffer? */
1324     for (i = 0; i < s->slice_context_count; i++)
1325         if (h->thread_context[i] && !h->thread_context[i]->bipred_scratchpad)
1326             h->thread_context[i]->bipred_scratchpad = av_malloc(16 * 6 * s->linesize);
1327
1328     /* Some macroblocks can be accessed before they're available in case
1329      * of lost slices, MBAFF or threading. */
1330     memset(h->slice_table, -1,
1331            (s->mb_height * s->mb_stride - 1) * sizeof(*h->slice_table));
1332
1333     // s->decode = (s->flags & CODEC_FLAG_PSNR) || !s->encoding ||
1334     //             s->current_picture.f.reference /* || h->contains_intra */ || 1;
1335
1336     /* We mark the current picture as non-reference after allocating it, so
1337      * that if we break out due to an error it can be released automatically
1338      * in the next ff_MPV_frame_start().
1339      * SVQ3 as well as most other codecs have only last/next/current and thus
1340      * get released even with set reference, besides SVQ3 and others do not
1341      * mark frames as reference later "naturally". */
1342     if (s->codec_id != AV_CODEC_ID_SVQ3)
1343         s->current_picture_ptr->f.reference = 0;
1344
1345     s->current_picture_ptr->field_poc[0]     =
1346         s->current_picture_ptr->field_poc[1] = INT_MAX;
1347
1348     h->next_output_pic = NULL;
1349
1350     assert(s->current_picture_ptr->long_ref == 0);
1351
1352     return 0;
1353 }
1354
1355 /**
1356  * Run setup operations that must be run after slice header decoding.
1357  * This includes finding the next displayed frame.
1358  *
1359  * @param h h264 master context
1360  * @param setup_finished enough NALs have been read that we can call
1361  * ff_thread_finish_setup()
1362  */
1363 static void decode_postinit(H264Context *h, int setup_finished)
1364 {
1365     MpegEncContext *const s = &h->s;
1366     Picture *out = s->current_picture_ptr;
1367     Picture *cur = s->current_picture_ptr;
1368     int i, pics, out_of_order, out_idx;
1369
1370     s->current_picture_ptr->f.qscale_type = FF_QSCALE_TYPE_H264;
1371     s->current_picture_ptr->f.pict_type   = s->pict_type;
1372
1373     if (h->next_output_pic)
1374         return;
1375
1376     if (cur->field_poc[0] == INT_MAX || cur->field_poc[1] == INT_MAX) {
1377         /* FIXME: if we have two PAFF fields in one packet, we can't start
1378          * the next thread here. If we have one field per packet, we can.
1379          * The check in decode_nal_units() is not good enough to find this
1380          * yet, so we assume the worst for now. */
1381         // if (setup_finished)
1382         //    ff_thread_finish_setup(s->avctx);
1383         return;
1384     }
1385
1386     cur->f.interlaced_frame = 0;
1387     cur->f.repeat_pict      = 0;
1388
1389     /* Signal interlacing information externally. */
1390     /* Prioritize picture timing SEI information over used
1391      * decoding process if it exists. */
1392
1393     if (h->sps.pic_struct_present_flag) {
1394         switch (h->sei_pic_struct) {
1395         case SEI_PIC_STRUCT_FRAME:
1396             break;
1397         case SEI_PIC_STRUCT_TOP_FIELD:
1398         case SEI_PIC_STRUCT_BOTTOM_FIELD:
1399             cur->f.interlaced_frame = 1;
1400             break;
1401         case SEI_PIC_STRUCT_TOP_BOTTOM:
1402         case SEI_PIC_STRUCT_BOTTOM_TOP:
1403             if (FIELD_OR_MBAFF_PICTURE)
1404                 cur->f.interlaced_frame = 1;
1405             else
1406                 // try to flag soft telecine progressive
1407                 cur->f.interlaced_frame = h->prev_interlaced_frame;
1408             break;
1409         case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
1410         case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
1411             /* Signal the possibility of telecined film externally
1412              * (pic_struct 5,6). From these hints, let the applications
1413              * decide if they apply deinterlacing. */
1414             cur->f.repeat_pict = 1;
1415             break;
1416         case SEI_PIC_STRUCT_FRAME_DOUBLING:
1417             // Force progressive here, doubling interlaced frame is a bad idea.
1418             cur->f.repeat_pict = 2;
1419             break;
1420         case SEI_PIC_STRUCT_FRAME_TRIPLING:
1421             cur->f.repeat_pict = 4;
1422             break;
1423         }
1424
1425         if ((h->sei_ct_type & 3) &&
1426             h->sei_pic_struct <= SEI_PIC_STRUCT_BOTTOM_TOP)
1427             cur->f.interlaced_frame = (h->sei_ct_type & (1 << 1)) != 0;
1428     } else {
1429         /* Derive interlacing flag from used decoding process. */
1430         cur->f.interlaced_frame = FIELD_OR_MBAFF_PICTURE;
1431     }
1432     h->prev_interlaced_frame = cur->f.interlaced_frame;
1433
1434     if (cur->field_poc[0] != cur->field_poc[1]) {
1435         /* Derive top_field_first from field pocs. */
1436         cur->f.top_field_first = cur->field_poc[0] < cur->field_poc[1];
1437     } else {
1438         if (cur->f.interlaced_frame || h->sps.pic_struct_present_flag) {
1439             /* Use picture timing SEI information. Even if it is a
1440              * information of a past frame, better than nothing. */
1441             if (h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM ||
1442                 h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM_TOP)
1443                 cur->f.top_field_first = 1;
1444             else
1445                 cur->f.top_field_first = 0;
1446         } else {
1447             /* Most likely progressive */
1448             cur->f.top_field_first = 0;
1449         }
1450     }
1451
1452     cur->mmco_reset = h->mmco_reset;
1453     h->mmco_reset = 0;
1454     // FIXME do something with unavailable reference frames
1455
1456     /* Sort B-frames into display order */
1457
1458     if (h->sps.bitstream_restriction_flag &&
1459         s->avctx->has_b_frames < h->sps.num_reorder_frames) {
1460         s->avctx->has_b_frames = h->sps.num_reorder_frames;
1461         s->low_delay           = 0;
1462     }
1463
1464     if (s->avctx->strict_std_compliance >= FF_COMPLIANCE_STRICT &&
1465         !h->sps.bitstream_restriction_flag) {
1466         s->avctx->has_b_frames = MAX_DELAYED_PIC_COUNT - 1;
1467         s->low_delay           = 0;
1468     }
1469
1470     for (i = 0; 1; i++) {
1471         if(i == MAX_DELAYED_PIC_COUNT || cur->poc < h->last_pocs[i]){
1472             if(i)
1473                 h->last_pocs[i-1] = cur->poc;
1474             break;
1475         } else if(i) {
1476             h->last_pocs[i-1]= h->last_pocs[i];
1477         }
1478     }
1479     out_of_order = MAX_DELAYED_PIC_COUNT - i;
1480     if(   cur->f.pict_type == AV_PICTURE_TYPE_B
1481        || (h->last_pocs[MAX_DELAYED_PIC_COUNT-2] > INT_MIN && h->last_pocs[MAX_DELAYED_PIC_COUNT-1] - h->last_pocs[MAX_DELAYED_PIC_COUNT-2] > 2))
1482         out_of_order = FFMAX(out_of_order, 1);
1483     if(s->avctx->has_b_frames < out_of_order && !h->sps.bitstream_restriction_flag){
1484         av_log(s->avctx, AV_LOG_VERBOSE, "Increasing reorder buffer to %d\n", out_of_order);
1485         s->avctx->has_b_frames = out_of_order;
1486         s->low_delay = 0;
1487     }
1488
1489     pics = 0;
1490     while (h->delayed_pic[pics])
1491         pics++;
1492
1493     av_assert0(pics <= MAX_DELAYED_PIC_COUNT);
1494
1495     h->delayed_pic[pics++] = cur;
1496     if (cur->f.reference == 0)
1497         cur->f.reference = DELAYED_PIC_REF;
1498
1499     out = h->delayed_pic[0];
1500     out_idx = 0;
1501     for (i = 1; h->delayed_pic[i] &&
1502                 !h->delayed_pic[i]->f.key_frame &&
1503                 !h->delayed_pic[i]->mmco_reset;
1504          i++)
1505         if (h->delayed_pic[i]->poc < out->poc) {
1506             out     = h->delayed_pic[i];
1507             out_idx = i;
1508         }
1509     if (s->avctx->has_b_frames == 0 &&
1510         (h->delayed_pic[0]->f.key_frame || h->delayed_pic[0]->mmco_reset))
1511         h->next_outputed_poc = INT_MIN;
1512     out_of_order = out->poc < h->next_outputed_poc;
1513
1514     if (out_of_order || pics > s->avctx->has_b_frames) {
1515         out->f.reference &= ~DELAYED_PIC_REF;
1516         // for frame threading, the owner must be the second field's thread or
1517         // else the first thread can release the picture and reuse it unsafely
1518         out->owner2       = s;
1519         for (i = out_idx; h->delayed_pic[i]; i++)
1520             h->delayed_pic[i] = h->delayed_pic[i + 1];
1521     }
1522     if (!out_of_order && pics > s->avctx->has_b_frames) {
1523         h->next_output_pic = out;
1524         if (out_idx == 0 && h->delayed_pic[0] && (h->delayed_pic[0]->f.key_frame || h->delayed_pic[0]->mmco_reset)) {
1525             h->next_outputed_poc = INT_MIN;
1526         } else
1527             h->next_outputed_poc = out->poc;
1528     } else {
1529         av_log(s->avctx, AV_LOG_DEBUG, "no picture %s\n", out_of_order ? "ooo" : "");
1530     }
1531
1532     if (h->next_output_pic && h->next_output_pic->sync) {
1533         h->sync |= 2;
1534     }
1535
1536     if (setup_finished)
1537         ff_thread_finish_setup(s->avctx);
1538 }
1539
1540 static av_always_inline void backup_mb_border(H264Context *h, uint8_t *src_y,
1541                                               uint8_t *src_cb, uint8_t *src_cr,
1542                                               int linesize, int uvlinesize,
1543                                               int simple)
1544 {
1545     MpegEncContext *const s = &h->s;
1546     uint8_t *top_border;
1547     int top_idx = 1;
1548     const int pixel_shift = h->pixel_shift;
1549     int chroma444 = CHROMA444;
1550     int chroma422 = CHROMA422;
1551
1552     src_y  -= linesize;
1553     src_cb -= uvlinesize;
1554     src_cr -= uvlinesize;
1555
1556     if (!simple && FRAME_MBAFF) {
1557         if (s->mb_y & 1) {
1558             if (!MB_MBAFF) {
1559                 top_border = h->top_borders[0][s->mb_x];
1560                 AV_COPY128(top_border, src_y + 15 * linesize);
1561                 if (pixel_shift)
1562                     AV_COPY128(top_border + 16, src_y + 15 * linesize + 16);
1563                 if (simple || !CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
1564                     if (chroma444) {
1565                         if (pixel_shift) {
1566                             AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
1567                             AV_COPY128(top_border + 48, src_cb + 15 * uvlinesize + 16);
1568                             AV_COPY128(top_border + 64, src_cr + 15 * uvlinesize);
1569                             AV_COPY128(top_border + 80, src_cr + 15 * uvlinesize + 16);
1570                         } else {
1571                             AV_COPY128(top_border + 16, src_cb + 15 * uvlinesize);
1572                             AV_COPY128(top_border + 32, src_cr + 15 * uvlinesize);
1573                         }
1574                     } else if (chroma422) {
1575                         if (pixel_shift) {
1576                             AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
1577                             AV_COPY128(top_border + 48, src_cr + 15 * uvlinesize);
1578                         } else {
1579                             AV_COPY64(top_border + 16, src_cb + 15 * uvlinesize);
1580                             AV_COPY64(top_border + 24, src_cr + 15 * uvlinesize);
1581                         }
1582                     } else {
1583                         if (pixel_shift) {
1584                             AV_COPY128(top_border + 32, src_cb + 7 * uvlinesize);
1585                             AV_COPY128(top_border + 48, src_cr + 7 * uvlinesize);
1586                         } else {
1587                             AV_COPY64(top_border + 16, src_cb + 7 * uvlinesize);
1588                             AV_COPY64(top_border + 24, src_cr + 7 * uvlinesize);
1589                         }
1590                     }
1591                 }
1592             }
1593         } else if (MB_MBAFF) {
1594             top_idx = 0;
1595         } else
1596             return;
1597     }
1598
1599     top_border = h->top_borders[top_idx][s->mb_x];
1600     /* There are two lines saved, the line above the top macroblock
1601      * of a pair, and the line above the bottom macroblock. */
1602     AV_COPY128(top_border, src_y + 16 * linesize);
1603     if (pixel_shift)
1604         AV_COPY128(top_border + 16, src_y + 16 * linesize + 16);
1605
1606     if (simple || !CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
1607         if (chroma444) {
1608             if (pixel_shift) {
1609                 AV_COPY128(top_border + 32, src_cb + 16 * linesize);
1610                 AV_COPY128(top_border + 48, src_cb + 16 * linesize + 16);
1611                 AV_COPY128(top_border + 64, src_cr + 16 * linesize);
1612                 AV_COPY128(top_border + 80, src_cr + 16 * linesize + 16);
1613             } else {
1614                 AV_COPY128(top_border + 16, src_cb + 16 * linesize);
1615                 AV_COPY128(top_border + 32, src_cr + 16 * linesize);
1616             }
1617         } else if (chroma422) {
1618             if (pixel_shift) {
1619                 AV_COPY128(top_border + 32, src_cb + 16 * uvlinesize);
1620                 AV_COPY128(top_border + 48, src_cr + 16 * uvlinesize);
1621             } else {
1622                 AV_COPY64(top_border + 16, src_cb + 16 * uvlinesize);
1623                 AV_COPY64(top_border + 24, src_cr + 16 * uvlinesize);
1624             }
1625         } else {
1626             if (pixel_shift) {
1627                 AV_COPY128(top_border + 32, src_cb + 8 * uvlinesize);
1628                 AV_COPY128(top_border + 48, src_cr + 8 * uvlinesize);
1629             } else {
1630                 AV_COPY64(top_border + 16, src_cb + 8 * uvlinesize);
1631                 AV_COPY64(top_border + 24, src_cr + 8 * uvlinesize);
1632             }
1633         }
1634     }
1635 }
1636
1637 static av_always_inline void xchg_mb_border(H264Context *h, uint8_t *src_y,
1638                                             uint8_t *src_cb, uint8_t *src_cr,
1639                                             int linesize, int uvlinesize,
1640                                             int xchg, int chroma444,
1641                                             int simple, int pixel_shift)
1642 {
1643     MpegEncContext *const s = &h->s;
1644     int deblock_topleft;
1645     int deblock_top;
1646     int top_idx = 1;
1647     uint8_t *top_border_m1;
1648     uint8_t *top_border;
1649
1650     if (!simple && FRAME_MBAFF) {
1651         if (s->mb_y & 1) {
1652             if (!MB_MBAFF)
1653                 return;
1654         } else {
1655             top_idx = MB_MBAFF ? 0 : 1;
1656         }
1657     }
1658
1659     if (h->deblocking_filter == 2) {
1660         deblock_topleft = h->slice_table[h->mb_xy - 1 - s->mb_stride] == h->slice_num;
1661         deblock_top     = h->top_type;
1662     } else {
1663         deblock_topleft = (s->mb_x > 0);
1664         deblock_top     = (s->mb_y > !!MB_FIELD);
1665     }
1666
1667     src_y  -= linesize   + 1 + pixel_shift;
1668     src_cb -= uvlinesize + 1 + pixel_shift;
1669     src_cr -= uvlinesize + 1 + pixel_shift;
1670
1671     top_border_m1 = h->top_borders[top_idx][s->mb_x - 1];
1672     top_border    = h->top_borders[top_idx][s->mb_x];
1673
1674 #define XCHG(a, b, xchg)                        \
1675     if (pixel_shift) {                          \
1676         if (xchg) {                             \
1677             AV_SWAP64(b + 0, a + 0);            \
1678             AV_SWAP64(b + 8, a + 8);            \
1679         } else {                                \
1680             AV_COPY128(b, a);                   \
1681         }                                       \
1682     } else if (xchg)                            \
1683         AV_SWAP64(b, a);                        \
1684     else                                        \
1685         AV_COPY64(b, a);
1686
1687     if (deblock_top) {
1688         if (deblock_topleft) {
1689             XCHG(top_border_m1 + (8 << pixel_shift),
1690                  src_y - (7 << pixel_shift), 1);
1691         }
1692         XCHG(top_border + (0 << pixel_shift), src_y + (1 << pixel_shift), xchg);
1693         XCHG(top_border + (8 << pixel_shift), src_y + (9 << pixel_shift), 1);
1694         if (s->mb_x + 1 < s->mb_width) {
1695             XCHG(h->top_borders[top_idx][s->mb_x + 1],
1696                  src_y + (17 << pixel_shift), 1);
1697         }
1698     }
1699     if (simple || !CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
1700         if (chroma444) {
1701             if (deblock_topleft) {
1702                 XCHG(top_border_m1 + (24 << pixel_shift), src_cb - (7 << pixel_shift), 1);
1703                 XCHG(top_border_m1 + (40 << pixel_shift), src_cr - (7 << pixel_shift), 1);
1704             }
1705             XCHG(top_border + (16 << pixel_shift), src_cb + (1 << pixel_shift), xchg);
1706             XCHG(top_border + (24 << pixel_shift), src_cb + (9 << pixel_shift), 1);
1707             XCHG(top_border + (32 << pixel_shift), src_cr + (1 << pixel_shift), xchg);
1708             XCHG(top_border + (40 << pixel_shift), src_cr + (9 << pixel_shift), 1);
1709             if (s->mb_x + 1 < s->mb_width) {
1710                 XCHG(h->top_borders[top_idx][s->mb_x + 1] + (16 << pixel_shift), src_cb + (17 << pixel_shift), 1);
1711                 XCHG(h->top_borders[top_idx][s->mb_x + 1] + (32 << pixel_shift), src_cr + (17 << pixel_shift), 1);
1712             }
1713         } else {
1714             if (deblock_top) {
1715                 if (deblock_topleft) {
1716                     XCHG(top_border_m1 + (16 << pixel_shift), src_cb - (7 << pixel_shift), 1);
1717                     XCHG(top_border_m1 + (24 << pixel_shift), src_cr - (7 << pixel_shift), 1);
1718                 }
1719                 XCHG(top_border + (16 << pixel_shift), src_cb + 1 + pixel_shift, 1);
1720                 XCHG(top_border + (24 << pixel_shift), src_cr + 1 + pixel_shift, 1);
1721             }
1722         }
1723     }
1724 }
1725
1726 static av_always_inline int dctcoef_get(DCTELEM *mb, int high_bit_depth,
1727                                         int index)
1728 {
1729     if (high_bit_depth) {
1730         return AV_RN32A(((int32_t *)mb) + index);
1731     } else
1732         return AV_RN16A(mb + index);
1733 }
1734
1735 static av_always_inline void dctcoef_set(DCTELEM *mb, int high_bit_depth,
1736                                          int index, int value)
1737 {
1738     if (high_bit_depth) {
1739         AV_WN32A(((int32_t *)mb) + index, value);
1740     } else
1741         AV_WN16A(mb + index, value);
1742 }
1743
1744 static av_always_inline void hl_decode_mb_predict_luma(H264Context *h,
1745                                                        int mb_type, int is_h264,
1746                                                        int simple,
1747                                                        int transform_bypass,
1748                                                        int pixel_shift,
1749                                                        int *block_offset,
1750                                                        int linesize,
1751                                                        uint8_t *dest_y, int p)
1752 {
1753     MpegEncContext *const s = &h->s;
1754     void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1755     void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
1756     int i;
1757     int qscale = p == 0 ? s->qscale : h->chroma_qp[p - 1];
1758     block_offset += 16 * p;
1759     if (IS_INTRA4x4(mb_type)) {
1760         if (simple || !s->encoding) {
1761             if (IS_8x8DCT(mb_type)) {
1762                 if (transform_bypass) {
1763                     idct_dc_add  =
1764                     idct_add     = s->dsp.add_pixels8;
1765                 } else {
1766                     idct_dc_add = h->h264dsp.h264_idct8_dc_add;
1767                     idct_add    = h->h264dsp.h264_idct8_add;
1768                 }
1769                 for (i = 0; i < 16; i += 4) {
1770                     uint8_t *const ptr = dest_y + block_offset[i];
1771                     const int dir      = h->intra4x4_pred_mode_cache[scan8[i]];
1772                     if (transform_bypass && h->sps.profile_idc == 244 && dir <= 1) {
1773                         h->hpc.pred8x8l_add[dir](ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
1774                     } else {
1775                         const int nnz = h->non_zero_count_cache[scan8[i + p * 16]];
1776                         h->hpc.pred8x8l[dir](ptr, (h->topleft_samples_available << i) & 0x8000,
1777                                              (h->topright_samples_available << i) & 0x4000, linesize);
1778                         if (nnz) {
1779                             if (nnz == 1 && dctcoef_get(h->mb, pixel_shift, i * 16 + p * 256))
1780                                 idct_dc_add(ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
1781                             else
1782                                 idct_add(ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
1783                         }
1784                     }
1785                 }
1786             } else {
1787                 if (transform_bypass) {
1788                     idct_dc_add  =
1789                         idct_add = s->dsp.add_pixels4;
1790                 } else {
1791                     idct_dc_add = h->h264dsp.h264_idct_dc_add;
1792                     idct_add    = h->h264dsp.h264_idct_add;
1793                 }
1794                 for (i = 0; i < 16; i++) {
1795                     uint8_t *const ptr = dest_y + block_offset[i];
1796                     const int dir      = h->intra4x4_pred_mode_cache[scan8[i]];
1797
1798                     if (transform_bypass && h->sps.profile_idc == 244 && dir <= 1) {
1799                         h->hpc.pred4x4_add[dir](ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
1800                     } else {
1801                         uint8_t *topright;
1802                         int nnz, tr;
1803                         uint64_t tr_high;
1804                         if (dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED) {
1805                             const int topright_avail = (h->topright_samples_available << i) & 0x8000;
1806                             av_assert2(s->mb_y || linesize <= block_offset[i]);
1807                             if (!topright_avail) {
1808                                 if (pixel_shift) {
1809                                     tr_high  = ((uint16_t *)ptr)[3 - linesize / 2] * 0x0001000100010001ULL;
1810                                     topright = (uint8_t *)&tr_high;
1811                                 } else {
1812                                     tr       = ptr[3 - linesize] * 0x01010101u;
1813                                     topright = (uint8_t *)&tr;
1814                                 }
1815                             } else
1816                                 topright = ptr + (4 << pixel_shift) - linesize;
1817                         } else
1818                             topright = NULL;
1819
1820                         h->hpc.pred4x4[dir](ptr, topright, linesize);
1821                         nnz = h->non_zero_count_cache[scan8[i + p * 16]];
1822                         if (nnz) {
1823                             if (is_h264) {
1824                                 if (nnz == 1 && dctcoef_get(h->mb, pixel_shift, i * 16 + p * 256))
1825                                     idct_dc_add(ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
1826                                 else
1827                                     idct_add(ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
1828                             } else if (CONFIG_SVQ3_DECODER)
1829                                 ff_svq3_add_idct_c(ptr, h->mb + i * 16 + p * 256, linesize, qscale, 0);
1830                         }
1831                     }
1832                 }
1833             }
1834         }
1835     } else {
1836         h->hpc.pred16x16[h->intra16x16_pred_mode](dest_y, linesize);
1837         if (is_h264) {
1838             if (h->non_zero_count_cache[scan8[LUMA_DC_BLOCK_INDEX + p]]) {
1839                 if (!transform_bypass)
1840                     h->h264dsp.h264_luma_dc_dequant_idct(h->mb + (p * 256 << pixel_shift),
1841                                                          h->mb_luma_dc[p],
1842                                                          h->dequant4_coeff[p][qscale][0]);
1843                 else {
1844                     static const uint8_t dc_mapping[16] = {
1845                          0 * 16,  1 * 16,  4 * 16,  5 * 16,
1846                          2 * 16,  3 * 16,  6 * 16,  7 * 16,
1847                          8 * 16,  9 * 16, 12 * 16, 13 * 16,
1848                         10 * 16, 11 * 16, 14 * 16, 15 * 16 };
1849                     for (i = 0; i < 16; i++)
1850                         dctcoef_set(h->mb + (p * 256 << pixel_shift),
1851                                     pixel_shift, dc_mapping[i],
1852                                     dctcoef_get(h->mb_luma_dc[p],
1853                                                 pixel_shift, i));
1854                 }
1855             }
1856         } else if (CONFIG_SVQ3_DECODER)
1857             ff_svq3_luma_dc_dequant_idct_c(h->mb + p * 256,
1858                                            h->mb_luma_dc[p], qscale);
1859     }
1860 }
1861
1862 static av_always_inline void hl_decode_mb_idct_luma(H264Context *h, int mb_type,
1863                                                     int is_h264, int simple,
1864                                                     int transform_bypass,
1865                                                     int pixel_shift,
1866                                                     int *block_offset,
1867                                                     int linesize,
1868                                                     uint8_t *dest_y, int p)
1869 {
1870     MpegEncContext *const s = &h->s;
1871     void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1872     int i;
1873     block_offset += 16 * p;
1874     if (!IS_INTRA4x4(mb_type)) {
1875         if (is_h264) {
1876             if (IS_INTRA16x16(mb_type)) {
1877                 if (transform_bypass) {
1878                     if (h->sps.profile_idc == 244 &&
1879                         (h->intra16x16_pred_mode == VERT_PRED8x8 ||
1880                          h->intra16x16_pred_mode == HOR_PRED8x8)) {
1881                         h->hpc.pred16x16_add[h->intra16x16_pred_mode](dest_y, block_offset,
1882                                                                       h->mb + (p * 256 << pixel_shift),
1883                                                                       linesize);
1884                     } else {
1885                         for (i = 0; i < 16; i++)
1886                             if (h->non_zero_count_cache[scan8[i + p * 16]] ||
1887                                 dctcoef_get(h->mb, pixel_shift, i * 16 + p * 256))
1888                                 s->dsp.add_pixels4(dest_y + block_offset[i],
1889                                                    h->mb + (i * 16 + p * 256 << pixel_shift),
1890                                                    linesize);
1891                     }
1892                 } else {
1893                     h->h264dsp.h264_idct_add16intra(dest_y, block_offset,
1894                                                     h->mb + (p * 256 << pixel_shift),
1895                                                     linesize,
1896                                                     h->non_zero_count_cache + p * 5 * 8);
1897                 }
1898             } else if (h->cbp & 15) {
1899                 if (transform_bypass) {
1900                     const int di = IS_8x8DCT(mb_type) ? 4 : 1;
1901                     idct_add = IS_8x8DCT(mb_type) ? s->dsp.add_pixels8
1902                                                   : s->dsp.add_pixels4;
1903                     for (i = 0; i < 16; i += di)
1904                         if (h->non_zero_count_cache[scan8[i + p * 16]])
1905                             idct_add(dest_y + block_offset[i],
1906                                      h->mb + (i * 16 + p * 256 << pixel_shift),
1907                                      linesize);
1908                 } else {
1909                     if (IS_8x8DCT(mb_type))
1910                         h->h264dsp.h264_idct8_add4(dest_y, block_offset,
1911                                                    h->mb + (p * 256 << pixel_shift),
1912                                                    linesize,
1913                                                    h->non_zero_count_cache + p * 5 * 8);
1914                     else
1915                         h->h264dsp.h264_idct_add16(dest_y, block_offset,
1916                                                    h->mb + (p * 256 << pixel_shift),
1917                                                    linesize,
1918                                                    h->non_zero_count_cache + p * 5 * 8);
1919                 }
1920             }
1921         } else if (CONFIG_SVQ3_DECODER) {
1922             for (i = 0; i < 16; i++)
1923                 if (h->non_zero_count_cache[scan8[i + p * 16]] || h->mb[i * 16 + p * 256]) {
1924                     // FIXME benchmark weird rule, & below
1925                     uint8_t *const ptr = dest_y + block_offset[i];
1926                     ff_svq3_add_idct_c(ptr, h->mb + i * 16 + p * 256, linesize,
1927                                        s->qscale, IS_INTRA(mb_type) ? 1 : 0);
1928                 }
1929         }
1930     }
1931 }
1932
1933 #define BITS   8
1934 #define SIMPLE 1
1935 #include "h264_mb_template.c"
1936
1937 #undef  BITS
1938 #define BITS   16
1939 #include "h264_mb_template.c"
1940
1941 #undef  SIMPLE
1942 #define SIMPLE 0
1943 #include "h264_mb_template.c"
1944
1945 void ff_h264_hl_decode_mb(H264Context *h)
1946 {
1947     MpegEncContext *const s = &h->s;
1948     const int mb_xy   = h->mb_xy;
1949     const int mb_type = s->current_picture.f.mb_type[mb_xy];
1950     int is_complex    = CONFIG_SMALL || h->is_complex || IS_INTRA_PCM(mb_type) || s->qscale == 0;
1951
1952     if (CHROMA444) {
1953         if (is_complex || h->pixel_shift)
1954             hl_decode_mb_444_complex(h);
1955         else
1956             hl_decode_mb_444_simple_8(h);
1957     } else if (is_complex) {
1958         hl_decode_mb_complex(h);
1959     } else if (h->pixel_shift) {
1960         hl_decode_mb_simple_16(h);
1961     } else
1962         hl_decode_mb_simple_8(h);
1963 }
1964
1965 static int pred_weight_table(H264Context *h)
1966 {
1967     MpegEncContext *const s = &h->s;
1968     int list, i;
1969     int luma_def, chroma_def;
1970
1971     h->use_weight             = 0;
1972     h->use_weight_chroma      = 0;
1973     h->luma_log2_weight_denom = get_ue_golomb(&s->gb);
1974     if (h->sps.chroma_format_idc)
1975         h->chroma_log2_weight_denom = get_ue_golomb(&s->gb);
1976     luma_def   = 1 << h->luma_log2_weight_denom;
1977     chroma_def = 1 << h->chroma_log2_weight_denom;
1978
1979     for (list = 0; list < 2; list++) {
1980         h->luma_weight_flag[list]   = 0;
1981         h->chroma_weight_flag[list] = 0;
1982         for (i = 0; i < h->ref_count[list]; i++) {
1983             int luma_weight_flag, chroma_weight_flag;
1984
1985             luma_weight_flag = get_bits1(&s->gb);
1986             if (luma_weight_flag) {
1987                 h->luma_weight[i][list][0] = get_se_golomb(&s->gb);
1988                 h->luma_weight[i][list][1] = get_se_golomb(&s->gb);
1989                 if (h->luma_weight[i][list][0] != luma_def ||
1990                     h->luma_weight[i][list][1] != 0) {
1991                     h->use_weight             = 1;
1992                     h->luma_weight_flag[list] = 1;
1993                 }
1994             } else {
1995                 h->luma_weight[i][list][0] = luma_def;
1996                 h->luma_weight[i][list][1] = 0;
1997             }
1998
1999             if (h->sps.chroma_format_idc) {
2000                 chroma_weight_flag = get_bits1(&s->gb);
2001                 if (chroma_weight_flag) {
2002                     int j;
2003                     for (j = 0; j < 2; j++) {
2004                         h->chroma_weight[i][list][j][0] = get_se_golomb(&s->gb);
2005                         h->chroma_weight[i][list][j][1] = get_se_golomb(&s->gb);
2006                         if (h->chroma_weight[i][list][j][0] != chroma_def ||
2007                             h->chroma_weight[i][list][j][1] != 0) {
2008                             h->use_weight_chroma = 1;
2009                             h->chroma_weight_flag[list] = 1;
2010                         }
2011                     }
2012                 } else {
2013                     int j;
2014                     for (j = 0; j < 2; j++) {
2015                         h->chroma_weight[i][list][j][0] = chroma_def;
2016                         h->chroma_weight[i][list][j][1] = 0;
2017                     }
2018                 }
2019             }
2020         }
2021         if (h->slice_type_nos != AV_PICTURE_TYPE_B)
2022             break;
2023     }
2024     h->use_weight = h->use_weight || h->use_weight_chroma;
2025     return 0;
2026 }
2027
2028 /**
2029  * Initialize implicit_weight table.
2030  * @param field  0/1 initialize the weight for interlaced MBAFF
2031  *                -1 initializes the rest
2032  */
2033 static void implicit_weight_table(H264Context *h, int field)
2034 {
2035     MpegEncContext *const s = &h->s;
2036     int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
2037
2038     for (i = 0; i < 2; i++) {
2039         h->luma_weight_flag[i]   = 0;
2040         h->chroma_weight_flag[i] = 0;
2041     }
2042
2043     if (field < 0) {
2044         if (s->picture_structure == PICT_FRAME) {
2045             cur_poc = s->current_picture_ptr->poc;
2046         } else {
2047             cur_poc = s->current_picture_ptr->field_poc[s->picture_structure - 1];
2048         }
2049         if (h->ref_count[0] == 1 && h->ref_count[1] == 1 && !FRAME_MBAFF &&
2050             h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2 * cur_poc) {
2051             h->use_weight = 0;
2052             h->use_weight_chroma = 0;
2053             return;
2054         }
2055         ref_start  = 0;
2056         ref_count0 = h->ref_count[0];
2057         ref_count1 = h->ref_count[1];
2058     } else {
2059         cur_poc    = s->current_picture_ptr->field_poc[field];
2060         ref_start  = 16;
2061         ref_count0 = 16 + 2 * h->ref_count[0];
2062         ref_count1 = 16 + 2 * h->ref_count[1];
2063     }
2064
2065     h->use_weight               = 2;
2066     h->use_weight_chroma        = 2;
2067     h->luma_log2_weight_denom   = 5;
2068     h->chroma_log2_weight_denom = 5;
2069
2070     for (ref0 = ref_start; ref0 < ref_count0; ref0++) {
2071         int poc0 = h->ref_list[0][ref0].poc;
2072         for (ref1 = ref_start; ref1 < ref_count1; ref1++) {
2073             int w = 32;
2074             if (!h->ref_list[0][ref0].long_ref && !h->ref_list[1][ref1].long_ref) {
2075                 int poc1 = h->ref_list[1][ref1].poc;
2076                 int td   = av_clip(poc1 - poc0, -128, 127);
2077                 if (td) {
2078                     int tb = av_clip(cur_poc - poc0, -128, 127);
2079                     int tx = (16384 + (FFABS(td) >> 1)) / td;
2080                     int dist_scale_factor = (tb * tx + 32) >> 8;
2081                     if (dist_scale_factor >= -64 && dist_scale_factor <= 128)
2082                         w = 64 - dist_scale_factor;
2083                 }
2084             }
2085             if (field < 0) {
2086                 h->implicit_weight[ref0][ref1][0] =
2087                 h->implicit_weight[ref0][ref1][1] = w;
2088             } else {
2089                 h->implicit_weight[ref0][ref1][field] = w;
2090             }
2091         }
2092     }
2093 }
2094
2095 /**
2096  * instantaneous decoder refresh.
2097  */
2098 static void idr(H264Context *h)
2099 {
2100     int i;
2101     ff_h264_remove_all_refs(h);
2102     h->prev_frame_num        = 0;
2103     h->prev_frame_num_offset = 0;
2104     h->prev_poc_msb          = 1<<16;
2105     h->prev_poc_lsb          = 0;
2106     for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
2107         h->last_pocs[i] = INT_MIN;
2108 }
2109
2110 /* forget old pics after a seek */
2111 static void flush_dpb(AVCodecContext *avctx)
2112 {
2113     H264Context *h = avctx->priv_data;
2114     int i;
2115     for (i=0; i<=MAX_DELAYED_PIC_COUNT; i++) {
2116         if (h->delayed_pic[i])
2117             h->delayed_pic[i]->f.reference = 0;
2118         h->delayed_pic[i] = NULL;
2119     }
2120     h->outputed_poc = h->next_outputed_poc = INT_MIN;
2121     h->prev_interlaced_frame = 1;
2122     idr(h);
2123     h->prev_frame_num = -1;
2124     if (h->s.current_picture_ptr)
2125         h->s.current_picture_ptr->f.reference = 0;
2126     h->s.first_field = 0;
2127     ff_h264_reset_sei(h);
2128     ff_mpeg_flush(avctx);
2129     h->recovery_frame= -1;
2130     h->sync= 0;
2131 }
2132
2133 static int init_poc(H264Context *h)
2134 {
2135     MpegEncContext *const s = &h->s;
2136     const int max_frame_num = 1 << h->sps.log2_max_frame_num;
2137     int field_poc[2];
2138     Picture *cur = s->current_picture_ptr;
2139
2140     h->frame_num_offset = h->prev_frame_num_offset;
2141     if (h->frame_num < h->prev_frame_num)
2142         h->frame_num_offset += max_frame_num;
2143
2144     if (h->sps.poc_type == 0) {
2145         const int max_poc_lsb = 1 << h->sps.log2_max_poc_lsb;
2146
2147         if (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb / 2)
2148             h->poc_msb = h->prev_poc_msb + max_poc_lsb;
2149         else if (h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb / 2)
2150             h->poc_msb = h->prev_poc_msb - max_poc_lsb;
2151         else
2152             h->poc_msb = h->prev_poc_msb;
2153         field_poc[0] =
2154         field_poc[1] = h->poc_msb + h->poc_lsb;
2155         if (s->picture_structure == PICT_FRAME)
2156             field_poc[1] += h->delta_poc_bottom;
2157     } else if (h->sps.poc_type == 1) {
2158         int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
2159         int i;
2160
2161         if (h->sps.poc_cycle_length != 0)
2162             abs_frame_num = h->frame_num_offset + h->frame_num;
2163         else
2164             abs_frame_num = 0;
2165
2166         if (h->nal_ref_idc == 0 && abs_frame_num > 0)
2167             abs_frame_num--;
2168
2169         expected_delta_per_poc_cycle = 0;
2170         for (i = 0; i < h->sps.poc_cycle_length; i++)
2171             // FIXME integrate during sps parse
2172             expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[i];
2173
2174         if (abs_frame_num > 0) {
2175             int poc_cycle_cnt          = (abs_frame_num - 1) / h->sps.poc_cycle_length;
2176             int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
2177
2178             expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
2179             for (i = 0; i <= frame_num_in_poc_cycle; i++)
2180                 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[i];
2181         } else
2182             expectedpoc = 0;
2183
2184         if (h->nal_ref_idc == 0)
2185             expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
2186
2187         field_poc[0] = expectedpoc + h->delta_poc[0];
2188         field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
2189
2190         if (s->picture_structure == PICT_FRAME)
2191             field_poc[1] += h->delta_poc[1];
2192     } else {
2193         int poc = 2 * (h->frame_num_offset + h->frame_num);
2194
2195         if (!h->nal_ref_idc)
2196             poc--;
2197
2198         field_poc[0] = poc;
2199         field_poc[1] = poc;
2200     }
2201
2202     if (s->picture_structure != PICT_BOTTOM_FIELD)
2203         s->current_picture_ptr->field_poc[0] = field_poc[0];
2204     if (s->picture_structure != PICT_TOP_FIELD)
2205         s->current_picture_ptr->field_poc[1] = field_poc[1];
2206     cur->poc = FFMIN(cur->field_poc[0], cur->field_poc[1]);
2207
2208     return 0;
2209 }
2210
2211 /**
2212  * initialize scan tables
2213  */
2214 static void init_scan_tables(H264Context *h)
2215 {
2216     int i;
2217     for (i = 0; i < 16; i++) {
2218 #define T(x) (x >> 2) | ((x << 2) & 0xF)
2219         h->zigzag_scan[i] = T(zigzag_scan[i]);
2220         h->field_scan[i]  = T(field_scan[i]);
2221 #undef T
2222     }
2223     for (i = 0; i < 64; i++) {
2224 #define T(x) (x >> 3) | ((x & 7) << 3)
2225         h->zigzag_scan8x8[i]       = T(ff_zigzag_direct[i]);
2226         h->zigzag_scan8x8_cavlc[i] = T(zigzag_scan8x8_cavlc[i]);
2227         h->field_scan8x8[i]        = T(field_scan8x8[i]);
2228         h->field_scan8x8_cavlc[i]  = T(field_scan8x8_cavlc[i]);
2229 #undef T
2230     }
2231     if (h->sps.transform_bypass) { // FIXME same ugly
2232         memcpy(h->zigzag_scan_q0          , zigzag_scan             , sizeof(h->zigzag_scan_q0         ));
2233         memcpy(h->zigzag_scan8x8_q0       , ff_zigzag_direct        , sizeof(h->zigzag_scan8x8_q0      ));
2234         memcpy(h->zigzag_scan8x8_cavlc_q0 , zigzag_scan8x8_cavlc    , sizeof(h->zigzag_scan8x8_cavlc_q0));
2235         memcpy(h->field_scan_q0           , field_scan              , sizeof(h->field_scan_q0          ));
2236         memcpy(h->field_scan8x8_q0        , field_scan8x8           , sizeof(h->field_scan8x8_q0       ));
2237         memcpy(h->field_scan8x8_cavlc_q0  , field_scan8x8_cavlc     , sizeof(h->field_scan8x8_cavlc_q0 ));
2238     } else {
2239         memcpy(h->zigzag_scan_q0          , h->zigzag_scan          , sizeof(h->zigzag_scan_q0         ));
2240         memcpy(h->zigzag_scan8x8_q0       , h->zigzag_scan8x8       , sizeof(h->zigzag_scan8x8_q0      ));
2241         memcpy(h->zigzag_scan8x8_cavlc_q0 , h->zigzag_scan8x8_cavlc , sizeof(h->zigzag_scan8x8_cavlc_q0));
2242         memcpy(h->field_scan_q0           , h->field_scan           , sizeof(h->field_scan_q0          ));
2243         memcpy(h->field_scan8x8_q0        , h->field_scan8x8        , sizeof(h->field_scan8x8_q0       ));
2244         memcpy(h->field_scan8x8_cavlc_q0  , h->field_scan8x8_cavlc  , sizeof(h->field_scan8x8_cavlc_q0 ));
2245     }
2246 }
2247
2248 static int field_end(H264Context *h, int in_setup)
2249 {
2250     MpegEncContext *const s     = &h->s;
2251     AVCodecContext *const avctx = s->avctx;
2252     int err = 0;
2253     s->mb_y = 0;
2254
2255     if (!in_setup && !s->droppable)
2256         ff_thread_report_progress(&s->current_picture_ptr->f, INT_MAX,
2257                                   s->picture_structure == PICT_BOTTOM_FIELD);
2258
2259     if (CONFIG_H264_VDPAU_DECODER &&
2260         s->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU)
2261         ff_vdpau_h264_set_reference_frames(s);
2262
2263     if (in_setup || !(avctx->active_thread_type & FF_THREAD_FRAME)) {
2264         if (!s->droppable) {
2265             err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
2266             h->prev_poc_msb = h->poc_msb;
2267             h->prev_poc_lsb = h->poc_lsb;
2268         }
2269         h->prev_frame_num_offset = h->frame_num_offset;
2270         h->prev_frame_num        = h->frame_num;
2271         h->outputed_poc          = h->next_outputed_poc;
2272     }
2273
2274     if (avctx->hwaccel) {
2275         if (avctx->hwaccel->end_frame(avctx) < 0)
2276             av_log(avctx, AV_LOG_ERROR,
2277                    "hardware accelerator failed to decode picture\n");
2278     }
2279
2280     if (CONFIG_H264_VDPAU_DECODER &&
2281         s->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU)
2282         ff_vdpau_h264_picture_complete(s);
2283
2284     /*
2285      * FIXME: Error handling code does not seem to support interlaced
2286      * when slices span multiple rows
2287      * The ff_er_add_slice calls don't work right for bottom
2288      * fields; they cause massive erroneous error concealing
2289      * Error marking covers both fields (top and bottom).
2290      * This causes a mismatched s->error_count
2291      * and a bad error table. Further, the error count goes to
2292      * INT_MAX when called for bottom field, because mb_y is
2293      * past end by one (callers fault) and resync_mb_y != 0
2294      * causes problems for the first MB line, too.
2295      */
2296     if (!FIELD_PICTURE)
2297         ff_er_frame_end(s);
2298
2299     ff_MPV_frame_end(s);
2300
2301     h->current_slice = 0;
2302
2303     return err;
2304 }
2305
2306 /**
2307  * Replicate H264 "master" context to thread contexts.
2308  */
2309 static void clone_slice(H264Context *dst, H264Context *src)
2310 {
2311     memcpy(dst->block_offset, src->block_offset, sizeof(dst->block_offset));
2312     dst->s.current_picture_ptr = src->s.current_picture_ptr;
2313     dst->s.current_picture     = src->s.current_picture;
2314     dst->s.linesize            = src->s.linesize;
2315     dst->s.uvlinesize          = src->s.uvlinesize;
2316     dst->s.first_field         = src->s.first_field;
2317
2318     dst->prev_poc_msb          = src->prev_poc_msb;
2319     dst->prev_poc_lsb          = src->prev_poc_lsb;
2320     dst->prev_frame_num_offset = src->prev_frame_num_offset;
2321     dst->prev_frame_num        = src->prev_frame_num;
2322     dst->short_ref_count       = src->short_ref_count;
2323
2324     memcpy(dst->short_ref,        src->short_ref,        sizeof(dst->short_ref));
2325     memcpy(dst->long_ref,         src->long_ref,         sizeof(dst->long_ref));
2326     memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
2327     memcpy(dst->ref_list,         src->ref_list,         sizeof(dst->ref_list));
2328
2329     memcpy(dst->dequant4_coeff,   src->dequant4_coeff,   sizeof(src->dequant4_coeff));
2330     memcpy(dst->dequant8_coeff,   src->dequant8_coeff,   sizeof(src->dequant8_coeff));
2331 }
2332
2333 /**
2334  * Compute profile from profile_idc and constraint_set?_flags.
2335  *
2336  * @param sps SPS
2337  *
2338  * @return profile as defined by FF_PROFILE_H264_*
2339  */
2340 int ff_h264_get_profile(SPS *sps)
2341 {
2342     int profile = sps->profile_idc;
2343
2344     switch (sps->profile_idc) {
2345     case FF_PROFILE_H264_BASELINE:
2346         // constraint_set1_flag set to 1
2347         profile |= (sps->constraint_set_flags & 1 << 1) ? FF_PROFILE_H264_CONSTRAINED : 0;
2348         break;
2349     case FF_PROFILE_H264_HIGH_10:
2350     case FF_PROFILE_H264_HIGH_422:
2351     case FF_PROFILE_H264_HIGH_444_PREDICTIVE:
2352         // constraint_set3_flag set to 1
2353         profile |= (sps->constraint_set_flags & 1 << 3) ? FF_PROFILE_H264_INTRA : 0;
2354         break;
2355     }
2356
2357     return profile;
2358 }
2359
2360 static int h264_set_parameter_from_sps(H264Context *h)
2361 {
2362     MpegEncContext *s = &h->s;
2363
2364     if (s->flags & CODEC_FLAG_LOW_DELAY ||
2365         (h->sps.bitstream_restriction_flag &&
2366          !h->sps.num_reorder_frames)) {
2367         if (s->avctx->has_b_frames > 1 || h->delayed_pic[0])
2368             av_log(h->s.avctx, AV_LOG_WARNING, "Delayed frames seen. "
2369                    "Reenabling low delay requires a codec flush.\n");
2370         else
2371             s->low_delay = 1;
2372     }
2373
2374     if (s->avctx->has_b_frames < 2)
2375         s->avctx->has_b_frames = !s->low_delay;
2376
2377     if (s->avctx->bits_per_raw_sample != h->sps.bit_depth_luma ||
2378         h->cur_chroma_format_idc      != h->sps.chroma_format_idc) {
2379         if (s->avctx->codec &&
2380             s->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU &&
2381             (h->sps.bit_depth_luma != 8 || h->sps.chroma_format_idc > 1)) {
2382             av_log(s->avctx, AV_LOG_ERROR,
2383                    "VDPAU decoding does not support video colorspace.\n");
2384             return AVERROR_INVALIDDATA;
2385         }
2386         if (h->sps.bit_depth_luma >= 8 && h->sps.bit_depth_luma <= 14 &&
2387             h->sps.bit_depth_luma != 11 && h->sps.bit_depth_luma != 13 &&
2388                 (h->sps.bit_depth_luma != 9 || !CHROMA422)) {
2389             s->avctx->bits_per_raw_sample = h->sps.bit_depth_luma;
2390             h->cur_chroma_format_idc      = h->sps.chroma_format_idc;
2391             h->pixel_shift                = h->sps.bit_depth_luma > 8;
2392
2393             ff_h264dsp_init(&h->h264dsp, h->sps.bit_depth_luma,
2394                             h->sps.chroma_format_idc);
2395             ff_h264_pred_init(&h->hpc, s->codec_id, h->sps.bit_depth_luma,
2396                               h->sps.chroma_format_idc);
2397             s->dsp.dct_bits = h->sps.bit_depth_luma > 8 ? 32 : 16;
2398             ff_dsputil_init(&s->dsp, s->avctx);
2399         } else {
2400             av_log(s->avctx, AV_LOG_ERROR, "Unsupported bit depth: %d\n",
2401                    h->sps.bit_depth_luma);
2402             return AVERROR_INVALIDDATA;
2403         }
2404     }
2405     return 0;
2406 }
2407
2408 /**
2409  * Decode a slice header.
2410  * This will also call ff_MPV_common_init() and frame_start() as needed.
2411  *
2412  * @param h h264context
2413  * @param h0 h264 master context (differs from 'h' when doing sliced based
2414  *           parallel decoding)
2415  *
2416  * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
2417  */
2418 static int decode_slice_header(H264Context *h, H264Context *h0)
2419 {
2420     MpegEncContext *const s  = &h->s;
2421     MpegEncContext *const s0 = &h0->s;
2422     unsigned int first_mb_in_slice;
2423     unsigned int pps_id;
2424     int num_ref_idx_active_override_flag, ret;
2425     unsigned int slice_type, tmp, i, j;
2426     int default_ref_list_done = 0;
2427     int last_pic_structure, last_pic_droppable;
2428     int must_reinit;
2429
2430     /* FIXME: 2tap qpel isn't implemented for high bit depth. */
2431     if ((s->avctx->flags2 & CODEC_FLAG2_FAST) &&
2432         !h->nal_ref_idc && !h->pixel_shift) {
2433         s->me.qpel_put = s->dsp.put_2tap_qpel_pixels_tab;
2434         s->me.qpel_avg = s->dsp.avg_2tap_qpel_pixels_tab;
2435     } else {
2436         s->me.qpel_put = s->dsp.put_h264_qpel_pixels_tab;
2437         s->me.qpel_avg = s->dsp.avg_h264_qpel_pixels_tab;
2438     }
2439
2440     first_mb_in_slice = get_ue_golomb_long(&s->gb);
2441
2442     if (first_mb_in_slice == 0) { // FIXME better field boundary detection
2443         if (h0->current_slice && FIELD_PICTURE) {
2444             field_end(h, 1);
2445         }
2446
2447         h0->current_slice = 0;
2448         if (!s0->first_field) {
2449             if (s->current_picture_ptr && !s->droppable &&
2450                 s->current_picture_ptr->owner2 == s) {
2451                 ff_thread_report_progress(&s->current_picture_ptr->f, INT_MAX,
2452                                           s->picture_structure == PICT_BOTTOM_FIELD);
2453             }
2454             s->current_picture_ptr = NULL;
2455         }
2456     }
2457
2458     slice_type = get_ue_golomb_31(&s->gb);
2459     if (slice_type > 9) {
2460         av_log(h->s.avctx, AV_LOG_ERROR,
2461                "slice type too large (%d) at %d %d\n",
2462                slice_type, s->mb_x, s->mb_y);
2463         return -1;
2464     }
2465     if (slice_type > 4) {
2466         slice_type -= 5;
2467         h->slice_type_fixed = 1;
2468     } else
2469         h->slice_type_fixed = 0;
2470
2471     slice_type = golomb_to_pict_type[slice_type];
2472     if (slice_type == AV_PICTURE_TYPE_I ||
2473         (h0->current_slice != 0 && slice_type == h0->last_slice_type)) {
2474         default_ref_list_done = 1;
2475     }
2476     h->slice_type     = slice_type;
2477     h->slice_type_nos = slice_type & 3;
2478
2479     // to make a few old functions happy, it's wrong though
2480     s->pict_type = h->slice_type;
2481
2482     pps_id = get_ue_golomb(&s->gb);
2483     if (pps_id >= MAX_PPS_COUNT) {
2484         av_log(h->s.avctx, AV_LOG_ERROR, "pps_id %d out of range\n", pps_id);
2485         return -1;
2486     }
2487     if (!h0->pps_buffers[pps_id]) {
2488         av_log(h->s.avctx, AV_LOG_ERROR,
2489                "non-existing PPS %u referenced\n",
2490                pps_id);
2491         return -1;
2492     }
2493     h->pps = *h0->pps_buffers[pps_id];
2494
2495     if (!h0->sps_buffers[h->pps.sps_id]) {
2496         av_log(h->s.avctx, AV_LOG_ERROR,
2497                "non-existing SPS %u referenced\n",
2498                h->pps.sps_id);
2499         return -1;
2500     }
2501
2502     if (h->pps.sps_id != h->current_sps_id ||
2503         h0->sps_buffers[h->pps.sps_id]->new) {
2504         h0->sps_buffers[h->pps.sps_id]->new = 0;
2505
2506         h->current_sps_id = h->pps.sps_id;
2507         h->sps            = *h0->sps_buffers[h->pps.sps_id];
2508     }
2509
2510     s->avctx->profile = ff_h264_get_profile(&h->sps);
2511     s->avctx->level   = h->sps.level_idc;
2512     s->avctx->refs    = h->sps.ref_frame_count;
2513
2514     must_reinit = (s->context_initialized &&
2515                     (   16*h->sps.mb_width != s->avctx->coded_width
2516                      || 16*h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag) != s->avctx->coded_height
2517                      || s->avctx->bits_per_raw_sample != h->sps.bit_depth_luma
2518                      || h->cur_chroma_format_idc != h->sps.chroma_format_idc
2519                      || av_cmp_q(h->sps.sar, s->avctx->sample_aspect_ratio)));
2520
2521     if(must_reinit && (h != h0 || (s->avctx->active_thread_type & FF_THREAD_FRAME))) {
2522         av_log_missing_feature(s->avctx,
2523                                 "Width/height/bit depth/chroma idc changing with threads", 0);
2524         return AVERROR_PATCHWELCOME;   // width / height changed during parallelized decoding
2525     }
2526
2527     s->mb_width  = h->sps.mb_width;
2528     s->mb_height = h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
2529
2530     h->b_stride = s->mb_width * 4;
2531
2532     s->chroma_y_shift = h->sps.chroma_format_idc <= 1; // 400 uses yuv420p
2533
2534     s->width  = 16 * s->mb_width;
2535     s->height = 16 * s->mb_height;
2536
2537     if(must_reinit) {
2538         free_tables(h, 0);
2539         flush_dpb(s->avctx);
2540         ff_MPV_common_end(s);
2541         h->list_count = 0;
2542         h->current_slice = 0;
2543     }
2544     if (!s->context_initialized) {
2545         if (h != h0) {
2546             av_log(h->s.avctx, AV_LOG_ERROR,
2547                    "Cannot (re-)initialize context during parallel decoding.\n");
2548             return -1;
2549         }
2550         if(   FFALIGN(s->avctx->width , 16                                 ) == s->width
2551            && FFALIGN(s->avctx->height, 16*(2 - h->sps.frame_mbs_only_flag)) == s->height
2552            && !h->sps.crop_right && !h->sps.crop_bottom
2553            && (s->avctx->width != s->width || s->avctx->height && s->height)
2554         ) {
2555             av_log(h->s.avctx, AV_LOG_DEBUG, "Using externally provided dimensions\n");
2556             s->avctx->coded_width  = s->width;
2557             s->avctx->coded_height = s->height;
2558         } else{
2559             avcodec_set_dimensions(s->avctx, s->width, s->height);
2560             s->avctx->width  -= (2>>CHROMA444)*FFMIN(h->sps.crop_right, (8<<CHROMA444)-1);
2561             s->avctx->height -= (1<<s->chroma_y_shift)*FFMIN(h->sps.crop_bottom, (16>>s->chroma_y_shift)-1) * (2 - h->sps.frame_mbs_only_flag);
2562         }
2563         s->avctx->sample_aspect_ratio = h->sps.sar;
2564         av_assert0(s->avctx->sample_aspect_ratio.den);
2565
2566         if ((ret = h264_set_parameter_from_sps(h)) < 0)
2567             return ret;
2568
2569         if (h->sps.video_signal_type_present_flag) {
2570             s->avctx->color_range = h->sps.full_range>0 ? AVCOL_RANGE_JPEG
2571                                                       : AVCOL_RANGE_MPEG;
2572             if (h->sps.colour_description_present_flag) {
2573                 s->avctx->color_primaries = h->sps.color_primaries;
2574                 s->avctx->color_trc       = h->sps.color_trc;
2575                 s->avctx->colorspace      = h->sps.colorspace;
2576             }
2577         }
2578
2579         if (h->sps.timing_info_present_flag) {
2580             int64_t den = h->sps.time_scale;
2581             if (h->x264_build < 44U)
2582                 den *= 2;
2583             av_reduce(&s->avctx->time_base.num, &s->avctx->time_base.den,
2584                       h->sps.num_units_in_tick, den, 1 << 30);
2585         }
2586
2587         switch (h->sps.bit_depth_luma) {
2588         case 9:
2589             if (CHROMA444) {
2590                 if (s->avctx->colorspace == AVCOL_SPC_RGB) {
2591                     s->avctx->pix_fmt = AV_PIX_FMT_GBRP9;
2592                 } else
2593                     s->avctx->pix_fmt = AV_PIX_FMT_YUV444P9;
2594             } else if (CHROMA422)
2595                 s->avctx->pix_fmt = AV_PIX_FMT_YUV422P9;
2596             else
2597                 s->avctx->pix_fmt = AV_PIX_FMT_YUV420P9;
2598             break;
2599         case 10:
2600             if (CHROMA444) {
2601                 if (s->avctx->colorspace == AVCOL_SPC_RGB) {
2602                     s->avctx->pix_fmt = AV_PIX_FMT_GBRP10;
2603                 } else
2604                     s->avctx->pix_fmt = AV_PIX_FMT_YUV444P10;
2605             } else if (CHROMA422)
2606                 s->avctx->pix_fmt = AV_PIX_FMT_YUV422P10;
2607             else
2608                 s->avctx->pix_fmt = AV_PIX_FMT_YUV420P10;
2609             break;
2610         case 12:
2611             if (CHROMA444) {
2612                 if (s->avctx->colorspace == AVCOL_SPC_RGB) {
2613                     s->avctx->pix_fmt = AV_PIX_FMT_GBRP12;
2614                 } else
2615                     s->avctx->pix_fmt = AV_PIX_FMT_YUV444P12;
2616             } else if (CHROMA422)
2617                 s->avctx->pix_fmt = AV_PIX_FMT_YUV422P12;
2618             else
2619                 s->avctx->pix_fmt = AV_PIX_FMT_YUV420P12;
2620             break;
2621         case 14:
2622             if (CHROMA444) {
2623                 if (s->avctx->colorspace == AVCOL_SPC_RGB) {
2624                     s->avctx->pix_fmt = AV_PIX_FMT_GBRP14;
2625                 } else
2626                     s->avctx->pix_fmt = AV_PIX_FMT_YUV444P14;
2627             } else if (CHROMA422)
2628                 s->avctx->pix_fmt = AV_PIX_FMT_YUV422P14;
2629             else
2630                 s->avctx->pix_fmt = AV_PIX_FMT_YUV420P14;
2631             break;
2632         case 8:
2633             if (CHROMA444) {
2634                     s->avctx->pix_fmt = s->avctx->color_range == AVCOL_RANGE_JPEG ? AV_PIX_FMT_YUVJ444P
2635                                                                                   : AV_PIX_FMT_YUV444P;
2636                     if (s->avctx->colorspace == AVCOL_SPC_RGB) {
2637                         s->avctx->pix_fmt = AV_PIX_FMT_GBR24P;
2638                         av_log(h->s.avctx, AV_LOG_DEBUG, "Detected GBR colorspace.\n");
2639                     } else if (s->avctx->colorspace == AVCOL_SPC_YCGCO) {
2640                         av_log(h->s.avctx, AV_LOG_WARNING, "Detected unsupported YCgCo colorspace.\n");
2641                     }
2642             } else if (CHROMA422) {
2643                 s->avctx->pix_fmt = s->avctx->color_range == AVCOL_RANGE_JPEG ? AV_PIX_FMT_YUVJ422P
2644                                                                               : AV_PIX_FMT_YUV422P;
2645             } else {
2646                 s->avctx->pix_fmt = s->avctx->get_format(s->avctx,
2647                                                          s->avctx->codec->pix_fmts ?
2648                                                          s->avctx->codec->pix_fmts :
2649                                                          s->avctx->color_range == AVCOL_RANGE_JPEG ?
2650                                                          hwaccel_pixfmt_list_h264_jpeg_420 :
2651                                                          ff_hwaccel_pixfmt_list_420);
2652             }
2653             break;
2654         default:
2655             av_log(s->avctx, AV_LOG_ERROR,
2656                    "Unsupported bit depth: %d\n", h->sps.bit_depth_luma);
2657             return AVERROR_INVALIDDATA;
2658         }
2659
2660         s->avctx->hwaccel = ff_find_hwaccel(s->avctx->codec->id,
2661                                             s->avctx->pix_fmt);
2662
2663         if (ff_MPV_common_init(s) < 0) {
2664             av_log(h->s.avctx, AV_LOG_ERROR, "ff_MPV_common_init() failed.\n");
2665             return -1;
2666         }
2667         s->first_field = 0;
2668         h->prev_interlaced_frame = 1;
2669
2670         init_scan_tables(h);
2671         if (ff_h264_alloc_tables(h) < 0) {
2672             av_log(h->s.avctx, AV_LOG_ERROR,
2673                    "Could not allocate memory for h264\n");
2674             return AVERROR(ENOMEM);
2675         }
2676         h->bipred_scratchpad = NULL;
2677
2678         if (!HAVE_THREADS || !(s->avctx->active_thread_type & FF_THREAD_SLICE)) {
2679             if (context_init(h) < 0) {
2680                 av_log(h->s.avctx, AV_LOG_ERROR, "context_init() failed.\n");
2681                 return -1;
2682             }
2683         } else {
2684             for (i = 1; i < s->slice_context_count; i++) {
2685                 H264Context *c;
2686                 c = h->thread_context[i] = av_malloc(sizeof(H264Context));
2687                 memcpy(c, h->s.thread_context[i], sizeof(MpegEncContext));
2688                 memset(&c->s + 1, 0, sizeof(H264Context) - sizeof(MpegEncContext));
2689                 c->h264dsp     = h->h264dsp;
2690                 c->sps         = h->sps;
2691                 c->pps         = h->pps;
2692                 c->pixel_shift = h->pixel_shift;
2693                 c->cur_chroma_format_idc = h->cur_chroma_format_idc;
2694                 init_scan_tables(c);
2695                 clone_tables(c, h, i);
2696             }
2697
2698             for (i = 0; i < s->slice_context_count; i++)
2699                 if (context_init(h->thread_context[i]) < 0) {
2700                     av_log(h->s.avctx, AV_LOG_ERROR,
2701                            "context_init() failed.\n");
2702                     return -1;
2703                 }
2704         }
2705     }
2706
2707     if (h == h0 && h->dequant_coeff_pps != pps_id) {
2708         h->dequant_coeff_pps = pps_id;
2709         init_dequant_tables(h);
2710     }
2711
2712     h->frame_num = get_bits(&s->gb, h->sps.log2_max_frame_num);
2713
2714     h->mb_mbaff        = 0;
2715     h->mb_aff_frame    = 0;
2716     last_pic_structure = s0->picture_structure;
2717     last_pic_droppable = s0->droppable;
2718     s->droppable       = h->nal_ref_idc == 0;
2719     if (h->sps.frame_mbs_only_flag) {
2720         s->picture_structure = PICT_FRAME;
2721     } else {
2722         if (!h->sps.direct_8x8_inference_flag && slice_type == AV_PICTURE_TYPE_B) {
2723             av_log(h->s.avctx, AV_LOG_ERROR, "This stream was generated by a broken encoder, invalid 8x8 inference\n");
2724             return -1;
2725         }
2726         if (get_bits1(&s->gb)) { // field_pic_flag
2727             s->picture_structure = PICT_TOP_FIELD + get_bits1(&s->gb); // bottom_field_flag
2728         } else {
2729             s->picture_structure = PICT_FRAME;
2730             h->mb_aff_frame      = h->sps.mb_aff;
2731         }
2732     }
2733     h->mb_field_decoding_flag = s->picture_structure != PICT_FRAME;
2734
2735     if (h0->current_slice != 0) {
2736         if (last_pic_structure != s->picture_structure ||
2737             last_pic_droppable != s->droppable) {
2738             av_log(h->s.avctx, AV_LOG_ERROR,
2739                    "Changing field mode (%d -> %d) between slices is not allowed\n",
2740                    last_pic_structure, s->picture_structure);
2741             s->picture_structure = last_pic_structure;
2742             s->droppable         = last_pic_droppable;
2743             return AVERROR_INVALIDDATA;
2744         } else if (!s0->current_picture_ptr) {
2745             av_log(s->avctx, AV_LOG_ERROR,
2746                    "unset current_picture_ptr on %d. slice\n",
2747                    h0->current_slice + 1);
2748             return AVERROR_INVALIDDATA;
2749         }
2750     } else {
2751         /* Shorten frame num gaps so we don't have to allocate reference
2752          * frames just to throw them away */
2753         if (h->frame_num != h->prev_frame_num && h->prev_frame_num >= 0) {
2754             int unwrap_prev_frame_num = h->prev_frame_num;
2755             int max_frame_num         = 1 << h->sps.log2_max_frame_num;
2756
2757             if (unwrap_prev_frame_num > h->frame_num)
2758                 unwrap_prev_frame_num -= max_frame_num;
2759
2760             if ((h->frame_num - unwrap_prev_frame_num) > h->sps.ref_frame_count) {
2761                 unwrap_prev_frame_num = (h->frame_num - h->sps.ref_frame_count) - 1;
2762                 if (unwrap_prev_frame_num < 0)
2763                     unwrap_prev_frame_num += max_frame_num;
2764
2765                 h->prev_frame_num = unwrap_prev_frame_num;
2766             }
2767         }
2768
2769         /* See if we have a decoded first field looking for a pair...
2770          * Here, we're using that to see if we should mark previously
2771          * decode frames as "finished".
2772          * We have to do that before the "dummy" in-between frame allocation,
2773          * since that can modify s->current_picture_ptr. */
2774         if (s0->first_field) {
2775             assert(s0->current_picture_ptr);
2776             assert(s0->current_picture_ptr->f.data[0]);
2777             assert(s0->current_picture_ptr->f.reference != DELAYED_PIC_REF);
2778
2779             /* Mark old field/frame as completed */
2780             if (!last_pic_droppable && s0->current_picture_ptr->owner2 == s0) {
2781                 ff_thread_report_progress(&s0->current_picture_ptr->f, INT_MAX,
2782                                           last_pic_structure == PICT_BOTTOM_FIELD);
2783             }
2784
2785             /* figure out if we have a complementary field pair */
2786             if (!FIELD_PICTURE || s->picture_structure == last_pic_structure) {
2787                 /* Previous field is unmatched. Don't display it, but let it
2788                  * remain for reference if marked as such. */
2789                 if (!last_pic_droppable && last_pic_structure != PICT_FRAME) {
2790                     ff_thread_report_progress(&s0->current_picture_ptr->f, INT_MAX,
2791                                               last_pic_structure == PICT_TOP_FIELD);
2792                 }
2793             } else {
2794                 if (s0->current_picture_ptr->frame_num != h->frame_num) {
2795                     /* This and previous field were reference, but had
2796                      * different frame_nums. Consider this field first in
2797                      * pair. Throw away previous field except for reference
2798                      * purposes. */
2799                     if (!last_pic_droppable && last_pic_structure != PICT_FRAME) {
2800                         ff_thread_report_progress(&s0->current_picture_ptr->f, INT_MAX,
2801                                                   last_pic_structure == PICT_TOP_FIELD);
2802                     }
2803                 } else {
2804                     /* Second field in complementary pair */
2805                     if (!((last_pic_structure   == PICT_TOP_FIELD &&
2806                            s->picture_structure == PICT_BOTTOM_FIELD) ||
2807                           (last_pic_structure   == PICT_BOTTOM_FIELD &&
2808                            s->picture_structure == PICT_TOP_FIELD))) {
2809                         av_log(s->avctx, AV_LOG_ERROR,
2810                                "Invalid field mode combination %d/%d\n",
2811                                last_pic_structure, s->picture_structure);
2812                         s->picture_structure = last_pic_structure;
2813                         s->droppable         = last_pic_droppable;
2814                         return AVERROR_INVALIDDATA;
2815                     } else if (last_pic_droppable != s->droppable) {
2816                         av_log(s->avctx, AV_LOG_ERROR,
2817                                "Cannot combine reference and non-reference fields in the same frame\n");
2818                         av_log_ask_for_sample(s->avctx, NULL);
2819                         s->picture_structure = last_pic_structure;
2820                         s->droppable         = last_pic_droppable;
2821                         return AVERROR_INVALIDDATA;
2822                     }
2823
2824                     /* Take ownership of this buffer. Note that if another thread owned
2825                      * the first field of this buffer, we're not operating on that pointer,
2826                      * so the original thread is still responsible for reporting progress
2827                      * on that first field (or if that was us, we just did that above).
2828                      * By taking ownership, we assign responsibility to ourselves to
2829                      * report progress on the second field. */
2830                     s0->current_picture_ptr->owner2 = s0;
2831                 }
2832             }
2833         }
2834
2835         while (h->frame_num != h->prev_frame_num && h->prev_frame_num >= 0 &&
2836                h->frame_num != (h->prev_frame_num + 1) % (1 << h->sps.log2_max_frame_num)) {
2837             Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
2838             av_log(h->s.avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n",
2839                    h->frame_num, h->prev_frame_num);
2840             if (ff_h264_frame_start(h) < 0)
2841                 return -1;
2842             h->prev_frame_num++;
2843             h->prev_frame_num %= 1 << h->sps.log2_max_frame_num;
2844             s->current_picture_ptr->frame_num = h->prev_frame_num;
2845             ff_thread_report_progress(&s->current_picture_ptr->f, INT_MAX, 0);
2846             ff_thread_report_progress(&s->current_picture_ptr->f, INT_MAX, 1);
2847             ff_generate_sliding_window_mmcos(h);
2848             if (ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index) < 0 &&
2849                 (s->avctx->err_recognition & AV_EF_EXPLODE))
2850                 return AVERROR_INVALIDDATA;
2851             /* Error concealment: if a ref is missing, copy the previous ref in its place.
2852              * FIXME: avoiding a memcpy would be nice, but ref handling makes many assumptions
2853              * about there being no actual duplicates.
2854              * FIXME: this doesn't copy padding for out-of-frame motion vectors.  Given we're
2855              * concealing a lost frame, this probably isn't noticeable by comparison, but it should
2856              * be fixed. */
2857             if (h->short_ref_count) {
2858                 if (prev) {
2859                     av_image_copy(h->short_ref[0]->f.data, h->short_ref[0]->f.linesize,
2860                                   (const uint8_t **)prev->f.data, prev->f.linesize,
2861                                   s->avctx->pix_fmt, s->mb_width * 16, s->mb_height * 16);
2862                     h->short_ref[0]->poc = prev->poc + 2;
2863                 }
2864                 h->short_ref[0]->frame_num = h->prev_frame_num;
2865             }
2866         }
2867
2868         /* See if we have a decoded first field looking for a pair...
2869          * We're using that to see whether to continue decoding in that
2870          * frame, or to allocate a new one. */
2871         if (s0->first_field) {
2872             assert(s0->current_picture_ptr);
2873             assert(s0->current_picture_ptr->f.data[0]);
2874             assert(s0->current_picture_ptr->f.reference != DELAYED_PIC_REF);
2875
2876             /* figure out if we have a complementary field pair */
2877             if (!FIELD_PICTURE || s->picture_structure == last_pic_structure) {
2878                 /* Previous field is unmatched. Don't display it, but let it
2879                  * remain for reference if marked as such. */
2880                 s0->current_picture_ptr = NULL;
2881                 s0->first_field         = FIELD_PICTURE;
2882             } else {
2883                 if (s0->current_picture_ptr->frame_num != h->frame_num) {
2884                     ff_thread_report_progress((AVFrame*)s0->current_picture_ptr, INT_MAX,
2885                                               s0->picture_structure==PICT_BOTTOM_FIELD);
2886                     /* This and the previous field had different frame_nums.
2887                      * Consider this field first in pair. Throw away previous
2888                      * one except for reference purposes. */
2889                     s0->first_field         = 1;
2890                     s0->current_picture_ptr = NULL;
2891                 } else {
2892                     /* Second field in complementary pair */
2893                     s0->first_field = 0;
2894                 }
2895             }
2896         } else {
2897             /* Frame or first field in a potentially complementary pair */
2898             s0->first_field = FIELD_PICTURE;
2899         }
2900
2901         if (!FIELD_PICTURE || s0->first_field) {
2902             if (ff_h264_frame_start(h) < 0) {
2903                 s0->first_field = 0;
2904                 return -1;
2905             }
2906         } else {
2907             ff_release_unused_pictures(s, 0);
2908         }
2909     }
2910     if (h != h0)
2911         clone_slice(h, h0);
2912
2913     s->current_picture_ptr->frame_num = h->frame_num; // FIXME frame_num cleanup
2914
2915     av_assert1(s->mb_num == s->mb_width * s->mb_height);
2916     if (first_mb_in_slice << FIELD_OR_MBAFF_PICTURE >= s->mb_num ||
2917         first_mb_in_slice >= s->mb_num) {
2918         av_log(h->s.avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
2919         return -1;
2920     }
2921     s->resync_mb_x = s->mb_x =  first_mb_in_slice % s->mb_width;
2922     s->resync_mb_y = s->mb_y = (first_mb_in_slice / s->mb_width) << FIELD_OR_MBAFF_PICTURE;
2923     if (s->picture_structure == PICT_BOTTOM_FIELD)
2924         s->resync_mb_y = s->mb_y = s->mb_y + 1;
2925     av_assert1(s->mb_y < s->mb_height);
2926
2927     if (s->picture_structure == PICT_FRAME) {
2928         h->curr_pic_num = h->frame_num;
2929         h->max_pic_num  = 1 << h->sps.log2_max_frame_num;
2930     } else {
2931         h->curr_pic_num = 2 * h->frame_num + 1;
2932         h->max_pic_num  = 1 << (h->sps.log2_max_frame_num + 1);
2933     }
2934
2935     if (h->nal_unit_type == NAL_IDR_SLICE)
2936         get_ue_golomb(&s->gb); /* idr_pic_id */
2937
2938     if (h->sps.poc_type == 0) {
2939         h->poc_lsb = get_bits(&s->gb, h->sps.log2_max_poc_lsb);
2940
2941         if (h->pps.pic_order_present == 1 && s->picture_structure == PICT_FRAME)
2942             h->delta_poc_bottom = get_se_golomb(&s->gb);
2943     }
2944
2945     if (h->sps.poc_type == 1 && !h->sps.delta_pic_order_always_zero_flag) {
2946         h->delta_poc[0] = get_se_golomb(&s->gb);
2947
2948         if (h->pps.pic_order_present == 1 && s->picture_structure == PICT_FRAME)
2949             h->delta_poc[1] = get_se_golomb(&s->gb);
2950     }
2951
2952     init_poc(h);
2953
2954     if (h->pps.redundant_pic_cnt_present)
2955         h->redundant_pic_count = get_ue_golomb(&s->gb);
2956
2957     // set defaults, might be overridden a few lines later
2958     h->ref_count[0] = h->pps.ref_count[0];
2959     h->ref_count[1] = h->pps.ref_count[1];
2960
2961     if (h->slice_type_nos != AV_PICTURE_TYPE_I) {
2962         unsigned max[2];
2963         max[0] = max[1] = s->picture_structure == PICT_FRAME ? 15 : 31;
2964
2965         if (h->slice_type_nos == AV_PICTURE_TYPE_B)
2966             h->direct_spatial_mv_pred = get_bits1(&s->gb);
2967         num_ref_idx_active_override_flag = get_bits1(&s->gb);
2968
2969         if (num_ref_idx_active_override_flag) {
2970             h->ref_count[0] = get_ue_golomb(&s->gb) + 1;
2971             if (h->slice_type_nos == AV_PICTURE_TYPE_B) {
2972                 h->ref_count[1] = get_ue_golomb(&s->gb) + 1;
2973             } else
2974                 // full range is spec-ok in this case, even for frames
2975                 h->ref_count[1] = 1;
2976         }
2977
2978         if (h->ref_count[0]-1 > max[0] || h->ref_count[1]-1 > max[1]){
2979             av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow %u > %u or %u > %u\n", h->ref_count[0]-1, max[0], h->ref_count[1]-1, max[1]);
2980             h->ref_count[0] = h->ref_count[1] = 1;
2981             return AVERROR_INVALIDDATA;
2982         }
2983
2984         if (h->slice_type_nos == AV_PICTURE_TYPE_B)
2985             h->list_count = 2;
2986         else
2987             h->list_count = 1;
2988     } else
2989         h->ref_count[1]= h->ref_count[0]= h->list_count= 0;
2990
2991     if (!default_ref_list_done)
2992         ff_h264_fill_default_ref_list(h);
2993
2994     if (h->slice_type_nos != AV_PICTURE_TYPE_I &&
2995         ff_h264_decode_ref_pic_list_reordering(h) < 0) {
2996         h->ref_count[1] = h->ref_count[0] = 0;
2997         return -1;
2998     }
2999
3000     if (h->slice_type_nos != AV_PICTURE_TYPE_I) {
3001         s->last_picture_ptr = &h->ref_list[0][0];
3002         s->last_picture_ptr->owner2 = s;
3003         ff_copy_picture(&s->last_picture, s->last_picture_ptr);
3004     }
3005     if (h->slice_type_nos == AV_PICTURE_TYPE_B) {
3006         s->next_picture_ptr = &h->ref_list[1][0];
3007         s->next_picture_ptr->owner2 = s;
3008         ff_copy_picture(&s->next_picture, s->next_picture_ptr);
3009     }
3010
3011     if ((h->pps.weighted_pred && h->slice_type_nos == AV_PICTURE_TYPE_P) ||
3012         (h->pps.weighted_bipred_idc == 1 &&
3013          h->slice_type_nos == AV_PICTURE_TYPE_B))
3014         pred_weight_table(h);
3015     else if (h->pps.weighted_bipred_idc == 2 &&
3016              h->slice_type_nos == AV_PICTURE_TYPE_B) {
3017         implicit_weight_table(h, -1);
3018     } else {
3019         h->use_weight = 0;
3020         for (i = 0; i < 2; i++) {
3021             h->luma_weight_flag[i]   = 0;
3022             h->chroma_weight_flag[i] = 0;
3023         }
3024     }
3025
3026     if (h->nal_ref_idc && ff_h264_decode_ref_pic_marking(h0, &s->gb) < 0 &&
3027         (s->avctx->err_recognition & AV_EF_EXPLODE))
3028         return AVERROR_INVALIDDATA;
3029
3030     if (FRAME_MBAFF) {
3031         ff_h264_fill_mbaff_ref_list(h);
3032
3033         if (h->pps.weighted_bipred_idc == 2 && h->slice_type_nos == AV_PICTURE_TYPE_B) {
3034             implicit_weight_table(h, 0);
3035             implicit_weight_table(h, 1);
3036         }
3037     }
3038
3039     if (h->slice_type_nos == AV_PICTURE_TYPE_B && !h->direct_spatial_mv_pred)
3040         ff_h264_direct_dist_scale_factor(h);
3041     ff_h264_direct_ref_list_init(h);
3042
3043     if (h->slice_type_nos != AV_PICTURE_TYPE_I && h->pps.cabac) {
3044         tmp = get_ue_golomb_31(&s->gb);
3045         if (tmp > 2) {
3046             av_log(s->avctx, AV_LOG_ERROR, "cabac_init_idc overflow\n");
3047             return -1;
3048         }
3049         h->cabac_init_idc = tmp;
3050     }
3051
3052     h->last_qscale_diff = 0;
3053     tmp = h->pps.init_qp + get_se_golomb(&s->gb);
3054     if (tmp > 51 + 6 * (h->sps.bit_depth_luma - 8)) {
3055         av_log(s->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
3056         return -1;
3057     }
3058     s->qscale       = tmp;
3059     h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
3060     h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
3061     // FIXME qscale / qp ... stuff
3062     if (h->slice_type == AV_PICTURE_TYPE_SP)
3063         get_bits1(&s->gb); /* sp_for_switch_flag */
3064     if (h->slice_type == AV_PICTURE_TYPE_SP ||
3065         h->slice_type == AV_PICTURE_TYPE_SI)
3066         get_se_golomb(&s->gb); /* slice_qs_delta */
3067
3068     h->deblocking_filter     = 1;
3069     h->slice_alpha_c0_offset = 52;
3070     h->slice_beta_offset     = 52;
3071     if (h->pps.deblocking_filter_parameters_present) {
3072         tmp = get_ue_golomb_31(&s->gb);
3073         if (tmp > 2) {
3074             av_log(s->avctx, AV_LOG_ERROR,
3075                    "deblocking_filter_idc %u out of range\n", tmp);
3076             return -1;
3077         }
3078         h->deblocking_filter = tmp;
3079         if (h->deblocking_filter < 2)
3080             h->deblocking_filter ^= 1;  // 1<->0
3081
3082         if (h->deblocking_filter) {
3083             h->slice_alpha_c0_offset += get_se_golomb(&s->gb) << 1;
3084             h->slice_beta_offset     += get_se_golomb(&s->gb) << 1;
3085             if (h->slice_alpha_c0_offset > 104U ||
3086                 h->slice_beta_offset     > 104U) {
3087                 av_log(s->avctx, AV_LOG_ERROR,
3088                        "deblocking filter parameters %d %d out of range\n",
3089                        h->slice_alpha_c0_offset, h->slice_beta_offset);
3090                 return -1;
3091             }
3092         }
3093     }
3094
3095     if (s->avctx->skip_loop_filter >= AVDISCARD_ALL ||
3096         (s->avctx->skip_loop_filter >= AVDISCARD_NONKEY &&
3097          h->slice_type_nos != AV_PICTURE_TYPE_I) ||
3098         (s->avctx->skip_loop_filter >= AVDISCARD_BIDIR  &&
3099          h->slice_type_nos == AV_PICTURE_TYPE_B) ||
3100         (s->avctx->skip_loop_filter >= AVDISCARD_NONREF &&
3101          h->nal_ref_idc == 0))
3102         h->deblocking_filter = 0;
3103
3104     if (h->deblocking_filter == 1 && h0->max_contexts > 1) {
3105         if (s->avctx->flags2 & CODEC_FLAG2_FAST) {
3106             /* Cheat slightly for speed:
3107              * Do not bother to deblock across slices. */
3108             h->deblocking_filter = 2;
3109         } else {
3110             h0->max_contexts = 1;
3111             if (!h0->single_decode_warning) {
3112                 av_log(s->avctx, AV_LOG_INFO,
3113                        "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
3114                 h0->single_decode_warning = 1;
3115             }
3116             if (h != h0) {
3117                 av_log(h->s.avctx, AV_LOG_ERROR,
3118                        "Deblocking switched inside frame.\n");
3119                 return 1;
3120             }
3121         }
3122     }
3123     h->qp_thresh = 15 + 52 -
3124                    FFMIN(h->slice_alpha_c0_offset, h->slice_beta_offset) -
3125                    FFMAX3(0,
3126                           h->pps.chroma_qp_index_offset[0],
3127                           h->pps.chroma_qp_index_offset[1]) +
3128                    6 * (h->sps.bit_depth_luma - 8);
3129
3130     h0->last_slice_type = slice_type;
3131     h->slice_num = ++h0->current_slice;
3132
3133     if (h->slice_num)
3134         h0->slice_row[(h->slice_num-1)&(MAX_SLICES-1)]= s->resync_mb_y;
3135     if (   h0->slice_row[h->slice_num&(MAX_SLICES-1)] + 3 >= s->resync_mb_y
3136         && h0->slice_row[h->slice_num&(MAX_SLICES-1)] <= s->resync_mb_y
3137         && h->slice_num >= MAX_SLICES) {
3138         //in case of ASO this check needs to be updated depending on how we decide to assign slice numbers in this case
3139         av_log(s->avctx, AV_LOG_WARNING, "Possibly too many slices (%d >= %d), increase MAX_SLICES and recompile if there are artifacts\n", h->slice_num, MAX_SLICES);
3140     }
3141
3142     for (j = 0; j < 2; j++) {
3143         int id_list[16];
3144         int *ref2frm = h->ref2frm[h->slice_num & (MAX_SLICES - 1)][j];
3145         for (i = 0; i < 16; i++) {
3146             id_list[i] = 60;
3147             if (h->ref_list[j][i].f.data[0]) {
3148                 int k;
3149                 uint8_t *base = h->ref_list[j][i].f.base[0];
3150                 for (k = 0; k < h->short_ref_count; k++)
3151                     if (h->short_ref[k]->f.base[0] == base) {
3152                         id_list[i] = k;
3153                         break;
3154                     }
3155                 for (k = 0; k < h->long_ref_count; k++)
3156                     if (h->long_ref[k] && h->long_ref[k]->f.base[0] == base) {
3157                         id_list[i] = h->short_ref_count + k;
3158                         break;
3159                     }
3160             }
3161         }
3162
3163         ref2frm[0]     =
3164             ref2frm[1] = -1;
3165         for (i = 0; i < 16; i++)
3166             ref2frm[i + 2] = 4 * id_list[i] +
3167                              (h->ref_list[j][i].f.reference & 3);
3168         ref2frm[18 + 0]     =
3169             ref2frm[18 + 1] = -1;
3170         for (i = 16; i < 48; i++)
3171             ref2frm[i + 4] = 4 * id_list[(i - 16) >> 1] +
3172                              (h->ref_list[j][i].f.reference & 3);
3173     }
3174
3175     // FIXME: fix draw_edges + PAFF + frame threads
3176     h->emu_edge_width  = (s->flags & CODEC_FLAG_EMU_EDGE ||
3177                           (!h->sps.frame_mbs_only_flag &&
3178                            s->avctx->active_thread_type))
3179                          ? 0 : 16;
3180     h->emu_edge_height = (FRAME_MBAFF || FIELD_PICTURE) ? 0 : h->emu_edge_width;
3181
3182     if (s->avctx->debug & FF_DEBUG_PICT_INFO) {
3183         av_log(h->s.avctx, AV_LOG_DEBUG,
3184                "slice:%d %s mb:%d %c%s%s pps:%u frame:%d poc:%d/%d ref:%d/%d qp:%d loop:%d:%d:%d weight:%d%s %s\n",
3185                h->slice_num,
3186                (s->picture_structure == PICT_FRAME ? "F" : s->picture_structure == PICT_TOP_FIELD ? "T" : "B"),
3187                first_mb_in_slice,
3188                av_get_picture_type_char(h->slice_type),
3189                h->slice_type_fixed ? " fix" : "",
3190                h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
3191                pps_id, h->frame_num,
3192                s->current_picture_ptr->field_poc[0],
3193                s->current_picture_ptr->field_poc[1],
3194                h->ref_count[0], h->ref_count[1],
3195                s->qscale,
3196                h->deblocking_filter,
3197                h->slice_alpha_c0_offset / 2 - 26, h->slice_beta_offset / 2 - 26,
3198                h->use_weight,
3199                h->use_weight == 1 && h->use_weight_chroma ? "c" : "",
3200                h->slice_type == AV_PICTURE_TYPE_B ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : "");
3201     }
3202
3203     return 0;
3204 }
3205
3206 int ff_h264_get_slice_type(const H264Context *h)
3207 {
3208     switch (h->slice_type) {
3209     case AV_PICTURE_TYPE_P:
3210         return 0;
3211     case AV_PICTURE_TYPE_B:
3212         return 1;
3213     case AV_PICTURE_TYPE_I:
3214         return 2;
3215     case AV_PICTURE_TYPE_SP:
3216         return 3;
3217     case AV_PICTURE_TYPE_SI:
3218         return 4;
3219     default:
3220         return -1;
3221     }
3222 }
3223
3224 static av_always_inline void fill_filter_caches_inter(H264Context *h,
3225                                                       MpegEncContext *const s,
3226                                                       int mb_type, int top_xy,
3227                                                       int left_xy[LEFT_MBS],
3228                                                       int top_type,
3229                                                       int left_type[LEFT_MBS],
3230                                                       int mb_xy, int list)
3231 {
3232     int b_stride = h->b_stride;
3233     int16_t(*mv_dst)[2] = &h->mv_cache[list][scan8[0]];
3234     int8_t *ref_cache = &h->ref_cache[list][scan8[0]];
3235     if (IS_INTER(mb_type) || IS_DIRECT(mb_type)) {
3236         if (USES_LIST(top_type, list)) {
3237             const int b_xy  = h->mb2b_xy[top_xy] + 3 * b_stride;
3238             const int b8_xy = 4 * top_xy + 2;
3239             int (*ref2frm)[64] = (void*)(h->ref2frm[h->slice_table[top_xy] & (MAX_SLICES - 1)][0] + (MB_MBAFF ? 20 : 2));
3240             AV_COPY128(mv_dst - 1 * 8, s->current_picture.f.motion_val[list][b_xy + 0]);
3241             ref_cache[0 - 1 * 8] =
3242             ref_cache[1 - 1 * 8] = ref2frm[list][s->current_picture.f.ref_index[list][b8_xy + 0]];
3243             ref_cache[2 - 1 * 8] =
3244             ref_cache[3 - 1 * 8] = ref2frm[list][s->current_picture.f.ref_index[list][b8_xy + 1]];
3245         } else {
3246             AV_ZERO128(mv_dst - 1 * 8);
3247             AV_WN32A(&ref_cache[0 - 1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
3248         }
3249
3250         if (!IS_INTERLACED(mb_type ^ left_type[LTOP])) {
3251             if (USES_LIST(left_type[LTOP], list)) {
3252                 const int b_xy  = h->mb2b_xy[left_xy[LTOP]] + 3;
3253                 const int b8_xy = 4 * left_xy[LTOP] + 1;
3254                 int (*ref2frm)[64] =(void*)( h->ref2frm[h->slice_table[left_xy[LTOP]] & (MAX_SLICES - 1)][0] + (MB_MBAFF ? 20 : 2));
3255                 AV_COPY32(mv_dst - 1 +  0, s->current_picture.f.motion_val[list][b_xy + b_stride * 0]);
3256                 AV_COPY32(mv_dst - 1 +  8, s->current_picture.f.motion_val[list][b_xy + b_stride * 1]);
3257                 AV_COPY32(mv_dst - 1 + 16, s->current_picture.f.motion_val[list][b_xy + b_stride * 2]);
3258                 AV_COPY32(mv_dst - 1 + 24, s->current_picture.f.motion_val[list][b_xy + b_stride * 3]);
3259                 ref_cache[-1 +  0] =
3260                 ref_cache[-1 +  8] = ref2frm[list][s->current_picture.f.ref_index[list][b8_xy + 2 * 0]];
3261                 ref_cache[-1 + 16] =
3262                 ref_cache[-1 + 24] = ref2frm[list][s->current_picture.f.ref_index[list][b8_xy + 2 * 1]];
3263             } else {
3264                 AV_ZERO32(mv_dst - 1 +  0);
3265                 AV_ZERO32(mv_dst - 1 +  8);
3266                 AV_ZERO32(mv_dst - 1 + 16);
3267                 AV_ZERO32(mv_dst - 1 + 24);
3268                 ref_cache[-1 +  0] =
3269                 ref_cache[-1 +  8] =
3270                 ref_cache[-1 + 16] =
3271                 ref_cache[-1 + 24] = LIST_NOT_USED;
3272             }
3273         }
3274     }
3275
3276     if (!USES_LIST(mb_type, list)) {
3277         fill_rectangle(mv_dst, 4, 4, 8, pack16to32(0, 0), 4);
3278         AV_WN32A(&ref_cache[0 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
3279         AV_WN32A(&ref_cache[1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
3280         AV_WN32A(&ref_cache[2 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
3281         AV_WN32A(&ref_cache[3 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
3282         return;
3283     }
3284
3285     {
3286         int8_t *ref = &s->current_picture.f.ref_index[list][4 * mb_xy];
3287         int (*ref2frm)[64] = (void*)(h->ref2frm[h->slice_num & (MAX_SLICES - 1)][0] + (MB_MBAFF ? 20 : 2));
3288         uint32_t ref01 = (pack16to32(ref2frm[list][ref[0]], ref2frm[list][ref[1]]) & 0x00FF00FF) * 0x0101;
3289         uint32_t ref23 = (pack16to32(ref2frm[list][ref[2]], ref2frm[list][ref[3]]) & 0x00FF00FF) * 0x0101;
3290         AV_WN32A(&ref_cache[0 * 8], ref01);
3291         AV_WN32A(&ref_cache[1 * 8], ref01);
3292         AV_WN32A(&ref_cache[2 * 8], ref23);
3293         AV_WN32A(&ref_cache[3 * 8], ref23);
3294     }
3295
3296     {
3297         int16_t(*mv_src)[2] = &s->current_picture.f.motion_val[list][4 * s->mb_x + 4 * s->mb_y * b_stride];
3298         AV_COPY128(mv_dst + 8 * 0, mv_src + 0 * b_stride);
3299         AV_COPY128(mv_dst + 8 * 1, mv_src + 1 * b_stride);
3300         AV_COPY128(mv_dst + 8 * 2, mv_src + 2 * b_stride);
3301         AV_COPY128(mv_dst + 8 * 3, mv_src + 3 * b_stride);
3302     }
3303 }
3304
3305 /**
3306  *
3307  * @return non zero if the loop filter can be skipped
3308  */
3309 static int fill_filter_caches(H264Context *h, int mb_type)
3310 {
3311     MpegEncContext *const s = &h->s;
3312     const int mb_xy = h->mb_xy;
3313     int top_xy, left_xy[LEFT_MBS];
3314     int top_type, left_type[LEFT_MBS];
3315     uint8_t *nnz;
3316     uint8_t *nnz_cache;
3317
3318     top_xy = mb_xy - (s->mb_stride << MB_FIELD);
3319
3320     /* Wow, what a mess, why didn't they simplify the interlacing & intra
3321      * stuff, I can't imagine that these complex rules are worth it. */
3322
3323     left_xy[LBOT] = left_xy[LTOP] = mb_xy - 1;
3324     if (FRAME_MBAFF) {
3325         const int left_mb_field_flag = IS_INTERLACED(s->current_picture.f.mb_type[mb_xy - 1]);
3326         const int curr_mb_field_flag = IS_INTERLACED(mb_type);
3327         if (s->mb_y & 1) {
3328             if (left_mb_field_flag != curr_mb_field_flag)
3329                 left_xy[LTOP] -= s->mb_stride;
3330         } else {
3331             if (curr_mb_field_flag)
3332                 top_xy += s->mb_stride &
3333                     (((s->current_picture.f.mb_type[top_xy] >> 7) & 1) - 1);
3334             if (left_mb_field_flag != curr_mb_field_flag)
3335                 left_xy[LBOT] += s->mb_stride;
3336         }
3337     }
3338
3339     h->top_mb_xy        = top_xy;
3340     h->left_mb_xy[LTOP] = left_xy[LTOP];
3341     h->left_mb_xy[LBOT] = left_xy[LBOT];
3342     {
3343         /* For sufficiently low qp, filtering wouldn't do anything.
3344          * This is a conservative estimate: could also check beta_offset
3345          * and more accurate chroma_qp. */
3346         int qp_thresh = h->qp_thresh; // FIXME strictly we should store qp_thresh for each mb of a slice
3347         int qp        = s->current_picture.f.qscale_table[mb_xy];
3348         if (qp <= qp_thresh &&
3349             (left_xy[LTOP] < 0 ||
3350              ((qp + s->current_picture.f.qscale_table[left_xy[LTOP]] + 1) >> 1) <= qp_thresh) &&
3351             (top_xy < 0 ||
3352              ((qp + s->current_picture.f.qscale_table[top_xy] + 1) >> 1) <= qp_thresh)) {
3353             if (!FRAME_MBAFF)
3354                 return 1;
3355             if ((left_xy[LTOP] < 0 ||
3356                  ((qp + s->current_picture.f.qscale_table[left_xy[LBOT]] + 1) >> 1) <= qp_thresh) &&
3357                 (top_xy < s->mb_stride ||
3358                  ((qp + s->current_picture.f.qscale_table[top_xy - s->mb_stride] + 1) >> 1) <= qp_thresh))
3359                 return 1;
3360         }
3361     }
3362
3363     top_type        = s->current_picture.f.mb_type[top_xy];
3364     left_type[LTOP] = s->current_picture.f.mb_type[left_xy[LTOP]];
3365     left_type[LBOT] = s->current_picture.f.mb_type[left_xy[LBOT]];
3366     if (h->deblocking_filter == 2) {
3367         if (h->slice_table[top_xy] != h->slice_num)
3368             top_type = 0;
3369         if (h->slice_table[left_xy[LBOT]] != h->slice_num)
3370             left_type[LTOP] = left_type[LBOT] = 0;
3371     } else {
3372         if (h->slice_table[top_xy] == 0xFFFF)
3373             top_type = 0;
3374         if (h->slice_table[left_xy[LBOT]] == 0xFFFF)
3375             left_type[LTOP] = left_type[LBOT] = 0;
3376     }
3377     h->top_type        = top_type;
3378     h->left_type[LTOP] = left_type[LTOP];
3379     h->left_type[LBOT] = left_type[LBOT];
3380
3381     if (IS_INTRA(mb_type))
3382         return 0;
3383
3384     fill_filter_caches_inter(h, s, mb_type, top_xy, left_xy,
3385                              top_type, left_type, mb_xy, 0);
3386     if (h->list_count == 2)
3387         fill_filter_caches_inter(h, s, mb_type, top_xy, left_xy,
3388                                  top_type, left_type, mb_xy, 1);
3389
3390     nnz       = h->non_zero_count[mb_xy];
3391     nnz_cache = h->non_zero_count_cache;
3392     AV_COPY32(&nnz_cache[4 + 8 * 1], &nnz[0]);
3393     AV_COPY32(&nnz_cache[4 + 8 * 2], &nnz[4]);
3394     AV_COPY32(&nnz_cache[4 + 8 * 3], &nnz[8]);
3395     AV_COPY32(&nnz_cache[4 + 8 * 4], &nnz[12]);
3396     h->cbp = h->cbp_table[mb_xy];
3397
3398     if (top_type) {
3399         nnz = h->non_zero_count[top_xy];
3400         AV_COPY32(&nnz_cache[4 + 8 * 0], &nnz[3 * 4]);
3401     }
3402
3403     if (left_type[LTOP]) {
3404         nnz = h->non_zero_count[left_xy[LTOP]];
3405         nnz_cache[3 + 8 * 1] = nnz[3 + 0 * 4];
3406         nnz_cache[3 + 8 * 2] = nnz[3 + 1 * 4];
3407         nnz_cache[3 + 8 * 3] = nnz[3 + 2 * 4];
3408         nnz_cache[3 + 8 * 4] = nnz[3 + 3 * 4];
3409     }
3410
3411     /* CAVLC 8x8dct requires NNZ values for residual decoding that differ
3412      * from what the loop filter needs */
3413     if (!CABAC && h->pps.transform_8x8_mode) {
3414         if (IS_8x8DCT(top_type)) {
3415             nnz_cache[4 + 8 * 0]     =
3416                 nnz_cache[5 + 8 * 0] = (h->cbp_table[top_xy] & 0x4000) >> 12;
3417             nnz_cache[6 + 8 * 0]     =
3418                 nnz_cache[7 + 8 * 0] = (h->cbp_table[top_xy] & 0x8000) >> 12;
3419         }
3420         if (IS_8x8DCT(left_type[LTOP])) {
3421             nnz_cache[3 + 8 * 1]     =
3422                 nnz_cache[3 + 8 * 2] = (h->cbp_table[left_xy[LTOP]] & 0x2000) >> 12; // FIXME check MBAFF
3423         }
3424         if (IS_8x8DCT(left_type[LBOT])) {
3425             nnz_cache[3 + 8 * 3]     =
3426                 nnz_cache[3 + 8 * 4] = (h->cbp_table[left_xy[LBOT]] & 0x8000) >> 12; // FIXME check MBAFF
3427         }
3428
3429         if (IS_8x8DCT(mb_type)) {
3430             nnz_cache[scan8[0]] =
3431             nnz_cache[scan8[1]] =
3432             nnz_cache[scan8[2]] =
3433             nnz_cache[scan8[3]] = (h->cbp & 0x1000) >> 12;
3434
3435             nnz_cache[scan8[0 + 4]] =
3436             nnz_cache[scan8[1 + 4]] =
3437             nnz_cache[scan8[2 + 4]] =
3438             nnz_cache[scan8[3 + 4]] = (h->cbp & 0x2000) >> 12;
3439
3440             nnz_cache[scan8[0 + 8]] =
3441             nnz_cache[scan8[1 + 8]] =
3442             nnz_cache[scan8[2 + 8]] =
3443             nnz_cache[scan8[3 + 8]] = (h->cbp & 0x4000) >> 12;
3444
3445             nnz_cache[scan8[0 + 12]] =
3446             nnz_cache[scan8[1 + 12]] =
3447             nnz_cache[scan8[2 + 12]] =
3448             nnz_cache[scan8[3 + 12]] = (h->cbp & 0x8000) >> 12;
3449         }
3450     }
3451
3452     return 0;
3453 }
3454
3455 static void loop_filter(H264Context *h, int start_x, int end_x)
3456 {
3457     MpegEncContext *const s = &h->s;
3458     uint8_t *dest_y, *dest_cb, *dest_cr;
3459     int linesize, uvlinesize, mb_x, mb_y;
3460     const int end_mb_y       = s->mb_y + FRAME_MBAFF;
3461     const int old_slice_type = h->slice_type;
3462     const int pixel_shift    = h->pixel_shift;
3463     const int block_h        = 16 >> s->chroma_y_shift;
3464
3465     if (h->deblocking_filter) {
3466         for (mb_x = start_x; mb_x < end_x; mb_x++)
3467             for (mb_y = end_mb_y - FRAME_MBAFF; mb_y <= end_mb_y; mb_y++) {
3468                 int mb_xy, mb_type;
3469                 mb_xy         = h->mb_xy = mb_x + mb_y * s->mb_stride;
3470                 h->slice_num  = h->slice_table[mb_xy];
3471                 mb_type       = s->current_picture.f.mb_type[mb_xy];
3472                 h->list_count = h->list_counts[mb_xy];
3473
3474                 if (FRAME_MBAFF)
3475                     h->mb_mbaff               =
3476                     h->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
3477
3478                 s->mb_x = mb_x;
3479                 s->mb_y = mb_y;
3480                 dest_y  = s->current_picture.f.data[0] +
3481                           ((mb_x << pixel_shift) + mb_y * s->linesize) * 16;
3482                 dest_cb = s->current_picture.f.data[1] +
3483                           (mb_x << pixel_shift) * (8 << CHROMA444) +
3484                           mb_y * s->uvlinesize * block_h;
3485                 dest_cr = s->current_picture.f.data[2] +
3486                           (mb_x << pixel_shift) * (8 << CHROMA444) +
3487                           mb_y * s->uvlinesize * block_h;
3488                 // FIXME simplify above
3489
3490                 if (MB_FIELD) {
3491                     linesize   = h->mb_linesize   = s->linesize   * 2;
3492                     uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
3493                     if (mb_y & 1) { // FIXME move out of this function?
3494                         dest_y  -= s->linesize   * 15;
3495                         dest_cb -= s->uvlinesize * (block_h - 1);
3496                         dest_cr -= s->uvlinesize * (block_h - 1);
3497                     }
3498                 } else {
3499                     linesize   = h->mb_linesize   = s->linesize;
3500                     uvlinesize = h->mb_uvlinesize = s->uvlinesize;
3501                 }
3502                 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize,
3503                                  uvlinesize, 0);
3504                 if (fill_filter_caches(h, mb_type))
3505                     continue;
3506                 h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.f.qscale_table[mb_xy]);
3507                 h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.f.qscale_table[mb_xy]);
3508
3509                 if (FRAME_MBAFF) {
3510                     ff_h264_filter_mb(h, mb_x, mb_y, dest_y, dest_cb, dest_cr,
3511                                       linesize, uvlinesize);
3512                 } else {
3513                     ff_h264_filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb,
3514                                            dest_cr, linesize, uvlinesize);
3515                 }
3516             }
3517     }
3518     h->slice_type   = old_slice_type;
3519     s->mb_x         = end_x;
3520     s->mb_y         = end_mb_y - FRAME_MBAFF;
3521     h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
3522     h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
3523 }
3524
3525 static void predict_field_decoding_flag(H264Context *h)
3526 {
3527     MpegEncContext *const s = &h->s;
3528     const int mb_xy = s->mb_x + s->mb_y * s->mb_stride;
3529     int mb_type     = (h->slice_table[mb_xy - 1] == h->slice_num) ?
3530                       s->current_picture.f.mb_type[mb_xy - 1] :
3531                       (h->slice_table[mb_xy - s->mb_stride] == h->slice_num) ?
3532                       s->current_picture.f.mb_type[mb_xy - s->mb_stride] : 0;
3533     h->mb_mbaff     = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
3534 }
3535
3536 /**
3537  * Draw edges and report progress for the last MB row.
3538  */
3539 static void decode_finish_row(H264Context *h)
3540 {
3541     MpegEncContext *const s = &h->s;
3542     int top            = 16 * (s->mb_y      >> FIELD_PICTURE);
3543     int pic_height     = 16 *  s->mb_height >> FIELD_PICTURE;
3544     int height         =  16      << FRAME_MBAFF;
3545     int deblock_border = (16 + 4) << FRAME_MBAFF;
3546
3547     if (h->deblocking_filter) {
3548         if ((top + height) >= pic_height)
3549             height += deblock_border;
3550         top -= deblock_border;
3551     }
3552
3553     if (top >= pic_height || (top + height) < h->emu_edge_height)
3554         return;
3555
3556     height = FFMIN(height, pic_height - top);
3557     if (top < h->emu_edge_height) {
3558         height = top + height;
3559         top    = 0;
3560     }
3561
3562     ff_draw_horiz_band(s, top, height);
3563
3564     if (s->droppable)
3565         return;
3566
3567     ff_thread_report_progress(&s->current_picture_ptr->f, top + height - 1,
3568                               s->picture_structure == PICT_BOTTOM_FIELD);
3569 }
3570
3571 static int decode_slice(struct AVCodecContext *avctx, void *arg)
3572 {
3573     H264Context *h = *(void **)arg;
3574     MpegEncContext *const s = &h->s;
3575     const int part_mask     = s->partitioned_frame ? (ER_AC_END | ER_AC_ERROR)
3576                                                    : 0x7F;
3577     int lf_x_start = s->mb_x;
3578
3579     s->mb_skip_run = -1;
3580
3581     h->is_complex = FRAME_MBAFF || s->picture_structure != PICT_FRAME ||
3582                     s->codec_id != AV_CODEC_ID_H264 ||
3583                     (CONFIG_GRAY && (s->flags & CODEC_FLAG_GRAY));
3584
3585     if (h->pps.cabac) {
3586         /* realign */
3587         align_get_bits(&s->gb);
3588
3589         /* init cabac */
3590         ff_init_cabac_decoder(&h->cabac,
3591                               s->gb.buffer + get_bits_count(&s->gb) / 8,
3592                               (get_bits_left(&s->gb) + 7) / 8);
3593
3594         ff_h264_init_cabac_states(h);
3595
3596         for (;;) {
3597             // START_TIMER
3598             int ret = ff_h264_decode_mb_cabac(h);
3599             int eos;
3600             // STOP_TIMER("decode_mb_cabac")
3601
3602             if (ret >= 0)
3603                 ff_h264_hl_decode_mb(h);
3604
3605             // FIXME optimal? or let mb_decode decode 16x32 ?
3606             if (ret >= 0 && FRAME_MBAFF) {
3607                 s->mb_y++;
3608
3609                 ret = ff_h264_decode_mb_cabac(h);
3610
3611                 if (ret >= 0)
3612                     ff_h264_hl_decode_mb(h);
3613                 s->mb_y--;
3614             }
3615             eos = get_cabac_terminate(&h->cabac);
3616
3617             if ((s->workaround_bugs & FF_BUG_TRUNCATED) &&
3618                 h->cabac.bytestream > h->cabac.bytestream_end + 2) {
3619                 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x - 1,
3620                                 s->mb_y, ER_MB_END & part_mask);
3621                 if (s->mb_x >= lf_x_start)
3622                     loop_filter(h, lf_x_start, s->mb_x + 1);
3623                 return 0;
3624             }
3625             if (h->cabac.bytestream > h->cabac.bytestream_end + 2 )
3626                 av_log(h->s.avctx, AV_LOG_DEBUG, "bytestream overread %td\n", h->cabac.bytestream_end - h->cabac.bytestream);
3627             if (ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 4) {
3628                 av_log(h->s.avctx, AV_LOG_ERROR,
3629                        "error while decoding MB %d %d, bytestream (%td)\n",
3630                        s->mb_x, s->mb_y,
3631                        h->cabac.bytestream_end - h->cabac.bytestream);
3632                 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x,
3633                                 s->mb_y, ER_MB_ERROR & part_mask);
3634                 return -1;
3635             }
3636
3637             if (++s->mb_x >= s->mb_width) {
3638                 loop_filter(h, lf_x_start, s->mb_x);
3639                 s->mb_x = lf_x_start = 0;
3640                 decode_finish_row(h);
3641                 ++s->mb_y;
3642                 if (FIELD_OR_MBAFF_PICTURE) {
3643                     ++s->mb_y;
3644                     if (FRAME_MBAFF && s->mb_y < s->mb_height)
3645                         predict_field_decoding_flag(h);
3646                 }
3647             }
3648
3649             if (eos || s->mb_y >= s->mb_height) {
3650                 tprintf(s->avctx, "slice end %d %d\n",
3651                         get_bits_count(&s->gb), s->gb.size_in_bits);
3652                 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x - 1,
3653                                 s->mb_y, ER_MB_END & part_mask);
3654                 if (s->mb_x > lf_x_start)
3655                     loop_filter(h, lf_x_start, s->mb_x);
3656                 return 0;
3657             }
3658         }
3659     } else {
3660         for (;;) {
3661             int ret = ff_h264_decode_mb_cavlc(h);
3662
3663             if (ret >= 0)
3664                 ff_h264_hl_decode_mb(h);
3665
3666             // FIXME optimal? or let mb_decode decode 16x32 ?
3667             if (ret >= 0 && FRAME_MBAFF) {
3668                 s->mb_y++;
3669                 ret = ff_h264_decode_mb_cavlc(h);
3670
3671                 if (ret >= 0)
3672                     ff_h264_hl_decode_mb(h);
3673                 s->mb_y--;
3674             }
3675
3676             if (ret < 0) {
3677                 av_log(h->s.avctx, AV_LOG_ERROR,
3678                        "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
3679                 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x,
3680                                 s->mb_y, ER_MB_ERROR & part_mask);
3681                 return -1;
3682             }
3683
3684             if (++s->mb_x >= s->mb_width) {
3685                 loop_filter(h, lf_x_start, s->mb_x);
3686                 s->mb_x = lf_x_start = 0;
3687                 decode_finish_row(h);
3688                 ++s->mb_y;
3689                 if (FIELD_OR_MBAFF_PICTURE) {
3690                     ++s->mb_y;
3691                     if (FRAME_MBAFF && s->mb_y < s->mb_height)
3692                         predict_field_decoding_flag(h);
3693                 }
3694                 if (s->mb_y >= s->mb_height) {
3695                     tprintf(s->avctx, "slice end %d %d\n",
3696                             get_bits_count(&s->gb), s->gb.size_in_bits);
3697
3698                     if (   get_bits_left(&s->gb) == 0
3699                         || get_bits_left(&s->gb) > 0 && !(s->avctx->err_recognition & AV_EF_AGGRESSIVE)) {
3700                         ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y,
3701                                         s->mb_x - 1, s->mb_y,
3702                                         ER_MB_END & part_mask);
3703
3704                         return 0;
3705                     } else {
3706                         ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y,
3707                                         s->mb_x, s->mb_y,
3708                                         ER_MB_END & part_mask);
3709
3710                         return -1;
3711                     }
3712                 }
3713             }
3714
3715             if (get_bits_left(&s->gb) <= 0 && s->mb_skip_run <= 0) {
3716                 tprintf(s->avctx, "slice end %d %d\n",
3717                         get_bits_count(&s->gb), s->gb.size_in_bits);
3718                 if (get_bits_left(&s->gb) == 0) {
3719                     ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y,
3720                                     s->mb_x - 1, s->mb_y,
3721                                     ER_MB_END & part_mask);
3722                     if (s->mb_x > lf_x_start)
3723                         loop_filter(h, lf_x_start, s->mb_x);
3724
3725                     return 0;
3726                 } else {
3727                     ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x,
3728                                     s->mb_y, ER_MB_ERROR & part_mask);
3729
3730                     return -1;
3731                 }
3732             }
3733         }
3734     }
3735 }
3736
3737 /**
3738  * Call decode_slice() for each context.
3739  *
3740  * @param h h264 master context
3741  * @param context_count number of contexts to execute
3742  */
3743 static int execute_decode_slices(H264Context *h, int context_count)
3744 {
3745     MpegEncContext *const s     = &h->s;
3746     AVCodecContext *const avctx = s->avctx;
3747     H264Context *hx;
3748     int i;
3749
3750     if (s->avctx->hwaccel ||
3751         s->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU)
3752         return 0;
3753     if (context_count == 1) {
3754         return decode_slice(avctx, &h);
3755     } else {
3756         for (i = 1; i < context_count; i++) {
3757             hx                    = h->thread_context[i];
3758             hx->s.err_recognition = avctx->err_recognition;
3759             hx->s.error_count     = 0;
3760             hx->x264_build        = h->x264_build;
3761         }
3762
3763         avctx->execute(avctx, decode_slice, h->thread_context,
3764                        NULL, context_count, sizeof(void *));
3765
3766         /* pull back stuff from slices to master context */
3767         hx                   = h->thread_context[context_count - 1];
3768         s->mb_x              = hx->s.mb_x;
3769         s->mb_y              = hx->s.mb_y;
3770         s->droppable         = hx->s.droppable;
3771         s->picture_structure = hx->s.picture_structure;
3772         for (i = 1; i < context_count; i++)
3773             h->s.error_count += h->thread_context[i]->s.error_count;
3774     }
3775
3776     return 0;
3777 }
3778
3779 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size)
3780 {
3781     MpegEncContext *const s     = &h->s;
3782     AVCodecContext *const avctx = s->avctx;
3783     H264Context *hx; ///< thread context
3784     int buf_index;
3785     int context_count;
3786     int next_avc;
3787     int pass = !(avctx->active_thread_type & FF_THREAD_FRAME);
3788     int nals_needed = 0; ///< number of NALs that need decoding before the next frame thread starts
3789     int nal_index;
3790     int idr_cleared=0;
3791
3792     h->nal_unit_type= 0;
3793
3794     if(!s->slice_context_count)
3795          s->slice_context_count= 1;
3796     h->max_contexts = s->slice_context_count;
3797     if (!(s->flags2 & CODEC_FLAG2_CHUNKS)) {
3798         h->current_slice = 0;
3799         if (!s->first_field)
3800             s->current_picture_ptr = NULL;
3801         ff_h264_reset_sei(h);
3802     }
3803
3804     if (h->nal_length_size == 4) {
3805         if (buf_size > 8 && AV_RB32(buf) == 1 && AV_RB32(buf+5) > (unsigned)buf_size) {
3806             h->is_avc = 0;
3807         }else if(buf_size > 3 && AV_RB32(buf) > 1 && AV_RB32(buf) <= (unsigned)buf_size)
3808             h->is_avc = 1;
3809     }
3810
3811     for (; pass <= 1; pass++) {
3812         buf_index     = 0;
3813         context_count = 0;
3814         next_avc      = h->is_avc ? 0 : buf_size;
3815         nal_index     = 0;
3816         for (;;) {
3817             int consumed;
3818             int dst_length;
3819             int bit_length;
3820             const uint8_t *ptr;
3821             int i, nalsize = 0;
3822             int err;
3823
3824             if (buf_index >= next_avc) {
3825                 if (buf_index >= buf_size - h->nal_length_size)
3826                     break;
3827                 nalsize = 0;
3828                 for (i = 0; i < h->nal_length_size; i++)
3829                     nalsize = (nalsize << 8) | buf[buf_index++];
3830                 if (nalsize <= 0 || nalsize > buf_size - buf_index) {
3831                     av_log(h->s.avctx, AV_LOG_ERROR,
3832                            "AVC: nal size %d\n", nalsize);
3833                     break;
3834                 }
3835                 next_avc = buf_index + nalsize;
3836             } else {
3837                 // start code prefix search
3838                 for (; buf_index + 3 < next_avc; buf_index++)
3839                     // This should always succeed in the first iteration.
3840                     if (buf[buf_index]     == 0 &&
3841                         buf[buf_index + 1] == 0 &&
3842                         buf[buf_index + 2] == 1)
3843                         break;
3844
3845                 if (buf_index + 3 >= buf_size) {
3846                     buf_index = buf_size;
3847                     break;
3848                 }
3849
3850                 buf_index += 3;
3851                 if (buf_index >= next_avc)
3852                     continue;
3853             }
3854
3855             hx = h->thread_context[context_count];
3856
3857             ptr = ff_h264_decode_nal(hx, buf + buf_index, &dst_length,
3858                                      &consumed, next_avc - buf_index);
3859             if (ptr == NULL || dst_length < 0) {
3860                 buf_index = -1;
3861                 goto end;
3862             }
3863             i = buf_index + consumed;
3864             if ((s->workaround_bugs & FF_BUG_AUTODETECT) && i + 3 < next_avc &&
3865                 buf[i]     == 0x00 && buf[i + 1] == 0x00 &&
3866                 buf[i + 2] == 0x01 && buf[i + 3] == 0xE0)
3867                 s->workaround_bugs |= FF_BUG_TRUNCATED;
3868
3869             if (!(s->workaround_bugs & FF_BUG_TRUNCATED))
3870                 while(dst_length > 0 && ptr[dst_length - 1] == 0)
3871                     dst_length--;
3872             bit_length = !dst_length ? 0
3873                                      : (8 * dst_length -
3874                                         decode_rbsp_trailing(h, ptr + dst_length - 1));
3875
3876             if (s->avctx->debug & FF_DEBUG_STARTCODE)
3877                 av_log(h->s.avctx, AV_LOG_DEBUG, "NAL %d/%d at %d/%d length %d pass %d\n", hx->nal_unit_type, hx->nal_ref_idc, buf_index, buf_size, dst_length, pass);
3878
3879             if (h->is_avc && (nalsize != consumed) && nalsize)
3880                 av_log(h->s.avctx, AV_LOG_DEBUG,
3881                        "AVC: Consumed only %d bytes instead of %d\n",
3882                        consumed, nalsize);
3883
3884             buf_index += consumed;
3885             nal_index++;
3886
3887             if (pass == 0) {
3888                 /* packets can sometimes contain multiple PPS/SPS,
3889                  * e.g. two PAFF field pictures in one packet, or a demuxer
3890                  * which splits NALs strangely if so, when frame threading we
3891                  * can't start the next thread until we've read all of them */
3892                 switch (hx->nal_unit_type) {
3893                 case NAL_SPS:
3894                 case NAL_PPS:
3895                     nals_needed = nal_index;
3896                     break;
3897                 case NAL_IDR_SLICE:
3898                 case NAL_SLICE:
3899                     init_get_bits(&hx->s.gb, ptr, bit_length);
3900                     if (!get_ue_golomb(&hx->s.gb))
3901                         nals_needed = nal_index;
3902                 }
3903                 continue;
3904             }
3905
3906             // FIXME do not discard SEI id
3907             if (avctx->skip_frame >= AVDISCARD_NONREF && h->nal_ref_idc == 0)
3908                 continue;
3909
3910 again:
3911             err = 0;
3912
3913             if (h->decoding_extradata) {
3914                 switch (hx->nal_unit_type) {
3915                 case NAL_IDR_SLICE:
3916                 case NAL_SLICE:
3917                 case NAL_DPA:
3918                 case NAL_DPB:
3919                 case NAL_DPC:
3920                 case NAL_AUXILIARY_SLICE:
3921                     av_log(h->s.avctx, AV_LOG_WARNING, "Ignoring NAL %d in global header\n", hx->nal_unit_type);
3922                     hx->nal_unit_type = NAL_FILLER_DATA;
3923                 }
3924             }
3925
3926             switch (hx->nal_unit_type) {
3927             case NAL_IDR_SLICE:
3928                 if (h->nal_unit_type != NAL_IDR_SLICE) {
3929                     av_log(h->s.avctx, AV_LOG_ERROR,
3930                            "Invalid mix of idr and non-idr slices\n");
3931                     buf_index = -1;
3932                     goto end;
3933                 }
3934                 if(!idr_cleared)
3935                     idr(h); // FIXME ensure we don't lose some frames if there is reordering
3936                 idr_cleared = 1;
3937             case NAL_SLICE:
3938                 init_get_bits(&hx->s.gb, ptr, bit_length);
3939                 hx->intra_gb_ptr        =
3940                     hx->inter_gb_ptr    = &hx->s.gb;
3941                 hx->s.data_partitioning = 0;
3942
3943                 if ((err = decode_slice_header(hx, h)))
3944                     break;
3945
3946                 if (h->sei_recovery_frame_cnt >= 0 && (h->frame_num != h->sei_recovery_frame_cnt || hx->slice_type_nos != AV_PICTURE_TYPE_I))
3947                     h->valid_recovery_point = 1;
3948
3949                 if (   h->sei_recovery_frame_cnt >= 0
3950                     && (   h->recovery_frame<0
3951                         || ((h->recovery_frame - h->frame_num) & ((1 << h->sps.log2_max_frame_num)-1)) > h->sei_recovery_frame_cnt)) {
3952                     h->recovery_frame = (h->frame_num + h->sei_recovery_frame_cnt) %
3953                                         (1 << h->sps.log2_max_frame_num);
3954
3955                     if (!h->valid_recovery_point)
3956                         h->recovery_frame = h->frame_num;
3957                 }
3958
3959                 s->current_picture_ptr->f.key_frame |=
3960                         (hx->nal_unit_type == NAL_IDR_SLICE);
3961
3962                 if (h->recovery_frame == h->frame_num) {
3963                     s->current_picture_ptr->sync |= 1;
3964                     h->recovery_frame = -1;
3965                 }
3966
3967                 h->sync |= !!s->current_picture_ptr->f.key_frame;
3968                 h->sync |= 3*!!(s->flags2 & CODEC_FLAG2_SHOW_ALL);
3969                 s->current_picture_ptr->sync |= h->sync;
3970
3971                 if (h->current_slice == 1) {
3972                     if (!(s->flags2 & CODEC_FLAG2_CHUNKS))
3973                         decode_postinit(h, nal_index >= nals_needed);
3974
3975                     if (s->avctx->hwaccel &&
3976                         s->avctx->hwaccel->start_frame(s->avctx, NULL, 0) < 0)
3977                         return -1;
3978                     if (CONFIG_H264_VDPAU_DECODER &&
3979                         s->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU)
3980                         ff_vdpau_h264_picture_start(s);
3981                 }
3982
3983                 if (hx->redundant_pic_count == 0 &&
3984                     (avctx->skip_frame < AVDISCARD_NONREF ||
3985                      hx->nal_ref_idc) &&
3986                     (avctx->skip_frame < AVDISCARD_BIDIR  ||
3987                      hx->slice_type_nos != AV_PICTURE_TYPE_B) &&
3988                     (avctx->skip_frame < AVDISCARD_NONKEY ||
3989                      hx->slice_type_nos == AV_PICTURE_TYPE_I) &&
3990                     avctx->skip_frame < AVDISCARD_ALL) {
3991                     if (avctx->hwaccel) {
3992                         if (avctx->hwaccel->decode_slice(avctx,
3993                                                          &buf[buf_index - consumed],
3994                                                          consumed) < 0)
3995                             return -1;
3996                     } else if (CONFIG_H264_VDPAU_DECODER &&
3997                                s->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU) {
3998                         static const uint8_t start_code[] = {
3999                             0x00, 0x00, 0x01 };
4000                         ff_vdpau_add_data_chunk(s, start_code,
4001                                                 sizeof(start_code));
4002                         ff_vdpau_add_data_chunk(s, &buf[buf_index - consumed],
4003                                                 consumed);
4004                     } else
4005                         context_count++;
4006                 }
4007                 break;
4008             case NAL_DPA:
4009                 init_get_bits(&hx->s.gb, ptr, bit_length);
4010                 hx->intra_gb_ptr =
4011                 hx->inter_gb_ptr = NULL;
4012
4013                 if ((err = decode_slice_header(hx, h)) < 0)
4014                     break;
4015
4016                 hx->s.data_partitioning = 1;
4017                 break;
4018             case NAL_DPB:
4019                 init_get_bits(&hx->intra_gb, ptr, bit_length);
4020                 hx->intra_gb_ptr = &hx->intra_gb;
4021                 break;
4022             case NAL_DPC:
4023                 init_get_bits(&hx->inter_gb, ptr, bit_length);
4024                 hx->inter_gb_ptr = &hx->inter_gb;
4025
4026                 av_log(h->s.avctx, AV_LOG_ERROR, "Partitioned H.264 support is incomplete\n");
4027                 break;
4028
4029                 if (hx->redundant_pic_count == 0 &&
4030                     hx->intra_gb_ptr &&
4031                     hx->s.data_partitioning &&
4032                     s->current_picture_ptr &&
4033                     s->context_initialized &&
4034                     (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc) &&
4035                     (avctx->skip_frame < AVDISCARD_BIDIR  ||
4036                      hx->slice_type_nos != AV_PICTURE_TYPE_B) &&
4037                     (avctx->skip_frame < AVDISCARD_NONKEY ||
4038                      hx->slice_type_nos == AV_PICTURE_TYPE_I) &&
4039                     avctx->skip_frame < AVDISCARD_ALL)
4040                     context_count++;
4041                 break;
4042             case NAL_SEI:
4043                 init_get_bits(&s->gb, ptr, bit_length);
4044                 ff_h264_decode_sei(h);
4045                 break;
4046             case NAL_SPS:
4047                 init_get_bits(&s->gb, ptr, bit_length);
4048                 if (ff_h264_decode_seq_parameter_set(h) < 0 && (h->is_avc ? (nalsize != consumed) && nalsize : 1)) {
4049                     av_log(h->s.avctx, AV_LOG_DEBUG,
4050                            "SPS decoding failure, trying again with the complete NAL\n");
4051                     if (h->is_avc)
4052                         av_assert0(next_avc - buf_index + consumed == nalsize);
4053                     if ((next_avc - buf_index + consumed - 1) >= INT_MAX/8)
4054                         break;
4055                     init_get_bits(&s->gb, &buf[buf_index + 1 - consumed],
4056                                   8*(next_avc - buf_index + consumed - 1));
4057                     ff_h264_decode_seq_parameter_set(h);
4058                 }
4059
4060                 break;
4061             case NAL_PPS:
4062                 init_get_bits(&s->gb, ptr, bit_length);
4063                 ff_h264_decode_picture_parameter_set(h, bit_length);
4064                 break;
4065             case NAL_AUD:
4066             case NAL_END_SEQUENCE:
4067             case NAL_END_STREAM:
4068             case NAL_FILLER_DATA:
4069             case NAL_SPS_EXT:
4070             case NAL_AUXILIARY_SLICE:
4071                 break;
4072             default:
4073                 av_log(avctx, AV_LOG_DEBUG, "Unknown NAL code: %d (%d bits)\n",
4074                        hx->nal_unit_type, bit_length);
4075             }
4076
4077             if (context_count == h->max_contexts) {
4078                 execute_decode_slices(h, context_count);
4079                 context_count = 0;
4080             }
4081
4082             if (err < 0)
4083                 av_log(h->s.avctx, AV_LOG_ERROR, "decode_slice_header error\n");
4084             else if (err == 1) {
4085                 /* Slice could not be decoded in parallel mode, copy down
4086                  * NAL unit stuff to context 0 and restart. Note that
4087                  * rbsp_buffer is not transferred, but since we no longer
4088                  * run in parallel mode this should not be an issue. */
4089                 h->nal_unit_type = hx->nal_unit_type;
4090                 h->nal_ref_idc   = hx->nal_ref_idc;
4091                 hx               = h;
4092                 goto again;
4093             }
4094         }
4095     }
4096     if (context_count)
4097         execute_decode_slices(h, context_count);
4098
4099 end:
4100     /* clean up */
4101     if (s->current_picture_ptr && s->current_picture_ptr->owner2 == s &&
4102         !s->droppable) {
4103         ff_thread_report_progress(&s->current_picture_ptr->f, INT_MAX,
4104                                   s->picture_structure == PICT_BOTTOM_FIELD);
4105     }
4106
4107     return buf_index;
4108 }
4109
4110 /**
4111  * Return the number of bytes consumed for building the current frame.
4112  */
4113 static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size)
4114 {
4115     if (pos == 0)
4116         pos = 1;          // avoid infinite loops (i doubt that is needed but ...)
4117     if (pos + 10 > buf_size)
4118         pos = buf_size;                   // oops ;)
4119
4120     return pos;
4121 }
4122
4123 static int decode_frame(AVCodecContext *avctx, void *data,
4124                         int *got_frame, AVPacket *avpkt)
4125 {
4126     const uint8_t *buf = avpkt->data;
4127     int buf_size       = avpkt->size;
4128     H264Context *h     = avctx->priv_data;
4129     MpegEncContext *s  = &h->s;
4130     AVFrame *pict      = data;
4131     int buf_index      = 0;
4132     Picture *out;
4133     int i, out_idx;
4134
4135     s->flags  = avctx->flags;
4136     s->flags2 = avctx->flags2;
4137
4138     /* end of stream, output what is still in the buffers */
4139     if (buf_size == 0) {
4140  out:
4141
4142         s->current_picture_ptr = NULL;
4143
4144         // FIXME factorize this with the output code below
4145         out     = h->delayed_pic[0];
4146         out_idx = 0;
4147         for (i = 1;
4148              h->delayed_pic[i] &&
4149              !h->delayed_pic[i]->f.key_frame &&
4150              !h->delayed_pic[i]->mmco_reset;
4151              i++)
4152             if (h->delayed_pic[i]->poc < out->poc) {
4153                 out     = h->delayed_pic[i];
4154                 out_idx = i;
4155             }
4156
4157         for (i = out_idx; h->delayed_pic[i]; i++)
4158             h->delayed_pic[i] = h->delayed_pic[i + 1];
4159
4160         if (out) {
4161             *got_frame = 1;
4162             *pict      = out->f;
4163         }
4164
4165         return buf_index;
4166     }
4167     if(h->is_avc && buf_size >= 9 && buf[0]==1 && buf[2]==0 && (buf[4]&0xFC)==0xFC && (buf[5]&0x1F) && buf[8]==0x67){
4168         int cnt= buf[5]&0x1f;
4169         const uint8_t *p= buf+6;
4170         while(cnt--){
4171             int nalsize= AV_RB16(p) + 2;
4172             if(nalsize > buf_size - (p-buf) || p[2]!=0x67)
4173                 goto not_extra;
4174             p += nalsize;
4175         }
4176         cnt = *(p++);
4177         if(!cnt)
4178             goto not_extra;
4179         while(cnt--){
4180             int nalsize= AV_RB16(p) + 2;
4181             if(nalsize > buf_size - (p-buf) || p[2]!=0x68)
4182                 goto not_extra;
4183             p += nalsize;
4184         }
4185
4186         return ff_h264_decode_extradata(h, buf, buf_size);
4187     }
4188 not_extra:
4189
4190     buf_index = decode_nal_units(h, buf, buf_size);
4191     if (buf_index < 0)
4192         return -1;
4193
4194     if (!s->current_picture_ptr && h->nal_unit_type == NAL_END_SEQUENCE) {
4195         av_assert0(buf_index <= buf_size);
4196         goto out;
4197     }
4198
4199     if (!(s->flags2 & CODEC_FLAG2_CHUNKS) && !s->current_picture_ptr) {
4200         if (avctx->skip_frame >= AVDISCARD_NONREF ||
4201             buf_size >= 4 && !memcmp("Q264", buf, 4))
4202             return buf_size;
4203         av_log(avctx, AV_LOG_ERROR, "no frame!\n");
4204         return -1;
4205     }
4206
4207     if (!(s->flags2 & CODEC_FLAG2_CHUNKS) ||
4208         (s->mb_y >= s->mb_height && s->mb_height)) {
4209         if (s->flags2 & CODEC_FLAG2_CHUNKS)
4210             decode_postinit(h, 1);
4211
4212         field_end(h, 0);
4213
4214         /* Wait for second field. */
4215         *got_frame = 0;
4216         if (h->next_output_pic && (h->next_output_pic->sync || h->sync>1)) {
4217             *got_frame = 1;
4218             *pict      = h->next_output_pic->f;
4219         }
4220     }
4221
4222     assert(pict->data[0] || !*got_frame);
4223     ff_print_debug_info(s, pict);
4224
4225     return get_consumed_bytes(s, buf_index, buf_size);
4226 }
4227
4228 av_cold void ff_h264_free_context(H264Context *h)
4229 {
4230     int i;
4231
4232     free_tables(h, 1); // FIXME cleanup init stuff perhaps
4233
4234     for (i = 0; i < MAX_SPS_COUNT; i++)
4235         av_freep(h->sps_buffers + i);
4236
4237     for (i = 0; i < MAX_PPS_COUNT; i++)
4238         av_freep(h->pps_buffers + i);
4239 }
4240
4241 static av_cold int h264_decode_end(AVCodecContext *avctx)
4242 {
4243     H264Context *h    = avctx->priv_data;
4244     MpegEncContext *s = &h->s;
4245
4246     ff_h264_remove_all_refs(h);
4247     ff_h264_free_context(h);
4248
4249     ff_MPV_common_end(s);
4250
4251     // memset(h, 0, sizeof(H264Context));
4252
4253     return 0;
4254 }
4255
4256 static const AVProfile profiles[] = {
4257     { FF_PROFILE_H264_BASELINE,             "Baseline"              },
4258     { FF_PROFILE_H264_CONSTRAINED_BASELINE, "Constrained Baseline"  },
4259     { FF_PROFILE_H264_MAIN,                 "Main"                  },
4260     { FF_PROFILE_H264_EXTENDED,             "Extended"              },
4261     { FF_PROFILE_H264_HIGH,                 "High"                  },
4262     { FF_PROFILE_H264_HIGH_10,              "High 10"               },
4263     { FF_PROFILE_H264_HIGH_10_INTRA,        "High 10 Intra"         },
4264     { FF_PROFILE_H264_HIGH_422,             "High 4:2:2"            },
4265     { FF_PROFILE_H264_HIGH_422_INTRA,       "High 4:2:2 Intra"      },
4266     { FF_PROFILE_H264_HIGH_444,             "High 4:4:4"            },
4267     { FF_PROFILE_H264_HIGH_444_PREDICTIVE,  "High 4:4:4 Predictive" },
4268     { FF_PROFILE_H264_HIGH_444_INTRA,       "High 4:4:4 Intra"      },
4269     { FF_PROFILE_H264_CAVLC_444,            "CAVLC 4:4:4"           },
4270     { FF_PROFILE_UNKNOWN },
4271 };
4272
4273 static const AVOption h264_options[] = {
4274     {"is_avc", "is avc", offsetof(H264Context, is_avc), FF_OPT_TYPE_INT, {.i64 = 0}, 0, 1, 0},
4275     {"nal_length_size", "nal_length_size", offsetof(H264Context, nal_length_size), FF_OPT_TYPE_INT, {.i64 = 0}, 0, 4, 0},
4276     {NULL}
4277 };
4278
4279 static const AVClass h264_class = {
4280     .class_name = "H264 Decoder",
4281     .item_name  = av_default_item_name,
4282     .option     = h264_options,
4283     .version    = LIBAVUTIL_VERSION_INT,
4284 };
4285
4286 static const AVClass h264_vdpau_class = {
4287     .class_name = "H264 VDPAU Decoder",
4288     .item_name  = av_default_item_name,
4289     .option     = h264_options,
4290     .version    = LIBAVUTIL_VERSION_INT,
4291 };
4292
4293 AVCodec ff_h264_decoder = {
4294     .name                  = "h264",
4295     .type                  = AVMEDIA_TYPE_VIDEO,
4296     .id                    = AV_CODEC_ID_H264,
4297     .priv_data_size        = sizeof(H264Context),
4298     .init                  = ff_h264_decode_init,
4299     .close                 = h264_decode_end,
4300     .decode                = decode_frame,
4301     .capabilities          = /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 |
4302                              CODEC_CAP_DELAY | CODEC_CAP_SLICE_THREADS |
4303                              CODEC_CAP_FRAME_THREADS,
4304     .flush                 = flush_dpb,
4305     .long_name             = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10"),
4306     .init_thread_copy      = ONLY_IF_THREADS_ENABLED(decode_init_thread_copy),
4307     .update_thread_context = ONLY_IF_THREADS_ENABLED(decode_update_thread_context),
4308     .profiles              = NULL_IF_CONFIG_SMALL(profiles),
4309     .priv_class            = &h264_class,
4310 };
4311
4312 #if CONFIG_H264_VDPAU_DECODER
4313 AVCodec ff_h264_vdpau_decoder = {
4314     .name           = "h264_vdpau",
4315     .type           = AVMEDIA_TYPE_VIDEO,
4316     .id             = AV_CODEC_ID_H264,
4317     .priv_data_size = sizeof(H264Context),
4318     .init           = ff_h264_decode_init,
4319     .close          = h264_decode_end,
4320     .decode         = decode_frame,
4321     .capabilities   = CODEC_CAP_DR1 | CODEC_CAP_DELAY | CODEC_CAP_HWACCEL_VDPAU,
4322     .flush          = flush_dpb,
4323     .long_name      = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10 (VDPAU acceleration)"),
4324     .pix_fmts       = (const enum AVPixelFormat[]) { AV_PIX_FMT_VDPAU_H264,
4325                                                    AV_PIX_FMT_NONE},
4326     .profiles       = NULL_IF_CONFIG_SMALL(profiles),
4327     .priv_class     = &h264_vdpau_class,
4328 };
4329 #endif