]> git.sesse.net Git - ffmpeg/blob - libavcodec/h264.c
Merge commit '8729698d50739524665090e083d1bfdf28235724'
[ffmpeg] / 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->vdsp.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->vdsp.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->vdsp.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->vdsp.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->vdsp.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->vdsp.prefetch(src[0] + off, s->linesize, 4);
758         if (chroma_idc == 3 /* yuv444 */) {
759             s->vdsp.prefetch(src[1] + off, s->linesize, 4);
760             s->vdsp.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->vdsp.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                             int parse_extradata);
976
977 static av_cold void common_init(H264Context *h)
978 {
979     MpegEncContext *const s = &h->s;
980
981     s->width    = s->avctx->width;
982     s->height   = s->avctx->height;
983     s->codec_id = s->avctx->codec->id;
984
985     s->avctx->bits_per_raw_sample = 8;
986     h->cur_chroma_format_idc = 1;
987
988     ff_h264dsp_init(&h->h264dsp,
989                     s->avctx->bits_per_raw_sample, h->cur_chroma_format_idc);
990     ff_h264_pred_init(&h->hpc, s->codec_id,
991                       s->avctx->bits_per_raw_sample, h->cur_chroma_format_idc);
992
993     h->dequant_coeff_pps = -1;
994     s->unrestricted_mv   = 1;
995
996     s->dsp.dct_bits = 16;
997     /* needed so that IDCT permutation is known early */
998     ff_dsputil_init(&s->dsp, s->avctx);
999     ff_videodsp_init(&s->vdsp, 8);
1000
1001     memset(h->pps.scaling_matrix4, 16, 6 * 16 * sizeof(uint8_t));
1002     memset(h->pps.scaling_matrix8, 16, 2 * 64 * sizeof(uint8_t));
1003 }
1004
1005 int ff_h264_decode_extradata(H264Context *h, const uint8_t *buf, int size)
1006 {
1007     AVCodecContext *avctx = h->s.avctx;
1008
1009     if (!buf || size <= 0)
1010         return -1;
1011
1012     if (buf[0] == 1) {
1013         int i, cnt, nalsize;
1014         const unsigned char *p = buf;
1015
1016         h->is_avc = 1;
1017
1018         if (size < 7) {
1019             av_log(avctx, AV_LOG_ERROR, "avcC too short\n");
1020             return -1;
1021         }
1022         /* sps and pps in the avcC always have length coded with 2 bytes,
1023          * so put a fake nal_length_size = 2 while parsing them */
1024         h->nal_length_size = 2;
1025         // Decode sps from avcC
1026         cnt = *(p + 5) & 0x1f; // Number of sps
1027         p  += 6;
1028         for (i = 0; i < cnt; i++) {
1029             nalsize = AV_RB16(p) + 2;
1030             if(nalsize > size - (p-buf))
1031                 return -1;
1032             if (decode_nal_units(h, p, nalsize, 1) < 0) {
1033                 av_log(avctx, AV_LOG_ERROR,
1034                        "Decoding sps %d from avcC failed\n", i);
1035                 return -1;
1036             }
1037             p += nalsize;
1038         }
1039         // Decode pps from avcC
1040         cnt = *(p++); // Number of pps
1041         for (i = 0; i < cnt; i++) {
1042             nalsize = AV_RB16(p) + 2;
1043             if(nalsize > size - (p-buf))
1044                 return -1;
1045             if (decode_nal_units(h, p, nalsize, 1) < 0) {
1046                 av_log(avctx, AV_LOG_ERROR,
1047                        "Decoding pps %d from avcC failed\n", i);
1048                 return -1;
1049             }
1050             p += nalsize;
1051         }
1052         // Now store right nal length size, that will be used to parse all other nals
1053         h->nal_length_size = (buf[4] & 0x03) + 1;
1054     } else {
1055         h->is_avc = 0;
1056         if (decode_nal_units(h, buf, size, 1) < 0)
1057             return -1;
1058     }
1059     return size;
1060 }
1061
1062 av_cold int ff_h264_decode_init(AVCodecContext *avctx)
1063 {
1064     H264Context *h = avctx->priv_data;
1065     MpegEncContext *const s = &h->s;
1066     int i;
1067
1068     ff_MPV_decode_defaults(s);
1069
1070     s->avctx = avctx;
1071     common_init(h);
1072
1073     s->out_format      = FMT_H264;
1074     s->workaround_bugs = avctx->workaround_bugs;
1075
1076     /* set defaults */
1077     // s->decode_mb = ff_h263_decode_mb;
1078     s->quarter_sample = 1;
1079     if (!avctx->has_b_frames)
1080         s->low_delay = 1;
1081
1082     avctx->chroma_sample_location = AVCHROMA_LOC_LEFT;
1083
1084     ff_h264_decode_init_vlc();
1085
1086     h->pixel_shift = 0;
1087     h->sps.bit_depth_luma = avctx->bits_per_raw_sample = 8;
1088
1089     h->thread_context[0] = h;
1090     h->outputed_poc      = h->next_outputed_poc = INT_MIN;
1091     for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
1092         h->last_pocs[i] = INT_MIN;
1093     h->prev_poc_msb = 1 << 16;
1094     h->prev_frame_num = -1;
1095     h->x264_build   = -1;
1096     ff_h264_reset_sei(h);
1097     if (avctx->codec_id == AV_CODEC_ID_H264) {
1098         if (avctx->ticks_per_frame == 1) {
1099             if(s->avctx->time_base.den < INT_MAX/2) {
1100                 s->avctx->time_base.den *= 2;
1101             } else
1102                 s->avctx->time_base.num /= 2;
1103         }
1104         avctx->ticks_per_frame = 2;
1105     }
1106
1107     if (avctx->extradata_size > 0 && avctx->extradata &&
1108         ff_h264_decode_extradata(h, avctx->extradata, avctx->extradata_size) < 0) {
1109         ff_h264_free_context(h);
1110         return -1;
1111     }
1112
1113     if (h->sps.bitstream_restriction_flag &&
1114         s->avctx->has_b_frames < h->sps.num_reorder_frames) {
1115         s->avctx->has_b_frames = h->sps.num_reorder_frames;
1116         s->low_delay           = 0;
1117     }
1118
1119     ff_init_cabac_states();
1120
1121     return 0;
1122 }
1123
1124 #define IN_RANGE(a, b, size) (((a) >= (b)) && ((a) < ((b) + (size))))
1125
1126 static void copy_picture_range(Picture **to, Picture **from, int count,
1127                                MpegEncContext *new_base,
1128                                MpegEncContext *old_base)
1129 {
1130     int i;
1131
1132     for (i = 0; i < count; i++) {
1133         assert((IN_RANGE(from[i], old_base, sizeof(*old_base)) ||
1134                 IN_RANGE(from[i], old_base->picture,
1135                          sizeof(Picture) * old_base->picture_count) ||
1136                 !from[i]));
1137         to[i] = REBASE_PICTURE(from[i], new_base, old_base);
1138     }
1139 }
1140
1141 static void copy_parameter_set(void **to, void **from, int count, int size)
1142 {
1143     int i;
1144
1145     for (i = 0; i < count; i++) {
1146         if (to[i] && !from[i])
1147             av_freep(&to[i]);
1148         else if (from[i] && !to[i])
1149             to[i] = av_malloc(size);
1150
1151         if (from[i])
1152             memcpy(to[i], from[i], size);
1153     }
1154 }
1155
1156 static int decode_init_thread_copy(AVCodecContext *avctx)
1157 {
1158     H264Context *h = avctx->priv_data;
1159
1160     if (!avctx->internal->is_copy)
1161         return 0;
1162     memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
1163     memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
1164
1165     h->s.context_initialized = 0;
1166
1167     return 0;
1168 }
1169
1170 #define copy_fields(to, from, start_field, end_field)                   \
1171     memcpy(&to->start_field, &from->start_field,                        \
1172            (char *)&to->end_field - (char *)&to->start_field)
1173
1174 static int h264_slice_header_init(H264Context *, int);
1175
1176 static int h264_set_parameter_from_sps(H264Context *h);
1177
1178 static int decode_update_thread_context(AVCodecContext *dst,
1179                                         const AVCodecContext *src)
1180 {
1181     H264Context *h = dst->priv_data, *h1 = src->priv_data;
1182     MpegEncContext *const s = &h->s, *const s1 = &h1->s;
1183     int inited = s->context_initialized, err;
1184     int i;
1185
1186     if (dst == src)
1187         return 0;
1188
1189     if (inited &&
1190         (s->width      != s1->width      ||
1191          s->height     != s1->height     ||
1192          s->mb_width   != s1->mb_width   ||
1193          s->mb_height  != s1->mb_height  ||
1194          h->sps.bit_depth_luma    != h1->sps.bit_depth_luma    ||
1195          h->sps.chroma_format_idc != h1->sps.chroma_format_idc ||
1196          h->sps.colorspace        != h1->sps.colorspace)) {
1197
1198         av_freep(&h->bipred_scratchpad);
1199
1200         s->width     = s1->width;
1201         s->height    = s1->height;
1202         s->mb_height = s1->mb_height;
1203         h->b_stride  = h1->b_stride;
1204         // SPS/PPS
1205         copy_parameter_set((void **)h->sps_buffers, (void **)h1->sps_buffers,
1206                         MAX_SPS_COUNT, sizeof(SPS));
1207         h->sps = h1->sps;
1208         copy_parameter_set((void **)h->pps_buffers, (void **)h1->pps_buffers,
1209                         MAX_PPS_COUNT, sizeof(PPS));
1210         h->pps = h1->pps;
1211
1212         if ((err = h264_slice_header_init(h, 1)) < 0) {
1213             av_log(h->s.avctx, AV_LOG_ERROR, "h264_slice_header_init() failed");
1214             return err;
1215         }
1216         h->context_reinitialized = 1;
1217
1218         /* update linesize on resize for h264. The h264 decoder doesn't
1219          * necessarily call ff_MPV_frame_start in the new thread */
1220         s->linesize   = s1->linesize;
1221         s->uvlinesize = s1->uvlinesize;
1222
1223         /* copy block_offset since frame_start may not be called */
1224         memcpy(h->block_offset, h1->block_offset, sizeof(h->block_offset));
1225         h264_set_parameter_from_sps(h);
1226     }
1227
1228     err = ff_mpeg_update_thread_context(dst, src);
1229     if (err)
1230         return err;
1231
1232     if (!inited) {
1233         for (i = 0; i < MAX_SPS_COUNT; i++)
1234             av_freep(h->sps_buffers + i);
1235
1236         for (i = 0; i < MAX_PPS_COUNT; i++)
1237             av_freep(h->pps_buffers + i);
1238
1239         // copy all fields after MpegEnc
1240         memcpy(&h->s + 1, &h1->s + 1,
1241                sizeof(H264Context) - sizeof(MpegEncContext));
1242         memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
1243         memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
1244
1245         if (s1->context_initialized) {
1246         if (ff_h264_alloc_tables(h) < 0) {
1247             av_log(dst, AV_LOG_ERROR, "Could not allocate memory for h264\n");
1248             return AVERROR(ENOMEM);
1249         }
1250         context_init(h);
1251
1252         }
1253
1254         for (i = 0; i < 2; i++) {
1255             h->rbsp_buffer[i]      = NULL;
1256             h->rbsp_buffer_size[i] = 0;
1257         }
1258         h->bipred_scratchpad = NULL;
1259
1260         h->thread_context[0] = h;
1261
1262         s->dsp.clear_blocks(h->mb);
1263         s->dsp.clear_blocks(h->mb + (24 * 16 << h->pixel_shift));
1264     }
1265
1266     /* frame_start may not be called for the next thread (if it's decoding
1267      * a bottom field) so this has to be allocated here */
1268     if (!h->bipred_scratchpad && s->linesize)
1269         h->bipred_scratchpad = av_malloc(16 * 6 * s->linesize);
1270
1271     // extradata/NAL handling
1272     h->is_avc = h1->is_avc;
1273
1274     // SPS/PPS
1275     copy_parameter_set((void **)h->sps_buffers, (void **)h1->sps_buffers,
1276                        MAX_SPS_COUNT, sizeof(SPS));
1277     h->sps = h1->sps;
1278     copy_parameter_set((void **)h->pps_buffers, (void **)h1->pps_buffers,
1279                        MAX_PPS_COUNT, sizeof(PPS));
1280     h->pps = h1->pps;
1281
1282     // Dequantization matrices
1283     // FIXME these are big - can they be only copied when PPS changes?
1284     copy_fields(h, h1, dequant4_buffer, dequant4_coeff);
1285
1286     for (i = 0; i < 6; i++)
1287         h->dequant4_coeff[i] = h->dequant4_buffer[0] +
1288                                (h1->dequant4_coeff[i] - h1->dequant4_buffer[0]);
1289
1290     for (i = 0; i < 6; i++)
1291         h->dequant8_coeff[i] = h->dequant8_buffer[0] +
1292                                (h1->dequant8_coeff[i] - h1->dequant8_buffer[0]);
1293
1294     h->dequant_coeff_pps = h1->dequant_coeff_pps;
1295
1296     // POC timing
1297     copy_fields(h, h1, poc_lsb, redundant_pic_count);
1298
1299     // reference lists
1300     copy_fields(h, h1, ref_count, list_count);
1301     copy_fields(h, h1, ref_list, intra_gb);
1302     copy_fields(h, h1, short_ref, cabac_init_idc);
1303
1304     copy_picture_range(h->short_ref, h1->short_ref, 32, s, s1);
1305     copy_picture_range(h->long_ref, h1->long_ref, 32, s, s1);
1306     copy_picture_range(h->delayed_pic, h1->delayed_pic,
1307                        MAX_DELAYED_PIC_COUNT + 2, s, s1);
1308
1309     h->last_slice_type = h1->last_slice_type;
1310     h->sync            = h1->sync;
1311
1312     if (!s->current_picture_ptr)
1313         return 0;
1314
1315     if (!s->droppable) {
1316         err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1317         h->prev_poc_msb = h->poc_msb;
1318         h->prev_poc_lsb = h->poc_lsb;
1319     }
1320     h->prev_frame_num_offset = h->frame_num_offset;
1321     h->prev_frame_num        = h->frame_num;
1322     h->outputed_poc          = h->next_outputed_poc;
1323
1324     return err;
1325 }
1326
1327 int ff_h264_frame_start(H264Context *h)
1328 {
1329     MpegEncContext *const s = &h->s;
1330     int i;
1331     const int pixel_shift = h->pixel_shift;
1332
1333     if (ff_MPV_frame_start(s, s->avctx) < 0)
1334         return -1;
1335     ff_er_frame_start(s);
1336     /*
1337      * ff_MPV_frame_start uses pict_type to derive key_frame.
1338      * This is incorrect for H.264; IDR markings must be used.
1339      * Zero here; IDR markings per slice in frame or fields are ORed in later.
1340      * See decode_nal_units().
1341      */
1342     s->current_picture_ptr->f.key_frame = 0;
1343     s->current_picture_ptr->sync        = 0;
1344     s->current_picture_ptr->mmco_reset  = 0;
1345
1346     assert(s->linesize && s->uvlinesize);
1347
1348     for (i = 0; i < 16; i++) {
1349         h->block_offset[i]           = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * s->linesize * ((scan8[i] - scan8[0]) >> 3);
1350         h->block_offset[48 + i]      = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * s->linesize * ((scan8[i] - scan8[0]) >> 3);
1351     }
1352     for (i = 0; i < 16; i++) {
1353         h->block_offset[16 + i]      =
1354         h->block_offset[32 + i]      = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * s->uvlinesize * ((scan8[i] - scan8[0]) >> 3);
1355         h->block_offset[48 + 16 + i] =
1356         h->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * s->uvlinesize * ((scan8[i] - scan8[0]) >> 3);
1357     }
1358
1359     /* can't be in alloc_tables because linesize isn't known there.
1360      * FIXME: redo bipred weight to not require extra buffer? */
1361     for (i = 0; i < s->slice_context_count; i++)
1362         if (h->thread_context[i] && !h->thread_context[i]->bipred_scratchpad)
1363             h->thread_context[i]->bipred_scratchpad = av_malloc(16 * 6 * s->linesize);
1364
1365     /* Some macroblocks can be accessed before they're available in case
1366      * of lost slices, MBAFF or threading. */
1367     memset(h->slice_table, -1,
1368            (s->mb_height * s->mb_stride - 1) * sizeof(*h->slice_table));
1369
1370     // s->decode = (s->flags & CODEC_FLAG_PSNR) || !s->encoding ||
1371     //             s->current_picture.f.reference /* || h->contains_intra */ || 1;
1372
1373     /* We mark the current picture as non-reference after allocating it, so
1374      * that if we break out due to an error it can be released automatically
1375      * in the next ff_MPV_frame_start().
1376      * SVQ3 as well as most other codecs have only last/next/current and thus
1377      * get released even with set reference, besides SVQ3 and others do not
1378      * mark frames as reference later "naturally". */
1379     if (s->codec_id != AV_CODEC_ID_SVQ3)
1380         s->current_picture_ptr->f.reference = 0;
1381
1382     s->current_picture_ptr->field_poc[0]     =
1383         s->current_picture_ptr->field_poc[1] = INT_MAX;
1384
1385     h->next_output_pic = NULL;
1386
1387     assert(s->current_picture_ptr->long_ref == 0);
1388
1389     return 0;
1390 }
1391
1392 /**
1393  * Run setup operations that must be run after slice header decoding.
1394  * This includes finding the next displayed frame.
1395  *
1396  * @param h h264 master context
1397  * @param setup_finished enough NALs have been read that we can call
1398  * ff_thread_finish_setup()
1399  */
1400 static void decode_postinit(H264Context *h, int setup_finished)
1401 {
1402     MpegEncContext *const s = &h->s;
1403     Picture *out = s->current_picture_ptr;
1404     Picture *cur = s->current_picture_ptr;
1405     int i, pics, out_of_order, out_idx;
1406
1407     s->current_picture_ptr->f.qscale_type = FF_QSCALE_TYPE_H264;
1408     s->current_picture_ptr->f.pict_type   = s->pict_type;
1409
1410     if (h->next_output_pic)
1411         return;
1412
1413     if (cur->field_poc[0] == INT_MAX || cur->field_poc[1] == INT_MAX) {
1414         /* FIXME: if we have two PAFF fields in one packet, we can't start
1415          * the next thread here. If we have one field per packet, we can.
1416          * The check in decode_nal_units() is not good enough to find this
1417          * yet, so we assume the worst for now. */
1418         // if (setup_finished)
1419         //    ff_thread_finish_setup(s->avctx);
1420         return;
1421     }
1422
1423     cur->f.interlaced_frame = 0;
1424     cur->f.repeat_pict      = 0;
1425
1426     /* Signal interlacing information externally. */
1427     /* Prioritize picture timing SEI information over used
1428      * decoding process if it exists. */
1429
1430     if (h->sps.pic_struct_present_flag) {
1431         switch (h->sei_pic_struct) {
1432         case SEI_PIC_STRUCT_FRAME:
1433             break;
1434         case SEI_PIC_STRUCT_TOP_FIELD:
1435         case SEI_PIC_STRUCT_BOTTOM_FIELD:
1436             cur->f.interlaced_frame = 1;
1437             break;
1438         case SEI_PIC_STRUCT_TOP_BOTTOM:
1439         case SEI_PIC_STRUCT_BOTTOM_TOP:
1440             if (FIELD_OR_MBAFF_PICTURE)
1441                 cur->f.interlaced_frame = 1;
1442             else
1443                 // try to flag soft telecine progressive
1444                 cur->f.interlaced_frame = h->prev_interlaced_frame;
1445             break;
1446         case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
1447         case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
1448             /* Signal the possibility of telecined film externally
1449              * (pic_struct 5,6). From these hints, let the applications
1450              * decide if they apply deinterlacing. */
1451             cur->f.repeat_pict = 1;
1452             break;
1453         case SEI_PIC_STRUCT_FRAME_DOUBLING:
1454             // Force progressive here, doubling interlaced frame is a bad idea.
1455             cur->f.repeat_pict = 2;
1456             break;
1457         case SEI_PIC_STRUCT_FRAME_TRIPLING:
1458             cur->f.repeat_pict = 4;
1459             break;
1460         }
1461
1462         if ((h->sei_ct_type & 3) &&
1463             h->sei_pic_struct <= SEI_PIC_STRUCT_BOTTOM_TOP)
1464             cur->f.interlaced_frame = (h->sei_ct_type & (1 << 1)) != 0;
1465     } else {
1466         /* Derive interlacing flag from used decoding process. */
1467         cur->f.interlaced_frame = FIELD_OR_MBAFF_PICTURE;
1468     }
1469     h->prev_interlaced_frame = cur->f.interlaced_frame;
1470
1471     if (cur->field_poc[0] != cur->field_poc[1]) {
1472         /* Derive top_field_first from field pocs. */
1473         cur->f.top_field_first = cur->field_poc[0] < cur->field_poc[1];
1474     } else {
1475         if (cur->f.interlaced_frame || h->sps.pic_struct_present_flag) {
1476             /* Use picture timing SEI information. Even if it is a
1477              * information of a past frame, better than nothing. */
1478             if (h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM ||
1479                 h->sei_pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM_TOP)
1480                 cur->f.top_field_first = 1;
1481             else
1482                 cur->f.top_field_first = 0;
1483         } else {
1484             /* Most likely progressive */
1485             cur->f.top_field_first = 0;
1486         }
1487     }
1488
1489     cur->mmco_reset = h->mmco_reset;
1490     h->mmco_reset = 0;
1491     // FIXME do something with unavailable reference frames
1492
1493     /* Sort B-frames into display order */
1494
1495     if (h->sps.bitstream_restriction_flag &&
1496         s->avctx->has_b_frames < h->sps.num_reorder_frames) {
1497         s->avctx->has_b_frames = h->sps.num_reorder_frames;
1498         s->low_delay           = 0;
1499     }
1500
1501     if (s->avctx->strict_std_compliance >= FF_COMPLIANCE_STRICT &&
1502         !h->sps.bitstream_restriction_flag) {
1503         s->avctx->has_b_frames = MAX_DELAYED_PIC_COUNT - 1;
1504         s->low_delay           = 0;
1505     }
1506
1507     for (i = 0; 1; i++) {
1508         if(i == MAX_DELAYED_PIC_COUNT || cur->poc < h->last_pocs[i]){
1509             if(i)
1510                 h->last_pocs[i-1] = cur->poc;
1511             break;
1512         } else if(i) {
1513             h->last_pocs[i-1]= h->last_pocs[i];
1514         }
1515     }
1516     out_of_order = MAX_DELAYED_PIC_COUNT - i;
1517     if(   cur->f.pict_type == AV_PICTURE_TYPE_B
1518        || (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))
1519         out_of_order = FFMAX(out_of_order, 1);
1520     if (out_of_order == MAX_DELAYED_PIC_COUNT) {
1521         av_log(s->avctx, AV_LOG_VERBOSE, "Invalid POC %d<%d\n", cur->poc, h->last_pocs[0]);
1522         for (i = 1; i < MAX_DELAYED_PIC_COUNT; i++)
1523             h->last_pocs[i] = INT_MIN;
1524         h->last_pocs[0] = cur->poc;
1525         cur->mmco_reset = 1;
1526     } else if(s->avctx->has_b_frames < out_of_order && !h->sps.bitstream_restriction_flag){
1527         av_log(s->avctx, AV_LOG_VERBOSE, "Increasing reorder buffer to %d\n", out_of_order);
1528         s->avctx->has_b_frames = out_of_order;
1529         s->low_delay = 0;
1530     }
1531
1532     pics = 0;
1533     while (h->delayed_pic[pics])
1534         pics++;
1535
1536     av_assert0(pics <= MAX_DELAYED_PIC_COUNT);
1537
1538     h->delayed_pic[pics++] = cur;
1539     if (cur->f.reference == 0)
1540         cur->f.reference = DELAYED_PIC_REF;
1541
1542     out = h->delayed_pic[0];
1543     out_idx = 0;
1544     for (i = 1; h->delayed_pic[i] &&
1545                 !h->delayed_pic[i]->f.key_frame &&
1546                 !h->delayed_pic[i]->mmco_reset;
1547          i++)
1548         if (h->delayed_pic[i]->poc < out->poc) {
1549             out     = h->delayed_pic[i];
1550             out_idx = i;
1551         }
1552     if (s->avctx->has_b_frames == 0 &&
1553         (h->delayed_pic[0]->f.key_frame || h->delayed_pic[0]->mmco_reset))
1554         h->next_outputed_poc = INT_MIN;
1555     out_of_order = out->poc < h->next_outputed_poc;
1556
1557     if (out_of_order || pics > s->avctx->has_b_frames) {
1558         out->f.reference &= ~DELAYED_PIC_REF;
1559         // for frame threading, the owner must be the second field's thread or
1560         // else the first thread can release the picture and reuse it unsafely
1561         out->owner2       = s;
1562         for (i = out_idx; h->delayed_pic[i]; i++)
1563             h->delayed_pic[i] = h->delayed_pic[i + 1];
1564     }
1565     if (!out_of_order && pics > s->avctx->has_b_frames) {
1566         h->next_output_pic = out;
1567         if (out_idx == 0 && h->delayed_pic[0] && (h->delayed_pic[0]->f.key_frame || h->delayed_pic[0]->mmco_reset)) {
1568             h->next_outputed_poc = INT_MIN;
1569         } else
1570             h->next_outputed_poc = out->poc;
1571     } else {
1572         av_log(s->avctx, AV_LOG_DEBUG, "no picture %s\n", out_of_order ? "ooo" : "");
1573     }
1574
1575     if (h->next_output_pic && h->next_output_pic->sync) {
1576         h->sync |= 2;
1577     }
1578
1579     if (setup_finished)
1580         ff_thread_finish_setup(s->avctx);
1581 }
1582
1583 static av_always_inline void backup_mb_border(H264Context *h, uint8_t *src_y,
1584                                               uint8_t *src_cb, uint8_t *src_cr,
1585                                               int linesize, int uvlinesize,
1586                                               int simple)
1587 {
1588     MpegEncContext *const s = &h->s;
1589     uint8_t *top_border;
1590     int top_idx = 1;
1591     const int pixel_shift = h->pixel_shift;
1592     int chroma444 = CHROMA444;
1593     int chroma422 = CHROMA422;
1594
1595     src_y  -= linesize;
1596     src_cb -= uvlinesize;
1597     src_cr -= uvlinesize;
1598
1599     if (!simple && FRAME_MBAFF) {
1600         if (s->mb_y & 1) {
1601             if (!MB_MBAFF) {
1602                 top_border = h->top_borders[0][s->mb_x];
1603                 AV_COPY128(top_border, src_y + 15 * linesize);
1604                 if (pixel_shift)
1605                     AV_COPY128(top_border + 16, src_y + 15 * linesize + 16);
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 + 15 * uvlinesize);
1610                             AV_COPY128(top_border + 48, src_cb + 15 * uvlinesize + 16);
1611                             AV_COPY128(top_border + 64, src_cr + 15 * uvlinesize);
1612                             AV_COPY128(top_border + 80, src_cr + 15 * uvlinesize + 16);
1613                         } else {
1614                             AV_COPY128(top_border + 16, src_cb + 15 * uvlinesize);
1615                             AV_COPY128(top_border + 32, src_cr + 15 * uvlinesize);
1616                         }
1617                     } else if (chroma422) {
1618                         if (pixel_shift) {
1619                             AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
1620                             AV_COPY128(top_border + 48, src_cr + 15 * uvlinesize);
1621                         } else {
1622                             AV_COPY64(top_border + 16, src_cb + 15 * uvlinesize);
1623                             AV_COPY64(top_border + 24, src_cr + 15 * uvlinesize);
1624                         }
1625                     } else {
1626                         if (pixel_shift) {
1627                             AV_COPY128(top_border + 32, src_cb + 7 * uvlinesize);
1628                             AV_COPY128(top_border + 48, src_cr + 7 * uvlinesize);
1629                         } else {
1630                             AV_COPY64(top_border + 16, src_cb + 7 * uvlinesize);
1631                             AV_COPY64(top_border + 24, src_cr + 7 * uvlinesize);
1632                         }
1633                     }
1634                 }
1635             }
1636         } else if (MB_MBAFF) {
1637             top_idx = 0;
1638         } else
1639             return;
1640     }
1641
1642     top_border = h->top_borders[top_idx][s->mb_x];
1643     /* There are two lines saved, the line above the top macroblock
1644      * of a pair, and the line above the bottom macroblock. */
1645     AV_COPY128(top_border, src_y + 16 * linesize);
1646     if (pixel_shift)
1647         AV_COPY128(top_border + 16, src_y + 16 * linesize + 16);
1648
1649     if (simple || !CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
1650         if (chroma444) {
1651             if (pixel_shift) {
1652                 AV_COPY128(top_border + 32, src_cb + 16 * linesize);
1653                 AV_COPY128(top_border + 48, src_cb + 16 * linesize + 16);
1654                 AV_COPY128(top_border + 64, src_cr + 16 * linesize);
1655                 AV_COPY128(top_border + 80, src_cr + 16 * linesize + 16);
1656             } else {
1657                 AV_COPY128(top_border + 16, src_cb + 16 * linesize);
1658                 AV_COPY128(top_border + 32, src_cr + 16 * linesize);
1659             }
1660         } else if (chroma422) {
1661             if (pixel_shift) {
1662                 AV_COPY128(top_border + 32, src_cb + 16 * uvlinesize);
1663                 AV_COPY128(top_border + 48, src_cr + 16 * uvlinesize);
1664             } else {
1665                 AV_COPY64(top_border + 16, src_cb + 16 * uvlinesize);
1666                 AV_COPY64(top_border + 24, src_cr + 16 * uvlinesize);
1667             }
1668         } else {
1669             if (pixel_shift) {
1670                 AV_COPY128(top_border + 32, src_cb + 8 * uvlinesize);
1671                 AV_COPY128(top_border + 48, src_cr + 8 * uvlinesize);
1672             } else {
1673                 AV_COPY64(top_border + 16, src_cb + 8 * uvlinesize);
1674                 AV_COPY64(top_border + 24, src_cr + 8 * uvlinesize);
1675             }
1676         }
1677     }
1678 }
1679
1680 static av_always_inline void xchg_mb_border(H264Context *h, uint8_t *src_y,
1681                                             uint8_t *src_cb, uint8_t *src_cr,
1682                                             int linesize, int uvlinesize,
1683                                             int xchg, int chroma444,
1684                                             int simple, int pixel_shift)
1685 {
1686     MpegEncContext *const s = &h->s;
1687     int deblock_topleft;
1688     int deblock_top;
1689     int top_idx = 1;
1690     uint8_t *top_border_m1;
1691     uint8_t *top_border;
1692
1693     if (!simple && FRAME_MBAFF) {
1694         if (s->mb_y & 1) {
1695             if (!MB_MBAFF)
1696                 return;
1697         } else {
1698             top_idx = MB_MBAFF ? 0 : 1;
1699         }
1700     }
1701
1702     if (h->deblocking_filter == 2) {
1703         deblock_topleft = h->slice_table[h->mb_xy - 1 - s->mb_stride] == h->slice_num;
1704         deblock_top     = h->top_type;
1705     } else {
1706         deblock_topleft = (s->mb_x > 0);
1707         deblock_top     = (s->mb_y > !!MB_FIELD);
1708     }
1709
1710     src_y  -= linesize   + 1 + pixel_shift;
1711     src_cb -= uvlinesize + 1 + pixel_shift;
1712     src_cr -= uvlinesize + 1 + pixel_shift;
1713
1714     top_border_m1 = h->top_borders[top_idx][s->mb_x - 1];
1715     top_border    = h->top_borders[top_idx][s->mb_x];
1716
1717 #define XCHG(a, b, xchg)                        \
1718     if (pixel_shift) {                          \
1719         if (xchg) {                             \
1720             AV_SWAP64(b + 0, a + 0);            \
1721             AV_SWAP64(b + 8, a + 8);            \
1722         } else {                                \
1723             AV_COPY128(b, a);                   \
1724         }                                       \
1725     } else if (xchg)                            \
1726         AV_SWAP64(b, a);                        \
1727     else                                        \
1728         AV_COPY64(b, a);
1729
1730     if (deblock_top) {
1731         if (deblock_topleft) {
1732             XCHG(top_border_m1 + (8 << pixel_shift),
1733                  src_y - (7 << pixel_shift), 1);
1734         }
1735         XCHG(top_border + (0 << pixel_shift), src_y + (1 << pixel_shift), xchg);
1736         XCHG(top_border + (8 << pixel_shift), src_y + (9 << pixel_shift), 1);
1737         if (s->mb_x + 1 < s->mb_width) {
1738             XCHG(h->top_borders[top_idx][s->mb_x + 1],
1739                  src_y + (17 << pixel_shift), 1);
1740         }
1741     }
1742     if (simple || !CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
1743         if (chroma444) {
1744             if (deblock_topleft) {
1745                 XCHG(top_border_m1 + (24 << pixel_shift), src_cb - (7 << pixel_shift), 1);
1746                 XCHG(top_border_m1 + (40 << pixel_shift), src_cr - (7 << pixel_shift), 1);
1747             }
1748             XCHG(top_border + (16 << pixel_shift), src_cb + (1 << pixel_shift), xchg);
1749             XCHG(top_border + (24 << pixel_shift), src_cb + (9 << pixel_shift), 1);
1750             XCHG(top_border + (32 << pixel_shift), src_cr + (1 << pixel_shift), xchg);
1751             XCHG(top_border + (40 << pixel_shift), src_cr + (9 << pixel_shift), 1);
1752             if (s->mb_x + 1 < s->mb_width) {
1753                 XCHG(h->top_borders[top_idx][s->mb_x + 1] + (16 << pixel_shift), src_cb + (17 << pixel_shift), 1);
1754                 XCHG(h->top_borders[top_idx][s->mb_x + 1] + (32 << pixel_shift), src_cr + (17 << pixel_shift), 1);
1755             }
1756         } else {
1757             if (deblock_top) {
1758                 if (deblock_topleft) {
1759                     XCHG(top_border_m1 + (16 << pixel_shift), src_cb - (7 << pixel_shift), 1);
1760                     XCHG(top_border_m1 + (24 << pixel_shift), src_cr - (7 << pixel_shift), 1);
1761                 }
1762                 XCHG(top_border + (16 << pixel_shift), src_cb + 1 + pixel_shift, 1);
1763                 XCHG(top_border + (24 << pixel_shift), src_cr + 1 + pixel_shift, 1);
1764             }
1765         }
1766     }
1767 }
1768
1769 static av_always_inline int dctcoef_get(DCTELEM *mb, int high_bit_depth,
1770                                         int index)
1771 {
1772     if (high_bit_depth) {
1773         return AV_RN32A(((int32_t *)mb) + index);
1774     } else
1775         return AV_RN16A(mb + index);
1776 }
1777
1778 static av_always_inline void dctcoef_set(DCTELEM *mb, int high_bit_depth,
1779                                          int index, int value)
1780 {
1781     if (high_bit_depth) {
1782         AV_WN32A(((int32_t *)mb) + index, value);
1783     } else
1784         AV_WN16A(mb + index, value);
1785 }
1786
1787 static av_always_inline void hl_decode_mb_predict_luma(H264Context *h,
1788                                                        int mb_type, int is_h264,
1789                                                        int simple,
1790                                                        int transform_bypass,
1791                                                        int pixel_shift,
1792                                                        int *block_offset,
1793                                                        int linesize,
1794                                                        uint8_t *dest_y, int p)
1795 {
1796     MpegEncContext *const s = &h->s;
1797     void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1798     void (*idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
1799     int i;
1800     int qscale = p == 0 ? s->qscale : h->chroma_qp[p - 1];
1801     block_offset += 16 * p;
1802     if (IS_INTRA4x4(mb_type)) {
1803         if (simple || !s->encoding) {
1804             if (IS_8x8DCT(mb_type)) {
1805                 if (transform_bypass) {
1806                     idct_dc_add  =
1807                     idct_add     = s->dsp.add_pixels8;
1808                 } else {
1809                     idct_dc_add = h->h264dsp.h264_idct8_dc_add;
1810                     idct_add    = h->h264dsp.h264_idct8_add;
1811                 }
1812                 for (i = 0; i < 16; i += 4) {
1813                     uint8_t *const ptr = dest_y + block_offset[i];
1814                     const int dir      = h->intra4x4_pred_mode_cache[scan8[i]];
1815                     if (transform_bypass && h->sps.profile_idc == 244 && dir <= 1) {
1816                         h->hpc.pred8x8l_add[dir](ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
1817                     } else {
1818                         const int nnz = h->non_zero_count_cache[scan8[i + p * 16]];
1819                         h->hpc.pred8x8l[dir](ptr, (h->topleft_samples_available << i) & 0x8000,
1820                                              (h->topright_samples_available << i) & 0x4000, linesize);
1821                         if (nnz) {
1822                             if (nnz == 1 && dctcoef_get(h->mb, pixel_shift, i * 16 + p * 256))
1823                                 idct_dc_add(ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
1824                             else
1825                                 idct_add(ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
1826                         }
1827                     }
1828                 }
1829             } else {
1830                 if (transform_bypass) {
1831                     idct_dc_add  =
1832                         idct_add = s->dsp.add_pixels4;
1833                 } else {
1834                     idct_dc_add = h->h264dsp.h264_idct_dc_add;
1835                     idct_add    = h->h264dsp.h264_idct_add;
1836                 }
1837                 for (i = 0; i < 16; i++) {
1838                     uint8_t *const ptr = dest_y + block_offset[i];
1839                     const int dir      = h->intra4x4_pred_mode_cache[scan8[i]];
1840
1841                     if (transform_bypass && h->sps.profile_idc == 244 && dir <= 1) {
1842                         h->hpc.pred4x4_add[dir](ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
1843                     } else {
1844                         uint8_t *topright;
1845                         int nnz, tr;
1846                         uint64_t tr_high;
1847                         if (dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED) {
1848                             const int topright_avail = (h->topright_samples_available << i) & 0x8000;
1849                             av_assert2(s->mb_y || linesize <= block_offset[i]);
1850                             if (!topright_avail) {
1851                                 if (pixel_shift) {
1852                                     tr_high  = ((uint16_t *)ptr)[3 - linesize / 2] * 0x0001000100010001ULL;
1853                                     topright = (uint8_t *)&tr_high;
1854                                 } else {
1855                                     tr       = ptr[3 - linesize] * 0x01010101u;
1856                                     topright = (uint8_t *)&tr;
1857                                 }
1858                             } else
1859                                 topright = ptr + (4 << pixel_shift) - linesize;
1860                         } else
1861                             topright = NULL;
1862
1863                         h->hpc.pred4x4[dir](ptr, topright, linesize);
1864                         nnz = h->non_zero_count_cache[scan8[i + p * 16]];
1865                         if (nnz) {
1866                             if (is_h264) {
1867                                 if (nnz == 1 && dctcoef_get(h->mb, pixel_shift, i * 16 + p * 256))
1868                                     idct_dc_add(ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
1869                                 else
1870                                     idct_add(ptr, h->mb + (i * 16 + p * 256 << pixel_shift), linesize);
1871                             } else if (CONFIG_SVQ3_DECODER)
1872                                 ff_svq3_add_idct_c(ptr, h->mb + i * 16 + p * 256, linesize, qscale, 0);
1873                         }
1874                     }
1875                 }
1876             }
1877         }
1878     } else {
1879         h->hpc.pred16x16[h->intra16x16_pred_mode](dest_y, linesize);
1880         if (is_h264) {
1881             if (h->non_zero_count_cache[scan8[LUMA_DC_BLOCK_INDEX + p]]) {
1882                 if (!transform_bypass)
1883                     h->h264dsp.h264_luma_dc_dequant_idct(h->mb + (p * 256 << pixel_shift),
1884                                                          h->mb_luma_dc[p],
1885                                                          h->dequant4_coeff[p][qscale][0]);
1886                 else {
1887                     static const uint8_t dc_mapping[16] = {
1888                          0 * 16,  1 * 16,  4 * 16,  5 * 16,
1889                          2 * 16,  3 * 16,  6 * 16,  7 * 16,
1890                          8 * 16,  9 * 16, 12 * 16, 13 * 16,
1891                         10 * 16, 11 * 16, 14 * 16, 15 * 16 };
1892                     for (i = 0; i < 16; i++)
1893                         dctcoef_set(h->mb + (p * 256 << pixel_shift),
1894                                     pixel_shift, dc_mapping[i],
1895                                     dctcoef_get(h->mb_luma_dc[p],
1896                                                 pixel_shift, i));
1897                 }
1898             }
1899         } else if (CONFIG_SVQ3_DECODER)
1900             ff_svq3_luma_dc_dequant_idct_c(h->mb + p * 256,
1901                                            h->mb_luma_dc[p], qscale);
1902     }
1903 }
1904
1905 static av_always_inline void hl_decode_mb_idct_luma(H264Context *h, int mb_type,
1906                                                     int is_h264, int simple,
1907                                                     int transform_bypass,
1908                                                     int pixel_shift,
1909                                                     int *block_offset,
1910                                                     int linesize,
1911                                                     uint8_t *dest_y, int p)
1912 {
1913     MpegEncContext *const s = &h->s;
1914     void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
1915     int i;
1916     block_offset += 16 * p;
1917     if (!IS_INTRA4x4(mb_type)) {
1918         if (is_h264) {
1919             if (IS_INTRA16x16(mb_type)) {
1920                 if (transform_bypass) {
1921                     if (h->sps.profile_idc == 244 &&
1922                         (h->intra16x16_pred_mode == VERT_PRED8x8 ||
1923                          h->intra16x16_pred_mode == HOR_PRED8x8)) {
1924                         h->hpc.pred16x16_add[h->intra16x16_pred_mode](dest_y, block_offset,
1925                                                                       h->mb + (p * 256 << pixel_shift),
1926                                                                       linesize);
1927                     } else {
1928                         for (i = 0; i < 16; i++)
1929                             if (h->non_zero_count_cache[scan8[i + p * 16]] ||
1930                                 dctcoef_get(h->mb, pixel_shift, i * 16 + p * 256))
1931                                 s->dsp.add_pixels4(dest_y + block_offset[i],
1932                                                    h->mb + (i * 16 + p * 256 << pixel_shift),
1933                                                    linesize);
1934                     }
1935                 } else {
1936                     h->h264dsp.h264_idct_add16intra(dest_y, block_offset,
1937                                                     h->mb + (p * 256 << pixel_shift),
1938                                                     linesize,
1939                                                     h->non_zero_count_cache + p * 5 * 8);
1940                 }
1941             } else if (h->cbp & 15) {
1942                 if (transform_bypass) {
1943                     const int di = IS_8x8DCT(mb_type) ? 4 : 1;
1944                     idct_add = IS_8x8DCT(mb_type) ? s->dsp.add_pixels8
1945                                                   : s->dsp.add_pixels4;
1946                     for (i = 0; i < 16; i += di)
1947                         if (h->non_zero_count_cache[scan8[i + p * 16]])
1948                             idct_add(dest_y + block_offset[i],
1949                                      h->mb + (i * 16 + p * 256 << pixel_shift),
1950                                      linesize);
1951                 } else {
1952                     if (IS_8x8DCT(mb_type))
1953                         h->h264dsp.h264_idct8_add4(dest_y, block_offset,
1954                                                    h->mb + (p * 256 << pixel_shift),
1955                                                    linesize,
1956                                                    h->non_zero_count_cache + p * 5 * 8);
1957                     else
1958                         h->h264dsp.h264_idct_add16(dest_y, block_offset,
1959                                                    h->mb + (p * 256 << pixel_shift),
1960                                                    linesize,
1961                                                    h->non_zero_count_cache + p * 5 * 8);
1962                 }
1963             }
1964         } else if (CONFIG_SVQ3_DECODER) {
1965             for (i = 0; i < 16; i++)
1966                 if (h->non_zero_count_cache[scan8[i + p * 16]] || h->mb[i * 16 + p * 256]) {
1967                     // FIXME benchmark weird rule, & below
1968                     uint8_t *const ptr = dest_y + block_offset[i];
1969                     ff_svq3_add_idct_c(ptr, h->mb + i * 16 + p * 256, linesize,
1970                                        s->qscale, IS_INTRA(mb_type) ? 1 : 0);
1971                 }
1972         }
1973     }
1974 }
1975
1976 #define BITS   8
1977 #define SIMPLE 1
1978 #include "h264_mb_template.c"
1979
1980 #undef  BITS
1981 #define BITS   16
1982 #include "h264_mb_template.c"
1983
1984 #undef  SIMPLE
1985 #define SIMPLE 0
1986 #include "h264_mb_template.c"
1987
1988 void ff_h264_hl_decode_mb(H264Context *h)
1989 {
1990     MpegEncContext *const s = &h->s;
1991     const int mb_xy   = h->mb_xy;
1992     const int mb_type = s->current_picture.f.mb_type[mb_xy];
1993     int is_complex    = CONFIG_SMALL || h->is_complex || IS_INTRA_PCM(mb_type) || s->qscale == 0;
1994
1995     if (CHROMA444) {
1996         if (is_complex || h->pixel_shift)
1997             hl_decode_mb_444_complex(h);
1998         else
1999             hl_decode_mb_444_simple_8(h);
2000     } else if (is_complex) {
2001         hl_decode_mb_complex(h);
2002     } else if (h->pixel_shift) {
2003         hl_decode_mb_simple_16(h);
2004     } else
2005         hl_decode_mb_simple_8(h);
2006 }
2007
2008 static int pred_weight_table(H264Context *h)
2009 {
2010     MpegEncContext *const s = &h->s;
2011     int list, i;
2012     int luma_def, chroma_def;
2013
2014     h->use_weight             = 0;
2015     h->use_weight_chroma      = 0;
2016     h->luma_log2_weight_denom = get_ue_golomb(&s->gb);
2017     if (h->sps.chroma_format_idc)
2018         h->chroma_log2_weight_denom = get_ue_golomb(&s->gb);
2019     luma_def   = 1 << h->luma_log2_weight_denom;
2020     chroma_def = 1 << h->chroma_log2_weight_denom;
2021
2022     for (list = 0; list < 2; list++) {
2023         h->luma_weight_flag[list]   = 0;
2024         h->chroma_weight_flag[list] = 0;
2025         for (i = 0; i < h->ref_count[list]; i++) {
2026             int luma_weight_flag, chroma_weight_flag;
2027
2028             luma_weight_flag = get_bits1(&s->gb);
2029             if (luma_weight_flag) {
2030                 h->luma_weight[i][list][0] = get_se_golomb(&s->gb);
2031                 h->luma_weight[i][list][1] = get_se_golomb(&s->gb);
2032                 if (h->luma_weight[i][list][0] != luma_def ||
2033                     h->luma_weight[i][list][1] != 0) {
2034                     h->use_weight             = 1;
2035                     h->luma_weight_flag[list] = 1;
2036                 }
2037             } else {
2038                 h->luma_weight[i][list][0] = luma_def;
2039                 h->luma_weight[i][list][1] = 0;
2040             }
2041
2042             if (h->sps.chroma_format_idc) {
2043                 chroma_weight_flag = get_bits1(&s->gb);
2044                 if (chroma_weight_flag) {
2045                     int j;
2046                     for (j = 0; j < 2; j++) {
2047                         h->chroma_weight[i][list][j][0] = get_se_golomb(&s->gb);
2048                         h->chroma_weight[i][list][j][1] = get_se_golomb(&s->gb);
2049                         if (h->chroma_weight[i][list][j][0] != chroma_def ||
2050                             h->chroma_weight[i][list][j][1] != 0) {
2051                             h->use_weight_chroma = 1;
2052                             h->chroma_weight_flag[list] = 1;
2053                         }
2054                     }
2055                 } else {
2056                     int j;
2057                     for (j = 0; j < 2; j++) {
2058                         h->chroma_weight[i][list][j][0] = chroma_def;
2059                         h->chroma_weight[i][list][j][1] = 0;
2060                     }
2061                 }
2062             }
2063         }
2064         if (h->slice_type_nos != AV_PICTURE_TYPE_B)
2065             break;
2066     }
2067     h->use_weight = h->use_weight || h->use_weight_chroma;
2068     return 0;
2069 }
2070
2071 /**
2072  * Initialize implicit_weight table.
2073  * @param field  0/1 initialize the weight for interlaced MBAFF
2074  *                -1 initializes the rest
2075  */
2076 static void implicit_weight_table(H264Context *h, int field)
2077 {
2078     MpegEncContext *const s = &h->s;
2079     int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
2080
2081     for (i = 0; i < 2; i++) {
2082         h->luma_weight_flag[i]   = 0;
2083         h->chroma_weight_flag[i] = 0;
2084     }
2085
2086     if (field < 0) {
2087         if (s->picture_structure == PICT_FRAME) {
2088             cur_poc = s->current_picture_ptr->poc;
2089         } else {
2090             cur_poc = s->current_picture_ptr->field_poc[s->picture_structure - 1];
2091         }
2092         if (h->ref_count[0] == 1 && h->ref_count[1] == 1 && !FRAME_MBAFF &&
2093             h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2 * cur_poc) {
2094             h->use_weight = 0;
2095             h->use_weight_chroma = 0;
2096             return;
2097         }
2098         ref_start  = 0;
2099         ref_count0 = h->ref_count[0];
2100         ref_count1 = h->ref_count[1];
2101     } else {
2102         cur_poc    = s->current_picture_ptr->field_poc[field];
2103         ref_start  = 16;
2104         ref_count0 = 16 + 2 * h->ref_count[0];
2105         ref_count1 = 16 + 2 * h->ref_count[1];
2106     }
2107
2108     h->use_weight               = 2;
2109     h->use_weight_chroma        = 2;
2110     h->luma_log2_weight_denom   = 5;
2111     h->chroma_log2_weight_denom = 5;
2112
2113     for (ref0 = ref_start; ref0 < ref_count0; ref0++) {
2114         int poc0 = h->ref_list[0][ref0].poc;
2115         for (ref1 = ref_start; ref1 < ref_count1; ref1++) {
2116             int w = 32;
2117             if (!h->ref_list[0][ref0].long_ref && !h->ref_list[1][ref1].long_ref) {
2118                 int poc1 = h->ref_list[1][ref1].poc;
2119                 int td   = av_clip(poc1 - poc0, -128, 127);
2120                 if (td) {
2121                     int tb = av_clip(cur_poc - poc0, -128, 127);
2122                     int tx = (16384 + (FFABS(td) >> 1)) / td;
2123                     int dist_scale_factor = (tb * tx + 32) >> 8;
2124                     if (dist_scale_factor >= -64 && dist_scale_factor <= 128)
2125                         w = 64 - dist_scale_factor;
2126                 }
2127             }
2128             if (field < 0) {
2129                 h->implicit_weight[ref0][ref1][0] =
2130                 h->implicit_weight[ref0][ref1][1] = w;
2131             } else {
2132                 h->implicit_weight[ref0][ref1][field] = w;
2133             }
2134         }
2135     }
2136 }
2137
2138 /**
2139  * instantaneous decoder refresh.
2140  */
2141 static void idr(H264Context *h)
2142 {
2143     int i;
2144     ff_h264_remove_all_refs(h);
2145     h->prev_frame_num        = 0;
2146     h->prev_frame_num_offset = 0;
2147     h->prev_poc_msb          = 1<<16;
2148     h->prev_poc_lsb          = 0;
2149     for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
2150         h->last_pocs[i] = INT_MIN;
2151 }
2152
2153 /* forget old pics after a seek */
2154 static void flush_change(H264Context *h)
2155 {
2156     h->outputed_poc = h->next_outputed_poc = INT_MIN;
2157     h->prev_interlaced_frame = 1;
2158     idr(h);
2159     h->prev_frame_num = -1;
2160     if (h->s.current_picture_ptr)
2161         h->s.current_picture_ptr->f.reference = 0;
2162     h->s.first_field = 0;
2163     memset(h->ref_list[0], 0, sizeof(h->ref_list[0]));
2164     memset(h->ref_list[1], 0, sizeof(h->ref_list[1]));
2165     memset(h->default_ref_list[0], 0, sizeof(h->default_ref_list[0]));
2166     memset(h->default_ref_list[1], 0, sizeof(h->default_ref_list[1]));
2167     ff_h264_reset_sei(h);
2168     h->recovery_frame= -1;
2169     h->sync= 0;
2170     h->list_count = 0;
2171     h->current_slice = 0;
2172 }
2173
2174 /* forget old pics after a seek */
2175 static void flush_dpb(AVCodecContext *avctx)
2176 {
2177     H264Context *h = avctx->priv_data;
2178     int i;
2179
2180     for (i = 0; i <= MAX_DELAYED_PIC_COUNT; i++) {
2181         if (h->delayed_pic[i])
2182             h->delayed_pic[i]->f.reference = 0;
2183         h->delayed_pic[i] = NULL;
2184     }
2185
2186     flush_change(h);
2187     ff_mpeg_flush(avctx);
2188 }
2189
2190 static int init_poc(H264Context *h)
2191 {
2192     MpegEncContext *const s = &h->s;
2193     const int max_frame_num = 1 << h->sps.log2_max_frame_num;
2194     int field_poc[2];
2195     Picture *cur = s->current_picture_ptr;
2196
2197     h->frame_num_offset = h->prev_frame_num_offset;
2198     if (h->frame_num < h->prev_frame_num)
2199         h->frame_num_offset += max_frame_num;
2200
2201     if (h->sps.poc_type == 0) {
2202         const int max_poc_lsb = 1 << h->sps.log2_max_poc_lsb;
2203
2204         if (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb / 2)
2205             h->poc_msb = h->prev_poc_msb + max_poc_lsb;
2206         else if (h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb / 2)
2207             h->poc_msb = h->prev_poc_msb - max_poc_lsb;
2208         else
2209             h->poc_msb = h->prev_poc_msb;
2210         field_poc[0] =
2211         field_poc[1] = h->poc_msb + h->poc_lsb;
2212         if (s->picture_structure == PICT_FRAME)
2213             field_poc[1] += h->delta_poc_bottom;
2214     } else if (h->sps.poc_type == 1) {
2215         int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
2216         int i;
2217
2218         if (h->sps.poc_cycle_length != 0)
2219             abs_frame_num = h->frame_num_offset + h->frame_num;
2220         else
2221             abs_frame_num = 0;
2222
2223         if (h->nal_ref_idc == 0 && abs_frame_num > 0)
2224             abs_frame_num--;
2225
2226         expected_delta_per_poc_cycle = 0;
2227         for (i = 0; i < h->sps.poc_cycle_length; i++)
2228             // FIXME integrate during sps parse
2229             expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[i];
2230
2231         if (abs_frame_num > 0) {
2232             int poc_cycle_cnt          = (abs_frame_num - 1) / h->sps.poc_cycle_length;
2233             int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
2234
2235             expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
2236             for (i = 0; i <= frame_num_in_poc_cycle; i++)
2237                 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[i];
2238         } else
2239             expectedpoc = 0;
2240
2241         if (h->nal_ref_idc == 0)
2242             expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
2243
2244         field_poc[0] = expectedpoc + h->delta_poc[0];
2245         field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
2246
2247         if (s->picture_structure == PICT_FRAME)
2248             field_poc[1] += h->delta_poc[1];
2249     } else {
2250         int poc = 2 * (h->frame_num_offset + h->frame_num);
2251
2252         if (!h->nal_ref_idc)
2253             poc--;
2254
2255         field_poc[0] = poc;
2256         field_poc[1] = poc;
2257     }
2258
2259     if (s->picture_structure != PICT_BOTTOM_FIELD)
2260         s->current_picture_ptr->field_poc[0] = field_poc[0];
2261     if (s->picture_structure != PICT_TOP_FIELD)
2262         s->current_picture_ptr->field_poc[1] = field_poc[1];
2263     cur->poc = FFMIN(cur->field_poc[0], cur->field_poc[1]);
2264
2265     return 0;
2266 }
2267
2268 /**
2269  * initialize scan tables
2270  */
2271 static void init_scan_tables(H264Context *h)
2272 {
2273     int i;
2274     for (i = 0; i < 16; i++) {
2275 #define T(x) (x >> 2) | ((x << 2) & 0xF)
2276         h->zigzag_scan[i] = T(zigzag_scan[i]);
2277         h->field_scan[i]  = T(field_scan[i]);
2278 #undef T
2279     }
2280     for (i = 0; i < 64; i++) {
2281 #define T(x) (x >> 3) | ((x & 7) << 3)
2282         h->zigzag_scan8x8[i]       = T(ff_zigzag_direct[i]);
2283         h->zigzag_scan8x8_cavlc[i] = T(zigzag_scan8x8_cavlc[i]);
2284         h->field_scan8x8[i]        = T(field_scan8x8[i]);
2285         h->field_scan8x8_cavlc[i]  = T(field_scan8x8_cavlc[i]);
2286 #undef T
2287     }
2288     if (h->sps.transform_bypass) { // FIXME same ugly
2289         memcpy(h->zigzag_scan_q0          , zigzag_scan             , sizeof(h->zigzag_scan_q0         ));
2290         memcpy(h->zigzag_scan8x8_q0       , ff_zigzag_direct        , sizeof(h->zigzag_scan8x8_q0      ));
2291         memcpy(h->zigzag_scan8x8_cavlc_q0 , zigzag_scan8x8_cavlc    , sizeof(h->zigzag_scan8x8_cavlc_q0));
2292         memcpy(h->field_scan_q0           , field_scan              , sizeof(h->field_scan_q0          ));
2293         memcpy(h->field_scan8x8_q0        , field_scan8x8           , sizeof(h->field_scan8x8_q0       ));
2294         memcpy(h->field_scan8x8_cavlc_q0  , field_scan8x8_cavlc     , sizeof(h->field_scan8x8_cavlc_q0 ));
2295     } else {
2296         memcpy(h->zigzag_scan_q0          , h->zigzag_scan          , sizeof(h->zigzag_scan_q0         ));
2297         memcpy(h->zigzag_scan8x8_q0       , h->zigzag_scan8x8       , sizeof(h->zigzag_scan8x8_q0      ));
2298         memcpy(h->zigzag_scan8x8_cavlc_q0 , h->zigzag_scan8x8_cavlc , sizeof(h->zigzag_scan8x8_cavlc_q0));
2299         memcpy(h->field_scan_q0           , h->field_scan           , sizeof(h->field_scan_q0          ));
2300         memcpy(h->field_scan8x8_q0        , h->field_scan8x8        , sizeof(h->field_scan8x8_q0       ));
2301         memcpy(h->field_scan8x8_cavlc_q0  , h->field_scan8x8_cavlc  , sizeof(h->field_scan8x8_cavlc_q0 ));
2302     }
2303 }
2304
2305 static int field_end(H264Context *h, int in_setup)
2306 {
2307     MpegEncContext *const s     = &h->s;
2308     AVCodecContext *const avctx = s->avctx;
2309     int err = 0;
2310     s->mb_y = 0;
2311
2312     if (!in_setup && !s->droppable)
2313         ff_thread_report_progress(&s->current_picture_ptr->f, INT_MAX,
2314                                   s->picture_structure == PICT_BOTTOM_FIELD);
2315
2316     if (CONFIG_H264_VDPAU_DECODER &&
2317         s->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU)
2318         ff_vdpau_h264_set_reference_frames(s);
2319
2320     if (in_setup || !(avctx->active_thread_type & FF_THREAD_FRAME)) {
2321         if (!s->droppable) {
2322             err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
2323             h->prev_poc_msb = h->poc_msb;
2324             h->prev_poc_lsb = h->poc_lsb;
2325         }
2326         h->prev_frame_num_offset = h->frame_num_offset;
2327         h->prev_frame_num        = h->frame_num;
2328         h->outputed_poc          = h->next_outputed_poc;
2329     }
2330
2331     if (avctx->hwaccel) {
2332         if (avctx->hwaccel->end_frame(avctx) < 0)
2333             av_log(avctx, AV_LOG_ERROR,
2334                    "hardware accelerator failed to decode picture\n");
2335     }
2336
2337     if (CONFIG_H264_VDPAU_DECODER &&
2338         s->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU)
2339         ff_vdpau_h264_picture_complete(s);
2340
2341     /*
2342      * FIXME: Error handling code does not seem to support interlaced
2343      * when slices span multiple rows
2344      * The ff_er_add_slice calls don't work right for bottom
2345      * fields; they cause massive erroneous error concealing
2346      * Error marking covers both fields (top and bottom).
2347      * This causes a mismatched s->error_count
2348      * and a bad error table. Further, the error count goes to
2349      * INT_MAX when called for bottom field, because mb_y is
2350      * past end by one (callers fault) and resync_mb_y != 0
2351      * causes problems for the first MB line, too.
2352      */
2353     if (!FIELD_PICTURE)
2354         ff_er_frame_end(s);
2355
2356     ff_MPV_frame_end(s);
2357
2358     h->current_slice = 0;
2359
2360     return err;
2361 }
2362
2363 /**
2364  * Replicate H264 "master" context to thread contexts.
2365  */
2366 static int clone_slice(H264Context *dst, H264Context *src)
2367 {
2368     int ret;
2369
2370     memcpy(dst->block_offset, src->block_offset, sizeof(dst->block_offset));
2371     dst->s.current_picture_ptr = src->s.current_picture_ptr;
2372     dst->s.current_picture     = src->s.current_picture;
2373     dst->s.linesize            = src->s.linesize;
2374     dst->s.uvlinesize          = src->s.uvlinesize;
2375     dst->s.first_field         = src->s.first_field;
2376
2377     if (!dst->s.edge_emu_buffer &&
2378         (ret = ff_mpv_frame_size_alloc(&dst->s, dst->s.linesize))) {
2379         av_log(dst->s.avctx, AV_LOG_ERROR,
2380                "Failed to allocate scratch buffers\n");
2381         return ret;
2382     }
2383
2384     dst->prev_poc_msb          = src->prev_poc_msb;
2385     dst->prev_poc_lsb          = src->prev_poc_lsb;
2386     dst->prev_frame_num_offset = src->prev_frame_num_offset;
2387     dst->prev_frame_num        = src->prev_frame_num;
2388     dst->short_ref_count       = src->short_ref_count;
2389
2390     memcpy(dst->short_ref,        src->short_ref,        sizeof(dst->short_ref));
2391     memcpy(dst->long_ref,         src->long_ref,         sizeof(dst->long_ref));
2392     memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
2393     memcpy(dst->ref_list,         src->ref_list,         sizeof(dst->ref_list));
2394
2395     memcpy(dst->dequant4_coeff,   src->dequant4_coeff,   sizeof(src->dequant4_coeff));
2396     memcpy(dst->dequant8_coeff,   src->dequant8_coeff,   sizeof(src->dequant8_coeff));
2397
2398     return 0;
2399 }
2400
2401 /**
2402  * Compute profile from profile_idc and constraint_set?_flags.
2403  *
2404  * @param sps SPS
2405  *
2406  * @return profile as defined by FF_PROFILE_H264_*
2407  */
2408 int ff_h264_get_profile(SPS *sps)
2409 {
2410     int profile = sps->profile_idc;
2411
2412     switch (sps->profile_idc) {
2413     case FF_PROFILE_H264_BASELINE:
2414         // constraint_set1_flag set to 1
2415         profile |= (sps->constraint_set_flags & 1 << 1) ? FF_PROFILE_H264_CONSTRAINED : 0;
2416         break;
2417     case FF_PROFILE_H264_HIGH_10:
2418     case FF_PROFILE_H264_HIGH_422:
2419     case FF_PROFILE_H264_HIGH_444_PREDICTIVE:
2420         // constraint_set3_flag set to 1
2421         profile |= (sps->constraint_set_flags & 1 << 3) ? FF_PROFILE_H264_INTRA : 0;
2422         break;
2423     }
2424
2425     return profile;
2426 }
2427
2428 static int h264_set_parameter_from_sps(H264Context *h)
2429 {
2430     MpegEncContext *s = &h->s;
2431
2432     if (s->flags & CODEC_FLAG_LOW_DELAY ||
2433         (h->sps.bitstream_restriction_flag &&
2434          !h->sps.num_reorder_frames)) {
2435         if (s->avctx->has_b_frames > 1 || h->delayed_pic[0])
2436             av_log(h->s.avctx, AV_LOG_WARNING, "Delayed frames seen. "
2437                    "Reenabling low delay requires a codec flush.\n");
2438         else
2439             s->low_delay = 1;
2440     }
2441
2442     if (s->avctx->has_b_frames < 2)
2443         s->avctx->has_b_frames = !s->low_delay;
2444
2445     if (s->avctx->bits_per_raw_sample != h->sps.bit_depth_luma ||
2446         h->cur_chroma_format_idc      != h->sps.chroma_format_idc) {
2447         if (s->avctx->codec &&
2448             s->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU &&
2449             (h->sps.bit_depth_luma != 8 || h->sps.chroma_format_idc > 1)) {
2450             av_log(s->avctx, AV_LOG_ERROR,
2451                    "VDPAU decoding does not support video colorspace.\n");
2452             return AVERROR_INVALIDDATA;
2453         }
2454         if (h->sps.bit_depth_luma >= 8 && h->sps.bit_depth_luma <= 14 &&
2455             h->sps.bit_depth_luma != 11 && h->sps.bit_depth_luma != 13 &&
2456                 (h->sps.bit_depth_luma != 9 || !CHROMA422)) {
2457             s->avctx->bits_per_raw_sample = h->sps.bit_depth_luma;
2458             h->cur_chroma_format_idc      = h->sps.chroma_format_idc;
2459             h->pixel_shift                = h->sps.bit_depth_luma > 8;
2460
2461             ff_h264dsp_init(&h->h264dsp, h->sps.bit_depth_luma,
2462                             h->sps.chroma_format_idc);
2463             ff_h264_pred_init(&h->hpc, s->codec_id, h->sps.bit_depth_luma,
2464                               h->sps.chroma_format_idc);
2465             s->dsp.dct_bits = h->sps.bit_depth_luma > 8 ? 32 : 16;
2466             ff_dsputil_init(&s->dsp, s->avctx);
2467             ff_videodsp_init(&s->vdsp, h->sps.bit_depth_luma);
2468         } else {
2469             av_log(s->avctx, AV_LOG_ERROR, "Unsupported bit depth: %d\n",
2470                    h->sps.bit_depth_luma);
2471             return AVERROR_INVALIDDATA;
2472         }
2473     }
2474     return 0;
2475 }
2476
2477 static enum PixelFormat get_pixel_format(H264Context *h)
2478 {
2479     MpegEncContext *const s  = &h->s;
2480     switch (h->sps.bit_depth_luma) {
2481     case 9:
2482         if (CHROMA444) {
2483             if (s->avctx->colorspace == AVCOL_SPC_RGB) {
2484                 return AV_PIX_FMT_GBRP9;
2485             } else
2486                 return AV_PIX_FMT_YUV444P9;
2487         } else if (CHROMA422)
2488             return AV_PIX_FMT_YUV422P9;
2489         else
2490             return AV_PIX_FMT_YUV420P9;
2491         break;
2492     case 10:
2493         if (CHROMA444) {
2494             if (s->avctx->colorspace == AVCOL_SPC_RGB) {
2495                 return AV_PIX_FMT_GBRP10;
2496             } else
2497                 return AV_PIX_FMT_YUV444P10;
2498         } else if (CHROMA422)
2499             return AV_PIX_FMT_YUV422P10;
2500         else
2501             return AV_PIX_FMT_YUV420P10;
2502         break;
2503     case 12:
2504         if (CHROMA444) {
2505             if (s->avctx->colorspace == AVCOL_SPC_RGB) {
2506                 return AV_PIX_FMT_GBRP12;
2507             } else
2508                 return AV_PIX_FMT_YUV444P12;
2509         } else if (CHROMA422)
2510             return AV_PIX_FMT_YUV422P12;
2511         else
2512             return AV_PIX_FMT_YUV420P12;
2513         break;
2514     case 14:
2515         if (CHROMA444) {
2516             if (s->avctx->colorspace == AVCOL_SPC_RGB) {
2517                 return AV_PIX_FMT_GBRP14;
2518             } else
2519                 return AV_PIX_FMT_YUV444P14;
2520         } else if (CHROMA422)
2521             return AV_PIX_FMT_YUV422P14;
2522         else
2523             return AV_PIX_FMT_YUV420P14;
2524         break;
2525     case 8:
2526         if (CHROMA444) {
2527             if (s->avctx->colorspace == AVCOL_SPC_RGB) {
2528                 av_log(h->s.avctx, AV_LOG_DEBUG, "Detected GBR colorspace.\n");
2529                 return AV_PIX_FMT_GBR24P;
2530             } else if (s->avctx->colorspace == AVCOL_SPC_YCGCO) {
2531                 av_log(h->s.avctx, AV_LOG_WARNING, "Detected unsupported YCgCo colorspace.\n");
2532             }
2533             return s->avctx->color_range == AVCOL_RANGE_JPEG ? AV_PIX_FMT_YUVJ444P
2534                                                                 : AV_PIX_FMT_YUV444P;
2535         } else if (CHROMA422) {
2536             return s->avctx->color_range == AVCOL_RANGE_JPEG ? AV_PIX_FMT_YUVJ422P
2537                                                              : AV_PIX_FMT_YUV422P;
2538         } else {
2539             return s->avctx->get_format(s->avctx, s->avctx->codec->pix_fmts ?
2540                                         s->avctx->codec->pix_fmts :
2541                                         s->avctx->color_range == AVCOL_RANGE_JPEG ?
2542                                         hwaccel_pixfmt_list_h264_jpeg_420 :
2543                                         ff_hwaccel_pixfmt_list_420);
2544         }
2545         break;
2546     default:
2547         av_log(s->avctx, AV_LOG_ERROR,
2548                "Unsupported bit depth: %d\n", h->sps.bit_depth_luma);
2549         return AVERROR_INVALIDDATA;
2550     }
2551 }
2552
2553 static int h264_slice_header_init(H264Context *h, int reinit)
2554 {
2555     MpegEncContext *const s  = &h->s;
2556     int i, ret;
2557
2558     if(   FFALIGN(s->avctx->width , 16                                 ) == s->width
2559         && FFALIGN(s->avctx->height, 16*(2 - h->sps.frame_mbs_only_flag)) == s->height
2560         && !h->sps.crop_right && !h->sps.crop_bottom
2561         && (s->avctx->width != s->width || s->avctx->height && s->height)
2562     ) {
2563         av_log(h->s.avctx, AV_LOG_DEBUG, "Using externally provided dimensions\n");
2564         s->avctx->coded_width  = s->width;
2565         s->avctx->coded_height = s->height;
2566     } else{
2567         avcodec_set_dimensions(s->avctx, s->width, s->height);
2568         s->avctx->width  -= (2>>CHROMA444)*FFMIN(h->sps.crop_right, (8<<CHROMA444)-1);
2569         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);
2570     }
2571
2572     s->avctx->sample_aspect_ratio = h->sps.sar;
2573     av_assert0(s->avctx->sample_aspect_ratio.den);
2574
2575     if (h->sps.timing_info_present_flag) {
2576         int64_t den = h->sps.time_scale;
2577         if (h->x264_build < 44U)
2578             den *= 2;
2579         av_reduce(&s->avctx->time_base.num, &s->avctx->time_base.den,
2580                   h->sps.num_units_in_tick, den, 1 << 30);
2581     }
2582
2583     s->avctx->hwaccel = ff_find_hwaccel(s->avctx->codec->id, s->avctx->pix_fmt);
2584
2585     if (reinit) {
2586         free_tables(h, 0);
2587         if ((ret = ff_MPV_common_frame_size_change(s)) < 0) {
2588             av_log(h->s.avctx, AV_LOG_ERROR, "ff_MPV_common_frame_size_change() failed.\n");
2589             return ret;
2590         }
2591     } else {
2592         if ((ret = ff_MPV_common_init(s) < 0)) {
2593             av_log(h->s.avctx, AV_LOG_ERROR, "ff_MPV_common_init() failed.\n");
2594             return ret;
2595         }
2596     }
2597     s->first_field = 0;
2598     h->prev_interlaced_frame = 1;
2599
2600     init_scan_tables(h);
2601     if (ff_h264_alloc_tables(h) < 0) {
2602         av_log(h->s.avctx, AV_LOG_ERROR,
2603                "Could not allocate memory for h264\n");
2604         return AVERROR(ENOMEM);
2605     }
2606
2607     if (!HAVE_THREADS || !(s->avctx->active_thread_type & FF_THREAD_SLICE)) {
2608         if (context_init(h) < 0) {
2609             av_log(h->s.avctx, AV_LOG_ERROR, "context_init() failed.\n");
2610             return -1;
2611         }
2612     } else {
2613         for (i = 1; i < s->slice_context_count; i++) {
2614             H264Context *c;
2615             c = h->thread_context[i] = av_malloc(sizeof(H264Context));
2616             memcpy(c, h->s.thread_context[i], sizeof(MpegEncContext));
2617             memset(&c->s + 1, 0, sizeof(H264Context) - sizeof(MpegEncContext));
2618             c->h264dsp     = h->h264dsp;
2619             c->sps         = h->sps;
2620             c->pps         = h->pps;
2621             c->pixel_shift = h->pixel_shift;
2622             c->cur_chroma_format_idc = h->cur_chroma_format_idc;
2623             init_scan_tables(c);
2624             clone_tables(c, h, i);
2625         }
2626
2627         for (i = 0; i < s->slice_context_count; i++)
2628             if (context_init(h->thread_context[i]) < 0) {
2629                 av_log(h->s.avctx, AV_LOG_ERROR, "context_init() failed.\n");
2630                 return -1;
2631             }
2632     }
2633
2634     return 0;
2635 }
2636
2637 /**
2638  * Decode a slice header.
2639  * This will also call ff_MPV_common_init() and frame_start() as needed.
2640  *
2641  * @param h h264context
2642  * @param h0 h264 master context (differs from 'h' when doing sliced based
2643  *           parallel decoding)
2644  *
2645  * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
2646  */
2647 static int decode_slice_header(H264Context *h, H264Context *h0)
2648 {
2649     MpegEncContext *const s  = &h->s;
2650     MpegEncContext *const s0 = &h0->s;
2651     unsigned int first_mb_in_slice;
2652     unsigned int pps_id;
2653     int num_ref_idx_active_override_flag, ret;
2654     unsigned int slice_type, tmp, i, j;
2655     int default_ref_list_done = 0;
2656     int last_pic_structure, last_pic_droppable;
2657     int must_reinit;
2658     int needs_reinit = 0;
2659
2660     /* FIXME: 2tap qpel isn't implemented for high bit depth. */
2661     if ((s->avctx->flags2 & CODEC_FLAG2_FAST) &&
2662         !h->nal_ref_idc && !h->pixel_shift) {
2663         s->me.qpel_put = s->dsp.put_2tap_qpel_pixels_tab;
2664         s->me.qpel_avg = s->dsp.avg_2tap_qpel_pixels_tab;
2665     } else {
2666         s->me.qpel_put = s->dsp.put_h264_qpel_pixels_tab;
2667         s->me.qpel_avg = s->dsp.avg_h264_qpel_pixels_tab;
2668     }
2669
2670     first_mb_in_slice = get_ue_golomb_long(&s->gb);
2671
2672     if (first_mb_in_slice == 0) { // FIXME better field boundary detection
2673         if (h0->current_slice && FIELD_PICTURE) {
2674             field_end(h, 1);
2675         }
2676
2677         h0->current_slice = 0;
2678         if (!s0->first_field) {
2679             if (s->current_picture_ptr && !s->droppable &&
2680                 s->current_picture_ptr->owner2 == s) {
2681                 ff_thread_report_progress(&s->current_picture_ptr->f, INT_MAX,
2682                                           s->picture_structure == PICT_BOTTOM_FIELD);
2683             }
2684             s->current_picture_ptr = NULL;
2685         }
2686     }
2687
2688     slice_type = get_ue_golomb_31(&s->gb);
2689     if (slice_type > 9) {
2690         av_log(h->s.avctx, AV_LOG_ERROR,
2691                "slice type too large (%d) at %d %d\n",
2692                slice_type, s->mb_x, s->mb_y);
2693         return -1;
2694     }
2695     if (slice_type > 4) {
2696         slice_type -= 5;
2697         h->slice_type_fixed = 1;
2698     } else
2699         h->slice_type_fixed = 0;
2700
2701     slice_type = golomb_to_pict_type[slice_type];
2702     if (slice_type == AV_PICTURE_TYPE_I ||
2703         (h0->current_slice != 0 && slice_type == h0->last_slice_type)) {
2704         default_ref_list_done = 1;
2705     }
2706     h->slice_type     = slice_type;
2707     h->slice_type_nos = slice_type & 3;
2708
2709     // to make a few old functions happy, it's wrong though
2710     s->pict_type = h->slice_type;
2711
2712     pps_id = get_ue_golomb(&s->gb);
2713     if (pps_id >= MAX_PPS_COUNT) {
2714         av_log(h->s.avctx, AV_LOG_ERROR, "pps_id %d out of range\n", pps_id);
2715         return -1;
2716     }
2717     if (!h0->pps_buffers[pps_id]) {
2718         av_log(h->s.avctx, AV_LOG_ERROR,
2719                "non-existing PPS %u referenced\n",
2720                pps_id);
2721         return -1;
2722     }
2723     h->pps = *h0->pps_buffers[pps_id];
2724
2725     if (!h0->sps_buffers[h->pps.sps_id]) {
2726         av_log(h->s.avctx, AV_LOG_ERROR,
2727                "non-existing SPS %u referenced\n",
2728                h->pps.sps_id);
2729         return -1;
2730     }
2731
2732     if (h->pps.sps_id != h->current_sps_id ||
2733         h->context_reinitialized           ||
2734         h0->sps_buffers[h->pps.sps_id]->new) {
2735         SPS *new_sps = h0->sps_buffers[h->pps.sps_id];
2736
2737         h0->sps_buffers[h->pps.sps_id]->new = 0;
2738
2739         if (h->sps.chroma_format_idc != new_sps->chroma_format_idc ||
2740             h->sps.bit_depth_luma    != new_sps->bit_depth_luma)
2741             needs_reinit = 1;
2742
2743         h->current_sps_id = h->pps.sps_id;
2744         h->sps            = *h0->sps_buffers[h->pps.sps_id];
2745
2746         if (s->mb_width  != h->sps.mb_width ||
2747             s->mb_height != h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag) ||
2748             s->avctx->bits_per_raw_sample != h->sps.bit_depth_luma ||
2749             h->cur_chroma_format_idc != h->sps.chroma_format_idc
2750         )
2751             needs_reinit = 1;
2752
2753         if ((ret = h264_set_parameter_from_sps(h)) < 0)
2754             return ret;
2755     }
2756
2757     s->avctx->profile = ff_h264_get_profile(&h->sps);
2758     s->avctx->level   = h->sps.level_idc;
2759     s->avctx->refs    = h->sps.ref_frame_count;
2760
2761     must_reinit = (s->context_initialized &&
2762                     (   16*h->sps.mb_width != s->avctx->coded_width
2763                      || 16*h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag) != s->avctx->coded_height
2764                      || s->avctx->bits_per_raw_sample != h->sps.bit_depth_luma
2765                      || h->cur_chroma_format_idc != h->sps.chroma_format_idc
2766                      || av_cmp_q(h->sps.sar, s->avctx->sample_aspect_ratio)));
2767
2768
2769     s->mb_width  = h->sps.mb_width;
2770     s->mb_height = h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
2771
2772     h->b_stride = s->mb_width * 4;
2773
2774     s->chroma_y_shift = h->sps.chroma_format_idc <= 1; // 400 uses yuv420p
2775
2776     s->width  = 16 * s->mb_width;
2777     s->height = 16 * s->mb_height;
2778
2779     if (h->sps.video_signal_type_present_flag) {
2780         s->avctx->color_range = h->sps.full_range>0 ? AVCOL_RANGE_JPEG
2781                                                     : AVCOL_RANGE_MPEG;
2782         if (h->sps.colour_description_present_flag) {
2783             if (s->avctx->colorspace != h->sps.colorspace)
2784                 needs_reinit = 1;
2785             s->avctx->color_primaries = h->sps.color_primaries;
2786             s->avctx->color_trc       = h->sps.color_trc;
2787             s->avctx->colorspace      = h->sps.colorspace;
2788         }
2789     }
2790
2791     if (s->context_initialized &&
2792         (
2793          needs_reinit                   ||
2794          must_reinit)) {
2795
2796         if (h != h0) {
2797             av_log(s->avctx, AV_LOG_ERROR, "changing width/height on "
2798                    "slice %d\n", h0->current_slice + 1);
2799             return AVERROR_INVALIDDATA;
2800         }
2801
2802         flush_change(h);
2803
2804         if ((ret = get_pixel_format(h)) < 0)
2805             return ret;
2806         s->avctx->pix_fmt = ret;
2807
2808         av_log(h->s.avctx, AV_LOG_INFO, "Reinit context to %dx%d, "
2809                "pix_fmt: %d\n", s->width, s->height, s->avctx->pix_fmt);
2810
2811         if ((ret = h264_slice_header_init(h, 1)) < 0) {
2812             av_log(h->s.avctx, AV_LOG_ERROR,
2813                    "h264_slice_header_init() failed\n");
2814             return ret;
2815         }
2816         h->context_reinitialized = 1;
2817     }
2818     if (!s->context_initialized) {
2819         if (h != h0) {
2820             av_log(h->s.avctx, AV_LOG_ERROR,
2821                    "Cannot (re-)initialize context during parallel decoding.\n");
2822             return -1;
2823         }
2824
2825         if ((ret = get_pixel_format(h)) < 0)
2826             return ret;
2827         s->avctx->pix_fmt = ret;
2828
2829         if ((ret = h264_slice_header_init(h, 0)) < 0) {
2830             av_log(h->s.avctx, AV_LOG_ERROR,
2831                    "h264_slice_header_init() failed\n");
2832             return ret;
2833         }
2834     }
2835
2836     if (h == h0 && h->dequant_coeff_pps != pps_id) {
2837         h->dequant_coeff_pps = pps_id;
2838         init_dequant_tables(h);
2839     }
2840
2841     h->frame_num = get_bits(&s->gb, h->sps.log2_max_frame_num);
2842
2843     h->mb_mbaff        = 0;
2844     h->mb_aff_frame    = 0;
2845     last_pic_structure = s0->picture_structure;
2846     last_pic_droppable = s0->droppable;
2847     s->droppable       = h->nal_ref_idc == 0;
2848     if (h->sps.frame_mbs_only_flag) {
2849         s->picture_structure = PICT_FRAME;
2850     } else {
2851         if (!h->sps.direct_8x8_inference_flag && slice_type == AV_PICTURE_TYPE_B) {
2852             av_log(h->s.avctx, AV_LOG_ERROR, "This stream was generated by a broken encoder, invalid 8x8 inference\n");
2853             return -1;
2854         }
2855         if (get_bits1(&s->gb)) { // field_pic_flag
2856             s->picture_structure = PICT_TOP_FIELD + get_bits1(&s->gb); // bottom_field_flag
2857         } else {
2858             s->picture_structure = PICT_FRAME;
2859             h->mb_aff_frame      = h->sps.mb_aff;
2860         }
2861     }
2862     h->mb_field_decoding_flag = s->picture_structure != PICT_FRAME;
2863
2864     if (h0->current_slice != 0) {
2865         if (last_pic_structure != s->picture_structure ||
2866             last_pic_droppable != s->droppable) {
2867             av_log(h->s.avctx, AV_LOG_ERROR,
2868                    "Changing field mode (%d -> %d) between slices is not allowed\n",
2869                    last_pic_structure, s->picture_structure);
2870             s->picture_structure = last_pic_structure;
2871             s->droppable         = last_pic_droppable;
2872             return AVERROR_INVALIDDATA;
2873         } else if (!s0->current_picture_ptr) {
2874             av_log(s->avctx, AV_LOG_ERROR,
2875                    "unset current_picture_ptr on %d. slice\n",
2876                    h0->current_slice + 1);
2877             return AVERROR_INVALIDDATA;
2878         }
2879     } else {
2880         /* Shorten frame num gaps so we don't have to allocate reference
2881          * frames just to throw them away */
2882         if (h->frame_num != h->prev_frame_num && h->prev_frame_num >= 0) {
2883             int unwrap_prev_frame_num = h->prev_frame_num;
2884             int max_frame_num         = 1 << h->sps.log2_max_frame_num;
2885
2886             if (unwrap_prev_frame_num > h->frame_num)
2887                 unwrap_prev_frame_num -= max_frame_num;
2888
2889             if ((h->frame_num - unwrap_prev_frame_num) > h->sps.ref_frame_count) {
2890                 unwrap_prev_frame_num = (h->frame_num - h->sps.ref_frame_count) - 1;
2891                 if (unwrap_prev_frame_num < 0)
2892                     unwrap_prev_frame_num += max_frame_num;
2893
2894                 h->prev_frame_num = unwrap_prev_frame_num;
2895             }
2896         }
2897
2898         /* See if we have a decoded first field looking for a pair...
2899          * Here, we're using that to see if we should mark previously
2900          * decode frames as "finished".
2901          * We have to do that before the "dummy" in-between frame allocation,
2902          * since that can modify s->current_picture_ptr. */
2903         if (s0->first_field) {
2904             assert(s0->current_picture_ptr);
2905             assert(s0->current_picture_ptr->f.data[0]);
2906             assert(s0->current_picture_ptr->f.reference != DELAYED_PIC_REF);
2907
2908             /* Mark old field/frame as completed */
2909             if (!last_pic_droppable && s0->current_picture_ptr->owner2 == s0) {
2910                 ff_thread_report_progress(&s0->current_picture_ptr->f, INT_MAX,
2911                                           last_pic_structure == PICT_BOTTOM_FIELD);
2912             }
2913
2914             /* figure out if we have a complementary field pair */
2915             if (!FIELD_PICTURE || s->picture_structure == last_pic_structure) {
2916                 /* Previous field is unmatched. Don't display it, but let it
2917                  * remain for reference if marked as such. */
2918                 if (!last_pic_droppable && last_pic_structure != PICT_FRAME) {
2919                     ff_thread_report_progress(&s0->current_picture_ptr->f, INT_MAX,
2920                                               last_pic_structure == PICT_TOP_FIELD);
2921                 }
2922             } else {
2923                 if (s0->current_picture_ptr->frame_num != h->frame_num) {
2924                     /* This and previous field were reference, but had
2925                      * different frame_nums. Consider this field first in
2926                      * pair. Throw away previous field except for reference
2927                      * purposes. */
2928                     if (!last_pic_droppable && last_pic_structure != PICT_FRAME) {
2929                         ff_thread_report_progress(&s0->current_picture_ptr->f, INT_MAX,
2930                                                   last_pic_structure == PICT_TOP_FIELD);
2931                     }
2932                 } else {
2933                     /* Second field in complementary pair */
2934                     if (!((last_pic_structure   == PICT_TOP_FIELD &&
2935                            s->picture_structure == PICT_BOTTOM_FIELD) ||
2936                           (last_pic_structure   == PICT_BOTTOM_FIELD &&
2937                            s->picture_structure == PICT_TOP_FIELD))) {
2938                         av_log(s->avctx, AV_LOG_ERROR,
2939                                "Invalid field mode combination %d/%d\n",
2940                                last_pic_structure, s->picture_structure);
2941                         s->picture_structure = last_pic_structure;
2942                         s->droppable         = last_pic_droppable;
2943                         return AVERROR_INVALIDDATA;
2944                     } else if (last_pic_droppable != s->droppable) {
2945                         av_log(s->avctx, AV_LOG_ERROR,
2946                                "Cannot combine reference and non-reference fields in the same frame\n");
2947                         av_log_ask_for_sample(s->avctx, NULL);
2948                         s->picture_structure = last_pic_structure;
2949                         s->droppable         = last_pic_droppable;
2950                         return AVERROR_PATCHWELCOME;
2951                     }
2952
2953                     /* Take ownership of this buffer. Note that if another thread owned
2954                      * the first field of this buffer, we're not operating on that pointer,
2955                      * so the original thread is still responsible for reporting progress
2956                      * on that first field (or if that was us, we just did that above).
2957                      * By taking ownership, we assign responsibility to ourselves to
2958                      * report progress on the second field. */
2959                     s0->current_picture_ptr->owner2 = s0;
2960                 }
2961             }
2962         }
2963
2964         while (h->frame_num != h->prev_frame_num && h->prev_frame_num >= 0 && !s0->first_field &&
2965                h->frame_num != (h->prev_frame_num + 1) % (1 << h->sps.log2_max_frame_num)) {
2966             Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
2967             av_log(h->s.avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n",
2968                    h->frame_num, h->prev_frame_num);
2969             if (ff_h264_frame_start(h) < 0)
2970                 return -1;
2971             h->prev_frame_num++;
2972             h->prev_frame_num %= 1 << h->sps.log2_max_frame_num;
2973             s->current_picture_ptr->frame_num = h->prev_frame_num;
2974             ff_thread_report_progress(&s->current_picture_ptr->f, INT_MAX, 0);
2975             ff_thread_report_progress(&s->current_picture_ptr->f, INT_MAX, 1);
2976             ff_generate_sliding_window_mmcos(h);
2977             if (ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index) < 0 &&
2978                 (s->avctx->err_recognition & AV_EF_EXPLODE))
2979                 return AVERROR_INVALIDDATA;
2980             /* Error concealment: if a ref is missing, copy the previous ref in its place.
2981              * FIXME: avoiding a memcpy would be nice, but ref handling makes many assumptions
2982              * about there being no actual duplicates.
2983              * FIXME: this doesn't copy padding for out-of-frame motion vectors.  Given we're
2984              * concealing a lost frame, this probably isn't noticeable by comparison, but it should
2985              * be fixed. */
2986             if (h->short_ref_count) {
2987                 if (prev) {
2988                     av_image_copy(h->short_ref[0]->f.data, h->short_ref[0]->f.linesize,
2989                                   (const uint8_t **)prev->f.data, prev->f.linesize,
2990                                   s->avctx->pix_fmt, s->mb_width * 16, s->mb_height * 16);
2991                     h->short_ref[0]->poc = prev->poc + 2;
2992                 }
2993                 h->short_ref[0]->frame_num = h->prev_frame_num;
2994             }
2995         }
2996
2997         /* See if we have a decoded first field looking for a pair...
2998          * We're using that to see whether to continue decoding in that
2999          * frame, or to allocate a new one. */
3000         if (s0->first_field) {
3001             assert(s0->current_picture_ptr);
3002             assert(s0->current_picture_ptr->f.data[0]);
3003             assert(s0->current_picture_ptr->f.reference != DELAYED_PIC_REF);
3004
3005             /* figure out if we have a complementary field pair */
3006             if (!FIELD_PICTURE || s->picture_structure == last_pic_structure) {
3007                 /* Previous field is unmatched. Don't display it, but let it
3008                  * remain for reference if marked as such. */
3009                 s0->current_picture_ptr = NULL;
3010                 s0->first_field         = FIELD_PICTURE;
3011             } else {
3012                 if (s0->current_picture_ptr->frame_num != h->frame_num) {
3013                     ff_thread_report_progress((AVFrame*)s0->current_picture_ptr, INT_MAX,
3014                                               s0->picture_structure==PICT_BOTTOM_FIELD);
3015                     /* This and the previous field had different frame_nums.
3016                      * Consider this field first in pair. Throw away previous
3017                      * one except for reference purposes. */
3018                     s0->first_field         = 1;
3019                     s0->current_picture_ptr = NULL;
3020                 } else {
3021                     /* Second field in complementary pair */
3022                     s0->first_field = 0;
3023                 }
3024             }
3025         } else {
3026             /* Frame or first field in a potentially complementary pair */
3027             s0->first_field = FIELD_PICTURE;
3028         }
3029
3030         if (!FIELD_PICTURE || s0->first_field) {
3031             if (ff_h264_frame_start(h) < 0) {
3032                 s0->first_field = 0;
3033                 return -1;
3034             }
3035         } else {
3036             ff_release_unused_pictures(s, 0);
3037         }
3038     }
3039     if (h != h0 && (ret = clone_slice(h, h0)) < 0)
3040         return ret;
3041
3042     s->current_picture_ptr->frame_num = h->frame_num; // FIXME frame_num cleanup
3043
3044     av_assert1(s->mb_num == s->mb_width * s->mb_height);
3045     if (first_mb_in_slice << FIELD_OR_MBAFF_PICTURE >= s->mb_num ||
3046         first_mb_in_slice >= s->mb_num) {
3047         av_log(h->s.avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
3048         return -1;
3049     }
3050     s->resync_mb_x = s->mb_x =  first_mb_in_slice % s->mb_width;
3051     s->resync_mb_y = s->mb_y = (first_mb_in_slice / s->mb_width) << FIELD_OR_MBAFF_PICTURE;
3052     if (s->picture_structure == PICT_BOTTOM_FIELD)
3053         s->resync_mb_y = s->mb_y = s->mb_y + 1;
3054     av_assert1(s->mb_y < s->mb_height);
3055
3056     if (s->picture_structure == PICT_FRAME) {
3057         h->curr_pic_num = h->frame_num;
3058         h->max_pic_num  = 1 << h->sps.log2_max_frame_num;
3059     } else {
3060         h->curr_pic_num = 2 * h->frame_num + 1;
3061         h->max_pic_num  = 1 << (h->sps.log2_max_frame_num + 1);
3062     }
3063
3064     if (h->nal_unit_type == NAL_IDR_SLICE)
3065         get_ue_golomb(&s->gb); /* idr_pic_id */
3066
3067     if (h->sps.poc_type == 0) {
3068         h->poc_lsb = get_bits(&s->gb, h->sps.log2_max_poc_lsb);
3069
3070         if (h->pps.pic_order_present == 1 && s->picture_structure == PICT_FRAME)
3071             h->delta_poc_bottom = get_se_golomb(&s->gb);
3072     }
3073
3074     if (h->sps.poc_type == 1 && !h->sps.delta_pic_order_always_zero_flag) {
3075         h->delta_poc[0] = get_se_golomb(&s->gb);
3076
3077         if (h->pps.pic_order_present == 1 && s->picture_structure == PICT_FRAME)
3078             h->delta_poc[1] = get_se_golomb(&s->gb);
3079     }
3080
3081     init_poc(h);
3082
3083     if (h->pps.redundant_pic_cnt_present)
3084         h->redundant_pic_count = get_ue_golomb(&s->gb);
3085
3086     // set defaults, might be overridden a few lines later
3087     h->ref_count[0] = h->pps.ref_count[0];
3088     h->ref_count[1] = h->pps.ref_count[1];
3089
3090     if (h->slice_type_nos != AV_PICTURE_TYPE_I) {
3091         unsigned max[2];
3092         max[0] = max[1] = s->picture_structure == PICT_FRAME ? 15 : 31;
3093
3094         if (h->slice_type_nos == AV_PICTURE_TYPE_B)
3095             h->direct_spatial_mv_pred = get_bits1(&s->gb);
3096         num_ref_idx_active_override_flag = get_bits1(&s->gb);
3097
3098         if (num_ref_idx_active_override_flag) {
3099             h->ref_count[0] = get_ue_golomb(&s->gb) + 1;
3100             if (h->slice_type_nos == AV_PICTURE_TYPE_B) {
3101                 h->ref_count[1] = get_ue_golomb(&s->gb) + 1;
3102             } else
3103                 // full range is spec-ok in this case, even for frames
3104                 h->ref_count[1] = 1;
3105         }
3106
3107         if (h->ref_count[0]-1 > max[0] || h->ref_count[1]-1 > max[1]){
3108             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]);
3109             h->ref_count[0] = h->ref_count[1] = 1;
3110             return AVERROR_INVALIDDATA;
3111         }
3112
3113         if (h->slice_type_nos == AV_PICTURE_TYPE_B)
3114             h->list_count = 2;
3115         else
3116             h->list_count = 1;
3117     } else
3118         h->ref_count[1]= h->ref_count[0]= h->list_count= 0;
3119
3120     if (!default_ref_list_done)
3121         ff_h264_fill_default_ref_list(h);
3122
3123     if (h->slice_type_nos != AV_PICTURE_TYPE_I &&
3124         ff_h264_decode_ref_pic_list_reordering(h) < 0) {
3125         h->ref_count[1] = h->ref_count[0] = 0;
3126         return -1;
3127     }
3128
3129     if (h->slice_type_nos != AV_PICTURE_TYPE_I) {
3130         s->last_picture_ptr = &h->ref_list[0][0];
3131         s->last_picture_ptr->owner2 = s;
3132         ff_copy_picture(&s->last_picture, s->last_picture_ptr);
3133     }
3134     if (h->slice_type_nos == AV_PICTURE_TYPE_B) {
3135         s->next_picture_ptr = &h->ref_list[1][0];
3136         s->next_picture_ptr->owner2 = s;
3137         ff_copy_picture(&s->next_picture, s->next_picture_ptr);
3138     }
3139
3140     if ((h->pps.weighted_pred && h->slice_type_nos == AV_PICTURE_TYPE_P) ||
3141         (h->pps.weighted_bipred_idc == 1 &&
3142          h->slice_type_nos == AV_PICTURE_TYPE_B))
3143         pred_weight_table(h);
3144     else if (h->pps.weighted_bipred_idc == 2 &&
3145              h->slice_type_nos == AV_PICTURE_TYPE_B) {
3146         implicit_weight_table(h, -1);
3147     } else {
3148         h->use_weight = 0;
3149         for (i = 0; i < 2; i++) {
3150             h->luma_weight_flag[i]   = 0;
3151             h->chroma_weight_flag[i] = 0;
3152         }
3153     }
3154
3155     if (h->nal_ref_idc && ff_h264_decode_ref_pic_marking(h0, &s->gb) < 0 &&
3156         (s->avctx->err_recognition & AV_EF_EXPLODE))
3157         return AVERROR_INVALIDDATA;
3158
3159     if (FRAME_MBAFF) {
3160         ff_h264_fill_mbaff_ref_list(h);
3161
3162         if (h->pps.weighted_bipred_idc == 2 && h->slice_type_nos == AV_PICTURE_TYPE_B) {
3163             implicit_weight_table(h, 0);
3164             implicit_weight_table(h, 1);
3165         }
3166     }
3167
3168     if (h->slice_type_nos == AV_PICTURE_TYPE_B && !h->direct_spatial_mv_pred)
3169         ff_h264_direct_dist_scale_factor(h);
3170     ff_h264_direct_ref_list_init(h);
3171
3172     if (h->slice_type_nos != AV_PICTURE_TYPE_I && h->pps.cabac) {
3173         tmp = get_ue_golomb_31(&s->gb);
3174         if (tmp > 2) {
3175             av_log(s->avctx, AV_LOG_ERROR, "cabac_init_idc overflow\n");
3176             return -1;
3177         }
3178         h->cabac_init_idc = tmp;
3179     }
3180
3181     h->last_qscale_diff = 0;
3182     tmp = h->pps.init_qp + get_se_golomb(&s->gb);
3183     if (tmp > 51 + 6 * (h->sps.bit_depth_luma - 8)) {
3184         av_log(s->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
3185         return -1;
3186     }
3187     s->qscale       = tmp;
3188     h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
3189     h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
3190     // FIXME qscale / qp ... stuff
3191     if (h->slice_type == AV_PICTURE_TYPE_SP)
3192         get_bits1(&s->gb); /* sp_for_switch_flag */
3193     if (h->slice_type == AV_PICTURE_TYPE_SP ||
3194         h->slice_type == AV_PICTURE_TYPE_SI)
3195         get_se_golomb(&s->gb); /* slice_qs_delta */
3196
3197     h->deblocking_filter     = 1;
3198     h->slice_alpha_c0_offset = 52;
3199     h->slice_beta_offset     = 52;
3200     if (h->pps.deblocking_filter_parameters_present) {
3201         tmp = get_ue_golomb_31(&s->gb);
3202         if (tmp > 2) {
3203             av_log(s->avctx, AV_LOG_ERROR,
3204                    "deblocking_filter_idc %u out of range\n", tmp);
3205             return -1;
3206         }
3207         h->deblocking_filter = tmp;
3208         if (h->deblocking_filter < 2)
3209             h->deblocking_filter ^= 1;  // 1<->0
3210
3211         if (h->deblocking_filter) {
3212             h->slice_alpha_c0_offset += get_se_golomb(&s->gb) << 1;
3213             h->slice_beta_offset     += get_se_golomb(&s->gb) << 1;
3214             if (h->slice_alpha_c0_offset > 104U ||
3215                 h->slice_beta_offset     > 104U) {
3216                 av_log(s->avctx, AV_LOG_ERROR,
3217                        "deblocking filter parameters %d %d out of range\n",
3218                        h->slice_alpha_c0_offset, h->slice_beta_offset);
3219                 return -1;
3220             }
3221         }
3222     }
3223
3224     if (s->avctx->skip_loop_filter >= AVDISCARD_ALL ||
3225         (s->avctx->skip_loop_filter >= AVDISCARD_NONKEY &&
3226          h->slice_type_nos != AV_PICTURE_TYPE_I) ||
3227         (s->avctx->skip_loop_filter >= AVDISCARD_BIDIR  &&
3228          h->slice_type_nos == AV_PICTURE_TYPE_B) ||
3229         (s->avctx->skip_loop_filter >= AVDISCARD_NONREF &&
3230          h->nal_ref_idc == 0))
3231         h->deblocking_filter = 0;
3232
3233     if (h->deblocking_filter == 1 && h0->max_contexts > 1) {
3234         if (s->avctx->flags2 & CODEC_FLAG2_FAST) {
3235             /* Cheat slightly for speed:
3236              * Do not bother to deblock across slices. */
3237             h->deblocking_filter = 2;
3238         } else {
3239             h0->max_contexts = 1;
3240             if (!h0->single_decode_warning) {
3241                 av_log(s->avctx, AV_LOG_INFO,
3242                        "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
3243                 h0->single_decode_warning = 1;
3244             }
3245             if (h != h0) {
3246                 av_log(h->s.avctx, AV_LOG_ERROR,
3247                        "Deblocking switched inside frame.\n");
3248                 return 1;
3249             }
3250         }
3251     }
3252     h->qp_thresh = 15 + 52 -
3253                    FFMIN(h->slice_alpha_c0_offset, h->slice_beta_offset) -
3254                    FFMAX3(0,
3255                           h->pps.chroma_qp_index_offset[0],
3256                           h->pps.chroma_qp_index_offset[1]) +
3257                    6 * (h->sps.bit_depth_luma - 8);
3258
3259     h0->last_slice_type = slice_type;
3260     h->slice_num = ++h0->current_slice;
3261
3262     if (h->slice_num)
3263         h0->slice_row[(h->slice_num-1)&(MAX_SLICES-1)]= s->resync_mb_y;
3264     if (   h0->slice_row[h->slice_num&(MAX_SLICES-1)] + 3 >= s->resync_mb_y
3265         && h0->slice_row[h->slice_num&(MAX_SLICES-1)] <= s->resync_mb_y
3266         && h->slice_num >= MAX_SLICES) {
3267         //in case of ASO this check needs to be updated depending on how we decide to assign slice numbers in this case
3268         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);
3269     }
3270
3271     for (j = 0; j < 2; j++) {
3272         int id_list[16];
3273         int *ref2frm = h->ref2frm[h->slice_num & (MAX_SLICES - 1)][j];
3274         for (i = 0; i < 16; i++) {
3275             id_list[i] = 60;
3276             if (h->ref_list[j][i].f.data[0]) {
3277                 int k;
3278                 uint8_t *base = h->ref_list[j][i].f.base[0];
3279                 for (k = 0; k < h->short_ref_count; k++)
3280                     if (h->short_ref[k]->f.base[0] == base) {
3281                         id_list[i] = k;
3282                         break;
3283                     }
3284                 for (k = 0; k < h->long_ref_count; k++)
3285                     if (h->long_ref[k] && h->long_ref[k]->f.base[0] == base) {
3286                         id_list[i] = h->short_ref_count + k;
3287                         break;
3288                     }
3289             }
3290         }
3291
3292         ref2frm[0]     =
3293             ref2frm[1] = -1;
3294         for (i = 0; i < 16; i++)
3295             ref2frm[i + 2] = 4 * id_list[i] +
3296                              (h->ref_list[j][i].f.reference & 3);
3297         ref2frm[18 + 0]     =
3298             ref2frm[18 + 1] = -1;
3299         for (i = 16; i < 48; i++)
3300             ref2frm[i + 4] = 4 * id_list[(i - 16) >> 1] +
3301                              (h->ref_list[j][i].f.reference & 3);
3302     }
3303
3304     // FIXME: fix draw_edges + PAFF + frame threads
3305     h->emu_edge_width  = (s->flags & CODEC_FLAG_EMU_EDGE ||
3306                           (!h->sps.frame_mbs_only_flag &&
3307                            s->avctx->active_thread_type))
3308                          ? 0 : 16;
3309     h->emu_edge_height = (FRAME_MBAFF || FIELD_PICTURE) ? 0 : h->emu_edge_width;
3310
3311     if (s->avctx->debug & FF_DEBUG_PICT_INFO) {
3312         av_log(h->s.avctx, AV_LOG_DEBUG,
3313                "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",
3314                h->slice_num,
3315                (s->picture_structure == PICT_FRAME ? "F" : s->picture_structure == PICT_TOP_FIELD ? "T" : "B"),
3316                first_mb_in_slice,
3317                av_get_picture_type_char(h->slice_type),
3318                h->slice_type_fixed ? " fix" : "",
3319                h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
3320                pps_id, h->frame_num,
3321                s->current_picture_ptr->field_poc[0],
3322                s->current_picture_ptr->field_poc[1],
3323                h->ref_count[0], h->ref_count[1],
3324                s->qscale,
3325                h->deblocking_filter,
3326                h->slice_alpha_c0_offset / 2 - 26, h->slice_beta_offset / 2 - 26,
3327                h->use_weight,
3328                h->use_weight == 1 && h->use_weight_chroma ? "c" : "",
3329                h->slice_type == AV_PICTURE_TYPE_B ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : "");
3330     }
3331
3332     return 0;
3333 }
3334
3335 int ff_h264_get_slice_type(const H264Context *h)
3336 {
3337     switch (h->slice_type) {
3338     case AV_PICTURE_TYPE_P:
3339         return 0;
3340     case AV_PICTURE_TYPE_B:
3341         return 1;
3342     case AV_PICTURE_TYPE_I:
3343         return 2;
3344     case AV_PICTURE_TYPE_SP:
3345         return 3;
3346     case AV_PICTURE_TYPE_SI:
3347         return 4;
3348     default:
3349         return -1;
3350     }
3351 }
3352
3353 static av_always_inline void fill_filter_caches_inter(H264Context *h,
3354                                                       MpegEncContext *const s,
3355                                                       int mb_type, int top_xy,
3356                                                       int left_xy[LEFT_MBS],
3357                                                       int top_type,
3358                                                       int left_type[LEFT_MBS],
3359                                                       int mb_xy, int list)
3360 {
3361     int b_stride = h->b_stride;
3362     int16_t(*mv_dst)[2] = &h->mv_cache[list][scan8[0]];
3363     int8_t *ref_cache = &h->ref_cache[list][scan8[0]];
3364     if (IS_INTER(mb_type) || IS_DIRECT(mb_type)) {
3365         if (USES_LIST(top_type, list)) {
3366             const int b_xy  = h->mb2b_xy[top_xy] + 3 * b_stride;
3367             const int b8_xy = 4 * top_xy + 2;
3368             int (*ref2frm)[64] = (void*)(h->ref2frm[h->slice_table[top_xy] & (MAX_SLICES - 1)][0] + (MB_MBAFF ? 20 : 2));
3369             AV_COPY128(mv_dst - 1 * 8, s->current_picture.f.motion_val[list][b_xy + 0]);
3370             ref_cache[0 - 1 * 8] =
3371             ref_cache[1 - 1 * 8] = ref2frm[list][s->current_picture.f.ref_index[list][b8_xy + 0]];
3372             ref_cache[2 - 1 * 8] =
3373             ref_cache[3 - 1 * 8] = ref2frm[list][s->current_picture.f.ref_index[list][b8_xy + 1]];
3374         } else {
3375             AV_ZERO128(mv_dst - 1 * 8);
3376             AV_WN32A(&ref_cache[0 - 1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
3377         }
3378
3379         if (!IS_INTERLACED(mb_type ^ left_type[LTOP])) {
3380             if (USES_LIST(left_type[LTOP], list)) {
3381                 const int b_xy  = h->mb2b_xy[left_xy[LTOP]] + 3;
3382                 const int b8_xy = 4 * left_xy[LTOP] + 1;
3383                 int (*ref2frm)[64] =(void*)( h->ref2frm[h->slice_table[left_xy[LTOP]] & (MAX_SLICES - 1)][0] + (MB_MBAFF ? 20 : 2));
3384                 AV_COPY32(mv_dst - 1 +  0, s->current_picture.f.motion_val[list][b_xy + b_stride * 0]);
3385                 AV_COPY32(mv_dst - 1 +  8, s->current_picture.f.motion_val[list][b_xy + b_stride * 1]);
3386                 AV_COPY32(mv_dst - 1 + 16, s->current_picture.f.motion_val[list][b_xy + b_stride * 2]);
3387                 AV_COPY32(mv_dst - 1 + 24, s->current_picture.f.motion_val[list][b_xy + b_stride * 3]);
3388                 ref_cache[-1 +  0] =
3389                 ref_cache[-1 +  8] = ref2frm[list][s->current_picture.f.ref_index[list][b8_xy + 2 * 0]];
3390                 ref_cache[-1 + 16] =
3391                 ref_cache[-1 + 24] = ref2frm[list][s->current_picture.f.ref_index[list][b8_xy + 2 * 1]];
3392             } else {
3393                 AV_ZERO32(mv_dst - 1 +  0);
3394                 AV_ZERO32(mv_dst - 1 +  8);
3395                 AV_ZERO32(mv_dst - 1 + 16);
3396                 AV_ZERO32(mv_dst - 1 + 24);
3397                 ref_cache[-1 +  0] =
3398                 ref_cache[-1 +  8] =
3399                 ref_cache[-1 + 16] =
3400                 ref_cache[-1 + 24] = LIST_NOT_USED;
3401             }
3402         }
3403     }
3404
3405     if (!USES_LIST(mb_type, list)) {
3406         fill_rectangle(mv_dst, 4, 4, 8, pack16to32(0, 0), 4);
3407         AV_WN32A(&ref_cache[0 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
3408         AV_WN32A(&ref_cache[1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
3409         AV_WN32A(&ref_cache[2 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
3410         AV_WN32A(&ref_cache[3 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
3411         return;
3412     }
3413
3414     {
3415         int8_t *ref = &s->current_picture.f.ref_index[list][4 * mb_xy];
3416         int (*ref2frm)[64] = (void*)(h->ref2frm[h->slice_num & (MAX_SLICES - 1)][0] + (MB_MBAFF ? 20 : 2));
3417         uint32_t ref01 = (pack16to32(ref2frm[list][ref[0]], ref2frm[list][ref[1]]) & 0x00FF00FF) * 0x0101;
3418         uint32_t ref23 = (pack16to32(ref2frm[list][ref[2]], ref2frm[list][ref[3]]) & 0x00FF00FF) * 0x0101;
3419         AV_WN32A(&ref_cache[0 * 8], ref01);
3420         AV_WN32A(&ref_cache[1 * 8], ref01);
3421         AV_WN32A(&ref_cache[2 * 8], ref23);
3422         AV_WN32A(&ref_cache[3 * 8], ref23);
3423     }
3424
3425     {
3426         int16_t(*mv_src)[2] = &s->current_picture.f.motion_val[list][4 * s->mb_x + 4 * s->mb_y * b_stride];
3427         AV_COPY128(mv_dst + 8 * 0, mv_src + 0 * b_stride);
3428         AV_COPY128(mv_dst + 8 * 1, mv_src + 1 * b_stride);
3429         AV_COPY128(mv_dst + 8 * 2, mv_src + 2 * b_stride);
3430         AV_COPY128(mv_dst + 8 * 3, mv_src + 3 * b_stride);
3431     }
3432 }
3433
3434 /**
3435  *
3436  * @return non zero if the loop filter can be skipped
3437  */
3438 static int fill_filter_caches(H264Context *h, int mb_type)
3439 {
3440     MpegEncContext *const s = &h->s;
3441     const int mb_xy = h->mb_xy;
3442     int top_xy, left_xy[LEFT_MBS];
3443     int top_type, left_type[LEFT_MBS];
3444     uint8_t *nnz;
3445     uint8_t *nnz_cache;
3446
3447     top_xy = mb_xy - (s->mb_stride << MB_FIELD);
3448
3449     /* Wow, what a mess, why didn't they simplify the interlacing & intra
3450      * stuff, I can't imagine that these complex rules are worth it. */
3451
3452     left_xy[LBOT] = left_xy[LTOP] = mb_xy - 1;
3453     if (FRAME_MBAFF) {
3454         const int left_mb_field_flag = IS_INTERLACED(s->current_picture.f.mb_type[mb_xy - 1]);
3455         const int curr_mb_field_flag = IS_INTERLACED(mb_type);
3456         if (s->mb_y & 1) {
3457             if (left_mb_field_flag != curr_mb_field_flag)
3458                 left_xy[LTOP] -= s->mb_stride;
3459         } else {
3460             if (curr_mb_field_flag)
3461                 top_xy += s->mb_stride &
3462                     (((s->current_picture.f.mb_type[top_xy] >> 7) & 1) - 1);
3463             if (left_mb_field_flag != curr_mb_field_flag)
3464                 left_xy[LBOT] += s->mb_stride;
3465         }
3466     }
3467
3468     h->top_mb_xy        = top_xy;
3469     h->left_mb_xy[LTOP] = left_xy[LTOP];
3470     h->left_mb_xy[LBOT] = left_xy[LBOT];
3471     {
3472         /* For sufficiently low qp, filtering wouldn't do anything.
3473          * This is a conservative estimate: could also check beta_offset
3474          * and more accurate chroma_qp. */
3475         int qp_thresh = h->qp_thresh; // FIXME strictly we should store qp_thresh for each mb of a slice
3476         int qp        = s->current_picture.f.qscale_table[mb_xy];
3477         if (qp <= qp_thresh &&
3478             (left_xy[LTOP] < 0 ||
3479              ((qp + s->current_picture.f.qscale_table[left_xy[LTOP]] + 1) >> 1) <= qp_thresh) &&
3480             (top_xy < 0 ||
3481              ((qp + s->current_picture.f.qscale_table[top_xy] + 1) >> 1) <= qp_thresh)) {
3482             if (!FRAME_MBAFF)
3483                 return 1;
3484             if ((left_xy[LTOP] < 0 ||
3485                  ((qp + s->current_picture.f.qscale_table[left_xy[LBOT]] + 1) >> 1) <= qp_thresh) &&
3486                 (top_xy < s->mb_stride ||
3487                  ((qp + s->current_picture.f.qscale_table[top_xy - s->mb_stride] + 1) >> 1) <= qp_thresh))
3488                 return 1;
3489         }
3490     }
3491
3492     top_type        = s->current_picture.f.mb_type[top_xy];
3493     left_type[LTOP] = s->current_picture.f.mb_type[left_xy[LTOP]];
3494     left_type[LBOT] = s->current_picture.f.mb_type[left_xy[LBOT]];
3495     if (h->deblocking_filter == 2) {
3496         if (h->slice_table[top_xy] != h->slice_num)
3497             top_type = 0;
3498         if (h->slice_table[left_xy[LBOT]] != h->slice_num)
3499             left_type[LTOP] = left_type[LBOT] = 0;
3500     } else {
3501         if (h->slice_table[top_xy] == 0xFFFF)
3502             top_type = 0;
3503         if (h->slice_table[left_xy[LBOT]] == 0xFFFF)
3504             left_type[LTOP] = left_type[LBOT] = 0;
3505     }
3506     h->top_type        = top_type;
3507     h->left_type[LTOP] = left_type[LTOP];
3508     h->left_type[LBOT] = left_type[LBOT];
3509
3510     if (IS_INTRA(mb_type))
3511         return 0;
3512
3513     fill_filter_caches_inter(h, s, mb_type, top_xy, left_xy,
3514                              top_type, left_type, mb_xy, 0);
3515     if (h->list_count == 2)
3516         fill_filter_caches_inter(h, s, mb_type, top_xy, left_xy,
3517                                  top_type, left_type, mb_xy, 1);
3518
3519     nnz       = h->non_zero_count[mb_xy];
3520     nnz_cache = h->non_zero_count_cache;
3521     AV_COPY32(&nnz_cache[4 + 8 * 1], &nnz[0]);
3522     AV_COPY32(&nnz_cache[4 + 8 * 2], &nnz[4]);
3523     AV_COPY32(&nnz_cache[4 + 8 * 3], &nnz[8]);
3524     AV_COPY32(&nnz_cache[4 + 8 * 4], &nnz[12]);
3525     h->cbp = h->cbp_table[mb_xy];
3526
3527     if (top_type) {
3528         nnz = h->non_zero_count[top_xy];
3529         AV_COPY32(&nnz_cache[4 + 8 * 0], &nnz[3 * 4]);
3530     }
3531
3532     if (left_type[LTOP]) {
3533         nnz = h->non_zero_count[left_xy[LTOP]];
3534         nnz_cache[3 + 8 * 1] = nnz[3 + 0 * 4];
3535         nnz_cache[3 + 8 * 2] = nnz[3 + 1 * 4];
3536         nnz_cache[3 + 8 * 3] = nnz[3 + 2 * 4];
3537         nnz_cache[3 + 8 * 4] = nnz[3 + 3 * 4];
3538     }
3539
3540     /* CAVLC 8x8dct requires NNZ values for residual decoding that differ
3541      * from what the loop filter needs */
3542     if (!CABAC && h->pps.transform_8x8_mode) {
3543         if (IS_8x8DCT(top_type)) {
3544             nnz_cache[4 + 8 * 0]     =
3545                 nnz_cache[5 + 8 * 0] = (h->cbp_table[top_xy] & 0x4000) >> 12;
3546             nnz_cache[6 + 8 * 0]     =
3547                 nnz_cache[7 + 8 * 0] = (h->cbp_table[top_xy] & 0x8000) >> 12;
3548         }
3549         if (IS_8x8DCT(left_type[LTOP])) {
3550             nnz_cache[3 + 8 * 1]     =
3551                 nnz_cache[3 + 8 * 2] = (h->cbp_table[left_xy[LTOP]] & 0x2000) >> 12; // FIXME check MBAFF
3552         }
3553         if (IS_8x8DCT(left_type[LBOT])) {
3554             nnz_cache[3 + 8 * 3]     =
3555                 nnz_cache[3 + 8 * 4] = (h->cbp_table[left_xy[LBOT]] & 0x8000) >> 12; // FIXME check MBAFF
3556         }
3557
3558         if (IS_8x8DCT(mb_type)) {
3559             nnz_cache[scan8[0]] =
3560             nnz_cache[scan8[1]] =
3561             nnz_cache[scan8[2]] =
3562             nnz_cache[scan8[3]] = (h->cbp & 0x1000) >> 12;
3563
3564             nnz_cache[scan8[0 + 4]] =
3565             nnz_cache[scan8[1 + 4]] =
3566             nnz_cache[scan8[2 + 4]] =
3567             nnz_cache[scan8[3 + 4]] = (h->cbp & 0x2000) >> 12;
3568
3569             nnz_cache[scan8[0 + 8]] =
3570             nnz_cache[scan8[1 + 8]] =
3571             nnz_cache[scan8[2 + 8]] =
3572             nnz_cache[scan8[3 + 8]] = (h->cbp & 0x4000) >> 12;
3573
3574             nnz_cache[scan8[0 + 12]] =
3575             nnz_cache[scan8[1 + 12]] =
3576             nnz_cache[scan8[2 + 12]] =
3577             nnz_cache[scan8[3 + 12]] = (h->cbp & 0x8000) >> 12;
3578         }
3579     }
3580
3581     return 0;
3582 }
3583
3584 static void loop_filter(H264Context *h, int start_x, int end_x)
3585 {
3586     MpegEncContext *const s = &h->s;
3587     uint8_t *dest_y, *dest_cb, *dest_cr;
3588     int linesize, uvlinesize, mb_x, mb_y;
3589     const int end_mb_y       = s->mb_y + FRAME_MBAFF;
3590     const int old_slice_type = h->slice_type;
3591     const int pixel_shift    = h->pixel_shift;
3592     const int block_h        = 16 >> s->chroma_y_shift;
3593
3594     if (h->deblocking_filter) {
3595         for (mb_x = start_x; mb_x < end_x; mb_x++)
3596             for (mb_y = end_mb_y - FRAME_MBAFF; mb_y <= end_mb_y; mb_y++) {
3597                 int mb_xy, mb_type;
3598                 mb_xy         = h->mb_xy = mb_x + mb_y * s->mb_stride;
3599                 h->slice_num  = h->slice_table[mb_xy];
3600                 mb_type       = s->current_picture.f.mb_type[mb_xy];
3601                 h->list_count = h->list_counts[mb_xy];
3602
3603                 if (FRAME_MBAFF)
3604                     h->mb_mbaff               =
3605                     h->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
3606
3607                 s->mb_x = mb_x;
3608                 s->mb_y = mb_y;
3609                 dest_y  = s->current_picture.f.data[0] +
3610                           ((mb_x << pixel_shift) + mb_y * s->linesize) * 16;
3611                 dest_cb = s->current_picture.f.data[1] +
3612                           (mb_x << pixel_shift) * (8 << CHROMA444) +
3613                           mb_y * s->uvlinesize * block_h;
3614                 dest_cr = s->current_picture.f.data[2] +
3615                           (mb_x << pixel_shift) * (8 << CHROMA444) +
3616                           mb_y * s->uvlinesize * block_h;
3617                 // FIXME simplify above
3618
3619                 if (MB_FIELD) {
3620                     linesize   = h->mb_linesize   = s->linesize   * 2;
3621                     uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
3622                     if (mb_y & 1) { // FIXME move out of this function?
3623                         dest_y  -= s->linesize   * 15;
3624                         dest_cb -= s->uvlinesize * (block_h - 1);
3625                         dest_cr -= s->uvlinesize * (block_h - 1);
3626                     }
3627                 } else {
3628                     linesize   = h->mb_linesize   = s->linesize;
3629                     uvlinesize = h->mb_uvlinesize = s->uvlinesize;
3630                 }
3631                 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize,
3632                                  uvlinesize, 0);
3633                 if (fill_filter_caches(h, mb_type))
3634                     continue;
3635                 h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.f.qscale_table[mb_xy]);
3636                 h->chroma_qp[1] = get_chroma_qp(h, 1, s->current_picture.f.qscale_table[mb_xy]);
3637
3638                 if (FRAME_MBAFF) {
3639                     ff_h264_filter_mb(h, mb_x, mb_y, dest_y, dest_cb, dest_cr,
3640                                       linesize, uvlinesize);
3641                 } else {
3642                     ff_h264_filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb,
3643                                            dest_cr, linesize, uvlinesize);
3644                 }
3645             }
3646     }
3647     h->slice_type   = old_slice_type;
3648     s->mb_x         = end_x;
3649     s->mb_y         = end_mb_y - FRAME_MBAFF;
3650     h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
3651     h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
3652 }
3653
3654 static void predict_field_decoding_flag(H264Context *h)
3655 {
3656     MpegEncContext *const s = &h->s;
3657     const int mb_xy = s->mb_x + s->mb_y * s->mb_stride;
3658     int mb_type     = (h->slice_table[mb_xy - 1] == h->slice_num) ?
3659                       s->current_picture.f.mb_type[mb_xy - 1] :
3660                       (h->slice_table[mb_xy - s->mb_stride] == h->slice_num) ?
3661                       s->current_picture.f.mb_type[mb_xy - s->mb_stride] : 0;
3662     h->mb_mbaff     = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
3663 }
3664
3665 /**
3666  * Draw edges and report progress for the last MB row.
3667  */
3668 static void decode_finish_row(H264Context *h)
3669 {
3670     MpegEncContext *const s = &h->s;
3671     int top            = 16 * (s->mb_y      >> FIELD_PICTURE);
3672     int pic_height     = 16 *  s->mb_height >> FIELD_PICTURE;
3673     int height         =  16      << FRAME_MBAFF;
3674     int deblock_border = (16 + 4) << FRAME_MBAFF;
3675
3676     if (h->deblocking_filter) {
3677         if ((top + height) >= pic_height)
3678             height += deblock_border;
3679         top -= deblock_border;
3680     }
3681
3682     if (top >= pic_height || (top + height) < h->emu_edge_height)
3683         return;
3684
3685     height = FFMIN(height, pic_height - top);
3686     if (top < h->emu_edge_height) {
3687         height = top + height;
3688         top    = 0;
3689     }
3690
3691     ff_draw_horiz_band(s, top, height);
3692
3693     if (s->droppable)
3694         return;
3695
3696     ff_thread_report_progress(&s->current_picture_ptr->f, top + height - 1,
3697                               s->picture_structure == PICT_BOTTOM_FIELD);
3698 }
3699
3700 static int decode_slice(struct AVCodecContext *avctx, void *arg)
3701 {
3702     H264Context *h = *(void **)arg;
3703     MpegEncContext *const s = &h->s;
3704     const int part_mask     = s->partitioned_frame ? (ER_AC_END | ER_AC_ERROR)
3705                                                    : 0x7F;
3706     int lf_x_start = s->mb_x;
3707
3708     s->mb_skip_run = -1;
3709
3710     h->is_complex = FRAME_MBAFF || s->picture_structure != PICT_FRAME ||
3711                     s->codec_id != AV_CODEC_ID_H264 ||
3712                     (CONFIG_GRAY && (s->flags & CODEC_FLAG_GRAY));
3713
3714     if (h->pps.cabac) {
3715         /* realign */
3716         align_get_bits(&s->gb);
3717
3718         /* init cabac */
3719         ff_init_cabac_decoder(&h->cabac,
3720                               s->gb.buffer + get_bits_count(&s->gb) / 8,
3721                               (get_bits_left(&s->gb) + 7) / 8);
3722
3723         ff_h264_init_cabac_states(h);
3724
3725         for (;;) {
3726             // START_TIMER
3727             int ret = ff_h264_decode_mb_cabac(h);
3728             int eos;
3729             // STOP_TIMER("decode_mb_cabac")
3730
3731             if (ret >= 0)
3732                 ff_h264_hl_decode_mb(h);
3733
3734             // FIXME optimal? or let mb_decode decode 16x32 ?
3735             if (ret >= 0 && FRAME_MBAFF) {
3736                 s->mb_y++;
3737
3738                 ret = ff_h264_decode_mb_cabac(h);
3739
3740                 if (ret >= 0)
3741                     ff_h264_hl_decode_mb(h);
3742                 s->mb_y--;
3743             }
3744             eos = get_cabac_terminate(&h->cabac);
3745
3746             if ((s->workaround_bugs & FF_BUG_TRUNCATED) &&
3747                 h->cabac.bytestream > h->cabac.bytestream_end + 2) {
3748                 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x - 1,
3749                                 s->mb_y, ER_MB_END & part_mask);
3750                 if (s->mb_x >= lf_x_start)
3751                     loop_filter(h, lf_x_start, s->mb_x + 1);
3752                 return 0;
3753             }
3754             if (h->cabac.bytestream > h->cabac.bytestream_end + 2 )
3755                 av_log(h->s.avctx, AV_LOG_DEBUG, "bytestream overread %td\n", h->cabac.bytestream_end - h->cabac.bytestream);
3756             if (ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 4) {
3757                 av_log(h->s.avctx, AV_LOG_ERROR,
3758                        "error while decoding MB %d %d, bytestream (%td)\n",
3759                        s->mb_x, s->mb_y,
3760                        h->cabac.bytestream_end - h->cabac.bytestream);
3761                 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x,
3762                                 s->mb_y, ER_MB_ERROR & part_mask);
3763                 return -1;
3764             }
3765
3766             if (++s->mb_x >= s->mb_width) {
3767                 loop_filter(h, lf_x_start, s->mb_x);
3768                 s->mb_x = lf_x_start = 0;
3769                 decode_finish_row(h);
3770                 ++s->mb_y;
3771                 if (FIELD_OR_MBAFF_PICTURE) {
3772                     ++s->mb_y;
3773                     if (FRAME_MBAFF && s->mb_y < s->mb_height)
3774                         predict_field_decoding_flag(h);
3775                 }
3776             }
3777
3778             if (eos || s->mb_y >= s->mb_height) {
3779                 tprintf(s->avctx, "slice end %d %d\n",
3780                         get_bits_count(&s->gb), s->gb.size_in_bits);
3781                 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x - 1,
3782                                 s->mb_y, ER_MB_END & part_mask);
3783                 if (s->mb_x > lf_x_start)
3784                     loop_filter(h, lf_x_start, s->mb_x);
3785                 return 0;
3786             }
3787         }
3788     } else {
3789         for (;;) {
3790             int ret = ff_h264_decode_mb_cavlc(h);
3791
3792             if (ret >= 0)
3793                 ff_h264_hl_decode_mb(h);
3794
3795             // FIXME optimal? or let mb_decode decode 16x32 ?
3796             if (ret >= 0 && FRAME_MBAFF) {
3797                 s->mb_y++;
3798                 ret = ff_h264_decode_mb_cavlc(h);
3799
3800                 if (ret >= 0)
3801                     ff_h264_hl_decode_mb(h);
3802                 s->mb_y--;
3803             }
3804
3805             if (ret < 0) {
3806                 av_log(h->s.avctx, AV_LOG_ERROR,
3807                        "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
3808                 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x,
3809                                 s->mb_y, ER_MB_ERROR & part_mask);
3810                 return -1;
3811             }
3812
3813             if (++s->mb_x >= s->mb_width) {
3814                 loop_filter(h, lf_x_start, s->mb_x);
3815                 s->mb_x = lf_x_start = 0;
3816                 decode_finish_row(h);
3817                 ++s->mb_y;
3818                 if (FIELD_OR_MBAFF_PICTURE) {
3819                     ++s->mb_y;
3820                     if (FRAME_MBAFF && s->mb_y < s->mb_height)
3821                         predict_field_decoding_flag(h);
3822                 }
3823                 if (s->mb_y >= s->mb_height) {
3824                     tprintf(s->avctx, "slice end %d %d\n",
3825                             get_bits_count(&s->gb), s->gb.size_in_bits);
3826
3827                     if (   get_bits_left(&s->gb) == 0
3828                         || get_bits_left(&s->gb) > 0 && !(s->avctx->err_recognition & AV_EF_AGGRESSIVE)) {
3829                         ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y,
3830                                         s->mb_x - 1, s->mb_y,
3831                                         ER_MB_END & part_mask);
3832
3833                         return 0;
3834                     } else {
3835                         ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y,
3836                                         s->mb_x, s->mb_y,
3837                                         ER_MB_END & part_mask);
3838
3839                         return -1;
3840                     }
3841                 }
3842             }
3843
3844             if (get_bits_left(&s->gb) <= 0 && s->mb_skip_run <= 0) {
3845                 tprintf(s->avctx, "slice end %d %d\n",
3846                         get_bits_count(&s->gb), s->gb.size_in_bits);
3847                 if (get_bits_left(&s->gb) == 0) {
3848                     ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y,
3849                                     s->mb_x - 1, s->mb_y,
3850                                     ER_MB_END & part_mask);
3851                     if (s->mb_x > lf_x_start)
3852                         loop_filter(h, lf_x_start, s->mb_x);
3853
3854                     return 0;
3855                 } else {
3856                     ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x,
3857                                     s->mb_y, ER_MB_ERROR & part_mask);
3858
3859                     return -1;
3860                 }
3861             }
3862         }
3863     }
3864 }
3865
3866 /**
3867  * Call decode_slice() for each context.
3868  *
3869  * @param h h264 master context
3870  * @param context_count number of contexts to execute
3871  */
3872 static int execute_decode_slices(H264Context *h, int context_count)
3873 {
3874     MpegEncContext *const s     = &h->s;
3875     AVCodecContext *const avctx = s->avctx;
3876     H264Context *hx;
3877     int i;
3878
3879     if (s->avctx->hwaccel ||
3880         s->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU)
3881         return 0;
3882     if (context_count == 1) {
3883         return decode_slice(avctx, &h);
3884     } else {
3885         for (i = 1; i < context_count; i++) {
3886             hx                    = h->thread_context[i];
3887             hx->s.err_recognition = avctx->err_recognition;
3888             hx->s.error_count     = 0;
3889             hx->x264_build        = h->x264_build;
3890         }
3891
3892         avctx->execute(avctx, decode_slice, h->thread_context,
3893                        NULL, context_count, sizeof(void *));
3894
3895         /* pull back stuff from slices to master context */
3896         hx                   = h->thread_context[context_count - 1];
3897         s->mb_x              = hx->s.mb_x;
3898         s->mb_y              = hx->s.mb_y;
3899         s->droppable         = hx->s.droppable;
3900         s->picture_structure = hx->s.picture_structure;
3901         for (i = 1; i < context_count; i++)
3902             h->s.error_count += h->thread_context[i]->s.error_count;
3903     }
3904
3905     return 0;
3906 }
3907
3908 static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size,
3909                             int parse_extradata)
3910 {
3911     MpegEncContext *const s     = &h->s;
3912     AVCodecContext *const avctx = s->avctx;
3913     H264Context *hx; ///< thread context
3914     int buf_index;
3915     int context_count;
3916     int next_avc;
3917     int pass = !(avctx->active_thread_type & FF_THREAD_FRAME);
3918     int nals_needed = 0; ///< number of NALs that need decoding before the next frame thread starts
3919     int nal_index;
3920     int idr_cleared=0;
3921     int first_slice = 0;
3922
3923     h->nal_unit_type= 0;
3924
3925     if(!s->slice_context_count)
3926          s->slice_context_count= 1;
3927     h->max_contexts = s->slice_context_count;
3928     if (!(s->flags2 & CODEC_FLAG2_CHUNKS)) {
3929         h->current_slice = 0;
3930         if (!s->first_field)
3931             s->current_picture_ptr = NULL;
3932         ff_h264_reset_sei(h);
3933     }
3934
3935     if (h->nal_length_size == 4) {
3936         if (buf_size > 8 && AV_RB32(buf) == 1 && AV_RB32(buf+5) > (unsigned)buf_size) {
3937             h->is_avc = 0;
3938         }else if(buf_size > 3 && AV_RB32(buf) > 1 && AV_RB32(buf) <= (unsigned)buf_size)
3939             h->is_avc = 1;
3940     }
3941
3942     for (; pass <= 1; pass++) {
3943         buf_index     = 0;
3944         context_count = 0;
3945         next_avc      = h->is_avc ? 0 : buf_size;
3946         nal_index     = 0;
3947         for (;;) {
3948             int consumed;
3949             int dst_length;
3950             int bit_length;
3951             const uint8_t *ptr;
3952             int i, nalsize = 0;
3953             int err;
3954
3955             if (buf_index >= next_avc) {
3956                 if (buf_index >= buf_size - h->nal_length_size)
3957                     break;
3958                 nalsize = 0;
3959                 for (i = 0; i < h->nal_length_size; i++)
3960                     nalsize = (nalsize << 8) | buf[buf_index++];
3961                 if (nalsize <= 0 || nalsize > buf_size - buf_index) {
3962                     av_log(h->s.avctx, AV_LOG_ERROR,
3963                            "AVC: nal size %d\n", nalsize);
3964                     break;
3965                 }
3966                 next_avc = buf_index + nalsize;
3967             } else {
3968                 // start code prefix search
3969                 for (; buf_index + 3 < next_avc; buf_index++)
3970                     // This should always succeed in the first iteration.
3971                     if (buf[buf_index]     == 0 &&
3972                         buf[buf_index + 1] == 0 &&
3973                         buf[buf_index + 2] == 1)
3974                         break;
3975
3976                 if (buf_index + 3 >= buf_size) {
3977                     buf_index = buf_size;
3978                     break;
3979                 }
3980
3981                 buf_index += 3;
3982                 if (buf_index >= next_avc)
3983                     continue;
3984             }
3985
3986             hx = h->thread_context[context_count];
3987
3988             ptr = ff_h264_decode_nal(hx, buf + buf_index, &dst_length,
3989                                      &consumed, next_avc - buf_index);
3990             if (ptr == NULL || dst_length < 0) {
3991                 buf_index = -1;
3992                 goto end;
3993             }
3994             i = buf_index + consumed;
3995             if ((s->workaround_bugs & FF_BUG_AUTODETECT) && i + 3 < next_avc &&
3996                 buf[i]     == 0x00 && buf[i + 1] == 0x00 &&
3997                 buf[i + 2] == 0x01 && buf[i + 3] == 0xE0)
3998                 s->workaround_bugs |= FF_BUG_TRUNCATED;
3999
4000             if (!(s->workaround_bugs & FF_BUG_TRUNCATED))
4001                 while(dst_length > 0 && ptr[dst_length - 1] == 0)
4002                     dst_length--;
4003             bit_length = !dst_length ? 0
4004                                      : (8 * dst_length -
4005                                         decode_rbsp_trailing(h, ptr + dst_length - 1));
4006
4007             if (s->avctx->debug & FF_DEBUG_STARTCODE)
4008                 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);
4009
4010             if (h->is_avc && (nalsize != consumed) && nalsize)
4011                 av_log(h->s.avctx, AV_LOG_DEBUG,
4012                        "AVC: Consumed only %d bytes instead of %d\n",
4013                        consumed, nalsize);
4014
4015             buf_index += consumed;
4016             nal_index++;
4017
4018             if (pass == 0) {
4019                 /* packets can sometimes contain multiple PPS/SPS,
4020                  * e.g. two PAFF field pictures in one packet, or a demuxer
4021                  * which splits NALs strangely if so, when frame threading we
4022                  * can't start the next thread until we've read all of them */
4023                 switch (hx->nal_unit_type) {
4024                 case NAL_SPS:
4025                 case NAL_PPS:
4026                     nals_needed = nal_index;
4027                     break;
4028                 case NAL_DPA:
4029                 case NAL_IDR_SLICE:
4030                 case NAL_SLICE:
4031                     init_get_bits(&hx->s.gb, ptr, bit_length);
4032                     if (!get_ue_golomb(&hx->s.gb) || !first_slice)
4033                         nals_needed = nal_index;
4034                     if (!first_slice)
4035                         first_slice = hx->nal_unit_type;
4036                 }
4037                 continue;
4038             }
4039
4040             if (!first_slice)
4041                 switch (hx->nal_unit_type) {
4042                 case NAL_DPA:
4043                 case NAL_IDR_SLICE:
4044                 case NAL_SLICE:
4045                     first_slice = hx->nal_unit_type;
4046                 }
4047
4048             // FIXME do not discard SEI id
4049             if (avctx->skip_frame >= AVDISCARD_NONREF && h->nal_ref_idc == 0)
4050                 continue;
4051
4052 again:
4053             /* Ignore per frame NAL unit type during extradata
4054              * parsing. Decoding slices is not possible in codec init
4055              * with frame-mt */
4056             if (parse_extradata) {
4057                 switch (hx->nal_unit_type) {
4058                 case NAL_IDR_SLICE:
4059                 case NAL_SLICE:
4060                 case NAL_DPA:
4061                 case NAL_DPB:
4062                 case NAL_DPC:
4063                 case NAL_AUXILIARY_SLICE:
4064                     av_log(h->s.avctx, AV_LOG_WARNING, "Ignoring NAL %d in global header/extradata\n", hx->nal_unit_type);
4065                     hx->nal_unit_type = NAL_FF_IGNORE;
4066                 }
4067             }
4068
4069             err = 0;
4070
4071             switch (hx->nal_unit_type) {
4072             case NAL_IDR_SLICE:
4073                 if (first_slice != NAL_IDR_SLICE) {
4074                     av_log(h->s.avctx, AV_LOG_ERROR,
4075                            "Invalid mix of idr and non-idr slices\n");
4076                     buf_index = -1;
4077                     goto end;
4078                 }
4079                 if(!idr_cleared)
4080                     idr(h); // FIXME ensure we don't lose some frames if there is reordering
4081                 idr_cleared = 1;
4082             case NAL_SLICE:
4083                 init_get_bits(&hx->s.gb, ptr, bit_length);
4084                 hx->intra_gb_ptr        =
4085                     hx->inter_gb_ptr    = &hx->s.gb;
4086                 hx->s.data_partitioning = 0;
4087
4088                 if ((err = decode_slice_header(hx, h)))
4089                     break;
4090
4091                 if (h->sei_recovery_frame_cnt >= 0 && (h->frame_num != h->sei_recovery_frame_cnt || hx->slice_type_nos != AV_PICTURE_TYPE_I))
4092                     h->valid_recovery_point = 1;
4093
4094                 if (   h->sei_recovery_frame_cnt >= 0
4095                     && (   h->recovery_frame<0
4096                         || ((h->recovery_frame - h->frame_num) & ((1 << h->sps.log2_max_frame_num)-1)) > h->sei_recovery_frame_cnt)) {
4097                     h->recovery_frame = (h->frame_num + h->sei_recovery_frame_cnt) %
4098                                         (1 << h->sps.log2_max_frame_num);
4099
4100                     if (!h->valid_recovery_point)
4101                         h->recovery_frame = h->frame_num;
4102                 }
4103
4104                 s->current_picture_ptr->f.key_frame |=
4105                         (hx->nal_unit_type == NAL_IDR_SLICE);
4106
4107                 if (h->recovery_frame == h->frame_num) {
4108                     s->current_picture_ptr->sync |= 1;
4109                     h->recovery_frame = -1;
4110                 }
4111
4112                 h->sync |= !!s->current_picture_ptr->f.key_frame;
4113                 h->sync |= 3*!!(s->flags2 & CODEC_FLAG2_SHOW_ALL);
4114                 s->current_picture_ptr->sync |= h->sync;
4115
4116                 if (h->current_slice == 1) {
4117                     if (!(s->flags2 & CODEC_FLAG2_CHUNKS))
4118                         decode_postinit(h, nal_index >= nals_needed);
4119
4120                     if (s->avctx->hwaccel &&
4121                         s->avctx->hwaccel->start_frame(s->avctx, NULL, 0) < 0)
4122                         return -1;
4123                     if (CONFIG_H264_VDPAU_DECODER &&
4124                         s->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU)
4125                         ff_vdpau_h264_picture_start(s);
4126                 }
4127
4128                 if (hx->redundant_pic_count == 0 &&
4129                     (avctx->skip_frame < AVDISCARD_NONREF ||
4130                      hx->nal_ref_idc) &&
4131                     (avctx->skip_frame < AVDISCARD_BIDIR  ||
4132                      hx->slice_type_nos != AV_PICTURE_TYPE_B) &&
4133                     (avctx->skip_frame < AVDISCARD_NONKEY ||
4134                      hx->slice_type_nos == AV_PICTURE_TYPE_I) &&
4135                     avctx->skip_frame < AVDISCARD_ALL) {
4136                     if (avctx->hwaccel) {
4137                         if (avctx->hwaccel->decode_slice(avctx,
4138                                                          &buf[buf_index - consumed],
4139                                                          consumed) < 0)
4140                             return -1;
4141                     } else if (CONFIG_H264_VDPAU_DECODER &&
4142                                s->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU) {
4143                         static const uint8_t start_code[] = {
4144                             0x00, 0x00, 0x01 };
4145                         ff_vdpau_add_data_chunk(s, start_code,
4146                                                 sizeof(start_code));
4147                         ff_vdpau_add_data_chunk(s, &buf[buf_index - consumed],
4148                                                 consumed);
4149                     } else
4150                         context_count++;
4151                 }
4152                 break;
4153             case NAL_DPA:
4154                 init_get_bits(&hx->s.gb, ptr, bit_length);
4155                 hx->intra_gb_ptr =
4156                 hx->inter_gb_ptr = NULL;
4157
4158                 if ((err = decode_slice_header(hx, h)) < 0)
4159                     break;
4160
4161                 hx->s.data_partitioning = 1;
4162                 break;
4163             case NAL_DPB:
4164                 init_get_bits(&hx->intra_gb, ptr, bit_length);
4165                 hx->intra_gb_ptr = &hx->intra_gb;
4166                 break;
4167             case NAL_DPC:
4168                 init_get_bits(&hx->inter_gb, ptr, bit_length);
4169                 hx->inter_gb_ptr = &hx->inter_gb;
4170
4171                 av_log(h->s.avctx, AV_LOG_ERROR, "Partitioned H.264 support is incomplete\n");
4172                 break;
4173
4174                 if (hx->redundant_pic_count == 0 &&
4175                     hx->intra_gb_ptr &&
4176                     hx->s.data_partitioning &&
4177                     s->current_picture_ptr &&
4178                     s->context_initialized &&
4179                     (avctx->skip_frame < AVDISCARD_NONREF || hx->nal_ref_idc) &&
4180                     (avctx->skip_frame < AVDISCARD_BIDIR  ||
4181                      hx->slice_type_nos != AV_PICTURE_TYPE_B) &&
4182                     (avctx->skip_frame < AVDISCARD_NONKEY ||
4183                      hx->slice_type_nos == AV_PICTURE_TYPE_I) &&
4184                     avctx->skip_frame < AVDISCARD_ALL)
4185                     context_count++;
4186                 break;
4187             case NAL_SEI:
4188                 init_get_bits(&s->gb, ptr, bit_length);
4189                 ff_h264_decode_sei(h);
4190                 break;
4191             case NAL_SPS:
4192                 init_get_bits(&s->gb, ptr, bit_length);
4193                 if (ff_h264_decode_seq_parameter_set(h) < 0 && (h->is_avc ? (nalsize != consumed) && nalsize : 1)) {
4194                     av_log(h->s.avctx, AV_LOG_DEBUG,
4195                            "SPS decoding failure, trying again with the complete NAL\n");
4196                     if (h->is_avc)
4197                         av_assert0(next_avc - buf_index + consumed == nalsize);
4198                     if ((next_avc - buf_index + consumed - 1) >= INT_MAX/8)
4199                         break;
4200                     init_get_bits(&s->gb, &buf[buf_index + 1 - consumed],
4201                                   8*(next_avc - buf_index + consumed - 1));
4202                     ff_h264_decode_seq_parameter_set(h);
4203                 }
4204
4205                 break;
4206             case NAL_PPS:
4207                 init_get_bits(&s->gb, ptr, bit_length);
4208                 ff_h264_decode_picture_parameter_set(h, bit_length);
4209                 break;
4210             case NAL_AUD:
4211             case NAL_END_SEQUENCE:
4212             case NAL_END_STREAM:
4213             case NAL_FILLER_DATA:
4214             case NAL_SPS_EXT:
4215             case NAL_AUXILIARY_SLICE:
4216                 break;
4217             case NAL_FF_IGNORE:
4218                 break;
4219             default:
4220                 av_log(avctx, AV_LOG_DEBUG, "Unknown NAL code: %d (%d bits)\n",
4221                        hx->nal_unit_type, bit_length);
4222             }
4223
4224             if (context_count == h->max_contexts) {
4225                 execute_decode_slices(h, context_count);
4226                 context_count = 0;
4227             }
4228
4229             if (err < 0)
4230                 av_log(h->s.avctx, AV_LOG_ERROR, "decode_slice_header error\n");
4231             else if (err == 1) {
4232                 /* Slice could not be decoded in parallel mode, copy down
4233                  * NAL unit stuff to context 0 and restart. Note that
4234                  * rbsp_buffer is not transferred, but since we no longer
4235                  * run in parallel mode this should not be an issue. */
4236                 h->nal_unit_type = hx->nal_unit_type;
4237                 h->nal_ref_idc   = hx->nal_ref_idc;
4238                 hx               = h;
4239                 goto again;
4240             }
4241         }
4242     }
4243     if (context_count)
4244         execute_decode_slices(h, context_count);
4245
4246 end:
4247     /* clean up */
4248     if (s->current_picture_ptr && s->current_picture_ptr->owner2 == s &&
4249         !s->droppable) {
4250         ff_thread_report_progress(&s->current_picture_ptr->f, INT_MAX,
4251                                   s->picture_structure == PICT_BOTTOM_FIELD);
4252     }
4253
4254     return buf_index;
4255 }
4256
4257 /**
4258  * Return the number of bytes consumed for building the current frame.
4259  */
4260 static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size)
4261 {
4262     if (pos == 0)
4263         pos = 1;          // avoid infinite loops (i doubt that is needed but ...)
4264     if (pos + 10 > buf_size)
4265         pos = buf_size;                   // oops ;)
4266
4267     return pos;
4268 }
4269
4270 static int decode_frame(AVCodecContext *avctx, void *data,
4271                         int *got_frame, AVPacket *avpkt)
4272 {
4273     const uint8_t *buf = avpkt->data;
4274     int buf_size       = avpkt->size;
4275     H264Context *h     = avctx->priv_data;
4276     MpegEncContext *s  = &h->s;
4277     AVFrame *pict      = data;
4278     int buf_index      = 0;
4279     Picture *out;
4280     int i, out_idx;
4281
4282     s->flags  = avctx->flags;
4283     s->flags2 = avctx->flags2;
4284
4285     /* end of stream, output what is still in the buffers */
4286     if (buf_size == 0) {
4287  out:
4288
4289         s->current_picture_ptr = NULL;
4290
4291         // FIXME factorize this with the output code below
4292         out     = h->delayed_pic[0];
4293         out_idx = 0;
4294         for (i = 1;
4295              h->delayed_pic[i] &&
4296              !h->delayed_pic[i]->f.key_frame &&
4297              !h->delayed_pic[i]->mmco_reset;
4298              i++)
4299             if (h->delayed_pic[i]->poc < out->poc) {
4300                 out     = h->delayed_pic[i];
4301                 out_idx = i;
4302             }
4303
4304         for (i = out_idx; h->delayed_pic[i]; i++)
4305             h->delayed_pic[i] = h->delayed_pic[i + 1];
4306
4307         if (out) {
4308             *got_frame = 1;
4309             *pict      = out->f;
4310         }
4311
4312         return buf_index;
4313     }
4314     if(h->is_avc && buf_size >= 9 && buf[0]==1 && buf[2]==0 && (buf[4]&0xFC)==0xFC && (buf[5]&0x1F) && buf[8]==0x67){
4315         int cnt= buf[5]&0x1f;
4316         const uint8_t *p= buf+6;
4317         while(cnt--){
4318             int nalsize= AV_RB16(p) + 2;
4319             if(nalsize > buf_size - (p-buf) || p[2]!=0x67)
4320                 goto not_extra;
4321             p += nalsize;
4322         }
4323         cnt = *(p++);
4324         if(!cnt)
4325             goto not_extra;
4326         while(cnt--){
4327             int nalsize= AV_RB16(p) + 2;
4328             if(nalsize > buf_size - (p-buf) || p[2]!=0x68)
4329                 goto not_extra;
4330             p += nalsize;
4331         }
4332
4333         return ff_h264_decode_extradata(h, buf, buf_size);
4334     }
4335 not_extra:
4336
4337     buf_index = decode_nal_units(h, buf, buf_size, 0);
4338     if (buf_index < 0)
4339         return -1;
4340
4341     if (!s->current_picture_ptr && h->nal_unit_type == NAL_END_SEQUENCE) {
4342         av_assert0(buf_index <= buf_size);
4343         goto out;
4344     }
4345
4346     if (!(s->flags2 & CODEC_FLAG2_CHUNKS) && !s->current_picture_ptr) {
4347         if (avctx->skip_frame >= AVDISCARD_NONREF ||
4348             buf_size >= 4 && !memcmp("Q264", buf, 4))
4349             return buf_size;
4350         av_log(avctx, AV_LOG_ERROR, "no frame!\n");
4351         return -1;
4352     }
4353
4354     if (!(s->flags2 & CODEC_FLAG2_CHUNKS) ||
4355         (s->mb_y >= s->mb_height && s->mb_height)) {
4356         if (s->flags2 & CODEC_FLAG2_CHUNKS)
4357             decode_postinit(h, 1);
4358
4359         field_end(h, 0);
4360         h->context_reinitialized = 0;
4361
4362         /* Wait for second field. */
4363         *got_frame = 0;
4364         if (h->next_output_pic && (h->next_output_pic->sync || h->sync>1)) {
4365             *got_frame = 1;
4366             *pict      = h->next_output_pic->f;
4367         }
4368     }
4369
4370     assert(pict->data[0] || !*got_frame);
4371     ff_print_debug_info(s, pict);
4372
4373     return get_consumed_bytes(s, buf_index, buf_size);
4374 }
4375
4376 av_cold void ff_h264_free_context(H264Context *h)
4377 {
4378     int i;
4379
4380     free_tables(h, 1); // FIXME cleanup init stuff perhaps
4381
4382     for (i = 0; i < MAX_SPS_COUNT; i++)
4383         av_freep(h->sps_buffers + i);
4384
4385     for (i = 0; i < MAX_PPS_COUNT; i++)
4386         av_freep(h->pps_buffers + i);
4387 }
4388
4389 static av_cold int h264_decode_end(AVCodecContext *avctx)
4390 {
4391     H264Context *h    = avctx->priv_data;
4392     MpegEncContext *s = &h->s;
4393
4394     ff_h264_remove_all_refs(h);
4395     ff_h264_free_context(h);
4396
4397     ff_MPV_common_end(s);
4398
4399     // memset(h, 0, sizeof(H264Context));
4400
4401     return 0;
4402 }
4403
4404 static const AVProfile profiles[] = {
4405     { FF_PROFILE_H264_BASELINE,             "Baseline"              },
4406     { FF_PROFILE_H264_CONSTRAINED_BASELINE, "Constrained Baseline"  },
4407     { FF_PROFILE_H264_MAIN,                 "Main"                  },
4408     { FF_PROFILE_H264_EXTENDED,             "Extended"              },
4409     { FF_PROFILE_H264_HIGH,                 "High"                  },
4410     { FF_PROFILE_H264_HIGH_10,              "High 10"               },
4411     { FF_PROFILE_H264_HIGH_10_INTRA,        "High 10 Intra"         },
4412     { FF_PROFILE_H264_HIGH_422,             "High 4:2:2"            },
4413     { FF_PROFILE_H264_HIGH_422_INTRA,       "High 4:2:2 Intra"      },
4414     { FF_PROFILE_H264_HIGH_444,             "High 4:4:4"            },
4415     { FF_PROFILE_H264_HIGH_444_PREDICTIVE,  "High 4:4:4 Predictive" },
4416     { FF_PROFILE_H264_HIGH_444_INTRA,       "High 4:4:4 Intra"      },
4417     { FF_PROFILE_H264_CAVLC_444,            "CAVLC 4:4:4"           },
4418     { FF_PROFILE_UNKNOWN },
4419 };
4420
4421 static const AVOption h264_options[] = {
4422     {"is_avc", "is avc", offsetof(H264Context, is_avc), FF_OPT_TYPE_INT, {.i64 = 0}, 0, 1, 0},
4423     {"nal_length_size", "nal_length_size", offsetof(H264Context, nal_length_size), FF_OPT_TYPE_INT, {.i64 = 0}, 0, 4, 0},
4424     {NULL}
4425 };
4426
4427 static const AVClass h264_class = {
4428     .class_name = "H264 Decoder",
4429     .item_name  = av_default_item_name,
4430     .option     = h264_options,
4431     .version    = LIBAVUTIL_VERSION_INT,
4432 };
4433
4434 static const AVClass h264_vdpau_class = {
4435     .class_name = "H264 VDPAU Decoder",
4436     .item_name  = av_default_item_name,
4437     .option     = h264_options,
4438     .version    = LIBAVUTIL_VERSION_INT,
4439 };
4440
4441 AVCodec ff_h264_decoder = {
4442     .name                  = "h264",
4443     .type                  = AVMEDIA_TYPE_VIDEO,
4444     .id                    = AV_CODEC_ID_H264,
4445     .priv_data_size        = sizeof(H264Context),
4446     .init                  = ff_h264_decode_init,
4447     .close                 = h264_decode_end,
4448     .decode                = decode_frame,
4449     .capabilities          = /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 |
4450                              CODEC_CAP_DELAY | CODEC_CAP_SLICE_THREADS |
4451                              CODEC_CAP_FRAME_THREADS,
4452     .flush                 = flush_dpb,
4453     .long_name             = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10"),
4454     .init_thread_copy      = ONLY_IF_THREADS_ENABLED(decode_init_thread_copy),
4455     .update_thread_context = ONLY_IF_THREADS_ENABLED(decode_update_thread_context),
4456     .profiles              = NULL_IF_CONFIG_SMALL(profiles),
4457     .priv_class            = &h264_class,
4458 };
4459
4460 #if CONFIG_H264_VDPAU_DECODER
4461 AVCodec ff_h264_vdpau_decoder = {
4462     .name           = "h264_vdpau",
4463     .type           = AVMEDIA_TYPE_VIDEO,
4464     .id             = AV_CODEC_ID_H264,
4465     .priv_data_size = sizeof(H264Context),
4466     .init           = ff_h264_decode_init,
4467     .close          = h264_decode_end,
4468     .decode         = decode_frame,
4469     .capabilities   = CODEC_CAP_DR1 | CODEC_CAP_DELAY | CODEC_CAP_HWACCEL_VDPAU,
4470     .flush          = flush_dpb,
4471     .long_name      = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10 (VDPAU acceleration)"),
4472     .pix_fmts       = (const enum AVPixelFormat[]) { AV_PIX_FMT_VDPAU_H264,
4473                                                    AV_PIX_FMT_NONE},
4474     .profiles       = NULL_IF_CONFIG_SMALL(profiles),
4475     .priv_class     = &h264_vdpau_class,
4476 };
4477 #endif