FFmpeg
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
h264_slice.c
Go to the documentation of this file.
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 #include "libavutil/avassert.h"
29 #include "libavutil/imgutils.h"
30 #include "libavutil/timer.h"
31 #include "internal.h"
32 #include "cabac.h"
33 #include "cabac_functions.h"
34 #include "error_resilience.h"
35 #include "avcodec.h"
36 #include "h264.h"
37 #include "h264data.h"
38 #include "h264chroma.h"
39 #include "h264_mvpred.h"
40 #include "golomb.h"
41 #include "mathops.h"
42 #include "mpegutils.h"
43 #include "rectangle.h"
44 #include "thread.h"
45 
46 
47 static const uint8_t rem6[QP_MAX_NUM + 1] = {
48  0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2,
49  3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5,
50  0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2,
51  3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5,
52  0, 1, 2, 3,
53 };
54 
55 static const uint8_t div6[QP_MAX_NUM + 1] = {
56  0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 3, 3, 3,
57  3, 3, 3, 4, 4, 4, 4, 4, 4, 5, 5, 5, 5, 5, 5, 6, 6, 6, 6, 6, 6,
58  7, 7, 7, 7, 7, 7, 8, 8, 8, 8, 8, 8, 9, 9, 9, 9, 9, 9, 10, 10, 10,
59  10,10,10,11,11,11,11,11,11,12,12,12,12,12,12,13,13,13, 13, 13, 13,
60  14,14,14,14,
61 };
62 
63 static const uint8_t field_scan[16+1] = {
64  0 + 0 * 4, 0 + 1 * 4, 1 + 0 * 4, 0 + 2 * 4,
65  0 + 3 * 4, 1 + 1 * 4, 1 + 2 * 4, 1 + 3 * 4,
66  2 + 0 * 4, 2 + 1 * 4, 2 + 2 * 4, 2 + 3 * 4,
67  3 + 0 * 4, 3 + 1 * 4, 3 + 2 * 4, 3 + 3 * 4,
68 };
69 
70 static const uint8_t field_scan8x8[64+1] = {
71  0 + 0 * 8, 0 + 1 * 8, 0 + 2 * 8, 1 + 0 * 8,
72  1 + 1 * 8, 0 + 3 * 8, 0 + 4 * 8, 1 + 2 * 8,
73  2 + 0 * 8, 1 + 3 * 8, 0 + 5 * 8, 0 + 6 * 8,
74  0 + 7 * 8, 1 + 4 * 8, 2 + 1 * 8, 3 + 0 * 8,
75  2 + 2 * 8, 1 + 5 * 8, 1 + 6 * 8, 1 + 7 * 8,
76  2 + 3 * 8, 3 + 1 * 8, 4 + 0 * 8, 3 + 2 * 8,
77  2 + 4 * 8, 2 + 5 * 8, 2 + 6 * 8, 2 + 7 * 8,
78  3 + 3 * 8, 4 + 1 * 8, 5 + 0 * 8, 4 + 2 * 8,
79  3 + 4 * 8, 3 + 5 * 8, 3 + 6 * 8, 3 + 7 * 8,
80  4 + 3 * 8, 5 + 1 * 8, 6 + 0 * 8, 5 + 2 * 8,
81  4 + 4 * 8, 4 + 5 * 8, 4 + 6 * 8, 4 + 7 * 8,
82  5 + 3 * 8, 6 + 1 * 8, 6 + 2 * 8, 5 + 4 * 8,
83  5 + 5 * 8, 5 + 6 * 8, 5 + 7 * 8, 6 + 3 * 8,
84  7 + 0 * 8, 7 + 1 * 8, 6 + 4 * 8, 6 + 5 * 8,
85  6 + 6 * 8, 6 + 7 * 8, 7 + 2 * 8, 7 + 3 * 8,
86  7 + 4 * 8, 7 + 5 * 8, 7 + 6 * 8, 7 + 7 * 8,
87 };
88 
89 static const uint8_t field_scan8x8_cavlc[64+1] = {
90  0 + 0 * 8, 1 + 1 * 8, 2 + 0 * 8, 0 + 7 * 8,
91  2 + 2 * 8, 2 + 3 * 8, 2 + 4 * 8, 3 + 3 * 8,
92  3 + 4 * 8, 4 + 3 * 8, 4 + 4 * 8, 5 + 3 * 8,
93  5 + 5 * 8, 7 + 0 * 8, 6 + 6 * 8, 7 + 4 * 8,
94  0 + 1 * 8, 0 + 3 * 8, 1 + 3 * 8, 1 + 4 * 8,
95  1 + 5 * 8, 3 + 1 * 8, 2 + 5 * 8, 4 + 1 * 8,
96  3 + 5 * 8, 5 + 1 * 8, 4 + 5 * 8, 6 + 1 * 8,
97  5 + 6 * 8, 7 + 1 * 8, 6 + 7 * 8, 7 + 5 * 8,
98  0 + 2 * 8, 0 + 4 * 8, 0 + 5 * 8, 2 + 1 * 8,
99  1 + 6 * 8, 4 + 0 * 8, 2 + 6 * 8, 5 + 0 * 8,
100  3 + 6 * 8, 6 + 0 * 8, 4 + 6 * 8, 6 + 2 * 8,
101  5 + 7 * 8, 6 + 4 * 8, 7 + 2 * 8, 7 + 6 * 8,
102  1 + 0 * 8, 1 + 2 * 8, 0 + 6 * 8, 3 + 0 * 8,
103  1 + 7 * 8, 3 + 2 * 8, 2 + 7 * 8, 4 + 2 * 8,
104  3 + 7 * 8, 5 + 2 * 8, 4 + 7 * 8, 5 + 4 * 8,
105  6 + 3 * 8, 6 + 5 * 8, 7 + 3 * 8, 7 + 7 * 8,
106 };
107 
108 // zigzag_scan8x8_cavlc[i] = zigzag_scan8x8[(i/4) + 16*(i%4)]
109 static const uint8_t zigzag_scan8x8_cavlc[64+1] = {
110  0 + 0 * 8, 1 + 1 * 8, 1 + 2 * 8, 2 + 2 * 8,
111  4 + 1 * 8, 0 + 5 * 8, 3 + 3 * 8, 7 + 0 * 8,
112  3 + 4 * 8, 1 + 7 * 8, 5 + 3 * 8, 6 + 3 * 8,
113  2 + 7 * 8, 6 + 4 * 8, 5 + 6 * 8, 7 + 5 * 8,
114  1 + 0 * 8, 2 + 0 * 8, 0 + 3 * 8, 3 + 1 * 8,
115  3 + 2 * 8, 0 + 6 * 8, 4 + 2 * 8, 6 + 1 * 8,
116  2 + 5 * 8, 2 + 6 * 8, 6 + 2 * 8, 5 + 4 * 8,
117  3 + 7 * 8, 7 + 3 * 8, 4 + 7 * 8, 7 + 6 * 8,
118  0 + 1 * 8, 3 + 0 * 8, 0 + 4 * 8, 4 + 0 * 8,
119  2 + 3 * 8, 1 + 5 * 8, 5 + 1 * 8, 5 + 2 * 8,
120  1 + 6 * 8, 3 + 5 * 8, 7 + 1 * 8, 4 + 5 * 8,
121  4 + 6 * 8, 7 + 4 * 8, 5 + 7 * 8, 6 + 7 * 8,
122  0 + 2 * 8, 2 + 1 * 8, 1 + 3 * 8, 5 + 0 * 8,
123  1 + 4 * 8, 2 + 4 * 8, 6 + 0 * 8, 4 + 3 * 8,
124  0 + 7 * 8, 4 + 4 * 8, 7 + 2 * 8, 3 + 6 * 8,
125  5 + 5 * 8, 6 + 5 * 8, 6 + 6 * 8, 7 + 7 * 8,
126 };
127 
128 static const uint8_t dequant4_coeff_init[6][3] = {
129  { 10, 13, 16 },
130  { 11, 14, 18 },
131  { 13, 16, 20 },
132  { 14, 18, 23 },
133  { 16, 20, 25 },
134  { 18, 23, 29 },
135 };
136 
137 static const uint8_t dequant8_coeff_init_scan[16] = {
138  0, 3, 4, 3, 3, 1, 5, 1, 4, 5, 2, 5, 3, 1, 5, 1
139 };
140 
141 static const uint8_t dequant8_coeff_init[6][6] = {
142  { 20, 18, 32, 19, 25, 24 },
143  { 22, 19, 35, 21, 28, 26 },
144  { 26, 23, 42, 24, 33, 31 },
145  { 28, 25, 45, 26, 35, 33 },
146  { 32, 28, 51, 30, 40, 38 },
147  { 36, 32, 58, 34, 46, 43 },
148 };
149 
150 static const enum AVPixelFormat h264_hwaccel_pixfmt_list_420[] = {
151 #if CONFIG_H264_DXVA2_HWACCEL
152  AV_PIX_FMT_DXVA2_VLD,
153 #endif
154 #if CONFIG_H264_VAAPI_HWACCEL
155  AV_PIX_FMT_VAAPI_VLD,
156 #endif
157 #if CONFIG_H264_VDA_HWACCEL
158  AV_PIX_FMT_VDA_VLD,
159  AV_PIX_FMT_VDA,
160 #endif
161 #if CONFIG_H264_VDPAU_HWACCEL
162  AV_PIX_FMT_VDPAU,
163 #endif
164  AV_PIX_FMT_YUV420P,
165  AV_PIX_FMT_NONE
166 };
167 
168 static const enum AVPixelFormat h264_hwaccel_pixfmt_list_jpeg_420[] = {
169 #if CONFIG_H264_DXVA2_HWACCEL
170  AV_PIX_FMT_DXVA2_VLD,
171 #endif
172 #if CONFIG_H264_VAAPI_HWACCEL
173  AV_PIX_FMT_VAAPI_VLD,
174 #endif
175 #if CONFIG_H264_VDA_HWACCEL
176  AV_PIX_FMT_VDA_VLD,
177  AV_PIX_FMT_VDA,
178 #endif
179 #if CONFIG_H264_VDPAU_HWACCEL
180  AV_PIX_FMT_VDPAU,
181 #endif
182  AV_PIX_FMT_YUVJ420P,
183  AV_PIX_FMT_NONE
184 };
185 
186 
187 static void release_unused_pictures(H264Context *h, int remove_current)
188 {
189  int i;
190 
191  /* release non reference frames */
192  for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
193  if (h->DPB[i].f.buf[0] && !h->DPB[i].reference &&
194  (remove_current || &h->DPB[i] != h->cur_pic_ptr)) {
195  ff_h264_unref_picture(h, &h->DPB[i]);
196  }
197  }
198 }
199 
200 static int alloc_scratch_buffers(H264Context *h, int linesize)
201 {
202  int alloc_size = FFALIGN(FFABS(linesize) + 32, 32);
203 
204  if (h->bipred_scratchpad)
205  return 0;
206 
207  h->bipred_scratchpad = av_malloc(16 * 6 * alloc_size);
208  // edge emu needs blocksize + filter length - 1
209  // (= 21x21 for h264)
210  h->edge_emu_buffer = av_mallocz(alloc_size * 2 * 21);
211 
212  if (!h->bipred_scratchpad || !h->edge_emu_buffer) {
213  av_freep(&h->bipred_scratchpad);
214  av_freep(&h->edge_emu_buffer);
215  return AVERROR(ENOMEM);
216  }
217 
218  return 0;
219 }
220 
221 static int init_table_pools(H264Context *h)
222 {
223  const int big_mb_num = h->mb_stride * (h->mb_height + 1) + 1;
224  const int mb_array_size = h->mb_stride * h->mb_height;
225  const int b4_stride = h->mb_width * 4 + 1;
226  const int b4_array_size = b4_stride * h->mb_height * 4;
227 
228  h->qscale_table_pool = av_buffer_pool_init(big_mb_num + h->mb_stride,
229  av_buffer_allocz);
230  h->mb_type_pool = av_buffer_pool_init((big_mb_num + h->mb_stride) *
231  sizeof(uint32_t), av_buffer_allocz);
232  h->motion_val_pool = av_buffer_pool_init(2 * (b4_array_size + 4) *
233  sizeof(int16_t), av_buffer_allocz);
234  h->ref_index_pool = av_buffer_pool_init(4 * mb_array_size, av_buffer_allocz);
235 
236  if (!h->qscale_table_pool || !h->mb_type_pool || !h->motion_val_pool ||
237  !h->ref_index_pool) {
238  av_buffer_pool_uninit(&h->qscale_table_pool);
239  av_buffer_pool_uninit(&h->mb_type_pool);
240  av_buffer_pool_uninit(&h->motion_val_pool);
241  av_buffer_pool_uninit(&h->ref_index_pool);
242  return AVERROR(ENOMEM);
243  }
244 
245  return 0;
246 }
247 
248 static int alloc_picture(H264Context *h, H264Picture *pic)
249 {
250  int i, ret = 0;
251 
252  av_assert0(!pic->f.data[0]);
253 
254  pic->tf.f = &pic->f;
255  ret = ff_thread_get_buffer(h->avctx, &pic->tf, pic->reference ?
256  AV_GET_BUFFER_FLAG_REF : 0);
257  if (ret < 0)
258  goto fail;
259 
260  h->linesize = pic->f.linesize[0];
261  h->uvlinesize = pic->f.linesize[1];
262  pic->crop = h->sps.crop;
263  pic->crop_top = h->sps.crop_top;
264  pic->crop_left= h->sps.crop_left;
265 
266  if (h->avctx->hwaccel) {
267  const AVHWAccel *hwaccel = h->avctx->hwaccel;
268  av_assert0(!pic->hwaccel_picture_private);
269  if (hwaccel->frame_priv_data_size) {
270  pic->hwaccel_priv_buf = av_buffer_allocz(hwaccel->frame_priv_data_size);
271  if (!pic->hwaccel_priv_buf)
272  return AVERROR(ENOMEM);
273  pic->hwaccel_picture_private = pic->hwaccel_priv_buf->data;
274  }
275  }
276  if (!h->avctx->hwaccel && CONFIG_GRAY && h->flags & CODEC_FLAG_GRAY && pic->f.data[2]) {
277  int h_chroma_shift, v_chroma_shift;
278  av_pix_fmt_get_chroma_sub_sample(pic->f.format,
279  &h_chroma_shift, &v_chroma_shift);
280 
281  for(i=0; i<FF_CEIL_RSHIFT(h->avctx->height, v_chroma_shift); i++) {
282  memset(pic->f.data[1] + pic->f.linesize[1]*i,
283  0x80, FF_CEIL_RSHIFT(h->avctx->width, h_chroma_shift));
284  memset(pic->f.data[2] + pic->f.linesize[2]*i,
285  0x80, FF_CEIL_RSHIFT(h->avctx->width, h_chroma_shift));
286  }
287  }
288 
289  if (!h->qscale_table_pool) {
290  ret = init_table_pools(h);
291  if (ret < 0)
292  goto fail;
293  }
294 
295  pic->qscale_table_buf = av_buffer_pool_get(h->qscale_table_pool);
296  pic->mb_type_buf = av_buffer_pool_get(h->mb_type_pool);
297  if (!pic->qscale_table_buf || !pic->mb_type_buf)
298  goto fail;
299 
300  pic->mb_type = (uint32_t*)pic->mb_type_buf->data + 2 * h->mb_stride + 1;
301  pic->qscale_table = pic->qscale_table_buf->data + 2 * h->mb_stride + 1;
302 
303  for (i = 0; i < 2; i++) {
304  pic->motion_val_buf[i] = av_buffer_pool_get(h->motion_val_pool);
305  pic->ref_index_buf[i] = av_buffer_pool_get(h->ref_index_pool);
306  if (!pic->motion_val_buf[i] || !pic->ref_index_buf[i])
307  goto fail;
308 
309  pic->motion_val[i] = (int16_t (*)[2])pic->motion_val_buf[i]->data + 4;
310  pic->ref_index[i] = pic->ref_index_buf[i]->data;
311  }
312 
313  return 0;
314 fail:
315  ff_h264_unref_picture(h, pic);
316  return (ret < 0) ? ret : AVERROR(ENOMEM);
317 }
318 
319 static inline int pic_is_unused(H264Context *h, H264Picture *pic)
320 {
321  if (!pic->f.buf[0])
322  return 1;
323  if (pic->needs_realloc && !(pic->reference & DELAYED_PIC_REF))
324  return 1;
325  return 0;
326 }
327 
328 static int find_unused_picture(H264Context *h)
329 {
330  int i;
331 
332  for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
333  if (pic_is_unused(h, &h->DPB[i]))
334  break;
335  }
336  if (i == H264_MAX_PICTURE_COUNT)
337  return AVERROR_INVALIDDATA;
338 
339  if (h->DPB[i].needs_realloc) {
340  h->DPB[i].needs_realloc = 0;
341  ff_h264_unref_picture(h, &h->DPB[i]);
342  }
343 
344  return i;
345 }
346 
347 
348 static void init_dequant8_coeff_table(H264Context *h)
349 {
350  int i, j, q, x;
351  const int max_qp = 51 + 6 * (h->sps.bit_depth_luma - 8);
352 
353  for (i = 0; i < 6; i++) {
354  h->dequant8_coeff[i] = h->dequant8_buffer[i];
355  for (j = 0; j < i; j++)
356  if (!memcmp(h->pps.scaling_matrix8[j], h->pps.scaling_matrix8[i],
357  64 * sizeof(uint8_t))) {
358  h->dequant8_coeff[i] = h->dequant8_buffer[j];
359  break;
360  }
361  if (j < i)
362  continue;
363 
364  for (q = 0; q < max_qp + 1; q++) {
365  int shift = div6[q];
366  int idx = rem6[q];
367  for (x = 0; x < 64; x++)
368  h->dequant8_coeff[i][q][(x >> 3) | ((x & 7) << 3)] =
369  ((uint32_t)dequant8_coeff_init[idx][dequant8_coeff_init_scan[((x >> 1) & 12) | (x & 3)]] *
370  h->pps.scaling_matrix8[i][x]) << shift;
371  }
372  }
373 }
374 
375 static void init_dequant4_coeff_table(H264Context *h)
376 {
377  int i, j, q, x;
378  const int max_qp = 51 + 6 * (h->sps.bit_depth_luma - 8);
379  for (i = 0; i < 6; i++) {
380  h->dequant4_coeff[i] = h->dequant4_buffer[i];
381  for (j = 0; j < i; j++)
382  if (!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i],
383  16 * sizeof(uint8_t))) {
384  h->dequant4_coeff[i] = h->dequant4_buffer[j];
385  break;
386  }
387  if (j < i)
388  continue;
389 
390  for (q = 0; q < max_qp + 1; q++) {
391  int shift = div6[q] + 2;
392  int idx = rem6[q];
393  for (x = 0; x < 16; x++)
394  h->dequant4_coeff[i][q][(x >> 2) | ((x << 2) & 0xF)] =
395  ((uint32_t)dequant4_coeff_init[idx][(x & 1) + ((x >> 2) & 1)] *
396  h->pps.scaling_matrix4[i][x]) << shift;
397  }
398  }
399 }
400 
401 void h264_init_dequant_tables(H264Context *h)
402 {
403  int i, x;
404  init_dequant4_coeff_table(h);
405  memset(h->dequant8_coeff, 0, sizeof(h->dequant8_coeff));
406 
407  if (h->pps.transform_8x8_mode)
408  init_dequant8_coeff_table(h);
409  if (h->sps.transform_bypass) {
410  for (i = 0; i < 6; i++)
411  for (x = 0; x < 16; x++)
412  h->dequant4_coeff[i][0][x] = 1 << 6;
413  if (h->pps.transform_8x8_mode)
414  for (i = 0; i < 6; i++)
415  for (x = 0; x < 64; x++)
416  h->dequant8_coeff[i][0][x] = 1 << 6;
417  }
418 }
419 
420 /**
421  * Mimic alloc_tables(), but for every context thread.
422  */
423 static void clone_tables(H264Context *dst, H264Context *src, int i)
424 {
425  dst->intra4x4_pred_mode = src->intra4x4_pred_mode + i * 8 * 2 * src->mb_stride;
426  dst->non_zero_count = src->non_zero_count;
427  dst->slice_table = src->slice_table;
428  dst->cbp_table = src->cbp_table;
429  dst->mb2b_xy = src->mb2b_xy;
430  dst->mb2br_xy = src->mb2br_xy;
431  dst->chroma_pred_mode_table = src->chroma_pred_mode_table;
432  dst->mvd_table[0] = src->mvd_table[0] + i * 8 * 2 * src->mb_stride;
433  dst->mvd_table[1] = src->mvd_table[1] + i * 8 * 2 * src->mb_stride;
434  dst->direct_table = src->direct_table;
435  dst->list_counts = src->list_counts;
436  dst->DPB = src->DPB;
437  dst->cur_pic_ptr = src->cur_pic_ptr;
438  dst->cur_pic = src->cur_pic;
439  dst->bipred_scratchpad = NULL;
440  dst->edge_emu_buffer = NULL;
441  ff_h264_pred_init(&dst->hpc, src->avctx->codec_id, src->sps.bit_depth_luma,
442  src->sps.chroma_format_idc);
443 }
444 
445 #define IN_RANGE(a, b, size) (((a) >= (b)) && ((a) < ((b) + (size))))
446 #undef REBASE_PICTURE
447 #define REBASE_PICTURE(pic, new_ctx, old_ctx) \
448  (((pic) && (pic) >= (old_ctx)->DPB && \
449  (pic) < (old_ctx)->DPB + H264_MAX_PICTURE_COUNT) ? \
450  &(new_ctx)->DPB[(pic) - (old_ctx)->DPB] : NULL)
451 
452 static void copy_picture_range(H264Picture **to, H264Picture **from, int count,
453  H264Context *new_base,
454  H264Context *old_base)
455 {
456  int i;
457 
458  for (i = 0; i < count; i++) {
459  assert((IN_RANGE(from[i], old_base, sizeof(*old_base)) ||
460  IN_RANGE(from[i], old_base->DPB,
461  sizeof(H264Picture) * H264_MAX_PICTURE_COUNT) ||
462  !from[i]));
463  to[i] = REBASE_PICTURE(from[i], new_base, old_base);
464  }
465 }
466 
467 static int copy_parameter_set(void **to, void **from, int count, int size)
468 {
469  int i;
470 
471  for (i = 0; i < count; i++) {
472  if (to[i] && !from[i]) {
473  av_freep(&to[i]);
474  } else if (from[i] && !to[i]) {
475  to[i] = av_malloc(size);
476  if (!to[i])
477  return AVERROR(ENOMEM);
478  }
479 
480  if (from[i])
481  memcpy(to[i], from[i], size);
482  }
483 
484  return 0;
485 }
486 
487 #define copy_fields(to, from, start_field, end_field) \
488  memcpy(&(to)->start_field, &(from)->start_field, \
489  (char *)&(to)->end_field - (char *)&(to)->start_field)
490 
491 static int h264_slice_header_init(H264Context *h, int reinit);
492 
493 int ff_h264_update_thread_context(AVCodecContext *dst,
494  const AVCodecContext *src)
495 {
496  H264Context *h = dst->priv_data, *h1 = src->priv_data;
497  int inited = h->context_initialized, err = 0;
498  int context_reinitialized = 0;
499  int i, ret;
500 
501  if (dst == src)
502  return 0;
503 
504  if (inited &&
505  (h->width != h1->width ||
506  h->height != h1->height ||
507  h->mb_width != h1->mb_width ||
508  h->mb_height != h1->mb_height ||
509  h->sps.bit_depth_luma != h1->sps.bit_depth_luma ||
510  h->sps.chroma_format_idc != h1->sps.chroma_format_idc ||
511  h->sps.colorspace != h1->sps.colorspace)) {
512 
513  /* set bits_per_raw_sample to the previous value. the check for changed
514  * bit depth in h264_set_parameter_from_sps() uses it and sets it to
515  * the current value */
516  h->avctx->bits_per_raw_sample = h->sps.bit_depth_luma;
517 
518  av_freep(&h->bipred_scratchpad);
519 
520  h->width = h1->width;
521  h->height = h1->height;
522  h->mb_height = h1->mb_height;
523  h->mb_width = h1->mb_width;
524  h->mb_num = h1->mb_num;
525  h->mb_stride = h1->mb_stride;
526  h->b_stride = h1->b_stride;
527  // SPS/PPS
528  if ((ret = copy_parameter_set((void **)h->sps_buffers,
529  (void **)h1->sps_buffers,
530  MAX_SPS_COUNT, sizeof(SPS))) < 0)
531  return ret;
532  h->sps = h1->sps;
533  if ((ret = copy_parameter_set((void **)h->pps_buffers,
534  (void **)h1->pps_buffers,
535  MAX_PPS_COUNT, sizeof(PPS))) < 0)
536  return ret;
537  h->pps = h1->pps;
538 
539  if ((err = h264_slice_header_init(h, 1)) < 0) {
540  av_log(h->avctx, AV_LOG_ERROR, "h264_slice_header_init() failed\n");
541  return err;
542  }
543  context_reinitialized = 1;
544 
545 #if 0
546  h264_set_parameter_from_sps(h);
547  //Note we set context_reinitialized which will cause h264_set_parameter_from_sps to be reexecuted
548  h->cur_chroma_format_idc = h1->cur_chroma_format_idc;
549 #endif
550  }
551  /* update linesize on resize for h264. The h264 decoder doesn't
552  * necessarily call ff_mpv_frame_start in the new thread */
553  h->linesize = h1->linesize;
554  h->uvlinesize = h1->uvlinesize;
555 
556  /* copy block_offset since frame_start may not be called */
557  memcpy(h->block_offset, h1->block_offset, sizeof(h->block_offset));
558 
559  if (!inited) {
560  for (i = 0; i < MAX_SPS_COUNT; i++)
561  av_freep(h->sps_buffers + i);
562 
563  for (i = 0; i < MAX_PPS_COUNT; i++)
564  av_freep(h->pps_buffers + i);
565 
566  av_freep(&h->rbsp_buffer[0]);
567  av_freep(&h->rbsp_buffer[1]);
568  memcpy(h, h1, offsetof(H264Context, intra_pcm_ptr));
569  memcpy(&h->cabac, &h1->cabac,
570  sizeof(H264Context) - offsetof(H264Context, cabac));
571  av_assert0((void*)&h->cabac == &h->mb_padding + 1);
572 
573  memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
574  memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
575 
576  memset(&h->er, 0, sizeof(h->er));
577  memset(&h->mb, 0, sizeof(h->mb));
578  memset(&h->mb_luma_dc, 0, sizeof(h->mb_luma_dc));
579  memset(&h->mb_padding, 0, sizeof(h->mb_padding));
580  memset(&h->cur_pic, 0, sizeof(h->cur_pic));
581 
582  h->avctx = dst;
583  h->DPB = NULL;
584  h->qscale_table_pool = NULL;
585  h->mb_type_pool = NULL;
586  h->ref_index_pool = NULL;
587  h->motion_val_pool = NULL;
588  for (i = 0; i < 2; i++) {
589  h->rbsp_buffer[i] = NULL;
590  h->rbsp_buffer_size[i] = 0;
591  }
592 
593  if (h1->context_initialized) {
594  h->context_initialized = 0;
595 
596  memset(&h->cur_pic, 0, sizeof(h->cur_pic));
597  av_frame_unref(&h->cur_pic.f);
598  h->cur_pic.tf.f = &h->cur_pic.f;
599 
600  ret = ff_h264_alloc_tables(h);
601  if (ret < 0) {
602  av_log(dst, AV_LOG_ERROR, "Could not allocate memory\n");
603  return ret;
604  }
605  ret = ff_h264_context_init(h);
606  if (ret < 0) {
607  av_log(dst, AV_LOG_ERROR, "context_init() failed.\n");
608  return ret;
609  }
610  }
611 
612  h->bipred_scratchpad = NULL;
613  h->edge_emu_buffer = NULL;
614 
615  h->thread_context[0] = h;
616  h->context_initialized = h1->context_initialized;
617  }
618 
619  h->avctx->coded_height = h1->avctx->coded_height;
620  h->avctx->coded_width = h1->avctx->coded_width;
621  h->avctx->width = h1->avctx->width;
622  h->avctx->height = h1->avctx->height;
623  h->coded_picture_number = h1->coded_picture_number;
624  h->first_field = h1->first_field;
625  h->picture_structure = h1->picture_structure;
626  h->qscale = h1->qscale;
627  h->droppable = h1->droppable;
628  h->low_delay = h1->low_delay;
629 
630  for (i = 0; h->DPB && i < H264_MAX_PICTURE_COUNT; i++) {
631  ff_h264_unref_picture(h, &h->DPB[i]);
632  if (h1->DPB && h1->DPB[i].f.buf[0] &&
633  (ret = ff_h264_ref_picture(h, &h->DPB[i], &h1->DPB[i])) < 0)
634  return ret;
635  }
636 
637  h->cur_pic_ptr = REBASE_PICTURE(h1->cur_pic_ptr, h, h1);
638  ff_h264_unref_picture(h, &h->cur_pic);
639  if (h1->cur_pic.f.buf[0] && (ret = ff_h264_ref_picture(h, &h->cur_pic, &h1->cur_pic)) < 0)
640  return ret;
641 
642  h->workaround_bugs = h1->workaround_bugs;
643  h->low_delay = h1->low_delay;
644  h->droppable = h1->droppable;
645 
646  // extradata/NAL handling
647  h->is_avc = h1->is_avc;
648 
649  // SPS/PPS
650  if ((ret = copy_parameter_set((void **)h->sps_buffers,
651  (void **)h1->sps_buffers,
652  MAX_SPS_COUNT, sizeof(SPS))) < 0)
653  return ret;
654  h->sps = h1->sps;
655  if ((ret = copy_parameter_set((void **)h->pps_buffers,
656  (void **)h1->pps_buffers,
657  MAX_PPS_COUNT, sizeof(PPS))) < 0)
658  return ret;
659  h->pps = h1->pps;
660 
661  // Dequantization matrices
662  // FIXME these are big - can they be only copied when PPS changes?
663  copy_fields(h, h1, dequant4_buffer, dequant4_coeff);
664 
665  for (i = 0; i < 6; i++)
666  h->dequant4_coeff[i] = h->dequant4_buffer[0] +
667  (h1->dequant4_coeff[i] - h1->dequant4_buffer[0]);
668 
669  for (i = 0; i < 6; i++)
670  h->dequant8_coeff[i] = h->dequant8_buffer[0] +
671  (h1->dequant8_coeff[i] - h1->dequant8_buffer[0]);
672 
673  h->dequant_coeff_pps = h1->dequant_coeff_pps;
674 
675  // POC timing
676  copy_fields(h, h1, poc_lsb, redundant_pic_count);
677 
678  // reference lists
679  copy_fields(h, h1, short_ref, cabac_init_idc);
680 
681  copy_picture_range(h->short_ref, h1->short_ref, 32, h, h1);
682  copy_picture_range(h->long_ref, h1->long_ref, 32, h, h1);
683  copy_picture_range(h->delayed_pic, h1->delayed_pic,
684  MAX_DELAYED_PIC_COUNT + 2, h, h1);
685 
686  h->frame_recovered = h1->frame_recovered;
687 
688  if (context_reinitialized)
689  ff_h264_set_parameter_from_sps(h);
690 
691  if (!h->cur_pic_ptr)
692  return 0;
693 
694  if (!h->droppable) {
695  err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
696  h->prev_poc_msb = h->poc_msb;
697  h->prev_poc_lsb = h->poc_lsb;
698  }
699  h->prev_frame_num_offset = h->frame_num_offset;
700  h->prev_frame_num = h->frame_num;
701  h->outputed_poc = h->next_outputed_poc;
702 
703  h->recovery_frame = h1->recovery_frame;
704 
705  return err;
706 }
707 
708 static int h264_frame_start(H264Context *h)
709 {
710  H264Picture *pic;
711  int i, ret;
712  const int pixel_shift = h->pixel_shift;
713  int c[4] = {
714  1<<(h->sps.bit_depth_luma-1),
715  1<<(h->sps.bit_depth_chroma-1),
716  1<<(h->sps.bit_depth_chroma-1),
717  -1
718  };
719 
720  if (!ff_thread_can_start_frame(h->avctx)) {
721  av_log(h->avctx, AV_LOG_ERROR, "Attempt to start a frame outside SETUP state\n");
722  return -1;
723  }
724 
725  release_unused_pictures(h, 1);
726  h->cur_pic_ptr = NULL;
727 
728  i = find_unused_picture(h);
729  if (i < 0) {
730  av_log(h->avctx, AV_LOG_ERROR, "no frame buffer available\n");
731  return i;
732  }
733  pic = &h->DPB[i];
734 
735  pic->reference = h->droppable ? 0 : h->picture_structure;
736  pic->f.coded_picture_number = h->coded_picture_number++;
737  pic->field_picture = h->picture_structure != PICT_FRAME;
738 
739  /*
740  * Zero key_frame here; IDR markings per slice in frame or fields are ORed
741  * in later.
742  * See decode_nal_units().
743  */
744  pic->f.key_frame = 0;
745  pic->mmco_reset = 0;
746  pic->recovered = 0;
747  pic->invalid_gap = 0;
748  pic->sei_recovery_frame_cnt = h->sei_recovery_frame_cnt;
749 
750  if ((ret = alloc_picture(h, pic)) < 0)
751  return ret;
752  if(!h->frame_recovered && !h->avctx->hwaccel &&
753  !(h->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU))
754  avpriv_color_frame(&pic->f, c);
755 
756  h->cur_pic_ptr = pic;
757  ff_h264_unref_picture(h, &h->cur_pic);
758  if (CONFIG_ERROR_RESILIENCE) {
759  ff_h264_set_erpic(&h->er.cur_pic, NULL);
760  }
761 
762  if ((ret = ff_h264_ref_picture(h, &h->cur_pic, h->cur_pic_ptr)) < 0)
763  return ret;
764 
765  if (CONFIG_ERROR_RESILIENCE) {
766  ff_er_frame_start(&h->er);
767  ff_h264_set_erpic(&h->er.last_pic, NULL);
768  ff_h264_set_erpic(&h->er.next_pic, NULL);
769  }
770 
771  assert(h->linesize && h->uvlinesize);
772 
773  for (i = 0; i < 16; i++) {
774  h->block_offset[i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * h->linesize * ((scan8[i] - scan8[0]) >> 3);
775  h->block_offset[48 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * h->linesize * ((scan8[i] - scan8[0]) >> 3);
776  }
777  for (i = 0; i < 16; i++) {
778  h->block_offset[16 + i] =
779  h->block_offset[32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * h->uvlinesize * ((scan8[i] - scan8[0]) >> 3);
780  h->block_offset[48 + 16 + i] =
781  h->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * h->uvlinesize * ((scan8[i] - scan8[0]) >> 3);
782  }
783 
784  /* We mark the current picture as non-reference after allocating it, so
785  * that if we break out due to an error it can be released automatically
786  * in the next ff_mpv_frame_start().
787  */
788  h->cur_pic_ptr->reference = 0;
789 
790  h->cur_pic_ptr->field_poc[0] = h->cur_pic_ptr->field_poc[1] = INT_MAX;
791 
792  h->next_output_pic = NULL;
793 
794  assert(h->cur_pic_ptr->long_ref == 0);
795 
796  return 0;
797 }
798 
799 static av_always_inline void backup_mb_border(H264Context *h, uint8_t *src_y,
800  uint8_t *src_cb, uint8_t *src_cr,
801  int linesize, int uvlinesize,
802  int simple)
803 {
804  uint8_t *top_border;
805  int top_idx = 1;
806  const int pixel_shift = h->pixel_shift;
807  int chroma444 = CHROMA444(h);
808  int chroma422 = CHROMA422(h);
809 
810  src_y -= linesize;
811  src_cb -= uvlinesize;
812  src_cr -= uvlinesize;
813 
814  if (!simple && FRAME_MBAFF(h)) {
815  if (h->mb_y & 1) {
816  if (!MB_MBAFF(h)) {
817  top_border = h->top_borders[0][h->mb_x];
818  AV_COPY128(top_border, src_y + 15 * linesize);
819  if (pixel_shift)
820  AV_COPY128(top_border + 16, src_y + 15 * linesize + 16);
821  if (simple || !CONFIG_GRAY || !(h->flags & CODEC_FLAG_GRAY)) {
822  if (chroma444) {
823  if (pixel_shift) {
824  AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
825  AV_COPY128(top_border + 48, src_cb + 15 * uvlinesize + 16);
826  AV_COPY128(top_border + 64, src_cr + 15 * uvlinesize);
827  AV_COPY128(top_border + 80, src_cr + 15 * uvlinesize + 16);
828  } else {
829  AV_COPY128(top_border + 16, src_cb + 15 * uvlinesize);
830  AV_COPY128(top_border + 32, src_cr + 15 * uvlinesize);
831  }
832  } else if (chroma422) {
833  if (pixel_shift) {
834  AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
835  AV_COPY128(top_border + 48, src_cr + 15 * uvlinesize);
836  } else {
837  AV_COPY64(top_border + 16, src_cb + 15 * uvlinesize);
838  AV_COPY64(top_border + 24, src_cr + 15 * uvlinesize);
839  }
840  } else {
841  if (pixel_shift) {
842  AV_COPY128(top_border + 32, src_cb + 7 * uvlinesize);
843  AV_COPY128(top_border + 48, src_cr + 7 * uvlinesize);
844  } else {
845  AV_COPY64(top_border + 16, src_cb + 7 * uvlinesize);
846  AV_COPY64(top_border + 24, src_cr + 7 * uvlinesize);
847  }
848  }
849  }
850  }
851  } else if (MB_MBAFF(h)) {
852  top_idx = 0;
853  } else
854  return;
855  }
856 
857  top_border = h->top_borders[top_idx][h->mb_x];
858  /* There are two lines saved, the line above the top macroblock
859  * of a pair, and the line above the bottom macroblock. */
860  AV_COPY128(top_border, src_y + 16 * linesize);
861  if (pixel_shift)
862  AV_COPY128(top_border + 16, src_y + 16 * linesize + 16);
863 
864  if (simple || !CONFIG_GRAY || !(h->flags & CODEC_FLAG_GRAY)) {
865  if (chroma444) {
866  if (pixel_shift) {
867  AV_COPY128(top_border + 32, src_cb + 16 * linesize);
868  AV_COPY128(top_border + 48, src_cb + 16 * linesize + 16);
869  AV_COPY128(top_border + 64, src_cr + 16 * linesize);
870  AV_COPY128(top_border + 80, src_cr + 16 * linesize + 16);
871  } else {
872  AV_COPY128(top_border + 16, src_cb + 16 * linesize);
873  AV_COPY128(top_border + 32, src_cr + 16 * linesize);
874  }
875  } else if (chroma422) {
876  if (pixel_shift) {
877  AV_COPY128(top_border + 32, src_cb + 16 * uvlinesize);
878  AV_COPY128(top_border + 48, src_cr + 16 * uvlinesize);
879  } else {
880  AV_COPY64(top_border + 16, src_cb + 16 * uvlinesize);
881  AV_COPY64(top_border + 24, src_cr + 16 * uvlinesize);
882  }
883  } else {
884  if (pixel_shift) {
885  AV_COPY128(top_border + 32, src_cb + 8 * uvlinesize);
886  AV_COPY128(top_border + 48, src_cr + 8 * uvlinesize);
887  } else {
888  AV_COPY64(top_border + 16, src_cb + 8 * uvlinesize);
889  AV_COPY64(top_border + 24, src_cr + 8 * uvlinesize);
890  }
891  }
892  }
893 }
894 
895 /**
896  * Initialize implicit_weight table.
897  * @param field 0/1 initialize the weight for interlaced MBAFF
898  * -1 initializes the rest
899  */
900 static void implicit_weight_table(H264Context *h, int field)
901 {
902  int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
903 
904  for (i = 0; i < 2; i++) {
905  h->luma_weight_flag[i] = 0;
906  h->chroma_weight_flag[i] = 0;
907  }
908 
909  if (field < 0) {
910  if (h->picture_structure == PICT_FRAME) {
911  cur_poc = h->cur_pic_ptr->poc;
912  } else {
913  cur_poc = h->cur_pic_ptr->field_poc[h->picture_structure - 1];
914  }
915  if (h->ref_count[0] == 1 && h->ref_count[1] == 1 && !FRAME_MBAFF(h) &&
916  h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2 * cur_poc) {
917  h->use_weight = 0;
918  h->use_weight_chroma = 0;
919  return;
920  }
921  ref_start = 0;
922  ref_count0 = h->ref_count[0];
923  ref_count1 = h->ref_count[1];
924  } else {
925  cur_poc = h->cur_pic_ptr->field_poc[field];
926  ref_start = 16;
927  ref_count0 = 16 + 2 * h->ref_count[0];
928  ref_count1 = 16 + 2 * h->ref_count[1];
929  }
930 
931  h->use_weight = 2;
932  h->use_weight_chroma = 2;
933  h->luma_log2_weight_denom = 5;
934  h->chroma_log2_weight_denom = 5;
935 
936  for (ref0 = ref_start; ref0 < ref_count0; ref0++) {
937  int poc0 = h->ref_list[0][ref0].poc;
938  for (ref1 = ref_start; ref1 < ref_count1; ref1++) {
939  int w = 32;
940  if (!h->ref_list[0][ref0].long_ref && !h->ref_list[1][ref1].long_ref) {
941  int poc1 = h->ref_list[1][ref1].poc;
942  int td = av_clip(poc1 - poc0, -128, 127);
943  if (td) {
944  int tb = av_clip(cur_poc - poc0, -128, 127);
945  int tx = (16384 + (FFABS(td) >> 1)) / td;
946  int dist_scale_factor = (tb * tx + 32) >> 8;
947  if (dist_scale_factor >= -64 && dist_scale_factor <= 128)
948  w = 64 - dist_scale_factor;
949  }
950  }
951  if (field < 0) {
952  h->implicit_weight[ref0][ref1][0] =
953  h->implicit_weight[ref0][ref1][1] = w;
954  } else {
955  h->implicit_weight[ref0][ref1][field] = w;
956  }
957  }
958  }
959 }
960 
961 /**
962  * initialize scan tables
963  */
964 static void init_scan_tables(H264Context *h)
965 {
966  int i;
967  for (i = 0; i < 16; i++) {
968 #define TRANSPOSE(x) ((x) >> 2) | (((x) << 2) & 0xF)
969  h->zigzag_scan[i] = TRANSPOSE(zigzag_scan[i]);
970  h->field_scan[i] = TRANSPOSE(field_scan[i]);
971 #undef TRANSPOSE
972  }
973  for (i = 0; i < 64; i++) {
974 #define TRANSPOSE(x) ((x) >> 3) | (((x) & 7) << 3)
975  h->zigzag_scan8x8[i] = TRANSPOSE(ff_zigzag_direct[i]);
976  h->zigzag_scan8x8_cavlc[i] = TRANSPOSE(zigzag_scan8x8_cavlc[i]);
977  h->field_scan8x8[i] = TRANSPOSE(field_scan8x8[i]);
978  h->field_scan8x8_cavlc[i] = TRANSPOSE(field_scan8x8_cavlc[i]);
979 #undef TRANSPOSE
980  }
981  if (h->sps.transform_bypass) { // FIXME same ugly
982  memcpy(h->zigzag_scan_q0 , zigzag_scan , sizeof(h->zigzag_scan_q0 ));
983  memcpy(h->zigzag_scan8x8_q0 , ff_zigzag_direct , sizeof(h->zigzag_scan8x8_q0 ));
984  memcpy(h->zigzag_scan8x8_cavlc_q0 , zigzag_scan8x8_cavlc , sizeof(h->zigzag_scan8x8_cavlc_q0));
985  memcpy(h->field_scan_q0 , field_scan , sizeof(h->field_scan_q0 ));
986  memcpy(h->field_scan8x8_q0 , field_scan8x8 , sizeof(h->field_scan8x8_q0 ));
987  memcpy(h->field_scan8x8_cavlc_q0 , field_scan8x8_cavlc , sizeof(h->field_scan8x8_cavlc_q0 ));
988  } else {
989  memcpy(h->zigzag_scan_q0 , h->zigzag_scan , sizeof(h->zigzag_scan_q0 ));
990  memcpy(h->zigzag_scan8x8_q0 , h->zigzag_scan8x8 , sizeof(h->zigzag_scan8x8_q0 ));
991  memcpy(h->zigzag_scan8x8_cavlc_q0 , h->zigzag_scan8x8_cavlc , sizeof(h->zigzag_scan8x8_cavlc_q0));
992  memcpy(h->field_scan_q0 , h->field_scan , sizeof(h->field_scan_q0 ));
993  memcpy(h->field_scan8x8_q0 , h->field_scan8x8 , sizeof(h->field_scan8x8_q0 ));
994  memcpy(h->field_scan8x8_cavlc_q0 , h->field_scan8x8_cavlc , sizeof(h->field_scan8x8_cavlc_q0 ));
995  }
996 }
997 
998 /**
999  * Replicate H264 "master" context to thread contexts.
1000  */
1001 static int clone_slice(H264Context *dst, H264Context *src)
1002 {
1003  memcpy(dst->block_offset, src->block_offset, sizeof(dst->block_offset));
1004  dst->cur_pic_ptr = src->cur_pic_ptr;
1005  dst->cur_pic = src->cur_pic;
1006  dst->linesize = src->linesize;
1007  dst->uvlinesize = src->uvlinesize;
1008  dst->first_field = src->first_field;
1009 
1010  dst->prev_poc_msb = src->prev_poc_msb;
1011  dst->prev_poc_lsb = src->prev_poc_lsb;
1012  dst->prev_frame_num_offset = src->prev_frame_num_offset;
1013  dst->prev_frame_num = src->prev_frame_num;
1014  dst->short_ref_count = src->short_ref_count;
1015 
1016  memcpy(dst->short_ref, src->short_ref, sizeof(dst->short_ref));
1017  memcpy(dst->long_ref, src->long_ref, sizeof(dst->long_ref));
1018  memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
1019 
1020  memcpy(dst->dequant4_coeff, src->dequant4_coeff, sizeof(src->dequant4_coeff));
1021  memcpy(dst->dequant8_coeff, src->dequant8_coeff, sizeof(src->dequant8_coeff));
1022 
1023  return 0;
1024 }
1025 
1026 static enum AVPixelFormat get_pixel_format(H264Context *h, int force_callback)
1027 {
1028  switch (h->sps.bit_depth_luma) {
1029  case 9:
1030  if (CHROMA444(h)) {
1031  if (h->avctx->colorspace == AVCOL_SPC_RGB) {
1032  return AV_PIX_FMT_GBRP9;
1033  } else
1034  return AV_PIX_FMT_YUV444P9;
1035  } else if (CHROMA422(h))
1036  return AV_PIX_FMT_YUV422P9;
1037  else
1038  return AV_PIX_FMT_YUV420P9;
1039  break;
1040  case 10:
1041  if (CHROMA444(h)) {
1042  if (h->avctx->colorspace == AVCOL_SPC_RGB) {
1043  return AV_PIX_FMT_GBRP10;
1044  } else
1045  return AV_PIX_FMT_YUV444P10;
1046  } else if (CHROMA422(h))
1047  return AV_PIX_FMT_YUV422P10;
1048  else
1049  return AV_PIX_FMT_YUV420P10;
1050  break;
1051  case 12:
1052  if (CHROMA444(h)) {
1053  if (h->avctx->colorspace == AVCOL_SPC_RGB) {
1054  return AV_PIX_FMT_GBRP12;
1055  } else
1056  return AV_PIX_FMT_YUV444P12;
1057  } else if (CHROMA422(h))
1058  return AV_PIX_FMT_YUV422P12;
1059  else
1060  return AV_PIX_FMT_YUV420P12;
1061  break;
1062  case 14:
1063  if (CHROMA444(h)) {
1064  if (h->avctx->colorspace == AVCOL_SPC_RGB) {
1065  return AV_PIX_FMT_GBRP14;
1066  } else
1067  return AV_PIX_FMT_YUV444P14;
1068  } else if (CHROMA422(h))
1069  return AV_PIX_FMT_YUV422P14;
1070  else
1071  return AV_PIX_FMT_YUV420P14;
1072  break;
1073  case 8:
1074  if (CHROMA444(h)) {
1075  if (h->avctx->colorspace == AVCOL_SPC_RGB) {
1076  av_log(h->avctx, AV_LOG_DEBUG, "Detected GBR colorspace.\n");
1077  return AV_PIX_FMT_GBR24P;
1078  } else if (h->avctx->colorspace == AVCOL_SPC_YCGCO) {
1079  av_log(h->avctx, AV_LOG_WARNING, "Detected unsupported YCgCo colorspace.\n");
1080  }
1081  return h->avctx->color_range == AVCOL_RANGE_JPEG ? AV_PIX_FMT_YUVJ444P
1082  : AV_PIX_FMT_YUV444P;
1083  } else if (CHROMA422(h)) {
1084  return h->avctx->color_range == AVCOL_RANGE_JPEG ? AV_PIX_FMT_YUVJ422P
1085  : AV_PIX_FMT_YUV422P;
1086  } else {
1087  int i;
1088  const enum AVPixelFormat * fmt = h->avctx->codec->pix_fmts ?
1089  h->avctx->codec->pix_fmts :
1090  h->avctx->color_range == AVCOL_RANGE_JPEG ?
1091  h264_hwaccel_pixfmt_list_jpeg_420 :
1092  h264_hwaccel_pixfmt_list_420;
1093 
1094  for (i=0; fmt[i] != AV_PIX_FMT_NONE; i++)
1095  if (fmt[i] == h->avctx->pix_fmt && !force_callback)
1096  return fmt[i];
1097  return ff_thread_get_format(h->avctx, fmt);
1098  }
1099  break;
1100  default:
1101  av_log(h->avctx, AV_LOG_ERROR,
1102  "Unsupported bit depth %d\n", h->sps.bit_depth_luma);
1103  return AVERROR_INVALIDDATA;
1104  }
1105 }
1106 
1107 /* export coded and cropped frame dimensions to AVCodecContext */
1108 static int init_dimensions(H264Context *h)
1109 {
1110  int width = h->width - (h->sps.crop_right + h->sps.crop_left);
1111  int height = h->height - (h->sps.crop_top + h->sps.crop_bottom);
1112  int crop_present = h->sps.crop_left || h->sps.crop_top ||
1113  h->sps.crop_right || h->sps.crop_bottom;
1114  av_assert0(h->sps.crop_right + h->sps.crop_left < (unsigned)h->width);
1115  av_assert0(h->sps.crop_top + h->sps.crop_bottom < (unsigned)h->height);
1116 
1117  /* handle container cropping */
1118  if (!crop_present &&
1119  FFALIGN(h->avctx->width, 16) == h->width &&
1120  FFALIGN(h->avctx->height, 16) == h->height) {
1121  width = h->avctx->width;
1122  height = h->avctx->height;
1123  }
1124 
1125  if (width <= 0 || height <= 0) {
1126  av_log(h->avctx, AV_LOG_ERROR, "Invalid cropped dimensions: %dx%d.\n",
1127  width, height);
1128  if (h->avctx->err_recognition & AV_EF_EXPLODE)
1129  return AVERROR_INVALIDDATA;
1130 
1131  av_log(h->avctx, AV_LOG_WARNING, "Ignoring cropping information.\n");
1132  h->sps.crop_bottom =
1133  h->sps.crop_top =
1134  h->sps.crop_right =
1135  h->sps.crop_left =
1136  h->sps.crop = 0;
1137 
1138  width = h->width;
1139  height = h->height;
1140  }
1141 
1142  h->avctx->coded_width = h->width;
1143  h->avctx->coded_height = h->height;
1144  h->avctx->width = width;
1145  h->avctx->height = height;
1146 
1147  return 0;
1148 }
1149 
1150 static int h264_slice_header_init(H264Context *h, int reinit)
1151 {
1152  int nb_slices = (HAVE_THREADS &&
1153  h->avctx->active_thread_type & FF_THREAD_SLICE) ?
1154  h->avctx->thread_count : 1;
1155  int i, ret;
1156 
1157  ff_set_sar(h->avctx, h->sps.sar);
1158  av_pix_fmt_get_chroma_sub_sample(h->avctx->pix_fmt,
1159  &h->chroma_x_shift, &h->chroma_y_shift);
1160 
1161  if (h->sps.timing_info_present_flag) {
1162  int64_t den = h->sps.time_scale;
1163  if (h->x264_build < 44U)
1164  den *= 2;
1165  av_reduce(&h->avctx->time_base.num, &h->avctx->time_base.den,
1166  h->sps.num_units_in_tick, den, 1 << 30);
1167  }
1168 
1169  if (reinit)
1170  ff_h264_free_tables(h, 0);
1171  h->first_field = 0;
1172  h->prev_interlaced_frame = 1;
1173 
1174  init_scan_tables(h);
1175  ret = ff_h264_alloc_tables(h);
1176  if (ret < 0) {
1177  av_log(h->avctx, AV_LOG_ERROR, "Could not allocate memory\n");
1178  goto fail;
1179  }
1180 
1181  if (nb_slices > H264_MAX_THREADS || (nb_slices > h->mb_height && h->mb_height)) {
1182  int max_slices;
1183  if (h->mb_height)
1184  max_slices = FFMIN(H264_MAX_THREADS, h->mb_height);
1185  else
1186  max_slices = H264_MAX_THREADS;
1187  av_log(h->avctx, AV_LOG_WARNING, "too many threads/slices %d,"
1188  " reducing to %d\n", nb_slices, max_slices);
1189  nb_slices = max_slices;
1190  }
1191  h->slice_context_count = nb_slices;
1192 
1193  if (!HAVE_THREADS || !(h->avctx->active_thread_type & FF_THREAD_SLICE)) {
1194  ret = ff_h264_context_init(h);
1195  if (ret < 0) {
1196  av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
1197  goto fail;
1198  }
1199  } else {
1200  for (i = 1; i < h->slice_context_count; i++) {
1201  H264Context *c;
1202  c = h->thread_context[i] = av_mallocz(sizeof(H264Context));
1203  if (!c) {
1204  ret = AVERROR(ENOMEM);
1205  goto fail;
1206  }
1207  c->avctx = h->avctx;
1208  if (CONFIG_ERROR_RESILIENCE) {
1209  c->mecc = h->mecc;
1210  }
1211  c->vdsp = h->vdsp;
1212  c->h264dsp = h->h264dsp;
1213  c->h264qpel = h->h264qpel;
1214  c->h264chroma = h->h264chroma;
1215  c->sps = h->sps;
1216  c->pps = h->pps;
1217  c->pixel_shift = h->pixel_shift;
1218  c->cur_chroma_format_idc = h->cur_chroma_format_idc;
1219  c->width = h->width;
1220  c->height = h->height;
1221  c->linesize = h->linesize;
1222  c->uvlinesize = h->uvlinesize;
1223  c->chroma_x_shift = h->chroma_x_shift;
1224  c->chroma_y_shift = h->chroma_y_shift;
1225  c->qscale = h->qscale;
1226  c->droppable = h->droppable;
1227  c->data_partitioning = h->data_partitioning;
1228  c->low_delay = h->low_delay;
1229  c->mb_width = h->mb_width;
1230  c->mb_height = h->mb_height;
1231  c->mb_stride = h->mb_stride;
1232  c->mb_num = h->mb_num;
1233  c->flags = h->flags;
1234  c->workaround_bugs = h->workaround_bugs;
1235  c->pict_type = h->pict_type;
1236 
1237  init_scan_tables(c);
1238  clone_tables(c, h, i);
1239  c->context_initialized = 1;
1240  }
1241 
1242  for (i = 0; i < h->slice_context_count; i++)
1243  if ((ret = ff_h264_context_init(h->thread_context[i])) < 0) {
1244  av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
1245  goto fail;
1246  }
1247  }
1248 
1249  h->context_initialized = 1;
1250 
1251  return 0;
1252 fail:
1253  ff_h264_free_tables(h, 0);
1254  h->context_initialized = 0;
1255  return ret;
1256 }
1257 
1258 static enum AVPixelFormat non_j_pixfmt(enum AVPixelFormat a)
1259 {
1260  switch (a) {
1261  case AV_PIX_FMT_YUVJ420P: return AV_PIX_FMT_YUV420P;
1262  case AV_PIX_FMT_YUVJ422P: return AV_PIX_FMT_YUV422P;
1263  case AV_PIX_FMT_YUVJ444P: return AV_PIX_FMT_YUV444P;
1264  default:
1265  return a;
1266  }
1267 }
1268 
1269 /**
1270  * Decode a slice header.
1271  * This will (re)intialize the decoder and call h264_frame_start() as needed.
1272  *
1273  * @param h h264context
1274  * @param h0 h264 master context (differs from 'h' when doing sliced based
1275  * parallel decoding)
1276  *
1277  * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
1278  */
1279 int ff_h264_decode_slice_header(H264Context *h, H264Context *h0)
1280 {
1281  unsigned int first_mb_in_slice;
1282  unsigned int pps_id;
1283  int ret;
1284  unsigned int slice_type, tmp, i, j;
1285  int last_pic_structure, last_pic_droppable;
1286  int must_reinit;
1287  int needs_reinit = 0;
1288  int field_pic_flag, bottom_field_flag;
1289 
1290  h->qpel_put = h->h264qpel.put_h264_qpel_pixels_tab;
1291  h->qpel_avg = h->h264qpel.avg_h264_qpel_pixels_tab;
1292 
1293  first_mb_in_slice = get_ue_golomb_long(&h->gb);
1294 
1295  if (first_mb_in_slice == 0) { // FIXME better field boundary detection
1296  if (h0->current_slice && h->cur_pic_ptr && FIELD_PICTURE(h)) {
1297  ff_h264_field_end(h, 1);
1298  }
1299 
1300  h0->current_slice = 0;
1301  if (!h0->first_field) {
1302  if (h->cur_pic_ptr && !h->droppable) {
1303  ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1304  h->picture_structure == PICT_BOTTOM_FIELD);
1305  }
1306  h->cur_pic_ptr = NULL;
1307  }
1308  }
1309 
1310  slice_type = get_ue_golomb_31(&h->gb);
1311  if (slice_type > 9) {
1312  av_log(h->avctx, AV_LOG_ERROR,
1313  "slice type %d too large at %d %d\n",
1314  slice_type, h->mb_x, h->mb_y);
1315  return AVERROR_INVALIDDATA;
1316  }
1317  if (slice_type > 4) {
1318  slice_type -= 5;
1319  h->slice_type_fixed = 1;
1320  } else
1321  h->slice_type_fixed = 0;
1322 
1323  slice_type = golomb_to_pict_type[slice_type];
1324  h->slice_type = slice_type;
1325  h->slice_type_nos = slice_type & 3;
1326 
1327  if (h->nal_unit_type == NAL_IDR_SLICE &&
1328  h->slice_type_nos != AV_PICTURE_TYPE_I) {
1329  av_log(h->avctx, AV_LOG_ERROR, "A non-intra slice in an IDR NAL unit.\n");
1330  return AVERROR_INVALIDDATA;
1331  }
1332 
1333  if (
1334  (h->avctx->skip_frame >= AVDISCARD_NONREF && !h->nal_ref_idc) ||
1335  (h->avctx->skip_frame >= AVDISCARD_BIDIR && h->slice_type_nos == AV_PICTURE_TYPE_B) ||
1336  (h->avctx->skip_frame >= AVDISCARD_NONINTRA && h->slice_type_nos != AV_PICTURE_TYPE_I) ||
1337  (h->avctx->skip_frame >= AVDISCARD_NONKEY && h->nal_unit_type != NAL_IDR_SLICE) ||
1338  h->avctx->skip_frame >= AVDISCARD_ALL) {
1339  return SLICE_SKIPED;
1340  }
1341 
1342  // to make a few old functions happy, it's wrong though
1343  h->pict_type = h->slice_type;
1344 
1345  pps_id = get_ue_golomb(&h->gb);
1346  if (pps_id >= MAX_PPS_COUNT) {
1347  av_log(h->avctx, AV_LOG_ERROR, "pps_id %u out of range\n", pps_id);
1348  return AVERROR_INVALIDDATA;
1349  }
1350  if (!h0->pps_buffers[pps_id]) {
1351  av_log(h->avctx, AV_LOG_ERROR,
1352  "non-existing PPS %u referenced\n",
1353  pps_id);
1354  return AVERROR_INVALIDDATA;
1355  }
1356  if (h0->au_pps_id >= 0 && pps_id != h0->au_pps_id) {
1357  av_log(h->avctx, AV_LOG_ERROR,
1358  "PPS change from %d to %d forbidden\n",
1359  h0->au_pps_id, pps_id);
1360  return AVERROR_INVALIDDATA;
1361  }
1362  h->pps = *h0->pps_buffers[pps_id];
1363 
1364  if (!h0->sps_buffers[h->pps.sps_id]) {
1365  av_log(h->avctx, AV_LOG_ERROR,
1366  "non-existing SPS %u referenced\n",
1367  h->pps.sps_id);
1368  return AVERROR_INVALIDDATA;
1369  }
1370 
1371  if (h->pps.sps_id != h->sps.sps_id ||
1372  h->pps.sps_id != h->current_sps_id ||
1373  h0->sps_buffers[h->pps.sps_id]->new) {
1374 
1375  h->sps = *h0->sps_buffers[h->pps.sps_id];
1376 
1377  if (h->mb_width != h->sps.mb_width ||
1378  h->mb_height != h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag) ||
1379  h->avctx->bits_per_raw_sample != h->sps.bit_depth_luma ||
1380  h->cur_chroma_format_idc != h->sps.chroma_format_idc
1381  )
1382  needs_reinit = 1;
1383 
1384  if (h->bit_depth_luma != h->sps.bit_depth_luma ||
1385  h->chroma_format_idc != h->sps.chroma_format_idc) {
1386  h->bit_depth_luma = h->sps.bit_depth_luma;
1387  h->chroma_format_idc = h->sps.chroma_format_idc;
1388  needs_reinit = 1;
1389  }
1390  if ((ret = ff_h264_set_parameter_from_sps(h)) < 0)
1391  return ret;
1392  }
1393 
1394  h->avctx->profile = ff_h264_get_profile(&h->sps);
1395  h->avctx->level = h->sps.level_idc;
1396  h->avctx->refs = h->sps.ref_frame_count;
1397 
1398  must_reinit = (h->context_initialized &&
1399  ( 16*h->sps.mb_width != h->avctx->coded_width
1400  || 16*h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag) != h->avctx->coded_height
1401  || h->avctx->bits_per_raw_sample != h->sps.bit_depth_luma
1402  || h->cur_chroma_format_idc != h->sps.chroma_format_idc
1403  || av_cmp_q(h->sps.sar, h->avctx->sample_aspect_ratio)
1404  || h->mb_width != h->sps.mb_width
1405  || h->mb_height != h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag)
1406  ));
1407  if (non_j_pixfmt(h0->avctx->pix_fmt) != non_j_pixfmt(get_pixel_format(h0, 0)))
1408  must_reinit = 1;
1409 
1410  h->mb_width = h->sps.mb_width;
1411  h->mb_height = h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
1412  h->mb_num = h->mb_width * h->mb_height;
1413  h->mb_stride = h->mb_width + 1;
1414 
1415  h->b_stride = h->mb_width * 4;
1416 
1417  h->chroma_y_shift = h->sps.chroma_format_idc <= 1; // 400 uses yuv420p
1418 
1419  h->width = 16 * h->mb_width;
1420  h->height = 16 * h->mb_height;
1421 
1422  ret = init_dimensions(h);
1423  if (ret < 0)
1424  return ret;
1425 
1426  if (h->sps.video_signal_type_present_flag) {
1427  h->avctx->color_range = h->sps.full_range>0 ? AVCOL_RANGE_JPEG
1428  : AVCOL_RANGE_MPEG;
1429  if (h->sps.colour_description_present_flag) {
1430  if (h->avctx->colorspace != h->sps.colorspace)
1431  needs_reinit = 1;
1432  h->avctx->color_primaries = h->sps.color_primaries;
1433  h->avctx->color_trc = h->sps.color_trc;
1434  h->avctx->colorspace = h->sps.colorspace;
1435  }
1436  }
1437 
1438  if (h->context_initialized &&
1439  (must_reinit || needs_reinit)) {
1440  if (h != h0) {
1441  av_log(h->avctx, AV_LOG_ERROR,
1442  "changing width %d -> %d / height %d -> %d on "
1443  "slice %d\n",
1444  h->width, h->avctx->coded_width,
1445  h->height, h->avctx->coded_height,
1446  h0->current_slice + 1);
1447  return AVERROR_INVALIDDATA;
1448  }
1449 
1450  ff_h264_flush_change(h);
1451 
1452  if ((ret = get_pixel_format(h, 1)) < 0)
1453  return ret;
1454  h->avctx->pix_fmt = ret;
1455 
1456  av_log(h->avctx, AV_LOG_INFO, "Reinit context to %dx%d, "
1457  "pix_fmt: %s\n", h->width, h->height, av_get_pix_fmt_name(h->avctx->pix_fmt));
1458 
1459  if ((ret = h264_slice_header_init(h, 1)) < 0) {
1460  av_log(h->avctx, AV_LOG_ERROR,
1461  "h264_slice_header_init() failed\n");
1462  return ret;
1463  }
1464  }
1465  if (!h->context_initialized) {
1466  if (h != h0) {
1467  av_log(h->avctx, AV_LOG_ERROR,
1468  "Cannot (re-)initialize context during parallel decoding.\n");
1469  return AVERROR_PATCHWELCOME;
1470  }
1471 
1472  if ((ret = get_pixel_format(h, 1)) < 0)
1473  return ret;
1474  h->avctx->pix_fmt = ret;
1475 
1476  if ((ret = h264_slice_header_init(h, 0)) < 0) {
1477  av_log(h->avctx, AV_LOG_ERROR,
1478  "h264_slice_header_init() failed\n");
1479  return ret;
1480  }
1481  }
1482 
1483  if (h == h0 && h->dequant_coeff_pps != pps_id) {
1484  h->dequant_coeff_pps = pps_id;
1485  h264_init_dequant_tables(h);
1486  }
1487 
1488  h->frame_num = get_bits(&h->gb, h->sps.log2_max_frame_num);
1489 
1490  h->mb_mbaff = 0;
1491  h->mb_aff_frame = 0;
1492  last_pic_structure = h0->picture_structure;
1493  last_pic_droppable = h0->droppable;
1494  h->droppable = h->nal_ref_idc == 0;
1495  if (h->sps.frame_mbs_only_flag) {
1496  h->picture_structure = PICT_FRAME;
1497  } else {
1498  if (!h->sps.direct_8x8_inference_flag && slice_type == AV_PICTURE_TYPE_B) {
1499  av_log(h->avctx, AV_LOG_ERROR, "This stream was generated by a broken encoder, invalid 8x8 inference\n");
1500  return -1;
1501  }
1502  field_pic_flag = get_bits1(&h->gb);
1503  if (field_pic_flag) {
1504  bottom_field_flag = get_bits1(&h->gb);
1505  h->picture_structure = PICT_TOP_FIELD + bottom_field_flag;
1506  } else {
1507  h->picture_structure = PICT_FRAME;
1508  h->mb_aff_frame = h->sps.mb_aff;
1509  }
1510  }
1511  h->mb_field_decoding_flag = h->picture_structure != PICT_FRAME;
1512 
1513  if (h0->current_slice != 0) {
1514  if (last_pic_structure != h->picture_structure ||
1515  last_pic_droppable != h->droppable) {
1516  av_log(h->avctx, AV_LOG_ERROR,
1517  "Changing field mode (%d -> %d) between slices is not allowed\n",
1518  last_pic_structure, h->picture_structure);
1519  h->picture_structure = last_pic_structure;
1520  h->droppable = last_pic_droppable;
1521  return AVERROR_INVALIDDATA;
1522  } else if (!h0->cur_pic_ptr) {
1523  av_log(h->avctx, AV_LOG_ERROR,
1524  "unset cur_pic_ptr on slice %d\n",
1525  h0->current_slice + 1);
1526  return AVERROR_INVALIDDATA;
1527  }
1528  } else {
1529  /* Shorten frame num gaps so we don't have to allocate reference
1530  * frames just to throw them away */
1531  if (h->frame_num != h->prev_frame_num) {
1532  int unwrap_prev_frame_num = h->prev_frame_num;
1533  int max_frame_num = 1 << h->sps.log2_max_frame_num;
1534 
1535  if (unwrap_prev_frame_num > h->frame_num)
1536  unwrap_prev_frame_num -= max_frame_num;
1537 
1538  if ((h->frame_num - unwrap_prev_frame_num) > h->sps.ref_frame_count) {
1539  unwrap_prev_frame_num = (h->frame_num - h->sps.ref_frame_count) - 1;
1540  if (unwrap_prev_frame_num < 0)
1541  unwrap_prev_frame_num += max_frame_num;
1542 
1543  h->prev_frame_num = unwrap_prev_frame_num;
1544  }
1545  }
1546 
1547  /* See if we have a decoded first field looking for a pair...
1548  * Here, we're using that to see if we should mark previously
1549  * decode frames as "finished".
1550  * We have to do that before the "dummy" in-between frame allocation,
1551  * since that can modify h->cur_pic_ptr. */
1552  if (h0->first_field) {
1553  assert(h0->cur_pic_ptr);
1554  assert(h0->cur_pic_ptr->f.buf[0]);
1555  assert(h0->cur_pic_ptr->reference != DELAYED_PIC_REF);
1556 
1557  /* Mark old field/frame as completed */
1558  if (h0->cur_pic_ptr->tf.owner == h0->avctx) {
1559  ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
1560  last_pic_structure == PICT_BOTTOM_FIELD);
1561  }
1562 
1563  /* figure out if we have a complementary field pair */
1564  if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
1565  /* Previous field is unmatched. Don't display it, but let it
1566  * remain for reference if marked as such. */
1567  if (last_pic_structure != PICT_FRAME) {
1568  ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
1569  last_pic_structure == PICT_TOP_FIELD);
1570  }
1571  } else {
1572  if (h0->cur_pic_ptr->frame_num != h->frame_num) {
1573  /* This and previous field were reference, but had
1574  * different frame_nums. Consider this field first in
1575  * pair. Throw away previous field except for reference
1576  * purposes. */
1577  if (last_pic_structure != PICT_FRAME) {
1578  ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
1579  last_pic_structure == PICT_TOP_FIELD);
1580  }
1581  } else {
1582  /* Second field in complementary pair */
1583  if (!((last_pic_structure == PICT_TOP_FIELD &&
1584  h->picture_structure == PICT_BOTTOM_FIELD) ||
1585  (last_pic_structure == PICT_BOTTOM_FIELD &&
1586  h->picture_structure == PICT_TOP_FIELD))) {
1587  av_log(h->avctx, AV_LOG_ERROR,
1588  "Invalid field mode combination %d/%d\n",
1589  last_pic_structure, h->picture_structure);
1590  h->picture_structure = last_pic_structure;
1591  h->droppable = last_pic_droppable;
1592  return AVERROR_INVALIDDATA;
1593  } else if (last_pic_droppable != h->droppable) {
1594  avpriv_request_sample(h->avctx,
1595  "Found reference and non-reference fields in the same frame, which");
1596  h->picture_structure = last_pic_structure;
1597  h->droppable = last_pic_droppable;
1598  return AVERROR_PATCHWELCOME;
1599  }
1600  }
1601  }
1602  }
1603 
1604  while (h->frame_num != h->prev_frame_num && !h0->first_field &&
1605  h->frame_num != (h->prev_frame_num + 1) % (1 << h->sps.log2_max_frame_num)) {
1606  H264Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
1607  av_log(h->avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n",
1608  h->frame_num, h->prev_frame_num);
1609  if (!h->sps.gaps_in_frame_num_allowed_flag)
1610  for(i=0; i<FF_ARRAY_ELEMS(h->last_pocs); i++)
1611  h->last_pocs[i] = INT_MIN;
1612  ret = h264_frame_start(h);
1613  if (ret < 0) {
1614  h0->first_field = 0;
1615  return ret;
1616  }
1617 
1618  h->prev_frame_num++;
1619  h->prev_frame_num %= 1 << h->sps.log2_max_frame_num;
1620  h->cur_pic_ptr->frame_num = h->prev_frame_num;
1621  h->cur_pic_ptr->invalid_gap = !h->sps.gaps_in_frame_num_allowed_flag;
1622  ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 0);
1623  ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 1);
1624  ret = ff_generate_sliding_window_mmcos(h, 1);
1625  if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1626  return ret;
1627  ret = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1628  if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1629  return ret;
1630  /* Error concealment: If a ref is missing, copy the previous ref
1631  * in its place.
1632  * FIXME: Avoiding a memcpy would be nice, but ref handling makes
1633  * many assumptions about there being no actual duplicates.
1634  * FIXME: This does not copy padding for out-of-frame motion
1635  * vectors. Given we are concealing a lost frame, this probably
1636  * is not noticeable by comparison, but it should be fixed. */
1637  if (h->short_ref_count) {
1638  if (prev) {
1639  av_image_copy(h->short_ref[0]->f.data,
1640  h->short_ref[0]->f.linesize,
1641  (const uint8_t **)prev->f.data,
1642  prev->f.linesize,
1643  h->avctx->pix_fmt,
1644  h->mb_width * 16,
1645  h->mb_height * 16);
1646  h->short_ref[0]->poc = prev->poc + 2;
1647  }
1648  h->short_ref[0]->frame_num = h->prev_frame_num;
1649  }
1650  }
1651 
1652  /* See if we have a decoded first field looking for a pair...
1653  * We're using that to see whether to continue decoding in that
1654  * frame, or to allocate a new one. */
1655  if (h0->first_field) {
1656  assert(h0->cur_pic_ptr);
1657  assert(h0->cur_pic_ptr->f.buf[0]);
1658  assert(h0->cur_pic_ptr->reference != DELAYED_PIC_REF);
1659 
1660  /* figure out if we have a complementary field pair */
1661  if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
1662  /* Previous field is unmatched. Don't display it, but let it
1663  * remain for reference if marked as such. */
1664  h0->cur_pic_ptr = NULL;
1665  h0->first_field = FIELD_PICTURE(h);
1666  } else {
1667  if (h0->cur_pic_ptr->frame_num != h->frame_num) {
1668  ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
1669  h0->picture_structure==PICT_BOTTOM_FIELD);
1670  /* This and the previous field had different frame_nums.
1671  * Consider this field first in pair. Throw away previous
1672  * one except for reference purposes. */
1673  h0->first_field = 1;
1674  h0->cur_pic_ptr = NULL;
1675  } else {
1676  /* Second field in complementary pair */
1677  h0->first_field = 0;
1678  }
1679  }
1680  } else {
1681  /* Frame or first field in a potentially complementary pair */
1682  h0->first_field = FIELD_PICTURE(h);
1683  }
1684 
1685  if (!FIELD_PICTURE(h) || h0->first_field) {
1686  if (h264_frame_start(h) < 0) {
1687  h0->first_field = 0;
1688  return AVERROR_INVALIDDATA;
1689  }
1690  } else {
1691  release_unused_pictures(h, 0);
1692  }
1693  /* Some macroblocks can be accessed before they're available in case
1694  * of lost slices, MBAFF or threading. */
1695  if (FIELD_PICTURE(h)) {
1696  for(i = (h->picture_structure == PICT_BOTTOM_FIELD); i<h->mb_height; i++)
1697  memset(h->slice_table + i*h->mb_stride, -1, (h->mb_stride - (i+1==h->mb_height)) * sizeof(*h->slice_table));
1698  } else {
1699  memset(h->slice_table, -1,
1700  (h->mb_height * h->mb_stride - 1) * sizeof(*h->slice_table));
1701  }
1702  h0->last_slice_type = -1;
1703  }
1704  if (h != h0 && (ret = clone_slice(h, h0)) < 0)
1705  return ret;
1706 
1707  /* can't be in alloc_tables because linesize isn't known there.
1708  * FIXME: redo bipred weight to not require extra buffer? */
1709  for (i = 0; i < h->slice_context_count; i++)
1710  if (h->thread_context[i]) {
1711  ret = alloc_scratch_buffers(h->thread_context[i], h->linesize);
1712  if (ret < 0)
1713  return ret;
1714  }
1715 
1716  h->cur_pic_ptr->frame_num = h->frame_num; // FIXME frame_num cleanup
1717 
1718  av_assert1(h->mb_num == h->mb_width * h->mb_height);
1719  if (first_mb_in_slice << FIELD_OR_MBAFF_PICTURE(h) >= h->mb_num ||
1720  first_mb_in_slice >= h->mb_num) {
1721  av_log(h->avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
1722  return AVERROR_INVALIDDATA;
1723  }
1724  h->resync_mb_x = h->mb_x = first_mb_in_slice % h->mb_width;
1725  h->resync_mb_y = h->mb_y = (first_mb_in_slice / h->mb_width) <<
1726  FIELD_OR_MBAFF_PICTURE(h);
1727  if (h->picture_structure == PICT_BOTTOM_FIELD)
1728  h->resync_mb_y = h->mb_y = h->mb_y + 1;
1729  av_assert1(h->mb_y < h->mb_height);
1730 
1731  if (h->picture_structure == PICT_FRAME) {
1732  h->curr_pic_num = h->frame_num;
1733  h->max_pic_num = 1 << h->sps.log2_max_frame_num;
1734  } else {
1735  h->curr_pic_num = 2 * h->frame_num + 1;
1736  h->max_pic_num = 1 << (h->sps.log2_max_frame_num + 1);
1737  }
1738 
1739  if (h->nal_unit_type == NAL_IDR_SLICE)
1740  get_ue_golomb(&h->gb); /* idr_pic_id */
1741 
1742  if (h->sps.poc_type == 0) {
1743  h->poc_lsb = get_bits(&h->gb, h->sps.log2_max_poc_lsb);
1744 
1745  if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME)
1746  h->delta_poc_bottom = get_se_golomb(&h->gb);
1747  }
1748 
1749  if (h->sps.poc_type == 1 && !h->sps.delta_pic_order_always_zero_flag) {
1750  h->delta_poc[0] = get_se_golomb(&h->gb);
1751 
1752  if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME)
1753  h->delta_poc[1] = get_se_golomb(&h->gb);
1754  }
1755 
1756  ff_init_poc(h, h->cur_pic_ptr->field_poc, &h->cur_pic_ptr->poc);
1757 
1758  if (h->pps.redundant_pic_cnt_present)
1759  h->redundant_pic_count = get_ue_golomb(&h->gb);
1760 
1761  ret = ff_set_ref_count(h);
1762  if (ret < 0)
1763  return ret;
1764 
1765  if (slice_type != AV_PICTURE_TYPE_I &&
1766  (h0->current_slice == 0 ||
1767  slice_type != h0->last_slice_type ||
1768  memcmp(h0->last_ref_count, h0->ref_count, sizeof(h0->ref_count)))) {
1769 
1770  ff_h264_fill_default_ref_list(h);
1771  }
1772 
1773  if (h->slice_type_nos != AV_PICTURE_TYPE_I) {
1774  ret = ff_h264_decode_ref_pic_list_reordering(h);
1775  if (ret < 0) {
1776  h->ref_count[1] = h->ref_count[0] = 0;
1777  return ret;
1778  }
1779  }
1780 
1781  if ((h->pps.weighted_pred && h->slice_type_nos == AV_PICTURE_TYPE_P) ||
1782  (h->pps.weighted_bipred_idc == 1 &&
1783  h->slice_type_nos == AV_PICTURE_TYPE_B))
1784  ff_pred_weight_table(h);
1785  else if (h->pps.weighted_bipred_idc == 2 &&
1786  h->slice_type_nos == AV_PICTURE_TYPE_B) {
1787  implicit_weight_table(h, -1);
1788  } else {
1789  h->use_weight = 0;
1790  for (i = 0; i < 2; i++) {
1791  h->luma_weight_flag[i] = 0;
1792  h->chroma_weight_flag[i] = 0;
1793  }
1794  }
1795 
1796  // If frame-mt is enabled, only update mmco tables for the first slice
1797  // in a field. Subsequent slices can temporarily clobber h->mmco_index
1798  // or h->mmco, which will cause ref list mix-ups and decoding errors
1799  // further down the line. This may break decoding if the first slice is
1800  // corrupt, thus we only do this if frame-mt is enabled.
1801  if (h->nal_ref_idc) {
1802  ret = ff_h264_decode_ref_pic_marking(h0, &h->gb,
1803  !(h->avctx->active_thread_type & FF_THREAD_FRAME) ||
1804  h0->current_slice == 0);
1805  if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1806  return AVERROR_INVALIDDATA;
1807  }
1808 
1809  if (FRAME_MBAFF(h)) {
1810  ff_h264_fill_mbaff_ref_list(h);
1811 
1812  if (h->pps.weighted_bipred_idc == 2 && h->slice_type_nos == AV_PICTURE_TYPE_B) {
1813  implicit_weight_table(h, 0);
1814  implicit_weight_table(h, 1);
1815  }
1816  }
1817 
1818  if (h->slice_type_nos == AV_PICTURE_TYPE_B && !h->direct_spatial_mv_pred)
1819  ff_h264_direct_dist_scale_factor(h);
1820  ff_h264_direct_ref_list_init(h);
1821 
1822  if (h->slice_type_nos != AV_PICTURE_TYPE_I && h->pps.cabac) {
1823  tmp = get_ue_golomb_31(&h->gb);
1824  if (tmp > 2) {
1825  av_log(h->avctx, AV_LOG_ERROR, "cabac_init_idc %u overflow\n", tmp);
1826  return AVERROR_INVALIDDATA;
1827  }
1828  h->cabac_init_idc = tmp;
1829  }
1830 
1831  h->last_qscale_diff = 0;
1832  tmp = h->pps.init_qp + get_se_golomb(&h->gb);
1833  if (tmp > 51 + 6 * (h->sps.bit_depth_luma - 8)) {
1834  av_log(h->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
1835  return AVERROR_INVALIDDATA;
1836  }
1837  h->qscale = tmp;
1838  h->chroma_qp[0] = get_chroma_qp(h, 0, h->qscale);
1839  h->chroma_qp[1] = get_chroma_qp(h, 1, h->qscale);
1840  // FIXME qscale / qp ... stuff
1841  if (h->slice_type == AV_PICTURE_TYPE_SP)
1842  get_bits1(&h->gb); /* sp_for_switch_flag */
1843  if (h->slice_type == AV_PICTURE_TYPE_SP ||
1844  h->slice_type == AV_PICTURE_TYPE_SI)
1845  get_se_golomb(&h->gb); /* slice_qs_delta */
1846 
1847  h->deblocking_filter = 1;
1848  h->slice_alpha_c0_offset = 0;
1849  h->slice_beta_offset = 0;
1850  if (h->pps.deblocking_filter_parameters_present) {
1851  tmp = get_ue_golomb_31(&h->gb);
1852  if (tmp > 2) {
1853  av_log(h->avctx, AV_LOG_ERROR,
1854  "deblocking_filter_idc %u out of range\n", tmp);
1855  return AVERROR_INVALIDDATA;
1856  }
1857  h->deblocking_filter = tmp;
1858  if (h->deblocking_filter < 2)
1859  h->deblocking_filter ^= 1; // 1<->0
1860 
1861  if (h->deblocking_filter) {
1862  h->slice_alpha_c0_offset = get_se_golomb(&h->gb) * 2;
1863  h->slice_beta_offset = get_se_golomb(&h->gb) * 2;
1864  if (h->slice_alpha_c0_offset > 12 ||
1865  h->slice_alpha_c0_offset < -12 ||
1866  h->slice_beta_offset > 12 ||
1867  h->slice_beta_offset < -12) {
1868  av_log(h->avctx, AV_LOG_ERROR,
1869  "deblocking filter parameters %d %d out of range\n",
1870  h->slice_alpha_c0_offset, h->slice_beta_offset);
1871  return AVERROR_INVALIDDATA;
1872  }
1873  }
1874  }
1875 
1876  if (h->avctx->skip_loop_filter >= AVDISCARD_ALL ||
1877  (h->avctx->skip_loop_filter >= AVDISCARD_NONKEY &&
1878  h->nal_unit_type != NAL_IDR_SLICE) ||
1879  (h->avctx->skip_loop_filter >= AVDISCARD_NONINTRA &&
1880  h->slice_type_nos != AV_PICTURE_TYPE_I) ||
1881  (h->avctx->skip_loop_filter >= AVDISCARD_BIDIR &&
1882  h->slice_type_nos == AV_PICTURE_TYPE_B) ||
1883  (h->avctx->skip_loop_filter >= AVDISCARD_NONREF &&
1884  h->nal_ref_idc == 0))
1885  h->deblocking_filter = 0;
1886 
1887  if (h->deblocking_filter == 1 && h0->max_contexts > 1) {
1888  if (h->avctx->flags2 & CODEC_FLAG2_FAST) {
1889  /* Cheat slightly for speed:
1890  * Do not bother to deblock across slices. */
1891  h->deblocking_filter = 2;
1892  } else {
1893  h0->max_contexts = 1;
1894  if (!h0->single_decode_warning) {
1895  av_log(h->avctx, AV_LOG_INFO,
1896  "Cannot parallelize slice decoding with deblocking filter type 1, decoding such frames in sequential order\n"
1897  "To parallelize slice decoding you need video encoded with disable_deblocking_filter_idc set to 2 (deblock only edges that do not cross slices).\n"
1898  "Setting the flags2 libavcodec option to +fast (-flags2 +fast) will disable deblocking across slices and enable parallel slice decoding "
1899  "but will generate non-standard-compliant output.\n");
1900  h0->single_decode_warning = 1;
1901  }
1902  if (h != h0) {
1903  av_log(h->avctx, AV_LOG_ERROR,
1904  "Deblocking switched inside frame.\n");
1905  return SLICE_SINGLETHREAD;
1906  }
1907  }
1908  }
1909  h->qp_thresh = 15 -
1910  FFMIN(h->slice_alpha_c0_offset, h->slice_beta_offset) -
1911  FFMAX3(0,
1912  h->pps.chroma_qp_index_offset[0],
1913  h->pps.chroma_qp_index_offset[1]) +
1914  6 * (h->sps.bit_depth_luma - 8);
1915 
1916  h0->last_slice_type = slice_type;
1917  memcpy(h0->last_ref_count, h0->ref_count, sizeof(h0->last_ref_count));
1918  h->slice_num = ++h0->current_slice;
1919 
1920  if (h->slice_num)
1921  h0->slice_row[(h->slice_num-1)&(MAX_SLICES-1)]= h->resync_mb_y;
1922  if ( h0->slice_row[h->slice_num&(MAX_SLICES-1)] + 3 >= h->resync_mb_y
1923  && h0->slice_row[h->slice_num&(MAX_SLICES-1)] <= h->resync_mb_y
1924  && h->slice_num >= MAX_SLICES) {
1925  //in case of ASO this check needs to be updated depending on how we decide to assign slice numbers in this case
1926  av_log(h->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);
1927  }
1928 
1929  for (j = 0; j < 2; j++) {
1930  int id_list[16];
1931  int *ref2frm = h->ref2frm[h->slice_num & (MAX_SLICES - 1)][j];
1932  for (i = 0; i < 16; i++) {
1933  id_list[i] = 60;
1934  if (j < h->list_count && i < h->ref_count[j] &&
1935  h->ref_list[j][i].f.buf[0]) {
1936  int k;
1937  AVBuffer *buf = h->ref_list[j][i].f.buf[0]->buffer;
1938  for (k = 0; k < h->short_ref_count; k++)
1939  if (h->short_ref[k]->f.buf[0]->buffer == buf) {
1940  id_list[i] = k;
1941  break;
1942  }
1943  for (k = 0; k < h->long_ref_count; k++)
1944  if (h->long_ref[k] && h->long_ref[k]->f.buf[0]->buffer == buf) {
1945  id_list[i] = h->short_ref_count + k;
1946  break;
1947  }
1948  }
1949  }
1950 
1951  ref2frm[0] =
1952  ref2frm[1] = -1;
1953  for (i = 0; i < 16; i++)
1954  ref2frm[i + 2] = 4 * id_list[i] + (h->ref_list[j][i].reference & 3);
1955  ref2frm[18 + 0] =
1956  ref2frm[18 + 1] = -1;
1957  for (i = 16; i < 48; i++)
1958  ref2frm[i + 4] = 4 * id_list[(i - 16) >> 1] +
1959  (h->ref_list[j][i].reference & 3);
1960  }
1961 
1962  if (h->ref_count[0]) ff_h264_set_erpic(&h->er.last_pic, &h->ref_list[0][0]);
1963  if (h->ref_count[1]) ff_h264_set_erpic(&h->er.next_pic, &h->ref_list[1][0]);
1964 
1965  h->er.ref_count = h->ref_count[0];
1966  h0->au_pps_id = pps_id;
1967  h->sps.new =
1968  h0->sps_buffers[h->pps.sps_id]->new = 0;
1969  h->current_sps_id = h->pps.sps_id;
1970 
1971  if (h->avctx->debug & FF_DEBUG_PICT_INFO) {
1972  av_log(h->avctx, AV_LOG_DEBUG,
1973  "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",
1974  h->slice_num,
1975  (h->picture_structure == PICT_FRAME ? "F" : h->picture_structure == PICT_TOP_FIELD ? "T" : "B"),
1976  first_mb_in_slice,
1977  av_get_picture_type_char(h->slice_type),
1978  h->slice_type_fixed ? " fix" : "",
1979  h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
1980  pps_id, h->frame_num,
1981  h->cur_pic_ptr->field_poc[0],
1982  h->cur_pic_ptr->field_poc[1],
1983  h->ref_count[0], h->ref_count[1],
1984  h->qscale,
1985  h->deblocking_filter,
1986  h->slice_alpha_c0_offset, h->slice_beta_offset,
1987  h->use_weight,
1988  h->use_weight == 1 && h->use_weight_chroma ? "c" : "",
1989  h->slice_type == AV_PICTURE_TYPE_B ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : "");
1990  }
1991 
1992  return 0;
1993 }
1994 
1995 int ff_h264_get_slice_type(const H264Context *h)
1996 {
1997  switch (h->slice_type) {
1998  case AV_PICTURE_TYPE_P:
1999  return 0;
2000  case AV_PICTURE_TYPE_B:
2001  return 1;
2002  case AV_PICTURE_TYPE_I:
2003  return 2;
2004  case AV_PICTURE_TYPE_SP:
2005  return 3;
2006  case AV_PICTURE_TYPE_SI:
2007  return 4;
2008  default:
2009  return AVERROR_INVALIDDATA;
2010  }
2011 }
2012 
2013 static av_always_inline void fill_filter_caches_inter(H264Context *h,
2014  int mb_type, int top_xy,
2015  int left_xy[LEFT_MBS],
2016  int top_type,
2017  int left_type[LEFT_MBS],
2018  int mb_xy, int list)
2019 {
2020  int b_stride = h->b_stride;
2021  int16_t(*mv_dst)[2] = &h->mv_cache[list][scan8[0]];
2022  int8_t *ref_cache = &h->ref_cache[list][scan8[0]];
2023  if (IS_INTER(mb_type) || IS_DIRECT(mb_type)) {
2024  if (USES_LIST(top_type, list)) {
2025  const int b_xy = h->mb2b_xy[top_xy] + 3 * b_stride;
2026  const int b8_xy = 4 * top_xy + 2;
2027  int (*ref2frm)[64] = (void*)(h->ref2frm[h->slice_table[top_xy] & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2));
2028  AV_COPY128(mv_dst - 1 * 8, h->cur_pic.motion_val[list][b_xy + 0]);
2029  ref_cache[0 - 1 * 8] =
2030  ref_cache[1 - 1 * 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 0]];
2031  ref_cache[2 - 1 * 8] =
2032  ref_cache[3 - 1 * 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 1]];
2033  } else {
2034  AV_ZERO128(mv_dst - 1 * 8);
2035  AV_WN32A(&ref_cache[0 - 1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2036  }
2037 
2038  if (!IS_INTERLACED(mb_type ^ left_type[LTOP])) {
2039  if (USES_LIST(left_type[LTOP], list)) {
2040  const int b_xy = h->mb2b_xy[left_xy[LTOP]] + 3;
2041  const int b8_xy = 4 * left_xy[LTOP] + 1;
2042  int (*ref2frm)[64] =(void*)( h->ref2frm[h->slice_table[left_xy[LTOP]] & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2));
2043  AV_COPY32(mv_dst - 1 + 0, h->cur_pic.motion_val[list][b_xy + b_stride * 0]);
2044  AV_COPY32(mv_dst - 1 + 8, h->cur_pic.motion_val[list][b_xy + b_stride * 1]);
2045  AV_COPY32(mv_dst - 1 + 16, h->cur_pic.motion_val[list][b_xy + b_stride * 2]);
2046  AV_COPY32(mv_dst - 1 + 24, h->cur_pic.motion_val[list][b_xy + b_stride * 3]);
2047  ref_cache[-1 + 0] =
2048  ref_cache[-1 + 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 2 * 0]];
2049  ref_cache[-1 + 16] =
2050  ref_cache[-1 + 24] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 2 * 1]];
2051  } else {
2052  AV_ZERO32(mv_dst - 1 + 0);
2053  AV_ZERO32(mv_dst - 1 + 8);
2054  AV_ZERO32(mv_dst - 1 + 16);
2055  AV_ZERO32(mv_dst - 1 + 24);
2056  ref_cache[-1 + 0] =
2057  ref_cache[-1 + 8] =
2058  ref_cache[-1 + 16] =
2059  ref_cache[-1 + 24] = LIST_NOT_USED;
2060  }
2061  }
2062  }
2063 
2064  if (!USES_LIST(mb_type, list)) {
2065  fill_rectangle(mv_dst, 4, 4, 8, pack16to32(0, 0), 4);
2066  AV_WN32A(&ref_cache[0 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2067  AV_WN32A(&ref_cache[1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2068  AV_WN32A(&ref_cache[2 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2069  AV_WN32A(&ref_cache[3 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2070  return;
2071  }
2072 
2073  {
2074  int8_t *ref = &h->cur_pic.ref_index[list][4 * mb_xy];
2075  int (*ref2frm)[64] = (void*)(h->ref2frm[h->slice_num & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2));
2076  uint32_t ref01 = (pack16to32(ref2frm[list][ref[0]], ref2frm[list][ref[1]]) & 0x00FF00FF) * 0x0101;
2077  uint32_t ref23 = (pack16to32(ref2frm[list][ref[2]], ref2frm[list][ref[3]]) & 0x00FF00FF) * 0x0101;
2078  AV_WN32A(&ref_cache[0 * 8], ref01);
2079  AV_WN32A(&ref_cache[1 * 8], ref01);
2080  AV_WN32A(&ref_cache[2 * 8], ref23);
2081  AV_WN32A(&ref_cache[3 * 8], ref23);
2082  }
2083 
2084  {
2085  int16_t(*mv_src)[2] = &h->cur_pic.motion_val[list][4 * h->mb_x + 4 * h->mb_y * b_stride];
2086  AV_COPY128(mv_dst + 8 * 0, mv_src + 0 * b_stride);
2087  AV_COPY128(mv_dst + 8 * 1, mv_src + 1 * b_stride);
2088  AV_COPY128(mv_dst + 8 * 2, mv_src + 2 * b_stride);
2089  AV_COPY128(mv_dst + 8 * 3, mv_src + 3 * b_stride);
2090  }
2091 }
2092 
2093 /**
2094  *
2095  * @return non zero if the loop filter can be skipped
2096  */
2097 static int fill_filter_caches(H264Context *h, int mb_type)
2098 {
2099  const int mb_xy = h->mb_xy;
2100  int top_xy, left_xy[LEFT_MBS];
2101  int top_type, left_type[LEFT_MBS];
2102  uint8_t *nnz;
2103  uint8_t *nnz_cache;
2104 
2105  top_xy = mb_xy - (h->mb_stride << MB_FIELD(h));
2106 
2107  /* Wow, what a mess, why didn't they simplify the interlacing & intra
2108  * stuff, I can't imagine that these complex rules are worth it. */
2109 
2110  left_xy[LBOT] = left_xy[LTOP] = mb_xy - 1;
2111  if (FRAME_MBAFF(h)) {
2112  const int left_mb_field_flag = IS_INTERLACED(h->cur_pic.mb_type[mb_xy - 1]);
2113  const int curr_mb_field_flag = IS_INTERLACED(mb_type);
2114  if (h->mb_y & 1) {
2115  if (left_mb_field_flag != curr_mb_field_flag)
2116  left_xy[LTOP] -= h->mb_stride;
2117  } else {
2118  if (curr_mb_field_flag)
2119  top_xy += h->mb_stride &
2120  (((h->cur_pic.mb_type[top_xy] >> 7) & 1) - 1);
2121  if (left_mb_field_flag != curr_mb_field_flag)
2122  left_xy[LBOT] += h->mb_stride;
2123  }
2124  }
2125 
2126  h->top_mb_xy = top_xy;
2127  h->left_mb_xy[LTOP] = left_xy[LTOP];
2128  h->left_mb_xy[LBOT] = left_xy[LBOT];
2129  {
2130  /* For sufficiently low qp, filtering wouldn't do anything.
2131  * This is a conservative estimate: could also check beta_offset
2132  * and more accurate chroma_qp. */
2133  int qp_thresh = h->qp_thresh; // FIXME strictly we should store qp_thresh for each mb of a slice
2134  int qp = h->cur_pic.qscale_table[mb_xy];
2135  if (qp <= qp_thresh &&
2136  (left_xy[LTOP] < 0 ||
2137  ((qp + h->cur_pic.qscale_table[left_xy[LTOP]] + 1) >> 1) <= qp_thresh) &&
2138  (top_xy < 0 ||
2139  ((qp + h->cur_pic.qscale_table[top_xy] + 1) >> 1) <= qp_thresh)) {
2140  if (!FRAME_MBAFF(h))
2141  return 1;
2142  if ((left_xy[LTOP] < 0 ||
2143  ((qp + h->cur_pic.qscale_table[left_xy[LBOT]] + 1) >> 1) <= qp_thresh) &&
2144  (top_xy < h->mb_stride ||
2145  ((qp + h->cur_pic.qscale_table[top_xy - h->mb_stride] + 1) >> 1) <= qp_thresh))
2146  return 1;
2147  }
2148  }
2149 
2150  top_type = h->cur_pic.mb_type[top_xy];
2151  left_type[LTOP] = h->cur_pic.mb_type[left_xy[LTOP]];
2152  left_type[LBOT] = h->cur_pic.mb_type[left_xy[LBOT]];
2153  if (h->deblocking_filter == 2) {
2154  if (h->slice_table[top_xy] != h->slice_num)
2155  top_type = 0;
2156  if (h->slice_table[left_xy[LBOT]] != h->slice_num)
2157  left_type[LTOP] = left_type[LBOT] = 0;
2158  } else {
2159  if (h->slice_table[top_xy] == 0xFFFF)
2160  top_type = 0;
2161  if (h->slice_table[left_xy[LBOT]] == 0xFFFF)
2162  left_type[LTOP] = left_type[LBOT] = 0;
2163  }
2164  h->top_type = top_type;
2165  h->left_type[LTOP] = left_type[LTOP];
2166  h->left_type[LBOT] = left_type[LBOT];
2167 
2168  if (IS_INTRA(mb_type))
2169  return 0;
2170 
2171  fill_filter_caches_inter(h, mb_type, top_xy, left_xy,
2172  top_type, left_type, mb_xy, 0);
2173  if (h->list_count == 2)
2174  fill_filter_caches_inter(h, mb_type, top_xy, left_xy,
2175  top_type, left_type, mb_xy, 1);
2176 
2177  nnz = h->non_zero_count[mb_xy];
2178  nnz_cache = h->non_zero_count_cache;
2179  AV_COPY32(&nnz_cache[4 + 8 * 1], &nnz[0]);
2180  AV_COPY32(&nnz_cache[4 + 8 * 2], &nnz[4]);
2181  AV_COPY32(&nnz_cache[4 + 8 * 3], &nnz[8]);
2182  AV_COPY32(&nnz_cache[4 + 8 * 4], &nnz[12]);
2183  h->cbp = h->cbp_table[mb_xy];
2184 
2185  if (top_type) {
2186  nnz = h->non_zero_count[top_xy];
2187  AV_COPY32(&nnz_cache[4 + 8 * 0], &nnz[3 * 4]);
2188  }
2189 
2190  if (left_type[LTOP]) {
2191  nnz = h->non_zero_count[left_xy[LTOP]];
2192  nnz_cache[3 + 8 * 1] = nnz[3 + 0 * 4];
2193  nnz_cache[3 + 8 * 2] = nnz[3 + 1 * 4];
2194  nnz_cache[3 + 8 * 3] = nnz[3 + 2 * 4];
2195  nnz_cache[3 + 8 * 4] = nnz[3 + 3 * 4];
2196  }
2197 
2198  /* CAVLC 8x8dct requires NNZ values for residual decoding that differ
2199  * from what the loop filter needs */
2200  if (!CABAC(h) && h->pps.transform_8x8_mode) {
2201  if (IS_8x8DCT(top_type)) {
2202  nnz_cache[4 + 8 * 0] =
2203  nnz_cache[5 + 8 * 0] = (h->cbp_table[top_xy] & 0x4000) >> 12;
2204  nnz_cache[6 + 8 * 0] =
2205  nnz_cache[7 + 8 * 0] = (h->cbp_table[top_xy] & 0x8000) >> 12;
2206  }
2207  if (IS_8x8DCT(left_type[LTOP])) {
2208  nnz_cache[3 + 8 * 1] =
2209  nnz_cache[3 + 8 * 2] = (h->cbp_table[left_xy[LTOP]] & 0x2000) >> 12; // FIXME check MBAFF
2210  }
2211  if (IS_8x8DCT(left_type[LBOT])) {
2212  nnz_cache[3 + 8 * 3] =
2213  nnz_cache[3 + 8 * 4] = (h->cbp_table[left_xy[LBOT]] & 0x8000) >> 12; // FIXME check MBAFF
2214  }
2215 
2216  if (IS_8x8DCT(mb_type)) {
2217  nnz_cache[scan8[0]] =
2218  nnz_cache[scan8[1]] =
2219  nnz_cache[scan8[2]] =
2220  nnz_cache[scan8[3]] = (h->cbp & 0x1000) >> 12;
2221 
2222  nnz_cache[scan8[0 + 4]] =
2223  nnz_cache[scan8[1 + 4]] =
2224  nnz_cache[scan8[2 + 4]] =
2225  nnz_cache[scan8[3 + 4]] = (h->cbp & 0x2000) >> 12;
2226 
2227  nnz_cache[scan8[0 + 8]] =
2228  nnz_cache[scan8[1 + 8]] =
2229  nnz_cache[scan8[2 + 8]] =
2230  nnz_cache[scan8[3 + 8]] = (h->cbp & 0x4000) >> 12;
2231 
2232  nnz_cache[scan8[0 + 12]] =
2233  nnz_cache[scan8[1 + 12]] =
2234  nnz_cache[scan8[2 + 12]] =
2235  nnz_cache[scan8[3 + 12]] = (h->cbp & 0x8000) >> 12;
2236  }
2237  }
2238 
2239  return 0;
2240 }
2241 
2242 static void loop_filter(H264Context *h, int start_x, int end_x)
2243 {
2244  uint8_t *dest_y, *dest_cb, *dest_cr;
2245  int linesize, uvlinesize, mb_x, mb_y;
2246  const int end_mb_y = h->mb_y + FRAME_MBAFF(h);
2247  const int old_slice_type = h->slice_type;
2248  const int pixel_shift = h->pixel_shift;
2249  const int block_h = 16 >> h->chroma_y_shift;
2250 
2251  if (h->deblocking_filter) {
2252  for (mb_x = start_x; mb_x < end_x; mb_x++)
2253  for (mb_y = end_mb_y - FRAME_MBAFF(h); mb_y <= end_mb_y; mb_y++) {
2254  int mb_xy, mb_type;
2255  mb_xy = h->mb_xy = mb_x + mb_y * h->mb_stride;
2256  h->slice_num = h->slice_table[mb_xy];
2257  mb_type = h->cur_pic.mb_type[mb_xy];
2258  h->list_count = h->list_counts[mb_xy];
2259 
2260  if (FRAME_MBAFF(h))
2261  h->mb_mbaff =
2262  h->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
2263 
2264  h->mb_x = mb_x;
2265  h->mb_y = mb_y;
2266  dest_y = h->cur_pic.f.data[0] +
2267  ((mb_x << pixel_shift) + mb_y * h->linesize) * 16;
2268  dest_cb = h->cur_pic.f.data[1] +
2269  (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
2270  mb_y * h->uvlinesize * block_h;
2271  dest_cr = h->cur_pic.f.data[2] +
2272  (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
2273  mb_y * h->uvlinesize * block_h;
2274  // FIXME simplify above
2275 
2276  if (MB_FIELD(h)) {
2277  linesize = h->mb_linesize = h->linesize * 2;
2278  uvlinesize = h->mb_uvlinesize = h->uvlinesize * 2;
2279  if (mb_y & 1) { // FIXME move out of this function?
2280  dest_y -= h->linesize * 15;
2281  dest_cb -= h->uvlinesize * (block_h - 1);
2282  dest_cr -= h->uvlinesize * (block_h - 1);
2283  }
2284  } else {
2285  linesize = h->mb_linesize = h->linesize;
2286  uvlinesize = h->mb_uvlinesize = h->uvlinesize;
2287  }
2288  backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize,
2289  uvlinesize, 0);
2290  if (fill_filter_caches(h, mb_type))
2291  continue;
2292  h->chroma_qp[0] = get_chroma_qp(h, 0, h->cur_pic.qscale_table[mb_xy]);
2293  h->chroma_qp[1] = get_chroma_qp(h, 1, h->cur_pic.qscale_table[mb_xy]);
2294 
2295  if (FRAME_MBAFF(h)) {
2296  ff_h264_filter_mb(h, mb_x, mb_y, dest_y, dest_cb, dest_cr,
2297  linesize, uvlinesize);
2298  } else {
2299  ff_h264_filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb,
2300  dest_cr, linesize, uvlinesize);
2301  }
2302  }
2303  }
2304  h->slice_type = old_slice_type;
2305  h->mb_x = end_x;
2306  h->mb_y = end_mb_y - FRAME_MBAFF(h);
2307  h->chroma_qp[0] = get_chroma_qp(h, 0, h->qscale);
2308  h->chroma_qp[1] = get_chroma_qp(h, 1, h->qscale);
2309 }
2310 
2311 static void predict_field_decoding_flag(H264Context *h)
2312 {
2313  const int mb_xy = h->mb_x + h->mb_y * h->mb_stride;
2314  int mb_type = (h->slice_table[mb_xy - 1] == h->slice_num) ?
2315  h->cur_pic.mb_type[mb_xy - 1] :
2316  (h->slice_table[mb_xy - h->mb_stride] == h->slice_num) ?
2317  h->cur_pic.mb_type[mb_xy - h->mb_stride] : 0;
2318  h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
2319 }
2320 
2321 /**
2322  * Draw edges and report progress for the last MB row.
2323  */
2324 static void decode_finish_row(H264Context *h)
2325 {
2326  int top = 16 * (h->mb_y >> FIELD_PICTURE(h));
2327  int pic_height = 16 * h->mb_height >> FIELD_PICTURE(h);
2328  int height = 16 << FRAME_MBAFF(h);
2329  int deblock_border = (16 + 4) << FRAME_MBAFF(h);
2330 
2331  if (h->deblocking_filter) {
2332  if ((top + height) >= pic_height)
2333  height += deblock_border;
2334  top -= deblock_border;
2335  }
2336 
2337  if (top >= pic_height || (top + height) < 0)
2338  return;
2339 
2340  height = FFMIN(height, pic_height - top);
2341  if (top < 0) {
2342  height = top + height;
2343  top = 0;
2344  }
2345 
2346  ff_h264_draw_horiz_band(h, top, height);
2347 
2348  if (h->droppable || h->er.error_occurred)
2349  return;
2350 
2351  ff_thread_report_progress(&h->cur_pic_ptr->tf, top + height - 1,
2352  h->picture_structure == PICT_BOTTOM_FIELD);
2353 }
2354 
2355 static void er_add_slice(H264Context *h, int startx, int starty,
2356  int endx, int endy, int status)
2357 {
2358  if (CONFIG_ERROR_RESILIENCE) {
2359  ERContext *er = &h->er;
2360 
2361  ff_er_add_slice(er, startx, starty, endx, endy, status);
2362  }
2363 }
2364 
2365 static int decode_slice(struct AVCodecContext *avctx, void *arg)
2366 {
2367  H264Context *h = *(void **)arg;
2368  int lf_x_start = h->mb_x;
2369 
2370  h->mb_skip_run = -1;
2371 
2372  av_assert0(h->block_offset[15] == (4 * ((scan8[15] - scan8[0]) & 7) << h->pixel_shift) + 4 * h->linesize * ((scan8[15] - scan8[0]) >> 3));
2373 
2374  h->is_complex = FRAME_MBAFF(h) || h->picture_structure != PICT_FRAME ||
2375  avctx->codec_id != AV_CODEC_ID_H264 ||
2376  (CONFIG_GRAY && (h->flags & CODEC_FLAG_GRAY));
2377 
2378  if (!(h->avctx->active_thread_type & FF_THREAD_SLICE) && h->picture_structure == PICT_FRAME && h->er.error_status_table) {
2379  const int start_i = av_clip(h->resync_mb_x + h->resync_mb_y * h->mb_width, 0, h->mb_num - 1);
2380  if (start_i) {
2381  int prev_status = h->er.error_status_table[h->er.mb_index2xy[start_i - 1]];
2382  prev_status &= ~ VP_START;
2383  if (prev_status != (ER_MV_END | ER_DC_END | ER_AC_END))
2384  h->er.error_occurred = 1;
2385  }
2386  }
2387 
2388  if (h->pps.cabac) {
2389  /* realign */
2390  align_get_bits(&h->gb);
2391 
2392  /* init cabac */
2393  ff_init_cabac_decoder(&h->cabac,
2394  h->gb.buffer + get_bits_count(&h->gb) / 8,
2395  (get_bits_left(&h->gb) + 7) / 8);
2396 
2397  ff_h264_init_cabac_states(h);
2398 
2399  for (;;) {
2400  // START_TIMER
2401  int ret = ff_h264_decode_mb_cabac(h);
2402  int eos;
2403  // STOP_TIMER("decode_mb_cabac")
2404 
2405  if (ret >= 0)
2406  ff_h264_hl_decode_mb(h);
2407 
2408  // FIXME optimal? or let mb_decode decode 16x32 ?
2409  if (ret >= 0 && FRAME_MBAFF(h)) {
2410  h->mb_y++;
2411 
2412  ret = ff_h264_decode_mb_cabac(h);
2413 
2414  if (ret >= 0)
2415  ff_h264_hl_decode_mb(h);
2416  h->mb_y--;
2417  }
2418  eos = get_cabac_terminate(&h->cabac);
2419 
2420  if ((h->workaround_bugs & FF_BUG_TRUNCATED) &&
2421  h->cabac.bytestream > h->cabac.bytestream_end + 2) {
2422  er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x - 1,
2423  h->mb_y, ER_MB_END);
2424  if (h->mb_x >= lf_x_start)
2425  loop_filter(h, lf_x_start, h->mb_x + 1);
2426  return 0;
2427  }
2428  if (h->cabac.bytestream > h->cabac.bytestream_end + 2 )
2429  av_log(h->avctx, AV_LOG_DEBUG, "bytestream overread %"PTRDIFF_SPECIFIER"\n", h->cabac.bytestream_end - h->cabac.bytestream);
2430  if (ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 4) {
2431  av_log(h->avctx, AV_LOG_ERROR,
2432  "error while decoding MB %d %d, bytestream %"PTRDIFF_SPECIFIER"\n",
2433  h->mb_x, h->mb_y,
2434  h->cabac.bytestream_end - h->cabac.bytestream);
2435  er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x,
2436  h->mb_y, ER_MB_ERROR);
2437  return AVERROR_INVALIDDATA;
2438  }
2439 
2440  if (++h->mb_x >= h->mb_width) {
2441  loop_filter(h, lf_x_start, h->mb_x);
2442  h->mb_x = lf_x_start = 0;
2443  decode_finish_row(h);
2444  ++h->mb_y;
2445  if (FIELD_OR_MBAFF_PICTURE(h)) {
2446  ++h->mb_y;
2447  if (FRAME_MBAFF(h) && h->mb_y < h->mb_height)
2448  predict_field_decoding_flag(h);
2449  }
2450  }
2451 
2452  if (eos || h->mb_y >= h->mb_height) {
2453  tprintf(h->avctx, "slice end %d %d\n",
2454  get_bits_count(&h->gb), h->gb.size_in_bits);
2455  er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x - 1,
2456  h->mb_y, ER_MB_END);
2457  if (h->mb_x > lf_x_start)
2458  loop_filter(h, lf_x_start, h->mb_x);
2459  return 0;
2460  }
2461  }
2462  } else {
2463  for (;;) {
2464  int ret = ff_h264_decode_mb_cavlc(h);
2465 
2466  if (ret >= 0)
2467  ff_h264_hl_decode_mb(h);
2468 
2469  // FIXME optimal? or let mb_decode decode 16x32 ?
2470  if (ret >= 0 && FRAME_MBAFF(h)) {
2471  h->mb_y++;
2472  ret = ff_h264_decode_mb_cavlc(h);
2473 
2474  if (ret >= 0)
2475  ff_h264_hl_decode_mb(h);
2476  h->mb_y--;
2477  }
2478 
2479  if (ret < 0) {
2480  av_log(h->avctx, AV_LOG_ERROR,
2481  "error while decoding MB %d %d\n", h->mb_x, h->mb_y);
2482  er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x,
2483  h->mb_y, ER_MB_ERROR);
2484  return ret;
2485  }
2486 
2487  if (++h->mb_x >= h->mb_width) {
2488  loop_filter(h, lf_x_start, h->mb_x);
2489  h->mb_x = lf_x_start = 0;
2490  decode_finish_row(h);
2491  ++h->mb_y;
2492  if (FIELD_OR_MBAFF_PICTURE(h)) {
2493  ++h->mb_y;
2494  if (FRAME_MBAFF(h) && h->mb_y < h->mb_height)
2495  predict_field_decoding_flag(h);
2496  }
2497  if (h->mb_y >= h->mb_height) {
2498  tprintf(h->avctx, "slice end %d %d\n",
2499  get_bits_count(&h->gb), h->gb.size_in_bits);
2500 
2501  if ( get_bits_left(&h->gb) == 0
2502  || get_bits_left(&h->gb) > 0 && !(h->avctx->err_recognition & AV_EF_AGGRESSIVE)) {
2503  er_add_slice(h, h->resync_mb_x, h->resync_mb_y,
2504  h->mb_x - 1, h->mb_y, ER_MB_END);
2505 
2506  return 0;
2507  } else {
2508  er_add_slice(h, h->resync_mb_x, h->resync_mb_y,
2509  h->mb_x, h->mb_y, ER_MB_END);
2510 
2511  return AVERROR_INVALIDDATA;
2512  }
2513  }
2514  }
2515 
2516  if (get_bits_left(&h->gb) <= 0 && h->mb_skip_run <= 0) {
2517  tprintf(h->avctx, "slice end %d %d\n",
2518  get_bits_count(&h->gb), h->gb.size_in_bits);
2519 
2520  if (get_bits_left(&h->gb) == 0) {
2521  er_add_slice(h, h->resync_mb_x, h->resync_mb_y,
2522  h->mb_x - 1, h->mb_y, ER_MB_END);
2523  if (h->mb_x > lf_x_start)
2524  loop_filter(h, lf_x_start, h->mb_x);
2525 
2526  return 0;
2527  } else {
2528  er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x,
2529  h->mb_y, ER_MB_ERROR);
2530 
2531  return AVERROR_INVALIDDATA;
2532  }
2533  }
2534  }
2535  }
2536 }
2537 
2538 /**
2539  * Call decode_slice() for each context.
2540  *
2541  * @param h h264 master context
2542  * @param context_count number of contexts to execute
2543  */
2544 int ff_h264_execute_decode_slices(H264Context *h, unsigned context_count)
2545 {
2546  AVCodecContext *const avctx = h->avctx;
2547  H264Context *hx;
2548  int i;
2549 
2550  av_assert0(h->mb_y < h->mb_height);
2551 
2552  if (h->avctx->hwaccel ||
2553  h->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU)
2554  return 0;
2555  if (context_count == 1) {
2556  return decode_slice(avctx, &h);
2557  } else {
2558  av_assert0(context_count > 0);
2559  for (i = 1; i < context_count; i++) {
2560  hx = h->thread_context[i];
2561  if (CONFIG_ERROR_RESILIENCE) {
2562  hx->er.error_count = 0;
2563  }
2564  hx->x264_build = h->x264_build;
2565  }
2566 
2567  avctx->execute(avctx, decode_slice, h->thread_context,
2568  NULL, context_count, sizeof(void *));
2569 
2570  /* pull back stuff from slices to master context */
2571  hx = h->thread_context[context_count - 1];
2572  h->mb_x = hx->mb_x;
2573  h->mb_y = hx->mb_y;
2574  h->droppable = hx->droppable;
2575  h->picture_structure = hx->picture_structure;
2576  if (CONFIG_ERROR_RESILIENCE) {
2577  for (i = 1; i < context_count; i++)
2578  h->er.error_count += h->thread_context[i]->er.error_count;
2579  }
2580  }
2581 
2582  return 0;
2583 }
Generated on Sun Sep 14 2014 18:55:56 for FFmpeg by   doxygen 1.8.2