You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

2785 lines
104KB

  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. * @file
  23. * H.264 / AVC / MPEG-4 part10 codec.
  24. * @author Michael Niedermayer <michaelni@gmx.at>
  25. */
  26. #include "libavutil/avassert.h"
  27. #include "libavutil/display.h"
  28. #include "libavutil/imgutils.h"
  29. #include "libavutil/stereo3d.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 "h264dec.h"
  38. #include "h264data.h"
  39. #include "h264chroma.h"
  40. #include "h264_mvpred.h"
  41. #include "h264_ps.h"
  42. #include "golomb.h"
  43. #include "mathops.h"
  44. #include "mpegutils.h"
  45. #include "mpegvideo.h"
  46. #include "rectangle.h"
  47. #include "thread.h"
  48. static const uint8_t field_scan[16+1] = {
  49. 0 + 0 * 4, 0 + 1 * 4, 1 + 0 * 4, 0 + 2 * 4,
  50. 0 + 3 * 4, 1 + 1 * 4, 1 + 2 * 4, 1 + 3 * 4,
  51. 2 + 0 * 4, 2 + 1 * 4, 2 + 2 * 4, 2 + 3 * 4,
  52. 3 + 0 * 4, 3 + 1 * 4, 3 + 2 * 4, 3 + 3 * 4,
  53. };
  54. static const uint8_t field_scan8x8[64+1] = {
  55. 0 + 0 * 8, 0 + 1 * 8, 0 + 2 * 8, 1 + 0 * 8,
  56. 1 + 1 * 8, 0 + 3 * 8, 0 + 4 * 8, 1 + 2 * 8,
  57. 2 + 0 * 8, 1 + 3 * 8, 0 + 5 * 8, 0 + 6 * 8,
  58. 0 + 7 * 8, 1 + 4 * 8, 2 + 1 * 8, 3 + 0 * 8,
  59. 2 + 2 * 8, 1 + 5 * 8, 1 + 6 * 8, 1 + 7 * 8,
  60. 2 + 3 * 8, 3 + 1 * 8, 4 + 0 * 8, 3 + 2 * 8,
  61. 2 + 4 * 8, 2 + 5 * 8, 2 + 6 * 8, 2 + 7 * 8,
  62. 3 + 3 * 8, 4 + 1 * 8, 5 + 0 * 8, 4 + 2 * 8,
  63. 3 + 4 * 8, 3 + 5 * 8, 3 + 6 * 8, 3 + 7 * 8,
  64. 4 + 3 * 8, 5 + 1 * 8, 6 + 0 * 8, 5 + 2 * 8,
  65. 4 + 4 * 8, 4 + 5 * 8, 4 + 6 * 8, 4 + 7 * 8,
  66. 5 + 3 * 8, 6 + 1 * 8, 6 + 2 * 8, 5 + 4 * 8,
  67. 5 + 5 * 8, 5 + 6 * 8, 5 + 7 * 8, 6 + 3 * 8,
  68. 7 + 0 * 8, 7 + 1 * 8, 6 + 4 * 8, 6 + 5 * 8,
  69. 6 + 6 * 8, 6 + 7 * 8, 7 + 2 * 8, 7 + 3 * 8,
  70. 7 + 4 * 8, 7 + 5 * 8, 7 + 6 * 8, 7 + 7 * 8,
  71. };
  72. static const uint8_t field_scan8x8_cavlc[64+1] = {
  73. 0 + 0 * 8, 1 + 1 * 8, 2 + 0 * 8, 0 + 7 * 8,
  74. 2 + 2 * 8, 2 + 3 * 8, 2 + 4 * 8, 3 + 3 * 8,
  75. 3 + 4 * 8, 4 + 3 * 8, 4 + 4 * 8, 5 + 3 * 8,
  76. 5 + 5 * 8, 7 + 0 * 8, 6 + 6 * 8, 7 + 4 * 8,
  77. 0 + 1 * 8, 0 + 3 * 8, 1 + 3 * 8, 1 + 4 * 8,
  78. 1 + 5 * 8, 3 + 1 * 8, 2 + 5 * 8, 4 + 1 * 8,
  79. 3 + 5 * 8, 5 + 1 * 8, 4 + 5 * 8, 6 + 1 * 8,
  80. 5 + 6 * 8, 7 + 1 * 8, 6 + 7 * 8, 7 + 5 * 8,
  81. 0 + 2 * 8, 0 + 4 * 8, 0 + 5 * 8, 2 + 1 * 8,
  82. 1 + 6 * 8, 4 + 0 * 8, 2 + 6 * 8, 5 + 0 * 8,
  83. 3 + 6 * 8, 6 + 0 * 8, 4 + 6 * 8, 6 + 2 * 8,
  84. 5 + 7 * 8, 6 + 4 * 8, 7 + 2 * 8, 7 + 6 * 8,
  85. 1 + 0 * 8, 1 + 2 * 8, 0 + 6 * 8, 3 + 0 * 8,
  86. 1 + 7 * 8, 3 + 2 * 8, 2 + 7 * 8, 4 + 2 * 8,
  87. 3 + 7 * 8, 5 + 2 * 8, 4 + 7 * 8, 5 + 4 * 8,
  88. 6 + 3 * 8, 6 + 5 * 8, 7 + 3 * 8, 7 + 7 * 8,
  89. };
  90. // zigzag_scan8x8_cavlc[i] = zigzag_scan8x8[(i/4) + 16*(i%4)]
  91. static const uint8_t zigzag_scan8x8_cavlc[64+1] = {
  92. 0 + 0 * 8, 1 + 1 * 8, 1 + 2 * 8, 2 + 2 * 8,
  93. 4 + 1 * 8, 0 + 5 * 8, 3 + 3 * 8, 7 + 0 * 8,
  94. 3 + 4 * 8, 1 + 7 * 8, 5 + 3 * 8, 6 + 3 * 8,
  95. 2 + 7 * 8, 6 + 4 * 8, 5 + 6 * 8, 7 + 5 * 8,
  96. 1 + 0 * 8, 2 + 0 * 8, 0 + 3 * 8, 3 + 1 * 8,
  97. 3 + 2 * 8, 0 + 6 * 8, 4 + 2 * 8, 6 + 1 * 8,
  98. 2 + 5 * 8, 2 + 6 * 8, 6 + 2 * 8, 5 + 4 * 8,
  99. 3 + 7 * 8, 7 + 3 * 8, 4 + 7 * 8, 7 + 6 * 8,
  100. 0 + 1 * 8, 3 + 0 * 8, 0 + 4 * 8, 4 + 0 * 8,
  101. 2 + 3 * 8, 1 + 5 * 8, 5 + 1 * 8, 5 + 2 * 8,
  102. 1 + 6 * 8, 3 + 5 * 8, 7 + 1 * 8, 4 + 5 * 8,
  103. 4 + 6 * 8, 7 + 4 * 8, 5 + 7 * 8, 6 + 7 * 8,
  104. 0 + 2 * 8, 2 + 1 * 8, 1 + 3 * 8, 5 + 0 * 8,
  105. 1 + 4 * 8, 2 + 4 * 8, 6 + 0 * 8, 4 + 3 * 8,
  106. 0 + 7 * 8, 4 + 4 * 8, 7 + 2 * 8, 3 + 6 * 8,
  107. 5 + 5 * 8, 6 + 5 * 8, 6 + 6 * 8, 7 + 7 * 8,
  108. };
  109. static void release_unused_pictures(H264Context *h, int remove_current)
  110. {
  111. int i;
  112. /* release non reference frames */
  113. for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
  114. if (h->DPB[i].f->buf[0] && !h->DPB[i].reference &&
  115. (remove_current || &h->DPB[i] != h->cur_pic_ptr)) {
  116. ff_h264_unref_picture(h, &h->DPB[i]);
  117. }
  118. }
  119. }
  120. static int alloc_scratch_buffers(H264SliceContext *sl, int linesize)
  121. {
  122. const H264Context *h = sl->h264;
  123. int alloc_size = FFALIGN(FFABS(linesize) + 32, 32);
  124. av_fast_malloc(&sl->bipred_scratchpad, &sl->bipred_scratchpad_allocated, 16 * 6 * alloc_size);
  125. // edge emu needs blocksize + filter length - 1
  126. // (= 21x21 for H.264)
  127. av_fast_malloc(&sl->edge_emu_buffer, &sl->edge_emu_buffer_allocated, alloc_size * 2 * 21);
  128. av_fast_mallocz(&sl->top_borders[0], &sl->top_borders_allocated[0],
  129. h->mb_width * 16 * 3 * sizeof(uint8_t) * 2);
  130. av_fast_mallocz(&sl->top_borders[1], &sl->top_borders_allocated[1],
  131. h->mb_width * 16 * 3 * sizeof(uint8_t) * 2);
  132. if (!sl->bipred_scratchpad || !sl->edge_emu_buffer ||
  133. !sl->top_borders[0] || !sl->top_borders[1]) {
  134. av_freep(&sl->bipred_scratchpad);
  135. av_freep(&sl->edge_emu_buffer);
  136. av_freep(&sl->top_borders[0]);
  137. av_freep(&sl->top_borders[1]);
  138. sl->bipred_scratchpad_allocated = 0;
  139. sl->edge_emu_buffer_allocated = 0;
  140. sl->top_borders_allocated[0] = 0;
  141. sl->top_borders_allocated[1] = 0;
  142. return AVERROR(ENOMEM);
  143. }
  144. return 0;
  145. }
  146. static int init_table_pools(H264Context *h)
  147. {
  148. const int big_mb_num = h->mb_stride * (h->mb_height + 1) + 1;
  149. const int mb_array_size = h->mb_stride * h->mb_height;
  150. const int b4_stride = h->mb_width * 4 + 1;
  151. const int b4_array_size = b4_stride * h->mb_height * 4;
  152. h->qscale_table_pool = av_buffer_pool_init(big_mb_num + h->mb_stride,
  153. av_buffer_allocz);
  154. h->mb_type_pool = av_buffer_pool_init((big_mb_num + h->mb_stride) *
  155. sizeof(uint32_t), av_buffer_allocz);
  156. h->motion_val_pool = av_buffer_pool_init(2 * (b4_array_size + 4) *
  157. sizeof(int16_t), av_buffer_allocz);
  158. h->ref_index_pool = av_buffer_pool_init(4 * mb_array_size, av_buffer_allocz);
  159. if (!h->qscale_table_pool || !h->mb_type_pool || !h->motion_val_pool ||
  160. !h->ref_index_pool) {
  161. av_buffer_pool_uninit(&h->qscale_table_pool);
  162. av_buffer_pool_uninit(&h->mb_type_pool);
  163. av_buffer_pool_uninit(&h->motion_val_pool);
  164. av_buffer_pool_uninit(&h->ref_index_pool);
  165. return AVERROR(ENOMEM);
  166. }
  167. return 0;
  168. }
  169. static int alloc_picture(H264Context *h, H264Picture *pic)
  170. {
  171. int i, ret = 0;
  172. av_assert0(!pic->f->data[0]);
  173. pic->tf.f = pic->f;
  174. ret = ff_thread_get_buffer(h->avctx, &pic->tf, pic->reference ?
  175. AV_GET_BUFFER_FLAG_REF : 0);
  176. if (ret < 0)
  177. goto fail;
  178. pic->crop = h->ps.sps->crop;
  179. pic->crop_top = h->ps.sps->crop_top;
  180. pic->crop_left= h->ps.sps->crop_left;
  181. if (h->avctx->hwaccel) {
  182. const AVHWAccel *hwaccel = h->avctx->hwaccel;
  183. av_assert0(!pic->hwaccel_picture_private);
  184. if (hwaccel->frame_priv_data_size) {
  185. pic->hwaccel_priv_buf = av_buffer_allocz(hwaccel->frame_priv_data_size);
  186. if (!pic->hwaccel_priv_buf)
  187. return AVERROR(ENOMEM);
  188. pic->hwaccel_picture_private = pic->hwaccel_priv_buf->data;
  189. }
  190. }
  191. if (CONFIG_GRAY && !h->avctx->hwaccel && h->flags & AV_CODEC_FLAG_GRAY && pic->f->data[2]) {
  192. int h_chroma_shift, v_chroma_shift;
  193. av_pix_fmt_get_chroma_sub_sample(pic->f->format,
  194. &h_chroma_shift, &v_chroma_shift);
  195. for(i=0; i<AV_CEIL_RSHIFT(pic->f->height, v_chroma_shift); i++) {
  196. memset(pic->f->data[1] + pic->f->linesize[1]*i,
  197. 0x80, AV_CEIL_RSHIFT(pic->f->width, h_chroma_shift));
  198. memset(pic->f->data[2] + pic->f->linesize[2]*i,
  199. 0x80, AV_CEIL_RSHIFT(pic->f->width, h_chroma_shift));
  200. }
  201. }
  202. if (!h->qscale_table_pool) {
  203. ret = init_table_pools(h);
  204. if (ret < 0)
  205. goto fail;
  206. }
  207. pic->qscale_table_buf = av_buffer_pool_get(h->qscale_table_pool);
  208. pic->mb_type_buf = av_buffer_pool_get(h->mb_type_pool);
  209. if (!pic->qscale_table_buf || !pic->mb_type_buf)
  210. goto fail;
  211. pic->mb_type = (uint32_t*)pic->mb_type_buf->data + 2 * h->mb_stride + 1;
  212. pic->qscale_table = pic->qscale_table_buf->data + 2 * h->mb_stride + 1;
  213. for (i = 0; i < 2; i++) {
  214. pic->motion_val_buf[i] = av_buffer_pool_get(h->motion_val_pool);
  215. pic->ref_index_buf[i] = av_buffer_pool_get(h->ref_index_pool);
  216. if (!pic->motion_val_buf[i] || !pic->ref_index_buf[i])
  217. goto fail;
  218. pic->motion_val[i] = (int16_t (*)[2])pic->motion_val_buf[i]->data + 4;
  219. pic->ref_index[i] = pic->ref_index_buf[i]->data;
  220. }
  221. return 0;
  222. fail:
  223. ff_h264_unref_picture(h, pic);
  224. return (ret < 0) ? ret : AVERROR(ENOMEM);
  225. }
  226. static int find_unused_picture(H264Context *h)
  227. {
  228. int i;
  229. for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
  230. if (!h->DPB[i].f->buf[0])
  231. return i;
  232. }
  233. return AVERROR_INVALIDDATA;
  234. }
  235. #define IN_RANGE(a, b, size) (((void*)(a) >= (void*)(b)) && ((void*)(a) < (void*)((b) + (size))))
  236. #define REBASE_PICTURE(pic, new_ctx, old_ctx) \
  237. (((pic) && (pic) >= (old_ctx)->DPB && \
  238. (pic) < (old_ctx)->DPB + H264_MAX_PICTURE_COUNT) ? \
  239. &(new_ctx)->DPB[(pic) - (old_ctx)->DPB] : NULL)
  240. static void copy_picture_range(H264Picture **to, H264Picture **from, int count,
  241. H264Context *new_base,
  242. H264Context *old_base)
  243. {
  244. int i;
  245. for (i = 0; i < count; i++) {
  246. av_assert1(!from[i] ||
  247. IN_RANGE(from[i], old_base, 1) ||
  248. IN_RANGE(from[i], old_base->DPB, H264_MAX_PICTURE_COUNT));
  249. to[i] = REBASE_PICTURE(from[i], new_base, old_base);
  250. }
  251. }
  252. static int h264_slice_header_init(H264Context *h);
  253. int ff_h264_update_thread_context(AVCodecContext *dst,
  254. const AVCodecContext *src)
  255. {
  256. H264Context *h = dst->priv_data, *h1 = src->priv_data;
  257. int inited = h->context_initialized, err = 0;
  258. int need_reinit = 0;
  259. int i, ret;
  260. if (dst == src)
  261. return 0;
  262. // We can't fail if SPS isn't set at it breaks current skip_frame code
  263. //if (!h1->ps.sps)
  264. // return AVERROR_INVALIDDATA;
  265. if (inited &&
  266. (h->width != h1->width ||
  267. h->height != h1->height ||
  268. h->mb_width != h1->mb_width ||
  269. h->mb_height != h1->mb_height ||
  270. !h->ps.sps ||
  271. h->ps.sps->bit_depth_luma != h1->ps.sps->bit_depth_luma ||
  272. h->ps.sps->chroma_format_idc != h1->ps.sps->chroma_format_idc ||
  273. h->ps.sps->colorspace != h1->ps.sps->colorspace)) {
  274. need_reinit = 1;
  275. }
  276. /* copy block_offset since frame_start may not be called */
  277. memcpy(h->block_offset, h1->block_offset, sizeof(h->block_offset));
  278. // SPS/PPS
  279. for (i = 0; i < FF_ARRAY_ELEMS(h->ps.sps_list); i++) {
  280. av_buffer_unref(&h->ps.sps_list[i]);
  281. if (h1->ps.sps_list[i]) {
  282. h->ps.sps_list[i] = av_buffer_ref(h1->ps.sps_list[i]);
  283. if (!h->ps.sps_list[i])
  284. return AVERROR(ENOMEM);
  285. }
  286. }
  287. for (i = 0; i < FF_ARRAY_ELEMS(h->ps.pps_list); i++) {
  288. av_buffer_unref(&h->ps.pps_list[i]);
  289. if (h1->ps.pps_list[i]) {
  290. h->ps.pps_list[i] = av_buffer_ref(h1->ps.pps_list[i]);
  291. if (!h->ps.pps_list[i])
  292. return AVERROR(ENOMEM);
  293. }
  294. }
  295. av_buffer_unref(&h->ps.pps_ref);
  296. av_buffer_unref(&h->ps.sps_ref);
  297. h->ps.pps = NULL;
  298. h->ps.sps = NULL;
  299. if (h1->ps.pps_ref) {
  300. h->ps.pps_ref = av_buffer_ref(h1->ps.pps_ref);
  301. if (!h->ps.pps_ref)
  302. return AVERROR(ENOMEM);
  303. h->ps.pps = (const PPS*)h->ps.pps_ref->data;
  304. }
  305. if (h1->ps.sps_ref) {
  306. h->ps.sps_ref = av_buffer_ref(h1->ps.sps_ref);
  307. if (!h->ps.sps_ref)
  308. return AVERROR(ENOMEM);
  309. h->ps.sps = (const SPS*)h->ps.sps_ref->data;
  310. }
  311. if (need_reinit || !inited) {
  312. h->width = h1->width;
  313. h->height = h1->height;
  314. h->mb_height = h1->mb_height;
  315. h->mb_width = h1->mb_width;
  316. h->mb_num = h1->mb_num;
  317. h->mb_stride = h1->mb_stride;
  318. h->b_stride = h1->b_stride;
  319. if (h->context_initialized || h1->context_initialized) {
  320. if ((err = h264_slice_header_init(h)) < 0) {
  321. av_log(h->avctx, AV_LOG_ERROR, "h264_slice_header_init() failed");
  322. return err;
  323. }
  324. }
  325. /* copy block_offset since frame_start may not be called */
  326. memcpy(h->block_offset, h1->block_offset, sizeof(h->block_offset));
  327. }
  328. h->avctx->coded_height = h1->avctx->coded_height;
  329. h->avctx->coded_width = h1->avctx->coded_width;
  330. h->avctx->width = h1->avctx->width;
  331. h->avctx->height = h1->avctx->height;
  332. h->coded_picture_number = h1->coded_picture_number;
  333. h->first_field = h1->first_field;
  334. h->picture_structure = h1->picture_structure;
  335. h->mb_aff_frame = h1->mb_aff_frame;
  336. h->droppable = h1->droppable;
  337. for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
  338. ff_h264_unref_picture(h, &h->DPB[i]);
  339. if (h1->DPB[i].f->buf[0] &&
  340. (ret = ff_h264_ref_picture(h, &h->DPB[i], &h1->DPB[i])) < 0)
  341. return ret;
  342. }
  343. h->cur_pic_ptr = REBASE_PICTURE(h1->cur_pic_ptr, h, h1);
  344. ff_h264_unref_picture(h, &h->cur_pic);
  345. if (h1->cur_pic.f->buf[0]) {
  346. ret = ff_h264_ref_picture(h, &h->cur_pic, &h1->cur_pic);
  347. if (ret < 0)
  348. return ret;
  349. }
  350. h->enable_er = h1->enable_er;
  351. h->workaround_bugs = h1->workaround_bugs;
  352. h->droppable = h1->droppable;
  353. // extradata/NAL handling
  354. h->is_avc = h1->is_avc;
  355. h->nal_length_size = h1->nal_length_size;
  356. h->sei.unregistered.x264_build = h1->sei.unregistered.x264_build;
  357. memcpy(&h->poc, &h1->poc, sizeof(h->poc));
  358. memcpy(h->default_ref, h1->default_ref, sizeof(h->default_ref));
  359. memcpy(h->short_ref, h1->short_ref, sizeof(h->short_ref));
  360. memcpy(h->long_ref, h1->long_ref, sizeof(h->long_ref));
  361. memcpy(h->delayed_pic, h1->delayed_pic, sizeof(h->delayed_pic));
  362. memcpy(h->last_pocs, h1->last_pocs, sizeof(h->last_pocs));
  363. h->next_output_pic = h1->next_output_pic;
  364. h->next_outputed_poc = h1->next_outputed_poc;
  365. memcpy(h->mmco, h1->mmco, sizeof(h->mmco));
  366. h->nb_mmco = h1->nb_mmco;
  367. h->mmco_reset = h1->mmco_reset;
  368. h->explicit_ref_marking = h1->explicit_ref_marking;
  369. h->long_ref_count = h1->long_ref_count;
  370. h->short_ref_count = h1->short_ref_count;
  371. copy_picture_range(h->short_ref, h1->short_ref, 32, h, h1);
  372. copy_picture_range(h->long_ref, h1->long_ref, 32, h, h1);
  373. copy_picture_range(h->delayed_pic, h1->delayed_pic,
  374. MAX_DELAYED_PIC_COUNT + 2, h, h1);
  375. h->frame_recovered = h1->frame_recovered;
  376. if (!h->cur_pic_ptr)
  377. return 0;
  378. if (!h->droppable) {
  379. err = ff_h264_execute_ref_pic_marking(h);
  380. h->poc.prev_poc_msb = h->poc.poc_msb;
  381. h->poc.prev_poc_lsb = h->poc.poc_lsb;
  382. }
  383. h->poc.prev_frame_num_offset = h->poc.frame_num_offset;
  384. h->poc.prev_frame_num = h->poc.frame_num;
  385. h->recovery_frame = h1->recovery_frame;
  386. return err;
  387. }
  388. static int h264_frame_start(H264Context *h)
  389. {
  390. H264Picture *pic;
  391. int i, ret;
  392. const int pixel_shift = h->pixel_shift;
  393. int c[4] = {
  394. 1<<(h->ps.sps->bit_depth_luma-1),
  395. 1<<(h->ps.sps->bit_depth_chroma-1),
  396. 1<<(h->ps.sps->bit_depth_chroma-1),
  397. -1
  398. };
  399. if (!ff_thread_can_start_frame(h->avctx)) {
  400. av_log(h->avctx, AV_LOG_ERROR, "Attempt to start a frame outside SETUP state\n");
  401. return -1;
  402. }
  403. release_unused_pictures(h, 1);
  404. h->cur_pic_ptr = NULL;
  405. i = find_unused_picture(h);
  406. if (i < 0) {
  407. av_log(h->avctx, AV_LOG_ERROR, "no frame buffer available\n");
  408. return i;
  409. }
  410. pic = &h->DPB[i];
  411. pic->reference = h->droppable ? 0 : h->picture_structure;
  412. pic->f->coded_picture_number = h->coded_picture_number++;
  413. pic->field_picture = h->picture_structure != PICT_FRAME;
  414. pic->frame_num = h->poc.frame_num;
  415. /*
  416. * Zero key_frame here; IDR markings per slice in frame or fields are ORed
  417. * in later.
  418. * See decode_nal_units().
  419. */
  420. pic->f->key_frame = 0;
  421. pic->mmco_reset = 0;
  422. pic->recovered = 0;
  423. pic->invalid_gap = 0;
  424. pic->sei_recovery_frame_cnt = h->sei.recovery_point.recovery_frame_cnt;
  425. pic->f->pict_type = h->slice_ctx[0].slice_type;
  426. if ((ret = alloc_picture(h, pic)) < 0)
  427. return ret;
  428. if(!h->frame_recovered && !h->avctx->hwaccel
  429. #if FF_API_CAP_VDPAU
  430. && !(h->avctx->codec->capabilities & AV_CODEC_CAP_HWACCEL_VDPAU)
  431. #endif
  432. )
  433. ff_color_frame(pic->f, c);
  434. h->cur_pic_ptr = pic;
  435. ff_h264_unref_picture(h, &h->cur_pic);
  436. if (CONFIG_ERROR_RESILIENCE) {
  437. ff_h264_set_erpic(&h->slice_ctx[0].er.cur_pic, NULL);
  438. }
  439. if ((ret = ff_h264_ref_picture(h, &h->cur_pic, h->cur_pic_ptr)) < 0)
  440. return ret;
  441. for (i = 0; i < h->nb_slice_ctx; i++) {
  442. h->slice_ctx[i].linesize = h->cur_pic_ptr->f->linesize[0];
  443. h->slice_ctx[i].uvlinesize = h->cur_pic_ptr->f->linesize[1];
  444. }
  445. if (CONFIG_ERROR_RESILIENCE && h->enable_er) {
  446. ff_er_frame_start(&h->slice_ctx[0].er);
  447. ff_h264_set_erpic(&h->slice_ctx[0].er.last_pic, NULL);
  448. ff_h264_set_erpic(&h->slice_ctx[0].er.next_pic, NULL);
  449. }
  450. for (i = 0; i < 16; i++) {
  451. h->block_offset[i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * pic->f->linesize[0] * ((scan8[i] - scan8[0]) >> 3);
  452. h->block_offset[48 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * pic->f->linesize[0] * ((scan8[i] - scan8[0]) >> 3);
  453. }
  454. for (i = 0; i < 16; i++) {
  455. h->block_offset[16 + i] =
  456. h->block_offset[32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * pic->f->linesize[1] * ((scan8[i] - scan8[0]) >> 3);
  457. h->block_offset[48 + 16 + i] =
  458. h->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * pic->f->linesize[1] * ((scan8[i] - scan8[0]) >> 3);
  459. }
  460. /* We mark the current picture as non-reference after allocating it, so
  461. * that if we break out due to an error it can be released automatically
  462. * in the next ff_mpv_frame_start().
  463. */
  464. h->cur_pic_ptr->reference = 0;
  465. h->cur_pic_ptr->field_poc[0] = h->cur_pic_ptr->field_poc[1] = INT_MAX;
  466. h->next_output_pic = NULL;
  467. h->postpone_filter = 0;
  468. h->mb_aff_frame = h->ps.sps->mb_aff && (h->picture_structure == PICT_FRAME);
  469. assert(h->cur_pic_ptr->long_ref == 0);
  470. return 0;
  471. }
  472. static av_always_inline void backup_mb_border(const H264Context *h, H264SliceContext *sl,
  473. uint8_t *src_y,
  474. uint8_t *src_cb, uint8_t *src_cr,
  475. int linesize, int uvlinesize,
  476. int simple)
  477. {
  478. uint8_t *top_border;
  479. int top_idx = 1;
  480. const int pixel_shift = h->pixel_shift;
  481. int chroma444 = CHROMA444(h);
  482. int chroma422 = CHROMA422(h);
  483. src_y -= linesize;
  484. src_cb -= uvlinesize;
  485. src_cr -= uvlinesize;
  486. if (!simple && FRAME_MBAFF(h)) {
  487. if (sl->mb_y & 1) {
  488. if (!MB_MBAFF(sl)) {
  489. top_border = sl->top_borders[0][sl->mb_x];
  490. AV_COPY128(top_border, src_y + 15 * linesize);
  491. if (pixel_shift)
  492. AV_COPY128(top_border + 16, src_y + 15 * linesize + 16);
  493. if (simple || !CONFIG_GRAY || !(h->flags & AV_CODEC_FLAG_GRAY)) {
  494. if (chroma444) {
  495. if (pixel_shift) {
  496. AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
  497. AV_COPY128(top_border + 48, src_cb + 15 * uvlinesize + 16);
  498. AV_COPY128(top_border + 64, src_cr + 15 * uvlinesize);
  499. AV_COPY128(top_border + 80, src_cr + 15 * uvlinesize + 16);
  500. } else {
  501. AV_COPY128(top_border + 16, src_cb + 15 * uvlinesize);
  502. AV_COPY128(top_border + 32, src_cr + 15 * uvlinesize);
  503. }
  504. } else if (chroma422) {
  505. if (pixel_shift) {
  506. AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
  507. AV_COPY128(top_border + 48, src_cr + 15 * uvlinesize);
  508. } else {
  509. AV_COPY64(top_border + 16, src_cb + 15 * uvlinesize);
  510. AV_COPY64(top_border + 24, src_cr + 15 * uvlinesize);
  511. }
  512. } else {
  513. if (pixel_shift) {
  514. AV_COPY128(top_border + 32, src_cb + 7 * uvlinesize);
  515. AV_COPY128(top_border + 48, src_cr + 7 * uvlinesize);
  516. } else {
  517. AV_COPY64(top_border + 16, src_cb + 7 * uvlinesize);
  518. AV_COPY64(top_border + 24, src_cr + 7 * uvlinesize);
  519. }
  520. }
  521. }
  522. }
  523. } else if (MB_MBAFF(sl)) {
  524. top_idx = 0;
  525. } else
  526. return;
  527. }
  528. top_border = sl->top_borders[top_idx][sl->mb_x];
  529. /* There are two lines saved, the line above the top macroblock
  530. * of a pair, and the line above the bottom macroblock. */
  531. AV_COPY128(top_border, src_y + 16 * linesize);
  532. if (pixel_shift)
  533. AV_COPY128(top_border + 16, src_y + 16 * linesize + 16);
  534. if (simple || !CONFIG_GRAY || !(h->flags & AV_CODEC_FLAG_GRAY)) {
  535. if (chroma444) {
  536. if (pixel_shift) {
  537. AV_COPY128(top_border + 32, src_cb + 16 * linesize);
  538. AV_COPY128(top_border + 48, src_cb + 16 * linesize + 16);
  539. AV_COPY128(top_border + 64, src_cr + 16 * linesize);
  540. AV_COPY128(top_border + 80, src_cr + 16 * linesize + 16);
  541. } else {
  542. AV_COPY128(top_border + 16, src_cb + 16 * linesize);
  543. AV_COPY128(top_border + 32, src_cr + 16 * linesize);
  544. }
  545. } else if (chroma422) {
  546. if (pixel_shift) {
  547. AV_COPY128(top_border + 32, src_cb + 16 * uvlinesize);
  548. AV_COPY128(top_border + 48, src_cr + 16 * uvlinesize);
  549. } else {
  550. AV_COPY64(top_border + 16, src_cb + 16 * uvlinesize);
  551. AV_COPY64(top_border + 24, src_cr + 16 * uvlinesize);
  552. }
  553. } else {
  554. if (pixel_shift) {
  555. AV_COPY128(top_border + 32, src_cb + 8 * uvlinesize);
  556. AV_COPY128(top_border + 48, src_cr + 8 * uvlinesize);
  557. } else {
  558. AV_COPY64(top_border + 16, src_cb + 8 * uvlinesize);
  559. AV_COPY64(top_border + 24, src_cr + 8 * uvlinesize);
  560. }
  561. }
  562. }
  563. }
  564. /**
  565. * Initialize implicit_weight table.
  566. * @param field 0/1 initialize the weight for interlaced MBAFF
  567. * -1 initializes the rest
  568. */
  569. static void implicit_weight_table(const H264Context *h, H264SliceContext *sl, int field)
  570. {
  571. int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
  572. for (i = 0; i < 2; i++) {
  573. sl->pwt.luma_weight_flag[i] = 0;
  574. sl->pwt.chroma_weight_flag[i] = 0;
  575. }
  576. if (field < 0) {
  577. if (h->picture_structure == PICT_FRAME) {
  578. cur_poc = h->cur_pic_ptr->poc;
  579. } else {
  580. cur_poc = h->cur_pic_ptr->field_poc[h->picture_structure - 1];
  581. }
  582. if (sl->ref_count[0] == 1 && sl->ref_count[1] == 1 && !FRAME_MBAFF(h) &&
  583. sl->ref_list[0][0].poc + (int64_t)sl->ref_list[1][0].poc == 2 * cur_poc) {
  584. sl->pwt.use_weight = 0;
  585. sl->pwt.use_weight_chroma = 0;
  586. return;
  587. }
  588. ref_start = 0;
  589. ref_count0 = sl->ref_count[0];
  590. ref_count1 = sl->ref_count[1];
  591. } else {
  592. cur_poc = h->cur_pic_ptr->field_poc[field];
  593. ref_start = 16;
  594. ref_count0 = 16 + 2 * sl->ref_count[0];
  595. ref_count1 = 16 + 2 * sl->ref_count[1];
  596. }
  597. sl->pwt.use_weight = 2;
  598. sl->pwt.use_weight_chroma = 2;
  599. sl->pwt.luma_log2_weight_denom = 5;
  600. sl->pwt.chroma_log2_weight_denom = 5;
  601. for (ref0 = ref_start; ref0 < ref_count0; ref0++) {
  602. int64_t poc0 = sl->ref_list[0][ref0].poc;
  603. for (ref1 = ref_start; ref1 < ref_count1; ref1++) {
  604. int w = 32;
  605. if (!sl->ref_list[0][ref0].parent->long_ref && !sl->ref_list[1][ref1].parent->long_ref) {
  606. int poc1 = sl->ref_list[1][ref1].poc;
  607. int td = av_clip_int8(poc1 - poc0);
  608. if (td) {
  609. int tb = av_clip_int8(cur_poc - poc0);
  610. int tx = (16384 + (FFABS(td) >> 1)) / td;
  611. int dist_scale_factor = (tb * tx + 32) >> 8;
  612. if (dist_scale_factor >= -64 && dist_scale_factor <= 128)
  613. w = 64 - dist_scale_factor;
  614. }
  615. }
  616. if (field < 0) {
  617. sl->pwt.implicit_weight[ref0][ref1][0] =
  618. sl->pwt.implicit_weight[ref0][ref1][1] = w;
  619. } else {
  620. sl->pwt.implicit_weight[ref0][ref1][field] = w;
  621. }
  622. }
  623. }
  624. }
  625. /**
  626. * initialize scan tables
  627. */
  628. static void init_scan_tables(H264Context *h)
  629. {
  630. int i;
  631. for (i = 0; i < 16; i++) {
  632. #define TRANSPOSE(x) ((x) >> 2) | (((x) << 2) & 0xF)
  633. h->zigzag_scan[i] = TRANSPOSE(ff_zigzag_scan[i]);
  634. h->field_scan[i] = TRANSPOSE(field_scan[i]);
  635. #undef TRANSPOSE
  636. }
  637. for (i = 0; i < 64; i++) {
  638. #define TRANSPOSE(x) ((x) >> 3) | (((x) & 7) << 3)
  639. h->zigzag_scan8x8[i] = TRANSPOSE(ff_zigzag_direct[i]);
  640. h->zigzag_scan8x8_cavlc[i] = TRANSPOSE(zigzag_scan8x8_cavlc[i]);
  641. h->field_scan8x8[i] = TRANSPOSE(field_scan8x8[i]);
  642. h->field_scan8x8_cavlc[i] = TRANSPOSE(field_scan8x8_cavlc[i]);
  643. #undef TRANSPOSE
  644. }
  645. if (h->ps.sps->transform_bypass) { // FIXME same ugly
  646. memcpy(h->zigzag_scan_q0 , ff_zigzag_scan , sizeof(h->zigzag_scan_q0 ));
  647. memcpy(h->zigzag_scan8x8_q0 , ff_zigzag_direct , sizeof(h->zigzag_scan8x8_q0 ));
  648. memcpy(h->zigzag_scan8x8_cavlc_q0 , zigzag_scan8x8_cavlc , sizeof(h->zigzag_scan8x8_cavlc_q0));
  649. memcpy(h->field_scan_q0 , field_scan , sizeof(h->field_scan_q0 ));
  650. memcpy(h->field_scan8x8_q0 , field_scan8x8 , sizeof(h->field_scan8x8_q0 ));
  651. memcpy(h->field_scan8x8_cavlc_q0 , field_scan8x8_cavlc , sizeof(h->field_scan8x8_cavlc_q0 ));
  652. } else {
  653. memcpy(h->zigzag_scan_q0 , h->zigzag_scan , sizeof(h->zigzag_scan_q0 ));
  654. memcpy(h->zigzag_scan8x8_q0 , h->zigzag_scan8x8 , sizeof(h->zigzag_scan8x8_q0 ));
  655. memcpy(h->zigzag_scan8x8_cavlc_q0 , h->zigzag_scan8x8_cavlc , sizeof(h->zigzag_scan8x8_cavlc_q0));
  656. memcpy(h->field_scan_q0 , h->field_scan , sizeof(h->field_scan_q0 ));
  657. memcpy(h->field_scan8x8_q0 , h->field_scan8x8 , sizeof(h->field_scan8x8_q0 ));
  658. memcpy(h->field_scan8x8_cavlc_q0 , h->field_scan8x8_cavlc , sizeof(h->field_scan8x8_cavlc_q0 ));
  659. }
  660. }
  661. static enum AVPixelFormat get_pixel_format(H264Context *h, int force_callback)
  662. {
  663. #define HWACCEL_MAX (CONFIG_H264_DXVA2_HWACCEL + \
  664. CONFIG_H264_D3D11VA_HWACCEL + \
  665. CONFIG_H264_VAAPI_HWACCEL + \
  666. (CONFIG_H264_VDA_HWACCEL * 2) + \
  667. CONFIG_H264_VIDEOTOOLBOX_HWACCEL + \
  668. CONFIG_H264_VDPAU_HWACCEL)
  669. enum AVPixelFormat pix_fmts[HWACCEL_MAX + 2], *fmt = pix_fmts;
  670. const enum AVPixelFormat *choices = pix_fmts;
  671. int i;
  672. switch (h->ps.sps->bit_depth_luma) {
  673. case 9:
  674. if (CHROMA444(h)) {
  675. if (h->avctx->colorspace == AVCOL_SPC_RGB) {
  676. *fmt++ = AV_PIX_FMT_GBRP9;
  677. } else
  678. *fmt++ = AV_PIX_FMT_YUV444P9;
  679. } else if (CHROMA422(h))
  680. *fmt++ = AV_PIX_FMT_YUV422P9;
  681. else
  682. *fmt++ = AV_PIX_FMT_YUV420P9;
  683. break;
  684. case 10:
  685. if (CHROMA444(h)) {
  686. if (h->avctx->colorspace == AVCOL_SPC_RGB) {
  687. *fmt++ = AV_PIX_FMT_GBRP10;
  688. } else
  689. *fmt++ = AV_PIX_FMT_YUV444P10;
  690. } else if (CHROMA422(h))
  691. *fmt++ = AV_PIX_FMT_YUV422P10;
  692. else
  693. *fmt++ = AV_PIX_FMT_YUV420P10;
  694. break;
  695. case 12:
  696. if (CHROMA444(h)) {
  697. if (h->avctx->colorspace == AVCOL_SPC_RGB) {
  698. *fmt++ = AV_PIX_FMT_GBRP12;
  699. } else
  700. *fmt++ = AV_PIX_FMT_YUV444P12;
  701. } else if (CHROMA422(h))
  702. *fmt++ = AV_PIX_FMT_YUV422P12;
  703. else
  704. *fmt++ = AV_PIX_FMT_YUV420P12;
  705. break;
  706. case 14:
  707. if (CHROMA444(h)) {
  708. if (h->avctx->colorspace == AVCOL_SPC_RGB) {
  709. *fmt++ = AV_PIX_FMT_GBRP14;
  710. } else
  711. *fmt++ = AV_PIX_FMT_YUV444P14;
  712. } else if (CHROMA422(h))
  713. *fmt++ = AV_PIX_FMT_YUV422P14;
  714. else
  715. *fmt++ = AV_PIX_FMT_YUV420P14;
  716. break;
  717. case 8:
  718. #if CONFIG_H264_VDPAU_HWACCEL
  719. *fmt++ = AV_PIX_FMT_VDPAU;
  720. #endif
  721. if (CHROMA444(h)) {
  722. if (h->avctx->colorspace == AVCOL_SPC_RGB)
  723. *fmt++ = AV_PIX_FMT_GBRP;
  724. else if (h->avctx->color_range == AVCOL_RANGE_JPEG)
  725. *fmt++ = AV_PIX_FMT_YUVJ444P;
  726. else
  727. *fmt++ = AV_PIX_FMT_YUV444P;
  728. } else if (CHROMA422(h)) {
  729. if (h->avctx->color_range == AVCOL_RANGE_JPEG)
  730. *fmt++ = AV_PIX_FMT_YUVJ422P;
  731. else
  732. *fmt++ = AV_PIX_FMT_YUV422P;
  733. } else {
  734. #if CONFIG_H264_DXVA2_HWACCEL
  735. *fmt++ = AV_PIX_FMT_DXVA2_VLD;
  736. #endif
  737. #if CONFIG_H264_D3D11VA_HWACCEL
  738. *fmt++ = AV_PIX_FMT_D3D11VA_VLD;
  739. #endif
  740. #if CONFIG_H264_VAAPI_HWACCEL
  741. *fmt++ = AV_PIX_FMT_VAAPI;
  742. #endif
  743. #if CONFIG_H264_VDA_HWACCEL
  744. *fmt++ = AV_PIX_FMT_VDA_VLD;
  745. *fmt++ = AV_PIX_FMT_VDA;
  746. #endif
  747. #if CONFIG_H264_VIDEOTOOLBOX_HWACCEL
  748. *fmt++ = AV_PIX_FMT_VIDEOTOOLBOX;
  749. #endif
  750. if (h->avctx->codec->pix_fmts)
  751. choices = h->avctx->codec->pix_fmts;
  752. else if (h->avctx->color_range == AVCOL_RANGE_JPEG)
  753. *fmt++ = AV_PIX_FMT_YUVJ420P;
  754. else
  755. *fmt++ = AV_PIX_FMT_YUV420P;
  756. }
  757. break;
  758. default:
  759. av_log(h->avctx, AV_LOG_ERROR,
  760. "Unsupported bit depth %d\n", h->ps.sps->bit_depth_luma);
  761. return AVERROR_INVALIDDATA;
  762. }
  763. *fmt = AV_PIX_FMT_NONE;
  764. for (i=0; choices[i] != AV_PIX_FMT_NONE; i++)
  765. if (choices[i] == h->avctx->pix_fmt && !force_callback)
  766. return choices[i];
  767. return ff_thread_get_format(h->avctx, choices);
  768. }
  769. /* export coded and cropped frame dimensions to AVCodecContext */
  770. static int init_dimensions(H264Context *h)
  771. {
  772. const SPS *sps = (const SPS*)h->ps.sps;
  773. int width = h->width - (sps->crop_right + sps->crop_left);
  774. int height = h->height - (sps->crop_top + sps->crop_bottom);
  775. av_assert0(sps->crop_right + sps->crop_left < (unsigned)h->width);
  776. av_assert0(sps->crop_top + sps->crop_bottom < (unsigned)h->height);
  777. /* handle container cropping */
  778. if (FFALIGN(h->avctx->width, 16) == FFALIGN(width, 16) &&
  779. FFALIGN(h->avctx->height, 16) == FFALIGN(height, 16) &&
  780. h->avctx->width <= width &&
  781. h->avctx->height <= height
  782. ) {
  783. width = h->avctx->width;
  784. height = h->avctx->height;
  785. }
  786. h->avctx->coded_width = h->width;
  787. h->avctx->coded_height = h->height;
  788. h->avctx->width = width;
  789. h->avctx->height = height;
  790. return 0;
  791. }
  792. static int h264_slice_header_init(H264Context *h)
  793. {
  794. const SPS *sps = h->ps.sps;
  795. int i, ret;
  796. ff_set_sar(h->avctx, sps->sar);
  797. av_pix_fmt_get_chroma_sub_sample(h->avctx->pix_fmt,
  798. &h->chroma_x_shift, &h->chroma_y_shift);
  799. if (sps->timing_info_present_flag) {
  800. int64_t den = sps->time_scale;
  801. if (h->sei.unregistered.x264_build < 44U)
  802. den *= 2;
  803. av_reduce(&h->avctx->framerate.den, &h->avctx->framerate.num,
  804. sps->num_units_in_tick * h->avctx->ticks_per_frame, den, 1 << 30);
  805. }
  806. ff_h264_free_tables(h);
  807. h->first_field = 0;
  808. h->prev_interlaced_frame = 1;
  809. init_scan_tables(h);
  810. ret = ff_h264_alloc_tables(h);
  811. if (ret < 0) {
  812. av_log(h->avctx, AV_LOG_ERROR, "Could not allocate memory\n");
  813. goto fail;
  814. }
  815. #if FF_API_CAP_VDPAU
  816. if (h->avctx->codec &&
  817. h->avctx->codec->capabilities & AV_CODEC_CAP_HWACCEL_VDPAU &&
  818. (sps->bit_depth_luma != 8 || sps->chroma_format_idc > 1)) {
  819. av_log(h->avctx, AV_LOG_ERROR,
  820. "VDPAU decoding does not support video colorspace.\n");
  821. ret = AVERROR_INVALIDDATA;
  822. goto fail;
  823. }
  824. #endif
  825. if (sps->bit_depth_luma < 8 || sps->bit_depth_luma > 14 ||
  826. sps->bit_depth_luma == 11 || sps->bit_depth_luma == 13
  827. ) {
  828. av_log(h->avctx, AV_LOG_ERROR, "Unsupported bit depth %d\n",
  829. sps->bit_depth_luma);
  830. ret = AVERROR_INVALIDDATA;
  831. goto fail;
  832. }
  833. h->cur_bit_depth_luma =
  834. h->avctx->bits_per_raw_sample = sps->bit_depth_luma;
  835. h->cur_chroma_format_idc = sps->chroma_format_idc;
  836. h->pixel_shift = sps->bit_depth_luma > 8;
  837. h->chroma_format_idc = sps->chroma_format_idc;
  838. h->bit_depth_luma = sps->bit_depth_luma;
  839. ff_h264dsp_init(&h->h264dsp, sps->bit_depth_luma,
  840. sps->chroma_format_idc);
  841. ff_h264chroma_init(&h->h264chroma, sps->bit_depth_chroma);
  842. ff_h264qpel_init(&h->h264qpel, sps->bit_depth_luma);
  843. ff_h264_pred_init(&h->hpc, h->avctx->codec_id, sps->bit_depth_luma,
  844. sps->chroma_format_idc);
  845. ff_videodsp_init(&h->vdsp, sps->bit_depth_luma);
  846. if (!HAVE_THREADS || !(h->avctx->active_thread_type & FF_THREAD_SLICE)) {
  847. ret = ff_h264_slice_context_init(h, &h->slice_ctx[0]);
  848. if (ret < 0) {
  849. av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
  850. goto fail;
  851. }
  852. } else {
  853. for (i = 0; i < h->nb_slice_ctx; i++) {
  854. H264SliceContext *sl = &h->slice_ctx[i];
  855. sl->h264 = h;
  856. sl->intra4x4_pred_mode = h->intra4x4_pred_mode + i * 8 * 2 * h->mb_stride;
  857. sl->mvd_table[0] = h->mvd_table[0] + i * 8 * 2 * h->mb_stride;
  858. sl->mvd_table[1] = h->mvd_table[1] + i * 8 * 2 * h->mb_stride;
  859. if ((ret = ff_h264_slice_context_init(h, sl)) < 0) {
  860. av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
  861. goto fail;
  862. }
  863. }
  864. }
  865. h->context_initialized = 1;
  866. return 0;
  867. fail:
  868. ff_h264_free_tables(h);
  869. h->context_initialized = 0;
  870. return ret;
  871. }
  872. static enum AVPixelFormat non_j_pixfmt(enum AVPixelFormat a)
  873. {
  874. switch (a) {
  875. case AV_PIX_FMT_YUVJ420P: return AV_PIX_FMT_YUV420P;
  876. case AV_PIX_FMT_YUVJ422P: return AV_PIX_FMT_YUV422P;
  877. case AV_PIX_FMT_YUVJ444P: return AV_PIX_FMT_YUV444P;
  878. default:
  879. return a;
  880. }
  881. }
  882. static int h264_init_ps(H264Context *h, const H264SliceContext *sl, int first_slice)
  883. {
  884. const SPS *sps;
  885. int needs_reinit = 0, must_reinit, ret;
  886. if (first_slice) {
  887. av_buffer_unref(&h->ps.pps_ref);
  888. h->ps.pps = NULL;
  889. h->ps.pps_ref = av_buffer_ref(h->ps.pps_list[sl->pps_id]);
  890. if (!h->ps.pps_ref)
  891. return AVERROR(ENOMEM);
  892. h->ps.pps = (const PPS*)h->ps.pps_ref->data;
  893. }
  894. if (h->ps.sps != (const SPS*)h->ps.sps_list[h->ps.pps->sps_id]->data) {
  895. av_buffer_unref(&h->ps.sps_ref);
  896. h->ps.sps = NULL;
  897. h->ps.sps_ref = av_buffer_ref(h->ps.sps_list[h->ps.pps->sps_id]);
  898. if (!h->ps.sps_ref)
  899. return AVERROR(ENOMEM);
  900. h->ps.sps = (const SPS*)h->ps.sps_ref->data;
  901. if (h->mb_width != h->ps.sps->mb_width ||
  902. h->mb_height != h->ps.sps->mb_height ||
  903. h->cur_bit_depth_luma != h->ps.sps->bit_depth_luma ||
  904. h->cur_chroma_format_idc != h->ps.sps->chroma_format_idc
  905. )
  906. needs_reinit = 1;
  907. if (h->bit_depth_luma != h->ps.sps->bit_depth_luma ||
  908. h->chroma_format_idc != h->ps.sps->chroma_format_idc)
  909. needs_reinit = 1;
  910. }
  911. sps = h->ps.sps;
  912. must_reinit = (h->context_initialized &&
  913. ( 16*sps->mb_width != h->avctx->coded_width
  914. || 16*sps->mb_height != h->avctx->coded_height
  915. || h->cur_bit_depth_luma != sps->bit_depth_luma
  916. || h->cur_chroma_format_idc != sps->chroma_format_idc
  917. || h->mb_width != sps->mb_width
  918. || h->mb_height != sps->mb_height
  919. ));
  920. if (h->avctx->pix_fmt == AV_PIX_FMT_NONE
  921. || (non_j_pixfmt(h->avctx->pix_fmt) != non_j_pixfmt(get_pixel_format(h, 0))))
  922. must_reinit = 1;
  923. if (first_slice && av_cmp_q(sps->sar, h->avctx->sample_aspect_ratio))
  924. must_reinit = 1;
  925. if (!h->setup_finished) {
  926. h->avctx->profile = ff_h264_get_profile(sps);
  927. h->avctx->level = sps->level_idc;
  928. h->avctx->refs = sps->ref_frame_count;
  929. h->mb_width = sps->mb_width;
  930. h->mb_height = sps->mb_height;
  931. h->mb_num = h->mb_width * h->mb_height;
  932. h->mb_stride = h->mb_width + 1;
  933. h->b_stride = h->mb_width * 4;
  934. h->chroma_y_shift = sps->chroma_format_idc <= 1; // 400 uses yuv420p
  935. h->width = 16 * h->mb_width;
  936. h->height = 16 * h->mb_height;
  937. ret = init_dimensions(h);
  938. if (ret < 0)
  939. return ret;
  940. if (sps->video_signal_type_present_flag) {
  941. h->avctx->color_range = sps->full_range > 0 ? AVCOL_RANGE_JPEG
  942. : AVCOL_RANGE_MPEG;
  943. if (sps->colour_description_present_flag) {
  944. if (h->avctx->colorspace != sps->colorspace)
  945. needs_reinit = 1;
  946. h->avctx->color_primaries = sps->color_primaries;
  947. h->avctx->color_trc = sps->color_trc;
  948. h->avctx->colorspace = sps->colorspace;
  949. }
  950. }
  951. }
  952. if (!h->context_initialized || must_reinit || needs_reinit) {
  953. int flush_changes = h->context_initialized;
  954. h->context_initialized = 0;
  955. if (sl != h->slice_ctx) {
  956. av_log(h->avctx, AV_LOG_ERROR,
  957. "changing width %d -> %d / height %d -> %d on "
  958. "slice %d\n",
  959. h->width, h->avctx->coded_width,
  960. h->height, h->avctx->coded_height,
  961. h->current_slice + 1);
  962. return AVERROR_INVALIDDATA;
  963. }
  964. av_assert1(first_slice);
  965. if (flush_changes)
  966. ff_h264_flush_change(h);
  967. if ((ret = get_pixel_format(h, 1)) < 0)
  968. return ret;
  969. h->avctx->pix_fmt = ret;
  970. av_log(h->avctx, AV_LOG_VERBOSE, "Reinit context to %dx%d, "
  971. "pix_fmt: %s\n", h->width, h->height, av_get_pix_fmt_name(h->avctx->pix_fmt));
  972. if ((ret = h264_slice_header_init(h)) < 0) {
  973. av_log(h->avctx, AV_LOG_ERROR,
  974. "h264_slice_header_init() failed\n");
  975. return ret;
  976. }
  977. }
  978. return 0;
  979. }
  980. static int h264_export_frame_props(H264Context *h)
  981. {
  982. const SPS *sps = h->ps.sps;
  983. H264Picture *cur = h->cur_pic_ptr;
  984. cur->f->interlaced_frame = 0;
  985. cur->f->repeat_pict = 0;
  986. /* Signal interlacing information externally. */
  987. /* Prioritize picture timing SEI information over used
  988. * decoding process if it exists. */
  989. if (sps->pic_struct_present_flag && h->sei.picture_timing.present) {
  990. H264SEIPictureTiming *pt = &h->sei.picture_timing;
  991. switch (pt->pic_struct) {
  992. case SEI_PIC_STRUCT_FRAME:
  993. break;
  994. case SEI_PIC_STRUCT_TOP_FIELD:
  995. case SEI_PIC_STRUCT_BOTTOM_FIELD:
  996. cur->f->interlaced_frame = 1;
  997. break;
  998. case SEI_PIC_STRUCT_TOP_BOTTOM:
  999. case SEI_PIC_STRUCT_BOTTOM_TOP:
  1000. if (FIELD_OR_MBAFF_PICTURE(h))
  1001. cur->f->interlaced_frame = 1;
  1002. else
  1003. // try to flag soft telecine progressive
  1004. cur->f->interlaced_frame = h->prev_interlaced_frame;
  1005. break;
  1006. case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
  1007. case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
  1008. /* Signal the possibility of telecined film externally
  1009. * (pic_struct 5,6). From these hints, let the applications
  1010. * decide if they apply deinterlacing. */
  1011. cur->f->repeat_pict = 1;
  1012. break;
  1013. case SEI_PIC_STRUCT_FRAME_DOUBLING:
  1014. cur->f->repeat_pict = 2;
  1015. break;
  1016. case SEI_PIC_STRUCT_FRAME_TRIPLING:
  1017. cur->f->repeat_pict = 4;
  1018. break;
  1019. }
  1020. if ((pt->ct_type & 3) &&
  1021. pt->pic_struct <= SEI_PIC_STRUCT_BOTTOM_TOP)
  1022. cur->f->interlaced_frame = (pt->ct_type & (1 << 1)) != 0;
  1023. } else {
  1024. /* Derive interlacing flag from used decoding process. */
  1025. cur->f->interlaced_frame = FIELD_OR_MBAFF_PICTURE(h);
  1026. }
  1027. h->prev_interlaced_frame = cur->f->interlaced_frame;
  1028. if (cur->field_poc[0] != cur->field_poc[1]) {
  1029. /* Derive top_field_first from field pocs. */
  1030. cur->f->top_field_first = cur->field_poc[0] < cur->field_poc[1];
  1031. } else {
  1032. if (sps->pic_struct_present_flag && h->sei.picture_timing.present) {
  1033. /* Use picture timing SEI information. Even if it is a
  1034. * information of a past frame, better than nothing. */
  1035. if (h->sei.picture_timing.pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM ||
  1036. h->sei.picture_timing.pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM_TOP)
  1037. cur->f->top_field_first = 1;
  1038. else
  1039. cur->f->top_field_first = 0;
  1040. } else if (cur->f->interlaced_frame) {
  1041. /* Default to top field first when pic_struct_present_flag
  1042. * is not set but interlaced frame detected */
  1043. cur->f->top_field_first = 1;
  1044. } else {
  1045. /* Most likely progressive */
  1046. cur->f->top_field_first = 0;
  1047. }
  1048. }
  1049. if (h->sei.frame_packing.present &&
  1050. h->sei.frame_packing.frame_packing_arrangement_type <= 6 &&
  1051. h->sei.frame_packing.content_interpretation_type > 0 &&
  1052. h->sei.frame_packing.content_interpretation_type < 3) {
  1053. H264SEIFramePacking *fp = &h->sei.frame_packing;
  1054. AVStereo3D *stereo = av_stereo3d_create_side_data(cur->f);
  1055. if (stereo) {
  1056. switch (fp->frame_packing_arrangement_type) {
  1057. case 0:
  1058. stereo->type = AV_STEREO3D_CHECKERBOARD;
  1059. break;
  1060. case 1:
  1061. stereo->type = AV_STEREO3D_COLUMNS;
  1062. break;
  1063. case 2:
  1064. stereo->type = AV_STEREO3D_LINES;
  1065. break;
  1066. case 3:
  1067. if (fp->quincunx_sampling_flag)
  1068. stereo->type = AV_STEREO3D_SIDEBYSIDE_QUINCUNX;
  1069. else
  1070. stereo->type = AV_STEREO3D_SIDEBYSIDE;
  1071. break;
  1072. case 4:
  1073. stereo->type = AV_STEREO3D_TOPBOTTOM;
  1074. break;
  1075. case 5:
  1076. stereo->type = AV_STEREO3D_FRAMESEQUENCE;
  1077. break;
  1078. case 6:
  1079. stereo->type = AV_STEREO3D_2D;
  1080. break;
  1081. }
  1082. if (fp->content_interpretation_type == 2)
  1083. stereo->flags = AV_STEREO3D_FLAG_INVERT;
  1084. }
  1085. }
  1086. if (h->sei.display_orientation.present &&
  1087. (h->sei.display_orientation.anticlockwise_rotation ||
  1088. h->sei.display_orientation.hflip ||
  1089. h->sei.display_orientation.vflip)) {
  1090. H264SEIDisplayOrientation *o = &h->sei.display_orientation;
  1091. double angle = o->anticlockwise_rotation * 360 / (double) (1 << 16);
  1092. AVFrameSideData *rotation = av_frame_new_side_data(cur->f,
  1093. AV_FRAME_DATA_DISPLAYMATRIX,
  1094. sizeof(int32_t) * 9);
  1095. if (rotation) {
  1096. av_display_rotation_set((int32_t *)rotation->data, angle);
  1097. av_display_matrix_flip((int32_t *)rotation->data,
  1098. o->hflip, o->vflip);
  1099. }
  1100. }
  1101. if (h->sei.afd.present) {
  1102. AVFrameSideData *sd = av_frame_new_side_data(cur->f, AV_FRAME_DATA_AFD,
  1103. sizeof(uint8_t));
  1104. if (sd) {
  1105. *sd->data = h->sei.afd.active_format_description;
  1106. h->sei.afd.present = 0;
  1107. }
  1108. }
  1109. if (h->sei.a53_caption.a53_caption) {
  1110. H264SEIA53Caption *a53 = &h->sei.a53_caption;
  1111. AVFrameSideData *sd = av_frame_new_side_data(cur->f,
  1112. AV_FRAME_DATA_A53_CC,
  1113. a53->a53_caption_size);
  1114. if (sd)
  1115. memcpy(sd->data, a53->a53_caption, a53->a53_caption_size);
  1116. av_freep(&a53->a53_caption);
  1117. a53->a53_caption_size = 0;
  1118. h->avctx->properties |= FF_CODEC_PROPERTY_CLOSED_CAPTIONS;
  1119. }
  1120. return 0;
  1121. }
  1122. static int h264_select_output_frame(H264Context *h)
  1123. {
  1124. const SPS *sps = h->ps.sps;
  1125. H264Picture *out = h->cur_pic_ptr;
  1126. H264Picture *cur = h->cur_pic_ptr;
  1127. int i, pics, out_of_order, out_idx;
  1128. cur->mmco_reset = h->mmco_reset;
  1129. h->mmco_reset = 0;
  1130. if (sps->bitstream_restriction_flag ||
  1131. h->avctx->strict_std_compliance >= FF_COMPLIANCE_STRICT) {
  1132. h->avctx->has_b_frames = FFMAX(h->avctx->has_b_frames, sps->num_reorder_frames);
  1133. }
  1134. for (i = 0; 1; i++) {
  1135. if(i == MAX_DELAYED_PIC_COUNT || cur->poc < h->last_pocs[i]){
  1136. if(i)
  1137. h->last_pocs[i-1] = cur->poc;
  1138. break;
  1139. } else if(i) {
  1140. h->last_pocs[i-1]= h->last_pocs[i];
  1141. }
  1142. }
  1143. out_of_order = MAX_DELAYED_PIC_COUNT - i;
  1144. if( cur->f->pict_type == AV_PICTURE_TYPE_B
  1145. || (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))
  1146. out_of_order = FFMAX(out_of_order, 1);
  1147. if (out_of_order == MAX_DELAYED_PIC_COUNT) {
  1148. av_log(h->avctx, AV_LOG_VERBOSE, "Invalid POC %d<%d\n", cur->poc, h->last_pocs[0]);
  1149. for (i = 1; i < MAX_DELAYED_PIC_COUNT; i++)
  1150. h->last_pocs[i] = INT_MIN;
  1151. h->last_pocs[0] = cur->poc;
  1152. cur->mmco_reset = 1;
  1153. } else if(h->avctx->has_b_frames < out_of_order && !sps->bitstream_restriction_flag){
  1154. int loglevel = h->avctx->frame_number > 1 ? AV_LOG_WARNING : AV_LOG_VERBOSE;
  1155. av_log(h->avctx, loglevel, "Increasing reorder buffer to %d\n", out_of_order);
  1156. h->avctx->has_b_frames = out_of_order;
  1157. }
  1158. pics = 0;
  1159. while (h->delayed_pic[pics])
  1160. pics++;
  1161. av_assert0(pics <= MAX_DELAYED_PIC_COUNT);
  1162. h->delayed_pic[pics++] = cur;
  1163. if (cur->reference == 0)
  1164. cur->reference = DELAYED_PIC_REF;
  1165. out = h->delayed_pic[0];
  1166. out_idx = 0;
  1167. for (i = 1; h->delayed_pic[i] &&
  1168. !h->delayed_pic[i]->f->key_frame &&
  1169. !h->delayed_pic[i]->mmco_reset;
  1170. i++)
  1171. if (h->delayed_pic[i]->poc < out->poc) {
  1172. out = h->delayed_pic[i];
  1173. out_idx = i;
  1174. }
  1175. if (h->avctx->has_b_frames == 0 &&
  1176. (h->delayed_pic[0]->f->key_frame || h->delayed_pic[0]->mmco_reset))
  1177. h->next_outputed_poc = INT_MIN;
  1178. out_of_order = out->poc < h->next_outputed_poc;
  1179. if (out_of_order || pics > h->avctx->has_b_frames) {
  1180. out->reference &= ~DELAYED_PIC_REF;
  1181. for (i = out_idx; h->delayed_pic[i]; i++)
  1182. h->delayed_pic[i] = h->delayed_pic[i + 1];
  1183. }
  1184. if (!out_of_order && pics > h->avctx->has_b_frames) {
  1185. h->next_output_pic = out;
  1186. if (out_idx == 0 && h->delayed_pic[0] && (h->delayed_pic[0]->f->key_frame || h->delayed_pic[0]->mmco_reset)) {
  1187. h->next_outputed_poc = INT_MIN;
  1188. } else
  1189. h->next_outputed_poc = out->poc;
  1190. if (out->recovered) {
  1191. // We have reached an recovery point and all frames after it in
  1192. // display order are "recovered".
  1193. h->frame_recovered |= FRAME_RECOVERED_SEI;
  1194. }
  1195. out->recovered |= !!(h->frame_recovered & FRAME_RECOVERED_SEI);
  1196. if (!out->recovered) {
  1197. if (!(h->avctx->flags & AV_CODEC_FLAG_OUTPUT_CORRUPT) &&
  1198. !(h->avctx->flags2 & AV_CODEC_FLAG2_SHOW_ALL)) {
  1199. h->next_output_pic = NULL;
  1200. } else {
  1201. out->f->flags |= AV_FRAME_FLAG_CORRUPT;
  1202. }
  1203. }
  1204. } else {
  1205. av_log(h->avctx, AV_LOG_DEBUG, "no picture %s\n", out_of_order ? "ooo" : "");
  1206. }
  1207. return 0;
  1208. }
  1209. /* This function is called right after decoding the slice header for a first
  1210. * slice in a field (or a frame). It decides whether we are decoding a new frame
  1211. * or a second field in a pair and does the necessary setup.
  1212. */
  1213. static int h264_field_start(H264Context *h, const H264SliceContext *sl,
  1214. const H2645NAL *nal, int first_slice)
  1215. {
  1216. int i;
  1217. const SPS *sps;
  1218. int last_pic_structure, last_pic_droppable, ret;
  1219. ret = h264_init_ps(h, sl, first_slice);
  1220. if (ret < 0)
  1221. return ret;
  1222. sps = h->ps.sps;
  1223. last_pic_droppable = h->droppable;
  1224. last_pic_structure = h->picture_structure;
  1225. h->droppable = (nal->ref_idc == 0);
  1226. h->picture_structure = sl->picture_structure;
  1227. h->poc.frame_num = sl->frame_num;
  1228. h->poc.poc_lsb = sl->poc_lsb;
  1229. h->poc.delta_poc_bottom = sl->delta_poc_bottom;
  1230. h->poc.delta_poc[0] = sl->delta_poc[0];
  1231. h->poc.delta_poc[1] = sl->delta_poc[1];
  1232. /* Shorten frame num gaps so we don't have to allocate reference
  1233. * frames just to throw them away */
  1234. if (h->poc.frame_num != h->poc.prev_frame_num) {
  1235. int unwrap_prev_frame_num = h->poc.prev_frame_num;
  1236. int max_frame_num = 1 << sps->log2_max_frame_num;
  1237. if (unwrap_prev_frame_num > h->poc.frame_num)
  1238. unwrap_prev_frame_num -= max_frame_num;
  1239. if ((h->poc.frame_num - unwrap_prev_frame_num) > sps->ref_frame_count) {
  1240. unwrap_prev_frame_num = (h->poc.frame_num - sps->ref_frame_count) - 1;
  1241. if (unwrap_prev_frame_num < 0)
  1242. unwrap_prev_frame_num += max_frame_num;
  1243. h->poc.prev_frame_num = unwrap_prev_frame_num;
  1244. }
  1245. }
  1246. /* See if we have a decoded first field looking for a pair...
  1247. * Here, we're using that to see if we should mark previously
  1248. * decode frames as "finished".
  1249. * We have to do that before the "dummy" in-between frame allocation,
  1250. * since that can modify h->cur_pic_ptr. */
  1251. if (h->first_field) {
  1252. int last_field = last_pic_structure == PICT_BOTTOM_FIELD;
  1253. av_assert0(h->cur_pic_ptr);
  1254. av_assert0(h->cur_pic_ptr->f->buf[0]);
  1255. assert(h->cur_pic_ptr->reference != DELAYED_PIC_REF);
  1256. /* Mark old field/frame as completed */
  1257. if (h->cur_pic_ptr->tf.owner[last_field] == h->avctx) {
  1258. ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, last_field);
  1259. }
  1260. /* figure out if we have a complementary field pair */
  1261. if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
  1262. /* Previous field is unmatched. Don't display it, but let it
  1263. * remain for reference if marked as such. */
  1264. if (last_pic_structure != PICT_FRAME) {
  1265. ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
  1266. last_pic_structure == PICT_TOP_FIELD);
  1267. }
  1268. } else {
  1269. if (h->cur_pic_ptr->frame_num != h->poc.frame_num) {
  1270. /* This and previous field were reference, but had
  1271. * different frame_nums. Consider this field first in
  1272. * pair. Throw away previous field except for reference
  1273. * purposes. */
  1274. if (last_pic_structure != PICT_FRAME) {
  1275. ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
  1276. last_pic_structure == PICT_TOP_FIELD);
  1277. }
  1278. } else {
  1279. /* Second field in complementary pair */
  1280. if (!((last_pic_structure == PICT_TOP_FIELD &&
  1281. h->picture_structure == PICT_BOTTOM_FIELD) ||
  1282. (last_pic_structure == PICT_BOTTOM_FIELD &&
  1283. h->picture_structure == PICT_TOP_FIELD))) {
  1284. av_log(h->avctx, AV_LOG_ERROR,
  1285. "Invalid field mode combination %d/%d\n",
  1286. last_pic_structure, h->picture_structure);
  1287. h->picture_structure = last_pic_structure;
  1288. h->droppable = last_pic_droppable;
  1289. return AVERROR_INVALIDDATA;
  1290. } else if (last_pic_droppable != h->droppable) {
  1291. avpriv_request_sample(h->avctx,
  1292. "Found reference and non-reference fields in the same frame, which");
  1293. h->picture_structure = last_pic_structure;
  1294. h->droppable = last_pic_droppable;
  1295. return AVERROR_PATCHWELCOME;
  1296. }
  1297. }
  1298. }
  1299. }
  1300. while (h->poc.frame_num != h->poc.prev_frame_num && !h->first_field &&
  1301. h->poc.frame_num != (h->poc.prev_frame_num + 1) % (1 << sps->log2_max_frame_num)) {
  1302. H264Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
  1303. av_log(h->avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n",
  1304. h->poc.frame_num, h->poc.prev_frame_num);
  1305. if (!sps->gaps_in_frame_num_allowed_flag)
  1306. for(i=0; i<FF_ARRAY_ELEMS(h->last_pocs); i++)
  1307. h->last_pocs[i] = INT_MIN;
  1308. ret = h264_frame_start(h);
  1309. if (ret < 0) {
  1310. h->first_field = 0;
  1311. return ret;
  1312. }
  1313. h->poc.prev_frame_num++;
  1314. h->poc.prev_frame_num %= 1 << sps->log2_max_frame_num;
  1315. h->cur_pic_ptr->frame_num = h->poc.prev_frame_num;
  1316. h->cur_pic_ptr->invalid_gap = !sps->gaps_in_frame_num_allowed_flag;
  1317. ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 0);
  1318. ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 1);
  1319. h->explicit_ref_marking = 0;
  1320. ret = ff_h264_execute_ref_pic_marking(h);
  1321. if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
  1322. return ret;
  1323. /* Error concealment: If a ref is missing, copy the previous ref
  1324. * in its place.
  1325. * FIXME: Avoiding a memcpy would be nice, but ref handling makes
  1326. * many assumptions about there being no actual duplicates.
  1327. * FIXME: This does not copy padding for out-of-frame motion
  1328. * vectors. Given we are concealing a lost frame, this probably
  1329. * is not noticeable by comparison, but it should be fixed. */
  1330. if (h->short_ref_count) {
  1331. if (prev &&
  1332. h->short_ref[0]->f->width == prev->f->width &&
  1333. h->short_ref[0]->f->height == prev->f->height &&
  1334. h->short_ref[0]->f->format == prev->f->format) {
  1335. ff_thread_await_progress(&prev->tf, INT_MAX, 0);
  1336. if (prev->field_picture)
  1337. ff_thread_await_progress(&prev->tf, INT_MAX, 1);
  1338. av_image_copy(h->short_ref[0]->f->data,
  1339. h->short_ref[0]->f->linesize,
  1340. (const uint8_t **)prev->f->data,
  1341. prev->f->linesize,
  1342. prev->f->format,
  1343. prev->f->width,
  1344. prev->f->height);
  1345. h->short_ref[0]->poc = prev->poc + 2;
  1346. }
  1347. h->short_ref[0]->frame_num = h->poc.prev_frame_num;
  1348. }
  1349. }
  1350. /* See if we have a decoded first field looking for a pair...
  1351. * We're using that to see whether to continue decoding in that
  1352. * frame, or to allocate a new one. */
  1353. if (h->first_field) {
  1354. av_assert0(h->cur_pic_ptr);
  1355. av_assert0(h->cur_pic_ptr->f->buf[0]);
  1356. assert(h->cur_pic_ptr->reference != DELAYED_PIC_REF);
  1357. /* figure out if we have a complementary field pair */
  1358. if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
  1359. /* Previous field is unmatched. Don't display it, but let it
  1360. * remain for reference if marked as such. */
  1361. h->missing_fields ++;
  1362. h->cur_pic_ptr = NULL;
  1363. h->first_field = FIELD_PICTURE(h);
  1364. } else {
  1365. h->missing_fields = 0;
  1366. if (h->cur_pic_ptr->frame_num != h->poc.frame_num) {
  1367. ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
  1368. h->picture_structure==PICT_BOTTOM_FIELD);
  1369. /* This and the previous field had different frame_nums.
  1370. * Consider this field first in pair. Throw away previous
  1371. * one except for reference purposes. */
  1372. h->first_field = 1;
  1373. h->cur_pic_ptr = NULL;
  1374. } else {
  1375. /* Second field in complementary pair */
  1376. h->first_field = 0;
  1377. }
  1378. }
  1379. } else {
  1380. /* Frame or first field in a potentially complementary pair */
  1381. h->first_field = FIELD_PICTURE(h);
  1382. }
  1383. if (!FIELD_PICTURE(h) || h->first_field) {
  1384. if (h264_frame_start(h) < 0) {
  1385. h->first_field = 0;
  1386. return AVERROR_INVALIDDATA;
  1387. }
  1388. } else {
  1389. int field = h->picture_structure == PICT_BOTTOM_FIELD;
  1390. release_unused_pictures(h, 0);
  1391. h->cur_pic_ptr->tf.owner[field] = h->avctx;
  1392. }
  1393. /* Some macroblocks can be accessed before they're available in case
  1394. * of lost slices, MBAFF or threading. */
  1395. if (FIELD_PICTURE(h)) {
  1396. for(i = (h->picture_structure == PICT_BOTTOM_FIELD); i<h->mb_height; i++)
  1397. memset(h->slice_table + i*h->mb_stride, -1, (h->mb_stride - (i+1==h->mb_height)) * sizeof(*h->slice_table));
  1398. } else {
  1399. memset(h->slice_table, -1,
  1400. (h->mb_height * h->mb_stride - 1) * sizeof(*h->slice_table));
  1401. }
  1402. ff_h264_init_poc(h->cur_pic_ptr->field_poc, &h->cur_pic_ptr->poc,
  1403. h->ps.sps, &h->poc, h->picture_structure, nal->ref_idc);
  1404. memcpy(h->mmco, sl->mmco, sl->nb_mmco * sizeof(*h->mmco));
  1405. h->nb_mmco = sl->nb_mmco;
  1406. h->explicit_ref_marking = sl->explicit_ref_marking;
  1407. h->picture_idr = nal->type == H264_NAL_IDR_SLICE;
  1408. if (h->sei.recovery_point.recovery_frame_cnt >= 0) {
  1409. const int sei_recovery_frame_cnt = h->sei.recovery_point.recovery_frame_cnt;
  1410. if (h->poc.frame_num != sei_recovery_frame_cnt || sl->slice_type_nos != AV_PICTURE_TYPE_I)
  1411. h->valid_recovery_point = 1;
  1412. if ( h->recovery_frame < 0
  1413. || av_mod_uintp2(h->recovery_frame - h->poc.frame_num, h->ps.sps->log2_max_frame_num) > sei_recovery_frame_cnt) {
  1414. h->recovery_frame = av_mod_uintp2(h->poc.frame_num + sei_recovery_frame_cnt, h->ps.sps->log2_max_frame_num);
  1415. if (!h->valid_recovery_point)
  1416. h->recovery_frame = h->poc.frame_num;
  1417. }
  1418. }
  1419. h->cur_pic_ptr->f->key_frame |= (nal->type == H264_NAL_IDR_SLICE);
  1420. if (nal->type == H264_NAL_IDR_SLICE ||
  1421. (h->recovery_frame == h->poc.frame_num && nal->ref_idc)) {
  1422. h->recovery_frame = -1;
  1423. h->cur_pic_ptr->recovered = 1;
  1424. }
  1425. // If we have an IDR, all frames after it in decoded order are
  1426. // "recovered".
  1427. if (nal->type == H264_NAL_IDR_SLICE)
  1428. h->frame_recovered |= FRAME_RECOVERED_IDR;
  1429. #if 1
  1430. h->cur_pic_ptr->recovered |= h->frame_recovered;
  1431. #else
  1432. h->cur_pic_ptr->recovered |= !!(h->frame_recovered & FRAME_RECOVERED_IDR);
  1433. #endif
  1434. /* Set the frame properties/side data. Only done for the second field in
  1435. * field coded frames, since some SEI information is present for each field
  1436. * and is merged by the SEI parsing code. */
  1437. if (!FIELD_PICTURE(h) || !h->first_field || h->missing_fields > 1) {
  1438. ret = h264_export_frame_props(h);
  1439. if (ret < 0)
  1440. return ret;
  1441. ret = h264_select_output_frame(h);
  1442. if (ret < 0)
  1443. return ret;
  1444. }
  1445. return 0;
  1446. }
  1447. static int h264_slice_header_parse(const H264Context *h, H264SliceContext *sl,
  1448. const H2645NAL *nal)
  1449. {
  1450. const SPS *sps;
  1451. const PPS *pps;
  1452. int ret;
  1453. unsigned int slice_type, tmp, i;
  1454. int field_pic_flag, bottom_field_flag;
  1455. int first_slice = sl == h->slice_ctx && !h->current_slice;
  1456. int picture_structure;
  1457. if (first_slice)
  1458. av_assert0(!h->setup_finished);
  1459. sl->first_mb_addr = get_ue_golomb_long(&sl->gb);
  1460. slice_type = get_ue_golomb_31(&sl->gb);
  1461. if (slice_type > 9) {
  1462. av_log(h->avctx, AV_LOG_ERROR,
  1463. "slice type %d too large at %d\n",
  1464. slice_type, sl->first_mb_addr);
  1465. return AVERROR_INVALIDDATA;
  1466. }
  1467. if (slice_type > 4) {
  1468. slice_type -= 5;
  1469. sl->slice_type_fixed = 1;
  1470. } else
  1471. sl->slice_type_fixed = 0;
  1472. slice_type = ff_h264_golomb_to_pict_type[slice_type];
  1473. sl->slice_type = slice_type;
  1474. sl->slice_type_nos = slice_type & 3;
  1475. if (nal->type == H264_NAL_IDR_SLICE &&
  1476. sl->slice_type_nos != AV_PICTURE_TYPE_I) {
  1477. av_log(h->avctx, AV_LOG_ERROR, "A non-intra slice in an IDR NAL unit.\n");
  1478. return AVERROR_INVALIDDATA;
  1479. }
  1480. sl->pps_id = get_ue_golomb(&sl->gb);
  1481. if (sl->pps_id >= MAX_PPS_COUNT) {
  1482. av_log(h->avctx, AV_LOG_ERROR, "pps_id %u out of range\n", sl->pps_id);
  1483. return AVERROR_INVALIDDATA;
  1484. }
  1485. if (!h->ps.pps_list[sl->pps_id]) {
  1486. av_log(h->avctx, AV_LOG_ERROR,
  1487. "non-existing PPS %u referenced\n",
  1488. sl->pps_id);
  1489. return AVERROR_INVALIDDATA;
  1490. }
  1491. pps = (const PPS*)h->ps.pps_list[sl->pps_id]->data;
  1492. if (!h->ps.sps_list[pps->sps_id]) {
  1493. av_log(h->avctx, AV_LOG_ERROR,
  1494. "non-existing SPS %u referenced\n", pps->sps_id);
  1495. return AVERROR_INVALIDDATA;
  1496. }
  1497. sps = (const SPS*)h->ps.sps_list[pps->sps_id]->data;
  1498. sl->frame_num = get_bits(&sl->gb, sps->log2_max_frame_num);
  1499. if (!first_slice) {
  1500. if (h->poc.frame_num != sl->frame_num) {
  1501. av_log(h->avctx, AV_LOG_ERROR, "Frame num change from %d to %d\n",
  1502. h->poc.frame_num, sl->frame_num);
  1503. return AVERROR_INVALIDDATA;
  1504. }
  1505. }
  1506. sl->mb_mbaff = 0;
  1507. if (sps->frame_mbs_only_flag) {
  1508. picture_structure = PICT_FRAME;
  1509. } else {
  1510. if (!sps->direct_8x8_inference_flag && slice_type == AV_PICTURE_TYPE_B) {
  1511. av_log(h->avctx, AV_LOG_ERROR, "This stream was generated by a broken encoder, invalid 8x8 inference\n");
  1512. return -1;
  1513. }
  1514. field_pic_flag = get_bits1(&sl->gb);
  1515. if (field_pic_flag) {
  1516. bottom_field_flag = get_bits1(&sl->gb);
  1517. picture_structure = PICT_TOP_FIELD + bottom_field_flag;
  1518. } else {
  1519. picture_structure = PICT_FRAME;
  1520. }
  1521. }
  1522. sl->picture_structure = picture_structure;
  1523. sl->mb_field_decoding_flag = picture_structure != PICT_FRAME;
  1524. if (picture_structure == PICT_FRAME) {
  1525. sl->curr_pic_num = sl->frame_num;
  1526. sl->max_pic_num = 1 << sps->log2_max_frame_num;
  1527. } else {
  1528. sl->curr_pic_num = 2 * sl->frame_num + 1;
  1529. sl->max_pic_num = 1 << (sps->log2_max_frame_num + 1);
  1530. }
  1531. if (nal->type == H264_NAL_IDR_SLICE)
  1532. get_ue_golomb_long(&sl->gb); /* idr_pic_id */
  1533. if (sps->poc_type == 0) {
  1534. sl->poc_lsb = get_bits(&sl->gb, sps->log2_max_poc_lsb);
  1535. if (pps->pic_order_present == 1 && picture_structure == PICT_FRAME)
  1536. sl->delta_poc_bottom = get_se_golomb(&sl->gb);
  1537. }
  1538. if (sps->poc_type == 1 && !sps->delta_pic_order_always_zero_flag) {
  1539. sl->delta_poc[0] = get_se_golomb(&sl->gb);
  1540. if (pps->pic_order_present == 1 && picture_structure == PICT_FRAME)
  1541. sl->delta_poc[1] = get_se_golomb(&sl->gb);
  1542. }
  1543. sl->redundant_pic_count = 0;
  1544. if (pps->redundant_pic_cnt_present)
  1545. sl->redundant_pic_count = get_ue_golomb(&sl->gb);
  1546. if (sl->slice_type_nos == AV_PICTURE_TYPE_B)
  1547. sl->direct_spatial_mv_pred = get_bits1(&sl->gb);
  1548. ret = ff_h264_parse_ref_count(&sl->list_count, sl->ref_count,
  1549. &sl->gb, pps, sl->slice_type_nos,
  1550. picture_structure, h->avctx);
  1551. if (ret < 0)
  1552. return ret;
  1553. if (sl->slice_type_nos != AV_PICTURE_TYPE_I) {
  1554. ret = ff_h264_decode_ref_pic_list_reordering(sl, h->avctx);
  1555. if (ret < 0) {
  1556. sl->ref_count[1] = sl->ref_count[0] = 0;
  1557. return ret;
  1558. }
  1559. }
  1560. sl->pwt.use_weight = 0;
  1561. for (i = 0; i < 2; i++) {
  1562. sl->pwt.luma_weight_flag[i] = 0;
  1563. sl->pwt.chroma_weight_flag[i] = 0;
  1564. }
  1565. if ((pps->weighted_pred && sl->slice_type_nos == AV_PICTURE_TYPE_P) ||
  1566. (pps->weighted_bipred_idc == 1 &&
  1567. sl->slice_type_nos == AV_PICTURE_TYPE_B))
  1568. ff_h264_pred_weight_table(&sl->gb, sps, sl->ref_count,
  1569. sl->slice_type_nos, &sl->pwt, h->avctx);
  1570. sl->explicit_ref_marking = 0;
  1571. if (nal->ref_idc) {
  1572. ret = ff_h264_decode_ref_pic_marking(sl, &sl->gb, nal, h->avctx);
  1573. if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
  1574. return AVERROR_INVALIDDATA;
  1575. }
  1576. if (sl->slice_type_nos != AV_PICTURE_TYPE_I && pps->cabac) {
  1577. tmp = get_ue_golomb_31(&sl->gb);
  1578. if (tmp > 2) {
  1579. av_log(h->avctx, AV_LOG_ERROR, "cabac_init_idc %u overflow\n", tmp);
  1580. return AVERROR_INVALIDDATA;
  1581. }
  1582. sl->cabac_init_idc = tmp;
  1583. }
  1584. sl->last_qscale_diff = 0;
  1585. tmp = pps->init_qp + get_se_golomb(&sl->gb);
  1586. if (tmp > 51 + 6 * (sps->bit_depth_luma - 8)) {
  1587. av_log(h->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
  1588. return AVERROR_INVALIDDATA;
  1589. }
  1590. sl->qscale = tmp;
  1591. sl->chroma_qp[0] = get_chroma_qp(pps, 0, sl->qscale);
  1592. sl->chroma_qp[1] = get_chroma_qp(pps, 1, sl->qscale);
  1593. // FIXME qscale / qp ... stuff
  1594. if (sl->slice_type == AV_PICTURE_TYPE_SP)
  1595. get_bits1(&sl->gb); /* sp_for_switch_flag */
  1596. if (sl->slice_type == AV_PICTURE_TYPE_SP ||
  1597. sl->slice_type == AV_PICTURE_TYPE_SI)
  1598. get_se_golomb(&sl->gb); /* slice_qs_delta */
  1599. sl->deblocking_filter = 1;
  1600. sl->slice_alpha_c0_offset = 0;
  1601. sl->slice_beta_offset = 0;
  1602. if (pps->deblocking_filter_parameters_present) {
  1603. tmp = get_ue_golomb_31(&sl->gb);
  1604. if (tmp > 2) {
  1605. av_log(h->avctx, AV_LOG_ERROR,
  1606. "deblocking_filter_idc %u out of range\n", tmp);
  1607. return AVERROR_INVALIDDATA;
  1608. }
  1609. sl->deblocking_filter = tmp;
  1610. if (sl->deblocking_filter < 2)
  1611. sl->deblocking_filter ^= 1; // 1<->0
  1612. if (sl->deblocking_filter) {
  1613. sl->slice_alpha_c0_offset = get_se_golomb(&sl->gb) * 2;
  1614. sl->slice_beta_offset = get_se_golomb(&sl->gb) * 2;
  1615. if (sl->slice_alpha_c0_offset > 12 ||
  1616. sl->slice_alpha_c0_offset < -12 ||
  1617. sl->slice_beta_offset > 12 ||
  1618. sl->slice_beta_offset < -12) {
  1619. av_log(h->avctx, AV_LOG_ERROR,
  1620. "deblocking filter parameters %d %d out of range\n",
  1621. sl->slice_alpha_c0_offset, sl->slice_beta_offset);
  1622. return AVERROR_INVALIDDATA;
  1623. }
  1624. }
  1625. }
  1626. return 0;
  1627. }
  1628. /* do all the per-slice initialization needed before we can start decoding the
  1629. * actual MBs */
  1630. static int h264_slice_init(H264Context *h, H264SliceContext *sl,
  1631. const H2645NAL *nal)
  1632. {
  1633. int i, j, ret = 0;
  1634. if (h->picture_idr && nal->type != H264_NAL_IDR_SLICE) {
  1635. av_log(h->avctx, AV_LOG_ERROR, "Invalid mix of IDR and non-IDR slices\n");
  1636. return AVERROR_INVALIDDATA;
  1637. }
  1638. av_assert1(h->mb_num == h->mb_width * h->mb_height);
  1639. if (sl->first_mb_addr << FIELD_OR_MBAFF_PICTURE(h) >= h->mb_num ||
  1640. sl->first_mb_addr >= h->mb_num) {
  1641. av_log(h->avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
  1642. return AVERROR_INVALIDDATA;
  1643. }
  1644. sl->resync_mb_x = sl->mb_x = sl->first_mb_addr % h->mb_width;
  1645. sl->resync_mb_y = sl->mb_y = (sl->first_mb_addr / h->mb_width) <<
  1646. FIELD_OR_MBAFF_PICTURE(h);
  1647. if (h->picture_structure == PICT_BOTTOM_FIELD)
  1648. sl->resync_mb_y = sl->mb_y = sl->mb_y + 1;
  1649. av_assert1(sl->mb_y < h->mb_height);
  1650. ret = ff_h264_build_ref_list(h, sl);
  1651. if (ret < 0)
  1652. return ret;
  1653. if (h->ps.pps->weighted_bipred_idc == 2 &&
  1654. sl->slice_type_nos == AV_PICTURE_TYPE_B) {
  1655. implicit_weight_table(h, sl, -1);
  1656. if (FRAME_MBAFF(h)) {
  1657. implicit_weight_table(h, sl, 0);
  1658. implicit_weight_table(h, sl, 1);
  1659. }
  1660. }
  1661. if (sl->slice_type_nos == AV_PICTURE_TYPE_B && !sl->direct_spatial_mv_pred)
  1662. ff_h264_direct_dist_scale_factor(h, sl);
  1663. ff_h264_direct_ref_list_init(h, sl);
  1664. if (h->avctx->skip_loop_filter >= AVDISCARD_ALL ||
  1665. (h->avctx->skip_loop_filter >= AVDISCARD_NONKEY &&
  1666. h->nal_unit_type != H264_NAL_IDR_SLICE) ||
  1667. (h->avctx->skip_loop_filter >= AVDISCARD_NONINTRA &&
  1668. sl->slice_type_nos != AV_PICTURE_TYPE_I) ||
  1669. (h->avctx->skip_loop_filter >= AVDISCARD_BIDIR &&
  1670. sl->slice_type_nos == AV_PICTURE_TYPE_B) ||
  1671. (h->avctx->skip_loop_filter >= AVDISCARD_NONREF &&
  1672. nal->ref_idc == 0))
  1673. sl->deblocking_filter = 0;
  1674. if (sl->deblocking_filter == 1 && h->nb_slice_ctx > 1) {
  1675. if (h->avctx->flags2 & AV_CODEC_FLAG2_FAST) {
  1676. /* Cheat slightly for speed:
  1677. * Do not bother to deblock across slices. */
  1678. sl->deblocking_filter = 2;
  1679. } else {
  1680. h->postpone_filter = 1;
  1681. }
  1682. }
  1683. sl->qp_thresh = 15 -
  1684. FFMIN(sl->slice_alpha_c0_offset, sl->slice_beta_offset) -
  1685. FFMAX3(0,
  1686. h->ps.pps->chroma_qp_index_offset[0],
  1687. h->ps.pps->chroma_qp_index_offset[1]) +
  1688. 6 * (h->ps.sps->bit_depth_luma - 8);
  1689. sl->slice_num = ++h->current_slice;
  1690. if (sl->slice_num)
  1691. h->slice_row[(sl->slice_num-1)&(MAX_SLICES-1)]= sl->resync_mb_y;
  1692. if ( h->slice_row[sl->slice_num&(MAX_SLICES-1)] + 3 >= sl->resync_mb_y
  1693. && h->slice_row[sl->slice_num&(MAX_SLICES-1)] <= sl->resync_mb_y
  1694. && sl->slice_num >= MAX_SLICES) {
  1695. //in case of ASO this check needs to be updated depending on how we decide to assign slice numbers in this case
  1696. av_log(h->avctx, AV_LOG_WARNING, "Possibly too many slices (%d >= %d), increase MAX_SLICES and recompile if there are artifacts\n", sl->slice_num, MAX_SLICES);
  1697. }
  1698. for (j = 0; j < 2; j++) {
  1699. int id_list[16];
  1700. int *ref2frm = h->ref2frm[sl->slice_num & (MAX_SLICES - 1)][j];
  1701. for (i = 0; i < 16; i++) {
  1702. id_list[i] = 60;
  1703. if (j < sl->list_count && i < sl->ref_count[j] &&
  1704. sl->ref_list[j][i].parent->f->buf[0]) {
  1705. int k;
  1706. AVBuffer *buf = sl->ref_list[j][i].parent->f->buf[0]->buffer;
  1707. for (k = 0; k < h->short_ref_count; k++)
  1708. if (h->short_ref[k]->f->buf[0]->buffer == buf) {
  1709. id_list[i] = k;
  1710. break;
  1711. }
  1712. for (k = 0; k < h->long_ref_count; k++)
  1713. if (h->long_ref[k] && h->long_ref[k]->f->buf[0]->buffer == buf) {
  1714. id_list[i] = h->short_ref_count + k;
  1715. break;
  1716. }
  1717. }
  1718. }
  1719. ref2frm[0] =
  1720. ref2frm[1] = -1;
  1721. for (i = 0; i < 16; i++)
  1722. ref2frm[i + 2] = 4 * id_list[i] + (sl->ref_list[j][i].reference & 3);
  1723. ref2frm[18 + 0] =
  1724. ref2frm[18 + 1] = -1;
  1725. for (i = 16; i < 48; i++)
  1726. ref2frm[i + 4] = 4 * id_list[(i - 16) >> 1] +
  1727. (sl->ref_list[j][i].reference & 3);
  1728. }
  1729. if (h->avctx->debug & FF_DEBUG_PICT_INFO) {
  1730. av_log(h->avctx, AV_LOG_DEBUG,
  1731. "slice:%d %s mb:%d %c%s%s frame:%d poc:%d/%d ref:%d/%d qp:%d loop:%d:%d:%d weight:%d%s %s\n",
  1732. sl->slice_num,
  1733. (h->picture_structure == PICT_FRAME ? "F" : h->picture_structure == PICT_TOP_FIELD ? "T" : "B"),
  1734. sl->mb_y * h->mb_width + sl->mb_x,
  1735. av_get_picture_type_char(sl->slice_type),
  1736. sl->slice_type_fixed ? " fix" : "",
  1737. nal->type == H264_NAL_IDR_SLICE ? " IDR" : "",
  1738. h->poc.frame_num,
  1739. h->cur_pic_ptr->field_poc[0],
  1740. h->cur_pic_ptr->field_poc[1],
  1741. sl->ref_count[0], sl->ref_count[1],
  1742. sl->qscale,
  1743. sl->deblocking_filter,
  1744. sl->slice_alpha_c0_offset, sl->slice_beta_offset,
  1745. sl->pwt.use_weight,
  1746. sl->pwt.use_weight == 1 && sl->pwt.use_weight_chroma ? "c" : "",
  1747. sl->slice_type == AV_PICTURE_TYPE_B ? (sl->direct_spatial_mv_pred ? "SPAT" : "TEMP") : "");
  1748. }
  1749. return 0;
  1750. }
  1751. int ff_h264_queue_decode_slice(H264Context *h, const H2645NAL *nal)
  1752. {
  1753. H264SliceContext *sl = h->slice_ctx + h->nb_slice_ctx_queued;
  1754. int first_slice = sl == h->slice_ctx && !h->current_slice;
  1755. int ret;
  1756. sl->gb = nal->gb;
  1757. ret = h264_slice_header_parse(h, sl, nal);
  1758. if (ret < 0)
  1759. return ret;
  1760. // discard redundant pictures
  1761. if (sl->redundant_pic_count > 0) {
  1762. sl->ref_count[0] = sl->ref_count[1] = 0;
  1763. return 0;
  1764. }
  1765. if (sl->first_mb_addr == 0 || !h->current_slice) {
  1766. if (h->setup_finished) {
  1767. av_log(h->avctx, AV_LOG_ERROR, "Too many fields\n");
  1768. return AVERROR_INVALIDDATA;
  1769. }
  1770. }
  1771. if (sl->first_mb_addr == 0) { // FIXME better field boundary detection
  1772. if (h->current_slice) {
  1773. // this slice starts a new field
  1774. // first decode any pending queued slices
  1775. if (h->nb_slice_ctx_queued) {
  1776. H264SliceContext tmp_ctx;
  1777. ret = ff_h264_execute_decode_slices(h);
  1778. if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
  1779. return ret;
  1780. memcpy(&tmp_ctx, h->slice_ctx, sizeof(tmp_ctx));
  1781. memcpy(h->slice_ctx, sl, sizeof(tmp_ctx));
  1782. memcpy(sl, &tmp_ctx, sizeof(tmp_ctx));
  1783. sl = h->slice_ctx;
  1784. }
  1785. if (h->cur_pic_ptr && FIELD_PICTURE(h) && h->first_field) {
  1786. ret = ff_h264_field_end(h, h->slice_ctx, 1);
  1787. if (ret < 0)
  1788. return ret;
  1789. } else if (h->cur_pic_ptr && !FIELD_PICTURE(h) && !h->first_field && h->nal_unit_type == H264_NAL_IDR_SLICE) {
  1790. av_log(h, AV_LOG_WARNING, "Broken frame packetizing\n");
  1791. ret = ff_h264_field_end(h, h->slice_ctx, 1);
  1792. ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 0);
  1793. ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 1);
  1794. h->cur_pic_ptr = NULL;
  1795. if (ret < 0)
  1796. return ret;
  1797. } else
  1798. return AVERROR_INVALIDDATA;
  1799. }
  1800. if (!h->first_field) {
  1801. if (h->cur_pic_ptr && !h->droppable) {
  1802. ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
  1803. h->picture_structure == PICT_BOTTOM_FIELD);
  1804. }
  1805. h->cur_pic_ptr = NULL;
  1806. }
  1807. }
  1808. if (!h->current_slice)
  1809. av_assert0(sl == h->slice_ctx);
  1810. if (h->current_slice == 0 && !h->first_field) {
  1811. if (
  1812. (h->avctx->skip_frame >= AVDISCARD_NONREF && !h->nal_ref_idc) ||
  1813. (h->avctx->skip_frame >= AVDISCARD_BIDIR && sl->slice_type_nos == AV_PICTURE_TYPE_B) ||
  1814. (h->avctx->skip_frame >= AVDISCARD_NONINTRA && sl->slice_type_nos != AV_PICTURE_TYPE_I) ||
  1815. (h->avctx->skip_frame >= AVDISCARD_NONKEY && h->nal_unit_type != H264_NAL_IDR_SLICE && h->sei.recovery_point.recovery_frame_cnt < 0) ||
  1816. h->avctx->skip_frame >= AVDISCARD_ALL) {
  1817. return 0;
  1818. }
  1819. }
  1820. if (!first_slice) {
  1821. const PPS *pps = (const PPS*)h->ps.pps_list[sl->pps_id]->data;
  1822. if (h->ps.pps->sps_id != pps->sps_id ||
  1823. h->ps.pps->transform_8x8_mode != pps->transform_8x8_mode /*||
  1824. (h->setup_finished && h->ps.pps != pps)*/) {
  1825. av_log(h->avctx, AV_LOG_ERROR, "PPS changed between slices\n");
  1826. return AVERROR_INVALIDDATA;
  1827. }
  1828. if (h->ps.sps != (const SPS*)h->ps.sps_list[h->ps.pps->sps_id]->data) {
  1829. av_log(h->avctx, AV_LOG_ERROR,
  1830. "SPS changed in the middle of the frame\n");
  1831. return AVERROR_INVALIDDATA;
  1832. }
  1833. }
  1834. if (h->current_slice == 0) {
  1835. ret = h264_field_start(h, sl, nal, first_slice);
  1836. if (ret < 0)
  1837. return ret;
  1838. } else {
  1839. if (h->picture_structure != sl->picture_structure ||
  1840. h->droppable != (nal->ref_idc == 0)) {
  1841. av_log(h->avctx, AV_LOG_ERROR,
  1842. "Changing field mode (%d -> %d) between slices is not allowed\n",
  1843. h->picture_structure, sl->picture_structure);
  1844. return AVERROR_INVALIDDATA;
  1845. } else if (!h->cur_pic_ptr) {
  1846. av_log(h->avctx, AV_LOG_ERROR,
  1847. "unset cur_pic_ptr on slice %d\n",
  1848. h->current_slice + 1);
  1849. return AVERROR_INVALIDDATA;
  1850. }
  1851. }
  1852. ret = h264_slice_init(h, sl, nal);
  1853. if (ret < 0)
  1854. return ret;
  1855. h->nb_slice_ctx_queued++;
  1856. return 0;
  1857. }
  1858. int ff_h264_get_slice_type(const H264SliceContext *sl)
  1859. {
  1860. switch (sl->slice_type) {
  1861. case AV_PICTURE_TYPE_P:
  1862. return 0;
  1863. case AV_PICTURE_TYPE_B:
  1864. return 1;
  1865. case AV_PICTURE_TYPE_I:
  1866. return 2;
  1867. case AV_PICTURE_TYPE_SP:
  1868. return 3;
  1869. case AV_PICTURE_TYPE_SI:
  1870. return 4;
  1871. default:
  1872. return AVERROR_INVALIDDATA;
  1873. }
  1874. }
  1875. static av_always_inline void fill_filter_caches_inter(const H264Context *h,
  1876. H264SliceContext *sl,
  1877. int mb_type, int top_xy,
  1878. int left_xy[LEFT_MBS],
  1879. int top_type,
  1880. int left_type[LEFT_MBS],
  1881. int mb_xy, int list)
  1882. {
  1883. int b_stride = h->b_stride;
  1884. int16_t(*mv_dst)[2] = &sl->mv_cache[list][scan8[0]];
  1885. int8_t *ref_cache = &sl->ref_cache[list][scan8[0]];
  1886. if (IS_INTER(mb_type) || IS_DIRECT(mb_type)) {
  1887. if (USES_LIST(top_type, list)) {
  1888. const int b_xy = h->mb2b_xy[top_xy] + 3 * b_stride;
  1889. const int b8_xy = 4 * top_xy + 2;
  1890. const int *ref2frm = &h->ref2frm[h->slice_table[top_xy] & (MAX_SLICES - 1)][list][(MB_MBAFF(sl) ? 20 : 2)];
  1891. AV_COPY128(mv_dst - 1 * 8, h->cur_pic.motion_val[list][b_xy + 0]);
  1892. ref_cache[0 - 1 * 8] =
  1893. ref_cache[1 - 1 * 8] = ref2frm[h->cur_pic.ref_index[list][b8_xy + 0]];
  1894. ref_cache[2 - 1 * 8] =
  1895. ref_cache[3 - 1 * 8] = ref2frm[h->cur_pic.ref_index[list][b8_xy + 1]];
  1896. } else {
  1897. AV_ZERO128(mv_dst - 1 * 8);
  1898. AV_WN32A(&ref_cache[0 - 1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
  1899. }
  1900. if (!IS_INTERLACED(mb_type ^ left_type[LTOP])) {
  1901. if (USES_LIST(left_type[LTOP], list)) {
  1902. const int b_xy = h->mb2b_xy[left_xy[LTOP]] + 3;
  1903. const int b8_xy = 4 * left_xy[LTOP] + 1;
  1904. const int *ref2frm = &h->ref2frm[h->slice_table[left_xy[LTOP]] & (MAX_SLICES - 1)][list][(MB_MBAFF(sl) ? 20 : 2)];
  1905. AV_COPY32(mv_dst - 1 + 0, h->cur_pic.motion_val[list][b_xy + b_stride * 0]);
  1906. AV_COPY32(mv_dst - 1 + 8, h->cur_pic.motion_val[list][b_xy + b_stride * 1]);
  1907. AV_COPY32(mv_dst - 1 + 16, h->cur_pic.motion_val[list][b_xy + b_stride * 2]);
  1908. AV_COPY32(mv_dst - 1 + 24, h->cur_pic.motion_val[list][b_xy + b_stride * 3]);
  1909. ref_cache[-1 + 0] =
  1910. ref_cache[-1 + 8] = ref2frm[h->cur_pic.ref_index[list][b8_xy + 2 * 0]];
  1911. ref_cache[-1 + 16] =
  1912. ref_cache[-1 + 24] = ref2frm[h->cur_pic.ref_index[list][b8_xy + 2 * 1]];
  1913. } else {
  1914. AV_ZERO32(mv_dst - 1 + 0);
  1915. AV_ZERO32(mv_dst - 1 + 8);
  1916. AV_ZERO32(mv_dst - 1 + 16);
  1917. AV_ZERO32(mv_dst - 1 + 24);
  1918. ref_cache[-1 + 0] =
  1919. ref_cache[-1 + 8] =
  1920. ref_cache[-1 + 16] =
  1921. ref_cache[-1 + 24] = LIST_NOT_USED;
  1922. }
  1923. }
  1924. }
  1925. if (!USES_LIST(mb_type, list)) {
  1926. fill_rectangle(mv_dst, 4, 4, 8, pack16to32(0, 0), 4);
  1927. AV_WN32A(&ref_cache[0 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
  1928. AV_WN32A(&ref_cache[1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
  1929. AV_WN32A(&ref_cache[2 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
  1930. AV_WN32A(&ref_cache[3 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
  1931. return;
  1932. }
  1933. {
  1934. int8_t *ref = &h->cur_pic.ref_index[list][4 * mb_xy];
  1935. const int *ref2frm = &h->ref2frm[sl->slice_num & (MAX_SLICES - 1)][list][(MB_MBAFF(sl) ? 20 : 2)];
  1936. uint32_t ref01 = (pack16to32(ref2frm[ref[0]], ref2frm[ref[1]]) & 0x00FF00FF) * 0x0101;
  1937. uint32_t ref23 = (pack16to32(ref2frm[ref[2]], ref2frm[ref[3]]) & 0x00FF00FF) * 0x0101;
  1938. AV_WN32A(&ref_cache[0 * 8], ref01);
  1939. AV_WN32A(&ref_cache[1 * 8], ref01);
  1940. AV_WN32A(&ref_cache[2 * 8], ref23);
  1941. AV_WN32A(&ref_cache[3 * 8], ref23);
  1942. }
  1943. {
  1944. int16_t(*mv_src)[2] = &h->cur_pic.motion_val[list][4 * sl->mb_x + 4 * sl->mb_y * b_stride];
  1945. AV_COPY128(mv_dst + 8 * 0, mv_src + 0 * b_stride);
  1946. AV_COPY128(mv_dst + 8 * 1, mv_src + 1 * b_stride);
  1947. AV_COPY128(mv_dst + 8 * 2, mv_src + 2 * b_stride);
  1948. AV_COPY128(mv_dst + 8 * 3, mv_src + 3 * b_stride);
  1949. }
  1950. }
  1951. /**
  1952. * @return non zero if the loop filter can be skipped
  1953. */
  1954. static int fill_filter_caches(const H264Context *h, H264SliceContext *sl, int mb_type)
  1955. {
  1956. const int mb_xy = sl->mb_xy;
  1957. int top_xy, left_xy[LEFT_MBS];
  1958. int top_type, left_type[LEFT_MBS];
  1959. uint8_t *nnz;
  1960. uint8_t *nnz_cache;
  1961. top_xy = mb_xy - (h->mb_stride << MB_FIELD(sl));
  1962. left_xy[LBOT] = left_xy[LTOP] = mb_xy - 1;
  1963. if (FRAME_MBAFF(h)) {
  1964. const int left_mb_field_flag = IS_INTERLACED(h->cur_pic.mb_type[mb_xy - 1]);
  1965. const int curr_mb_field_flag = IS_INTERLACED(mb_type);
  1966. if (sl->mb_y & 1) {
  1967. if (left_mb_field_flag != curr_mb_field_flag)
  1968. left_xy[LTOP] -= h->mb_stride;
  1969. } else {
  1970. if (curr_mb_field_flag)
  1971. top_xy += h->mb_stride &
  1972. (((h->cur_pic.mb_type[top_xy] >> 7) & 1) - 1);
  1973. if (left_mb_field_flag != curr_mb_field_flag)
  1974. left_xy[LBOT] += h->mb_stride;
  1975. }
  1976. }
  1977. sl->top_mb_xy = top_xy;
  1978. sl->left_mb_xy[LTOP] = left_xy[LTOP];
  1979. sl->left_mb_xy[LBOT] = left_xy[LBOT];
  1980. {
  1981. /* For sufficiently low qp, filtering wouldn't do anything.
  1982. * This is a conservative estimate: could also check beta_offset
  1983. * and more accurate chroma_qp. */
  1984. int qp_thresh = sl->qp_thresh; // FIXME strictly we should store qp_thresh for each mb of a slice
  1985. int qp = h->cur_pic.qscale_table[mb_xy];
  1986. if (qp <= qp_thresh &&
  1987. (left_xy[LTOP] < 0 ||
  1988. ((qp + h->cur_pic.qscale_table[left_xy[LTOP]] + 1) >> 1) <= qp_thresh) &&
  1989. (top_xy < 0 ||
  1990. ((qp + h->cur_pic.qscale_table[top_xy] + 1) >> 1) <= qp_thresh)) {
  1991. if (!FRAME_MBAFF(h))
  1992. return 1;
  1993. if ((left_xy[LTOP] < 0 ||
  1994. ((qp + h->cur_pic.qscale_table[left_xy[LBOT]] + 1) >> 1) <= qp_thresh) &&
  1995. (top_xy < h->mb_stride ||
  1996. ((qp + h->cur_pic.qscale_table[top_xy - h->mb_stride] + 1) >> 1) <= qp_thresh))
  1997. return 1;
  1998. }
  1999. }
  2000. top_type = h->cur_pic.mb_type[top_xy];
  2001. left_type[LTOP] = h->cur_pic.mb_type[left_xy[LTOP]];
  2002. left_type[LBOT] = h->cur_pic.mb_type[left_xy[LBOT]];
  2003. if (sl->deblocking_filter == 2) {
  2004. if (h->slice_table[top_xy] != sl->slice_num)
  2005. top_type = 0;
  2006. if (h->slice_table[left_xy[LBOT]] != sl->slice_num)
  2007. left_type[LTOP] = left_type[LBOT] = 0;
  2008. } else {
  2009. if (h->slice_table[top_xy] == 0xFFFF)
  2010. top_type = 0;
  2011. if (h->slice_table[left_xy[LBOT]] == 0xFFFF)
  2012. left_type[LTOP] = left_type[LBOT] = 0;
  2013. }
  2014. sl->top_type = top_type;
  2015. sl->left_type[LTOP] = left_type[LTOP];
  2016. sl->left_type[LBOT] = left_type[LBOT];
  2017. if (IS_INTRA(mb_type))
  2018. return 0;
  2019. fill_filter_caches_inter(h, sl, mb_type, top_xy, left_xy,
  2020. top_type, left_type, mb_xy, 0);
  2021. if (sl->list_count == 2)
  2022. fill_filter_caches_inter(h, sl, mb_type, top_xy, left_xy,
  2023. top_type, left_type, mb_xy, 1);
  2024. nnz = h->non_zero_count[mb_xy];
  2025. nnz_cache = sl->non_zero_count_cache;
  2026. AV_COPY32(&nnz_cache[4 + 8 * 1], &nnz[0]);
  2027. AV_COPY32(&nnz_cache[4 + 8 * 2], &nnz[4]);
  2028. AV_COPY32(&nnz_cache[4 + 8 * 3], &nnz[8]);
  2029. AV_COPY32(&nnz_cache[4 + 8 * 4], &nnz[12]);
  2030. sl->cbp = h->cbp_table[mb_xy];
  2031. if (top_type) {
  2032. nnz = h->non_zero_count[top_xy];
  2033. AV_COPY32(&nnz_cache[4 + 8 * 0], &nnz[3 * 4]);
  2034. }
  2035. if (left_type[LTOP]) {
  2036. nnz = h->non_zero_count[left_xy[LTOP]];
  2037. nnz_cache[3 + 8 * 1] = nnz[3 + 0 * 4];
  2038. nnz_cache[3 + 8 * 2] = nnz[3 + 1 * 4];
  2039. nnz_cache[3 + 8 * 3] = nnz[3 + 2 * 4];
  2040. nnz_cache[3 + 8 * 4] = nnz[3 + 3 * 4];
  2041. }
  2042. /* CAVLC 8x8dct requires NNZ values for residual decoding that differ
  2043. * from what the loop filter needs */
  2044. if (!CABAC(h) && h->ps.pps->transform_8x8_mode) {
  2045. if (IS_8x8DCT(top_type)) {
  2046. nnz_cache[4 + 8 * 0] =
  2047. nnz_cache[5 + 8 * 0] = (h->cbp_table[top_xy] & 0x4000) >> 12;
  2048. nnz_cache[6 + 8 * 0] =
  2049. nnz_cache[7 + 8 * 0] = (h->cbp_table[top_xy] & 0x8000) >> 12;
  2050. }
  2051. if (IS_8x8DCT(left_type[LTOP])) {
  2052. nnz_cache[3 + 8 * 1] =
  2053. nnz_cache[3 + 8 * 2] = (h->cbp_table[left_xy[LTOP]] & 0x2000) >> 12; // FIXME check MBAFF
  2054. }
  2055. if (IS_8x8DCT(left_type[LBOT])) {
  2056. nnz_cache[3 + 8 * 3] =
  2057. nnz_cache[3 + 8 * 4] = (h->cbp_table[left_xy[LBOT]] & 0x8000) >> 12; // FIXME check MBAFF
  2058. }
  2059. if (IS_8x8DCT(mb_type)) {
  2060. nnz_cache[scan8[0]] =
  2061. nnz_cache[scan8[1]] =
  2062. nnz_cache[scan8[2]] =
  2063. nnz_cache[scan8[3]] = (sl->cbp & 0x1000) >> 12;
  2064. nnz_cache[scan8[0 + 4]] =
  2065. nnz_cache[scan8[1 + 4]] =
  2066. nnz_cache[scan8[2 + 4]] =
  2067. nnz_cache[scan8[3 + 4]] = (sl->cbp & 0x2000) >> 12;
  2068. nnz_cache[scan8[0 + 8]] =
  2069. nnz_cache[scan8[1 + 8]] =
  2070. nnz_cache[scan8[2 + 8]] =
  2071. nnz_cache[scan8[3 + 8]] = (sl->cbp & 0x4000) >> 12;
  2072. nnz_cache[scan8[0 + 12]] =
  2073. nnz_cache[scan8[1 + 12]] =
  2074. nnz_cache[scan8[2 + 12]] =
  2075. nnz_cache[scan8[3 + 12]] = (sl->cbp & 0x8000) >> 12;
  2076. }
  2077. }
  2078. return 0;
  2079. }
  2080. static void loop_filter(const H264Context *h, H264SliceContext *sl, int start_x, int end_x)
  2081. {
  2082. uint8_t *dest_y, *dest_cb, *dest_cr;
  2083. int linesize, uvlinesize, mb_x, mb_y;
  2084. const int end_mb_y = sl->mb_y + FRAME_MBAFF(h);
  2085. const int old_slice_type = sl->slice_type;
  2086. const int pixel_shift = h->pixel_shift;
  2087. const int block_h = 16 >> h->chroma_y_shift;
  2088. if (h->postpone_filter)
  2089. return;
  2090. if (sl->deblocking_filter) {
  2091. for (mb_x = start_x; mb_x < end_x; mb_x++)
  2092. for (mb_y = end_mb_y - FRAME_MBAFF(h); mb_y <= end_mb_y; mb_y++) {
  2093. int mb_xy, mb_type;
  2094. mb_xy = sl->mb_xy = mb_x + mb_y * h->mb_stride;
  2095. mb_type = h->cur_pic.mb_type[mb_xy];
  2096. if (FRAME_MBAFF(h))
  2097. sl->mb_mbaff =
  2098. sl->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
  2099. sl->mb_x = mb_x;
  2100. sl->mb_y = mb_y;
  2101. dest_y = h->cur_pic.f->data[0] +
  2102. ((mb_x << pixel_shift) + mb_y * sl->linesize) * 16;
  2103. dest_cb = h->cur_pic.f->data[1] +
  2104. (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
  2105. mb_y * sl->uvlinesize * block_h;
  2106. dest_cr = h->cur_pic.f->data[2] +
  2107. (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
  2108. mb_y * sl->uvlinesize * block_h;
  2109. // FIXME simplify above
  2110. if (MB_FIELD(sl)) {
  2111. linesize = sl->mb_linesize = sl->linesize * 2;
  2112. uvlinesize = sl->mb_uvlinesize = sl->uvlinesize * 2;
  2113. if (mb_y & 1) { // FIXME move out of this function?
  2114. dest_y -= sl->linesize * 15;
  2115. dest_cb -= sl->uvlinesize * (block_h - 1);
  2116. dest_cr -= sl->uvlinesize * (block_h - 1);
  2117. }
  2118. } else {
  2119. linesize = sl->mb_linesize = sl->linesize;
  2120. uvlinesize = sl->mb_uvlinesize = sl->uvlinesize;
  2121. }
  2122. backup_mb_border(h, sl, dest_y, dest_cb, dest_cr, linesize,
  2123. uvlinesize, 0);
  2124. if (fill_filter_caches(h, sl, mb_type))
  2125. continue;
  2126. sl->chroma_qp[0] = get_chroma_qp(h->ps.pps, 0, h->cur_pic.qscale_table[mb_xy]);
  2127. sl->chroma_qp[1] = get_chroma_qp(h->ps.pps, 1, h->cur_pic.qscale_table[mb_xy]);
  2128. if (FRAME_MBAFF(h)) {
  2129. ff_h264_filter_mb(h, sl, mb_x, mb_y, dest_y, dest_cb, dest_cr,
  2130. linesize, uvlinesize);
  2131. } else {
  2132. ff_h264_filter_mb_fast(h, sl, mb_x, mb_y, dest_y, dest_cb,
  2133. dest_cr, linesize, uvlinesize);
  2134. }
  2135. }
  2136. }
  2137. sl->slice_type = old_slice_type;
  2138. sl->mb_x = end_x;
  2139. sl->mb_y = end_mb_y - FRAME_MBAFF(h);
  2140. sl->chroma_qp[0] = get_chroma_qp(h->ps.pps, 0, sl->qscale);
  2141. sl->chroma_qp[1] = get_chroma_qp(h->ps.pps, 1, sl->qscale);
  2142. }
  2143. static void predict_field_decoding_flag(const H264Context *h, H264SliceContext *sl)
  2144. {
  2145. const int mb_xy = sl->mb_x + sl->mb_y * h->mb_stride;
  2146. int mb_type = (h->slice_table[mb_xy - 1] == sl->slice_num) ?
  2147. h->cur_pic.mb_type[mb_xy - 1] :
  2148. (h->slice_table[mb_xy - h->mb_stride] == sl->slice_num) ?
  2149. h->cur_pic.mb_type[mb_xy - h->mb_stride] : 0;
  2150. sl->mb_mbaff = sl->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
  2151. }
  2152. /**
  2153. * Draw edges and report progress for the last MB row.
  2154. */
  2155. static void decode_finish_row(const H264Context *h, H264SliceContext *sl)
  2156. {
  2157. int top = 16 * (sl->mb_y >> FIELD_PICTURE(h));
  2158. int pic_height = 16 * h->mb_height >> FIELD_PICTURE(h);
  2159. int height = 16 << FRAME_MBAFF(h);
  2160. int deblock_border = (16 + 4) << FRAME_MBAFF(h);
  2161. if (sl->deblocking_filter) {
  2162. if ((top + height) >= pic_height)
  2163. height += deblock_border;
  2164. top -= deblock_border;
  2165. }
  2166. if (top >= pic_height || (top + height) < 0)
  2167. return;
  2168. height = FFMIN(height, pic_height - top);
  2169. if (top < 0) {
  2170. height = top + height;
  2171. top = 0;
  2172. }
  2173. ff_h264_draw_horiz_band(h, sl, top, height);
  2174. if (h->droppable || sl->h264->slice_ctx[0].er.error_occurred)
  2175. return;
  2176. ff_thread_report_progress(&h->cur_pic_ptr->tf, top + height - 1,
  2177. h->picture_structure == PICT_BOTTOM_FIELD);
  2178. }
  2179. static void er_add_slice(H264SliceContext *sl,
  2180. int startx, int starty,
  2181. int endx, int endy, int status)
  2182. {
  2183. if (!sl->h264->enable_er)
  2184. return;
  2185. if (CONFIG_ERROR_RESILIENCE) {
  2186. ERContext *er = &sl->h264->slice_ctx[0].er;
  2187. ff_er_add_slice(er, startx, starty, endx, endy, status);
  2188. }
  2189. }
  2190. static int decode_slice(struct AVCodecContext *avctx, void *arg)
  2191. {
  2192. H264SliceContext *sl = arg;
  2193. const H264Context *h = sl->h264;
  2194. int lf_x_start = sl->mb_x;
  2195. int orig_deblock = sl->deblocking_filter;
  2196. int ret;
  2197. sl->linesize = h->cur_pic_ptr->f->linesize[0];
  2198. sl->uvlinesize = h->cur_pic_ptr->f->linesize[1];
  2199. ret = alloc_scratch_buffers(sl, sl->linesize);
  2200. if (ret < 0)
  2201. return ret;
  2202. sl->mb_skip_run = -1;
  2203. av_assert0(h->block_offset[15] == (4 * ((scan8[15] - scan8[0]) & 7) << h->pixel_shift) + 4 * sl->linesize * ((scan8[15] - scan8[0]) >> 3));
  2204. if (h->postpone_filter)
  2205. sl->deblocking_filter = 0;
  2206. sl->is_complex = FRAME_MBAFF(h) || h->picture_structure != PICT_FRAME ||
  2207. (CONFIG_GRAY && (h->flags & AV_CODEC_FLAG_GRAY));
  2208. if (!(h->avctx->active_thread_type & FF_THREAD_SLICE) && h->picture_structure == PICT_FRAME && h->slice_ctx[0].er.error_status_table) {
  2209. const int start_i = av_clip(sl->resync_mb_x + sl->resync_mb_y * h->mb_width, 0, h->mb_num - 1);
  2210. if (start_i) {
  2211. int prev_status = h->slice_ctx[0].er.error_status_table[h->slice_ctx[0].er.mb_index2xy[start_i - 1]];
  2212. prev_status &= ~ VP_START;
  2213. if (prev_status != (ER_MV_END | ER_DC_END | ER_AC_END))
  2214. h->slice_ctx[0].er.error_occurred = 1;
  2215. }
  2216. }
  2217. if (h->ps.pps->cabac) {
  2218. /* realign */
  2219. align_get_bits(&sl->gb);
  2220. /* init cabac */
  2221. ret = ff_init_cabac_decoder(&sl->cabac,
  2222. sl->gb.buffer + get_bits_count(&sl->gb) / 8,
  2223. (get_bits_left(&sl->gb) + 7) / 8);
  2224. if (ret < 0)
  2225. return ret;
  2226. ff_h264_init_cabac_states(h, sl);
  2227. for (;;) {
  2228. // START_TIMER
  2229. int ret, eos;
  2230. if (sl->mb_x + sl->mb_y * h->mb_width >= sl->next_slice_idx) {
  2231. av_log(h->avctx, AV_LOG_ERROR, "Slice overlaps with next at %d\n",
  2232. sl->next_slice_idx);
  2233. er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
  2234. sl->mb_y, ER_MB_ERROR);
  2235. return AVERROR_INVALIDDATA;
  2236. }
  2237. ret = ff_h264_decode_mb_cabac(h, sl);
  2238. // STOP_TIMER("decode_mb_cabac")
  2239. if (ret >= 0)
  2240. ff_h264_hl_decode_mb(h, sl);
  2241. // FIXME optimal? or let mb_decode decode 16x32 ?
  2242. if (ret >= 0 && FRAME_MBAFF(h)) {
  2243. sl->mb_y++;
  2244. ret = ff_h264_decode_mb_cabac(h, sl);
  2245. if (ret >= 0)
  2246. ff_h264_hl_decode_mb(h, sl);
  2247. sl->mb_y--;
  2248. }
  2249. eos = get_cabac_terminate(&sl->cabac);
  2250. if ((h->workaround_bugs & FF_BUG_TRUNCATED) &&
  2251. sl->cabac.bytestream > sl->cabac.bytestream_end + 2) {
  2252. er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x - 1,
  2253. sl->mb_y, ER_MB_END);
  2254. if (sl->mb_x >= lf_x_start)
  2255. loop_filter(h, sl, lf_x_start, sl->mb_x + 1);
  2256. goto finish;
  2257. }
  2258. if (sl->cabac.bytestream > sl->cabac.bytestream_end + 2 )
  2259. av_log(h->avctx, AV_LOG_DEBUG, "bytestream overread %"PTRDIFF_SPECIFIER"\n", sl->cabac.bytestream_end - sl->cabac.bytestream);
  2260. if (ret < 0 || sl->cabac.bytestream > sl->cabac.bytestream_end + 4) {
  2261. av_log(h->avctx, AV_LOG_ERROR,
  2262. "error while decoding MB %d %d, bytestream %"PTRDIFF_SPECIFIER"\n",
  2263. sl->mb_x, sl->mb_y,
  2264. sl->cabac.bytestream_end - sl->cabac.bytestream);
  2265. er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
  2266. sl->mb_y, ER_MB_ERROR);
  2267. return AVERROR_INVALIDDATA;
  2268. }
  2269. if (++sl->mb_x >= h->mb_width) {
  2270. loop_filter(h, sl, lf_x_start, sl->mb_x);
  2271. sl->mb_x = lf_x_start = 0;
  2272. decode_finish_row(h, sl);
  2273. ++sl->mb_y;
  2274. if (FIELD_OR_MBAFF_PICTURE(h)) {
  2275. ++sl->mb_y;
  2276. if (FRAME_MBAFF(h) && sl->mb_y < h->mb_height)
  2277. predict_field_decoding_flag(h, sl);
  2278. }
  2279. }
  2280. if (eos || sl->mb_y >= h->mb_height) {
  2281. ff_tlog(h->avctx, "slice end %d %d\n",
  2282. get_bits_count(&sl->gb), sl->gb.size_in_bits);
  2283. er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x - 1,
  2284. sl->mb_y, ER_MB_END);
  2285. if (sl->mb_x > lf_x_start)
  2286. loop_filter(h, sl, lf_x_start, sl->mb_x);
  2287. goto finish;
  2288. }
  2289. }
  2290. } else {
  2291. for (;;) {
  2292. int ret;
  2293. if (sl->mb_x + sl->mb_y * h->mb_width >= sl->next_slice_idx) {
  2294. av_log(h->avctx, AV_LOG_ERROR, "Slice overlaps with next at %d\n",
  2295. sl->next_slice_idx);
  2296. er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
  2297. sl->mb_y, ER_MB_ERROR);
  2298. return AVERROR_INVALIDDATA;
  2299. }
  2300. ret = ff_h264_decode_mb_cavlc(h, sl);
  2301. if (ret >= 0)
  2302. ff_h264_hl_decode_mb(h, sl);
  2303. // FIXME optimal? or let mb_decode decode 16x32 ?
  2304. if (ret >= 0 && FRAME_MBAFF(h)) {
  2305. sl->mb_y++;
  2306. ret = ff_h264_decode_mb_cavlc(h, sl);
  2307. if (ret >= 0)
  2308. ff_h264_hl_decode_mb(h, sl);
  2309. sl->mb_y--;
  2310. }
  2311. if (ret < 0) {
  2312. av_log(h->avctx, AV_LOG_ERROR,
  2313. "error while decoding MB %d %d\n", sl->mb_x, sl->mb_y);
  2314. er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
  2315. sl->mb_y, ER_MB_ERROR);
  2316. return ret;
  2317. }
  2318. if (++sl->mb_x >= h->mb_width) {
  2319. loop_filter(h, sl, lf_x_start, sl->mb_x);
  2320. sl->mb_x = lf_x_start = 0;
  2321. decode_finish_row(h, sl);
  2322. ++sl->mb_y;
  2323. if (FIELD_OR_MBAFF_PICTURE(h)) {
  2324. ++sl->mb_y;
  2325. if (FRAME_MBAFF(h) && sl->mb_y < h->mb_height)
  2326. predict_field_decoding_flag(h, sl);
  2327. }
  2328. if (sl->mb_y >= h->mb_height) {
  2329. ff_tlog(h->avctx, "slice end %d %d\n",
  2330. get_bits_count(&sl->gb), sl->gb.size_in_bits);
  2331. if ( get_bits_left(&sl->gb) == 0
  2332. || get_bits_left(&sl->gb) > 0 && !(h->avctx->err_recognition & AV_EF_AGGRESSIVE)) {
  2333. er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
  2334. sl->mb_x - 1, sl->mb_y, ER_MB_END);
  2335. goto finish;
  2336. } else {
  2337. er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
  2338. sl->mb_x, sl->mb_y, ER_MB_END);
  2339. return AVERROR_INVALIDDATA;
  2340. }
  2341. }
  2342. }
  2343. if (get_bits_left(&sl->gb) <= 0 && sl->mb_skip_run <= 0) {
  2344. ff_tlog(h->avctx, "slice end %d %d\n",
  2345. get_bits_count(&sl->gb), sl->gb.size_in_bits);
  2346. if (get_bits_left(&sl->gb) == 0) {
  2347. er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
  2348. sl->mb_x - 1, sl->mb_y, ER_MB_END);
  2349. if (sl->mb_x > lf_x_start)
  2350. loop_filter(h, sl, lf_x_start, sl->mb_x);
  2351. goto finish;
  2352. } else {
  2353. er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
  2354. sl->mb_y, ER_MB_ERROR);
  2355. return AVERROR_INVALIDDATA;
  2356. }
  2357. }
  2358. }
  2359. }
  2360. finish:
  2361. sl->deblocking_filter = orig_deblock;
  2362. return 0;
  2363. }
  2364. /**
  2365. * Call decode_slice() for each context.
  2366. *
  2367. * @param h h264 master context
  2368. */
  2369. int ff_h264_execute_decode_slices(H264Context *h)
  2370. {
  2371. AVCodecContext *const avctx = h->avctx;
  2372. H264SliceContext *sl;
  2373. int context_count = h->nb_slice_ctx_queued;
  2374. int ret = 0;
  2375. int i, j;
  2376. h->slice_ctx[0].next_slice_idx = INT_MAX;
  2377. if (h->avctx->hwaccel || context_count < 1
  2378. #if FF_API_CAP_VDPAU
  2379. || h->avctx->codec->capabilities & AV_CODEC_CAP_HWACCEL_VDPAU
  2380. #endif
  2381. )
  2382. return 0;
  2383. av_assert0(context_count && h->slice_ctx[context_count - 1].mb_y < h->mb_height);
  2384. if (context_count == 1) {
  2385. h->slice_ctx[0].next_slice_idx = h->mb_width * h->mb_height;
  2386. h->postpone_filter = 0;
  2387. ret = decode_slice(avctx, &h->slice_ctx[0]);
  2388. h->mb_y = h->slice_ctx[0].mb_y;
  2389. if (ret < 0)
  2390. goto finish;
  2391. } else {
  2392. av_assert0(context_count > 0);
  2393. for (i = 0; i < context_count; i++) {
  2394. int next_slice_idx = h->mb_width * h->mb_height;
  2395. int slice_idx;
  2396. sl = &h->slice_ctx[i];
  2397. if (CONFIG_ERROR_RESILIENCE) {
  2398. sl->er.error_count = 0;
  2399. }
  2400. /* make sure none of those slices overlap */
  2401. slice_idx = sl->mb_y * h->mb_width + sl->mb_x;
  2402. for (j = 0; j < context_count; j++) {
  2403. H264SliceContext *sl2 = &h->slice_ctx[j];
  2404. int slice_idx2 = sl2->mb_y * h->mb_width + sl2->mb_x;
  2405. if (i == j || slice_idx2 < slice_idx)
  2406. continue;
  2407. next_slice_idx = FFMIN(next_slice_idx, slice_idx2);
  2408. }
  2409. sl->next_slice_idx = next_slice_idx;
  2410. }
  2411. avctx->execute(avctx, decode_slice, h->slice_ctx,
  2412. NULL, context_count, sizeof(h->slice_ctx[0]));
  2413. /* pull back stuff from slices to master context */
  2414. sl = &h->slice_ctx[context_count - 1];
  2415. h->mb_y = sl->mb_y;
  2416. if (CONFIG_ERROR_RESILIENCE) {
  2417. for (i = 1; i < context_count; i++)
  2418. h->slice_ctx[0].er.error_count += h->slice_ctx[i].er.error_count;
  2419. }
  2420. if (h->postpone_filter) {
  2421. h->postpone_filter = 0;
  2422. for (i = 0; i < context_count; i++) {
  2423. int y_end, x_end;
  2424. sl = &h->slice_ctx[i];
  2425. y_end = FFMIN(sl->mb_y + 1, h->mb_height);
  2426. x_end = (sl->mb_y >= h->mb_height) ? h->mb_width : sl->mb_x;
  2427. for (j = sl->resync_mb_y; j < y_end; j += 1 + FIELD_OR_MBAFF_PICTURE(h)) {
  2428. sl->mb_y = j;
  2429. loop_filter(h, sl, j > sl->resync_mb_y ? 0 : sl->resync_mb_x,
  2430. j == y_end - 1 ? x_end : h->mb_width);
  2431. }
  2432. }
  2433. }
  2434. }
  2435. finish:
  2436. h->nb_slice_ctx_queued = 0;
  2437. return ret;
  2438. }