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.

2880 lines
108KB

  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 "internal.h"
  31. #include "cabac.h"
  32. #include "cabac_functions.h"
  33. #include "error_resilience.h"
  34. #include "avcodec.h"
  35. #include "h264.h"
  36. #include "h264dec.h"
  37. #include "h264data.h"
  38. #include "h264chroma.h"
  39. #include "h264_mvpred.h"
  40. #include "h264_ps.h"
  41. #include "golomb.h"
  42. #include "mathops.h"
  43. #include "mpegutils.h"
  44. #include "mpegvideo.h"
  45. #include "rectangle.h"
  46. #include "thread.h"
  47. static const uint8_t field_scan[16+1] = {
  48. 0 + 0 * 4, 0 + 1 * 4, 1 + 0 * 4, 0 + 2 * 4,
  49. 0 + 3 * 4, 1 + 1 * 4, 1 + 2 * 4, 1 + 3 * 4,
  50. 2 + 0 * 4, 2 + 1 * 4, 2 + 2 * 4, 2 + 3 * 4,
  51. 3 + 0 * 4, 3 + 1 * 4, 3 + 2 * 4, 3 + 3 * 4,
  52. };
  53. static const uint8_t field_scan8x8[64+1] = {
  54. 0 + 0 * 8, 0 + 1 * 8, 0 + 2 * 8, 1 + 0 * 8,
  55. 1 + 1 * 8, 0 + 3 * 8, 0 + 4 * 8, 1 + 2 * 8,
  56. 2 + 0 * 8, 1 + 3 * 8, 0 + 5 * 8, 0 + 6 * 8,
  57. 0 + 7 * 8, 1 + 4 * 8, 2 + 1 * 8, 3 + 0 * 8,
  58. 2 + 2 * 8, 1 + 5 * 8, 1 + 6 * 8, 1 + 7 * 8,
  59. 2 + 3 * 8, 3 + 1 * 8, 4 + 0 * 8, 3 + 2 * 8,
  60. 2 + 4 * 8, 2 + 5 * 8, 2 + 6 * 8, 2 + 7 * 8,
  61. 3 + 3 * 8, 4 + 1 * 8, 5 + 0 * 8, 4 + 2 * 8,
  62. 3 + 4 * 8, 3 + 5 * 8, 3 + 6 * 8, 3 + 7 * 8,
  63. 4 + 3 * 8, 5 + 1 * 8, 6 + 0 * 8, 5 + 2 * 8,
  64. 4 + 4 * 8, 4 + 5 * 8, 4 + 6 * 8, 4 + 7 * 8,
  65. 5 + 3 * 8, 6 + 1 * 8, 6 + 2 * 8, 5 + 4 * 8,
  66. 5 + 5 * 8, 5 + 6 * 8, 5 + 7 * 8, 6 + 3 * 8,
  67. 7 + 0 * 8, 7 + 1 * 8, 6 + 4 * 8, 6 + 5 * 8,
  68. 6 + 6 * 8, 6 + 7 * 8, 7 + 2 * 8, 7 + 3 * 8,
  69. 7 + 4 * 8, 7 + 5 * 8, 7 + 6 * 8, 7 + 7 * 8,
  70. };
  71. static const uint8_t field_scan8x8_cavlc[64+1] = {
  72. 0 + 0 * 8, 1 + 1 * 8, 2 + 0 * 8, 0 + 7 * 8,
  73. 2 + 2 * 8, 2 + 3 * 8, 2 + 4 * 8, 3 + 3 * 8,
  74. 3 + 4 * 8, 4 + 3 * 8, 4 + 4 * 8, 5 + 3 * 8,
  75. 5 + 5 * 8, 7 + 0 * 8, 6 + 6 * 8, 7 + 4 * 8,
  76. 0 + 1 * 8, 0 + 3 * 8, 1 + 3 * 8, 1 + 4 * 8,
  77. 1 + 5 * 8, 3 + 1 * 8, 2 + 5 * 8, 4 + 1 * 8,
  78. 3 + 5 * 8, 5 + 1 * 8, 4 + 5 * 8, 6 + 1 * 8,
  79. 5 + 6 * 8, 7 + 1 * 8, 6 + 7 * 8, 7 + 5 * 8,
  80. 0 + 2 * 8, 0 + 4 * 8, 0 + 5 * 8, 2 + 1 * 8,
  81. 1 + 6 * 8, 4 + 0 * 8, 2 + 6 * 8, 5 + 0 * 8,
  82. 3 + 6 * 8, 6 + 0 * 8, 4 + 6 * 8, 6 + 2 * 8,
  83. 5 + 7 * 8, 6 + 4 * 8, 7 + 2 * 8, 7 + 6 * 8,
  84. 1 + 0 * 8, 1 + 2 * 8, 0 + 6 * 8, 3 + 0 * 8,
  85. 1 + 7 * 8, 3 + 2 * 8, 2 + 7 * 8, 4 + 2 * 8,
  86. 3 + 7 * 8, 5 + 2 * 8, 4 + 7 * 8, 5 + 4 * 8,
  87. 6 + 3 * 8, 6 + 5 * 8, 7 + 3 * 8, 7 + 7 * 8,
  88. };
  89. // zigzag_scan8x8_cavlc[i] = zigzag_scan8x8[(i/4) + 16*(i%4)]
  90. static const uint8_t zigzag_scan8x8_cavlc[64+1] = {
  91. 0 + 0 * 8, 1 + 1 * 8, 1 + 2 * 8, 2 + 2 * 8,
  92. 4 + 1 * 8, 0 + 5 * 8, 3 + 3 * 8, 7 + 0 * 8,
  93. 3 + 4 * 8, 1 + 7 * 8, 5 + 3 * 8, 6 + 3 * 8,
  94. 2 + 7 * 8, 6 + 4 * 8, 5 + 6 * 8, 7 + 5 * 8,
  95. 1 + 0 * 8, 2 + 0 * 8, 0 + 3 * 8, 3 + 1 * 8,
  96. 3 + 2 * 8, 0 + 6 * 8, 4 + 2 * 8, 6 + 1 * 8,
  97. 2 + 5 * 8, 2 + 6 * 8, 6 + 2 * 8, 5 + 4 * 8,
  98. 3 + 7 * 8, 7 + 3 * 8, 4 + 7 * 8, 7 + 6 * 8,
  99. 0 + 1 * 8, 3 + 0 * 8, 0 + 4 * 8, 4 + 0 * 8,
  100. 2 + 3 * 8, 1 + 5 * 8, 5 + 1 * 8, 5 + 2 * 8,
  101. 1 + 6 * 8, 3 + 5 * 8, 7 + 1 * 8, 4 + 5 * 8,
  102. 4 + 6 * 8, 7 + 4 * 8, 5 + 7 * 8, 6 + 7 * 8,
  103. 0 + 2 * 8, 2 + 1 * 8, 1 + 3 * 8, 5 + 0 * 8,
  104. 1 + 4 * 8, 2 + 4 * 8, 6 + 0 * 8, 4 + 3 * 8,
  105. 0 + 7 * 8, 4 + 4 * 8, 7 + 2 * 8, 3 + 6 * 8,
  106. 5 + 5 * 8, 6 + 5 * 8, 6 + 6 * 8, 7 + 7 * 8,
  107. };
  108. static void release_unused_pictures(H264Context *h, int remove_current)
  109. {
  110. int i;
  111. /* release non reference frames */
  112. for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
  113. if (h->DPB[i].f->buf[0] && !h->DPB[i].reference &&
  114. (remove_current || &h->DPB[i] != h->cur_pic_ptr)) {
  115. ff_h264_unref_picture(h, &h->DPB[i]);
  116. }
  117. }
  118. }
  119. static int alloc_scratch_buffers(H264SliceContext *sl, int linesize)
  120. {
  121. const H264Context *h = sl->h264;
  122. int alloc_size = FFALIGN(FFABS(linesize) + 32, 32);
  123. av_fast_malloc(&sl->bipred_scratchpad, &sl->bipred_scratchpad_allocated, 16 * 6 * alloc_size);
  124. // edge emu needs blocksize + filter length - 1
  125. // (= 21x21 for H.264)
  126. av_fast_malloc(&sl->edge_emu_buffer, &sl->edge_emu_buffer_allocated, alloc_size * 2 * 21);
  127. av_fast_mallocz(&sl->top_borders[0], &sl->top_borders_allocated[0],
  128. h->mb_width * 16 * 3 * sizeof(uint8_t) * 2);
  129. av_fast_mallocz(&sl->top_borders[1], &sl->top_borders_allocated[1],
  130. h->mb_width * 16 * 3 * sizeof(uint8_t) * 2);
  131. if (!sl->bipred_scratchpad || !sl->edge_emu_buffer ||
  132. !sl->top_borders[0] || !sl->top_borders[1]) {
  133. av_freep(&sl->bipred_scratchpad);
  134. av_freep(&sl->edge_emu_buffer);
  135. av_freep(&sl->top_borders[0]);
  136. av_freep(&sl->top_borders[1]);
  137. sl->bipred_scratchpad_allocated = 0;
  138. sl->edge_emu_buffer_allocated = 0;
  139. sl->top_borders_allocated[0] = 0;
  140. sl->top_borders_allocated[1] = 0;
  141. return AVERROR(ENOMEM);
  142. }
  143. return 0;
  144. }
  145. static int init_table_pools(H264Context *h)
  146. {
  147. const int big_mb_num = h->mb_stride * (h->mb_height + 1) + 1;
  148. const int mb_array_size = h->mb_stride * h->mb_height;
  149. const int b4_stride = h->mb_width * 4 + 1;
  150. const int b4_array_size = b4_stride * h->mb_height * 4;
  151. h->qscale_table_pool = av_buffer_pool_init(big_mb_num + h->mb_stride,
  152. av_buffer_allocz);
  153. h->mb_type_pool = av_buffer_pool_init((big_mb_num + h->mb_stride) *
  154. sizeof(uint32_t), av_buffer_allocz);
  155. h->motion_val_pool = av_buffer_pool_init(2 * (b4_array_size + 4) *
  156. sizeof(int16_t), av_buffer_allocz);
  157. h->ref_index_pool = av_buffer_pool_init(4 * mb_array_size, av_buffer_allocz);
  158. if (!h->qscale_table_pool || !h->mb_type_pool || !h->motion_val_pool ||
  159. !h->ref_index_pool) {
  160. av_buffer_pool_uninit(&h->qscale_table_pool);
  161. av_buffer_pool_uninit(&h->mb_type_pool);
  162. av_buffer_pool_uninit(&h->motion_val_pool);
  163. av_buffer_pool_uninit(&h->ref_index_pool);
  164. return AVERROR(ENOMEM);
  165. }
  166. return 0;
  167. }
  168. static int alloc_picture(H264Context *h, H264Picture *pic)
  169. {
  170. int i, ret = 0;
  171. av_assert0(!pic->f->data[0]);
  172. pic->tf.f = pic->f;
  173. ret = ff_thread_get_buffer(h->avctx, &pic->tf, pic->reference ?
  174. AV_GET_BUFFER_FLAG_REF : 0);
  175. if (ret < 0)
  176. goto fail;
  177. if (h->avctx->hwaccel) {
  178. const AVHWAccel *hwaccel = h->avctx->hwaccel;
  179. av_assert0(!pic->hwaccel_picture_private);
  180. if (hwaccel->frame_priv_data_size) {
  181. pic->hwaccel_priv_buf = av_buffer_allocz(hwaccel->frame_priv_data_size);
  182. if (!pic->hwaccel_priv_buf)
  183. return AVERROR(ENOMEM);
  184. pic->hwaccel_picture_private = pic->hwaccel_priv_buf->data;
  185. }
  186. }
  187. if (CONFIG_GRAY && !h->avctx->hwaccel && h->flags & AV_CODEC_FLAG_GRAY && pic->f->data[2]) {
  188. int h_chroma_shift, v_chroma_shift;
  189. av_pix_fmt_get_chroma_sub_sample(pic->f->format,
  190. &h_chroma_shift, &v_chroma_shift);
  191. for(i=0; i<AV_CEIL_RSHIFT(pic->f->height, v_chroma_shift); i++) {
  192. memset(pic->f->data[1] + pic->f->linesize[1]*i,
  193. 0x80, AV_CEIL_RSHIFT(pic->f->width, h_chroma_shift));
  194. memset(pic->f->data[2] + pic->f->linesize[2]*i,
  195. 0x80, AV_CEIL_RSHIFT(pic->f->width, h_chroma_shift));
  196. }
  197. }
  198. if (!h->qscale_table_pool) {
  199. ret = init_table_pools(h);
  200. if (ret < 0)
  201. goto fail;
  202. }
  203. pic->qscale_table_buf = av_buffer_pool_get(h->qscale_table_pool);
  204. pic->mb_type_buf = av_buffer_pool_get(h->mb_type_pool);
  205. if (!pic->qscale_table_buf || !pic->mb_type_buf)
  206. goto fail;
  207. pic->mb_type = (uint32_t*)pic->mb_type_buf->data + 2 * h->mb_stride + 1;
  208. pic->qscale_table = pic->qscale_table_buf->data + 2 * h->mb_stride + 1;
  209. for (i = 0; i < 2; i++) {
  210. pic->motion_val_buf[i] = av_buffer_pool_get(h->motion_val_pool);
  211. pic->ref_index_buf[i] = av_buffer_pool_get(h->ref_index_pool);
  212. if (!pic->motion_val_buf[i] || !pic->ref_index_buf[i])
  213. goto fail;
  214. pic->motion_val[i] = (int16_t (*)[2])pic->motion_val_buf[i]->data + 4;
  215. pic->ref_index[i] = pic->ref_index_buf[i]->data;
  216. }
  217. pic->pps_buf = av_buffer_ref(h->ps.pps_ref);
  218. if (!pic->pps_buf)
  219. goto fail;
  220. pic->pps = (const PPS*)pic->pps_buf->data;
  221. pic->mb_width = h->mb_width;
  222. pic->mb_height = h->mb_height;
  223. pic->mb_stride = h->mb_stride;
  224. return 0;
  225. fail:
  226. ff_h264_unref_picture(h, pic);
  227. return (ret < 0) ? ret : AVERROR(ENOMEM);
  228. }
  229. static int find_unused_picture(H264Context *h)
  230. {
  231. int i;
  232. for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
  233. if (!h->DPB[i].f->buf[0])
  234. return i;
  235. }
  236. return AVERROR_INVALIDDATA;
  237. }
  238. #define IN_RANGE(a, b, size) (((void*)(a) >= (void*)(b)) && ((void*)(a) < (void*)((b) + (size))))
  239. #define REBASE_PICTURE(pic, new_ctx, old_ctx) \
  240. (((pic) && (pic) >= (old_ctx)->DPB && \
  241. (pic) < (old_ctx)->DPB + H264_MAX_PICTURE_COUNT) ? \
  242. &(new_ctx)->DPB[(pic) - (old_ctx)->DPB] : NULL)
  243. static void copy_picture_range(H264Picture **to, H264Picture **from, int count,
  244. H264Context *new_base,
  245. H264Context *old_base)
  246. {
  247. int i;
  248. for (i = 0; i < count; i++) {
  249. av_assert1(!from[i] ||
  250. IN_RANGE(from[i], old_base, 1) ||
  251. IN_RANGE(from[i], old_base->DPB, H264_MAX_PICTURE_COUNT));
  252. to[i] = REBASE_PICTURE(from[i], new_base, old_base);
  253. }
  254. }
  255. static int h264_slice_header_init(H264Context *h);
  256. int ff_h264_update_thread_context(AVCodecContext *dst,
  257. const AVCodecContext *src)
  258. {
  259. H264Context *h = dst->priv_data, *h1 = src->priv_data;
  260. int inited = h->context_initialized, err = 0;
  261. int need_reinit = 0;
  262. int i, ret;
  263. if (dst == src)
  264. return 0;
  265. // We can't fail if SPS isn't set at it breaks current skip_frame code
  266. //if (!h1->ps.sps)
  267. // return AVERROR_INVALIDDATA;
  268. if (inited &&
  269. (h->width != h1->width ||
  270. h->height != h1->height ||
  271. h->mb_width != h1->mb_width ||
  272. h->mb_height != h1->mb_height ||
  273. !h->ps.sps ||
  274. h->ps.sps->bit_depth_luma != h1->ps.sps->bit_depth_luma ||
  275. h->ps.sps->chroma_format_idc != h1->ps.sps->chroma_format_idc ||
  276. h->ps.sps->colorspace != h1->ps.sps->colorspace)) {
  277. need_reinit = 1;
  278. }
  279. /* copy block_offset since frame_start may not be called */
  280. memcpy(h->block_offset, h1->block_offset, sizeof(h->block_offset));
  281. // SPS/PPS
  282. for (i = 0; i < FF_ARRAY_ELEMS(h->ps.sps_list); i++) {
  283. ret = av_buffer_replace(&h->ps.sps_list[i], h1->ps.sps_list[i]);
  284. if (ret < 0)
  285. return ret;
  286. }
  287. for (i = 0; i < FF_ARRAY_ELEMS(h->ps.pps_list); i++) {
  288. ret = av_buffer_replace(&h->ps.pps_list[i], h1->ps.pps_list[i]);
  289. if (ret < 0)
  290. return ret;
  291. }
  292. ret = av_buffer_replace(&h->ps.pps_ref, h1->ps.pps_ref);
  293. if (ret < 0)
  294. return ret;
  295. h->ps.pps = NULL;
  296. h->ps.sps = NULL;
  297. if (h1->ps.pps_ref) {
  298. h->ps.pps = (const PPS*)h->ps.pps_ref->data;
  299. h->ps.sps = h->ps.pps->sps;
  300. }
  301. if (need_reinit || !inited) {
  302. h->width = h1->width;
  303. h->height = h1->height;
  304. h->mb_height = h1->mb_height;
  305. h->mb_width = h1->mb_width;
  306. h->mb_num = h1->mb_num;
  307. h->mb_stride = h1->mb_stride;
  308. h->b_stride = h1->b_stride;
  309. h->x264_build = h1->x264_build;
  310. if (h->context_initialized || h1->context_initialized) {
  311. if ((err = h264_slice_header_init(h)) < 0) {
  312. av_log(h->avctx, AV_LOG_ERROR, "h264_slice_header_init() failed");
  313. return err;
  314. }
  315. }
  316. /* copy block_offset since frame_start may not be called */
  317. memcpy(h->block_offset, h1->block_offset, sizeof(h->block_offset));
  318. }
  319. h->avctx->coded_height = h1->avctx->coded_height;
  320. h->avctx->coded_width = h1->avctx->coded_width;
  321. h->avctx->width = h1->avctx->width;
  322. h->avctx->height = h1->avctx->height;
  323. h->width_from_caller = h1->width_from_caller;
  324. h->height_from_caller = h1->height_from_caller;
  325. h->coded_picture_number = h1->coded_picture_number;
  326. h->first_field = h1->first_field;
  327. h->picture_structure = h1->picture_structure;
  328. h->mb_aff_frame = h1->mb_aff_frame;
  329. h->droppable = h1->droppable;
  330. for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
  331. ff_h264_unref_picture(h, &h->DPB[i]);
  332. if (h1->DPB[i].f->buf[0] &&
  333. (ret = ff_h264_ref_picture(h, &h->DPB[i], &h1->DPB[i])) < 0)
  334. return ret;
  335. }
  336. h->cur_pic_ptr = REBASE_PICTURE(h1->cur_pic_ptr, h, h1);
  337. ff_h264_unref_picture(h, &h->cur_pic);
  338. if (h1->cur_pic.f->buf[0]) {
  339. ret = ff_h264_ref_picture(h, &h->cur_pic, &h1->cur_pic);
  340. if (ret < 0)
  341. return ret;
  342. }
  343. h->enable_er = h1->enable_er;
  344. h->workaround_bugs = h1->workaround_bugs;
  345. h->droppable = h1->droppable;
  346. // extradata/NAL handling
  347. h->is_avc = h1->is_avc;
  348. h->nal_length_size = h1->nal_length_size;
  349. memcpy(&h->poc, &h1->poc, sizeof(h->poc));
  350. memcpy(h->short_ref, h1->short_ref, sizeof(h->short_ref));
  351. memcpy(h->long_ref, h1->long_ref, sizeof(h->long_ref));
  352. memcpy(h->delayed_pic, h1->delayed_pic, sizeof(h->delayed_pic));
  353. memcpy(h->last_pocs, h1->last_pocs, sizeof(h->last_pocs));
  354. h->next_output_pic = h1->next_output_pic;
  355. h->next_outputed_poc = h1->next_outputed_poc;
  356. memcpy(h->mmco, h1->mmco, sizeof(h->mmco));
  357. h->nb_mmco = h1->nb_mmco;
  358. h->mmco_reset = h1->mmco_reset;
  359. h->explicit_ref_marking = h1->explicit_ref_marking;
  360. h->long_ref_count = h1->long_ref_count;
  361. h->short_ref_count = h1->short_ref_count;
  362. copy_picture_range(h->short_ref, h1->short_ref, 32, h, h1);
  363. copy_picture_range(h->long_ref, h1->long_ref, 32, h, h1);
  364. copy_picture_range(h->delayed_pic, h1->delayed_pic,
  365. MAX_DELAYED_PIC_COUNT + 2, h, h1);
  366. h->frame_recovered = h1->frame_recovered;
  367. ret = av_buffer_replace(&h->sei.a53_caption.buf_ref, h1->sei.a53_caption.buf_ref);
  368. if (ret < 0)
  369. return ret;
  370. for (i = 0; i < h->sei.unregistered.nb_buf_ref; i++)
  371. av_buffer_unref(&h->sei.unregistered.buf_ref[i]);
  372. h->sei.unregistered.nb_buf_ref = 0;
  373. if (h1->sei.unregistered.nb_buf_ref) {
  374. ret = av_reallocp_array(&h->sei.unregistered.buf_ref,
  375. h1->sei.unregistered.nb_buf_ref,
  376. sizeof(*h->sei.unregistered.buf_ref));
  377. if (ret < 0)
  378. return ret;
  379. for (i = 0; i < h1->sei.unregistered.nb_buf_ref; i++) {
  380. h->sei.unregistered.buf_ref[i] = av_buffer_ref(h1->sei.unregistered.buf_ref[i]);
  381. if (!h->sei.unregistered.buf_ref[i])
  382. return AVERROR(ENOMEM);
  383. h->sei.unregistered.nb_buf_ref++;
  384. }
  385. }
  386. h->sei.unregistered.x264_build = h1->sei.unregistered.x264_build;
  387. if (!h->cur_pic_ptr)
  388. return 0;
  389. if (!h->droppable) {
  390. err = ff_h264_execute_ref_pic_marking(h);
  391. h->poc.prev_poc_msb = h->poc.poc_msb;
  392. h->poc.prev_poc_lsb = h->poc.poc_lsb;
  393. }
  394. h->poc.prev_frame_num_offset = h->poc.frame_num_offset;
  395. h->poc.prev_frame_num = h->poc.frame_num;
  396. h->recovery_frame = h1->recovery_frame;
  397. return err;
  398. }
  399. static int h264_frame_start(H264Context *h)
  400. {
  401. H264Picture *pic;
  402. int i, ret;
  403. const int pixel_shift = h->pixel_shift;
  404. if (!ff_thread_can_start_frame(h->avctx)) {
  405. av_log(h->avctx, AV_LOG_ERROR, "Attempt to start a frame outside SETUP state\n");
  406. return -1;
  407. }
  408. release_unused_pictures(h, 1);
  409. h->cur_pic_ptr = NULL;
  410. i = find_unused_picture(h);
  411. if (i < 0) {
  412. av_log(h->avctx, AV_LOG_ERROR, "no frame buffer available\n");
  413. return i;
  414. }
  415. pic = &h->DPB[i];
  416. pic->reference = h->droppable ? 0 : h->picture_structure;
  417. pic->f->coded_picture_number = h->coded_picture_number++;
  418. pic->field_picture = h->picture_structure != PICT_FRAME;
  419. pic->frame_num = h->poc.frame_num;
  420. /*
  421. * Zero key_frame here; IDR markings per slice in frame or fields are ORed
  422. * in later.
  423. * See decode_nal_units().
  424. */
  425. pic->f->key_frame = 0;
  426. pic->mmco_reset = 0;
  427. pic->recovered = 0;
  428. pic->invalid_gap = 0;
  429. pic->sei_recovery_frame_cnt = h->sei.recovery_point.recovery_frame_cnt;
  430. pic->f->pict_type = h->slice_ctx[0].slice_type;
  431. pic->f->crop_left = h->crop_left;
  432. pic->f->crop_right = h->crop_right;
  433. pic->f->crop_top = h->crop_top;
  434. pic->f->crop_bottom = h->crop_bottom;
  435. if ((ret = alloc_picture(h, pic)) < 0)
  436. return ret;
  437. h->cur_pic_ptr = pic;
  438. ff_h264_unref_picture(h, &h->cur_pic);
  439. if (CONFIG_ERROR_RESILIENCE) {
  440. ff_h264_set_erpic(&h->slice_ctx[0].er.cur_pic, NULL);
  441. }
  442. if ((ret = ff_h264_ref_picture(h, &h->cur_pic, h->cur_pic_ptr)) < 0)
  443. return ret;
  444. for (i = 0; i < h->nb_slice_ctx; i++) {
  445. h->slice_ctx[i].linesize = h->cur_pic_ptr->f->linesize[0];
  446. h->slice_ctx[i].uvlinesize = h->cur_pic_ptr->f->linesize[1];
  447. }
  448. if (CONFIG_ERROR_RESILIENCE && h->enable_er) {
  449. ff_er_frame_start(&h->slice_ctx[0].er);
  450. ff_h264_set_erpic(&h->slice_ctx[0].er.last_pic, NULL);
  451. ff_h264_set_erpic(&h->slice_ctx[0].er.next_pic, NULL);
  452. }
  453. for (i = 0; i < 16; i++) {
  454. h->block_offset[i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * pic->f->linesize[0] * ((scan8[i] - scan8[0]) >> 3);
  455. h->block_offset[48 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * pic->f->linesize[0] * ((scan8[i] - scan8[0]) >> 3);
  456. }
  457. for (i = 0; i < 16; i++) {
  458. h->block_offset[16 + i] =
  459. h->block_offset[32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * pic->f->linesize[1] * ((scan8[i] - scan8[0]) >> 3);
  460. h->block_offset[48 + 16 + i] =
  461. h->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * pic->f->linesize[1] * ((scan8[i] - scan8[0]) >> 3);
  462. }
  463. /* We mark the current picture as non-reference after allocating it, so
  464. * that if we break out due to an error it can be released automatically
  465. * in the next ff_mpv_frame_start().
  466. */
  467. h->cur_pic_ptr->reference = 0;
  468. h->cur_pic_ptr->field_poc[0] = h->cur_pic_ptr->field_poc[1] = INT_MAX;
  469. h->next_output_pic = NULL;
  470. h->postpone_filter = 0;
  471. h->mb_aff_frame = h->ps.sps->mb_aff && (h->picture_structure == PICT_FRAME);
  472. if (h->sei.unregistered.x264_build >= 0)
  473. h->x264_build = h->sei.unregistered.x264_build;
  474. assert(h->cur_pic_ptr->long_ref == 0);
  475. return 0;
  476. }
  477. static av_always_inline void backup_mb_border(const H264Context *h, H264SliceContext *sl,
  478. uint8_t *src_y,
  479. uint8_t *src_cb, uint8_t *src_cr,
  480. int linesize, int uvlinesize,
  481. int simple)
  482. {
  483. uint8_t *top_border;
  484. int top_idx = 1;
  485. const int pixel_shift = h->pixel_shift;
  486. int chroma444 = CHROMA444(h);
  487. int chroma422 = CHROMA422(h);
  488. src_y -= linesize;
  489. src_cb -= uvlinesize;
  490. src_cr -= uvlinesize;
  491. if (!simple && FRAME_MBAFF(h)) {
  492. if (sl->mb_y & 1) {
  493. if (!MB_MBAFF(sl)) {
  494. top_border = sl->top_borders[0][sl->mb_x];
  495. AV_COPY128(top_border, src_y + 15 * linesize);
  496. if (pixel_shift)
  497. AV_COPY128(top_border + 16, src_y + 15 * linesize + 16);
  498. if (simple || !CONFIG_GRAY || !(h->flags & AV_CODEC_FLAG_GRAY)) {
  499. if (chroma444) {
  500. if (pixel_shift) {
  501. AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
  502. AV_COPY128(top_border + 48, src_cb + 15 * uvlinesize + 16);
  503. AV_COPY128(top_border + 64, src_cr + 15 * uvlinesize);
  504. AV_COPY128(top_border + 80, src_cr + 15 * uvlinesize + 16);
  505. } else {
  506. AV_COPY128(top_border + 16, src_cb + 15 * uvlinesize);
  507. AV_COPY128(top_border + 32, src_cr + 15 * uvlinesize);
  508. }
  509. } else if (chroma422) {
  510. if (pixel_shift) {
  511. AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
  512. AV_COPY128(top_border + 48, src_cr + 15 * uvlinesize);
  513. } else {
  514. AV_COPY64(top_border + 16, src_cb + 15 * uvlinesize);
  515. AV_COPY64(top_border + 24, src_cr + 15 * uvlinesize);
  516. }
  517. } else {
  518. if (pixel_shift) {
  519. AV_COPY128(top_border + 32, src_cb + 7 * uvlinesize);
  520. AV_COPY128(top_border + 48, src_cr + 7 * uvlinesize);
  521. } else {
  522. AV_COPY64(top_border + 16, src_cb + 7 * uvlinesize);
  523. AV_COPY64(top_border + 24, src_cr + 7 * uvlinesize);
  524. }
  525. }
  526. }
  527. }
  528. } else if (MB_MBAFF(sl)) {
  529. top_idx = 0;
  530. } else
  531. return;
  532. }
  533. top_border = sl->top_borders[top_idx][sl->mb_x];
  534. /* There are two lines saved, the line above the top macroblock
  535. * of a pair, and the line above the bottom macroblock. */
  536. AV_COPY128(top_border, src_y + 16 * linesize);
  537. if (pixel_shift)
  538. AV_COPY128(top_border + 16, src_y + 16 * linesize + 16);
  539. if (simple || !CONFIG_GRAY || !(h->flags & AV_CODEC_FLAG_GRAY)) {
  540. if (chroma444) {
  541. if (pixel_shift) {
  542. AV_COPY128(top_border + 32, src_cb + 16 * linesize);
  543. AV_COPY128(top_border + 48, src_cb + 16 * linesize + 16);
  544. AV_COPY128(top_border + 64, src_cr + 16 * linesize);
  545. AV_COPY128(top_border + 80, src_cr + 16 * linesize + 16);
  546. } else {
  547. AV_COPY128(top_border + 16, src_cb + 16 * linesize);
  548. AV_COPY128(top_border + 32, src_cr + 16 * linesize);
  549. }
  550. } else if (chroma422) {
  551. if (pixel_shift) {
  552. AV_COPY128(top_border + 32, src_cb + 16 * uvlinesize);
  553. AV_COPY128(top_border + 48, src_cr + 16 * uvlinesize);
  554. } else {
  555. AV_COPY64(top_border + 16, src_cb + 16 * uvlinesize);
  556. AV_COPY64(top_border + 24, src_cr + 16 * uvlinesize);
  557. }
  558. } else {
  559. if (pixel_shift) {
  560. AV_COPY128(top_border + 32, src_cb + 8 * uvlinesize);
  561. AV_COPY128(top_border + 48, src_cr + 8 * uvlinesize);
  562. } else {
  563. AV_COPY64(top_border + 16, src_cb + 8 * uvlinesize);
  564. AV_COPY64(top_border + 24, src_cr + 8 * uvlinesize);
  565. }
  566. }
  567. }
  568. }
  569. /**
  570. * Initialize implicit_weight table.
  571. * @param field 0/1 initialize the weight for interlaced MBAFF
  572. * -1 initializes the rest
  573. */
  574. static void implicit_weight_table(const H264Context *h, H264SliceContext *sl, int field)
  575. {
  576. int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
  577. for (i = 0; i < 2; i++) {
  578. sl->pwt.luma_weight_flag[i] = 0;
  579. sl->pwt.chroma_weight_flag[i] = 0;
  580. }
  581. if (field < 0) {
  582. if (h->picture_structure == PICT_FRAME) {
  583. cur_poc = h->cur_pic_ptr->poc;
  584. } else {
  585. cur_poc = h->cur_pic_ptr->field_poc[h->picture_structure - 1];
  586. }
  587. if (sl->ref_count[0] == 1 && sl->ref_count[1] == 1 && !FRAME_MBAFF(h) &&
  588. sl->ref_list[0][0].poc + (int64_t)sl->ref_list[1][0].poc == 2LL * cur_poc) {
  589. sl->pwt.use_weight = 0;
  590. sl->pwt.use_weight_chroma = 0;
  591. return;
  592. }
  593. ref_start = 0;
  594. ref_count0 = sl->ref_count[0];
  595. ref_count1 = sl->ref_count[1];
  596. } else {
  597. cur_poc = h->cur_pic_ptr->field_poc[field];
  598. ref_start = 16;
  599. ref_count0 = 16 + 2 * sl->ref_count[0];
  600. ref_count1 = 16 + 2 * sl->ref_count[1];
  601. }
  602. sl->pwt.use_weight = 2;
  603. sl->pwt.use_weight_chroma = 2;
  604. sl->pwt.luma_log2_weight_denom = 5;
  605. sl->pwt.chroma_log2_weight_denom = 5;
  606. for (ref0 = ref_start; ref0 < ref_count0; ref0++) {
  607. int64_t poc0 = sl->ref_list[0][ref0].poc;
  608. for (ref1 = ref_start; ref1 < ref_count1; ref1++) {
  609. int w = 32;
  610. if (!sl->ref_list[0][ref0].parent->long_ref && !sl->ref_list[1][ref1].parent->long_ref) {
  611. int poc1 = sl->ref_list[1][ref1].poc;
  612. int td = av_clip_int8(poc1 - poc0);
  613. if (td) {
  614. int tb = av_clip_int8(cur_poc - poc0);
  615. int tx = (16384 + (FFABS(td) >> 1)) / td;
  616. int dist_scale_factor = (tb * tx + 32) >> 8;
  617. if (dist_scale_factor >= -64 && dist_scale_factor <= 128)
  618. w = 64 - dist_scale_factor;
  619. }
  620. }
  621. if (field < 0) {
  622. sl->pwt.implicit_weight[ref0][ref1][0] =
  623. sl->pwt.implicit_weight[ref0][ref1][1] = w;
  624. } else {
  625. sl->pwt.implicit_weight[ref0][ref1][field] = w;
  626. }
  627. }
  628. }
  629. }
  630. /**
  631. * initialize scan tables
  632. */
  633. static void init_scan_tables(H264Context *h)
  634. {
  635. int i;
  636. for (i = 0; i < 16; i++) {
  637. #define TRANSPOSE(x) ((x) >> 2) | (((x) << 2) & 0xF)
  638. h->zigzag_scan[i] = TRANSPOSE(ff_zigzag_scan[i]);
  639. h->field_scan[i] = TRANSPOSE(field_scan[i]);
  640. #undef TRANSPOSE
  641. }
  642. for (i = 0; i < 64; i++) {
  643. #define TRANSPOSE(x) ((x) >> 3) | (((x) & 7) << 3)
  644. h->zigzag_scan8x8[i] = TRANSPOSE(ff_zigzag_direct[i]);
  645. h->zigzag_scan8x8_cavlc[i] = TRANSPOSE(zigzag_scan8x8_cavlc[i]);
  646. h->field_scan8x8[i] = TRANSPOSE(field_scan8x8[i]);
  647. h->field_scan8x8_cavlc[i] = TRANSPOSE(field_scan8x8_cavlc[i]);
  648. #undef TRANSPOSE
  649. }
  650. if (h->ps.sps->transform_bypass) { // FIXME same ugly
  651. memcpy(h->zigzag_scan_q0 , ff_zigzag_scan , sizeof(h->zigzag_scan_q0 ));
  652. memcpy(h->zigzag_scan8x8_q0 , ff_zigzag_direct , sizeof(h->zigzag_scan8x8_q0 ));
  653. memcpy(h->zigzag_scan8x8_cavlc_q0 , zigzag_scan8x8_cavlc , sizeof(h->zigzag_scan8x8_cavlc_q0));
  654. memcpy(h->field_scan_q0 , field_scan , sizeof(h->field_scan_q0 ));
  655. memcpy(h->field_scan8x8_q0 , field_scan8x8 , sizeof(h->field_scan8x8_q0 ));
  656. memcpy(h->field_scan8x8_cavlc_q0 , field_scan8x8_cavlc , sizeof(h->field_scan8x8_cavlc_q0 ));
  657. } else {
  658. memcpy(h->zigzag_scan_q0 , h->zigzag_scan , sizeof(h->zigzag_scan_q0 ));
  659. memcpy(h->zigzag_scan8x8_q0 , h->zigzag_scan8x8 , sizeof(h->zigzag_scan8x8_q0 ));
  660. memcpy(h->zigzag_scan8x8_cavlc_q0 , h->zigzag_scan8x8_cavlc , sizeof(h->zigzag_scan8x8_cavlc_q0));
  661. memcpy(h->field_scan_q0 , h->field_scan , sizeof(h->field_scan_q0 ));
  662. memcpy(h->field_scan8x8_q0 , h->field_scan8x8 , sizeof(h->field_scan8x8_q0 ));
  663. memcpy(h->field_scan8x8_cavlc_q0 , h->field_scan8x8_cavlc , sizeof(h->field_scan8x8_cavlc_q0 ));
  664. }
  665. }
  666. static enum AVPixelFormat get_pixel_format(H264Context *h, int force_callback)
  667. {
  668. #define HWACCEL_MAX (CONFIG_H264_DXVA2_HWACCEL + \
  669. (CONFIG_H264_D3D11VA_HWACCEL * 2) + \
  670. CONFIG_H264_NVDEC_HWACCEL + \
  671. CONFIG_H264_VAAPI_HWACCEL + \
  672. CONFIG_H264_VIDEOTOOLBOX_HWACCEL + \
  673. CONFIG_H264_VDPAU_HWACCEL)
  674. enum AVPixelFormat pix_fmts[HWACCEL_MAX + 2], *fmt = pix_fmts;
  675. const enum AVPixelFormat *choices = pix_fmts;
  676. int i;
  677. switch (h->ps.sps->bit_depth_luma) {
  678. case 9:
  679. if (CHROMA444(h)) {
  680. if (h->avctx->colorspace == AVCOL_SPC_RGB) {
  681. *fmt++ = AV_PIX_FMT_GBRP9;
  682. } else
  683. *fmt++ = AV_PIX_FMT_YUV444P9;
  684. } else if (CHROMA422(h))
  685. *fmt++ = AV_PIX_FMT_YUV422P9;
  686. else
  687. *fmt++ = AV_PIX_FMT_YUV420P9;
  688. break;
  689. case 10:
  690. if (CHROMA444(h)) {
  691. if (h->avctx->colorspace == AVCOL_SPC_RGB) {
  692. *fmt++ = AV_PIX_FMT_GBRP10;
  693. } else
  694. *fmt++ = AV_PIX_FMT_YUV444P10;
  695. } else if (CHROMA422(h))
  696. *fmt++ = AV_PIX_FMT_YUV422P10;
  697. else
  698. *fmt++ = AV_PIX_FMT_YUV420P10;
  699. break;
  700. case 12:
  701. if (CHROMA444(h)) {
  702. if (h->avctx->colorspace == AVCOL_SPC_RGB) {
  703. *fmt++ = AV_PIX_FMT_GBRP12;
  704. } else
  705. *fmt++ = AV_PIX_FMT_YUV444P12;
  706. } else if (CHROMA422(h))
  707. *fmt++ = AV_PIX_FMT_YUV422P12;
  708. else
  709. *fmt++ = AV_PIX_FMT_YUV420P12;
  710. break;
  711. case 14:
  712. if (CHROMA444(h)) {
  713. if (h->avctx->colorspace == AVCOL_SPC_RGB) {
  714. *fmt++ = AV_PIX_FMT_GBRP14;
  715. } else
  716. *fmt++ = AV_PIX_FMT_YUV444P14;
  717. } else if (CHROMA422(h))
  718. *fmt++ = AV_PIX_FMT_YUV422P14;
  719. else
  720. *fmt++ = AV_PIX_FMT_YUV420P14;
  721. break;
  722. case 8:
  723. #if CONFIG_H264_VDPAU_HWACCEL
  724. *fmt++ = AV_PIX_FMT_VDPAU;
  725. #endif
  726. #if CONFIG_H264_NVDEC_HWACCEL
  727. *fmt++ = AV_PIX_FMT_CUDA;
  728. #endif
  729. if (CHROMA444(h)) {
  730. if (h->avctx->colorspace == AVCOL_SPC_RGB)
  731. *fmt++ = AV_PIX_FMT_GBRP;
  732. else if (h->avctx->color_range == AVCOL_RANGE_JPEG)
  733. *fmt++ = AV_PIX_FMT_YUVJ444P;
  734. else
  735. *fmt++ = AV_PIX_FMT_YUV444P;
  736. } else if (CHROMA422(h)) {
  737. if (h->avctx->color_range == AVCOL_RANGE_JPEG)
  738. *fmt++ = AV_PIX_FMT_YUVJ422P;
  739. else
  740. *fmt++ = AV_PIX_FMT_YUV422P;
  741. } else {
  742. #if CONFIG_H264_DXVA2_HWACCEL
  743. *fmt++ = AV_PIX_FMT_DXVA2_VLD;
  744. #endif
  745. #if CONFIG_H264_D3D11VA_HWACCEL
  746. *fmt++ = AV_PIX_FMT_D3D11VA_VLD;
  747. *fmt++ = AV_PIX_FMT_D3D11;
  748. #endif
  749. #if CONFIG_H264_VAAPI_HWACCEL
  750. *fmt++ = AV_PIX_FMT_VAAPI;
  751. #endif
  752. #if CONFIG_H264_VIDEOTOOLBOX_HWACCEL
  753. *fmt++ = AV_PIX_FMT_VIDEOTOOLBOX;
  754. #endif
  755. if (h->avctx->codec->pix_fmts)
  756. choices = h->avctx->codec->pix_fmts;
  757. else if (h->avctx->color_range == AVCOL_RANGE_JPEG)
  758. *fmt++ = AV_PIX_FMT_YUVJ420P;
  759. else
  760. *fmt++ = AV_PIX_FMT_YUV420P;
  761. }
  762. break;
  763. default:
  764. av_log(h->avctx, AV_LOG_ERROR,
  765. "Unsupported bit depth %d\n", h->ps.sps->bit_depth_luma);
  766. return AVERROR_INVALIDDATA;
  767. }
  768. *fmt = AV_PIX_FMT_NONE;
  769. for (i=0; choices[i] != AV_PIX_FMT_NONE; i++)
  770. if (choices[i] == h->avctx->pix_fmt && !force_callback)
  771. return choices[i];
  772. return ff_thread_get_format(h->avctx, choices);
  773. }
  774. /* export coded and cropped frame dimensions to AVCodecContext */
  775. static void init_dimensions(H264Context *h)
  776. {
  777. const SPS *sps = (const SPS*)h->ps.sps;
  778. int cr = sps->crop_right;
  779. int cl = sps->crop_left;
  780. int ct = sps->crop_top;
  781. int cb = sps->crop_bottom;
  782. int width = h->width - (cr + cl);
  783. int height = h->height - (ct + cb);
  784. av_assert0(sps->crop_right + sps->crop_left < (unsigned)h->width);
  785. av_assert0(sps->crop_top + sps->crop_bottom < (unsigned)h->height);
  786. /* handle container cropping */
  787. if (h->width_from_caller > 0 && h->height_from_caller > 0 &&
  788. !sps->crop_top && !sps->crop_left &&
  789. FFALIGN(h->width_from_caller, 16) == FFALIGN(width, 16) &&
  790. FFALIGN(h->height_from_caller, 16) == FFALIGN(height, 16) &&
  791. h->width_from_caller <= width &&
  792. h->height_from_caller <= height) {
  793. width = h->width_from_caller;
  794. height = h->height_from_caller;
  795. cl = 0;
  796. ct = 0;
  797. cr = h->width - width;
  798. cb = h->height - height;
  799. } else {
  800. h->width_from_caller = 0;
  801. h->height_from_caller = 0;
  802. }
  803. h->avctx->coded_width = h->width;
  804. h->avctx->coded_height = h->height;
  805. h->avctx->width = width;
  806. h->avctx->height = height;
  807. h->crop_right = cr;
  808. h->crop_left = cl;
  809. h->crop_top = ct;
  810. h->crop_bottom = cb;
  811. }
  812. static int h264_slice_header_init(H264Context *h)
  813. {
  814. const SPS *sps = h->ps.sps;
  815. int i, ret;
  816. if (!sps) {
  817. ret = AVERROR_INVALIDDATA;
  818. goto fail;
  819. }
  820. ff_set_sar(h->avctx, sps->sar);
  821. av_pix_fmt_get_chroma_sub_sample(h->avctx->pix_fmt,
  822. &h->chroma_x_shift, &h->chroma_y_shift);
  823. if (sps->timing_info_present_flag) {
  824. int64_t den = sps->time_scale;
  825. if (h->x264_build < 44U)
  826. den *= 2;
  827. av_reduce(&h->avctx->framerate.den, &h->avctx->framerate.num,
  828. sps->num_units_in_tick * h->avctx->ticks_per_frame, den, 1 << 30);
  829. }
  830. ff_h264_free_tables(h);
  831. h->first_field = 0;
  832. h->prev_interlaced_frame = 1;
  833. init_scan_tables(h);
  834. ret = ff_h264_alloc_tables(h);
  835. if (ret < 0) {
  836. av_log(h->avctx, AV_LOG_ERROR, "Could not allocate memory\n");
  837. goto fail;
  838. }
  839. if (sps->bit_depth_luma < 8 || sps->bit_depth_luma > 14 ||
  840. sps->bit_depth_luma == 11 || sps->bit_depth_luma == 13
  841. ) {
  842. av_log(h->avctx, AV_LOG_ERROR, "Unsupported bit depth %d\n",
  843. sps->bit_depth_luma);
  844. ret = AVERROR_INVALIDDATA;
  845. goto fail;
  846. }
  847. h->cur_bit_depth_luma =
  848. h->avctx->bits_per_raw_sample = sps->bit_depth_luma;
  849. h->cur_chroma_format_idc = sps->chroma_format_idc;
  850. h->pixel_shift = sps->bit_depth_luma > 8;
  851. h->chroma_format_idc = sps->chroma_format_idc;
  852. h->bit_depth_luma = sps->bit_depth_luma;
  853. ff_h264dsp_init(&h->h264dsp, sps->bit_depth_luma,
  854. sps->chroma_format_idc);
  855. ff_h264chroma_init(&h->h264chroma, sps->bit_depth_chroma);
  856. ff_h264qpel_init(&h->h264qpel, sps->bit_depth_luma);
  857. ff_h264_pred_init(&h->hpc, h->avctx->codec_id, sps->bit_depth_luma,
  858. sps->chroma_format_idc);
  859. ff_videodsp_init(&h->vdsp, sps->bit_depth_luma);
  860. if (!HAVE_THREADS || !(h->avctx->active_thread_type & FF_THREAD_SLICE)) {
  861. ret = ff_h264_slice_context_init(h, &h->slice_ctx[0]);
  862. if (ret < 0) {
  863. av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
  864. goto fail;
  865. }
  866. } else {
  867. for (i = 0; i < h->nb_slice_ctx; i++) {
  868. H264SliceContext *sl = &h->slice_ctx[i];
  869. sl->h264 = h;
  870. sl->intra4x4_pred_mode = h->intra4x4_pred_mode + i * 8 * 2 * h->mb_stride;
  871. sl->mvd_table[0] = h->mvd_table[0] + i * 8 * 2 * h->mb_stride;
  872. sl->mvd_table[1] = h->mvd_table[1] + i * 8 * 2 * h->mb_stride;
  873. if ((ret = ff_h264_slice_context_init(h, sl)) < 0) {
  874. av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
  875. goto fail;
  876. }
  877. }
  878. }
  879. h->context_initialized = 1;
  880. return 0;
  881. fail:
  882. ff_h264_free_tables(h);
  883. h->context_initialized = 0;
  884. return ret;
  885. }
  886. static enum AVPixelFormat non_j_pixfmt(enum AVPixelFormat a)
  887. {
  888. switch (a) {
  889. case AV_PIX_FMT_YUVJ420P: return AV_PIX_FMT_YUV420P;
  890. case AV_PIX_FMT_YUVJ422P: return AV_PIX_FMT_YUV422P;
  891. case AV_PIX_FMT_YUVJ444P: return AV_PIX_FMT_YUV444P;
  892. default:
  893. return a;
  894. }
  895. }
  896. static int h264_init_ps(H264Context *h, const H264SliceContext *sl, int first_slice)
  897. {
  898. const SPS *sps;
  899. int needs_reinit = 0, must_reinit, ret;
  900. if (first_slice) {
  901. av_buffer_unref(&h->ps.pps_ref);
  902. h->ps.pps = NULL;
  903. h->ps.pps_ref = av_buffer_ref(h->ps.pps_list[sl->pps_id]);
  904. if (!h->ps.pps_ref)
  905. return AVERROR(ENOMEM);
  906. h->ps.pps = (const PPS*)h->ps.pps_ref->data;
  907. }
  908. if (h->ps.sps != h->ps.pps->sps) {
  909. h->ps.sps = (const SPS*)h->ps.pps->sps;
  910. if (h->mb_width != h->ps.sps->mb_width ||
  911. h->mb_height != h->ps.sps->mb_height ||
  912. h->cur_bit_depth_luma != h->ps.sps->bit_depth_luma ||
  913. h->cur_chroma_format_idc != h->ps.sps->chroma_format_idc
  914. )
  915. needs_reinit = 1;
  916. if (h->bit_depth_luma != h->ps.sps->bit_depth_luma ||
  917. h->chroma_format_idc != h->ps.sps->chroma_format_idc)
  918. needs_reinit = 1;
  919. }
  920. sps = h->ps.sps;
  921. must_reinit = (h->context_initialized &&
  922. ( 16*sps->mb_width != h->avctx->coded_width
  923. || 16*sps->mb_height != h->avctx->coded_height
  924. || h->cur_bit_depth_luma != sps->bit_depth_luma
  925. || h->cur_chroma_format_idc != sps->chroma_format_idc
  926. || h->mb_width != sps->mb_width
  927. || h->mb_height != sps->mb_height
  928. ));
  929. if (h->avctx->pix_fmt == AV_PIX_FMT_NONE
  930. || (non_j_pixfmt(h->avctx->pix_fmt) != non_j_pixfmt(get_pixel_format(h, 0))))
  931. must_reinit = 1;
  932. if (first_slice && av_cmp_q(sps->sar, h->avctx->sample_aspect_ratio))
  933. must_reinit = 1;
  934. if (!h->setup_finished) {
  935. h->avctx->profile = ff_h264_get_profile(sps);
  936. h->avctx->level = sps->level_idc;
  937. h->avctx->refs = sps->ref_frame_count;
  938. h->mb_width = sps->mb_width;
  939. h->mb_height = sps->mb_height;
  940. h->mb_num = h->mb_width * h->mb_height;
  941. h->mb_stride = h->mb_width + 1;
  942. h->b_stride = h->mb_width * 4;
  943. h->chroma_y_shift = sps->chroma_format_idc <= 1; // 400 uses yuv420p
  944. h->width = 16 * h->mb_width;
  945. h->height = 16 * h->mb_height;
  946. init_dimensions(h);
  947. if (sps->video_signal_type_present_flag) {
  948. h->avctx->color_range = sps->full_range > 0 ? AVCOL_RANGE_JPEG
  949. : AVCOL_RANGE_MPEG;
  950. if (sps->colour_description_present_flag) {
  951. if (h->avctx->colorspace != sps->colorspace)
  952. needs_reinit = 1;
  953. h->avctx->color_primaries = sps->color_primaries;
  954. h->avctx->color_trc = sps->color_trc;
  955. h->avctx->colorspace = sps->colorspace;
  956. }
  957. }
  958. if (h->sei.alternative_transfer.present &&
  959. av_color_transfer_name(h->sei.alternative_transfer.preferred_transfer_characteristics) &&
  960. h->sei.alternative_transfer.preferred_transfer_characteristics != AVCOL_TRC_UNSPECIFIED) {
  961. h->avctx->color_trc = h->sei.alternative_transfer.preferred_transfer_characteristics;
  962. }
  963. }
  964. h->avctx->chroma_sample_location = sps->chroma_location;
  965. if (!h->context_initialized || must_reinit || needs_reinit) {
  966. int flush_changes = h->context_initialized;
  967. h->context_initialized = 0;
  968. if (sl != h->slice_ctx) {
  969. av_log(h->avctx, AV_LOG_ERROR,
  970. "changing width %d -> %d / height %d -> %d on "
  971. "slice %d\n",
  972. h->width, h->avctx->coded_width,
  973. h->height, h->avctx->coded_height,
  974. h->current_slice + 1);
  975. return AVERROR_INVALIDDATA;
  976. }
  977. av_assert1(first_slice);
  978. if (flush_changes)
  979. ff_h264_flush_change(h);
  980. if ((ret = get_pixel_format(h, 1)) < 0)
  981. return ret;
  982. h->avctx->pix_fmt = ret;
  983. av_log(h->avctx, AV_LOG_VERBOSE, "Reinit context to %dx%d, "
  984. "pix_fmt: %s\n", h->width, h->height, av_get_pix_fmt_name(h->avctx->pix_fmt));
  985. if ((ret = h264_slice_header_init(h)) < 0) {
  986. av_log(h->avctx, AV_LOG_ERROR,
  987. "h264_slice_header_init() failed\n");
  988. return ret;
  989. }
  990. }
  991. return 0;
  992. }
  993. static int h264_export_frame_props(H264Context *h)
  994. {
  995. const SPS *sps = h->ps.sps;
  996. H264Picture *cur = h->cur_pic_ptr;
  997. AVFrame *out = cur->f;
  998. out->interlaced_frame = 0;
  999. out->repeat_pict = 0;
  1000. /* Signal interlacing information externally. */
  1001. /* Prioritize picture timing SEI information over used
  1002. * decoding process if it exists. */
  1003. if (h->sei.picture_timing.present) {
  1004. int ret = ff_h264_sei_process_picture_timing(&h->sei.picture_timing, sps,
  1005. h->avctx);
  1006. if (ret < 0) {
  1007. av_log(h->avctx, AV_LOG_ERROR, "Error processing a picture timing SEI\n");
  1008. if (h->avctx->err_recognition & AV_EF_EXPLODE)
  1009. return ret;
  1010. h->sei.picture_timing.present = 0;
  1011. }
  1012. }
  1013. if (sps->pic_struct_present_flag && h->sei.picture_timing.present) {
  1014. H264SEIPictureTiming *pt = &h->sei.picture_timing;
  1015. switch (pt->pic_struct) {
  1016. case H264_SEI_PIC_STRUCT_FRAME:
  1017. break;
  1018. case H264_SEI_PIC_STRUCT_TOP_FIELD:
  1019. case H264_SEI_PIC_STRUCT_BOTTOM_FIELD:
  1020. out->interlaced_frame = 1;
  1021. break;
  1022. case H264_SEI_PIC_STRUCT_TOP_BOTTOM:
  1023. case H264_SEI_PIC_STRUCT_BOTTOM_TOP:
  1024. if (FIELD_OR_MBAFF_PICTURE(h))
  1025. out->interlaced_frame = 1;
  1026. else
  1027. // try to flag soft telecine progressive
  1028. out->interlaced_frame = h->prev_interlaced_frame;
  1029. break;
  1030. case H264_SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
  1031. case H264_SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
  1032. /* Signal the possibility of telecined film externally
  1033. * (pic_struct 5,6). From these hints, let the applications
  1034. * decide if they apply deinterlacing. */
  1035. out->repeat_pict = 1;
  1036. break;
  1037. case H264_SEI_PIC_STRUCT_FRAME_DOUBLING:
  1038. out->repeat_pict = 2;
  1039. break;
  1040. case H264_SEI_PIC_STRUCT_FRAME_TRIPLING:
  1041. out->repeat_pict = 4;
  1042. break;
  1043. }
  1044. if ((pt->ct_type & 3) &&
  1045. pt->pic_struct <= H264_SEI_PIC_STRUCT_BOTTOM_TOP)
  1046. out->interlaced_frame = (pt->ct_type & (1 << 1)) != 0;
  1047. } else {
  1048. /* Derive interlacing flag from used decoding process. */
  1049. out->interlaced_frame = FIELD_OR_MBAFF_PICTURE(h);
  1050. }
  1051. h->prev_interlaced_frame = out->interlaced_frame;
  1052. if (cur->field_poc[0] != cur->field_poc[1]) {
  1053. /* Derive top_field_first from field pocs. */
  1054. out->top_field_first = cur->field_poc[0] < cur->field_poc[1];
  1055. } else {
  1056. if (sps->pic_struct_present_flag && h->sei.picture_timing.present) {
  1057. /* Use picture timing SEI information. Even if it is a
  1058. * information of a past frame, better than nothing. */
  1059. if (h->sei.picture_timing.pic_struct == H264_SEI_PIC_STRUCT_TOP_BOTTOM ||
  1060. h->sei.picture_timing.pic_struct == H264_SEI_PIC_STRUCT_TOP_BOTTOM_TOP)
  1061. out->top_field_first = 1;
  1062. else
  1063. out->top_field_first = 0;
  1064. } else if (out->interlaced_frame) {
  1065. /* Default to top field first when pic_struct_present_flag
  1066. * is not set but interlaced frame detected */
  1067. out->top_field_first = 1;
  1068. } else {
  1069. /* Most likely progressive */
  1070. out->top_field_first = 0;
  1071. }
  1072. }
  1073. if (h->sei.frame_packing.present &&
  1074. h->sei.frame_packing.arrangement_type <= 6 &&
  1075. h->sei.frame_packing.content_interpretation_type > 0 &&
  1076. h->sei.frame_packing.content_interpretation_type < 3) {
  1077. H264SEIFramePacking *fp = &h->sei.frame_packing;
  1078. AVStereo3D *stereo = av_stereo3d_create_side_data(out);
  1079. if (stereo) {
  1080. switch (fp->arrangement_type) {
  1081. case H264_SEI_FPA_TYPE_CHECKERBOARD:
  1082. stereo->type = AV_STEREO3D_CHECKERBOARD;
  1083. break;
  1084. case H264_SEI_FPA_TYPE_INTERLEAVE_COLUMN:
  1085. stereo->type = AV_STEREO3D_COLUMNS;
  1086. break;
  1087. case H264_SEI_FPA_TYPE_INTERLEAVE_ROW:
  1088. stereo->type = AV_STEREO3D_LINES;
  1089. break;
  1090. case H264_SEI_FPA_TYPE_SIDE_BY_SIDE:
  1091. if (fp->quincunx_sampling_flag)
  1092. stereo->type = AV_STEREO3D_SIDEBYSIDE_QUINCUNX;
  1093. else
  1094. stereo->type = AV_STEREO3D_SIDEBYSIDE;
  1095. break;
  1096. case H264_SEI_FPA_TYPE_TOP_BOTTOM:
  1097. stereo->type = AV_STEREO3D_TOPBOTTOM;
  1098. break;
  1099. case H264_SEI_FPA_TYPE_INTERLEAVE_TEMPORAL:
  1100. stereo->type = AV_STEREO3D_FRAMESEQUENCE;
  1101. break;
  1102. case H264_SEI_FPA_TYPE_2D:
  1103. stereo->type = AV_STEREO3D_2D;
  1104. break;
  1105. }
  1106. if (fp->content_interpretation_type == 2)
  1107. stereo->flags = AV_STEREO3D_FLAG_INVERT;
  1108. if (fp->arrangement_type == H264_SEI_FPA_TYPE_INTERLEAVE_TEMPORAL) {
  1109. if (fp->current_frame_is_frame0_flag)
  1110. stereo->view = AV_STEREO3D_VIEW_LEFT;
  1111. else
  1112. stereo->view = AV_STEREO3D_VIEW_RIGHT;
  1113. }
  1114. }
  1115. }
  1116. if (h->sei.display_orientation.present &&
  1117. (h->sei.display_orientation.anticlockwise_rotation ||
  1118. h->sei.display_orientation.hflip ||
  1119. h->sei.display_orientation.vflip)) {
  1120. H264SEIDisplayOrientation *o = &h->sei.display_orientation;
  1121. double angle = o->anticlockwise_rotation * 360 / (double) (1 << 16);
  1122. AVFrameSideData *rotation = av_frame_new_side_data(out,
  1123. AV_FRAME_DATA_DISPLAYMATRIX,
  1124. sizeof(int32_t) * 9);
  1125. if (rotation) {
  1126. av_display_rotation_set((int32_t *)rotation->data, angle);
  1127. av_display_matrix_flip((int32_t *)rotation->data,
  1128. o->hflip, o->vflip);
  1129. }
  1130. }
  1131. if (h->sei.afd.present) {
  1132. AVFrameSideData *sd = av_frame_new_side_data(out, AV_FRAME_DATA_AFD,
  1133. sizeof(uint8_t));
  1134. if (sd) {
  1135. *sd->data = h->sei.afd.active_format_description;
  1136. h->sei.afd.present = 0;
  1137. }
  1138. }
  1139. if (h->sei.a53_caption.buf_ref) {
  1140. H264SEIA53Caption *a53 = &h->sei.a53_caption;
  1141. AVFrameSideData *sd = av_frame_new_side_data_from_buf(out, AV_FRAME_DATA_A53_CC, a53->buf_ref);
  1142. if (!sd)
  1143. av_buffer_unref(&a53->buf_ref);
  1144. a53->buf_ref = NULL;
  1145. h->avctx->properties |= FF_CODEC_PROPERTY_CLOSED_CAPTIONS;
  1146. }
  1147. for (int i = 0; i < h->sei.unregistered.nb_buf_ref; i++) {
  1148. H264SEIUnregistered *unreg = &h->sei.unregistered;
  1149. if (unreg->buf_ref[i]) {
  1150. AVFrameSideData *sd = av_frame_new_side_data_from_buf(out,
  1151. AV_FRAME_DATA_SEI_UNREGISTERED,
  1152. unreg->buf_ref[i]);
  1153. if (!sd)
  1154. av_buffer_unref(&unreg->buf_ref[i]);
  1155. unreg->buf_ref[i] = NULL;
  1156. }
  1157. }
  1158. h->sei.unregistered.nb_buf_ref = 0;
  1159. if (h->sei.picture_timing.timecode_cnt > 0) {
  1160. uint32_t *tc_sd;
  1161. char tcbuf[AV_TIMECODE_STR_SIZE];
  1162. AVFrameSideData *tcside = av_frame_new_side_data(out,
  1163. AV_FRAME_DATA_S12M_TIMECODE,
  1164. sizeof(uint32_t)*4);
  1165. if (!tcside)
  1166. return AVERROR(ENOMEM);
  1167. tc_sd = (uint32_t*)tcside->data;
  1168. tc_sd[0] = h->sei.picture_timing.timecode_cnt;
  1169. for (int i = 0; i < tc_sd[0]; i++) {
  1170. int drop = h->sei.picture_timing.timecode[i].dropframe;
  1171. int hh = h->sei.picture_timing.timecode[i].hours;
  1172. int mm = h->sei.picture_timing.timecode[i].minutes;
  1173. int ss = h->sei.picture_timing.timecode[i].seconds;
  1174. int ff = h->sei.picture_timing.timecode[i].frame;
  1175. tc_sd[i + 1] = av_timecode_get_smpte(h->avctx->framerate, drop, hh, mm, ss, ff);
  1176. av_timecode_make_smpte_tc_string2(tcbuf, h->avctx->framerate, tc_sd[i + 1], 0, 0);
  1177. av_dict_set(&out->metadata, "timecode", tcbuf, 0);
  1178. }
  1179. h->sei.picture_timing.timecode_cnt = 0;
  1180. }
  1181. return 0;
  1182. }
  1183. static int h264_select_output_frame(H264Context *h)
  1184. {
  1185. const SPS *sps = h->ps.sps;
  1186. H264Picture *out = h->cur_pic_ptr;
  1187. H264Picture *cur = h->cur_pic_ptr;
  1188. int i, pics, out_of_order, out_idx;
  1189. cur->mmco_reset = h->mmco_reset;
  1190. h->mmco_reset = 0;
  1191. if (sps->bitstream_restriction_flag ||
  1192. h->avctx->strict_std_compliance >= FF_COMPLIANCE_STRICT) {
  1193. h->avctx->has_b_frames = FFMAX(h->avctx->has_b_frames, sps->num_reorder_frames);
  1194. }
  1195. for (i = 0; 1; i++) {
  1196. if(i == MAX_DELAYED_PIC_COUNT || cur->poc < h->last_pocs[i]){
  1197. if(i)
  1198. h->last_pocs[i-1] = cur->poc;
  1199. break;
  1200. } else if(i) {
  1201. h->last_pocs[i-1]= h->last_pocs[i];
  1202. }
  1203. }
  1204. out_of_order = MAX_DELAYED_PIC_COUNT - i;
  1205. if( cur->f->pict_type == AV_PICTURE_TYPE_B
  1206. || (h->last_pocs[MAX_DELAYED_PIC_COUNT-2] > INT_MIN && h->last_pocs[MAX_DELAYED_PIC_COUNT-1] - (int64_t)h->last_pocs[MAX_DELAYED_PIC_COUNT-2] > 2))
  1207. out_of_order = FFMAX(out_of_order, 1);
  1208. if (out_of_order == MAX_DELAYED_PIC_COUNT) {
  1209. av_log(h->avctx, AV_LOG_VERBOSE, "Invalid POC %d<%d\n", cur->poc, h->last_pocs[0]);
  1210. for (i = 1; i < MAX_DELAYED_PIC_COUNT; i++)
  1211. h->last_pocs[i] = INT_MIN;
  1212. h->last_pocs[0] = cur->poc;
  1213. cur->mmco_reset = 1;
  1214. } else if(h->avctx->has_b_frames < out_of_order && !sps->bitstream_restriction_flag){
  1215. int loglevel = h->avctx->frame_number > 1 ? AV_LOG_WARNING : AV_LOG_VERBOSE;
  1216. av_log(h->avctx, loglevel, "Increasing reorder buffer to %d\n", out_of_order);
  1217. h->avctx->has_b_frames = out_of_order;
  1218. }
  1219. pics = 0;
  1220. while (h->delayed_pic[pics])
  1221. pics++;
  1222. av_assert0(pics <= MAX_DELAYED_PIC_COUNT);
  1223. h->delayed_pic[pics++] = cur;
  1224. if (cur->reference == 0)
  1225. cur->reference = DELAYED_PIC_REF;
  1226. out = h->delayed_pic[0];
  1227. out_idx = 0;
  1228. for (i = 1; h->delayed_pic[i] &&
  1229. !h->delayed_pic[i]->f->key_frame &&
  1230. !h->delayed_pic[i]->mmco_reset;
  1231. i++)
  1232. if (h->delayed_pic[i]->poc < out->poc) {
  1233. out = h->delayed_pic[i];
  1234. out_idx = i;
  1235. }
  1236. if (h->avctx->has_b_frames == 0 &&
  1237. (h->delayed_pic[0]->f->key_frame || h->delayed_pic[0]->mmco_reset))
  1238. h->next_outputed_poc = INT_MIN;
  1239. out_of_order = out->poc < h->next_outputed_poc;
  1240. if (out_of_order || pics > h->avctx->has_b_frames) {
  1241. out->reference &= ~DELAYED_PIC_REF;
  1242. for (i = out_idx; h->delayed_pic[i]; i++)
  1243. h->delayed_pic[i] = h->delayed_pic[i + 1];
  1244. }
  1245. if (!out_of_order && pics > h->avctx->has_b_frames) {
  1246. h->next_output_pic = out;
  1247. if (out_idx == 0 && h->delayed_pic[0] && (h->delayed_pic[0]->f->key_frame || h->delayed_pic[0]->mmco_reset)) {
  1248. h->next_outputed_poc = INT_MIN;
  1249. } else
  1250. h->next_outputed_poc = out->poc;
  1251. if (out->recovered) {
  1252. // We have reached an recovery point and all frames after it in
  1253. // display order are "recovered".
  1254. h->frame_recovered |= FRAME_RECOVERED_SEI;
  1255. }
  1256. out->recovered |= !!(h->frame_recovered & FRAME_RECOVERED_SEI);
  1257. if (!out->recovered) {
  1258. if (!(h->avctx->flags & AV_CODEC_FLAG_OUTPUT_CORRUPT) &&
  1259. !(h->avctx->flags2 & AV_CODEC_FLAG2_SHOW_ALL)) {
  1260. h->next_output_pic = NULL;
  1261. } else {
  1262. out->f->flags |= AV_FRAME_FLAG_CORRUPT;
  1263. }
  1264. }
  1265. } else {
  1266. av_log(h->avctx, AV_LOG_DEBUG, "no picture %s\n", out_of_order ? "ooo" : "");
  1267. }
  1268. return 0;
  1269. }
  1270. /* This function is called right after decoding the slice header for a first
  1271. * slice in a field (or a frame). It decides whether we are decoding a new frame
  1272. * or a second field in a pair and does the necessary setup.
  1273. */
  1274. static int h264_field_start(H264Context *h, const H264SliceContext *sl,
  1275. const H2645NAL *nal, int first_slice)
  1276. {
  1277. int i;
  1278. const SPS *sps;
  1279. int last_pic_structure, last_pic_droppable, ret;
  1280. ret = h264_init_ps(h, sl, first_slice);
  1281. if (ret < 0)
  1282. return ret;
  1283. sps = h->ps.sps;
  1284. if (sps && sps->bitstream_restriction_flag &&
  1285. h->avctx->has_b_frames < sps->num_reorder_frames) {
  1286. h->avctx->has_b_frames = sps->num_reorder_frames;
  1287. }
  1288. last_pic_droppable = h->droppable;
  1289. last_pic_structure = h->picture_structure;
  1290. h->droppable = (nal->ref_idc == 0);
  1291. h->picture_structure = sl->picture_structure;
  1292. h->poc.frame_num = sl->frame_num;
  1293. h->poc.poc_lsb = sl->poc_lsb;
  1294. h->poc.delta_poc_bottom = sl->delta_poc_bottom;
  1295. h->poc.delta_poc[0] = sl->delta_poc[0];
  1296. h->poc.delta_poc[1] = sl->delta_poc[1];
  1297. /* Shorten frame num gaps so we don't have to allocate reference
  1298. * frames just to throw them away */
  1299. if (h->poc.frame_num != h->poc.prev_frame_num) {
  1300. int unwrap_prev_frame_num = h->poc.prev_frame_num;
  1301. int max_frame_num = 1 << sps->log2_max_frame_num;
  1302. if (unwrap_prev_frame_num > h->poc.frame_num)
  1303. unwrap_prev_frame_num -= max_frame_num;
  1304. if ((h->poc.frame_num - unwrap_prev_frame_num) > sps->ref_frame_count) {
  1305. unwrap_prev_frame_num = (h->poc.frame_num - sps->ref_frame_count) - 1;
  1306. if (unwrap_prev_frame_num < 0)
  1307. unwrap_prev_frame_num += max_frame_num;
  1308. h->poc.prev_frame_num = unwrap_prev_frame_num;
  1309. }
  1310. }
  1311. /* See if we have a decoded first field looking for a pair...
  1312. * Here, we're using that to see if we should mark previously
  1313. * decode frames as "finished".
  1314. * We have to do that before the "dummy" in-between frame allocation,
  1315. * since that can modify h->cur_pic_ptr. */
  1316. if (h->first_field) {
  1317. int last_field = last_pic_structure == PICT_BOTTOM_FIELD;
  1318. av_assert0(h->cur_pic_ptr);
  1319. av_assert0(h->cur_pic_ptr->f->buf[0]);
  1320. assert(h->cur_pic_ptr->reference != DELAYED_PIC_REF);
  1321. /* Mark old field/frame as completed */
  1322. if (h->cur_pic_ptr->tf.owner[last_field] == h->avctx) {
  1323. ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, last_field);
  1324. }
  1325. /* figure out if we have a complementary field pair */
  1326. if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
  1327. /* Previous field is unmatched. Don't display it, but let it
  1328. * remain for reference if marked as such. */
  1329. if (last_pic_structure != PICT_FRAME) {
  1330. ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
  1331. last_pic_structure == PICT_TOP_FIELD);
  1332. }
  1333. } else {
  1334. if (h->cur_pic_ptr->frame_num != h->poc.frame_num) {
  1335. /* This and previous field were reference, but had
  1336. * different frame_nums. Consider this field first in
  1337. * pair. Throw away previous field except for reference
  1338. * purposes. */
  1339. if (last_pic_structure != PICT_FRAME) {
  1340. ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
  1341. last_pic_structure == PICT_TOP_FIELD);
  1342. }
  1343. } else {
  1344. /* Second field in complementary pair */
  1345. if (!((last_pic_structure == PICT_TOP_FIELD &&
  1346. h->picture_structure == PICT_BOTTOM_FIELD) ||
  1347. (last_pic_structure == PICT_BOTTOM_FIELD &&
  1348. h->picture_structure == PICT_TOP_FIELD))) {
  1349. av_log(h->avctx, AV_LOG_ERROR,
  1350. "Invalid field mode combination %d/%d\n",
  1351. last_pic_structure, h->picture_structure);
  1352. h->picture_structure = last_pic_structure;
  1353. h->droppable = last_pic_droppable;
  1354. return AVERROR_INVALIDDATA;
  1355. } else if (last_pic_droppable != h->droppable) {
  1356. avpriv_request_sample(h->avctx,
  1357. "Found reference and non-reference fields in the same frame, which");
  1358. h->picture_structure = last_pic_structure;
  1359. h->droppable = last_pic_droppable;
  1360. return AVERROR_PATCHWELCOME;
  1361. }
  1362. }
  1363. }
  1364. }
  1365. while (h->poc.frame_num != h->poc.prev_frame_num && !h->first_field &&
  1366. h->poc.frame_num != (h->poc.prev_frame_num + 1) % (1 << sps->log2_max_frame_num)) {
  1367. H264Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
  1368. av_log(h->avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n",
  1369. h->poc.frame_num, h->poc.prev_frame_num);
  1370. if (!sps->gaps_in_frame_num_allowed_flag)
  1371. for(i=0; i<FF_ARRAY_ELEMS(h->last_pocs); i++)
  1372. h->last_pocs[i] = INT_MIN;
  1373. ret = h264_frame_start(h);
  1374. if (ret < 0) {
  1375. h->first_field = 0;
  1376. return ret;
  1377. }
  1378. h->poc.prev_frame_num++;
  1379. h->poc.prev_frame_num %= 1 << sps->log2_max_frame_num;
  1380. h->cur_pic_ptr->frame_num = h->poc.prev_frame_num;
  1381. h->cur_pic_ptr->invalid_gap = !sps->gaps_in_frame_num_allowed_flag;
  1382. ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 0);
  1383. ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 1);
  1384. h->explicit_ref_marking = 0;
  1385. ret = ff_h264_execute_ref_pic_marking(h);
  1386. if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
  1387. return ret;
  1388. /* Error concealment: If a ref is missing, copy the previous ref
  1389. * in its place.
  1390. * FIXME: Avoiding a memcpy would be nice, but ref handling makes
  1391. * many assumptions about there being no actual duplicates.
  1392. * FIXME: This does not copy padding for out-of-frame motion
  1393. * vectors. Given we are concealing a lost frame, this probably
  1394. * is not noticeable by comparison, but it should be fixed. */
  1395. if (h->short_ref_count) {
  1396. int c[4] = {
  1397. 1<<(h->ps.sps->bit_depth_luma-1),
  1398. 1<<(h->ps.sps->bit_depth_chroma-1),
  1399. 1<<(h->ps.sps->bit_depth_chroma-1),
  1400. -1
  1401. };
  1402. if (prev &&
  1403. h->short_ref[0]->f->width == prev->f->width &&
  1404. h->short_ref[0]->f->height == prev->f->height &&
  1405. h->short_ref[0]->f->format == prev->f->format) {
  1406. ff_thread_await_progress(&prev->tf, INT_MAX, 0);
  1407. if (prev->field_picture)
  1408. ff_thread_await_progress(&prev->tf, INT_MAX, 1);
  1409. ff_thread_release_buffer(h->avctx, &h->short_ref[0]->tf);
  1410. h->short_ref[0]->tf.f = h->short_ref[0]->f;
  1411. ret = ff_thread_ref_frame(&h->short_ref[0]->tf, &prev->tf);
  1412. if (ret < 0)
  1413. return ret;
  1414. h->short_ref[0]->poc = prev->poc + 2U;
  1415. ff_thread_report_progress(&h->short_ref[0]->tf, INT_MAX, 0);
  1416. if (h->short_ref[0]->field_picture)
  1417. ff_thread_report_progress(&h->short_ref[0]->tf, INT_MAX, 1);
  1418. } else if (!h->frame_recovered && !h->avctx->hwaccel)
  1419. ff_color_frame(h->short_ref[0]->f, c);
  1420. h->short_ref[0]->frame_num = h->poc.prev_frame_num;
  1421. }
  1422. }
  1423. /* See if we have a decoded first field looking for a pair...
  1424. * We're using that to see whether to continue decoding in that
  1425. * frame, or to allocate a new one. */
  1426. if (h->first_field) {
  1427. av_assert0(h->cur_pic_ptr);
  1428. av_assert0(h->cur_pic_ptr->f->buf[0]);
  1429. assert(h->cur_pic_ptr->reference != DELAYED_PIC_REF);
  1430. /* figure out if we have a complementary field pair */
  1431. if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
  1432. /* Previous field is unmatched. Don't display it, but let it
  1433. * remain for reference if marked as such. */
  1434. h->missing_fields ++;
  1435. h->cur_pic_ptr = NULL;
  1436. h->first_field = FIELD_PICTURE(h);
  1437. } else {
  1438. h->missing_fields = 0;
  1439. if (h->cur_pic_ptr->frame_num != h->poc.frame_num) {
  1440. ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
  1441. h->picture_structure==PICT_BOTTOM_FIELD);
  1442. /* This and the previous field had different frame_nums.
  1443. * Consider this field first in pair. Throw away previous
  1444. * one except for reference purposes. */
  1445. h->first_field = 1;
  1446. h->cur_pic_ptr = NULL;
  1447. } else if (h->cur_pic_ptr->reference & DELAYED_PIC_REF) {
  1448. /* This frame was already output, we cannot draw into it
  1449. * anymore.
  1450. */
  1451. h->first_field = 1;
  1452. h->cur_pic_ptr = NULL;
  1453. } else {
  1454. /* Second field in complementary pair */
  1455. h->first_field = 0;
  1456. }
  1457. }
  1458. } else {
  1459. /* Frame or first field in a potentially complementary pair */
  1460. h->first_field = FIELD_PICTURE(h);
  1461. }
  1462. if (!FIELD_PICTURE(h) || h->first_field) {
  1463. if (h264_frame_start(h) < 0) {
  1464. h->first_field = 0;
  1465. return AVERROR_INVALIDDATA;
  1466. }
  1467. } else {
  1468. int field = h->picture_structure == PICT_BOTTOM_FIELD;
  1469. release_unused_pictures(h, 0);
  1470. h->cur_pic_ptr->tf.owner[field] = h->avctx;
  1471. }
  1472. /* Some macroblocks can be accessed before they're available in case
  1473. * of lost slices, MBAFF or threading. */
  1474. if (FIELD_PICTURE(h)) {
  1475. for(i = (h->picture_structure == PICT_BOTTOM_FIELD); i<h->mb_height; i++)
  1476. memset(h->slice_table + i*h->mb_stride, -1, (h->mb_stride - (i+1==h->mb_height)) * sizeof(*h->slice_table));
  1477. } else {
  1478. memset(h->slice_table, -1,
  1479. (h->mb_height * h->mb_stride - 1) * sizeof(*h->slice_table));
  1480. }
  1481. ret = ff_h264_init_poc(h->cur_pic_ptr->field_poc, &h->cur_pic_ptr->poc,
  1482. h->ps.sps, &h->poc, h->picture_structure, nal->ref_idc);
  1483. if (ret < 0)
  1484. return ret;
  1485. memcpy(h->mmco, sl->mmco, sl->nb_mmco * sizeof(*h->mmco));
  1486. h->nb_mmco = sl->nb_mmco;
  1487. h->explicit_ref_marking = sl->explicit_ref_marking;
  1488. h->picture_idr = nal->type == H264_NAL_IDR_SLICE;
  1489. if (h->sei.recovery_point.recovery_frame_cnt >= 0) {
  1490. const int sei_recovery_frame_cnt = h->sei.recovery_point.recovery_frame_cnt;
  1491. if (h->poc.frame_num != sei_recovery_frame_cnt || sl->slice_type_nos != AV_PICTURE_TYPE_I)
  1492. h->valid_recovery_point = 1;
  1493. if ( h->recovery_frame < 0
  1494. || av_mod_uintp2(h->recovery_frame - h->poc.frame_num, h->ps.sps->log2_max_frame_num) > sei_recovery_frame_cnt) {
  1495. h->recovery_frame = av_mod_uintp2(h->poc.frame_num + sei_recovery_frame_cnt, h->ps.sps->log2_max_frame_num);
  1496. if (!h->valid_recovery_point)
  1497. h->recovery_frame = h->poc.frame_num;
  1498. }
  1499. }
  1500. h->cur_pic_ptr->f->key_frame |= (nal->type == H264_NAL_IDR_SLICE);
  1501. if (nal->type == H264_NAL_IDR_SLICE ||
  1502. (h->recovery_frame == h->poc.frame_num && nal->ref_idc)) {
  1503. h->recovery_frame = -1;
  1504. h->cur_pic_ptr->recovered = 1;
  1505. }
  1506. // If we have an IDR, all frames after it in decoded order are
  1507. // "recovered".
  1508. if (nal->type == H264_NAL_IDR_SLICE)
  1509. h->frame_recovered |= FRAME_RECOVERED_IDR;
  1510. #if 1
  1511. h->cur_pic_ptr->recovered |= h->frame_recovered;
  1512. #else
  1513. h->cur_pic_ptr->recovered |= !!(h->frame_recovered & FRAME_RECOVERED_IDR);
  1514. #endif
  1515. /* Set the frame properties/side data. Only done for the second field in
  1516. * field coded frames, since some SEI information is present for each field
  1517. * and is merged by the SEI parsing code. */
  1518. if (!FIELD_PICTURE(h) || !h->first_field || h->missing_fields > 1) {
  1519. ret = h264_export_frame_props(h);
  1520. if (ret < 0)
  1521. return ret;
  1522. ret = h264_select_output_frame(h);
  1523. if (ret < 0)
  1524. return ret;
  1525. }
  1526. return 0;
  1527. }
  1528. static int h264_slice_header_parse(const H264Context *h, H264SliceContext *sl,
  1529. const H2645NAL *nal)
  1530. {
  1531. const SPS *sps;
  1532. const PPS *pps;
  1533. int ret;
  1534. unsigned int slice_type, tmp, i;
  1535. int field_pic_flag, bottom_field_flag;
  1536. int first_slice = sl == h->slice_ctx && !h->current_slice;
  1537. int picture_structure;
  1538. if (first_slice)
  1539. av_assert0(!h->setup_finished);
  1540. sl->first_mb_addr = get_ue_golomb_long(&sl->gb);
  1541. slice_type = get_ue_golomb_31(&sl->gb);
  1542. if (slice_type > 9) {
  1543. av_log(h->avctx, AV_LOG_ERROR,
  1544. "slice type %d too large at %d\n",
  1545. slice_type, sl->first_mb_addr);
  1546. return AVERROR_INVALIDDATA;
  1547. }
  1548. if (slice_type > 4) {
  1549. slice_type -= 5;
  1550. sl->slice_type_fixed = 1;
  1551. } else
  1552. sl->slice_type_fixed = 0;
  1553. slice_type = ff_h264_golomb_to_pict_type[slice_type];
  1554. sl->slice_type = slice_type;
  1555. sl->slice_type_nos = slice_type & 3;
  1556. if (nal->type == H264_NAL_IDR_SLICE &&
  1557. sl->slice_type_nos != AV_PICTURE_TYPE_I) {
  1558. av_log(h->avctx, AV_LOG_ERROR, "A non-intra slice in an IDR NAL unit.\n");
  1559. return AVERROR_INVALIDDATA;
  1560. }
  1561. sl->pps_id = get_ue_golomb(&sl->gb);
  1562. if (sl->pps_id >= MAX_PPS_COUNT) {
  1563. av_log(h->avctx, AV_LOG_ERROR, "pps_id %u out of range\n", sl->pps_id);
  1564. return AVERROR_INVALIDDATA;
  1565. }
  1566. if (!h->ps.pps_list[sl->pps_id]) {
  1567. av_log(h->avctx, AV_LOG_ERROR,
  1568. "non-existing PPS %u referenced\n",
  1569. sl->pps_id);
  1570. return AVERROR_INVALIDDATA;
  1571. }
  1572. pps = (const PPS*)h->ps.pps_list[sl->pps_id]->data;
  1573. sps = pps->sps;
  1574. sl->frame_num = get_bits(&sl->gb, sps->log2_max_frame_num);
  1575. if (!first_slice) {
  1576. if (h->poc.frame_num != sl->frame_num) {
  1577. av_log(h->avctx, AV_LOG_ERROR, "Frame num change from %d to %d\n",
  1578. h->poc.frame_num, sl->frame_num);
  1579. return AVERROR_INVALIDDATA;
  1580. }
  1581. }
  1582. sl->mb_mbaff = 0;
  1583. if (sps->frame_mbs_only_flag) {
  1584. picture_structure = PICT_FRAME;
  1585. } else {
  1586. if (!sps->direct_8x8_inference_flag && slice_type == AV_PICTURE_TYPE_B) {
  1587. av_log(h->avctx, AV_LOG_ERROR, "This stream was generated by a broken encoder, invalid 8x8 inference\n");
  1588. return -1;
  1589. }
  1590. field_pic_flag = get_bits1(&sl->gb);
  1591. if (field_pic_flag) {
  1592. bottom_field_flag = get_bits1(&sl->gb);
  1593. picture_structure = PICT_TOP_FIELD + bottom_field_flag;
  1594. } else {
  1595. picture_structure = PICT_FRAME;
  1596. }
  1597. }
  1598. sl->picture_structure = picture_structure;
  1599. sl->mb_field_decoding_flag = picture_structure != PICT_FRAME;
  1600. if (picture_structure == PICT_FRAME) {
  1601. sl->curr_pic_num = sl->frame_num;
  1602. sl->max_pic_num = 1 << sps->log2_max_frame_num;
  1603. } else {
  1604. sl->curr_pic_num = 2 * sl->frame_num + 1;
  1605. sl->max_pic_num = 1 << (sps->log2_max_frame_num + 1);
  1606. }
  1607. if (nal->type == H264_NAL_IDR_SLICE)
  1608. get_ue_golomb_long(&sl->gb); /* idr_pic_id */
  1609. if (sps->poc_type == 0) {
  1610. sl->poc_lsb = get_bits(&sl->gb, sps->log2_max_poc_lsb);
  1611. if (pps->pic_order_present == 1 && picture_structure == PICT_FRAME)
  1612. sl->delta_poc_bottom = get_se_golomb(&sl->gb);
  1613. }
  1614. if (sps->poc_type == 1 && !sps->delta_pic_order_always_zero_flag) {
  1615. sl->delta_poc[0] = get_se_golomb(&sl->gb);
  1616. if (pps->pic_order_present == 1 && picture_structure == PICT_FRAME)
  1617. sl->delta_poc[1] = get_se_golomb(&sl->gb);
  1618. }
  1619. sl->redundant_pic_count = 0;
  1620. if (pps->redundant_pic_cnt_present)
  1621. sl->redundant_pic_count = get_ue_golomb(&sl->gb);
  1622. if (sl->slice_type_nos == AV_PICTURE_TYPE_B)
  1623. sl->direct_spatial_mv_pred = get_bits1(&sl->gb);
  1624. ret = ff_h264_parse_ref_count(&sl->list_count, sl->ref_count,
  1625. &sl->gb, pps, sl->slice_type_nos,
  1626. picture_structure, h->avctx);
  1627. if (ret < 0)
  1628. return ret;
  1629. if (sl->slice_type_nos != AV_PICTURE_TYPE_I) {
  1630. ret = ff_h264_decode_ref_pic_list_reordering(sl, h->avctx);
  1631. if (ret < 0) {
  1632. sl->ref_count[1] = sl->ref_count[0] = 0;
  1633. return ret;
  1634. }
  1635. }
  1636. sl->pwt.use_weight = 0;
  1637. for (i = 0; i < 2; i++) {
  1638. sl->pwt.luma_weight_flag[i] = 0;
  1639. sl->pwt.chroma_weight_flag[i] = 0;
  1640. }
  1641. if ((pps->weighted_pred && sl->slice_type_nos == AV_PICTURE_TYPE_P) ||
  1642. (pps->weighted_bipred_idc == 1 &&
  1643. sl->slice_type_nos == AV_PICTURE_TYPE_B)) {
  1644. ret = ff_h264_pred_weight_table(&sl->gb, sps, sl->ref_count,
  1645. sl->slice_type_nos, &sl->pwt,
  1646. picture_structure, h->avctx);
  1647. if (ret < 0)
  1648. return ret;
  1649. }
  1650. sl->explicit_ref_marking = 0;
  1651. if (nal->ref_idc) {
  1652. ret = ff_h264_decode_ref_pic_marking(sl, &sl->gb, nal, h->avctx);
  1653. if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
  1654. return AVERROR_INVALIDDATA;
  1655. }
  1656. if (sl->slice_type_nos != AV_PICTURE_TYPE_I && pps->cabac) {
  1657. tmp = get_ue_golomb_31(&sl->gb);
  1658. if (tmp > 2) {
  1659. av_log(h->avctx, AV_LOG_ERROR, "cabac_init_idc %u overflow\n", tmp);
  1660. return AVERROR_INVALIDDATA;
  1661. }
  1662. sl->cabac_init_idc = tmp;
  1663. }
  1664. sl->last_qscale_diff = 0;
  1665. tmp = pps->init_qp + (unsigned)get_se_golomb(&sl->gb);
  1666. if (tmp > 51 + 6 * (sps->bit_depth_luma - 8)) {
  1667. av_log(h->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
  1668. return AVERROR_INVALIDDATA;
  1669. }
  1670. sl->qscale = tmp;
  1671. sl->chroma_qp[0] = get_chroma_qp(pps, 0, sl->qscale);
  1672. sl->chroma_qp[1] = get_chroma_qp(pps, 1, sl->qscale);
  1673. // FIXME qscale / qp ... stuff
  1674. if (sl->slice_type == AV_PICTURE_TYPE_SP)
  1675. get_bits1(&sl->gb); /* sp_for_switch_flag */
  1676. if (sl->slice_type == AV_PICTURE_TYPE_SP ||
  1677. sl->slice_type == AV_PICTURE_TYPE_SI)
  1678. get_se_golomb(&sl->gb); /* slice_qs_delta */
  1679. sl->deblocking_filter = 1;
  1680. sl->slice_alpha_c0_offset = 0;
  1681. sl->slice_beta_offset = 0;
  1682. if (pps->deblocking_filter_parameters_present) {
  1683. tmp = get_ue_golomb_31(&sl->gb);
  1684. if (tmp > 2) {
  1685. av_log(h->avctx, AV_LOG_ERROR,
  1686. "deblocking_filter_idc %u out of range\n", tmp);
  1687. return AVERROR_INVALIDDATA;
  1688. }
  1689. sl->deblocking_filter = tmp;
  1690. if (sl->deblocking_filter < 2)
  1691. sl->deblocking_filter ^= 1; // 1<->0
  1692. if (sl->deblocking_filter) {
  1693. int slice_alpha_c0_offset_div2 = get_se_golomb(&sl->gb);
  1694. int slice_beta_offset_div2 = get_se_golomb(&sl->gb);
  1695. if (slice_alpha_c0_offset_div2 > 6 ||
  1696. slice_alpha_c0_offset_div2 < -6 ||
  1697. slice_beta_offset_div2 > 6 ||
  1698. slice_beta_offset_div2 < -6) {
  1699. av_log(h->avctx, AV_LOG_ERROR,
  1700. "deblocking filter parameters %d %d out of range\n",
  1701. slice_alpha_c0_offset_div2, slice_beta_offset_div2);
  1702. return AVERROR_INVALIDDATA;
  1703. }
  1704. sl->slice_alpha_c0_offset = slice_alpha_c0_offset_div2 * 2;
  1705. sl->slice_beta_offset = slice_beta_offset_div2 * 2;
  1706. }
  1707. }
  1708. return 0;
  1709. }
  1710. /* do all the per-slice initialization needed before we can start decoding the
  1711. * actual MBs */
  1712. static int h264_slice_init(H264Context *h, H264SliceContext *sl,
  1713. const H2645NAL *nal)
  1714. {
  1715. int i, j, ret = 0;
  1716. if (h->picture_idr && nal->type != H264_NAL_IDR_SLICE) {
  1717. av_log(h->avctx, AV_LOG_ERROR, "Invalid mix of IDR and non-IDR slices\n");
  1718. return AVERROR_INVALIDDATA;
  1719. }
  1720. av_assert1(h->mb_num == h->mb_width * h->mb_height);
  1721. if (sl->first_mb_addr << FIELD_OR_MBAFF_PICTURE(h) >= h->mb_num ||
  1722. sl->first_mb_addr >= h->mb_num) {
  1723. av_log(h->avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
  1724. return AVERROR_INVALIDDATA;
  1725. }
  1726. sl->resync_mb_x = sl->mb_x = sl->first_mb_addr % h->mb_width;
  1727. sl->resync_mb_y = sl->mb_y = (sl->first_mb_addr / h->mb_width) <<
  1728. FIELD_OR_MBAFF_PICTURE(h);
  1729. if (h->picture_structure == PICT_BOTTOM_FIELD)
  1730. sl->resync_mb_y = sl->mb_y = sl->mb_y + 1;
  1731. av_assert1(sl->mb_y < h->mb_height);
  1732. ret = ff_h264_build_ref_list(h, sl);
  1733. if (ret < 0)
  1734. return ret;
  1735. if (h->ps.pps->weighted_bipred_idc == 2 &&
  1736. sl->slice_type_nos == AV_PICTURE_TYPE_B) {
  1737. implicit_weight_table(h, sl, -1);
  1738. if (FRAME_MBAFF(h)) {
  1739. implicit_weight_table(h, sl, 0);
  1740. implicit_weight_table(h, sl, 1);
  1741. }
  1742. }
  1743. if (sl->slice_type_nos == AV_PICTURE_TYPE_B && !sl->direct_spatial_mv_pred)
  1744. ff_h264_direct_dist_scale_factor(h, sl);
  1745. if (!h->setup_finished)
  1746. ff_h264_direct_ref_list_init(h, sl);
  1747. if (h->avctx->skip_loop_filter >= AVDISCARD_ALL ||
  1748. (h->avctx->skip_loop_filter >= AVDISCARD_NONKEY &&
  1749. h->nal_unit_type != H264_NAL_IDR_SLICE) ||
  1750. (h->avctx->skip_loop_filter >= AVDISCARD_NONINTRA &&
  1751. sl->slice_type_nos != AV_PICTURE_TYPE_I) ||
  1752. (h->avctx->skip_loop_filter >= AVDISCARD_BIDIR &&
  1753. sl->slice_type_nos == AV_PICTURE_TYPE_B) ||
  1754. (h->avctx->skip_loop_filter >= AVDISCARD_NONREF &&
  1755. nal->ref_idc == 0))
  1756. sl->deblocking_filter = 0;
  1757. if (sl->deblocking_filter == 1 && h->nb_slice_ctx > 1) {
  1758. if (h->avctx->flags2 & AV_CODEC_FLAG2_FAST) {
  1759. /* Cheat slightly for speed:
  1760. * Do not bother to deblock across slices. */
  1761. sl->deblocking_filter = 2;
  1762. } else {
  1763. h->postpone_filter = 1;
  1764. }
  1765. }
  1766. sl->qp_thresh = 15 -
  1767. FFMIN(sl->slice_alpha_c0_offset, sl->slice_beta_offset) -
  1768. FFMAX3(0,
  1769. h->ps.pps->chroma_qp_index_offset[0],
  1770. h->ps.pps->chroma_qp_index_offset[1]) +
  1771. 6 * (h->ps.sps->bit_depth_luma - 8);
  1772. sl->slice_num = ++h->current_slice;
  1773. if (sl->slice_num)
  1774. h->slice_row[(sl->slice_num-1)&(MAX_SLICES-1)]= sl->resync_mb_y;
  1775. if ( h->slice_row[sl->slice_num&(MAX_SLICES-1)] + 3 >= sl->resync_mb_y
  1776. && h->slice_row[sl->slice_num&(MAX_SLICES-1)] <= sl->resync_mb_y
  1777. && sl->slice_num >= MAX_SLICES) {
  1778. //in case of ASO this check needs to be updated depending on how we decide to assign slice numbers in this case
  1779. 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);
  1780. }
  1781. for (j = 0; j < 2; j++) {
  1782. int id_list[16];
  1783. int *ref2frm = h->ref2frm[sl->slice_num & (MAX_SLICES - 1)][j];
  1784. for (i = 0; i < 16; i++) {
  1785. id_list[i] = 60;
  1786. if (j < sl->list_count && i < sl->ref_count[j] &&
  1787. sl->ref_list[j][i].parent->f->buf[0]) {
  1788. int k;
  1789. AVBuffer *buf = sl->ref_list[j][i].parent->f->buf[0]->buffer;
  1790. for (k = 0; k < h->short_ref_count; k++)
  1791. if (h->short_ref[k]->f->buf[0]->buffer == buf) {
  1792. id_list[i] = k;
  1793. break;
  1794. }
  1795. for (k = 0; k < h->long_ref_count; k++)
  1796. if (h->long_ref[k] && h->long_ref[k]->f->buf[0]->buffer == buf) {
  1797. id_list[i] = h->short_ref_count + k;
  1798. break;
  1799. }
  1800. }
  1801. }
  1802. ref2frm[0] =
  1803. ref2frm[1] = -1;
  1804. for (i = 0; i < 16; i++)
  1805. ref2frm[i + 2] = 4 * id_list[i] + (sl->ref_list[j][i].reference & 3);
  1806. ref2frm[18 + 0] =
  1807. ref2frm[18 + 1] = -1;
  1808. for (i = 16; i < 48; i++)
  1809. ref2frm[i + 4] = 4 * id_list[(i - 16) >> 1] +
  1810. (sl->ref_list[j][i].reference & 3);
  1811. }
  1812. if (h->avctx->debug & FF_DEBUG_PICT_INFO) {
  1813. av_log(h->avctx, AV_LOG_DEBUG,
  1814. "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",
  1815. sl->slice_num,
  1816. (h->picture_structure == PICT_FRAME ? "F" : h->picture_structure == PICT_TOP_FIELD ? "T" : "B"),
  1817. sl->mb_y * h->mb_width + sl->mb_x,
  1818. av_get_picture_type_char(sl->slice_type),
  1819. sl->slice_type_fixed ? " fix" : "",
  1820. nal->type == H264_NAL_IDR_SLICE ? " IDR" : "",
  1821. h->poc.frame_num,
  1822. h->cur_pic_ptr->field_poc[0],
  1823. h->cur_pic_ptr->field_poc[1],
  1824. sl->ref_count[0], sl->ref_count[1],
  1825. sl->qscale,
  1826. sl->deblocking_filter,
  1827. sl->slice_alpha_c0_offset, sl->slice_beta_offset,
  1828. sl->pwt.use_weight,
  1829. sl->pwt.use_weight == 1 && sl->pwt.use_weight_chroma ? "c" : "",
  1830. sl->slice_type == AV_PICTURE_TYPE_B ? (sl->direct_spatial_mv_pred ? "SPAT" : "TEMP") : "");
  1831. }
  1832. return 0;
  1833. }
  1834. int ff_h264_queue_decode_slice(H264Context *h, const H2645NAL *nal)
  1835. {
  1836. H264SliceContext *sl = h->slice_ctx + h->nb_slice_ctx_queued;
  1837. int first_slice = sl == h->slice_ctx && !h->current_slice;
  1838. int ret;
  1839. sl->gb = nal->gb;
  1840. ret = h264_slice_header_parse(h, sl, nal);
  1841. if (ret < 0)
  1842. return ret;
  1843. // discard redundant pictures
  1844. if (sl->redundant_pic_count > 0) {
  1845. sl->ref_count[0] = sl->ref_count[1] = 0;
  1846. return 0;
  1847. }
  1848. if (sl->first_mb_addr == 0 || !h->current_slice) {
  1849. if (h->setup_finished) {
  1850. av_log(h->avctx, AV_LOG_ERROR, "Too many fields\n");
  1851. return AVERROR_INVALIDDATA;
  1852. }
  1853. }
  1854. if (sl->first_mb_addr == 0) { // FIXME better field boundary detection
  1855. if (h->current_slice) {
  1856. // this slice starts a new field
  1857. // first decode any pending queued slices
  1858. if (h->nb_slice_ctx_queued) {
  1859. H264SliceContext tmp_ctx;
  1860. ret = ff_h264_execute_decode_slices(h);
  1861. if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
  1862. return ret;
  1863. memcpy(&tmp_ctx, h->slice_ctx, sizeof(tmp_ctx));
  1864. memcpy(h->slice_ctx, sl, sizeof(tmp_ctx));
  1865. memcpy(sl, &tmp_ctx, sizeof(tmp_ctx));
  1866. sl = h->slice_ctx;
  1867. }
  1868. if (h->cur_pic_ptr && FIELD_PICTURE(h) && h->first_field) {
  1869. ret = ff_h264_field_end(h, h->slice_ctx, 1);
  1870. if (ret < 0)
  1871. return ret;
  1872. } else if (h->cur_pic_ptr && !FIELD_PICTURE(h) && !h->first_field && h->nal_unit_type == H264_NAL_IDR_SLICE) {
  1873. av_log(h, AV_LOG_WARNING, "Broken frame packetizing\n");
  1874. ret = ff_h264_field_end(h, h->slice_ctx, 1);
  1875. ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 0);
  1876. ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 1);
  1877. h->cur_pic_ptr = NULL;
  1878. if (ret < 0)
  1879. return ret;
  1880. } else
  1881. return AVERROR_INVALIDDATA;
  1882. }
  1883. if (!h->first_field) {
  1884. if (h->cur_pic_ptr && !h->droppable) {
  1885. ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
  1886. h->picture_structure == PICT_BOTTOM_FIELD);
  1887. }
  1888. h->cur_pic_ptr = NULL;
  1889. }
  1890. }
  1891. if (!h->current_slice)
  1892. av_assert0(sl == h->slice_ctx);
  1893. if (h->current_slice == 0 && !h->first_field) {
  1894. if (
  1895. (h->avctx->skip_frame >= AVDISCARD_NONREF && !h->nal_ref_idc) ||
  1896. (h->avctx->skip_frame >= AVDISCARD_BIDIR && sl->slice_type_nos == AV_PICTURE_TYPE_B) ||
  1897. (h->avctx->skip_frame >= AVDISCARD_NONINTRA && sl->slice_type_nos != AV_PICTURE_TYPE_I) ||
  1898. (h->avctx->skip_frame >= AVDISCARD_NONKEY && h->nal_unit_type != H264_NAL_IDR_SLICE && h->sei.recovery_point.recovery_frame_cnt < 0) ||
  1899. h->avctx->skip_frame >= AVDISCARD_ALL) {
  1900. return 0;
  1901. }
  1902. }
  1903. if (!first_slice) {
  1904. const PPS *pps = (const PPS*)h->ps.pps_list[sl->pps_id]->data;
  1905. if (h->ps.pps->sps_id != pps->sps_id ||
  1906. h->ps.pps->transform_8x8_mode != pps->transform_8x8_mode /*||
  1907. (h->setup_finished && h->ps.pps != pps)*/) {
  1908. av_log(h->avctx, AV_LOG_ERROR, "PPS changed between slices\n");
  1909. return AVERROR_INVALIDDATA;
  1910. }
  1911. if (h->ps.sps != pps->sps) {
  1912. av_log(h->avctx, AV_LOG_ERROR,
  1913. "SPS changed in the middle of the frame\n");
  1914. return AVERROR_INVALIDDATA;
  1915. }
  1916. }
  1917. if (h->current_slice == 0) {
  1918. ret = h264_field_start(h, sl, nal, first_slice);
  1919. if (ret < 0)
  1920. return ret;
  1921. } else {
  1922. if (h->picture_structure != sl->picture_structure ||
  1923. h->droppable != (nal->ref_idc == 0)) {
  1924. av_log(h->avctx, AV_LOG_ERROR,
  1925. "Changing field mode (%d -> %d) between slices is not allowed\n",
  1926. h->picture_structure, sl->picture_structure);
  1927. return AVERROR_INVALIDDATA;
  1928. } else if (!h->cur_pic_ptr) {
  1929. av_log(h->avctx, AV_LOG_ERROR,
  1930. "unset cur_pic_ptr on slice %d\n",
  1931. h->current_slice + 1);
  1932. return AVERROR_INVALIDDATA;
  1933. }
  1934. }
  1935. ret = h264_slice_init(h, sl, nal);
  1936. if (ret < 0)
  1937. return ret;
  1938. h->nb_slice_ctx_queued++;
  1939. return 0;
  1940. }
  1941. int ff_h264_get_slice_type(const H264SliceContext *sl)
  1942. {
  1943. switch (sl->slice_type) {
  1944. case AV_PICTURE_TYPE_P:
  1945. return 0;
  1946. case AV_PICTURE_TYPE_B:
  1947. return 1;
  1948. case AV_PICTURE_TYPE_I:
  1949. return 2;
  1950. case AV_PICTURE_TYPE_SP:
  1951. return 3;
  1952. case AV_PICTURE_TYPE_SI:
  1953. return 4;
  1954. default:
  1955. return AVERROR_INVALIDDATA;
  1956. }
  1957. }
  1958. static av_always_inline void fill_filter_caches_inter(const H264Context *h,
  1959. H264SliceContext *sl,
  1960. int mb_type, int top_xy,
  1961. int left_xy[LEFT_MBS],
  1962. int top_type,
  1963. int left_type[LEFT_MBS],
  1964. int mb_xy, int list)
  1965. {
  1966. int b_stride = h->b_stride;
  1967. int16_t(*mv_dst)[2] = &sl->mv_cache[list][scan8[0]];
  1968. int8_t *ref_cache = &sl->ref_cache[list][scan8[0]];
  1969. if (IS_INTER(mb_type) || IS_DIRECT(mb_type)) {
  1970. if (USES_LIST(top_type, list)) {
  1971. const int b_xy = h->mb2b_xy[top_xy] + 3 * b_stride;
  1972. const int b8_xy = 4 * top_xy + 2;
  1973. const int *ref2frm = &h->ref2frm[h->slice_table[top_xy] & (MAX_SLICES - 1)][list][(MB_MBAFF(sl) ? 20 : 2)];
  1974. AV_COPY128(mv_dst - 1 * 8, h->cur_pic.motion_val[list][b_xy + 0]);
  1975. ref_cache[0 - 1 * 8] =
  1976. ref_cache[1 - 1 * 8] = ref2frm[h->cur_pic.ref_index[list][b8_xy + 0]];
  1977. ref_cache[2 - 1 * 8] =
  1978. ref_cache[3 - 1 * 8] = ref2frm[h->cur_pic.ref_index[list][b8_xy + 1]];
  1979. } else {
  1980. AV_ZERO128(mv_dst - 1 * 8);
  1981. AV_WN32A(&ref_cache[0 - 1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
  1982. }
  1983. if (!IS_INTERLACED(mb_type ^ left_type[LTOP])) {
  1984. if (USES_LIST(left_type[LTOP], list)) {
  1985. const int b_xy = h->mb2b_xy[left_xy[LTOP]] + 3;
  1986. const int b8_xy = 4 * left_xy[LTOP] + 1;
  1987. const int *ref2frm = &h->ref2frm[h->slice_table[left_xy[LTOP]] & (MAX_SLICES - 1)][list][(MB_MBAFF(sl) ? 20 : 2)];
  1988. AV_COPY32(mv_dst - 1 + 0, h->cur_pic.motion_val[list][b_xy + b_stride * 0]);
  1989. AV_COPY32(mv_dst - 1 + 8, h->cur_pic.motion_val[list][b_xy + b_stride * 1]);
  1990. AV_COPY32(mv_dst - 1 + 16, h->cur_pic.motion_val[list][b_xy + b_stride * 2]);
  1991. AV_COPY32(mv_dst - 1 + 24, h->cur_pic.motion_val[list][b_xy + b_stride * 3]);
  1992. ref_cache[-1 + 0] =
  1993. ref_cache[-1 + 8] = ref2frm[h->cur_pic.ref_index[list][b8_xy + 2 * 0]];
  1994. ref_cache[-1 + 16] =
  1995. ref_cache[-1 + 24] = ref2frm[h->cur_pic.ref_index[list][b8_xy + 2 * 1]];
  1996. } else {
  1997. AV_ZERO32(mv_dst - 1 + 0);
  1998. AV_ZERO32(mv_dst - 1 + 8);
  1999. AV_ZERO32(mv_dst - 1 + 16);
  2000. AV_ZERO32(mv_dst - 1 + 24);
  2001. ref_cache[-1 + 0] =
  2002. ref_cache[-1 + 8] =
  2003. ref_cache[-1 + 16] =
  2004. ref_cache[-1 + 24] = LIST_NOT_USED;
  2005. }
  2006. }
  2007. }
  2008. if (!USES_LIST(mb_type, list)) {
  2009. fill_rectangle(mv_dst, 4, 4, 8, pack16to32(0, 0), 4);
  2010. AV_WN32A(&ref_cache[0 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
  2011. AV_WN32A(&ref_cache[1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
  2012. AV_WN32A(&ref_cache[2 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
  2013. AV_WN32A(&ref_cache[3 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
  2014. return;
  2015. }
  2016. {
  2017. int8_t *ref = &h->cur_pic.ref_index[list][4 * mb_xy];
  2018. const int *ref2frm = &h->ref2frm[sl->slice_num & (MAX_SLICES - 1)][list][(MB_MBAFF(sl) ? 20 : 2)];
  2019. uint32_t ref01 = (pack16to32(ref2frm[ref[0]], ref2frm[ref[1]]) & 0x00FF00FF) * 0x0101;
  2020. uint32_t ref23 = (pack16to32(ref2frm[ref[2]], ref2frm[ref[3]]) & 0x00FF00FF) * 0x0101;
  2021. AV_WN32A(&ref_cache[0 * 8], ref01);
  2022. AV_WN32A(&ref_cache[1 * 8], ref01);
  2023. AV_WN32A(&ref_cache[2 * 8], ref23);
  2024. AV_WN32A(&ref_cache[3 * 8], ref23);
  2025. }
  2026. {
  2027. int16_t(*mv_src)[2] = &h->cur_pic.motion_val[list][4 * sl->mb_x + 4 * sl->mb_y * b_stride];
  2028. AV_COPY128(mv_dst + 8 * 0, mv_src + 0 * b_stride);
  2029. AV_COPY128(mv_dst + 8 * 1, mv_src + 1 * b_stride);
  2030. AV_COPY128(mv_dst + 8 * 2, mv_src + 2 * b_stride);
  2031. AV_COPY128(mv_dst + 8 * 3, mv_src + 3 * b_stride);
  2032. }
  2033. }
  2034. /**
  2035. * @return non zero if the loop filter can be skipped
  2036. */
  2037. static int fill_filter_caches(const H264Context *h, H264SliceContext *sl, int mb_type)
  2038. {
  2039. const int mb_xy = sl->mb_xy;
  2040. int top_xy, left_xy[LEFT_MBS];
  2041. int top_type, left_type[LEFT_MBS];
  2042. uint8_t *nnz;
  2043. uint8_t *nnz_cache;
  2044. top_xy = mb_xy - (h->mb_stride << MB_FIELD(sl));
  2045. left_xy[LBOT] = left_xy[LTOP] = mb_xy - 1;
  2046. if (FRAME_MBAFF(h)) {
  2047. const int left_mb_field_flag = IS_INTERLACED(h->cur_pic.mb_type[mb_xy - 1]);
  2048. const int curr_mb_field_flag = IS_INTERLACED(mb_type);
  2049. if (sl->mb_y & 1) {
  2050. if (left_mb_field_flag != curr_mb_field_flag)
  2051. left_xy[LTOP] -= h->mb_stride;
  2052. } else {
  2053. if (curr_mb_field_flag)
  2054. top_xy += h->mb_stride &
  2055. (((h->cur_pic.mb_type[top_xy] >> 7) & 1) - 1);
  2056. if (left_mb_field_flag != curr_mb_field_flag)
  2057. left_xy[LBOT] += h->mb_stride;
  2058. }
  2059. }
  2060. sl->top_mb_xy = top_xy;
  2061. sl->left_mb_xy[LTOP] = left_xy[LTOP];
  2062. sl->left_mb_xy[LBOT] = left_xy[LBOT];
  2063. {
  2064. /* For sufficiently low qp, filtering wouldn't do anything.
  2065. * This is a conservative estimate: could also check beta_offset
  2066. * and more accurate chroma_qp. */
  2067. int qp_thresh = sl->qp_thresh; // FIXME strictly we should store qp_thresh for each mb of a slice
  2068. int qp = h->cur_pic.qscale_table[mb_xy];
  2069. if (qp <= qp_thresh &&
  2070. (left_xy[LTOP] < 0 ||
  2071. ((qp + h->cur_pic.qscale_table[left_xy[LTOP]] + 1) >> 1) <= qp_thresh) &&
  2072. (top_xy < 0 ||
  2073. ((qp + h->cur_pic.qscale_table[top_xy] + 1) >> 1) <= qp_thresh)) {
  2074. if (!FRAME_MBAFF(h))
  2075. return 1;
  2076. if ((left_xy[LTOP] < 0 ||
  2077. ((qp + h->cur_pic.qscale_table[left_xy[LBOT]] + 1) >> 1) <= qp_thresh) &&
  2078. (top_xy < h->mb_stride ||
  2079. ((qp + h->cur_pic.qscale_table[top_xy - h->mb_stride] + 1) >> 1) <= qp_thresh))
  2080. return 1;
  2081. }
  2082. }
  2083. top_type = h->cur_pic.mb_type[top_xy];
  2084. left_type[LTOP] = h->cur_pic.mb_type[left_xy[LTOP]];
  2085. left_type[LBOT] = h->cur_pic.mb_type[left_xy[LBOT]];
  2086. if (sl->deblocking_filter == 2) {
  2087. if (h->slice_table[top_xy] != sl->slice_num)
  2088. top_type = 0;
  2089. if (h->slice_table[left_xy[LBOT]] != sl->slice_num)
  2090. left_type[LTOP] = left_type[LBOT] = 0;
  2091. } else {
  2092. if (h->slice_table[top_xy] == 0xFFFF)
  2093. top_type = 0;
  2094. if (h->slice_table[left_xy[LBOT]] == 0xFFFF)
  2095. left_type[LTOP] = left_type[LBOT] = 0;
  2096. }
  2097. sl->top_type = top_type;
  2098. sl->left_type[LTOP] = left_type[LTOP];
  2099. sl->left_type[LBOT] = left_type[LBOT];
  2100. if (IS_INTRA(mb_type))
  2101. return 0;
  2102. fill_filter_caches_inter(h, sl, mb_type, top_xy, left_xy,
  2103. top_type, left_type, mb_xy, 0);
  2104. if (sl->list_count == 2)
  2105. fill_filter_caches_inter(h, sl, mb_type, top_xy, left_xy,
  2106. top_type, left_type, mb_xy, 1);
  2107. nnz = h->non_zero_count[mb_xy];
  2108. nnz_cache = sl->non_zero_count_cache;
  2109. AV_COPY32(&nnz_cache[4 + 8 * 1], &nnz[0]);
  2110. AV_COPY32(&nnz_cache[4 + 8 * 2], &nnz[4]);
  2111. AV_COPY32(&nnz_cache[4 + 8 * 3], &nnz[8]);
  2112. AV_COPY32(&nnz_cache[4 + 8 * 4], &nnz[12]);
  2113. sl->cbp = h->cbp_table[mb_xy];
  2114. if (top_type) {
  2115. nnz = h->non_zero_count[top_xy];
  2116. AV_COPY32(&nnz_cache[4 + 8 * 0], &nnz[3 * 4]);
  2117. }
  2118. if (left_type[LTOP]) {
  2119. nnz = h->non_zero_count[left_xy[LTOP]];
  2120. nnz_cache[3 + 8 * 1] = nnz[3 + 0 * 4];
  2121. nnz_cache[3 + 8 * 2] = nnz[3 + 1 * 4];
  2122. nnz_cache[3 + 8 * 3] = nnz[3 + 2 * 4];
  2123. nnz_cache[3 + 8 * 4] = nnz[3 + 3 * 4];
  2124. }
  2125. /* CAVLC 8x8dct requires NNZ values for residual decoding that differ
  2126. * from what the loop filter needs */
  2127. if (!CABAC(h) && h->ps.pps->transform_8x8_mode) {
  2128. if (IS_8x8DCT(top_type)) {
  2129. nnz_cache[4 + 8 * 0] =
  2130. nnz_cache[5 + 8 * 0] = (h->cbp_table[top_xy] & 0x4000) >> 12;
  2131. nnz_cache[6 + 8 * 0] =
  2132. nnz_cache[7 + 8 * 0] = (h->cbp_table[top_xy] & 0x8000) >> 12;
  2133. }
  2134. if (IS_8x8DCT(left_type[LTOP])) {
  2135. nnz_cache[3 + 8 * 1] =
  2136. nnz_cache[3 + 8 * 2] = (h->cbp_table[left_xy[LTOP]] & 0x2000) >> 12; // FIXME check MBAFF
  2137. }
  2138. if (IS_8x8DCT(left_type[LBOT])) {
  2139. nnz_cache[3 + 8 * 3] =
  2140. nnz_cache[3 + 8 * 4] = (h->cbp_table[left_xy[LBOT]] & 0x8000) >> 12; // FIXME check MBAFF
  2141. }
  2142. if (IS_8x8DCT(mb_type)) {
  2143. nnz_cache[scan8[0]] =
  2144. nnz_cache[scan8[1]] =
  2145. nnz_cache[scan8[2]] =
  2146. nnz_cache[scan8[3]] = (sl->cbp & 0x1000) >> 12;
  2147. nnz_cache[scan8[0 + 4]] =
  2148. nnz_cache[scan8[1 + 4]] =
  2149. nnz_cache[scan8[2 + 4]] =
  2150. nnz_cache[scan8[3 + 4]] = (sl->cbp & 0x2000) >> 12;
  2151. nnz_cache[scan8[0 + 8]] =
  2152. nnz_cache[scan8[1 + 8]] =
  2153. nnz_cache[scan8[2 + 8]] =
  2154. nnz_cache[scan8[3 + 8]] = (sl->cbp & 0x4000) >> 12;
  2155. nnz_cache[scan8[0 + 12]] =
  2156. nnz_cache[scan8[1 + 12]] =
  2157. nnz_cache[scan8[2 + 12]] =
  2158. nnz_cache[scan8[3 + 12]] = (sl->cbp & 0x8000) >> 12;
  2159. }
  2160. }
  2161. return 0;
  2162. }
  2163. static void loop_filter(const H264Context *h, H264SliceContext *sl, int start_x, int end_x)
  2164. {
  2165. uint8_t *dest_y, *dest_cb, *dest_cr;
  2166. int linesize, uvlinesize, mb_x, mb_y;
  2167. const int end_mb_y = sl->mb_y + FRAME_MBAFF(h);
  2168. const int old_slice_type = sl->slice_type;
  2169. const int pixel_shift = h->pixel_shift;
  2170. const int block_h = 16 >> h->chroma_y_shift;
  2171. if (h->postpone_filter)
  2172. return;
  2173. if (sl->deblocking_filter) {
  2174. for (mb_x = start_x; mb_x < end_x; mb_x++)
  2175. for (mb_y = end_mb_y - FRAME_MBAFF(h); mb_y <= end_mb_y; mb_y++) {
  2176. int mb_xy, mb_type;
  2177. mb_xy = sl->mb_xy = mb_x + mb_y * h->mb_stride;
  2178. mb_type = h->cur_pic.mb_type[mb_xy];
  2179. if (FRAME_MBAFF(h))
  2180. sl->mb_mbaff =
  2181. sl->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
  2182. sl->mb_x = mb_x;
  2183. sl->mb_y = mb_y;
  2184. dest_y = h->cur_pic.f->data[0] +
  2185. ((mb_x << pixel_shift) + mb_y * sl->linesize) * 16;
  2186. dest_cb = h->cur_pic.f->data[1] +
  2187. (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
  2188. mb_y * sl->uvlinesize * block_h;
  2189. dest_cr = h->cur_pic.f->data[2] +
  2190. (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
  2191. mb_y * sl->uvlinesize * block_h;
  2192. // FIXME simplify above
  2193. if (MB_FIELD(sl)) {
  2194. linesize = sl->mb_linesize = sl->linesize * 2;
  2195. uvlinesize = sl->mb_uvlinesize = sl->uvlinesize * 2;
  2196. if (mb_y & 1) { // FIXME move out of this function?
  2197. dest_y -= sl->linesize * 15;
  2198. dest_cb -= sl->uvlinesize * (block_h - 1);
  2199. dest_cr -= sl->uvlinesize * (block_h - 1);
  2200. }
  2201. } else {
  2202. linesize = sl->mb_linesize = sl->linesize;
  2203. uvlinesize = sl->mb_uvlinesize = sl->uvlinesize;
  2204. }
  2205. backup_mb_border(h, sl, dest_y, dest_cb, dest_cr, linesize,
  2206. uvlinesize, 0);
  2207. if (fill_filter_caches(h, sl, mb_type))
  2208. continue;
  2209. sl->chroma_qp[0] = get_chroma_qp(h->ps.pps, 0, h->cur_pic.qscale_table[mb_xy]);
  2210. sl->chroma_qp[1] = get_chroma_qp(h->ps.pps, 1, h->cur_pic.qscale_table[mb_xy]);
  2211. if (FRAME_MBAFF(h)) {
  2212. ff_h264_filter_mb(h, sl, mb_x, mb_y, dest_y, dest_cb, dest_cr,
  2213. linesize, uvlinesize);
  2214. } else {
  2215. ff_h264_filter_mb_fast(h, sl, mb_x, mb_y, dest_y, dest_cb,
  2216. dest_cr, linesize, uvlinesize);
  2217. }
  2218. }
  2219. }
  2220. sl->slice_type = old_slice_type;
  2221. sl->mb_x = end_x;
  2222. sl->mb_y = end_mb_y - FRAME_MBAFF(h);
  2223. sl->chroma_qp[0] = get_chroma_qp(h->ps.pps, 0, sl->qscale);
  2224. sl->chroma_qp[1] = get_chroma_qp(h->ps.pps, 1, sl->qscale);
  2225. }
  2226. static void predict_field_decoding_flag(const H264Context *h, H264SliceContext *sl)
  2227. {
  2228. const int mb_xy = sl->mb_x + sl->mb_y * h->mb_stride;
  2229. int mb_type = (h->slice_table[mb_xy - 1] == sl->slice_num) ?
  2230. h->cur_pic.mb_type[mb_xy - 1] :
  2231. (h->slice_table[mb_xy - h->mb_stride] == sl->slice_num) ?
  2232. h->cur_pic.mb_type[mb_xy - h->mb_stride] : 0;
  2233. sl->mb_mbaff = sl->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
  2234. }
  2235. /**
  2236. * Draw edges and report progress for the last MB row.
  2237. */
  2238. static void decode_finish_row(const H264Context *h, H264SliceContext *sl)
  2239. {
  2240. int top = 16 * (sl->mb_y >> FIELD_PICTURE(h));
  2241. int pic_height = 16 * h->mb_height >> FIELD_PICTURE(h);
  2242. int height = 16 << FRAME_MBAFF(h);
  2243. int deblock_border = (16 + 4) << FRAME_MBAFF(h);
  2244. if (sl->deblocking_filter) {
  2245. if ((top + height) >= pic_height)
  2246. height += deblock_border;
  2247. top -= deblock_border;
  2248. }
  2249. if (top >= pic_height || (top + height) < 0)
  2250. return;
  2251. height = FFMIN(height, pic_height - top);
  2252. if (top < 0) {
  2253. height = top + height;
  2254. top = 0;
  2255. }
  2256. ff_h264_draw_horiz_band(h, sl, top, height);
  2257. if (h->droppable || sl->h264->slice_ctx[0].er.error_occurred)
  2258. return;
  2259. ff_thread_report_progress(&h->cur_pic_ptr->tf, top + height - 1,
  2260. h->picture_structure == PICT_BOTTOM_FIELD);
  2261. }
  2262. static void er_add_slice(H264SliceContext *sl,
  2263. int startx, int starty,
  2264. int endx, int endy, int status)
  2265. {
  2266. if (!sl->h264->enable_er)
  2267. return;
  2268. if (CONFIG_ERROR_RESILIENCE) {
  2269. ERContext *er = &sl->h264->slice_ctx[0].er;
  2270. ff_er_add_slice(er, startx, starty, endx, endy, status);
  2271. }
  2272. }
  2273. static int decode_slice(struct AVCodecContext *avctx, void *arg)
  2274. {
  2275. H264SliceContext *sl = arg;
  2276. const H264Context *h = sl->h264;
  2277. int lf_x_start = sl->mb_x;
  2278. int orig_deblock = sl->deblocking_filter;
  2279. int ret;
  2280. sl->linesize = h->cur_pic_ptr->f->linesize[0];
  2281. sl->uvlinesize = h->cur_pic_ptr->f->linesize[1];
  2282. ret = alloc_scratch_buffers(sl, sl->linesize);
  2283. if (ret < 0)
  2284. return ret;
  2285. sl->mb_skip_run = -1;
  2286. av_assert0(h->block_offset[15] == (4 * ((scan8[15] - scan8[0]) & 7) << h->pixel_shift) + 4 * sl->linesize * ((scan8[15] - scan8[0]) >> 3));
  2287. if (h->postpone_filter)
  2288. sl->deblocking_filter = 0;
  2289. sl->is_complex = FRAME_MBAFF(h) || h->picture_structure != PICT_FRAME ||
  2290. (CONFIG_GRAY && (h->flags & AV_CODEC_FLAG_GRAY));
  2291. if (!(h->avctx->active_thread_type & FF_THREAD_SLICE) && h->picture_structure == PICT_FRAME && h->slice_ctx[0].er.error_status_table) {
  2292. const int start_i = av_clip(sl->resync_mb_x + sl->resync_mb_y * h->mb_width, 0, h->mb_num - 1);
  2293. if (start_i) {
  2294. int prev_status = h->slice_ctx[0].er.error_status_table[h->slice_ctx[0].er.mb_index2xy[start_i - 1]];
  2295. prev_status &= ~ VP_START;
  2296. if (prev_status != (ER_MV_END | ER_DC_END | ER_AC_END))
  2297. h->slice_ctx[0].er.error_occurred = 1;
  2298. }
  2299. }
  2300. if (h->ps.pps->cabac) {
  2301. /* realign */
  2302. align_get_bits(&sl->gb);
  2303. /* init cabac */
  2304. ret = ff_init_cabac_decoder(&sl->cabac,
  2305. sl->gb.buffer + get_bits_count(&sl->gb) / 8,
  2306. (get_bits_left(&sl->gb) + 7) / 8);
  2307. if (ret < 0)
  2308. return ret;
  2309. ff_h264_init_cabac_states(h, sl);
  2310. for (;;) {
  2311. int ret, eos;
  2312. if (sl->mb_x + sl->mb_y * h->mb_width >= sl->next_slice_idx) {
  2313. av_log(h->avctx, AV_LOG_ERROR, "Slice overlaps with next at %d\n",
  2314. sl->next_slice_idx);
  2315. er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
  2316. sl->mb_y, ER_MB_ERROR);
  2317. return AVERROR_INVALIDDATA;
  2318. }
  2319. ret = ff_h264_decode_mb_cabac(h, sl);
  2320. if (ret >= 0)
  2321. ff_h264_hl_decode_mb(h, sl);
  2322. // FIXME optimal? or let mb_decode decode 16x32 ?
  2323. if (ret >= 0 && FRAME_MBAFF(h)) {
  2324. sl->mb_y++;
  2325. ret = ff_h264_decode_mb_cabac(h, sl);
  2326. if (ret >= 0)
  2327. ff_h264_hl_decode_mb(h, sl);
  2328. sl->mb_y--;
  2329. }
  2330. eos = get_cabac_terminate(&sl->cabac);
  2331. if ((h->workaround_bugs & FF_BUG_TRUNCATED) &&
  2332. sl->cabac.bytestream > sl->cabac.bytestream_end + 2) {
  2333. er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x - 1,
  2334. sl->mb_y, ER_MB_END);
  2335. if (sl->mb_x >= lf_x_start)
  2336. loop_filter(h, sl, lf_x_start, sl->mb_x + 1);
  2337. goto finish;
  2338. }
  2339. if (sl->cabac.bytestream > sl->cabac.bytestream_end + 2 )
  2340. av_log(h->avctx, AV_LOG_DEBUG, "bytestream overread %"PTRDIFF_SPECIFIER"\n", sl->cabac.bytestream_end - sl->cabac.bytestream);
  2341. if (ret < 0 || sl->cabac.bytestream > sl->cabac.bytestream_end + 4) {
  2342. av_log(h->avctx, AV_LOG_ERROR,
  2343. "error while decoding MB %d %d, bytestream %"PTRDIFF_SPECIFIER"\n",
  2344. sl->mb_x, sl->mb_y,
  2345. sl->cabac.bytestream_end - sl->cabac.bytestream);
  2346. er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
  2347. sl->mb_y, ER_MB_ERROR);
  2348. return AVERROR_INVALIDDATA;
  2349. }
  2350. if (++sl->mb_x >= h->mb_width) {
  2351. loop_filter(h, sl, lf_x_start, sl->mb_x);
  2352. sl->mb_x = lf_x_start = 0;
  2353. decode_finish_row(h, sl);
  2354. ++sl->mb_y;
  2355. if (FIELD_OR_MBAFF_PICTURE(h)) {
  2356. ++sl->mb_y;
  2357. if (FRAME_MBAFF(h) && sl->mb_y < h->mb_height)
  2358. predict_field_decoding_flag(h, sl);
  2359. }
  2360. }
  2361. if (eos || sl->mb_y >= h->mb_height) {
  2362. ff_tlog(h->avctx, "slice end %d %d\n",
  2363. get_bits_count(&sl->gb), sl->gb.size_in_bits);
  2364. er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x - 1,
  2365. sl->mb_y, ER_MB_END);
  2366. if (sl->mb_x > lf_x_start)
  2367. loop_filter(h, sl, lf_x_start, sl->mb_x);
  2368. goto finish;
  2369. }
  2370. }
  2371. } else {
  2372. for (;;) {
  2373. int ret;
  2374. if (sl->mb_x + sl->mb_y * h->mb_width >= sl->next_slice_idx) {
  2375. av_log(h->avctx, AV_LOG_ERROR, "Slice overlaps with next at %d\n",
  2376. sl->next_slice_idx);
  2377. er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
  2378. sl->mb_y, ER_MB_ERROR);
  2379. return AVERROR_INVALIDDATA;
  2380. }
  2381. ret = ff_h264_decode_mb_cavlc(h, sl);
  2382. if (ret >= 0)
  2383. ff_h264_hl_decode_mb(h, sl);
  2384. // FIXME optimal? or let mb_decode decode 16x32 ?
  2385. if (ret >= 0 && FRAME_MBAFF(h)) {
  2386. sl->mb_y++;
  2387. ret = ff_h264_decode_mb_cavlc(h, sl);
  2388. if (ret >= 0)
  2389. ff_h264_hl_decode_mb(h, sl);
  2390. sl->mb_y--;
  2391. }
  2392. if (ret < 0) {
  2393. av_log(h->avctx, AV_LOG_ERROR,
  2394. "error while decoding MB %d %d\n", sl->mb_x, sl->mb_y);
  2395. er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
  2396. sl->mb_y, ER_MB_ERROR);
  2397. return ret;
  2398. }
  2399. if (++sl->mb_x >= h->mb_width) {
  2400. loop_filter(h, sl, lf_x_start, sl->mb_x);
  2401. sl->mb_x = lf_x_start = 0;
  2402. decode_finish_row(h, sl);
  2403. ++sl->mb_y;
  2404. if (FIELD_OR_MBAFF_PICTURE(h)) {
  2405. ++sl->mb_y;
  2406. if (FRAME_MBAFF(h) && sl->mb_y < h->mb_height)
  2407. predict_field_decoding_flag(h, sl);
  2408. }
  2409. if (sl->mb_y >= h->mb_height) {
  2410. ff_tlog(h->avctx, "slice end %d %d\n",
  2411. get_bits_count(&sl->gb), sl->gb.size_in_bits);
  2412. if ( get_bits_left(&sl->gb) == 0
  2413. || get_bits_left(&sl->gb) > 0 && !(h->avctx->err_recognition & AV_EF_AGGRESSIVE)) {
  2414. er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
  2415. sl->mb_x - 1, sl->mb_y, ER_MB_END);
  2416. goto finish;
  2417. } else {
  2418. er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
  2419. sl->mb_x, sl->mb_y, ER_MB_END);
  2420. return AVERROR_INVALIDDATA;
  2421. }
  2422. }
  2423. }
  2424. if (get_bits_left(&sl->gb) <= 0 && sl->mb_skip_run <= 0) {
  2425. ff_tlog(h->avctx, "slice end %d %d\n",
  2426. get_bits_count(&sl->gb), sl->gb.size_in_bits);
  2427. if (get_bits_left(&sl->gb) == 0) {
  2428. er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
  2429. sl->mb_x - 1, sl->mb_y, ER_MB_END);
  2430. if (sl->mb_x > lf_x_start)
  2431. loop_filter(h, sl, lf_x_start, sl->mb_x);
  2432. goto finish;
  2433. } else {
  2434. er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
  2435. sl->mb_y, ER_MB_ERROR);
  2436. return AVERROR_INVALIDDATA;
  2437. }
  2438. }
  2439. }
  2440. }
  2441. finish:
  2442. sl->deblocking_filter = orig_deblock;
  2443. return 0;
  2444. }
  2445. /**
  2446. * Call decode_slice() for each context.
  2447. *
  2448. * @param h h264 master context
  2449. */
  2450. int ff_h264_execute_decode_slices(H264Context *h)
  2451. {
  2452. AVCodecContext *const avctx = h->avctx;
  2453. H264SliceContext *sl;
  2454. int context_count = h->nb_slice_ctx_queued;
  2455. int ret = 0;
  2456. int i, j;
  2457. h->slice_ctx[0].next_slice_idx = INT_MAX;
  2458. if (h->avctx->hwaccel || context_count < 1)
  2459. return 0;
  2460. av_assert0(context_count && h->slice_ctx[context_count - 1].mb_y < h->mb_height);
  2461. if (context_count == 1) {
  2462. h->slice_ctx[0].next_slice_idx = h->mb_width * h->mb_height;
  2463. h->postpone_filter = 0;
  2464. ret = decode_slice(avctx, &h->slice_ctx[0]);
  2465. h->mb_y = h->slice_ctx[0].mb_y;
  2466. if (ret < 0)
  2467. goto finish;
  2468. } else {
  2469. av_assert0(context_count > 0);
  2470. for (i = 0; i < context_count; i++) {
  2471. int next_slice_idx = h->mb_width * h->mb_height;
  2472. int slice_idx;
  2473. sl = &h->slice_ctx[i];
  2474. if (CONFIG_ERROR_RESILIENCE) {
  2475. sl->er.error_count = 0;
  2476. }
  2477. /* make sure none of those slices overlap */
  2478. slice_idx = sl->mb_y * h->mb_width + sl->mb_x;
  2479. for (j = 0; j < context_count; j++) {
  2480. H264SliceContext *sl2 = &h->slice_ctx[j];
  2481. int slice_idx2 = sl2->mb_y * h->mb_width + sl2->mb_x;
  2482. if (i == j || slice_idx2 < slice_idx)
  2483. continue;
  2484. next_slice_idx = FFMIN(next_slice_idx, slice_idx2);
  2485. }
  2486. sl->next_slice_idx = next_slice_idx;
  2487. }
  2488. avctx->execute(avctx, decode_slice, h->slice_ctx,
  2489. NULL, context_count, sizeof(h->slice_ctx[0]));
  2490. /* pull back stuff from slices to master context */
  2491. sl = &h->slice_ctx[context_count - 1];
  2492. h->mb_y = sl->mb_y;
  2493. if (CONFIG_ERROR_RESILIENCE) {
  2494. for (i = 1; i < context_count; i++)
  2495. h->slice_ctx[0].er.error_count += h->slice_ctx[i].er.error_count;
  2496. }
  2497. if (h->postpone_filter) {
  2498. h->postpone_filter = 0;
  2499. for (i = 0; i < context_count; i++) {
  2500. int y_end, x_end;
  2501. sl = &h->slice_ctx[i];
  2502. y_end = FFMIN(sl->mb_y + 1, h->mb_height);
  2503. x_end = (sl->mb_y >= h->mb_height) ? h->mb_width : sl->mb_x;
  2504. for (j = sl->resync_mb_y; j < y_end; j += 1 + FIELD_OR_MBAFF_PICTURE(h)) {
  2505. sl->mb_y = j;
  2506. loop_filter(h, sl, j > sl->resync_mb_y ? 0 : sl->resync_mb_x,
  2507. j == y_end - 1 ? x_end : h->mb_width);
  2508. }
  2509. }
  2510. }
  2511. }
  2512. finish:
  2513. h->nb_slice_ctx_queued = 0;
  2514. return ret;
  2515. }