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.

2815 lines
106KB

  1. /*
  2. * H.26L/H.264/AVC/JVT/14496-10/... decoder
  3. * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
  4. *
  5. * This file is part of FFmpeg.
  6. *
  7. * FFmpeg is free software; you can redistribute it and/or
  8. * modify it under the terms of the GNU Lesser General Public
  9. * License as published by the Free Software Foundation; either
  10. * version 2.1 of the License, or (at your option) any later version.
  11. *
  12. * FFmpeg is distributed in the hope that it will be useful,
  13. * but WITHOUT ANY WARRANTY; without even the implied warranty of
  14. * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  15. * Lesser General Public License for more details.
  16. *
  17. * You should have received a copy of the GNU Lesser General Public
  18. * License along with FFmpeg; if not, write to the Free Software
  19. * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  20. */
  21. /**
  22. * @file
  23. * H.264 / AVC / MPEG-4 part10 codec.
  24. * @author Michael Niedermayer <michaelni@gmx.at>
  25. */
  26. #include "libavutil/avassert.h"
  27. #include "libavutil/display.h"
  28. #include "libavutil/imgutils.h"
  29. #include "libavutil/stereo3d.h"
  30. #include "libavutil/timer.h"
  31. #include "internal.h"
  32. #include "cabac.h"
  33. #include "cabac_functions.h"
  34. #include "error_resilience.h"
  35. #include "avcodec.h"
  36. #include "h264.h"
  37. #include "h264dec.h"
  38. #include "h264data.h"
  39. #include "h264chroma.h"
  40. #include "h264_mvpred.h"
  41. #include "h264_ps.h"
  42. #include "golomb.h"
  43. #include "mathops.h"
  44. #include "mpegutils.h"
  45. #include "mpegvideo.h"
  46. #include "rectangle.h"
  47. #include "thread.h"
  48. static const uint8_t field_scan[16+1] = {
  49. 0 + 0 * 4, 0 + 1 * 4, 1 + 0 * 4, 0 + 2 * 4,
  50. 0 + 3 * 4, 1 + 1 * 4, 1 + 2 * 4, 1 + 3 * 4,
  51. 2 + 0 * 4, 2 + 1 * 4, 2 + 2 * 4, 2 + 3 * 4,
  52. 3 + 0 * 4, 3 + 1 * 4, 3 + 2 * 4, 3 + 3 * 4,
  53. };
  54. static const uint8_t field_scan8x8[64+1] = {
  55. 0 + 0 * 8, 0 + 1 * 8, 0 + 2 * 8, 1 + 0 * 8,
  56. 1 + 1 * 8, 0 + 3 * 8, 0 + 4 * 8, 1 + 2 * 8,
  57. 2 + 0 * 8, 1 + 3 * 8, 0 + 5 * 8, 0 + 6 * 8,
  58. 0 + 7 * 8, 1 + 4 * 8, 2 + 1 * 8, 3 + 0 * 8,
  59. 2 + 2 * 8, 1 + 5 * 8, 1 + 6 * 8, 1 + 7 * 8,
  60. 2 + 3 * 8, 3 + 1 * 8, 4 + 0 * 8, 3 + 2 * 8,
  61. 2 + 4 * 8, 2 + 5 * 8, 2 + 6 * 8, 2 + 7 * 8,
  62. 3 + 3 * 8, 4 + 1 * 8, 5 + 0 * 8, 4 + 2 * 8,
  63. 3 + 4 * 8, 3 + 5 * 8, 3 + 6 * 8, 3 + 7 * 8,
  64. 4 + 3 * 8, 5 + 1 * 8, 6 + 0 * 8, 5 + 2 * 8,
  65. 4 + 4 * 8, 4 + 5 * 8, 4 + 6 * 8, 4 + 7 * 8,
  66. 5 + 3 * 8, 6 + 1 * 8, 6 + 2 * 8, 5 + 4 * 8,
  67. 5 + 5 * 8, 5 + 6 * 8, 5 + 7 * 8, 6 + 3 * 8,
  68. 7 + 0 * 8, 7 + 1 * 8, 6 + 4 * 8, 6 + 5 * 8,
  69. 6 + 6 * 8, 6 + 7 * 8, 7 + 2 * 8, 7 + 3 * 8,
  70. 7 + 4 * 8, 7 + 5 * 8, 7 + 6 * 8, 7 + 7 * 8,
  71. };
  72. static const uint8_t field_scan8x8_cavlc[64+1] = {
  73. 0 + 0 * 8, 1 + 1 * 8, 2 + 0 * 8, 0 + 7 * 8,
  74. 2 + 2 * 8, 2 + 3 * 8, 2 + 4 * 8, 3 + 3 * 8,
  75. 3 + 4 * 8, 4 + 3 * 8, 4 + 4 * 8, 5 + 3 * 8,
  76. 5 + 5 * 8, 7 + 0 * 8, 6 + 6 * 8, 7 + 4 * 8,
  77. 0 + 1 * 8, 0 + 3 * 8, 1 + 3 * 8, 1 + 4 * 8,
  78. 1 + 5 * 8, 3 + 1 * 8, 2 + 5 * 8, 4 + 1 * 8,
  79. 3 + 5 * 8, 5 + 1 * 8, 4 + 5 * 8, 6 + 1 * 8,
  80. 5 + 6 * 8, 7 + 1 * 8, 6 + 7 * 8, 7 + 5 * 8,
  81. 0 + 2 * 8, 0 + 4 * 8, 0 + 5 * 8, 2 + 1 * 8,
  82. 1 + 6 * 8, 4 + 0 * 8, 2 + 6 * 8, 5 + 0 * 8,
  83. 3 + 6 * 8, 6 + 0 * 8, 4 + 6 * 8, 6 + 2 * 8,
  84. 5 + 7 * 8, 6 + 4 * 8, 7 + 2 * 8, 7 + 6 * 8,
  85. 1 + 0 * 8, 1 + 2 * 8, 0 + 6 * 8, 3 + 0 * 8,
  86. 1 + 7 * 8, 3 + 2 * 8, 2 + 7 * 8, 4 + 2 * 8,
  87. 3 + 7 * 8, 5 + 2 * 8, 4 + 7 * 8, 5 + 4 * 8,
  88. 6 + 3 * 8, 6 + 5 * 8, 7 + 3 * 8, 7 + 7 * 8,
  89. };
  90. // zigzag_scan8x8_cavlc[i] = zigzag_scan8x8[(i/4) + 16*(i%4)]
  91. static const uint8_t zigzag_scan8x8_cavlc[64+1] = {
  92. 0 + 0 * 8, 1 + 1 * 8, 1 + 2 * 8, 2 + 2 * 8,
  93. 4 + 1 * 8, 0 + 5 * 8, 3 + 3 * 8, 7 + 0 * 8,
  94. 3 + 4 * 8, 1 + 7 * 8, 5 + 3 * 8, 6 + 3 * 8,
  95. 2 + 7 * 8, 6 + 4 * 8, 5 + 6 * 8, 7 + 5 * 8,
  96. 1 + 0 * 8, 2 + 0 * 8, 0 + 3 * 8, 3 + 1 * 8,
  97. 3 + 2 * 8, 0 + 6 * 8, 4 + 2 * 8, 6 + 1 * 8,
  98. 2 + 5 * 8, 2 + 6 * 8, 6 + 2 * 8, 5 + 4 * 8,
  99. 3 + 7 * 8, 7 + 3 * 8, 4 + 7 * 8, 7 + 6 * 8,
  100. 0 + 1 * 8, 3 + 0 * 8, 0 + 4 * 8, 4 + 0 * 8,
  101. 2 + 3 * 8, 1 + 5 * 8, 5 + 1 * 8, 5 + 2 * 8,
  102. 1 + 6 * 8, 3 + 5 * 8, 7 + 1 * 8, 4 + 5 * 8,
  103. 4 + 6 * 8, 7 + 4 * 8, 5 + 7 * 8, 6 + 7 * 8,
  104. 0 + 2 * 8, 2 + 1 * 8, 1 + 3 * 8, 5 + 0 * 8,
  105. 1 + 4 * 8, 2 + 4 * 8, 6 + 0 * 8, 4 + 3 * 8,
  106. 0 + 7 * 8, 4 + 4 * 8, 7 + 2 * 8, 3 + 6 * 8,
  107. 5 + 5 * 8, 6 + 5 * 8, 6 + 6 * 8, 7 + 7 * 8,
  108. };
  109. static void release_unused_pictures(H264Context *h, int remove_current)
  110. {
  111. int i;
  112. /* release non reference frames */
  113. for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
  114. if (h->DPB[i].f->buf[0] && !h->DPB[i].reference &&
  115. (remove_current || &h->DPB[i] != h->cur_pic_ptr)) {
  116. ff_h264_unref_picture(h, &h->DPB[i]);
  117. }
  118. }
  119. }
  120. static int alloc_scratch_buffers(H264SliceContext *sl, int linesize)
  121. {
  122. const H264Context *h = sl->h264;
  123. int alloc_size = FFALIGN(FFABS(linesize) + 32, 32);
  124. av_fast_malloc(&sl->bipred_scratchpad, &sl->bipred_scratchpad_allocated, 16 * 6 * alloc_size);
  125. // edge emu needs blocksize + filter length - 1
  126. // (= 21x21 for H.264)
  127. av_fast_malloc(&sl->edge_emu_buffer, &sl->edge_emu_buffer_allocated, alloc_size * 2 * 21);
  128. av_fast_mallocz(&sl->top_borders[0], &sl->top_borders_allocated[0],
  129. h->mb_width * 16 * 3 * sizeof(uint8_t) * 2);
  130. av_fast_mallocz(&sl->top_borders[1], &sl->top_borders_allocated[1],
  131. h->mb_width * 16 * 3 * sizeof(uint8_t) * 2);
  132. if (!sl->bipred_scratchpad || !sl->edge_emu_buffer ||
  133. !sl->top_borders[0] || !sl->top_borders[1]) {
  134. av_freep(&sl->bipred_scratchpad);
  135. av_freep(&sl->edge_emu_buffer);
  136. av_freep(&sl->top_borders[0]);
  137. av_freep(&sl->top_borders[1]);
  138. sl->bipred_scratchpad_allocated = 0;
  139. sl->edge_emu_buffer_allocated = 0;
  140. sl->top_borders_allocated[0] = 0;
  141. sl->top_borders_allocated[1] = 0;
  142. return AVERROR(ENOMEM);
  143. }
  144. return 0;
  145. }
  146. static int init_table_pools(H264Context *h)
  147. {
  148. const int big_mb_num = h->mb_stride * (h->mb_height + 1) + 1;
  149. const int mb_array_size = h->mb_stride * h->mb_height;
  150. const int b4_stride = h->mb_width * 4 + 1;
  151. const int b4_array_size = b4_stride * h->mb_height * 4;
  152. h->qscale_table_pool = av_buffer_pool_init(big_mb_num + h->mb_stride,
  153. av_buffer_allocz);
  154. h->mb_type_pool = av_buffer_pool_init((big_mb_num + h->mb_stride) *
  155. sizeof(uint32_t), av_buffer_allocz);
  156. h->motion_val_pool = av_buffer_pool_init(2 * (b4_array_size + 4) *
  157. sizeof(int16_t), av_buffer_allocz);
  158. h->ref_index_pool = av_buffer_pool_init(4 * mb_array_size, av_buffer_allocz);
  159. if (!h->qscale_table_pool || !h->mb_type_pool || !h->motion_val_pool ||
  160. !h->ref_index_pool) {
  161. av_buffer_pool_uninit(&h->qscale_table_pool);
  162. av_buffer_pool_uninit(&h->mb_type_pool);
  163. av_buffer_pool_uninit(&h->motion_val_pool);
  164. av_buffer_pool_uninit(&h->ref_index_pool);
  165. return AVERROR(ENOMEM);
  166. }
  167. return 0;
  168. }
  169. static int alloc_picture(H264Context *h, H264Picture *pic)
  170. {
  171. int i, ret = 0;
  172. av_assert0(!pic->f->data[0]);
  173. pic->tf.f = pic->f;
  174. ret = ff_thread_get_buffer(h->avctx, &pic->tf, pic->reference ?
  175. AV_GET_BUFFER_FLAG_REF : 0);
  176. if (ret < 0)
  177. goto fail;
  178. if (h->avctx->hwaccel) {
  179. const AVHWAccel *hwaccel = h->avctx->hwaccel;
  180. av_assert0(!pic->hwaccel_picture_private);
  181. if (hwaccel->frame_priv_data_size) {
  182. pic->hwaccel_priv_buf = av_buffer_allocz(hwaccel->frame_priv_data_size);
  183. if (!pic->hwaccel_priv_buf)
  184. return AVERROR(ENOMEM);
  185. pic->hwaccel_picture_private = pic->hwaccel_priv_buf->data;
  186. }
  187. }
  188. if (CONFIG_GRAY && !h->avctx->hwaccel && h->flags & AV_CODEC_FLAG_GRAY && pic->f->data[2]) {
  189. int h_chroma_shift, v_chroma_shift;
  190. av_pix_fmt_get_chroma_sub_sample(pic->f->format,
  191. &h_chroma_shift, &v_chroma_shift);
  192. for(i=0; i<AV_CEIL_RSHIFT(pic->f->height, v_chroma_shift); i++) {
  193. memset(pic->f->data[1] + pic->f->linesize[1]*i,
  194. 0x80, AV_CEIL_RSHIFT(pic->f->width, h_chroma_shift));
  195. memset(pic->f->data[2] + pic->f->linesize[2]*i,
  196. 0x80, AV_CEIL_RSHIFT(pic->f->width, h_chroma_shift));
  197. }
  198. }
  199. if (!h->qscale_table_pool) {
  200. ret = init_table_pools(h);
  201. if (ret < 0)
  202. goto fail;
  203. }
  204. pic->qscale_table_buf = av_buffer_pool_get(h->qscale_table_pool);
  205. pic->mb_type_buf = av_buffer_pool_get(h->mb_type_pool);
  206. if (!pic->qscale_table_buf || !pic->mb_type_buf)
  207. goto fail;
  208. pic->mb_type = (uint32_t*)pic->mb_type_buf->data + 2 * h->mb_stride + 1;
  209. pic->qscale_table = pic->qscale_table_buf->data + 2 * h->mb_stride + 1;
  210. for (i = 0; i < 2; i++) {
  211. pic->motion_val_buf[i] = av_buffer_pool_get(h->motion_val_pool);
  212. pic->ref_index_buf[i] = av_buffer_pool_get(h->ref_index_pool);
  213. if (!pic->motion_val_buf[i] || !pic->ref_index_buf[i])
  214. goto fail;
  215. pic->motion_val[i] = (int16_t (*)[2])pic->motion_val_buf[i]->data + 4;
  216. pic->ref_index[i] = pic->ref_index_buf[i]->data;
  217. }
  218. return 0;
  219. fail:
  220. ff_h264_unref_picture(h, pic);
  221. return (ret < 0) ? ret : AVERROR(ENOMEM);
  222. }
  223. static int find_unused_picture(H264Context *h)
  224. {
  225. int i;
  226. for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
  227. if (!h->DPB[i].f->buf[0])
  228. return i;
  229. }
  230. return AVERROR_INVALIDDATA;
  231. }
  232. #define IN_RANGE(a, b, size) (((void*)(a) >= (void*)(b)) && ((void*)(a) < (void*)((b) + (size))))
  233. #define REBASE_PICTURE(pic, new_ctx, old_ctx) \
  234. (((pic) && (pic) >= (old_ctx)->DPB && \
  235. (pic) < (old_ctx)->DPB + H264_MAX_PICTURE_COUNT) ? \
  236. &(new_ctx)->DPB[(pic) - (old_ctx)->DPB] : NULL)
  237. static void copy_picture_range(H264Picture **to, H264Picture **from, int count,
  238. H264Context *new_base,
  239. H264Context *old_base)
  240. {
  241. int i;
  242. for (i = 0; i < count; i++) {
  243. av_assert1(!from[i] ||
  244. IN_RANGE(from[i], old_base, 1) ||
  245. IN_RANGE(from[i], old_base->DPB, H264_MAX_PICTURE_COUNT));
  246. to[i] = REBASE_PICTURE(from[i], new_base, old_base);
  247. }
  248. }
  249. static int h264_slice_header_init(H264Context *h);
  250. int ff_h264_update_thread_context(AVCodecContext *dst,
  251. const AVCodecContext *src)
  252. {
  253. H264Context *h = dst->priv_data, *h1 = src->priv_data;
  254. int inited = h->context_initialized, err = 0;
  255. int need_reinit = 0;
  256. int i, ret;
  257. if (dst == src)
  258. return 0;
  259. // We can't fail if SPS isn't set at it breaks current skip_frame code
  260. //if (!h1->ps.sps)
  261. // return AVERROR_INVALIDDATA;
  262. if (inited &&
  263. (h->width != h1->width ||
  264. h->height != h1->height ||
  265. h->mb_width != h1->mb_width ||
  266. h->mb_height != h1->mb_height ||
  267. !h->ps.sps ||
  268. h->ps.sps->bit_depth_luma != h1->ps.sps->bit_depth_luma ||
  269. h->ps.sps->chroma_format_idc != h1->ps.sps->chroma_format_idc ||
  270. h->ps.sps->colorspace != h1->ps.sps->colorspace)) {
  271. need_reinit = 1;
  272. }
  273. /* copy block_offset since frame_start may not be called */
  274. memcpy(h->block_offset, h1->block_offset, sizeof(h->block_offset));
  275. // SPS/PPS
  276. for (i = 0; i < FF_ARRAY_ELEMS(h->ps.sps_list); i++) {
  277. av_buffer_unref(&h->ps.sps_list[i]);
  278. if (h1->ps.sps_list[i]) {
  279. h->ps.sps_list[i] = av_buffer_ref(h1->ps.sps_list[i]);
  280. if (!h->ps.sps_list[i])
  281. return AVERROR(ENOMEM);
  282. }
  283. }
  284. for (i = 0; i < FF_ARRAY_ELEMS(h->ps.pps_list); i++) {
  285. av_buffer_unref(&h->ps.pps_list[i]);
  286. if (h1->ps.pps_list[i]) {
  287. h->ps.pps_list[i] = av_buffer_ref(h1->ps.pps_list[i]);
  288. if (!h->ps.pps_list[i])
  289. return AVERROR(ENOMEM);
  290. }
  291. }
  292. av_buffer_unref(&h->ps.pps_ref);
  293. av_buffer_unref(&h->ps.sps_ref);
  294. h->ps.pps = NULL;
  295. h->ps.sps = NULL;
  296. if (h1->ps.pps_ref) {
  297. h->ps.pps_ref = av_buffer_ref(h1->ps.pps_ref);
  298. if (!h->ps.pps_ref)
  299. return AVERROR(ENOMEM);
  300. h->ps.pps = (const PPS*)h->ps.pps_ref->data;
  301. }
  302. if (h1->ps.sps_ref) {
  303. h->ps.sps_ref = av_buffer_ref(h1->ps.sps_ref);
  304. if (!h->ps.sps_ref)
  305. return AVERROR(ENOMEM);
  306. h->ps.sps = (const SPS*)h->ps.sps_ref->data;
  307. }
  308. if (need_reinit || !inited) {
  309. h->width = h1->width;
  310. h->height = h1->height;
  311. h->mb_height = h1->mb_height;
  312. h->mb_width = h1->mb_width;
  313. h->mb_num = h1->mb_num;
  314. h->mb_stride = h1->mb_stride;
  315. h->b_stride = h1->b_stride;
  316. if (h->context_initialized || h1->context_initialized) {
  317. if ((err = h264_slice_header_init(h)) < 0) {
  318. av_log(h->avctx, AV_LOG_ERROR, "h264_slice_header_init() failed");
  319. return err;
  320. }
  321. }
  322. /* copy block_offset since frame_start may not be called */
  323. memcpy(h->block_offset, h1->block_offset, sizeof(h->block_offset));
  324. }
  325. h->avctx->coded_height = h1->avctx->coded_height;
  326. h->avctx->coded_width = h1->avctx->coded_width;
  327. h->avctx->width = h1->avctx->width;
  328. h->avctx->height = h1->avctx->height;
  329. h->width_from_caller = h1->width_from_caller;
  330. h->height_from_caller = h1->height_from_caller;
  331. h->coded_picture_number = h1->coded_picture_number;
  332. h->first_field = h1->first_field;
  333. h->picture_structure = h1->picture_structure;
  334. h->mb_aff_frame = h1->mb_aff_frame;
  335. h->droppable = h1->droppable;
  336. for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
  337. ff_h264_unref_picture(h, &h->DPB[i]);
  338. if (h1->DPB[i].f->buf[0] &&
  339. (ret = ff_h264_ref_picture(h, &h->DPB[i], &h1->DPB[i])) < 0)
  340. return ret;
  341. }
  342. h->cur_pic_ptr = REBASE_PICTURE(h1->cur_pic_ptr, h, h1);
  343. ff_h264_unref_picture(h, &h->cur_pic);
  344. if (h1->cur_pic.f->buf[0]) {
  345. ret = ff_h264_ref_picture(h, &h->cur_pic, &h1->cur_pic);
  346. if (ret < 0)
  347. return ret;
  348. }
  349. h->enable_er = h1->enable_er;
  350. h->workaround_bugs = h1->workaround_bugs;
  351. h->x264_build = h1->x264_build;
  352. h->droppable = h1->droppable;
  353. // extradata/NAL handling
  354. h->is_avc = h1->is_avc;
  355. h->nal_length_size = h1->nal_length_size;
  356. memcpy(&h->poc, &h1->poc, sizeof(h->poc));
  357. memcpy(h->short_ref, h1->short_ref, sizeof(h->short_ref));
  358. memcpy(h->long_ref, h1->long_ref, sizeof(h->long_ref));
  359. memcpy(h->delayed_pic, h1->delayed_pic, sizeof(h->delayed_pic));
  360. memcpy(h->last_pocs, h1->last_pocs, sizeof(h->last_pocs));
  361. h->next_output_pic = h1->next_output_pic;
  362. h->next_outputed_poc = h1->next_outputed_poc;
  363. memcpy(h->mmco, h1->mmco, sizeof(h->mmco));
  364. h->nb_mmco = h1->nb_mmco;
  365. h->mmco_reset = h1->mmco_reset;
  366. h->explicit_ref_marking = h1->explicit_ref_marking;
  367. h->long_ref_count = h1->long_ref_count;
  368. h->short_ref_count = h1->short_ref_count;
  369. copy_picture_range(h->short_ref, h1->short_ref, 32, h, h1);
  370. copy_picture_range(h->long_ref, h1->long_ref, 32, h, h1);
  371. copy_picture_range(h->delayed_pic, h1->delayed_pic,
  372. MAX_DELAYED_PIC_COUNT + 2, h, h1);
  373. h->frame_recovered = h1->frame_recovered;
  374. if (!h->cur_pic_ptr)
  375. return 0;
  376. if (!h->droppable) {
  377. err = ff_h264_execute_ref_pic_marking(h);
  378. h->poc.prev_poc_msb = h->poc.poc_msb;
  379. h->poc.prev_poc_lsb = h->poc.poc_lsb;
  380. }
  381. h->poc.prev_frame_num_offset = h->poc.frame_num_offset;
  382. h->poc.prev_frame_num = h->poc.frame_num;
  383. h->recovery_frame = h1->recovery_frame;
  384. return err;
  385. }
  386. static int h264_frame_start(H264Context *h)
  387. {
  388. H264Picture *pic;
  389. int i, ret;
  390. const int pixel_shift = h->pixel_shift;
  391. int c[4] = {
  392. 1<<(h->ps.sps->bit_depth_luma-1),
  393. 1<<(h->ps.sps->bit_depth_chroma-1),
  394. 1<<(h->ps.sps->bit_depth_chroma-1),
  395. -1
  396. };
  397. if (!ff_thread_can_start_frame(h->avctx)) {
  398. av_log(h->avctx, AV_LOG_ERROR, "Attempt to start a frame outside SETUP state\n");
  399. return -1;
  400. }
  401. release_unused_pictures(h, 1);
  402. h->cur_pic_ptr = NULL;
  403. i = find_unused_picture(h);
  404. if (i < 0) {
  405. av_log(h->avctx, AV_LOG_ERROR, "no frame buffer available\n");
  406. return i;
  407. }
  408. pic = &h->DPB[i];
  409. pic->reference = h->droppable ? 0 : h->picture_structure;
  410. pic->f->coded_picture_number = h->coded_picture_number++;
  411. pic->field_picture = h->picture_structure != PICT_FRAME;
  412. pic->frame_num = h->poc.frame_num;
  413. /*
  414. * Zero key_frame here; IDR markings per slice in frame or fields are ORed
  415. * in later.
  416. * See decode_nal_units().
  417. */
  418. pic->f->key_frame = 0;
  419. pic->mmco_reset = 0;
  420. pic->recovered = 0;
  421. pic->invalid_gap = 0;
  422. pic->sei_recovery_frame_cnt = h->sei.recovery_point.recovery_frame_cnt;
  423. pic->f->pict_type = h->slice_ctx[0].slice_type;
  424. pic->f->crop_left = h->crop_left;
  425. pic->f->crop_right = h->crop_right;
  426. pic->f->crop_top = h->crop_top;
  427. pic->f->crop_bottom = h->crop_bottom;
  428. if ((ret = alloc_picture(h, pic)) < 0)
  429. return ret;
  430. if(!h->frame_recovered && !h->avctx->hwaccel)
  431. ff_color_frame(pic->f, c);
  432. h->cur_pic_ptr = pic;
  433. ff_h264_unref_picture(h, &h->cur_pic);
  434. if (CONFIG_ERROR_RESILIENCE) {
  435. ff_h264_set_erpic(&h->slice_ctx[0].er.cur_pic, NULL);
  436. }
  437. if ((ret = ff_h264_ref_picture(h, &h->cur_pic, h->cur_pic_ptr)) < 0)
  438. return ret;
  439. for (i = 0; i < h->nb_slice_ctx; i++) {
  440. h->slice_ctx[i].linesize = h->cur_pic_ptr->f->linesize[0];
  441. h->slice_ctx[i].uvlinesize = h->cur_pic_ptr->f->linesize[1];
  442. }
  443. if (CONFIG_ERROR_RESILIENCE && h->enable_er) {
  444. ff_er_frame_start(&h->slice_ctx[0].er);
  445. ff_h264_set_erpic(&h->slice_ctx[0].er.last_pic, NULL);
  446. ff_h264_set_erpic(&h->slice_ctx[0].er.next_pic, NULL);
  447. }
  448. for (i = 0; i < 16; i++) {
  449. h->block_offset[i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * pic->f->linesize[0] * ((scan8[i] - scan8[0]) >> 3);
  450. h->block_offset[48 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * pic->f->linesize[0] * ((scan8[i] - scan8[0]) >> 3);
  451. }
  452. for (i = 0; i < 16; i++) {
  453. h->block_offset[16 + i] =
  454. h->block_offset[32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * pic->f->linesize[1] * ((scan8[i] - scan8[0]) >> 3);
  455. h->block_offset[48 + 16 + i] =
  456. h->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * pic->f->linesize[1] * ((scan8[i] - scan8[0]) >> 3);
  457. }
  458. /* We mark the current picture as non-reference after allocating it, so
  459. * that if we break out due to an error it can be released automatically
  460. * in the next ff_mpv_frame_start().
  461. */
  462. h->cur_pic_ptr->reference = 0;
  463. h->cur_pic_ptr->field_poc[0] = h->cur_pic_ptr->field_poc[1] = INT_MAX;
  464. h->next_output_pic = NULL;
  465. h->postpone_filter = 0;
  466. h->mb_aff_frame = h->ps.sps->mb_aff && (h->picture_structure == PICT_FRAME);
  467. if (h->sei.unregistered.x264_build >= 0)
  468. h->x264_build = h->sei.unregistered.x264_build;
  469. assert(h->cur_pic_ptr->long_ref == 0);
  470. return 0;
  471. }
  472. static av_always_inline void backup_mb_border(const H264Context *h, H264SliceContext *sl,
  473. uint8_t *src_y,
  474. uint8_t *src_cb, uint8_t *src_cr,
  475. int linesize, int uvlinesize,
  476. int simple)
  477. {
  478. uint8_t *top_border;
  479. int top_idx = 1;
  480. const int pixel_shift = h->pixel_shift;
  481. int chroma444 = CHROMA444(h);
  482. int chroma422 = CHROMA422(h);
  483. src_y -= linesize;
  484. src_cb -= uvlinesize;
  485. src_cr -= uvlinesize;
  486. if (!simple && FRAME_MBAFF(h)) {
  487. if (sl->mb_y & 1) {
  488. if (!MB_MBAFF(sl)) {
  489. top_border = sl->top_borders[0][sl->mb_x];
  490. AV_COPY128(top_border, src_y + 15 * linesize);
  491. if (pixel_shift)
  492. AV_COPY128(top_border + 16, src_y + 15 * linesize + 16);
  493. if (simple || !CONFIG_GRAY || !(h->flags & AV_CODEC_FLAG_GRAY)) {
  494. if (chroma444) {
  495. if (pixel_shift) {
  496. AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
  497. AV_COPY128(top_border + 48, src_cb + 15 * uvlinesize + 16);
  498. AV_COPY128(top_border + 64, src_cr + 15 * uvlinesize);
  499. AV_COPY128(top_border + 80, src_cr + 15 * uvlinesize + 16);
  500. } else {
  501. AV_COPY128(top_border + 16, src_cb + 15 * uvlinesize);
  502. AV_COPY128(top_border + 32, src_cr + 15 * uvlinesize);
  503. }
  504. } else if (chroma422) {
  505. if (pixel_shift) {
  506. AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
  507. AV_COPY128(top_border + 48, src_cr + 15 * uvlinesize);
  508. } else {
  509. AV_COPY64(top_border + 16, src_cb + 15 * uvlinesize);
  510. AV_COPY64(top_border + 24, src_cr + 15 * uvlinesize);
  511. }
  512. } else {
  513. if (pixel_shift) {
  514. AV_COPY128(top_border + 32, src_cb + 7 * uvlinesize);
  515. AV_COPY128(top_border + 48, src_cr + 7 * uvlinesize);
  516. } else {
  517. AV_COPY64(top_border + 16, src_cb + 7 * uvlinesize);
  518. AV_COPY64(top_border + 24, src_cr + 7 * uvlinesize);
  519. }
  520. }
  521. }
  522. }
  523. } else if (MB_MBAFF(sl)) {
  524. top_idx = 0;
  525. } else
  526. return;
  527. }
  528. top_border = sl->top_borders[top_idx][sl->mb_x];
  529. /* There are two lines saved, the line above the top macroblock
  530. * of a pair, and the line above the bottom macroblock. */
  531. AV_COPY128(top_border, src_y + 16 * linesize);
  532. if (pixel_shift)
  533. AV_COPY128(top_border + 16, src_y + 16 * linesize + 16);
  534. if (simple || !CONFIG_GRAY || !(h->flags & AV_CODEC_FLAG_GRAY)) {
  535. if (chroma444) {
  536. if (pixel_shift) {
  537. AV_COPY128(top_border + 32, src_cb + 16 * linesize);
  538. AV_COPY128(top_border + 48, src_cb + 16 * linesize + 16);
  539. AV_COPY128(top_border + 64, src_cr + 16 * linesize);
  540. AV_COPY128(top_border + 80, src_cr + 16 * linesize + 16);
  541. } else {
  542. AV_COPY128(top_border + 16, src_cb + 16 * linesize);
  543. AV_COPY128(top_border + 32, src_cr + 16 * linesize);
  544. }
  545. } else if (chroma422) {
  546. if (pixel_shift) {
  547. AV_COPY128(top_border + 32, src_cb + 16 * uvlinesize);
  548. AV_COPY128(top_border + 48, src_cr + 16 * uvlinesize);
  549. } else {
  550. AV_COPY64(top_border + 16, src_cb + 16 * uvlinesize);
  551. AV_COPY64(top_border + 24, src_cr + 16 * uvlinesize);
  552. }
  553. } else {
  554. if (pixel_shift) {
  555. AV_COPY128(top_border + 32, src_cb + 8 * uvlinesize);
  556. AV_COPY128(top_border + 48, src_cr + 8 * uvlinesize);
  557. } else {
  558. AV_COPY64(top_border + 16, src_cb + 8 * uvlinesize);
  559. AV_COPY64(top_border + 24, src_cr + 8 * uvlinesize);
  560. }
  561. }
  562. }
  563. }
  564. /**
  565. * Initialize implicit_weight table.
  566. * @param field 0/1 initialize the weight for interlaced MBAFF
  567. * -1 initializes the rest
  568. */
  569. static void implicit_weight_table(const H264Context *h, H264SliceContext *sl, int field)
  570. {
  571. int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
  572. for (i = 0; i < 2; i++) {
  573. sl->pwt.luma_weight_flag[i] = 0;
  574. sl->pwt.chroma_weight_flag[i] = 0;
  575. }
  576. if (field < 0) {
  577. if (h->picture_structure == PICT_FRAME) {
  578. cur_poc = h->cur_pic_ptr->poc;
  579. } else {
  580. cur_poc = h->cur_pic_ptr->field_poc[h->picture_structure - 1];
  581. }
  582. if (sl->ref_count[0] == 1 && sl->ref_count[1] == 1 && !FRAME_MBAFF(h) &&
  583. sl->ref_list[0][0].poc + (int64_t)sl->ref_list[1][0].poc == 2 * cur_poc) {
  584. sl->pwt.use_weight = 0;
  585. sl->pwt.use_weight_chroma = 0;
  586. return;
  587. }
  588. ref_start = 0;
  589. ref_count0 = sl->ref_count[0];
  590. ref_count1 = sl->ref_count[1];
  591. } else {
  592. cur_poc = h->cur_pic_ptr->field_poc[field];
  593. ref_start = 16;
  594. ref_count0 = 16 + 2 * sl->ref_count[0];
  595. ref_count1 = 16 + 2 * sl->ref_count[1];
  596. }
  597. sl->pwt.use_weight = 2;
  598. sl->pwt.use_weight_chroma = 2;
  599. sl->pwt.luma_log2_weight_denom = 5;
  600. sl->pwt.chroma_log2_weight_denom = 5;
  601. for (ref0 = ref_start; ref0 < ref_count0; ref0++) {
  602. int64_t poc0 = sl->ref_list[0][ref0].poc;
  603. for (ref1 = ref_start; ref1 < ref_count1; ref1++) {
  604. int w = 32;
  605. if (!sl->ref_list[0][ref0].parent->long_ref && !sl->ref_list[1][ref1].parent->long_ref) {
  606. int poc1 = sl->ref_list[1][ref1].poc;
  607. int td = av_clip_int8(poc1 - poc0);
  608. if (td) {
  609. int tb = av_clip_int8(cur_poc - poc0);
  610. int tx = (16384 + (FFABS(td) >> 1)) / td;
  611. int dist_scale_factor = (tb * tx + 32) >> 8;
  612. if (dist_scale_factor >= -64 && dist_scale_factor <= 128)
  613. w = 64 - dist_scale_factor;
  614. }
  615. }
  616. if (field < 0) {
  617. sl->pwt.implicit_weight[ref0][ref1][0] =
  618. sl->pwt.implicit_weight[ref0][ref1][1] = w;
  619. } else {
  620. sl->pwt.implicit_weight[ref0][ref1][field] = w;
  621. }
  622. }
  623. }
  624. }
  625. /**
  626. * initialize scan tables
  627. */
  628. static void init_scan_tables(H264Context *h)
  629. {
  630. int i;
  631. for (i = 0; i < 16; i++) {
  632. #define TRANSPOSE(x) ((x) >> 2) | (((x) << 2) & 0xF)
  633. h->zigzag_scan[i] = TRANSPOSE(ff_zigzag_scan[i]);
  634. h->field_scan[i] = TRANSPOSE(field_scan[i]);
  635. #undef TRANSPOSE
  636. }
  637. for (i = 0; i < 64; i++) {
  638. #define TRANSPOSE(x) ((x) >> 3) | (((x) & 7) << 3)
  639. h->zigzag_scan8x8[i] = TRANSPOSE(ff_zigzag_direct[i]);
  640. h->zigzag_scan8x8_cavlc[i] = TRANSPOSE(zigzag_scan8x8_cavlc[i]);
  641. h->field_scan8x8[i] = TRANSPOSE(field_scan8x8[i]);
  642. h->field_scan8x8_cavlc[i] = TRANSPOSE(field_scan8x8_cavlc[i]);
  643. #undef TRANSPOSE
  644. }
  645. if (h->ps.sps->transform_bypass) { // FIXME same ugly
  646. memcpy(h->zigzag_scan_q0 , ff_zigzag_scan , sizeof(h->zigzag_scan_q0 ));
  647. memcpy(h->zigzag_scan8x8_q0 , ff_zigzag_direct , sizeof(h->zigzag_scan8x8_q0 ));
  648. memcpy(h->zigzag_scan8x8_cavlc_q0 , zigzag_scan8x8_cavlc , sizeof(h->zigzag_scan8x8_cavlc_q0));
  649. memcpy(h->field_scan_q0 , field_scan , sizeof(h->field_scan_q0 ));
  650. memcpy(h->field_scan8x8_q0 , field_scan8x8 , sizeof(h->field_scan8x8_q0 ));
  651. memcpy(h->field_scan8x8_cavlc_q0 , field_scan8x8_cavlc , sizeof(h->field_scan8x8_cavlc_q0 ));
  652. } else {
  653. memcpy(h->zigzag_scan_q0 , h->zigzag_scan , sizeof(h->zigzag_scan_q0 ));
  654. memcpy(h->zigzag_scan8x8_q0 , h->zigzag_scan8x8 , sizeof(h->zigzag_scan8x8_q0 ));
  655. memcpy(h->zigzag_scan8x8_cavlc_q0 , h->zigzag_scan8x8_cavlc , sizeof(h->zigzag_scan8x8_cavlc_q0));
  656. memcpy(h->field_scan_q0 , h->field_scan , sizeof(h->field_scan_q0 ));
  657. memcpy(h->field_scan8x8_q0 , h->field_scan8x8 , sizeof(h->field_scan8x8_q0 ));
  658. memcpy(h->field_scan8x8_cavlc_q0 , h->field_scan8x8_cavlc , sizeof(h->field_scan8x8_cavlc_q0 ));
  659. }
  660. }
  661. static enum AVPixelFormat get_pixel_format(H264Context *h, int force_callback)
  662. {
  663. #define HWACCEL_MAX (CONFIG_H264_DXVA2_HWACCEL + \
  664. (CONFIG_H264_D3D11VA_HWACCEL * 2) + \
  665. CONFIG_H264_NVDEC_HWACCEL + \
  666. CONFIG_H264_VAAPI_HWACCEL + \
  667. CONFIG_H264_VIDEOTOOLBOX_HWACCEL + \
  668. CONFIG_H264_VDPAU_HWACCEL)
  669. enum AVPixelFormat pix_fmts[HWACCEL_MAX + 2], *fmt = pix_fmts;
  670. const enum AVPixelFormat *choices = pix_fmts;
  671. int i;
  672. switch (h->ps.sps->bit_depth_luma) {
  673. case 9:
  674. if (CHROMA444(h)) {
  675. if (h->avctx->colorspace == AVCOL_SPC_RGB) {
  676. *fmt++ = AV_PIX_FMT_GBRP9;
  677. } else
  678. *fmt++ = AV_PIX_FMT_YUV444P9;
  679. } else if (CHROMA422(h))
  680. *fmt++ = AV_PIX_FMT_YUV422P9;
  681. else
  682. *fmt++ = AV_PIX_FMT_YUV420P9;
  683. break;
  684. case 10:
  685. if (CHROMA444(h)) {
  686. if (h->avctx->colorspace == AVCOL_SPC_RGB) {
  687. *fmt++ = AV_PIX_FMT_GBRP10;
  688. } else
  689. *fmt++ = AV_PIX_FMT_YUV444P10;
  690. } else if (CHROMA422(h))
  691. *fmt++ = AV_PIX_FMT_YUV422P10;
  692. else
  693. *fmt++ = AV_PIX_FMT_YUV420P10;
  694. break;
  695. case 12:
  696. if (CHROMA444(h)) {
  697. if (h->avctx->colorspace == AVCOL_SPC_RGB) {
  698. *fmt++ = AV_PIX_FMT_GBRP12;
  699. } else
  700. *fmt++ = AV_PIX_FMT_YUV444P12;
  701. } else if (CHROMA422(h))
  702. *fmt++ = AV_PIX_FMT_YUV422P12;
  703. else
  704. *fmt++ = AV_PIX_FMT_YUV420P12;
  705. break;
  706. case 14:
  707. if (CHROMA444(h)) {
  708. if (h->avctx->colorspace == AVCOL_SPC_RGB) {
  709. *fmt++ = AV_PIX_FMT_GBRP14;
  710. } else
  711. *fmt++ = AV_PIX_FMT_YUV444P14;
  712. } else if (CHROMA422(h))
  713. *fmt++ = AV_PIX_FMT_YUV422P14;
  714. else
  715. *fmt++ = AV_PIX_FMT_YUV420P14;
  716. break;
  717. case 8:
  718. #if CONFIG_H264_VDPAU_HWACCEL
  719. *fmt++ = AV_PIX_FMT_VDPAU;
  720. #endif
  721. #if CONFIG_H264_NVDEC_HWACCEL
  722. *fmt++ = AV_PIX_FMT_CUDA;
  723. #endif
  724. if (CHROMA444(h)) {
  725. if (h->avctx->colorspace == AVCOL_SPC_RGB)
  726. *fmt++ = AV_PIX_FMT_GBRP;
  727. else if (h->avctx->color_range == AVCOL_RANGE_JPEG)
  728. *fmt++ = AV_PIX_FMT_YUVJ444P;
  729. else
  730. *fmt++ = AV_PIX_FMT_YUV444P;
  731. } else if (CHROMA422(h)) {
  732. if (h->avctx->color_range == AVCOL_RANGE_JPEG)
  733. *fmt++ = AV_PIX_FMT_YUVJ422P;
  734. else
  735. *fmt++ = AV_PIX_FMT_YUV422P;
  736. } else {
  737. #if CONFIG_H264_DXVA2_HWACCEL
  738. *fmt++ = AV_PIX_FMT_DXVA2_VLD;
  739. #endif
  740. #if CONFIG_H264_D3D11VA_HWACCEL
  741. *fmt++ = AV_PIX_FMT_D3D11VA_VLD;
  742. *fmt++ = AV_PIX_FMT_D3D11;
  743. #endif
  744. #if CONFIG_H264_VAAPI_HWACCEL
  745. *fmt++ = AV_PIX_FMT_VAAPI;
  746. #endif
  747. #if CONFIG_H264_VIDEOTOOLBOX_HWACCEL
  748. *fmt++ = AV_PIX_FMT_VIDEOTOOLBOX;
  749. #endif
  750. if (h->avctx->codec->pix_fmts)
  751. choices = h->avctx->codec->pix_fmts;
  752. else if (h->avctx->color_range == AVCOL_RANGE_JPEG)
  753. *fmt++ = AV_PIX_FMT_YUVJ420P;
  754. else
  755. *fmt++ = AV_PIX_FMT_YUV420P;
  756. }
  757. break;
  758. default:
  759. av_log(h->avctx, AV_LOG_ERROR,
  760. "Unsupported bit depth %d\n", h->ps.sps->bit_depth_luma);
  761. return AVERROR_INVALIDDATA;
  762. }
  763. *fmt = AV_PIX_FMT_NONE;
  764. for (i=0; choices[i] != AV_PIX_FMT_NONE; i++)
  765. if (choices[i] == h->avctx->pix_fmt && !force_callback)
  766. return choices[i];
  767. return ff_thread_get_format(h->avctx, choices);
  768. }
  769. /* export coded and cropped frame dimensions to AVCodecContext */
  770. static int init_dimensions(H264Context *h)
  771. {
  772. const SPS *sps = (const SPS*)h->ps.sps;
  773. int cr = sps->crop_right;
  774. int cl = sps->crop_left;
  775. int ct = sps->crop_top;
  776. int cb = sps->crop_bottom;
  777. int width = h->width - (cr + cl);
  778. int height = h->height - (ct + cb);
  779. av_assert0(sps->crop_right + sps->crop_left < (unsigned)h->width);
  780. av_assert0(sps->crop_top + sps->crop_bottom < (unsigned)h->height);
  781. /* handle container cropping */
  782. if (h->width_from_caller > 0 && h->height_from_caller > 0 &&
  783. !sps->crop_top && !sps->crop_left &&
  784. FFALIGN(h->width_from_caller, 16) == FFALIGN(width, 16) &&
  785. FFALIGN(h->height_from_caller, 16) == FFALIGN(height, 16) &&
  786. h->width_from_caller <= width &&
  787. h->height_from_caller <= height) {
  788. width = h->width_from_caller;
  789. height = h->height_from_caller;
  790. cl = 0;
  791. ct = 0;
  792. cr = h->width - width;
  793. cb = h->height - height;
  794. } else {
  795. h->width_from_caller = 0;
  796. h->height_from_caller = 0;
  797. }
  798. h->avctx->coded_width = h->width;
  799. h->avctx->coded_height = h->height;
  800. h->avctx->width = width;
  801. h->avctx->height = height;
  802. h->crop_right = cr;
  803. h->crop_left = cl;
  804. h->crop_top = ct;
  805. h->crop_bottom = cb;
  806. return 0;
  807. }
  808. static int h264_slice_header_init(H264Context *h)
  809. {
  810. const SPS *sps = h->ps.sps;
  811. int i, ret;
  812. ff_set_sar(h->avctx, sps->sar);
  813. av_pix_fmt_get_chroma_sub_sample(h->avctx->pix_fmt,
  814. &h->chroma_x_shift, &h->chroma_y_shift);
  815. if (sps->timing_info_present_flag) {
  816. int64_t den = sps->time_scale;
  817. if (h->x264_build < 44U)
  818. den *= 2;
  819. av_reduce(&h->avctx->framerate.den, &h->avctx->framerate.num,
  820. sps->num_units_in_tick * h->avctx->ticks_per_frame, den, 1 << 30);
  821. }
  822. ff_h264_free_tables(h);
  823. h->first_field = 0;
  824. h->prev_interlaced_frame = 1;
  825. init_scan_tables(h);
  826. ret = ff_h264_alloc_tables(h);
  827. if (ret < 0) {
  828. av_log(h->avctx, AV_LOG_ERROR, "Could not allocate memory\n");
  829. goto fail;
  830. }
  831. if (sps->bit_depth_luma < 8 || sps->bit_depth_luma > 14 ||
  832. sps->bit_depth_luma == 11 || sps->bit_depth_luma == 13
  833. ) {
  834. av_log(h->avctx, AV_LOG_ERROR, "Unsupported bit depth %d\n",
  835. sps->bit_depth_luma);
  836. ret = AVERROR_INVALIDDATA;
  837. goto fail;
  838. }
  839. h->cur_bit_depth_luma =
  840. h->avctx->bits_per_raw_sample = sps->bit_depth_luma;
  841. h->cur_chroma_format_idc = sps->chroma_format_idc;
  842. h->pixel_shift = sps->bit_depth_luma > 8;
  843. h->chroma_format_idc = sps->chroma_format_idc;
  844. h->bit_depth_luma = sps->bit_depth_luma;
  845. ff_h264dsp_init(&h->h264dsp, sps->bit_depth_luma,
  846. sps->chroma_format_idc);
  847. ff_h264chroma_init(&h->h264chroma, sps->bit_depth_chroma);
  848. ff_h264qpel_init(&h->h264qpel, sps->bit_depth_luma);
  849. ff_h264_pred_init(&h->hpc, h->avctx->codec_id, sps->bit_depth_luma,
  850. sps->chroma_format_idc);
  851. ff_videodsp_init(&h->vdsp, sps->bit_depth_luma);
  852. if (!HAVE_THREADS || !(h->avctx->active_thread_type & FF_THREAD_SLICE)) {
  853. ret = ff_h264_slice_context_init(h, &h->slice_ctx[0]);
  854. if (ret < 0) {
  855. av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
  856. goto fail;
  857. }
  858. } else {
  859. for (i = 0; i < h->nb_slice_ctx; i++) {
  860. H264SliceContext *sl = &h->slice_ctx[i];
  861. sl->h264 = h;
  862. sl->intra4x4_pred_mode = h->intra4x4_pred_mode + i * 8 * 2 * h->mb_stride;
  863. sl->mvd_table[0] = h->mvd_table[0] + i * 8 * 2 * h->mb_stride;
  864. sl->mvd_table[1] = h->mvd_table[1] + i * 8 * 2 * h->mb_stride;
  865. if ((ret = ff_h264_slice_context_init(h, sl)) < 0) {
  866. av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
  867. goto fail;
  868. }
  869. }
  870. }
  871. h->context_initialized = 1;
  872. return 0;
  873. fail:
  874. ff_h264_free_tables(h);
  875. h->context_initialized = 0;
  876. return ret;
  877. }
  878. static enum AVPixelFormat non_j_pixfmt(enum AVPixelFormat a)
  879. {
  880. switch (a) {
  881. case AV_PIX_FMT_YUVJ420P: return AV_PIX_FMT_YUV420P;
  882. case AV_PIX_FMT_YUVJ422P: return AV_PIX_FMT_YUV422P;
  883. case AV_PIX_FMT_YUVJ444P: return AV_PIX_FMT_YUV444P;
  884. default:
  885. return a;
  886. }
  887. }
  888. static int h264_init_ps(H264Context *h, const H264SliceContext *sl, int first_slice)
  889. {
  890. const SPS *sps;
  891. int needs_reinit = 0, must_reinit, ret;
  892. if (first_slice) {
  893. av_buffer_unref(&h->ps.pps_ref);
  894. h->ps.pps = NULL;
  895. h->ps.pps_ref = av_buffer_ref(h->ps.pps_list[sl->pps_id]);
  896. if (!h->ps.pps_ref)
  897. return AVERROR(ENOMEM);
  898. h->ps.pps = (const PPS*)h->ps.pps_ref->data;
  899. }
  900. if (h->ps.sps != (const SPS*)h->ps.sps_list[h->ps.pps->sps_id]->data) {
  901. av_buffer_unref(&h->ps.sps_ref);
  902. h->ps.sps = NULL;
  903. h->ps.sps_ref = av_buffer_ref(h->ps.sps_list[h->ps.pps->sps_id]);
  904. if (!h->ps.sps_ref)
  905. return AVERROR(ENOMEM);
  906. h->ps.sps = (const SPS*)h->ps.sps_ref->data;
  907. if (h->mb_width != h->ps.sps->mb_width ||
  908. h->mb_height != h->ps.sps->mb_height ||
  909. h->cur_bit_depth_luma != h->ps.sps->bit_depth_luma ||
  910. h->cur_chroma_format_idc != h->ps.sps->chroma_format_idc
  911. )
  912. needs_reinit = 1;
  913. if (h->bit_depth_luma != h->ps.sps->bit_depth_luma ||
  914. h->chroma_format_idc != h->ps.sps->chroma_format_idc)
  915. needs_reinit = 1;
  916. }
  917. sps = h->ps.sps;
  918. must_reinit = (h->context_initialized &&
  919. ( 16*sps->mb_width != h->avctx->coded_width
  920. || 16*sps->mb_height != h->avctx->coded_height
  921. || h->cur_bit_depth_luma != sps->bit_depth_luma
  922. || h->cur_chroma_format_idc != sps->chroma_format_idc
  923. || h->mb_width != sps->mb_width
  924. || h->mb_height != sps->mb_height
  925. ));
  926. if (h->avctx->pix_fmt == AV_PIX_FMT_NONE
  927. || (non_j_pixfmt(h->avctx->pix_fmt) != non_j_pixfmt(get_pixel_format(h, 0))))
  928. must_reinit = 1;
  929. if (first_slice && av_cmp_q(sps->sar, h->avctx->sample_aspect_ratio))
  930. must_reinit = 1;
  931. if (!h->setup_finished) {
  932. h->avctx->profile = ff_h264_get_profile(sps);
  933. h->avctx->level = sps->level_idc;
  934. h->avctx->refs = sps->ref_frame_count;
  935. h->mb_width = sps->mb_width;
  936. h->mb_height = sps->mb_height;
  937. h->mb_num = h->mb_width * h->mb_height;
  938. h->mb_stride = h->mb_width + 1;
  939. h->b_stride = h->mb_width * 4;
  940. h->chroma_y_shift = sps->chroma_format_idc <= 1; // 400 uses yuv420p
  941. h->width = 16 * h->mb_width;
  942. h->height = 16 * h->mb_height;
  943. ret = init_dimensions(h);
  944. if (ret < 0)
  945. return ret;
  946. if (sps->video_signal_type_present_flag) {
  947. h->avctx->color_range = sps->full_range > 0 ? AVCOL_RANGE_JPEG
  948. : AVCOL_RANGE_MPEG;
  949. if (sps->colour_description_present_flag) {
  950. if (h->avctx->colorspace != sps->colorspace)
  951. needs_reinit = 1;
  952. h->avctx->color_primaries = sps->color_primaries;
  953. h->avctx->color_trc = sps->color_trc;
  954. h->avctx->colorspace = sps->colorspace;
  955. }
  956. }
  957. }
  958. if (!h->context_initialized || must_reinit || needs_reinit) {
  959. int flush_changes = h->context_initialized;
  960. h->context_initialized = 0;
  961. if (sl != h->slice_ctx) {
  962. av_log(h->avctx, AV_LOG_ERROR,
  963. "changing width %d -> %d / height %d -> %d on "
  964. "slice %d\n",
  965. h->width, h->avctx->coded_width,
  966. h->height, h->avctx->coded_height,
  967. h->current_slice + 1);
  968. return AVERROR_INVALIDDATA;
  969. }
  970. av_assert1(first_slice);
  971. if (flush_changes)
  972. ff_h264_flush_change(h);
  973. if ((ret = get_pixel_format(h, 1)) < 0)
  974. return ret;
  975. h->avctx->pix_fmt = ret;
  976. av_log(h->avctx, AV_LOG_VERBOSE, "Reinit context to %dx%d, "
  977. "pix_fmt: %s\n", h->width, h->height, av_get_pix_fmt_name(h->avctx->pix_fmt));
  978. if ((ret = h264_slice_header_init(h)) < 0) {
  979. av_log(h->avctx, AV_LOG_ERROR,
  980. "h264_slice_header_init() failed\n");
  981. return ret;
  982. }
  983. }
  984. return 0;
  985. }
  986. static int h264_export_frame_props(H264Context *h)
  987. {
  988. const SPS *sps = h->ps.sps;
  989. H264Picture *cur = h->cur_pic_ptr;
  990. cur->f->interlaced_frame = 0;
  991. cur->f->repeat_pict = 0;
  992. /* Signal interlacing information externally. */
  993. /* Prioritize picture timing SEI information over used
  994. * decoding process if it exists. */
  995. if (sps->pic_struct_present_flag && h->sei.picture_timing.present) {
  996. H264SEIPictureTiming *pt = &h->sei.picture_timing;
  997. switch (pt->pic_struct) {
  998. case H264_SEI_PIC_STRUCT_FRAME:
  999. break;
  1000. case H264_SEI_PIC_STRUCT_TOP_FIELD:
  1001. case H264_SEI_PIC_STRUCT_BOTTOM_FIELD:
  1002. cur->f->interlaced_frame = 1;
  1003. break;
  1004. case H264_SEI_PIC_STRUCT_TOP_BOTTOM:
  1005. case H264_SEI_PIC_STRUCT_BOTTOM_TOP:
  1006. if (FIELD_OR_MBAFF_PICTURE(h))
  1007. cur->f->interlaced_frame = 1;
  1008. else
  1009. // try to flag soft telecine progressive
  1010. cur->f->interlaced_frame = h->prev_interlaced_frame;
  1011. break;
  1012. case H264_SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
  1013. case H264_SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
  1014. /* Signal the possibility of telecined film externally
  1015. * (pic_struct 5,6). From these hints, let the applications
  1016. * decide if they apply deinterlacing. */
  1017. cur->f->repeat_pict = 1;
  1018. break;
  1019. case H264_SEI_PIC_STRUCT_FRAME_DOUBLING:
  1020. cur->f->repeat_pict = 2;
  1021. break;
  1022. case H264_SEI_PIC_STRUCT_FRAME_TRIPLING:
  1023. cur->f->repeat_pict = 4;
  1024. break;
  1025. }
  1026. if ((pt->ct_type & 3) &&
  1027. pt->pic_struct <= H264_SEI_PIC_STRUCT_BOTTOM_TOP)
  1028. cur->f->interlaced_frame = (pt->ct_type & (1 << 1)) != 0;
  1029. } else {
  1030. /* Derive interlacing flag from used decoding process. */
  1031. cur->f->interlaced_frame = FIELD_OR_MBAFF_PICTURE(h);
  1032. }
  1033. h->prev_interlaced_frame = cur->f->interlaced_frame;
  1034. if (cur->field_poc[0] != cur->field_poc[1]) {
  1035. /* Derive top_field_first from field pocs. */
  1036. cur->f->top_field_first = cur->field_poc[0] < cur->field_poc[1];
  1037. } else {
  1038. if (sps->pic_struct_present_flag && h->sei.picture_timing.present) {
  1039. /* Use picture timing SEI information. Even if it is a
  1040. * information of a past frame, better than nothing. */
  1041. if (h->sei.picture_timing.pic_struct == H264_SEI_PIC_STRUCT_TOP_BOTTOM ||
  1042. h->sei.picture_timing.pic_struct == H264_SEI_PIC_STRUCT_TOP_BOTTOM_TOP)
  1043. cur->f->top_field_first = 1;
  1044. else
  1045. cur->f->top_field_first = 0;
  1046. } else if (cur->f->interlaced_frame) {
  1047. /* Default to top field first when pic_struct_present_flag
  1048. * is not set but interlaced frame detected */
  1049. cur->f->top_field_first = 1;
  1050. } else {
  1051. /* Most likely progressive */
  1052. cur->f->top_field_first = 0;
  1053. }
  1054. }
  1055. if (h->sei.frame_packing.present &&
  1056. h->sei.frame_packing.arrangement_type <= 6 &&
  1057. h->sei.frame_packing.content_interpretation_type > 0 &&
  1058. h->sei.frame_packing.content_interpretation_type < 3) {
  1059. H264SEIFramePacking *fp = &h->sei.frame_packing;
  1060. AVStereo3D *stereo = av_stereo3d_create_side_data(cur->f);
  1061. if (stereo) {
  1062. switch (fp->arrangement_type) {
  1063. case H264_SEI_FPA_TYPE_CHECKERBOARD:
  1064. stereo->type = AV_STEREO3D_CHECKERBOARD;
  1065. break;
  1066. case H264_SEI_FPA_TYPE_INTERLEAVE_COLUMN:
  1067. stereo->type = AV_STEREO3D_COLUMNS;
  1068. break;
  1069. case H264_SEI_FPA_TYPE_INTERLEAVE_ROW:
  1070. stereo->type = AV_STEREO3D_LINES;
  1071. break;
  1072. case H264_SEI_FPA_TYPE_SIDE_BY_SIDE:
  1073. if (fp->quincunx_sampling_flag)
  1074. stereo->type = AV_STEREO3D_SIDEBYSIDE_QUINCUNX;
  1075. else
  1076. stereo->type = AV_STEREO3D_SIDEBYSIDE;
  1077. break;
  1078. case H264_SEI_FPA_TYPE_TOP_BOTTOM:
  1079. stereo->type = AV_STEREO3D_TOPBOTTOM;
  1080. break;
  1081. case H264_SEI_FPA_TYPE_INTERLEAVE_TEMPORAL:
  1082. stereo->type = AV_STEREO3D_FRAMESEQUENCE;
  1083. break;
  1084. case H264_SEI_FPA_TYPE_2D:
  1085. stereo->type = AV_STEREO3D_2D;
  1086. break;
  1087. }
  1088. if (fp->content_interpretation_type == 2)
  1089. stereo->flags = AV_STEREO3D_FLAG_INVERT;
  1090. if (fp->arrangement_type == H264_SEI_FPA_TYPE_INTERLEAVE_TEMPORAL) {
  1091. if (fp->current_frame_is_frame0_flag)
  1092. stereo->view = AV_STEREO3D_VIEW_LEFT;
  1093. else
  1094. stereo->view = AV_STEREO3D_VIEW_RIGHT;
  1095. }
  1096. }
  1097. }
  1098. if (h->sei.display_orientation.present &&
  1099. (h->sei.display_orientation.anticlockwise_rotation ||
  1100. h->sei.display_orientation.hflip ||
  1101. h->sei.display_orientation.vflip)) {
  1102. H264SEIDisplayOrientation *o = &h->sei.display_orientation;
  1103. double angle = o->anticlockwise_rotation * 360 / (double) (1 << 16);
  1104. AVFrameSideData *rotation = av_frame_new_side_data(cur->f,
  1105. AV_FRAME_DATA_DISPLAYMATRIX,
  1106. sizeof(int32_t) * 9);
  1107. if (rotation) {
  1108. av_display_rotation_set((int32_t *)rotation->data, angle);
  1109. av_display_matrix_flip((int32_t *)rotation->data,
  1110. o->hflip, o->vflip);
  1111. }
  1112. }
  1113. if (h->sei.afd.present) {
  1114. AVFrameSideData *sd = av_frame_new_side_data(cur->f, AV_FRAME_DATA_AFD,
  1115. sizeof(uint8_t));
  1116. if (sd) {
  1117. *sd->data = h->sei.afd.active_format_description;
  1118. h->sei.afd.present = 0;
  1119. }
  1120. }
  1121. if (h->sei.a53_caption.a53_caption) {
  1122. H264SEIA53Caption *a53 = &h->sei.a53_caption;
  1123. AVFrameSideData *sd = av_frame_new_side_data(cur->f,
  1124. AV_FRAME_DATA_A53_CC,
  1125. a53->a53_caption_size);
  1126. if (sd)
  1127. memcpy(sd->data, a53->a53_caption, a53->a53_caption_size);
  1128. av_freep(&a53->a53_caption);
  1129. a53->a53_caption_size = 0;
  1130. h->avctx->properties |= FF_CODEC_PROPERTY_CLOSED_CAPTIONS;
  1131. }
  1132. if (h->sei.alternative_transfer.present &&
  1133. av_color_transfer_name(h->sei.alternative_transfer.preferred_transfer_characteristics) &&
  1134. h->sei.alternative_transfer.preferred_transfer_characteristics != AVCOL_TRC_UNSPECIFIED) {
  1135. h->avctx->color_trc = cur->f->color_trc = h->sei.alternative_transfer.preferred_transfer_characteristics;
  1136. }
  1137. return 0;
  1138. }
  1139. static int h264_select_output_frame(H264Context *h)
  1140. {
  1141. const SPS *sps = h->ps.sps;
  1142. H264Picture *out = h->cur_pic_ptr;
  1143. H264Picture *cur = h->cur_pic_ptr;
  1144. int i, pics, out_of_order, out_idx;
  1145. cur->mmco_reset = h->mmco_reset;
  1146. h->mmco_reset = 0;
  1147. if (sps->bitstream_restriction_flag ||
  1148. h->avctx->strict_std_compliance >= FF_COMPLIANCE_STRICT) {
  1149. h->avctx->has_b_frames = FFMAX(h->avctx->has_b_frames, sps->num_reorder_frames);
  1150. }
  1151. for (i = 0; 1; i++) {
  1152. if(i == MAX_DELAYED_PIC_COUNT || cur->poc < h->last_pocs[i]){
  1153. if(i)
  1154. h->last_pocs[i-1] = cur->poc;
  1155. break;
  1156. } else if(i) {
  1157. h->last_pocs[i-1]= h->last_pocs[i];
  1158. }
  1159. }
  1160. out_of_order = MAX_DELAYED_PIC_COUNT - i;
  1161. if( cur->f->pict_type == AV_PICTURE_TYPE_B
  1162. || (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))
  1163. out_of_order = FFMAX(out_of_order, 1);
  1164. if (out_of_order == MAX_DELAYED_PIC_COUNT) {
  1165. av_log(h->avctx, AV_LOG_VERBOSE, "Invalid POC %d<%d\n", cur->poc, h->last_pocs[0]);
  1166. for (i = 1; i < MAX_DELAYED_PIC_COUNT; i++)
  1167. h->last_pocs[i] = INT_MIN;
  1168. h->last_pocs[0] = cur->poc;
  1169. cur->mmco_reset = 1;
  1170. } else if(h->avctx->has_b_frames < out_of_order && !sps->bitstream_restriction_flag){
  1171. int loglevel = h->avctx->frame_number > 1 ? AV_LOG_WARNING : AV_LOG_VERBOSE;
  1172. av_log(h->avctx, loglevel, "Increasing reorder buffer to %d\n", out_of_order);
  1173. h->avctx->has_b_frames = out_of_order;
  1174. }
  1175. pics = 0;
  1176. while (h->delayed_pic[pics])
  1177. pics++;
  1178. av_assert0(pics <= MAX_DELAYED_PIC_COUNT);
  1179. h->delayed_pic[pics++] = cur;
  1180. if (cur->reference == 0)
  1181. cur->reference = DELAYED_PIC_REF;
  1182. out = h->delayed_pic[0];
  1183. out_idx = 0;
  1184. for (i = 1; h->delayed_pic[i] &&
  1185. !h->delayed_pic[i]->f->key_frame &&
  1186. !h->delayed_pic[i]->mmco_reset;
  1187. i++)
  1188. if (h->delayed_pic[i]->poc < out->poc) {
  1189. out = h->delayed_pic[i];
  1190. out_idx = i;
  1191. }
  1192. if (h->avctx->has_b_frames == 0 &&
  1193. (h->delayed_pic[0]->f->key_frame || h->delayed_pic[0]->mmco_reset))
  1194. h->next_outputed_poc = INT_MIN;
  1195. out_of_order = out->poc < h->next_outputed_poc;
  1196. if (out_of_order || pics > h->avctx->has_b_frames) {
  1197. out->reference &= ~DELAYED_PIC_REF;
  1198. for (i = out_idx; h->delayed_pic[i]; i++)
  1199. h->delayed_pic[i] = h->delayed_pic[i + 1];
  1200. }
  1201. if (!out_of_order && pics > h->avctx->has_b_frames) {
  1202. h->next_output_pic = out;
  1203. if (out_idx == 0 && h->delayed_pic[0] && (h->delayed_pic[0]->f->key_frame || h->delayed_pic[0]->mmco_reset)) {
  1204. h->next_outputed_poc = INT_MIN;
  1205. } else
  1206. h->next_outputed_poc = out->poc;
  1207. if (out->recovered) {
  1208. // We have reached an recovery point and all frames after it in
  1209. // display order are "recovered".
  1210. h->frame_recovered |= FRAME_RECOVERED_SEI;
  1211. }
  1212. out->recovered |= !!(h->frame_recovered & FRAME_RECOVERED_SEI);
  1213. if (!out->recovered) {
  1214. if (!(h->avctx->flags & AV_CODEC_FLAG_OUTPUT_CORRUPT) &&
  1215. !(h->avctx->flags2 & AV_CODEC_FLAG2_SHOW_ALL)) {
  1216. h->next_output_pic = NULL;
  1217. } else {
  1218. out->f->flags |= AV_FRAME_FLAG_CORRUPT;
  1219. }
  1220. }
  1221. } else {
  1222. av_log(h->avctx, AV_LOG_DEBUG, "no picture %s\n", out_of_order ? "ooo" : "");
  1223. }
  1224. return 0;
  1225. }
  1226. /* This function is called right after decoding the slice header for a first
  1227. * slice in a field (or a frame). It decides whether we are decoding a new frame
  1228. * or a second field in a pair and does the necessary setup.
  1229. */
  1230. static int h264_field_start(H264Context *h, const H264SliceContext *sl,
  1231. const H2645NAL *nal, int first_slice)
  1232. {
  1233. int i;
  1234. const SPS *sps;
  1235. int last_pic_structure, last_pic_droppable, ret;
  1236. ret = h264_init_ps(h, sl, first_slice);
  1237. if (ret < 0)
  1238. return ret;
  1239. sps = h->ps.sps;
  1240. last_pic_droppable = h->droppable;
  1241. last_pic_structure = h->picture_structure;
  1242. h->droppable = (nal->ref_idc == 0);
  1243. h->picture_structure = sl->picture_structure;
  1244. h->poc.frame_num = sl->frame_num;
  1245. h->poc.poc_lsb = sl->poc_lsb;
  1246. h->poc.delta_poc_bottom = sl->delta_poc_bottom;
  1247. h->poc.delta_poc[0] = sl->delta_poc[0];
  1248. h->poc.delta_poc[1] = sl->delta_poc[1];
  1249. /* Shorten frame num gaps so we don't have to allocate reference
  1250. * frames just to throw them away */
  1251. if (h->poc.frame_num != h->poc.prev_frame_num) {
  1252. int unwrap_prev_frame_num = h->poc.prev_frame_num;
  1253. int max_frame_num = 1 << sps->log2_max_frame_num;
  1254. if (unwrap_prev_frame_num > h->poc.frame_num)
  1255. unwrap_prev_frame_num -= max_frame_num;
  1256. if ((h->poc.frame_num - unwrap_prev_frame_num) > sps->ref_frame_count) {
  1257. unwrap_prev_frame_num = (h->poc.frame_num - sps->ref_frame_count) - 1;
  1258. if (unwrap_prev_frame_num < 0)
  1259. unwrap_prev_frame_num += max_frame_num;
  1260. h->poc.prev_frame_num = unwrap_prev_frame_num;
  1261. }
  1262. }
  1263. /* See if we have a decoded first field looking for a pair...
  1264. * Here, we're using that to see if we should mark previously
  1265. * decode frames as "finished".
  1266. * We have to do that before the "dummy" in-between frame allocation,
  1267. * since that can modify h->cur_pic_ptr. */
  1268. if (h->first_field) {
  1269. int last_field = last_pic_structure == PICT_BOTTOM_FIELD;
  1270. av_assert0(h->cur_pic_ptr);
  1271. av_assert0(h->cur_pic_ptr->f->buf[0]);
  1272. assert(h->cur_pic_ptr->reference != DELAYED_PIC_REF);
  1273. /* Mark old field/frame as completed */
  1274. if (h->cur_pic_ptr->tf.owner[last_field] == h->avctx) {
  1275. ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, last_field);
  1276. }
  1277. /* figure out if we have a complementary field pair */
  1278. if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
  1279. /* Previous field is unmatched. Don't display it, but let it
  1280. * remain for reference if marked as such. */
  1281. if (last_pic_structure != PICT_FRAME) {
  1282. ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
  1283. last_pic_structure == PICT_TOP_FIELD);
  1284. }
  1285. } else {
  1286. if (h->cur_pic_ptr->frame_num != h->poc.frame_num) {
  1287. /* This and previous field were reference, but had
  1288. * different frame_nums. Consider this field first in
  1289. * pair. Throw away previous field except for reference
  1290. * purposes. */
  1291. if (last_pic_structure != PICT_FRAME) {
  1292. ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
  1293. last_pic_structure == PICT_TOP_FIELD);
  1294. }
  1295. } else {
  1296. /* Second field in complementary pair */
  1297. if (!((last_pic_structure == PICT_TOP_FIELD &&
  1298. h->picture_structure == PICT_BOTTOM_FIELD) ||
  1299. (last_pic_structure == PICT_BOTTOM_FIELD &&
  1300. h->picture_structure == PICT_TOP_FIELD))) {
  1301. av_log(h->avctx, AV_LOG_ERROR,
  1302. "Invalid field mode combination %d/%d\n",
  1303. last_pic_structure, h->picture_structure);
  1304. h->picture_structure = last_pic_structure;
  1305. h->droppable = last_pic_droppable;
  1306. return AVERROR_INVALIDDATA;
  1307. } else if (last_pic_droppable != h->droppable) {
  1308. avpriv_request_sample(h->avctx,
  1309. "Found reference and non-reference fields in the same frame, which");
  1310. h->picture_structure = last_pic_structure;
  1311. h->droppable = last_pic_droppable;
  1312. return AVERROR_PATCHWELCOME;
  1313. }
  1314. }
  1315. }
  1316. }
  1317. while (h->poc.frame_num != h->poc.prev_frame_num && !h->first_field &&
  1318. h->poc.frame_num != (h->poc.prev_frame_num + 1) % (1 << sps->log2_max_frame_num)) {
  1319. H264Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
  1320. av_log(h->avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n",
  1321. h->poc.frame_num, h->poc.prev_frame_num);
  1322. if (!sps->gaps_in_frame_num_allowed_flag)
  1323. for(i=0; i<FF_ARRAY_ELEMS(h->last_pocs); i++)
  1324. h->last_pocs[i] = INT_MIN;
  1325. ret = h264_frame_start(h);
  1326. if (ret < 0) {
  1327. h->first_field = 0;
  1328. return ret;
  1329. }
  1330. h->poc.prev_frame_num++;
  1331. h->poc.prev_frame_num %= 1 << sps->log2_max_frame_num;
  1332. h->cur_pic_ptr->frame_num = h->poc.prev_frame_num;
  1333. h->cur_pic_ptr->invalid_gap = !sps->gaps_in_frame_num_allowed_flag;
  1334. ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 0);
  1335. ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 1);
  1336. h->explicit_ref_marking = 0;
  1337. ret = ff_h264_execute_ref_pic_marking(h);
  1338. if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
  1339. return ret;
  1340. /* Error concealment: If a ref is missing, copy the previous ref
  1341. * in its place.
  1342. * FIXME: Avoiding a memcpy would be nice, but ref handling makes
  1343. * many assumptions about there being no actual duplicates.
  1344. * FIXME: This does not copy padding for out-of-frame motion
  1345. * vectors. Given we are concealing a lost frame, this probably
  1346. * is not noticeable by comparison, but it should be fixed. */
  1347. if (h->short_ref_count) {
  1348. if (prev &&
  1349. h->short_ref[0]->f->width == prev->f->width &&
  1350. h->short_ref[0]->f->height == prev->f->height &&
  1351. h->short_ref[0]->f->format == prev->f->format) {
  1352. ff_thread_await_progress(&prev->tf, INT_MAX, 0);
  1353. if (prev->field_picture)
  1354. ff_thread_await_progress(&prev->tf, INT_MAX, 1);
  1355. av_image_copy(h->short_ref[0]->f->data,
  1356. h->short_ref[0]->f->linesize,
  1357. (const uint8_t **)prev->f->data,
  1358. prev->f->linesize,
  1359. prev->f->format,
  1360. prev->f->width,
  1361. prev->f->height);
  1362. h->short_ref[0]->poc = prev->poc + 2;
  1363. }
  1364. h->short_ref[0]->frame_num = h->poc.prev_frame_num;
  1365. }
  1366. }
  1367. /* See if we have a decoded first field looking for a pair...
  1368. * We're using that to see whether to continue decoding in that
  1369. * frame, or to allocate a new one. */
  1370. if (h->first_field) {
  1371. av_assert0(h->cur_pic_ptr);
  1372. av_assert0(h->cur_pic_ptr->f->buf[0]);
  1373. assert(h->cur_pic_ptr->reference != DELAYED_PIC_REF);
  1374. /* figure out if we have a complementary field pair */
  1375. if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
  1376. /* Previous field is unmatched. Don't display it, but let it
  1377. * remain for reference if marked as such. */
  1378. h->missing_fields ++;
  1379. h->cur_pic_ptr = NULL;
  1380. h->first_field = FIELD_PICTURE(h);
  1381. } else {
  1382. h->missing_fields = 0;
  1383. if (h->cur_pic_ptr->frame_num != h->poc.frame_num) {
  1384. ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
  1385. h->picture_structure==PICT_BOTTOM_FIELD);
  1386. /* This and the previous field had different frame_nums.
  1387. * Consider this field first in pair. Throw away previous
  1388. * one except for reference purposes. */
  1389. h->first_field = 1;
  1390. h->cur_pic_ptr = NULL;
  1391. } else if (h->cur_pic_ptr->reference & DELAYED_PIC_REF) {
  1392. /* This frame was already output, we cannot draw into it
  1393. * anymore.
  1394. */
  1395. h->first_field = 1;
  1396. h->cur_pic_ptr = NULL;
  1397. } else {
  1398. /* Second field in complementary pair */
  1399. h->first_field = 0;
  1400. }
  1401. }
  1402. } else {
  1403. /* Frame or first field in a potentially complementary pair */
  1404. h->first_field = FIELD_PICTURE(h);
  1405. }
  1406. if (!FIELD_PICTURE(h) || h->first_field) {
  1407. if (h264_frame_start(h) < 0) {
  1408. h->first_field = 0;
  1409. return AVERROR_INVALIDDATA;
  1410. }
  1411. } else {
  1412. int field = h->picture_structure == PICT_BOTTOM_FIELD;
  1413. release_unused_pictures(h, 0);
  1414. h->cur_pic_ptr->tf.owner[field] = h->avctx;
  1415. }
  1416. /* Some macroblocks can be accessed before they're available in case
  1417. * of lost slices, MBAFF or threading. */
  1418. if (FIELD_PICTURE(h)) {
  1419. for(i = (h->picture_structure == PICT_BOTTOM_FIELD); i<h->mb_height; i++)
  1420. memset(h->slice_table + i*h->mb_stride, -1, (h->mb_stride - (i+1==h->mb_height)) * sizeof(*h->slice_table));
  1421. } else {
  1422. memset(h->slice_table, -1,
  1423. (h->mb_height * h->mb_stride - 1) * sizeof(*h->slice_table));
  1424. }
  1425. ret = ff_h264_init_poc(h->cur_pic_ptr->field_poc, &h->cur_pic_ptr->poc,
  1426. h->ps.sps, &h->poc, h->picture_structure, nal->ref_idc);
  1427. if (ret < 0)
  1428. return ret;
  1429. memcpy(h->mmco, sl->mmco, sl->nb_mmco * sizeof(*h->mmco));
  1430. h->nb_mmco = sl->nb_mmco;
  1431. h->explicit_ref_marking = sl->explicit_ref_marking;
  1432. h->picture_idr = nal->type == H264_NAL_IDR_SLICE;
  1433. if (h->sei.recovery_point.recovery_frame_cnt >= 0) {
  1434. const int sei_recovery_frame_cnt = h->sei.recovery_point.recovery_frame_cnt;
  1435. if (h->poc.frame_num != sei_recovery_frame_cnt || sl->slice_type_nos != AV_PICTURE_TYPE_I)
  1436. h->valid_recovery_point = 1;
  1437. if ( h->recovery_frame < 0
  1438. || av_mod_uintp2(h->recovery_frame - h->poc.frame_num, h->ps.sps->log2_max_frame_num) > sei_recovery_frame_cnt) {
  1439. h->recovery_frame = av_mod_uintp2(h->poc.frame_num + sei_recovery_frame_cnt, h->ps.sps->log2_max_frame_num);
  1440. if (!h->valid_recovery_point)
  1441. h->recovery_frame = h->poc.frame_num;
  1442. }
  1443. }
  1444. h->cur_pic_ptr->f->key_frame |= (nal->type == H264_NAL_IDR_SLICE);
  1445. if (nal->type == H264_NAL_IDR_SLICE ||
  1446. (h->recovery_frame == h->poc.frame_num && nal->ref_idc)) {
  1447. h->recovery_frame = -1;
  1448. h->cur_pic_ptr->recovered = 1;
  1449. }
  1450. // If we have an IDR, all frames after it in decoded order are
  1451. // "recovered".
  1452. if (nal->type == H264_NAL_IDR_SLICE)
  1453. h->frame_recovered |= FRAME_RECOVERED_IDR;
  1454. #if 1
  1455. h->cur_pic_ptr->recovered |= h->frame_recovered;
  1456. #else
  1457. h->cur_pic_ptr->recovered |= !!(h->frame_recovered & FRAME_RECOVERED_IDR);
  1458. #endif
  1459. /* Set the frame properties/side data. Only done for the second field in
  1460. * field coded frames, since some SEI information is present for each field
  1461. * and is merged by the SEI parsing code. */
  1462. if (!FIELD_PICTURE(h) || !h->first_field || h->missing_fields > 1) {
  1463. ret = h264_export_frame_props(h);
  1464. if (ret < 0)
  1465. return ret;
  1466. ret = h264_select_output_frame(h);
  1467. if (ret < 0)
  1468. return ret;
  1469. }
  1470. return 0;
  1471. }
  1472. static int h264_slice_header_parse(const H264Context *h, H264SliceContext *sl,
  1473. const H2645NAL *nal)
  1474. {
  1475. const SPS *sps;
  1476. const PPS *pps;
  1477. int ret;
  1478. unsigned int slice_type, tmp, i;
  1479. int field_pic_flag, bottom_field_flag;
  1480. int first_slice = sl == h->slice_ctx && !h->current_slice;
  1481. int picture_structure;
  1482. if (first_slice)
  1483. av_assert0(!h->setup_finished);
  1484. sl->first_mb_addr = get_ue_golomb_long(&sl->gb);
  1485. slice_type = get_ue_golomb_31(&sl->gb);
  1486. if (slice_type > 9) {
  1487. av_log(h->avctx, AV_LOG_ERROR,
  1488. "slice type %d too large at %d\n",
  1489. slice_type, sl->first_mb_addr);
  1490. return AVERROR_INVALIDDATA;
  1491. }
  1492. if (slice_type > 4) {
  1493. slice_type -= 5;
  1494. sl->slice_type_fixed = 1;
  1495. } else
  1496. sl->slice_type_fixed = 0;
  1497. slice_type = ff_h264_golomb_to_pict_type[slice_type];
  1498. sl->slice_type = slice_type;
  1499. sl->slice_type_nos = slice_type & 3;
  1500. if (nal->type == H264_NAL_IDR_SLICE &&
  1501. sl->slice_type_nos != AV_PICTURE_TYPE_I) {
  1502. av_log(h->avctx, AV_LOG_ERROR, "A non-intra slice in an IDR NAL unit.\n");
  1503. return AVERROR_INVALIDDATA;
  1504. }
  1505. sl->pps_id = get_ue_golomb(&sl->gb);
  1506. if (sl->pps_id >= MAX_PPS_COUNT) {
  1507. av_log(h->avctx, AV_LOG_ERROR, "pps_id %u out of range\n", sl->pps_id);
  1508. return AVERROR_INVALIDDATA;
  1509. }
  1510. if (!h->ps.pps_list[sl->pps_id]) {
  1511. av_log(h->avctx, AV_LOG_ERROR,
  1512. "non-existing PPS %u referenced\n",
  1513. sl->pps_id);
  1514. return AVERROR_INVALIDDATA;
  1515. }
  1516. pps = (const PPS*)h->ps.pps_list[sl->pps_id]->data;
  1517. if (!h->ps.sps_list[pps->sps_id]) {
  1518. av_log(h->avctx, AV_LOG_ERROR,
  1519. "non-existing SPS %u referenced\n", pps->sps_id);
  1520. return AVERROR_INVALIDDATA;
  1521. }
  1522. sps = (const SPS*)h->ps.sps_list[pps->sps_id]->data;
  1523. sl->frame_num = get_bits(&sl->gb, sps->log2_max_frame_num);
  1524. if (!first_slice) {
  1525. if (h->poc.frame_num != sl->frame_num) {
  1526. av_log(h->avctx, AV_LOG_ERROR, "Frame num change from %d to %d\n",
  1527. h->poc.frame_num, sl->frame_num);
  1528. return AVERROR_INVALIDDATA;
  1529. }
  1530. }
  1531. sl->mb_mbaff = 0;
  1532. if (sps->frame_mbs_only_flag) {
  1533. picture_structure = PICT_FRAME;
  1534. } else {
  1535. if (!sps->direct_8x8_inference_flag && slice_type == AV_PICTURE_TYPE_B) {
  1536. av_log(h->avctx, AV_LOG_ERROR, "This stream was generated by a broken encoder, invalid 8x8 inference\n");
  1537. return -1;
  1538. }
  1539. field_pic_flag = get_bits1(&sl->gb);
  1540. if (field_pic_flag) {
  1541. bottom_field_flag = get_bits1(&sl->gb);
  1542. picture_structure = PICT_TOP_FIELD + bottom_field_flag;
  1543. } else {
  1544. picture_structure = PICT_FRAME;
  1545. }
  1546. }
  1547. sl->picture_structure = picture_structure;
  1548. sl->mb_field_decoding_flag = picture_structure != PICT_FRAME;
  1549. if (picture_structure == PICT_FRAME) {
  1550. sl->curr_pic_num = sl->frame_num;
  1551. sl->max_pic_num = 1 << sps->log2_max_frame_num;
  1552. } else {
  1553. sl->curr_pic_num = 2 * sl->frame_num + 1;
  1554. sl->max_pic_num = 1 << (sps->log2_max_frame_num + 1);
  1555. }
  1556. if (nal->type == H264_NAL_IDR_SLICE)
  1557. get_ue_golomb_long(&sl->gb); /* idr_pic_id */
  1558. if (sps->poc_type == 0) {
  1559. sl->poc_lsb = get_bits(&sl->gb, sps->log2_max_poc_lsb);
  1560. if (pps->pic_order_present == 1 && picture_structure == PICT_FRAME)
  1561. sl->delta_poc_bottom = get_se_golomb(&sl->gb);
  1562. }
  1563. if (sps->poc_type == 1 && !sps->delta_pic_order_always_zero_flag) {
  1564. sl->delta_poc[0] = get_se_golomb(&sl->gb);
  1565. if (pps->pic_order_present == 1 && picture_structure == PICT_FRAME)
  1566. sl->delta_poc[1] = get_se_golomb(&sl->gb);
  1567. }
  1568. sl->redundant_pic_count = 0;
  1569. if (pps->redundant_pic_cnt_present)
  1570. sl->redundant_pic_count = get_ue_golomb(&sl->gb);
  1571. if (sl->slice_type_nos == AV_PICTURE_TYPE_B)
  1572. sl->direct_spatial_mv_pred = get_bits1(&sl->gb);
  1573. ret = ff_h264_parse_ref_count(&sl->list_count, sl->ref_count,
  1574. &sl->gb, pps, sl->slice_type_nos,
  1575. picture_structure, h->avctx);
  1576. if (ret < 0)
  1577. return ret;
  1578. if (sl->slice_type_nos != AV_PICTURE_TYPE_I) {
  1579. ret = ff_h264_decode_ref_pic_list_reordering(sl, h->avctx);
  1580. if (ret < 0) {
  1581. sl->ref_count[1] = sl->ref_count[0] = 0;
  1582. return ret;
  1583. }
  1584. }
  1585. sl->pwt.use_weight = 0;
  1586. for (i = 0; i < 2; i++) {
  1587. sl->pwt.luma_weight_flag[i] = 0;
  1588. sl->pwt.chroma_weight_flag[i] = 0;
  1589. }
  1590. if ((pps->weighted_pred && sl->slice_type_nos == AV_PICTURE_TYPE_P) ||
  1591. (pps->weighted_bipred_idc == 1 &&
  1592. sl->slice_type_nos == AV_PICTURE_TYPE_B)) {
  1593. ret = ff_h264_pred_weight_table(&sl->gb, sps, sl->ref_count,
  1594. sl->slice_type_nos, &sl->pwt,
  1595. picture_structure, h->avctx);
  1596. if (ret < 0)
  1597. return ret;
  1598. }
  1599. sl->explicit_ref_marking = 0;
  1600. if (nal->ref_idc) {
  1601. ret = ff_h264_decode_ref_pic_marking(sl, &sl->gb, nal, h->avctx);
  1602. if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
  1603. return AVERROR_INVALIDDATA;
  1604. }
  1605. if (sl->slice_type_nos != AV_PICTURE_TYPE_I && pps->cabac) {
  1606. tmp = get_ue_golomb_31(&sl->gb);
  1607. if (tmp > 2) {
  1608. av_log(h->avctx, AV_LOG_ERROR, "cabac_init_idc %u overflow\n", tmp);
  1609. return AVERROR_INVALIDDATA;
  1610. }
  1611. sl->cabac_init_idc = tmp;
  1612. }
  1613. sl->last_qscale_diff = 0;
  1614. tmp = pps->init_qp + (unsigned)get_se_golomb(&sl->gb);
  1615. if (tmp > 51 + 6 * (sps->bit_depth_luma - 8)) {
  1616. av_log(h->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
  1617. return AVERROR_INVALIDDATA;
  1618. }
  1619. sl->qscale = tmp;
  1620. sl->chroma_qp[0] = get_chroma_qp(pps, 0, sl->qscale);
  1621. sl->chroma_qp[1] = get_chroma_qp(pps, 1, sl->qscale);
  1622. // FIXME qscale / qp ... stuff
  1623. if (sl->slice_type == AV_PICTURE_TYPE_SP)
  1624. get_bits1(&sl->gb); /* sp_for_switch_flag */
  1625. if (sl->slice_type == AV_PICTURE_TYPE_SP ||
  1626. sl->slice_type == AV_PICTURE_TYPE_SI)
  1627. get_se_golomb(&sl->gb); /* slice_qs_delta */
  1628. sl->deblocking_filter = 1;
  1629. sl->slice_alpha_c0_offset = 0;
  1630. sl->slice_beta_offset = 0;
  1631. if (pps->deblocking_filter_parameters_present) {
  1632. tmp = get_ue_golomb_31(&sl->gb);
  1633. if (tmp > 2) {
  1634. av_log(h->avctx, AV_LOG_ERROR,
  1635. "deblocking_filter_idc %u out of range\n", tmp);
  1636. return AVERROR_INVALIDDATA;
  1637. }
  1638. sl->deblocking_filter = tmp;
  1639. if (sl->deblocking_filter < 2)
  1640. sl->deblocking_filter ^= 1; // 1<->0
  1641. if (sl->deblocking_filter) {
  1642. int slice_alpha_c0_offset_div2 = get_se_golomb(&sl->gb);
  1643. int slice_beta_offset_div2 = get_se_golomb(&sl->gb);
  1644. if (slice_alpha_c0_offset_div2 > 6 ||
  1645. slice_alpha_c0_offset_div2 < -6 ||
  1646. slice_beta_offset_div2 > 6 ||
  1647. slice_beta_offset_div2 < -6) {
  1648. av_log(h->avctx, AV_LOG_ERROR,
  1649. "deblocking filter parameters %d %d out of range\n",
  1650. slice_alpha_c0_offset_div2, slice_beta_offset_div2);
  1651. return AVERROR_INVALIDDATA;
  1652. }
  1653. sl->slice_alpha_c0_offset = slice_alpha_c0_offset_div2 * 2;
  1654. sl->slice_beta_offset = slice_beta_offset_div2 * 2;
  1655. }
  1656. }
  1657. return 0;
  1658. }
  1659. /* do all the per-slice initialization needed before we can start decoding the
  1660. * actual MBs */
  1661. static int h264_slice_init(H264Context *h, H264SliceContext *sl,
  1662. const H2645NAL *nal)
  1663. {
  1664. int i, j, ret = 0;
  1665. if (h->picture_idr && nal->type != H264_NAL_IDR_SLICE) {
  1666. av_log(h->avctx, AV_LOG_ERROR, "Invalid mix of IDR and non-IDR slices\n");
  1667. return AVERROR_INVALIDDATA;
  1668. }
  1669. av_assert1(h->mb_num == h->mb_width * h->mb_height);
  1670. if (sl->first_mb_addr << FIELD_OR_MBAFF_PICTURE(h) >= h->mb_num ||
  1671. sl->first_mb_addr >= h->mb_num) {
  1672. av_log(h->avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
  1673. return AVERROR_INVALIDDATA;
  1674. }
  1675. sl->resync_mb_x = sl->mb_x = sl->first_mb_addr % h->mb_width;
  1676. sl->resync_mb_y = sl->mb_y = (sl->first_mb_addr / h->mb_width) <<
  1677. FIELD_OR_MBAFF_PICTURE(h);
  1678. if (h->picture_structure == PICT_BOTTOM_FIELD)
  1679. sl->resync_mb_y = sl->mb_y = sl->mb_y + 1;
  1680. av_assert1(sl->mb_y < h->mb_height);
  1681. ret = ff_h264_build_ref_list(h, sl);
  1682. if (ret < 0)
  1683. return ret;
  1684. if (h->ps.pps->weighted_bipred_idc == 2 &&
  1685. sl->slice_type_nos == AV_PICTURE_TYPE_B) {
  1686. implicit_weight_table(h, sl, -1);
  1687. if (FRAME_MBAFF(h)) {
  1688. implicit_weight_table(h, sl, 0);
  1689. implicit_weight_table(h, sl, 1);
  1690. }
  1691. }
  1692. if (sl->slice_type_nos == AV_PICTURE_TYPE_B && !sl->direct_spatial_mv_pred)
  1693. ff_h264_direct_dist_scale_factor(h, sl);
  1694. if (!h->setup_finished)
  1695. ff_h264_direct_ref_list_init(h, sl);
  1696. if (h->avctx->skip_loop_filter >= AVDISCARD_ALL ||
  1697. (h->avctx->skip_loop_filter >= AVDISCARD_NONKEY &&
  1698. h->nal_unit_type != H264_NAL_IDR_SLICE) ||
  1699. (h->avctx->skip_loop_filter >= AVDISCARD_NONINTRA &&
  1700. sl->slice_type_nos != AV_PICTURE_TYPE_I) ||
  1701. (h->avctx->skip_loop_filter >= AVDISCARD_BIDIR &&
  1702. sl->slice_type_nos == AV_PICTURE_TYPE_B) ||
  1703. (h->avctx->skip_loop_filter >= AVDISCARD_NONREF &&
  1704. nal->ref_idc == 0))
  1705. sl->deblocking_filter = 0;
  1706. if (sl->deblocking_filter == 1 && h->nb_slice_ctx > 1) {
  1707. if (h->avctx->flags2 & AV_CODEC_FLAG2_FAST) {
  1708. /* Cheat slightly for speed:
  1709. * Do not bother to deblock across slices. */
  1710. sl->deblocking_filter = 2;
  1711. } else {
  1712. h->postpone_filter = 1;
  1713. }
  1714. }
  1715. sl->qp_thresh = 15 -
  1716. FFMIN(sl->slice_alpha_c0_offset, sl->slice_beta_offset) -
  1717. FFMAX3(0,
  1718. h->ps.pps->chroma_qp_index_offset[0],
  1719. h->ps.pps->chroma_qp_index_offset[1]) +
  1720. 6 * (h->ps.sps->bit_depth_luma - 8);
  1721. sl->slice_num = ++h->current_slice;
  1722. if (sl->slice_num)
  1723. h->slice_row[(sl->slice_num-1)&(MAX_SLICES-1)]= sl->resync_mb_y;
  1724. if ( h->slice_row[sl->slice_num&(MAX_SLICES-1)] + 3 >= sl->resync_mb_y
  1725. && h->slice_row[sl->slice_num&(MAX_SLICES-1)] <= sl->resync_mb_y
  1726. && sl->slice_num >= MAX_SLICES) {
  1727. //in case of ASO this check needs to be updated depending on how we decide to assign slice numbers in this case
  1728. 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);
  1729. }
  1730. for (j = 0; j < 2; j++) {
  1731. int id_list[16];
  1732. int *ref2frm = h->ref2frm[sl->slice_num & (MAX_SLICES - 1)][j];
  1733. for (i = 0; i < 16; i++) {
  1734. id_list[i] = 60;
  1735. if (j < sl->list_count && i < sl->ref_count[j] &&
  1736. sl->ref_list[j][i].parent->f->buf[0]) {
  1737. int k;
  1738. AVBuffer *buf = sl->ref_list[j][i].parent->f->buf[0]->buffer;
  1739. for (k = 0; k < h->short_ref_count; k++)
  1740. if (h->short_ref[k]->f->buf[0]->buffer == buf) {
  1741. id_list[i] = k;
  1742. break;
  1743. }
  1744. for (k = 0; k < h->long_ref_count; k++)
  1745. if (h->long_ref[k] && h->long_ref[k]->f->buf[0]->buffer == buf) {
  1746. id_list[i] = h->short_ref_count + k;
  1747. break;
  1748. }
  1749. }
  1750. }
  1751. ref2frm[0] =
  1752. ref2frm[1] = -1;
  1753. for (i = 0; i < 16; i++)
  1754. ref2frm[i + 2] = 4 * id_list[i] + (sl->ref_list[j][i].reference & 3);
  1755. ref2frm[18 + 0] =
  1756. ref2frm[18 + 1] = -1;
  1757. for (i = 16; i < 48; i++)
  1758. ref2frm[i + 4] = 4 * id_list[(i - 16) >> 1] +
  1759. (sl->ref_list[j][i].reference & 3);
  1760. }
  1761. if (h->avctx->debug & FF_DEBUG_PICT_INFO) {
  1762. av_log(h->avctx, AV_LOG_DEBUG,
  1763. "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",
  1764. sl->slice_num,
  1765. (h->picture_structure == PICT_FRAME ? "F" : h->picture_structure == PICT_TOP_FIELD ? "T" : "B"),
  1766. sl->mb_y * h->mb_width + sl->mb_x,
  1767. av_get_picture_type_char(sl->slice_type),
  1768. sl->slice_type_fixed ? " fix" : "",
  1769. nal->type == H264_NAL_IDR_SLICE ? " IDR" : "",
  1770. h->poc.frame_num,
  1771. h->cur_pic_ptr->field_poc[0],
  1772. h->cur_pic_ptr->field_poc[1],
  1773. sl->ref_count[0], sl->ref_count[1],
  1774. sl->qscale,
  1775. sl->deblocking_filter,
  1776. sl->slice_alpha_c0_offset, sl->slice_beta_offset,
  1777. sl->pwt.use_weight,
  1778. sl->pwt.use_weight == 1 && sl->pwt.use_weight_chroma ? "c" : "",
  1779. sl->slice_type == AV_PICTURE_TYPE_B ? (sl->direct_spatial_mv_pred ? "SPAT" : "TEMP") : "");
  1780. }
  1781. return 0;
  1782. }
  1783. int ff_h264_queue_decode_slice(H264Context *h, const H2645NAL *nal)
  1784. {
  1785. H264SliceContext *sl = h->slice_ctx + h->nb_slice_ctx_queued;
  1786. int first_slice = sl == h->slice_ctx && !h->current_slice;
  1787. int ret;
  1788. sl->gb = nal->gb;
  1789. ret = h264_slice_header_parse(h, sl, nal);
  1790. if (ret < 0)
  1791. return ret;
  1792. // discard redundant pictures
  1793. if (sl->redundant_pic_count > 0) {
  1794. sl->ref_count[0] = sl->ref_count[1] = 0;
  1795. return 0;
  1796. }
  1797. if (sl->first_mb_addr == 0 || !h->current_slice) {
  1798. if (h->setup_finished) {
  1799. av_log(h->avctx, AV_LOG_ERROR, "Too many fields\n");
  1800. return AVERROR_INVALIDDATA;
  1801. }
  1802. }
  1803. if (sl->first_mb_addr == 0) { // FIXME better field boundary detection
  1804. if (h->current_slice) {
  1805. // this slice starts a new field
  1806. // first decode any pending queued slices
  1807. if (h->nb_slice_ctx_queued) {
  1808. H264SliceContext tmp_ctx;
  1809. ret = ff_h264_execute_decode_slices(h);
  1810. if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
  1811. return ret;
  1812. memcpy(&tmp_ctx, h->slice_ctx, sizeof(tmp_ctx));
  1813. memcpy(h->slice_ctx, sl, sizeof(tmp_ctx));
  1814. memcpy(sl, &tmp_ctx, sizeof(tmp_ctx));
  1815. sl = h->slice_ctx;
  1816. }
  1817. if (h->cur_pic_ptr && FIELD_PICTURE(h) && h->first_field) {
  1818. ret = ff_h264_field_end(h, h->slice_ctx, 1);
  1819. if (ret < 0)
  1820. return ret;
  1821. } else if (h->cur_pic_ptr && !FIELD_PICTURE(h) && !h->first_field && h->nal_unit_type == H264_NAL_IDR_SLICE) {
  1822. av_log(h, AV_LOG_WARNING, "Broken frame packetizing\n");
  1823. ret = ff_h264_field_end(h, h->slice_ctx, 1);
  1824. ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 0);
  1825. ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 1);
  1826. h->cur_pic_ptr = NULL;
  1827. if (ret < 0)
  1828. return ret;
  1829. } else
  1830. return AVERROR_INVALIDDATA;
  1831. }
  1832. if (!h->first_field) {
  1833. if (h->cur_pic_ptr && !h->droppable) {
  1834. ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
  1835. h->picture_structure == PICT_BOTTOM_FIELD);
  1836. }
  1837. h->cur_pic_ptr = NULL;
  1838. }
  1839. }
  1840. if (!h->current_slice)
  1841. av_assert0(sl == h->slice_ctx);
  1842. if (h->current_slice == 0 && !h->first_field) {
  1843. if (
  1844. (h->avctx->skip_frame >= AVDISCARD_NONREF && !h->nal_ref_idc) ||
  1845. (h->avctx->skip_frame >= AVDISCARD_BIDIR && sl->slice_type_nos == AV_PICTURE_TYPE_B) ||
  1846. (h->avctx->skip_frame >= AVDISCARD_NONINTRA && sl->slice_type_nos != AV_PICTURE_TYPE_I) ||
  1847. (h->avctx->skip_frame >= AVDISCARD_NONKEY && h->nal_unit_type != H264_NAL_IDR_SLICE && h->sei.recovery_point.recovery_frame_cnt < 0) ||
  1848. h->avctx->skip_frame >= AVDISCARD_ALL) {
  1849. return 0;
  1850. }
  1851. }
  1852. if (!first_slice) {
  1853. const PPS *pps = (const PPS*)h->ps.pps_list[sl->pps_id]->data;
  1854. if (h->ps.pps->sps_id != pps->sps_id ||
  1855. h->ps.pps->transform_8x8_mode != pps->transform_8x8_mode /*||
  1856. (h->setup_finished && h->ps.pps != pps)*/) {
  1857. av_log(h->avctx, AV_LOG_ERROR, "PPS changed between slices\n");
  1858. return AVERROR_INVALIDDATA;
  1859. }
  1860. if (h->ps.sps != (const SPS*)h->ps.sps_list[h->ps.pps->sps_id]->data) {
  1861. av_log(h->avctx, AV_LOG_ERROR,
  1862. "SPS changed in the middle of the frame\n");
  1863. return AVERROR_INVALIDDATA;
  1864. }
  1865. }
  1866. if (h->current_slice == 0) {
  1867. ret = h264_field_start(h, sl, nal, first_slice);
  1868. if (ret < 0)
  1869. return ret;
  1870. } else {
  1871. if (h->picture_structure != sl->picture_structure ||
  1872. h->droppable != (nal->ref_idc == 0)) {
  1873. av_log(h->avctx, AV_LOG_ERROR,
  1874. "Changing field mode (%d -> %d) between slices is not allowed\n",
  1875. h->picture_structure, sl->picture_structure);
  1876. return AVERROR_INVALIDDATA;
  1877. } else if (!h->cur_pic_ptr) {
  1878. av_log(h->avctx, AV_LOG_ERROR,
  1879. "unset cur_pic_ptr on slice %d\n",
  1880. h->current_slice + 1);
  1881. return AVERROR_INVALIDDATA;
  1882. }
  1883. }
  1884. ret = h264_slice_init(h, sl, nal);
  1885. if (ret < 0)
  1886. return ret;
  1887. h->nb_slice_ctx_queued++;
  1888. return 0;
  1889. }
  1890. int ff_h264_get_slice_type(const H264SliceContext *sl)
  1891. {
  1892. switch (sl->slice_type) {
  1893. case AV_PICTURE_TYPE_P:
  1894. return 0;
  1895. case AV_PICTURE_TYPE_B:
  1896. return 1;
  1897. case AV_PICTURE_TYPE_I:
  1898. return 2;
  1899. case AV_PICTURE_TYPE_SP:
  1900. return 3;
  1901. case AV_PICTURE_TYPE_SI:
  1902. return 4;
  1903. default:
  1904. return AVERROR_INVALIDDATA;
  1905. }
  1906. }
  1907. static av_always_inline void fill_filter_caches_inter(const H264Context *h,
  1908. H264SliceContext *sl,
  1909. int mb_type, int top_xy,
  1910. int left_xy[LEFT_MBS],
  1911. int top_type,
  1912. int left_type[LEFT_MBS],
  1913. int mb_xy, int list)
  1914. {
  1915. int b_stride = h->b_stride;
  1916. int16_t(*mv_dst)[2] = &sl->mv_cache[list][scan8[0]];
  1917. int8_t *ref_cache = &sl->ref_cache[list][scan8[0]];
  1918. if (IS_INTER(mb_type) || IS_DIRECT(mb_type)) {
  1919. if (USES_LIST(top_type, list)) {
  1920. const int b_xy = h->mb2b_xy[top_xy] + 3 * b_stride;
  1921. const int b8_xy = 4 * top_xy + 2;
  1922. const int *ref2frm = &h->ref2frm[h->slice_table[top_xy] & (MAX_SLICES - 1)][list][(MB_MBAFF(sl) ? 20 : 2)];
  1923. AV_COPY128(mv_dst - 1 * 8, h->cur_pic.motion_val[list][b_xy + 0]);
  1924. ref_cache[0 - 1 * 8] =
  1925. ref_cache[1 - 1 * 8] = ref2frm[h->cur_pic.ref_index[list][b8_xy + 0]];
  1926. ref_cache[2 - 1 * 8] =
  1927. ref_cache[3 - 1 * 8] = ref2frm[h->cur_pic.ref_index[list][b8_xy + 1]];
  1928. } else {
  1929. AV_ZERO128(mv_dst - 1 * 8);
  1930. AV_WN32A(&ref_cache[0 - 1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
  1931. }
  1932. if (!IS_INTERLACED(mb_type ^ left_type[LTOP])) {
  1933. if (USES_LIST(left_type[LTOP], list)) {
  1934. const int b_xy = h->mb2b_xy[left_xy[LTOP]] + 3;
  1935. const int b8_xy = 4 * left_xy[LTOP] + 1;
  1936. const int *ref2frm = &h->ref2frm[h->slice_table[left_xy[LTOP]] & (MAX_SLICES - 1)][list][(MB_MBAFF(sl) ? 20 : 2)];
  1937. AV_COPY32(mv_dst - 1 + 0, h->cur_pic.motion_val[list][b_xy + b_stride * 0]);
  1938. AV_COPY32(mv_dst - 1 + 8, h->cur_pic.motion_val[list][b_xy + b_stride * 1]);
  1939. AV_COPY32(mv_dst - 1 + 16, h->cur_pic.motion_val[list][b_xy + b_stride * 2]);
  1940. AV_COPY32(mv_dst - 1 + 24, h->cur_pic.motion_val[list][b_xy + b_stride * 3]);
  1941. ref_cache[-1 + 0] =
  1942. ref_cache[-1 + 8] = ref2frm[h->cur_pic.ref_index[list][b8_xy + 2 * 0]];
  1943. ref_cache[-1 + 16] =
  1944. ref_cache[-1 + 24] = ref2frm[h->cur_pic.ref_index[list][b8_xy + 2 * 1]];
  1945. } else {
  1946. AV_ZERO32(mv_dst - 1 + 0);
  1947. AV_ZERO32(mv_dst - 1 + 8);
  1948. AV_ZERO32(mv_dst - 1 + 16);
  1949. AV_ZERO32(mv_dst - 1 + 24);
  1950. ref_cache[-1 + 0] =
  1951. ref_cache[-1 + 8] =
  1952. ref_cache[-1 + 16] =
  1953. ref_cache[-1 + 24] = LIST_NOT_USED;
  1954. }
  1955. }
  1956. }
  1957. if (!USES_LIST(mb_type, list)) {
  1958. fill_rectangle(mv_dst, 4, 4, 8, pack16to32(0, 0), 4);
  1959. AV_WN32A(&ref_cache[0 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
  1960. AV_WN32A(&ref_cache[1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
  1961. AV_WN32A(&ref_cache[2 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
  1962. AV_WN32A(&ref_cache[3 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
  1963. return;
  1964. }
  1965. {
  1966. int8_t *ref = &h->cur_pic.ref_index[list][4 * mb_xy];
  1967. const int *ref2frm = &h->ref2frm[sl->slice_num & (MAX_SLICES - 1)][list][(MB_MBAFF(sl) ? 20 : 2)];
  1968. uint32_t ref01 = (pack16to32(ref2frm[ref[0]], ref2frm[ref[1]]) & 0x00FF00FF) * 0x0101;
  1969. uint32_t ref23 = (pack16to32(ref2frm[ref[2]], ref2frm[ref[3]]) & 0x00FF00FF) * 0x0101;
  1970. AV_WN32A(&ref_cache[0 * 8], ref01);
  1971. AV_WN32A(&ref_cache[1 * 8], ref01);
  1972. AV_WN32A(&ref_cache[2 * 8], ref23);
  1973. AV_WN32A(&ref_cache[3 * 8], ref23);
  1974. }
  1975. {
  1976. int16_t(*mv_src)[2] = &h->cur_pic.motion_val[list][4 * sl->mb_x + 4 * sl->mb_y * b_stride];
  1977. AV_COPY128(mv_dst + 8 * 0, mv_src + 0 * b_stride);
  1978. AV_COPY128(mv_dst + 8 * 1, mv_src + 1 * b_stride);
  1979. AV_COPY128(mv_dst + 8 * 2, mv_src + 2 * b_stride);
  1980. AV_COPY128(mv_dst + 8 * 3, mv_src + 3 * b_stride);
  1981. }
  1982. }
  1983. /**
  1984. * @return non zero if the loop filter can be skipped
  1985. */
  1986. static int fill_filter_caches(const H264Context *h, H264SliceContext *sl, int mb_type)
  1987. {
  1988. const int mb_xy = sl->mb_xy;
  1989. int top_xy, left_xy[LEFT_MBS];
  1990. int top_type, left_type[LEFT_MBS];
  1991. uint8_t *nnz;
  1992. uint8_t *nnz_cache;
  1993. top_xy = mb_xy - (h->mb_stride << MB_FIELD(sl));
  1994. left_xy[LBOT] = left_xy[LTOP] = mb_xy - 1;
  1995. if (FRAME_MBAFF(h)) {
  1996. const int left_mb_field_flag = IS_INTERLACED(h->cur_pic.mb_type[mb_xy - 1]);
  1997. const int curr_mb_field_flag = IS_INTERLACED(mb_type);
  1998. if (sl->mb_y & 1) {
  1999. if (left_mb_field_flag != curr_mb_field_flag)
  2000. left_xy[LTOP] -= h->mb_stride;
  2001. } else {
  2002. if (curr_mb_field_flag)
  2003. top_xy += h->mb_stride &
  2004. (((h->cur_pic.mb_type[top_xy] >> 7) & 1) - 1);
  2005. if (left_mb_field_flag != curr_mb_field_flag)
  2006. left_xy[LBOT] += h->mb_stride;
  2007. }
  2008. }
  2009. sl->top_mb_xy = top_xy;
  2010. sl->left_mb_xy[LTOP] = left_xy[LTOP];
  2011. sl->left_mb_xy[LBOT] = left_xy[LBOT];
  2012. {
  2013. /* For sufficiently low qp, filtering wouldn't do anything.
  2014. * This is a conservative estimate: could also check beta_offset
  2015. * and more accurate chroma_qp. */
  2016. int qp_thresh = sl->qp_thresh; // FIXME strictly we should store qp_thresh for each mb of a slice
  2017. int qp = h->cur_pic.qscale_table[mb_xy];
  2018. if (qp <= qp_thresh &&
  2019. (left_xy[LTOP] < 0 ||
  2020. ((qp + h->cur_pic.qscale_table[left_xy[LTOP]] + 1) >> 1) <= qp_thresh) &&
  2021. (top_xy < 0 ||
  2022. ((qp + h->cur_pic.qscale_table[top_xy] + 1) >> 1) <= qp_thresh)) {
  2023. if (!FRAME_MBAFF(h))
  2024. return 1;
  2025. if ((left_xy[LTOP] < 0 ||
  2026. ((qp + h->cur_pic.qscale_table[left_xy[LBOT]] + 1) >> 1) <= qp_thresh) &&
  2027. (top_xy < h->mb_stride ||
  2028. ((qp + h->cur_pic.qscale_table[top_xy - h->mb_stride] + 1) >> 1) <= qp_thresh))
  2029. return 1;
  2030. }
  2031. }
  2032. top_type = h->cur_pic.mb_type[top_xy];
  2033. left_type[LTOP] = h->cur_pic.mb_type[left_xy[LTOP]];
  2034. left_type[LBOT] = h->cur_pic.mb_type[left_xy[LBOT]];
  2035. if (sl->deblocking_filter == 2) {
  2036. if (h->slice_table[top_xy] != sl->slice_num)
  2037. top_type = 0;
  2038. if (h->slice_table[left_xy[LBOT]] != sl->slice_num)
  2039. left_type[LTOP] = left_type[LBOT] = 0;
  2040. } else {
  2041. if (h->slice_table[top_xy] == 0xFFFF)
  2042. top_type = 0;
  2043. if (h->slice_table[left_xy[LBOT]] == 0xFFFF)
  2044. left_type[LTOP] = left_type[LBOT] = 0;
  2045. }
  2046. sl->top_type = top_type;
  2047. sl->left_type[LTOP] = left_type[LTOP];
  2048. sl->left_type[LBOT] = left_type[LBOT];
  2049. if (IS_INTRA(mb_type))
  2050. return 0;
  2051. fill_filter_caches_inter(h, sl, mb_type, top_xy, left_xy,
  2052. top_type, left_type, mb_xy, 0);
  2053. if (sl->list_count == 2)
  2054. fill_filter_caches_inter(h, sl, mb_type, top_xy, left_xy,
  2055. top_type, left_type, mb_xy, 1);
  2056. nnz = h->non_zero_count[mb_xy];
  2057. nnz_cache = sl->non_zero_count_cache;
  2058. AV_COPY32(&nnz_cache[4 + 8 * 1], &nnz[0]);
  2059. AV_COPY32(&nnz_cache[4 + 8 * 2], &nnz[4]);
  2060. AV_COPY32(&nnz_cache[4 + 8 * 3], &nnz[8]);
  2061. AV_COPY32(&nnz_cache[4 + 8 * 4], &nnz[12]);
  2062. sl->cbp = h->cbp_table[mb_xy];
  2063. if (top_type) {
  2064. nnz = h->non_zero_count[top_xy];
  2065. AV_COPY32(&nnz_cache[4 + 8 * 0], &nnz[3 * 4]);
  2066. }
  2067. if (left_type[LTOP]) {
  2068. nnz = h->non_zero_count[left_xy[LTOP]];
  2069. nnz_cache[3 + 8 * 1] = nnz[3 + 0 * 4];
  2070. nnz_cache[3 + 8 * 2] = nnz[3 + 1 * 4];
  2071. nnz_cache[3 + 8 * 3] = nnz[3 + 2 * 4];
  2072. nnz_cache[3 + 8 * 4] = nnz[3 + 3 * 4];
  2073. }
  2074. /* CAVLC 8x8dct requires NNZ values for residual decoding that differ
  2075. * from what the loop filter needs */
  2076. if (!CABAC(h) && h->ps.pps->transform_8x8_mode) {
  2077. if (IS_8x8DCT(top_type)) {
  2078. nnz_cache[4 + 8 * 0] =
  2079. nnz_cache[5 + 8 * 0] = (h->cbp_table[top_xy] & 0x4000) >> 12;
  2080. nnz_cache[6 + 8 * 0] =
  2081. nnz_cache[7 + 8 * 0] = (h->cbp_table[top_xy] & 0x8000) >> 12;
  2082. }
  2083. if (IS_8x8DCT(left_type[LTOP])) {
  2084. nnz_cache[3 + 8 * 1] =
  2085. nnz_cache[3 + 8 * 2] = (h->cbp_table[left_xy[LTOP]] & 0x2000) >> 12; // FIXME check MBAFF
  2086. }
  2087. if (IS_8x8DCT(left_type[LBOT])) {
  2088. nnz_cache[3 + 8 * 3] =
  2089. nnz_cache[3 + 8 * 4] = (h->cbp_table[left_xy[LBOT]] & 0x8000) >> 12; // FIXME check MBAFF
  2090. }
  2091. if (IS_8x8DCT(mb_type)) {
  2092. nnz_cache[scan8[0]] =
  2093. nnz_cache[scan8[1]] =
  2094. nnz_cache[scan8[2]] =
  2095. nnz_cache[scan8[3]] = (sl->cbp & 0x1000) >> 12;
  2096. nnz_cache[scan8[0 + 4]] =
  2097. nnz_cache[scan8[1 + 4]] =
  2098. nnz_cache[scan8[2 + 4]] =
  2099. nnz_cache[scan8[3 + 4]] = (sl->cbp & 0x2000) >> 12;
  2100. nnz_cache[scan8[0 + 8]] =
  2101. nnz_cache[scan8[1 + 8]] =
  2102. nnz_cache[scan8[2 + 8]] =
  2103. nnz_cache[scan8[3 + 8]] = (sl->cbp & 0x4000) >> 12;
  2104. nnz_cache[scan8[0 + 12]] =
  2105. nnz_cache[scan8[1 + 12]] =
  2106. nnz_cache[scan8[2 + 12]] =
  2107. nnz_cache[scan8[3 + 12]] = (sl->cbp & 0x8000) >> 12;
  2108. }
  2109. }
  2110. return 0;
  2111. }
  2112. static void loop_filter(const H264Context *h, H264SliceContext *sl, int start_x, int end_x)
  2113. {
  2114. uint8_t *dest_y, *dest_cb, *dest_cr;
  2115. int linesize, uvlinesize, mb_x, mb_y;
  2116. const int end_mb_y = sl->mb_y + FRAME_MBAFF(h);
  2117. const int old_slice_type = sl->slice_type;
  2118. const int pixel_shift = h->pixel_shift;
  2119. const int block_h = 16 >> h->chroma_y_shift;
  2120. if (h->postpone_filter)
  2121. return;
  2122. if (sl->deblocking_filter) {
  2123. for (mb_x = start_x; mb_x < end_x; mb_x++)
  2124. for (mb_y = end_mb_y - FRAME_MBAFF(h); mb_y <= end_mb_y; mb_y++) {
  2125. int mb_xy, mb_type;
  2126. mb_xy = sl->mb_xy = mb_x + mb_y * h->mb_stride;
  2127. mb_type = h->cur_pic.mb_type[mb_xy];
  2128. if (FRAME_MBAFF(h))
  2129. sl->mb_mbaff =
  2130. sl->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
  2131. sl->mb_x = mb_x;
  2132. sl->mb_y = mb_y;
  2133. dest_y = h->cur_pic.f->data[0] +
  2134. ((mb_x << pixel_shift) + mb_y * sl->linesize) * 16;
  2135. dest_cb = h->cur_pic.f->data[1] +
  2136. (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
  2137. mb_y * sl->uvlinesize * block_h;
  2138. dest_cr = h->cur_pic.f->data[2] +
  2139. (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
  2140. mb_y * sl->uvlinesize * block_h;
  2141. // FIXME simplify above
  2142. if (MB_FIELD(sl)) {
  2143. linesize = sl->mb_linesize = sl->linesize * 2;
  2144. uvlinesize = sl->mb_uvlinesize = sl->uvlinesize * 2;
  2145. if (mb_y & 1) { // FIXME move out of this function?
  2146. dest_y -= sl->linesize * 15;
  2147. dest_cb -= sl->uvlinesize * (block_h - 1);
  2148. dest_cr -= sl->uvlinesize * (block_h - 1);
  2149. }
  2150. } else {
  2151. linesize = sl->mb_linesize = sl->linesize;
  2152. uvlinesize = sl->mb_uvlinesize = sl->uvlinesize;
  2153. }
  2154. backup_mb_border(h, sl, dest_y, dest_cb, dest_cr, linesize,
  2155. uvlinesize, 0);
  2156. if (fill_filter_caches(h, sl, mb_type))
  2157. continue;
  2158. sl->chroma_qp[0] = get_chroma_qp(h->ps.pps, 0, h->cur_pic.qscale_table[mb_xy]);
  2159. sl->chroma_qp[1] = get_chroma_qp(h->ps.pps, 1, h->cur_pic.qscale_table[mb_xy]);
  2160. if (FRAME_MBAFF(h)) {
  2161. ff_h264_filter_mb(h, sl, mb_x, mb_y, dest_y, dest_cb, dest_cr,
  2162. linesize, uvlinesize);
  2163. } else {
  2164. ff_h264_filter_mb_fast(h, sl, mb_x, mb_y, dest_y, dest_cb,
  2165. dest_cr, linesize, uvlinesize);
  2166. }
  2167. }
  2168. }
  2169. sl->slice_type = old_slice_type;
  2170. sl->mb_x = end_x;
  2171. sl->mb_y = end_mb_y - FRAME_MBAFF(h);
  2172. sl->chroma_qp[0] = get_chroma_qp(h->ps.pps, 0, sl->qscale);
  2173. sl->chroma_qp[1] = get_chroma_qp(h->ps.pps, 1, sl->qscale);
  2174. }
  2175. static void predict_field_decoding_flag(const H264Context *h, H264SliceContext *sl)
  2176. {
  2177. const int mb_xy = sl->mb_x + sl->mb_y * h->mb_stride;
  2178. int mb_type = (h->slice_table[mb_xy - 1] == sl->slice_num) ?
  2179. h->cur_pic.mb_type[mb_xy - 1] :
  2180. (h->slice_table[mb_xy - h->mb_stride] == sl->slice_num) ?
  2181. h->cur_pic.mb_type[mb_xy - h->mb_stride] : 0;
  2182. sl->mb_mbaff = sl->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
  2183. }
  2184. /**
  2185. * Draw edges and report progress for the last MB row.
  2186. */
  2187. static void decode_finish_row(const H264Context *h, H264SliceContext *sl)
  2188. {
  2189. int top = 16 * (sl->mb_y >> FIELD_PICTURE(h));
  2190. int pic_height = 16 * h->mb_height >> FIELD_PICTURE(h);
  2191. int height = 16 << FRAME_MBAFF(h);
  2192. int deblock_border = (16 + 4) << FRAME_MBAFF(h);
  2193. if (sl->deblocking_filter) {
  2194. if ((top + height) >= pic_height)
  2195. height += deblock_border;
  2196. top -= deblock_border;
  2197. }
  2198. if (top >= pic_height || (top + height) < 0)
  2199. return;
  2200. height = FFMIN(height, pic_height - top);
  2201. if (top < 0) {
  2202. height = top + height;
  2203. top = 0;
  2204. }
  2205. ff_h264_draw_horiz_band(h, sl, top, height);
  2206. if (h->droppable || sl->h264->slice_ctx[0].er.error_occurred)
  2207. return;
  2208. ff_thread_report_progress(&h->cur_pic_ptr->tf, top + height - 1,
  2209. h->picture_structure == PICT_BOTTOM_FIELD);
  2210. }
  2211. static void er_add_slice(H264SliceContext *sl,
  2212. int startx, int starty,
  2213. int endx, int endy, int status)
  2214. {
  2215. if (!sl->h264->enable_er)
  2216. return;
  2217. if (CONFIG_ERROR_RESILIENCE) {
  2218. ERContext *er = &sl->h264->slice_ctx[0].er;
  2219. ff_er_add_slice(er, startx, starty, endx, endy, status);
  2220. }
  2221. }
  2222. static int decode_slice(struct AVCodecContext *avctx, void *arg)
  2223. {
  2224. H264SliceContext *sl = arg;
  2225. const H264Context *h = sl->h264;
  2226. int lf_x_start = sl->mb_x;
  2227. int orig_deblock = sl->deblocking_filter;
  2228. int ret;
  2229. sl->linesize = h->cur_pic_ptr->f->linesize[0];
  2230. sl->uvlinesize = h->cur_pic_ptr->f->linesize[1];
  2231. ret = alloc_scratch_buffers(sl, sl->linesize);
  2232. if (ret < 0)
  2233. return ret;
  2234. sl->mb_skip_run = -1;
  2235. av_assert0(h->block_offset[15] == (4 * ((scan8[15] - scan8[0]) & 7) << h->pixel_shift) + 4 * sl->linesize * ((scan8[15] - scan8[0]) >> 3));
  2236. if (h->postpone_filter)
  2237. sl->deblocking_filter = 0;
  2238. sl->is_complex = FRAME_MBAFF(h) || h->picture_structure != PICT_FRAME ||
  2239. (CONFIG_GRAY && (h->flags & AV_CODEC_FLAG_GRAY));
  2240. if (!(h->avctx->active_thread_type & FF_THREAD_SLICE) && h->picture_structure == PICT_FRAME && h->slice_ctx[0].er.error_status_table) {
  2241. const int start_i = av_clip(sl->resync_mb_x + sl->resync_mb_y * h->mb_width, 0, h->mb_num - 1);
  2242. if (start_i) {
  2243. int prev_status = h->slice_ctx[0].er.error_status_table[h->slice_ctx[0].er.mb_index2xy[start_i - 1]];
  2244. prev_status &= ~ VP_START;
  2245. if (prev_status != (ER_MV_END | ER_DC_END | ER_AC_END))
  2246. h->slice_ctx[0].er.error_occurred = 1;
  2247. }
  2248. }
  2249. if (h->ps.pps->cabac) {
  2250. /* realign */
  2251. align_get_bits(&sl->gb);
  2252. /* init cabac */
  2253. ret = ff_init_cabac_decoder(&sl->cabac,
  2254. sl->gb.buffer + get_bits_count(&sl->gb) / 8,
  2255. (get_bits_left(&sl->gb) + 7) / 8);
  2256. if (ret < 0)
  2257. return ret;
  2258. ff_h264_init_cabac_states(h, sl);
  2259. for (;;) {
  2260. // START_TIMER
  2261. int ret, eos;
  2262. if (sl->mb_x + sl->mb_y * h->mb_width >= sl->next_slice_idx) {
  2263. av_log(h->avctx, AV_LOG_ERROR, "Slice overlaps with next at %d\n",
  2264. sl->next_slice_idx);
  2265. er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
  2266. sl->mb_y, ER_MB_ERROR);
  2267. return AVERROR_INVALIDDATA;
  2268. }
  2269. ret = ff_h264_decode_mb_cabac(h, sl);
  2270. // STOP_TIMER("decode_mb_cabac")
  2271. if (ret >= 0)
  2272. ff_h264_hl_decode_mb(h, sl);
  2273. // FIXME optimal? or let mb_decode decode 16x32 ?
  2274. if (ret >= 0 && FRAME_MBAFF(h)) {
  2275. sl->mb_y++;
  2276. ret = ff_h264_decode_mb_cabac(h, sl);
  2277. if (ret >= 0)
  2278. ff_h264_hl_decode_mb(h, sl);
  2279. sl->mb_y--;
  2280. }
  2281. eos = get_cabac_terminate(&sl->cabac);
  2282. if ((h->workaround_bugs & FF_BUG_TRUNCATED) &&
  2283. sl->cabac.bytestream > sl->cabac.bytestream_end + 2) {
  2284. er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x - 1,
  2285. sl->mb_y, ER_MB_END);
  2286. if (sl->mb_x >= lf_x_start)
  2287. loop_filter(h, sl, lf_x_start, sl->mb_x + 1);
  2288. goto finish;
  2289. }
  2290. if (sl->cabac.bytestream > sl->cabac.bytestream_end + 2 )
  2291. av_log(h->avctx, AV_LOG_DEBUG, "bytestream overread %"PTRDIFF_SPECIFIER"\n", sl->cabac.bytestream_end - sl->cabac.bytestream);
  2292. if (ret < 0 || sl->cabac.bytestream > sl->cabac.bytestream_end + 4) {
  2293. av_log(h->avctx, AV_LOG_ERROR,
  2294. "error while decoding MB %d %d, bytestream %"PTRDIFF_SPECIFIER"\n",
  2295. sl->mb_x, sl->mb_y,
  2296. sl->cabac.bytestream_end - sl->cabac.bytestream);
  2297. er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
  2298. sl->mb_y, ER_MB_ERROR);
  2299. return AVERROR_INVALIDDATA;
  2300. }
  2301. if (++sl->mb_x >= h->mb_width) {
  2302. loop_filter(h, sl, lf_x_start, sl->mb_x);
  2303. sl->mb_x = lf_x_start = 0;
  2304. decode_finish_row(h, sl);
  2305. ++sl->mb_y;
  2306. if (FIELD_OR_MBAFF_PICTURE(h)) {
  2307. ++sl->mb_y;
  2308. if (FRAME_MBAFF(h) && sl->mb_y < h->mb_height)
  2309. predict_field_decoding_flag(h, sl);
  2310. }
  2311. }
  2312. if (eos || sl->mb_y >= h->mb_height) {
  2313. ff_tlog(h->avctx, "slice end %d %d\n",
  2314. get_bits_count(&sl->gb), sl->gb.size_in_bits);
  2315. er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x - 1,
  2316. sl->mb_y, ER_MB_END);
  2317. if (sl->mb_x > lf_x_start)
  2318. loop_filter(h, sl, lf_x_start, sl->mb_x);
  2319. goto finish;
  2320. }
  2321. }
  2322. } else {
  2323. for (;;) {
  2324. int ret;
  2325. if (sl->mb_x + sl->mb_y * h->mb_width >= sl->next_slice_idx) {
  2326. av_log(h->avctx, AV_LOG_ERROR, "Slice overlaps with next at %d\n",
  2327. sl->next_slice_idx);
  2328. er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
  2329. sl->mb_y, ER_MB_ERROR);
  2330. return AVERROR_INVALIDDATA;
  2331. }
  2332. ret = ff_h264_decode_mb_cavlc(h, sl);
  2333. if (ret >= 0)
  2334. ff_h264_hl_decode_mb(h, sl);
  2335. // FIXME optimal? or let mb_decode decode 16x32 ?
  2336. if (ret >= 0 && FRAME_MBAFF(h)) {
  2337. sl->mb_y++;
  2338. ret = ff_h264_decode_mb_cavlc(h, sl);
  2339. if (ret >= 0)
  2340. ff_h264_hl_decode_mb(h, sl);
  2341. sl->mb_y--;
  2342. }
  2343. if (ret < 0) {
  2344. av_log(h->avctx, AV_LOG_ERROR,
  2345. "error while decoding MB %d %d\n", sl->mb_x, sl->mb_y);
  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 ret;
  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. if (sl->mb_y >= h->mb_height) {
  2361. ff_tlog(h->avctx, "slice end %d %d\n",
  2362. get_bits_count(&sl->gb), sl->gb.size_in_bits);
  2363. if ( get_bits_left(&sl->gb) == 0
  2364. || get_bits_left(&sl->gb) > 0 && !(h->avctx->err_recognition & AV_EF_AGGRESSIVE)) {
  2365. er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
  2366. sl->mb_x - 1, sl->mb_y, ER_MB_END);
  2367. goto finish;
  2368. } else {
  2369. er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
  2370. sl->mb_x, sl->mb_y, ER_MB_END);
  2371. return AVERROR_INVALIDDATA;
  2372. }
  2373. }
  2374. }
  2375. if (get_bits_left(&sl->gb) <= 0 && sl->mb_skip_run <= 0) {
  2376. ff_tlog(h->avctx, "slice end %d %d\n",
  2377. get_bits_count(&sl->gb), sl->gb.size_in_bits);
  2378. if (get_bits_left(&sl->gb) == 0) {
  2379. er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
  2380. sl->mb_x - 1, sl->mb_y, ER_MB_END);
  2381. if (sl->mb_x > lf_x_start)
  2382. loop_filter(h, sl, lf_x_start, sl->mb_x);
  2383. goto finish;
  2384. } else {
  2385. er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
  2386. sl->mb_y, ER_MB_ERROR);
  2387. return AVERROR_INVALIDDATA;
  2388. }
  2389. }
  2390. }
  2391. }
  2392. finish:
  2393. sl->deblocking_filter = orig_deblock;
  2394. return 0;
  2395. }
  2396. /**
  2397. * Call decode_slice() for each context.
  2398. *
  2399. * @param h h264 master context
  2400. */
  2401. int ff_h264_execute_decode_slices(H264Context *h)
  2402. {
  2403. AVCodecContext *const avctx = h->avctx;
  2404. H264SliceContext *sl;
  2405. int context_count = h->nb_slice_ctx_queued;
  2406. int ret = 0;
  2407. int i, j;
  2408. h->slice_ctx[0].next_slice_idx = INT_MAX;
  2409. if (h->avctx->hwaccel || context_count < 1)
  2410. return 0;
  2411. av_assert0(context_count && h->slice_ctx[context_count - 1].mb_y < h->mb_height);
  2412. if (context_count == 1) {
  2413. h->slice_ctx[0].next_slice_idx = h->mb_width * h->mb_height;
  2414. h->postpone_filter = 0;
  2415. ret = decode_slice(avctx, &h->slice_ctx[0]);
  2416. h->mb_y = h->slice_ctx[0].mb_y;
  2417. if (ret < 0)
  2418. goto finish;
  2419. } else {
  2420. av_assert0(context_count > 0);
  2421. for (i = 0; i < context_count; i++) {
  2422. int next_slice_idx = h->mb_width * h->mb_height;
  2423. int slice_idx;
  2424. sl = &h->slice_ctx[i];
  2425. if (CONFIG_ERROR_RESILIENCE) {
  2426. sl->er.error_count = 0;
  2427. }
  2428. /* make sure none of those slices overlap */
  2429. slice_idx = sl->mb_y * h->mb_width + sl->mb_x;
  2430. for (j = 0; j < context_count; j++) {
  2431. H264SliceContext *sl2 = &h->slice_ctx[j];
  2432. int slice_idx2 = sl2->mb_y * h->mb_width + sl2->mb_x;
  2433. if (i == j || slice_idx2 < slice_idx)
  2434. continue;
  2435. next_slice_idx = FFMIN(next_slice_idx, slice_idx2);
  2436. }
  2437. sl->next_slice_idx = next_slice_idx;
  2438. }
  2439. avctx->execute(avctx, decode_slice, h->slice_ctx,
  2440. NULL, context_count, sizeof(h->slice_ctx[0]));
  2441. /* pull back stuff from slices to master context */
  2442. sl = &h->slice_ctx[context_count - 1];
  2443. h->mb_y = sl->mb_y;
  2444. if (CONFIG_ERROR_RESILIENCE) {
  2445. for (i = 1; i < context_count; i++)
  2446. h->slice_ctx[0].er.error_count += h->slice_ctx[i].er.error_count;
  2447. }
  2448. if (h->postpone_filter) {
  2449. h->postpone_filter = 0;
  2450. for (i = 0; i < context_count; i++) {
  2451. int y_end, x_end;
  2452. sl = &h->slice_ctx[i];
  2453. y_end = FFMIN(sl->mb_y + 1, h->mb_height);
  2454. x_end = (sl->mb_y >= h->mb_height) ? h->mb_width : sl->mb_x;
  2455. for (j = sl->resync_mb_y; j < y_end; j += 1 + FIELD_OR_MBAFF_PICTURE(h)) {
  2456. sl->mb_y = j;
  2457. loop_filter(h, sl, j > sl->resync_mb_y ? 0 : sl->resync_mb_x,
  2458. j == y_end - 1 ? x_end : h->mb_width);
  2459. }
  2460. }
  2461. }
  2462. }
  2463. finish:
  2464. h->nb_slice_ctx_queued = 0;
  2465. return ret;
  2466. }