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.

905 lines
28KB

  1. /*
  2. * H.26L/H.264/AVC/JVT/14496-10/... encoder/decoder
  3. * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
  4. *
  5. * This file is part of Libav.
  6. *
  7. * Libav 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. * Libav 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 Libav; 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 / MPEG4 part10 codec.
  24. * @author Michael Niedermayer <michaelni@gmx.at>
  25. */
  26. #ifndef AVCODEC_H264_H
  27. #define AVCODEC_H264_H
  28. #include "libavutil/intreadwrite.h"
  29. #include "dsputil.h"
  30. #include "cabac.h"
  31. #include "mpegvideo.h"
  32. #include "h264dsp.h"
  33. #include "h264pred.h"
  34. #include "rectangle.h"
  35. #define interlaced_dct interlaced_dct_is_a_bad_name
  36. #define mb_intra mb_intra_is_not_initialized_see_mb_type
  37. #define MAX_SPS_COUNT 32
  38. #define MAX_PPS_COUNT 256
  39. #define MAX_MMCO_COUNT 66
  40. #define MAX_DELAYED_PIC_COUNT 16
  41. /* Compiling in interlaced support reduces the speed
  42. * of progressive decoding by about 2%. */
  43. #define ALLOW_INTERLACE
  44. #define FMO 0
  45. /**
  46. * The maximum number of slices supported by the decoder.
  47. * must be a power of 2
  48. */
  49. #define MAX_SLICES 16
  50. #ifdef ALLOW_INTERLACE
  51. #define MB_MBAFF h->mb_mbaff
  52. #define MB_FIELD h->mb_field_decoding_flag
  53. #define FRAME_MBAFF h->mb_aff_frame
  54. #define FIELD_PICTURE (s->picture_structure != PICT_FRAME)
  55. #define LEFT_MBS 2
  56. #define LTOP 0
  57. #define LBOT 1
  58. #define LEFT(i) (i)
  59. #else
  60. #define MB_MBAFF 0
  61. #define MB_FIELD 0
  62. #define FRAME_MBAFF 0
  63. #define FIELD_PICTURE 0
  64. #undef IS_INTERLACED
  65. #define IS_INTERLACED(mb_type) 0
  66. #define LEFT_MBS 1
  67. #define LTOP 0
  68. #define LBOT 0
  69. #define LEFT(i) 0
  70. #endif
  71. #define FIELD_OR_MBAFF_PICTURE (FRAME_MBAFF || FIELD_PICTURE)
  72. #ifndef CABAC
  73. #define CABAC h->pps.cabac
  74. #endif
  75. #define CHROMA422 (h->sps.chroma_format_idc == 2)
  76. #define CHROMA444 (h->sps.chroma_format_idc == 3)
  77. #define EXTENDED_SAR 255
  78. #define MB_TYPE_REF0 MB_TYPE_ACPRED // dirty but it fits in 16 bit
  79. #define MB_TYPE_8x8DCT 0x01000000
  80. #define IS_REF0(a) ((a) & MB_TYPE_REF0)
  81. #define IS_8x8DCT(a) ((a) & MB_TYPE_8x8DCT)
  82. /**
  83. * Value of Picture.reference when Picture is not a reference picture, but
  84. * is held for delayed output.
  85. */
  86. #define DELAYED_PIC_REF 4
  87. #define QP_MAX_NUM (51 + 2 * 6) // The maximum supported qp
  88. /* NAL unit types */
  89. enum {
  90. NAL_SLICE = 1,
  91. NAL_DPA,
  92. NAL_DPB,
  93. NAL_DPC,
  94. NAL_IDR_SLICE,
  95. NAL_SEI,
  96. NAL_SPS,
  97. NAL_PPS,
  98. NAL_AUD,
  99. NAL_END_SEQUENCE,
  100. NAL_END_STREAM,
  101. NAL_FILLER_DATA,
  102. NAL_SPS_EXT,
  103. NAL_AUXILIARY_SLICE = 19
  104. };
  105. /**
  106. * SEI message types
  107. */
  108. typedef enum {
  109. SEI_BUFFERING_PERIOD = 0, ///< buffering period (H.264, D.1.1)
  110. SEI_TYPE_PIC_TIMING = 1, ///< picture timing
  111. SEI_TYPE_USER_DATA_UNREGISTERED = 5, ///< unregistered user data
  112. SEI_TYPE_RECOVERY_POINT = 6 ///< recovery point (frame # to decoder sync)
  113. } SEI_Type;
  114. /**
  115. * pic_struct in picture timing SEI message
  116. */
  117. typedef enum {
  118. SEI_PIC_STRUCT_FRAME = 0, ///< 0: %frame
  119. SEI_PIC_STRUCT_TOP_FIELD = 1, ///< 1: top field
  120. SEI_PIC_STRUCT_BOTTOM_FIELD = 2, ///< 2: bottom field
  121. SEI_PIC_STRUCT_TOP_BOTTOM = 3, ///< 3: top field, bottom field, in that order
  122. SEI_PIC_STRUCT_BOTTOM_TOP = 4, ///< 4: bottom field, top field, in that order
  123. SEI_PIC_STRUCT_TOP_BOTTOM_TOP = 5, ///< 5: top field, bottom field, top field repeated, in that order
  124. SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM = 6, ///< 6: bottom field, top field, bottom field repeated, in that order
  125. SEI_PIC_STRUCT_FRAME_DOUBLING = 7, ///< 7: %frame doubling
  126. SEI_PIC_STRUCT_FRAME_TRIPLING = 8 ///< 8: %frame tripling
  127. } SEI_PicStructType;
  128. /**
  129. * Sequence parameter set
  130. */
  131. typedef struct SPS {
  132. int profile_idc;
  133. int level_idc;
  134. int chroma_format_idc;
  135. int transform_bypass; ///< qpprime_y_zero_transform_bypass_flag
  136. int log2_max_frame_num; ///< log2_max_frame_num_minus4 + 4
  137. int poc_type; ///< pic_order_cnt_type
  138. int log2_max_poc_lsb; ///< log2_max_pic_order_cnt_lsb_minus4
  139. int delta_pic_order_always_zero_flag;
  140. int offset_for_non_ref_pic;
  141. int offset_for_top_to_bottom_field;
  142. int poc_cycle_length; ///< num_ref_frames_in_pic_order_cnt_cycle
  143. int ref_frame_count; ///< num_ref_frames
  144. int gaps_in_frame_num_allowed_flag;
  145. int mb_width; ///< pic_width_in_mbs_minus1 + 1
  146. int mb_height; ///< pic_height_in_map_units_minus1 + 1
  147. int frame_mbs_only_flag;
  148. int mb_aff; ///< mb_adaptive_frame_field_flag
  149. int direct_8x8_inference_flag;
  150. int crop; ///< frame_cropping_flag
  151. unsigned int crop_left; ///< frame_cropping_rect_left_offset
  152. unsigned int crop_right; ///< frame_cropping_rect_right_offset
  153. unsigned int crop_top; ///< frame_cropping_rect_top_offset
  154. unsigned int crop_bottom; ///< frame_cropping_rect_bottom_offset
  155. int vui_parameters_present_flag;
  156. AVRational sar;
  157. int video_signal_type_present_flag;
  158. int full_range;
  159. int colour_description_present_flag;
  160. enum AVColorPrimaries color_primaries;
  161. enum AVColorTransferCharacteristic color_trc;
  162. enum AVColorSpace colorspace;
  163. int timing_info_present_flag;
  164. uint32_t num_units_in_tick;
  165. uint32_t time_scale;
  166. int fixed_frame_rate_flag;
  167. short offset_for_ref_frame[256]; // FIXME dyn aloc?
  168. int bitstream_restriction_flag;
  169. int num_reorder_frames;
  170. int scaling_matrix_present;
  171. uint8_t scaling_matrix4[6][16];
  172. uint8_t scaling_matrix8[6][64];
  173. int nal_hrd_parameters_present_flag;
  174. int vcl_hrd_parameters_present_flag;
  175. int pic_struct_present_flag;
  176. int time_offset_length;
  177. int cpb_cnt; ///< See H.264 E.1.2
  178. int initial_cpb_removal_delay_length; ///< initial_cpb_removal_delay_length_minus1 + 1
  179. int cpb_removal_delay_length; ///< cpb_removal_delay_length_minus1 + 1
  180. int dpb_output_delay_length; ///< dpb_output_delay_length_minus1 + 1
  181. int bit_depth_luma; ///< bit_depth_luma_minus8 + 8
  182. int bit_depth_chroma; ///< bit_depth_chroma_minus8 + 8
  183. int residual_color_transform_flag; ///< residual_colour_transform_flag
  184. int constraint_set_flags; ///< constraint_set[0-3]_flag
  185. } SPS;
  186. /**
  187. * Picture parameter set
  188. */
  189. typedef struct PPS {
  190. unsigned int sps_id;
  191. int cabac; ///< entropy_coding_mode_flag
  192. int pic_order_present; ///< pic_order_present_flag
  193. int slice_group_count; ///< num_slice_groups_minus1 + 1
  194. int mb_slice_group_map_type;
  195. unsigned int ref_count[2]; ///< num_ref_idx_l0/1_active_minus1 + 1
  196. int weighted_pred; ///< weighted_pred_flag
  197. int weighted_bipred_idc;
  198. int init_qp; ///< pic_init_qp_minus26 + 26
  199. int init_qs; ///< pic_init_qs_minus26 + 26
  200. int chroma_qp_index_offset[2];
  201. int deblocking_filter_parameters_present; ///< deblocking_filter_parameters_present_flag
  202. int constrained_intra_pred; ///< constrained_intra_pred_flag
  203. int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag
  204. int transform_8x8_mode; ///< transform_8x8_mode_flag
  205. uint8_t scaling_matrix4[6][16];
  206. uint8_t scaling_matrix8[6][64];
  207. uint8_t chroma_qp_table[2][64]; ///< pre-scaled (with chroma_qp_index_offset) version of qp_table
  208. int chroma_qp_diff;
  209. } PPS;
  210. /**
  211. * Memory management control operation opcode.
  212. */
  213. typedef enum MMCOOpcode {
  214. MMCO_END = 0,
  215. MMCO_SHORT2UNUSED,
  216. MMCO_LONG2UNUSED,
  217. MMCO_SHORT2LONG,
  218. MMCO_SET_MAX_LONG,
  219. MMCO_RESET,
  220. MMCO_LONG,
  221. } MMCOOpcode;
  222. /**
  223. * Memory management control operation.
  224. */
  225. typedef struct MMCO {
  226. MMCOOpcode opcode;
  227. int short_pic_num; ///< pic_num without wrapping (pic_num & max_pic_num)
  228. int long_arg; ///< index, pic_num, or num long refs depending on opcode
  229. } MMCO;
  230. /**
  231. * H264Context
  232. */
  233. typedef struct H264Context {
  234. MpegEncContext s;
  235. H264DSPContext h264dsp;
  236. int pixel_shift; ///< 0 for 8-bit H264, 1 for high-bit-depth H264
  237. int chroma_qp[2]; // QPc
  238. int qp_thresh; ///< QP threshold to skip loopfilter
  239. int prev_mb_skipped;
  240. int next_mb_skipped;
  241. // prediction stuff
  242. int chroma_pred_mode;
  243. int intra16x16_pred_mode;
  244. int topleft_mb_xy;
  245. int top_mb_xy;
  246. int topright_mb_xy;
  247. int left_mb_xy[LEFT_MBS];
  248. int topleft_type;
  249. int top_type;
  250. int topright_type;
  251. int left_type[LEFT_MBS];
  252. const uint8_t *left_block;
  253. int topleft_partition;
  254. int8_t intra4x4_pred_mode_cache[5 * 8];
  255. int8_t(*intra4x4_pred_mode);
  256. H264PredContext hpc;
  257. unsigned int topleft_samples_available;
  258. unsigned int top_samples_available;
  259. unsigned int topright_samples_available;
  260. unsigned int left_samples_available;
  261. uint8_t (*top_borders[2])[(16 * 3) * 2];
  262. /**
  263. * non zero coeff count cache.
  264. * is 64 if not available.
  265. */
  266. DECLARE_ALIGNED(8, uint8_t, non_zero_count_cache)[15 * 8];
  267. uint8_t (*non_zero_count)[48];
  268. /**
  269. * Motion vector cache.
  270. */
  271. DECLARE_ALIGNED(16, int16_t, mv_cache)[2][5 * 8][2];
  272. DECLARE_ALIGNED(8, int8_t, ref_cache)[2][5 * 8];
  273. #define LIST_NOT_USED -1 // FIXME rename?
  274. #define PART_NOT_AVAILABLE -2
  275. /**
  276. * number of neighbors (top and/or left) that used 8x8 dct
  277. */
  278. int neighbor_transform_size;
  279. /**
  280. * block_offset[ 0..23] for frame macroblocks
  281. * block_offset[24..47] for field macroblocks
  282. */
  283. int block_offset[2 * (16 * 3)];
  284. uint32_t *mb2b_xy; // FIXME are these 4 a good idea?
  285. uint32_t *mb2br_xy;
  286. int b_stride; // FIXME use s->b4_stride
  287. int mb_linesize; ///< may be equal to s->linesize or s->linesize * 2, for mbaff
  288. int mb_uvlinesize;
  289. int emu_edge_width;
  290. int emu_edge_height;
  291. SPS sps; ///< current sps
  292. /**
  293. * current pps
  294. */
  295. PPS pps; // FIXME move to Picture perhaps? (->no) do we need that?
  296. uint32_t dequant4_buffer[6][QP_MAX_NUM + 1][16]; // FIXME should these be moved down?
  297. uint32_t dequant8_buffer[6][QP_MAX_NUM + 1][64];
  298. uint32_t(*dequant4_coeff[6])[16];
  299. uint32_t(*dequant8_coeff[6])[64];
  300. int slice_num;
  301. uint16_t *slice_table; ///< slice_table_base + 2*mb_stride + 1
  302. int slice_type;
  303. int slice_type_nos; ///< S free slice type (SI/SP are remapped to I/P)
  304. int slice_type_fixed;
  305. // interlacing specific flags
  306. int mb_aff_frame;
  307. int mb_field_decoding_flag;
  308. int mb_mbaff; ///< mb_aff_frame && mb_field_decoding_flag
  309. DECLARE_ALIGNED(8, uint16_t, sub_mb_type)[4];
  310. // Weighted pred stuff
  311. int use_weight;
  312. int use_weight_chroma;
  313. int luma_log2_weight_denom;
  314. int chroma_log2_weight_denom;
  315. // The following 2 can be changed to int8_t but that causes 10cpu cycles speedloss
  316. int luma_weight[48][2][2];
  317. int chroma_weight[48][2][2][2];
  318. int implicit_weight[48][48][2];
  319. int direct_spatial_mv_pred;
  320. int col_parity;
  321. int col_fieldoff;
  322. int dist_scale_factor[16];
  323. int dist_scale_factor_field[2][32];
  324. int map_col_to_list0[2][16 + 32];
  325. int map_col_to_list0_field[2][2][16 + 32];
  326. /**
  327. * num_ref_idx_l0/1_active_minus1 + 1
  328. */
  329. unsigned int ref_count[2]; ///< counts frames or fields, depending on current mb mode
  330. unsigned int list_count;
  331. uint8_t *list_counts; ///< Array of list_count per MB specifying the slice type
  332. Picture ref_list[2][48]; /**< 0..15: frame refs, 16..47: mbaff field refs.
  333. * Reordered version of default_ref_list
  334. * according to picture reordering in slice header */
  335. int ref2frm[MAX_SLICES][2][64]; ///< reference to frame number lists, used in the loop filter, the first 2 are for -2,-1
  336. // data partitioning
  337. GetBitContext intra_gb;
  338. GetBitContext inter_gb;
  339. GetBitContext *intra_gb_ptr;
  340. GetBitContext *inter_gb_ptr;
  341. DECLARE_ALIGNED(16, DCTELEM, mb)[16 * 48 * 2]; ///< as a dct coeffecient is int32_t in high depth, we need to reserve twice the space.
  342. DECLARE_ALIGNED(16, DCTELEM, mb_luma_dc)[3][16 * 2];
  343. DCTELEM mb_padding[256 * 2]; ///< as mb is addressed by scantable[i] and scantable is uint8_t we can either check that i is not too large or ensure that there is some unused stuff after mb
  344. /**
  345. * Cabac
  346. */
  347. CABACContext cabac;
  348. uint8_t cabac_state[1024];
  349. /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0, 1, 2), 0x0? luma_cbp */
  350. uint16_t *cbp_table;
  351. int cbp;
  352. int top_cbp;
  353. int left_cbp;
  354. /* chroma_pred_mode for i4x4 or i16x16, else 0 */
  355. uint8_t *chroma_pred_mode_table;
  356. int last_qscale_diff;
  357. uint8_t (*mvd_table[2])[2];
  358. DECLARE_ALIGNED(16, uint8_t, mvd_cache)[2][5 * 8][2];
  359. uint8_t *direct_table;
  360. uint8_t direct_cache[5 * 8];
  361. uint8_t zigzag_scan[16];
  362. uint8_t zigzag_scan8x8[64];
  363. uint8_t zigzag_scan8x8_cavlc[64];
  364. uint8_t field_scan[16];
  365. uint8_t field_scan8x8[64];
  366. uint8_t field_scan8x8_cavlc[64];
  367. const uint8_t *zigzag_scan_q0;
  368. const uint8_t *zigzag_scan8x8_q0;
  369. const uint8_t *zigzag_scan8x8_cavlc_q0;
  370. const uint8_t *field_scan_q0;
  371. const uint8_t *field_scan8x8_q0;
  372. const uint8_t *field_scan8x8_cavlc_q0;
  373. int x264_build;
  374. int mb_xy;
  375. int is_complex;
  376. // deblock
  377. int deblocking_filter; ///< disable_deblocking_filter_idc with 1 <-> 0
  378. int slice_alpha_c0_offset;
  379. int slice_beta_offset;
  380. // =============================================================
  381. // Things below are not used in the MB or more inner code
  382. int nal_ref_idc;
  383. int nal_unit_type;
  384. uint8_t *rbsp_buffer[2];
  385. unsigned int rbsp_buffer_size[2];
  386. /**
  387. * Used to parse AVC variant of h264
  388. */
  389. int is_avc; ///< this flag is != 0 if codec is avc1
  390. int nal_length_size; ///< Number of bytes used for nal length (1, 2 or 4)
  391. int got_first; ///< this flag is != 0 if we've parsed a frame
  392. SPS *sps_buffers[MAX_SPS_COUNT];
  393. PPS *pps_buffers[MAX_PPS_COUNT];
  394. int dequant_coeff_pps; ///< reinit tables when pps changes
  395. uint16_t *slice_table_base;
  396. // POC stuff
  397. int poc_lsb;
  398. int poc_msb;
  399. int delta_poc_bottom;
  400. int delta_poc[2];
  401. int frame_num;
  402. int prev_poc_msb; ///< poc_msb of the last reference pic for POC type 0
  403. int prev_poc_lsb; ///< poc_lsb of the last reference pic for POC type 0
  404. int frame_num_offset; ///< for POC type 2
  405. int prev_frame_num_offset; ///< for POC type 2
  406. int prev_frame_num; ///< frame_num of the last pic for POC type 1/2
  407. /**
  408. * frame_num for frames or 2 * frame_num + 1 for field pics.
  409. */
  410. int curr_pic_num;
  411. /**
  412. * max_frame_num or 2 * max_frame_num for field pics.
  413. */
  414. int max_pic_num;
  415. int redundant_pic_count;
  416. Picture *short_ref[32];
  417. Picture *long_ref[32];
  418. Picture default_ref_list[2][32]; ///< base reference list for all slices of a coded picture
  419. Picture *delayed_pic[MAX_DELAYED_PIC_COUNT + 2]; // FIXME size?
  420. int last_pocs[MAX_DELAYED_PIC_COUNT];
  421. Picture *next_output_pic;
  422. int outputed_poc;
  423. int next_outputed_poc;
  424. /**
  425. * memory management control operations buffer.
  426. */
  427. MMCO mmco[MAX_MMCO_COUNT];
  428. int mmco_index;
  429. int mmco_reset;
  430. int long_ref_count; ///< number of actual long term references
  431. int short_ref_count; ///< number of actual short term references
  432. int cabac_init_idc;
  433. /**
  434. * @name Members for slice based multithreading
  435. * @{
  436. */
  437. struct H264Context *thread_context[MAX_THREADS];
  438. /**
  439. * current slice number, used to initalize slice_num of each thread/context
  440. */
  441. int current_slice;
  442. /**
  443. * Max number of threads / contexts.
  444. * This is equal to AVCodecContext.thread_count unless
  445. * multithreaded decoding is impossible, in which case it is
  446. * reduced to 1.
  447. */
  448. int max_contexts;
  449. /**
  450. * 1 if the single thread fallback warning has already been
  451. * displayed, 0 otherwise.
  452. */
  453. int single_decode_warning;
  454. int last_slice_type;
  455. /** @} */
  456. /**
  457. * pic_struct in picture timing SEI message
  458. */
  459. SEI_PicStructType sei_pic_struct;
  460. /**
  461. * Complement sei_pic_struct
  462. * SEI_PIC_STRUCT_TOP_BOTTOM and SEI_PIC_STRUCT_BOTTOM_TOP indicate interlaced frames.
  463. * However, soft telecined frames may have these values.
  464. * This is used in an attempt to flag soft telecine progressive.
  465. */
  466. int prev_interlaced_frame;
  467. /**
  468. * Bit set of clock types for fields/frames in picture timing SEI message.
  469. * For each found ct_type, appropriate bit is set (e.g., bit 1 for
  470. * interlaced).
  471. */
  472. int sei_ct_type;
  473. /**
  474. * dpb_output_delay in picture timing SEI message, see H.264 C.2.2
  475. */
  476. int sei_dpb_output_delay;
  477. /**
  478. * cpb_removal_delay in picture timing SEI message, see H.264 C.1.2
  479. */
  480. int sei_cpb_removal_delay;
  481. /**
  482. * recovery_frame_cnt from SEI message
  483. *
  484. * Set to -1 if no recovery point SEI message found or to number of frames
  485. * before playback synchronizes. Frames having recovery point are key
  486. * frames.
  487. */
  488. int sei_recovery_frame_cnt;
  489. int luma_weight_flag[2]; ///< 7.4.3.2 luma_weight_lX_flag
  490. int chroma_weight_flag[2]; ///< 7.4.3.2 chroma_weight_lX_flag
  491. // Timestamp stuff
  492. int sei_buffering_period_present; ///< Buffering period SEI flag
  493. int initial_cpb_removal_delay[32]; ///< Initial timestamps for CPBs
  494. int cur_chroma_format_idc;
  495. } H264Context;
  496. extern const uint8_t ff_h264_chroma_qp[3][QP_MAX_NUM + 1]; ///< One chroma qp table for each supported bit depth (8, 9, 10).
  497. extern const uint16_t ff_h264_mb_sizes[4];
  498. /**
  499. * Decode SEI
  500. */
  501. int ff_h264_decode_sei(H264Context *h);
  502. /**
  503. * Decode SPS
  504. */
  505. int ff_h264_decode_seq_parameter_set(H264Context *h);
  506. /**
  507. * compute profile from sps
  508. */
  509. int ff_h264_get_profile(SPS *sps);
  510. /**
  511. * Decode PPS
  512. */
  513. int ff_h264_decode_picture_parameter_set(H264Context *h, int bit_length);
  514. /**
  515. * Decode a network abstraction layer unit.
  516. * @param consumed is the number of bytes used as input
  517. * @param length is the length of the array
  518. * @param dst_length is the number of decoded bytes FIXME here
  519. * or a decode rbsp tailing?
  520. * @return decoded bytes, might be src+1 if no escapes
  521. */
  522. const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src,
  523. int *dst_length, int *consumed, int length);
  524. /**
  525. * Free any data that may have been allocated in the H264 context
  526. * like SPS, PPS etc.
  527. */
  528. av_cold void ff_h264_free_context(H264Context *h);
  529. /**
  530. * Reconstruct bitstream slice_type.
  531. */
  532. int ff_h264_get_slice_type(const H264Context *h);
  533. /**
  534. * Allocate tables.
  535. * needs width/height
  536. */
  537. int ff_h264_alloc_tables(H264Context *h);
  538. /**
  539. * Fill the default_ref_list.
  540. */
  541. int ff_h264_fill_default_ref_list(H264Context *h);
  542. int ff_h264_decode_ref_pic_list_reordering(H264Context *h);
  543. void ff_h264_fill_mbaff_ref_list(H264Context *h);
  544. void ff_h264_remove_all_refs(H264Context *h);
  545. /**
  546. * Execute the reference picture marking (memory management control operations).
  547. */
  548. int ff_h264_execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count);
  549. int ff_h264_decode_ref_pic_marking(H264Context *h, GetBitContext *gb);
  550. void ff_generate_sliding_window_mmcos(H264Context *h);
  551. /**
  552. * Check if the top & left blocks are available if needed & change the
  553. * dc mode so it only uses the available blocks.
  554. */
  555. int ff_h264_check_intra4x4_pred_mode(H264Context *h);
  556. /**
  557. * Check if the top & left blocks are available if needed & change the
  558. * dc mode so it only uses the available blocks.
  559. */
  560. int ff_h264_check_intra_pred_mode(H264Context *h, int mode, int is_chroma);
  561. void ff_h264_hl_decode_mb(H264Context *h);
  562. int ff_h264_frame_start(H264Context *h);
  563. int ff_h264_decode_extradata(H264Context *h);
  564. av_cold int ff_h264_decode_init(AVCodecContext *avctx);
  565. av_cold void ff_h264_decode_init_vlc(void);
  566. /**
  567. * Decode a macroblock
  568. * @return 0 if OK, ER_AC_ERROR / ER_DC_ERROR / ER_MV_ERROR on error
  569. */
  570. int ff_h264_decode_mb_cavlc(H264Context *h);
  571. /**
  572. * Decode a CABAC coded macroblock
  573. * @return 0 if OK, ER_AC_ERROR / ER_DC_ERROR / ER_MV_ERROR on error
  574. */
  575. int ff_h264_decode_mb_cabac(H264Context *h);
  576. void ff_h264_init_cabac_states(H264Context *h);
  577. void ff_h264_direct_dist_scale_factor(H264Context *const h);
  578. void ff_h264_direct_ref_list_init(H264Context *const h);
  579. void ff_h264_pred_direct_motion(H264Context *const h, int *mb_type);
  580. void ff_h264_filter_mb_fast(H264Context *h, int mb_x, int mb_y,
  581. uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr,
  582. unsigned int linesize, unsigned int uvlinesize);
  583. void ff_h264_filter_mb(H264Context *h, int mb_x, int mb_y,
  584. uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr,
  585. unsigned int linesize, unsigned int uvlinesize);
  586. /**
  587. * Reset SEI values at the beginning of the frame.
  588. *
  589. * @param h H.264 context.
  590. */
  591. void ff_h264_reset_sei(H264Context *h);
  592. /*
  593. * o-o o-o
  594. * / / /
  595. * o-o o-o
  596. * ,---'
  597. * o-o o-o
  598. * / / /
  599. * o-o o-o
  600. */
  601. /* Scan8 organization:
  602. * 0 1 2 3 4 5 6 7
  603. * 0 DY y y y y y
  604. * 1 y Y Y Y Y
  605. * 2 y Y Y Y Y
  606. * 3 y Y Y Y Y
  607. * 4 y Y Y Y Y
  608. * 5 DU u u u u u
  609. * 6 u U U U U
  610. * 7 u U U U U
  611. * 8 u U U U U
  612. * 9 u U U U U
  613. * 10 DV v v v v v
  614. * 11 v V V V V
  615. * 12 v V V V V
  616. * 13 v V V V V
  617. * 14 v V V V V
  618. * DY/DU/DV are for luma/chroma DC.
  619. */
  620. #define LUMA_DC_BLOCK_INDEX 48
  621. #define CHROMA_DC_BLOCK_INDEX 49
  622. // This table must be here because scan8[constant] must be known at compiletime
  623. static const uint8_t scan8[16 * 3 + 3] = {
  624. 4 + 1 * 8, 5 + 1 * 8, 4 + 2 * 8, 5 + 2 * 8,
  625. 6 + 1 * 8, 7 + 1 * 8, 6 + 2 * 8, 7 + 2 * 8,
  626. 4 + 3 * 8, 5 + 3 * 8, 4 + 4 * 8, 5 + 4 * 8,
  627. 6 + 3 * 8, 7 + 3 * 8, 6 + 4 * 8, 7 + 4 * 8,
  628. 4 + 6 * 8, 5 + 6 * 8, 4 + 7 * 8, 5 + 7 * 8,
  629. 6 + 6 * 8, 7 + 6 * 8, 6 + 7 * 8, 7 + 7 * 8,
  630. 4 + 8 * 8, 5 + 8 * 8, 4 + 9 * 8, 5 + 9 * 8,
  631. 6 + 8 * 8, 7 + 8 * 8, 6 + 9 * 8, 7 + 9 * 8,
  632. 4 + 11 * 8, 5 + 11 * 8, 4 + 12 * 8, 5 + 12 * 8,
  633. 6 + 11 * 8, 7 + 11 * 8, 6 + 12 * 8, 7 + 12 * 8,
  634. 4 + 13 * 8, 5 + 13 * 8, 4 + 14 * 8, 5 + 14 * 8,
  635. 6 + 13 * 8, 7 + 13 * 8, 6 + 14 * 8, 7 + 14 * 8,
  636. 0 + 0 * 8, 0 + 5 * 8, 0 + 10 * 8
  637. };
  638. static av_always_inline uint32_t pack16to32(int a, int b)
  639. {
  640. #if HAVE_BIGENDIAN
  641. return (b & 0xFFFF) + (a << 16);
  642. #else
  643. return (a & 0xFFFF) + (b << 16);
  644. #endif
  645. }
  646. static av_always_inline uint16_t pack8to16(int a, int b)
  647. {
  648. #if HAVE_BIGENDIAN
  649. return (b & 0xFF) + (a << 8);
  650. #else
  651. return (a & 0xFF) + (b << 8);
  652. #endif
  653. }
  654. /**
  655. * Get the chroma qp.
  656. */
  657. static av_always_inline int get_chroma_qp(H264Context *h, int t, int qscale)
  658. {
  659. return h->pps.chroma_qp_table[t][qscale];
  660. }
  661. /**
  662. * Get the predicted intra4x4 prediction mode.
  663. */
  664. static av_always_inline int pred_intra_mode(H264Context *h, int n)
  665. {
  666. const int index8 = scan8[n];
  667. const int left = h->intra4x4_pred_mode_cache[index8 - 1];
  668. const int top = h->intra4x4_pred_mode_cache[index8 - 8];
  669. const int min = FFMIN(left, top);
  670. tprintf(h->s.avctx, "mode:%d %d min:%d\n", left, top, min);
  671. if (min < 0)
  672. return DC_PRED;
  673. else
  674. return min;
  675. }
  676. static av_always_inline void write_back_intra_pred_mode(H264Context *h)
  677. {
  678. int8_t *i4x4 = h->intra4x4_pred_mode + h->mb2br_xy[h->mb_xy];
  679. int8_t *i4x4_cache = h->intra4x4_pred_mode_cache;
  680. AV_COPY32(i4x4, i4x4_cache + 4 + 8 * 4);
  681. i4x4[4] = i4x4_cache[7 + 8 * 3];
  682. i4x4[5] = i4x4_cache[7 + 8 * 2];
  683. i4x4[6] = i4x4_cache[7 + 8 * 1];
  684. }
  685. static av_always_inline void write_back_non_zero_count(H264Context *h)
  686. {
  687. const int mb_xy = h->mb_xy;
  688. uint8_t *nnz = h->non_zero_count[mb_xy];
  689. uint8_t *nnz_cache = h->non_zero_count_cache;
  690. AV_COPY32(&nnz[ 0], &nnz_cache[4 + 8 * 1]);
  691. AV_COPY32(&nnz[ 4], &nnz_cache[4 + 8 * 2]);
  692. AV_COPY32(&nnz[ 8], &nnz_cache[4 + 8 * 3]);
  693. AV_COPY32(&nnz[12], &nnz_cache[4 + 8 * 4]);
  694. AV_COPY32(&nnz[16], &nnz_cache[4 + 8 * 6]);
  695. AV_COPY32(&nnz[20], &nnz_cache[4 + 8 * 7]);
  696. AV_COPY32(&nnz[32], &nnz_cache[4 + 8 * 11]);
  697. AV_COPY32(&nnz[36], &nnz_cache[4 + 8 * 12]);
  698. if (!h->s.chroma_y_shift) {
  699. AV_COPY32(&nnz[24], &nnz_cache[4 + 8 * 8]);
  700. AV_COPY32(&nnz[28], &nnz_cache[4 + 8 * 9]);
  701. AV_COPY32(&nnz[40], &nnz_cache[4 + 8 * 13]);
  702. AV_COPY32(&nnz[44], &nnz_cache[4 + 8 * 14]);
  703. }
  704. }
  705. static av_always_inline void write_back_motion_list(H264Context *h,
  706. MpegEncContext *const s,
  707. int b_stride,
  708. int b_xy, int b8_xy,
  709. int mb_type, int list)
  710. {
  711. int16_t(*mv_dst)[2] = &s->current_picture.f.motion_val[list][b_xy];
  712. int16_t(*mv_src)[2] = &h->mv_cache[list][scan8[0]];
  713. AV_COPY128(mv_dst + 0 * b_stride, mv_src + 8 * 0);
  714. AV_COPY128(mv_dst + 1 * b_stride, mv_src + 8 * 1);
  715. AV_COPY128(mv_dst + 2 * b_stride, mv_src + 8 * 2);
  716. AV_COPY128(mv_dst + 3 * b_stride, mv_src + 8 * 3);
  717. if (CABAC) {
  718. uint8_t (*mvd_dst)[2] = &h->mvd_table[list][FMO ? 8 * h->mb_xy
  719. : h->mb2br_xy[h->mb_xy]];
  720. uint8_t(*mvd_src)[2] = &h->mvd_cache[list][scan8[0]];
  721. if (IS_SKIP(mb_type)) {
  722. AV_ZERO128(mvd_dst);
  723. } else {
  724. AV_COPY64(mvd_dst, mvd_src + 8 * 3);
  725. AV_COPY16(mvd_dst + 3 + 3, mvd_src + 3 + 8 * 0);
  726. AV_COPY16(mvd_dst + 3 + 2, mvd_src + 3 + 8 * 1);
  727. AV_COPY16(mvd_dst + 3 + 1, mvd_src + 3 + 8 * 2);
  728. }
  729. }
  730. {
  731. int8_t *ref_index = &s->current_picture.f.ref_index[list][b8_xy];
  732. int8_t *ref_cache = h->ref_cache[list];
  733. ref_index[0 + 0 * 2] = ref_cache[scan8[0]];
  734. ref_index[1 + 0 * 2] = ref_cache[scan8[4]];
  735. ref_index[0 + 1 * 2] = ref_cache[scan8[8]];
  736. ref_index[1 + 1 * 2] = ref_cache[scan8[12]];
  737. }
  738. }
  739. static av_always_inline void write_back_motion(H264Context *h, int mb_type)
  740. {
  741. MpegEncContext *const s = &h->s;
  742. const int b_stride = h->b_stride;
  743. const int b_xy = 4 * s->mb_x + 4 * s->mb_y * h->b_stride; // try mb2b(8)_xy
  744. const int b8_xy = 4 * h->mb_xy;
  745. if (USES_LIST(mb_type, 0)) {
  746. write_back_motion_list(h, s, b_stride, b_xy, b8_xy, mb_type, 0);
  747. } else {
  748. fill_rectangle(&s->current_picture.f.ref_index[0][b8_xy],
  749. 2, 2, 2, (uint8_t)LIST_NOT_USED, 1);
  750. }
  751. if (USES_LIST(mb_type, 1))
  752. write_back_motion_list(h, s, b_stride, b_xy, b8_xy, mb_type, 1);
  753. if (h->slice_type_nos == AV_PICTURE_TYPE_B && CABAC) {
  754. if (IS_8X8(mb_type)) {
  755. uint8_t *direct_table = &h->direct_table[4 * h->mb_xy];
  756. direct_table[1] = h->sub_mb_type[1] >> 1;
  757. direct_table[2] = h->sub_mb_type[2] >> 1;
  758. direct_table[3] = h->sub_mb_type[3] >> 1;
  759. }
  760. }
  761. }
  762. static av_always_inline int get_dct8x8_allowed(H264Context *h)
  763. {
  764. if (h->sps.direct_8x8_inference_flag)
  765. return !(AV_RN64A(h->sub_mb_type) &
  766. ((MB_TYPE_16x8 | MB_TYPE_8x16 | MB_TYPE_8x8) *
  767. 0x0001000100010001ULL));
  768. else
  769. return !(AV_RN64A(h->sub_mb_type) &
  770. ((MB_TYPE_16x8 | MB_TYPE_8x16 | MB_TYPE_8x8 | MB_TYPE_DIRECT2) *
  771. 0x0001000100010001ULL));
  772. }
  773. #endif /* AVCODEC_H264_H */