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.

899 lines
27KB

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