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.

952 lines
29KB

  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 "cabac.h"
  30. #include "error_resilience.h"
  31. #include "get_bits.h"
  32. #include "mpegvideo.h"
  33. #include "h264chroma.h"
  34. #include "h264dsp.h"
  35. #include "h264pred.h"
  36. #include "h264qpel.h"
  37. #include "rectangle.h"
  38. #define MAX_SPS_COUNT 32
  39. #define MAX_PPS_COUNT 256
  40. #define MAX_MMCO_COUNT 66
  41. #define MAX_DELAYED_PIC_COUNT 16
  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 (h->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 CHROMA422 (h->sps.chroma_format_idc == 2)
  77. #define CHROMA444 (h->sps.chroma_format_idc == 3)
  78. #define EXTENDED_SAR 255
  79. #define MB_TYPE_REF0 MB_TYPE_ACPRED // dirty but it fits in 16 bit
  80. #define MB_TYPE_8x8DCT 0x01000000
  81. #define IS_REF0(a) ((a) & MB_TYPE_REF0)
  82. #define IS_8x8DCT(a) ((a) & MB_TYPE_8x8DCT)
  83. #define QP_MAX_NUM (51 + 2 * 6) // The maximum supported qp
  84. /* NAL unit types */
  85. enum {
  86. NAL_SLICE = 1,
  87. NAL_DPA,
  88. NAL_DPB,
  89. NAL_DPC,
  90. NAL_IDR_SLICE,
  91. NAL_SEI,
  92. NAL_SPS,
  93. NAL_PPS,
  94. NAL_AUD,
  95. NAL_END_SEQUENCE,
  96. NAL_END_STREAM,
  97. NAL_FILLER_DATA,
  98. NAL_SPS_EXT,
  99. NAL_AUXILIARY_SLICE = 19,
  100. NAL_FF_IGNORE = 0xff0f001,
  101. };
  102. /**
  103. * SEI message types
  104. */
  105. typedef enum {
  106. SEI_BUFFERING_PERIOD = 0, ///< buffering period (H.264, D.1.1)
  107. SEI_TYPE_PIC_TIMING = 1, ///< picture timing
  108. SEI_TYPE_USER_DATA_UNREGISTERED = 5, ///< unregistered user data
  109. SEI_TYPE_RECOVERY_POINT = 6 ///< recovery point (frame # to decoder sync)
  110. } SEI_Type;
  111. /**
  112. * pic_struct in picture timing SEI message
  113. */
  114. typedef enum {
  115. SEI_PIC_STRUCT_FRAME = 0, ///< 0: %frame
  116. SEI_PIC_STRUCT_TOP_FIELD = 1, ///< 1: top field
  117. SEI_PIC_STRUCT_BOTTOM_FIELD = 2, ///< 2: bottom field
  118. SEI_PIC_STRUCT_TOP_BOTTOM = 3, ///< 3: top field, bottom field, in that order
  119. SEI_PIC_STRUCT_BOTTOM_TOP = 4, ///< 4: bottom field, top field, in that order
  120. SEI_PIC_STRUCT_TOP_BOTTOM_TOP = 5, ///< 5: top field, bottom field, top field repeated, in that order
  121. SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM = 6, ///< 6: bottom field, top field, bottom field repeated, in that order
  122. SEI_PIC_STRUCT_FRAME_DOUBLING = 7, ///< 7: %frame doubling
  123. SEI_PIC_STRUCT_FRAME_TRIPLING = 8 ///< 8: %frame tripling
  124. } SEI_PicStructType;
  125. /**
  126. * Sequence parameter set
  127. */
  128. typedef struct SPS {
  129. int profile_idc;
  130. int level_idc;
  131. int chroma_format_idc;
  132. int transform_bypass; ///< qpprime_y_zero_transform_bypass_flag
  133. int log2_max_frame_num; ///< log2_max_frame_num_minus4 + 4
  134. int poc_type; ///< pic_order_cnt_type
  135. int log2_max_poc_lsb; ///< log2_max_pic_order_cnt_lsb_minus4
  136. int delta_pic_order_always_zero_flag;
  137. int offset_for_non_ref_pic;
  138. int offset_for_top_to_bottom_field;
  139. int poc_cycle_length; ///< num_ref_frames_in_pic_order_cnt_cycle
  140. int ref_frame_count; ///< num_ref_frames
  141. int gaps_in_frame_num_allowed_flag;
  142. int mb_width; ///< pic_width_in_mbs_minus1 + 1
  143. int mb_height; ///< pic_height_in_map_units_minus1 + 1
  144. int frame_mbs_only_flag;
  145. int mb_aff; ///< mb_adaptive_frame_field_flag
  146. int direct_8x8_inference_flag;
  147. int crop; ///< frame_cropping_flag
  148. unsigned int crop_left; ///< frame_cropping_rect_left_offset
  149. unsigned int crop_right; ///< frame_cropping_rect_right_offset
  150. unsigned int crop_top; ///< frame_cropping_rect_top_offset
  151. unsigned int crop_bottom; ///< frame_cropping_rect_bottom_offset
  152. int vui_parameters_present_flag;
  153. AVRational sar;
  154. int video_signal_type_present_flag;
  155. int full_range;
  156. int colour_description_present_flag;
  157. enum AVColorPrimaries color_primaries;
  158. enum AVColorTransferCharacteristic color_trc;
  159. enum AVColorSpace colorspace;
  160. int timing_info_present_flag;
  161. uint32_t num_units_in_tick;
  162. uint32_t time_scale;
  163. int fixed_frame_rate_flag;
  164. short offset_for_ref_frame[256]; // FIXME dyn aloc?
  165. int bitstream_restriction_flag;
  166. int num_reorder_frames;
  167. int scaling_matrix_present;
  168. uint8_t scaling_matrix4[6][16];
  169. uint8_t scaling_matrix8[6][64];
  170. int nal_hrd_parameters_present_flag;
  171. int vcl_hrd_parameters_present_flag;
  172. int pic_struct_present_flag;
  173. int time_offset_length;
  174. int cpb_cnt; ///< See H.264 E.1.2
  175. int initial_cpb_removal_delay_length; ///< initial_cpb_removal_delay_length_minus1 + 1
  176. int cpb_removal_delay_length; ///< cpb_removal_delay_length_minus1 + 1
  177. int dpb_output_delay_length; ///< dpb_output_delay_length_minus1 + 1
  178. int bit_depth_luma; ///< bit_depth_luma_minus8 + 8
  179. int bit_depth_chroma; ///< bit_depth_chroma_minus8 + 8
  180. int residual_color_transform_flag; ///< residual_colour_transform_flag
  181. int constraint_set_flags; ///< constraint_set[0-3]_flag
  182. int new; ///< flag to keep track if the decoder context needs re-init due to changed SPS
  183. } SPS;
  184. /**
  185. * Picture parameter set
  186. */
  187. typedef struct PPS {
  188. unsigned int sps_id;
  189. int cabac; ///< entropy_coding_mode_flag
  190. int pic_order_present; ///< pic_order_present_flag
  191. int slice_group_count; ///< num_slice_groups_minus1 + 1
  192. int mb_slice_group_map_type;
  193. unsigned int ref_count[2]; ///< num_ref_idx_l0/1_active_minus1 + 1
  194. int weighted_pred; ///< weighted_pred_flag
  195. int weighted_bipred_idc;
  196. int init_qp; ///< pic_init_qp_minus26 + 26
  197. int init_qs; ///< pic_init_qs_minus26 + 26
  198. int chroma_qp_index_offset[2];
  199. int deblocking_filter_parameters_present; ///< deblocking_filter_parameters_present_flag
  200. int constrained_intra_pred; ///< constrained_intra_pred_flag
  201. int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag
  202. int transform_8x8_mode; ///< transform_8x8_mode_flag
  203. uint8_t scaling_matrix4[6][16];
  204. uint8_t scaling_matrix8[6][64];
  205. uint8_t chroma_qp_table[2][64]; ///< pre-scaled (with chroma_qp_index_offset) version of qp_table
  206. int chroma_qp_diff;
  207. } PPS;
  208. /**
  209. * Memory management control operation opcode.
  210. */
  211. typedef enum MMCOOpcode {
  212. MMCO_END = 0,
  213. MMCO_SHORT2UNUSED,
  214. MMCO_LONG2UNUSED,
  215. MMCO_SHORT2LONG,
  216. MMCO_SET_MAX_LONG,
  217. MMCO_RESET,
  218. MMCO_LONG,
  219. } MMCOOpcode;
  220. /**
  221. * Memory management control operation.
  222. */
  223. typedef struct MMCO {
  224. MMCOOpcode opcode;
  225. int short_pic_num; ///< pic_num without wrapping (pic_num & max_pic_num)
  226. int long_arg; ///< index, pic_num, or num long refs depending on opcode
  227. } MMCO;
  228. /**
  229. * H264Context
  230. */
  231. typedef struct H264Context {
  232. AVCodecContext *avctx;
  233. DSPContext dsp;
  234. VideoDSPContext vdsp;
  235. H264DSPContext h264dsp;
  236. H264ChromaContext h264chroma;
  237. H264QpelContext h264qpel;
  238. MotionEstContext me;
  239. ParseContext parse_context;
  240. GetBitContext gb;
  241. ERContext er;
  242. Picture *DPB;
  243. Picture *cur_pic_ptr;
  244. Picture cur_pic;
  245. int pixel_shift; ///< 0 for 8-bit H264, 1 for high-bit-depth H264
  246. int chroma_qp[2]; // QPc
  247. int qp_thresh; ///< QP threshold to skip loopfilter
  248. int width, height;
  249. int linesize, uvlinesize;
  250. int chroma_x_shift, chroma_y_shift;
  251. int qscale;
  252. int droppable;
  253. int data_partitioning;
  254. int coded_picture_number;
  255. int low_delay;
  256. int context_initialized;
  257. int flags;
  258. int workaround_bugs;
  259. int prev_mb_skipped;
  260. int next_mb_skipped;
  261. // prediction stuff
  262. int chroma_pred_mode;
  263. int intra16x16_pred_mode;
  264. int topleft_mb_xy;
  265. int top_mb_xy;
  266. int topright_mb_xy;
  267. int left_mb_xy[LEFT_MBS];
  268. int topleft_type;
  269. int top_type;
  270. int topright_type;
  271. int left_type[LEFT_MBS];
  272. const uint8_t *left_block;
  273. int topleft_partition;
  274. int8_t intra4x4_pred_mode_cache[5 * 8];
  275. int8_t(*intra4x4_pred_mode);
  276. H264PredContext hpc;
  277. unsigned int topleft_samples_available;
  278. unsigned int top_samples_available;
  279. unsigned int topright_samples_available;
  280. unsigned int left_samples_available;
  281. uint8_t (*top_borders[2])[(16 * 3) * 2];
  282. /**
  283. * non zero coeff count cache.
  284. * is 64 if not available.
  285. */
  286. DECLARE_ALIGNED(8, uint8_t, non_zero_count_cache)[15 * 8];
  287. uint8_t (*non_zero_count)[48];
  288. /**
  289. * Motion vector cache.
  290. */
  291. DECLARE_ALIGNED(16, int16_t, mv_cache)[2][5 * 8][2];
  292. DECLARE_ALIGNED(8, int8_t, ref_cache)[2][5 * 8];
  293. #define LIST_NOT_USED -1 // FIXME rename?
  294. #define PART_NOT_AVAILABLE -2
  295. /**
  296. * number of neighbors (top and/or left) that used 8x8 dct
  297. */
  298. int neighbor_transform_size;
  299. /**
  300. * block_offset[ 0..23] for frame macroblocks
  301. * block_offset[24..47] for field macroblocks
  302. */
  303. int block_offset[2 * (16 * 3)];
  304. uint32_t *mb2b_xy; // FIXME are these 4 a good idea?
  305. uint32_t *mb2br_xy;
  306. int b_stride; // FIXME use s->b4_stride
  307. int mb_linesize; ///< may be equal to s->linesize or s->linesize * 2, for mbaff
  308. int mb_uvlinesize;
  309. unsigned current_sps_id; ///< id of the current SPS
  310. SPS sps; ///< current sps
  311. /**
  312. * current pps
  313. */
  314. PPS pps; // FIXME move to Picture perhaps? (->no) do we need that?
  315. uint32_t dequant4_buffer[6][QP_MAX_NUM + 1][16]; // FIXME should these be moved down?
  316. uint32_t dequant8_buffer[6][QP_MAX_NUM + 1][64];
  317. uint32_t(*dequant4_coeff[6])[16];
  318. uint32_t(*dequant8_coeff[6])[64];
  319. int slice_num;
  320. uint16_t *slice_table; ///< slice_table_base + 2*mb_stride + 1
  321. int slice_type;
  322. int slice_type_nos; ///< S free slice type (SI/SP are remapped to I/P)
  323. int slice_type_fixed;
  324. // interlacing specific flags
  325. int mb_aff_frame;
  326. int mb_field_decoding_flag;
  327. int mb_mbaff; ///< mb_aff_frame && mb_field_decoding_flag
  328. int picture_structure;
  329. int first_field;
  330. DECLARE_ALIGNED(8, uint16_t, sub_mb_type)[4];
  331. // Weighted pred stuff
  332. int use_weight;
  333. int use_weight_chroma;
  334. int luma_log2_weight_denom;
  335. int chroma_log2_weight_denom;
  336. // The following 2 can be changed to int8_t but that causes 10cpu cycles speedloss
  337. int luma_weight[48][2][2];
  338. int chroma_weight[48][2][2][2];
  339. int implicit_weight[48][48][2];
  340. int direct_spatial_mv_pred;
  341. int col_parity;
  342. int col_fieldoff;
  343. int dist_scale_factor[32];
  344. int dist_scale_factor_field[2][32];
  345. int map_col_to_list0[2][16 + 32];
  346. int map_col_to_list0_field[2][2][16 + 32];
  347. /**
  348. * num_ref_idx_l0/1_active_minus1 + 1
  349. */
  350. unsigned int ref_count[2]; ///< counts frames or fields, depending on current mb mode
  351. unsigned int list_count;
  352. uint8_t *list_counts; ///< Array of list_count per MB specifying the slice type
  353. Picture ref_list[2][48]; /**< 0..15: frame refs, 16..47: mbaff field refs.
  354. * Reordered version of default_ref_list
  355. * according to picture reordering in slice header */
  356. int ref2frm[MAX_SLICES][2][64]; ///< reference to frame number lists, used in the loop filter, the first 2 are for -2,-1
  357. // data partitioning
  358. GetBitContext intra_gb;
  359. GetBitContext inter_gb;
  360. GetBitContext *intra_gb_ptr;
  361. GetBitContext *inter_gb_ptr;
  362. const uint8_t *intra_pcm_ptr;
  363. DECLARE_ALIGNED(16, int16_t, mb)[16 * 48 * 2]; ///< as a dct coeffecient is int32_t in high depth, we need to reserve twice the space.
  364. DECLARE_ALIGNED(16, int16_t, mb_luma_dc)[3][16 * 2];
  365. int16_t 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
  366. /**
  367. * Cabac
  368. */
  369. CABACContext cabac;
  370. uint8_t cabac_state[1024];
  371. /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0, 1, 2), 0x0? luma_cbp */
  372. uint16_t *cbp_table;
  373. int cbp;
  374. int top_cbp;
  375. int left_cbp;
  376. /* chroma_pred_mode for i4x4 or i16x16, else 0 */
  377. uint8_t *chroma_pred_mode_table;
  378. int last_qscale_diff;
  379. uint8_t (*mvd_table[2])[2];
  380. DECLARE_ALIGNED(16, uint8_t, mvd_cache)[2][5 * 8][2];
  381. uint8_t *direct_table;
  382. uint8_t direct_cache[5 * 8];
  383. uint8_t zigzag_scan[16];
  384. uint8_t zigzag_scan8x8[64];
  385. uint8_t zigzag_scan8x8_cavlc[64];
  386. uint8_t field_scan[16];
  387. uint8_t field_scan8x8[64];
  388. uint8_t field_scan8x8_cavlc[64];
  389. const uint8_t *zigzag_scan_q0;
  390. const uint8_t *zigzag_scan8x8_q0;
  391. const uint8_t *zigzag_scan8x8_cavlc_q0;
  392. const uint8_t *field_scan_q0;
  393. const uint8_t *field_scan8x8_q0;
  394. const uint8_t *field_scan8x8_cavlc_q0;
  395. int x264_build;
  396. int mb_x, mb_y;
  397. int resync_mb_x;
  398. int resync_mb_y;
  399. int mb_skip_run;
  400. int mb_height, mb_width;
  401. int mb_stride;
  402. int mb_num;
  403. int mb_xy;
  404. int is_complex;
  405. // deblock
  406. int deblocking_filter; ///< disable_deblocking_filter_idc with 1 <-> 0
  407. int slice_alpha_c0_offset;
  408. int slice_beta_offset;
  409. // =============================================================
  410. // Things below are not used in the MB or more inner code
  411. int nal_ref_idc;
  412. int nal_unit_type;
  413. uint8_t *rbsp_buffer[2];
  414. unsigned int rbsp_buffer_size[2];
  415. /**
  416. * Used to parse AVC variant of h264
  417. */
  418. int is_avc; ///< this flag is != 0 if codec is avc1
  419. int nal_length_size; ///< Number of bytes used for nal length (1, 2 or 4)
  420. int got_first; ///< this flag is != 0 if we've parsed a frame
  421. int bit_depth_luma; ///< luma bit depth from sps to detect changes
  422. int chroma_format_idc; ///< chroma format from sps to detect changes
  423. SPS *sps_buffers[MAX_SPS_COUNT];
  424. PPS *pps_buffers[MAX_PPS_COUNT];
  425. int dequant_coeff_pps; ///< reinit tables when pps changes
  426. uint16_t *slice_table_base;
  427. // POC stuff
  428. int poc_lsb;
  429. int poc_msb;
  430. int delta_poc_bottom;
  431. int delta_poc[2];
  432. int frame_num;
  433. int prev_poc_msb; ///< poc_msb of the last reference pic for POC type 0
  434. int prev_poc_lsb; ///< poc_lsb of the last reference pic for POC type 0
  435. int frame_num_offset; ///< for POC type 2
  436. int prev_frame_num_offset; ///< for POC type 2
  437. int prev_frame_num; ///< frame_num of the last pic for POC type 1/2
  438. /**
  439. * frame_num for frames or 2 * frame_num + 1 for field pics.
  440. */
  441. int curr_pic_num;
  442. /**
  443. * max_frame_num or 2 * max_frame_num for field pics.
  444. */
  445. int max_pic_num;
  446. int redundant_pic_count;
  447. Picture default_ref_list[2][32]; ///< base reference list for all slices of a coded picture
  448. Picture *short_ref[32];
  449. Picture *long_ref[32];
  450. Picture *delayed_pic[MAX_DELAYED_PIC_COUNT + 2]; // FIXME size?
  451. int last_pocs[MAX_DELAYED_PIC_COUNT];
  452. Picture *next_output_pic;
  453. int outputed_poc;
  454. int next_outputed_poc;
  455. /**
  456. * memory management control operations buffer.
  457. */
  458. MMCO mmco[MAX_MMCO_COUNT];
  459. int mmco_index;
  460. int mmco_reset;
  461. int long_ref_count; ///< number of actual long term references
  462. int short_ref_count; ///< number of actual short term references
  463. int cabac_init_idc;
  464. /**
  465. * @name Members for slice based multithreading
  466. * @{
  467. */
  468. struct H264Context *thread_context[MAX_THREADS];
  469. /**
  470. * current slice number, used to initalize slice_num of each thread/context
  471. */
  472. int current_slice;
  473. /**
  474. * Max number of threads / contexts.
  475. * This is equal to AVCodecContext.thread_count unless
  476. * multithreaded decoding is impossible, in which case it is
  477. * reduced to 1.
  478. */
  479. int max_contexts;
  480. int slice_context_count;
  481. /**
  482. * 1 if the single thread fallback warning has already been
  483. * displayed, 0 otherwise.
  484. */
  485. int single_decode_warning;
  486. enum AVPictureType pict_type;
  487. int last_slice_type;
  488. /** @} */
  489. /**
  490. * pic_struct in picture timing SEI message
  491. */
  492. SEI_PicStructType sei_pic_struct;
  493. /**
  494. * Complement sei_pic_struct
  495. * SEI_PIC_STRUCT_TOP_BOTTOM and SEI_PIC_STRUCT_BOTTOM_TOP indicate interlaced frames.
  496. * However, soft telecined frames may have these values.
  497. * This is used in an attempt to flag soft telecine progressive.
  498. */
  499. int prev_interlaced_frame;
  500. /**
  501. * Bit set of clock types for fields/frames in picture timing SEI message.
  502. * For each found ct_type, appropriate bit is set (e.g., bit 1 for
  503. * interlaced).
  504. */
  505. int sei_ct_type;
  506. /**
  507. * dpb_output_delay in picture timing SEI message, see H.264 C.2.2
  508. */
  509. int sei_dpb_output_delay;
  510. /**
  511. * cpb_removal_delay in picture timing SEI message, see H.264 C.1.2
  512. */
  513. int sei_cpb_removal_delay;
  514. /**
  515. * recovery_frame_cnt from SEI message
  516. *
  517. * Set to -1 if no recovery point SEI message found or to number of frames
  518. * before playback synchronizes. Frames having recovery point are key
  519. * frames.
  520. */
  521. int sei_recovery_frame_cnt;
  522. int luma_weight_flag[2]; ///< 7.4.3.2 luma_weight_lX_flag
  523. int chroma_weight_flag[2]; ///< 7.4.3.2 chroma_weight_lX_flag
  524. // Timestamp stuff
  525. int sei_buffering_period_present; ///< Buffering period SEI flag
  526. int initial_cpb_removal_delay[32]; ///< Initial timestamps for CPBs
  527. int cur_chroma_format_idc;
  528. uint8_t *bipred_scratchpad;
  529. uint8_t *edge_emu_buffer;
  530. int16_t *dc_val_base;
  531. AVBufferPool *qscale_table_pool;
  532. AVBufferPool *mb_type_pool;
  533. AVBufferPool *motion_val_pool;
  534. AVBufferPool *ref_index_pool;
  535. } H264Context;
  536. 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).
  537. extern const uint16_t ff_h264_mb_sizes[4];
  538. /**
  539. * Decode SEI
  540. */
  541. int ff_h264_decode_sei(H264Context *h);
  542. /**
  543. * Decode SPS
  544. */
  545. int ff_h264_decode_seq_parameter_set(H264Context *h);
  546. /**
  547. * compute profile from sps
  548. */
  549. int ff_h264_get_profile(SPS *sps);
  550. /**
  551. * Decode PPS
  552. */
  553. int ff_h264_decode_picture_parameter_set(H264Context *h, int bit_length);
  554. /**
  555. * Decode a network abstraction layer unit.
  556. * @param consumed is the number of bytes used as input
  557. * @param length is the length of the array
  558. * @param dst_length is the number of decoded bytes FIXME here
  559. * or a decode rbsp tailing?
  560. * @return decoded bytes, might be src+1 if no escapes
  561. */
  562. const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src,
  563. int *dst_length, int *consumed, int length);
  564. /**
  565. * Free any data that may have been allocated in the H264 context
  566. * like SPS, PPS etc.
  567. */
  568. void ff_h264_free_context(H264Context *h);
  569. /**
  570. * Reconstruct bitstream slice_type.
  571. */
  572. int ff_h264_get_slice_type(const H264Context *h);
  573. /**
  574. * Allocate tables.
  575. * needs width/height
  576. */
  577. int ff_h264_alloc_tables(H264Context *h);
  578. /**
  579. * Fill the default_ref_list.
  580. */
  581. int ff_h264_fill_default_ref_list(H264Context *h);
  582. int ff_h264_decode_ref_pic_list_reordering(H264Context *h);
  583. void ff_h264_fill_mbaff_ref_list(H264Context *h);
  584. void ff_h264_remove_all_refs(H264Context *h);
  585. /**
  586. * Execute the reference picture marking (memory management control operations).
  587. */
  588. int ff_h264_execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count);
  589. int ff_h264_decode_ref_pic_marking(H264Context *h, GetBitContext *gb,
  590. int first_slice);
  591. int ff_generate_sliding_window_mmcos(H264Context *h, int first_slice);
  592. /**
  593. * Check if the top & left blocks are available if needed & change the
  594. * dc mode so it only uses the available blocks.
  595. */
  596. int ff_h264_check_intra4x4_pred_mode(H264Context *h);
  597. /**
  598. * Check if the top & left blocks are available if needed & change the
  599. * dc mode so it only uses the available blocks.
  600. */
  601. int ff_h264_check_intra_pred_mode(H264Context *h, int mode, int is_chroma);
  602. void ff_h264_hl_decode_mb(H264Context *h);
  603. int ff_h264_frame_start(H264Context *h);
  604. int ff_h264_decode_extradata(H264Context *h);
  605. int ff_h264_decode_init(AVCodecContext *avctx);
  606. void ff_h264_decode_init_vlc(void);
  607. /**
  608. * Decode a macroblock
  609. * @return 0 if OK, ER_AC_ERROR / ER_DC_ERROR / ER_MV_ERROR on error
  610. */
  611. int ff_h264_decode_mb_cavlc(H264Context *h);
  612. /**
  613. * Decode a CABAC coded macroblock
  614. * @return 0 if OK, ER_AC_ERROR / ER_DC_ERROR / ER_MV_ERROR on error
  615. */
  616. int ff_h264_decode_mb_cabac(H264Context *h);
  617. void ff_h264_init_cabac_states(H264Context *h);
  618. void ff_h264_direct_dist_scale_factor(H264Context *const h);
  619. void ff_h264_direct_ref_list_init(H264Context *const h);
  620. void ff_h264_pred_direct_motion(H264Context *const h, int *mb_type);
  621. void ff_h264_filter_mb_fast(H264Context *h, int mb_x, int mb_y,
  622. uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr,
  623. unsigned int linesize, unsigned int uvlinesize);
  624. void ff_h264_filter_mb(H264Context *h, int mb_x, int mb_y,
  625. uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr,
  626. unsigned int linesize, unsigned int uvlinesize);
  627. /**
  628. * Reset SEI values at the beginning of the frame.
  629. *
  630. * @param h H.264 context.
  631. */
  632. void ff_h264_reset_sei(H264Context *h);
  633. /*
  634. * o-o o-o
  635. * / / /
  636. * o-o o-o
  637. * ,---'
  638. * o-o o-o
  639. * / / /
  640. * o-o o-o
  641. */
  642. /* Scan8 organization:
  643. * 0 1 2 3 4 5 6 7
  644. * 0 DY y y y y y
  645. * 1 y Y Y Y Y
  646. * 2 y Y Y Y Y
  647. * 3 y Y Y Y Y
  648. * 4 y Y Y Y Y
  649. * 5 DU u u u u u
  650. * 6 u U U U U
  651. * 7 u U U U U
  652. * 8 u U U U U
  653. * 9 u U U U U
  654. * 10 DV v v v v v
  655. * 11 v V V V V
  656. * 12 v V V V V
  657. * 13 v V V V V
  658. * 14 v V V V V
  659. * DY/DU/DV are for luma/chroma DC.
  660. */
  661. #define LUMA_DC_BLOCK_INDEX 48
  662. #define CHROMA_DC_BLOCK_INDEX 49
  663. // This table must be here because scan8[constant] must be known at compiletime
  664. static const uint8_t scan8[16 * 3 + 3] = {
  665. 4 + 1 * 8, 5 + 1 * 8, 4 + 2 * 8, 5 + 2 * 8,
  666. 6 + 1 * 8, 7 + 1 * 8, 6 + 2 * 8, 7 + 2 * 8,
  667. 4 + 3 * 8, 5 + 3 * 8, 4 + 4 * 8, 5 + 4 * 8,
  668. 6 + 3 * 8, 7 + 3 * 8, 6 + 4 * 8, 7 + 4 * 8,
  669. 4 + 6 * 8, 5 + 6 * 8, 4 + 7 * 8, 5 + 7 * 8,
  670. 6 + 6 * 8, 7 + 6 * 8, 6 + 7 * 8, 7 + 7 * 8,
  671. 4 + 8 * 8, 5 + 8 * 8, 4 + 9 * 8, 5 + 9 * 8,
  672. 6 + 8 * 8, 7 + 8 * 8, 6 + 9 * 8, 7 + 9 * 8,
  673. 4 + 11 * 8, 5 + 11 * 8, 4 + 12 * 8, 5 + 12 * 8,
  674. 6 + 11 * 8, 7 + 11 * 8, 6 + 12 * 8, 7 + 12 * 8,
  675. 4 + 13 * 8, 5 + 13 * 8, 4 + 14 * 8, 5 + 14 * 8,
  676. 6 + 13 * 8, 7 + 13 * 8, 6 + 14 * 8, 7 + 14 * 8,
  677. 0 + 0 * 8, 0 + 5 * 8, 0 + 10 * 8
  678. };
  679. static av_always_inline uint32_t pack16to32(int a, int b)
  680. {
  681. #if HAVE_BIGENDIAN
  682. return (b & 0xFFFF) + (a << 16);
  683. #else
  684. return (a & 0xFFFF) + (b << 16);
  685. #endif
  686. }
  687. static av_always_inline uint16_t pack8to16(int a, int b)
  688. {
  689. #if HAVE_BIGENDIAN
  690. return (b & 0xFF) + (a << 8);
  691. #else
  692. return (a & 0xFF) + (b << 8);
  693. #endif
  694. }
  695. /**
  696. * Get the chroma qp.
  697. */
  698. static av_always_inline int get_chroma_qp(H264Context *h, int t, int qscale)
  699. {
  700. return h->pps.chroma_qp_table[t][qscale];
  701. }
  702. /**
  703. * Get the predicted intra4x4 prediction mode.
  704. */
  705. static av_always_inline int pred_intra_mode(H264Context *h, int n)
  706. {
  707. const int index8 = scan8[n];
  708. const int left = h->intra4x4_pred_mode_cache[index8 - 1];
  709. const int top = h->intra4x4_pred_mode_cache[index8 - 8];
  710. const int min = FFMIN(left, top);
  711. tprintf(h->avctx, "mode:%d %d min:%d\n", left, top, min);
  712. if (min < 0)
  713. return DC_PRED;
  714. else
  715. return min;
  716. }
  717. static av_always_inline void write_back_intra_pred_mode(H264Context *h)
  718. {
  719. int8_t *i4x4 = h->intra4x4_pred_mode + h->mb2br_xy[h->mb_xy];
  720. int8_t *i4x4_cache = h->intra4x4_pred_mode_cache;
  721. AV_COPY32(i4x4, i4x4_cache + 4 + 8 * 4);
  722. i4x4[4] = i4x4_cache[7 + 8 * 3];
  723. i4x4[5] = i4x4_cache[7 + 8 * 2];
  724. i4x4[6] = i4x4_cache[7 + 8 * 1];
  725. }
  726. static av_always_inline void write_back_non_zero_count(H264Context *h)
  727. {
  728. const int mb_xy = h->mb_xy;
  729. uint8_t *nnz = h->non_zero_count[mb_xy];
  730. uint8_t *nnz_cache = h->non_zero_count_cache;
  731. AV_COPY32(&nnz[ 0], &nnz_cache[4 + 8 * 1]);
  732. AV_COPY32(&nnz[ 4], &nnz_cache[4 + 8 * 2]);
  733. AV_COPY32(&nnz[ 8], &nnz_cache[4 + 8 * 3]);
  734. AV_COPY32(&nnz[12], &nnz_cache[4 + 8 * 4]);
  735. AV_COPY32(&nnz[16], &nnz_cache[4 + 8 * 6]);
  736. AV_COPY32(&nnz[20], &nnz_cache[4 + 8 * 7]);
  737. AV_COPY32(&nnz[32], &nnz_cache[4 + 8 * 11]);
  738. AV_COPY32(&nnz[36], &nnz_cache[4 + 8 * 12]);
  739. if (!h->chroma_y_shift) {
  740. AV_COPY32(&nnz[24], &nnz_cache[4 + 8 * 8]);
  741. AV_COPY32(&nnz[28], &nnz_cache[4 + 8 * 9]);
  742. AV_COPY32(&nnz[40], &nnz_cache[4 + 8 * 13]);
  743. AV_COPY32(&nnz[44], &nnz_cache[4 + 8 * 14]);
  744. }
  745. }
  746. static av_always_inline void write_back_motion_list(H264Context *h,
  747. int b_stride,
  748. int b_xy, int b8_xy,
  749. int mb_type, int list)
  750. {
  751. int16_t(*mv_dst)[2] = &h->cur_pic.motion_val[list][b_xy];
  752. int16_t(*mv_src)[2] = &h->mv_cache[list][scan8[0]];
  753. AV_COPY128(mv_dst + 0 * b_stride, mv_src + 8 * 0);
  754. AV_COPY128(mv_dst + 1 * b_stride, mv_src + 8 * 1);
  755. AV_COPY128(mv_dst + 2 * b_stride, mv_src + 8 * 2);
  756. AV_COPY128(mv_dst + 3 * b_stride, mv_src + 8 * 3);
  757. if (CABAC) {
  758. uint8_t (*mvd_dst)[2] = &h->mvd_table[list][FMO ? 8 * h->mb_xy
  759. : h->mb2br_xy[h->mb_xy]];
  760. uint8_t(*mvd_src)[2] = &h->mvd_cache[list][scan8[0]];
  761. if (IS_SKIP(mb_type)) {
  762. AV_ZERO128(mvd_dst);
  763. } else {
  764. AV_COPY64(mvd_dst, mvd_src + 8 * 3);
  765. AV_COPY16(mvd_dst + 3 + 3, mvd_src + 3 + 8 * 0);
  766. AV_COPY16(mvd_dst + 3 + 2, mvd_src + 3 + 8 * 1);
  767. AV_COPY16(mvd_dst + 3 + 1, mvd_src + 3 + 8 * 2);
  768. }
  769. }
  770. {
  771. int8_t *ref_index = &h->cur_pic.ref_index[list][b8_xy];
  772. int8_t *ref_cache = h->ref_cache[list];
  773. ref_index[0 + 0 * 2] = ref_cache[scan8[0]];
  774. ref_index[1 + 0 * 2] = ref_cache[scan8[4]];
  775. ref_index[0 + 1 * 2] = ref_cache[scan8[8]];
  776. ref_index[1 + 1 * 2] = ref_cache[scan8[12]];
  777. }
  778. }
  779. static av_always_inline void write_back_motion(H264Context *h, int mb_type)
  780. {
  781. const int b_stride = h->b_stride;
  782. const int b_xy = 4 * h->mb_x + 4 * h->mb_y * h->b_stride; // try mb2b(8)_xy
  783. const int b8_xy = 4 * h->mb_xy;
  784. if (USES_LIST(mb_type, 0)) {
  785. write_back_motion_list(h, b_stride, b_xy, b8_xy, mb_type, 0);
  786. } else {
  787. fill_rectangle(&h->cur_pic.ref_index[0][b8_xy],
  788. 2, 2, 2, (uint8_t)LIST_NOT_USED, 1);
  789. }
  790. if (USES_LIST(mb_type, 1))
  791. write_back_motion_list(h, b_stride, b_xy, b8_xy, mb_type, 1);
  792. if (h->slice_type_nos == AV_PICTURE_TYPE_B && CABAC) {
  793. if (IS_8X8(mb_type)) {
  794. uint8_t *direct_table = &h->direct_table[4 * h->mb_xy];
  795. direct_table[1] = h->sub_mb_type[1] >> 1;
  796. direct_table[2] = h->sub_mb_type[2] >> 1;
  797. direct_table[3] = h->sub_mb_type[3] >> 1;
  798. }
  799. }
  800. }
  801. static av_always_inline int get_dct8x8_allowed(H264Context *h)
  802. {
  803. if (h->sps.direct_8x8_inference_flag)
  804. return !(AV_RN64A(h->sub_mb_type) &
  805. ((MB_TYPE_16x8 | MB_TYPE_8x16 | MB_TYPE_8x8) *
  806. 0x0001000100010001ULL));
  807. else
  808. return !(AV_RN64A(h->sub_mb_type) &
  809. ((MB_TYPE_16x8 | MB_TYPE_8x16 | MB_TYPE_8x8 | MB_TYPE_DIRECT2) *
  810. 0x0001000100010001ULL));
  811. }
  812. void ff_h264_draw_horiz_band(H264Context *h, int y, int height);
  813. #endif /* AVCODEC_H264_H */