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.

931 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/buffer.h"
  29. #include "libavutil/intreadwrite.h"
  30. #include "libavutil/thread.h"
  31. #include "cabac.h"
  32. #include "error_resilience.h"
  33. #include "get_bits.h"
  34. #include "h264_parse.h"
  35. #include "h264_sei.h"
  36. #include "h2645_parse.h"
  37. #include "h264chroma.h"
  38. #include "h264dsp.h"
  39. #include "h264pred.h"
  40. #include "h264qpel.h"
  41. #include "internal.h"
  42. #include "mpegutils.h"
  43. #include "parser.h"
  44. #include "qpeldsp.h"
  45. #include "rectangle.h"
  46. #include "videodsp.h"
  47. #define H264_MAX_PICTURE_COUNT 32
  48. #define MAX_SPS_COUNT 32
  49. #define MAX_PPS_COUNT 256
  50. #define MAX_MMCO_COUNT 66
  51. #define MAX_DELAYED_PIC_COUNT 16
  52. /* Compiling in interlaced support reduces the speed
  53. * of progressive decoding by about 2%. */
  54. #define ALLOW_INTERLACE
  55. #define FMO 0
  56. /**
  57. * The maximum number of slices supported by the decoder.
  58. * must be a power of 2
  59. */
  60. #define MAX_SLICES 32
  61. #ifdef ALLOW_INTERLACE
  62. #define MB_MBAFF(h) h->mb_mbaff
  63. #define MB_FIELD(h) h->mb_field_decoding_flag
  64. #define FRAME_MBAFF(h) h->mb_aff_frame
  65. #define FIELD_PICTURE(h) (h->picture_structure != PICT_FRAME)
  66. #define LEFT_MBS 2
  67. #define LTOP 0
  68. #define LBOT 1
  69. #define LEFT(i) (i)
  70. #else
  71. #define MB_MBAFF(h) 0
  72. #define MB_FIELD(h) 0
  73. #define FRAME_MBAFF(h) 0
  74. #define FIELD_PICTURE(h) 0
  75. #undef IS_INTERLACED
  76. #define IS_INTERLACED(mb_type) 0
  77. #define LEFT_MBS 1
  78. #define LTOP 0
  79. #define LBOT 0
  80. #define LEFT(i) 0
  81. #endif
  82. #define FIELD_OR_MBAFF_PICTURE(h) (FRAME_MBAFF(h) || FIELD_PICTURE(h))
  83. #ifndef CABAC
  84. #define CABAC(h) h->ps.pps->cabac
  85. #endif
  86. #define CHROMA422(h) (h->ps.sps->chroma_format_idc == 2)
  87. #define CHROMA444(h) (h->ps.sps->chroma_format_idc == 3)
  88. #define EXTENDED_SAR 255
  89. #define MB_TYPE_REF0 MB_TYPE_ACPRED // dirty but it fits in 16 bit
  90. #define MB_TYPE_8x8DCT 0x01000000
  91. #define IS_REF0(a) ((a) & MB_TYPE_REF0)
  92. #define IS_8x8DCT(a) ((a) & MB_TYPE_8x8DCT)
  93. #define QP_MAX_NUM (51 + 2 * 6) // The maximum supported qp
  94. /* NAL unit types */
  95. enum {
  96. NAL_SLICE = 1,
  97. NAL_DPA = 2,
  98. NAL_DPB = 3,
  99. NAL_DPC = 4,
  100. NAL_IDR_SLICE = 5,
  101. NAL_SEI = 6,
  102. NAL_SPS = 7,
  103. NAL_PPS = 8,
  104. NAL_AUD = 9,
  105. NAL_END_SEQUENCE = 10,
  106. NAL_END_STREAM = 11,
  107. NAL_FILLER_DATA = 12,
  108. NAL_SPS_EXT = 13,
  109. NAL_AUXILIARY_SLICE = 19,
  110. NAL_FF_IGNORE = 0xff0f001,
  111. };
  112. /**
  113. * Sequence parameter set
  114. */
  115. typedef struct SPS {
  116. unsigned int sps_id;
  117. int profile_idc;
  118. int level_idc;
  119. int chroma_format_idc;
  120. int transform_bypass; ///< qpprime_y_zero_transform_bypass_flag
  121. int log2_max_frame_num; ///< log2_max_frame_num_minus4 + 4
  122. int poc_type; ///< pic_order_cnt_type
  123. int log2_max_poc_lsb; ///< log2_max_pic_order_cnt_lsb_minus4
  124. int delta_pic_order_always_zero_flag;
  125. int offset_for_non_ref_pic;
  126. int offset_for_top_to_bottom_field;
  127. int poc_cycle_length; ///< num_ref_frames_in_pic_order_cnt_cycle
  128. int ref_frame_count; ///< num_ref_frames
  129. int gaps_in_frame_num_allowed_flag;
  130. int mb_width; ///< pic_width_in_mbs_minus1 + 1
  131. int mb_height; ///< pic_height_in_map_units_minus1 + 1
  132. int frame_mbs_only_flag;
  133. int mb_aff; ///< mb_adaptive_frame_field_flag
  134. int direct_8x8_inference_flag;
  135. int crop; ///< frame_cropping_flag
  136. /* those 4 are already in luma samples */
  137. unsigned int crop_left; ///< frame_cropping_rect_left_offset
  138. unsigned int crop_right; ///< frame_cropping_rect_right_offset
  139. unsigned int crop_top; ///< frame_cropping_rect_top_offset
  140. unsigned int crop_bottom; ///< frame_cropping_rect_bottom_offset
  141. int vui_parameters_present_flag;
  142. AVRational sar;
  143. int video_signal_type_present_flag;
  144. int full_range;
  145. int colour_description_present_flag;
  146. enum AVColorPrimaries color_primaries;
  147. enum AVColorTransferCharacteristic color_trc;
  148. enum AVColorSpace colorspace;
  149. int timing_info_present_flag;
  150. uint32_t num_units_in_tick;
  151. uint32_t time_scale;
  152. int fixed_frame_rate_flag;
  153. short offset_for_ref_frame[256]; // FIXME dyn aloc?
  154. int bitstream_restriction_flag;
  155. int num_reorder_frames;
  156. int scaling_matrix_present;
  157. uint8_t scaling_matrix4[6][16];
  158. uint8_t scaling_matrix8[6][64];
  159. int nal_hrd_parameters_present_flag;
  160. int vcl_hrd_parameters_present_flag;
  161. int pic_struct_present_flag;
  162. int time_offset_length;
  163. int cpb_cnt; ///< See H.264 E.1.2
  164. int initial_cpb_removal_delay_length; ///< initial_cpb_removal_delay_length_minus1 + 1
  165. int cpb_removal_delay_length; ///< cpb_removal_delay_length_minus1 + 1
  166. int dpb_output_delay_length; ///< dpb_output_delay_length_minus1 + 1
  167. int bit_depth_luma; ///< bit_depth_luma_minus8 + 8
  168. int bit_depth_chroma; ///< bit_depth_chroma_minus8 + 8
  169. int residual_color_transform_flag; ///< residual_colour_transform_flag
  170. int constraint_set_flags; ///< constraint_set[0-3]_flag
  171. } SPS;
  172. /**
  173. * Picture parameter set
  174. */
  175. typedef struct PPS {
  176. unsigned int sps_id;
  177. int cabac; ///< entropy_coding_mode_flag
  178. int pic_order_present; ///< pic_order_present_flag
  179. int slice_group_count; ///< num_slice_groups_minus1 + 1
  180. int mb_slice_group_map_type;
  181. unsigned int ref_count[2]; ///< num_ref_idx_l0/1_active_minus1 + 1
  182. int weighted_pred; ///< weighted_pred_flag
  183. int weighted_bipred_idc;
  184. int init_qp; ///< pic_init_qp_minus26 + 26
  185. int init_qs; ///< pic_init_qs_minus26 + 26
  186. int chroma_qp_index_offset[2];
  187. int deblocking_filter_parameters_present; ///< deblocking_filter_parameters_present_flag
  188. int constrained_intra_pred; ///< constrained_intra_pred_flag
  189. int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag
  190. int transform_8x8_mode; ///< transform_8x8_mode_flag
  191. uint8_t scaling_matrix4[6][16];
  192. uint8_t scaling_matrix8[6][64];
  193. uint8_t chroma_qp_table[2][64]; ///< pre-scaled (with chroma_qp_index_offset) version of qp_table
  194. int chroma_qp_diff;
  195. uint32_t dequant4_buffer[6][QP_MAX_NUM + 1][16];
  196. uint32_t dequant8_buffer[6][QP_MAX_NUM + 1][64];
  197. uint32_t(*dequant4_coeff[6])[16];
  198. uint32_t(*dequant8_coeff[6])[64];
  199. } PPS;
  200. typedef struct H264ParamSets {
  201. AVBufferRef *sps_list[MAX_SPS_COUNT];
  202. AVBufferRef *pps_list[MAX_PPS_COUNT];
  203. /* currently active parameters sets */
  204. const PPS *pps;
  205. // FIXME this should properly be const
  206. SPS *sps;
  207. } H264ParamSets;
  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. typedef struct H264Picture {
  229. AVFrame *f;
  230. ThreadFrame tf;
  231. AVBufferRef *qscale_table_buf;
  232. int8_t *qscale_table;
  233. AVBufferRef *motion_val_buf[2];
  234. int16_t (*motion_val[2])[2];
  235. AVBufferRef *mb_type_buf;
  236. uint32_t *mb_type;
  237. AVBufferRef *hwaccel_priv_buf;
  238. void *hwaccel_picture_private; ///< hardware accelerator private data
  239. AVBufferRef *ref_index_buf[2];
  240. int8_t *ref_index[2];
  241. int field_poc[2]; ///< top/bottom POC
  242. int poc; ///< frame POC
  243. int frame_num; ///< frame_num (raw frame_num from slice header)
  244. int mmco_reset; /**< MMCO_RESET set this 1. Reordering code must
  245. not mix pictures before and after MMCO_RESET. */
  246. int pic_id; /**< pic_num (short -> no wrap version of pic_num,
  247. pic_num & max_pic_num; long -> long_pic_num) */
  248. int long_ref; ///< 1->long term reference 0->short term reference
  249. int ref_poc[2][2][32]; ///< POCs of the frames used as reference (FIXME need per slice)
  250. int ref_count[2][2]; ///< number of entries in ref_poc (FIXME need per slice)
  251. int mbaff; ///< 1 -> MBAFF frame 0-> not MBAFF
  252. int field_picture; ///< whether or not picture was encoded in separate fields
  253. int reference;
  254. int recovered; ///< picture at IDR or recovery point + recovery count
  255. } H264Picture;
  256. typedef struct H264Ref {
  257. uint8_t *data[3];
  258. int linesize[3];
  259. int reference;
  260. int poc;
  261. int pic_id;
  262. H264Picture *parent;
  263. } H264Ref;
  264. typedef struct H264SliceContext {
  265. struct H264Context *h264;
  266. GetBitContext gb;
  267. ERContext er;
  268. int slice_num;
  269. int slice_type;
  270. int slice_type_nos; ///< S free slice type (SI/SP are remapped to I/P)
  271. int slice_type_fixed;
  272. int qscale;
  273. int chroma_qp[2]; // QPc
  274. int qp_thresh; ///< QP threshold to skip loopfilter
  275. int last_qscale_diff;
  276. // deblock
  277. int deblocking_filter; ///< disable_deblocking_filter_idc with 1 <-> 0
  278. int slice_alpha_c0_offset;
  279. int slice_beta_offset;
  280. H264PredWeightTable pwt;
  281. int prev_mb_skipped;
  282. int next_mb_skipped;
  283. int chroma_pred_mode;
  284. int intra16x16_pred_mode;
  285. int8_t intra4x4_pred_mode_cache[5 * 8];
  286. int8_t(*intra4x4_pred_mode);
  287. int topleft_mb_xy;
  288. int top_mb_xy;
  289. int topright_mb_xy;
  290. int left_mb_xy[LEFT_MBS];
  291. int topleft_type;
  292. int top_type;
  293. int topright_type;
  294. int left_type[LEFT_MBS];
  295. const uint8_t *left_block;
  296. int topleft_partition;
  297. unsigned int topleft_samples_available;
  298. unsigned int top_samples_available;
  299. unsigned int topright_samples_available;
  300. unsigned int left_samples_available;
  301. ptrdiff_t linesize, uvlinesize;
  302. ptrdiff_t mb_linesize; ///< may be equal to s->linesize or s->linesize * 2, for mbaff
  303. ptrdiff_t mb_uvlinesize;
  304. int mb_x, mb_y;
  305. int mb_xy;
  306. int resync_mb_x;
  307. int resync_mb_y;
  308. // index of the first MB of the next slice
  309. int next_slice_idx;
  310. int mb_skip_run;
  311. int is_complex;
  312. int mb_field_decoding_flag;
  313. int mb_mbaff; ///< mb_aff_frame && mb_field_decoding_flag
  314. int redundant_pic_count;
  315. /**
  316. * number of neighbors (top and/or left) that used 8x8 dct
  317. */
  318. int neighbor_transform_size;
  319. int direct_spatial_mv_pred;
  320. int col_parity;
  321. int col_fieldoff;
  322. int cbp;
  323. int top_cbp;
  324. int left_cbp;
  325. int dist_scale_factor[32];
  326. int dist_scale_factor_field[2][32];
  327. int map_col_to_list0[2][16 + 32];
  328. int map_col_to_list0_field[2][2][16 + 32];
  329. /**
  330. * num_ref_idx_l0/1_active_minus1 + 1
  331. */
  332. unsigned int ref_count[2]; ///< counts frames or fields, depending on current mb mode
  333. unsigned int list_count;
  334. H264Ref 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. const uint8_t *intra_pcm_ptr;
  338. int16_t *dc_val_base;
  339. uint8_t *bipred_scratchpad;
  340. uint8_t *edge_emu_buffer;
  341. uint8_t (*top_borders[2])[(16 * 3) * 2];
  342. int bipred_scratchpad_allocated;
  343. int edge_emu_buffer_allocated;
  344. int top_borders_allocated[2];
  345. /**
  346. * non zero coeff count cache.
  347. * is 64 if not available.
  348. */
  349. DECLARE_ALIGNED(8, uint8_t, non_zero_count_cache)[15 * 8];
  350. /**
  351. * Motion vector cache.
  352. */
  353. DECLARE_ALIGNED(16, int16_t, mv_cache)[2][5 * 8][2];
  354. DECLARE_ALIGNED(8, int8_t, ref_cache)[2][5 * 8];
  355. DECLARE_ALIGNED(16, uint8_t, mvd_cache)[2][5 * 8][2];
  356. uint8_t direct_cache[5 * 8];
  357. DECLARE_ALIGNED(8, uint16_t, sub_mb_type)[4];
  358. ///< as a dct coeffecient is int32_t in high depth, we need to reserve twice the space.
  359. DECLARE_ALIGNED(16, int16_t, mb)[16 * 48 * 2];
  360. DECLARE_ALIGNED(16, int16_t, mb_luma_dc)[3][16 * 2];
  361. ///< as mb is addressed by scantable[i] and scantable is uint8_t we can either
  362. ///< check that i is not too large or ensure that there is some unused stuff after mb
  363. int16_t mb_padding[256 * 2];
  364. uint8_t (*mvd_table[2])[2];
  365. /**
  366. * Cabac
  367. */
  368. CABACContext cabac;
  369. uint8_t cabac_state[1024];
  370. int cabac_init_idc;
  371. } H264SliceContext;
  372. /**
  373. * H264Context
  374. */
  375. typedef struct H264Context {
  376. const AVClass *class;
  377. AVCodecContext *avctx;
  378. VideoDSPContext vdsp;
  379. H264DSPContext h264dsp;
  380. H264ChromaContext h264chroma;
  381. H264QpelContext h264qpel;
  382. GetBitContext gb;
  383. H264Picture DPB[H264_MAX_PICTURE_COUNT];
  384. H264Picture *cur_pic_ptr;
  385. H264Picture cur_pic;
  386. H264SliceContext *slice_ctx;
  387. int nb_slice_ctx;
  388. H2645Packet pkt;
  389. int pixel_shift; ///< 0 for 8-bit H264, 1 for high-bit-depth H264
  390. /* coded dimensions -- 16 * mb w/h */
  391. int width, height;
  392. int chroma_x_shift, chroma_y_shift;
  393. int droppable;
  394. int coded_picture_number;
  395. int context_initialized;
  396. int flags;
  397. int workaround_bugs;
  398. /* Set when slice threading is used and at least one slice uses deblocking
  399. * mode 1 (i.e. across slice boundaries). Then we disable the loop filter
  400. * during normal MB decoding and execute it serially at the end.
  401. */
  402. int postpone_filter;
  403. int8_t(*intra4x4_pred_mode);
  404. H264PredContext hpc;
  405. uint8_t (*non_zero_count)[48];
  406. #define LIST_NOT_USED -1 // FIXME rename?
  407. #define PART_NOT_AVAILABLE -2
  408. /**
  409. * block_offset[ 0..23] for frame macroblocks
  410. * block_offset[24..47] for field macroblocks
  411. */
  412. int block_offset[2 * (16 * 3)];
  413. uint32_t *mb2b_xy; // FIXME are these 4 a good idea?
  414. uint32_t *mb2br_xy;
  415. int b_stride; // FIXME use s->b4_stride
  416. uint16_t *slice_table; ///< slice_table_base + 2*mb_stride + 1
  417. // interlacing specific flags
  418. int mb_aff_frame;
  419. int picture_structure;
  420. int first_field;
  421. uint8_t *list_counts; ///< Array of list_count per MB specifying the slice type
  422. /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0, 1, 2), 0x0? luma_cbp */
  423. uint16_t *cbp_table;
  424. /* chroma_pred_mode for i4x4 or i16x16, else 0 */
  425. uint8_t *chroma_pred_mode_table;
  426. uint8_t (*mvd_table[2])[2];
  427. uint8_t *direct_table;
  428. uint8_t zigzag_scan[16];
  429. uint8_t zigzag_scan8x8[64];
  430. uint8_t zigzag_scan8x8_cavlc[64];
  431. uint8_t field_scan[16];
  432. uint8_t field_scan8x8[64];
  433. uint8_t field_scan8x8_cavlc[64];
  434. const uint8_t *zigzag_scan_q0;
  435. const uint8_t *zigzag_scan8x8_q0;
  436. const uint8_t *zigzag_scan8x8_cavlc_q0;
  437. const uint8_t *field_scan_q0;
  438. const uint8_t *field_scan8x8_q0;
  439. const uint8_t *field_scan8x8_cavlc_q0;
  440. int mb_y;
  441. int mb_height, mb_width;
  442. int mb_stride;
  443. int mb_num;
  444. // =============================================================
  445. // Things below are not used in the MB or more inner code
  446. int nal_ref_idc;
  447. int nal_unit_type;
  448. /**
  449. * Used to parse AVC variant of h264
  450. */
  451. int is_avc; ///< this flag is != 0 if codec is avc1
  452. int nal_length_size; ///< Number of bytes used for nal length (1, 2 or 4)
  453. int bit_depth_luma; ///< luma bit depth from sps to detect changes
  454. int chroma_format_idc; ///< chroma format from sps to detect changes
  455. H264ParamSets ps;
  456. uint16_t *slice_table_base;
  457. H264POCContext poc;
  458. /**
  459. * frame_num for frames or 2 * frame_num + 1 for field pics.
  460. */
  461. int curr_pic_num;
  462. /**
  463. * max_frame_num or 2 * max_frame_num for field pics.
  464. */
  465. int max_pic_num;
  466. H264Picture *short_ref[32];
  467. H264Picture *long_ref[32];
  468. H264Picture *delayed_pic[MAX_DELAYED_PIC_COUNT + 2]; // FIXME size?
  469. int last_pocs[MAX_DELAYED_PIC_COUNT];
  470. H264Picture *next_output_pic;
  471. int next_outputed_poc;
  472. /**
  473. * memory management control operations buffer.
  474. */
  475. MMCO mmco[MAX_MMCO_COUNT];
  476. int mmco_index;
  477. int mmco_reset;
  478. int long_ref_count; ///< number of actual long term references
  479. int short_ref_count; ///< number of actual short term references
  480. /**
  481. * @name Members for slice based multithreading
  482. * @{
  483. */
  484. /**
  485. * current slice number, used to initalize slice_num of each thread/context
  486. */
  487. int current_slice;
  488. /** @} */
  489. /**
  490. * Complement sei_pic_struct
  491. * SEI_PIC_STRUCT_TOP_BOTTOM and SEI_PIC_STRUCT_BOTTOM_TOP indicate interlaced frames.
  492. * However, soft telecined frames may have these values.
  493. * This is used in an attempt to flag soft telecine progressive.
  494. */
  495. int prev_interlaced_frame;
  496. /**
  497. * recovery_frame is the frame_num at which the next frame should
  498. * be fully constructed.
  499. *
  500. * Set to -1 when not expecting a recovery point.
  501. */
  502. int recovery_frame;
  503. /**
  504. * We have seen an IDR, so all the following frames in coded order are correctly
  505. * decodable.
  506. */
  507. #define FRAME_RECOVERED_IDR (1 << 0)
  508. /**
  509. * Sufficient number of frames have been decoded since a SEI recovery point,
  510. * so all the following frames in presentation order are correct.
  511. */
  512. #define FRAME_RECOVERED_SEI (1 << 1)
  513. int frame_recovered; ///< Initial frame has been completely recovered
  514. /* for frame threading, this is set to 1
  515. * after finish_setup() has been called, so we cannot modify
  516. * some context properties (which are supposed to stay constant between
  517. * slices) anymore */
  518. int setup_finished;
  519. int enable_er;
  520. H264SEIContext sei;
  521. AVBufferPool *qscale_table_pool;
  522. AVBufferPool *mb_type_pool;
  523. AVBufferPool *motion_val_pool;
  524. AVBufferPool *ref_index_pool;
  525. int ref2frm[MAX_SLICES][2][64]; ///< reference to frame number lists, used in the loop filter, the first 2 are for -2,-1
  526. } H264Context;
  527. extern const uint16_t ff_h264_mb_sizes[4];
  528. /**
  529. * Decode SPS
  530. */
  531. int ff_h264_decode_seq_parameter_set(GetBitContext *gb, AVCodecContext *avctx,
  532. H264ParamSets *ps);
  533. /**
  534. * Decode PPS
  535. */
  536. int ff_h264_decode_picture_parameter_set(GetBitContext *gb, AVCodecContext *avctx,
  537. H264ParamSets *ps, int bit_length);
  538. /**
  539. * Reconstruct bitstream slice_type.
  540. */
  541. int ff_h264_get_slice_type(const H264SliceContext *sl);
  542. /**
  543. * Allocate tables.
  544. * needs width/height
  545. */
  546. int ff_h264_alloc_tables(H264Context *h);
  547. int ff_h264_decode_ref_pic_list_reordering(const H264Context *h, H264SliceContext *sl);
  548. void ff_h264_fill_mbaff_ref_list(H264SliceContext *sl);
  549. void ff_h264_remove_all_refs(H264Context *h);
  550. /**
  551. * Execute the reference picture marking (memory management control operations).
  552. */
  553. int ff_h264_execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count);
  554. int ff_h264_decode_ref_pic_marking(H264Context *h, GetBitContext *gb,
  555. int first_slice);
  556. int ff_generate_sliding_window_mmcos(H264Context *h, int first_slice);
  557. void ff_h264_hl_decode_mb(const H264Context *h, H264SliceContext *sl);
  558. int ff_h264_decode_init(AVCodecContext *avctx);
  559. void ff_h264_decode_init_vlc(void);
  560. /**
  561. * Decode a macroblock
  562. * @return 0 if OK, ER_AC_ERROR / ER_DC_ERROR / ER_MV_ERROR on error
  563. */
  564. int ff_h264_decode_mb_cavlc(const H264Context *h, H264SliceContext *sl);
  565. /**
  566. * Decode a CABAC coded macroblock
  567. * @return 0 if OK, ER_AC_ERROR / ER_DC_ERROR / ER_MV_ERROR on error
  568. */
  569. int ff_h264_decode_mb_cabac(const H264Context *h, H264SliceContext *sl);
  570. void ff_h264_init_cabac_states(const H264Context *h, H264SliceContext *sl);
  571. void ff_h264_init_dequant_tables(H264Context *h);
  572. void ff_h264_direct_dist_scale_factor(const H264Context *const h, H264SliceContext *sl);
  573. void ff_h264_direct_ref_list_init(const H264Context *const h, H264SliceContext *sl);
  574. void ff_h264_pred_direct_motion(const H264Context *const h, H264SliceContext *sl,
  575. int *mb_type);
  576. void ff_h264_filter_mb_fast(const H264Context *h, H264SliceContext *sl, int mb_x, int mb_y,
  577. uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr,
  578. unsigned int linesize, unsigned int uvlinesize);
  579. void ff_h264_filter_mb(const H264Context *h, H264SliceContext *sl, int mb_x, int mb_y,
  580. uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr,
  581. unsigned int linesize, unsigned int uvlinesize);
  582. /*
  583. * o-o o-o
  584. * / / /
  585. * o-o o-o
  586. * ,---'
  587. * o-o o-o
  588. * / / /
  589. * o-o o-o
  590. */
  591. /* Scan8 organization:
  592. * 0 1 2 3 4 5 6 7
  593. * 0 DY y y y y y
  594. * 1 y Y Y Y Y
  595. * 2 y Y Y Y Y
  596. * 3 y Y Y Y Y
  597. * 4 y Y Y Y Y
  598. * 5 DU u u u u u
  599. * 6 u U U U U
  600. * 7 u U U U U
  601. * 8 u U U U U
  602. * 9 u U U U U
  603. * 10 DV v v v v v
  604. * 11 v V V V V
  605. * 12 v V V V V
  606. * 13 v V V V V
  607. * 14 v V V V V
  608. * DY/DU/DV are for luma/chroma DC.
  609. */
  610. #define LUMA_DC_BLOCK_INDEX 48
  611. #define CHROMA_DC_BLOCK_INDEX 49
  612. // This table must be here because scan8[constant] must be known at compiletime
  613. static const uint8_t scan8[16 * 3 + 3] = {
  614. 4 + 1 * 8, 5 + 1 * 8, 4 + 2 * 8, 5 + 2 * 8,
  615. 6 + 1 * 8, 7 + 1 * 8, 6 + 2 * 8, 7 + 2 * 8,
  616. 4 + 3 * 8, 5 + 3 * 8, 4 + 4 * 8, 5 + 4 * 8,
  617. 6 + 3 * 8, 7 + 3 * 8, 6 + 4 * 8, 7 + 4 * 8,
  618. 4 + 6 * 8, 5 + 6 * 8, 4 + 7 * 8, 5 + 7 * 8,
  619. 6 + 6 * 8, 7 + 6 * 8, 6 + 7 * 8, 7 + 7 * 8,
  620. 4 + 8 * 8, 5 + 8 * 8, 4 + 9 * 8, 5 + 9 * 8,
  621. 6 + 8 * 8, 7 + 8 * 8, 6 + 9 * 8, 7 + 9 * 8,
  622. 4 + 11 * 8, 5 + 11 * 8, 4 + 12 * 8, 5 + 12 * 8,
  623. 6 + 11 * 8, 7 + 11 * 8, 6 + 12 * 8, 7 + 12 * 8,
  624. 4 + 13 * 8, 5 + 13 * 8, 4 + 14 * 8, 5 + 14 * 8,
  625. 6 + 13 * 8, 7 + 13 * 8, 6 + 14 * 8, 7 + 14 * 8,
  626. 0 + 0 * 8, 0 + 5 * 8, 0 + 10 * 8
  627. };
  628. static av_always_inline uint32_t pack16to32(int a, int b)
  629. {
  630. #if HAVE_BIGENDIAN
  631. return (b & 0xFFFF) + (a << 16);
  632. #else
  633. return (a & 0xFFFF) + (b << 16);
  634. #endif
  635. }
  636. static av_always_inline uint16_t pack8to16(int a, int b)
  637. {
  638. #if HAVE_BIGENDIAN
  639. return (b & 0xFF) + (a << 8);
  640. #else
  641. return (a & 0xFF) + (b << 8);
  642. #endif
  643. }
  644. /**
  645. * Get the chroma qp.
  646. */
  647. static av_always_inline int get_chroma_qp(const H264Context *h, int t, int qscale)
  648. {
  649. return h->ps.pps->chroma_qp_table[t][qscale];
  650. }
  651. /**
  652. * Get the predicted intra4x4 prediction mode.
  653. */
  654. static av_always_inline int pred_intra_mode(const H264Context *h,
  655. H264SliceContext *sl, int n)
  656. {
  657. const int index8 = scan8[n];
  658. const int left = sl->intra4x4_pred_mode_cache[index8 - 1];
  659. const int top = sl->intra4x4_pred_mode_cache[index8 - 8];
  660. const int min = FFMIN(left, top);
  661. ff_tlog(h->avctx, "mode:%d %d min:%d\n", left, top, min);
  662. if (min < 0)
  663. return DC_PRED;
  664. else
  665. return min;
  666. }
  667. static av_always_inline void write_back_intra_pred_mode(const H264Context *h,
  668. H264SliceContext *sl)
  669. {
  670. int8_t *i4x4 = sl->intra4x4_pred_mode + h->mb2br_xy[sl->mb_xy];
  671. int8_t *i4x4_cache = sl->intra4x4_pred_mode_cache;
  672. AV_COPY32(i4x4, i4x4_cache + 4 + 8 * 4);
  673. i4x4[4] = i4x4_cache[7 + 8 * 3];
  674. i4x4[5] = i4x4_cache[7 + 8 * 2];
  675. i4x4[6] = i4x4_cache[7 + 8 * 1];
  676. }
  677. static av_always_inline void write_back_non_zero_count(const H264Context *h,
  678. H264SliceContext *sl)
  679. {
  680. const int mb_xy = sl->mb_xy;
  681. uint8_t *nnz = h->non_zero_count[mb_xy];
  682. uint8_t *nnz_cache = sl->non_zero_count_cache;
  683. AV_COPY32(&nnz[ 0], &nnz_cache[4 + 8 * 1]);
  684. AV_COPY32(&nnz[ 4], &nnz_cache[4 + 8 * 2]);
  685. AV_COPY32(&nnz[ 8], &nnz_cache[4 + 8 * 3]);
  686. AV_COPY32(&nnz[12], &nnz_cache[4 + 8 * 4]);
  687. AV_COPY32(&nnz[16], &nnz_cache[4 + 8 * 6]);
  688. AV_COPY32(&nnz[20], &nnz_cache[4 + 8 * 7]);
  689. AV_COPY32(&nnz[32], &nnz_cache[4 + 8 * 11]);
  690. AV_COPY32(&nnz[36], &nnz_cache[4 + 8 * 12]);
  691. if (!h->chroma_y_shift) {
  692. AV_COPY32(&nnz[24], &nnz_cache[4 + 8 * 8]);
  693. AV_COPY32(&nnz[28], &nnz_cache[4 + 8 * 9]);
  694. AV_COPY32(&nnz[40], &nnz_cache[4 + 8 * 13]);
  695. AV_COPY32(&nnz[44], &nnz_cache[4 + 8 * 14]);
  696. }
  697. }
  698. static av_always_inline void write_back_motion_list(const H264Context *h,
  699. H264SliceContext *sl,
  700. int b_stride,
  701. int b_xy, int b8_xy,
  702. int mb_type, int list)
  703. {
  704. int16_t(*mv_dst)[2] = &h->cur_pic.motion_val[list][b_xy];
  705. int16_t(*mv_src)[2] = &sl->mv_cache[list][scan8[0]];
  706. AV_COPY128(mv_dst + 0 * b_stride, mv_src + 8 * 0);
  707. AV_COPY128(mv_dst + 1 * b_stride, mv_src + 8 * 1);
  708. AV_COPY128(mv_dst + 2 * b_stride, mv_src + 8 * 2);
  709. AV_COPY128(mv_dst + 3 * b_stride, mv_src + 8 * 3);
  710. if (CABAC(h)) {
  711. uint8_t (*mvd_dst)[2] = &sl->mvd_table[list][FMO ? 8 * sl->mb_xy
  712. : h->mb2br_xy[sl->mb_xy]];
  713. uint8_t(*mvd_src)[2] = &sl->mvd_cache[list][scan8[0]];
  714. if (IS_SKIP(mb_type)) {
  715. AV_ZERO128(mvd_dst);
  716. } else {
  717. AV_COPY64(mvd_dst, mvd_src + 8 * 3);
  718. AV_COPY16(mvd_dst + 3 + 3, mvd_src + 3 + 8 * 0);
  719. AV_COPY16(mvd_dst + 3 + 2, mvd_src + 3 + 8 * 1);
  720. AV_COPY16(mvd_dst + 3 + 1, mvd_src + 3 + 8 * 2);
  721. }
  722. }
  723. {
  724. int8_t *ref_index = &h->cur_pic.ref_index[list][b8_xy];
  725. int8_t *ref_cache = sl->ref_cache[list];
  726. ref_index[0 + 0 * 2] = ref_cache[scan8[0]];
  727. ref_index[1 + 0 * 2] = ref_cache[scan8[4]];
  728. ref_index[0 + 1 * 2] = ref_cache[scan8[8]];
  729. ref_index[1 + 1 * 2] = ref_cache[scan8[12]];
  730. }
  731. }
  732. static av_always_inline void write_back_motion(const H264Context *h,
  733. H264SliceContext *sl,
  734. int mb_type)
  735. {
  736. const int b_stride = h->b_stride;
  737. const int b_xy = 4 * sl->mb_x + 4 * sl->mb_y * h->b_stride; // try mb2b(8)_xy
  738. const int b8_xy = 4 * sl->mb_xy;
  739. if (USES_LIST(mb_type, 0)) {
  740. write_back_motion_list(h, sl, b_stride, b_xy, b8_xy, mb_type, 0);
  741. } else {
  742. fill_rectangle(&h->cur_pic.ref_index[0][b8_xy],
  743. 2, 2, 2, (uint8_t)LIST_NOT_USED, 1);
  744. }
  745. if (USES_LIST(mb_type, 1))
  746. write_back_motion_list(h, sl, b_stride, b_xy, b8_xy, mb_type, 1);
  747. if (sl->slice_type_nos == AV_PICTURE_TYPE_B && CABAC(h)) {
  748. if (IS_8X8(mb_type)) {
  749. uint8_t *direct_table = &h->direct_table[4 * sl->mb_xy];
  750. direct_table[1] = sl->sub_mb_type[1] >> 1;
  751. direct_table[2] = sl->sub_mb_type[2] >> 1;
  752. direct_table[3] = sl->sub_mb_type[3] >> 1;
  753. }
  754. }
  755. }
  756. static av_always_inline int get_dct8x8_allowed(const H264Context *h, H264SliceContext *sl)
  757. {
  758. if (h->ps.sps->direct_8x8_inference_flag)
  759. return !(AV_RN64A(sl->sub_mb_type) &
  760. ((MB_TYPE_16x8 | MB_TYPE_8x16 | MB_TYPE_8x8) *
  761. 0x0001000100010001ULL));
  762. else
  763. return !(AV_RN64A(sl->sub_mb_type) &
  764. ((MB_TYPE_16x8 | MB_TYPE_8x16 | MB_TYPE_8x8 | MB_TYPE_DIRECT2) *
  765. 0x0001000100010001ULL));
  766. }
  767. int ff_h264_field_end(H264Context *h, H264SliceContext *sl, int in_setup);
  768. int ff_h264_ref_picture(H264Context *h, H264Picture *dst, H264Picture *src);
  769. void ff_h264_unref_picture(H264Context *h, H264Picture *pic);
  770. int ff_h264_slice_context_init(H264Context *h, H264SliceContext *sl);
  771. void ff_h264_draw_horiz_band(const H264Context *h, H264SliceContext *sl, int y, int height);
  772. int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl);
  773. int ff_h264_execute_decode_slices(H264Context *h, unsigned context_count);
  774. int ff_h264_update_thread_context(AVCodecContext *dst,
  775. const AVCodecContext *src);
  776. void ff_h264_flush_change(H264Context *h);
  777. void ff_h264_free_tables(H264Context *h);
  778. #endif /* AVCODEC_H264_H */