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.

1310 lines
48KB

  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 libavcodec/h264.h
  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 "dsputil.h"
  29. #include "cabac.h"
  30. #include "mpegvideo.h"
  31. #include "h264pred.h"
  32. #include "rectangle.h"
  33. #define interlaced_dct interlaced_dct_is_a_bad_name
  34. #define mb_intra mb_intra_is_not_initialized_see_mb_type
  35. #define LUMA_DC_BLOCK_INDEX 25
  36. #define CHROMA_DC_BLOCK_INDEX 26
  37. #define CHROMA_DC_COEFF_TOKEN_VLC_BITS 8
  38. #define COEFF_TOKEN_VLC_BITS 8
  39. #define TOTAL_ZEROS_VLC_BITS 9
  40. #define CHROMA_DC_TOTAL_ZEROS_VLC_BITS 3
  41. #define RUN_VLC_BITS 3
  42. #define RUN7_VLC_BITS 6
  43. #define MAX_SPS_COUNT 32
  44. #define MAX_PPS_COUNT 256
  45. #define MAX_MMCO_COUNT 66
  46. #define MAX_DELAYED_PIC_COUNT 16
  47. /* Compiling in interlaced support reduces the speed
  48. * of progressive decoding by about 2%. */
  49. #define ALLOW_INTERLACE
  50. #define ALLOW_NOCHROMA
  51. /**
  52. * The maximum number of slices supported by the decoder.
  53. * must be a power of 2
  54. */
  55. #define MAX_SLICES 16
  56. #ifdef ALLOW_INTERLACE
  57. #define MB_MBAFF h->mb_mbaff
  58. #define MB_FIELD h->mb_field_decoding_flag
  59. #define FRAME_MBAFF h->mb_aff_frame
  60. #define FIELD_PICTURE (s->picture_structure != PICT_FRAME)
  61. #else
  62. #define MB_MBAFF 0
  63. #define MB_FIELD 0
  64. #define FRAME_MBAFF 0
  65. #define FIELD_PICTURE 0
  66. #undef IS_INTERLACED
  67. #define IS_INTERLACED(mb_type) 0
  68. #endif
  69. #define FIELD_OR_MBAFF_PICTURE (FRAME_MBAFF || FIELD_PICTURE)
  70. #ifdef ALLOW_NOCHROMA
  71. #define CHROMA h->sps.chroma_format_idc
  72. #else
  73. #define CHROMA 1
  74. #endif
  75. #ifndef CABAC
  76. #define CABAC h->pps.cabac
  77. #endif
  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. /**
  84. * Value of Picture.reference when Picture is not a reference picture, but
  85. * is held for delayed output.
  86. */
  87. #define DELAYED_PIC_REF 4
  88. /* NAL unit types */
  89. enum {
  90. NAL_SLICE=1,
  91. NAL_DPA,
  92. NAL_DPB,
  93. NAL_DPC,
  94. NAL_IDR_SLICE,
  95. NAL_SEI,
  96. NAL_SPS,
  97. NAL_PPS,
  98. NAL_AUD,
  99. NAL_END_SEQUENCE,
  100. NAL_END_STREAM,
  101. NAL_FILLER_DATA,
  102. NAL_SPS_EXT,
  103. NAL_AUXILIARY_SLICE=19
  104. };
  105. /**
  106. * SEI message types
  107. */
  108. typedef enum {
  109. SEI_BUFFERING_PERIOD = 0, ///< buffering period (H.264, D.1.1)
  110. SEI_TYPE_PIC_TIMING = 1, ///< picture timing
  111. SEI_TYPE_USER_DATA_UNREGISTERED = 5, ///< unregistered user data
  112. SEI_TYPE_RECOVERY_POINT = 6 ///< recovery point (frame # to decoder sync)
  113. } SEI_Type;
  114. /**
  115. * pic_struct in picture timing SEI message
  116. */
  117. typedef enum {
  118. SEI_PIC_STRUCT_FRAME = 0, ///< 0: %frame
  119. SEI_PIC_STRUCT_TOP_FIELD = 1, ///< 1: top field
  120. SEI_PIC_STRUCT_BOTTOM_FIELD = 2, ///< 2: bottom field
  121. SEI_PIC_STRUCT_TOP_BOTTOM = 3, ///< 3: top field, bottom field, in that order
  122. SEI_PIC_STRUCT_BOTTOM_TOP = 4, ///< 4: bottom field, top field, in that order
  123. SEI_PIC_STRUCT_TOP_BOTTOM_TOP = 5, ///< 5: top field, bottom field, top field repeated, in that order
  124. SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM = 6, ///< 6: bottom field, top field, bottom field repeated, in that order
  125. SEI_PIC_STRUCT_FRAME_DOUBLING = 7, ///< 7: %frame doubling
  126. SEI_PIC_STRUCT_FRAME_TRIPLING = 8 ///< 8: %frame tripling
  127. } SEI_PicStructType;
  128. /**
  129. * Sequence parameter set
  130. */
  131. typedef struct SPS{
  132. int profile_idc;
  133. int level_idc;
  134. int chroma_format_idc;
  135. int transform_bypass; ///< qpprime_y_zero_transform_bypass_flag
  136. int log2_max_frame_num; ///< log2_max_frame_num_minus4 + 4
  137. int poc_type; ///< pic_order_cnt_type
  138. int log2_max_poc_lsb; ///< log2_max_pic_order_cnt_lsb_minus4
  139. int delta_pic_order_always_zero_flag;
  140. int offset_for_non_ref_pic;
  141. int offset_for_top_to_bottom_field;
  142. int poc_cycle_length; ///< num_ref_frames_in_pic_order_cnt_cycle
  143. int ref_frame_count; ///< num_ref_frames
  144. int gaps_in_frame_num_allowed_flag;
  145. int mb_width; ///< pic_width_in_mbs_minus1 + 1
  146. int mb_height; ///< pic_height_in_map_units_minus1 + 1
  147. int frame_mbs_only_flag;
  148. int mb_aff; ///<mb_adaptive_frame_field_flag
  149. int direct_8x8_inference_flag;
  150. int crop; ///< frame_cropping_flag
  151. unsigned int crop_left; ///< frame_cropping_rect_left_offset
  152. unsigned int crop_right; ///< frame_cropping_rect_right_offset
  153. unsigned int crop_top; ///< frame_cropping_rect_top_offset
  154. unsigned int crop_bottom; ///< frame_cropping_rect_bottom_offset
  155. int vui_parameters_present_flag;
  156. AVRational sar;
  157. int video_signal_type_present_flag;
  158. int full_range;
  159. int colour_description_present_flag;
  160. enum AVColorPrimaries color_primaries;
  161. enum AVColorTransferCharacteristic color_trc;
  162. enum AVColorSpace colorspace;
  163. int timing_info_present_flag;
  164. uint32_t num_units_in_tick;
  165. uint32_t time_scale;
  166. int fixed_frame_rate_flag;
  167. short offset_for_ref_frame[256]; //FIXME dyn aloc?
  168. int bitstream_restriction_flag;
  169. int num_reorder_frames;
  170. int scaling_matrix_present;
  171. uint8_t scaling_matrix4[6][16];
  172. uint8_t scaling_matrix8[2][64];
  173. int nal_hrd_parameters_present_flag;
  174. int vcl_hrd_parameters_present_flag;
  175. int pic_struct_present_flag;
  176. int time_offset_length;
  177. int cpb_cnt; ///< See H.264 E.1.2
  178. int initial_cpb_removal_delay_length; ///< initial_cpb_removal_delay_length_minus1 +1
  179. int cpb_removal_delay_length; ///< cpb_removal_delay_length_minus1 + 1
  180. int dpb_output_delay_length; ///< dpb_output_delay_length_minus1 + 1
  181. int bit_depth_luma; ///< bit_depth_luma_minus8 + 8
  182. int bit_depth_chroma; ///< bit_depth_chroma_minus8 + 8
  183. int residual_color_transform_flag; ///< residual_colour_transform_flag
  184. }SPS;
  185. /**
  186. * Picture parameter set
  187. */
  188. typedef struct PPS{
  189. unsigned int sps_id;
  190. int cabac; ///< entropy_coding_mode_flag
  191. int pic_order_present; ///< pic_order_present_flag
  192. int slice_group_count; ///< num_slice_groups_minus1 + 1
  193. int mb_slice_group_map_type;
  194. unsigned int ref_count[2]; ///< num_ref_idx_l0/1_active_minus1 + 1
  195. int weighted_pred; ///< weighted_pred_flag
  196. int weighted_bipred_idc;
  197. int init_qp; ///< pic_init_qp_minus26 + 26
  198. int init_qs; ///< pic_init_qs_minus26 + 26
  199. int chroma_qp_index_offset[2];
  200. int deblocking_filter_parameters_present; ///< deblocking_filter_parameters_present_flag
  201. int constrained_intra_pred; ///< constrained_intra_pred_flag
  202. int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag
  203. int transform_8x8_mode; ///< transform_8x8_mode_flag
  204. uint8_t scaling_matrix4[6][16];
  205. uint8_t scaling_matrix8[2][64];
  206. uint8_t chroma_qp_table[2][64]; ///< pre-scaled (with chroma_qp_index_offset) version of qp_table
  207. int chroma_qp_diff;
  208. }PPS;
  209. /**
  210. * Memory management control operation opcode.
  211. */
  212. typedef enum MMCOOpcode{
  213. MMCO_END=0,
  214. MMCO_SHORT2UNUSED,
  215. MMCO_LONG2UNUSED,
  216. MMCO_SHORT2LONG,
  217. MMCO_SET_MAX_LONG,
  218. MMCO_RESET,
  219. MMCO_LONG,
  220. } MMCOOpcode;
  221. /**
  222. * Memory management control operation.
  223. */
  224. typedef struct MMCO{
  225. MMCOOpcode opcode;
  226. int short_pic_num; ///< pic_num without wrapping (pic_num & max_pic_num)
  227. int long_arg; ///< index, pic_num, or num long refs depending on opcode
  228. } MMCO;
  229. /**
  230. * H264Context
  231. */
  232. typedef struct H264Context{
  233. MpegEncContext s;
  234. int nal_ref_idc;
  235. int nal_unit_type;
  236. uint8_t *rbsp_buffer[2];
  237. unsigned int rbsp_buffer_size[2];
  238. /**
  239. * Used to parse AVC variant of h264
  240. */
  241. int is_avc; ///< this flag is != 0 if codec is avc1
  242. int got_avcC; ///< flag used to parse avcC data only once
  243. int nal_length_size; ///< Number of bytes used for nal length (1, 2 or 4)
  244. int chroma_qp[2]; //QPc
  245. int qp_thresh; ///< QP threshold to skip loopfilter
  246. int prev_mb_skipped;
  247. int next_mb_skipped;
  248. //prediction stuff
  249. int chroma_pred_mode;
  250. int intra16x16_pred_mode;
  251. int top_mb_xy;
  252. int left_mb_xy[2];
  253. int8_t intra4x4_pred_mode_cache[5*8];
  254. int8_t (*intra4x4_pred_mode)[8];
  255. H264PredContext hpc;
  256. unsigned int topleft_samples_available;
  257. unsigned int top_samples_available;
  258. unsigned int topright_samples_available;
  259. unsigned int left_samples_available;
  260. uint8_t (*top_borders[2])[16+2*8];
  261. uint8_t left_border[2*(17+2*9)];
  262. /**
  263. * non zero coeff count cache.
  264. * is 64 if not available.
  265. */
  266. DECLARE_ALIGNED_8(uint8_t, non_zero_count_cache[6*8]);
  267. uint8_t (*non_zero_count)[16];
  268. /**
  269. * Motion vector cache.
  270. */
  271. DECLARE_ALIGNED_8(int16_t, mv_cache[2][5*8][2]);
  272. DECLARE_ALIGNED_8(int8_t, ref_cache[2][5*8]);
  273. #define LIST_NOT_USED -1 //FIXME rename?
  274. #define PART_NOT_AVAILABLE -2
  275. /**
  276. * is 1 if the specific list MV&references are set to 0,0,-2.
  277. */
  278. int mv_cache_clean[2];
  279. /**
  280. * number of neighbors (top and/or left) that used 8x8 dct
  281. */
  282. int neighbor_transform_size;
  283. /**
  284. * block_offset[ 0..23] for frame macroblocks
  285. * block_offset[24..47] for field macroblocks
  286. */
  287. int block_offset[2*(16+8)];
  288. uint32_t *mb2b_xy; //FIXME are these 4 a good idea?
  289. uint32_t *mb2b8_xy;
  290. int b_stride; //FIXME use s->b4_stride
  291. int b8_stride;
  292. int mb_linesize; ///< may be equal to s->linesize or s->linesize*2, for mbaff
  293. int mb_uvlinesize;
  294. int emu_edge_width;
  295. int emu_edge_height;
  296. int halfpel_flag;
  297. int thirdpel_flag;
  298. int unknown_svq3_flag;
  299. int next_slice_index;
  300. SPS *sps_buffers[MAX_SPS_COUNT];
  301. SPS sps; ///< current sps
  302. PPS *pps_buffers[MAX_PPS_COUNT];
  303. /**
  304. * current pps
  305. */
  306. PPS pps; //FIXME move to Picture perhaps? (->no) do we need that?
  307. uint32_t dequant4_buffer[6][52][16];
  308. uint32_t dequant8_buffer[2][52][64];
  309. uint32_t (*dequant4_coeff[6])[16];
  310. uint32_t (*dequant8_coeff[2])[64];
  311. int dequant_coeff_pps; ///< reinit tables when pps changes
  312. int slice_num;
  313. uint16_t *slice_table_base;
  314. uint16_t *slice_table; ///< slice_table_base + 2*mb_stride + 1
  315. int slice_type;
  316. int slice_type_nos; ///< S free slice type (SI/SP are remapped to I/P)
  317. int slice_type_fixed;
  318. //interlacing specific flags
  319. int mb_aff_frame;
  320. int mb_field_decoding_flag;
  321. int mb_mbaff; ///< mb_aff_frame && mb_field_decoding_flag
  322. DECLARE_ALIGNED_8(uint16_t, sub_mb_type[4]);
  323. //POC stuff
  324. int poc_lsb;
  325. int poc_msb;
  326. int delta_poc_bottom;
  327. int delta_poc[2];
  328. int frame_num;
  329. int prev_poc_msb; ///< poc_msb of the last reference pic for POC type 0
  330. int prev_poc_lsb; ///< poc_lsb of the last reference pic for POC type 0
  331. int frame_num_offset; ///< for POC type 2
  332. int prev_frame_num_offset; ///< for POC type 2
  333. int prev_frame_num; ///< frame_num of the last pic for POC type 1/2
  334. /**
  335. * frame_num for frames or 2*frame_num+1 for field pics.
  336. */
  337. int curr_pic_num;
  338. /**
  339. * max_frame_num or 2*max_frame_num for field pics.
  340. */
  341. int max_pic_num;
  342. //Weighted pred stuff
  343. int use_weight;
  344. int use_weight_chroma;
  345. int luma_log2_weight_denom;
  346. int chroma_log2_weight_denom;
  347. int luma_weight[2][48];
  348. int luma_offset[2][48];
  349. int chroma_weight[2][48][2];
  350. int chroma_offset[2][48][2];
  351. int implicit_weight[48][48];
  352. //deblock
  353. int deblocking_filter; ///< disable_deblocking_filter_idc with 1<->0
  354. int slice_alpha_c0_offset;
  355. int slice_beta_offset;
  356. int redundant_pic_count;
  357. int direct_spatial_mv_pred;
  358. int dist_scale_factor[16];
  359. int dist_scale_factor_field[2][32];
  360. int map_col_to_list0[2][16+32];
  361. int map_col_to_list0_field[2][2][16+32];
  362. /**
  363. * num_ref_idx_l0/1_active_minus1 + 1
  364. */
  365. unsigned int ref_count[2]; ///< counts frames or fields, depending on current mb mode
  366. unsigned int list_count;
  367. Picture *short_ref[32];
  368. Picture *long_ref[32];
  369. Picture default_ref_list[2][32]; ///< base reference list for all slices of a coded picture
  370. Picture ref_list[2][48]; /**< 0..15: frame refs, 16..47: mbaff field refs.
  371. Reordered version of default_ref_list
  372. according to picture reordering in slice header */
  373. int ref2frm[MAX_SLICES][2][64]; ///< reference to frame number lists, used in the loop filter, the first 2 are for -2,-1
  374. Picture *delayed_pic[MAX_DELAYED_PIC_COUNT+2]; //FIXME size?
  375. int outputed_poc;
  376. /**
  377. * memory management control operations buffer.
  378. */
  379. MMCO mmco[MAX_MMCO_COUNT];
  380. int mmco_index;
  381. int long_ref_count; ///< number of actual long term references
  382. int short_ref_count; ///< number of actual short term references
  383. //data partitioning
  384. GetBitContext intra_gb;
  385. GetBitContext inter_gb;
  386. GetBitContext *intra_gb_ptr;
  387. GetBitContext *inter_gb_ptr;
  388. DECLARE_ALIGNED_16(DCTELEM, mb[16*24]);
  389. DCTELEM mb_padding[256]; ///< 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
  390. /**
  391. * Cabac
  392. */
  393. CABACContext cabac;
  394. uint8_t cabac_state[460];
  395. int cabac_init_idc;
  396. /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0,1,2), 0x0? luma_cbp */
  397. uint16_t *cbp_table;
  398. int cbp;
  399. int top_cbp;
  400. int left_cbp;
  401. /* chroma_pred_mode for i4x4 or i16x16, else 0 */
  402. uint8_t *chroma_pred_mode_table;
  403. int last_qscale_diff;
  404. int16_t (*mvd_table[2])[2];
  405. DECLARE_ALIGNED_8(int16_t, mvd_cache[2][5*8][2]);
  406. uint8_t *direct_table;
  407. uint8_t direct_cache[5*8];
  408. uint8_t zigzag_scan[16];
  409. uint8_t zigzag_scan8x8[64];
  410. uint8_t zigzag_scan8x8_cavlc[64];
  411. uint8_t field_scan[16];
  412. uint8_t field_scan8x8[64];
  413. uint8_t field_scan8x8_cavlc[64];
  414. const uint8_t *zigzag_scan_q0;
  415. const uint8_t *zigzag_scan8x8_q0;
  416. const uint8_t *zigzag_scan8x8_cavlc_q0;
  417. const uint8_t *field_scan_q0;
  418. const uint8_t *field_scan8x8_q0;
  419. const uint8_t *field_scan8x8_cavlc_q0;
  420. int x264_build;
  421. /**
  422. * @defgroup multithreading Members for slice based multithreading
  423. * @{
  424. */
  425. struct H264Context *thread_context[MAX_THREADS];
  426. /**
  427. * current slice number, used to initalize slice_num of each thread/context
  428. */
  429. int current_slice;
  430. /**
  431. * Max number of threads / contexts.
  432. * This is equal to AVCodecContext.thread_count unless
  433. * multithreaded decoding is impossible, in which case it is
  434. * reduced to 1.
  435. */
  436. int max_contexts;
  437. /**
  438. * 1 if the single thread fallback warning has already been
  439. * displayed, 0 otherwise.
  440. */
  441. int single_decode_warning;
  442. int last_slice_type;
  443. /** @} */
  444. int mb_xy;
  445. uint32_t svq3_watermark_key;
  446. /**
  447. * pic_struct in picture timing SEI message
  448. */
  449. SEI_PicStructType sei_pic_struct;
  450. /**
  451. * Complement sei_pic_struct
  452. * SEI_PIC_STRUCT_TOP_BOTTOM and SEI_PIC_STRUCT_BOTTOM_TOP indicate interlaced frames.
  453. * However, soft telecined frames may have these values.
  454. * This is used in an attempt to flag soft telecine progressive.
  455. */
  456. int prev_interlaced_frame;
  457. /**
  458. * Bit set of clock types for fields/frames in picture timing SEI message.
  459. * For each found ct_type, appropriate bit is set (e.g., bit 1 for
  460. * interlaced).
  461. */
  462. int sei_ct_type;
  463. /**
  464. * dpb_output_delay in picture timing SEI message, see H.264 C.2.2
  465. */
  466. int sei_dpb_output_delay;
  467. /**
  468. * cpb_removal_delay in picture timing SEI message, see H.264 C.1.2
  469. */
  470. int sei_cpb_removal_delay;
  471. /**
  472. * recovery_frame_cnt from SEI message
  473. *
  474. * Set to -1 if no recovery point SEI message found or to number of frames
  475. * before playback synchronizes. Frames having recovery point are key
  476. * frames.
  477. */
  478. int sei_recovery_frame_cnt;
  479. int is_complex;
  480. int luma_weight_flag[2]; ///< 7.4.3.2 luma_weight_lX_flag
  481. int chroma_weight_flag[2]; ///< 7.4.3.2 chroma_weight_lX_flag
  482. // Timestamp stuff
  483. int sei_buffering_period_present; ///< Buffering period SEI flag
  484. int initial_cpb_removal_delay[32]; ///< Initial timestamps for CPBs
  485. }H264Context;
  486. extern const uint8_t ff_h264_chroma_qp[52];
  487. /**
  488. * Decode SEI
  489. */
  490. int ff_h264_decode_sei(H264Context *h);
  491. /**
  492. * Decode SPS
  493. */
  494. int ff_h264_decode_seq_parameter_set(H264Context *h);
  495. /**
  496. * Decode PPS
  497. */
  498. int ff_h264_decode_picture_parameter_set(H264Context *h, int bit_length);
  499. /**
  500. * Decodes a network abstraction layer unit.
  501. * @param consumed is the number of bytes used as input
  502. * @param length is the length of the array
  503. * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp tailing?
  504. * @returns decoded bytes, might be src+1 if no escapes
  505. */
  506. const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src, int *dst_length, int *consumed, int length);
  507. /**
  508. * identifies the exact end of the bitstream
  509. * @return the length of the trailing, or 0 if damaged
  510. */
  511. int ff_h264_decode_rbsp_trailing(H264Context *h, const uint8_t *src);
  512. /**
  513. * frees any data that may have been allocated in the H264 context like SPS, PPS etc.
  514. */
  515. av_cold void ff_h264_free_context(H264Context *h);
  516. /**
  517. * reconstructs bitstream slice_type.
  518. */
  519. int ff_h264_get_slice_type(H264Context *h);
  520. /**
  521. * allocates tables.
  522. * needs width/height
  523. */
  524. int ff_h264_alloc_tables(H264Context *h);
  525. /**
  526. * fills the default_ref_list.
  527. */
  528. int ff_h264_fill_default_ref_list(H264Context *h);
  529. int ff_h264_decode_ref_pic_list_reordering(H264Context *h);
  530. void ff_h264_fill_mbaff_ref_list(H264Context *h);
  531. void ff_h264_remove_all_refs(H264Context *h);
  532. /**
  533. * Executes the reference picture marking (memory management control operations).
  534. */
  535. int ff_h264_execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count);
  536. int ff_h264_decode_ref_pic_marking(H264Context *h, GetBitContext *gb);
  537. /**
  538. * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
  539. */
  540. int ff_h264_check_intra4x4_pred_mode(H264Context *h);
  541. /**
  542. * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
  543. */
  544. int ff_h264_check_intra_pred_mode(H264Context *h, int mode);
  545. void ff_h264_write_back_intra_pred_mode(H264Context *h);
  546. void ff_h264_hl_decode_mb(H264Context *h);
  547. int ff_h264_frame_start(H264Context *h);
  548. av_cold int ff_h264_decode_init(AVCodecContext *avctx);
  549. av_cold int ff_h264_decode_end(AVCodecContext *avctx);
  550. av_cold void ff_h264_decode_init_vlc(void);
  551. /**
  552. * decodes a macroblock
  553. * @returns 0 if OK, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
  554. */
  555. int ff_h264_decode_mb_cavlc(H264Context *h);
  556. /**
  557. * decodes a CABAC coded macroblock
  558. * @returns 0 if OK, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
  559. */
  560. int ff_h264_decode_mb_cabac(H264Context *h);
  561. void ff_h264_init_cabac_states(H264Context *h);
  562. void ff_h264_direct_dist_scale_factor(H264Context * const h);
  563. void ff_h264_direct_ref_list_init(H264Context * const h);
  564. void ff_h264_pred_direct_motion(H264Context * const h, int *mb_type);
  565. 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);
  566. 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);
  567. /**
  568. * Reset SEI values at the beginning of the frame.
  569. *
  570. * @param h H.264 context.
  571. */
  572. void ff_h264_reset_sei(H264Context *h);
  573. /*
  574. o-o o-o
  575. / / /
  576. o-o o-o
  577. ,---'
  578. o-o o-o
  579. / / /
  580. o-o o-o
  581. */
  582. //This table must be here because scan8[constant] must be known at compiletime
  583. static const uint8_t scan8[16 + 2*4]={
  584. 4+1*8, 5+1*8, 4+2*8, 5+2*8,
  585. 6+1*8, 7+1*8, 6+2*8, 7+2*8,
  586. 4+3*8, 5+3*8, 4+4*8, 5+4*8,
  587. 6+3*8, 7+3*8, 6+4*8, 7+4*8,
  588. 1+1*8, 2+1*8,
  589. 1+2*8, 2+2*8,
  590. 1+4*8, 2+4*8,
  591. 1+5*8, 2+5*8,
  592. };
  593. static av_always_inline uint32_t pack16to32(int a, int b){
  594. #if HAVE_BIGENDIAN
  595. return (b&0xFFFF) + (a<<16);
  596. #else
  597. return (a&0xFFFF) + (b<<16);
  598. #endif
  599. }
  600. /**
  601. * gets the chroma qp.
  602. */
  603. static inline int get_chroma_qp(H264Context *h, int t, int qscale){
  604. return h->pps.chroma_qp_table[t][qscale];
  605. }
  606. static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my);
  607. static void fill_caches(H264Context *h, int mb_type, int for_deblock){
  608. MpegEncContext * const s = &h->s;
  609. const int mb_xy= h->mb_xy;
  610. int topleft_xy, top_xy, topright_xy, left_xy[2];
  611. int topleft_type, top_type, topright_type, left_type[2];
  612. const uint8_t * left_block;
  613. int topleft_partition= -1;
  614. int i;
  615. static const uint8_t left_block_options[4][8]={
  616. {0,1,2,3,7,10,8,11},
  617. {2,2,3,3,8,11,8,11},
  618. {0,0,1,1,7,10,7,10},
  619. {0,2,0,2,7,10,7,10}
  620. };
  621. top_xy = mb_xy - (s->mb_stride << FIELD_PICTURE);
  622. //FIXME deblocking could skip the intra and nnz parts.
  623. if(for_deblock && (h->slice_num == 1 || h->slice_table[mb_xy] == h->slice_table[top_xy]) && !FRAME_MBAFF)
  624. return;
  625. /* Wow, what a mess, why didn't they simplify the interlacing & intra
  626. * stuff, I can't imagine that these complex rules are worth it. */
  627. topleft_xy = top_xy - 1;
  628. topright_xy= top_xy + 1;
  629. left_xy[1] = left_xy[0] = mb_xy-1;
  630. left_block = left_block_options[0];
  631. if(FRAME_MBAFF){
  632. const int pair_xy = s->mb_x + (s->mb_y & ~1)*s->mb_stride;
  633. const int top_pair_xy = pair_xy - s->mb_stride;
  634. const int topleft_pair_xy = top_pair_xy - 1;
  635. const int topright_pair_xy = top_pair_xy + 1;
  636. const int topleft_mb_field_flag = IS_INTERLACED(s->current_picture.mb_type[topleft_pair_xy]);
  637. const int top_mb_field_flag = IS_INTERLACED(s->current_picture.mb_type[top_pair_xy]);
  638. const int topright_mb_field_flag = IS_INTERLACED(s->current_picture.mb_type[topright_pair_xy]);
  639. const int left_mb_field_flag = IS_INTERLACED(s->current_picture.mb_type[pair_xy-1]);
  640. const int curr_mb_field_flag = IS_INTERLACED(mb_type);
  641. const int bottom = (s->mb_y & 1);
  642. tprintf(s->avctx, "fill_caches: curr_mb_field_flag:%d, left_mb_field_flag:%d, topleft_mb_field_flag:%d, top_mb_field_flag:%d, topright_mb_field_flag:%d\n", curr_mb_field_flag, left_mb_field_flag, topleft_mb_field_flag, top_mb_field_flag, topright_mb_field_flag);
  643. if (curr_mb_field_flag && (bottom || top_mb_field_flag)){
  644. top_xy -= s->mb_stride;
  645. }
  646. if (curr_mb_field_flag && (bottom || topleft_mb_field_flag)){
  647. topleft_xy -= s->mb_stride;
  648. } else if(bottom && !curr_mb_field_flag && left_mb_field_flag) {
  649. topleft_xy += s->mb_stride;
  650. // take top left mv from the middle of the mb, as opposed to all other modes which use the bottom right partition
  651. topleft_partition = 0;
  652. }
  653. if (curr_mb_field_flag && (bottom || topright_mb_field_flag)){
  654. topright_xy -= s->mb_stride;
  655. }
  656. if (left_mb_field_flag != curr_mb_field_flag) {
  657. left_xy[1] = left_xy[0] = pair_xy - 1;
  658. if (curr_mb_field_flag) {
  659. left_xy[1] += s->mb_stride;
  660. left_block = left_block_options[3];
  661. } else {
  662. left_block= left_block_options[2 - bottom];
  663. }
  664. }
  665. }
  666. h->top_mb_xy = top_xy;
  667. h->left_mb_xy[0] = left_xy[0];
  668. h->left_mb_xy[1] = left_xy[1];
  669. if(for_deblock){
  670. topleft_type = 0;
  671. topright_type = 0;
  672. top_type = h->slice_table[top_xy ] < 0xFFFF ? s->current_picture.mb_type[top_xy] : 0;
  673. left_type[0] = h->slice_table[left_xy[0] ] < 0xFFFF ? s->current_picture.mb_type[left_xy[0]] : 0;
  674. left_type[1] = h->slice_table[left_xy[1] ] < 0xFFFF ? s->current_picture.mb_type[left_xy[1]] : 0;
  675. if(MB_MBAFF && !IS_INTRA(mb_type)){
  676. int list;
  677. for(list=0; list<h->list_count; list++){
  678. //These values where changed for ease of performing MC, we need to change them back
  679. //FIXME maybe we can make MC and loop filter use the same values or prevent
  680. //the MC code from changing ref_cache and rather use a temporary array.
  681. if(USES_LIST(mb_type,list)){
  682. int8_t *ref = &s->current_picture.ref_index[list][h->mb2b8_xy[mb_xy]];
  683. *(uint32_t*)&h->ref_cache[list][scan8[ 0]] =
  684. *(uint32_t*)&h->ref_cache[list][scan8[ 2]] = (pack16to32(ref[0],ref[1])&0x00FF00FF)*0x0101;
  685. ref += h->b8_stride;
  686. *(uint32_t*)&h->ref_cache[list][scan8[ 8]] =
  687. *(uint32_t*)&h->ref_cache[list][scan8[10]] = (pack16to32(ref[0],ref[1])&0x00FF00FF)*0x0101;
  688. }
  689. }
  690. }
  691. }else{
  692. topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
  693. top_type = h->slice_table[top_xy ] == h->slice_num ? s->current_picture.mb_type[top_xy] : 0;
  694. topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
  695. left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
  696. left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
  697. if(IS_INTRA(mb_type)){
  698. int type_mask= h->pps.constrained_intra_pred ? IS_INTRA(-1) : -1;
  699. h->topleft_samples_available=
  700. h->top_samples_available=
  701. h->left_samples_available= 0xFFFF;
  702. h->topright_samples_available= 0xEEEA;
  703. if(!(top_type & type_mask)){
  704. h->topleft_samples_available= 0xB3FF;
  705. h->top_samples_available= 0x33FF;
  706. h->topright_samples_available= 0x26EA;
  707. }
  708. if(IS_INTERLACED(mb_type) != IS_INTERLACED(left_type[0])){
  709. if(IS_INTERLACED(mb_type)){
  710. if(!(left_type[0] & type_mask)){
  711. h->topleft_samples_available&= 0xDFFF;
  712. h->left_samples_available&= 0x5FFF;
  713. }
  714. if(!(left_type[1] & type_mask)){
  715. h->topleft_samples_available&= 0xFF5F;
  716. h->left_samples_available&= 0xFF5F;
  717. }
  718. }else{
  719. int left_typei = h->slice_table[left_xy[0] + s->mb_stride ] == h->slice_num
  720. ? s->current_picture.mb_type[left_xy[0] + s->mb_stride] : 0;
  721. assert(left_xy[0] == left_xy[1]);
  722. if(!((left_typei & type_mask) && (left_type[0] & type_mask))){
  723. h->topleft_samples_available&= 0xDF5F;
  724. h->left_samples_available&= 0x5F5F;
  725. }
  726. }
  727. }else{
  728. if(!(left_type[0] & type_mask)){
  729. h->topleft_samples_available&= 0xDF5F;
  730. h->left_samples_available&= 0x5F5F;
  731. }
  732. }
  733. if(!(topleft_type & type_mask))
  734. h->topleft_samples_available&= 0x7FFF;
  735. if(!(topright_type & type_mask))
  736. h->topright_samples_available&= 0xFBFF;
  737. if(IS_INTRA4x4(mb_type)){
  738. if(IS_INTRA4x4(top_type)){
  739. h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
  740. h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
  741. h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
  742. h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
  743. }else{
  744. int pred;
  745. if(!(top_type & type_mask))
  746. pred= -1;
  747. else{
  748. pred= 2;
  749. }
  750. h->intra4x4_pred_mode_cache[4+8*0]=
  751. h->intra4x4_pred_mode_cache[5+8*0]=
  752. h->intra4x4_pred_mode_cache[6+8*0]=
  753. h->intra4x4_pred_mode_cache[7+8*0]= pred;
  754. }
  755. for(i=0; i<2; i++){
  756. if(IS_INTRA4x4(left_type[i])){
  757. h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
  758. h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
  759. }else{
  760. int pred;
  761. if(!(left_type[i] & type_mask))
  762. pred= -1;
  763. else{
  764. pred= 2;
  765. }
  766. h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
  767. h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
  768. }
  769. }
  770. }
  771. }
  772. }
  773. /*
  774. 0 . T T. T T T T
  775. 1 L . .L . . . .
  776. 2 L . .L . . . .
  777. 3 . T TL . . . .
  778. 4 L . .L . . . .
  779. 5 L . .. . . . .
  780. */
  781. //FIXME constraint_intra_pred & partitioning & nnz (let us hope this is just a typo in the spec)
  782. if(top_type){
  783. h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][4];
  784. h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][5];
  785. h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][6];
  786. h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3];
  787. h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][9];
  788. h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8];
  789. h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][12];
  790. h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11];
  791. }else{
  792. h->non_zero_count_cache[4+8*0]=
  793. h->non_zero_count_cache[5+8*0]=
  794. h->non_zero_count_cache[6+8*0]=
  795. h->non_zero_count_cache[7+8*0]=
  796. h->non_zero_count_cache[1+8*0]=
  797. h->non_zero_count_cache[2+8*0]=
  798. h->non_zero_count_cache[1+8*3]=
  799. h->non_zero_count_cache[2+8*3]= CABAC && !IS_INTRA(mb_type) ? 0 : 64;
  800. }
  801. for (i=0; i<2; i++) {
  802. if(left_type[i]){
  803. h->non_zero_count_cache[3+8*1 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[0+2*i]];
  804. h->non_zero_count_cache[3+8*2 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[1+2*i]];
  805. h->non_zero_count_cache[0+8*1 + 8*i]= h->non_zero_count[left_xy[i]][left_block[4+2*i]];
  806. h->non_zero_count_cache[0+8*4 + 8*i]= h->non_zero_count[left_xy[i]][left_block[5+2*i]];
  807. }else{
  808. h->non_zero_count_cache[3+8*1 + 2*8*i]=
  809. h->non_zero_count_cache[3+8*2 + 2*8*i]=
  810. h->non_zero_count_cache[0+8*1 + 8*i]=
  811. h->non_zero_count_cache[0+8*4 + 8*i]= CABAC && !IS_INTRA(mb_type) ? 0 : 64;
  812. }
  813. }
  814. if( CABAC ) {
  815. // top_cbp
  816. if(top_type) {
  817. h->top_cbp = h->cbp_table[top_xy];
  818. } else if(IS_INTRA(mb_type)) {
  819. h->top_cbp = 0x1C0;
  820. } else {
  821. h->top_cbp = 0;
  822. }
  823. // left_cbp
  824. if (left_type[0]) {
  825. h->left_cbp = h->cbp_table[left_xy[0]] & 0x1f0;
  826. } else if(IS_INTRA(mb_type)) {
  827. h->left_cbp = 0x1C0;
  828. } else {
  829. h->left_cbp = 0;
  830. }
  831. if (left_type[0]) {
  832. h->left_cbp |= ((h->cbp_table[left_xy[0]]>>((left_block[0]&(~1))+1))&0x1) << 1;
  833. }
  834. if (left_type[1]) {
  835. h->left_cbp |= ((h->cbp_table[left_xy[1]]>>((left_block[2]&(~1))+1))&0x1) << 3;
  836. }
  837. }
  838. #if 1
  839. if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
  840. int list;
  841. for(list=0; list<h->list_count; list++){
  842. if(!USES_LIST(mb_type, list) && !IS_DIRECT(mb_type) && !h->deblocking_filter){
  843. /*if(!h->mv_cache_clean[list]){
  844. memset(h->mv_cache [list], 0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
  845. memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
  846. h->mv_cache_clean[list]= 1;
  847. }*/
  848. continue;
  849. }
  850. h->mv_cache_clean[list]= 0;
  851. if(USES_LIST(top_type, list)){
  852. const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
  853. const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
  854. *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0];
  855. *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];
  856. *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];
  857. *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];
  858. h->ref_cache[list][scan8[0] + 0 - 1*8]=
  859. h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
  860. h->ref_cache[list][scan8[0] + 2 - 1*8]=
  861. h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
  862. }else{
  863. *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]=
  864. *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]=
  865. *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]=
  866. *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;
  867. *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
  868. }
  869. for(i=0; i<2; i++){
  870. int cache_idx = scan8[0] - 1 + i*2*8;
  871. if(USES_LIST(left_type[i], list)){
  872. const int b_xy= h->mb2b_xy[left_xy[i]] + 3;
  873. const int b8_xy= h->mb2b8_xy[left_xy[i]] + 1;
  874. *(uint32_t*)h->mv_cache[list][cache_idx ]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[0+i*2]];
  875. *(uint32_t*)h->mv_cache[list][cache_idx+8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[1+i*2]];
  876. h->ref_cache[list][cache_idx ]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0+i*2]>>1)];
  877. h->ref_cache[list][cache_idx+8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[1+i*2]>>1)];
  878. }else{
  879. *(uint32_t*)h->mv_cache [list][cache_idx ]=
  880. *(uint32_t*)h->mv_cache [list][cache_idx+8]= 0;
  881. h->ref_cache[list][cache_idx ]=
  882. h->ref_cache[list][cache_idx+8]= left_type[i] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
  883. }
  884. }
  885. if(for_deblock || ((IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred) && !FRAME_MBAFF))
  886. continue;
  887. if(USES_LIST(topleft_type, list)){
  888. const int b_xy = h->mb2b_xy[topleft_xy] + 3 + h->b_stride + (topleft_partition & 2*h->b_stride);
  889. const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + (topleft_partition & h->b8_stride);
  890. *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
  891. h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
  892. }else{
  893. *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
  894. h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
  895. }
  896. if(USES_LIST(topright_type, list)){
  897. const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
  898. const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
  899. *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
  900. h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
  901. }else{
  902. *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
  903. h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
  904. }
  905. if((IS_SKIP(mb_type) || IS_DIRECT(mb_type)) && !FRAME_MBAFF)
  906. continue;
  907. h->ref_cache[list][scan8[5 ]+1] =
  908. h->ref_cache[list][scan8[7 ]+1] =
  909. h->ref_cache[list][scan8[13]+1] = //FIXME remove past 3 (init somewhere else)
  910. h->ref_cache[list][scan8[4 ]] =
  911. h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
  912. *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
  913. *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
  914. *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
  915. *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
  916. *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
  917. if( CABAC ) {
  918. /* XXX beurk, Load mvd */
  919. if(USES_LIST(top_type, list)){
  920. const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
  921. *(uint32_t*)h->mvd_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 0];
  922. *(uint32_t*)h->mvd_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 1];
  923. *(uint32_t*)h->mvd_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 2];
  924. *(uint32_t*)h->mvd_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 3];
  925. }else{
  926. *(uint32_t*)h->mvd_cache [list][scan8[0] + 0 - 1*8]=
  927. *(uint32_t*)h->mvd_cache [list][scan8[0] + 1 - 1*8]=
  928. *(uint32_t*)h->mvd_cache [list][scan8[0] + 2 - 1*8]=
  929. *(uint32_t*)h->mvd_cache [list][scan8[0] + 3 - 1*8]= 0;
  930. }
  931. if(USES_LIST(left_type[0], list)){
  932. const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
  933. *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 0*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[0]];
  934. *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[1]];
  935. }else{
  936. *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 0*8]=
  937. *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 1*8]= 0;
  938. }
  939. if(USES_LIST(left_type[1], list)){
  940. const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
  941. *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 2*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[2]];
  942. *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 3*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[3]];
  943. }else{
  944. *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 2*8]=
  945. *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 3*8]= 0;
  946. }
  947. *(uint32_t*)h->mvd_cache [list][scan8[5 ]+1]=
  948. *(uint32_t*)h->mvd_cache [list][scan8[7 ]+1]=
  949. *(uint32_t*)h->mvd_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
  950. *(uint32_t*)h->mvd_cache [list][scan8[4 ]]=
  951. *(uint32_t*)h->mvd_cache [list][scan8[12]]= 0;
  952. if(h->slice_type_nos == FF_B_TYPE){
  953. fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, 0, 1);
  954. if(IS_DIRECT(top_type)){
  955. *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0x01010101;
  956. }else if(IS_8X8(top_type)){
  957. int b8_xy = h->mb2b8_xy[top_xy] + h->b8_stride;
  958. h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy];
  959. h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 1];
  960. }else{
  961. *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0;
  962. }
  963. if(IS_DIRECT(left_type[0]))
  964. h->direct_cache[scan8[0] - 1 + 0*8]= 1;
  965. else if(IS_8X8(left_type[0]))
  966. h->direct_cache[scan8[0] - 1 + 0*8]= h->direct_table[h->mb2b8_xy[left_xy[0]] + 1 + h->b8_stride*(left_block[0]>>1)];
  967. else
  968. h->direct_cache[scan8[0] - 1 + 0*8]= 0;
  969. if(IS_DIRECT(left_type[1]))
  970. h->direct_cache[scan8[0] - 1 + 2*8]= 1;
  971. else if(IS_8X8(left_type[1]))
  972. h->direct_cache[scan8[0] - 1 + 2*8]= h->direct_table[h->mb2b8_xy[left_xy[1]] + 1 + h->b8_stride*(left_block[2]>>1)];
  973. else
  974. h->direct_cache[scan8[0] - 1 + 2*8]= 0;
  975. }
  976. }
  977. if(FRAME_MBAFF){
  978. #define MAP_MVS\
  979. MAP_F2F(scan8[0] - 1 - 1*8, topleft_type)\
  980. MAP_F2F(scan8[0] + 0 - 1*8, top_type)\
  981. MAP_F2F(scan8[0] + 1 - 1*8, top_type)\
  982. MAP_F2F(scan8[0] + 2 - 1*8, top_type)\
  983. MAP_F2F(scan8[0] + 3 - 1*8, top_type)\
  984. MAP_F2F(scan8[0] + 4 - 1*8, topright_type)\
  985. MAP_F2F(scan8[0] - 1 + 0*8, left_type[0])\
  986. MAP_F2F(scan8[0] - 1 + 1*8, left_type[0])\
  987. MAP_F2F(scan8[0] - 1 + 2*8, left_type[1])\
  988. MAP_F2F(scan8[0] - 1 + 3*8, left_type[1])
  989. if(MB_FIELD){
  990. #define MAP_F2F(idx, mb_type)\
  991. if(!IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
  992. h->ref_cache[list][idx] <<= 1;\
  993. h->mv_cache[list][idx][1] /= 2;\
  994. h->mvd_cache[list][idx][1] /= 2;\
  995. }
  996. MAP_MVS
  997. #undef MAP_F2F
  998. }else{
  999. #define MAP_F2F(idx, mb_type)\
  1000. if(IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
  1001. h->ref_cache[list][idx] >>= 1;\
  1002. h->mv_cache[list][idx][1] <<= 1;\
  1003. h->mvd_cache[list][idx][1] <<= 1;\
  1004. }
  1005. MAP_MVS
  1006. #undef MAP_F2F
  1007. }
  1008. }
  1009. }
  1010. }
  1011. #endif
  1012. h->neighbor_transform_size= !!IS_8x8DCT(top_type) + !!IS_8x8DCT(left_type[0]);
  1013. }
  1014. /**
  1015. * gets the predicted intra4x4 prediction mode.
  1016. */
  1017. static inline int pred_intra_mode(H264Context *h, int n){
  1018. const int index8= scan8[n];
  1019. const int left= h->intra4x4_pred_mode_cache[index8 - 1];
  1020. const int top = h->intra4x4_pred_mode_cache[index8 - 8];
  1021. const int min= FFMIN(left, top);
  1022. tprintf(h->s.avctx, "mode:%d %d min:%d\n", left ,top, min);
  1023. if(min<0) return DC_PRED;
  1024. else return min;
  1025. }
  1026. static inline void write_back_non_zero_count(H264Context *h){
  1027. const int mb_xy= h->mb_xy;
  1028. h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[7+8*1];
  1029. h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[7+8*2];
  1030. h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[7+8*3];
  1031. h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];
  1032. h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[4+8*4];
  1033. h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[5+8*4];
  1034. h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[6+8*4];
  1035. h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[1+8*2];
  1036. h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
  1037. h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[2+8*1];
  1038. h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[1+8*5];
  1039. h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
  1040. h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[2+8*4];
  1041. }
  1042. static inline void write_back_motion(H264Context *h, int mb_type){
  1043. MpegEncContext * const s = &h->s;
  1044. const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
  1045. const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
  1046. int list;
  1047. if(!USES_LIST(mb_type, 0))
  1048. fill_rectangle(&s->current_picture.ref_index[0][b8_xy], 2, 2, h->b8_stride, (uint8_t)LIST_NOT_USED, 1);
  1049. for(list=0; list<h->list_count; list++){
  1050. int y;
  1051. if(!USES_LIST(mb_type, list))
  1052. continue;
  1053. for(y=0; y<4; y++){
  1054. *(uint64_t*)s->current_picture.motion_val[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mv_cache[list][scan8[0]+0 + 8*y];
  1055. *(uint64_t*)s->current_picture.motion_val[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mv_cache[list][scan8[0]+2 + 8*y];
  1056. }
  1057. if( CABAC ) {
  1058. if(IS_SKIP(mb_type))
  1059. fill_rectangle(h->mvd_table[list][b_xy], 4, 4, h->b_stride, 0, 4);
  1060. else
  1061. for(y=0; y<4; y++){
  1062. *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+0 + 8*y];
  1063. *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+2 + 8*y];
  1064. }
  1065. }
  1066. {
  1067. int8_t *ref_index = &s->current_picture.ref_index[list][b8_xy];
  1068. ref_index[0+0*h->b8_stride]= h->ref_cache[list][scan8[0]];
  1069. ref_index[1+0*h->b8_stride]= h->ref_cache[list][scan8[4]];
  1070. ref_index[0+1*h->b8_stride]= h->ref_cache[list][scan8[8]];
  1071. ref_index[1+1*h->b8_stride]= h->ref_cache[list][scan8[12]];
  1072. }
  1073. }
  1074. if(h->slice_type_nos == FF_B_TYPE && CABAC){
  1075. if(IS_8X8(mb_type)){
  1076. uint8_t *direct_table = &h->direct_table[b8_xy];
  1077. direct_table[1+0*h->b8_stride] = IS_DIRECT(h->sub_mb_type[1]) ? 1 : 0;
  1078. direct_table[0+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[2]) ? 1 : 0;
  1079. direct_table[1+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[3]) ? 1 : 0;
  1080. }
  1081. }
  1082. }
  1083. static inline int get_dct8x8_allowed(H264Context *h){
  1084. if(h->sps.direct_8x8_inference_flag)
  1085. return !(*(uint64_t*)h->sub_mb_type & ((MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_8x8 )*0x0001000100010001ULL));
  1086. else
  1087. return !(*(uint64_t*)h->sub_mb_type & ((MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_8x8|MB_TYPE_DIRECT2)*0x0001000100010001ULL));
  1088. }
  1089. static void predict_field_decoding_flag(H264Context *h){
  1090. MpegEncContext * const s = &h->s;
  1091. const int mb_xy= h->mb_xy;
  1092. int mb_type = (h->slice_table[mb_xy-1] == h->slice_num)
  1093. ? s->current_picture.mb_type[mb_xy-1]
  1094. : (h->slice_table[mb_xy-s->mb_stride] == h->slice_num)
  1095. ? s->current_picture.mb_type[mb_xy-s->mb_stride]
  1096. : 0;
  1097. h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
  1098. }
  1099. /**
  1100. * decodes a P_SKIP or B_SKIP macroblock
  1101. */
  1102. static void decode_mb_skip(H264Context *h){
  1103. MpegEncContext * const s = &h->s;
  1104. const int mb_xy= h->mb_xy;
  1105. int mb_type=0;
  1106. memset(h->non_zero_count[mb_xy], 0, 16);
  1107. memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui
  1108. if(MB_FIELD)
  1109. mb_type|= MB_TYPE_INTERLACED;
  1110. if( h->slice_type_nos == FF_B_TYPE )
  1111. {
  1112. // just for fill_caches. pred_direct_motion will set the real mb_type
  1113. mb_type|= MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2|MB_TYPE_SKIP;
  1114. fill_caches(h, mb_type, 0); //FIXME check what is needed and what not ...
  1115. ff_h264_pred_direct_motion(h, &mb_type);
  1116. mb_type|= MB_TYPE_SKIP;
  1117. }
  1118. else
  1119. {
  1120. int mx, my;
  1121. mb_type|= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0|MB_TYPE_SKIP;
  1122. fill_caches(h, mb_type, 0); //FIXME check what is needed and what not ...
  1123. pred_pskip_motion(h, &mx, &my);
  1124. fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
  1125. fill_rectangle( h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4);
  1126. }
  1127. write_back_motion(h, mb_type);
  1128. s->current_picture.mb_type[mb_xy]= mb_type;
  1129. s->current_picture.qscale_table[mb_xy]= s->qscale;
  1130. h->slice_table[ mb_xy ]= h->slice_num;
  1131. h->prev_mb_skipped= 1;
  1132. }
  1133. #include "h264_mvpred.h" //For pred_pskip_motion()
  1134. #endif /* AVCODEC_H264_H */