Fixed DXVA2 H264 hwaccel compilation.
[libav.git] / libavcodec / h264.h
CommitLineData
26b4fe82
AJ
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
26b4fe82
AJ
20 */
21
22/**
bad5537e 23 * @file libavcodec/h264.h
26b4fe82
AJ
24 * H.264 / AVC / MPEG4 part10 codec.
25 * @author Michael Niedermayer <michaelni@gmx.at>
26 */
27
98790382
SS
28#ifndef AVCODEC_H264_H
29#define AVCODEC_H264_H
26b4fe82 30
19769ece 31#include "libavutil/intreadwrite.h"
26b4fe82
AJ
32#include "dsputil.h"
33#include "cabac.h"
34#include "mpegvideo.h"
c92a30bb 35#include "h264pred.h"
e1e94902 36#include "rectangle.h"
26b4fe82
AJ
37
38#define interlaced_dct interlaced_dct_is_a_bad_name
90b5b51e 39#define mb_intra mb_intra_is_not_initialized_see_mb_type
26b4fe82
AJ
40
41#define LUMA_DC_BLOCK_INDEX 25
42#define CHROMA_DC_BLOCK_INDEX 26
43
44#define CHROMA_DC_COEFF_TOKEN_VLC_BITS 8
45#define COEFF_TOKEN_VLC_BITS 8
46#define TOTAL_ZEROS_VLC_BITS 9
47#define CHROMA_DC_TOTAL_ZEROS_VLC_BITS 3
48#define RUN_VLC_BITS 3
49#define RUN7_VLC_BITS 6
50
51#define MAX_SPS_COUNT 32
52#define MAX_PPS_COUNT 256
53
54#define MAX_MMCO_COUNT 66
55
64b9d48f
AS
56#define MAX_DELAYED_PIC_COUNT 16
57
26b4fe82
AJ
58/* Compiling in interlaced support reduces the speed
59 * of progressive decoding by about 2%. */
60#define ALLOW_INTERLACE
61
0af6967e
MN
62#define ALLOW_NOCHROMA
63
d43c1922
MN
64#define FMO 0
65
b735aeea
MN
66/**
67 * The maximum number of slices supported by the decoder.
68 * must be a power of 2
69 */
70#define MAX_SLICES 16
71
26b4fe82
AJ
72#ifdef ALLOW_INTERLACE
73#define MB_MBAFF h->mb_mbaff
74#define MB_FIELD h->mb_field_decoding_flag
75#define FRAME_MBAFF h->mb_aff_frame
d6c52130 76#define FIELD_PICTURE (s->picture_structure != PICT_FRAME)
26b4fe82
AJ
77#else
78#define MB_MBAFF 0
79#define MB_FIELD 0
80#define FRAME_MBAFF 0
bbb3edb8 81#define FIELD_PICTURE 0
26b4fe82
AJ
82#undef IS_INTERLACED
83#define IS_INTERLACED(mb_type) 0
84#endif
f3e53d9f 85#define FIELD_OR_MBAFF_PICTURE (FRAME_MBAFF || FIELD_PICTURE)
26b4fe82 86
0af6967e
MN
87#ifdef ALLOW_NOCHROMA
88#define CHROMA h->sps.chroma_format_idc
89#else
90#define CHROMA 1
91#endif
92
ddd60f28
MN
93#ifndef CABAC
94#define CABAC h->pps.cabac
95#endif
96
44ee6135
MN
97#define EXTENDED_SAR 255
98
ce9ff9cc
DB
99#define MB_TYPE_REF0 MB_TYPE_ACPRED //dirty but it fits in 16 bit
100#define MB_TYPE_8x8DCT 0x01000000
101#define IS_REF0(a) ((a) & MB_TYPE_REF0)
102#define IS_8x8DCT(a) ((a) & MB_TYPE_8x8DCT)
103
ea6f00c4
MN
104/**
105 * Value of Picture.reference when Picture is not a reference picture, but
106 * is held for delayed output.
107 */
108#define DELAYED_PIC_REF 4
109
110
44ee6135
MN
111/* NAL unit types */
112enum {
2791c608
MN
113 NAL_SLICE=1,
114 NAL_DPA,
115 NAL_DPB,
116 NAL_DPC,
117 NAL_IDR_SLICE,
118 NAL_SEI,
119 NAL_SPS,
120 NAL_PPS,
121 NAL_AUD,
122 NAL_END_SEQUENCE,
123 NAL_END_STREAM,
124 NAL_FILLER_DATA,
125 NAL_SPS_EXT,
126 NAL_AUXILIARY_SLICE=19
44ee6135
MN
127};
128
26b4fe82 129/**
dd02ea11
IS
130 * SEI message types
131 */
132typedef enum {
2ea89d92 133 SEI_BUFFERING_PERIOD = 0, ///< buffering period (H.264, D.1.1)
dd02ea11
IS
134 SEI_TYPE_PIC_TIMING = 1, ///< picture timing
135 SEI_TYPE_USER_DATA_UNREGISTERED = 5, ///< unregistered user data
136 SEI_TYPE_RECOVERY_POINT = 6 ///< recovery point (frame # to decoder sync)
137} SEI_Type;
138
139/**
b09a7c05
140 * pic_struct in picture timing SEI message
141 */
142typedef enum {
143 SEI_PIC_STRUCT_FRAME = 0, ///< 0: %frame
144 SEI_PIC_STRUCT_TOP_FIELD = 1, ///< 1: top field
145 SEI_PIC_STRUCT_BOTTOM_FIELD = 2, ///< 2: bottom field
146 SEI_PIC_STRUCT_TOP_BOTTOM = 3, ///< 3: top field, bottom field, in that order
147 SEI_PIC_STRUCT_BOTTOM_TOP = 4, ///< 4: bottom field, top field, in that order
148 SEI_PIC_STRUCT_TOP_BOTTOM_TOP = 5, ///< 5: top field, bottom field, top field repeated, in that order
149 SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM = 6, ///< 6: bottom field, top field, bottom field repeated, in that order
150 SEI_PIC_STRUCT_FRAME_DOUBLING = 7, ///< 7: %frame doubling
151 SEI_PIC_STRUCT_FRAME_TRIPLING = 8 ///< 8: %frame tripling
152} SEI_PicStructType;
153
154/**
26b4fe82
AJ
155 * Sequence parameter set
156 */
157typedef struct SPS{
158
159 int profile_idc;
160 int level_idc;
0af6967e 161 int chroma_format_idc;
26b4fe82
AJ
162 int transform_bypass; ///< qpprime_y_zero_transform_bypass_flag
163 int log2_max_frame_num; ///< log2_max_frame_num_minus4 + 4
164 int poc_type; ///< pic_order_cnt_type
165 int log2_max_poc_lsb; ///< log2_max_pic_order_cnt_lsb_minus4
166 int delta_pic_order_always_zero_flag;
167 int offset_for_non_ref_pic;
168 int offset_for_top_to_bottom_field;
169 int poc_cycle_length; ///< num_ref_frames_in_pic_order_cnt_cycle
170 int ref_frame_count; ///< num_ref_frames
171 int gaps_in_frame_num_allowed_flag;
bcc3476c
JD
172 int mb_width; ///< pic_width_in_mbs_minus1 + 1
173 int mb_height; ///< pic_height_in_map_units_minus1 + 1
26b4fe82
AJ
174 int frame_mbs_only_flag;
175 int mb_aff; ///<mb_adaptive_frame_field_flag
176 int direct_8x8_inference_flag;
177 int crop; ///< frame_cropping_flag
59eeddf0
MN
178 unsigned int crop_left; ///< frame_cropping_rect_left_offset
179 unsigned int crop_right; ///< frame_cropping_rect_right_offset
180 unsigned int crop_top; ///< frame_cropping_rect_top_offset
181 unsigned int crop_bottom; ///< frame_cropping_rect_bottom_offset
26b4fe82
AJ
182 int vui_parameters_present_flag;
183 AVRational sar;
c4dffe7e
DC
184 int video_signal_type_present_flag;
185 int full_range;
186 int colour_description_present_flag;
187 enum AVColorPrimaries color_primaries;
188 enum AVColorTransferCharacteristic color_trc;
189 enum AVColorSpace colorspace;
26b4fe82
AJ
190 int timing_info_present_flag;
191 uint32_t num_units_in_tick;
192 uint32_t time_scale;
193 int fixed_frame_rate_flag;
194 short offset_for_ref_frame[256]; //FIXME dyn aloc?
195 int bitstream_restriction_flag;
196 int num_reorder_frames;
197 int scaling_matrix_present;
198 uint8_t scaling_matrix4[6][16];
199 uint8_t scaling_matrix8[2][64];
b09a7c05
200 int nal_hrd_parameters_present_flag;
201 int vcl_hrd_parameters_present_flag;
202 int pic_struct_present_flag;
203 int time_offset_length;
24a414e0 204 int cpb_cnt; ///< See H.264 E.1.2
82e38c29 205 int initial_cpb_removal_delay_length; ///< initial_cpb_removal_delay_length_minus1 +1
b09a7c05
206 int cpb_removal_delay_length; ///< cpb_removal_delay_length_minus1 + 1
207 int dpb_output_delay_length; ///< dpb_output_delay_length_minus1 + 1
cb99c652
GB
208 int bit_depth_luma; ///< bit_depth_luma_minus8 + 8
209 int bit_depth_chroma; ///< bit_depth_chroma_minus8 + 8
210 int residual_color_transform_flag; ///< residual_colour_transform_flag
26b4fe82
AJ
211}SPS;
212
213/**
214 * Picture parameter set
215 */
216typedef struct PPS{
217 unsigned int sps_id;
218 int cabac; ///< entropy_coding_mode_flag
219 int pic_order_present; ///< pic_order_present_flag
220 int slice_group_count; ///< num_slice_groups_minus1 + 1
221 int mb_slice_group_map_type;
222 unsigned int ref_count[2]; ///< num_ref_idx_l0/1_active_minus1 + 1
223 int weighted_pred; ///< weighted_pred_flag
224 int weighted_bipred_idc;
225 int init_qp; ///< pic_init_qp_minus26 + 26
226 int init_qs; ///< pic_init_qs_minus26 + 26
4691a77d 227 int chroma_qp_index_offset[2];
26b4fe82
AJ
228 int deblocking_filter_parameters_present; ///< deblocking_filter_parameters_present_flag
229 int constrained_intra_pred; ///< constrained_intra_pred_flag
230 int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag
231 int transform_8x8_mode; ///< transform_8x8_mode_flag
232 uint8_t scaling_matrix4[6][16];
233 uint8_t scaling_matrix8[2][64];
5a78bfbd 234 uint8_t chroma_qp_table[2][64]; ///< pre-scaled (with chroma_qp_index_offset) version of qp_table
4691a77d 235 int chroma_qp_diff;
26b4fe82
AJ
236}PPS;
237
238/**
239 * Memory management control operation opcode.
240 */
241typedef enum MMCOOpcode{
242 MMCO_END=0,
243 MMCO_SHORT2UNUSED,
244 MMCO_LONG2UNUSED,
245 MMCO_SHORT2LONG,
246 MMCO_SET_MAX_LONG,
247 MMCO_RESET,
248 MMCO_LONG,
249} MMCOOpcode;
250
251/**
252 * Memory management control operation.
253 */
254typedef struct MMCO{
255 MMCOOpcode opcode;
0d0447ea
JD
256 int short_pic_num; ///< pic_num without wrapping (pic_num & max_pic_num)
257 int long_arg; ///< index, pic_num, or num long refs depending on opcode
26b4fe82
AJ
258} MMCO;
259
260/**
261 * H264Context
262 */
263typedef struct H264Context{
264 MpegEncContext s;
265 int nal_ref_idc;
266 int nal_unit_type;
24456882
267 uint8_t *rbsp_buffer[2];
268 unsigned int rbsp_buffer_size[2];
26b4fe82
AJ
269
270 /**
271 * Used to parse AVC variant of h264
272 */
273 int is_avc; ///< this flag is != 0 if codec is avc1
26b4fe82
AJ
274 int nal_length_size; ///< Number of bytes used for nal length (1, 2 or 4)
275
4691a77d 276 int chroma_qp[2]; //QPc
26b4fe82 277
2e36c931
MN
278 int qp_thresh; ///< QP threshold to skip loopfilter
279
26b4fe82
AJ
280 int prev_mb_skipped;
281 int next_mb_skipped;
282
283 //prediction stuff
284 int chroma_pred_mode;
285 int intra16x16_pred_mode;
286
c1bb66ac 287 int topleft_mb_xy;
26b4fe82 288 int top_mb_xy;
c1bb66ac 289 int topright_mb_xy;
26b4fe82
AJ
290 int left_mb_xy[2];
291
c1bb66ac 292 int topleft_type;
99344d43 293 int top_type;
c1bb66ac 294 int topright_type;
99344d43
MN
295 int left_type[2];
296
c1bb66ac
MN
297 const uint8_t * left_block;
298 int topleft_partition;
299
26b4fe82 300 int8_t intra4x4_pred_mode_cache[5*8];
5b0fb524 301 int8_t (*intra4x4_pred_mode);
c92a30bb 302 H264PredContext hpc;
26b4fe82
AJ
303 unsigned int topleft_samples_available;
304 unsigned int top_samples_available;
305 unsigned int topright_samples_available;
306 unsigned int left_samples_available;
307 uint8_t (*top_borders[2])[16+2*8];
26b4fe82
AJ
308
309 /**
310 * non zero coeff count cache.
311 * is 64 if not available.
312 */
c6727809 313 DECLARE_ALIGNED_8(uint8_t, non_zero_count_cache)[6*8];
b6303e6d
MN
314
315 /*
316 .UU.YYYY
317 .UU.YYYY
318 .vv.YYYY
319 .VV.YYYY
320 */
c988f975 321 uint8_t (*non_zero_count)[32];
26b4fe82
AJ
322
323 /**
324 * Motion vector cache.
325 */
0b69d625 326 DECLARE_ALIGNED_16(int16_t, mv_cache)[2][5*8][2];
c6727809 327 DECLARE_ALIGNED_8(int8_t, ref_cache)[2][5*8];
26b4fe82
AJ
328#define LIST_NOT_USED -1 //FIXME rename?
329#define PART_NOT_AVAILABLE -2
330
331 /**
332 * is 1 if the specific list MV&references are set to 0,0,-2.
333 */
334 int mv_cache_clean[2];
335
336 /**
337 * number of neighbors (top and/or left) that used 8x8 dct
338 */
339 int neighbor_transform_size;
340
341 /**
342 * block_offset[ 0..23] for frame macroblocks
343 * block_offset[24..47] for field macroblocks
344 */
345 int block_offset[2*(16+8)];
346
347 uint32_t *mb2b_xy; //FIXME are these 4 a good idea?
d43c1922 348 uint32_t *mb2br_xy;
26b4fe82 349 int b_stride; //FIXME use s->b4_stride
26b4fe82
AJ
350
351 int mb_linesize; ///< may be equal to s->linesize or s->linesize*2, for mbaff
352 int mb_uvlinesize;
353
354 int emu_edge_width;
355 int emu_edge_height;
356
7978debd 357 SPS *sps_buffers[MAX_SPS_COUNT];
26b4fe82
AJ
358 SPS sps; ///< current sps
359
7978debd 360 PPS *pps_buffers[MAX_PPS_COUNT];
26b4fe82
AJ
361 /**
362 * current pps
363 */
364 PPS pps; //FIXME move to Picture perhaps? (->no) do we need that?
365
366 uint32_t dequant4_buffer[6][52][16];
367 uint32_t dequant8_buffer[2][52][64];
368 uint32_t (*dequant4_coeff[6])[16];
369 uint32_t (*dequant8_coeff[2])[64];
370 int dequant_coeff_pps; ///< reinit tables when pps changes
371
372 int slice_num;
b735aeea
MN
373 uint16_t *slice_table_base;
374 uint16_t *slice_table; ///< slice_table_base + 2*mb_stride + 1
26b4fe82 375 int slice_type;
e3e6f18f 376 int slice_type_nos; ///< S free slice type (SI/SP are remapped to I/P)
26b4fe82
AJ
377 int slice_type_fixed;
378
379 //interlacing specific flags
380 int mb_aff_frame;
381 int mb_field_decoding_flag;
382 int mb_mbaff; ///< mb_aff_frame && mb_field_decoding_flag
383
c6727809 384 DECLARE_ALIGNED_8(uint16_t, sub_mb_type)[4];
26b4fe82
AJ
385
386 //POC stuff
387 int poc_lsb;
388 int poc_msb;
389 int delta_poc_bottom;
390 int delta_poc[2];
391 int frame_num;
392 int prev_poc_msb; ///< poc_msb of the last reference pic for POC type 0
393 int prev_poc_lsb; ///< poc_lsb of the last reference pic for POC type 0
394 int frame_num_offset; ///< for POC type 2
395 int prev_frame_num_offset; ///< for POC type 2
396 int prev_frame_num; ///< frame_num of the last pic for POC type 1/2
397
398 /**
f57e2af6 399 * frame_num for frames or 2*frame_num+1 for field pics.
26b4fe82
AJ
400 */
401 int curr_pic_num;
402
403 /**
404 * max_frame_num or 2*max_frame_num for field pics.
405 */
406 int max_pic_num;
407
408 //Weighted pred stuff
409 int use_weight;
410 int use_weight_chroma;
411 int luma_log2_weight_denom;
412 int chroma_log2_weight_denom;
af2b0df4 413 //The following 2 can be changed to int8_t but that causes 10cpu cycles speedloss
70118abd
MN
414 int luma_weight[2][48][2];
415 int chroma_weight[2][48][2][2];
26b4fe82
AJ
416 int implicit_weight[48][48];
417
418 //deblock
419 int deblocking_filter; ///< disable_deblocking_filter_idc with 1<->0
420 int slice_alpha_c0_offset;
421 int slice_beta_offset;
422
423 int redundant_pic_count;
424
425 int direct_spatial_mv_pred;
3a06e864
MN
426 int col_parity;
427 int col_fieldoff;
26b4fe82 428 int dist_scale_factor[16];
8b1fd554 429 int dist_scale_factor_field[2][32];
f4d3382d
MN
430 int map_col_to_list0[2][16+32];
431 int map_col_to_list0_field[2][2][16+32];
26b4fe82
AJ
432
433 /**
434 * num_ref_idx_l0/1_active_minus1 + 1
435 */
436 unsigned int ref_count[2]; ///< counts frames or fields, depending on current mb mode
437 unsigned int list_count;
c988f975 438 uint8_t *list_counts; ///< Array of list_count per MB specifying the slice type
26b4fe82
AJ
439 Picture *short_ref[32];
440 Picture *long_ref[32];
bcc3476c
JD
441 Picture default_ref_list[2][32]; ///< base reference list for all slices of a coded picture
442 Picture ref_list[2][48]; /**< 0..15: frame refs, 16..47: mbaff field refs.
443 Reordered version of default_ref_list
444 according to picture reordering in slice header */
b735aeea 445 int ref2frm[MAX_SLICES][2][64]; ///< reference to frame number lists, used in the loop filter, the first 2 are for -2,-1
64b9d48f 446 Picture *delayed_pic[MAX_DELAYED_PIC_COUNT+2]; //FIXME size?
df8a7dff 447 int outputed_poc;
26b4fe82
AJ
448
449 /**
450 * memory management control operations buffer.
451 */
452 MMCO mmco[MAX_MMCO_COUNT];
453 int mmco_index;
454
455 int long_ref_count; ///< number of actual long term references
456 int short_ref_count; ///< number of actual short term references
457
458 //data partitioning
459 GetBitContext intra_gb;
460 GetBitContext inter_gb;
461 GetBitContext *intra_gb_ptr;
462 GetBitContext *inter_gb_ptr;
463
c6727809 464 DECLARE_ALIGNED_16(DCTELEM, mb)[16*24];
2cab6401 465 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
26b4fe82
AJ
466
467 /**
468 * Cabac
469 */
470 CABACContext cabac;
471 uint8_t cabac_state[460];
472 int cabac_init_idc;
473
474 /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0,1,2), 0x0? luma_cbp */
475 uint16_t *cbp_table;
476 int cbp;
477 int top_cbp;
478 int left_cbp;
479 /* chroma_pred_mode for i4x4 or i16x16, else 0 */
480 uint8_t *chroma_pred_mode_table;
481 int last_qscale_diff;
b5bd0700
MN
482 uint8_t (*mvd_table[2])[2];
483 DECLARE_ALIGNED_16(uint8_t, mvd_cache)[2][5*8][2];
26b4fe82
AJ
484 uint8_t *direct_table;
485 uint8_t direct_cache[5*8];
486
487 uint8_t zigzag_scan[16];
488 uint8_t zigzag_scan8x8[64];
489 uint8_t zigzag_scan8x8_cavlc[64];
490 uint8_t field_scan[16];
491 uint8_t field_scan8x8[64];
492 uint8_t field_scan8x8_cavlc[64];
493 const uint8_t *zigzag_scan_q0;
494 const uint8_t *zigzag_scan8x8_q0;
495 const uint8_t *zigzag_scan8x8_cavlc_q0;
496 const uint8_t *field_scan_q0;
497 const uint8_t *field_scan8x8_q0;
498 const uint8_t *field_scan8x8_cavlc_q0;
499
500 int x264_build;
afebe2f7
501
502 /**
503 * @defgroup multithreading Members for slice based multithreading
504 * @{
505 */
506 struct H264Context *thread_context[MAX_THREADS];
507
508 /**
509 * current slice number, used to initalize slice_num of each thread/context
510 */
511 int current_slice;
512
513 /**
514 * Max number of threads / contexts.
515 * This is equal to AVCodecContext.thread_count unless
516 * multithreaded decoding is impossible, in which case it is
517 * reduced to 1.
518 */
519 int max_contexts;
520
521 /**
522 * 1 if the single thread fallback warning has already been
523 * displayed, 0 otherwise.
524 */
525 int single_decode_warning;
526
527 int last_slice_type;
528 /** @} */
529
64514ee8
AS
530 int mb_xy;
531
b09a7c05
532 /**
533 * pic_struct in picture timing SEI message
534 */
535 SEI_PicStructType sei_pic_struct;
89db0bae 536
37a558fe 537 /**
b19d493f
HY
538 * Complement sei_pic_struct
539 * SEI_PIC_STRUCT_TOP_BOTTOM and SEI_PIC_STRUCT_BOTTOM_TOP indicate interlaced frames.
540 * However, soft telecined frames may have these values.
541 * This is used in an attempt to flag soft telecine progressive.
542 */
543 int prev_interlaced_frame;
544
545 /**
70e01da3
IS
546 * Bit set of clock types for fields/frames in picture timing SEI message.
547 * For each found ct_type, appropriate bit is set (e.g., bit 1 for
548 * interlaced).
549 */
550 int sei_ct_type;
551
552 /**
ff594f81
IS
553 * dpb_output_delay in picture timing SEI message, see H.264 C.2.2
554 */
555 int sei_dpb_output_delay;
556
557 /**
cf6065ca
IS
558 * cpb_removal_delay in picture timing SEI message, see H.264 C.1.2
559 */
560 int sei_cpb_removal_delay;
561
562 /**
37a558fe
IS
563 * recovery_frame_cnt from SEI message
564 *
565 * Set to -1 if no recovery point SEI message found or to number of frames
566 * before playback synchronizes. Frames having recovery point are key
567 * frames.
568 */
569 int sei_recovery_frame_cnt;
570
89db0bae 571 int is_complex;
cb99c652
GB
572
573 int luma_weight_flag[2]; ///< 7.4.3.2 luma_weight_lX_flag
574 int chroma_weight_flag[2]; ///< 7.4.3.2 chroma_weight_lX_flag
2ea89d92
IS
575
576 // Timestamp stuff
577 int sei_buffering_period_present; ///< Buffering period SEI flag
578 int initial_cpb_removal_delay[32]; ///< Initial timestamps for CPBs
d7f5e520
MN
579
580 //SVQ3 specific fields
581 int halfpel_flag;
582 int thirdpel_flag;
583 int unknown_svq3_flag;
584 int next_slice_index;
585 uint32_t svq3_watermark_key;
26b4fe82
AJ
586}H264Context;
587
889fce8e
MN
588
589extern const uint8_t ff_h264_chroma_qp[52];
590
881b5b80
RC
591void ff_svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
592
593void ff_svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
889fce8e 594
1790a5e9
IS
595/**
596 * Decode SEI
597 */
598int ff_h264_decode_sei(H264Context *h);
599
600/**
601 * Decode SPS
602 */
603int ff_h264_decode_seq_parameter_set(H264Context *h);
604
605/**
606 * Decode PPS
607 */
608int ff_h264_decode_picture_parameter_set(H264Context *h, int bit_length);
609
610/**
611 * Decodes a network abstraction layer unit.
612 * @param consumed is the number of bytes used as input
613 * @param length is the length of the array
614 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp tailing?
615 * @returns decoded bytes, might be src+1 if no escapes
616 */
617const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src, int *dst_length, int *consumed, int length);
618
619/**
620 * identifies the exact end of the bitstream
621 * @return the length of the trailing, or 0 if damaged
622 */
623int ff_h264_decode_rbsp_trailing(H264Context *h, const uint8_t *src);
624
15861962
RD
625/**
626 * frees any data that may have been allocated in the H264 context like SPS, PPS etc.
627 */
cbf1eae9 628av_cold void ff_h264_free_context(H264Context *h);
15861962 629
75dd6938
LA
630/**
631 * reconstructs bitstream slice_type.
632 */
0dc343d4 633int ff_h264_get_slice_type(const H264Context *h);
75dd6938 634
903d58f6
MN
635/**
636 * allocates tables.
637 * needs width/height
638 */
639int ff_h264_alloc_tables(H264Context *h);
640
641/**
ea6f00c4
MN
642 * fills the default_ref_list.
643 */
644int ff_h264_fill_default_ref_list(H264Context *h);
645
646int ff_h264_decode_ref_pic_list_reordering(H264Context *h);
647void ff_h264_fill_mbaff_ref_list(H264Context *h);
648void ff_h264_remove_all_refs(H264Context *h);
649
650/**
651 * Executes the reference picture marking (memory management control operations).
652 */
653int ff_h264_execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count);
654
655int ff_h264_decode_ref_pic_marking(H264Context *h, GetBitContext *gb);
656
657
658/**
903d58f6
MN
659 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
660 */
2bedc0e8
MN
661int ff_h264_check_intra4x4_pred_mode(H264Context *h);
662
663/**
664 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
665 */
903d58f6
MN
666int ff_h264_check_intra_pred_mode(H264Context *h, int mode);
667
668void ff_h264_write_back_intra_pred_mode(H264Context *h);
669void ff_h264_hl_decode_mb(H264Context *h);
670int ff_h264_frame_start(H264Context *h);
671av_cold int ff_h264_decode_init(AVCodecContext *avctx);
672av_cold int ff_h264_decode_end(AVCodecContext *avctx);
e1e94902
MN
673av_cold void ff_h264_decode_init_vlc(void);
674
675/**
676 * decodes a macroblock
677 * @returns 0 if OK, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
678 */
679int ff_h264_decode_mb_cavlc(H264Context *h);
903d58f6 680
cc51b282
MN
681/**
682 * decodes a CABAC coded macroblock
683 * @returns 0 if OK, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
684 */
685int ff_h264_decode_mb_cabac(H264Context *h);
686
687void ff_h264_init_cabac_states(H264Context *h);
688
943f69a6
MN
689void ff_h264_direct_dist_scale_factor(H264Context * const h);
690void ff_h264_direct_ref_list_init(H264Context * const h);
691void ff_h264_pred_direct_motion(H264Context * const h, int *mb_type);
692
082cf971
MN
693void 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);
694void 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);
695
9c095463
MN
696/**
697 * Reset SEI values at the beginning of the frame.
698 *
699 * @param h H.264 context.
700 */
701void ff_h264_reset_sei(H264Context *h);
702
703
903d58f6
MN
704/*
705o-o o-o
706 / / /
707o-o o-o
708 ,---'
709o-o o-o
710 / / /
711o-o o-o
712*/
713//This table must be here because scan8[constant] must be known at compiletime
714static const uint8_t scan8[16 + 2*4]={
715 4+1*8, 5+1*8, 4+2*8, 5+2*8,
716 6+1*8, 7+1*8, 6+2*8, 7+2*8,
717 4+3*8, 5+3*8, 4+4*8, 5+4*8,
718 6+3*8, 7+3*8, 6+4*8, 7+4*8,
719 1+1*8, 2+1*8,
720 1+2*8, 2+2*8,
721 1+4*8, 2+4*8,
722 1+5*8, 2+5*8,
723};
724
725static av_always_inline uint32_t pack16to32(int a, int b){
726#if HAVE_BIGENDIAN
727 return (b&0xFFFF) + (a<<16);
728#else
729 return (a&0xFFFF) + (b<<16);
730#endif
731}
732
b5bd0700
MN
733static av_always_inline uint16_t pack8to16(int a, int b){
734#if HAVE_BIGENDIAN
735 return (b&0xFF) + (a<<8);
736#else
737 return (a&0xFF) + (b<<8);
738#endif
739}
740
903d58f6 741/**
082cf971
MN
742 * gets the chroma qp.
743 */
744static inline int get_chroma_qp(H264Context *h, int t, int qscale){
745 return h->pps.chroma_qp_table[t][qscale];
746}
747
e1e94902
MN
748static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my);
749
c1bb66ac 750static void fill_decode_neighbors(H264Context *h, int mb_type){
e1e94902
MN
751 MpegEncContext * const s = &h->s;
752 const int mb_xy= h->mb_xy;
753 int topleft_xy, top_xy, topright_xy, left_xy[2];
b6303e6d
MN
754 static const uint8_t left_block_options[4][16]={
755 {0,1,2,3,7,10,8,11,7+0*8, 7+1*8, 7+2*8, 7+3*8, 2+0*8, 2+3*8, 2+1*8, 2+2*8},
756 {2,2,3,3,8,11,8,11,7+2*8, 7+2*8, 7+3*8, 7+3*8, 2+1*8, 2+2*8, 2+1*8, 2+2*8},
757 {0,0,1,1,7,10,7,10,7+0*8, 7+0*8, 7+1*8, 7+1*8, 2+0*8, 2+3*8, 2+0*8, 2+3*8},
758 {0,2,0,2,7,10,7,10,7+0*8, 7+2*8, 7+0*8, 7+2*8, 2+0*8, 2+3*8, 2+0*8, 2+3*8}
e1e94902
MN
759 };
760
c1bb66ac
MN
761 h->topleft_partition= -1;
762
806ac67b 763 top_xy = mb_xy - (s->mb_stride << MB_FIELD);
e1e94902 764
e1e94902
MN
765 /* Wow, what a mess, why didn't they simplify the interlacing & intra
766 * stuff, I can't imagine that these complex rules are worth it. */
767
768 topleft_xy = top_xy - 1;
769 topright_xy= top_xy + 1;
770 left_xy[1] = left_xy[0] = mb_xy-1;
c1bb66ac 771 h->left_block = left_block_options[0];
e1e94902 772 if(FRAME_MBAFF){
806ac67b 773 const int left_mb_field_flag = IS_INTERLACED(s->current_picture.mb_type[mb_xy-1]);
e1e94902 774 const int curr_mb_field_flag = IS_INTERLACED(mb_type);
806ac67b
MN
775 if(s->mb_y&1){
776 if (left_mb_field_flag != curr_mb_field_flag) {
777 left_xy[1] = left_xy[0] = mb_xy - s->mb_stride - 1;
778 if (curr_mb_field_flag) {
779 left_xy[1] += s->mb_stride;
c1bb66ac 780 h->left_block = left_block_options[3];
806ac67b
MN
781 } else {
782 topleft_xy += s->mb_stride;
783 // take top left mv from the middle of the mb, as opposed to all other modes which use the bottom right partition
c1bb66ac
MN
784 h->topleft_partition = 0;
785 h->left_block = left_block_options[1];
806ac67b
MN
786 }
787 }
788 }else{
789 if(curr_mb_field_flag){
790 topleft_xy += s->mb_stride & (((s->current_picture.mb_type[top_xy - 1]>>7)&1)-1);
791 topright_xy += s->mb_stride & (((s->current_picture.mb_type[top_xy + 1]>>7)&1)-1);
792 top_xy += s->mb_stride & (((s->current_picture.mb_type[top_xy ]>>7)&1)-1);
793 }
794 if (left_mb_field_flag != curr_mb_field_flag) {
806ac67b
MN
795 if (curr_mb_field_flag) {
796 left_xy[1] += s->mb_stride;
c1bb66ac 797 h->left_block = left_block_options[3];
806ac67b 798 } else {
c1bb66ac 799 h->left_block = left_block_options[2];
806ac67b 800 }
e1e94902
MN
801 }
802 }
803 }
804
c1bb66ac
MN
805 h->topleft_mb_xy = topleft_xy;
806 h->top_mb_xy = top_xy;
807 h->topright_mb_xy= topright_xy;
e1e94902
MN
808 h->left_mb_xy[0] = left_xy[0];
809 h->left_mb_xy[1] = left_xy[1];
c1bb66ac 810 //FIXME do we need all in the context?
82fb5bb2
MN
811
812 h->topleft_type = s->current_picture.mb_type[topleft_xy] ;
813 h->top_type = s->current_picture.mb_type[top_xy] ;
814 h->topright_type= s->current_picture.mb_type[topright_xy];
815 h->left_type[0] = s->current_picture.mb_type[left_xy[0]] ;
816 h->left_type[1] = s->current_picture.mb_type[left_xy[1]] ;
817
818 if(h->slice_table[topleft_xy ] != h->slice_num) h->topleft_type = 0;
819 if(h->slice_table[top_xy ] != h->slice_num) h->top_type = 0;
820 if(h->slice_table[topright_xy] != h->slice_num) h->topright_type= 0;
ce9c6916 821 if(h->slice_table[left_xy[0] ] != h->slice_num) h->left_type[0] = h->left_type[1] = 0;
c1bb66ac
MN
822}
823
824static void fill_decode_caches(H264Context *h, int mb_type){
825 MpegEncContext * const s = &h->s;
c1bb66ac
MN
826 int topleft_xy, top_xy, topright_xy, left_xy[2];
827 int topleft_type, top_type, topright_type, left_type[2];
828 const uint8_t * left_block= h->left_block;
829 int i;
830
831 topleft_xy = h->topleft_mb_xy ;
832 top_xy = h->top_mb_xy ;
833 topright_xy = h->topright_mb_xy;
834 left_xy[0] = h->left_mb_xy[0] ;
835 left_xy[1] = h->left_mb_xy[1] ;
836 topleft_type = h->topleft_type ;
837 top_type = h->top_type ;
838 topright_type= h->topright_type ;
839 left_type[0] = h->left_type[0] ;
840 left_type[1] = h->left_type[1] ;
e1e94902 841
2e4362af 842 if(!IS_SKIP(mb_type)){
7a93858a
MN
843 if(IS_INTRA(mb_type)){
844 int type_mask= h->pps.constrained_intra_pred ? IS_INTRA(-1) : -1;
845 h->topleft_samples_available=
846 h->top_samples_available=
847 h->left_samples_available= 0xFFFF;
848 h->topright_samples_available= 0xEEEA;
849
850 if(!(top_type & type_mask)){
851 h->topleft_samples_available= 0xB3FF;
852 h->top_samples_available= 0x33FF;
853 h->topright_samples_available= 0x26EA;
854 }
855 if(IS_INTERLACED(mb_type) != IS_INTERLACED(left_type[0])){
856 if(IS_INTERLACED(mb_type)){
857 if(!(left_type[0] & type_mask)){
858 h->topleft_samples_available&= 0xDFFF;
859 h->left_samples_available&= 0x5FFF;
860 }
861 if(!(left_type[1] & type_mask)){
862 h->topleft_samples_available&= 0xFF5F;
863 h->left_samples_available&= 0xFF5F;
864 }
865 }else{
8f8497ae
MN
866 int left_typei = s->current_picture.mb_type[left_xy[0] + s->mb_stride];
867
7a93858a
MN
868 assert(left_xy[0] == left_xy[1]);
869 if(!((left_typei & type_mask) && (left_type[0] & type_mask))){
870 h->topleft_samples_available&= 0xDF5F;
871 h->left_samples_available&= 0x5F5F;
872 }
e1e94902
MN
873 }
874 }else{
7a93858a 875 if(!(left_type[0] & type_mask)){
e1e94902
MN
876 h->topleft_samples_available&= 0xDF5F;
877 h->left_samples_available&= 0x5F5F;
878 }
879 }
e1e94902 880
7a93858a
MN
881 if(!(topleft_type & type_mask))
882 h->topleft_samples_available&= 0x7FFF;
e1e94902 883
7a93858a
MN
884 if(!(topright_type & type_mask))
885 h->topright_samples_available&= 0xFBFF;
e1e94902 886
7a93858a
MN
887 if(IS_INTRA4x4(mb_type)){
888 if(IS_INTRA4x4(top_type)){
662a5b23 889 AV_COPY32(h->intra4x4_pred_mode_cache+4+8*0, h->intra4x4_pred_mode + h->mb2br_xy[top_xy]);
e1e94902 890 }else{
7a93858a
MN
891 h->intra4x4_pred_mode_cache[4+8*0]=
892 h->intra4x4_pred_mode_cache[5+8*0]=
893 h->intra4x4_pred_mode_cache[6+8*0]=
013202d7 894 h->intra4x4_pred_mode_cache[7+8*0]= 2 - 3*!(top_type & type_mask);
7a93858a
MN
895 }
896 for(i=0; i<2; i++){
897 if(IS_INTRA4x4(left_type[i])){
5b0fb524 898 int8_t *mode= h->intra4x4_pred_mode + h->mb2br_xy[left_xy[i]];
662a5b23
MN
899 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= mode[6-left_block[0+2*i]];
900 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= mode[6-left_block[1+2*i]];
7a93858a 901 }else{
7a93858a 902 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
013202d7 903 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= 2 - 3*!(left_type[i] & type_mask);
7a93858a 904 }
e1e94902
MN
905 }
906 }
907 }
e1e94902
MN
908
909
910/*
9110 . T T. T T T T
9121 L . .L . . . .
9132 L . .L . . . .
9143 . T TL . . . .
9154 L . .L . . . .
9165 L . .. . . . .
917*/
918//FIXME constraint_intra_pred & partitioning & nnz (let us hope this is just a typo in the spec)
919 if(top_type){
19769ece 920 AV_COPY32(&h->non_zero_count_cache[4+8*0], &h->non_zero_count[top_xy][4+3*8]);
7a93858a
MN
921 h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][1+1*8];
922 h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][2+1*8];
e1e94902 923
7a93858a
MN
924 h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][1+2*8];
925 h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][2+2*8];
599fe45b 926 }else {
7a93858a
MN
927 h->non_zero_count_cache[1+8*0]=
928 h->non_zero_count_cache[2+8*0]=
e1e94902 929
7a93858a
MN
930 h->non_zero_count_cache[1+8*3]=
931 h->non_zero_count_cache[2+8*3]=
19769ece 932 AV_WN32A(&h->non_zero_count_cache[4+8*0], CABAC && !IS_INTRA(mb_type) ? 0 : 0x40404040);
e1e94902
MN
933 }
934
935 for (i=0; i<2; i++) {
936 if(left_type[i]){
b6303e6d
MN
937 h->non_zero_count_cache[3+8*1 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[8+0+2*i]];
938 h->non_zero_count_cache[3+8*2 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[8+1+2*i]];
7a93858a
MN
939 h->non_zero_count_cache[0+8*1 + 8*i]= h->non_zero_count[left_xy[i]][left_block[8+4+2*i]];
940 h->non_zero_count_cache[0+8*4 + 8*i]= h->non_zero_count[left_xy[i]][left_block[8+5+2*i]];
599fe45b 941 }else{
7a93858a
MN
942 h->non_zero_count_cache[3+8*1 + 2*8*i]=
943 h->non_zero_count_cache[3+8*2 + 2*8*i]=
944 h->non_zero_count_cache[0+8*1 + 8*i]=
945 h->non_zero_count_cache[0+8*4 + 8*i]= CABAC && !IS_INTRA(mb_type) ? 0 : 64;
e1e94902
MN
946 }
947 }
948
599fe45b 949 if( CABAC ) {
e1e94902
MN
950 // top_cbp
951 if(top_type) {
952 h->top_cbp = h->cbp_table[top_xy];
e1e94902 953 } else {
59b5370f 954 h->top_cbp = IS_INTRA(mb_type) ? 0x1CF : 0x00F;
e1e94902
MN
955 }
956 // left_cbp
957 if (left_type[0]) {
cf7b67bc
MN
958 h->left_cbp = (h->cbp_table[left_xy[0]] & 0x1f0)
959 | ((h->cbp_table[left_xy[0]]>>(left_block[0]&(~1)))&2)
960 | (((h->cbp_table[left_xy[1]]>>(left_block[2]&(~1)))&2) << 2);
59b5370f
MN
961 } else {
962 h->left_cbp = IS_INTRA(mb_type) ? 0x1CF : 0x00F;
e1e94902
MN
963 }
964 }
2e4362af 965 }
e1e94902
MN
966
967#if 1
da452aca 968 if(IS_INTER(mb_type) || (IS_DIRECT(mb_type) && h->direct_spatial_mv_pred)){
e1e94902
MN
969 int list;
970 for(list=0; list<h->list_count; list++){
5ca43c25 971 if(!USES_LIST(mb_type, list)){
e1e94902
MN
972 /*if(!h->mv_cache_clean[list]){
973 memset(h->mv_cache [list], 0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
974 memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
975 h->mv_cache_clean[list]= 1;
976 }*/
977 continue;
978 }
da452aca
MN
979 assert(!(IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred));
980
e1e94902
MN
981 h->mv_cache_clean[list]= 0;
982
983 if(USES_LIST(top_type, list)){
984 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
0b69d625 985 AV_COPY128(h->mv_cache[list][scan8[0] + 0 - 1*8], s->current_picture.motion_val[list][b_xy + 0]);
7a93858a 986 h->ref_cache[list][scan8[0] + 0 - 1*8]=
358b5b1a 987 h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][4*top_xy + 2];
7a93858a 988 h->ref_cache[list][scan8[0] + 2 - 1*8]=
358b5b1a 989 h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][4*top_xy + 3];
e1e94902 990 }else{
0b69d625 991 AV_ZERO128(h->mv_cache[list][scan8[0] + 0 - 1*8]);
19769ece 992 AV_WN32A(&h->ref_cache[list][scan8[0] + 0 - 1*8], ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101);
e1e94902
MN
993 }
994
16b802fe 995 if(mb_type & (MB_TYPE_16x8|MB_TYPE_8x8)){
e1e94902
MN
996 for(i=0; i<2; i++){
997 int cache_idx = scan8[0] - 1 + i*2*8;
998 if(USES_LIST(left_type[i], list)){
999 const int b_xy= h->mb2b_xy[left_xy[i]] + 3;
358b5b1a 1000 const int b8_xy= 4*left_xy[i] + 1;
19769ece
MR
1001 AV_COPY32(h->mv_cache[list][cache_idx ], s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[0+i*2]]);
1002 AV_COPY32(h->mv_cache[list][cache_idx+8], s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[1+i*2]]);
358b5b1a
MN
1003 h->ref_cache[list][cache_idx ]= s->current_picture.ref_index[list][b8_xy + (left_block[0+i*2]&~1)];
1004 h->ref_cache[list][cache_idx+8]= s->current_picture.ref_index[list][b8_xy + (left_block[1+i*2]&~1)];
e1e94902 1005 }else{
19769ece
MR
1006 AV_ZERO32(h->mv_cache [list][cache_idx ]);
1007 AV_ZERO32(h->mv_cache [list][cache_idx+8]);
e1e94902 1008 h->ref_cache[list][cache_idx ]=
aebf3123 1009 h->ref_cache[list][cache_idx+8]= (left_type[i]) ? LIST_NOT_USED : PART_NOT_AVAILABLE;
e1e94902
MN
1010 }
1011 }
16b802fe
MN
1012 }else{
1013 if(USES_LIST(left_type[0], list)){
1014 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
1015 const int b8_xy= 4*left_xy[0] + 1;
1016 AV_COPY32(h->mv_cache[list][scan8[0] - 1], s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[0]]);
1017 h->ref_cache[list][scan8[0] - 1]= s->current_picture.ref_index[list][b8_xy + (left_block[0]&~1)];
1018 }else{
1019 AV_ZERO32(h->mv_cache [list][scan8[0] - 1]);
1020 h->ref_cache[list][scan8[0] - 1]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
1021 }
1022 }
e1e94902 1023
e1e94902
MN
1024 if(USES_LIST(topright_type, list)){
1025 const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
19769ece 1026 AV_COPY32(h->mv_cache[list][scan8[0] + 4 - 1*8], s->current_picture.motion_val[list][b_xy]);
358b5b1a 1027 h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][4*topright_xy + 2];
e1e94902 1028 }else{
19769ece 1029 AV_ZERO32(h->mv_cache [list][scan8[0] + 4 - 1*8]);
e1e94902
MN
1030 h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
1031 }
cf41a02b
MN
1032 if(h->ref_cache[list][scan8[0] + 4 - 1*8] < 0){
1033 if(USES_LIST(topleft_type, list)){
1034 const int b_xy = h->mb2b_xy [topleft_xy] + 3 + h->b_stride + (h->topleft_partition & 2*h->b_stride);
1035 const int b8_xy= 4*topleft_xy + 1 + (h->topleft_partition & 2);
1036 AV_COPY32(h->mv_cache[list][scan8[0] - 1 - 1*8], s->current_picture.motion_val[list][b_xy]);
1037 h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
1038 }else{
1039 AV_ZERO32(h->mv_cache[list][scan8[0] - 1 - 1*8]);
1040 h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
1041 }
1042 }
e1e94902 1043
bb770c5b 1044 if((mb_type&(MB_TYPE_SKIP|MB_TYPE_DIRECT2)) && !FRAME_MBAFF)
e1e94902
MN
1045 continue;
1046
e2b28acf 1047 if(!(mb_type&(MB_TYPE_SKIP|MB_TYPE_DIRECT2))) {
e1e94902
MN
1048 h->ref_cache[list][scan8[4 ]] =
1049 h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
19769ece
MR
1050 AV_ZERO32(h->mv_cache [list][scan8[4 ]]);
1051 AV_ZERO32(h->mv_cache [list][scan8[12]]);
e1e94902 1052
cb9285a2 1053 if( CABAC ) {
e1e94902
MN
1054 /* XXX beurk, Load mvd */
1055 if(USES_LIST(top_type, list)){
e1c88a21 1056 const int b_xy= h->mb2br_xy[top_xy];
b5bd0700 1057 AV_COPY64(h->mvd_cache[list][scan8[0] + 0 - 1*8], h->mvd_table[list][b_xy + 0]);
e1e94902 1058 }else{
b5bd0700 1059 AV_ZERO64(h->mvd_cache[list][scan8[0] + 0 - 1*8]);
e1e94902
MN
1060 }
1061 if(USES_LIST(left_type[0], list)){
e1c88a21
MN
1062 const int b_xy= h->mb2br_xy[left_xy[0]] + 6;
1063 AV_COPY16(h->mvd_cache[list][scan8[0] - 1 + 0*8], h->mvd_table[list][b_xy - left_block[0]]);
1064 AV_COPY16(h->mvd_cache[list][scan8[0] - 1 + 1*8], h->mvd_table[list][b_xy - left_block[1]]);
e1e94902 1065 }else{
b5bd0700
MN
1066 AV_ZERO16(h->mvd_cache [list][scan8[0] - 1 + 0*8]);
1067 AV_ZERO16(h->mvd_cache [list][scan8[0] - 1 + 1*8]);
e1e94902
MN
1068 }
1069 if(USES_LIST(left_type[1], list)){
e1c88a21
MN
1070 const int b_xy= h->mb2br_xy[left_xy[1]] + 6;
1071 AV_COPY16(h->mvd_cache[list][scan8[0] - 1 + 2*8], h->mvd_table[list][b_xy - left_block[2]]);
1072 AV_COPY16(h->mvd_cache[list][scan8[0] - 1 + 3*8], h->mvd_table[list][b_xy - left_block[3]]);
e1e94902 1073 }else{
b5bd0700
MN
1074 AV_ZERO16(h->mvd_cache [list][scan8[0] - 1 + 2*8]);
1075 AV_ZERO16(h->mvd_cache [list][scan8[0] - 1 + 3*8]);
e1e94902 1076 }
b5bd0700
MN
1077 AV_ZERO16(h->mvd_cache [list][scan8[4 ]]);
1078 AV_ZERO16(h->mvd_cache [list][scan8[12]]);
e1e94902 1079 if(h->slice_type_nos == FF_B_TYPE){
2dc380ca 1080 fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, MB_TYPE_16x16>>1, 1);
e1e94902
MN
1081
1082 if(IS_DIRECT(top_type)){
19769ece 1083 AV_WN32A(&h->direct_cache[scan8[0] - 1*8], 0x01010101*(MB_TYPE_DIRECT2>>1));
e1e94902 1084 }else if(IS_8X8(top_type)){
5e350863
MN
1085 int b8_xy = 4*top_xy;
1086 h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy + 2];
1087 h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 3];
e1e94902 1088 }else{
19769ece 1089 AV_WN32A(&h->direct_cache[scan8[0] - 1*8], 0x01010101*(MB_TYPE_16x16>>1));
e1e94902
MN
1090 }
1091
1092 if(IS_DIRECT(left_type[0]))
2dc380ca 1093 h->direct_cache[scan8[0] - 1 + 0*8]= MB_TYPE_DIRECT2>>1;
e1e94902 1094 else if(IS_8X8(left_type[0]))
5e350863 1095 h->direct_cache[scan8[0] - 1 + 0*8]= h->direct_table[4*left_xy[0] + 1 + (left_block[0]&~1)];
e1e94902 1096 else
2dc380ca 1097 h->direct_cache[scan8[0] - 1 + 0*8]= MB_TYPE_16x16>>1;
e1e94902
MN
1098
1099 if(IS_DIRECT(left_type[1]))
2dc380ca 1100 h->direct_cache[scan8[0] - 1 + 2*8]= MB_TYPE_DIRECT2>>1;
e1e94902 1101 else if(IS_8X8(left_type[1]))
5e350863 1102 h->direct_cache[scan8[0] - 1 + 2*8]= h->direct_table[4*left_xy[1] + 1 + (left_block[2]&~1)];
e1e94902 1103 else
2dc380ca 1104 h->direct_cache[scan8[0] - 1 + 2*8]= MB_TYPE_16x16>>1;
e1e94902
MN
1105 }
1106 }
cb9285a2 1107 }
e1e94902
MN
1108 if(FRAME_MBAFF){
1109#define MAP_MVS\
1110 MAP_F2F(scan8[0] - 1 - 1*8, topleft_type)\
1111 MAP_F2F(scan8[0] + 0 - 1*8, top_type)\
1112 MAP_F2F(scan8[0] + 1 - 1*8, top_type)\
1113 MAP_F2F(scan8[0] + 2 - 1*8, top_type)\
1114 MAP_F2F(scan8[0] + 3 - 1*8, top_type)\
1115 MAP_F2F(scan8[0] + 4 - 1*8, topright_type)\
1116 MAP_F2F(scan8[0] - 1 + 0*8, left_type[0])\
1117 MAP_F2F(scan8[0] - 1 + 1*8, left_type[0])\
1118 MAP_F2F(scan8[0] - 1 + 2*8, left_type[1])\
1119 MAP_F2F(scan8[0] - 1 + 3*8, left_type[1])
1120 if(MB_FIELD){
1121#define MAP_F2F(idx, mb_type)\
1122 if(!IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
1123 h->ref_cache[list][idx] <<= 1;\
1124 h->mv_cache[list][idx][1] /= 2;\
9127a369 1125 h->mvd_cache[list][idx][1] >>=1;\
e1e94902
MN
1126 }
1127 MAP_MVS
1128#undef MAP_F2F
1129 }else{
1130#define MAP_F2F(idx, mb_type)\
1131 if(IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
1132 h->ref_cache[list][idx] >>= 1;\
1133 h->mv_cache[list][idx][1] <<= 1;\
1134 h->mvd_cache[list][idx][1] <<= 1;\
1135 }
1136 MAP_MVS
1137#undef MAP_F2F
1138 }
1139 }
1140 }
1141 }
1142#endif
1143
7a93858a 1144 h->neighbor_transform_size= !!IS_8x8DCT(top_type) + !!IS_8x8DCT(left_type[0]);
f432b43b
MN
1145}
1146
44a5e7b6
MN
1147/**
1148 *
1149 * @returns non zero if the loop filter can be skiped
1150 */
1151static int fill_filter_caches(H264Context *h, int mb_type){
599fe45b
MN
1152 MpegEncContext * const s = &h->s;
1153 const int mb_xy= h->mb_xy;
1154 int top_xy, left_xy[2];
1155 int top_type, left_type[2];
599fe45b
MN
1156
1157 top_xy = mb_xy - (s->mb_stride << MB_FIELD);
1158
1159 //FIXME deblocking could skip the intra and nnz parts.
1160
1161 /* Wow, what a mess, why didn't they simplify the interlacing & intra
1162 * stuff, I can't imagine that these complex rules are worth it. */
1163
1164 left_xy[1] = left_xy[0] = mb_xy-1;
1165 if(FRAME_MBAFF){
1166 const int left_mb_field_flag = IS_INTERLACED(s->current_picture.mb_type[mb_xy-1]);
1167 const int curr_mb_field_flag = IS_INTERLACED(mb_type);
1168 if(s->mb_y&1){
1169 if (left_mb_field_flag != curr_mb_field_flag) {
8652e44a 1170 left_xy[0] -= s->mb_stride;
599fe45b
MN
1171 }
1172 }else{
1173 if(curr_mb_field_flag){
1174 top_xy += s->mb_stride & (((s->current_picture.mb_type[top_xy ]>>7)&1)-1);
1175 }
1176 if (left_mb_field_flag != curr_mb_field_flag) {
8652e44a 1177 left_xy[1] += s->mb_stride;
599fe45b
MN
1178 }
1179 }
1180 }
1181
1182 h->top_mb_xy = top_xy;
1183 h->left_mb_xy[0] = left_xy[0];
1184 h->left_mb_xy[1] = left_xy[1];
1185 {
1186 //for sufficiently low qp, filtering wouldn't do anything
1187 //this is a conservative estimate: could also check beta_offset and more accurate chroma_qp
1188 int qp_thresh = h->qp_thresh; //FIXME strictly we should store qp_thresh for each mb of a slice
1189 int qp = s->current_picture.qscale_table[mb_xy];
1190 if(qp <= qp_thresh
1191 && (left_xy[0]<0 || ((qp + s->current_picture.qscale_table[left_xy[0]] + 1)>>1) <= qp_thresh)
1192 && (top_xy < 0 || ((qp + s->current_picture.qscale_table[top_xy ] + 1)>>1) <= qp_thresh)){
1193 if(!FRAME_MBAFF)
1194 return 1;
1195 if( (left_xy[0]< 0 || ((qp + s->current_picture.qscale_table[left_xy[1] ] + 1)>>1) <= qp_thresh)
1196 && (top_xy < s->mb_stride || ((qp + s->current_picture.qscale_table[top_xy -s->mb_stride] + 1)>>1) <= qp_thresh))
1197 return 1;
1198 }
1199 }
1200
1201 if(h->deblocking_filter == 2){
1202 h->top_type = top_type = h->slice_table[top_xy ] == h->slice_num ? s->current_picture.mb_type[top_xy] : 0;
1203 h->left_type[0]= left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
1204 h->left_type[1]= left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
1205 }else{
1206 h->top_type = top_type = h->slice_table[top_xy ] < 0xFFFF ? s->current_picture.mb_type[top_xy] : 0;
1207 h->left_type[0]= left_type[0] = h->slice_table[left_xy[0] ] < 0xFFFF ? s->current_picture.mb_type[left_xy[0]] : 0;
1208 h->left_type[1]= left_type[1] = h->slice_table[left_xy[1] ] < 0xFFFF ? s->current_picture.mb_type[left_xy[1]] : 0;
1209 }
1210 if(IS_INTRA(mb_type))
1211 return 0;
1212
1213 AV_COPY64(&h->non_zero_count_cache[0+8*1], &h->non_zero_count[mb_xy][ 0]);
1214 AV_COPY64(&h->non_zero_count_cache[0+8*2], &h->non_zero_count[mb_xy][ 8]);
19769ece
MR
1215 AV_COPY32(&h->non_zero_count_cache[0+8*5], &h->non_zero_count[mb_xy][16]);
1216 AV_COPY32(&h->non_zero_count_cache[4+8*3], &h->non_zero_count[mb_xy][20]);
599fe45b
MN
1217 AV_COPY64(&h->non_zero_count_cache[0+8*4], &h->non_zero_count[mb_xy][24]);
1218
1219 h->cbp= h->cbp_table[mb_xy];
1220
1221 {
1222 int list;
1223 for(list=0; list<h->list_count; list++){
1224 int8_t *ref;
1225 int y, b_stride;
1226 int16_t (*mv_dst)[2];
1227 int16_t (*mv_src)[2];
1228
1229 if(!USES_LIST(mb_type, list)){
1230 fill_rectangle( h->mv_cache[list][scan8[0]], 4, 4, 8, pack16to32(0,0), 4);
19769ece
MR
1231 AV_WN32A(&h->ref_cache[list][scan8[ 0]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
1232 AV_WN32A(&h->ref_cache[list][scan8[ 2]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
1233 AV_WN32A(&h->ref_cache[list][scan8[ 8]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
1234 AV_WN32A(&h->ref_cache[list][scan8[10]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
599fe45b
MN
1235 continue;
1236 }
1237
358b5b1a 1238 ref = &s->current_picture.ref_index[list][4*mb_xy];
599fe45b 1239 {
056c5021 1240 int (*ref2frm)[64] = h->ref2frm[ h->slice_num&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
19769ece
MR
1241 AV_WN32A(&h->ref_cache[list][scan8[ 0]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
1242 AV_WN32A(&h->ref_cache[list][scan8[ 2]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
358b5b1a 1243 ref += 2;
19769ece
MR
1244 AV_WN32A(&h->ref_cache[list][scan8[ 8]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
1245 AV_WN32A(&h->ref_cache[list][scan8[10]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
599fe45b
MN
1246 }
1247
1248 b_stride = h->b_stride;
1249 mv_dst = &h->mv_cache[list][scan8[0]];
1250 mv_src = &s->current_picture.motion_val[list][4*s->mb_x + 4*s->mb_y*b_stride];
1251 for(y=0; y<4; y++){
1252 AV_COPY128(mv_dst + 8*y, mv_src + y*b_stride);
1253 }
1254
1255 }
1256 }
1257
1258
1259/*
12600 . T T. T T T T
12611 L . .L . . . .
12622 L . .L . . . .
12633 . T TL . . . .
12644 L . .L . . . .
12655 L . .. . . . .
1266*/
1267//FIXME constraint_intra_pred & partitioning & nnz (let us hope this is just a typo in the spec)
1268 if(top_type){
19769ece 1269 AV_COPY32(&h->non_zero_count_cache[4+8*0], &h->non_zero_count[top_xy][4+3*8]);
599fe45b
MN
1270 }
1271
1272 if(left_type[0]){
1273 h->non_zero_count_cache[3+8*1]= h->non_zero_count[left_xy[0]][7+0*8];
1274 h->non_zero_count_cache[3+8*2]= h->non_zero_count[left_xy[0]][7+1*8];
1275 h->non_zero_count_cache[3+8*3]= h->non_zero_count[left_xy[0]][7+2*8];
1276 h->non_zero_count_cache[3+8*4]= h->non_zero_count[left_xy[0]][7+3*8];
1277 }
1278
1279 // CAVLC 8x8dct requires NNZ values for residual decoding that differ from what the loop filter needs
1280 if(!CABAC && h->pps.transform_8x8_mode){
1281 if(IS_8x8DCT(top_type)){
1282 h->non_zero_count_cache[4+8*0]=
1283 h->non_zero_count_cache[5+8*0]= h->cbp_table[top_xy] & 4;
1284 h->non_zero_count_cache[6+8*0]=
1285 h->non_zero_count_cache[7+8*0]= h->cbp_table[top_xy] & 8;
1286 }
1287 if(IS_8x8DCT(left_type[0])){
1288 h->non_zero_count_cache[3+8*1]=
1289 h->non_zero_count_cache[3+8*2]= h->cbp_table[left_xy[0]]&2; //FIXME check MBAFF
1290 }
1291 if(IS_8x8DCT(left_type[1])){
1292 h->non_zero_count_cache[3+8*3]=
1293 h->non_zero_count_cache[3+8*4]= h->cbp_table[left_xy[1]]&8; //FIXME check MBAFF
1294 }
1295
1296 if(IS_8x8DCT(mb_type)){
1297 h->non_zero_count_cache[scan8[0 ]]= h->non_zero_count_cache[scan8[1 ]]=
1298 h->non_zero_count_cache[scan8[2 ]]= h->non_zero_count_cache[scan8[3 ]]= h->cbp & 1;
1299
1300 h->non_zero_count_cache[scan8[0+ 4]]= h->non_zero_count_cache[scan8[1+ 4]]=
1301 h->non_zero_count_cache[scan8[2+ 4]]= h->non_zero_count_cache[scan8[3+ 4]]= h->cbp & 2;
1302
1303 h->non_zero_count_cache[scan8[0+ 8]]= h->non_zero_count_cache[scan8[1+ 8]]=
1304 h->non_zero_count_cache[scan8[2+ 8]]= h->non_zero_count_cache[scan8[3+ 8]]= h->cbp & 4;
1305
1306 h->non_zero_count_cache[scan8[0+12]]= h->non_zero_count_cache[scan8[1+12]]=
1307 h->non_zero_count_cache[scan8[2+12]]= h->non_zero_count_cache[scan8[3+12]]= h->cbp & 8;
1308 }
1309 }
1310
1311 if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
1312 int list;
1313 for(list=0; list<h->list_count; list++){
1314 if(USES_LIST(top_type, list)){
1315 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
358b5b1a 1316 const int b8_xy= 4*top_xy + 2;
056c5021 1317 int (*ref2frm)[64] = h->ref2frm[ h->slice_table[top_xy]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
599fe45b
MN
1318 AV_COPY128(h->mv_cache[list][scan8[0] + 0 - 1*8], s->current_picture.motion_val[list][b_xy + 0]);
1319 h->ref_cache[list][scan8[0] + 0 - 1*8]=
1320 h->ref_cache[list][scan8[0] + 1 - 1*8]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 0]];
1321 h->ref_cache[list][scan8[0] + 2 - 1*8]=
1322 h->ref_cache[list][scan8[0] + 3 - 1*8]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 1]];
1323 }else{
1324 AV_ZERO128(h->mv_cache[list][scan8[0] + 0 - 1*8]);
19769ece 1325 AV_WN32A(&h->ref_cache[list][scan8[0] + 0 - 1*8], ((LIST_NOT_USED)&0xFF)*0x01010101u);
599fe45b
MN
1326 }
1327
1328 if(!IS_INTERLACED(mb_type^left_type[0])){
1329 if(USES_LIST(left_type[0], list)){
1330 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
358b5b1a 1331 const int b8_xy= 4*left_xy[0] + 1;
056c5021 1332 int (*ref2frm)[64] = h->ref2frm[ h->slice_table[left_xy[0]]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
19769ece
MR
1333 AV_COPY32(h->mv_cache[list][scan8[0] - 1 + 0 ], s->current_picture.motion_val[list][b_xy + h->b_stride*0]);
1334 AV_COPY32(h->mv_cache[list][scan8[0] - 1 + 8 ], s->current_picture.motion_val[list][b_xy + h->b_stride*1]);
1335 AV_COPY32(h->mv_cache[list][scan8[0] - 1 +16 ], s->current_picture.motion_val[list][b_xy + h->b_stride*2]);
1336 AV_COPY32(h->mv_cache[list][scan8[0] - 1 +24 ], s->current_picture.motion_val[list][b_xy + h->b_stride*3]);
599fe45b 1337 h->ref_cache[list][scan8[0] - 1 + 0 ]=
358b5b1a 1338 h->ref_cache[list][scan8[0] - 1 + 8 ]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 2*0]];
599fe45b 1339 h->ref_cache[list][scan8[0] - 1 +16 ]=
358b5b1a 1340 h->ref_cache[list][scan8[0] - 1 +24 ]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 2*1]];
599fe45b 1341 }else{
19769ece
MR
1342 AV_ZERO32(h->mv_cache [list][scan8[0] - 1 + 0 ]);
1343 AV_ZERO32(h->mv_cache [list][scan8[0] - 1 + 8 ]);
1344 AV_ZERO32(h->mv_cache [list][scan8[0] - 1 +16 ]);
1345 AV_ZERO32(h->mv_cache [list][scan8[0] - 1 +24 ]);
599fe45b
MN
1346 h->ref_cache[list][scan8[0] - 1 + 0 ]=
1347 h->ref_cache[list][scan8[0] - 1 + 8 ]=
1348 h->ref_cache[list][scan8[0] - 1 + 16 ]=
1349 h->ref_cache[list][scan8[0] - 1 + 24 ]= LIST_NOT_USED;
1350 }
1351 }
1352 }
1353 }
1354
1355 return 0;
f432b43b
MN
1356}
1357
e1e94902
MN
1358/**
1359 * gets the predicted intra4x4 prediction mode.
1360 */
1361static inline int pred_intra_mode(H264Context *h, int n){
1362 const int index8= scan8[n];
1363 const int left= h->intra4x4_pred_mode_cache[index8 - 1];
1364 const int top = h->intra4x4_pred_mode_cache[index8 - 8];
1365 const int min= FFMIN(left, top);
1366
1367 tprintf(h->s.avctx, "mode:%d %d min:%d\n", left ,top, min);
1368
1369 if(min<0) return DC_PRED;
1370 else return min;
1371}
1372
1373static inline void write_back_non_zero_count(H264Context *h){
1374 const int mb_xy= h->mb_xy;
1375
0b69d625
AS
1376 AV_COPY64(&h->non_zero_count[mb_xy][ 0], &h->non_zero_count_cache[0+8*1]);
1377 AV_COPY64(&h->non_zero_count[mb_xy][ 8], &h->non_zero_count_cache[0+8*2]);
19769ece
MR
1378 AV_COPY32(&h->non_zero_count[mb_xy][16], &h->non_zero_count_cache[0+8*5]);
1379 AV_COPY32(&h->non_zero_count[mb_xy][20], &h->non_zero_count_cache[4+8*3]);
0b69d625 1380 AV_COPY64(&h->non_zero_count[mb_xy][24], &h->non_zero_count_cache[0+8*4]);
e1e94902
MN
1381}
1382
1383static inline void write_back_motion(H264Context *h, int mb_type){
1384 MpegEncContext * const s = &h->s;
358b5b1a
MN
1385 const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride; //try mb2b(8)_xy
1386 const int b8_xy= 4*h->mb_xy;
e1e94902
MN
1387 int list;
1388
1389 if(!USES_LIST(mb_type, 0))
358b5b1a 1390 fill_rectangle(&s->current_picture.ref_index[0][b8_xy], 2, 2, 2, (uint8_t)LIST_NOT_USED, 1);
e1e94902
MN
1391
1392 for(list=0; list<h->list_count; list++){
0b69d625
AS
1393 int y, b_stride;
1394 int16_t (*mv_dst)[2];
1395 int16_t (*mv_src)[2];
1396
e1e94902
MN
1397 if(!USES_LIST(mb_type, list))
1398 continue;
1399
0b69d625
AS
1400 b_stride = h->b_stride;
1401 mv_dst = &s->current_picture.motion_val[list][b_xy];
1402 mv_src = &h->mv_cache[list][scan8[0]];
e1e94902 1403 for(y=0; y<4; y++){
0b69d625 1404 AV_COPY128(mv_dst + y*b_stride, mv_src + 8*y);
e1e94902 1405 }
ddd60f28 1406 if( CABAC ) {
e1c88a21 1407 uint8_t (*mvd_dst)[2] = &h->mvd_table[list][FMO ? 8*h->mb_xy : h->mb2br_xy[h->mb_xy]];
b5bd0700 1408 uint8_t (*mvd_src)[2] = &h->mvd_cache[list][scan8[0]];
e1e94902 1409 if(IS_SKIP(mb_type))
e1c88a21
MN
1410 AV_ZERO128(mvd_dst);
1411 else{
1412 AV_COPY64(mvd_dst, mvd_src + 8*3);
c2186cbd
MN
1413 AV_COPY16(mvd_dst + 3 + 3, mvd_src + 3 + 8*0);
1414 AV_COPY16(mvd_dst + 3 + 2, mvd_src + 3 + 8*1);
1415 AV_COPY16(mvd_dst + 3 + 1, mvd_src + 3 + 8*2);
e1e94902
MN
1416 }
1417 }
1418
1419 {
1420 int8_t *ref_index = &s->current_picture.ref_index[list][b8_xy];
358b5b1a
MN
1421 ref_index[0+0*2]= h->ref_cache[list][scan8[0]];
1422 ref_index[1+0*2]= h->ref_cache[list][scan8[4]];
1423 ref_index[0+1*2]= h->ref_cache[list][scan8[8]];
1424 ref_index[1+1*2]= h->ref_cache[list][scan8[12]];
e1e94902
MN
1425 }
1426 }
1427
ddd60f28 1428 if(h->slice_type_nos == FF_B_TYPE && CABAC){
e1e94902 1429 if(IS_8X8(mb_type)){
5e350863
MN
1430 uint8_t *direct_table = &h->direct_table[4*h->mb_xy];
1431 direct_table[1] = h->sub_mb_type[1]>>1;
1432 direct_table[2] = h->sub_mb_type[2]>>1;
1433 direct_table[3] = h->sub_mb_type[3]>>1;
e1e94902
MN
1434 }
1435 }
1436}
1437
1438static inline int get_dct8x8_allowed(H264Context *h){
1439 if(h->sps.direct_8x8_inference_flag)
19769ece 1440 return !(AV_RN64A(h->sub_mb_type) & ((MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_8x8 )*0x0001000100010001ULL));
e1e94902 1441 else
19769ece 1442 return !(AV_RN64A(h->sub_mb_type) & ((MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_8x8|MB_TYPE_DIRECT2)*0x0001000100010001ULL));
e1e94902
MN
1443}
1444
e1e94902
MN
1445/**
1446 * decodes a P_SKIP or B_SKIP macroblock
1447 */
1448static void decode_mb_skip(H264Context *h){
1449 MpegEncContext * const s = &h->s;
1450 const int mb_xy= h->mb_xy;
1451 int mb_type=0;
1452
c988f975 1453 memset(h->non_zero_count[mb_xy], 0, 32);
e1e94902
MN
1454 memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui
1455
1456 if(MB_FIELD)
1457 mb_type|= MB_TYPE_INTERLACED;
1458
1459 if( h->slice_type_nos == FF_B_TYPE )
1460 {
1461 // just for fill_caches. pred_direct_motion will set the real mb_type
5ca43c25 1462 mb_type|= MB_TYPE_L0L1|MB_TYPE_DIRECT2|MB_TYPE_SKIP;
c1bb66ac
MN
1463 if(h->direct_spatial_mv_pred){
1464 fill_decode_neighbors(h, mb_type);
f432b43b 1465 fill_decode_caches(h, mb_type); //FIXME check what is needed and what not ...
c1bb66ac 1466 }
e1e94902
MN
1467 ff_h264_pred_direct_motion(h, &mb_type);
1468 mb_type|= MB_TYPE_SKIP;
1469 }
1470 else
1471 {
1472 int mx, my;
1473 mb_type|= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0|MB_TYPE_SKIP;
1474
c1bb66ac 1475 fill_decode_neighbors(h, mb_type);
f432b43b 1476 fill_decode_caches(h, mb_type); //FIXME check what is needed and what not ...
e1e94902
MN
1477 pred_pskip_motion(h, &mx, &my);
1478 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
1479 fill_rectangle( h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4);
1480 }
1481
1482 write_back_motion(h, mb_type);
1483 s->current_picture.mb_type[mb_xy]= mb_type;
1484 s->current_picture.qscale_table[mb_xy]= s->qscale;
1485 h->slice_table[ mb_xy ]= h->slice_num;
1486 h->prev_mb_skipped= 1;
1487}
1488
7c2de274
MN
1489#include "h264_mvpred.h" //For pred_pskip_motion()
1490
98790382 1491#endif /* AVCODEC_H264_H */