The cvs version 1.103 of h264.c brokes 13 conformance streams, this
[libav.git] / libavcodec / h264.c
CommitLineData
0da71265
MN
1/*
2 * H.26L/H.264/AVC/JVT/14496-10/... encoder/decoder
3 * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
4 *
5 * This library is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU Lesser General Public
7 * License as published by the Free Software Foundation; either
8 * version 2 of the License, or (at your option) any later version.
9 *
10 * This library is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * Lesser General Public License for more details.
14 *
15 * You should have received a copy of the GNU Lesser General Public
16 * License along with this library; if not, write to the Free Software
17 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
18 *
19 */
20
21/**
22 * @file h264.c
23 * H.264 / AVC / MPEG4 part10 codec.
24 * @author Michael Niedermayer <michaelni@gmx.at>
25 */
26
27#include "common.h"
28#include "dsputil.h"
29#include "avcodec.h"
30#include "mpegvideo.h"
31#include "h264data.h"
32#include "golomb.h"
33
e5017ab8
LA
34#include "cabac.h"
35
0da71265
MN
36#undef NDEBUG
37#include <assert.h>
38
39#define interlaced_dct interlaced_dct_is_a_bad_name
40#define mb_intra mb_intra_isnt_initalized_see_mb_type
41
42#define LUMA_DC_BLOCK_INDEX 25
43#define CHROMA_DC_BLOCK_INDEX 26
44
45#define CHROMA_DC_COEFF_TOKEN_VLC_BITS 8
46#define COEFF_TOKEN_VLC_BITS 8
47#define TOTAL_ZEROS_VLC_BITS 9
48#define CHROMA_DC_TOTAL_ZEROS_VLC_BITS 3
49#define RUN_VLC_BITS 3
50#define RUN7_VLC_BITS 6
51
52#define MAX_SPS_COUNT 32
53#define MAX_PPS_COUNT 256
54
55#define MAX_MMCO_COUNT 66
56
57/**
58 * Sequence parameter set
59 */
60typedef struct SPS{
61
62 int profile_idc;
63 int level_idc;
0da71265
MN
64 int log2_max_frame_num; ///< log2_max_frame_num_minus4 + 4
65 int poc_type; ///< pic_order_cnt_type
66 int log2_max_poc_lsb; ///< log2_max_pic_order_cnt_lsb_minus4
67 int delta_pic_order_always_zero_flag;
68 int offset_for_non_ref_pic;
69 int offset_for_top_to_bottom_field;
70 int poc_cycle_length; ///< num_ref_frames_in_pic_order_cnt_cycle
71 int ref_frame_count; ///< num_ref_frames
a15e68de 72 int gaps_in_frame_num_allowed_flag;
0da71265
MN
73 int mb_width; ///< frame_width_in_mbs_minus1 + 1
74 int mb_height; ///< frame_height_in_mbs_minus1 + 1
75 int frame_mbs_only_flag;
76 int mb_aff; ///<mb_adaptive_frame_field_flag
77 int direct_8x8_inference_flag;
a15e68de
MN
78 int crop; ///< frame_cropping_flag
79 int crop_left; ///< frame_cropping_rect_left_offset
80 int crop_right; ///< frame_cropping_rect_right_offset
81 int crop_top; ///< frame_cropping_rect_top_offset
82 int crop_bottom; ///< frame_cropping_rect_bottom_offset
0da71265 83 int vui_parameters_present_flag;
5ff85f1d 84 AVRational sar;
ec587a50
MR
85 int timing_info_present_flag;
86 uint32_t num_units_in_tick;
87 uint32_t time_scale;
88 int fixed_frame_rate_flag;
0da71265 89 short offset_for_ref_frame[256]; //FIXME dyn aloc?
4e4d983e
LM
90 int bitstream_restriction_flag;
91 int num_reorder_frames;
0da71265
MN
92}SPS;
93
94/**
95 * Picture parameter set
96 */
97typedef struct PPS{
98 int sps_id;
99 int cabac; ///< entropy_coding_mode_flag
100 int pic_order_present; ///< pic_order_present_flag
101 int slice_group_count; ///< num_slice_groups_minus1 + 1
102 int mb_slice_group_map_type;
103 int ref_count[2]; ///< num_ref_idx_l0/1_active_minus1 + 1
104 int weighted_pred; ///< weighted_pred_flag
105 int weighted_bipred_idc;
106 int init_qp; ///< pic_init_qp_minus26 + 26
107 int init_qs; ///< pic_init_qs_minus26 + 26
108 int chroma_qp_index_offset;
109 int deblocking_filter_parameters_present; ///< deblocking_filter_parameters_present_flag
110 int constrained_intra_pred; ///< constrained_intra_pred_flag
111 int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag
0da71265
MN
112}PPS;
113
114/**
115 * Memory management control operation opcode.
116 */
117typedef enum MMCOOpcode{
118 MMCO_END=0,
119 MMCO_SHORT2UNUSED,
120 MMCO_LONG2UNUSED,
121 MMCO_SHORT2LONG,
122 MMCO_SET_MAX_LONG,
123 MMCO_RESET,
124 MMCO_LONG,
125} MMCOOpcode;
126
127/**
128 * Memory management control operation.
129 */
130typedef struct MMCO{
131 MMCOOpcode opcode;
132 int short_frame_num;
133 int long_index;
134} MMCO;
135
136/**
137 * H264Context
138 */
139typedef struct H264Context{
140 MpegEncContext s;
141 int nal_ref_idc;
142 int nal_unit_type;
143#define NAL_SLICE 1
144#define NAL_DPA 2
145#define NAL_DPB 3
146#define NAL_DPC 4
147#define NAL_IDR_SLICE 5
148#define NAL_SEI 6
149#define NAL_SPS 7
150#define NAL_PPS 8
151#define NAL_PICTURE_DELIMITER 9
152#define NAL_FILTER_DATA 10
153 uint8_t *rbsp_buffer;
154 int rbsp_buffer_size;
155
4770b1b4
RT
156 /**
157 * Used to parse AVC variant of h264
158 */
159 int is_avc; ///< this flag is != 0 if codec is avc1
160 int got_avcC; ///< flag used to parse avcC data only once
161 int nal_length_size; ///< Number of bytes used for nal length (1, 2 or 4)
162
0da71265
MN
163 int chroma_qp; //QPc
164
165 int prev_mb_skiped; //FIXME remove (IMHO not used)
166
167 //prediction stuff
168 int chroma_pred_mode;
169 int intra16x16_pred_mode;
170
171 int8_t intra4x4_pred_mode_cache[5*8];
172 int8_t (*intra4x4_pred_mode)[8];
173 void (*pred4x4 [9+3])(uint8_t *src, uint8_t *topright, int stride);//FIXME move to dsp?
174 void (*pred8x8 [4+3])(uint8_t *src, int stride);
175 void (*pred16x16[4+3])(uint8_t *src, int stride);
176 unsigned int topleft_samples_available;
177 unsigned int top_samples_available;
178 unsigned int topright_samples_available;
179 unsigned int left_samples_available;
53c05b1e
MN
180 uint8_t (*top_border)[16+2*8];
181 uint8_t left_border[17+2*9];
0da71265
MN
182
183 /**
184 * non zero coeff count cache.
185 * is 64 if not available.
186 */
bd9ea6a4 187 uint8_t non_zero_count_cache[6*8] __align8;
53c05b1e 188 uint8_t (*non_zero_count)[16];
0da71265
MN
189
190 /**
191 * Motion vector cache.
192 */
bd9ea6a4
MN
193 int16_t mv_cache[2][5*8][2] __align8;
194 int8_t ref_cache[2][5*8] __align8;
0da71265
MN
195#define LIST_NOT_USED -1 //FIXME rename?
196#define PART_NOT_AVAILABLE -2
197
198 /**
199 * is 1 if the specific list MV&references are set to 0,0,-2.
200 */
201 int mv_cache_clean[2];
202
6867a90b
LLL
203 /**
204 * block_offset[ 0..23] for frame macroblocks
205 * block_offset[24..47] for field macroblocks
206 */
207 int block_offset[2*(16+8)];
0da71265
MN
208
209 uint16_t *mb2b_xy; //FIXME are these 4 a good idea?
210 uint16_t *mb2b8_xy;
650cec0c 211 int b_stride; //FIXME use s->b4_stride
0da71265
MN
212 int b8_stride;
213
8b82a956
MN
214 int halfpel_flag;
215 int thirdpel_flag;
216
da3b9756
MM
217 int unknown_svq3_flag;
218 int next_slice_index;
219
0da71265
MN
220 SPS sps_buffer[MAX_SPS_COUNT];
221 SPS sps; ///< current sps
222
223 PPS pps_buffer[MAX_PPS_COUNT];
224 /**
225 * current pps
226 */
227 PPS pps; //FIXME move tp Picture perhaps? (->no) do we need that?
228
229 int slice_num;
230 uint8_t *slice_table_base;
231 uint8_t *slice_table; ///< slice_table_base + mb_stride + 1
232 int slice_type;
233 int slice_type_fixed;
234
235 //interlacing specific flags
236 int mb_field_decoding_flag;
237
238 int sub_mb_type[4];
239
240 //POC stuff
241 int poc_lsb;
242 int poc_msb;
243 int delta_poc_bottom;
244 int delta_poc[2];
245 int frame_num;
246 int prev_poc_msb; ///< poc_msb of the last reference pic for POC type 0
247 int prev_poc_lsb; ///< poc_lsb of the last reference pic for POC type 0
248 int frame_num_offset; ///< for POC type 2
249 int prev_frame_num_offset; ///< for POC type 2
250 int prev_frame_num; ///< frame_num of the last pic for POC type 1/2
251
252 /**
253 * frame_num for frames or 2*frame_num for field pics.
254 */
255 int curr_pic_num;
256
257 /**
258 * max_frame_num or 2*max_frame_num for field pics.
259 */
260 int max_pic_num;
261
262 //Weighted pred stuff
9f2d1b4f
LM
263 int use_weight;
264 int use_weight_chroma;
0da71265
MN
265 int luma_log2_weight_denom;
266 int chroma_log2_weight_denom;
267 int luma_weight[2][16];
268 int luma_offset[2][16];
269 int chroma_weight[2][16][2];
270 int chroma_offset[2][16][2];
9f2d1b4f 271 int implicit_weight[16][16];
0da71265
MN
272
273 //deblock
53c05b1e 274 int deblocking_filter; ///< disable_deblocking_filter_idc with 1<->0
980a82b7
MN
275 int slice_alpha_c0_offset;
276 int slice_beta_offset;
0da71265
MN
277
278 int redundant_pic_count;
279
280 int direct_spatial_mv_pred;
5ad984c9 281 int dist_scale_factor[16];
2f944356 282 int map_col_to_list0[2][16];
0da71265
MN
283
284 /**
285 * num_ref_idx_l0/1_active_minus1 + 1
286 */
287 int ref_count[2];// FIXME split for AFF
288 Picture *short_ref[16];
289 Picture *long_ref[16];
290 Picture default_ref_list[2][32];
291 Picture ref_list[2][32]; //FIXME size?
292 Picture field_ref_list[2][32]; //FIXME size?
a4dae92b 293 Picture *delayed_pic[16]; //FIXME size?
4e4d983e 294 Picture *delayed_output_pic;
0da71265
MN
295
296 /**
297 * memory management control operations buffer.
298 */
299 MMCO mmco[MAX_MMCO_COUNT];
300 int mmco_index;
301
302 int long_ref_count; ///< number of actual long term references
303 int short_ref_count; ///< number of actual short term references
304
305 //data partitioning
306 GetBitContext intra_gb;
307 GetBitContext inter_gb;
308 GetBitContext *intra_gb_ptr;
309 GetBitContext *inter_gb_ptr;
310
311 DCTELEM mb[16*24] __align8;
e5017ab8
LA
312
313 /**
314 * Cabac
315 */
316 CABACContext cabac;
317 uint8_t cabac_state[399];
318 int cabac_init_idc;
319
320 /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0,1,2), 0x0? luma_cbp */
321 uint16_t *cbp_table;
3981c385
MN
322 int top_cbp;
323 int left_cbp;
693399ba 324 /* chroma_pred_mode for i4x4 or i16x16, else 0 */
e5017ab8
LA
325 uint8_t *chroma_pred_mode_table;
326 int last_qscale_diff;
9e528114 327 int16_t (*mvd_table[2])[2];
bd9ea6a4 328 int16_t mvd_cache[2][5*8][2] __align8;
5ad984c9
LM
329 uint8_t *direct_table;
330 uint8_t direct_cache[5*8];
e5017ab8 331
0da71265
MN
332}H264Context;
333
334static VLC coeff_token_vlc[4];
335static VLC chroma_dc_coeff_token_vlc;
336
337static VLC total_zeros_vlc[15];
338static VLC chroma_dc_total_zeros_vlc[3];
339
340static VLC run_vlc[6];
341static VLC run7_vlc;
342
8b82a956
MN
343static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
344static void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
53c05b1e 345static void filter_mb( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr);
8b82a956 346
377ec888
MN
347static inline uint32_t pack16to32(int a, int b){
348#ifdef WORDS_BIGENDIAN
349 return (b&0xFFFF) + (a<<16);
350#else
351 return (a&0xFFFF) + (b<<16);
352#endif
353}
354
0da71265
MN
355/**
356 * fill a rectangle.
5175b937
LLL
357 * @param h height of the rectangle, should be a constant
358 * @param w width of the rectangle, should be a constant
0da71265
MN
359 * @param size the size of val (1 or 4), should be a constant
360 */
af6e2fed
MN
361static inline void fill_rectangle(void *vp, int w, int h, int stride, uint32_t val, int size){ //FIXME ensure this IS inlined
362 uint8_t *p= (uint8_t*)vp;
0da71265
MN
363 assert(size==1 || size==4);
364
365 w *= size;
366 stride *= size;
367
bd9ea6a4 368 assert((((int)vp)&(FFMIN(w, STRIDE_ALIGN)-1)) == 0);
0da71265
MN
369//FIXME check what gcc generates for 64 bit on x86 and possible write a 32 bit ver of it
370 if(w==2 && h==2){
371 *(uint16_t*)(p + 0)=
372 *(uint16_t*)(p + stride)= size==4 ? val : val*0x0101;
373 }else if(w==2 && h==4){
374 *(uint16_t*)(p + 0*stride)=
375 *(uint16_t*)(p + 1*stride)=
376 *(uint16_t*)(p + 2*stride)=
377 *(uint16_t*)(p + 3*stride)= size==4 ? val : val*0x0101;
f7a8c179
MN
378 }else if(w==4 && h==1){
379 *(uint32_t*)(p + 0*stride)= size==4 ? val : val*0x01010101;
0da71265
MN
380 }else if(w==4 && h==2){
381 *(uint32_t*)(p + 0*stride)=
382 *(uint32_t*)(p + 1*stride)= size==4 ? val : val*0x01010101;
383 }else if(w==4 && h==4){
384 *(uint32_t*)(p + 0*stride)=
385 *(uint32_t*)(p + 1*stride)=
386 *(uint32_t*)(p + 2*stride)=
387 *(uint32_t*)(p + 3*stride)= size==4 ? val : val*0x01010101;
388 }else if(w==8 && h==1){
389 *(uint32_t*)(p + 0)=
390 *(uint32_t*)(p + 4)= size==4 ? val : val*0x01010101;
391 }else if(w==8 && h==2){
392 *(uint32_t*)(p + 0 + 0*stride)=
393 *(uint32_t*)(p + 4 + 0*stride)=
394 *(uint32_t*)(p + 0 + 1*stride)=
395 *(uint32_t*)(p + 4 + 1*stride)= size==4 ? val : val*0x01010101;
396 }else if(w==8 && h==4){
397 *(uint64_t*)(p + 0*stride)=
398 *(uint64_t*)(p + 1*stride)=
399 *(uint64_t*)(p + 2*stride)=
400 *(uint64_t*)(p + 3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
401 }else if(w==16 && h==2){
402 *(uint64_t*)(p + 0+0*stride)=
403 *(uint64_t*)(p + 8+0*stride)=
404 *(uint64_t*)(p + 0+1*stride)=
405 *(uint64_t*)(p + 8+1*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
406 }else if(w==16 && h==4){
407 *(uint64_t*)(p + 0+0*stride)=
408 *(uint64_t*)(p + 8+0*stride)=
409 *(uint64_t*)(p + 0+1*stride)=
410 *(uint64_t*)(p + 8+1*stride)=
411 *(uint64_t*)(p + 0+2*stride)=
412 *(uint64_t*)(p + 8+2*stride)=
413 *(uint64_t*)(p + 0+3*stride)=
414 *(uint64_t*)(p + 8+3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
415 }else
416 assert(0);
417}
418
46f2f05f 419static inline void fill_caches(H264Context *h, int mb_type, int for_deblock){
0da71265 420 MpegEncContext * const s = &h->s;
7bc9090a 421 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
0da71265
MN
422 int topleft_xy, top_xy, topright_xy, left_xy[2];
423 int topleft_type, top_type, topright_type, left_type[2];
6867a90b 424 int left_block[8];
0da71265
MN
425 int i;
426
427 //wow what a mess, why didnt they simplify the interlacing&intra stuff, i cant imagine that these complex rules are worth it
428
6867a90b
LLL
429 top_xy = mb_xy - s->mb_stride;
430 topleft_xy = top_xy - 1;
431 topright_xy= top_xy + 1;
432 left_xy[1] = left_xy[0] = mb_xy-1;
433 left_block[0]= 0;
434 left_block[1]= 1;
435 left_block[2]= 2;
436 left_block[3]= 3;
437 left_block[4]= 7;
438 left_block[5]= 10;
439 left_block[6]= 8;
440 left_block[7]= 11;
0da71265 441 if(h->sps.mb_aff){
6867a90b
LLL
442 const int pair_xy = s->mb_x + (s->mb_y & ~1)*s->mb_stride;
443 const int top_pair_xy = pair_xy - s->mb_stride;
444 const int topleft_pair_xy = top_pair_xy - 1;
445 const int topright_pair_xy = top_pair_xy + 1;
446 const int topleft_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[topleft_pair_xy]);
447 const int top_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[top_pair_xy]);
448 const int topright_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[topright_pair_xy]);
449 const int left_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[pair_xy-1]);
450 const int curr_mb_frame_flag = !IS_INTERLACED(mb_type);
451 const int bottom = (s->mb_y & 1);
452 tprintf("fill_caches: curr_mb_frame_flag:%d, left_mb_frame_flag:%d, topleft_mb_frame_flag:%d, top_mb_frame_flag:%d, topright_mb_frame_flag:%d\n", curr_mb_frame_flag, left_mb_frame_flag, topleft_mb_frame_flag, top_mb_frame_flag, topright_mb_frame_flag);
453 if (bottom
454 ? !curr_mb_frame_flag // bottom macroblock
455 : (!curr_mb_frame_flag && !top_mb_frame_flag) // top macroblock
456 ) {
457 top_xy -= s->mb_stride;
458 }
459 if (bottom
460 ? !curr_mb_frame_flag // bottom macroblock
461 : (!curr_mb_frame_flag && !topleft_mb_frame_flag) // top macroblock
462 ) {
463 topleft_xy -= s->mb_stride;
464 }
465 if (bottom
466 ? !curr_mb_frame_flag // bottom macroblock
467 : (!curr_mb_frame_flag && !topright_mb_frame_flag) // top macroblock
468 ) {
469 topright_xy -= s->mb_stride;
470 }
471 if (left_mb_frame_flag != curr_mb_frame_flag) {
472 left_xy[1] = left_xy[0] = pair_xy - 1;
473 if (curr_mb_frame_flag) {
474 if (bottom) {
475 left_block[0]= 2;
476 left_block[1]= 2;
477 left_block[2]= 3;
478 left_block[3]= 3;
479 left_block[4]= 8;
480 left_block[5]= 11;
481 left_block[6]= 8;
482 left_block[7]= 11;
483 } else {
484 left_block[0]= 0;
485 left_block[1]= 0;
486 left_block[2]= 1;
487 left_block[3]= 1;
488 left_block[4]= 7;
489 left_block[5]= 10;
490 left_block[6]= 7;
491 left_block[7]= 10;
492 }
493 } else {
494 left_xy[1] += s->mb_stride;
495 //left_block[0]= 0;
496 left_block[1]= 2;
497 left_block[2]= 0;
498 left_block[3]= 2;
499 //left_block[4]= 7;
500 left_block[5]= 10;
501 left_block[6]= 7;
502 left_block[7]= 10;
503 }
504 }
0da71265
MN
505 }
506
46f2f05f
MN
507 if(for_deblock){
508 topleft_type = h->slice_table[topleft_xy ] < 255 ? s->current_picture.mb_type[topleft_xy] : 0;
509 top_type = h->slice_table[top_xy ] < 255 ? s->current_picture.mb_type[top_xy] : 0;
510 topright_type= h->slice_table[topright_xy] < 255 ? s->current_picture.mb_type[topright_xy]: 0;
511 left_type[0] = h->slice_table[left_xy[0] ] < 255 ? s->current_picture.mb_type[left_xy[0]] : 0;
512 left_type[1] = h->slice_table[left_xy[1] ] < 255 ? s->current_picture.mb_type[left_xy[1]] : 0;
513 }else{
514 topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
515 top_type = h->slice_table[top_xy ] == h->slice_num ? s->current_picture.mb_type[top_xy] : 0;
516 topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
517 left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
518 left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
519 }
0da71265
MN
520
521 if(IS_INTRA(mb_type)){
522 h->topleft_samples_available=
523 h->top_samples_available=
524 h->left_samples_available= 0xFFFF;
525 h->topright_samples_available= 0xEEEA;
526
527 if(!IS_INTRA(top_type) && (top_type==0 || h->pps.constrained_intra_pred)){
528 h->topleft_samples_available= 0xB3FF;
529 h->top_samples_available= 0x33FF;
530 h->topright_samples_available= 0x26EA;
531 }
532 for(i=0; i<2; i++){
533 if(!IS_INTRA(left_type[i]) && (left_type[i]==0 || h->pps.constrained_intra_pred)){
534 h->topleft_samples_available&= 0xDF5F;
535 h->left_samples_available&= 0x5F5F;
536 }
537 }
538
539 if(!IS_INTRA(topleft_type) && (topleft_type==0 || h->pps.constrained_intra_pred))
540 h->topleft_samples_available&= 0x7FFF;
541
542 if(!IS_INTRA(topright_type) && (topright_type==0 || h->pps.constrained_intra_pred))
543 h->topright_samples_available&= 0xFBFF;
544
545 if(IS_INTRA4x4(mb_type)){
546 if(IS_INTRA4x4(top_type)){
547 h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
548 h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
549 h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
550 h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
551 }else{
552 int pred;
6fbcaaa0 553 if(!top_type || (IS_INTER(top_type) && h->pps.constrained_intra_pred))
0da71265 554 pred= -1;
6fbcaaa0
LLL
555 else{
556 pred= 2;
0da71265
MN
557 }
558 h->intra4x4_pred_mode_cache[4+8*0]=
559 h->intra4x4_pred_mode_cache[5+8*0]=
560 h->intra4x4_pred_mode_cache[6+8*0]=
561 h->intra4x4_pred_mode_cache[7+8*0]= pred;
562 }
563 for(i=0; i<2; i++){
564 if(IS_INTRA4x4(left_type[i])){
565 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
566 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
567 }else{
568 int pred;
6fbcaaa0 569 if(!left_type[i] || (IS_INTER(left_type[i]) && h->pps.constrained_intra_pred))
0da71265 570 pred= -1;
6fbcaaa0
LLL
571 else{
572 pred= 2;
0da71265
MN
573 }
574 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
575 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
576 }
577 }
578 }
579 }
580
581
582/*
5830 . T T. T T T T
5841 L . .L . . . .
5852 L . .L . . . .
5863 . T TL . . . .
5874 L . .L . . . .
5885 L . .. . . . .
589*/
590//FIXME constraint_intra_pred & partitioning & nnz (lets hope this is just a typo in the spec)
591 if(top_type){
6867a90b
LLL
592 h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][4];
593 h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][5];
594 h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][6];
53c05b1e 595 h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3];
0da71265 596
6867a90b 597 h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][9];
53c05b1e 598 h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8];
0da71265 599
6867a90b 600 h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][12];
53c05b1e 601 h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11];
3981c385
MN
602
603 h->top_cbp= h->cbp_table[top_xy];
0da71265
MN
604 }else{
605 h->non_zero_count_cache[4+8*0]=
606 h->non_zero_count_cache[5+8*0]=
607 h->non_zero_count_cache[6+8*0]=
608 h->non_zero_count_cache[7+8*0]=
609
610 h->non_zero_count_cache[1+8*0]=
611 h->non_zero_count_cache[2+8*0]=
612
613 h->non_zero_count_cache[1+8*3]=
3981c385
MN
614 h->non_zero_count_cache[2+8*3]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
615
616 if(IS_INTRA(mb_type)) h->top_cbp= 0x1C0;
617 else h->top_cbp= 0;
0da71265
MN
618 }
619
6867a90b
LLL
620 for (i=0; i<2; i++) {
621 if(left_type[i]){
622 h->non_zero_count_cache[3+8*1 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[0+2*i]];
623 h->non_zero_count_cache[3+8*2 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[1+2*i]];
624 h->non_zero_count_cache[0+8*1 + 8*i]= h->non_zero_count[left_xy[i]][left_block[4+2*i]];
625 h->non_zero_count_cache[0+8*4 + 8*i]= h->non_zero_count[left_xy[i]][left_block[5+2*i]];
626 h->left_cbp= h->cbp_table[left_xy[i]]; //FIXME interlacing
627 }else{
628 h->non_zero_count_cache[3+8*1 + 2*8*i]=
629 h->non_zero_count_cache[3+8*2 + 2*8*i]=
630 h->non_zero_count_cache[0+8*1 + 8*i]=
631 h->non_zero_count_cache[0+8*4 + 8*i]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
632
633 if(IS_INTRA(mb_type)) h->left_cbp= 0x1C0;//FIXME interlacing
634 else h->left_cbp= 0;
635 }
0da71265 636 }
6867a90b 637
0da71265 638#if 1
5ad984c9
LM
639 //FIXME direct mb can skip much of this
640 if(IS_INTER(mb_type) || (IS_DIRECT(mb_type) && h->direct_spatial_mv_pred)){
0da71265
MN
641 int list;
642 for(list=0; list<2; list++){
b75e10d5 643 if(!USES_LIST(mb_type, list) && !IS_DIRECT(mb_type) && !for_deblock){
0da71265
MN
644 /*if(!h->mv_cache_clean[list]){
645 memset(h->mv_cache [list], 0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
646 memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
647 h->mv_cache_clean[list]= 1;
648 }*/
5ad984c9 649 continue;
0da71265
MN
650 }
651 h->mv_cache_clean[list]= 0;
652
653 if(IS_INTER(topleft_type)){
654 const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
655 const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + h->b8_stride;
656 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
657 h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
658 }else{
659 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
660 h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
661 }
662
663 if(IS_INTER(top_type)){
664 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
665 const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
666 *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0];
667 *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];
668 *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];
669 *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];
670 h->ref_cache[list][scan8[0] + 0 - 1*8]=
671 h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
672 h->ref_cache[list][scan8[0] + 2 - 1*8]=
673 h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
674 }else{
675 *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]=
676 *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]=
677 *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]=
678 *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;
679 *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
680 }
681
682 if(IS_INTER(topright_type)){
683 const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
684 const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
685 *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
686 h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
687 }else{
688 *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
689 h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
690 }
691
692 //FIXME unify cleanup or sth
693 if(IS_INTER(left_type[0])){
694 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
695 const int b8_xy= h->mb2b8_xy[left_xy[0]] + 1;
696 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 + 0*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[0]];
697 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 + 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[1]];
698 h->ref_cache[list][scan8[0] - 1 + 0*8]=
699 h->ref_cache[list][scan8[0] - 1 + 1*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0]>>1)];
700 }else{
701 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 0*8]=
702 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 1*8]= 0;
703 h->ref_cache[list][scan8[0] - 1 + 0*8]=
704 h->ref_cache[list][scan8[0] - 1 + 1*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
705 }
706
707 if(IS_INTER(left_type[1])){
708 const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
709 const int b8_xy= h->mb2b8_xy[left_xy[1]] + 1;
710 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 + 2*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[2]];
711 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 + 3*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[3]];
712 h->ref_cache[list][scan8[0] - 1 + 2*8]=
713 h->ref_cache[list][scan8[0] - 1 + 3*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[2]>>1)];
714 }else{
715 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 2*8]=
716 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 3*8]= 0;
717 h->ref_cache[list][scan8[0] - 1 + 2*8]=
718 h->ref_cache[list][scan8[0] - 1 + 3*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
719 }
720
46f2f05f
MN
721 if(for_deblock)
722 continue;
723
0da71265
MN
724 h->ref_cache[list][scan8[5 ]+1] =
725 h->ref_cache[list][scan8[7 ]+1] =
726 h->ref_cache[list][scan8[13]+1] = //FIXME remove past 3 (init somewher else)
727 h->ref_cache[list][scan8[4 ]] =
728 h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
729 *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
730 *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
731 *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewher else)
732 *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
733 *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
9e528114
LA
734
735 if( h->pps.cabac ) {
736 /* XXX beurk, Load mvd */
737 if(IS_INTER(topleft_type)){
738 const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
739 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy];
740 }else{
741 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 - 1*8]= 0;
742 }
743
744 if(IS_INTER(top_type)){
745 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
746 *(uint32_t*)h->mvd_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 0];
747 *(uint32_t*)h->mvd_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 1];
748 *(uint32_t*)h->mvd_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 2];
749 *(uint32_t*)h->mvd_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 3];
750 }else{
751 *(uint32_t*)h->mvd_cache [list][scan8[0] + 0 - 1*8]=
752 *(uint32_t*)h->mvd_cache [list][scan8[0] + 1 - 1*8]=
753 *(uint32_t*)h->mvd_cache [list][scan8[0] + 2 - 1*8]=
754 *(uint32_t*)h->mvd_cache [list][scan8[0] + 3 - 1*8]= 0;
755 }
756 if(IS_INTER(left_type[0])){
757 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
758 *(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]];
759 *(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]];
760 }else{
761 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 0*8]=
762 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 1*8]= 0;
763 }
764 if(IS_INTER(left_type[1])){
765 const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
766 *(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]];
767 *(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]];
768 }else{
769 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 2*8]=
770 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 3*8]= 0;
771 }
772 *(uint32_t*)h->mvd_cache [list][scan8[5 ]+1]=
773 *(uint32_t*)h->mvd_cache [list][scan8[7 ]+1]=
774 *(uint32_t*)h->mvd_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewher else)
775 *(uint32_t*)h->mvd_cache [list][scan8[4 ]]=
776 *(uint32_t*)h->mvd_cache [list][scan8[12]]= 0;
5ad984c9
LM
777
778 if(h->slice_type == B_TYPE){
779 fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, 0, 1);
780
781 if(IS_DIRECT(top_type)){
782 *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0x01010101;
783 }else if(IS_8X8(top_type)){
784 int b8_xy = h->mb2b8_xy[top_xy] + h->b8_stride;
785 h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy];
786 h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 1];
787 }else{
788 *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0;
789 }
790
791 //FIXME interlacing
792 if(IS_DIRECT(left_type[0])){
793 h->direct_cache[scan8[0] - 1 + 0*8]=
794 h->direct_cache[scan8[0] - 1 + 2*8]= 1;
795 }else if(IS_8X8(left_type[0])){
796 int b8_xy = h->mb2b8_xy[left_xy[0]] + 1;
797 h->direct_cache[scan8[0] - 1 + 0*8]= h->direct_table[b8_xy];
798 h->direct_cache[scan8[0] - 1 + 2*8]= h->direct_table[b8_xy + h->b8_stride];
799 }else{
800 h->direct_cache[scan8[0] - 1 + 0*8]=
801 h->direct_cache[scan8[0] - 1 + 2*8]= 0;
802 }
803 }
9e528114 804 }
0da71265 805 }
0da71265
MN
806 }
807#endif
808}
809
810static inline void write_back_intra_pred_mode(H264Context *h){
811 MpegEncContext * const s = &h->s;
7bc9090a 812 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
0da71265
MN
813
814 h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
815 h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
816 h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
817 h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
818 h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
819 h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
820 h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
821}
822
823/**
824 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
825 */
826static inline int check_intra4x4_pred_mode(H264Context *h){
827 MpegEncContext * const s = &h->s;
828 static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
829 static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
830 int i;
831
832 if(!(h->top_samples_available&0x8000)){
833 for(i=0; i<4; i++){
834 int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
835 if(status<0){
9b879566 836 av_log(h->s.avctx, AV_LOG_ERROR, "top block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y);
0da71265
MN
837 return -1;
838 } else if(status){
839 h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
840 }
841 }
842 }
843
844 if(!(h->left_samples_available&0x8000)){
845 for(i=0; i<4; i++){
846 int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
847 if(status<0){
9b879566 848 av_log(h->s.avctx, AV_LOG_ERROR, "left block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y);
0da71265
MN
849 return -1;
850 } else if(status){
851 h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
852 }
853 }
854 }
855
856 return 0;
857} //FIXME cleanup like next
858
859/**
860 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
861 */
862static inline int check_intra_pred_mode(H264Context *h, int mode){
863 MpegEncContext * const s = &h->s;
864 static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
865 static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
866
5175b937
LLL
867 if(mode < 0 || mode > 6) {
868 av_log(h->s.avctx, AV_LOG_ERROR, "out of range intra chroma pred mode at %d %d\n", s->mb_x, s->mb_y);
7440fe83 869 return -1;
5175b937 870 }
7440fe83 871
0da71265
MN
872 if(!(h->top_samples_available&0x8000)){
873 mode= top[ mode ];
874 if(mode<0){
9b879566 875 av_log(h->s.avctx, AV_LOG_ERROR, "top block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);
0da71265
MN
876 return -1;
877 }
878 }
879
880 if(!(h->left_samples_available&0x8000)){
881 mode= left[ mode ];
882 if(mode<0){
9b879566 883 av_log(h->s.avctx, AV_LOG_ERROR, "left block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);
0da71265
MN
884 return -1;
885 }
886 }
887
888 return mode;
889}
890
891/**
892 * gets the predicted intra4x4 prediction mode.
893 */
894static inline int pred_intra_mode(H264Context *h, int n){
895 const int index8= scan8[n];
896 const int left= h->intra4x4_pred_mode_cache[index8 - 1];
897 const int top = h->intra4x4_pred_mode_cache[index8 - 8];
898 const int min= FFMIN(left, top);
899
95c26348 900 tprintf("mode:%d %d min:%d\n", left ,top, min);
0da71265
MN
901
902 if(min<0) return DC_PRED;
903 else return min;
904}
905
906static inline void write_back_non_zero_count(H264Context *h){
907 MpegEncContext * const s = &h->s;
7bc9090a 908 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
0da71265 909
6867a90b
LLL
910 h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[7+8*1];
911 h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[7+8*2];
912 h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[7+8*3];
53c05b1e 913 h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];
6867a90b
LLL
914 h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[4+8*4];
915 h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[5+8*4];
916 h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[6+8*4];
53c05b1e 917
6867a90b 918 h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[1+8*2];
53c05b1e 919 h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
6867a90b 920 h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[2+8*1];
53c05b1e 921
6867a90b 922 h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[1+8*5];
53c05b1e 923 h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
6867a90b 924 h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[2+8*4];
0da71265
MN
925}
926
927/**
928 * gets the predicted number of non zero coefficients.
929 * @param n block index
930 */
931static inline int pred_non_zero_count(H264Context *h, int n){
932 const int index8= scan8[n];
933 const int left= h->non_zero_count_cache[index8 - 1];
934 const int top = h->non_zero_count_cache[index8 - 8];
935 int i= left + top;
936
937 if(i<64) i= (i+1)>>1;
938
95c26348 939 tprintf("pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
0da71265
MN
940
941 return i&31;
942}
943
1924f3ce
MN
944static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
945 const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];
946
947 if(topright_ref != PART_NOT_AVAILABLE){
948 *C= h->mv_cache[list][ i - 8 + part_width ];
949 return topright_ref;
950 }else{
95c26348
MN
951 tprintf("topright MV not available\n");
952
1924f3ce
MN
953 *C= h->mv_cache[list][ i - 8 - 1 ];
954 return h->ref_cache[list][ i - 8 - 1 ];
955 }
956}
957
0da71265
MN
958/**
959 * gets the predicted MV.
960 * @param n the block index
961 * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)
962 * @param mx the x component of the predicted motion vector
963 * @param my the y component of the predicted motion vector
964 */
965static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
0da71265
MN
966 const int index8= scan8[n];
967 const int top_ref= h->ref_cache[list][ index8 - 8 ];
0da71265
MN
968 const int left_ref= h->ref_cache[list][ index8 - 1 ];
969 const int16_t * const A= h->mv_cache[list][ index8 - 1 ];
970 const int16_t * const B= h->mv_cache[list][ index8 - 8 ];
1924f3ce
MN
971 const int16_t * C;
972 int diagonal_ref, match_count;
973
0da71265 974 assert(part_width==1 || part_width==2 || part_width==4);
1924f3ce 975
0da71265
MN
976/* mv_cache
977 B . . A T T T T
978 U . . L . . , .
979 U . . L . . . .
980 U . . L . . , .
981 . . . L . . . .
982*/
1924f3ce
MN
983
984 diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);
985 match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);
827c91bf 986 tprintf("pred_motion match_count=%d\n", match_count);
1924f3ce
MN
987 if(match_count > 1){ //most common
988 *mx= mid_pred(A[0], B[0], C[0]);
989 *my= mid_pred(A[1], B[1], C[1]);
990 }else if(match_count==1){
991 if(left_ref==ref){
992 *mx= A[0];
993 *my= A[1];
994 }else if(top_ref==ref){
995 *mx= B[0];
996 *my= B[1];
0da71265 997 }else{
1924f3ce
MN
998 *mx= C[0];
999 *my= C[1];
0da71265
MN
1000 }
1001 }else{
1924f3ce 1002 if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
0da71265 1003 *mx= A[0];
1924f3ce 1004 *my= A[1];
0da71265 1005 }else{
1924f3ce
MN
1006 *mx= mid_pred(A[0], B[0], C[0]);
1007 *my= mid_pred(A[1], B[1], C[1]);
0da71265 1008 }
0da71265 1009 }
1924f3ce 1010
af6e2fed 1011 tprintf("pred_motion (%2d %2d %2d) (%2d %2d %2d) (%2d %2d %2d) -> (%2d %2d %2d) at %2d %2d %d list %d\n", top_ref, B[0], B[1], diagonal_ref, C[0], C[1], left_ref, A[0], A[1], ref, *mx, *my, h->s.mb_x, h->s.mb_y, n, list);
0da71265
MN
1012}
1013
1014/**
1015 * gets the directionally predicted 16x8 MV.
1016 * @param n the block index
1017 * @param mx the x component of the predicted motion vector
1018 * @param my the y component of the predicted motion vector
1019 */
1020static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
0da71265
MN
1021 if(n==0){
1022 const int top_ref= h->ref_cache[list][ scan8[0] - 8 ];
1023 const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
1024
579448a0 1025 tprintf("pred_16x8: (%2d %2d %2d) at %2d %2d %d list %d\n", top_ref, B[0], B[1], h->s.mb_x, h->s.mb_y, n, list);
0da71265
MN
1026
1027 if(top_ref == ref){
1028 *mx= B[0];
1029 *my= B[1];
1030 return;
1031 }
1032 }else{
1033 const int left_ref= h->ref_cache[list][ scan8[8] - 1 ];
1034 const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];
1035
579448a0 1036 tprintf("pred_16x8: (%2d %2d %2d) at %2d %2d %d list %d\n", left_ref, A[0], A[1], h->s.mb_x, h->s.mb_y, n, list);
0da71265
MN
1037
1038 if(left_ref == ref){
1039 *mx= A[0];
1040 *my= A[1];
1041 return;
1042 }
1043 }
1044
1045 //RARE
1046 pred_motion(h, n, 4, list, ref, mx, my);
1047}
1048
1049/**
1050 * gets the directionally predicted 8x16 MV.
1051 * @param n the block index
1052 * @param mx the x component of the predicted motion vector
1053 * @param my the y component of the predicted motion vector
1054 */
1055static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
0da71265
MN
1056 if(n==0){
1057 const int left_ref= h->ref_cache[list][ scan8[0] - 1 ];
1058 const int16_t * const A= h->mv_cache[list][ scan8[0] - 1 ];
1059
579448a0 1060 tprintf("pred_8x16: (%2d %2d %2d) at %2d %2d %d list %d\n", left_ref, A[0], A[1], h->s.mb_x, h->s.mb_y, n, list);
0da71265
MN
1061
1062 if(left_ref == ref){
1063 *mx= A[0];
1064 *my= A[1];
1065 return;
1066 }
1067 }else{
1924f3ce
MN
1068 const int16_t * C;
1069 int diagonal_ref;
1070
1071 diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);
0da71265 1072
579448a0 1073 tprintf("pred_8x16: (%2d %2d %2d) at %2d %2d %d list %d\n", diagonal_ref, C[0], C[1], h->s.mb_x, h->s.mb_y, n, list);
0da71265 1074
1924f3ce 1075 if(diagonal_ref == ref){
0da71265
MN
1076 *mx= C[0];
1077 *my= C[1];
1078 return;
1079 }
0da71265
MN
1080 }
1081
1082 //RARE
1083 pred_motion(h, n, 2, list, ref, mx, my);
1084}
1085
1086static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
0da71265
MN
1087 const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
1088 const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];
1089
5175b937 1090 tprintf("pred_pskip: (%d) (%d) at %2d %2d\n", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
0da71265
MN
1091
1092 if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
1093 || (top_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0)
1094 || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){
1095
1096 *mx = *my = 0;
1097 return;
1098 }
1099
1100 pred_motion(h, 0, 4, 0, 0, mx, my);
1101
1102 return;
1103}
1104
5ad984c9
LM
1105static inline void direct_dist_scale_factor(H264Context * const h){
1106 const int poc = h->s.current_picture_ptr->poc;
1107 const int poc1 = h->ref_list[1][0].poc;
1108 int i;
1109 for(i=0; i<h->ref_count[0]; i++){
1110 int poc0 = h->ref_list[0][i].poc;
1111 int td = clip(poc1 - poc0, -128, 127);
1112 if(td == 0 /* FIXME || pic0 is a long-term ref */){
1113 h->dist_scale_factor[i] = 256;
1114 }else{
1115 int tb = clip(poc - poc0, -128, 127);
1116 int tx = (16384 + (ABS(td) >> 1)) / td;
1117 h->dist_scale_factor[i] = clip((tb*tx + 32) >> 6, -1024, 1023);
1118 }
1119 }
1120}
2f944356
LM
1121static inline void direct_ref_list_init(H264Context * const h){
1122 MpegEncContext * const s = &h->s;
1123 Picture * const ref1 = &h->ref_list[1][0];
1124 Picture * const cur = s->current_picture_ptr;
1125 int list, i, j;
1126 if(cur->pict_type == I_TYPE)
1127 cur->ref_count[0] = 0;
1128 if(cur->pict_type != B_TYPE)
1129 cur->ref_count[1] = 0;
1130 for(list=0; list<2; list++){
1131 cur->ref_count[list] = h->ref_count[list];
1132 for(j=0; j<h->ref_count[list]; j++)
1133 cur->ref_poc[list][j] = h->ref_list[list][j].poc;
1134 }
1135 if(cur->pict_type != B_TYPE || h->direct_spatial_mv_pred)
1136 return;
1137 for(list=0; list<2; list++){
1138 for(i=0; i<ref1->ref_count[list]; i++){
1139 const int poc = ref1->ref_poc[list][i];
1140 h->map_col_to_list0[list][i] = PART_NOT_AVAILABLE;
1141 for(j=0; j<h->ref_count[list]; j++)
1142 if(h->ref_list[list][j].poc == poc){
1143 h->map_col_to_list0[list][i] = j;
1144 break;
1145 }
1146 }
1147 }
1148}
5ad984c9
LM
1149
1150static inline void pred_direct_motion(H264Context * const h, int *mb_type){
1151 MpegEncContext * const s = &h->s;
1152 const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
1153 const int b8_xy = 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1154 const int b4_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1155 const int mb_type_col = h->ref_list[1][0].mb_type[mb_xy];
1156 const int16_t (*l1mv0)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[0][b4_xy];
1157 const int8_t *l1ref0 = &h->ref_list[1][0].ref_index[0][b8_xy];
2f944356 1158 const int8_t *l1ref1 = &h->ref_list[1][0].ref_index[1][b8_xy];
5ad984c9
LM
1159 const int is_b8x8 = IS_8X8(*mb_type);
1160 int sub_mb_type;
1161 int i8, i4;
1162
1163 if(IS_8X8(mb_type_col) && !h->sps.direct_8x8_inference_flag){
1164 /* FIXME save sub mb types from previous frames (or derive from MVs)
1165 * so we know exactly what block size to use */
1166 sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */
3622988f 1167 *mb_type = MB_TYPE_8x8|MB_TYPE_L0L1;
5ad984c9
LM
1168 }else if(!is_b8x8 && (IS_16X16(mb_type_col) || IS_INTRA(mb_type_col))){
1169 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1170 *mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */
1171 }else{
1172 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
3622988f 1173 *mb_type = MB_TYPE_8x8|MB_TYPE_L0L1;
5ad984c9
LM
1174 }
1175 if(!is_b8x8)
1176 *mb_type |= MB_TYPE_DIRECT2;
1177
827c91bf
LLL
1178 tprintf("mb_type = %08x, sub_mb_type = %08x, is_b8x8 = %d, mb_type_col = %08x\n", *mb_type, sub_mb_type, is_b8x8, mb_type_col);
1179
5ad984c9
LM
1180 if(h->direct_spatial_mv_pred){
1181 int ref[2];
1182 int mv[2][2];
1183 int list;
1184
1185 /* ref = min(neighbors) */
1186 for(list=0; list<2; list++){
1187 int refa = h->ref_cache[list][scan8[0] - 1];
1188 int refb = h->ref_cache[list][scan8[0] - 8];
1189 int refc = h->ref_cache[list][scan8[0] - 8 + 4];
1190 if(refc == -2)
1191 refc = h->ref_cache[list][scan8[0] - 8 - 1];
1192 ref[list] = refa;
1193 if(ref[list] < 0 || (refb < ref[list] && refb >= 0))
1194 ref[list] = refb;
1195 if(ref[list] < 0 || (refc < ref[list] && refc >= 0))
1196 ref[list] = refc;
1197 if(ref[list] < 0)
1198 ref[list] = -1;
1199 }
1200
1201 if(ref[0] < 0 && ref[1] < 0){
1202 ref[0] = ref[1] = 0;
1203 mv[0][0] = mv[0][1] =
1204 mv[1][0] = mv[1][1] = 0;
1205 }else{
1206 for(list=0; list<2; list++){
1207 if(ref[list] >= 0)
1208 pred_motion(h, 0, 4, list, ref[list], &mv[list][0], &mv[list][1]);
1209 else
1210 mv[list][0] = mv[list][1] = 0;
1211 }
1212 }
1213
1214 if(ref[1] < 0){
1215 *mb_type &= ~MB_TYPE_P0L1;
1216 sub_mb_type &= ~MB_TYPE_P0L1;
1217 }else if(ref[0] < 0){
1218 *mb_type &= ~MB_TYPE_P0L0;
1219 sub_mb_type &= ~MB_TYPE_P0L0;
1220 }
1221
1222 if(IS_16X16(*mb_type)){
1223 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref[0], 1);
1224 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, ref[1], 1);
1225 if(!IS_INTRA(mb_type_col) && l1ref0[0] == 0 &&
1226 ABS(l1mv0[0][0]) <= 1 && ABS(l1mv0[0][1]) <= 1){
1227 if(ref[0] > 0)
1228 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1229 else
1230 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
1231 if(ref[1] > 0)
1232 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1233 else
1234 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
1235 }else{
1236 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1237 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1238 }
1239 }else{
1240 for(i8=0; i8<4; i8++){
1241 const int x8 = i8&1;
1242 const int y8 = i8>>1;
1243
1244 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1245 continue;
1246 h->sub_mb_type[i8] = sub_mb_type;
1247
1248 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1249 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1250 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref[0], 1);
1251 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, ref[1], 1);
1252
1253 /* col_zero_flag */
1254 if(!IS_INTRA(mb_type_col) && l1ref0[x8 + y8*h->b8_stride] == 0){
1255 for(i4=0; i4<4; i4++){
1256 const int16_t *mv_col = l1mv0[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1257 if(ABS(mv_col[0]) <= 1 && ABS(mv_col[1]) <= 1){
1258 if(ref[0] == 0)
1259 *(uint32_t*)h->mv_cache[0][scan8[i8*4+i4]] = 0;
1260 if(ref[1] == 0)
1261 *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = 0;
1262 }
1263 }
1264 }
1265 }
1266 }
1267 }else{ /* direct temporal mv pred */
5ad984c9
LM
1268 if(IS_16X16(*mb_type)){
1269 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1);
1270 if(IS_INTRA(mb_type_col)){
1271 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
1272 fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
1273 fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
1274 }else{
2f944356
LM
1275 const int ref0 = l1ref0[0] >= 0 ? h->map_col_to_list0[0][l1ref0[0]]
1276 : h->map_col_to_list0[1][l1ref1[0]];
5ad984c9
LM
1277 const int dist_scale_factor = h->dist_scale_factor[ref0];
1278 const int16_t *mv_col = l1mv0[0];
1279 int mv_l0[2];
1280 mv_l0[0] = (dist_scale_factor * mv_col[0] + 128) >> 8;
1281 mv_l0[1] = (dist_scale_factor * mv_col[1] + 128) >> 8;
1282 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref0, 1);
1283 fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv_l0[0],mv_l0[1]), 4);
1284 fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]), 4);
1285 }
1286 }else{
1287 for(i8=0; i8<4; i8++){
1288 const int x8 = i8&1;
1289 const int y8 = i8>>1;
1290 int ref0, dist_scale_factor;
1291
1292 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1293 continue;
1294 h->sub_mb_type[i8] = sub_mb_type;
1295 if(IS_INTRA(mb_type_col)){
1296 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1297 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1298 fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1299 fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1300 continue;
1301 }
1302
1303 ref0 = l1ref0[x8 + y8*h->b8_stride];
2f944356
LM
1304 if(ref0 >= 0)
1305 ref0 = h->map_col_to_list0[0][ref0];
1306 else
1307 ref0 = h->map_col_to_list0[1][l1ref1[x8 + y8*h->b8_stride]];
5ad984c9
LM
1308 dist_scale_factor = h->dist_scale_factor[ref0];
1309
1310 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1311 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1312 for(i4=0; i4<4; i4++){
1313 const int16_t *mv_col = l1mv0[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1314 int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]];
1315 mv_l0[0] = (dist_scale_factor * mv_col[0] + 128) >> 8;
1316 mv_l0[1] = (dist_scale_factor * mv_col[1] + 128) >> 8;
1317 *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] =
1318 pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
1319 }
1320 }
1321 }
1322 }
1323}
1324
0da71265
MN
1325static inline void write_back_motion(H264Context *h, int mb_type){
1326 MpegEncContext * const s = &h->s;
0da71265
MN
1327 const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1328 const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1329 int list;
1330
1331 for(list=0; list<2; list++){
1332 int y;
b75e10d5 1333 if(!USES_LIST(mb_type, list)){
0da71265
MN
1334 if(1){ //FIXME skip or never read if mb_type doesnt use it
1335 for(y=0; y<4; y++){
1336 *(uint64_t*)s->current_picture.motion_val[list][b_xy + 0 + y*h->b_stride]=
1337 *(uint64_t*)s->current_picture.motion_val[list][b_xy + 2 + y*h->b_stride]= 0;
1338 }
9e528114
LA
1339 if( h->pps.cabac ) {
1340 /* FIXME needed ? */
1341 for(y=0; y<4; y++){
1342 *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]=
1343 *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= 0;
1344 }
1345 }
0da71265 1346 for(y=0; y<2; y++){
7d7f635d 1347 *(uint16_t*)&s->current_picture.ref_index[list][b8_xy + y*h->b8_stride]= (LIST_NOT_USED&0xFF)*0x0101;
0da71265
MN
1348 }
1349 }
5ad984c9 1350 continue;
0da71265
MN
1351 }
1352
1353 for(y=0; y<4; y++){
1354 *(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];
1355 *(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];
1356 }
9e528114
LA
1357 if( h->pps.cabac ) {
1358 for(y=0; y<4; y++){
1359 *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+0 + 8*y];
1360 *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+2 + 8*y];
1361 }
1362 }
0da71265
MN
1363 for(y=0; y<2; y++){
1364 s->current_picture.ref_index[list][b8_xy + 0 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+0 + 16*y];
1365 s->current_picture.ref_index[list][b8_xy + 1 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+2 + 16*y];
1366 }
1367 }
5ad984c9
LM
1368
1369 if(h->slice_type == B_TYPE && h->pps.cabac){
1370 if(IS_8X8(mb_type)){
1371 h->direct_table[b8_xy+1+0*h->b8_stride] = IS_DIRECT(h->sub_mb_type[1]) ? 1 : 0;
1372 h->direct_table[b8_xy+0+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[2]) ? 1 : 0;
1373 h->direct_table[b8_xy+1+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[3]) ? 1 : 0;
1374 }
1375 }
0da71265
MN
1376}
1377
1378/**
1379 * Decodes a network abstraction layer unit.
1380 * @param consumed is the number of bytes used as input
1381 * @param length is the length of the array
1382 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp ttailing?
1383 * @returns decoded bytes, might be src+1 if no escapes
1384 */
1385static uint8_t *decode_nal(H264Context *h, uint8_t *src, int *dst_length, int *consumed, int length){
1386 int i, si, di;
1387 uint8_t *dst;
1388
1389// src[0]&0x80; //forbidden bit
1390 h->nal_ref_idc= src[0]>>5;
1391 h->nal_unit_type= src[0]&0x1F;
1392
1393 src++; length--;
1394#if 0
1395 for(i=0; i<length; i++)
1396 printf("%2X ", src[i]);
1397#endif
1398 for(i=0; i+1<length; i+=2){
1399 if(src[i]) continue;
1400 if(i>0 && src[i-1]==0) i--;
1401 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1402 if(src[i+2]!=3){
1403 /* startcode, so we must be past the end */
1404 length=i;
1405 }
1406 break;
1407 }
1408 }
1409
1410 if(i>=length-1){ //no escaped 0
1411 *dst_length= length;
1412 *consumed= length+1; //+1 for the header
1413 return src;
1414 }
1415
1416 h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length);
1417 dst= h->rbsp_buffer;
1418
1419//printf("deoding esc\n");
1420 si=di=0;
1421 while(si<length){
1422 //remove escapes (very rare 1:2^22)
1423 if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1424 if(src[si+2]==3){ //escape
1425 dst[di++]= 0;
1426 dst[di++]= 0;
1427 si+=3;
c8470cc1 1428 continue;
0da71265
MN
1429 }else //next start code
1430 break;
1431 }
1432
1433 dst[di++]= src[si++];
1434 }
1435
1436 *dst_length= di;
1437 *consumed= si + 1;//+1 for the header
1438//FIXME store exact number of bits in the getbitcontext (its needed for decoding)
1439 return dst;
1440}
1441
e5017ab8 1442#if 0
0da71265
MN
1443/**
1444 * @param src the data which should be escaped
1445 * @param dst the target buffer, dst+1 == src is allowed as a special case
1446 * @param length the length of the src data
1447 * @param dst_length the length of the dst array
1448 * @returns length of escaped data in bytes or -1 if an error occured
1449 */
1450static int encode_nal(H264Context *h, uint8_t *dst, uint8_t *src, int length, int dst_length){
1451 int i, escape_count, si, di;
1452 uint8_t *temp;
1453
1454 assert(length>=0);
1455 assert(dst_length>0);
1456
1457 dst[0]= (h->nal_ref_idc<<5) + h->nal_unit_type;
1458
1459 if(length==0) return 1;
1460
1461 escape_count= 0;
1462 for(i=0; i<length; i+=2){
1463 if(src[i]) continue;
1464 if(i>0 && src[i-1]==0)
1465 i--;
1466 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1467 escape_count++;
1468 i+=2;
1469 }
1470 }
1471
1472 if(escape_count==0){
1473 if(dst+1 != src)
1474 memcpy(dst+1, src, length);
1475 return length + 1;
1476 }
1477
1478 if(length + escape_count + 1> dst_length)
1479 return -1;
1480
1481 //this should be damn rare (hopefully)
1482
1483 h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length + escape_count);
1484 temp= h->rbsp_buffer;
1485//printf("encoding esc\n");
1486
1487 si= 0;
1488 di= 0;
1489 while(si < length){
1490 if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1491 temp[di++]= 0; si++;
1492 temp[di++]= 0; si++;
1493 temp[di++]= 3;
1494 temp[di++]= src[si++];
1495 }
1496 else
1497 temp[di++]= src[si++];
1498 }
1499 memcpy(dst+1, temp, length+escape_count);
1500
1501 assert(di == length+escape_count);
1502
1503 return di + 1;
1504}
1505
1506/**
1507 * write 1,10,100,1000,... for alignment, yes its exactly inverse to mpeg4
1508 */
1509static void encode_rbsp_trailing(PutBitContext *pb){
1510 int length;
1511 put_bits(pb, 1, 1);
fe455f33 1512 length= (-put_bits_count(pb))&7;
0da71265
MN
1513 if(length) put_bits(pb, length, 0);
1514}
e5017ab8 1515#endif
0da71265
MN
1516
1517/**
1518 * identifies the exact end of the bitstream
1519 * @return the length of the trailing, or 0 if damaged
1520 */
1521static int decode_rbsp_trailing(uint8_t *src){
1522 int v= *src;
1523 int r;
1524
95c26348 1525 tprintf("rbsp trailing %X\n", v);
0da71265
MN
1526
1527 for(r=1; r<9; r++){
1528 if(v&1) return r;
1529 v>>=1;
1530 }
1531 return 0;
1532}
1533
1534/**
1535 * idct tranforms the 16 dc values and dequantize them.
1536 * @param qp quantization parameter
1537 */
1538static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp){
1539 const int qmul= dequant_coeff[qp][0];
1540#define stride 16
1541 int i;
1542 int temp[16]; //FIXME check if this is a good idea
1543 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
1544 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1545
1546//memset(block, 64, 2*256);
1547//return;
1548 for(i=0; i<4; i++){
1549 const int offset= y_offset[i];
1550 const int z0= block[offset+stride*0] + block[offset+stride*4];
1551 const int z1= block[offset+stride*0] - block[offset+stride*4];
1552 const int z2= block[offset+stride*1] - block[offset+stride*5];
1553 const int z3= block[offset+stride*1] + block[offset+stride*5];
1554
1555 temp[4*i+0]= z0+z3;
1556 temp[4*i+1]= z1+z2;
1557 temp[4*i+2]= z1-z2;
1558 temp[4*i+3]= z0-z3;
1559 }
1560
1561 for(i=0; i<4; i++){
1562 const int offset= x_offset[i];
1563 const int z0= temp[4*0+i] + temp[4*2+i];
1564 const int z1= temp[4*0+i] - temp[4*2+i];
1565 const int z2= temp[4*1+i] - temp[4*3+i];
1566 const int z3= temp[4*1+i] + temp[4*3+i];
1567
1568 block[stride*0 +offset]= ((z0 + z3)*qmul + 2)>>2; //FIXME think about merging this into decode_resdual
1569 block[stride*2 +offset]= ((z1 + z2)*qmul + 2)>>2;
1570 block[stride*8 +offset]= ((z1 - z2)*qmul + 2)>>2;
1571 block[stride*10+offset]= ((z0 - z3)*qmul + 2)>>2;
1572 }
1573}
1574
e5017ab8 1575#if 0
0da71265
MN
1576/**
1577 * dct tranforms the 16 dc values.
1578 * @param qp quantization parameter ??? FIXME
1579 */
1580static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
1581// const int qmul= dequant_coeff[qp][0];
1582 int i;
1583 int temp[16]; //FIXME check if this is a good idea
1584 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
1585 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1586
1587 for(i=0; i<4; i++){
1588 const int offset= y_offset[i];
1589 const int z0= block[offset+stride*0] + block[offset+stride*4];
1590 const int z1= block[offset+stride*0] - block[offset+stride*4];
1591 const int z2= block[offset+stride*1] - block[offset+stride*5];
1592 const int z3= block[offset+stride*1] + block[offset+stride*5];
1593
1594 temp[4*i+0]= z0+z3;
1595 temp[4*i+1]= z1+z2;
1596 temp[4*i+2]= z1-z2;
1597 temp[4*i+3]= z0-z3;
1598 }
1599
1600 for(i=0; i<4; i++){
1601 const int offset= x_offset[i];
1602 const int z0= temp[4*0+i] + temp[4*2+i];
1603 const int z1= temp[4*0+i] - temp[4*2+i];
1604 const int z2= temp[4*1+i] - temp[4*3+i];
1605 const int z3= temp[4*1+i] + temp[4*3+i];
1606
1607 block[stride*0 +offset]= (z0 + z3)>>1;
1608 block[stride*2 +offset]= (z1 + z2)>>1;
1609 block[stride*8 +offset]= (z1 - z2)>>1;
1610 block[stride*10+offset]= (z0 - z3)>>1;
1611 }
1612}
e5017ab8
LA
1613#endif
1614
0da71265
MN
1615#undef xStride
1616#undef stride
1617
1618static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp){
1619 const int qmul= dequant_coeff[qp][0];
1620 const int stride= 16*2;
1621 const int xStride= 16;
1622 int a,b,c,d,e;
1623
1624 a= block[stride*0 + xStride*0];
1625 b= block[stride*0 + xStride*1];
1626 c= block[stride*1 + xStride*0];
1627 d= block[stride*1 + xStride*1];
1628
1629 e= a-b;
1630 a= a+b;
1631 b= c-d;
1632 c= c+d;
1633
1634 block[stride*0 + xStride*0]= ((a+c)*qmul + 0)>>1;
1635 block[stride*0 + xStride*1]= ((e+b)*qmul + 0)>>1;
1636 block[stride*1 + xStride*0]= ((a-c)*qmul + 0)>>1;
1637 block[stride*1 + xStride*1]= ((e-b)*qmul + 0)>>1;
1638}
1639
e5017ab8 1640#if 0
0da71265
MN
1641static void chroma_dc_dct_c(DCTELEM *block){
1642 const int stride= 16*2;
1643 const int xStride= 16;
1644 int a,b,c,d,e;
1645
1646 a= block[stride*0 + xStride*0];
1647 b= block[stride*0 + xStride*1];
1648 c= block[stride*1 + xStride*0];
1649 d= block[stride*1 + xStride*1];
1650
1651 e= a-b;
1652 a= a+b;
1653 b= c-d;
1654 c= c+d;
1655
1656 block[stride*0 + xStride*0]= (a+c);
1657 block[stride*0 + xStride*1]= (e+b);
1658 block[stride*1 + xStride*0]= (a-c);
1659 block[stride*1 + xStride*1]= (e-b);
1660}
e5017ab8 1661#endif
0da71265
MN
1662
1663/**
1664 * gets the chroma qp.
1665 */
1666static inline int get_chroma_qp(H264Context *h, int qscale){
1667
1668 return chroma_qp[clip(qscale + h->pps.chroma_qp_index_offset, 0, 51)];
1669}
1670
1671
e5017ab8 1672#if 0
0da71265
MN
1673static void h264_diff_dct_c(DCTELEM *block, uint8_t *src1, uint8_t *src2, int stride){
1674 int i;
1675 //FIXME try int temp instead of block
1676
1677 for(i=0; i<4; i++){
1678 const int d0= src1[0 + i*stride] - src2[0 + i*stride];
1679 const int d1= src1[1 + i*stride] - src2[1 + i*stride];
1680 const int d2= src1[2 + i*stride] - src2[2 + i*stride];
1681 const int d3= src1[3 + i*stride] - src2[3 + i*stride];
1682 const int z0= d0 + d3;
1683 const int z3= d0 - d3;
1684 const int z1= d1 + d2;
1685 const int z2= d1 - d2;
1686
1687 block[0 + 4*i]= z0 + z1;
1688 block[1 + 4*i]= 2*z3 + z2;
1689 block[2 + 4*i]= z0 - z1;
1690 block[3 + 4*i]= z3 - 2*z2;
1691 }
1692
1693 for(i=0; i<4; i++){
1694 const int z0= block[0*4 + i] + block[3*4 + i];
1695 const int z3= block[0*4 + i] - block[3*4 + i];
1696 const int z1= block[1*4 + i] + block[2*4 + i];
1697 const int z2= block[1*4 + i] - block[2*4 + i];
1698
1699 block[0*4 + i]= z0 + z1;
1700 block[1*4 + i]= 2*z3 + z2;
1701 block[2*4 + i]= z0 - z1;
1702 block[3*4 + i]= z3 - 2*z2;
1703 }
1704}
e5017ab8 1705#endif
0da71265
MN
1706
1707//FIXME need to check that this doesnt overflow signed 32 bit for low qp, iam not sure, its very close
1708//FIXME check that gcc inlines this (and optimizes intra & seperate_dc stuff away)
1709static inline int quantize_c(DCTELEM *block, uint8_t *scantable, int qscale, int intra, int seperate_dc){
1710 int i;
1711 const int * const quant_table= quant_coeff[qscale];
1712 const int bias= intra ? (1<<QUANT_SHIFT)/3 : (1<<QUANT_SHIFT)/6;
1713 const unsigned int threshold1= (1<<QUANT_SHIFT) - bias - 1;
1714 const unsigned int threshold2= (threshold1<<1);
1715 int last_non_zero;
1716
1717 if(seperate_dc){
1718 if(qscale<=18){
1719 //avoid overflows
1720 const int dc_bias= intra ? (1<<(QUANT_SHIFT-2))/3 : (1<<(QUANT_SHIFT-2))/6;
1721 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT-2)) - dc_bias - 1;
1722 const unsigned int dc_threshold2= (dc_threshold1<<1);
1723
1724 int level= block[0]*quant_coeff[qscale+18][0];
1725 if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1726 if(level>0){
1727 level= (dc_bias + level)>>(QUANT_SHIFT-2);
1728 block[0]= level;
1729 }else{
1730 level= (dc_bias - level)>>(QUANT_SHIFT-2);
1731 block[0]= -level;
1732 }
1733// last_non_zero = i;
1734 }else{
1735 block[0]=0;
1736 }
1737 }else{
1738 const int dc_bias= intra ? (1<<(QUANT_SHIFT+1))/3 : (1<<(QUANT_SHIFT+1))/6;
1739 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT+1)) - dc_bias - 1;
1740 const unsigned int dc_threshold2= (dc_threshold1<<1);
1741
1742 int level= block[0]*quant_table[0];
1743 if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1744 if(level>0){
1745 level= (dc_bias + level)>>(QUANT_SHIFT+1);
1746 block[0]= level;
1747 }else{
1748 level= (dc_bias - level)>>(QUANT_SHIFT+1);
1749 block[0]= -level;
1750 }
1751// last_non_zero = i;
1752 }else{
1753 block[0]=0;
1754 }
1755 }
1756 last_non_zero= 0;
1757 i=1;
1758 }else{
1759 last_non_zero= -1;
1760 i=0;
1761 }
1762
1763 for(; i<16; i++){
1764 const int j= scantable[i];
1765 int level= block[j]*quant_table[j];
1766
1767// if( bias+level >= (1<<(QMAT_SHIFT - 3))
1768// || bias-level >= (1<<(QMAT_SHIFT - 3))){
1769 if(((unsigned)(level+threshold1))>threshold2){
1770 if(level>0){
1771 level= (bias + level)>>QUANT_SHIFT;
1772 block[j]= level;
1773 }else{
1774 level= (bias - level)>>QUANT_SHIFT;
1775 block[j]= -level;
1776 }
1777 last_non_zero = i;
1778 }else{
1779 block[j]=0;
1780 }
1781 }
1782
1783 return last_non_zero;
1784}
1785
1786static void pred4x4_vertical_c(uint8_t *src, uint8_t *topright, int stride){
1787 const uint32_t a= ((uint32_t*)(src-stride))[0];
1788 ((uint32_t*)(src+0*stride))[0]= a;
1789 ((uint32_t*)(src+1*stride))[0]= a;
1790 ((uint32_t*)(src+2*stride))[0]= a;
1791 ((uint32_t*)(src+3*stride))[0]= a;
1792}
1793
1794static void pred4x4_horizontal_c(uint8_t *src, uint8_t *topright, int stride){
1795 ((uint32_t*)(src+0*stride))[0]= src[-1+0*stride]*0x01010101;
1796 ((uint32_t*)(src+1*stride))[0]= src[-1+1*stride]*0x01010101;
1797 ((uint32_t*)(src+2*stride))[0]= src[-1+2*stride]*0x01010101;
1798 ((uint32_t*)(src+3*stride))[0]= src[-1+3*stride]*0x01010101;
1799}
1800
1801static void pred4x4_dc_c(uint8_t *src, uint8_t *topright, int stride){
1802 const int dc= ( src[-stride] + src[1-stride] + src[2-stride] + src[3-stride]
1803 + src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 4) >>3;
1804
1805 ((uint32_t*)(src+0*stride))[0]=
1806 ((uint32_t*)(src+1*stride))[0]=
1807 ((uint32_t*)(src+2*stride))[0]=
1808 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1809}
1810
1811static void pred4x4_left_dc_c(uint8_t *src, uint8_t *topright, int stride){
1812 const int dc= ( src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 2) >>2;
1813
1814 ((uint32_t*)(src+0*stride))[0]=
1815 ((uint32_t*)(src+1*stride))[0]=
1816 ((uint32_t*)(src+2*stride))[0]=
1817 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1818}
1819
1820static void pred4x4_top_dc_c(uint8_t *src, uint8_t *topright, int stride){
1821 const int dc= ( src[-stride] + src[1-stride] + src[2-stride] + src[3-stride] + 2) >>2;
1822
1823 ((uint32_t*)(src+0*stride))[0]=
1824 ((uint32_t*)(src+1*stride))[0]=
1825 ((uint32_t*)(src+2*stride))[0]=
1826 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1827}
1828
1829static void pred4x4_128_dc_c(uint8_t *src, uint8_t *topright, int stride){
1830 ((uint32_t*)(src+0*stride))[0]=
1831 ((uint32_t*)(src+1*stride))[0]=
1832 ((uint32_t*)(src+2*stride))[0]=
1833 ((uint32_t*)(src+3*stride))[0]= 128U*0x01010101U;
1834}
1835
1836
1837#define LOAD_TOP_RIGHT_EDGE\
1838 const int t4= topright[0];\
1839 const int t5= topright[1];\
1840 const int t6= topright[2];\
1841 const int t7= topright[3];\
1842
1843#define LOAD_LEFT_EDGE\
1844 const int l0= src[-1+0*stride];\
1845 const int l1= src[-1+1*stride];\
1846 const int l2= src[-1+2*stride];\
1847 const int l3= src[-1+3*stride];\
1848
1849#define LOAD_TOP_EDGE\
1850 const int t0= src[ 0-1*stride];\
1851 const int t1= src[ 1-1*stride];\
1852 const int t2= src[ 2-1*stride];\
1853 const int t3= src[ 3-1*stride];\
1854
1855static void pred4x4_down_right_c(uint8_t *src, uint8_t *topright, int stride){
1856 const int lt= src[-1-1*stride];
1857 LOAD_TOP_EDGE
1858 LOAD_LEFT_EDGE
1859
1860 src[0+3*stride]=(l3 + 2*l2 + l1 + 2)>>2;
1861 src[0+2*stride]=
1862 src[1+3*stride]=(l2 + 2*l1 + l0 + 2)>>2;
1863 src[0+1*stride]=
1864 src[1+2*stride]=
1865 src[2+3*stride]=(l1 + 2*l0 + lt + 2)>>2;
1866 src[0+0*stride]=
1867 src[1+1*stride]=
1868 src[2+2*stride]=
1869 src[3+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1870 src[1+0*stride]=
1871 src[2+1*stride]=
1872 src[3+2*stride]=(lt + 2*t0 + t1 + 2)>>2;
1873 src[2+0*stride]=
1874 src[3+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1875 src[3+0*stride]=(t1 + 2*t2 + t3 + 2)>>2;
4cfbf61b 1876}
0da71265
MN
1877
1878static void pred4x4_down_left_c(uint8_t *src, uint8_t *topright, int stride){
1879 LOAD_TOP_EDGE
1880 LOAD_TOP_RIGHT_EDGE
1881// LOAD_LEFT_EDGE
1882
1883 src[0+0*stride]=(t0 + t2 + 2*t1 + 2)>>2;
1884 src[1+0*stride]=
1885 src[0+1*stride]=(t1 + t3 + 2*t2 + 2)>>2;
1886 src[2+0*stride]=
1887 src[1+1*stride]=
1888 src[0+2*stride]=(t2 + t4 + 2*t3 + 2)>>2;
1889 src[3+0*stride]=
1890 src[2+1*stride]=
1891 src[1+2*stride]=
1892 src[0+3*stride]=(t3 + t5 + 2*t4 + 2)>>2;
1893 src[3+1*stride]=
1894 src[2+2*stride]=
1895 src[1+3*stride]=(t4 + t6 + 2*t5 + 2)>>2;
1896 src[3+2*stride]=
1897 src[2+3*stride]=(t5 + t7 + 2*t6 + 2)>>2;
1898 src[3+3*stride]=(t6 + 3*t7 + 2)>>2;
4cfbf61b 1899}
0da71265
MN
1900
1901static void pred4x4_vertical_right_c(uint8_t *src, uint8_t *topright, int stride){
1902 const int lt= src[-1-1*stride];
1903 LOAD_TOP_EDGE
1904 LOAD_LEFT_EDGE
1905 const __attribute__((unused)) int unu= l3;
1906
1907 src[0+0*stride]=
1908 src[1+2*stride]=(lt + t0 + 1)>>1;
1909 src[1+0*stride]=
1910 src[2+2*stride]=(t0 + t1 + 1)>>1;
1911 src[2+0*stride]=
1912 src[3+2*stride]=(t1 + t2 + 1)>>1;
1913 src[3+0*stride]=(t2 + t3 + 1)>>1;
1914 src[0+1*stride]=
1915 src[1+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1916 src[1+1*stride]=
1917 src[2+3*stride]=(lt + 2*t0 + t1 + 2)>>2;
1918 src[2+1*stride]=
1919 src[3+3*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1920 src[3+1*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1921 src[0+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1922 src[0+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
4cfbf61b 1923}
0da71265
MN
1924
1925static void pred4x4_vertical_left_c(uint8_t *src, uint8_t *topright, int stride){
1926 LOAD_TOP_EDGE
1927 LOAD_TOP_RIGHT_EDGE
1928 const __attribute__((unused)) int unu= t7;
1929
1930 src[0+0*stride]=(t0 + t1 + 1)>>1;
1931 src[1+0*stride]=
1932 src[0+2*stride]=(t1 + t2 + 1)>>1;
1933 src[2+0*stride]=
1934 src[1+2*stride]=(t2 + t3 + 1)>>1;
1935 src[3+0*stride]=
1936 src[2+2*stride]=(t3 + t4+ 1)>>1;
1937 src[3+2*stride]=(t4 + t5+ 1)>>1;
1938 src[0+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1939 src[1+1*stride]=
1940 src[0+3*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1941 src[2+1*stride]=
1942 src[1+3*stride]=(t2 + 2*t3 + t4 + 2)>>2;
1943 src[3+1*stride]=
1944 src[2+3*stride]=(t3 + 2*t4 + t5 + 2)>>2;
1945 src[3+3*stride]=(t4 + 2*t5 + t6 + 2)>>2;
4cfbf61b 1946}
0da71265
MN
1947
1948static void pred4x4_horizontal_up_c(uint8_t *src, uint8_t *topright, int stride){
1949 LOAD_LEFT_EDGE
1950
1951 src[0+0*stride]=(l0 + l1 + 1)>>1;
1952 src[1+0*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1953 src[2+0*stride]=
1954 src[0+1*stride]=(l1 + l2 + 1)>>1;
1955 src[3+0*stride]=
1956 src[1+1*stride]=(l1 + 2*l2 + l3 + 2)>>2;
1957 src[2+1*stride]=
1958 src[0+2*stride]=(l2 + l3 + 1)>>1;
1959 src[3+1*stride]=
1960 src[1+2*stride]=(l2 + 2*l3 + l3 + 2)>>2;
1961 src[3+2*stride]=
1962 src[1+3*stride]=
1963 src[0+3*stride]=
1964 src[2+2*stride]=
1965 src[2+3*stride]=
1966 src[3+3*stride]=l3;
4cfbf61b 1967}
0da71265
MN
1968
1969static void pred4x4_horizontal_down_c(uint8_t *src, uint8_t *topright, int stride){
1970 const int lt= src[-1-1*stride];
1971 LOAD_TOP_EDGE
1972 LOAD_LEFT_EDGE
1973 const __attribute__((unused)) int unu= t3;
1974
1975 src[0+0*stride]=
1976 src[2+1*stride]=(lt + l0 + 1)>>1;
1977 src[1+0*stride]=
1978 src[3+1*stride]=(l0 + 2*lt + t0 + 2)>>2;
1979 src[2+0*stride]=(lt + 2*t0 + t1 + 2)>>2;
1980 src[3+0*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1981 src[0+1*stride]=
1982 src[2+2*stride]=(l0 + l1 + 1)>>1;
1983 src[1+1*stride]=
1984 src[3+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1985 src[0+2*stride]=
1986 src[2+3*stride]=(l1 + l2+ 1)>>1;
1987 src[1+2*stride]=
1988 src[3+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1989 src[0+3*stride]=(l2 + l3 + 1)>>1;
1990 src[1+3*stride]=(l1 + 2*l2 + l3 + 2)>>2;
4cfbf61b 1991}
0da71265
MN
1992
1993static void pred16x16_vertical_c(uint8_t *src, int stride){
1994 int i;
1995 const uint32_t a= ((uint32_t*)(src-stride))[0];
1996 const uint32_t b= ((uint32_t*)(src-stride))[1];
1997 const uint32_t c= ((uint32_t*)(src-stride))[2];
1998 const uint32_t d= ((uint32_t*)(src-stride))[3];
1999
2000 for(i=0; i<16; i++){
2001 ((uint32_t*)(src+i*stride))[0]= a;
2002 ((uint32_t*)(src+i*stride))[1]= b;
2003 ((uint32_t*)(src+i*stride))[2]= c;
2004 ((uint32_t*)(src+i*stride))[3]= d;
2005 }
2006}
2007
2008static void pred16x16_horizontal_c(uint8_t *src, int stride){
2009 int i;
2010
2011 for(i=0; i<16; i++){
2012 ((uint32_t*)(src+i*stride))[0]=
2013 ((uint32_t*)(src+i*stride))[1]=
2014 ((uint32_t*)(src+i*stride))[2]=
2015 ((uint32_t*)(src+i*stride))[3]= src[-1+i*stride]*0x01010101;
2016 }
2017}
2018
2019static void pred16x16_dc_c(uint8_t *src, int stride){
2020 int i, dc=0;
2021
2022 for(i=0;i<16; i++){
2023 dc+= src[-1+i*stride];
2024 }
2025
2026 for(i=0;i<16; i++){
2027 dc+= src[i-stride];
2028 }
2029
2030 dc= 0x01010101*((dc + 16)>>5);
2031
2032 for(i=0; i<16; i++){
2033 ((uint32_t*)(src+i*stride))[0]=
2034 ((uint32_t*)(src+i*stride))[1]=
2035 ((uint32_t*)(src+i*stride))[2]=
2036 ((uint32_t*)(src+i*stride))[3]= dc;
2037 }
2038}
2039
2040static void pred16x16_left_dc_c(uint8_t *src, int stride){
2041 int i, dc=0;
2042
2043 for(i=0;i<16; i++){
2044 dc+= src[-1+i*stride];
2045 }
2046
2047 dc= 0x01010101*((dc + 8)>>4);
2048
2049 for(i=0; i<16; i++){
2050 ((uint32_t*)(src+i*stride))[0]=
2051 ((uint32_t*)(src+i*stride))[1]=
2052 ((uint32_t*)(src+i*stride))[2]=
2053 ((uint32_t*)(src+i*stride))[3]= dc;
2054 }
2055}
2056
2057static void pred16x16_top_dc_c(uint8_t *src, int stride){
2058 int i, dc=0;
2059
2060 for(i=0;i<16; i++){
2061 dc+= src[i-stride];
2062 }
2063 dc= 0x01010101*((dc + 8)>>4);
2064
2065 for(i=0; i<16; i++){
2066 ((uint32_t*)(src+i*stride))[0]=
2067 ((uint32_t*)(src+i*stride))[1]=
2068 ((uint32_t*)(src+i*stride))[2]=
2069 ((uint32_t*)(src+i*stride))[3]= dc;
2070 }
2071}
2072
2073static void pred16x16_128_dc_c(uint8_t *src, int stride){
2074 int i;
2075
2076 for(i=0; i<16; i++){
2077 ((uint32_t*)(src+i*stride))[0]=
2078 ((uint32_t*)(src+i*stride))[1]=
2079 ((uint32_t*)(src+i*stride))[2]=
2080 ((uint32_t*)(src+i*stride))[3]= 0x01010101U*128U;
2081 }
2082}
2083
8b82a956 2084static inline void pred16x16_plane_compat_c(uint8_t *src, int stride, const int svq3){
30f73fc7
MN
2085 int i, j, k;
2086 int a;
2087 uint8_t *cm = cropTbl + MAX_NEG_CROP;
2088 const uint8_t * const src0 = src+7-stride;
2089 const uint8_t *src1 = src+8*stride-1;
2090 const uint8_t *src2 = src1-2*stride; // == src+6*stride-1;
2091 int H = src0[1] - src0[-1];
2092 int V = src1[0] - src2[ 0];
2093 for(k=2; k<=8; ++k) {
2094 src1 += stride; src2 -= stride;
2095 H += k*(src0[k] - src0[-k]);
2096 V += k*(src1[0] - src2[ 0]);
2097 }
8b82a956
MN
2098 if(svq3){
2099 H = ( 5*(H/4) ) / 16;
2100 V = ( 5*(V/4) ) / 16;
2e26c8d2
MM
2101
2102 /* required for 100% accuracy */
2103 i = H; H = V; V = i;
8b82a956
MN
2104 }else{
2105 H = ( 5*H+32 ) >> 6;
2106 V = ( 5*V+32 ) >> 6;
2107 }
30f73fc7
MN
2108
2109 a = 16*(src1[0] + src2[16] + 1) - 7*(V+H);
2110 for(j=16; j>0; --j) {
2111 int b = a;
2112 a += V;
2113 for(i=-16; i<0; i+=4) {
2114 src[16+i] = cm[ (b ) >> 5 ];
2115 src[17+i] = cm[ (b+ H) >> 5 ];
2116 src[18+i] = cm[ (b+2*H) >> 5 ];
2117 src[19+i] = cm[ (b+3*H) >> 5 ];
2118 b += 4*H;
0da71265 2119 }
30f73fc7
MN
2120 src += stride;
2121 }
0da71265
MN
2122}
2123
8b82a956
MN
2124static void pred16x16_plane_c(uint8_t *src, int stride){
2125 pred16x16_plane_compat_c(src, stride, 0);
2126}
2127
0da71265
MN
2128static void pred8x8_vertical_c(uint8_t *src, int stride){
2129 int i;
2130 const uint32_t a= ((uint32_t*)(src-stride))[0];
2131 const uint32_t b= ((uint32_t*)(src-stride))[1];
2132
2133 for(i=0; i<8; i++){
2134 ((uint32_t*)(src+i*stride))[0]= a;
2135 ((uint32_t*)(src+i*stride))[1]= b;
2136 }
2137}
2138
2139static void pred8x8_horizontal_c(uint8_t *src, int stride){
2140 int i;
2141
2142 for(i=0; i<8; i++){
2143 ((uint32_t*)(src+i*stride))[0]=
2144 ((uint32_t*)(src+i*stride))[1]= src[-1+i*stride]*0x01010101;
2145 }
2146}
2147
2148static void pred8x8_128_dc_c(uint8_t *src, int stride){
2149 int i;
2150
2151 for(i=0; i<4; i++){
2152 ((uint32_t*)(src+i*stride))[0]=
2153 ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
2154 }
2155 for(i=4; i<8; i++){
2156 ((uint32_t*)(src+i*stride))[0]=
2157 ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
2158 }
2159}
2160
2161static void pred8x8_left_dc_c(uint8_t *src, int stride){
2162 int i;
2163 int dc0, dc2;
2164
2165 dc0=dc2=0;
2166 for(i=0;i<4; i++){
2167 dc0+= src[-1+i*stride];
2168 dc2+= src[-1+(i+4)*stride];
2169 }
2170 dc0= 0x01010101*((dc0 + 2)>>2);
2171 dc2= 0x01010101*((dc2 + 2)>>2);
2172
2173 for(i=0; i<4; i++){
2174 ((uint32_t*)(src+i*stride))[0]=
2175 ((uint32_t*)(src+i*stride))[1]= dc0;
2176 }
2177 for(i=4; i<8; i++){
2178 ((uint32_t*)(src+i*stride))[0]=
2179 ((uint32_t*)(src+i*stride))[1]= dc2;
2180 }
2181}
2182
2183static void pred8x8_top_dc_c(uint8_t *src, int stride){
2184 int i;
2185 int dc0, dc1;
2186
2187 dc0=dc1=0;
2188 for(i=0;i<4; i++){
2189 dc0+= src[i-stride];
2190 dc1+= src[4+i-stride];
2191 }
2192 dc0= 0x01010101*((dc0 + 2)>>2);
2193 dc1= 0x01010101*((dc1 + 2)>>2);
2194
2195 for(i=0; i<4; i++){
2196 ((uint32_t*)(src+i*stride))[0]= dc0;
2197 ((uint32_t*)(src+i*stride))[1]= dc1;
2198 }
2199 for(i=4; i<8; i++){
2200 ((uint32_t*)(src+i*stride))[0]= dc0;
2201 ((uint32_t*)(src+i*stride))[1]= dc1;
2202 }
2203}
2204
2205
2206static void pred8x8_dc_c(uint8_t *src, int stride){
2207 int i;
2208 int dc0, dc1, dc2, dc3;
2209
2210 dc0=dc1=dc2=0;
2211 for(i=0;i<4; i++){
2212 dc0+= src[-1+i*stride] + src[i-stride];
2213 dc1+= src[4+i-stride];
2214 dc2+= src[-1+(i+4)*stride];
2215 }
2216 dc3= 0x01010101*((dc1 + dc2 + 4)>>3);
2217 dc0= 0x01010101*((dc0 + 4)>>3);
2218 dc1= 0x01010101*((dc1 + 2)>>2);
2219 dc2= 0x01010101*((dc2 + 2)>>2);
2220
2221 for(i=0; i<4; i++){
2222 ((uint32_t*)(src+i*stride))[0]= dc0;
2223 ((uint32_t*)(src+i*stride))[1]= dc1;
2224 }
2225 for(i=4; i<8; i++){
2226 ((uint32_t*)(src+i*stride))[0]= dc2;
2227 ((uint32_t*)(src+i*stride))[1]= dc3;
2228 }
2229}
2230
2231static void pred8x8_plane_c(uint8_t *src, int stride){
30f73fc7
MN
2232 int j, k;
2233 int a;
2234 uint8_t *cm = cropTbl + MAX_NEG_CROP;
2235 const uint8_t * const src0 = src+3-stride;
2236 const uint8_t *src1 = src+4*stride-1;
2237 const uint8_t *src2 = src1-2*stride; // == src+2*stride-1;
2238 int H = src0[1] - src0[-1];
2239 int V = src1[0] - src2[ 0];
2240 for(k=2; k<=4; ++k) {
2241 src1 += stride; src2 -= stride;
2242 H += k*(src0[k] - src0[-k]);
2243 V += k*(src1[0] - src2[ 0]);
2244 }
2245 H = ( 17*H+16 ) >> 5;
2246 V = ( 17*V+16 ) >> 5;
2247
2248 a = 16*(src1[0] + src2[8]+1) - 3*(V+H);
2249 for(j=8; j>0; --j) {
2250 int b = a;
2251 a += V;
2252 src[0] = cm[ (b ) >> 5 ];
2253 src[1] = cm[ (b+ H) >> 5 ];
2254 src[2] = cm[ (b+2*H) >> 5 ];
2255 src[3] = cm[ (b+3*H) >> 5 ];
2256 src[4] = cm[ (b+4*H) >> 5 ];
2257 src[5] = cm[ (b+5*H) >> 5 ];
2258 src[6] = cm[ (b+6*H) >> 5 ];
2259 src[7] = cm[ (b+7*H) >> 5 ];
2260 src += stride;
2261 }
0da71265
MN
2262}
2263
2264static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
2265 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2266 int src_x_offset, int src_y_offset,
2267 qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
2268 MpegEncContext * const s = &h->s;
2269 const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
2270 const int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
2271 const int luma_xy= (mx&3) + ((my&3)<<2);
2272 uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*s->linesize;
2273 uint8_t * src_cb= pic->data[1] + (mx>>3) + (my>>3)*s->uvlinesize;
2274 uint8_t * src_cr= pic->data[2] + (mx>>3) + (my>>3)*s->uvlinesize;
2275 int extra_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16; //FIXME increase edge?, IMHO not worth it
2276 int extra_height= extra_width;
2277 int emu=0;
2278 const int full_mx= mx>>2;
2279 const int full_my= my>>2;
2280
2281 assert(pic->data[0]);
2282
2283 if(mx&7) extra_width -= 3;
2284 if(my&7) extra_height -= 3;
2285
2286 if( full_mx < 0-extra_width
2287 || full_my < 0-extra_height
2288 || full_mx + 16/*FIXME*/ > s->width + extra_width
2289 || full_my + 16/*FIXME*/ > s->height + extra_height){
c009df3f 2290 ff_emulated_edge_mc(s->edge_emu_buffer, src_y - 2 - 2*s->linesize, s->linesize, 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, s->width, s->height);
0da71265
MN
2291 src_y= s->edge_emu_buffer + 2 + 2*s->linesize;
2292 emu=1;
2293 }
2294
2295 qpix_op[luma_xy](dest_y, src_y, s->linesize); //FIXME try variable height perhaps?
2296 if(!square){
2297 qpix_op[luma_xy](dest_y + delta, src_y + delta, s->linesize);
2298 }
2299
2300 if(s->flags&CODEC_FLAG_GRAY) return;
2301
2302 if(emu){
c009df3f 2303 ff_emulated_edge_mc(s->edge_emu_buffer, src_cb, s->uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), s->width>>1, s->height>>1);
0da71265
MN
2304 src_cb= s->edge_emu_buffer;
2305 }
2306 chroma_op(dest_cb, src_cb, s->uvlinesize, chroma_height, mx&7, my&7);
2307
2308 if(emu){
c009df3f 2309 ff_emulated_edge_mc(s->edge_emu_buffer, src_cr, s->uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), s->width>>1, s->height>>1);
0da71265
MN
2310 src_cr= s->edge_emu_buffer;
2311 }
2312 chroma_op(dest_cr, src_cr, s->uvlinesize, chroma_height, mx&7, my&7);
2313}
2314
9f2d1b4f 2315static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
0da71265
MN
2316 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2317 int x_offset, int y_offset,
2318 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2319 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
2320 int list0, int list1){
2321 MpegEncContext * const s = &h->s;
2322 qpel_mc_func *qpix_op= qpix_put;
2323 h264_chroma_mc_func chroma_op= chroma_put;
2324
2325 dest_y += 2*x_offset + 2*y_offset*s-> linesize;
2326 dest_cb += x_offset + y_offset*s->uvlinesize;
2327 dest_cr += x_offset + y_offset*s->uvlinesize;
2328 x_offset += 8*s->mb_x;
2329 y_offset += 8*s->mb_y;
2330
2331 if(list0){
1924f3ce 2332 Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
0da71265
MN
2333 mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
2334 dest_y, dest_cb, dest_cr, x_offset, y_offset,
2335 qpix_op, chroma_op);
2336
2337 qpix_op= qpix_avg;
2338 chroma_op= chroma_avg;
2339 }
2340
2341 if(list1){
1924f3ce 2342 Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
0da71265
MN
2343 mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
2344 dest_y, dest_cb, dest_cr, x_offset, y_offset,
2345 qpix_op, chroma_op);
2346 }
2347}
2348
9f2d1b4f
LM
2349static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
2350 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2351 int x_offset, int y_offset,
2352 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2353 h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
2354 h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
2355 int list0, int list1){
2356 MpegEncContext * const s = &h->s;
2357
2358 dest_y += 2*x_offset + 2*y_offset*s-> linesize;
2359 dest_cb += x_offset + y_offset*s->uvlinesize;
2360 dest_cr += x_offset + y_offset*s->uvlinesize;
2361 x_offset += 8*s->mb_x;
2362 y_offset += 8*s->mb_y;
2363
2364 if(list0 && list1){
2365 /* don't optimize for luma-only case, since B-frames usually
2366 * use implicit weights => chroma too. */
2367 uint8_t *tmp_cb = s->obmc_scratchpad;
2368 uint8_t *tmp_cr = tmp_cb + 8*s->uvlinesize;
2369 uint8_t *tmp_y = tmp_cr + 8*s->uvlinesize;
2370 int refn0 = h->ref_cache[0][ scan8[n] ];
2371 int refn1 = h->ref_cache[1][ scan8[n] ];
2372
2373 mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
2374 dest_y, dest_cb, dest_cr,
2375 x_offset, y_offset, qpix_put, chroma_put);
2376 mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
2377 tmp_y, tmp_cb, tmp_cr,
2378 x_offset, y_offset, qpix_put, chroma_put);
2379
2380 if(h->use_weight == 2){
2381 int weight0 = h->implicit_weight[refn0][refn1];
2382 int weight1 = 64 - weight0;
2383 luma_weight_avg( dest_y, tmp_y, s-> linesize, 5, weight0, weight1, 0, 0);
2384 chroma_weight_avg(dest_cb, tmp_cb, s->uvlinesize, 5, weight0, weight1, 0, 0);
2385 chroma_weight_avg(dest_cr, tmp_cr, s->uvlinesize, 5, weight0, weight1, 0, 0);
2386 }else{
2387 luma_weight_avg(dest_y, tmp_y, s->linesize, h->luma_log2_weight_denom,
2388 h->luma_weight[0][refn0], h->luma_weight[1][refn1],
2389 h->luma_offset[0][refn0], h->luma_offset[1][refn1]);
2390 chroma_weight_avg(dest_cb, tmp_cb, s->uvlinesize, h->chroma_log2_weight_denom,
2391 h->chroma_weight[0][refn0][0], h->chroma_weight[1][refn1][0],
2392 h->chroma_offset[0][refn0][0], h->chroma_offset[1][refn1][0]);
2393 chroma_weight_avg(dest_cr, tmp_cr, s->uvlinesize, h->chroma_log2_weight_denom,
2394 h->chroma_weight[0][refn0][1], h->chroma_weight[1][refn1][1],
2395 h->chroma_offset[0][refn0][1], h->chroma_offset[1][refn1][1]);
2396 }
2397 }else{
2398 int list = list1 ? 1 : 0;
2399 int refn = h->ref_cache[list][ scan8[n] ];
2400 Picture *ref= &h->ref_list[list][refn];
2401 mc_dir_part(h, ref, n, square, chroma_height, delta, list,
2402 dest_y, dest_cb, dest_cr, x_offset, y_offset,
2403 qpix_put, chroma_put);
2404
2405 luma_weight_op(dest_y, s->linesize, h->luma_log2_weight_denom,
2406 h->luma_weight[list][refn], h->luma_offset[list][refn]);
2407 if(h->use_weight_chroma){
2408 chroma_weight_op(dest_cb, s->uvlinesize, h->chroma_log2_weight_denom,
2409 h->chroma_weight[list][refn][0], h->chroma_offset[list][refn][0]);
2410 chroma_weight_op(dest_cr, s->uvlinesize, h->chroma_log2_weight_denom,
2411 h->chroma_weight[list][refn][1], h->chroma_offset[list][refn][1]);
2412 }
2413 }
2414}
2415
2416static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
2417 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2418 int x_offset, int y_offset,
2419 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2420 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
2421 h264_weight_func *weight_op, h264_biweight_func *weight_avg,
2422 int list0, int list1){
2423 if((h->use_weight==2 && list0 && list1
2424 && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ] != 32))
2425 || h->use_weight==1)
2426 mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
2427 x_offset, y_offset, qpix_put, chroma_put,
2428 weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
2429 else
2430 mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
2431 x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
2432}
2433
0da71265
MN
2434static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2435 qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
9f2d1b4f
LM
2436 qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
2437 h264_weight_func *weight_op, h264_biweight_func *weight_avg){
0da71265 2438 MpegEncContext * const s = &h->s;
7bc9090a 2439 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
0da71265
MN
2440 const int mb_type= s->current_picture.mb_type[mb_xy];
2441
2442 assert(IS_INTER(mb_type));
2443
2444 if(IS_16X16(mb_type)){
2445 mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
2446 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
9f2d1b4f 2447 &weight_op[0], &weight_avg[0],
0da71265
MN
2448 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2449 }else if(IS_16X8(mb_type)){
2450 mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
2451 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
9f2d1b4f 2452 &weight_op[1], &weight_avg[1],
0da71265
MN
2453 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2454 mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
2455 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
9f2d1b4f 2456 &weight_op[1], &weight_avg[1],
0da71265
MN
2457 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2458 }else if(IS_8X16(mb_type)){
2459 mc_part(h, 0, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 0, 0,
2460 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
9f2d1b4f 2461 &weight_op[2], &weight_avg[2],
0da71265
MN
2462 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2463 mc_part(h, 4, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 4, 0,
2464 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
9f2d1b4f 2465 &weight_op[2], &weight_avg[2],
0da71265
MN
2466 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2467 }else{
2468 int i;
2469
2470 assert(IS_8X8(mb_type));
2471
2472 for(i=0; i<4; i++){
2473 const int sub_mb_type= h->sub_mb_type[i];
2474 const int n= 4*i;
2475 int x_offset= (i&1)<<2;
2476 int y_offset= (i&2)<<1;
2477
2478 if(IS_SUB_8X8(sub_mb_type)){
2479 mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2480 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
9f2d1b4f 2481 &weight_op[3], &weight_avg[3],
0da71265
MN
2482 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2483 }else if(IS_SUB_8X4(sub_mb_type)){
2484 mc_part(h, n , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2485 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
9f2d1b4f 2486 &weight_op[4], &weight_avg[4],
0da71265
MN
2487 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2488 mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
2489 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
9f2d1b4f 2490 &weight_op[4], &weight_avg[4],
0da71265
MN
2491 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2492 }else if(IS_SUB_4X8(sub_mb_type)){
2493 mc_part(h, n , 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2494 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
9f2d1b4f 2495 &weight_op[5], &weight_avg[5],
0da71265
MN
2496 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2497 mc_part(h, n+1, 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
2498 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
9f2d1b4f 2499 &weight_op[5], &weight_avg[5],
0da71265
MN
2500 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2501 }else{
2502 int j;
2503 assert(IS_SUB_4X4(sub_mb_type));
2504 for(j=0; j<4; j++){
2505 int sub_x_offset= x_offset + 2*(j&1);
2506 int sub_y_offset= y_offset + (j&2);
2507 mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
2508 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
9f2d1b4f 2509 &weight_op[6], &weight_avg[6],
0da71265
MN
2510 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2511 }
2512 }
2513 }
2514 }
2515}
2516
2517static void decode_init_vlc(H264Context *h){
2518 static int done = 0;
2519
2520 if (!done) {
2521 int i;
2522 done = 1;
2523
2524 init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5,
2525 &chroma_dc_coeff_token_len [0], 1, 1,
073c2593 2526 &chroma_dc_coeff_token_bits[0], 1, 1, 1);
0da71265
MN
2527
2528 for(i=0; i<4; i++){
2529 init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17,
2530 &coeff_token_len [i][0], 1, 1,
073c2593 2531 &coeff_token_bits[i][0], 1, 1, 1);
0da71265
MN
2532 }
2533
2534 for(i=0; i<3; i++){
2535 init_vlc(&chroma_dc_total_zeros_vlc[i], CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
2536 &chroma_dc_total_zeros_len [i][0], 1, 1,
073c2593 2537 &chroma_dc_total_zeros_bits[i][0], 1, 1, 1);
0da71265
MN
2538 }
2539 for(i=0; i<15; i++){
2540 init_vlc(&total_zeros_vlc[i], TOTAL_ZEROS_VLC_BITS, 16,
2541 &total_zeros_len [i][0], 1, 1,
073c2593 2542 &total_zeros_bits[i][0], 1, 1, 1);
0da71265
MN
2543 }
2544
2545 for(i=0; i<6; i++){
2546 init_vlc(&run_vlc[i], RUN_VLC_BITS, 7,
2547 &run_len [i][0], 1, 1,
073c2593 2548 &run_bits[i][0], 1, 1, 1);
0da71265
MN
2549 }
2550 init_vlc(&run7_vlc, RUN7_VLC_BITS, 16,
2551 &run_len [6][0], 1, 1,
073c2593 2552 &run_bits[6][0], 1, 1, 1);
0da71265
MN
2553 }
2554}
2555
2556/**
2557 * Sets the intra prediction function pointers.
2558 */
2559static void init_pred_ptrs(H264Context *h){
2560// MpegEncContext * const s = &h->s;
2561
2562 h->pred4x4[VERT_PRED ]= pred4x4_vertical_c;
2563 h->pred4x4[HOR_PRED ]= pred4x4_horizontal_c;
2564 h->pred4x4[DC_PRED ]= pred4x4_dc_c;
2565 h->pred4x4[DIAG_DOWN_LEFT_PRED ]= pred4x4_down_left_c;
2566 h->pred4x4[DIAG_DOWN_RIGHT_PRED]= pred4x4_down_right_c;
2567 h->pred4x4[VERT_RIGHT_PRED ]= pred4x4_vertical_right_c;
2568 h->pred4x4[HOR_DOWN_PRED ]= pred4x4_horizontal_down_c;
2569 h->pred4x4[VERT_LEFT_PRED ]= pred4x4_vertical_left_c;
2570 h->pred4x4[HOR_UP_PRED ]= pred4x4_horizontal_up_c;
2571 h->pred4x4[LEFT_DC_PRED ]= pred4x4_left_dc_c;
2572 h->pred4x4[TOP_DC_PRED ]= pred4x4_top_dc_c;
2573 h->pred4x4[DC_128_PRED ]= pred4x4_128_dc_c;
2574
2575 h->pred8x8[DC_PRED8x8 ]= pred8x8_dc_c;
2576 h->pred8x8[VERT_PRED8x8 ]= pred8x8_vertical_c;
2577 h->pred8x8[HOR_PRED8x8 ]= pred8x8_horizontal_c;
2578 h->pred8x8[PLANE_PRED8x8 ]= pred8x8_plane_c;
2579 h->pred8x8[LEFT_DC_PRED8x8]= pred8x8_left_dc_c;
2580 h->pred8x8[TOP_DC_PRED8x8 ]= pred8x8_top_dc_c;
2581 h->pred8x8[DC_128_PRED8x8 ]= pred8x8_128_dc_c;
2582
2583 h->pred16x16[DC_PRED8x8 ]= pred16x16_dc_c;
2584 h->pred16x16[VERT_PRED8x8 ]= pred16x16_vertical_c;
2585 h->pred16x16[HOR_PRED8x8 ]= pred16x16_horizontal_c;
2586 h->pred16x16[PLANE_PRED8x8 ]= pred16x16_plane_c;
2587 h->pred16x16[LEFT_DC_PRED8x8]= pred16x16_left_dc_c;
2588 h->pred16x16[TOP_DC_PRED8x8 ]= pred16x16_top_dc_c;
2589 h->pred16x16[DC_128_PRED8x8 ]= pred16x16_128_dc_c;
2590}
2591
0da71265 2592static void free_tables(H264Context *h){
0da71265 2593 av_freep(&h->intra4x4_pred_mode);
e5017ab8
LA
2594 av_freep(&h->chroma_pred_mode_table);
2595 av_freep(&h->cbp_table);
9e528114
LA
2596 av_freep(&h->mvd_table[0]);
2597 av_freep(&h->mvd_table[1]);
5ad984c9 2598 av_freep(&h->direct_table);
0da71265
MN
2599 av_freep(&h->non_zero_count);
2600 av_freep(&h->slice_table_base);
53c05b1e 2601 av_freep(&h->top_border);
0da71265 2602 h->slice_table= NULL;
e5017ab8 2603
0da71265
MN
2604 av_freep(&h->mb2b_xy);
2605 av_freep(&h->mb2b8_xy);
9f2d1b4f
LM
2606
2607 av_freep(&h->s.obmc_scratchpad);
0da71265
MN
2608}
2609
2610/**
2611 * allocates tables.
2612 * needs widzh/height
2613 */
2614static int alloc_tables(H264Context *h){
2615 MpegEncContext * const s = &h->s;
7bc9090a 2616 const int big_mb_num= s->mb_stride * (s->mb_height+1);
0da71265
MN
2617 int x,y;
2618
2619 CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8 * sizeof(uint8_t))
e5017ab8 2620
53c05b1e 2621 CHECKED_ALLOCZ(h->non_zero_count , big_mb_num * 16 * sizeof(uint8_t))
0da71265 2622 CHECKED_ALLOCZ(h->slice_table_base , big_mb_num * sizeof(uint8_t))
53c05b1e 2623 CHECKED_ALLOCZ(h->top_border , s->mb_width * (16+8+8) * sizeof(uint8_t))
5d0e4cb8 2624 CHECKED_ALLOCZ(h->cbp_table, big_mb_num * sizeof(uint16_t))
0da71265 2625
e5017ab8
LA
2626 if( h->pps.cabac ) {
2627 CHECKED_ALLOCZ(h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t))
9e528114
LA
2628 CHECKED_ALLOCZ(h->mvd_table[0], 32*big_mb_num * sizeof(uint16_t));
2629 CHECKED_ALLOCZ(h->mvd_table[1], 32*big_mb_num * sizeof(uint16_t));
5ad984c9 2630 CHECKED_ALLOCZ(h->direct_table, 32*big_mb_num * sizeof(uint8_t));
e5017ab8
LA
2631 }
2632
0da71265 2633 memset(h->slice_table_base, -1, big_mb_num * sizeof(uint8_t));
7bc9090a 2634 h->slice_table= h->slice_table_base + s->mb_stride + 1;
0da71265
MN
2635
2636 CHECKED_ALLOCZ(h->mb2b_xy , big_mb_num * sizeof(uint16_t));
2637 CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint16_t));
2638 for(y=0; y<s->mb_height; y++){
2639 for(x=0; x<s->mb_width; x++){
7bc9090a 2640 const int mb_xy= x + y*s->mb_stride;
0da71265
MN
2641 const int b_xy = 4*x + 4*y*h->b_stride;
2642 const int b8_xy= 2*x + 2*y*h->b8_stride;
2643
2644 h->mb2b_xy [mb_xy]= b_xy;
2645 h->mb2b8_xy[mb_xy]= b8_xy;
2646 }
2647 }
9f2d1b4f 2648
9c6221ae
GV
2649 s->obmc_scratchpad = NULL;
2650
0da71265
MN
2651 return 0;
2652fail:
2653 free_tables(h);
2654 return -1;
2655}
2656
2657static void common_init(H264Context *h){
2658 MpegEncContext * const s = &h->s;
0da71265
MN
2659
2660 s->width = s->avctx->width;
2661 s->height = s->avctx->height;
2662 s->codec_id= s->avctx->codec->id;
2663
2664 init_pred_ptrs(h);
2665
9a41c2c7 2666 s->unrestricted_mv=1;
0da71265
MN
2667 s->decode=1; //FIXME
2668}
2669
2670static int decode_init(AVCodecContext *avctx){
2671 H264Context *h= avctx->priv_data;
2672 MpegEncContext * const s = &h->s;
2673
3edcacde
MN
2674 MPV_decode_defaults(s);
2675
0da71265
MN
2676 s->avctx = avctx;
2677 common_init(h);
2678
2679 s->out_format = FMT_H264;
2680 s->workaround_bugs= avctx->workaround_bugs;
2681
2682 // set defaults
0da71265
MN
2683// s->decode_mb= ff_h263_decode_mb;
2684 s->low_delay= 1;
2685 avctx->pix_fmt= PIX_FMT_YUV420P;
2686
2687 decode_init_vlc(h);
2688
26165f99
MR
2689 if(avctx->extradata_size > 0 && avctx->extradata &&
2690 *(char *)avctx->extradata == 1){
4770b1b4
RT
2691 h->is_avc = 1;
2692 h->got_avcC = 0;
26165f99
MR
2693 } else {
2694 h->is_avc = 0;
4770b1b4
RT
2695 }
2696
0da71265
MN
2697 return 0;
2698}
2699
2700static void frame_start(H264Context *h){
2701 MpegEncContext * const s = &h->s;
2702 int i;
2703
2704 MPV_frame_start(s, s->avctx);
2705 ff_er_frame_start(s);
0da71265
MN
2706
2707 assert(s->linesize && s->uvlinesize);
2708
2709 for(i=0; i<16; i++){
2710 h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
6867a90b 2711 h->block_offset[24+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
0da71265
MN
2712 }
2713 for(i=0; i<4; i++){
2714 h->block_offset[16+i]=
2715 h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
6867a90b
LLL
2716 h->block_offset[24+16+i]=
2717 h->block_offset[24+20+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
0da71265
MN
2718 }
2719
934b0821
LM
2720 /* can't be in alloc_tables because linesize isn't known there.
2721 * FIXME: redo bipred weight to not require extra buffer? */
2722 if(!s->obmc_scratchpad)
2723 s->obmc_scratchpad = av_malloc(16*s->linesize + 2*8*s->uvlinesize);
2724
0da71265
MN
2725// s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
2726}
2727
53c05b1e
MN
2728static inline void backup_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){
2729 MpegEncContext * const s = &h->s;
2730 int i;
2731
2732 src_y -= linesize;
2733 src_cb -= uvlinesize;
2734 src_cr -= uvlinesize;
2735
6867a90b
LLL
2736 // There is two lines saved, the line above the the top macroblock of a pair,
2737 // and the line above the bottom macroblock
53c05b1e
MN
2738 h->left_border[0]= h->top_border[s->mb_x][15];
2739 for(i=1; i<17; i++){
2740 h->left_border[i]= src_y[15+i* linesize];
2741 }
2742
2743 *(uint64_t*)(h->top_border[s->mb_x]+0)= *(uint64_t*)(src_y + 16*linesize);
2744 *(uint64_t*)(h->top_border[s->mb_x]+8)= *(uint64_t*)(src_y +8+16*linesize);
2745
2746 if(!(s->flags&CODEC_FLAG_GRAY)){
2747 h->left_border[17 ]= h->top_border[s->mb_x][16+7];
2748 h->left_border[17+9]= h->top_border[s->mb_x][24+7];
2749 for(i=1; i<9; i++){
2750 h->left_border[i+17 ]= src_cb[7+i*uvlinesize];
2751 h->left_border[i+17+9]= src_cr[7+i*uvlinesize];
2752 }
2753 *(uint64_t*)(h->top_border[s->mb_x]+16)= *(uint64_t*)(src_cb+8*uvlinesize);
2754 *(uint64_t*)(h->top_border[s->mb_x]+24)= *(uint64_t*)(src_cr+8*uvlinesize);
2755 }
2756}
2757
2758static inline void xchg_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize, int xchg){
2759 MpegEncContext * const s = &h->s;
2760 int temp8, i;
2761 uint64_t temp64;
d89dc06a
LM
2762 int deblock_left = (s->mb_x > 0);
2763 int deblock_top = (s->mb_y > 0);
53c05b1e
MN
2764
2765 src_y -= linesize + 1;
2766 src_cb -= uvlinesize + 1;
2767 src_cr -= uvlinesize + 1;
2768
2769#define XCHG(a,b,t,xchg)\
2770t= a;\
2771if(xchg)\
2772 a= b;\
2773b= t;
d89dc06a
LM
2774
2775 if(deblock_left){
2776 for(i = !deblock_top; i<17; i++){
2777 XCHG(h->left_border[i ], src_y [i* linesize], temp8, xchg);
2778 }
2779 }
2780
2781 if(deblock_top){
2782 XCHG(*(uint64_t*)(h->top_border[s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
2783 XCHG(*(uint64_t*)(h->top_border[s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
53c05b1e 2784 }
53c05b1e
MN
2785
2786 if(!(s->flags&CODEC_FLAG_GRAY)){
d89dc06a
LM
2787 if(deblock_left){
2788 for(i = !deblock_top; i<9; i++){
2789 XCHG(h->left_border[i+17 ], src_cb[i*uvlinesize], temp8, xchg);
2790 XCHG(h->left_border[i+17+9], src_cr[i*uvlinesize], temp8, xchg);
2791 }
2792 }
2793 if(deblock_top){
2794 XCHG(*(uint64_t*)(h->top_border[s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
2795 XCHG(*(uint64_t*)(h->top_border[s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
53c05b1e 2796 }
53c05b1e
MN
2797 }
2798}
2799
0da71265
MN
2800static void hl_decode_mb(H264Context *h){
2801 MpegEncContext * const s = &h->s;
2802 const int mb_x= s->mb_x;
2803 const int mb_y= s->mb_y;
7bc9090a 2804 const int mb_xy= mb_x + mb_y*s->mb_stride;
0da71265
MN
2805 const int mb_type= s->current_picture.mb_type[mb_xy];
2806 uint8_t *dest_y, *dest_cb, *dest_cr;
2807 int linesize, uvlinesize /*dct_offset*/;
2808 int i;
6867a90b 2809 int *block_offset = &h->block_offset[0];
0da71265
MN
2810
2811 if(!s->decode)
2812 return;
2813
0da71265
MN
2814 dest_y = s->current_picture.data[0] + (mb_y * 16* s->linesize ) + mb_x * 16;
2815 dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2816 dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2817
2818 if (h->mb_field_decoding_flag) {
2819 linesize = s->linesize * 2;
2820 uvlinesize = s->uvlinesize * 2;
6867a90b 2821 block_offset = &h->block_offset[24];
0da71265
MN
2822 if(mb_y&1){ //FIXME move out of this func?
2823 dest_y -= s->linesize*15;
6867a90b
LLL
2824 dest_cb-= s->uvlinesize*7;
2825 dest_cr-= s->uvlinesize*7;
0da71265
MN
2826 }
2827 } else {
2828 linesize = s->linesize;
2829 uvlinesize = s->uvlinesize;
2830// dct_offset = s->linesize * 16;
2831 }
2832
6fbcaaa0
LLL
2833 if (IS_INTRA_PCM(mb_type)) {
2834 unsigned int x, y;
2835
2836 // The pixels are stored in h->mb array in the same order as levels,
2837 // copy them in output in the correct order.
2838 for(i=0; i<16; i++) {
2839 for (y=0; y<4; y++) {
2840 for (x=0; x<4; x++) {
6867a90b 2841 *(dest_y + block_offset[i] + y*linesize + x) = h->mb[i*16+y*4+x];
6fbcaaa0
LLL
2842 }
2843 }
2844 }
2845 for(i=16; i<16+4; i++) {
2846 for (y=0; y<4; y++) {
2847 for (x=0; x<4; x++) {
6867a90b 2848 *(dest_cb + block_offset[i] + y*uvlinesize + x) = h->mb[i*16+y*4+x];
6fbcaaa0
LLL
2849 }
2850 }
2851 }
2852 for(i=20; i<20+4; i++) {
2853 for (y=0; y<4; y++) {
2854 for (x=0; x<4; x++) {
6867a90b 2855 *(dest_cr + block_offset[i] + y*uvlinesize + x) = h->mb[i*16+y*4+x];
6fbcaaa0
LLL
2856 }
2857 }
2858 }
e7e09b49
LLL
2859 } else {
2860 if(IS_INTRA(mb_type)){
2861 if(h->deblocking_filter)
2862 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1);
53c05b1e 2863
e7e09b49
LLL
2864 if(!(s->flags&CODEC_FLAG_GRAY)){
2865 h->pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
2866 h->pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
2867 }
0da71265 2868
e7e09b49
LLL
2869 if(IS_INTRA4x4(mb_type)){
2870 if(!s->encoding){
2871 for(i=0; i<16; i++){
6867a90b 2872 uint8_t * const ptr= dest_y + block_offset[i];
e7e09b49
LLL
2873 uint8_t *topright;
2874 const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
2875 int tr;
2876
2877 if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
2878 const int topright_avail= (h->topright_samples_available<<i)&0x8000;
6867a90b 2879 assert(mb_y || linesize <= block_offset[i]);
e7e09b49
LLL
2880 if(!topright_avail){
2881 tr= ptr[3 - linesize]*0x01010101;
2882 topright= (uint8_t*) &tr;
2883 }else if(i==5 && h->deblocking_filter){
2884 tr= *(uint32_t*)h->top_border[mb_x+1];
2885 topright= (uint8_t*) &tr;
2886 }else
2887 topright= ptr + 4 - linesize;
a9799653 2888 }else
e7e09b49
LLL
2889 topright= NULL;
2890
2891 h->pred4x4[ dir ](ptr, topright, linesize);
2892 if(h->non_zero_count_cache[ scan8[i] ]){
2893 if(s->codec_id == CODEC_ID_H264)
2894 s->dsp.h264_idct_add(ptr, h->mb + i*16, linesize);
2895 else
2896 svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
2897 }
8b82a956 2898 }
0da71265 2899 }
e7e09b49
LLL
2900 }else{
2901 h->pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
2902 if(s->codec_id == CODEC_ID_H264)
2903 h264_luma_dc_dequant_idct_c(h->mb, s->qscale);
2904 else
2905 svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
0da71265 2906 }
e7e09b49
LLL
2907 if(h->deblocking_filter)
2908 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);
2909 }else if(s->codec_id == CODEC_ID_H264){
2910 hl_motion(h, dest_y, dest_cb, dest_cr,
2911 s->dsp.put_h264_qpel_pixels_tab, s->dsp.put_h264_chroma_pixels_tab,
2912 s->dsp.avg_h264_qpel_pixels_tab, s->dsp.avg_h264_chroma_pixels_tab,
2913 s->dsp.weight_h264_pixels_tab, s->dsp.biweight_h264_pixels_tab);
0da71265 2914 }
e7e09b49
LLL
2915
2916
2917 if(!IS_INTRA4x4(mb_type)){
2918 if(s->codec_id == CODEC_ID_H264){
2919 for(i=0; i<16; i++){
2920 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
6867a90b 2921 uint8_t * const ptr= dest_y + block_offset[i];
e7e09b49
LLL
2922 s->dsp.h264_idct_add(ptr, h->mb + i*16, linesize);
2923 }
4704097a 2924 }
e7e09b49
LLL
2925 }else{
2926 for(i=0; i<16; i++){
2927 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
6867a90b 2928 uint8_t * const ptr= dest_y + block_offset[i];
e7e09b49
LLL
2929 svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
2930 }
4704097a 2931 }
0da71265
MN
2932 }
2933 }
0da71265 2934
e7e09b49
LLL
2935 if(!(s->flags&CODEC_FLAG_GRAY)){
2936 chroma_dc_dequant_idct_c(h->mb + 16*16, h->chroma_qp);
2937 chroma_dc_dequant_idct_c(h->mb + 16*16+4*16, h->chroma_qp);
2938 if(s->codec_id == CODEC_ID_H264){
2939 for(i=16; i<16+4; i++){
2940 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
6867a90b 2941 uint8_t * const ptr= dest_cb + block_offset[i];
e7e09b49
LLL
2942 s->dsp.h264_idct_add(ptr, h->mb + i*16, uvlinesize);
2943 }
4704097a 2944 }
e7e09b49
LLL
2945 for(i=20; i<20+4; i++){
2946 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
6867a90b 2947 uint8_t * const ptr= dest_cr + block_offset[i];
e7e09b49
LLL
2948 s->dsp.h264_idct_add(ptr, h->mb + i*16, uvlinesize);
2949 }
4704097a 2950 }
e7e09b49
LLL
2951 }else{
2952 for(i=16; i<16+4; i++){
2953 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
6867a90b 2954 uint8_t * const ptr= dest_cb + block_offset[i];
e7e09b49
LLL
2955 svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
2956 }
4704097a 2957 }
e7e09b49
LLL
2958 for(i=20; i<20+4; i++){
2959 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
6867a90b 2960 uint8_t * const ptr= dest_cr + block_offset[i];
e7e09b49
LLL
2961 svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
2962 }
4704097a 2963 }
0da71265
MN
2964 }
2965 }
2966 }
53c05b1e
MN
2967 if(h->deblocking_filter) {
2968 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
46f2f05f 2969 fill_caches(h, mb_type, 1); //FIXME dont fill stuff which isnt used by filter_mb
53c05b1e
MN
2970 filter_mb(h, mb_x, mb_y, dest_y, dest_cb, dest_cr);
2971 }
0da71265
MN
2972}
2973
0da71265
MN
2974/**
2975 * fills the default_ref_list.
2976 */
2977static int fill_default_ref_list(H264Context *h){
2978 MpegEncContext * const s = &h->s;
2979 int i;
827c91bf 2980 int smallest_poc_greater_than_current = -1;
0da71265
MN
2981 Picture sorted_short_ref[16];
2982
2983 if(h->slice_type==B_TYPE){
2984 int out_i;
2985 int limit= -1;
2986
827c91bf 2987 /* sort frame according to poc in B slice */
0da71265
MN
2988 for(out_i=0; out_i<h->short_ref_count; out_i++){
2989 int best_i=-1;
792bb815 2990 int best_poc=INT_MAX;
0da71265
MN
2991
2992 for(i=0; i<h->short_ref_count; i++){
2993 const int poc= h->short_ref[i]->poc;
2994 if(poc > limit && poc < best_poc){
2995 best_poc= poc;
2996 best_i= i;
2997 }
2998 }
2999
3000 assert(best_i != -1);
3001
3002 limit= best_poc;
3003 sorted_short_ref[out_i]= *h->short_ref[best_i];
827c91bf
LLL
3004 tprintf("sorted poc: %d->%d poc:%d fn:%d\n", best_i, out_i, sorted_short_ref[out_i].poc, sorted_short_ref[out_i].frame_num);
3005 if (-1 == smallest_poc_greater_than_current) {
3006 if (h->short_ref[best_i]->poc >= s->current_picture_ptr->poc) {
3007 smallest_poc_greater_than_current = out_i;
3008 }
3009 }
0da71265
MN
3010 }
3011 }
3012
3013 if(s->picture_structure == PICT_FRAME){
3014 if(h->slice_type==B_TYPE){
0da71265 3015 int list;
827c91bf 3016 tprintf("current poc: %d, smallest_poc_greater_than_current: %d\n", s->current_picture_ptr->poc, smallest_poc_greater_than_current);
0da71265 3017
827c91bf 3018 // find the largest poc
0da71265 3019 for(list=0; list<2; list++){
827c91bf 3020 int index = 0;
3b33943e
MN
3021 int j= -99;
3022 int step= list ? -1 : 1;
827c91bf 3023
3b33943e
MN
3024 for(i=0; i<h->short_ref_count && index < h->ref_count[list]; i++, j+=step) {
3025 while(j<0 || j>= h->short_ref_count){
3026 step = -step;
3027 j= smallest_poc_greater_than_current + (step>>1);
0da71265 3028 }
3b33943e
MN
3029 if(sorted_short_ref[j].reference != 3) continue;
3030 h->default_ref_list[list][index ]= sorted_short_ref[j];
3031 h->default_ref_list[list][index++].pic_id= sorted_short_ref[j].frame_num;
0da71265
MN
3032 }
3033
827c91bf
LLL
3034 for(i = 0; i < 16 && index < h->ref_count[ list ]; i++){
3035 if(h->long_ref[i] == NULL) continue;
0da71265
MN
3036 if(h->long_ref[i]->reference != 3) continue;
3037
3038 h->default_ref_list[ list ][index ]= *h->long_ref[i];
3039 h->default_ref_list[ list ][index++].pic_id= i;;
3040 }
3041
3b33943e 3042 if(list && (smallest_poc_greater_than_current<=0 || smallest_poc_greater_than_current>=h->short_ref_count) && (1 < index)){
827c91bf
LLL
3043 // swap the two first elements of L1 when
3044 // L0 and L1 are identical
0da71265
MN
3045 Picture temp= h->default_ref_list[1][0];
3046 h->default_ref_list[1][0] = h->default_ref_list[1][1];
3047 h->default_ref_list[1][0] = temp;
3048 }
3049
3050 if(index < h->ref_count[ list ])
3051 memset(&h->default_ref_list[list][index], 0, sizeof(Picture)*(h->ref_count[ list ] - index));
3052 }
3053 }else{
3054 int index=0;
2f944356 3055 for(i=0; i<h->short_ref_count; i++){
0da71265
MN
3056 if(h->short_ref[i]->reference != 3) continue; //FIXME refernce field shit
3057 h->default_ref_list[0][index ]= *h->short_ref[i];
3058 h->default_ref_list[0][index++].pic_id= h->short_ref[i]->frame_num;
3059 }
2f944356 3060 for(i = 0; i < 16; i++){
827c91bf 3061 if(h->long_ref[i] == NULL) continue;
0da71265
MN
3062 if(h->long_ref[i]->reference != 3) continue;
3063 h->default_ref_list[0][index ]= *h->long_ref[i];
3064 h->default_ref_list[0][index++].pic_id= i;;
3065 }
3066 if(index < h->ref_count[0])
3067 memset(&h->default_ref_list[0][index], 0, sizeof(Picture)*(h->ref_count[0] - index));
3068 }
3069 }else{ //FIELD
3070 if(h->slice_type==B_TYPE){
3071 }else{
3072 //FIXME second field balh
3073 }
3074 }
827c91bf
LLL
3075#ifdef TRACE
3076 for (i=0; i<h->ref_count[0]; i++) {
3077 tprintf("List0: %s fn:%d 0x%p\n", (h->default_ref_list[0][i].long_ref ? "LT" : "ST"), h->default_ref_list[0][i].pic_id, h->default_ref_list[0][i].data[0]);
3078 }
3079 if(h->slice_type==B_TYPE){
3080 for (i=0; i<h->ref_count[1]; i++) {
3081 tprintf("List1: %s fn:%d 0x%p\n", (h->default_ref_list[1][i].long_ref ? "LT" : "ST"), h->default_ref_list[1][i].pic_id, h->default_ref_list[0][i].data[0]);
3082 }
3083 }
3084#endif
0da71265
MN
3085 return 0;
3086}
3087
827c91bf
LLL
3088static void print_short_term(H264Context *h);
3089static void print_long_term(H264Context *h);
3090
0da71265
MN
3091static int decode_ref_pic_list_reordering(H264Context *h){
3092 MpegEncContext * const s = &h->s;
3093 int list;
3094
827c91bf
LLL
3095 print_short_term(h);
3096 print_long_term(h);
0da71265
MN
3097 if(h->slice_type==I_TYPE || h->slice_type==SI_TYPE) return 0; //FIXME move beofre func
3098
3099 for(list=0; list<2; list++){
3100 memcpy(h->ref_list[list], h->default_ref_list[list], sizeof(Picture)*h->ref_count[list]);
3101
3102 if(get_bits1(&s->gb)){
3103 int pred= h->curr_pic_num;
3104 int index;
3105
3106 for(index=0; ; index++){
3107 int reordering_of_pic_nums_idc= get_ue_golomb(&s->gb);
3108 int pic_id;
3109 int i;
2f944356 3110 Picture *ref = NULL;
0da71265 3111
0bc42cad
LM
3112 if(reordering_of_pic_nums_idc==3)
3113 break;
0da71265
MN
3114
3115 if(index >= h->ref_count[list]){
9b879566 3116 av_log(h->s.avctx, AV_LOG_ERROR, "reference count overflow\n");
0da71265
MN
3117 return -1;
3118 }
3119
3120 if(reordering_of_pic_nums_idc<3){
3121 if(reordering_of_pic_nums_idc<2){
3122 const int abs_diff_pic_num= get_ue_golomb(&s->gb) + 1;
3123
3124 if(abs_diff_pic_num >= h->max_pic_num){
9b879566 3125 av_log(h->s.avctx, AV_LOG_ERROR, "abs_diff_pic_num overflow\n");
0da71265
MN
3126 return -1;
3127 }
3128
3129 if(reordering_of_pic_nums_idc == 0) pred-= abs_diff_pic_num;
3130 else pred+= abs_diff_pic_num;
3131 pred &= h->max_pic_num - 1;
3132
2f944356
LM
3133 for(i= h->short_ref_count-1; i>=0; i--){
3134 ref = h->short_ref[i];
3135 if(ref->data[0] != NULL && ref->frame_num == pred && ref->long_ref == 0) // ignore non existing pictures by testing data[0] pointer
0da71265
MN
3136 break;
3137 }
3138 }else{
3139 pic_id= get_ue_golomb(&s->gb); //long_term_pic_idx
2f944356 3140 ref = h->long_ref[pic_id];
0da71265
MN
3141 }
3142
0d315f28 3143 if (i < 0) {
9b879566 3144 av_log(h->s.avctx, AV_LOG_ERROR, "reference picture missing during reorder\n");
0da71265 3145 memset(&h->ref_list[list][index], 0, sizeof(Picture)); //FIXME
2f944356
LM
3146 } else {
3147 h->ref_list[list][index]= *ref;
0da71265 3148 }
0bc42cad 3149 }else{
9b879566 3150 av_log(h->s.avctx, AV_LOG_ERROR, "illegal reordering_of_pic_nums_idc\n");
0da71265
MN
3151 return -1;
3152 }
3153 }
3154 }
3155
3156 if(h->slice_type!=B_TYPE) break;
3157 }
5ad984c9
LM
3158
3159 if(h->slice_type==B_TYPE && !h->direct_spatial_mv_pred)
3160 direct_dist_scale_factor(h);
2f944356 3161 direct_ref_list_init(h);
0da71265
MN
3162 return 0;
3163}
3164
3165static int pred_weight_table(H264Context *h){
3166 MpegEncContext * const s = &h->s;
3167 int list, i;
9f2d1b4f 3168 int luma_def, chroma_def;
0da71265 3169
9f2d1b4f
LM
3170 h->use_weight= 0;
3171 h->use_weight_chroma= 0;
0da71265
MN
3172 h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
3173 h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
9f2d1b4f
LM
3174 luma_def = 1<<h->luma_log2_weight_denom;
3175 chroma_def = 1<<h->chroma_log2_weight_denom;
0da71265
MN
3176
3177 for(list=0; list<2; list++){
3178 for(i=0; i<h->ref_count[list]; i++){
3179 int luma_weight_flag, chroma_weight_flag;
3180
3181 luma_weight_flag= get_bits1(&s->gb);
3182 if(luma_weight_flag){
3183 h->luma_weight[list][i]= get_se_golomb(&s->gb);
3184 h->luma_offset[list][i]= get_se_golomb(&s->gb);
9f2d1b4f
LM
3185 if( h->luma_weight[list][i] != luma_def
3186 || h->luma_offset[list][i] != 0)
3187 h->use_weight= 1;
3188 }else{
3189 h->luma_weight[list][i]= luma_def;
3190 h->luma_offset[list][i]= 0;
0da71265
MN
3191 }
3192
3193 chroma_weight_flag= get_bits1(&s->gb);
3194 if(chroma_weight_flag){
3195 int j;
3196 for(j=0; j<2; j++){
3197 h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);
3198 h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);
9f2d1b4f
LM
3199 if( h->chroma_weight[list][i][j] != chroma_def
3200 || h->chroma_offset[list][i][j] != 0)
3201 h->use_weight_chroma= 1;
3202 }
3203 }else{
3204 int j;
3205 for(j=0; j<2; j++){
3206 h->chroma_weight[list][i][j]= chroma_def;
3207 h->chroma_offset[list][i][j]= 0;
0da71265
MN
3208 }
3209 }
3210 }
3211 if(h->slice_type != B_TYPE) break;
3212 }
9f2d1b4f 3213 h->use_weight= h->use_weight || h->use_weight_chroma;
0da71265
MN
3214 return 0;
3215}
3216
9f2d1b4f
LM
3217static void implicit_weight_table(H264Context *h){
3218 MpegEncContext * const s = &h->s;
9f2d1b4f
LM
3219 int ref0, ref1;
3220 int cur_poc = s->current_picture_ptr->poc;
3221
3222 if( h->ref_count[0] == 1 && h->ref_count[1] == 1
3223 && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
3224 h->use_weight= 0;
3225 h->use_weight_chroma= 0;
3226 return;
3227 }
3228
3229 h->use_weight= 2;
3230 h->use_weight_chroma= 2;
3231 h->luma_log2_weight_denom= 5;
3232 h->chroma_log2_weight_denom= 5;
3233
3234 /* FIXME: MBAFF */
3235 for(ref0=0; ref0 < h->ref_count[0]; ref0++){
3236 int poc0 = h->ref_list[0][ref0].poc;
3237 for(ref1=0; ref1 < h->ref_count[1]; ref1++){
738386a5 3238 int poc1 = h->ref_list[1][ref1].poc;
9f2d1b4f
LM
3239 int td = clip(poc1 - poc0, -128, 127);
3240 if(td){
3241 int tb = clip(cur_poc - poc0, -128, 127);
3242 int tx = (16384 + (ABS(td) >> 1)) / td;
3243 int dist_scale_factor = clip((tb*tx + 32) >> 6, -1024, 1023) >> 2;
3244 if(dist_scale_factor < -64 || dist_scale_factor > 128)
3245 h->implicit_weight[ref0][ref1] = 32;
3246 else
3247 h->implicit_weight[ref0][ref1] = 64 - dist_scale_factor;
3248 }else
3249 h->implicit_weight[ref0][ref1] = 32;
3250 }
3251 }
3252}
3253
4e4d983e
LM
3254static inline void unreference_pic(H264Context *h, Picture *pic){
3255 int i;
3256 pic->reference=0;
3257 if(pic == h->delayed_output_pic)
3258 pic->reference=1;
3259 else{
3260 for(i = 0; h->delayed_pic[i]; i++)
3261 if(pic == h->delayed_pic[i]){
3262 pic->reference=1;
3263 break;
3264 }
3265 }
3266}
3267
0da71265 3268/**
5175b937 3269 * instantaneous decoder refresh.
0da71265
MN
3270 */
3271static void idr(H264Context *h){
4e4d983e 3272 int i;
0da71265 3273
dc032f33
LLL
3274 for(i=0; i<16; i++){
3275 if (h->long_ref[i] != NULL) {
4e4d983e 3276 unreference_pic(h, h->long_ref[i]);
dc032f33
LLL
3277 h->long_ref[i]= NULL;
3278 }
0da71265
MN
3279 }
3280 h->long_ref_count=0;
3281
3282 for(i=0; i<h->short_ref_count; i++){
4e4d983e 3283 unreference_pic(h, h->short_ref[i]);
0da71265
MN
3284 h->short_ref[i]= NULL;
3285 }
3286 h->short_ref_count=0;
3287}
3288
0da71265
MN
3289/**
3290 *
3291 * @return the removed picture or NULL if an error occures
3292 */
3293static Picture * remove_short(H264Context *h, int frame_num){
1924f3ce 3294 MpegEncContext * const s = &h->s;
0da71265
MN
3295 int i;
3296
1924f3ce 3297 if(s->avctx->debug&FF_DEBUG_MMCO)
9b879566 3298 av_log(h->s.avctx, AV_LOG_DEBUG, "remove short %d count %d\n", frame_num, h->short_ref_count);
1924f3ce 3299
0da71265
MN
3300 for(i=0; i<h->short_ref_count; i++){
3301 Picture *pic= h->short_ref[i];
1924f3ce 3302 if(s->avctx->debug&FF_DEBUG_MMCO)
9b879566 3303 av_log(h->s.avctx, AV_LOG_DEBUG, "%d %d %p\n", i, pic->frame_num, pic);
0da71265
MN
3304 if(pic->frame_num == frame_num){
3305 h->short_ref[i]= NULL;
3306 memmove(&h->short_ref[i], &h->short_ref[i+1], (h->short_ref_count - i - 1)*sizeof(Picture*));
3307 h->short_ref_count--;
3308 return pic;
3309 }
3310 }
3311 return NULL;
3312}
3313
3314/**
3315 *
3316 * @return the removed picture or NULL if an error occures
3317 */
3318static Picture * remove_long(H264Context *h, int i){
3319 Picture *pic;
3320
0da71265 3321 pic= h->long_ref[i];
0da71265 3322 h->long_ref[i]= NULL;
827c91bf 3323 if(pic) h->long_ref_count--;
0da71265
MN
3324
3325 return pic;
3326}
3327
3328/**
827c91bf
LLL
3329 * print short term list
3330 */
3331static void print_short_term(H264Context *h) {
3332 uint32_t i;
3333 if(h->s.avctx->debug&FF_DEBUG_MMCO) {
3334 av_log(h->s.avctx, AV_LOG_DEBUG, "short term list:\n");
3335 for(i=0; i<h->short_ref_count; i++){
3336 Picture *pic= h->short_ref[i];
3337 av_log(h->s.avctx, AV_LOG_DEBUG, "%d fn:%d poc:%d %p\n", i, pic->frame_num, pic->poc, pic->data[0]);
3338 }
3339 }
3340}
3341
3342/**
3343 * print long term list
3344 */
3345static void print_long_term(H264Context *h) {
3346 uint32_t i;
3347 if(h->s.avctx->debug&FF_DEBUG_MMCO) {
3348 av_log(h->s.avctx, AV_LOG_DEBUG, "long term list:\n");
3349 for(i = 0; i < 16; i++){
3350 Picture *pic= h->long_ref[i];
3351 if (pic) {
3352 av_log(h->s.avctx, AV_LOG_DEBUG, "%d fn:%d poc:%d %p\n", i, pic->frame_num, pic->poc, pic->data[0]);
3353 }
3354 }
3355 }
3356}
3357
3358/**
0da71265
MN
3359 * Executes the reference picture marking (memory management control operations).
3360 */
3361static int execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count){
3362 MpegEncContext * const s = &h->s;
827c91bf 3363 int i, j;
0da71265
MN
3364 int current_is_long=0;
3365 Picture *pic;
3366
3367 if((s->avctx->debug&FF_DEBUG_MMCO) && mmco_count==0)
9b879566 3368 av_log(h->s.avctx, AV_LOG_DEBUG, "no mmco here\n");
0da71265
MN
3369
3370 for(i=0; i<mmco_count; i++){
3371 if(s->avctx->debug&FF_DEBUG_MMCO)
9b879566 3372 av_log(h->s.avctx, AV_LOG_DEBUG, "mmco:%d %d %d\n", h->mmco[i].opcode, h->mmco[i].short_frame_num, h->mmco[i].long_index);
0da71265
MN
3373
3374 switch(mmco[i].opcode){
3375 case MMCO_SHORT2UNUSED:
3376 pic= remove_short(h, mmco[i].short_frame_num);
3377 if(pic==NULL) return -1;
4e4d983e 3378 unreference_pic(h, pic);
0da71265
MN
3379 break;
3380 case MMCO_SHORT2LONG:
3381 pic= remove_long(h, mmco[i].long_index);
4e4d983e 3382 if(pic) unreference_pic(h, pic);
0da71265
MN
3383
3384 h->long_ref[ mmco[i].long_index ]= remove_short(h, mmco[i].short_frame_num);
3385 h->long_ref[ mmco[i].long_index ]->long_ref=1;
dc032f33 3386 h->long_ref_count++;
0da71265
MN
3387 break;
3388 case MMCO_LONG2UNUSED:
3389 pic= remove_long(h, mmco[i].long_index);
3390 if(pic==NULL) return -1;
4e4d983e 3391 unreference_pic(h, pic);
0da71265
MN
3392 break;
3393 case MMCO_LONG:
3394 pic= remove_long(h, mmco[i].long_index);
4e4d983e 3395 if(pic) unreference_pic(h, pic);
0da71265
MN
3396
3397 h->long_ref[ mmco[i].long_index ]= s->current_picture_ptr;
3398 h->long_ref[ mmco[i].long_index ]->long_ref=1;
3399 h->long_ref_count++;
3400
3401 current_is_long=1;
3402 break;
3403 case MMCO_SET_MAX_LONG:
3404 assert(mmco[i].long_index <= 16);
827c91bf
LLL
3405 // just remove the long term which index is greater than new max
3406 for(j = mmco[i].long_index; j<16; j++){
3407 pic = remove_long(h, j);
4e4d983e 3408 if (pic) unreference_pic(h, pic);
0da71265
MN
3409 }
3410 break;
3411 case MMCO_RESET:
3412 while(h->short_ref_count){
3413 pic= remove_short(h, h->short_ref[0]->frame_num);
4e4d983e 3414 unreference_pic(h, pic);
0da71265 3415 }
827c91bf
LLL
3416 for(j = 0; j < 16; j++) {
3417 pic= remove_long(h, j);
4e4d983e 3418 if(pic) unreference_pic(h, pic);
0da71265
MN
3419 }
3420 break;
3421 default: assert(0);
3422 }
3423 }
3424
3425 if(!current_is_long){
3426 pic= remove_short(h, s->current_picture_ptr->frame_num);
3427 if(pic){
4e4d983e 3428 unreference_pic(h, pic);
9b879566 3429 av_log(h->s.avctx, AV_LOG_ERROR, "illegal short term buffer state detected\n");
0da71265
MN
3430 }
3431
3432 if(h->short_ref_count)
1924f3ce
MN
3433 memmove(&h->short_ref[1], &h->short_ref[0], h->short_ref_count*sizeof(Picture*));
3434
3435 h->short_ref[0]= s->current_picture_ptr;
0da71265
MN
3436 h->short_ref[0]->long_ref=0;
3437 h->short_ref_count++;
3438 }
3439
827c91bf
LLL
3440 print_short_term(h);
3441 print_long_term(h);
0da71265
MN
3442 return 0;
3443}
3444
3445static int decode_ref_pic_marking(H264Context *h){
3446 MpegEncContext * const s = &h->s;
3447 int i;
3448
3449 if(h->nal_unit_type == NAL_IDR_SLICE){ //FIXME fields
3450 s->broken_link= get_bits1(&s->gb) -1;
3451 h->mmco[0].long_index= get_bits1(&s->gb) - 1; // current_long_term_idx
3452 if(h->mmco[0].long_index == -1)
3453 h->mmco_index= 0;
3454 else{
3455 h->mmco[0].opcode= MMCO_LONG;
3456 h->mmco_index= 1;
3457 }
3458 }else{
3459 if(get_bits1(&s->gb)){ // adaptive_ref_pic_marking_mode_flag
579448a0 3460 for(i= 0; i<MAX_MMCO_COUNT; i++) {
0da71265
MN
3461 MMCOOpcode opcode= get_ue_golomb(&s->gb);;
3462
3463 h->mmco[i].opcode= opcode;
3464 if(opcode==MMCO_SHORT2UNUSED || opcode==MMCO_SHORT2LONG){
3465 h->mmco[i].short_frame_num= (h->frame_num - get_ue_golomb(&s->gb) - 1) & ((1<<h->sps.log2_max_frame_num)-1); //FIXME fields
3466/* if(h->mmco[i].short_frame_num >= h->short_ref_count || h->short_ref[ h->mmco[i].short_frame_num ] == NULL){
3467 fprintf(stderr, "illegal short ref in memory management control operation %d\n", mmco);
3468 return -1;
3469 }*/
3470 }
3471 if(opcode==MMCO_SHORT2LONG || opcode==MMCO_LONG2UNUSED || opcode==MMCO_LONG || opcode==MMCO_SET_MAX_LONG){
3472 h->mmco[i].long_index= get_ue_golomb(&s->gb);
3473 if(/*h->mmco[i].long_index >= h->long_ref_count || h->long_ref[ h->mmco[i].long_index ] == NULL*/ h->mmco[i].long_index >= 16){
9b879566 3474 av_log(h->s.avctx, AV_LOG_ERROR, "illegal long ref in memory management control operation %d\n", opcode);
0da71265
MN
3475 return -1;
3476 }
3477 }
3478
3479 if(opcode > MMCO_LONG){
9b879566 3480 av_log(h->s.avctx, AV_LOG_ERROR, "illegal memory management control operation %d\n", opcode);
0da71265
MN
3481 return -1;
3482 }
792bb815
MN
3483 if(opcode == MMCO_END)
3484 break;
0da71265
MN
3485 }
3486 h->mmco_index= i;
3487 }else{
3488 assert(h->long_ref_count + h->short_ref_count <= h->sps.ref_frame_count);
3489
3490 if(h->long_ref_count + h->short_ref_count == h->sps.ref_frame_count){ //FIXME fields
3491 h->mmco[0].opcode= MMCO_SHORT2UNUSED;
3492 h->mmco[0].short_frame_num= h->short_ref[ h->short_ref_count - 1 ]->frame_num;
3493 h->mmco_index= 1;
3494 }else
3495 h->mmco_index= 0;
3496 }
3497 }
3498
3499 return 0;
3500}
3501
3502static int init_poc(H264Context *h){
3503 MpegEncContext * const s = &h->s;
3504 const int max_frame_num= 1<<h->sps.log2_max_frame_num;
3505 int field_poc[2];
3506
3507 if(h->nal_unit_type == NAL_IDR_SLICE){
3508 h->frame_num_offset= 0;
3509 }else{
3510 if(h->frame_num < h->prev_frame_num)
3511 h->frame_num_offset= h->prev_frame_num_offset + max_frame_num;
3512 else
3513 h->frame_num_offset= h->prev_frame_num_offset;
3514 }
3515
3516 if(h->sps.poc_type==0){
3517 const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
3518
3519 if (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
3520 h->poc_msb = h->prev_poc_msb + max_poc_lsb;
3521 else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
3522 h->poc_msb = h->prev_poc_msb - max_poc_lsb;
3523 else
3524 h->poc_msb = h->prev_poc_msb;
3525//printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
3526 field_poc[0] =
3527 field_poc[1] = h->poc_msb + h->poc_lsb;
3528 if(s->picture_structure == PICT_FRAME)
3529 field_poc[1] += h->delta_poc_bottom;
3530 }else if(h->sps.poc_type==1){
3531 int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
3532 int i;
3533
3534 if(h->sps.poc_cycle_length != 0)
3535 abs_frame_num = h->frame_num_offset + h->frame_num;
3536 else
3537 abs_frame_num = 0;
3538
3539 if(h->nal_ref_idc==0 && abs_frame_num > 0)
3540 abs_frame_num--;
3541
3542 expected_delta_per_poc_cycle = 0;
3543 for(i=0; i < h->sps.poc_cycle_length; i++)
3544 expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
3545
3546 if(abs_frame_num > 0){
3547 int poc_cycle_cnt = (abs_frame_num - 1) / h->sps.poc_cycle_length;
3548 int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
3549
3550 expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
3551 for(i = 0; i <= frame_num_in_poc_cycle; i++)
3552 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
3553 } else
3554 expectedpoc = 0;
3555
3556 if(h->nal_ref_idc == 0)
3557 expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
3558
3559 field_poc[0] = expectedpoc + h->delta_poc[0];
3560 field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
3561
3562 if(s->picture_structure == PICT_FRAME)
3563 field_poc[1] += h->delta_poc[1];
3564 }else{
3565 int poc;
3566 if(h->nal_unit_type == NAL_IDR_SLICE){
3567 poc= 0;
3568 }else{
3569 if(h->nal_ref_idc) poc= 2*(h->frame_num_offset + h->frame_num);
3570 else poc= 2*(h->frame_num_offset + h->frame_num) - 1;
3571 }
3572 field_poc[0]= poc;
3573 field_poc[1]= poc;
3574 }
3575
3576 if(s->picture_structure != PICT_BOTTOM_FIELD)
3577 s->current_picture_ptr->field_poc[0]= field_poc[0];
3578 if(s->picture_structure != PICT_TOP_FIELD)
3579 s->current_picture_ptr->field_poc[1]= field_poc[1];
3580 if(s->picture_structure == PICT_FRAME) // FIXME field pix?
3581 s->current_picture_ptr->poc= FFMIN(field_poc[0], field_poc[1]);
3582
3583 return 0;
3584}
3585
3586/**
3587 * decodes a slice header.
3588 * this will allso call MPV_common_init() and frame_start() as needed
3589 */
3590static int decode_slice_header(H264Context *h){
3591 MpegEncContext * const s = &h->s;
3592 int first_mb_in_slice, pps_id;
3593 int num_ref_idx_active_override_flag;
3594 static const uint8_t slice_type_map[5]= {P_TYPE, B_TYPE, I_TYPE, SP_TYPE, SI_TYPE};
0bf79634
LLL
3595 int slice_type;
3596 int default_ref_list_done = 0;
0da71265
MN
3597
3598 s->current_picture.reference= h->nal_ref_idc != 0;
2f944356 3599 s->dropable= h->nal_ref_idc == 0;
0da71265
MN
3600
3601 first_mb_in_slice= get_ue_golomb(&s->gb);
3602
0bf79634
LLL
3603 slice_type= get_ue_golomb(&s->gb);
3604 if(slice_type > 9){
9b879566 3605 av_log(h->s.avctx, AV_LOG_ERROR, "slice type too large (%d) at %d %d\n", h->slice_type, s->mb_x, s->mb_y);
5175b937 3606 return -1;
0da71265 3607 }
0bf79634
LLL
3608 if(slice_type > 4){
3609 slice_type -= 5;
0da71265
MN
3610 h->slice_type_fixed=1;
3611 }else
3612 h->slice_type_fixed=0;
3613
0bf79634
LLL
3614 slice_type= slice_type_map[ slice_type ];
3615 if (slice_type == I_TYPE
3616 || (h->slice_num != 0 && slice_type == h->slice_type) ) {
3617 default_ref_list_done = 1;
3618 }
3619 h->slice_type= slice_type;
3620
0da71265
MN
3621 s->pict_type= h->slice_type; // to make a few old func happy, its wrong though
3622
3623 pps_id= get_ue_golomb(&s->gb);
3624 if(pps_id>255){
9b879566 3625 av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
0da71265
MN
3626 return -1;
3627 }
3628 h->pps= h->pps_buffer[pps_id];
8b92b792 3629 if(h->pps.slice_group_count == 0){
9b879566 3630 av_log(h->s.avctx, AV_LOG_ERROR, "non existing PPS referenced\n");
8b92b792
MN
3631 return -1;
3632 }
3633
0da71265 3634 h->sps= h->sps_buffer[ h->pps.sps_id ];
8b92b792 3635 if(h->sps.log2_max_frame_num == 0){
9b879566 3636 av_log(h->s.avctx, AV_LOG_ERROR, "non existing SPS referenced\n");
8b92b792
MN
3637 return -1;
3638 }
0da71265
MN
3639
3640 s->mb_width= h->sps.mb_width;
6867a90b 3641 s->mb_height= h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
0da71265 3642
650cec0c
LM
3643 h->b_stride= s->mb_width*4 + 1;
3644 h->b8_stride= s->mb_width*2 + 1;
0da71265 3645
a15e68de 3646 s->width = 16*s->mb_width - 2*(h->sps.crop_left + h->sps.crop_right );
0da71265 3647 if(h->sps.frame_mbs_only_flag)
a15e68de 3648 s->height= 16*s->mb_height - 2*(h->sps.crop_top + h->sps.crop_bottom);
0da71265 3649 else
a15e68de 3650 s->height= 16*s->mb_height - 4*(h->sps.crop_top + h->sps.crop_bottom); //FIXME recheck
0da71265 3651
0da71265 3652 if (s->context_initialized
5ff85f1d 3653 && ( s->width != s->avctx->width || s->height != s->avctx->height)) {
0da71265
MN
3654 free_tables(h);
3655 MPV_common_end(s);
3656 }
3657 if (!s->context_initialized) {
3658 if (MPV_common_init(s) < 0)
3659 return -1;
3660
3661 alloc_tables(h);
3662
3663 s->avctx->width = s->width;
3664 s->avctx->height = s->height;
5ff85f1d 3665 s->avctx->sample_aspect_ratio= h->sps.sar;
eed4ebd2
MN
3666 if(!s->avctx->sample_aspect_ratio.den)
3667