e73f32d7a2332337a288094ccf97c55473e9c0d1
[libav.git] / libavcodec / h264.c
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
34 #undef NDEBUG
35 #include <assert.h>
36
37 #define interlaced_dct interlaced_dct_is_a_bad_name
38 #define mb_intra mb_intra_isnt_initalized_see_mb_type
39
40 #define LUMA_DC_BLOCK_INDEX 25
41 #define CHROMA_DC_BLOCK_INDEX 26
42
43 #define CHROMA_DC_COEFF_TOKEN_VLC_BITS 8
44 #define COEFF_TOKEN_VLC_BITS 8
45 #define TOTAL_ZEROS_VLC_BITS 9
46 #define CHROMA_DC_TOTAL_ZEROS_VLC_BITS 3
47 #define RUN_VLC_BITS 3
48 #define RUN7_VLC_BITS 6
49
50 #define MAX_SPS_COUNT 32
51 #define MAX_PPS_COUNT 256
52
53 #define MAX_MMCO_COUNT 66
54
55 /**
56 * Sequence parameter set
57 */
58 typedef struct SPS{
59
60 int profile_idc;
61 int level_idc;
62 int multiple_slice_groups; ///< more_than_one_slice_group_allowed_flag
63 int arbitrary_slice_order; ///< arbitrary_slice_order_allowed_flag
64 int redundant_slices; ///< redundant_slices_allowed_flag
65 int log2_max_frame_num; ///< log2_max_frame_num_minus4 + 4
66 int poc_type; ///< pic_order_cnt_type
67 int log2_max_poc_lsb; ///< log2_max_pic_order_cnt_lsb_minus4
68 int delta_pic_order_always_zero_flag;
69 int offset_for_non_ref_pic;
70 int offset_for_top_to_bottom_field;
71 int poc_cycle_length; ///< num_ref_frames_in_pic_order_cnt_cycle
72 int ref_frame_count; ///< num_ref_frames
73 int required_frame_num_update_behaviour_flag;
74 int mb_width; ///< frame_width_in_mbs_minus1 + 1
75 int mb_height; ///< frame_height_in_mbs_minus1 + 1
76 int frame_mbs_only_flag;
77 int mb_aff; ///<mb_adaptive_frame_field_flag
78 int direct_8x8_inference_flag;
79 int vui_parameters_present_flag;
80 int sar_width;
81 int sar_height;
82 short offset_for_ref_frame[256]; //FIXME dyn aloc?
83 }SPS;
84
85 /**
86 * Picture parameter set
87 */
88 typedef struct PPS{
89 int sps_id;
90 int cabac; ///< entropy_coding_mode_flag
91 int pic_order_present; ///< pic_order_present_flag
92 int slice_group_count; ///< num_slice_groups_minus1 + 1
93 int mb_slice_group_map_type;
94 int ref_count[2]; ///< num_ref_idx_l0/1_active_minus1 + 1
95 int weighted_pred; ///< weighted_pred_flag
96 int weighted_bipred_idc;
97 int init_qp; ///< pic_init_qp_minus26 + 26
98 int init_qs; ///< pic_init_qs_minus26 + 26
99 int chroma_qp_index_offset;
100 int deblocking_filter_parameters_present; ///< deblocking_filter_parameters_present_flag
101 int constrained_intra_pred; ///< constrained_intra_pred_flag
102 int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag
103 int crop; ///< frame_cropping_flag
104 int crop_left; ///< frame_cropping_rect_left_offset
105 int crop_right; ///< frame_cropping_rect_right_offset
106 int crop_top; ///< frame_cropping_rect_top_offset
107 int crop_bottom; ///< frame_cropping_rect_bottom_offset
108 }PPS;
109
110 /**
111 * Memory management control operation opcode.
112 */
113 typedef enum MMCOOpcode{
114 MMCO_END=0,
115 MMCO_SHORT2UNUSED,
116 MMCO_LONG2UNUSED,
117 MMCO_SHORT2LONG,
118 MMCO_SET_MAX_LONG,
119 MMCO_RESET,
120 MMCO_LONG,
121 } MMCOOpcode;
122
123 /**
124 * Memory management control operation.
125 */
126 typedef struct MMCO{
127 MMCOOpcode opcode;
128 int short_frame_num;
129 int long_index;
130 } MMCO;
131
132 /**
133 * H264Context
134 */
135 typedef struct H264Context{
136 MpegEncContext s;
137 int nal_ref_idc;
138 int nal_unit_type;
139 #define NAL_SLICE 1
140 #define NAL_DPA 2
141 #define NAL_DPB 3
142 #define NAL_DPC 4
143 #define NAL_IDR_SLICE 5
144 #define NAL_SEI 6
145 #define NAL_SPS 7
146 #define NAL_PPS 8
147 #define NAL_PICTURE_DELIMITER 9
148 #define NAL_FILTER_DATA 10
149 uint8_t *rbsp_buffer;
150 int rbsp_buffer_size;
151
152 int chroma_qp; //QPc
153
154 int prev_mb_skiped; //FIXME remove (IMHO not used)
155
156 //prediction stuff
157 int chroma_pred_mode;
158 int intra16x16_pred_mode;
159
160 int8_t intra4x4_pred_mode_cache[5*8];
161 int8_t (*intra4x4_pred_mode)[8];
162 void (*pred4x4 [9+3])(uint8_t *src, uint8_t *topright, int stride);//FIXME move to dsp?
163 void (*pred8x8 [4+3])(uint8_t *src, int stride);
164 void (*pred16x16[4+3])(uint8_t *src, int stride);
165 unsigned int topleft_samples_available;
166 unsigned int top_samples_available;
167 unsigned int topright_samples_available;
168 unsigned int left_samples_available;
169
170 /**
171 * non zero coeff count cache.
172 * is 64 if not available.
173 */
174 uint8_t non_zero_count_cache[6*8];
175 uint8_t (*non_zero_count)[16];
176
177 /**
178 * Motion vector cache.
179 */
180 int16_t mv_cache[2][5*8][2];
181 int8_t ref_cache[2][5*8];
182 #define LIST_NOT_USED -1 //FIXME rename?
183 #define PART_NOT_AVAILABLE -2
184
185 /**
186 * is 1 if the specific list MV&references are set to 0,0,-2.
187 */
188 int mv_cache_clean[2];
189
190 int block_offset[16+8];
191 int chroma_subblock_offset[16]; //FIXME remove
192
193 uint16_t *mb2b_xy; //FIXME are these 4 a good idea?
194 uint16_t *mb2b8_xy;
195 int b_stride;
196 int b8_stride;
197
198 int halfpel_flag;
199 int thirdpel_flag;
200
201 int unknown_svq3_flag;
202 int next_slice_index;
203
204 SPS sps_buffer[MAX_SPS_COUNT];
205 SPS sps; ///< current sps
206
207 PPS pps_buffer[MAX_PPS_COUNT];
208 /**
209 * current pps
210 */
211 PPS pps; //FIXME move tp Picture perhaps? (->no) do we need that?
212
213 int slice_num;
214 uint8_t *slice_table_base;
215 uint8_t *slice_table; ///< slice_table_base + mb_stride + 1
216 int slice_type;
217 int slice_type_fixed;
218
219 //interlacing specific flags
220 int mb_field_decoding_flag;
221
222 int sub_mb_type[4];
223
224 //POC stuff
225 int poc_lsb;
226 int poc_msb;
227 int delta_poc_bottom;
228 int delta_poc[2];
229 int frame_num;
230 int prev_poc_msb; ///< poc_msb of the last reference pic for POC type 0
231 int prev_poc_lsb; ///< poc_lsb of the last reference pic for POC type 0
232 int frame_num_offset; ///< for POC type 2
233 int prev_frame_num_offset; ///< for POC type 2
234 int prev_frame_num; ///< frame_num of the last pic for POC type 1/2
235
236 /**
237 * frame_num for frames or 2*frame_num for field pics.
238 */
239 int curr_pic_num;
240
241 /**
242 * max_frame_num or 2*max_frame_num for field pics.
243 */
244 int max_pic_num;
245
246 //Weighted pred stuff
247 int luma_log2_weight_denom;
248 int chroma_log2_weight_denom;
249 int luma_weight[2][16];
250 int luma_offset[2][16];
251 int chroma_weight[2][16][2];
252 int chroma_offset[2][16][2];
253
254 //deblock
255 int disable_deblocking_filter_idc;
256 int slice_alpha_c0_offset_div2;
257 int slice_beta_offset_div2;
258
259 int redundant_pic_count;
260
261 int direct_spatial_mv_pred;
262
263 /**
264 * num_ref_idx_l0/1_active_minus1 + 1
265 */
266 int ref_count[2];// FIXME split for AFF
267 Picture *short_ref[16];
268 Picture *long_ref[16];
269 Picture default_ref_list[2][32];
270 Picture ref_list[2][32]; //FIXME size?
271 Picture field_ref_list[2][32]; //FIXME size?
272
273 /**
274 * memory management control operations buffer.
275 */
276 MMCO mmco[MAX_MMCO_COUNT];
277 int mmco_index;
278
279 int long_ref_count; ///< number of actual long term references
280 int short_ref_count; ///< number of actual short term references
281
282 //data partitioning
283 GetBitContext intra_gb;
284 GetBitContext inter_gb;
285 GetBitContext *intra_gb_ptr;
286 GetBitContext *inter_gb_ptr;
287
288 DCTELEM mb[16*24] __align8;
289 }H264Context;
290
291 static VLC coeff_token_vlc[4];
292 static VLC chroma_dc_coeff_token_vlc;
293
294 static VLC total_zeros_vlc[15];
295 static VLC chroma_dc_total_zeros_vlc[3];
296
297 static VLC run_vlc[6];
298 static VLC run7_vlc;
299
300 static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
301 static void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
302
303 static inline uint32_t pack16to32(int a, int b){
304 #ifdef WORDS_BIGENDIAN
305 return (b&0xFFFF) + (a<<16);
306 #else
307 return (a&0xFFFF) + (b<<16);
308 #endif
309 }
310
311 /**
312 * fill a rectangle.
313 * @param h height of the recatangle, should be a constant
314 * @param w width of the recatangle, should be a constant
315 * @param size the size of val (1 or 4), should be a constant
316 */
317 static inline void fill_rectangle(void *vp, int w, int h, int stride, uint32_t val, int size){ //FIXME ensure this IS inlined
318 uint8_t *p= (uint8_t*)vp;
319 assert(size==1 || size==4);
320
321 w *= size;
322 stride *= size;
323
324 //FIXME check what gcc generates for 64 bit on x86 and possible write a 32 bit ver of it
325 if(w==2 && h==2){
326 *(uint16_t*)(p + 0)=
327 *(uint16_t*)(p + stride)= size==4 ? val : val*0x0101;
328 }else if(w==2 && h==4){
329 *(uint16_t*)(p + 0*stride)=
330 *(uint16_t*)(p + 1*stride)=
331 *(uint16_t*)(p + 2*stride)=
332 *(uint16_t*)(p + 3*stride)= size==4 ? val : val*0x0101;
333 }else if(w==4 && h==1){
334 *(uint32_t*)(p + 0*stride)= size==4 ? val : val*0x01010101;
335 }else if(w==4 && h==2){
336 *(uint32_t*)(p + 0*stride)=
337 *(uint32_t*)(p + 1*stride)= size==4 ? val : val*0x01010101;
338 }else if(w==4 && h==4){
339 *(uint32_t*)(p + 0*stride)=
340 *(uint32_t*)(p + 1*stride)=
341 *(uint32_t*)(p + 2*stride)=
342 *(uint32_t*)(p + 3*stride)= size==4 ? val : val*0x01010101;
343 }else if(w==8 && h==1){
344 *(uint32_t*)(p + 0)=
345 *(uint32_t*)(p + 4)= size==4 ? val : val*0x01010101;
346 }else if(w==8 && h==2){
347 *(uint32_t*)(p + 0 + 0*stride)=
348 *(uint32_t*)(p + 4 + 0*stride)=
349 *(uint32_t*)(p + 0 + 1*stride)=
350 *(uint32_t*)(p + 4 + 1*stride)= size==4 ? val : val*0x01010101;
351 }else if(w==8 && h==4){
352 *(uint64_t*)(p + 0*stride)=
353 *(uint64_t*)(p + 1*stride)=
354 *(uint64_t*)(p + 2*stride)=
355 *(uint64_t*)(p + 3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
356 }else if(w==16 && h==2){
357 *(uint64_t*)(p + 0+0*stride)=
358 *(uint64_t*)(p + 8+0*stride)=
359 *(uint64_t*)(p + 0+1*stride)=
360 *(uint64_t*)(p + 8+1*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
361 }else if(w==16 && h==4){
362 *(uint64_t*)(p + 0+0*stride)=
363 *(uint64_t*)(p + 8+0*stride)=
364 *(uint64_t*)(p + 0+1*stride)=
365 *(uint64_t*)(p + 8+1*stride)=
366 *(uint64_t*)(p + 0+2*stride)=
367 *(uint64_t*)(p + 8+2*stride)=
368 *(uint64_t*)(p + 0+3*stride)=
369 *(uint64_t*)(p + 8+3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
370 }else
371 assert(0);
372 }
373
374 static inline void fill_caches(H264Context *h, int mb_type){
375 MpegEncContext * const s = &h->s;
376 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
377 int topleft_xy, top_xy, topright_xy, left_xy[2];
378 int topleft_type, top_type, topright_type, left_type[2];
379 int left_block[4];
380 int i;
381
382 //wow what a mess, why didnt they simplify the interlacing&intra stuff, i cant imagine that these complex rules are worth it
383
384 if(h->sps.mb_aff){
385 //FIXME
386 }else{
387 topleft_xy = mb_xy-1 - s->mb_stride;
388 top_xy = mb_xy - s->mb_stride;
389 topright_xy= mb_xy+1 - s->mb_stride;
390 left_xy[0] = mb_xy-1;
391 left_xy[1] = mb_xy-1;
392 left_block[0]= 0;
393 left_block[1]= 1;
394 left_block[2]= 2;
395 left_block[3]= 3;
396 }
397
398 topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
399 top_type = h->slice_table[top_xy ] == h->slice_num ? s->current_picture.mb_type[top_xy] : 0;
400 topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
401 left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
402 left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
403
404 if(IS_INTRA(mb_type)){
405 h->topleft_samples_available=
406 h->top_samples_available=
407 h->left_samples_available= 0xFFFF;
408 h->topright_samples_available= 0xEEEA;
409
410 if(!IS_INTRA(top_type) && (top_type==0 || h->pps.constrained_intra_pred)){
411 h->topleft_samples_available= 0xB3FF;
412 h->top_samples_available= 0x33FF;
413 h->topright_samples_available= 0x26EA;
414 }
415 for(i=0; i<2; i++){
416 if(!IS_INTRA(left_type[i]) && (left_type[i]==0 || h->pps.constrained_intra_pred)){
417 h->topleft_samples_available&= 0xDF5F;
418 h->left_samples_available&= 0x5F5F;
419 }
420 }
421
422 if(!IS_INTRA(topleft_type) && (topleft_type==0 || h->pps.constrained_intra_pred))
423 h->topleft_samples_available&= 0x7FFF;
424
425 if(!IS_INTRA(topright_type) && (topright_type==0 || h->pps.constrained_intra_pred))
426 h->topright_samples_available&= 0xFBFF;
427
428 if(IS_INTRA4x4(mb_type)){
429 if(IS_INTRA4x4(top_type)){
430 h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
431 h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
432 h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
433 h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
434 }else{
435 int pred;
436 if(IS_INTRA16x16(top_type) || (IS_INTER(top_type) && !h->pps.constrained_intra_pred))
437 pred= 2;
438 else{
439 pred= -1;
440 }
441 h->intra4x4_pred_mode_cache[4+8*0]=
442 h->intra4x4_pred_mode_cache[5+8*0]=
443 h->intra4x4_pred_mode_cache[6+8*0]=
444 h->intra4x4_pred_mode_cache[7+8*0]= pred;
445 }
446 for(i=0; i<2; i++){
447 if(IS_INTRA4x4(left_type[i])){
448 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
449 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
450 }else{
451 int pred;
452 if(IS_INTRA16x16(left_type[i]) || (IS_INTER(left_type[i]) && !h->pps.constrained_intra_pred))
453 pred= 2;
454 else{
455 pred= -1;
456 }
457 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
458 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
459 }
460 }
461 }
462 }
463
464
465 /*
466 0 . T T. T T T T
467 1 L . .L . . . .
468 2 L . .L . . . .
469 3 . T TL . . . .
470 4 L . .L . . . .
471 5 L . .. . . . .
472 */
473 //FIXME constraint_intra_pred & partitioning & nnz (lets hope this is just a typo in the spec)
474 if(top_type){
475 h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][0];
476 h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][1];
477 h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][2];
478 h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3];
479
480 h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][7];
481 h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8];
482
483 h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][10];
484 h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11];
485 }else{
486 h->non_zero_count_cache[4+8*0]=
487 h->non_zero_count_cache[5+8*0]=
488 h->non_zero_count_cache[6+8*0]=
489 h->non_zero_count_cache[7+8*0]=
490
491 h->non_zero_count_cache[1+8*0]=
492 h->non_zero_count_cache[2+8*0]=
493
494 h->non_zero_count_cache[1+8*3]=
495 h->non_zero_count_cache[2+8*3]= 64;
496 }
497
498 if(left_type[0]){
499 h->non_zero_count_cache[3+8*1]= h->non_zero_count[left_xy[0]][6];
500 h->non_zero_count_cache[3+8*2]= h->non_zero_count[left_xy[0]][5];
501 h->non_zero_count_cache[0+8*1]= h->non_zero_count[left_xy[0]][9]; //FIXME left_block
502 h->non_zero_count_cache[0+8*4]= h->non_zero_count[left_xy[0]][12];
503 }else{
504 h->non_zero_count_cache[3+8*1]=
505 h->non_zero_count_cache[3+8*2]=
506 h->non_zero_count_cache[0+8*1]=
507 h->non_zero_count_cache[0+8*4]= 64;
508 }
509
510 if(left_type[1]){
511 h->non_zero_count_cache[3+8*3]= h->non_zero_count[left_xy[1]][4];
512 h->non_zero_count_cache[3+8*4]= h->non_zero_count[left_xy[1]][3];
513 h->non_zero_count_cache[0+8*2]= h->non_zero_count[left_xy[1]][8];
514 h->non_zero_count_cache[0+8*5]= h->non_zero_count[left_xy[1]][11];
515 }else{
516 h->non_zero_count_cache[3+8*3]=
517 h->non_zero_count_cache[3+8*4]=
518 h->non_zero_count_cache[0+8*2]=
519 h->non_zero_count_cache[0+8*5]= 64;
520 }
521
522 #if 1
523 if(IS_INTER(mb_type)){
524 int list;
525 for(list=0; list<2; list++){
526 if((!IS_8X8(mb_type)) && !USES_LIST(mb_type, list)){
527 /*if(!h->mv_cache_clean[list]){
528 memset(h->mv_cache [list], 0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
529 memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
530 h->mv_cache_clean[list]= 1;
531 }*/
532 continue; //FIXME direct mode ...
533 }
534 h->mv_cache_clean[list]= 0;
535
536 if(IS_INTER(topleft_type)){
537 const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
538 const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + h->b8_stride;
539 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
540 h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
541 }else{
542 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
543 h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
544 }
545
546 if(IS_INTER(top_type)){
547 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
548 const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
549 *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0];
550 *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];
551 *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];
552 *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];
553 h->ref_cache[list][scan8[0] + 0 - 1*8]=
554 h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
555 h->ref_cache[list][scan8[0] + 2 - 1*8]=
556 h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
557 }else{
558 *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]=
559 *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]=
560 *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]=
561 *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;
562 *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
563 }
564
565 if(IS_INTER(topright_type)){
566 const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
567 const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
568 *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
569 h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
570 }else{
571 *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
572 h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
573 }
574
575 //FIXME unify cleanup or sth
576 if(IS_INTER(left_type[0])){
577 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
578 const int b8_xy= h->mb2b8_xy[left_xy[0]] + 1;
579 *(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]];
580 *(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]];
581 h->ref_cache[list][scan8[0] - 1 + 0*8]=
582 h->ref_cache[list][scan8[0] - 1 + 1*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0]>>1)];
583 }else{
584 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 0*8]=
585 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 1*8]= 0;
586 h->ref_cache[list][scan8[0] - 1 + 0*8]=
587 h->ref_cache[list][scan8[0] - 1 + 1*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
588 }
589
590 if(IS_INTER(left_type[1])){
591 const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
592 const int b8_xy= h->mb2b8_xy[left_xy[1]] + 1;
593 *(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]];
594 *(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]];
595 h->ref_cache[list][scan8[0] - 1 + 2*8]=
596 h->ref_cache[list][scan8[0] - 1 + 3*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[2]>>1)];
597 }else{
598 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 2*8]=
599 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 3*8]= 0;
600 h->ref_cache[list][scan8[0] - 1 + 2*8]=
601 h->ref_cache[list][scan8[0] - 1 + 3*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
602 }
603
604 h->ref_cache[list][scan8[5 ]+1] =
605 h->ref_cache[list][scan8[7 ]+1] =
606 h->ref_cache[list][scan8[13]+1] = //FIXME remove past 3 (init somewher else)
607 h->ref_cache[list][scan8[4 ]] =
608 h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
609 *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
610 *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
611 *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewher else)
612 *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
613 *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
614 }
615 //FIXME
616
617 }
618 #endif
619 }
620
621 static inline void write_back_intra_pred_mode(H264Context *h){
622 MpegEncContext * const s = &h->s;
623 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
624
625 h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
626 h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
627 h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
628 h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
629 h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
630 h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
631 h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
632 }
633
634 /**
635 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
636 */
637 static inline int check_intra4x4_pred_mode(H264Context *h){
638 MpegEncContext * const s = &h->s;
639 static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
640 static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
641 int i;
642
643 if(!(h->top_samples_available&0x8000)){
644 for(i=0; i<4; i++){
645 int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
646 if(status<0){
647 fprintf(stderr, "top block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y);
648 return -1;
649 } else if(status){
650 h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
651 }
652 }
653 }
654
655 if(!(h->left_samples_available&0x8000)){
656 for(i=0; i<4; i++){
657 int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
658 if(status<0){
659 fprintf(stderr, "left block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y);
660 return -1;
661 } else if(status){
662 h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
663 }
664 }
665 }
666
667 return 0;
668 } //FIXME cleanup like next
669
670 /**
671 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
672 */
673 static inline int check_intra_pred_mode(H264Context *h, int mode){
674 MpegEncContext * const s = &h->s;
675 static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
676 static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
677
678 if(!(h->top_samples_available&0x8000)){
679 mode= top[ mode ];
680 if(mode<0){
681 fprintf(stderr, "top block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);
682 return -1;
683 }
684 }
685
686 if(!(h->left_samples_available&0x8000)){
687 mode= left[ mode ];
688 if(mode<0){
689 fprintf(stderr, "left block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);
690 return -1;
691 }
692 }
693
694 return mode;
695 }
696
697 /**
698 * gets the predicted intra4x4 prediction mode.
699 */
700 static inline int pred_intra_mode(H264Context *h, int n){
701 const int index8= scan8[n];
702 const int left= h->intra4x4_pred_mode_cache[index8 - 1];
703 const int top = h->intra4x4_pred_mode_cache[index8 - 8];
704 const int min= FFMIN(left, top);
705
706 tprintf("mode:%d %d min:%d\n", left ,top, min);
707
708 if(min<0) return DC_PRED;
709 else return min;
710 }
711
712 static inline void write_back_non_zero_count(H264Context *h){
713 MpegEncContext * const s = &h->s;
714 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
715
716 h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[4+8*4];
717 h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[5+8*4];
718 h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[6+8*4];
719 h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];
720 h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[7+8*3];
721 h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[7+8*2];
722 h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[7+8*1];
723
724 h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[1+8*2];
725 h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
726 h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[2+8*1];
727
728 h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[1+8*5];
729 h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
730 h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[2+8*4];
731 }
732
733 /**
734 * gets the predicted number of non zero coefficients.
735 * @param n block index
736 */
737 static inline int pred_non_zero_count(H264Context *h, int n){
738 const int index8= scan8[n];
739 const int left= h->non_zero_count_cache[index8 - 1];
740 const int top = h->non_zero_count_cache[index8 - 8];
741 int i= left + top;
742
743 if(i<64) i= (i+1)>>1;
744
745 tprintf("pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
746
747 return i&31;
748 }
749
750 static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
751 const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];
752
753 if(topright_ref != PART_NOT_AVAILABLE){
754 *C= h->mv_cache[list][ i - 8 + part_width ];
755 return topright_ref;
756 }else{
757 tprintf("topright MV not available\n");
758
759 *C= h->mv_cache[list][ i - 8 - 1 ];
760 return h->ref_cache[list][ i - 8 - 1 ];
761 }
762 }
763
764 /**
765 * gets the predicted MV.
766 * @param n the block index
767 * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)
768 * @param mx the x component of the predicted motion vector
769 * @param my the y component of the predicted motion vector
770 */
771 static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
772 const int index8= scan8[n];
773 const int top_ref= h->ref_cache[list][ index8 - 8 ];
774 const int left_ref= h->ref_cache[list][ index8 - 1 ];
775 const int16_t * const A= h->mv_cache[list][ index8 - 1 ];
776 const int16_t * const B= h->mv_cache[list][ index8 - 8 ];
777 const int16_t * C;
778 int diagonal_ref, match_count;
779
780 assert(part_width==1 || part_width==2 || part_width==4);
781
782 /* mv_cache
783 B . . A T T T T
784 U . . L . . , .
785 U . . L . . . .
786 U . . L . . , .
787 . . . L . . . .
788 */
789
790 diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);
791 match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);
792
793 if(match_count > 1){ //most common
794 *mx= mid_pred(A[0], B[0], C[0]);
795 *my= mid_pred(A[1], B[1], C[1]);
796 }else if(match_count==1){
797 if(left_ref==ref){
798 *mx= A[0];
799 *my= A[1];
800 }else if(top_ref==ref){
801 *mx= B[0];
802 *my= B[1];
803 }else{
804 *mx= C[0];
805 *my= C[1];
806 }
807 }else{
808 if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
809 *mx= A[0];
810 *my= A[1];
811 }else{
812 *mx= mid_pred(A[0], B[0], C[0]);
813 *my= mid_pred(A[1], B[1], C[1]);
814 }
815 }
816
817 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);
818 }
819
820 /**
821 * gets the directionally predicted 16x8 MV.
822 * @param n the block index
823 * @param mx the x component of the predicted motion vector
824 * @param my the y component of the predicted motion vector
825 */
826 static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
827 if(n==0){
828 const int top_ref= h->ref_cache[list][ scan8[0] - 8 ];
829 const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
830
831 tprintf("pred_16x8: (%2d %2d %2d) at %2d %2d %d list %d", top_ref, B[0], B[1], h->s.mb_x, h->s.mb_y, n, list);
832
833 if(top_ref == ref){
834 *mx= B[0];
835 *my= B[1];
836 return;
837 }
838 }else{
839 const int left_ref= h->ref_cache[list][ scan8[8] - 1 ];
840 const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];
841
842 tprintf("pred_16x8: (%2d %2d %2d) at %2d %2d %d list %d", left_ref, A[0], A[1], h->s.mb_x, h->s.mb_y, n, list);
843
844 if(left_ref == ref){
845 *mx= A[0];
846 *my= A[1];
847 return;
848 }
849 }
850
851 //RARE
852 pred_motion(h, n, 4, list, ref, mx, my);
853 }
854
855 /**
856 * gets the directionally predicted 8x16 MV.
857 * @param n the block index
858 * @param mx the x component of the predicted motion vector
859 * @param my the y component of the predicted motion vector
860 */
861 static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
862 if(n==0){
863 const int left_ref= h->ref_cache[list][ scan8[0] - 1 ];
864 const int16_t * const A= h->mv_cache[list][ scan8[0] - 1 ];
865
866 tprintf("pred_8x16: (%2d %2d %2d) at %2d %2d %d list %d", left_ref, A[0], A[1], h->s.mb_x, h->s.mb_y, n, list);
867
868 if(left_ref == ref){
869 *mx= A[0];
870 *my= A[1];
871 return;
872 }
873 }else{
874 const int16_t * C;
875 int diagonal_ref;
876
877 diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);
878
879 tprintf("pred_8x16: (%2d %2d %2d) at %2d %2d %d list %d", diagonal_ref, C[0], C[1], h->s.mb_x, h->s.mb_y, n, list);
880
881 if(diagonal_ref == ref){
882 *mx= C[0];
883 *my= C[1];
884 return;
885 }
886 }
887
888 //RARE
889 pred_motion(h, n, 2, list, ref, mx, my);
890 }
891
892 static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
893 const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
894 const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];
895
896 tprintf("pred_pskip: (%d) (%d) at %2d %2d", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
897
898 if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
899 || (top_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0)
900 || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){
901
902 *mx = *my = 0;
903 return;
904 }
905
906 pred_motion(h, 0, 4, 0, 0, mx, my);
907
908 return;
909 }
910
911 static inline void write_back_motion(H264Context *h, int mb_type){
912 MpegEncContext * const s = &h->s;
913 const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
914 const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
915 int list;
916
917 for(list=0; list<2; list++){
918 int y;
919 if((!IS_8X8(mb_type)) && !USES_LIST(mb_type, list)){
920 if(1){ //FIXME skip or never read if mb_type doesnt use it
921 for(y=0; y<4; y++){
922 *(uint64_t*)s->current_picture.motion_val[list][b_xy + 0 + y*h->b_stride]=
923 *(uint64_t*)s->current_picture.motion_val[list][b_xy + 2 + y*h->b_stride]= 0;
924 }
925 for(y=0; y<2; y++){
926 *(uint16_t*)s->current_picture.motion_val[list][b8_xy + y*h->b8_stride]= (LIST_NOT_USED&0xFF)*0x0101;
927 }
928 }
929 continue; //FIXME direct mode ...
930 }
931
932 for(y=0; y<4; y++){
933 *(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];
934 *(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];
935 }
936 for(y=0; y<2; y++){
937 s->current_picture.ref_index[list][b8_xy + 0 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+0 + 16*y];
938 s->current_picture.ref_index[list][b8_xy + 1 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+2 + 16*y];
939 }
940 }
941 }
942
943 /**
944 * Decodes a network abstraction layer unit.
945 * @param consumed is the number of bytes used as input
946 * @param length is the length of the array
947 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp ttailing?
948 * @returns decoded bytes, might be src+1 if no escapes
949 */
950 static uint8_t *decode_nal(H264Context *h, uint8_t *src, int *dst_length, int *consumed, int length){
951 int i, si, di;
952 uint8_t *dst;
953
954 // src[0]&0x80; //forbidden bit
955 h->nal_ref_idc= src[0]>>5;
956 h->nal_unit_type= src[0]&0x1F;
957
958 src++; length--;
959 #if 0
960 for(i=0; i<length; i++)
961 printf("%2X ", src[i]);
962 #endif
963 for(i=0; i+1<length; i+=2){
964 if(src[i]) continue;
965 if(i>0 && src[i-1]==0) i--;
966 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
967 if(src[i+2]!=3){
968 /* startcode, so we must be past the end */
969 length=i;
970 }
971 break;
972 }
973 }
974
975 if(i>=length-1){ //no escaped 0
976 *dst_length= length;
977 *consumed= length+1; //+1 for the header
978 return src;
979 }
980
981 h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length);
982 dst= h->rbsp_buffer;
983
984 //printf("deoding esc\n");
985 si=di=0;
986 while(si<length){
987 //remove escapes (very rare 1:2^22)
988 if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
989 if(src[si+2]==3){ //escape
990 dst[di++]= 0;
991 dst[di++]= 0;
992 si+=3;
993 }else //next start code
994 break;
995 }
996
997 dst[di++]= src[si++];
998 }
999
1000 *dst_length= di;
1001 *consumed= si + 1;//+1 for the header
1002 //FIXME store exact number of bits in the getbitcontext (its needed for decoding)
1003 return dst;
1004 }
1005
1006 /**
1007 * @param src the data which should be escaped
1008 * @param dst the target buffer, dst+1 == src is allowed as a special case
1009 * @param length the length of the src data
1010 * @param dst_length the length of the dst array
1011 * @returns length of escaped data in bytes or -1 if an error occured
1012 */
1013 static int encode_nal(H264Context *h, uint8_t *dst, uint8_t *src, int length, int dst_length){
1014 int i, escape_count, si, di;
1015 uint8_t *temp;
1016
1017 assert(length>=0);
1018 assert(dst_length>0);
1019
1020 dst[0]= (h->nal_ref_idc<<5) + h->nal_unit_type;
1021
1022 if(length==0) return 1;
1023
1024 escape_count= 0;
1025 for(i=0; i<length; i+=2){
1026 if(src[i]) continue;
1027 if(i>0 && src[i-1]==0)
1028 i--;
1029 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1030 escape_count++;
1031 i+=2;
1032 }
1033 }
1034
1035 if(escape_count==0){
1036 if(dst+1 != src)
1037 memcpy(dst+1, src, length);
1038 return length + 1;
1039 }
1040
1041 if(length + escape_count + 1> dst_length)
1042 return -1;
1043
1044 //this should be damn rare (hopefully)
1045
1046 h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length + escape_count);
1047 temp= h->rbsp_buffer;
1048 //printf("encoding esc\n");
1049
1050 si= 0;
1051 di= 0;
1052 while(si < length){
1053 if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1054 temp[di++]= 0; si++;
1055 temp[di++]= 0; si++;
1056 temp[di++]= 3;
1057 temp[di++]= src[si++];
1058 }
1059 else
1060 temp[di++]= src[si++];
1061 }
1062 memcpy(dst+1, temp, length+escape_count);
1063
1064 assert(di == length+escape_count);
1065
1066 return di + 1;
1067 }
1068
1069 /**
1070 * write 1,10,100,1000,... for alignment, yes its exactly inverse to mpeg4
1071 */
1072 static void encode_rbsp_trailing(PutBitContext *pb){
1073 int length;
1074 put_bits(pb, 1, 1);
1075 length= (-get_bit_count(pb))&7;
1076 if(length) put_bits(pb, length, 0);
1077 }
1078
1079 /**
1080 * identifies the exact end of the bitstream
1081 * @return the length of the trailing, or 0 if damaged
1082 */
1083 static int decode_rbsp_trailing(uint8_t *src){
1084 int v= *src;
1085 int r;
1086
1087 tprintf("rbsp trailing %X\n", v);
1088
1089 for(r=1; r<9; r++){
1090 if(v&1) return r;
1091 v>>=1;
1092 }
1093 return 0;
1094 }
1095
1096 /**
1097 * idct tranforms the 16 dc values and dequantize them.
1098 * @param qp quantization parameter
1099 */
1100 static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp){
1101 const int qmul= dequant_coeff[qp][0];
1102 #define stride 16
1103 int i;
1104 int temp[16]; //FIXME check if this is a good idea
1105 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
1106 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1107
1108 //memset(block, 64, 2*256);
1109 //return;
1110 for(i=0; i<4; i++){
1111 const int offset= y_offset[i];
1112 const int z0= block[offset+stride*0] + block[offset+stride*4];
1113 const int z1= block[offset+stride*0] - block[offset+stride*4];
1114 const int z2= block[offset+stride*1] - block[offset+stride*5];
1115 const int z3= block[offset+stride*1] + block[offset+stride*5];
1116
1117 temp[4*i+0]= z0+z3;
1118 temp[4*i+1]= z1+z2;
1119 temp[4*i+2]= z1-z2;
1120 temp[4*i+3]= z0-z3;
1121 }
1122
1123 for(i=0; i<4; i++){
1124 const int offset= x_offset[i];
1125 const int z0= temp[4*0+i] + temp[4*2+i];
1126 const int z1= temp[4*0+i] - temp[4*2+i];
1127 const int z2= temp[4*1+i] - temp[4*3+i];
1128 const int z3= temp[4*1+i] + temp[4*3+i];
1129
1130 block[stride*0 +offset]= ((z0 + z3)*qmul + 2)>>2; //FIXME think about merging this into decode_resdual
1131 block[stride*2 +offset]= ((z1 + z2)*qmul + 2)>>2;
1132 block[stride*8 +offset]= ((z1 - z2)*qmul + 2)>>2;
1133 block[stride*10+offset]= ((z0 - z3)*qmul + 2)>>2;
1134 }
1135 }
1136
1137 /**
1138 * dct tranforms the 16 dc values.
1139 * @param qp quantization parameter ??? FIXME
1140 */
1141 static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
1142 // const int qmul= dequant_coeff[qp][0];
1143 int i;
1144 int temp[16]; //FIXME check if this is a good idea
1145 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
1146 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1147
1148 for(i=0; i<4; i++){
1149 const int offset= y_offset[i];
1150 const int z0= block[offset+stride*0] + block[offset+stride*4];
1151 const int z1= block[offset+stride*0] - block[offset+stride*4];
1152 const int z2= block[offset+stride*1] - block[offset+stride*5];
1153 const int z3= block[offset+stride*1] + block[offset+stride*5];
1154
1155 temp[4*i+0]= z0+z3;
1156 temp[4*i+1]= z1+z2;
1157 temp[4*i+2]= z1-z2;
1158 temp[4*i+3]= z0-z3;
1159 }
1160
1161 for(i=0; i<4; i++){
1162 const int offset= x_offset[i];
1163 const int z0= temp[4*0+i] + temp[4*2+i];
1164 const int z1= temp[4*0+i] - temp[4*2+i];
1165 const int z2= temp[4*1+i] - temp[4*3+i];
1166 const int z3= temp[4*1+i] + temp[4*3+i];
1167
1168 block[stride*0 +offset]= (z0 + z3)>>1;
1169 block[stride*2 +offset]= (z1 + z2)>>1;
1170 block[stride*8 +offset]= (z1 - z2)>>1;
1171 block[stride*10+offset]= (z0 - z3)>>1;
1172 }
1173 }
1174 #undef xStride
1175 #undef stride
1176
1177 static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp){
1178 const int qmul= dequant_coeff[qp][0];
1179 const int stride= 16*2;
1180 const int xStride= 16;
1181 int a,b,c,d,e;
1182
1183 a= block[stride*0 + xStride*0];
1184 b= block[stride*0 + xStride*1];
1185 c= block[stride*1 + xStride*0];
1186 d= block[stride*1 + xStride*1];
1187
1188 e= a-b;
1189 a= a+b;
1190 b= c-d;
1191 c= c+d;
1192
1193 block[stride*0 + xStride*0]= ((a+c)*qmul + 0)>>1;
1194 block[stride*0 + xStride*1]= ((e+b)*qmul + 0)>>1;
1195 block[stride*1 + xStride*0]= ((a-c)*qmul + 0)>>1;
1196 block[stride*1 + xStride*1]= ((e-b)*qmul + 0)>>1;
1197 }
1198
1199 static void chroma_dc_dct_c(DCTELEM *block){
1200 const int stride= 16*2;
1201 const int xStride= 16;
1202 int a,b,c,d,e;
1203
1204 a= block[stride*0 + xStride*0];
1205 b= block[stride*0 + xStride*1];
1206 c= block[stride*1 + xStride*0];
1207 d= block[stride*1 + xStride*1];
1208
1209 e= a-b;
1210 a= a+b;
1211 b= c-d;
1212 c= c+d;
1213
1214 block[stride*0 + xStride*0]= (a+c);
1215 block[stride*0 + xStride*1]= (e+b);
1216 block[stride*1 + xStride*0]= (a-c);
1217 block[stride*1 + xStride*1]= (e-b);
1218 }
1219
1220 /**
1221 * gets the chroma qp.
1222 */
1223 static inline int get_chroma_qp(H264Context *h, int qscale){
1224
1225 return chroma_qp[clip(qscale + h->pps.chroma_qp_index_offset, 0, 51)];
1226 }
1227
1228
1229 /**
1230 *
1231 */
1232 static void h264_add_idct_c(uint8_t *dst, DCTELEM *block, int stride){
1233 int i;
1234 uint8_t *cm = cropTbl + MAX_NEG_CROP;
1235
1236 block[0] += 32;
1237 #if 1
1238 for(i=0; i<4; i++){
1239 const int z0= block[i + 4*0] + block[i + 4*2];
1240 const int z1= block[i + 4*0] - block[i + 4*2];
1241 const int z2= (block[i + 4*1]>>1) - block[i + 4*3];
1242 const int z3= block[i + 4*1] + (block[i + 4*3]>>1);
1243
1244 block[i + 4*0]= z0 + z3;
1245 block[i + 4*1]= z1 + z2;
1246 block[i + 4*2]= z1 - z2;
1247 block[i + 4*3]= z0 - z3;
1248 }
1249
1250 for(i=0; i<4; i++){
1251 const int z0= block[0 + 4*i] + block[2 + 4*i];
1252 const int z1= block[0 + 4*i] - block[2 + 4*i];
1253 const int z2= (block[1 + 4*i]>>1) - block[3 + 4*i];
1254 const int z3= block[1 + 4*i] + (block[3 + 4*i]>>1);
1255
1256 dst[0 + i*stride]= cm[ dst[0 + i*stride] + ((z0 + z3) >> 6) ];
1257 dst[1 + i*stride]= cm[ dst[1 + i*stride] + ((z1 + z2) >> 6) ];
1258 dst[2 + i*stride]= cm[ dst[2 + i*stride] + ((z1 - z2) >> 6) ];
1259 dst[3 + i*stride]= cm[ dst[3 + i*stride] + ((z0 - z3) >> 6) ];
1260 }
1261 #else
1262 for(i=0; i<4; i++){
1263 const int z0= block[0 + 4*i] + block[2 + 4*i];
1264 const int z1= block[0 + 4*i] - block[2 + 4*i];
1265 const int z2= (block[1 + 4*i]>>1) - block[3 + 4*i];
1266 const int z3= block[1 + 4*i] + (block[3 + 4*i]>>1);
1267
1268 block[0 + 4*i]= z0 + z3;
1269 block[1 + 4*i]= z1 + z2;
1270 block[2 + 4*i]= z1 - z2;
1271 block[3 + 4*i]= z0 - z3;
1272 }
1273
1274 for(i=0; i<4; i++){
1275 const int z0= block[i + 4*0] + block[i + 4*2];
1276 const int z1= block[i + 4*0] - block[i + 4*2];
1277 const int z2= (block[i + 4*1]>>1) - block[i + 4*3];
1278 const int z3= block[i + 4*1] + (block[i + 4*3]>>1);
1279
1280 dst[i + 0*stride]= cm[ dst[i + 0*stride] + ((z0 + z3) >> 6) ];
1281 dst[i + 1*stride]= cm[ dst[i + 1*stride] + ((z1 + z2) >> 6) ];
1282 dst[i + 2*stride]= cm[ dst[i + 2*stride] + ((z1 - z2) >> 6) ];
1283 dst[i + 3*stride]= cm[ dst[i + 3*stride] + ((z0 - z3) >> 6) ];
1284 }
1285 #endif
1286 }
1287
1288 static void h264_diff_dct_c(DCTELEM *block, uint8_t *src1, uint8_t *src2, int stride){
1289 int i;
1290 //FIXME try int temp instead of block
1291
1292 for(i=0; i<4; i++){
1293 const int d0= src1[0 + i*stride] - src2[0 + i*stride];
1294 const int d1= src1[1 + i*stride] - src2[1 + i*stride];
1295 const int d2= src1[2 + i*stride] - src2[2 + i*stride];
1296 const int d3= src1[3 + i*stride] - src2[3 + i*stride];
1297 const int z0= d0 + d3;
1298 const int z3= d0 - d3;
1299 const int z1= d1 + d2;
1300 const int z2= d1 - d2;
1301
1302 block[0 + 4*i]= z0 + z1;
1303 block[1 + 4*i]= 2*z3 + z2;
1304 block[2 + 4*i]= z0 - z1;
1305 block[3 + 4*i]= z3 - 2*z2;
1306 }
1307
1308 for(i=0; i<4; i++){
1309 const int z0= block[0*4 + i] + block[3*4 + i];
1310 const int z3= block[0*4 + i] - block[3*4 + i];
1311 const int z1= block[1*4 + i] + block[2*4 + i];
1312 const int z2= block[1*4 + i] - block[2*4 + i];
1313
1314 block[0*4 + i]= z0 + z1;
1315 block[1*4 + i]= 2*z3 + z2;
1316 block[2*4 + i]= z0 - z1;
1317 block[3*4 + i]= z3 - 2*z2;
1318 }
1319 }
1320
1321 //FIXME need to check that this doesnt overflow signed 32 bit for low qp, iam not sure, its very close
1322 //FIXME check that gcc inlines this (and optimizes intra & seperate_dc stuff away)
1323 static inline int quantize_c(DCTELEM *block, uint8_t *scantable, int qscale, int intra, int seperate_dc){
1324 int i;
1325 const int * const quant_table= quant_coeff[qscale];
1326 const int bias= intra ? (1<<QUANT_SHIFT)/3 : (1<<QUANT_SHIFT)/6;
1327 const unsigned int threshold1= (1<<QUANT_SHIFT) - bias - 1;
1328 const unsigned int threshold2= (threshold1<<1);
1329 int last_non_zero;
1330
1331 if(seperate_dc){
1332 if(qscale<=18){
1333 //avoid overflows
1334 const int dc_bias= intra ? (1<<(QUANT_SHIFT-2))/3 : (1<<(QUANT_SHIFT-2))/6;
1335 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT-2)) - dc_bias - 1;
1336 const unsigned int dc_threshold2= (dc_threshold1<<1);
1337
1338 int level= block[0]*quant_coeff[qscale+18][0];
1339 if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1340 if(level>0){
1341 level= (dc_bias + level)>>(QUANT_SHIFT-2);
1342 block[0]= level;
1343 }else{
1344 level= (dc_bias - level)>>(QUANT_SHIFT-2);
1345 block[0]= -level;
1346 }
1347 // last_non_zero = i;
1348 }else{
1349 block[0]=0;
1350 }
1351 }else{
1352 const int dc_bias= intra ? (1<<(QUANT_SHIFT+1))/3 : (1<<(QUANT_SHIFT+1))/6;
1353 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT+1)) - dc_bias - 1;
1354 const unsigned int dc_threshold2= (dc_threshold1<<1);
1355
1356 int level= block[0]*quant_table[0];
1357 if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1358 if(level>0){
1359 level= (dc_bias + level)>>(QUANT_SHIFT+1);
1360 block[0]= level;
1361 }else{
1362 level= (dc_bias - level)>>(QUANT_SHIFT+1);
1363 block[0]= -level;
1364 }
1365 // last_non_zero = i;
1366 }else{
1367 block[0]=0;
1368 }
1369 }
1370 last_non_zero= 0;
1371 i=1;
1372 }else{
1373 last_non_zero= -1;
1374 i=0;
1375 }
1376
1377 for(; i<16; i++){
1378 const int j= scantable[i];
1379 int level= block[j]*quant_table[j];
1380
1381 // if( bias+level >= (1<<(QMAT_SHIFT - 3))
1382 // || bias-level >= (1<<(QMAT_SHIFT - 3))){
1383 if(((unsigned)(level+threshold1))>threshold2){
1384 if(level>0){
1385 level= (bias + level)>>QUANT_SHIFT;
1386 block[j]= level;
1387 }else{
1388 level= (bias - level)>>QUANT_SHIFT;
1389 block[j]= -level;
1390 }
1391 last_non_zero = i;
1392 }else{
1393 block[j]=0;
1394 }
1395 }
1396
1397 return last_non_zero;
1398 }
1399
1400 static void pred4x4_vertical_c(uint8_t *src, uint8_t *topright, int stride){
1401 const uint32_t a= ((uint32_t*)(src-stride))[0];
1402 ((uint32_t*)(src+0*stride))[0]= a;
1403 ((uint32_t*)(src+1*stride))[0]= a;
1404 ((uint32_t*)(src+2*stride))[0]= a;
1405 ((uint32_t*)(src+3*stride))[0]= a;
1406 }
1407
1408 static void pred4x4_horizontal_c(uint8_t *src, uint8_t *topright, int stride){
1409 ((uint32_t*)(src+0*stride))[0]= src[-1+0*stride]*0x01010101;
1410 ((uint32_t*)(src+1*stride))[0]= src[-1+1*stride]*0x01010101;
1411 ((uint32_t*)(src+2*stride))[0]= src[-1+2*stride]*0x01010101;
1412 ((uint32_t*)(src+3*stride))[0]= src[-1+3*stride]*0x01010101;
1413 }
1414
1415 static void pred4x4_dc_c(uint8_t *src, uint8_t *topright, int stride){
1416 const int dc= ( src[-stride] + src[1-stride] + src[2-stride] + src[3-stride]
1417 + src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 4) >>3;
1418
1419 ((uint32_t*)(src+0*stride))[0]=
1420 ((uint32_t*)(src+1*stride))[0]=
1421 ((uint32_t*)(src+2*stride))[0]=
1422 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1423 }
1424
1425 static void pred4x4_left_dc_c(uint8_t *src, uint8_t *topright, int stride){
1426 const int dc= ( src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 2) >>2;
1427
1428 ((uint32_t*)(src+0*stride))[0]=
1429 ((uint32_t*)(src+1*stride))[0]=
1430 ((uint32_t*)(src+2*stride))[0]=
1431 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1432 }
1433
1434 static void pred4x4_top_dc_c(uint8_t *src, uint8_t *topright, int stride){
1435 const int dc= ( src[-stride] + src[1-stride] + src[2-stride] + src[3-stride] + 2) >>2;
1436
1437 ((uint32_t*)(src+0*stride))[0]=
1438 ((uint32_t*)(src+1*stride))[0]=
1439 ((uint32_t*)(src+2*stride))[0]=
1440 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1441 }
1442
1443 static void pred4x4_128_dc_c(uint8_t *src, uint8_t *topright, int stride){
1444 ((uint32_t*)(src+0*stride))[0]=
1445 ((uint32_t*)(src+1*stride))[0]=
1446 ((uint32_t*)(src+2*stride))[0]=
1447 ((uint32_t*)(src+3*stride))[0]= 128U*0x01010101U;
1448 }
1449
1450
1451 #define LOAD_TOP_RIGHT_EDGE\
1452 const int t4= topright[0];\
1453 const int t5= topright[1];\
1454 const int t6= topright[2];\
1455 const int t7= topright[3];\
1456
1457 #define LOAD_LEFT_EDGE\
1458 const int l0= src[-1+0*stride];\
1459 const int l1= src[-1+1*stride];\
1460 const int l2= src[-1+2*stride];\
1461 const int l3= src[-1+3*stride];\
1462
1463 #define LOAD_TOP_EDGE\
1464 const int t0= src[ 0-1*stride];\
1465 const int t1= src[ 1-1*stride];\
1466 const int t2= src[ 2-1*stride];\
1467 const int t3= src[ 3-1*stride];\
1468
1469 static void pred4x4_down_right_c(uint8_t *src, uint8_t *topright, int stride){
1470 const int lt= src[-1-1*stride];
1471 LOAD_TOP_EDGE
1472 LOAD_LEFT_EDGE
1473
1474 src[0+3*stride]=(l3 + 2*l2 + l1 + 2)>>2;
1475 src[0+2*stride]=
1476 src[1+3*stride]=(l2 + 2*l1 + l0 + 2)>>2;
1477 src[0+1*stride]=
1478 src[1+2*stride]=
1479 src[2+3*stride]=(l1 + 2*l0 + lt + 2)>>2;
1480 src[0+0*stride]=
1481 src[1+1*stride]=
1482 src[2+2*stride]=
1483 src[3+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1484 src[1+0*stride]=
1485 src[2+1*stride]=
1486 src[3+2*stride]=(lt + 2*t0 + t1 + 2)>>2;
1487 src[2+0*stride]=
1488 src[3+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1489 src[3+0*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1490 }
1491
1492 static void pred4x4_down_left_c(uint8_t *src, uint8_t *topright, int stride){
1493 LOAD_TOP_EDGE
1494 LOAD_TOP_RIGHT_EDGE
1495 // LOAD_LEFT_EDGE
1496
1497 src[0+0*stride]=(t0 + t2 + 2*t1 + 2)>>2;
1498 src[1+0*stride]=
1499 src[0+1*stride]=(t1 + t3 + 2*t2 + 2)>>2;
1500 src[2+0*stride]=
1501 src[1+1*stride]=
1502 src[0+2*stride]=(t2 + t4 + 2*t3 + 2)>>2;
1503 src[3+0*stride]=
1504 src[2+1*stride]=
1505 src[1+2*stride]=
1506 src[0+3*stride]=(t3 + t5 + 2*t4 + 2)>>2;
1507 src[3+1*stride]=
1508 src[2+2*stride]=
1509 src[1+3*stride]=(t4 + t6 + 2*t5 + 2)>>2;
1510 src[3+2*stride]=
1511 src[2+3*stride]=(t5 + t7 + 2*t6 + 2)>>2;
1512 src[3+3*stride]=(t6 + 3*t7 + 2)>>2;
1513 }
1514
1515 static void pred4x4_vertical_right_c(uint8_t *src, uint8_t *topright, int stride){
1516 const int lt= src[-1-1*stride];
1517 LOAD_TOP_EDGE
1518 LOAD_LEFT_EDGE
1519 const __attribute__((unused)) int unu= l3;
1520
1521 src[0+0*stride]=
1522 src[1+2*stride]=(lt + t0 + 1)>>1;
1523 src[1+0*stride]=
1524 src[2+2*stride]=(t0 + t1 + 1)>>1;
1525 src[2+0*stride]=
1526 src[3+2*stride]=(t1 + t2 + 1)>>1;
1527 src[3+0*stride]=(t2 + t3 + 1)>>1;
1528 src[0+1*stride]=
1529 src[1+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1530 src[1+1*stride]=
1531 src[2+3*stride]=(lt + 2*t0 + t1 + 2)>>2;
1532 src[2+1*stride]=
1533 src[3+3*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1534 src[3+1*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1535 src[0+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1536 src[0+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1537 }
1538
1539 static void pred4x4_vertical_left_c(uint8_t *src, uint8_t *topright, int stride){
1540 LOAD_TOP_EDGE
1541 LOAD_TOP_RIGHT_EDGE
1542 const __attribute__((unused)) int unu= t7;
1543
1544 src[0+0*stride]=(t0 + t1 + 1)>>1;
1545 src[1+0*stride]=
1546 src[0+2*stride]=(t1 + t2 + 1)>>1;
1547 src[2+0*stride]=
1548 src[1+2*stride]=(t2 + t3 + 1)>>1;
1549 src[3+0*stride]=
1550 src[2+2*stride]=(t3 + t4+ 1)>>1;
1551 src[3+2*stride]=(t4 + t5+ 1)>>1;
1552 src[0+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1553 src[1+1*stride]=
1554 src[0+3*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1555 src[2+1*stride]=
1556 src[1+3*stride]=(t2 + 2*t3 + t4 + 2)>>2;
1557 src[3+1*stride]=
1558 src[2+3*stride]=(t3 + 2*t4 + t5 + 2)>>2;
1559 src[3+3*stride]=(t4 + 2*t5 + t6 + 2)>>2;
1560 }
1561
1562 static void pred4x4_horizontal_up_c(uint8_t *src, uint8_t *topright, int stride){
1563 LOAD_LEFT_EDGE
1564
1565 src[0+0*stride]=(l0 + l1 + 1)>>1;
1566 src[1+0*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1567 src[2+0*stride]=
1568 src[0+1*stride]=(l1 + l2 + 1)>>1;
1569 src[3+0*stride]=
1570 src[1+1*stride]=(l1 + 2*l2 + l3 + 2)>>2;
1571 src[2+1*stride]=
1572 src[0+2*stride]=(l2 + l3 + 1)>>1;
1573 src[3+1*stride]=
1574 src[1+2*stride]=(l2 + 2*l3 + l3 + 2)>>2;
1575 src[3+2*stride]=
1576 src[1+3*stride]=
1577 src[0+3*stride]=
1578 src[2+2*stride]=
1579 src[2+3*stride]=
1580 src[3+3*stride]=l3;
1581 }
1582
1583 static void pred4x4_horizontal_down_c(uint8_t *src, uint8_t *topright, int stride){
1584 const int lt= src[-1-1*stride];
1585 LOAD_TOP_EDGE
1586 LOAD_LEFT_EDGE
1587 const __attribute__((unused)) int unu= t3;
1588
1589 src[0+0*stride]=
1590 src[2+1*stride]=(lt + l0 + 1)>>1;
1591 src[1+0*stride]=
1592 src[3+1*stride]=(l0 + 2*lt + t0 + 2)>>2;
1593 src[2+0*stride]=(lt + 2*t0 + t1 + 2)>>2;
1594 src[3+0*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1595 src[0+1*stride]=
1596 src[2+2*stride]=(l0 + l1 + 1)>>1;
1597 src[1+1*stride]=
1598 src[3+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1599 src[0+2*stride]=
1600 src[2+3*stride]=(l1 + l2+ 1)>>1;
1601 src[1+2*stride]=
1602 src[3+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1603 src[0+3*stride]=(l2 + l3 + 1)>>1;
1604 src[1+3*stride]=(l1 + 2*l2 + l3 + 2)>>2;
1605 }
1606
1607 static void pred16x16_vertical_c(uint8_t *src, int stride){
1608 int i;
1609 const uint32_t a= ((uint32_t*)(src-stride))[0];
1610 const uint32_t b= ((uint32_t*)(src-stride))[1];
1611 const uint32_t c= ((uint32_t*)(src-stride))[2];
1612 const uint32_t d= ((uint32_t*)(src-stride))[3];
1613
1614 for(i=0; i<16; i++){
1615 ((uint32_t*)(src+i*stride))[0]= a;
1616 ((uint32_t*)(src+i*stride))[1]= b;
1617 ((uint32_t*)(src+i*stride))[2]= c;
1618 ((uint32_t*)(src+i*stride))[3]= d;
1619 }
1620 }
1621
1622 static void pred16x16_horizontal_c(uint8_t *src, int stride){
1623 int i;
1624
1625 for(i=0; i<16; i++){
1626 ((uint32_t*)(src+i*stride))[0]=
1627 ((uint32_t*)(src+i*stride))[1]=
1628 ((uint32_t*)(src+i*stride))[2]=
1629 ((uint32_t*)(src+i*stride))[3]= src[-1+i*stride]*0x01010101;
1630 }
1631 }
1632
1633 static void pred16x16_dc_c(uint8_t *src, int stride){
1634 int i, dc=0;
1635
1636 for(i=0;i<16; i++){
1637 dc+= src[-1+i*stride];
1638 }
1639
1640 for(i=0;i<16; i++){
1641 dc+= src[i-stride];
1642 }
1643
1644 dc= 0x01010101*((dc + 16)>>5);
1645
1646 for(i=0; i<16; i++){
1647 ((uint32_t*)(src+i*stride))[0]=
1648 ((uint32_t*)(src+i*stride))[1]=
1649 ((uint32_t*)(src+i*stride))[2]=
1650 ((uint32_t*)(src+i*stride))[3]= dc;
1651 }
1652 }
1653
1654 static void pred16x16_left_dc_c(uint8_t *src, int stride){
1655 int i, dc=0;
1656
1657 for(i=0;i<16; i++){
1658 dc+= src[-1+i*stride];
1659 }
1660
1661 dc= 0x01010101*((dc + 8)>>4);
1662
1663 for(i=0; i<16; i++){
1664 ((uint32_t*)(src+i*stride))[0]=
1665 ((uint32_t*)(src+i*stride))[1]=
1666 ((uint32_t*)(src+i*stride))[2]=
1667 ((uint32_t*)(src+i*stride))[3]= dc;
1668 }
1669 }
1670
1671 static void pred16x16_top_dc_c(uint8_t *src, int stride){
1672 int i, dc=0;
1673
1674 for(i=0;i<16; i++){
1675 dc+= src[i-stride];
1676 }
1677 dc= 0x01010101*((dc + 8)>>4);
1678
1679 for(i=0; i<16; i++){
1680 ((uint32_t*)(src+i*stride))[0]=
1681 ((uint32_t*)(src+i*stride))[1]=
1682 ((uint32_t*)(src+i*stride))[2]=
1683 ((uint32_t*)(src+i*stride))[3]= dc;
1684 }
1685 }
1686
1687 static void pred16x16_128_dc_c(uint8_t *src, int stride){
1688 int i;
1689
1690 for(i=0; i<16; i++){
1691 ((uint32_t*)(src+i*stride))[0]=
1692 ((uint32_t*)(src+i*stride))[1]=
1693 ((uint32_t*)(src+i*stride))[2]=
1694 ((uint32_t*)(src+i*stride))[3]= 0x01010101U*128U;
1695 }
1696 }
1697
1698 static inline void pred16x16_plane_compat_c(uint8_t *src, int stride, const int svq3){
1699 int i, j, k;
1700 int a;
1701 uint8_t *cm = cropTbl + MAX_NEG_CROP;
1702 const uint8_t * const src0 = src+7-stride;
1703 const uint8_t *src1 = src+8*stride-1;
1704 const uint8_t *src2 = src1-2*stride; // == src+6*stride-1;
1705 int H = src0[1] - src0[-1];
1706 int V = src1[0] - src2[ 0];
1707 for(k=2; k<=8; ++k) {
1708 src1 += stride; src2 -= stride;
1709 H += k*(src0[k] - src0[-k]);
1710 V += k*(src1[0] - src2[ 0]);
1711 }
1712 if(svq3){
1713 H = ( 5*(H/4) ) / 16;
1714 V = ( 5*(V/4) ) / 16;
1715 }else{
1716 H = ( 5*H+32 ) >> 6;
1717 V = ( 5*V+32 ) >> 6;
1718 }
1719
1720 a = 16*(src1[0] + src2[16] + 1) - 7*(V+H);
1721 for(j=16; j>0; --j) {
1722 int b = a;
1723 a += V;
1724 for(i=-16; i<0; i+=4) {
1725 src[16+i] = cm[ (b ) >> 5 ];
1726 src[17+i] = cm[ (b+ H) >> 5 ];
1727 src[18+i] = cm[ (b+2*H) >> 5 ];
1728 src[19+i] = cm[ (b+3*H) >> 5 ];
1729 b += 4*H;
1730 }
1731 src += stride;
1732 }
1733 }
1734
1735 static void pred16x16_plane_c(uint8_t *src, int stride){
1736 pred16x16_plane_compat_c(src, stride, 0);
1737 }
1738
1739 static void pred8x8_vertical_c(uint8_t *src, int stride){
1740 int i;
1741 const uint32_t a= ((uint32_t*)(src-stride))[0];
1742 const uint32_t b= ((uint32_t*)(src-stride))[1];
1743
1744 for(i=0; i<8; i++){
1745 ((uint32_t*)(src+i*stride))[0]= a;
1746 ((uint32_t*)(src+i*stride))[1]= b;
1747 }
1748 }
1749
1750 static void pred8x8_horizontal_c(uint8_t *src, int stride){
1751 int i;
1752
1753 for(i=0; i<8; i++){
1754 ((uint32_t*)(src+i*stride))[0]=
1755 ((uint32_t*)(src+i*stride))[1]= src[-1+i*stride]*0x01010101;
1756 }
1757 }
1758
1759 static void pred8x8_128_dc_c(uint8_t *src, int stride){
1760 int i;
1761
1762 for(i=0; i<4; i++){
1763 ((uint32_t*)(src+i*stride))[0]=
1764 ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
1765 }
1766 for(i=4; i<8; i++){
1767 ((uint32_t*)(src+i*stride))[0]=
1768 ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
1769 }
1770 }
1771
1772 static void pred8x8_left_dc_c(uint8_t *src, int stride){
1773 int i;
1774 int dc0, dc2;
1775
1776 dc0=dc2=0;
1777 for(i=0;i<4; i++){
1778 dc0+= src[-1+i*stride];
1779 dc2+= src[-1+(i+4)*stride];
1780 }
1781 dc0= 0x01010101*((dc0 + 2)>>2);
1782 dc2= 0x01010101*((dc2 + 2)>>2);
1783
1784 for(i=0; i<4; i++){
1785 ((uint32_t*)(src+i*stride))[0]=
1786 ((uint32_t*)(src+i*stride))[1]= dc0;
1787 }
1788 for(i=4; i<8; i++){
1789 ((uint32_t*)(src+i*stride))[0]=
1790 ((uint32_t*)(src+i*stride))[1]= dc2;
1791 }
1792 }
1793
1794 static void pred8x8_top_dc_c(uint8_t *src, int stride){
1795 int i;
1796 int dc0, dc1;
1797
1798 dc0=dc1=0;
1799 for(i=0;i<4; i++){
1800 dc0+= src[i-stride];
1801 dc1+= src[4+i-stride];
1802 }
1803 dc0= 0x01010101*((dc0 + 2)>>2);
1804 dc1= 0x01010101*((dc1 + 2)>>2);
1805
1806 for(i=0; i<4; i++){
1807 ((uint32_t*)(src+i*stride))[0]= dc0;
1808 ((uint32_t*)(src+i*stride))[1]= dc1;
1809 }
1810 for(i=4; i<8; i++){
1811 ((uint32_t*)(src+i*stride))[0]= dc0;
1812 ((uint32_t*)(src+i*stride))[1]= dc1;
1813 }
1814 }
1815
1816
1817 static void pred8x8_dc_c(uint8_t *src, int stride){
1818 int i;
1819 int dc0, dc1, dc2, dc3;
1820
1821 dc0=dc1=dc2=0;
1822 for(i=0;i<4; i++){
1823 dc0+= src[-1+i*stride] + src[i-stride];
1824 dc1+= src[4+i-stride];
1825 dc2+= src[-1+(i+4)*stride];
1826 }
1827 dc3= 0x01010101*((dc1 + dc2 + 4)>>3);
1828 dc0= 0x01010101*((dc0 + 4)>>3);
1829 dc1= 0x01010101*((dc1 + 2)>>2);
1830 dc2= 0x01010101*((dc2 + 2)>>2);
1831
1832 for(i=0; i<4; i++){
1833 ((uint32_t*)(src+i*stride))[0]= dc0;
1834 ((uint32_t*)(src+i*stride))[1]= dc1;
1835 }
1836 for(i=4; i<8; i++){
1837 ((uint32_t*)(src+i*stride))[0]= dc2;
1838 ((uint32_t*)(src+i*stride))[1]= dc3;
1839 }
1840 }
1841
1842 static void pred8x8_plane_c(uint8_t *src, int stride){
1843 int j, k;
1844 int a;
1845 uint8_t *cm = cropTbl + MAX_NEG_CROP;
1846 const uint8_t * const src0 = src+3-stride;
1847 const uint8_t *src1 = src+4*stride-1;
1848 const uint8_t *src2 = src1-2*stride; // == src+2*stride-1;
1849 int H = src0[1] - src0[-1];
1850 int V = src1[0] - src2[ 0];
1851 for(k=2; k<=4; ++k) {
1852 src1 += stride; src2 -= stride;
1853 H += k*(src0[k] - src0[-k]);
1854 V += k*(src1[0] - src2[ 0]);
1855 }
1856 H = ( 17*H+16 ) >> 5;
1857 V = ( 17*V+16 ) >> 5;
1858
1859 a = 16*(src1[0] + src2[8]+1) - 3*(V+H);
1860 for(j=8; j>0; --j) {
1861 int b = a;
1862 a += V;
1863 src[0] = cm[ (b ) >> 5 ];
1864 src[1] = cm[ (b+ H) >> 5 ];
1865 src[2] = cm[ (b+2*H) >> 5 ];
1866 src[3] = cm[ (b+3*H) >> 5 ];
1867 src[4] = cm[ (b+4*H) >> 5 ];
1868 src[5] = cm[ (b+5*H) >> 5 ];
1869 src[6] = cm[ (b+6*H) >> 5 ];
1870 src[7] = cm[ (b+7*H) >> 5 ];
1871 src += stride;
1872 }
1873 }
1874
1875 static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
1876 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1877 int src_x_offset, int src_y_offset,
1878 qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
1879 MpegEncContext * const s = &h->s;
1880 const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
1881 const int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
1882 const int luma_xy= (mx&3) + ((my&3)<<2);
1883 uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*s->linesize;
1884 uint8_t * src_cb= pic->data[1] + (mx>>3) + (my>>3)*s->uvlinesize;
1885 uint8_t * src_cr= pic->data[2] + (mx>>3) + (my>>3)*s->uvlinesize;
1886 int extra_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16; //FIXME increase edge?, IMHO not worth it
1887 int extra_height= extra_width;
1888 int emu=0;
1889 const int full_mx= mx>>2;
1890 const int full_my= my>>2;
1891
1892 assert(pic->data[0]);
1893
1894 if(mx&7) extra_width -= 3;
1895 if(my&7) extra_height -= 3;
1896
1897 if( full_mx < 0-extra_width
1898 || full_my < 0-extra_height
1899 || full_mx + 16/*FIXME*/ > s->width + extra_width
1900 || full_my + 16/*FIXME*/ > s->height + extra_height){
1901 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);
1902 src_y= s->edge_emu_buffer + 2 + 2*s->linesize;
1903 emu=1;
1904 }
1905
1906 qpix_op[luma_xy](dest_y, src_y, s->linesize); //FIXME try variable height perhaps?
1907 if(!square){
1908 qpix_op[luma_xy](dest_y + delta, src_y + delta, s->linesize);
1909 }
1910
1911 if(s->flags&CODEC_FLAG_GRAY) return;
1912
1913 if(emu){
1914 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);
1915 src_cb= s->edge_emu_buffer;
1916 }
1917 chroma_op(dest_cb, src_cb, s->uvlinesize, chroma_height, mx&7, my&7);
1918
1919 if(emu){
1920 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);
1921 src_cr= s->edge_emu_buffer;
1922 }
1923 chroma_op(dest_cr, src_cr, s->uvlinesize, chroma_height, mx&7, my&7);
1924 }
1925
1926 static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
1927 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1928 int x_offset, int y_offset,
1929 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
1930 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
1931 int list0, int list1){
1932 MpegEncContext * const s = &h->s;
1933 qpel_mc_func *qpix_op= qpix_put;
1934 h264_chroma_mc_func chroma_op= chroma_put;
1935
1936 dest_y += 2*x_offset + 2*y_offset*s-> linesize;
1937 dest_cb += x_offset + y_offset*s->uvlinesize;
1938 dest_cr += x_offset + y_offset*s->uvlinesize;
1939 x_offset += 8*s->mb_x;
1940 y_offset += 8*s->mb_y;
1941
1942 if(list0){
1943 Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
1944 mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
1945 dest_y, dest_cb, dest_cr, x_offset, y_offset,
1946 qpix_op, chroma_op);
1947
1948 qpix_op= qpix_avg;
1949 chroma_op= chroma_avg;
1950 }
1951
1952 if(list1){
1953 Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
1954 mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
1955 dest_y, dest_cb, dest_cr, x_offset, y_offset,
1956 qpix_op, chroma_op);
1957 }
1958 }
1959
1960 static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1961 qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
1962 qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg)){
1963 MpegEncContext * const s = &h->s;
1964 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
1965 const int mb_type= s->current_picture.mb_type[mb_xy];
1966
1967 assert(IS_INTER(mb_type));
1968
1969 if(IS_16X16(mb_type)){
1970 mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
1971 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
1972 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1973 }else if(IS_16X8(mb_type)){
1974 mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
1975 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
1976 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1977 mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
1978 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
1979 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
1980 }else if(IS_8X16(mb_type)){
1981 mc_part(h, 0, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 0, 0,
1982 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
1983 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1984 mc_part(h, 4, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 4, 0,
1985 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
1986 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
1987 }else{
1988 int i;
1989
1990 assert(IS_8X8(mb_type));
1991
1992 for(i=0; i<4; i++){
1993 const int sub_mb_type= h->sub_mb_type[i];
1994 const int n= 4*i;
1995 int x_offset= (i&1)<<2;
1996 int y_offset= (i&2)<<1;
1997
1998 if(IS_SUB_8X8(sub_mb_type)){
1999 mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2000 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2001 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2002 }else if(IS_SUB_8X4(sub_mb_type)){
2003 mc_part(h, n , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2004 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2005 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2006 mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
2007 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2008 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2009 }else if(IS_SUB_4X8(sub_mb_type)){
2010 mc_part(h, n , 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2011 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2012 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2013 mc_part(h, n+1, 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
2014 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2015 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2016 }else{
2017 int j;
2018 assert(IS_SUB_4X4(sub_mb_type));
2019 for(j=0; j<4; j++){
2020 int sub_x_offset= x_offset + 2*(j&1);
2021 int sub_y_offset= y_offset + (j&2);
2022 mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
2023 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2024 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2025 }
2026 }
2027 }
2028 }
2029 }
2030
2031 static void decode_init_vlc(H264Context *h){
2032 static int done = 0;
2033
2034 if (!done) {
2035 int i;
2036 done = 1;
2037
2038 init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5,
2039 &chroma_dc_coeff_token_len [0], 1, 1,
2040 &chroma_dc_coeff_token_bits[0], 1, 1);
2041
2042 for(i=0; i<4; i++){
2043 init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17,
2044 &coeff_token_len [i][0], 1, 1,
2045 &coeff_token_bits[i][0], 1, 1);
2046 }
2047
2048 for(i=0; i<3; i++){
2049 init_vlc(&chroma_dc_total_zeros_vlc[i], CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
2050 &chroma_dc_total_zeros_len [i][0], 1, 1,
2051 &chroma_dc_total_zeros_bits[i][0], 1, 1);
2052 }
2053 for(i=0; i<15; i++){
2054 init_vlc(&total_zeros_vlc[i], TOTAL_ZEROS_VLC_BITS, 16,
2055 &total_zeros_len [i][0], 1, 1,
2056 &total_zeros_bits[i][0], 1, 1);
2057 }
2058
2059 for(i=0; i<6; i++){
2060 init_vlc(&run_vlc[i], RUN_VLC_BITS, 7,
2061 &run_len [i][0], 1, 1,
2062 &run_bits[i][0], 1, 1);
2063 }
2064 init_vlc(&run7_vlc, RUN7_VLC_BITS, 16,
2065 &run_len [6][0], 1, 1,
2066 &run_bits[6][0], 1, 1);
2067 }
2068 }
2069
2070 /**
2071 * Sets the intra prediction function pointers.
2072 */
2073 static void init_pred_ptrs(H264Context *h){
2074 // MpegEncContext * const s = &h->s;
2075
2076 h->pred4x4[VERT_PRED ]= pred4x4_vertical_c;
2077 h->pred4x4[HOR_PRED ]= pred4x4_horizontal_c;
2078 h->pred4x4[DC_PRED ]= pred4x4_dc_c;
2079 h->pred4x4[DIAG_DOWN_LEFT_PRED ]= pred4x4_down_left_c;
2080 h->pred4x4[DIAG_DOWN_RIGHT_PRED]= pred4x4_down_right_c;
2081 h->pred4x4[VERT_RIGHT_PRED ]= pred4x4_vertical_right_c;
2082 h->pred4x4[HOR_DOWN_PRED ]= pred4x4_horizontal_down_c;
2083 h->pred4x4[VERT_LEFT_PRED ]= pred4x4_vertical_left_c;
2084 h->pred4x4[HOR_UP_PRED ]= pred4x4_horizontal_up_c;
2085 h->pred4x4[LEFT_DC_PRED ]= pred4x4_left_dc_c;
2086 h->pred4x4[TOP_DC_PRED ]= pred4x4_top_dc_c;
2087 h->pred4x4[DC_128_PRED ]= pred4x4_128_dc_c;
2088
2089 h->pred8x8[DC_PRED8x8 ]= pred8x8_dc_c;
2090 h->pred8x8[VERT_PRED8x8 ]= pred8x8_vertical_c;
2091 h->pred8x8[HOR_PRED8x8 ]= pred8x8_horizontal_c;
2092 h->pred8x8[PLANE_PRED8x8 ]= pred8x8_plane_c;
2093 h->pred8x8[LEFT_DC_PRED8x8]= pred8x8_left_dc_c;
2094 h->pred8x8[TOP_DC_PRED8x8 ]= pred8x8_top_dc_c;
2095 h->pred8x8[DC_128_PRED8x8 ]= pred8x8_128_dc_c;
2096
2097 h->pred16x16[DC_PRED8x8 ]= pred16x16_dc_c;
2098 h->pred16x16[VERT_PRED8x8 ]= pred16x16_vertical_c;
2099 h->pred16x16[HOR_PRED8x8 ]= pred16x16_horizontal_c;
2100 h->pred16x16[PLANE_PRED8x8 ]= pred16x16_plane_c;
2101 h->pred16x16[LEFT_DC_PRED8x8]= pred16x16_left_dc_c;
2102 h->pred16x16[TOP_DC_PRED8x8 ]= pred16x16_top_dc_c;
2103 h->pred16x16[DC_128_PRED8x8 ]= pred16x16_128_dc_c;
2104 }
2105
2106 //FIXME factorize
2107 #define CHECKED_ALLOCZ(p, size)\
2108 {\
2109 p= av_mallocz(size);\
2110 if(p==NULL){\
2111 perror("malloc");\
2112 goto fail;\
2113 }\
2114 }
2115
2116 static void free_tables(H264Context *h){
2117 av_freep(&h->intra4x4_pred_mode);
2118 av_freep(&h->non_zero_count);
2119 av_freep(&h->slice_table_base);
2120 h->slice_table= NULL;
2121
2122 av_freep(&h->mb2b_xy);
2123 av_freep(&h->mb2b8_xy);
2124 }
2125
2126 /**
2127 * allocates tables.
2128 * needs widzh/height
2129 */
2130 static int alloc_tables(H264Context *h){
2131 MpegEncContext * const s = &h->s;
2132 const int big_mb_num= s->mb_stride * (s->mb_height+1);
2133 int x,y;
2134
2135 CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8 * sizeof(uint8_t))
2136 CHECKED_ALLOCZ(h->non_zero_count , big_mb_num * 16 * sizeof(uint8_t))
2137 CHECKED_ALLOCZ(h->slice_table_base , big_mb_num * sizeof(uint8_t))
2138
2139 memset(h->slice_table_base, -1, big_mb_num * sizeof(uint8_t));
2140 h->slice_table= h->slice_table_base + s->mb_stride + 1;
2141
2142 CHECKED_ALLOCZ(h->mb2b_xy , big_mb_num * sizeof(uint16_t));
2143 CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint16_t));
2144 for(y=0; y<s->mb_height; y++){
2145 for(x=0; x<s->mb_width; x++){
2146 const int mb_xy= x + y*s->mb_stride;
2147 const int b_xy = 4*x + 4*y*h->b_stride;
2148 const int b8_xy= 2*x + 2*y*h->b8_stride;
2149
2150 h->mb2b_xy [mb_xy]= b_xy;
2151 h->mb2b8_xy[mb_xy]= b8_xy;
2152 }
2153 }
2154
2155 return 0;
2156 fail:
2157 free_tables(h);
2158 return -1;
2159 }
2160
2161 static void common_init(H264Context *h){
2162 MpegEncContext * const s = &h->s;
2163
2164 s->width = s->avctx->width;
2165 s->height = s->avctx->height;
2166 s->codec_id= s->avctx->codec->id;
2167
2168 init_pred_ptrs(h);
2169
2170 s->decode=1; //FIXME
2171 }
2172
2173 static int decode_init(AVCodecContext *avctx){
2174 H264Context *h= avctx->priv_data;
2175 MpegEncContext * const s = &h->s;
2176
2177 s->avctx = avctx;
2178 common_init(h);
2179
2180 s->out_format = FMT_H264;
2181 s->workaround_bugs= avctx->workaround_bugs;
2182
2183 // set defaults
2184 s->progressive_sequence=1;
2185 // s->decode_mb= ff_h263_decode_mb;
2186 s->low_delay= 1;
2187 avctx->pix_fmt= PIX_FMT_YUV420P;
2188
2189 decode_init_vlc(h);
2190
2191 return 0;
2192 }
2193
2194 static void frame_start(H264Context *h){
2195 MpegEncContext * const s = &h->s;
2196 int i;
2197
2198 MPV_frame_start(s, s->avctx);
2199 ff_er_frame_start(s);
2200 h->mmco_index=0;
2201
2202 assert(s->linesize && s->uvlinesize);
2203
2204 for(i=0; i<16; i++){
2205 h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
2206 h->chroma_subblock_offset[i]= 2*((scan8[i] - scan8[0])&7) + 2*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2207 }
2208 for(i=0; i<4; i++){
2209 h->block_offset[16+i]=
2210 h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2211 }
2212
2213 // s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
2214 }
2215
2216 static void hl_decode_mb(H264Context *h){
2217 MpegEncContext * const s = &h->s;
2218 const int mb_x= s->mb_x;
2219 const int mb_y= s->mb_y;
2220 const int mb_xy= mb_x + mb_y*s->mb_stride;
2221 const int mb_type= s->current_picture.mb_type[mb_xy];
2222 uint8_t *dest_y, *dest_cb, *dest_cr;
2223 int linesize, uvlinesize /*dct_offset*/;
2224 int i;
2225
2226 if(!s->decode)
2227 return;
2228
2229 if(s->mb_skiped){
2230 }
2231
2232 dest_y = s->current_picture.data[0] + (mb_y * 16* s->linesize ) + mb_x * 16;
2233 dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2234 dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2235
2236 if (h->mb_field_decoding_flag) {
2237 linesize = s->linesize * 2;
2238 uvlinesize = s->uvlinesize * 2;
2239 if(mb_y&1){ //FIXME move out of this func?
2240 dest_y -= s->linesize*15;
2241 dest_cb-= s->linesize*7;
2242 dest_cr-= s->linesize*7;
2243 }
2244 } else {
2245 linesize = s->linesize;
2246 uvlinesize = s->uvlinesize;
2247 // dct_offset = s->linesize * 16;
2248 }
2249
2250 if(IS_INTRA(mb_type)){
2251 if(!(s->flags&CODEC_FLAG_GRAY)){
2252 h->pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
2253 h->pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
2254 }
2255
2256 if(IS_INTRA4x4(mb_type)){
2257 if(!s->encoding){
2258 for(i=0; i<16; i++){
2259 uint8_t * const ptr= dest_y + h->block_offset[i];
2260 uint8_t *topright= ptr + 4 - linesize;
2261 const int topright_avail= (h->topright_samples_available<<i)&0x8000;
2262 const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
2263 int tr;
2264
2265 if(!topright_avail){
2266 tr= ptr[3 - linesize]*0x01010101;
2267 topright= (uint8_t*) &tr;
2268 }
2269
2270 h->pred4x4[ dir ](ptr, topright, linesize);
2271 if(h->non_zero_count_cache[ scan8[i] ]){
2272 if(s->codec_id == CODEC_ID_H264)
2273 h264_add_idct_c(ptr, h->mb + i*16, linesize);
2274 else
2275 svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
2276 }
2277 }
2278 }
2279 }else{
2280 h->pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
2281 if(s->codec_id == CODEC_ID_H264)
2282 h264_luma_dc_dequant_idct_c(h->mb, s->qscale);
2283 else
2284 svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
2285 }
2286 }else if(s->codec_id == CODEC_ID_H264){
2287 hl_motion(h, dest_y, dest_cb, dest_cr,
2288 s->dsp.put_h264_qpel_pixels_tab, s->dsp.put_h264_chroma_pixels_tab,
2289 s->dsp.avg_h264_qpel_pixels_tab, s->dsp.avg_h264_chroma_pixels_tab);
2290 }
2291
2292
2293 if(!IS_INTRA4x4(mb_type)){
2294 if(s->codec_id == CODEC_ID_H264){
2295 for(i=0; i<16; i++){
2296 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
2297 uint8_t * const ptr= dest_y + h->block_offset[i];
2298 h264_add_idct_c(ptr, h->mb + i*16, linesize);
2299 }
2300 }
2301 }else{
2302 for(i=0; i<16; i++){
2303 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
2304 uint8_t * const ptr= dest_y + h->block_offset[i];
2305 svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
2306 }
2307 }
2308 }
2309 }
2310
2311 if(!(s->flags&CODEC_FLAG_GRAY)){
2312 chroma_dc_dequant_idct_c(h->mb + 16*16, h->chroma_qp);
2313 chroma_dc_dequant_idct_c(h->mb + 16*16+4*16, h->chroma_qp);
2314 if(s->codec_id == CODEC_ID_H264){
2315 for(i=16; i<16+4; i++){
2316 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2317 uint8_t * const ptr= dest_cb + h->block_offset[i];
2318 h264_add_idct_c(ptr, h->mb + i*16, uvlinesize);
2319 }
2320 }
2321 for(i=20; i<20+4; i++){
2322 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2323 uint8_t * const ptr= dest_cr + h->block_offset[i];
2324 h264_add_idct_c(ptr, h->mb + i*16, uvlinesize);
2325 }
2326 }
2327 }else{
2328 for(i=16; i<16+4; i++){
2329 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2330 uint8_t * const ptr= dest_cb + h->block_offset[i];
2331 svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
2332 }
2333 }
2334 for(i=20; i<20+4; i++){
2335 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2336 uint8_t * const ptr= dest_cr + h->block_offset[i];
2337 svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
2338 }
2339 }
2340 }
2341 }
2342 }
2343
2344 static void decode_mb_cabac(H264Context *h){
2345 // MpegEncContext * const s = &h->s;
2346 }
2347
2348 /**
2349 * fills the default_ref_list.
2350 */
2351 static int fill_default_ref_list(H264Context *h){
2352 MpegEncContext * const s = &h->s;
2353 int i;
2354 Picture sorted_short_ref[16];
2355
2356 if(h->slice_type==B_TYPE){
2357 int out_i;
2358 int limit= -1;
2359
2360 for(out_i=0; out_i<h->short_ref_count; out_i++){
2361 int best_i=-1;
2362 int best_poc=-1;
2363
2364 for(i=0; i<h->short_ref_count; i++){
2365 const int poc= h->short_ref[i]->poc;
2366 if(poc > limit && poc < best_poc){
2367 best_poc= poc;
2368 best_i= i;
2369 }
2370 }
2371
2372 assert(best_i != -1);
2373
2374 limit= best_poc;
2375 sorted_short_ref[out_i]= *h->short_ref[best_i];
2376 }
2377 }
2378
2379 if(s->picture_structure == PICT_FRAME){
2380 if(h->slice_type==B_TYPE){
2381 const int current_poc= s->current_picture_ptr->poc;
2382 int list;
2383
2384 for(list=0; list<2; list++){
2385 int index=0;
2386
2387 for(i=0; i<h->short_ref_count && index < h->ref_count[list]; i++){
2388 const int i2= list ? h->short_ref_count - i - 1 : i;
2389 const int poc= sorted_short_ref[i2].poc;
2390
2391 if(sorted_short_ref[i2].reference != 3) continue; //FIXME refernce field shit
2392
2393 if((list==1 && poc > current_poc) || (list==0 && poc < current_poc)){
2394 h->default_ref_list[list][index ]= sorted_short_ref[i2];
2395 h->default_ref_list[list][index++].pic_id= sorted_short_ref[i2].frame_num;
2396 }
2397 }
2398
2399 for(i=0; i<h->long_ref_count && index < h->ref_count[ list ]; i++){
2400 if(h->long_ref[i]->reference != 3) continue;
2401
2402 h->default_ref_list[ list ][index ]= *h->long_ref[i];
2403 h->default_ref_list[ list ][index++].pic_id= i;;
2404 }
2405
2406 if(h->long_ref_count > 1 && h->short_ref_count==0){
2407 Picture temp= h->default_ref_list[1][0];
2408 h->default_ref_list[1][0] = h->default_ref_list[1][1];
2409 h->default_ref_list[1][0] = temp;
2410 }
2411
2412 if(index < h->ref_count[ list ])
2413 memset(&h->default_ref_list[list][index], 0, sizeof(Picture)*(h->ref_count[ list ] - index));
2414 }
2415 }else{
2416 int index=0;
2417 for(i=0; i<h->short_ref_count && index < h->ref_count[0]; i++){
2418 if(h->short_ref[i]->reference != 3) continue; //FIXME refernce field shit
2419 h->default_ref_list[0][index ]= *h->short_ref[i];
2420 h->default_ref_list[0][index++].pic_id= h->short_ref[i]->frame_num;
2421 }
2422 for(i=0; i<h->long_ref_count && index < h->ref_count[0]; i++){
2423 if(h->long_ref[i]->reference != 3) continue;
2424 h->default_ref_list[0][index ]= *h->long_ref[i];
2425 h->default_ref_list[0][index++].pic_id= i;;
2426 }
2427 if(index < h->ref_count[0])
2428 memset(&h->default_ref_list[0][index], 0, sizeof(Picture)*(h->ref_count[0] - index));
2429 }
2430 }else{ //FIELD
2431 if(h->slice_type==B_TYPE){
2432 }else{
2433 //FIXME second field balh
2434 }
2435 }
2436 return 0;
2437 }
2438
2439 static int decode_ref_pic_list_reordering(H264Context *h){
2440 MpegEncContext * const s = &h->s;
2441 int list;
2442
2443 if(h->slice_type==I_TYPE || h->slice_type==SI_TYPE) return 0; //FIXME move beofre func
2444
2445 for(list=0; list<2; list++){
2446 memcpy(h->ref_list[list], h->default_ref_list[list], sizeof(Picture)*h->ref_count[list]);
2447
2448 if(get_bits1(&s->gb)){
2449 int pred= h->curr_pic_num;
2450 int index;
2451
2452 for(index=0; ; index++){
2453 int reordering_of_pic_nums_idc= get_ue_golomb(&s->gb);
2454 int pic_id;
2455 int i;
2456
2457
2458 if(index >= h->ref_count[list]){
2459 fprintf(stderr, "reference count overflow\n");
2460 return -1;
2461 }
2462
2463 if(reordering_of_pic_nums_idc<3){
2464 if(reordering_of_pic_nums_idc<2){
2465 const int abs_diff_pic_num= get_ue_golomb(&s->gb) + 1;
2466
2467 if(abs_diff_pic_num >= h->max_pic_num){
2468 fprintf(stderr, "abs_diff_pic_num overflow\n");
2469 return -1;
2470 }
2471
2472 if(reordering_of_pic_nums_idc == 0) pred-= abs_diff_pic_num;
2473 else pred+= abs_diff_pic_num;
2474 pred &= h->max_pic_num - 1;
2475
2476 for(i= h->ref_count[list]-1; i>=index; i--){
2477 if(h->ref_list[list][i].pic_id == pred && h->ref_list[list][i].long_ref==0)
2478 break;
2479 }
2480 }else{
2481 pic_id= get_ue_golomb(&s->gb); //long_term_pic_idx
2482
2483 for(i= h->ref_count[list]-1; i>=index; i--){
2484 if(h->ref_list[list][i].pic_id == pic_id && h->ref_list[list][i].long_ref==1)
2485 break;
2486 }
2487 }
2488
2489 if(i < index){
2490 fprintf(stderr, "reference picture missing during reorder\n");
2491 memset(&h->ref_list[list][index], 0, sizeof(Picture)); //FIXME
2492 }else if(i > index){
2493 Picture tmp= h->ref_list[list][i];
2494 for(; i>index; i--){
2495 h->ref_list[list][i]= h->ref_list[list][i-1];
2496 }
2497 h->ref_list[list][index]= tmp;
2498 }
2499 }else if(reordering_of_pic_nums_idc==3)
2500 break;
2501 else{
2502 fprintf(stderr, "illegal reordering_of_pic_nums_idc\n");
2503 return -1;
2504 }
2505 }
2506 }
2507
2508 if(h->slice_type!=B_TYPE) break;
2509 }
2510 return 0;
2511 }
2512
2513 static int pred_weight_table(H264Context *h){
2514 MpegEncContext * const s = &h->s;
2515 int list, i;
2516
2517 h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
2518 h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
2519
2520 for(list=0; list<2; list++){
2521 for(i=0; i<h->ref_count[list]; i++){
2522 int luma_weight_flag, chroma_weight_flag;
2523
2524 luma_weight_flag= get_bits1(&s->gb);
2525 if(luma_weight_flag){
2526 h->luma_weight[list][i]= get_se_golomb(&s->gb);
2527 h->luma_offset[list][i]= get_se_golomb(&s->gb);
2528 }
2529
2530 chroma_weight_flag= get_bits1(&s->gb);
2531 if(chroma_weight_flag){
2532 int j;
2533 for(j=0; j<2; j++){
2534 h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);
2535 h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);
2536 }
2537 }
2538 }
2539 if(h->slice_type != B_TYPE) break;
2540 }
2541 return 0;
2542 }
2543
2544 /**
2545 * instantaneos decoder refresh.
2546 */
2547 static void idr(H264Context *h){
2548 int i;
2549
2550 for(i=0; i<h->long_ref_count; i++){
2551 h->long_ref[i]->reference=0;
2552 h->long_ref[i]= NULL;
2553 }
2554 h->long_ref_count=0;
2555
2556 for(i=0; i<h->short_ref_count; i++){
2557 h->short_ref[i]->reference=0;
2558 h->short_ref[i]= NULL;
2559 }
2560 h->short_ref_count=0;
2561 }
2562
2563 /**
2564 *
2565 * @return the removed picture or NULL if an error occures
2566 */
2567 static Picture * remove_short(H264Context *h, int frame_num){
2568 MpegEncContext * const s = &h->s;
2569 int i;
2570
2571 if(s->avctx->debug&FF_DEBUG_MMCO)
2572 printf("remove short %d count %d\n", frame_num, h->short_ref_count);
2573
2574 for(i=0; i<h->short_ref_count; i++){
2575 Picture *pic= h->short_ref[i];
2576 if(s->avctx->debug&FF_DEBUG_MMCO)
2577 printf("%d %d %p\n", i, pic->frame_num, pic);
2578 if(pic->frame_num == frame_num){
2579 h->short_ref[i]= NULL;
2580 memmove(&h->short_ref[i], &h->short_ref[i+1], (h->short_ref_count - i - 1)*sizeof(Picture*));
2581 h->short_ref_count--;
2582 return pic;
2583 }
2584 }
2585 return NULL;
2586 }
2587
2588 /**
2589 *
2590 * @return the removed picture or NULL if an error occures
2591 */
2592 static Picture * remove_long(H264Context *h, int i){
2593 Picture *pic;
2594
2595 if(i >= h->long_ref_count) return NULL;
2596 pic= h->long_ref[i];
2597 if(pic==NULL) return NULL;
2598
2599 h->long_ref[i]= NULL;
2600 memmove(&h->long_ref[i], &h->long_ref[i+1], (h->long_ref_count - i - 1)*sizeof(Picture*));
2601 h->long_ref_count--;
2602
2603 return pic;
2604 }
2605
2606 /**
2607 * Executes the reference picture marking (memory management control operations).
2608 */
2609 static int execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count){
2610 MpegEncContext * const s = &h->s;
2611 int i;
2612 int current_is_long=0;
2613 Picture *pic;
2614
2615 if((s->avctx->debug&FF_DEBUG_MMCO) && mmco_count==0)
2616 printf("no mmco here\n");
2617
2618 for(i=0; i<mmco_count; i++){
2619 if(s->avctx->debug&FF_DEBUG_MMCO)
2620 printf("mmco:%d %d %d\n", h->mmco[i].opcode, h->mmco[i].short_frame_num, h->mmco[i].long_index);
2621
2622 switch(mmco[i].opcode){
2623 case MMCO_SHORT2UNUSED:
2624 pic= remove_short(h, mmco[i].short_frame_num);
2625 if(pic==NULL) return -1;
2626 pic->reference= 0;
2627 break;
2628 case MMCO_SHORT2LONG:
2629 pic= remove_long(h, mmco[i].long_index);
2630 if(pic) pic->reference=0;
2631
2632 h->long_ref[ mmco[i].long_index ]= remove_short(h, mmco[i].short_frame_num);
2633 h->long_ref[ mmco[i].long_index ]->long_ref=1;
2634 break;
2635 case MMCO_LONG2UNUSED:
2636 pic= remove_long(h, mmco[i].long_index);
2637 if(pic==NULL) return -1;
2638 pic->reference= 0;
2639 break;
2640 case MMCO_LONG:
2641 pic= remove_long(h, mmco[i].long_index);
2642 if(pic) pic->reference=0;
2643
2644 h->long_ref[ mmco[i].long_index ]= s->current_picture_ptr;
2645 h->long_ref[ mmco[i].long_index ]->long_ref=1;
2646 h->long_ref_count++;
2647
2648 current_is_long=1;
2649 break;
2650 case MMCO_SET_MAX_LONG:
2651 assert(mmco[i].long_index <= 16);
2652 while(mmco[i].long_index < h->long_ref_count){
2653 pic= remove_long(h, mmco[i].long_index);
2654 pic->reference=0;
2655 }
2656 while(mmco[i].long_index > h->long_ref_count){
2657 h->long_ref[ h->long_ref_count++ ]= NULL;
2658 }
2659 break;
2660 case MMCO_RESET:
2661 while(h->short_ref_count){
2662 pic= remove_short(h, h->short_ref[0]->frame_num);
2663 pic->reference=0;
2664 }
2665 while(h->long_ref_count){
2666 pic= remove_long(h, h->long_ref_count-1);
2667 pic->reference=0;
2668 }
2669 break;
2670 default: assert(0);
2671 }
2672 }
2673
2674 if(!current_is_long){
2675 pic= remove_short(h, s->current_picture_ptr->frame_num);
2676 if(pic){
2677 pic->reference=0;
2678 fprintf(stderr, "illegal short term buffer state detected\n");
2679 }
2680
2681 if(h->short_ref_count)
2682 memmove(&h->short_ref[1], &h->short_ref[0], h->short_ref_count*sizeof(Picture*));
2683
2684 h->short_ref[0]= s->current_picture_ptr;
2685 h->short_ref[0]->long_ref=0;
2686 h->short_ref_count++;
2687 }
2688
2689 return 0;
2690 }
2691
2692 static int decode_ref_pic_marking(H264Context *h){
2693 MpegEncContext * const s = &h->s;
2694 int i;
2695
2696 if(h->nal_unit_type == NAL_IDR_SLICE){ //FIXME fields
2697 s->broken_link= get_bits1(&s->gb) -1;
2698 h->mmco[0].long_index= get_bits1(&s->gb) - 1; // current_long_term_idx
2699 if(h->mmco[0].long_index == -1)
2700 h->mmco_index= 0;
2701 else{
2702 h->mmco[0].opcode= MMCO_LONG;
2703 h->mmco_index= 1;
2704 }
2705 }else{
2706 if(get_bits1(&s->gb)){ // adaptive_ref_pic_marking_mode_flag
2707 for(i= h->mmco_index; i<MAX_MMCO_COUNT; i++) {
2708 MMCOOpcode opcode= get_ue_golomb(&s->gb);;
2709
2710 h->mmco[i].opcode= opcode;
2711 if(opcode==MMCO_SHORT2UNUSED || opcode==MMCO_SHORT2LONG){
2712 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
2713 /* if(h->mmco[i].short_frame_num >= h->short_ref_count || h->short_ref[ h->mmco[i].short_frame_num ] == NULL){
2714 fprintf(stderr, "illegal short ref in memory management control operation %d\n", mmco);
2715 return -1;
2716 }*/
2717 }
2718 if(opcode==MMCO_SHORT2LONG || opcode==MMCO_LONG2UNUSED || opcode==MMCO_LONG || opcode==MMCO_SET_MAX_LONG){
2719 h->mmco[i].long_index= get_ue_golomb(&s->gb);
2720 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){
2721 fprintf(stderr, "illegal long ref in memory management control operation %d\n", opcode);
2722 return -1;
2723 }
2724 }
2725
2726 if(opcode > MMCO_LONG){
2727 fprintf(stderr, "illegal memory management control operation %d\n", opcode);
2728 return -1;
2729 }
2730 }
2731 h->mmco_index= i;
2732 }else{
2733 assert(h->long_ref_count + h->short_ref_count <= h->sps.ref_frame_count);
2734
2735 if(h->long_ref_count + h->short_ref_count == h->sps.ref_frame_count){ //FIXME fields
2736 h->mmco[0].opcode= MMCO_SHORT2UNUSED;
2737 h->mmco[0].short_frame_num= h->short_ref[ h->short_ref_count - 1 ]->frame_num;
2738 h->mmco_index= 1;
2739 }else
2740 h->mmco_index= 0;
2741 }
2742 }
2743
2744 return 0;
2745 }
2746
2747 static int init_poc(H264Context *h){
2748 MpegEncContext * const s = &h->s;
2749 const int max_frame_num= 1<<h->sps.log2_max_frame_num;
2750 int field_poc[2];
2751
2752 if(h->nal_unit_type == NAL_IDR_SLICE){
2753 h->frame_num_offset= 0;
2754 }else{
2755 if(h->frame_num < h->prev_frame_num)
2756 h->frame_num_offset= h->prev_frame_num_offset + max_frame_num;
2757 else
2758 h->frame_num_offset= h->prev_frame_num_offset;
2759 }
2760
2761 if(h->sps.poc_type==0){
2762 const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
2763
2764 if (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
2765 h->poc_msb = h->prev_poc_msb + max_poc_lsb;
2766 else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
2767 h->poc_msb = h->prev_poc_msb - max_poc_lsb;
2768 else
2769 h->poc_msb = h->prev_poc_msb;
2770 //printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
2771 field_poc[0] =
2772 field_poc[1] = h->poc_msb + h->poc_lsb;
2773 if(s->picture_structure == PICT_FRAME)
2774 field_poc[1] += h->delta_poc_bottom;
2775 }else if(h->sps.poc_type==1){
2776 int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
2777 int i;
2778
2779 if(h->sps.poc_cycle_length != 0)
2780 abs_frame_num = h->frame_num_offset + h->frame_num;
2781 else
2782 abs_frame_num = 0;
2783
2784 if(h->nal_ref_idc==0 && abs_frame_num > 0)
2785 abs_frame_num--;
2786
2787 expected_delta_per_poc_cycle = 0;
2788 for(i=0; i < h->sps.poc_cycle_length; i++)
2789 expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
2790
2791 if(abs_frame_num > 0){
2792 int poc_cycle_cnt = (abs_frame_num - 1) / h->sps.poc_cycle_length;
2793 int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
2794
2795 expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
2796 for(i = 0; i <= frame_num_in_poc_cycle; i++)
2797 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
2798 } else
2799 expectedpoc = 0;
2800
2801 if(h->nal_ref_idc == 0)
2802 expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
2803
2804 field_poc[0] = expectedpoc + h->delta_poc[0];
2805 field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
2806
2807 if(s->picture_structure == PICT_FRAME)
2808 field_poc[1] += h->delta_poc[1];
2809 }else{
2810 int poc;
2811 if(h->nal_unit_type == NAL_IDR_SLICE){
2812 poc= 0;
2813 }else{
2814 if(h->nal_ref_idc) poc= 2*(h->frame_num_offset + h->frame_num);
2815 else poc= 2*(h->frame_num_offset + h->frame_num) - 1;
2816 }
2817 field_poc[0]= poc;
2818 field_poc[1]= poc;
2819 }
2820
2821 if(s->picture_structure != PICT_BOTTOM_FIELD)
2822 s->current_picture_ptr->field_poc[0]= field_poc[0];
2823 if(s->picture_structure != PICT_TOP_FIELD)
2824 s->current_picture_ptr->field_poc[1]= field_poc[1];
2825 if(s->picture_structure == PICT_FRAME) // FIXME field pix?
2826 s->current_picture_ptr->poc= FFMIN(field_poc[0], field_poc[1]);
2827
2828 return 0;
2829 }
2830
2831 /**
2832 * decodes a slice header.
2833 * this will allso call MPV_common_init() and frame_start() as needed
2834 */
2835 static int decode_slice_header(H264Context *h){
2836 MpegEncContext * const s = &h->s;
2837 int first_mb_in_slice, pps_id;
2838 int num_ref_idx_active_override_flag;
2839 static const uint8_t slice_type_map[5]= {P_TYPE, B_TYPE, I_TYPE, SP_TYPE, SI_TYPE};
2840 float new_aspect;
2841
2842 s->current_picture.reference= h->nal_ref_idc != 0;
2843
2844 first_mb_in_slice= get_ue_golomb(&s->gb);
2845
2846 h->slice_type= get_ue_golomb(&s->gb);
2847 if(h->slice_type > 9){
2848 fprintf(stderr, "slice type too large (%d) at %d %d\n", h->slice_type, s->mb_x, s->mb_y);
2849 }
2850 if(h->slice_type > 4){
2851 h->slice_type -= 5;
2852 h->slice_type_fixed=1;
2853 }else
2854 h->slice_type_fixed=0;
2855
2856 h->slice_type= slice_type_map[ h->slice_type ];
2857
2858 s->pict_type= h->slice_type; // to make a few old func happy, its wrong though
2859
2860 pps_id= get_ue_golomb(&s->gb);
2861 if(pps_id>255){
2862 fprintf(stderr, "pps_id out of range\n");
2863 return -1;
2864 }
2865 h->pps= h->pps_buffer[pps_id];
2866 if(h->pps.slice_group_count == 0){
2867 fprintf(stderr, "non existing PPS referenced\n");
2868 return -1;
2869 }
2870
2871 h->sps= h->sps_buffer[ h->pps.sps_id ];
2872 if(h->sps.log2_max_frame_num == 0){
2873 fprintf(stderr, "non existing SPS referenced\n");
2874 return -1;
2875 }
2876
2877 s->mb_width= h->sps.mb_width;
2878 s->mb_height= h->sps.mb_height;
2879
2880 h->b_stride= s->mb_width*4;
2881 h->b8_stride= s->mb_width*2;
2882
2883 s->mb_x = first_mb_in_slice % s->mb_width;
2884 s->mb_y = first_mb_in_slice / s->mb_width; //FIXME AFFW
2885
2886 s->width = 16*s->mb_width - 2*(h->pps.crop_left + h->pps.crop_right );
2887 if(h->sps.frame_mbs_only_flag)
2888 s->height= 16*s->mb_height - 2*(h->pps.crop_top + h->pps.crop_bottom);
2889 else
2890 s->height= 16*s->mb_height - 4*(h->pps.crop_top + h->pps.crop_bottom); //FIXME recheck
2891
2892 if(h->pps.crop_left || h->pps.crop_top){
2893 fprintf(stderr, "insane croping not completly supported, this could look slightly wrong ...\n");
2894 }
2895
2896 if(s->aspected_height) //FIXME emms at end of slice ?
2897 new_aspect= h->sps.sar_width*s->width / (float)(s->height*h->sps.sar_height);
2898 else
2899 new_aspect=0;
2900
2901 if (s->context_initialized
2902 && ( s->width != s->avctx->width || s->height != s->avctx->height
2903 || ABS(new_aspect - s->avctx->aspect_ratio) > 0.001)) {
2904 free_tables(h);
2905 MPV_common_end(s);
2906 }
2907 if (!s->context_initialized) {
2908 if (MPV_common_init(s) < 0)
2909 return -1;
2910
2911 alloc_tables(h);
2912
2913 s->avctx->width = s->width;
2914 s->avctx->height = s->height;
2915 s->avctx->aspect_ratio= new_aspect;
2916 }
2917
2918 if(first_mb_in_slice == 0){
2919 frame_start(h);
2920 }
2921
2922 s->current_picture_ptr->frame_num= //FIXME frame_num cleanup
2923 h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
2924
2925 if(h->sps.frame_mbs_only_flag){
2926 s->picture_structure= PICT_FRAME;
2927 }else{
2928 if(get_bits1(&s->gb)) //field_pic_flag
2929 s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
2930 else
2931 s->picture_structure= PICT_FRAME;
2932 }
2933
2934 if(s->picture_structure==PICT_FRAME){
2935 h->curr_pic_num= h->frame_num;
2936 h->max_pic_num= 1<< h->sps.log2_max_frame_num;
2937 }else{
2938 h->curr_pic_num= 2*h->frame_num;
2939 h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
2940 }
2941
2942 if(h->nal_unit_type == NAL_IDR_SLICE){
2943 int idr_pic_id= get_ue_golomb(&s->gb);
2944 }
2945
2946 if(h->sps.poc_type==0){
2947 h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
2948
2949 if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
2950 h->delta_poc_bottom= get_se_golomb(&s->gb);
2951 }
2952 }
2953
2954 if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
2955 h->delta_poc[0]= get_se_golomb(&s->gb);
2956
2957 if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
2958 h->delta_poc[1]= get_se_golomb(&s->gb);
2959 }
2960
2961 init_poc(h);
2962
2963 if(h->pps.redundant_pic_cnt_present){
2964 h->redundant_pic_count= get_ue_golomb(&s->gb);
2965 }
2966
2967 //set defaults, might be overriden a few line later
2968 h->ref_count[0]= h->pps.ref_count[0];
2969 h->ref_count[1]= h->pps.ref_count[1];
2970
2971 if(h->slice_type == P_TYPE || h->slice_type == SP_TYPE || h->slice_type == B_TYPE){
2972 if(h->slice_type == B_TYPE){
2973 h->direct_spatial_mv_pred= get_bits1(&s->gb);
2974 }
2975 num_ref_idx_active_override_flag= get_bits1(&s->gb);
2976
2977 if(num_ref_idx_active_override_flag){
2978 h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
2979 if(h->slice_type==B_TYPE)
2980 h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
2981
2982 if(h->ref_count[0] > 32 || h->ref_count[1] > 32){
2983 fprintf(stderr, "reference overflow\n");
2984 return -1;
2985 }
2986 }
2987 }
2988
2989 if(first_mb_in_slice == 0){
2990 fill_default_ref_list(h);
2991 }
2992
2993 decode_ref_pic_list_reordering(h);
2994
2995 if( (h->pps.weighted_pred && (h->slice_type == P_TYPE || h->slice_type == SP_TYPE ))
2996 || (h->pps.weighted_bipred_idc==1 && h->slice_type==B_TYPE ) )
2997 pred_weight_table(h);
2998
2999 if(s->current_picture.reference)
3000 decode_ref_pic_marking(h);
3001 //FIXME CABAC stuff
3002
3003 s->qscale = h->pps.init_qp + get_se_golomb(&s->gb); //slice_qp_delta
3004 //FIXME qscale / qp ... stuff
3005 if(h->slice_type == SP_TYPE){
3006 int sp_for_switch_flag= get_bits1(&s->gb);
3007 }
3008 if(h->slice_type==SP_TYPE || h->slice_type == SI_TYPE){
3009 int slice_qs_delta= get_se_golomb(&s->gb);
3010 }
3011
3012 if( h->pps.deblocking_filter_parameters_present ) {
3013 h->disable_deblocking_filter_idc= get_ue_golomb(&s->gb);
3014 if( h->disable_deblocking_filter_idc != 1 ) {
3015 h->slice_alpha_c0_offset_div2= get_se_golomb(&s->gb);
3016 h->slice_beta_offset_div2= get_se_golomb(&s->gb);
3017 }
3018 }else
3019 h->disable_deblocking_filter_idc= 0;
3020
3021 #if 0 //FMO
3022 if( h->pps.num_slice_groups > 1 && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
3023 slice_group_change_cycle= get_bits(&s->gb, ?);
3024 #endif
3025
3026 if(s->avctx->debug&FF_DEBUG_PICT_INFO){
3027 printf("mb:%d %c pps:%d frame:%d poc:%d/%d ref:%d/%d qp:%d loop:%d\n",
3028 first_mb_in_slice,
3029 av_get_pict_type_char(h->slice_type),
3030 pps_id, h->frame_num,
3031 s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
3032 h->ref_count[0], h->ref_count[1],
3033 s->qscale,
3034 h->disable_deblocking_filter_idc
3035 );
3036 }
3037
3038 return 0;
3039 }
3040
3041 /**
3042 *
3043 */
3044 static inline int get_level_prefix(GetBitContext *gb){
3045 unsigned int buf;
3046 int log;
3047
3048 OPEN_READER(re, gb);
3049 UPDATE_CACHE(re, gb);
3050 buf=GET_CACHE(re, gb);
3051
3052 log= 32 - av_log2(buf);
3053 #ifdef TRACE
3054 print_bin(buf>>(32-log), log);
3055 printf("%5d %2d %3d lpr @%5d in %s get_level_prefix\n", buf>>(32-log), log, log-1, get_bits_count(gb), __FILE__);
3056 #endif
3057
3058 LAST_SKIP_BITS(re, gb, log);
3059 CLOSE_READER(re, gb);
3060
3061 return log-1;
3062 }
3063
3064 /**
3065 * decodes a residual block.
3066 * @param n block index
3067 * @param scantable scantable
3068 * @param max_coeff number of coefficients in the block
3069 * @return <0 if an error occured
3070 */
3071 static int decode_residual(H264Context *h, GetBitContext *gb, DCTELEM *block, int n, const uint8_t *scantable, int qp, int max_coeff){
3072 MpegEncContext * const s = &h->s;
3073 const uint16_t *qmul= dequant_coeff[qp];
3074 static const int coeff_token_table_index[17]= {0, 0, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 3, 3, 3};
3075 int level[16], run[16];
3076 int suffix_length, zeros_left, coeff_num, coeff_token, total_coeff, i, trailing_ones;
3077
3078 //FIXME put trailing_onex into the context
3079
3080 if(n == CHROMA_DC_BLOCK_INDEX){
3081 coeff_token= get_vlc2(gb, chroma_dc_coeff_token_vlc.table, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 1);
3082 total_coeff= coeff_token>>2;
3083 }else{
3084 if(n == LUMA_DC_BLOCK_INDEX){
3085 total_coeff= pred_non_zero_count(h, 0);
3086 coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
3087 total_coeff= coeff_token>>2;
3088 }else{
3089 total_coeff= pred_non_zero_count(h, n);
3090 coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
3091 total_coeff= coeff_token>>2;
3092 h->non_zero_count_cache[ scan8[n] ]= total_coeff;
3093 }
3094 }
3095
3096 //FIXME set last_non_zero?
3097
3098 if(total_coeff==0)
3099 return 0;
3100
3101 trailing_ones= coeff_token&3;
3102 tprintf("trailing:%d, total:%d\n", trailing_ones, total_coeff);
3103 assert(total_coeff<=16);
3104
3105 for(i=0; i<trailing_ones; i++){
3106 level[i]= 1 - 2*get_bits1(gb);
3107 }
3108
3109 suffix_length= total_coeff > 10 && trailing_ones < 3;
3110
3111 for(; i<total_coeff; i++){
3112 const int prefix= get_level_prefix(gb);