f186fc26b2a3897ed84538f1ed363a9e820e476e
[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 log2_max_frame_num; ///< log2_max_frame_num_minus4 + 4
63 int poc_type; ///< pic_order_cnt_type
64 int log2_max_poc_lsb; ///< log2_max_pic_order_cnt_lsb_minus4
65 int delta_pic_order_always_zero_flag;
66 int offset_for_non_ref_pic;
67 int offset_for_top_to_bottom_field;
68 int poc_cycle_length; ///< num_ref_frames_in_pic_order_cnt_cycle
69 int ref_frame_count; ///< num_ref_frames
70 int gaps_in_frame_num_allowed_flag;
71 int mb_width; ///< frame_width_in_mbs_minus1 + 1
72 int mb_height; ///< frame_height_in_mbs_minus1 + 1
73 int frame_mbs_only_flag;
74 int mb_aff; ///<mb_adaptive_frame_field_flag
75 int direct_8x8_inference_flag;
76 int crop; ///< frame_cropping_flag
77 int crop_left; ///< frame_cropping_rect_left_offset
78 int crop_right; ///< frame_cropping_rect_right_offset
79 int crop_top; ///< frame_cropping_rect_top_offset
80 int crop_bottom; ///< frame_cropping_rect_bottom_offset
81 int vui_parameters_present_flag;
82 int sar_width;
83 int sar_height;
84 short offset_for_ref_frame[256]; //FIXME dyn aloc?
85 }SPS;
86
87 /**
88 * Picture parameter set
89 */
90 typedef struct PPS{
91 int sps_id;
92 int cabac; ///< entropy_coding_mode_flag
93 int pic_order_present; ///< pic_order_present_flag
94 int slice_group_count; ///< num_slice_groups_minus1 + 1
95 int mb_slice_group_map_type;
96 int ref_count[2]; ///< num_ref_idx_l0/1_active_minus1 + 1
97 int weighted_pred; ///< weighted_pred_flag
98 int weighted_bipred_idc;
99 int init_qp; ///< pic_init_qp_minus26 + 26
100 int init_qs; ///< pic_init_qs_minus26 + 26
101 int chroma_qp_index_offset;
102 int deblocking_filter_parameters_present; ///< deblocking_filter_parameters_present_flag
103 int constrained_intra_pred; ///< constrained_intra_pred_flag
104 int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag
105 }PPS;
106
107 /**
108 * Memory management control operation opcode.
109 */
110 typedef enum MMCOOpcode{
111 MMCO_END=0,
112 MMCO_SHORT2UNUSED,
113 MMCO_LONG2UNUSED,
114 MMCO_SHORT2LONG,
115 MMCO_SET_MAX_LONG,
116 MMCO_RESET,
117 MMCO_LONG,
118 } MMCOOpcode;
119
120 /**
121 * Memory management control operation.
122 */
123 typedef struct MMCO{
124 MMCOOpcode opcode;
125 int short_frame_num;
126 int long_index;
127 } MMCO;
128
129 /**
130 * H264Context
131 */
132 typedef struct H264Context{
133 MpegEncContext s;
134 int nal_ref_idc;
135 int nal_unit_type;
136 #define NAL_SLICE 1
137 #define NAL_DPA 2
138 #define NAL_DPB 3
139 #define NAL_DPC 4
140 #define NAL_IDR_SLICE 5
141 #define NAL_SEI 6
142 #define NAL_SPS 7
143 #define NAL_PPS 8
144 #define NAL_PICTURE_DELIMITER 9
145 #define NAL_FILTER_DATA 10
146 uint8_t *rbsp_buffer;
147 int rbsp_buffer_size;
148
149 int chroma_qp; //QPc
150
151 int prev_mb_skiped; //FIXME remove (IMHO not used)
152
153 //prediction stuff
154 int chroma_pred_mode;
155 int intra16x16_pred_mode;
156
157 int8_t intra4x4_pred_mode_cache[5*8];
158 int8_t (*intra4x4_pred_mode)[8];
159 void (*pred4x4 [9+3])(uint8_t *src, uint8_t *topright, int stride);//FIXME move to dsp?
160 void (*pred8x8 [4+3])(uint8_t *src, int stride);
161 void (*pred16x16[4+3])(uint8_t *src, int stride);
162 unsigned int topleft_samples_available;
163 unsigned int top_samples_available;
164 unsigned int topright_samples_available;
165 unsigned int left_samples_available;
166
167 /**
168 * non zero coeff count cache.
169 * is 64 if not available.
170 */
171 uint8_t non_zero_count_cache[6*8];
172 uint8_t (*non_zero_count)[16];
173
174 /**
175 * Motion vector cache.
176 */
177 int16_t mv_cache[2][5*8][2];
178 int8_t ref_cache[2][5*8];
179 #define LIST_NOT_USED -1 //FIXME rename?
180 #define PART_NOT_AVAILABLE -2
181
182 /**
183 * is 1 if the specific list MV&references are set to 0,0,-2.
184 */
185 int mv_cache_clean[2];
186
187 int block_offset[16+8];
188 int chroma_subblock_offset[16]; //FIXME remove
189
190 uint16_t *mb2b_xy; //FIXME are these 4 a good idea?
191 uint16_t *mb2b8_xy;
192 int b_stride;
193 int b8_stride;
194
195 int halfpel_flag;
196 int thirdpel_flag;
197
198 int unknown_svq3_flag;
199 int next_slice_index;
200
201 SPS sps_buffer[MAX_SPS_COUNT];
202 SPS sps; ///< current sps
203
204 PPS pps_buffer[MAX_PPS_COUNT];
205 /**
206 * current pps
207 */
208 PPS pps; //FIXME move tp Picture perhaps? (->no) do we need that?
209
210 int slice_num;
211 uint8_t *slice_table_base;
212 uint8_t *slice_table; ///< slice_table_base + mb_stride + 1
213 int slice_type;
214 int slice_type_fixed;
215
216 //interlacing specific flags
217 int mb_field_decoding_flag;
218
219 int sub_mb_type[4];
220
221 //POC stuff
222 int poc_lsb;
223 int poc_msb;
224 int delta_poc_bottom;
225 int delta_poc[2];
226 int frame_num;
227 int prev_poc_msb; ///< poc_msb of the last reference pic for POC type 0
228 int prev_poc_lsb; ///< poc_lsb of the last reference pic for POC type 0
229 int frame_num_offset; ///< for POC type 2
230 int prev_frame_num_offset; ///< for POC type 2
231 int prev_frame_num; ///< frame_num of the last pic for POC type 1/2
232
233 /**
234 * frame_num for frames or 2*frame_num for field pics.
235 */
236 int curr_pic_num;
237
238 /**
239 * max_frame_num or 2*max_frame_num for field pics.
240 */
241 int max_pic_num;
242
243 //Weighted pred stuff
244 int luma_log2_weight_denom;
245 int chroma_log2_weight_denom;
246 int luma_weight[2][16];
247 int luma_offset[2][16];
248 int chroma_weight[2][16][2];
249 int chroma_offset[2][16][2];
250
251 //deblock
252 int disable_deblocking_filter_idc;
253 int slice_alpha_c0_offset_div2;
254 int slice_beta_offset_div2;
255
256 int redundant_pic_count;
257
258 int direct_spatial_mv_pred;
259
260 /**
261 * num_ref_idx_l0/1_active_minus1 + 1
262 */
263 int ref_count[2];// FIXME split for AFF
264 Picture *short_ref[16];
265 Picture *long_ref[16];
266 Picture default_ref_list[2][32];
267 Picture ref_list[2][32]; //FIXME size?
268 Picture field_ref_list[2][32]; //FIXME size?
269
270 /**
271 * memory management control operations buffer.
272 */
273 MMCO mmco[MAX_MMCO_COUNT];
274 int mmco_index;
275
276 int long_ref_count; ///< number of actual long term references
277 int short_ref_count; ///< number of actual short term references
278
279 //data partitioning
280 GetBitContext intra_gb;
281 GetBitContext inter_gb;
282 GetBitContext *intra_gb_ptr;
283 GetBitContext *inter_gb_ptr;
284
285 DCTELEM mb[16*24] __align8;
286 }H264Context;
287
288 static VLC coeff_token_vlc[4];
289 static VLC chroma_dc_coeff_token_vlc;
290
291 static VLC total_zeros_vlc[15];
292 static VLC chroma_dc_total_zeros_vlc[3];
293
294 static VLC run_vlc[6];
295 static VLC run7_vlc;
296
297 static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
298 static void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
299
300 static inline uint32_t pack16to32(int a, int b){
301 #ifdef WORDS_BIGENDIAN
302 return (b&0xFFFF) + (a<<16);
303 #else
304 return (a&0xFFFF) + (b<<16);
305 #endif
306 }
307
308 /**
309 * fill a rectangle.
310 * @param h height of the recatangle, should be a constant
311 * @param w width of the recatangle, should be a constant
312 * @param size the size of val (1 or 4), should be a constant
313 */
314 static inline void fill_rectangle(void *vp, int w, int h, int stride, uint32_t val, int size){ //FIXME ensure this IS inlined
315 uint8_t *p= (uint8_t*)vp;
316 assert(size==1 || size==4);
317
318 w *= size;
319 stride *= size;
320
321 //FIXME check what gcc generates for 64 bit on x86 and possible write a 32 bit ver of it
322 if(w==2 && h==2){
323 *(uint16_t*)(p + 0)=
324 *(uint16_t*)(p + stride)= size==4 ? val : val*0x0101;
325 }else if(w==2 && h==4){
326 *(uint16_t*)(p + 0*stride)=
327 *(uint16_t*)(p + 1*stride)=
328 *(uint16_t*)(p + 2*stride)=
329 *(uint16_t*)(p + 3*stride)= size==4 ? val : val*0x0101;
330 }else if(w==4 && h==1){
331 *(uint32_t*)(p + 0*stride)= size==4 ? val : val*0x01010101;
332 }else if(w==4 && h==2){
333 *(uint32_t*)(p + 0*stride)=
334 *(uint32_t*)(p + 1*stride)= size==4 ? val : val*0x01010101;
335 }else if(w==4 && h==4){
336 *(uint32_t*)(p + 0*stride)=
337 *(uint32_t*)(p + 1*stride)=
338 *(uint32_t*)(p + 2*stride)=
339 *(uint32_t*)(p + 3*stride)= size==4 ? val : val*0x01010101;
340 }else if(w==8 && h==1){
341 *(uint32_t*)(p + 0)=
342 *(uint32_t*)(p + 4)= size==4 ? val : val*0x01010101;
343 }else if(w==8 && h==2){
344 *(uint32_t*)(p + 0 + 0*stride)=
345 *(uint32_t*)(p + 4 + 0*stride)=
346 *(uint32_t*)(p + 0 + 1*stride)=
347 *(uint32_t*)(p + 4 + 1*stride)= size==4 ? val : val*0x01010101;
348 }else if(w==8 && h==4){
349 *(uint64_t*)(p + 0*stride)=
350 *(uint64_t*)(p + 1*stride)=
351 *(uint64_t*)(p + 2*stride)=
352 *(uint64_t*)(p + 3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
353 }else if(w==16 && h==2){
354 *(uint64_t*)(p + 0+0*stride)=
355 *(uint64_t*)(p + 8+0*stride)=
356 *(uint64_t*)(p + 0+1*stride)=
357 *(uint64_t*)(p + 8+1*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
358 }else if(w==16 && h==4){
359 *(uint64_t*)(p + 0+0*stride)=
360 *(uint64_t*)(p + 8+0*stride)=
361 *(uint64_t*)(p + 0+1*stride)=
362 *(uint64_t*)(p + 8+1*stride)=
363 *(uint64_t*)(p + 0+2*stride)=
364 *(uint64_t*)(p + 8+2*stride)=
365 *(uint64_t*)(p + 0+3*stride)=
366 *(uint64_t*)(p + 8+3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
367 }else
368 assert(0);
369 }
370
371 static inline void fill_caches(H264Context *h, int mb_type){
372 MpegEncContext * const s = &h->s;
373 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
374 int topleft_xy, top_xy, topright_xy, left_xy[2];
375 int topleft_type, top_type, topright_type, left_type[2];
376 int left_block[4];
377 int i;
378
379 //wow what a mess, why didnt they simplify the interlacing&intra stuff, i cant imagine that these complex rules are worth it
380
381 if(h->sps.mb_aff){
382 //FIXME
383 topleft_xy = 0; /* avoid warning */
384 top_xy = 0; /* avoid warning */
385 topright_xy = 0; /* avoid warning */
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 if(match_count > 1){ //most common
793 *mx= mid_pred(A[0], B[0], C[0]);
794 *my= mid_pred(A[1], B[1], C[1]);
795 }else if(match_count==1){
796 if(left_ref==ref){
797 *mx= A[0];
798 *my= A[1];
799 }else if(top_ref==ref){
800 *mx= B[0];
801 *my= B[1];
802 }else{
803 *mx= C[0];
804 *my= C[1];
805 }
806 }else{
807 if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
808 *mx= A[0];
809 *my= A[1];
810 }else{
811 *mx= mid_pred(A[0], B[0], C[0]);
812 *my= mid_pred(A[1], B[1], C[1]);
813 }
814 }
815
816 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);
817 }
818
819 /**
820 * gets the directionally predicted 16x8 MV.
821 * @param n the block index
822 * @param mx the x component of the predicted motion vector
823 * @param my the y component of the predicted motion vector
824 */
825 static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
826 if(n==0){
827 const int top_ref= h->ref_cache[list][ scan8[0] - 8 ];
828 const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
829
830 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);
831
832 if(top_ref == ref){
833 *mx= B[0];
834 *my= B[1];
835 return;
836 }
837 }else{
838 const int left_ref= h->ref_cache[list][ scan8[8] - 1 ];
839 const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];
840
841 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);
842
843 if(left_ref == ref){
844 *mx= A[0];
845 *my= A[1];
846 return;
847 }
848 }
849
850 //RARE
851 pred_motion(h, n, 4, list, ref, mx, my);
852 }
853
854 /**
855 * gets the directionally predicted 8x16 MV.
856 * @param n the block index
857 * @param mx the x component of the predicted motion vector
858 * @param my the y component of the predicted motion vector
859 */
860 static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
861 if(n==0){
862 const int left_ref= h->ref_cache[list][ scan8[0] - 1 ];
863 const int16_t * const A= h->mv_cache[list][ scan8[0] - 1 ];
864
865 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);
866
867 if(left_ref == ref){
868 *mx= A[0];
869 *my= A[1];
870 return;
871 }
872 }else{
873 const int16_t * C;
874 int diagonal_ref;
875
876 diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);
877
878 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);
879
880 if(diagonal_ref == ref){
881 *mx= C[0];
882 *my= C[1];
883 return;
884 }
885 }
886
887 //RARE
888 pred_motion(h, n, 2, list, ref, mx, my);
889 }
890
891 static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
892 const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
893 const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];
894
895 tprintf("pred_pskip: (%d) (%d) at %2d %2d", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
896
897 if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
898 || (top_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0)
899 || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){
900
901 *mx = *my = 0;
902 return;
903 }
904
905 pred_motion(h, 0, 4, 0, 0, mx, my);
906
907 return;
908 }
909
910 static inline void write_back_motion(H264Context *h, int mb_type){
911 MpegEncContext * const s = &h->s;
912 const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
913 const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
914 int list;
915
916 for(list=0; list<2; list++){
917 int y;
918 if((!IS_8X8(mb_type)) && !USES_LIST(mb_type, list)){
919 if(1){ //FIXME skip or never read if mb_type doesnt use it
920 for(y=0; y<4; y++){
921 *(uint64_t*)s->current_picture.motion_val[list][b_xy + 0 + y*h->b_stride]=
922 *(uint64_t*)s->current_picture.motion_val[list][b_xy + 2 + y*h->b_stride]= 0;
923 }
924 for(y=0; y<2; y++){
925 *(uint16_t*)s->current_picture.motion_val[list][b8_xy + y*h->b8_stride]= (LIST_NOT_USED&0xFF)*0x0101;
926 }
927 }
928 continue; //FIXME direct mode ...
929 }
930
931 for(y=0; y<4; y++){
932 *(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];
933 *(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];
934 }
935 for(y=0; y<2; y++){
936 s->current_picture.ref_index[list][b8_xy + 0 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+0 + 16*y];
937 s->current_picture.ref_index[list][b8_xy + 1 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+2 + 16*y];
938 }
939 }
940 }
941
942 /**
943 * Decodes a network abstraction layer unit.
944 * @param consumed is the number of bytes used as input
945 * @param length is the length of the array
946 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp ttailing?
947 * @returns decoded bytes, might be src+1 if no escapes
948 */
949 static uint8_t *decode_nal(H264Context *h, uint8_t *src, int *dst_length, int *consumed, int length){
950 int i, si, di;
951 uint8_t *dst;
952
953 // src[0]&0x80; //forbidden bit
954 h->nal_ref_idc= src[0]>>5;
955 h->nal_unit_type= src[0]&0x1F;
956
957 src++; length--;
958 #if 0
959 for(i=0; i<length; i++)
960 printf("%2X ", src[i]);
961 #endif
962 for(i=0; i+1<length; i+=2){
963 if(src[i]) continue;
964 if(i>0 && src[i-1]==0) i--;
965 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
966 if(src[i+2]!=3){
967 /* startcode, so we must be past the end */
968 length=i;
969 }
970 break;
971 }
972 }
973
974 if(i>=length-1){ //no escaped 0
975 *dst_length= length;
976 *consumed= length+1; //+1 for the header
977 return src;
978 }
979
980 h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length);
981 dst= h->rbsp_buffer;
982
983 //printf("deoding esc\n");
984 si=di=0;
985 while(si<length){
986 //remove escapes (very rare 1:2^22)
987 if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
988 if(src[si+2]==3){ //escape
989 dst[di++]= 0;
990 dst[di++]= 0;
991 si+=3;
992 }else //next start code
993 break;
994 }
995
996 dst[di++]= src[si++];
997 }
998
999 *dst_length= di;
1000 *consumed= si + 1;//+1 for the header
1001 //FIXME store exact number of bits in the getbitcontext (its needed for decoding)
1002 return dst;
1003 }
1004
1005 /**
1006 * @param src the data which should be escaped
1007 * @param dst the target buffer, dst+1 == src is allowed as a special case
1008 * @param length the length of the src data
1009 * @param dst_length the length of the dst array
1010 * @returns length of escaped data in bytes or -1 if an error occured
1011 */
1012 static int encode_nal(H264Context *h, uint8_t *dst, uint8_t *src, int length, int dst_length){
1013 int i, escape_count, si, di;
1014 uint8_t *temp;
1015
1016 assert(length>=0);
1017 assert(dst_length>0);
1018
1019 dst[0]= (h->nal_ref_idc<<5) + h->nal_unit_type;
1020
1021 if(length==0) return 1;
1022
1023 escape_count= 0;
1024 for(i=0; i<length; i+=2){
1025 if(src[i]) continue;
1026 if(i>0 && src[i-1]==0)
1027 i--;
1028 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1029 escape_count++;
1030 i+=2;
1031 }
1032 }
1033
1034 if(escape_count==0){
1035 if(dst+1 != src)
1036 memcpy(dst+1, src, length);
1037 return length + 1;
1038 }
1039
1040 if(length + escape_count + 1> dst_length)
1041 return -1;
1042
1043 //this should be damn rare (hopefully)
1044
1045 h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length + escape_count);
1046 temp= h->rbsp_buffer;
1047 //printf("encoding esc\n");
1048
1049 si= 0;
1050 di= 0;
1051 while(si < length){
1052 if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1053 temp[di++]= 0; si++;
1054 temp[di++]= 0; si++;
1055 temp[di++]= 3;
1056 temp[di++]= src[si++];
1057 }
1058 else
1059 temp[di++]= src[si++];
1060 }
1061 memcpy(dst+1, temp, length+escape_count);
1062
1063 assert(di == length+escape_count);
1064
1065 return di + 1;
1066 }
1067
1068 /**
1069 * write 1,10,100,1000,... for alignment, yes its exactly inverse to mpeg4
1070 */
1071 static void encode_rbsp_trailing(PutBitContext *pb){
1072 int length;
1073 put_bits(pb, 1, 1);
1074 length= (-get_bit_count(pb))&7;
1075 if(length) put_bits(pb, length, 0);
1076 }
1077
1078 /**
1079 * identifies the exact end of the bitstream
1080 * @return the length of the trailing, or 0 if damaged
1081 */
1082 static int decode_rbsp_trailing(uint8_t *src){
1083 int v= *src;
1084 int r;
1085
1086 tprintf("rbsp trailing %X\n", v);
1087
1088 for(r=1; r<9; r++){
1089 if(v&1) return r;
1090 v>>=1;
1091 }
1092 return 0;
1093 }
1094
1095 /**
1096 * idct tranforms the 16 dc values and dequantize them.
1097 * @param qp quantization parameter
1098 */
1099 static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp){
1100 const int qmul= dequant_coeff[qp][0];
1101 #define stride 16
1102 int i;
1103 int temp[16]; //FIXME check if this is a good idea
1104 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
1105 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1106
1107 //memset(block, 64, 2*256);
1108 //return;
1109 for(i=0; i<4; i++){
1110 const int offset= y_offset[i];
1111 const int z0= block[offset+stride*0] + block[offset+stride*4];
1112 const int z1= block[offset+stride*0] - block[offset+stride*4];
1113 const int z2= block[offset+stride*1] - block[offset+stride*5];
1114 const int z3= block[offset+stride*1] + block[offset+stride*5];
1115
1116 temp[4*i+0]= z0+z3;
1117 temp[4*i+1]= z1+z2;
1118 temp[4*i+2]= z1-z2;
1119 temp[4*i+3]= z0-z3;
1120 }
1121
1122 for(i=0; i<4; i++){
1123 const int offset= x_offset[i];
1124 const int z0= temp[4*0+i] + temp[4*2+i];
1125 const int z1= temp[4*0+i] - temp[4*2+i];
1126 const int z2= temp[4*1+i] - temp[4*3+i];
1127 const int z3= temp[4*1+i] + temp[4*3+i];
1128
1129 block[stride*0 +offset]= ((z0 + z3)*qmul + 2)>>2; //FIXME think about merging this into decode_resdual
1130 block[stride*2 +offset]= ((z1 + z2)*qmul + 2)>>2;
1131 block[stride*8 +offset]= ((z1 - z2)*qmul + 2)>>2;
1132 block[stride*10+offset]= ((z0 - z3)*qmul + 2)>>2;
1133 }
1134 }
1135
1136 /**
1137 * dct tranforms the 16 dc values.
1138 * @param qp quantization parameter ??? FIXME
1139 */
1140 static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
1141 // const int qmul= dequant_coeff[qp][0];
1142 int i;
1143 int temp[16]; //FIXME check if this is a good idea
1144 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
1145 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1146
1147 for(i=0; i<4; i++){
1148 const int offset= y_offset[i];
1149 const int z0= block[offset+stride*0] + block[offset+stride*4];
1150 const int z1= block[offset+stride*0] - block[offset+stride*4];
1151 const int z2= block[offset+stride*1] - block[offset+stride*5];
1152 const int z3= block[offset+stride*1] + block[offset+stride*5];
1153
1154 temp[4*i+0]= z0+z3;
1155 temp[4*i+1]= z1+z2;
1156 temp[4*i+2]= z1-z2;
1157 temp[4*i+3]= z0-z3;
1158 }
1159
1160 for(i=0; i<4; i++){
1161 const int offset= x_offset[i];
1162 const int z0= temp[4*0+i] + temp[4*2+i];
1163 const int z1= temp[4*0+i] - temp[4*2+i];
1164 const int z2= temp[4*1+i] - temp[4*3+i];
1165 const int z3= temp[4*1+i] + temp[4*3+i];
1166
1167 block[stride*0 +offset]= (z0 + z3)>>1;
1168 block[stride*2 +offset]= (z1 + z2)>>1;
1169 block[stride*8 +offset]= (z1 - z2)>>1;
1170 block[stride*10+offset]= (z0 - z3)>>1;
1171 }
1172 }
1173 #undef xStride
1174 #undef stride
1175
1176 static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp){
1177 const int qmul= dequant_coeff[qp][0];
1178 const int stride= 16*2;
1179 const int xStride= 16;
1180 int a,b,c,d,e;
1181
1182 a= block[stride*0 + xStride*0];
1183 b= block[stride*0 + xStride*1];
1184 c= block[stride*1 + xStride*0];
1185 d= block[stride*1 + xStride*1];
1186
1187 e= a-b;
1188 a= a+b;
1189 b= c-d;
1190 c= c+d;
1191
1192 block[stride*0 + xStride*0]= ((a+c)*qmul + 0)>>1;
1193 block[stride*0 + xStride*1]= ((e+b)*qmul + 0)>>1;
1194 block[stride*1 + xStride*0]= ((a-c)*qmul + 0)>>1;
1195 block[stride*1 + xStride*1]= ((e-b)*qmul + 0)>>1;
1196 }
1197
1198 static void chroma_dc_dct_c(DCTELEM *block){
1199 const int stride= 16*2;
1200 const int xStride= 16;
1201 int a,b,c,d,e;
1202
1203 a= block[stride*0 + xStride*0];
1204 b= block[stride*0 + xStride*1];
1205 c= block[stride*1 + xStride*0];
1206 d= block[stride*1 + xStride*1];
1207
1208 e= a-b;
1209 a= a+b;
1210 b= c-d;
1211 c= c+d;
1212
1213 block[stride*0 + xStride*0]= (a+c);
1214 block[stride*0 + xStride*1]= (e+b);
1215 block[stride*1 + xStride*0]= (a-c);
1216 block[stride*1 + xStride*1]= (e-b);
1217 }
1218
1219 /**
1220 * gets the chroma qp.
1221 */
1222 static inline int get_chroma_qp(H264Context *h, int qscale){
1223
1224 return chroma_qp[clip(qscale + h->pps.chroma_qp_index_offset, 0, 51)];
1225 }
1226
1227
1228 /**
1229 *
1230 */
1231 static void h264_add_idct_c(uint8_t *dst, DCTELEM *block, int stride){
1232 int i;
1233 uint8_t *cm = cropTbl + MAX_NEG_CROP;
1234
1235 block[0] += 32;
1236 #if 1
1237 for(i=0; i<4; i++){
1238 const int z0= block[i + 4*0] + block[i + 4*2];
1239 const int z1= block[i + 4*0] - block[i + 4*2];
1240 const int z2= (block[i + 4*1]>>1) - block[i + 4*3];
1241 const int z3= block[i + 4*1] + (block[i + 4*3]>>1);
1242
1243 block[i + 4*0]= z0 + z3;
1244 block[i + 4*1]= z1 + z2;
1245 block[i + 4*2]= z1 - z2;
1246 block[i + 4*3]= z0 - z3;
1247 }
1248
1249 for(i=0; i<4; i++){
1250 const int z0= block[0 + 4*i] + block[2 + 4*i];
1251 const int z1= block[0 + 4*i] - block[2 + 4*i];
1252 const int z2= (block[1 + 4*i]>>1) - block[3 + 4*i];
1253 const int z3= block[1 + 4*i] + (block[3 + 4*i]>>1);
1254
1255 dst[0 + i*stride]= cm[ dst[0 + i*stride] + ((z0 + z3) >> 6) ];
1256 dst[1 + i*stride]= cm[ dst[1 + i*stride] + ((z1 + z2) >> 6) ];
1257 dst[2 + i*stride]= cm[ dst[2 + i*stride] + ((z1 - z2) >> 6) ];
1258 dst[3 + i*stride]= cm[ dst[3 + i*stride] + ((z0 - z3) >> 6) ];
1259 }
1260 #else
1261 for(i=0; i<4; i++){
1262 const int z0= block[0 + 4*i] + block[2 + 4*i];
1263 const int z1= block[0 + 4*i] - block[2 + 4*i];
1264 const int z2= (block[1 + 4*i]>>1) - block[3 + 4*i];
1265 const int z3= block[1 + 4*i] + (block[3 + 4*i]>>1);
1266
1267 block[0 + 4*i]= z0 + z3;
1268 block[1 + 4*i]= z1 + z2;
1269 block[2 + 4*i]= z1 - z2;
1270 block[3 + 4*i]= z0 - z3;
1271 }
1272
1273 for(i=0; i<4; i++){
1274 const int z0= block[i + 4*0] + block[i + 4*2];
1275 const int z1= block[i + 4*0] - block[i + 4*2];
1276 const int z2= (block[i + 4*1]>>1) - block[i + 4*3];
1277 const int z3= block[i + 4*1] + (block[i + 4*3]>>1);
1278
1279 dst[i + 0*stride]= cm[ dst[i + 0*stride] + ((z0 + z3) >> 6) ];
1280 dst[i + 1*stride]= cm[ dst[i + 1*stride] + ((z1 + z2) >> 6) ];
1281 dst[i + 2*stride]= cm[ dst[i + 2*stride] + ((z1 - z2) >> 6) ];
1282 dst[i + 3*stride]= cm[ dst[i + 3*stride] + ((z0 - z3) >> 6) ];
1283 }
1284 #endif
1285 }
1286
1287 static void h264_diff_dct_c(DCTELEM *block, uint8_t *src1, uint8_t *src2, int stride){
1288 int i;
1289 //FIXME try int temp instead of block
1290
1291 for(i=0; i<4; i++){
1292 const int d0= src1[0 + i*stride] - src2[0 + i*stride];
1293 const int d1= src1[1 + i*stride] - src2[1 + i*stride];
1294 const int d2= src1[2 + i*stride] - src2[2 + i*stride];
1295 const int d3= src1[3 + i*stride] - src2[3 + i*stride];
1296 const int z0= d0 + d3;
1297 const int z3= d0 - d3;
1298 const int z1= d1 + d2;
1299 const int z2= d1 - d2;
1300
1301 block[0 + 4*i]= z0 + z1;
1302 block[1 + 4*i]= 2*z3 + z2;
1303 block[2 + 4*i]= z0 - z1;
1304 block[3 + 4*i]= z3 - 2*z2;
1305 }
1306
1307 for(i=0; i<4; i++){
1308 const int z0= block[0*4 + i] + block[3*4 + i];
1309 const int z3= block[0*4 + i] - block[3*4 + i];
1310 const int z1= block[1*4 + i] + block[2*4 + i];
1311 const int z2= block[1*4 + i] - block[2*4 + i];
1312
1313 block[0*4 + i]= z0 + z1;
1314 block[1*4 + i]= 2*z3 + z2;
1315 block[2*4 + i]= z0 - z1;
1316 block[3*4 + i]= z3 - 2*z2;
1317 }
1318 }
1319
1320 //FIXME need to check that this doesnt overflow signed 32 bit for low qp, iam not sure, its very close
1321 //FIXME check that gcc inlines this (and optimizes intra & seperate_dc stuff away)
1322 static inline int quantize_c(DCTELEM *block, uint8_t *scantable, int qscale, int intra, int seperate_dc){
1323 int i;
1324 const int * const quant_table= quant_coeff[qscale];
1325 const int bias= intra ? (1<<QUANT_SHIFT)/3 : (1<<QUANT_SHIFT)/6;
1326 const unsigned int threshold1= (1<<QUANT_SHIFT) - bias - 1;
1327 const unsigned int threshold2= (threshold1<<1);
1328 int last_non_zero;
1329
1330 if(seperate_dc){
1331 if(qscale<=18){
1332 //avoid overflows
1333 const int dc_bias= intra ? (1<<(QUANT_SHIFT-2))/3 : (1<<(QUANT_SHIFT-2))/6;
1334 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT-2)) - dc_bias - 1;
1335 const unsigned int dc_threshold2= (dc_threshold1<<1);
1336
1337 int level= block[0]*quant_coeff[qscale+18][0];
1338 if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1339 if(level>0){
1340 level= (dc_bias + level)>>(QUANT_SHIFT-2);
1341 block[0]= level;
1342 }else{
1343 level= (dc_bias - level)>>(QUANT_SHIFT-2);
1344 block[0]= -level;
1345 }
1346 // last_non_zero = i;
1347 }else{
1348 block[0]=0;
1349 }
1350 }else{
1351 const int dc_bias= intra ? (1<<(QUANT_SHIFT+1))/3 : (1<<(QUANT_SHIFT+1))/6;
1352 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT+1)) - dc_bias - 1;
1353 const unsigned int dc_threshold2= (dc_threshold1<<1);
1354
1355 int level= block[0]*quant_table[0];
1356 if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1357 if(level>0){
1358 level= (dc_bias + level)>>(QUANT_SHIFT+1);
1359 block[0]= level;
1360 }else{
1361 level= (dc_bias - level)>>(QUANT_SHIFT+1);
1362 block[0]= -level;
1363 }
1364 // last_non_zero = i;
1365 }else{
1366 block[0]=0;
1367 }
1368 }
1369 last_non_zero= 0;
1370 i=1;
1371 }else{
1372 last_non_zero= -1;
1373 i=0;
1374 }
1375
1376 for(; i<16; i++){
1377 const int j= scantable[i];
1378 int level= block[j]*quant_table[j];
1379
1380 // if( bias+level >= (1<<(QMAT_SHIFT - 3))
1381 // || bias-level >= (1<<(QMAT_SHIFT - 3))){
1382 if(((unsigned)(level+threshold1))>threshold2){
1383 if(level>0){
1384 level= (bias + level)>>QUANT_SHIFT;
1385 block[j]= level;
1386 }else{
1387 level= (bias - level)>>QUANT_SHIFT;
1388 block[j]= -level;
1389 }
1390 last_non_zero = i;
1391 }else{
1392 block[j]=0;
1393 }
1394 }
1395
1396 return last_non_zero;
1397 }
1398
1399 static void pred4x4_vertical_c(uint8_t *src, uint8_t *topright, int stride){
1400 const uint32_t a= ((uint32_t*)(src-stride))[0];
1401 ((uint32_t*)(src+0*stride))[0]= a;
1402 ((uint32_t*)(src+1*stride))[0]= a;
1403 ((uint32_t*)(src+2*stride))[0]= a;
1404 ((uint32_t*)(src+3*stride))[0]= a;
1405 }
1406
1407 static void pred4x4_horizontal_c(uint8_t *src, uint8_t *topright, int stride){
1408 ((uint32_t*)(src+0*stride))[0]= src[-1+0*stride]*0x01010101;
1409 ((uint32_t*)(src+1*stride))[0]= src[-1+1*stride]*0x01010101;
1410 ((uint32_t*)(src+2*stride))[0]= src[-1+2*stride]*0x01010101;
1411 ((uint32_t*)(src+3*stride))[0]= src[-1+3*stride]*0x01010101;
1412 }
1413
1414 static void pred4x4_dc_c(uint8_t *src, uint8_t *topright, int stride){
1415 const int dc= ( src[-stride] + src[1-stride] + src[2-stride] + src[3-stride]
1416 + src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 4) >>3;
1417
1418 ((uint32_t*)(src+0*stride))[0]=
1419 ((uint32_t*)(src+1*stride))[0]=
1420 ((uint32_t*)(src+2*stride))[0]=
1421 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1422 }
1423
1424 static void pred4x4_left_dc_c(uint8_t *src, uint8_t *topright, int stride){
1425 const int dc= ( src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 2) >>2;
1426
1427 ((uint32_t*)(src+0*stride))[0]=
1428 ((uint32_t*)(src+1*stride))[0]=
1429 ((uint32_t*)(src+2*stride))[0]=
1430 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1431 }
1432
1433 static void pred4x4_top_dc_c(uint8_t *src, uint8_t *topright, int stride){
1434 const int dc= ( src[-stride] + src[1-stride] + src[2-stride] + src[3-stride] + 2) >>2;
1435
1436 ((uint32_t*)(src+0*stride))[0]=
1437 ((uint32_t*)(src+1*stride))[0]=
1438 ((uint32_t*)(src+2*stride))[0]=
1439 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1440 }
1441
1442 static void pred4x4_128_dc_c(uint8_t *src, uint8_t *topright, int stride){
1443 ((uint32_t*)(src+0*stride))[0]=
1444 ((uint32_t*)(src+1*stride))[0]=
1445 ((uint32_t*)(src+2*stride))[0]=
1446 ((uint32_t*)(src+3*stride))[0]= 128U*0x01010101U;
1447 }
1448
1449
1450 #define LOAD_TOP_RIGHT_EDGE\
1451 const int t4= topright[0];\
1452 const int t5= topright[1];\
1453 const int t6= topright[2];\
1454 const int t7= topright[3];\
1455
1456 #define LOAD_LEFT_EDGE\
1457 const int l0= src[-1+0*stride];\
1458 const int l1= src[-1+1*stride];\
1459 const int l2= src[-1+2*stride];\
1460 const int l3= src[-1+3*stride];\
1461
1462 #define LOAD_TOP_EDGE\
1463 const int t0= src[ 0-1*stride];\
1464 const int t1= src[ 1-1*stride];\
1465 const int t2= src[ 2-1*stride];\
1466 const int t3= src[ 3-1*stride];\
1467
1468 static void pred4x4_down_right_c(uint8_t *src, uint8_t *topright, int stride){
1469 const int lt= src[-1-1*stride];
1470 LOAD_TOP_EDGE
1471 LOAD_LEFT_EDGE
1472
1473 src[0+3*stride]=(l3 + 2*l2 + l1 + 2)>>2;
1474 src[0+2*stride]=
1475 src[1+3*stride]=(l2 + 2*l1 + l0 + 2)>>2;
1476 src[0+1*stride]=
1477 src[1+2*stride]=
1478 src[2+3*stride]=(l1 + 2*l0 + lt + 2)>>2;
1479 src[0+0*stride]=
1480 src[1+1*stride]=
1481 src[2+2*stride]=
1482 src[3+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1483 src[1+0*stride]=
1484 src[2+1*stride]=
1485 src[3+2*stride]=(lt + 2*t0 + t1 + 2)>>2;
1486 src[2+0*stride]=
1487 src[3+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1488 src[3+0*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1489 }
1490
1491 static void pred4x4_down_left_c(uint8_t *src, uint8_t *topright, int stride){
1492 LOAD_TOP_EDGE
1493 LOAD_TOP_RIGHT_EDGE
1494 // LOAD_LEFT_EDGE
1495
1496 src[0+0*stride]=(t0 + t2 + 2*t1 + 2)>>2;
1497 src[1+0*stride]=
1498 src[0+1*stride]=(t1 + t3 + 2*t2 + 2)>>2;
1499 src[2+0*stride]=
1500 src[1+1*stride]=
1501 src[0+2*stride]=(t2 + t4 + 2*t3 + 2)>>2;
1502 src[3+0*stride]=
1503 src[2+1*stride]=
1504 src[1+2*stride]=
1505 src[0+3*stride]=(t3 + t5 + 2*t4 + 2)>>2;
1506 src[3+1*stride]=
1507 src[2+2*stride]=
1508 src[1+3*stride]=(t4 + t6 + 2*t5 + 2)>>2;
1509 src[3+2*stride]=
1510 src[2+3*stride]=(t5 + t7 + 2*t6 + 2)>>2;
1511 src[3+3*stride]=(t6 + 3*t7 + 2)>>2;
1512 }
1513
1514 static void pred4x4_vertical_right_c(uint8_t *src, uint8_t *topright, int stride){
1515 const int lt= src[-1-1*stride];
1516 LOAD_TOP_EDGE
1517 LOAD_LEFT_EDGE
1518 const __attribute__((unused)) int unu= l3;
1519
1520 src[0+0*stride]=
1521 src[1+2*stride]=(lt + t0 + 1)>>1;
1522 src[1+0*stride]=
1523 src[2+2*stride]=(t0 + t1 + 1)>>1;
1524 src[2+0*stride]=
1525 src[3+2*stride]=(t1 + t2 + 1)>>1;
1526 src[3+0*stride]=(t2 + t3 + 1)>>1;
1527 src[0+1*stride]=
1528 src[1+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1529 src[1+1*stride]=
1530 src[2+3*stride]=(lt + 2*t0 + t1 + 2)>>2;
1531 src[2+1*stride]=
1532 src[3+3*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1533 src[3+1*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1534 src[0+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1535 src[0+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1536 }
1537
1538 static void pred4x4_vertical_left_c(uint8_t *src, uint8_t *topright, int stride){
1539 LOAD_TOP_EDGE
1540 LOAD_TOP_RIGHT_EDGE
1541 const __attribute__((unused)) int unu= t7;
1542
1543 src[0+0*stride]=(t0 + t1 + 1)>>1;
1544 src[1+0*stride]=
1545 src[0+2*stride]=(t1 + t2 + 1)>>1;
1546 src[2+0*stride]=
1547 src[1+2*stride]=(t2 + t3 + 1)>>1;
1548 src[3+0*stride]=
1549 src[2+2*stride]=(t3 + t4+ 1)>>1;
1550 src[3+2*stride]=(t4 + t5+ 1)>>1;
1551 src[0+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1552 src[1+1*stride]=
1553 src[0+3*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1554 src[2+1*stride]=
1555 src[1+3*stride]=(t2 + 2*t3 + t4 + 2)>>2;
1556 src[3+1*stride]=
1557 src[2+3*stride]=(t3 + 2*t4 + t5 + 2)>>2;
1558 src[3+3*stride]=(t4 + 2*t5 + t6 + 2)>>2;
1559 }
1560
1561 static void pred4x4_horizontal_up_c(uint8_t *src, uint8_t *topright, int stride){
1562 LOAD_LEFT_EDGE
1563
1564 src[0+0*stride]=(l0 + l1 + 1)>>1;
1565 src[1+0*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1566 src[2+0*stride]=
1567 src[0+1*stride]=(l1 + l2 + 1)>>1;
1568 src[3+0*stride]=
1569 src[1+1*stride]=(l1 + 2*l2 + l3 + 2)>>2;
1570 src[2+1*stride]=
1571 src[0+2*stride]=(l2 + l3 + 1)>>1;
1572 src[3+1*stride]=
1573 src[1+2*stride]=(l2 + 2*l3 + l3 + 2)>>2;
1574 src[3+2*stride]=
1575 src[1+3*stride]=
1576 src[0+3*stride]=
1577 src[2+2*stride]=
1578 src[2+3*stride]=
1579 src[3+3*stride]=l3;
1580 }
1581
1582 static void pred4x4_horizontal_down_c(uint8_t *src, uint8_t *topright, int stride){
1583 const int lt= src[-1-1*stride];
1584 LOAD_TOP_EDGE
1585 LOAD_LEFT_EDGE
1586 const __attribute__((unused)) int unu= t3;
1587
1588 src[0+0*stride]=
1589 src[2+1*stride]=(lt + l0 + 1)>>1;
1590 src[1+0*stride]=
1591 src[3+1*stride]=(l0 + 2*lt + t0 + 2)>>2;
1592 src[2+0*stride]=(lt + 2*t0 + t1 + 2)>>2;
1593 src[3+0*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1594 src[0+1*stride]=
1595 src[2+2*stride]=(l0 + l1 + 1)>>1;
1596 src[1+1*stride]=
1597 src[3+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1598 src[0+2*stride]=
1599 src[2+3*stride]=(l1 + l2+ 1)>>1;
1600 src[1+2*stride]=
1601 src[3+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1602 src[0+3*stride]=(l2 + l3 + 1)>>1;
1603 src[1+3*stride]=(l1 + 2*l2 + l3 + 2)>>2;
1604 }
1605
1606 static void pred16x16_vertical_c(uint8_t *src, int stride){
1607 int i;
1608 const uint32_t a= ((uint32_t*)(src-stride))[0];
1609 const uint32_t b= ((uint32_t*)(src-stride))[1];
1610 const uint32_t c= ((uint32_t*)(src-stride))[2];
1611 const uint32_t d= ((uint32_t*)(src-stride))[3];
1612
1613 for(i=0; i<16; i++){
1614 ((uint32_t*)(src+i*stride))[0]= a;
1615 ((uint32_t*)(src+i*stride))[1]= b;
1616 ((uint32_t*)(src+i*stride))[2]= c;
1617 ((uint32_t*)(src+i*stride))[3]= d;
1618 }
1619 }
1620
1621 static void pred16x16_horizontal_c(uint8_t *src, int stride){
1622 int i;
1623
1624 for(i=0; i<16; i++){
1625 ((uint32_t*)(src+i*stride))[0]=
1626 ((uint32_t*)(src+i*stride))[1]=
1627 ((uint32_t*)(src+i*stride))[2]=
1628 ((uint32_t*)(src+i*stride))[3]= src[-1+i*stride]*0x01010101;
1629 }
1630 }
1631
1632 static void pred16x16_dc_c(uint8_t *src, int stride){
1633 int i, dc=0;
1634
1635 for(i=0;i<16; i++){
1636 dc+= src[-1+i*stride];
1637 }
1638
1639 for(i=0;i<16; i++){
1640 dc+= src[i-stride];
1641 }
1642
1643 dc= 0x01010101*((dc + 16)>>5);
1644
1645 for(i=0; i<16; i++){
1646 ((uint32_t*)(src+i*stride))[0]=
1647 ((uint32_t*)(src+i*stride))[1]=
1648 ((uint32_t*)(src+i*stride))[2]=
1649 ((uint32_t*)(src+i*stride))[3]= dc;
1650 }
1651 }
1652
1653 static void pred16x16_left_dc_c(uint8_t *src, int stride){
1654 int i, dc=0;
1655
1656 for(i=0;i<16; i++){
1657 dc+= src[-1+i*stride];
1658 }
1659
1660 dc= 0x01010101*((dc + 8)>>4);
1661
1662 for(i=0; i<16; i++){
1663 ((uint32_t*)(src+i*stride))[0]=
1664 ((uint32_t*)(src+i*stride))[1]=
1665 ((uint32_t*)(src+i*stride))[2]=
1666 ((uint32_t*)(src+i*stride))[3]= dc;
1667 }
1668 }
1669
1670 static void pred16x16_top_dc_c(uint8_t *src, int stride){
1671 int i, dc=0;
1672
1673 for(i=0;i<16; i++){
1674 dc+= src[i-stride];
1675 }
1676 dc= 0x01010101*((dc + 8)>>4);
1677
1678 for(i=0; i<16; i++){
1679 ((uint32_t*)(src+i*stride))[0]=
1680 ((uint32_t*)(src+i*stride))[1]=
1681 ((uint32_t*)(src+i*stride))[2]=
1682 ((uint32_t*)(src+i*stride))[3]= dc;
1683 }
1684 }
1685
1686 static void pred16x16_128_dc_c(uint8_t *src, int stride){
1687 int i;
1688
1689 for(i=0; i<16; i++){
1690 ((uint32_t*)(src+i*stride))[0]=
1691 ((uint32_t*)(src+i*stride))[1]=
1692 ((uint32_t*)(src+i*stride))[2]=
1693 ((uint32_t*)(src+i*stride))[3]= 0x01010101U*128U;
1694 }
1695 }
1696
1697 static inline void pred16x16_plane_compat_c(uint8_t *src, int stride, const int svq3){
1698 int i, j, k;
1699 int a;
1700 uint8_t *cm = cropTbl + MAX_NEG_CROP;
1701 const uint8_t * const src0 = src+7-stride;
1702 const uint8_t *src1 = src+8*stride-1;
1703 const uint8_t *src2 = src1-2*stride; // == src+6*stride-1;
1704 int H = src0[1] - src0[-1];
1705 int V = src1[0] - src2[ 0];
1706 for(k=2; k<=8; ++k) {
1707 src1 += stride; src2 -= stride;
1708 H += k*(src0[k] - src0[-k]);
1709 V += k*(src1[0] - src2[ 0]);
1710 }
1711 if(svq3){
1712 H = ( 5*(H/4) ) / 16;
1713 V = ( 5*(V/4) ) / 16;
1714
1715 /* required for 100% accuracy */
1716 i = H; H = V; V = i;
1717 }else{
1718 H = ( 5*H+32 ) >> 6;
1719 V = ( 5*V+32 ) >> 6;
1720 }
1721
1722 a = 16*(src1[0] + src2[16] + 1) - 7*(V+H);
1723 for(j=16; j>0; --j) {
1724 int b = a;
1725 a += V;
1726 for(i=-16; i<0; i+=4) {
1727 src[16+i] = cm[ (b ) >> 5 ];
1728 src[17+i] = cm[ (b+ H) >> 5 ];
1729 src[18+i] = cm[ (b+2*H) >> 5 ];
1730 src[19+i] = cm[ (b+3*H) >> 5 ];
1731 b += 4*H;
1732 }
1733 src += stride;
1734 }
1735 }
1736
1737 static void pred16x16_plane_c(uint8_t *src, int stride){
1738 pred16x16_plane_compat_c(src, stride, 0);
1739 }
1740
1741 static void pred8x8_vertical_c(uint8_t *src, int stride){
1742 int i;
1743 const uint32_t a= ((uint32_t*)(src-stride))[0];
1744 const uint32_t b= ((uint32_t*)(src-stride))[1];
1745
1746 for(i=0; i<8; i++){
1747 ((uint32_t*)(src+i*stride))[0]= a;
1748 ((uint32_t*)(src+i*stride))[1]= b;
1749 }
1750 }
1751
1752 static void pred8x8_horizontal_c(uint8_t *src, int stride){
1753 int i;
1754
1755 for(i=0; i<8; i++){
1756 ((uint32_t*)(src+i*stride))[0]=
1757 ((uint32_t*)(src+i*stride))[1]= src[-1+i*stride]*0x01010101;
1758 }
1759 }
1760
1761 static void pred8x8_128_dc_c(uint8_t *src, int stride){
1762 int i;
1763
1764 for(i=0; i<4; i++){
1765 ((uint32_t*)(src+i*stride))[0]=
1766 ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
1767 }
1768 for(i=4; i<8; i++){
1769 ((uint32_t*)(src+i*stride))[0]=
1770 ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
1771 }
1772 }
1773
1774 static void pred8x8_left_dc_c(uint8_t *src, int stride){
1775 int i;
1776 int dc0, dc2;
1777
1778 dc0=dc2=0;
1779 for(i=0;i<4; i++){
1780 dc0+= src[-1+i*stride];
1781 dc2+= src[-1+(i+4)*stride];
1782 }
1783 dc0= 0x01010101*((dc0 + 2)>>2);
1784 dc2= 0x01010101*((dc2 + 2)>>2);
1785
1786 for(i=0; i<4; i++){
1787 ((uint32_t*)(src+i*stride))[0]=
1788 ((uint32_t*)(src+i*stride))[1]= dc0;
1789 }
1790 for(i=4; i<8; i++){
1791 ((uint32_t*)(src+i*stride))[0]=
1792 ((uint32_t*)(src+i*stride))[1]= dc2;
1793 }
1794 }
1795
1796 static void pred8x8_top_dc_c(uint8_t *src, int stride){
1797 int i;
1798 int dc0, dc1;
1799
1800 dc0=dc1=0;
1801 for(i=0;i<4; i++){
1802 dc0+= src[i-stride];
1803 dc1+= src[4+i-stride];
1804 }
1805 dc0= 0x01010101*((dc0 + 2)>>2);
1806 dc1= 0x01010101*((dc1 + 2)>>2);
1807
1808 for(i=0; i<4; i++){
1809 ((uint32_t*)(src+i*stride))[0]= dc0;
1810 ((uint32_t*)(src+i*stride))[1]= dc1;
1811 }
1812 for(i=4; i<8; i++){
1813 ((uint32_t*)(src+i*stride))[0]= dc0;
1814 ((uint32_t*)(src+i*stride))[1]= dc1;
1815 }
1816 }
1817
1818
1819 static void pred8x8_dc_c(uint8_t *src, int stride){
1820 int i;
1821 int dc0, dc1, dc2, dc3;
1822
1823 dc0=dc1=dc2=0;
1824 for(i=0;i<4; i++){
1825 dc0+= src[-1+i*stride] + src[i-stride];
1826 dc1+= src[4+i-stride];
1827 dc2+= src[-1+(i+4)*stride];
1828 }
1829 dc3= 0x01010101*((dc1 + dc2 + 4)>>3);
1830 dc0= 0x01010101*((dc0 + 4)>>3);
1831 dc1= 0x01010101*((dc1 + 2)>>2);
1832 dc2= 0x01010101*((dc2 + 2)>>2);
1833
1834 for(i=0; i<4; i++){
1835 ((uint32_t*)(src+i*stride))[0]= dc0;
1836 ((uint32_t*)(src+i*stride))[1]= dc1;
1837 }
1838 for(i=4; i<8; i++){
1839 ((uint32_t*)(src+i*stride))[0]= dc2;
1840 ((uint32_t*)(src+i*stride))[1]= dc3;
1841 }
1842 }
1843
1844 static void pred8x8_plane_c(uint8_t *src, int stride){
1845 int j, k;
1846 int a;
1847 uint8_t *cm = cropTbl + MAX_NEG_CROP;
1848 const uint8_t * const src0 = src+3-stride;
1849 const uint8_t *src1 = src+4*stride-1;
1850 const uint8_t *src2 = src1-2*stride; // == src+2*stride-1;
1851 int H = src0[1] - src0[-1];
1852 int V = src1[0] - src2[ 0];
1853 for(k=2; k<=4; ++k) {
1854 src1 += stride; src2 -= stride;
1855 H += k*(src0[k] - src0[-k]);
1856 V += k*(src1[0] - src2[ 0]);
1857 }
1858 H = ( 17*H+16 ) >> 5;
1859 V = ( 17*V+16 ) >> 5;
1860
1861 a = 16*(src1[0] + src2[8]+1) - 3*(V+H);
1862 for(j=8; j>0; --j) {
1863 int b = a;
1864 a += V;
1865 src[0] = cm[ (b ) >> 5 ];
1866 src[1] = cm[ (b+ H) >> 5 ];
1867 src[2] = cm[ (b+2*H) >> 5 ];
1868 src[3] = cm[ (b+3*H) >> 5 ];
1869 src[4] = cm[ (b+4*H) >> 5 ];
1870 src[5] = cm[ (b+5*H) >> 5 ];
1871 src[6] = cm[ (b+6*H) >> 5 ];
1872 src[7] = cm[ (b+7*H) >> 5 ];
1873 src += stride;
1874 }
1875 }
1876
1877 static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
1878 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1879 int src_x_offset, int src_y_offset,
1880 qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
1881 MpegEncContext * const s = &h->s;
1882 const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
1883 const int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
1884 const int luma_xy= (mx&3) + ((my&3)<<2);
1885 uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*s->linesize;
1886 uint8_t * src_cb= pic->data[1] + (mx>>3) + (my>>3)*s->uvlinesize;
1887 uint8_t * src_cr= pic->data[2] + (mx>>3) + (my>>3)*s->uvlinesize;
1888 int extra_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16; //FIXME increase edge?, IMHO not worth it
1889 int extra_height= extra_width;
1890 int emu=0;
1891 const int full_mx= mx>>2;
1892 const int full_my= my>>2;
1893
1894 assert(pic->data[0]);
1895
1896 if(mx&7) extra_width -= 3;
1897 if(my&7) extra_height -= 3;
1898
1899 if( full_mx < 0-extra_width
1900 || full_my < 0-extra_height
1901 || full_mx + 16/*FIXME*/ > s->width + extra_width
1902 || full_my + 16/*FIXME*/ > s->height + extra_height){
1903 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);
1904 src_y= s->edge_emu_buffer + 2 + 2*s->linesize;
1905 emu=1;
1906 }
1907
1908 qpix_op[luma_xy](dest_y, src_y, s->linesize); //FIXME try variable height perhaps?
1909 if(!square){
1910 qpix_op[luma_xy](dest_y + delta, src_y + delta, s->linesize);
1911 }
1912
1913 if(s->flags&CODEC_FLAG_GRAY) return;
1914
1915 if(emu){
1916 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);
1917 src_cb= s->edge_emu_buffer;
1918 }
1919 chroma_op(dest_cb, src_cb, s->uvlinesize, chroma_height, mx&7, my&7);
1920
1921 if(emu){
1922 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);
1923 src_cr= s->edge_emu_buffer;
1924 }
1925 chroma_op(dest_cr, src_cr, s->uvlinesize, chroma_height, mx&7, my&7);
1926 }
1927
1928 static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
1929 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1930 int x_offset, int y_offset,
1931 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
1932 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
1933 int list0, int list1){
1934 MpegEncContext * const s = &h->s;
1935 qpel_mc_func *qpix_op= qpix_put;
1936 h264_chroma_mc_func chroma_op= chroma_put;
1937
1938 dest_y += 2*x_offset + 2*y_offset*s-> linesize;
1939 dest_cb += x_offset + y_offset*s->uvlinesize;
1940 dest_cr += x_offset + y_offset*s->uvlinesize;
1941 x_offset += 8*s->mb_x;
1942 y_offset += 8*s->mb_y;
1943
1944 if(list0){
1945 Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
1946 mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
1947 dest_y, dest_cb, dest_cr, x_offset, y_offset,
1948 qpix_op, chroma_op);
1949
1950 qpix_op= qpix_avg;
1951 chroma_op= chroma_avg;
1952 }
1953
1954 if(list1){
1955 Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
1956 mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
1957 dest_y, dest_cb, dest_cr, x_offset, y_offset,
1958 qpix_op, chroma_op);
1959 }
1960 }
1961
1962 static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1963 qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
1964 qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg)){
1965 MpegEncContext * const s = &h->s;
1966 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
1967 const int mb_type= s->current_picture.mb_type[mb_xy];
1968
1969 assert(IS_INTER(mb_type));
1970
1971 if(IS_16X16(mb_type)){
1972 mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
1973 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
1974 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1975 }else if(IS_16X8(mb_type)){
1976 mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
1977 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
1978 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1979 mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
1980 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
1981 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
1982 }else if(IS_8X16(mb_type)){
1983 mc_part(h, 0, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 0, 0,
1984 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
1985 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1986 mc_part(h, 4, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 4, 0,
1987 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
1988 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
1989 }else{
1990 int i;
1991
1992 assert(IS_8X8(mb_type));
1993
1994 for(i=0; i<4; i++){
1995 const int sub_mb_type= h->sub_mb_type[i];
1996 const int n= 4*i;
1997 int x_offset= (i&1)<<2;
1998 int y_offset= (i&2)<<1;
1999
2000 if(IS_SUB_8X8(sub_mb_type)){
2001 mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2002 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2003 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2004 }else if(IS_SUB_8X4(sub_mb_type)){
2005 mc_part(h, n , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2006 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2007 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2008 mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
2009 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2010 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2011 }else if(IS_SUB_4X8(sub_mb_type)){
2012 mc_part(h, n , 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2013 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2014 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2015 mc_part(h, n+1, 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
2016 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2017 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2018 }else{
2019 int j;
2020 assert(IS_SUB_4X4(sub_mb_type));
2021 for(j=0; j<4; j++){
2022 int sub_x_offset= x_offset + 2*(j&1);
2023 int sub_y_offset= y_offset + (j&2);
2024 mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
2025 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2026 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2027 }
2028 }
2029 }
2030 }
2031 }
2032
2033 static void decode_init_vlc(H264Context *h){
2034 static int done = 0;
2035
2036 if (!done) {
2037 int i;
2038 done = 1;
2039
2040 init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5,
2041 &chroma_dc_coeff_token_len [0], 1, 1,
2042 &chroma_dc_coeff_token_bits[0], 1, 1);
2043
2044 for(i=0; i<4; i++){
2045 init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17,
2046 &coeff_token_len [i][0], 1, 1,
2047 &coeff_token_bits[i][0], 1, 1);
2048 }
2049
2050 for(i=0; i<3; i++){
2051 init_vlc(&chroma_dc_total_zeros_vlc[i], CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
2052 &chroma_dc_total_zeros_len [i][0], 1, 1,
2053 &chroma_dc_total_zeros_bits[i][0], 1, 1);
2054 }
2055 for(i=0; i<15; i++){
2056 init_vlc(&total_zeros_vlc[i], TOTAL_ZEROS_VLC_BITS, 16,
2057 &total_zeros_len [i][0], 1, 1,
2058 &total_zeros_bits[i][0], 1, 1);
2059 }
2060
2061 for(i=0; i<6; i++){
2062 init_vlc(&run_vlc[i], RUN_VLC_BITS, 7,
2063 &run_len [i][0], 1, 1,
2064 &run_bits[i][0], 1, 1);
2065 }
2066 init_vlc(&run7_vlc, RUN7_VLC_BITS, 16,
2067 &run_len [6][0], 1, 1,
2068 &run_bits[6][0], 1, 1);
2069 }
2070 }
2071
2072 /**
2073 * Sets the intra prediction function pointers.
2074 */
2075 static void init_pred_ptrs(H264Context *h){
2076 // MpegEncContext * const s = &h->s;
2077
2078 h->pred4x4[VERT_PRED ]= pred4x4_vertical_c;
2079 h->pred4x4[HOR_PRED ]= pred4x4_horizontal_c;
2080 h->pred4x4[DC_PRED ]= pred4x4_dc_c;
2081 h->pred4x4[DIAG_DOWN_LEFT_PRED ]= pred4x4_down_left_c;
2082 h->pred4x4[DIAG_DOWN_RIGHT_PRED]= pred4x4_down_right_c;
2083 h->pred4x4[VERT_RIGHT_PRED ]= pred4x4_vertical_right_c;
2084 h->pred4x4[HOR_DOWN_PRED ]= pred4x4_horizontal_down_c;
2085 h->pred4x4[VERT_LEFT_PRED ]= pred4x4_vertical_left_c;
2086 h->pred4x4[HOR_UP_PRED ]= pred4x4_horizontal_up_c;
2087 h->pred4x4[LEFT_DC_PRED ]= pred4x4_left_dc_c;
2088 h->pred4x4[TOP_DC_PRED ]= pred4x4_top_dc_c;
2089 h->pred4x4[DC_128_PRED ]= pred4x4_128_dc_c;
2090
2091 h->pred8x8[DC_PRED8x8 ]= pred8x8_dc_c;
2092 h->pred8x8[VERT_PRED8x8 ]= pred8x8_vertical_c;
2093 h->pred8x8[HOR_PRED8x8 ]= pred8x8_horizontal_c;
2094 h->pred8x8[PLANE_PRED8x8 ]= pred8x8_plane_c;
2095 h->pred8x8[LEFT_DC_PRED8x8]= pred8x8_left_dc_c;
2096 h->pred8x8[TOP_DC_PRED8x8 ]= pred8x8_top_dc_c;
2097 h->pred8x8[DC_128_PRED8x8 ]= pred8x8_128_dc_c;
2098
2099 h->pred16x16[DC_PRED8x8 ]= pred16x16_dc_c;
2100 h->pred16x16[VERT_PRED8x8 ]= pred16x16_vertical_c;
2101 h->pred16x16[HOR_PRED8x8 ]= pred16x16_horizontal_c;
2102 h->pred16x16[PLANE_PRED8x8 ]= pred16x16_plane_c;
2103 h->pred16x16[LEFT_DC_PRED8x8]= pred16x16_left_dc_c;
2104 h->pred16x16[TOP_DC_PRED8x8 ]= pred16x16_top_dc_c;
2105 h->pred16x16[DC_128_PRED8x8 ]= pred16x16_128_dc_c;
2106 }
2107
2108 static void free_tables(H264Context *h){
2109 av_freep(&h->intra4x4_pred_mode);
2110 av_freep(&h->non_zero_count);
2111 av_freep(&h->slice_table_base);
2112 h->slice_table= NULL;
2113
2114 av_freep(&h->mb2b_xy);
2115 av_freep(&h->mb2b8_xy);
2116 }
2117
2118 /**
2119 * allocates tables.
2120 * needs widzh/height
2121 */
2122 static int alloc_tables(H264Context *h){
2123 MpegEncContext * const s = &h->s;
2124 const int big_mb_num= s->mb_stride * (s->mb_height+1);
2125 int x,y;
2126
2127 CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8 * sizeof(uint8_t))
2128 CHECKED_ALLOCZ(h->non_zero_count , big_mb_num * 16 * sizeof(uint8_t))
2129 CHECKED_ALLOCZ(h->slice_table_base , big_mb_num * sizeof(uint8_t))
2130
2131 memset(h->slice_table_base, -1, big_mb_num * sizeof(uint8_t));
2132 h->slice_table= h->slice_table_base + s->mb_stride + 1;
2133
2134 CHECKED_ALLOCZ(h->mb2b_xy , big_mb_num * sizeof(uint16_t));
2135 CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint16_t));
2136 for(y=0; y<s->mb_height; y++){
2137 for(x=0; x<s->mb_width; x++){
2138 const int mb_xy= x + y*s->mb_stride;
2139 const int b_xy = 4*x + 4*y*h->b_stride;
2140 const int b8_xy= 2*x + 2*y*h->b8_stride;
2141
2142 h->mb2b_xy [mb_xy]= b_xy;
2143 h->mb2b8_xy[mb_xy]= b8_xy;
2144 }
2145 }
2146
2147 return 0;
2148 fail:
2149 free_tables(h);
2150 return -1;
2151 }
2152
2153 static void common_init(H264Context *h){
2154 MpegEncContext * const s = &h->s;
2155
2156 s->width = s->avctx->width;
2157 s->height = s->avctx->height;
2158 s->codec_id= s->avctx->codec->id;
2159
2160 init_pred_ptrs(h);
2161
2162 s->decode=1; //FIXME
2163 }
2164
2165 static int decode_init(AVCodecContext *avctx){
2166 H264Context *h= avctx->priv_data;
2167 MpegEncContext * const s = &h->s;
2168
2169 s->avctx = avctx;
2170 common_init(h);
2171
2172 s->out_format = FMT_H264;
2173 s->workaround_bugs= avctx->workaround_bugs;
2174
2175 // set defaults
2176 s->progressive_sequence=1;
2177 // s->decode_mb= ff_h263_decode_mb;
2178 s->low_delay= 1;
2179 avctx->pix_fmt= PIX_FMT_YUV420P;
2180
2181 decode_init_vlc(h);
2182
2183 return 0;
2184 }
2185
2186 static void frame_start(H264Context *h){
2187 MpegEncContext * const s = &h->s;
2188 int i;
2189
2190 MPV_frame_start(s, s->avctx);
2191 ff_er_frame_start(s);
2192 h->mmco_index=0;
2193
2194 assert(s->linesize && s->uvlinesize);
2195
2196 for(i=0; i<16; i++){
2197 h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
2198 h->chroma_subblock_offset[i]= 2*((scan8[i] - scan8[0])&7) + 2*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2199 }
2200 for(i=0; i<4; i++){
2201 h->block_offset[16+i]=
2202 h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2203 }
2204
2205 // s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
2206 }
2207
2208 static void hl_decode_mb(H264Context *h){
2209 MpegEncContext * const s = &h->s;
2210 const int mb_x= s->mb_x;
2211 const int mb_y= s->mb_y;
2212 const int mb_xy= mb_x + mb_y*s->mb_stride;
2213 const int mb_type= s->current_picture.mb_type[mb_xy];
2214 uint8_t *dest_y, *dest_cb, *dest_cr;
2215 int linesize, uvlinesize /*dct_offset*/;
2216 int i;
2217
2218 if(!s->decode)
2219 return;
2220
2221 if(s->mb_skiped){
2222 }
2223
2224 dest_y = s->current_picture.data[0] + (mb_y * 16* s->linesize ) + mb_x * 16;
2225 dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2226 dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2227
2228 if (h->mb_field_decoding_flag) {
2229 linesize = s->linesize * 2;
2230 uvlinesize = s->uvlinesize * 2;
2231 if(mb_y&1){ //FIXME move out of this func?
2232 dest_y -= s->linesize*15;
2233 dest_cb-= s->linesize*7;
2234 dest_cr-= s->linesize*7;
2235 }
2236 } else {
2237 linesize = s->linesize;
2238 uvlinesize = s->uvlinesize;
2239 // dct_offset = s->linesize * 16;
2240 }
2241
2242 if(IS_INTRA(mb_type)){
2243 if(!(s->flags&CODEC_FLAG_GRAY)){
2244 h->pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
2245 h->pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
2246 }
2247
2248 if(IS_INTRA4x4(mb_type)){
2249 if(!s->encoding){
2250 for(i=0; i<16; i++){
2251 uint8_t * const ptr= dest_y + h->block_offset[i];
2252 uint8_t *topright= ptr + 4 - linesize;
2253 const int topright_avail= (h->topright_samples_available<<i)&0x8000;
2254 const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
2255 int tr;
2256
2257 if(!topright_avail){
2258 tr= ptr[3 - linesize]*0x01010101;
2259 topright= (uint8_t*) &tr;
2260 }
2261
2262 h->pred4x4[ dir ](ptr, topright, linesize);
2263 if(h->non_zero_count_cache[ scan8[i] ]){
2264 if(s->codec_id == CODEC_ID_H264)
2265 h264_add_idct_c(ptr, h->mb + i*16, linesize);
2266 else
2267 svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
2268 }
2269 }
2270 }
2271 }else{
2272 h->pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
2273 if(s->codec_id == CODEC_ID_H264)
2274 h264_luma_dc_dequant_idct_c(h->mb, s->qscale);
2275 else
2276 svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
2277 }
2278 }else if(s->codec_id == CODEC_ID_H264){
2279 hl_motion(h, dest_y, dest_cb, dest_cr,
2280 s->dsp.put_h264_qpel_pixels_tab, s->dsp.put_h264_chroma_pixels_tab,
2281 s->dsp.avg_h264_qpel_pixels_tab, s->dsp.avg_h264_chroma_pixels_tab);
2282 }
2283
2284
2285 if(!IS_INTRA4x4(mb_type)){
2286 if(s->codec_id == CODEC_ID_H264){
2287 for(i=0; i<16; i++){
2288 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
2289 uint8_t * const ptr= dest_y + h->block_offset[i];
2290 h264_add_idct_c(ptr, h->mb + i*16, linesize);
2291 }
2292 }
2293 }else{
2294 for(i=0; i<16; i++){
2295 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
2296 uint8_t * const ptr= dest_y + h->block_offset[i];
2297 svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
2298 }
2299 }
2300 }
2301 }
2302
2303 if(!(s->flags&CODEC_FLAG_GRAY)){
2304 chroma_dc_dequant_idct_c(h->mb + 16*16, h->chroma_qp);
2305 chroma_dc_dequant_idct_c(h->mb + 16*16+4*16, h->chroma_qp);
2306 if(s->codec_id == CODEC_ID_H264){
2307 for(i=16; i<16+4; i++){
2308 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2309 uint8_t * const ptr= dest_cb + h->block_offset[i];
2310 h264_add_idct_c(ptr, h->mb + i*16, uvlinesize);
2311 }
2312 }
2313 for(i=20; i<20+4; i++){
2314 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2315 uint8_t * const ptr= dest_cr + h->block_offset[i];
2316 h264_add_idct_c(ptr, h->mb + i*16, uvlinesize);
2317 }
2318 }
2319 }else{
2320 for(i=16; i<16+4; i++){
2321 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2322 uint8_t * const ptr= dest_cb + h->block_offset[i];
2323 svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
2324 }
2325 }
2326 for(i=20; i<20+4; i++){
2327 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2328 uint8_t * const ptr= dest_cr + h->block_offset[i];
2329 svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
2330 }
2331 }
2332 }
2333 }
2334 }
2335
2336 static void decode_mb_cabac(H264Context *h){
2337 // MpegEncContext * const s = &h->s;
2338 }
2339
2340 /**
2341 * fills the default_ref_list.
2342 */
2343 static int fill_default_ref_list(H264Context *h){
2344 MpegEncContext * const s = &h->s;
2345 int i;
2346 Picture sorted_short_ref[16];
2347
2348 if(h->slice_type==B_TYPE){
2349 int out_i;
2350 int limit= -1;
2351
2352 for(out_i=0; out_i<h->short_ref_count; out_i++){
2353 int best_i=-1;
2354 int best_poc=-1;
2355
2356 for(i=0; i<h->short_ref_count; i++){
2357 const int poc= h->short_ref[i]->poc;
2358 if(poc > limit && poc < best_poc){
2359 best_poc= poc;
2360 best_i= i;
2361 }
2362 }
2363
2364 assert(best_i != -1);
2365
2366 limit= best_poc;
2367 sorted_short_ref[out_i]= *h->short_ref[best_i];
2368 }
2369 }
2370
2371 if(s->picture_structure == PICT_FRAME){
2372 if(h->slice_type==B_TYPE){
2373 const int current_poc= s->current_picture_ptr->poc;
2374 int list;
2375
2376 for(list=0; list<2; list++){
2377 int index=0;
2378
2379 for(i=0; i<h->short_ref_count && index < h->ref_count[list]; i++){
2380 const int i2= list ? h->short_ref_count - i - 1 : i;
2381 const int poc= sorted_short_ref[i2].poc;
2382
2383 if(sorted_short_ref[i2].reference != 3) continue; //FIXME refernce field shit
2384
2385 if((list==1 && poc > current_poc) || (list==0 && poc < current_poc)){
2386 h->default_ref_list[list][index ]= sorted_short_ref[i2];
2387 h->default_ref_list[list][index++].pic_id= sorted_short_ref[i2].frame_num;
2388 }
2389 }
2390
2391 for(i=0; i<h->long_ref_count && index < h->ref_count[ list ]; i++){
2392 if(h->long_ref[i]->reference != 3) continue;
2393
2394 h->default_ref_list[ list ][index ]= *h->long_ref[i];
2395 h->default_ref_list[ list ][index++].pic_id= i;;
2396 }
2397
2398 if(h->long_ref_count > 1 && h->short_ref_count==0){
2399 Picture temp= h->default_ref_list[1][0];
2400 h->default_ref_list[1][0] = h->default_ref_list[1][1];
2401 h->default_ref_list[1][0] = temp;
2402 }
2403
2404 if(index < h->ref_count[ list ])
2405 memset(&h->default_ref_list[list][index], 0, sizeof(Picture)*(h->ref_count[ list ] - index));
2406 }
2407 }else{
2408 int index=0;
2409 for(i=0; i<h->short_ref_count && index < h->ref_count[0]; i++){
2410 if(h->short_ref[i]->reference != 3) continue; //FIXME refernce field shit
2411 h->default_ref_list[0][index ]= *h->short_ref[i];
2412 h->default_ref_list[0][index++].pic_id= h->short_ref[i]->frame_num;
2413 }
2414 for(i=0; i<h->long_ref_count && index < h->ref_count[0]; i++){
2415 if(h->long_ref[i]->reference != 3) continue;
2416 h->default_ref_list[0][index ]= *h->long_ref[i];
2417 h->default_ref_list[0][index++].pic_id= i;;
2418 }
2419 if(index < h->ref_count[0])
2420 memset(&h->default_ref_list[0][index], 0, sizeof(Picture)*(h->ref_count[0] - index));
2421 }
2422 }else{ //FIELD
2423 if(h->slice_type==B_TYPE){
2424 }else{
2425 //FIXME second field balh
2426 }
2427 }
2428 return 0;
2429 }
2430
2431 static int decode_ref_pic_list_reordering(H264Context *h){
2432 MpegEncContext * const s = &h->s;
2433 int list;
2434
2435 if(h->slice_type==I_TYPE || h->slice_type==SI_TYPE) return 0; //FIXME move beofre func
2436
2437 for(list=0; list<2; list++){
2438 memcpy(h->ref_list[list], h->default_ref_list[list], sizeof(Picture)*h->ref_count[list]);
2439
2440 if(get_bits1(&s->gb)){
2441 int pred= h->curr_pic_num;
2442 int index;
2443
2444 for(index=0; ; index++){
2445 int reordering_of_pic_nums_idc= get_ue_golomb(&s->gb);
2446 int pic_id;
2447 int i;
2448
2449
2450 if(index >= h->ref_count[list]){
2451 fprintf(stderr, "reference count overflow\n");
2452 return -1;
2453 }
2454
2455 if(reordering_of_pic_nums_idc<3){
2456 if(reordering_of_pic_nums_idc<2){
2457 const int abs_diff_pic_num= get_ue_golomb(&s->gb) + 1;
2458
2459 if(abs_diff_pic_num >= h->max_pic_num){
2460 fprintf(stderr, "abs_diff_pic_num overflow\n");
2461 return -1;
2462 }
2463
2464 if(reordering_of_pic_nums_idc == 0) pred-= abs_diff_pic_num;
2465 else pred+= abs_diff_pic_num;
2466 pred &= h->max_pic_num - 1;
2467
2468 for(i= h->ref_count[list]-1; i>=index; i--){
2469 if(h->ref_list[list][i].pic_id == pred && h->ref_list[list][i].long_ref==0)
2470 break;
2471 }
2472 }else{
2473 pic_id= get_ue_golomb(&s->gb); //long_term_pic_idx
2474
2475 for(i= h->ref_count[list]-1; i>=index; i--){
2476 if(h->ref_list[list][i].pic_id == pic_id && h->ref_list[list][i].long_ref==1)
2477 break;
2478 }
2479 }
2480
2481 if(i < index){
2482 fprintf(stderr, "reference picture missing during reorder\n");
2483 memset(&h->ref_list[list][index], 0, sizeof(Picture)); //FIXME
2484 }else if(i > index){
2485 Picture tmp= h->ref_list[list][i];
2486 for(; i>index; i--){
2487 h->ref_list[list][i]= h->ref_list[list][i-1];
2488 }
2489 h->ref_list[list][index]= tmp;
2490 }
2491 }else if(reordering_of_pic_nums_idc==3)
2492 break;
2493 else{
2494 fprintf(stderr, "illegal reordering_of_pic_nums_idc\n");
2495 return -1;
2496 }
2497 }
2498 }
2499
2500 if(h->slice_type!=B_TYPE) break;
2501 }
2502 return 0;
2503 }
2504
2505 static int pred_weight_table(H264Context *h){
2506 MpegEncContext * const s = &h->s;
2507 int list, i;
2508
2509 h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
2510 h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
2511
2512 for(list=0; list<2; list++){
2513 for(i=0; i<h->ref_count[list]; i++){
2514 int luma_weight_flag, chroma_weight_flag;
2515
2516 luma_weight_flag= get_bits1(&s->gb);
2517 if(luma_weight_flag){
2518 h->luma_weight[list][i]= get_se_golomb(&s->gb);
2519 h->luma_offset[list][i]= get_se_golomb(&s->gb);
2520 }
2521
2522 chroma_weight_flag= get_bits1(&s->gb);
2523 if(chroma_weight_flag){
2524 int j;
2525 for(j=0; j<2; j++){
2526 h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);
2527 h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);
2528 }
2529 }
2530 }
2531 if(h->slice_type != B_TYPE) break;
2532 }
2533 return 0;
2534 }
2535
2536 /**
2537 * instantaneos decoder refresh.
2538 */
2539 static void idr(H264Context *h){
2540 int i;
2541
2542 for(i=0; i<h->long_ref_count; i++){
2543 h->long_ref[i]->reference=0;
2544 h->long_ref[i]= NULL;
2545 }
2546 h->long_ref_count=0;
2547
2548 for(i=0; i<h->short_ref_count; i++){
2549 h->short_ref[i]->reference=0;
2550 h->short_ref[i]= NULL;
2551 }
2552 h->short_ref_count=0;
2553 }
2554
2555 /**
2556 *
2557 * @return the removed picture or NULL if an error occures
2558 */
2559 static Picture * remove_short(H264Context *h, int frame_num){
2560 MpegEncContext * const s = &h->s;
2561 int i;
2562
2563 if(s->avctx->debug&FF_DEBUG_MMCO)
2564 printf("remove short %d count %d\n", frame_num, h->short_ref_count);
2565
2566 for(i=0; i<h->short_ref_count; i++){
2567 Picture *pic= h->short_ref[i];
2568 if(s->avctx->debug&FF_DEBUG_MMCO)
2569 printf("%d %d %p\n", i, pic->frame_num, pic);
2570 if(pic->frame_num == frame_num){
2571 h->short_ref[i]= NULL;
2572 memmove(&h->short_ref[i], &h->short_ref[i+1], (h->short_ref_count - i - 1)*sizeof(Picture*));
2573 h->short_ref_count--;
2574 return pic;
2575 }
2576 }
2577 return NULL;
2578 }
2579
2580 /**
2581 *
2582 * @return the removed picture or NULL if an error occures
2583 */
2584 static Picture * remove_long(H264Context *h, int i){
2585 Picture *pic;
2586
2587 if(i >= h->long_ref_count) return NULL;
2588 pic= h->long_ref[i];
2589 if(pic==NULL) return NULL;
2590
2591 h->long_ref[i]= NULL;
2592 memmove(&h->long_ref[i], &h->long_ref[i+1], (h->long_ref_count - i - 1)*sizeof(Picture*));
2593 h->long_ref_count--;
2594
2595 return pic;
2596 }
2597
2598 /**
2599 * Executes the reference picture marking (memory management control operations).
2600 */
2601 static int execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count){
2602 MpegEncContext * const s = &h->s;
2603 int i;
2604 int current_is_long=0;
2605 Picture *pic;
2606
2607 if((s->avctx->debug&FF_DEBUG_MMCO) && mmco_count==0)
2608 printf("no mmco here\n");
2609
2610 for(i=0; i<mmco_count; i++){
2611 if(s->avctx->debug&FF_DEBUG_MMCO)
2612 printf("mmco:%d %d %d\n", h->mmco[i].opcode, h->mmco[i].short_frame_num, h->mmco[i].long_index);
2613
2614 switch(mmco[i].opcode){
2615 case MMCO_SHORT2UNUSED:
2616 pic= remove_short(h, mmco[i].short_frame_num);
2617 if(pic==NULL) return -1;
2618 pic->reference= 0;
2619 break;
2620 case MMCO_SHORT2LONG:
2621 pic= remove_long(h, mmco[i].long_index);
2622 if(pic) pic->reference=0;
2623
2624 h->long_ref[ mmco[i].long_index ]= remove_short(h, mmco[i].short_frame_num);
2625 h->long_ref[ mmco[i].long_index ]->long_ref=1;
2626 break;
2627 case MMCO_LONG2UNUSED:
2628 pic= remove_long(h, mmco[i].long_index);
2629 if(pic==NULL) return -1;
2630 pic->reference= 0;
2631 break;
2632 case MMCO_LONG:
2633 pic= remove_long(h, mmco[i].long_index);
2634 if(pic) pic->reference=0;
2635
2636 h->long_ref[ mmco[i].long_index ]= s->current_picture_ptr;
2637 h->long_ref[ mmco[i].long_index ]->long_ref=1;
2638 h->long_ref_count++;
2639
2640 current_is_long=1;
2641 break;
2642 case MMCO_SET_MAX_LONG:
2643 assert(mmco[i].long_index <= 16);
2644 while(mmco[i].long_index < h->long_ref_count){
2645 pic= remove_long(h, mmco[i].long_index);
2646 pic->reference=0;
2647 }
2648 while(mmco[i].long_index > h->long_ref_count){
2649 h->long_ref[ h->long_ref_count++ ]= NULL;
2650 }
2651 break;
2652 case MMCO_RESET:
2653 while(h->short_ref_count){
2654 pic= remove_short(h, h->short_ref[0]->frame_num);
2655 pic->reference=0;
2656 }
2657 while(h->long_ref_count){
2658 pic= remove_long(h, h->long_ref_count-1);
2659 pic->reference=0;
2660 }
2661 break;
2662 default: assert(0);
2663 }
2664 }
2665
2666 if(!current_is_long){
2667 pic= remove_short(h, s->current_picture_ptr->frame_num);
2668 if(pic){
2669 pic->reference=0;
2670 fprintf(stderr, "illegal short term buffer state detected\n");
2671 }
2672
2673 if(h->short_ref_count)
2674 memmove(&h->short_ref[1], &h->short_ref[0], h->short_ref_count*sizeof(Picture*));
2675
2676 h->short_ref[0]= s->current_picture_ptr;
2677 h->short_ref[0]->long_ref=0;
2678 h->short_ref_count++;
2679 }
2680
2681 return 0;
2682 }
2683
2684 static int decode_ref_pic_marking(H264Context *h){
2685 MpegEncContext * const s = &h->s;
2686 int i;
2687
2688 if(h->nal_unit_type == NAL_IDR_SLICE){ //FIXME fields
2689 s->broken_link= get_bits1(&s->gb) -1;
2690 h->mmco[0].long_index= get_bits1(&s->gb) - 1; // current_long_term_idx
2691 if(h->mmco[0].long_index == -1)
2692 h->mmco_index= 0;
2693 else{
2694 h->mmco[0].opcode= MMCO_LONG;
2695 h->mmco_index= 1;
2696 }
2697 }else{
2698 if(get_bits1(&s->gb)){ // adaptive_ref_pic_marking_mode_flag
2699 for(i= h->mmco_index; i<MAX_MMCO_COUNT; i++) {
2700 MMCOOpcode opcode= get_ue_golomb(&s->gb);;
2701
2702 h->mmco[i].opcode= opcode;
2703 if(opcode==MMCO_SHORT2UNUSED || opcode==MMCO_SHORT2LONG){
2704 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
2705 /* if(h->mmco[i].short_frame_num >= h->short_ref_count || h->short_ref[ h->mmco[i].short_frame_num ] == NULL){
2706 fprintf(stderr, "illegal short ref in memory management control operation %d\n", mmco);
2707 return -1;
2708 }*/
2709 }
2710 if(opcode==MMCO_SHORT2LONG || opcode==MMCO_LONG2UNUSED || opcode==MMCO_LONG || opcode==MMCO_SET_MAX_LONG){
2711 h->mmco[i].long_index= get_ue_golomb(&s->gb);
2712 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){
2713 fprintf(stderr, "illegal long ref in memory management control operation %d\n", opcode);
2714 return -1;
2715 }
2716 }
2717
2718 if(opcode > MMCO_LONG){
2719 fprintf(stderr, "illegal memory management control operation %d\n", opcode);
2720 return -1;