2 * H.26L/H.264/AVC/JVT/14496-10/... encoder/decoder
3 * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
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.
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.
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
23 * H.264 / AVC / MPEG4 part10 codec.
24 * @author Michael Niedermayer <michaelni@gmx.at>
30 #include "mpegvideo.h"
37 #define interlaced_dct interlaced_dct_is_a_bad_name
38 #define mb_intra mb_intra_isnt_initalized_see_mb_type
40 #define LUMA_DC_BLOCK_INDEX 25
41 #define CHROMA_DC_BLOCK_INDEX 26
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
50 #define MAX_SPS_COUNT 32
51 #define MAX_PPS_COUNT 256
53 #define MAX_MMCO_COUNT 66
56 * Sequence parameter set
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;
82 short offset_for_ref_frame[256]; //FIXME dyn aloc?
86 * Picture parameter set
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
111 * Memory management control operation opcode.
113 typedef enum MMCOOpcode{
124 * Memory management control operation.
135 typedef struct H264Context{
143 #define NAL_IDR_SLICE 5
147 #define NAL_PICTURE_DELIMITER 9
148 #define NAL_FILTER_DATA 10
149 uint8_t *rbsp_buffer;
150 int rbsp_buffer_size;
154 int prev_mb_skiped; //FIXME remove (IMHO not used)
157 int chroma_pred_mode;
158 int intra16x16_pred_mode;
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;
171 * non zero coeff count cache.
172 * is 64 if not available.
174 uint8_t non_zero_count_cache[6*8];
175 uint8_t (*non_zero_count)[16];
178 * Motion vector cache.
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
186 * is 1 if the specific list MV&references are set to 0,0,-2.
188 int mv_cache_clean[2];
190 int block_offset[16+8];
191 int chroma_subblock_offset[16]; //FIXME remove
193 uint16_t *mb2b_xy; //FIXME are these 4 a good idea?
201 SPS sps_buffer[MAX_SPS_COUNT];
202 SPS sps; ///< current sps
204 PPS pps_buffer[MAX_PPS_COUNT];
208 PPS pps; //FIXME move tp Picture perhaps? (->no) do we need that?
211 uint8_t *slice_table_base;
212 uint8_t *slice_table; ///< slice_table_base + mb_stride + 1
214 int slice_type_fixed;
216 //interlacing specific flags
217 int mb_field_decoding_flag;
224 int delta_poc_bottom;
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
234 * frame_num for frames or 2*frame_num for field pics.
239 * max_frame_num or 2*max_frame_num for field pics.
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];
252 int disable_deblocking_filter_idc;
253 int slice_alpha_c0_offset_div2;
254 int slice_beta_offset_div2;
256 int redundant_pic_count;
258 int direct_spatial_mv_pred;
261 * num_ref_idx_l0/1_active_minus1 + 1
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?
271 * memory management control operations buffer.
273 MMCO mmco[MAX_MMCO_COUNT];
276 int long_ref_count; ///< number of actual long term references
277 int short_ref_count; ///< number of actual short term references
280 GetBitContext intra_gb;
281 GetBitContext inter_gb;
282 GetBitContext *intra_gb_ptr;
283 GetBitContext *inter_gb_ptr;
285 DCTELEM mb[16*24] __align8;
288 static VLC coeff_token_vlc[4];
289 static VLC chroma_dc_coeff_token_vlc;
291 static VLC total_zeros_vlc[15];
292 static VLC chroma_dc_total_zeros_vlc[3];
294 static VLC run_vlc[6];
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);
302 * @param h height of the recatangle, should be a constant
303 * @param w width of the recatangle, should be a constant
304 * @param size the size of val (1 or 4), should be a constant
306 static inline void fill_rectangle(void *vp, int w, int h, int stride, uint32_t val, int size){ //FIXME ensure this IS inlined
307 uint8_t *p= (uint8_t*)vp;
308 assert(size==1 || size==4);
313 //FIXME check what gcc generates for 64 bit on x86 and possible write a 32 bit ver of it
316 *(uint16_t*)(p + stride)= size==4 ? val : val*0x0101;
317 }else if(w==2 && h==4){
318 *(uint16_t*)(p + 0*stride)=
319 *(uint16_t*)(p + 1*stride)=
320 *(uint16_t*)(p + 2*stride)=
321 *(uint16_t*)(p + 3*stride)= size==4 ? val : val*0x0101;
322 }else if(w==4 && h==2){
323 *(uint32_t*)(p + 0*stride)=
324 *(uint32_t*)(p + 1*stride)= size==4 ? val : val*0x01010101;
325 }else if(w==4 && h==4){
326 *(uint32_t*)(p + 0*stride)=
327 *(uint32_t*)(p + 1*stride)=
328 *(uint32_t*)(p + 2*stride)=
329 *(uint32_t*)(p + 3*stride)= size==4 ? val : val*0x01010101;
330 }else if(w==8 && h==1){
332 *(uint32_t*)(p + 4)= size==4 ? val : val*0x01010101;
333 }else if(w==8 && h==2){
334 *(uint32_t*)(p + 0 + 0*stride)=
335 *(uint32_t*)(p + 4 + 0*stride)=
336 *(uint32_t*)(p + 0 + 1*stride)=
337 *(uint32_t*)(p + 4 + 1*stride)= size==4 ? val : val*0x01010101;
338 }else if(w==8 && h==4){
339 *(uint64_t*)(p + 0*stride)=
340 *(uint64_t*)(p + 1*stride)=
341 *(uint64_t*)(p + 2*stride)=
342 *(uint64_t*)(p + 3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
343 }else if(w==16 && h==2){
344 *(uint64_t*)(p + 0+0*stride)=
345 *(uint64_t*)(p + 8+0*stride)=
346 *(uint64_t*)(p + 0+1*stride)=
347 *(uint64_t*)(p + 8+1*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
348 }else if(w==16 && h==4){
349 *(uint64_t*)(p + 0+0*stride)=
350 *(uint64_t*)(p + 8+0*stride)=
351 *(uint64_t*)(p + 0+1*stride)=
352 *(uint64_t*)(p + 8+1*stride)=
353 *(uint64_t*)(p + 0+2*stride)=
354 *(uint64_t*)(p + 8+2*stride)=
355 *(uint64_t*)(p + 0+3*stride)=
356 *(uint64_t*)(p + 8+3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
361 static inline void fill_caches(H264Context *h, int mb_type){
362 MpegEncContext * const s = &h->s;
363 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
364 int topleft_xy, top_xy, topright_xy, left_xy[2];
365 int topleft_type, top_type, topright_type, left_type[2];
369 //wow what a mess, why didnt they simplify the interlacing&intra stuff, i cant imagine that these complex rules are worth it
374 topleft_xy = mb_xy-1 - s->mb_stride;
375 top_xy = mb_xy - s->mb_stride;
376 topright_xy= mb_xy+1 - s->mb_stride;
377 left_xy[0] = mb_xy-1;
378 left_xy[1] = mb_xy-1;
385 topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
386 top_type = h->slice_table[top_xy ] == h->slice_num ? s->current_picture.mb_type[top_xy] : 0;
387 topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
388 left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
389 left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
391 if(IS_INTRA(mb_type)){
392 h->topleft_samples_available=
393 h->top_samples_available=
394 h->left_samples_available= 0xFFFF;
395 h->topright_samples_available= 0xEEEA;
397 if(!IS_INTRA(top_type) && (top_type==0 || h->pps.constrained_intra_pred)){
398 h->topleft_samples_available= 0xB3FF;
399 h->top_samples_available= 0x33FF;
400 h->topright_samples_available= 0x26EA;
403 if(!IS_INTRA(left_type[i]) && (left_type[i]==0 || h->pps.constrained_intra_pred)){
404 h->topleft_samples_available&= 0xDF5F;
405 h->left_samples_available&= 0x5F5F;
409 if(!IS_INTRA(topleft_type) && (topleft_type==0 || h->pps.constrained_intra_pred))
410 h->topleft_samples_available&= 0x7FFF;
412 if(!IS_INTRA(topright_type) && (topright_type==0 || h->pps.constrained_intra_pred))
413 h->topright_samples_available&= 0xFBFF;
415 if(IS_INTRA4x4(mb_type)){
416 if(IS_INTRA4x4(top_type)){
417 h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
418 h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
419 h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
420 h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
423 if(IS_INTRA16x16(top_type) || (IS_INTER(top_type) && !h->pps.constrained_intra_pred))
428 h->intra4x4_pred_mode_cache[4+8*0]=
429 h->intra4x4_pred_mode_cache[5+8*0]=
430 h->intra4x4_pred_mode_cache[6+8*0]=
431 h->intra4x4_pred_mode_cache[7+8*0]= pred;
434 if(IS_INTRA4x4(left_type[i])){
435 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
436 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
439 if(IS_INTRA16x16(left_type[i]) || (IS_INTER(left_type[i]) && !h->pps.constrained_intra_pred))
444 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
445 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
460 //FIXME constraint_intra_pred & partitioning & nnz (lets hope this is just a typo in the spec)
462 h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][0];
463 h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][1];
464 h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][2];
465 h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3];
467 h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][7];
468 h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8];
470 h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][10];
471 h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11];
473 h->non_zero_count_cache[4+8*0]=
474 h->non_zero_count_cache[5+8*0]=
475 h->non_zero_count_cache[6+8*0]=
476 h->non_zero_count_cache[7+8*0]=
478 h->non_zero_count_cache[1+8*0]=
479 h->non_zero_count_cache[2+8*0]=
481 h->non_zero_count_cache[1+8*3]=
482 h->non_zero_count_cache[2+8*3]= 64;
486 h->non_zero_count_cache[3+8*1]= h->non_zero_count[left_xy[0]][6];
487 h->non_zero_count_cache[3+8*2]= h->non_zero_count[left_xy[0]][5];
488 h->non_zero_count_cache[0+8*1]= h->non_zero_count[left_xy[0]][9]; //FIXME left_block
489 h->non_zero_count_cache[0+8*4]= h->non_zero_count[left_xy[0]][12];
491 h->non_zero_count_cache[3+8*1]=
492 h->non_zero_count_cache[3+8*2]=
493 h->non_zero_count_cache[0+8*1]=
494 h->non_zero_count_cache[0+8*4]= 64;
498 h->non_zero_count_cache[3+8*3]= h->non_zero_count[left_xy[1]][4];
499 h->non_zero_count_cache[3+8*4]= h->non_zero_count[left_xy[1]][3];
500 h->non_zero_count_cache[0+8*2]= h->non_zero_count[left_xy[1]][8];
501 h->non_zero_count_cache[0+8*5]= h->non_zero_count[left_xy[1]][11];
503 h->non_zero_count_cache[3+8*3]=
504 h->non_zero_count_cache[3+8*4]=
505 h->non_zero_count_cache[0+8*2]=
506 h->non_zero_count_cache[0+8*5]= 64;
510 if(IS_INTER(mb_type)){
512 for(list=0; list<2; list++){
513 if((!IS_8X8(mb_type)) && !USES_LIST(mb_type, list)){
514 /*if(!h->mv_cache_clean[list]){
515 memset(h->mv_cache [list], 0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
516 memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
517 h->mv_cache_clean[list]= 1;
519 continue; //FIXME direct mode ...
521 h->mv_cache_clean[list]= 0;
523 if(IS_INTER(topleft_type)){
524 const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
525 const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + h->b8_stride;
526 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
527 h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
529 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
530 h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
533 if(IS_INTER(top_type)){
534 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
535 const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
536 *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0];
537 *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];
538 *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];
539 *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];
540 h->ref_cache[list][scan8[0] + 0 - 1*8]=
541 h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
542 h->ref_cache[list][scan8[0] + 2 - 1*8]=
543 h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
545 *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]=
546 *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]=
547 *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]=
548 *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;
549 *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
552 if(IS_INTER(topright_type)){
553 const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
554 const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
555 *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
556 h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
558 *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
559 h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
562 //FIXME unify cleanup or sth
563 if(IS_INTER(left_type[0])){
564 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
565 const int b8_xy= h->mb2b8_xy[left_xy[0]] + 1;
566 *(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]];
567 *(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]];
568 h->ref_cache[list][scan8[0] - 1 + 0*8]=
569 h->ref_cache[list][scan8[0] - 1 + 1*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0]>>1)];
571 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 0*8]=
572 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 1*8]= 0;
573 h->ref_cache[list][scan8[0] - 1 + 0*8]=
574 h->ref_cache[list][scan8[0] - 1 + 1*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
577 if(IS_INTER(left_type[1])){
578 const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
579 const int b8_xy= h->mb2b8_xy[left_xy[1]] + 1;
580 *(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]];
581 *(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]];
582 h->ref_cache[list][scan8[0] - 1 + 2*8]=
583 h->ref_cache[list][scan8[0] - 1 + 3*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[2]>>1)];
585 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 2*8]=
586 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 3*8]= 0;
587 h->ref_cache[list][scan8[0] - 1 + 2*8]=
588 h->ref_cache[list][scan8[0] - 1 + 3*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
591 h->ref_cache[list][scan8[5 ]+1] =
592 h->ref_cache[list][scan8[7 ]+1] =
593 h->ref_cache[list][scan8[13]+1] = //FIXME remove past 3 (init somewher else)
594 h->ref_cache[list][scan8[4 ]] =
595 h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
596 *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
597 *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
598 *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewher else)
599 *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
600 *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
608 static inline void write_back_intra_pred_mode(H264Context *h){
609 MpegEncContext * const s = &h->s;
610 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
612 h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
613 h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
614 h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
615 h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
616 h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
617 h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
618 h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
622 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
624 static inline int check_intra4x4_pred_mode(H264Context *h){
625 MpegEncContext * const s = &h->s;
626 static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
627 static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
630 if(!(h->top_samples_available&0x8000)){
632 int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
634 fprintf(stderr, "top block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y);
637 h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
642 if(!(h->left_samples_available&0x8000)){
644 int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
646 fprintf(stderr, "left block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y);
649 h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
655 } //FIXME cleanup like next
658 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
660 static inline int check_intra_pred_mode(H264Context *h, int mode){
661 MpegEncContext * const s = &h->s;
662 static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
663 static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
665 if(!(h->top_samples_available&0x8000)){
668 fprintf(stderr, "top block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);
673 if(!(h->left_samples_available&0x8000)){
676 fprintf(stderr, "left block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);
685 * gets the predicted intra4x4 prediction mode.
687 static inline int pred_intra_mode(H264Context *h, int n){
688 const int index8= scan8[n];
689 const int left= h->intra4x4_pred_mode_cache[index8 - 1];
690 const int top = h->intra4x4_pred_mode_cache[index8 - 8];
691 const int min= FFMIN(left, top);
693 tprintf("mode:%d %d min:%d\n", left ,top, min);
695 if(min<0) return DC_PRED;
699 static inline void write_back_non_zero_count(H264Context *h){
700 MpegEncContext * const s = &h->s;
701 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
703 h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[4+8*4];
704 h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[5+8*4];
705 h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[6+8*4];
706 h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];
707 h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[7+8*3];
708 h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[7+8*2];
709 h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[7+8*1];
711 h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[1+8*2];
712 h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
713 h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[2+8*1];
715 h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[1+8*5];
716 h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
717 h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[2+8*4];
721 * gets the predicted number of non zero coefficients.
722 * @param n block index
724 static inline int pred_non_zero_count(H264Context *h, int n){
725 const int index8= scan8[n];
726 const int left= h->non_zero_count_cache[index8 - 1];
727 const int top = h->non_zero_count_cache[index8 - 8];
730 if(i<64) i= (i+1)>>1;
732 tprintf("pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
737 static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
738 const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];
740 if(topright_ref != PART_NOT_AVAILABLE){
741 *C= h->mv_cache[list][ i - 8 + part_width ];
744 tprintf("topright MV not available\n");
746 *C= h->mv_cache[list][ i - 8 - 1 ];
747 return h->ref_cache[list][ i - 8 - 1 ];
752 * gets the predicted MV.
753 * @param n the block index
754 * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)
755 * @param mx the x component of the predicted motion vector
756 * @param my the y component of the predicted motion vector
758 static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
759 const int index8= scan8[n];
760 const int top_ref= h->ref_cache[list][ index8 - 8 ];
761 const int left_ref= h->ref_cache[list][ index8 - 1 ];
762 const int16_t * const A= h->mv_cache[list][ index8 - 1 ];
763 const int16_t * const B= h->mv_cache[list][ index8 - 8 ];
765 int diagonal_ref, match_count;
767 assert(part_width==1 || part_width==2 || part_width==4);
777 diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);
778 match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);
780 if(match_count > 1){ //most common
781 *mx= mid_pred(A[0], B[0], C[0]);
782 *my= mid_pred(A[1], B[1], C[1]);
783 }else if(match_count==1){
787 }else if(top_ref==ref){
795 if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
799 *mx= mid_pred(A[0], B[0], C[0]);
800 *my= mid_pred(A[1], B[1], C[1]);
804 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);
808 * gets the directionally predicted 16x8 MV.
809 * @param n the block index
810 * @param mx the x component of the predicted motion vector
811 * @param my the y component of the predicted motion vector
813 static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
815 const int top_ref= h->ref_cache[list][ scan8[0] - 8 ];
816 const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
818 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);
826 const int left_ref= h->ref_cache[list][ scan8[8] - 1 ];
827 const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];
829 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);
839 pred_motion(h, n, 4, list, ref, mx, my);
843 * gets the directionally predicted 8x16 MV.
844 * @param n the block index
845 * @param mx the x component of the predicted motion vector
846 * @param my the y component of the predicted motion vector
848 static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
850 const int left_ref= h->ref_cache[list][ scan8[0] - 1 ];
851 const int16_t * const A= h->mv_cache[list][ scan8[0] - 1 ];
853 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);
864 diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);
866 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);
868 if(diagonal_ref == ref){
876 pred_motion(h, n, 2, list, ref, mx, my);
879 static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
880 const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
881 const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];
883 tprintf("pred_pskip: (%d) (%d) at %2d %2d", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
885 if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
886 || (top_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0)
887 || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){
893 pred_motion(h, 0, 4, 0, 0, mx, my);
898 static inline void write_back_motion(H264Context *h, int mb_type){
899 MpegEncContext * const s = &h->s;
900 const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
901 const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
904 for(list=0; list<2; list++){
906 if((!IS_8X8(mb_type)) && !USES_LIST(mb_type, list)){
907 if(1){ //FIXME skip or never read if mb_type doesnt use it
909 *(uint64_t*)s->current_picture.motion_val[list][b_xy + 0 + y*h->b_stride]=
910 *(uint64_t*)s->current_picture.motion_val[list][b_xy + 2 + y*h->b_stride]= 0;
913 *(uint16_t*)s->current_picture.motion_val[list][b8_xy + y*h->b8_stride]= (LIST_NOT_USED&0xFF)*0x0101;
916 continue; //FIXME direct mode ...
920 *(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];
921 *(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];
924 s->current_picture.ref_index[list][b8_xy + 0 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+0 + 16*y];
925 s->current_picture.ref_index[list][b8_xy + 1 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+2 + 16*y];
931 * Decodes a network abstraction layer unit.
932 * @param consumed is the number of bytes used as input
933 * @param length is the length of the array
934 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp ttailing?
935 * @returns decoded bytes, might be src+1 if no escapes
937 static uint8_t *decode_nal(H264Context *h, uint8_t *src, int *dst_length, int *consumed, int length){
941 // src[0]&0x80; //forbidden bit
942 h->nal_ref_idc= src[0]>>5;
943 h->nal_unit_type= src[0]&0x1F;
947 for(i=0; i<length; i++)
948 printf("%2X ", src[i]);
950 for(i=0; i+1<length; i+=2){
952 if(i>0 && src[i-1]==0) i--;
953 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
955 /* startcode, so we must be past the end */
962 if(i>=length-1){ //no escaped 0
964 *consumed= length+1; //+1 for the header
968 h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length);
971 //printf("deoding esc\n");
974 //remove escapes (very rare 1:2^22)
975 if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
976 if(src[si+2]==3){ //escape
980 }else //next start code
984 dst[di++]= src[si++];
988 *consumed= si + 1;//+1 for the header
989 //FIXME store exact number of bits in the getbitcontext (its needed for decoding)
994 * @param src the data which should be escaped
995 * @param dst the target buffer, dst+1 == src is allowed as a special case
996 * @param length the length of the src data
997 * @param dst_length the length of the dst array
998 * @returns length of escaped data in bytes or -1 if an error occured
1000 static int encode_nal(H264Context *h, uint8_t *dst, uint8_t *src, int length, int dst_length){
1001 int i, escape_count, si, di;
1005 assert(dst_length>0);
1007 dst[0]= (h->nal_ref_idc<<5) + h->nal_unit_type;
1009 if(length==0) return 1;
1012 for(i=0; i<length; i+=2){
1013 if(src[i]) continue;
1014 if(i>0 && src[i-1]==0)
1016 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1022 if(escape_count==0){
1024 memcpy(dst+1, src, length);
1028 if(length + escape_count + 1> dst_length)
1031 //this should be damn rare (hopefully)
1033 h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length + escape_count);
1034 temp= h->rbsp_buffer;
1035 //printf("encoding esc\n");
1040 if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1041 temp[di++]= 0; si++;
1042 temp[di++]= 0; si++;
1044 temp[di++]= src[si++];
1047 temp[di++]= src[si++];
1049 memcpy(dst+1, temp, length+escape_count);
1051 assert(di == length+escape_count);
1057 * write 1,10,100,1000,... for alignment, yes its exactly inverse to mpeg4
1059 static void encode_rbsp_trailing(PutBitContext *pb){
1062 length= (-get_bit_count(pb))&7;
1063 if(length) put_bits(pb, length, 0);
1067 * identifies the exact end of the bitstream
1068 * @return the length of the trailing, or 0 if damaged
1070 static int decode_rbsp_trailing(uint8_t *src){
1074 tprintf("rbsp trailing %X\n", v);
1084 * idct tranforms the 16 dc values and dequantize them.
1085 * @param qp quantization parameter
1087 static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp){
1088 const int qmul= dequant_coeff[qp][0];
1091 int temp[16]; //FIXME check if this is a good idea
1092 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
1093 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1095 //memset(block, 64, 2*256);
1098 const int offset= y_offset[i];
1099 const int z0= block[offset+stride*0] + block[offset+stride*4];
1100 const int z1= block[offset+stride*0] - block[offset+stride*4];
1101 const int z2= block[offset+stride*1] - block[offset+stride*5];
1102 const int z3= block[offset+stride*1] + block[offset+stride*5];
1111 const int offset= x_offset[i];
1112 const int z0= temp[4*0+i] + temp[4*2+i];
1113 const int z1= temp[4*0+i] - temp[4*2+i];
1114 const int z2= temp[4*1+i] - temp[4*3+i];
1115 const int z3= temp[4*1+i] + temp[4*3+i];
1117 block[stride*0 +offset]= ((z0 + z3)*qmul + 2)>>2; //FIXME think about merging this into decode_resdual
1118 block[stride*2 +offset]= ((z1 + z2)*qmul + 2)>>2;
1119 block[stride*8 +offset]= ((z1 - z2)*qmul + 2)>>2;
1120 block[stride*10+offset]= ((z0 - z3)*qmul + 2)>>2;
1125 * dct tranforms the 16 dc values.
1126 * @param qp quantization parameter ??? FIXME
1128 static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
1129 // const int qmul= dequant_coeff[qp][0];
1131 int temp[16]; //FIXME check if this is a good idea
1132 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
1133 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1136 const int offset= y_offset[i];
1137 const int z0= block[offset+stride*0] + block[offset+stride*4];
1138 const int z1= block[offset+stride*0] - block[offset+stride*4];
1139 const int z2= block[offset+stride*1] - block[offset+stride*5];
1140 const int z3= block[offset+stride*1] + block[offset+stride*5];
1149 const int offset= x_offset[i];
1150 const int z0= temp[4*0+i] + temp[4*2+i];
1151 const int z1= temp[4*0+i] - temp[4*2+i];
1152 const int z2= temp[4*1+i] - temp[4*3+i];
1153 const int z3= temp[4*1+i] + temp[4*3+i];
1155 block[stride*0 +offset]= (z0 + z3)>>1;
1156 block[stride*2 +offset]= (z1 + z2)>>1;
1157 block[stride*8 +offset]= (z1 - z2)>>1;
1158 block[stride*10+offset]= (z0 - z3)>>1;
1164 static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp){
1165 const int qmul= dequant_coeff[qp][0];
1166 const int stride= 16*2;
1167 const int xStride= 16;
1170 a= block[stride*0 + xStride*0];
1171 b= block[stride*0 + xStride*1];
1172 c= block[stride*1 + xStride*0];
1173 d= block[stride*1 + xStride*1];
1180 block[stride*0 + xStride*0]= ((a+c)*qmul + 0)>>1;
1181 block[stride*0 + xStride*1]= ((e+b)*qmul + 0)>>1;
1182 block[stride*1 + xStride*0]= ((a-c)*qmul + 0)>>1;
1183 block[stride*1 + xStride*1]= ((e-b)*qmul + 0)>>1;
1186 static void chroma_dc_dct_c(DCTELEM *block){
1187 const int stride= 16*2;
1188 const int xStride= 16;
1191 a= block[stride*0 + xStride*0];
1192 b= block[stride*0 + xStride*1];
1193 c= block[stride*1 + xStride*0];
1194 d= block[stride*1 + xStride*1];
1201 block[stride*0 + xStride*0]= (a+c);
1202 block[stride*0 + xStride*1]= (e+b);
1203 block[stride*1 + xStride*0]= (a-c);
1204 block[stride*1 + xStride*1]= (e-b);
1208 * gets the chroma qp.
1210 static inline int get_chroma_qp(H264Context *h, int qscale){
1212 return chroma_qp[clip(qscale + h->pps.chroma_qp_index_offset, 0, 51)];
1219 static void h264_add_idct_c(uint8_t *dst, DCTELEM *block, int stride){
1221 uint8_t *cm = cropTbl + MAX_NEG_CROP;
1226 const int z0= block[i + 4*0] + block[i + 4*2];
1227 const int z1= block[i + 4*0] - block[i + 4*2];
1228 const int z2= (block[i + 4*1]>>1) - block[i + 4*3];
1229 const int z3= block[i + 4*1] + (block[i + 4*3]>>1);
1231 block[i + 4*0]= z0 + z3;
1232 block[i + 4*1]= z1 + z2;
1233 block[i + 4*2]= z1 - z2;
1234 block[i + 4*3]= z0 - z3;
1238 const int z0= block[0 + 4*i] + block[2 + 4*i];
1239 const int z1= block[0 + 4*i] - block[2 + 4*i];
1240 const int z2= (block[1 + 4*i]>>1) - block[3 + 4*i];
1241 const int z3= block[1 + 4*i] + (block[3 + 4*i]>>1);
1243 dst[0 + i*stride]= cm[ dst[0 + i*stride] + ((z0 + z3) >> 6) ];
1244 dst[1 + i*stride]= cm[ dst[1 + i*stride] + ((z1 + z2) >> 6) ];
1245 dst[2 + i*stride]= cm[ dst[2 + i*stride] + ((z1 - z2) >> 6) ];
1246 dst[3 + i*stride]= cm[ dst[3 + i*stride] + ((z0 - z3) >> 6) ];
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);
1255 block[0 + 4*i]= z0 + z3;
1256 block[1 + 4*i]= z1 + z2;
1257 block[2 + 4*i]= z1 - z2;
1258 block[3 + 4*i]= z0 - z3;
1262 const int z0= block[i + 4*0] + block[i + 4*2];
1263 const int z1= block[i + 4*0] - block[i + 4*2];
1264 const int z2= (block[i + 4*1]>>1) - block[i + 4*3];
1265 const int z3= block[i + 4*1] + (block[i + 4*3]>>1);
1267 dst[i + 0*stride]= cm[ dst[i + 0*stride] + ((z0 + z3) >> 6) ];
1268 dst[i + 1*stride]= cm[ dst[i + 1*stride] + ((z1 + z2) >> 6) ];
1269 dst[i + 2*stride]= cm[ dst[i + 2*stride] + ((z1 - z2) >> 6) ];
1270 dst[i + 3*stride]= cm[ dst[i + 3*stride] + ((z0 - z3) >> 6) ];
1275 static void h264_diff_dct_c(DCTELEM *block, uint8_t *src1, uint8_t *src2, int stride){
1277 //FIXME try int temp instead of block
1280 const int d0= src1[0 + i*stride] - src2[0 + i*stride];
1281 const int d1= src1[1 + i*stride] - src2[1 + i*stride];
1282 const int d2= src1[2 + i*stride] - src2[2 + i*stride];
1283 const int d3= src1[3 + i*stride] - src2[3 + i*stride];
1284 const int z0= d0 + d3;
1285 const int z3= d0 - d3;
1286 const int z1= d1 + d2;
1287 const int z2= d1 - d2;
1289 block[0 + 4*i]= z0 + z1;
1290 block[1 + 4*i]= 2*z3 + z2;
1291 block[2 + 4*i]= z0 - z1;
1292 block[3 + 4*i]= z3 - 2*z2;
1296 const int z0= block[0*4 + i] + block[3*4 + i];
1297 const int z3= block[0*4 + i] - block[3*4 + i];
1298 const int z1= block[1*4 + i] + block[2*4 + i];
1299 const int z2= block[1*4 + i] - block[2*4 + i];
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;
1308 //FIXME need to check that this doesnt overflow signed 32 bit for low qp, iam not sure, its very close
1309 //FIXME check that gcc inlines this (and optimizes intra & seperate_dc stuff away)
1310 static inline int quantize_c(DCTELEM *block, uint8_t *scantable, int qscale, int intra, int seperate_dc){
1312 const int * const quant_table= quant_coeff[qscale];
1313 const int bias= intra ? (1<<QUANT_SHIFT)/3 : (1<<QUANT_SHIFT)/6;
1314 const unsigned int threshold1= (1<<QUANT_SHIFT) - bias - 1;
1315 const unsigned int threshold2= (threshold1<<1);
1321 const int dc_bias= intra ? (1<<(QUANT_SHIFT-2))/3 : (1<<(QUANT_SHIFT-2))/6;
1322 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT-2)) - dc_bias - 1;
1323 const unsigned int dc_threshold2= (dc_threshold1<<1);
1325 int level= block[0]*quant_coeff[qscale+18][0];
1326 if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1328 level= (dc_bias + level)>>(QUANT_SHIFT-2);
1331 level= (dc_bias - level)>>(QUANT_SHIFT-2);
1334 // last_non_zero = i;
1339 const int dc_bias= intra ? (1<<(QUANT_SHIFT+1))/3 : (1<<(QUANT_SHIFT+1))/6;
1340 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT+1)) - dc_bias - 1;
1341 const unsigned int dc_threshold2= (dc_threshold1<<1);
1343 int level= block[0]*quant_table[0];
1344 if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1346 level= (dc_bias + level)>>(QUANT_SHIFT+1);
1349 level= (dc_bias - level)>>(QUANT_SHIFT+1);
1352 // last_non_zero = i;
1365 const int j= scantable[i];
1366 int level= block[j]*quant_table[j];
1368 // if( bias+level >= (1<<(QMAT_SHIFT - 3))
1369 // || bias-level >= (1<<(QMAT_SHIFT - 3))){
1370 if(((unsigned)(level+threshold1))>threshold2){
1372 level= (bias + level)>>QUANT_SHIFT;
1375 level= (bias - level)>>QUANT_SHIFT;
1384 return last_non_zero;
1387 static void pred4x4_vertical_c(uint8_t *src, uint8_t *topright, int stride){
1388 const uint32_t a= ((uint32_t*)(src-stride))[0];
1389 ((uint32_t*)(src+0*stride))[0]= a;
1390 ((uint32_t*)(src+1*stride))[0]= a;
1391 ((uint32_t*)(src+2*stride))[0]= a;
1392 ((uint32_t*)(src+3*stride))[0]= a;
1395 static void pred4x4_horizontal_c(uint8_t *src, uint8_t *topright, int stride){
1396 ((uint32_t*)(src+0*stride))[0]= src[-1+0*stride]*0x01010101;
1397 ((uint32_t*)(src+1*stride))[0]= src[-1+1*stride]*0x01010101;
1398 ((uint32_t*)(src+2*stride))[0]= src[-1+2*stride]*0x01010101;
1399 ((uint32_t*)(src+3*stride))[0]= src[-1+3*stride]*0x01010101;
1402 static void pred4x4_dc_c(uint8_t *src, uint8_t *topright, int stride){
1403 const int dc= ( src[-stride] + src[1-stride] + src[2-stride] + src[3-stride]
1404 + src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 4) >>3;
1406 ((uint32_t*)(src+0*stride))[0]=
1407 ((uint32_t*)(src+1*stride))[0]=
1408 ((uint32_t*)(src+2*stride))[0]=
1409 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1412 static void pred4x4_left_dc_c(uint8_t *src, uint8_t *topright, int stride){
1413 const int dc= ( src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 2) >>2;
1415 ((uint32_t*)(src+0*stride))[0]=
1416 ((uint32_t*)(src+1*stride))[0]=
1417 ((uint32_t*)(src+2*stride))[0]=
1418 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1421 static void pred4x4_top_dc_c(uint8_t *src, uint8_t *topright, int stride){
1422 const int dc= ( src[-stride] + src[1-stride] + src[2-stride] + src[3-stride] + 2) >>2;
1424 ((uint32_t*)(src+0*stride))[0]=
1425 ((uint32_t*)(src+1*stride))[0]=
1426 ((uint32_t*)(src+2*stride))[0]=
1427 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1430 static void pred4x4_128_dc_c(uint8_t *src, uint8_t *topright, int stride){
1431 ((uint32_t*)(src+0*stride))[0]=
1432 ((uint32_t*)(src+1*stride))[0]=
1433 ((uint32_t*)(src+2*stride))[0]=
1434 ((uint32_t*)(src+3*stride))[0]= 128U*0x01010101U;
1438 #define LOAD_TOP_RIGHT_EDGE\
1439 const int t4= topright[0];\
1440 const int t5= topright[1];\
1441 const int t6= topright[2];\
1442 const int t7= topright[3];\
1444 #define LOAD_LEFT_EDGE\
1445 const int l0= src[-1+0*stride];\
1446 const int l1= src[-1+1*stride];\
1447 const int l2= src[-1+2*stride];\
1448 const int l3= src[-1+3*stride];\
1450 #define LOAD_TOP_EDGE\
1451 const int t0= src[ 0-1*stride];\
1452 const int t1= src[ 1-1*stride];\
1453 const int t2= src[ 2-1*stride];\
1454 const int t3= src[ 3-1*stride];\
1456 static void pred4x4_down_right_c(uint8_t *src, uint8_t *topright, int stride){
1457 const int lt= src[-1-1*stride];
1461 src[0+3*stride]=(l3 + 2*l2 + l1 + 2)>>2;
1463 src[1+3*stride]=(l2 + 2*l1 + l0 + 2)>>2;
1466 src[2+3*stride]=(l1 + 2*l0 + lt + 2)>>2;
1470 src[3+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1473 src[3+2*stride]=(lt + 2*t0 + t1 + 2)>>2;
1475 src[3+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1476 src[3+0*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1479 static void pred4x4_down_left_c(uint8_t *src, uint8_t *topright, int stride){
1484 src[0+0*stride]=(t0 + t2 + 2*t1 + 2)>>2;
1486 src[0+1*stride]=(t1 + t3 + 2*t2 + 2)>>2;
1489 src[0+2*stride]=(t2 + t4 + 2*t3 + 2)>>2;
1493 src[0+3*stride]=(t3 + t5 + 2*t4 + 2)>>2;
1496 src[1+3*stride]=(t4 + t6 + 2*t5 + 2)>>2;
1498 src[2+3*stride]=(t5 + t7 + 2*t6 + 2)>>2;
1499 src[3+3*stride]=(t6 + 3*t7 + 2)>>2;
1502 static void pred4x4_vertical_right_c(uint8_t *src, uint8_t *topright, int stride){
1503 const int lt= src[-1-1*stride];
1506 const __attribute__((unused)) int unu= l3;
1509 src[1+2*stride]=(lt + t0 + 1)>>1;
1511 src[2+2*stride]=(t0 + t1 + 1)>>1;
1513 src[3+2*stride]=(t1 + t2 + 1)>>1;
1514 src[3+0*stride]=(t2 + t3 + 1)>>1;
1516 src[1+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1518 src[2+3*stride]=(lt + 2*t0 + t1 + 2)>>2;
1520 src[3+3*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1521 src[3+1*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1522 src[0+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1523 src[0+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1526 static void pred4x4_vertical_left_c(uint8_t *src, uint8_t *topright, int stride){
1529 const __attribute__((unused)) int unu= t7;
1531 src[0+0*stride]=(t0 + t1 + 1)>>1;
1533 src[0+2*stride]=(t1 + t2 + 1)>>1;
1535 src[1+2*stride]=(t2 + t3 + 1)>>1;
1537 src[2+2*stride]=(t3 + t4+ 1)>>1;
1538 src[3+2*stride]=(t4 + t5+ 1)>>1;
1539 src[0+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1541 src[0+3*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1543 src[1+3*stride]=(t2 + 2*t3 + t4 + 2)>>2;
1545 src[2+3*stride]=(t3 + 2*t4 + t5 + 2)>>2;
1546 src[3+3*stride]=(t4 + 2*t5 + t6 + 2)>>2;
1549 static void pred4x4_horizontal_up_c(uint8_t *src, uint8_t *topright, int stride){
1552 src[0+0*stride]=(l0 + l1 + 1)>>1;
1553 src[1+0*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1555 src[0+1*stride]=(l1 + l2 + 1)>>1;
1557 src[1+1*stride]=(l1 + 2*l2 + l3 + 2)>>2;
1559 src[0+2*stride]=(l2 + l3 + 1)>>1;
1561 src[1+2*stride]=(l2 + 2*l3 + l3 + 2)>>2;
1570 static void pred4x4_horizontal_down_c(uint8_t *src, uint8_t *topright, int stride){
1571 const int lt= src[-1-1*stride];
1574 const __attribute__((unused)) int unu= t3;
1577 src[2+1*stride]=(lt + l0 + 1)>>1;
1579 src[3+1*stride]=(l0 + 2*lt + t0 + 2)>>2;
1580 src[2+0*stride]=(lt + 2*t0 + t1 + 2)>>2;
1581 src[3+0*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1583 src[2+2*stride]=(l0 + l1 + 1)>>1;
1585 src[3+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1587 src[2+3*stride]=(l1 + l2+ 1)>>1;
1589 src[3+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1590 src[0+3*stride]=(l2 + l3 + 1)>>1;
1591 src[1+3*stride]=(l1 + 2*l2 + l3 + 2)>>2;
1594 static void pred16x16_vertical_c(uint8_t *src, int stride){
1596 const uint32_t a= ((uint32_t*)(src-stride))[0];
1597 const uint32_t b= ((uint32_t*)(src-stride))[1];
1598 const uint32_t c= ((uint32_t*)(src-stride))[2];
1599 const uint32_t d= ((uint32_t*)(src-stride))[3];
1601 for(i=0; i<16; i++){
1602 ((uint32_t*)(src+i*stride))[0]= a;
1603 ((uint32_t*)(src+i*stride))[1]= b;
1604 ((uint32_t*)(src+i*stride))[2]= c;
1605 ((uint32_t*)(src+i*stride))[3]= d;
1609 static void pred16x16_horizontal_c(uint8_t *src, int stride){
1612 for(i=0; i<16; i++){
1613 ((uint32_t*)(src+i*stride))[0]=
1614 ((uint32_t*)(src+i*stride))[1]=
1615 ((uint32_t*)(src+i*stride))[2]=
1616 ((uint32_t*)(src+i*stride))[3]= src[-1+i*stride]*0x01010101;
1620 static void pred16x16_dc_c(uint8_t *src, int stride){
1624 dc+= src[-1+i*stride];
1631 dc= 0x01010101*((dc + 16)>>5);
1633 for(i=0; i<16; i++){
1634 ((uint32_t*)(src+i*stride))[0]=
1635 ((uint32_t*)(src+i*stride))[1]=
1636 ((uint32_t*)(src+i*stride))[2]=
1637 ((uint32_t*)(src+i*stride))[3]= dc;
1641 static void pred16x16_left_dc_c(uint8_t *src, int stride){
1645 dc+= src[-1+i*stride];
1648 dc= 0x01010101*((dc + 8)>>4);
1650 for(i=0; i<16; i++){
1651 ((uint32_t*)(src+i*stride))[0]=
1652 ((uint32_t*)(src+i*stride))[1]=
1653 ((uint32_t*)(src+i*stride))[2]=
1654 ((uint32_t*)(src+i*stride))[3]= dc;
1658 static void pred16x16_top_dc_c(uint8_t *src, int stride){
1664 dc= 0x01010101*((dc + 8)>>4);
1666 for(i=0; i<16; i++){
1667 ((uint32_t*)(src+i*stride))[0]=
1668 ((uint32_t*)(src+i*stride))[1]=
1669 ((uint32_t*)(src+i*stride))[2]=
1670 ((uint32_t*)(src+i*stride))[3]= dc;
1674 static void pred16x16_128_dc_c(uint8_t *src, int stride){
1677 for(i=0; i<16; i++){
1678 ((uint32_t*)(src+i*stride))[0]=
1679 ((uint32_t*)(src+i*stride))[1]=
1680 ((uint32_t*)(src+i*stride))[2]=
1681 ((uint32_t*)(src+i*stride))[3]= 0x01010101U*128U;
1685 static inline void pred16x16_plane_compat_c(uint8_t *src, int stride, const int svq3){
1688 uint8_t *cm = cropTbl + MAX_NEG_CROP;
1689 const uint8_t * const src0 = src+7-stride;
1690 const uint8_t *src1 = src+8*stride-1;
1691 const uint8_t *src2 = src1-2*stride; // == src+6*stride-1;
1692 int H = src0[1] - src0[-1];
1693 int V = src1[0] - src2[ 0];
1694 for(k=2; k<=8; ++k) {
1695 src1 += stride; src2 -= stride;
1696 H += k*(src0[k] - src0[-k]);
1697 V += k*(src1[0] - src2[ 0]);
1700 H = ( 5*(H/4) ) / 16;
1701 V = ( 5*(V/4) ) / 16;
1703 H = ( 5*H+32 ) >> 6;
1704 V = ( 5*V+32 ) >> 6;
1707 a = 16*(src1[0] + src2[16] + 1) - 7*(V+H);
1708 for(j=16; j>0; --j) {
1711 for(i=-16; i<0; i+=4) {
1712 src[16+i] = cm[ (b ) >> 5 ];
1713 src[17+i] = cm[ (b+ H) >> 5 ];
1714 src[18+i] = cm[ (b+2*H) >> 5 ];
1715 src[19+i] = cm[ (b+3*H) >> 5 ];
1722 static void pred16x16_plane_c(uint8_t *src, int stride){
1723 pred16x16_plane_compat_c(src, stride, 0);
1726 static void pred8x8_vertical_c(uint8_t *src, int stride){
1728 const uint32_t a= ((uint32_t*)(src-stride))[0];
1729 const uint32_t b= ((uint32_t*)(src-stride))[1];
1732 ((uint32_t*)(src+i*stride))[0]= a;
1733 ((uint32_t*)(src+i*stride))[1]= b;
1737 static void pred8x8_horizontal_c(uint8_t *src, int stride){
1741 ((uint32_t*)(src+i*stride))[0]=
1742 ((uint32_t*)(src+i*stride))[1]= src[-1+i*stride]*0x01010101;
1746 static void pred8x8_128_dc_c(uint8_t *src, int stride){
1750 ((uint32_t*)(src+i*stride))[0]=
1751 ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
1754 ((uint32_t*)(src+i*stride))[0]=
1755 ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
1759 static void pred8x8_left_dc_c(uint8_t *src, int stride){
1765 dc0+= src[-1+i*stride];
1766 dc2+= src[-1+(i+4)*stride];
1768 dc0= 0x01010101*((dc0 + 2)>>2);
1769 dc2= 0x01010101*((dc2 + 2)>>2);
1772 ((uint32_t*)(src+i*stride))[0]=
1773 ((uint32_t*)(src+i*stride))[1]= dc0;
1776 ((uint32_t*)(src+i*stride))[0]=
1777 ((uint32_t*)(src+i*stride))[1]= dc2;
1781 static void pred8x8_top_dc_c(uint8_t *src, int stride){
1787 dc0+= src[i-stride];
1788 dc1+= src[4+i-stride];
1790 dc0= 0x01010101*((dc0 + 2)>>2);
1791 dc1= 0x01010101*((dc1 + 2)>>2);
1794 ((uint32_t*)(src+i*stride))[0]= dc0;
1795 ((uint32_t*)(src+i*stride))[1]= dc1;
1798 ((uint32_t*)(src+i*stride))[0]= dc0;
1799 ((uint32_t*)(src+i*stride))[1]= dc1;
1804 static void pred8x8_dc_c(uint8_t *src, int stride){
1806 int dc0, dc1, dc2, dc3;
1810 dc0+= src[-1+i*stride] + src[i-stride];
1811 dc1+= src[4+i-stride];
1812 dc2+= src[-1+(i+4)*stride];
1814 dc3= 0x01010101*((dc1 + dc2 + 4)>>3);
1815 dc0= 0x01010101*((dc0 + 4)>>3);
1816 dc1= 0x01010101*((dc1 + 2)>>2);
1817 dc2= 0x01010101*((dc2 + 2)>>2);
1820 ((uint32_t*)(src+i*stride))[0]= dc0;
1821 ((uint32_t*)(src+i*stride))[1]= dc1;
1824 ((uint32_t*)(src+i*stride))[0]= dc2;
1825 ((uint32_t*)(src+i*stride))[1]= dc3;
1829 static void pred8x8_plane_c(uint8_t *src, int stride){
1832 uint8_t *cm = cropTbl + MAX_NEG_CROP;
1833 const uint8_t * const src0 = src+3-stride;
1834 const uint8_t *src1 = src+4*stride-1;
1835 const uint8_t *src2 = src1-2*stride; // == src+2*stride-1;
1836 int H = src0[1] - src0[-1];
1837 int V = src1[0] - src2[ 0];
1838 for(k=2; k<=4; ++k) {
1839 src1 += stride; src2 -= stride;
1840 H += k*(src0[k] - src0[-k]);
1841 V += k*(src1[0] - src2[ 0]);
1843 H = ( 17*H+16 ) >> 5;
1844 V = ( 17*V+16 ) >> 5;
1846 a = 16*(src1[0] + src2[8]+1) - 3*(V+H);
1847 for(j=8; j>0; --j) {
1850 src[0] = cm[ (b ) >> 5 ];
1851 src[1] = cm[ (b+ H) >> 5 ];
1852 src[2] = cm[ (b+2*H) >> 5 ];
1853 src[3] = cm[ (b+3*H) >> 5 ];
1854 src[4] = cm[ (b+4*H) >> 5 ];
1855 src[5] = cm[ (b+5*H) >> 5 ];
1856 src[6] = cm[ (b+6*H) >> 5 ];
1857 src[7] = cm[ (b+7*H) >> 5 ];
1862 static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
1863 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1864 int src_x_offset, int src_y_offset,
1865 qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
1866 MpegEncContext * const s = &h->s;
1867 const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
1868 const int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
1869 const int luma_xy= (mx&3) + ((my&3)<<2);
1870 uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*s->linesize;
1871 uint8_t * src_cb= pic->data[1] + (mx>>3) + (my>>3)*s->uvlinesize;
1872 uint8_t * src_cr= pic->data[2] + (mx>>3) + (my>>3)*s->uvlinesize;
1873 int extra_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16; //FIXME increase edge?, IMHO not worth it
1874 int extra_height= extra_width;
1876 const int full_mx= mx>>2;
1877 const int full_my= my>>2;
1879 assert(pic->data[0]);
1881 if(mx&7) extra_width -= 3;
1882 if(my&7) extra_height -= 3;
1884 if( full_mx < 0-extra_width
1885 || full_my < 0-extra_height
1886 || full_mx + 16/*FIXME*/ > s->width + extra_width
1887 || full_my + 16/*FIXME*/ > s->height + extra_height){
1888 ff_emulated_edge_mc(s, src_y - 2 - 2*s->linesize, s->linesize, 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, s->width, s->height);
1889 src_y= s->edge_emu_buffer + 2 + 2*s->linesize;
1893 qpix_op[luma_xy](dest_y, src_y, s->linesize); //FIXME try variable height perhaps?
1895 qpix_op[luma_xy](dest_y + delta, src_y + delta, s->linesize);
1898 if(s->flags&CODEC_FLAG_GRAY) return;
1901 ff_emulated_edge_mc(s, src_cb, s->uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), s->width>>1, s->height>>1);
1902 src_cb= s->edge_emu_buffer;
1904 chroma_op(dest_cb, src_cb, s->uvlinesize, chroma_height, mx&7, my&7);
1907 ff_emulated_edge_mc(s, src_cr, s->uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), s->width>>1, s->height>>1);
1908 src_cr= s->edge_emu_buffer;
1910 chroma_op(dest_cr, src_cr, s->uvlinesize, chroma_height, mx&7, my&7);
1913 static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
1914 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1915 int x_offset, int y_offset,
1916 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
1917 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
1918 int list0, int list1){
1919 MpegEncContext * const s = &h->s;
1920 qpel_mc_func *qpix_op= qpix_put;
1921 h264_chroma_mc_func chroma_op= chroma_put;
1923 dest_y += 2*x_offset + 2*y_offset*s-> linesize;
1924 dest_cb += x_offset + y_offset*s->uvlinesize;
1925 dest_cr += x_offset + y_offset*s->uvlinesize;
1926 x_offset += 8*s->mb_x;
1927 y_offset += 8*s->mb_y;
1930 Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
1931 mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
1932 dest_y, dest_cb, dest_cr, x_offset, y_offset,
1933 qpix_op, chroma_op);
1936 chroma_op= chroma_avg;
1940 Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
1941 mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
1942 dest_y, dest_cb, dest_cr, x_offset, y_offset,
1943 qpix_op, chroma_op);
1947 static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1948 qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
1949 qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg)){
1950 MpegEncContext * const s = &h->s;
1951 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
1952 const int mb_type= s->current_picture.mb_type[mb_xy];
1954 assert(IS_INTER(mb_type));
1956 if(IS_16X16(mb_type)){
1957 mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
1958 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
1959 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1960 }else if(IS_16X8(mb_type)){
1961 mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
1962 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
1963 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1964 mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
1965 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
1966 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
1967 }else if(IS_8X16(mb_type)){
1968 mc_part(h, 0, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 0, 0,
1969 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
1970 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1971 mc_part(h, 4, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 4, 0,
1972 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
1973 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
1977 assert(IS_8X8(mb_type));
1980 const int sub_mb_type= h->sub_mb_type[i];
1982 int x_offset= (i&1)<<2;
1983 int y_offset= (i&2)<<1;
1985 if(IS_SUB_8X8(sub_mb_type)){
1986 mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
1987 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
1988 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1989 }else if(IS_SUB_8X4(sub_mb_type)){
1990 mc_part(h, n , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
1991 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
1992 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1993 mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
1994 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
1995 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1996 }else if(IS_SUB_4X8(sub_mb_type)){
1997 mc_part(h, n , 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
1998 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
1999 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2000 mc_part(h, n+1, 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
2001 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2002 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2005 assert(IS_SUB_4X4(sub_mb_type));
2007 int sub_x_offset= x_offset + 2*(j&1);
2008 int sub_y_offset= y_offset + (j&2);
2009 mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
2010 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2011 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2018 static void decode_init_vlc(H264Context *h){
2019 static int done = 0;
2025 init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5,
2026 &chroma_dc_coeff_token_len [0], 1, 1,
2027 &chroma_dc_coeff_token_bits[0], 1, 1);
2030 init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17,
2031 &coeff_token_len [i][0], 1, 1,
2032 &coeff_token_bits[i][0], 1, 1);
2036 init_vlc(&chroma_dc_total_zeros_vlc[i], CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
2037 &chroma_dc_total_zeros_len [i][0], 1, 1,
2038 &chroma_dc_total_zeros_bits[i][0], 1, 1);
2040 for(i=0; i<15; i++){
2041 init_vlc(&total_zeros_vlc[i], TOTAL_ZEROS_VLC_BITS, 16,
2042 &total_zeros_len [i][0], 1, 1,
2043 &total_zeros_bits[i][0], 1, 1);
2047 init_vlc(&run_vlc[i], RUN_VLC_BITS, 7,
2048 &run_len [i][0], 1, 1,
2049 &run_bits[i][0], 1, 1);
2051 init_vlc(&run7_vlc, RUN7_VLC_BITS, 16,
2052 &run_len [6][0], 1, 1,
2053 &run_bits[6][0], 1, 1);
2058 * Sets the intra prediction function pointers.
2060 static void init_pred_ptrs(H264Context *h){
2061 // MpegEncContext * const s = &h->s;
2063 h->pred4x4[VERT_PRED ]= pred4x4_vertical_c;
2064 h->pred4x4[HOR_PRED ]= pred4x4_horizontal_c;
2065 h->pred4x4[DC_PRED ]= pred4x4_dc_c;
2066 h->pred4x4[DIAG_DOWN_LEFT_PRED ]= pred4x4_down_left_c;
2067 h->pred4x4[DIAG_DOWN_RIGHT_PRED]= pred4x4_down_right_c;
2068 h->pred4x4[VERT_RIGHT_PRED ]= pred4x4_vertical_right_c;
2069 h->pred4x4[HOR_DOWN_PRED ]= pred4x4_horizontal_down_c;
2070 h->pred4x4[VERT_LEFT_PRED ]= pred4x4_vertical_left_c;
2071 h->pred4x4[HOR_UP_PRED ]= pred4x4_horizontal_up_c;
2072 h->pred4x4[LEFT_DC_PRED ]= pred4x4_left_dc_c;
2073 h->pred4x4[TOP_DC_PRED ]= pred4x4_top_dc_c;
2074 h->pred4x4[DC_128_PRED ]= pred4x4_128_dc_c;
2076 h->pred8x8[DC_PRED8x8 ]= pred8x8_dc_c;
2077 h->pred8x8[VERT_PRED8x8 ]= pred8x8_vertical_c;
2078 h->pred8x8[HOR_PRED8x8 ]= pred8x8_horizontal_c;
2079 h->pred8x8[PLANE_PRED8x8 ]= pred8x8_plane_c;
2080 h->pred8x8[LEFT_DC_PRED8x8]= pred8x8_left_dc_c;
2081 h->pred8x8[TOP_DC_PRED8x8 ]= pred8x8_top_dc_c;
2082 h->pred8x8[DC_128_PRED8x8 ]= pred8x8_128_dc_c;
2084 h->pred16x16[DC_PRED8x8 ]= pred16x16_dc_c;
2085 h->pred16x16[VERT_PRED8x8 ]= pred16x16_vertical_c;
2086 h->pred16x16[HOR_PRED8x8 ]= pred16x16_horizontal_c;
2087 h->pred16x16[PLANE_PRED8x8 ]= pred16x16_plane_c;
2088 h->pred16x16[LEFT_DC_PRED8x8]= pred16x16_left_dc_c;
2089 h->pred16x16[TOP_DC_PRED8x8 ]= pred16x16_top_dc_c;
2090 h->pred16x16[DC_128_PRED8x8 ]= pred16x16_128_dc_c;
2094 #define CHECKED_ALLOCZ(p, size)\
2096 p= av_mallocz(size);\
2103 static void free_tables(H264Context *h){
2104 av_freep(&h->intra4x4_pred_mode);
2105 av_freep(&h->non_zero_count);
2106 av_freep(&h->slice_table_base);
2107 h->slice_table= NULL;
2109 av_freep(&h->mb2b_xy);
2110 av_freep(&h->mb2b8_xy);
2115 * needs widzh/height
2117 static int alloc_tables(H264Context *h){
2118 MpegEncContext * const s = &h->s;
2119 const int big_mb_num= s->mb_stride * (s->mb_height+1);
2122 CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8 * sizeof(uint8_t))
2123 CHECKED_ALLOCZ(h->non_zero_count , big_mb_num * 16 * sizeof(uint8_t))
2124 CHECKED_ALLOCZ(h->slice_table_base , big_mb_num * sizeof(uint8_t))
2126 memset(h->slice_table_base, -1, big_mb_num * sizeof(uint8_t));
2127 h->slice_table= h->slice_table_base + s->mb_stride + 1;
2129 CHECKED_ALLOCZ(h->mb2b_xy , big_mb_num * sizeof(uint16_t));
2130 CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint16_t));
2131 for(y=0; y<s->mb_height; y++){
2132 for(x=0; x<s->mb_width; x++){
2133 const int mb_xy= x + y*s->mb_stride;
2134 const int b_xy = 4*x + 4*y*h->b_stride;
2135 const int b8_xy= 2*x + 2*y*h->b8_stride;
2137 h->mb2b_xy [mb_xy]= b_xy;
2138 h->mb2b8_xy[mb_xy]= b8_xy;
2148 static void common_init(H264Context *h){
2149 MpegEncContext * const s = &h->s;
2151 s->width = s->avctx->width;
2152 s->height = s->avctx->height;
2153 s->codec_id= s->avctx->codec->id;
2157 s->decode=1; //FIXME
2160 static int decode_init(AVCodecContext *avctx){
2161 H264Context *h= avctx->priv_data;
2162 MpegEncContext * const s = &h->s;
2167 s->out_format = FMT_H264;
2168 s->workaround_bugs= avctx->workaround_bugs;
2171 s->progressive_sequence=1;
2172 // s->decode_mb= ff_h263_decode_mb;
2174 avctx->pix_fmt= PIX_FMT_YUV420P;
2181 static void frame_start(H264Context *h){
2182 MpegEncContext * const s = &h->s;
2185 MPV_frame_start(s, s->avctx);
2186 ff_er_frame_start(s);
2189 assert(s->linesize && s->uvlinesize);
2191 for(i=0; i<16; i++){
2192 h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
2193 h->chroma_subblock_offset[i]= 2*((scan8[i] - scan8[0])&7) + 2*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2196 h->block_offset[16+i]=
2197 h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2200 // s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
2203 static void hl_decode_mb(H264Context *h){
2204 MpegEncContext * const s = &h->s;
2205 const int mb_x= s->mb_x;
2206 const int mb_y= s->mb_y;
2207 const int mb_xy= mb_x + mb_y*s->mb_stride;
2208 const int mb_type= s->current_picture.mb_type[mb_xy];
2209 uint8_t *dest_y, *dest_cb, *dest_cr;
2210 int linesize, uvlinesize /*dct_offset*/;
2219 dest_y = s->current_picture.data[0] + (mb_y * 16* s->linesize ) + mb_x * 16;
2220 dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2221 dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2223 if (h->mb_field_decoding_flag) {
2224 linesize = s->linesize * 2;
2225 uvlinesize = s->uvlinesize * 2;
2226 if(mb_y&1){ //FIXME move out of this func?
2227 dest_y -= s->linesize*15;
2228 dest_cb-= s->linesize*7;
2229 dest_cr-= s->linesize*7;
2232 linesize = s->linesize;
2233 uvlinesize = s->uvlinesize;
2234 // dct_offset = s->linesize * 16;
2237 if(IS_INTRA(mb_type)){
2238 if(!(s->flags&CODEC_FLAG_GRAY)){
2239 h->pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
2240 h->pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
2243 if(IS_INTRA4x4(mb_type)){
2245 for(i=0; i<16; i++){
2246 uint8_t * const ptr= dest_y + h->block_offset[i];
2247 uint8_t *topright= ptr + 4 - linesize;
2248 const int topright_avail= (h->topright_samples_available<<i)&0x8000;
2249 const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
2252 if(!topright_avail){
2253 tr= ptr[3 - linesize]*0x01010101;
2254 topright= (uint8_t*) &tr;
2257 h->pred4x4[ dir ](ptr, topright, linesize);
2258 if(h->non_zero_count_cache[ scan8[i] ]){
2259 if(s->codec_id == CODEC_ID_H264)
2260 h264_add_idct_c(ptr, h->mb + i*16, linesize);
2262 svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
2267 h->pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
2268 if(s->codec_id == CODEC_ID_H264)
2269 h264_luma_dc_dequant_idct_c(h->mb, s->qscale);
2271 svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
2273 }else if(s->codec_id == CODEC_ID_H264){
2274 hl_motion(h, dest_y, dest_cb, dest_cr,
2275 s->dsp.put_h264_qpel_pixels_tab, s->dsp.put_h264_chroma_pixels_tab,
2276 s->dsp.avg_h264_qpel_pixels_tab, s->dsp.avg_h264_chroma_pixels_tab);
2280 if(!IS_INTRA4x4(mb_type)){
2281 for(i=0; i<16; i++){
2282 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
2283 uint8_t * const ptr= dest_y + h->block_offset[i];
2284 if(s->codec_id == CODEC_ID_H264)
2285 h264_add_idct_c(ptr, h->mb + i*16, linesize);
2287 svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
2292 if(!(s->flags&CODEC_FLAG_GRAY)){
2293 chroma_dc_dequant_idct_c(h->mb + 16*16, h->chroma_qp);
2294 chroma_dc_dequant_idct_c(h->mb + 16*16+4*16, h->chroma_qp);
2295 for(i=16; i<16+4; i++){
2296 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2297 uint8_t * const ptr= dest_cb + h->block_offset[i];
2298 if(s->codec_id == CODEC_ID_H264)
2299 h264_add_idct_c(ptr, h->mb + i*16, uvlinesize);
2301 svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
2304 for(i=20; i<20+4; i++){
2305 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2306 uint8_t * const ptr= dest_cr + h->block_offset[i];
2307 if(s->codec_id == CODEC_ID_H264)
2308 h264_add_idct_c(ptr, h->mb + i*16, uvlinesize);
2310 svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
2316 static void decode_mb_cabac(H264Context *h){
2317 // MpegEncContext * const s = &h->s;
2321 * fills the default_ref_list.
2323 static int fill_default_ref_list(H264Context *h){
2324 MpegEncContext * const s = &h->s;
2326 Picture sorted_short_ref[16];
2328 if(h->slice_type==B_TYPE){
2332 for(out_i=0; out_i<h->short_ref_count; out_i++){
2336 for(i=0; i<h->short_ref_count; i++){
2337 const int poc= h->short_ref[i]->poc;
2338 if(poc > limit && poc < best_poc){
2344 assert(best_i != -1);
2347 sorted_short_ref[out_i]= *h->short_ref[best_i];
2351 if(s->picture_structure == PICT_FRAME){
2352 if(h->slice_type==B_TYPE){
2353 const int current_poc= s->current_picture_ptr->poc;
2356 for(list=0; list<2; list++){
2359 for(i=0; i<h->short_ref_count && index < h->ref_count[list]; i++){
2360 const int i2= list ? h->short_ref_count - i - 1 : i;
2361 const int poc= sorted_short_ref[i2].poc;
2363 if(sorted_short_ref[i2].reference != 3) continue; //FIXME refernce field shit
2365 if((list==1 && poc > current_poc) || (list==0 && poc < current_poc)){
2366 h->default_ref_list[list][index ]= sorted_short_ref[i2];
2367 h->default_ref_list[list][index++].pic_id= sorted_short_ref[i2].frame_num;
2371 for(i=0; i<h->long_ref_count && index < h->ref_count[ list ]; i++){
2372 if(h->long_ref[i]->reference != 3) continue;
2374 h->default_ref_list[ list ][index ]= *h->long_ref[i];
2375 h->default_ref_list[ list ][index++].pic_id= i;;
2378 if(h->long_ref_count > 1 && h->short_ref_count==0){
2379 Picture temp= h->default_ref_list[1][0];
2380 h->default_ref_list[1][0] = h->default_ref_list[1][1];
2381 h->default_ref_list[1][0] = temp;
2384 if(index < h->ref_count[ list ])
2385 memset(&h->default_ref_list[list][index], 0, sizeof(Picture)*(h->ref_count[ list ] - index));
2389 for(i=0; i<h->short_ref_count && index < h->ref_count[0]; i++){
2390 if(h->short_ref[i]->reference != 3) continue; //FIXME refernce field shit
2391 h->default_ref_list[0][index ]= *h->short_ref[i];
2392 h->default_ref_list[0][index++].pic_id= h->short_ref[i]->frame_num;
2394 for(i=0; i<h->long_ref_count && index < h->ref_count[0]; i++){
2395 if(h->long_ref[i]->reference != 3) continue;
2396 h->default_ref_list[0][index ]= *h->long_ref[i];
2397 h->default_ref_list[0][index++].pic_id= i;;
2399 if(index < h->ref_count[0])
2400 memset(&h->default_ref_list[0][index], 0, sizeof(Picture)*(h->ref_count[0] - index));
2403 if(h->slice_type==B_TYPE){
2405 //FIXME second field balh
2411 static int decode_ref_pic_list_reordering(H264Context *h){
2412 MpegEncContext * const s = &h->s;
2415 if(h->slice_type==I_TYPE || h->slice_type==SI_TYPE) return 0; //FIXME move beofre func
2417 for(list=0; list<2; list++){
2418 memcpy(h->ref_list[list], h->default_ref_list[list], sizeof(Picture)*h->ref_count[list]);
2420 if(get_bits1(&s->gb)){
2421 int pred= h->curr_pic_num;
2424 for(index=0; ; index++){
2425 int reordering_of_pic_nums_idc= get_ue_golomb(&s->gb);
2430 if(index >= h->ref_count[list]){
2431 fprintf(stderr, "reference count overflow\n");
2435 if(reordering_of_pic_nums_idc<3){
2436 if(reordering_of_pic_nums_idc<2){
2437 const int abs_diff_pic_num= get_ue_golomb(&s->gb) + 1;
2439 if(abs_diff_pic_num >= h->max_pic_num){
2440 fprintf(stderr, "abs_diff_pic_num overflow\n");
2444 if(reordering_of_pic_nums_idc == 0) pred-= abs_diff_pic_num;
2445 else pred+= abs_diff_pic_num;
2446 pred &= h->max_pic_num - 1;
2448 for(i= h->ref_count[list]-1; i>=index; i--){
2449 if(h->ref_list[list][i].pic_id == pred && h->ref_list[list][i].long_ref==0)
2453 pic_id= get_ue_golomb(&s->gb); //long_term_pic_idx
2455 for(i= h->ref_count[list]-1; i>=index; i--){
2456 if(h->ref_list[list][i].pic_id == pic_id && h->ref_list[list][i].long_ref==1)
2462 fprintf(stderr, "reference picture missing during reorder\n");
2463 memset(&h->ref_list[list][index], 0, sizeof(Picture)); //FIXME
2464 }else if(i > index){
2465 Picture tmp= h->ref_list[list][i];
2466 for(; i>index; i--){
2467 h->ref_list[list][i]= h->ref_list[list][i-1];
2469 h->ref_list[list][index]= tmp;
2471 }else if(reordering_of_pic_nums_idc==3)
2474 fprintf(stderr, "illegal reordering_of_pic_nums_idc\n");
2480 if(h->slice_type!=B_TYPE) break;
2485 static int pred_weight_table(H264Context *h){
2486 MpegEncContext * const s = &h->s;
2489 h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
2490 h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
2492 for(list=0; list<2; list++){
2493 for(i=0; i<h->ref_count[list]; i++){
2494 int luma_weight_flag, chroma_weight_flag;
2496 luma_weight_flag= get_bits1(&s->gb);
2497 if(luma_weight_flag){
2498 h->luma_weight[list][i]= get_se_golomb(&s->gb);
2499 h->luma_offset[list][i]= get_se_golomb(&s->gb);
2502 chroma_weight_flag= get_bits1(&s->gb);
2503 if(chroma_weight_flag){
2506 h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);
2507 h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);
2511 if(h->slice_type != B_TYPE) break;
2517 * instantaneos decoder refresh.
2519 static void idr(H264Context *h){
2522 for(i=0; i<h->long_ref_count; i++){
2523 h->long_ref[i]->reference=0;
2524 h->long_ref[i]= NULL;
2526 h->long_ref_count=0;
2528 for(i=0; i<h->short_ref_count; i++){
2529 h->short_ref[i]->reference=0;
2530 h->short_ref[i]= NULL;
2532 h->short_ref_count=0;
2537 * @return the removed picture or NULL if an error occures
2539 static Picture * remove_short(H264Context *h, int frame_num){
2540 MpegEncContext * const s = &h->s;
2543 if(s->avctx->debug&FF_DEBUG_MMCO)
2544 printf("remove short %d count %d\n", frame_num, h->short_ref_count);
2546 for(i=0; i<h->short_ref_count; i++){
2547 Picture *pic= h->short_ref[i];
2548 if(s->avctx->debug&FF_DEBUG_MMCO)
2549 printf("%d %d %X\n", i, pic->frame_num, (int)pic);
2550 if(pic->frame_num == frame_num){
2551 h->short_ref[i]= NULL;
2552 memmove(&h->short_ref[i], &h->short_ref[i+1], (h->short_ref_count - i - 1)*sizeof(Picture*));
2553 h->short_ref_count--;
2562 * @return the removed picture or NULL if an error occures
2564 static Picture * remove_long(H264Context *h, int i){
2567 if(i >= h->long_ref_count) return NULL;
2568 pic= h->long_ref[i];
2569 if(pic==NULL) return NULL;
2571 h->long_ref[i]= NULL;
2572 memmove(&h->long_ref[i], &h->long_ref[i+1], (h->long_ref_count - i - 1)*sizeof(Picture*));
2573 h->long_ref_count--;
2579 * Executes the reference picture marking (memory management control operations).
2581 static int execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count){
2582 MpegEncContext * const s = &h->s;
2584 int current_is_long=0;
2587 if((s->avctx->debug&FF_DEBUG_MMCO) && mmco_count==0)
2588 printf("no mmco here\n");
2590 for(i=0; i<mmco_count; i++){
2591 if(s->avctx->debug&FF_DEBUG_MMCO)
2592 printf("mmco:%d %d %d\n", h->mmco[i].opcode, h->mmco[i].short_frame_num, h->mmco[i].long_index);
2594 switch(mmco[i].opcode){
2595 case MMCO_SHORT2UNUSED:
2596 pic= remove_short(h, mmco[i].short_frame_num);
2597 if(pic==NULL) return -1;
2600 case MMCO_SHORT2LONG:
2601 pic= remove_long(h, mmco[i].long_index);
2602 if(pic) pic->reference=0;
2604 h->long_ref[ mmco[i].long_index ]= remove_short(h, mmco[i].short_frame_num);
2605 h->long_ref[ mmco[i].long_index ]->long_ref=1;
2607 case MMCO_LONG2UNUSED:
2608 pic= remove_long(h, mmco[i].long_index);
2609 if(pic==NULL) return -1;
2613 pic= remove_long(h, mmco[i].long_index);
2614 if(pic) pic->reference=0;
2616 h->long_ref[ mmco[i].long_index ]= s->current_picture_ptr;
2617 h->long_ref[ mmco[i].long_index ]->long_ref=1;
2618 h->long_ref_count++;
2622 case MMCO_SET_MAX_LONG:
2623 assert(mmco[i].long_index <= 16);
2624 while(mmco[i].long_index < h->long_ref_count){
2625 pic= remove_long(h, mmco[i].long_index);
2628 while(mmco[i].long_index > h->long_ref_count){
2629 h->long_ref[ h->long_ref_count++ ]= NULL;
2633 while(h->short_ref_count){
2634 pic= remove_short(h, h->short_ref[0]->frame_num);
2637 while(h->long_ref_count){
2638 pic= remove_long(h, h->long_ref_count-1);
2646 if(!current_is_long){
2647 pic= remove_short(h, s->current_picture_ptr->frame_num);
2650 fprintf(stderr, "illegal short term buffer state detected\n");
2653 if(h->short_ref_count)
2654 memmove(&h->short_ref[1], &h->short_ref[0], h->short_ref_count*sizeof(Picture*));
2656 h->short_ref[0]= s->current_picture_ptr;
2657 h->short_ref[0]->long_ref=0;
2658 h->short_ref_count++;
2664 static int decode_ref_pic_marking(H264Context *h){
2665 MpegEncContext * const s = &h->s;
2668 if(h->nal_unit_type == NAL_IDR_SLICE){ //FIXME fields
2669 s->broken_link= get_bits1(&s->gb) -1;
2670 h->mmco[0].long_index= get_bits1(&s->gb) - 1; // current_long_term_idx
2671 if(h->mmco[0].long_index == -1)
2674 h->mmco[0].opcode= MMCO_LONG;
2678 if(get_bits1(&s->gb)){ // adaptive_ref_pic_marking_mode_flag
2679 for(i= h->mmco_index; i<MAX_MMCO_COUNT; i++) {
2680 MMCOOpcode opcode= get_ue_golomb(&s->gb);;
2682 h->mmco[i].opcode= opcode;
2683 if(opcode==MMCO_SHORT2UNUSED || opcode==MMCO_SHORT2LONG){
2684 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
2685 /* if(h->mmco[i].short_frame_num >= h->short_ref_count || h->short_ref[ h->mmco[i].short_frame_num ] == NULL){
2686 fprintf(stderr, "illegal short ref in memory management control operation %d\n", mmco);
2690 if(opcode==MMCO_SHORT2LONG || opcode==MMCO_LONG2UNUSED || opcode==MMCO_LONG || opcode==MMCO_SET_MAX_LONG){
2691 h->mmco[i].long_index= get_ue_golomb(&s->gb);
2692 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){
2693 fprintf(stderr, "illegal long ref in memory management control operation %d\n", opcode);
2698 if(opcode > MMCO_LONG){
2699 fprintf(stderr, "illegal memory management control operation %d\n", opcode);
2705 assert(h->long_ref_count + h->short_ref_count <= h->sps.ref_frame_count);
2707 if(h->long_ref_count + h->short_ref_count == h->sps.ref_frame_count){ //FIXME fields
2708 h->mmco[0].opcode= MMCO_SHORT2UNUSED;
2709 h->mmco[0].short_frame_num= h->short_ref[ h->short_ref_count - 1 ]->frame_num;
2719 static int init_poc(H264Context *h){
2720 MpegEncContext * const s = &h->s;
2721 const int max_frame_num= 1<<h->sps.log2_max_frame_num;
2724 if(h->nal_unit_type == NAL_IDR_SLICE){
2725 h->frame_num_offset= 0;
2727 if(h->frame_num < h->prev_frame_num)
2728 h->frame_num_offset= h->prev_frame_num_offset + max_frame_num;
2730 h->frame_num_offset= h->prev_frame_num_offset;
2733 if(h->sps.poc_type==0){
2734 const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
2736 if (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
2737 h->poc_msb = h->prev_poc_msb + max_poc_lsb;
2738 else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
2739 h->poc_msb = h->prev_poc_msb - max_poc_lsb;
2741 h->poc_msb = h->prev_poc_msb;
2742 //printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
2744 field_poc[1] = h->poc_msb + h->poc_lsb;
2745 if(s->picture_structure == PICT_FRAME)
2746 field_poc[1] += h->delta_poc_bottom;
2747 }else if(h->sps.poc_type==1){
2748 int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
2751 if(h->sps.poc_cycle_length != 0)
2752 abs_frame_num = h->frame_num_offset + h->frame_num;
2756 if(h->nal_ref_idc==0 && abs_frame_num > 0)
2759 expected_delta_per_poc_cycle = 0;
2760 for(i=0; i < h->sps.poc_cycle_length; i++)
2761 expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
2763 if(abs_frame_num > 0){
2764 int poc_cycle_cnt = (abs_frame_num - 1) / h->sps.poc_cycle_length;
2765 int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
2767 expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
2768 for(i = 0; i <= frame_num_in_poc_cycle; i++)
2769 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
2773 if(h->nal_ref_idc == 0)
2774 expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
2776 field_poc[0] = expectedpoc + h->delta_poc[0];
2777 field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
2779 if(s->picture_structure == PICT_FRAME)
2780 field_poc[1] += h->delta_poc[1];
2783 if(h->nal_unit_type == NAL_IDR_SLICE){
2786 if(h->nal_ref_idc) poc= 2*(h->frame_num_offset + h->frame_num);
2787 else poc= 2*(h->frame_num_offset + h->frame_num) - 1;
2793 if(s->picture_structure != PICT_BOTTOM_FIELD)
2794 s->current_picture_ptr->field_poc[0]= field_poc[0];
2795 if(s->picture_structure != PICT_TOP_FIELD)
2796 s->current_picture_ptr->field_poc[1]= field_poc[1];
2797 if(s->picture_structure == PICT_FRAME) // FIXME field pix?
2798 s->current_picture_ptr->poc= FFMIN(field_poc[0], field_poc[1]);
2804 * decodes a slice header.
2805 * this will allso call MPV_common_init() and frame_start() as needed
2807 static int decode_slice_header(H264Context *h){
2808 MpegEncContext * const s = &h->s;
2809 int first_mb_in_slice, pps_id;
2810 int num_ref_idx_active_override_flag;
2811 static const uint8_t slice_type_map[5]= {P_TYPE, B_TYPE, I_TYPE, SP_TYPE, SI_TYPE};
2814 s->current_picture.reference= h->nal_ref_idc != 0;
2816 first_mb_in_slice= get_ue_golomb(&s->gb);
2818 h->slice_type= get_ue_golomb(&s->gb);
2819 if(h->slice_type > 9){
2820 fprintf(stderr, "slice type too large (%d) at %d %d\n", h->slice_type, s->mb_x, s->mb_y);
2822 if(h->slice_type > 4){
2824 h->slice_type_fixed=1;
2826 h->slice_type_fixed=0;
2828 h->slice_type= slice_type_map[ h->slice_type ];
2830 s->pict_type= h->slice_type; // to make a few old func happy, its wrong though
2832 pps_id= get_ue_golomb(&s->gb);
2834 fprintf(stderr, "pps_id out of range\n");
2837 h->pps= h->pps_buffer[pps_id];
2838 if(h->pps.slice_group_count == 0){
2839 fprintf(stderr, "non existing PPS referenced\n");
2843 h->sps= h->sps_buffer[ h->pps.sps_id ];
2844 if(h->sps.log2_max_frame_num == 0){
2845 fprintf(stderr, "non existing SPS referenced\n");
2849 s->mb_width= h->sps.mb_width;
2850 s->mb_height= h->sps.mb_height;
2852 h->b_stride= s->mb_width*4;
2853 h->b8_stride= s->mb_width*2;
2855 s->mb_x = first_mb_in_slice % s->mb_width;
2856 s->mb_y = first_mb_in_slice / s->mb_width; //FIXME AFFW
2858 s->width = 16*s->mb_width - 2*(h->pps.crop_left + h->pps.crop_right );
2859 if(h->sps.frame_mbs_only_flag)
2860 s->height= 16*s->mb_height - 2*(h->pps.crop_top + h->pps.crop_bottom);
2862 s->height= 16*s->mb_height - 4*(h->pps.crop_top + h->pps.crop_bottom); //FIXME recheck
2864 if(h->pps.crop_left || h->pps.crop_top){
2865 fprintf(stderr, "insane croping not completly supported, this could look slightly wrong ...\n");
2868 if(s->aspected_height) //FIXME emms at end of slice ?
2869 new_aspect= h->sps.sar_width*s->width / (float)(s->height*h->sps.sar_height);
2873 if (s->context_initialized
2874 && ( s->width != s->avctx->width || s->height != s->avctx->height
2875 || ABS(new_aspect - s->avctx->aspect_ratio) > 0.001)) {
2879 if (!s->context_initialized) {
2880 if (MPV_common_init(s) < 0)
2885 s->avctx->width = s->width;
2886 s->avctx->height = s->height;
2887 s->avctx->aspect_ratio= new_aspect;
2890 if(first_mb_in_slice == 0){
2894 s->current_picture_ptr->frame_num= //FIXME frame_num cleanup
2895 h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
2897 if(h->sps.frame_mbs_only_flag){
2898 s->picture_structure= PICT_FRAME;
2900 if(get_bits1(&s->gb)) //field_pic_flag
2901 s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
2903 s->picture_structure= PICT_FRAME;
2906 if(s->picture_structure==PICT_FRAME){
2907 h->curr_pic_num= h->frame_num;
2908 h->max_pic_num= 1<< h->sps.log2_max_frame_num;
2910 h->curr_pic_num= 2*h->frame_num;
2911 h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
2914 if(h->nal_unit_type == NAL_IDR_SLICE){
2915 int idr_pic_id= get_ue_golomb(&s->gb);
2918 if(h->sps.poc_type==0){
2919 h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
2921 if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
2922 h->delta_poc_bottom= get_se_golomb(&s->gb);
2926 if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
2927 h->delta_poc[0]= get_se_golomb(&s->gb);
2929 if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
2930 h->delta_poc[1]= get_se_golomb(&s->gb);
2935 if(h->pps.redundant_pic_cnt_present){
2936 h->redundant_pic_count= get_ue_golomb(&s->gb);
2939 //set defaults, might be overriden a few line later
2940 h->ref_count[0]= h->pps.ref_count[0];
2941 h->ref_count[1]= h->pps.ref_count[1];
2943 if(h->slice_type == P_TYPE || h->slice_type == SP_TYPE || h->slice_type == B_TYPE){
2944 if(h->slice_type == B_TYPE){
2945 h->direct_spatial_mv_pred= get_bits1(&s->gb);
2947 num_ref_idx_active_override_flag= get_bits1(&s->gb);
2949 if(num_ref_idx_active_override_flag){
2950 h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
2951 if(h->slice_type==B_TYPE)
2952 h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
2954 if(h->ref_count[0] > 32 || h->ref_count[1] > 32){
2955 fprintf(stderr, "reference overflow\n");
2961 if(first_mb_in_slice == 0){
2962 fill_default_ref_list(h);
2965 decode_ref_pic_list_reordering(h);
2967 if( (h->pps.weighted_pred && (h->slice_type == P_TYPE || h->slice_type == SP_TYPE ))
2968 || (h->pps.weighted_bipred_idc==1 && h->slice_type==B_TYPE ) )
2969 pred_weight_table(h);
2971 if(s->current_picture.reference)
2972 decode_ref_pic_marking(h);
2975 s->qscale = h->pps.init_qp + get_se_golomb(&s->gb); //slice_qp_delta
2976 //FIXME qscale / qp ... stuff
2977 if(h->slice_type == SP_TYPE){
2978 int sp_for_switch_flag= get_bits1(&s->gb);
2980 if(h->slice_type==SP_TYPE || h->slice_type == SI_TYPE){
2981 int slice_qs_delta= get_se_golomb(&s->gb);
2984 if( h->pps.deblocking_filter_parameters_present ) {
2985 h->disable_deblocking_filter_idc= get_ue_golomb(&s->gb);
2986 if( h->disable_deblocking_filter_idc != 1 ) {
2987 h->slice_alpha_c0_offset_div2= get_se_golomb(&s->gb);
2988 h->slice_beta_offset_div2= get_se_golomb(&s->gb);
2991 h->disable_deblocking_filter_idc= 0;
2994 if( h->pps.num_slice_groups > 1 && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
2995 slice_group_change_cycle= get_bits(&s->gb, ?);
2998 if(s->avctx->debug&FF_DEBUG_PICT_INFO){
2999 printf("mb:%d %c pps:%d frame:%d poc:%d/%d ref:%d/%d qp:%d loop:%d\n",
3001 ff_get_pict_type_char(h->slice_type),
3002 pps_id, h->frame_num,
3003 s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
3004 h->ref_count[0], h->ref_count[1],
3006 h->disable_deblocking_filter_idc
3016 static inline int get_level_prefix(GetBitContext *gb){
3020 OPEN_READER(re, gb);
3021 UPDATE_CACHE(re, gb);
3022 buf=GET_CACHE(re, gb);
3024 log= 32 - av_log2(buf);
3026 print_bin(buf>>(32-log), log);
3027 printf("%5d %2d %3d lpr @%5d in %s get_level_prefix\n", buf>>(32-log), log, log-1, get_bits_count(gb), __FILE__);
3030 LAST_SKIP_BITS(re, gb, log);
3031 CLOSE_READER(re, gb);
3037 * decodes a residual block.
3038 * @param n block index
3039 * @param scantable scantable
3040 * @param max_coeff number of coefficients in the block
3041 * @return <0 if an error occured
3043 static int decode_residual(H264Context *h, GetBitContext *gb, DCTELEM *block, int n, const uint8_t *scantable, int qp, int max_coeff){
3044 MpegEncContext * const s = &h->s;
3045 const uint16_t *qmul= dequant_coeff[qp];
3046 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};
3047 int level[16], run[16];
3048 int suffix_length, zeros_left, coeff_num, coeff_token, total_coeff, i, trailing_ones;
3050 //FIXME put trailing_onex into the context
3052 if(n == CHROMA_DC_BLOCK_INDEX){
3053 coeff_token= get_vlc2(gb, chroma_dc_coeff_token_vlc.table, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 1);
3054 total_coeff= coeff_token>>2;
3056 if(n == LUMA_DC_BLOCK_INDEX){
3057 total_coeff= pred_non_zero_count(h, 0);
3058 coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
3059 total_coeff= coeff_token>>2;
3061 total_coeff= pred_non_zero_count(h, n);
3062 coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
3063 total_coeff= coeff_token>>2;
3064 h->non_zero_count_cache[ scan8[n] ]= total_coeff;
3068 //FIXME set last_non_zero?
3073 trailing_ones= coeff_token&3;
3074 tprintf("trailing:%d, total:%d\n", trailing_ones, total_coeff);
3075 assert(total_coeff<=16);
3077 for(i=0; i<trailing_ones; i++){
3078 level[i]= 1 - 2*get_bits1(gb);
3081 suffix_length= total_coeff > 10 && trailing_ones < 3;
3083 for(; i<total_coeff; i++){
3084 const int prefix= get_level_prefix(gb);
3085 int level_code, mask;
3087 if(prefix<14){ //FIXME try to build a large unified VLC table for all this
3089 level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
3091 level_code= (prefix<<suffix_length); //part
3092 }else if(prefix==14){
3094 level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
3096 level_code= prefix + get_bits(gb, 4); //part
3097 }else if(prefix==15){
3098 level_code= (prefix<<suffix_length) + get_bits(gb, 12); //part
3099 if(suffix_length==0) level_code+=15; //FIXME doesnt make (much)sense
3101 fprintf(stderr, "prefix too large at %d %d\n", s->mb_x, s->mb_y);
3105 if(i==trailing_ones && i<3) level_code+= 2; //FIXME split first iteration
3107 mask= -(level_code&1);
3108 level[i]= (((2+level_code)>>1) ^ mask) - mask;
3110 if(suffix_length==0) suffix_length=1; //FIXME split first iteration
3113 if(ABS(level[i]) > (3<<(suffix_length-1)) && suffix_length<6) suffix_length++;
3115 if((2+level_code)>>1) > (3<<(suffix_length-1)) && suffix_length<6) suffix_length++;
3116 ? == prefix > 2 or sth
3118 tprintf("level: %d suffix_length:%d\n", level[i], suffix_length);
3121 if(total_coeff == max_coeff)
3124 if(n == CHROMA_DC_BLOCK_INDEX)
3125 zeros_left= get_vlc2(gb, chroma_dc_total_zeros_vlc[ total_coeff-1 ].table, CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 1);
3127 zeros_left= get_vlc2(gb, total_zeros_vlc[ total_coeff-1 ].table, TOTAL_ZEROS_VLC_BITS, 1);
3130 for(i=0; i<total_coeff-1; i++){
3133 else if(zeros_left < 7){
3134 run[i]= get_vlc2(gb, run_vlc[zeros_left-1].table, RUN_VLC_BITS, 1);
3136 run[i]= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2);
3138 zeros_left -= run[i];
3142 fprintf(stderr, "negative number of zero coeffs at %d %d\n", s->mb_x, s->mb_y);
3146 for(; i<total_coeff-1; i++){
3154 for(i=total_coeff-1; i>=0; i--){ //FIXME merge into rundecode?
3157 coeff_num += run[i] + 1; //FIXME add 1 earlier ?
3158 j= scantable[ coeff_num ];
3163 for(i=total_coeff-1; i>=0; i--){ //FIXME merge into rundecode?
3166 coeff_num += run[i] + 1; //FIXME add 1 earlier ?
3167 j= scantable[ coeff_num ];
3169 block[j]= level[i] * qmul[j];
3170 // printf("%d %d ", block[j], qmul[j]);
3177 * decodes a macroblock
3178 * @returns 0 if ok, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
3180 static int decode_mb(H264Context *h){
3181 MpegEncContext * const s = &h->s;
3182 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
3183 int mb_type, partition_count, cbp;
3185 memset(h->mb, 0, sizeof(int16_t)*24*16); //FIXME avoid if allready clear (move after skip handlong?
3187 tprintf("pic:%d mb:%d/%d\n", h->frame_num, s->mb_x, s->mb_y);
3189 if(h->slice_type != I_TYPE && h->slice_type != SI_TYPE){
3190 if(s->mb_skip_run==-1)
3191 s->mb_skip_run= get_ue_golomb(&s->gb);
3193 if (s->mb_skip_run--) {
3197 mb_type= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0;
3199 memset(h->non_zero_count[mb_xy], 0, 16);
3200 memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui
3202 if(h->sps.mb_aff && s->mb_skip_run==0 && (s->mb_y&1)==0){
3203 h->mb_field_decoding_flag= get_bits1(&s->gb);
3206 if(h->mb_field_decoding_flag)
3207 mb_type|= MB_TYPE_INTERLACED;
3209 fill_caches(h, mb_type); //FIXME check what is needed and what not ...
3210 pred_pskip_motion(h, &mx, &my);
3211 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
3212 fill_rectangle( h->mv_cache[0][scan8[0]], 4, 4, 8, (mx&0xFFFF)+(my<<16), 4);
3213 write_back_motion(h, mb_type);
3215 s->current_picture.mb_type[mb_xy]= mb_type; //FIXME SKIP type
3216 h->slice_table[ mb_xy ]= h->slice_num;
3218 h->prev_mb_skiped= 1;
3222 if(h->sps.mb_aff /* && !field pic FIXME needed? */){
3224 h->mb_field_decoding_flag = get_bits1(&s->gb);
3226 h->mb_field_decoding_flag=0; //FIXME som ed note ?!
3228 h->prev_mb_skiped= 0;
3230 mb_type= get_ue_golomb(&s->gb);
3231 if(h->slice_type == B_TYPE){
3233 partition_count= b_mb_type_info[mb_type].partition_count;
3234 mb_type= b_mb_type_info[mb_type].type;
3237 goto decode_intra_mb;
3239 }else if(h->slice_type == P_TYPE /*|| h->slice_type == SP_TYPE */){
3241 partition_count= p_mb_type_info[mb_type].partition_count;
3242 mb_type= p_mb_type_info[mb_type].type;
3245 goto decode_intra_mb;
3248 assert(h->slice_type == I_TYPE);
3251 fprintf(stderr, "mb_type %d in %c slice to large at %d %d\n", mb_type, ff_get_pict_type_char(h->slice_type), s->mb_x, s->mb_y);
3255 cbp= i_mb_type_info[mb_type].cbp;
3256 h->intra16x16_pred_mode= i_mb_type_info[mb_type].pred_mode;
3257 mb_type= i_mb_type_info[mb_type].type;
3260 if(h->mb_field_decoding_flag)
3261 mb_type |= MB_TYPE_INTERLACED;
3263 s->current_picture.mb_type[mb_xy]= mb_type;
3264 h->slice_table[ mb_xy ]= h->slice_num;
3266 if(IS_INTRA_PCM(mb_type)){
3270 // we assume these blocks are very rare so we dont optimize it
3271 align_get_bits(&s->gb);
3273 ptr= s->gb.buffer + get_bits_count(&s->gb);
3275 for(y=0; y<16; y++){
3276 const int index= 4*(y&3) + 64*(y>>2);
3277 for(x=0; x<16; x++){
3278 h->mb[index + (x&3) + 16*(x>>2)]= *(ptr++);
3282 const int index= 256 + 4*(y&3) + 32*(y>>2);
3284 h->mb[index + (x&3) + 16*(x>>2)]= *(ptr++);
3288 const int index= 256 + 64 + 4*(y&3) + 32*(y>>2);
3290 h->mb[index + (x&3) + 16*(x>>2)]= *(ptr++);
3294 skip_bits(&s->gb, 384); //FIXME check /fix the bitstream readers
3296 memset(h->non_zero_count[mb_xy], 16, 16);
3301 fill_caches(h, mb_type);
3304 if(IS_INTRA(mb_type)){
3305 // init_top_left_availability(h);
3306 if(IS_INTRA4x4(mb_type)){
3309 // fill_intra4x4_pred_table(h);
3310 for(i=0; i<16; i++){
3311 const int mode_coded= !get_bits1(&s->gb);
3312 const int predicted_mode= pred_intra_mode(h, i);
3316 const int rem_mode= get_bits(&s->gb, 3);
3317 if(rem_mode<predicted_mode)
3322 mode= predicted_mode;
3325 h->intra4x4_pred_mode_cache[ scan8[i] ] = mode;
3327 write_back_intra_pred_mode(h);
3328 if( check_intra4x4_pred_mode(h) < 0)
3331 h->intra16x16_pred_mode= check_intra_pred_mode(h, h->intra16x16_pred_mode);
3332 if(h->intra16x16_pred_mode < 0)
3335 h->chroma_pred_mode= get_ue_golomb(&s->gb);
3337 h->chroma_pred_mode= check_intra_pred_mode(h, h->chroma_pred_mode);
3338 if(h->chroma_pred_mode < 0)
3340 }else if(partition_count==4){
3341 int i, j, sub_partition_count[4], list, ref[2][4];
3343 if(h->slice_type == B_TYPE){
3345 h->sub_mb_type[i]= get_ue_golomb(&s->gb);
3346 if(h->sub_mb_type[i] >=13){
3347 fprintf(stderr, "B sub_mb_type %d out of range at %d %d\n", h->sub_mb_type[i], s->mb_x, s->mb_y);
3350 sub_partition_count[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
3351 h->sub_mb_type[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].type;
3354 assert(h->slice_type == P_TYPE || h->slice_type == SP_TYPE); //FIXME SP correct ?
3356 h->sub_mb_type[i]= get_ue_golomb(&s->gb);
3357 if(h->sub_mb_type[i] >=4){
3358 fprintf(stderr, "P sub_mb_type %d out of range at %d %d\n", h->sub_mb_type[i], s->mb_x, s->mb_y);
3361 sub_partition_count[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
3362 h->sub_mb_type[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].type;
3366 for(list=0; list<2; list++){
3367 const int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list];
3368 if(ref_count == 0) continue;
3370 if(IS_DIR(h->sub_mb_type[i], 0, list) && !IS_DIRECT(h->sub_mb_type[i])){
3371 ref[list][i] = get_te0_golomb(&s->gb, ref_count); //FIXME init to 0 before and skip?
3379 for(list=0; list<2; list++){
3380 const int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list];
3381 if(ref_count == 0) continue;
3384 h->ref_cache[list][ scan8[4*i] ]=h->ref_cache[list][ scan8[4*i]+1 ]=
3385 h->ref_cache[list][ scan8[4*i]+8 ]=h->ref_cache[list][ scan8[4*i]+9 ]= ref[list][i];
3387 if(IS_DIR(h->sub_mb_type[i], 0, list) && !IS_DIRECT(h->sub_mb_type[i])){
3388 const int sub_mb_type= h->sub_mb_type[i];
3389 const int block_width= (sub_mb_type & (MB_TYPE_16x16|MB_TYPE_16x8)) ? 2 : 1;
3390 for(j=0; j<sub_partition_count[i]; j++){
3392 const int index= 4*i + block_width*j;
3393 int16_t (* mv_cache)[2]= &h->mv_cache[list][ scan8[index] ];
3394 pred_motion(h, index, block_width, list, h->ref_cache[list][ scan8[index] ], &mx, &my);
3395 mx += get_se_golomb(&s->gb);
3396 my += get_se_golomb(&s->gb);
3397 tprintf("final mv:%d %d\n", mx, my);
3399 if(IS_SUB_8X8(sub_mb_type)){
3400 mv_cache[ 0 ][0]= mv_cache[ 1 ][0]=
3401 mv_cache[ 8 ][0]= mv_cache[ 9 ][0]= mx;
3402 mv_cache[ 0 ][1]= mv_cache[ 1 ][1]=
3403 mv_cache[ 8 ][1]= mv_cache[ 9 ][1]= my;
3404 }else if(IS_SUB_8X4(sub_mb_type)){
3405 mv_cache[ 0 ][0]= mv_cache[ 1 ][0]= mx;
3406 mv_cache[ 0 ][1]= mv_cache[ 1 ][1]= my;
3407 }else if(IS_SUB_4X8(sub_mb_type)){
3408 mv_cache[ 0 ][0]= mv_cache[ 8 ][0]= mx;
3409 mv_cache[ 0 ][1]= mv_cache[ 8 ][1]= my;
3411 assert(IS_SUB_4X4(sub_mb_type));
3412 mv_cache[ 0 ][0]= mx;
3413 mv_cache[ 0 ][1]= my;
3417 uint32_t *p= (uint32_t *)&h->mv_cache[list][ scan8[4*i] ][0];
3423 }else if(!IS_DIRECT(mb_type)){
3424 int list, mx, my, i;
3425 //FIXME we should set ref_idx_l? to 0 if we use that later ...
3426 if(IS_16X16(mb_type)){
3427 for(list=0; list<2; list++){
3428 if(h->ref_count[0]>0){
3429 if(IS_DIR(mb_type, 0, list)){
3430 const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
3431 fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, val, 1);
3435 for(list=0; list<2; list++){
3436 if(IS_DIR(mb_type, 0, list)){
3437 pred_motion(h, 0, 4, list, h->ref_cache[list][ scan8[0] ], &mx, &my);
3438 mx += get_se_golomb(&s->gb);
3439 my += get_se_golomb(&s->gb);
3440 tprintf("final mv:%d %d\n", mx, my);
3442 fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, (mx&0xFFFF) + (my<<16), 4);
3446 else if(IS_16X8(mb_type)){
3447 for(list=0; list<2; list++){
3448 if(h->ref_count[list]>0){
3450 if(IS_DIR(mb_type, i, list)){
3451 const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
3452 fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, val, 1);
3457 for(list=0; list<2; list++){
3459 if(IS_DIR(mb_type, i, list)){
3460 pred_16x8_motion(h, 8*i, list, h->ref_cache[list][scan8[0] + 16*i], &mx, &my);
3461 mx += get_se_golomb(&s->gb);
3462 my += get_se_golomb(&s->gb);
3463 tprintf("final mv:%d %d\n", mx, my);
3465 fill_rectangle(h->mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, (mx&0xFFFF) + (my<<16), 4);
3470 assert(IS_8X16(mb_type));
3471 for(list=0; list<2; list++){
3472 if(h->ref_count[list]>0){
3474 if(IS_DIR(mb_type, i, list)){ //FIXME optimize
3475 const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
3476 fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, val, 1);
3481 for(list=0; list<2; list++){
3483 if(IS_DIR(mb_type, i, list)){
3484 pred_8x16_motion(h, i*4, list, h->ref_cache[list][ scan8[0] + 2*i ], &mx, &my);
3485 mx += get_se_golomb(&s->gb);
3486 my += get_se_golomb(&s->gb);
3487 tprintf("final mv:%d %d\n", mx, my);
3489 fill_rectangle(h->mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, (mx&0xFFFF) + (my<<16), 4);
3496 if(IS_INTER(mb_type))
3497 write_back_motion(h, mb_type);
3499 if(!IS_INTRA16x16(mb_type)){
3500 cbp= get_ue_golomb(&s->gb);
3502 fprintf(stderr, "cbp too large (%d) at %d %d\n", cbp, s->mb_x, s->mb_y);
3506 if(IS_INTRA4x4(mb_type))
3507 cbp= golomb_to_intra4x4_cbp[cbp];
3509 cbp= golomb_to_inter_cbp[cbp];
3512 if(cbp || IS_INTRA16x16(mb_type)){
3513 int i8x8, i4x4, chroma_idx;
3514 int chroma_qp, dquant;
3515 GetBitContext *gb= IS_INTRA(mb_type) ? h->intra_gb_ptr : h->inter_gb_ptr;
3516 const uint8_t *scan, *dc_scan;
3518 // fill_non_zero_count_cache(h);
3520 if(IS_INTERLACED(mb_type)){
3522 dc_scan= luma_dc_field_scan;
3525 dc_scan= luma_dc_zigzag_scan;
3528 dquant= get_se_golomb(&s->gb);
3530 if( dquant > 25 || dquant < -26 ){
3531 fprintf(stderr, "dquant out of range (%d) at %d %d\n", dquant, s->mb_x, s->mb_y);
3535 s->qscale += dquant;
3536 if(((unsigned)s->qscale) > 51){
3537 if(s->qscale<0) s->qscale+= 52;
3538 else s->qscale-= 52;
3541 h->chroma_qp= chroma_qp= get_chroma_qp(h, s->qscale);
3542 if(IS_INTRA16x16(mb_type)){
3543 if( decode_residual(h, h->intra_gb_ptr, h->mb, LUMA_DC_BLOCK_INDEX, dc_scan, s->qscale, 16) < 0){
3544 return -1; //FIXME continue if partotioned and other retirn -1 too
3547 assert((cbp&15) == 0 || (cbp&15) == 15);
3550 for(i8x8=0; i8x8<4; i8x8++){
3551 for(i4x4=0; i4x4<4; i4x4++){
3552 const int index= i4x4 + 4*i8x8;
3553 if( decode_residual(h, h->intra_gb_ptr, h->mb + 16*index, index, scan + 1, s->qscale, 15) < 0 ){
3559 memset(&h->non_zero_count_cache[8], 0, 8*4); //FIXME stupid & slow
3562 for(i8x8=0; i8x8<4; i8x8++){
3563 if(cbp & (1<<i8x8)){
3564 for(i4x4=0; i4x4<4; i4x4++){
3565 const int index= i4x4 + 4*i8x8;
3567 if( decode_residual(h, gb, h->mb + 16*index, index, scan, s->qscale, 16) <0 ){
3572 uint8_t * const nnz= &h->non_zero_count_cache[ scan8[4*i8x8] ];
3573 nnz[0] = nnz[1] = nnz[8] = nnz[9] = 0;
3579 for(chroma_idx=0; chroma_idx<2; chroma_idx++)
3580 if( decode_residual(h, gb, h->mb + 256 + 16*4*chroma_idx, CHROMA_DC_BLOCK_INDEX, chroma_dc_scan, chroma_qp, 4) < 0){
3586 for(chroma_idx=0; chroma_idx<2; chroma_idx++){
3587 for(i4x4=0; i4x4<4; i4x4++){
3588 const int index= 16 + 4*chroma_idx + i4x4;
3589 if( decode_residual(h, gb, h->mb + 16*index, index, scan + 1, chroma_qp, 15) < 0){
3595 uint8_t * const nnz= &h->non_zero_count_cache[0];
3596 nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =
3597 nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
3600 memset(&h->non_zero_count_cache[8], 0, 8*5);
3602 write_back_non_zero_count(h);
3607 static int decode_slice(H264Context *h){
3608 MpegEncContext * const s = &h->s;
3609 const int part_mask= s->partitioned_frame ? (AC_END|AC_ERROR) : 0x7F;
3615 int ret= decode_mb(h);
3619 if(ret>=0 && h->sps.mb_aff){ //FIXME optimal? or let mb_decode decode 16x32 ?
3628 fprintf(stderr, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
3629 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask);
3634 if(++s->mb_x >= s->mb_width){
3636 ff_draw_horiz_band(s, 16*s->mb_y, 16);
3637 if(++s->mb_y >= s->mb_height){
3638 tprintf("slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
3640 if(get_bits_count(&s->gb) == s->gb.size_in_bits){
3641 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
3645 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
3652 if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->mb_skip_run<=0){
3653 if(get_bits_count(&s->gb) == s->gb.size_in_bits){
3654 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
3658 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask);
3666 for(;s->mb_y < s->mb_height; s->mb_y++){
3667 for(;s->mb_x < s->mb_width; s->mb_x++){
3668 int ret= decode_mb(h);
3673 fprintf(stderr, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
3674 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask);
3679 if(++s->mb_x >= s->mb_width){
3681 if(++s->mb_y >= s->mb_height){
3682 if(get_bits_count(s->gb) == s->gb.size_in_bits){
3683 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
3687 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
3694 if(get_bits_count(s->?gb) >= s->gb?.size_in_bits){
3695 if(get_bits_count(s->gb) == s->gb.size_in_bits){
3696 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
3700 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask);
3707 ff_draw_horiz_band(s, 16*s->mb_y, 16);
3710 return -1; //not reached
3713 static inline int decode_vui_parameters(H264Context *h, SPS *sps){
3714 MpegEncContext * const s = &h->s;
3715 int aspect_ratio_info_present_flag, aspect_ratio_idc;
3717 aspect_ratio_info_present_flag= get_bits1(&s->gb);
3719 if( aspect_ratio_info_present_flag ) {
3720 aspect_ratio_idc= get_bits(&s->gb, 8);
3721 if( aspect_ratio_idc == EXTENDED_SAR ) {
3722 sps->sar_width= get_bits(&s->gb, 16);
3723 sps->sar_height= get_bits(&s->gb, 16);
3724 }else if(aspect_ratio_idc < 16){
3725 sps->sar_width= pixel_aspect[aspect_ratio_idc][0];
3726 sps->sar_height= pixel_aspect[aspect_ratio_idc][1];
3728 fprintf(stderr, "illegal aspect ratio\n");
3735 // s->avctx->aspect_ratio= sar_width*s->width / (float)(s->height*sar_height);
3737 | overscan_info_present_flag |0 |u(1) |
3738 | if( overscan_info_present_flag ) | | |
3739 | overscan_appropriate_flag |0 |u(1) |
3740 | video_signal_type_present_flag |0 |u(1) |
3741 | if( video_signal_type_present_flag ) { | | |
3742 | video_format |0 |u(3) |
3743 | video_full_range_flag |0 |u(1) |
3744 | colour_description_present_flag |0 |u(1) |
3745 | if( colour_description_present_flag ) { | | |
3746 | colour_primaries |0 |u(8) |
3747 | transfer_characteristics |0 |u(8) |
3748 | matrix_coefficients |0 |u(8) |
3751 | chroma_location_info_present_flag |0 |u(1) |
3752 | if ( chroma_location_info_present_flag ) { | | |
3753 | chroma_sample_location_type_top_field |0 |ue(v) |
3754 | chroma_sample_location_type_bottom_field |0 |ue(v) |
3756 | timing_info_present_flag |0 |u(1) |
3757 | if( timing_info_present_flag ) { | | |
3758 | num_units_in_tick |0 |u(32) |
3759 | time_scale |0 |u(32) |
3760 | fixed_frame_rate_flag |0 |u(1) |
3762 | nal_hrd_parameters_present_flag |0 |u(1) |
3763 | if( nal_hrd_parameters_present_flag = = 1) | | |
3764 | hrd_parameters( ) | | |
3765 | vcl_hrd_parameters_present_flag |0 |u(1) |
3766 | if( vcl_hrd_parameters_present_flag = = 1) | | |
3767 | hrd_parameters( ) | | |
3768 | if( ( nal_hrd_parameters_present_flag = = 1 | || | |
3770 |( vcl_hrd_parameters_present_flag = = 1 ) ) | | |
3771 | low_delay_hrd_flag |0 |u(1) |
3772 | bitstream_restriction_flag |0 |u(1) |
3773 | if( bitstream_restriction_flag ) { |0 |u(1) |
3774 | motion_vectors_over_pic_boundaries_flag |0 |u(1) |
3775 | max_bytes_per_pic_denom |0 |ue(v) |
3776 | max_bits_per_mb_denom |0 |ue(v) |
3777 | log2_max_mv_length_horizontal |0 |ue(v) |
3778 | log2_max_mv_length_vertical |0 |ue(v) |
3779 | num_reorder_frames |0 |ue(v) |
3780 | max_dec_frame_buffering |0 |ue(v) |
3787 static inline int decode_seq_parameter_set(H264Context *h){
3788 MpegEncContext * const s = &h->s;
3789 int profile_idc, level_idc, multiple_slice_groups, arbitrary_slice_order, redundant_slices;
3793 profile_idc= get_bits(&s->gb, 8);
3794 level_idc= get_bits(&s->gb, 8);
3795 multiple_slice_groups= get_bits1(&s->gb);
3796 arbitrary_slice_order= get_bits1(&s->gb);
3797 redundant_slices= get_bits1(&s->gb);
3799 sps_id= get_ue_golomb(&s->gb);
3801 sps= &h->sps_buffer[ sps_id ];
3803 sps->profile_idc= profile_idc;
3804 sps->level_idc= level_idc;
3805 sps->multiple_slice_groups= multiple_slice_groups;
3806 sps->arbitrary_slice_order= arbitrary_slice_order;
3807 sps->redundant_slices= redundant_slices;
3809 sps->log2_max_frame_num= get_ue_golomb(&s->gb) + 4;
3811 sps->poc_type= get_ue_golomb(&s->gb);
3813 if(sps->poc_type == 0){ //FIXME #define
3814 sps->log2_max_poc_lsb= get_ue_golomb(&s->gb) + 4;
3815 } else if(sps->poc_type == 1){//FIXME #define
3816 sps->delta_pic_order_always_zero_flag= get_bits1(&s->gb);
3817 sps->offset_for_non_ref_pic= get_se_golomb(&s->gb);
3818 sps->offset_for_top_to_bottom_field= get_se_golomb(&s->gb);
3819 sps->poc_cycle_length= get_ue_golomb(&s->gb);
3821 for(i=0; i<sps->poc_cycle_length; i++)
3822 sps->offset_for_ref_frame[i]= get_se_golomb(&s->gb);
3824 if(sps->poc_type > 2){
3825 fprintf(stderr, "illegal POC type %d\n", sps->poc_type);
3829 sps->ref_frame_count= get_ue_golomb(&s->gb);
3830 sps->required_frame_num_update_behaviour_flag= get_bits1(&s->gb);
3831 sps->mb_width= get_ue_golomb(&s->gb) + 1;
3832 sps->mb_height= get_ue_golomb(&s->gb) + 1;
3833 sps->frame_mbs_only_flag= get_bits1(&s->gb);
3834 if(!sps->frame_mbs_only_flag)
3835 sps->mb_aff= get_bits1(&s->gb);
3839 sps->direct_8x8_inference_flag= get_bits1(&s->gb);
3841 sps->vui_parameters_present_flag= get_bits1(&s->gb);
3842 if( sps->vui_parameters_present_flag )
3843 decode_vui_parameters(h, sps);
3845 if(s->avctx->debug&FF_DEBUG_PICT_INFO){
3846 printf("sps:%d profile:%d/%d poc:%d ref:%d %dx%d %s %s %s\n",
3847 sps_id, sps->profile_idc, sps->level_idc,
3849 sps->ref_frame_count,
3850 sps->mb_width, sps->mb_height,
3851 sps->frame_mbs_only_flag ? "FRM" : (sps->mb_aff ? "MB-AFF" : "PIC-AFF"),
3852 sps->direct_8x8_inference_flag ? "8B8" : "",
3853 sps->vui_parameters_present_flag ? "VUI" : ""
3859 static inline int decode_picture_parameter_set(H264Context *h){
3860 MpegEncContext * const s = &h->s;
3861 int pps_id= get_ue_golomb(&s->gb);
3862 PPS *pps= &h->pps_buffer[pps_id];
3864 pps->sps_id= get_ue_golomb(&s->gb);
3865 pps->cabac= get_bits1(&s->gb);
3866 pps->pic_order_present= get_bits1(&s->gb);
3867 pps->slice_group_count= get_ue_golomb(&s->gb) + 1;
3868 if(pps->slice_group_count > 1 ){
3869 pps->mb_slice_group_map_type= get_ue_golomb(&s->gb);
3870 fprintf(stderr, "FMO not supported\n");
3871 switch(pps->mb_slice_group_map_type){
3874 | for( i = 0; i <= num_slice_groups_minus1; i++ ) | | |
3875 | run_length[ i ] |1 |ue(v) |
3880 | for( i = 0; i < num_slice_groups_minus1; i++ ) | | |
3882 | top_left_mb[ i ] |1 |ue(v) |
3883 | bottom_right_mb[ i ] |1 |ue(v) |
3891 | slice_group_change_direction_flag |1 |u(1) |
3892 | slice_group_change_rate_minus1 |1 |ue(v) |
3897 | slice_group_id_cnt_minus1 |1 |ue(v) |
3898 | for( i = 0; i <= slice_group_id_cnt_minus1; i++ | | |
3900 | slice_group_id[ i ] |1 |u(v) |
3905 pps->ref_count[0]= get_ue_golomb(&s->gb) + 1;
3906 pps->ref_count[1]= get_ue_golomb(&s->gb) + 1;
3907 if(pps->ref_count[0] > 32 || pps->ref_count[1] > 32){
3908 fprintf(stderr, "reference overflow (pps)\n");
3912 pps->weighted_pred= get_bits1(&s->gb);
3913 pps->weighted_bipred_idc= get_bits(&s->gb, 2);
3914 pps->init_qp= get_se_golomb(&s->gb) + 26;
3915 pps->init_qs= get_se_golomb(&s->gb) + 26;
3916 pps->chroma_qp_index_offset= get_se_golomb(&s->gb);
3917 pps->deblocking_filter_parameters_present= get_bits1(&s->gb);
3918 pps->constrained_intra_pred= get_bits1(&s->gb);
3919 pps->redundant_pic_cnt_present = get_bits1(&s->gb);
3920 pps->crop= get_bits1(&s->gb);
3922 pps->crop_left = get_ue_golomb(&s->gb);
3923 pps->crop_right = get_ue_golomb(&s->gb);
3924 pps->crop_top = get_ue_golomb(&s->gb);
3925 pps->crop_bottom= get_ue_golomb(&s->gb);
3930 pps->crop_bottom= 0;
3933 if(s->avctx->debug&FF_DEBUG_PICT_INFO){
3934 printf("pps:%d sps:%d %s slice_groups:%d ref:%d/%d %s qp:%d/%d/%d %s %s %s crop:%d/%d/%d/%d\n",
3935 pps_id, pps->sps_id,
3936 pps->cabac ? "CABAC" : "CAVLC",
3937 pps->slice_group_count,
3938 pps->ref_count[0], pps->ref_count[1],
3939 pps->weighted_pred ? "weighted" : "",
3940 pps->init_qp, pps->init_qs, pps->chroma_qp_index_offset,
3941 pps->deblocking_filter_parameters_present ? "LPAR" : "",
3942 pps->constrained_intra_pred ? "CONSTR" : "",
3943 pps->redundant_pic_cnt_present ? "REDU" : "",
3944 pps->crop_left, pps->crop_right,
3945 pps->crop_top, pps->crop_bottom
3953 * finds the end of the current frame in the bitstream.
3954 * @return the position of the first byte of the next frame, or -1
3956 static int find_frame_end(MpegEncContext *s, uint8_t *buf, int buf_size){
3957 ParseContext *pc= &s->parse_context;
3960 //printf("first %02X%02X%02X%02X\n", buf[0], buf[1],buf[2],buf[3]);
3961 // mb_addr= pc->mb_addr - 1;
3963 //FIXME this will fail with slices
3964 for(i=0; i<buf_size; i++){
3965 state= (state<<8) | buf[i];
3966 if((state&0xFFFFFF1F) == 0x101 || (state&0xFFFFFF1F) == 0x102 || (state&0xFFFFFF1F) == 0x105){
3967 if(pc->frame_start_found){
3969 pc->frame_start_found= 0;
3972 pc->frame_start_found= 1;
3977 return END_NOT_FOUND;
3980 static int decode_nal_units(H264Context *h, uint8_t *buf, int buf_size){
3981 MpegEncContext * const s = &h->s;
3982 AVCodecContext * const avctx= s->avctx;
3986 for(i=0; i<32; i++){
3987 printf("%X ", buf[i]);
3996 // start code prefix search
3997 for(; buf_index + 3 < buf_size; buf_index++){
3998 // this should allways succeed in the first iteration
3999 if(buf[buf_index] == 0 && buf[buf_index+1] == 0 && buf[buf_index+2] == 1)
4003 if(buf_index+3 >= buf_size) break;
4007 ptr= decode_nal(h, buf + buf_index, &dst_length, &consumed, buf_size - buf_index);
4008 if(ptr[dst_length - 1] == 0) dst_length--;
4009 bit_length= 8*dst_length - decode_rbsp_trailing(ptr + dst_length - 1);
4011 if(s->avctx->debug&FF_DEBUG_STARTCODE){
4012 printf("NAL %d at %d length %d\n", h->nal_unit_type, buf_index, dst_length);
4015 buf_index += consumed;
4017 if(h->nal_ref_idc < s->hurry_up)
4020 switch(h->nal_unit_type){
4022 idr(h); //FIXME ensure we dont loose some frames if there is reordering
4024 init_get_bits(&s->gb, ptr, bit_length);
4026 h->inter_gb_ptr= &s->gb;
4027 s->data_partitioning = 0;
4029 if(decode_slice_header(h) < 0) return -1;
4030 if(h->redundant_pic_count==0)
4034 init_get_bits(&s->gb, ptr, bit_length);
4036 h->inter_gb_ptr= NULL;
4037 s->data_partitioning = 1;
4039 if(decode_slice_header(h) < 0) return -1;
4042 init_get_bits(&h->intra_gb, ptr, bit_length);
4043 h->intra_gb_ptr= &h->intra_gb;
4046 init_get_bits(&h->inter_gb, ptr, bit_length);
4047 h->inter_gb_ptr= &h->inter_gb;
4049 if(h->redundant_pic_count==0 && h->intra_gb_ptr && s->data_partitioning)
4055 init_get_bits(&s->gb, ptr, bit_length);
4056 decode_seq_parameter_set(h);
4058 if(s->flags& CODEC_FLAG_LOW_DELAY)
4061 avctx->has_b_frames= !s->low_delay;
4064 init_get_bits(&s->gb, ptr, bit_length);
4066 decode_picture_parameter_set(h);
4069 case NAL_PICTURE_DELIMITER:
4071 case NAL_FILTER_DATA:
4075 //FIXME move after where irt is set
4076 s->current_picture.pict_type= s->pict_type;
4077 s->current_picture.key_frame= s->pict_type == I_TYPE;
4080 if(!s->current_picture_ptr) return buf_index; //no frame
4082 h->prev_frame_num_offset= h->frame_num_offset;
4083 h->prev_frame_num= h->frame_num;
4084 if(s->current_picture_ptr->reference){
4085 h->prev_poc_msb= h->poc_msb;
4086 h->prev_poc_lsb= h->poc_lsb;
4088 if(s->current_picture_ptr->reference)
4089 execute_ref_pic_marking(h, h->mmco, h->mmco_index);
4091 assert(h->mmco_index==0);
4100 * retunrs the number of bytes consumed for building the current frame
4102 static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size){
4103 if(s->flags&CODEC_FLAG_TRUNCATED){
4104 pos -= s->parse_context.last_index;
4105 if(pos<0) pos=0; // FIXME remove (uneeded?)
4109 if(pos==0) pos=1; //avoid infinite loops (i doubt thats needed but ...)
4110 if(pos+10>buf_size) pos=buf_size; // oops ;)
4116 static int decode_frame(AVCodecContext *avctx,
4117 void *data, int *data_size,
4118 uint8_t *buf, int buf_size)
4120 H264Context *h = avctx->priv_data;
4121 MpegEncContext *s = &h->s;
4122 AVFrame *pict = data;
4125 s->flags= avctx->flags;
4129 /* no supplementary picture */
4130 if (buf_size == 0) {
4134 if(s->flags&CODEC_FLAG_TRUNCATED){
4135 int next= find_frame_end(s, buf, buf_size);
4137 if( ff_combine_frame(s, next, &buf, &buf_size) < 0 )
4139 //printf("next:%d buf_size:%d last_index:%d\n", next, buf_size, s->parse_context.last_index);
4142 if(s->avctx->extradata_size && s->picture_number==0){
4143 if(0 < decode_nal_units(h, s->avctx->extradata, s->avctx->extradata_size) )
4147 buf_index=decode_nal_units(h, buf, buf_size);
4151 //FIXME do something with unavailable reference frames
4153 // if(ret==FRAME_SKIPED) return get_consumed_bytes(s, buf_index, buf_size);
4155 if(s->pict_type==B_TYPE || s->low_delay){
4156 *pict= *(AVFrame*)&s->current_picture;
4158 *pict= *(AVFrame*)&s->last_picture;
4161 if(!s->current_picture_ptr){
4162 fprintf(stderr, "error, NO frame\n");
4166 *pict= *(AVFrame*)&s->current_picture; //FIXME
4167 ff_print_debug_info(s, s->current_picture_ptr);
4168 assert(pict->data[0]);
4169 //printf("out %d\n", (int)pict->data[0]);
4172 /* Return the Picture timestamp as the frame number */
4173 /* we substract 1 because it is added on utils.c */
4174 avctx->frame_number = s->picture_number - 1;
4177 /* dont output the last pic after seeking */
4178 if(s->last_picture_ptr || s->low_delay)
4179 //Note this isnt a issue as a IDR pic should flush teh buffers
4181 *data_size = sizeof(AVFrame);
4182 return get_consumed_bytes(s, buf_index, buf_size);
4185 static inline void fill_mb_avail(H264Context *h){
4186 MpegEncContext * const s = &h->s;
4187 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
4190 h->mb_avail[0]= s->mb_x && h->slice_table[mb_xy - s->mb_stride - 1] == h->slice_num;
4191 h->mb_avail[1]= h->slice_table[mb_xy - s->mb_stride ] == h->slice_num;
4192 h->mb_avail[2]= s->mb_x+1 < s->mb_width && h->slice_table[mb_xy - s->mb_stride + 1] == h->slice_num;
4198 h->mb_avail[3]= s->mb_x && h->slice_table[mb_xy - 1] == h->slice_num;
4199 h->mb_avail[4]= 1; //FIXME move out
4200 h->mb_avail[5]= 0; //FIXME move out
4206 #define SIZE (COUNT*40)
4212 // int int_temp[10000];
4214 AVCodecContext avctx;
4216 dsputil_init(&dsp, &avctx);
4218 init_put_bits(&pb, temp, SIZE, NULL, NULL);
4219 printf("testing unsigned exp golomb\n");
4220 for(i=0; i<COUNT; i++){
4222 set_ue_golomb(&pb, i);
4223 STOP_TIMER("set_ue_golomb");
4225 flush_put_bits(&pb);
4227 init_get_bits(&gb, temp, 8*SIZE);
4228 for(i=0; i<COUNT; i++){
4231 s= show_bits(&gb, 24);
4234 j= get_ue_golomb(&gb);
4236 printf("missmatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
4239 STOP_TIMER("get_ue_golomb");
4243 init_put_bits(&pb, temp, SIZE, NULL, NULL);
4244 printf("testing signed exp golomb\n");
4245 for(i=0; i<COUNT; i++){
4247 set_se_golomb(&pb, i - COUNT/2);
4248 STOP_TIMER("set_se_golomb");
4250 flush_put_bits(&pb);
4252 init_get_bits(&gb, temp, 8*SIZE);
4253 for(i=0; i<COUNT; i++){
4256 s= show_bits(&gb, 24);
4259 j= get_se_golomb(&gb);
4260 if(j != i - COUNT/2){
4261 printf("missmatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
4264 STOP_TIMER("get_se_golomb");
4267 printf("testing 4x4 (I)DCT\n");
4270 uint8_t src[16], ref[16];
4271 uint64_t error= 0, max_error=0;
4273 for(i=0; i<COUNT; i++){
4275 // printf("%d %d %d\n", r1, r2, (r2-r1)*16);
4276 for(j=0; j<16; j++){
4277 ref[j]= random()%255;
4278 src[j]= random()%255;
4281 h264_diff_dct_c(block, src, ref, 4);
4284 for(j=0; j<16; j++){
4285 // printf("%d ", block[j]);
4286 block[j]= block[j]*4;
4287 if(j&1) block[j]= (block[j]*4 + 2)/5;
4288 if(j&4) block[j]= (block[j]*4 + 2)/5;
4292 h264_add_idct_c(ref, block, 4);
4293 /* for(j=0; j<16; j++){
4294 printf("%d ", ref[j]);
4298 for(j=0; j<16; j++){
4299 int diff= ABS(src[j] - ref[j]);
4302 max_error= FFMAX(max_error, diff);
4305 printf("error=%f max_error=%d\n", ((float)error)/COUNT/16, (int)max_error );
4307 printf("testing quantizer\n");
4308 for(qp=0; qp<52; qp++){
4310 src1_block[i]= src2_block[i]= random()%255;
4314 printf("Testing NAL layer\n");
4316 uint8_t bitstream[COUNT];
4317 uint8_t nal[COUNT*2];
4319 memset(&h, 0, sizeof(H264Context));
4321 for(i=0; i<COUNT; i++){
4329 for(j=0; j<COUNT; j++){
4330 bitstream[j]= (random() % 255) + 1;
4333 for(j=0; j<zeros; j++){
4334 int pos= random() % COUNT;
4335 while(bitstream[pos] == 0){
4344 nal_length= encode_nal(&h, nal, bitstream, COUNT, COUNT*2);
4346 printf("encoding failed\n");
4350 out= decode_nal(&h, nal, &out_length, &consumed, nal_length);
4354 if(out_length != COUNT){
4355 printf("incorrect length %d %d\n", out_length, COUNT);
4359 if(consumed != nal_length){
4360 printf("incorrect consumed length %d %d\n", nal_length, consumed);
4364 if(memcmp(bitstream, out, COUNT)){
4365 printf("missmatch\n");
4370 printf("Testing RBSP\n");
4378 static int decode_end(AVCodecContext *avctx)
4380 H264Context *h = avctx->priv_data;
4381 MpegEncContext *s = &h->s;
4383 free_tables(h); //FIXME cleanup init stuff perhaps
4386 // memset(h, 0, sizeof(H264Context));
4392 AVCodec h264_decoder = {
4396 sizeof(H264Context),
4401 /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 | CODEC_CAP_TRUNCATED,