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 log2_max_frame_num; ///< log2_max_frame_num_minus4 + 4
63 int poc_type; ///< pic_order_cnt_type
64 int log2_max_poc_lsb; ///< log2_max_pic_order_cnt_lsb_minus4
65 int delta_pic_order_always_zero_flag;
66 int offset_for_non_ref_pic;
67 int offset_for_top_to_bottom_field;
68 int poc_cycle_length; ///< num_ref_frames_in_pic_order_cnt_cycle
69 int ref_frame_count; ///< num_ref_frames
70 int gaps_in_frame_num_allowed_flag;
71 int mb_width; ///< frame_width_in_mbs_minus1 + 1
72 int mb_height; ///< frame_height_in_mbs_minus1 + 1
73 int frame_mbs_only_flag;
74 int mb_aff; ///<mb_adaptive_frame_field_flag
75 int direct_8x8_inference_flag;
76 int crop; ///< frame_cropping_flag
77 int crop_left; ///< frame_cropping_rect_left_offset
78 int crop_right; ///< frame_cropping_rect_right_offset
79 int crop_top; ///< frame_cropping_rect_top_offset
80 int crop_bottom; ///< frame_cropping_rect_bottom_offset
81 int vui_parameters_present_flag;
83 short offset_for_ref_frame[256]; //FIXME dyn aloc?
87 * Picture parameter set
91 int cabac; ///< entropy_coding_mode_flag
92 int pic_order_present; ///< pic_order_present_flag
93 int slice_group_count; ///< num_slice_groups_minus1 + 1
94 int mb_slice_group_map_type;
95 int ref_count[2]; ///< num_ref_idx_l0/1_active_minus1 + 1
96 int weighted_pred; ///< weighted_pred_flag
97 int weighted_bipred_idc;
98 int init_qp; ///< pic_init_qp_minus26 + 26
99 int init_qs; ///< pic_init_qs_minus26 + 26
100 int chroma_qp_index_offset;
101 int deblocking_filter_parameters_present; ///< deblocking_filter_parameters_present_flag
102 int constrained_intra_pred; ///< constrained_intra_pred_flag
103 int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag
107 * Memory management control operation opcode.
109 typedef enum MMCOOpcode{
120 * Memory management control operation.
131 typedef struct H264Context{
139 #define NAL_IDR_SLICE 5
143 #define NAL_PICTURE_DELIMITER 9
144 #define NAL_FILTER_DATA 10
145 uint8_t *rbsp_buffer;
146 int rbsp_buffer_size;
150 int prev_mb_skiped; //FIXME remove (IMHO not used)
153 int chroma_pred_mode;
154 int intra16x16_pred_mode;
156 int8_t intra4x4_pred_mode_cache[5*8];
157 int8_t (*intra4x4_pred_mode)[8];
158 void (*pred4x4 [9+3])(uint8_t *src, uint8_t *topright, int stride);//FIXME move to dsp?
159 void (*pred8x8 [4+3])(uint8_t *src, int stride);
160 void (*pred16x16[4+3])(uint8_t *src, int stride);
161 unsigned int topleft_samples_available;
162 unsigned int top_samples_available;
163 unsigned int topright_samples_available;
164 unsigned int left_samples_available;
167 * non zero coeff count cache.
168 * is 64 if not available.
170 uint8_t non_zero_count_cache[6*8];
171 uint8_t (*non_zero_count)[16+4+4]; /* store all values for deblocking filter */
174 * Motion vector cache.
176 int16_t mv_cache[2][5*8][2];
177 int8_t ref_cache[2][5*8];
178 #define LIST_NOT_USED -1 //FIXME rename?
179 #define PART_NOT_AVAILABLE -2
182 * is 1 if the specific list MV&references are set to 0,0,-2.
184 int mv_cache_clean[2];
186 int block_offset[16+8];
187 int chroma_subblock_offset[16]; //FIXME remove
189 uint16_t *mb2b_xy; //FIXME are these 4 a good idea?
197 int unknown_svq3_flag;
198 int next_slice_index;
200 SPS sps_buffer[MAX_SPS_COUNT];
201 SPS sps; ///< current sps
203 PPS pps_buffer[MAX_PPS_COUNT];
207 PPS pps; //FIXME move tp Picture perhaps? (->no) do we need that?
210 uint8_t *slice_table_base;
211 uint8_t *slice_table; ///< slice_table_base + mb_stride + 1
213 int slice_type_fixed;
215 //interlacing specific flags
216 int mb_field_decoding_flag;
223 int delta_poc_bottom;
226 int prev_poc_msb; ///< poc_msb of the last reference pic for POC type 0
227 int prev_poc_lsb; ///< poc_lsb of the last reference pic for POC type 0
228 int frame_num_offset; ///< for POC type 2
229 int prev_frame_num_offset; ///< for POC type 2
230 int prev_frame_num; ///< frame_num of the last pic for POC type 1/2
233 * frame_num for frames or 2*frame_num for field pics.
238 * max_frame_num or 2*max_frame_num for field pics.
242 //Weighted pred stuff
243 int luma_log2_weight_denom;
244 int chroma_log2_weight_denom;
245 int luma_weight[2][16];
246 int luma_offset[2][16];
247 int chroma_weight[2][16][2];
248 int chroma_offset[2][16][2];
251 int disable_deblocking_filter_idc;
252 int slice_alpha_c0_offset;
253 int slice_beta_offset;
255 int redundant_pic_count;
257 int direct_spatial_mv_pred;
260 * num_ref_idx_l0/1_active_minus1 + 1
262 int ref_count[2];// FIXME split for AFF
263 Picture *short_ref[16];
264 Picture *long_ref[16];
265 Picture default_ref_list[2][32];
266 Picture ref_list[2][32]; //FIXME size?
267 Picture field_ref_list[2][32]; //FIXME size?
270 * memory management control operations buffer.
272 MMCO mmco[MAX_MMCO_COUNT];
275 int long_ref_count; ///< number of actual long term references
276 int short_ref_count; ///< number of actual short term references
279 GetBitContext intra_gb;
280 GetBitContext inter_gb;
281 GetBitContext *intra_gb_ptr;
282 GetBitContext *inter_gb_ptr;
284 DCTELEM mb[16*24] __align8;
287 static VLC coeff_token_vlc[4];
288 static VLC chroma_dc_coeff_token_vlc;
290 static VLC total_zeros_vlc[15];
291 static VLC chroma_dc_total_zeros_vlc[3];
293 static VLC run_vlc[6];
296 static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
297 static void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
299 static inline uint32_t pack16to32(int a, int b){
300 #ifdef WORDS_BIGENDIAN
301 return (b&0xFFFF) + (a<<16);
303 return (a&0xFFFF) + (b<<16);
309 * @param h height of the recatangle, should be a constant
310 * @param w width of the recatangle, should be a constant
311 * @param size the size of val (1 or 4), should be a constant
313 static inline void fill_rectangle(void *vp, int w, int h, int stride, uint32_t val, int size){ //FIXME ensure this IS inlined
314 uint8_t *p= (uint8_t*)vp;
315 assert(size==1 || size==4);
320 //FIXME check what gcc generates for 64 bit on x86 and possible write a 32 bit ver of it
323 *(uint16_t*)(p + stride)= size==4 ? val : val*0x0101;
324 }else if(w==2 && h==4){
325 *(uint16_t*)(p + 0*stride)=
326 *(uint16_t*)(p + 1*stride)=
327 *(uint16_t*)(p + 2*stride)=
328 *(uint16_t*)(p + 3*stride)= size==4 ? val : val*0x0101;
329 }else if(w==4 && h==1){
330 *(uint32_t*)(p + 0*stride)= size==4 ? val : val*0x01010101;
331 }else if(w==4 && h==2){
332 *(uint32_t*)(p + 0*stride)=
333 *(uint32_t*)(p + 1*stride)= size==4 ? val : val*0x01010101;
334 }else if(w==4 && h==4){
335 *(uint32_t*)(p + 0*stride)=
336 *(uint32_t*)(p + 1*stride)=
337 *(uint32_t*)(p + 2*stride)=
338 *(uint32_t*)(p + 3*stride)= size==4 ? val : val*0x01010101;
339 }else if(w==8 && h==1){
341 *(uint32_t*)(p + 4)= size==4 ? val : val*0x01010101;
342 }else if(w==8 && h==2){
343 *(uint32_t*)(p + 0 + 0*stride)=
344 *(uint32_t*)(p + 4 + 0*stride)=
345 *(uint32_t*)(p + 0 + 1*stride)=
346 *(uint32_t*)(p + 4 + 1*stride)= size==4 ? val : val*0x01010101;
347 }else if(w==8 && h==4){
348 *(uint64_t*)(p + 0*stride)=
349 *(uint64_t*)(p + 1*stride)=
350 *(uint64_t*)(p + 2*stride)=
351 *(uint64_t*)(p + 3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
352 }else if(w==16 && h==2){
353 *(uint64_t*)(p + 0+0*stride)=
354 *(uint64_t*)(p + 8+0*stride)=
355 *(uint64_t*)(p + 0+1*stride)=
356 *(uint64_t*)(p + 8+1*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
357 }else if(w==16 && h==4){
358 *(uint64_t*)(p + 0+0*stride)=
359 *(uint64_t*)(p + 8+0*stride)=
360 *(uint64_t*)(p + 0+1*stride)=
361 *(uint64_t*)(p + 8+1*stride)=
362 *(uint64_t*)(p + 0+2*stride)=
363 *(uint64_t*)(p + 8+2*stride)=
364 *(uint64_t*)(p + 0+3*stride)=
365 *(uint64_t*)(p + 8+3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
370 static inline void fill_caches(H264Context *h, int mb_type){
371 MpegEncContext * const s = &h->s;
372 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
373 int topleft_xy, top_xy, topright_xy, left_xy[2];
374 int topleft_type, top_type, topright_type, left_type[2];
378 //wow what a mess, why didnt they simplify the interlacing&intra stuff, i cant imagine that these complex rules are worth it
382 topleft_xy = 0; /* avoid warning */
383 top_xy = 0; /* avoid warning */
384 topright_xy = 0; /* avoid warning */
386 topleft_xy = mb_xy-1 - s->mb_stride;
387 top_xy = mb_xy - s->mb_stride;
388 topright_xy= mb_xy+1 - s->mb_stride;
389 left_xy[0] = mb_xy-1;
390 left_xy[1] = mb_xy-1;
397 topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
398 top_type = h->slice_table[top_xy ] == h->slice_num ? s->current_picture.mb_type[top_xy] : 0;
399 topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
400 left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
401 left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
403 if(IS_INTRA(mb_type)){
404 h->topleft_samples_available=
405 h->top_samples_available=
406 h->left_samples_available= 0xFFFF;
407 h->topright_samples_available= 0xEEEA;
409 if(!IS_INTRA(top_type) && (top_type==0 || h->pps.constrained_intra_pred)){
410 h->topleft_samples_available= 0xB3FF;
411 h->top_samples_available= 0x33FF;
412 h->topright_samples_available= 0x26EA;
415 if(!IS_INTRA(left_type[i]) && (left_type[i]==0 || h->pps.constrained_intra_pred)){
416 h->topleft_samples_available&= 0xDF5F;
417 h->left_samples_available&= 0x5F5F;
421 if(!IS_INTRA(topleft_type) && (topleft_type==0 || h->pps.constrained_intra_pred))
422 h->topleft_samples_available&= 0x7FFF;
424 if(!IS_INTRA(topright_type) && (topright_type==0 || h->pps.constrained_intra_pred))
425 h->topright_samples_available&= 0xFBFF;
427 if(IS_INTRA4x4(mb_type)){
428 if(IS_INTRA4x4(top_type)){
429 h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
430 h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
431 h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
432 h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
435 if(IS_INTRA16x16(top_type) || (IS_INTER(top_type) && !h->pps.constrained_intra_pred))
440 h->intra4x4_pred_mode_cache[4+8*0]=
441 h->intra4x4_pred_mode_cache[5+8*0]=
442 h->intra4x4_pred_mode_cache[6+8*0]=
443 h->intra4x4_pred_mode_cache[7+8*0]= pred;
446 if(IS_INTRA4x4(left_type[i])){
447 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
448 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
451 if(IS_INTRA16x16(left_type[i]) || (IS_INTER(left_type[i]) && !h->pps.constrained_intra_pred))
456 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
457 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
472 //FIXME constraint_intra_pred & partitioning & nnz (lets hope this is just a typo in the spec)
474 h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][10];
475 h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][11];
476 h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][14];
477 h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][15];
479 h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][18];
480 h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][19];
482 h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][22];
483 h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][23];
485 h->non_zero_count_cache[4+8*0]=
486 h->non_zero_count_cache[5+8*0]=
487 h->non_zero_count_cache[6+8*0]=
488 h->non_zero_count_cache[7+8*0]=
490 h->non_zero_count_cache[1+8*0]=
491 h->non_zero_count_cache[2+8*0]=
493 h->non_zero_count_cache[1+8*3]=
494 h->non_zero_count_cache[2+8*3]= 64;
498 h->non_zero_count_cache[3+8*1]= h->non_zero_count[left_xy[0]][5];
499 h->non_zero_count_cache[3+8*2]= h->non_zero_count[left_xy[0]][7];
500 h->non_zero_count_cache[0+8*1]= h->non_zero_count[left_xy[0]][17]; //FIXME left_block
501 h->non_zero_count_cache[0+8*4]= h->non_zero_count[left_xy[0]][21];
503 h->non_zero_count_cache[3+8*1]=
504 h->non_zero_count_cache[3+8*2]=
505 h->non_zero_count_cache[0+8*1]=
506 h->non_zero_count_cache[0+8*4]= 64;
510 h->non_zero_count_cache[3+8*3]= h->non_zero_count[left_xy[1]][13];
511 h->non_zero_count_cache[3+8*4]= h->non_zero_count[left_xy[1]][15];
512 h->non_zero_count_cache[0+8*2]= h->non_zero_count[left_xy[1]][19];
513 h->non_zero_count_cache[0+8*5]= h->non_zero_count[left_xy[1]][23];
515 h->non_zero_count_cache[3+8*3]=
516 h->non_zero_count_cache[3+8*4]=
517 h->non_zero_count_cache[0+8*2]=
518 h->non_zero_count_cache[0+8*5]= 64;
522 if(IS_INTER(mb_type)){
524 for(list=0; list<2; list++){
525 if((!IS_8X8(mb_type)) && !USES_LIST(mb_type, list)){
526 /*if(!h->mv_cache_clean[list]){
527 memset(h->mv_cache [list], 0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
528 memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
529 h->mv_cache_clean[list]= 1;
531 continue; //FIXME direct mode ...
533 h->mv_cache_clean[list]= 0;
535 if(IS_INTER(topleft_type)){
536 const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
537 const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + h->b8_stride;
538 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
539 h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
541 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
542 h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
545 if(IS_INTER(top_type)){
546 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
547 const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
548 *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0];
549 *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];
550 *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];
551 *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];
552 h->ref_cache[list][scan8[0] + 0 - 1*8]=
553 h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
554 h->ref_cache[list][scan8[0] + 2 - 1*8]=
555 h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
557 *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]=
558 *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]=
559 *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]=
560 *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;
561 *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
564 if(IS_INTER(topright_type)){
565 const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
566 const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
567 *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
568 h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
570 *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
571 h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
574 //FIXME unify cleanup or sth
575 if(IS_INTER(left_type[0])){
576 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
577 const int b8_xy= h->mb2b8_xy[left_xy[0]] + 1;
578 *(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]];
579 *(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]];
580 h->ref_cache[list][scan8[0] - 1 + 0*8]=
581 h->ref_cache[list][scan8[0] - 1 + 1*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0]>>1)];
583 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 0*8]=
584 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 1*8]= 0;
585 h->ref_cache[list][scan8[0] - 1 + 0*8]=
586 h->ref_cache[list][scan8[0] - 1 + 1*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
589 if(IS_INTER(left_type[1])){
590 const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
591 const int b8_xy= h->mb2b8_xy[left_xy[1]] + 1;
592 *(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]];
593 *(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]];
594 h->ref_cache[list][scan8[0] - 1 + 2*8]=
595 h->ref_cache[list][scan8[0] - 1 + 3*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[2]>>1)];
597 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 2*8]=
598 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 3*8]= 0;
599 h->ref_cache[list][scan8[0] - 1 + 2*8]=
600 h->ref_cache[list][scan8[0] - 1 + 3*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
603 h->ref_cache[list][scan8[5 ]+1] =
604 h->ref_cache[list][scan8[7 ]+1] =
605 h->ref_cache[list][scan8[13]+1] = //FIXME remove past 3 (init somewher else)
606 h->ref_cache[list][scan8[4 ]] =
607 h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
608 *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
609 *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
610 *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewher else)
611 *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
612 *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
620 static inline void write_back_intra_pred_mode(H264Context *h){
621 MpegEncContext * const s = &h->s;
622 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
624 h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
625 h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
626 h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
627 h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
628 h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
629 h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
630 h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
634 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
636 static inline int check_intra4x4_pred_mode(H264Context *h){
637 MpegEncContext * const s = &h->s;
638 static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
639 static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
642 if(!(h->top_samples_available&0x8000)){
644 int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
646 av_log(h->s.avctx, AV_LOG_ERROR, "top block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y);
649 h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
654 if(!(h->left_samples_available&0x8000)){
656 int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
658 av_log(h->s.avctx, AV_LOG_ERROR, "left block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y);
661 h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
667 } //FIXME cleanup like next
670 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
672 static inline int check_intra_pred_mode(H264Context *h, int mode){
673 MpegEncContext * const s = &h->s;
674 static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
675 static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
677 if(!(h->top_samples_available&0x8000)){
680 av_log(h->s.avctx, AV_LOG_ERROR, "top block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);
685 if(!(h->left_samples_available&0x8000)){
688 av_log(h->s.avctx, AV_LOG_ERROR, "left block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);
697 * gets the predicted intra4x4 prediction mode.
699 static inline int pred_intra_mode(H264Context *h, int n){
700 const int index8= scan8[n];
701 const int left= h->intra4x4_pred_mode_cache[index8 - 1];
702 const int top = h->intra4x4_pred_mode_cache[index8 - 8];
703 const int min= FFMIN(left, top);
705 tprintf("mode:%d %d min:%d\n", left ,top, min);
707 if(min<0) return DC_PRED;
711 static inline void write_back_non_zero_count(H264Context *h){
712 MpegEncContext * const s = &h->s;
713 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
716 for( n = 0; n < 16+4+4; n++ )
717 h->non_zero_count[mb_xy][n] = h->non_zero_count_cache[scan8[n]];
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);
779 if(match_count > 1){ //most common
780 *mx= mid_pred(A[0], B[0], C[0]);
781 *my= mid_pred(A[1], B[1], C[1]);
782 }else if(match_count==1){
786 }else if(top_ref==ref){
794 if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
798 *mx= mid_pred(A[0], B[0], C[0]);
799 *my= mid_pred(A[1], B[1], C[1]);
803 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);
807 * gets the directionally predicted 16x8 MV.
808 * @param n the block index
809 * @param mx the x component of the predicted motion vector
810 * @param my the y component of the predicted motion vector
812 static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
814 const int top_ref= h->ref_cache[list][ scan8[0] - 8 ];
815 const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
817 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);
825 const int left_ref= h->ref_cache[list][ scan8[8] - 1 ];
826 const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];
828 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);
838 pred_motion(h, n, 4, list, ref, mx, my);
842 * gets the directionally predicted 8x16 MV.
843 * @param n the block index
844 * @param mx the x component of the predicted motion vector
845 * @param my the y component of the predicted motion vector
847 static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
849 const int left_ref= h->ref_cache[list][ scan8[0] - 1 ];
850 const int16_t * const A= h->mv_cache[list][ scan8[0] - 1 ];
852 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);
863 diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);
865 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);
867 if(diagonal_ref == ref){
875 pred_motion(h, n, 2, list, ref, mx, my);
878 static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
879 const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
880 const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];
882 tprintf("pred_pskip: (%d) (%d) at %2d %2d", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
884 if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
885 || (top_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0)
886 || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){
892 pred_motion(h, 0, 4, 0, 0, mx, my);
897 static inline void write_back_motion(H264Context *h, int mb_type){
898 MpegEncContext * const s = &h->s;
899 const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
900 const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
903 for(list=0; list<2; list++){
905 if((!IS_8X8(mb_type)) && !USES_LIST(mb_type, list)){
906 if(1){ //FIXME skip or never read if mb_type doesnt use it
908 *(uint64_t*)s->current_picture.motion_val[list][b_xy + 0 + y*h->b_stride]=
909 *(uint64_t*)s->current_picture.motion_val[list][b_xy + 2 + y*h->b_stride]= 0;
912 *(uint16_t*)s->current_picture.motion_val[list][b8_xy + y*h->b8_stride]= (LIST_NOT_USED&0xFF)*0x0101;
915 continue; //FIXME direct mode ...
919 *(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];
920 *(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];
923 s->current_picture.ref_index[list][b8_xy + 0 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+0 + 16*y];
924 s->current_picture.ref_index[list][b8_xy + 1 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+2 + 16*y];
930 * Decodes a network abstraction layer unit.
931 * @param consumed is the number of bytes used as input
932 * @param length is the length of the array
933 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp ttailing?
934 * @returns decoded bytes, might be src+1 if no escapes
936 static uint8_t *decode_nal(H264Context *h, uint8_t *src, int *dst_length, int *consumed, int length){
940 // src[0]&0x80; //forbidden bit
941 h->nal_ref_idc= src[0]>>5;
942 h->nal_unit_type= src[0]&0x1F;
946 for(i=0; i<length; i++)
947 printf("%2X ", src[i]);
949 for(i=0; i+1<length; i+=2){
951 if(i>0 && src[i-1]==0) i--;
952 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
954 /* startcode, so we must be past the end */
961 if(i>=length-1){ //no escaped 0
963 *consumed= length+1; //+1 for the header
967 h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length);
970 //printf("deoding esc\n");
973 //remove escapes (very rare 1:2^22)
974 if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
975 if(src[si+2]==3){ //escape
979 }else //next start code
983 dst[di++]= src[si++];
987 *consumed= si + 1;//+1 for the header
988 //FIXME store exact number of bits in the getbitcontext (its needed for decoding)
993 * @param src the data which should be escaped
994 * @param dst the target buffer, dst+1 == src is allowed as a special case
995 * @param length the length of the src data
996 * @param dst_length the length of the dst array
997 * @returns length of escaped data in bytes or -1 if an error occured
999 static int encode_nal(H264Context *h, uint8_t *dst, uint8_t *src, int length, int dst_length){
1000 int i, escape_count, si, di;
1004 assert(dst_length>0);
1006 dst[0]= (h->nal_ref_idc<<5) + h->nal_unit_type;
1008 if(length==0) return 1;
1011 for(i=0; i<length; i+=2){
1012 if(src[i]) continue;
1013 if(i>0 && src[i-1]==0)
1015 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1021 if(escape_count==0){
1023 memcpy(dst+1, src, length);
1027 if(length + escape_count + 1> dst_length)
1030 //this should be damn rare (hopefully)
1032 h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length + escape_count);
1033 temp= h->rbsp_buffer;
1034 //printf("encoding esc\n");
1039 if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1040 temp[di++]= 0; si++;
1041 temp[di++]= 0; si++;
1043 temp[di++]= src[si++];
1046 temp[di++]= src[si++];
1048 memcpy(dst+1, temp, length+escape_count);
1050 assert(di == length+escape_count);
1056 * write 1,10,100,1000,... for alignment, yes its exactly inverse to mpeg4
1058 static void encode_rbsp_trailing(PutBitContext *pb){
1061 length= (-put_bits_count(pb))&7;
1062 if(length) put_bits(pb, length, 0);
1066 * identifies the exact end of the bitstream
1067 * @return the length of the trailing, or 0 if damaged
1069 static int decode_rbsp_trailing(uint8_t *src){
1073 tprintf("rbsp trailing %X\n", v);
1083 * idct tranforms the 16 dc values and dequantize them.
1084 * @param qp quantization parameter
1086 static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp){
1087 const int qmul= dequant_coeff[qp][0];
1090 int temp[16]; //FIXME check if this is a good idea
1091 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
1092 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1094 //memset(block, 64, 2*256);
1097 const int offset= y_offset[i];
1098 const int z0= block[offset+stride*0] + block[offset+stride*4];
1099 const int z1= block[offset+stride*0] - block[offset+stride*4];
1100 const int z2= block[offset+stride*1] - block[offset+stride*5];
1101 const int z3= block[offset+stride*1] + block[offset+stride*5];
1110 const int offset= x_offset[i];
1111 const int z0= temp[4*0+i] + temp[4*2+i];
1112 const int z1= temp[4*0+i] - temp[4*2+i];
1113 const int z2= temp[4*1+i] - temp[4*3+i];
1114 const int z3= temp[4*1+i] + temp[4*3+i];
1116 block[stride*0 +offset]= ((z0 + z3)*qmul + 2)>>2; //FIXME think about merging this into decode_resdual
1117 block[stride*2 +offset]= ((z1 + z2)*qmul + 2)>>2;
1118 block[stride*8 +offset]= ((z1 - z2)*qmul + 2)>>2;
1119 block[stride*10+offset]= ((z0 - z3)*qmul + 2)>>2;
1124 * dct tranforms the 16 dc values.
1125 * @param qp quantization parameter ??? FIXME
1127 static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
1128 // const int qmul= dequant_coeff[qp][0];
1130 int temp[16]; //FIXME check if this is a good idea
1131 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
1132 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1135 const int offset= y_offset[i];
1136 const int z0= block[offset+stride*0] + block[offset+stride*4];
1137 const int z1= block[offset+stride*0] - block[offset+stride*4];
1138 const int z2= block[offset+stride*1] - block[offset+stride*5];
1139 const int z3= block[offset+stride*1] + block[offset+stride*5];
1148 const int offset= x_offset[i];
1149 const int z0= temp[4*0+i] + temp[4*2+i];
1150 const int z1= temp[4*0+i] - temp[4*2+i];
1151 const int z2= temp[4*1+i] - temp[4*3+i];
1152 const int z3= temp[4*1+i] + temp[4*3+i];
1154 block[stride*0 +offset]= (z0 + z3)>>1;
1155 block[stride*2 +offset]= (z1 + z2)>>1;
1156 block[stride*8 +offset]= (z1 - z2)>>1;
1157 block[stride*10+offset]= (z0 - z3)>>1;
1163 static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp){
1164 const int qmul= dequant_coeff[qp][0];
1165 const int stride= 16*2;
1166 const int xStride= 16;
1169 a= block[stride*0 + xStride*0];
1170 b= block[stride*0 + xStride*1];
1171 c= block[stride*1 + xStride*0];
1172 d= block[stride*1 + xStride*1];
1179 block[stride*0 + xStride*0]= ((a+c)*qmul + 0)>>1;
1180 block[stride*0 + xStride*1]= ((e+b)*qmul + 0)>>1;
1181 block[stride*1 + xStride*0]= ((a-c)*qmul + 0)>>1;
1182 block[stride*1 + xStride*1]= ((e-b)*qmul + 0)>>1;
1185 static void chroma_dc_dct_c(DCTELEM *block){
1186 const int stride= 16*2;
1187 const int xStride= 16;
1190 a= block[stride*0 + xStride*0];
1191 b= block[stride*0 + xStride*1];
1192 c= block[stride*1 + xStride*0];
1193 d= block[stride*1 + xStride*1];
1200 block[stride*0 + xStride*0]= (a+c);
1201 block[stride*0 + xStride*1]= (e+b);
1202 block[stride*1 + xStride*0]= (a-c);
1203 block[stride*1 + xStride*1]= (e-b);
1207 * gets the chroma qp.
1209 static inline int get_chroma_qp(H264Context *h, int qscale){
1211 return chroma_qp[clip(qscale + h->pps.chroma_qp_index_offset, 0, 51)];
1218 static void h264_add_idct_c(uint8_t *dst, DCTELEM *block, int stride){
1220 uint8_t *cm = cropTbl + MAX_NEG_CROP;
1225 const int z0= block[i + 4*0] + block[i + 4*2];
1226 const int z1= block[i + 4*0] - block[i + 4*2];
1227 const int z2= (block[i + 4*1]>>1) - block[i + 4*3];
1228 const int z3= block[i + 4*1] + (block[i + 4*3]>>1);
1230 block[i + 4*0]= z0 + z3;
1231 block[i + 4*1]= z1 + z2;
1232 block[i + 4*2]= z1 - z2;
1233 block[i + 4*3]= z0 - z3;
1237 const int z0= block[0 + 4*i] + block[2 + 4*i];
1238 const int z1= block[0 + 4*i] - block[2 + 4*i];
1239 const int z2= (block[1 + 4*i]>>1) - block[3 + 4*i];
1240 const int z3= block[1 + 4*i] + (block[3 + 4*i]>>1);
1242 dst[0 + i*stride]= cm[ dst[0 + i*stride] + ((z0 + z3) >> 6) ];
1243 dst[1 + i*stride]= cm[ dst[1 + i*stride] + ((z1 + z2) >> 6) ];
1244 dst[2 + i*stride]= cm[ dst[2 + i*stride] + ((z1 - z2) >> 6) ];
1245 dst[3 + i*stride]= cm[ dst[3 + i*stride] + ((z0 - z3) >> 6) ];
1249 const int z0= block[0 + 4*i] + block[2 + 4*i];
1250 const int z1= block[0 + 4*i] - block[2 + 4*i];
1251 const int z2= (block[1 + 4*i]>>1) - block[3 + 4*i];
1252 const int z3= block[1 + 4*i] + (block[3 + 4*i]>>1);
1254 block[0 + 4*i]= z0 + z3;
1255 block[1 + 4*i]= z1 + z2;
1256 block[2 + 4*i]= z1 - z2;
1257 block[3 + 4*i]= z0 - z3;
1261 const int z0= block[i + 4*0] + block[i + 4*2];
1262 const int z1= block[i + 4*0] - block[i + 4*2];
1263 const int z2= (block[i + 4*1]>>1) - block[i + 4*3];
1264 const int z3= block[i + 4*1] + (block[i + 4*3]>>1);
1266 dst[i + 0*stride]= cm[ dst[i + 0*stride] + ((z0 + z3) >> 6) ];
1267 dst[i + 1*stride]= cm[ dst[i + 1*stride] + ((z1 + z2) >> 6) ];
1268 dst[i + 2*stride]= cm[ dst[i + 2*stride] + ((z1 - z2) >> 6) ];
1269 dst[i + 3*stride]= cm[ dst[i + 3*stride] + ((z0 - z3) >> 6) ];
1274 static void h264_diff_dct_c(DCTELEM *block, uint8_t *src1, uint8_t *src2, int stride){
1276 //FIXME try int temp instead of block
1279 const int d0= src1[0 + i*stride] - src2[0 + i*stride];
1280 const int d1= src1[1 + i*stride] - src2[1 + i*stride];
1281 const int d2= src1[2 + i*stride] - src2[2 + i*stride];
1282 const int d3= src1[3 + i*stride] - src2[3 + i*stride];
1283 const int z0= d0 + d3;
1284 const int z3= d0 - d3;
1285 const int z1= d1 + d2;
1286 const int z2= d1 - d2;
1288 block[0 + 4*i]= z0 + z1;
1289 block[1 + 4*i]= 2*z3 + z2;
1290 block[2 + 4*i]= z0 - z1;
1291 block[3 + 4*i]= z3 - 2*z2;
1295 const int z0= block[0*4 + i] + block[3*4 + i];
1296 const int z3= block[0*4 + i] - block[3*4 + i];
1297 const int z1= block[1*4 + i] + block[2*4 + i];
1298 const int z2= block[1*4 + i] - block[2*4 + i];
1300 block[0*4 + i]= z0 + z1;
1301 block[1*4 + i]= 2*z3 + z2;
1302 block[2*4 + i]= z0 - z1;
1303 block[3*4 + i]= z3 - 2*z2;
1307 //FIXME need to check that this doesnt overflow signed 32 bit for low qp, iam not sure, its very close
1308 //FIXME check that gcc inlines this (and optimizes intra & seperate_dc stuff away)
1309 static inline int quantize_c(DCTELEM *block, uint8_t *scantable, int qscale, int intra, int seperate_dc){
1311 const int * const quant_table= quant_coeff[qscale];
1312 const int bias= intra ? (1<<QUANT_SHIFT)/3 : (1<<QUANT_SHIFT)/6;
1313 const unsigned int threshold1= (1<<QUANT_SHIFT) - bias - 1;
1314 const unsigned int threshold2= (threshold1<<1);
1320 const int dc_bias= intra ? (1<<(QUANT_SHIFT-2))/3 : (1<<(QUANT_SHIFT-2))/6;
1321 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT-2)) - dc_bias - 1;
1322 const unsigned int dc_threshold2= (dc_threshold1<<1);
1324 int level= block[0]*quant_coeff[qscale+18][0];
1325 if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1327 level= (dc_bias + level)>>(QUANT_SHIFT-2);
1330 level= (dc_bias - level)>>(QUANT_SHIFT-2);
1333 // last_non_zero = i;
1338 const int dc_bias= intra ? (1<<(QUANT_SHIFT+1))/3 : (1<<(QUANT_SHIFT+1))/6;
1339 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT+1)) - dc_bias - 1;
1340 const unsigned int dc_threshold2= (dc_threshold1<<1);
1342 int level= block[0]*quant_table[0];
1343 if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1345 level= (dc_bias + level)>>(QUANT_SHIFT+1);
1348 level= (dc_bias - level)>>(QUANT_SHIFT+1);
1351 // last_non_zero = i;
1364 const int j= scantable[i];
1365 int level= block[j]*quant_table[j];
1367 // if( bias+level >= (1<<(QMAT_SHIFT - 3))
1368 // || bias-level >= (1<<(QMAT_SHIFT - 3))){
1369 if(((unsigned)(level+threshold1))>threshold2){
1371 level= (bias + level)>>QUANT_SHIFT;
1374 level= (bias - level)>>QUANT_SHIFT;
1383 return last_non_zero;
1386 static void pred4x4_vertical_c(uint8_t *src, uint8_t *topright, int stride){
1387 const uint32_t a= ((uint32_t*)(src-stride))[0];
1388 ((uint32_t*)(src+0*stride))[0]= a;
1389 ((uint32_t*)(src+1*stride))[0]= a;
1390 ((uint32_t*)(src+2*stride))[0]= a;
1391 ((uint32_t*)(src+3*stride))[0]= a;
1394 static void pred4x4_horizontal_c(uint8_t *src, uint8_t *topright, int stride){
1395 ((uint32_t*)(src+0*stride))[0]= src[-1+0*stride]*0x01010101;
1396 ((uint32_t*)(src+1*stride))[0]= src[-1+1*stride]*0x01010101;
1397 ((uint32_t*)(src+2*stride))[0]= src[-1+2*stride]*0x01010101;
1398 ((uint32_t*)(src+3*stride))[0]= src[-1+3*stride]*0x01010101;
1401 static void pred4x4_dc_c(uint8_t *src, uint8_t *topright, int stride){
1402 const int dc= ( src[-stride] + src[1-stride] + src[2-stride] + src[3-stride]
1403 + src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 4) >>3;
1405 ((uint32_t*)(src+0*stride))[0]=
1406 ((uint32_t*)(src+1*stride))[0]=
1407 ((uint32_t*)(src+2*stride))[0]=
1408 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1411 static void pred4x4_left_dc_c(uint8_t *src, uint8_t *topright, int stride){
1412 const int dc= ( src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 2) >>2;
1414 ((uint32_t*)(src+0*stride))[0]=
1415 ((uint32_t*)(src+1*stride))[0]=
1416 ((uint32_t*)(src+2*stride))[0]=
1417 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1420 static void pred4x4_top_dc_c(uint8_t *src, uint8_t *topright, int stride){
1421 const int dc= ( src[-stride] + src[1-stride] + src[2-stride] + src[3-stride] + 2) >>2;
1423 ((uint32_t*)(src+0*stride))[0]=
1424 ((uint32_t*)(src+1*stride))[0]=
1425 ((uint32_t*)(src+2*stride))[0]=
1426 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1429 static void pred4x4_128_dc_c(uint8_t *src, uint8_t *topright, int stride){
1430 ((uint32_t*)(src+0*stride))[0]=
1431 ((uint32_t*)(src+1*stride))[0]=
1432 ((uint32_t*)(src+2*stride))[0]=
1433 ((uint32_t*)(src+3*stride))[0]= 128U*0x01010101U;
1437 #define LOAD_TOP_RIGHT_EDGE\
1438 const int t4= topright[0];\
1439 const int t5= topright[1];\
1440 const int t6= topright[2];\
1441 const int t7= topright[3];\
1443 #define LOAD_LEFT_EDGE\
1444 const int l0= src[-1+0*stride];\
1445 const int l1= src[-1+1*stride];\
1446 const int l2= src[-1+2*stride];\
1447 const int l3= src[-1+3*stride];\
1449 #define LOAD_TOP_EDGE\
1450 const int t0= src[ 0-1*stride];\
1451 const int t1= src[ 1-1*stride];\
1452 const int t2= src[ 2-1*stride];\
1453 const int t3= src[ 3-1*stride];\
1455 static void pred4x4_down_right_c(uint8_t *src, uint8_t *topright, int stride){
1456 const int lt= src[-1-1*stride];
1460 src[0+3*stride]=(l3 + 2*l2 + l1 + 2)>>2;
1462 src[1+3*stride]=(l2 + 2*l1 + l0 + 2)>>2;
1465 src[2+3*stride]=(l1 + 2*l0 + lt + 2)>>2;
1469 src[3+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1472 src[3+2*stride]=(lt + 2*t0 + t1 + 2)>>2;
1474 src[3+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1475 src[3+0*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1478 static void pred4x4_down_left_c(uint8_t *src, uint8_t *topright, int stride){
1483 src[0+0*stride]=(t0 + t2 + 2*t1 + 2)>>2;
1485 src[0+1*stride]=(t1 + t3 + 2*t2 + 2)>>2;
1488 src[0+2*stride]=(t2 + t4 + 2*t3 + 2)>>2;
1492 src[0+3*stride]=(t3 + t5 + 2*t4 + 2)>>2;
1495 src[1+3*stride]=(t4 + t6 + 2*t5 + 2)>>2;
1497 src[2+3*stride]=(t5 + t7 + 2*t6 + 2)>>2;
1498 src[3+3*stride]=(t6 + 3*t7 + 2)>>2;
1501 static void pred4x4_vertical_right_c(uint8_t *src, uint8_t *topright, int stride){
1502 const int lt= src[-1-1*stride];
1505 const __attribute__((unused)) int unu= l3;
1508 src[1+2*stride]=(lt + t0 + 1)>>1;
1510 src[2+2*stride]=(t0 + t1 + 1)>>1;
1512 src[3+2*stride]=(t1 + t2 + 1)>>1;
1513 src[3+0*stride]=(t2 + t3 + 1)>>1;
1515 src[1+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1517 src[2+3*stride]=(lt + 2*t0 + t1 + 2)>>2;
1519 src[3+3*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1520 src[3+1*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1521 src[0+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1522 src[0+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1525 static void pred4x4_vertical_left_c(uint8_t *src, uint8_t *topright, int stride){
1528 const __attribute__((unused)) int unu= t7;
1530 src[0+0*stride]=(t0 + t1 + 1)>>1;
1532 src[0+2*stride]=(t1 + t2 + 1)>>1;
1534 src[1+2*stride]=(t2 + t3 + 1)>>1;
1536 src[2+2*stride]=(t3 + t4+ 1)>>1;
1537 src[3+2*stride]=(t4 + t5+ 1)>>1;
1538 src[0+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1540 src[0+3*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1542 src[1+3*stride]=(t2 + 2*t3 + t4 + 2)>>2;
1544 src[2+3*stride]=(t3 + 2*t4 + t5 + 2)>>2;
1545 src[3+3*stride]=(t4 + 2*t5 + t6 + 2)>>2;
1548 static void pred4x4_horizontal_up_c(uint8_t *src, uint8_t *topright, int stride){
1551 src[0+0*stride]=(l0 + l1 + 1)>>1;
1552 src[1+0*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1554 src[0+1*stride]=(l1 + l2 + 1)>>1;
1556 src[1+1*stride]=(l1 + 2*l2 + l3 + 2)>>2;
1558 src[0+2*stride]=(l2 + l3 + 1)>>1;
1560 src[1+2*stride]=(l2 + 2*l3 + l3 + 2)>>2;
1569 static void pred4x4_horizontal_down_c(uint8_t *src, uint8_t *topright, int stride){
1570 const int lt= src[-1-1*stride];
1573 const __attribute__((unused)) int unu= t3;
1576 src[2+1*stride]=(lt + l0 + 1)>>1;
1578 src[3+1*stride]=(l0 + 2*lt + t0 + 2)>>2;
1579 src[2+0*stride]=(lt + 2*t0 + t1 + 2)>>2;
1580 src[3+0*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1582 src[2+2*stride]=(l0 + l1 + 1)>>1;
1584 src[3+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1586 src[2+3*stride]=(l1 + l2+ 1)>>1;
1588 src[3+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1589 src[0+3*stride]=(l2 + l3 + 1)>>1;
1590 src[1+3*stride]=(l1 + 2*l2 + l3 + 2)>>2;
1593 static void pred16x16_vertical_c(uint8_t *src, int stride){
1595 const uint32_t a= ((uint32_t*)(src-stride))[0];
1596 const uint32_t b= ((uint32_t*)(src-stride))[1];
1597 const uint32_t c= ((uint32_t*)(src-stride))[2];
1598 const uint32_t d= ((uint32_t*)(src-stride))[3];
1600 for(i=0; i<16; i++){
1601 ((uint32_t*)(src+i*stride))[0]= a;
1602 ((uint32_t*)(src+i*stride))[1]= b;
1603 ((uint32_t*)(src+i*stride))[2]= c;
1604 ((uint32_t*)(src+i*stride))[3]= d;
1608 static void pred16x16_horizontal_c(uint8_t *src, int stride){
1611 for(i=0; i<16; i++){
1612 ((uint32_t*)(src+i*stride))[0]=
1613 ((uint32_t*)(src+i*stride))[1]=
1614 ((uint32_t*)(src+i*stride))[2]=
1615 ((uint32_t*)(src+i*stride))[3]= src[-1+i*stride]*0x01010101;
1619 static void pred16x16_dc_c(uint8_t *src, int stride){
1623 dc+= src[-1+i*stride];
1630 dc= 0x01010101*((dc + 16)>>5);
1632 for(i=0; i<16; i++){
1633 ((uint32_t*)(src+i*stride))[0]=
1634 ((uint32_t*)(src+i*stride))[1]=
1635 ((uint32_t*)(src+i*stride))[2]=
1636 ((uint32_t*)(src+i*stride))[3]= dc;
1640 static void pred16x16_left_dc_c(uint8_t *src, int stride){
1644 dc+= src[-1+i*stride];
1647 dc= 0x01010101*((dc + 8)>>4);
1649 for(i=0; i<16; i++){
1650 ((uint32_t*)(src+i*stride))[0]=
1651 ((uint32_t*)(src+i*stride))[1]=
1652 ((uint32_t*)(src+i*stride))[2]=
1653 ((uint32_t*)(src+i*stride))[3]= dc;
1657 static void pred16x16_top_dc_c(uint8_t *src, int stride){
1663 dc= 0x01010101*((dc + 8)>>4);
1665 for(i=0; i<16; i++){
1666 ((uint32_t*)(src+i*stride))[0]=
1667 ((uint32_t*)(src+i*stride))[1]=
1668 ((uint32_t*)(src+i*stride))[2]=
1669 ((uint32_t*)(src+i*stride))[3]= dc;
1673 static void pred16x16_128_dc_c(uint8_t *src, int stride){
1676 for(i=0; i<16; i++){
1677 ((uint32_t*)(src+i*stride))[0]=
1678 ((uint32_t*)(src+i*stride))[1]=
1679 ((uint32_t*)(src+i*stride))[2]=
1680 ((uint32_t*)(src+i*stride))[3]= 0x01010101U*128U;
1684 static inline void pred16x16_plane_compat_c(uint8_t *src, int stride, const int svq3){
1687 uint8_t *cm = cropTbl + MAX_NEG_CROP;
1688 const uint8_t * const src0 = src+7-stride;
1689 const uint8_t *src1 = src+8*stride-1;
1690 const uint8_t *src2 = src1-2*stride; // == src+6*stride-1;
1691 int H = src0[1] - src0[-1];
1692 int V = src1[0] - src2[ 0];
1693 for(k=2; k<=8; ++k) {
1694 src1 += stride; src2 -= stride;
1695 H += k*(src0[k] - src0[-k]);
1696 V += k*(src1[0] - src2[ 0]);
1699 H = ( 5*(H/4) ) / 16;
1700 V = ( 5*(V/4) ) / 16;
1702 /* required for 100% accuracy */
1703 i = H; H = V; V = i;
1705 H = ( 5*H+32 ) >> 6;
1706 V = ( 5*V+32 ) >> 6;
1709 a = 16*(src1[0] + src2[16] + 1) - 7*(V+H);
1710 for(j=16; j>0; --j) {
1713 for(i=-16; i<0; i+=4) {
1714 src[16+i] = cm[ (b ) >> 5 ];
1715 src[17+i] = cm[ (b+ H) >> 5 ];
1716 src[18+i] = cm[ (b+2*H) >> 5 ];
1717 src[19+i] = cm[ (b+3*H) >> 5 ];
1724 static void pred16x16_plane_c(uint8_t *src, int stride){
1725 pred16x16_plane_compat_c(src, stride, 0);
1728 static void pred8x8_vertical_c(uint8_t *src, int stride){
1730 const uint32_t a= ((uint32_t*)(src-stride))[0];
1731 const uint32_t b= ((uint32_t*)(src-stride))[1];
1734 ((uint32_t*)(src+i*stride))[0]= a;
1735 ((uint32_t*)(src+i*stride))[1]= b;
1739 static void pred8x8_horizontal_c(uint8_t *src, int stride){
1743 ((uint32_t*)(src+i*stride))[0]=
1744 ((uint32_t*)(src+i*stride))[1]= src[-1+i*stride]*0x01010101;
1748 static void pred8x8_128_dc_c(uint8_t *src, int stride){
1752 ((uint32_t*)(src+i*stride))[0]=
1753 ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
1756 ((uint32_t*)(src+i*stride))[0]=
1757 ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
1761 static void pred8x8_left_dc_c(uint8_t *src, int stride){
1767 dc0+= src[-1+i*stride];
1768 dc2+= src[-1+(i+4)*stride];
1770 dc0= 0x01010101*((dc0 + 2)>>2);
1771 dc2= 0x01010101*((dc2 + 2)>>2);
1774 ((uint32_t*)(src+i*stride))[0]=
1775 ((uint32_t*)(src+i*stride))[1]= dc0;
1778 ((uint32_t*)(src+i*stride))[0]=
1779 ((uint32_t*)(src+i*stride))[1]= dc2;
1783 static void pred8x8_top_dc_c(uint8_t *src, int stride){
1789 dc0+= src[i-stride];
1790 dc1+= src[4+i-stride];
1792 dc0= 0x01010101*((dc0 + 2)>>2);
1793 dc1= 0x01010101*((dc1 + 2)>>2);
1796 ((uint32_t*)(src+i*stride))[0]= dc0;
1797 ((uint32_t*)(src+i*stride))[1]= dc1;
1800 ((uint32_t*)(src+i*stride))[0]= dc0;
1801 ((uint32_t*)(src+i*stride))[1]= dc1;
1806 static void pred8x8_dc_c(uint8_t *src, int stride){
1808 int dc0, dc1, dc2, dc3;
1812 dc0+= src[-1+i*stride] + src[i-stride];
1813 dc1+= src[4+i-stride];
1814 dc2+= src[-1+(i+4)*stride];
1816 dc3= 0x01010101*((dc1 + dc2 + 4)>>3);
1817 dc0= 0x01010101*((dc0 + 4)>>3);
1818 dc1= 0x01010101*((dc1 + 2)>>2);
1819 dc2= 0x01010101*((dc2 + 2)>>2);
1822 ((uint32_t*)(src+i*stride))[0]= dc0;
1823 ((uint32_t*)(src+i*stride))[1]= dc1;
1826 ((uint32_t*)(src+i*stride))[0]= dc2;
1827 ((uint32_t*)(src+i*stride))[1]= dc3;
1831 static void pred8x8_plane_c(uint8_t *src, int stride){
1834 uint8_t *cm = cropTbl + MAX_NEG_CROP;
1835 const uint8_t * const src0 = src+3-stride;
1836 const uint8_t *src1 = src+4*stride-1;
1837 const uint8_t *src2 = src1-2*stride; // == src+2*stride-1;
1838 int H = src0[1] - src0[-1];
1839 int V = src1[0] - src2[ 0];
1840 for(k=2; k<=4; ++k) {
1841 src1 += stride; src2 -= stride;
1842 H += k*(src0[k] - src0[-k]);
1843 V += k*(src1[0] - src2[ 0]);
1845 H = ( 17*H+16 ) >> 5;
1846 V = ( 17*V+16 ) >> 5;
1848 a = 16*(src1[0] + src2[8]+1) - 3*(V+H);
1849 for(j=8; j>0; --j) {
1852 src[0] = cm[ (b ) >> 5 ];
1853 src[1] = cm[ (b+ H) >> 5 ];
1854 src[2] = cm[ (b+2*H) >> 5 ];
1855 src[3] = cm[ (b+3*H) >> 5 ];
1856 src[4] = cm[ (b+4*H) >> 5 ];
1857 src[5] = cm[ (b+5*H) >> 5 ];
1858 src[6] = cm[ (b+6*H) >> 5 ];
1859 src[7] = cm[ (b+7*H) >> 5 ];
1864 static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
1865 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1866 int src_x_offset, int src_y_offset,
1867 qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
1868 MpegEncContext * const s = &h->s;
1869 const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
1870 const int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
1871 const int luma_xy= (mx&3) + ((my&3)<<2);
1872 uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*s->linesize;
1873 uint8_t * src_cb= pic->data[1] + (mx>>3) + (my>>3)*s->uvlinesize;
1874 uint8_t * src_cr= pic->data[2] + (mx>>3) + (my>>3)*s->uvlinesize;
1875 int extra_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16; //FIXME increase edge?, IMHO not worth it
1876 int extra_height= extra_width;
1878 const int full_mx= mx>>2;
1879 const int full_my= my>>2;
1881 assert(pic->data[0]);
1883 if(mx&7) extra_width -= 3;
1884 if(my&7) extra_height -= 3;
1886 if( full_mx < 0-extra_width
1887 || full_my < 0-extra_height
1888 || full_mx + 16/*FIXME*/ > s->width + extra_width
1889 || full_my + 16/*FIXME*/ > s->height + extra_height){
1890 ff_emulated_edge_mc(s->edge_emu_buffer, src_y - 2 - 2*s->linesize, s->linesize, 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, s->width, s->height);
1891 src_y= s->edge_emu_buffer + 2 + 2*s->linesize;
1895 qpix_op[luma_xy](dest_y, src_y, s->linesize); //FIXME try variable height perhaps?
1897 qpix_op[luma_xy](dest_y + delta, src_y + delta, s->linesize);
1900 if(s->flags&CODEC_FLAG_GRAY) return;
1903 ff_emulated_edge_mc(s->edge_emu_buffer, src_cb, s->uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), s->width>>1, s->height>>1);
1904 src_cb= s->edge_emu_buffer;
1906 chroma_op(dest_cb, src_cb, s->uvlinesize, chroma_height, mx&7, my&7);
1909 ff_emulated_edge_mc(s->edge_emu_buffer, src_cr, s->uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), s->width>>1, s->height>>1);
1910 src_cr= s->edge_emu_buffer;
1912 chroma_op(dest_cr, src_cr, s->uvlinesize, chroma_height, mx&7, my&7);
1915 static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
1916 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1917 int x_offset, int y_offset,
1918 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
1919 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
1920 int list0, int list1){
1921 MpegEncContext * const s = &h->s;
1922 qpel_mc_func *qpix_op= qpix_put;
1923 h264_chroma_mc_func chroma_op= chroma_put;
1925 dest_y += 2*x_offset + 2*y_offset*s-> linesize;
1926 dest_cb += x_offset + y_offset*s->uvlinesize;
1927 dest_cr += x_offset + y_offset*s->uvlinesize;
1928 x_offset += 8*s->mb_x;
1929 y_offset += 8*s->mb_y;
1932 Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
1933 mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
1934 dest_y, dest_cb, dest_cr, x_offset, y_offset,
1935 qpix_op, chroma_op);
1938 chroma_op= chroma_avg;
1942 Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
1943 mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
1944 dest_y, dest_cb, dest_cr, x_offset, y_offset,
1945 qpix_op, chroma_op);
1949 static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1950 qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
1951 qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg)){
1952 MpegEncContext * const s = &h->s;
1953 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
1954 const int mb_type= s->current_picture.mb_type[mb_xy];
1956 assert(IS_INTER(mb_type));
1958 if(IS_16X16(mb_type)){
1959 mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
1960 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
1961 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1962 }else if(IS_16X8(mb_type)){
1963 mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
1964 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
1965 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1966 mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
1967 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
1968 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
1969 }else if(IS_8X16(mb_type)){
1970 mc_part(h, 0, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 0, 0,
1971 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
1972 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1973 mc_part(h, 4, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 4, 0,
1974 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
1975 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
1979 assert(IS_8X8(mb_type));
1982 const int sub_mb_type= h->sub_mb_type[i];
1984 int x_offset= (i&1)<<2;
1985 int y_offset= (i&2)<<1;
1987 if(IS_SUB_8X8(sub_mb_type)){
1988 mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
1989 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
1990 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1991 }else if(IS_SUB_8X4(sub_mb_type)){
1992 mc_part(h, n , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
1993 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
1994 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1995 mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
1996 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
1997 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
1998 }else if(IS_SUB_4X8(sub_mb_type)){
1999 mc_part(h, n , 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2000 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2001 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2002 mc_part(h, n+1, 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
2003 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2004 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2007 assert(IS_SUB_4X4(sub_mb_type));
2009 int sub_x_offset= x_offset + 2*(j&1);
2010 int sub_y_offset= y_offset + (j&2);
2011 mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
2012 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2013 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2020 static void decode_init_vlc(H264Context *h){
2021 static int done = 0;
2027 init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5,
2028 &chroma_dc_coeff_token_len [0], 1, 1,
2029 &chroma_dc_coeff_token_bits[0], 1, 1);
2032 init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17,
2033 &coeff_token_len [i][0], 1, 1,
2034 &coeff_token_bits[i][0], 1, 1);
2038 init_vlc(&chroma_dc_total_zeros_vlc[i], CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
2039 &chroma_dc_total_zeros_len [i][0], 1, 1,
2040 &chroma_dc_total_zeros_bits[i][0], 1, 1);
2042 for(i=0; i<15; i++){
2043 init_vlc(&total_zeros_vlc[i], TOTAL_ZEROS_VLC_BITS, 16,
2044 &total_zeros_len [i][0], 1, 1,
2045 &total_zeros_bits[i][0], 1, 1);
2049 init_vlc(&run_vlc[i], RUN_VLC_BITS, 7,
2050 &run_len [i][0], 1, 1,
2051 &run_bits[i][0], 1, 1);
2053 init_vlc(&run7_vlc, RUN7_VLC_BITS, 16,
2054 &run_len [6][0], 1, 1,
2055 &run_bits[6][0], 1, 1);
2060 * Sets the intra prediction function pointers.
2062 static void init_pred_ptrs(H264Context *h){
2063 // MpegEncContext * const s = &h->s;
2065 h->pred4x4[VERT_PRED ]= pred4x4_vertical_c;
2066 h->pred4x4[HOR_PRED ]= pred4x4_horizontal_c;
2067 h->pred4x4[DC_PRED ]= pred4x4_dc_c;
2068 h->pred4x4[DIAG_DOWN_LEFT_PRED ]= pred4x4_down_left_c;
2069 h->pred4x4[DIAG_DOWN_RIGHT_PRED]= pred4x4_down_right_c;
2070 h->pred4x4[VERT_RIGHT_PRED ]= pred4x4_vertical_right_c;
2071 h->pred4x4[HOR_DOWN_PRED ]= pred4x4_horizontal_down_c;
2072 h->pred4x4[VERT_LEFT_PRED ]= pred4x4_vertical_left_c;
2073 h->pred4x4[HOR_UP_PRED ]= pred4x4_horizontal_up_c;
2074 h->pred4x4[LEFT_DC_PRED ]= pred4x4_left_dc_c;
2075 h->pred4x4[TOP_DC_PRED ]= pred4x4_top_dc_c;
2076 h->pred4x4[DC_128_PRED ]= pred4x4_128_dc_c;
2078 h->pred8x8[DC_PRED8x8 ]= pred8x8_dc_c;
2079 h->pred8x8[VERT_PRED8x8 ]= pred8x8_vertical_c;
2080 h->pred8x8[HOR_PRED8x8 ]= pred8x8_horizontal_c;
2081 h->pred8x8[PLANE_PRED8x8 ]= pred8x8_plane_c;
2082 h->pred8x8[LEFT_DC_PRED8x8]= pred8x8_left_dc_c;
2083 h->pred8x8[TOP_DC_PRED8x8 ]= pred8x8_top_dc_c;
2084 h->pred8x8[DC_128_PRED8x8 ]= pred8x8_128_dc_c;
2086 h->pred16x16[DC_PRED8x8 ]= pred16x16_dc_c;
2087 h->pred16x16[VERT_PRED8x8 ]= pred16x16_vertical_c;
2088 h->pred16x16[HOR_PRED8x8 ]= pred16x16_horizontal_c;
2089 h->pred16x16[PLANE_PRED8x8 ]= pred16x16_plane_c;
2090 h->pred16x16[LEFT_DC_PRED8x8]= pred16x16_left_dc_c;
2091 h->pred16x16[TOP_DC_PRED8x8 ]= pred16x16_top_dc_c;
2092 h->pred16x16[DC_128_PRED8x8 ]= pred16x16_128_dc_c;
2095 static void free_tables(H264Context *h){
2096 av_freep(&h->intra4x4_pred_mode);
2097 av_freep(&h->non_zero_count);
2098 av_freep(&h->slice_table_base);
2099 h->slice_table= NULL;
2101 av_freep(&h->mb2b_xy);
2102 av_freep(&h->mb2b8_xy);
2107 * needs widzh/height
2109 static int alloc_tables(H264Context *h){
2110 MpegEncContext * const s = &h->s;
2111 const int big_mb_num= s->mb_stride * (s->mb_height+1);
2114 CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8 * sizeof(uint8_t))
2115 CHECKED_ALLOCZ(h->non_zero_count , big_mb_num * (16+4+4) * sizeof(uint8_t))
2116 CHECKED_ALLOCZ(h->slice_table_base , big_mb_num * sizeof(uint8_t))
2118 memset(h->slice_table_base, -1, big_mb_num * sizeof(uint8_t));
2119 h->slice_table= h->slice_table_base + s->mb_stride + 1;
2121 CHECKED_ALLOCZ(h->mb2b_xy , big_mb_num * sizeof(uint16_t));
2122 CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint16_t));
2123 for(y=0; y<s->mb_height; y++){
2124 for(x=0; x<s->mb_width; x++){
2125 const int mb_xy= x + y*s->mb_stride;
2126 const int b_xy = 4*x + 4*y*h->b_stride;
2127 const int b8_xy= 2*x + 2*y*h->b8_stride;
2129 h->mb2b_xy [mb_xy]= b_xy;
2130 h->mb2b8_xy[mb_xy]= b8_xy;
2140 static void common_init(H264Context *h){
2141 MpegEncContext * const s = &h->s;
2143 s->width = s->avctx->width;
2144 s->height = s->avctx->height;
2145 s->codec_id= s->avctx->codec->id;
2149 s->unrestricted_mv=1;
2150 s->decode=1; //FIXME
2153 static int decode_init(AVCodecContext *avctx){
2154 H264Context *h= avctx->priv_data;
2155 MpegEncContext * const s = &h->s;
2157 MPV_decode_defaults(s);
2162 s->out_format = FMT_H264;
2163 s->workaround_bugs= avctx->workaround_bugs;
2166 // s->decode_mb= ff_h263_decode_mb;
2168 avctx->pix_fmt= PIX_FMT_YUV420P;
2175 static void frame_start(H264Context *h){
2176 MpegEncContext * const s = &h->s;
2179 MPV_frame_start(s, s->avctx);
2180 ff_er_frame_start(s);
2183 assert(s->linesize && s->uvlinesize);
2185 for(i=0; i<16; i++){
2186 h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
2187 h->chroma_subblock_offset[i]= 2*((scan8[i] - scan8[0])&7) + 2*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2190 h->block_offset[16+i]=
2191 h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2194 // s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
2197 static void hl_decode_mb(H264Context *h){
2198 MpegEncContext * const s = &h->s;
2199 const int mb_x= s->mb_x;
2200 const int mb_y= s->mb_y;
2201 const int mb_xy= mb_x + mb_y*s->mb_stride;
2202 const int mb_type= s->current_picture.mb_type[mb_xy];
2203 uint8_t *dest_y, *dest_cb, *dest_cr;
2204 int linesize, uvlinesize /*dct_offset*/;
2213 dest_y = s->current_picture.data[0] + (mb_y * 16* s->linesize ) + mb_x * 16;
2214 dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2215 dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2217 if (h->mb_field_decoding_flag) {
2218 linesize = s->linesize * 2;
2219 uvlinesize = s->uvlinesize * 2;
2220 if(mb_y&1){ //FIXME move out of this func?
2221 dest_y -= s->linesize*15;
2222 dest_cb-= s->linesize*7;
2223 dest_cr-= s->linesize*7;
2226 linesize = s->linesize;
2227 uvlinesize = s->uvlinesize;
2228 // dct_offset = s->linesize * 16;
2231 if(IS_INTRA(mb_type)){
2232 if(!(s->flags&CODEC_FLAG_GRAY)){
2233 h->pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
2234 h->pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
2237 if(IS_INTRA4x4(mb_type)){
2239 for(i=0; i<16; i++){
2240 uint8_t * const ptr= dest_y + h->block_offset[i];
2241 uint8_t *topright= ptr + 4 - linesize;
2242 const int topright_avail= (h->topright_samples_available<<i)&0x8000;
2243 const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
2246 if(!topright_avail){
2247 tr= ptr[3 - linesize]*0x01010101;
2248 topright= (uint8_t*) &tr;
2251 h->pred4x4[ dir ](ptr, topright, linesize);
2252 if(h->non_zero_count_cache[ scan8[i] ]){
2253 if(s->codec_id == CODEC_ID_H264)
2254 h264_add_idct_c(ptr, h->mb + i*16, linesize);
2256 svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
2261 h->pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
2262 if(s->codec_id == CODEC_ID_H264)
2263 h264_luma_dc_dequant_idct_c(h->mb, s->qscale);
2265 svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
2267 }else if(s->codec_id == CODEC_ID_H264){
2268 hl_motion(h, dest_y, dest_cb, dest_cr,
2269 s->dsp.put_h264_qpel_pixels_tab, s->dsp.put_h264_chroma_pixels_tab,
2270 s->dsp.avg_h264_qpel_pixels_tab, s->dsp.avg_h264_chroma_pixels_tab);
2274 if(!IS_INTRA4x4(mb_type)){
2275 if(s->codec_id == CODEC_ID_H264){
2276 for(i=0; i<16; i++){
2277 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
2278 uint8_t * const ptr= dest_y + h->block_offset[i];
2279 h264_add_idct_c(ptr, h->mb + i*16, linesize);
2283 for(i=0; i<16; i++){
2284 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
2285 uint8_t * const ptr= dest_y + h->block_offset[i];
2286 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 if(s->codec_id == CODEC_ID_H264){
2296 for(i=16; i<16+4; i++){
2297 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2298 uint8_t * const ptr= dest_cb + h->block_offset[i];
2299 h264_add_idct_c(ptr, h->mb + i*16, uvlinesize);
2302 for(i=20; i<20+4; i++){
2303 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2304 uint8_t * const ptr= dest_cr + h->block_offset[i];
2305 h264_add_idct_c(ptr, h->mb + i*16, uvlinesize);
2309 for(i=16; i<16+4; i++){
2310 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2311 uint8_t * const ptr= dest_cb + h->block_offset[i];
2312 svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
2315 for(i=20; i<20+4; i++){
2316 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2317 uint8_t * const ptr= dest_cr + h->block_offset[i];
2318 svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
2325 static void decode_mb_cabac(H264Context *h){
2326 // MpegEncContext * const s = &h->s;
2330 * fills the default_ref_list.
2332 static int fill_default_ref_list(H264Context *h){
2333 MpegEncContext * const s = &h->s;
2335 Picture sorted_short_ref[16];
2337 if(h->slice_type==B_TYPE){
2341 for(out_i=0; out_i<h->short_ref_count; out_i++){
2345 for(i=0; i<h->short_ref_count; i++){
2346 const int poc= h->short_ref[i]->poc;
2347 if(poc > limit && poc < best_poc){
2353 assert(best_i != -1);
2356 sorted_short_ref[out_i]= *h->short_ref[best_i];
2360 if(s->picture_structure == PICT_FRAME){
2361 if(h->slice_type==B_TYPE){
2362 const int current_poc= s->current_picture_ptr->poc;
2365 for(list=0; list<2; list++){
2368 for(i=0; i<h->short_ref_count && index < h->ref_count[list]; i++){
2369 const int i2= list ? h->short_ref_count - i - 1 : i;
2370 const int poc= sorted_short_ref[i2].poc;
2372 if(sorted_short_ref[i2].reference != 3) continue; //FIXME refernce field shit
2374 if((list==1 && poc > current_poc) || (list==0 && poc < current_poc)){
2375 h->default_ref_list[list][index ]= sorted_short_ref[i2];
2376 h->default_ref_list[list][index++].pic_id= sorted_short_ref[i2].frame_num;
2380 for(i=0; i<h->long_ref_count && index < h->ref_count[ list ]; i++){
2381 if(h->long_ref[i]->reference != 3) continue;
2383 h->default_ref_list[ list ][index ]= *h->long_ref[i];
2384 h->default_ref_list[ list ][index++].pic_id= i;;
2387 if(h->long_ref_count > 1 && h->short_ref_count==0){
2388 Picture temp= h->default_ref_list[1][0];
2389 h->default_ref_list[1][0] = h->default_ref_list[1][1];
2390 h->default_ref_list[1][0] = temp;
2393 if(index < h->ref_count[ list ])
2394 memset(&h->default_ref_list[list][index], 0, sizeof(Picture)*(h->ref_count[ list ] - index));
2398 for(i=0; i<h->short_ref_count && index < h->ref_count[0]; i++){
2399 if(h->short_ref[i]->reference != 3) continue; //FIXME refernce field shit
2400 h->default_ref_list[0][index ]= *h->short_ref[i];
2401 h->default_ref_list[0][index++].pic_id= h->short_ref[i]->frame_num;
2403 for(i=0; i<h->long_ref_count && index < h->ref_count[0]; i++){
2404 if(h->long_ref[i]->reference != 3) continue;
2405 h->default_ref_list[0][index ]= *h->long_ref[i];
2406 h->default_ref_list[0][index++].pic_id= i;;
2408 if(index < h->ref_count[0])
2409 memset(&h->default_ref_list[0][index], 0, sizeof(Picture)*(h->ref_count[0] - index));
2412 if(h->slice_type==B_TYPE){
2414 //FIXME second field balh
2420 static int decode_ref_pic_list_reordering(H264Context *h){
2421 MpegEncContext * const s = &h->s;
2424 if(h->slice_type==I_TYPE || h->slice_type==SI_TYPE) return 0; //FIXME move beofre func
2426 for(list=0; list<2; list++){
2427 memcpy(h->ref_list[list], h->default_ref_list[list], sizeof(Picture)*h->ref_count[list]);
2429 if(get_bits1(&s->gb)){
2430 int pred= h->curr_pic_num;
2433 for(index=0; ; index++){
2434 int reordering_of_pic_nums_idc= get_ue_golomb(&s->gb);
2439 if(index >= h->ref_count[list]){
2440 av_log(h->s.avctx, AV_LOG_ERROR, "reference count overflow\n");
2444 if(reordering_of_pic_nums_idc<3){
2445 if(reordering_of_pic_nums_idc<2){
2446 const int abs_diff_pic_num= get_ue_golomb(&s->gb) + 1;
2448 if(abs_diff_pic_num >= h->max_pic_num){
2449 av_log(h->s.avctx, AV_LOG_ERROR, "abs_diff_pic_num overflow\n");
2453 if(reordering_of_pic_nums_idc == 0) pred-= abs_diff_pic_num;
2454 else pred+= abs_diff_pic_num;
2455 pred &= h->max_pic_num - 1;
2457 for(i= h->ref_count[list]-1; i>=index; i--){
2458 if(h->ref_list[list][i].pic_id == pred && h->ref_list[list][i].long_ref==0)
2462 pic_id= get_ue_golomb(&s->gb); //long_term_pic_idx
2464 for(i= h->ref_count[list]-1; i>=index; i--){
2465 if(h->ref_list[list][i].pic_id == pic_id && h->ref_list[list][i].long_ref==1)
2471 av_log(h->s.avctx, AV_LOG_ERROR, "reference picture missing during reorder\n");
2472 memset(&h->ref_list[list][index], 0, sizeof(Picture)); //FIXME
2473 }else if(i > index){
2474 Picture tmp= h->ref_list[list][i];
2475 for(; i>index; i--){
2476 h->ref_list[list][i]= h->ref_list[list][i-1];
2478 h->ref_list[list][index]= tmp;
2480 }else if(reordering_of_pic_nums_idc==3)
2483 av_log(h->s.avctx, AV_LOG_ERROR, "illegal reordering_of_pic_nums_idc\n");
2489 if(h->slice_type!=B_TYPE) break;
2494 static int pred_weight_table(H264Context *h){
2495 MpegEncContext * const s = &h->s;
2498 h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
2499 h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
2501 for(list=0; list<2; list++){
2502 for(i=0; i<h->ref_count[list]; i++){
2503 int luma_weight_flag, chroma_weight_flag;
2505 luma_weight_flag= get_bits1(&s->gb);
2506 if(luma_weight_flag){
2507 h->luma_weight[list][i]= get_se_golomb(&s->gb);
2508 h->luma_offset[list][i]= get_se_golomb(&s->gb);
2511 chroma_weight_flag= get_bits1(&s->gb);
2512 if(chroma_weight_flag){
2515 h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);
2516 h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);
2520 if(h->slice_type != B_TYPE) break;
2526 * instantaneos decoder refresh.
2528 static void idr(H264Context *h){
2531 for(i=0; i<h->long_ref_count; i++){
2532 h->long_ref[i]->reference=0;
2533 h->long_ref[i]= NULL;
2535 h->long_ref_count=0;
2537 for(i=0; i<h->short_ref_count; i++){
2538 h->short_ref[i]->reference=0;
2539 h->short_ref[i]= NULL;
2541 h->short_ref_count=0;
2546 * @return the removed picture or NULL if an error occures
2548 static Picture * remove_short(H264Context *h, int frame_num){
2549 MpegEncContext * const s = &h->s;
2552 if(s->avctx->debug&FF_DEBUG_MMCO)
2553 av_log(h->s.avctx, AV_LOG_DEBUG, "remove short %d count %d\n", frame_num, h->short_ref_count);
2555 for(i=0; i<h->short_ref_count; i++){
2556 Picture *pic= h->short_ref[i];
2557 if(s->avctx->debug&FF_DEBUG_MMCO)
2558 av_log(h->s.avctx, AV_LOG_DEBUG, "%d %d %p\n", i, pic->frame_num, pic);
2559 if(pic->frame_num == frame_num){
2560 h->short_ref[i]= NULL;
2561 memmove(&h->short_ref[i], &h->short_ref[i+1], (h->short_ref_count - i - 1)*sizeof(Picture*));
2562 h->short_ref_count--;
2571 * @return the removed picture or NULL if an error occures
2573 static Picture * remove_long(H264Context *h, int i){
2576 if(i >= h->long_ref_count) return NULL;
2577 pic= h->long_ref[i];
2578 if(pic==NULL) return NULL;
2580 h->long_ref[i]= NULL;
2581 memmove(&h->long_ref[i], &h->long_ref[i+1], (h->long_ref_count - i - 1)*sizeof(Picture*));
2582 h->long_ref_count--;
2588 * Executes the reference picture marking (memory management control operations).
2590 static int execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count){
2591 MpegEncContext * const s = &h->s;
2593 int current_is_long=0;
2596 if((s->avctx->debug&FF_DEBUG_MMCO) && mmco_count==0)
2597 av_log(h->s.avctx, AV_LOG_DEBUG, "no mmco here\n");
2599 for(i=0; i<mmco_count; i++){
2600 if(s->avctx->debug&FF_DEBUG_MMCO)
2601 av_log(h->s.avctx, AV_LOG_DEBUG, "mmco:%d %d %d\n", h->mmco[i].opcode, h->mmco[i].short_frame_num, h->mmco[i].long_index);
2603 switch(mmco[i].opcode){
2604 case MMCO_SHORT2UNUSED:
2605 pic= remove_short(h, mmco[i].short_frame_num);
2606 if(pic==NULL) return -1;
2609 case MMCO_SHORT2LONG:
2610 pic= remove_long(h, mmco[i].long_index);
2611 if(pic) pic->reference=0;
2613 h->long_ref[ mmco[i].long_index ]= remove_short(h, mmco[i].short_frame_num);
2614 h->long_ref[ mmco[i].long_index ]->long_ref=1;
2616 case MMCO_LONG2UNUSED:
2617 pic= remove_long(h, mmco[i].long_index);
2618 if(pic==NULL) return -1;
2622 pic= remove_long(h, mmco[i].long_index);
2623 if(pic) pic->reference=0;
2625 h->long_ref[ mmco[i].long_index ]= s->current_picture_ptr;
2626 h->long_ref[ mmco[i].long_index ]->long_ref=1;
2627 h->long_ref_count++;
2631 case MMCO_SET_MAX_LONG:
2632 assert(mmco[i].long_index <= 16);
2633 while(mmco[i].long_index < h->long_ref_count){
2634 pic= remove_long(h, mmco[i].long_index);
2637 while(mmco[i].long_index > h->long_ref_count){
2638 h->long_ref[ h->long_ref_count++ ]= NULL;
2642 while(h->short_ref_count){
2643 pic= remove_short(h, h->short_ref[0]->frame_num);
2646 while(h->long_ref_count){
2647 pic= remove_long(h, h->long_ref_count-1);
2655 if(!current_is_long){
2656 pic= remove_short(h, s->current_picture_ptr->frame_num);
2659 av_log(h->s.avctx, AV_LOG_ERROR, "illegal short term buffer state detected\n");
2662 if(h->short_ref_count)
2663 memmove(&h->short_ref[1], &h->short_ref[0], h->short_ref_count*sizeof(Picture*));
2665 h->short_ref[0]= s->current_picture_ptr;
2666 h->short_ref[0]->long_ref=0;
2667 h->short_ref_count++;
2673 static int decode_ref_pic_marking(H264Context *h){
2674 MpegEncContext * const s = &h->s;
2677 if(h->nal_unit_type == NAL_IDR_SLICE){ //FIXME fields
2678 s->broken_link= get_bits1(&s->gb) -1;
2679 h->mmco[0].long_index= get_bits1(&s->gb) - 1; // current_long_term_idx
2680 if(h->mmco[0].long_index == -1)
2683 h->mmco[0].opcode= MMCO_LONG;
2687 if(get_bits1(&s->gb)){ // adaptive_ref_pic_marking_mode_flag
2688 for(i= h->mmco_index; i<MAX_MMCO_COUNT; i++) {
2689 MMCOOpcode opcode= get_ue_golomb(&s->gb);;
2691 h->mmco[i].opcode= opcode;
2692 if(opcode==MMCO_SHORT2UNUSED || opcode==MMCO_SHORT2LONG){
2693 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
2694 /* if(h->mmco[i].short_frame_num >= h->short_ref_count || h->short_ref[ h->mmco[i].short_frame_num ] == NULL){
2695 fprintf(stderr, "illegal short ref in memory management control operation %d\n", mmco);
2699 if(opcode==MMCO_SHORT2LONG || opcode==MMCO_LONG2UNUSED || opcode==MMCO_LONG || opcode==MMCO_SET_MAX_LONG){
2700 h->mmco[i].long_index= get_ue_golomb(&s->gb);
2701 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){
2702 av_log(h->s.avctx, AV_LOG_ERROR, "illegal long ref in memory management control operation %d\n", opcode);
2707 if(opcode > MMCO_LONG){
2708 av_log(h->s.avctx, AV_LOG_ERROR, "illegal memory management control operation %d\n", opcode);
2714 assert(h->long_ref_count + h->short_ref_count <= h->sps.ref_frame_count);
2716 if(h->long_ref_count + h->short_ref_count == h->sps.ref_frame_count){ //FIXME fields
2717 h->mmco[0].opcode= MMCO_SHORT2UNUSED;
2718 h->mmco[0].short_frame_num= h->short_ref[ h->short_ref_count - 1 ]->frame_num;
2728 static int init_poc(H264Context *h){
2729 MpegEncContext * const s = &h->s;
2730 const int max_frame_num= 1<<h->sps.log2_max_frame_num;
2733 if(h->nal_unit_type == NAL_IDR_SLICE){
2734 h->frame_num_offset= 0;
2736 if(h->frame_num < h->prev_frame_num)
2737 h->frame_num_offset= h->prev_frame_num_offset + max_frame_num;
2739 h->frame_num_offset= h->prev_frame_num_offset;
2742 if(h->sps.poc_type==0){
2743 const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
2745 if (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
2746 h->poc_msb = h->prev_poc_msb + max_poc_lsb;
2747 else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
2748 h->poc_msb = h->prev_poc_msb - max_poc_lsb;
2750 h->poc_msb = h->prev_poc_msb;
2751 //printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
2753 field_poc[1] = h->poc_msb + h->poc_lsb;
2754 if(s->picture_structure == PICT_FRAME)
2755 field_poc[1] += h->delta_poc_bottom;
2756 }else if(h->sps.poc_type==1){
2757 int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
2760 if(h->sps.poc_cycle_length != 0)
2761 abs_frame_num = h->frame_num_offset + h->frame_num;
2765 if(h->nal_ref_idc==0 && abs_frame_num > 0)
2768 expected_delta_per_poc_cycle = 0;
2769 for(i=0; i < h->sps.poc_cycle_length; i++)
2770 expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
2772 if(abs_frame_num > 0){
2773 int poc_cycle_cnt = (abs_frame_num - 1) / h->sps.poc_cycle_length;
2774 int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
2776 expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
2777 for(i = 0; i <= frame_num_in_poc_cycle; i++)
2778 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
2782 if(h->nal_ref_idc == 0)
2783 expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
2785 field_poc[0] = expectedpoc + h->delta_poc[0];
2786 field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
2788 if(s->picture_structure == PICT_FRAME)
2789 field_poc[1] += h->delta_poc[1];
2792 if(h->nal_unit_type == NAL_IDR_SLICE){
2795 if(h->nal_ref_idc) poc= 2*(h->frame_num_offset + h->frame_num);
2796 else poc= 2*(h->frame_num_offset + h->frame_num) - 1;
2802 if(s->picture_structure != PICT_BOTTOM_FIELD)
2803 s->current_picture_ptr->field_poc[0]= field_poc[0];
2804 if(s->picture_structure != PICT_TOP_FIELD)
2805 s->current_picture_ptr->field_poc[1]= field_poc[1];
2806 if(s->picture_structure == PICT_FRAME) // FIXME field pix?
2807 s->current_picture_ptr->poc= FFMIN(field_poc[0], field_poc[1]);
2813 * decodes a slice header.
2814 * this will allso call MPV_common_init() and frame_start() as needed
2816 static int decode_slice_header(H264Context *h){
2817 MpegEncContext * const s = &h->s;
2818 int first_mb_in_slice, pps_id;
2819 int num_ref_idx_active_override_flag;
2820 static const uint8_t slice_type_map[5]= {P_TYPE, B_TYPE, I_TYPE, SP_TYPE, SI_TYPE};
2822 s->current_picture.reference= h->nal_ref_idc != 0;
2824 first_mb_in_slice= get_ue_golomb(&s->gb);
2826 h->slice_type= get_ue_golomb(&s->gb);
2827 if(h->slice_type > 9){
2828 av_log(h->s.avctx, AV_LOG_ERROR, "slice type too large (%d) at %d %d\n", h->slice_type, s->mb_x, s->mb_y);
2830 if(h->slice_type > 4){
2832 h->slice_type_fixed=1;
2834 h->slice_type_fixed=0;
2836 h->slice_type= slice_type_map[ h->slice_type ];
2838 s->pict_type= h->slice_type; // to make a few old func happy, its wrong though
2840 pps_id= get_ue_golomb(&s->gb);
2842 av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
2845 h->pps= h->pps_buffer[pps_id];
2846 if(h->pps.slice_group_count == 0){
2847 av_log(h->s.avctx, AV_LOG_ERROR, "non existing PPS referenced\n");
2851 h->sps= h->sps_buffer[ h->pps.sps_id ];
2852 if(h->sps.log2_max_frame_num == 0){
2853 av_log(h->s.avctx, AV_LOG_ERROR, "non existing SPS referenced\n");
2857 s->mb_width= h->sps.mb_width;
2858 s->mb_height= h->sps.mb_height;
2860 h->b_stride= s->mb_width*4;
2861 h->b8_stride= s->mb_width*2;
2863 s->mb_x = first_mb_in_slice % s->mb_width;
2864 s->mb_y = first_mb_in_slice / s->mb_width; //FIXME AFFW
2866 s->width = 16*s->mb_width - 2*(h->sps.crop_left + h->sps.crop_right );
2867 if(h->sps.frame_mbs_only_flag)
2868 s->height= 16*s->mb_height - 2*(h->sps.crop_top + h->sps.crop_bottom);
2870 s->height= 16*s->mb_height - 4*(h->sps.crop_top + h->sps.crop_bottom); //FIXME recheck
2872 if (s->context_initialized
2873 && ( s->width != s->avctx->width || s->height != s->avctx->height)) {
2877 if (!s->context_initialized) {
2878 if (MPV_common_init(s) < 0)
2883 s->avctx->width = s->width;
2884 s->avctx->height = s->height;
2885 s->avctx->sample_aspect_ratio= h->sps.sar;
2888 if(first_mb_in_slice == 0){
2892 s->current_picture_ptr->frame_num= //FIXME frame_num cleanup
2893 h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
2895 if(h->sps.frame_mbs_only_flag){
2896 s->picture_structure= PICT_FRAME;
2898 if(get_bits1(&s->gb)) //field_pic_flag
2899 s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
2901 s->picture_structure= PICT_FRAME;
2904 if(s->picture_structure==PICT_FRAME){
2905 h->curr_pic_num= h->frame_num;
2906 h->max_pic_num= 1<< h->sps.log2_max_frame_num;
2908 h->curr_pic_num= 2*h->frame_num;
2909 h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
2912 if(h->nal_unit_type == NAL_IDR_SLICE){
2913 get_ue_golomb(&s->gb); /* idr_pic_id */
2916 if(h->sps.poc_type==0){
2917 h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
2919 if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
2920 h->delta_poc_bottom= get_se_golomb(&s->gb);
2924 if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
2925 h->delta_poc[0]= get_se_golomb(&s->gb);
2927 if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
2928 h->delta_poc[1]= get_se_golomb(&s->gb);
2933 if(h->pps.redundant_pic_cnt_present){
2934 h->redundant_pic_count= get_ue_golomb(&s->gb);
2937 //set defaults, might be overriden a few line later
2938 h->ref_count[0]= h->pps.ref_count[0];
2939 h->ref_count[1]= h->pps.ref_count[1];
2941 if(h->slice_type == P_TYPE || h->slice_type == SP_TYPE || h->slice_type == B_TYPE){
2942 if(h->slice_type == B_TYPE){
2943 h->direct_spatial_mv_pred= get_bits1(&s->gb);
2945 num_ref_idx_active_override_flag= get_bits1(&s->gb);
2947 if(num_ref_idx_active_override_flag){
2948 h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
2949 if(h->slice_type==B_TYPE)
2950 h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
2952 if(h->ref_count[0] > 32 || h->ref_count[1] > 32){
2953 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
2959 if(first_mb_in_slice == 0){
2960 fill_default_ref_list(h);
2963 decode_ref_pic_list_reordering(h);
2965 if( (h->pps.weighted_pred && (h->slice_type == P_TYPE || h->slice_type == SP_TYPE ))
2966 || (h->pps.weighted_bipred_idc==1 && h->slice_type==B_TYPE ) )
2967 pred_weight_table(h);
2969 if(s->current_picture.reference)
2970 decode_ref_pic_marking(h);
2973 s->qscale = h->pps.init_qp + get_se_golomb(&s->gb); //slice_qp_delta
2974 //FIXME qscale / qp ... stuff
2975 if(h->slice_type == SP_TYPE){
2976 get_bits1(&s->gb); /* sp_for_switch_flag */
2978 if(h->slice_type==SP_TYPE || h->slice_type == SI_TYPE){
2979 get_se_golomb(&s->gb); /* slice_qs_delta */
2982 if( h->pps.deblocking_filter_parameters_present ) {
2983 h->disable_deblocking_filter_idc= get_ue_golomb(&s->gb);
2984 if( h->disable_deblocking_filter_idc != 1 ) {
2985 h->slice_alpha_c0_offset = get_se_golomb(&s->gb) << 1;
2986 h->slice_beta_offset = get_se_golomb(&s->gb) << 1;
2988 h->slice_alpha_c0_offset = 0;
2989 h->slice_beta_offset = 0;
2992 h->disable_deblocking_filter_idc = 0;
2993 h->slice_alpha_c0_offset = 0;
2994 h->slice_beta_offset = 0;
2998 if( h->pps.num_slice_groups > 1 && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
2999 slice_group_change_cycle= get_bits(&s->gb, ?);
3002 if(s->avctx->debug&FF_DEBUG_PICT_INFO){
3003 av_log(h->s.avctx, AV_LOG_DEBUG, "mb:%d %c pps:%d frame:%d poc:%d/%d ref:%d/%d qp:%d loop:%d\n",
3005 av_get_pict_type_char(h->slice_type),
3006 pps_id, h->frame_num,
3007 s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
3008 h->ref_count[0], h->ref_count[1],
3010 h->disable_deblocking_filter_idc
3020 static inline int get_level_prefix(GetBitContext *gb){
3024 OPEN_READER(re, gb);
3025 UPDATE_CACHE(re, gb);
3026 buf=GET_CACHE(re, gb);
3028 log= 32 - av_log2(buf);
3030 print_bin(buf>>(32-log), log);
3031 printf("%5d %2d %3d lpr @%5d in %s get_level_prefix\n", buf>>(32-log), log, log-1, get_bits_count(gb), __FILE__);
3034 LAST_SKIP_BITS(re, gb, log);
3035 CLOSE_READER(re, gb);
3041 * decodes a residual block.
3042 * @param n block index
3043 * @param scantable scantable
3044 * @param max_coeff number of coefficients in the block
3045 * @return <0 if an error occured
3047 static int decode_residual(H264Context *h, GetBitContext *gb, DCTELEM *block, int n, const uint8_t *scantable, int qp, int max_coeff){
3048 MpegEncContext * const s = &h->s;
3049 const uint16_t *qmul= dequant_coeff[qp];
3050 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};
3051 int level[16], run[16];
3052 int suffix_length, zeros_left, coeff_num, coeff_token, total_coeff, i, trailing_ones;
3054 //FIXME put trailing_onex into the context
3056 if(n == CHROMA_DC_BLOCK_INDEX){
3057 coeff_token= get_vlc2(gb, chroma_dc_coeff_token_vlc.table, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 1);
3058 total_coeff= coeff_token>>2;
3060 if(n == LUMA_DC_BLOCK_INDEX){
3061 total_coeff= pred_non_zero_count(h, 0);
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;
3065 total_coeff= pred_non_zero_count(h, n);
3066 coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
3067 total_coeff= coeff_token>>2;
3068 h->non_zero_count_cache[ scan8[n] ]= total_coeff;
3072 //FIXME set last_non_zero?
3077 trailing_ones= coeff_token&3;
3078 tprintf("trailing:%d, total:%d\n", trailing_ones, total_coeff);
3079 assert(total_coeff<=16);
3081 for(i=0; i<trailing_ones; i++){
3082 level[i]= 1 - 2*get_bits1(gb);
3085 suffix_length= total_coeff > 10 && trailing_ones < 3;
3087 for(; i<total_coeff; i++){
3088 const int prefix= get_level_prefix(gb);
3089 int level_code, mask;
3091 if(prefix<14){ //FIXME try to build a large unified VLC table for all this
3093 level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
3095 level_code= (prefix<<suffix_length); //part
3096 }else if(prefix==14){
3098 level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
3100 level_code= prefix + get_bits(gb, 4); //part
3101 }else if(prefix==15){
3102 level_code= (prefix<<suffix_length) + get_bits(gb, 12); //part
3103 if(suffix_length==0) level_code+=15; //FIXME doesnt make (much)sense
3105 av_log(h->s.avctx, AV_LOG_ERROR, "prefix too large at %d %d\n", s->mb_x, s->mb_y);
3109 if(i==trailing_ones && i<3) level_code+= 2; //FIXME split first iteration
3111 mask= -(level_code&1);
3112 level[i]= (((2+level_code)>>1) ^ mask) - mask;
3114 if(suffix_length==0) suffix_length=1; //FIXME split first iteration
3117 if(ABS(level[i]) > (3<<(suffix_length-1)) && suffix_length<6) suffix_length++;
3119 if((2+level_code)>>1) > (3<<(suffix_length-1)) && suffix_length<6) suffix_length++;
3120 /* ? == prefix > 2 or sth */
3122 tprintf("level: %d suffix_length:%d\n", level[i], suffix_length);
3125 if(total_coeff == max_coeff)
3128 if(n == CHROMA_DC_BLOCK_INDEX)
3129 zeros_left= get_vlc2(gb, chroma_dc_total_zeros_vlc[ total_coeff-1 ].table, CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 1);
3131 zeros_left= get_vlc2(gb, total_zeros_vlc[ total_coeff-1 ].table, TOTAL_ZEROS_VLC_BITS, 1);
3134 for(i=0; i<total_coeff-1; i++){
3137 else if(zeros_left < 7){
3138 run[i]= get_vlc2(gb, run_vlc[zeros_left-1].table, RUN_VLC_BITS, 1);
3140 run[i]= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2);
3142 zeros_left -= run[i];
3146 av_log(h->s.avctx, AV_LOG_ERROR, "negative number of zero coeffs at %d %d\n", s->mb_x, s->mb_y);
3150 for(; i<total_coeff-1; i++){
3158 for(i=total_coeff-1; i>=0; i--){ //FIXME merge into rundecode?
3161 coeff_num += run[i] + 1; //FIXME add 1 earlier ?
3162 j= scantable[ coeff_num ];
3167 for(i=total_coeff-1; i>=0; i--){ //FIXME merge into rundecode?
3170 coeff_num += run[i] + 1; //FIXME add 1 earlier ?
3171 j= scantable[ coeff_num ];
3173 block[j]= level[i] * qmul[j];
3174 // printf("%d %d ", block[j], qmul[j]);
3181 * decodes a macroblock
3182 * @returns 0 if ok, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
3184 static int decode_mb(H264Context *h){
3185 MpegEncContext * const s = &h->s;
3186 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
3187 int mb_type, partition_count, cbp;
3189 s->dsp.clear_blocks(h->mb); //FIXME avoid if allready clear (move after skip handlong?
3191 tprintf("pic:%d mb:%d/%d\n", h->frame_num, s->mb_x, s->mb_y);
3192 cbp = 0; /* avoid warning. FIXME: find a solution without slowing
3194 if(h->slice_type != I_TYPE && h->slice_type != SI_TYPE){
3195 if(s->mb_skip_run==-1)
3196 s->mb_skip_run= get_ue_golomb(&s->gb);
3198 if (s->mb_skip_run--) {
3202 mb_type= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0;
3204 memset(h->non_zero_count[mb_xy], 0, 16+4+4);
3205 memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui
3207 if(h->sps.mb_aff && s->mb_skip_run==0 && (s->mb_y&1)==0){
3208 h->mb_field_decoding_flag= get_bits1(&s->gb);
3211 if(h->mb_field_decoding_flag)
3212 mb_type|= MB_TYPE_INTERLACED;
3214 fill_caches(h, mb_type); //FIXME check what is needed and what not ...
3215 pred_pskip_motion(h, &mx, &my);
3216 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
3217 fill_rectangle( h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4);
3218 write_back_motion(h, mb_type);
3220 s->current_picture.mb_type[mb_xy]= mb_type; //FIXME SKIP type
3221 s->current_picture.qscale_table[mb_xy]= s->qscale;
3222 h->slice_table[ mb_xy ]= h->slice_num;
3224 h->prev_mb_skiped= 1;
3228 if(h->sps.mb_aff /* && !field pic FIXME needed? */){
3230 h->mb_field_decoding_flag = get_bits1(&s->gb);
3232 h->mb_field_decoding_flag=0; //FIXME som ed note ?!
3234 h->prev_mb_skiped= 0;
3236 mb_type= get_ue_golomb(&s->gb);
3237 if(h->slice_type == B_TYPE){
3239 partition_count= b_mb_type_info[mb_type].partition_count;
3240 mb_type= b_mb_type_info[mb_type].type;
3243 goto decode_intra_mb;
3245 }else if(h->slice_type == P_TYPE /*|| h->slice_type == SP_TYPE */){
3247 partition_count= p_mb_type_info[mb_type].partition_count;
3248 mb_type= p_mb_type_info[mb_type].type;
3251 goto decode_intra_mb;
3254 assert(h->slice_type == I_TYPE);
3257 av_log(h->s.avctx, AV_LOG_ERROR, "mb_type %d in %c slice to large at %d %d\n", mb_type, av_get_pict_type_char(h->slice_type), s->mb_x, s->mb_y);
3261 cbp= i_mb_type_info[mb_type].cbp;
3262 h->intra16x16_pred_mode= i_mb_type_info[mb_type].pred_mode;
3263 mb_type= i_mb_type_info[mb_type].type;
3266 if(h->mb_field_decoding_flag)
3267 mb_type |= MB_TYPE_INTERLACED;
3269 s->current_picture.mb_type[mb_xy]= mb_type;
3270 h->slice_table[ mb_xy ]= h->slice_num;
3272 if(IS_INTRA_PCM(mb_type)){
3276 // we assume these blocks are very rare so we dont optimize it
3277 align_get_bits(&s->gb);
3279 ptr= s->gb.buffer + get_bits_count(&s->gb);
3281 for(y=0; y<16; y++){
3282 const int index= 4*(y&3) + 64*(y>>2);
3283 for(x=0; x<16; x++){
3284 h->mb[index + (x&3) + 16*(x>>2)]= *(ptr++);
3288 const int index= 256 + 4*(y&3) + 32*(y>>2);
3290 h->mb[index + (x&3) + 16*(x>>2)]= *(ptr++);
3294 const int index= 256 + 64 + 4*(y&3) + 32*(y>>2);
3296 h->mb[index + (x&3) + 16*(x>>2)]= *(ptr++);
3300 skip_bits(&s->gb, 384); //FIXME check /fix the bitstream readers
3302 memset(h->non_zero_count[mb_xy], 16, 16+4+4);
3303 s->current_picture.qscale_table[mb_xy]= s->qscale;
3308 fill_caches(h, mb_type);
3311 if(IS_INTRA(mb_type)){
3312 // init_top_left_availability(h);
3313 if(IS_INTRA4x4(mb_type)){
3316 // fill_intra4x4_pred_table(h);
3317 for(i=0; i<16; i++){
3318 const int mode_coded= !get_bits1(&s->gb);
3319 const int predicted_mode= pred_intra_mode(h, i);
3323 const int rem_mode= get_bits(&s->gb, 3);
3324 if(rem_mode<predicted_mode)
3329 mode= predicted_mode;
3332 h->intra4x4_pred_mode_cache[ scan8[i] ] = mode;
3334 write_back_intra_pred_mode(h);
3335 if( check_intra4x4_pred_mode(h) < 0)
3338 h->intra16x16_pred_mode= check_intra_pred_mode(h, h->intra16x16_pred_mode);
3339 if(h->intra16x16_pred_mode < 0)
3342 h->chroma_pred_mode= get_ue_golomb(&s->gb);
3344 h->chroma_pred_mode= check_intra_pred_mode(h, h->chroma_pred_mode);
3345 if(h->chroma_pred_mode < 0)
3347 }else if(partition_count==4){
3348 int i, j, sub_partition_count[4], list, ref[2][4];
3350 if(h->slice_type == B_TYPE){
3352 h->sub_mb_type[i]= get_ue_golomb(&s->gb);
3353 if(h->sub_mb_type[i] >=13){
3354 av_log(h->s.avctx, AV_LOG_ERROR, "B sub_mb_type %d out of range at %d %d\n", h->sub_mb_type[i], s->mb_x, s->mb_y);
3357 sub_partition_count[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
3358 h->sub_mb_type[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].type;
3361 assert(h->slice_type == P_TYPE || h->slice_type == SP_TYPE); //FIXME SP correct ?
3363 h->sub_mb_type[i]= get_ue_golomb(&s->gb);
3364 if(h->sub_mb_type[i] >=4){
3365 av_log(h->s.avctx, AV_LOG_ERROR, "P sub_mb_type %d out of range at %d %d\n", h->sub_mb_type[i], s->mb_x, s->mb_y);
3368 sub_partition_count[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
3369 h->sub_mb_type[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].type;
3373 for(list=0; list<2; list++){
3374 const int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list];
3375 if(ref_count == 0) continue;
3377 if(IS_DIR(h->sub_mb_type[i], 0, list) && !IS_DIRECT(h->sub_mb_type[i])){
3378 ref[list][i] = get_te0_golomb(&s->gb, ref_count); //FIXME init to 0 before and skip?
3386 for(list=0; list<2; list++){
3387 const int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list];
3388 if(ref_count == 0) continue;
3391 h->ref_cache[list][ scan8[4*i] ]=h->ref_cache[list][ scan8[4*i]+1 ]=
3392 h->ref_cache[list][ scan8[4*i]+8 ]=h->ref_cache[list][ scan8[4*i]+9 ]= ref[list][i];
3394 if(IS_DIR(h->sub_mb_type[i], 0, list) && !IS_DIRECT(h->sub_mb_type[i])){
3395 const int sub_mb_type= h->sub_mb_type[i];
3396 const int block_width= (sub_mb_type & (MB_TYPE_16x16|MB_TYPE_16x8)) ? 2 : 1;
3397 for(j=0; j<sub_partition_count[i]; j++){
3399 const int index= 4*i + block_width*j;
3400 int16_t (* mv_cache)[2]= &h->mv_cache[list][ scan8[index] ];
3401 pred_motion(h, index, block_width, list, h->ref_cache[list][ scan8[index] ], &mx, &my);
3402 mx += get_se_golomb(&s->gb);
3403 my += get_se_golomb(&s->gb);
3404 tprintf("final mv:%d %d\n", mx, my);
3406 if(IS_SUB_8X8(sub_mb_type)){
3407 mv_cache[ 0 ][0]= mv_cache[ 1 ][0]=
3408 mv_cache[ 8 ][0]= mv_cache[ 9 ][0]= mx;
3409 mv_cache[ 0 ][1]= mv_cache[ 1 ][1]=
3410 mv_cache[ 8 ][1]= mv_cache[ 9 ][1]= my;
3411 }else if(IS_SUB_8X4(sub_mb_type)){
3412 mv_cache[ 0 ][0]= mv_cache[ 1 ][0]= mx;
3413 mv_cache[ 0 ][1]= mv_cache[ 1 ][1]= my;
3414 }else if(IS_SUB_4X8(sub_mb_type)){
3415 mv_cache[ 0 ][0]= mv_cache[ 8 ][0]= mx;
3416 mv_cache[ 0 ][1]= mv_cache[ 8 ][1]= my;
3418 assert(IS_SUB_4X4(sub_mb_type));
3419 mv_cache[ 0 ][0]= mx;
3420 mv_cache[ 0 ][1]= my;
3424 uint32_t *p= (uint32_t *)&h->mv_cache[list][ scan8[4*i] ][0];
3430 }else if(!IS_DIRECT(mb_type)){
3431 int list, mx, my, i;
3432 //FIXME we should set ref_idx_l? to 0 if we use that later ...
3433 if(IS_16X16(mb_type)){
3434 for(list=0; list<2; list++){
3435 if(h->ref_count[0]>0){
3436 if(IS_DIR(mb_type, 0, list)){
3437 const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
3438 fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, val, 1);
3442 for(list=0; list<2; list++){
3443 if(IS_DIR(mb_type, 0, list)){
3444 pred_motion(h, 0, 4, list, h->ref_cache[list][ scan8[0] ], &mx, &my);
3445 mx += get_se_golomb(&s->gb);
3446 my += get_se_golomb(&s->gb);
3447 tprintf("final mv:%d %d\n", mx, my);
3449 fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, pack16to32(mx,my), 4);
3453 else if(IS_16X8(mb_type)){
3454 for(list=0; list<2; list++){
3455 if(h->ref_count[list]>0){
3457 if(IS_DIR(mb_type, i, list)){
3458 const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
3459 fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, val, 1);
3464 for(list=0; list<2; list++){
3466 if(IS_DIR(mb_type, i, list)){
3467 pred_16x8_motion(h, 8*i, list, h->ref_cache[list][scan8[0] + 16*i], &mx, &my);
3468 mx += get_se_golomb(&s->gb);
3469 my += get_se_golomb(&s->gb);
3470 tprintf("final mv:%d %d\n", mx, my);
3472 fill_rectangle(h->mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, pack16to32(mx,my), 4);
3477 assert(IS_8X16(mb_type));
3478 for(list=0; list<2; list++){
3479 if(h->ref_count[list]>0){
3481 if(IS_DIR(mb_type, i, list)){ //FIXME optimize
3482 const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
3483 fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, val, 1);
3488 for(list=0; list<2; list++){
3490 if(IS_DIR(mb_type, i, list)){
3491 pred_8x16_motion(h, i*4, list, h->ref_cache[list][ scan8[0] + 2*i ], &mx, &my);
3492 mx += get_se_golomb(&s->gb);
3493 my += get_se_golomb(&s->gb);
3494 tprintf("final mv:%d %d\n", mx, my);
3496 fill_rectangle(h->mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, pack16to32(mx,my), 4);
3503 if(IS_INTER(mb_type))
3504 write_back_motion(h, mb_type);
3506 if(!IS_INTRA16x16(mb_type)){
3507 cbp= get_ue_golomb(&s->gb);
3509 av_log(h->s.avctx, AV_LOG_ERROR, "cbp too large (%d) at %d %d\n", cbp, s->mb_x, s->mb_y);
3513 if(IS_INTRA4x4(mb_type))
3514 cbp= golomb_to_intra4x4_cbp[cbp];
3516 cbp= golomb_to_inter_cbp[cbp];
3519 if(cbp || IS_INTRA16x16(mb_type)){
3520 int i8x8, i4x4, chroma_idx;
3521 int chroma_qp, dquant;
3522 GetBitContext *gb= IS_INTRA(mb_type) ? h->intra_gb_ptr : h->inter_gb_ptr;
3523 const uint8_t *scan, *dc_scan;
3525 // fill_non_zero_count_cache(h);
3527 if(IS_INTERLACED(mb_type)){
3529 dc_scan= luma_dc_field_scan;
3532 dc_scan= luma_dc_zigzag_scan;
3535 dquant= get_se_golomb(&s->gb);
3537 if( dquant > 25 || dquant < -26 ){
3538 av_log(h->s.avctx, AV_LOG_ERROR, "dquant out of range (%d) at %d %d\n", dquant, s->mb_x, s->mb_y);
3542 s->qscale += dquant;
3543 if(((unsigned)s->qscale) > 51){
3544 if(s->qscale<0) s->qscale+= 52;
3545 else s->qscale-= 52;
3548 h->chroma_qp= chroma_qp= get_chroma_qp(h, s->qscale);
3549 if(IS_INTRA16x16(mb_type)){
3550 if( decode_residual(h, h->intra_gb_ptr, h->mb, LUMA_DC_BLOCK_INDEX, dc_scan, s->qscale, 16) < 0){
3551 return -1; //FIXME continue if partotioned and other retirn -1 too
3554 assert((cbp&15) == 0 || (cbp&15) == 15);
3557 for(i8x8=0; i8x8<4; i8x8++){
3558 for(i4x4=0; i4x4<4; i4x4++){
3559 const int index= i4x4 + 4*i8x8;
3560 if( decode_residual(h, h->intra_gb_ptr, h->mb + 16*index, index, scan + 1, s->qscale, 15) < 0 ){
3566 fill_rectangle(&h->non_zero_count_cache[scan8[0]], 4, 4, 8, 0, 1);
3569 for(i8x8=0; i8x8<4; i8x8++){
3570 if(cbp & (1<<i8x8)){
3571 for(i4x4=0; i4x4<4; i4x4++){
3572 const int index= i4x4 + 4*i8x8;
3574 if( decode_residual(h, gb, h->mb + 16*index, index, scan, s->qscale, 16) <0 ){
3579 uint8_t * const nnz= &h->non_zero_count_cache[ scan8[4*i8x8] ];
3580 nnz[0] = nnz[1] = nnz[8] = nnz[9] = 0;
3586 for(chroma_idx=0; chroma_idx<2; chroma_idx++)
3587 if( decode_residual(h, gb, h->mb + 256 + 16*4*chroma_idx, CHROMA_DC_BLOCK_INDEX, chroma_dc_scan, chroma_qp, 4) < 0){
3593 for(chroma_idx=0; chroma_idx<2; chroma_idx++){
3594 for(i4x4=0; i4x4<4; i4x4++){
3595 const int index= 16 + 4*chroma_idx + i4x4;
3596 if( decode_residual(h, gb, h->mb + 16*index, index, scan + 1, chroma_qp, 15) < 0){
3602 uint8_t * const nnz= &h->non_zero_count_cache[0];
3603 nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =
3604 nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
3607 memset(&h->non_zero_count_cache[8], 0, 8*5);
3609 s->current_picture.qscale_table[mb_xy]= s->qscale;
3610 write_back_non_zero_count(h);
3615 static void filter_mb_edgev( H264Context *h, uint8_t *pix, int stride, int bS[4], int qp ) {
3617 const int index_a = clip( qp + h->slice_alpha_c0_offset, 0, 51 );
3618 const int alpha = alpha_table[index_a];
3619 const int beta = beta_table[clip( qp + h->slice_beta_offset, 0, 51 )];
3621 for( i = 0; i < 4; i++ ) {
3627 /* 4px edge length */
3628 for( d = 0; d < 4; d++ ) {
3629 const uint8_t p0 = pix[-1];
3630 const uint8_t p1 = pix[-2];
3631 const uint8_t p2 = pix[-3];
3633 const uint8_t q0 = pix[0];
3634 const uint8_t q1 = pix[1];
3635 const uint8_t q2 = pix[2];
3637 if( abs( p0 - q0 ) >= alpha ||
3638 abs( p1 - p0 ) >= beta ||
3639 abs( q1 - q0 ) >= beta ) {
3645 const int tc0 = tc0_table[index_a][bS[i] - 1];
3649 if( abs( p2 - p0 ) < beta ) {
3650 pix[-2] = p1 + clip( ( p2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( p1 << 1 ) ) >> 1, -tc0, tc0 );
3653 if( abs( q2 - q0 ) < beta ) {
3654 pix[1] = q1 + clip( ( q2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( q1 << 1 ) ) >> 1, -tc0, tc0 );
3658 i_delta = clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );
3659 pix[-1] = clip( p0 + i_delta, 0, 255 ); /* p0' */
3660 pix[0] = clip( q0 - i_delta, 0, 255 ); /* q0' */
3664 const int c = abs( p0 - q0 ) < (( alpha >> 2 ) + 2 );
3666 if( abs( p2 - p0 ) < beta && c )
3668 const uint8_t p3 = pix[-4];
3670 pix[-1] = ( p2 + 2*p1 + 2*p0 + 2*q0 + q1 + 4 ) >> 3;
3671 pix[-2] = ( p2 + p1 + p0 + q0 + 2 ) >> 2;
3672 pix[-3] = ( 2*p3 + 3*p2 + p1 + p0 + q0 + 4 ) >> 3;
3675 pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
3677 if( abs( q2 - q0 ) < beta && c )
3679 const uint8_t q3 = pix[3];
3681 pix[0] = ( p1 + 2*p0 + 2*q0 + 2*q1 + q2 + 4 ) >> 3;
3682 pix[1] = ( p0 + q0 + q1 + q2 + 2 ) >> 2;
3683 pix[2] = ( 2*q3 + 3*q2 + q1 + q0 + p0 + 4 ) >> 3;
3686 pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
3693 static void filter_mb_edgecv( H264Context *h, uint8_t *pix, int stride, int bS[4], int qp ) {
3695 const int index_a = clip( qp + h->slice_alpha_c0_offset, 0, 51 );
3696 const int alpha = alpha_table[index_a];
3697 const int beta = beta_table[clip( qp + h->slice_beta_offset, 0, 51 )];
3699 for( i = 0; i < 4; i++ ) {
3705 /* 2px edge length (because we use same bS than the one for luma) */
3706 for( d = 0; d < 2; d++ )
3708 const uint8_t p0 = pix[-1];
3709 const uint8_t p1 = pix[-2];
3710 const uint8_t q0 = pix[0];
3711 const uint8_t q1 = pix[1];
3713 if( abs( p0 - q0 ) >= alpha ||
3714 abs( p1 - p0 ) >= beta ||
3715 abs( q1 - q0 ) >= beta ) {
3721 const int tc = tc0_table[index_a][bS[i] - 1] + 1;
3722 const int i_delta = clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );
3724 pix[-1] = clip( p0 + i_delta, 0, 255 ); /* p0' */
3725 pix[0] = clip( q0 - i_delta, 0, 255 ); /* q0' */
3727 pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2; /* p0' */
3728 pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2; /* q0' */
3735 static void filter_mb_edgeh( H264Context *h, uint8_t *pix, int stride, int bS[4], int qp ) {
3737 const int index_a = clip( qp + h->slice_alpha_c0_offset, 0, 51 );
3738 const int alpha = alpha_table[index_a];
3739 const int beta = beta_table[clip( qp + h->slice_beta_offset, 0, 51 )];
3740 const int pix_next = stride;
3742 for( i = 0; i < 4; i++ ) {
3748 /* 4px edge length */
3749 for( d = 0; d < 4; d++ ) {
3750 const uint8_t p0 = pix[-1*pix_next];
3751 const uint8_t p1 = pix[-2*pix_next];
3752 const uint8_t p2 = pix[-3*pix_next];
3753 const uint8_t q0 = pix[0];
3754 const uint8_t q1 = pix[1*pix_next];
3755 const uint8_t q2 = pix[2*pix_next];
3757 if( abs( p0 - q0 ) >= alpha ||
3758 abs( p1 - p0 ) >= beta ||
3759 abs( q1 - q0 ) >= beta ) {
3765 const int tc0 = tc0_table[index_a][bS[i] - 1];
3769 if( abs( p2 - p0 ) < beta ) {
3770 pix[-2*pix_next] = p1 + clip( ( p2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( p1 << 1 ) ) >> 1, -tc0, tc0 );
3773 if( abs( q2 - q0 ) < beta ) {
3774 pix[pix_next] = q1 + clip( ( q2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( q1 << 1 ) ) >> 1, -tc0, tc0 );
3778 i_delta = clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );
3779 pix[-pix_next] = clip( p0 + i_delta, 0, 255 ); /* p0' */
3780 pix[0] = clip( q0 - i_delta, 0, 255 ); /* q0' */
3784 const uint8_t p3 = pix[-4*pix_next];
3785 const uint8_t q3 = pix[ 3*pix_next];
3786 const int c = abs( p0 - q0 ) < (( alpha >> 2 ) + 2 );
3788 if( abs( p2 - p0 ) < beta && c ) {
3790 pix[-1*pix_next] = ( p2 + 2*p1 + 2*p0 + 2*q0 + q1 + 4 ) >> 3;
3791 pix[-2*pix_next] = ( p2 + p1 + p0 + q0 + 2 ) >> 2;
3792 pix[-3*pix_next] = ( 2*p3 + 3*p2 + p1 + p0 + q0 + 4 ) >> 3;
3795 pix[-1*pix_next] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
3797 if( abs( q2 - q0 ) < beta && c ) {
3799 pix[0*pix_next] = ( p1 + 2*p0 + 2*q0 + 2*q1 + q2 + 4 ) >> 3;
3800 pix[1*pix_next] = ( p0 + q0 + q1 + q2 + 2 ) >> 2;
3801 pix[2*pix_next] = ( 2*q3 + 3*q2 + q1 + q0 + p0 + 4 ) >> 3;
3804 pix[0*pix_next] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
3812 static void filter_mb_edgech( H264Context *h, uint8_t *pix, int stride, int bS[4], int qp ) {
3814 const int index_a = clip( qp + h->slice_alpha_c0_offset, 0, 51 );
3815 const int alpha = alpha_table[index_a];
3816 const int beta = beta_table[clip( qp + h->slice_beta_offset, 0, 51 )];
3817 const int pix_next = stride;
3819 for( i = 0; i < 4; i++ )
3826 /* 2px edge length (see deblocking_filter_edgecv) */
3827 for( d = 0; d < 2; d++ ) {
3828 const uint8_t p0 = pix[-1*pix_next];
3829 const uint8_t p1 = pix[-2*pix_next];
3830 const uint8_t q0 = pix[0];
3831 const uint8_t q1 = pix[1*pix_next];
3833 if( abs( p0 - q0 ) >= alpha ||
3834 abs( p1 - p0 ) >= beta ||
3835 abs( q1 - q0 ) >= beta ) {
3841 int tc = tc0_table[index_a][bS[i] - 1] + 1;
3842 int i_delta = clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );
3844 pix[-pix_next] = clip( p0 + i_delta, 0, 255 ); /* p0' */
3845 pix[0] = clip( q0 - i_delta, 0, 255 ); /* q0' */
3849 pix[-pix_next] = ( 2*p1 + p0 + q1 + 2 ) >> 2; /* p0' */
3850 pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2; /* q0' */
3857 static void filter_mb( H264Context *h, int mb_x, int mb_y ) {
3858 MpegEncContext * const s = &h->s;
3859 const int mb_xy= mb_x + mb_y*s->mb_stride;
3860 uint8_t *img_y = s->current_picture.data[0] + (mb_y * 16* s->linesize ) + mb_x * 16;
3861 uint8_t *img_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
3862 uint8_t *img_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
3863 int linesize, uvlinesize;
3866 /* FIXME what's that ? */
3871 /* FIXME Implement deblocking filter for field MB */
3872 if( h->sps.mb_aff ) {
3875 linesize = s->linesize;
3876 uvlinesize = s->uvlinesize;
3878 /* dir : 0 -> vertical edge, 1 -> horizontal edge */
3879 for( dir = 0; dir < 2; dir++ )
3884 /* test picture boundary */
3885 if( ( dir == 0 && mb_x == 0 ) || ( dir == 1 && mb_y == 0 ) ) {
3888 /* FIXME test slice boundary */
3889 if( h->disable_deblocking_filter_idc == 2 ) {
3893 for( edge = start; edge < 4; edge++ ) {
3894 /* mbn_xy: neighbour macroblock (how that works for field ?) */
3895 int mbn_xy = edge > 0 ? mb_xy : ( dir == 0 ? mb_xy -1 : mb_xy - s->mb_stride );
3899 if( IS_INTRA( s->current_picture.mb_type[mb_xy] ) ||
3900 IS_INTRA( s->current_picture.mb_type[mbn_xy] ) ) {
3901 bS[0] = bS[1] = bS[2] = bS[3] = ( edge == 0 ? 4 : 3 );
3904 for( i = 0; i < 4; i++ ) {
3905 static const uint8_t block_idx_xy[4][4] = {
3906 { 0, 2, 8, 10}, { 1, 3, 9, 11},
3907 { 4, 6, 12, 14}, { 5, 7, 13, 15}
3910 int x = dir == 0 ? edge : i;
3911 int y = dir == 0 ? i : edge;
3912 int xn = (x - (dir == 0 ? 1 : 0 ))&0x03;
3913 int yn = (y - (dir == 0 ? 0 : 1 ))&0x03;
3915 if( h->non_zero_count[mb_xy][block_idx_xy[x][y]] != 0 ||
3916 h->non_zero_count[mbn_xy][block_idx_xy[xn][yn]] != 0 ) {
3919 else if( h->slice_type == P_TYPE ) {
3920 const int b8_xy = h->mb2b8_xy[mb_xy]+(y/2)*h->b8_stride+(x/2);
3921 const int b8n_xy= h->mb2b8_xy[mbn_xy]+(yn/2)*h->b8_stride+(xn/2);
3922 const int b_xy = h->mb2b_xy[mb_xy]+y*h->b_stride+x;
3923 const int bn_xy = h->mb2b_xy[mbn_xy]+yn*h->b_stride+xn;
3924 if( s->current_picture.ref_index[0][b8_xy] != s->current_picture.ref_index[0][b8n_xy] ||
3925 abs( s->current_picture.motion_val[0][b_xy][0] - s->current_picture.motion_val[0][bn_xy][0] ) >= 4 ||
3926 abs( s->current_picture.motion_val[0][b_xy][1] - s->current_picture.motion_val[0][bn_xy][1] ) >= 4 )
3932 /* FIXME Add support for B frame */
3939 qp = ( s->current_picture.qscale_table[mb_xy] + s->current_picture.qscale_table[mbn_xy] + 1 ) >> 1;
3941 filter_mb_edgev( h, &img_y[4*edge], linesize, bS, qp );
3943 int chroma_qp = ( get_chroma_qp( h, s->current_picture.qscale_table[mb_xy] ) +
3944 get_chroma_qp( h, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1;
3945 filter_mb_edgecv( h, &img_cb[2*edge], uvlinesize, bS, chroma_qp );
3946 filter_mb_edgecv( h, &img_cr[2*edge], uvlinesize, bS, chroma_qp );
3949 filter_mb_edgeh( h, &img_y[4*edge*linesize], linesize, bS, qp );
3951 int chroma_qp = ( get_chroma_qp( h, s->current_picture.qscale_table[mb_xy] ) +
3952 get_chroma_qp( h, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1;
3953 filter_mb_edgech( h, &img_cb[2*edge*uvlinesize], uvlinesize, bS, chroma_qp );
3954 filter_mb_edgech( h, &img_cr[2*edge*uvlinesize], uvlinesize, bS, chroma_qp );
3961 static int decode_slice(H264Context *h){
3962 MpegEncContext * const s = &h->s;
3963 const int part_mask= s->partitioned_frame ? (AC_END|AC_ERROR) : 0x7F;
3969 int ret= decode_mb(h);
3973 if(ret>=0 && h->sps.mb_aff){ //FIXME optimal? or let mb_decode decode 16x32 ?
3982 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
3983 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);
3988 if(++s->mb_x >= s->mb_width){
3990 ff_draw_horiz_band(s, 16*s->mb_y, 16);
3991 if(++s->mb_y >= s->mb_height){
3992 tprintf("slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
3994 if(get_bits_count(&s->gb) == s->gb.size_in_bits){
3995 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);
3999 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);
4006 if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->mb_skip_run<=0){
4007 if(get_bits_count(&s->gb) == s->gb.size_in_bits){
4008 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);
4012 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);
4020 for(;s->mb_y < s->mb_height; s->mb_y++){
4021 for(;s->mb_x < s->mb_width; s->mb_x++){
4022 int ret= decode_mb(h);
4027 fprintf(stderr, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
4028 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);
4033 if(++s->mb_x >= s->mb_width){
4035 if(++s->mb_y >= s->mb_height){
4036 if(get_bits_count(s->gb) == s->gb.size_in_bits){
4037 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);
4041 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);
4048 if(get_bits_count(s->?gb) >= s->gb?.size_in_bits){
4049 if(get_bits_count(s->gb) == s->gb.size_in_bits){
4050 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);
4054 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);
4061 ff_draw_horiz_band(s, 16*s->mb_y, 16);
4064 return -1; //not reached
4067 static void filter_frame(H264Context *h) {
4071 for( mb_y = 0; mb_y < h->s.mb_height; mb_y++ ) {
4072 for( mb_x = 0; mb_x < h->s.mb_width; mb_x++ ) {
4073 filter_mb( h, mb_x, mb_y );
4078 static inline int decode_vui_parameters(H264Context *h, SPS *sps){
4079 MpegEncContext * const s = &h->s;
4080 int aspect_ratio_info_present_flag, aspect_ratio_idc;
4082 aspect_ratio_info_present_flag= get_bits1(&s->gb);
4084 if( aspect_ratio_info_present_flag ) {
4085 aspect_ratio_idc= get_bits(&s->gb, 8);
4086 if( aspect_ratio_idc == EXTENDED_SAR ) {
4087 sps->sar.num= get_bits(&s->gb, 16);
4088 sps->sar.den= get_bits(&s->gb, 16);
4089 }else if(aspect_ratio_idc < 16){
4090 sps->sar= pixel_aspect[aspect_ratio_idc];
4092 av_log(h->s.avctx, AV_LOG_ERROR, "illegal aspect ratio\n");
4099 // s->avctx->aspect_ratio= sar_width*s->width / (float)(s->height*sar_height);
4101 | overscan_info_present_flag |0 |u(1) |
4102 | if( overscan_info_present_flag ) | | |
4103 | overscan_appropriate_flag |0 |u(1) |
4104 | video_signal_type_present_flag |0 |u(1) |
4105 | if( video_signal_type_present_flag ) { | | |
4106 | video_format |0 |u(3) |
4107 | video_full_range_flag |0 |u(1) |
4108 | colour_description_present_flag |0 |u(1) |
4109 | if( colour_description_present_flag ) { | | |
4110 | colour_primaries |0 |u(8) |
4111 | transfer_characteristics |0 |u(8) |
4112 | matrix_coefficients |0 |u(8) |
4115 | chroma_location_info_present_flag |0 |u(1) |
4116 | if ( chroma_location_info_present_flag ) { | | |
4117 | chroma_sample_location_type_top_field |0 |ue(v) |
4118 | chroma_sample_location_type_bottom_field |0 |ue(v) |
4120 | timing_info_present_flag |0 |u(1) |
4121 | if( timing_info_present_flag ) { | | |
4122 | num_units_in_tick |0 |u(32) |
4123 | time_scale |0 |u(32) |
4124 | fixed_frame_rate_flag |0 |u(1) |
4126 | nal_hrd_parameters_present_flag |0 |u(1) |
4127 | if( nal_hrd_parameters_present_flag = = 1) | | |
4128 | hrd_parameters( ) | | |
4129 | vcl_hrd_parameters_present_flag |0 |u(1) |
4130 | if( vcl_hrd_parameters_present_flag = = 1) | | |
4131 | hrd_parameters( ) | | |
4132 | if( ( nal_hrd_parameters_present_flag = = 1 | || | |
4134 |( vcl_hrd_parameters_present_flag = = 1 ) ) | | |
4135 | low_delay_hrd_flag |0 |u(1) |
4136 | bitstream_restriction_flag |0 |u(1) |
4137 | if( bitstream_restriction_flag ) { |0 |u(1) |
4138 | motion_vectors_over_pic_boundaries_flag |0 |u(1) |
4139 | max_bytes_per_pic_denom |0 |ue(v) |
4140 | max_bits_per_mb_denom |0 |ue(v) |
4141 | log2_max_mv_length_horizontal |0 |ue(v) |
4142 | log2_max_mv_length_vertical |0 |ue(v) |
4143 | num_reorder_frames |0 |ue(v) |
4144 | max_dec_frame_buffering |0 |ue(v) |
4151 static inline int decode_seq_parameter_set(H264Context *h){
4152 MpegEncContext * const s = &h->s;
4153 int profile_idc, level_idc;
4157 profile_idc= get_bits(&s->gb, 8);
4158 get_bits1(&s->gb); //constraint_set0_flag
4159 get_bits1(&s->gb); //constraint_set1_flag
4160 get_bits1(&s->gb); //constraint_set2_flag
4161 get_bits(&s->gb, 5); // reserved
4162 level_idc= get_bits(&s->gb, 8);
4163 sps_id= get_ue_golomb(&s->gb);
4165 sps= &h->sps_buffer[ sps_id ];
4166 sps->profile_idc= profile_idc;
4167 sps->level_idc= level_idc;
4169 sps->log2_max_frame_num= get_ue_golomb(&s->gb) + 4;
4170 sps->poc_type= get_ue_golomb(&s->gb);
4172 if(sps->poc_type == 0){ //FIXME #define
4173 sps->log2_max_poc_lsb= get_ue_golomb(&s->gb) + 4;
4174 } else if(sps->poc_type == 1){//FIXME #define
4175 sps->delta_pic_order_always_zero_flag= get_bits1(&s->gb);
4176 sps->offset_for_non_ref_pic= get_se_golomb(&s->gb);
4177 sps->offset_for_top_to_bottom_field= get_se_golomb(&s->gb);
4178 sps->poc_cycle_length= get_ue_golomb(&s->gb);
4180 for(i=0; i<sps->poc_cycle_length; i++)
4181 sps->offset_for_ref_frame[i]= get_se_golomb(&s->gb);
4183 if(sps->poc_type > 2){
4184 av_log(h->s.avctx, AV_LOG_ERROR, "illegal POC type %d\n", sps->poc_type);
4188 sps->ref_frame_count= get_ue_golomb(&s->gb);
4189 sps->gaps_in_frame_num_allowed_flag= get_bits1(&s->gb);
4190 sps->mb_width= get_ue_golomb(&s->gb) + 1;
4191 sps->mb_height= get_ue_golomb(&s->gb) + 1;
4192 sps->frame_mbs_only_flag= get_bits1(&s->gb);
4193 if(!sps->frame_mbs_only_flag)
4194 sps->mb_aff= get_bits1(&s->gb);
4198 sps->direct_8x8_inference_flag= get_bits1(&s->gb);
4200 sps->crop= get_bits1(&s->gb);
4202 sps->crop_left = get_ue_golomb(&s->gb);
4203 sps->crop_right = get_ue_golomb(&s->gb);
4204 sps->crop_top = get_ue_golomb(&s->gb);
4205 sps->crop_bottom= get_ue_golomb(&s->gb);
4206 if(sps->crop_left || sps->crop_top){
4207 av_log(h->s.avctx, AV_LOG_ERROR, "insane cropping not completly supported, this could look slightly wrong ...\n");
4213 sps->crop_bottom= 0;
4216 sps->vui_parameters_present_flag= get_bits1(&s->gb);
4217 if( sps->vui_parameters_present_flag )
4218 decode_vui_parameters(h, sps);
4220 if(s->avctx->debug&FF_DEBUG_PICT_INFO){
4221 av_log(h->s.avctx, AV_LOG_DEBUG, "sps:%d profile:%d/%d poc:%d ref:%d %dx%d %s %s crop:%d/%d/%d/%d %s\n",
4222 sps_id, sps->profile_idc, sps->level_idc,
4224 sps->ref_frame_count,
4225 sps->mb_width, sps->mb_height,
4226 sps->frame_mbs_only_flag ? "FRM" : (sps->mb_aff ? "MB-AFF" : "PIC-AFF"),
4227 sps->direct_8x8_inference_flag ? "8B8" : "",
4228 sps->crop_left, sps->crop_right,
4229 sps->crop_top, sps->crop_bottom,
4230 sps->vui_parameters_present_flag ? "VUI" : ""
4236 static inline int decode_picture_parameter_set(H264Context *h){
4237 MpegEncContext * const s = &h->s;
4238 int pps_id= get_ue_golomb(&s->gb);
4239 PPS *pps= &h->pps_buffer[pps_id];
4241 pps->sps_id= get_ue_golomb(&s->gb);
4242 pps->cabac= get_bits1(&s->gb);
4243 pps->pic_order_present= get_bits1(&s->gb);
4244 pps->slice_group_count= get_ue_golomb(&s->gb) + 1;
4245 if(pps->slice_group_count > 1 ){
4246 pps->mb_slice_group_map_type= get_ue_golomb(&s->gb);
4247 av_log(h->s.avctx, AV_LOG_ERROR, "FMO not supported\n");
4248 switch(pps->mb_slice_group_map_type){
4251 | for( i = 0; i <= num_slice_groups_minus1; i++ ) | | |
4252 | run_length[ i ] |1 |ue(v) |
4257 | for( i = 0; i < num_slice_groups_minus1; i++ ) | | |
4259 | top_left_mb[ i ] |1 |ue(v) |
4260 | bottom_right_mb[ i ] |1 |ue(v) |
4268 | slice_group_change_direction_flag |1 |u(1) |
4269 | slice_group_change_rate_minus1 |1 |ue(v) |
4274 | slice_group_id_cnt_minus1 |1 |ue(v) |
4275 | for( i = 0; i <= slice_group_id_cnt_minus1; i++ | | |
4277 | slice_group_id[ i ] |1 |u(v) |
4282 pps->ref_count[0]= get_ue_golomb(&s->gb) + 1;
4283 pps->ref_count[1]= get_ue_golomb(&s->gb) + 1;
4284 if(pps->ref_count[0] > 32 || pps->ref_count[1] > 32){
4285 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow (pps)\n");
4289 pps->weighted_pred= get_bits1(&s->gb);
4290 pps->weighted_bipred_idc= get_bits(&s->gb, 2);
4291 pps->init_qp= get_se_golomb(&s->gb) + 26;
4292 pps->init_qs= get_se_golomb(&s->gb) + 26;
4293 pps->chroma_qp_index_offset= get_se_golomb(&s->gb);
4294 pps->deblocking_filter_parameters_present= get_bits1(&s->gb);
4295 pps->constrained_intra_pred= get_bits1(&s->gb);
4296 pps->redundant_pic_cnt_present = get_bits1(&s->gb);
4298 if(s->avctx->debug&FF_DEBUG_PICT_INFO){
4299 av_log(h->s.avctx, AV_LOG_DEBUG, "pps:%d sps:%d %s slice_groups:%d ref:%d/%d %s qp:%d/%d/%d %s %s %s\n",
4300 pps_id, pps->sps_id,
4301 pps->cabac ? "CABAC" : "CAVLC",
4302 pps->slice_group_count,
4303 pps->ref_count[0], pps->ref_count[1],
4304 pps->weighted_pred ? "weighted" : "",
4305 pps->init_qp, pps->init_qs, pps->chroma_qp_index_offset,
4306 pps->deblocking_filter_parameters_present ? "LPAR" : "",
4307 pps->constrained_intra_pred ? "CONSTR" : "",
4308 pps->redundant_pic_cnt_present ? "REDU" : ""
4316 * finds the end of the current frame in the bitstream.
4317 * @return the position of the first byte of the next frame, or -1
4319 static int find_frame_end(MpegEncContext *s, uint8_t *buf, int buf_size){
4320 ParseContext *pc= &s->parse_context;
4323 //printf("first %02X%02X%02X%02X\n", buf[0], buf[1],buf[2],buf[3]);
4324 // mb_addr= pc->mb_addr - 1;
4326 //FIXME this will fail with slices
4327 for(i=0; i<buf_size; i++){
4328 state= (state<<8) | buf[i];
4329 if((state&0xFFFFFF1F) == 0x101 || (state&0xFFFFFF1F) == 0x102 || (state&0xFFFFFF1F) == 0x105){
4330 if(pc->frame_start_found){
4332 pc->frame_start_found= 0;
4335 pc->frame_start_found= 1;
4340 return END_NOT_FOUND;
4343 static int decode_nal_units(H264Context *h, uint8_t *buf, int buf_size){
4344 MpegEncContext * const s = &h->s;
4345 AVCodecContext * const avctx= s->avctx;
4349 for(i=0; i<32; i++){
4350 printf("%X ", buf[i]);
4359 // start code prefix search
4360 for(; buf_index + 3 < buf_size; buf_index++){
4361 // this should allways succeed in the first iteration
4362 if(buf[buf_index] == 0 && buf[buf_index+1] == 0 && buf[buf_index+2] == 1)
4366 if(buf_index+3 >= buf_size) break;
4370 ptr= decode_nal(h, buf + buf_index, &dst_length, &consumed, buf_size - buf_index);
4371 if(ptr[dst_length - 1] == 0) dst_length--;
4372 bit_length= 8*dst_length - decode_rbsp_trailing(ptr + dst_length - 1);
4374 if(s->avctx->debug&FF_DEBUG_STARTCODE){
4375 av_log(h->s.avctx, AV_LOG_DEBUG, "NAL %d at %d length %d\n", h->nal_unit_type, buf_index, dst_length);
4378 buf_index += consumed;
4380 if(h->nal_ref_idc < s->hurry_up)
4383 switch(h->nal_unit_type){
4385 idr(h); //FIXME ensure we dont loose some frames if there is reordering
4387 init_get_bits(&s->gb, ptr, bit_length);
4389 h->inter_gb_ptr= &s->gb;
4390 s->data_partitioning = 0;
4392 if(decode_slice_header(h) < 0) return -1;
4393 if(h->redundant_pic_count==0)
4397 init_get_bits(&s->gb, ptr, bit_length);
4399 h->inter_gb_ptr= NULL;
4400 s->data_partitioning = 1;
4402 if(decode_slice_header(h) < 0) return -1;
4405 init_get_bits(&h->intra_gb, ptr, bit_length);
4406 h->intra_gb_ptr= &h->intra_gb;
4409 init_get_bits(&h->inter_gb, ptr, bit_length);
4410 h->inter_gb_ptr= &h->inter_gb;
4412 if(h->redundant_pic_count==0 && h->intra_gb_ptr && s->data_partitioning)
4418 init_get_bits(&s->gb, ptr, bit_length);
4419 decode_seq_parameter_set(h);
4421 if(s->flags& CODEC_FLAG_LOW_DELAY)
4424 avctx->has_b_frames= !s->low_delay;
4427 init_get_bits(&s->gb, ptr, bit_length);
4429 decode_picture_parameter_set(h);
4432 case NAL_PICTURE_DELIMITER:
4434 case NAL_FILTER_DATA:
4438 //FIXME move after where irt is set
4439 s->current_picture.pict_type= s->pict_type;
4440 s->current_picture.key_frame= s->pict_type == I_TYPE;
4443 if(!s->current_picture_ptr) return buf_index; //no frame
4445 h->prev_frame_num_offset= h->frame_num_offset;
4446 h->prev_frame_num= h->frame_num;
4447 if(s->current_picture_ptr->reference){
4448 h->prev_poc_msb= h->poc_msb;
4449 h->prev_poc_lsb= h->poc_lsb;
4451 if(s->current_picture_ptr->reference)
4452 execute_ref_pic_marking(h, h->mmco, h->mmco_index);
4454 assert(h->mmco_index==0);
4458 if( h->disable_deblocking_filter_idc != 1 ) {
4468 * retunrs the number of bytes consumed for building the current frame
4470 static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size){
4471 if(s->flags&CODEC_FLAG_TRUNCATED){
4472 pos -= s->parse_context.last_index;
4473 if(pos<0) pos=0; // FIXME remove (uneeded?)
4477 if(pos==0) pos=1; //avoid infinite loops (i doubt thats needed but ...)
4478 if(pos+10>buf_size) pos=buf_size; // oops ;)
4484 static int decode_frame(AVCodecContext *avctx,
4485 void *data, int *data_size,
4486 uint8_t *buf, int buf_size)
4488 H264Context *h = avctx->priv_data;
4489 MpegEncContext *s = &h->s;
4490 AVFrame *pict = data;
4493 s->flags= avctx->flags;
4494 s->flags2= avctx->flags2;
4498 /* no supplementary picture */
4499 if (buf_size == 0) {
4503 if(s->flags&CODEC_FLAG_TRUNCATED){
4504 int next= find_frame_end(s, buf, buf_size);
4506 if( ff_combine_frame(s, next, &buf, &buf_size) < 0 )
4508 //printf("next:%d buf_size:%d last_index:%d\n", next, buf_size, s->parse_context.last_index);
4511 if(s->avctx->extradata_size && s->picture_number==0){
4512 if(0 < decode_nal_units(h, s->avctx->extradata, s->avctx->extradata_size) )
4516 buf_index=decode_nal_units(h, buf, buf_size);
4520 //FIXME do something with unavailable reference frames
4522 // if(ret==FRAME_SKIPED) return get_consumed_bytes(s, buf_index, buf_size);
4524 if(s->pict_type==B_TYPE || s->low_delay){
4525 *pict= *(AVFrame*)&s->current_picture;
4527 *pict= *(AVFrame*)&s->last_picture;
4530 if(!s->current_picture_ptr){
4531 av_log(h->s.avctx, AV_LOG_DEBUG, "error, NO frame\n");
4535 *pict= *(AVFrame*)&s->current_picture; //FIXME
4536 ff_print_debug_info(s, pict);
4537 assert(pict->data[0]);
4538 //printf("out %d\n", (int)pict->data[0]);
4541 /* Return the Picture timestamp as the frame number */
4542 /* we substract 1 because it is added on utils.c */
4543 avctx->frame_number = s->picture_number - 1;
4546 /* dont output the last pic after seeking */
4547 if(s->last_picture_ptr || s->low_delay)
4548 //Note this isnt a issue as a IDR pic should flush teh buffers
4550 *data_size = sizeof(AVFrame);
4551 return get_consumed_bytes(s, buf_index, buf_size);
4554 static inline void fill_mb_avail(H264Context *h){
4555 MpegEncContext * const s = &h->s;
4556 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
4559 h->mb_avail[0]= s->mb_x && h->slice_table[mb_xy - s->mb_stride - 1] == h->slice_num;
4560 h->mb_avail[1]= h->slice_table[mb_xy - s->mb_stride ] == h->slice_num;
4561 h->mb_avail[2]= s->mb_x+1 < s->mb_width && h->slice_table[mb_xy - s->mb_stride + 1] == h->slice_num;
4567 h->mb_avail[3]= s->mb_x && h->slice_table[mb_xy - 1] == h->slice_num;
4568 h->mb_avail[4]= 1; //FIXME move out
4569 h->mb_avail[5]= 0; //FIXME move out
4575 #define SIZE (COUNT*40)
4581 // int int_temp[10000];
4583 AVCodecContext avctx;
4585 dsputil_init(&dsp, &avctx);
4587 init_put_bits(&pb, temp, SIZE);
4588 printf("testing unsigned exp golomb\n");
4589 for(i=0; i<COUNT; i++){
4591 set_ue_golomb(&pb, i);
4592 STOP_TIMER("set_ue_golomb");
4594 flush_put_bits(&pb);
4596 init_get_bits(&gb, temp, 8*SIZE);
4597 for(i=0; i<COUNT; i++){
4600 s= show_bits(&gb, 24);
4603 j= get_ue_golomb(&gb);
4605 printf("missmatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
4608 STOP_TIMER("get_ue_golomb");
4612 init_put_bits(&pb, temp, SIZE);
4613 printf("testing signed exp golomb\n");
4614 for(i=0; i<COUNT; i++){
4616 set_se_golomb(&pb, i - COUNT/2);
4617 STOP_TIMER("set_se_golomb");
4619 flush_put_bits(&pb);
4621 init_get_bits(&gb, temp, 8*SIZE);
4622 for(i=0; i<COUNT; i++){
4625 s= show_bits(&gb, 24);
4628 j= get_se_golomb(&gb);
4629 if(j != i - COUNT/2){
4630 printf("missmatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
4633 STOP_TIMER("get_se_golomb");
4636 printf("testing 4x4 (I)DCT\n");
4639 uint8_t src[16], ref[16];
4640 uint64_t error= 0, max_error=0;
4642 for(i=0; i<COUNT; i++){
4644 // printf("%d %d %d\n", r1, r2, (r2-r1)*16);
4645 for(j=0; j<16; j++){
4646 ref[j]= random()%255;
4647 src[j]= random()%255;
4650 h264_diff_dct_c(block, src, ref, 4);
4653 for(j=0; j<16; j++){
4654 // printf("%d ", block[j]);
4655 block[j]= block[j]*4;
4656 if(j&1) block[j]= (block[j]*4 + 2)/5;
4657 if(j&4) block[j]= (block[j]*4 + 2)/5;
4661 h264_add_idct_c(ref, block, 4);
4662 /* for(j=0; j<16; j++){
4663 printf("%d ", ref[j]);
4667 for(j=0; j<16; j++){
4668 int diff= ABS(src[j] - ref[j]);
4671 max_error= FFMAX(max_error, diff);
4674 printf("error=%f max_error=%d\n", ((float)error)/COUNT/16, (int)max_error );
4676 printf("testing quantizer\n");
4677 for(qp=0; qp<52; qp++){
4679 src1_block[i]= src2_block[i]= random()%255;
4683 printf("Testing NAL layer\n");
4685 uint8_t bitstream[COUNT];
4686 uint8_t nal[COUNT*2];
4688 memset(&h, 0, sizeof(H264Context));
4690 for(i=0; i<COUNT; i++){
4698 for(j=0; j<COUNT; j++){
4699 bitstream[j]= (random() % 255) + 1;
4702 for(j=0; j<zeros; j++){
4703 int pos= random() % COUNT;
4704 while(bitstream[pos] == 0){
4713 nal_length= encode_nal(&h, nal, bitstream, COUNT, COUNT*2);
4715 printf("encoding failed\n");
4719 out= decode_nal(&h, nal, &out_length, &consumed, nal_length);
4723 if(out_length != COUNT){
4724 printf("incorrect length %d %d\n", out_length, COUNT);
4728 if(consumed != nal_length){
4729 printf("incorrect consumed length %d %d\n", nal_length, consumed);
4733 if(memcmp(bitstream, out, COUNT)){
4734 printf("missmatch\n");
4739 printf("Testing RBSP\n");
4747 static int decode_end(AVCodecContext *avctx)
4749 H264Context *h = avctx->priv_data;
4750 MpegEncContext *s = &h->s;
4752 free_tables(h); //FIXME cleanup init stuff perhaps
4755 // memset(h, 0, sizeof(H264Context));
4761 AVCodec h264_decoder = {
4765 sizeof(H264Context),
4770 /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 | CODEC_CAP_TRUNCATED,