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];
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_div2;
253 int slice_beta_offset_div2;
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][0];
475 h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][1];
476 h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][2];
477 h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3];
479 h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][7];
480 h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8];
482 h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][10];
483 h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11];
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]][6];
499 h->non_zero_count_cache[3+8*2]= h->non_zero_count[left_xy[0]][5];
500 h->non_zero_count_cache[0+8*1]= h->non_zero_count[left_xy[0]][9]; //FIXME left_block
501 h->non_zero_count_cache[0+8*4]= h->non_zero_count[left_xy[0]][12];
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]][4];
511 h->non_zero_count_cache[3+8*4]= h->non_zero_count[left_xy[1]][3];
512 h->non_zero_count_cache[0+8*2]= h->non_zero_count[left_xy[1]][8];
513 h->non_zero_count_cache[0+8*5]= h->non_zero_count[left_xy[1]][11];
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;
715 h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[4+8*4];
716 h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[5+8*4];
717 h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[6+8*4];
718 h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];
719 h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[7+8*3];
720 h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[7+8*2];
721 h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[7+8*1];
723 h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[1+8*2];
724 h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
725 h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[2+8*1];
727 h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[1+8*5];
728 h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
729 h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[2+8*4];
733 * gets the predicted number of non zero coefficients.
734 * @param n block index
736 static inline int pred_non_zero_count(H264Context *h, int n){
737 const int index8= scan8[n];
738 const int left= h->non_zero_count_cache[index8 - 1];
739 const int top = h->non_zero_count_cache[index8 - 8];
742 if(i<64) i= (i+1)>>1;
744 tprintf("pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
749 static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
750 const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];
752 if(topright_ref != PART_NOT_AVAILABLE){
753 *C= h->mv_cache[list][ i - 8 + part_width ];
756 tprintf("topright MV not available\n");
758 *C= h->mv_cache[list][ i - 8 - 1 ];
759 return h->ref_cache[list][ i - 8 - 1 ];
764 * gets the predicted MV.
765 * @param n the block index
766 * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)
767 * @param mx the x component of the predicted motion vector
768 * @param my the y component of the predicted motion vector
770 static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
771 const int index8= scan8[n];
772 const int top_ref= h->ref_cache[list][ index8 - 8 ];
773 const int left_ref= h->ref_cache[list][ index8 - 1 ];
774 const int16_t * const A= h->mv_cache[list][ index8 - 1 ];
775 const int16_t * const B= h->mv_cache[list][ index8 - 8 ];
777 int diagonal_ref, match_count;
779 assert(part_width==1 || part_width==2 || part_width==4);
789 diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);
790 match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);
791 if(match_count > 1){ //most common
792 *mx= mid_pred(A[0], B[0], C[0]);
793 *my= mid_pred(A[1], B[1], C[1]);
794 }else if(match_count==1){
798 }else if(top_ref==ref){
806 if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
810 *mx= mid_pred(A[0], B[0], C[0]);
811 *my= mid_pred(A[1], B[1], C[1]);
815 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);
819 * gets the directionally predicted 16x8 MV.
820 * @param n the block index
821 * @param mx the x component of the predicted motion vector
822 * @param my the y component of the predicted motion vector
824 static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
826 const int top_ref= h->ref_cache[list][ scan8[0] - 8 ];
827 const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
829 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);
837 const int left_ref= h->ref_cache[list][ scan8[8] - 1 ];
838 const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];
840 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);
850 pred_motion(h, n, 4, list, ref, mx, my);
854 * gets the directionally predicted 8x16 MV.
855 * @param n the block index
856 * @param mx the x component of the predicted motion vector
857 * @param my the y component of the predicted motion vector
859 static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
861 const int left_ref= h->ref_cache[list][ scan8[0] - 1 ];
862 const int16_t * const A= h->mv_cache[list][ scan8[0] - 1 ];
864 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);
875 diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);
877 tprintf("pred_8x16: (%2d %2d %2d) at %2d %2d %d list %d", diagonal_ref, C[0], C[1], h->s.mb_x, h->s.mb_y, n, list);
879 if(diagonal_ref == ref){
887 pred_motion(h, n, 2, list, ref, mx, my);
890 static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
891 const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
892 const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];
894 tprintf("pred_pskip: (%d) (%d) at %2d %2d", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
896 if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
897 || (top_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0)
898 || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){
904 pred_motion(h, 0, 4, 0, 0, mx, my);
909 static inline void write_back_motion(H264Context *h, int mb_type){
910 MpegEncContext * const s = &h->s;
911 const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
912 const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
915 for(list=0; list<2; list++){
917 if((!IS_8X8(mb_type)) && !USES_LIST(mb_type, list)){
918 if(1){ //FIXME skip or never read if mb_type doesnt use it
920 *(uint64_t*)s->current_picture.motion_val[list][b_xy + 0 + y*h->b_stride]=
921 *(uint64_t*)s->current_picture.motion_val[list][b_xy + 2 + y*h->b_stride]= 0;
924 *(uint16_t*)s->current_picture.motion_val[list][b8_xy + y*h->b8_stride]= (LIST_NOT_USED&0xFF)*0x0101;
927 continue; //FIXME direct mode ...
931 *(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];
932 *(uint64_t*)s->current_picture.motion_val[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mv_cache[list][scan8[0]+2 + 8*y];
935 s->current_picture.ref_index[list][b8_xy + 0 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+0 + 16*y];
936 s->current_picture.ref_index[list][b8_xy + 1 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+2 + 16*y];
942 * Decodes a network abstraction layer unit.
943 * @param consumed is the number of bytes used as input
944 * @param length is the length of the array
945 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp ttailing?
946 * @returns decoded bytes, might be src+1 if no escapes
948 static uint8_t *decode_nal(H264Context *h, uint8_t *src, int *dst_length, int *consumed, int length){
952 // src[0]&0x80; //forbidden bit
953 h->nal_ref_idc= src[0]>>5;
954 h->nal_unit_type= src[0]&0x1F;
958 for(i=0; i<length; i++)
959 printf("%2X ", src[i]);
961 for(i=0; i+1<length; i+=2){
963 if(i>0 && src[i-1]==0) i--;
964 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
966 /* startcode, so we must be past the end */
973 if(i>=length-1){ //no escaped 0
975 *consumed= length+1; //+1 for the header
979 h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length);
982 //printf("deoding esc\n");
985 //remove escapes (very rare 1:2^22)
986 if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
987 if(src[si+2]==3){ //escape
991 }else //next start code
995 dst[di++]= src[si++];
999 *consumed= si + 1;//+1 for the header
1000 //FIXME store exact number of bits in the getbitcontext (its needed for decoding)
1005 * @param src the data which should be escaped
1006 * @param dst the target buffer, dst+1 == src is allowed as a special case
1007 * @param length the length of the src data
1008 * @param dst_length the length of the dst array
1009 * @returns length of escaped data in bytes or -1 if an error occured
1011 static int encode_nal(H264Context *h, uint8_t *dst, uint8_t *src, int length, int dst_length){
1012 int i, escape_count, si, di;
1016 assert(dst_length>0);
1018 dst[0]= (h->nal_ref_idc<<5) + h->nal_unit_type;
1020 if(length==0) return 1;
1023 for(i=0; i<length; i+=2){
1024 if(src[i]) continue;
1025 if(i>0 && src[i-1]==0)
1027 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1033 if(escape_count==0){
1035 memcpy(dst+1, src, length);
1039 if(length + escape_count + 1> dst_length)
1042 //this should be damn rare (hopefully)
1044 h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length + escape_count);
1045 temp= h->rbsp_buffer;
1046 //printf("encoding esc\n");
1051 if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1052 temp[di++]= 0; si++;
1053 temp[di++]= 0; si++;
1055 temp[di++]= src[si++];
1058 temp[di++]= src[si++];
1060 memcpy(dst+1, temp, length+escape_count);
1062 assert(di == length+escape_count);
1068 * write 1,10,100,1000,... for alignment, yes its exactly inverse to mpeg4
1070 static void encode_rbsp_trailing(PutBitContext *pb){
1073 length= (-get_bit_count(pb))&7;
1074 if(length) put_bits(pb, length, 0);
1078 * identifies the exact end of the bitstream
1079 * @return the length of the trailing, or 0 if damaged
1081 static int decode_rbsp_trailing(uint8_t *src){
1085 tprintf("rbsp trailing %X\n", v);
1095 * idct tranforms the 16 dc values and dequantize them.
1096 * @param qp quantization parameter
1098 static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp){
1099 const int qmul= dequant_coeff[qp][0];
1102 int temp[16]; //FIXME check if this is a good idea
1103 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
1104 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1106 //memset(block, 64, 2*256);
1109 const int offset= y_offset[i];
1110 const int z0= block[offset+stride*0] + block[offset+stride*4];
1111 const int z1= block[offset+stride*0] - block[offset+stride*4];
1112 const int z2= block[offset+stride*1] - block[offset+stride*5];
1113 const int z3= block[offset+stride*1] + block[offset+stride*5];
1122 const int offset= x_offset[i];
1123 const int z0= temp[4*0+i] + temp[4*2+i];
1124 const int z1= temp[4*0+i] - temp[4*2+i];
1125 const int z2= temp[4*1+i] - temp[4*3+i];
1126 const int z3= temp[4*1+i] + temp[4*3+i];
1128 block[stride*0 +offset]= ((z0 + z3)*qmul + 2)>>2; //FIXME think about merging this into decode_resdual
1129 block[stride*2 +offset]= ((z1 + z2)*qmul + 2)>>2;
1130 block[stride*8 +offset]= ((z1 - z2)*qmul + 2)>>2;
1131 block[stride*10+offset]= ((z0 - z3)*qmul + 2)>>2;
1136 * dct tranforms the 16 dc values.
1137 * @param qp quantization parameter ??? FIXME
1139 static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
1140 // const int qmul= dequant_coeff[qp][0];
1142 int temp[16]; //FIXME check if this is a good idea
1143 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
1144 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1147 const int offset= y_offset[i];
1148 const int z0= block[offset+stride*0] + block[offset+stride*4];
1149 const int z1= block[offset+stride*0] - block[offset+stride*4];
1150 const int z2= block[offset+stride*1] - block[offset+stride*5];
1151 const int z3= block[offset+stride*1] + block[offset+stride*5];
1160 const int offset= x_offset[i];
1161 const int z0= temp[4*0+i] + temp[4*2+i];
1162 const int z1= temp[4*0+i] - temp[4*2+i];
1163 const int z2= temp[4*1+i] - temp[4*3+i];
1164 const int z3= temp[4*1+i] + temp[4*3+i];
1166 block[stride*0 +offset]= (z0 + z3)>>1;
1167 block[stride*2 +offset]= (z1 + z2)>>1;
1168 block[stride*8 +offset]= (z1 - z2)>>1;
1169 block[stride*10+offset]= (z0 - z3)>>1;
1175 static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp){
1176 const int qmul= dequant_coeff[qp][0];
1177 const int stride= 16*2;
1178 const int xStride= 16;
1181 a= block[stride*0 + xStride*0];
1182 b= block[stride*0 + xStride*1];
1183 c= block[stride*1 + xStride*0];
1184 d= block[stride*1 + xStride*1];
1191 block[stride*0 + xStride*0]= ((a+c)*qmul + 0)>>1;
1192 block[stride*0 + xStride*1]= ((e+b)*qmul + 0)>>1;
1193 block[stride*1 + xStride*0]= ((a-c)*qmul + 0)>>1;
1194 block[stride*1 + xStride*1]= ((e-b)*qmul + 0)>>1;
1197 static void chroma_dc_dct_c(DCTELEM *block){
1198 const int stride= 16*2;
1199 const int xStride= 16;
1202 a= block[stride*0 + xStride*0];
1203 b= block[stride*0 + xStride*1];
1204 c= block[stride*1 + xStride*0];
1205 d= block[stride*1 + xStride*1];
1212 block[stride*0 + xStride*0]= (a+c);
1213 block[stride*0 + xStride*1]= (e+b);
1214 block[stride*1 + xStride*0]= (a-c);
1215 block[stride*1 + xStride*1]= (e-b);
1219 * gets the chroma qp.
1221 static inline int get_chroma_qp(H264Context *h, int qscale){
1223 return chroma_qp[clip(qscale + h->pps.chroma_qp_index_offset, 0, 51)];
1230 static void h264_add_idct_c(uint8_t *dst, DCTELEM *block, int stride){
1232 uint8_t *cm = cropTbl + MAX_NEG_CROP;
1237 const int z0= block[i + 4*0] + block[i + 4*2];
1238 const int z1= block[i + 4*0] - block[i + 4*2];
1239 const int z2= (block[i + 4*1]>>1) - block[i + 4*3];
1240 const int z3= block[i + 4*1] + (block[i + 4*3]>>1);
1242 block[i + 4*0]= z0 + z3;
1243 block[i + 4*1]= z1 + z2;
1244 block[i + 4*2]= z1 - z2;
1245 block[i + 4*3]= z0 - z3;
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 dst[0 + i*stride]= cm[ dst[0 + i*stride] + ((z0 + z3) >> 6) ];
1255 dst[1 + i*stride]= cm[ dst[1 + i*stride] + ((z1 + z2) >> 6) ];
1256 dst[2 + i*stride]= cm[ dst[2 + i*stride] + ((z1 - z2) >> 6) ];
1257 dst[3 + i*stride]= cm[ dst[3 + i*stride] + ((z0 - z3) >> 6) ];
1261 const int z0= block[0 + 4*i] + block[2 + 4*i];
1262 const int z1= block[0 + 4*i] - block[2 + 4*i];
1263 const int z2= (block[1 + 4*i]>>1) - block[3 + 4*i];
1264 const int z3= block[1 + 4*i] + (block[3 + 4*i]>>1);
1266 block[0 + 4*i]= z0 + z3;
1267 block[1 + 4*i]= z1 + z2;
1268 block[2 + 4*i]= z1 - z2;
1269 block[3 + 4*i]= z0 - z3;
1273 const int z0= block[i + 4*0] + block[i + 4*2];
1274 const int z1= block[i + 4*0] - block[i + 4*2];
1275 const int z2= (block[i + 4*1]>>1) - block[i + 4*3];
1276 const int z3= block[i + 4*1] + (block[i + 4*3]>>1);
1278 dst[i + 0*stride]= cm[ dst[i + 0*stride] + ((z0 + z3) >> 6) ];
1279 dst[i + 1*stride]= cm[ dst[i + 1*stride] + ((z1 + z2) >> 6) ];
1280 dst[i + 2*stride]= cm[ dst[i + 2*stride] + ((z1 - z2) >> 6) ];
1281 dst[i + 3*stride]= cm[ dst[i + 3*stride] + ((z0 - z3) >> 6) ];
1286 static void h264_diff_dct_c(DCTELEM *block, uint8_t *src1, uint8_t *src2, int stride){
1288 //FIXME try int temp instead of block
1291 const int d0= src1[0 + i*stride] - src2[0 + i*stride];
1292 const int d1= src1[1 + i*stride] - src2[1 + i*stride];
1293 const int d2= src1[2 + i*stride] - src2[2 + i*stride];
1294 const int d3= src1[3 + i*stride] - src2[3 + i*stride];
1295 const int z0= d0 + d3;
1296 const int z3= d0 - d3;
1297 const int z1= d1 + d2;
1298 const int z2= d1 - d2;
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 const int z0= block[0*4 + i] + block[3*4 + i];
1308 const int z3= block[0*4 + i] - block[3*4 + i];
1309 const int z1= block[1*4 + i] + block[2*4 + i];
1310 const int z2= block[1*4 + i] - block[2*4 + i];
1312 block[0*4 + i]= z0 + z1;
1313 block[1*4 + i]= 2*z3 + z2;
1314 block[2*4 + i]= z0 - z1;
1315 block[3*4 + i]= z3 - 2*z2;
1319 //FIXME need to check that this doesnt overflow signed 32 bit for low qp, iam not sure, its very close
1320 //FIXME check that gcc inlines this (and optimizes intra & seperate_dc stuff away)
1321 static inline int quantize_c(DCTELEM *block, uint8_t *scantable, int qscale, int intra, int seperate_dc){
1323 const int * const quant_table= quant_coeff[qscale];
1324 const int bias= intra ? (1<<QUANT_SHIFT)/3 : (1<<QUANT_SHIFT)/6;
1325 const unsigned int threshold1= (1<<QUANT_SHIFT) - bias - 1;
1326 const unsigned int threshold2= (threshold1<<1);
1332 const int dc_bias= intra ? (1<<(QUANT_SHIFT-2))/3 : (1<<(QUANT_SHIFT-2))/6;
1333 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT-2)) - dc_bias - 1;
1334 const unsigned int dc_threshold2= (dc_threshold1<<1);
1336 int level= block[0]*quant_coeff[qscale+18][0];
1337 if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1339 level= (dc_bias + level)>>(QUANT_SHIFT-2);
1342 level= (dc_bias - level)>>(QUANT_SHIFT-2);
1345 // last_non_zero = i;
1350 const int dc_bias= intra ? (1<<(QUANT_SHIFT+1))/3 : (1<<(QUANT_SHIFT+1))/6;
1351 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT+1)) - dc_bias - 1;
1352 const unsigned int dc_threshold2= (dc_threshold1<<1);
1354 int level= block[0]*quant_table[0];
1355 if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1357 level= (dc_bias + level)>>(QUANT_SHIFT+1);
1360 level= (dc_bias - level)>>(QUANT_SHIFT+1);
1363 // last_non_zero = i;
1376 const int j= scantable[i];
1377 int level= block[j]*quant_table[j];
1379 // if( bias+level >= (1<<(QMAT_SHIFT - 3))
1380 // || bias-level >= (1<<(QMAT_SHIFT - 3))){
1381 if(((unsigned)(level+threshold1))>threshold2){
1383 level= (bias + level)>>QUANT_SHIFT;
1386 level= (bias - level)>>QUANT_SHIFT;
1395 return last_non_zero;
1398 static void pred4x4_vertical_c(uint8_t *src, uint8_t *topright, int stride){
1399 const uint32_t a= ((uint32_t*)(src-stride))[0];
1400 ((uint32_t*)(src+0*stride))[0]= a;
1401 ((uint32_t*)(src+1*stride))[0]= a;
1402 ((uint32_t*)(src+2*stride))[0]= a;
1403 ((uint32_t*)(src+3*stride))[0]= a;
1406 static void pred4x4_horizontal_c(uint8_t *src, uint8_t *topright, int stride){
1407 ((uint32_t*)(src+0*stride))[0]= src[-1+0*stride]*0x01010101;
1408 ((uint32_t*)(src+1*stride))[0]= src[-1+1*stride]*0x01010101;
1409 ((uint32_t*)(src+2*stride))[0]= src[-1+2*stride]*0x01010101;
1410 ((uint32_t*)(src+3*stride))[0]= src[-1+3*stride]*0x01010101;
1413 static void pred4x4_dc_c(uint8_t *src, uint8_t *topright, int stride){
1414 const int dc= ( src[-stride] + src[1-stride] + src[2-stride] + src[3-stride]
1415 + src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 4) >>3;
1417 ((uint32_t*)(src+0*stride))[0]=
1418 ((uint32_t*)(src+1*stride))[0]=
1419 ((uint32_t*)(src+2*stride))[0]=
1420 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1423 static void pred4x4_left_dc_c(uint8_t *src, uint8_t *topright, int stride){
1424 const int dc= ( src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 2) >>2;
1426 ((uint32_t*)(src+0*stride))[0]=
1427 ((uint32_t*)(src+1*stride))[0]=
1428 ((uint32_t*)(src+2*stride))[0]=
1429 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1432 static void pred4x4_top_dc_c(uint8_t *src, uint8_t *topright, int stride){
1433 const int dc= ( src[-stride] + src[1-stride] + src[2-stride] + src[3-stride] + 2) >>2;
1435 ((uint32_t*)(src+0*stride))[0]=
1436 ((uint32_t*)(src+1*stride))[0]=
1437 ((uint32_t*)(src+2*stride))[0]=
1438 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1441 static void pred4x4_128_dc_c(uint8_t *src, uint8_t *topright, int stride){
1442 ((uint32_t*)(src+0*stride))[0]=
1443 ((uint32_t*)(src+1*stride))[0]=
1444 ((uint32_t*)(src+2*stride))[0]=
1445 ((uint32_t*)(src+3*stride))[0]= 128U*0x01010101U;
1449 #define LOAD_TOP_RIGHT_EDGE\
1450 const int t4= topright[0];\
1451 const int t5= topright[1];\
1452 const int t6= topright[2];\
1453 const int t7= topright[3];\
1455 #define LOAD_LEFT_EDGE\
1456 const int l0= src[-1+0*stride];\
1457 const int l1= src[-1+1*stride];\
1458 const int l2= src[-1+2*stride];\
1459 const int l3= src[-1+3*stride];\
1461 #define LOAD_TOP_EDGE\
1462 const int t0= src[ 0-1*stride];\
1463 const int t1= src[ 1-1*stride];\
1464 const int t2= src[ 2-1*stride];\
1465 const int t3= src[ 3-1*stride];\
1467 static void pred4x4_down_right_c(uint8_t *src, uint8_t *topright, int stride){
1468 const int lt= src[-1-1*stride];
1472 src[0+3*stride]=(l3 + 2*l2 + l1 + 2)>>2;
1474 src[1+3*stride]=(l2 + 2*l1 + l0 + 2)>>2;
1477 src[2+3*stride]=(l1 + 2*l0 + lt + 2)>>2;
1481 src[3+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1484 src[3+2*stride]=(lt + 2*t0 + t1 + 2)>>2;
1486 src[3+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1487 src[3+0*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1490 static void pred4x4_down_left_c(uint8_t *src, uint8_t *topright, int stride){
1495 src[0+0*stride]=(t0 + t2 + 2*t1 + 2)>>2;
1497 src[0+1*stride]=(t1 + t3 + 2*t2 + 2)>>2;
1500 src[0+2*stride]=(t2 + t4 + 2*t3 + 2)>>2;
1504 src[0+3*stride]=(t3 + t5 + 2*t4 + 2)>>2;
1507 src[1+3*stride]=(t4 + t6 + 2*t5 + 2)>>2;
1509 src[2+3*stride]=(t5 + t7 + 2*t6 + 2)>>2;
1510 src[3+3*stride]=(t6 + 3*t7 + 2)>>2;
1513 static void pred4x4_vertical_right_c(uint8_t *src, uint8_t *topright, int stride){
1514 const int lt= src[-1-1*stride];
1517 const __attribute__((unused)) int unu= l3;
1520 src[1+2*stride]=(lt + t0 + 1)>>1;
1522 src[2+2*stride]=(t0 + t1 + 1)>>1;
1524 src[3+2*stride]=(t1 + t2 + 1)>>1;
1525 src[3+0*stride]=(t2 + t3 + 1)>>1;
1527 src[1+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1529 src[2+3*stride]=(lt + 2*t0 + t1 + 2)>>2;
1531 src[3+3*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1532 src[3+1*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1533 src[0+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1534 src[0+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1537 static void pred4x4_vertical_left_c(uint8_t *src, uint8_t *topright, int stride){
1540 const __attribute__((unused)) int unu= t7;
1542 src[0+0*stride]=(t0 + t1 + 1)>>1;
1544 src[0+2*stride]=(t1 + t2 + 1)>>1;
1546 src[1+2*stride]=(t2 + t3 + 1)>>1;
1548 src[2+2*stride]=(t3 + t4+ 1)>>1;
1549 src[3+2*stride]=(t4 + t5+ 1)>>1;
1550 src[0+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1552 src[0+3*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1554 src[1+3*stride]=(t2 + 2*t3 + t4 + 2)>>2;
1556 src[2+3*stride]=(t3 + 2*t4 + t5 + 2)>>2;
1557 src[3+3*stride]=(t4 + 2*t5 + t6 + 2)>>2;
1560 static void pred4x4_horizontal_up_c(uint8_t *src, uint8_t *topright, int stride){
1563 src[0+0*stride]=(l0 + l1 + 1)>>1;
1564 src[1+0*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1566 src[0+1*stride]=(l1 + l2 + 1)>>1;
1568 src[1+1*stride]=(l1 + 2*l2 + l3 + 2)>>2;
1570 src[0+2*stride]=(l2 + l3 + 1)>>1;
1572 src[1+2*stride]=(l2 + 2*l3 + l3 + 2)>>2;
1581 static void pred4x4_horizontal_down_c(uint8_t *src, uint8_t *topright, int stride){
1582 const int lt= src[-1-1*stride];
1585 const __attribute__((unused)) int unu= t3;
1588 src[2+1*stride]=(lt + l0 + 1)>>1;
1590 src[3+1*stride]=(l0 + 2*lt + t0 + 2)>>2;
1591 src[2+0*stride]=(lt + 2*t0 + t1 + 2)>>2;
1592 src[3+0*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1594 src[2+2*stride]=(l0 + l1 + 1)>>1;
1596 src[3+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1598 src[2+3*stride]=(l1 + l2+ 1)>>1;
1600 src[3+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1601 src[0+3*stride]=(l2 + l3 + 1)>>1;
1602 src[1+3*stride]=(l1 + 2*l2 + l3 + 2)>>2;
1605 static void pred16x16_vertical_c(uint8_t *src, int stride){
1607 const uint32_t a= ((uint32_t*)(src-stride))[0];
1608 const uint32_t b= ((uint32_t*)(src-stride))[1];
1609 const uint32_t c= ((uint32_t*)(src-stride))[2];
1610 const uint32_t d= ((uint32_t*)(src-stride))[3];
1612 for(i=0; i<16; i++){
1613 ((uint32_t*)(src+i*stride))[0]= a;
1614 ((uint32_t*)(src+i*stride))[1]= b;
1615 ((uint32_t*)(src+i*stride))[2]= c;
1616 ((uint32_t*)(src+i*stride))[3]= d;
1620 static void pred16x16_horizontal_c(uint8_t *src, int stride){
1623 for(i=0; i<16; i++){
1624 ((uint32_t*)(src+i*stride))[0]=
1625 ((uint32_t*)(src+i*stride))[1]=
1626 ((uint32_t*)(src+i*stride))[2]=
1627 ((uint32_t*)(src+i*stride))[3]= src[-1+i*stride]*0x01010101;
1631 static void pred16x16_dc_c(uint8_t *src, int stride){
1635 dc+= src[-1+i*stride];
1642 dc= 0x01010101*((dc + 16)>>5);
1644 for(i=0; i<16; i++){
1645 ((uint32_t*)(src+i*stride))[0]=
1646 ((uint32_t*)(src+i*stride))[1]=
1647 ((uint32_t*)(src+i*stride))[2]=
1648 ((uint32_t*)(src+i*stride))[3]= dc;
1652 static void pred16x16_left_dc_c(uint8_t *src, int stride){
1656 dc+= src[-1+i*stride];
1659 dc= 0x01010101*((dc + 8)>>4);
1661 for(i=0; i<16; i++){
1662 ((uint32_t*)(src+i*stride))[0]=
1663 ((uint32_t*)(src+i*stride))[1]=
1664 ((uint32_t*)(src+i*stride))[2]=
1665 ((uint32_t*)(src+i*stride))[3]= dc;
1669 static void pred16x16_top_dc_c(uint8_t *src, int stride){
1675 dc= 0x01010101*((dc + 8)>>4);
1677 for(i=0; i<16; i++){
1678 ((uint32_t*)(src+i*stride))[0]=
1679 ((uint32_t*)(src+i*stride))[1]=
1680 ((uint32_t*)(src+i*stride))[2]=
1681 ((uint32_t*)(src+i*stride))[3]= dc;
1685 static void pred16x16_128_dc_c(uint8_t *src, int stride){
1688 for(i=0; i<16; i++){
1689 ((uint32_t*)(src+i*stride))[0]=
1690 ((uint32_t*)(src+i*stride))[1]=
1691 ((uint32_t*)(src+i*stride))[2]=
1692 ((uint32_t*)(src+i*stride))[3]= 0x01010101U*128U;
1696 static inline void pred16x16_plane_compat_c(uint8_t *src, int stride, const int svq3){
1699 uint8_t *cm = cropTbl + MAX_NEG_CROP;
1700 const uint8_t * const src0 = src+7-stride;
1701 const uint8_t *src1 = src+8*stride-1;
1702 const uint8_t *src2 = src1-2*stride; // == src+6*stride-1;
1703 int H = src0[1] - src0[-1];
1704 int V = src1[0] - src2[ 0];
1705 for(k=2; k<=8; ++k) {
1706 src1 += stride; src2 -= stride;
1707 H += k*(src0[k] - src0[-k]);
1708 V += k*(src1[0] - src2[ 0]);
1711 H = ( 5*(H/4) ) / 16;
1712 V = ( 5*(V/4) ) / 16;
1714 /* required for 100% accuracy */
1715 i = H; H = V; V = i;
1717 H = ( 5*H+32 ) >> 6;
1718 V = ( 5*V+32 ) >> 6;
1721 a = 16*(src1[0] + src2[16] + 1) - 7*(V+H);
1722 for(j=16; j>0; --j) {
1725 for(i=-16; i<0; i+=4) {
1726 src[16+i] = cm[ (b ) >> 5 ];
1727 src[17+i] = cm[ (b+ H) >> 5 ];
1728 src[18+i] = cm[ (b+2*H) >> 5 ];
1729 src[19+i] = cm[ (b+3*H) >> 5 ];
1736 static void pred16x16_plane_c(uint8_t *src, int stride){
1737 pred16x16_plane_compat_c(src, stride, 0);
1740 static void pred8x8_vertical_c(uint8_t *src, int stride){
1742 const uint32_t a= ((uint32_t*)(src-stride))[0];
1743 const uint32_t b= ((uint32_t*)(src-stride))[1];
1746 ((uint32_t*)(src+i*stride))[0]= a;
1747 ((uint32_t*)(src+i*stride))[1]= b;
1751 static void pred8x8_horizontal_c(uint8_t *src, int stride){
1755 ((uint32_t*)(src+i*stride))[0]=
1756 ((uint32_t*)(src+i*stride))[1]= src[-1+i*stride]*0x01010101;
1760 static void pred8x8_128_dc_c(uint8_t *src, int stride){
1764 ((uint32_t*)(src+i*stride))[0]=
1765 ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
1768 ((uint32_t*)(src+i*stride))[0]=
1769 ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
1773 static void pred8x8_left_dc_c(uint8_t *src, int stride){
1779 dc0+= src[-1+i*stride];
1780 dc2+= src[-1+(i+4)*stride];
1782 dc0= 0x01010101*((dc0 + 2)>>2);
1783 dc2= 0x01010101*((dc2 + 2)>>2);
1786 ((uint32_t*)(src+i*stride))[0]=
1787 ((uint32_t*)(src+i*stride))[1]= dc0;
1790 ((uint32_t*)(src+i*stride))[0]=
1791 ((uint32_t*)(src+i*stride))[1]= dc2;
1795 static void pred8x8_top_dc_c(uint8_t *src, int stride){
1801 dc0+= src[i-stride];
1802 dc1+= src[4+i-stride];
1804 dc0= 0x01010101*((dc0 + 2)>>2);
1805 dc1= 0x01010101*((dc1 + 2)>>2);
1808 ((uint32_t*)(src+i*stride))[0]= dc0;
1809 ((uint32_t*)(src+i*stride))[1]= dc1;
1812 ((uint32_t*)(src+i*stride))[0]= dc0;
1813 ((uint32_t*)(src+i*stride))[1]= dc1;
1818 static void pred8x8_dc_c(uint8_t *src, int stride){
1820 int dc0, dc1, dc2, dc3;
1824 dc0+= src[-1+i*stride] + src[i-stride];
1825 dc1+= src[4+i-stride];
1826 dc2+= src[-1+(i+4)*stride];
1828 dc3= 0x01010101*((dc1 + dc2 + 4)>>3);
1829 dc0= 0x01010101*((dc0 + 4)>>3);
1830 dc1= 0x01010101*((dc1 + 2)>>2);
1831 dc2= 0x01010101*((dc2 + 2)>>2);
1834 ((uint32_t*)(src+i*stride))[0]= dc0;
1835 ((uint32_t*)(src+i*stride))[1]= dc1;
1838 ((uint32_t*)(src+i*stride))[0]= dc2;
1839 ((uint32_t*)(src+i*stride))[1]= dc3;
1843 static void pred8x8_plane_c(uint8_t *src, int stride){
1846 uint8_t *cm = cropTbl + MAX_NEG_CROP;
1847 const uint8_t * const src0 = src+3-stride;
1848 const uint8_t *src1 = src+4*stride-1;
1849 const uint8_t *src2 = src1-2*stride; // == src+2*stride-1;
1850 int H = src0[1] - src0[-1];
1851 int V = src1[0] - src2[ 0];
1852 for(k=2; k<=4; ++k) {
1853 src1 += stride; src2 -= stride;
1854 H += k*(src0[k] - src0[-k]);
1855 V += k*(src1[0] - src2[ 0]);
1857 H = ( 17*H+16 ) >> 5;
1858 V = ( 17*V+16 ) >> 5;
1860 a = 16*(src1[0] + src2[8]+1) - 3*(V+H);
1861 for(j=8; j>0; --j) {
1864 src[0] = cm[ (b ) >> 5 ];
1865 src[1] = cm[ (b+ H) >> 5 ];
1866 src[2] = cm[ (b+2*H) >> 5 ];
1867 src[3] = cm[ (b+3*H) >> 5 ];
1868 src[4] = cm[ (b+4*H) >> 5 ];
1869 src[5] = cm[ (b+5*H) >> 5 ];
1870 src[6] = cm[ (b+6*H) >> 5 ];
1871 src[7] = cm[ (b+7*H) >> 5 ];
1876 static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
1877 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1878 int src_x_offset, int src_y_offset,
1879 qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
1880 MpegEncContext * const s = &h->s;
1881 const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
1882 const int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
1883 const int luma_xy= (mx&3) + ((my&3)<<2);
1884 uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*s->linesize;
1885 uint8_t * src_cb= pic->data[1] + (mx>>3) + (my>>3)*s->uvlinesize;
1886 uint8_t * src_cr= pic->data[2] + (mx>>3) + (my>>3)*s->uvlinesize;
1887 int extra_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16; //FIXME increase edge?, IMHO not worth it
1888 int extra_height= extra_width;
1890 const int full_mx= mx>>2;
1891 const int full_my= my>>2;
1893 assert(pic->data[0]);
1895 if(mx&7) extra_width -= 3;
1896 if(my&7) extra_height -= 3;
1898 if( full_mx < 0-extra_width
1899 || full_my < 0-extra_height
1900 || full_mx + 16/*FIXME*/ > s->width + extra_width
1901 || full_my + 16/*FIXME*/ > s->height + extra_height){
1902 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);
1903 src_y= s->edge_emu_buffer + 2 + 2*s->linesize;
1907 qpix_op[luma_xy](dest_y, src_y, s->linesize); //FIXME try variable height perhaps?
1909 qpix_op[luma_xy](dest_y + delta, src_y + delta, s->linesize);
1912 if(s->flags&CODEC_FLAG_GRAY) return;
1915 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);
1916 src_cb= s->edge_emu_buffer;
1918 chroma_op(dest_cb, src_cb, s->uvlinesize, chroma_height, mx&7, my&7);
1921 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);
1922 src_cr= s->edge_emu_buffer;
1924 chroma_op(dest_cr, src_cr, s->uvlinesize, chroma_height, mx&7, my&7);
1927 static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
1928 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1929 int x_offset, int y_offset,
1930 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
1931 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
1932 int list0, int list1){
1933 MpegEncContext * const s = &h->s;
1934 qpel_mc_func *qpix_op= qpix_put;
1935 h264_chroma_mc_func chroma_op= chroma_put;
1937 dest_y += 2*x_offset + 2*y_offset*s-> linesize;
1938 dest_cb += x_offset + y_offset*s->uvlinesize;
1939 dest_cr += x_offset + y_offset*s->uvlinesize;
1940 x_offset += 8*s->mb_x;
1941 y_offset += 8*s->mb_y;
1944 Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
1945 mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
1946 dest_y, dest_cb, dest_cr, x_offset, y_offset,
1947 qpix_op, chroma_op);
1950 chroma_op= chroma_avg;
1954 Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
1955 mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
1956 dest_y, dest_cb, dest_cr, x_offset, y_offset,
1957 qpix_op, chroma_op);
1961 static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1962 qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
1963 qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg)){
1964 MpegEncContext * const s = &h->s;
1965 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
1966 const int mb_type= s->current_picture.mb_type[mb_xy];
1968 assert(IS_INTER(mb_type));
1970 if(IS_16X16(mb_type)){
1971 mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
1972 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
1973 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1974 }else if(IS_16X8(mb_type)){
1975 mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
1976 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
1977 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1978 mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
1979 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
1980 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
1981 }else if(IS_8X16(mb_type)){
1982 mc_part(h, 0, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 0, 0,
1983 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
1984 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
1985 mc_part(h, 4, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 4, 0,
1986 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
1987 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
1991 assert(IS_8X8(mb_type));
1994 const int sub_mb_type= h->sub_mb_type[i];
1996 int x_offset= (i&1)<<2;
1997 int y_offset= (i&2)<<1;
1999 if(IS_SUB_8X8(sub_mb_type)){
2000 mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2001 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2002 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2003 }else if(IS_SUB_8X4(sub_mb_type)){
2004 mc_part(h, n , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2005 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2006 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2007 mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
2008 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2009 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2010 }else if(IS_SUB_4X8(sub_mb_type)){
2011 mc_part(h, n , 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset, 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));
2014 mc_part(h, n+1, 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
2015 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2016 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2019 assert(IS_SUB_4X4(sub_mb_type));
2021 int sub_x_offset= x_offset + 2*(j&1);
2022 int sub_y_offset= y_offset + (j&2);
2023 mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
2024 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2025 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2032 static void decode_init_vlc(H264Context *h){
2033 static int done = 0;
2039 init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5,
2040 &chroma_dc_coeff_token_len [0], 1, 1,
2041 &chroma_dc_coeff_token_bits[0], 1, 1);
2044 init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17,
2045 &coeff_token_len [i][0], 1, 1,
2046 &coeff_token_bits[i][0], 1, 1);
2050 init_vlc(&chroma_dc_total_zeros_vlc[i], CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
2051 &chroma_dc_total_zeros_len [i][0], 1, 1,
2052 &chroma_dc_total_zeros_bits[i][0], 1, 1);
2054 for(i=0; i<15; i++){
2055 init_vlc(&total_zeros_vlc[i], TOTAL_ZEROS_VLC_BITS, 16,
2056 &total_zeros_len [i][0], 1, 1,
2057 &total_zeros_bits[i][0], 1, 1);
2061 init_vlc(&run_vlc[i], RUN_VLC_BITS, 7,
2062 &run_len [i][0], 1, 1,
2063 &run_bits[i][0], 1, 1);
2065 init_vlc(&run7_vlc, RUN7_VLC_BITS, 16,
2066 &run_len [6][0], 1, 1,
2067 &run_bits[6][0], 1, 1);
2072 * Sets the intra prediction function pointers.
2074 static void init_pred_ptrs(H264Context *h){
2075 // MpegEncContext * const s = &h->s;
2077 h->pred4x4[VERT_PRED ]= pred4x4_vertical_c;
2078 h->pred4x4[HOR_PRED ]= pred4x4_horizontal_c;
2079 h->pred4x4[DC_PRED ]= pred4x4_dc_c;
2080 h->pred4x4[DIAG_DOWN_LEFT_PRED ]= pred4x4_down_left_c;
2081 h->pred4x4[DIAG_DOWN_RIGHT_PRED]= pred4x4_down_right_c;
2082 h->pred4x4[VERT_RIGHT_PRED ]= pred4x4_vertical_right_c;
2083 h->pred4x4[HOR_DOWN_PRED ]= pred4x4_horizontal_down_c;
2084 h->pred4x4[VERT_LEFT_PRED ]= pred4x4_vertical_left_c;
2085 h->pred4x4[HOR_UP_PRED ]= pred4x4_horizontal_up_c;
2086 h->pred4x4[LEFT_DC_PRED ]= pred4x4_left_dc_c;
2087 h->pred4x4[TOP_DC_PRED ]= pred4x4_top_dc_c;
2088 h->pred4x4[DC_128_PRED ]= pred4x4_128_dc_c;
2090 h->pred8x8[DC_PRED8x8 ]= pred8x8_dc_c;
2091 h->pred8x8[VERT_PRED8x8 ]= pred8x8_vertical_c;
2092 h->pred8x8[HOR_PRED8x8 ]= pred8x8_horizontal_c;
2093 h->pred8x8[PLANE_PRED8x8 ]= pred8x8_plane_c;
2094 h->pred8x8[LEFT_DC_PRED8x8]= pred8x8_left_dc_c;
2095 h->pred8x8[TOP_DC_PRED8x8 ]= pred8x8_top_dc_c;
2096 h->pred8x8[DC_128_PRED8x8 ]= pred8x8_128_dc_c;
2098 h->pred16x16[DC_PRED8x8 ]= pred16x16_dc_c;
2099 h->pred16x16[VERT_PRED8x8 ]= pred16x16_vertical_c;
2100 h->pred16x16[HOR_PRED8x8 ]= pred16x16_horizontal_c;
2101 h->pred16x16[PLANE_PRED8x8 ]= pred16x16_plane_c;
2102 h->pred16x16[LEFT_DC_PRED8x8]= pred16x16_left_dc_c;
2103 h->pred16x16[TOP_DC_PRED8x8 ]= pred16x16_top_dc_c;
2104 h->pred16x16[DC_128_PRED8x8 ]= pred16x16_128_dc_c;
2107 static void free_tables(H264Context *h){
2108 av_freep(&h->intra4x4_pred_mode);
2109 av_freep(&h->non_zero_count);
2110 av_freep(&h->slice_table_base);
2111 h->slice_table= NULL;
2113 av_freep(&h->mb2b_xy);
2114 av_freep(&h->mb2b8_xy);
2119 * needs widzh/height
2121 static int alloc_tables(H264Context *h){
2122 MpegEncContext * const s = &h->s;
2123 const int big_mb_num= s->mb_stride * (s->mb_height+1);
2126 CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8 * sizeof(uint8_t))
2127 CHECKED_ALLOCZ(h->non_zero_count , big_mb_num * 16 * sizeof(uint8_t))
2128 CHECKED_ALLOCZ(h->slice_table_base , big_mb_num * sizeof(uint8_t))
2130 memset(h->slice_table_base, -1, big_mb_num * sizeof(uint8_t));
2131 h->slice_table= h->slice_table_base + s->mb_stride + 1;
2133 CHECKED_ALLOCZ(h->mb2b_xy , big_mb_num * sizeof(uint16_t));
2134 CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint16_t));
2135 for(y=0; y<s->mb_height; y++){
2136 for(x=0; x<s->mb_width; x++){
2137 const int mb_xy= x + y*s->mb_stride;
2138 const int b_xy = 4*x + 4*y*h->b_stride;
2139 const int b8_xy= 2*x + 2*y*h->b8_stride;
2141 h->mb2b_xy [mb_xy]= b_xy;
2142 h->mb2b8_xy[mb_xy]= b8_xy;
2152 static void common_init(H264Context *h){
2153 MpegEncContext * const s = &h->s;
2155 s->width = s->avctx->width;
2156 s->height = s->avctx->height;
2157 s->codec_id= s->avctx->codec->id;
2161 s->decode=1; //FIXME
2164 static int decode_init(AVCodecContext *avctx){
2165 H264Context *h= avctx->priv_data;
2166 MpegEncContext * const s = &h->s;
2171 s->out_format = FMT_H264;
2172 s->workaround_bugs= avctx->workaround_bugs;
2175 s->progressive_sequence=1;
2176 // s->decode_mb= ff_h263_decode_mb;
2178 avctx->pix_fmt= PIX_FMT_YUV420P;
2185 static void frame_start(H264Context *h){
2186 MpegEncContext * const s = &h->s;
2189 MPV_frame_start(s, s->avctx);
2190 ff_er_frame_start(s);
2193 assert(s->linesize && s->uvlinesize);
2195 for(i=0; i<16; i++){
2196 h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
2197 h->chroma_subblock_offset[i]= 2*((scan8[i] - scan8[0])&7) + 2*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2200 h->block_offset[16+i]=
2201 h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2204 // s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
2207 static void hl_decode_mb(H264Context *h){
2208 MpegEncContext * const s = &h->s;
2209 const int mb_x= s->mb_x;
2210 const int mb_y= s->mb_y;
2211 const int mb_xy= mb_x + mb_y*s->mb_stride;
2212 const int mb_type= s->current_picture.mb_type[mb_xy];
2213 uint8_t *dest_y, *dest_cb, *dest_cr;
2214 int linesize, uvlinesize /*dct_offset*/;
2223 dest_y = s->current_picture.data[0] + (mb_y * 16* s->linesize ) + mb_x * 16;
2224 dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2225 dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2227 if (h->mb_field_decoding_flag) {
2228 linesize = s->linesize * 2;
2229 uvlinesize = s->uvlinesize * 2;
2230 if(mb_y&1){ //FIXME move out of this func?
2231 dest_y -= s->linesize*15;
2232 dest_cb-= s->linesize*7;
2233 dest_cr-= s->linesize*7;
2236 linesize = s->linesize;
2237 uvlinesize = s->uvlinesize;
2238 // dct_offset = s->linesize * 16;
2241 if(IS_INTRA(mb_type)){
2242 if(!(s->flags&CODEC_FLAG_GRAY)){
2243 h->pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
2244 h->pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
2247 if(IS_INTRA4x4(mb_type)){
2249 for(i=0; i<16; i++){
2250 uint8_t * const ptr= dest_y + h->block_offset[i];
2251 uint8_t *topright= ptr + 4 - linesize;
2252 const int topright_avail= (h->topright_samples_available<<i)&0x8000;
2253 const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
2256 if(!topright_avail){
2257 tr= ptr[3 - linesize]*0x01010101;
2258 topright= (uint8_t*) &tr;
2261 h->pred4x4[ dir ](ptr, topright, linesize);
2262 if(h->non_zero_count_cache[ scan8[i] ]){
2263 if(s->codec_id == CODEC_ID_H264)
2264 h264_add_idct_c(ptr, h->mb + i*16, linesize);
2266 svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
2271 h->pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
2272 if(s->codec_id == CODEC_ID_H264)
2273 h264_luma_dc_dequant_idct_c(h->mb, s->qscale);
2275 svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
2277 }else if(s->codec_id == CODEC_ID_H264){
2278 hl_motion(h, dest_y, dest_cb, dest_cr,
2279 s->dsp.put_h264_qpel_pixels_tab, s->dsp.put_h264_chroma_pixels_tab,
2280 s->dsp.avg_h264_qpel_pixels_tab, s->dsp.avg_h264_chroma_pixels_tab);
2284 if(!IS_INTRA4x4(mb_type)){
2285 if(s->codec_id == CODEC_ID_H264){
2286 for(i=0; i<16; i++){
2287 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
2288 uint8_t * const ptr= dest_y + h->block_offset[i];
2289 h264_add_idct_c(ptr, h->mb + i*16, linesize);
2293 for(i=0; i<16; i++){
2294 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
2295 uint8_t * const ptr= dest_y + h->block_offset[i];
2296 svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
2302 if(!(s->flags&CODEC_FLAG_GRAY)){
2303 chroma_dc_dequant_idct_c(h->mb + 16*16, h->chroma_qp);
2304 chroma_dc_dequant_idct_c(h->mb + 16*16+4*16, h->chroma_qp);
2305 if(s->codec_id == CODEC_ID_H264){
2306 for(i=16; i<16+4; i++){
2307 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2308 uint8_t * const ptr= dest_cb + h->block_offset[i];
2309 h264_add_idct_c(ptr, h->mb + i*16, uvlinesize);
2312 for(i=20; i<20+4; i++){
2313 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2314 uint8_t * const ptr= dest_cr + h->block_offset[i];
2315 h264_add_idct_c(ptr, h->mb + i*16, uvlinesize);
2319 for(i=16; i<16+4; i++){
2320 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2321 uint8_t * const ptr= dest_cb + h->block_offset[i];
2322 svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
2325 for(i=20; i<20+4; i++){
2326 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2327 uint8_t * const ptr= dest_cr + h->block_offset[i];
2328 svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
2335 static void decode_mb_cabac(H264Context *h){
2336 // MpegEncContext * const s = &h->s;
2340 * fills the default_ref_list.
2342 static int fill_default_ref_list(H264Context *h){
2343 MpegEncContext * const s = &h->s;
2345 Picture sorted_short_ref[16];
2347 if(h->slice_type==B_TYPE){
2351 for(out_i=0; out_i<h->short_ref_count; out_i++){
2355 for(i=0; i<h->short_ref_count; i++){
2356 const int poc= h->short_ref[i]->poc;
2357 if(poc > limit && poc < best_poc){
2363 assert(best_i != -1);
2366 sorted_short_ref[out_i]= *h->short_ref[best_i];
2370 if(s->picture_structure == PICT_FRAME){
2371 if(h->slice_type==B_TYPE){
2372 const int current_poc= s->current_picture_ptr->poc;
2375 for(list=0; list<2; list++){
2378 for(i=0; i<h->short_ref_count && index < h->ref_count[list]; i++){
2379 const int i2= list ? h->short_ref_count - i - 1 : i;
2380 const int poc= sorted_short_ref[i2].poc;
2382 if(sorted_short_ref[i2].reference != 3) continue; //FIXME refernce field shit
2384 if((list==1 && poc > current_poc) || (list==0 && poc < current_poc)){
2385 h->default_ref_list[list][index ]= sorted_short_ref[i2];
2386 h->default_ref_list[list][index++].pic_id= sorted_short_ref[i2].frame_num;
2390 for(i=0; i<h->long_ref_count && index < h->ref_count[ list ]; i++){
2391 if(h->long_ref[i]->reference != 3) continue;
2393 h->default_ref_list[ list ][index ]= *h->long_ref[i];
2394 h->default_ref_list[ list ][index++].pic_id= i;;
2397 if(h->long_ref_count > 1 && h->short_ref_count==0){
2398 Picture temp= h->default_ref_list[1][0];
2399 h->default_ref_list[1][0] = h->default_ref_list[1][1];
2400 h->default_ref_list[1][0] = temp;
2403 if(index < h->ref_count[ list ])
2404 memset(&h->default_ref_list[list][index], 0, sizeof(Picture)*(h->ref_count[ list ] - index));
2408 for(i=0; i<h->short_ref_count && index < h->ref_count[0]; i++){
2409 if(h->short_ref[i]->reference != 3) continue; //FIXME refernce field shit
2410 h->default_ref_list[0][index ]= *h->short_ref[i];
2411 h->default_ref_list[0][index++].pic_id= h->short_ref[i]->frame_num;
2413 for(i=0; i<h->long_ref_count && index < h->ref_count[0]; i++){
2414 if(h->long_ref[i]->reference != 3) continue;
2415 h->default_ref_list[0][index ]= *h->long_ref[i];
2416 h->default_ref_list[0][index++].pic_id= i;;
2418 if(index < h->ref_count[0])
2419 memset(&h->default_ref_list[0][index], 0, sizeof(Picture)*(h->ref_count[0] - index));
2422 if(h->slice_type==B_TYPE){
2424 //FIXME second field balh
2430 static int decode_ref_pic_list_reordering(H264Context *h){
2431 MpegEncContext * const s = &h->s;
2434 if(h->slice_type==I_TYPE || h->slice_type==SI_TYPE) return 0; //FIXME move beofre func
2436 for(list=0; list<2; list++){
2437 memcpy(h->ref_list[list], h->default_ref_list[list], sizeof(Picture)*h->ref_count[list]);
2439 if(get_bits1(&s->gb)){
2440 int pred= h->curr_pic_num;
2443 for(index=0; ; index++){
2444 int reordering_of_pic_nums_idc= get_ue_golomb(&s->gb);
2449 if(index >= h->ref_count[list]){
2450 av_log(h->s.avctx, AV_LOG_ERROR, "reference count overflow\n");
2454 if(reordering_of_pic_nums_idc<3){
2455 if(reordering_of_pic_nums_idc<2){
2456 const int abs_diff_pic_num= get_ue_golomb(&s->gb) + 1;
2458 if(abs_diff_pic_num >= h->max_pic_num){
2459 av_log(h->s.avctx, AV_LOG_ERROR, "abs_diff_pic_num overflow\n");
2463 if(reordering_of_pic_nums_idc == 0) pred-= abs_diff_pic_num;
2464 else pred+= abs_diff_pic_num;
2465 pred &= h->max_pic_num - 1;
2467 for(i= h->ref_count[list]-1; i>=index; i--){
2468 if(h->ref_list[list][i].pic_id == pred && h->ref_list[list][i].long_ref==0)
2472 pic_id= get_ue_golomb(&s->gb); //long_term_pic_idx
2474 for(i= h->ref_count[list]-1; i>=index; i--){
2475 if(h->ref_list[list][i].pic_id == pic_id && h->ref_list[list][i].long_ref==1)
2481 av_log(h->s.avctx, AV_LOG_ERROR, "reference picture missing during reorder\n");
2482 memset(&h->ref_list[list][index], 0, sizeof(Picture)); //FIXME
2483 }else if(i > index){
2484 Picture tmp= h->ref_list[list][i];
2485 for(; i>index; i--){
2486 h->ref_list[list][i]= h->ref_list[list][i-1];
2488 h->ref_list[list][index]= tmp;
2490 }else if(reordering_of_pic_nums_idc==3)
2493 av_log(h->s.avctx, AV_LOG_ERROR, "illegal reordering_of_pic_nums_idc\n");
2499 if(h->slice_type!=B_TYPE) break;
2504 static int pred_weight_table(H264Context *h){
2505 MpegEncContext * const s = &h->s;
2508 h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
2509 h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
2511 for(list=0; list<2; list++){
2512 for(i=0; i<h->ref_count[list]; i++){
2513 int luma_weight_flag, chroma_weight_flag;
2515 luma_weight_flag= get_bits1(&s->gb);
2516 if(luma_weight_flag){
2517 h->luma_weight[list][i]= get_se_golomb(&s->gb);
2518 h->luma_offset[list][i]= get_se_golomb(&s->gb);
2521 chroma_weight_flag= get_bits1(&s->gb);
2522 if(chroma_weight_flag){
2525 h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);
2526 h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);
2530 if(h->slice_type != B_TYPE) break;
2536 * instantaneos decoder refresh.
2538 static void idr(H264Context *h){
2541 for(i=0; i<h->long_ref_count; i++){
2542 h->long_ref[i]->reference=0;
2543 h->long_ref[i]= NULL;
2545 h->long_ref_count=0;
2547 for(i=0; i<h->short_ref_count; i++){
2548 h->short_ref[i]->reference=0;
2549 h->short_ref[i]= NULL;
2551 h->short_ref_count=0;
2556 * @return the removed picture or NULL if an error occures
2558 static Picture * remove_short(H264Context *h, int frame_num){
2559 MpegEncContext * const s = &h->s;
2562 if(s->avctx->debug&FF_DEBUG_MMCO)
2563 av_log(h->s.avctx, AV_LOG_DEBUG, "remove short %d count %d\n", frame_num, h->short_ref_count);
2565 for(i=0; i<h->short_ref_count; i++){
2566 Picture *pic= h->short_ref[i];
2567 if(s->avctx->debug&FF_DEBUG_MMCO)
2568 av_log(h->s.avctx, AV_LOG_DEBUG, "%d %d %p\n", i, pic->frame_num, pic);
2569 if(pic->frame_num == frame_num){
2570 h->short_ref[i]= NULL;
2571 memmove(&h->short_ref[i], &h->short_ref[i+1], (h->short_ref_count - i - 1)*sizeof(Picture*));
2572 h->short_ref_count--;
2581 * @return the removed picture or NULL if an error occures
2583 static Picture * remove_long(H264Context *h, int i){
2586 if(i >= h->long_ref_count) return NULL;
2587 pic= h->long_ref[i];
2588 if(pic==NULL) return NULL;
2590 h->long_ref[i]= NULL;
2591 memmove(&h->long_ref[i], &h->long_ref[i+1], (h->long_ref_count - i - 1)*sizeof(Picture*));
2592 h->long_ref_count--;
2598 * Executes the reference picture marking (memory management control operations).
2600 static int execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count){
2601 MpegEncContext * const s = &h->s;
2603 int current_is_long=0;
2606 if((s->avctx->debug&FF_DEBUG_MMCO) && mmco_count==0)
2607 av_log(h->s.avctx, AV_LOG_DEBUG, "no mmco here\n");
2609 for(i=0; i<mmco_count; i++){
2610 if(s->avctx->debug&FF_DEBUG_MMCO)
2611 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);
2613 switch(mmco[i].opcode){
2614 case MMCO_SHORT2UNUSED:
2615 pic= remove_short(h, mmco[i].short_frame_num);
2616 if(pic==NULL) return -1;
2619 case MMCO_SHORT2LONG:
2620 pic= remove_long(h, mmco[i].long_index);
2621 if(pic) pic->reference=0;
2623 h->long_ref[ mmco[i].long_index ]= remove_short(h, mmco[i].short_frame_num);
2624 h->long_ref[ mmco[i].long_index ]->long_ref=1;
2626 case MMCO_LONG2UNUSED:
2627 pic= remove_long(h, mmco[i].long_index);
2628 if(pic==NULL) return -1;
2632 pic= remove_long(h, mmco[i].long_index);
2633 if(pic) pic->reference=0;
2635 h->long_ref[ mmco[i].long_index ]= s->current_picture_ptr;
2636 h->long_ref[ mmco[i].long_index ]->long_ref=1;
2637 h->long_ref_count++;
2641 case MMCO_SET_MAX_LONG:
2642 assert(mmco[i].long_index <= 16);
2643 while(mmco[i].long_index < h->long_ref_count){
2644 pic= remove_long(h, mmco[i].long_index);
2647 while(mmco[i].long_index > h->long_ref_count){
2648 h->long_ref[ h->long_ref_count++ ]= NULL;
2652 while(h->short_ref_count){
2653 pic= remove_short(h, h->short_ref[0]->frame_num);
2656 while(h->long_ref_count){
2657 pic= remove_long(h, h->long_ref_count-1);
2665 if(!current_is_long){
2666 pic= remove_short(h, s->current_picture_ptr->frame_num);
2669 av_log(h->s.avctx, AV_LOG_ERROR, "illegal short term buffer state detected\n");
2672 if(h->short_ref_count)
2673 memmove(&h->short_ref[1], &h->short_ref[0], h->short_ref_count*sizeof(Picture*));
2675 h->short_ref[0]= s->current_picture_ptr;
2676 h->short_ref[0]->long_ref=0;
2677 h->short_ref_count++;
2683 static int decode_ref_pic_marking(H264Context *h){
2684 MpegEncContext * const s = &h->s;
2687 if(h->nal_unit_type == NAL_IDR_SLICE){ //FIXME fields
2688 s->broken_link= get_bits1(&s->gb) -1;
2689 h->mmco[0].long_index= get_bits1(&s->gb) - 1; // current_long_term_idx
2690 if(h->mmco[0].long_index == -1)
2693 h->mmco[0].opcode= MMCO_LONG;
2697 if(get_bits1(&s->gb)){ // adaptive_ref_pic_marking_mode_flag
2698 for(i= h->mmco_index; i<MAX_MMCO_COUNT; i++) {
2699 MMCOOpcode opcode= get_ue_golomb(&s->gb);;
2701 h->mmco[i].opcode= opcode;
2702 if(opcode==MMCO_SHORT2UNUSED || opcode==MMCO_SHORT2LONG){
2703 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
2704 /* if(h->mmco[i].short_frame_num >= h->short_ref_count || h->short_ref[ h->mmco[i].short_frame_num ] == NULL){
2705 fprintf(stderr, "illegal short ref in memory management control operation %d\n", mmco);
2709 if(opcode==MMCO_SHORT2LONG || opcode==MMCO_LONG2UNUSED || opcode==MMCO_LONG || opcode==MMCO_SET_MAX_LONG){
2710 h->mmco[i].long_index= get_ue_golomb(&s->gb);
2711 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){
2712 av_log(h->s.avctx, AV_LOG_ERROR, "illegal long ref in memory management control operation %d\n", opcode);
2717 if(opcode > MMCO_LONG){
2718 av_log(h->s.avctx, AV_LOG_ERROR, "illegal memory management control operation %d\n", opcode);
2724 assert(h->long_ref_count + h->short_ref_count <= h->sps.ref_frame_count);
2726 if(h->long_ref_count + h->short_ref_count == h->sps.ref_frame_count){ //FIXME fields
2727 h->mmco[0].opcode= MMCO_SHORT2UNUSED;
2728 h->mmco[0].short_frame_num= h->short_ref[ h->short_ref_count - 1 ]->frame_num;
2738 static int init_poc(H264Context *h){
2739 MpegEncContext * const s = &h->s;
2740 const int max_frame_num= 1<<h->sps.log2_max_frame_num;
2743 if(h->nal_unit_type == NAL_IDR_SLICE){
2744 h->frame_num_offset= 0;
2746 if(h->frame_num < h->prev_frame_num)
2747 h->frame_num_offset= h->prev_frame_num_offset + max_frame_num;
2749 h->frame_num_offset= h->prev_frame_num_offset;
2752 if(h->sps.poc_type==0){
2753 const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
2755 if (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
2756 h->poc_msb = h->prev_poc_msb + max_poc_lsb;
2757 else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
2758 h->poc_msb = h->prev_poc_msb - max_poc_lsb;
2760 h->poc_msb = h->prev_poc_msb;
2761 //printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
2763 field_poc[1] = h->poc_msb + h->poc_lsb;
2764 if(s->picture_structure == PICT_FRAME)
2765 field_poc[1] += h->delta_poc_bottom;
2766 }else if(h->sps.poc_type==1){
2767 int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
2770 if(h->sps.poc_cycle_length != 0)
2771 abs_frame_num = h->frame_num_offset + h->frame_num;
2775 if(h->nal_ref_idc==0 && abs_frame_num > 0)
2778 expected_delta_per_poc_cycle = 0;
2779 for(i=0; i < h->sps.poc_cycle_length; i++)
2780 expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
2782 if(abs_frame_num > 0){
2783 int poc_cycle_cnt = (abs_frame_num - 1) / h->sps.poc_cycle_length;
2784 int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
2786 expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
2787 for(i = 0; i <= frame_num_in_poc_cycle; i++)
2788 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
2792 if(h->nal_ref_idc == 0)
2793 expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
2795 field_poc[0] = expectedpoc + h->delta_poc[0];
2796 field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
2798 if(s->picture_structure == PICT_FRAME)
2799 field_poc[1] += h->delta_poc[1];
2802 if(h->nal_unit_type == NAL_IDR_SLICE){
2805 if(h->nal_ref_idc) poc= 2*(h->frame_num_offset + h->frame_num);
2806 else poc= 2*(h->frame_num_offset + h->frame_num) - 1;
2812 if(s->picture_structure != PICT_BOTTOM_FIELD)
2813 s->current_picture_ptr->field_poc[0]= field_poc[0];
2814 if(s->picture_structure != PICT_TOP_FIELD)
2815 s->current_picture_ptr->field_poc[1]= field_poc[1];
2816 if(s->picture_structure == PICT_FRAME) // FIXME field pix?
2817 s->current_picture_ptr->poc= FFMIN(field_poc[0], field_poc[1]);
2823 * decodes a slice header.
2824 * this will allso call MPV_common_init() and frame_start() as needed
2826 static int decode_slice_header(H264Context *h){
2827 MpegEncContext * const s = &h->s;
2828 int first_mb_in_slice, pps_id;
2829 int num_ref_idx_active_override_flag;
2830 static const uint8_t slice_type_map[5]= {P_TYPE, B_TYPE, I_TYPE, SP_TYPE, SI_TYPE};
2832 s->current_picture.reference= h->nal_ref_idc != 0;
2834 first_mb_in_slice= get_ue_golomb(&s->gb);
2836 h->slice_type= get_ue_golomb(&s->gb);
2837 if(h->slice_type > 9){
2838 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);
2840 if(h->slice_type > 4){
2842 h->slice_type_fixed=1;
2844 h->slice_type_fixed=0;
2846 h->slice_type= slice_type_map[ h->slice_type ];
2848 s->pict_type= h->slice_type; // to make a few old func happy, its wrong though
2850 pps_id= get_ue_golomb(&s->gb);
2852 av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
2855 h->pps= h->pps_buffer[pps_id];
2856 if(h->pps.slice_group_count == 0){
2857 av_log(h->s.avctx, AV_LOG_ERROR, "non existing PPS referenced\n");
2861 h->sps= h->sps_buffer[ h->pps.sps_id ];
2862 if(h->sps.log2_max_frame_num == 0){
2863 av_log(h->s.avctx, AV_LOG_ERROR, "non existing SPS referenced\n");
2867 s->mb_width= h->sps.mb_width;
2868 s->mb_height= h->sps.mb_height;
2870 h->b_stride= s->mb_width*4;
2871 h->b8_stride= s->mb_width*2;
2873 s->mb_x = first_mb_in_slice % s->mb_width;
2874 s->mb_y = first_mb_in_slice / s->mb_width; //FIXME AFFW
2876 s->width = 16*s->mb_width - 2*(h->sps.crop_left + h->sps.crop_right );
2877 if(h->sps.frame_mbs_only_flag)
2878 s->height= 16*s->mb_height - 2*(h->sps.crop_top + h->sps.crop_bottom);
2880 s->height= 16*s->mb_height - 4*(h->sps.crop_top + h->sps.crop_bottom); //FIXME recheck
2882 if (s->context_initialized
2883 && ( s->width != s->avctx->width || s->height != s->avctx->height)) {
2887 if (!s->context_initialized) {
2888 if (MPV_common_init(s) < 0)
2893 s->avctx->width = s->width;
2894 s->avctx->height = s->height;
2895 s->avctx->sample_aspect_ratio= h->sps.sar;
2898 if(first_mb_in_slice == 0){
2902 s->current_picture_ptr->frame_num= //FIXME frame_num cleanup
2903 h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
2905 if(h->sps.frame_mbs_only_flag){
2906 s->picture_structure= PICT_FRAME;
2908 if(get_bits1(&s->gb)) //field_pic_flag
2909 s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
2911 s->picture_structure= PICT_FRAME;
2914 if(s->picture_structure==PICT_FRAME){
2915 h->curr_pic_num= h->frame_num;
2916 h->max_pic_num= 1<< h->sps.log2_max_frame_num;
2918 h->curr_pic_num= 2*h->frame_num;
2919 h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
2922 if(h->nal_unit_type == NAL_IDR_SLICE){
2923 get_ue_golomb(&s->gb); /* idr_pic_id */
2926 if(h->sps.poc_type==0){
2927 h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
2929 if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
2930 h->delta_poc_bottom= get_se_golomb(&s->gb);
2934 if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
2935 h->delta_poc[0]= get_se_golomb(&s->gb);
2937 if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
2938 h->delta_poc[1]= get_se_golomb(&s->gb);
2943 if(h->pps.redundant_pic_cnt_present){
2944 h->redundant_pic_count= get_ue_golomb(&s->gb);
2947 //set defaults, might be overriden a few line later
2948 h->ref_count[0]= h->pps.ref_count[0];
2949 h->ref_count[1]= h->pps.ref_count[1];
2951 if(h->slice_type == P_TYPE || h->slice_type == SP_TYPE || h->slice_type == B_TYPE){
2952 if(h->slice_type == B_TYPE){
2953 h->direct_spatial_mv_pred= get_bits1(&s->gb);
2955 num_ref_idx_active_override_flag= get_bits1(&s->gb);
2957 if(num_ref_idx_active_override_flag){
2958 h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
2959 if(h->slice_type==B_TYPE)
2960 h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
2962 if(h->ref_count[0] > 32 || h->ref_count[1] > 32){
2963 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
2969 if(first_mb_in_slice == 0){
2970 fill_default_ref_list(h);
2973 decode_ref_pic_list_reordering(h);
2975 if( (h->pps.weighted_pred && (h->slice_type == P_TYPE || h->slice_type == SP_TYPE ))
2976 || (h->pps.weighted_bipred_idc==1 && h->slice_type==B_TYPE ) )
2977 pred_weight_table(h);
2979 if(s->current_picture.reference)
2980 decode_ref_pic_marking(h);
2983 s->qscale = h->pps.init_qp + get_se_golomb(&s->gb); //slice_qp_delta
2984 //FIXME qscale / qp ... stuff
2985 if(h->slice_type == SP_TYPE){
2986 get_bits1(&s->gb); /* sp_for_switch_flag */
2988 if(h->slice_type==SP_TYPE || h->slice_type == SI_TYPE){
2989 get_se_golomb(&s->gb); /* slice_qs_delta */
2992 if( h->pps.deblocking_filter_parameters_present ) {
2993 h->disable_deblocking_filter_idc= get_ue_golomb(&s->gb);
2994 if( h->disable_deblocking_filter_idc != 1 ) {
2995 h->slice_alpha_c0_offset_div2= get_se_golomb(&s->gb);
2996 h->slice_beta_offset_div2= get_se_golomb(&s->gb);
2999 h->disable_deblocking_filter_idc= 0;
3002 if( h->pps.num_slice_groups > 1 && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
3003 slice_group_change_cycle= get_bits(&s->gb, ?);
3006 if(s->avctx->debug&FF_DEBUG_PICT_INFO){
3007 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",
3009 av_get_pict_type_char(h->slice_type),
3010 pps_id, h->frame_num,
3011 s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
3012 h->ref_count[0], h->ref_count[1],
3014 h->disable_deblocking_filter_idc
3024 static inline int get_level_prefix(GetBitContext *gb){
3028 OPEN_READER(re, gb);
3029 UPDATE_CACHE(re, gb);
3030 buf=GET_CACHE(re, gb);
3032 log= 32 - av_log2(buf);
3034 print_bin(buf>>(32-log), log);
3035 printf("%5d %2d %3d lpr @%5d in %s get_level_prefix\n", buf>>(32-log), log, log-1, get_bits_count(gb), __FILE__);
3038 LAST_SKIP_BITS(re, gb, log);
3039 CLOSE_READER(re, gb);
3045 * decodes a residual block.
3046 * @param n block index
3047 * @param scantable scantable
3048 * @param max_coeff number of coefficients in the block
3049 * @return <0 if an error occured
3051 static int decode_residual(H264Context *h, GetBitContext *gb, DCTELEM *block, int n, const uint8_t *scantable, int qp, int max_coeff){
3052 MpegEncContext * const s = &h->s;
3053 const uint16_t *qmul= dequant_coeff[qp];
3054 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};
3055 int level[16], run[16];
3056 int suffix_length, zeros_left, coeff_num, coeff_token, total_coeff, i, trailing_ones;
3058 //FIXME put trailing_onex into the context
3060 if(n == CHROMA_DC_BLOCK_INDEX){
3061 coeff_token= get_vlc2(gb, chroma_dc_coeff_token_vlc.table, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 1);
3062 total_coeff= coeff_token>>2;
3064 if(n == LUMA_DC_BLOCK_INDEX){
3065 total_coeff= pred_non_zero_count(h, 0);
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;
3069 total_coeff= pred_non_zero_count(h, n);
3070 coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
3071 total_coeff= coeff_token>>2;
3072 h->non_zero_count_cache[ scan8[n] ]= total_coeff;
3076 //FIXME set last_non_zero?
3081 trailing_ones= coeff_token&3;
3082 tprintf("trailing:%d, total:%d\n", trailing_ones, total_coeff);
3083 assert(total_coeff<=16);
3085 for(i=0; i<trailing_ones; i++){
3086 level[i]= 1 - 2*get_bits1(gb);
3089 suffix_length= total_coeff > 10 && trailing_ones < 3;
3091 for(; i<total_coeff; i++){
3092 const int prefix= get_level_prefix(gb);
3093 int level_code, mask;
3095 if(prefix<14){ //FIXME try to build a large unified VLC table for all this
3097 level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
3099 level_code= (prefix<<suffix_length); //part
3100 }else if(prefix==14){
3102 level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
3104 level_code= prefix + get_bits(gb, 4); //part
3105 }else if(prefix==15){
3106 level_code= (prefix<<suffix_length) + get_bits(gb, 12); //part
3107 if(suffix_length==0) level_code+=15; //FIXME doesnt make (much)sense
3109 av_log(h->s.avctx, AV_LOG_ERROR, "prefix too large at %d %d\n", s->mb_x, s->mb_y);
3113 if(i==trailing_ones && i<3) level_code+= 2; //FIXME split first iteration
3115 mask= -(level_code&1);
3116 level[i]= (((2+level_code)>>1) ^ mask) - mask;
3118 if(suffix_length==0) suffix_length=1; //FIXME split first iteration
3121 if(ABS(level[i]) > (3<<(suffix_length-1)) && suffix_length<6) suffix_length++;
3123 if((2+level_code)>>1) > (3<<(suffix_length-1)) && suffix_length<6) suffix_length++;
3124 ? == prefix > 2 or sth
3126 tprintf("level: %d suffix_length:%d\n", level[i], suffix_length);
3129 if(total_coeff == max_coeff)
3132 if(n == CHROMA_DC_BLOCK_INDEX)
3133 zeros_left= get_vlc2(gb, chroma_dc_total_zeros_vlc[ total_coeff-1 ].table, CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 1);
3135 zeros_left= get_vlc2(gb, total_zeros_vlc[ total_coeff-1 ].table, TOTAL_ZEROS_VLC_BITS, 1);
3138 for(i=0; i<total_coeff-1; i++){
3141 else if(zeros_left < 7){
3142 run[i]= get_vlc2(gb, run_vlc[zeros_left-1].table, RUN_VLC_BITS, 1);
3144 run[i]= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2);
3146 zeros_left -= run[i];
3150 av_log(h->s.avctx, AV_LOG_ERROR, "negative number of zero coeffs at %d %d\n", s->mb_x, s->mb_y);
3154 for(; i<total_coeff-1; i++){
3162 for(i=total_coeff-1; i>=0; i--){ //FIXME merge into rundecode?
3165 coeff_num += run[i] + 1; //FIXME add 1 earlier ?
3166 j= scantable[ coeff_num ];
3171 for(i=total_coeff-1; i>=0; i--){ //FIXME merge into rundecode?
3174 coeff_num += run[i] + 1; //FIXME add 1 earlier ?
3175 j= scantable[ coeff_num ];
3177 block[j]= level[i] * qmul[j];
3178 // printf("%d %d ", block[j], qmul[j]);
3185 * decodes a macroblock
3186 * @returns 0 if ok, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
3188 static int decode_mb(H264Context *h){
3189 MpegEncContext * const s = &h->s;
3190 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
3191 int mb_type, partition_count, cbp;
3193 s->dsp.clear_blocks(h->mb); //FIXME avoid if allready clear (move after skip handlong?
3195 tprintf("pic:%d mb:%d/%d\n", h->frame_num, s->mb_x, s->mb_y);
3196 cbp = 0; /* avoid warning. FIXME: find a solution without slowing
3198 if(h->slice_type != I_TYPE && h->slice_type != SI_TYPE){
3199 if(s->mb_skip_run==-1)
3200 s->mb_skip_run= get_ue_golomb(&s->gb);
3202 if (s->mb_skip_run--) {
3206 mb_type= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0;
3208 memset(h->non_zero_count[mb_xy], 0, 16);
3209 memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui
3211 if(h->sps.mb_aff && s->mb_skip_run==0 && (s->mb_y&1)==0){
3212 h->mb_field_decoding_flag= get_bits1(&s->gb);
3215 if(h->mb_field_decoding_flag)
3216 mb_type|= MB_TYPE_INTERLACED;
3218 fill_caches(h, mb_type); //FIXME check what is needed and what not ...
3219 pred_pskip_motion(h, &mx, &my);
3220 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
3221 fill_rectangle( h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4);
3222 write_back_motion(h, mb_type);
3224 s->current_picture.mb_type[mb_xy]= mb_type; //FIXME SKIP type
3225 h->slice_table[ mb_xy ]= h->slice_num;
3227 h->prev_mb_skiped= 1;
3231 if(h->sps.mb_aff /* && !field pic FIXME needed? */){
3233 h->mb_field_decoding_flag = get_bits1(&s->gb);
3235 h->mb_field_decoding_flag=0; //FIXME som ed note ?!
3237 h->prev_mb_skiped= 0;
3239 mb_type= get_ue_golomb(&s->gb);
3240 if(h->slice_type == B_TYPE){
3242 partition_count= b_mb_type_info[mb_type].partition_count;
3243 mb_type= b_mb_type_info[mb_type].type;
3246 goto decode_intra_mb;
3248 }else if(h->slice_type == P_TYPE /*|| h->slice_type == SP_TYPE */){
3250 partition_count= p_mb_type_info[mb_type].partition_count;
3251 mb_type= p_mb_type_info[mb_type].type;
3254 goto decode_intra_mb;
3257 assert(h->slice_type == I_TYPE);
3260 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);
3264 cbp= i_mb_type_info[mb_type].cbp;
3265 h->intra16x16_pred_mode= i_mb_type_info[mb_type].pred_mode;
3266 mb_type= i_mb_type_info[mb_type].type;
3269 if(h->mb_field_decoding_flag)
3270 mb_type |= MB_TYPE_INTERLACED;
3272 s->current_picture.mb_type[mb_xy]= mb_type;
3273 h->slice_table[ mb_xy ]= h->slice_num;
3275 if(IS_INTRA_PCM(mb_type)){
3279 // we assume these blocks are very rare so we dont optimize it
3280 align_get_bits(&s->gb);
3282 ptr= s->gb.buffer + get_bits_count(&s->gb);
3284 for(y=0; y<16; y++){
3285 const int index= 4*(y&3) + 64*(y>>2);
3286 for(x=0; x<16; x++){
3287 h->mb[index + (x&3) + 16*(x>>2)]= *(ptr++);
3291 const int index= 256 + 4*(y&3) + 32*(y>>2);
3293 h->mb[index + (x&3) + 16*(x>>2)]= *(ptr++);
3297 const int index= 256 + 64 + 4*(y&3) + 32*(y>>2);
3299 h->mb[index + (x&3) + 16*(x>>2)]= *(ptr++);
3303 skip_bits(&s->gb, 384); //FIXME check /fix the bitstream readers
3305 memset(h->non_zero_count[mb_xy], 16, 16);
3310 fill_caches(h, mb_type);
3313 if(IS_INTRA(mb_type)){
3314 // init_top_left_availability(h);
3315 if(IS_INTRA4x4(mb_type)){
3318 // fill_intra4x4_pred_table(h);
3319 for(i=0; i<16; i++){
3320 const int mode_coded= !get_bits1(&s->gb);
3321 const int predicted_mode= pred_intra_mode(h, i);
3325 const int rem_mode= get_bits(&s->gb, 3);
3326 if(rem_mode<predicted_mode)
3331 mode= predicted_mode;
3334 h->intra4x4_pred_mode_cache[ scan8[i] ] = mode;
3336 write_back_intra_pred_mode(h);
3337 if( check_intra4x4_pred_mode(h) < 0)
3340 h->intra16x16_pred_mode= check_intra_pred_mode(h, h->intra16x16_pred_mode);
3341 if(h->intra16x16_pred_mode < 0)
3344 h->chroma_pred_mode= get_ue_golomb(&s->gb);
3346 h->chroma_pred_mode= check_intra_pred_mode(h, h->chroma_pred_mode);
3347 if(h->chroma_pred_mode < 0)
3349 }else if(partition_count==4){
3350 int i, j, sub_partition_count[4], list, ref[2][4];
3352 if(h->slice_type == B_TYPE){
3354 h->sub_mb_type[i]= get_ue_golomb(&s->gb);
3355 if(h->sub_mb_type[i] >=13){
3356 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);
3359 sub_partition_count[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
3360 h->sub_mb_type[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].type;
3363 assert(h->slice_type == P_TYPE || h->slice_type == SP_TYPE); //FIXME SP correct ?
3365 h->sub_mb_type[i]= get_ue_golomb(&s->gb);
3366 if(h->sub_mb_type[i] >=4){
3367 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);
3370 sub_partition_count[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
3371 h->sub_mb_type[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].type;
3375 for(list=0; list<2; list++){
3376 const int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list];
3377 if(ref_count == 0) continue;
3379 if(IS_DIR(h->sub_mb_type[i], 0, list) && !IS_DIRECT(h->sub_mb_type[i])){
3380 ref[list][i] = get_te0_golomb(&s->gb, ref_count); //FIXME init to 0 before and skip?
3388 for(list=0; list<2; list++){
3389 const int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list];
3390 if(ref_count == 0) continue;
3393 h->ref_cache[list][ scan8[4*i] ]=h->ref_cache[list][ scan8[4*i]+1 ]=
3394 h->ref_cache[list][ scan8[4*i]+8 ]=h->ref_cache[list][ scan8[4*i]+9 ]= ref[list][i];
3396 if(IS_DIR(h->sub_mb_type[i], 0, list) && !IS_DIRECT(h->sub_mb_type[i])){
3397 const int sub_mb_type= h->sub_mb_type[i];
3398 const int block_width= (sub_mb_type & (MB_TYPE_16x16|MB_TYPE_16x8)) ? 2 : 1;
3399 for(j=0; j<sub_partition_count[i]; j++){
3401 const int index= 4*i + block_width*j;
3402 int16_t (* mv_cache)[2]= &h->mv_cache[list][ scan8[index] ];
3403 pred_motion(h, index, block_width, list, h->ref_cache[list][ scan8[index] ], &mx, &my);
3404 mx += get_se_golomb(&s->gb);
3405 my += get_se_golomb(&s->gb);
3406 tprintf("final mv:%d %d\n", mx, my);
3408 if(IS_SUB_8X8(sub_mb_type)){
3409 mv_cache[ 0 ][0]= mv_cache[ 1 ][0]=
3410 mv_cache[ 8 ][0]= mv_cache[ 9 ][0]= mx;
3411 mv_cache[ 0 ][1]= mv_cache[ 1 ][1]=
3412 mv_cache[ 8 ][1]= mv_cache[ 9 ][1]= my;
3413 }else if(IS_SUB_8X4(sub_mb_type)){
3414 mv_cache[ 0 ][0]= mv_cache[ 1 ][0]= mx;
3415 mv_cache[ 0 ][1]= mv_cache[ 1 ][1]= my;
3416 }else if(IS_SUB_4X8(sub_mb_type)){
3417 mv_cache[ 0 ][0]= mv_cache[ 8 ][0]= mx;
3418 mv_cache[ 0 ][1]= mv_cache[ 8 ][1]= my;
3420 assert(IS_SUB_4X4(sub_mb_type));
3421 mv_cache[ 0 ][0]= mx;
3422 mv_cache[ 0 ][1]= my;
3426 uint32_t *p= (uint32_t *)&h->mv_cache[list][ scan8[4*i] ][0];
3432 }else if(!IS_DIRECT(mb_type)){
3433 int list, mx, my, i;
3434 //FIXME we should set ref_idx_l? to 0 if we use that later ...
3435 if(IS_16X16(mb_type)){
3436 for(list=0; list<2; list++){
3437 if(h->ref_count[0]>0){
3438 if(IS_DIR(mb_type, 0, list)){
3439 const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
3440 fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, val, 1);
3444 for(list=0; list<2; list++){
3445 if(IS_DIR(mb_type, 0, list)){
3446 pred_motion(h, 0, 4, list, h->ref_cache[list][ scan8[0] ], &mx, &my);
3447 mx += get_se_golomb(&s->gb);
3448 my += get_se_golomb(&s->gb);
3449 tprintf("final mv:%d %d\n", mx, my);
3451 fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, pack16to32(mx,my), 4);
3455 else if(IS_16X8(mb_type)){
3456 for(list=0; list<2; list++){
3457 if(h->ref_count[list]>0){
3459 if(IS_DIR(mb_type, i, list)){
3460 const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
3461 fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, val, 1);
3466 for(list=0; list<2; list++){
3468 if(IS_DIR(mb_type, i, list)){
3469 pred_16x8_motion(h, 8*i, list, h->ref_cache[list][scan8[0] + 16*i], &mx, &my);
3470 mx += get_se_golomb(&s->gb);
3471 my += get_se_golomb(&s->gb);
3472 tprintf("final mv:%d %d\n", mx, my);
3474 fill_rectangle(h->mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, pack16to32(mx,my), 4);
3479 assert(IS_8X16(mb_type));
3480 for(list=0; list<2; list++){
3481 if(h->ref_count[list]>0){
3483 if(IS_DIR(mb_type, i, list)){ //FIXME optimize
3484 const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
3485 fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, val, 1);
3490 for(list=0; list<2; list++){
3492 if(IS_DIR(mb_type, i, list)){
3493 pred_8x16_motion(h, i*4, list, h->ref_cache[list][ scan8[0] + 2*i ], &mx, &my);
3494 mx += get_se_golomb(&s->gb);
3495 my += get_se_golomb(&s->gb);
3496 tprintf("final mv:%d %d\n", mx, my);
3498 fill_rectangle(h->mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, pack16to32(mx,my), 4);
3505 if(IS_INTER(mb_type))
3506 write_back_motion(h, mb_type);
3508 if(!IS_INTRA16x16(mb_type)){
3509 cbp= get_ue_golomb(&s->gb);
3511 av_log(h->s.avctx, AV_LOG_ERROR, "cbp too large (%d) at %d %d\n", cbp, s->mb_x, s->mb_y);
3515 if(IS_INTRA4x4(mb_type))
3516 cbp= golomb_to_intra4x4_cbp[cbp];
3518 cbp= golomb_to_inter_cbp[cbp];
3521 if(cbp || IS_INTRA16x16(mb_type)){
3522 int i8x8, i4x4, chroma_idx;
3523 int chroma_qp, dquant;
3524 GetBitContext *gb= IS_INTRA(mb_type) ? h->intra_gb_ptr : h->inter_gb_ptr;
3525 const uint8_t *scan, *dc_scan;
3527 // fill_non_zero_count_cache(h);
3529 if(IS_INTERLACED(mb_type)){
3531 dc_scan= luma_dc_field_scan;
3534 dc_scan= luma_dc_zigzag_scan;
3537 dquant= get_se_golomb(&s->gb);
3539 if( dquant > 25 || dquant < -26 ){
3540 av_log(h->s.avctx, AV_LOG_ERROR, "dquant out of range (%d) at %d %d\n", dquant, s->mb_x, s->mb_y);
3544 s->qscale += dquant;
3545 if(((unsigned)s->qscale) > 51){
3546 if(s->qscale<0) s->qscale+= 52;
3547 else s->qscale-= 52;
3550 h->chroma_qp= chroma_qp= get_chroma_qp(h, s->qscale);
3551 if(IS_INTRA16x16(mb_type)){
3552 if( decode_residual(h, h->intra_gb_ptr, h->mb, LUMA_DC_BLOCK_INDEX, dc_scan, s->qscale, 16) < 0){
3553 return -1; //FIXME continue if partotioned and other retirn -1 too
3556 assert((cbp&15) == 0 || (cbp&15) == 15);
3559 for(i8x8=0; i8x8<4; i8x8++){
3560 for(i4x4=0; i4x4<4; i4x4++){
3561 const int index= i4x4 + 4*i8x8;
3562 if( decode_residual(h, h->intra_gb_ptr, h->mb + 16*index, index, scan + 1, s->qscale, 15) < 0 ){
3568 fill_rectangle(&h->non_zero_count_cache[scan8[0]], 4, 4, 8, 0, 1);
3571 for(i8x8=0; i8x8<4; i8x8++){
3572 if(cbp & (1<<i8x8)){
3573 for(i4x4=0; i4x4<4; i4x4++){
3574 const int index= i4x4 + 4*i8x8;
3576 if( decode_residual(h, gb, h->mb + 16*index, index, scan, s->qscale, 16) <0 ){
3581 uint8_t * const nnz= &h->non_zero_count_cache[ scan8[4*i8x8] ];
3582 nnz[0] = nnz[1] = nnz[8] = nnz[9] = 0;
3588 for(chroma_idx=0; chroma_idx<2; chroma_idx++)
3589 if( decode_residual(h, gb, h->mb + 256 + 16*4*chroma_idx, CHROMA_DC_BLOCK_INDEX, chroma_dc_scan, chroma_qp, 4) < 0){
3595 for(chroma_idx=0; chroma_idx<2; chroma_idx++){
3596 for(i4x4=0; i4x4<4; i4x4++){
3597 const int index= 16 + 4*chroma_idx + i4x4;
3598 if( decode_residual(h, gb, h->mb + 16*index, index, scan + 1, chroma_qp, 15) < 0){
3604 uint8_t * const nnz= &h->non_zero_count_cache[0];
3605 nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =
3606 nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
3609 memset(&h->non_zero_count_cache[8], 0, 8*5);
3611 write_back_non_zero_count(h);
3616 static int decode_slice(H264Context *h){
3617 MpegEncContext * const s = &h->s;
3618 const int part_mask= s->partitioned_frame ? (AC_END|AC_ERROR) : 0x7F;
3624 int ret= decode_mb(h);
3628 if(ret>=0 && h->sps.mb_aff){ //FIXME optimal? or let mb_decode decode 16x32 ?
3637 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
3638 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);
3643 if(++s->mb_x >= s->mb_width){
3645 ff_draw_horiz_band(s, 16*s->mb_y, 16);
3646 if(++s->mb_y >= s->mb_height){
3647 tprintf("slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
3649 if(get_bits_count(&s->gb) == s->gb.size_in_bits){
3650 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);
3654 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);
3661 if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->mb_skip_run<=0){
3662 if(get_bits_count(&s->gb) == s->gb.size_in_bits){
3663 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);
3667 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);
3675 for(;s->mb_y < s->mb_height; s->mb_y++){
3676 for(;s->mb_x < s->mb_width; s->mb_x++){
3677 int ret= decode_mb(h);
3682 fprintf(stderr, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
3683 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);
3688 if(++s->mb_x >= s->mb_width){
3690 if(++s->mb_y >= s->mb_height){
3691 if(get_bits_count(s->gb) == s->gb.size_in_bits){
3692 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);
3696 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);
3703 if(get_bits_count(s->?gb) >= s->gb?.size_in_bits){
3704 if(get_bits_count(s->gb) == s->gb.size_in_bits){
3705 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);
3709 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);
3716 ff_draw_horiz_band(s, 16*s->mb_y, 16);
3719 return -1; //not reached
3722 static inline int decode_vui_parameters(H264Context *h, SPS *sps){
3723 MpegEncContext * const s = &h->s;
3724 int aspect_ratio_info_present_flag, aspect_ratio_idc;
3726 aspect_ratio_info_present_flag= get_bits1(&s->gb);
3728 if( aspect_ratio_info_present_flag ) {
3729 aspect_ratio_idc= get_bits(&s->gb, 8);
3730 if( aspect_ratio_idc == EXTENDED_SAR ) {
3731 sps->sar.num= get_bits(&s->gb, 16);
3732 sps->sar.den= get_bits(&s->gb, 16);
3733 }else if(aspect_ratio_idc < 16){
3734 sps->sar= pixel_aspect[aspect_ratio_idc];
3736 av_log(h->s.avctx, AV_LOG_ERROR, "illegal aspect ratio\n");
3743 // s->avctx->aspect_ratio= sar_width*s->width / (float)(s->height*sar_height);
3745 | overscan_info_present_flag |0 |u(1) |
3746 | if( overscan_info_present_flag ) | | |
3747 | overscan_appropriate_flag |0 |u(1) |
3748 | video_signal_type_present_flag |0 |u(1) |
3749 | if( video_signal_type_present_flag ) { | | |
3750 | video_format |0 |u(3) |
3751 | video_full_range_flag |0 |u(1) |
3752 | colour_description_present_flag |0 |u(1) |
3753 | if( colour_description_present_flag ) { | | |
3754 | colour_primaries |0 |u(8) |
3755 | transfer_characteristics |0 |u(8) |
3756 | matrix_coefficients |0 |u(8) |
3759 | chroma_location_info_present_flag |0 |u(1) |
3760 | if ( chroma_location_info_present_flag ) { | | |
3761 | chroma_sample_location_type_top_field |0 |ue(v) |
3762 | chroma_sample_location_type_bottom_field |0 |ue(v) |
3764 | timing_info_present_flag |0 |u(1) |
3765 | if( timing_info_present_flag ) { | | |
3766 | num_units_in_tick |0 |u(32) |
3767 | time_scale |0 |u(32) |
3768 | fixed_frame_rate_flag |0 |u(1) |
3770 | nal_hrd_parameters_present_flag |0 |u(1) |
3771 | if( nal_hrd_parameters_present_flag = = 1) | | |
3772 | hrd_parameters( ) | | |
3773 | vcl_hrd_parameters_present_flag |0 |u(1) |
3774 | if( vcl_hrd_parameters_present_flag = = 1) | | |
3775 | hrd_parameters( ) | | |
3776 | if( ( nal_hrd_parameters_present_flag = = 1 | || | |
3778 |( vcl_hrd_parameters_present_flag = = 1 ) ) | | |
3779 | low_delay_hrd_flag |0 |u(1) |
3780 | bitstream_restriction_flag |0 |u(1) |
3781 | if( bitstream_restriction_flag ) { |0 |u(1) |
3782 | motion_vectors_over_pic_boundaries_flag |0 |u(1) |
3783 | max_bytes_per_pic_denom |0 |ue(v) |
3784 | max_bits_per_mb_denom |0 |ue(v) |
3785 | log2_max_mv_length_horizontal |0 |ue(v) |
3786 | log2_max_mv_length_vertical |0 |ue(v) |
3787 | num_reorder_frames |0 |ue(v) |
3788 | max_dec_frame_buffering |0 |ue(v) |
3795 static inline int decode_seq_parameter_set(H264Context *h){
3796 MpegEncContext * const s = &h->s;
3797 int profile_idc, level_idc;
3801 profile_idc= get_bits(&s->gb, 8);
3802 get_bits1(&s->gb); //constraint_set0_flag
3803 get_bits1(&s->gb); //constraint_set1_flag
3804 get_bits1(&s->gb); //constraint_set2_flag
3805 get_bits(&s->gb, 5); // reserved
3806 level_idc= get_bits(&s->gb, 8);
3807 sps_id= get_ue_golomb(&s->gb);
3809 sps= &h->sps_buffer[ sps_id ];
3810 sps->profile_idc= profile_idc;
3811 sps->level_idc= level_idc;
3813 sps->log2_max_frame_num= get_ue_golomb(&s->gb) + 4;
3814 sps->poc_type= get_ue_golomb(&s->gb);
3816 if(sps->poc_type == 0){ //FIXME #define
3817 sps->log2_max_poc_lsb= get_ue_golomb(&s->gb) + 4;
3818 } else if(sps->poc_type == 1){//FIXME #define
3819 sps->delta_pic_order_always_zero_flag= get_bits1(&s->gb);
3820 sps->offset_for_non_ref_pic= get_se_golomb(&s->gb);
3821 sps->offset_for_top_to_bottom_field= get_se_golomb(&s->gb);
3822 sps->poc_cycle_length= get_ue_golomb(&s->gb);
3824 for(i=0; i<sps->poc_cycle_length; i++)
3825 sps->offset_for_ref_frame[i]= get_se_golomb(&s->gb);
3827 if(sps->poc_type > 2){
3828 av_log(h->s.avctx, AV_LOG_ERROR, "illegal POC type %d\n", sps->poc_type);
3832 sps->ref_frame_count= get_ue_golomb(&s->gb);
3833 sps->gaps_in_frame_num_allowed_flag= get_bits1(&s->gb);
3834 sps->mb_width= get_ue_golomb(&s->gb) + 1;
3835 sps->mb_height= get_ue_golomb(&s->gb) + 1;
3836 sps->frame_mbs_only_flag= get_bits1(&s->gb);
3837 if(!sps->frame_mbs_only_flag)
3838 sps->mb_aff= get_bits1(&s->gb);
3842 sps->direct_8x8_inference_flag= get_bits1(&s->gb);
3844 sps->crop= get_bits1(&s->gb);
3846 sps->crop_left = get_ue_golomb(&s->gb);
3847 sps->crop_right = get_ue_golomb(&s->gb);
3848 sps->crop_top = get_ue_golomb(&s->gb);
3849 sps->crop_bottom= get_ue_golomb(&s->gb);
3850 if(sps->crop_left || sps->crop_top){
3851 av_log(h->s.avctx, AV_LOG_ERROR, "insane cropping not completly supported, this could look slightly wrong ...\n");
3857 sps->crop_bottom= 0;
3860 sps->vui_parameters_present_flag= get_bits1(&s->gb);
3861 if( sps->vui_parameters_present_flag )
3862 decode_vui_parameters(h, sps);
3864 if(s->avctx->debug&FF_DEBUG_PICT_INFO){
3865 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",
3866 sps_id, sps->profile_idc, sps->level_idc,
3868 sps->ref_frame_count,
3869 sps->mb_width, sps->mb_height,
3870 sps->frame_mbs_only_flag ? "FRM" : (sps->mb_aff ? "MB-AFF" : "PIC-AFF"),
3871 sps->direct_8x8_inference_flag ? "8B8" : "",
3872 sps->crop_left, sps->crop_right,
3873 sps->crop_top, sps->crop_bottom,
3874 sps->vui_parameters_present_flag ? "VUI" : ""
3880 static inline int decode_picture_parameter_set(H264Context *h){
3881 MpegEncContext * const s = &h->s;
3882 int pps_id= get_ue_golomb(&s->gb);
3883 PPS *pps= &h->pps_buffer[pps_id];
3885 pps->sps_id= get_ue_golomb(&s->gb);
3886 pps->cabac= get_bits1(&s->gb);
3887 pps->pic_order_present= get_bits1(&s->gb);
3888 pps->slice_group_count= get_ue_golomb(&s->gb) + 1;
3889 if(pps->slice_group_count > 1 ){
3890 pps->mb_slice_group_map_type= get_ue_golomb(&s->gb);
3891 av_log(h->s.avctx, AV_LOG_ERROR, "FMO not supported\n");
3892 switch(pps->mb_slice_group_map_type){
3895 | for( i = 0; i <= num_slice_groups_minus1; i++ ) | | |
3896 | run_length[ i ] |1 |ue(v) |
3901 | for( i = 0; i < num_slice_groups_minus1; i++ ) | | |
3903 | top_left_mb[ i ] |1 |ue(v) |
3904 | bottom_right_mb[ i ] |1 |ue(v) |
3912 | slice_group_change_direction_flag |1 |u(1) |
3913 | slice_group_change_rate_minus1 |1 |ue(v) |
3918 | slice_group_id_cnt_minus1 |1 |ue(v) |
3919 | for( i = 0; i <= slice_group_id_cnt_minus1; i++ | | |
3921 | slice_group_id[ i ] |1 |u(v) |
3926 pps->ref_count[0]= get_ue_golomb(&s->gb) + 1;
3927 pps->ref_count[1]= get_ue_golomb(&s->gb) + 1;
3928 if(pps->ref_count[0] > 32 || pps->ref_count[1] > 32){
3929 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow (pps)\n");
3933 pps->weighted_pred= get_bits1(&s->gb);
3934 pps->weighted_bipred_idc= get_bits(&s->gb, 2);
3935 pps->init_qp= get_se_golomb(&s->gb) + 26;
3936 pps->init_qs= get_se_golomb(&s->gb) + 26;
3937 pps->chroma_qp_index_offset= get_se_golomb(&s->gb);
3938 pps->deblocking_filter_parameters_present= get_bits1(&s->gb);
3939 pps->constrained_intra_pred= get_bits1(&s->gb);
3940 pps->redundant_pic_cnt_present = get_bits1(&s->gb);
3942 if(s->avctx->debug&FF_DEBUG_PICT_INFO){
3943 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",
3944 pps_id, pps->sps_id,
3945 pps->cabac ? "CABAC" : "CAVLC",
3946 pps->slice_group_count,
3947 pps->ref_count[0], pps->ref_count[1],
3948 pps->weighted_pred ? "weighted" : "",
3949 pps->init_qp, pps->init_qs, pps->chroma_qp_index_offset,
3950 pps->deblocking_filter_parameters_present ? "LPAR" : "",
3951 pps->constrained_intra_pred ? "CONSTR" : "",
3952 pps->redundant_pic_cnt_present ? "REDU" : ""
3960 * finds the end of the current frame in the bitstream.
3961 * @return the position of the first byte of the next frame, or -1
3963 static int find_frame_end(MpegEncContext *s, uint8_t *buf, int buf_size){
3964 ParseContext *pc= &s->parse_context;
3967 //printf("first %02X%02X%02X%02X\n", buf[0], buf[1],buf[2],buf[3]);
3968 // mb_addr= pc->mb_addr - 1;
3970 //FIXME this will fail with slices
3971 for(i=0; i<buf_size; i++){
3972 state= (state<<8) | buf[i];
3973 if((state&0xFFFFFF1F) == 0x101 || (state&0xFFFFFF1F) == 0x102 || (state&0xFFFFFF1F) == 0x105){
3974 if(pc->frame_start_found){
3976 pc->frame_start_found= 0;
3979 pc->frame_start_found= 1;
3984 return END_NOT_FOUND;
3987 static int decode_nal_units(H264Context *h, uint8_t *buf, int buf_size){
3988 MpegEncContext * const s = &h->s;
3989 AVCodecContext * const avctx= s->avctx;
3993 for(i=0; i<32; i++){
3994 printf("%X ", buf[i]);
4003 // start code prefix search
4004 for(; buf_index + 3 < buf_size; buf_index++){
4005 // this should allways succeed in the first iteration
4006 if(buf[buf_index] == 0 && buf[buf_index+1] == 0 && buf[buf_index+2] == 1)
4010 if(buf_index+3 >= buf_size) break;
4014 ptr= decode_nal(h, buf + buf_index, &dst_length, &consumed, buf_size - buf_index);
4015 if(ptr[dst_length - 1] == 0) dst_length--;
4016 bit_length= 8*dst_length - decode_rbsp_trailing(ptr + dst_length - 1);
4018 if(s->avctx->debug&FF_DEBUG_STARTCODE){
4019 av_log(h->s.avctx, AV_LOG_DEBUG, "NAL %d at %d length %d\n", h->nal_unit_type, buf_index, dst_length);
4022 buf_index += consumed;
4024 if(h->nal_ref_idc < s->hurry_up)
4027 switch(h->nal_unit_type){
4029 idr(h); //FIXME ensure we dont loose some frames if there is reordering
4031 init_get_bits(&s->gb, ptr, bit_length);
4033 h->inter_gb_ptr= &s->gb;
4034 s->data_partitioning = 0;
4036 if(decode_slice_header(h) < 0) return -1;
4037 if(h->redundant_pic_count==0)
4041 init_get_bits(&s->gb, ptr, bit_length);
4043 h->inter_gb_ptr= NULL;
4044 s->data_partitioning = 1;
4046 if(decode_slice_header(h) < 0) return -1;
4049 init_get_bits(&h->intra_gb, ptr, bit_length);
4050 h->intra_gb_ptr= &h->intra_gb;
4053 init_get_bits(&h->inter_gb, ptr, bit_length);
4054 h->inter_gb_ptr= &h->inter_gb;
4056 if(h->redundant_pic_count==0 && h->intra_gb_ptr && s->data_partitioning)
4062 init_get_bits(&s->gb, ptr, bit_length);
4063 decode_seq_parameter_set(h);
4065 if(s->flags& CODEC_FLAG_LOW_DELAY)
4068 avctx->has_b_frames= !s->low_delay;
4071 init_get_bits(&s->gb, ptr, bit_length);
4073 decode_picture_parameter_set(h);
4076 case NAL_PICTURE_DELIMITER:
4078 case NAL_FILTER_DATA:
4082 //FIXME move after where irt is set
4083 s->current_picture.pict_type= s->pict_type;
4084 s->current_picture.key_frame= s->pict_type == I_TYPE;
4087 if(!s->current_picture_ptr) return buf_index; //no frame
4089 h->prev_frame_num_offset= h->frame_num_offset;
4090 h->prev_frame_num= h->frame_num;
4091 if(s->current_picture_ptr->reference){
4092 h->prev_poc_msb= h->poc_msb;
4093 h->prev_poc_lsb= h->poc_lsb;
4095 if(s->current_picture_ptr->reference)
4096 execute_ref_pic_marking(h, h->mmco, h->mmco_index);
4098 assert(h->mmco_index==0);
4107 * retunrs the number of bytes consumed for building the current frame
4109 static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size){
4110 if(s->flags&CODEC_FLAG_TRUNCATED){
4111 pos -= s->parse_context.last_index;
4112 if(pos<0) pos=0; // FIXME remove (uneeded?)
4116 if(pos==0) pos=1; //avoid infinite loops (i doubt thats needed but ...)
4117 if(pos+10>buf_size) pos=buf_size; // oops ;)
4123 static int decode_frame(AVCodecContext *avctx,
4124 void *data, int *data_size,
4125 uint8_t *buf, int buf_size)
4127 H264Context *h = avctx->priv_data;
4128 MpegEncContext *s = &h->s;
4129 AVFrame *pict = data;
4132 s->flags= avctx->flags;
4136 /* no supplementary picture */
4137 if (buf_size == 0) {
4141 if(s->flags&CODEC_FLAG_TRUNCATED){
4142 int next= find_frame_end(s, buf, buf_size);
4144 if( ff_combine_frame(s, next, &buf, &buf_size) < 0 )
4146 //printf("next:%d buf_size:%d last_index:%d\n", next, buf_size, s->parse_context.last_index);
4149 if(s->avctx->extradata_size && s->picture_number==0){
4150 if(0 < decode_nal_units(h, s->avctx->extradata, s->avctx->extradata_size) )
4154 buf_index=decode_nal_units(h, buf, buf_size);
4158 //FIXME do something with unavailable reference frames
4160 // if(ret==FRAME_SKIPED) return get_consumed_bytes(s, buf_index, buf_size);
4162 if(s->pict_type==B_TYPE || s->low_delay){
4163 *pict= *(AVFrame*)&s->current_picture;
4165 *pict= *(AVFrame*)&s->last_picture;
4168 if(!s->current_picture_ptr){
4169 av_log(h->s.avctx, AV_LOG_DEBUG, "error, NO frame\n");
4173 *pict= *(AVFrame*)&s->current_picture; //FIXME
4174 ff_print_debug_info(s, s->current_picture_ptr);
4175 assert(pict->data[0]);
4176 //printf("out %d\n", (int)pict->data[0]);
4179 /* Return the Picture timestamp as the frame number */
4180 /* we substract 1 because it is added on utils.c */
4181 avctx->frame_number = s->picture_number - 1;
4184 /* dont output the last pic after seeking */
4185 if(s->last_picture_ptr || s->low_delay)
4186 //Note this isnt a issue as a IDR pic should flush teh buffers
4188 *data_size = sizeof(AVFrame);
4189 return get_consumed_bytes(s, buf_index, buf_size);
4192 static inline void fill_mb_avail(H264Context *h){
4193 MpegEncContext * const s = &h->s;
4194 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
4197 h->mb_avail[0]= s->mb_x && h->slice_table[mb_xy - s->mb_stride - 1] == h->slice_num;
4198 h->mb_avail[1]= h->slice_table[mb_xy - s->mb_stride ] == h->slice_num;
4199 h->mb_avail[2]= s->mb_x+1 < s->mb_width && h->slice_table[mb_xy - s->mb_stride + 1] == h->slice_num;
4205 h->mb_avail[3]= s->mb_x && h->slice_table[mb_xy - 1] == h->slice_num;
4206 h->mb_avail[4]= 1; //FIXME move out
4207 h->mb_avail[5]= 0; //FIXME move out
4213 #define SIZE (COUNT*40)
4219 // int int_temp[10000];
4221 AVCodecContext avctx;
4223 dsputil_init(&dsp, &avctx);
4225 init_put_bits(&pb, temp, SIZE);
4226 printf("testing unsigned exp golomb\n");
4227 for(i=0; i<COUNT; i++){
4229 set_ue_golomb(&pb, i);
4230 STOP_TIMER("set_ue_golomb");
4232 flush_put_bits(&pb);
4234 init_get_bits(&gb, temp, 8*SIZE);
4235 for(i=0; i<COUNT; i++){
4238 s= show_bits(&gb, 24);
4241 j= get_ue_golomb(&gb);
4243 printf("missmatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
4246 STOP_TIMER("get_ue_golomb");
4250 init_put_bits(&pb, temp, SIZE);
4251 printf("testing signed exp golomb\n");
4252 for(i=0; i<COUNT; i++){
4254 set_se_golomb(&pb, i - COUNT/2);
4255 STOP_TIMER("set_se_golomb");
4257 flush_put_bits(&pb);
4259 init_get_bits(&gb, temp, 8*SIZE);
4260 for(i=0; i<COUNT; i++){
4263 s= show_bits(&gb, 24);
4266 j= get_se_golomb(&gb);
4267 if(j != i - COUNT/2){
4268 printf("missmatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
4271 STOP_TIMER("get_se_golomb");
4274 printf("testing 4x4 (I)DCT\n");
4277 uint8_t src[16], ref[16];
4278 uint64_t error= 0, max_error=0;
4280 for(i=0; i<COUNT; i++){
4282 // printf("%d %d %d\n", r1, r2, (r2-r1)*16);
4283 for(j=0; j<16; j++){
4284 ref[j]= random()%255;
4285 src[j]= random()%255;
4288 h264_diff_dct_c(block, src, ref, 4);
4291 for(j=0; j<16; j++){
4292 // printf("%d ", block[j]);
4293 block[j]= block[j]*4;
4294 if(j&1) block[j]= (block[j]*4 + 2)/5;
4295 if(j&4) block[j]= (block[j]*4 + 2)/5;
4299 h264_add_idct_c(ref, block, 4);
4300 /* for(j=0; j<16; j++){
4301 printf("%d ", ref[j]);
4305 for(j=0; j<16; j++){
4306 int diff= ABS(src[j] - ref[j]);
4309 max_error= FFMAX(max_error, diff);
4312 printf("error=%f max_error=%d\n", ((float)error)/COUNT/16, (int)max_error );
4314 printf("testing quantizer\n");
4315 for(qp=0; qp<52; qp++){
4317 src1_block[i]= src2_block[i]= random()%255;
4321 printf("Testing NAL layer\n");
4323 uint8_t bitstream[COUNT];
4324 uint8_t nal[COUNT*2];
4326 memset(&h, 0, sizeof(H264Context));
4328 for(i=0; i<COUNT; i++){
4336 for(j=0; j<COUNT; j++){
4337 bitstream[j]= (random() % 255) + 1;
4340 for(j=0; j<zeros; j++){
4341 int pos= random() % COUNT;
4342 while(bitstream[pos] == 0){
4351 nal_length= encode_nal(&h, nal, bitstream, COUNT, COUNT*2);
4353 printf("encoding failed\n");
4357 out= decode_nal(&h, nal, &out_length, &consumed, nal_length);
4361 if(out_length != COUNT){
4362 printf("incorrect length %d %d\n", out_length, COUNT);
4366 if(consumed != nal_length){
4367 printf("incorrect consumed length %d %d\n", nal_length, consumed);
4371 if(memcmp(bitstream, out, COUNT)){
4372 printf("missmatch\n");
4377 printf("Testing RBSP\n");
4385 static int decode_end(AVCodecContext *avctx)
4387 H264Context *h = avctx->priv_data;
4388 MpegEncContext *s = &h->s;
4390 free_tables(h); //FIXME cleanup init stuff perhaps
4393 // memset(h, 0, sizeof(H264Context));
4399 AVCodec h264_decoder = {
4403 sizeof(H264Context),
4408 /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 | CODEC_CAP_TRUNCATED,