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"
39 #define interlaced_dct interlaced_dct_is_a_bad_name
40 #define mb_intra mb_intra_isnt_initalized_see_mb_type
42 #define LUMA_DC_BLOCK_INDEX 25
43 #define CHROMA_DC_BLOCK_INDEX 26
45 #define CHROMA_DC_COEFF_TOKEN_VLC_BITS 8
46 #define COEFF_TOKEN_VLC_BITS 8
47 #define TOTAL_ZEROS_VLC_BITS 9
48 #define CHROMA_DC_TOTAL_ZEROS_VLC_BITS 3
49 #define RUN_VLC_BITS 3
50 #define RUN7_VLC_BITS 6
52 #define MAX_SPS_COUNT 32
53 #define MAX_PPS_COUNT 256
55 #define MAX_MMCO_COUNT 66
58 * Sequence parameter set
64 int log2_max_frame_num; ///< log2_max_frame_num_minus4 + 4
65 int poc_type; ///< pic_order_cnt_type
66 int log2_max_poc_lsb; ///< log2_max_pic_order_cnt_lsb_minus4
67 int delta_pic_order_always_zero_flag;
68 int offset_for_non_ref_pic;
69 int offset_for_top_to_bottom_field;
70 int poc_cycle_length; ///< num_ref_frames_in_pic_order_cnt_cycle
71 int ref_frame_count; ///< num_ref_frames
72 int gaps_in_frame_num_allowed_flag;
73 int mb_width; ///< frame_width_in_mbs_minus1 + 1
74 int mb_height; ///< frame_height_in_mbs_minus1 + 1
75 int frame_mbs_only_flag;
76 int mb_aff; ///<mb_adaptive_frame_field_flag
77 int direct_8x8_inference_flag;
78 int crop; ///< frame_cropping_flag
79 int crop_left; ///< frame_cropping_rect_left_offset
80 int crop_right; ///< frame_cropping_rect_right_offset
81 int crop_top; ///< frame_cropping_rect_top_offset
82 int crop_bottom; ///< frame_cropping_rect_bottom_offset
83 int vui_parameters_present_flag;
85 int timing_info_present_flag;
86 uint32_t num_units_in_tick;
88 int fixed_frame_rate_flag;
89 short offset_for_ref_frame[256]; //FIXME dyn aloc?
93 * Picture parameter set
97 int cabac; ///< entropy_coding_mode_flag
98 int pic_order_present; ///< pic_order_present_flag
99 int slice_group_count; ///< num_slice_groups_minus1 + 1
100 int mb_slice_group_map_type;
101 int ref_count[2]; ///< num_ref_idx_l0/1_active_minus1 + 1
102 int weighted_pred; ///< weighted_pred_flag
103 int weighted_bipred_idc;
104 int init_qp; ///< pic_init_qp_minus26 + 26
105 int init_qs; ///< pic_init_qs_minus26 + 26
106 int chroma_qp_index_offset;
107 int deblocking_filter_parameters_present; ///< deblocking_filter_parameters_present_flag
108 int constrained_intra_pred; ///< constrained_intra_pred_flag
109 int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag
113 * Memory management control operation opcode.
115 typedef enum MMCOOpcode{
126 * Memory management control operation.
137 typedef struct H264Context{
145 #define NAL_IDR_SLICE 5
149 #define NAL_PICTURE_DELIMITER 9
150 #define NAL_FILTER_DATA 10
151 uint8_t *rbsp_buffer;
152 int rbsp_buffer_size;
155 * Used to parse AVC variant of h264
157 int is_avc; ///< this flag is != 0 if codec is avc1
158 int got_avcC; ///< flag used to parse avcC data only once
159 int nal_length_size; ///< Number of bytes used for nal length (1, 2 or 4)
163 int prev_mb_skiped; //FIXME remove (IMHO not used)
166 int chroma_pred_mode;
167 int intra16x16_pred_mode;
169 int8_t intra4x4_pred_mode_cache[5*8];
170 int8_t (*intra4x4_pred_mode)[8];
171 void (*pred4x4 [9+3])(uint8_t *src, uint8_t *topright, int stride);//FIXME move to dsp?
172 void (*pred8x8 [4+3])(uint8_t *src, int stride);
173 void (*pred16x16[4+3])(uint8_t *src, int stride);
174 unsigned int topleft_samples_available;
175 unsigned int top_samples_available;
176 unsigned int topright_samples_available;
177 unsigned int left_samples_available;
178 uint8_t (*top_border)[16+2*8];
179 uint8_t left_border[17+2*9];
182 * non zero coeff count cache.
183 * is 64 if not available.
185 uint8_t non_zero_count_cache[6*8];
186 uint8_t (*non_zero_count)[16];
189 * Motion vector cache.
191 int16_t mv_cache[2][5*8][2];
192 int8_t ref_cache[2][5*8];
193 #define LIST_NOT_USED -1 //FIXME rename?
194 #define PART_NOT_AVAILABLE -2
197 * is 1 if the specific list MV&references are set to 0,0,-2.
199 int mv_cache_clean[2];
201 int block_offset[16+8];
202 int chroma_subblock_offset[16]; //FIXME remove
204 uint16_t *mb2b_xy; //FIXME are these 4 a good idea?
212 int unknown_svq3_flag;
213 int next_slice_index;
215 SPS sps_buffer[MAX_SPS_COUNT];
216 SPS sps; ///< current sps
218 PPS pps_buffer[MAX_PPS_COUNT];
222 PPS pps; //FIXME move tp Picture perhaps? (->no) do we need that?
225 uint8_t *slice_table_base;
226 uint8_t *slice_table; ///< slice_table_base + mb_stride + 1
228 int slice_type_fixed;
230 //interlacing specific flags
231 int mb_field_decoding_flag;
238 int delta_poc_bottom;
241 int prev_poc_msb; ///< poc_msb of the last reference pic for POC type 0
242 int prev_poc_lsb; ///< poc_lsb of the last reference pic for POC type 0
243 int frame_num_offset; ///< for POC type 2
244 int prev_frame_num_offset; ///< for POC type 2
245 int prev_frame_num; ///< frame_num of the last pic for POC type 1/2
248 * frame_num for frames or 2*frame_num for field pics.
253 * max_frame_num or 2*max_frame_num for field pics.
257 //Weighted pred stuff
258 int luma_log2_weight_denom;
259 int chroma_log2_weight_denom;
260 int luma_weight[2][16];
261 int luma_offset[2][16];
262 int chroma_weight[2][16][2];
263 int chroma_offset[2][16][2];
266 int deblocking_filter; ///< disable_deblocking_filter_idc with 1<->0
267 int slice_alpha_c0_offset;
268 int slice_beta_offset;
270 int redundant_pic_count;
272 int direct_spatial_mv_pred;
275 * num_ref_idx_l0/1_active_minus1 + 1
277 int ref_count[2];// FIXME split for AFF
278 Picture *short_ref[16];
279 Picture *long_ref[16];
280 Picture default_ref_list[2][32];
281 Picture ref_list[2][32]; //FIXME size?
282 Picture field_ref_list[2][32]; //FIXME size?
285 * memory management control operations buffer.
287 MMCO mmco[MAX_MMCO_COUNT];
290 int long_ref_count; ///< number of actual long term references
291 int short_ref_count; ///< number of actual short term references
294 GetBitContext intra_gb;
295 GetBitContext inter_gb;
296 GetBitContext *intra_gb_ptr;
297 GetBitContext *inter_gb_ptr;
299 DCTELEM mb[16*24] __align8;
305 uint8_t cabac_state[399];
308 /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0,1,2), 0x0? luma_cbp */
312 /* chroma_pred_mode for i4x4 or i16x16, else 0 */
313 uint8_t *chroma_pred_mode_table;
314 int last_qscale_diff;
315 int16_t (*mvd_table[2])[2];
316 int16_t mvd_cache[2][5*8][2];
320 static VLC coeff_token_vlc[4];
321 static VLC chroma_dc_coeff_token_vlc;
323 static VLC total_zeros_vlc[15];
324 static VLC chroma_dc_total_zeros_vlc[3];
326 static VLC run_vlc[6];
329 static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
330 static void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
331 static void filter_mb( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr);
333 static inline uint32_t pack16to32(int a, int b){
334 #ifdef WORDS_BIGENDIAN
335 return (b&0xFFFF) + (a<<16);
337 return (a&0xFFFF) + (b<<16);
343 * @param h height of the recatangle, should be a constant
344 * @param w width of the recatangle, should be a constant
345 * @param size the size of val (1 or 4), should be a constant
347 static inline void fill_rectangle(void *vp, int w, int h, int stride, uint32_t val, int size){ //FIXME ensure this IS inlined
348 uint8_t *p= (uint8_t*)vp;
349 assert(size==1 || size==4);
354 //FIXME check what gcc generates for 64 bit on x86 and possible write a 32 bit ver of it
357 *(uint16_t*)(p + stride)= size==4 ? val : val*0x0101;
358 }else if(w==2 && h==4){
359 *(uint16_t*)(p + 0*stride)=
360 *(uint16_t*)(p + 1*stride)=
361 *(uint16_t*)(p + 2*stride)=
362 *(uint16_t*)(p + 3*stride)= size==4 ? val : val*0x0101;
363 }else if(w==4 && h==1){
364 *(uint32_t*)(p + 0*stride)= size==4 ? val : val*0x01010101;
365 }else if(w==4 && h==2){
366 *(uint32_t*)(p + 0*stride)=
367 *(uint32_t*)(p + 1*stride)= size==4 ? val : val*0x01010101;
368 }else if(w==4 && h==4){
369 *(uint32_t*)(p + 0*stride)=
370 *(uint32_t*)(p + 1*stride)=
371 *(uint32_t*)(p + 2*stride)=
372 *(uint32_t*)(p + 3*stride)= size==4 ? val : val*0x01010101;
373 }else if(w==8 && h==1){
375 *(uint32_t*)(p + 4)= size==4 ? val : val*0x01010101;
376 }else if(w==8 && h==2){
377 *(uint32_t*)(p + 0 + 0*stride)=
378 *(uint32_t*)(p + 4 + 0*stride)=
379 *(uint32_t*)(p + 0 + 1*stride)=
380 *(uint32_t*)(p + 4 + 1*stride)= size==4 ? val : val*0x01010101;
381 }else if(w==8 && h==4){
382 *(uint64_t*)(p + 0*stride)=
383 *(uint64_t*)(p + 1*stride)=
384 *(uint64_t*)(p + 2*stride)=
385 *(uint64_t*)(p + 3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
386 }else if(w==16 && h==2){
387 *(uint64_t*)(p + 0+0*stride)=
388 *(uint64_t*)(p + 8+0*stride)=
389 *(uint64_t*)(p + 0+1*stride)=
390 *(uint64_t*)(p + 8+1*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
391 }else if(w==16 && h==4){
392 *(uint64_t*)(p + 0+0*stride)=
393 *(uint64_t*)(p + 8+0*stride)=
394 *(uint64_t*)(p + 0+1*stride)=
395 *(uint64_t*)(p + 8+1*stride)=
396 *(uint64_t*)(p + 0+2*stride)=
397 *(uint64_t*)(p + 8+2*stride)=
398 *(uint64_t*)(p + 0+3*stride)=
399 *(uint64_t*)(p + 8+3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
404 static inline void fill_caches(H264Context *h, int mb_type){
405 MpegEncContext * const s = &h->s;
406 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
407 int topleft_xy, top_xy, topright_xy, left_xy[2];
408 int topleft_type, top_type, topright_type, left_type[2];
412 //wow what a mess, why didnt they simplify the interlacing&intra stuff, i cant imagine that these complex rules are worth it
416 topleft_xy = 0; /* avoid warning */
417 top_xy = 0; /* avoid warning */
418 topright_xy = 0; /* avoid warning */
420 topleft_xy = mb_xy-1 - s->mb_stride;
421 top_xy = mb_xy - s->mb_stride;
422 topright_xy= mb_xy+1 - s->mb_stride;
423 left_xy[0] = mb_xy-1;
424 left_xy[1] = mb_xy-1;
431 topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
432 top_type = h->slice_table[top_xy ] == h->slice_num ? s->current_picture.mb_type[top_xy] : 0;
433 topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
434 left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
435 left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
437 if(IS_INTRA(mb_type)){
438 h->topleft_samples_available=
439 h->top_samples_available=
440 h->left_samples_available= 0xFFFF;
441 h->topright_samples_available= 0xEEEA;
443 if(!IS_INTRA(top_type) && (top_type==0 || h->pps.constrained_intra_pred)){
444 h->topleft_samples_available= 0xB3FF;
445 h->top_samples_available= 0x33FF;
446 h->topright_samples_available= 0x26EA;
449 if(!IS_INTRA(left_type[i]) && (left_type[i]==0 || h->pps.constrained_intra_pred)){
450 h->topleft_samples_available&= 0xDF5F;
451 h->left_samples_available&= 0x5F5F;
455 if(!IS_INTRA(topleft_type) && (topleft_type==0 || h->pps.constrained_intra_pred))
456 h->topleft_samples_available&= 0x7FFF;
458 if(!IS_INTRA(topright_type) && (topright_type==0 || h->pps.constrained_intra_pred))
459 h->topright_samples_available&= 0xFBFF;
461 if(IS_INTRA4x4(mb_type)){
462 if(IS_INTRA4x4(top_type)){
463 h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
464 h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
465 h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
466 h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
469 if(IS_INTRA16x16(top_type) || (IS_INTER(top_type) && !h->pps.constrained_intra_pred))
474 h->intra4x4_pred_mode_cache[4+8*0]=
475 h->intra4x4_pred_mode_cache[5+8*0]=
476 h->intra4x4_pred_mode_cache[6+8*0]=
477 h->intra4x4_pred_mode_cache[7+8*0]= pred;
480 if(IS_INTRA4x4(left_type[i])){
481 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
482 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
485 if(IS_INTRA16x16(left_type[i]) || (IS_INTER(left_type[i]) && !h->pps.constrained_intra_pred))
490 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
491 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
506 //FIXME constraint_intra_pred & partitioning & nnz (lets hope this is just a typo in the spec)
508 h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][0];
509 h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][1];
510 h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][2];
511 h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3];
513 h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][7];
514 h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8];
516 h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][10];
517 h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11];
519 h->top_cbp= h->cbp_table[top_xy];
521 h->non_zero_count_cache[4+8*0]=
522 h->non_zero_count_cache[5+8*0]=
523 h->non_zero_count_cache[6+8*0]=
524 h->non_zero_count_cache[7+8*0]=
526 h->non_zero_count_cache[1+8*0]=
527 h->non_zero_count_cache[2+8*0]=
529 h->non_zero_count_cache[1+8*3]=
530 h->non_zero_count_cache[2+8*3]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
532 if(IS_INTRA(mb_type)) h->top_cbp= 0x1C0;
537 h->non_zero_count_cache[3+8*1]= h->non_zero_count[left_xy[0]][6];
538 h->non_zero_count_cache[3+8*2]= h->non_zero_count[left_xy[0]][5];
539 h->non_zero_count_cache[0+8*1]= h->non_zero_count[left_xy[0]][9]; //FIXME left_block
540 h->non_zero_count_cache[0+8*4]= h->non_zero_count[left_xy[0]][12];
541 h->left_cbp= h->cbp_table[left_xy[0]]; //FIXME interlacing
543 h->non_zero_count_cache[3+8*1]=
544 h->non_zero_count_cache[3+8*2]=
545 h->non_zero_count_cache[0+8*1]=
546 h->non_zero_count_cache[0+8*4]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
548 if(IS_INTRA(mb_type)) h->left_cbp= 0x1C0;//FIXME interlacing
553 h->non_zero_count_cache[3+8*3]= h->non_zero_count[left_xy[1]][4];
554 h->non_zero_count_cache[3+8*4]= h->non_zero_count[left_xy[1]][3];
555 h->non_zero_count_cache[0+8*2]= h->non_zero_count[left_xy[1]][8];
556 h->non_zero_count_cache[0+8*5]= h->non_zero_count[left_xy[1]][11];
558 h->non_zero_count_cache[3+8*3]=
559 h->non_zero_count_cache[3+8*4]=
560 h->non_zero_count_cache[0+8*2]=
561 h->non_zero_count_cache[0+8*5]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
565 if(IS_INTER(mb_type)){
567 for(list=0; list<2; list++){
568 if((!IS_8X8(mb_type)) && !USES_LIST(mb_type, list)){
569 /*if(!h->mv_cache_clean[list]){
570 memset(h->mv_cache [list], 0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
571 memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
572 h->mv_cache_clean[list]= 1;
574 continue; //FIXME direct mode ...
576 h->mv_cache_clean[list]= 0;
578 if(IS_INTER(topleft_type)){
579 const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
580 const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + h->b8_stride;
581 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
582 h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
584 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
585 h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
588 if(IS_INTER(top_type)){
589 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
590 const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
591 *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0];
592 *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];
593 *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];
594 *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];
595 h->ref_cache[list][scan8[0] + 0 - 1*8]=
596 h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
597 h->ref_cache[list][scan8[0] + 2 - 1*8]=
598 h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
600 *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]=
601 *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]=
602 *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]=
603 *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;
604 *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
607 if(IS_INTER(topright_type)){
608 const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
609 const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
610 *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
611 h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
613 *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
614 h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
617 //FIXME unify cleanup or sth
618 if(IS_INTER(left_type[0])){
619 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
620 const int b8_xy= h->mb2b8_xy[left_xy[0]] + 1;
621 *(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]];
622 *(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]];
623 h->ref_cache[list][scan8[0] - 1 + 0*8]=
624 h->ref_cache[list][scan8[0] - 1 + 1*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0]>>1)];
626 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 0*8]=
627 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 1*8]= 0;
628 h->ref_cache[list][scan8[0] - 1 + 0*8]=
629 h->ref_cache[list][scan8[0] - 1 + 1*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
632 if(IS_INTER(left_type[1])){
633 const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
634 const int b8_xy= h->mb2b8_xy[left_xy[1]] + 1;
635 *(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]];
636 *(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]];
637 h->ref_cache[list][scan8[0] - 1 + 2*8]=
638 h->ref_cache[list][scan8[0] - 1 + 3*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[2]>>1)];
640 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 2*8]=
641 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 3*8]= 0;
642 h->ref_cache[list][scan8[0] - 1 + 2*8]=
643 h->ref_cache[list][scan8[0] - 1 + 3*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
646 h->ref_cache[list][scan8[5 ]+1] =
647 h->ref_cache[list][scan8[7 ]+1] =
648 h->ref_cache[list][scan8[13]+1] = //FIXME remove past 3 (init somewher else)
649 h->ref_cache[list][scan8[4 ]] =
650 h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
651 *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
652 *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
653 *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewher else)
654 *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
655 *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
658 /* XXX beurk, Load mvd */
659 if(IS_INTER(topleft_type)){
660 const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
661 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy];
663 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 - 1*8]= 0;
666 if(IS_INTER(top_type)){
667 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
668 *(uint32_t*)h->mvd_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 0];
669 *(uint32_t*)h->mvd_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 1];
670 *(uint32_t*)h->mvd_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 2];
671 *(uint32_t*)h->mvd_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 3];
673 *(uint32_t*)h->mvd_cache [list][scan8[0] + 0 - 1*8]=
674 *(uint32_t*)h->mvd_cache [list][scan8[0] + 1 - 1*8]=
675 *(uint32_t*)h->mvd_cache [list][scan8[0] + 2 - 1*8]=
676 *(uint32_t*)h->mvd_cache [list][scan8[0] + 3 - 1*8]= 0;
678 if(IS_INTER(left_type[0])){
679 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
680 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 0*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[0]];
681 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[1]];
683 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 0*8]=
684 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 1*8]= 0;
686 if(IS_INTER(left_type[1])){
687 const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
688 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 2*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[2]];
689 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 3*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[3]];
691 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 2*8]=
692 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 3*8]= 0;
694 *(uint32_t*)h->mvd_cache [list][scan8[5 ]+1]=
695 *(uint32_t*)h->mvd_cache [list][scan8[7 ]+1]=
696 *(uint32_t*)h->mvd_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewher else)
697 *(uint32_t*)h->mvd_cache [list][scan8[4 ]]=
698 *(uint32_t*)h->mvd_cache [list][scan8[12]]= 0;
706 static inline void write_back_intra_pred_mode(H264Context *h){
707 MpegEncContext * const s = &h->s;
708 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
710 h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
711 h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
712 h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
713 h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
714 h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
715 h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
716 h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
720 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
722 static inline int check_intra4x4_pred_mode(H264Context *h){
723 MpegEncContext * const s = &h->s;
724 static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
725 static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
728 if(!(h->top_samples_available&0x8000)){
730 int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
732 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);
735 h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
740 if(!(h->left_samples_available&0x8000)){
742 int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
744 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);
747 h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
753 } //FIXME cleanup like next
756 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
758 static inline int check_intra_pred_mode(H264Context *h, int mode){
759 MpegEncContext * const s = &h->s;
760 static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
761 static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
763 if(mode < 0 || mode > 6)
766 if(!(h->top_samples_available&0x8000)){
769 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);
774 if(!(h->left_samples_available&0x8000)){
777 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);
786 * gets the predicted intra4x4 prediction mode.
788 static inline int pred_intra_mode(H264Context *h, int n){
789 const int index8= scan8[n];
790 const int left= h->intra4x4_pred_mode_cache[index8 - 1];
791 const int top = h->intra4x4_pred_mode_cache[index8 - 8];
792 const int min= FFMIN(left, top);
794 tprintf("mode:%d %d min:%d\n", left ,top, min);
796 if(min<0) return DC_PRED;
800 static inline void write_back_non_zero_count(H264Context *h){
801 MpegEncContext * const s = &h->s;
802 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
804 h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[4+8*4];
805 h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[5+8*4];
806 h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[6+8*4];
807 h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];
808 h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[7+8*3];
809 h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[7+8*2];
810 h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[7+8*1];
812 h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[1+8*2];
813 h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
814 h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[2+8*1];
816 h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[1+8*5];
817 h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
818 h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[2+8*4];
822 * gets the predicted number of non zero coefficients.
823 * @param n block index
825 static inline int pred_non_zero_count(H264Context *h, int n){
826 const int index8= scan8[n];
827 const int left= h->non_zero_count_cache[index8 - 1];
828 const int top = h->non_zero_count_cache[index8 - 8];
831 if(i<64) i= (i+1)>>1;
833 tprintf("pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
838 static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
839 const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];
841 if(topright_ref != PART_NOT_AVAILABLE){
842 *C= h->mv_cache[list][ i - 8 + part_width ];
845 tprintf("topright MV not available\n");
847 *C= h->mv_cache[list][ i - 8 - 1 ];
848 return h->ref_cache[list][ i - 8 - 1 ];
853 * gets the predicted MV.
854 * @param n the block index
855 * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)
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_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
860 const int index8= scan8[n];
861 const int top_ref= h->ref_cache[list][ index8 - 8 ];
862 const int left_ref= h->ref_cache[list][ index8 - 1 ];
863 const int16_t * const A= h->mv_cache[list][ index8 - 1 ];
864 const int16_t * const B= h->mv_cache[list][ index8 - 8 ];
866 int diagonal_ref, match_count;
868 assert(part_width==1 || part_width==2 || part_width==4);
878 diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);
879 match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);
880 if(match_count > 1){ //most common
881 *mx= mid_pred(A[0], B[0], C[0]);
882 *my= mid_pred(A[1], B[1], C[1]);
883 }else if(match_count==1){
887 }else if(top_ref==ref){
895 if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
899 *mx= mid_pred(A[0], B[0], C[0]);
900 *my= mid_pred(A[1], B[1], C[1]);
904 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);
908 * gets the directionally predicted 16x8 MV.
909 * @param n the block index
910 * @param mx the x component of the predicted motion vector
911 * @param my the y component of the predicted motion vector
913 static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
915 const int top_ref= h->ref_cache[list][ scan8[0] - 8 ];
916 const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
918 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);
926 const int left_ref= h->ref_cache[list][ scan8[8] - 1 ];
927 const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];
929 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);
939 pred_motion(h, n, 4, list, ref, mx, my);
943 * gets the directionally predicted 8x16 MV.
944 * @param n the block index
945 * @param mx the x component of the predicted motion vector
946 * @param my the y component of the predicted motion vector
948 static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
950 const int left_ref= h->ref_cache[list][ scan8[0] - 1 ];
951 const int16_t * const A= h->mv_cache[list][ scan8[0] - 1 ];
953 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);
964 diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);
966 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);
968 if(diagonal_ref == ref){
976 pred_motion(h, n, 2, list, ref, mx, my);
979 static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
980 const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
981 const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];
983 tprintf("pred_pskip: (%d) (%d) at %2d %2d", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
985 if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
986 || (top_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0)
987 || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){
993 pred_motion(h, 0, 4, 0, 0, mx, my);
998 static inline void write_back_motion(H264Context *h, int mb_type){
999 MpegEncContext * const s = &h->s;
1000 const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1001 const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1004 for(list=0; list<2; list++){
1006 if((!IS_8X8(mb_type)) && !USES_LIST(mb_type, list)){
1007 if(1){ //FIXME skip or never read if mb_type doesnt use it
1009 *(uint64_t*)s->current_picture.motion_val[list][b_xy + 0 + y*h->b_stride]=
1010 *(uint64_t*)s->current_picture.motion_val[list][b_xy + 2 + y*h->b_stride]= 0;
1012 if( h->pps.cabac ) {
1013 /* FIXME needed ? */
1015 *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]=
1016 *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= 0;
1020 *(uint16_t*)s->current_picture.motion_val[list][b8_xy + y*h->b8_stride]= (LIST_NOT_USED&0xFF)*0x0101;
1023 continue; //FIXME direct mode ...
1027 *(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];
1028 *(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];
1030 if( h->pps.cabac ) {
1032 *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+0 + 8*y];
1033 *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+2 + 8*y];
1037 s->current_picture.ref_index[list][b8_xy + 0 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+0 + 16*y];
1038 s->current_picture.ref_index[list][b8_xy + 1 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+2 + 16*y];
1044 * Decodes a network abstraction layer unit.
1045 * @param consumed is the number of bytes used as input
1046 * @param length is the length of the array
1047 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp ttailing?
1048 * @returns decoded bytes, might be src+1 if no escapes
1050 static uint8_t *decode_nal(H264Context *h, uint8_t *src, int *dst_length, int *consumed, int length){
1054 // src[0]&0x80; //forbidden bit
1055 h->nal_ref_idc= src[0]>>5;
1056 h->nal_unit_type= src[0]&0x1F;
1060 for(i=0; i<length; i++)
1061 printf("%2X ", src[i]);
1063 for(i=0; i+1<length; i+=2){
1064 if(src[i]) continue;
1065 if(i>0 && src[i-1]==0) i--;
1066 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1068 /* startcode, so we must be past the end */
1075 if(i>=length-1){ //no escaped 0
1076 *dst_length= length;
1077 *consumed= length+1; //+1 for the header
1081 h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length);
1082 dst= h->rbsp_buffer;
1084 //printf("deoding esc\n");
1087 //remove escapes (very rare 1:2^22)
1088 if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1089 if(src[si+2]==3){ //escape
1094 }else //next start code
1098 dst[di++]= src[si++];
1102 *consumed= si + 1;//+1 for the header
1103 //FIXME store exact number of bits in the getbitcontext (its needed for decoding)
1109 * @param src the data which should be escaped
1110 * @param dst the target buffer, dst+1 == src is allowed as a special case
1111 * @param length the length of the src data
1112 * @param dst_length the length of the dst array
1113 * @returns length of escaped data in bytes or -1 if an error occured
1115 static int encode_nal(H264Context *h, uint8_t *dst, uint8_t *src, int length, int dst_length){
1116 int i, escape_count, si, di;
1120 assert(dst_length>0);
1122 dst[0]= (h->nal_ref_idc<<5) + h->nal_unit_type;
1124 if(length==0) return 1;
1127 for(i=0; i<length; i+=2){
1128 if(src[i]) continue;
1129 if(i>0 && src[i-1]==0)
1131 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1137 if(escape_count==0){
1139 memcpy(dst+1, src, length);
1143 if(length + escape_count + 1> dst_length)
1146 //this should be damn rare (hopefully)
1148 h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length + escape_count);
1149 temp= h->rbsp_buffer;
1150 //printf("encoding esc\n");
1155 if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1156 temp[di++]= 0; si++;
1157 temp[di++]= 0; si++;
1159 temp[di++]= src[si++];
1162 temp[di++]= src[si++];
1164 memcpy(dst+1, temp, length+escape_count);
1166 assert(di == length+escape_count);
1172 * write 1,10,100,1000,... for alignment, yes its exactly inverse to mpeg4
1174 static void encode_rbsp_trailing(PutBitContext *pb){
1177 length= (-put_bits_count(pb))&7;
1178 if(length) put_bits(pb, length, 0);
1183 * identifies the exact end of the bitstream
1184 * @return the length of the trailing, or 0 if damaged
1186 static int decode_rbsp_trailing(uint8_t *src){
1190 tprintf("rbsp trailing %X\n", v);
1200 * idct tranforms the 16 dc values and dequantize them.
1201 * @param qp quantization parameter
1203 static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp){
1204 const int qmul= dequant_coeff[qp][0];
1207 int temp[16]; //FIXME check if this is a good idea
1208 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
1209 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1211 //memset(block, 64, 2*256);
1214 const int offset= y_offset[i];
1215 const int z0= block[offset+stride*0] + block[offset+stride*4];
1216 const int z1= block[offset+stride*0] - block[offset+stride*4];
1217 const int z2= block[offset+stride*1] - block[offset+stride*5];
1218 const int z3= block[offset+stride*1] + block[offset+stride*5];
1227 const int offset= x_offset[i];
1228 const int z0= temp[4*0+i] + temp[4*2+i];
1229 const int z1= temp[4*0+i] - temp[4*2+i];
1230 const int z2= temp[4*1+i] - temp[4*3+i];
1231 const int z3= temp[4*1+i] + temp[4*3+i];
1233 block[stride*0 +offset]= ((z0 + z3)*qmul + 2)>>2; //FIXME think about merging this into decode_resdual
1234 block[stride*2 +offset]= ((z1 + z2)*qmul + 2)>>2;
1235 block[stride*8 +offset]= ((z1 - z2)*qmul + 2)>>2;
1236 block[stride*10+offset]= ((z0 - z3)*qmul + 2)>>2;
1242 * dct tranforms the 16 dc values.
1243 * @param qp quantization parameter ??? FIXME
1245 static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
1246 // const int qmul= dequant_coeff[qp][0];
1248 int temp[16]; //FIXME check if this is a good idea
1249 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
1250 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1253 const int offset= y_offset[i];
1254 const int z0= block[offset+stride*0] + block[offset+stride*4];
1255 const int z1= block[offset+stride*0] - block[offset+stride*4];
1256 const int z2= block[offset+stride*1] - block[offset+stride*5];
1257 const int z3= block[offset+stride*1] + block[offset+stride*5];
1266 const int offset= x_offset[i];
1267 const int z0= temp[4*0+i] + temp[4*2+i];
1268 const int z1= temp[4*0+i] - temp[4*2+i];
1269 const int z2= temp[4*1+i] - temp[4*3+i];
1270 const int z3= temp[4*1+i] + temp[4*3+i];
1272 block[stride*0 +offset]= (z0 + z3)>>1;
1273 block[stride*2 +offset]= (z1 + z2)>>1;
1274 block[stride*8 +offset]= (z1 - z2)>>1;
1275 block[stride*10+offset]= (z0 - z3)>>1;
1283 static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp){
1284 const int qmul= dequant_coeff[qp][0];
1285 const int stride= 16*2;
1286 const int xStride= 16;
1289 a= block[stride*0 + xStride*0];
1290 b= block[stride*0 + xStride*1];
1291 c= block[stride*1 + xStride*0];
1292 d= block[stride*1 + xStride*1];
1299 block[stride*0 + xStride*0]= ((a+c)*qmul + 0)>>1;
1300 block[stride*0 + xStride*1]= ((e+b)*qmul + 0)>>1;
1301 block[stride*1 + xStride*0]= ((a-c)*qmul + 0)>>1;
1302 block[stride*1 + xStride*1]= ((e-b)*qmul + 0)>>1;
1306 static void chroma_dc_dct_c(DCTELEM *block){
1307 const int stride= 16*2;
1308 const int xStride= 16;
1311 a= block[stride*0 + xStride*0];
1312 b= block[stride*0 + xStride*1];
1313 c= block[stride*1 + xStride*0];
1314 d= block[stride*1 + xStride*1];
1321 block[stride*0 + xStride*0]= (a+c);
1322 block[stride*0 + xStride*1]= (e+b);
1323 block[stride*1 + xStride*0]= (a-c);
1324 block[stride*1 + xStride*1]= (e-b);
1329 * gets the chroma qp.
1331 static inline int get_chroma_qp(H264Context *h, int qscale){
1333 return chroma_qp[clip(qscale + h->pps.chroma_qp_index_offset, 0, 51)];
1338 static void h264_diff_dct_c(DCTELEM *block, uint8_t *src1, uint8_t *src2, int stride){
1340 //FIXME try int temp instead of block
1343 const int d0= src1[0 + i*stride] - src2[0 + i*stride];
1344 const int d1= src1[1 + i*stride] - src2[1 + i*stride];
1345 const int d2= src1[2 + i*stride] - src2[2 + i*stride];
1346 const int d3= src1[3 + i*stride] - src2[3 + i*stride];
1347 const int z0= d0 + d3;
1348 const int z3= d0 - d3;
1349 const int z1= d1 + d2;
1350 const int z2= d1 - d2;
1352 block[0 + 4*i]= z0 + z1;
1353 block[1 + 4*i]= 2*z3 + z2;
1354 block[2 + 4*i]= z0 - z1;
1355 block[3 + 4*i]= z3 - 2*z2;
1359 const int z0= block[0*4 + i] + block[3*4 + i];
1360 const int z3= block[0*4 + i] - block[3*4 + i];
1361 const int z1= block[1*4 + i] + block[2*4 + i];
1362 const int z2= block[1*4 + i] - block[2*4 + i];
1364 block[0*4 + i]= z0 + z1;
1365 block[1*4 + i]= 2*z3 + z2;
1366 block[2*4 + i]= z0 - z1;
1367 block[3*4 + i]= z3 - 2*z2;
1372 //FIXME need to check that this doesnt overflow signed 32 bit for low qp, iam not sure, its very close
1373 //FIXME check that gcc inlines this (and optimizes intra & seperate_dc stuff away)
1374 static inline int quantize_c(DCTELEM *block, uint8_t *scantable, int qscale, int intra, int seperate_dc){
1376 const int * const quant_table= quant_coeff[qscale];
1377 const int bias= intra ? (1<<QUANT_SHIFT)/3 : (1<<QUANT_SHIFT)/6;
1378 const unsigned int threshold1= (1<<QUANT_SHIFT) - bias - 1;
1379 const unsigned int threshold2= (threshold1<<1);
1385 const int dc_bias= intra ? (1<<(QUANT_SHIFT-2))/3 : (1<<(QUANT_SHIFT-2))/6;
1386 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT-2)) - dc_bias - 1;
1387 const unsigned int dc_threshold2= (dc_threshold1<<1);
1389 int level= block[0]*quant_coeff[qscale+18][0];
1390 if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1392 level= (dc_bias + level)>>(QUANT_SHIFT-2);
1395 level= (dc_bias - level)>>(QUANT_SHIFT-2);
1398 // last_non_zero = i;
1403 const int dc_bias= intra ? (1<<(QUANT_SHIFT+1))/3 : (1<<(QUANT_SHIFT+1))/6;
1404 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT+1)) - dc_bias - 1;
1405 const unsigned int dc_threshold2= (dc_threshold1<<1);
1407 int level= block[0]*quant_table[0];
1408 if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1410 level= (dc_bias + level)>>(QUANT_SHIFT+1);
1413 level= (dc_bias - level)>>(QUANT_SHIFT+1);
1416 // last_non_zero = i;
1429 const int j= scantable[i];
1430 int level= block[j]*quant_table[j];
1432 // if( bias+level >= (1<<(QMAT_SHIFT - 3))
1433 // || bias-level >= (1<<(QMAT_SHIFT - 3))){
1434 if(((unsigned)(level+threshold1))>threshold2){
1436 level= (bias + level)>>QUANT_SHIFT;
1439 level= (bias - level)>>QUANT_SHIFT;
1448 return last_non_zero;
1451 static void pred4x4_vertical_c(uint8_t *src, uint8_t *topright, int stride){
1452 const uint32_t a= ((uint32_t*)(src-stride))[0];
1453 ((uint32_t*)(src+0*stride))[0]= a;
1454 ((uint32_t*)(src+1*stride))[0]= a;
1455 ((uint32_t*)(src+2*stride))[0]= a;
1456 ((uint32_t*)(src+3*stride))[0]= a;
1459 static void pred4x4_horizontal_c(uint8_t *src, uint8_t *topright, int stride){
1460 ((uint32_t*)(src+0*stride))[0]= src[-1+0*stride]*0x01010101;
1461 ((uint32_t*)(src+1*stride))[0]= src[-1+1*stride]*0x01010101;
1462 ((uint32_t*)(src+2*stride))[0]= src[-1+2*stride]*0x01010101;
1463 ((uint32_t*)(src+3*stride))[0]= src[-1+3*stride]*0x01010101;
1466 static void pred4x4_dc_c(uint8_t *src, uint8_t *topright, int stride){
1467 const int dc= ( src[-stride] + src[1-stride] + src[2-stride] + src[3-stride]
1468 + src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 4) >>3;
1470 ((uint32_t*)(src+0*stride))[0]=
1471 ((uint32_t*)(src+1*stride))[0]=
1472 ((uint32_t*)(src+2*stride))[0]=
1473 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1476 static void pred4x4_left_dc_c(uint8_t *src, uint8_t *topright, int stride){
1477 const int dc= ( src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 2) >>2;
1479 ((uint32_t*)(src+0*stride))[0]=
1480 ((uint32_t*)(src+1*stride))[0]=
1481 ((uint32_t*)(src+2*stride))[0]=
1482 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1485 static void pred4x4_top_dc_c(uint8_t *src, uint8_t *topright, int stride){
1486 const int dc= ( src[-stride] + src[1-stride] + src[2-stride] + src[3-stride] + 2) >>2;
1488 ((uint32_t*)(src+0*stride))[0]=
1489 ((uint32_t*)(src+1*stride))[0]=
1490 ((uint32_t*)(src+2*stride))[0]=
1491 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1494 static void pred4x4_128_dc_c(uint8_t *src, uint8_t *topright, int stride){
1495 ((uint32_t*)(src+0*stride))[0]=
1496 ((uint32_t*)(src+1*stride))[0]=
1497 ((uint32_t*)(src+2*stride))[0]=
1498 ((uint32_t*)(src+3*stride))[0]= 128U*0x01010101U;
1502 #define LOAD_TOP_RIGHT_EDGE\
1503 const int t4= topright[0];\
1504 const int t5= topright[1];\
1505 const int t6= topright[2];\
1506 const int t7= topright[3];\
1508 #define LOAD_LEFT_EDGE\
1509 const int l0= src[-1+0*stride];\
1510 const int l1= src[-1+1*stride];\
1511 const int l2= src[-1+2*stride];\
1512 const int l3= src[-1+3*stride];\
1514 #define LOAD_TOP_EDGE\
1515 const int t0= src[ 0-1*stride];\
1516 const int t1= src[ 1-1*stride];\
1517 const int t2= src[ 2-1*stride];\
1518 const int t3= src[ 3-1*stride];\
1520 static void pred4x4_down_right_c(uint8_t *src, uint8_t *topright, int stride){
1521 const int lt= src[-1-1*stride];
1525 src[0+3*stride]=(l3 + 2*l2 + l1 + 2)>>2;
1527 src[1+3*stride]=(l2 + 2*l1 + l0 + 2)>>2;
1530 src[2+3*stride]=(l1 + 2*l0 + lt + 2)>>2;
1534 src[3+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1537 src[3+2*stride]=(lt + 2*t0 + t1 + 2)>>2;
1539 src[3+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1540 src[3+0*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1543 static void pred4x4_down_left_c(uint8_t *src, uint8_t *topright, int stride){
1548 src[0+0*stride]=(t0 + t2 + 2*t1 + 2)>>2;
1550 src[0+1*stride]=(t1 + t3 + 2*t2 + 2)>>2;
1553 src[0+2*stride]=(t2 + t4 + 2*t3 + 2)>>2;
1557 src[0+3*stride]=(t3 + t5 + 2*t4 + 2)>>2;
1560 src[1+3*stride]=(t4 + t6 + 2*t5 + 2)>>2;
1562 src[2+3*stride]=(t5 + t7 + 2*t6 + 2)>>2;
1563 src[3+3*stride]=(t6 + 3*t7 + 2)>>2;
1566 static void pred4x4_vertical_right_c(uint8_t *src, uint8_t *topright, int stride){
1567 const int lt= src[-1-1*stride];
1570 const __attribute__((unused)) int unu= l3;
1573 src[1+2*stride]=(lt + t0 + 1)>>1;
1575 src[2+2*stride]=(t0 + t1 + 1)>>1;
1577 src[3+2*stride]=(t1 + t2 + 1)>>1;
1578 src[3+0*stride]=(t2 + t3 + 1)>>1;
1580 src[1+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1582 src[2+3*stride]=(lt + 2*t0 + t1 + 2)>>2;
1584 src[3+3*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1585 src[3+1*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1586 src[0+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1587 src[0+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1590 static void pred4x4_vertical_left_c(uint8_t *src, uint8_t *topright, int stride){
1593 const __attribute__((unused)) int unu= t7;
1595 src[0+0*stride]=(t0 + t1 + 1)>>1;
1597 src[0+2*stride]=(t1 + t2 + 1)>>1;
1599 src[1+2*stride]=(t2 + t3 + 1)>>1;
1601 src[2+2*stride]=(t3 + t4+ 1)>>1;
1602 src[3+2*stride]=(t4 + t5+ 1)>>1;
1603 src[0+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1605 src[0+3*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1607 src[1+3*stride]=(t2 + 2*t3 + t4 + 2)>>2;
1609 src[2+3*stride]=(t3 + 2*t4 + t5 + 2)>>2;
1610 src[3+3*stride]=(t4 + 2*t5 + t6 + 2)>>2;
1613 static void pred4x4_horizontal_up_c(uint8_t *src, uint8_t *topright, int stride){
1616 src[0+0*stride]=(l0 + l1 + 1)>>1;
1617 src[1+0*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1619 src[0+1*stride]=(l1 + l2 + 1)>>1;
1621 src[1+1*stride]=(l1 + 2*l2 + l3 + 2)>>2;
1623 src[0+2*stride]=(l2 + l3 + 1)>>1;
1625 src[1+2*stride]=(l2 + 2*l3 + l3 + 2)>>2;
1634 static void pred4x4_horizontal_down_c(uint8_t *src, uint8_t *topright, int stride){
1635 const int lt= src[-1-1*stride];
1638 const __attribute__((unused)) int unu= t3;
1641 src[2+1*stride]=(lt + l0 + 1)>>1;
1643 src[3+1*stride]=(l0 + 2*lt + t0 + 2)>>2;
1644 src[2+0*stride]=(lt + 2*t0 + t1 + 2)>>2;
1645 src[3+0*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1647 src[2+2*stride]=(l0 + l1 + 1)>>1;
1649 src[3+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1651 src[2+3*stride]=(l1 + l2+ 1)>>1;
1653 src[3+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1654 src[0+3*stride]=(l2 + l3 + 1)>>1;
1655 src[1+3*stride]=(l1 + 2*l2 + l3 + 2)>>2;
1658 static void pred16x16_vertical_c(uint8_t *src, int stride){
1660 const uint32_t a= ((uint32_t*)(src-stride))[0];
1661 const uint32_t b= ((uint32_t*)(src-stride))[1];
1662 const uint32_t c= ((uint32_t*)(src-stride))[2];
1663 const uint32_t d= ((uint32_t*)(src-stride))[3];
1665 for(i=0; i<16; i++){
1666 ((uint32_t*)(src+i*stride))[0]= a;
1667 ((uint32_t*)(src+i*stride))[1]= b;
1668 ((uint32_t*)(src+i*stride))[2]= c;
1669 ((uint32_t*)(src+i*stride))[3]= d;
1673 static void pred16x16_horizontal_c(uint8_t *src, int stride){
1676 for(i=0; i<16; i++){
1677 ((uint32_t*)(src+i*stride))[0]=
1678 ((uint32_t*)(src+i*stride))[1]=
1679 ((uint32_t*)(src+i*stride))[2]=
1680 ((uint32_t*)(src+i*stride))[3]= src[-1+i*stride]*0x01010101;
1684 static void pred16x16_dc_c(uint8_t *src, int stride){
1688 dc+= src[-1+i*stride];
1695 dc= 0x01010101*((dc + 16)>>5);
1697 for(i=0; i<16; i++){
1698 ((uint32_t*)(src+i*stride))[0]=
1699 ((uint32_t*)(src+i*stride))[1]=
1700 ((uint32_t*)(src+i*stride))[2]=
1701 ((uint32_t*)(src+i*stride))[3]= dc;
1705 static void pred16x16_left_dc_c(uint8_t *src, int stride){
1709 dc+= src[-1+i*stride];
1712 dc= 0x01010101*((dc + 8)>>4);
1714 for(i=0; i<16; i++){
1715 ((uint32_t*)(src+i*stride))[0]=
1716 ((uint32_t*)(src+i*stride))[1]=
1717 ((uint32_t*)(src+i*stride))[2]=
1718 ((uint32_t*)(src+i*stride))[3]= dc;
1722 static void pred16x16_top_dc_c(uint8_t *src, int stride){
1728 dc= 0x01010101*((dc + 8)>>4);
1730 for(i=0; i<16; i++){
1731 ((uint32_t*)(src+i*stride))[0]=
1732 ((uint32_t*)(src+i*stride))[1]=
1733 ((uint32_t*)(src+i*stride))[2]=
1734 ((uint32_t*)(src+i*stride))[3]= dc;
1738 static void pred16x16_128_dc_c(uint8_t *src, int stride){
1741 for(i=0; i<16; i++){
1742 ((uint32_t*)(src+i*stride))[0]=
1743 ((uint32_t*)(src+i*stride))[1]=
1744 ((uint32_t*)(src+i*stride))[2]=
1745 ((uint32_t*)(src+i*stride))[3]= 0x01010101U*128U;
1749 static inline void pred16x16_plane_compat_c(uint8_t *src, int stride, const int svq3){
1752 uint8_t *cm = cropTbl + MAX_NEG_CROP;
1753 const uint8_t * const src0 = src+7-stride;
1754 const uint8_t *src1 = src+8*stride-1;
1755 const uint8_t *src2 = src1-2*stride; // == src+6*stride-1;
1756 int H = src0[1] - src0[-1];
1757 int V = src1[0] - src2[ 0];
1758 for(k=2; k<=8; ++k) {
1759 src1 += stride; src2 -= stride;
1760 H += k*(src0[k] - src0[-k]);
1761 V += k*(src1[0] - src2[ 0]);
1764 H = ( 5*(H/4) ) / 16;
1765 V = ( 5*(V/4) ) / 16;
1767 /* required for 100% accuracy */
1768 i = H; H = V; V = i;
1770 H = ( 5*H+32 ) >> 6;
1771 V = ( 5*V+32 ) >> 6;
1774 a = 16*(src1[0] + src2[16] + 1) - 7*(V+H);
1775 for(j=16; j>0; --j) {
1778 for(i=-16; i<0; i+=4) {
1779 src[16+i] = cm[ (b ) >> 5 ];
1780 src[17+i] = cm[ (b+ H) >> 5 ];
1781 src[18+i] = cm[ (b+2*H) >> 5 ];
1782 src[19+i] = cm[ (b+3*H) >> 5 ];
1789 static void pred16x16_plane_c(uint8_t *src, int stride){
1790 pred16x16_plane_compat_c(src, stride, 0);
1793 static void pred8x8_vertical_c(uint8_t *src, int stride){
1795 const uint32_t a= ((uint32_t*)(src-stride))[0];
1796 const uint32_t b= ((uint32_t*)(src-stride))[1];
1799 ((uint32_t*)(src+i*stride))[0]= a;
1800 ((uint32_t*)(src+i*stride))[1]= b;
1804 static void pred8x8_horizontal_c(uint8_t *src, int stride){
1808 ((uint32_t*)(src+i*stride))[0]=
1809 ((uint32_t*)(src+i*stride))[1]= src[-1+i*stride]*0x01010101;
1813 static void pred8x8_128_dc_c(uint8_t *src, int stride){
1817 ((uint32_t*)(src+i*stride))[0]=
1818 ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
1821 ((uint32_t*)(src+i*stride))[0]=
1822 ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
1826 static void pred8x8_left_dc_c(uint8_t *src, int stride){
1832 dc0+= src[-1+i*stride];
1833 dc2+= src[-1+(i+4)*stride];
1835 dc0= 0x01010101*((dc0 + 2)>>2);
1836 dc2= 0x01010101*((dc2 + 2)>>2);
1839 ((uint32_t*)(src+i*stride))[0]=
1840 ((uint32_t*)(src+i*stride))[1]= dc0;
1843 ((uint32_t*)(src+i*stride))[0]=
1844 ((uint32_t*)(src+i*stride))[1]= dc2;
1848 static void pred8x8_top_dc_c(uint8_t *src, int stride){
1854 dc0+= src[i-stride];
1855 dc1+= src[4+i-stride];
1857 dc0= 0x01010101*((dc0 + 2)>>2);
1858 dc1= 0x01010101*((dc1 + 2)>>2);
1861 ((uint32_t*)(src+i*stride))[0]= dc0;
1862 ((uint32_t*)(src+i*stride))[1]= dc1;
1865 ((uint32_t*)(src+i*stride))[0]= dc0;
1866 ((uint32_t*)(src+i*stride))[1]= dc1;
1871 static void pred8x8_dc_c(uint8_t *src, int stride){
1873 int dc0, dc1, dc2, dc3;
1877 dc0+= src[-1+i*stride] + src[i-stride];
1878 dc1+= src[4+i-stride];
1879 dc2+= src[-1+(i+4)*stride];
1881 dc3= 0x01010101*((dc1 + dc2 + 4)>>3);
1882 dc0= 0x01010101*((dc0 + 4)>>3);
1883 dc1= 0x01010101*((dc1 + 2)>>2);
1884 dc2= 0x01010101*((dc2 + 2)>>2);
1887 ((uint32_t*)(src+i*stride))[0]= dc0;
1888 ((uint32_t*)(src+i*stride))[1]= dc1;
1891 ((uint32_t*)(src+i*stride))[0]= dc2;
1892 ((uint32_t*)(src+i*stride))[1]= dc3;
1896 static void pred8x8_plane_c(uint8_t *src, int stride){
1899 uint8_t *cm = cropTbl + MAX_NEG_CROP;
1900 const uint8_t * const src0 = src+3-stride;
1901 const uint8_t *src1 = src+4*stride-1;
1902 const uint8_t *src2 = src1-2*stride; // == src+2*stride-1;
1903 int H = src0[1] - src0[-1];
1904 int V = src1[0] - src2[ 0];
1905 for(k=2; k<=4; ++k) {
1906 src1 += stride; src2 -= stride;
1907 H += k*(src0[k] - src0[-k]);
1908 V += k*(src1[0] - src2[ 0]);
1910 H = ( 17*H+16 ) >> 5;
1911 V = ( 17*V+16 ) >> 5;
1913 a = 16*(src1[0] + src2[8]+1) - 3*(V+H);
1914 for(j=8; j>0; --j) {
1917 src[0] = cm[ (b ) >> 5 ];
1918 src[1] = cm[ (b+ H) >> 5 ];
1919 src[2] = cm[ (b+2*H) >> 5 ];
1920 src[3] = cm[ (b+3*H) >> 5 ];
1921 src[4] = cm[ (b+4*H) >> 5 ];
1922 src[5] = cm[ (b+5*H) >> 5 ];
1923 src[6] = cm[ (b+6*H) >> 5 ];
1924 src[7] = cm[ (b+7*H) >> 5 ];
1929 static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
1930 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1931 int src_x_offset, int src_y_offset,
1932 qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
1933 MpegEncContext * const s = &h->s;
1934 const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
1935 const int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
1936 const int luma_xy= (mx&3) + ((my&3)<<2);
1937 uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*s->linesize;
1938 uint8_t * src_cb= pic->data[1] + (mx>>3) + (my>>3)*s->uvlinesize;
1939 uint8_t * src_cr= pic->data[2] + (mx>>3) + (my>>3)*s->uvlinesize;
1940 int extra_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16; //FIXME increase edge?, IMHO not worth it
1941 int extra_height= extra_width;
1943 const int full_mx= mx>>2;
1944 const int full_my= my>>2;
1946 assert(pic->data[0]);
1948 if(mx&7) extra_width -= 3;
1949 if(my&7) extra_height -= 3;
1951 if( full_mx < 0-extra_width
1952 || full_my < 0-extra_height
1953 || full_mx + 16/*FIXME*/ > s->width + extra_width
1954 || full_my + 16/*FIXME*/ > s->height + extra_height){
1955 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);
1956 src_y= s->edge_emu_buffer + 2 + 2*s->linesize;
1960 qpix_op[luma_xy](dest_y, src_y, s->linesize); //FIXME try variable height perhaps?
1962 qpix_op[luma_xy](dest_y + delta, src_y + delta, s->linesize);
1965 if(s->flags&CODEC_FLAG_GRAY) return;
1968 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);
1969 src_cb= s->edge_emu_buffer;
1971 chroma_op(dest_cb, src_cb, s->uvlinesize, chroma_height, mx&7, my&7);
1974 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);
1975 src_cr= s->edge_emu_buffer;
1977 chroma_op(dest_cr, src_cr, s->uvlinesize, chroma_height, mx&7, my&7);
1980 static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
1981 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1982 int x_offset, int y_offset,
1983 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
1984 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
1985 int list0, int list1){
1986 MpegEncContext * const s = &h->s;
1987 qpel_mc_func *qpix_op= qpix_put;
1988 h264_chroma_mc_func chroma_op= chroma_put;
1990 dest_y += 2*x_offset + 2*y_offset*s-> linesize;
1991 dest_cb += x_offset + y_offset*s->uvlinesize;
1992 dest_cr += x_offset + y_offset*s->uvlinesize;
1993 x_offset += 8*s->mb_x;
1994 y_offset += 8*s->mb_y;
1997 Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
1998 mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
1999 dest_y, dest_cb, dest_cr, x_offset, y_offset,
2000 qpix_op, chroma_op);
2003 chroma_op= chroma_avg;
2007 Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
2008 mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
2009 dest_y, dest_cb, dest_cr, x_offset, y_offset,
2010 qpix_op, chroma_op);
2014 static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2015 qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
2016 qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg)){
2017 MpegEncContext * const s = &h->s;
2018 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
2019 const int mb_type= s->current_picture.mb_type[mb_xy];
2021 assert(IS_INTER(mb_type));
2023 if(IS_16X16(mb_type)){
2024 mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
2025 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
2026 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2027 }else if(IS_16X8(mb_type)){
2028 mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
2029 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
2030 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2031 mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
2032 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
2033 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2034 }else if(IS_8X16(mb_type)){
2035 mc_part(h, 0, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 0, 0,
2036 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2037 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2038 mc_part(h, 4, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 4, 0,
2039 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2040 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2044 assert(IS_8X8(mb_type));
2047 const int sub_mb_type= h->sub_mb_type[i];
2049 int x_offset= (i&1)<<2;
2050 int y_offset= (i&2)<<1;
2052 if(IS_SUB_8X8(sub_mb_type)){
2053 mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2054 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2055 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2056 }else if(IS_SUB_8X4(sub_mb_type)){
2057 mc_part(h, n , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2058 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2059 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2060 mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
2061 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2062 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2063 }else if(IS_SUB_4X8(sub_mb_type)){
2064 mc_part(h, n , 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2065 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2066 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2067 mc_part(h, n+1, 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
2068 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2069 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2072 assert(IS_SUB_4X4(sub_mb_type));
2074 int sub_x_offset= x_offset + 2*(j&1);
2075 int sub_y_offset= y_offset + (j&2);
2076 mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
2077 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2078 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2085 static void decode_init_vlc(H264Context *h){
2086 static int done = 0;
2092 init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5,
2093 &chroma_dc_coeff_token_len [0], 1, 1,
2094 &chroma_dc_coeff_token_bits[0], 1, 1);
2097 init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17,
2098 &coeff_token_len [i][0], 1, 1,
2099 &coeff_token_bits[i][0], 1, 1);
2103 init_vlc(&chroma_dc_total_zeros_vlc[i], CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
2104 &chroma_dc_total_zeros_len [i][0], 1, 1,
2105 &chroma_dc_total_zeros_bits[i][0], 1, 1);
2107 for(i=0; i<15; i++){
2108 init_vlc(&total_zeros_vlc[i], TOTAL_ZEROS_VLC_BITS, 16,
2109 &total_zeros_len [i][0], 1, 1,
2110 &total_zeros_bits[i][0], 1, 1);
2114 init_vlc(&run_vlc[i], RUN_VLC_BITS, 7,
2115 &run_len [i][0], 1, 1,
2116 &run_bits[i][0], 1, 1);
2118 init_vlc(&run7_vlc, RUN7_VLC_BITS, 16,
2119 &run_len [6][0], 1, 1,
2120 &run_bits[6][0], 1, 1);
2125 * Sets the intra prediction function pointers.
2127 static void init_pred_ptrs(H264Context *h){
2128 // MpegEncContext * const s = &h->s;
2130 h->pred4x4[VERT_PRED ]= pred4x4_vertical_c;
2131 h->pred4x4[HOR_PRED ]= pred4x4_horizontal_c;
2132 h->pred4x4[DC_PRED ]= pred4x4_dc_c;
2133 h->pred4x4[DIAG_DOWN_LEFT_PRED ]= pred4x4_down_left_c;
2134 h->pred4x4[DIAG_DOWN_RIGHT_PRED]= pred4x4_down_right_c;
2135 h->pred4x4[VERT_RIGHT_PRED ]= pred4x4_vertical_right_c;
2136 h->pred4x4[HOR_DOWN_PRED ]= pred4x4_horizontal_down_c;
2137 h->pred4x4[VERT_LEFT_PRED ]= pred4x4_vertical_left_c;
2138 h->pred4x4[HOR_UP_PRED ]= pred4x4_horizontal_up_c;
2139 h->pred4x4[LEFT_DC_PRED ]= pred4x4_left_dc_c;
2140 h->pred4x4[TOP_DC_PRED ]= pred4x4_top_dc_c;
2141 h->pred4x4[DC_128_PRED ]= pred4x4_128_dc_c;
2143 h->pred8x8[DC_PRED8x8 ]= pred8x8_dc_c;
2144 h->pred8x8[VERT_PRED8x8 ]= pred8x8_vertical_c;
2145 h->pred8x8[HOR_PRED8x8 ]= pred8x8_horizontal_c;
2146 h->pred8x8[PLANE_PRED8x8 ]= pred8x8_plane_c;
2147 h->pred8x8[LEFT_DC_PRED8x8]= pred8x8_left_dc_c;
2148 h->pred8x8[TOP_DC_PRED8x8 ]= pred8x8_top_dc_c;
2149 h->pred8x8[DC_128_PRED8x8 ]= pred8x8_128_dc_c;
2151 h->pred16x16[DC_PRED8x8 ]= pred16x16_dc_c;
2152 h->pred16x16[VERT_PRED8x8 ]= pred16x16_vertical_c;
2153 h->pred16x16[HOR_PRED8x8 ]= pred16x16_horizontal_c;
2154 h->pred16x16[PLANE_PRED8x8 ]= pred16x16_plane_c;
2155 h->pred16x16[LEFT_DC_PRED8x8]= pred16x16_left_dc_c;
2156 h->pred16x16[TOP_DC_PRED8x8 ]= pred16x16_top_dc_c;
2157 h->pred16x16[DC_128_PRED8x8 ]= pred16x16_128_dc_c;
2160 static void free_tables(H264Context *h){
2161 av_freep(&h->intra4x4_pred_mode);
2162 av_freep(&h->chroma_pred_mode_table);
2163 av_freep(&h->cbp_table);
2164 av_freep(&h->mvd_table[0]);
2165 av_freep(&h->mvd_table[1]);
2166 av_freep(&h->non_zero_count);
2167 av_freep(&h->slice_table_base);
2168 av_freep(&h->top_border);
2169 h->slice_table= NULL;
2171 av_freep(&h->mb2b_xy);
2172 av_freep(&h->mb2b8_xy);
2177 * needs widzh/height
2179 static int alloc_tables(H264Context *h){
2180 MpegEncContext * const s = &h->s;
2181 const int big_mb_num= s->mb_stride * (s->mb_height+1);
2184 CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8 * sizeof(uint8_t))
2186 CHECKED_ALLOCZ(h->non_zero_count , big_mb_num * 16 * sizeof(uint8_t))
2187 CHECKED_ALLOCZ(h->slice_table_base , big_mb_num * sizeof(uint8_t))
2188 CHECKED_ALLOCZ(h->top_border , s->mb_width * (16+8+8) * sizeof(uint8_t))
2190 if( h->pps.cabac ) {
2191 CHECKED_ALLOCZ(h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t))
2192 CHECKED_ALLOCZ(h->cbp_table, big_mb_num * sizeof(uint16_t))
2193 CHECKED_ALLOCZ(h->mvd_table[0], 32*big_mb_num * sizeof(uint16_t));
2194 CHECKED_ALLOCZ(h->mvd_table[1], 32*big_mb_num * sizeof(uint16_t));
2197 memset(h->slice_table_base, -1, big_mb_num * sizeof(uint8_t));
2198 h->slice_table= h->slice_table_base + s->mb_stride + 1;
2200 CHECKED_ALLOCZ(h->mb2b_xy , big_mb_num * sizeof(uint16_t));
2201 CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint16_t));
2202 for(y=0; y<s->mb_height; y++){
2203 for(x=0; x<s->mb_width; x++){
2204 const int mb_xy= x + y*s->mb_stride;
2205 const int b_xy = 4*x + 4*y*h->b_stride;
2206 const int b8_xy= 2*x + 2*y*h->b8_stride;
2208 h->mb2b_xy [mb_xy]= b_xy;
2209 h->mb2b8_xy[mb_xy]= b8_xy;
2219 static void common_init(H264Context *h){
2220 MpegEncContext * const s = &h->s;
2222 s->width = s->avctx->width;
2223 s->height = s->avctx->height;
2224 s->codec_id= s->avctx->codec->id;
2228 s->unrestricted_mv=1;
2229 s->decode=1; //FIXME
2232 static int decode_init(AVCodecContext *avctx){
2233 H264Context *h= avctx->priv_data;
2234 MpegEncContext * const s = &h->s;
2236 MPV_decode_defaults(s);
2241 s->out_format = FMT_H264;
2242 s->workaround_bugs= avctx->workaround_bugs;
2245 // s->decode_mb= ff_h263_decode_mb;
2247 avctx->pix_fmt= PIX_FMT_YUV420P;
2251 if(avctx->codec_tag != 0x31637661) // avc1
2254 if((avctx->extradata_size == 0) || (avctx->extradata == NULL)) {
2255 av_log(avctx, AV_LOG_ERROR, "AVC codec requires avcC data\n");
2265 static void frame_start(H264Context *h){
2266 MpegEncContext * const s = &h->s;
2269 MPV_frame_start(s, s->avctx);
2270 ff_er_frame_start(s);
2273 assert(s->linesize && s->uvlinesize);
2275 for(i=0; i<16; i++){
2276 h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
2277 h->chroma_subblock_offset[i]= 2*((scan8[i] - scan8[0])&7) + 2*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2280 h->block_offset[16+i]=
2281 h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2284 // s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
2287 static inline void backup_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){
2288 MpegEncContext * const s = &h->s;
2292 src_cb -= uvlinesize;
2293 src_cr -= uvlinesize;
2295 h->left_border[0]= h->top_border[s->mb_x][15];
2296 for(i=1; i<17; i++){
2297 h->left_border[i]= src_y[15+i* linesize];
2300 *(uint64_t*)(h->top_border[s->mb_x]+0)= *(uint64_t*)(src_y + 16*linesize);
2301 *(uint64_t*)(h->top_border[s->mb_x]+8)= *(uint64_t*)(src_y +8+16*linesize);
2303 if(!(s->flags&CODEC_FLAG_GRAY)){
2304 h->left_border[17 ]= h->top_border[s->mb_x][16+7];
2305 h->left_border[17+9]= h->top_border[s->mb_x][24+7];
2307 h->left_border[i+17 ]= src_cb[7+i*uvlinesize];
2308 h->left_border[i+17+9]= src_cr[7+i*uvlinesize];
2310 *(uint64_t*)(h->top_border[s->mb_x]+16)= *(uint64_t*)(src_cb+8*uvlinesize);
2311 *(uint64_t*)(h->top_border[s->mb_x]+24)= *(uint64_t*)(src_cr+8*uvlinesize);
2315 static inline void xchg_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize, int xchg){
2316 MpegEncContext * const s = &h->s;
2319 int deblock_left = (s->mb_x > 0);
2320 int deblock_top = (s->mb_y > 0);
2322 src_y -= linesize + 1;
2323 src_cb -= uvlinesize + 1;
2324 src_cr -= uvlinesize + 1;
2326 #define XCHG(a,b,t,xchg)\
2333 for(i = !deblock_top; i<17; i++){
2334 XCHG(h->left_border[i ], src_y [i* linesize], temp8, xchg);
2339 XCHG(*(uint64_t*)(h->top_border[s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
2340 XCHG(*(uint64_t*)(h->top_border[s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
2343 if(!(s->flags&CODEC_FLAG_GRAY)){
2345 for(i = !deblock_top; i<9; i++){
2346 XCHG(h->left_border[i+17 ], src_cb[i*uvlinesize], temp8, xchg);
2347 XCHG(h->left_border[i+17+9], src_cr[i*uvlinesize], temp8, xchg);
2351 XCHG(*(uint64_t*)(h->top_border[s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
2352 XCHG(*(uint64_t*)(h->top_border[s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
2357 static void hl_decode_mb(H264Context *h){
2358 MpegEncContext * const s = &h->s;
2359 const int mb_x= s->mb_x;
2360 const int mb_y= s->mb_y;
2361 const int mb_xy= mb_x + mb_y*s->mb_stride;
2362 const int mb_type= s->current_picture.mb_type[mb_xy];
2363 uint8_t *dest_y, *dest_cb, *dest_cr;
2364 int linesize, uvlinesize /*dct_offset*/;
2373 dest_y = s->current_picture.data[0] + (mb_y * 16* s->linesize ) + mb_x * 16;
2374 dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2375 dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2377 if (h->mb_field_decoding_flag) {
2378 linesize = s->linesize * 2;
2379 uvlinesize = s->uvlinesize * 2;
2380 if(mb_y&1){ //FIXME move out of this func?
2381 dest_y -= s->linesize*15;
2382 dest_cb-= s->linesize*7;
2383 dest_cr-= s->linesize*7;
2386 linesize = s->linesize;
2387 uvlinesize = s->uvlinesize;
2388 // dct_offset = s->linesize * 16;
2391 if(IS_INTRA(mb_type)){
2392 if(h->deblocking_filter)
2393 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1);
2395 if(!(s->flags&CODEC_FLAG_GRAY)){
2396 h->pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
2397 h->pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
2400 if(IS_INTRA4x4(mb_type)){
2402 for(i=0; i<16; i++){
2403 uint8_t * const ptr= dest_y + h->block_offset[i];
2404 uint8_t *topright= ptr + 4 - linesize;
2405 const int topright_avail= (h->topright_samples_available<<i)&0x8000;
2406 const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
2409 if(!topright_avail){
2410 tr= ptr[3 - linesize]*0x01010101;
2411 topright= (uint8_t*) &tr;
2412 }else if(i==5 && h->deblocking_filter){
2413 tr= *(uint32_t*)h->top_border[mb_x+1];
2414 topright= (uint8_t*) &tr;
2417 h->pred4x4[ dir ](ptr, topright, linesize);
2418 if(h->non_zero_count_cache[ scan8[i] ]){
2419 if(s->codec_id == CODEC_ID_H264)
2420 s->dsp.h264_idct_add(ptr, h->mb + i*16, linesize);
2422 svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
2427 h->pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
2428 if(s->codec_id == CODEC_ID_H264)
2429 h264_luma_dc_dequant_idct_c(h->mb, s->qscale);
2431 svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
2433 if(h->deblocking_filter)
2434 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);
2435 }else if(s->codec_id == CODEC_ID_H264){
2436 hl_motion(h, dest_y, dest_cb, dest_cr,
2437 s->dsp.put_h264_qpel_pixels_tab, s->dsp.put_h264_chroma_pixels_tab,
2438 s->dsp.avg_h264_qpel_pixels_tab, s->dsp.avg_h264_chroma_pixels_tab);
2442 if(!IS_INTRA4x4(mb_type)){
2443 if(s->codec_id == CODEC_ID_H264){
2444 for(i=0; i<16; i++){
2445 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
2446 uint8_t * const ptr= dest_y + h->block_offset[i];
2447 s->dsp.h264_idct_add(ptr, h->mb + i*16, linesize);
2451 for(i=0; i<16; i++){
2452 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
2453 uint8_t * const ptr= dest_y + h->block_offset[i];
2454 svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
2460 if(!(s->flags&CODEC_FLAG_GRAY)){
2461 chroma_dc_dequant_idct_c(h->mb + 16*16, h->chroma_qp);
2462 chroma_dc_dequant_idct_c(h->mb + 16*16+4*16, h->chroma_qp);
2463 if(s->codec_id == CODEC_ID_H264){
2464 for(i=16; i<16+4; i++){
2465 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2466 uint8_t * const ptr= dest_cb + h->block_offset[i];
2467 s->dsp.h264_idct_add(ptr, h->mb + i*16, uvlinesize);
2470 for(i=20; i<20+4; i++){
2471 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2472 uint8_t * const ptr= dest_cr + h->block_offset[i];
2473 s->dsp.h264_idct_add(ptr, h->mb + i*16, uvlinesize);
2477 for(i=16; i<16+4; i++){
2478 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2479 uint8_t * const ptr= dest_cb + h->block_offset[i];
2480 svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
2483 for(i=20; i<20+4; i++){
2484 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2485 uint8_t * const ptr= dest_cr + h->block_offset[i];
2486 svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
2491 if(h->deblocking_filter) {
2492 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2493 filter_mb(h, mb_x, mb_y, dest_y, dest_cb, dest_cr);
2498 * fills the default_ref_list.
2500 static int fill_default_ref_list(H264Context *h){
2501 MpegEncContext * const s = &h->s;
2503 Picture sorted_short_ref[16];
2505 if(h->slice_type==B_TYPE){
2509 for(out_i=0; out_i<h->short_ref_count; out_i++){
2511 int best_poc=INT_MAX;
2513 for(i=0; i<h->short_ref_count; i++){
2514 const int poc= h->short_ref[i]->poc;
2515 if(poc > limit && poc < best_poc){
2521 assert(best_i != -1);
2524 sorted_short_ref[out_i]= *h->short_ref[best_i];
2528 if(s->picture_structure == PICT_FRAME){
2529 if(h->slice_type==B_TYPE){
2530 const int current_poc= s->current_picture_ptr->poc;
2533 for(list=0; list<2; list++){
2536 for(i=0; i<h->short_ref_count && index < h->ref_count[list]; i++){
2537 const int i2= list ? h->short_ref_count - i - 1 : i;
2538 const int poc= sorted_short_ref[i2].poc;
2540 if(sorted_short_ref[i2].reference != 3) continue; //FIXME refernce field shit
2542 if((list==1 && poc > current_poc) || (list==0 && poc < current_poc)){
2543 h->default_ref_list[list][index ]= sorted_short_ref[i2];
2544 h->default_ref_list[list][index++].pic_id= sorted_short_ref[i2].frame_num;
2548 for(i=0; i<h->long_ref_count && index < h->ref_count[ list ]; i++){
2549 if(h->long_ref[i]->reference != 3) continue;
2551 h->default_ref_list[ list ][index ]= *h->long_ref[i];
2552 h->default_ref_list[ list ][index++].pic_id= i;;
2555 if(h->long_ref_count > 1 && h->short_ref_count==0){
2556 Picture temp= h->default_ref_list[1][0];
2557 h->default_ref_list[1][0] = h->default_ref_list[1][1];
2558 h->default_ref_list[1][0] = temp;
2561 if(index < h->ref_count[ list ])
2562 memset(&h->default_ref_list[list][index], 0, sizeof(Picture)*(h->ref_count[ list ] - index));
2566 for(i=0; i<h->short_ref_count && index < h->ref_count[0]; i++){
2567 if(h->short_ref[i]->reference != 3) continue; //FIXME refernce field shit
2568 h->default_ref_list[0][index ]= *h->short_ref[i];
2569 h->default_ref_list[0][index++].pic_id= h->short_ref[i]->frame_num;
2571 for(i=0; i<h->long_ref_count && index < h->ref_count[0]; i++){
2572 if(h->long_ref[i]->reference != 3) continue;
2573 h->default_ref_list[0][index ]= *h->long_ref[i];
2574 h->default_ref_list[0][index++].pic_id= i;;
2576 if(index < h->ref_count[0])
2577 memset(&h->default_ref_list[0][index], 0, sizeof(Picture)*(h->ref_count[0] - index));
2580 if(h->slice_type==B_TYPE){
2582 //FIXME second field balh
2588 static int decode_ref_pic_list_reordering(H264Context *h){
2589 MpegEncContext * const s = &h->s;
2592 if(h->slice_type==I_TYPE || h->slice_type==SI_TYPE) return 0; //FIXME move beofre func
2594 for(list=0; list<2; list++){
2595 memcpy(h->ref_list[list], h->default_ref_list[list], sizeof(Picture)*h->ref_count[list]);
2597 if(get_bits1(&s->gb)){
2598 int pred= h->curr_pic_num;
2601 for(index=0; ; index++){
2602 int reordering_of_pic_nums_idc= get_ue_golomb(&s->gb);
2606 if(reordering_of_pic_nums_idc==3)
2609 if(index >= h->ref_count[list]){
2610 av_log(h->s.avctx, AV_LOG_ERROR, "reference count overflow\n");
2614 if(reordering_of_pic_nums_idc<3){
2615 if(reordering_of_pic_nums_idc<2){
2616 const int abs_diff_pic_num= get_ue_golomb(&s->gb) + 1;
2618 if(abs_diff_pic_num >= h->max_pic_num){
2619 av_log(h->s.avctx, AV_LOG_ERROR, "abs_diff_pic_num overflow\n");
2623 if(reordering_of_pic_nums_idc == 0) pred-= abs_diff_pic_num;
2624 else pred+= abs_diff_pic_num;
2625 pred &= h->max_pic_num - 1;
2627 for(i= h->ref_count[list]-1; i>=index; i--){
2628 if(h->ref_list[list][i].pic_id == pred && h->ref_list[list][i].long_ref==0)
2632 pic_id= get_ue_golomb(&s->gb); //long_term_pic_idx
2634 for(i= h->ref_count[list]-1; i>=index; i--){
2635 if(h->ref_list[list][i].pic_id == pic_id && h->ref_list[list][i].long_ref==1)
2641 av_log(h->s.avctx, AV_LOG_ERROR, "reference picture missing during reorder\n");
2642 memset(&h->ref_list[list][index], 0, sizeof(Picture)); //FIXME
2643 }else if(i > index){
2644 Picture tmp= h->ref_list[list][i];
2645 for(; i>index; i--){
2646 h->ref_list[list][i]= h->ref_list[list][i-1];
2648 h->ref_list[list][index]= tmp;
2651 av_log(h->s.avctx, AV_LOG_ERROR, "illegal reordering_of_pic_nums_idc\n");
2657 if(h->slice_type!=B_TYPE) break;
2662 static int pred_weight_table(H264Context *h){
2663 MpegEncContext * const s = &h->s;
2666 h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
2667 h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
2669 for(list=0; list<2; list++){
2670 for(i=0; i<h->ref_count[list]; i++){
2671 int luma_weight_flag, chroma_weight_flag;
2673 luma_weight_flag= get_bits1(&s->gb);
2674 if(luma_weight_flag){
2675 h->luma_weight[list][i]= get_se_golomb(&s->gb);
2676 h->luma_offset[list][i]= get_se_golomb(&s->gb);
2679 chroma_weight_flag= get_bits1(&s->gb);
2680 if(chroma_weight_flag){
2683 h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);
2684 h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);
2688 if(h->slice_type != B_TYPE) break;
2694 * instantaneos decoder refresh.
2696 static void idr(H264Context *h){
2699 for(i=0; i<h->long_ref_count; i++){
2700 h->long_ref[i]->reference=0;
2701 h->long_ref[i]= NULL;
2703 h->long_ref_count=0;
2705 for(i=0; i<h->short_ref_count; i++){
2706 h->short_ref[i]->reference=0;
2707 h->short_ref[i]= NULL;
2709 h->short_ref_count=0;
2714 * @return the removed picture or NULL if an error occures
2716 static Picture * remove_short(H264Context *h, int frame_num){
2717 MpegEncContext * const s = &h->s;
2720 if(s->avctx->debug&FF_DEBUG_MMCO)
2721 av_log(h->s.avctx, AV_LOG_DEBUG, "remove short %d count %d\n", frame_num, h->short_ref_count);
2723 for(i=0; i<h->short_ref_count; i++){
2724 Picture *pic= h->short_ref[i];
2725 if(s->avctx->debug&FF_DEBUG_MMCO)
2726 av_log(h->s.avctx, AV_LOG_DEBUG, "%d %d %p\n", i, pic->frame_num, pic);
2727 if(pic->frame_num == frame_num){
2728 h->short_ref[i]= NULL;
2729 memmove(&h->short_ref[i], &h->short_ref[i+1], (h->short_ref_count - i - 1)*sizeof(Picture*));
2730 h->short_ref_count--;
2739 * @return the removed picture or NULL if an error occures
2741 static Picture * remove_long(H264Context *h, int i){
2744 if(i >= h->long_ref_count) return NULL;
2745 pic= h->long_ref[i];
2746 if(pic==NULL) return NULL;
2748 h->long_ref[i]= NULL;
2749 memmove(&h->long_ref[i], &h->long_ref[i+1], (h->long_ref_count - i - 1)*sizeof(Picture*));
2750 h->long_ref_count--;
2756 * Executes the reference picture marking (memory management control operations).
2758 static int execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count){
2759 MpegEncContext * const s = &h->s;
2761 int current_is_long=0;
2764 if((s->avctx->debug&FF_DEBUG_MMCO) && mmco_count==0)
2765 av_log(h->s.avctx, AV_LOG_DEBUG, "no mmco here\n");
2767 for(i=0; i<mmco_count; i++){
2768 if(s->avctx->debug&FF_DEBUG_MMCO)
2769 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);
2771 switch(mmco[i].opcode){
2772 case MMCO_SHORT2UNUSED:
2773 pic= remove_short(h, mmco[i].short_frame_num);
2774 if(pic==NULL) return -1;
2777 case MMCO_SHORT2LONG:
2778 pic= remove_long(h, mmco[i].long_index);
2779 if(pic) pic->reference=0;
2781 h->long_ref[ mmco[i].long_index ]= remove_short(h, mmco[i].short_frame_num);
2782 h->long_ref[ mmco[i].long_index ]->long_ref=1;
2784 case MMCO_LONG2UNUSED:
2785 pic= remove_long(h, mmco[i].long_index);
2786 if(pic==NULL) return -1;
2790 pic= remove_long(h, mmco[i].long_index);
2791 if(pic) pic->reference=0;
2793 h->long_ref[ mmco[i].long_index ]= s->current_picture_ptr;
2794 h->long_ref[ mmco[i].long_index ]->long_ref=1;
2795 h->long_ref_count++;
2799 case MMCO_SET_MAX_LONG:
2800 assert(mmco[i].long_index <= 16);
2801 while(mmco[i].long_index < h->long_ref_count){
2802 pic= remove_long(h, mmco[i].long_index);
2805 while(mmco[i].long_index > h->long_ref_count){
2806 h->long_ref[ h->long_ref_count++ ]= NULL;
2810 while(h->short_ref_count){
2811 pic= remove_short(h, h->short_ref[0]->frame_num);
2814 while(h->long_ref_count){
2815 pic= remove_long(h, h->long_ref_count-1);
2823 if(!current_is_long){
2824 pic= remove_short(h, s->current_picture_ptr->frame_num);
2827 av_log(h->s.avctx, AV_LOG_ERROR, "illegal short term buffer state detected\n");
2830 if(h->short_ref_count)
2831 memmove(&h->short_ref[1], &h->short_ref[0], h->short_ref_count*sizeof(Picture*));
2833 h->short_ref[0]= s->current_picture_ptr;
2834 h->short_ref[0]->long_ref=0;
2835 h->short_ref_count++;
2841 static int decode_ref_pic_marking(H264Context *h){
2842 MpegEncContext * const s = &h->s;
2845 if(h->nal_unit_type == NAL_IDR_SLICE){ //FIXME fields
2846 s->broken_link= get_bits1(&s->gb) -1;
2847 h->mmco[0].long_index= get_bits1(&s->gb) - 1; // current_long_term_idx
2848 if(h->mmco[0].long_index == -1)
2851 h->mmco[0].opcode= MMCO_LONG;
2855 if(get_bits1(&s->gb)){ // adaptive_ref_pic_marking_mode_flag
2856 for(i= h->mmco_index; i<MAX_MMCO_COUNT; i++) {
2857 MMCOOpcode opcode= get_ue_golomb(&s->gb);;
2859 h->mmco[i].opcode= opcode;
2860 if(opcode==MMCO_SHORT2UNUSED || opcode==MMCO_SHORT2LONG){
2861 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
2862 /* if(h->mmco[i].short_frame_num >= h->short_ref_count || h->short_ref[ h->mmco[i].short_frame_num ] == NULL){
2863 fprintf(stderr, "illegal short ref in memory management control operation %d\n", mmco);
2867 if(opcode==MMCO_SHORT2LONG || opcode==MMCO_LONG2UNUSED || opcode==MMCO_LONG || opcode==MMCO_SET_MAX_LONG){
2868 h->mmco[i].long_index= get_ue_golomb(&s->gb);
2869 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){
2870 av_log(h->s.avctx, AV_LOG_ERROR, "illegal long ref in memory management control operation %d\n", opcode);
2875 if(opcode > MMCO_LONG){
2876 av_log(h->s.avctx, AV_LOG_ERROR, "illegal memory management control operation %d\n", opcode);
2879 if(opcode == MMCO_END)
2884 assert(h->long_ref_count + h->short_ref_count <= h->sps.ref_frame_count);
2886 if(h->long_ref_count + h->short_ref_count == h->sps.ref_frame_count){ //FIXME fields
2887 h->mmco[0].opcode= MMCO_SHORT2UNUSED;
2888 h->mmco[0].short_frame_num= h->short_ref[ h->short_ref_count - 1 ]->frame_num;
2898 static int init_poc(H264Context *h){
2899 MpegEncContext * const s = &h->s;
2900 const int max_frame_num= 1<<h->sps.log2_max_frame_num;
2903 if(h->nal_unit_type == NAL_IDR_SLICE){
2904 h->frame_num_offset= 0;
2906 if(h->frame_num < h->prev_frame_num)
2907 h->frame_num_offset= h->prev_frame_num_offset + max_frame_num;
2909 h->frame_num_offset= h->prev_frame_num_offset;
2912 if(h->sps.poc_type==0){
2913 const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
2915 if (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
2916 h->poc_msb = h->prev_poc_msb + max_poc_lsb;
2917 else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
2918 h->poc_msb = h->prev_poc_msb - max_poc_lsb;
2920 h->poc_msb = h->prev_poc_msb;
2921 //printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
2923 field_poc[1] = h->poc_msb + h->poc_lsb;
2924 if(s->picture_structure == PICT_FRAME)
2925 field_poc[1] += h->delta_poc_bottom;
2926 }else if(h->sps.poc_type==1){
2927 int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
2930 if(h->sps.poc_cycle_length != 0)
2931 abs_frame_num = h->frame_num_offset + h->frame_num;
2935 if(h->nal_ref_idc==0 && abs_frame_num > 0)
2938 expected_delta_per_poc_cycle = 0;
2939 for(i=0; i < h->sps.poc_cycle_length; i++)
2940 expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
2942 if(abs_frame_num > 0){
2943 int poc_cycle_cnt = (abs_frame_num - 1) / h->sps.poc_cycle_length;
2944 int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
2946 expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
2947 for(i = 0; i <= frame_num_in_poc_cycle; i++)
2948 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
2952 if(h->nal_ref_idc == 0)
2953 expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
2955 field_poc[0] = expectedpoc + h->delta_poc[0];
2956 field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
2958 if(s->picture_structure == PICT_FRAME)
2959 field_poc[1] += h->delta_poc[1];
2962 if(h->nal_unit_type == NAL_IDR_SLICE){
2965 if(h->nal_ref_idc) poc= 2*(h->frame_num_offset + h->frame_num);
2966 else poc= 2*(h->frame_num_offset + h->frame_num) - 1;
2972 if(s->picture_structure != PICT_BOTTOM_FIELD)
2973 s->current_picture_ptr->field_poc[0]= field_poc[0];
2974 if(s->picture_structure != PICT_TOP_FIELD)
2975 s->current_picture_ptr->field_poc[1]= field_poc[1];
2976 if(s->picture_structure == PICT_FRAME) // FIXME field pix?
2977 s->current_picture_ptr->poc= FFMIN(field_poc[0], field_poc[1]);
2983 * decodes a slice header.
2984 * this will allso call MPV_common_init() and frame_start() as needed
2986 static int decode_slice_header(H264Context *h){
2987 MpegEncContext * const s = &h->s;
2988 int first_mb_in_slice, pps_id;
2989 int num_ref_idx_active_override_flag;
2990 static const uint8_t slice_type_map[5]= {P_TYPE, B_TYPE, I_TYPE, SP_TYPE, SI_TYPE};
2992 s->current_picture.reference= h->nal_ref_idc != 0;
2994 first_mb_in_slice= get_ue_golomb(&s->gb);
2996 h->slice_type= get_ue_golomb(&s->gb);
2997 if(h->slice_type > 9){
2998 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);
3000 if(h->slice_type > 4){
3002 h->slice_type_fixed=1;
3004 h->slice_type_fixed=0;
3006 h->slice_type= slice_type_map[ h->slice_type ];
3008 s->pict_type= h->slice_type; // to make a few old func happy, its wrong though
3010 pps_id= get_ue_golomb(&s->gb);
3012 av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
3015 h->pps= h->pps_buffer[pps_id];
3016 if(h->pps.slice_group_count == 0){
3017 av_log(h->s.avctx, AV_LOG_ERROR, "non existing PPS referenced\n");
3021 h->sps= h->sps_buffer[ h->pps.sps_id ];
3022 if(h->sps.log2_max_frame_num == 0){
3023 av_log(h->s.avctx, AV_LOG_ERROR, "non existing SPS referenced\n");
3027 s->mb_width= h->sps.mb_width;
3028 s->mb_height= h->sps.mb_height;
3030 h->b_stride= s->mb_width*4;
3031 h->b8_stride= s->mb_width*2;
3033 s->mb_x = first_mb_in_slice % s->mb_width;
3034 s->mb_y = first_mb_in_slice / s->mb_width; //FIXME AFFW
3036 s->width = 16*s->mb_width - 2*(h->sps.crop_left + h->sps.crop_right );
3037 if(h->sps.frame_mbs_only_flag)
3038 s->height= 16*s->mb_height - 2*(h->sps.crop_top + h->sps.crop_bottom);
3040 s->height= 16*s->mb_height - 4*(h->sps.crop_top + h->sps.crop_bottom); //FIXME recheck
3042 if (s->context_initialized
3043 && ( s->width != s->avctx->width || s->height != s->avctx->height)) {
3047 if (!s->context_initialized) {
3048 if (MPV_common_init(s) < 0)
3053 s->avctx->width = s->width;
3054 s->avctx->height = s->height;
3055 s->avctx->sample_aspect_ratio= h->sps.sar;
3057 if(h->sps.timing_info_present_flag && h->sps.fixed_frame_rate_flag){
3058 s->avctx->frame_rate = h->sps.time_scale;
3059 s->avctx->frame_rate_base = h->sps.num_units_in_tick;
3063 if(first_mb_in_slice == 0){
3067 s->current_picture_ptr->frame_num= //FIXME frame_num cleanup
3068 h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
3070 if(h->sps.frame_mbs_only_flag){
3071 s->picture_structure= PICT_FRAME;
3073 if(get_bits1(&s->gb)) //field_pic_flag
3074 s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
3076 s->picture_structure= PICT_FRAME;
3079 if(s->picture_structure==PICT_FRAME){
3080 h->curr_pic_num= h->frame_num;
3081 h->max_pic_num= 1<< h->sps.log2_max_frame_num;
3083 h->curr_pic_num= 2*h->frame_num;
3084 h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
3087 if(h->nal_unit_type == NAL_IDR_SLICE){
3088 get_ue_golomb(&s->gb); /* idr_pic_id */
3091 if(h->sps.poc_type==0){
3092 h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
3094 if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
3095 h->delta_poc_bottom= get_se_golomb(&s->gb);
3099 if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
3100 h->delta_poc[0]= get_se_golomb(&s->gb);
3102 if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
3103 h->delta_poc[1]= get_se_golomb(&s->gb);
3108 if(h->pps.redundant_pic_cnt_present){
3109 h->redundant_pic_count= get_ue_golomb(&s->gb);
3112 //set defaults, might be overriden a few line later
3113 h->ref_count[0]= h->pps.ref_count[0];
3114 h->ref_count[1]= h->pps.ref_count[1];
3116 if(h->slice_type == P_TYPE || h->slice_type == SP_TYPE || h->slice_type == B_TYPE){
3117 if(h->slice_type == B_TYPE){
3118 h->direct_spatial_mv_pred= get_bits1(&s->gb);
3120 num_ref_idx_active_override_flag= get_bits1(&s->gb);
3122 if(num_ref_idx_active_override_flag){
3123 h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
3124 if(h->slice_type==B_TYPE)
3125 h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
3127 if(h->ref_count[0] > 32 || h->ref_count[1] > 32){
3128 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
3134 if(first_mb_in_slice == 0){
3135 fill_default_ref_list(h);
3138 decode_ref_pic_list_reordering(h);
3140 if( (h->pps.weighted_pred && (h->slice_type == P_TYPE || h->slice_type == SP_TYPE ))
3141 || (h->pps.weighted_bipred_idc==1 && h->slice_type==B_TYPE ) )
3142 pred_weight_table(h);
3144 if(s->current_picture.reference)
3145 decode_ref_pic_marking(h);
3147 if( h->slice_type != I_TYPE && h->slice_type != SI_TYPE && h->pps.cabac )
3148 h->cabac_init_idc = get_ue_golomb(&s->gb);
3150 h->last_qscale_diff = 0;
3151 s->qscale = h->pps.init_qp + get_se_golomb(&s->gb);
3152 if(s->qscale<0 || s->qscale>51){
3153 av_log(s->avctx, AV_LOG_ERROR, "QP %d out of range\n", s->qscale);
3156 //FIXME qscale / qp ... stuff
3157 if(h->slice_type == SP_TYPE){
3158 get_bits1(&s->gb); /* sp_for_switch_flag */
3160 if(h->slice_type==SP_TYPE || h->slice_type == SI_TYPE){
3161 get_se_golomb(&s->gb); /* slice_qs_delta */
3164 h->deblocking_filter = 1;
3165 h->slice_alpha_c0_offset = 0;
3166 h->slice_beta_offset = 0;
3167 if( h->pps.deblocking_filter_parameters_present ) {
3168 h->deblocking_filter= get_ue_golomb(&s->gb);
3169 if(h->deblocking_filter < 2)
3170 h->deblocking_filter^= 1; // 1<->0
3172 if( h->deblocking_filter ) {
3173 h->slice_alpha_c0_offset = get_se_golomb(&s->gb) << 1;
3174 h->slice_beta_offset = get_se_golomb(&s->gb) << 1;
3179 if( h->pps.num_slice_groups > 1 && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
3180 slice_group_change_cycle= get_bits(&s->gb, ?);
3183 if(s->avctx->debug&FF_DEBUG_PICT_INFO){
3184 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",
3186 av_get_pict_type_char(h->slice_type),
3187 pps_id, h->frame_num,
3188 s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
3189 h->ref_count[0], h->ref_count[1],
3191 h->deblocking_filter
3201 static inline int get_level_prefix(GetBitContext *gb){
3205 OPEN_READER(re, gb);
3206 UPDATE_CACHE(re, gb);
3207 buf=GET_CACHE(re, gb);
3209 log= 32 - av_log2(buf);
3211 print_bin(buf>>(32-log), log);
3212 av_log(NULL, AV_LOG_DEBUG, "%5d %2d %3d lpr @%5d in %s get_level_prefix\n", buf>>(32-log), log, log-1, get_bits_count(gb), __FILE__);
3215 LAST_SKIP_BITS(re, gb, log);
3216 CLOSE_READER(re, gb);
3222 * decodes a residual block.
3223 * @param n block index
3224 * @param scantable scantable
3225 * @param max_coeff number of coefficients in the block
3226 * @return <0 if an error occured
3228 static int decode_residual(H264Context *h, GetBitContext *gb, DCTELEM *block, int n, const uint8_t *scantable, int qp, int max_coeff){
3229 MpegEncContext * const s = &h->s;
3230 const uint16_t *qmul= dequant_coeff[qp];
3231 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};
3232 int level[16], run[16];
3233 int suffix_length, zeros_left, coeff_num, coeff_token, total_coeff, i, trailing_ones;
3235 //FIXME put trailing_onex into the context
3237 if(n == CHROMA_DC_BLOCK_INDEX){
3238 coeff_token= get_vlc2(gb, chroma_dc_coeff_token_vlc.table, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 1);
3239 total_coeff= coeff_token>>2;
3241 if(n == LUMA_DC_BLOCK_INDEX){
3242 total_coeff= pred_non_zero_count(h, 0);
3243 coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
3244 total_coeff= coeff_token>>2;
3246 total_coeff= pred_non_zero_count(h, n);
3247 coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
3248 total_coeff= coeff_token>>2;
3249 h->non_zero_count_cache[ scan8[n] ]= total_coeff;
3253 //FIXME set last_non_zero?
3258 trailing_ones= coeff_token&3;
3259 tprintf("trailing:%d, total:%d\n", trailing_ones, total_coeff);
3260 assert(total_coeff<=16);
3262 for(i=0; i<trailing_ones; i++){
3263 level[i]= 1 - 2*get_bits1(gb);
3266 suffix_length= total_coeff > 10 && trailing_ones < 3;
3268 for(; i<total_coeff; i++){
3269 const int prefix= get_level_prefix(gb);
3270 int level_code, mask;
3272 if(prefix<14){ //FIXME try to build a large unified VLC table for all this
3274 level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
3276 level_code= (prefix<<suffix_length); //part
3277 }else if(prefix==14){
3279 level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
3281 level_code= prefix + get_bits(gb, 4); //part
3282 }else if(prefix==15){
3283 level_code= (prefix<<suffix_length) + get_bits(gb, 12); //part
3284 if(suffix_length==0) level_code+=15; //FIXME doesnt make (much)sense
3286 av_log(h->s.avctx, AV_LOG_ERROR, "prefix too large at %d %d\n", s->mb_x, s->mb_y);
3290 if(i==trailing_ones && i<3) level_code+= 2; //FIXME split first iteration
3292 mask= -(level_code&1);
3293 level[i]= (((2+level_code)>>1) ^ mask) - mask;
3295 if(suffix_length==0) suffix_length=1; //FIXME split first iteration
3298 if(ABS(level[i]) > (3<<(suffix_length-1)) && suffix_length<6) suffix_length++;
3300 if((2+level_code)>>1) > (3<<(suffix_length-1)) && suffix_length<6) suffix_length++;
3301 /* ? == prefix > 2 or sth */
3303 tprintf("level: %d suffix_length:%d\n", level[i], suffix_length);
3306 if(total_coeff == max_coeff)
3309 if(n == CHROMA_DC_BLOCK_INDEX)
3310 zeros_left= get_vlc2(gb, chroma_dc_total_zeros_vlc[ total_coeff-1 ].table, CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 1);
3312 zeros_left= get_vlc2(gb, total_zeros_vlc[ total_coeff-1 ].table, TOTAL_ZEROS_VLC_BITS, 1);
3315 for(i=0; i<total_coeff-1; i++){
3318 else if(zeros_left < 7){
3319 run[i]= get_vlc2(gb, run_vlc[zeros_left-1].table, RUN_VLC_BITS, 1);
3321 run[i]= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2);
3323 zeros_left -= run[i];
3327 av_log(h->s.avctx, AV_LOG_ERROR, "negative number of zero coeffs at %d %d\n", s->mb_x, s->mb_y);
3331 for(; i<total_coeff-1; i++){
3339 for(i=total_coeff-1; i>=0; i--){ //FIXME merge into rundecode?
3342 coeff_num += run[i] + 1; //FIXME add 1 earlier ?
3343 j= scantable[ coeff_num ];
3348 for(i=total_coeff-1; i>=0; i--){ //FIXME merge into rundecode?
3351 coeff_num += run[i] + 1; //FIXME add 1 earlier ?
3352 j= scantable[ coeff_num ];
3354 block[j]= level[i] * qmul[j];
3355 // printf("%d %d ", block[j], qmul[j]);
3362 * decodes a macroblock
3363 * @returns 0 if ok, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
3365 static int decode_mb_cavlc(H264Context *h){
3366 MpegEncContext * const s = &h->s;
3367 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
3368 int mb_type, partition_count, cbp;
3370 s->dsp.clear_blocks(h->mb); //FIXME avoid if allready clear (move after skip handlong?
3372 tprintf("pic:%d mb:%d/%d\n", h->frame_num, s->mb_x, s->mb_y);
3373 cbp = 0; /* avoid warning. FIXME: find a solution without slowing
3375 if(h->slice_type != I_TYPE && h->slice_type != SI_TYPE){
3376 if(s->mb_skip_run==-1)
3377 s->mb_skip_run= get_ue_golomb(&s->gb);
3379 if (s->mb_skip_run--) {
3383 mb_type= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0;
3385 memset(h->non_zero_count[mb_xy], 0, 16);
3386 memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui
3388 if(h->sps.mb_aff && s->mb_skip_run==0 && (s->mb_y&1)==0){
3389 h->mb_field_decoding_flag= get_bits1(&s->gb);
3392 if(h->mb_field_decoding_flag)
3393 mb_type|= MB_TYPE_INTERLACED;
3395 fill_caches(h, mb_type); //FIXME check what is needed and what not ...
3396 pred_pskip_motion(h, &mx, &my);
3397 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
3398 fill_rectangle( h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4);
3399 write_back_motion(h, mb_type);
3401 s->current_picture.mb_type[mb_xy]= mb_type; //FIXME SKIP type
3402 s->current_picture.qscale_table[mb_xy]= s->qscale;
3403 h->slice_table[ mb_xy ]= h->slice_num;
3405 h->prev_mb_skiped= 1;
3409 if(h->sps.mb_aff /* && !field pic FIXME needed? */){
3411 h->mb_field_decoding_flag = get_bits1(&s->gb);
3413 h->mb_field_decoding_flag=0; //FIXME som ed note ?!
3415 h->prev_mb_skiped= 0;
3417 mb_type= get_ue_golomb(&s->gb);
3418 if(h->slice_type == B_TYPE){
3420 partition_count= b_mb_type_info[mb_type].partition_count;
3421 mb_type= b_mb_type_info[mb_type].type;
3424 goto decode_intra_mb;
3426 }else if(h->slice_type == P_TYPE /*|| h->slice_type == SP_TYPE */){
3428 partition_count= p_mb_type_info[mb_type].partition_count;
3429 mb_type= p_mb_type_info[mb_type].type;
3432 goto decode_intra_mb;
3435 assert(h->slice_type == I_TYPE);
3438 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);
3442 cbp= i_mb_type_info[mb_type].cbp;
3443 h->intra16x16_pred_mode= i_mb_type_info[mb_type].pred_mode;
3444 mb_type= i_mb_type_info[mb_type].type;
3447 if(h->mb_field_decoding_flag)
3448 mb_type |= MB_TYPE_INTERLACED;
3450 s->current_picture.mb_type[mb_xy]= mb_type;
3451 h->slice_table[ mb_xy ]= h->slice_num;
3453 if(IS_INTRA_PCM(mb_type)){
3457 // we assume these blocks are very rare so we dont optimize it
3458 align_get_bits(&s->gb);
3460 ptr= s->gb.buffer + get_bits_count(&s->gb);
3462 for(y=0; y<16; y++){
3463 const int index= 4*(y&3) + 64*(y>>2);
3464 for(x=0; x<16; x++){
3465 h->mb[index + (x&3) + 16*(x>>2)]= *(ptr++);
3469 const int index= 256 + 4*(y&3) + 32*(y>>2);
3471 h->mb[index + (x&3) + 16*(x>>2)]= *(ptr++);
3475 const int index= 256 + 64 + 4*(y&3) + 32*(y>>2);
3477 h->mb[index + (x&3) + 16*(x>>2)]= *(ptr++);
3481 skip_bits(&s->gb, 384); //FIXME check /fix the bitstream readers
3483 //FIXME deblock filter, non_zero_count_cache init ...
3484 memset(h->non_zero_count[mb_xy], 16, 16);
3485 s->current_picture.qscale_table[mb_xy]= s->qscale;
3490 fill_caches(h, mb_type);
3493 if(IS_INTRA(mb_type)){
3494 // init_top_left_availability(h);
3495 if(IS_INTRA4x4(mb_type)){
3498 // fill_intra4x4_pred_table(h);
3499 for(i=0; i<16; i++){
3500 const int mode_coded= !get_bits1(&s->gb);
3501 const int predicted_mode= pred_intra_mode(h, i);
3505 const int rem_mode= get_bits(&s->gb, 3);
3506 if(rem_mode<predicted_mode)
3511 mode= predicted_mode;
3514 h->intra4x4_pred_mode_cache[ scan8[i] ] = mode;
3516 write_back_intra_pred_mode(h);
3517 if( check_intra4x4_pred_mode(h) < 0)
3520 h->intra16x16_pred_mode= check_intra_pred_mode(h, h->intra16x16_pred_mode);
3521 if(h->intra16x16_pred_mode < 0)
3524 h->chroma_pred_mode= get_ue_golomb(&s->gb);
3526 h->chroma_pred_mode= check_intra_pred_mode(h, h->chroma_pred_mode);
3527 if(h->chroma_pred_mode < 0)
3529 }else if(partition_count==4){
3530 int i, j, sub_partition_count[4], list, ref[2][4];
3532 if(h->slice_type == B_TYPE){
3534 h->sub_mb_type[i]= get_ue_golomb(&s->gb);
3535 if(h->sub_mb_type[i] >=13){
3536 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);
3539 sub_partition_count[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
3540 h->sub_mb_type[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].type;
3543 assert(h->slice_type == P_TYPE || h->slice_type == SP_TYPE); //FIXME SP correct ?
3545 h->sub_mb_type[i]= get_ue_golomb(&s->gb);
3546 if(h->sub_mb_type[i] >=4){
3547 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);
3550 sub_partition_count[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
3551 h->sub_mb_type[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].type;
3555 for(list=0; list<2; list++){
3556 const int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list];
3557 if(ref_count == 0) continue;
3559 if(IS_DIR(h->sub_mb_type[i], 0, list) && !IS_DIRECT(h->sub_mb_type[i])){
3560 ref[list][i] = get_te0_golomb(&s->gb, ref_count); //FIXME init to 0 before and skip?
3568 for(list=0; list<2; list++){
3569 const int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list];
3570 if(ref_count == 0) continue;
3573 h->ref_cache[list][ scan8[4*i] ]=h->ref_cache[list][ scan8[4*i]+1 ]=
3574 h->ref_cache[list][ scan8[4*i]+8 ]=h->ref_cache[list][ scan8[4*i]+9 ]= ref[list][i];
3576 if(IS_DIR(h->sub_mb_type[i], 0, list) && !IS_DIRECT(h->sub_mb_type[i])){
3577 const int sub_mb_type= h->sub_mb_type[i];
3578 const int block_width= (sub_mb_type & (MB_TYPE_16x16|MB_TYPE_16x8)) ? 2 : 1;
3579 for(j=0; j<sub_partition_count[i]; j++){
3581 const int index= 4*i + block_width*j;
3582 int16_t (* mv_cache)[2]= &h->mv_cache[list][ scan8[index] ];
3583 pred_motion(h, index, block_width, list, h->ref_cache[list][ scan8[index] ], &mx, &my);
3584 mx += get_se_golomb(&s->gb);
3585 my += get_se_golomb(&s->gb);
3586 tprintf("final mv:%d %d\n", mx, my);
3588 if(IS_SUB_8X8(sub_mb_type)){
3589 mv_cache[ 0 ][0]= mv_cache[ 1 ][0]=
3590 mv_cache[ 8 ][0]= mv_cache[ 9 ][0]= mx;
3591 mv_cache[ 0 ][1]= mv_cache[ 1 ][1]=
3592 mv_cache[ 8 ][1]= mv_cache[ 9 ][1]= my;
3593 }else if(IS_SUB_8X4(sub_mb_type)){
3594 mv_cache[ 0 ][0]= mv_cache[ 1 ][0]= mx;
3595 mv_cache[ 0 ][1]= mv_cache[ 1 ][1]= my;
3596 }else if(IS_SUB_4X8(sub_mb_type)){
3597 mv_cache[ 0 ][0]= mv_cache[ 8 ][0]= mx;
3598 mv_cache[ 0 ][1]= mv_cache[ 8 ][1]= my;
3600 assert(IS_SUB_4X4(sub_mb_type));
3601 mv_cache[ 0 ][0]= mx;
3602 mv_cache[ 0 ][1]= my;
3606 uint32_t *p= (uint32_t *)&h->mv_cache[list][ scan8[4*i] ][0];
3612 }else if(!IS_DIRECT(mb_type)){
3613 int list, mx, my, i;
3614 //FIXME we should set ref_idx_l? to 0 if we use that later ...
3615 if(IS_16X16(mb_type)){
3616 for(list=0; list<2; list++){
3617 if(h->ref_count[0]>0){
3618 if(IS_DIR(mb_type, 0, list)){
3619 const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
3620 fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, val, 1);
3624 for(list=0; list<2; list++){
3625 if(IS_DIR(mb_type, 0, list)){
3626 pred_motion(h, 0, 4, list, h->ref_cache[list][ scan8[0] ], &mx, &my);
3627 mx += get_se_golomb(&s->gb);
3628 my += get_se_golomb(&s->gb);
3629 tprintf("final mv:%d %d\n", mx, my);
3631 fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, pack16to32(mx,my), 4);
3635 else if(IS_16X8(mb_type)){
3636 for(list=0; list<2; list++){
3637 if(h->ref_count[list]>0){
3639 if(IS_DIR(mb_type, i, list)){
3640 const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
3641 fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, val, 1);
3646 for(list=0; list<2; list++){
3648 if(IS_DIR(mb_type, i, list)){
3649 pred_16x8_motion(h, 8*i, list, h->ref_cache[list][scan8[0] + 16*i], &mx, &my);
3650 mx += get_se_golomb(&s->gb);
3651 my += get_se_golomb(&s->gb);
3652 tprintf("final mv:%d %d\n", mx, my);
3654 fill_rectangle(h->mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, pack16to32(mx,my), 4);
3659 assert(IS_8X16(mb_type));
3660 for(list=0; list<2; list++){
3661 if(h->ref_count[list]>0){
3663 if(IS_DIR(mb_type, i, list)){ //FIXME optimize
3664 const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
3665 fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, val, 1);
3670 for(list=0; list<2; list++){
3672 if(IS_DIR(mb_type, i, list)){
3673 pred_8x16_motion(h, i*4, list, h->ref_cache[list][ scan8[0] + 2*i ], &mx, &my);
3674 mx += get_se_golomb(&s->gb);
3675 my += get_se_golomb(&s->gb);
3676 tprintf("final mv:%d %d\n", mx, my);
3678 fill_rectangle(h->mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, pack16to32(mx,my), 4);
3685 if(IS_INTER(mb_type))
3686 write_back_motion(h, mb_type);
3688 if(!IS_INTRA16x16(mb_type)){
3689 cbp= get_ue_golomb(&s->gb);
3691 av_log(h->s.avctx, AV_LOG_ERROR, "cbp too large (%d) at %d %d\n", cbp, s->mb_x, s->mb_y);
3695 if(IS_INTRA4x4(mb_type))
3696 cbp= golomb_to_intra4x4_cbp[cbp];
3698 cbp= golomb_to_inter_cbp[cbp];
3701 if(cbp || IS_INTRA16x16(mb_type)){
3702 int i8x8, i4x4, chroma_idx;
3703 int chroma_qp, dquant;
3704 GetBitContext *gb= IS_INTRA(mb_type) ? h->intra_gb_ptr : h->inter_gb_ptr;
3705 const uint8_t *scan, *dc_scan;
3707 // fill_non_zero_count_cache(h);
3709 if(IS_INTERLACED(mb_type)){
3711 dc_scan= luma_dc_field_scan;
3714 dc_scan= luma_dc_zigzag_scan;
3717 dquant= get_se_golomb(&s->gb);
3719 if( dquant > 25 || dquant < -26 ){
3720 av_log(h->s.avctx, AV_LOG_ERROR, "dquant out of range (%d) at %d %d\n", dquant, s->mb_x, s->mb_y);
3724 s->qscale += dquant;
3725 if(((unsigned)s->qscale) > 51){
3726 if(s->qscale<0) s->qscale+= 52;
3727 else s->qscale-= 52;
3730 h->chroma_qp= chroma_qp= get_chroma_qp(h, s->qscale);
3731 if(IS_INTRA16x16(mb_type)){
3732 if( decode_residual(h, h->intra_gb_ptr, h->mb, LUMA_DC_BLOCK_INDEX, dc_scan, s->qscale, 16) < 0){
3733 return -1; //FIXME continue if partotioned and other retirn -1 too
3736 assert((cbp&15) == 0 || (cbp&15) == 15);
3739 for(i8x8=0; i8x8<4; i8x8++){
3740 for(i4x4=0; i4x4<4; i4x4++){
3741 const int index= i4x4 + 4*i8x8;
3742 if( decode_residual(h, h->intra_gb_ptr, h->mb + 16*index, index, scan + 1, s->qscale, 15) < 0 ){
3748 fill_rectangle(&h->non_zero_count_cache[scan8[0]], 4, 4, 8, 0, 1);
3751 for(i8x8=0; i8x8<4; i8x8++){
3752 if(cbp & (1<<i8x8)){
3753 for(i4x4=0; i4x4<4; i4x4++){
3754 const int index= i4x4 + 4*i8x8;
3756 if( decode_residual(h, gb, h->mb + 16*index, index, scan, s->qscale, 16) <0 ){
3761 uint8_t * const nnz= &h->non_zero_count_cache[ scan8[4*i8x8] ];
3762 nnz[0] = nnz[1] = nnz[8] = nnz[9] = 0;
3768 for(chroma_idx=0; chroma_idx<2; chroma_idx++)
3769 if( decode_residual(h, gb, h->mb + 256 + 16*4*chroma_idx, CHROMA_DC_BLOCK_INDEX, chroma_dc_scan, chroma_qp, 4) < 0){
3775 for(chroma_idx=0; chroma_idx<2; chroma_idx++){
3776 for(i4x4=0; i4x4<4; i4x4++){
3777 const int index= 16 + 4*chroma_idx + i4x4;
3778 if( decode_residual(h, gb, h->mb + 16*index, index, scan + 1, chroma_qp, 15) < 0){
3784 uint8_t * const nnz= &h->non_zero_count_cache[0];
3785 nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =
3786 nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
3789 uint8_t * const nnz= &h->non_zero_count_cache[0];
3790 fill_rectangle(&nnz[scan8[0]], 4, 4, 8, 0, 1);
3791 nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =
3792 nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
3794 s->current_picture.qscale_table[mb_xy]= s->qscale;
3795 write_back_non_zero_count(h);
3800 static int decode_cabac_intra_mb_type(H264Context *h, int ctx_base, int intra_slice) {
3801 uint8_t *state= &h->cabac_state[ctx_base];
3805 MpegEncContext * const s = &h->s;
3806 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
3808 if( s->mb_x > 0 && !IS_INTRA4x4( s->current_picture.mb_type[mb_xy-1] ) )
3810 if( s->mb_y > 0 && !IS_INTRA4x4( s->current_picture.mb_type[mb_xy-s->mb_stride] ) )
3812 if( get_cabac( &h->cabac, &state[ctx] ) == 0 )
3813 return 0; /* I4x4 */
3816 if( get_cabac( &h->cabac, &state[0] ) == 0 )
3817 return 0; /* I4x4 */
3820 if( get_cabac_terminate( &h->cabac ) )
3821 return 25; /* PCM */
3823 mb_type = 1; /* I16x16 */
3824 if( get_cabac( &h->cabac, &state[1] ) )
3825 mb_type += 12; /* cbp_luma != 0 */
3827 if( get_cabac( &h->cabac, &state[2] ) ) {
3828 if( get_cabac( &h->cabac, &state[2+intra_slice] ) )
3829 mb_type += 4 * 2; /* cbp_chroma == 2 */
3831 mb_type += 4 * 1; /* cbp_chroma == 1 */
3833 if( get_cabac( &h->cabac, &state[3+intra_slice] ) )
3835 if( get_cabac( &h->cabac, &state[3+2*intra_slice] ) )
3840 static int decode_cabac_mb_type( H264Context *h ) {
3841 MpegEncContext * const s = &h->s;
3843 if( h->slice_type == I_TYPE ) {
3844 return decode_cabac_intra_mb_type(h, 3, 1);
3845 } else if( h->slice_type == P_TYPE ) {
3846 if( get_cabac( &h->cabac, &h->cabac_state[14] ) == 0 ) {
3848 if( get_cabac( &h->cabac, &h->cabac_state[15] ) == 0 ) {
3849 if( get_cabac( &h->cabac, &h->cabac_state[16] ) == 0 )
3850 return 0; /* P_L0_D16x16; */
3852 return 3; /* P_8x8; */
3854 if( get_cabac( &h->cabac, &h->cabac_state[17] ) == 0 )
3855 return 2; /* P_L0_D8x16; */
3857 return 1; /* P_L0_D16x8; */
3860 return decode_cabac_intra_mb_type(h, 17, 0) + 5;
3862 } else if( h->slice_type == B_TYPE ) {
3863 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
3867 if( s->mb_x > 0 && !IS_SKIP( s->current_picture.mb_type[mb_xy-1] )
3868 && !IS_DIRECT( s->current_picture.mb_type[mb_xy-1] ) )
3870 if( s->mb_y > 0 && !IS_SKIP( s->current_picture.mb_type[mb_xy-s->mb_stride] )
3871 && !IS_DIRECT( s->current_picture.mb_type[mb_xy-s->mb_stride] ) )
3874 if( !get_cabac( &h->cabac, &h->cabac_state[27+ctx] ) )
3875 return 0; /* B_Direct_16x16 */
3877 if( !get_cabac( &h->cabac, &h->cabac_state[27+3] ) ) {
3878 return 1 + get_cabac( &h->cabac, &h->cabac_state[27+5] ); /* B_L[01]_16x16 */
3881 bits = get_cabac( &h->cabac, &h->cabac_state[27+4] ) << 3;
3882 bits|= get_cabac( &h->cabac, &h->cabac_state[27+5] ) << 2;
3883 bits|= get_cabac( &h->cabac, &h->cabac_state[27+5] ) << 1;
3884 bits|= get_cabac( &h->cabac, &h->cabac_state[27+5] );
3886 return bits + 3; /* B_Bi_16x16 through B_L1_L0_16x8 */
3887 else if( bits == 13 ) {
3888 return decode_cabac_intra_mb_type(h, 32, 0) + 23;
3889 } else if( bits == 14 )
3890 return 11; /* B_L1_L0_8x16 */
3891 else if( bits == 15 )
3892 return 22; /* B_8x8 */
3894 bits= ( bits<<1 ) | get_cabac( &h->cabac, &h->cabac_state[27+5] );
3895 return bits - 4; /* B_L0_Bi_* through B_Bi_Bi_* */
3897 /* TODO SI/SP frames? */
3902 static int decode_cabac_mb_skip( H264Context *h) {
3903 MpegEncContext * const s = &h->s;
3904 const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
3905 const int mba_xy = mb_xy - 1;
3906 const int mbb_xy = mb_xy - s->mb_stride;
3909 if( s->mb_x > 0 && !IS_SKIP( s->current_picture.mb_type[mba_xy] ) )
3911 if( s->mb_y > 0 && !IS_SKIP( s->current_picture.mb_type[mbb_xy] ) )
3914 if( h->slice_type == P_TYPE || h->slice_type == SP_TYPE)
3915 return get_cabac( &h->cabac, &h->cabac_state[11+ctx] );
3917 return get_cabac( &h->cabac, &h->cabac_state[24+ctx] );
3920 static int decode_cabac_mb_intra4x4_pred_mode( H264Context *h, int pred_mode ) {
3923 if( get_cabac( &h->cabac, &h->cabac_state[68] ) )
3926 if( get_cabac( &h->cabac, &h->cabac_state[69] ) )
3928 if( get_cabac( &h->cabac, &h->cabac_state[69] ) )
3930 if( get_cabac( &h->cabac, &h->cabac_state[69] ) )
3932 if( mode >= pred_mode )
3938 static int decode_cabac_mb_chroma_pre_mode( H264Context *h) {
3939 MpegEncContext * const s = &h->s;
3940 const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
3941 const int mba_xy = mb_xy - 1;
3942 const int mbb_xy = mb_xy - s->mb_stride;
3946 /* No need to test for IS_INTRA4x4 and IS_INTRA16x16, as we set chroma_pred_mode_table to 0 */
3947 if( s->mb_x > 0 && h->chroma_pred_mode_table[mba_xy] != 0 )
3950 if( s->mb_y > 0 && h->chroma_pred_mode_table[mbb_xy] != 0 )
3953 if( get_cabac( &h->cabac, &h->cabac_state[64+ctx] ) == 0 )
3956 if( get_cabac( &h->cabac, &h->cabac_state[64+3] ) == 0 )
3958 if( get_cabac( &h->cabac, &h->cabac_state[64+3] ) == 0 )
3964 static const uint8_t block_idx_x[16] = {
3965 0, 1, 0, 1, 2, 3, 2, 3, 0, 1, 0, 1, 2, 3, 2, 3
3967 static const uint8_t block_idx_y[16] = {
3968 0, 0, 1, 1, 0, 0, 1, 1, 2, 2, 3, 3, 2, 2, 3, 3
3970 static const uint8_t block_idx_xy[4][4] = {
3977 static int decode_cabac_mb_cbp_luma( H264Context *h) {
3978 MpegEncContext * const s = &h->s;
3979 const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
3984 h->cbp_table[mb_xy] = 0; /* FIXME aaahahahah beurk */
3986 for( i8x8 = 0; i8x8 < 4; i8x8++ ) {
3992 x = block_idx_x[4*i8x8];
3993 y = block_idx_y[4*i8x8];
3997 else if( s->mb_x > 0 )
4002 else if( s->mb_y > 0 )
4003 mbb_xy = mb_xy - s->mb_stride;
4005 /* No need to test for skip as we put 0 for skip block */
4007 int i8x8a = block_idx_xy[(x-1)&0x03][y]/4;
4008 if( ((h->cbp_table[mba_xy] >> i8x8a)&0x01) == 0 )
4013 int i8x8b = block_idx_xy[x][(y-1)&0x03]/4;
4014 if( ((h->cbp_table[mbb_xy] >> i8x8b)&0x01) == 0 )
4018 if( get_cabac( &h->cabac, &h->cabac_state[73 + ctx] ) ) {
4020 h->cbp_table[mb_xy] = cbp; /* FIXME aaahahahah beurk */
4025 static int decode_cabac_mb_cbp_chroma( H264Context *h) {
4029 cbp_a = (h->left_cbp>>4)&0x03;
4030 cbp_b = (h-> top_cbp>>4)&0x03;
4033 if( cbp_a > 0 ) ctx++;
4034 if( cbp_b > 0 ) ctx += 2;
4035 if( get_cabac( &h->cabac, &h->cabac_state[77 + ctx] ) == 0 )
4039 if( cbp_a == 2 ) ctx++;
4040 if( cbp_b == 2 ) ctx += 2;
4041 return 1 + get_cabac( &h->cabac, &h->cabac_state[77 + ctx] );
4043 static int decode_cabac_mb_dqp( H264Context *h) {
4044 MpegEncContext * const s = &h->s;
4050 mbn_xy = s->mb_x + s->mb_y*s->mb_stride - 1;
4052 mbn_xy = s->mb_width - 1 + (s->mb_y-1)*s->mb_stride;
4054 if( mbn_xy >= 0 && h->last_qscale_diff != 0 && ( IS_INTRA16x16(s->current_picture.mb_type[mbn_xy] ) || (h->cbp_table[mbn_xy]&0x3f) ) )
4057 while( get_cabac( &h->cabac, &h->cabac_state[60 + ctx] ) ) {
4068 return -(val + 1)/2;
4070 static int decode_cabac_p_mb_sub_type( H264Context *h ) {
4071 if( get_cabac( &h->cabac, &h->cabac_state[21] ) )
4073 if( !get_cabac( &h->cabac, &h->cabac_state[22] ) )
4075 if( get_cabac( &h->cabac, &h->cabac_state[23] ) )
4079 static int decode_cabac_b_mb_sub_type( H264Context *h ) {
4081 if( !get_cabac( &h->cabac, &h->cabac_state[36] ) )
4082 return 0; /* B_Direct_8x8 */
4083 if( !get_cabac( &h->cabac, &h->cabac_state[37] ) )
4084 return 1 + get_cabac( &h->cabac, &h->cabac_state[39] ); /* B_L0_8x8, B_L1_8x8 */
4086 if( get_cabac( &h->cabac, &h->cabac_state[38] ) ) {
4087 if( get_cabac( &h->cabac, &h->cabac_state[39] ) )
4088 return 11 + get_cabac( &h->cabac, &h->cabac_state[39] ); /* B_L1_4x4, B_Bi_4x4 */
4091 type += 2*get_cabac( &h->cabac, &h->cabac_state[39] );
4092 type += get_cabac( &h->cabac, &h->cabac_state[39] );
4096 static int decode_cabac_mb_ref( H264Context *h, int list, int n ) {
4097 int refa = h->ref_cache[list][scan8[n] - 1];
4098 int refb = h->ref_cache[list][scan8[n] - 8];
4107 while( get_cabac( &h->cabac, &h->cabac_state[54+ctx] ) ) {
4117 static int decode_cabac_mb_mvd( H264Context *h, int list, int n, int l ) {
4118 int amvd = abs( h->mvd_cache[list][scan8[n] - 1][l] ) +
4119 abs( h->mvd_cache[list][scan8[n] - 8][l] );
4120 int ctxbase = (l == 0) ? 40 : 47;
4126 else if( amvd > 32 )
4131 while( mvd < 9 && get_cabac( &h->cabac, &h->cabac_state[ctxbase+ctx] ) ) {
4141 while( get_cabac_bypass( &h->cabac ) ) {
4146 if( get_cabac_bypass( &h->cabac ) )
4150 if( mvd != 0 && get_cabac_bypass( &h->cabac ) )
4156 static int get_cabac_cbf_ctx( H264Context *h, int cat, int idx ) {
4161 nza = h->left_cbp&0x100;
4162 nzb = h-> top_cbp&0x100;
4163 } else if( cat == 1 || cat == 2 ) {
4164 nza = h->non_zero_count_cache[scan8[idx] - 1];
4165 nzb = h->non_zero_count_cache[scan8[idx] - 8];
4166 } else if( cat == 3 ) {
4167 nza = (h->left_cbp>>(6+idx))&0x01;
4168 nzb = (h-> top_cbp>>(6+idx))&0x01;
4171 nza = h->non_zero_count_cache[scan8[16+idx] - 1];
4172 nzb = h->non_zero_count_cache[scan8[16+idx] - 8];
4181 return ctx + 4 * cat;
4184 static int decode_cabac_residual( H264Context *h, DCTELEM *block, int cat, int n, const uint8_t *scantable, int qp, int max_coeff) {
4185 const int mb_xy = h->s.mb_x + h->s.mb_y*h->s.mb_stride;
4186 const uint16_t *qmul= dequant_coeff[qp];
4187 static const int significant_coeff_flag_offset[5] = { 0, 15, 29, 44, 47 };
4188 static const int coeff_abs_level_m1_offset[5] = {227+ 0, 227+10, 227+20, 227+30, 227+39 };
4194 int coeff_count = 0;
4197 int abslevelgt1 = 0;
4199 /* cat: 0-> DC 16x16 n = 0
4200 * 1-> AC 16x16 n = luma4x4idx
4201 * 2-> Luma4x4 n = luma4x4idx
4202 * 3-> DC Chroma n = iCbCr
4203 * 4-> AC Chroma n = 4 * iCbCr + chroma4x4idx
4206 /* read coded block flag */
4207 if( get_cabac( &h->cabac, &h->cabac_state[85 + get_cabac_cbf_ctx( h, cat, n ) ] ) == 0 ) {
4208 if( cat == 1 || cat == 2 )
4209 h->non_zero_count_cache[scan8[n]] = 0;
4211 h->non_zero_count_cache[scan8[16+n]] = 0;
4216 for(last= 0; last < max_coeff - 1; last++) {
4217 if( get_cabac( &h->cabac, &h->cabac_state[105+significant_coeff_flag_offset[cat]+last] )) {
4218 index[coeff_count++] = last;
4219 if( get_cabac( &h->cabac, &h->cabac_state[166+significant_coeff_flag_offset[cat]+last] ) ) {
4225 if( last == max_coeff -1 ) {
4226 index[coeff_count++] = last;
4229 if( cat == 0 && coeff_count > 0 )
4230 h->cbp_table[mb_xy] |= 0x100;
4231 else if( cat == 1 || cat == 2 )
4232 h->non_zero_count_cache[scan8[n]] = coeff_count;
4233 else if( cat == 3 && coeff_count > 0 )
4234 h->cbp_table[mb_xy] |= 0x40 << n;
4236 h->non_zero_count_cache[scan8[16+n]] = coeff_count;
4238 for( i = coeff_count - 1; i >= 0; i-- ) {
4239 int ctx = (abslevelgt1 != 0 ? 0 : FFMIN( 4, abslevel1 + 1 )) + coeff_abs_level_m1_offset[cat];
4241 if( get_cabac( &h->cabac, &h->cabac_state[ctx] ) == 0 ) {
4242 if( get_cabac_bypass( &h->cabac ) )
4250 ctx = 5 + FFMIN( 4, abslevelgt1 ) + coeff_abs_level_m1_offset[cat];
4251 while( coeff_abs < 15 && get_cabac( &h->cabac, &h->cabac_state[ctx] ) ) {
4255 if( coeff_abs >= 15 ) {
4257 while( get_cabac_bypass( &h->cabac ) ) {
4258 coeff_abs += 1 << j;
4263 if( get_cabac_bypass( &h->cabac ) )
4264 coeff_abs += 1 << j ;
4268 if( get_cabac_bypass( &h->cabac ) )
4269 coeff[i] = -coeff_abs;
4271 coeff[i] = coeff_abs;
4277 if( cat == 0 || cat == 3 ) { /* DC */
4278 for(i = 0; i < coeff_count; i++) {
4279 block[scantable[ index[i] ]] = coeff[i];
4283 for(i = 0; i < coeff_count; i++) {
4284 int j= scantable[index[i]];
4285 block[j] = coeff[i] * qmul[j];
4292 * decodes a macroblock
4293 * @returns 0 if ok, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
4295 static int decode_mb_cabac(H264Context *h) {
4296 MpegEncContext * const s = &h->s;
4297 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
4298 int mb_type, partition_count, cbp = 0;
4300 s->dsp.clear_blocks(h->mb); //FIXME avoid if allready clear (move after skip handlong?)
4302 if( h->sps.mb_aff ) {
4303 av_log( h->s.avctx, AV_LOG_ERROR, "Fields not supported with CABAC\n" );
4307 if( h->slice_type != I_TYPE && h->slice_type != SI_TYPE ) {
4308 /* read skip flags */
4309 if( decode_cabac_mb_skip( h ) ) {
4314 mb_type= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0|MB_TYPE_SKIP;
4316 memset(h->non_zero_count[mb_xy], 0, 16);
4317 memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui
4319 if(h->sps.mb_aff && s->mb_skip_run==0 && (s->mb_y&1)==0){
4320 h->mb_field_decoding_flag= get_bits1(&s->gb);
4322 if(h->mb_field_decoding_flag)
4323 mb_type|= MB_TYPE_INTERLACED;
4326 fill_caches(h, mb_type); //FIXME check what is needed and what not ...
4327 pred_pskip_motion(h, &mx, &my);
4328 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
4329 fill_rectangle( h->mvd_cache[0][scan8[0]], 4, 4, 8, pack16to32(0,0), 4);
4330 fill_rectangle( h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4);
4331 write_back_motion(h, mb_type);
4333 s->current_picture.mb_type[mb_xy]= mb_type; //FIXME SKIP type
4334 s->current_picture.qscale_table[mb_xy]= s->qscale;
4335 h->slice_table[ mb_xy ]= h->slice_num;
4336 h->cbp_table[mb_xy] = 0;
4337 h->chroma_pred_mode_table[mb_xy] = 0;
4338 h->last_qscale_diff = 0;
4340 h->prev_mb_skiped= 1;
4346 h->prev_mb_skiped = 0;
4348 if( ( mb_type = decode_cabac_mb_type( h ) ) < 0 ) {
4349 av_log( h->s.avctx, AV_LOG_ERROR, "decode_cabac_mb_type failed\n" );
4353 if( h->slice_type == B_TYPE ) {
4355 partition_count= b_mb_type_info[mb_type].partition_count;
4356 mb_type= b_mb_type_info[mb_type].type;
4359 goto decode_intra_mb;
4361 } else if( h->slice_type == P_TYPE ) {
4363 partition_count= p_mb_type_info[mb_type].partition_count;
4364 mb_type= p_mb_type_info[mb_type].type;
4367 goto decode_intra_mb;
4370 assert(h->slice_type == I_TYPE);
4372 partition_count = 0;
4373 cbp= i_mb_type_info[mb_type].cbp;
4374 h->intra16x16_pred_mode= i_mb_type_info[mb_type].pred_mode;
4375 mb_type= i_mb_type_info[mb_type].type;
4378 if(h->mb_field_decoding_flag)
4379 mb_type |= MB_TYPE_INTERLACED;
4382 s->current_picture.mb_type[mb_xy]= mb_type;
4383 h->slice_table[ mb_xy ]= h->slice_num;
4385 if(IS_INTRA_PCM(mb_type)) {
4388 h->cbp_table[mb_xy] = 0xf +4*2; //FIXME ?!
4389 h->cbp_table[mb_xy] |= 0x1C0;
4390 h->chroma_pred_mode_table[mb_xy] = 0;
4391 s->current_picture.qscale_table[mb_xy]= s->qscale;
4395 fill_caches(h, mb_type);
4397 if( IS_INTRA( mb_type ) ) {
4398 if( IS_INTRA4x4( mb_type ) ) {
4400 for( i = 0; i < 16; i++ ) {
4401 int pred = pred_intra_mode( h, i );
4402 h->intra4x4_pred_mode_cache[ scan8[i] ] = decode_cabac_mb_intra4x4_pred_mode( h, pred );
4404 //av_log( s->avctx, AV_LOG_ERROR, "i4x4 pred=%d mode=%d\n", pred, h->intra4x4_pred_mode_cache[ scan8[i] ] );
4406 write_back_intra_pred_mode(h);
4407 if( check_intra4x4_pred_mode(h) < 0 ) return -1;
4409 h->intra16x16_pred_mode= check_intra_pred_mode( h, h->intra16x16_pred_mode );
4410 if( h->intra16x16_pred_mode < 0 ) return -1;
4412 h->chroma_pred_mode_table[mb_xy] =
4413 h->chroma_pred_mode = decode_cabac_mb_chroma_pre_mode( h );
4415 h->chroma_pred_mode= check_intra_pred_mode( h, h->chroma_pred_mode );
4416 if( h->chroma_pred_mode < 0 ) return -1;
4417 } else if( partition_count == 4 ) {
4418 int i, j, sub_partition_count[4], list, ref[2][4];
4420 if( h->slice_type == B_TYPE ) {
4421 for( i = 0; i < 4; i++ ) {
4422 h->sub_mb_type[i] = decode_cabac_b_mb_sub_type( h );
4423 sub_partition_count[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
4424 h->sub_mb_type[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].type;
4427 for( i = 0; i < 4; i++ ) {
4428 h->sub_mb_type[i] = decode_cabac_p_mb_sub_type( h );
4429 sub_partition_count[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
4430 h->sub_mb_type[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].type;
4434 for( list = 0; list < 2; list++ ) {
4435 if( h->ref_count[list] > 0 ) {
4436 for( i = 0; i < 4; i++ ) {
4437 if(IS_DIR(h->sub_mb_type[i], 0, list) && !IS_DIRECT(h->sub_mb_type[i])){
4438 if( h->ref_count[list] > 1 )
4439 ref[list][i] = decode_cabac_mb_ref( h, list, 4*i );
4445 h->ref_cache[list][ scan8[4*i]+1 ]=
4446 h->ref_cache[list][ scan8[4*i]+8 ]=h->ref_cache[list][ scan8[4*i]+9 ]= ref[list][i];
4451 for(list=0; list<2; list++){
4453 h->ref_cache[list][ scan8[4*i] ]=h->ref_cache[list][ scan8[4*i]+1 ];
4455 if(IS_DIR(h->sub_mb_type[i], 0, list) && !IS_DIRECT(h->sub_mb_type[i])){
4456 const int sub_mb_type= h->sub_mb_type[i];
4457 const int block_width= (sub_mb_type & (MB_TYPE_16x16|MB_TYPE_16x8)) ? 2 : 1;
4458 for(j=0; j<sub_partition_count[i]; j++){
4461 const int index= 4*i + block_width*j;
4462 int16_t (* mv_cache)[2]= &h->mv_cache[list][ scan8[index] ];
4463 int16_t (* mvd_cache)[2]= &h->mvd_cache[list][ scan8[index] ];
4464 pred_motion(h, index, block_width, list, h->ref_cache[list][ scan8[index] ], &mpx, &mpy);
4466 mx = mpx + decode_cabac_mb_mvd( h, list, index, 0 );
4467 my = mpy + decode_cabac_mb_mvd( h, list, index, 1 );
4468 tprintf("final mv:%d %d\n", mx, my);
4470 if(IS_SUB_8X8(sub_mb_type)){
4471 mv_cache[ 0 ][0]= mv_cache[ 1 ][0]=
4472 mv_cache[ 8 ][0]= mv_cache[ 9 ][0]= mx;
4473 mv_cache[ 0 ][1]= mv_cache[ 1 ][1]=
4474 mv_cache[ 8 ][1]= mv_cache[ 9 ][1]= my;
4476 mvd_cache[ 0 ][0]= mvd_cache[ 1 ][0]=
4477 mvd_cache[ 8 ][0]= mvd_cache[ 9 ][0]= mx - mpx;
4478 mvd_cache[ 0 ][1]= mvd_cache[ 1 ][1]=
4479 mvd_cache[ 8 ][1]= mvd_cache[ 9 ][1]= my - mpy;
4480 }else if(IS_SUB_8X4(sub_mb_type)){
4481 mv_cache[ 0 ][0]= mv_cache[ 1 ][0]= mx;
4482 mv_cache[ 0 ][1]= mv_cache[ 1 ][1]= my;
4484 mvd_cache[ 0 ][0]= mvd_cache[ 1 ][0]= mx- mpx;
4485 mvd_cache[ 0 ][1]= mvd_cache[ 1 ][1]= my - mpy;
4486 }else if(IS_SUB_4X8(sub_mb_type)){
4487 mv_cache[ 0 ][0]= mv_cache[ 8 ][0]= mx;
4488 mv_cache[ 0 ][1]= mv_cache[ 8 ][1]= my;
4490 mvd_cache[ 0 ][0]= mvd_cache[ 8 ][0]= mx - mpx;
4491 mvd_cache[ 0 ][1]= mvd_cache[ 8 ][1]= my - mpy;
4493 assert(IS_SUB_4X4(sub_mb_type));
4494 mv_cache[ 0 ][0]= mx;
4495 mv_cache[ 0 ][1]= my;
4497 mvd_cache[ 0 ][0]= mx - mpx;
4498 mvd_cache[ 0 ][1]= my - mpy;
4502 uint32_t *p= (uint32_t *)&h->mv_cache[list][ scan8[4*i] ][0];
4503 uint32_t *pd= (uint32_t *)&h->mvd_cache[list][ scan8[4*i] ][0];
4504 p[0] = p[1] = p[8] = p[9] = 0;
4505 pd[0]= pd[1]= pd[8]= pd[9]= 0;
4509 } else if( !IS_DIRECT(mb_type) ) {
4510 int list, mx, my, i, mpx, mpy;
4511 if(IS_16X16(mb_type)){
4512 for(list=0; list<2; list++){
4513 if(IS_DIR(mb_type, 0, list)){
4514 if(h->ref_count[list] > 0 ){
4515 const int ref = h->ref_count[list] > 1 ? decode_cabac_mb_ref( h, list, 0 ) : 0;
4516 fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, ref, 1);
4520 for(list=0; list<2; list++){
4521 if(IS_DIR(mb_type, 0, list)){
4522 pred_motion(h, 0, 4, list, h->ref_cache[list][ scan8[0] ], &mpx, &mpy);
4524 mx = mpx + decode_cabac_mb_mvd( h, list, 0, 0 );
4525 my = mpy + decode_cabac_mb_mvd( h, list, 0, 1 );
4526 tprintf("final mv:%d %d\n", mx, my);
4528 fill_rectangle(h->mvd_cache[list][ scan8[0] ], 4, 4, 8, pack16to32(mx-mpx,my-mpy), 4);
4529 fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, pack16to32(mx,my), 4);
4533 else if(IS_16X8(mb_type)){
4534 for(list=0; list<2; list++){
4535 if(h->ref_count[list]>0){
4537 if(IS_DIR(mb_type, i, list)){
4538 const int ref= h->ref_count[list] > 1 ? decode_cabac_mb_ref( h, list, 8*i ) : 0;
4539 fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, ref, 1);
4544 for(list=0; list<2; list++){
4546 if(IS_DIR(mb_type, i, list)){
4547 pred_16x8_motion(h, 8*i, list, h->ref_cache[list][scan8[0] + 16*i], &mpx, &mpy);
4548 mx = mpx + decode_cabac_mb_mvd( h, list, 8*i, 0 );
4549 my = mpy + decode_cabac_mb_mvd( h, list, 8*i, 1 );
4550 tprintf("final mv:%d %d\n", mx, my);
4552 fill_rectangle(h->mvd_cache[list][ scan8[0] + 16*i ], 4, 2, 8, pack16to32(mx-mpx,my-mpy), 4);
4553 fill_rectangle(h->mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, pack16to32(mx,my), 4);
4558 assert(IS_8X16(mb_type));
4559 for(list=0; list<2; list++){
4560 if(h->ref_count[list]>0){
4562 if(IS_DIR(mb_type, i, list)){ //FIXME optimize
4563 const int ref= h->ref_count[list] > 1 ? decode_cabac_mb_ref( h, list, 4*i ) : 0;
4564 fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, ref, 1);
4569 for(list=0; list<2; list++){
4571 if(IS_DIR(mb_type, i, list)){
4572 pred_8x16_motion(h, i*4, list, h->ref_cache[list][ scan8[0] + 2*i ], &mpx, &mpy);
4573 mx = mpx + decode_cabac_mb_mvd( h, list, 4*i, 0 );
4574 my = mpy + decode_cabac_mb_mvd( h, list, 4*i, 1 );
4576 tprintf("final mv:%d %d\n", mx, my);
4577 fill_rectangle(h->mvd_cache[list][ scan8[0] + 2*i ], 2, 4, 8, pack16to32(mx-mpx,my-mpy), 4);
4578 fill_rectangle(h->mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, pack16to32(mx,my), 4);
4585 if( IS_INTER( mb_type ) ) {
4586 h->chroma_pred_mode_table[mb_xy] = 0;
4587 write_back_motion( h, mb_type );
4590 if( !IS_INTRA16x16( mb_type ) ) {
4591 cbp = decode_cabac_mb_cbp_luma( h );
4592 cbp |= decode_cabac_mb_cbp_chroma( h ) << 4;
4595 h->cbp_table[mb_xy] = cbp;
4597 if( cbp || IS_INTRA16x16( mb_type ) ) {
4598 const uint8_t *scan, *dc_scan;
4601 if(IS_INTERLACED(mb_type)){
4603 dc_scan= luma_dc_field_scan;
4606 dc_scan= luma_dc_zigzag_scan;
4609 h->last_qscale_diff = dqp = decode_cabac_mb_dqp( h );
4611 if(((unsigned)s->qscale) > 51){
4612 if(s->qscale<0) s->qscale+= 52;
4613 else s->qscale-= 52;
4615 h->chroma_qp = get_chroma_qp(h, s->qscale);
4617 if( IS_INTRA16x16( mb_type ) ) {
4619 //av_log( s->avctx, AV_LOG_ERROR, "INTRA16x16 DC\n" );
4620 if( decode_cabac_residual( h, h->mb, 0, 0, dc_scan, s->qscale, 16) < 0)
4623 for( i = 0; i < 16; i++ ) {
4624 //av_log( s->avctx, AV_LOG_ERROR, "INTRA16x16 AC:%d\n", i );
4625 if( decode_cabac_residual(h, h->mb + 16*i, 1, i, scan + 1, s->qscale, 15) < 0 )
4629 fill_rectangle(&h->non_zero_count_cache[scan8[0]], 4, 4, 8, 0, 1);
4633 for( i8x8 = 0; i8x8 < 4; i8x8++ ) {
4634 if( cbp & (1<<i8x8) ) {
4635 for( i4x4 = 0; i4x4 < 4; i4x4++ ) {
4636 const int index = 4*i8x8 + i4x4;
4637 //av_log( s->avctx, AV_LOG_ERROR, "Luma4x4: %d\n", index );
4638 if( decode_cabac_residual(h, h->mb + 16*index, 2, index, scan, s->qscale, 16) < 0 )
4642 uint8_t * const nnz= &h->non_zero_count_cache[ scan8[4*i8x8] ];
4643 nnz[0] = nnz[1] = nnz[8] = nnz[9] = 0;
4650 for( c = 0; c < 2; c++ ) {
4651 //av_log( s->avctx, AV_LOG_ERROR, "INTRA C%d-DC\n",c );
4652 if( decode_cabac_residual(h, h->mb + 256 + 16*4*c, 3, c, chroma_dc_scan, h->chroma_qp, 4) < 0)
4659 for( c = 0; c < 2; c++ ) {
4660 for( i = 0; i < 4; i++ ) {
4661 const int index = 16 + 4 * c + i;
4662 //av_log( s->avctx, AV_LOG_ERROR, "INTRA C%d-AC %d\n",c, index - 16 );
4663 if( decode_cabac_residual(h, h->mb + 16*index, 4, index - 16, scan + 1, h->chroma_qp, 15) < 0)
4668 uint8_t * const nnz= &h->non_zero_count_cache[0];
4669 nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =
4670 nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
4673 memset( &h->non_zero_count_cache[8], 0, 8*5 );
4676 s->current_picture.qscale_table[mb_xy]= s->qscale;
4677 write_back_non_zero_count(h);
4683 static void filter_mb_edgev( H264Context *h, uint8_t *pix, int stride, int bS[4], int qp ) {
4685 const int index_a = clip( qp + h->slice_alpha_c0_offset, 0, 51 );
4686 const int alpha = alpha_table[index_a];
4687 const int beta = beta_table[clip( qp + h->slice_beta_offset, 0, 51 )];
4689 for( i = 0; i < 4; i++ ) {
4696 const int tc0 = tc0_table[index_a][bS[i] - 1];
4697 /* 4px edge length */
4698 for( d = 0; d < 4; d++ ) {
4699 const int p0 = pix[-1];
4700 const int p1 = pix[-2];
4701 const int p2 = pix[-3];
4702 const int q0 = pix[0];
4703 const int q1 = pix[1];
4704 const int q2 = pix[2];
4706 if( ABS( p0 - q0 ) < alpha &&
4707 ABS( p1 - p0 ) < beta &&
4708 ABS( q1 - q0 ) < beta ) {
4712 if( ABS( p2 - p0 ) < beta ) {
4713 pix[-2] = p1 + clip( ( p2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( p1 << 1 ) ) >> 1, -tc0, tc0 );
4716 if( ABS( q2 - q0 ) < beta ) {
4717 pix[1] = q1 + clip( ( q2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( q1 << 1 ) ) >> 1, -tc0, tc0 );
4721 i_delta = clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );
4722 pix[-1] = clip_uint8( p0 + i_delta ); /* p0' */
4723 pix[0] = clip_uint8( q0 - i_delta ); /* q0' */
4728 /* 4px edge length */
4729 for( d = 0; d < 4; d++ ) {
4730 const int p0 = pix[-1];
4731 const int p1 = pix[-2];
4732 const int p2 = pix[-3];
4734 const int q0 = pix[0];
4735 const int q1 = pix[1];
4736 const int q2 = pix[2];
4738 if( ABS( p0 - q0 ) < alpha &&
4739 ABS( p1 - p0 ) < beta &&
4740 ABS( q1 - q0 ) < beta ) {
4742 if(ABS( p0 - q0 ) < (( alpha >> 2 ) + 2 )){
4743 if( ABS( p2 - p0 ) < beta)
4745 const int p3 = pix[-4];
4747 pix[-1] = ( p2 + 2*p1 + 2*p0 + 2*q0 + q1 + 4 ) >> 3;
4748 pix[-2] = ( p2 + p1 + p0 + q0 + 2 ) >> 2;
4749 pix[-3] = ( 2*p3 + 3*p2 + p1 + p0 + q0 + 4 ) >> 3;
4752 pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
4754 if( ABS( q2 - q0 ) < beta)
4756 const int q3 = pix[3];
4758 pix[0] = ( p1 + 2*p0 + 2*q0 + 2*q1 + q2 + 4 ) >> 3;
4759 pix[1] = ( p0 + q0 + q1 + q2 + 2 ) >> 2;
4760 pix[2] = ( 2*q3 + 3*q2 + q1 + q0 + p0 + 4 ) >> 3;
4763 pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
4767 pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
4768 pix[ 0] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
4776 static void filter_mb_edgecv( H264Context *h, uint8_t *pix, int stride, int bS[4], int qp ) {
4778 const int index_a = clip( qp + h->slice_alpha_c0_offset, 0, 51 );
4779 const int alpha = alpha_table[index_a];
4780 const int beta = beta_table[clip( qp + h->slice_beta_offset, 0, 51 )];
4782 for( i = 0; i < 4; i++ ) {
4789 const int tc = tc0_table[index_a][bS[i] - 1] + 1;
4790 /* 2px edge length (because we use same bS than the one for luma) */
4791 for( d = 0; d < 2; d++ ){
4792 const int p0 = pix[-1];
4793 const int p1 = pix[-2];
4794 const int q0 = pix[0];
4795 const int q1 = pix[1];
4797 if( ABS( p0 - q0 ) < alpha &&
4798 ABS( p1 - p0 ) < beta &&
4799 ABS( q1 - q0 ) < beta ) {
4800 const int i_delta = clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );
4802 pix[-1] = clip_uint8( p0 + i_delta ); /* p0' */
4803 pix[0] = clip_uint8( q0 - i_delta ); /* q0' */
4808 /* 2px edge length (because we use same bS than the one for luma) */
4809 for( d = 0; d < 2; d++ ){
4810 const int p0 = pix[-1];
4811 const int p1 = pix[-2];
4812 const int q0 = pix[0];
4813 const int q1 = pix[1];
4815 if( ABS( p0 - q0 ) < alpha &&
4816 ABS( p1 - p0 ) < beta &&
4817 ABS( q1 - q0 ) < beta ) {
4819 pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2; /* p0' */
4820 pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2; /* q0' */
4828 static void filter_mb_edgeh( H264Context *h, uint8_t *pix, int stride, int bS[4], int qp ) {
4830 const int index_a = clip( qp + h->slice_alpha_c0_offset, 0, 51 );
4831 const int alpha = alpha_table[index_a];
4832 const int beta = beta_table[clip( qp + h->slice_beta_offset, 0, 51 )];
4833 const int pix_next = stride;
4835 for( i = 0; i < 4; i++ ) {
4842 const int tc0 = tc0_table[index_a][bS[i] - 1];
4843 /* 4px edge length */
4844 for( d = 0; d < 4; d++ ) {
4845 const int p0 = pix[-1*pix_next];
4846 const int p1 = pix[-2*pix_next];
4847 const int p2 = pix[-3*pix_next];
4848 const int q0 = pix[0];
4849 const int q1 = pix[1*pix_next];
4850 const int q2 = pix[2*pix_next];
4852 if( ABS( p0 - q0 ) < alpha &&
4853 ABS( p1 - p0 ) < beta &&
4854 ABS( q1 - q0 ) < beta ) {
4859 if( ABS( p2 - p0 ) < beta ) {
4860 pix[-2*pix_next] = p1 + clip( ( p2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( p1 << 1 ) ) >> 1, -tc0, tc0 );
4863 if( ABS( q2 - q0 ) < beta ) {
4864 pix[pix_next] = q1 + clip( ( q2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( q1 << 1 ) ) >> 1, -tc0, tc0 );
4868 i_delta = clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );
4869 pix[-pix_next] = clip_uint8( p0 + i_delta ); /* p0' */
4870 pix[0] = clip_uint8( q0 - i_delta ); /* q0' */
4875 /* 4px edge length */
4876 for( d = 0; d < 4; d++ ) {
4877 const int p0 = pix[-1*pix_next];
4878 const int p1 = pix[-2*pix_next];
4879 const int p2 = pix[-3*pix_next];
4880 const int q0 = pix[0];
4881 const int q1 = pix[1*pix_next];
4882 const int q2 = pix[2*pix_next];
4884 if( ABS( p0 - q0 ) < alpha &&
4885 ABS( p1 - p0 ) < beta &&
4886 ABS( q1 - q0 ) < beta ) {
4888 const int p3 = pix[-4*pix_next];
4889 const int q3 = pix[ 3*pix_next];
4891 if(ABS( p0 - q0 ) < (( alpha >> 2 ) + 2 )){
4892 if( ABS( p2 - p0 ) < beta) {
4894 pix[-1*pix_next] = ( p2 + 2*p1 + 2*p0 + 2*q0 + q1 + 4 ) >> 3;
4895 pix[-2*pix_next] = ( p2 + p1 + p0 + q0 + 2 ) >> 2;
4896 pix[-3*pix_next] = ( 2*p3 + 3*p2 + p1 + p0 + q0 + 4 ) >> 3;
4899 pix[-1*pix_next] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
4901 if( ABS( q2 - q0 ) < beta) {
4903 pix[0*pix_next] = ( p1 + 2*p0 + 2*q0 + 2*q1 + q2 + 4 ) >> 3;
4904 pix[1*pix_next] = ( p0 + q0 + q1 + q2 + 2 ) >> 2;
4905 pix[2*pix_next] = ( 2*q3 + 3*q2 + q1 + q0 + p0 + 4 ) >> 3;
4908 pix[0*pix_next] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
4912 pix[-1*pix_next] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
4913 pix[ 0*pix_next] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
4922 static void filter_mb_edgech( H264Context *h, uint8_t *pix, int stride, int bS[4], int qp ) {
4924 const int index_a = clip( qp + h->slice_alpha_c0_offset, 0, 51 );
4925 const int alpha = alpha_table[index_a];
4926 const int beta = beta_table[clip( qp + h->slice_beta_offset, 0, 51 )];
4927 const int pix_next = stride;
4929 for( i = 0; i < 4; i++ )
4937 int tc = tc0_table[index_a][bS[i] - 1] + 1;
4938 /* 2px edge length (see deblocking_filter_edgecv) */
4939 for( d = 0; d < 2; d++ ) {
4940 const int p0 = pix[-1*pix_next];
4941 const int p1 = pix[-2*pix_next];
4942 const int q0 = pix[0];
4943 const int q1 = pix[1*pix_next];
4945 if( ABS( p0 - q0 ) < alpha &&
4946 ABS( p1 - p0 ) < beta &&
4947 ABS( q1 - q0 ) < beta ) {
4949 int i_delta = clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );
4951 pix[-pix_next] = clip_uint8( p0 + i_delta ); /* p0' */
4952 pix[0] = clip_uint8( q0 - i_delta ); /* q0' */
4957 /* 2px edge length (see deblocking_filter_edgecv) */
4958 for( d = 0; d < 2; d++ ) {
4959 const int p0 = pix[-1*pix_next];
4960 const int p1 = pix[-2*pix_next];
4961 const int q0 = pix[0];
4962 const int q1 = pix[1*pix_next];
4964 if( ABS( p0 - q0 ) < alpha &&
4965 ABS( p1 - p0 ) < beta &&
4966 ABS( q1 - q0 ) < beta ) {
4968 pix[-pix_next] = ( 2*p1 + p0 + q1 + 2 ) >> 2; /* p0' */
4969 pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2; /* q0' */
4977 static void filter_mb( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr) {
4978 MpegEncContext * const s = &h->s;
4979 const int mb_xy= mb_x + mb_y*s->mb_stride;
4980 int linesize, uvlinesize;
4983 /* FIXME Implement deblocking filter for field MB */
4984 if( h->sps.mb_aff ) {
4987 linesize = s->linesize;
4988 uvlinesize = s->uvlinesize;
4990 /* dir : 0 -> vertical edge, 1 -> horizontal edge */
4991 for( dir = 0; dir < 2; dir++ )
4996 /* test picture boundary */
4997 if( ( dir == 0 && mb_x == 0 ) || ( dir == 1 && mb_y == 0 ) ) {
5000 /* FIXME test slice boundary */
5001 if( h->deblocking_filter == 2 ) {
5005 for( edge = start; edge < 4; edge++ ) {
5006 /* mbn_xy: neighbour macroblock (how that works for field ?) */
5007 int mbn_xy = edge > 0 ? mb_xy : ( dir == 0 ? mb_xy -1 : mb_xy - s->mb_stride );
5011 if( IS_INTRA( s->current_picture.mb_type[mb_xy] ) ||
5012 IS_INTRA( s->current_picture.mb_type[mbn_xy] ) ) {
5013 bS[0] = bS[1] = bS[2] = bS[3] = ( edge == 0 ? 4 : 3 );
5016 for( i = 0; i < 4; i++ ) {
5017 int x = dir == 0 ? edge : i;
5018 int y = dir == 0 ? i : edge;
5019 int b_idx= 8 + 4 + x + 8*y;
5020 int bn_idx= b_idx - (dir ? 8:1);
5022 if( h->non_zero_count_cache[b_idx] != 0 ||
5023 h->non_zero_count_cache[bn_idx] != 0 ) {
5026 else if( h->slice_type == P_TYPE ) {
5027 if( h->ref_cache[0][b_idx] != h->ref_cache[0][bn_idx] ||
5028 ABS( h->mv_cache[0][b_idx][0] - h->mv_cache[0][bn_idx][0] ) >= 4 ||
5029 ABS( h->mv_cache[0][b_idx][1] - h->mv_cache[0][bn_idx][1] ) >= 4 )
5035 /* FIXME Add support for B frame */
5040 if(bS[0]+bS[1]+bS[2]+bS[3] == 0)
5045 qp = ( s->qscale + s->current_picture.qscale_table[mbn_xy] + 1 ) >> 1;
5047 filter_mb_edgev( h, &img_y[4*edge], linesize, bS, qp );
5048 if( (edge&1) == 0 ) {
5049 int chroma_qp = ( h->chroma_qp +
5050 get_chroma_qp( h, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1;
5051 filter_mb_edgecv( h, &img_cb[2*edge], uvlinesize, bS, chroma_qp );
5052 filter_mb_edgecv( h, &img_cr[2*edge], uvlinesize, bS, chroma_qp );
5055 filter_mb_edgeh( h, &img_y[4*edge*linesize], linesize, bS, qp );
5056 if( (edge&1) == 0 ) {
5057 int chroma_qp = ( h->chroma_qp +
5058 get_chroma_qp( h, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1;
5059 filter_mb_edgech( h, &img_cb[2*edge*uvlinesize], uvlinesize, bS, chroma_qp );
5060 filter_mb_edgech( h, &img_cr[2*edge*uvlinesize], uvlinesize, bS, chroma_qp );
5067 static int decode_slice(H264Context *h){
5068 MpegEncContext * const s = &h->s;
5069 const int part_mask= s->partitioned_frame ? (AC_END|AC_ERROR) : 0x7F;
5073 if( h->pps.cabac ) {
5077 align_get_bits( &s->gb );
5080 ff_init_cabac_states( &h->cabac, ff_h264_lps_range, ff_h264_mps_state, ff_h264_lps_state, 64 );
5081 ff_init_cabac_decoder( &h->cabac,
5082 s->gb.buffer + get_bits_count(&s->gb)/8,
5083 ( s->gb.size_in_bits - get_bits_count(&s->gb) + 7)/8);
5084 /* calculate pre-state */
5085 for( i= 0; i < 399; i++ ) {
5087 if( h->slice_type == I_TYPE )
5088 pre = clip( ((cabac_context_init_I[i][0] * s->qscale) >>4 ) + cabac_context_init_I[i][1], 1, 126 );
5090 pre = clip( ((cabac_context_init_PB[h->cabac_init_idc][i][0] * s->qscale) >>4 ) + cabac_context_init_PB[h->cabac_init_idc][i][1], 1, 126 );
5093 h->cabac_state[i] = 2 * ( 63 - pre ) + 0;
5095 h->cabac_state[i] = 2 * ( pre - 64 ) + 1;
5099 int ret = decode_mb_cabac(h);
5100 int eos = get_cabac_terminate( &h->cabac ); /* End of Slice flag */
5102 if(ret>=0) hl_decode_mb(h);
5104 /* XXX: useless as decode_mb_cabac it doesn't support that ... */
5105 if( ret >= 0 && h->sps.mb_aff ) { //FIXME optimal? or let mb_decode decode 16x32 ?
5108 if(ret>=0) ret = decode_mb_cabac(h);
5109 eos = get_cabac_terminate( &h->cabac );
5115 if( ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 1) {
5116 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
5117 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);
5121 if( ++s->mb_x >= s->mb_width ) {
5123 ff_draw_horiz_band(s, 16*s->mb_y, 16);
5124 if( ++s->mb_y >= s->mb_height ) {
5125 tprintf("slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
5129 if( eos || s->mb_y >= s->mb_height ) {
5130 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);
5134 /* TODO test over-reading in cabac code */
5135 else if( read too much in h->cabac ) {
5136 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);
5144 int ret = decode_mb_cavlc(h);
5146 if(ret>=0) hl_decode_mb(h);
5148 if(ret>=0 && h->sps.mb_aff){ //FIXME optimal? or let mb_decode decode 16x32 ?
5150 ret = decode_mb_cavlc(h);
5152 if(ret>=0) hl_decode_mb(h);
5157 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
5158 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);
5163 if(++s->mb_x >= s->mb_width){
5165 ff_draw_horiz_band(s, 16*s->mb_y, 16);
5166 if(++s->mb_y >= s->mb_height){
5167 tprintf("slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
5169 if(get_bits_count(&s->gb) == s->gb.size_in_bits ) {
5170 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);
5174 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);
5181 if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->mb_skip_run<=0){
5182 if(get_bits_count(&s->gb) == s->gb.size_in_bits ){
5183 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);
5187 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);
5196 for(;s->mb_y < s->mb_height; s->mb_y++){
5197 for(;s->mb_x < s->mb_width; s->mb_x++){
5198 int ret= decode_mb(h);
5203 fprintf(stderr, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
5204 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);
5209 if(++s->mb_x >= s->mb_width){
5211 if(++s->mb_y >= s->mb_height){
5212 if(get_bits_count(s->gb) == s->gb.size_in_bits){
5213 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);
5217 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);
5224 if(get_bits_count(s->?gb) >= s->gb?.size_in_bits){
5225 if(get_bits_count(s->gb) == s->gb.size_in_bits){
5226 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);
5230 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);
5237 ff_draw_horiz_band(s, 16*s->mb_y, 16);
5240 return -1; //not reached
5243 static inline int decode_vui_parameters(H264Context *h, SPS *sps){
5244 MpegEncContext * const s = &h->s;
5245 int aspect_ratio_info_present_flag, aspect_ratio_idc;
5247 aspect_ratio_info_present_flag= get_bits1(&s->gb);
5249 if( aspect_ratio_info_present_flag ) {
5250 aspect_ratio_idc= get_bits(&s->gb, 8);
5251 if( aspect_ratio_idc == EXTENDED_SAR ) {
5252 sps->sar.num= get_bits(&s->gb, 16);
5253 sps->sar.den= get_bits(&s->gb, 16);
5254 }else if(aspect_ratio_idc < 16){
5255 sps->sar= pixel_aspect[aspect_ratio_idc];
5257 av_log(h->s.avctx, AV_LOG_ERROR, "illegal aspect ratio\n");
5264 // s->avctx->aspect_ratio= sar_width*s->width / (float)(s->height*sar_height);
5266 if(get_bits1(&s->gb)){ /* overscan_info_present_flag */
5267 get_bits1(&s->gb); /* overscan_appropriate_flag */
5270 if(get_bits1(&s->gb)){ /* video_signal_type_present_flag */
5271 get_bits(&s->gb, 3); /* video_format */
5272 get_bits1(&s->gb); /* video_full_range_flag */
5273 if(get_bits1(&s->gb)){ /* colour_description_present_flag */
5274 get_bits(&s->gb, 8); /* colour_primaries */
5275 get_bits(&s->gb, 8); /* transfer_characteristics */
5276 get_bits(&s->gb, 8); /* matrix_coefficients */
5280 if(get_bits1(&s->gb)){ /* chroma_location_info_present_flag */
5281 get_ue_golomb(&s->gb); /* chroma_sample_location_type_top_field */
5282 get_ue_golomb(&s->gb); /* chroma_sample_location_type_bottom_field */
5285 sps->timing_info_present_flag = get_bits1(&s->gb);
5286 if(sps->timing_info_present_flag){
5287 sps->num_units_in_tick = get_bits_long(&s->gb, 32);
5288 sps->time_scale = get_bits_long(&s->gb, 32);
5289 sps->fixed_frame_rate_flag = get_bits1(&s->gb);
5293 | nal_hrd_parameters_present_flag |0 |u(1) |
5294 | if( nal_hrd_parameters_present_flag = = 1) | | |
5295 | hrd_parameters( ) | | |
5296 | vcl_hrd_parameters_present_flag |0 |u(1) |
5297 | if( vcl_hrd_parameters_present_flag = = 1) | | |
5298 | hrd_parameters( ) | | |
5299 | if( ( nal_hrd_parameters_present_flag = = 1 | || | |
5301 |( vcl_hrd_parameters_present_flag = = 1 ) ) | | |
5302 | low_delay_hrd_flag |0 |u(1) |
5303 | bitstream_restriction_flag |0 |u(1) |
5304 | if( bitstream_restriction_flag ) { |0 |u(1) |
5305 | motion_vectors_over_pic_boundaries_flag |0 |u(1) |
5306 | max_bytes_per_pic_denom |0 |ue(v) |
5307 | max_bits_per_mb_denom |0 |ue(v) |
5308 | log2_max_mv_length_horizontal |0 |ue(v) |
5309 | log2_max_mv_length_vertical |0 |ue(v) |
5310 | num_reorder_frames |0 |ue(v) |
5311 | max_dec_frame_buffering |0 |ue(v) |
5318 static inline int decode_seq_parameter_set(H264Context *h){
5319 MpegEncContext * const s = &h->s;
5320 int profile_idc, level_idc;
5324 profile_idc= get_bits(&s->gb, 8);
5325 get_bits1(&s->gb); //constraint_set0_flag
5326 get_bits1(&s->gb); //constraint_set1_flag
5327 get_bits1(&s->gb); //constraint_set2_flag
5328 get_bits1(&s->gb); //constraint_set3_flag
5329 get_bits(&s->gb, 4); // reserved
5330 level_idc= get_bits(&s->gb, 8);
5331 sps_id= get_ue_golomb(&s->gb);
5333 sps= &h->sps_buffer[ sps_id ];
5334 sps->profile_idc= profile_idc;
5335 sps->level_idc= level_idc;
5337 sps->log2_max_frame_num= get_ue_golomb(&s->gb) + 4;
5338 sps->poc_type= get_ue_golomb(&s->gb);
5340 if(sps->poc_type == 0){ //FIXME #define
5341 sps->log2_max_poc_lsb= get_ue_golomb(&s->gb) + 4;
5342 } else if(sps->poc_type == 1){//FIXME #define
5343 sps->delta_pic_order_always_zero_flag= get_bits1(&s->gb);
5344 sps->offset_for_non_ref_pic= get_se_golomb(&s->gb);
5345 sps->offset_for_top_to_bottom_field= get_se_golomb(&s->gb);
5346 sps->poc_cycle_length= get_ue_golomb(&s->gb);
5348 for(i=0; i<sps->poc_cycle_length; i++)
5349 sps->offset_for_ref_frame[i]= get_se_golomb(&s->gb);
5351 if(sps->poc_type > 2){
5352 av_log(h->s.avctx, AV_LOG_ERROR, "illegal POC type %d\n", sps->poc_type);
5356 sps->ref_frame_count= get_ue_golomb(&s->gb);
5357 if(sps->ref_frame_count > MAX_PICTURE_COUNT-2){
5358 av_log(h->s.avctx, AV_LOG_ERROR, "too many reference frames\n");
5360 sps->gaps_in_frame_num_allowed_flag= get_bits1(&s->gb);
5361 sps->mb_width= get_ue_golomb(&s->gb) + 1;
5362 sps->mb_height= get_ue_golomb(&s->gb) + 1;
5363 sps->frame_mbs_only_flag= get_bits1(&s->gb);
5364 if(!sps->frame_mbs_only_flag)
5365 sps->mb_aff= get_bits1(&s->gb);
5369 sps->direct_8x8_inference_flag= get_bits1(&s->gb);
5371 sps->crop= get_bits1(&s->gb);
5373 sps->crop_left = get_ue_golomb(&s->gb);
5374 sps->crop_right = get_ue_golomb(&s->gb);
5375 sps->crop_top = get_ue_golomb(&s->gb);
5376 sps->crop_bottom= get_ue_golomb(&s->gb);
5377 if(sps->crop_left || sps->crop_top){
5378 av_log(h->s.avctx, AV_LOG_ERROR, "insane cropping not completly supported, this could look slightly wrong ...\n");
5384 sps->crop_bottom= 0;
5387 sps->vui_parameters_present_flag= get_bits1(&s->gb);
5388 if( sps->vui_parameters_present_flag )
5389 decode_vui_parameters(h, sps);
5391 if(s->avctx->debug&FF_DEBUG_PICT_INFO){
5392 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",
5393 sps_id, sps->profile_idc, sps->level_idc,
5395 sps->ref_frame_count,
5396 sps->mb_width, sps->mb_height,
5397 sps->frame_mbs_only_flag ? "FRM" : (sps->mb_aff ? "MB-AFF" : "PIC-AFF"),
5398 sps->direct_8x8_inference_flag ? "8B8" : "",
5399 sps->crop_left, sps->crop_right,
5400 sps->crop_top, sps->crop_bottom,
5401 sps->vui_parameters_present_flag ? "VUI" : ""
5407 static inline int decode_picture_parameter_set(H264Context *h){
5408 MpegEncContext * const s = &h->s;
5409 int pps_id= get_ue_golomb(&s->gb);
5410 PPS *pps= &h->pps_buffer[pps_id];
5412 pps->sps_id= get_ue_golomb(&s->gb);
5413 pps->cabac= get_bits1(&s->gb);
5414 pps->pic_order_present= get_bits1(&s->gb);
5415 pps->slice_group_count= get_ue_golomb(&s->gb) + 1;
5416 if(pps->slice_group_count > 1 ){
5417 pps->mb_slice_group_map_type= get_ue_golomb(&s->gb);
5418 av_log(h->s.avctx, AV_LOG_ERROR, "FMO not supported\n");
5419 switch(pps->mb_slice_group_map_type){
5422 | for( i = 0; i <= num_slice_groups_minus1; i++ ) | | |
5423 | run_length[ i ] |1 |ue(v) |
5428 | for( i = 0; i < num_slice_groups_minus1; i++ ) | | |
5430 | top_left_mb[ i ] |1 |ue(v) |
5431 | bottom_right_mb[ i ] |1 |ue(v) |
5439 | slice_group_change_direction_flag |1 |u(1) |
5440 | slice_group_change_rate_minus1 |1 |ue(v) |
5445 | slice_group_id_cnt_minus1 |1 |ue(v) |
5446 | for( i = 0; i <= slice_group_id_cnt_minus1; i++ | | |
5448 | slice_group_id[ i ] |1 |u(v) |
5453 pps->ref_count[0]= get_ue_golomb(&s->gb) + 1;
5454 pps->ref_count[1]= get_ue_golomb(&s->gb) + 1;
5455 if(pps->ref_count[0] > 32 || pps->ref_count[1] > 32){
5456 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow (pps)\n");
5460 pps->weighted_pred= get_bits1(&s->gb);
5461 pps->weighted_bipred_idc= get_bits(&s->gb, 2);
5462 pps->init_qp= get_se_golomb(&s->gb) + 26;
5463 pps->init_qs= get_se_golomb(&s->gb) + 26;
5464 pps->chroma_qp_index_offset= get_se_golomb(&s->gb);
5465 pps->deblocking_filter_parameters_present= get_bits1(&s->gb);
5466 pps->constrained_intra_pred= get_bits1(&s->gb);
5467 pps->redundant_pic_cnt_present = get_bits1(&s->gb);
5469 if(s->avctx->debug&FF_DEBUG_PICT_INFO){
5470 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",
5471 pps_id, pps->sps_id,
5472 pps->cabac ? "CABAC" : "CAVLC",
5473 pps->slice_group_count,
5474 pps->ref_count[0], pps->ref_count[1],
5475 pps->weighted_pred ? "weighted" : "",
5476 pps->init_qp, pps->init_qs, pps->chroma_qp_index_offset,
5477 pps->deblocking_filter_parameters_present ? "LPAR" : "",
5478 pps->constrained_intra_pred ? "CONSTR" : "",
5479 pps->redundant_pic_cnt_present ? "REDU" : ""
5487 * finds the end of the current frame in the bitstream.
5488 * @return the position of the first byte of the next frame, or -1
5490 static int find_frame_end(ParseContext *pc, const uint8_t *buf, int buf_size){
5493 //printf("first %02X%02X%02X%02X\n", buf[0], buf[1],buf[2],buf[3]);
5494 // mb_addr= pc->mb_addr - 1;
5496 //FIXME this will fail with slices
5497 for(i=0; i<buf_size; i++){
5498 state= (state<<8) | buf[i];
5499 if((state&0xFFFFFF1F) == 0x101 || (state&0xFFFFFF1F) == 0x102 || (state&0xFFFFFF1F) == 0x105){
5500 if(pc->frame_start_found){
5502 pc->frame_start_found= 0;
5505 pc->frame_start_found= 1;
5510 return END_NOT_FOUND;
5513 static int h264_parse(AVCodecParserContext *s,
5514 AVCodecContext *avctx,
5515 uint8_t **poutbuf, int *poutbuf_size,
5516 const uint8_t *buf, int buf_size)
5518 ParseContext *pc = s->priv_data;
5521 next= find_frame_end(pc, buf, buf_size);
5523 if (ff_combine_frame(pc, next, (uint8_t **)&buf, &buf_size) < 0) {
5529 *poutbuf = (uint8_t *)buf;
5530 *poutbuf_size = buf_size;
5534 static int decode_nal_units(H264Context *h, uint8_t *buf, int buf_size){
5535 MpegEncContext * const s = &h->s;
5536 AVCodecContext * const avctx= s->avctx;
5540 for(i=0; i<32; i++){
5541 printf("%X ", buf[i]);
5552 if(buf_index >= buf_size) break;
5554 for(i = 0; i < h->nal_length_size; i++)
5555 nalsize = (nalsize << 8) | buf[buf_index++];
5557 // start code prefix search
5558 for(; buf_index + 3 < buf_size; buf_index++){
5559 // this should allways succeed in the first iteration
5560 if(buf[buf_index] == 0 && buf[buf_index+1] == 0 && buf[buf_index+2] == 1)
5564 if(buf_index+3 >= buf_size) break;
5569 ptr= decode_nal(h, buf + buf_index, &dst_length, &consumed, buf_size - buf_index);
5570 if(ptr[dst_length - 1] == 0) dst_length--;
5571 bit_length= 8*dst_length - decode_rbsp_trailing(ptr + dst_length - 1);
5573 if(s->avctx->debug&FF_DEBUG_STARTCODE){
5574 av_log(h->s.avctx, AV_LOG_DEBUG, "NAL %d at %d length %d\n", h->nal_unit_type, buf_index, dst_length);
5577 if (h->is_avc && (nalsize != consumed))
5578 av_log(h->s.avctx, AV_LOG_ERROR, "AVC: Consumed only %d bytes instead of %d\n", consumed, nalsize);
5580 buf_index += consumed;
5582 if( s->hurry_up == 1 && h->nal_ref_idc == 0 )
5585 switch(h->nal_unit_type){
5587 idr(h); //FIXME ensure we dont loose some frames if there is reordering
5589 init_get_bits(&s->gb, ptr, bit_length);
5591 h->inter_gb_ptr= &s->gb;
5592 s->data_partitioning = 0;
5594 if(decode_slice_header(h) < 0) return -1;
5595 if(h->redundant_pic_count==0 && s->hurry_up < 5 )
5599 init_get_bits(&s->gb, ptr, bit_length);
5601 h->inter_gb_ptr= NULL;
5602 s->data_partitioning = 1;
5604 if(decode_slice_header(h) < 0) return -1;
5607 init_get_bits(&h->intra_gb, ptr, bit_length);
5608 h->intra_gb_ptr= &h->intra_gb;
5611 init_get_bits(&h->inter_gb, ptr, bit_length);
5612 h->inter_gb_ptr= &h->inter_gb;
5614 if(h->redundant_pic_count==0 && h->intra_gb_ptr && s->data_partitioning && s->hurry_up < 5 )
5620 init_get_bits(&s->gb, ptr, bit_length);
5621 decode_seq_parameter_set(h);
5623 if(s->flags& CODEC_FLAG_LOW_DELAY)
5626 avctx->has_b_frames= !s->low_delay;
5629 init_get_bits(&s->gb, ptr, bit_length);
5631 decode_picture_parameter_set(h);
5634 case NAL_PICTURE_DELIMITER:
5636 case NAL_FILTER_DATA:
5639 av_log(avctx, AV_LOG_ERROR, "Unknown NAL code: %d\n", h->nal_unit_type);
5642 //FIXME move after where irt is set
5643 s->current_picture.pict_type= s->pict_type;
5644 s->current_picture.key_frame= s->pict_type == I_TYPE;
5647 if(!s->current_picture_ptr) return buf_index; //no frame
5649 h->prev_frame_num_offset= h->frame_num_offset;
5650 h->prev_frame_num= h->frame_num;
5651 if(s->current_picture_ptr->reference){
5652 h->prev_poc_msb= h->poc_msb;
5653 h->prev_poc_lsb= h->poc_lsb;
5655 if(s->current_picture_ptr->reference)
5656 execute_ref_pic_marking(h, h->mmco, h->mmco_index);
5658 assert(h->mmco_index==0);
5668 * retunrs the number of bytes consumed for building the current frame
5670 static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size){
5671 if(s->flags&CODEC_FLAG_TRUNCATED){
5672 pos -= s->parse_context.last_index;
5673 if(pos<0) pos=0; // FIXME remove (uneeded?)
5677 if(pos==0) pos=1; //avoid infinite loops (i doubt thats needed but ...)
5678 if(pos+10>buf_size) pos=buf_size; // oops ;)
5684 static int decode_frame(AVCodecContext *avctx,
5685 void *data, int *data_size,
5686 uint8_t *buf, int buf_size)
5688 H264Context *h = avctx->priv_data;
5689 MpegEncContext *s = &h->s;
5690 AVFrame *pict = data;
5693 s->flags= avctx->flags;
5694 s->flags2= avctx->flags2;
5696 /* no supplementary picture */
5697 if (buf_size == 0) {
5701 if(s->flags&CODEC_FLAG_TRUNCATED){
5702 int next= find_frame_end(&s->parse_context, buf, buf_size);
5704 if( ff_combine_frame(&s->parse_context, next, &buf, &buf_size) < 0 )
5706 //printf("next:%d buf_size:%d last_index:%d\n", next, buf_size, s->parse_context.last_index);
5709 if(h->is_avc && !h->got_avcC) {
5710 int i, cnt, nalsize;
5711 unsigned char *p = avctx->extradata;
5712 if(avctx->extradata_size < 7) {
5713 av_log(avctx, AV_LOG_ERROR, "avcC too short\n");
5717 av_log(avctx, AV_LOG_ERROR, "Unknown avcC version %d\n", *p);
5720 /* sps and pps in the avcC always have length coded with 2 bytes,
5721 so put a fake nal_length_size = 2 while parsing them */
5722 h->nal_length_size = 2;
5723 // Decode sps from avcC
5724 cnt = *(p+5) & 0x1f; // Number of sps
5726 for (i = 0; i < cnt; i++) {
5727 nalsize = BE_16(p) + 2;
5728 if(decode_nal_units(h, p, nalsize) != nalsize) {
5729 av_log(avctx, AV_LOG_ERROR, "Decoding sps %d from avcC failed\n", i);
5734 // Decode pps from avcC
5735 cnt = *(p++); // Number of pps
5736 for (i = 0; i < cnt; i++) {
5737 nalsize = BE_16(p) + 2;
5738 if(decode_nal_units(h, p, nalsize) != nalsize) {
5739 av_log(avctx, AV_LOG_ERROR, "Decoding pps %d from avcC failed\n", i);
5744 // Now store right nal length size, that will be use to parse all other nals
5745 h->nal_length_size = ((*(((char*)(avctx->extradata))+4))&0x03)+1;
5746 // Do not reparse avcC
5750 if(!h->is_avc && s->avctx->extradata_size && s->picture_number==0){
5751 if(0 < decode_nal_units(h, s->avctx->extradata, s->avctx->extradata_size) )
5755 buf_index=decode_nal_units(h, buf, buf_size);
5759 //FIXME do something with unavailable reference frames
5761 // if(ret==FRAME_SKIPED) return get_consumed_bytes(s, buf_index, buf_size);
5763 if(s->pict_type==B_TYPE || s->low_delay){
5764 *pict= *(AVFrame*)&s->current_picture;
5766 *pict= *(AVFrame*)&s->last_picture;
5769 if(!s->current_picture_ptr){
5770 av_log(h->s.avctx, AV_LOG_DEBUG, "error, NO frame\n");
5774 *pict= *(AVFrame*)&s->current_picture; //FIXME
5775 ff_print_debug_info(s, pict);
5776 assert(pict->data[0]);
5777 //printf("out %d\n", (int)pict->data[0]);
5780 /* Return the Picture timestamp as the frame number */
5781 /* we substract 1 because it is added on utils.c */
5782 avctx->frame_number = s->picture_number - 1;
5785 /* dont output the last pic after seeking */
5786 if(s->last_picture_ptr || s->low_delay)
5787 //Note this isnt a issue as a IDR pic should flush teh buffers
5789 *data_size = sizeof(AVFrame);
5790 return get_consumed_bytes(s, buf_index, buf_size);
5793 static inline void fill_mb_avail(H264Context *h){
5794 MpegEncContext * const s = &h->s;
5795 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
5798 h->mb_avail[0]= s->mb_x && h->slice_table[mb_xy - s->mb_stride - 1] == h->slice_num;
5799 h->mb_avail[1]= h->slice_table[mb_xy - s->mb_stride ] == h->slice_num;
5800 h->mb_avail[2]= s->mb_x+1 < s->mb_width && h->slice_table[mb_xy - s->mb_stride + 1] == h->slice_num;
5806 h->mb_avail[3]= s->mb_x && h->slice_table[mb_xy - 1] == h->slice_num;
5807 h->mb_avail[4]= 1; //FIXME move out
5808 h->mb_avail[5]= 0; //FIXME move out
5814 #define SIZE (COUNT*40)
5820 // int int_temp[10000];
5822 AVCodecContext avctx;
5824 dsputil_init(&dsp, &avctx);
5826 init_put_bits(&pb, temp, SIZE);
5827 printf("testing unsigned exp golomb\n");
5828 for(i=0; i<COUNT; i++){
5830 set_ue_golomb(&pb, i);
5831 STOP_TIMER("set_ue_golomb");
5833 flush_put_bits(&pb);
5835 init_get_bits(&gb, temp, 8*SIZE);
5836 for(i=0; i<COUNT; i++){
5839 s= show_bits(&gb, 24);
5842 j= get_ue_golomb(&gb);
5844 printf("missmatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
5847 STOP_TIMER("get_ue_golomb");
5851 init_put_bits(&pb, temp, SIZE);
5852 printf("testing signed exp golomb\n");
5853 for(i=0; i<COUNT; i++){
5855 set_se_golomb(&pb, i - COUNT/2);
5856 STOP_TIMER("set_se_golomb");
5858 flush_put_bits(&pb);
5860 init_get_bits(&gb, temp, 8*SIZE);
5861 for(i=0; i<COUNT; i++){
5864 s= show_bits(&gb, 24);
5867 j= get_se_golomb(&gb);
5868 if(j != i - COUNT/2){
5869 printf("missmatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
5872 STOP_TIMER("get_se_golomb");
5875 printf("testing 4x4 (I)DCT\n");
5878 uint8_t src[16], ref[16];
5879 uint64_t error= 0, max_error=0;
5881 for(i=0; i<COUNT; i++){
5883 // printf("%d %d %d\n", r1, r2, (r2-r1)*16);
5884 for(j=0; j<16; j++){
5885 ref[j]= random()%255;
5886 src[j]= random()%255;
5889 h264_diff_dct_c(block, src, ref, 4);
5892 for(j=0; j<16; j++){
5893 // printf("%d ", block[j]);
5894 block[j]= block[j]*4;
5895 if(j&1) block[j]= (block[j]*4 + 2)/5;
5896 if(j&4) block[j]= (block[j]*4 + 2)/5;
5900 s->dsp.h264_idct_add(ref, block, 4);
5901 /* for(j=0; j<16; j++){
5902 printf("%d ", ref[j]);
5906 for(j=0; j<16; j++){
5907 int diff= ABS(src[j] - ref[j]);
5910 max_error= FFMAX(max_error, diff);
5913 printf("error=%f max_error=%d\n", ((float)error)/COUNT/16, (int)max_error );
5915 printf("testing quantizer\n");
5916 for(qp=0; qp<52; qp++){
5918 src1_block[i]= src2_block[i]= random()%255;
5922 printf("Testing NAL layer\n");
5924 uint8_t bitstream[COUNT];
5925 uint8_t nal[COUNT*2];
5927 memset(&h, 0, sizeof(H264Context));
5929 for(i=0; i<COUNT; i++){
5937 for(j=0; j<COUNT; j++){
5938 bitstream[j]= (random() % 255) + 1;
5941 for(j=0; j<zeros; j++){
5942 int pos= random() % COUNT;
5943 while(bitstream[pos] == 0){
5952 nal_length= encode_nal(&h, nal, bitstream, COUNT, COUNT*2);
5954 printf("encoding failed\n");
5958 out= decode_nal(&h, nal, &out_length, &consumed, nal_length);
5962 if(out_length != COUNT){
5963 printf("incorrect length %d %d\n", out_length, COUNT);
5967 if(consumed != nal_length){
5968 printf("incorrect consumed length %d %d\n", nal_length, consumed);
5972 if(memcmp(bitstream, out, COUNT)){
5973 printf("missmatch\n");
5978 printf("Testing RBSP\n");
5986 static int decode_end(AVCodecContext *avctx)
5988 H264Context *h = avctx->priv_data;
5989 MpegEncContext *s = &h->s;
5991 free_tables(h); //FIXME cleanup init stuff perhaps
5994 // memset(h, 0, sizeof(H264Context));
6000 AVCodec h264_decoder = {
6004 sizeof(H264Context),
6009 /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 | CODEC_CAP_TRUNCATED,
6012 AVCodecParser h264_parser = {
6014 sizeof(ParseContext),