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?
90 int bitstream_restriction_flag;
91 int num_reorder_frames;
95 * Picture parameter set
99 int cabac; ///< entropy_coding_mode_flag
100 int pic_order_present; ///< pic_order_present_flag
101 int slice_group_count; ///< num_slice_groups_minus1 + 1
102 int mb_slice_group_map_type;
103 int ref_count[2]; ///< num_ref_idx_l0/1_active_minus1 + 1
104 int weighted_pred; ///< weighted_pred_flag
105 int weighted_bipred_idc;
106 int init_qp; ///< pic_init_qp_minus26 + 26
107 int init_qs; ///< pic_init_qs_minus26 + 26
108 int chroma_qp_index_offset;
109 int deblocking_filter_parameters_present; ///< deblocking_filter_parameters_present_flag
110 int constrained_intra_pred; ///< constrained_intra_pred_flag
111 int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag
115 * Memory management control operation opcode.
117 typedef enum MMCOOpcode{
128 * Memory management control operation.
139 typedef struct H264Context{
147 #define NAL_IDR_SLICE 5
151 #define NAL_PICTURE_DELIMITER 9
152 #define NAL_FILTER_DATA 10
153 uint8_t *rbsp_buffer;
154 int rbsp_buffer_size;
157 * Used to parse AVC variant of h264
159 int is_avc; ///< this flag is != 0 if codec is avc1
160 int got_avcC; ///< flag used to parse avcC data only once
161 int nal_length_size; ///< Number of bytes used for nal length (1, 2 or 4)
165 int prev_mb_skiped; //FIXME remove (IMHO not used)
168 int chroma_pred_mode;
169 int intra16x16_pred_mode;
174 int8_t intra4x4_pred_mode_cache[5*8];
175 int8_t (*intra4x4_pred_mode)[8];
176 void (*pred4x4 [9+3])(uint8_t *src, uint8_t *topright, int stride);//FIXME move to dsp?
177 void (*pred8x8 [4+3])(uint8_t *src, int stride);
178 void (*pred16x16[4+3])(uint8_t *src, int stride);
179 unsigned int topleft_samples_available;
180 unsigned int top_samples_available;
181 unsigned int topright_samples_available;
182 unsigned int left_samples_available;
183 uint8_t (*top_borders[2])[16+2*8];
184 uint8_t left_border[2*(17+2*9)];
187 * non zero coeff count cache.
188 * is 64 if not available.
190 uint8_t non_zero_count_cache[6*8] __align8;
191 uint8_t (*non_zero_count)[16];
194 * Motion vector cache.
196 int16_t mv_cache[2][5*8][2] __align8;
197 int8_t ref_cache[2][5*8] __align8;
198 #define LIST_NOT_USED -1 //FIXME rename?
199 #define PART_NOT_AVAILABLE -2
202 * is 1 if the specific list MV&references are set to 0,0,-2.
204 int mv_cache_clean[2];
207 * block_offset[ 0..23] for frame macroblocks
208 * block_offset[24..47] for field macroblocks
210 int block_offset[2*(16+8)];
212 uint16_t *mb2b_xy; //FIXME are these 4 a good idea?
214 int b_stride; //FIXME use s->b4_stride
220 int unknown_svq3_flag;
221 int next_slice_index;
223 SPS sps_buffer[MAX_SPS_COUNT];
224 SPS sps; ///< current sps
226 PPS pps_buffer[MAX_PPS_COUNT];
230 PPS pps; //FIXME move tp Picture perhaps? (->no) do we need that?
233 uint8_t *slice_table_base;
234 uint8_t *slice_table; ///< slice_table_base + mb_stride + 1
236 int slice_type_fixed;
238 //interlacing specific flags
240 int mb_field_decoding_flag;
247 int delta_poc_bottom;
250 int prev_poc_msb; ///< poc_msb of the last reference pic for POC type 0
251 int prev_poc_lsb; ///< poc_lsb of the last reference pic for POC type 0
252 int frame_num_offset; ///< for POC type 2
253 int prev_frame_num_offset; ///< for POC type 2
254 int prev_frame_num; ///< frame_num of the last pic for POC type 1/2
257 * frame_num for frames or 2*frame_num for field pics.
262 * max_frame_num or 2*max_frame_num for field pics.
266 //Weighted pred stuff
268 int use_weight_chroma;
269 int luma_log2_weight_denom;
270 int chroma_log2_weight_denom;
271 int luma_weight[2][16];
272 int luma_offset[2][16];
273 int chroma_weight[2][16][2];
274 int chroma_offset[2][16][2];
275 int implicit_weight[16][16];
278 int deblocking_filter; ///< disable_deblocking_filter_idc with 1<->0
279 int slice_alpha_c0_offset;
280 int slice_beta_offset;
282 int redundant_pic_count;
284 int direct_spatial_mv_pred;
285 int dist_scale_factor[16];
286 int map_col_to_list0[2][16];
289 * num_ref_idx_l0/1_active_minus1 + 1
291 int ref_count[2];// FIXME split for AFF
292 Picture *short_ref[32];
293 Picture *long_ref[32];
294 Picture default_ref_list[2][32];
295 Picture ref_list[2][32]; //FIXME size?
296 Picture field_ref_list[2][32]; //FIXME size?
297 Picture *delayed_pic[16]; //FIXME size?
298 Picture *delayed_output_pic;
301 * memory management control operations buffer.
303 MMCO mmco[MAX_MMCO_COUNT];
306 int long_ref_count; ///< number of actual long term references
307 int short_ref_count; ///< number of actual short term references
310 GetBitContext intra_gb;
311 GetBitContext inter_gb;
312 GetBitContext *intra_gb_ptr;
313 GetBitContext *inter_gb_ptr;
315 DCTELEM mb[16*24] __align8;
321 uint8_t cabac_state[399];
324 /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0,1,2), 0x0? luma_cbp */
328 /* chroma_pred_mode for i4x4 or i16x16, else 0 */
329 uint8_t *chroma_pred_mode_table;
330 int last_qscale_diff;
331 int16_t (*mvd_table[2])[2];
332 int16_t mvd_cache[2][5*8][2] __align8;
333 uint8_t *direct_table;
334 uint8_t direct_cache[5*8];
338 static VLC coeff_token_vlc[4];
339 static VLC chroma_dc_coeff_token_vlc;
341 static VLC total_zeros_vlc[15];
342 static VLC chroma_dc_total_zeros_vlc[3];
344 static VLC run_vlc[6];
347 static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
348 static void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
349 static void filter_mb( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr, unsigned int linesize, unsigned int uvlinesize);
351 static inline uint32_t pack16to32(int a, int b){
352 #ifdef WORDS_BIGENDIAN
353 return (b&0xFFFF) + (a<<16);
355 return (a&0xFFFF) + (b<<16);
361 * @param h height of the rectangle, should be a constant
362 * @param w width of the rectangle, should be a constant
363 * @param size the size of val (1 or 4), should be a constant
365 static inline void fill_rectangle(void *vp, int w, int h, int stride, uint32_t val, int size){ //FIXME ensure this IS inlined
366 uint8_t *p= (uint8_t*)vp;
367 assert(size==1 || size==4);
372 assert((((int)vp)&(FFMIN(w, STRIDE_ALIGN)-1)) == 0);
373 //FIXME check what gcc generates for 64 bit on x86 and possible write a 32 bit ver of it
376 *(uint16_t*)(p + stride)= size==4 ? val : val*0x0101;
377 }else if(w==2 && h==4){
378 *(uint16_t*)(p + 0*stride)=
379 *(uint16_t*)(p + 1*stride)=
380 *(uint16_t*)(p + 2*stride)=
381 *(uint16_t*)(p + 3*stride)= size==4 ? val : val*0x0101;
382 }else if(w==4 && h==1){
383 *(uint32_t*)(p + 0*stride)= size==4 ? val : val*0x01010101;
384 }else if(w==4 && h==2){
385 *(uint32_t*)(p + 0*stride)=
386 *(uint32_t*)(p + 1*stride)= size==4 ? val : val*0x01010101;
387 }else if(w==4 && h==4){
388 *(uint32_t*)(p + 0*stride)=
389 *(uint32_t*)(p + 1*stride)=
390 *(uint32_t*)(p + 2*stride)=
391 *(uint32_t*)(p + 3*stride)= size==4 ? val : val*0x01010101;
392 }else if(w==8 && h==1){
394 *(uint32_t*)(p + 4)= size==4 ? val : val*0x01010101;
395 }else if(w==8 && h==2){
396 *(uint32_t*)(p + 0 + 0*stride)=
397 *(uint32_t*)(p + 4 + 0*stride)=
398 *(uint32_t*)(p + 0 + 1*stride)=
399 *(uint32_t*)(p + 4 + 1*stride)= size==4 ? val : val*0x01010101;
400 }else if(w==8 && h==4){
401 *(uint64_t*)(p + 0*stride)=
402 *(uint64_t*)(p + 1*stride)=
403 *(uint64_t*)(p + 2*stride)=
404 *(uint64_t*)(p + 3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
405 }else if(w==16 && h==2){
406 *(uint64_t*)(p + 0+0*stride)=
407 *(uint64_t*)(p + 8+0*stride)=
408 *(uint64_t*)(p + 0+1*stride)=
409 *(uint64_t*)(p + 8+1*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
410 }else if(w==16 && h==4){
411 *(uint64_t*)(p + 0+0*stride)=
412 *(uint64_t*)(p + 8+0*stride)=
413 *(uint64_t*)(p + 0+1*stride)=
414 *(uint64_t*)(p + 8+1*stride)=
415 *(uint64_t*)(p + 0+2*stride)=
416 *(uint64_t*)(p + 8+2*stride)=
417 *(uint64_t*)(p + 0+3*stride)=
418 *(uint64_t*)(p + 8+3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
423 static inline void fill_caches(H264Context *h, int mb_type, int for_deblock){
424 MpegEncContext * const s = &h->s;
425 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
426 int topleft_xy, top_xy, topright_xy, left_xy[2];
427 int topleft_type, top_type, topright_type, left_type[2];
431 //wow what a mess, why didnt they simplify the interlacing&intra stuff, i cant imagine that these complex rules are worth it
433 top_xy = mb_xy - s->mb_stride;
434 topleft_xy = top_xy - 1;
435 topright_xy= top_xy + 1;
436 left_xy[1] = left_xy[0] = mb_xy-1;
446 const int pair_xy = s->mb_x + (s->mb_y & ~1)*s->mb_stride;
447 const int top_pair_xy = pair_xy - s->mb_stride;
448 const int topleft_pair_xy = top_pair_xy - 1;
449 const int topright_pair_xy = top_pair_xy + 1;
450 const int topleft_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[topleft_pair_xy]);
451 const int top_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[top_pair_xy]);
452 const int topright_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[topright_pair_xy]);
453 const int left_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[pair_xy-1]);
454 const int curr_mb_frame_flag = !IS_INTERLACED(mb_type);
455 const int bottom = (s->mb_y & 1);
456 tprintf("fill_caches: curr_mb_frame_flag:%d, left_mb_frame_flag:%d, topleft_mb_frame_flag:%d, top_mb_frame_flag:%d, topright_mb_frame_flag:%d\n", curr_mb_frame_flag, left_mb_frame_flag, topleft_mb_frame_flag, top_mb_frame_flag, topright_mb_frame_flag);
458 ? !curr_mb_frame_flag // bottom macroblock
459 : (!curr_mb_frame_flag && !top_mb_frame_flag) // top macroblock
461 top_xy -= s->mb_stride;
464 ? !curr_mb_frame_flag // bottom macroblock
465 : (!curr_mb_frame_flag && !topleft_mb_frame_flag) // top macroblock
467 topleft_xy -= s->mb_stride;
470 ? !curr_mb_frame_flag // bottom macroblock
471 : (!curr_mb_frame_flag && !topright_mb_frame_flag) // top macroblock
473 topright_xy -= s->mb_stride;
475 if (left_mb_frame_flag != curr_mb_frame_flag) {
476 left_xy[1] = left_xy[0] = pair_xy - 1;
477 if (curr_mb_frame_flag) {
498 left_xy[1] += s->mb_stride;
512 h->top_mb_xy = top_xy;
513 h->left_mb_xy[0] = left_xy[0];
514 h->left_mb_xy[1] = left_xy[1];
517 topleft_type = h->slice_table[topleft_xy ] < 255 ? s->current_picture.mb_type[topleft_xy] : 0;
518 top_type = h->slice_table[top_xy ] < 255 ? s->current_picture.mb_type[top_xy] : 0;
519 topright_type= h->slice_table[topright_xy] < 255 ? s->current_picture.mb_type[topright_xy]: 0;
520 left_type[0] = h->slice_table[left_xy[0] ] < 255 ? s->current_picture.mb_type[left_xy[0]] : 0;
521 left_type[1] = h->slice_table[left_xy[1] ] < 255 ? s->current_picture.mb_type[left_xy[1]] : 0;
523 topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
524 top_type = h->slice_table[top_xy ] == h->slice_num ? s->current_picture.mb_type[top_xy] : 0;
525 topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
526 left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
527 left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
530 if(IS_INTRA(mb_type)){
531 h->topleft_samples_available=
532 h->top_samples_available=
533 h->left_samples_available= 0xFFFF;
534 h->topright_samples_available= 0xEEEA;
536 if(!IS_INTRA(top_type) && (top_type==0 || h->pps.constrained_intra_pred)){
537 h->topleft_samples_available= 0xB3FF;
538 h->top_samples_available= 0x33FF;
539 h->topright_samples_available= 0x26EA;
542 if(!IS_INTRA(left_type[i]) && (left_type[i]==0 || h->pps.constrained_intra_pred)){
543 h->topleft_samples_available&= 0xDF5F;
544 h->left_samples_available&= 0x5F5F;
548 if(!IS_INTRA(topleft_type) && (topleft_type==0 || h->pps.constrained_intra_pred))
549 h->topleft_samples_available&= 0x7FFF;
551 if(!IS_INTRA(topright_type) && (topright_type==0 || h->pps.constrained_intra_pred))
552 h->topright_samples_available&= 0xFBFF;
554 if(IS_INTRA4x4(mb_type)){
555 if(IS_INTRA4x4(top_type)){
556 h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
557 h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
558 h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
559 h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
562 if(!top_type || (IS_INTER(top_type) && h->pps.constrained_intra_pred))
567 h->intra4x4_pred_mode_cache[4+8*0]=
568 h->intra4x4_pred_mode_cache[5+8*0]=
569 h->intra4x4_pred_mode_cache[6+8*0]=
570 h->intra4x4_pred_mode_cache[7+8*0]= pred;
573 if(IS_INTRA4x4(left_type[i])){
574 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
575 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
578 if(!left_type[i] || (IS_INTER(left_type[i]) && h->pps.constrained_intra_pred))
583 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
584 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
599 //FIXME constraint_intra_pred & partitioning & nnz (lets hope this is just a typo in the spec)
601 h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][4];
602 h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][5];
603 h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][6];
604 h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3];
606 h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][9];
607 h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8];
609 h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][12];
610 h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11];
612 h->top_cbp= h->cbp_table[top_xy];
614 h->non_zero_count_cache[4+8*0]=
615 h->non_zero_count_cache[5+8*0]=
616 h->non_zero_count_cache[6+8*0]=
617 h->non_zero_count_cache[7+8*0]=
619 h->non_zero_count_cache[1+8*0]=
620 h->non_zero_count_cache[2+8*0]=
622 h->non_zero_count_cache[1+8*3]=
623 h->non_zero_count_cache[2+8*3]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
625 if(IS_INTRA(mb_type)) h->top_cbp= 0x1C0;
629 for (i=0; i<2; i++) {
631 h->non_zero_count_cache[3+8*1 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[0+2*i]];
632 h->non_zero_count_cache[3+8*2 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[1+2*i]];
633 h->non_zero_count_cache[0+8*1 + 8*i]= h->non_zero_count[left_xy[i]][left_block[4+2*i]];
634 h->non_zero_count_cache[0+8*4 + 8*i]= h->non_zero_count[left_xy[i]][left_block[5+2*i]];
635 h->left_cbp= h->cbp_table[left_xy[i]]; //FIXME interlacing
637 h->non_zero_count_cache[3+8*1 + 2*8*i]=
638 h->non_zero_count_cache[3+8*2 + 2*8*i]=
639 h->non_zero_count_cache[0+8*1 + 8*i]=
640 h->non_zero_count_cache[0+8*4 + 8*i]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
642 if(IS_INTRA(mb_type)) h->left_cbp= 0x1C0;//FIXME interlacing
648 //FIXME direct mb can skip much of this
649 if(IS_INTER(mb_type) || (IS_DIRECT(mb_type) && h->direct_spatial_mv_pred)){
651 for(list=0; list<2; list++){
652 if(!USES_LIST(mb_type, list) && !IS_DIRECT(mb_type) && !for_deblock){
653 /*if(!h->mv_cache_clean[list]){
654 memset(h->mv_cache [list], 0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
655 memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
656 h->mv_cache_clean[list]= 1;
660 h->mv_cache_clean[list]= 0;
662 if(IS_INTER(topleft_type)){
663 const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
664 const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + h->b8_stride;
665 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
666 h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
668 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
669 h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
672 if(IS_INTER(top_type)){
673 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
674 const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
675 *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0];
676 *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];
677 *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];
678 *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];
679 h->ref_cache[list][scan8[0] + 0 - 1*8]=
680 h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
681 h->ref_cache[list][scan8[0] + 2 - 1*8]=
682 h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
684 *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]=
685 *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]=
686 *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]=
687 *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;
688 *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
691 if(IS_INTER(topright_type)){
692 const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
693 const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
694 *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
695 h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
697 *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
698 h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
701 //FIXME unify cleanup or sth
702 if(IS_INTER(left_type[0])){
703 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
704 const int b8_xy= h->mb2b8_xy[left_xy[0]] + 1;
705 *(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]];
706 *(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]];
707 h->ref_cache[list][scan8[0] - 1 + 0*8]=
708 h->ref_cache[list][scan8[0] - 1 + 1*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0]>>1)];
710 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 0*8]=
711 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 1*8]= 0;
712 h->ref_cache[list][scan8[0] - 1 + 0*8]=
713 h->ref_cache[list][scan8[0] - 1 + 1*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
716 if(IS_INTER(left_type[1])){
717 const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
718 const int b8_xy= h->mb2b8_xy[left_xy[1]] + 1;
719 *(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]];
720 *(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]];
721 h->ref_cache[list][scan8[0] - 1 + 2*8]=
722 h->ref_cache[list][scan8[0] - 1 + 3*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[2]>>1)];
724 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 2*8]=
725 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 3*8]= 0;
726 h->ref_cache[list][scan8[0] - 1 + 2*8]=
727 h->ref_cache[list][scan8[0] - 1 + 3*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
733 h->ref_cache[list][scan8[5 ]+1] =
734 h->ref_cache[list][scan8[7 ]+1] =
735 h->ref_cache[list][scan8[13]+1] = //FIXME remove past 3 (init somewher else)
736 h->ref_cache[list][scan8[4 ]] =
737 h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
738 *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
739 *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
740 *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewher else)
741 *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
742 *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
745 /* XXX beurk, Load mvd */
746 if(IS_INTER(topleft_type)){
747 const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
748 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy];
750 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 - 1*8]= 0;
753 if(IS_INTER(top_type)){
754 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
755 *(uint32_t*)h->mvd_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 0];
756 *(uint32_t*)h->mvd_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 1];
757 *(uint32_t*)h->mvd_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 2];
758 *(uint32_t*)h->mvd_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 3];
760 *(uint32_t*)h->mvd_cache [list][scan8[0] + 0 - 1*8]=
761 *(uint32_t*)h->mvd_cache [list][scan8[0] + 1 - 1*8]=
762 *(uint32_t*)h->mvd_cache [list][scan8[0] + 2 - 1*8]=
763 *(uint32_t*)h->mvd_cache [list][scan8[0] + 3 - 1*8]= 0;
765 if(IS_INTER(left_type[0])){
766 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
767 *(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]];
768 *(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]];
770 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 0*8]=
771 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 1*8]= 0;
773 if(IS_INTER(left_type[1])){
774 const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
775 *(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]];
776 *(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]];
778 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 2*8]=
779 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 3*8]= 0;
781 *(uint32_t*)h->mvd_cache [list][scan8[5 ]+1]=
782 *(uint32_t*)h->mvd_cache [list][scan8[7 ]+1]=
783 *(uint32_t*)h->mvd_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewher else)
784 *(uint32_t*)h->mvd_cache [list][scan8[4 ]]=
785 *(uint32_t*)h->mvd_cache [list][scan8[12]]= 0;
787 if(h->slice_type == B_TYPE){
788 fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, 0, 1);
790 if(IS_DIRECT(top_type)){
791 *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0x01010101;
792 }else if(IS_8X8(top_type)){
793 int b8_xy = h->mb2b8_xy[top_xy] + h->b8_stride;
794 h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy];
795 h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 1];
797 *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0;
801 if(IS_DIRECT(left_type[0])){
802 h->direct_cache[scan8[0] - 1 + 0*8]=
803 h->direct_cache[scan8[0] - 1 + 2*8]= 1;
804 }else if(IS_8X8(left_type[0])){
805 int b8_xy = h->mb2b8_xy[left_xy[0]] + 1;
806 h->direct_cache[scan8[0] - 1 + 0*8]= h->direct_table[b8_xy];
807 h->direct_cache[scan8[0] - 1 + 2*8]= h->direct_table[b8_xy + h->b8_stride];
809 h->direct_cache[scan8[0] - 1 + 0*8]=
810 h->direct_cache[scan8[0] - 1 + 2*8]= 0;
819 static inline void write_back_intra_pred_mode(H264Context *h){
820 MpegEncContext * const s = &h->s;
821 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
823 h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
824 h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
825 h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
826 h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
827 h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
828 h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
829 h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
833 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
835 static inline int check_intra4x4_pred_mode(H264Context *h){
836 MpegEncContext * const s = &h->s;
837 static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
838 static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
841 if(!(h->top_samples_available&0x8000)){
843 int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
845 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);
848 h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
853 if(!(h->left_samples_available&0x8000)){
855 int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
857 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);
860 h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
866 } //FIXME cleanup like next
869 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
871 static inline int check_intra_pred_mode(H264Context *h, int mode){
872 MpegEncContext * const s = &h->s;
873 static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
874 static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
876 if(mode < 0 || mode > 6) {
877 av_log(h->s.avctx, AV_LOG_ERROR, "out of range intra chroma pred mode at %d %d\n", s->mb_x, s->mb_y);
881 if(!(h->top_samples_available&0x8000)){
884 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);
889 if(!(h->left_samples_available&0x8000)){
892 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);
901 * gets the predicted intra4x4 prediction mode.
903 static inline int pred_intra_mode(H264Context *h, int n){
904 const int index8= scan8[n];
905 const int left= h->intra4x4_pred_mode_cache[index8 - 1];
906 const int top = h->intra4x4_pred_mode_cache[index8 - 8];
907 const int min= FFMIN(left, top);
909 tprintf("mode:%d %d min:%d\n", left ,top, min);
911 if(min<0) return DC_PRED;
915 static inline void write_back_non_zero_count(H264Context *h){
916 MpegEncContext * const s = &h->s;
917 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
919 h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[7+8*1];
920 h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[7+8*2];
921 h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[7+8*3];
922 h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];
923 h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[4+8*4];
924 h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[5+8*4];
925 h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[6+8*4];
927 h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[1+8*2];
928 h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
929 h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[2+8*1];
931 h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[1+8*5];
932 h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
933 h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[2+8*4];
937 * gets the predicted number of non zero coefficients.
938 * @param n block index
940 static inline int pred_non_zero_count(H264Context *h, int n){
941 const int index8= scan8[n];
942 const int left= h->non_zero_count_cache[index8 - 1];
943 const int top = h->non_zero_count_cache[index8 - 8];
946 if(i<64) i= (i+1)>>1;
948 tprintf("pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
953 static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
954 const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];
956 if(topright_ref != PART_NOT_AVAILABLE){
957 *C= h->mv_cache[list][ i - 8 + part_width ];
960 tprintf("topright MV not available\n");
962 *C= h->mv_cache[list][ i - 8 - 1 ];
963 return h->ref_cache[list][ i - 8 - 1 ];
968 * gets the predicted MV.
969 * @param n the block index
970 * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)
971 * @param mx the x component of the predicted motion vector
972 * @param my the y component of the predicted motion vector
974 static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
975 const int index8= scan8[n];
976 const int top_ref= h->ref_cache[list][ index8 - 8 ];
977 const int left_ref= h->ref_cache[list][ index8 - 1 ];
978 const int16_t * const A= h->mv_cache[list][ index8 - 1 ];
979 const int16_t * const B= h->mv_cache[list][ index8 - 8 ];
981 int diagonal_ref, match_count;
983 assert(part_width==1 || part_width==2 || part_width==4);
993 diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);
994 match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);
995 tprintf("pred_motion match_count=%d\n", match_count);
996 if(match_count > 1){ //most common
997 *mx= mid_pred(A[0], B[0], C[0]);
998 *my= mid_pred(A[1], B[1], C[1]);
999 }else if(match_count==1){
1003 }else if(top_ref==ref){
1011 if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
1015 *mx= mid_pred(A[0], B[0], C[0]);
1016 *my= mid_pred(A[1], B[1], C[1]);
1020 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);
1024 * gets the directionally predicted 16x8 MV.
1025 * @param n the block index
1026 * @param mx the x component of the predicted motion vector
1027 * @param my the y component of the predicted motion vector
1029 static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
1031 const int top_ref= h->ref_cache[list][ scan8[0] - 8 ];
1032 const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
1034 tprintf("pred_16x8: (%2d %2d %2d) at %2d %2d %d list %d\n", top_ref, B[0], B[1], h->s.mb_x, h->s.mb_y, n, list);
1042 const int left_ref= h->ref_cache[list][ scan8[8] - 1 ];
1043 const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];
1045 tprintf("pred_16x8: (%2d %2d %2d) at %2d %2d %d list %d\n", left_ref, A[0], A[1], h->s.mb_x, h->s.mb_y, n, list);
1047 if(left_ref == ref){
1055 pred_motion(h, n, 4, list, ref, mx, my);
1059 * gets the directionally predicted 8x16 MV.
1060 * @param n the block index
1061 * @param mx the x component of the predicted motion vector
1062 * @param my the y component of the predicted motion vector
1064 static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
1066 const int left_ref= h->ref_cache[list][ scan8[0] - 1 ];
1067 const int16_t * const A= h->mv_cache[list][ scan8[0] - 1 ];
1069 tprintf("pred_8x16: (%2d %2d %2d) at %2d %2d %d list %d\n", left_ref, A[0], A[1], h->s.mb_x, h->s.mb_y, n, list);
1071 if(left_ref == ref){
1080 diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);
1082 tprintf("pred_8x16: (%2d %2d %2d) at %2d %2d %d list %d\n", diagonal_ref, C[0], C[1], h->s.mb_x, h->s.mb_y, n, list);
1084 if(diagonal_ref == ref){
1092 pred_motion(h, n, 2, list, ref, mx, my);
1095 static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
1096 const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
1097 const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];
1099 tprintf("pred_pskip: (%d) (%d) at %2d %2d\n", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
1101 if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
1102 || (top_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0)
1103 || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){
1109 pred_motion(h, 0, 4, 0, 0, mx, my);
1114 static inline void direct_dist_scale_factor(H264Context * const h){
1115 const int poc = h->s.current_picture_ptr->poc;
1116 const int poc1 = h->ref_list[1][0].poc;
1118 for(i=0; i<h->ref_count[0]; i++){
1119 int poc0 = h->ref_list[0][i].poc;
1120 int td = clip(poc1 - poc0, -128, 127);
1121 if(td == 0 /* FIXME || pic0 is a long-term ref */){
1122 h->dist_scale_factor[i] = 256;
1124 int tb = clip(poc - poc0, -128, 127);
1125 int tx = (16384 + (ABS(td) >> 1)) / td;
1126 h->dist_scale_factor[i] = clip((tb*tx + 32) >> 6, -1024, 1023);
1130 static inline void direct_ref_list_init(H264Context * const h){
1131 MpegEncContext * const s = &h->s;
1132 Picture * const ref1 = &h->ref_list[1][0];
1133 Picture * const cur = s->current_picture_ptr;
1135 if(cur->pict_type == I_TYPE)
1136 cur->ref_count[0] = 0;
1137 if(cur->pict_type != B_TYPE)
1138 cur->ref_count[1] = 0;
1139 for(list=0; list<2; list++){
1140 cur->ref_count[list] = h->ref_count[list];
1141 for(j=0; j<h->ref_count[list]; j++)
1142 cur->ref_poc[list][j] = h->ref_list[list][j].poc;
1144 if(cur->pict_type != B_TYPE || h->direct_spatial_mv_pred)
1146 for(list=0; list<2; list++){
1147 for(i=0; i<ref1->ref_count[list]; i++){
1148 const int poc = ref1->ref_poc[list][i];
1149 h->map_col_to_list0[list][i] = PART_NOT_AVAILABLE;
1150 for(j=0; j<h->ref_count[list]; j++)
1151 if(h->ref_list[list][j].poc == poc){
1152 h->map_col_to_list0[list][i] = j;
1159 static inline void pred_direct_motion(H264Context * const h, int *mb_type){
1160 MpegEncContext * const s = &h->s;
1161 const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
1162 const int b8_xy = 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1163 const int b4_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1164 const int mb_type_col = h->ref_list[1][0].mb_type[mb_xy];
1165 const int16_t (*l1mv0)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[0][b4_xy];
1166 const int8_t *l1ref0 = &h->ref_list[1][0].ref_index[0][b8_xy];
1167 const int8_t *l1ref1 = &h->ref_list[1][0].ref_index[1][b8_xy];
1168 const int is_b8x8 = IS_8X8(*mb_type);
1172 if(IS_8X8(mb_type_col) && !h->sps.direct_8x8_inference_flag){
1173 /* FIXME save sub mb types from previous frames (or derive from MVs)
1174 * so we know exactly what block size to use */
1175 sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */
1176 *mb_type = MB_TYPE_8x8|MB_TYPE_L0L1;
1177 }else if(!is_b8x8 && (IS_16X16(mb_type_col) || IS_INTRA(mb_type_col))){
1178 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1179 *mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */
1181 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1182 *mb_type = MB_TYPE_8x8|MB_TYPE_L0L1;
1185 *mb_type |= MB_TYPE_DIRECT2;
1187 tprintf("mb_type = %08x, sub_mb_type = %08x, is_b8x8 = %d, mb_type_col = %08x\n", *mb_type, sub_mb_type, is_b8x8, mb_type_col);
1189 if(h->direct_spatial_mv_pred){
1194 /* ref = min(neighbors) */
1195 for(list=0; list<2; list++){
1196 int refa = h->ref_cache[list][scan8[0] - 1];
1197 int refb = h->ref_cache[list][scan8[0] - 8];
1198 int refc = h->ref_cache[list][scan8[0] - 8 + 4];
1200 refc = h->ref_cache[list][scan8[0] - 8 - 1];
1202 if(ref[list] < 0 || (refb < ref[list] && refb >= 0))
1204 if(ref[list] < 0 || (refc < ref[list] && refc >= 0))
1210 if(ref[0] < 0 && ref[1] < 0){
1211 ref[0] = ref[1] = 0;
1212 mv[0][0] = mv[0][1] =
1213 mv[1][0] = mv[1][1] = 0;
1215 for(list=0; list<2; list++){
1217 pred_motion(h, 0, 4, list, ref[list], &mv[list][0], &mv[list][1]);
1219 mv[list][0] = mv[list][1] = 0;
1224 *mb_type &= ~MB_TYPE_P0L1;
1225 sub_mb_type &= ~MB_TYPE_P0L1;
1226 }else if(ref[0] < 0){
1227 *mb_type &= ~MB_TYPE_P0L0;
1228 sub_mb_type &= ~MB_TYPE_P0L0;
1231 if(IS_16X16(*mb_type)){
1232 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref[0], 1);
1233 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, ref[1], 1);
1234 if(!IS_INTRA(mb_type_col) && l1ref0[0] == 0 &&
1235 ABS(l1mv0[0][0]) <= 1 && ABS(l1mv0[0][1]) <= 1){
1237 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1239 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
1241 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1243 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
1245 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1246 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1249 for(i8=0; i8<4; i8++){
1250 const int x8 = i8&1;
1251 const int y8 = i8>>1;
1253 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1255 h->sub_mb_type[i8] = sub_mb_type;
1257 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1258 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1259 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref[0], 1);
1260 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, ref[1], 1);
1263 if(!IS_INTRA(mb_type_col) && l1ref0[x8 + y8*h->b8_stride] == 0){
1264 for(i4=0; i4<4; i4++){
1265 const int16_t *mv_col = l1mv0[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1266 if(ABS(mv_col[0]) <= 1 && ABS(mv_col[1]) <= 1){
1268 *(uint32_t*)h->mv_cache[0][scan8[i8*4+i4]] = 0;
1270 *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = 0;
1276 }else{ /* direct temporal mv pred */
1277 if(IS_16X16(*mb_type)){
1278 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1);
1279 if(IS_INTRA(mb_type_col)){
1280 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
1281 fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
1282 fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
1284 const int ref0 = l1ref0[0] >= 0 ? h->map_col_to_list0[0][l1ref0[0]]
1285 : h->map_col_to_list0[1][l1ref1[0]];
1286 const int dist_scale_factor = h->dist_scale_factor[ref0];
1287 const int16_t *mv_col = l1mv0[0];
1289 mv_l0[0] = (dist_scale_factor * mv_col[0] + 128) >> 8;
1290 mv_l0[1] = (dist_scale_factor * mv_col[1] + 128) >> 8;
1291 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref0, 1);
1292 fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv_l0[0],mv_l0[1]), 4);
1293 fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]), 4);
1296 for(i8=0; i8<4; i8++){
1297 const int x8 = i8&1;
1298 const int y8 = i8>>1;
1299 int ref0, dist_scale_factor;
1301 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1303 h->sub_mb_type[i8] = sub_mb_type;
1304 if(IS_INTRA(mb_type_col)){
1305 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1306 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1307 fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1308 fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1312 ref0 = l1ref0[x8 + y8*h->b8_stride];
1314 ref0 = h->map_col_to_list0[0][ref0];
1316 ref0 = h->map_col_to_list0[1][l1ref1[x8 + y8*h->b8_stride]];
1317 dist_scale_factor = h->dist_scale_factor[ref0];
1319 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1320 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1321 for(i4=0; i4<4; i4++){
1322 const int16_t *mv_col = l1mv0[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1323 int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]];
1324 mv_l0[0] = (dist_scale_factor * mv_col[0] + 128) >> 8;
1325 mv_l0[1] = (dist_scale_factor * mv_col[1] + 128) >> 8;
1326 *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] =
1327 pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
1334 static inline void write_back_motion(H264Context *h, int mb_type){
1335 MpegEncContext * const s = &h->s;
1336 const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1337 const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1340 for(list=0; list<2; list++){
1342 if(!USES_LIST(mb_type, list)){
1343 if(1){ //FIXME skip or never read if mb_type doesnt use it
1345 *(uint64_t*)s->current_picture.motion_val[list][b_xy + 0 + y*h->b_stride]=
1346 *(uint64_t*)s->current_picture.motion_val[list][b_xy + 2 + y*h->b_stride]= 0;
1348 if( h->pps.cabac ) {
1349 /* FIXME needed ? */
1351 *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]=
1352 *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= 0;
1356 *(uint16_t*)&s->current_picture.ref_index[list][b8_xy + y*h->b8_stride]= (LIST_NOT_USED&0xFF)*0x0101;
1363 *(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];
1364 *(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];
1366 if( h->pps.cabac ) {
1368 *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+0 + 8*y];
1369 *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+2 + 8*y];
1373 s->current_picture.ref_index[list][b8_xy + 0 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+0 + 16*y];
1374 s->current_picture.ref_index[list][b8_xy + 1 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+2 + 16*y];
1378 if(h->slice_type == B_TYPE && h->pps.cabac){
1379 if(IS_8X8(mb_type)){
1380 h->direct_table[b8_xy+1+0*h->b8_stride] = IS_DIRECT(h->sub_mb_type[1]) ? 1 : 0;
1381 h->direct_table[b8_xy+0+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[2]) ? 1 : 0;
1382 h->direct_table[b8_xy+1+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[3]) ? 1 : 0;
1388 * Decodes a network abstraction layer unit.
1389 * @param consumed is the number of bytes used as input
1390 * @param length is the length of the array
1391 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp ttailing?
1392 * @returns decoded bytes, might be src+1 if no escapes
1394 static uint8_t *decode_nal(H264Context *h, uint8_t *src, int *dst_length, int *consumed, int length){
1398 // src[0]&0x80; //forbidden bit
1399 h->nal_ref_idc= src[0]>>5;
1400 h->nal_unit_type= src[0]&0x1F;
1404 for(i=0; i<length; i++)
1405 printf("%2X ", src[i]);
1407 for(i=0; i+1<length; i+=2){
1408 if(src[i]) continue;
1409 if(i>0 && src[i-1]==0) i--;
1410 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1412 /* startcode, so we must be past the end */
1419 if(i>=length-1){ //no escaped 0
1420 *dst_length= length;
1421 *consumed= length+1; //+1 for the header
1425 h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length);
1426 dst= h->rbsp_buffer;
1428 //printf("deoding esc\n");
1431 //remove escapes (very rare 1:2^22)
1432 if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1433 if(src[si+2]==3){ //escape
1438 }else //next start code
1442 dst[di++]= src[si++];
1446 *consumed= si + 1;//+1 for the header
1447 //FIXME store exact number of bits in the getbitcontext (its needed for decoding)
1453 * @param src the data which should be escaped
1454 * @param dst the target buffer, dst+1 == src is allowed as a special case
1455 * @param length the length of the src data
1456 * @param dst_length the length of the dst array
1457 * @returns length of escaped data in bytes or -1 if an error occured
1459 static int encode_nal(H264Context *h, uint8_t *dst, uint8_t *src, int length, int dst_length){
1460 int i, escape_count, si, di;
1464 assert(dst_length>0);
1466 dst[0]= (h->nal_ref_idc<<5) + h->nal_unit_type;
1468 if(length==0) return 1;
1471 for(i=0; i<length; i+=2){
1472 if(src[i]) continue;
1473 if(i>0 && src[i-1]==0)
1475 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1481 if(escape_count==0){
1483 memcpy(dst+1, src, length);
1487 if(length + escape_count + 1> dst_length)
1490 //this should be damn rare (hopefully)
1492 h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length + escape_count);
1493 temp= h->rbsp_buffer;
1494 //printf("encoding esc\n");
1499 if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1500 temp[di++]= 0; si++;
1501 temp[di++]= 0; si++;
1503 temp[di++]= src[si++];
1506 temp[di++]= src[si++];
1508 memcpy(dst+1, temp, length+escape_count);
1510 assert(di == length+escape_count);
1516 * write 1,10,100,1000,... for alignment, yes its exactly inverse to mpeg4
1518 static void encode_rbsp_trailing(PutBitContext *pb){
1521 length= (-put_bits_count(pb))&7;
1522 if(length) put_bits(pb, length, 0);
1527 * identifies the exact end of the bitstream
1528 * @return the length of the trailing, or 0 if damaged
1530 static int decode_rbsp_trailing(uint8_t *src){
1534 tprintf("rbsp trailing %X\n", v);
1544 * idct tranforms the 16 dc values and dequantize them.
1545 * @param qp quantization parameter
1547 static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp){
1548 const int qmul= dequant_coeff[qp][0];
1551 int temp[16]; //FIXME check if this is a good idea
1552 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
1553 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1555 //memset(block, 64, 2*256);
1558 const int offset= y_offset[i];
1559 const int z0= block[offset+stride*0] + block[offset+stride*4];
1560 const int z1= block[offset+stride*0] - block[offset+stride*4];
1561 const int z2= block[offset+stride*1] - block[offset+stride*5];
1562 const int z3= block[offset+stride*1] + block[offset+stride*5];
1571 const int offset= x_offset[i];
1572 const int z0= temp[4*0+i] + temp[4*2+i];
1573 const int z1= temp[4*0+i] - temp[4*2+i];
1574 const int z2= temp[4*1+i] - temp[4*3+i];
1575 const int z3= temp[4*1+i] + temp[4*3+i];
1577 block[stride*0 +offset]= ((z0 + z3)*qmul + 2)>>2; //FIXME think about merging this into decode_resdual
1578 block[stride*2 +offset]= ((z1 + z2)*qmul + 2)>>2;
1579 block[stride*8 +offset]= ((z1 - z2)*qmul + 2)>>2;
1580 block[stride*10+offset]= ((z0 - z3)*qmul + 2)>>2;
1586 * dct tranforms the 16 dc values.
1587 * @param qp quantization parameter ??? FIXME
1589 static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
1590 // const int qmul= dequant_coeff[qp][0];
1592 int temp[16]; //FIXME check if this is a good idea
1593 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
1594 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1597 const int offset= y_offset[i];
1598 const int z0= block[offset+stride*0] + block[offset+stride*4];
1599 const int z1= block[offset+stride*0] - block[offset+stride*4];
1600 const int z2= block[offset+stride*1] - block[offset+stride*5];
1601 const int z3= block[offset+stride*1] + block[offset+stride*5];
1610 const int offset= x_offset[i];
1611 const int z0= temp[4*0+i] + temp[4*2+i];
1612 const int z1= temp[4*0+i] - temp[4*2+i];
1613 const int z2= temp[4*1+i] - temp[4*3+i];
1614 const int z3= temp[4*1+i] + temp[4*3+i];
1616 block[stride*0 +offset]= (z0 + z3)>>1;
1617 block[stride*2 +offset]= (z1 + z2)>>1;
1618 block[stride*8 +offset]= (z1 - z2)>>1;
1619 block[stride*10+offset]= (z0 - z3)>>1;
1627 static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp){
1628 const int qmul= dequant_coeff[qp][0];
1629 const int stride= 16*2;
1630 const int xStride= 16;
1633 a= block[stride*0 + xStride*0];
1634 b= block[stride*0 + xStride*1];
1635 c= block[stride*1 + xStride*0];
1636 d= block[stride*1 + xStride*1];
1643 block[stride*0 + xStride*0]= ((a+c)*qmul + 0)>>1;
1644 block[stride*0 + xStride*1]= ((e+b)*qmul + 0)>>1;
1645 block[stride*1 + xStride*0]= ((a-c)*qmul + 0)>>1;
1646 block[stride*1 + xStride*1]= ((e-b)*qmul + 0)>>1;
1650 static void chroma_dc_dct_c(DCTELEM *block){
1651 const int stride= 16*2;
1652 const int xStride= 16;
1655 a= block[stride*0 + xStride*0];
1656 b= block[stride*0 + xStride*1];
1657 c= block[stride*1 + xStride*0];
1658 d= block[stride*1 + xStride*1];
1665 block[stride*0 + xStride*0]= (a+c);
1666 block[stride*0 + xStride*1]= (e+b);
1667 block[stride*1 + xStride*0]= (a-c);
1668 block[stride*1 + xStride*1]= (e-b);
1673 * gets the chroma qp.
1675 static inline int get_chroma_qp(int chroma_qp_index_offset, int qscale){
1677 return chroma_qp[clip(qscale + chroma_qp_index_offset, 0, 51)];
1682 static void h264_diff_dct_c(DCTELEM *block, uint8_t *src1, uint8_t *src2, int stride){
1684 //FIXME try int temp instead of block
1687 const int d0= src1[0 + i*stride] - src2[0 + i*stride];
1688 const int d1= src1[1 + i*stride] - src2[1 + i*stride];
1689 const int d2= src1[2 + i*stride] - src2[2 + i*stride];
1690 const int d3= src1[3 + i*stride] - src2[3 + i*stride];
1691 const int z0= d0 + d3;
1692 const int z3= d0 - d3;
1693 const int z1= d1 + d2;
1694 const int z2= d1 - d2;
1696 block[0 + 4*i]= z0 + z1;
1697 block[1 + 4*i]= 2*z3 + z2;
1698 block[2 + 4*i]= z0 - z1;
1699 block[3 + 4*i]= z3 - 2*z2;
1703 const int z0= block[0*4 + i] + block[3*4 + i];
1704 const int z3= block[0*4 + i] - block[3*4 + i];
1705 const int z1= block[1*4 + i] + block[2*4 + i];
1706 const int z2= block[1*4 + i] - block[2*4 + i];
1708 block[0*4 + i]= z0 + z1;
1709 block[1*4 + i]= 2*z3 + z2;
1710 block[2*4 + i]= z0 - z1;
1711 block[3*4 + i]= z3 - 2*z2;
1716 //FIXME need to check that this doesnt overflow signed 32 bit for low qp, iam not sure, its very close
1717 //FIXME check that gcc inlines this (and optimizes intra & seperate_dc stuff away)
1718 static inline int quantize_c(DCTELEM *block, uint8_t *scantable, int qscale, int intra, int seperate_dc){
1720 const int * const quant_table= quant_coeff[qscale];
1721 const int bias= intra ? (1<<QUANT_SHIFT)/3 : (1<<QUANT_SHIFT)/6;
1722 const unsigned int threshold1= (1<<QUANT_SHIFT) - bias - 1;
1723 const unsigned int threshold2= (threshold1<<1);
1729 const int dc_bias= intra ? (1<<(QUANT_SHIFT-2))/3 : (1<<(QUANT_SHIFT-2))/6;
1730 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT-2)) - dc_bias - 1;
1731 const unsigned int dc_threshold2= (dc_threshold1<<1);
1733 int level= block[0]*quant_coeff[qscale+18][0];
1734 if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1736 level= (dc_bias + level)>>(QUANT_SHIFT-2);
1739 level= (dc_bias - level)>>(QUANT_SHIFT-2);
1742 // last_non_zero = i;
1747 const int dc_bias= intra ? (1<<(QUANT_SHIFT+1))/3 : (1<<(QUANT_SHIFT+1))/6;
1748 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT+1)) - dc_bias - 1;
1749 const unsigned int dc_threshold2= (dc_threshold1<<1);
1751 int level= block[0]*quant_table[0];
1752 if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1754 level= (dc_bias + level)>>(QUANT_SHIFT+1);
1757 level= (dc_bias - level)>>(QUANT_SHIFT+1);
1760 // last_non_zero = i;
1773 const int j= scantable[i];
1774 int level= block[j]*quant_table[j];
1776 // if( bias+level >= (1<<(QMAT_SHIFT - 3))
1777 // || bias-level >= (1<<(QMAT_SHIFT - 3))){
1778 if(((unsigned)(level+threshold1))>threshold2){
1780 level= (bias + level)>>QUANT_SHIFT;
1783 level= (bias - level)>>QUANT_SHIFT;
1792 return last_non_zero;
1795 static void pred4x4_vertical_c(uint8_t *src, uint8_t *topright, int stride){
1796 const uint32_t a= ((uint32_t*)(src-stride))[0];
1797 ((uint32_t*)(src+0*stride))[0]= a;
1798 ((uint32_t*)(src+1*stride))[0]= a;
1799 ((uint32_t*)(src+2*stride))[0]= a;
1800 ((uint32_t*)(src+3*stride))[0]= a;
1803 static void pred4x4_horizontal_c(uint8_t *src, uint8_t *topright, int stride){
1804 ((uint32_t*)(src+0*stride))[0]= src[-1+0*stride]*0x01010101;
1805 ((uint32_t*)(src+1*stride))[0]= src[-1+1*stride]*0x01010101;
1806 ((uint32_t*)(src+2*stride))[0]= src[-1+2*stride]*0x01010101;
1807 ((uint32_t*)(src+3*stride))[0]= src[-1+3*stride]*0x01010101;
1810 static void pred4x4_dc_c(uint8_t *src, uint8_t *topright, int stride){
1811 const int dc= ( src[-stride] + src[1-stride] + src[2-stride] + src[3-stride]
1812 + src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 4) >>3;
1814 ((uint32_t*)(src+0*stride))[0]=
1815 ((uint32_t*)(src+1*stride))[0]=
1816 ((uint32_t*)(src+2*stride))[0]=
1817 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1820 static void pred4x4_left_dc_c(uint8_t *src, uint8_t *topright, int stride){
1821 const int dc= ( src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 2) >>2;
1823 ((uint32_t*)(src+0*stride))[0]=
1824 ((uint32_t*)(src+1*stride))[0]=
1825 ((uint32_t*)(src+2*stride))[0]=
1826 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1829 static void pred4x4_top_dc_c(uint8_t *src, uint8_t *topright, int stride){
1830 const int dc= ( src[-stride] + src[1-stride] + src[2-stride] + src[3-stride] + 2) >>2;
1832 ((uint32_t*)(src+0*stride))[0]=
1833 ((uint32_t*)(src+1*stride))[0]=
1834 ((uint32_t*)(src+2*stride))[0]=
1835 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1838 static void pred4x4_128_dc_c(uint8_t *src, uint8_t *topright, int stride){
1839 ((uint32_t*)(src+0*stride))[0]=
1840 ((uint32_t*)(src+1*stride))[0]=
1841 ((uint32_t*)(src+2*stride))[0]=
1842 ((uint32_t*)(src+3*stride))[0]= 128U*0x01010101U;
1846 #define LOAD_TOP_RIGHT_EDGE\
1847 const int t4= topright[0];\
1848 const int t5= topright[1];\
1849 const int t6= topright[2];\
1850 const int t7= topright[3];\
1852 #define LOAD_LEFT_EDGE\
1853 const int l0= src[-1+0*stride];\
1854 const int l1= src[-1+1*stride];\
1855 const int l2= src[-1+2*stride];\
1856 const int l3= src[-1+3*stride];\
1858 #define LOAD_TOP_EDGE\
1859 const int t0= src[ 0-1*stride];\
1860 const int t1= src[ 1-1*stride];\
1861 const int t2= src[ 2-1*stride];\
1862 const int t3= src[ 3-1*stride];\
1864 static void pred4x4_down_right_c(uint8_t *src, uint8_t *topright, int stride){
1865 const int lt= src[-1-1*stride];
1869 src[0+3*stride]=(l3 + 2*l2 + l1 + 2)>>2;
1871 src[1+3*stride]=(l2 + 2*l1 + l0 + 2)>>2;
1874 src[2+3*stride]=(l1 + 2*l0 + lt + 2)>>2;
1878 src[3+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1881 src[3+2*stride]=(lt + 2*t0 + t1 + 2)>>2;
1883 src[3+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1884 src[3+0*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1887 static void pred4x4_down_left_c(uint8_t *src, uint8_t *topright, int stride){
1892 src[0+0*stride]=(t0 + t2 + 2*t1 + 2)>>2;
1894 src[0+1*stride]=(t1 + t3 + 2*t2 + 2)>>2;
1897 src[0+2*stride]=(t2 + t4 + 2*t3 + 2)>>2;
1901 src[0+3*stride]=(t3 + t5 + 2*t4 + 2)>>2;
1904 src[1+3*stride]=(t4 + t6 + 2*t5 + 2)>>2;
1906 src[2+3*stride]=(t5 + t7 + 2*t6 + 2)>>2;
1907 src[3+3*stride]=(t6 + 3*t7 + 2)>>2;
1910 static void pred4x4_vertical_right_c(uint8_t *src, uint8_t *topright, int stride){
1911 const int lt= src[-1-1*stride];
1914 const __attribute__((unused)) int unu= l3;
1917 src[1+2*stride]=(lt + t0 + 1)>>1;
1919 src[2+2*stride]=(t0 + t1 + 1)>>1;
1921 src[3+2*stride]=(t1 + t2 + 1)>>1;
1922 src[3+0*stride]=(t2 + t3 + 1)>>1;
1924 src[1+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1926 src[2+3*stride]=(lt + 2*t0 + t1 + 2)>>2;
1928 src[3+3*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1929 src[3+1*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1930 src[0+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1931 src[0+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1934 static void pred4x4_vertical_left_c(uint8_t *src, uint8_t *topright, int stride){
1937 const __attribute__((unused)) int unu= t7;
1939 src[0+0*stride]=(t0 + t1 + 1)>>1;
1941 src[0+2*stride]=(t1 + t2 + 1)>>1;
1943 src[1+2*stride]=(t2 + t3 + 1)>>1;
1945 src[2+2*stride]=(t3 + t4+ 1)>>1;
1946 src[3+2*stride]=(t4 + t5+ 1)>>1;
1947 src[0+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1949 src[0+3*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1951 src[1+3*stride]=(t2 + 2*t3 + t4 + 2)>>2;
1953 src[2+3*stride]=(t3 + 2*t4 + t5 + 2)>>2;
1954 src[3+3*stride]=(t4 + 2*t5 + t6 + 2)>>2;
1957 static void pred4x4_horizontal_up_c(uint8_t *src, uint8_t *topright, int stride){
1960 src[0+0*stride]=(l0 + l1 + 1)>>1;
1961 src[1+0*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1963 src[0+1*stride]=(l1 + l2 + 1)>>1;
1965 src[1+1*stride]=(l1 + 2*l2 + l3 + 2)>>2;
1967 src[0+2*stride]=(l2 + l3 + 1)>>1;
1969 src[1+2*stride]=(l2 + 2*l3 + l3 + 2)>>2;
1978 static void pred4x4_horizontal_down_c(uint8_t *src, uint8_t *topright, int stride){
1979 const int lt= src[-1-1*stride];
1982 const __attribute__((unused)) int unu= t3;
1985 src[2+1*stride]=(lt + l0 + 1)>>1;
1987 src[3+1*stride]=(l0 + 2*lt + t0 + 2)>>2;
1988 src[2+0*stride]=(lt + 2*t0 + t1 + 2)>>2;
1989 src[3+0*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1991 src[2+2*stride]=(l0 + l1 + 1)>>1;
1993 src[3+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1995 src[2+3*stride]=(l1 + l2+ 1)>>1;
1997 src[3+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1998 src[0+3*stride]=(l2 + l3 + 1)>>1;
1999 src[1+3*stride]=(l1 + 2*l2 + l3 + 2)>>2;
2002 static void pred16x16_vertical_c(uint8_t *src, int stride){
2004 const uint32_t a= ((uint32_t*)(src-stride))[0];
2005 const uint32_t b= ((uint32_t*)(src-stride))[1];
2006 const uint32_t c= ((uint32_t*)(src-stride))[2];
2007 const uint32_t d= ((uint32_t*)(src-stride))[3];
2009 for(i=0; i<16; i++){
2010 ((uint32_t*)(src+i*stride))[0]= a;
2011 ((uint32_t*)(src+i*stride))[1]= b;
2012 ((uint32_t*)(src+i*stride))[2]= c;
2013 ((uint32_t*)(src+i*stride))[3]= d;
2017 static void pred16x16_horizontal_c(uint8_t *src, int stride){
2020 for(i=0; i<16; i++){
2021 ((uint32_t*)(src+i*stride))[0]=
2022 ((uint32_t*)(src+i*stride))[1]=
2023 ((uint32_t*)(src+i*stride))[2]=
2024 ((uint32_t*)(src+i*stride))[3]= src[-1+i*stride]*0x01010101;
2028 static void pred16x16_dc_c(uint8_t *src, int stride){
2032 dc+= src[-1+i*stride];
2039 dc= 0x01010101*((dc + 16)>>5);
2041 for(i=0; i<16; i++){
2042 ((uint32_t*)(src+i*stride))[0]=
2043 ((uint32_t*)(src+i*stride))[1]=
2044 ((uint32_t*)(src+i*stride))[2]=
2045 ((uint32_t*)(src+i*stride))[3]= dc;
2049 static void pred16x16_left_dc_c(uint8_t *src, int stride){
2053 dc+= src[-1+i*stride];
2056 dc= 0x01010101*((dc + 8)>>4);
2058 for(i=0; i<16; i++){
2059 ((uint32_t*)(src+i*stride))[0]=
2060 ((uint32_t*)(src+i*stride))[1]=
2061 ((uint32_t*)(src+i*stride))[2]=
2062 ((uint32_t*)(src+i*stride))[3]= dc;
2066 static void pred16x16_top_dc_c(uint8_t *src, int stride){
2072 dc= 0x01010101*((dc + 8)>>4);
2074 for(i=0; i<16; i++){
2075 ((uint32_t*)(src+i*stride))[0]=
2076 ((uint32_t*)(src+i*stride))[1]=
2077 ((uint32_t*)(src+i*stride))[2]=
2078 ((uint32_t*)(src+i*stride))[3]= dc;
2082 static void pred16x16_128_dc_c(uint8_t *src, int stride){
2085 for(i=0; i<16; i++){
2086 ((uint32_t*)(src+i*stride))[0]=
2087 ((uint32_t*)(src+i*stride))[1]=
2088 ((uint32_t*)(src+i*stride))[2]=
2089 ((uint32_t*)(src+i*stride))[3]= 0x01010101U*128U;
2093 static inline void pred16x16_plane_compat_c(uint8_t *src, int stride, const int svq3){
2096 uint8_t *cm = cropTbl + MAX_NEG_CROP;
2097 const uint8_t * const src0 = src+7-stride;
2098 const uint8_t *src1 = src+8*stride-1;
2099 const uint8_t *src2 = src1-2*stride; // == src+6*stride-1;
2100 int H = src0[1] - src0[-1];
2101 int V = src1[0] - src2[ 0];
2102 for(k=2; k<=8; ++k) {
2103 src1 += stride; src2 -= stride;
2104 H += k*(src0[k] - src0[-k]);
2105 V += k*(src1[0] - src2[ 0]);
2108 H = ( 5*(H/4) ) / 16;
2109 V = ( 5*(V/4) ) / 16;
2111 /* required for 100% accuracy */
2112 i = H; H = V; V = i;
2114 H = ( 5*H+32 ) >> 6;
2115 V = ( 5*V+32 ) >> 6;
2118 a = 16*(src1[0] + src2[16] + 1) - 7*(V+H);
2119 for(j=16; j>0; --j) {
2122 for(i=-16; i<0; i+=4) {
2123 src[16+i] = cm[ (b ) >> 5 ];
2124 src[17+i] = cm[ (b+ H) >> 5 ];
2125 src[18+i] = cm[ (b+2*H) >> 5 ];
2126 src[19+i] = cm[ (b+3*H) >> 5 ];
2133 static void pred16x16_plane_c(uint8_t *src, int stride){
2134 pred16x16_plane_compat_c(src, stride, 0);
2137 static void pred8x8_vertical_c(uint8_t *src, int stride){
2139 const uint32_t a= ((uint32_t*)(src-stride))[0];
2140 const uint32_t b= ((uint32_t*)(src-stride))[1];
2143 ((uint32_t*)(src+i*stride))[0]= a;
2144 ((uint32_t*)(src+i*stride))[1]= b;
2148 static void pred8x8_horizontal_c(uint8_t *src, int stride){
2152 ((uint32_t*)(src+i*stride))[0]=
2153 ((uint32_t*)(src+i*stride))[1]= src[-1+i*stride]*0x01010101;
2157 static void pred8x8_128_dc_c(uint8_t *src, int stride){
2161 ((uint32_t*)(src+i*stride))[0]=
2162 ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
2165 ((uint32_t*)(src+i*stride))[0]=
2166 ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
2170 static void pred8x8_left_dc_c(uint8_t *src, int stride){
2176 dc0+= src[-1+i*stride];
2177 dc2+= src[-1+(i+4)*stride];
2179 dc0= 0x01010101*((dc0 + 2)>>2);
2180 dc2= 0x01010101*((dc2 + 2)>>2);
2183 ((uint32_t*)(src+i*stride))[0]=
2184 ((uint32_t*)(src+i*stride))[1]= dc0;
2187 ((uint32_t*)(src+i*stride))[0]=
2188 ((uint32_t*)(src+i*stride))[1]= dc2;
2192 static void pred8x8_top_dc_c(uint8_t *src, int stride){
2198 dc0+= src[i-stride];
2199 dc1+= src[4+i-stride];
2201 dc0= 0x01010101*((dc0 + 2)>>2);
2202 dc1= 0x01010101*((dc1 + 2)>>2);
2205 ((uint32_t*)(src+i*stride))[0]= dc0;
2206 ((uint32_t*)(src+i*stride))[1]= dc1;
2209 ((uint32_t*)(src+i*stride))[0]= dc0;
2210 ((uint32_t*)(src+i*stride))[1]= dc1;
2215 static void pred8x8_dc_c(uint8_t *src, int stride){
2217 int dc0, dc1, dc2, dc3;
2221 dc0+= src[-1+i*stride] + src[i-stride];
2222 dc1+= src[4+i-stride];
2223 dc2+= src[-1+(i+4)*stride];
2225 dc3= 0x01010101*((dc1 + dc2 + 4)>>3);
2226 dc0= 0x01010101*((dc0 + 4)>>3);
2227 dc1= 0x01010101*((dc1 + 2)>>2);
2228 dc2= 0x01010101*((dc2 + 2)>>2);
2231 ((uint32_t*)(src+i*stride))[0]= dc0;
2232 ((uint32_t*)(src+i*stride))[1]= dc1;
2235 ((uint32_t*)(src+i*stride))[0]= dc2;
2236 ((uint32_t*)(src+i*stride))[1]= dc3;
2240 static void pred8x8_plane_c(uint8_t *src, int stride){
2243 uint8_t *cm = cropTbl + MAX_NEG_CROP;
2244 const uint8_t * const src0 = src+3-stride;
2245 const uint8_t *src1 = src+4*stride-1;
2246 const uint8_t *src2 = src1-2*stride; // == src+2*stride-1;
2247 int H = src0[1] - src0[-1];
2248 int V = src1[0] - src2[ 0];
2249 for(k=2; k<=4; ++k) {
2250 src1 += stride; src2 -= stride;
2251 H += k*(src0[k] - src0[-k]);
2252 V += k*(src1[0] - src2[ 0]);
2254 H = ( 17*H+16 ) >> 5;
2255 V = ( 17*V+16 ) >> 5;
2257 a = 16*(src1[0] + src2[8]+1) - 3*(V+H);
2258 for(j=8; j>0; --j) {
2261 src[0] = cm[ (b ) >> 5 ];
2262 src[1] = cm[ (b+ H) >> 5 ];
2263 src[2] = cm[ (b+2*H) >> 5 ];
2264 src[3] = cm[ (b+3*H) >> 5 ];
2265 src[4] = cm[ (b+4*H) >> 5 ];
2266 src[5] = cm[ (b+5*H) >> 5 ];
2267 src[6] = cm[ (b+6*H) >> 5 ];
2268 src[7] = cm[ (b+7*H) >> 5 ];
2273 static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
2274 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2275 int src_x_offset, int src_y_offset,
2276 qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
2277 MpegEncContext * const s = &h->s;
2278 const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
2279 const int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
2280 const int luma_xy= (mx&3) + ((my&3)<<2);
2281 uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*s->linesize;
2282 uint8_t * src_cb= pic->data[1] + (mx>>3) + (my>>3)*s->uvlinesize;
2283 uint8_t * src_cr= pic->data[2] + (mx>>3) + (my>>3)*s->uvlinesize;
2284 int extra_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16; //FIXME increase edge?, IMHO not worth it
2285 int extra_height= extra_width;
2287 const int full_mx= mx>>2;
2288 const int full_my= my>>2;
2290 assert(pic->data[0]);
2292 if(mx&7) extra_width -= 3;
2293 if(my&7) extra_height -= 3;
2295 if( full_mx < 0-extra_width
2296 || full_my < 0-extra_height
2297 || full_mx + 16/*FIXME*/ > s->width + extra_width
2298 || full_my + 16/*FIXME*/ > s->height + extra_height){
2299 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);
2300 src_y= s->edge_emu_buffer + 2 + 2*s->linesize;
2304 qpix_op[luma_xy](dest_y, src_y, s->linesize); //FIXME try variable height perhaps?
2306 qpix_op[luma_xy](dest_y + delta, src_y + delta, s->linesize);
2309 if(s->flags&CODEC_FLAG_GRAY) return;
2312 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);
2313 src_cb= s->edge_emu_buffer;
2315 chroma_op(dest_cb, src_cb, s->uvlinesize, chroma_height, mx&7, my&7);
2318 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);
2319 src_cr= s->edge_emu_buffer;
2321 chroma_op(dest_cr, src_cr, s->uvlinesize, chroma_height, mx&7, my&7);
2324 static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
2325 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2326 int x_offset, int y_offset,
2327 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2328 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
2329 int list0, int list1){
2330 MpegEncContext * const s = &h->s;
2331 qpel_mc_func *qpix_op= qpix_put;
2332 h264_chroma_mc_func chroma_op= chroma_put;
2334 dest_y += 2*x_offset + 2*y_offset*s-> linesize;
2335 dest_cb += x_offset + y_offset*s->uvlinesize;
2336 dest_cr += x_offset + y_offset*s->uvlinesize;
2337 x_offset += 8*s->mb_x;
2338 y_offset += 8*s->mb_y;
2341 Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
2342 mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
2343 dest_y, dest_cb, dest_cr, x_offset, y_offset,
2344 qpix_op, chroma_op);
2347 chroma_op= chroma_avg;
2351 Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
2352 mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
2353 dest_y, dest_cb, dest_cr, x_offset, y_offset,
2354 qpix_op, chroma_op);
2358 static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
2359 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2360 int x_offset, int y_offset,
2361 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2362 h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
2363 h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
2364 int list0, int list1){
2365 MpegEncContext * const s = &h->s;
2367 dest_y += 2*x_offset + 2*y_offset*s-> linesize;
2368 dest_cb += x_offset + y_offset*s->uvlinesize;
2369 dest_cr += x_offset + y_offset*s->uvlinesize;
2370 x_offset += 8*s->mb_x;
2371 y_offset += 8*s->mb_y;
2374 /* don't optimize for luma-only case, since B-frames usually
2375 * use implicit weights => chroma too. */
2376 uint8_t *tmp_cb = s->obmc_scratchpad;
2377 uint8_t *tmp_cr = tmp_cb + 8*s->uvlinesize;
2378 uint8_t *tmp_y = tmp_cr + 8*s->uvlinesize;
2379 int refn0 = h->ref_cache[0][ scan8[n] ];
2380 int refn1 = h->ref_cache[1][ scan8[n] ];
2382 mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
2383 dest_y, dest_cb, dest_cr,
2384 x_offset, y_offset, qpix_put, chroma_put);
2385 mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
2386 tmp_y, tmp_cb, tmp_cr,
2387 x_offset, y_offset, qpix_put, chroma_put);
2389 if(h->use_weight == 2){
2390 int weight0 = h->implicit_weight[refn0][refn1];
2391 int weight1 = 64 - weight0;
2392 luma_weight_avg( dest_y, tmp_y, s-> linesize, 5, weight0, weight1, 0, 0);
2393 chroma_weight_avg(dest_cb, tmp_cb, s->uvlinesize, 5, weight0, weight1, 0, 0);
2394 chroma_weight_avg(dest_cr, tmp_cr, s->uvlinesize, 5, weight0, weight1, 0, 0);
2396 luma_weight_avg(dest_y, tmp_y, s->linesize, h->luma_log2_weight_denom,
2397 h->luma_weight[0][refn0], h->luma_weight[1][refn1],
2398 h->luma_offset[0][refn0], h->luma_offset[1][refn1]);
2399 chroma_weight_avg(dest_cb, tmp_cb, s->uvlinesize, h->chroma_log2_weight_denom,
2400 h->chroma_weight[0][refn0][0], h->chroma_weight[1][refn1][0],
2401 h->chroma_offset[0][refn0][0], h->chroma_offset[1][refn1][0]);
2402 chroma_weight_avg(dest_cr, tmp_cr, s->uvlinesize, h->chroma_log2_weight_denom,
2403 h->chroma_weight[0][refn0][1], h->chroma_weight[1][refn1][1],
2404 h->chroma_offset[0][refn0][1], h->chroma_offset[1][refn1][1]);
2407 int list = list1 ? 1 : 0;
2408 int refn = h->ref_cache[list][ scan8[n] ];
2409 Picture *ref= &h->ref_list[list][refn];
2410 mc_dir_part(h, ref, n, square, chroma_height, delta, list,
2411 dest_y, dest_cb, dest_cr, x_offset, y_offset,
2412 qpix_put, chroma_put);
2414 luma_weight_op(dest_y, s->linesize, h->luma_log2_weight_denom,
2415 h->luma_weight[list][refn], h->luma_offset[list][refn]);
2416 if(h->use_weight_chroma){
2417 chroma_weight_op(dest_cb, s->uvlinesize, h->chroma_log2_weight_denom,
2418 h->chroma_weight[list][refn][0], h->chroma_offset[list][refn][0]);
2419 chroma_weight_op(dest_cr, s->uvlinesize, h->chroma_log2_weight_denom,
2420 h->chroma_weight[list][refn][1], h->chroma_offset[list][refn][1]);
2425 static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
2426 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2427 int x_offset, int y_offset,
2428 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2429 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
2430 h264_weight_func *weight_op, h264_biweight_func *weight_avg,
2431 int list0, int list1){
2432 if((h->use_weight==2 && list0 && list1
2433 && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ] != 32))
2434 || h->use_weight==1)
2435 mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
2436 x_offset, y_offset, qpix_put, chroma_put,
2437 weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
2439 mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
2440 x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
2443 static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2444 qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
2445 qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
2446 h264_weight_func *weight_op, h264_biweight_func *weight_avg){
2447 MpegEncContext * const s = &h->s;
2448 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
2449 const int mb_type= s->current_picture.mb_type[mb_xy];
2451 assert(IS_INTER(mb_type));
2453 if(IS_16X16(mb_type)){
2454 mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
2455 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
2456 &weight_op[0], &weight_avg[0],
2457 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2458 }else if(IS_16X8(mb_type)){
2459 mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
2460 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
2461 &weight_op[1], &weight_avg[1],
2462 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2463 mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
2464 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
2465 &weight_op[1], &weight_avg[1],
2466 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2467 }else if(IS_8X16(mb_type)){
2468 mc_part(h, 0, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 0, 0,
2469 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2470 &weight_op[2], &weight_avg[2],
2471 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2472 mc_part(h, 4, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 4, 0,
2473 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2474 &weight_op[2], &weight_avg[2],
2475 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2479 assert(IS_8X8(mb_type));
2482 const int sub_mb_type= h->sub_mb_type[i];
2484 int x_offset= (i&1)<<2;
2485 int y_offset= (i&2)<<1;
2487 if(IS_SUB_8X8(sub_mb_type)){
2488 mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2489 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2490 &weight_op[3], &weight_avg[3],
2491 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2492 }else if(IS_SUB_8X4(sub_mb_type)){
2493 mc_part(h, n , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2494 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2495 &weight_op[4], &weight_avg[4],
2496 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2497 mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
2498 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2499 &weight_op[4], &weight_avg[4],
2500 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2501 }else if(IS_SUB_4X8(sub_mb_type)){
2502 mc_part(h, n , 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2503 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2504 &weight_op[5], &weight_avg[5],
2505 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2506 mc_part(h, n+1, 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
2507 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2508 &weight_op[5], &weight_avg[5],
2509 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2512 assert(IS_SUB_4X4(sub_mb_type));
2514 int sub_x_offset= x_offset + 2*(j&1);
2515 int sub_y_offset= y_offset + (j&2);
2516 mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
2517 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2518 &weight_op[6], &weight_avg[6],
2519 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2526 static void decode_init_vlc(H264Context *h){
2527 static int done = 0;
2533 init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5,
2534 &chroma_dc_coeff_token_len [0], 1, 1,
2535 &chroma_dc_coeff_token_bits[0], 1, 1, 1);
2538 init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17,
2539 &coeff_token_len [i][0], 1, 1,
2540 &coeff_token_bits[i][0], 1, 1, 1);
2544 init_vlc(&chroma_dc_total_zeros_vlc[i], CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
2545 &chroma_dc_total_zeros_len [i][0], 1, 1,
2546 &chroma_dc_total_zeros_bits[i][0], 1, 1, 1);
2548 for(i=0; i<15; i++){
2549 init_vlc(&total_zeros_vlc[i], TOTAL_ZEROS_VLC_BITS, 16,
2550 &total_zeros_len [i][0], 1, 1,
2551 &total_zeros_bits[i][0], 1, 1, 1);
2555 init_vlc(&run_vlc[i], RUN_VLC_BITS, 7,
2556 &run_len [i][0], 1, 1,
2557 &run_bits[i][0], 1, 1, 1);
2559 init_vlc(&run7_vlc, RUN7_VLC_BITS, 16,
2560 &run_len [6][0], 1, 1,
2561 &run_bits[6][0], 1, 1, 1);
2566 * Sets the intra prediction function pointers.
2568 static void init_pred_ptrs(H264Context *h){
2569 // MpegEncContext * const s = &h->s;
2571 h->pred4x4[VERT_PRED ]= pred4x4_vertical_c;
2572 h->pred4x4[HOR_PRED ]= pred4x4_horizontal_c;
2573 h->pred4x4[DC_PRED ]= pred4x4_dc_c;
2574 h->pred4x4[DIAG_DOWN_LEFT_PRED ]= pred4x4_down_left_c;
2575 h->pred4x4[DIAG_DOWN_RIGHT_PRED]= pred4x4_down_right_c;
2576 h->pred4x4[VERT_RIGHT_PRED ]= pred4x4_vertical_right_c;
2577 h->pred4x4[HOR_DOWN_PRED ]= pred4x4_horizontal_down_c;
2578 h->pred4x4[VERT_LEFT_PRED ]= pred4x4_vertical_left_c;
2579 h->pred4x4[HOR_UP_PRED ]= pred4x4_horizontal_up_c;
2580 h->pred4x4[LEFT_DC_PRED ]= pred4x4_left_dc_c;
2581 h->pred4x4[TOP_DC_PRED ]= pred4x4_top_dc_c;
2582 h->pred4x4[DC_128_PRED ]= pred4x4_128_dc_c;
2584 h->pred8x8[DC_PRED8x8 ]= pred8x8_dc_c;
2585 h->pred8x8[VERT_PRED8x8 ]= pred8x8_vertical_c;
2586 h->pred8x8[HOR_PRED8x8 ]= pred8x8_horizontal_c;
2587 h->pred8x8[PLANE_PRED8x8 ]= pred8x8_plane_c;
2588 h->pred8x8[LEFT_DC_PRED8x8]= pred8x8_left_dc_c;
2589 h->pred8x8[TOP_DC_PRED8x8 ]= pred8x8_top_dc_c;
2590 h->pred8x8[DC_128_PRED8x8 ]= pred8x8_128_dc_c;
2592 h->pred16x16[DC_PRED8x8 ]= pred16x16_dc_c;
2593 h->pred16x16[VERT_PRED8x8 ]= pred16x16_vertical_c;
2594 h->pred16x16[HOR_PRED8x8 ]= pred16x16_horizontal_c;
2595 h->pred16x16[PLANE_PRED8x8 ]= pred16x16_plane_c;
2596 h->pred16x16[LEFT_DC_PRED8x8]= pred16x16_left_dc_c;
2597 h->pred16x16[TOP_DC_PRED8x8 ]= pred16x16_top_dc_c;
2598 h->pred16x16[DC_128_PRED8x8 ]= pred16x16_128_dc_c;
2601 static void free_tables(H264Context *h){
2602 av_freep(&h->intra4x4_pred_mode);
2603 av_freep(&h->chroma_pred_mode_table);
2604 av_freep(&h->cbp_table);
2605 av_freep(&h->mvd_table[0]);
2606 av_freep(&h->mvd_table[1]);
2607 av_freep(&h->direct_table);
2608 av_freep(&h->non_zero_count);
2609 av_freep(&h->slice_table_base);
2610 av_freep(&h->top_borders[1]);
2611 av_freep(&h->top_borders[0]);
2612 h->slice_table= NULL;
2614 av_freep(&h->mb2b_xy);
2615 av_freep(&h->mb2b8_xy);
2617 av_freep(&h->s.obmc_scratchpad);
2622 * needs widzh/height
2624 static int alloc_tables(H264Context *h){
2625 MpegEncContext * const s = &h->s;
2626 const int big_mb_num= s->mb_stride * (s->mb_height+1);
2629 CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8 * sizeof(uint8_t))
2631 CHECKED_ALLOCZ(h->non_zero_count , big_mb_num * 16 * sizeof(uint8_t))
2632 CHECKED_ALLOCZ(h->slice_table_base , big_mb_num * sizeof(uint8_t))
2633 CHECKED_ALLOCZ(h->top_borders[0] , s->mb_width * (16+8+8) * sizeof(uint8_t))
2634 CHECKED_ALLOCZ(h->top_borders[1] , s->mb_width * (16+8+8) * sizeof(uint8_t))
2635 CHECKED_ALLOCZ(h->cbp_table, big_mb_num * sizeof(uint16_t))
2637 if( h->pps.cabac ) {
2638 CHECKED_ALLOCZ(h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t))
2639 CHECKED_ALLOCZ(h->mvd_table[0], 32*big_mb_num * sizeof(uint16_t));
2640 CHECKED_ALLOCZ(h->mvd_table[1], 32*big_mb_num * sizeof(uint16_t));
2641 CHECKED_ALLOCZ(h->direct_table, 32*big_mb_num * sizeof(uint8_t));
2644 memset(h->slice_table_base, -1, big_mb_num * sizeof(uint8_t));
2645 h->slice_table= h->slice_table_base + s->mb_stride + 1;
2647 CHECKED_ALLOCZ(h->mb2b_xy , big_mb_num * sizeof(uint16_t));
2648 CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint16_t));
2649 for(y=0; y<s->mb_height; y++){
2650 for(x=0; x<s->mb_width; x++){
2651 const int mb_xy= x + y*s->mb_stride;
2652 const int b_xy = 4*x + 4*y*h->b_stride;
2653 const int b8_xy= 2*x + 2*y*h->b8_stride;
2655 h->mb2b_xy [mb_xy]= b_xy;
2656 h->mb2b8_xy[mb_xy]= b8_xy;
2660 s->obmc_scratchpad = NULL;
2668 static void common_init(H264Context *h){
2669 MpegEncContext * const s = &h->s;
2671 s->width = s->avctx->width;
2672 s->height = s->avctx->height;
2673 s->codec_id= s->avctx->codec->id;
2677 s->unrestricted_mv=1;
2678 s->decode=1; //FIXME
2681 static int decode_init(AVCodecContext *avctx){
2682 H264Context *h= avctx->priv_data;
2683 MpegEncContext * const s = &h->s;
2685 MPV_decode_defaults(s);
2690 s->out_format = FMT_H264;
2691 s->workaround_bugs= avctx->workaround_bugs;
2694 // s->decode_mb= ff_h263_decode_mb;
2696 avctx->pix_fmt= PIX_FMT_YUV420P;
2700 if(avctx->extradata_size > 0 && avctx->extradata &&
2701 *(char *)avctx->extradata == 1){
2711 static void frame_start(H264Context *h){
2712 MpegEncContext * const s = &h->s;
2715 MPV_frame_start(s, s->avctx);
2716 ff_er_frame_start(s);
2718 assert(s->linesize && s->uvlinesize);
2720 for(i=0; i<16; i++){
2721 h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
2722 h->block_offset[24+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
2725 h->block_offset[16+i]=
2726 h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2727 h->block_offset[24+16+i]=
2728 h->block_offset[24+20+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2731 /* can't be in alloc_tables because linesize isn't known there.
2732 * FIXME: redo bipred weight to not require extra buffer? */
2733 if(!s->obmc_scratchpad)
2734 s->obmc_scratchpad = av_malloc(16*s->linesize + 2*8*s->uvlinesize);
2736 // s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
2739 static inline void backup_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){
2740 MpegEncContext * const s = &h->s;
2744 src_cb -= uvlinesize;
2745 src_cr -= uvlinesize;
2747 // There is two lines saved, the line above the the top macroblock of a pair,
2748 // and the line above the bottom macroblock
2749 h->left_border[0]= h->top_borders[0][s->mb_x][15];
2750 for(i=1; i<17; i++){
2751 h->left_border[i]= src_y[15+i* linesize];
2754 *(uint64_t*)(h->top_borders[0][s->mb_x]+0)= *(uint64_t*)(src_y + 16*linesize);
2755 *(uint64_t*)(h->top_borders[0][s->mb_x]+8)= *(uint64_t*)(src_y +8+16*linesize);
2757 if(!(s->flags&CODEC_FLAG_GRAY)){
2758 h->left_border[17 ]= h->top_borders[0][s->mb_x][16+7];
2759 h->left_border[17+9]= h->top_borders[0][s->mb_x][24+7];
2761 h->left_border[i+17 ]= src_cb[7+i*uvlinesize];
2762 h->left_border[i+17+9]= src_cr[7+i*uvlinesize];
2764 *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+8*uvlinesize);
2765 *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+8*uvlinesize);
2769 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){
2770 MpegEncContext * const s = &h->s;
2773 int deblock_left = (s->mb_x > 0);
2774 int deblock_top = (s->mb_y > 0);
2776 src_y -= linesize + 1;
2777 src_cb -= uvlinesize + 1;
2778 src_cr -= uvlinesize + 1;
2780 #define XCHG(a,b,t,xchg)\
2787 for(i = !deblock_top; i<17; i++){
2788 XCHG(h->left_border[i ], src_y [i* linesize], temp8, xchg);
2793 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
2794 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
2797 if(!(s->flags&CODEC_FLAG_GRAY)){
2799 for(i = !deblock_top; i<9; i++){
2800 XCHG(h->left_border[i+17 ], src_cb[i*uvlinesize], temp8, xchg);
2801 XCHG(h->left_border[i+17+9], src_cr[i*uvlinesize], temp8, xchg);
2805 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
2806 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
2811 static inline void backup_pair_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){
2812 MpegEncContext * const s = &h->s;
2815 src_y -= 2 * linesize;
2816 src_cb -= 2 * uvlinesize;
2817 src_cr -= 2 * uvlinesize;
2819 // There is two lines saved, the line above the the top macroblock of a pair,
2820 // and the line above the bottom macroblock
2821 h->left_border[0]= h->top_borders[0][s->mb_x][15];
2822 h->left_border[1]= h->top_borders[1][s->mb_x][15];
2823 for(i=2; i<34; i++){
2824 h->left_border[i]= src_y[15+i* linesize];
2827 *(uint64_t*)(h->top_borders[0][s->mb_x]+0)= *(uint64_t*)(src_y + 32*linesize);
2828 *(uint64_t*)(h->top_borders[0][s->mb_x]+8)= *(uint64_t*)(src_y +8+32*linesize);
2829 *(uint64_t*)(h->top_borders[1][s->mb_x]+0)= *(uint64_t*)(src_y + 33*linesize);
2830 *(uint64_t*)(h->top_borders[1][s->mb_x]+8)= *(uint64_t*)(src_y +8+33*linesize);
2832 if(!(s->flags&CODEC_FLAG_GRAY)){
2833 h->left_border[34 ]= h->top_borders[0][s->mb_x][16+7];
2834 h->left_border[34+ 1]= h->top_borders[1][s->mb_x][16+7];
2835 h->left_border[34+18 ]= h->top_borders[0][s->mb_x][24+7];
2836 h->left_border[34+18+1]= h->top_borders[1][s->mb_x][24+7];
2837 for(i=2; i<18; i++){
2838 h->left_border[i+34 ]= src_cb[7+i*uvlinesize];
2839 h->left_border[i+34+18]= src_cr[7+i*uvlinesize];
2841 *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+16*uvlinesize);
2842 *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+16*uvlinesize);
2843 *(uint64_t*)(h->top_borders[1][s->mb_x]+16)= *(uint64_t*)(src_cb+17*uvlinesize);
2844 *(uint64_t*)(h->top_borders[1][s->mb_x]+24)= *(uint64_t*)(src_cr+17*uvlinesize);
2848 static inline void xchg_pair_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize, int xchg){
2849 MpegEncContext * const s = &h->s;
2852 int deblock_left = (s->mb_x > 0);
2853 int deblock_top = (s->mb_y > 0);
2855 tprintf("xchg_pair_border: src_y:%p src_cb:%p src_cr:%p ls:%d uvls:%d\n", src_y, src_cb, src_cr, linesize, uvlinesize);
2857 src_y -= 2 * linesize + 1;
2858 src_cb -= 2 * uvlinesize + 1;
2859 src_cr -= 2 * uvlinesize + 1;
2861 #define XCHG(a,b,t,xchg)\
2868 for(i = (!deblock_top)<<1; i<34; i++){
2869 XCHG(h->left_border[i ], src_y [i* linesize], temp8, xchg);
2874 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
2875 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
2876 XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+0), *(uint64_t*)(src_y +1 +linesize), temp64, xchg);
2877 XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+8), *(uint64_t*)(src_y +9 +linesize), temp64, 1);
2880 if(!(s->flags&CODEC_FLAG_GRAY)){
2882 for(i = (!deblock_top) << 1; i<18; i++){
2883 XCHG(h->left_border[i+34 ], src_cb[i*uvlinesize], temp8, xchg);
2884 XCHG(h->left_border[i+34+18], src_cr[i*uvlinesize], temp8, xchg);
2888 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
2889 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
2890 XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+16), *(uint64_t*)(src_cb+1 +uvlinesize), temp64, 1);
2891 XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+24), *(uint64_t*)(src_cr+1 +uvlinesize), temp64, 1);
2896 static void hl_decode_mb(H264Context *h){
2897 MpegEncContext * const s = &h->s;
2898 const int mb_x= s->mb_x;
2899 const int mb_y= s->mb_y;
2900 const int mb_xy= mb_x + mb_y*s->mb_stride;
2901 const int mb_type= s->current_picture.mb_type[mb_xy];
2902 uint8_t *dest_y, *dest_cb, *dest_cr;
2903 int linesize, uvlinesize /*dct_offset*/;
2905 int *block_offset = &h->block_offset[0];
2906 const unsigned int bottom = mb_y & 1;
2911 dest_y = s->current_picture.data[0] + (mb_y * 16* s->linesize ) + mb_x * 16;
2912 dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2913 dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2915 if (h->mb_field_decoding_flag) {
2916 linesize = s->linesize * 2;
2917 uvlinesize = s->uvlinesize * 2;
2918 block_offset = &h->block_offset[24];
2919 if(mb_y&1){ //FIXME move out of this func?
2920 dest_y -= s->linesize*15;
2921 dest_cb-= s->uvlinesize*7;
2922 dest_cr-= s->uvlinesize*7;
2925 linesize = s->linesize;
2926 uvlinesize = s->uvlinesize;
2927 // dct_offset = s->linesize * 16;
2930 if (IS_INTRA_PCM(mb_type)) {
2933 // The pixels are stored in h->mb array in the same order as levels,
2934 // copy them in output in the correct order.
2935 for(i=0; i<16; i++) {
2936 for (y=0; y<4; y++) {
2937 for (x=0; x<4; x++) {
2938 *(dest_y + block_offset[i] + y*linesize + x) = h->mb[i*16+y*4+x];
2942 for(i=16; i<16+4; i++) {
2943 for (y=0; y<4; y++) {
2944 for (x=0; x<4; x++) {
2945 *(dest_cb + block_offset[i] + y*uvlinesize + x) = h->mb[i*16+y*4+x];
2949 for(i=20; i<20+4; i++) {
2950 for (y=0; y<4; y++) {
2951 for (x=0; x<4; x++) {
2952 *(dest_cr + block_offset[i] + y*uvlinesize + x) = h->mb[i*16+y*4+x];
2957 if(IS_INTRA(mb_type)){
2958 if(h->deblocking_filter) {
2959 if (h->mb_aff_frame) {
2961 xchg_pair_border(h, dest_y, dest_cb, dest_cr, s->linesize, s->uvlinesize, 1);
2963 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1);
2967 if(!(s->flags&CODEC_FLAG_GRAY)){
2968 h->pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
2969 h->pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
2972 if(IS_INTRA4x4(mb_type)){
2974 for(i=0; i<16; i++){
2975 uint8_t * const ptr= dest_y + block_offset[i];
2977 const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
2980 if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
2981 const int topright_avail= (h->topright_samples_available<<i)&0x8000;
2982 assert(mb_y || linesize <= block_offset[i]);
2983 if(!topright_avail){
2984 tr= ptr[3 - linesize]*0x01010101;
2985 topright= (uint8_t*) &tr;
2986 }else if(i==5 && h->deblocking_filter){
2987 tr= *(uint32_t*)h->top_borders[h->mb_aff_frame ? IS_INTERLACED(mb_type) ? bottom : 1 : 0][mb_x+1];
2988 topright= (uint8_t*) &tr;
2990 topright= ptr + 4 - linesize;
2994 h->pred4x4[ dir ](ptr, topright, linesize);
2995 if(h->non_zero_count_cache[ scan8[i] ]){
2996 if(s->codec_id == CODEC_ID_H264)
2997 s->dsp.h264_idct_add(ptr, h->mb + i*16, linesize);
2999 svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
3004 h->pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
3005 if(s->codec_id == CODEC_ID_H264)
3006 h264_luma_dc_dequant_idct_c(h->mb, s->qscale);
3008 svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
3010 if(h->deblocking_filter) {
3011 if (h->mb_aff_frame) {
3013 uint8_t *pair_dest_y = s->current_picture.data[0] + ((mb_y-1) * 16* s->linesize ) + mb_x * 16;
3014 uint8_t *pair_dest_cb = s->current_picture.data[1] + ((mb_y-1) * 8 * s->uvlinesize) + mb_x * 8;
3015 uint8_t *pair_dest_cr = s->current_picture.data[2] + ((mb_y-1) * 8 * s->uvlinesize) + mb_x * 8;
3017 xchg_pair_border(h, pair_dest_y, pair_dest_cb, pair_dest_cr, s->linesize, s->uvlinesize, 0);
3021 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);
3024 }else if(s->codec_id == CODEC_ID_H264){
3025 hl_motion(h, dest_y, dest_cb, dest_cr,
3026 s->dsp.put_h264_qpel_pixels_tab, s->dsp.put_h264_chroma_pixels_tab,
3027 s->dsp.avg_h264_qpel_pixels_tab, s->dsp.avg_h264_chroma_pixels_tab,
3028 s->dsp.weight_h264_pixels_tab, s->dsp.biweight_h264_pixels_tab);
3032 if(!IS_INTRA4x4(mb_type)){
3033 if(s->codec_id == CODEC_ID_H264){
3034 for(i=0; i<16; i++){
3035 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
3036 uint8_t * const ptr= dest_y + block_offset[i];
3037 s->dsp.h264_idct_add(ptr, h->mb + i*16, linesize);
3041 for(i=0; i<16; i++){
3042 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
3043 uint8_t * const ptr= dest_y + block_offset[i];
3044 svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
3050 if(!(s->flags&CODEC_FLAG_GRAY)){
3051 chroma_dc_dequant_idct_c(h->mb + 16*16, h->chroma_qp);
3052 chroma_dc_dequant_idct_c(h->mb + 16*16+4*16, h->chroma_qp);
3053 if(s->codec_id == CODEC_ID_H264){
3054 for(i=16; i<16+4; i++){
3055 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
3056 uint8_t * const ptr= dest_cb + block_offset[i];
3057 s->dsp.h264_idct_add(ptr, h->mb + i*16, uvlinesize);
3060 for(i=20; i<20+4; i++){
3061 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
3062 uint8_t * const ptr= dest_cr + block_offset[i];
3063 s->dsp.h264_idct_add(ptr, h->mb + i*16, uvlinesize);
3067 for(i=16; i<16+4; i++){
3068 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
3069 uint8_t * const ptr= dest_cb + block_offset[i];
3070 svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
3073 for(i=20; i<20+4; i++){
3074 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
3075 uint8_t * const ptr= dest_cr + block_offset[i];
3076 svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
3082 if(h->deblocking_filter) {
3083 if (h->mb_aff_frame) {
3084 const int mb_y = s->mb_y - 1;
3085 uint8_t *pair_dest_y, *pair_dest_cb, *pair_dest_cr;
3086 const int mb_xy= mb_x + mb_y*s->mb_stride;
3087 const int mb_type_top = s->current_picture.mb_type[mb_xy];
3088 const int mb_type_bottom= s->current_picture.mb_type[mb_xy+s->mb_stride];
3089 uint8_t tmp = s->current_picture.data[1][384];
3090 if (!bottom) return;
3091 pair_dest_y = s->current_picture.data[0] + (mb_y * 16* s->linesize ) + mb_x * 16;
3092 pair_dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
3093 pair_dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
3095 backup_pair_border(h, pair_dest_y, pair_dest_cb, pair_dest_cr, s->linesize, s->uvlinesize);
3096 // TODO deblock a pair
3099 tprintf("call mbaff filter_mb mb_x:%d mb_y:%d pair_dest_y = %p, dest_y = %p\n", mb_x, mb_y, pair_dest_y, dest_y);
3100 fill_caches(h, mb_type_top, 1); //FIXME dont fill stuff which isnt used by filter_mb
3101 filter_mb(h, mb_x, mb_y, pair_dest_y, pair_dest_cb, pair_dest_cr, linesize, uvlinesize);
3102 if (tmp != s->current_picture.data[1][384]) {
3103 tprintf("modified pixel 8,1 (1)\n");
3107 tprintf("call mbaff filter_mb\n");
3108 fill_caches(h, mb_type_bottom, 1); //FIXME dont fill stuff which isnt used by filter_mb
3109 filter_mb(h, mb_x, mb_y+1, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
3110 if (tmp != s->current_picture.data[1][384]) {
3111 tprintf("modified pixel 8,1 (2)\n");
3114 tprintf("call filter_mb\n");
3115 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
3116 fill_caches(h, mb_type, 1); //FIXME dont fill stuff which isnt used by filter_mb
3117 filter_mb(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
3123 * fills the default_ref_list.
3125 static int fill_default_ref_list(H264Context *h){
3126 MpegEncContext * const s = &h->s;
3128 int smallest_poc_greater_than_current = -1;
3129 Picture sorted_short_ref[32];
3131 if(h->slice_type==B_TYPE){
3135 /* sort frame according to poc in B slice */
3136 for(out_i=0; out_i<h->short_ref_count; out_i++){
3138 int best_poc=INT_MAX;
3140 for(i=0; i<h->short_ref_count; i++){
3141 const int poc= h->short_ref[i]->poc;
3142 if(poc > limit && poc < best_poc){
3148 assert(best_i != -1);
3151 sorted_short_ref[out_i]= *h->short_ref[best_i];
3152 tprintf("sorted poc: %d->%d poc:%d fn:%d\n", best_i, out_i, sorted_short_ref[out_i].poc, sorted_short_ref[out_i].frame_num);
3153 if (-1 == smallest_poc_greater_than_current) {
3154 if (h->short_ref[best_i]->poc >= s->current_picture_ptr->poc) {
3155 smallest_poc_greater_than_current = out_i;
3161 if(s->picture_structure == PICT_FRAME){
3162 if(h->slice_type==B_TYPE){
3164 tprintf("current poc: %d, smallest_poc_greater_than_current: %d\n", s->current_picture_ptr->poc, smallest_poc_greater_than_current);
3166 // find the largest poc
3167 for(list=0; list<2; list++){
3170 int step= list ? -1 : 1;
3172 for(i=0; i<h->short_ref_count && index < h->ref_count[list]; i++, j+=step) {
3173 while(j<0 || j>= h->short_ref_count){
3175 j= smallest_poc_greater_than_current + (step>>1);
3177 if(sorted_short_ref[j].reference != 3) continue;
3178 h->default_ref_list[list][index ]= sorted_short_ref[j];
3179 h->default_ref_list[list][index++].pic_id= sorted_short_ref[j].frame_num;
3182 for(i = 0; i < 16 && index < h->ref_count[ list ]; i++){
3183 if(h->long_ref[i] == NULL) continue;
3184 if(h->long_ref[i]->reference != 3) continue;
3186 h->default_ref_list[ list ][index ]= *h->long_ref[i];
3187 h->default_ref_list[ list ][index++].pic_id= i;;
3190 if(list && (smallest_poc_greater_than_current<=0 || smallest_poc_greater_than_current>=h->short_ref_count) && (1 < index)){
3191 // swap the two first elements of L1 when
3192 // L0 and L1 are identical
3193 Picture temp= h->default_ref_list[1][0];
3194 h->default_ref_list[1][0] = h->default_ref_list[1][1];
3195 h->default_ref_list[1][0] = temp;
3198 if(index < h->ref_count[ list ])
3199 memset(&h->default_ref_list[list][index], 0, sizeof(Picture)*(h->ref_count[ list ] - index));
3203 for(i=0; i<h->short_ref_count; i++){
3204 if(h->short_ref[i]->reference != 3) continue; //FIXME refernce field shit
3205 h->default_ref_list[0][index ]= *h->short_ref[i];
3206 h->default_ref_list[0][index++].pic_id= h->short_ref[i]->frame_num;
3208 for(i = 0; i < 16; i++){
3209 if(h->long_ref[i] == NULL) continue;
3210 if(h->long_ref[i]->reference != 3) continue;
3211 h->default_ref_list[0][index ]= *h->long_ref[i];
3212 h->default_ref_list[0][index++].pic_id= i;;
3214 if(index < h->ref_count[0])
3215 memset(&h->default_ref_list[0][index], 0, sizeof(Picture)*(h->ref_count[0] - index));
3218 if(h->slice_type==B_TYPE){
3220 //FIXME second field balh
3224 for (i=0; i<h->ref_count[0]; i++) {
3225 tprintf("List0: %s fn:%d 0x%p\n", (h->default_ref_list[0][i].long_ref ? "LT" : "ST"), h->default_ref_list[0][i].pic_id, h->default_ref_list[0][i].data[0]);
3227 if(h->slice_type==B_TYPE){
3228 for (i=0; i<h->ref_count[1]; i++) {
3229 tprintf("List1: %s fn:%d 0x%p\n", (h->default_ref_list[1][i].long_ref ? "LT" : "ST"), h->default_ref_list[1][i].pic_id, h->default_ref_list[0][i].data[0]);
3236 static void print_short_term(H264Context *h);
3237 static void print_long_term(H264Context *h);
3239 static int decode_ref_pic_list_reordering(H264Context *h){
3240 MpegEncContext * const s = &h->s;
3243 print_short_term(h);
3245 if(h->slice_type==I_TYPE || h->slice_type==SI_TYPE) return 0; //FIXME move beofre func
3247 for(list=0; list<2; list++){
3248 memcpy(h->ref_list[list], h->default_ref_list[list], sizeof(Picture)*h->ref_count[list]);
3250 if(get_bits1(&s->gb)){
3251 int pred= h->curr_pic_num;
3254 for(index=0; ; index++){
3255 int reordering_of_pic_nums_idc= get_ue_golomb(&s->gb);
3258 Picture *ref = NULL;
3260 if(reordering_of_pic_nums_idc==3)
3263 if(index >= h->ref_count[list]){
3264 av_log(h->s.avctx, AV_LOG_ERROR, "reference count overflow\n");
3268 if(reordering_of_pic_nums_idc<3){
3269 if(reordering_of_pic_nums_idc<2){
3270 const int abs_diff_pic_num= get_ue_golomb(&s->gb) + 1;
3272 if(abs_diff_pic_num >= h->max_pic_num){
3273 av_log(h->s.avctx, AV_LOG_ERROR, "abs_diff_pic_num overflow\n");
3277 if(reordering_of_pic_nums_idc == 0) pred-= abs_diff_pic_num;
3278 else pred+= abs_diff_pic_num;
3279 pred &= h->max_pic_num - 1;
3281 for(i= h->short_ref_count-1; i>=0; i--){
3282 ref = h->short_ref[i];
3283 if(ref->data[0] != NULL && ref->frame_num == pred && ref->long_ref == 0) // ignore non existing pictures by testing data[0] pointer
3287 pic_id= get_ue_golomb(&s->gb); //long_term_pic_idx
3288 ref = h->long_ref[pic_id];
3292 av_log(h->s.avctx, AV_LOG_ERROR, "reference picture missing during reorder\n");
3293 memset(&h->ref_list[list][index], 0, sizeof(Picture)); //FIXME
3295 h->ref_list[list][index]= *ref;
3298 av_log(h->s.avctx, AV_LOG_ERROR, "illegal reordering_of_pic_nums_idc\n");
3304 if(h->slice_type!=B_TYPE) break;
3307 if(h->slice_type==B_TYPE && !h->direct_spatial_mv_pred)
3308 direct_dist_scale_factor(h);
3309 direct_ref_list_init(h);
3313 static int pred_weight_table(H264Context *h){
3314 MpegEncContext * const s = &h->s;
3316 int luma_def, chroma_def;
3319 h->use_weight_chroma= 0;
3320 h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
3321 h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
3322 luma_def = 1<<h->luma_log2_weight_denom;
3323 chroma_def = 1<<h->chroma_log2_weight_denom;
3325 for(list=0; list<2; list++){
3326 for(i=0; i<h->ref_count[list]; i++){
3327 int luma_weight_flag, chroma_weight_flag;
3329 luma_weight_flag= get_bits1(&s->gb);
3330 if(luma_weight_flag){
3331 h->luma_weight[list][i]= get_se_golomb(&s->gb);
3332 h->luma_offset[list][i]= get_se_golomb(&s->gb);
3333 if( h->luma_weight[list][i] != luma_def
3334 || h->luma_offset[list][i] != 0)
3337 h->luma_weight[list][i]= luma_def;
3338 h->luma_offset[list][i]= 0;
3341 chroma_weight_flag= get_bits1(&s->gb);
3342 if(chroma_weight_flag){
3345 h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);
3346 h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);
3347 if( h->chroma_weight[list][i][j] != chroma_def
3348 || h->chroma_offset[list][i][j] != 0)
3349 h->use_weight_chroma= 1;
3354 h->chroma_weight[list][i][j]= chroma_def;
3355 h->chroma_offset[list][i][j]= 0;
3359 if(h->slice_type != B_TYPE) break;
3361 h->use_weight= h->use_weight || h->use_weight_chroma;
3365 static void implicit_weight_table(H264Context *h){
3366 MpegEncContext * const s = &h->s;
3368 int cur_poc = s->current_picture_ptr->poc;
3370 if( h->ref_count[0] == 1 && h->ref_count[1] == 1
3371 && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
3373 h->use_weight_chroma= 0;
3378 h->use_weight_chroma= 2;
3379 h->luma_log2_weight_denom= 5;
3380 h->chroma_log2_weight_denom= 5;
3383 for(ref0=0; ref0 < h->ref_count[0]; ref0++){
3384 int poc0 = h->ref_list[0][ref0].poc;
3385 for(ref1=0; ref1 < h->ref_count[1]; ref1++){
3386 int poc1 = h->ref_list[1][ref1].poc;
3387 int td = clip(poc1 - poc0, -128, 127);
3389 int tb = clip(cur_poc - poc0, -128, 127);
3390 int tx = (16384 + (ABS(td) >> 1)) / td;
3391 int dist_scale_factor = clip((tb*tx + 32) >> 6, -1024, 1023) >> 2;
3392 if(dist_scale_factor < -64 || dist_scale_factor > 128)
3393 h->implicit_weight[ref0][ref1] = 32;
3395 h->implicit_weight[ref0][ref1] = 64 - dist_scale_factor;
3397 h->implicit_weight[ref0][ref1] = 32;
3402 static inline void unreference_pic(H264Context *h, Picture *pic){
3405 if(pic == h->delayed_output_pic)
3408 for(i = 0; h->delayed_pic[i]; i++)
3409 if(pic == h->delayed_pic[i]){
3417 * instantaneous decoder refresh.
3419 static void idr(H264Context *h){
3422 for(i=0; i<16; i++){
3423 if (h->long_ref[i] != NULL) {
3424 unreference_pic(h, h->long_ref[i]);
3425 h->long_ref[i]= NULL;
3428 h->long_ref_count=0;
3430 for(i=0; i<h->short_ref_count; i++){
3431 unreference_pic(h, h->short_ref[i]);
3432 h->short_ref[i]= NULL;
3434 h->short_ref_count=0;
3439 * @return the removed picture or NULL if an error occures
3441 static Picture * remove_short(H264Context *h, int frame_num){
3442 MpegEncContext * const s = &h->s;
3445 if(s->avctx->debug&FF_DEBUG_MMCO)
3446 av_log(h->s.avctx, AV_LOG_DEBUG, "remove short %d count %d\n", frame_num, h->short_ref_count);
3448 for(i=0; i<h->short_ref_count; i++){
3449 Picture *pic= h->short_ref[i];
3450 if(s->avctx->debug&FF_DEBUG_MMCO)
3451 av_log(h->s.avctx, AV_LOG_DEBUG, "%d %d %p\n", i, pic->frame_num, pic);
3452 if(pic->frame_num == frame_num){
3453 h->short_ref[i]= NULL;
3454 memmove(&h->short_ref[i], &h->short_ref[i+1], (h->short_ref_count - i - 1)*sizeof(Picture*));
3455 h->short_ref_count--;
3464 * @return the removed picture or NULL if an error occures
3466 static Picture * remove_long(H264Context *h, int i){
3469 pic= h->long_ref[i];
3470 h->long_ref[i]= NULL;
3471 if(pic) h->long_ref_count--;
3477 * print short term list
3479 static void print_short_term(H264Context *h) {
3481 if(h->s.avctx->debug&FF_DEBUG_MMCO) {
3482 av_log(h->s.avctx, AV_LOG_DEBUG, "short term list:\n");
3483 for(i=0; i<h->short_ref_count; i++){
3484 Picture *pic= h->short_ref[i];
3485 av_log(h->s.avctx, AV_LOG_DEBUG, "%d fn:%d poc:%d %p\n", i, pic->frame_num, pic->poc, pic->data[0]);
3491 * print long term list
3493 static void print_long_term(H264Context *h) {
3495 if(h->s.avctx->debug&FF_DEBUG_MMCO) {
3496 av_log(h->s.avctx, AV_LOG_DEBUG, "long term list:\n");
3497 for(i = 0; i < 16; i++){
3498 Picture *pic= h->long_ref[i];
3500 av_log(h->s.avctx, AV_LOG_DEBUG, "%d fn:%d poc:%d %p\n", i, pic->frame_num, pic->poc, pic->data[0]);
3507 * Executes the reference picture marking (memory management control operations).
3509 static int execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count){
3510 MpegEncContext * const s = &h->s;
3512 int current_is_long=0;
3515 if((s->avctx->debug&FF_DEBUG_MMCO) && mmco_count==0)
3516 av_log(h->s.avctx, AV_LOG_DEBUG, "no mmco here\n");
3518 for(i=0; i<mmco_count; i++){
3519 if(s->avctx->debug&FF_DEBUG_MMCO)
3520 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);
3522 switch(mmco[i].opcode){
3523 case MMCO_SHORT2UNUSED:
3524 pic= remove_short(h, mmco[i].short_frame_num);
3525 if(pic==NULL) return -1;
3526 unreference_pic(h, pic);
3528 case MMCO_SHORT2LONG:
3529 pic= remove_long(h, mmco[i].long_index);
3530 if(pic) unreference_pic(h, pic);
3532 h->long_ref[ mmco[i].long_index ]= remove_short(h, mmco[i].short_frame_num);
3533 h->long_ref[ mmco[i].long_index ]->long_ref=1;
3534 h->long_ref_count++;
3536 case MMCO_LONG2UNUSED:
3537 pic= remove_long(h, mmco[i].long_index);
3538 if(pic==NULL) return -1;
3539 unreference_pic(h, pic);
3542 pic= remove_long(h, mmco[i].long_index);
3543 if(pic) unreference_pic(h, pic);
3545 h->long_ref[ mmco[i].long_index ]= s->current_picture_ptr;
3546 h->long_ref[ mmco[i].long_index ]->long_ref=1;
3547 h->long_ref_count++;
3551 case MMCO_SET_MAX_LONG:
3552 assert(mmco[i].long_index <= 16);
3553 // just remove the long term which index is greater than new max
3554 for(j = mmco[i].long_index; j<16; j++){
3555 pic = remove_long(h, j);
3556 if (pic) unreference_pic(h, pic);
3560 while(h->short_ref_count){
3561 pic= remove_short(h, h->short_ref[0]->frame_num);
3562 unreference_pic(h, pic);
3564 for(j = 0; j < 16; j++) {
3565 pic= remove_long(h, j);
3566 if(pic) unreference_pic(h, pic);
3573 if(!current_is_long){
3574 pic= remove_short(h, s->current_picture_ptr->frame_num);
3576 unreference_pic(h, pic);
3577 av_log(h->s.avctx, AV_LOG_ERROR, "illegal short term buffer state detected\n");
3580 if(h->short_ref_count)
3581 memmove(&h->short_ref[1], &h->short_ref[0], h->short_ref_count*sizeof(Picture*));
3583 h->short_ref[0]= s->current_picture_ptr;
3584 h->short_ref[0]->long_ref=0;
3585 h->short_ref_count++;
3588 print_short_term(h);
3593 static int decode_ref_pic_marking(H264Context *h){
3594 MpegEncContext * const s = &h->s;
3597 if(h->nal_unit_type == NAL_IDR_SLICE){ //FIXME fields
3598 s->broken_link= get_bits1(&s->gb) -1;
3599 h->mmco[0].long_index= get_bits1(&s->gb) - 1; // current_long_term_idx
3600 if(h->mmco[0].long_index == -1)
3603 h->mmco[0].opcode= MMCO_LONG;
3607 if(get_bits1(&s->gb)){ // adaptive_ref_pic_marking_mode_flag
3608 for(i= 0; i<MAX_MMCO_COUNT; i++) {
3609 MMCOOpcode opcode= get_ue_golomb(&s->gb);;
3611 h->mmco[i].opcode= opcode;
3612 if(opcode==MMCO_SHORT2UNUSED || opcode==MMCO_SHORT2LONG){
3613 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
3614 /* if(h->mmco[i].short_frame_num >= h->short_ref_count || h->short_ref[ h->mmco[i].short_frame_num ] == NULL){
3615 fprintf(stderr, "illegal short ref in memory management control operation %d\n", mmco);
3619 if(opcode==MMCO_SHORT2LONG || opcode==MMCO_LONG2UNUSED || opcode==MMCO_LONG || opcode==MMCO_SET_MAX_LONG){
3620 h->mmco[i].long_index= get_ue_golomb(&s->gb);
3621 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){
3622 av_log(h->s.avctx, AV_LOG_ERROR, "illegal long ref in memory management control operation %d\n", opcode);
3627 if(opcode > MMCO_LONG){
3628 av_log(h->s.avctx, AV_LOG_ERROR, "illegal memory management control operation %d\n", opcode);
3631 if(opcode == MMCO_END)
3636 assert(h->long_ref_count + h->short_ref_count <= h->sps.ref_frame_count);
3638 if(h->long_ref_count + h->short_ref_count == h->sps.ref_frame_count){ //FIXME fields
3639 h->mmco[0].opcode= MMCO_SHORT2UNUSED;
3640 h->mmco[0].short_frame_num= h->short_ref[ h->short_ref_count - 1 ]->frame_num;
3650 static int init_poc(H264Context *h){
3651 MpegEncContext * const s = &h->s;
3652 const int max_frame_num= 1<<h->sps.log2_max_frame_num;
3655 if(h->nal_unit_type == NAL_IDR_SLICE){
3656 h->frame_num_offset= 0;
3658 if(h->frame_num < h->prev_frame_num)
3659 h->frame_num_offset= h->prev_frame_num_offset + max_frame_num;
3661 h->frame_num_offset= h->prev_frame_num_offset;
3664 if(h->sps.poc_type==0){
3665 const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
3667 if (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
3668 h->poc_msb = h->prev_poc_msb + max_poc_lsb;
3669 else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
3670 h->poc_msb = h->prev_poc_msb - max_poc_lsb;
3672 h->poc_msb = h->prev_poc_msb;
3673 //printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
3675 field_poc[1] = h->poc_msb + h->poc_lsb;
3676 if(s->picture_structure == PICT_FRAME)
3677 field_poc[1] += h->delta_poc_bottom;
3678 }else if(h->sps.poc_type==1){
3679 int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
3682 if(h->sps.poc_cycle_length != 0)
3683 abs_frame_num = h->frame_num_offset + h->frame_num;
3687 if(h->nal_ref_idc==0 && abs_frame_num > 0)
3690 expected_delta_per_poc_cycle = 0;
3691 for(i=0; i < h->sps.poc_cycle_length; i++)
3692 expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
3694 if(abs_frame_num > 0){
3695 int poc_cycle_cnt = (abs_frame_num - 1) / h->sps.poc_cycle_length;
3696 int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
3698 expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
3699 for(i = 0; i <= frame_num_in_poc_cycle; i++)
3700 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
3704 if(h->nal_ref_idc == 0)
3705 expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
3707 field_poc[0] = expectedpoc + h->delta_poc[0];
3708 field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
3710 if(s->picture_structure == PICT_FRAME)
3711 field_poc[1] += h->delta_poc[1];
3714 if(h->nal_unit_type == NAL_IDR_SLICE){
3717 if(h->nal_ref_idc) poc= 2*(h->frame_num_offset + h->frame_num);
3718 else poc= 2*(h->frame_num_offset + h->frame_num) - 1;
3724 if(s->picture_structure != PICT_BOTTOM_FIELD)
3725 s->current_picture_ptr->field_poc[0]= field_poc[0];
3726 if(s->picture_structure != PICT_TOP_FIELD)
3727 s->current_picture_ptr->field_poc[1]= field_poc[1];
3728 if(s->picture_structure == PICT_FRAME) // FIXME field pix?
3729 s->current_picture_ptr->poc= FFMIN(field_poc[0], field_poc[1]);
3735 * decodes a slice header.
3736 * this will allso call MPV_common_init() and frame_start() as needed
3738 static int decode_slice_header(H264Context *h){
3739 MpegEncContext * const s = &h->s;
3740 int first_mb_in_slice, pps_id;
3741 int num_ref_idx_active_override_flag;
3742 static const uint8_t slice_type_map[5]= {P_TYPE, B_TYPE, I_TYPE, SP_TYPE, SI_TYPE};
3744 int default_ref_list_done = 0;
3746 s->current_picture.reference= h->nal_ref_idc != 0;
3747 s->dropable= h->nal_ref_idc == 0;
3749 first_mb_in_slice= get_ue_golomb(&s->gb);
3751 slice_type= get_ue_golomb(&s->gb);
3753 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);
3758 h->slice_type_fixed=1;
3760 h->slice_type_fixed=0;
3762 slice_type= slice_type_map[ slice_type ];
3763 if (slice_type == I_TYPE
3764 || (h->slice_num != 0 && slice_type == h->slice_type) ) {
3765 default_ref_list_done = 1;
3767 h->slice_type= slice_type;
3769 s->pict_type= h->slice_type; // to make a few old func happy, its wrong though
3771 pps_id= get_ue_golomb(&s->gb);
3773 av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
3776 h->pps= h->pps_buffer[pps_id];
3777 if(h->pps.slice_group_count == 0){
3778 av_log(h->s.avctx, AV_LOG_ERROR, "non existing PPS referenced\n");
3782 h->sps= h->sps_buffer[ h->pps.sps_id ];
3783 if(h->sps.log2_max_frame_num == 0){
3784 av_log(h->s.avctx, AV_LOG_ERROR, "non existing SPS referenced\n");
3788 s->mb_width= h->sps.mb_width;
3789 s->mb_height= h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
3791 h->b_stride= s->mb_width*4 + 1;
3792 h->b8_stride= s->mb_width*2 + 1;
3794 s->width = 16*s->mb_width - 2*(h->sps.crop_left + h->sps.crop_right );
3795 if(h->sps.frame_mbs_only_flag)
3796 s->height= 16*s->mb_height - 2*(h->sps.crop_top + h->sps.crop_bottom);
3798 s->height= 16*s->mb_height - 4*(h->sps.crop_top + h->sps.crop_bottom); //FIXME recheck
3800 if (s->context_initialized
3801 && ( s->width != s->avctx->width || s->height != s->avctx->height)) {
3805 if (!s->context_initialized) {
3806 if (MPV_common_init(s) < 0)
3811 s->avctx->width = s->width;
3812 s->avctx->height = s->height;
3813 s->avctx->sample_aspect_ratio= h->sps.sar;
3814 if(!s->avctx->sample_aspect_ratio.den)
3815 s->avctx->sample_aspect_ratio.den = 1;
3817 if(h->sps.timing_info_present_flag && h->sps.fixed_frame_rate_flag){
3818 s->avctx->frame_rate = h->sps.time_scale;
3819 s->avctx->frame_rate_base = h->sps.num_units_in_tick;
3823 if(h->slice_num == 0){
3827 s->current_picture_ptr->frame_num= //FIXME frame_num cleanup
3828 h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
3830 h->mb_aff_frame = 0;
3831 if(h->sps.frame_mbs_only_flag){
3832 s->picture_structure= PICT_FRAME;
3834 if(get_bits1(&s->gb)) { //field_pic_flag
3835 s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
3837 s->picture_structure= PICT_FRAME;
3838 first_mb_in_slice <<= 1;
3839 h->mb_aff_frame = h->sps.mb_aff;
3843 s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
3844 s->resync_mb_y = s->mb_y = first_mb_in_slice / s->mb_width;
3846 if(s->picture_structure==PICT_FRAME){
3847 h->curr_pic_num= h->frame_num;
3848 h->max_pic_num= 1<< h->sps.log2_max_frame_num;
3850 h->curr_pic_num= 2*h->frame_num;
3851 h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
3854 if(h->nal_unit_type == NAL_IDR_SLICE){
3855 get_ue_golomb(&s->gb); /* idr_pic_id */
3858 if(h->sps.poc_type==0){
3859 h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
3861 if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
3862 h->delta_poc_bottom= get_se_golomb(&s->gb);
3866 if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
3867 h->delta_poc[0]= get_se_golomb(&s->gb);
3869 if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
3870 h->delta_poc[1]= get_se_golomb(&s->gb);
3875 if(h->pps.redundant_pic_cnt_present){
3876 h->redundant_pic_count= get_ue_golomb(&s->gb);
3879 //set defaults, might be overriden a few line later
3880 h->ref_count[0]= h->pps.ref_count[0];
3881 h->ref_count[1]= h->pps.ref_count[1];
3883 if(h->slice_type == P_TYPE || h->slice_type == SP_TYPE || h->slice_type == B_TYPE){
3884 if(h->slice_type == B_TYPE){
3885 h->direct_spatial_mv_pred= get_bits1(&s->gb);
3887 num_ref_idx_active_override_flag= get_bits1(&s->gb);
3889 if(num_ref_idx_active_override_flag){
3890 h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
3891 if(h->slice_type==B_TYPE)
3892 h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
3894 if(h->ref_count[0] > 32 || h->ref_count[1] > 32){
3895 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
3901 if(!default_ref_list_done){
3902 fill_default_ref_list(h);
3905 decode_ref_pic_list_reordering(h);
3907 if( (h->pps.weighted_pred && (h->slice_type == P_TYPE || h->slice_type == SP_TYPE ))
3908 || (h->pps.weighted_bipred_idc==1 && h->slice_type==B_TYPE ) )
3909 pred_weight_table(h);
3910 else if(h->pps.weighted_bipred_idc==2 && h->slice_type==B_TYPE)
3911 implicit_weight_table(h);
3915 if(s->current_picture.reference)
3916 decode_ref_pic_marking(h);
3918 if( h->slice_type != I_TYPE && h->slice_type != SI_TYPE && h->pps.cabac )
3919 h->cabac_init_idc = get_ue_golomb(&s->gb);
3921 h->last_qscale_diff = 0;
3922 s->qscale = h->pps.init_qp + get_se_golomb(&s->gb);
3923 if(s->qscale<0 || s->qscale>51){
3924 av_log(s->avctx, AV_LOG_ERROR, "QP %d out of range\n", s->qscale);
3927 h->chroma_qp = get_chroma_qp(h->pps.chroma_qp_index_offset, s->qscale);
3928 //FIXME qscale / qp ... stuff
3929 if(h->slice_type == SP_TYPE){
3930 get_bits1(&s->gb); /* sp_for_switch_flag */
3932 if(h->slice_type==SP_TYPE || h->slice_type == SI_TYPE){
3933 get_se_golomb(&s->gb); /* slice_qs_delta */
3936 h->deblocking_filter = 1;
3937 h->slice_alpha_c0_offset = 0;
3938 h->slice_beta_offset = 0;
3939 if( h->pps.deblocking_filter_parameters_present ) {
3940 h->deblocking_filter= get_ue_golomb(&s->gb);
3941 if(h->deblocking_filter < 2)
3942 h->deblocking_filter^= 1; // 1<->0
3944 if( h->deblocking_filter ) {
3945 h->slice_alpha_c0_offset = get_se_golomb(&s->gb) << 1;
3946 h->slice_beta_offset = get_se_golomb(&s->gb) << 1;
3951 if( h->pps.num_slice_groups > 1 && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
3952 slice_group_change_cycle= get_bits(&s->gb, ?);
3957 if(s->avctx->debug&FF_DEBUG_PICT_INFO){
3958 av_log(h->s.avctx, AV_LOG_DEBUG, "slice:%d %s mb:%d %c pps:%d frame:%d poc:%d/%d ref:%d/%d qp:%d loop:%d weight:%d%s\n",
3960 (s->picture_structure==PICT_FRAME ? "F" : s->picture_structure==PICT_TOP_FIELD ? "T" : "B"),
3962 av_get_pict_type_char(h->slice_type),
3963 pps_id, h->frame_num,
3964 s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
3965 h->ref_count[0], h->ref_count[1],
3967 h->deblocking_filter,
3969 h->use_weight==1 && h->use_weight_chroma ? "c" : ""
3979 static inline int get_level_prefix(GetBitContext *gb){
3983 OPEN_READER(re, gb);
3984 UPDATE_CACHE(re, gb);
3985 buf=GET_CACHE(re, gb);
3987 log= 32 - av_log2(buf);
3989 print_bin(buf>>(32-log), log);
3990 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__);
3993 LAST_SKIP_BITS(re, gb, log);
3994 CLOSE_READER(re, gb);
4000 * decodes a residual block.
4001 * @param n block index
4002 * @param scantable scantable
4003 * @param max_coeff number of coefficients in the block
4004 * @return <0 if an error occured
4006 static int decode_residual(H264Context *h, GetBitContext *gb, DCTELEM *block, int n, const uint8_t *scantable, int qp, int max_coeff){
4007 MpegEncContext * const s = &h->s;
4008 const uint16_t *qmul= dequant_coeff[qp];
4009 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};
4010 int level[16], run[16];
4011 int suffix_length, zeros_left, coeff_num, coeff_token, total_coeff, i, trailing_ones;
4013 //FIXME put trailing_onex into the context
4015 if(n == CHROMA_DC_BLOCK_INDEX){
4016 coeff_token= get_vlc2(gb, chroma_dc_coeff_token_vlc.table, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 1);
4017 total_coeff= coeff_token>>2;
4019 if(n == LUMA_DC_BLOCK_INDEX){
4020 total_coeff= pred_non_zero_count(h, 0);
4021 coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
4022 total_coeff= coeff_token>>2;
4024 total_coeff= pred_non_zero_count(h, n);
4025 coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
4026 total_coeff= coeff_token>>2;
4027 h->non_zero_count_cache[ scan8[n] ]= total_coeff;
4031 //FIXME set last_non_zero?
4036 trailing_ones= coeff_token&3;
4037 tprintf("trailing:%d, total:%d\n", trailing_ones, total_coeff);
4038 assert(total_coeff<=16);
4040 for(i=0; i<trailing_ones; i++){
4041 level[i]= 1 - 2*get_bits1(gb);
4044 suffix_length= total_coeff > 10 && trailing_ones < 3;
4046 for(; i<total_coeff; i++){
4047 const int prefix= get_level_prefix(gb);
4048 int level_code, mask;
4050 if(prefix<14){ //FIXME try to build a large unified VLC table for all this
4052 level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
4054 level_code= (prefix<<suffix_length); //part
4055 }else if(prefix==14){
4057 level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
4059 level_code= prefix + get_bits(gb, 4); //part
4060 }else if(prefix==15){
4061 level_code= (prefix<<suffix_length) + get_bits(gb, 12); //part
4062 if(suffix_length==0) level_code+=15; //FIXME doesnt make (much)sense
4064 av_log(h->s.avctx, AV_LOG_ERROR, "prefix too large at %d %d\n", s->mb_x, s->mb_y);
4068 if(i==trailing_ones && i<3) level_code+= 2; //FIXME split first iteration
4070 mask= -(level_code&1);
4071 level[i]= (((2+level_code)>>1) ^ mask) - mask;
4073 if(suffix_length==0) suffix_length=1; //FIXME split first iteration
4076 if(ABS(level[i]) > (3<<(suffix_length-1)) && suffix_length<6) suffix_length++;
4078 if((2+level_code)>>1) > (3<<(suffix_length-1)) && suffix_length<6) suffix_length++;
4079 /* ? == prefix > 2 or sth */
4081 tprintf("level: %d suffix_length:%d\n", level[i], suffix_length);
4084 if(total_coeff == max_coeff)
4087 if(n == CHROMA_DC_BLOCK_INDEX)
4088 zeros_left= get_vlc2(gb, chroma_dc_total_zeros_vlc[ total_coeff-1 ].table, CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 1);
4090 zeros_left= get_vlc2(gb, total_zeros_vlc[ total_coeff-1 ].table, TOTAL_ZEROS_VLC_BITS, 1);
4093 for(i=0; i<total_coeff-1; i++){
4096 else if(zeros_left < 7){
4097 run[i]= get_vlc2(gb, run_vlc[zeros_left-1].table, RUN_VLC_BITS, 1);
4099 run[i]= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2);
4101 zeros_left -= run[i];
4105 av_log(h->s.avctx, AV_LOG_ERROR, "negative number of zero coeffs at %d %d\n", s->mb_x, s->mb_y);
4109 for(; i<total_coeff-1; i++){
4117 for(i=total_coeff-1; i>=0; i--){ //FIXME merge into rundecode?
4120 coeff_num += run[i] + 1; //FIXME add 1 earlier ?
4121 j= scantable[ coeff_num ];
4126 for(i=total_coeff-1; i>=0; i--){ //FIXME merge into rundecode?
4129 coeff_num += run[i] + 1; //FIXME add 1 earlier ?
4130 j= scantable[ coeff_num ];
4132 block[j]= level[i] * qmul[j];
4133 // printf("%d %d ", block[j], qmul[j]);
4140 * decodes a P_SKIP or B_SKIP macroblock
4142 static void decode_mb_skip(H264Context *h){
4143 MpegEncContext * const s = &h->s;
4144 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
4147 memset(h->non_zero_count[mb_xy], 0, 16);
4148 memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui
4150 if(h->mb_aff_frame && s->mb_skip_run==0 && (s->mb_y&1)==0){
4151 h->mb_field_decoding_flag= get_bits1(&s->gb);
4153 if(h->mb_field_decoding_flag)
4154 mb_type|= MB_TYPE_INTERLACED;
4156 if( h->slice_type == B_TYPE )
4158 // just for fill_caches. pred_direct_motion will set the real mb_type
4159 mb_type= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2|MB_TYPE_SKIP;
4161 fill_caches(h, mb_type, 0); //FIXME check what is needed and what not ...
4162 pred_direct_motion(h, &mb_type);
4164 fill_rectangle(h->mvd_cache[0][scan8[0]], 4, 4, 8, 0, 4);
4165 fill_rectangle(h->mvd_cache[1][scan8[0]], 4, 4, 8, 0, 4);
4171 mb_type= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0|MB_TYPE_SKIP;
4173 fill_caches(h, mb_type, 0); //FIXME check what is needed and what not ...
4174 pred_pskip_motion(h, &mx, &my);
4175 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
4176 fill_rectangle( h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4);
4178 fill_rectangle(h->mvd_cache[0][scan8[0]], 4, 4, 8, 0, 4);
4181 write_back_motion(h, mb_type);
4182 s->current_picture.mb_type[mb_xy]= mb_type|MB_TYPE_SKIP;
4183 s->current_picture.qscale_table[mb_xy]= s->qscale;
4184 h->slice_table[ mb_xy ]= h->slice_num;
4185 h->prev_mb_skiped= 1;
4189 * decodes a macroblock
4190 * @returns 0 if ok, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
4192 static int decode_mb_cavlc(H264Context *h){
4193 MpegEncContext * const s = &h->s;
4194 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
4195 int mb_type, partition_count, cbp;
4197 s->dsp.clear_blocks(h->mb); //FIXME avoid if allready clear (move after skip handlong?
4199 tprintf("pic:%d mb:%d/%d\n", h->frame_num, s->mb_x, s->mb_y);
4200 cbp = 0; /* avoid warning. FIXME: find a solution without slowing
4202 if(h->slice_type != I_TYPE && h->slice_type != SI_TYPE){
4203 if(s->mb_skip_run==-1)
4204 s->mb_skip_run= get_ue_golomb(&s->gb);
4206 if (s->mb_skip_run--) {
4211 if(h->mb_aff_frame){
4212 if ( ((s->mb_y&1) == 0) || h->prev_mb_skiped)
4213 h->mb_field_decoding_flag = get_bits1(&s->gb);
4215 h->mb_field_decoding_flag= (s->picture_structure!=PICT_FRAME);
4217 h->prev_mb_skiped= 0;
4219 mb_type= get_ue_golomb(&s->gb);
4220 if(h->slice_type == B_TYPE){
4222 partition_count= b_mb_type_info[mb_type].partition_count;
4223 mb_type= b_mb_type_info[mb_type].type;
4226 goto decode_intra_mb;
4228 }else if(h->slice_type == P_TYPE /*|| h->slice_type == SP_TYPE */){
4230 partition_count= p_mb_type_info[mb_type].partition_count;
4231 mb_type= p_mb_type_info[mb_type].type;
4234 goto decode_intra_mb;
4237 assert(h->slice_type == I_TYPE);
4240 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);
4244 cbp= i_mb_type_info[mb_type].cbp;
4245 h->intra16x16_pred_mode= i_mb_type_info[mb_type].pred_mode;
4246 mb_type= i_mb_type_info[mb_type].type;
4249 if(h->mb_field_decoding_flag)
4250 mb_type |= MB_TYPE_INTERLACED;
4252 s->current_picture.mb_type[mb_xy]= mb_type;
4253 h->slice_table[ mb_xy ]= h->slice_num;
4255 if(IS_INTRA_PCM(mb_type)){
4258 // we assume these blocks are very rare so we dont optimize it
4259 align_get_bits(&s->gb);
4261 // The pixels are stored in the same order as levels in h->mb array.
4262 for(y=0; y<16; y++){
4263 const int index= 4*(y&3) + 32*((y>>2)&1) + 128*(y>>3);
4264 for(x=0; x<16; x++){
4265 tprintf("LUMA ICPM LEVEL (%3d)\n", show_bits(&s->gb, 8));
4266 h->mb[index + (x&3) + 16*((x>>2)&1) + 64*(x>>3)]= get_bits(&s->gb, 8);
4270 const int index= 256 + 4*(y&3) + 32*(y>>2);
4272 tprintf("CHROMA U ICPM LEVEL (%3d)\n", show_bits(&s->gb, 8));
4273 h->mb[index + (x&3) + 16*(x>>2)]= get_bits(&s->gb, 8);
4277 const int index= 256 + 64 + 4*(y&3) + 32*(y>>2);
4279 tprintf("CHROMA V ICPM LEVEL (%3d)\n", show_bits(&s->gb, 8));
4280 h->mb[index + (x&3) + 16*(x>>2)]= get_bits(&s->gb, 8);
4284 // In deblocking, the quantiser is 0
4285 s->current_picture.qscale_table[mb_xy]= 0;
4286 h->chroma_qp = get_chroma_qp(h->pps.chroma_qp_index_offset, 0);
4287 // All coeffs are presents
4288 memset(h->non_zero_count[mb_xy], 16, 16);
4293 fill_caches(h, mb_type, 0);
4296 if(IS_INTRA(mb_type)){
4297 // init_top_left_availability(h);
4298 if(IS_INTRA4x4(mb_type)){
4301 // fill_intra4x4_pred_table(h);
4302 for(i=0; i<16; i++){
4303 const int mode_coded= !get_bits1(&s->gb);
4304 const int predicted_mode= pred_intra_mode(h, i);
4308 const int rem_mode= get_bits(&s->gb, 3);
4309 if(rem_mode<predicted_mode)
4314 mode= predicted_mode;
4317 h->intra4x4_pred_mode_cache[ scan8[i] ] = mode;
4319 write_back_intra_pred_mode(h);
4320 if( check_intra4x4_pred_mode(h) < 0)
4323 h->intra16x16_pred_mode= check_intra_pred_mode(h, h->intra16x16_pred_mode);
4324 if(h->intra16x16_pred_mode < 0)
4327 h->chroma_pred_mode= get_ue_golomb(&s->gb);
4329 h->chroma_pred_mode= check_intra_pred_mode(h, h->chroma_pred_mode);
4330 if(h->chroma_pred_mode < 0)
4332 }else if(partition_count==4){
4333 int i, j, sub_partition_count[4], list, ref[2][4];
4335 if(h->slice_type == B_TYPE){
4337 h->sub_mb_type[i]= get_ue_golomb(&s->gb);
4338 if(h->sub_mb_type[i] >=13){
4339 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);
4342 sub_partition_count[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
4343 h->sub_mb_type[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].type;
4345 if( IS_DIRECT(h->sub_mb_type[0]) || IS_DIRECT(h->sub_mb_type[1])
4346 || IS_DIRECT(h->sub_mb_type[2]) || IS_DIRECT(h->sub_mb_type[3]))
4347 pred_direct_motion(h, &mb_type);
4349 assert(h->slice_type == P_TYPE || h->slice_type == SP_TYPE); //FIXME SP correct ?
4351 h->sub_mb_type[i]= get_ue_golomb(&s->gb);
4352 if(h->sub_mb_type[i] >=4){
4353 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);
4356 sub_partition_count[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
4357 h->sub_mb_type[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].type;
4361 for(list=0; list<2; list++){
4362 const int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list];
4363 if(ref_count == 0) continue;
4365 if(IS_DIRECT(h->sub_mb_type[i])) continue;
4366 if(IS_DIR(h->sub_mb_type[i], 0, list)){
4367 ref[list][i] = get_te0_golomb(&s->gb, ref_count); //FIXME init to 0 before and skip?
4375 for(list=0; list<2; list++){
4376 const int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list];
4377 if(ref_count == 0) continue;
4380 if(IS_DIRECT(h->sub_mb_type[i])) continue;
4381 h->ref_cache[list][ scan8[4*i] ]=h->ref_cache[list][ scan8[4*i]+1 ]=
4382 h->ref_cache[list][ scan8[4*i]+8 ]=h->ref_cache[list][ scan8[4*i]+9 ]= ref[list][i];
4384 if(IS_DIR(h->sub_mb_type[i], 0, list)){
4385 const int sub_mb_type= h->sub_mb_type[i];
4386 const int block_width= (sub_mb_type & (MB_TYPE_16x16|MB_TYPE_16x8)) ? 2 : 1;
4387 for(j=0; j<sub_partition_count[i]; j++){
4389 const int index= 4*i + block_width*j;
4390 int16_t (* mv_cache)[2]= &h->mv_cache[list][ scan8[index] ];
4391 pred_motion(h, index, block_width, list, h->ref_cache[list][ scan8[index] ], &mx, &my);
4392 mx += get_se_golomb(&s->gb);
4393 my += get_se_golomb(&s->gb);
4394 tprintf("final mv:%d %d\n", mx, my);
4396 if(IS_SUB_8X8(sub_mb_type)){
4397 mv_cache[ 0 ][0]= mv_cache[ 1 ][0]=
4398 mv_cache[ 8 ][0]= mv_cache[ 9 ][0]= mx;
4399 mv_cache[ 0 ][1]= mv_cache[ 1 ][1]=
4400 mv_cache[ 8 ][1]= mv_cache[ 9 ][1]= my;
4401 }else if(IS_SUB_8X4(sub_mb_type)){
4402 mv_cache[ 0 ][0]= mv_cache[ 1 ][0]= mx;
4403 mv_cache[ 0 ][1]= mv_cache[ 1 ][1]= my;
4404 }else if(IS_SUB_4X8(sub_mb_type)){
4405 mv_cache[ 0 ][0]= mv_cache[ 8 ][0]= mx;
4406 mv_cache[ 0 ][1]= mv_cache[ 8 ][1]= my;
4408 assert(IS_SUB_4X4(sub_mb_type));
4409 mv_cache[ 0 ][0]= mx;
4410 mv_cache[ 0 ][1]= my;
4414 uint32_t *p= (uint32_t *)&h->mv_cache[list][ scan8[4*i] ][0];
4420 }else if(IS_DIRECT(mb_type)){
4421 pred_direct_motion(h, &mb_type);
4422 s->current_picture.mb_type[mb_xy]= mb_type;
4424 int list, mx, my, i;
4425 //FIXME we should set ref_idx_l? to 0 if we use that later ...
4426 if(IS_16X16(mb_type)){
4427 for(list=0; list<2; list++){
4428 if(h->ref_count[list]>0){
4429 if(IS_DIR(mb_type, 0, list)){
4430 const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
4431 fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, val, 1);
4433 fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, (LIST_NOT_USED&0xFF), 1);
4436 for(list=0; list<2; list++){
4437 if(IS_DIR(mb_type, 0, list)){
4438 pred_motion(h, 0, 4, list, h->ref_cache[list][ scan8[0] ], &mx, &my);
4439 mx += get_se_golomb(&s->gb);
4440 my += get_se_golomb(&s->gb);
4441 tprintf("final mv:%d %d\n", mx, my);
4443 fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, pack16to32(mx,my), 4);
4445 fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, 0, 4);
4448 else if(IS_16X8(mb_type)){
4449 for(list=0; list<2; list++){
4450 if(h->ref_count[list]>0){
4452 if(IS_DIR(mb_type, i, list)){
4453 const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
4454 fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, val, 1);
4456 fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, (LIST_NOT_USED&0xFF), 1);
4460 for(list=0; list<2; list++){
4462 if(IS_DIR(mb_type, i, list)){
4463 pred_16x8_motion(h, 8*i, list, h->ref_cache[list][scan8[0] + 16*i], &mx, &my);
4464 mx += get_se_golomb(&s->gb);
4465 my += get_se_golomb(&s->gb);
4466 tprintf("final mv:%d %d\n", mx, my);
4468 fill_rectangle(h->mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, pack16to32(mx,my), 4);
4470 fill_rectangle(h->mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, 0, 4);
4474 assert(IS_8X16(mb_type));
4475 for(list=0; list<2; list++){
4476 if(h->ref_count[list]>0){
4478 if(IS_DIR(mb_type, i, list)){ //FIXME optimize
4479 const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
4480 fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, val, 1);
4482 fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, (LIST_NOT_USED&0xFF), 1);
4486 for(list=0; list<2; list++){
4488 if(IS_DIR(mb_type, i, list)){
4489 pred_8x16_motion(h, i*4, list, h->ref_cache[list][ scan8[0] + 2*i ], &mx, &my);
4490 mx += get_se_golomb(&s->gb);
4491 my += get_se_golomb(&s->gb);
4492 tprintf("final mv:%d %d\n", mx, my);
4494 fill_rectangle(h->mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, pack16to32(mx,my), 4);
4496 fill_rectangle(h->mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, 0, 4);
4502 if(IS_INTER(mb_type))
4503 write_back_motion(h, mb_type);
4505 if(!IS_INTRA16x16(mb_type)){
4506 cbp= get_ue_golomb(&s->gb);
4508 av_log(h->s.avctx, AV_LOG_ERROR, "cbp too large (%d) at %d %d\n", cbp, s->mb_x, s->mb_y);
4512 if(IS_INTRA4x4(mb_type))
4513 cbp= golomb_to_intra4x4_cbp[cbp];
4515 cbp= golomb_to_inter_cbp[cbp];
4518 if(cbp || IS_INTRA16x16(mb_type)){
4519 int i8x8, i4x4, chroma_idx;
4520 int chroma_qp, dquant;
4521 GetBitContext *gb= IS_INTRA(mb_type) ? h->intra_gb_ptr : h->inter_gb_ptr;
4522 const uint8_t *scan, *dc_scan;
4524 // fill_non_zero_count_cache(h);
4526 if(IS_INTERLACED(mb_type)){
4528 dc_scan= luma_dc_field_scan;
4531 dc_scan= luma_dc_zigzag_scan;
4534 dquant= get_se_golomb(&s->gb);
4536 if( dquant > 25 || dquant < -26 ){
4537 av_log(h->s.avctx, AV_LOG_ERROR, "dquant out of range (%d) at %d %d\n", dquant, s->mb_x, s->mb_y);
4541 s->qscale += dquant;
4542 if(((unsigned)s->qscale) > 51){
4543 if(s->qscale<0) s->qscale+= 52;
4544 else s->qscale-= 52;
4547 h->chroma_qp= chroma_qp= get_chroma_qp(h->pps.chroma_qp_index_offset, s->qscale);
4548 if(IS_INTRA16x16(mb_type)){
4549 if( decode_residual(h, h->intra_gb_ptr, h->mb, LUMA_DC_BLOCK_INDEX, dc_scan, s->qscale, 16) < 0){
4550 return -1; //FIXME continue if partotioned and other retirn -1 too
4553 assert((cbp&15) == 0 || (cbp&15) == 15);
4556 for(i8x8=0; i8x8<4; i8x8++){
4557 for(i4x4=0; i4x4<4; i4x4++){
4558 const int index= i4x4 + 4*i8x8;
4559 if( decode_residual(h, h->intra_gb_ptr, h->mb + 16*index, index, scan + 1, s->qscale, 15) < 0 ){
4565 fill_rectangle(&h->non_zero_count_cache[scan8[0]], 4, 4, 8, 0, 1);
4568 for(i8x8=0; i8x8<4; i8x8++){
4569 if(cbp & (1<<i8x8)){
4570 for(i4x4=0; i4x4<4; i4x4++){
4571 const int index= i4x4 + 4*i8x8;
4573 if( decode_residual(h, gb, h->mb + 16*index, index, scan, s->qscale, 16) <0 ){
4578 uint8_t * const nnz= &h->non_zero_count_cache[ scan8[4*i8x8] ];
4579 nnz[0] = nnz[1] = nnz[8] = nnz[9] = 0;
4585 for(chroma_idx=0; chroma_idx<2; chroma_idx++)
4586 if( decode_residual(h, gb, h->mb + 256 + 16*4*chroma_idx, CHROMA_DC_BLOCK_INDEX, chroma_dc_scan, chroma_qp, 4) < 0){
4592 for(chroma_idx=0; chroma_idx<2; chroma_idx++){
4593 for(i4x4=0; i4x4<4; i4x4++){
4594 const int index= 16 + 4*chroma_idx + i4x4;
4595 if( decode_residual(h, gb, h->mb + 16*index, index, scan + 1, chroma_qp, 15) < 0){
4601 uint8_t * const nnz= &h->non_zero_count_cache[0];
4602 nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =
4603 nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
4606 uint8_t * const nnz= &h->non_zero_count_cache[0];
4607 fill_rectangle(&nnz[scan8[0]], 4, 4, 8, 0, 1);
4608 nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =
4609 nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
4611 s->current_picture.qscale_table[mb_xy]= s->qscale;
4612 write_back_non_zero_count(h);
4617 static int decode_cabac_intra_mb_type(H264Context *h, int ctx_base, int intra_slice) {
4618 uint8_t *state= &h->cabac_state[ctx_base];
4622 MpegEncContext * const s = &h->s;
4623 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
4624 const int mba_xy = mb_xy - 1;
4625 const int mbb_xy = mb_xy - s->mb_stride;
4627 if( h->slice_table[mba_xy] == h->slice_num && !IS_INTRA4x4( s->current_picture.mb_type[mba_xy] ) )
4629 if( h->slice_table[mbb_xy] == h->slice_num && !IS_INTRA4x4( s->current_picture.mb_type[mbb_xy] ) )
4631 if( get_cabac( &h->cabac, &state[ctx] ) == 0 )
4632 return 0; /* I4x4 */
4635 if( get_cabac( &h->cabac, &state[0] ) == 0 )
4636 return 0; /* I4x4 */
4639 if( get_cabac_terminate( &h->cabac ) )
4640 return 25; /* PCM */
4642 mb_type = 1; /* I16x16 */
4643 if( get_cabac( &h->cabac, &state[1] ) )
4644 mb_type += 12; /* cbp_luma != 0 */
4646 if( get_cabac( &h->cabac, &state[2] ) ) {
4647 if( get_cabac( &h->cabac, &state[2+intra_slice] ) )
4648 mb_type += 4 * 2; /* cbp_chroma == 2 */
4650 mb_type += 4 * 1; /* cbp_chroma == 1 */
4652 if( get_cabac( &h->cabac, &state[3+intra_slice] ) )
4654 if( get_cabac( &h->cabac, &state[3+2*intra_slice] ) )
4659 static int decode_cabac_mb_type( H264Context *h ) {
4660 MpegEncContext * const s = &h->s;
4662 if( h->slice_type == I_TYPE ) {
4663 return decode_cabac_intra_mb_type(h, 3, 1);
4664 } else if( h->slice_type == P_TYPE ) {
4665 if( get_cabac( &h->cabac, &h->cabac_state[14] ) == 0 ) {
4667 if( get_cabac( &h->cabac, &h->cabac_state[15] ) == 0 ) {
4668 if( get_cabac( &h->cabac, &h->cabac_state[16] ) == 0 )
4669 return 0; /* P_L0_D16x16; */
4671 return 3; /* P_8x8; */
4673 if( get_cabac( &h->cabac, &h->cabac_state[17] ) == 0 )
4674 return 2; /* P_L0_D8x16; */
4676 return 1; /* P_L0_D16x8; */
4679 return decode_cabac_intra_mb_type(h, 17, 0) + 5;
4681 } else if( h->slice_type == B_TYPE ) {
4682 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
4683 const int mba_xy = mb_xy - 1;
4684 const int mbb_xy = mb_xy - s->mb_stride;
4688 if( h->slice_table[mba_xy] == h->slice_num && !IS_SKIP( s->current_picture.mb_type[mba_xy] )
4689 && !IS_DIRECT( s->current_picture.mb_type[mba_xy] ) )
4691 if( h->slice_table[mbb_xy] == h->slice_num && !IS_SKIP( s->current_picture.mb_type[mbb_xy] )
4692 && !IS_DIRECT( s->current_picture.mb_type[mbb_xy] ) )
4695 if( !get_cabac( &h->cabac, &h->cabac_state[27+ctx] ) )
4696 return 0; /* B_Direct_16x16 */
4698 if( !get_cabac( &h->cabac, &h->cabac_state[27+3] ) ) {
4699 return 1 + get_cabac( &h->cabac, &h->cabac_state[27+5] ); /* B_L[01]_16x16 */
4702 bits = get_cabac( &h->cabac, &h->cabac_state[27+4] ) << 3;
4703 bits|= get_cabac( &h->cabac, &h->cabac_state[27+5] ) << 2;
4704 bits|= get_cabac( &h->cabac, &h->cabac_state[27+5] ) << 1;
4705 bits|= get_cabac( &h->cabac, &h->cabac_state[27+5] );
4707 return bits + 3; /* B_Bi_16x16 through B_L1_L0_16x8 */
4708 else if( bits == 13 ) {
4709 return decode_cabac_intra_mb_type(h, 32, 0) + 23;
4710 } else if( bits == 14 )
4711 return 11; /* B_L1_L0_8x16 */
4712 else if( bits == 15 )
4713 return 22; /* B_8x8 */
4715 bits= ( bits<<1 ) | get_cabac( &h->cabac, &h->cabac_state[27+5] );
4716 return bits - 4; /* B_L0_Bi_* through B_Bi_Bi_* */
4718 /* TODO SI/SP frames? */
4723 static int decode_cabac_mb_skip( H264Context *h) {
4724 MpegEncContext * const s = &h->s;
4725 const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
4726 const int mba_xy = mb_xy - 1;
4727 const int mbb_xy = mb_xy - s->mb_stride;
4730 if( h->slice_table[mba_xy] == h->slice_num && !IS_SKIP( s->current_picture.mb_type[mba_xy] ))
4732 if( h->slice_table[mbb_xy] == h->slice_num && !IS_SKIP( s->current_picture.mb_type[mbb_xy] ))
4735 if( h->slice_type == P_TYPE || h->slice_type == SP_TYPE)
4736 return get_cabac( &h->cabac, &h->cabac_state[11+ctx] );
4738 return get_cabac( &h->cabac, &h->cabac_state[24+ctx] );
4741 static int decode_cabac_mb_intra4x4_pred_mode( H264Context *h, int pred_mode ) {
4744 if( get_cabac( &h->cabac, &h->cabac_state[68] ) )
4747 if( get_cabac( &h->cabac, &h->cabac_state[69] ) )
4749 if( get_cabac( &h->cabac, &h->cabac_state[69] ) )
4751 if( get_cabac( &h->cabac, &h->cabac_state[69] ) )
4753 if( mode >= pred_mode )
4759 static int decode_cabac_mb_chroma_pre_mode( H264Context *h) {
4760 MpegEncContext * const s = &h->s;
4761 const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
4762 const int mba_xy = mb_xy - 1;
4763 const int mbb_xy = mb_xy - s->mb_stride;
4767 /* No need to test for IS_INTRA4x4 and IS_INTRA16x16, as we set chroma_pred_mode_table to 0 */
4768 if( h->slice_table[mba_xy] == h->slice_num && h->chroma_pred_mode_table[mba_xy] != 0 )
4771 if( h->slice_table[mbb_xy] == h->slice_num && h->chroma_pred_mode_table[mbb_xy] != 0 )
4774 if( get_cabac( &h->cabac, &h->cabac_state[64+ctx] ) == 0 )
4777 if( get_cabac( &h->cabac, &h->cabac_state[64+3] ) == 0 )
4779 if( get_cabac( &h->cabac, &h->cabac_state[64+3] ) == 0 )
4785 static const uint8_t block_idx_x[16] = {
4786 0, 1, 0, 1, 2, 3, 2, 3, 0, 1, 0, 1, 2, 3, 2, 3
4788 static const uint8_t block_idx_y[16] = {
4789 0, 0, 1, 1, 0, 0, 1, 1, 2, 2, 3, 3, 2, 2, 3, 3
4791 static const uint8_t block_idx_xy[4][4] = {
4798 static int decode_cabac_mb_cbp_luma( H264Context *h) {
4799 MpegEncContext * const s = &h->s;
4800 const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
4805 h->cbp_table[mb_xy] = 0; /* FIXME aaahahahah beurk */
4807 for( i8x8 = 0; i8x8 < 4; i8x8++ ) {
4813 x = block_idx_x[4*i8x8];
4814 y = block_idx_y[4*i8x8];
4818 else if( s->mb_x > 0 ) {
4820 if (h->slice_table[mba_xy] != h->slice_num) {
4827 else if( s->mb_y > 0 ) {
4828 mbb_xy = mb_xy - s->mb_stride;
4829 if (h->slice_table[mbb_xy] != h->slice_num) {
4834 /* No need to test for skip as we put 0 for skip block */
4835 /* No need to test for IPCM as we put 1 for IPCM block */
4837 int i8x8a = block_idx_xy[(x-1)&0x03][y]/4;
4838 if( ((h->cbp_table[mba_xy] >> i8x8a)&0x01) == 0 )
4843 int i8x8b = block_idx_xy[x][(y-1)&0x03]/4;
4844 if( ((h->cbp_table[mbb_xy] >> i8x8b)&0x01) == 0 )
4848 if( get_cabac( &h->cabac, &h->cabac_state[73 + ctx] ) ) {
4850 h->cbp_table[mb_xy] = cbp; /* FIXME aaahahahah beurk */
4855 static int decode_cabac_mb_cbp_chroma( H264Context *h) {
4859 cbp_a = (h->left_cbp>>4)&0x03;
4860 cbp_b = (h-> top_cbp>>4)&0x03;
4863 if( cbp_a > 0 ) ctx++;
4864 if( cbp_b > 0 ) ctx += 2;
4865 if( get_cabac( &h->cabac, &h->cabac_state[77 + ctx] ) == 0 )
4869 if( cbp_a == 2 ) ctx++;
4870 if( cbp_b == 2 ) ctx += 2;
4871 return 1 + get_cabac( &h->cabac, &h->cabac_state[77 + ctx] );
4873 static int decode_cabac_mb_dqp( H264Context *h) {
4874 MpegEncContext * const s = &h->s;
4880 mbn_xy = s->mb_x + s->mb_y*s->mb_stride - 1;
4882 mbn_xy = s->mb_width - 1 + (s->mb_y-1)*s->mb_stride;
4884 if( h->last_qscale_diff != 0 && ( IS_INTRA16x16(s->current_picture.mb_type[mbn_xy] ) || (h->cbp_table[mbn_xy]&0x3f) ) )
4887 while( get_cabac( &h->cabac, &h->cabac_state[60 + ctx] ) ) {
4898 return -(val + 1)/2;
4900 static int decode_cabac_p_mb_sub_type( H264Context *h ) {
4901 if( get_cabac( &h->cabac, &h->cabac_state[21] ) )
4903 if( !get_cabac( &h->cabac, &h->cabac_state[22] ) )
4905 if( get_cabac( &h->cabac, &h->cabac_state[23] ) )
4909 static int decode_cabac_b_mb_sub_type( H264Context *h ) {
4911 if( !get_cabac( &h->cabac, &h->cabac_state[36] ) )
4912 return 0; /* B_Direct_8x8 */
4913 if( !get_cabac( &h->cabac, &h->cabac_state[37] ) )
4914 return 1 + get_cabac( &h->cabac, &h->cabac_state[39] ); /* B_L0_8x8, B_L1_8x8 */
4916 if( get_cabac( &h->cabac, &h->cabac_state[38] ) ) {
4917 if( get_cabac( &h->cabac, &h->cabac_state[39] ) )
4918 return 11 + get_cabac( &h->cabac, &h->cabac_state[39] ); /* B_L1_4x4, B_Bi_4x4 */
4921 type += 2*get_cabac( &h->cabac, &h->cabac_state[39] );
4922 type += get_cabac( &h->cabac, &h->cabac_state[39] );
4926 static int decode_cabac_mb_ref( H264Context *h, int list, int n ) {
4927 int refa = h->ref_cache[list][scan8[n] - 1];
4928 int refb = h->ref_cache[list][scan8[n] - 8];
4932 if( h->slice_type == B_TYPE) {
4933 if( refa > 0 && !h->direct_cache[scan8[n] - 1] )
4935 if( refb > 0 && !h->direct_cache[scan8[n] - 8] )
4944 while( get_cabac( &h->cabac, &h->cabac_state[54+ctx] ) ) {
4954 static int decode_cabac_mb_mvd( H264Context *h, int list, int n, int l ) {
4955 int amvd = abs( h->mvd_cache[list][scan8[n] - 1][l] ) +
4956 abs( h->mvd_cache[list][scan8[n] - 8][l] );
4957 int ctxbase = (l == 0) ? 40 : 47;
4962 else if( amvd > 32 )
4967 if(!get_cabac(&h->cabac, &h->cabac_state[ctxbase+ctx]))
4972 while( mvd < 9 && get_cabac( &h->cabac, &h->cabac_state[ctxbase+ctx] ) ) {
4980 while( get_cabac_bypass( &h->cabac ) ) {
4985 if( get_cabac_bypass( &h->cabac ) )
4989 if( get_cabac_bypass( &h->cabac ) ) return -mvd;
4993 static int inline get_cabac_cbf_ctx( H264Context *h, int cat, int idx ) {
4998 nza = h->left_cbp&0x100;
4999 nzb = h-> top_cbp&0x100;
5000 } else if( cat == 1 || cat == 2 ) {
5001 nza = h->non_zero_count_cache[scan8[idx] - 1];
5002 nzb = h->non_zero_count_cache[scan8[idx] - 8];
5003 } else if( cat == 3 ) {
5004 nza = (h->left_cbp>>(6+idx))&0x01;
5005 nzb = (h-> top_cbp>>(6+idx))&0x01;
5008 nza = h->non_zero_count_cache[scan8[16+idx] - 1];
5009 nzb = h->non_zero_count_cache[scan8[16+idx] - 8];
5018 return ctx + 4 * cat;
5021 static int inline decode_cabac_residual( H264Context *h, DCTELEM *block, int cat, int n, const uint8_t *scantable, int qp, int max_coeff) {
5022 const int mb_xy = h->s.mb_x + h->s.mb_y*h->s.mb_stride;
5023 const uint16_t *qmul= dequant_coeff[qp];
5024 static const int significant_coeff_flag_offset[5] = { 0, 15, 29, 44, 47 };
5025 static const int coeff_abs_level_m1_offset[5] = {227+ 0, 227+10, 227+20, 227+30, 227+39 };
5030 int coeff_count = 0;
5033 int abslevelgt1 = 0;
5035 /* cat: 0-> DC 16x16 n = 0
5036 * 1-> AC 16x16 n = luma4x4idx
5037 * 2-> Luma4x4 n = luma4x4idx
5038 * 3-> DC Chroma n = iCbCr
5039 * 4-> AC Chroma n = 4 * iCbCr + chroma4x4idx
5042 /* read coded block flag */
5043 if( get_cabac( &h->cabac, &h->cabac_state[85 + get_cabac_cbf_ctx( h, cat, n ) ] ) == 0 ) {
5044 if( cat == 1 || cat == 2 )
5045 h->non_zero_count_cache[scan8[n]] = 0;
5047 h->non_zero_count_cache[scan8[16+n]] = 0;
5052 for(last= 0; last < max_coeff - 1; last++) {
5053 if( get_cabac( &h->cabac, &h->cabac_state[105+significant_coeff_flag_offset[cat]+last] )) {
5054 index[coeff_count++] = last;
5055 if( get_cabac( &h->cabac, &h->cabac_state[166+significant_coeff_flag_offset[cat]+last] ) ) {
5061 if( last == max_coeff -1 ) {
5062 index[coeff_count++] = last;
5064 assert(coeff_count > 0);
5067 h->cbp_table[mb_xy] |= 0x100;
5068 else if( cat == 1 || cat == 2 )
5069 h->non_zero_count_cache[scan8[n]] = coeff_count;
5071 h->cbp_table[mb_xy] |= 0x40 << n;
5074 h->non_zero_count_cache[scan8[16+n]] = coeff_count;
5077 for( i = coeff_count - 1; i >= 0; i-- ) {
5078 int ctx = (abslevelgt1 != 0 ? 0 : FFMIN( 4, abslevel1 )) + coeff_abs_level_m1_offset[cat];
5079 int j= scantable[index[i]];
5081 if( get_cabac( &h->cabac, &h->cabac_state[ctx] ) == 0 ) {
5082 if( cat == 0 || cat == 3 ) {
5083 if( get_cabac_bypass( &h->cabac ) ) block[j] = -1;
5086 if( get_cabac_bypass( &h->cabac ) ) block[j] = -qmul[j];
5087 else block[j] = qmul[j];
5093 ctx = 5 + FFMIN( 4, abslevelgt1 ) + coeff_abs_level_m1_offset[cat];
5094 while( coeff_abs < 15 && get_cabac( &h->cabac, &h->cabac_state[ctx] ) ) {
5098 if( coeff_abs >= 15 ) {
5100 while( get_cabac_bypass( &h->cabac ) ) {
5101 coeff_abs += 1 << j;
5106 if( get_cabac_bypass( &h->cabac ) )
5107 coeff_abs += 1 << j ;
5111 if( cat == 0 || cat == 3 ) {
5112 if( get_cabac_bypass( &h->cabac ) ) block[j] = -coeff_abs;
5113 else block[j] = coeff_abs;
5115 if( get_cabac_bypass( &h->cabac ) ) block[j] = -coeff_abs * qmul[j];
5116 else block[j] = coeff_abs * qmul[j];
5126 * decodes a macroblock
5127 * @returns 0 if ok, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
5129 static int decode_mb_cabac(H264Context *h) {
5130 MpegEncContext * const s = &h->s;
5131 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
5132 int mb_type, partition_count, cbp = 0;
5134 s->dsp.clear_blocks(h->mb); //FIXME avoid if allready clear (move after skip handlong?)
5136 if( h->sps.mb_aff ) {
5137 av_log( h->s.avctx, AV_LOG_ERROR, "Fields not supported with CABAC\n" );
5141 tprintf("pic:%d mb:%d/%d\n", h->frame_num, s->mb_x, s->mb_y);
5142 if( h->slice_type != I_TYPE && h->slice_type != SI_TYPE ) {
5143 /* read skip flags */
5144 if( decode_cabac_mb_skip( h ) ) {
5147 h->cbp_table[mb_xy] = 0;
5148 h->chroma_pred_mode_table[mb_xy] = 0;
5149 h->last_qscale_diff = 0;
5155 h->prev_mb_skiped = 0;
5157 if( ( mb_type = decode_cabac_mb_type( h ) ) < 0 ) {
5158 av_log( h->s.avctx, AV_LOG_ERROR, "decode_cabac_mb_type failed\n" );
5162 if( h->slice_type == B_TYPE ) {
5164 partition_count= b_mb_type_info[mb_type].partition_count;
5165 mb_type= b_mb_type_info[mb_type].type;
5168 goto decode_intra_mb;
5170 } else if( h->slice_type == P_TYPE ) {
5172 partition_count= p_mb_type_info[mb_type].partition_count;
5173 mb_type= p_mb_type_info[mb_type].type;
5176 goto decode_intra_mb;
5179 assert(h->slice_type == I_TYPE);
5181 partition_count = 0;
5182 cbp= i_mb_type_info[mb_type].cbp;
5183 h->intra16x16_pred_mode= i_mb_type_info[mb_type].pred_mode;
5184 mb_type= i_mb_type_info[mb_type].type;
5187 if(h->mb_field_decoding_flag)
5188 mb_type |= MB_TYPE_INTERLACED;
5191 s->current_picture.mb_type[mb_xy]= mb_type;
5192 h->slice_table[ mb_xy ]= h->slice_num;
5194 if(IS_INTRA_PCM(mb_type)) {
5198 // We assume these blocks are very rare so we dont optimize it.
5199 // FIXME The two following lines get the bitstream position in the cabac
5200 // decode, I think it should be done by a function in cabac.h (or cabac.c).
5201 ptr= h->cabac.bytestream;
5202 if (h->cabac.low&0x1) ptr-=CABAC_BITS/8;
5204 // The pixels are stored in the same order as levels in h->mb array.
5205 for(y=0; y<16; y++){
5206 const int index= 4*(y&3) + 32*((y>>2)&1) + 128*(y>>3);
5207 for(x=0; x<16; x++){
5208 tprintf("LUMA ICPM LEVEL (%3d)\n", *ptr);
5209 h->mb[index + (x&3) + 16*((x>>2)&1) + 64*(x>>3)]= *ptr++;
5213 const int index= 256 + 4*(y&3) + 32*(y>>2);
5215 tprintf("CHROMA U ICPM LEVEL (%3d)\n", *ptr);
5216 h->mb[index + (x&3) + 16*(x>>2)]= *ptr++;
5220 const int index= 256 + 64 + 4*(y&3) + 32*(y>>2);
5222 tprintf("CHROMA V ICPM LEVEL (%3d)\n", *ptr);
5223 h->mb[index + (x&3) + 16*(x>>2)]= *ptr++;
5227 ff_init_cabac_decoder(&h->cabac, ptr, h->cabac.bytestream_end - ptr);
5229 // All blocks are presents
5230 h->cbp_table[mb_xy] = 0x1ef;
5231 h->chroma_pred_mode_table[mb_xy] = 0;
5232 // In deblocking, the quantiser is 0
5233 s->current_picture.qscale_table[mb_xy]= 0;
5234 h->chroma_qp = get_chroma_qp(h->pps.chroma_qp_index_offset, 0);
5235 // All coeffs are presents
5236 memset(h->non_zero_count[mb_xy], 16, 16);
5240 fill_caches(h, mb_type, 0);
5242 if( IS_INTRA( mb_type ) ) {
5243 if( IS_INTRA4x4( mb_type ) ) {
5245 for( i = 0; i < 16; i++ ) {
5246 int pred = pred_intra_mode( h, i );
5247 h->intra4x4_pred_mode_cache[ scan8[i] ] = decode_cabac_mb_intra4x4_pred_mode( h, pred );
5249 //av_log( s->avctx, AV_LOG_ERROR, "i4x4 pred=%d mode=%d\n", pred, h->intra4x4_pred_mode_cache[ scan8[i] ] );
5251 write_back_intra_pred_mode(h);
5252 if( check_intra4x4_pred_mode(h) < 0 ) return -1;
5254 h->intra16x16_pred_mode= check_intra_pred_mode( h, h->intra16x16_pred_mode );
5255 if( h->intra16x16_pred_mode < 0 ) return -1;
5257 h->chroma_pred_mode_table[mb_xy] =
5258 h->chroma_pred_mode = decode_cabac_mb_chroma_pre_mode( h );
5260 h->chroma_pred_mode= check_intra_pred_mode( h, h->chroma_pred_mode );
5261 if( h->chroma_pred_mode < 0 ) return -1;
5262 } else if( partition_count == 4 ) {
5263 int i, j, sub_partition_count[4], list, ref[2][4];
5265 if( h->slice_type == B_TYPE ) {
5266 for( i = 0; i < 4; i++ ) {
5267 h->sub_mb_type[i] = decode_cabac_b_mb_sub_type( h );
5268 sub_partition_count[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
5269 h->sub_mb_type[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].type;
5271 if( IS_DIRECT(h->sub_mb_type[0]) || IS_DIRECT(h->sub_mb_type[1])
5272 || IS_DIRECT(h->sub_mb_type[2]) || IS_DIRECT(h->sub_mb_type[3])) {
5273 pred_direct_motion(h, &mb_type);
5274 if( h->ref_count[0] > 1 || h->ref_count[1] > 1 ) {
5275 for( i = 0; i < 4; i++ )
5276 if( IS_DIRECT(h->sub_mb_type[i]) )
5277 fill_rectangle( &h->direct_cache[scan8[4*i]], 2, 2, 8, 1, 1 );
5281 for( i = 0; i < 4; i++ ) {
5282 h->sub_mb_type[i] = decode_cabac_p_mb_sub_type( h );
5283 sub_partition_count[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
5284 h->sub_mb_type[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].type;
5288 for( list = 0; list < 2; list++ ) {
5289 if( h->ref_count[list] > 0 ) {
5290 for( i = 0; i < 4; i++ ) {
5291 if(IS_DIRECT(h->sub_mb_type[i])) continue;
5292 if(IS_DIR(h->sub_mb_type[i], 0, list)){
5293 if( h->ref_count[list] > 1 )
5294 ref[list][i] = decode_cabac_mb_ref( h, list, 4*i );
5300 h->ref_cache[list][ scan8[4*i]+1 ]=
5301 h->ref_cache[list][ scan8[4*i]+8 ]=h->ref_cache[list][ scan8[4*i]+9 ]= ref[list][i];
5306 for(list=0; list<2; list++){
5308 if(IS_DIRECT(h->sub_mb_type[i])){
5309 fill_rectangle(h->mvd_cache[list][scan8[4*i]], 2, 2, 8, 0, 4);
5312 h->ref_cache[list][ scan8[4*i] ]=h->ref_cache[list][ scan8[4*i]+1 ];
5314 if(IS_DIR(h->sub_mb_type[i], 0, list) && !IS_DIRECT(h->sub_mb_type[i])){
5315 const int sub_mb_type= h->sub_mb_type[i];
5316 const int block_width= (sub_mb_type & (MB_TYPE_16x16|MB_TYPE_16x8)) ? 2 : 1;
5317 for(j=0; j<sub_partition_count[i]; j++){
5320 const int index= 4*i + block_width*j;
5321 int16_t (* mv_cache)[2]= &h->mv_cache[list][ scan8[index] ];
5322 int16_t (* mvd_cache)[2]= &h->mvd_cache[list][ scan8[index] ];
5323 pred_motion(h, index, block_width, list, h->ref_cache[list][ scan8[index] ], &mpx, &mpy);
5325 mx = mpx + decode_cabac_mb_mvd( h, list, index, 0 );
5326 my = mpy + decode_cabac_mb_mvd( h, list, index, 1 );
5327 tprintf("final mv:%d %d\n", mx, my);
5329 if(IS_SUB_8X8(sub_mb_type)){
5330 mv_cache[ 0 ][0]= mv_cache[ 1 ][0]=
5331 mv_cache[ 8 ][0]= mv_cache[ 9 ][0]= mx;
5332 mv_cache[ 0 ][1]= mv_cache[ 1 ][1]=
5333 mv_cache[ 8 ][1]= mv_cache[ 9 ][1]= my;
5335 mvd_cache[ 0 ][0]= mvd_cache[ 1 ][0]=
5336 mvd_cache[ 8 ][0]= mvd_cache[ 9 ][0]= mx - mpx;
5337 mvd_cache[ 0 ][1]= mvd_cache[ 1 ][1]=
5338 mvd_cache[ 8 ][1]= mvd_cache[ 9 ][1]= my - mpy;
5339 }else if(IS_SUB_8X4(sub_mb_type)){
5340 mv_cache[ 0 ][0]= mv_cache[ 1 ][0]= mx;
5341 mv_cache[ 0 ][1]= mv_cache[ 1 ][1]= my;
5343 mvd_cache[ 0 ][0]= mvd_cache[ 1 ][0]= mx- mpx;
5344 mvd_cache[ 0 ][1]= mvd_cache[ 1 ][1]= my - mpy;
5345 }else if(IS_SUB_4X8(sub_mb_type)){
5346 mv_cache[ 0 ][0]= mv_cache[ 8 ][0]= mx;
5347 mv_cache[ 0 ][1]= mv_cache[ 8 ][1]= my;
5349 mvd_cache[ 0 ][0]= mvd_cache[ 8 ][0]= mx - mpx;
5350 mvd_cache[ 0 ][1]= mvd_cache[ 8 ][1]= my - mpy;
5352 assert(IS_SUB_4X4(sub_mb_type));
5353 mv_cache[ 0 ][0]= mx;
5354 mv_cache[ 0 ][1]= my;
5356 mvd_cache[ 0 ][0]= mx - mpx;
5357 mvd_cache[ 0 ][1]= my - mpy;
5361 uint32_t *p= (uint32_t *)&h->mv_cache[list][ scan8[4*i] ][0];
5362 uint32_t *pd= (uint32_t *)&h->mvd_cache[list][ scan8[4*i] ][0];
5363 p[0] = p[1] = p[8] = p[9] = 0;
5364 pd[0]= pd[1]= pd[8]= pd[9]= 0;
5368 } else if( IS_DIRECT(mb_type) ) {
5369 pred_direct_motion(h, &mb_type);
5370 s->current_picture.mb_type[mb_xy]= mb_type;
5371 fill_rectangle(h->mvd_cache[0][scan8[0]], 4, 4, 8, 0, 4);
5372 fill_rectangle(h->mvd_cache[1][scan8[0]], 4, 4, 8, 0, 4);
5374 int list, mx, my, i, mpx, mpy;
5375 if(IS_16X16(mb_type)){
5376 for(list=0; list<2; list++){
5377 if(IS_DIR(mb_type, 0, list)){
5378 if(h->ref_count[list] > 0 ){
5379 const int ref = h->ref_count[list] > 1 ? decode_cabac_mb_ref( h, list, 0 ) : 0;
5380 fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, ref, 1);
5383 fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, (uint8_t)LIST_NOT_USED, 1);
5385 for(list=0; list<2; list++){
5386 if(IS_DIR(mb_type, 0, list)){
5387 pred_motion(h, 0, 4, list, h->ref_cache[list][ scan8[0] ], &mpx, &mpy);
5389 mx = mpx + decode_cabac_mb_mvd( h, list, 0, 0 );
5390 my = mpy + decode_cabac_mb_mvd( h, list, 0, 1 );
5391 tprintf("final mv:%d %d\n", mx, my);
5393 fill_rectangle(h->mvd_cache[list][ scan8[0] ], 4, 4, 8, pack16to32(mx-mpx,my-mpy), 4);
5394 fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, pack16to32(mx,my), 4);
5396 fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, 0, 4);
5399 else if(IS_16X8(mb_type)){
5400 for(list=0; list<2; list++){
5401 if(h->ref_count[list]>0){
5403 if(IS_DIR(mb_type, i, list)){
5404 const int ref= h->ref_count[list] > 1 ? decode_cabac_mb_ref( h, list, 8*i ) : 0;
5405 fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, ref, 1);
5407 fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, (LIST_NOT_USED&0xFF), 1);
5411 for(list=0; list<2; list++){
5413 if(IS_DIR(mb_type, i, list)){
5414 pred_16x8_motion(h, 8*i, list, h->ref_cache[list][scan8[0] + 16*i], &mpx, &mpy);
5415 mx = mpx + decode_cabac_mb_mvd( h, list, 8*i, 0 );
5416 my = mpy + decode_cabac_mb_mvd( h, list, 8*i, 1 );
5417 tprintf("final mv:%d %d\n", mx, my);
5419 fill_rectangle(h->mvd_cache[list][ scan8[0] + 16*i ], 4, 2, 8, pack16to32(mx-mpx,my-mpy), 4);
5420 fill_rectangle(h->mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, pack16to32(mx,my), 4);
5422 fill_rectangle(h->mvd_cache[list][ scan8[0] + 16*i ], 4, 2, 8, 0, 4);
5423 fill_rectangle(h-> mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, 0, 4);
5428 assert(IS_8X16(mb_type));
5429 for(list=0; list<2; list++){
5430 if(h->ref_count[list]>0){
5432 if(IS_DIR(mb_type, i, list)){ //FIXME optimize
5433 const int ref= h->ref_count[list] > 1 ? decode_cabac_mb_ref( h, list, 4*i ) : 0;
5434 fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, ref, 1);
5436 fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, (LIST_NOT_USED&0xFF), 1);
5440 for(list=0; list<2; list++){
5442 if(IS_DIR(mb_type, i, list)){
5443 pred_8x16_motion(h, i*4, list, h->ref_cache[list][ scan8[0] + 2*i ], &mpx, &mpy);
5444 mx = mpx + decode_cabac_mb_mvd( h, list, 4*i, 0 );
5445 my = mpy + decode_cabac_mb_mvd( h, list, 4*i, 1 );
5447 tprintf("final mv:%d %d\n", mx, my);
5448 fill_rectangle(h->mvd_cache[list][ scan8[0] + 2*i ], 2, 4, 8, pack16to32(mx-mpx,my-mpy), 4);
5449 fill_rectangle(h->mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, pack16to32(mx,my), 4);
5451 fill_rectangle(h->mvd_cache[list][ scan8[0] + 2*i ], 2, 4, 8, 0, 4);
5452 fill_rectangle(h-> mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, 0, 4);
5459 if( IS_INTER( mb_type ) ) {
5460 h->chroma_pred_mode_table[mb_xy] = 0;
5461 write_back_motion( h, mb_type );
5464 if( !IS_INTRA16x16( mb_type ) ) {
5465 cbp = decode_cabac_mb_cbp_luma( h );
5466 cbp |= decode_cabac_mb_cbp_chroma( h ) << 4;
5469 h->cbp_table[mb_xy] = cbp;
5471 if( cbp || IS_INTRA16x16( mb_type ) ) {
5472 const uint8_t *scan, *dc_scan;
5475 if(IS_INTERLACED(mb_type)){
5477 dc_scan= luma_dc_field_scan;
5480 dc_scan= luma_dc_zigzag_scan;
5483 h->last_qscale_diff = dqp = decode_cabac_mb_dqp( h );
5485 if(((unsigned)s->qscale) > 51){
5486 if(s->qscale<0) s->qscale+= 52;
5487 else s->qscale-= 52;
5489 h->chroma_qp = get_chroma_qp(h->pps.chroma_qp_index_offset, s->qscale);
5491 if( IS_INTRA16x16( mb_type ) ) {
5493 //av_log( s->avctx, AV_LOG_ERROR, "INTRA16x16 DC\n" );
5494 if( decode_cabac_residual( h, h->mb, 0, 0, dc_scan, s->qscale, 16) < 0)
5497 for( i = 0; i < 16; i++ ) {
5498 //av_log( s->avctx, AV_LOG_ERROR, "INTRA16x16 AC:%d\n", i );
5499 if( decode_cabac_residual(h, h->mb + 16*i, 1, i, scan + 1, s->qscale, 15) < 0 )
5503 fill_rectangle(&h->non_zero_count_cache[scan8[0]], 4, 4, 8, 0, 1);
5507 for( i8x8 = 0; i8x8 < 4; i8x8++ ) {
5508 if( cbp & (1<<i8x8) ) {
5509 for( i4x4 = 0; i4x4 < 4; i4x4++ ) {
5510 const int index = 4*i8x8 + i4x4;
5511 //av_log( s->avctx, AV_LOG_ERROR, "Luma4x4: %d\n", index );
5512 if( decode_cabac_residual(h, h->mb + 16*index, 2, index, scan, s->qscale, 16) < 0 )
5516 uint8_t * const nnz= &h->non_zero_count_cache[ scan8[4*i8x8] ];
5517 nnz[0] = nnz[1] = nnz[8] = nnz[9] = 0;
5524 for( c = 0; c < 2; c++ ) {
5525 //av_log( s->avctx, AV_LOG_ERROR, "INTRA C%d-DC\n",c );
5526 if( decode_cabac_residual(h, h->mb + 256 + 16*4*c, 3, c, chroma_dc_scan, h->chroma_qp, 4) < 0)
5533 for( c = 0; c < 2; c++ ) {
5534 for( i = 0; i < 4; i++ ) {
5535 const int index = 16 + 4 * c + i;
5536 //av_log( s->avctx, AV_LOG_ERROR, "INTRA C%d-AC %d\n",c, index - 16 );
5537 if( decode_cabac_residual(h, h->mb + 16*index, 4, index - 16, scan + 1, h->chroma_qp, 15) < 0)
5542 uint8_t * const nnz= &h->non_zero_count_cache[0];
5543 nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =
5544 nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
5547 uint8_t * const nnz= &h->non_zero_count_cache[0];
5548 fill_rectangle(&nnz[scan8[0]], 4, 4, 8, 0, 1);
5549 nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =
5550 nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
5553 s->current_picture.qscale_table[mb_xy]= s->qscale;
5554 write_back_non_zero_count(h);
5560 static void filter_mb_edgev( H264Context *h, uint8_t *pix, int stride, int bS[4], int qp ) {
5562 const int index_a = clip( qp + h->slice_alpha_c0_offset, 0, 51 );
5563 const int alpha = alpha_table[index_a];
5564 const int beta = beta_table[clip( qp + h->slice_beta_offset, 0, 51 )];
5566 for( i = 0; i < 4; i++ ) {
5573 const int tc0 = tc0_table[index_a][bS[i] - 1];
5574 /* 4px edge length */
5575 for( d = 0; d < 4; d++ ) {
5576 const int p0 = pix[-1];
5577 const int p1 = pix[-2];
5578 const int p2 = pix[-3];
5579 const int q0 = pix[0];
5580 const int q1 = pix[1];
5581 const int q2 = pix[2];
5583 if( ABS( p0 - q0 ) < alpha &&
5584 ABS( p1 - p0 ) < beta &&
5585 ABS( q1 - q0 ) < beta ) {
5589 if( ABS( p2 - p0 ) < beta ) {
5590 pix[-2] = p1 + clip( ( p2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( p1 << 1 ) ) >> 1, -tc0, tc0 );
5593 if( ABS( q2 - q0 ) < beta ) {
5594 pix[1] = q1 + clip( ( q2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( q1 << 1 ) ) >> 1, -tc0, tc0 );
5598 i_delta = clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );
5599 pix[-1] = clip_uint8( p0 + i_delta ); /* p0' */
5600 pix[0] = clip_uint8( q0 - i_delta ); /* q0' */
5601 tprintf("filter_mb_edgev i:%d d:%d, qp:%d, indexA:%d, alpha:%d, beta:%d, tc:%d\n# bS:%d -> [%02x, %02x, %02x, %02x, %02x, %02x] =>[%02x, %02x, %02x, %02x]\n", i, d, qp, index_a, alpha, beta, tc, bS[i], pix[-3], p1, p0, q0, q1, pix[2], pix[-2], pix[-1], pix[0], pix[1]);
5606 /* 4px edge length */
5607 for( d = 0; d < 4; d++ ) {
5608 const int p0 = pix[-1];
5609 const int p1 = pix[-2];
5610 const int p2 = pix[-3];
5612 const int q0 = pix[0];
5613 const int q1 = pix[1];
5614 const int q2 = pix[2];
5616 if( ABS( p0 - q0 ) < alpha &&
5617 ABS( p1 - p0 ) < beta &&
5618 ABS( q1 - q0 ) < beta ) {
5620 if(ABS( p0 - q0 ) < (( alpha >> 2 ) + 2 )){
5621 if( ABS( p2 - p0 ) < beta)
5623 const int p3 = pix[-4];
5625 pix[-1] = ( p2 + 2*p1 + 2*p0 + 2*q0 + q1 + 4 ) >> 3;
5626 pix[-2] = ( p2 + p1 + p0 + q0 + 2 ) >> 2;
5627 pix[-3] = ( 2*p3 + 3*p2 + p1 + p0 + q0 + 4 ) >> 3;
5630 pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
5632 if( ABS( q2 - q0 ) < beta)
5634 const int q3 = pix[3];
5636 pix[0] = ( p1 + 2*p0 + 2*q0 + 2*q1 + q2 + 4 ) >> 3;
5637 pix[1] = ( p0 + q0 + q1 + q2 + 2 ) >> 2;
5638 pix[2] = ( 2*q3 + 3*q2 + q1 + q0 + p0 + 4 ) >> 3;
5641 pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
5645 pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
5646 pix[ 0] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
5648 tprintf("filter_mb_edgev i:%d d:%d\n# bS:4 -> [%02x, %02x, %02x, %02x, %02x, %02x] =>[%02x, %02x, %02x, %02x]\n", i, d, p2, p1, p0, q0, q1, q2, pix[-2], pix[-1], pix[0], pix[1]);
5655 static void filter_mb_edgecv( H264Context *h, uint8_t *pix, int stride, int bS[4], int qp ) {
5657 const int index_a = clip( qp + h->slice_alpha_c0_offset, 0, 51 );
5658 const int alpha = alpha_table[index_a];
5659 const int beta = beta_table[clip( qp + h->slice_beta_offset, 0, 51 )];
5661 for( i = 0; i < 4; i++ ) {
5668 const int tc = tc0_table[index_a][bS[i] - 1] + 1;
5669 /* 2px edge length (because we use same bS than the one for luma) */
5670 for( d = 0; d < 2; d++ ){
5671 const int p0 = pix[-1];
5672 const int p1 = pix[-2];
5673 const int q0 = pix[0];
5674 const int q1 = pix[1];
5676 if( ABS( p0 - q0 ) < alpha &&
5677 ABS( p1 - p0 ) < beta &&
5678 ABS( q1 - q0 ) < beta ) {
5679 const int i_delta = clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );
5681 pix[-1] = clip_uint8( p0 + i_delta ); /* p0' */
5682 pix[0] = clip_uint8( q0 - i_delta ); /* q0' */
5683 tprintf("filter_mb_edgecv i:%d d:%d, qp:%d, indexA:%d, alpha:%d, beta:%d, tc:%d\n# bS:%d -> [%02x, %02x, %02x, %02x, %02x, %02x] =>[%02x, %02x, %02x, %02x]\n", i, d, qp, index_a, alpha, beta, tc, bS[i], pix[-3], p1, p0, q0, q1, pix[2], p1, pix[-1], pix[0], q1);
5688 /* 2px edge length (because we use same bS than the one for luma) */
5689 for( d = 0; d < 2; d++ ){
5690 const int p0 = pix[-1];
5691 const int p1 = pix[-2];
5692 const int q0 = pix[0];
5693 const int q1 = pix[1];
5695 if( ABS( p0 - q0 ) < alpha &&
5696 ABS( p1 - p0 ) < beta &&
5697 ABS( q1 - q0 ) < beta ) {
5699 pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2; /* p0' */
5700 pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2; /* q0' */
5701 tprintf("filter_mb_edgecv i:%d d:%d\n# bS:4 -> [%02x, %02x, %02x, %02x, %02x, %02x] =>[%02x, %02x, %02x, %02x]\n", i, d, pix[-3], p1, p0, q0, q1, pix[2], p1, pix[-1], pix[0], q1);
5709 static void filter_mb_mbaff_edgev( H264Context *h, uint8_t *pix, int stride, int bS[8], int qp[2] ) {
5711 for( i = 0; i < 16; i++, pix += stride) {
5717 int bS_index = (i >> 1);
5718 if (h->mb_field_decoding_flag) {
5720 bS_index |= (i & 1);
5723 if( bS[bS_index] == 0 ) {
5727 qp_index = h->mb_field_decoding_flag ? (i & 1) : (i >> 3);
5728 index_a = clip( qp[qp_index] + h->slice_alpha_c0_offset, 0, 51 );
5729 alpha = alpha_table[index_a];
5730 beta = beta_table[clip( qp[qp_index] + h->slice_beta_offset, 0, 51 )];
5733 if( bS[bS_index] < 4 ) {
5734 const int tc0 = tc0_table[index_a][bS[bS_index] - 1];
5735 /* 4px edge length */
5736 const int p0 = pix[-1];
5737 const int p1 = pix[-2];
5738 const int p2 = pix[-3];
5739 const int q0 = pix[0];
5740 const int q1 = pix[1];
5741 const int q2 = pix[2];
5743 if( ABS( p0 - q0 ) < alpha &&
5744 ABS( p1 - p0 ) < beta &&
5745 ABS( q1 - q0 ) < beta ) {
5749 if( ABS( p2 - p0 ) < beta ) {
5750 pix[-2] = p1 + clip( ( p2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( p1 << 1 ) ) >> 1, -tc0, tc0 );
5753 if( ABS( q2 - q0 ) < beta ) {
5754 pix[1] = q1 + clip( ( q2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( q1 << 1 ) ) >> 1, -tc0, tc0 );
5758 i_delta = clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );
5759 pix[-1] = clip_uint8( p0 + i_delta ); /* p0' */
5760 pix[0] = clip_uint8( q0 - i_delta ); /* q0' */
5761 tprintf("filter_mb_mbaff_edgev i:%d, qp:%d, indexA:%d, alpha:%d, beta:%d, tc:%d\n# bS:%d -> [%02x, %02x, %02x, %02x, %02x, %02x] =>[%02x, %02x, %02x, %02x]\n", i, qp[qp_index], index_a, alpha, beta, tc, bS[bS_index], pix[-3], p1, p0, q0, q1, pix[2], p1, pix[-1], pix[0], q1);
5764 /* 4px edge length */
5765 const int p0 = pix[-1];
5766 const int p1 = pix[-2];
5767 const int p2 = pix[-3];
5769 const int q0 = pix[0];
5770 const int q1 = pix[1];
5771 const int q2 = pix[2];
5773 if( ABS( p0 - q0 ) < alpha &&
5774 ABS( p1 - p0 ) < beta &&
5775 ABS( q1 - q0 ) < beta ) {
5777 if(ABS( p0 - q0 ) < (( alpha >> 2 ) + 2 )){
5778 if( ABS( p2 - p0 ) < beta)
5780 const int p3 = pix[-4];
5782 pix[-1] = ( p2 + 2*p1 + 2*p0 + 2*q0 + q1 + 4 ) >> 3;
5783 pix[-2] = ( p2 + p1 + p0 + q0 + 2 ) >> 2;
5784 pix[-3] = ( 2*p3 + 3*p2 + p1 + p0 + q0 + 4 ) >> 3;
5787 pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
5789 if( ABS( q2 - q0 ) < beta)
5791 const int q3 = pix[3];
5793 pix[0] = ( p1 + 2*p0 + 2*q0 + 2*q1 + q2 + 4 ) >> 3;
5794 pix[1] = ( p0 + q0 + q1 + q2 + 2 ) >> 2;
5795 pix[2] = ( 2*q3 + 3*q2 + q1 + q0 + p0 + 4 ) >> 3;
5798 pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
5802 pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
5803 pix[ 0] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
5805 tprintf("filter_mb_mbaff_edgev i:%d, qp:%d, indexA:%d, alpha:%d, beta:%d\n# bS:4 -> [%02x, %02x, %02x, %02x, %02x, %02x] =>[%02x, %02x, %02x, %02x, %02x, %02x]\n", i, qp[qp_index], index_a, alpha, beta, p2, p1, p0, q0, q1, q2, pix[-3], pix[-2], pix[-1], pix[0], pix[1], pix[2]);
5810 static void filter_mb_mbaff_edgecv( H264Context *h, uint8_t *pix, int stride, int bS[4], int qp[2] ) {
5812 for( i = 0; i < 8; i++, pix += stride) {
5820 if( bS[bS_index] == 0 ) {
5824 qp_index = h->mb_field_decoding_flag ? (i & 1) : (i >> 3);
5825 index_a = clip( qp[qp_index] + h->slice_alpha_c0_offset, 0, 51 );
5826 alpha = alpha_table[index_a];
5827 beta = beta_table[clip( qp[qp_index] + h->slice_beta_offset, 0, 51 )];
5828 if( bS[bS_index] < 4 ) {
5829 const int tc = tc0_table[index_a][bS[bS_index] - 1] + 1;
5830 /* 2px edge length (because we use same bS than the one for luma) */
5831 const int p0 = pix[-1];
5832 const int p1 = pix[-2];
5833 const int q0 = pix[0];
5834 const int q1 = pix[1];
5836 if( ABS( p0 - q0 ) < alpha &&
5837 ABS( p1 - p0 ) < beta &&
5838 ABS( q1 - q0 ) < beta ) {
5839 const int i_delta = clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );
5841 pix[-1] = clip_uint8( p0 + i_delta ); /* p0' */
5842 pix[0] = clip_uint8( q0 - i_delta ); /* q0' */
5843 tprintf("filter_mb_mbaff_edgecv i:%d, qp:%d, indexA:%d, alpha:%d, beta:%d, tc:%d\n# bS:%d -> [%02x, %02x, %02x, %02x, %02x, %02x] =>[%02x, %02x, %02x, %02x]\n", i, qp[qp_index], index_a, alpha, beta, tc, bS[bS_index], pix[-3], p1, p0, q0, q1, pix[2], p1, pix[-1], pix[0], q1);
5846 const int p0 = pix[-1];
5847 const int p1 = pix[-2];
5848 const int q0 = pix[0];
5849 const int q1 = pix[1];
5851 if( ABS( p0 - q0 ) < alpha &&
5852 ABS( p1 - p0 ) < beta &&
5853 ABS( q1 - q0 ) < beta ) {
5855 pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2; /* p0' */
5856 pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2; /* q0' */
5857 tprintf("filter_mb_mbaff_edgecv i:%d\n# bS:4 -> [%02x, %02x, %02x, %02x, %02x, %02x] =>[%02x, %02x, %02x, %02x, %02x, %02x]\n", i, pix[-3], p1, p0, q0, q1, pix[2], pix[-3], pix[-2], pix[-1], pix[0], pix[1], pix[2]);
5863 static void filter_mb_edgeh( H264Context *h, uint8_t *pix, int stride, int bS[4], int qp ) {
5865 const int index_a = clip( qp + h->slice_alpha_c0_offset, 0, 51 );
5866 const int alpha = alpha_table[index_a];
5867 const int beta = beta_table[clip( qp + h->slice_beta_offset, 0, 51 )];
5868 const int pix_next = stride;
5870 for( i = 0; i < 4; i++ ) {
5877 const int tc0 = tc0_table[index_a][bS[i] - 1];
5878 /* 4px edge length */
5879 for( d = 0; d < 4; d++ ) {
5880 const int p0 = pix[-1*pix_next];
5881 const int p1 = pix[-2*pix_next];
5882 const int p2 = pix[-3*pix_next];
5883 const int q0 = pix[0];
5884 const int q1 = pix[1*pix_next];
5885 const int q2 = pix[2*pix_next];
5887 if( ABS( p0 - q0 ) < alpha &&
5888 ABS( p1 - p0 ) < beta &&
5889 ABS( q1 - q0 ) < beta ) {
5894 if( ABS( p2 - p0 ) < beta ) {
5895 pix[-2*pix_next] = p1 + clip( ( p2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( p1 << 1 ) ) >> 1, -tc0, tc0 );
5898 if( ABS( q2 - q0 ) < beta ) {
5899 pix[pix_next] = q1 + clip( ( q2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( q1 << 1 ) ) >> 1, -tc0, tc0 );
5903 i_delta = clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );
5904 pix[-pix_next] = clip_uint8( p0 + i_delta ); /* p0' */
5905 pix[0] = clip_uint8( q0 - i_delta ); /* q0' */
5906 tprintf("filter_mb_edgeh i:%d d:%d, qp:%d, indexA:%d, alpha:%d, beta:%d, tc:%d\n# bS:%d -> [%02x, %02x, %02x, %02x, %02x, %02x] =>[%02x, %02x, %02x, %02x]\n", i, d, qp, index_a, alpha, beta, tc, bS[i], p2, p1, p0, q0, q1, q2, pix[-2*pix_next], pix[-pix_next], pix[0], pix[pix_next]);
5911 /* 4px edge length */
5912 for( d = 0; d < 4; d++ ) {
5913 const int p0 = pix[-1*pix_next];
5914 const int p1 = pix[-2*pix_next];
5915 const int p2 = pix[-3*pix_next];
5916 const int q0 = pix[0];
5917 const int q1 = pix[1*pix_next];
5918 const int q2 = pix[2*pix_next];
5920 if( ABS( p0 - q0 ) < alpha &&
5921 ABS( p1 - p0 ) < beta &&
5922 ABS( q1 - q0 ) < beta ) {
5924 const int p3 = pix[-4*pix_next];
5925 const int q3 = pix[ 3*pix_next];
5927 if(ABS( p0 - q0 ) < (( alpha >> 2 ) + 2 )){
5928 if( ABS( p2 - p0 ) < beta) {
5930 pix[-1*pix_next] = ( p2 + 2*p1 + 2*p0 + 2*q0 + q1 + 4 ) >> 3;
5931 pix[-2*pix_next] = ( p2 + p1 + p0 + q0 + 2 ) >> 2;
5932 pix[-3*pix_next] = ( 2*p3 + 3*p2 + p1 + p0 + q0 + 4 ) >> 3;
5935 pix[-1*pix_next] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
5937 if( ABS( q2 - q0 ) < beta) {
5939 pix[0*pix_next] = ( p1 + 2*p0 + 2*q0 + 2*q1 + q2 + 4 ) >> 3;
5940 pix[1*pix_next] = ( p0 + q0 + q1 + q2 + 2 ) >> 2;
5941 pix[2*pix_next] = ( 2*q3 + 3*q2 + q1 + q0 + p0 + 4 ) >> 3;
5944 pix[0*pix_next] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
5948 pix[-1*pix_next] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
5949 pix[ 0*pix_next] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
5951 tprintf("filter_mb_edgeh i:%d d:%d, qp:%d, indexA:%d, alpha:%d, beta:%d\n# bS:%d -> [%02x, %02x, %02x, %02x, %02x, %02x] =>[%02x, %02x, %02x, %02x]\n", i, d, qp, index_a, alpha, beta, bS[i], p2, p1, p0, q0, q1, q2, pix[-2*pix_next], pix[-pix_next], pix[0], pix[pix_next]);
5959 static void filter_mb_edgech( H264Context *h, uint8_t *pix, int stride, int bS[4], int qp ) {
5961 const int index_a = clip( qp + h->slice_alpha_c0_offset, 0, 51 );
5962 const int alpha = alpha_table[index_a];
5963 const int beta = beta_table[clip( qp + h->slice_beta_offset, 0, 51 )];
5964 const int pix_next = stride;
5966 for( i = 0; i < 4; i++ )
5974 int tc = tc0_table[index_a][bS[i] - 1] + 1;
5975 /* 2px edge length (see deblocking_filter_edgecv) */
5976 for( d = 0; d < 2; d++ ) {
5977 const int p0 = pix[-1*pix_next];
5978 const int p1 = pix[-2*pix_next];
5979 const int q0 = pix[0];
5980 const int q1 = pix[1*pix_next];
5982 if( ABS( p0 - q0 ) < alpha &&
5983 ABS( p1 - p0 ) < beta &&
5984 ABS( q1 - q0 ) < beta ) {
5986 int i_delta = clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );
5988 pix[-pix_next] = clip_uint8( p0 + i_delta ); /* p0' */
5989 pix[0] = clip_uint8( q0 - i_delta ); /* q0' */
5990 tprintf("filter_mb_edgech i:%d d:%d, qp:%d, indexA:%d, alpha:%d, beta:%d, tc:%d\n# bS:%d -> [%02x, %02x, %02x, %02x, %02x, %02x] =>[%02x, %02x, %02x, %02x]\n", i, d, qp, index_a, alpha, beta, tc, bS[i], pix[-3*pix_next], p1, p0, q0, q1, pix[2*pix_next], pix[-2*pix_next], pix[-pix_next], pix[0], pix[pix_next]);
5995 /* 2px edge length (see deblocking_filter_edgecv) */
5996 for( d = 0; d < 2; d++ ) {
5997 const int p0 = pix[-1*pix_next];
5998 const int p1 = pix[-2*pix_next];
5999 const int q0 = pix[0];
6000 const int q1 = pix[1*pix_next];
6002 if( ABS( p0 - q0 ) < alpha &&
6003 ABS( p1 - p0 ) < beta &&
6004 ABS( q1 - q0 ) < beta ) {
6006 pix[-pix_next] = ( 2*p1 + p0 + q1 + 2 ) >> 2; /* p0' */
6007 pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2; /* q0' */
6008 tprintf("filter_mb_edgech i:%d d:%d, qp:%d, indexA:%d, alpha:%d, beta:%d\n# bS:%d -> [%02x, %02x, %02x, %02x, %02x, %02x] =>[%02x, %02x, %02x, %02x]\n", i, d, qp, index_a, alpha, beta, bS[i], pix[-3*pix_next], p1, p0, q0, q1, pix[2*pix_next], pix[-2*pix_next], pix[-pix_next], pix[0], pix[pix_next]);
6016 static void filter_mb( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr, unsigned int linesize, unsigned int uvlinesize) {
6017 MpegEncContext * const s = &h->s;
6018 const int mb_xy= mb_x + mb_y*s->mb_stride;
6019 int first_vertical_edge_done = 0;
6023 // left mb is in picture
6024 && h->slice_table[mb_xy-1] != 255
6025 // and current and left pair do not have the same interlaced type
6026 && (IS_INTERLACED(s->current_picture.mb_type[mb_xy]) != IS_INTERLACED(s->current_picture.mb_type[mb_xy-1]))
6027 // and left mb is in the same slice if deblocking_filter == 2
6028 && (h->deblocking_filter!=2 || h->slice_table[mb_xy-1] == h->slice_table[mb_xy])) {
6029 /* First vertical edge is different in MBAFF frames
6030 * There are 8 differents bS to compute and 2 differents Qp
6037 first_vertical_edge_done = 1;
6038 for( i = 0; i < 8; i++ ) {
6040 int b_idx= 8 + 4 + 8*y;
6041 int bn_idx= b_idx - 1;
6043 int mbn_xy = h->mb_field_decoding_flag ? h->left_mb_xy[i>>2] : h->left_mb_xy[i&1];
6045 if( IS_INTRA( s->current_picture.mb_type[mb_xy] ) ||
6046 IS_INTRA( s->current_picture.mb_type[mbn_xy] ) ) {
6048 } else if( h->non_zero_count_cache[b_idx] != 0 ||
6049 h->non_zero_count_cache[bn_idx] != 0 ) {
6052 /* FIXME: A given frame may occupy more than one position in
6053 * the reference list. So we should compare the frame numbers,
6054 * not the indices in the ref list. */
6057 for( l = 0; l < 1 + (h->slice_type == B_TYPE); l++ ) {
6058 if( h->ref_cache[l][b_idx] != h->ref_cache[l][bn_idx] ||
6059 ABS( h->mv_cache[l][b_idx][0] - h->mv_cache[l][bn_idx][0] ) >= 4 ||
6060 ABS( h->mv_cache[l][b_idx][1] - h->mv_cache[l][bn_idx][1] ) >= 4 ) {
6067 if(bS[0]+bS[1]+bS[2]+bS[3] != 0) {
6068 // Do not use s->qscale as luma quantiser because it has not the same
6069 // value in IPCM macroblocks.
6070 qp[0] = ( s->current_picture.qscale_table[mb_xy] + s->current_picture.qscale_table[h->left_mb_xy[0]] + 1 ) >> 1;
6071 chroma_qp[0] = ( get_chroma_qp( h->pps.chroma_qp_index_offset, s->current_picture.qscale_table[mb_xy] ) +
6072 get_chroma_qp( h->pps.chroma_qp_index_offset, s->current_picture.qscale_table[h->left_mb_xy[0]] ) + 1 ) >> 1;
6073 qp[1] = ( s->current_picture.qscale_table[mb_xy] + s->current_picture.qscale_table[h->left_mb_xy[1]] + 1 ) >> 1;
6074 chroma_qp[1] = ( get_chroma_qp( h->pps.chroma_qp_index_offset, s->current_picture.qscale_table[mb_xy] ) +
6075 get_chroma_qp( h->pps.chroma_qp_index_offset, s->current_picture.qscale_table[h->left_mb_xy[1]] ) + 1 ) >> 1;
6078 tprintf("filter mb:%d/%d MBAFF, QPy:%d/%d, QPc:%d/%d ls:%d uvls:%d", mb_x, mb_y, qp[0], qp[1], chroma_qp[0], chroma_qp[1], linesize, uvlinesize);
6079 { int i; for (i = 0; i < 8; i++) tprintf(" bS[%d]:%d", i, bS[i]); tprintf("\n"); }
6080 filter_mb_mbaff_edgev ( h, &img_y [0], linesize, bS, qp );
6081 filter_mb_mbaff_edgecv( h, &img_cb[0], uvlinesize, bS, chroma_qp );
6082 filter_mb_mbaff_edgecv( h, &img_cr[0], uvlinesize, bS, chroma_qp );
6085 /* dir : 0 -> vertical edge, 1 -> horizontal edge */
6086 for( dir = 0; dir < 2; dir++ )
6089 const int mbm_xy = dir == 0 ? mb_xy -1 : h->top_mb_xy;
6090 int start = h->slice_table[mbm_xy] == 255 ? 1 : 0;
6092 if (first_vertical_edge_done) {
6094 first_vertical_edge_done = 0;
6097 if (h->deblocking_filter==2 && h->slice_table[mbm_xy] != h->slice_table[mb_xy])
6101 for( edge = start; edge < 4; edge++ ) {
6102 /* mbn_xy: neighbour macroblock */
6103 int mbn_xy = edge > 0 ? mb_xy : mbm_xy;
6107 if (h->mb_aff_frame && (dir == 1) && (edge == 0) && ((mb_y & 1) == 0)
6108 && !IS_INTERLACED(s->current_picture.mb_type[mb_xy])
6109 && IS_INTERLACED(s->current_picture.mb_type[mbn_xy])
6111 // This is a special case in the norm where the filtering must
6112 // be done twice (one each of the field) even if we are in a
6113 // frame macroblock.
6115 unsigned int tmp_linesize = 2 * linesize;
6116 unsigned int tmp_uvlinesize = 2 * uvlinesize;
6117 int mbn_xy = mb_xy - 2 * s->mb_stride;
6121 if( IS_INTRA( s->current_picture.mb_type[mb_xy] ) ||
6122 IS_INTRA( s->current_picture.mb_type[mbn_xy] ) ) {
6123 bS[0] = bS[1] = bS[2] = bS[3] = 3;
6129 // Do not use s->qscale as luma quantiser because it has not the same
6130 // value in IPCM macroblocks.
6131 qp = ( s->current_picture.qscale_table[mb_xy] + s->current_picture.qscale_table[mbn_xy] + 1 ) >> 1;
6132 tprintf("filter mb:%d/%d dir:%d edge:%d, QPy:%d ls:%d uvls:%d", mb_x, mb_y, dir, edge, qp, tmp_linesize, tmp_uvlinesize);
6133 { int i; for (i = 0; i < 4; i++) tprintf(" bS[%d]:%d", i, bS[i]); tprintf("\n"); }
6134 filter_mb_edgeh( h, &img_y[0], tmp_linesize, bS, qp );
6135 chroma_qp = ( h->chroma_qp +
6136 get_chroma_qp( h->pps.chroma_qp_index_offset, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1;
6137 filter_mb_edgech( h, &img_cb[0], tmp_uvlinesize, bS, chroma_qp );
6138 filter_mb_edgech( h, &img_cr[0], tmp_uvlinesize, bS, chroma_qp );
6141 mbn_xy += s->mb_stride;
6142 if( IS_INTRA( s->current_picture.mb_type[mb_xy] ) ||
6143 IS_INTRA( s->current_picture.mb_type[mbn_xy] ) ) {
6144 bS[0] = bS[1] = bS[2] = bS[3] = 3;
6150 // Do not use s->qscale as luma quantiser because it has not the same
6151 // value in IPCM macroblocks.
6152 qp = ( s->current_picture.qscale_table[mb_xy] + s->current_picture.qscale_table[mbn_xy] + 1 ) >> 1;
6153 tprintf("filter mb:%d/%d dir:%d edge:%d, QPy:%d ls:%d uvls:%d", mb_x, mb_y, dir, edge, qp, tmp_linesize, tmp_uvlinesize);
6154 { int i; for (i = 0; i < 4; i++) tprintf(" bS[%d]:%d", i, bS[i]); tprintf("\n"); }
6155 filter_mb_edgeh( h, &img_y[linesize], tmp_linesize, bS, qp );
6156 chroma_qp = ( h->chroma_qp +
6157 get_chroma_qp( h->pps.chroma_qp_index_offset, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1;
6158 filter_mb_edgech( h, &img_cb[uvlinesize], tmp_uvlinesize, bS, chroma_qp );
6159 filter_mb_edgech( h, &img_cr[uvlinesize], tmp_uvlinesize, bS, chroma_qp );
6162 if( IS_INTRA( s->current_picture.mb_type[mb_xy] ) ||
6163 IS_INTRA( s->current_picture.mb_type[mbn_xy] ) ) {
6166 if ( (!IS_INTERLACED(s->current_picture.mb_type[mb_xy]) && !IS_INTERLACED(s->current_picture.mb_type[mbm_xy]))
6167 || ((h->mb_aff_frame || (s->picture_structure != PICT_FRAME)) && (dir == 0))
6176 bS[0] = bS[1] = bS[2] = bS[3] = value;
6179 for( i = 0; i < 4; i++ ) {
6180 int x = dir == 0 ? edge : i;
6181 int y = dir == 0 ? i : edge;
6182 int b_idx= 8 + 4 + x + 8*y;
6183 int bn_idx= b_idx - (dir ? 8:1);
6185 if( h->non_zero_count_cache[b_idx] != 0 ||
6186 h->non_zero_count_cache[bn_idx] != 0 ) {
6191 /* FIXME: A given frame may occupy more than one position in
6192 * the reference list. So we should compare the frame numbers,
6193 * not the indices in the ref list. */
6196 for( l = 0; l < 1 + (h->slice_type == B_TYPE); l++ ) {
6197 if( h->ref_cache[l][b_idx] != h->ref_cache[l][bn_idx] ||
6198 ABS( h->mv_cache[l][b_idx][0] - h->mv_cache[l][bn_idx][0] ) >= 4 ||
6199 ABS( h->mv_cache[l][b_idx][1] - h->mv_cache[l][bn_idx][1] ) >= 4 ) {
6207 if(bS[0]+bS[1]+bS[2]+bS[3] == 0)
6212 // Do not use s->qscale as luma quantiser because it has not the same
6213 // value in IPCM macroblocks.
6214 qp = ( s->current_picture.qscale_table[mb_xy] + s->current_picture.qscale_table[mbn_xy] + 1 ) >> 1;
6215 //tprintf("filter mb:%d/%d dir:%d edge:%d, QPy:%d, QPc:%d, QPcn:%d\n", mb_x, mb_y, dir, edge, qp, h->chroma_qp, s->current_picture.qscale_table[mbn_xy]);
6216 tprintf("filter mb:%d/%d dir:%d edge:%d, QPy:%d ls:%d uvls:%d", mb_x, mb_y, dir, edge, qp, linesize, uvlinesize);
6217 { int i; for (i = 0; i < 4; i++) tprintf(" bS[%d]:%d", i, bS[i]); tprintf("\n"); }
6219 filter_mb_edgev( h, &img_y[4*edge], linesize, bS, qp );
6220 if( (edge&1) == 0 ) {
6221 int chroma_qp = ( h->chroma_qp +
6222 get_chroma_qp( h->pps.chroma_qp_index_offset, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1;
6223 filter_mb_edgecv( h, &img_cb[2*edge], uvlinesize, bS, chroma_qp );
6224 filter_mb_edgecv( h, &img_cr[2*edge], uvlinesize, bS, chroma_qp );
6227 filter_mb_edgeh( h, &img_y[4*edge*linesize], linesize, bS, qp );
6228 if( (edge&1) == 0 ) {
6229 int chroma_qp = ( h->chroma_qp +
6230 get_chroma_qp( h->pps.chroma_qp_index_offset, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1;
6231 filter_mb_edgech( h, &img_cb[2*edge*uvlinesize], uvlinesize, bS, chroma_qp );
6232 filter_mb_edgech( h, &img_cr[2*edge*uvlinesize], uvlinesize, bS, chroma_qp );
6239 static int decode_slice(H264Context *h){
6240 MpegEncContext * const s = &h->s;
6241 const int part_mask= s->partitioned_frame ? (AC_END|AC_ERROR) : 0x7F;
6245 if( h->pps.cabac ) {
6249 align_get_bits( &s->gb );
6252 ff_init_cabac_states( &h->cabac, ff_h264_lps_range, ff_h264_mps_state, ff_h264_lps_state, 64 );
6253 ff_init_cabac_decoder( &h->cabac,
6254 s->gb.buffer + get_bits_count(&s->gb)/8,
6255 ( s->gb.size_in_bits - get_bits_count(&s->gb) + 7)/8);
6256 /* calculate pre-state */
6257 for( i= 0; i < 399; i++ ) {
6259 if( h->slice_type == I_TYPE )
6260 pre = clip( ((cabac_context_init_I[i][0] * s->qscale) >>4 ) + cabac_context_init_I[i][1], 1, 126 );
6262 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 );
6265 h->cabac_state[i] = 2 * ( 63 - pre ) + 0;
6267 h->cabac_state[i] = 2 * ( pre - 64 ) + 1;
6271 int ret = decode_mb_cabac(h);
6274 if(ret>=0) hl_decode_mb(h);
6276 /* XXX: useless as decode_mb_cabac it doesn't support that ... */
6277 if( ret >= 0 && h->mb_aff_frame ) { //FIXME optimal? or let mb_decode decode 16x32 ?
6280 if(ret>=0) ret = decode_mb_cabac(h);
6285 eos = get_cabac_terminate( &h->cabac );
6287 if( ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 1) {
6288 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
6289 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);
6293 if( ++s->mb_x >= s->mb_width ) {
6295 ff_draw_horiz_band(s, 16*s->mb_y, 16);
6297 if(h->mb_aff_frame) {
6302 if( eos || s->mb_y >= s->mb_height ) {
6303 tprintf("slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
6304 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);
6311 int ret = decode_mb_cavlc(h);
6313 if(ret>=0) hl_decode_mb(h);
6315 if(ret>=0 && h->mb_aff_frame){ //FIXME optimal? or let mb_decode decode 16x32 ?
6317 ret = decode_mb_cavlc(h);
6319 if(ret>=0) hl_decode_mb(h);
6324 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
6325 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);
6330 if(++s->mb_x >= s->mb_width){
6332 ff_draw_horiz_band(s, 16*s->mb_y, 16);
6334 if(h->mb_aff_frame) {
6337 if(s->mb_y >= s->mb_height){
6338 tprintf("slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
6340 if(get_bits_count(&s->gb) == s->gb.size_in_bits ) {
6341 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);
6345 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);
6352 if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->mb_skip_run<=0){
6353 tprintf("slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
6354 if(get_bits_count(&s->gb) == s->gb.size_in_bits ){
6355 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);
6359 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);
6368 for(;s->mb_y < s->mb_height; s->mb_y++){
6369 for(;s->mb_x < s->mb_width; s->mb_x++){
6370 int ret= decode_mb(h);
6375 fprintf(stderr, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
6376 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);
6381 if(++s->mb_x >= s->mb_width){
6383 if(++s->mb_y >= s->mb_height){
6384 if(get_bits_count(s->gb) == s->gb.size_in_bits){
6385 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);
6389 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);
6396 if(get_bits_count(s->?gb) >= s->gb?.size_in_bits){
6397 if(get_bits_count(s->gb) == s->gb.size_in_bits){
6398 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);
6402 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);
6409 ff_draw_horiz_band(s, 16*s->mb_y, 16);
6412 return -1; //not reached
6415 static inline void decode_hrd_parameters(H264Context *h, SPS *sps){
6416 MpegEncContext * const s = &h->s;
6418 cpb_count = get_ue_golomb(&s->gb) + 1;
6419 get_bits(&s->gb, 4); /* bit_rate_scale */
6420 get_bits(&s->gb, 4); /* cpb_size_scale */
6421 for(i=0; i<cpb_count; i++){
6422 get_ue_golomb(&s->gb); /* bit_rate_value_minus1 */
6423 get_ue_golomb(&s->gb); /* cpb_size_value_minus1 */
6424 get_bits1(&s->gb); /* cbr_flag */
6426 get_bits(&s->gb, 5); /* initial_cpb_removal_delay_length_minus1 */
6427 get_bits(&s->gb, 5); /* cpb_removal_delay_length_minus1 */
6428 get_bits(&s->gb, 5); /* dpb_output_delay_length_minus1 */
6429 get_bits(&s->gb, 5); /* time_offset_length */
6432 static inline int decode_vui_parameters(H264Context *h, SPS *sps){
6433 MpegEncContext * const s = &h->s;
6434 int aspect_ratio_info_present_flag, aspect_ratio_idc;
6435 int nal_hrd_parameters_present_flag, vcl_hrd_parameters_present_flag;
6437 aspect_ratio_info_present_flag= get_bits1(&s->gb);
6439 if( aspect_ratio_info_present_flag ) {
6440 aspect_ratio_idc= get_bits(&s->gb, 8);
6441 if( aspect_ratio_idc == EXTENDED_SAR ) {
6442 sps->sar.num= get_bits(&s->gb, 16);
6443 sps->sar.den= get_bits(&s->gb, 16);
6444 }else if(aspect_ratio_idc < 16){
6445 sps->sar= pixel_aspect[aspect_ratio_idc];
6447 av_log(h->s.avctx, AV_LOG_ERROR, "illegal aspect ratio\n");
6454 // s->avctx->aspect_ratio= sar_width*s->width / (float)(s->height*sar_height);
6456 if(get_bits1(&s->gb)){ /* overscan_info_present_flag */
6457 get_bits1(&s->gb); /* overscan_appropriate_flag */
6460 if(get_bits1(&s->gb)){ /* video_signal_type_present_flag */
6461 get_bits(&s->gb, 3); /* video_format */
6462 get_bits1(&s->gb); /* video_full_range_flag */
6463 if(get_bits1(&s->gb)){ /* colour_description_present_flag */
6464 get_bits(&s->gb, 8); /* colour_primaries */
6465 get_bits(&s->gb, 8); /* transfer_characteristics */
6466 get_bits(&s->gb, 8); /* matrix_coefficients */
6470 if(get_bits1(&s->gb)){ /* chroma_location_info_present_flag */
6471 get_ue_golomb(&s->gb); /* chroma_sample_location_type_top_field */
6472 get_ue_golomb(&s->gb); /* chroma_sample_location_type_bottom_field */
6475 sps->timing_info_present_flag = get_bits1(&s->gb);
6476 if(sps->timing_info_present_flag){
6477 sps->num_units_in_tick = get_bits_long(&s->gb, 32);
6478 sps->time_scale = get_bits_long(&s->gb, 32);
6479 sps->fixed_frame_rate_flag = get_bits1(&s->gb);
6482 nal_hrd_parameters_present_flag = get_bits1(&s->gb);
6483 if(nal_hrd_parameters_present_flag)
6484 decode_hrd_parameters(h, sps);
6485 vcl_hrd_parameters_present_flag = get_bits1(&s->gb);
6486 if(vcl_hrd_parameters_present_flag)
6487 decode_hrd_parameters(h, sps);
6488 if(nal_hrd_parameters_present_flag || vcl_hrd_parameters_present_flag)
6489 get_bits1(&s->gb); /* low_delay_hrd_flag */
6490 get_bits1(&s->gb); /* pic_struct_present_flag */
6492 sps->bitstream_restriction_flag = get_bits1(&s->gb);
6493 if(sps->bitstream_restriction_flag){
6494 get_bits1(&s->gb); /* motion_vectors_over_pic_boundaries_flag */
6495 get_ue_golomb(&s->gb); /* max_bytes_per_pic_denom */
6496 get_ue_golomb(&s->gb); /* max_bits_per_mb_denom */
6497 get_ue_golomb(&s->gb); /* log2_max_mv_length_horizontal */
6498 get_ue_golomb(&s->gb); /* log2_max_mv_length_vertical */
6499 sps->num_reorder_frames = get_ue_golomb(&s->gb);
6500 get_ue_golomb(&s->gb); /* max_dec_frame_buffering */
6506 static inline int decode_seq_parameter_set(H264Context *h){
6507 MpegEncContext * const s = &h->s;
6508 int profile_idc, level_idc;
6512 profile_idc= get_bits(&s->gb, 8);
6513 get_bits1(&s->gb); //constraint_set0_flag
6514 get_bits1(&s->gb); //constraint_set1_flag
6515 get_bits1(&s->gb); //constraint_set2_flag
6516 get_bits1(&s->gb); //constraint_set3_flag
6517 get_bits(&s->gb, 4); // reserved
6518 level_idc= get_bits(&s->gb, 8);
6519 sps_id= get_ue_golomb(&s->gb);
6521 sps= &h->sps_buffer[ sps_id ];
6522 sps->profile_idc= profile_idc;
6523 sps->level_idc= level_idc;
6525 sps->log2_max_frame_num= get_ue_golomb(&s->gb) + 4;
6526 sps->poc_type= get_ue_golomb(&s->gb);
6528 if(sps->poc_type == 0){ //FIXME #define
6529 sps->log2_max_poc_lsb= get_ue_golomb(&s->gb) + 4;
6530 } else if(sps->poc_type == 1){//FIXME #define
6531 sps->delta_pic_order_always_zero_flag= get_bits1(&s->gb);
6532 sps->offset_for_non_ref_pic= get_se_golomb(&s->gb);
6533 sps->offset_for_top_to_bottom_field= get_se_golomb(&s->gb);
6534 sps->poc_cycle_length= get_ue_golomb(&s->gb);
6536 for(i=0; i<sps->poc_cycle_length; i++)
6537 sps->offset_for_ref_frame[i]= get_se_golomb(&s->gb);
6539 if(sps->poc_type > 2){
6540 av_log(h->s.avctx, AV_LOG_ERROR, "illegal POC type %d\n", sps->poc_type);
6544 sps->ref_frame_count= get_ue_golomb(&s->gb);
6545 if(sps->ref_frame_count > MAX_PICTURE_COUNT-2){
6546 av_log(h->s.avctx, AV_LOG_ERROR, "too many reference frames\n");
6548 sps->gaps_in_frame_num_allowed_flag= get_bits1(&s->gb);
6549 sps->mb_width= get_ue_golomb(&s->gb) + 1;
6550 sps->mb_height= get_ue_golomb(&s->gb) + 1;
6551 if((unsigned)sps->mb_width >= INT_MAX/16 || (unsigned)sps->mb_height >= INT_MAX/16 ||
6552 avcodec_check_dimensions(NULL, 16*sps->mb_width, 16*sps->mb_height))
6555 sps->frame_mbs_only_flag= get_bits1(&s->gb);
6556 if(!sps->frame_mbs_only_flag)
6557 sps->mb_aff= get_bits1(&s->gb);
6561 sps->direct_8x8_inference_flag= get_bits1(&s->gb);
6563 sps->crop= get_bits1(&s->gb);
6565 sps->crop_left = get_ue_golomb(&s->gb);
6566 sps->crop_right = get_ue_golomb(&s->gb);
6567 sps->crop_top = get_ue_golomb(&s->gb);
6568 sps->crop_bottom= get_ue_golomb(&s->gb);
6569 if(sps->crop_left || sps->crop_top){
6570 av_log(h->s.avctx, AV_LOG_ERROR, "insane cropping not completly supported, this could look slightly wrong ...\n");
6576 sps->crop_bottom= 0;
6579 sps->vui_parameters_present_flag= get_bits1(&s->gb);
6580 if( sps->vui_parameters_present_flag )
6581 decode_vui_parameters(h, sps);
6583 if(s->avctx->debug&FF_DEBUG_PICT_INFO){
6584 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",
6585 sps_id, sps->profile_idc, sps->level_idc,
6587 sps->ref_frame_count,
6588 sps->mb_width, sps->mb_height,
6589 sps->frame_mbs_only_flag ? "FRM" : (sps->mb_aff ? "MB-AFF" : "PIC-AFF"),
6590 sps->direct_8x8_inference_flag ? "8B8" : "",
6591 sps->crop_left, sps->crop_right,
6592 sps->crop_top, sps->crop_bottom,
6593 sps->vui_parameters_present_flag ? "VUI" : ""
6599 static inline int decode_picture_parameter_set(H264Context *h){
6600 MpegEncContext * const s = &h->s;
6601 int pps_id= get_ue_golomb(&s->gb);
6602 PPS *pps= &h->pps_buffer[pps_id];
6604 pps->sps_id= get_ue_golomb(&s->gb);
6605 pps->cabac= get_bits1(&s->gb);
6606 pps->pic_order_present= get_bits1(&s->gb);
6607 pps->slice_group_count= get_ue_golomb(&s->gb) + 1;
6608 if(pps->slice_group_count > 1 ){
6609 pps->mb_slice_group_map_type= get_ue_golomb(&s->gb);
6610 av_log(h->s.avctx, AV_LOG_ERROR, "FMO not supported\n");
6611 switch(pps->mb_slice_group_map_type){
6614 | for( i = 0; i <= num_slice_groups_minus1; i++ ) | | |
6615 | run_length[ i ] |1 |ue(v) |
6620 | for( i = 0; i < num_slice_groups_minus1; i++ ) | | |
6622 | top_left_mb[ i ] |1 |ue(v) |
6623 | bottom_right_mb[ i ] |1 |ue(v) |
6631 | slice_group_change_direction_flag |1 |u(1) |
6632 | slice_group_change_rate_minus1 |1 |ue(v) |
6637 | slice_group_id_cnt_minus1 |1 |ue(v) |
6638 | for( i = 0; i <= slice_group_id_cnt_minus1; i++ | | |
6640 | slice_group_id[ i ] |1 |u(v) |
6645 pps->ref_count[0]= get_ue_golomb(&s->gb) + 1;
6646 pps->ref_count[1]= get_ue_golomb(&s->gb) + 1;
6647 if(pps->ref_count[0] > 32 || pps->ref_count[1] > 32){
6648 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow (pps)\n");
6652 pps->weighted_pred= get_bits1(&s->gb);
6653 pps->weighted_bipred_idc= get_bits(&s->gb, 2);
6654 pps->init_qp= get_se_golomb(&s->gb) + 26;
6655 pps->init_qs= get_se_golomb(&s->gb) + 26;
6656 pps->chroma_qp_index_offset= get_se_golomb(&s->gb);
6657 pps->deblocking_filter_parameters_present= get_bits1(&s->gb);
6658 pps->constrained_intra_pred= get_bits1(&s->gb);
6659 pps->redundant_pic_cnt_present = get_bits1(&s->gb);
6661 if(s->avctx->debug&FF_DEBUG_PICT_INFO){
6662 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",
6663 pps_id, pps->sps_id,
6664 pps->cabac ? "CABAC" : "CAVLC",
6665 pps->slice_group_count,
6666 pps->ref_count[0], pps->ref_count[1],
6667 pps->weighted_pred ? "weighted" : "",
6668 pps->init_qp, pps->init_qs, pps->chroma_qp_index_offset,
6669 pps->deblocking_filter_parameters_present ? "LPAR" : "",
6670 pps->constrained_intra_pred ? "CONSTR" : "",
6671 pps->redundant_pic_cnt_present ? "REDU" : ""
6679 * finds the end of the current frame in the bitstream.
6680 * @return the position of the first byte of the next frame, or -1
6682 static int find_frame_end(H264Context *h, const uint8_t *buf, int buf_size){
6685 ParseContext *pc = &(h->s.parse_context);
6686 //printf("first %02X%02X%02X%02X\n", buf[0], buf[1],buf[2],buf[3]);
6687 // mb_addr= pc->mb_addr - 1;
6689 for(i=0; i<=buf_size; i++){
6690 if((state&0xFFFFFF1F) == 0x101 || (state&0xFFFFFF1F) == 0x102 || (state&0xFFFFFF1F) == 0x105){
6691 tprintf("find_frame_end new startcode = %08x, frame_start_found = %d, pos = %d\n", state, pc->frame_start_found, i);
6692 if(pc->frame_start_found){
6693 // If there isn't one more byte in the buffer
6694 // the test on first_mb_in_slice cannot be done yet
6695 // do it at next call.
6696 if (i >= buf_size) break;
6697 if (buf[i] & 0x80) {
6698 // first_mb_in_slice is 0, probably the first nal of a new
6700 tprintf("find_frame_end frame_end_found, state = %08x, pos = %d\n", state, i);
6702 pc->frame_start_found= 0;
6706 pc->frame_start_found = 1;
6709 state= (state<<8) | buf[i];
6713 return END_NOT_FOUND;
6716 static int h264_parse(AVCodecParserContext *s,
6717 AVCodecContext *avctx,
6718 uint8_t **poutbuf, int *poutbuf_size,
6719 const uint8_t *buf, int buf_size)
6721 H264Context *h = s->priv_data;
6722 ParseContext *pc = &h->s.parse_context;
6725 next= find_frame_end(h, buf, buf_size);
6727 if (ff_combine_frame(pc, next, (uint8_t **)&buf, &buf_size) < 0) {
6733 *poutbuf = (uint8_t *)buf;
6734 *poutbuf_size = buf_size;
6738 static int decode_nal_units(H264Context *h, uint8_t *buf, int buf_size){
6739 MpegEncContext * const s = &h->s;
6740 AVCodecContext * const avctx= s->avctx;
6744 for(i=0; i<32; i++){
6745 printf("%X ", buf[i]);
6757 if(buf_index >= buf_size) break;
6759 for(i = 0; i < h->nal_length_size; i++)
6760 nalsize = (nalsize << 8) | buf[buf_index++];
6762 // start code prefix search
6763 for(; buf_index + 3 < buf_size; buf_index++){
6764 // this should allways succeed in the first iteration
6765 if(buf[buf_index] == 0 && buf[buf_index+1] == 0 && buf[buf_index+2] == 1)
6769 if(buf_index+3 >= buf_size) break;
6774 ptr= decode_nal(h, buf + buf_index, &dst_length, &consumed, h->is_avc ? nalsize : buf_size - buf_index);
6775 if(ptr[dst_length - 1] == 0) dst_length--;
6776 bit_length= 8*dst_length - decode_rbsp_trailing(ptr + dst_length - 1);
6778 if(s->avctx->debug&FF_DEBUG_STARTCODE){
6779 av_log(h->s.avctx, AV_LOG_DEBUG, "NAL %d at %d/%d length %d\n", h->nal_unit_type, buf_index, buf_size, dst_length);
6782 if (h->is_avc && (nalsize != consumed))
6783 av_log(h->s.avctx, AV_LOG_ERROR, "AVC: Consumed only %d bytes instead of %d\n", consumed, nalsize);
6785 buf_index += consumed;
6787 if( s->hurry_up == 1 && h->nal_ref_idc == 0 )
6790 switch(h->nal_unit_type){
6792 idr(h); //FIXME ensure we dont loose some frames if there is reordering
6794 init_get_bits(&s->gb, ptr, bit_length);
6796 h->inter_gb_ptr= &s->gb;
6797 s->data_partitioning = 0;
6799 if(decode_slice_header(h) < 0) return -1;
6800 if(h->redundant_pic_count==0 && s->hurry_up < 5 )
6804 init_get_bits(&s->gb, ptr, bit_length);
6806 h->inter_gb_ptr= NULL;
6807 s->data_partitioning = 1;
6809 if(decode_slice_header(h) < 0) return -1;
6812 init_get_bits(&h->intra_gb, ptr, bit_length);
6813 h->intra_gb_ptr= &h->intra_gb;
6816 init_get_bits(&h->inter_gb, ptr, bit_length);
6817 h->inter_gb_ptr= &h->inter_gb;
6819 if(h->redundant_pic_count==0 && h->intra_gb_ptr && s->data_partitioning && s->hurry_up < 5 )
6825 init_get_bits(&s->gb, ptr, bit_length);
6826 decode_seq_parameter_set(h);
6828 if(s->flags& CODEC_FLAG_LOW_DELAY)
6831 if(avctx->has_b_frames < 2)
6832 avctx->has_b_frames= !s->low_delay;
6835 init_get_bits(&s->gb, ptr, bit_length);
6837 decode_picture_parameter_set(h);
6840 case NAL_PICTURE_DELIMITER:
6842 case NAL_FILTER_DATA:
6845 av_log(avctx, AV_LOG_ERROR, "Unknown NAL code: %d\n", h->nal_unit_type);
6849 if(!s->current_picture_ptr) return buf_index; //no frame
6851 s->current_picture_ptr->pict_type= s->pict_type;
6852 s->current_picture_ptr->key_frame= s->pict_type == I_TYPE && h->nal_unit_type == NAL_IDR_SLICE;
6854 h->prev_frame_num_offset= h->frame_num_offset;
6855 h->prev_frame_num= h->frame_num;
6856 if(s->current_picture_ptr->reference){
6857 h->prev_poc_msb= h->poc_msb;
6858 h->prev_poc_lsb= h->poc_lsb;
6860 if(s->current_picture_ptr->reference)
6861 execute_ref_pic_marking(h, h->mmco, h->mmco_index);
6871 * retunrs the number of bytes consumed for building the current frame
6873 static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size){
6874 if(s->flags&CODEC_FLAG_TRUNCATED){
6875 pos -= s->parse_context.last_index;
6876 if(pos<0) pos=0; // FIXME remove (uneeded?)
6880 if(pos==0) pos=1; //avoid infinite loops (i doubt thats needed but ...)
6881 if(pos+10>buf_size) pos=buf_size; // oops ;)
6887 static int decode_frame(AVCodecContext *avctx,
6888 void *data, int *data_size,
6889 uint8_t *buf, int buf_size)
6891 H264Context *h = avctx->priv_data;
6892 MpegEncContext *s = &h->s;
6893 AVFrame *pict = data;
6896 s->flags= avctx->flags;
6897 s->flags2= avctx->flags2;
6899 /* no supplementary picture */
6900 if (buf_size == 0) {
6904 if(s->flags&CODEC_FLAG_TRUNCATED){
6905 int next= find_frame_end(h, buf, buf_size);
6907 if( ff_combine_frame(&s->parse_context, next, &buf, &buf_size) < 0 )
6909 //printf("next:%d buf_size:%d last_index:%d\n", next, buf_size, s->parse_context.last_index);
6912 if(h->is_avc && !h->got_avcC) {
6913 int i, cnt, nalsize;
6914 unsigned char *p = avctx->extradata;
6915 if(avctx->extradata_size < 7) {
6916 av_log(avctx, AV_LOG_ERROR, "avcC too short\n");
6920 av_log(avctx, AV_LOG_ERROR, "Unknown avcC version %d\n", *p);
6923 /* sps and pps in the avcC always have length coded with 2 bytes,
6924 so put a fake nal_length_size = 2 while parsing them */
6925 h->nal_length_size = 2;
6926 // Decode sps from avcC
6927 cnt = *(p+5) & 0x1f; // Number of sps
6929 for (i = 0; i < cnt; i++) {
6930 nalsize = BE_16(p) + 2;
6931 if(decode_nal_units(h, p, nalsize) != nalsize) {
6932 av_log(avctx, AV_LOG_ERROR, "Decoding sps %d from avcC failed\n", i);
6937 // Decode pps from avcC
6938 cnt = *(p++); // Number of pps
6939 for (i = 0; i < cnt; i++) {
6940 nalsize = BE_16(p) + 2;
6941 if(decode_nal_units(h, p, nalsize) != nalsize) {
6942 av_log(avctx, AV_LOG_ERROR, "Decoding pps %d from avcC failed\n", i);
6947 // Now store right nal length size, that will be use to parse all other nals
6948 h->nal_length_size = ((*(((char*)(avctx->extradata))+4))&0x03)+1;
6949 // Do not reparse avcC
6953 if(!h->is_avc && s->avctx->extradata_size && s->picture_number==0){
6954 if(0 < decode_nal_units(h, s->avctx->extradata, s->avctx->extradata_size) )
6958 buf_index=decode_nal_units(h, buf, buf_size);
6962 //FIXME do something with unavailable reference frames
6964 // if(ret==FRAME_SKIPED) return get_consumed_bytes(s, buf_index, buf_size);
6965 if(!s->current_picture_ptr){
6966 av_log(h->s.avctx, AV_LOG_DEBUG, "error, NO frame\n");
6971 Picture *out = s->current_picture_ptr;
6972 #if 0 //decode order
6973 *data_size = sizeof(AVFrame);
6975 /* Sort B-frames into display order */
6976 Picture *cur = s->current_picture_ptr;
6977 Picture *prev = h->delayed_output_pic;
6982 int dropped_frame = 0;
6985 if(h->sps.bitstream_restriction_flag
6986 && s->avctx->has_b_frames < h->sps.num_reorder_frames){
6987 s->avctx->has_b_frames = h->sps.num_reorder_frames;
6991 while(h->delayed_pic[pics]) pics++;
6992 h->delayed_pic[pics++] = cur;
6993 if(cur->reference == 0)
6996 for(i=0; h->delayed_pic[i]; i++)
6997 if(h->delayed_pic[i]->key_frame)
7000 out = h->delayed_pic[0];
7001 for(i=1; h->delayed_pic[i] && !h->delayed_pic[i]->key_frame; i++)
7002 if(h->delayed_pic[i]->poc < out->poc){
7003 out = h->delayed_pic[i];
7007 out_of_order = !cross_idr && prev && out->poc < prev->poc;
7008 if(prev && pics <= s->avctx->has_b_frames)
7010 else if((out_of_order && pics-1 == s->avctx->has_b_frames)
7012 ((!cross_idr && prev && out->poc > prev->poc + 2)
7013 || cur->pict_type == B_TYPE)))
7016 s->avctx->has_b_frames++;
7019 else if(out_of_order)
7022 if(out_of_order || pics > s->avctx->has_b_frames){
7023 dropped_frame = (out != h->delayed_pic[out_idx]);
7024 for(i=out_idx; h->delayed_pic[i]; i++)
7025 h->delayed_pic[i] = h->delayed_pic[i+1];
7028 if(prev == out && !dropped_frame)
7031 *data_size = sizeof(AVFrame);
7032 if(prev && prev != out && prev->reference == 1)
7033 prev->reference = 0;
7034 h->delayed_output_pic = out;
7037 *pict= *(AVFrame*)out;
7040 assert(pict->data[0]);
7041 ff_print_debug_info(s, pict);
7042 //printf("out %d\n", (int)pict->data[0]);
7045 /* Return the Picture timestamp as the frame number */
7046 /* we substract 1 because it is added on utils.c */
7047 avctx->frame_number = s->picture_number - 1;
7049 return get_consumed_bytes(s, buf_index, buf_size);
7052 static inline void fill_mb_avail(H264Context *h){
7053 MpegEncContext * const s = &h->s;
7054 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
7057 h->mb_avail[0]= s->mb_x && h->slice_table[mb_xy - s->mb_stride - 1] == h->slice_num;
7058 h->mb_avail[1]= h->slice_table[mb_xy - s->mb_stride ] == h->slice_num;
7059 h->mb_avail[2]= s->mb_x+1 < s->mb_width && h->slice_table[mb_xy - s->mb_stride + 1] == h->slice_num;
7065 h->mb_avail[3]= s->mb_x && h->slice_table[mb_xy - 1] == h->slice_num;
7066 h->mb_avail[4]= 1; //FIXME move out
7067 h->mb_avail[5]= 0; //FIXME move out
7073 #define SIZE (COUNT*40)
7079 // int int_temp[10000];
7081 AVCodecContext avctx;
7083 dsputil_init(&dsp, &avctx);
7085 init_put_bits(&pb, temp, SIZE);
7086 printf("testing unsigned exp golomb\n");
7087 for(i=0; i<COUNT; i++){
7089 set_ue_golomb(&pb, i);
7090 STOP_TIMER("set_ue_golomb");
7092 flush_put_bits(&pb);
7094 init_get_bits(&gb, temp, 8*SIZE);
7095 for(i=0; i<COUNT; i++){
7098 s= show_bits(&gb, 24);
7101 j= get_ue_golomb(&gb);
7103 printf("missmatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
7106 STOP_TIMER("get_ue_golomb");
7110 init_put_bits(&pb, temp, SIZE);
7111 printf("testing signed exp golomb\n");
7112 for(i=0; i<COUNT; i++){
7114 set_se_golomb(&pb, i - COUNT/2);
7115 STOP_TIMER("set_se_golomb");
7117 flush_put_bits(&pb);
7119 init_get_bits(&gb, temp, 8*SIZE);
7120 for(i=0; i<COUNT; i++){
7123 s= show_bits(&gb, 24);
7126 j= get_se_golomb(&gb);
7127 if(j != i - COUNT/2){
7128 printf("missmatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
7131 STOP_TIMER("get_se_golomb");
7134 printf("testing 4x4 (I)DCT\n");
7137 uint8_t src[16], ref[16];
7138 uint64_t error= 0, max_error=0;
7140 for(i=0; i<COUNT; i++){
7142 // printf("%d %d %d\n", r1, r2, (r2-r1)*16);
7143 for(j=0; j<16; j++){
7144 ref[j]= random()%255;
7145 src[j]= random()%255;
7148 h264_diff_dct_c(block, src, ref, 4);
7151 for(j=0; j<16; j++){
7152 // printf("%d ", block[j]);
7153 block[j]= block[j]*4;
7154 if(j&1) block[j]= (block[j]*4 + 2)/5;
7155 if(j&4) block[j]= (block[j]*4 + 2)/5;
7159 s->dsp.h264_idct_add(ref, block, 4);
7160 /* for(j=0; j<16; j++){
7161 printf("%d ", ref[j]);
7165 for(j=0; j<16; j++){
7166 int diff= ABS(src[j] - ref[j]);
7169 max_error= FFMAX(max_error, diff);
7172 printf("error=%f max_error=%d\n", ((float)error)/COUNT/16, (int)max_error );
7174 printf("testing quantizer\n");
7175 for(qp=0; qp<52; qp++){
7177 src1_block[i]= src2_block[i]= random()%255;
7181 printf("Testing NAL layer\n");
7183 uint8_t bitstream[COUNT];
7184 uint8_t nal[COUNT*2];
7186 memset(&h, 0, sizeof(H264Context));
7188 for(i=0; i<COUNT; i++){
7196 for(j=0; j<COUNT; j++){
7197 bitstream[j]= (random() % 255) + 1;
7200 for(j=0; j<zeros; j++){
7201 int pos= random() % COUNT;
7202 while(bitstream[pos] == 0){
7211 nal_length= encode_nal(&h, nal, bitstream, COUNT, COUNT*2);
7213 printf("encoding failed\n");
7217 out= decode_nal(&h, nal, &out_length, &consumed, nal_length);
7221 if(out_length != COUNT){
7222 printf("incorrect length %d %d\n", out_length, COUNT);
7226 if(consumed != nal_length){
7227 printf("incorrect consumed length %d %d\n", nal_length, consumed);
7231 if(memcmp(bitstream, out, COUNT)){
7232 printf("missmatch\n");
7237 printf("Testing RBSP\n");
7245 static int decode_end(AVCodecContext *avctx)
7247 H264Context *h = avctx->priv_data;
7248 MpegEncContext *s = &h->s;
7250 free_tables(h); //FIXME cleanup init stuff perhaps
7253 // memset(h, 0, sizeof(H264Context));
7259 AVCodec h264_decoder = {
7263 sizeof(H264Context),
7268 /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 | CODEC_CAP_TRUNCATED | CODEC_CAP_DELAY,
7271 AVCodecParser h264_parser = {
7273 sizeof(H264Context),