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 short offset_for_ref_frame[256]; //FIXME dyn aloc?
89 * Picture parameter set
93 int cabac; ///< entropy_coding_mode_flag
94 int pic_order_present; ///< pic_order_present_flag
95 int slice_group_count; ///< num_slice_groups_minus1 + 1
96 int mb_slice_group_map_type;
97 int ref_count[2]; ///< num_ref_idx_l0/1_active_minus1 + 1
98 int weighted_pred; ///< weighted_pred_flag
99 int weighted_bipred_idc;
100 int init_qp; ///< pic_init_qp_minus26 + 26
101 int init_qs; ///< pic_init_qs_minus26 + 26
102 int chroma_qp_index_offset;
103 int deblocking_filter_parameters_present; ///< deblocking_filter_parameters_present_flag
104 int constrained_intra_pred; ///< constrained_intra_pred_flag
105 int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag
109 * Memory management control operation opcode.
111 typedef enum MMCOOpcode{
122 * Memory management control operation.
133 typedef struct H264Context{
141 #define NAL_IDR_SLICE 5
145 #define NAL_PICTURE_DELIMITER 9
146 #define NAL_FILTER_DATA 10
147 uint8_t *rbsp_buffer;
148 int rbsp_buffer_size;
152 int prev_mb_skiped; //FIXME remove (IMHO not used)
155 int chroma_pred_mode;
156 int intra16x16_pred_mode;
158 int8_t intra4x4_pred_mode_cache[5*8];
159 int8_t (*intra4x4_pred_mode)[8];
160 void (*pred4x4 [9+3])(uint8_t *src, uint8_t *topright, int stride);//FIXME move to dsp?
161 void (*pred8x8 [4+3])(uint8_t *src, int stride);
162 void (*pred16x16[4+3])(uint8_t *src, int stride);
163 unsigned int topleft_samples_available;
164 unsigned int top_samples_available;
165 unsigned int topright_samples_available;
166 unsigned int left_samples_available;
167 uint8_t (*top_border)[16+2*8];
168 uint8_t left_border[17+2*9];
171 * non zero coeff count cache.
172 * is 64 if not available.
174 uint8_t non_zero_count_cache[6*8];
175 uint8_t (*non_zero_count)[16];
178 * Motion vector cache.
180 int16_t mv_cache[2][5*8][2];
181 int8_t ref_cache[2][5*8];
182 #define LIST_NOT_USED -1 //FIXME rename?
183 #define PART_NOT_AVAILABLE -2
186 * is 1 if the specific list MV&references are set to 0,0,-2.
188 int mv_cache_clean[2];
190 int block_offset[16+8];
191 int chroma_subblock_offset[16]; //FIXME remove
193 uint16_t *mb2b_xy; //FIXME are these 4 a good idea?
201 int unknown_svq3_flag;
202 int next_slice_index;
204 SPS sps_buffer[MAX_SPS_COUNT];
205 SPS sps; ///< current sps
207 PPS pps_buffer[MAX_PPS_COUNT];
211 PPS pps; //FIXME move tp Picture perhaps? (->no) do we need that?
214 uint8_t *slice_table_base;
215 uint8_t *slice_table; ///< slice_table_base + mb_stride + 1
217 int slice_type_fixed;
219 //interlacing specific flags
220 int mb_field_decoding_flag;
227 int delta_poc_bottom;
230 int prev_poc_msb; ///< poc_msb of the last reference pic for POC type 0
231 int prev_poc_lsb; ///< poc_lsb of the last reference pic for POC type 0
232 int frame_num_offset; ///< for POC type 2
233 int prev_frame_num_offset; ///< for POC type 2
234 int prev_frame_num; ///< frame_num of the last pic for POC type 1/2
237 * frame_num for frames or 2*frame_num for field pics.
242 * max_frame_num or 2*max_frame_num for field pics.
246 //Weighted pred stuff
247 int luma_log2_weight_denom;
248 int chroma_log2_weight_denom;
249 int luma_weight[2][16];
250 int luma_offset[2][16];
251 int chroma_weight[2][16][2];
252 int chroma_offset[2][16][2];
255 int deblocking_filter; ///< disable_deblocking_filter_idc with 1<->0
256 int slice_alpha_c0_offset;
257 int slice_beta_offset;
259 int redundant_pic_count;
261 int direct_spatial_mv_pred;
264 * num_ref_idx_l0/1_active_minus1 + 1
266 int ref_count[2];// FIXME split for AFF
267 Picture *short_ref[16];
268 Picture *long_ref[16];
269 Picture default_ref_list[2][32];
270 Picture ref_list[2][32]; //FIXME size?
271 Picture field_ref_list[2][32]; //FIXME size?
274 * memory management control operations buffer.
276 MMCO mmco[MAX_MMCO_COUNT];
279 int long_ref_count; ///< number of actual long term references
280 int short_ref_count; ///< number of actual short term references
283 GetBitContext intra_gb;
284 GetBitContext inter_gb;
285 GetBitContext *intra_gb_ptr;
286 GetBitContext *inter_gb_ptr;
288 DCTELEM mb[16*24] __align8;
294 uint8_t cabac_state[399];
297 /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0,1,2), 0x0? luma_cbp */
299 /* chroma_pred_mode for i4x4 or i16x16, else 0 */
300 uint8_t *chroma_pred_mode_table;
301 int last_qscale_diff;
302 int16_t (*mvd_table[2])[2];
303 int16_t mvd_cache[2][5*8][2];
307 static VLC coeff_token_vlc[4];
308 static VLC chroma_dc_coeff_token_vlc;
310 static VLC total_zeros_vlc[15];
311 static VLC chroma_dc_total_zeros_vlc[3];
313 static VLC run_vlc[6];
316 static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
317 static void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
318 static void filter_mb( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr);
320 static inline uint32_t pack16to32(int a, int b){
321 #ifdef WORDS_BIGENDIAN
322 return (b&0xFFFF) + (a<<16);
324 return (a&0xFFFF) + (b<<16);
330 * @param h height of the recatangle, should be a constant
331 * @param w width of the recatangle, should be a constant
332 * @param size the size of val (1 or 4), should be a constant
334 static inline void fill_rectangle(void *vp, int w, int h, int stride, uint32_t val, int size){ //FIXME ensure this IS inlined
335 uint8_t *p= (uint8_t*)vp;
336 assert(size==1 || size==4);
341 //FIXME check what gcc generates for 64 bit on x86 and possible write a 32 bit ver of it
344 *(uint16_t*)(p + stride)= size==4 ? val : val*0x0101;
345 }else if(w==2 && h==4){
346 *(uint16_t*)(p + 0*stride)=
347 *(uint16_t*)(p + 1*stride)=
348 *(uint16_t*)(p + 2*stride)=
349 *(uint16_t*)(p + 3*stride)= size==4 ? val : val*0x0101;
350 }else if(w==4 && h==1){
351 *(uint32_t*)(p + 0*stride)= size==4 ? val : val*0x01010101;
352 }else if(w==4 && h==2){
353 *(uint32_t*)(p + 0*stride)=
354 *(uint32_t*)(p + 1*stride)= size==4 ? val : val*0x01010101;
355 }else if(w==4 && h==4){
356 *(uint32_t*)(p + 0*stride)=
357 *(uint32_t*)(p + 1*stride)=
358 *(uint32_t*)(p + 2*stride)=
359 *(uint32_t*)(p + 3*stride)= size==4 ? val : val*0x01010101;
360 }else if(w==8 && h==1){
362 *(uint32_t*)(p + 4)= size==4 ? val : val*0x01010101;
363 }else if(w==8 && h==2){
364 *(uint32_t*)(p + 0 + 0*stride)=
365 *(uint32_t*)(p + 4 + 0*stride)=
366 *(uint32_t*)(p + 0 + 1*stride)=
367 *(uint32_t*)(p + 4 + 1*stride)= size==4 ? val : val*0x01010101;
368 }else if(w==8 && h==4){
369 *(uint64_t*)(p + 0*stride)=
370 *(uint64_t*)(p + 1*stride)=
371 *(uint64_t*)(p + 2*stride)=
372 *(uint64_t*)(p + 3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
373 }else if(w==16 && h==2){
374 *(uint64_t*)(p + 0+0*stride)=
375 *(uint64_t*)(p + 8+0*stride)=
376 *(uint64_t*)(p + 0+1*stride)=
377 *(uint64_t*)(p + 8+1*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
378 }else if(w==16 && h==4){
379 *(uint64_t*)(p + 0+0*stride)=
380 *(uint64_t*)(p + 8+0*stride)=
381 *(uint64_t*)(p + 0+1*stride)=
382 *(uint64_t*)(p + 8+1*stride)=
383 *(uint64_t*)(p + 0+2*stride)=
384 *(uint64_t*)(p + 8+2*stride)=
385 *(uint64_t*)(p + 0+3*stride)=
386 *(uint64_t*)(p + 8+3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
391 static inline void fill_caches(H264Context *h, int mb_type){
392 MpegEncContext * const s = &h->s;
393 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
394 int topleft_xy, top_xy, topright_xy, left_xy[2];
395 int topleft_type, top_type, topright_type, left_type[2];
399 //wow what a mess, why didnt they simplify the interlacing&intra stuff, i cant imagine that these complex rules are worth it
403 topleft_xy = 0; /* avoid warning */
404 top_xy = 0; /* avoid warning */
405 topright_xy = 0; /* avoid warning */
407 topleft_xy = mb_xy-1 - s->mb_stride;
408 top_xy = mb_xy - s->mb_stride;
409 topright_xy= mb_xy+1 - s->mb_stride;
410 left_xy[0] = mb_xy-1;
411 left_xy[1] = mb_xy-1;
418 topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
419 top_type = h->slice_table[top_xy ] == h->slice_num ? s->current_picture.mb_type[top_xy] : 0;
420 topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
421 left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
422 left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
424 if(IS_INTRA(mb_type)){
425 h->topleft_samples_available=
426 h->top_samples_available=
427 h->left_samples_available= 0xFFFF;
428 h->topright_samples_available= 0xEEEA;
430 if(!IS_INTRA(top_type) && (top_type==0 || h->pps.constrained_intra_pred)){
431 h->topleft_samples_available= 0xB3FF;
432 h->top_samples_available= 0x33FF;
433 h->topright_samples_available= 0x26EA;
436 if(!IS_INTRA(left_type[i]) && (left_type[i]==0 || h->pps.constrained_intra_pred)){
437 h->topleft_samples_available&= 0xDF5F;
438 h->left_samples_available&= 0x5F5F;
442 if(!IS_INTRA(topleft_type) && (topleft_type==0 || h->pps.constrained_intra_pred))
443 h->topleft_samples_available&= 0x7FFF;
445 if(!IS_INTRA(topright_type) && (topright_type==0 || h->pps.constrained_intra_pred))
446 h->topright_samples_available&= 0xFBFF;
448 if(IS_INTRA4x4(mb_type)){
449 if(IS_INTRA4x4(top_type)){
450 h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
451 h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
452 h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
453 h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
456 if(IS_INTRA16x16(top_type) || (IS_INTER(top_type) && !h->pps.constrained_intra_pred))
461 h->intra4x4_pred_mode_cache[4+8*0]=
462 h->intra4x4_pred_mode_cache[5+8*0]=
463 h->intra4x4_pred_mode_cache[6+8*0]=
464 h->intra4x4_pred_mode_cache[7+8*0]= pred;
467 if(IS_INTRA4x4(left_type[i])){
468 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
469 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
472 if(IS_INTRA16x16(left_type[i]) || (IS_INTER(left_type[i]) && !h->pps.constrained_intra_pred))
477 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
478 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
493 //FIXME constraint_intra_pred & partitioning & nnz (lets hope this is just a typo in the spec)
495 h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][0];
496 h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][1];
497 h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][2];
498 h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3];
500 h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][7];
501 h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8];
503 h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][10];
504 h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11];
506 h->non_zero_count_cache[4+8*0]=
507 h->non_zero_count_cache[5+8*0]=
508 h->non_zero_count_cache[6+8*0]=
509 h->non_zero_count_cache[7+8*0]=
511 h->non_zero_count_cache[1+8*0]=
512 h->non_zero_count_cache[2+8*0]=
514 h->non_zero_count_cache[1+8*3]=
515 h->non_zero_count_cache[2+8*3]= 64;
519 h->non_zero_count_cache[3+8*1]= h->non_zero_count[left_xy[0]][6];
520 h->non_zero_count_cache[3+8*2]= h->non_zero_count[left_xy[0]][5];
521 h->non_zero_count_cache[0+8*1]= h->non_zero_count[left_xy[0]][9]; //FIXME left_block
522 h->non_zero_count_cache[0+8*4]= h->non_zero_count[left_xy[0]][12];
524 h->non_zero_count_cache[3+8*1]=
525 h->non_zero_count_cache[3+8*2]=
526 h->non_zero_count_cache[0+8*1]=
527 h->non_zero_count_cache[0+8*4]= 64;
531 h->non_zero_count_cache[3+8*3]= h->non_zero_count[left_xy[1]][4];
532 h->non_zero_count_cache[3+8*4]= h->non_zero_count[left_xy[1]][3];
533 h->non_zero_count_cache[0+8*2]= h->non_zero_count[left_xy[1]][8];
534 h->non_zero_count_cache[0+8*5]= h->non_zero_count[left_xy[1]][11];
536 h->non_zero_count_cache[3+8*3]=
537 h->non_zero_count_cache[3+8*4]=
538 h->non_zero_count_cache[0+8*2]=
539 h->non_zero_count_cache[0+8*5]= 64;
543 if(IS_INTER(mb_type)){
545 for(list=0; list<2; list++){
546 if((!IS_8X8(mb_type)) && !USES_LIST(mb_type, list)){
547 /*if(!h->mv_cache_clean[list]){
548 memset(h->mv_cache [list], 0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
549 memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
550 h->mv_cache_clean[list]= 1;
552 continue; //FIXME direct mode ...
554 h->mv_cache_clean[list]= 0;
556 if(IS_INTER(topleft_type)){
557 const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
558 const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + h->b8_stride;
559 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
560 h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
562 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
563 h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
566 if(IS_INTER(top_type)){
567 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
568 const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
569 *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0];
570 *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];
571 *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];
572 *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];
573 h->ref_cache[list][scan8[0] + 0 - 1*8]=
574 h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
575 h->ref_cache[list][scan8[0] + 2 - 1*8]=
576 h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
578 *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]=
579 *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]=
580 *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]=
581 *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;
582 *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
585 if(IS_INTER(topright_type)){
586 const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
587 const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
588 *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
589 h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
591 *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
592 h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
595 //FIXME unify cleanup or sth
596 if(IS_INTER(left_type[0])){
597 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
598 const int b8_xy= h->mb2b8_xy[left_xy[0]] + 1;
599 *(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]];
600 *(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]];
601 h->ref_cache[list][scan8[0] - 1 + 0*8]=
602 h->ref_cache[list][scan8[0] - 1 + 1*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0]>>1)];
604 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 0*8]=
605 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 1*8]= 0;
606 h->ref_cache[list][scan8[0] - 1 + 0*8]=
607 h->ref_cache[list][scan8[0] - 1 + 1*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
610 if(IS_INTER(left_type[1])){
611 const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
612 const int b8_xy= h->mb2b8_xy[left_xy[1]] + 1;
613 *(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]];
614 *(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]];
615 h->ref_cache[list][scan8[0] - 1 + 2*8]=
616 h->ref_cache[list][scan8[0] - 1 + 3*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[2]>>1)];
618 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 2*8]=
619 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 3*8]= 0;
620 h->ref_cache[list][scan8[0] - 1 + 2*8]=
621 h->ref_cache[list][scan8[0] - 1 + 3*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
624 h->ref_cache[list][scan8[5 ]+1] =
625 h->ref_cache[list][scan8[7 ]+1] =
626 h->ref_cache[list][scan8[13]+1] = //FIXME remove past 3 (init somewher else)
627 h->ref_cache[list][scan8[4 ]] =
628 h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
629 *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
630 *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
631 *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewher else)
632 *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
633 *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
636 /* XXX beurk, Load mvd */
637 if(IS_INTER(topleft_type)){
638 const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
639 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy];
641 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 - 1*8]= 0;
644 if(IS_INTER(top_type)){
645 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
646 *(uint32_t*)h->mvd_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 0];
647 *(uint32_t*)h->mvd_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 1];
648 *(uint32_t*)h->mvd_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 2];
649 *(uint32_t*)h->mvd_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 3];
651 *(uint32_t*)h->mvd_cache [list][scan8[0] + 0 - 1*8]=
652 *(uint32_t*)h->mvd_cache [list][scan8[0] + 1 - 1*8]=
653 *(uint32_t*)h->mvd_cache [list][scan8[0] + 2 - 1*8]=
654 *(uint32_t*)h->mvd_cache [list][scan8[0] + 3 - 1*8]= 0;
656 if(IS_INTER(left_type[0])){
657 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
658 *(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]];
659 *(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]];
661 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 0*8]=
662 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 1*8]= 0;
664 if(IS_INTER(left_type[1])){
665 const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
666 *(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]];
667 *(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]];
669 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 2*8]=
670 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 3*8]= 0;
672 *(uint32_t*)h->mvd_cache [list][scan8[5 ]+1]=
673 *(uint32_t*)h->mvd_cache [list][scan8[7 ]+1]=
674 *(uint32_t*)h->mvd_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewher else)
675 *(uint32_t*)h->mvd_cache [list][scan8[4 ]]=
676 *(uint32_t*)h->mvd_cache [list][scan8[12]]= 0;
684 static inline void write_back_intra_pred_mode(H264Context *h){
685 MpegEncContext * const s = &h->s;
686 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
688 h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
689 h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
690 h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
691 h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
692 h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
693 h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
694 h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
698 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
700 static inline int check_intra4x4_pred_mode(H264Context *h){
701 MpegEncContext * const s = &h->s;
702 static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
703 static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
706 if(!(h->top_samples_available&0x8000)){
708 int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
710 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);
713 h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
718 if(!(h->left_samples_available&0x8000)){
720 int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
722 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);
725 h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
731 } //FIXME cleanup like next
734 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
736 static inline int check_intra_pred_mode(H264Context *h, int mode){
737 MpegEncContext * const s = &h->s;
738 static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
739 static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
741 if(!(h->top_samples_available&0x8000)){
744 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);
749 if(!(h->left_samples_available&0x8000)){
752 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);
761 * gets the predicted intra4x4 prediction mode.
763 static inline int pred_intra_mode(H264Context *h, int n){
764 const int index8= scan8[n];
765 const int left= h->intra4x4_pred_mode_cache[index8 - 1];
766 const int top = h->intra4x4_pred_mode_cache[index8 - 8];
767 const int min= FFMIN(left, top);
769 tprintf("mode:%d %d min:%d\n", left ,top, min);
771 if(min<0) return DC_PRED;
775 static inline void write_back_non_zero_count(H264Context *h){
776 MpegEncContext * const s = &h->s;
777 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
779 h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[4+8*4];
780 h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[5+8*4];
781 h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[6+8*4];
782 h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];
783 h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[7+8*3];
784 h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[7+8*2];
785 h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[7+8*1];
787 h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[1+8*2];
788 h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
789 h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[2+8*1];
791 h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[1+8*5];
792 h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
793 h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[2+8*4];
797 * gets the predicted number of non zero coefficients.
798 * @param n block index
800 static inline int pred_non_zero_count(H264Context *h, int n){
801 const int index8= scan8[n];
802 const int left= h->non_zero_count_cache[index8 - 1];
803 const int top = h->non_zero_count_cache[index8 - 8];
806 if(i<64) i= (i+1)>>1;
808 tprintf("pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
813 static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
814 const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];
816 if(topright_ref != PART_NOT_AVAILABLE){
817 *C= h->mv_cache[list][ i - 8 + part_width ];
820 tprintf("topright MV not available\n");
822 *C= h->mv_cache[list][ i - 8 - 1 ];
823 return h->ref_cache[list][ i - 8 - 1 ];
828 * gets the predicted MV.
829 * @param n the block index
830 * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)
831 * @param mx the x component of the predicted motion vector
832 * @param my the y component of the predicted motion vector
834 static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
835 const int index8= scan8[n];
836 const int top_ref= h->ref_cache[list][ index8 - 8 ];
837 const int left_ref= h->ref_cache[list][ index8 - 1 ];
838 const int16_t * const A= h->mv_cache[list][ index8 - 1 ];
839 const int16_t * const B= h->mv_cache[list][ index8 - 8 ];
841 int diagonal_ref, match_count;
843 assert(part_width==1 || part_width==2 || part_width==4);
853 diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);
854 match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);
855 if(match_count > 1){ //most common
856 *mx= mid_pred(A[0], B[0], C[0]);
857 *my= mid_pred(A[1], B[1], C[1]);
858 }else if(match_count==1){
862 }else if(top_ref==ref){
870 if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
874 *mx= mid_pred(A[0], B[0], C[0]);
875 *my= mid_pred(A[1], B[1], C[1]);
879 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);
883 * gets the directionally predicted 16x8 MV.
884 * @param n the block index
885 * @param mx the x component of the predicted motion vector
886 * @param my the y component of the predicted motion vector
888 static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
890 const int top_ref= h->ref_cache[list][ scan8[0] - 8 ];
891 const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
893 tprintf("pred_16x8: (%2d %2d %2d) at %2d %2d %d list %d", top_ref, B[0], B[1], h->s.mb_x, h->s.mb_y, n, list);
901 const int left_ref= h->ref_cache[list][ scan8[8] - 1 ];
902 const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];
904 tprintf("pred_16x8: (%2d %2d %2d) at %2d %2d %d list %d", left_ref, A[0], A[1], h->s.mb_x, h->s.mb_y, n, list);
914 pred_motion(h, n, 4, list, ref, mx, my);
918 * gets the directionally predicted 8x16 MV.
919 * @param n the block index
920 * @param mx the x component of the predicted motion vector
921 * @param my the y component of the predicted motion vector
923 static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
925 const int left_ref= h->ref_cache[list][ scan8[0] - 1 ];
926 const int16_t * const A= h->mv_cache[list][ scan8[0] - 1 ];
928 tprintf("pred_8x16: (%2d %2d %2d) at %2d %2d %d list %d", left_ref, A[0], A[1], h->s.mb_x, h->s.mb_y, n, list);
939 diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);
941 tprintf("pred_8x16: (%2d %2d %2d) at %2d %2d %d list %d", diagonal_ref, C[0], C[1], h->s.mb_x, h->s.mb_y, n, list);
943 if(diagonal_ref == ref){
951 pred_motion(h, n, 2, list, ref, mx, my);
954 static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
955 const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
956 const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];
958 tprintf("pred_pskip: (%d) (%d) at %2d %2d", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
960 if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
961 || (top_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0)
962 || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){
968 pred_motion(h, 0, 4, 0, 0, mx, my);
973 static inline void write_back_motion(H264Context *h, int mb_type){
974 MpegEncContext * const s = &h->s;
975 const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
976 const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
979 for(list=0; list<2; list++){
981 if((!IS_8X8(mb_type)) && !USES_LIST(mb_type, list)){
982 if(1){ //FIXME skip or never read if mb_type doesnt use it
984 *(uint64_t*)s->current_picture.motion_val[list][b_xy + 0 + y*h->b_stride]=
985 *(uint64_t*)s->current_picture.motion_val[list][b_xy + 2 + y*h->b_stride]= 0;
990 *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]=
991 *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= 0;
995 *(uint16_t*)s->current_picture.motion_val[list][b8_xy + y*h->b8_stride]= (LIST_NOT_USED&0xFF)*0x0101;
998 continue; //FIXME direct mode ...
1002 *(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];
1003 *(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];
1005 if( h->pps.cabac ) {
1007 *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+0 + 8*y];
1008 *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+2 + 8*y];
1012 s->current_picture.ref_index[list][b8_xy + 0 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+0 + 16*y];
1013 s->current_picture.ref_index[list][b8_xy + 1 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+2 + 16*y];
1019 * Decodes a network abstraction layer unit.
1020 * @param consumed is the number of bytes used as input
1021 * @param length is the length of the array
1022 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp ttailing?
1023 * @returns decoded bytes, might be src+1 if no escapes
1025 static uint8_t *decode_nal(H264Context *h, uint8_t *src, int *dst_length, int *consumed, int length){
1029 // src[0]&0x80; //forbidden bit
1030 h->nal_ref_idc= src[0]>>5;
1031 h->nal_unit_type= src[0]&0x1F;
1035 for(i=0; i<length; i++)
1036 printf("%2X ", src[i]);
1038 for(i=0; i+1<length; i+=2){
1039 if(src[i]) continue;
1040 if(i>0 && src[i-1]==0) i--;
1041 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1043 /* startcode, so we must be past the end */
1050 if(i>=length-1){ //no escaped 0
1051 *dst_length= length;
1052 *consumed= length+1; //+1 for the header
1056 h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length);
1057 dst= h->rbsp_buffer;
1059 //printf("deoding esc\n");
1062 //remove escapes (very rare 1:2^22)
1063 if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1064 if(src[si+2]==3){ //escape
1069 }else //next start code
1073 dst[di++]= src[si++];
1077 *consumed= si + 1;//+1 for the header
1078 //FIXME store exact number of bits in the getbitcontext (its needed for decoding)
1084 * @param src the data which should be escaped
1085 * @param dst the target buffer, dst+1 == src is allowed as a special case
1086 * @param length the length of the src data
1087 * @param dst_length the length of the dst array
1088 * @returns length of escaped data in bytes or -1 if an error occured
1090 static int encode_nal(H264Context *h, uint8_t *dst, uint8_t *src, int length, int dst_length){
1091 int i, escape_count, si, di;
1095 assert(dst_length>0);
1097 dst[0]= (h->nal_ref_idc<<5) + h->nal_unit_type;
1099 if(length==0) return 1;
1102 for(i=0; i<length; i+=2){
1103 if(src[i]) continue;
1104 if(i>0 && src[i-1]==0)
1106 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1112 if(escape_count==0){
1114 memcpy(dst+1, src, length);
1118 if(length + escape_count + 1> dst_length)
1121 //this should be damn rare (hopefully)
1123 h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length + escape_count);
1124 temp= h->rbsp_buffer;
1125 //printf("encoding esc\n");
1130 if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1131 temp[di++]= 0; si++;
1132 temp[di++]= 0; si++;
1134 temp[di++]= src[si++];
1137 temp[di++]= src[si++];
1139 memcpy(dst+1, temp, length+escape_count);
1141 assert(di == length+escape_count);
1147 * write 1,10,100,1000,... for alignment, yes its exactly inverse to mpeg4
1149 static void encode_rbsp_trailing(PutBitContext *pb){
1152 length= (-put_bits_count(pb))&7;
1153 if(length) put_bits(pb, length, 0);
1158 * identifies the exact end of the bitstream
1159 * @return the length of the trailing, or 0 if damaged
1161 static int decode_rbsp_trailing(uint8_t *src){
1165 tprintf("rbsp trailing %X\n", v);
1175 * idct tranforms the 16 dc values and dequantize them.
1176 * @param qp quantization parameter
1178 static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp){
1179 const int qmul= dequant_coeff[qp][0];
1182 int temp[16]; //FIXME check if this is a good idea
1183 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
1184 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1186 //memset(block, 64, 2*256);
1189 const int offset= y_offset[i];
1190 const int z0= block[offset+stride*0] + block[offset+stride*4];
1191 const int z1= block[offset+stride*0] - block[offset+stride*4];
1192 const int z2= block[offset+stride*1] - block[offset+stride*5];
1193 const int z3= block[offset+stride*1] + block[offset+stride*5];
1202 const int offset= x_offset[i];
1203 const int z0= temp[4*0+i] + temp[4*2+i];
1204 const int z1= temp[4*0+i] - temp[4*2+i];
1205 const int z2= temp[4*1+i] - temp[4*3+i];
1206 const int z3= temp[4*1+i] + temp[4*3+i];
1208 block[stride*0 +offset]= ((z0 + z3)*qmul + 2)>>2; //FIXME think about merging this into decode_resdual
1209 block[stride*2 +offset]= ((z1 + z2)*qmul + 2)>>2;
1210 block[stride*8 +offset]= ((z1 - z2)*qmul + 2)>>2;
1211 block[stride*10+offset]= ((z0 - z3)*qmul + 2)>>2;
1217 * dct tranforms the 16 dc values.
1218 * @param qp quantization parameter ??? FIXME
1220 static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
1221 // const int qmul= dequant_coeff[qp][0];
1223 int temp[16]; //FIXME check if this is a good idea
1224 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
1225 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1228 const int offset= y_offset[i];
1229 const int z0= block[offset+stride*0] + block[offset+stride*4];
1230 const int z1= block[offset+stride*0] - block[offset+stride*4];
1231 const int z2= block[offset+stride*1] - block[offset+stride*5];
1232 const int z3= block[offset+stride*1] + block[offset+stride*5];
1241 const int offset= x_offset[i];
1242 const int z0= temp[4*0+i] + temp[4*2+i];
1243 const int z1= temp[4*0+i] - temp[4*2+i];
1244 const int z2= temp[4*1+i] - temp[4*3+i];
1245 const int z3= temp[4*1+i] + temp[4*3+i];
1247 block[stride*0 +offset]= (z0 + z3)>>1;
1248 block[stride*2 +offset]= (z1 + z2)>>1;
1249 block[stride*8 +offset]= (z1 - z2)>>1;
1250 block[stride*10+offset]= (z0 - z3)>>1;
1258 static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp){
1259 const int qmul= dequant_coeff[qp][0];
1260 const int stride= 16*2;
1261 const int xStride= 16;
1264 a= block[stride*0 + xStride*0];
1265 b= block[stride*0 + xStride*1];
1266 c= block[stride*1 + xStride*0];
1267 d= block[stride*1 + xStride*1];
1274 block[stride*0 + xStride*0]= ((a+c)*qmul + 0)>>1;
1275 block[stride*0 + xStride*1]= ((e+b)*qmul + 0)>>1;
1276 block[stride*1 + xStride*0]= ((a-c)*qmul + 0)>>1;
1277 block[stride*1 + xStride*1]= ((e-b)*qmul + 0)>>1;
1281 static void chroma_dc_dct_c(DCTELEM *block){
1282 const int stride= 16*2;
1283 const int xStride= 16;
1286 a= block[stride*0 + xStride*0];
1287 b= block[stride*0 + xStride*1];
1288 c= block[stride*1 + xStride*0];
1289 d= block[stride*1 + xStride*1];
1296 block[stride*0 + xStride*0]= (a+c);
1297 block[stride*0 + xStride*1]= (e+b);
1298 block[stride*1 + xStride*0]= (a-c);
1299 block[stride*1 + xStride*1]= (e-b);
1304 * gets the chroma qp.
1306 static inline int get_chroma_qp(H264Context *h, int qscale){
1308 return chroma_qp[clip(qscale + h->pps.chroma_qp_index_offset, 0, 51)];
1315 static void h264_add_idct_c(uint8_t *dst, DCTELEM *block, int stride){
1317 uint8_t *cm = cropTbl + MAX_NEG_CROP;
1322 const int z0= block[0 + 4*i] + block[2 + 4*i];
1323 const int z1= block[0 + 4*i] - block[2 + 4*i];
1324 const int z2= (block[1 + 4*i]>>1) - block[3 + 4*i];
1325 const int z3= block[1 + 4*i] + (block[3 + 4*i]>>1);
1327 block[0 + 4*i]= z0 + z3;
1328 block[1 + 4*i]= z1 + z2;
1329 block[2 + 4*i]= z1 - z2;
1330 block[3 + 4*i]= z0 - z3;
1334 const int z0= block[i + 4*0] + block[i + 4*2];
1335 const int z1= block[i + 4*0] - block[i + 4*2];
1336 const int z2= (block[i + 4*1]>>1) - block[i + 4*3];
1337 const int z3= block[i + 4*1] + (block[i + 4*3]>>1);
1339 dst[i + 0*stride]= cm[ dst[i + 0*stride] + ((z0 + z3) >> 6) ];
1340 dst[i + 1*stride]= cm[ dst[i + 1*stride] + ((z1 + z2) >> 6) ];
1341 dst[i + 2*stride]= cm[ dst[i + 2*stride] + ((z1 - z2) >> 6) ];
1342 dst[i + 3*stride]= cm[ dst[i + 3*stride] + ((z0 - z3) >> 6) ];
1347 static void h264_diff_dct_c(DCTELEM *block, uint8_t *src1, uint8_t *src2, int stride){
1349 //FIXME try int temp instead of block
1352 const int d0= src1[0 + i*stride] - src2[0 + i*stride];
1353 const int d1= src1[1 + i*stride] - src2[1 + i*stride];
1354 const int d2= src1[2 + i*stride] - src2[2 + i*stride];
1355 const int d3= src1[3 + i*stride] - src2[3 + i*stride];
1356 const int z0= d0 + d3;
1357 const int z3= d0 - d3;
1358 const int z1= d1 + d2;
1359 const int z2= d1 - d2;
1361 block[0 + 4*i]= z0 + z1;
1362 block[1 + 4*i]= 2*z3 + z2;
1363 block[2 + 4*i]= z0 - z1;
1364 block[3 + 4*i]= z3 - 2*z2;
1368 const int z0= block[0*4 + i] + block[3*4 + i];
1369 const int z3= block[0*4 + i] - block[3*4 + i];
1370 const int z1= block[1*4 + i] + block[2*4 + i];
1371 const int z2= block[1*4 + i] - block[2*4 + i];
1373 block[0*4 + i]= z0 + z1;
1374 block[1*4 + i]= 2*z3 + z2;
1375 block[2*4 + i]= z0 - z1;
1376 block[3*4 + i]= z3 - 2*z2;
1381 //FIXME need to check that this doesnt overflow signed 32 bit for low qp, iam not sure, its very close
1382 //FIXME check that gcc inlines this (and optimizes intra & seperate_dc stuff away)
1383 static inline int quantize_c(DCTELEM *block, uint8_t *scantable, int qscale, int intra, int seperate_dc){
1385 const int * const quant_table= quant_coeff[qscale];
1386 const int bias= intra ? (1<<QUANT_SHIFT)/3 : (1<<QUANT_SHIFT)/6;
1387 const unsigned int threshold1= (1<<QUANT_SHIFT) - bias - 1;
1388 const unsigned int threshold2= (threshold1<<1);
1394 const int dc_bias= intra ? (1<<(QUANT_SHIFT-2))/3 : (1<<(QUANT_SHIFT-2))/6;
1395 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT-2)) - dc_bias - 1;
1396 const unsigned int dc_threshold2= (dc_threshold1<<1);
1398 int level= block[0]*quant_coeff[qscale+18][0];
1399 if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1401 level= (dc_bias + level)>>(QUANT_SHIFT-2);
1404 level= (dc_bias - level)>>(QUANT_SHIFT-2);
1407 // last_non_zero = i;
1412 const int dc_bias= intra ? (1<<(QUANT_SHIFT+1))/3 : (1<<(QUANT_SHIFT+1))/6;
1413 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT+1)) - dc_bias - 1;
1414 const unsigned int dc_threshold2= (dc_threshold1<<1);
1416 int level= block[0]*quant_table[0];
1417 if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1419 level= (dc_bias + level)>>(QUANT_SHIFT+1);
1422 level= (dc_bias - level)>>(QUANT_SHIFT+1);
1425 // last_non_zero = i;
1438 const int j= scantable[i];
1439 int level= block[j]*quant_table[j];
1441 // if( bias+level >= (1<<(QMAT_SHIFT - 3))
1442 // || bias-level >= (1<<(QMAT_SHIFT - 3))){
1443 if(((unsigned)(level+threshold1))>threshold2){
1445 level= (bias + level)>>QUANT_SHIFT;
1448 level= (bias - level)>>QUANT_SHIFT;
1457 return last_non_zero;
1460 static void pred4x4_vertical_c(uint8_t *src, uint8_t *topright, int stride){
1461 const uint32_t a= ((uint32_t*)(src-stride))[0];
1462 ((uint32_t*)(src+0*stride))[0]= a;
1463 ((uint32_t*)(src+1*stride))[0]= a;
1464 ((uint32_t*)(src+2*stride))[0]= a;
1465 ((uint32_t*)(src+3*stride))[0]= a;
1468 static void pred4x4_horizontal_c(uint8_t *src, uint8_t *topright, int stride){
1469 ((uint32_t*)(src+0*stride))[0]= src[-1+0*stride]*0x01010101;
1470 ((uint32_t*)(src+1*stride))[0]= src[-1+1*stride]*0x01010101;
1471 ((uint32_t*)(src+2*stride))[0]= src[-1+2*stride]*0x01010101;
1472 ((uint32_t*)(src+3*stride))[0]= src[-1+3*stride]*0x01010101;
1475 static void pred4x4_dc_c(uint8_t *src, uint8_t *topright, int stride){
1476 const int dc= ( src[-stride] + src[1-stride] + src[2-stride] + src[3-stride]
1477 + src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 4) >>3;
1479 ((uint32_t*)(src+0*stride))[0]=
1480 ((uint32_t*)(src+1*stride))[0]=
1481 ((uint32_t*)(src+2*stride))[0]=
1482 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1485 static void pred4x4_left_dc_c(uint8_t *src, uint8_t *topright, int stride){
1486 const int dc= ( src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 2) >>2;
1488 ((uint32_t*)(src+0*stride))[0]=
1489 ((uint32_t*)(src+1*stride))[0]=
1490 ((uint32_t*)(src+2*stride))[0]=
1491 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1494 static void pred4x4_top_dc_c(uint8_t *src, uint8_t *topright, int stride){
1495 const int dc= ( src[-stride] + src[1-stride] + src[2-stride] + src[3-stride] + 2) >>2;
1497 ((uint32_t*)(src+0*stride))[0]=
1498 ((uint32_t*)(src+1*stride))[0]=
1499 ((uint32_t*)(src+2*stride))[0]=
1500 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1503 static void pred4x4_128_dc_c(uint8_t *src, uint8_t *topright, int stride){
1504 ((uint32_t*)(src+0*stride))[0]=
1505 ((uint32_t*)(src+1*stride))[0]=
1506 ((uint32_t*)(src+2*stride))[0]=
1507 ((uint32_t*)(src+3*stride))[0]= 128U*0x01010101U;
1511 #define LOAD_TOP_RIGHT_EDGE\
1512 const int t4= topright[0];\
1513 const int t5= topright[1];\
1514 const int t6= topright[2];\
1515 const int t7= topright[3];\
1517 #define LOAD_LEFT_EDGE\
1518 const int l0= src[-1+0*stride];\
1519 const int l1= src[-1+1*stride];\
1520 const int l2= src[-1+2*stride];\
1521 const int l3= src[-1+3*stride];\
1523 #define LOAD_TOP_EDGE\
1524 const int t0= src[ 0-1*stride];\
1525 const int t1= src[ 1-1*stride];\
1526 const int t2= src[ 2-1*stride];\
1527 const int t3= src[ 3-1*stride];\
1529 static void pred4x4_down_right_c(uint8_t *src, uint8_t *topright, int stride){
1530 const int lt= src[-1-1*stride];
1534 src[0+3*stride]=(l3 + 2*l2 + l1 + 2)>>2;
1536 src[1+3*stride]=(l2 + 2*l1 + l0 + 2)>>2;
1539 src[2+3*stride]=(l1 + 2*l0 + lt + 2)>>2;
1543 src[3+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1546 src[3+2*stride]=(lt + 2*t0 + t1 + 2)>>2;
1548 src[3+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1549 src[3+0*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1552 static void pred4x4_down_left_c(uint8_t *src, uint8_t *topright, int stride){
1557 src[0+0*stride]=(t0 + t2 + 2*t1 + 2)>>2;
1559 src[0+1*stride]=(t1 + t3 + 2*t2 + 2)>>2;
1562 src[0+2*stride]=(t2 + t4 + 2*t3 + 2)>>2;
1566 src[0+3*stride]=(t3 + t5 + 2*t4 + 2)>>2;
1569 src[1+3*stride]=(t4 + t6 + 2*t5 + 2)>>2;
1571 src[2+3*stride]=(t5 + t7 + 2*t6 + 2)>>2;
1572 src[3+3*stride]=(t6 + 3*t7 + 2)>>2;
1575 static void pred4x4_vertical_right_c(uint8_t *src, uint8_t *topright, int stride){
1576 const int lt= src[-1-1*stride];
1579 const __attribute__((unused)) int unu= l3;
1582 src[1+2*stride]=(lt + t0 + 1)>>1;
1584 src[2+2*stride]=(t0 + t1 + 1)>>1;
1586 src[3+2*stride]=(t1 + t2 + 1)>>1;
1587 src[3+0*stride]=(t2 + t3 + 1)>>1;
1589 src[1+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1591 src[2+3*stride]=(lt + 2*t0 + t1 + 2)>>2;
1593 src[3+3*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1594 src[3+1*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1595 src[0+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1596 src[0+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1599 static void pred4x4_vertical_left_c(uint8_t *src, uint8_t *topright, int stride){
1602 const __attribute__((unused)) int unu= t7;
1604 src[0+0*stride]=(t0 + t1 + 1)>>1;
1606 src[0+2*stride]=(t1 + t2 + 1)>>1;
1608 src[1+2*stride]=(t2 + t3 + 1)>>1;
1610 src[2+2*stride]=(t3 + t4+ 1)>>1;
1611 src[3+2*stride]=(t4 + t5+ 1)>>1;
1612 src[0+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1614 src[0+3*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1616 src[1+3*stride]=(t2 + 2*t3 + t4 + 2)>>2;
1618 src[2+3*stride]=(t3 + 2*t4 + t5 + 2)>>2;
1619 src[3+3*stride]=(t4 + 2*t5 + t6 + 2)>>2;
1622 static void pred4x4_horizontal_up_c(uint8_t *src, uint8_t *topright, int stride){
1625 src[0+0*stride]=(l0 + l1 + 1)>>1;
1626 src[1+0*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1628 src[0+1*stride]=(l1 + l2 + 1)>>1;
1630 src[1+1*stride]=(l1 + 2*l2 + l3 + 2)>>2;
1632 src[0+2*stride]=(l2 + l3 + 1)>>1;
1634 src[1+2*stride]=(l2 + 2*l3 + l3 + 2)>>2;
1643 static void pred4x4_horizontal_down_c(uint8_t *src, uint8_t *topright, int stride){
1644 const int lt= src[-1-1*stride];
1647 const __attribute__((unused)) int unu= t3;
1650 src[2+1*stride]=(lt + l0 + 1)>>1;
1652 src[3+1*stride]=(l0 + 2*lt + t0 + 2)>>2;
1653 src[2+0*stride]=(lt + 2*t0 + t1 + 2)>>2;
1654 src[3+0*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1656 src[2+2*stride]=(l0 + l1 + 1)>>1;
1658 src[3+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1660 src[2+3*stride]=(l1 + l2+ 1)>>1;
1662 src[3+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1663 src[0+3*stride]=(l2 + l3 + 1)>>1;
1664 src[1+3*stride]=(l1 + 2*l2 + l3 + 2)>>2;
1667 static void pred16x16_vertical_c(uint8_t *src, int stride){
1669 const uint32_t a= ((uint32_t*)(src-stride))[0];
1670 const uint32_t b= ((uint32_t*)(src-stride))[1];
1671 const uint32_t c= ((uint32_t*)(src-stride))[2];
1672 const uint32_t d= ((uint32_t*)(src-stride))[3];
1674 for(i=0; i<16; i++){
1675 ((uint32_t*)(src+i*stride))[0]= a;
1676 ((uint32_t*)(src+i*stride))[1]= b;
1677 ((uint32_t*)(src+i*stride))[2]= c;
1678 ((uint32_t*)(src+i*stride))[3]= d;
1682 static void pred16x16_horizontal_c(uint8_t *src, int stride){
1685 for(i=0; i<16; i++){
1686 ((uint32_t*)(src+i*stride))[0]=
1687 ((uint32_t*)(src+i*stride))[1]=
1688 ((uint32_t*)(src+i*stride))[2]=
1689 ((uint32_t*)(src+i*stride))[3]= src[-1+i*stride]*0x01010101;
1693 static void pred16x16_dc_c(uint8_t *src, int stride){
1697 dc+= src[-1+i*stride];
1704 dc= 0x01010101*((dc + 16)>>5);
1706 for(i=0; i<16; i++){
1707 ((uint32_t*)(src+i*stride))[0]=
1708 ((uint32_t*)(src+i*stride))[1]=
1709 ((uint32_t*)(src+i*stride))[2]=
1710 ((uint32_t*)(src+i*stride))[3]= dc;
1714 static void pred16x16_left_dc_c(uint8_t *src, int stride){
1718 dc+= src[-1+i*stride];
1721 dc= 0x01010101*((dc + 8)>>4);
1723 for(i=0; i<16; i++){
1724 ((uint32_t*)(src+i*stride))[0]=
1725 ((uint32_t*)(src+i*stride))[1]=
1726 ((uint32_t*)(src+i*stride))[2]=
1727 ((uint32_t*)(src+i*stride))[3]= dc;
1731 static void pred16x16_top_dc_c(uint8_t *src, int stride){
1737 dc= 0x01010101*((dc + 8)>>4);
1739 for(i=0; i<16; i++){
1740 ((uint32_t*)(src+i*stride))[0]=
1741 ((uint32_t*)(src+i*stride))[1]=
1742 ((uint32_t*)(src+i*stride))[2]=
1743 ((uint32_t*)(src+i*stride))[3]= dc;
1747 static void pred16x16_128_dc_c(uint8_t *src, int stride){
1750 for(i=0; i<16; i++){
1751 ((uint32_t*)(src+i*stride))[0]=
1752 ((uint32_t*)(src+i*stride))[1]=
1753 ((uint32_t*)(src+i*stride))[2]=
1754 ((uint32_t*)(src+i*stride))[3]= 0x01010101U*128U;
1758 static inline void pred16x16_plane_compat_c(uint8_t *src, int stride, const int svq3){
1761 uint8_t *cm = cropTbl + MAX_NEG_CROP;
1762 const uint8_t * const src0 = src+7-stride;
1763 const uint8_t *src1 = src+8*stride-1;
1764 const uint8_t *src2 = src1-2*stride; // == src+6*stride-1;
1765 int H = src0[1] - src0[-1];
1766 int V = src1[0] - src2[ 0];
1767 for(k=2; k<=8; ++k) {
1768 src1 += stride; src2 -= stride;
1769 H += k*(src0[k] - src0[-k]);
1770 V += k*(src1[0] - src2[ 0]);
1773 H = ( 5*(H/4) ) / 16;
1774 V = ( 5*(V/4) ) / 16;
1776 /* required for 100% accuracy */
1777 i = H; H = V; V = i;
1779 H = ( 5*H+32 ) >> 6;
1780 V = ( 5*V+32 ) >> 6;
1783 a = 16*(src1[0] + src2[16] + 1) - 7*(V+H);
1784 for(j=16; j>0; --j) {
1787 for(i=-16; i<0; i+=4) {
1788 src[16+i] = cm[ (b ) >> 5 ];
1789 src[17+i] = cm[ (b+ H) >> 5 ];
1790 src[18+i] = cm[ (b+2*H) >> 5 ];
1791 src[19+i] = cm[ (b+3*H) >> 5 ];
1798 static void pred16x16_plane_c(uint8_t *src, int stride){
1799 pred16x16_plane_compat_c(src, stride, 0);
1802 static void pred8x8_vertical_c(uint8_t *src, int stride){
1804 const uint32_t a= ((uint32_t*)(src-stride))[0];
1805 const uint32_t b= ((uint32_t*)(src-stride))[1];
1808 ((uint32_t*)(src+i*stride))[0]= a;
1809 ((uint32_t*)(src+i*stride))[1]= b;
1813 static void pred8x8_horizontal_c(uint8_t *src, int stride){
1817 ((uint32_t*)(src+i*stride))[0]=
1818 ((uint32_t*)(src+i*stride))[1]= src[-1+i*stride]*0x01010101;
1822 static void pred8x8_128_dc_c(uint8_t *src, int stride){
1826 ((uint32_t*)(src+i*stride))[0]=
1827 ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
1830 ((uint32_t*)(src+i*stride))[0]=
1831 ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
1835 static void pred8x8_left_dc_c(uint8_t *src, int stride){
1841 dc0+= src[-1+i*stride];
1842 dc2+= src[-1+(i+4)*stride];
1844 dc0= 0x01010101*((dc0 + 2)>>2);
1845 dc2= 0x01010101*((dc2 + 2)>>2);
1848 ((uint32_t*)(src+i*stride))[0]=
1849 ((uint32_t*)(src+i*stride))[1]= dc0;
1852 ((uint32_t*)(src+i*stride))[0]=
1853 ((uint32_t*)(src+i*stride))[1]= dc2;
1857 static void pred8x8_top_dc_c(uint8_t *src, int stride){
1863 dc0+= src[i-stride];
1864 dc1+= src[4+i-stride];
1866 dc0= 0x01010101*((dc0 + 2)>>2);
1867 dc1= 0x01010101*((dc1 + 2)>>2);
1870 ((uint32_t*)(src+i*stride))[0]= dc0;
1871 ((uint32_t*)(src+i*stride))[1]= dc1;
1874 ((uint32_t*)(src+i*stride))[0]= dc0;
1875 ((uint32_t*)(src+i*stride))[1]= dc1;
1880 static void pred8x8_dc_c(uint8_t *src, int stride){
1882 int dc0, dc1, dc2, dc3;
1886 dc0+= src[-1+i*stride] + src[i-stride];
1887 dc1+= src[4+i-stride];
1888 dc2+= src[-1+(i+4)*stride];
1890 dc3= 0x01010101*((dc1 + dc2 + 4)>>3);
1891 dc0= 0x01010101*((dc0 + 4)>>3);
1892 dc1= 0x01010101*((dc1 + 2)>>2);
1893 dc2= 0x01010101*((dc2 + 2)>>2);
1896 ((uint32_t*)(src+i*stride))[0]= dc0;
1897 ((uint32_t*)(src+i*stride))[1]= dc1;
1900 ((uint32_t*)(src+i*stride))[0]= dc2;
1901 ((uint32_t*)(src+i*stride))[1]= dc3;
1905 static void pred8x8_plane_c(uint8_t *src, int stride){
1908 uint8_t *cm = cropTbl + MAX_NEG_CROP;
1909 const uint8_t * const src0 = src+3-stride;
1910 const uint8_t *src1 = src+4*stride-1;
1911 const uint8_t *src2 = src1-2*stride; // == src+2*stride-1;
1912 int H = src0[1] - src0[-1];
1913 int V = src1[0] - src2[ 0];
1914 for(k=2; k<=4; ++k) {
1915 src1 += stride; src2 -= stride;
1916 H += k*(src0[k] - src0[-k]);
1917 V += k*(src1[0] - src2[ 0]);
1919 H = ( 17*H+16 ) >> 5;
1920 V = ( 17*V+16 ) >> 5;
1922 a = 16*(src1[0] + src2[8]+1) - 3*(V+H);
1923 for(j=8; j>0; --j) {
1926 src[0] = cm[ (b ) >> 5 ];
1927 src[1] = cm[ (b+ H) >> 5 ];
1928 src[2] = cm[ (b+2*H) >> 5 ];
1929 src[3] = cm[ (b+3*H) >> 5 ];
1930 src[4] = cm[ (b+4*H) >> 5 ];
1931 src[5] = cm[ (b+5*H) >> 5 ];
1932 src[6] = cm[ (b+6*H) >> 5 ];
1933 src[7] = cm[ (b+7*H) >> 5 ];
1938 static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
1939 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1940 int src_x_offset, int src_y_offset,
1941 qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
1942 MpegEncContext * const s = &h->s;
1943 const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
1944 const int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
1945 const int luma_xy= (mx&3) + ((my&3)<<2);
1946 uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*s->linesize;
1947 uint8_t * src_cb= pic->data[1] + (mx>>3) + (my>>3)*s->uvlinesize;
1948 uint8_t * src_cr= pic->data[2] + (mx>>3) + (my>>3)*s->uvlinesize;
1949 int extra_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16; //FIXME increase edge?, IMHO not worth it
1950 int extra_height= extra_width;
1952 const int full_mx= mx>>2;
1953 const int full_my= my>>2;
1955 assert(pic->data[0]);
1957 if(mx&7) extra_width -= 3;
1958 if(my&7) extra_height -= 3;
1960 if( full_mx < 0-extra_width
1961 || full_my < 0-extra_height
1962 || full_mx + 16/*FIXME*/ > s->width + extra_width
1963 || full_my + 16/*FIXME*/ > s->height + extra_height){
1964 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);
1965 src_y= s->edge_emu_buffer + 2 + 2*s->linesize;
1969 qpix_op[luma_xy](dest_y, src_y, s->linesize); //FIXME try variable height perhaps?
1971 qpix_op[luma_xy](dest_y + delta, src_y + delta, s->linesize);
1974 if(s->flags&CODEC_FLAG_GRAY) return;
1977 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);
1978 src_cb= s->edge_emu_buffer;
1980 chroma_op(dest_cb, src_cb, s->uvlinesize, chroma_height, mx&7, my&7);
1983 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);
1984 src_cr= s->edge_emu_buffer;
1986 chroma_op(dest_cr, src_cr, s->uvlinesize, chroma_height, mx&7, my&7);
1989 static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
1990 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1991 int x_offset, int y_offset,
1992 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
1993 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
1994 int list0, int list1){
1995 MpegEncContext * const s = &h->s;
1996 qpel_mc_func *qpix_op= qpix_put;
1997 h264_chroma_mc_func chroma_op= chroma_put;
1999 dest_y += 2*x_offset + 2*y_offset*s-> linesize;
2000 dest_cb += x_offset + y_offset*s->uvlinesize;
2001 dest_cr += x_offset + y_offset*s->uvlinesize;
2002 x_offset += 8*s->mb_x;
2003 y_offset += 8*s->mb_y;
2006 Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
2007 mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
2008 dest_y, dest_cb, dest_cr, x_offset, y_offset,
2009 qpix_op, chroma_op);
2012 chroma_op= chroma_avg;
2016 Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
2017 mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
2018 dest_y, dest_cb, dest_cr, x_offset, y_offset,
2019 qpix_op, chroma_op);
2023 static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2024 qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
2025 qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg)){
2026 MpegEncContext * const s = &h->s;
2027 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
2028 const int mb_type= s->current_picture.mb_type[mb_xy];
2030 assert(IS_INTER(mb_type));
2032 if(IS_16X16(mb_type)){
2033 mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
2034 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
2035 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2036 }else if(IS_16X8(mb_type)){
2037 mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
2038 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
2039 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2040 mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
2041 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
2042 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2043 }else if(IS_8X16(mb_type)){
2044 mc_part(h, 0, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 0, 0,
2045 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2046 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2047 mc_part(h, 4, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 4, 0,
2048 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2049 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2053 assert(IS_8X8(mb_type));
2056 const int sub_mb_type= h->sub_mb_type[i];
2058 int x_offset= (i&1)<<2;
2059 int y_offset= (i&2)<<1;
2061 if(IS_SUB_8X8(sub_mb_type)){
2062 mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2063 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2064 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2065 }else if(IS_SUB_8X4(sub_mb_type)){
2066 mc_part(h, n , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2067 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2068 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2069 mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
2070 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2071 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2072 }else if(IS_SUB_4X8(sub_mb_type)){
2073 mc_part(h, n , 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2074 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2075 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2076 mc_part(h, n+1, 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
2077 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2078 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2081 assert(IS_SUB_4X4(sub_mb_type));
2083 int sub_x_offset= x_offset + 2*(j&1);
2084 int sub_y_offset= y_offset + (j&2);
2085 mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
2086 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2087 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2094 static void decode_init_vlc(H264Context *h){
2095 static int done = 0;
2101 init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5,
2102 &chroma_dc_coeff_token_len [0], 1, 1,
2103 &chroma_dc_coeff_token_bits[0], 1, 1);
2106 init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17,
2107 &coeff_token_len [i][0], 1, 1,
2108 &coeff_token_bits[i][0], 1, 1);
2112 init_vlc(&chroma_dc_total_zeros_vlc[i], CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
2113 &chroma_dc_total_zeros_len [i][0], 1, 1,
2114 &chroma_dc_total_zeros_bits[i][0], 1, 1);
2116 for(i=0; i<15; i++){
2117 init_vlc(&total_zeros_vlc[i], TOTAL_ZEROS_VLC_BITS, 16,
2118 &total_zeros_len [i][0], 1, 1,
2119 &total_zeros_bits[i][0], 1, 1);
2123 init_vlc(&run_vlc[i], RUN_VLC_BITS, 7,
2124 &run_len [i][0], 1, 1,
2125 &run_bits[i][0], 1, 1);
2127 init_vlc(&run7_vlc, RUN7_VLC_BITS, 16,
2128 &run_len [6][0], 1, 1,
2129 &run_bits[6][0], 1, 1);
2134 * Sets the intra prediction function pointers.
2136 static void init_pred_ptrs(H264Context *h){
2137 // MpegEncContext * const s = &h->s;
2139 h->pred4x4[VERT_PRED ]= pred4x4_vertical_c;
2140 h->pred4x4[HOR_PRED ]= pred4x4_horizontal_c;
2141 h->pred4x4[DC_PRED ]= pred4x4_dc_c;
2142 h->pred4x4[DIAG_DOWN_LEFT_PRED ]= pred4x4_down_left_c;
2143 h->pred4x4[DIAG_DOWN_RIGHT_PRED]= pred4x4_down_right_c;
2144 h->pred4x4[VERT_RIGHT_PRED ]= pred4x4_vertical_right_c;
2145 h->pred4x4[HOR_DOWN_PRED ]= pred4x4_horizontal_down_c;
2146 h->pred4x4[VERT_LEFT_PRED ]= pred4x4_vertical_left_c;
2147 h->pred4x4[HOR_UP_PRED ]= pred4x4_horizontal_up_c;
2148 h->pred4x4[LEFT_DC_PRED ]= pred4x4_left_dc_c;
2149 h->pred4x4[TOP_DC_PRED ]= pred4x4_top_dc_c;
2150 h->pred4x4[DC_128_PRED ]= pred4x4_128_dc_c;
2152 h->pred8x8[DC_PRED8x8 ]= pred8x8_dc_c;
2153 h->pred8x8[VERT_PRED8x8 ]= pred8x8_vertical_c;
2154 h->pred8x8[HOR_PRED8x8 ]= pred8x8_horizontal_c;
2155 h->pred8x8[PLANE_PRED8x8 ]= pred8x8_plane_c;
2156 h->pred8x8[LEFT_DC_PRED8x8]= pred8x8_left_dc_c;
2157 h->pred8x8[TOP_DC_PRED8x8 ]= pred8x8_top_dc_c;
2158 h->pred8x8[DC_128_PRED8x8 ]= pred8x8_128_dc_c;
2160 h->pred16x16[DC_PRED8x8 ]= pred16x16_dc_c;
2161 h->pred16x16[VERT_PRED8x8 ]= pred16x16_vertical_c;
2162 h->pred16x16[HOR_PRED8x8 ]= pred16x16_horizontal_c;
2163 h->pred16x16[PLANE_PRED8x8 ]= pred16x16_plane_c;
2164 h->pred16x16[LEFT_DC_PRED8x8]= pred16x16_left_dc_c;
2165 h->pred16x16[TOP_DC_PRED8x8 ]= pred16x16_top_dc_c;
2166 h->pred16x16[DC_128_PRED8x8 ]= pred16x16_128_dc_c;
2169 static void free_tables(H264Context *h){
2170 av_freep(&h->intra4x4_pred_mode);
2171 av_freep(&h->chroma_pred_mode_table);
2172 av_freep(&h->cbp_table);
2173 av_freep(&h->mvd_table[0]);
2174 av_freep(&h->mvd_table[1]);
2175 av_freep(&h->non_zero_count);
2176 av_freep(&h->slice_table_base);
2177 av_freep(&h->top_border);
2178 h->slice_table= NULL;
2180 av_freep(&h->mb2b_xy);
2181 av_freep(&h->mb2b8_xy);
2186 * needs widzh/height
2188 static int alloc_tables(H264Context *h){
2189 MpegEncContext * const s = &h->s;
2190 const int big_mb_num= s->mb_stride * (s->mb_height+1);
2193 CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8 * sizeof(uint8_t))
2195 CHECKED_ALLOCZ(h->non_zero_count , big_mb_num * 16 * sizeof(uint8_t))
2196 CHECKED_ALLOCZ(h->slice_table_base , big_mb_num * sizeof(uint8_t))
2197 CHECKED_ALLOCZ(h->top_border , s->mb_width * (16+8+8) * sizeof(uint8_t))
2199 if( h->pps.cabac ) {
2200 CHECKED_ALLOCZ(h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t))
2201 CHECKED_ALLOCZ(h->cbp_table, big_mb_num * sizeof(uint16_t))
2202 CHECKED_ALLOCZ(h->mvd_table[0], 32*big_mb_num * sizeof(uint16_t));
2203 CHECKED_ALLOCZ(h->mvd_table[1], 32*big_mb_num * sizeof(uint16_t));
2206 memset(h->slice_table_base, -1, big_mb_num * sizeof(uint8_t));
2207 h->slice_table= h->slice_table_base + s->mb_stride + 1;
2209 CHECKED_ALLOCZ(h->mb2b_xy , big_mb_num * sizeof(uint16_t));
2210 CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint16_t));
2211 for(y=0; y<s->mb_height; y++){
2212 for(x=0; x<s->mb_width; x++){
2213 const int mb_xy= x + y*s->mb_stride;
2214 const int b_xy = 4*x + 4*y*h->b_stride;
2215 const int b8_xy= 2*x + 2*y*h->b8_stride;
2217 h->mb2b_xy [mb_xy]= b_xy;
2218 h->mb2b8_xy[mb_xy]= b8_xy;
2228 static void common_init(H264Context *h){
2229 MpegEncContext * const s = &h->s;
2231 s->width = s->avctx->width;
2232 s->height = s->avctx->height;
2233 s->codec_id= s->avctx->codec->id;
2237 s->unrestricted_mv=1;
2238 s->decode=1; //FIXME
2241 static int decode_init(AVCodecContext *avctx){
2242 H264Context *h= avctx->priv_data;
2243 MpegEncContext * const s = &h->s;
2245 MPV_decode_defaults(s);
2250 s->out_format = FMT_H264;
2251 s->workaround_bugs= avctx->workaround_bugs;
2254 // s->decode_mb= ff_h263_decode_mb;
2256 avctx->pix_fmt= PIX_FMT_YUV420P;
2263 static void frame_start(H264Context *h){
2264 MpegEncContext * const s = &h->s;
2267 MPV_frame_start(s, s->avctx);
2268 ff_er_frame_start(s);
2271 assert(s->linesize && s->uvlinesize);
2273 for(i=0; i<16; i++){
2274 h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
2275 h->chroma_subblock_offset[i]= 2*((scan8[i] - scan8[0])&7) + 2*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2278 h->block_offset[16+i]=
2279 h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2282 // s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
2285 static inline void backup_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){
2286 MpegEncContext * const s = &h->s;
2290 src_cb -= uvlinesize;
2291 src_cr -= uvlinesize;
2293 h->left_border[0]= h->top_border[s->mb_x][15];
2294 for(i=1; i<17; i++){
2295 h->left_border[i]= src_y[15+i* linesize];
2298 *(uint64_t*)(h->top_border[s->mb_x]+0)= *(uint64_t*)(src_y + 16*linesize);
2299 *(uint64_t*)(h->top_border[s->mb_x]+8)= *(uint64_t*)(src_y +8+16*linesize);
2301 if(!(s->flags&CODEC_FLAG_GRAY)){
2302 h->left_border[17 ]= h->top_border[s->mb_x][16+7];
2303 h->left_border[17+9]= h->top_border[s->mb_x][24+7];
2305 h->left_border[i+17 ]= src_cb[7+i*uvlinesize];
2306 h->left_border[i+17+9]= src_cr[7+i*uvlinesize];
2308 *(uint64_t*)(h->top_border[s->mb_x]+16)= *(uint64_t*)(src_cb+8*uvlinesize);
2309 *(uint64_t*)(h->top_border[s->mb_x]+24)= *(uint64_t*)(src_cr+8*uvlinesize);
2313 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){
2314 MpegEncContext * const s = &h->s;
2318 src_y -= linesize + 1;
2319 src_cb -= uvlinesize + 1;
2320 src_cr -= uvlinesize + 1;
2322 #define XCHG(a,b,t,xchg)\
2328 for(i=0; i<17; i++){
2329 XCHG(h->left_border[i ], src_y [i* linesize], temp8, xchg);
2332 XCHG(*(uint64_t*)(h->top_border[s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
2333 XCHG(*(uint64_t*)(h->top_border[s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
2335 if(!(s->flags&CODEC_FLAG_GRAY)){
2337 XCHG(h->left_border[i+17 ], src_cb[i*uvlinesize], temp8, xchg);
2338 XCHG(h->left_border[i+17+9], src_cr[i*uvlinesize], temp8, xchg);
2340 XCHG(*(uint64_t*)(h->top_border[s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
2341 XCHG(*(uint64_t*)(h->top_border[s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
2345 static void hl_decode_mb(H264Context *h){
2346 MpegEncContext * const s = &h->s;
2347 const int mb_x= s->mb_x;
2348 const int mb_y= s->mb_y;
2349 const int mb_xy= mb_x + mb_y*s->mb_stride;
2350 const int mb_type= s->current_picture.mb_type[mb_xy];
2351 uint8_t *dest_y, *dest_cb, *dest_cr;
2352 int linesize, uvlinesize /*dct_offset*/;
2361 dest_y = s->current_picture.data[0] + (mb_y * 16* s->linesize ) + mb_x * 16;
2362 dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2363 dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2365 if (h->mb_field_decoding_flag) {
2366 linesize = s->linesize * 2;
2367 uvlinesize = s->uvlinesize * 2;
2368 if(mb_y&1){ //FIXME move out of this func?
2369 dest_y -= s->linesize*15;
2370 dest_cb-= s->linesize*7;
2371 dest_cr-= s->linesize*7;
2374 linesize = s->linesize;
2375 uvlinesize = s->uvlinesize;
2376 // dct_offset = s->linesize * 16;
2379 if(IS_INTRA(mb_type)){
2380 if(h->deblocking_filter)
2381 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1);
2383 if(!(s->flags&CODEC_FLAG_GRAY)){
2384 h->pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
2385 h->pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
2388 if(IS_INTRA4x4(mb_type)){
2390 for(i=0; i<16; i++){
2391 uint8_t * const ptr= dest_y + h->block_offset[i];
2392 uint8_t *topright= ptr + 4 - linesize;
2393 const int topright_avail= (h->topright_samples_available<<i)&0x8000;
2394 const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
2397 if(!topright_avail){
2398 tr= ptr[3 - linesize]*0x01010101;
2399 topright= (uint8_t*) &tr;
2400 }else if(i==5 && h->deblocking_filter){
2401 tr= *(uint32_t*)h->top_border[mb_x+1];
2402 topright= (uint8_t*) &tr;
2405 h->pred4x4[ dir ](ptr, topright, linesize);
2406 if(h->non_zero_count_cache[ scan8[i] ]){
2407 if(s->codec_id == CODEC_ID_H264)
2408 h264_add_idct_c(ptr, h->mb + i*16, linesize);
2410 svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
2415 h->pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
2416 if(s->codec_id == CODEC_ID_H264)
2417 h264_luma_dc_dequant_idct_c(h->mb, s->qscale);
2419 svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
2421 if(h->deblocking_filter)
2422 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);
2423 }else if(s->codec_id == CODEC_ID_H264){
2424 hl_motion(h, dest_y, dest_cb, dest_cr,
2425 s->dsp.put_h264_qpel_pixels_tab, s->dsp.put_h264_chroma_pixels_tab,
2426 s->dsp.avg_h264_qpel_pixels_tab, s->dsp.avg_h264_chroma_pixels_tab);
2430 if(!IS_INTRA4x4(mb_type)){
2431 if(s->codec_id == CODEC_ID_H264){
2432 for(i=0; i<16; i++){
2433 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
2434 uint8_t * const ptr= dest_y + h->block_offset[i];
2435 h264_add_idct_c(ptr, h->mb + i*16, linesize);
2439 for(i=0; i<16; i++){
2440 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
2441 uint8_t * const ptr= dest_y + h->block_offset[i];
2442 svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
2448 if(!(s->flags&CODEC_FLAG_GRAY)){
2449 chroma_dc_dequant_idct_c(h->mb + 16*16, h->chroma_qp);
2450 chroma_dc_dequant_idct_c(h->mb + 16*16+4*16, h->chroma_qp);
2451 if(s->codec_id == CODEC_ID_H264){
2452 for(i=16; i<16+4; i++){
2453 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2454 uint8_t * const ptr= dest_cb + h->block_offset[i];
2455 h264_add_idct_c(ptr, h->mb + i*16, uvlinesize);
2458 for(i=20; i<20+4; i++){
2459 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2460 uint8_t * const ptr= dest_cr + h->block_offset[i];
2461 h264_add_idct_c(ptr, h->mb + i*16, uvlinesize);
2465 for(i=16; i<16+4; i++){
2466 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2467 uint8_t * const ptr= dest_cb + h->block_offset[i];
2468 svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
2471 for(i=20; i<20+4; i++){
2472 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2473 uint8_t * const ptr= dest_cr + h->block_offset[i];
2474 svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
2479 if(h->deblocking_filter) {
2480 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2481 filter_mb(h, mb_x, mb_y, dest_y, dest_cb, dest_cr);
2486 * fills the default_ref_list.
2488 static int fill_default_ref_list(H264Context *h){
2489 MpegEncContext * const s = &h->s;
2491 Picture sorted_short_ref[16];
2493 if(h->slice_type==B_TYPE){
2497 for(out_i=0; out_i<h->short_ref_count; out_i++){
2501 for(i=0; i<h->short_ref_count; i++){
2502 const int poc= h->short_ref[i]->poc;
2503 if(poc > limit && poc < best_poc){
2509 assert(best_i != -1);
2512 sorted_short_ref[out_i]= *h->short_ref[best_i];
2516 if(s->picture_structure == PICT_FRAME){
2517 if(h->slice_type==B_TYPE){
2518 const int current_poc= s->current_picture_ptr->poc;
2521 for(list=0; list<2; list++){
2524 for(i=0; i<h->short_ref_count && index < h->ref_count[list]; i++){
2525 const int i2= list ? h->short_ref_count - i - 1 : i;
2526 const int poc= sorted_short_ref[i2].poc;
2528 if(sorted_short_ref[i2].reference != 3) continue; //FIXME refernce field shit
2530 if((list==1 && poc > current_poc) || (list==0 && poc < current_poc)){
2531 h->default_ref_list[list][index ]= sorted_short_ref[i2];
2532 h->default_ref_list[list][index++].pic_id= sorted_short_ref[i2].frame_num;
2536 for(i=0; i<h->long_ref_count && index < h->ref_count[ list ]; i++){
2537 if(h->long_ref[i]->reference != 3) continue;
2539 h->default_ref_list[ list ][index ]= *h->long_ref[i];
2540 h->default_ref_list[ list ][index++].pic_id= i;;
2543 if(h->long_ref_count > 1 && h->short_ref_count==0){
2544 Picture temp= h->default_ref_list[1][0];
2545 h->default_ref_list[1][0] = h->default_ref_list[1][1];
2546 h->default_ref_list[1][0] = temp;
2549 if(index < h->ref_count[ list ])
2550 memset(&h->default_ref_list[list][index], 0, sizeof(Picture)*(h->ref_count[ list ] - index));
2554 for(i=0; i<h->short_ref_count && index < h->ref_count[0]; i++){
2555 if(h->short_ref[i]->reference != 3) continue; //FIXME refernce field shit
2556 h->default_ref_list[0][index ]= *h->short_ref[i];
2557 h->default_ref_list[0][index++].pic_id= h->short_ref[i]->frame_num;
2559 for(i=0; i<h->long_ref_count && index < h->ref_count[0]; i++){
2560 if(h->long_ref[i]->reference != 3) continue;
2561 h->default_ref_list[0][index ]= *h->long_ref[i];
2562 h->default_ref_list[0][index++].pic_id= i;;
2564 if(index < h->ref_count[0])
2565 memset(&h->default_ref_list[0][index], 0, sizeof(Picture)*(h->ref_count[0] - index));
2568 if(h->slice_type==B_TYPE){
2570 //FIXME second field balh
2576 static int decode_ref_pic_list_reordering(H264Context *h){
2577 MpegEncContext * const s = &h->s;
2580 if(h->slice_type==I_TYPE || h->slice_type==SI_TYPE) return 0; //FIXME move beofre func
2582 for(list=0; list<2; list++){
2583 memcpy(h->ref_list[list], h->default_ref_list[list], sizeof(Picture)*h->ref_count[list]);
2585 if(get_bits1(&s->gb)){
2586 int pred= h->curr_pic_num;
2589 for(index=0; ; index++){
2590 int reordering_of_pic_nums_idc= get_ue_golomb(&s->gb);
2595 if(index >= h->ref_count[list]){
2596 av_log(h->s.avctx, AV_LOG_ERROR, "reference count overflow\n");
2600 if(reordering_of_pic_nums_idc<3){
2601 if(reordering_of_pic_nums_idc<2){
2602 const int abs_diff_pic_num= get_ue_golomb(&s->gb) + 1;
2604 if(abs_diff_pic_num >= h->max_pic_num){
2605 av_log(h->s.avctx, AV_LOG_ERROR, "abs_diff_pic_num overflow\n");
2609 if(reordering_of_pic_nums_idc == 0) pred-= abs_diff_pic_num;
2610 else pred+= abs_diff_pic_num;
2611 pred &= h->max_pic_num - 1;
2613 for(i= h->ref_count[list]-1; i>=index; i--){
2614 if(h->ref_list[list][i].pic_id == pred && h->ref_list[list][i].long_ref==0)
2618 pic_id= get_ue_golomb(&s->gb); //long_term_pic_idx
2620 for(i= h->ref_count[list]-1; i>=index; i--){
2621 if(h->ref_list[list][i].pic_id == pic_id && h->ref_list[list][i].long_ref==1)
2627 av_log(h->s.avctx, AV_LOG_ERROR, "reference picture missing during reorder\n");
2628 memset(&h->ref_list[list][index], 0, sizeof(Picture)); //FIXME
2629 }else if(i > index){
2630 Picture tmp= h->ref_list[list][i];
2631 for(; i>index; i--){
2632 h->ref_list[list][i]= h->ref_list[list][i-1];
2634 h->ref_list[list][index]= tmp;
2636 }else if(reordering_of_pic_nums_idc==3)
2639 av_log(h->s.avctx, AV_LOG_ERROR, "illegal reordering_of_pic_nums_idc\n");
2645 if(h->slice_type!=B_TYPE) break;
2650 static int pred_weight_table(H264Context *h){
2651 MpegEncContext * const s = &h->s;
2654 h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
2655 h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
2657 for(list=0; list<2; list++){
2658 for(i=0; i<h->ref_count[list]; i++){
2659 int luma_weight_flag, chroma_weight_flag;
2661 luma_weight_flag= get_bits1(&s->gb);
2662 if(luma_weight_flag){
2663 h->luma_weight[list][i]= get_se_golomb(&s->gb);
2664 h->luma_offset[list][i]= get_se_golomb(&s->gb);
2667 chroma_weight_flag= get_bits1(&s->gb);
2668 if(chroma_weight_flag){
2671 h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);
2672 h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);
2676 if(h->slice_type != B_TYPE) break;
2682 * instantaneos decoder refresh.
2684 static void idr(H264Context *h){
2687 for(i=0; i<h->long_ref_count; i++){
2688 h->long_ref[i]->reference=0;
2689 h->long_ref[i]= NULL;
2691 h->long_ref_count=0;
2693 for(i=0; i<h->short_ref_count; i++){
2694 h->short_ref[i]->reference=0;
2695 h->short_ref[i]= NULL;
2697 h->short_ref_count=0;
2702 * @return the removed picture or NULL if an error occures
2704 static Picture * remove_short(H264Context *h, int frame_num){
2705 MpegEncContext * const s = &h->s;
2708 if(s->avctx->debug&FF_DEBUG_MMCO)
2709 av_log(h->s.avctx, AV_LOG_DEBUG, "remove short %d count %d\n", frame_num, h->short_ref_count);
2711 for(i=0; i<h->short_ref_count; i++){
2712 Picture *pic= h->short_ref[i];
2713 if(s->avctx->debug&FF_DEBUG_MMCO)
2714 av_log(h->s.avctx, AV_LOG_DEBUG, "%d %d %p\n", i, pic->frame_num, pic);
2715 if(pic->frame_num == frame_num){
2716 h->short_ref[i]= NULL;
2717 memmove(&h->short_ref[i], &h->short_ref[i+1], (h->short_ref_count - i - 1)*sizeof(Picture*));
2718 h->short_ref_count--;
2727 * @return the removed picture or NULL if an error occures
2729 static Picture * remove_long(H264Context *h, int i){
2732 if(i >= h->long_ref_count) return NULL;
2733 pic= h->long_ref[i];
2734 if(pic==NULL) return NULL;
2736 h->long_ref[i]= NULL;
2737 memmove(&h->long_ref[i], &h->long_ref[i+1], (h->long_ref_count - i - 1)*sizeof(Picture*));
2738 h->long_ref_count--;
2744 * Executes the reference picture marking (memory management control operations).
2746 static int execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count){
2747 MpegEncContext * const s = &h->s;
2749 int current_is_long=0;
2752 if((s->avctx->debug&FF_DEBUG_MMCO) && mmco_count==0)
2753 av_log(h->s.avctx, AV_LOG_DEBUG, "no mmco here\n");
2755 for(i=0; i<mmco_count; i++){
2756 if(s->avctx->debug&FF_DEBUG_MMCO)
2757 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);
2759 switch(mmco[i].opcode){
2760 case MMCO_SHORT2UNUSED:
2761 pic= remove_short(h, mmco[i].short_frame_num);
2762 if(pic==NULL) return -1;
2765 case MMCO_SHORT2LONG:
2766 pic= remove_long(h, mmco[i].long_index);
2767 if(pic) pic->reference=0;
2769 h->long_ref[ mmco[i].long_index ]= remove_short(h, mmco[i].short_frame_num);
2770 h->long_ref[ mmco[i].long_index ]->long_ref=1;
2772 case MMCO_LONG2UNUSED:
2773 pic= remove_long(h, mmco[i].long_index);
2774 if(pic==NULL) return -1;
2778 pic= remove_long(h, mmco[i].long_index);
2779 if(pic) pic->reference=0;
2781 h->long_ref[ mmco[i].long_index ]= s->current_picture_ptr;
2782 h->long_ref[ mmco[i].long_index ]->long_ref=1;
2783 h->long_ref_count++;
2787 case MMCO_SET_MAX_LONG:
2788 assert(mmco[i].long_index <= 16);
2789 while(mmco[i].long_index < h->long_ref_count){
2790 pic= remove_long(h, mmco[i].long_index);
2793 while(mmco[i].long_index > h->long_ref_count){
2794 h->long_ref[ h->long_ref_count++ ]= NULL;
2798 while(h->short_ref_count){
2799 pic= remove_short(h, h->short_ref[0]->frame_num);
2802 while(h->long_ref_count){
2803 pic= remove_long(h, h->long_ref_count-1);
2811 if(!current_is_long){
2812 pic= remove_short(h, s->current_picture_ptr->frame_num);
2815 av_log(h->s.avctx, AV_LOG_ERROR, "illegal short term buffer state detected\n");
2818 if(h->short_ref_count)
2819 memmove(&h->short_ref[1], &h->short_ref[0], h->short_ref_count*sizeof(Picture*));
2821 h->short_ref[0]= s->current_picture_ptr;
2822 h->short_ref[0]->long_ref=0;
2823 h->short_ref_count++;
2829 static int decode_ref_pic_marking(H264Context *h){
2830 MpegEncContext * const s = &h->s;
2833 if(h->nal_unit_type == NAL_IDR_SLICE){ //FIXME fields
2834 s->broken_link= get_bits1(&s->gb) -1;
2835 h->mmco[0].long_index= get_bits1(&s->gb) - 1; // current_long_term_idx
2836 if(h->mmco[0].long_index == -1)
2839 h->mmco[0].opcode= MMCO_LONG;
2843 if(get_bits1(&s->gb)){ // adaptive_ref_pic_marking_mode_flag
2844 for(i= h->mmco_index; i<MAX_MMCO_COUNT; i++) {
2845 MMCOOpcode opcode= get_ue_golomb(&s->gb);;
2847 h->mmco[i].opcode= opcode;
2848 if(opcode==MMCO_SHORT2UNUSED || opcode==MMCO_SHORT2LONG){
2849 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
2850 /* if(h->mmco[i].short_frame_num >= h->short_ref_count || h->short_ref[ h->mmco[i].short_frame_num ] == NULL){
2851 fprintf(stderr, "illegal short ref in memory management control operation %d\n", mmco);
2855 if(opcode==MMCO_SHORT2LONG || opcode==MMCO_LONG2UNUSED || opcode==MMCO_LONG || opcode==MMCO_SET_MAX_LONG){
2856 h->mmco[i].long_index= get_ue_golomb(&s->gb);
2857 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){
2858 av_log(h->s.avctx, AV_LOG_ERROR, "illegal long ref in memory management control operation %d\n", opcode);
2863 if(opcode > MMCO_LONG){
2864 av_log(h->s.avctx, AV_LOG_ERROR, "illegal memory management control operation %d\n", opcode);
2870 assert(h->long_ref_count + h->short_ref_count <= h->sps.ref_frame_count);
2872 if(h->long_ref_count + h->short_ref_count == h->sps.ref_frame_count){ //FIXME fields
2873 h->mmco[0].opcode= MMCO_SHORT2UNUSED;
2874 h->mmco[0].short_frame_num= h->short_ref[ h->short_ref_count - 1 ]->frame_num;
2884 static int init_poc(H264Context *h){
2885 MpegEncContext * const s = &h->s;
2886 const int max_frame_num= 1<<h->sps.log2_max_frame_num;
2889 if(h->nal_unit_type == NAL_IDR_SLICE){
2890 h->frame_num_offset= 0;
2892 if(h->frame_num < h->prev_frame_num)
2893 h->frame_num_offset= h->prev_frame_num_offset + max_frame_num;
2895 h->frame_num_offset= h->prev_frame_num_offset;
2898 if(h->sps.poc_type==0){
2899 const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
2901 if (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
2902 h->poc_msb = h->prev_poc_msb + max_poc_lsb;
2903 else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
2904 h->poc_msb = h->prev_poc_msb - max_poc_lsb;
2906 h->poc_msb = h->prev_poc_msb;
2907 //printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
2909 field_poc[1] = h->poc_msb + h->poc_lsb;
2910 if(s->picture_structure == PICT_FRAME)
2911 field_poc[1] += h->delta_poc_bottom;
2912 }else if(h->sps.poc_type==1){
2913 int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
2916 if(h->sps.poc_cycle_length != 0)
2917 abs_frame_num = h->frame_num_offset + h->frame_num;
2921 if(h->nal_ref_idc==0 && abs_frame_num > 0)
2924 expected_delta_per_poc_cycle = 0;
2925 for(i=0; i < h->sps.poc_cycle_length; i++)
2926 expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
2928 if(abs_frame_num > 0){
2929 int poc_cycle_cnt = (abs_frame_num - 1) / h->sps.poc_cycle_length;
2930 int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
2932 expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
2933 for(i = 0; i <= frame_num_in_poc_cycle; i++)
2934 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
2938 if(h->nal_ref_idc == 0)
2939 expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
2941 field_poc[0] = expectedpoc + h->delta_poc[0];
2942 field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
2944 if(s->picture_structure == PICT_FRAME)
2945 field_poc[1] += h->delta_poc[1];
2948 if(h->nal_unit_type == NAL_IDR_SLICE){
2951 if(h->nal_ref_idc) poc= 2*(h->frame_num_offset + h->frame_num);
2952 else poc= 2*(h->frame_num_offset + h->frame_num) - 1;
2958 if(s->picture_structure != PICT_BOTTOM_FIELD)
2959 s->current_picture_ptr->field_poc[0]= field_poc[0];
2960 if(s->picture_structure != PICT_TOP_FIELD)
2961 s->current_picture_ptr->field_poc[1]= field_poc[1];
2962 if(s->picture_structure == PICT_FRAME) // FIXME field pix?
2963 s->current_picture_ptr->poc= FFMIN(field_poc[0], field_poc[1]);
2969 * decodes a slice header.
2970 * this will allso call MPV_common_init() and frame_start() as needed
2972 static int decode_slice_header(H264Context *h){
2973 MpegEncContext * const s = &h->s;
2974 int first_mb_in_slice, pps_id;
2975 int num_ref_idx_active_override_flag;
2976 static const uint8_t slice_type_map[5]= {P_TYPE, B_TYPE, I_TYPE, SP_TYPE, SI_TYPE};
2978 s->current_picture.reference= h->nal_ref_idc != 0;
2980 first_mb_in_slice= get_ue_golomb(&s->gb);
2982 h->slice_type= get_ue_golomb(&s->gb);
2983 if(h->slice_type > 9){
2984 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);
2986 if(h->slice_type > 4){
2988 h->slice_type_fixed=1;
2990 h->slice_type_fixed=0;
2992 h->slice_type= slice_type_map[ h->slice_type ];
2994 s->pict_type= h->slice_type; // to make a few old func happy, its wrong though
2996 pps_id= get_ue_golomb(&s->gb);
2998 av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
3001 h->pps= h->pps_buffer[pps_id];
3002 if(h->pps.slice_group_count == 0){
3003 av_log(h->s.avctx, AV_LOG_ERROR, "non existing PPS referenced\n");
3007 h->sps= h->sps_buffer[ h->pps.sps_id ];
3008 if(h->sps.log2_max_frame_num == 0){
3009 av_log(h->s.avctx, AV_LOG_ERROR, "non existing SPS referenced\n");
3013 s->mb_width= h->sps.mb_width;
3014 s->mb_height= h->sps.mb_height;
3016 h->b_stride= s->mb_width*4;
3017 h->b8_stride= s->mb_width*2;
3019 s->mb_x = first_mb_in_slice % s->mb_width;
3020 s->mb_y = first_mb_in_slice / s->mb_width; //FIXME AFFW
3022 s->width = 16*s->mb_width - 2*(h->sps.crop_left + h->sps.crop_right );
3023 if(h->sps.frame_mbs_only_flag)
3024 s->height= 16*s->mb_height - 2*(h->sps.crop_top + h->sps.crop_bottom);
3026 s->height= 16*s->mb_height - 4*(h->sps.crop_top + h->sps.crop_bottom); //FIXME recheck
3028 if (s->context_initialized
3029 && ( s->width != s->avctx->width || s->height != s->avctx->height)) {
3033 if (!s->context_initialized) {
3034 if (MPV_common_init(s) < 0)
3039 s->avctx->width = s->width;
3040 s->avctx->height = s->height;
3041 s->avctx->sample_aspect_ratio= h->sps.sar;
3044 if(first_mb_in_slice == 0){
3048 s->current_picture_ptr->frame_num= //FIXME frame_num cleanup
3049 h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
3051 if(h->sps.frame_mbs_only_flag){
3052 s->picture_structure= PICT_FRAME;
3054 if(get_bits1(&s->gb)) //field_pic_flag
3055 s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
3057 s->picture_structure= PICT_FRAME;
3060 if(s->picture_structure==PICT_FRAME){
3061 h->curr_pic_num= h->frame_num;
3062 h->max_pic_num= 1<< h->sps.log2_max_frame_num;
3064 h->curr_pic_num= 2*h->frame_num;
3065 h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
3068 if(h->nal_unit_type == NAL_IDR_SLICE){
3069 get_ue_golomb(&s->gb); /* idr_pic_id */
3072 if(h->sps.poc_type==0){
3073 h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
3075 if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
3076 h->delta_poc_bottom= get_se_golomb(&s->gb);
3080 if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
3081 h->delta_poc[0]= get_se_golomb(&s->gb);
3083 if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
3084 h->delta_poc[1]= get_se_golomb(&s->gb);
3089 if(h->pps.redundant_pic_cnt_present){
3090 h->redundant_pic_count= get_ue_golomb(&s->gb);
3093 //set defaults, might be overriden a few line later
3094 h->ref_count[0]= h->pps.ref_count[0];
3095 h->ref_count[1]= h->pps.ref_count[1];
3097 if(h->slice_type == P_TYPE || h->slice_type == SP_TYPE || h->slice_type == B_TYPE){
3098 if(h->slice_type == B_TYPE){
3099 h->direct_spatial_mv_pred= get_bits1(&s->gb);
3101 num_ref_idx_active_override_flag= get_bits1(&s->gb);
3103 if(num_ref_idx_active_override_flag){
3104 h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
3105 if(h->slice_type==B_TYPE)
3106 h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
3108 if(h->ref_count[0] > 32 || h->ref_count[1] > 32){
3109 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
3115 if(first_mb_in_slice == 0){
3116 fill_default_ref_list(h);
3119 decode_ref_pic_list_reordering(h);
3121 if( (h->pps.weighted_pred && (h->slice_type == P_TYPE || h->slice_type == SP_TYPE ))
3122 || (h->pps.weighted_bipred_idc==1 && h->slice_type==B_TYPE ) )
3123 pred_weight_table(h);
3125 if(s->current_picture.reference)
3126 decode_ref_pic_marking(h);
3128 if( h->slice_type != I_TYPE && h->slice_type != SI_TYPE && h->pps.cabac )
3129 h->cabac_init_idc = get_ue_golomb(&s->gb);
3131 h->last_qscale_diff = 0;
3132 s->qscale = h->pps.init_qp + get_se_golomb(&s->gb);
3133 if(s->qscale<0 || s->qscale>51){
3134 av_log(s->avctx, AV_LOG_ERROR, "QP %d out of range\n", s->qscale);
3137 //FIXME qscale / qp ... stuff
3138 if(h->slice_type == SP_TYPE){
3139 get_bits1(&s->gb); /* sp_for_switch_flag */
3141 if(h->slice_type==SP_TYPE || h->slice_type == SI_TYPE){
3142 get_se_golomb(&s->gb); /* slice_qs_delta */
3145 h->deblocking_filter = 1;
3146 h->slice_alpha_c0_offset = 0;
3147 h->slice_beta_offset = 0;
3148 if( h->pps.deblocking_filter_parameters_present ) {
3149 h->deblocking_filter= get_ue_golomb(&s->gb);
3150 if(h->deblocking_filter < 2)
3151 h->deblocking_filter^= 1; // 1<->0
3153 if( h->deblocking_filter ) {
3154 h->slice_alpha_c0_offset = get_se_golomb(&s->gb) << 1;
3155 h->slice_beta_offset = get_se_golomb(&s->gb) << 1;
3160 if( h->pps.num_slice_groups > 1 && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
3161 slice_group_change_cycle= get_bits(&s->gb, ?);
3164 if(s->avctx->debug&FF_DEBUG_PICT_INFO){
3165 av_log(h->s.avctx, AV_LOG_DEBUG, "mb:%d %c pps:%d frame:%d poc:%d/%d ref:%d/%d qp:%d loop:%d\n",
3167 av_get_pict_type_char(h->slice_type),
3168 pps_id, h->frame_num,
3169 s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
3170 h->ref_count[0], h->ref_count[1],
3172 h->deblocking_filter
3182 static inline int get_level_prefix(GetBitContext *gb){
3186 OPEN_READER(re, gb);
3187 UPDATE_CACHE(re, gb);
3188 buf=GET_CACHE(re, gb);
3190 log= 32 - av_log2(buf);
3192 print_bin(buf>>(32-log), log);
3193 printf("%5d %2d %3d lpr @%5d in %s get_level_prefix\n", buf>>(32-log), log, log-1, get_bits_count(gb), __FILE__);
3196 LAST_SKIP_BITS(re, gb, log);
3197 CLOSE_READER(re, gb);
3203 * decodes a residual block.
3204 * @param n block index
3205 * @param scantable scantable
3206 * @param max_coeff number of coefficients in the block
3207 * @return <0 if an error occured
3209 static int decode_residual(H264Context *h, GetBitContext *gb, DCTELEM *block, int n, const uint8_t *scantable, int qp, int max_coeff){
3210 MpegEncContext * const s = &h->s;
3211 const uint16_t *qmul= dequant_coeff[qp];
3212 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};
3213 int level[16], run[16];
3214 int suffix_length, zeros_left, coeff_num, coeff_token, total_coeff, i, trailing_ones;
3216 //FIXME put trailing_onex into the context
3218 if(n == CHROMA_DC_BLOCK_INDEX){
3219 coeff_token= get_vlc2(gb, chroma_dc_coeff_token_vlc.table, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 1);
3220 total_coeff= coeff_token>>2;
3222 if(n == LUMA_DC_BLOCK_INDEX){
3223 total_coeff= pred_non_zero_count(h, 0);
3224 coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
3225 total_coeff= coeff_token>>2;
3227 total_coeff= pred_non_zero_count(h, n);
3228 coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
3229 total_coeff= coeff_token>>2;
3230 h->non_zero_count_cache[ scan8[n] ]= total_coeff;
3234 //FIXME set last_non_zero?
3239 trailing_ones= coeff_token&3;
3240 tprintf("trailing:%d, total:%d\n", trailing_ones, total_coeff);
3241 assert(total_coeff<=16);
3243 for(i=0; i<trailing_ones; i++){
3244 level[i]= 1 - 2*get_bits1(gb);
3247 suffix_length= total_coeff > 10 && trailing_ones < 3;
3249 for(; i<total_coeff; i++){
3250 const int prefix= get_level_prefix(gb);
3251 int level_code, mask;
3253 if(prefix<14){ //FIXME try to build a large unified VLC table for all this
3255 level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
3257 level_code= (prefix<<suffix_length); //part
3258 }else if(prefix==14){
3260 level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
3262 level_code= prefix + get_bits(gb, 4); //part
3263 }else if(prefix==15){
3264 level_code= (prefix<<suffix_length) + get_bits(gb, 12); //part
3265 if(suffix_length==0) level_code+=15; //FIXME doesnt make (much)sense
3267 av_log(h->s.avctx, AV_LOG_ERROR, "prefix too large at %d %d\n", s->mb_x, s->mb_y);
3271 if(i==trailing_ones && i<3) level_code+= 2; //FIXME split first iteration
3273 mask= -(level_code&1);
3274 level[i]= (((2+level_code)>>1) ^ mask) - mask;
3276 if(suffix_length==0) suffix_length=1; //FIXME split first iteration
3279 if(ABS(level[i]) > (3<<(suffix_length-1)) && suffix_length<6) suffix_length++;
3281 if((2+level_code)>>1) > (3<<(suffix_length-1)) && suffix_length<6) suffix_length++;
3282 /* ? == prefix > 2 or sth */
3284 tprintf("level: %d suffix_length:%d\n", level[i], suffix_length);
3287 if(total_coeff == max_coeff)
3290 if(n == CHROMA_DC_BLOCK_INDEX)
3291 zeros_left= get_vlc2(gb, chroma_dc_total_zeros_vlc[ total_coeff-1 ].table, CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 1);
3293 zeros_left= get_vlc2(gb, total_zeros_vlc[ total_coeff-1 ].table, TOTAL_ZEROS_VLC_BITS, 1);
3296 for(i=0; i<total_coeff-1; i++){
3299 else if(zeros_left < 7){
3300 run[i]= get_vlc2(gb, run_vlc[zeros_left-1].table, RUN_VLC_BITS, 1);
3302 run[i]= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2);
3304 zeros_left -= run[i];
3308 av_log(h->s.avctx, AV_LOG_ERROR, "negative number of zero coeffs at %d %d\n", s->mb_x, s->mb_y);
3312 for(; i<total_coeff-1; i++){
3320 for(i=total_coeff-1; i>=0; i--){ //FIXME merge into rundecode?
3323 coeff_num += run[i] + 1; //FIXME add 1 earlier ?
3324 j= scantable[ coeff_num ];
3329 for(i=total_coeff-1; i>=0; i--){ //FIXME merge into rundecode?
3332 coeff_num += run[i] + 1; //FIXME add 1 earlier ?
3333 j= scantable[ coeff_num ];
3335 block[j]= level[i] * qmul[j];
3336 // printf("%d %d ", block[j], qmul[j]);
3343 * decodes a macroblock
3344 * @returns 0 if ok, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
3346 static int decode_mb_cavlc(H264Context *h){
3347 MpegEncContext * const s = &h->s;
3348 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
3349 int mb_type, partition_count, cbp;
3351 s->dsp.clear_blocks(h->mb); //FIXME avoid if allready clear (move after skip handlong?
3353 tprintf("pic:%d mb:%d/%d\n", h->frame_num, s->mb_x, s->mb_y);
3354 cbp = 0; /* avoid warning. FIXME: find a solution without slowing
3356 if(h->slice_type != I_TYPE && h->slice_type != SI_TYPE){
3357 if(s->mb_skip_run==-1)
3358 s->mb_skip_run= get_ue_golomb(&s->gb);
3360 if (s->mb_skip_run--) {
3364 mb_type= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0;
3366 memset(h->non_zero_count[mb_xy], 0, 16);
3367 memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui
3369 if(h->sps.mb_aff && s->mb_skip_run==0 && (s->mb_y&1)==0){
3370 h->mb_field_decoding_flag= get_bits1(&s->gb);
3373 if(h->mb_field_decoding_flag)
3374 mb_type|= MB_TYPE_INTERLACED;
3376 fill_caches(h, mb_type); //FIXME check what is needed and what not ...
3377 pred_pskip_motion(h, &mx, &my);
3378 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
3379 fill_rectangle( h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4);
3380 write_back_motion(h, mb_type);
3382 s->current_picture.mb_type[mb_xy]= mb_type; //FIXME SKIP type
3383 s->current_picture.qscale_table[mb_xy]= s->qscale;
3384 h->slice_table[ mb_xy ]= h->slice_num;
3386 h->prev_mb_skiped= 1;
3390 if(h->sps.mb_aff /* && !field pic FIXME needed? */){
3392 h->mb_field_decoding_flag = get_bits1(&s->gb);
3394 h->mb_field_decoding_flag=0; //FIXME som ed note ?!
3396 h->prev_mb_skiped= 0;
3398 mb_type= get_ue_golomb(&s->gb);
3399 if(h->slice_type == B_TYPE){
3401 partition_count= b_mb_type_info[mb_type].partition_count;
3402 mb_type= b_mb_type_info[mb_type].type;
3405 goto decode_intra_mb;
3407 }else if(h->slice_type == P_TYPE /*|| h->slice_type == SP_TYPE */){
3409 partition_count= p_mb_type_info[mb_type].partition_count;
3410 mb_type= p_mb_type_info[mb_type].type;
3413 goto decode_intra_mb;
3416 assert(h->slice_type == I_TYPE);
3419 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);
3423 cbp= i_mb_type_info[mb_type].cbp;
3424 h->intra16x16_pred_mode= i_mb_type_info[mb_type].pred_mode;
3425 mb_type= i_mb_type_info[mb_type].type;
3428 if(h->mb_field_decoding_flag)
3429 mb_type |= MB_TYPE_INTERLACED;
3431 s->current_picture.mb_type[mb_xy]= mb_type;
3432 h->slice_table[ mb_xy ]= h->slice_num;
3434 if(IS_INTRA_PCM(mb_type)){
3438 // we assume these blocks are very rare so we dont optimize it
3439 align_get_bits(&s->gb);
3441 ptr= s->gb.buffer + get_bits_count(&s->gb);
3443 for(y=0; y<16; y++){
3444 const int index= 4*(y&3) + 64*(y>>2);
3445 for(x=0; x<16; x++){
3446 h->mb[index + (x&3) + 16*(x>>2)]= *(ptr++);
3450 const int index= 256 + 4*(y&3) + 32*(y>>2);
3452 h->mb[index + (x&3) + 16*(x>>2)]= *(ptr++);
3456 const int index= 256 + 64 + 4*(y&3) + 32*(y>>2);
3458 h->mb[index + (x&3) + 16*(x>>2)]= *(ptr++);
3462 skip_bits(&s->gb, 384); //FIXME check /fix the bitstream readers
3464 //FIXME deblock filter, non_zero_count_cache init ...
3465 memset(h->non_zero_count[mb_xy], 16, 16);
3466 s->current_picture.qscale_table[mb_xy]= s->qscale;
3471 fill_caches(h, mb_type);
3474 if(IS_INTRA(mb_type)){
3475 // init_top_left_availability(h);
3476 if(IS_INTRA4x4(mb_type)){
3479 // fill_intra4x4_pred_table(h);
3480 for(i=0; i<16; i++){
3481 const int mode_coded= !get_bits1(&s->gb);
3482 const int predicted_mode= pred_intra_mode(h, i);
3486 const int rem_mode= get_bits(&s->gb, 3);
3487 if(rem_mode<predicted_mode)
3492 mode= predicted_mode;
3495 h->intra4x4_pred_mode_cache[ scan8[i] ] = mode;
3497 write_back_intra_pred_mode(h);
3498 if( check_intra4x4_pred_mode(h) < 0)
3501 h->intra16x16_pred_mode= check_intra_pred_mode(h, h->intra16x16_pred_mode);
3502 if(h->intra16x16_pred_mode < 0)
3505 h->chroma_pred_mode= get_ue_golomb(&s->gb);
3507 h->chroma_pred_mode= check_intra_pred_mode(h, h->chroma_pred_mode);
3508 if(h->chroma_pred_mode < 0)
3510 }else if(partition_count==4){
3511 int i, j, sub_partition_count[4], list, ref[2][4];
3513 if(h->slice_type == B_TYPE){
3515 h->sub_mb_type[i]= get_ue_golomb(&s->gb);
3516 if(h->sub_mb_type[i] >=13){
3517 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);
3520 sub_partition_count[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
3521 h->sub_mb_type[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].type;
3524 assert(h->slice_type == P_TYPE || h->slice_type == SP_TYPE); //FIXME SP correct ?
3526 h->sub_mb_type[i]= get_ue_golomb(&s->gb);
3527 if(h->sub_mb_type[i] >=4){
3528 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);
3531 sub_partition_count[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
3532 h->sub_mb_type[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].type;
3536 for(list=0; list<2; list++){
3537 const int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list];
3538 if(ref_count == 0) continue;
3540 if(IS_DIR(h->sub_mb_type[i], 0, list) && !IS_DIRECT(h->sub_mb_type[i])){
3541 ref[list][i] = get_te0_golomb(&s->gb, ref_count); //FIXME init to 0 before and skip?
3549 for(list=0; list<2; list++){
3550 const int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list];
3551 if(ref_count == 0) continue;
3554 h->ref_cache[list][ scan8[4*i] ]=h->ref_cache[list][ scan8[4*i]+1 ]=
3555 h->ref_cache[list][ scan8[4*i]+8 ]=h->ref_cache[list][ scan8[4*i]+9 ]= ref[list][i];
3557 if(IS_DIR(h->sub_mb_type[i], 0, list) && !IS_DIRECT(h->sub_mb_type[i])){
3558 const int sub_mb_type= h->sub_mb_type[i];
3559 const int block_width= (sub_mb_type & (MB_TYPE_16x16|MB_TYPE_16x8)) ? 2 : 1;
3560 for(j=0; j<sub_partition_count[i]; j++){
3562 const int index= 4*i + block_width*j;
3563 int16_t (* mv_cache)[2]= &h->mv_cache[list][ scan8[index] ];
3564 pred_motion(h, index, block_width, list, h->ref_cache[list][ scan8[index] ], &mx, &my);
3565 mx += get_se_golomb(&s->gb);
3566 my += get_se_golomb(&s->gb);
3567 tprintf("final mv:%d %d\n", mx, my);
3569 if(IS_SUB_8X8(sub_mb_type)){
3570 mv_cache[ 0 ][0]= mv_cache[ 1 ][0]=
3571 mv_cache[ 8 ][0]= mv_cache[ 9 ][0]= mx;
3572 mv_cache[ 0 ][1]= mv_cache[ 1 ][1]=
3573 mv_cache[ 8 ][1]= mv_cache[ 9 ][1]= my;
3574 }else if(IS_SUB_8X4(sub_mb_type)){
3575 mv_cache[ 0 ][0]= mv_cache[ 1 ][0]= mx;
3576 mv_cache[ 0 ][1]= mv_cache[ 1 ][1]= my;
3577 }else if(IS_SUB_4X8(sub_mb_type)){
3578 mv_cache[ 0 ][0]= mv_cache[ 8 ][0]= mx;
3579 mv_cache[ 0 ][1]= mv_cache[ 8 ][1]= my;
3581 assert(IS_SUB_4X4(sub_mb_type));
3582 mv_cache[ 0 ][0]= mx;
3583 mv_cache[ 0 ][1]= my;
3587 uint32_t *p= (uint32_t *)&h->mv_cache[list][ scan8[4*i] ][0];
3593 }else if(!IS_DIRECT(mb_type)){
3594 int list, mx, my, i;
3595 //FIXME we should set ref_idx_l? to 0 if we use that later ...
3596 if(IS_16X16(mb_type)){
3597 for(list=0; list<2; list++){
3598 if(h->ref_count[0]>0){
3599 if(IS_DIR(mb_type, 0, list)){
3600 const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
3601 fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, val, 1);
3605 for(list=0; list<2; list++){
3606 if(IS_DIR(mb_type, 0, list)){
3607 pred_motion(h, 0, 4, list, h->ref_cache[list][ scan8[0] ], &mx, &my);
3608 mx += get_se_golomb(&s->gb);
3609 my += get_se_golomb(&s->gb);
3610 tprintf("final mv:%d %d\n", mx, my);
3612 fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, pack16to32(mx,my), 4);
3616 else if(IS_16X8(mb_type)){
3617 for(list=0; list<2; list++){
3618 if(h->ref_count[list]>0){
3620 if(IS_DIR(mb_type, i, list)){
3621 const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
3622 fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, val, 1);
3627 for(list=0; list<2; list++){
3629 if(IS_DIR(mb_type, i, list)){
3630 pred_16x8_motion(h, 8*i, list, h->ref_cache[list][scan8[0] + 16*i], &mx, &my);
3631 mx += get_se_golomb(&s->gb);
3632 my += get_se_golomb(&s->gb);
3633 tprintf("final mv:%d %d\n", mx, my);
3635 fill_rectangle(h->mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, pack16to32(mx,my), 4);
3640 assert(IS_8X16(mb_type));
3641 for(list=0; list<2; list++){
3642 if(h->ref_count[list]>0){
3644 if(IS_DIR(mb_type, i, list)){ //FIXME optimize
3645 const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
3646 fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, val, 1);
3651 for(list=0; list<2; list++){
3653 if(IS_DIR(mb_type, i, list)){
3654 pred_8x16_motion(h, i*4, list, h->ref_cache[list][ scan8[0] + 2*i ], &mx, &my);
3655 mx += get_se_golomb(&s->gb);
3656 my += get_se_golomb(&s->gb);
3657 tprintf("final mv:%d %d\n", mx, my);
3659 fill_rectangle(h->mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, pack16to32(mx,my), 4);
3666 if(IS_INTER(mb_type))
3667 write_back_motion(h, mb_type);
3669 if(!IS_INTRA16x16(mb_type)){
3670 cbp= get_ue_golomb(&s->gb);
3672 av_log(h->s.avctx, AV_LOG_ERROR, "cbp too large (%d) at %d %d\n", cbp, s->mb_x, s->mb_y);
3676 if(IS_INTRA4x4(mb_type))
3677 cbp= golomb_to_intra4x4_cbp[cbp];
3679 cbp= golomb_to_inter_cbp[cbp];
3682 if(cbp || IS_INTRA16x16(mb_type)){
3683 int i8x8, i4x4, chroma_idx;
3684 int chroma_qp, dquant;
3685 GetBitContext *gb= IS_INTRA(mb_type) ? h->intra_gb_ptr : h->inter_gb_ptr;
3686 const uint8_t *scan, *dc_scan;
3688 // fill_non_zero_count_cache(h);
3690 if(IS_INTERLACED(mb_type)){
3692 dc_scan= luma_dc_field_scan;
3695 dc_scan= luma_dc_zigzag_scan;
3698 dquant= get_se_golomb(&s->gb);
3700 if( dquant > 25 || dquant < -26 ){
3701 av_log(h->s.avctx, AV_LOG_ERROR, "dquant out of range (%d) at %d %d\n", dquant, s->mb_x, s->mb_y);
3705 s->qscale += dquant;
3706 if(((unsigned)s->qscale) > 51){
3707 if(s->qscale<0) s->qscale+= 52;
3708 else s->qscale-= 52;
3711 h->chroma_qp= chroma_qp= get_chroma_qp(h, s->qscale);
3712 if(IS_INTRA16x16(mb_type)){
3713 if( decode_residual(h, h->intra_gb_ptr, h->mb, LUMA_DC_BLOCK_INDEX, dc_scan, s->qscale, 16) < 0){
3714 return -1; //FIXME continue if partotioned and other retirn -1 too
3717 assert((cbp&15) == 0 || (cbp&15) == 15);
3720 for(i8x8=0; i8x8<4; i8x8++){
3721 for(i4x4=0; i4x4<4; i4x4++){
3722 const int index= i4x4 + 4*i8x8;
3723 if( decode_residual(h, h->intra_gb_ptr, h->mb + 16*index, index, scan + 1, s->qscale, 15) < 0 ){
3729 fill_rectangle(&h->non_zero_count_cache[scan8[0]], 4, 4, 8, 0, 1);
3732 for(i8x8=0; i8x8<4; i8x8++){
3733 if(cbp & (1<<i8x8)){
3734 for(i4x4=0; i4x4<4; i4x4++){
3735 const int index= i4x4 + 4*i8x8;
3737 if( decode_residual(h, gb, h->mb + 16*index, index, scan, s->qscale, 16) <0 ){
3742 uint8_t * const nnz= &h->non_zero_count_cache[ scan8[4*i8x8] ];
3743 nnz[0] = nnz[1] = nnz[8] = nnz[9] = 0;
3749 for(chroma_idx=0; chroma_idx<2; chroma_idx++)
3750 if( decode_residual(h, gb, h->mb + 256 + 16*4*chroma_idx, CHROMA_DC_BLOCK_INDEX, chroma_dc_scan, chroma_qp, 4) < 0){
3756 for(chroma_idx=0; chroma_idx<2; chroma_idx++){
3757 for(i4x4=0; i4x4<4; i4x4++){
3758 const int index= 16 + 4*chroma_idx + i4x4;
3759 if( decode_residual(h, gb, h->mb + 16*index, index, scan + 1, chroma_qp, 15) < 0){
3765 uint8_t * const nnz= &h->non_zero_count_cache[0];
3766 nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =
3767 nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
3770 uint8_t * const nnz= &h->non_zero_count_cache[0];
3771 fill_rectangle(&nnz[scan8[0]], 4, 4, 8, 0, 1);
3772 nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =
3773 nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
3775 s->current_picture.qscale_table[mb_xy]= s->qscale;
3776 write_back_non_zero_count(h);
3781 static int decode_cabac_mb_type( H264Context *h ) {
3782 MpegEncContext * const s = &h->s;
3784 if( h->slice_type == I_TYPE ) {
3785 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
3789 if( s->mb_x > 0 && !IS_INTRA4x4( s->current_picture.mb_type[mb_xy-1] ) )
3791 if( s->mb_y > 0 && !IS_INTRA4x4( s->current_picture.mb_type[mb_xy-s->mb_stride] ) )
3794 if( get_cabac( &h->cabac, &h->cabac_state[3+ctx] ) == 0 )
3795 return 0; /* I4x4 */
3797 if( get_cabac_terminate( &h->cabac ) )
3798 return 25; /* PCM */
3800 mb_type = 1; /* I16x16 */
3801 if( get_cabac( &h->cabac, &h->cabac_state[3+3] ) )
3802 mb_type += 12; /* cbp_luma != 0 */
3804 if( get_cabac( &h->cabac, &h->cabac_state[3+4] ) ) {
3805 if( get_cabac( &h->cabac, &h->cabac_state[3+5] ) )
3806 mb_type += 4 * 2; /* cbp_chroma == 2 */
3808 mb_type += 4 * 1; /* cbp_chroma == 1 */
3810 if( get_cabac( &h->cabac, &h->cabac_state[3+6] ) )
3812 if( get_cabac( &h->cabac, &h->cabac_state[3+7] ) )
3816 } else if( h->slice_type == P_TYPE ) {
3817 if( get_cabac( &h->cabac, &h->cabac_state[14] ) == 0 ) {
3819 if( get_cabac( &h->cabac, &h->cabac_state[15] ) == 0 ) {
3820 if( get_cabac( &h->cabac, &h->cabac_state[16] ) == 0 )
3821 return 0; /* P_L0_D16x16; */
3823 return 3; /* P_8x8; */
3825 if( get_cabac( &h->cabac, &h->cabac_state[17] ) == 0 )
3826 return 2; /* P_L0_D8x16; */
3828 return 1; /* P_L0_D16x8; */
3833 if( get_cabac( &h->cabac, &h->cabac_state[17] ) == 0 )
3834 return 5+0; /* I_4x4 */
3835 if( get_cabac_terminate( &h->cabac ) )
3836 return 5+25; /*I_PCM */
3837 mb_type = 5+1; /* I16x16 */
3838 if( get_cabac( &h->cabac, &h->cabac_state[17+1] ) )
3839 mb_type += 12; /* cbp_luma != 0 */
3841 if( get_cabac( &h->cabac, &h->cabac_state[17+2] ) ) {
3842 if( get_cabac( &h->cabac, &h->cabac_state[17+2] ) )
3843 mb_type += 4 * 2; /* cbp_chroma == 2 */
3845 mb_type += 4 * 1; /* cbp_chroma == 1 */
3847 if( get_cabac( &h->cabac, &h->cabac_state[17+3] ) )
3849 if( get_cabac( &h->cabac, &h->cabac_state[17+3] ) )
3855 /* TODO do others frames types */
3860 static int decode_cabac_mb_skip( H264Context *h) {
3861 MpegEncContext * const s = &h->s;
3862 const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
3863 const int mba_xy = mb_xy - 1;
3864 const int mbb_xy = mb_xy - s->mb_stride;
3867 if( s->mb_x > 0 && !IS_SKIP( s->current_picture.mb_type[mba_xy] ) )
3869 if( s->mb_y > 0 && !IS_SKIP( s->current_picture.mb_type[mbb_xy] ) )
3872 if( h->slice_type == P_TYPE || h->slice_type == SP_TYPE)
3873 return get_cabac( &h->cabac, &h->cabac_state[11+ctx] );
3875 return get_cabac( &h->cabac, &h->cabac_state[24+ctx] );
3878 static int decode_cabac_mb_intra4x4_pred_mode( H264Context *h, int pred_mode ) {
3881 if( get_cabac( &h->cabac, &h->cabac_state[68] ) )
3884 if( get_cabac( &h->cabac, &h->cabac_state[69] ) )
3886 if( get_cabac( &h->cabac, &h->cabac_state[69] ) )
3888 if( get_cabac( &h->cabac, &h->cabac_state[69] ) )
3890 if( mode >= pred_mode )
3896 static int decode_cabac_mb_chroma_pre_mode( H264Context *h) {
3897 MpegEncContext * const s = &h->s;
3898 const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
3899 const int mba_xy = mb_xy - 1;
3900 const int mbb_xy = mb_xy - s->mb_stride;
3904 /* No need to test for IS_INTRA4x4 and IS_INTRA16x16, as we set chroma_pred_mode_table to 0 */
3905 if( s->mb_x > 0 && h->chroma_pred_mode_table[mba_xy] != 0 )
3908 if( s->mb_y > 0 && h->chroma_pred_mode_table[mbb_xy] != 0 )
3911 if( get_cabac( &h->cabac, &h->cabac_state[64+ctx] ) == 0 )
3914 if( get_cabac( &h->cabac, &h->cabac_state[64+3] ) == 0 )
3916 if( get_cabac( &h->cabac, &h->cabac_state[64+3] ) == 0 )
3922 static const uint8_t block_idx_x[16] = {
3923 0, 1, 0, 1, 2, 3, 2, 3, 0, 1, 0, 1, 2, 3, 2, 3
3925 static const uint8_t block_idx_y[16] = {
3926 0, 0, 1, 1, 0, 0, 1, 1, 2, 2, 3, 3, 2, 2, 3, 3
3928 static const uint8_t block_idx_xy[4][4] = {
3935 static int decode_cabac_mb_cbp_luma( H264Context *h) {
3936 MpegEncContext * const s = &h->s;
3937 const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
3942 h->cbp_table[mb_xy] = 0; /* FIXME aaahahahah beurk */
3944 for( i8x8 = 0; i8x8 < 4; i8x8++ ) {
3950 x = block_idx_x[4*i8x8];
3951 y = block_idx_y[4*i8x8];
3955 else if( s->mb_x > 0 )
3960 else if( s->mb_y > 0 )
3961 mbb_xy = mb_xy - s->mb_stride;
3963 /* No need to test for skip as we put 0 for skip block */
3965 int i8x8a = block_idx_xy[(x-1)&0x03][y]/4;
3966 if( ((h->cbp_table[mba_xy] >> i8x8a)&0x01) == 0 )
3971 int i8x8b = block_idx_xy[x][(y-1)&0x03]/4;
3972 if( ((h->cbp_table[mbb_xy] >> i8x8b)&0x01) == 0 )
3976 if( get_cabac( &h->cabac, &h->cabac_state[73 + ctx] ) ) {
3978 h->cbp_table[mb_xy] = cbp; /* FIXME aaahahahah beurk */
3983 static int decode_cabac_mb_cbp_chroma( H264Context *h) {
3984 MpegEncContext * const s = &h->s;
3985 const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
3989 /* No need to test for skip */
3991 cbp_a = (h->cbp_table[mb_xy-1]>>4)&0x03;
3996 cbp_b = (h->cbp_table[mb_xy-s->mb_stride]>>4)&0x03;
4001 if( cbp_a > 0 ) ctx++;
4002 if( cbp_b > 0 ) ctx += 2;
4003 if( get_cabac( &h->cabac, &h->cabac_state[77 + ctx] ) == 0 )
4007 if( cbp_a == 2 ) ctx++;
4008 if( cbp_b == 2 ) ctx += 2;
4009 if( get_cabac( &h->cabac, &h->cabac_state[77 + ctx] ) )
4014 static int decode_cabac_mb_dqp( H264Context *h) {
4015 MpegEncContext * const s = &h->s;
4021 mbn_xy = s->mb_x + s->mb_y*s->mb_stride - 1;
4023 mbn_xy = s->mb_width - 1 + (s->mb_y-1)*s->mb_stride;
4025 if( mbn_xy >= 0 && h->last_qscale_diff != 0 && ( IS_INTRA16x16(s->current_picture.mb_type[mbn_xy] ) || (h->cbp_table[mbn_xy]&0x3f) ) )
4028 while( get_cabac( &h->cabac, &h->cabac_state[60 + ctx] ) ) {
4039 return -(val + 1)/2;
4041 static int decode_cabac_mb_sub_type( H264Context *h ) {
4042 if( get_cabac( &h->cabac, &h->cabac_state[21] ) )
4044 if( !get_cabac( &h->cabac, &h->cabac_state[22] ) )
4046 if( get_cabac( &h->cabac, &h->cabac_state[23] ) )
4051 static int decode_cabac_mb_ref( H264Context *h, int list, int n ) {
4052 int refa = h->ref_cache[list][scan8[n] - 1];
4053 int refb = h->ref_cache[list][scan8[n] - 8];
4062 while( get_cabac( &h->cabac, &h->cabac_state[54+ctx] ) ) {
4072 static int decode_cabac_mb_mvd( H264Context *h, int list, int n, int l ) {
4073 int amvd = abs( h->mvd_cache[list][scan8[n] - 1][l] ) +
4074 abs( h->mvd_cache[list][scan8[n] - 8][l] );
4075 int ctxbase = (l == 0) ? 40 : 47;
4081 else if( amvd > 32 )
4086 while( mvd < 9 && get_cabac( &h->cabac, &h->cabac_state[ctxbase+ctx] ) ) {
4096 while( get_cabac_bypass( &h->cabac ) ) {
4101 if( get_cabac_bypass( &h->cabac ) )
4105 if( mvd != 0 && get_cabac_bypass( &h->cabac ) )
4111 static int get_cabac_cbf_ctx( H264Context *h, int cat, int idx ) {
4112 MpegEncContext * const s = &h->s;
4113 const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
4124 if( IS_INTRA16x16(s->current_picture.mb_type[mba_xy] ) )
4125 nza = h->cbp_table[mba_xy]&0x100;
4128 mbb_xy = mb_xy - s->mb_stride;
4129 if( IS_INTRA16x16(s->current_picture.mb_type[mbb_xy] ) )
4130 nzb = h->cbp_table[mbb_xy]&0x100;
4132 } else if( cat == 1 || cat == 2 ) {
4136 x = block_idx_x[idx];
4137 y = block_idx_y[idx];
4141 else if( s->mb_x > 0 )
4146 else if( s->mb_y > 0 )
4147 mbb_xy = mb_xy - s->mb_stride;
4149 /* No need to test for skip */
4151 i8x8a = block_idx_xy[(x-1)&0x03][y]/4;
4153 if( !IS_INTRA_PCM(s->current_picture.mb_type[mba_xy] ) &&
4154 ((h->cbp_table[mba_xy]&0x0f)>>i8x8a))
4155 nza = h->non_zero_count_cache[scan8[idx] - 1];
4159 i8x8b = block_idx_xy[x][(y-1)&0x03]/4;
4161 if( !IS_INTRA_PCM(s->current_picture.mb_type[mbb_xy] ) &&
4162 ((h->cbp_table[mbb_xy]&0x0f)>>i8x8b))
4163 nzb = h->non_zero_count_cache[scan8[idx] - 8];
4165 } else if( cat == 3 ) {
4169 if( !IS_INTRA_PCM(s->current_picture.mb_type[mba_xy] ) &&
4170 (h->cbp_table[mba_xy]&0x30) )
4171 nza = (h->cbp_table[mba_xy]>>(6+idx))&0x01;
4174 mbb_xy = mb_xy - s->mb_stride;
4176 if( !IS_INTRA_PCM(s->current_picture.mb_type[mbb_xy] ) &&
4177 (h->cbp_table[mbb_xy]&0x30) )
4178 nzb = (h->cbp_table[mbb_xy]>>(6+idx))&0x01;
4180 } else if( cat == 4 ) {
4181 int idxc = idx % 4 ;
4182 if( idxc == 1 || idxc == 3 )
4184 else if( s->mb_x > 0 )
4187 if( idxc == 2 || idxc == 3 )
4189 else if( s->mb_y > 0 )
4190 mbb_xy = mb_xy - s->mb_stride;
4193 !IS_INTRA_PCM(s->current_picture.mb_type[mba_xy] ) &&
4194 (h->cbp_table[mba_xy]&0x30) == 0x20 )
4195 nza = h->non_zero_count_cache[scan8[16+idx] - 1];
4198 !IS_INTRA_PCM(s->current_picture.mb_type[mbb_xy] ) &&
4199 (h->cbp_table[mbb_xy]&0x30) == 0x20 )
4200 nzb = h->non_zero_count_cache[scan8[16+idx] - 8];
4203 if( ( mba_xy < 0 && IS_INTRA( s->current_picture.mb_type[mb_xy] ) ) ||
4204 ( mba_xy >= 0 && IS_INTRA_PCM(s->current_picture.mb_type[mba_xy] ) ) ||
4208 if( ( mbb_xy < 0 && IS_INTRA( s->current_picture.mb_type[mb_xy] ) ) ||
4209 ( mbb_xy >= 0 && IS_INTRA_PCM(s->current_picture.mb_type[mbb_xy] ) ) ||
4213 return ctx + 4 * cat;
4216 static int decode_cabac_residual( H264Context *h, DCTELEM *block, int cat, int n, const uint8_t *scantable, int qp, int max_coeff) {
4217 const int mb_xy = h->s.mb_x + h->s.mb_y*h->s.mb_stride;
4218 const uint16_t *qmul= dequant_coeff[qp];
4219 static const int significant_coeff_flag_offset[5] = { 0, 15, 29, 44, 47 };
4220 static const int last_significant_coeff_flag_offset[5] = { 0, 15, 29, 44, 47 };
4221 static const int coeff_abs_level_m1_offset[5] = { 0, 10, 20, 30, 39 };
4226 int coeff_count = 0;
4231 int abslevelgt1 = 0;
4233 /* cat: 0-> DC 16x16 n = 0
4234 * 1-> AC 16x16 n = luma4x4idx
4235 * 2-> Luma4x4 n = luma4x4idx
4236 * 3-> DC Chroma n = iCbCr
4237 * 4-> AC Chroma n = 4 * iCbCr + chroma4x4idx
4240 /* read coded block flag */
4241 if( get_cabac( &h->cabac, &h->cabac_state[85 + get_cabac_cbf_ctx( h, cat, n ) ] ) == 0 ) {
4242 if( cat == 1 || cat == 2 )
4243 h->non_zero_count_cache[scan8[n]] = 0;
4245 h->non_zero_count_cache[scan8[16+n]] = 0;
4250 while( last < max_coeff - 1 ) {
4251 int ctx = FFMIN( last, max_coeff - 2 );
4253 if( get_cabac( &h->cabac, &h->cabac_state[105+significant_coeff_flag_offset[cat]+ctx] ) == 0 ) {
4259 if( get_cabac( &h->cabac, &h->cabac_state[166+last_significant_coeff_flag_offset[cat]+ctx] ) ) {
4260 while( last < max_coeff ) {
4267 if( last == max_coeff -1 ) {
4272 if( cat == 0 && coeff_count > 0 )
4273 h->cbp_table[mb_xy] |= 0x100;
4274 else if( cat == 1 || cat == 2 )
4275 h->non_zero_count_cache[scan8[n]] = coeff_count;
4276 else if( cat == 3 && coeff_count > 0 )
4277 h->cbp_table[mb_xy] |= 0x40 << n;
4279 h->non_zero_count_cache[scan8[16+n]] = coeff_count;
4281 for( i = coeff_count - 1; i >= 0; i-- ) {
4284 int ctx = (abslevelgt1 != 0 ? 0 : FFMIN( 4, abslevel1 + 1 )) + coeff_abs_level_m1_offset[cat];
4286 if( get_cabac( &h->cabac, &h->cabac_state[227+ctx] ) == 0 ) {
4290 ctx = 5 + FFMIN( 4, abslevelgt1 ) + coeff_abs_level_m1_offset[cat];
4291 while( coeff_abs_m1 < 14 && get_cabac( &h->cabac, &h->cabac_state[227+ctx] ) ) {
4296 if( coeff_abs_m1 >= 14 ) {
4298 while( get_cabac_bypass( &h->cabac ) ) {
4299 coeff_abs_m1 += 1 << j;
4304 if( get_cabac_bypass( &h->cabac ) )
4305 coeff_abs_m1 += 1 << j ;
4308 if( get_cabac_bypass( &h->cabac ) )
4309 coeff[i] = -1 *( coeff_abs_m1 + 1 );
4311 coeff[i] = coeff_abs_m1 + 1;
4313 if( coeff_abs_m1 == 0 )
4319 if( cat == 0 || cat == 3 ) { /* DC */
4321 for( i = 0, j = 0; j < coeff_count; i++ ) {
4323 block[scantable[i]] = coeff[j];
4331 for( i = 0, j = 0; j < coeff_count; i++ ) {
4333 block[scantable[i]] = coeff[j] * qmul[scantable[i]];
4343 * decodes a macroblock
4344 * @returns 0 if ok, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
4346 static int decode_mb_cabac(H264Context *h) {
4347 MpegEncContext * const s = &h->s;
4348 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
4349 int mb_type, partition_count, cbp = 0;
4351 s->dsp.clear_blocks(h->mb); //FIXME avoid if allready clear (move after skip handlong?)
4353 if( h->slice_type == B_TYPE ) {
4354 av_log( h->s.avctx, AV_LOG_ERROR, "B-frame not supported with CABAC\n" );
4357 if( h->sps.mb_aff ) {
4358 av_log( h->s.avctx, AV_LOG_ERROR, "Fields not supported with CABAC\n" );
4362 if( h->slice_type != I_TYPE && h->slice_type != SI_TYPE ) {
4363 /* read skip flags */
4364 if( decode_cabac_mb_skip( h ) ) {
4368 mb_type= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0|MB_TYPE_SKIP;
4370 memset(h->non_zero_count[mb_xy], 0, 16);
4371 memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui
4373 if(h->sps.mb_aff && s->mb_skip_run==0 && (s->mb_y&1)==0){
4374 h->mb_field_decoding_flag= get_bits1(&s->gb);
4376 if(h->mb_field_decoding_flag)
4377 mb_type|= MB_TYPE_INTERLACED;
4380 fill_caches(h, mb_type); //FIXME check what is needed and what not ...
4381 pred_pskip_motion(h, &mx, &my);
4382 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
4383 fill_rectangle( h->mvd_cache[0][scan8[0]], 4, 4, 8, pack16to32(0,0), 4);
4384 fill_rectangle( h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4);
4385 write_back_motion(h, mb_type);
4387 s->current_picture.mb_type[mb_xy]= mb_type; //FIXME SKIP type
4388 s->current_picture.qscale_table[mb_xy]= s->qscale;
4389 h->slice_table[ mb_xy ]= h->slice_num;
4390 h->cbp_table[mb_xy] = 0;
4391 h->chroma_pred_mode_table[mb_xy] = 0;
4392 h->last_qscale_diff = 0;
4394 h->prev_mb_skiped= 1;
4400 h->prev_mb_skiped = 0;
4402 if( ( mb_type = decode_cabac_mb_type( h ) ) < 0 ) {
4403 av_log( h->s.avctx, AV_LOG_ERROR, "decode_cabac_mb_type failed\n" );
4407 if( h->slice_type == P_TYPE ) {
4409 partition_count= p_mb_type_info[mb_type].partition_count;
4410 mb_type= p_mb_type_info[mb_type].type;
4413 goto decode_intra_mb;
4416 assert(h->slice_type == I_TYPE);
4418 partition_count = 0;
4419 cbp= i_mb_type_info[mb_type].cbp;
4420 h->intra16x16_pred_mode= i_mb_type_info[mb_type].pred_mode;
4421 mb_type= i_mb_type_info[mb_type].type;
4424 if(h->mb_field_decoding_flag)
4425 mb_type |= MB_TYPE_INTERLACED;
4428 s->current_picture.mb_type[mb_xy]= mb_type;
4429 h->slice_table[ mb_xy ]= h->slice_num;
4431 if(IS_INTRA_PCM(mb_type)) {
4433 h->cbp_table[mb_xy] = 0xf +4*2;
4434 h->chroma_pred_mode_table[mb_xy] = 0;
4435 s->current_picture.qscale_table[mb_xy]= s->qscale;
4439 fill_caches(h, mb_type);
4441 if( IS_INTRA( mb_type ) ) {
4442 if( IS_INTRA4x4( mb_type ) ) {
4444 for( i = 0; i < 16; i++ ) {
4445 int pred = pred_intra_mode( h, i );
4446 h->intra4x4_pred_mode_cache[ scan8[i] ] = decode_cabac_mb_intra4x4_pred_mode( h, pred );
4448 //av_log( s->avctx, AV_LOG_ERROR, "i4x4 pred=%d mode=%d\n", pred, h->intra4x4_pred_mode_cache[ scan8[i] ] );
4450 write_back_intra_pred_mode(h);
4451 if( check_intra4x4_pred_mode(h) < 0 ) return -1;
4453 h->intra16x16_pred_mode= check_intra_pred_mode( h, h->intra16x16_pred_mode );
4454 if( h->intra16x16_pred_mode < 0 ) return -1;
4456 h->chroma_pred_mode_table[mb_xy] =
4457 h->chroma_pred_mode = decode_cabac_mb_chroma_pre_mode( h );
4459 h->chroma_pred_mode= check_intra_pred_mode( h, h->chroma_pred_mode );
4460 if( h->chroma_pred_mode < 0 ) return -1;
4461 } else if( partition_count == 4 ) {
4462 int i, j, sub_partition_count[4], list, ref[2][4];
4465 for( i = 0; i < 4; i++ ) {
4466 h->sub_mb_type[i] = decode_cabac_mb_sub_type( h );
4467 sub_partition_count[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
4468 h->sub_mb_type[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].type;
4471 for( list = 0; list < 2; list++ ) {
4472 if( h->ref_count[list] > 0 ) {
4473 for( i = 0; i < 4; i++ ) {
4474 if(IS_DIR(h->sub_mb_type[i], 0, list) && !IS_DIRECT(h->sub_mb_type[i])){
4475 if( h->ref_count[list] > 1 )
4476 ref[list][i] = decode_cabac_mb_ref( h, list, 4*i );
4482 h->ref_cache[list][ scan8[4*i] ]=h->ref_cache[list][ scan8[4*i]+1 ]=
4483 h->ref_cache[list][ scan8[4*i]+8 ]=h->ref_cache[list][ scan8[4*i]+9 ]= ref[list][i];
4488 for(list=0; list<2; list++){
4491 //h->ref_cache[list][ scan8[4*i] ]=h->ref_cache[list][ scan8[4*i]+1 ]=
4492 //h->ref_cache[list][ scan8[4*i]+8 ]=h->ref_cache[list][ scan8[4*i]+9 ]= ref[list][i];
4494 if(IS_DIR(h->sub_mb_type[i], 0, list) && !IS_DIRECT(h->sub_mb_type[i])){
4495 const int sub_mb_type= h->sub_mb_type[i];
4496 const int block_width= (sub_mb_type & (MB_TYPE_16x16|MB_TYPE_16x8)) ? 2 : 1;
4497 for(j=0; j<sub_partition_count[i]; j++){
4500 const int index= 4*i + block_width*j;
4501 int16_t (* mv_cache)[2]= &h->mv_cache[list][ scan8[index] ];
4502 int16_t (* mvd_cache)[2]= &h->mvd_cache[list][ scan8[index] ];
4503 pred_motion(h, index, block_width, list, h->ref_cache[list][ scan8[index] ], &mpx, &mpy);
4505 mx = mpx + decode_cabac_mb_mvd( h, list, index, 0 );
4506 my = mpy + decode_cabac_mb_mvd( h, list, index, 1 );
4507 tprintf("final mv:%d %d\n", mx, my);
4509 if(IS_SUB_8X8(sub_mb_type)){
4510 mv_cache[ 0 ][0]= mv_cache[ 1 ][0]=
4511 mv_cache[ 8 ][0]= mv_cache[ 9 ][0]= mx;
4512 mv_cache[ 0 ][1]= mv_cache[ 1 ][1]=
4513 mv_cache[ 8 ][1]= mv_cache[ 9 ][1]= my;
4515 mvd_cache[ 0 ][0]= mvd_cache[ 1 ][0]=
4516 mvd_cache[ 8 ][0]= mvd_cache[ 9 ][0]= mx - mpx;
4517 mvd_cache[ 0 ][1]= mvd_cache[ 1 ][1]=
4518 mvd_cache[ 8 ][1]= mvd_cache[ 9 ][1]= my - mpy;
4519 }else if(IS_SUB_8X4(sub_mb_type)){
4520 mv_cache[ 0 ][0]= mv_cache[ 1 ][0]= mx;
4521 mv_cache[ 0 ][1]= mv_cache[ 1 ][1]= my;
4523 mvd_cache[ 0 ][0]= mvd_cache[ 1 ][0]= mx- mpx;
4524 mvd_cache[ 0 ][1]= mvd_cache[ 1 ][1]= my - mpy;
4525 }else if(IS_SUB_4X8(sub_mb_type)){
4526 mv_cache[ 0 ][0]= mv_cache[ 8 ][0]= mx;
4527 mv_cache[ 0 ][1]= mv_cache[ 8 ][1]= my;
4529 mvd_cache[ 0 ][0]= mvd_cache[ 8 ][0]= mx - mpx;
4530 mvd_cache[ 0 ][1]= mvd_cache[ 8 ][1]= my - mpy;
4532 assert(IS_SUB_4X4(sub_mb_type));
4533 mv_cache[ 0 ][0]= mx;
4534 mv_cache[ 0 ][1]= my;
4536 mvd_cache[ 0 ][0]= mx - mpx;
4537 mvd_cache[ 0 ][1]= my - mpy;
4541 uint32_t *p= (uint32_t *)&h->mv_cache[list][ scan8[4*i] ][0];
4542 uint32_t *pd= (uint32_t *)&h->mvd_cache[list][ scan8[4*i] ][0];
4543 p[0] = p[1] = p[8] = p[9] = 0;
4544 pd[0]= pd[1]= pd[8]= pd[9]= 0;
4548 } else if( !IS_DIRECT(mb_type) ) {
4549 int list, mx, my, i, mpx, mpy;
4550 if(IS_16X16(mb_type)){
4551 for(list=0; list<2; list++){
4552 if(IS_DIR(mb_type, 0, list)){
4553 if(h->ref_count[list] > 0 ){
4554 const int ref = h->ref_count[list] > 1 ? decode_cabac_mb_ref( h, list, 0 ) : 0;
4555 fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, ref, 1);
4559 for(list=0; list<2; list++){
4560 if(IS_DIR(mb_type, 0, list)){
4561 pred_motion(h, 0, 4, list, h->ref_cache[list][ scan8[0] ], &mpx, &mpy);
4563 mx = mpx + decode_cabac_mb_mvd( h, list, 0, 0 );
4564 my = mpy + decode_cabac_mb_mvd( h, list, 0, 1 );
4565 tprintf("final mv:%d %d\n", mx, my);
4567 fill_rectangle(h->mvd_cache[list][ scan8[0] ], 4, 4, 8, pack16to32(mx-mpx,my-mpy), 4);
4568 fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, pack16to32(mx,my), 4);
4572 else if(IS_16X8(mb_type)){
4573 for(list=0; list<2; list++){
4574 if(h->ref_count[list]>0){
4576 if(IS_DIR(mb_type, i, list)){
4577 const int ref= h->ref_count[list] > 1 ? decode_cabac_mb_ref( h, list, 8*i ) : 0;
4578 fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, ref, 1);
4583 for(list=0; list<2; list++){
4585 if(IS_DIR(mb_type, i, list)){
4586 pred_16x8_motion(h, 8*i, list, h->ref_cache[list][scan8[0] + 16*i], &mpx, &mpy);
4587 mx = mpx + decode_cabac_mb_mvd( h, list, 8*i, 0 );
4588 my = mpy + decode_cabac_mb_mvd( h, list, 8*i, 1 );
4589 tprintf("final mv:%d %d\n", mx, my);
4591 fill_rectangle(h->mvd_cache[list][ scan8[0] + 16*i ], 4, 2, 8, pack16to32(mx-mpx,my-mpy), 4);
4592 fill_rectangle(h->mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, pack16to32(mx,my), 4);
4597 assert(IS_8X16(mb_type));
4598 for(list=0; list<2; list++){
4599 if(h->ref_count[list]>0){
4601 if(IS_DIR(mb_type, i, list)){ //FIXME optimize
4602 const int ref= h->ref_count[list] > 1 ? decode_cabac_mb_ref( h, list, 4*i ) : 0;
4603 fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, ref, 1);
4608 for(list=0; list<2; list++){
4610 if(IS_DIR(mb_type, i, list)){
4611 pred_8x16_motion(h, i*4, list, h->ref_cache[list][ scan8[0] + 2*i ], &mpx, &mpy);
4612 mx = mpx + decode_cabac_mb_mvd( h, list, 4*i, 0 );
4613 my = mpy + decode_cabac_mb_mvd( h, list, 4*i, 1 );
4615 tprintf("final mv:%d %d\n", mx, my);
4616 fill_rectangle(h->mvd_cache[list][ scan8[0] + 2*i ], 2, 4, 8, pack16to32(mx-mpx,my-mpy), 4);
4617 fill_rectangle(h->mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, pack16to32(mx,my), 4);
4624 if( IS_INTER( mb_type ) ) {
4625 h->chroma_pred_mode_table[mb_xy] = 0;
4626 write_back_motion( h, mb_type );
4629 if( !IS_INTRA16x16( mb_type ) ) {
4630 cbp = decode_cabac_mb_cbp_luma( h );
4631 cbp |= decode_cabac_mb_cbp_chroma( h ) << 4;
4634 h->cbp_table[mb_xy] = cbp;
4636 if( cbp || IS_INTRA16x16( mb_type ) ) {
4637 const uint8_t *scan, *dc_scan;
4640 if(IS_INTERLACED(mb_type)){
4642 dc_scan= luma_dc_field_scan;
4645 dc_scan= luma_dc_zigzag_scan;
4648 h->last_qscale_diff = dqp = decode_cabac_mb_dqp( h );
4650 if(((unsigned)s->qscale) > 51){
4651 if(s->qscale<0) s->qscale+= 52;
4652 else s->qscale-= 52;
4654 h->chroma_qp = get_chroma_qp(h, s->qscale);
4656 if( IS_INTRA16x16( mb_type ) ) {
4658 //av_log( s->avctx, AV_LOG_ERROR, "INTRA16x16 DC\n" );
4659 if( decode_cabac_residual( h, h->mb, 0, 0, dc_scan, s->qscale, 16) < 0)
4662 for( i = 0; i < 16; i++ ) {
4663 //av_log( s->avctx, AV_LOG_ERROR, "INTRA16x16 AC:%d\n", i );
4664 if( decode_cabac_residual(h, h->mb + 16*i, 1, i, scan + 1, s->qscale, 15) < 0 )
4668 fill_rectangle(&h->non_zero_count_cache[scan8[0]], 4, 4, 8, 0, 1);
4672 for( i8x8 = 0; i8x8 < 4; i8x8++ ) {
4673 if( cbp & (1<<i8x8) ) {
4674 for( i4x4 = 0; i4x4 < 4; i4x4++ ) {
4675 const int index = 4*i8x8 + i4x4;
4676 //av_log( s->avctx, AV_LOG_ERROR, "Luma4x4: %d\n", index );
4677 if( decode_cabac_residual(h, h->mb + 16*index, 2, index, scan, s->qscale, 16) < 0 )
4681 uint8_t * const nnz= &h->non_zero_count_cache[ scan8[4*i8x8] ];
4682 nnz[0] = nnz[1] = nnz[8] = nnz[9] = 0;
4689 for( c = 0; c < 2; c++ ) {
4690 //av_log( s->avctx, AV_LOG_ERROR, "INTRA C%d-DC\n",c );
4691 if( decode_cabac_residual(h, h->mb + 256 + 16*4*c, 3, c, chroma_dc_scan, h->chroma_qp, 4) < 0)
4698 for( c = 0; c < 2; c++ ) {
4699 for( i = 0; i < 4; i++ ) {
4700 const int index = 16 + 4 * c + i;
4701 //av_log( s->avctx, AV_LOG_ERROR, "INTRA C%d-AC %d\n",c, index - 16 );
4702 if( decode_cabac_residual(h, h->mb + 16*index, 4, index - 16, scan + 1, h->chroma_qp, 15) < 0)
4707 uint8_t * const nnz= &h->non_zero_count_cache[0];
4708 nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =
4709 nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
4712 memset( &h->non_zero_count_cache[8], 0, 8*5 );
4715 s->current_picture.qscale_table[mb_xy]= s->qscale;
4716 write_back_non_zero_count(h);
4722 static void filter_mb_edgev( H264Context *h, uint8_t *pix, int stride, int bS[4], int qp ) {
4724 const int index_a = clip( qp + h->slice_alpha_c0_offset, 0, 51 );
4725 const int alpha = alpha_table[index_a];
4726 const int beta = beta_table[clip( qp + h->slice_beta_offset, 0, 51 )];
4728 for( i = 0; i < 4; i++ ) {
4735 const int tc0 = tc0_table[index_a][bS[i] - 1];
4736 /* 4px edge length */
4737 for( d = 0; d < 4; d++ ) {
4738 const int p0 = pix[-1];
4739 const int p1 = pix[-2];
4740 const int p2 = pix[-3];
4741 const int q0 = pix[0];
4742 const int q1 = pix[1];
4743 const int q2 = pix[2];
4745 if( ABS( p0 - q0 ) < alpha &&
4746 ABS( p1 - p0 ) < beta &&
4747 ABS( q1 - q0 ) < beta ) {
4751 if( ABS( p2 - p0 ) < beta ) {
4752 pix[-2] = p1 + clip( ( p2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( p1 << 1 ) ) >> 1, -tc0, tc0 );
4755 if( ABS( q2 - q0 ) < beta ) {
4756 pix[1] = q1 + clip( ( q2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( q1 << 1 ) ) >> 1, -tc0, tc0 );
4760 i_delta = clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );
4761 pix[-1] = clip_uint8( p0 + i_delta ); /* p0' */
4762 pix[0] = clip_uint8( q0 - i_delta ); /* q0' */
4767 /* 4px edge length */
4768 for( d = 0; d < 4; d++ ) {
4769 const int p0 = pix[-1];
4770 const int p1 = pix[-2];
4771 const int p2 = pix[-3];
4773 const int q0 = pix[0];
4774 const int q1 = pix[1];
4775 const int q2 = pix[2];
4777 if( ABS( p0 - q0 ) < alpha &&
4778 ABS( p1 - p0 ) < beta &&
4779 ABS( q1 - q0 ) < beta ) {
4781 if(ABS( p0 - q0 ) < (( alpha >> 2 ) + 2 )){
4782 if( ABS( p2 - p0 ) < beta)
4784 const int p3 = pix[-4];
4786 pix[-1] = ( p2 + 2*p1 + 2*p0 + 2*q0 + q1 + 4 ) >> 3;
4787 pix[-2] = ( p2 + p1 + p0 + q0 + 2 ) >> 2;
4788 pix[-3] = ( 2*p3 + 3*p2 + p1 + p0 + q0 + 4 ) >> 3;
4791 pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
4793 if( ABS( q2 - q0 ) < beta)
4795 const int q3 = pix[3];
4797 pix[0] = ( p1 + 2*p0 + 2*q0 + 2*q1 + q2 + 4 ) >> 3;
4798 pix[1] = ( p0 + q0 + q1 + q2 + 2 ) >> 2;
4799 pix[2] = ( 2*q3 + 3*q2 + q1 + q0 + p0 + 4 ) >> 3;
4802 pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
4806 pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
4807 pix[ 0] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
4815 static void filter_mb_edgecv( H264Context *h, uint8_t *pix, int stride, int bS[4], int qp ) {
4817 const int index_a = clip( qp + h->slice_alpha_c0_offset, 0, 51 );
4818 const int alpha = alpha_table[index_a];
4819 const int beta = beta_table[clip( qp + h->slice_beta_offset, 0, 51 )];
4821 for( i = 0; i < 4; i++ ) {
4828 const int tc = tc0_table[index_a][bS[i] - 1] + 1;
4829 /* 2px edge length (because we use same bS than the one for luma) */
4830 for( d = 0; d < 2; d++ ){
4831 const int p0 = pix[-1];
4832 const int p1 = pix[-2];
4833 const int q0 = pix[0];
4834 const int q1 = pix[1];
4836 if( ABS( p0 - q0 ) < alpha &&
4837 ABS( p1 - p0 ) < beta &&
4838 ABS( q1 - q0 ) < beta ) {
4839 const int i_delta = clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );
4841 pix[-1] = clip_uint8( p0 + i_delta ); /* p0' */
4842 pix[0] = clip_uint8( q0 - i_delta ); /* q0' */
4847 /* 2px edge length (because we use same bS than the one for luma) */
4848 for( d = 0; d < 2; d++ ){
4849 const int p0 = pix[-1];
4850 const int p1 = pix[-2];
4851 const int q0 = pix[0];
4852 const int q1 = pix[1];
4854 if( ABS( p0 - q0 ) < alpha &&
4855 ABS( p1 - p0 ) < beta &&
4856 ABS( q1 - q0 ) < beta ) {
4858 pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2; /* p0' */
4859 pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2; /* q0' */
4867 static void filter_mb_edgeh( H264Context *h, uint8_t *pix, int stride, int bS[4], int qp ) {
4869 const int index_a = clip( qp + h->slice_alpha_c0_offset, 0, 51 );
4870 const int alpha = alpha_table[index_a];
4871 const int beta = beta_table[clip( qp + h->slice_beta_offset, 0, 51 )];
4872 const int pix_next = stride;
4874 for( i = 0; i < 4; i++ ) {
4881 const int tc0 = tc0_table[index_a][bS[i] - 1];
4882 /* 4px edge length */
4883 for( d = 0; d < 4; d++ ) {
4884 const int p0 = pix[-1*pix_next];
4885 const int p1 = pix[-2*pix_next];
4886 const int p2 = pix[-3*pix_next];
4887 const int q0 = pix[0];
4888 const int q1 = pix[1*pix_next];
4889 const int q2 = pix[2*pix_next];
4891 if( ABS( p0 - q0 ) < alpha &&
4892 ABS( p1 - p0 ) < beta &&
4893 ABS( q1 - q0 ) < beta ) {
4898 if( ABS( p2 - p0 ) < beta ) {
4899 pix[-2*pix_next] = p1 + clip( ( p2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( p1 << 1 ) ) >> 1, -tc0, tc0 );
4902 if( ABS( q2 - q0 ) < beta ) {
4903 pix[pix_next] = q1 + clip( ( q2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( q1 << 1 ) ) >> 1, -tc0, tc0 );
4907 i_delta = clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );
4908 pix[-pix_next] = clip_uint8( p0 + i_delta ); /* p0' */
4909 pix[0] = clip_uint8( q0 - i_delta ); /* q0' */
4914 /* 4px edge length */
4915 for( d = 0; d < 4; d++ ) {
4916 const int p0 = pix[-1*pix_next];
4917 const int p1 = pix[-2*pix_next];
4918 const int p2 = pix[-3*pix_next];
4919 const int q0 = pix[0];
4920 const int q1 = pix[1*pix_next];
4921 const int q2 = pix[2*pix_next];
4923 if( ABS( p0 - q0 ) < alpha &&
4924 ABS( p1 - p0 ) < beta &&
4925 ABS( q1 - q0 ) < beta ) {
4927 const int p3 = pix[-4*pix_next];
4928 const int q3 = pix[ 3*pix_next];
4930 if(ABS( p0 - q0 ) < (( alpha >> 2 ) + 2 )){
4931 if( ABS( p2 - p0 ) < beta) {
4933 pix[-1*pix_next] = ( p2 + 2*p1 + 2*p0 + 2*q0 + q1 + 4 ) >> 3;
4934 pix[-2*pix_next] = ( p2 + p1 + p0 + q0 + 2 ) >> 2;
4935 pix[-3*pix_next] = ( 2*p3 + 3*p2 + p1 + p0 + q0 + 4 ) >> 3;
4938 pix[-1*pix_next] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
4940 if( ABS( q2 - q0 ) < beta) {
4942 pix[0*pix_next] = ( p1 + 2*p0 + 2*q0 + 2*q1 + q2 + 4 ) >> 3;
4943 pix[1*pix_next] = ( p0 + q0 + q1 + q2 + 2 ) >> 2;
4944 pix[2*pix_next] = ( 2*q3 + 3*q2 + q1 + q0 + p0 + 4 ) >> 3;
4947 pix[0*pix_next] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
4951 pix[-1*pix_next] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
4952 pix[ 0*pix_next] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
4961 static void filter_mb_edgech( H264Context *h, uint8_t *pix, int stride, int bS[4], int qp ) {
4963 const int index_a = clip( qp + h->slice_alpha_c0_offset, 0, 51 );
4964 const int alpha = alpha_table[index_a];
4965 const int beta = beta_table[clip( qp + h->slice_beta_offset, 0, 51 )];
4966 const int pix_next = stride;
4968 for( i = 0; i < 4; i++ )
4976 int tc = tc0_table[index_a][bS[i] - 1] + 1;
4977 /* 2px edge length (see deblocking_filter_edgecv) */
4978 for( d = 0; d < 2; d++ ) {
4979 const int p0 = pix[-1*pix_next];
4980 const int p1 = pix[-2*pix_next];
4981 const int q0 = pix[0];
4982 const int q1 = pix[1*pix_next];
4984 if( ABS( p0 - q0 ) < alpha &&
4985 ABS( p1 - p0 ) < beta &&
4986 ABS( q1 - q0 ) < beta ) {
4988 int i_delta = clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );
4990 pix[-pix_next] = clip_uint8( p0 + i_delta ); /* p0' */
4991 pix[0] = clip_uint8( q0 - i_delta ); /* q0' */
4996 /* 2px edge length (see deblocking_filter_edgecv) */
4997 for( d = 0; d < 2; d++ ) {
4998 const int p0 = pix[-1*pix_next];
4999 const int p1 = pix[-2*pix_next];
5000 const int q0 = pix[0];
5001 const int q1 = pix[1*pix_next];
5003 if( ABS( p0 - q0 ) < alpha &&
5004 ABS( p1 - p0 ) < beta &&
5005 ABS( q1 - q0 ) < beta ) {
5007 pix[-pix_next] = ( 2*p1 + p0 + q1 + 2 ) >> 2; /* p0' */
5008 pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2; /* q0' */
5016 static void filter_mb( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr) {
5017 MpegEncContext * const s = &h->s;
5018 const int mb_xy= mb_x + mb_y*s->mb_stride;
5019 int linesize, uvlinesize;
5022 /* FIXME Implement deblocking filter for field MB */
5023 if( h->sps.mb_aff ) {
5026 linesize = s->linesize;
5027 uvlinesize = s->uvlinesize;
5029 /* dir : 0 -> vertical edge, 1 -> horizontal edge */
5030 for( dir = 0; dir < 2; dir++ )
5035 /* test picture boundary */
5036 if( ( dir == 0 && mb_x == 0 ) || ( dir == 1 && mb_y == 0 ) ) {
5039 /* FIXME test slice boundary */
5040 if( h->deblocking_filter == 2 ) {
5044 for( edge = start; edge < 4; edge++ ) {
5045 /* mbn_xy: neighbour macroblock (how that works for field ?) */
5046 int mbn_xy = edge > 0 ? mb_xy : ( dir == 0 ? mb_xy -1 : mb_xy - s->mb_stride );
5050 if( IS_INTRA( s->current_picture.mb_type[mb_xy] ) ||
5051 IS_INTRA( s->current_picture.mb_type[mbn_xy] ) ) {
5052 bS[0] = bS[1] = bS[2] = bS[3] = ( edge == 0 ? 4 : 3 );
5055 for( i = 0; i < 4; i++ ) {
5056 int x = dir == 0 ? edge : i;
5057 int y = dir == 0 ? i : edge;
5058 int b_idx= 8 + 4 + x + 8*y;
5059 int bn_idx= b_idx - (dir ? 8:1);
5061 if( h->non_zero_count_cache[b_idx] != 0 ||
5062 h->non_zero_count_cache[bn_idx] != 0 ) {
5065 else if( h->slice_type == P_TYPE ) {
5066 if( h->ref_cache[0][b_idx] != h->ref_cache[0][bn_idx] ||
5067 ABS( h->mv_cache[0][b_idx][0] - h->mv_cache[0][bn_idx][0] ) >= 4 ||
5068 ABS( h->mv_cache[0][b_idx][1] - h->mv_cache[0][bn_idx][1] ) >= 4 )
5074 /* FIXME Add support for B frame */
5079 if(bS[0]+bS[1]+bS[2]+bS[3] == 0)
5084 qp = ( s->qscale + s->current_picture.qscale_table[mbn_xy] + 1 ) >> 1;
5086 filter_mb_edgev( h, &img_y[4*edge], linesize, bS, qp );
5087 if( (edge&1) == 0 ) {
5088 int chroma_qp = ( h->chroma_qp +
5089 get_chroma_qp( h, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1;
5090 filter_mb_edgecv( h, &img_cb[2*edge], uvlinesize, bS, chroma_qp );
5091 filter_mb_edgecv( h, &img_cr[2*edge], uvlinesize, bS, chroma_qp );
5094 filter_mb_edgeh( h, &img_y[4*edge*linesize], linesize, bS, qp );
5095 if( (edge&1) == 0 ) {
5096 int chroma_qp = ( h->chroma_qp +
5097 get_chroma_qp( h, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1;
5098 filter_mb_edgech( h, &img_cb[2*edge*uvlinesize], uvlinesize, bS, chroma_qp );
5099 filter_mb_edgech( h, &img_cr[2*edge*uvlinesize], uvlinesize, bS, chroma_qp );
5106 static int decode_slice(H264Context *h){
5107 MpegEncContext * const s = &h->s;
5108 const int part_mask= s->partitioned_frame ? (AC_END|AC_ERROR) : 0x7F;
5112 if( h->pps.cabac ) {
5116 align_get_bits( &s->gb );
5119 ff_init_cabac_states( &h->cabac, ff_h264_lps_range, ff_h264_mps_state, ff_h264_lps_state, 64 );
5120 ff_init_cabac_decoder( &h->cabac,
5121 s->gb.buffer + get_bits_count(&s->gb)/8,
5122 ( s->gb.size_in_bits - get_bits_count(&s->gb) ) );
5123 /* calculate pre-state */
5124 for( i= 0; i < 399; i++ ) {
5126 if( h->slice_type == I_TYPE )
5127 pre = clip( ((cabac_context_init_I[i][0] * s->qscale) >>4 ) + cabac_context_init_I[i][1], 1, 126 );
5129 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 );
5132 h->cabac_state[i] = 2 * ( 63 - pre ) + 0;
5134 h->cabac_state[i] = 2 * ( pre - 64 ) + 1;
5138 int ret = decode_mb_cabac(h);
5139 int eos = get_cabac_terminate( &h->cabac ); /* End of Slice flag */
5143 /* XXX: useless as decode_mb_cabac it doesn't support that ... */
5144 if( ret >= 0 && h->sps.mb_aff ) { //FIXME optimal? or let mb_decode decode 16x32 ?
5147 ret = decode_mb_cabac(h);
5148 eos = get_cabac_terminate( &h->cabac );
5155 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
5156 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);
5160 if( ++s->mb_x >= s->mb_width ) {
5162 ff_draw_horiz_band(s, 16*s->mb_y, 16);
5163 if( ++s->mb_y >= s->mb_height ) {
5164 tprintf("slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
5168 if( eos || s->mb_y >= s->mb_height ) {
5169 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);
5173 /* TODO test over-reading in cabac code */
5174 else if( read too much in h->cabac ) {
5175 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);
5183 int ret = decode_mb_cavlc(h);
5187 if(ret>=0 && h->sps.mb_aff){ //FIXME optimal? or let mb_decode decode 16x32 ?
5189 ret = decode_mb_cavlc(h);
5196 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
5197 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);
5202 if(++s->mb_x >= s->mb_width){
5204 ff_draw_horiz_band(s, 16*s->mb_y, 16);
5205 if(++s->mb_y >= s->mb_height){
5206 tprintf("slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
5208 if(get_bits_count(&s->gb) == s->gb.size_in_bits ) {
5209 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);
5213 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);
5220 if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->mb_skip_run<=0){
5221 if(get_bits_count(&s->gb) == s->gb.size_in_bits ){
5222 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);
5226 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);
5235 for(;s->mb_y < s->mb_height; s->mb_y++){
5236 for(;s->mb_x < s->mb_width; s->mb_x++){
5237 int ret= decode_mb(h);
5242 fprintf(stderr, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
5243 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);
5248 if(++s->mb_x >= s->mb_width){
5250 if(++s->mb_y >= s->mb_height){
5251 if(get_bits_count(s->gb) == s->gb.size_in_bits){
5252 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);
5256 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);
5263 if(get_bits_count(s->?gb) >= s->gb?.size_in_bits){
5264 if(get_bits_count(s->gb) == s->gb.size_in_bits){
5265 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);
5269 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);
5276 ff_draw_horiz_band(s, 16*s->mb_y, 16);
5279 return -1; //not reached
5282 static inline int decode_vui_parameters(H264Context *h, SPS *sps){
5283 MpegEncContext * const s = &h->s;
5284 int aspect_ratio_info_present_flag, aspect_ratio_idc;
5286 aspect_ratio_info_present_flag= get_bits1(&s->gb);
5288 if( aspect_ratio_info_present_flag ) {
5289 aspect_ratio_idc= get_bits(&s->gb, 8);
5290 if( aspect_ratio_idc == EXTENDED_SAR ) {
5291 sps->sar.num= get_bits(&s->gb, 16);
5292 sps->sar.den= get_bits(&s->gb, 16);
5293 }else if(aspect_ratio_idc < 16){
5294 sps->sar= pixel_aspect[aspect_ratio_idc];
5296 av_log(h->s.avctx, AV_LOG_ERROR, "illegal aspect ratio\n");
5303 // s->avctx->aspect_ratio= sar_width*s->width / (float)(s->height*sar_height);
5305 | overscan_info_present_flag |0 |u(1) |
5306 | if( overscan_info_present_flag ) | | |
5307 | overscan_appropriate_flag |0 |u(1) |
5308 | video_signal_type_present_flag |0 |u(1) |
5309 | if( video_signal_type_present_flag ) { | | |
5310 | video_format |0 |u(3) |
5311 | video_full_range_flag |0 |u(1) |
5312 | colour_description_present_flag |0 |u(1) |
5313 | if( colour_description_present_flag ) { | | |
5314 | colour_primaries |0 |u(8) |
5315 | transfer_characteristics |0 |u(8) |
5316 | matrix_coefficients |0 |u(8) |
5319 | chroma_location_info_present_flag |0 |u(1) |
5320 | if ( chroma_location_info_present_flag ) { | | |
5321 | chroma_sample_location_type_top_field |0 |ue(v) |
5322 | chroma_sample_location_type_bottom_field |0 |ue(v) |
5324 | timing_info_present_flag |0 |u(1) |
5325 | if( timing_info_present_flag ) { | | |
5326 | num_units_in_tick |0 |u(32) |
5327 | time_scale |0 |u(32) |
5328 | fixed_frame_rate_flag |0 |u(1) |
5330 | nal_hrd_parameters_present_flag |0 |u(1) |
5331 | if( nal_hrd_parameters_present_flag = = 1) | | |
5332 | hrd_parameters( ) | | |
5333 | vcl_hrd_parameters_present_flag |0 |u(1) |
5334 | if( vcl_hrd_parameters_present_flag = = 1) | | |
5335 | hrd_parameters( ) | | |
5336 | if( ( nal_hrd_parameters_present_flag = = 1 | || | |
5338 |( vcl_hrd_parameters_present_flag = = 1 ) ) | | |
5339 | low_delay_hrd_flag |0 |u(1) |
5340 | bitstream_restriction_flag |0 |u(1) |
5341 | if( bitstream_restriction_flag ) { |0 |u(1) |
5342 | motion_vectors_over_pic_boundaries_flag |0 |u(1) |
5343 | max_bytes_per_pic_denom |0 |ue(v) |
5344 | max_bits_per_mb_denom |0 |ue(v) |
5345 | log2_max_mv_length_horizontal |0 |ue(v) |
5346 | log2_max_mv_length_vertical |0 |ue(v) |
5347 | num_reorder_frames |0 |ue(v) |
5348 | max_dec_frame_buffering |0 |ue(v) |
5355 static inline int decode_seq_parameter_set(H264Context *h){
5356 MpegEncContext * const s = &h->s;
5357 int profile_idc, level_idc;
5361 profile_idc= get_bits(&s->gb, 8);
5362 get_bits1(&s->gb); //constraint_set0_flag
5363 get_bits1(&s->gb); //constraint_set1_flag
5364 get_bits1(&s->gb); //constraint_set2_flag
5365 get_bits(&s->gb, 5); // reserved
5366 level_idc= get_bits(&s->gb, 8);
5367 sps_id= get_ue_golomb(&s->gb);
5369 sps= &h->sps_buffer[ sps_id ];
5370 sps->profile_idc= profile_idc;
5371 sps->level_idc= level_idc;
5373 sps->log2_max_frame_num= get_ue_golomb(&s->gb) + 4;
5374 sps->poc_type= get_ue_golomb(&s->gb);
5376 if(sps->poc_type == 0){ //FIXME #define
5377 sps->log2_max_poc_lsb= get_ue_golomb(&s->gb) + 4;
5378 } else if(sps->poc_type == 1){//FIXME #define
5379 sps->delta_pic_order_always_zero_flag= get_bits1(&s->gb);
5380 sps->offset_for_non_ref_pic= get_se_golomb(&s->gb);
5381 sps->offset_for_top_to_bottom_field= get_se_golomb(&s->gb);
5382 sps->poc_cycle_length= get_ue_golomb(&s->gb);
5384 for(i=0; i<sps->poc_cycle_length; i++)
5385 sps->offset_for_ref_frame[i]= get_se_golomb(&s->gb);
5387 if(sps->poc_type > 2){
5388 av_log(h->s.avctx, AV_LOG_ERROR, "illegal POC type %d\n", sps->poc_type);
5392 sps->ref_frame_count= get_ue_golomb(&s->gb);
5393 sps->gaps_in_frame_num_allowed_flag= get_bits1(&s->gb);
5394 sps->mb_width= get_ue_golomb(&s->gb) + 1;
5395 sps->mb_height= get_ue_golomb(&s->gb) + 1;
5396 sps->frame_mbs_only_flag= get_bits1(&s->gb);
5397 if(!sps->frame_mbs_only_flag)
5398 sps->mb_aff= get_bits1(&s->gb);
5402 sps->direct_8x8_inference_flag= get_bits1(&s->gb);
5404 sps->crop= get_bits1(&s->gb);
5406 sps->crop_left = get_ue_golomb(&s->gb);
5407 sps->crop_right = get_ue_golomb(&s->gb);
5408 sps->crop_top = get_ue_golomb(&s->gb);
5409 sps->crop_bottom= get_ue_golomb(&s->gb);
5410 if(sps->crop_left || sps->crop_top){
5411 av_log(h->s.avctx, AV_LOG_ERROR, "insane cropping not completly supported, this could look slightly wrong ...\n");
5417 sps->crop_bottom= 0;
5420 sps->vui_parameters_present_flag= get_bits1(&s->gb);
5421 if( sps->vui_parameters_present_flag )
5422 decode_vui_parameters(h, sps);
5424 if(s->avctx->debug&FF_DEBUG_PICT_INFO){
5425 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",
5426 sps_id, sps->profile_idc, sps->level_idc,
5428 sps->ref_frame_count,
5429 sps->mb_width, sps->mb_height,
5430 sps->frame_mbs_only_flag ? "FRM" : (sps->mb_aff ? "MB-AFF" : "PIC-AFF"),
5431 sps->direct_8x8_inference_flag ? "8B8" : "",
5432 sps->crop_left, sps->crop_right,
5433 sps->crop_top, sps->crop_bottom,
5434 sps->vui_parameters_present_flag ? "VUI" : ""
5440 static inline int decode_picture_parameter_set(H264Context *h){
5441 MpegEncContext * const s = &h->s;
5442 int pps_id= get_ue_golomb(&s->gb);
5443 PPS *pps= &h->pps_buffer[pps_id];
5445 pps->sps_id= get_ue_golomb(&s->gb);
5446 pps->cabac= get_bits1(&s->gb);
5447 pps->pic_order_present= get_bits1(&s->gb);
5448 pps->slice_group_count= get_ue_golomb(&s->gb) + 1;
5449 if(pps->slice_group_count > 1 ){
5450 pps->mb_slice_group_map_type= get_ue_golomb(&s->gb);
5451 av_log(h->s.avctx, AV_LOG_ERROR, "FMO not supported\n");
5452 switch(pps->mb_slice_group_map_type){
5455 | for( i = 0; i <= num_slice_groups_minus1; i++ ) | | |
5456 | run_length[ i ] |1 |ue(v) |
5461 | for( i = 0; i < num_slice_groups_minus1; i++ ) | | |
5463 | top_left_mb[ i ] |1 |ue(v) |
5464 | bottom_right_mb[ i ] |1 |ue(v) |
5472 | slice_group_change_direction_flag |1 |u(1) |
5473 | slice_group_change_rate_minus1 |1 |ue(v) |
5478 | slice_group_id_cnt_minus1 |1 |ue(v) |
5479 | for( i = 0; i <= slice_group_id_cnt_minus1; i++ | | |
5481 | slice_group_id[ i ] |1 |u(v) |
5486 pps->ref_count[0]= get_ue_golomb(&s->gb) + 1;
5487 pps->ref_count[1]= get_ue_golomb(&s->gb) + 1;
5488 if(pps->ref_count[0] > 32 || pps->ref_count[1] > 32){
5489 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow (pps)\n");
5493 pps->weighted_pred= get_bits1(&s->gb);
5494 pps->weighted_bipred_idc= get_bits(&s->gb, 2);
5495 pps->init_qp= get_se_golomb(&s->gb) + 26;
5496 pps->init_qs= get_se_golomb(&s->gb) + 26;
5497 pps->chroma_qp_index_offset= get_se_golomb(&s->gb);
5498 pps->deblocking_filter_parameters_present= get_bits1(&s->gb);
5499 pps->constrained_intra_pred= get_bits1(&s->gb);
5500 pps->redundant_pic_cnt_present = get_bits1(&s->gb);
5502 if(s->avctx->debug&FF_DEBUG_PICT_INFO){
5503 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",
5504 pps_id, pps->sps_id,
5505 pps->cabac ? "CABAC" : "CAVLC",
5506 pps->slice_group_count,
5507 pps->ref_count[0], pps->ref_count[1],
5508 pps->weighted_pred ? "weighted" : "",
5509 pps->init_qp, pps->init_qs, pps->chroma_qp_index_offset,
5510 pps->deblocking_filter_parameters_present ? "LPAR" : "",
5511 pps->constrained_intra_pred ? "CONSTR" : "",
5512 pps->redundant_pic_cnt_present ? "REDU" : ""
5520 * finds the end of the current frame in the bitstream.
5521 * @return the position of the first byte of the next frame, or -1
5523 static int find_frame_end(ParseContext *pc, const uint8_t *buf, int buf_size){
5526 //printf("first %02X%02X%02X%02X\n", buf[0], buf[1],buf[2],buf[3]);
5527 // mb_addr= pc->mb_addr - 1;
5529 //FIXME this will fail with slices
5530 for(i=0; i<buf_size; i++){
5531 state= (state<<8) | buf[i];
5532 if((state&0xFFFFFF1F) == 0x101 || (state&0xFFFFFF1F) == 0x102 || (state&0xFFFFFF1F) == 0x105){
5533 if(pc->frame_start_found){
5535 pc->frame_start_found= 0;
5538 pc->frame_start_found= 1;
5543 return END_NOT_FOUND;
5546 static int h264_parse(AVCodecParserContext *s,
5547 AVCodecContext *avctx,
5548 uint8_t **poutbuf, int *poutbuf_size,
5549 const uint8_t *buf, int buf_size)
5551 ParseContext *pc = s->priv_data;
5554 next= find_frame_end(pc, buf, buf_size);
5556 if (ff_combine_frame(pc, next, (uint8_t **)&buf, &buf_size) < 0) {
5562 *poutbuf = (uint8_t *)buf;
5563 *poutbuf_size = buf_size;
5567 static int decode_nal_units(H264Context *h, uint8_t *buf, int buf_size){
5568 MpegEncContext * const s = &h->s;
5569 AVCodecContext * const avctx= s->avctx;
5573 for(i=0; i<32; i++){
5574 printf("%X ", buf[i]);
5583 // start code prefix search
5584 for(; buf_index + 3 < buf_size; buf_index++){
5585 // this should allways succeed in the first iteration
5586 if(buf[buf_index] == 0 && buf[buf_index+1] == 0 && buf[buf_index+2] == 1)
5590 if(buf_index+3 >= buf_size) break;
5594 ptr= decode_nal(h, buf + buf_index, &dst_length, &consumed, buf_size - buf_index);
5595 if(ptr[dst_length - 1] == 0) dst_length--;
5596 bit_length= 8*dst_length - decode_rbsp_trailing(ptr + dst_length - 1);
5598 if(s->avctx->debug&FF_DEBUG_STARTCODE){
5599 av_log(h->s.avctx, AV_LOG_DEBUG, "NAL %d at %d length %d\n", h->nal_unit_type, buf_index, dst_length);
5602 buf_index += consumed;
5604 if( s->hurry_up == 1 && h->nal_ref_idc == 0 )
5607 switch(h->nal_unit_type){
5609 idr(h); //FIXME ensure we dont loose some frames if there is reordering
5611 init_get_bits(&s->gb, ptr, bit_length);
5613 h->inter_gb_ptr= &s->gb;
5614 s->data_partitioning = 0;
5616 if(decode_slice_header(h) < 0) return -1;
5617 if(h->redundant_pic_count==0 && s->hurry_up < 5 )
5621 init_get_bits(&s->gb, ptr, bit_length);
5623 h->inter_gb_ptr= NULL;
5624 s->data_partitioning = 1;
5626 if(decode_slice_header(h) < 0) return -1;
5629 init_get_bits(&h->intra_gb, ptr, bit_length);
5630 h->intra_gb_ptr= &h->intra_gb;
5633 init_get_bits(&h->inter_gb, ptr, bit_length);
5634 h->inter_gb_ptr= &h->inter_gb;
5636 if(h->redundant_pic_count==0 && h->intra_gb_ptr && s->data_partitioning && s->hurry_up < 5 )
5642 init_get_bits(&s->gb, ptr, bit_length);
5643 decode_seq_parameter_set(h);
5645 if(s->flags& CODEC_FLAG_LOW_DELAY)
5648 avctx->has_b_frames= !s->low_delay;
5651 init_get_bits(&s->gb, ptr, bit_length);
5653 decode_picture_parameter_set(h);
5656 case NAL_PICTURE_DELIMITER:
5658 case NAL_FILTER_DATA:
5662 //FIXME move after where irt is set
5663 s->current_picture.pict_type= s->pict_type;
5664 s->current_picture.key_frame= s->pict_type == I_TYPE;
5667 if(!s->current_picture_ptr) return buf_index; //no frame
5669 h->prev_frame_num_offset= h->frame_num_offset;
5670 h->prev_frame_num= h->frame_num;
5671 if(s->current_picture_ptr->reference){
5672 h->prev_poc_msb= h->poc_msb;
5673 h->prev_poc_lsb= h->poc_lsb;
5675 if(s->current_picture_ptr->reference)
5676 execute_ref_pic_marking(h, h->mmco, h->mmco_index);
5678 assert(h->mmco_index==0);
5688 * retunrs the number of bytes consumed for building the current frame
5690 static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size){
5691 if(s->flags&CODEC_FLAG_TRUNCATED){
5692 pos -= s->parse_context.last_index;
5693 if(pos<0) pos=0; // FIXME remove (uneeded?)
5697 if(pos==0) pos=1; //avoid infinite loops (i doubt thats needed but ...)
5698 if(pos+10>buf_size) pos=buf_size; // oops ;)
5704 static int decode_frame(AVCodecContext *avctx,
5705 void *data, int *data_size,
5706 uint8_t *buf, int buf_size)
5708 H264Context *h = avctx->priv_data;
5709 MpegEncContext *s = &h->s;
5710 AVFrame *pict = data;
5713 s->flags= avctx->flags;
5714 s->flags2= avctx->flags2;
5718 /* no supplementary picture */
5719 if (buf_size == 0) {
5723 if(s->flags&CODEC_FLAG_TRUNCATED){
5724 int next= find_frame_end(&s->parse_context, buf, buf_size);
5726 if( ff_combine_frame(&s->parse_context, next, &buf, &buf_size) < 0 )
5728 //printf("next:%d buf_size:%d last_index:%d\n", next, buf_size, s->parse_context.last_index);
5731 if(s->avctx->extradata_size && s->picture_number==0){
5732 if(0 < decode_nal_units(h, s->avctx->extradata, s->avctx->extradata_size) )
5736 buf_index=decode_nal_units(h, buf, buf_size);
5740 //FIXME do something with unavailable reference frames
5742 // if(ret==FRAME_SKIPED) return get_consumed_bytes(s, buf_index, buf_size);
5744 if(s->pict_type==B_TYPE || s->low_delay){
5745 *pict= *(AVFrame*)&s->current_picture;
5747 *pict= *(AVFrame*)&s->last_picture;
5750 if(!s->current_picture_ptr){
5751 av_log(h->s.avctx, AV_LOG_DEBUG, "error, NO frame\n");
5755 *pict= *(AVFrame*)&s->current_picture; //FIXME
5756 ff_print_debug_info(s, pict);
5757 assert(pict->data[0]);
5758 //printf("out %d\n", (int)pict->data[0]);
5761 /* Return the Picture timestamp as the frame number */
5762 /* we substract 1 because it is added on utils.c */
5763 avctx->frame_number = s->picture_number - 1;
5766 /* dont output the last pic after seeking */
5767 if(s->last_picture_ptr || s->low_delay)
5768 //Note this isnt a issue as a IDR pic should flush teh buffers
5770 *data_size = sizeof(AVFrame);
5771 return get_consumed_bytes(s, buf_index, buf_size);
5774 static inline void fill_mb_avail(H264Context *h){
5775 MpegEncContext * const s = &h->s;
5776 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
5779 h->mb_avail[0]= s->mb_x && h->slice_table[mb_xy - s->mb_stride - 1] == h->slice_num;
5780 h->mb_avail[1]= h->slice_table[mb_xy - s->mb_stride ] == h->slice_num;
5781 h->mb_avail[2]= s->mb_x+1 < s->mb_width && h->slice_table[mb_xy - s->mb_stride + 1] == h->slice_num;
5787 h->mb_avail[3]= s->mb_x && h->slice_table[mb_xy - 1] == h->slice_num;
5788 h->mb_avail[4]= 1; //FIXME move out
5789 h->mb_avail[5]= 0; //FIXME move out
5795 #define SIZE (COUNT*40)
5801 // int int_temp[10000];
5803 AVCodecContext avctx;
5805 dsputil_init(&dsp, &avctx);
5807 init_put_bits(&pb, temp, SIZE);
5808 printf("testing unsigned exp golomb\n");
5809 for(i=0; i<COUNT; i++){
5811 set_ue_golomb(&pb, i);
5812 STOP_TIMER("set_ue_golomb");
5814 flush_put_bits(&pb);
5816 init_get_bits(&gb, temp, 8*SIZE);
5817 for(i=0; i<COUNT; i++){
5820 s= show_bits(&gb, 24);
5823 j= get_ue_golomb(&gb);
5825 printf("missmatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
5828 STOP_TIMER("get_ue_golomb");
5832 init_put_bits(&pb, temp, SIZE);
5833 printf("testing signed exp golomb\n");
5834 for(i=0; i<COUNT; i++){
5836 set_se_golomb(&pb, i - COUNT/2);
5837 STOP_TIMER("set_se_golomb");
5839 flush_put_bits(&pb);
5841 init_get_bits(&gb, temp, 8*SIZE);
5842 for(i=0; i<COUNT; i++){
5845 s= show_bits(&gb, 24);
5848 j= get_se_golomb(&gb);
5849 if(j != i - COUNT/2){
5850 printf("missmatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
5853 STOP_TIMER("get_se_golomb");
5856 printf("testing 4x4 (I)DCT\n");
5859 uint8_t src[16], ref[16];
5860 uint64_t error= 0, max_error=0;
5862 for(i=0; i<COUNT; i++){
5864 // printf("%d %d %d\n", r1, r2, (r2-r1)*16);
5865 for(j=0; j<16; j++){
5866 ref[j]= random()%255;
5867 src[j]= random()%255;
5870 h264_diff_dct_c(block, src, ref, 4);
5873 for(j=0; j<16; j++){
5874 // printf("%d ", block[j]);
5875 block[j]= block[j]*4;
5876 if(j&1) block[j]= (block[j]*4 + 2)/5;
5877 if(j&4) block[j]= (block[j]*4 + 2)/5;
5881 h264_add_idct_c(ref, block, 4);
5882 /* for(j=0; j<16; j++){
5883 printf("%d ", ref[j]);
5887 for(j=0; j<16; j++){
5888 int diff= ABS(src[j] - ref[j]);
5891 max_error= FFMAX(max_error, diff);
5894 printf("error=%f max_error=%d\n", ((float)error)/COUNT/16, (int)max_error );
5896 printf("testing quantizer\n");
5897 for(qp=0; qp<52; qp++){
5899 src1_block[i]= src2_block[i]= random()%255;
5903 printf("Testing NAL layer\n");
5905 uint8_t bitstream[COUNT];
5906 uint8_t nal[COUNT*2];
5908 memset(&h, 0, sizeof(H264Context));
5910 for(i=0; i<COUNT; i++){
5918 for(j=0; j<COUNT; j++){
5919 bitstream[j]= (random() % 255) + 1;
5922 for(j=0; j<zeros; j++){
5923 int pos= random() % COUNT;
5924 while(bitstream[pos] == 0){
5933 nal_length= encode_nal(&h, nal, bitstream, COUNT, COUNT*2);
5935 printf("encoding failed\n");
5939 out= decode_nal(&h, nal, &out_length, &consumed, nal_length);
5943 if(out_length != COUNT){
5944 printf("incorrect length %d %d\n", out_length, COUNT);
5948 if(consumed != nal_length){
5949 printf("incorrect consumed length %d %d\n", nal_length, consumed);
5953 if(memcmp(bitstream, out, COUNT)){
5954 printf("missmatch\n");
5959 printf("Testing RBSP\n");
5967 static int decode_end(AVCodecContext *avctx)
5969 H264Context *h = avctx->priv_data;
5970 MpegEncContext *s = &h->s;
5972 free_tables(h); //FIXME cleanup init stuff perhaps
5975 // memset(h, 0, sizeof(H264Context));
5981 AVCodec h264_decoder = {
5985 sizeof(H264Context),
5990 /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 | CODEC_CAP_TRUNCATED,
5993 AVCodecParser h264_parser = {
5995 sizeof(ParseContext),