2 * H.26L/H.264/AVC/JVT/14496-10/... encoder/decoder
3 * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
5 * This library is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU Lesser General Public
7 * License as published by the Free Software Foundation; either
8 * version 2 of the License, or (at your option) any later version.
10 * This library is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * Lesser General Public License for more details.
15 * You should have received a copy of the GNU Lesser General Public
16 * License along with this library; if not, write to the Free Software
17 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
23 * H.264 / AVC / MPEG4 part10 codec.
24 * @author Michael Niedermayer <michaelni@gmx.at>
30 #include "mpegvideo.h"
39 #define interlaced_dct interlaced_dct_is_a_bad_name
40 #define mb_intra mb_intra_isnt_initalized_see_mb_type
42 #define LUMA_DC_BLOCK_INDEX 25
43 #define CHROMA_DC_BLOCK_INDEX 26
45 #define CHROMA_DC_COEFF_TOKEN_VLC_BITS 8
46 #define COEFF_TOKEN_VLC_BITS 8
47 #define TOTAL_ZEROS_VLC_BITS 9
48 #define CHROMA_DC_TOTAL_ZEROS_VLC_BITS 3
49 #define RUN_VLC_BITS 3
50 #define RUN7_VLC_BITS 6
52 #define MAX_SPS_COUNT 32
53 #define MAX_PPS_COUNT 256
55 #define MAX_MMCO_COUNT 66
58 * Sequence parameter set
64 int log2_max_frame_num; ///< log2_max_frame_num_minus4 + 4
65 int poc_type; ///< pic_order_cnt_type
66 int log2_max_poc_lsb; ///< log2_max_pic_order_cnt_lsb_minus4
67 int delta_pic_order_always_zero_flag;
68 int offset_for_non_ref_pic;
69 int offset_for_top_to_bottom_field;
70 int poc_cycle_length; ///< num_ref_frames_in_pic_order_cnt_cycle
71 int ref_frame_count; ///< num_ref_frames
72 int gaps_in_frame_num_allowed_flag;
73 int mb_width; ///< frame_width_in_mbs_minus1 + 1
74 int mb_height; ///< frame_height_in_mbs_minus1 + 1
75 int frame_mbs_only_flag;
76 int mb_aff; ///<mb_adaptive_frame_field_flag
77 int direct_8x8_inference_flag;
78 int crop; ///< frame_cropping_flag
79 int crop_left; ///< frame_cropping_rect_left_offset
80 int crop_right; ///< frame_cropping_rect_right_offset
81 int crop_top; ///< frame_cropping_rect_top_offset
82 int crop_bottom; ///< frame_cropping_rect_bottom_offset
83 int vui_parameters_present_flag;
85 int timing_info_present_flag;
86 uint32_t num_units_in_tick;
88 int fixed_frame_rate_flag;
89 short offset_for_ref_frame[256]; //FIXME dyn aloc?
93 * Picture parameter set
97 int cabac; ///< entropy_coding_mode_flag
98 int pic_order_present; ///< pic_order_present_flag
99 int slice_group_count; ///< num_slice_groups_minus1 + 1
100 int mb_slice_group_map_type;
101 int ref_count[2]; ///< num_ref_idx_l0/1_active_minus1 + 1
102 int weighted_pred; ///< weighted_pred_flag
103 int weighted_bipred_idc;
104 int init_qp; ///< pic_init_qp_minus26 + 26
105 int init_qs; ///< pic_init_qs_minus26 + 26
106 int chroma_qp_index_offset;
107 int deblocking_filter_parameters_present; ///< deblocking_filter_parameters_present_flag
108 int constrained_intra_pred; ///< constrained_intra_pred_flag
109 int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag
113 * Memory management control operation opcode.
115 typedef enum MMCOOpcode{
126 * Memory management control operation.
137 typedef struct H264Context{
145 #define NAL_IDR_SLICE 5
149 #define NAL_PICTURE_DELIMITER 9
150 #define NAL_FILTER_DATA 10
151 uint8_t *rbsp_buffer;
152 int rbsp_buffer_size;
155 * Used to parse AVC variant of h264
157 int is_avc; ///< this flag is != 0 if codec is avc1
158 int got_avcC; ///< flag used to parse avcC data only once
159 int nal_length_size; ///< Number of bytes used for nal length (1, 2 or 4)
163 int prev_mb_skiped; //FIXME remove (IMHO not used)
166 int chroma_pred_mode;
167 int intra16x16_pred_mode;
169 int8_t intra4x4_pred_mode_cache[5*8];
170 int8_t (*intra4x4_pred_mode)[8];
171 void (*pred4x4 [9+3])(uint8_t *src, uint8_t *topright, int stride);//FIXME move to dsp?
172 void (*pred8x8 [4+3])(uint8_t *src, int stride);
173 void (*pred16x16[4+3])(uint8_t *src, int stride);
174 unsigned int topleft_samples_available;
175 unsigned int top_samples_available;
176 unsigned int topright_samples_available;
177 unsigned int left_samples_available;
178 uint8_t (*top_border)[16+2*8];
179 uint8_t left_border[17+2*9];
182 * non zero coeff count cache.
183 * is 64 if not available.
185 uint8_t non_zero_count_cache[6*8];
186 uint8_t (*non_zero_count)[16];
189 * Motion vector cache.
191 int16_t mv_cache[2][5*8][2];
192 int8_t ref_cache[2][5*8];
193 #define LIST_NOT_USED -1 //FIXME rename?
194 #define PART_NOT_AVAILABLE -2
197 * is 1 if the specific list MV&references are set to 0,0,-2.
199 int mv_cache_clean[2];
201 int block_offset[16+8];
202 int chroma_subblock_offset[16]; //FIXME remove
204 uint16_t *mb2b_xy; //FIXME are these 4 a good idea?
212 int unknown_svq3_flag;
213 int next_slice_index;
215 SPS sps_buffer[MAX_SPS_COUNT];
216 SPS sps; ///< current sps
218 PPS pps_buffer[MAX_PPS_COUNT];
222 PPS pps; //FIXME move tp Picture perhaps? (->no) do we need that?
225 uint8_t *slice_table_base;
226 uint8_t *slice_table; ///< slice_table_base + mb_stride + 1
228 int slice_type_fixed;
230 //interlacing specific flags
231 int mb_field_decoding_flag;
238 int delta_poc_bottom;
241 int prev_poc_msb; ///< poc_msb of the last reference pic for POC type 0
242 int prev_poc_lsb; ///< poc_lsb of the last reference pic for POC type 0
243 int frame_num_offset; ///< for POC type 2
244 int prev_frame_num_offset; ///< for POC type 2
245 int prev_frame_num; ///< frame_num of the last pic for POC type 1/2
248 * frame_num for frames or 2*frame_num for field pics.
253 * max_frame_num or 2*max_frame_num for field pics.
257 //Weighted pred stuff
258 int luma_log2_weight_denom;
259 int chroma_log2_weight_denom;
260 int luma_weight[2][16];
261 int luma_offset[2][16];
262 int chroma_weight[2][16][2];
263 int chroma_offset[2][16][2];
266 int deblocking_filter; ///< disable_deblocking_filter_idc with 1<->0
267 int slice_alpha_c0_offset;
268 int slice_beta_offset;
270 int redundant_pic_count;
272 int direct_spatial_mv_pred;
275 * num_ref_idx_l0/1_active_minus1 + 1
277 int ref_count[2];// FIXME split for AFF
278 Picture *short_ref[16];
279 Picture *long_ref[16];
280 Picture default_ref_list[2][32];
281 Picture ref_list[2][32]; //FIXME size?
282 Picture field_ref_list[2][32]; //FIXME size?
285 * memory management control operations buffer.
287 MMCO mmco[MAX_MMCO_COUNT];
290 int long_ref_count; ///< number of actual long term references
291 int short_ref_count; ///< number of actual short term references
294 GetBitContext intra_gb;
295 GetBitContext inter_gb;
296 GetBitContext *intra_gb_ptr;
297 GetBitContext *inter_gb_ptr;
299 DCTELEM mb[16*24] __align8;
305 uint8_t cabac_state[399];
308 /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0,1,2), 0x0? luma_cbp */
310 /* chroma_pred_mode for i4x4 or i16x16, else 0 */
311 uint8_t *chroma_pred_mode_table;
312 int last_qscale_diff;
313 int16_t (*mvd_table[2])[2];
314 int16_t mvd_cache[2][5*8][2];
318 static VLC coeff_token_vlc[4];
319 static VLC chroma_dc_coeff_token_vlc;
321 static VLC total_zeros_vlc[15];
322 static VLC chroma_dc_total_zeros_vlc[3];
324 static VLC run_vlc[6];
327 static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
328 static void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
329 static void filter_mb( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr);
331 static inline uint32_t pack16to32(int a, int b){
332 #ifdef WORDS_BIGENDIAN
333 return (b&0xFFFF) + (a<<16);
335 return (a&0xFFFF) + (b<<16);
341 * @param h height of the recatangle, should be a constant
342 * @param w width of the recatangle, should be a constant
343 * @param size the size of val (1 or 4), should be a constant
345 static inline void fill_rectangle(void *vp, int w, int h, int stride, uint32_t val, int size){ //FIXME ensure this IS inlined
346 uint8_t *p= (uint8_t*)vp;
347 assert(size==1 || size==4);
352 //FIXME check what gcc generates for 64 bit on x86 and possible write a 32 bit ver of it
355 *(uint16_t*)(p + stride)= size==4 ? val : val*0x0101;
356 }else if(w==2 && h==4){
357 *(uint16_t*)(p + 0*stride)=
358 *(uint16_t*)(p + 1*stride)=
359 *(uint16_t*)(p + 2*stride)=
360 *(uint16_t*)(p + 3*stride)= size==4 ? val : val*0x0101;
361 }else if(w==4 && h==1){
362 *(uint32_t*)(p + 0*stride)= size==4 ? val : val*0x01010101;
363 }else if(w==4 && h==2){
364 *(uint32_t*)(p + 0*stride)=
365 *(uint32_t*)(p + 1*stride)= size==4 ? val : val*0x01010101;
366 }else if(w==4 && h==4){
367 *(uint32_t*)(p + 0*stride)=
368 *(uint32_t*)(p + 1*stride)=
369 *(uint32_t*)(p + 2*stride)=
370 *(uint32_t*)(p + 3*stride)= size==4 ? val : val*0x01010101;
371 }else if(w==8 && h==1){
373 *(uint32_t*)(p + 4)= size==4 ? val : val*0x01010101;
374 }else if(w==8 && h==2){
375 *(uint32_t*)(p + 0 + 0*stride)=
376 *(uint32_t*)(p + 4 + 0*stride)=
377 *(uint32_t*)(p + 0 + 1*stride)=
378 *(uint32_t*)(p + 4 + 1*stride)= size==4 ? val : val*0x01010101;
379 }else if(w==8 && h==4){
380 *(uint64_t*)(p + 0*stride)=
381 *(uint64_t*)(p + 1*stride)=
382 *(uint64_t*)(p + 2*stride)=
383 *(uint64_t*)(p + 3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
384 }else if(w==16 && h==2){
385 *(uint64_t*)(p + 0+0*stride)=
386 *(uint64_t*)(p + 8+0*stride)=
387 *(uint64_t*)(p + 0+1*stride)=
388 *(uint64_t*)(p + 8+1*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
389 }else if(w==16 && h==4){
390 *(uint64_t*)(p + 0+0*stride)=
391 *(uint64_t*)(p + 8+0*stride)=
392 *(uint64_t*)(p + 0+1*stride)=
393 *(uint64_t*)(p + 8+1*stride)=
394 *(uint64_t*)(p + 0+2*stride)=
395 *(uint64_t*)(p + 8+2*stride)=
396 *(uint64_t*)(p + 0+3*stride)=
397 *(uint64_t*)(p + 8+3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
402 static inline void fill_caches(H264Context *h, int mb_type){
403 MpegEncContext * const s = &h->s;
404 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
405 int topleft_xy, top_xy, topright_xy, left_xy[2];
406 int topleft_type, top_type, topright_type, left_type[2];
410 //wow what a mess, why didnt they simplify the interlacing&intra stuff, i cant imagine that these complex rules are worth it
414 topleft_xy = 0; /* avoid warning */
415 top_xy = 0; /* avoid warning */
416 topright_xy = 0; /* avoid warning */
418 topleft_xy = mb_xy-1 - s->mb_stride;
419 top_xy = mb_xy - s->mb_stride;
420 topright_xy= mb_xy+1 - s->mb_stride;
421 left_xy[0] = mb_xy-1;
422 left_xy[1] = mb_xy-1;
429 topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
430 top_type = h->slice_table[top_xy ] == h->slice_num ? s->current_picture.mb_type[top_xy] : 0;
431 topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
432 left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
433 left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
435 if(IS_INTRA(mb_type)){
436 h->topleft_samples_available=
437 h->top_samples_available=
438 h->left_samples_available= 0xFFFF;
439 h->topright_samples_available= 0xEEEA;
441 if(!IS_INTRA(top_type) && (top_type==0 || h->pps.constrained_intra_pred)){
442 h->topleft_samples_available= 0xB3FF;
443 h->top_samples_available= 0x33FF;
444 h->topright_samples_available= 0x26EA;
447 if(!IS_INTRA(left_type[i]) && (left_type[i]==0 || h->pps.constrained_intra_pred)){
448 h->topleft_samples_available&= 0xDF5F;
449 h->left_samples_available&= 0x5F5F;
453 if(!IS_INTRA(topleft_type) && (topleft_type==0 || h->pps.constrained_intra_pred))
454 h->topleft_samples_available&= 0x7FFF;
456 if(!IS_INTRA(topright_type) && (topright_type==0 || h->pps.constrained_intra_pred))
457 h->topright_samples_available&= 0xFBFF;
459 if(IS_INTRA4x4(mb_type)){
460 if(IS_INTRA4x4(top_type)){
461 h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
462 h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
463 h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
464 h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
467 if(IS_INTRA16x16(top_type) || (IS_INTER(top_type) && !h->pps.constrained_intra_pred))
472 h->intra4x4_pred_mode_cache[4+8*0]=
473 h->intra4x4_pred_mode_cache[5+8*0]=
474 h->intra4x4_pred_mode_cache[6+8*0]=
475 h->intra4x4_pred_mode_cache[7+8*0]= pred;
478 if(IS_INTRA4x4(left_type[i])){
479 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
480 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
483 if(IS_INTRA16x16(left_type[i]) || (IS_INTER(left_type[i]) && !h->pps.constrained_intra_pred))
488 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
489 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
504 //FIXME constraint_intra_pred & partitioning & nnz (lets hope this is just a typo in the spec)
506 h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][0];
507 h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][1];
508 h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][2];
509 h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3];
511 h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][7];
512 h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8];
514 h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][10];
515 h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11];
517 h->non_zero_count_cache[4+8*0]=
518 h->non_zero_count_cache[5+8*0]=
519 h->non_zero_count_cache[6+8*0]=
520 h->non_zero_count_cache[7+8*0]=
522 h->non_zero_count_cache[1+8*0]=
523 h->non_zero_count_cache[2+8*0]=
525 h->non_zero_count_cache[1+8*3]=
526 h->non_zero_count_cache[2+8*3]= 64;
530 h->non_zero_count_cache[3+8*1]= h->non_zero_count[left_xy[0]][6];
531 h->non_zero_count_cache[3+8*2]= h->non_zero_count[left_xy[0]][5];
532 h->non_zero_count_cache[0+8*1]= h->non_zero_count[left_xy[0]][9]; //FIXME left_block
533 h->non_zero_count_cache[0+8*4]= h->non_zero_count[left_xy[0]][12];
535 h->non_zero_count_cache[3+8*1]=
536 h->non_zero_count_cache[3+8*2]=
537 h->non_zero_count_cache[0+8*1]=
538 h->non_zero_count_cache[0+8*4]= 64;
542 h->non_zero_count_cache[3+8*3]= h->non_zero_count[left_xy[1]][4];
543 h->non_zero_count_cache[3+8*4]= h->non_zero_count[left_xy[1]][3];
544 h->non_zero_count_cache[0+8*2]= h->non_zero_count[left_xy[1]][8];
545 h->non_zero_count_cache[0+8*5]= h->non_zero_count[left_xy[1]][11];
547 h->non_zero_count_cache[3+8*3]=
548 h->non_zero_count_cache[3+8*4]=
549 h->non_zero_count_cache[0+8*2]=
550 h->non_zero_count_cache[0+8*5]= 64;
554 if(IS_INTER(mb_type)){
556 for(list=0; list<2; list++){
557 if((!IS_8X8(mb_type)) && !USES_LIST(mb_type, list)){
558 /*if(!h->mv_cache_clean[list]){
559 memset(h->mv_cache [list], 0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
560 memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
561 h->mv_cache_clean[list]= 1;
563 continue; //FIXME direct mode ...
565 h->mv_cache_clean[list]= 0;
567 if(IS_INTER(topleft_type)){
568 const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
569 const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + h->b8_stride;
570 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
571 h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
573 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
574 h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
577 if(IS_INTER(top_type)){
578 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
579 const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
580 *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0];
581 *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];
582 *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];
583 *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];
584 h->ref_cache[list][scan8[0] + 0 - 1*8]=
585 h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
586 h->ref_cache[list][scan8[0] + 2 - 1*8]=
587 h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
589 *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]=
590 *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]=
591 *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]=
592 *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;
593 *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
596 if(IS_INTER(topright_type)){
597 const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
598 const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
599 *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
600 h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
602 *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
603 h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
606 //FIXME unify cleanup or sth
607 if(IS_INTER(left_type[0])){
608 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
609 const int b8_xy= h->mb2b8_xy[left_xy[0]] + 1;
610 *(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]];
611 *(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]];
612 h->ref_cache[list][scan8[0] - 1 + 0*8]=
613 h->ref_cache[list][scan8[0] - 1 + 1*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0]>>1)];
615 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 0*8]=
616 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 1*8]= 0;
617 h->ref_cache[list][scan8[0] - 1 + 0*8]=
618 h->ref_cache[list][scan8[0] - 1 + 1*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
621 if(IS_INTER(left_type[1])){
622 const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
623 const int b8_xy= h->mb2b8_xy[left_xy[1]] + 1;
624 *(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]];
625 *(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]];
626 h->ref_cache[list][scan8[0] - 1 + 2*8]=
627 h->ref_cache[list][scan8[0] - 1 + 3*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[2]>>1)];
629 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 2*8]=
630 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 3*8]= 0;
631 h->ref_cache[list][scan8[0] - 1 + 2*8]=
632 h->ref_cache[list][scan8[0] - 1 + 3*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
635 h->ref_cache[list][scan8[5 ]+1] =
636 h->ref_cache[list][scan8[7 ]+1] =
637 h->ref_cache[list][scan8[13]+1] = //FIXME remove past 3 (init somewher else)
638 h->ref_cache[list][scan8[4 ]] =
639 h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
640 *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
641 *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
642 *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewher else)
643 *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
644 *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
647 /* XXX beurk, Load mvd */
648 if(IS_INTER(topleft_type)){
649 const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
650 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy];
652 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 - 1*8]= 0;
655 if(IS_INTER(top_type)){
656 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
657 *(uint32_t*)h->mvd_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 0];
658 *(uint32_t*)h->mvd_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 1];
659 *(uint32_t*)h->mvd_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 2];
660 *(uint32_t*)h->mvd_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 3];
662 *(uint32_t*)h->mvd_cache [list][scan8[0] + 0 - 1*8]=
663 *(uint32_t*)h->mvd_cache [list][scan8[0] + 1 - 1*8]=
664 *(uint32_t*)h->mvd_cache [list][scan8[0] + 2 - 1*8]=
665 *(uint32_t*)h->mvd_cache [list][scan8[0] + 3 - 1*8]= 0;
667 if(IS_INTER(left_type[0])){
668 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
669 *(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]];
670 *(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]];
672 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 0*8]=
673 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 1*8]= 0;
675 if(IS_INTER(left_type[1])){
676 const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
677 *(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]];
678 *(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]];
680 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 2*8]=
681 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 3*8]= 0;
683 *(uint32_t*)h->mvd_cache [list][scan8[5 ]+1]=
684 *(uint32_t*)h->mvd_cache [list][scan8[7 ]+1]=
685 *(uint32_t*)h->mvd_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewher else)
686 *(uint32_t*)h->mvd_cache [list][scan8[4 ]]=
687 *(uint32_t*)h->mvd_cache [list][scan8[12]]= 0;
695 static inline void write_back_intra_pred_mode(H264Context *h){
696 MpegEncContext * const s = &h->s;
697 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
699 h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
700 h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
701 h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
702 h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
703 h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
704 h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
705 h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
709 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
711 static inline int check_intra4x4_pred_mode(H264Context *h){
712 MpegEncContext * const s = &h->s;
713 static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
714 static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
717 if(!(h->top_samples_available&0x8000)){
719 int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
721 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);
724 h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
729 if(!(h->left_samples_available&0x8000)){
731 int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
733 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);
736 h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
742 } //FIXME cleanup like next
745 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
747 static inline int check_intra_pred_mode(H264Context *h, int mode){
748 MpegEncContext * const s = &h->s;
749 static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
750 static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
752 if(mode < 0 || mode > 6)
755 if(!(h->top_samples_available&0x8000)){
758 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);
763 if(!(h->left_samples_available&0x8000)){
766 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);
775 * gets the predicted intra4x4 prediction mode.
777 static inline int pred_intra_mode(H264Context *h, int n){
778 const int index8= scan8[n];
779 const int left= h->intra4x4_pred_mode_cache[index8 - 1];
780 const int top = h->intra4x4_pred_mode_cache[index8 - 8];
781 const int min= FFMIN(left, top);
783 tprintf("mode:%d %d min:%d\n", left ,top, min);
785 if(min<0) return DC_PRED;
789 static inline void write_back_non_zero_count(H264Context *h){
790 MpegEncContext * const s = &h->s;
791 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
793 h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[4+8*4];
794 h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[5+8*4];
795 h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[6+8*4];
796 h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];
797 h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[7+8*3];
798 h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[7+8*2];
799 h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[7+8*1];
801 h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[1+8*2];
802 h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
803 h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[2+8*1];
805 h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[1+8*5];
806 h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
807 h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[2+8*4];
811 * gets the predicted number of non zero coefficients.
812 * @param n block index
814 static inline int pred_non_zero_count(H264Context *h, int n){
815 const int index8= scan8[n];
816 const int left= h->non_zero_count_cache[index8 - 1];
817 const int top = h->non_zero_count_cache[index8 - 8];
820 if(i<64) i= (i+1)>>1;
822 tprintf("pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
827 static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
828 const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];
830 if(topright_ref != PART_NOT_AVAILABLE){
831 *C= h->mv_cache[list][ i - 8 + part_width ];
834 tprintf("topright MV not available\n");
836 *C= h->mv_cache[list][ i - 8 - 1 ];
837 return h->ref_cache[list][ i - 8 - 1 ];
842 * gets the predicted MV.
843 * @param n the block index
844 * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)
845 * @param mx the x component of the predicted motion vector
846 * @param my the y component of the predicted motion vector
848 static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
849 const int index8= scan8[n];
850 const int top_ref= h->ref_cache[list][ index8 - 8 ];
851 const int left_ref= h->ref_cache[list][ index8 - 1 ];
852 const int16_t * const A= h->mv_cache[list][ index8 - 1 ];
853 const int16_t * const B= h->mv_cache[list][ index8 - 8 ];
855 int diagonal_ref, match_count;
857 assert(part_width==1 || part_width==2 || part_width==4);
867 diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);
868 match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);
869 if(match_count > 1){ //most common
870 *mx= mid_pred(A[0], B[0], C[0]);
871 *my= mid_pred(A[1], B[1], C[1]);
872 }else if(match_count==1){
876 }else if(top_ref==ref){
884 if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
888 *mx= mid_pred(A[0], B[0], C[0]);
889 *my= mid_pred(A[1], B[1], C[1]);
893 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);
897 * gets the directionally predicted 16x8 MV.
898 * @param n the block index
899 * @param mx the x component of the predicted motion vector
900 * @param my the y component of the predicted motion vector
902 static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
904 const int top_ref= h->ref_cache[list][ scan8[0] - 8 ];
905 const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
907 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);
915 const int left_ref= h->ref_cache[list][ scan8[8] - 1 ];
916 const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];
918 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);
928 pred_motion(h, n, 4, list, ref, mx, my);
932 * gets the directionally predicted 8x16 MV.
933 * @param n the block index
934 * @param mx the x component of the predicted motion vector
935 * @param my the y component of the predicted motion vector
937 static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
939 const int left_ref= h->ref_cache[list][ scan8[0] - 1 ];
940 const int16_t * const A= h->mv_cache[list][ scan8[0] - 1 ];
942 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);
953 diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);
955 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);
957 if(diagonal_ref == ref){
965 pred_motion(h, n, 2, list, ref, mx, my);
968 static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
969 const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
970 const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];
972 tprintf("pred_pskip: (%d) (%d) at %2d %2d", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
974 if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
975 || (top_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0)
976 || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){
982 pred_motion(h, 0, 4, 0, 0, mx, my);
987 static inline void write_back_motion(H264Context *h, int mb_type){
988 MpegEncContext * const s = &h->s;
989 const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
990 const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
993 for(list=0; list<2; list++){
995 if((!IS_8X8(mb_type)) && !USES_LIST(mb_type, list)){
996 if(1){ //FIXME skip or never read if mb_type doesnt use it
998 *(uint64_t*)s->current_picture.motion_val[list][b_xy + 0 + y*h->b_stride]=
999 *(uint64_t*)s->current_picture.motion_val[list][b_xy + 2 + y*h->b_stride]= 0;
1001 if( h->pps.cabac ) {
1002 /* FIXME needed ? */
1004 *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]=
1005 *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= 0;
1009 *(uint16_t*)s->current_picture.motion_val[list][b8_xy + y*h->b8_stride]= (LIST_NOT_USED&0xFF)*0x0101;
1012 continue; //FIXME direct mode ...
1016 *(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];
1017 *(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];
1019 if( h->pps.cabac ) {
1021 *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+0 + 8*y];
1022 *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+2 + 8*y];
1026 s->current_picture.ref_index[list][b8_xy + 0 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+0 + 16*y];
1027 s->current_picture.ref_index[list][b8_xy + 1 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+2 + 16*y];
1033 * Decodes a network abstraction layer unit.
1034 * @param consumed is the number of bytes used as input
1035 * @param length is the length of the array
1036 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp ttailing?
1037 * @returns decoded bytes, might be src+1 if no escapes
1039 static uint8_t *decode_nal(H264Context *h, uint8_t *src, int *dst_length, int *consumed, int length){
1043 // src[0]&0x80; //forbidden bit
1044 h->nal_ref_idc= src[0]>>5;
1045 h->nal_unit_type= src[0]&0x1F;
1049 for(i=0; i<length; i++)
1050 printf("%2X ", src[i]);
1052 for(i=0; i+1<length; i+=2){
1053 if(src[i]) continue;
1054 if(i>0 && src[i-1]==0) i--;
1055 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1057 /* startcode, so we must be past the end */
1064 if(i>=length-1){ //no escaped 0
1065 *dst_length= length;
1066 *consumed= length+1; //+1 for the header
1070 h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length);
1071 dst= h->rbsp_buffer;
1073 //printf("deoding esc\n");
1076 //remove escapes (very rare 1:2^22)
1077 if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1078 if(src[si+2]==3){ //escape
1083 }else //next start code
1087 dst[di++]= src[si++];
1091 *consumed= si + 1;//+1 for the header
1092 //FIXME store exact number of bits in the getbitcontext (its needed for decoding)
1098 * @param src the data which should be escaped
1099 * @param dst the target buffer, dst+1 == src is allowed as a special case
1100 * @param length the length of the src data
1101 * @param dst_length the length of the dst array
1102 * @returns length of escaped data in bytes or -1 if an error occured
1104 static int encode_nal(H264Context *h, uint8_t *dst, uint8_t *src, int length, int dst_length){
1105 int i, escape_count, si, di;
1109 assert(dst_length>0);
1111 dst[0]= (h->nal_ref_idc<<5) + h->nal_unit_type;
1113 if(length==0) return 1;
1116 for(i=0; i<length; i+=2){
1117 if(src[i]) continue;
1118 if(i>0 && src[i-1]==0)
1120 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1126 if(escape_count==0){
1128 memcpy(dst+1, src, length);
1132 if(length + escape_count + 1> dst_length)
1135 //this should be damn rare (hopefully)
1137 h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length + escape_count);
1138 temp= h->rbsp_buffer;
1139 //printf("encoding esc\n");
1144 if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1145 temp[di++]= 0; si++;
1146 temp[di++]= 0; si++;
1148 temp[di++]= src[si++];
1151 temp[di++]= src[si++];
1153 memcpy(dst+1, temp, length+escape_count);
1155 assert(di == length+escape_count);
1161 * write 1,10,100,1000,... for alignment, yes its exactly inverse to mpeg4
1163 static void encode_rbsp_trailing(PutBitContext *pb){
1166 length= (-put_bits_count(pb))&7;
1167 if(length) put_bits(pb, length, 0);
1172 * identifies the exact end of the bitstream
1173 * @return the length of the trailing, or 0 if damaged
1175 static int decode_rbsp_trailing(uint8_t *src){
1179 tprintf("rbsp trailing %X\n", v);
1189 * idct tranforms the 16 dc values and dequantize them.
1190 * @param qp quantization parameter
1192 static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp){
1193 const int qmul= dequant_coeff[qp][0];
1196 int temp[16]; //FIXME check if this is a good idea
1197 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
1198 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1200 //memset(block, 64, 2*256);
1203 const int offset= y_offset[i];
1204 const int z0= block[offset+stride*0] + block[offset+stride*4];
1205 const int z1= block[offset+stride*0] - block[offset+stride*4];
1206 const int z2= block[offset+stride*1] - block[offset+stride*5];
1207 const int z3= block[offset+stride*1] + block[offset+stride*5];
1216 const int offset= x_offset[i];
1217 const int z0= temp[4*0+i] + temp[4*2+i];
1218 const int z1= temp[4*0+i] - temp[4*2+i];
1219 const int z2= temp[4*1+i] - temp[4*3+i];
1220 const int z3= temp[4*1+i] + temp[4*3+i];
1222 block[stride*0 +offset]= ((z0 + z3)*qmul + 2)>>2; //FIXME think about merging this into decode_resdual
1223 block[stride*2 +offset]= ((z1 + z2)*qmul + 2)>>2;
1224 block[stride*8 +offset]= ((z1 - z2)*qmul + 2)>>2;
1225 block[stride*10+offset]= ((z0 - z3)*qmul + 2)>>2;
1231 * dct tranforms the 16 dc values.
1232 * @param qp quantization parameter ??? FIXME
1234 static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
1235 // const int qmul= dequant_coeff[qp][0];
1237 int temp[16]; //FIXME check if this is a good idea
1238 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
1239 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1242 const int offset= y_offset[i];
1243 const int z0= block[offset+stride*0] + block[offset+stride*4];
1244 const int z1= block[offset+stride*0] - block[offset+stride*4];
1245 const int z2= block[offset+stride*1] - block[offset+stride*5];
1246 const int z3= block[offset+stride*1] + block[offset+stride*5];
1255 const int offset= x_offset[i];
1256 const int z0= temp[4*0+i] + temp[4*2+i];
1257 const int z1= temp[4*0+i] - temp[4*2+i];
1258 const int z2= temp[4*1+i] - temp[4*3+i];
1259 const int z3= temp[4*1+i] + temp[4*3+i];
1261 block[stride*0 +offset]= (z0 + z3)>>1;
1262 block[stride*2 +offset]= (z1 + z2)>>1;
1263 block[stride*8 +offset]= (z1 - z2)>>1;
1264 block[stride*10+offset]= (z0 - z3)>>1;
1272 static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp){
1273 const int qmul= dequant_coeff[qp][0];
1274 const int stride= 16*2;
1275 const int xStride= 16;
1278 a= block[stride*0 + xStride*0];
1279 b= block[stride*0 + xStride*1];
1280 c= block[stride*1 + xStride*0];
1281 d= block[stride*1 + xStride*1];
1288 block[stride*0 + xStride*0]= ((a+c)*qmul + 0)>>1;
1289 block[stride*0 + xStride*1]= ((e+b)*qmul + 0)>>1;
1290 block[stride*1 + xStride*0]= ((a-c)*qmul + 0)>>1;
1291 block[stride*1 + xStride*1]= ((e-b)*qmul + 0)>>1;
1295 static void chroma_dc_dct_c(DCTELEM *block){
1296 const int stride= 16*2;
1297 const int xStride= 16;
1300 a= block[stride*0 + xStride*0];
1301 b= block[stride*0 + xStride*1];
1302 c= block[stride*1 + xStride*0];
1303 d= block[stride*1 + xStride*1];
1310 block[stride*0 + xStride*0]= (a+c);
1311 block[stride*0 + xStride*1]= (e+b);
1312 block[stride*1 + xStride*0]= (a-c);
1313 block[stride*1 + xStride*1]= (e-b);
1318 * gets the chroma qp.
1320 static inline int get_chroma_qp(H264Context *h, int qscale){
1322 return chroma_qp[clip(qscale + h->pps.chroma_qp_index_offset, 0, 51)];
1329 static void h264_add_idct_c(uint8_t *dst, DCTELEM *block, int stride){
1331 uint8_t *cm = cropTbl + MAX_NEG_CROP;
1336 const int z0= block[0 + 4*i] + block[2 + 4*i];
1337 const int z1= block[0 + 4*i] - block[2 + 4*i];
1338 const int z2= (block[1 + 4*i]>>1) - block[3 + 4*i];
1339 const int z3= block[1 + 4*i] + (block[3 + 4*i]>>1);
1341 block[0 + 4*i]= z0 + z3;
1342 block[1 + 4*i]= z1 + z2;
1343 block[2 + 4*i]= z1 - z2;
1344 block[3 + 4*i]= z0 - z3;
1348 const int z0= block[i + 4*0] + block[i + 4*2];
1349 const int z1= block[i + 4*0] - block[i + 4*2];
1350 const int z2= (block[i + 4*1]>>1) - block[i + 4*3];
1351 const int z3= block[i + 4*1] + (block[i + 4*3]>>1);
1353 dst[i + 0*stride]= cm[ dst[i + 0*stride] + ((z0 + z3) >> 6) ];
1354 dst[i + 1*stride]= cm[ dst[i + 1*stride] + ((z1 + z2) >> 6) ];
1355 dst[i + 2*stride]= cm[ dst[i + 2*stride] + ((z1 - z2) >> 6) ];
1356 dst[i + 3*stride]= cm[ dst[i + 3*stride] + ((z0 - z3) >> 6) ];
1361 static void h264_diff_dct_c(DCTELEM *block, uint8_t *src1, uint8_t *src2, int stride){
1363 //FIXME try int temp instead of block
1366 const int d0= src1[0 + i*stride] - src2[0 + i*stride];
1367 const int d1= src1[1 + i*stride] - src2[1 + i*stride];
1368 const int d2= src1[2 + i*stride] - src2[2 + i*stride];
1369 const int d3= src1[3 + i*stride] - src2[3 + i*stride];
1370 const int z0= d0 + d3;
1371 const int z3= d0 - d3;
1372 const int z1= d1 + d2;
1373 const int z2= d1 - d2;
1375 block[0 + 4*i]= z0 + z1;
1376 block[1 + 4*i]= 2*z3 + z2;
1377 block[2 + 4*i]= z0 - z1;
1378 block[3 + 4*i]= z3 - 2*z2;
1382 const int z0= block[0*4 + i] + block[3*4 + i];
1383 const int z3= block[0*4 + i] - block[3*4 + i];
1384 const int z1= block[1*4 + i] + block[2*4 + i];
1385 const int z2= block[1*4 + i] - block[2*4 + i];
1387 block[0*4 + i]= z0 + z1;
1388 block[1*4 + i]= 2*z3 + z2;
1389 block[2*4 + i]= z0 - z1;
1390 block[3*4 + i]= z3 - 2*z2;
1395 //FIXME need to check that this doesnt overflow signed 32 bit for low qp, iam not sure, its very close
1396 //FIXME check that gcc inlines this (and optimizes intra & seperate_dc stuff away)
1397 static inline int quantize_c(DCTELEM *block, uint8_t *scantable, int qscale, int intra, int seperate_dc){
1399 const int * const quant_table= quant_coeff[qscale];
1400 const int bias= intra ? (1<<QUANT_SHIFT)/3 : (1<<QUANT_SHIFT)/6;
1401 const unsigned int threshold1= (1<<QUANT_SHIFT) - bias - 1;
1402 const unsigned int threshold2= (threshold1<<1);
1408 const int dc_bias= intra ? (1<<(QUANT_SHIFT-2))/3 : (1<<(QUANT_SHIFT-2))/6;
1409 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT-2)) - dc_bias - 1;
1410 const unsigned int dc_threshold2= (dc_threshold1<<1);
1412 int level= block[0]*quant_coeff[qscale+18][0];
1413 if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1415 level= (dc_bias + level)>>(QUANT_SHIFT-2);
1418 level= (dc_bias - level)>>(QUANT_SHIFT-2);
1421 // last_non_zero = i;
1426 const int dc_bias= intra ? (1<<(QUANT_SHIFT+1))/3 : (1<<(QUANT_SHIFT+1))/6;
1427 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT+1)) - dc_bias - 1;
1428 const unsigned int dc_threshold2= (dc_threshold1<<1);
1430 int level= block[0]*quant_table[0];
1431 if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1433 level= (dc_bias + level)>>(QUANT_SHIFT+1);
1436 level= (dc_bias - level)>>(QUANT_SHIFT+1);
1439 // last_non_zero = i;
1452 const int j= scantable[i];
1453 int level= block[j]*quant_table[j];
1455 // if( bias+level >= (1<<(QMAT_SHIFT - 3))
1456 // || bias-level >= (1<<(QMAT_SHIFT - 3))){
1457 if(((unsigned)(level+threshold1))>threshold2){
1459 level= (bias + level)>>QUANT_SHIFT;
1462 level= (bias - level)>>QUANT_SHIFT;
1471 return last_non_zero;
1474 static void pred4x4_vertical_c(uint8_t *src, uint8_t *topright, int stride){
1475 const uint32_t a= ((uint32_t*)(src-stride))[0];
1476 ((uint32_t*)(src+0*stride))[0]= a;
1477 ((uint32_t*)(src+1*stride))[0]= a;
1478 ((uint32_t*)(src+2*stride))[0]= a;
1479 ((uint32_t*)(src+3*stride))[0]= a;
1482 static void pred4x4_horizontal_c(uint8_t *src, uint8_t *topright, int stride){
1483 ((uint32_t*)(src+0*stride))[0]= src[-1+0*stride]*0x01010101;
1484 ((uint32_t*)(src+1*stride))[0]= src[-1+1*stride]*0x01010101;
1485 ((uint32_t*)(src+2*stride))[0]= src[-1+2*stride]*0x01010101;
1486 ((uint32_t*)(src+3*stride))[0]= src[-1+3*stride]*0x01010101;
1489 static void pred4x4_dc_c(uint8_t *src, uint8_t *topright, int stride){
1490 const int dc= ( src[-stride] + src[1-stride] + src[2-stride] + src[3-stride]
1491 + src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 4) >>3;
1493 ((uint32_t*)(src+0*stride))[0]=
1494 ((uint32_t*)(src+1*stride))[0]=
1495 ((uint32_t*)(src+2*stride))[0]=
1496 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1499 static void pred4x4_left_dc_c(uint8_t *src, uint8_t *topright, int stride){
1500 const int dc= ( src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 2) >>2;
1502 ((uint32_t*)(src+0*stride))[0]=
1503 ((uint32_t*)(src+1*stride))[0]=
1504 ((uint32_t*)(src+2*stride))[0]=
1505 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1508 static void pred4x4_top_dc_c(uint8_t *src, uint8_t *topright, int stride){
1509 const int dc= ( src[-stride] + src[1-stride] + src[2-stride] + src[3-stride] + 2) >>2;
1511 ((uint32_t*)(src+0*stride))[0]=
1512 ((uint32_t*)(src+1*stride))[0]=
1513 ((uint32_t*)(src+2*stride))[0]=
1514 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1517 static void pred4x4_128_dc_c(uint8_t *src, uint8_t *topright, int stride){
1518 ((uint32_t*)(src+0*stride))[0]=
1519 ((uint32_t*)(src+1*stride))[0]=
1520 ((uint32_t*)(src+2*stride))[0]=
1521 ((uint32_t*)(src+3*stride))[0]= 128U*0x01010101U;
1525 #define LOAD_TOP_RIGHT_EDGE\
1526 const int t4= topright[0];\
1527 const int t5= topright[1];\
1528 const int t6= topright[2];\
1529 const int t7= topright[3];\
1531 #define LOAD_LEFT_EDGE\
1532 const int l0= src[-1+0*stride];\
1533 const int l1= src[-1+1*stride];\
1534 const int l2= src[-1+2*stride];\
1535 const int l3= src[-1+3*stride];\
1537 #define LOAD_TOP_EDGE\
1538 const int t0= src[ 0-1*stride];\
1539 const int t1= src[ 1-1*stride];\
1540 const int t2= src[ 2-1*stride];\
1541 const int t3= src[ 3-1*stride];\
1543 static void pred4x4_down_right_c(uint8_t *src, uint8_t *topright, int stride){
1544 const int lt= src[-1-1*stride];
1548 src[0+3*stride]=(l3 + 2*l2 + l1 + 2)>>2;
1550 src[1+3*stride]=(l2 + 2*l1 + l0 + 2)>>2;
1553 src[2+3*stride]=(l1 + 2*l0 + lt + 2)>>2;
1557 src[3+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1560 src[3+2*stride]=(lt + 2*t0 + t1 + 2)>>2;
1562 src[3+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1563 src[3+0*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1566 static void pred4x4_down_left_c(uint8_t *src, uint8_t *topright, int stride){
1571 src[0+0*stride]=(t0 + t2 + 2*t1 + 2)>>2;
1573 src[0+1*stride]=(t1 + t3 + 2*t2 + 2)>>2;
1576 src[0+2*stride]=(t2 + t4 + 2*t3 + 2)>>2;
1580 src[0+3*stride]=(t3 + t5 + 2*t4 + 2)>>2;
1583 src[1+3*stride]=(t4 + t6 + 2*t5 + 2)>>2;
1585 src[2+3*stride]=(t5 + t7 + 2*t6 + 2)>>2;
1586 src[3+3*stride]=(t6 + 3*t7 + 2)>>2;
1589 static void pred4x4_vertical_right_c(uint8_t *src, uint8_t *topright, int stride){
1590 const int lt= src[-1-1*stride];
1593 const __attribute__((unused)) int unu= l3;
1596 src[1+2*stride]=(lt + t0 + 1)>>1;
1598 src[2+2*stride]=(t0 + t1 + 1)>>1;
1600 src[3+2*stride]=(t1 + t2 + 1)>>1;
1601 src[3+0*stride]=(t2 + t3 + 1)>>1;
1603 src[1+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1605 src[2+3*stride]=(lt + 2*t0 + t1 + 2)>>2;
1607 src[3+3*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1608 src[3+1*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1609 src[0+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1610 src[0+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1613 static void pred4x4_vertical_left_c(uint8_t *src, uint8_t *topright, int stride){
1616 const __attribute__((unused)) int unu= t7;
1618 src[0+0*stride]=(t0 + t1 + 1)>>1;
1620 src[0+2*stride]=(t1 + t2 + 1)>>1;
1622 src[1+2*stride]=(t2 + t3 + 1)>>1;
1624 src[2+2*stride]=(t3 + t4+ 1)>>1;
1625 src[3+2*stride]=(t4 + t5+ 1)>>1;
1626 src[0+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1628 src[0+3*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1630 src[1+3*stride]=(t2 + 2*t3 + t4 + 2)>>2;
1632 src[2+3*stride]=(t3 + 2*t4 + t5 + 2)>>2;
1633 src[3+3*stride]=(t4 + 2*t5 + t6 + 2)>>2;
1636 static void pred4x4_horizontal_up_c(uint8_t *src, uint8_t *topright, int stride){
1639 src[0+0*stride]=(l0 + l1 + 1)>>1;
1640 src[1+0*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1642 src[0+1*stride]=(l1 + l2 + 1)>>1;
1644 src[1+1*stride]=(l1 + 2*l2 + l3 + 2)>>2;
1646 src[0+2*stride]=(l2 + l3 + 1)>>1;
1648 src[1+2*stride]=(l2 + 2*l3 + l3 + 2)>>2;
1657 static void pred4x4_horizontal_down_c(uint8_t *src, uint8_t *topright, int stride){
1658 const int lt= src[-1-1*stride];
1661 const __attribute__((unused)) int unu= t3;
1664 src[2+1*stride]=(lt + l0 + 1)>>1;
1666 src[3+1*stride]=(l0 + 2*lt + t0 + 2)>>2;
1667 src[2+0*stride]=(lt + 2*t0 + t1 + 2)>>2;
1668 src[3+0*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1670 src[2+2*stride]=(l0 + l1 + 1)>>1;
1672 src[3+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1674 src[2+3*stride]=(l1 + l2+ 1)>>1;
1676 src[3+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1677 src[0+3*stride]=(l2 + l3 + 1)>>1;
1678 src[1+3*stride]=(l1 + 2*l2 + l3 + 2)>>2;
1681 static void pred16x16_vertical_c(uint8_t *src, int stride){
1683 const uint32_t a= ((uint32_t*)(src-stride))[0];
1684 const uint32_t b= ((uint32_t*)(src-stride))[1];
1685 const uint32_t c= ((uint32_t*)(src-stride))[2];
1686 const uint32_t d= ((uint32_t*)(src-stride))[3];
1688 for(i=0; i<16; i++){
1689 ((uint32_t*)(src+i*stride))[0]= a;
1690 ((uint32_t*)(src+i*stride))[1]= b;
1691 ((uint32_t*)(src+i*stride))[2]= c;
1692 ((uint32_t*)(src+i*stride))[3]= d;
1696 static void pred16x16_horizontal_c(uint8_t *src, int stride){
1699 for(i=0; i<16; i++){
1700 ((uint32_t*)(src+i*stride))[0]=
1701 ((uint32_t*)(src+i*stride))[1]=
1702 ((uint32_t*)(src+i*stride))[2]=
1703 ((uint32_t*)(src+i*stride))[3]= src[-1+i*stride]*0x01010101;
1707 static void pred16x16_dc_c(uint8_t *src, int stride){
1711 dc+= src[-1+i*stride];
1718 dc= 0x01010101*((dc + 16)>>5);
1720 for(i=0; i<16; i++){
1721 ((uint32_t*)(src+i*stride))[0]=
1722 ((uint32_t*)(src+i*stride))[1]=
1723 ((uint32_t*)(src+i*stride))[2]=
1724 ((uint32_t*)(src+i*stride))[3]= dc;
1728 static void pred16x16_left_dc_c(uint8_t *src, int stride){
1732 dc+= src[-1+i*stride];
1735 dc= 0x01010101*((dc + 8)>>4);
1737 for(i=0; i<16; i++){
1738 ((uint32_t*)(src+i*stride))[0]=
1739 ((uint32_t*)(src+i*stride))[1]=
1740 ((uint32_t*)(src+i*stride))[2]=
1741 ((uint32_t*)(src+i*stride))[3]= dc;
1745 static void pred16x16_top_dc_c(uint8_t *src, int stride){
1751 dc= 0x01010101*((dc + 8)>>4);
1753 for(i=0; i<16; i++){
1754 ((uint32_t*)(src+i*stride))[0]=
1755 ((uint32_t*)(src+i*stride))[1]=
1756 ((uint32_t*)(src+i*stride))[2]=
1757 ((uint32_t*)(src+i*stride))[3]= dc;
1761 static void pred16x16_128_dc_c(uint8_t *src, int stride){
1764 for(i=0; i<16; i++){
1765 ((uint32_t*)(src+i*stride))[0]=
1766 ((uint32_t*)(src+i*stride))[1]=
1767 ((uint32_t*)(src+i*stride))[2]=
1768 ((uint32_t*)(src+i*stride))[3]= 0x01010101U*128U;
1772 static inline void pred16x16_plane_compat_c(uint8_t *src, int stride, const int svq3){
1775 uint8_t *cm = cropTbl + MAX_NEG_CROP;
1776 const uint8_t * const src0 = src+7-stride;
1777 const uint8_t *src1 = src+8*stride-1;
1778 const uint8_t *src2 = src1-2*stride; // == src+6*stride-1;
1779 int H = src0[1] - src0[-1];
1780 int V = src1[0] - src2[ 0];
1781 for(k=2; k<=8; ++k) {
1782 src1 += stride; src2 -= stride;
1783 H += k*(src0[k] - src0[-k]);
1784 V += k*(src1[0] - src2[ 0]);
1787 H = ( 5*(H/4) ) / 16;
1788 V = ( 5*(V/4) ) / 16;
1790 /* required for 100% accuracy */
1791 i = H; H = V; V = i;
1793 H = ( 5*H+32 ) >> 6;
1794 V = ( 5*V+32 ) >> 6;
1797 a = 16*(src1[0] + src2[16] + 1) - 7*(V+H);
1798 for(j=16; j>0; --j) {
1801 for(i=-16; i<0; i+=4) {
1802 src[16+i] = cm[ (b ) >> 5 ];
1803 src[17+i] = cm[ (b+ H) >> 5 ];
1804 src[18+i] = cm[ (b+2*H) >> 5 ];
1805 src[19+i] = cm[ (b+3*H) >> 5 ];
1812 static void pred16x16_plane_c(uint8_t *src, int stride){
1813 pred16x16_plane_compat_c(src, stride, 0);
1816 static void pred8x8_vertical_c(uint8_t *src, int stride){
1818 const uint32_t a= ((uint32_t*)(src-stride))[0];
1819 const uint32_t b= ((uint32_t*)(src-stride))[1];
1822 ((uint32_t*)(src+i*stride))[0]= a;
1823 ((uint32_t*)(src+i*stride))[1]= b;
1827 static void pred8x8_horizontal_c(uint8_t *src, int stride){
1831 ((uint32_t*)(src+i*stride))[0]=
1832 ((uint32_t*)(src+i*stride))[1]= src[-1+i*stride]*0x01010101;
1836 static void pred8x8_128_dc_c(uint8_t *src, int stride){
1840 ((uint32_t*)(src+i*stride))[0]=
1841 ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
1844 ((uint32_t*)(src+i*stride))[0]=
1845 ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
1849 static void pred8x8_left_dc_c(uint8_t *src, int stride){
1855 dc0+= src[-1+i*stride];
1856 dc2+= src[-1+(i+4)*stride];
1858 dc0= 0x01010101*((dc0 + 2)>>2);
1859 dc2= 0x01010101*((dc2 + 2)>>2);
1862 ((uint32_t*)(src+i*stride))[0]=
1863 ((uint32_t*)(src+i*stride))[1]= dc0;
1866 ((uint32_t*)(src+i*stride))[0]=
1867 ((uint32_t*)(src+i*stride))[1]= dc2;
1871 static void pred8x8_top_dc_c(uint8_t *src, int stride){
1877 dc0+= src[i-stride];
1878 dc1+= src[4+i-stride];
1880 dc0= 0x01010101*((dc0 + 2)>>2);
1881 dc1= 0x01010101*((dc1 + 2)>>2);
1884 ((uint32_t*)(src+i*stride))[0]= dc0;
1885 ((uint32_t*)(src+i*stride))[1]= dc1;
1888 ((uint32_t*)(src+i*stride))[0]= dc0;
1889 ((uint32_t*)(src+i*stride))[1]= dc1;
1894 static void pred8x8_dc_c(uint8_t *src, int stride){
1896 int dc0, dc1, dc2, dc3;
1900 dc0+= src[-1+i*stride] + src[i-stride];
1901 dc1+= src[4+i-stride];
1902 dc2+= src[-1+(i+4)*stride];
1904 dc3= 0x01010101*((dc1 + dc2 + 4)>>3);
1905 dc0= 0x01010101*((dc0 + 4)>>3);
1906 dc1= 0x01010101*((dc1 + 2)>>2);
1907 dc2= 0x01010101*((dc2 + 2)>>2);
1910 ((uint32_t*)(src+i*stride))[0]= dc0;
1911 ((uint32_t*)(src+i*stride))[1]= dc1;
1914 ((uint32_t*)(src+i*stride))[0]= dc2;
1915 ((uint32_t*)(src+i*stride))[1]= dc3;
1919 static void pred8x8_plane_c(uint8_t *src, int stride){
1922 uint8_t *cm = cropTbl + MAX_NEG_CROP;
1923 const uint8_t * const src0 = src+3-stride;
1924 const uint8_t *src1 = src+4*stride-1;
1925 const uint8_t *src2 = src1-2*stride; // == src+2*stride-1;
1926 int H = src0[1] - src0[-1];
1927 int V = src1[0] - src2[ 0];
1928 for(k=2; k<=4; ++k) {
1929 src1 += stride; src2 -= stride;
1930 H += k*(src0[k] - src0[-k]);
1931 V += k*(src1[0] - src2[ 0]);
1933 H = ( 17*H+16 ) >> 5;
1934 V = ( 17*V+16 ) >> 5;
1936 a = 16*(src1[0] + src2[8]+1) - 3*(V+H);
1937 for(j=8; j>0; --j) {
1940 src[0] = cm[ (b ) >> 5 ];
1941 src[1] = cm[ (b+ H) >> 5 ];
1942 src[2] = cm[ (b+2*H) >> 5 ];
1943 src[3] = cm[ (b+3*H) >> 5 ];
1944 src[4] = cm[ (b+4*H) >> 5 ];
1945 src[5] = cm[ (b+5*H) >> 5 ];
1946 src[6] = cm[ (b+6*H) >> 5 ];
1947 src[7] = cm[ (b+7*H) >> 5 ];
1952 static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
1953 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
1954 int src_x_offset, int src_y_offset,
1955 qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
1956 MpegEncContext * const s = &h->s;
1957 const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
1958 const int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
1959 const int luma_xy= (mx&3) + ((my&3)<<2);
1960 uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*s->linesize;
1961 uint8_t * src_cb= pic->data[1] + (mx>>3) + (my>>3)*s->uvlinesize;
1962 uint8_t * src_cr= pic->data[2] + (mx>>3) + (my>>3)*s->uvlinesize;
1963 int extra_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16; //FIXME increase edge?, IMHO not worth it
1964 int extra_height= extra_width;
1966 const int full_mx= mx>>2;
1967 const int full_my= my>>2;
1969 assert(pic->data[0]);
1971 if(mx&7) extra_width -= 3;
1972 if(my&7) extra_height -= 3;
1974 if( full_mx < 0-extra_width
1975 || full_my < 0-extra_height
1976 || full_mx + 16/*FIXME*/ > s->width + extra_width
1977 || full_my + 16/*FIXME*/ > s->height + extra_height){
1978 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);
1979 src_y= s->edge_emu_buffer + 2 + 2*s->linesize;
1983 qpix_op[luma_xy](dest_y, src_y, s->linesize); //FIXME try variable height perhaps?
1985 qpix_op[luma_xy](dest_y + delta, src_y + delta, s->linesize);
1988 if(s->flags&CODEC_FLAG_GRAY) return;
1991 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);
1992 src_cb= s->edge_emu_buffer;
1994 chroma_op(dest_cb, src_cb, s->uvlinesize, chroma_height, mx&7, my&7);
1997 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);
1998 src_cr= s->edge_emu_buffer;
2000 chroma_op(dest_cr, src_cr, s->uvlinesize, chroma_height, mx&7, my&7);
2003 static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
2004 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2005 int x_offset, int y_offset,
2006 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2007 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
2008 int list0, int list1){
2009 MpegEncContext * const s = &h->s;
2010 qpel_mc_func *qpix_op= qpix_put;
2011 h264_chroma_mc_func chroma_op= chroma_put;
2013 dest_y += 2*x_offset + 2*y_offset*s-> linesize;
2014 dest_cb += x_offset + y_offset*s->uvlinesize;
2015 dest_cr += x_offset + y_offset*s->uvlinesize;
2016 x_offset += 8*s->mb_x;
2017 y_offset += 8*s->mb_y;
2020 Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
2021 mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
2022 dest_y, dest_cb, dest_cr, x_offset, y_offset,
2023 qpix_op, chroma_op);
2026 chroma_op= chroma_avg;
2030 Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
2031 mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
2032 dest_y, dest_cb, dest_cr, x_offset, y_offset,
2033 qpix_op, chroma_op);
2037 static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2038 qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
2039 qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg)){
2040 MpegEncContext * const s = &h->s;
2041 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
2042 const int mb_type= s->current_picture.mb_type[mb_xy];
2044 assert(IS_INTER(mb_type));
2046 if(IS_16X16(mb_type)){
2047 mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
2048 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
2049 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2050 }else if(IS_16X8(mb_type)){
2051 mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
2052 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
2053 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2054 mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
2055 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
2056 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2057 }else if(IS_8X16(mb_type)){
2058 mc_part(h, 0, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 0, 0,
2059 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2060 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2061 mc_part(h, 4, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 4, 0,
2062 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2063 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2067 assert(IS_8X8(mb_type));
2070 const int sub_mb_type= h->sub_mb_type[i];
2072 int x_offset= (i&1)<<2;
2073 int y_offset= (i&2)<<1;
2075 if(IS_SUB_8X8(sub_mb_type)){
2076 mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2077 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2078 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2079 }else if(IS_SUB_8X4(sub_mb_type)){
2080 mc_part(h, n , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2081 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2082 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2083 mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
2084 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2085 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2086 }else if(IS_SUB_4X8(sub_mb_type)){
2087 mc_part(h, n , 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2088 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2089 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2090 mc_part(h, n+1, 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
2091 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2092 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2095 assert(IS_SUB_4X4(sub_mb_type));
2097 int sub_x_offset= x_offset + 2*(j&1);
2098 int sub_y_offset= y_offset + (j&2);
2099 mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
2100 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2101 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2108 static void decode_init_vlc(H264Context *h){
2109 static int done = 0;
2115 init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5,
2116 &chroma_dc_coeff_token_len [0], 1, 1,
2117 &chroma_dc_coeff_token_bits[0], 1, 1);
2120 init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17,
2121 &coeff_token_len [i][0], 1, 1,
2122 &coeff_token_bits[i][0], 1, 1);
2126 init_vlc(&chroma_dc_total_zeros_vlc[i], CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
2127 &chroma_dc_total_zeros_len [i][0], 1, 1,
2128 &chroma_dc_total_zeros_bits[i][0], 1, 1);
2130 for(i=0; i<15; i++){
2131 init_vlc(&total_zeros_vlc[i], TOTAL_ZEROS_VLC_BITS, 16,
2132 &total_zeros_len [i][0], 1, 1,
2133 &total_zeros_bits[i][0], 1, 1);
2137 init_vlc(&run_vlc[i], RUN_VLC_BITS, 7,
2138 &run_len [i][0], 1, 1,
2139 &run_bits[i][0], 1, 1);
2141 init_vlc(&run7_vlc, RUN7_VLC_BITS, 16,
2142 &run_len [6][0], 1, 1,
2143 &run_bits[6][0], 1, 1);
2148 * Sets the intra prediction function pointers.
2150 static void init_pred_ptrs(H264Context *h){
2151 // MpegEncContext * const s = &h->s;
2153 h->pred4x4[VERT_PRED ]= pred4x4_vertical_c;
2154 h->pred4x4[HOR_PRED ]= pred4x4_horizontal_c;
2155 h->pred4x4[DC_PRED ]= pred4x4_dc_c;
2156 h->pred4x4[DIAG_DOWN_LEFT_PRED ]= pred4x4_down_left_c;
2157 h->pred4x4[DIAG_DOWN_RIGHT_PRED]= pred4x4_down_right_c;
2158 h->pred4x4[VERT_RIGHT_PRED ]= pred4x4_vertical_right_c;
2159 h->pred4x4[HOR_DOWN_PRED ]= pred4x4_horizontal_down_c;
2160 h->pred4x4[VERT_LEFT_PRED ]= pred4x4_vertical_left_c;
2161 h->pred4x4[HOR_UP_PRED ]= pred4x4_horizontal_up_c;
2162 h->pred4x4[LEFT_DC_PRED ]= pred4x4_left_dc_c;
2163 h->pred4x4[TOP_DC_PRED ]= pred4x4_top_dc_c;
2164 h->pred4x4[DC_128_PRED ]= pred4x4_128_dc_c;
2166 h->pred8x8[DC_PRED8x8 ]= pred8x8_dc_c;
2167 h->pred8x8[VERT_PRED8x8 ]= pred8x8_vertical_c;
2168 h->pred8x8[HOR_PRED8x8 ]= pred8x8_horizontal_c;
2169 h->pred8x8[PLANE_PRED8x8 ]= pred8x8_plane_c;
2170 h->pred8x8[LEFT_DC_PRED8x8]= pred8x8_left_dc_c;
2171 h->pred8x8[TOP_DC_PRED8x8 ]= pred8x8_top_dc_c;
2172 h->pred8x8[DC_128_PRED8x8 ]= pred8x8_128_dc_c;
2174 h->pred16x16[DC_PRED8x8 ]= pred16x16_dc_c;
2175 h->pred16x16[VERT_PRED8x8 ]= pred16x16_vertical_c;
2176 h->pred16x16[HOR_PRED8x8 ]= pred16x16_horizontal_c;
2177 h->pred16x16[PLANE_PRED8x8 ]= pred16x16_plane_c;
2178 h->pred16x16[LEFT_DC_PRED8x8]= pred16x16_left_dc_c;
2179 h->pred16x16[TOP_DC_PRED8x8 ]= pred16x16_top_dc_c;
2180 h->pred16x16[DC_128_PRED8x8 ]= pred16x16_128_dc_c;
2183 static void free_tables(H264Context *h){
2184 av_freep(&h->intra4x4_pred_mode);
2185 av_freep(&h->chroma_pred_mode_table);
2186 av_freep(&h->cbp_table);
2187 av_freep(&h->mvd_table[0]);
2188 av_freep(&h->mvd_table[1]);
2189 av_freep(&h->non_zero_count);
2190 av_freep(&h->slice_table_base);
2191 av_freep(&h->top_border);
2192 h->slice_table= NULL;
2194 av_freep(&h->mb2b_xy);
2195 av_freep(&h->mb2b8_xy);
2200 * needs widzh/height
2202 static int alloc_tables(H264Context *h){
2203 MpegEncContext * const s = &h->s;
2204 const int big_mb_num= s->mb_stride * (s->mb_height+1);
2207 CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8 * sizeof(uint8_t))
2209 CHECKED_ALLOCZ(h->non_zero_count , big_mb_num * 16 * sizeof(uint8_t))
2210 CHECKED_ALLOCZ(h->slice_table_base , big_mb_num * sizeof(uint8_t))
2211 CHECKED_ALLOCZ(h->top_border , s->mb_width * (16+8+8) * sizeof(uint8_t))
2213 if( h->pps.cabac ) {
2214 CHECKED_ALLOCZ(h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t))
2215 CHECKED_ALLOCZ(h->cbp_table, big_mb_num * sizeof(uint16_t))
2216 CHECKED_ALLOCZ(h->mvd_table[0], 32*big_mb_num * sizeof(uint16_t));
2217 CHECKED_ALLOCZ(h->mvd_table[1], 32*big_mb_num * sizeof(uint16_t));
2220 memset(h->slice_table_base, -1, big_mb_num * sizeof(uint8_t));
2221 h->slice_table= h->slice_table_base + s->mb_stride + 1;
2223 CHECKED_ALLOCZ(h->mb2b_xy , big_mb_num * sizeof(uint16_t));
2224 CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint16_t));
2225 for(y=0; y<s->mb_height; y++){
2226 for(x=0; x<s->mb_width; x++){
2227 const int mb_xy= x + y*s->mb_stride;
2228 const int b_xy = 4*x + 4*y*h->b_stride;
2229 const int b8_xy= 2*x + 2*y*h->b8_stride;
2231 h->mb2b_xy [mb_xy]= b_xy;
2232 h->mb2b8_xy[mb_xy]= b8_xy;
2242 static void common_init(H264Context *h){
2243 MpegEncContext * const s = &h->s;
2245 s->width = s->avctx->width;
2246 s->height = s->avctx->height;
2247 s->codec_id= s->avctx->codec->id;
2251 s->unrestricted_mv=1;
2252 s->decode=1; //FIXME
2255 static int decode_init(AVCodecContext *avctx){
2256 H264Context *h= avctx->priv_data;
2257 MpegEncContext * const s = &h->s;
2259 MPV_decode_defaults(s);
2264 s->out_format = FMT_H264;
2265 s->workaround_bugs= avctx->workaround_bugs;
2268 // s->decode_mb= ff_h263_decode_mb;
2270 avctx->pix_fmt= PIX_FMT_YUV420P;
2274 if(avctx->codec_tag != 0x31637661) // avc1
2277 if((avctx->extradata_size == 0) || (avctx->extradata == NULL)) {
2278 av_log(avctx, AV_LOG_ERROR, "AVC codec requires avcC data\n");
2288 static void frame_start(H264Context *h){
2289 MpegEncContext * const s = &h->s;
2292 MPV_frame_start(s, s->avctx);
2293 ff_er_frame_start(s);
2296 assert(s->linesize && s->uvlinesize);
2298 for(i=0; i<16; i++){
2299 h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
2300 h->chroma_subblock_offset[i]= 2*((scan8[i] - scan8[0])&7) + 2*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2303 h->block_offset[16+i]=
2304 h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2307 // s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
2310 static inline void backup_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){
2311 MpegEncContext * const s = &h->s;
2315 src_cb -= uvlinesize;
2316 src_cr -= uvlinesize;
2318 h->left_border[0]= h->top_border[s->mb_x][15];
2319 for(i=1; i<17; i++){
2320 h->left_border[i]= src_y[15+i* linesize];
2323 *(uint64_t*)(h->top_border[s->mb_x]+0)= *(uint64_t*)(src_y + 16*linesize);
2324 *(uint64_t*)(h->top_border[s->mb_x]+8)= *(uint64_t*)(src_y +8+16*linesize);
2326 if(!(s->flags&CODEC_FLAG_GRAY)){
2327 h->left_border[17 ]= h->top_border[s->mb_x][16+7];
2328 h->left_border[17+9]= h->top_border[s->mb_x][24+7];
2330 h->left_border[i+17 ]= src_cb[7+i*uvlinesize];
2331 h->left_border[i+17+9]= src_cr[7+i*uvlinesize];
2333 *(uint64_t*)(h->top_border[s->mb_x]+16)= *(uint64_t*)(src_cb+8*uvlinesize);
2334 *(uint64_t*)(h->top_border[s->mb_x]+24)= *(uint64_t*)(src_cr+8*uvlinesize);
2338 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){
2339 MpegEncContext * const s = &h->s;
2342 int deblock_left = (s->mb_x > 0);
2343 int deblock_top = (s->mb_y > 0);
2345 src_y -= linesize + 1;
2346 src_cb -= uvlinesize + 1;
2347 src_cr -= uvlinesize + 1;
2349 #define XCHG(a,b,t,xchg)\
2356 for(i = !deblock_top; i<17; i++){
2357 XCHG(h->left_border[i ], src_y [i* linesize], temp8, xchg);
2362 XCHG(*(uint64_t*)(h->top_border[s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
2363 XCHG(*(uint64_t*)(h->top_border[s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
2366 if(!(s->flags&CODEC_FLAG_GRAY)){
2368 for(i = !deblock_top; i<9; i++){
2369 XCHG(h->left_border[i+17 ], src_cb[i*uvlinesize], temp8, xchg);
2370 XCHG(h->left_border[i+17+9], src_cr[i*uvlinesize], temp8, xchg);
2374 XCHG(*(uint64_t*)(h->top_border[s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
2375 XCHG(*(uint64_t*)(h->top_border[s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
2380 static void hl_decode_mb(H264Context *h){
2381 MpegEncContext * const s = &h->s;
2382 const int mb_x= s->mb_x;
2383 const int mb_y= s->mb_y;
2384 const int mb_xy= mb_x + mb_y*s->mb_stride;
2385 const int mb_type= s->current_picture.mb_type[mb_xy];
2386 uint8_t *dest_y, *dest_cb, *dest_cr;
2387 int linesize, uvlinesize /*dct_offset*/;
2396 dest_y = s->current_picture.data[0] + (mb_y * 16* s->linesize ) + mb_x * 16;
2397 dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2398 dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2400 if (h->mb_field_decoding_flag) {
2401 linesize = s->linesize * 2;
2402 uvlinesize = s->uvlinesize * 2;
2403 if(mb_y&1){ //FIXME move out of this func?
2404 dest_y -= s->linesize*15;
2405 dest_cb-= s->linesize*7;
2406 dest_cr-= s->linesize*7;
2409 linesize = s->linesize;
2410 uvlinesize = s->uvlinesize;
2411 // dct_offset = s->linesize * 16;
2414 if(IS_INTRA(mb_type)){
2415 if(h->deblocking_filter)
2416 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1);
2418 if(!(s->flags&CODEC_FLAG_GRAY)){
2419 h->pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
2420 h->pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
2423 if(IS_INTRA4x4(mb_type)){
2425 for(i=0; i<16; i++){
2426 uint8_t * const ptr= dest_y + h->block_offset[i];
2427 uint8_t *topright= ptr + 4 - linesize;
2428 const int topright_avail= (h->topright_samples_available<<i)&0x8000;
2429 const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
2432 if(!topright_avail){
2433 tr= ptr[3 - linesize]*0x01010101;
2434 topright= (uint8_t*) &tr;
2435 }else if(i==5 && h->deblocking_filter){
2436 tr= *(uint32_t*)h->top_border[mb_x+1];
2437 topright= (uint8_t*) &tr;
2440 h->pred4x4[ dir ](ptr, topright, linesize);
2441 if(h->non_zero_count_cache[ scan8[i] ]){
2442 if(s->codec_id == CODEC_ID_H264)
2443 h264_add_idct_c(ptr, h->mb + i*16, linesize);
2445 svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
2450 h->pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
2451 if(s->codec_id == CODEC_ID_H264)
2452 h264_luma_dc_dequant_idct_c(h->mb, s->qscale);
2454 svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
2456 if(h->deblocking_filter)
2457 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);
2458 }else if(s->codec_id == CODEC_ID_H264){
2459 hl_motion(h, dest_y, dest_cb, dest_cr,
2460 s->dsp.put_h264_qpel_pixels_tab, s->dsp.put_h264_chroma_pixels_tab,
2461 s->dsp.avg_h264_qpel_pixels_tab, s->dsp.avg_h264_chroma_pixels_tab);
2465 if(!IS_INTRA4x4(mb_type)){
2466 if(s->codec_id == CODEC_ID_H264){
2467 for(i=0; i<16; i++){
2468 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
2469 uint8_t * const ptr= dest_y + h->block_offset[i];
2470 h264_add_idct_c(ptr, h->mb + i*16, linesize);
2474 for(i=0; i<16; i++){
2475 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
2476 uint8_t * const ptr= dest_y + h->block_offset[i];
2477 svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
2483 if(!(s->flags&CODEC_FLAG_GRAY)){
2484 chroma_dc_dequant_idct_c(h->mb + 16*16, h->chroma_qp);
2485 chroma_dc_dequant_idct_c(h->mb + 16*16+4*16, h->chroma_qp);
2486 if(s->codec_id == CODEC_ID_H264){
2487 for(i=16; i<16+4; i++){
2488 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2489 uint8_t * const ptr= dest_cb + h->block_offset[i];
2490 h264_add_idct_c(ptr, h->mb + i*16, uvlinesize);
2493 for(i=20; i<20+4; i++){
2494 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2495 uint8_t * const ptr= dest_cr + h->block_offset[i];
2496 h264_add_idct_c(ptr, h->mb + i*16, uvlinesize);
2500 for(i=16; i<16+4; i++){
2501 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2502 uint8_t * const ptr= dest_cb + h->block_offset[i];
2503 svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
2506 for(i=20; i<20+4; i++){
2507 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2508 uint8_t * const ptr= dest_cr + h->block_offset[i];
2509 svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
2514 if(h->deblocking_filter) {
2515 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2516 filter_mb(h, mb_x, mb_y, dest_y, dest_cb, dest_cr);
2521 * fills the default_ref_list.
2523 static int fill_default_ref_list(H264Context *h){
2524 MpegEncContext * const s = &h->s;
2526 Picture sorted_short_ref[16];
2528 if(h->slice_type==B_TYPE){
2532 for(out_i=0; out_i<h->short_ref_count; out_i++){
2534 int best_poc=INT_MAX;
2536 for(i=0; i<h->short_ref_count; i++){
2537 const int poc= h->short_ref[i]->poc;
2538 if(poc > limit && poc < best_poc){
2544 assert(best_i != -1);
2547 sorted_short_ref[out_i]= *h->short_ref[best_i];
2551 if(s->picture_structure == PICT_FRAME){
2552 if(h->slice_type==B_TYPE){
2553 const int current_poc= s->current_picture_ptr->poc;
2556 for(list=0; list<2; list++){
2559 for(i=0; i<h->short_ref_count && index < h->ref_count[list]; i++){
2560 const int i2= list ? h->short_ref_count - i - 1 : i;
2561 const int poc= sorted_short_ref[i2].poc;
2563 if(sorted_short_ref[i2].reference != 3) continue; //FIXME refernce field shit
2565 if((list==1 && poc > current_poc) || (list==0 && poc < current_poc)){
2566 h->default_ref_list[list][index ]= sorted_short_ref[i2];
2567 h->default_ref_list[list][index++].pic_id= sorted_short_ref[i2].frame_num;
2571 for(i=0; i<h->long_ref_count && index < h->ref_count[ list ]; i++){
2572 if(h->long_ref[i]->reference != 3) continue;
2574 h->default_ref_list[ list ][index ]= *h->long_ref[i];
2575 h->default_ref_list[ list ][index++].pic_id= i;;
2578 if(h->long_ref_count > 1 && h->short_ref_count==0){
2579 Picture temp= h->default_ref_list[1][0];
2580 h->default_ref_list[1][0] = h->default_ref_list[1][1];
2581 h->default_ref_list[1][0] = temp;
2584 if(index < h->ref_count[ list ])
2585 memset(&h->default_ref_list[list][index], 0, sizeof(Picture)*(h->ref_count[ list ] - index));
2589 for(i=0; i<h->short_ref_count && index < h->ref_count[0]; i++){
2590 if(h->short_ref[i]->reference != 3) continue; //FIXME refernce field shit
2591 h->default_ref_list[0][index ]= *h->short_ref[i];
2592 h->default_ref_list[0][index++].pic_id= h->short_ref[i]->frame_num;
2594 for(i=0; i<h->long_ref_count && index < h->ref_count[0]; i++){
2595 if(h->long_ref[i]->reference != 3) continue;
2596 h->default_ref_list[0][index ]= *h->long_ref[i];
2597 h->default_ref_list[0][index++].pic_id= i;;
2599 if(index < h->ref_count[0])
2600 memset(&h->default_ref_list[0][index], 0, sizeof(Picture)*(h->ref_count[0] - index));
2603 if(h->slice_type==B_TYPE){
2605 //FIXME second field balh
2611 static int decode_ref_pic_list_reordering(H264Context *h){
2612 MpegEncContext * const s = &h->s;
2615 if(h->slice_type==I_TYPE || h->slice_type==SI_TYPE) return 0; //FIXME move beofre func
2617 for(list=0; list<2; list++){
2618 memcpy(h->ref_list[list], h->default_ref_list[list], sizeof(Picture)*h->ref_count[list]);
2620 if(get_bits1(&s->gb)){
2621 int pred= h->curr_pic_num;
2624 for(index=0; ; index++){
2625 int reordering_of_pic_nums_idc= get_ue_golomb(&s->gb);
2630 if(index >= h->ref_count[list]){
2631 av_log(h->s.avctx, AV_LOG_ERROR, "reference count overflow\n");
2635 if(reordering_of_pic_nums_idc<3){
2636 if(reordering_of_pic_nums_idc<2){
2637 const int abs_diff_pic_num= get_ue_golomb(&s->gb) + 1;
2639 if(abs_diff_pic_num >= h->max_pic_num){
2640 av_log(h->s.avctx, AV_LOG_ERROR, "abs_diff_pic_num overflow\n");
2644 if(reordering_of_pic_nums_idc == 0) pred-= abs_diff_pic_num;
2645 else pred+= abs_diff_pic_num;
2646 pred &= h->max_pic_num - 1;
2648 for(i= h->ref_count[list]-1; i>=index; i--){
2649 if(h->ref_list[list][i].pic_id == pred && h->ref_list[list][i].long_ref==0)
2653 pic_id= get_ue_golomb(&s->gb); //long_term_pic_idx
2655 for(i= h->ref_count[list]-1; i>=index; i--){
2656 if(h->ref_list[list][i].pic_id == pic_id && h->ref_list[list][i].long_ref==1)
2662 av_log(h->s.avctx, AV_LOG_ERROR, "reference picture missing during reorder\n");
2663 memset(&h->ref_list[list][index], 0, sizeof(Picture)); //FIXME
2664 }else if(i > index){
2665 Picture tmp= h->ref_list[list][i];
2666 for(; i>index; i--){
2667 h->ref_list[list][i]= h->ref_list[list][i-1];
2669 h->ref_list[list][index]= tmp;
2671 }else if(reordering_of_pic_nums_idc==3)
2674 av_log(h->s.avctx, AV_LOG_ERROR, "illegal reordering_of_pic_nums_idc\n");
2680 if(h->slice_type!=B_TYPE) break;
2685 static int pred_weight_table(H264Context *h){
2686 MpegEncContext * const s = &h->s;
2689 h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
2690 h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
2692 for(list=0; list<2; list++){
2693 for(i=0; i<h->ref_count[list]; i++){
2694 int luma_weight_flag, chroma_weight_flag;
2696 luma_weight_flag= get_bits1(&s->gb);
2697 if(luma_weight_flag){
2698 h->luma_weight[list][i]= get_se_golomb(&s->gb);
2699 h->luma_offset[list][i]= get_se_golomb(&s->gb);
2702 chroma_weight_flag= get_bits1(&s->gb);
2703 if(chroma_weight_flag){
2706 h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);
2707 h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);
2711 if(h->slice_type != B_TYPE) break;
2717 * instantaneos decoder refresh.
2719 static void idr(H264Context *h){
2722 for(i=0; i<h->long_ref_count; i++){
2723 h->long_ref[i]->reference=0;
2724 h->long_ref[i]= NULL;
2726 h->long_ref_count=0;
2728 for(i=0; i<h->short_ref_count; i++){
2729 h->short_ref[i]->reference=0;
2730 h->short_ref[i]= NULL;
2732 h->short_ref_count=0;
2737 * @return the removed picture or NULL if an error occures
2739 static Picture * remove_short(H264Context *h, int frame_num){
2740 MpegEncContext * const s = &h->s;
2743 if(s->avctx->debug&FF_DEBUG_MMCO)
2744 av_log(h->s.avctx, AV_LOG_DEBUG, "remove short %d count %d\n", frame_num, h->short_ref_count);
2746 for(i=0; i<h->short_ref_count; i++){
2747 Picture *pic= h->short_ref[i];
2748 if(s->avctx->debug&FF_DEBUG_MMCO)
2749 av_log(h->s.avctx, AV_LOG_DEBUG, "%d %d %p\n", i, pic->frame_num, pic);
2750 if(pic->frame_num == frame_num){
2751 h->short_ref[i]= NULL;
2752 memmove(&h->short_ref[i], &h->short_ref[i+1], (h->short_ref_count - i - 1)*sizeof(Picture*));
2753 h->short_ref_count--;
2762 * @return the removed picture or NULL if an error occures
2764 static Picture * remove_long(H264Context *h, int i){
2767 if(i >= h->long_ref_count) return NULL;
2768 pic= h->long_ref[i];
2769 if(pic==NULL) return NULL;
2771 h->long_ref[i]= NULL;
2772 memmove(&h->long_ref[i], &h->long_ref[i+1], (h->long_ref_count - i - 1)*sizeof(Picture*));
2773 h->long_ref_count--;
2779 * Executes the reference picture marking (memory management control operations).
2781 static int execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count){
2782 MpegEncContext * const s = &h->s;
2784 int current_is_long=0;
2787 if((s->avctx->debug&FF_DEBUG_MMCO) && mmco_count==0)
2788 av_log(h->s.avctx, AV_LOG_DEBUG, "no mmco here\n");
2790 for(i=0; i<mmco_count; i++){
2791 if(s->avctx->debug&FF_DEBUG_MMCO)
2792 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);
2794 switch(mmco[i].opcode){
2795 case MMCO_SHORT2UNUSED:
2796 pic= remove_short(h, mmco[i].short_frame_num);
2797 if(pic==NULL) return -1;
2800 case MMCO_SHORT2LONG:
2801 pic= remove_long(h, mmco[i].long_index);
2802 if(pic) pic->reference=0;
2804 h->long_ref[ mmco[i].long_index ]= remove_short(h, mmco[i].short_frame_num);
2805 h->long_ref[ mmco[i].long_index ]->long_ref=1;
2807 case MMCO_LONG2UNUSED:
2808 pic= remove_long(h, mmco[i].long_index);
2809 if(pic==NULL) return -1;
2813 pic= remove_long(h, mmco[i].long_index);
2814 if(pic) pic->reference=0;
2816 h->long_ref[ mmco[i].long_index ]= s->current_picture_ptr;
2817 h->long_ref[ mmco[i].long_index ]->long_ref=1;
2818 h->long_ref_count++;
2822 case MMCO_SET_MAX_LONG:
2823 assert(mmco[i].long_index <= 16);
2824 while(mmco[i].long_index < h->long_ref_count){
2825 pic= remove_long(h, mmco[i].long_index);
2828 while(mmco[i].long_index > h->long_ref_count){
2829 h->long_ref[ h->long_ref_count++ ]= NULL;
2833 while(h->short_ref_count){
2834 pic= remove_short(h, h->short_ref[0]->frame_num);
2837 while(h->long_ref_count){
2838 pic= remove_long(h, h->long_ref_count-1);
2846 if(!current_is_long){
2847 pic= remove_short(h, s->current_picture_ptr->frame_num);
2850 av_log(h->s.avctx, AV_LOG_ERROR, "illegal short term buffer state detected\n");
2853 if(h->short_ref_count)
2854 memmove(&h->short_ref[1], &h->short_ref[0], h->short_ref_count*sizeof(Picture*));
2856 h->short_ref[0]= s->current_picture_ptr;
2857 h->short_ref[0]->long_ref=0;
2858 h->short_ref_count++;
2864 static int decode_ref_pic_marking(H264Context *h){
2865 MpegEncContext * const s = &h->s;
2868 if(h->nal_unit_type == NAL_IDR_SLICE){ //FIXME fields
2869 s->broken_link= get_bits1(&s->gb) -1;
2870 h->mmco[0].long_index= get_bits1(&s->gb) - 1; // current_long_term_idx
2871 if(h->mmco[0].long_index == -1)
2874 h->mmco[0].opcode= MMCO_LONG;
2878 if(get_bits1(&s->gb)){ // adaptive_ref_pic_marking_mode_flag
2879 for(i= h->mmco_index; i<MAX_MMCO_COUNT; i++) {
2880 MMCOOpcode opcode= get_ue_golomb(&s->gb);;
2882 h->mmco[i].opcode= opcode;
2883 if(opcode==MMCO_SHORT2UNUSED || opcode==MMCO_SHORT2LONG){
2884 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
2885 /* if(h->mmco[i].short_frame_num >= h->short_ref_count || h->short_ref[ h->mmco[i].short_frame_num ] == NULL){
2886 fprintf(stderr, "illegal short ref in memory management control operation %d\n", mmco);
2890 if(opcode==MMCO_SHORT2LONG || opcode==MMCO_LONG2UNUSED || opcode==MMCO_LONG || opcode==MMCO_SET_MAX_LONG){
2891 h->mmco[i].long_index= get_ue_golomb(&s->gb);
2892 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){
2893 av_log(h->s.avctx, AV_LOG_ERROR, "illegal long ref in memory management control operation %d\n", opcode);
2898 if(opcode > MMCO_LONG){
2899 av_log(h->s.avctx, AV_LOG_ERROR, "illegal memory management control operation %d\n", opcode);
2902 if(opcode == MMCO_END)
2907 assert(h->long_ref_count + h->short_ref_count <= h->sps.ref_frame_count);
2909 if(h->long_ref_count + h->short_ref_count == h->sps.ref_frame_count){ //FIXME fields
2910 h->mmco[0].opcode= MMCO_SHORT2UNUSED;
2911 h->mmco[0].short_frame_num= h->short_ref[ h->short_ref_count - 1 ]->frame_num;
2921 static int init_poc(H264Context *h){
2922 MpegEncContext * const s = &h->s;
2923 const int max_frame_num= 1<<h->sps.log2_max_frame_num;
2926 if(h->nal_unit_type == NAL_IDR_SLICE){
2927 h->frame_num_offset= 0;
2929 if(h->frame_num < h->prev_frame_num)
2930 h->frame_num_offset= h->prev_frame_num_offset + max_frame_num;
2932 h->frame_num_offset= h->prev_frame_num_offset;
2935 if(h->sps.poc_type==0){
2936 const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
2938 if (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
2939 h->poc_msb = h->prev_poc_msb + max_poc_lsb;
2940 else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
2941 h->poc_msb = h->prev_poc_msb - max_poc_lsb;
2943 h->poc_msb = h->prev_poc_msb;
2944 //printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
2946 field_poc[1] = h->poc_msb + h->poc_lsb;
2947 if(s->picture_structure == PICT_FRAME)
2948 field_poc[1] += h->delta_poc_bottom;
2949 }else if(h->sps.poc_type==1){
2950 int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
2953 if(h->sps.poc_cycle_length != 0)
2954 abs_frame_num = h->frame_num_offset + h->frame_num;
2958 if(h->nal_ref_idc==0 && abs_frame_num > 0)
2961 expected_delta_per_poc_cycle = 0;
2962 for(i=0; i < h->sps.poc_cycle_length; i++)
2963 expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
2965 if(abs_frame_num > 0){
2966 int poc_cycle_cnt = (abs_frame_num - 1) / h->sps.poc_cycle_length;
2967 int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
2969 expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
2970 for(i = 0; i <= frame_num_in_poc_cycle; i++)
2971 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
2975 if(h->nal_ref_idc == 0)
2976 expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
2978 field_poc[0] = expectedpoc + h->delta_poc[0];
2979 field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
2981 if(s->picture_structure == PICT_FRAME)
2982 field_poc[1] += h->delta_poc[1];
2985 if(h->nal_unit_type == NAL_IDR_SLICE){
2988 if(h->nal_ref_idc) poc= 2*(h->frame_num_offset + h->frame_num);
2989 else poc= 2*(h->frame_num_offset + h->frame_num) - 1;
2995 if(s->picture_structure != PICT_BOTTOM_FIELD)
2996 s->current_picture_ptr->field_poc[0]= field_poc[0];
2997 if(s->picture_structure != PICT_TOP_FIELD)
2998 s->current_picture_ptr->field_poc[1]= field_poc[1];
2999 if(s->picture_structure == PICT_FRAME) // FIXME field pix?
3000 s->current_picture_ptr->poc= FFMIN(field_poc[0], field_poc[1]);
3006 * decodes a slice header.
3007 * this will allso call MPV_common_init() and frame_start() as needed
3009 static int decode_slice_header(H264Context *h){
3010 MpegEncContext * const s = &h->s;
3011 int first_mb_in_slice, pps_id;
3012 int num_ref_idx_active_override_flag;
3013 static const uint8_t slice_type_map[5]= {P_TYPE, B_TYPE, I_TYPE, SP_TYPE, SI_TYPE};
3015 s->current_picture.reference= h->nal_ref_idc != 0;
3017 first_mb_in_slice= get_ue_golomb(&s->gb);
3019 h->slice_type= get_ue_golomb(&s->gb);
3020 if(h->slice_type > 9){
3021 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);
3023 if(h->slice_type > 4){
3025 h->slice_type_fixed=1;
3027 h->slice_type_fixed=0;
3029 h->slice_type= slice_type_map[ h->slice_type ];
3031 s->pict_type= h->slice_type; // to make a few old func happy, its wrong though
3033 pps_id= get_ue_golomb(&s->gb);
3035 av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
3038 h->pps= h->pps_buffer[pps_id];
3039 if(h->pps.slice_group_count == 0){
3040 av_log(h->s.avctx, AV_LOG_ERROR, "non existing PPS referenced\n");
3044 h->sps= h->sps_buffer[ h->pps.sps_id ];
3045 if(h->sps.log2_max_frame_num == 0){
3046 av_log(h->s.avctx, AV_LOG_ERROR, "non existing SPS referenced\n");
3050 s->mb_width= h->sps.mb_width;
3051 s->mb_height= h->sps.mb_height;
3053 h->b_stride= s->mb_width*4;
3054 h->b8_stride= s->mb_width*2;
3056 s->mb_x = first_mb_in_slice % s->mb_width;
3057 s->mb_y = first_mb_in_slice / s->mb_width; //FIXME AFFW
3059 s->width = 16*s->mb_width - 2*(h->sps.crop_left + h->sps.crop_right );
3060 if(h->sps.frame_mbs_only_flag)
3061 s->height= 16*s->mb_height - 2*(h->sps.crop_top + h->sps.crop_bottom);
3063 s->height= 16*s->mb_height - 4*(h->sps.crop_top + h->sps.crop_bottom); //FIXME recheck
3065 if (s->context_initialized
3066 && ( s->width != s->avctx->width || s->height != s->avctx->height)) {
3070 if (!s->context_initialized) {
3071 if (MPV_common_init(s) < 0)
3076 s->avctx->width = s->width;
3077 s->avctx->height = s->height;
3078 s->avctx->sample_aspect_ratio= h->sps.sar;
3080 if(h->sps.timing_info_present_flag && h->sps.fixed_frame_rate_flag){
3081 s->avctx->frame_rate = h->sps.time_scale;
3082 s->avctx->frame_rate_base = h->sps.num_units_in_tick;
3086 if(first_mb_in_slice == 0){
3090 s->current_picture_ptr->frame_num= //FIXME frame_num cleanup
3091 h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
3093 if(h->sps.frame_mbs_only_flag){
3094 s->picture_structure= PICT_FRAME;
3096 if(get_bits1(&s->gb)) //field_pic_flag
3097 s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
3099 s->picture_structure= PICT_FRAME;
3102 if(s->picture_structure==PICT_FRAME){
3103 h->curr_pic_num= h->frame_num;
3104 h->max_pic_num= 1<< h->sps.log2_max_frame_num;
3106 h->curr_pic_num= 2*h->frame_num;
3107 h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
3110 if(h->nal_unit_type == NAL_IDR_SLICE){
3111 get_ue_golomb(&s->gb); /* idr_pic_id */
3114 if(h->sps.poc_type==0){
3115 h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
3117 if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
3118 h->delta_poc_bottom= get_se_golomb(&s->gb);
3122 if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
3123 h->delta_poc[0]= get_se_golomb(&s->gb);
3125 if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
3126 h->delta_poc[1]= get_se_golomb(&s->gb);
3131 if(h->pps.redundant_pic_cnt_present){
3132 h->redundant_pic_count= get_ue_golomb(&s->gb);
3135 //set defaults, might be overriden a few line later
3136 h->ref_count[0]= h->pps.ref_count[0];
3137 h->ref_count[1]= h->pps.ref_count[1];
3139 if(h->slice_type == P_TYPE || h->slice_type == SP_TYPE || h->slice_type == B_TYPE){
3140 if(h->slice_type == B_TYPE){
3141 h->direct_spatial_mv_pred= get_bits1(&s->gb);
3143 num_ref_idx_active_override_flag= get_bits1(&s->gb);
3145 if(num_ref_idx_active_override_flag){
3146 h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
3147 if(h->slice_type==B_TYPE)
3148 h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
3150 if(h->ref_count[0] > 32 || h->ref_count[1] > 32){
3151 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
3157 if(first_mb_in_slice == 0){
3158 fill_default_ref_list(h);
3161 decode_ref_pic_list_reordering(h);
3163 if( (h->pps.weighted_pred && (h->slice_type == P_TYPE || h->slice_type == SP_TYPE ))
3164 || (h->pps.weighted_bipred_idc==1 && h->slice_type==B_TYPE ) )
3165 pred_weight_table(h);
3167 if(s->current_picture.reference)
3168 decode_ref_pic_marking(h);
3170 if( h->slice_type != I_TYPE && h->slice_type != SI_TYPE && h->pps.cabac )
3171 h->cabac_init_idc = get_ue_golomb(&s->gb);
3173 h->last_qscale_diff = 0;
3174 s->qscale = h->pps.init_qp + get_se_golomb(&s->gb);
3175 if(s->qscale<0 || s->qscale>51){
3176 av_log(s->avctx, AV_LOG_ERROR, "QP %d out of range\n", s->qscale);
3179 //FIXME qscale / qp ... stuff
3180 if(h->slice_type == SP_TYPE){
3181 get_bits1(&s->gb); /* sp_for_switch_flag */
3183 if(h->slice_type==SP_TYPE || h->slice_type == SI_TYPE){
3184 get_se_golomb(&s->gb); /* slice_qs_delta */
3187 h->deblocking_filter = 1;
3188 h->slice_alpha_c0_offset = 0;
3189 h->slice_beta_offset = 0;
3190 if( h->pps.deblocking_filter_parameters_present ) {
3191 h->deblocking_filter= get_ue_golomb(&s->gb);
3192 if(h->deblocking_filter < 2)
3193 h->deblocking_filter^= 1; // 1<->0
3195 if( h->deblocking_filter ) {
3196 h->slice_alpha_c0_offset = get_se_golomb(&s->gb) << 1;
3197 h->slice_beta_offset = get_se_golomb(&s->gb) << 1;
3202 if( h->pps.num_slice_groups > 1 && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
3203 slice_group_change_cycle= get_bits(&s->gb, ?);
3206 if(s->avctx->debug&FF_DEBUG_PICT_INFO){
3207 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",
3209 av_get_pict_type_char(h->slice_type),
3210 pps_id, h->frame_num,
3211 s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
3212 h->ref_count[0], h->ref_count[1],
3214 h->deblocking_filter
3224 static inline int get_level_prefix(GetBitContext *gb){
3228 OPEN_READER(re, gb);
3229 UPDATE_CACHE(re, gb);
3230 buf=GET_CACHE(re, gb);
3232 log= 32 - av_log2(buf);
3234 print_bin(buf>>(32-log), log);
3235 printf("%5d %2d %3d lpr @%5d in %s get_level_prefix\n", buf>>(32-log), log, log-1, get_bits_count(gb), __FILE__);
3238 LAST_SKIP_BITS(re, gb, log);
3239 CLOSE_READER(re, gb);
3245 * decodes a residual block.
3246 * @param n block index
3247 * @param scantable scantable
3248 * @param max_coeff number of coefficients in the block
3249 * @return <0 if an error occured
3251 static int decode_residual(H264Context *h, GetBitContext *gb, DCTELEM *block, int n, const uint8_t *scantable, int qp, int max_coeff){
3252 MpegEncContext * const s = &h->s;
3253 const uint16_t *qmul= dequant_coeff[qp];
3254 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};
3255 int level[16], run[16];
3256 int suffix_length, zeros_left, coeff_num, coeff_token, total_coeff, i, trailing_ones;
3258 //FIXME put trailing_onex into the context
3260 if(n == CHROMA_DC_BLOCK_INDEX){
3261 coeff_token= get_vlc2(gb, chroma_dc_coeff_token_vlc.table, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 1);
3262 total_coeff= coeff_token>>2;
3264 if(n == LUMA_DC_BLOCK_INDEX){
3265 total_coeff= pred_non_zero_count(h, 0);
3266 coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
3267 total_coeff= coeff_token>>2;
3269 total_coeff= pred_non_zero_count(h, n);
3270 coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
3271 total_coeff= coeff_token>>2;
3272 h->non_zero_count_cache[ scan8[n] ]= total_coeff;
3276 //FIXME set last_non_zero?
3281 trailing_ones= coeff_token&3;
3282 tprintf("trailing:%d, total:%d\n", trailing_ones, total_coeff);
3283 assert(total_coeff<=16);
3285 for(i=0; i<trailing_ones; i++){
3286 level[i]= 1 - 2*get_bits1(gb);
3289 suffix_length= total_coeff > 10 && trailing_ones < 3;
3291 for(; i<total_coeff; i++){
3292 const int prefix= get_level_prefix(gb);
3293 int level_code, mask;
3295 if(prefix<14){ //FIXME try to build a large unified VLC table for all this
3297 level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
3299 level_code= (prefix<<suffix_length); //part
3300 }else if(prefix==14){
3302 level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
3304 level_code= prefix + get_bits(gb, 4); //part
3305 }else if(prefix==15){
3306 level_code= (prefix<<suffix_length) + get_bits(gb, 12); //part
3307 if(suffix_length==0) level_code+=15; //FIXME doesnt make (much)sense
3309 av_log(h->s.avctx, AV_LOG_ERROR, "prefix too large at %d %d\n", s->mb_x, s->mb_y);
3313 if(i==trailing_ones && i<3) level_code+= 2; //FIXME split first iteration
3315 mask= -(level_code&1);
3316 level[i]= (((2+level_code)>>1) ^ mask) - mask;
3318 if(suffix_length==0) suffix_length=1; //FIXME split first iteration
3321 if(ABS(level[i]) > (3<<(suffix_length-1)) && suffix_length<6) suffix_length++;
3323 if((2+level_code)>>1) > (3<<(suffix_length-1)) && suffix_length<6) suffix_length++;
3324 /* ? == prefix > 2 or sth */
3326 tprintf("level: %d suffix_length:%d\n", level[i], suffix_length);
3329 if(total_coeff == max_coeff)
3332 if(n == CHROMA_DC_BLOCK_INDEX)
3333 zeros_left= get_vlc2(gb, chroma_dc_total_zeros_vlc[ total_coeff-1 ].table, CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 1);
3335 zeros_left= get_vlc2(gb, total_zeros_vlc[ total_coeff-1 ].table, TOTAL_ZEROS_VLC_BITS, 1);
3338 for(i=0; i<total_coeff-1; i++){
3341 else if(zeros_left < 7){
3342 run[i]= get_vlc2(gb, run_vlc[zeros_left-1].table, RUN_VLC_BITS, 1);
3344 run[i]= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2);
3346 zeros_left -= run[i];
3350 av_log(h->s.avctx, AV_LOG_ERROR, "negative number of zero coeffs at %d %d\n", s->mb_x, s->mb_y);
3354 for(; i<total_coeff-1; i++){
3362 for(i=total_coeff-1; i>=0; i--){ //FIXME merge into rundecode?
3365 coeff_num += run[i] + 1; //FIXME add 1 earlier ?
3366 j= scantable[ coeff_num ];
3371 for(i=total_coeff-1; i>=0; i--){ //FIXME merge into rundecode?
3374 coeff_num += run[i] + 1; //FIXME add 1 earlier ?
3375 j= scantable[ coeff_num ];
3377 block[j]= level[i] * qmul[j];
3378 // printf("%d %d ", block[j], qmul[j]);
3385 * decodes a macroblock
3386 * @returns 0 if ok, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
3388 static int decode_mb_cavlc(H264Context *h){
3389 MpegEncContext * const s = &h->s;
3390 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
3391 int mb_type, partition_count, cbp;
3393 s->dsp.clear_blocks(h->mb); //FIXME avoid if allready clear (move after skip handlong?
3395 tprintf("pic:%d mb:%d/%d\n", h->frame_num, s->mb_x, s->mb_y);
3396 cbp = 0; /* avoid warning. FIXME: find a solution without slowing
3398 if(h->slice_type != I_TYPE && h->slice_type != SI_TYPE){
3399 if(s->mb_skip_run==-1)
3400 s->mb_skip_run= get_ue_golomb(&s->gb);
3402 if (s->mb_skip_run--) {
3406 mb_type= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0;
3408 memset(h->non_zero_count[mb_xy], 0, 16);
3409 memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui
3411 if(h->sps.mb_aff && s->mb_skip_run==0 && (s->mb_y&1)==0){
3412 h->mb_field_decoding_flag= get_bits1(&s->gb);
3415 if(h->mb_field_decoding_flag)
3416 mb_type|= MB_TYPE_INTERLACED;
3418 fill_caches(h, mb_type); //FIXME check what is needed and what not ...
3419 pred_pskip_motion(h, &mx, &my);
3420 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
3421 fill_rectangle( h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4);
3422 write_back_motion(h, mb_type);
3424 s->current_picture.mb_type[mb_xy]= mb_type; //FIXME SKIP type
3425 s->current_picture.qscale_table[mb_xy]= s->qscale;
3426 h->slice_table[ mb_xy ]= h->slice_num;
3428 h->prev_mb_skiped= 1;
3432 if(h->sps.mb_aff /* && !field pic FIXME needed? */){
3434 h->mb_field_decoding_flag = get_bits1(&s->gb);
3436 h->mb_field_decoding_flag=0; //FIXME som ed note ?!
3438 h->prev_mb_skiped= 0;
3440 mb_type= get_ue_golomb(&s->gb);
3441 if(h->slice_type == B_TYPE){
3443 partition_count= b_mb_type_info[mb_type].partition_count;
3444 mb_type= b_mb_type_info[mb_type].type;
3447 goto decode_intra_mb;
3449 }else if(h->slice_type == P_TYPE /*|| h->slice_type == SP_TYPE */){
3451 partition_count= p_mb_type_info[mb_type].partition_count;
3452 mb_type= p_mb_type_info[mb_type].type;
3455 goto decode_intra_mb;
3458 assert(h->slice_type == I_TYPE);
3461 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);
3465 cbp= i_mb_type_info[mb_type].cbp;
3466 h->intra16x16_pred_mode= i_mb_type_info[mb_type].pred_mode;
3467 mb_type= i_mb_type_info[mb_type].type;
3470 if(h->mb_field_decoding_flag)
3471 mb_type |= MB_TYPE_INTERLACED;
3473 s->current_picture.mb_type[mb_xy]= mb_type;
3474 h->slice_table[ mb_xy ]= h->slice_num;
3476 if(IS_INTRA_PCM(mb_type)){
3480 // we assume these blocks are very rare so we dont optimize it
3481 align_get_bits(&s->gb);
3483 ptr= s->gb.buffer + get_bits_count(&s->gb);
3485 for(y=0; y<16; y++){
3486 const int index= 4*(y&3) + 64*(y>>2);
3487 for(x=0; x<16; x++){
3488 h->mb[index + (x&3) + 16*(x>>2)]= *(ptr++);
3492 const int index= 256 + 4*(y&3) + 32*(y>>2);
3494 h->mb[index + (x&3) + 16*(x>>2)]= *(ptr++);
3498 const int index= 256 + 64 + 4*(y&3) + 32*(y>>2);
3500 h->mb[index + (x&3) + 16*(x>>2)]= *(ptr++);
3504 skip_bits(&s->gb, 384); //FIXME check /fix the bitstream readers
3506 //FIXME deblock filter, non_zero_count_cache init ...
3507 memset(h->non_zero_count[mb_xy], 16, 16);
3508 s->current_picture.qscale_table[mb_xy]= s->qscale;
3513 fill_caches(h, mb_type);
3516 if(IS_INTRA(mb_type)){
3517 // init_top_left_availability(h);
3518 if(IS_INTRA4x4(mb_type)){
3521 // fill_intra4x4_pred_table(h);
3522 for(i=0; i<16; i++){
3523 const int mode_coded= !get_bits1(&s->gb);
3524 const int predicted_mode= pred_intra_mode(h, i);
3528 const int rem_mode= get_bits(&s->gb, 3);
3529 if(rem_mode<predicted_mode)
3534 mode= predicted_mode;
3537 h->intra4x4_pred_mode_cache[ scan8[i] ] = mode;
3539 write_back_intra_pred_mode(h);
3540 if( check_intra4x4_pred_mode(h) < 0)
3543 h->intra16x16_pred_mode= check_intra_pred_mode(h, h->intra16x16_pred_mode);
3544 if(h->intra16x16_pred_mode < 0)
3547 h->chroma_pred_mode= get_ue_golomb(&s->gb);
3549 h->chroma_pred_mode= check_intra_pred_mode(h, h->chroma_pred_mode);
3550 if(h->chroma_pred_mode < 0)
3552 }else if(partition_count==4){
3553 int i, j, sub_partition_count[4], list, ref[2][4];
3555 if(h->slice_type == B_TYPE){
3557 h->sub_mb_type[i]= get_ue_golomb(&s->gb);
3558 if(h->sub_mb_type[i] >=13){
3559 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);
3562 sub_partition_count[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
3563 h->sub_mb_type[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].type;
3566 assert(h->slice_type == P_TYPE || h->slice_type == SP_TYPE); //FIXME SP correct ?
3568 h->sub_mb_type[i]= get_ue_golomb(&s->gb);
3569 if(h->sub_mb_type[i] >=4){
3570 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);
3573 sub_partition_count[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
3574 h->sub_mb_type[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].type;
3578 for(list=0; list<2; list++){
3579 const int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list];
3580 if(ref_count == 0) continue;
3582 if(IS_DIR(h->sub_mb_type[i], 0, list) && !IS_DIRECT(h->sub_mb_type[i])){
3583 ref[list][i] = get_te0_golomb(&s->gb, ref_count); //FIXME init to 0 before and skip?
3591 for(list=0; list<2; list++){
3592 const int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list];
3593 if(ref_count == 0) continue;
3596 h->ref_cache[list][ scan8[4*i] ]=h->ref_cache[list][ scan8[4*i]+1 ]=
3597 h->ref_cache[list][ scan8[4*i]+8 ]=h->ref_cache[list][ scan8[4*i]+9 ]= ref[list][i];
3599 if(IS_DIR(h->sub_mb_type[i], 0, list) && !IS_DIRECT(h->sub_mb_type[i])){
3600 const int sub_mb_type= h->sub_mb_type[i];
3601 const int block_width= (sub_mb_type & (MB_TYPE_16x16|MB_TYPE_16x8)) ? 2 : 1;
3602 for(j=0; j<sub_partition_count[i]; j++){
3604 const int index= 4*i + block_width*j;
3605 int16_t (* mv_cache)[2]= &h->mv_cache[list][ scan8[index] ];
3606 pred_motion(h, index, block_width, list, h->ref_cache[list][ scan8[index] ], &mx, &my);
3607 mx += get_se_golomb(&s->gb);
3608 my += get_se_golomb(&s->gb);
3609 tprintf("final mv:%d %d\n", mx, my);
3611 if(IS_SUB_8X8(sub_mb_type)){
3612 mv_cache[ 0 ][0]= mv_cache[ 1 ][0]=
3613 mv_cache[ 8 ][0]= mv_cache[ 9 ][0]= mx;
3614 mv_cache[ 0 ][1]= mv_cache[ 1 ][1]=
3615 mv_cache[ 8 ][1]= mv_cache[ 9 ][1]= my;
3616 }else if(IS_SUB_8X4(sub_mb_type)){
3617 mv_cache[ 0 ][0]= mv_cache[ 1 ][0]= mx;
3618 mv_cache[ 0 ][1]= mv_cache[ 1 ][1]= my;
3619 }else if(IS_SUB_4X8(sub_mb_type)){
3620 mv_cache[ 0 ][0]= mv_cache[ 8 ][0]= mx;
3621 mv_cache[ 0 ][1]= mv_cache[ 8 ][1]= my;
3623 assert(IS_SUB_4X4(sub_mb_type));
3624 mv_cache[ 0 ][0]= mx;
3625 mv_cache[ 0 ][1]= my;
3629 uint32_t *p= (uint32_t *)&h->mv_cache[list][ scan8[4*i] ][0];
3635 }else if(!IS_DIRECT(mb_type)){
3636 int list, mx, my, i;
3637 //FIXME we should set ref_idx_l? to 0 if we use that later ...
3638 if(IS_16X16(mb_type)){
3639 for(list=0; list<2; list++){
3640 if(h->ref_count[0]>0){
3641 if(IS_DIR(mb_type, 0, list)){
3642 const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
3643 fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, val, 1);
3647 for(list=0; list<2; list++){
3648 if(IS_DIR(mb_type, 0, list)){
3649 pred_motion(h, 0, 4, list, h->ref_cache[list][ scan8[0] ], &mx, &my);
3650 mx += get_se_golomb(&s->gb);
3651 my += get_se_golomb(&s->gb);
3652 tprintf("final mv:%d %d\n", mx, my);
3654 fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, pack16to32(mx,my), 4);
3658 else if(IS_16X8(mb_type)){
3659 for(list=0; list<2; list++){
3660 if(h->ref_count[list]>0){
3662 if(IS_DIR(mb_type, i, list)){
3663 const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
3664 fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, val, 1);
3669 for(list=0; list<2; list++){
3671 if(IS_DIR(mb_type, i, list)){
3672 pred_16x8_motion(h, 8*i, list, h->ref_cache[list][scan8[0] + 16*i], &mx, &my);
3673 mx += get_se_golomb(&s->gb);
3674 my += get_se_golomb(&s->gb);
3675 tprintf("final mv:%d %d\n", mx, my);
3677 fill_rectangle(h->mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, pack16to32(mx,my), 4);
3682 assert(IS_8X16(mb_type));
3683 for(list=0; list<2; list++){
3684 if(h->ref_count[list]>0){
3686 if(IS_DIR(mb_type, i, list)){ //FIXME optimize
3687 const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
3688 fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, val, 1);
3693 for(list=0; list<2; list++){
3695 if(IS_DIR(mb_type, i, list)){
3696 pred_8x16_motion(h, i*4, list, h->ref_cache[list][ scan8[0] + 2*i ], &mx, &my);
3697 mx += get_se_golomb(&s->gb);
3698 my += get_se_golomb(&s->gb);
3699 tprintf("final mv:%d %d\n", mx, my);
3701 fill_rectangle(h->mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, pack16to32(mx,my), 4);
3708 if(IS_INTER(mb_type))
3709 write_back_motion(h, mb_type);
3711 if(!IS_INTRA16x16(mb_type)){
3712 cbp= get_ue_golomb(&s->gb);
3714 av_log(h->s.avctx, AV_LOG_ERROR, "cbp too large (%d) at %d %d\n", cbp, s->mb_x, s->mb_y);
3718 if(IS_INTRA4x4(mb_type))
3719 cbp= golomb_to_intra4x4_cbp[cbp];
3721 cbp= golomb_to_inter_cbp[cbp];
3724 if(cbp || IS_INTRA16x16(mb_type)){
3725 int i8x8, i4x4, chroma_idx;
3726 int chroma_qp, dquant;
3727 GetBitContext *gb= IS_INTRA(mb_type) ? h->intra_gb_ptr : h->inter_gb_ptr;
3728 const uint8_t *scan, *dc_scan;
3730 // fill_non_zero_count_cache(h);
3732 if(IS_INTERLACED(mb_type)){
3734 dc_scan= luma_dc_field_scan;
3737 dc_scan= luma_dc_zigzag_scan;
3740 dquant= get_se_golomb(&s->gb);
3742 if( dquant > 25 || dquant < -26 ){
3743 av_log(h->s.avctx, AV_LOG_ERROR, "dquant out of range (%d) at %d %d\n", dquant, s->mb_x, s->mb_y);
3747 s->qscale += dquant;
3748 if(((unsigned)s->qscale) > 51){
3749 if(s->qscale<0) s->qscale+= 52;
3750 else s->qscale-= 52;
3753 h->chroma_qp= chroma_qp= get_chroma_qp(h, s->qscale);
3754 if(IS_INTRA16x16(mb_type)){
3755 if( decode_residual(h, h->intra_gb_ptr, h->mb, LUMA_DC_BLOCK_INDEX, dc_scan, s->qscale, 16) < 0){
3756 return -1; //FIXME continue if partotioned and other retirn -1 too
3759 assert((cbp&15) == 0 || (cbp&15) == 15);
3762 for(i8x8=0; i8x8<4; i8x8++){
3763 for(i4x4=0; i4x4<4; i4x4++){
3764 const int index= i4x4 + 4*i8x8;
3765 if( decode_residual(h, h->intra_gb_ptr, h->mb + 16*index, index, scan + 1, s->qscale, 15) < 0 ){
3771 fill_rectangle(&h->non_zero_count_cache[scan8[0]], 4, 4, 8, 0, 1);
3774 for(i8x8=0; i8x8<4; i8x8++){
3775 if(cbp & (1<<i8x8)){
3776 for(i4x4=0; i4x4<4; i4x4++){
3777 const int index= i4x4 + 4*i8x8;
3779 if( decode_residual(h, gb, h->mb + 16*index, index, scan, s->qscale, 16) <0 ){
3784 uint8_t * const nnz= &h->non_zero_count_cache[ scan8[4*i8x8] ];
3785 nnz[0] = nnz[1] = nnz[8] = nnz[9] = 0;
3791 for(chroma_idx=0; chroma_idx<2; chroma_idx++)
3792 if( decode_residual(h, gb, h->mb + 256 + 16*4*chroma_idx, CHROMA_DC_BLOCK_INDEX, chroma_dc_scan, chroma_qp, 4) < 0){
3798 for(chroma_idx=0; chroma_idx<2; chroma_idx++){
3799 for(i4x4=0; i4x4<4; i4x4++){
3800 const int index= 16 + 4*chroma_idx + i4x4;
3801 if( decode_residual(h, gb, h->mb + 16*index, index, scan + 1, chroma_qp, 15) < 0){
3807 uint8_t * const nnz= &h->non_zero_count_cache[0];
3808 nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =
3809 nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
3812 uint8_t * const nnz= &h->non_zero_count_cache[0];
3813 fill_rectangle(&nnz[scan8[0]], 4, 4, 8, 0, 1);
3814 nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =
3815 nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
3817 s->current_picture.qscale_table[mb_xy]= s->qscale;
3818 write_back_non_zero_count(h);
3823 static int decode_cabac_mb_type( H264Context *h ) {
3824 MpegEncContext * const s = &h->s;
3826 if( h->slice_type == I_TYPE ) {
3827 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
3831 if( s->mb_x > 0 && !IS_INTRA4x4( s->current_picture.mb_type[mb_xy-1] ) )
3833 if( s->mb_y > 0 && !IS_INTRA4x4( s->current_picture.mb_type[mb_xy-s->mb_stride] ) )
3836 if( get_cabac( &h->cabac, &h->cabac_state[3+ctx] ) == 0 )
3837 return 0; /* I4x4 */
3839 if( get_cabac_terminate( &h->cabac ) )
3840 return 25; /* PCM */
3842 mb_type = 1; /* I16x16 */
3843 if( get_cabac( &h->cabac, &h->cabac_state[3+3] ) )
3844 mb_type += 12; /* cbp_luma != 0 */
3846 if( get_cabac( &h->cabac, &h->cabac_state[3+4] ) ) {
3847 if( get_cabac( &h->cabac, &h->cabac_state[3+5] ) )
3848 mb_type += 4 * 2; /* cbp_chroma == 2 */
3850 mb_type += 4 * 1; /* cbp_chroma == 1 */
3852 if( get_cabac( &h->cabac, &h->cabac_state[3+6] ) )
3854 if( get_cabac( &h->cabac, &h->cabac_state[3+7] ) )
3858 } else if( h->slice_type == P_TYPE ) {
3859 if( get_cabac( &h->cabac, &h->cabac_state[14] ) == 0 ) {
3861 if( get_cabac( &h->cabac, &h->cabac_state[15] ) == 0 ) {
3862 if( get_cabac( &h->cabac, &h->cabac_state[16] ) == 0 )
3863 return 0; /* P_L0_D16x16; */
3865 return 3; /* P_8x8; */
3867 if( get_cabac( &h->cabac, &h->cabac_state[17] ) == 0 )
3868 return 2; /* P_L0_D8x16; */
3870 return 1; /* P_L0_D16x8; */
3875 if( get_cabac( &h->cabac, &h->cabac_state[17] ) == 0 )
3876 return 5+0; /* I_4x4 */
3877 if( get_cabac_terminate( &h->cabac ) )
3878 return 5+25; /*I_PCM */
3879 mb_type = 5+1; /* I16x16 */
3880 if( get_cabac( &h->cabac, &h->cabac_state[17+1] ) )
3881 mb_type += 12; /* cbp_luma != 0 */
3883 if( get_cabac( &h->cabac, &h->cabac_state[17+2] ) ) {
3884 if( get_cabac( &h->cabac, &h->cabac_state[17+2] ) )
3885 mb_type += 4 * 2; /* cbp_chroma == 2 */
3887 mb_type += 4 * 1; /* cbp_chroma == 1 */
3889 if( get_cabac( &h->cabac, &h->cabac_state[17+3] ) )
3891 if( get_cabac( &h->cabac, &h->cabac_state[17+3] ) )
3897 /* TODO do others frames types */
3902 static int decode_cabac_mb_skip( H264Context *h) {
3903 MpegEncContext * const s = &h->s;
3904 const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
3905 const int mba_xy = mb_xy - 1;
3906 const int mbb_xy = mb_xy - s->mb_stride;
3909 if( s->mb_x > 0 && !IS_SKIP( s->current_picture.mb_type[mba_xy] ) )
3911 if( s->mb_y > 0 && !IS_SKIP( s->current_picture.mb_type[mbb_xy] ) )
3914 if( h->slice_type == P_TYPE || h->slice_type == SP_TYPE)
3915 return get_cabac( &h->cabac, &h->cabac_state[11+ctx] );
3917 return get_cabac( &h->cabac, &h->cabac_state[24+ctx] );
3920 static int decode_cabac_mb_intra4x4_pred_mode( H264Context *h, int pred_mode ) {
3923 if( get_cabac( &h->cabac, &h->cabac_state[68] ) )
3926 if( get_cabac( &h->cabac, &h->cabac_state[69] ) )
3928 if( get_cabac( &h->cabac, &h->cabac_state[69] ) )
3930 if( get_cabac( &h->cabac, &h->cabac_state[69] ) )
3932 if( mode >= pred_mode )
3938 static int decode_cabac_mb_chroma_pre_mode( H264Context *h) {
3939 MpegEncContext * const s = &h->s;
3940 const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
3941 const int mba_xy = mb_xy - 1;
3942 const int mbb_xy = mb_xy - s->mb_stride;
3946 /* No need to test for IS_INTRA4x4 and IS_INTRA16x16, as we set chroma_pred_mode_table to 0 */
3947 if( s->mb_x > 0 && h->chroma_pred_mode_table[mba_xy] != 0 )
3950 if( s->mb_y > 0 && h->chroma_pred_mode_table[mbb_xy] != 0 )
3953 if( get_cabac( &h->cabac, &h->cabac_state[64+ctx] ) == 0 )
3956 if( get_cabac( &h->cabac, &h->cabac_state[64+3] ) == 0 )
3958 if( get_cabac( &h->cabac, &h->cabac_state[64+3] ) == 0 )
3964 static const uint8_t block_idx_x[16] = {
3965 0, 1, 0, 1, 2, 3, 2, 3, 0, 1, 0, 1, 2, 3, 2, 3
3967 static const uint8_t block_idx_y[16] = {
3968 0, 0, 1, 1, 0, 0, 1, 1, 2, 2, 3, 3, 2, 2, 3, 3
3970 static const uint8_t block_idx_xy[4][4] = {
3977 static int decode_cabac_mb_cbp_luma( H264Context *h) {
3978 MpegEncContext * const s = &h->s;
3979 const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
3984 h->cbp_table[mb_xy] = 0; /* FIXME aaahahahah beurk */
3986 for( i8x8 = 0; i8x8 < 4; i8x8++ ) {
3992 x = block_idx_x[4*i8x8];
3993 y = block_idx_y[4*i8x8];
3997 else if( s->mb_x > 0 )
4002 else if( s->mb_y > 0 )
4003 mbb_xy = mb_xy - s->mb_stride;
4005 /* No need to test for skip as we put 0 for skip block */
4007 int i8x8a = block_idx_xy[(x-1)&0x03][y]/4;
4008 if( ((h->cbp_table[mba_xy] >> i8x8a)&0x01) == 0 )
4013 int i8x8b = block_idx_xy[x][(y-1)&0x03]/4;
4014 if( ((h->cbp_table[mbb_xy] >> i8x8b)&0x01) == 0 )
4018 if( get_cabac( &h->cabac, &h->cabac_state[73 + ctx] ) ) {
4020 h->cbp_table[mb_xy] = cbp; /* FIXME aaahahahah beurk */
4025 static int decode_cabac_mb_cbp_chroma( H264Context *h) {
4026 MpegEncContext * const s = &h->s;
4027 const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
4031 /* No need to test for skip */
4033 cbp_a = (h->cbp_table[mb_xy-1]>>4)&0x03;
4038 cbp_b = (h->cbp_table[mb_xy-s->mb_stride]>>4)&0x03;
4043 if( cbp_a > 0 ) ctx++;
4044 if( cbp_b > 0 ) ctx += 2;
4045 if( get_cabac( &h->cabac, &h->cabac_state[77 + ctx] ) == 0 )
4049 if( cbp_a == 2 ) ctx++;
4050 if( cbp_b == 2 ) ctx += 2;
4051 if( get_cabac( &h->cabac, &h->cabac_state[77 + ctx] ) )
4056 static int decode_cabac_mb_dqp( H264Context *h) {
4057 MpegEncContext * const s = &h->s;
4063 mbn_xy = s->mb_x + s->mb_y*s->mb_stride - 1;
4065 mbn_xy = s->mb_width - 1 + (s->mb_y-1)*s->mb_stride;
4067 if( mbn_xy >= 0 && h->last_qscale_diff != 0 && ( IS_INTRA16x16(s->current_picture.mb_type[mbn_xy] ) || (h->cbp_table[mbn_xy]&0x3f) ) )
4070 while( get_cabac( &h->cabac, &h->cabac_state[60 + ctx] ) ) {
4081 return -(val + 1)/2;
4083 static int decode_cabac_mb_sub_type( H264Context *h ) {
4084 if( get_cabac( &h->cabac, &h->cabac_state[21] ) )
4086 if( !get_cabac( &h->cabac, &h->cabac_state[22] ) )
4088 if( get_cabac( &h->cabac, &h->cabac_state[23] ) )
4093 static int decode_cabac_mb_ref( H264Context *h, int list, int n ) {
4094 int refa = h->ref_cache[list][scan8[n] - 1];
4095 int refb = h->ref_cache[list][scan8[n] - 8];
4104 while( get_cabac( &h->cabac, &h->cabac_state[54+ctx] ) ) {
4114 static int decode_cabac_mb_mvd( H264Context *h, int list, int n, int l ) {
4115 int amvd = abs( h->mvd_cache[list][scan8[n] - 1][l] ) +
4116 abs( h->mvd_cache[list][scan8[n] - 8][l] );
4117 int ctxbase = (l == 0) ? 40 : 47;
4123 else if( amvd > 32 )
4128 while( mvd < 9 && get_cabac( &h->cabac, &h->cabac_state[ctxbase+ctx] ) ) {
4138 while( get_cabac_bypass( &h->cabac ) ) {
4143 if( get_cabac_bypass( &h->cabac ) )
4147 if( mvd != 0 && get_cabac_bypass( &h->cabac ) )
4153 static int get_cabac_cbf_ctx( H264Context *h, int cat, int idx ) {
4154 MpegEncContext * const s = &h->s;
4155 const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
4166 if( IS_INTRA16x16(s->current_picture.mb_type[mba_xy] ) )
4167 nza = h->cbp_table[mba_xy]&0x100;
4170 mbb_xy = mb_xy - s->mb_stride;
4171 if( IS_INTRA16x16(s->current_picture.mb_type[mbb_xy] ) )
4172 nzb = h->cbp_table[mbb_xy]&0x100;
4174 } else if( cat == 1 || cat == 2 ) {
4178 x = block_idx_x[idx];
4179 y = block_idx_y[idx];
4183 else if( s->mb_x > 0 )
4188 else if( s->mb_y > 0 )
4189 mbb_xy = mb_xy - s->mb_stride;
4191 /* No need to test for skip */
4193 i8x8a = block_idx_xy[(x-1)&0x03][y]/4;
4195 if( !IS_INTRA_PCM(s->current_picture.mb_type[mba_xy] ) &&
4196 ((h->cbp_table[mba_xy]&0x0f)>>i8x8a))
4197 nza = h->non_zero_count_cache[scan8[idx] - 1];
4201 i8x8b = block_idx_xy[x][(y-1)&0x03]/4;
4203 if( !IS_INTRA_PCM(s->current_picture.mb_type[mbb_xy] ) &&
4204 ((h->cbp_table[mbb_xy]&0x0f)>>i8x8b))
4205 nzb = h->non_zero_count_cache[scan8[idx] - 8];
4207 } else if( cat == 3 ) {
4211 if( !IS_INTRA_PCM(s->current_picture.mb_type[mba_xy] ) &&
4212 (h->cbp_table[mba_xy]&0x30) )
4213 nza = (h->cbp_table[mba_xy]>>(6+idx))&0x01;
4216 mbb_xy = mb_xy - s->mb_stride;
4218 if( !IS_INTRA_PCM(s->current_picture.mb_type[mbb_xy] ) &&
4219 (h->cbp_table[mbb_xy]&0x30) )
4220 nzb = (h->cbp_table[mbb_xy]>>(6+idx))&0x01;
4222 } else if( cat == 4 ) {
4223 int idxc = idx % 4 ;
4224 if( idxc == 1 || idxc == 3 )
4226 else if( s->mb_x > 0 )
4229 if( idxc == 2 || idxc == 3 )
4231 else if( s->mb_y > 0 )
4232 mbb_xy = mb_xy - s->mb_stride;
4235 !IS_INTRA_PCM(s->current_picture.mb_type[mba_xy] ) &&
4236 (h->cbp_table[mba_xy]&0x30) == 0x20 )
4237 nza = h->non_zero_count_cache[scan8[16+idx] - 1];
4240 !IS_INTRA_PCM(s->current_picture.mb_type[mbb_xy] ) &&
4241 (h->cbp_table[mbb_xy]&0x30) == 0x20 )
4242 nzb = h->non_zero_count_cache[scan8[16+idx] - 8];
4245 if( ( mba_xy < 0 && IS_INTRA( s->current_picture.mb_type[mb_xy] ) ) ||
4246 ( mba_xy >= 0 && IS_INTRA_PCM(s->current_picture.mb_type[mba_xy] ) ) ||
4250 if( ( mbb_xy < 0 && IS_INTRA( s->current_picture.mb_type[mb_xy] ) ) ||
4251 ( mbb_xy >= 0 && IS_INTRA_PCM(s->current_picture.mb_type[mbb_xy] ) ) ||
4255 return ctx + 4 * cat;
4258 static int decode_cabac_residual( H264Context *h, DCTELEM *block, int cat, int n, const uint8_t *scantable, int qp, int max_coeff) {
4259 const int mb_xy = h->s.mb_x + h->s.mb_y*h->s.mb_stride;
4260 const uint16_t *qmul= dequant_coeff[qp];
4261 static const int significant_coeff_flag_offset[5] = { 0, 15, 29, 44, 47 };
4262 static const int last_significant_coeff_flag_offset[5] = { 0, 15, 29, 44, 47 };
4263 static const int coeff_abs_level_m1_offset[5] = { 0, 10, 20, 30, 39 };
4268 int coeff_count = 0;
4273 int abslevelgt1 = 0;
4275 /* cat: 0-> DC 16x16 n = 0
4276 * 1-> AC 16x16 n = luma4x4idx
4277 * 2-> Luma4x4 n = luma4x4idx
4278 * 3-> DC Chroma n = iCbCr
4279 * 4-> AC Chroma n = 4 * iCbCr + chroma4x4idx
4282 /* read coded block flag */
4283 if( get_cabac( &h->cabac, &h->cabac_state[85 + get_cabac_cbf_ctx( h, cat, n ) ] ) == 0 ) {
4284 if( cat == 1 || cat == 2 )
4285 h->non_zero_count_cache[scan8[n]] = 0;
4287 h->non_zero_count_cache[scan8[16+n]] = 0;
4292 while( last < max_coeff - 1 ) {
4293 int ctx = FFMIN( last, max_coeff - 2 );
4295 if( get_cabac( &h->cabac, &h->cabac_state[105+significant_coeff_flag_offset[cat]+ctx] ) == 0 ) {
4301 if( get_cabac( &h->cabac, &h->cabac_state[166+last_significant_coeff_flag_offset[cat]+ctx] ) ) {
4302 while( last < max_coeff ) {
4309 if( last == max_coeff -1 ) {
4314 if( cat == 0 && coeff_count > 0 )
4315 h->cbp_table[mb_xy] |= 0x100;
4316 else if( cat == 1 || cat == 2 )
4317 h->non_zero_count_cache[scan8[n]] = coeff_count;
4318 else if( cat == 3 && coeff_count > 0 )
4319 h->cbp_table[mb_xy] |= 0x40 << n;
4321 h->non_zero_count_cache[scan8[16+n]] = coeff_count;
4323 for( i = coeff_count - 1; i >= 0; i-- ) {
4326 int ctx = (abslevelgt1 != 0 ? 0 : FFMIN( 4, abslevel1 + 1 )) + coeff_abs_level_m1_offset[cat];
4328 if( get_cabac( &h->cabac, &h->cabac_state[227+ctx] ) == 0 ) {
4332 ctx = 5 + FFMIN( 4, abslevelgt1 ) + coeff_abs_level_m1_offset[cat];
4333 while( coeff_abs_m1 < 14 && get_cabac( &h->cabac, &h->cabac_state[227+ctx] ) ) {
4338 if( coeff_abs_m1 >= 14 ) {
4340 while( get_cabac_bypass( &h->cabac ) ) {
4341 coeff_abs_m1 += 1 << j;
4346 if( get_cabac_bypass( &h->cabac ) )
4347 coeff_abs_m1 += 1 << j ;
4350 if( get_cabac_bypass( &h->cabac ) )
4351 coeff[i] = -1 *( coeff_abs_m1 + 1 );
4353 coeff[i] = coeff_abs_m1 + 1;
4355 if( coeff_abs_m1 == 0 )
4361 if( cat == 0 || cat == 3 ) { /* DC */
4363 for( i = 0, j = 0; j < coeff_count; i++ ) {
4365 block[scantable[i]] = coeff[j];
4373 for( i = 0, j = 0; j < coeff_count; i++ ) {
4375 block[scantable[i]] = coeff[j] * qmul[scantable[i]];
4385 * decodes a macroblock
4386 * @returns 0 if ok, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
4388 static int decode_mb_cabac(H264Context *h) {
4389 MpegEncContext * const s = &h->s;
4390 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
4391 int mb_type, partition_count, cbp = 0;
4393 s->dsp.clear_blocks(h->mb); //FIXME avoid if allready clear (move after skip handlong?)
4395 if( h->slice_type == B_TYPE ) {
4396 av_log( h->s.avctx, AV_LOG_ERROR, "B-frame not supported with CABAC\n" );
4399 if( h->sps.mb_aff ) {
4400 av_log( h->s.avctx, AV_LOG_ERROR, "Fields not supported with CABAC\n" );
4404 if( h->slice_type != I_TYPE && h->slice_type != SI_TYPE ) {
4405 /* read skip flags */
4406 if( decode_cabac_mb_skip( h ) ) {
4410 mb_type= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0|MB_TYPE_SKIP;
4412 memset(h->non_zero_count[mb_xy], 0, 16);
4413 memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui
4415 if(h->sps.mb_aff && s->mb_skip_run==0 && (s->mb_y&1)==0){
4416 h->mb_field_decoding_flag= get_bits1(&s->gb);
4418 if(h->mb_field_decoding_flag)
4419 mb_type|= MB_TYPE_INTERLACED;
4422 fill_caches(h, mb_type); //FIXME check what is needed and what not ...
4423 pred_pskip_motion(h, &mx, &my);
4424 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
4425 fill_rectangle( h->mvd_cache[0][scan8[0]], 4, 4, 8, pack16to32(0,0), 4);
4426 fill_rectangle( h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4);
4427 write_back_motion(h, mb_type);
4429 s->current_picture.mb_type[mb_xy]= mb_type; //FIXME SKIP type
4430 s->current_picture.qscale_table[mb_xy]= s->qscale;
4431 h->slice_table[ mb_xy ]= h->slice_num;
4432 h->cbp_table[mb_xy] = 0;
4433 h->chroma_pred_mode_table[mb_xy] = 0;
4434 h->last_qscale_diff = 0;
4436 h->prev_mb_skiped= 1;
4442 h->prev_mb_skiped = 0;
4444 if( ( mb_type = decode_cabac_mb_type( h ) ) < 0 ) {
4445 av_log( h->s.avctx, AV_LOG_ERROR, "decode_cabac_mb_type failed\n" );
4449 if( h->slice_type == P_TYPE ) {
4451 partition_count= p_mb_type_info[mb_type].partition_count;
4452 mb_type= p_mb_type_info[mb_type].type;
4455 goto decode_intra_mb;
4458 assert(h->slice_type == I_TYPE);
4460 partition_count = 0;
4461 cbp= i_mb_type_info[mb_type].cbp;
4462 h->intra16x16_pred_mode= i_mb_type_info[mb_type].pred_mode;
4463 mb_type= i_mb_type_info[mb_type].type;
4466 if(h->mb_field_decoding_flag)
4467 mb_type |= MB_TYPE_INTERLACED;
4470 s->current_picture.mb_type[mb_xy]= mb_type;
4471 h->slice_table[ mb_xy ]= h->slice_num;
4473 if(IS_INTRA_PCM(mb_type)) {
4475 h->cbp_table[mb_xy] = 0xf +4*2;
4476 h->chroma_pred_mode_table[mb_xy] = 0;
4477 s->current_picture.qscale_table[mb_xy]= s->qscale;
4481 fill_caches(h, mb_type);
4483 if( IS_INTRA( mb_type ) ) {
4484 if( IS_INTRA4x4( mb_type ) ) {
4486 for( i = 0; i < 16; i++ ) {
4487 int pred = pred_intra_mode( h, i );
4488 h->intra4x4_pred_mode_cache[ scan8[i] ] = decode_cabac_mb_intra4x4_pred_mode( h, pred );
4490 //av_log( s->avctx, AV_LOG_ERROR, "i4x4 pred=%d mode=%d\n", pred, h->intra4x4_pred_mode_cache[ scan8[i] ] );
4492 write_back_intra_pred_mode(h);
4493 if( check_intra4x4_pred_mode(h) < 0 ) return -1;
4495 h->intra16x16_pred_mode= check_intra_pred_mode( h, h->intra16x16_pred_mode );
4496 if( h->intra16x16_pred_mode < 0 ) return -1;
4498 h->chroma_pred_mode_table[mb_xy] =
4499 h->chroma_pred_mode = decode_cabac_mb_chroma_pre_mode( h );
4501 h->chroma_pred_mode= check_intra_pred_mode( h, h->chroma_pred_mode );
4502 if( h->chroma_pred_mode < 0 ) return -1;
4503 } else if( partition_count == 4 ) {
4504 int i, j, sub_partition_count[4], list, ref[2][4];
4507 for( i = 0; i < 4; i++ ) {
4508 h->sub_mb_type[i] = decode_cabac_mb_sub_type( h );
4509 sub_partition_count[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
4510 h->sub_mb_type[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].type;
4513 for( list = 0; list < 2; list++ ) {
4514 if( h->ref_count[list] > 0 ) {
4515 for( i = 0; i < 4; i++ ) {
4516 if(IS_DIR(h->sub_mb_type[i], 0, list) && !IS_DIRECT(h->sub_mb_type[i])){
4517 if( h->ref_count[list] > 1 )
4518 ref[list][i] = decode_cabac_mb_ref( h, list, 4*i );
4524 h->ref_cache[list][ scan8[4*i]+1 ]=
4525 h->ref_cache[list][ scan8[4*i]+8 ]=h->ref_cache[list][ scan8[4*i]+9 ]= ref[list][i];
4530 for(list=0; list<2; list++){
4532 h->ref_cache[list][ scan8[4*i] ]=h->ref_cache[list][ scan8[4*i]+1 ];
4534 if(IS_DIR(h->sub_mb_type[i], 0, list) && !IS_DIRECT(h->sub_mb_type[i])){
4535 const int sub_mb_type= h->sub_mb_type[i];
4536 const int block_width= (sub_mb_type & (MB_TYPE_16x16|MB_TYPE_16x8)) ? 2 : 1;
4537 for(j=0; j<sub_partition_count[i]; j++){
4540 const int index= 4*i + block_width*j;
4541 int16_t (* mv_cache)[2]= &h->mv_cache[list][ scan8[index] ];
4542 int16_t (* mvd_cache)[2]= &h->mvd_cache[list][ scan8[index] ];
4543 pred_motion(h, index, block_width, list, h->ref_cache[list][ scan8[index] ], &mpx, &mpy);
4545 mx = mpx + decode_cabac_mb_mvd( h, list, index, 0 );
4546 my = mpy + decode_cabac_mb_mvd( h, list, index, 1 );
4547 tprintf("final mv:%d %d\n", mx, my);
4549 if(IS_SUB_8X8(sub_mb_type)){
4550 mv_cache[ 0 ][0]= mv_cache[ 1 ][0]=
4551 mv_cache[ 8 ][0]= mv_cache[ 9 ][0]= mx;
4552 mv_cache[ 0 ][1]= mv_cache[ 1 ][1]=
4553 mv_cache[ 8 ][1]= mv_cache[ 9 ][1]= my;
4555 mvd_cache[ 0 ][0]= mvd_cache[ 1 ][0]=
4556 mvd_cache[ 8 ][0]= mvd_cache[ 9 ][0]= mx - mpx;
4557 mvd_cache[ 0 ][1]= mvd_cache[ 1 ][1]=
4558 mvd_cache[ 8 ][1]= mvd_cache[ 9 ][1]= my - mpy;
4559 }else if(IS_SUB_8X4(sub_mb_type)){
4560 mv_cache[ 0 ][0]= mv_cache[ 1 ][0]= mx;
4561 mv_cache[ 0 ][1]= mv_cache[ 1 ][1]= my;
4563 mvd_cache[ 0 ][0]= mvd_cache[ 1 ][0]= mx- mpx;
4564 mvd_cache[ 0 ][1]= mvd_cache[ 1 ][1]= my - mpy;
4565 }else if(IS_SUB_4X8(sub_mb_type)){
4566 mv_cache[ 0 ][0]= mv_cache[ 8 ][0]= mx;
4567 mv_cache[ 0 ][1]= mv_cache[ 8 ][1]= my;
4569 mvd_cache[ 0 ][0]= mvd_cache[ 8 ][0]= mx - mpx;
4570 mvd_cache[ 0 ][1]= mvd_cache[ 8 ][1]= my - mpy;
4572 assert(IS_SUB_4X4(sub_mb_type));
4573 mv_cache[ 0 ][0]= mx;
4574 mv_cache[ 0 ][1]= my;
4576 mvd_cache[ 0 ][0]= mx - mpx;
4577 mvd_cache[ 0 ][1]= my - mpy;
4581 uint32_t *p= (uint32_t *)&h->mv_cache[list][ scan8[4*i] ][0];
4582 uint32_t *pd= (uint32_t *)&h->mvd_cache[list][ scan8[4*i] ][0];
4583 p[0] = p[1] = p[8] = p[9] = 0;
4584 pd[0]= pd[1]= pd[8]= pd[9]= 0;
4588 } else if( !IS_DIRECT(mb_type) ) {
4589 int list, mx, my, i, mpx, mpy;
4590 if(IS_16X16(mb_type)){
4591 for(list=0; list<2; list++){
4592 if(IS_DIR(mb_type, 0, list)){
4593 if(h->ref_count[list] > 0 ){
4594 const int ref = h->ref_count[list] > 1 ? decode_cabac_mb_ref( h, list, 0 ) : 0;
4595 fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, ref, 1);
4599 for(list=0; list<2; list++){
4600 if(IS_DIR(mb_type, 0, list)){
4601 pred_motion(h, 0, 4, list, h->ref_cache[list][ scan8[0] ], &mpx, &mpy);
4603 mx = mpx + decode_cabac_mb_mvd( h, list, 0, 0 );
4604 my = mpy + decode_cabac_mb_mvd( h, list, 0, 1 );
4605 tprintf("final mv:%d %d\n", mx, my);
4607 fill_rectangle(h->mvd_cache[list][ scan8[0] ], 4, 4, 8, pack16to32(mx-mpx,my-mpy), 4);
4608 fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, pack16to32(mx,my), 4);
4612 else if(IS_16X8(mb_type)){
4613 for(list=0; list<2; list++){
4614 if(h->ref_count[list]>0){
4616 if(IS_DIR(mb_type, i, list)){
4617 const int ref= h->ref_count[list] > 1 ? decode_cabac_mb_ref( h, list, 8*i ) : 0;
4618 fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, ref, 1);
4623 for(list=0; list<2; list++){
4625 if(IS_DIR(mb_type, i, list)){
4626 pred_16x8_motion(h, 8*i, list, h->ref_cache[list][scan8[0] + 16*i], &mpx, &mpy);
4627 mx = mpx + decode_cabac_mb_mvd( h, list, 8*i, 0 );
4628 my = mpy + decode_cabac_mb_mvd( h, list, 8*i, 1 );
4629 tprintf("final mv:%d %d\n", mx, my);
4631 fill_rectangle(h->mvd_cache[list][ scan8[0] + 16*i ], 4, 2, 8, pack16to32(mx-mpx,my-mpy), 4);
4632 fill_rectangle(h->mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, pack16to32(mx,my), 4);
4637 assert(IS_8X16(mb_type));
4638 for(list=0; list<2; list++){
4639 if(h->ref_count[list]>0){
4641 if(IS_DIR(mb_type, i, list)){ //FIXME optimize
4642 const int ref= h->ref_count[list] > 1 ? decode_cabac_mb_ref( h, list, 4*i ) : 0;
4643 fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, ref, 1);
4648 for(list=0; list<2; list++){
4650 if(IS_DIR(mb_type, i, list)){
4651 pred_8x16_motion(h, i*4, list, h->ref_cache[list][ scan8[0] + 2*i ], &mpx, &mpy);
4652 mx = mpx + decode_cabac_mb_mvd( h, list, 4*i, 0 );
4653 my = mpy + decode_cabac_mb_mvd( h, list, 4*i, 1 );
4655 tprintf("final mv:%d %d\n", mx, my);
4656 fill_rectangle(h->mvd_cache[list][ scan8[0] + 2*i ], 2, 4, 8, pack16to32(mx-mpx,my-mpy), 4);
4657 fill_rectangle(h->mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, pack16to32(mx,my), 4);
4664 if( IS_INTER( mb_type ) ) {
4665 h->chroma_pred_mode_table[mb_xy] = 0;
4666 write_back_motion( h, mb_type );
4669 if( !IS_INTRA16x16( mb_type ) ) {
4670 cbp = decode_cabac_mb_cbp_luma( h );
4671 cbp |= decode_cabac_mb_cbp_chroma( h ) << 4;
4674 h->cbp_table[mb_xy] = cbp;
4676 if( cbp || IS_INTRA16x16( mb_type ) ) {
4677 const uint8_t *scan, *dc_scan;
4680 if(IS_INTERLACED(mb_type)){
4682 dc_scan= luma_dc_field_scan;
4685 dc_scan= luma_dc_zigzag_scan;
4688 h->last_qscale_diff = dqp = decode_cabac_mb_dqp( h );
4690 if(((unsigned)s->qscale) > 51){
4691 if(s->qscale<0) s->qscale+= 52;
4692 else s->qscale-= 52;
4694 h->chroma_qp = get_chroma_qp(h, s->qscale);
4696 if( IS_INTRA16x16( mb_type ) ) {
4698 //av_log( s->avctx, AV_LOG_ERROR, "INTRA16x16 DC\n" );
4699 if( decode_cabac_residual( h, h->mb, 0, 0, dc_scan, s->qscale, 16) < 0)
4702 for( i = 0; i < 16; i++ ) {
4703 //av_log( s->avctx, AV_LOG_ERROR, "INTRA16x16 AC:%d\n", i );
4704 if( decode_cabac_residual(h, h->mb + 16*i, 1, i, scan + 1, s->qscale, 15) < 0 )
4708 fill_rectangle(&h->non_zero_count_cache[scan8[0]], 4, 4, 8, 0, 1);
4712 for( i8x8 = 0; i8x8 < 4; i8x8++ ) {
4713 if( cbp & (1<<i8x8) ) {
4714 for( i4x4 = 0; i4x4 < 4; i4x4++ ) {
4715 const int index = 4*i8x8 + i4x4;
4716 //av_log( s->avctx, AV_LOG_ERROR, "Luma4x4: %d\n", index );
4717 if( decode_cabac_residual(h, h->mb + 16*index, 2, index, scan, s->qscale, 16) < 0 )
4721 uint8_t * const nnz= &h->non_zero_count_cache[ scan8[4*i8x8] ];
4722 nnz[0] = nnz[1] = nnz[8] = nnz[9] = 0;
4729 for( c = 0; c < 2; c++ ) {
4730 //av_log( s->avctx, AV_LOG_ERROR, "INTRA C%d-DC\n",c );
4731 if( decode_cabac_residual(h, h->mb + 256 + 16*4*c, 3, c, chroma_dc_scan, h->chroma_qp, 4) < 0)
4738 for( c = 0; c < 2; c++ ) {
4739 for( i = 0; i < 4; i++ ) {
4740 const int index = 16 + 4 * c + i;
4741 //av_log( s->avctx, AV_LOG_ERROR, "INTRA C%d-AC %d\n",c, index - 16 );
4742 if( decode_cabac_residual(h, h->mb + 16*index, 4, index - 16, scan + 1, h->chroma_qp, 15) < 0)
4747 uint8_t * const nnz= &h->non_zero_count_cache[0];
4748 nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =
4749 nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
4752 memset( &h->non_zero_count_cache[8], 0, 8*5 );
4755 s->current_picture.qscale_table[mb_xy]= s->qscale;
4756 write_back_non_zero_count(h);
4762 static void filter_mb_edgev( H264Context *h, uint8_t *pix, int stride, int bS[4], int qp ) {
4764 const int index_a = clip( qp + h->slice_alpha_c0_offset, 0, 51 );
4765 const int alpha = alpha_table[index_a];
4766 const int beta = beta_table[clip( qp + h->slice_beta_offset, 0, 51 )];
4768 for( i = 0; i < 4; i++ ) {
4775 const int tc0 = tc0_table[index_a][bS[i] - 1];
4776 /* 4px edge length */
4777 for( d = 0; d < 4; d++ ) {
4778 const int p0 = pix[-1];
4779 const int p1 = pix[-2];
4780 const int p2 = pix[-3];
4781 const int q0 = pix[0];
4782 const int q1 = pix[1];
4783 const int q2 = pix[2];
4785 if( ABS( p0 - q0 ) < alpha &&
4786 ABS( p1 - p0 ) < beta &&
4787 ABS( q1 - q0 ) < beta ) {
4791 if( ABS( p2 - p0 ) < beta ) {
4792 pix[-2] = p1 + clip( ( p2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( p1 << 1 ) ) >> 1, -tc0, tc0 );
4795 if( ABS( q2 - q0 ) < beta ) {
4796 pix[1] = q1 + clip( ( q2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( q1 << 1 ) ) >> 1, -tc0, tc0 );
4800 i_delta = clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );
4801 pix[-1] = clip_uint8( p0 + i_delta ); /* p0' */
4802 pix[0] = clip_uint8( q0 - i_delta ); /* q0' */
4807 /* 4px edge length */
4808 for( d = 0; d < 4; d++ ) {
4809 const int p0 = pix[-1];
4810 const int p1 = pix[-2];
4811 const int p2 = pix[-3];
4813 const int q0 = pix[0];
4814 const int q1 = pix[1];
4815 const int q2 = pix[2];
4817 if( ABS( p0 - q0 ) < alpha &&
4818 ABS( p1 - p0 ) < beta &&
4819 ABS( q1 - q0 ) < beta ) {
4821 if(ABS( p0 - q0 ) < (( alpha >> 2 ) + 2 )){
4822 if( ABS( p2 - p0 ) < beta)
4824 const int p3 = pix[-4];
4826 pix[-1] = ( p2 + 2*p1 + 2*p0 + 2*q0 + q1 + 4 ) >> 3;
4827 pix[-2] = ( p2 + p1 + p0 + q0 + 2 ) >> 2;
4828 pix[-3] = ( 2*p3 + 3*p2 + p1 + p0 + q0 + 4 ) >> 3;
4831 pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
4833 if( ABS( q2 - q0 ) < beta)
4835 const int q3 = pix[3];
4837 pix[0] = ( p1 + 2*p0 + 2*q0 + 2*q1 + q2 + 4 ) >> 3;
4838 pix[1] = ( p0 + q0 + q1 + q2 + 2 ) >> 2;
4839 pix[2] = ( 2*q3 + 3*q2 + q1 + q0 + p0 + 4 ) >> 3;
4842 pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
4846 pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
4847 pix[ 0] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
4855 static void filter_mb_edgecv( H264Context *h, uint8_t *pix, int stride, int bS[4], int qp ) {
4857 const int index_a = clip( qp + h->slice_alpha_c0_offset, 0, 51 );
4858 const int alpha = alpha_table[index_a];
4859 const int beta = beta_table[clip( qp + h->slice_beta_offset, 0, 51 )];
4861 for( i = 0; i < 4; i++ ) {
4868 const int tc = tc0_table[index_a][bS[i] - 1] + 1;
4869 /* 2px edge length (because we use same bS than the one for luma) */
4870 for( d = 0; d < 2; d++ ){
4871 const int p0 = pix[-1];
4872 const int p1 = pix[-2];
4873 const int q0 = pix[0];
4874 const int q1 = pix[1];
4876 if( ABS( p0 - q0 ) < alpha &&
4877 ABS( p1 - p0 ) < beta &&
4878 ABS( q1 - q0 ) < beta ) {
4879 const int i_delta = clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );
4881 pix[-1] = clip_uint8( p0 + i_delta ); /* p0' */
4882 pix[0] = clip_uint8( q0 - i_delta ); /* q0' */
4887 /* 2px edge length (because we use same bS than the one for luma) */
4888 for( d = 0; d < 2; d++ ){
4889 const int p0 = pix[-1];
4890 const int p1 = pix[-2];
4891 const int q0 = pix[0];
4892 const int q1 = pix[1];
4894 if( ABS( p0 - q0 ) < alpha &&
4895 ABS( p1 - p0 ) < beta &&
4896 ABS( q1 - q0 ) < beta ) {
4898 pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2; /* p0' */
4899 pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2; /* q0' */
4907 static void filter_mb_edgeh( H264Context *h, uint8_t *pix, int stride, int bS[4], int qp ) {
4909 const int index_a = clip( qp + h->slice_alpha_c0_offset, 0, 51 );
4910 const int alpha = alpha_table[index_a];
4911 const int beta = beta_table[clip( qp + h->slice_beta_offset, 0, 51 )];
4912 const int pix_next = stride;
4914 for( i = 0; i < 4; i++ ) {
4921 const int tc0 = tc0_table[index_a][bS[i] - 1];
4922 /* 4px edge length */
4923 for( d = 0; d < 4; d++ ) {
4924 const int p0 = pix[-1*pix_next];
4925 const int p1 = pix[-2*pix_next];
4926 const int p2 = pix[-3*pix_next];
4927 const int q0 = pix[0];
4928 const int q1 = pix[1*pix_next];
4929 const int q2 = pix[2*pix_next];
4931 if( ABS( p0 - q0 ) < alpha &&
4932 ABS( p1 - p0 ) < beta &&
4933 ABS( q1 - q0 ) < beta ) {
4938 if( ABS( p2 - p0 ) < beta ) {
4939 pix[-2*pix_next] = p1 + clip( ( p2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( p1 << 1 ) ) >> 1, -tc0, tc0 );
4942 if( ABS( q2 - q0 ) < beta ) {
4943 pix[pix_next] = q1 + clip( ( q2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( q1 << 1 ) ) >> 1, -tc0, tc0 );
4947 i_delta = clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );
4948 pix[-pix_next] = clip_uint8( p0 + i_delta ); /* p0' */
4949 pix[0] = clip_uint8( q0 - i_delta ); /* q0' */
4954 /* 4px edge length */
4955 for( d = 0; d < 4; d++ ) {
4956 const int p0 = pix[-1*pix_next];
4957 const int p1 = pix[-2*pix_next];
4958 const int p2 = pix[-3*pix_next];
4959 const int q0 = pix[0];
4960 const int q1 = pix[1*pix_next];
4961 const int q2 = pix[2*pix_next];
4963 if( ABS( p0 - q0 ) < alpha &&
4964 ABS( p1 - p0 ) < beta &&
4965 ABS( q1 - q0 ) < beta ) {
4967 const int p3 = pix[-4*pix_next];
4968 const int q3 = pix[ 3*pix_next];
4970 if(ABS( p0 - q0 ) < (( alpha >> 2 ) + 2 )){
4971 if( ABS( p2 - p0 ) < beta) {
4973 pix[-1*pix_next] = ( p2 + 2*p1 + 2*p0 + 2*q0 + q1 + 4 ) >> 3;
4974 pix[-2*pix_next] = ( p2 + p1 + p0 + q0 + 2 ) >> 2;
4975 pix[-3*pix_next] = ( 2*p3 + 3*p2 + p1 + p0 + q0 + 4 ) >> 3;
4978 pix[-1*pix_next] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
4980 if( ABS( q2 - q0 ) < beta) {
4982 pix[0*pix_next] = ( p1 + 2*p0 + 2*q0 + 2*q1 + q2 + 4 ) >> 3;
4983 pix[1*pix_next] = ( p0 + q0 + q1 + q2 + 2 ) >> 2;
4984 pix[2*pix_next] = ( 2*q3 + 3*q2 + q1 + q0 + p0 + 4 ) >> 3;
4987 pix[0*pix_next] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
4991 pix[-1*pix_next] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
4992 pix[ 0*pix_next] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
5001 static void filter_mb_edgech( H264Context *h, uint8_t *pix, int stride, int bS[4], int qp ) {
5003 const int index_a = clip( qp + h->slice_alpha_c0_offset, 0, 51 );
5004 const int alpha = alpha_table[index_a];
5005 const int beta = beta_table[clip( qp + h->slice_beta_offset, 0, 51 )];
5006 const int pix_next = stride;
5008 for( i = 0; i < 4; i++ )
5016 int tc = tc0_table[index_a][bS[i] - 1] + 1;
5017 /* 2px edge length (see deblocking_filter_edgecv) */
5018 for( d = 0; d < 2; d++ ) {
5019 const int p0 = pix[-1*pix_next];
5020 const int p1 = pix[-2*pix_next];
5021 const int q0 = pix[0];
5022 const int q1 = pix[1*pix_next];
5024 if( ABS( p0 - q0 ) < alpha &&
5025 ABS( p1 - p0 ) < beta &&
5026 ABS( q1 - q0 ) < beta ) {
5028 int i_delta = clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );
5030 pix[-pix_next] = clip_uint8( p0 + i_delta ); /* p0' */
5031 pix[0] = clip_uint8( q0 - i_delta ); /* q0' */
5036 /* 2px edge length (see deblocking_filter_edgecv) */
5037 for( d = 0; d < 2; d++ ) {
5038 const int p0 = pix[-1*pix_next];
5039 const int p1 = pix[-2*pix_next];
5040 const int q0 = pix[0];
5041 const int q1 = pix[1*pix_next];
5043 if( ABS( p0 - q0 ) < alpha &&
5044 ABS( p1 - p0 ) < beta &&
5045 ABS( q1 - q0 ) < beta ) {
5047 pix[-pix_next] = ( 2*p1 + p0 + q1 + 2 ) >> 2; /* p0' */
5048 pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2; /* q0' */
5056 static void filter_mb( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr) {
5057 MpegEncContext * const s = &h->s;
5058 const int mb_xy= mb_x + mb_y*s->mb_stride;
5059 int linesize, uvlinesize;
5062 /* FIXME Implement deblocking filter for field MB */
5063 if( h->sps.mb_aff ) {
5066 linesize = s->linesize;
5067 uvlinesize = s->uvlinesize;
5069 /* dir : 0 -> vertical edge, 1 -> horizontal edge */
5070 for( dir = 0; dir < 2; dir++ )
5075 /* test picture boundary */
5076 if( ( dir == 0 && mb_x == 0 ) || ( dir == 1 && mb_y == 0 ) ) {
5079 /* FIXME test slice boundary */
5080 if( h->deblocking_filter == 2 ) {
5084 for( edge = start; edge < 4; edge++ ) {
5085 /* mbn_xy: neighbour macroblock (how that works for field ?) */
5086 int mbn_xy = edge > 0 ? mb_xy : ( dir == 0 ? mb_xy -1 : mb_xy - s->mb_stride );
5090 if( IS_INTRA( s->current_picture.mb_type[mb_xy] ) ||
5091 IS_INTRA( s->current_picture.mb_type[mbn_xy] ) ) {
5092 bS[0] = bS[1] = bS[2] = bS[3] = ( edge == 0 ? 4 : 3 );
5095 for( i = 0; i < 4; i++ ) {
5096 int x = dir == 0 ? edge : i;
5097 int y = dir == 0 ? i : edge;
5098 int b_idx= 8 + 4 + x + 8*y;
5099 int bn_idx= b_idx - (dir ? 8:1);
5101 if( h->non_zero_count_cache[b_idx] != 0 ||
5102 h->non_zero_count_cache[bn_idx] != 0 ) {
5105 else if( h->slice_type == P_TYPE ) {
5106 if( h->ref_cache[0][b_idx] != h->ref_cache[0][bn_idx] ||
5107 ABS( h->mv_cache[0][b_idx][0] - h->mv_cache[0][bn_idx][0] ) >= 4 ||
5108 ABS( h->mv_cache[0][b_idx][1] - h->mv_cache[0][bn_idx][1] ) >= 4 )
5114 /* FIXME Add support for B frame */
5119 if(bS[0]+bS[1]+bS[2]+bS[3] == 0)
5124 qp = ( s->qscale + s->current_picture.qscale_table[mbn_xy] + 1 ) >> 1;
5126 filter_mb_edgev( h, &img_y[4*edge], linesize, bS, qp );
5127 if( (edge&1) == 0 ) {
5128 int chroma_qp = ( h->chroma_qp +
5129 get_chroma_qp( h, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1;
5130 filter_mb_edgecv( h, &img_cb[2*edge], uvlinesize, bS, chroma_qp );
5131 filter_mb_edgecv( h, &img_cr[2*edge], uvlinesize, bS, chroma_qp );
5134 filter_mb_edgeh( h, &img_y[4*edge*linesize], linesize, bS, qp );
5135 if( (edge&1) == 0 ) {
5136 int chroma_qp = ( h->chroma_qp +
5137 get_chroma_qp( h, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1;
5138 filter_mb_edgech( h, &img_cb[2*edge*uvlinesize], uvlinesize, bS, chroma_qp );
5139 filter_mb_edgech( h, &img_cr[2*edge*uvlinesize], uvlinesize, bS, chroma_qp );
5146 static int decode_slice(H264Context *h){
5147 MpegEncContext * const s = &h->s;
5148 const int part_mask= s->partitioned_frame ? (AC_END|AC_ERROR) : 0x7F;
5152 if( h->pps.cabac ) {
5156 align_get_bits( &s->gb );
5159 ff_init_cabac_states( &h->cabac, ff_h264_lps_range, ff_h264_mps_state, ff_h264_lps_state, 64 );
5160 ff_init_cabac_decoder( &h->cabac,
5161 s->gb.buffer + get_bits_count(&s->gb)/8,
5162 ( s->gb.size_in_bits - get_bits_count(&s->gb) + 7)/8);
5163 /* calculate pre-state */
5164 for( i= 0; i < 399; i++ ) {
5166 if( h->slice_type == I_TYPE )
5167 pre = clip( ((cabac_context_init_I[i][0] * s->qscale) >>4 ) + cabac_context_init_I[i][1], 1, 126 );
5169 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 );
5172 h->cabac_state[i] = 2 * ( 63 - pre ) + 0;
5174 h->cabac_state[i] = 2 * ( pre - 64 ) + 1;
5178 int ret = decode_mb_cabac(h);
5179 int eos = get_cabac_terminate( &h->cabac ); /* End of Slice flag */
5181 if(ret>=0) hl_decode_mb(h);
5183 /* XXX: useless as decode_mb_cabac it doesn't support that ... */
5184 if( ret >= 0 && h->sps.mb_aff ) { //FIXME optimal? or let mb_decode decode 16x32 ?
5187 if(ret>=0) ret = decode_mb_cabac(h);
5188 eos = get_cabac_terminate( &h->cabac );
5194 if( ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 1) {
5195 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
5196 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);
5200 if( ++s->mb_x >= s->mb_width ) {
5202 ff_draw_horiz_band(s, 16*s->mb_y, 16);
5203 if( ++s->mb_y >= s->mb_height ) {
5204 tprintf("slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
5208 if( eos || s->mb_y >= s->mb_height ) {
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 /* TODO test over-reading in cabac code */
5214 else if( read too much in h->cabac ) {
5215 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);
5223 int ret = decode_mb_cavlc(h);
5225 if(ret>=0) hl_decode_mb(h);
5227 if(ret>=0 && h->sps.mb_aff){ //FIXME optimal? or let mb_decode decode 16x32 ?
5229 ret = decode_mb_cavlc(h);
5231 if(ret>=0) hl_decode_mb(h);
5236 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
5237 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);
5242 if(++s->mb_x >= s->mb_width){
5244 ff_draw_horiz_band(s, 16*s->mb_y, 16);
5245 if(++s->mb_y >= s->mb_height){
5246 tprintf("slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
5248 if(get_bits_count(&s->gb) == s->gb.size_in_bits ) {
5249 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);
5253 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);
5260 if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->mb_skip_run<=0){
5261 if(get_bits_count(&s->gb) == s->gb.size_in_bits ){
5262 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);
5266 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);
5275 for(;s->mb_y < s->mb_height; s->mb_y++){
5276 for(;s->mb_x < s->mb_width; s->mb_x++){
5277 int ret= decode_mb(h);
5282 fprintf(stderr, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
5283 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);
5288 if(++s->mb_x >= s->mb_width){
5290 if(++s->mb_y >= s->mb_height){
5291 if(get_bits_count(s->gb) == s->gb.size_in_bits){
5292 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);
5296 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);
5303 if(get_bits_count(s->?gb) >= s->gb?.size_in_bits){
5304 if(get_bits_count(s->gb) == s->gb.size_in_bits){
5305 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);
5309 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);
5316 ff_draw_horiz_band(s, 16*s->mb_y, 16);
5319 return -1; //not reached
5322 static inline int decode_vui_parameters(H264Context *h, SPS *sps){
5323 MpegEncContext * const s = &h->s;
5324 int aspect_ratio_info_present_flag, aspect_ratio_idc;
5326 aspect_ratio_info_present_flag= get_bits1(&s->gb);
5328 if( aspect_ratio_info_present_flag ) {
5329 aspect_ratio_idc= get_bits(&s->gb, 8);
5330 if( aspect_ratio_idc == EXTENDED_SAR ) {
5331 sps->sar.num= get_bits(&s->gb, 16);
5332 sps->sar.den= get_bits(&s->gb, 16);
5333 }else if(aspect_ratio_idc < 16){
5334 sps->sar= pixel_aspect[aspect_ratio_idc];
5336 av_log(h->s.avctx, AV_LOG_ERROR, "illegal aspect ratio\n");
5343 // s->avctx->aspect_ratio= sar_width*s->width / (float)(s->height*sar_height);
5345 if(get_bits1(&s->gb)){ /* overscan_info_present_flag */
5346 get_bits1(&s->gb); /* overscan_appropriate_flag */
5349 if(get_bits1(&s->gb)){ /* video_signal_type_present_flag */
5350 get_bits(&s->gb, 3); /* video_format */
5351 get_bits1(&s->gb); /* video_full_range_flag */
5352 if(get_bits1(&s->gb)){ /* colour_description_present_flag */
5353 get_bits(&s->gb, 8); /* colour_primaries */
5354 get_bits(&s->gb, 8); /* transfer_characteristics */
5355 get_bits(&s->gb, 8); /* matrix_coefficients */
5359 if(get_bits1(&s->gb)){ /* chroma_location_info_present_flag */
5360 get_ue_golomb(&s->gb); /* chroma_sample_location_type_top_field */
5361 get_ue_golomb(&s->gb); /* chroma_sample_location_type_bottom_field */
5364 sps->timing_info_present_flag = get_bits1(&s->gb);
5365 if(sps->timing_info_present_flag){
5366 sps->num_units_in_tick = get_bits_long(&s->gb, 32);
5367 sps->time_scale = get_bits_long(&s->gb, 32);
5368 sps->fixed_frame_rate_flag = get_bits1(&s->gb);
5372 | nal_hrd_parameters_present_flag |0 |u(1) |
5373 | if( nal_hrd_parameters_present_flag = = 1) | | |
5374 | hrd_parameters( ) | | |
5375 | vcl_hrd_parameters_present_flag |0 |u(1) |
5376 | if( vcl_hrd_parameters_present_flag = = 1) | | |
5377 | hrd_parameters( ) | | |
5378 | if( ( nal_hrd_parameters_present_flag = = 1 | || | |
5380 |( vcl_hrd_parameters_present_flag = = 1 ) ) | | |
5381 | low_delay_hrd_flag |0 |u(1) |
5382 | bitstream_restriction_flag |0 |u(1) |
5383 | if( bitstream_restriction_flag ) { |0 |u(1) |
5384 | motion_vectors_over_pic_boundaries_flag |0 |u(1) |
5385 | max_bytes_per_pic_denom |0 |ue(v) |
5386 | max_bits_per_mb_denom |0 |ue(v) |
5387 | log2_max_mv_length_horizontal |0 |ue(v) |
5388 | log2_max_mv_length_vertical |0 |ue(v) |
5389 | num_reorder_frames |0 |ue(v) |
5390 | max_dec_frame_buffering |0 |ue(v) |
5397 static inline int decode_seq_parameter_set(H264Context *h){
5398 MpegEncContext * const s = &h->s;
5399 int profile_idc, level_idc;
5403 profile_idc= get_bits(&s->gb, 8);
5404 get_bits1(&s->gb); //constraint_set0_flag
5405 get_bits1(&s->gb); //constraint_set1_flag
5406 get_bits1(&s->gb); //constraint_set2_flag
5407 get_bits(&s->gb, 5); // reserved
5408 level_idc= get_bits(&s->gb, 8);
5409 sps_id= get_ue_golomb(&s->gb);
5411 sps= &h->sps_buffer[ sps_id ];
5412 sps->profile_idc= profile_idc;
5413 sps->level_idc= level_idc;
5415 sps->log2_max_frame_num= get_ue_golomb(&s->gb) + 4;
5416 sps->poc_type= get_ue_golomb(&s->gb);
5418 if(sps->poc_type == 0){ //FIXME #define
5419 sps->log2_max_poc_lsb= get_ue_golomb(&s->gb) + 4;
5420 } else if(sps->poc_type == 1){//FIXME #define
5421 sps->delta_pic_order_always_zero_flag= get_bits1(&s->gb);
5422 sps->offset_for_non_ref_pic= get_se_golomb(&s->gb);
5423 sps->offset_for_top_to_bottom_field= get_se_golomb(&s->gb);
5424 sps->poc_cycle_length= get_ue_golomb(&s->gb);
5426 for(i=0; i<sps->poc_cycle_length; i++)
5427 sps->offset_for_ref_frame[i]= get_se_golomb(&s->gb);
5429 if(sps->poc_type > 2){
5430 av_log(h->s.avctx, AV_LOG_ERROR, "illegal POC type %d\n", sps->poc_type);
5434 sps->ref_frame_count= get_ue_golomb(&s->gb);
5435 if(sps->ref_frame_count > MAX_PICTURE_COUNT-2){
5436 av_log(h->s.avctx, AV_LOG_ERROR, "too many reference frames\n");
5438 sps->gaps_in_frame_num_allowed_flag= get_bits1(&s->gb);
5439 sps->mb_width= get_ue_golomb(&s->gb) + 1;
5440 sps->mb_height= get_ue_golomb(&s->gb) + 1;
5441 sps->frame_mbs_only_flag= get_bits1(&s->gb);
5442 if(!sps->frame_mbs_only_flag)
5443 sps->mb_aff= get_bits1(&s->gb);
5447 sps->direct_8x8_inference_flag= get_bits1(&s->gb);
5449 sps->crop= get_bits1(&s->gb);
5451 sps->crop_left = get_ue_golomb(&s->gb);
5452 sps->crop_right = get_ue_golomb(&s->gb);
5453 sps->crop_top = get_ue_golomb(&s->gb);
5454 sps->crop_bottom= get_ue_golomb(&s->gb);
5455 if(sps->crop_left || sps->crop_top){
5456 av_log(h->s.avctx, AV_LOG_ERROR, "insane cropping not completly supported, this could look slightly wrong ...\n");
5462 sps->crop_bottom= 0;
5465 sps->vui_parameters_present_flag= get_bits1(&s->gb);
5466 if( sps->vui_parameters_present_flag )
5467 decode_vui_parameters(h, sps);
5469 if(s->avctx->debug&FF_DEBUG_PICT_INFO){
5470 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",
5471 sps_id, sps->profile_idc, sps->level_idc,
5473 sps->ref_frame_count,
5474 sps->mb_width, sps->mb_height,
5475 sps->frame_mbs_only_flag ? "FRM" : (sps->mb_aff ? "MB-AFF" : "PIC-AFF"),
5476 sps->direct_8x8_inference_flag ? "8B8" : "",
5477 sps->crop_left, sps->crop_right,
5478 sps->crop_top, sps->crop_bottom,
5479 sps->vui_parameters_present_flag ? "VUI" : ""
5485 static inline int decode_picture_parameter_set(H264Context *h){
5486 MpegEncContext * const s = &h->s;
5487 int pps_id= get_ue_golomb(&s->gb);
5488 PPS *pps= &h->pps_buffer[pps_id];
5490 pps->sps_id= get_ue_golomb(&s->gb);
5491 pps->cabac= get_bits1(&s->gb);
5492 pps->pic_order_present= get_bits1(&s->gb);
5493 pps->slice_group_count= get_ue_golomb(&s->gb) + 1;
5494 if(pps->slice_group_count > 1 ){
5495 pps->mb_slice_group_map_type= get_ue_golomb(&s->gb);
5496 av_log(h->s.avctx, AV_LOG_ERROR, "FMO not supported\n");
5497 switch(pps->mb_slice_group_map_type){
5500 | for( i = 0; i <= num_slice_groups_minus1; i++ ) | | |
5501 | run_length[ i ] |1 |ue(v) |
5506 | for( i = 0; i < num_slice_groups_minus1; i++ ) | | |
5508 | top_left_mb[ i ] |1 |ue(v) |
5509 | bottom_right_mb[ i ] |1 |ue(v) |
5517 | slice_group_change_direction_flag |1 |u(1) |
5518 | slice_group_change_rate_minus1 |1 |ue(v) |
5523 | slice_group_id_cnt_minus1 |1 |ue(v) |
5524 | for( i = 0; i <= slice_group_id_cnt_minus1; i++ | | |
5526 | slice_group_id[ i ] |1 |u(v) |
5531 pps->ref_count[0]= get_ue_golomb(&s->gb) + 1;
5532 pps->ref_count[1]= get_ue_golomb(&s->gb) + 1;
5533 if(pps->ref_count[0] > 32 || pps->ref_count[1] > 32){
5534 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow (pps)\n");
5538 pps->weighted_pred= get_bits1(&s->gb);
5539 pps->weighted_bipred_idc= get_bits(&s->gb, 2);
5540 pps->init_qp= get_se_golomb(&s->gb) + 26;
5541 pps->init_qs= get_se_golomb(&s->gb) + 26;
5542 pps->chroma_qp_index_offset= get_se_golomb(&s->gb);
5543 pps->deblocking_filter_parameters_present= get_bits1(&s->gb);
5544 pps->constrained_intra_pred= get_bits1(&s->gb);
5545 pps->redundant_pic_cnt_present = get_bits1(&s->gb);
5547 if(s->avctx->debug&FF_DEBUG_PICT_INFO){
5548 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",
5549 pps_id, pps->sps_id,
5550 pps->cabac ? "CABAC" : "CAVLC",
5551 pps->slice_group_count,
5552 pps->ref_count[0], pps->ref_count[1],
5553 pps->weighted_pred ? "weighted" : "",
5554 pps->init_qp, pps->init_qs, pps->chroma_qp_index_offset,
5555 pps->deblocking_filter_parameters_present ? "LPAR" : "",
5556 pps->constrained_intra_pred ? "CONSTR" : "",
5557 pps->redundant_pic_cnt_present ? "REDU" : ""
5565 * finds the end of the current frame in the bitstream.
5566 * @return the position of the first byte of the next frame, or -1
5568 static int find_frame_end(ParseContext *pc, const uint8_t *buf, int buf_size){
5571 //printf("first %02X%02X%02X%02X\n", buf[0], buf[1],buf[2],buf[3]);
5572 // mb_addr= pc->mb_addr - 1;
5574 //FIXME this will fail with slices
5575 for(i=0; i<buf_size; i++){
5576 state= (state<<8) | buf[i];
5577 if((state&0xFFFFFF1F) == 0x101 || (state&0xFFFFFF1F) == 0x102 || (state&0xFFFFFF1F) == 0x105){
5578 if(pc->frame_start_found){
5580 pc->frame_start_found= 0;
5583 pc->frame_start_found= 1;
5588 return END_NOT_FOUND;
5591 static int h264_parse(AVCodecParserContext *s,
5592 AVCodecContext *avctx,
5593 uint8_t **poutbuf, int *poutbuf_size,
5594 const uint8_t *buf, int buf_size)
5596 ParseContext *pc = s->priv_data;
5599 next= find_frame_end(pc, buf, buf_size);
5601 if (ff_combine_frame(pc, next, (uint8_t **)&buf, &buf_size) < 0) {
5607 *poutbuf = (uint8_t *)buf;
5608 *poutbuf_size = buf_size;
5612 static int decode_nal_units(H264Context *h, uint8_t *buf, int buf_size){
5613 MpegEncContext * const s = &h->s;
5614 AVCodecContext * const avctx= s->avctx;
5618 for(i=0; i<32; i++){
5619 printf("%X ", buf[i]);
5630 if(buf_index >= buf_size) break;
5632 for(i = 0; i < h->nal_length_size; i++)
5633 nalsize = (nalsize << 8) | buf[buf_index++];
5635 // start code prefix search
5636 for(; buf_index + 3 < buf_size; buf_index++){
5637 // this should allways succeed in the first iteration
5638 if(buf[buf_index] == 0 && buf[buf_index+1] == 0 && buf[buf_index+2] == 1)
5642 if(buf_index+3 >= buf_size) break;
5647 ptr= decode_nal(h, buf + buf_index, &dst_length, &consumed, buf_size - buf_index);
5648 if(ptr[dst_length - 1] == 0) dst_length--;
5649 bit_length= 8*dst_length - decode_rbsp_trailing(ptr + dst_length - 1);
5651 if(s->avctx->debug&FF_DEBUG_STARTCODE){
5652 av_log(h->s.avctx, AV_LOG_DEBUG, "NAL %d at %d length %d\n", h->nal_unit_type, buf_index, dst_length);
5655 if (h->is_avc && (nalsize != consumed))
5656 av_log(h->s.avctx, AV_LOG_ERROR, "AVC: Consumed only %d bytes instead of %d\n", consumed, nalsize);
5658 buf_index += consumed;
5660 if( s->hurry_up == 1 && h->nal_ref_idc == 0 )
5663 switch(h->nal_unit_type){
5665 idr(h); //FIXME ensure we dont loose some frames if there is reordering
5667 init_get_bits(&s->gb, ptr, bit_length);
5669 h->inter_gb_ptr= &s->gb;
5670 s->data_partitioning = 0;
5672 if(decode_slice_header(h) < 0) return -1;
5673 if(h->redundant_pic_count==0 && s->hurry_up < 5 )
5677 init_get_bits(&s->gb, ptr, bit_length);
5679 h->inter_gb_ptr= NULL;
5680 s->data_partitioning = 1;
5682 if(decode_slice_header(h) < 0) return -1;
5685 init_get_bits(&h->intra_gb, ptr, bit_length);
5686 h->intra_gb_ptr= &h->intra_gb;
5689 init_get_bits(&h->inter_gb, ptr, bit_length);
5690 h->inter_gb_ptr= &h->inter_gb;
5692 if(h->redundant_pic_count==0 && h->intra_gb_ptr && s->data_partitioning && s->hurry_up < 5 )
5698 init_get_bits(&s->gb, ptr, bit_length);
5699 decode_seq_parameter_set(h);
5701 if(s->flags& CODEC_FLAG_LOW_DELAY)
5704 avctx->has_b_frames= !s->low_delay;
5707 init_get_bits(&s->gb, ptr, bit_length);
5709 decode_picture_parameter_set(h);
5712 case NAL_PICTURE_DELIMITER:
5714 case NAL_FILTER_DATA:
5717 av_log(avctx, AV_LOG_ERROR, "Unknown NAL code: %d\n", h->nal_unit_type);
5720 //FIXME move after where irt is set
5721 s->current_picture.pict_type= s->pict_type;
5722 s->current_picture.key_frame= s->pict_type == I_TYPE;
5725 if(!s->current_picture_ptr) return buf_index; //no frame
5727 h->prev_frame_num_offset= h->frame_num_offset;
5728 h->prev_frame_num= h->frame_num;
5729 if(s->current_picture_ptr->reference){
5730 h->prev_poc_msb= h->poc_msb;
5731 h->prev_poc_lsb= h->poc_lsb;
5733 if(s->current_picture_ptr->reference)
5734 execute_ref_pic_marking(h, h->mmco, h->mmco_index);
5736 assert(h->mmco_index==0);
5746 * retunrs the number of bytes consumed for building the current frame
5748 static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size){
5749 if(s->flags&CODEC_FLAG_TRUNCATED){
5750 pos -= s->parse_context.last_index;
5751 if(pos<0) pos=0; // FIXME remove (uneeded?)
5755 if(pos==0) pos=1; //avoid infinite loops (i doubt thats needed but ...)
5756 if(pos+10>buf_size) pos=buf_size; // oops ;)
5762 static int decode_frame(AVCodecContext *avctx,
5763 void *data, int *data_size,
5764 uint8_t *buf, int buf_size)
5766 H264Context *h = avctx->priv_data;
5767 MpegEncContext *s = &h->s;
5768 AVFrame *pict = data;
5771 s->flags= avctx->flags;
5772 s->flags2= avctx->flags2;
5774 /* no supplementary picture */
5775 if (buf_size == 0) {
5779 if(s->flags&CODEC_FLAG_TRUNCATED){
5780 int next= find_frame_end(&s->parse_context, buf, buf_size);
5782 if( ff_combine_frame(&s->parse_context, next, &buf, &buf_size) < 0 )
5784 //printf("next:%d buf_size:%d last_index:%d\n", next, buf_size, s->parse_context.last_index);
5787 if(h->is_avc && !h->got_avcC) {
5788 int i, cnt, nalsize;
5789 unsigned char *p = avctx->extradata;
5790 if(avctx->extradata_size < 7) {
5791 av_log(avctx, AV_LOG_ERROR, "avcC too short\n");
5795 av_log(avctx, AV_LOG_ERROR, "Unknown avcC version %d\n", *p);
5798 /* sps and pps in the avcC always have length coded with 2 bytes,
5799 so put a fake nal_length_size = 2 while parsing them */
5800 h->nal_length_size = 2;
5801 // Decode sps from avcC
5802 cnt = *(p+5) & 0x1f; // Number of sps
5804 for (i = 0; i < cnt; i++) {
5805 nalsize = BE_16(p) + 2;
5806 if(decode_nal_units(h, p, nalsize) != nalsize) {
5807 av_log(avctx, AV_LOG_ERROR, "Decoding sps %d from avcC failed\n", i);
5812 // Decode pps from avcC
5813 cnt = *(p++); // Number of pps
5814 for (i = 0; i < cnt; i++) {
5815 nalsize = BE_16(p) + 2;
5816 if(decode_nal_units(h, p, nalsize) != nalsize) {
5817 av_log(avctx, AV_LOG_ERROR, "Decoding pps %d from avcC failed\n", i);
5822 // Now store right nal length size, that will be use to parse all other nals
5823 h->nal_length_size = ((*(((char*)(avctx->extradata))+4))&0x03)+1;
5824 // Do not reparse avcC
5828 if(!h->is_avc && s->avctx->extradata_size && s->picture_number==0){
5829 if(0 < decode_nal_units(h, s->avctx->extradata, s->avctx->extradata_size) )
5833 buf_index=decode_nal_units(h, buf, buf_size);
5837 //FIXME do something with unavailable reference frames
5839 // if(ret==FRAME_SKIPED) return get_consumed_bytes(s, buf_index, buf_size);
5841 if(s->pict_type==B_TYPE || s->low_delay){
5842 *pict= *(AVFrame*)&s->current_picture;
5844 *pict= *(AVFrame*)&s->last_picture;
5847 if(!s->current_picture_ptr){
5848 av_log(h->s.avctx, AV_LOG_DEBUG, "error, NO frame\n");
5852 *pict= *(AVFrame*)&s->current_picture; //FIXME
5853 ff_print_debug_info(s, pict);
5854 assert(pict->data[0]);
5855 //printf("out %d\n", (int)pict->data[0]);
5858 /* Return the Picture timestamp as the frame number */
5859 /* we substract 1 because it is added on utils.c */
5860 avctx->frame_number = s->picture_number - 1;
5863 /* dont output the last pic after seeking */
5864 if(s->last_picture_ptr || s->low_delay)
5865 //Note this isnt a issue as a IDR pic should flush teh buffers
5867 *data_size = sizeof(AVFrame);
5868 return get_consumed_bytes(s, buf_index, buf_size);
5871 static inline void fill_mb_avail(H264Context *h){
5872 MpegEncContext * const s = &h->s;
5873 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
5876 h->mb_avail[0]= s->mb_x && h->slice_table[mb_xy - s->mb_stride - 1] == h->slice_num;
5877 h->mb_avail[1]= h->slice_table[mb_xy - s->mb_stride ] == h->slice_num;
5878 h->mb_avail[2]= s->mb_x+1 < s->mb_width && h->slice_table[mb_xy - s->mb_stride + 1] == h->slice_num;
5884 h->mb_avail[3]= s->mb_x && h->slice_table[mb_xy - 1] == h->slice_num;
5885 h->mb_avail[4]= 1; //FIXME move out
5886 h->mb_avail[5]= 0; //FIXME move out
5892 #define SIZE (COUNT*40)
5898 // int int_temp[10000];
5900 AVCodecContext avctx;
5902 dsputil_init(&dsp, &avctx);
5904 init_put_bits(&pb, temp, SIZE);
5905 printf("testing unsigned exp golomb\n");
5906 for(i=0; i<COUNT; i++){
5908 set_ue_golomb(&pb, i);
5909 STOP_TIMER("set_ue_golomb");
5911 flush_put_bits(&pb);
5913 init_get_bits(&gb, temp, 8*SIZE);
5914 for(i=0; i<COUNT; i++){
5917 s= show_bits(&gb, 24);
5920 j= get_ue_golomb(&gb);
5922 printf("missmatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
5925 STOP_TIMER("get_ue_golomb");
5929 init_put_bits(&pb, temp, SIZE);
5930 printf("testing signed exp golomb\n");
5931 for(i=0; i<COUNT; i++){
5933 set_se_golomb(&pb, i - COUNT/2);
5934 STOP_TIMER("set_se_golomb");
5936 flush_put_bits(&pb);
5938 init_get_bits(&gb, temp, 8*SIZE);
5939 for(i=0; i<COUNT; i++){
5942 s= show_bits(&gb, 24);
5945 j= get_se_golomb(&gb);
5946 if(j != i - COUNT/2){
5947 printf("missmatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
5950 STOP_TIMER("get_se_golomb");
5953 printf("testing 4x4 (I)DCT\n");
5956 uint8_t src[16], ref[16];
5957 uint64_t error= 0, max_error=0;
5959 for(i=0; i<COUNT; i++){
5961 // printf("%d %d %d\n", r1, r2, (r2-r1)*16);
5962 for(j=0; j<16; j++){
5963 ref[j]= random()%255;
5964 src[j]= random()%255;
5967 h264_diff_dct_c(block, src, ref, 4);
5970 for(j=0; j<16; j++){
5971 // printf("%d ", block[j]);
5972 block[j]= block[j]*4;
5973 if(j&1) block[j]= (block[j]*4 + 2)/5;
5974 if(j&4) block[j]= (block[j]*4 + 2)/5;
5978 h264_add_idct_c(ref, block, 4);
5979 /* for(j=0; j<16; j++){
5980 printf("%d ", ref[j]);
5984 for(j=0; j<16; j++){
5985 int diff= ABS(src[j] - ref[j]);
5988 max_error= FFMAX(max_error, diff);
5991 printf("error=%f max_error=%d\n", ((float)error)/COUNT/16, (int)max_error );
5993 printf("testing quantizer\n");
5994 for(qp=0; qp<52; qp++){
5996 src1_block[i]= src2_block[i]= random()%255;
6000 printf("Testing NAL layer\n");
6002 uint8_t bitstream[COUNT];
6003 uint8_t nal[COUNT*2];
6005 memset(&h, 0, sizeof(H264Context));
6007 for(i=0; i<COUNT; i++){
6015 for(j=0; j<COUNT; j++){
6016 bitstream[j]= (random() % 255) + 1;
6019 for(j=0; j<zeros; j++){
6020 int pos= random() % COUNT;
6021 while(bitstream[pos] == 0){
6030 nal_length= encode_nal(&h, nal, bitstream, COUNT, COUNT*2);
6032 printf("encoding failed\n");
6036 out= decode_nal(&h, nal, &out_length, &consumed, nal_length);
6040 if(out_length != COUNT){
6041 printf("incorrect length %d %d\n", out_length, COUNT);
6045 if(consumed != nal_length){
6046 printf("incorrect consumed length %d %d\n", nal_length, consumed);
6050 if(memcmp(bitstream, out, COUNT)){
6051 printf("missmatch\n");
6056 printf("Testing RBSP\n");
6064 static int decode_end(AVCodecContext *avctx)
6066 H264Context *h = avctx->priv_data;
6067 MpegEncContext *s = &h->s;
6069 free_tables(h); //FIXME cleanup init stuff perhaps
6072 // memset(h, 0, sizeof(H264Context));
6078 AVCodec h264_decoder = {
6082 sizeof(H264Context),
6087 /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 | CODEC_CAP_TRUNCATED,
6090 AVCodecParser h264_parser = {
6092 sizeof(ParseContext),