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?
206 int b_stride; //FIXME use s->b4_stride
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;
273 int dist_scale_factor[16];
276 * num_ref_idx_l0/1_active_minus1 + 1
278 int ref_count[2];// FIXME split for AFF
279 Picture *short_ref[16];
280 Picture *long_ref[16];
281 Picture default_ref_list[2][32];
282 Picture ref_list[2][32]; //FIXME size?
283 Picture field_ref_list[2][32]; //FIXME size?
286 * memory management control operations buffer.
288 MMCO mmco[MAX_MMCO_COUNT];
291 int long_ref_count; ///< number of actual long term references
292 int short_ref_count; ///< number of actual short term references
295 GetBitContext intra_gb;
296 GetBitContext inter_gb;
297 GetBitContext *intra_gb_ptr;
298 GetBitContext *inter_gb_ptr;
300 DCTELEM mb[16*24] __align8;
306 uint8_t cabac_state[399];
309 /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0,1,2), 0x0? luma_cbp */
313 /* chroma_pred_mode for i4x4 or i16x16, else 0 */
314 uint8_t *chroma_pred_mode_table;
315 int last_qscale_diff;
316 int16_t (*mvd_table[2])[2];
317 int16_t mvd_cache[2][5*8][2];
318 uint8_t *direct_table;
319 uint8_t direct_cache[5*8];
323 static VLC coeff_token_vlc[4];
324 static VLC chroma_dc_coeff_token_vlc;
326 static VLC total_zeros_vlc[15];
327 static VLC chroma_dc_total_zeros_vlc[3];
329 static VLC run_vlc[6];
332 static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
333 static void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
334 static void filter_mb( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr);
336 static inline uint32_t pack16to32(int a, int b){
337 #ifdef WORDS_BIGENDIAN
338 return (b&0xFFFF) + (a<<16);
340 return (a&0xFFFF) + (b<<16);
346 * @param h height of the rectangle, should be a constant
347 * @param w width of the rectangle, should be a constant
348 * @param size the size of val (1 or 4), should be a constant
350 static inline void fill_rectangle(void *vp, int w, int h, int stride, uint32_t val, int size){ //FIXME ensure this IS inlined
351 uint8_t *p= (uint8_t*)vp;
352 assert(size==1 || size==4);
357 //FIXME check what gcc generates for 64 bit on x86 and possible write a 32 bit ver of it
360 *(uint16_t*)(p + stride)= size==4 ? val : val*0x0101;
361 }else if(w==2 && h==4){
362 *(uint16_t*)(p + 0*stride)=
363 *(uint16_t*)(p + 1*stride)=
364 *(uint16_t*)(p + 2*stride)=
365 *(uint16_t*)(p + 3*stride)= size==4 ? val : val*0x0101;
366 }else if(w==4 && h==1){
367 *(uint32_t*)(p + 0*stride)= size==4 ? val : val*0x01010101;
368 }else if(w==4 && h==2){
369 *(uint32_t*)(p + 0*stride)=
370 *(uint32_t*)(p + 1*stride)= size==4 ? val : val*0x01010101;
371 }else if(w==4 && h==4){
372 *(uint32_t*)(p + 0*stride)=
373 *(uint32_t*)(p + 1*stride)=
374 *(uint32_t*)(p + 2*stride)=
375 *(uint32_t*)(p + 3*stride)= size==4 ? val : val*0x01010101;
376 }else if(w==8 && h==1){
378 *(uint32_t*)(p + 4)= size==4 ? val : val*0x01010101;
379 }else if(w==8 && h==2){
380 *(uint32_t*)(p + 0 + 0*stride)=
381 *(uint32_t*)(p + 4 + 0*stride)=
382 *(uint32_t*)(p + 0 + 1*stride)=
383 *(uint32_t*)(p + 4 + 1*stride)= size==4 ? val : val*0x01010101;
384 }else if(w==8 && h==4){
385 *(uint64_t*)(p + 0*stride)=
386 *(uint64_t*)(p + 1*stride)=
387 *(uint64_t*)(p + 2*stride)=
388 *(uint64_t*)(p + 3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
389 }else if(w==16 && h==2){
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)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
394 }else if(w==16 && h==4){
395 *(uint64_t*)(p + 0+0*stride)=
396 *(uint64_t*)(p + 8+0*stride)=
397 *(uint64_t*)(p + 0+1*stride)=
398 *(uint64_t*)(p + 8+1*stride)=
399 *(uint64_t*)(p + 0+2*stride)=
400 *(uint64_t*)(p + 8+2*stride)=
401 *(uint64_t*)(p + 0+3*stride)=
402 *(uint64_t*)(p + 8+3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
407 static inline void fill_caches(H264Context *h, int mb_type){
408 MpegEncContext * const s = &h->s;
409 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
410 int topleft_xy, top_xy, topright_xy, left_xy[2];
411 int topleft_type, top_type, topright_type, left_type[2];
415 //wow what a mess, why didnt they simplify the interlacing&intra stuff, i cant imagine that these complex rules are worth it
419 topleft_xy = 0; /* avoid warning */
420 top_xy = 0; /* avoid warning */
421 topright_xy = 0; /* avoid warning */
423 topleft_xy = mb_xy-1 - s->mb_stride;
424 top_xy = mb_xy - s->mb_stride;
425 topright_xy= mb_xy+1 - s->mb_stride;
426 left_xy[0] = mb_xy-1;
427 left_xy[1] = mb_xy-1;
434 topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
435 top_type = h->slice_table[top_xy ] == h->slice_num ? s->current_picture.mb_type[top_xy] : 0;
436 topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
437 left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
438 left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
440 if(IS_INTRA(mb_type)){
441 h->topleft_samples_available=
442 h->top_samples_available=
443 h->left_samples_available= 0xFFFF;
444 h->topright_samples_available= 0xEEEA;
446 if(!IS_INTRA(top_type) && (top_type==0 || h->pps.constrained_intra_pred)){
447 h->topleft_samples_available= 0xB3FF;
448 h->top_samples_available= 0x33FF;
449 h->topright_samples_available= 0x26EA;
452 if(!IS_INTRA(left_type[i]) && (left_type[i]==0 || h->pps.constrained_intra_pred)){
453 h->topleft_samples_available&= 0xDF5F;
454 h->left_samples_available&= 0x5F5F;
458 if(!IS_INTRA(topleft_type) && (topleft_type==0 || h->pps.constrained_intra_pred))
459 h->topleft_samples_available&= 0x7FFF;
461 if(!IS_INTRA(topright_type) && (topright_type==0 || h->pps.constrained_intra_pred))
462 h->topright_samples_available&= 0xFBFF;
464 if(IS_INTRA4x4(mb_type)){
465 if(IS_INTRA4x4(top_type)){
466 h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
467 h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
468 h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
469 h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
472 if(IS_INTRA16x16(top_type) || (IS_INTER(top_type) && !h->pps.constrained_intra_pred))
477 h->intra4x4_pred_mode_cache[4+8*0]=
478 h->intra4x4_pred_mode_cache[5+8*0]=
479 h->intra4x4_pred_mode_cache[6+8*0]=
480 h->intra4x4_pred_mode_cache[7+8*0]= pred;
483 if(IS_INTRA4x4(left_type[i])){
484 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
485 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
488 if(IS_INTRA16x16(left_type[i]) || (IS_INTER(left_type[i]) && !h->pps.constrained_intra_pred))
493 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
494 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
509 //FIXME constraint_intra_pred & partitioning & nnz (lets hope this is just a typo in the spec)
511 h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][0];
512 h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][1];
513 h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][2];
514 h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3];
516 h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][7];
517 h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8];
519 h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][10];
520 h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11];
522 h->top_cbp= h->cbp_table[top_xy];
524 h->non_zero_count_cache[4+8*0]=
525 h->non_zero_count_cache[5+8*0]=
526 h->non_zero_count_cache[6+8*0]=
527 h->non_zero_count_cache[7+8*0]=
529 h->non_zero_count_cache[1+8*0]=
530 h->non_zero_count_cache[2+8*0]=
532 h->non_zero_count_cache[1+8*3]=
533 h->non_zero_count_cache[2+8*3]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
535 if(IS_INTRA(mb_type)) h->top_cbp= 0x1C0;
540 h->non_zero_count_cache[3+8*1]= h->non_zero_count[left_xy[0]][6];
541 h->non_zero_count_cache[3+8*2]= h->non_zero_count[left_xy[0]][5];
542 h->non_zero_count_cache[0+8*1]= h->non_zero_count[left_xy[0]][9]; //FIXME left_block
543 h->non_zero_count_cache[0+8*4]= h->non_zero_count[left_xy[0]][12];
544 h->left_cbp= h->cbp_table[left_xy[0]]; //FIXME interlacing
546 h->non_zero_count_cache[3+8*1]=
547 h->non_zero_count_cache[3+8*2]=
548 h->non_zero_count_cache[0+8*1]=
549 h->non_zero_count_cache[0+8*4]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
551 if(IS_INTRA(mb_type)) h->left_cbp= 0x1C0;//FIXME interlacing
556 h->non_zero_count_cache[3+8*3]= h->non_zero_count[left_xy[1]][4];
557 h->non_zero_count_cache[3+8*4]= h->non_zero_count[left_xy[1]][3];
558 h->non_zero_count_cache[0+8*2]= h->non_zero_count[left_xy[1]][8];
559 h->non_zero_count_cache[0+8*5]= h->non_zero_count[left_xy[1]][11];
561 h->non_zero_count_cache[3+8*3]=
562 h->non_zero_count_cache[3+8*4]=
563 h->non_zero_count_cache[0+8*2]=
564 h->non_zero_count_cache[0+8*5]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
568 //FIXME direct mb can skip much of this
569 if(IS_INTER(mb_type) || (IS_DIRECT(mb_type) && h->direct_spatial_mv_pred)){
571 for(list=0; list<2; list++){
572 if((!IS_8X8(mb_type)) && !USES_LIST(mb_type, list) && !IS_DIRECT(mb_type)){
573 /*if(!h->mv_cache_clean[list]){
574 memset(h->mv_cache [list], 0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
575 memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
576 h->mv_cache_clean[list]= 1;
580 h->mv_cache_clean[list]= 0;
582 if(IS_INTER(topleft_type)){
583 const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
584 const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + h->b8_stride;
585 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
586 h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
588 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
589 h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
592 if(IS_INTER(top_type)){
593 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
594 const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
595 *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0];
596 *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];
597 *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];
598 *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];
599 h->ref_cache[list][scan8[0] + 0 - 1*8]=
600 h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
601 h->ref_cache[list][scan8[0] + 2 - 1*8]=
602 h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
604 *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]=
605 *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]=
606 *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]=
607 *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;
608 *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
611 if(IS_INTER(topright_type)){
612 const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
613 const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
614 *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
615 h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
617 *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
618 h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
621 //FIXME unify cleanup or sth
622 if(IS_INTER(left_type[0])){
623 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
624 const int b8_xy= h->mb2b8_xy[left_xy[0]] + 1;
625 *(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]];
626 *(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]];
627 h->ref_cache[list][scan8[0] - 1 + 0*8]=
628 h->ref_cache[list][scan8[0] - 1 + 1*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0]>>1)];
630 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 0*8]=
631 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 1*8]= 0;
632 h->ref_cache[list][scan8[0] - 1 + 0*8]=
633 h->ref_cache[list][scan8[0] - 1 + 1*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
636 if(IS_INTER(left_type[1])){
637 const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
638 const int b8_xy= h->mb2b8_xy[left_xy[1]] + 1;
639 *(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]];
640 *(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]];
641 h->ref_cache[list][scan8[0] - 1 + 2*8]=
642 h->ref_cache[list][scan8[0] - 1 + 3*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[2]>>1)];
644 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 2*8]=
645 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 3*8]= 0;
646 h->ref_cache[list][scan8[0] - 1 + 2*8]=
647 h->ref_cache[list][scan8[0] - 1 + 3*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
650 h->ref_cache[list][scan8[5 ]+1] =
651 h->ref_cache[list][scan8[7 ]+1] =
652 h->ref_cache[list][scan8[13]+1] = //FIXME remove past 3 (init somewher else)
653 h->ref_cache[list][scan8[4 ]] =
654 h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
655 *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
656 *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
657 *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewher else)
658 *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
659 *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
662 /* XXX beurk, Load mvd */
663 if(IS_INTER(topleft_type)){
664 const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
665 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy];
667 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 - 1*8]= 0;
670 if(IS_INTER(top_type)){
671 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
672 *(uint32_t*)h->mvd_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 0];
673 *(uint32_t*)h->mvd_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 1];
674 *(uint32_t*)h->mvd_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 2];
675 *(uint32_t*)h->mvd_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 3];
677 *(uint32_t*)h->mvd_cache [list][scan8[0] + 0 - 1*8]=
678 *(uint32_t*)h->mvd_cache [list][scan8[0] + 1 - 1*8]=
679 *(uint32_t*)h->mvd_cache [list][scan8[0] + 2 - 1*8]=
680 *(uint32_t*)h->mvd_cache [list][scan8[0] + 3 - 1*8]= 0;
682 if(IS_INTER(left_type[0])){
683 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
684 *(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]];
685 *(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]];
687 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 0*8]=
688 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 1*8]= 0;
690 if(IS_INTER(left_type[1])){
691 const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
692 *(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]];
693 *(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]];
695 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 2*8]=
696 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 3*8]= 0;
698 *(uint32_t*)h->mvd_cache [list][scan8[5 ]+1]=
699 *(uint32_t*)h->mvd_cache [list][scan8[7 ]+1]=
700 *(uint32_t*)h->mvd_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewher else)
701 *(uint32_t*)h->mvd_cache [list][scan8[4 ]]=
702 *(uint32_t*)h->mvd_cache [list][scan8[12]]= 0;
704 if(h->slice_type == B_TYPE){
705 fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, 0, 1);
707 if(IS_DIRECT(top_type)){
708 *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0x01010101;
709 }else if(IS_8X8(top_type)){
710 int b8_xy = h->mb2b8_xy[top_xy] + h->b8_stride;
711 h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy];
712 h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 1];
714 *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0;
718 if(IS_DIRECT(left_type[0])){
719 h->direct_cache[scan8[0] - 1 + 0*8]=
720 h->direct_cache[scan8[0] - 1 + 2*8]= 1;
721 }else if(IS_8X8(left_type[0])){
722 int b8_xy = h->mb2b8_xy[left_xy[0]] + 1;
723 h->direct_cache[scan8[0] - 1 + 0*8]= h->direct_table[b8_xy];
724 h->direct_cache[scan8[0] - 1 + 2*8]= h->direct_table[b8_xy + h->b8_stride];
726 h->direct_cache[scan8[0] - 1 + 0*8]=
727 h->direct_cache[scan8[0] - 1 + 2*8]= 0;
736 static inline void write_back_intra_pred_mode(H264Context *h){
737 MpegEncContext * const s = &h->s;
738 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
740 h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
741 h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
742 h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
743 h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
744 h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
745 h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
746 h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
750 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
752 static inline int check_intra4x4_pred_mode(H264Context *h){
753 MpegEncContext * const s = &h->s;
754 static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
755 static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
758 if(!(h->top_samples_available&0x8000)){
760 int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
762 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);
765 h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
770 if(!(h->left_samples_available&0x8000)){
772 int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
774 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);
777 h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
783 } //FIXME cleanup like next
786 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
788 static inline int check_intra_pred_mode(H264Context *h, int mode){
789 MpegEncContext * const s = &h->s;
790 static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
791 static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
793 if(mode < 0 || mode > 6) {
794 av_log(h->s.avctx, AV_LOG_ERROR, "out of range intra chroma pred mode at %d %d\n", s->mb_x, s->mb_y);
798 if(!(h->top_samples_available&0x8000)){
801 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);
806 if(!(h->left_samples_available&0x8000)){
809 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);
818 * gets the predicted intra4x4 prediction mode.
820 static inline int pred_intra_mode(H264Context *h, int n){
821 const int index8= scan8[n];
822 const int left= h->intra4x4_pred_mode_cache[index8 - 1];
823 const int top = h->intra4x4_pred_mode_cache[index8 - 8];
824 const int min= FFMIN(left, top);
826 tprintf("mode:%d %d min:%d\n", left ,top, min);
828 if(min<0) return DC_PRED;
832 static inline void write_back_non_zero_count(H264Context *h){
833 MpegEncContext * const s = &h->s;
834 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
836 h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[4+8*4];
837 h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[5+8*4];
838 h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[6+8*4];
839 h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];
840 h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[7+8*3];
841 h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[7+8*2];
842 h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[7+8*1];
844 h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[1+8*2];
845 h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
846 h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[2+8*1];
848 h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[1+8*5];
849 h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
850 h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[2+8*4];
854 * gets the predicted number of non zero coefficients.
855 * @param n block index
857 static inline int pred_non_zero_count(H264Context *h, int n){
858 const int index8= scan8[n];
859 const int left= h->non_zero_count_cache[index8 - 1];
860 const int top = h->non_zero_count_cache[index8 - 8];
863 if(i<64) i= (i+1)>>1;
865 tprintf("pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
870 static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
871 const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];
873 if(topright_ref != PART_NOT_AVAILABLE){
874 *C= h->mv_cache[list][ i - 8 + part_width ];
877 tprintf("topright MV not available\n");
879 *C= h->mv_cache[list][ i - 8 - 1 ];
880 return h->ref_cache[list][ i - 8 - 1 ];
885 * gets the predicted MV.
886 * @param n the block index
887 * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)
888 * @param mx the x component of the predicted motion vector
889 * @param my the y component of the predicted motion vector
891 static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
892 const int index8= scan8[n];
893 const int top_ref= h->ref_cache[list][ index8 - 8 ];
894 const int left_ref= h->ref_cache[list][ index8 - 1 ];
895 const int16_t * const A= h->mv_cache[list][ index8 - 1 ];
896 const int16_t * const B= h->mv_cache[list][ index8 - 8 ];
898 int diagonal_ref, match_count;
900 assert(part_width==1 || part_width==2 || part_width==4);
910 diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);
911 match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);
912 if(match_count > 1){ //most common
913 *mx= mid_pred(A[0], B[0], C[0]);
914 *my= mid_pred(A[1], B[1], C[1]);
915 }else if(match_count==1){
919 }else if(top_ref==ref){
927 if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
931 *mx= mid_pred(A[0], B[0], C[0]);
932 *my= mid_pred(A[1], B[1], C[1]);
936 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);
940 * gets the directionally predicted 16x8 MV.
941 * @param n the block index
942 * @param mx the x component of the predicted motion vector
943 * @param my the y component of the predicted motion vector
945 static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
947 const int top_ref= h->ref_cache[list][ scan8[0] - 8 ];
948 const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
950 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);
958 const int left_ref= h->ref_cache[list][ scan8[8] - 1 ];
959 const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];
961 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);
971 pred_motion(h, n, 4, list, ref, mx, my);
975 * gets the directionally predicted 8x16 MV.
976 * @param n the block index
977 * @param mx the x component of the predicted motion vector
978 * @param my the y component of the predicted motion vector
980 static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
982 const int left_ref= h->ref_cache[list][ scan8[0] - 1 ];
983 const int16_t * const A= h->mv_cache[list][ scan8[0] - 1 ];
985 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);
996 diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);
998 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);
1000 if(diagonal_ref == ref){
1008 pred_motion(h, n, 2, list, ref, mx, my);
1011 static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
1012 const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
1013 const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];
1015 tprintf("pred_pskip: (%d) (%d) at %2d %2d\n", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
1017 if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
1018 || (top_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0)
1019 || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){
1025 pred_motion(h, 0, 4, 0, 0, mx, my);
1030 static inline void direct_dist_scale_factor(H264Context * const h){
1031 const int poc = h->s.current_picture_ptr->poc;
1032 const int poc1 = h->ref_list[1][0].poc;
1034 for(i=0; i<h->ref_count[0]; i++){
1035 int poc0 = h->ref_list[0][i].poc;
1036 int td = clip(poc1 - poc0, -128, 127);
1037 if(td == 0 /* FIXME || pic0 is a long-term ref */){
1038 h->dist_scale_factor[i] = 256;
1040 int tb = clip(poc - poc0, -128, 127);
1041 int tx = (16384 + (ABS(td) >> 1)) / td;
1042 h->dist_scale_factor[i] = clip((tb*tx + 32) >> 6, -1024, 1023);
1047 static inline void pred_direct_motion(H264Context * const h, int *mb_type){
1048 MpegEncContext * const s = &h->s;
1049 const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
1050 const int b8_xy = 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1051 const int b4_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1052 const int mb_type_col = h->ref_list[1][0].mb_type[mb_xy];
1053 const int16_t (*l1mv0)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[0][b4_xy];
1054 const int8_t *l1ref0 = &h->ref_list[1][0].ref_index[0][b8_xy];
1055 const int is_b8x8 = IS_8X8(*mb_type);
1059 if(IS_8X8(mb_type_col) && !h->sps.direct_8x8_inference_flag){
1060 /* FIXME save sub mb types from previous frames (or derive from MVs)
1061 * so we know exactly what block size to use */
1062 sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */
1063 *mb_type = MB_TYPE_8x8;
1064 }else if(!is_b8x8 && (IS_16X16(mb_type_col) || IS_INTRA(mb_type_col))){
1065 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1066 *mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */
1068 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1069 *mb_type = MB_TYPE_8x8;
1072 *mb_type |= MB_TYPE_DIRECT2;
1074 if(h->direct_spatial_mv_pred){
1079 /* ref = min(neighbors) */
1080 for(list=0; list<2; list++){
1081 int refa = h->ref_cache[list][scan8[0] - 1];
1082 int refb = h->ref_cache[list][scan8[0] - 8];
1083 int refc = h->ref_cache[list][scan8[0] - 8 + 4];
1085 refc = h->ref_cache[list][scan8[0] - 8 - 1];
1087 if(ref[list] < 0 || (refb < ref[list] && refb >= 0))
1089 if(ref[list] < 0 || (refc < ref[list] && refc >= 0))
1095 if(ref[0] < 0 && ref[1] < 0){
1096 ref[0] = ref[1] = 0;
1097 mv[0][0] = mv[0][1] =
1098 mv[1][0] = mv[1][1] = 0;
1100 for(list=0; list<2; list++){
1102 pred_motion(h, 0, 4, list, ref[list], &mv[list][0], &mv[list][1]);
1104 mv[list][0] = mv[list][1] = 0;
1109 *mb_type &= ~MB_TYPE_P0L1;
1110 sub_mb_type &= ~MB_TYPE_P0L1;
1111 }else if(ref[0] < 0){
1112 *mb_type &= ~MB_TYPE_P0L0;
1113 sub_mb_type &= ~MB_TYPE_P0L0;
1116 if(IS_16X16(*mb_type)){
1117 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref[0], 1);
1118 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, ref[1], 1);
1119 if(!IS_INTRA(mb_type_col) && l1ref0[0] == 0 &&
1120 ABS(l1mv0[0][0]) <= 1 && ABS(l1mv0[0][1]) <= 1){
1122 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1124 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
1126 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1128 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
1130 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1131 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1134 for(i8=0; i8<4; i8++){
1135 const int x8 = i8&1;
1136 const int y8 = i8>>1;
1138 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1140 h->sub_mb_type[i8] = sub_mb_type;
1142 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1143 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1144 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref[0], 1);
1145 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, ref[1], 1);
1148 if(!IS_INTRA(mb_type_col) && l1ref0[x8 + y8*h->b8_stride] == 0){
1149 for(i4=0; i4<4; i4++){
1150 const int16_t *mv_col = l1mv0[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1151 if(ABS(mv_col[0]) <= 1 && ABS(mv_col[1]) <= 1){
1153 *(uint32_t*)h->mv_cache[0][scan8[i8*4+i4]] = 0;
1155 *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = 0;
1161 }else{ /* direct temporal mv pred */
1162 /* FIXME assumes that L1ref0 used the same ref lists as current frame */
1163 if(IS_16X16(*mb_type)){
1164 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1);
1165 if(IS_INTRA(mb_type_col)){
1166 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
1167 fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
1168 fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
1170 const int ref0 = l1ref0[0];
1171 const int dist_scale_factor = h->dist_scale_factor[ref0];
1172 const int16_t *mv_col = l1mv0[0];
1174 mv_l0[0] = (dist_scale_factor * mv_col[0] + 128) >> 8;
1175 mv_l0[1] = (dist_scale_factor * mv_col[1] + 128) >> 8;
1176 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref0, 1);
1177 fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv_l0[0],mv_l0[1]), 4);
1178 fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]), 4);
1181 for(i8=0; i8<4; i8++){
1182 const int x8 = i8&1;
1183 const int y8 = i8>>1;
1184 int ref0, dist_scale_factor;
1186 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1188 h->sub_mb_type[i8] = sub_mb_type;
1189 if(IS_INTRA(mb_type_col)){
1190 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1191 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1192 fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1193 fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1197 ref0 = l1ref0[x8 + y8*h->b8_stride];
1198 dist_scale_factor = h->dist_scale_factor[ref0];
1200 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1201 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1202 for(i4=0; i4<4; i4++){
1203 const int16_t *mv_col = l1mv0[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1204 int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]];
1205 mv_l0[0] = (dist_scale_factor * mv_col[0] + 128) >> 8;
1206 mv_l0[1] = (dist_scale_factor * mv_col[1] + 128) >> 8;
1207 *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] =
1208 pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
1215 static inline void write_back_motion(H264Context *h, int mb_type){
1216 MpegEncContext * const s = &h->s;
1217 const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1218 const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1221 for(list=0; list<2; list++){
1223 if((!IS_8X8(mb_type)) && !USES_LIST(mb_type, list)){
1224 if(1){ //FIXME skip or never read if mb_type doesnt use it
1226 *(uint64_t*)s->current_picture.motion_val[list][b_xy + 0 + y*h->b_stride]=
1227 *(uint64_t*)s->current_picture.motion_val[list][b_xy + 2 + y*h->b_stride]= 0;
1229 if( h->pps.cabac ) {
1230 /* FIXME needed ? */
1232 *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]=
1233 *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= 0;
1237 *(uint16_t*)&s->current_picture.ref_index[list][b8_xy + y*h->b8_stride]= (LIST_NOT_USED&0xFF)*0x0101;
1244 *(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];
1245 *(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];
1247 if( h->pps.cabac ) {
1249 *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+0 + 8*y];
1250 *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+2 + 8*y];
1254 s->current_picture.ref_index[list][b8_xy + 0 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+0 + 16*y];
1255 s->current_picture.ref_index[list][b8_xy + 1 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+2 + 16*y];
1259 if(h->slice_type == B_TYPE && h->pps.cabac){
1260 if(IS_8X8(mb_type)){
1261 h->direct_table[b8_xy+1+0*h->b8_stride] = IS_DIRECT(h->sub_mb_type[1]) ? 1 : 0;
1262 h->direct_table[b8_xy+0+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[2]) ? 1 : 0;
1263 h->direct_table[b8_xy+1+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[3]) ? 1 : 0;
1269 * Decodes a network abstraction layer unit.
1270 * @param consumed is the number of bytes used as input
1271 * @param length is the length of the array
1272 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp ttailing?
1273 * @returns decoded bytes, might be src+1 if no escapes
1275 static uint8_t *decode_nal(H264Context *h, uint8_t *src, int *dst_length, int *consumed, int length){
1279 // src[0]&0x80; //forbidden bit
1280 h->nal_ref_idc= src[0]>>5;
1281 h->nal_unit_type= src[0]&0x1F;
1285 for(i=0; i<length; i++)
1286 printf("%2X ", src[i]);
1288 for(i=0; i+1<length; i+=2){
1289 if(src[i]) continue;
1290 if(i>0 && src[i-1]==0) i--;
1291 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1293 /* startcode, so we must be past the end */
1300 if(i>=length-1){ //no escaped 0
1301 *dst_length= length;
1302 *consumed= length+1; //+1 for the header
1306 h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length);
1307 dst= h->rbsp_buffer;
1309 //printf("deoding esc\n");
1312 //remove escapes (very rare 1:2^22)
1313 if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1314 if(src[si+2]==3){ //escape
1319 }else //next start code
1323 dst[di++]= src[si++];
1327 *consumed= si + 1;//+1 for the header
1328 //FIXME store exact number of bits in the getbitcontext (its needed for decoding)
1334 * @param src the data which should be escaped
1335 * @param dst the target buffer, dst+1 == src is allowed as a special case
1336 * @param length the length of the src data
1337 * @param dst_length the length of the dst array
1338 * @returns length of escaped data in bytes or -1 if an error occured
1340 static int encode_nal(H264Context *h, uint8_t *dst, uint8_t *src, int length, int dst_length){
1341 int i, escape_count, si, di;
1345 assert(dst_length>0);
1347 dst[0]= (h->nal_ref_idc<<5) + h->nal_unit_type;
1349 if(length==0) return 1;
1352 for(i=0; i<length; i+=2){
1353 if(src[i]) continue;
1354 if(i>0 && src[i-1]==0)
1356 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1362 if(escape_count==0){
1364 memcpy(dst+1, src, length);
1368 if(length + escape_count + 1> dst_length)
1371 //this should be damn rare (hopefully)
1373 h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length + escape_count);
1374 temp= h->rbsp_buffer;
1375 //printf("encoding esc\n");
1380 if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1381 temp[di++]= 0; si++;
1382 temp[di++]= 0; si++;
1384 temp[di++]= src[si++];
1387 temp[di++]= src[si++];
1389 memcpy(dst+1, temp, length+escape_count);
1391 assert(di == length+escape_count);
1397 * write 1,10,100,1000,... for alignment, yes its exactly inverse to mpeg4
1399 static void encode_rbsp_trailing(PutBitContext *pb){
1402 length= (-put_bits_count(pb))&7;
1403 if(length) put_bits(pb, length, 0);
1408 * identifies the exact end of the bitstream
1409 * @return the length of the trailing, or 0 if damaged
1411 static int decode_rbsp_trailing(uint8_t *src){
1415 tprintf("rbsp trailing %X\n", v);
1425 * idct tranforms the 16 dc values and dequantize them.
1426 * @param qp quantization parameter
1428 static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp){
1429 const int qmul= dequant_coeff[qp][0];
1432 int temp[16]; //FIXME check if this is a good idea
1433 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
1434 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1436 //memset(block, 64, 2*256);
1439 const int offset= y_offset[i];
1440 const int z0= block[offset+stride*0] + block[offset+stride*4];
1441 const int z1= block[offset+stride*0] - block[offset+stride*4];
1442 const int z2= block[offset+stride*1] - block[offset+stride*5];
1443 const int z3= block[offset+stride*1] + block[offset+stride*5];
1452 const int offset= x_offset[i];
1453 const int z0= temp[4*0+i] + temp[4*2+i];
1454 const int z1= temp[4*0+i] - temp[4*2+i];
1455 const int z2= temp[4*1+i] - temp[4*3+i];
1456 const int z3= temp[4*1+i] + temp[4*3+i];
1458 block[stride*0 +offset]= ((z0 + z3)*qmul + 2)>>2; //FIXME think about merging this into decode_resdual
1459 block[stride*2 +offset]= ((z1 + z2)*qmul + 2)>>2;
1460 block[stride*8 +offset]= ((z1 - z2)*qmul + 2)>>2;
1461 block[stride*10+offset]= ((z0 - z3)*qmul + 2)>>2;
1467 * dct tranforms the 16 dc values.
1468 * @param qp quantization parameter ??? FIXME
1470 static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
1471 // const int qmul= dequant_coeff[qp][0];
1473 int temp[16]; //FIXME check if this is a good idea
1474 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
1475 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1478 const int offset= y_offset[i];
1479 const int z0= block[offset+stride*0] + block[offset+stride*4];
1480 const int z1= block[offset+stride*0] - block[offset+stride*4];
1481 const int z2= block[offset+stride*1] - block[offset+stride*5];
1482 const int z3= block[offset+stride*1] + block[offset+stride*5];
1491 const int offset= x_offset[i];
1492 const int z0= temp[4*0+i] + temp[4*2+i];
1493 const int z1= temp[4*0+i] - temp[4*2+i];
1494 const int z2= temp[4*1+i] - temp[4*3+i];
1495 const int z3= temp[4*1+i] + temp[4*3+i];
1497 block[stride*0 +offset]= (z0 + z3)>>1;
1498 block[stride*2 +offset]= (z1 + z2)>>1;
1499 block[stride*8 +offset]= (z1 - z2)>>1;
1500 block[stride*10+offset]= (z0 - z3)>>1;
1508 static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp){
1509 const int qmul= dequant_coeff[qp][0];
1510 const int stride= 16*2;
1511 const int xStride= 16;
1514 a= block[stride*0 + xStride*0];
1515 b= block[stride*0 + xStride*1];
1516 c= block[stride*1 + xStride*0];
1517 d= block[stride*1 + xStride*1];
1524 block[stride*0 + xStride*0]= ((a+c)*qmul + 0)>>1;
1525 block[stride*0 + xStride*1]= ((e+b)*qmul + 0)>>1;
1526 block[stride*1 + xStride*0]= ((a-c)*qmul + 0)>>1;
1527 block[stride*1 + xStride*1]= ((e-b)*qmul + 0)>>1;
1531 static void chroma_dc_dct_c(DCTELEM *block){
1532 const int stride= 16*2;
1533 const int xStride= 16;
1536 a= block[stride*0 + xStride*0];
1537 b= block[stride*0 + xStride*1];
1538 c= block[stride*1 + xStride*0];
1539 d= block[stride*1 + xStride*1];
1546 block[stride*0 + xStride*0]= (a+c);
1547 block[stride*0 + xStride*1]= (e+b);
1548 block[stride*1 + xStride*0]= (a-c);
1549 block[stride*1 + xStride*1]= (e-b);
1554 * gets the chroma qp.
1556 static inline int get_chroma_qp(H264Context *h, int qscale){
1558 return chroma_qp[clip(qscale + h->pps.chroma_qp_index_offset, 0, 51)];
1563 static void h264_diff_dct_c(DCTELEM *block, uint8_t *src1, uint8_t *src2, int stride){
1565 //FIXME try int temp instead of block
1568 const int d0= src1[0 + i*stride] - src2[0 + i*stride];
1569 const int d1= src1[1 + i*stride] - src2[1 + i*stride];
1570 const int d2= src1[2 + i*stride] - src2[2 + i*stride];
1571 const int d3= src1[3 + i*stride] - src2[3 + i*stride];
1572 const int z0= d0 + d3;
1573 const int z3= d0 - d3;
1574 const int z1= d1 + d2;
1575 const int z2= d1 - d2;
1577 block[0 + 4*i]= z0 + z1;
1578 block[1 + 4*i]= 2*z3 + z2;
1579 block[2 + 4*i]= z0 - z1;
1580 block[3 + 4*i]= z3 - 2*z2;
1584 const int z0= block[0*4 + i] + block[3*4 + i];
1585 const int z3= block[0*4 + i] - block[3*4 + i];
1586 const int z1= block[1*4 + i] + block[2*4 + i];
1587 const int z2= block[1*4 + i] - block[2*4 + i];
1589 block[0*4 + i]= z0 + z1;
1590 block[1*4 + i]= 2*z3 + z2;
1591 block[2*4 + i]= z0 - z1;
1592 block[3*4 + i]= z3 - 2*z2;
1597 //FIXME need to check that this doesnt overflow signed 32 bit for low qp, iam not sure, its very close
1598 //FIXME check that gcc inlines this (and optimizes intra & seperate_dc stuff away)
1599 static inline int quantize_c(DCTELEM *block, uint8_t *scantable, int qscale, int intra, int seperate_dc){
1601 const int * const quant_table= quant_coeff[qscale];
1602 const int bias= intra ? (1<<QUANT_SHIFT)/3 : (1<<QUANT_SHIFT)/6;
1603 const unsigned int threshold1= (1<<QUANT_SHIFT) - bias - 1;
1604 const unsigned int threshold2= (threshold1<<1);
1610 const int dc_bias= intra ? (1<<(QUANT_SHIFT-2))/3 : (1<<(QUANT_SHIFT-2))/6;
1611 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT-2)) - dc_bias - 1;
1612 const unsigned int dc_threshold2= (dc_threshold1<<1);
1614 int level= block[0]*quant_coeff[qscale+18][0];
1615 if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1617 level= (dc_bias + level)>>(QUANT_SHIFT-2);
1620 level= (dc_bias - level)>>(QUANT_SHIFT-2);
1623 // last_non_zero = i;
1628 const int dc_bias= intra ? (1<<(QUANT_SHIFT+1))/3 : (1<<(QUANT_SHIFT+1))/6;
1629 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT+1)) - dc_bias - 1;
1630 const unsigned int dc_threshold2= (dc_threshold1<<1);
1632 int level= block[0]*quant_table[0];
1633 if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1635 level= (dc_bias + level)>>(QUANT_SHIFT+1);
1638 level= (dc_bias - level)>>(QUANT_SHIFT+1);
1641 // last_non_zero = i;
1654 const int j= scantable[i];
1655 int level= block[j]*quant_table[j];
1657 // if( bias+level >= (1<<(QMAT_SHIFT - 3))
1658 // || bias-level >= (1<<(QMAT_SHIFT - 3))){
1659 if(((unsigned)(level+threshold1))>threshold2){
1661 level= (bias + level)>>QUANT_SHIFT;
1664 level= (bias - level)>>QUANT_SHIFT;
1673 return last_non_zero;
1676 static void pred4x4_vertical_c(uint8_t *src, uint8_t *topright, int stride){
1677 const uint32_t a= ((uint32_t*)(src-stride))[0];
1678 ((uint32_t*)(src+0*stride))[0]= a;
1679 ((uint32_t*)(src+1*stride))[0]= a;
1680 ((uint32_t*)(src+2*stride))[0]= a;
1681 ((uint32_t*)(src+3*stride))[0]= a;
1684 static void pred4x4_horizontal_c(uint8_t *src, uint8_t *topright, int stride){
1685 ((uint32_t*)(src+0*stride))[0]= src[-1+0*stride]*0x01010101;
1686 ((uint32_t*)(src+1*stride))[0]= src[-1+1*stride]*0x01010101;
1687 ((uint32_t*)(src+2*stride))[0]= src[-1+2*stride]*0x01010101;
1688 ((uint32_t*)(src+3*stride))[0]= src[-1+3*stride]*0x01010101;
1691 static void pred4x4_dc_c(uint8_t *src, uint8_t *topright, int stride){
1692 const int dc= ( src[-stride] + src[1-stride] + src[2-stride] + src[3-stride]
1693 + src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 4) >>3;
1695 ((uint32_t*)(src+0*stride))[0]=
1696 ((uint32_t*)(src+1*stride))[0]=
1697 ((uint32_t*)(src+2*stride))[0]=
1698 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1701 static void pred4x4_left_dc_c(uint8_t *src, uint8_t *topright, int stride){
1702 const int dc= ( src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 2) >>2;
1704 ((uint32_t*)(src+0*stride))[0]=
1705 ((uint32_t*)(src+1*stride))[0]=
1706 ((uint32_t*)(src+2*stride))[0]=
1707 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1710 static void pred4x4_top_dc_c(uint8_t *src, uint8_t *topright, int stride){
1711 const int dc= ( src[-stride] + src[1-stride] + src[2-stride] + src[3-stride] + 2) >>2;
1713 ((uint32_t*)(src+0*stride))[0]=
1714 ((uint32_t*)(src+1*stride))[0]=
1715 ((uint32_t*)(src+2*stride))[0]=
1716 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1719 static void pred4x4_128_dc_c(uint8_t *src, uint8_t *topright, int stride){
1720 ((uint32_t*)(src+0*stride))[0]=
1721 ((uint32_t*)(src+1*stride))[0]=
1722 ((uint32_t*)(src+2*stride))[0]=
1723 ((uint32_t*)(src+3*stride))[0]= 128U*0x01010101U;
1727 #define LOAD_TOP_RIGHT_EDGE\
1728 const int t4= topright[0];\
1729 const int t5= topright[1];\
1730 const int t6= topright[2];\
1731 const int t7= topright[3];\
1733 #define LOAD_LEFT_EDGE\
1734 const int l0= src[-1+0*stride];\
1735 const int l1= src[-1+1*stride];\
1736 const int l2= src[-1+2*stride];\
1737 const int l3= src[-1+3*stride];\
1739 #define LOAD_TOP_EDGE\
1740 const int t0= src[ 0-1*stride];\
1741 const int t1= src[ 1-1*stride];\
1742 const int t2= src[ 2-1*stride];\
1743 const int t3= src[ 3-1*stride];\
1745 static void pred4x4_down_right_c(uint8_t *src, uint8_t *topright, int stride){
1746 const int lt= src[-1-1*stride];
1750 src[0+3*stride]=(l3 + 2*l2 + l1 + 2)>>2;
1752 src[1+3*stride]=(l2 + 2*l1 + l0 + 2)>>2;
1755 src[2+3*stride]=(l1 + 2*l0 + lt + 2)>>2;
1759 src[3+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1762 src[3+2*stride]=(lt + 2*t0 + t1 + 2)>>2;
1764 src[3+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1765 src[3+0*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1768 static void pred4x4_down_left_c(uint8_t *src, uint8_t *topright, int stride){
1773 src[0+0*stride]=(t0 + t2 + 2*t1 + 2)>>2;
1775 src[0+1*stride]=(t1 + t3 + 2*t2 + 2)>>2;
1778 src[0+2*stride]=(t2 + t4 + 2*t3 + 2)>>2;
1782 src[0+3*stride]=(t3 + t5 + 2*t4 + 2)>>2;
1785 src[1+3*stride]=(t4 + t6 + 2*t5 + 2)>>2;
1787 src[2+3*stride]=(t5 + t7 + 2*t6 + 2)>>2;
1788 src[3+3*stride]=(t6 + 3*t7 + 2)>>2;
1791 static void pred4x4_vertical_right_c(uint8_t *src, uint8_t *topright, int stride){
1792 const int lt= src[-1-1*stride];
1795 const __attribute__((unused)) int unu= l3;
1798 src[1+2*stride]=(lt + t0 + 1)>>1;
1800 src[2+2*stride]=(t0 + t1 + 1)>>1;
1802 src[3+2*stride]=(t1 + t2 + 1)>>1;
1803 src[3+0*stride]=(t2 + t3 + 1)>>1;
1805 src[1+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1807 src[2+3*stride]=(lt + 2*t0 + t1 + 2)>>2;
1809 src[3+3*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1810 src[3+1*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1811 src[0+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1812 src[0+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1815 static void pred4x4_vertical_left_c(uint8_t *src, uint8_t *topright, int stride){
1818 const __attribute__((unused)) int unu= t7;
1820 src[0+0*stride]=(t0 + t1 + 1)>>1;
1822 src[0+2*stride]=(t1 + t2 + 1)>>1;
1824 src[1+2*stride]=(t2 + t3 + 1)>>1;
1826 src[2+2*stride]=(t3 + t4+ 1)>>1;
1827 src[3+2*stride]=(t4 + t5+ 1)>>1;
1828 src[0+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1830 src[0+3*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1832 src[1+3*stride]=(t2 + 2*t3 + t4 + 2)>>2;
1834 src[2+3*stride]=(t3 + 2*t4 + t5 + 2)>>2;
1835 src[3+3*stride]=(t4 + 2*t5 + t6 + 2)>>2;
1838 static void pred4x4_horizontal_up_c(uint8_t *src, uint8_t *topright, int stride){
1841 src[0+0*stride]=(l0 + l1 + 1)>>1;
1842 src[1+0*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1844 src[0+1*stride]=(l1 + l2 + 1)>>1;
1846 src[1+1*stride]=(l1 + 2*l2 + l3 + 2)>>2;
1848 src[0+2*stride]=(l2 + l3 + 1)>>1;
1850 src[1+2*stride]=(l2 + 2*l3 + l3 + 2)>>2;
1859 static void pred4x4_horizontal_down_c(uint8_t *src, uint8_t *topright, int stride){
1860 const int lt= src[-1-1*stride];
1863 const __attribute__((unused)) int unu= t3;
1866 src[2+1*stride]=(lt + l0 + 1)>>1;
1868 src[3+1*stride]=(l0 + 2*lt + t0 + 2)>>2;
1869 src[2+0*stride]=(lt + 2*t0 + t1 + 2)>>2;
1870 src[3+0*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1872 src[2+2*stride]=(l0 + l1 + 1)>>1;
1874 src[3+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1876 src[2+3*stride]=(l1 + l2+ 1)>>1;
1878 src[3+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1879 src[0+3*stride]=(l2 + l3 + 1)>>1;
1880 src[1+3*stride]=(l1 + 2*l2 + l3 + 2)>>2;
1883 static void pred16x16_vertical_c(uint8_t *src, int stride){
1885 const uint32_t a= ((uint32_t*)(src-stride))[0];
1886 const uint32_t b= ((uint32_t*)(src-stride))[1];
1887 const uint32_t c= ((uint32_t*)(src-stride))[2];
1888 const uint32_t d= ((uint32_t*)(src-stride))[3];
1890 for(i=0; i<16; i++){
1891 ((uint32_t*)(src+i*stride))[0]= a;
1892 ((uint32_t*)(src+i*stride))[1]= b;
1893 ((uint32_t*)(src+i*stride))[2]= c;
1894 ((uint32_t*)(src+i*stride))[3]= d;
1898 static void pred16x16_horizontal_c(uint8_t *src, int stride){
1901 for(i=0; i<16; i++){
1902 ((uint32_t*)(src+i*stride))[0]=
1903 ((uint32_t*)(src+i*stride))[1]=
1904 ((uint32_t*)(src+i*stride))[2]=
1905 ((uint32_t*)(src+i*stride))[3]= src[-1+i*stride]*0x01010101;
1909 static void pred16x16_dc_c(uint8_t *src, int stride){
1913 dc+= src[-1+i*stride];
1920 dc= 0x01010101*((dc + 16)>>5);
1922 for(i=0; i<16; i++){
1923 ((uint32_t*)(src+i*stride))[0]=
1924 ((uint32_t*)(src+i*stride))[1]=
1925 ((uint32_t*)(src+i*stride))[2]=
1926 ((uint32_t*)(src+i*stride))[3]= dc;
1930 static void pred16x16_left_dc_c(uint8_t *src, int stride){
1934 dc+= src[-1+i*stride];
1937 dc= 0x01010101*((dc + 8)>>4);
1939 for(i=0; i<16; i++){
1940 ((uint32_t*)(src+i*stride))[0]=
1941 ((uint32_t*)(src+i*stride))[1]=
1942 ((uint32_t*)(src+i*stride))[2]=
1943 ((uint32_t*)(src+i*stride))[3]= dc;
1947 static void pred16x16_top_dc_c(uint8_t *src, int stride){
1953 dc= 0x01010101*((dc + 8)>>4);
1955 for(i=0; i<16; i++){
1956 ((uint32_t*)(src+i*stride))[0]=
1957 ((uint32_t*)(src+i*stride))[1]=
1958 ((uint32_t*)(src+i*stride))[2]=
1959 ((uint32_t*)(src+i*stride))[3]= dc;
1963 static void pred16x16_128_dc_c(uint8_t *src, int stride){
1966 for(i=0; i<16; i++){
1967 ((uint32_t*)(src+i*stride))[0]=
1968 ((uint32_t*)(src+i*stride))[1]=
1969 ((uint32_t*)(src+i*stride))[2]=
1970 ((uint32_t*)(src+i*stride))[3]= 0x01010101U*128U;
1974 static inline void pred16x16_plane_compat_c(uint8_t *src, int stride, const int svq3){
1977 uint8_t *cm = cropTbl + MAX_NEG_CROP;
1978 const uint8_t * const src0 = src+7-stride;
1979 const uint8_t *src1 = src+8*stride-1;
1980 const uint8_t *src2 = src1-2*stride; // == src+6*stride-1;
1981 int H = src0[1] - src0[-1];
1982 int V = src1[0] - src2[ 0];
1983 for(k=2; k<=8; ++k) {
1984 src1 += stride; src2 -= stride;
1985 H += k*(src0[k] - src0[-k]);
1986 V += k*(src1[0] - src2[ 0]);
1989 H = ( 5*(H/4) ) / 16;
1990 V = ( 5*(V/4) ) / 16;
1992 /* required for 100% accuracy */
1993 i = H; H = V; V = i;
1995 H = ( 5*H+32 ) >> 6;
1996 V = ( 5*V+32 ) >> 6;
1999 a = 16*(src1[0] + src2[16] + 1) - 7*(V+H);
2000 for(j=16; j>0; --j) {
2003 for(i=-16; i<0; i+=4) {
2004 src[16+i] = cm[ (b ) >> 5 ];
2005 src[17+i] = cm[ (b+ H) >> 5 ];
2006 src[18+i] = cm[ (b+2*H) >> 5 ];
2007 src[19+i] = cm[ (b+3*H) >> 5 ];
2014 static void pred16x16_plane_c(uint8_t *src, int stride){
2015 pred16x16_plane_compat_c(src, stride, 0);
2018 static void pred8x8_vertical_c(uint8_t *src, int stride){
2020 const uint32_t a= ((uint32_t*)(src-stride))[0];
2021 const uint32_t b= ((uint32_t*)(src-stride))[1];
2024 ((uint32_t*)(src+i*stride))[0]= a;
2025 ((uint32_t*)(src+i*stride))[1]= b;
2029 static void pred8x8_horizontal_c(uint8_t *src, int stride){
2033 ((uint32_t*)(src+i*stride))[0]=
2034 ((uint32_t*)(src+i*stride))[1]= src[-1+i*stride]*0x01010101;
2038 static void pred8x8_128_dc_c(uint8_t *src, int stride){
2042 ((uint32_t*)(src+i*stride))[0]=
2043 ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
2046 ((uint32_t*)(src+i*stride))[0]=
2047 ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
2051 static void pred8x8_left_dc_c(uint8_t *src, int stride){
2057 dc0+= src[-1+i*stride];
2058 dc2+= src[-1+(i+4)*stride];
2060 dc0= 0x01010101*((dc0 + 2)>>2);
2061 dc2= 0x01010101*((dc2 + 2)>>2);
2064 ((uint32_t*)(src+i*stride))[0]=
2065 ((uint32_t*)(src+i*stride))[1]= dc0;
2068 ((uint32_t*)(src+i*stride))[0]=
2069 ((uint32_t*)(src+i*stride))[1]= dc2;
2073 static void pred8x8_top_dc_c(uint8_t *src, int stride){
2079 dc0+= src[i-stride];
2080 dc1+= src[4+i-stride];
2082 dc0= 0x01010101*((dc0 + 2)>>2);
2083 dc1= 0x01010101*((dc1 + 2)>>2);
2086 ((uint32_t*)(src+i*stride))[0]= dc0;
2087 ((uint32_t*)(src+i*stride))[1]= dc1;
2090 ((uint32_t*)(src+i*stride))[0]= dc0;
2091 ((uint32_t*)(src+i*stride))[1]= dc1;
2096 static void pred8x8_dc_c(uint8_t *src, int stride){
2098 int dc0, dc1, dc2, dc3;
2102 dc0+= src[-1+i*stride] + src[i-stride];
2103 dc1+= src[4+i-stride];
2104 dc2+= src[-1+(i+4)*stride];
2106 dc3= 0x01010101*((dc1 + dc2 + 4)>>3);
2107 dc0= 0x01010101*((dc0 + 4)>>3);
2108 dc1= 0x01010101*((dc1 + 2)>>2);
2109 dc2= 0x01010101*((dc2 + 2)>>2);
2112 ((uint32_t*)(src+i*stride))[0]= dc0;
2113 ((uint32_t*)(src+i*stride))[1]= dc1;
2116 ((uint32_t*)(src+i*stride))[0]= dc2;
2117 ((uint32_t*)(src+i*stride))[1]= dc3;
2121 static void pred8x8_plane_c(uint8_t *src, int stride){
2124 uint8_t *cm = cropTbl + MAX_NEG_CROP;
2125 const uint8_t * const src0 = src+3-stride;
2126 const uint8_t *src1 = src+4*stride-1;
2127 const uint8_t *src2 = src1-2*stride; // == src+2*stride-1;
2128 int H = src0[1] - src0[-1];
2129 int V = src1[0] - src2[ 0];
2130 for(k=2; k<=4; ++k) {
2131 src1 += stride; src2 -= stride;
2132 H += k*(src0[k] - src0[-k]);
2133 V += k*(src1[0] - src2[ 0]);
2135 H = ( 17*H+16 ) >> 5;
2136 V = ( 17*V+16 ) >> 5;
2138 a = 16*(src1[0] + src2[8]+1) - 3*(V+H);
2139 for(j=8; j>0; --j) {
2142 src[0] = cm[ (b ) >> 5 ];
2143 src[1] = cm[ (b+ H) >> 5 ];
2144 src[2] = cm[ (b+2*H) >> 5 ];
2145 src[3] = cm[ (b+3*H) >> 5 ];
2146 src[4] = cm[ (b+4*H) >> 5 ];
2147 src[5] = cm[ (b+5*H) >> 5 ];
2148 src[6] = cm[ (b+6*H) >> 5 ];
2149 src[7] = cm[ (b+7*H) >> 5 ];
2154 static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
2155 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2156 int src_x_offset, int src_y_offset,
2157 qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
2158 MpegEncContext * const s = &h->s;
2159 const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
2160 const int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
2161 const int luma_xy= (mx&3) + ((my&3)<<2);
2162 uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*s->linesize;
2163 uint8_t * src_cb= pic->data[1] + (mx>>3) + (my>>3)*s->uvlinesize;
2164 uint8_t * src_cr= pic->data[2] + (mx>>3) + (my>>3)*s->uvlinesize;
2165 int extra_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16; //FIXME increase edge?, IMHO not worth it
2166 int extra_height= extra_width;
2168 const int full_mx= mx>>2;
2169 const int full_my= my>>2;
2171 assert(pic->data[0]);
2173 if(mx&7) extra_width -= 3;
2174 if(my&7) extra_height -= 3;
2176 if( full_mx < 0-extra_width
2177 || full_my < 0-extra_height
2178 || full_mx + 16/*FIXME*/ > s->width + extra_width
2179 || full_my + 16/*FIXME*/ > s->height + extra_height){
2180 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);
2181 src_y= s->edge_emu_buffer + 2 + 2*s->linesize;
2185 qpix_op[luma_xy](dest_y, src_y, s->linesize); //FIXME try variable height perhaps?
2187 qpix_op[luma_xy](dest_y + delta, src_y + delta, s->linesize);
2190 if(s->flags&CODEC_FLAG_GRAY) return;
2193 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);
2194 src_cb= s->edge_emu_buffer;
2196 chroma_op(dest_cb, src_cb, s->uvlinesize, chroma_height, mx&7, my&7);
2199 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);
2200 src_cr= s->edge_emu_buffer;
2202 chroma_op(dest_cr, src_cr, s->uvlinesize, chroma_height, mx&7, my&7);
2205 static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
2206 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2207 int x_offset, int y_offset,
2208 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2209 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
2210 int list0, int list1){
2211 MpegEncContext * const s = &h->s;
2212 qpel_mc_func *qpix_op= qpix_put;
2213 h264_chroma_mc_func chroma_op= chroma_put;
2215 dest_y += 2*x_offset + 2*y_offset*s-> linesize;
2216 dest_cb += x_offset + y_offset*s->uvlinesize;
2217 dest_cr += x_offset + y_offset*s->uvlinesize;
2218 x_offset += 8*s->mb_x;
2219 y_offset += 8*s->mb_y;
2222 Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
2223 mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
2224 dest_y, dest_cb, dest_cr, x_offset, y_offset,
2225 qpix_op, chroma_op);
2228 chroma_op= chroma_avg;
2232 Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
2233 mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
2234 dest_y, dest_cb, dest_cr, x_offset, y_offset,
2235 qpix_op, chroma_op);
2239 static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2240 qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
2241 qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg)){
2242 MpegEncContext * const s = &h->s;
2243 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
2244 const int mb_type= s->current_picture.mb_type[mb_xy];
2246 assert(IS_INTER(mb_type));
2248 if(IS_16X16(mb_type)){
2249 mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
2250 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
2251 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2252 }else if(IS_16X8(mb_type)){
2253 mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
2254 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
2255 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2256 mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
2257 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
2258 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2259 }else if(IS_8X16(mb_type)){
2260 mc_part(h, 0, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 0, 0,
2261 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2262 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2263 mc_part(h, 4, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 4, 0,
2264 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2265 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2269 assert(IS_8X8(mb_type));
2272 const int sub_mb_type= h->sub_mb_type[i];
2274 int x_offset= (i&1)<<2;
2275 int y_offset= (i&2)<<1;
2277 if(IS_SUB_8X8(sub_mb_type)){
2278 mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2279 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2280 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2281 }else if(IS_SUB_8X4(sub_mb_type)){
2282 mc_part(h, n , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2283 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2284 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2285 mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
2286 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2287 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2288 }else if(IS_SUB_4X8(sub_mb_type)){
2289 mc_part(h, n , 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2290 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2291 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2292 mc_part(h, n+1, 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
2293 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2294 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2297 assert(IS_SUB_4X4(sub_mb_type));
2299 int sub_x_offset= x_offset + 2*(j&1);
2300 int sub_y_offset= y_offset + (j&2);
2301 mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
2302 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2303 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2310 static void decode_init_vlc(H264Context *h){
2311 static int done = 0;
2317 init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5,
2318 &chroma_dc_coeff_token_len [0], 1, 1,
2319 &chroma_dc_coeff_token_bits[0], 1, 1, 1);
2322 init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17,
2323 &coeff_token_len [i][0], 1, 1,
2324 &coeff_token_bits[i][0], 1, 1, 1);
2328 init_vlc(&chroma_dc_total_zeros_vlc[i], CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
2329 &chroma_dc_total_zeros_len [i][0], 1, 1,
2330 &chroma_dc_total_zeros_bits[i][0], 1, 1, 1);
2332 for(i=0; i<15; i++){
2333 init_vlc(&total_zeros_vlc[i], TOTAL_ZEROS_VLC_BITS, 16,
2334 &total_zeros_len [i][0], 1, 1,
2335 &total_zeros_bits[i][0], 1, 1, 1);
2339 init_vlc(&run_vlc[i], RUN_VLC_BITS, 7,
2340 &run_len [i][0], 1, 1,
2341 &run_bits[i][0], 1, 1, 1);
2343 init_vlc(&run7_vlc, RUN7_VLC_BITS, 16,
2344 &run_len [6][0], 1, 1,
2345 &run_bits[6][0], 1, 1, 1);
2350 * Sets the intra prediction function pointers.
2352 static void init_pred_ptrs(H264Context *h){
2353 // MpegEncContext * const s = &h->s;
2355 h->pred4x4[VERT_PRED ]= pred4x4_vertical_c;
2356 h->pred4x4[HOR_PRED ]= pred4x4_horizontal_c;
2357 h->pred4x4[DC_PRED ]= pred4x4_dc_c;
2358 h->pred4x4[DIAG_DOWN_LEFT_PRED ]= pred4x4_down_left_c;
2359 h->pred4x4[DIAG_DOWN_RIGHT_PRED]= pred4x4_down_right_c;
2360 h->pred4x4[VERT_RIGHT_PRED ]= pred4x4_vertical_right_c;
2361 h->pred4x4[HOR_DOWN_PRED ]= pred4x4_horizontal_down_c;
2362 h->pred4x4[VERT_LEFT_PRED ]= pred4x4_vertical_left_c;
2363 h->pred4x4[HOR_UP_PRED ]= pred4x4_horizontal_up_c;
2364 h->pred4x4[LEFT_DC_PRED ]= pred4x4_left_dc_c;
2365 h->pred4x4[TOP_DC_PRED ]= pred4x4_top_dc_c;
2366 h->pred4x4[DC_128_PRED ]= pred4x4_128_dc_c;
2368 h->pred8x8[DC_PRED8x8 ]= pred8x8_dc_c;
2369 h->pred8x8[VERT_PRED8x8 ]= pred8x8_vertical_c;
2370 h->pred8x8[HOR_PRED8x8 ]= pred8x8_horizontal_c;
2371 h->pred8x8[PLANE_PRED8x8 ]= pred8x8_plane_c;
2372 h->pred8x8[LEFT_DC_PRED8x8]= pred8x8_left_dc_c;
2373 h->pred8x8[TOP_DC_PRED8x8 ]= pred8x8_top_dc_c;
2374 h->pred8x8[DC_128_PRED8x8 ]= pred8x8_128_dc_c;
2376 h->pred16x16[DC_PRED8x8 ]= pred16x16_dc_c;
2377 h->pred16x16[VERT_PRED8x8 ]= pred16x16_vertical_c;
2378 h->pred16x16[HOR_PRED8x8 ]= pred16x16_horizontal_c;
2379 h->pred16x16[PLANE_PRED8x8 ]= pred16x16_plane_c;
2380 h->pred16x16[LEFT_DC_PRED8x8]= pred16x16_left_dc_c;
2381 h->pred16x16[TOP_DC_PRED8x8 ]= pred16x16_top_dc_c;
2382 h->pred16x16[DC_128_PRED8x8 ]= pred16x16_128_dc_c;
2385 static void free_tables(H264Context *h){
2386 av_freep(&h->intra4x4_pred_mode);
2387 av_freep(&h->chroma_pred_mode_table);
2388 av_freep(&h->cbp_table);
2389 av_freep(&h->mvd_table[0]);
2390 av_freep(&h->mvd_table[1]);
2391 av_freep(&h->direct_table);
2392 av_freep(&h->non_zero_count);
2393 av_freep(&h->slice_table_base);
2394 av_freep(&h->top_border);
2395 h->slice_table= NULL;
2397 av_freep(&h->mb2b_xy);
2398 av_freep(&h->mb2b8_xy);
2403 * needs widzh/height
2405 static int alloc_tables(H264Context *h){
2406 MpegEncContext * const s = &h->s;
2407 const int big_mb_num= s->mb_stride * (s->mb_height+1);
2410 CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8 * sizeof(uint8_t))
2412 CHECKED_ALLOCZ(h->non_zero_count , big_mb_num * 16 * sizeof(uint8_t))
2413 CHECKED_ALLOCZ(h->slice_table_base , big_mb_num * sizeof(uint8_t))
2414 CHECKED_ALLOCZ(h->top_border , s->mb_width * (16+8+8) * sizeof(uint8_t))
2415 CHECKED_ALLOCZ(h->cbp_table, big_mb_num * sizeof(uint16_t))
2417 if( h->pps.cabac ) {
2418 CHECKED_ALLOCZ(h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t))
2419 CHECKED_ALLOCZ(h->mvd_table[0], 32*big_mb_num * sizeof(uint16_t));
2420 CHECKED_ALLOCZ(h->mvd_table[1], 32*big_mb_num * sizeof(uint16_t));
2421 CHECKED_ALLOCZ(h->direct_table, 32*big_mb_num * sizeof(uint8_t));
2424 memset(h->slice_table_base, -1, big_mb_num * sizeof(uint8_t));
2425 h->slice_table= h->slice_table_base + s->mb_stride + 1;
2427 CHECKED_ALLOCZ(h->mb2b_xy , big_mb_num * sizeof(uint16_t));
2428 CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint16_t));
2429 for(y=0; y<s->mb_height; y++){
2430 for(x=0; x<s->mb_width; x++){
2431 const int mb_xy= x + y*s->mb_stride;
2432 const int b_xy = 4*x + 4*y*h->b_stride;
2433 const int b8_xy= 2*x + 2*y*h->b8_stride;
2435 h->mb2b_xy [mb_xy]= b_xy;
2436 h->mb2b8_xy[mb_xy]= b8_xy;
2446 static void common_init(H264Context *h){
2447 MpegEncContext * const s = &h->s;
2449 s->width = s->avctx->width;
2450 s->height = s->avctx->height;
2451 s->codec_id= s->avctx->codec->id;
2455 s->unrestricted_mv=1;
2456 s->decode=1; //FIXME
2459 static int decode_init(AVCodecContext *avctx){
2460 H264Context *h= avctx->priv_data;
2461 MpegEncContext * const s = &h->s;
2463 MPV_decode_defaults(s);
2468 s->out_format = FMT_H264;
2469 s->workaround_bugs= avctx->workaround_bugs;
2472 // s->decode_mb= ff_h263_decode_mb;
2474 avctx->pix_fmt= PIX_FMT_YUV420P;
2478 if(avctx->codec_tag != 0x31637661 && avctx->codec_tag != 0x31435641) // avc1
2481 if((avctx->extradata_size == 0) || (avctx->extradata == NULL)) {
2482 av_log(avctx, AV_LOG_ERROR, "AVC codec requires avcC data\n");
2492 static void frame_start(H264Context *h){
2493 MpegEncContext * const s = &h->s;
2496 MPV_frame_start(s, s->avctx);
2497 ff_er_frame_start(s);
2500 assert(s->linesize && s->uvlinesize);
2502 for(i=0; i<16; i++){
2503 h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
2504 h->chroma_subblock_offset[i]= 2*((scan8[i] - scan8[0])&7) + 2*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2507 h->block_offset[16+i]=
2508 h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2511 // s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
2514 static inline void backup_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){
2515 MpegEncContext * const s = &h->s;
2519 src_cb -= uvlinesize;
2520 src_cr -= uvlinesize;
2522 h->left_border[0]= h->top_border[s->mb_x][15];
2523 for(i=1; i<17; i++){
2524 h->left_border[i]= src_y[15+i* linesize];
2527 *(uint64_t*)(h->top_border[s->mb_x]+0)= *(uint64_t*)(src_y + 16*linesize);
2528 *(uint64_t*)(h->top_border[s->mb_x]+8)= *(uint64_t*)(src_y +8+16*linesize);
2530 if(!(s->flags&CODEC_FLAG_GRAY)){
2531 h->left_border[17 ]= h->top_border[s->mb_x][16+7];
2532 h->left_border[17+9]= h->top_border[s->mb_x][24+7];
2534 h->left_border[i+17 ]= src_cb[7+i*uvlinesize];
2535 h->left_border[i+17+9]= src_cr[7+i*uvlinesize];
2537 *(uint64_t*)(h->top_border[s->mb_x]+16)= *(uint64_t*)(src_cb+8*uvlinesize);
2538 *(uint64_t*)(h->top_border[s->mb_x]+24)= *(uint64_t*)(src_cr+8*uvlinesize);
2542 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){
2543 MpegEncContext * const s = &h->s;
2546 int deblock_left = (s->mb_x > 0);
2547 int deblock_top = (s->mb_y > 0);
2549 src_y -= linesize + 1;
2550 src_cb -= uvlinesize + 1;
2551 src_cr -= uvlinesize + 1;
2553 #define XCHG(a,b,t,xchg)\
2560 for(i = !deblock_top; i<17; i++){
2561 XCHG(h->left_border[i ], src_y [i* linesize], temp8, xchg);
2566 XCHG(*(uint64_t*)(h->top_border[s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
2567 XCHG(*(uint64_t*)(h->top_border[s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
2570 if(!(s->flags&CODEC_FLAG_GRAY)){
2572 for(i = !deblock_top; i<9; i++){
2573 XCHG(h->left_border[i+17 ], src_cb[i*uvlinesize], temp8, xchg);
2574 XCHG(h->left_border[i+17+9], src_cr[i*uvlinesize], temp8, xchg);
2578 XCHG(*(uint64_t*)(h->top_border[s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
2579 XCHG(*(uint64_t*)(h->top_border[s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
2584 static void hl_decode_mb(H264Context *h){
2585 MpegEncContext * const s = &h->s;
2586 const int mb_x= s->mb_x;
2587 const int mb_y= s->mb_y;
2588 const int mb_xy= mb_x + mb_y*s->mb_stride;
2589 const int mb_type= s->current_picture.mb_type[mb_xy];
2590 uint8_t *dest_y, *dest_cb, *dest_cr;
2591 int linesize, uvlinesize /*dct_offset*/;
2600 dest_y = s->current_picture.data[0] + (mb_y * 16* s->linesize ) + mb_x * 16;
2601 dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2602 dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2604 if (h->mb_field_decoding_flag) {
2605 linesize = s->linesize * 2;
2606 uvlinesize = s->uvlinesize * 2;
2607 if(mb_y&1){ //FIXME move out of this func?
2608 dest_y -= s->linesize*15;
2609 dest_cb-= s->linesize*7;
2610 dest_cr-= s->linesize*7;
2613 linesize = s->linesize;
2614 uvlinesize = s->uvlinesize;
2615 // dct_offset = s->linesize * 16;
2618 if(IS_INTRA(mb_type)){
2619 if(h->deblocking_filter)
2620 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1);
2622 if(!(s->flags&CODEC_FLAG_GRAY)){
2623 h->pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
2624 h->pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
2627 if(IS_INTRA4x4(mb_type)){
2629 for(i=0; i<16; i++){
2630 uint8_t * const ptr= dest_y + h->block_offset[i];
2632 const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
2635 if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
2636 const int topright_avail= (h->topright_samples_available<<i)&0x8000;
2637 assert(mb_y || linesize <= h->block_offset[i]);
2638 if(!topright_avail){
2639 tr= ptr[3 - linesize]*0x01010101;
2640 topright= (uint8_t*) &tr;
2641 }else if(i==5 && h->deblocking_filter){
2642 tr= *(uint32_t*)h->top_border[mb_x+1];
2643 topright= (uint8_t*) &tr;
2645 topright= ptr + 4 - linesize;
2649 h->pred4x4[ dir ](ptr, topright, linesize);
2650 if(h->non_zero_count_cache[ scan8[i] ]){
2651 if(s->codec_id == CODEC_ID_H264)
2652 s->dsp.h264_idct_add(ptr, h->mb + i*16, linesize);
2654 svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
2659 h->pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
2660 if(s->codec_id == CODEC_ID_H264)
2661 h264_luma_dc_dequant_idct_c(h->mb, s->qscale);
2663 svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
2665 if(h->deblocking_filter)
2666 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);
2667 }else if(s->codec_id == CODEC_ID_H264){
2668 hl_motion(h, dest_y, dest_cb, dest_cr,
2669 s->dsp.put_h264_qpel_pixels_tab, s->dsp.put_h264_chroma_pixels_tab,
2670 s->dsp.avg_h264_qpel_pixels_tab, s->dsp.avg_h264_chroma_pixels_tab);
2674 if(!IS_INTRA4x4(mb_type)){
2675 if(s->codec_id == CODEC_ID_H264){
2676 for(i=0; i<16; i++){
2677 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
2678 uint8_t * const ptr= dest_y + h->block_offset[i];
2679 s->dsp.h264_idct_add(ptr, h->mb + i*16, linesize);
2683 for(i=0; i<16; i++){
2684 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
2685 uint8_t * const ptr= dest_y + h->block_offset[i];
2686 svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
2692 if(!(s->flags&CODEC_FLAG_GRAY)){
2693 chroma_dc_dequant_idct_c(h->mb + 16*16, h->chroma_qp);
2694 chroma_dc_dequant_idct_c(h->mb + 16*16+4*16, h->chroma_qp);
2695 if(s->codec_id == CODEC_ID_H264){
2696 for(i=16; i<16+4; i++){
2697 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2698 uint8_t * const ptr= dest_cb + h->block_offset[i];
2699 s->dsp.h264_idct_add(ptr, h->mb + i*16, uvlinesize);
2702 for(i=20; i<20+4; i++){
2703 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2704 uint8_t * const ptr= dest_cr + h->block_offset[i];
2705 s->dsp.h264_idct_add(ptr, h->mb + i*16, uvlinesize);
2709 for(i=16; i<16+4; i++){
2710 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2711 uint8_t * const ptr= dest_cb + h->block_offset[i];
2712 svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
2715 for(i=20; i<20+4; i++){
2716 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2717 uint8_t * const ptr= dest_cr + h->block_offset[i];
2718 svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
2723 if(h->deblocking_filter) {
2724 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2725 filter_mb(h, mb_x, mb_y, dest_y, dest_cb, dest_cr);
2730 * fills the default_ref_list.
2732 static int fill_default_ref_list(H264Context *h){
2733 MpegEncContext * const s = &h->s;
2735 Picture sorted_short_ref[16];
2737 if(h->slice_type==B_TYPE){
2741 for(out_i=0; out_i<h->short_ref_count; out_i++){
2743 int best_poc=INT_MAX;
2745 for(i=0; i<h->short_ref_count; i++){
2746 const int poc= h->short_ref[i]->poc;
2747 if(poc > limit && poc < best_poc){
2753 assert(best_i != -1);
2756 sorted_short_ref[out_i]= *h->short_ref[best_i];
2760 if(s->picture_structure == PICT_FRAME){
2761 if(h->slice_type==B_TYPE){
2762 const int current_poc= s->current_picture_ptr->poc;
2765 for(list=0; list<2; list++){
2768 for(i=0; i<h->short_ref_count && index < h->ref_count[list]; i++){
2769 const int i2= list ? i : h->short_ref_count - i - 1;
2770 const int poc= sorted_short_ref[i2].poc;
2772 if(sorted_short_ref[i2].reference != 3) continue; //FIXME refernce field shit
2774 if((list==1 && poc > current_poc) || (list==0 && poc < current_poc)){
2775 h->default_ref_list[list][index ]= sorted_short_ref[i2];
2776 h->default_ref_list[list][index++].pic_id= sorted_short_ref[i2].frame_num;
2780 for(i=0; i<h->long_ref_count && index < h->ref_count[ list ]; i++){
2781 if(h->long_ref[i]->reference != 3) continue;
2783 h->default_ref_list[ list ][index ]= *h->long_ref[i];
2784 h->default_ref_list[ list ][index++].pic_id= i;;
2787 if(h->long_ref_count > 1 && h->short_ref_count==0){
2788 Picture temp= h->default_ref_list[1][0];
2789 h->default_ref_list[1][0] = h->default_ref_list[1][1];
2790 h->default_ref_list[1][0] = temp;
2793 if(index < h->ref_count[ list ])
2794 memset(&h->default_ref_list[list][index], 0, sizeof(Picture)*(h->ref_count[ list ] - index));
2798 for(i=0; i<h->short_ref_count && index < h->ref_count[0]; i++){
2799 if(h->short_ref[i]->reference != 3) continue; //FIXME refernce field shit
2800 h->default_ref_list[0][index ]= *h->short_ref[i];
2801 h->default_ref_list[0][index++].pic_id= h->short_ref[i]->frame_num;
2803 for(i=0; i<h->long_ref_count && index < h->ref_count[0]; i++){
2804 if(h->long_ref[i]->reference != 3) continue;
2805 h->default_ref_list[0][index ]= *h->long_ref[i];
2806 h->default_ref_list[0][index++].pic_id= i;;
2808 if(index < h->ref_count[0])
2809 memset(&h->default_ref_list[0][index], 0, sizeof(Picture)*(h->ref_count[0] - index));
2812 if(h->slice_type==B_TYPE){
2814 //FIXME second field balh
2820 static int decode_ref_pic_list_reordering(H264Context *h){
2821 MpegEncContext * const s = &h->s;
2824 if(h->slice_type==I_TYPE || h->slice_type==SI_TYPE) return 0; //FIXME move beofre func
2826 for(list=0; list<2; list++){
2827 memcpy(h->ref_list[list], h->default_ref_list[list], sizeof(Picture)*h->ref_count[list]);
2829 if(get_bits1(&s->gb)){
2830 int pred= h->curr_pic_num;
2833 for(index=0; ; index++){
2834 int reordering_of_pic_nums_idc= get_ue_golomb(&s->gb);
2838 if(reordering_of_pic_nums_idc==3)
2841 if(index >= h->ref_count[list]){
2842 av_log(h->s.avctx, AV_LOG_ERROR, "reference count overflow\n");
2846 if(reordering_of_pic_nums_idc<3){
2847 if(reordering_of_pic_nums_idc<2){
2848 const int abs_diff_pic_num= get_ue_golomb(&s->gb) + 1;
2850 if(abs_diff_pic_num >= h->max_pic_num){
2851 av_log(h->s.avctx, AV_LOG_ERROR, "abs_diff_pic_num overflow\n");
2855 if(reordering_of_pic_nums_idc == 0) pred-= abs_diff_pic_num;
2856 else pred+= abs_diff_pic_num;
2857 pred &= h->max_pic_num - 1;
2859 for(i= h->ref_count[list]-1; i>=index; i--){
2860 if(h->ref_list[list][i].pic_id == pred && h->ref_list[list][i].long_ref==0)
2864 pic_id= get_ue_golomb(&s->gb); //long_term_pic_idx
2866 for(i= h->ref_count[list]-1; i>=index; i--){
2867 if(h->ref_list[list][i].pic_id == pic_id && h->ref_list[list][i].long_ref==1)
2873 av_log(h->s.avctx, AV_LOG_ERROR, "reference picture missing during reorder\n");
2874 memset(&h->ref_list[list][index], 0, sizeof(Picture)); //FIXME
2875 }else if(i > index){
2876 Picture tmp= h->ref_list[list][i];
2877 for(; i>index; i--){
2878 h->ref_list[list][i]= h->ref_list[list][i-1];
2880 h->ref_list[list][index]= tmp;
2883 av_log(h->s.avctx, AV_LOG_ERROR, "illegal reordering_of_pic_nums_idc\n");
2889 if(h->slice_type!=B_TYPE) break;
2892 if(h->slice_type==B_TYPE && !h->direct_spatial_mv_pred)
2893 direct_dist_scale_factor(h);
2897 static int pred_weight_table(H264Context *h){
2898 MpegEncContext * const s = &h->s;
2901 h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
2902 h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
2904 for(list=0; list<2; list++){
2905 for(i=0; i<h->ref_count[list]; i++){
2906 int luma_weight_flag, chroma_weight_flag;
2908 luma_weight_flag= get_bits1(&s->gb);
2909 if(luma_weight_flag){
2910 h->luma_weight[list][i]= get_se_golomb(&s->gb);
2911 h->luma_offset[list][i]= get_se_golomb(&s->gb);
2914 chroma_weight_flag= get_bits1(&s->gb);
2915 if(chroma_weight_flag){
2918 h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);
2919 h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);
2923 if(h->slice_type != B_TYPE) break;
2929 * instantaneous decoder refresh.
2931 static void idr(H264Context *h){
2934 for(i=0; i<h->long_ref_count; i++){
2935 h->long_ref[i]->reference=0;
2936 h->long_ref[i]= NULL;
2938 h->long_ref_count=0;
2940 for(i=0; i<h->short_ref_count; i++){
2941 h->short_ref[i]->reference=0;
2942 h->short_ref[i]= NULL;
2944 h->short_ref_count=0;
2949 * @return the removed picture or NULL if an error occures
2951 static Picture * remove_short(H264Context *h, int frame_num){
2952 MpegEncContext * const s = &h->s;
2955 if(s->avctx->debug&FF_DEBUG_MMCO)
2956 av_log(h->s.avctx, AV_LOG_DEBUG, "remove short %d count %d\n", frame_num, h->short_ref_count);
2958 for(i=0; i<h->short_ref_count; i++){
2959 Picture *pic= h->short_ref[i];
2960 if(s->avctx->debug&FF_DEBUG_MMCO)
2961 av_log(h->s.avctx, AV_LOG_DEBUG, "%d %d %p\n", i, pic->frame_num, pic);
2962 if(pic->frame_num == frame_num){
2963 h->short_ref[i]= NULL;
2964 memmove(&h->short_ref[i], &h->short_ref[i+1], (h->short_ref_count - i - 1)*sizeof(Picture*));
2965 h->short_ref_count--;
2974 * @return the removed picture or NULL if an error occures
2976 static Picture * remove_long(H264Context *h, int i){
2979 if(i >= h->long_ref_count) return NULL;
2980 pic= h->long_ref[i];
2981 if(pic==NULL) return NULL;
2983 h->long_ref[i]= NULL;
2984 memmove(&h->long_ref[i], &h->long_ref[i+1], (h->long_ref_count - i - 1)*sizeof(Picture*));
2985 h->long_ref_count--;
2991 * Executes the reference picture marking (memory management control operations).
2993 static int execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count){
2994 MpegEncContext * const s = &h->s;
2996 int current_is_long=0;
2999 if((s->avctx->debug&FF_DEBUG_MMCO) && mmco_count==0)
3000 av_log(h->s.avctx, AV_LOG_DEBUG, "no mmco here\n");
3002 for(i=0; i<mmco_count; i++){
3003 if(s->avctx->debug&FF_DEBUG_MMCO)
3004 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);
3006 switch(mmco[i].opcode){
3007 case MMCO_SHORT2UNUSED:
3008 pic= remove_short(h, mmco[i].short_frame_num);
3009 if(pic==NULL) return -1;
3012 case MMCO_SHORT2LONG:
3013 pic= remove_long(h, mmco[i].long_index);
3014 if(pic) pic->reference=0;
3016 h->long_ref[ mmco[i].long_index ]= remove_short(h, mmco[i].short_frame_num);
3017 h->long_ref[ mmco[i].long_index ]->long_ref=1;
3019 case MMCO_LONG2UNUSED:
3020 pic= remove_long(h, mmco[i].long_index);
3021 if(pic==NULL) return -1;
3025 pic= remove_long(h, mmco[i].long_index);
3026 if(pic) pic->reference=0;
3028 h->long_ref[ mmco[i].long_index ]= s->current_picture_ptr;
3029 h->long_ref[ mmco[i].long_index ]->long_ref=1;
3030 h->long_ref_count++;
3034 case MMCO_SET_MAX_LONG:
3035 assert(mmco[i].long_index <= 16);
3036 while(mmco[i].long_index < h->long_ref_count){
3037 pic= remove_long(h, mmco[i].long_index);
3040 while(mmco[i].long_index > h->long_ref_count){
3041 h->long_ref[ h->long_ref_count++ ]= NULL;
3045 while(h->short_ref_count){
3046 pic= remove_short(h, h->short_ref[0]->frame_num);
3049 while(h->long_ref_count){
3050 pic= remove_long(h, h->long_ref_count-1);
3058 if(!current_is_long){
3059 pic= remove_short(h, s->current_picture_ptr->frame_num);
3062 av_log(h->s.avctx, AV_LOG_ERROR, "illegal short term buffer state detected\n");
3065 if(h->short_ref_count)
3066 memmove(&h->short_ref[1], &h->short_ref[0], h->short_ref_count*sizeof(Picture*));
3068 h->short_ref[0]= s->current_picture_ptr;
3069 h->short_ref[0]->long_ref=0;
3070 h->short_ref_count++;
3076 static int decode_ref_pic_marking(H264Context *h){
3077 MpegEncContext * const s = &h->s;
3080 if(h->nal_unit_type == NAL_IDR_SLICE){ //FIXME fields
3081 s->broken_link= get_bits1(&s->gb) -1;
3082 h->mmco[0].long_index= get_bits1(&s->gb) - 1; // current_long_term_idx
3083 if(h->mmco[0].long_index == -1)
3086 h->mmco[0].opcode= MMCO_LONG;
3090 if(get_bits1(&s->gb)){ // adaptive_ref_pic_marking_mode_flag
3091 for(i= h->mmco_index; i<MAX_MMCO_COUNT; i++) {
3092 MMCOOpcode opcode= get_ue_golomb(&s->gb);;
3094 h->mmco[i].opcode= opcode;
3095 if(opcode==MMCO_SHORT2UNUSED || opcode==MMCO_SHORT2LONG){
3096 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
3097 /* if(h->mmco[i].short_frame_num >= h->short_ref_count || h->short_ref[ h->mmco[i].short_frame_num ] == NULL){
3098 fprintf(stderr, "illegal short ref in memory management control operation %d\n", mmco);
3102 if(opcode==MMCO_SHORT2LONG || opcode==MMCO_LONG2UNUSED || opcode==MMCO_LONG || opcode==MMCO_SET_MAX_LONG){
3103 h->mmco[i].long_index= get_ue_golomb(&s->gb);
3104 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){
3105 av_log(h->s.avctx, AV_LOG_ERROR, "illegal long ref in memory management control operation %d\n", opcode);
3110 if(opcode > MMCO_LONG){
3111 av_log(h->s.avctx, AV_LOG_ERROR, "illegal memory management control operation %d\n", opcode);
3114 if(opcode == MMCO_END)
3119 assert(h->long_ref_count + h->short_ref_count <= h->sps.ref_frame_count);
3121 if(h->long_ref_count + h->short_ref_count == h->sps.ref_frame_count){ //FIXME fields
3122 h->mmco[0].opcode= MMCO_SHORT2UNUSED;
3123 h->mmco[0].short_frame_num= h->short_ref[ h->short_ref_count - 1 ]->frame_num;
3133 static int init_poc(H264Context *h){
3134 MpegEncContext * const s = &h->s;
3135 const int max_frame_num= 1<<h->sps.log2_max_frame_num;
3138 if(h->nal_unit_type == NAL_IDR_SLICE){
3139 h->frame_num_offset= 0;
3141 if(h->frame_num < h->prev_frame_num)
3142 h->frame_num_offset= h->prev_frame_num_offset + max_frame_num;
3144 h->frame_num_offset= h->prev_frame_num_offset;
3147 if(h->sps.poc_type==0){
3148 const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
3150 if (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
3151 h->poc_msb = h->prev_poc_msb + max_poc_lsb;
3152 else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
3153 h->poc_msb = h->prev_poc_msb - max_poc_lsb;
3155 h->poc_msb = h->prev_poc_msb;
3156 //printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
3158 field_poc[1] = h->poc_msb + h->poc_lsb;
3159 if(s->picture_structure == PICT_FRAME)
3160 field_poc[1] += h->delta_poc_bottom;
3161 }else if(h->sps.poc_type==1){
3162 int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
3165 if(h->sps.poc_cycle_length != 0)
3166 abs_frame_num = h->frame_num_offset + h->frame_num;
3170 if(h->nal_ref_idc==0 && abs_frame_num > 0)
3173 expected_delta_per_poc_cycle = 0;
3174 for(i=0; i < h->sps.poc_cycle_length; i++)
3175 expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
3177 if(abs_frame_num > 0){
3178 int poc_cycle_cnt = (abs_frame_num - 1) / h->sps.poc_cycle_length;
3179 int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
3181 expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
3182 for(i = 0; i <= frame_num_in_poc_cycle; i++)
3183 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
3187 if(h->nal_ref_idc == 0)
3188 expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
3190 field_poc[0] = expectedpoc + h->delta_poc[0];
3191 field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
3193 if(s->picture_structure == PICT_FRAME)
3194 field_poc[1] += h->delta_poc[1];
3197 if(h->nal_unit_type == NAL_IDR_SLICE){
3200 if(h->nal_ref_idc) poc= 2*(h->frame_num_offset + h->frame_num);
3201 else poc= 2*(h->frame_num_offset + h->frame_num) - 1;
3207 if(s->picture_structure != PICT_BOTTOM_FIELD)
3208 s->current_picture_ptr->field_poc[0]= field_poc[0];
3209 if(s->picture_structure != PICT_TOP_FIELD)
3210 s->current_picture_ptr->field_poc[1]= field_poc[1];
3211 if(s->picture_structure == PICT_FRAME) // FIXME field pix?
3212 s->current_picture_ptr->poc= FFMIN(field_poc[0], field_poc[1]);
3218 * decodes a slice header.
3219 * this will allso call MPV_common_init() and frame_start() as needed
3221 static int decode_slice_header(H264Context *h){
3222 MpegEncContext * const s = &h->s;
3223 int first_mb_in_slice, pps_id;
3224 int num_ref_idx_active_override_flag;
3225 static const uint8_t slice_type_map[5]= {P_TYPE, B_TYPE, I_TYPE, SP_TYPE, SI_TYPE};
3227 s->current_picture.reference= h->nal_ref_idc != 0;
3229 first_mb_in_slice= get_ue_golomb(&s->gb);
3231 h->slice_type= get_ue_golomb(&s->gb);
3232 if(h->slice_type > 9){
3233 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);
3236 if(h->slice_type > 4){
3238 h->slice_type_fixed=1;
3240 h->slice_type_fixed=0;
3242 h->slice_type= slice_type_map[ h->slice_type ];
3244 s->pict_type= h->slice_type; // to make a few old func happy, its wrong though
3246 pps_id= get_ue_golomb(&s->gb);
3248 av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
3251 h->pps= h->pps_buffer[pps_id];
3252 if(h->pps.slice_group_count == 0){
3253 av_log(h->s.avctx, AV_LOG_ERROR, "non existing PPS referenced\n");
3257 h->sps= h->sps_buffer[ h->pps.sps_id ];
3258 if(h->sps.log2_max_frame_num == 0){
3259 av_log(h->s.avctx, AV_LOG_ERROR, "non existing SPS referenced\n");
3263 s->mb_width= h->sps.mb_width;
3264 s->mb_height= h->sps.mb_height;
3266 h->b_stride= s->mb_width*4 + 1;
3267 h->b8_stride= s->mb_width*2 + 1;
3269 s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
3270 s->resync_mb_y = s->mb_y = first_mb_in_slice / s->mb_width; //FIXME AFFW
3272 s->width = 16*s->mb_width - 2*(h->sps.crop_left + h->sps.crop_right );
3273 if(h->sps.frame_mbs_only_flag)
3274 s->height= 16*s->mb_height - 2*(h->sps.crop_top + h->sps.crop_bottom);
3276 s->height= 16*s->mb_height - 4*(h->sps.crop_top + h->sps.crop_bottom); //FIXME recheck
3278 if (s->context_initialized
3279 && ( s->width != s->avctx->width || s->height != s->avctx->height)) {
3283 if (!s->context_initialized) {
3284 if (MPV_common_init(s) < 0)
3289 s->avctx->width = s->width;
3290 s->avctx->height = s->height;
3291 s->avctx->sample_aspect_ratio= h->sps.sar;
3293 if(h->sps.timing_info_present_flag && h->sps.fixed_frame_rate_flag){
3294 s->avctx->frame_rate = h->sps.time_scale;
3295 s->avctx->frame_rate_base = h->sps.num_units_in_tick;
3299 if(h->slice_num == 0){
3303 s->current_picture_ptr->frame_num= //FIXME frame_num cleanup
3304 h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
3306 if(h->sps.frame_mbs_only_flag){
3307 s->picture_structure= PICT_FRAME;
3309 if(get_bits1(&s->gb)) //field_pic_flag
3310 s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
3312 s->picture_structure= PICT_FRAME;
3315 if(s->picture_structure==PICT_FRAME){
3316 h->curr_pic_num= h->frame_num;
3317 h->max_pic_num= 1<< h->sps.log2_max_frame_num;
3319 h->curr_pic_num= 2*h->frame_num;
3320 h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
3323 if(h->nal_unit_type == NAL_IDR_SLICE){
3324 get_ue_golomb(&s->gb); /* idr_pic_id */
3327 if(h->sps.poc_type==0){
3328 h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
3330 if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
3331 h->delta_poc_bottom= get_se_golomb(&s->gb);
3335 if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
3336 h->delta_poc[0]= get_se_golomb(&s->gb);
3338 if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
3339 h->delta_poc[1]= get_se_golomb(&s->gb);
3344 if(h->pps.redundant_pic_cnt_present){
3345 h->redundant_pic_count= get_ue_golomb(&s->gb);
3348 //set defaults, might be overriden a few line later
3349 h->ref_count[0]= h->pps.ref_count[0];
3350 h->ref_count[1]= h->pps.ref_count[1];
3352 if(h->slice_type == P_TYPE || h->slice_type == SP_TYPE || h->slice_type == B_TYPE){
3353 if(h->slice_type == B_TYPE){
3354 h->direct_spatial_mv_pred= get_bits1(&s->gb);
3356 num_ref_idx_active_override_flag= get_bits1(&s->gb);
3358 if(num_ref_idx_active_override_flag){
3359 h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
3360 if(h->slice_type==B_TYPE)
3361 h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
3363 if(h->ref_count[0] > 32 || h->ref_count[1] > 32){
3364 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
3370 if(h->slice_num == 0){
3371 fill_default_ref_list(h);
3374 decode_ref_pic_list_reordering(h);
3376 if( (h->pps.weighted_pred && (h->slice_type == P_TYPE || h->slice_type == SP_TYPE ))
3377 || (h->pps.weighted_bipred_idc==1 && h->slice_type==B_TYPE ) )
3378 pred_weight_table(h);
3380 if(s->current_picture.reference)
3381 decode_ref_pic_marking(h);
3383 if( h->slice_type != I_TYPE && h->slice_type != SI_TYPE && h->pps.cabac )
3384 h->cabac_init_idc = get_ue_golomb(&s->gb);
3386 h->last_qscale_diff = 0;
3387 s->qscale = h->pps.init_qp + get_se_golomb(&s->gb);
3388 if(s->qscale<0 || s->qscale>51){
3389 av_log(s->avctx, AV_LOG_ERROR, "QP %d out of range\n", s->qscale);
3392 //FIXME qscale / qp ... stuff
3393 if(h->slice_type == SP_TYPE){
3394 get_bits1(&s->gb); /* sp_for_switch_flag */
3396 if(h->slice_type==SP_TYPE || h->slice_type == SI_TYPE){
3397 get_se_golomb(&s->gb); /* slice_qs_delta */
3400 h->deblocking_filter = 1;
3401 h->slice_alpha_c0_offset = 0;
3402 h->slice_beta_offset = 0;
3403 if( h->pps.deblocking_filter_parameters_present ) {
3404 h->deblocking_filter= get_ue_golomb(&s->gb);
3405 if(h->deblocking_filter < 2)
3406 h->deblocking_filter^= 1; // 1<->0
3408 if( h->deblocking_filter ) {
3409 h->slice_alpha_c0_offset = get_se_golomb(&s->gb) << 1;
3410 h->slice_beta_offset = get_se_golomb(&s->gb) << 1;
3415 if( h->pps.num_slice_groups > 1 && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
3416 slice_group_change_cycle= get_bits(&s->gb, ?);
3421 if(s->avctx->debug&FF_DEBUG_PICT_INFO){
3422 av_log(h->s.avctx, AV_LOG_DEBUG, "slice:%d mb:%d %c pps:%d frame:%d poc:%d/%d ref:%d/%d qp:%d loop:%d\n",
3423 h->slice_num, first_mb_in_slice,
3424 av_get_pict_type_char(h->slice_type),
3425 pps_id, h->frame_num,
3426 s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
3427 h->ref_count[0], h->ref_count[1],
3429 h->deblocking_filter
3439 static inline int get_level_prefix(GetBitContext *gb){
3443 OPEN_READER(re, gb);
3444 UPDATE_CACHE(re, gb);
3445 buf=GET_CACHE(re, gb);
3447 log= 32 - av_log2(buf);
3449 print_bin(buf>>(32-log), log);
3450 av_log(NULL, AV_LOG_DEBUG, "%5d %2d %3d lpr @%5d in %s get_level_prefix\n", buf>>(32-log), log, log-1, get_bits_count(gb), __FILE__);
3453 LAST_SKIP_BITS(re, gb, log);
3454 CLOSE_READER(re, gb);
3460 * decodes a residual block.
3461 * @param n block index
3462 * @param scantable scantable
3463 * @param max_coeff number of coefficients in the block
3464 * @return <0 if an error occured
3466 static int decode_residual(H264Context *h, GetBitContext *gb, DCTELEM *block, int n, const uint8_t *scantable, int qp, int max_coeff){
3467 MpegEncContext * const s = &h->s;
3468 const uint16_t *qmul= dequant_coeff[qp];
3469 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};
3470 int level[16], run[16];
3471 int suffix_length, zeros_left, coeff_num, coeff_token, total_coeff, i, trailing_ones;
3473 //FIXME put trailing_onex into the context
3475 if(n == CHROMA_DC_BLOCK_INDEX){
3476 coeff_token= get_vlc2(gb, chroma_dc_coeff_token_vlc.table, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 1);
3477 total_coeff= coeff_token>>2;
3479 if(n == LUMA_DC_BLOCK_INDEX){
3480 total_coeff= pred_non_zero_count(h, 0);
3481 coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
3482 total_coeff= coeff_token>>2;
3484 total_coeff= pred_non_zero_count(h, n);
3485 coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
3486 total_coeff= coeff_token>>2;
3487 h->non_zero_count_cache[ scan8[n] ]= total_coeff;
3491 //FIXME set last_non_zero?
3496 trailing_ones= coeff_token&3;
3497 tprintf("trailing:%d, total:%d\n", trailing_ones, total_coeff);
3498 assert(total_coeff<=16);
3500 for(i=0; i<trailing_ones; i++){
3501 level[i]= 1 - 2*get_bits1(gb);
3504 suffix_length= total_coeff > 10 && trailing_ones < 3;
3506 for(; i<total_coeff; i++){
3507 const int prefix= get_level_prefix(gb);
3508 int level_code, mask;
3510 if(prefix<14){ //FIXME try to build a large unified VLC table for all this
3512 level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
3514 level_code= (prefix<<suffix_length); //part
3515 }else if(prefix==14){
3517 level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
3519 level_code= prefix + get_bits(gb, 4); //part
3520 }else if(prefix==15){
3521 level_code= (prefix<<suffix_length) + get_bits(gb, 12); //part
3522 if(suffix_length==0) level_code+=15; //FIXME doesnt make (much)sense
3524 av_log(h->s.avctx, AV_LOG_ERROR, "prefix too large at %d %d\n", s->mb_x, s->mb_y);
3528 if(i==trailing_ones && i<3) level_code+= 2; //FIXME split first iteration
3530 mask= -(level_code&1);
3531 level[i]= (((2+level_code)>>1) ^ mask) - mask;
3533 if(suffix_length==0) suffix_length=1; //FIXME split first iteration
3536 if(ABS(level[i]) > (3<<(suffix_length-1)) && suffix_length<6) suffix_length++;
3538 if((2+level_code)>>1) > (3<<(suffix_length-1)) && suffix_length<6) suffix_length++;
3539 /* ? == prefix > 2 or sth */
3541 tprintf("level: %d suffix_length:%d\n", level[i], suffix_length);
3544 if(total_coeff == max_coeff)
3547 if(n == CHROMA_DC_BLOCK_INDEX)
3548 zeros_left= get_vlc2(gb, chroma_dc_total_zeros_vlc[ total_coeff-1 ].table, CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 1);
3550 zeros_left= get_vlc2(gb, total_zeros_vlc[ total_coeff-1 ].table, TOTAL_ZEROS_VLC_BITS, 1);
3553 for(i=0; i<total_coeff-1; i++){
3556 else if(zeros_left < 7){
3557 run[i]= get_vlc2(gb, run_vlc[zeros_left-1].table, RUN_VLC_BITS, 1);
3559 run[i]= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2);
3561 zeros_left -= run[i];
3565 av_log(h->s.avctx, AV_LOG_ERROR, "negative number of zero coeffs at %d %d\n", s->mb_x, s->mb_y);
3569 for(; i<total_coeff-1; i++){
3577 for(i=total_coeff-1; i>=0; i--){ //FIXME merge into rundecode?
3580 coeff_num += run[i] + 1; //FIXME add 1 earlier ?
3581 j= scantable[ coeff_num ];
3586 for(i=total_coeff-1; i>=0; i--){ //FIXME merge into rundecode?
3589 coeff_num += run[i] + 1; //FIXME add 1 earlier ?
3590 j= scantable[ coeff_num ];
3592 block[j]= level[i] * qmul[j];
3593 // printf("%d %d ", block[j], qmul[j]);
3600 * decodes a P_SKIP or B_SKIP macroblock
3602 static void decode_mb_skip(H264Context *h){
3603 MpegEncContext * const s = &h->s;
3604 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
3607 memset(h->non_zero_count[mb_xy], 0, 16);
3608 memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui
3610 if( h->slice_type == B_TYPE )
3612 // just for fill_caches. pred_direct_motion will set the real mb_type
3613 mb_type= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2|MB_TYPE_SKIP;
3616 fill_caches(h, mb_type); //FIXME check what is needed and what not ...
3617 pred_direct_motion(h, &mb_type);
3619 fill_rectangle(h->mvd_cache[0][scan8[0]], 4, 4, 8, 0, 4);
3620 fill_rectangle(h->mvd_cache[1][scan8[0]], 4, 4, 8, 0, 4);
3626 mb_type= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0|MB_TYPE_SKIP;
3628 if(h->sps.mb_aff && s->mb_skip_run==0 && (s->mb_y&1)==0){
3629 h->mb_field_decoding_flag= get_bits1(&s->gb);
3631 if(h->mb_field_decoding_flag)
3632 mb_type|= MB_TYPE_INTERLACED;
3634 fill_caches(h, mb_type); //FIXME check what is needed and what not ...
3635 pred_pskip_motion(h, &mx, &my);
3636 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
3637 fill_rectangle( h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4);
3639 fill_rectangle(h->mvd_cache[0][scan8[0]], 4, 4, 8, 0, 4);
3642 write_back_motion(h, mb_type);
3643 s->current_picture.mb_type[mb_xy]= mb_type|MB_TYPE_SKIP;
3644 s->current_picture.qscale_table[mb_xy]= s->qscale;
3645 h->slice_table[ mb_xy ]= h->slice_num;
3646 h->prev_mb_skiped= 1;
3650 * decodes a macroblock
3651 * @returns 0 if ok, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
3653 static int decode_mb_cavlc(H264Context *h){
3654 MpegEncContext * const s = &h->s;
3655 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
3656 int mb_type, partition_count, cbp;
3658 s->dsp.clear_blocks(h->mb); //FIXME avoid if allready clear (move after skip handlong?
3660 tprintf("pic:%d mb:%d/%d\n", h->frame_num, s->mb_x, s->mb_y);
3661 cbp = 0; /* avoid warning. FIXME: find a solution without slowing
3663 if(h->slice_type != I_TYPE && h->slice_type != SI_TYPE){
3664 if(s->mb_skip_run==-1)
3665 s->mb_skip_run= get_ue_golomb(&s->gb);
3667 if (s->mb_skip_run--) {
3672 if(h->sps.mb_aff /* && !field pic FIXME needed? */){
3674 h->mb_field_decoding_flag = get_bits1(&s->gb);
3676 h->mb_field_decoding_flag=0; //FIXME som ed note ?!
3678 h->prev_mb_skiped= 0;
3680 mb_type= get_ue_golomb(&s->gb);
3681 if(h->slice_type == B_TYPE){
3683 partition_count= b_mb_type_info[mb_type].partition_count;
3684 mb_type= b_mb_type_info[mb_type].type;
3687 goto decode_intra_mb;
3689 }else if(h->slice_type == P_TYPE /*|| h->slice_type == SP_TYPE */){
3691 partition_count= p_mb_type_info[mb_type].partition_count;
3692 mb_type= p_mb_type_info[mb_type].type;
3695 goto decode_intra_mb;
3698 assert(h->slice_type == I_TYPE);
3701 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);
3705 cbp= i_mb_type_info[mb_type].cbp;
3706 h->intra16x16_pred_mode= i_mb_type_info[mb_type].pred_mode;
3707 mb_type= i_mb_type_info[mb_type].type;
3710 if(h->mb_field_decoding_flag)
3711 mb_type |= MB_TYPE_INTERLACED;
3713 s->current_picture.mb_type[mb_xy]= mb_type;
3714 h->slice_table[ mb_xy ]= h->slice_num;
3716 if(IS_INTRA_PCM(mb_type)){
3720 // we assume these blocks are very rare so we dont optimize it
3721 align_get_bits(&s->gb);
3723 ptr= s->gb.buffer + get_bits_count(&s->gb);
3725 for(y=0; y<16; y++){
3726 const int index= 4*(y&3) + 64*(y>>2);
3727 for(x=0; x<16; x++){
3728 h->mb[index + (x&3) + 16*(x>>2)]= *(ptr++);
3732 const int index= 256 + 4*(y&3) + 32*(y>>2);
3734 h->mb[index + (x&3) + 16*(x>>2)]= *(ptr++);
3738 const int index= 256 + 64 + 4*(y&3) + 32*(y>>2);
3740 h->mb[index + (x&3) + 16*(x>>2)]= *(ptr++);
3744 skip_bits(&s->gb, 384); //FIXME check /fix the bitstream readers
3746 //FIXME deblock filter, non_zero_count_cache init ...
3747 memset(h->non_zero_count[mb_xy], 16, 16);
3748 s->current_picture.qscale_table[mb_xy]= s->qscale;
3753 fill_caches(h, mb_type);
3756 if(IS_INTRA(mb_type)){
3757 // init_top_left_availability(h);
3758 if(IS_INTRA4x4(mb_type)){
3761 // fill_intra4x4_pred_table(h);
3762 for(i=0; i<16; i++){
3763 const int mode_coded= !get_bits1(&s->gb);
3764 const int predicted_mode= pred_intra_mode(h, i);
3768 const int rem_mode= get_bits(&s->gb, 3);
3769 if(rem_mode<predicted_mode)
3774 mode= predicted_mode;
3777 h->intra4x4_pred_mode_cache[ scan8[i] ] = mode;
3779 write_back_intra_pred_mode(h);
3780 if( check_intra4x4_pred_mode(h) < 0)
3783 h->intra16x16_pred_mode= check_intra_pred_mode(h, h->intra16x16_pred_mode);
3784 if(h->intra16x16_pred_mode < 0)
3787 h->chroma_pred_mode= get_ue_golomb(&s->gb);
3789 h->chroma_pred_mode= check_intra_pred_mode(h, h->chroma_pred_mode);
3790 if(h->chroma_pred_mode < 0)
3792 }else if(partition_count==4){
3793 int i, j, sub_partition_count[4], list, ref[2][4];
3795 if(h->slice_type == B_TYPE){
3797 h->sub_mb_type[i]= get_ue_golomb(&s->gb);
3798 if(h->sub_mb_type[i] >=13){
3799 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);
3802 sub_partition_count[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
3803 h->sub_mb_type[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].type;
3805 if( IS_DIRECT(h->sub_mb_type[0]) || IS_DIRECT(h->sub_mb_type[1])
3806 || IS_DIRECT(h->sub_mb_type[2]) || IS_DIRECT(h->sub_mb_type[3]))
3807 pred_direct_motion(h, &mb_type);
3809 assert(h->slice_type == P_TYPE || h->slice_type == SP_TYPE); //FIXME SP correct ?
3811 h->sub_mb_type[i]= get_ue_golomb(&s->gb);
3812 if(h->sub_mb_type[i] >=4){
3813 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);
3816 sub_partition_count[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
3817 h->sub_mb_type[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].type;
3821 for(list=0; list<2; list++){
3822 const int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list];
3823 if(ref_count == 0) continue;
3825 if(IS_DIRECT(h->sub_mb_type[i])) continue;
3826 if(IS_DIR(h->sub_mb_type[i], 0, list)){
3827 ref[list][i] = get_te0_golomb(&s->gb, ref_count); //FIXME init to 0 before and skip?
3835 for(list=0; list<2; list++){
3836 const int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list];
3837 if(ref_count == 0) continue;
3840 if(IS_DIRECT(h->sub_mb_type[i])) continue;
3841 h->ref_cache[list][ scan8[4*i] ]=h->ref_cache[list][ scan8[4*i]+1 ]=
3842 h->ref_cache[list][ scan8[4*i]+8 ]=h->ref_cache[list][ scan8[4*i]+9 ]= ref[list][i];
3844 if(IS_DIR(h->sub_mb_type[i], 0, list)){
3845 const int sub_mb_type= h->sub_mb_type[i];
3846 const int block_width= (sub_mb_type & (MB_TYPE_16x16|MB_TYPE_16x8)) ? 2 : 1;
3847 for(j=0; j<sub_partition_count[i]; j++){
3849 const int index= 4*i + block_width*j;
3850 int16_t (* mv_cache)[2]= &h->mv_cache[list][ scan8[index] ];
3851 pred_motion(h, index, block_width, list, h->ref_cache[list][ scan8[index] ], &mx, &my);
3852 mx += get_se_golomb(&s->gb);
3853 my += get_se_golomb(&s->gb);
3854 tprintf("final mv:%d %d\n", mx, my);
3856 if(IS_SUB_8X8(sub_mb_type)){
3857 mv_cache[ 0 ][0]= mv_cache[ 1 ][0]=
3858 mv_cache[ 8 ][0]= mv_cache[ 9 ][0]= mx;
3859 mv_cache[ 0 ][1]= mv_cache[ 1 ][1]=
3860 mv_cache[ 8 ][1]= mv_cache[ 9 ][1]= my;
3861 }else if(IS_SUB_8X4(sub_mb_type)){
3862 mv_cache[ 0 ][0]= mv_cache[ 1 ][0]= mx;
3863 mv_cache[ 0 ][1]= mv_cache[ 1 ][1]= my;
3864 }else if(IS_SUB_4X8(sub_mb_type)){
3865 mv_cache[ 0 ][0]= mv_cache[ 8 ][0]= mx;
3866 mv_cache[ 0 ][1]= mv_cache[ 8 ][1]= my;
3868 assert(IS_SUB_4X4(sub_mb_type));
3869 mv_cache[ 0 ][0]= mx;
3870 mv_cache[ 0 ][1]= my;
3874 uint32_t *p= (uint32_t *)&h->mv_cache[list][ scan8[4*i] ][0];
3880 }else if(IS_DIRECT(mb_type)){
3881 pred_direct_motion(h, &mb_type);
3882 s->current_picture.mb_type[mb_xy]= mb_type;
3884 int list, mx, my, i;
3885 //FIXME we should set ref_idx_l? to 0 if we use that later ...
3886 if(IS_16X16(mb_type)){
3887 for(list=0; list<2; list++){
3888 if(h->ref_count[list]>0){
3889 if(IS_DIR(mb_type, 0, list)){
3890 const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
3891 fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, val, 1);
3895 for(list=0; list<2; list++){
3896 if(IS_DIR(mb_type, 0, list)){
3897 pred_motion(h, 0, 4, list, h->ref_cache[list][ scan8[0] ], &mx, &my);
3898 mx += get_se_golomb(&s->gb);
3899 my += get_se_golomb(&s->gb);
3900 tprintf("final mv:%d %d\n", mx, my);
3902 fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, pack16to32(mx,my), 4);
3906 else if(IS_16X8(mb_type)){
3907 for(list=0; list<2; list++){
3908 if(h->ref_count[list]>0){
3910 if(IS_DIR(mb_type, i, list)){
3911 const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
3912 fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, val, 1);
3913 }else // needed only for mixed refs (e.g. B_L0_L1_16x8)
3914 fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, (LIST_NOT_USED&0xFF), 1);
3918 for(list=0; list<2; list++){
3920 if(IS_DIR(mb_type, i, list)){
3921 pred_16x8_motion(h, 8*i, list, h->ref_cache[list][scan8[0] + 16*i], &mx, &my);
3922 mx += get_se_golomb(&s->gb);
3923 my += get_se_golomb(&s->gb);
3924 tprintf("final mv:%d %d\n", mx, my);
3926 fill_rectangle(h->mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, pack16to32(mx,my), 4);
3928 fill_rectangle(h->mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, 0, 4);
3932 assert(IS_8X16(mb_type));
3933 for(list=0; list<2; list++){
3934 if(h->ref_count[list]>0){
3936 if(IS_DIR(mb_type, i, list)){ //FIXME optimize
3937 const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
3938 fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, val, 1);
3939 }else // needed only for mixed refs
3940 fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, (LIST_NOT_USED&0xFF), 1);
3944 for(list=0; list<2; list++){
3946 if(IS_DIR(mb_type, i, list)){
3947 pred_8x16_motion(h, i*4, list, h->ref_cache[list][ scan8[0] + 2*i ], &mx, &my);
3948 mx += get_se_golomb(&s->gb);
3949 my += get_se_golomb(&s->gb);
3950 tprintf("final mv:%d %d\n", mx, my);
3952 fill_rectangle(h->mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, pack16to32(mx,my), 4);
3954 fill_rectangle(h->mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, 0, 4);
3960 if(IS_INTER(mb_type))
3961 write_back_motion(h, mb_type);
3963 if(!IS_INTRA16x16(mb_type)){
3964 cbp= get_ue_golomb(&s->gb);
3966 av_log(h->s.avctx, AV_LOG_ERROR, "cbp too large (%d) at %d %d\n", cbp, s->mb_x, s->mb_y);
3970 if(IS_INTRA4x4(mb_type))
3971 cbp= golomb_to_intra4x4_cbp[cbp];
3973 cbp= golomb_to_inter_cbp[cbp];
3976 if(cbp || IS_INTRA16x16(mb_type)){
3977 int i8x8, i4x4, chroma_idx;
3978 int chroma_qp, dquant;
3979 GetBitContext *gb= IS_INTRA(mb_type) ? h->intra_gb_ptr : h->inter_gb_ptr;
3980 const uint8_t *scan, *dc_scan;
3982 // fill_non_zero_count_cache(h);
3984 if(IS_INTERLACED(mb_type)){
3986 dc_scan= luma_dc_field_scan;
3989 dc_scan= luma_dc_zigzag_scan;
3992 dquant= get_se_golomb(&s->gb);
3994 if( dquant > 25 || dquant < -26 ){
3995 av_log(h->s.avctx, AV_LOG_ERROR, "dquant out of range (%d) at %d %d\n", dquant, s->mb_x, s->mb_y);
3999 s->qscale += dquant;
4000 if(((unsigned)s->qscale) > 51){
4001 if(s->qscale<0) s->qscale+= 52;
4002 else s->qscale-= 52;
4005 h->chroma_qp= chroma_qp= get_chroma_qp(h, s->qscale);
4006 if(IS_INTRA16x16(mb_type)){
4007 if( decode_residual(h, h->intra_gb_ptr, h->mb, LUMA_DC_BLOCK_INDEX, dc_scan, s->qscale, 16) < 0){
4008 return -1; //FIXME continue if partotioned and other retirn -1 too
4011 assert((cbp&15) == 0 || (cbp&15) == 15);
4014 for(i8x8=0; i8x8<4; i8x8++){
4015 for(i4x4=0; i4x4<4; i4x4++){
4016 const int index= i4x4 + 4*i8x8;
4017 if( decode_residual(h, h->intra_gb_ptr, h->mb + 16*index, index, scan + 1, s->qscale, 15) < 0 ){
4023 fill_rectangle(&h->non_zero_count_cache[scan8[0]], 4, 4, 8, 0, 1);
4026 for(i8x8=0; i8x8<4; i8x8++){
4027 if(cbp & (1<<i8x8)){
4028 for(i4x4=0; i4x4<4; i4x4++){
4029 const int index= i4x4 + 4*i8x8;
4031 if( decode_residual(h, gb, h->mb + 16*index, index, scan, s->qscale, 16) <0 ){
4036 uint8_t * const nnz= &h->non_zero_count_cache[ scan8[4*i8x8] ];
4037 nnz[0] = nnz[1] = nnz[8] = nnz[9] = 0;
4043 for(chroma_idx=0; chroma_idx<2; chroma_idx++)
4044 if( decode_residual(h, gb, h->mb + 256 + 16*4*chroma_idx, CHROMA_DC_BLOCK_INDEX, chroma_dc_scan, chroma_qp, 4) < 0){
4050 for(chroma_idx=0; chroma_idx<2; chroma_idx++){
4051 for(i4x4=0; i4x4<4; i4x4++){
4052 const int index= 16 + 4*chroma_idx + i4x4;
4053 if( decode_residual(h, gb, h->mb + 16*index, index, scan + 1, chroma_qp, 15) < 0){
4059 uint8_t * const nnz= &h->non_zero_count_cache[0];
4060 nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =
4061 nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
4064 uint8_t * const nnz= &h->non_zero_count_cache[0];
4065 fill_rectangle(&nnz[scan8[0]], 4, 4, 8, 0, 1);
4066 nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =
4067 nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
4069 s->current_picture.qscale_table[mb_xy]= s->qscale;
4070 write_back_non_zero_count(h);
4075 static int decode_cabac_intra_mb_type(H264Context *h, int ctx_base, int intra_slice) {
4076 uint8_t *state= &h->cabac_state[ctx_base];
4080 MpegEncContext * const s = &h->s;
4081 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
4083 if( s->mb_x > 0 && !IS_INTRA4x4( s->current_picture.mb_type[mb_xy-1] ) )
4085 if( s->mb_y > 0 && !IS_INTRA4x4( s->current_picture.mb_type[mb_xy-s->mb_stride] ) )
4087 if( get_cabac( &h->cabac, &state[ctx] ) == 0 )
4088 return 0; /* I4x4 */
4091 if( get_cabac( &h->cabac, &state[0] ) == 0 )
4092 return 0; /* I4x4 */
4095 if( get_cabac_terminate( &h->cabac ) )
4096 return 25; /* PCM */
4098 mb_type = 1; /* I16x16 */
4099 if( get_cabac( &h->cabac, &state[1] ) )
4100 mb_type += 12; /* cbp_luma != 0 */
4102 if( get_cabac( &h->cabac, &state[2] ) ) {
4103 if( get_cabac( &h->cabac, &state[2+intra_slice] ) )
4104 mb_type += 4 * 2; /* cbp_chroma == 2 */
4106 mb_type += 4 * 1; /* cbp_chroma == 1 */
4108 if( get_cabac( &h->cabac, &state[3+intra_slice] ) )
4110 if( get_cabac( &h->cabac, &state[3+2*intra_slice] ) )
4115 static int decode_cabac_mb_type( H264Context *h ) {
4116 MpegEncContext * const s = &h->s;
4118 if( h->slice_type == I_TYPE ) {
4119 return decode_cabac_intra_mb_type(h, 3, 1);
4120 } else if( h->slice_type == P_TYPE ) {
4121 if( get_cabac( &h->cabac, &h->cabac_state[14] ) == 0 ) {
4123 if( get_cabac( &h->cabac, &h->cabac_state[15] ) == 0 ) {
4124 if( get_cabac( &h->cabac, &h->cabac_state[16] ) == 0 )
4125 return 0; /* P_L0_D16x16; */
4127 return 3; /* P_8x8; */
4129 if( get_cabac( &h->cabac, &h->cabac_state[17] ) == 0 )
4130 return 2; /* P_L0_D8x16; */
4132 return 1; /* P_L0_D16x8; */
4135 return decode_cabac_intra_mb_type(h, 17, 0) + 5;
4137 } else if( h->slice_type == B_TYPE ) {
4138 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
4142 if( s->mb_x > 0 && !IS_SKIP( s->current_picture.mb_type[mb_xy-1] )
4143 && !IS_DIRECT( s->current_picture.mb_type[mb_xy-1] ) )
4145 if( s->mb_y > 0 && !IS_SKIP( s->current_picture.mb_type[mb_xy-s->mb_stride] )
4146 && !IS_DIRECT( s->current_picture.mb_type[mb_xy-s->mb_stride] ) )
4149 if( !get_cabac( &h->cabac, &h->cabac_state[27+ctx] ) )
4150 return 0; /* B_Direct_16x16 */
4152 if( !get_cabac( &h->cabac, &h->cabac_state[27+3] ) ) {
4153 return 1 + get_cabac( &h->cabac, &h->cabac_state[27+5] ); /* B_L[01]_16x16 */
4156 bits = get_cabac( &h->cabac, &h->cabac_state[27+4] ) << 3;
4157 bits|= get_cabac( &h->cabac, &h->cabac_state[27+5] ) << 2;
4158 bits|= get_cabac( &h->cabac, &h->cabac_state[27+5] ) << 1;
4159 bits|= get_cabac( &h->cabac, &h->cabac_state[27+5] );
4161 return bits + 3; /* B_Bi_16x16 through B_L1_L0_16x8 */
4162 else if( bits == 13 ) {
4163 return decode_cabac_intra_mb_type(h, 32, 0) + 23;
4164 } else if( bits == 14 )
4165 return 11; /* B_L1_L0_8x16 */
4166 else if( bits == 15 )
4167 return 22; /* B_8x8 */
4169 bits= ( bits<<1 ) | get_cabac( &h->cabac, &h->cabac_state[27+5] );
4170 return bits - 4; /* B_L0_Bi_* through B_Bi_Bi_* */
4172 /* TODO SI/SP frames? */
4177 static int decode_cabac_mb_skip( H264Context *h) {
4178 MpegEncContext * const s = &h->s;
4179 const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
4180 const int mba_xy = mb_xy - 1;
4181 const int mbb_xy = mb_xy - s->mb_stride;
4184 if( s->mb_x > 0 && !IS_SKIP( s->current_picture.mb_type[mba_xy] ) )
4186 if( s->mb_y > 0 && !IS_SKIP( s->current_picture.mb_type[mbb_xy] ) )
4189 if( h->slice_type == P_TYPE || h->slice_type == SP_TYPE)
4190 return get_cabac( &h->cabac, &h->cabac_state[11+ctx] );
4192 return get_cabac( &h->cabac, &h->cabac_state[24+ctx] );
4195 static int decode_cabac_mb_intra4x4_pred_mode( H264Context *h, int pred_mode ) {
4198 if( get_cabac( &h->cabac, &h->cabac_state[68] ) )
4201 if( get_cabac( &h->cabac, &h->cabac_state[69] ) )
4203 if( get_cabac( &h->cabac, &h->cabac_state[69] ) )
4205 if( get_cabac( &h->cabac, &h->cabac_state[69] ) )
4207 if( mode >= pred_mode )
4213 static int decode_cabac_mb_chroma_pre_mode( H264Context *h) {
4214 MpegEncContext * const s = &h->s;
4215 const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
4216 const int mba_xy = mb_xy - 1;
4217 const int mbb_xy = mb_xy - s->mb_stride;
4221 /* No need to test for IS_INTRA4x4 and IS_INTRA16x16, as we set chroma_pred_mode_table to 0 */
4222 if( s->mb_x > 0 && h->chroma_pred_mode_table[mba_xy] != 0 )
4225 if( s->mb_y > 0 && h->chroma_pred_mode_table[mbb_xy] != 0 )
4228 if( get_cabac( &h->cabac, &h->cabac_state[64+ctx] ) == 0 )
4231 if( get_cabac( &h->cabac, &h->cabac_state[64+3] ) == 0 )
4233 if( get_cabac( &h->cabac, &h->cabac_state[64+3] ) == 0 )
4239 static const uint8_t block_idx_x[16] = {
4240 0, 1, 0, 1, 2, 3, 2, 3, 0, 1, 0, 1, 2, 3, 2, 3
4242 static const uint8_t block_idx_y[16] = {
4243 0, 0, 1, 1, 0, 0, 1, 1, 2, 2, 3, 3, 2, 2, 3, 3
4245 static const uint8_t block_idx_xy[4][4] = {
4252 static int decode_cabac_mb_cbp_luma( H264Context *h) {
4253 MpegEncContext * const s = &h->s;
4254 const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
4259 h->cbp_table[mb_xy] = 0; /* FIXME aaahahahah beurk */
4261 for( i8x8 = 0; i8x8 < 4; i8x8++ ) {
4267 x = block_idx_x[4*i8x8];
4268 y = block_idx_y[4*i8x8];
4272 else if( s->mb_x > 0 )
4277 else if( s->mb_y > 0 )
4278 mbb_xy = mb_xy - s->mb_stride;
4280 /* No need to test for skip as we put 0 for skip block */
4282 int i8x8a = block_idx_xy[(x-1)&0x03][y]/4;
4283 if( ((h->cbp_table[mba_xy] >> i8x8a)&0x01) == 0 )
4288 int i8x8b = block_idx_xy[x][(y-1)&0x03]/4;
4289 if( ((h->cbp_table[mbb_xy] >> i8x8b)&0x01) == 0 )
4293 if( get_cabac( &h->cabac, &h->cabac_state[73 + ctx] ) ) {
4295 h->cbp_table[mb_xy] = cbp; /* FIXME aaahahahah beurk */
4300 static int decode_cabac_mb_cbp_chroma( H264Context *h) {
4304 cbp_a = (h->left_cbp>>4)&0x03;
4305 cbp_b = (h-> top_cbp>>4)&0x03;
4308 if( cbp_a > 0 ) ctx++;
4309 if( cbp_b > 0 ) ctx += 2;
4310 if( get_cabac( &h->cabac, &h->cabac_state[77 + ctx] ) == 0 )
4314 if( cbp_a == 2 ) ctx++;
4315 if( cbp_b == 2 ) ctx += 2;
4316 return 1 + get_cabac( &h->cabac, &h->cabac_state[77 + ctx] );
4318 static int decode_cabac_mb_dqp( H264Context *h) {
4319 MpegEncContext * const s = &h->s;
4325 mbn_xy = s->mb_x + s->mb_y*s->mb_stride - 1;
4327 mbn_xy = s->mb_width - 1 + (s->mb_y-1)*s->mb_stride;
4329 if( mbn_xy >= 0 && h->last_qscale_diff != 0 && ( IS_INTRA16x16(s->current_picture.mb_type[mbn_xy] ) || (h->cbp_table[mbn_xy]&0x3f) ) )
4332 while( get_cabac( &h->cabac, &h->cabac_state[60 + ctx] ) ) {
4343 return -(val + 1)/2;
4345 static int decode_cabac_p_mb_sub_type( H264Context *h ) {
4346 if( get_cabac( &h->cabac, &h->cabac_state[21] ) )
4348 if( !get_cabac( &h->cabac, &h->cabac_state[22] ) )
4350 if( get_cabac( &h->cabac, &h->cabac_state[23] ) )
4354 static int decode_cabac_b_mb_sub_type( H264Context *h ) {
4356 if( !get_cabac( &h->cabac, &h->cabac_state[36] ) )
4357 return 0; /* B_Direct_8x8 */
4358 if( !get_cabac( &h->cabac, &h->cabac_state[37] ) )
4359 return 1 + get_cabac( &h->cabac, &h->cabac_state[39] ); /* B_L0_8x8, B_L1_8x8 */
4361 if( get_cabac( &h->cabac, &h->cabac_state[38] ) ) {
4362 if( get_cabac( &h->cabac, &h->cabac_state[39] ) )
4363 return 11 + get_cabac( &h->cabac, &h->cabac_state[39] ); /* B_L1_4x4, B_Bi_4x4 */
4366 type += 2*get_cabac( &h->cabac, &h->cabac_state[39] );
4367 type += get_cabac( &h->cabac, &h->cabac_state[39] );
4371 static int decode_cabac_mb_ref( H264Context *h, int list, int n ) {
4372 int refa = h->ref_cache[list][scan8[n] - 1];
4373 int refb = h->ref_cache[list][scan8[n] - 8];
4377 if( h->slice_type == B_TYPE) {
4378 if( refa > 0 && !h->direct_cache[scan8[n] - 1] )
4380 if( refb > 0 && !h->direct_cache[scan8[n] - 8] )
4389 while( get_cabac( &h->cabac, &h->cabac_state[54+ctx] ) ) {
4399 static int decode_cabac_mb_mvd( H264Context *h, int list, int n, int l ) {
4400 int amvd = abs( h->mvd_cache[list][scan8[n] - 1][l] ) +
4401 abs( h->mvd_cache[list][scan8[n] - 8][l] );
4402 int ctxbase = (l == 0) ? 40 : 47;
4407 else if( amvd > 32 )
4412 if(!get_cabac(&h->cabac, &h->cabac_state[ctxbase+ctx]))
4417 while( mvd < 9 && get_cabac( &h->cabac, &h->cabac_state[ctxbase+ctx] ) ) {
4425 while( get_cabac_bypass( &h->cabac ) ) {
4430 if( get_cabac_bypass( &h->cabac ) )
4434 if( get_cabac_bypass( &h->cabac ) ) return -mvd;
4438 static int inline get_cabac_cbf_ctx( H264Context *h, int cat, int idx ) {
4443 nza = h->left_cbp&0x100;
4444 nzb = h-> top_cbp&0x100;
4445 } else if( cat == 1 || cat == 2 ) {
4446 nza = h->non_zero_count_cache[scan8[idx] - 1];
4447 nzb = h->non_zero_count_cache[scan8[idx] - 8];
4448 } else if( cat == 3 ) {
4449 nza = (h->left_cbp>>(6+idx))&0x01;
4450 nzb = (h-> top_cbp>>(6+idx))&0x01;
4453 nza = h->non_zero_count_cache[scan8[16+idx] - 1];
4454 nzb = h->non_zero_count_cache[scan8[16+idx] - 8];
4463 return ctx + 4 * cat;
4466 static int inline decode_cabac_residual( H264Context *h, DCTELEM *block, int cat, int n, const uint8_t *scantable, int qp, int max_coeff) {
4467 const int mb_xy = h->s.mb_x + h->s.mb_y*h->s.mb_stride;
4468 const uint16_t *qmul= dequant_coeff[qp];
4469 static const int significant_coeff_flag_offset[5] = { 0, 15, 29, 44, 47 };
4470 static const int coeff_abs_level_m1_offset[5] = {227+ 0, 227+10, 227+20, 227+30, 227+39 };
4475 int coeff_count = 0;
4478 int abslevelgt1 = 0;
4480 /* cat: 0-> DC 16x16 n = 0
4481 * 1-> AC 16x16 n = luma4x4idx
4482 * 2-> Luma4x4 n = luma4x4idx
4483 * 3-> DC Chroma n = iCbCr
4484 * 4-> AC Chroma n = 4 * iCbCr + chroma4x4idx
4487 /* read coded block flag */
4488 if( get_cabac( &h->cabac, &h->cabac_state[85 + get_cabac_cbf_ctx( h, cat, n ) ] ) == 0 ) {
4489 if( cat == 1 || cat == 2 )
4490 h->non_zero_count_cache[scan8[n]] = 0;
4492 h->non_zero_count_cache[scan8[16+n]] = 0;
4497 for(last= 0; last < max_coeff - 1; last++) {
4498 if( get_cabac( &h->cabac, &h->cabac_state[105+significant_coeff_flag_offset[cat]+last] )) {
4499 index[coeff_count++] = last;
4500 if( get_cabac( &h->cabac, &h->cabac_state[166+significant_coeff_flag_offset[cat]+last] ) ) {
4506 if( last == max_coeff -1 ) {
4507 index[coeff_count++] = last;
4509 assert(coeff_count > 0);
4512 h->cbp_table[mb_xy] |= 0x100;
4513 else if( cat == 1 || cat == 2 )
4514 h->non_zero_count_cache[scan8[n]] = coeff_count;
4516 h->cbp_table[mb_xy] |= 0x40 << n;
4519 h->non_zero_count_cache[scan8[16+n]] = coeff_count;
4522 for( i = coeff_count - 1; i >= 0; i-- ) {
4523 int ctx = (abslevelgt1 != 0 ? 0 : FFMIN( 4, abslevel1 )) + coeff_abs_level_m1_offset[cat];
4524 int j= scantable[index[i]];
4526 if( get_cabac( &h->cabac, &h->cabac_state[ctx] ) == 0 ) {
4527 if( cat == 0 || cat == 3 ) {
4528 if( get_cabac_bypass( &h->cabac ) ) block[j] = -1;
4531 if( get_cabac_bypass( &h->cabac ) ) block[j] = -qmul[j];
4532 else block[j] = qmul[j];
4538 ctx = 5 + FFMIN( 4, abslevelgt1 ) + coeff_abs_level_m1_offset[cat];
4539 while( coeff_abs < 15 && get_cabac( &h->cabac, &h->cabac_state[ctx] ) ) {
4543 if( coeff_abs >= 15 ) {
4545 while( get_cabac_bypass( &h->cabac ) ) {
4546 coeff_abs += 1 << j;
4551 if( get_cabac_bypass( &h->cabac ) )
4552 coeff_abs += 1 << j ;
4556 if( cat == 0 || cat == 3 ) {
4557 if( get_cabac_bypass( &h->cabac ) ) block[j] = -coeff_abs;
4558 else block[j] = coeff_abs;
4560 if( get_cabac_bypass( &h->cabac ) ) block[j] = -coeff_abs * qmul[j];
4561 else block[j] = coeff_abs * qmul[j];
4571 * decodes a macroblock
4572 * @returns 0 if ok, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
4574 static int decode_mb_cabac(H264Context *h) {
4575 MpegEncContext * const s = &h->s;
4576 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
4577 int mb_type, partition_count, cbp = 0;
4579 s->dsp.clear_blocks(h->mb); //FIXME avoid if allready clear (move after skip handlong?)
4581 if( h->sps.mb_aff ) {
4582 av_log( h->s.avctx, AV_LOG_ERROR, "Fields not supported with CABAC\n" );
4586 if( h->slice_type != I_TYPE && h->slice_type != SI_TYPE ) {
4587 /* read skip flags */
4588 if( decode_cabac_mb_skip( h ) ) {
4591 h->cbp_table[mb_xy] = 0;
4592 h->chroma_pred_mode_table[mb_xy] = 0;
4593 h->last_qscale_diff = 0;
4599 h->prev_mb_skiped = 0;
4601 if( ( mb_type = decode_cabac_mb_type( h ) ) < 0 ) {
4602 av_log( h->s.avctx, AV_LOG_ERROR, "decode_cabac_mb_type failed\n" );
4606 if( h->slice_type == B_TYPE ) {
4608 partition_count= b_mb_type_info[mb_type].partition_count;
4609 mb_type= b_mb_type_info[mb_type].type;
4612 goto decode_intra_mb;
4614 } else if( h->slice_type == P_TYPE ) {
4616 partition_count= p_mb_type_info[mb_type].partition_count;
4617 mb_type= p_mb_type_info[mb_type].type;
4620 goto decode_intra_mb;
4623 assert(h->slice_type == I_TYPE);
4625 partition_count = 0;
4626 cbp= i_mb_type_info[mb_type].cbp;
4627 h->intra16x16_pred_mode= i_mb_type_info[mb_type].pred_mode;
4628 mb_type= i_mb_type_info[mb_type].type;
4631 if(h->mb_field_decoding_flag)
4632 mb_type |= MB_TYPE_INTERLACED;
4635 s->current_picture.mb_type[mb_xy]= mb_type;
4636 h->slice_table[ mb_xy ]= h->slice_num;
4638 if(IS_INTRA_PCM(mb_type)) {
4641 h->cbp_table[mb_xy] = 0xf +4*2; //FIXME ?!
4642 h->cbp_table[mb_xy] |= 0x1C0;
4643 h->chroma_pred_mode_table[mb_xy] = 0;
4644 s->current_picture.qscale_table[mb_xy]= s->qscale;
4648 fill_caches(h, mb_type);
4650 if( IS_INTRA( mb_type ) ) {
4651 if( IS_INTRA4x4( mb_type ) ) {
4653 for( i = 0; i < 16; i++ ) {
4654 int pred = pred_intra_mode( h, i );
4655 h->intra4x4_pred_mode_cache[ scan8[i] ] = decode_cabac_mb_intra4x4_pred_mode( h, pred );
4657 //av_log( s->avctx, AV_LOG_ERROR, "i4x4 pred=%d mode=%d\n", pred, h->intra4x4_pred_mode_cache[ scan8[i] ] );
4659 write_back_intra_pred_mode(h);
4660 if( check_intra4x4_pred_mode(h) < 0 ) return -1;
4662 h->intra16x16_pred_mode= check_intra_pred_mode( h, h->intra16x16_pred_mode );
4663 if( h->intra16x16_pred_mode < 0 ) return -1;
4665 h->chroma_pred_mode_table[mb_xy] =
4666 h->chroma_pred_mode = decode_cabac_mb_chroma_pre_mode( h );
4668 h->chroma_pred_mode= check_intra_pred_mode( h, h->chroma_pred_mode );
4669 if( h->chroma_pred_mode < 0 ) return -1;
4670 } else if( partition_count == 4 ) {
4671 int i, j, sub_partition_count[4], list, ref[2][4];
4673 if( h->slice_type == B_TYPE ) {
4674 for( i = 0; i < 4; i++ ) {
4675 h->sub_mb_type[i] = decode_cabac_b_mb_sub_type( h );
4676 sub_partition_count[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
4677 h->sub_mb_type[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].type;
4679 if( IS_DIRECT(h->sub_mb_type[0]) || IS_DIRECT(h->sub_mb_type[1])
4680 || IS_DIRECT(h->sub_mb_type[2]) || IS_DIRECT(h->sub_mb_type[3])) {
4681 pred_direct_motion(h, &mb_type);
4682 if( h->ref_count[0] > 1 || h->ref_count[1] > 1 ) {
4683 for( i = 0; i < 4; i++ )
4684 if( IS_DIRECT(h->sub_mb_type[i]) )
4685 fill_rectangle( &h->direct_cache[scan8[4*i]], 2, 2, 8, 1, 1 );
4689 for( i = 0; i < 4; i++ ) {
4690 h->sub_mb_type[i] = decode_cabac_p_mb_sub_type( h );
4691 sub_partition_count[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
4692 h->sub_mb_type[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].type;
4696 for( list = 0; list < 2; list++ ) {
4697 if( h->ref_count[list] > 0 ) {
4698 for( i = 0; i < 4; i++ ) {
4699 if(IS_DIRECT(h->sub_mb_type[i])) continue;
4700 if(IS_DIR(h->sub_mb_type[i], 0, list)){
4701 if( h->ref_count[list] > 1 )
4702 ref[list][i] = decode_cabac_mb_ref( h, list, 4*i );
4708 h->ref_cache[list][ scan8[4*i]+1 ]=
4709 h->ref_cache[list][ scan8[4*i]+8 ]=h->ref_cache[list][ scan8[4*i]+9 ]= ref[list][i];
4714 for(list=0; list<2; list++){
4716 if(IS_DIRECT(h->sub_mb_type[i])){
4717 fill_rectangle(h->mvd_cache[list][scan8[4*i]], 2, 2, 8, 0, 4);
4720 h->ref_cache[list][ scan8[4*i] ]=h->ref_cache[list][ scan8[4*i]+1 ];
4722 if(IS_DIR(h->sub_mb_type[i], 0, list) && !IS_DIRECT(h->sub_mb_type[i])){
4723 const int sub_mb_type= h->sub_mb_type[i];
4724 const int block_width= (sub_mb_type & (MB_TYPE_16x16|MB_TYPE_16x8)) ? 2 : 1;
4725 for(j=0; j<sub_partition_count[i]; j++){
4728 const int index= 4*i + block_width*j;
4729 int16_t (* mv_cache)[2]= &h->mv_cache[list][ scan8[index] ];
4730 int16_t (* mvd_cache)[2]= &h->mvd_cache[list][ scan8[index] ];
4731 pred_motion(h, index, block_width, list, h->ref_cache[list][ scan8[index] ], &mpx, &mpy);
4733 mx = mpx + decode_cabac_mb_mvd( h, list, index, 0 );
4734 my = mpy + decode_cabac_mb_mvd( h, list, index, 1 );
4735 tprintf("final mv:%d %d\n", mx, my);
4737 if(IS_SUB_8X8(sub_mb_type)){
4738 mv_cache[ 0 ][0]= mv_cache[ 1 ][0]=
4739 mv_cache[ 8 ][0]= mv_cache[ 9 ][0]= mx;
4740 mv_cache[ 0 ][1]= mv_cache[ 1 ][1]=
4741 mv_cache[ 8 ][1]= mv_cache[ 9 ][1]= my;
4743 mvd_cache[ 0 ][0]= mvd_cache[ 1 ][0]=
4744 mvd_cache[ 8 ][0]= mvd_cache[ 9 ][0]= mx - mpx;
4745 mvd_cache[ 0 ][1]= mvd_cache[ 1 ][1]=
4746 mvd_cache[ 8 ][1]= mvd_cache[ 9 ][1]= my - mpy;
4747 }else if(IS_SUB_8X4(sub_mb_type)){
4748 mv_cache[ 0 ][0]= mv_cache[ 1 ][0]= mx;
4749 mv_cache[ 0 ][1]= mv_cache[ 1 ][1]= my;
4751 mvd_cache[ 0 ][0]= mvd_cache[ 1 ][0]= mx- mpx;
4752 mvd_cache[ 0 ][1]= mvd_cache[ 1 ][1]= my - mpy;
4753 }else if(IS_SUB_4X8(sub_mb_type)){
4754 mv_cache[ 0 ][0]= mv_cache[ 8 ][0]= mx;
4755 mv_cache[ 0 ][1]= mv_cache[ 8 ][1]= my;
4757 mvd_cache[ 0 ][0]= mvd_cache[ 8 ][0]= mx - mpx;
4758 mvd_cache[ 0 ][1]= mvd_cache[ 8 ][1]= my - mpy;
4760 assert(IS_SUB_4X4(sub_mb_type));
4761 mv_cache[ 0 ][0]= mx;
4762 mv_cache[ 0 ][1]= my;
4764 mvd_cache[ 0 ][0]= mx - mpx;
4765 mvd_cache[ 0 ][1]= my - mpy;
4769 uint32_t *p= (uint32_t *)&h->mv_cache[list][ scan8[4*i] ][0];
4770 uint32_t *pd= (uint32_t *)&h->mvd_cache[list][ scan8[4*i] ][0];
4771 p[0] = p[1] = p[8] = p[9] = 0;
4772 pd[0]= pd[1]= pd[8]= pd[9]= 0;
4776 } else if( IS_DIRECT(mb_type) ) {
4777 pred_direct_motion(h, &mb_type);
4778 s->current_picture.mb_type[mb_xy]= mb_type;
4779 fill_rectangle(h->mvd_cache[0][scan8[0]], 4, 4, 8, 0, 4);
4780 fill_rectangle(h->mvd_cache[1][scan8[0]], 4, 4, 8, 0, 4);
4782 int list, mx, my, i, mpx, mpy;
4783 if(IS_16X16(mb_type)){
4784 for(list=0; list<2; list++){
4785 if(IS_DIR(mb_type, 0, list)){
4786 if(h->ref_count[list] > 0 ){
4787 const int ref = h->ref_count[list] > 1 ? decode_cabac_mb_ref( h, list, 0 ) : 0;
4788 fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, ref, 1);
4792 for(list=0; list<2; list++){
4793 if(IS_DIR(mb_type, 0, list)){
4794 pred_motion(h, 0, 4, list, h->ref_cache[list][ scan8[0] ], &mpx, &mpy);
4796 mx = mpx + decode_cabac_mb_mvd( h, list, 0, 0 );
4797 my = mpy + decode_cabac_mb_mvd( h, list, 0, 1 );
4798 tprintf("final mv:%d %d\n", mx, my);
4800 fill_rectangle(h->mvd_cache[list][ scan8[0] ], 4, 4, 8, pack16to32(mx-mpx,my-mpy), 4);
4801 fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, pack16to32(mx,my), 4);
4805 else if(IS_16X8(mb_type)){
4806 for(list=0; list<2; list++){
4807 if(h->ref_count[list]>0){
4809 if(IS_DIR(mb_type, i, list)){
4810 const int ref= h->ref_count[list] > 1 ? decode_cabac_mb_ref( h, list, 8*i ) : 0;
4811 fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, ref, 1);
4813 fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, (LIST_NOT_USED&0xFF), 1);
4817 for(list=0; list<2; list++){
4819 if(IS_DIR(mb_type, i, list)){
4820 pred_16x8_motion(h, 8*i, list, h->ref_cache[list][scan8[0] + 16*i], &mpx, &mpy);
4821 mx = mpx + decode_cabac_mb_mvd( h, list, 8*i, 0 );
4822 my = mpy + decode_cabac_mb_mvd( h, list, 8*i, 1 );
4823 tprintf("final mv:%d %d\n", mx, my);
4825 fill_rectangle(h->mvd_cache[list][ scan8[0] + 16*i ], 4, 2, 8, pack16to32(mx-mpx,my-mpy), 4);
4826 fill_rectangle(h->mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, pack16to32(mx,my), 4);
4827 }else{ // needed only for mixed refs
4828 fill_rectangle(h->mvd_cache[list][ scan8[0] + 16*i ], 4, 2, 8, 0, 4);
4829 fill_rectangle(h-> mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, 0, 4);
4834 assert(IS_8X16(mb_type));
4835 for(list=0; list<2; list++){
4836 if(h->ref_count[list]>0){
4838 if(IS_DIR(mb_type, i, list)){ //FIXME optimize
4839 const int ref= h->ref_count[list] > 1 ? decode_cabac_mb_ref( h, list, 4*i ) : 0;
4840 fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, ref, 1);
4842 fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, (LIST_NOT_USED&0xFF), 1);
4846 for(list=0; list<2; list++){
4848 if(IS_DIR(mb_type, i, list)){
4849 pred_8x16_motion(h, i*4, list, h->ref_cache[list][ scan8[0] + 2*i ], &mpx, &mpy);
4850 mx = mpx + decode_cabac_mb_mvd( h, list, 4*i, 0 );
4851 my = mpy + decode_cabac_mb_mvd( h, list, 4*i, 1 );
4853 tprintf("final mv:%d %d\n", mx, my);
4854 fill_rectangle(h->mvd_cache[list][ scan8[0] + 2*i ], 2, 4, 8, pack16to32(mx-mpx,my-mpy), 4);
4855 fill_rectangle(h->mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, pack16to32(mx,my), 4);
4856 }else{ // needed only for mixed refs
4857 fill_rectangle(h->mvd_cache[list][ scan8[0] + 2*i ], 2, 4, 8, 0, 4);
4858 fill_rectangle(h-> mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, 0, 4);
4865 if( IS_INTER( mb_type ) ) {
4866 h->chroma_pred_mode_table[mb_xy] = 0;
4867 write_back_motion( h, mb_type );
4870 if( !IS_INTRA16x16( mb_type ) ) {
4871 cbp = decode_cabac_mb_cbp_luma( h );
4872 cbp |= decode_cabac_mb_cbp_chroma( h ) << 4;
4875 h->cbp_table[mb_xy] = cbp;
4877 if( cbp || IS_INTRA16x16( mb_type ) ) {
4878 const uint8_t *scan, *dc_scan;
4881 if(IS_INTERLACED(mb_type)){
4883 dc_scan= luma_dc_field_scan;
4886 dc_scan= luma_dc_zigzag_scan;
4889 h->last_qscale_diff = dqp = decode_cabac_mb_dqp( h );
4891 if(((unsigned)s->qscale) > 51){
4892 if(s->qscale<0) s->qscale+= 52;
4893 else s->qscale-= 52;
4895 h->chroma_qp = get_chroma_qp(h, s->qscale);
4897 if( IS_INTRA16x16( mb_type ) ) {
4899 //av_log( s->avctx, AV_LOG_ERROR, "INTRA16x16 DC\n" );
4900 if( decode_cabac_residual( h, h->mb, 0, 0, dc_scan, s->qscale, 16) < 0)
4903 for( i = 0; i < 16; i++ ) {
4904 //av_log( s->avctx, AV_LOG_ERROR, "INTRA16x16 AC:%d\n", i );
4905 if( decode_cabac_residual(h, h->mb + 16*i, 1, i, scan + 1, s->qscale, 15) < 0 )
4909 fill_rectangle(&h->non_zero_count_cache[scan8[0]], 4, 4, 8, 0, 1);
4913 for( i8x8 = 0; i8x8 < 4; i8x8++ ) {
4914 if( cbp & (1<<i8x8) ) {
4915 for( i4x4 = 0; i4x4 < 4; i4x4++ ) {
4916 const int index = 4*i8x8 + i4x4;
4917 //av_log( s->avctx, AV_LOG_ERROR, "Luma4x4: %d\n", index );
4918 if( decode_cabac_residual(h, h->mb + 16*index, 2, index, scan, s->qscale, 16) < 0 )
4922 uint8_t * const nnz= &h->non_zero_count_cache[ scan8[4*i8x8] ];
4923 nnz[0] = nnz[1] = nnz[8] = nnz[9] = 0;
4930 for( c = 0; c < 2; c++ ) {
4931 //av_log( s->avctx, AV_LOG_ERROR, "INTRA C%d-DC\n",c );
4932 if( decode_cabac_residual(h, h->mb + 256 + 16*4*c, 3, c, chroma_dc_scan, h->chroma_qp, 4) < 0)
4939 for( c = 0; c < 2; c++ ) {
4940 for( i = 0; i < 4; i++ ) {
4941 const int index = 16 + 4 * c + i;
4942 //av_log( s->avctx, AV_LOG_ERROR, "INTRA C%d-AC %d\n",c, index - 16 );
4943 if( decode_cabac_residual(h, h->mb + 16*index, 4, index - 16, scan + 1, h->chroma_qp, 15) < 0)
4948 uint8_t * const nnz= &h->non_zero_count_cache[0];
4949 nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =
4950 nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
4953 uint8_t * const nnz= &h->non_zero_count_cache[0];
4954 fill_rectangle(&nnz[scan8[0]], 4, 4, 8, 0, 1);
4955 nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =
4956 nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
4959 s->current_picture.qscale_table[mb_xy]= s->qscale;
4960 write_back_non_zero_count(h);
4966 static void filter_mb_edgev( H264Context *h, uint8_t *pix, int stride, int bS[4], int qp ) {
4968 const int index_a = clip( qp + h->slice_alpha_c0_offset, 0, 51 );
4969 const int alpha = alpha_table[index_a];
4970 const int beta = beta_table[clip( qp + h->slice_beta_offset, 0, 51 )];
4972 for( i = 0; i < 4; i++ ) {
4979 const int tc0 = tc0_table[index_a][bS[i] - 1];
4980 /* 4px edge length */
4981 for( d = 0; d < 4; d++ ) {
4982 const int p0 = pix[-1];
4983 const int p1 = pix[-2];
4984 const int p2 = pix[-3];
4985 const int q0 = pix[0];
4986 const int q1 = pix[1];
4987 const int q2 = pix[2];
4989 if( ABS( p0 - q0 ) < alpha &&
4990 ABS( p1 - p0 ) < beta &&
4991 ABS( q1 - q0 ) < beta ) {
4995 if( ABS( p2 - p0 ) < beta ) {
4996 pix[-2] = p1 + clip( ( p2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( p1 << 1 ) ) >> 1, -tc0, tc0 );
4999 if( ABS( q2 - q0 ) < beta ) {
5000 pix[1] = q1 + clip( ( q2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( q1 << 1 ) ) >> 1, -tc0, tc0 );
5004 i_delta = clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );
5005 pix[-1] = clip_uint8( p0 + i_delta ); /* p0' */
5006 pix[0] = clip_uint8( q0 - i_delta ); /* q0' */
5011 /* 4px edge length */
5012 for( d = 0; d < 4; d++ ) {
5013 const int p0 = pix[-1];
5014 const int p1 = pix[-2];
5015 const int p2 = pix[-3];
5017 const int q0 = pix[0];
5018 const int q1 = pix[1];
5019 const int q2 = pix[2];
5021 if( ABS( p0 - q0 ) < alpha &&
5022 ABS( p1 - p0 ) < beta &&
5023 ABS( q1 - q0 ) < beta ) {
5025 if(ABS( p0 - q0 ) < (( alpha >> 2 ) + 2 )){
5026 if( ABS( p2 - p0 ) < beta)
5028 const int p3 = pix[-4];
5030 pix[-1] = ( p2 + 2*p1 + 2*p0 + 2*q0 + q1 + 4 ) >> 3;
5031 pix[-2] = ( p2 + p1 + p0 + q0 + 2 ) >> 2;
5032 pix[-3] = ( 2*p3 + 3*p2 + p1 + p0 + q0 + 4 ) >> 3;
5035 pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
5037 if( ABS( q2 - q0 ) < beta)
5039 const int q3 = pix[3];
5041 pix[0] = ( p1 + 2*p0 + 2*q0 + 2*q1 + q2 + 4 ) >> 3;
5042 pix[1] = ( p0 + q0 + q1 + q2 + 2 ) >> 2;
5043 pix[2] = ( 2*q3 + 3*q2 + q1 + q0 + p0 + 4 ) >> 3;
5046 pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
5050 pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
5051 pix[ 0] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
5059 static void filter_mb_edgecv( H264Context *h, uint8_t *pix, int stride, int bS[4], int qp ) {
5061 const int index_a = clip( qp + h->slice_alpha_c0_offset, 0, 51 );
5062 const int alpha = alpha_table[index_a];
5063 const int beta = beta_table[clip( qp + h->slice_beta_offset, 0, 51 )];
5065 for( i = 0; i < 4; i++ ) {
5072 const int tc = tc0_table[index_a][bS[i] - 1] + 1;
5073 /* 2px edge length (because we use same bS than the one for luma) */
5074 for( d = 0; d < 2; d++ ){
5075 const int p0 = pix[-1];
5076 const int p1 = pix[-2];
5077 const int q0 = pix[0];
5078 const int q1 = pix[1];
5080 if( ABS( p0 - q0 ) < alpha &&
5081 ABS( p1 - p0 ) < beta &&
5082 ABS( q1 - q0 ) < beta ) {
5083 const int i_delta = clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );
5085 pix[-1] = clip_uint8( p0 + i_delta ); /* p0' */
5086 pix[0] = clip_uint8( q0 - i_delta ); /* q0' */
5091 /* 2px edge length (because we use same bS than the one for luma) */
5092 for( d = 0; d < 2; d++ ){
5093 const int p0 = pix[-1];
5094 const int p1 = pix[-2];
5095 const int q0 = pix[0];
5096 const int q1 = pix[1];
5098 if( ABS( p0 - q0 ) < alpha &&
5099 ABS( p1 - p0 ) < beta &&
5100 ABS( q1 - q0 ) < beta ) {
5102 pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2; /* p0' */
5103 pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2; /* q0' */
5111 static void filter_mb_edgeh( H264Context *h, uint8_t *pix, int stride, int bS[4], int qp ) {
5113 const int index_a = clip( qp + h->slice_alpha_c0_offset, 0, 51 );
5114 const int alpha = alpha_table[index_a];
5115 const int beta = beta_table[clip( qp + h->slice_beta_offset, 0, 51 )];
5116 const int pix_next = stride;
5118 for( i = 0; i < 4; i++ ) {
5125 const int tc0 = tc0_table[index_a][bS[i] - 1];
5126 /* 4px edge length */
5127 for( d = 0; d < 4; d++ ) {
5128 const int p0 = pix[-1*pix_next];
5129 const int p1 = pix[-2*pix_next];
5130 const int p2 = pix[-3*pix_next];
5131 const int q0 = pix[0];
5132 const int q1 = pix[1*pix_next];
5133 const int q2 = pix[2*pix_next];
5135 if( ABS( p0 - q0 ) < alpha &&
5136 ABS( p1 - p0 ) < beta &&
5137 ABS( q1 - q0 ) < beta ) {
5142 if( ABS( p2 - p0 ) < beta ) {
5143 pix[-2*pix_next] = p1 + clip( ( p2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( p1 << 1 ) ) >> 1, -tc0, tc0 );
5146 if( ABS( q2 - q0 ) < beta ) {
5147 pix[pix_next] = q1 + clip( ( q2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( q1 << 1 ) ) >> 1, -tc0, tc0 );
5151 i_delta = clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );
5152 pix[-pix_next] = clip_uint8( p0 + i_delta ); /* p0' */
5153 pix[0] = clip_uint8( q0 - i_delta ); /* q0' */
5158 /* 4px edge length */
5159 for( d = 0; d < 4; d++ ) {
5160 const int p0 = pix[-1*pix_next];
5161 const int p1 = pix[-2*pix_next];
5162 const int p2 = pix[-3*pix_next];
5163 const int q0 = pix[0];
5164 const int q1 = pix[1*pix_next];
5165 const int q2 = pix[2*pix_next];
5167 if( ABS( p0 - q0 ) < alpha &&
5168 ABS( p1 - p0 ) < beta &&
5169 ABS( q1 - q0 ) < beta ) {
5171 const int p3 = pix[-4*pix_next];
5172 const int q3 = pix[ 3*pix_next];
5174 if(ABS( p0 - q0 ) < (( alpha >> 2 ) + 2 )){
5175 if( ABS( p2 - p0 ) < beta) {
5177 pix[-1*pix_next] = ( p2 + 2*p1 + 2*p0 + 2*q0 + q1 + 4 ) >> 3;
5178 pix[-2*pix_next] = ( p2 + p1 + p0 + q0 + 2 ) >> 2;
5179 pix[-3*pix_next] = ( 2*p3 + 3*p2 + p1 + p0 + q0 + 4 ) >> 3;
5182 pix[-1*pix_next] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
5184 if( ABS( q2 - q0 ) < beta) {
5186 pix[0*pix_next] = ( p1 + 2*p0 + 2*q0 + 2*q1 + q2 + 4 ) >> 3;
5187 pix[1*pix_next] = ( p0 + q0 + q1 + q2 + 2 ) >> 2;
5188 pix[2*pix_next] = ( 2*q3 + 3*q2 + q1 + q0 + p0 + 4 ) >> 3;
5191 pix[0*pix_next] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
5195 pix[-1*pix_next] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
5196 pix[ 0*pix_next] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
5205 static void filter_mb_edgech( H264Context *h, uint8_t *pix, int stride, int bS[4], int qp ) {
5207 const int index_a = clip( qp + h->slice_alpha_c0_offset, 0, 51 );
5208 const int alpha = alpha_table[index_a];
5209 const int beta = beta_table[clip( qp + h->slice_beta_offset, 0, 51 )];
5210 const int pix_next = stride;
5212 for( i = 0; i < 4; i++ )
5220 int tc = tc0_table[index_a][bS[i] - 1] + 1;
5221 /* 2px edge length (see deblocking_filter_edgecv) */
5222 for( d = 0; d < 2; d++ ) {
5223 const int p0 = pix[-1*pix_next];
5224 const int p1 = pix[-2*pix_next];
5225 const int q0 = pix[0];
5226 const int q1 = pix[1*pix_next];
5228 if( ABS( p0 - q0 ) < alpha &&
5229 ABS( p1 - p0 ) < beta &&
5230 ABS( q1 - q0 ) < beta ) {
5232 int i_delta = clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );
5234 pix[-pix_next] = clip_uint8( p0 + i_delta ); /* p0' */
5235 pix[0] = clip_uint8( q0 - i_delta ); /* q0' */
5240 /* 2px edge length (see deblocking_filter_edgecv) */
5241 for( d = 0; d < 2; d++ ) {
5242 const int p0 = pix[-1*pix_next];
5243 const int p1 = pix[-2*pix_next];
5244 const int q0 = pix[0];
5245 const int q1 = pix[1*pix_next];
5247 if( ABS( p0 - q0 ) < alpha &&
5248 ABS( p1 - p0 ) < beta &&
5249 ABS( q1 - q0 ) < beta ) {
5251 pix[-pix_next] = ( 2*p1 + p0 + q1 + 2 ) >> 2; /* p0' */
5252 pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2; /* q0' */
5260 static void filter_mb( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr) {
5261 MpegEncContext * const s = &h->s;
5262 const int mb_xy= mb_x + mb_y*s->mb_stride;
5263 int linesize, uvlinesize;
5266 /* FIXME Implement deblocking filter for field MB */
5267 if( h->sps.mb_aff ) {
5270 linesize = s->linesize;
5271 uvlinesize = s->uvlinesize;
5273 /* dir : 0 -> vertical edge, 1 -> horizontal edge */
5274 for( dir = 0; dir < 2; dir++ )
5279 /* test picture boundary */
5280 if( ( dir == 0 && mb_x == 0 ) || ( dir == 1 && mb_y == 0 ) ) {
5283 /* FIXME test slice boundary */
5284 if( h->deblocking_filter == 2 ) {
5288 for( edge = start; edge < 4; edge++ ) {
5289 /* mbn_xy: neighbour macroblock (how that works for field ?) */
5290 int mbn_xy = edge > 0 ? mb_xy : ( dir == 0 ? mb_xy -1 : mb_xy - s->mb_stride );
5294 if( IS_INTRA( s->current_picture.mb_type[mb_xy] ) ||
5295 IS_INTRA( s->current_picture.mb_type[mbn_xy] ) ) {
5296 bS[0] = bS[1] = bS[2] = bS[3] = ( edge == 0 ? 4 : 3 );
5299 for( i = 0; i < 4; i++ ) {
5300 int x = dir == 0 ? edge : i;
5301 int y = dir == 0 ? i : edge;
5302 int b_idx= 8 + 4 + x + 8*y;
5303 int bn_idx= b_idx - (dir ? 8:1);
5305 if( h->non_zero_count_cache[b_idx] != 0 ||
5306 h->non_zero_count_cache[bn_idx] != 0 ) {
5309 else if( h->slice_type == P_TYPE ) {
5310 if( h->ref_cache[0][b_idx] != h->ref_cache[0][bn_idx] ||
5311 ABS( h->mv_cache[0][b_idx][0] - h->mv_cache[0][bn_idx][0] ) >= 4 ||
5312 ABS( h->mv_cache[0][b_idx][1] - h->mv_cache[0][bn_idx][1] ) >= 4 )
5318 /* FIXME Add support for B frame */
5323 if(bS[0]+bS[1]+bS[2]+bS[3] == 0)
5328 qp = ( s->qscale + s->current_picture.qscale_table[mbn_xy] + 1 ) >> 1;
5330 filter_mb_edgev( h, &img_y[4*edge], linesize, bS, qp );
5331 if( (edge&1) == 0 ) {
5332 int chroma_qp = ( h->chroma_qp +
5333 get_chroma_qp( h, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1;
5334 filter_mb_edgecv( h, &img_cb[2*edge], uvlinesize, bS, chroma_qp );
5335 filter_mb_edgecv( h, &img_cr[2*edge], uvlinesize, bS, chroma_qp );
5338 filter_mb_edgeh( h, &img_y[4*edge*linesize], linesize, bS, qp );
5339 if( (edge&1) == 0 ) {
5340 int chroma_qp = ( h->chroma_qp +
5341 get_chroma_qp( h, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1;
5342 filter_mb_edgech( h, &img_cb[2*edge*uvlinesize], uvlinesize, bS, chroma_qp );
5343 filter_mb_edgech( h, &img_cr[2*edge*uvlinesize], uvlinesize, bS, chroma_qp );
5350 static int decode_slice(H264Context *h){
5351 MpegEncContext * const s = &h->s;
5352 const int part_mask= s->partitioned_frame ? (AC_END|AC_ERROR) : 0x7F;
5356 if( h->pps.cabac ) {
5360 align_get_bits( &s->gb );
5363 ff_init_cabac_states( &h->cabac, ff_h264_lps_range, ff_h264_mps_state, ff_h264_lps_state, 64 );
5364 ff_init_cabac_decoder( &h->cabac,
5365 s->gb.buffer + get_bits_count(&s->gb)/8,
5366 ( s->gb.size_in_bits - get_bits_count(&s->gb) + 7)/8);
5367 /* calculate pre-state */
5368 for( i= 0; i < 399; i++ ) {
5370 if( h->slice_type == I_TYPE )
5371 pre = clip( ((cabac_context_init_I[i][0] * s->qscale) >>4 ) + cabac_context_init_I[i][1], 1, 126 );
5373 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 );
5376 h->cabac_state[i] = 2 * ( 63 - pre ) + 0;
5378 h->cabac_state[i] = 2 * ( pre - 64 ) + 1;
5382 int ret = decode_mb_cabac(h);
5383 int eos = get_cabac_terminate( &h->cabac ); /* End of Slice flag */
5385 if(ret>=0) hl_decode_mb(h);
5387 /* XXX: useless as decode_mb_cabac it doesn't support that ... */
5388 if( ret >= 0 && h->sps.mb_aff ) { //FIXME optimal? or let mb_decode decode 16x32 ?
5391 if(ret>=0) ret = decode_mb_cabac(h);
5392 eos = get_cabac_terminate( &h->cabac );
5398 if( ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 1) {
5399 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
5400 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);
5404 if( ++s->mb_x >= s->mb_width ) {
5406 ff_draw_horiz_band(s, 16*s->mb_y, 16);
5410 if( eos || s->mb_y >= s->mb_height ) {
5411 tprintf("slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
5412 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);
5416 /* TODO test over-reading in cabac code */
5417 else if( read too much in h->cabac ) {
5418 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);
5426 int ret = decode_mb_cavlc(h);
5428 if(ret>=0) hl_decode_mb(h);
5430 if(ret>=0 && h->sps.mb_aff){ //FIXME optimal? or let mb_decode decode 16x32 ?
5432 ret = decode_mb_cavlc(h);
5434 if(ret>=0) hl_decode_mb(h);
5439 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
5440 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);
5445 if(++s->mb_x >= s->mb_width){
5447 ff_draw_horiz_band(s, 16*s->mb_y, 16);
5448 if(++s->mb_y >= s->mb_height){
5449 tprintf("slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
5451 if(get_bits_count(&s->gb) == s->gb.size_in_bits ) {
5452 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);
5456 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);
5463 if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->mb_skip_run<=0){
5464 tprintf("slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
5465 if(get_bits_count(&s->gb) == s->gb.size_in_bits ){
5466 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);
5470 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);
5479 for(;s->mb_y < s->mb_height; s->mb_y++){
5480 for(;s->mb_x < s->mb_width; s->mb_x++){
5481 int ret= decode_mb(h);
5486 fprintf(stderr, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
5487 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);
5492 if(++s->mb_x >= s->mb_width){
5494 if(++s->mb_y >= s->mb_height){
5495 if(get_bits_count(s->gb) == s->gb.size_in_bits){
5496 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);
5500 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);
5507 if(get_bits_count(s->?gb) >= s->gb?.size_in_bits){
5508 if(get_bits_count(s->gb) == s->gb.size_in_bits){
5509 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);
5513 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);
5520 ff_draw_horiz_band(s, 16*s->mb_y, 16);
5523 return -1; //not reached
5526 static inline int decode_vui_parameters(H264Context *h, SPS *sps){
5527 MpegEncContext * const s = &h->s;
5528 int aspect_ratio_info_present_flag, aspect_ratio_idc;
5530 aspect_ratio_info_present_flag= get_bits1(&s->gb);
5532 if( aspect_ratio_info_present_flag ) {
5533 aspect_ratio_idc= get_bits(&s->gb, 8);
5534 if( aspect_ratio_idc == EXTENDED_SAR ) {
5535 sps->sar.num= get_bits(&s->gb, 16);
5536 sps->sar.den= get_bits(&s->gb, 16);
5537 }else if(aspect_ratio_idc < 16){
5538 sps->sar= pixel_aspect[aspect_ratio_idc];
5540 av_log(h->s.avctx, AV_LOG_ERROR, "illegal aspect ratio\n");
5547 // s->avctx->aspect_ratio= sar_width*s->width / (float)(s->height*sar_height);
5549 if(get_bits1(&s->gb)){ /* overscan_info_present_flag */
5550 get_bits1(&s->gb); /* overscan_appropriate_flag */
5553 if(get_bits1(&s->gb)){ /* video_signal_type_present_flag */
5554 get_bits(&s->gb, 3); /* video_format */
5555 get_bits1(&s->gb); /* video_full_range_flag */
5556 if(get_bits1(&s->gb)){ /* colour_description_present_flag */
5557 get_bits(&s->gb, 8); /* colour_primaries */
5558 get_bits(&s->gb, 8); /* transfer_characteristics */
5559 get_bits(&s->gb, 8); /* matrix_coefficients */
5563 if(get_bits1(&s->gb)){ /* chroma_location_info_present_flag */
5564 get_ue_golomb(&s->gb); /* chroma_sample_location_type_top_field */
5565 get_ue_golomb(&s->gb); /* chroma_sample_location_type_bottom_field */
5568 sps->timing_info_present_flag = get_bits1(&s->gb);
5569 if(sps->timing_info_present_flag){
5570 sps->num_units_in_tick = get_bits_long(&s->gb, 32);
5571 sps->time_scale = get_bits_long(&s->gb, 32);
5572 sps->fixed_frame_rate_flag = get_bits1(&s->gb);
5576 | nal_hrd_parameters_present_flag |0 |u(1) |
5577 | if( nal_hrd_parameters_present_flag = = 1) | | |
5578 | hrd_parameters( ) | | |
5579 | vcl_hrd_parameters_present_flag |0 |u(1) |
5580 | if( vcl_hrd_parameters_present_flag = = 1) | | |
5581 | hrd_parameters( ) | | |
5582 | if( ( nal_hrd_parameters_present_flag = = 1 | || | |
5584 |( vcl_hrd_parameters_present_flag = = 1 ) ) | | |
5585 | low_delay_hrd_flag |0 |u(1) |
5586 | bitstream_restriction_flag |0 |u(1) |
5587 | if( bitstream_restriction_flag ) { |0 |u(1) |
5588 | motion_vectors_over_pic_boundaries_flag |0 |u(1) |
5589 | max_bytes_per_pic_denom |0 |ue(v) |
5590 | max_bits_per_mb_denom |0 |ue(v) |
5591 | log2_max_mv_length_horizontal |0 |ue(v) |
5592 | log2_max_mv_length_vertical |0 |ue(v) |
5593 | num_reorder_frames |0 |ue(v) |
5594 | max_dec_frame_buffering |0 |ue(v) |
5601 static inline int decode_seq_parameter_set(H264Context *h){
5602 MpegEncContext * const s = &h->s;
5603 int profile_idc, level_idc;
5607 profile_idc= get_bits(&s->gb, 8);
5608 get_bits1(&s->gb); //constraint_set0_flag
5609 get_bits1(&s->gb); //constraint_set1_flag
5610 get_bits1(&s->gb); //constraint_set2_flag
5611 get_bits1(&s->gb); //constraint_set3_flag
5612 get_bits(&s->gb, 4); // reserved
5613 level_idc= get_bits(&s->gb, 8);
5614 sps_id= get_ue_golomb(&s->gb);
5616 sps= &h->sps_buffer[ sps_id ];
5617 sps->profile_idc= profile_idc;
5618 sps->level_idc= level_idc;
5620 sps->log2_max_frame_num= get_ue_golomb(&s->gb) + 4;
5621 sps->poc_type= get_ue_golomb(&s->gb);
5623 if(sps->poc_type == 0){ //FIXME #define
5624 sps->log2_max_poc_lsb= get_ue_golomb(&s->gb) + 4;
5625 } else if(sps->poc_type == 1){//FIXME #define
5626 sps->delta_pic_order_always_zero_flag= get_bits1(&s->gb);
5627 sps->offset_for_non_ref_pic= get_se_golomb(&s->gb);
5628 sps->offset_for_top_to_bottom_field= get_se_golomb(&s->gb);
5629 sps->poc_cycle_length= get_ue_golomb(&s->gb);
5631 for(i=0; i<sps->poc_cycle_length; i++)
5632 sps->offset_for_ref_frame[i]= get_se_golomb(&s->gb);
5634 if(sps->poc_type > 2){
5635 av_log(h->s.avctx, AV_LOG_ERROR, "illegal POC type %d\n", sps->poc_type);
5639 sps->ref_frame_count= get_ue_golomb(&s->gb);
5640 if(sps->ref_frame_count > MAX_PICTURE_COUNT-2){
5641 av_log(h->s.avctx, AV_LOG_ERROR, "too many reference frames\n");
5643 sps->gaps_in_frame_num_allowed_flag= get_bits1(&s->gb);
5644 sps->mb_width= get_ue_golomb(&s->gb) + 1;
5645 sps->mb_height= get_ue_golomb(&s->gb) + 1;
5646 sps->frame_mbs_only_flag= get_bits1(&s->gb);
5647 if(!sps->frame_mbs_only_flag)
5648 sps->mb_aff= get_bits1(&s->gb);
5652 sps->direct_8x8_inference_flag= get_bits1(&s->gb);
5654 sps->crop= get_bits1(&s->gb);
5656 sps->crop_left = get_ue_golomb(&s->gb);
5657 sps->crop_right = get_ue_golomb(&s->gb);
5658 sps->crop_top = get_ue_golomb(&s->gb);
5659 sps->crop_bottom= get_ue_golomb(&s->gb);
5660 if(sps->crop_left || sps->crop_top){
5661 av_log(h->s.avctx, AV_LOG_ERROR, "insane cropping not completly supported, this could look slightly wrong ...\n");
5667 sps->crop_bottom= 0;
5670 sps->vui_parameters_present_flag= get_bits1(&s->gb);
5671 if( sps->vui_parameters_present_flag )
5672 decode_vui_parameters(h, sps);
5674 if(s->avctx->debug&FF_DEBUG_PICT_INFO){
5675 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",
5676 sps_id, sps->profile_idc, sps->level_idc,
5678 sps->ref_frame_count,
5679 sps->mb_width, sps->mb_height,
5680 sps->frame_mbs_only_flag ? "FRM" : (sps->mb_aff ? "MB-AFF" : "PIC-AFF"),
5681 sps->direct_8x8_inference_flag ? "8B8" : "",
5682 sps->crop_left, sps->crop_right,
5683 sps->crop_top, sps->crop_bottom,
5684 sps->vui_parameters_present_flag ? "VUI" : ""
5690 static inline int decode_picture_parameter_set(H264Context *h){
5691 MpegEncContext * const s = &h->s;
5692 int pps_id= get_ue_golomb(&s->gb);
5693 PPS *pps= &h->pps_buffer[pps_id];
5695 pps->sps_id= get_ue_golomb(&s->gb);
5696 pps->cabac= get_bits1(&s->gb);
5697 pps->pic_order_present= get_bits1(&s->gb);
5698 pps->slice_group_count= get_ue_golomb(&s->gb) + 1;
5699 if(pps->slice_group_count > 1 ){
5700 pps->mb_slice_group_map_type= get_ue_golomb(&s->gb);
5701 av_log(h->s.avctx, AV_LOG_ERROR, "FMO not supported\n");
5702 switch(pps->mb_slice_group_map_type){
5705 | for( i = 0; i <= num_slice_groups_minus1; i++ ) | | |
5706 | run_length[ i ] |1 |ue(v) |
5711 | for( i = 0; i < num_slice_groups_minus1; i++ ) | | |
5713 | top_left_mb[ i ] |1 |ue(v) |
5714 | bottom_right_mb[ i ] |1 |ue(v) |
5722 | slice_group_change_direction_flag |1 |u(1) |
5723 | slice_group_change_rate_minus1 |1 |ue(v) |
5728 | slice_group_id_cnt_minus1 |1 |ue(v) |
5729 | for( i = 0; i <= slice_group_id_cnt_minus1; i++ | | |
5731 | slice_group_id[ i ] |1 |u(v) |
5736 pps->ref_count[0]= get_ue_golomb(&s->gb) + 1;
5737 pps->ref_count[1]= get_ue_golomb(&s->gb) + 1;
5738 if(pps->ref_count[0] > 32 || pps->ref_count[1] > 32){
5739 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow (pps)\n");
5743 pps->weighted_pred= get_bits1(&s->gb);
5744 pps->weighted_bipred_idc= get_bits(&s->gb, 2);
5745 pps->init_qp= get_se_golomb(&s->gb) + 26;
5746 pps->init_qs= get_se_golomb(&s->gb) + 26;
5747 pps->chroma_qp_index_offset= get_se_golomb(&s->gb);
5748 pps->deblocking_filter_parameters_present= get_bits1(&s->gb);
5749 pps->constrained_intra_pred= get_bits1(&s->gb);
5750 pps->redundant_pic_cnt_present = get_bits1(&s->gb);
5752 if(s->avctx->debug&FF_DEBUG_PICT_INFO){
5753 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",
5754 pps_id, pps->sps_id,
5755 pps->cabac ? "CABAC" : "CAVLC",
5756 pps->slice_group_count,
5757 pps->ref_count[0], pps->ref_count[1],
5758 pps->weighted_pred ? "weighted" : "",
5759 pps->init_qp, pps->init_qs, pps->chroma_qp_index_offset,
5760 pps->deblocking_filter_parameters_present ? "LPAR" : "",
5761 pps->constrained_intra_pred ? "CONSTR" : "",
5762 pps->redundant_pic_cnt_present ? "REDU" : ""
5770 * finds the end of the current frame in the bitstream.
5771 * @return the position of the first byte of the next frame, or -1
5773 static int find_frame_end(H264Context *h, const uint8_t *buf, int buf_size){
5776 ParseContext *pc = &(h->s.parse_context);
5777 //printf("first %02X%02X%02X%02X\n", buf[0], buf[1],buf[2],buf[3]);
5778 // mb_addr= pc->mb_addr - 1;
5780 for(i=0; i<=buf_size; i++){
5781 if((state&0xFFFFFF1F) == 0x101 || (state&0xFFFFFF1F) == 0x102 || (state&0xFFFFFF1F) == 0x105){
5782 tprintf("find_frame_end new startcode = %08x, frame_start_found = %d, pos = %d\n", state, pc->frame_start_found, i);
5783 if(pc->frame_start_found){
5784 // If there isn't one more byte in the buffer
5785 // the test on first_mb_in_slice cannot be done yet
5786 // do it at next call.
5787 if (i >= buf_size) break;
5788 if (buf[i] & 0x80) {
5789 // first_mb_in_slice is 0, probably the first nal of a new
5791 tprintf("find_frame_end frame_end_found, state = %08x, pos = %d\n", state, i);
5793 pc->frame_start_found= 0;
5797 pc->frame_start_found = 1;
5800 state= (state<<8) | buf[i];
5804 return END_NOT_FOUND;
5807 static int h264_parse(AVCodecParserContext *s,
5808 AVCodecContext *avctx,
5809 uint8_t **poutbuf, int *poutbuf_size,
5810 const uint8_t *buf, int buf_size)
5812 H264Context *h = s->priv_data;
5813 ParseContext *pc = &h->s.parse_context;
5816 next= find_frame_end(h, buf, buf_size);
5818 if (ff_combine_frame(pc, next, (uint8_t **)&buf, &buf_size) < 0) {
5824 *poutbuf = (uint8_t *)buf;
5825 *poutbuf_size = buf_size;
5829 static int decode_nal_units(H264Context *h, uint8_t *buf, int buf_size){
5830 MpegEncContext * const s = &h->s;
5831 AVCodecContext * const avctx= s->avctx;
5835 for(i=0; i<32; i++){
5836 printf("%X ", buf[i]);
5848 if(buf_index >= buf_size) break;
5850 for(i = 0; i < h->nal_length_size; i++)
5851 nalsize = (nalsize << 8) | buf[buf_index++];
5853 // start code prefix search
5854 for(; buf_index + 3 < buf_size; buf_index++){
5855 // this should allways succeed in the first iteration
5856 if(buf[buf_index] == 0 && buf[buf_index+1] == 0 && buf[buf_index+2] == 1)
5860 if(buf_index+3 >= buf_size) break;
5865 ptr= decode_nal(h, buf + buf_index, &dst_length, &consumed, buf_size - buf_index);
5866 if(ptr[dst_length - 1] == 0) dst_length--;
5867 bit_length= 8*dst_length - decode_rbsp_trailing(ptr + dst_length - 1);
5869 if(s->avctx->debug&FF_DEBUG_STARTCODE){
5870 av_log(h->s.avctx, AV_LOG_DEBUG, "NAL %d at %d length %d\n", h->nal_unit_type, buf_index, dst_length);
5873 if (h->is_avc && (nalsize != consumed))
5874 av_log(h->s.avctx, AV_LOG_ERROR, "AVC: Consumed only %d bytes instead of %d\n", consumed, nalsize);
5876 buf_index += consumed;
5878 if( s->hurry_up == 1 && h->nal_ref_idc == 0 )
5881 switch(h->nal_unit_type){
5883 idr(h); //FIXME ensure we dont loose some frames if there is reordering
5885 init_get_bits(&s->gb, ptr, bit_length);
5887 h->inter_gb_ptr= &s->gb;
5888 s->data_partitioning = 0;
5890 if(decode_slice_header(h) < 0) return -1;
5891 if(h->redundant_pic_count==0 && s->hurry_up < 5 )
5895 init_get_bits(&s->gb, ptr, bit_length);
5897 h->inter_gb_ptr= NULL;
5898 s->data_partitioning = 1;
5900 if(decode_slice_header(h) < 0) return -1;
5903 init_get_bits(&h->intra_gb, ptr, bit_length);
5904 h->intra_gb_ptr= &h->intra_gb;
5907 init_get_bits(&h->inter_gb, ptr, bit_length);
5908 h->inter_gb_ptr= &h->inter_gb;
5910 if(h->redundant_pic_count==0 && h->intra_gb_ptr && s->data_partitioning && s->hurry_up < 5 )
5916 init_get_bits(&s->gb, ptr, bit_length);
5917 decode_seq_parameter_set(h);
5919 if(s->flags& CODEC_FLAG_LOW_DELAY)
5922 avctx->has_b_frames= !s->low_delay;
5925 init_get_bits(&s->gb, ptr, bit_length);
5927 decode_picture_parameter_set(h);
5930 case NAL_PICTURE_DELIMITER:
5932 case NAL_FILTER_DATA:
5935 av_log(avctx, AV_LOG_ERROR, "Unknown NAL code: %d\n", h->nal_unit_type);
5938 //FIXME move after where irt is set
5939 s->current_picture.pict_type= s->pict_type;
5940 s->current_picture.key_frame= s->pict_type == I_TYPE;
5943 if(!s->current_picture_ptr) return buf_index; //no frame
5945 h->prev_frame_num_offset= h->frame_num_offset;
5946 h->prev_frame_num= h->frame_num;
5947 if(s->current_picture_ptr->reference){
5948 h->prev_poc_msb= h->poc_msb;
5949 h->prev_poc_lsb= h->poc_lsb;
5951 if(s->current_picture_ptr->reference)
5952 execute_ref_pic_marking(h, h->mmco, h->mmco_index);
5954 assert(h->mmco_index==0);
5964 * retunrs the number of bytes consumed for building the current frame
5966 static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size){
5967 if(s->flags&CODEC_FLAG_TRUNCATED){
5968 pos -= s->parse_context.last_index;
5969 if(pos<0) pos=0; // FIXME remove (uneeded?)
5973 if(pos==0) pos=1; //avoid infinite loops (i doubt thats needed but ...)
5974 if(pos+10>buf_size) pos=buf_size; // oops ;)
5980 static int decode_frame(AVCodecContext *avctx,
5981 void *data, int *data_size,
5982 uint8_t *buf, int buf_size)
5984 H264Context *h = avctx->priv_data;
5985 MpegEncContext *s = &h->s;
5986 AVFrame *pict = data;
5989 s->flags= avctx->flags;
5990 s->flags2= avctx->flags2;
5992 /* no supplementary picture */
5993 if (buf_size == 0) {
5997 if(s->flags&CODEC_FLAG_TRUNCATED){
5998 int next= find_frame_end(h, buf, buf_size);
6000 if( ff_combine_frame(&s->parse_context, next, &buf, &buf_size) < 0 )
6002 //printf("next:%d buf_size:%d last_index:%d\n", next, buf_size, s->parse_context.last_index);
6005 if(h->is_avc && !h->got_avcC) {
6006 int i, cnt, nalsize;
6007 unsigned char *p = avctx->extradata;
6008 if(avctx->extradata_size < 7) {
6009 av_log(avctx, AV_LOG_ERROR, "avcC too short\n");
6013 av_log(avctx, AV_LOG_ERROR, "Unknown avcC version %d\n", *p);
6016 /* sps and pps in the avcC always have length coded with 2 bytes,
6017 so put a fake nal_length_size = 2 while parsing them */
6018 h->nal_length_size = 2;
6019 // Decode sps from avcC
6020 cnt = *(p+5) & 0x1f; // Number of sps
6022 for (i = 0; i < cnt; i++) {
6023 nalsize = BE_16(p) + 2;
6024 if(decode_nal_units(h, p, nalsize) != nalsize) {
6025 av_log(avctx, AV_LOG_ERROR, "Decoding sps %d from avcC failed\n", i);
6030 // Decode pps from avcC
6031 cnt = *(p++); // Number of pps
6032 for (i = 0; i < cnt; i++) {
6033 nalsize = BE_16(p) + 2;
6034 if(decode_nal_units(h, p, nalsize) != nalsize) {
6035 av_log(avctx, AV_LOG_ERROR, "Decoding pps %d from avcC failed\n", i);
6040 // Now store right nal length size, that will be use to parse all other nals
6041 h->nal_length_size = ((*(((char*)(avctx->extradata))+4))&0x03)+1;
6042 // Do not reparse avcC
6046 if(!h->is_avc && s->avctx->extradata_size && s->picture_number==0){
6047 if(0 < decode_nal_units(h, s->avctx->extradata, s->avctx->extradata_size) )
6051 buf_index=decode_nal_units(h, buf, buf_size);
6055 //FIXME do something with unavailable reference frames
6057 // if(ret==FRAME_SKIPED) return get_consumed_bytes(s, buf_index, buf_size);
6059 if(s->pict_type==B_TYPE || s->low_delay){
6060 *pict= *(AVFrame*)&s->current_picture;
6062 *pict= *(AVFrame*)&s->last_picture;
6065 if(!s->current_picture_ptr){
6066 av_log(h->s.avctx, AV_LOG_DEBUG, "error, NO frame\n");
6070 *pict= *(AVFrame*)&s->current_picture; //FIXME
6071 ff_print_debug_info(s, pict);
6072 assert(pict->data[0]);
6073 //printf("out %d\n", (int)pict->data[0]);
6076 /* Return the Picture timestamp as the frame number */
6077 /* we substract 1 because it is added on utils.c */
6078 avctx->frame_number = s->picture_number - 1;
6081 /* dont output the last pic after seeking */
6082 if(s->last_picture_ptr || s->low_delay)
6083 //Note this isnt a issue as a IDR pic should flush the buffers
6085 *data_size = sizeof(AVFrame);
6086 return get_consumed_bytes(s, buf_index, buf_size);
6089 static inline void fill_mb_avail(H264Context *h){
6090 MpegEncContext * const s = &h->s;
6091 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
6094 h->mb_avail[0]= s->mb_x && h->slice_table[mb_xy - s->mb_stride - 1] == h->slice_num;
6095 h->mb_avail[1]= h->slice_table[mb_xy - s->mb_stride ] == h->slice_num;
6096 h->mb_avail[2]= s->mb_x+1 < s->mb_width && h->slice_table[mb_xy - s->mb_stride + 1] == h->slice_num;
6102 h->mb_avail[3]= s->mb_x && h->slice_table[mb_xy - 1] == h->slice_num;
6103 h->mb_avail[4]= 1; //FIXME move out
6104 h->mb_avail[5]= 0; //FIXME move out
6110 #define SIZE (COUNT*40)
6116 // int int_temp[10000];
6118 AVCodecContext avctx;
6120 dsputil_init(&dsp, &avctx);
6122 init_put_bits(&pb, temp, SIZE);
6123 printf("testing unsigned exp golomb\n");
6124 for(i=0; i<COUNT; i++){
6126 set_ue_golomb(&pb, i);
6127 STOP_TIMER("set_ue_golomb");
6129 flush_put_bits(&pb);
6131 init_get_bits(&gb, temp, 8*SIZE);
6132 for(i=0; i<COUNT; i++){
6135 s= show_bits(&gb, 24);
6138 j= get_ue_golomb(&gb);
6140 printf("missmatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
6143 STOP_TIMER("get_ue_golomb");
6147 init_put_bits(&pb, temp, SIZE);
6148 printf("testing signed exp golomb\n");
6149 for(i=0; i<COUNT; i++){
6151 set_se_golomb(&pb, i - COUNT/2);
6152 STOP_TIMER("set_se_golomb");
6154 flush_put_bits(&pb);
6156 init_get_bits(&gb, temp, 8*SIZE);
6157 for(i=0; i<COUNT; i++){
6160 s= show_bits(&gb, 24);
6163 j= get_se_golomb(&gb);
6164 if(j != i - COUNT/2){
6165 printf("missmatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
6168 STOP_TIMER("get_se_golomb");
6171 printf("testing 4x4 (I)DCT\n");
6174 uint8_t src[16], ref[16];
6175 uint64_t error= 0, max_error=0;
6177 for(i=0; i<COUNT; i++){
6179 // printf("%d %d %d\n", r1, r2, (r2-r1)*16);
6180 for(j=0; j<16; j++){
6181 ref[j]= random()%255;
6182 src[j]= random()%255;
6185 h264_diff_dct_c(block, src, ref, 4);
6188 for(j=0; j<16; j++){
6189 // printf("%d ", block[j]);
6190 block[j]= block[j]*4;
6191 if(j&1) block[j]= (block[j]*4 + 2)/5;
6192 if(j&4) block[j]= (block[j]*4 + 2)/5;
6196 s->dsp.h264_idct_add(ref, block, 4);
6197 /* for(j=0; j<16; j++){
6198 printf("%d ", ref[j]);
6202 for(j=0; j<16; j++){
6203 int diff= ABS(src[j] - ref[j]);
6206 max_error= FFMAX(max_error, diff);
6209 printf("error=%f max_error=%d\n", ((float)error)/COUNT/16, (int)max_error );
6211 printf("testing quantizer\n");
6212 for(qp=0; qp<52; qp++){
6214 src1_block[i]= src2_block[i]= random()%255;
6218 printf("Testing NAL layer\n");
6220 uint8_t bitstream[COUNT];
6221 uint8_t nal[COUNT*2];
6223 memset(&h, 0, sizeof(H264Context));
6225 for(i=0; i<COUNT; i++){
6233 for(j=0; j<COUNT; j++){
6234 bitstream[j]= (random() % 255) + 1;
6237 for(j=0; j<zeros; j++){
6238 int pos= random() % COUNT;
6239 while(bitstream[pos] == 0){
6248 nal_length= encode_nal(&h, nal, bitstream, COUNT, COUNT*2);
6250 printf("encoding failed\n");
6254 out= decode_nal(&h, nal, &out_length, &consumed, nal_length);
6258 if(out_length != COUNT){
6259 printf("incorrect length %d %d\n", out_length, COUNT);
6263 if(consumed != nal_length){
6264 printf("incorrect consumed length %d %d\n", nal_length, consumed);
6268 if(memcmp(bitstream, out, COUNT)){
6269 printf("missmatch\n");
6274 printf("Testing RBSP\n");
6282 static int decode_end(AVCodecContext *avctx)
6284 H264Context *h = avctx->priv_data;
6285 MpegEncContext *s = &h->s;
6287 free_tables(h); //FIXME cleanup init stuff perhaps
6290 // memset(h, 0, sizeof(H264Context));
6296 AVCodec h264_decoder = {
6300 sizeof(H264Context),
6305 /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 | CODEC_CAP_TRUNCATED,
6308 AVCodecParser h264_parser = {
6310 sizeof(H264Context),