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] __align8;
186 uint8_t (*non_zero_count)[16];
189 * Motion vector cache.
191 int16_t mv_cache[2][5*8][2] __align8;
192 int8_t ref_cache[2][5*8] __align8;
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
259 int use_weight_chroma;
260 int luma_log2_weight_denom;
261 int chroma_log2_weight_denom;
262 int luma_weight[2][16];
263 int luma_offset[2][16];
264 int chroma_weight[2][16][2];
265 int chroma_offset[2][16][2];
266 int implicit_weight[16][16];
269 int deblocking_filter; ///< disable_deblocking_filter_idc with 1<->0
270 int slice_alpha_c0_offset;
271 int slice_beta_offset;
273 int redundant_pic_count;
275 int direct_spatial_mv_pred;
276 int dist_scale_factor[16];
279 * num_ref_idx_l0/1_active_minus1 + 1
281 int ref_count[2];// FIXME split for AFF
282 Picture *short_ref[16];
283 Picture *long_ref[16];
284 Picture default_ref_list[2][32];
285 Picture ref_list[2][32]; //FIXME size?
286 Picture field_ref_list[2][32]; //FIXME size?
287 Picture *delayed_pic[16]; //FIXME size?
290 * memory management control operations buffer.
292 MMCO mmco[MAX_MMCO_COUNT];
295 int long_ref_count; ///< number of actual long term references
296 int short_ref_count; ///< number of actual short term references
299 GetBitContext intra_gb;
300 GetBitContext inter_gb;
301 GetBitContext *intra_gb_ptr;
302 GetBitContext *inter_gb_ptr;
304 DCTELEM mb[16*24] __align8;
310 uint8_t cabac_state[399];
313 /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0,1,2), 0x0? luma_cbp */
317 /* chroma_pred_mode for i4x4 or i16x16, else 0 */
318 uint8_t *chroma_pred_mode_table;
319 int last_qscale_diff;
320 int16_t (*mvd_table[2])[2];
321 int16_t mvd_cache[2][5*8][2] __align8;
322 uint8_t *direct_table;
323 uint8_t direct_cache[5*8];
327 static VLC coeff_token_vlc[4];
328 static VLC chroma_dc_coeff_token_vlc;
330 static VLC total_zeros_vlc[15];
331 static VLC chroma_dc_total_zeros_vlc[3];
333 static VLC run_vlc[6];
336 static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
337 static void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
338 static void filter_mb( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr);
340 static inline uint32_t pack16to32(int a, int b){
341 #ifdef WORDS_BIGENDIAN
342 return (b&0xFFFF) + (a<<16);
344 return (a&0xFFFF) + (b<<16);
350 * @param h height of the rectangle, should be a constant
351 * @param w width of the rectangle, should be a constant
352 * @param size the size of val (1 or 4), should be a constant
354 static inline void fill_rectangle(void *vp, int w, int h, int stride, uint32_t val, int size){ //FIXME ensure this IS inlined
355 uint8_t *p= (uint8_t*)vp;
356 assert(size==1 || size==4);
361 assert((((int)vp)&(FFMIN(w, STRIDE_ALIGN)-1)) == 0);
362 //FIXME check what gcc generates for 64 bit on x86 and possible write a 32 bit ver of it
365 *(uint16_t*)(p + stride)= size==4 ? val : val*0x0101;
366 }else if(w==2 && h==4){
367 *(uint16_t*)(p + 0*stride)=
368 *(uint16_t*)(p + 1*stride)=
369 *(uint16_t*)(p + 2*stride)=
370 *(uint16_t*)(p + 3*stride)= size==4 ? val : val*0x0101;
371 }else if(w==4 && h==1){
372 *(uint32_t*)(p + 0*stride)= size==4 ? val : val*0x01010101;
373 }else if(w==4 && h==2){
374 *(uint32_t*)(p + 0*stride)=
375 *(uint32_t*)(p + 1*stride)= size==4 ? val : val*0x01010101;
376 }else if(w==4 && h==4){
377 *(uint32_t*)(p + 0*stride)=
378 *(uint32_t*)(p + 1*stride)=
379 *(uint32_t*)(p + 2*stride)=
380 *(uint32_t*)(p + 3*stride)= size==4 ? val : val*0x01010101;
381 }else if(w==8 && h==1){
383 *(uint32_t*)(p + 4)= size==4 ? val : val*0x01010101;
384 }else if(w==8 && h==2){
385 *(uint32_t*)(p + 0 + 0*stride)=
386 *(uint32_t*)(p + 4 + 0*stride)=
387 *(uint32_t*)(p + 0 + 1*stride)=
388 *(uint32_t*)(p + 4 + 1*stride)= size==4 ? val : val*0x01010101;
389 }else if(w==8 && h==4){
390 *(uint64_t*)(p + 0*stride)=
391 *(uint64_t*)(p + 1*stride)=
392 *(uint64_t*)(p + 2*stride)=
393 *(uint64_t*)(p + 3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
394 }else if(w==16 && h==2){
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)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
399 }else if(w==16 && h==4){
400 *(uint64_t*)(p + 0+0*stride)=
401 *(uint64_t*)(p + 8+0*stride)=
402 *(uint64_t*)(p + 0+1*stride)=
403 *(uint64_t*)(p + 8+1*stride)=
404 *(uint64_t*)(p + 0+2*stride)=
405 *(uint64_t*)(p + 8+2*stride)=
406 *(uint64_t*)(p + 0+3*stride)=
407 *(uint64_t*)(p + 8+3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
412 static inline void fill_caches(H264Context *h, int mb_type, int for_deblock){
413 MpegEncContext * const s = &h->s;
414 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
415 int topleft_xy, top_xy, topright_xy, left_xy[2];
416 int topleft_type, top_type, topright_type, left_type[2];
420 //wow what a mess, why didnt they simplify the interlacing&intra stuff, i cant imagine that these complex rules are worth it
424 topleft_xy = 0; /* avoid warning */
425 top_xy = 0; /* avoid warning */
426 topright_xy = 0; /* avoid warning */
428 topleft_xy = mb_xy-1 - s->mb_stride;
429 top_xy = mb_xy - s->mb_stride;
430 topright_xy= mb_xy+1 - s->mb_stride;
431 left_xy[0] = mb_xy-1;
432 left_xy[1] = mb_xy-1;
440 topleft_type = h->slice_table[topleft_xy ] < 255 ? s->current_picture.mb_type[topleft_xy] : 0;
441 top_type = h->slice_table[top_xy ] < 255 ? s->current_picture.mb_type[top_xy] : 0;
442 topright_type= h->slice_table[topright_xy] < 255 ? s->current_picture.mb_type[topright_xy]: 0;
443 left_type[0] = h->slice_table[left_xy[0] ] < 255 ? s->current_picture.mb_type[left_xy[0]] : 0;
444 left_type[1] = h->slice_table[left_xy[1] ] < 255 ? s->current_picture.mb_type[left_xy[1]] : 0;
446 topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
447 top_type = h->slice_table[top_xy ] == h->slice_num ? s->current_picture.mb_type[top_xy] : 0;
448 topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
449 left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
450 left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
453 if(IS_INTRA(mb_type)){
454 h->topleft_samples_available=
455 h->top_samples_available=
456 h->left_samples_available= 0xFFFF;
457 h->topright_samples_available= 0xEEEA;
459 if(!IS_INTRA(top_type) && (top_type==0 || h->pps.constrained_intra_pred)){
460 h->topleft_samples_available= 0xB3FF;
461 h->top_samples_available= 0x33FF;
462 h->topright_samples_available= 0x26EA;
465 if(!IS_INTRA(left_type[i]) && (left_type[i]==0 || h->pps.constrained_intra_pred)){
466 h->topleft_samples_available&= 0xDF5F;
467 h->left_samples_available&= 0x5F5F;
471 if(!IS_INTRA(topleft_type) && (topleft_type==0 || h->pps.constrained_intra_pred))
472 h->topleft_samples_available&= 0x7FFF;
474 if(!IS_INTRA(topright_type) && (topright_type==0 || h->pps.constrained_intra_pred))
475 h->topright_samples_available&= 0xFBFF;
477 if(IS_INTRA4x4(mb_type)){
478 if(IS_INTRA4x4(top_type)){
479 h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
480 h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
481 h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
482 h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
485 if(!top_type || (IS_INTER(top_type) && h->pps.constrained_intra_pred))
490 h->intra4x4_pred_mode_cache[4+8*0]=
491 h->intra4x4_pred_mode_cache[5+8*0]=
492 h->intra4x4_pred_mode_cache[6+8*0]=
493 h->intra4x4_pred_mode_cache[7+8*0]= pred;
496 if(IS_INTRA4x4(left_type[i])){
497 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
498 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
501 if(!left_type[i] || (IS_INTER(left_type[i]) && h->pps.constrained_intra_pred))
506 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
507 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
522 //FIXME constraint_intra_pred & partitioning & nnz (lets hope this is just a typo in the spec)
524 h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][0];
525 h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][1];
526 h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][2];
527 h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3];
529 h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][7];
530 h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8];
532 h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][10];
533 h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11];
535 h->top_cbp= h->cbp_table[top_xy];
537 h->non_zero_count_cache[4+8*0]=
538 h->non_zero_count_cache[5+8*0]=
539 h->non_zero_count_cache[6+8*0]=
540 h->non_zero_count_cache[7+8*0]=
542 h->non_zero_count_cache[1+8*0]=
543 h->non_zero_count_cache[2+8*0]=
545 h->non_zero_count_cache[1+8*3]=
546 h->non_zero_count_cache[2+8*3]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
548 if(IS_INTRA(mb_type)) h->top_cbp= 0x1C0;
553 h->non_zero_count_cache[3+8*1]= h->non_zero_count[left_xy[0]][6];
554 h->non_zero_count_cache[3+8*2]= h->non_zero_count[left_xy[0]][5];
555 h->non_zero_count_cache[0+8*1]= h->non_zero_count[left_xy[0]][9]; //FIXME left_block
556 h->non_zero_count_cache[0+8*4]= h->non_zero_count[left_xy[0]][12];
557 h->left_cbp= h->cbp_table[left_xy[0]]; //FIXME interlacing
559 h->non_zero_count_cache[3+8*1]=
560 h->non_zero_count_cache[3+8*2]=
561 h->non_zero_count_cache[0+8*1]=
562 h->non_zero_count_cache[0+8*4]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
564 if(IS_INTRA(mb_type)) h->left_cbp= 0x1C0;//FIXME interlacing
569 h->non_zero_count_cache[3+8*3]= h->non_zero_count[left_xy[1]][4];
570 h->non_zero_count_cache[3+8*4]= h->non_zero_count[left_xy[1]][3];
571 h->non_zero_count_cache[0+8*2]= h->non_zero_count[left_xy[1]][8];
572 h->non_zero_count_cache[0+8*5]= h->non_zero_count[left_xy[1]][11];
574 h->non_zero_count_cache[3+8*3]=
575 h->non_zero_count_cache[3+8*4]=
576 h->non_zero_count_cache[0+8*2]=
577 h->non_zero_count_cache[0+8*5]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
581 //FIXME direct mb can skip much of this
582 if(IS_INTER(mb_type) || (IS_DIRECT(mb_type) && h->direct_spatial_mv_pred)){
584 for(list=0; list<2; list++){
585 if(!USES_LIST(mb_type, list) && !IS_DIRECT(mb_type) && !for_deblock){
586 /*if(!h->mv_cache_clean[list]){
587 memset(h->mv_cache [list], 0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
588 memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
589 h->mv_cache_clean[list]= 1;
593 h->mv_cache_clean[list]= 0;
595 if(IS_INTER(topleft_type)){
596 const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
597 const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + h->b8_stride;
598 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
599 h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
601 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
602 h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
605 if(IS_INTER(top_type)){
606 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
607 const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
608 *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0];
609 *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];
610 *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];
611 *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];
612 h->ref_cache[list][scan8[0] + 0 - 1*8]=
613 h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
614 h->ref_cache[list][scan8[0] + 2 - 1*8]=
615 h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
617 *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]=
618 *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]=
619 *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]=
620 *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;
621 *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
624 if(IS_INTER(topright_type)){
625 const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
626 const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
627 *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
628 h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
630 *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
631 h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
634 //FIXME unify cleanup or sth
635 if(IS_INTER(left_type[0])){
636 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
637 const int b8_xy= h->mb2b8_xy[left_xy[0]] + 1;
638 *(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]];
639 *(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]];
640 h->ref_cache[list][scan8[0] - 1 + 0*8]=
641 h->ref_cache[list][scan8[0] - 1 + 1*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0]>>1)];
643 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 0*8]=
644 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 1*8]= 0;
645 h->ref_cache[list][scan8[0] - 1 + 0*8]=
646 h->ref_cache[list][scan8[0] - 1 + 1*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
649 if(IS_INTER(left_type[1])){
650 const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
651 const int b8_xy= h->mb2b8_xy[left_xy[1]] + 1;
652 *(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]];
653 *(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]];
654 h->ref_cache[list][scan8[0] - 1 + 2*8]=
655 h->ref_cache[list][scan8[0] - 1 + 3*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[2]>>1)];
657 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 2*8]=
658 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 3*8]= 0;
659 h->ref_cache[list][scan8[0] - 1 + 2*8]=
660 h->ref_cache[list][scan8[0] - 1 + 3*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
666 h->ref_cache[list][scan8[5 ]+1] =
667 h->ref_cache[list][scan8[7 ]+1] =
668 h->ref_cache[list][scan8[13]+1] = //FIXME remove past 3 (init somewher else)
669 h->ref_cache[list][scan8[4 ]] =
670 h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
671 *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
672 *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
673 *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewher else)
674 *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
675 *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
678 /* XXX beurk, Load mvd */
679 if(IS_INTER(topleft_type)){
680 const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
681 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy];
683 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 - 1*8]= 0;
686 if(IS_INTER(top_type)){
687 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
688 *(uint32_t*)h->mvd_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 0];
689 *(uint32_t*)h->mvd_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 1];
690 *(uint32_t*)h->mvd_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 2];
691 *(uint32_t*)h->mvd_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 3];
693 *(uint32_t*)h->mvd_cache [list][scan8[0] + 0 - 1*8]=
694 *(uint32_t*)h->mvd_cache [list][scan8[0] + 1 - 1*8]=
695 *(uint32_t*)h->mvd_cache [list][scan8[0] + 2 - 1*8]=
696 *(uint32_t*)h->mvd_cache [list][scan8[0] + 3 - 1*8]= 0;
698 if(IS_INTER(left_type[0])){
699 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
700 *(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]];
701 *(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]];
703 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 0*8]=
704 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 1*8]= 0;
706 if(IS_INTER(left_type[1])){
707 const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
708 *(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]];
709 *(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]];
711 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 2*8]=
712 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 3*8]= 0;
714 *(uint32_t*)h->mvd_cache [list][scan8[5 ]+1]=
715 *(uint32_t*)h->mvd_cache [list][scan8[7 ]+1]=
716 *(uint32_t*)h->mvd_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewher else)
717 *(uint32_t*)h->mvd_cache [list][scan8[4 ]]=
718 *(uint32_t*)h->mvd_cache [list][scan8[12]]= 0;
720 if(h->slice_type == B_TYPE){
721 fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, 0, 1);
723 if(IS_DIRECT(top_type)){
724 *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0x01010101;
725 }else if(IS_8X8(top_type)){
726 int b8_xy = h->mb2b8_xy[top_xy] + h->b8_stride;
727 h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy];
728 h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 1];
730 *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0;
734 if(IS_DIRECT(left_type[0])){
735 h->direct_cache[scan8[0] - 1 + 0*8]=
736 h->direct_cache[scan8[0] - 1 + 2*8]= 1;
737 }else if(IS_8X8(left_type[0])){
738 int b8_xy = h->mb2b8_xy[left_xy[0]] + 1;
739 h->direct_cache[scan8[0] - 1 + 0*8]= h->direct_table[b8_xy];
740 h->direct_cache[scan8[0] - 1 + 2*8]= h->direct_table[b8_xy + h->b8_stride];
742 h->direct_cache[scan8[0] - 1 + 0*8]=
743 h->direct_cache[scan8[0] - 1 + 2*8]= 0;
752 static inline void write_back_intra_pred_mode(H264Context *h){
753 MpegEncContext * const s = &h->s;
754 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
756 h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
757 h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
758 h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
759 h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
760 h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
761 h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
762 h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
766 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
768 static inline int check_intra4x4_pred_mode(H264Context *h){
769 MpegEncContext * const s = &h->s;
770 static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
771 static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
774 if(!(h->top_samples_available&0x8000)){
776 int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
778 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);
781 h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
786 if(!(h->left_samples_available&0x8000)){
788 int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
790 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);
793 h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
799 } //FIXME cleanup like next
802 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
804 static inline int check_intra_pred_mode(H264Context *h, int mode){
805 MpegEncContext * const s = &h->s;
806 static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
807 static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
809 if(mode < 0 || mode > 6) {
810 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);
814 if(!(h->top_samples_available&0x8000)){
817 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);
822 if(!(h->left_samples_available&0x8000)){
825 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);
834 * gets the predicted intra4x4 prediction mode.
836 static inline int pred_intra_mode(H264Context *h, int n){
837 const int index8= scan8[n];
838 const int left= h->intra4x4_pred_mode_cache[index8 - 1];
839 const int top = h->intra4x4_pred_mode_cache[index8 - 8];
840 const int min= FFMIN(left, top);
842 tprintf("mode:%d %d min:%d\n", left ,top, min);
844 if(min<0) return DC_PRED;
848 static inline void write_back_non_zero_count(H264Context *h){
849 MpegEncContext * const s = &h->s;
850 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
852 h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[4+8*4];
853 h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[5+8*4];
854 h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[6+8*4];
855 h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];
856 h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[7+8*3];
857 h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[7+8*2];
858 h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[7+8*1];
860 h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[1+8*2];
861 h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
862 h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[2+8*1];
864 h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[1+8*5];
865 h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
866 h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[2+8*4];
870 * gets the predicted number of non zero coefficients.
871 * @param n block index
873 static inline int pred_non_zero_count(H264Context *h, int n){
874 const int index8= scan8[n];
875 const int left= h->non_zero_count_cache[index8 - 1];
876 const int top = h->non_zero_count_cache[index8 - 8];
879 if(i<64) i= (i+1)>>1;
881 tprintf("pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
886 static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
887 const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];
889 if(topright_ref != PART_NOT_AVAILABLE){
890 *C= h->mv_cache[list][ i - 8 + part_width ];
893 tprintf("topright MV not available\n");
895 *C= h->mv_cache[list][ i - 8 - 1 ];
896 return h->ref_cache[list][ i - 8 - 1 ];
901 * gets the predicted MV.
902 * @param n the block index
903 * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)
904 * @param mx the x component of the predicted motion vector
905 * @param my the y component of the predicted motion vector
907 static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
908 const int index8= scan8[n];
909 const int top_ref= h->ref_cache[list][ index8 - 8 ];
910 const int left_ref= h->ref_cache[list][ index8 - 1 ];
911 const int16_t * const A= h->mv_cache[list][ index8 - 1 ];
912 const int16_t * const B= h->mv_cache[list][ index8 - 8 ];
914 int diagonal_ref, match_count;
916 assert(part_width==1 || part_width==2 || part_width==4);
926 diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);
927 match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);
928 tprintf("pred_motion match_count=%d\n", match_count);
929 if(match_count > 1){ //most common
930 *mx= mid_pred(A[0], B[0], C[0]);
931 *my= mid_pred(A[1], B[1], C[1]);
932 }else if(match_count==1){
936 }else if(top_ref==ref){
944 if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
948 *mx= mid_pred(A[0], B[0], C[0]);
949 *my= mid_pred(A[1], B[1], C[1]);
953 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);
957 * gets the directionally predicted 16x8 MV.
958 * @param n the block index
959 * @param mx the x component of the predicted motion vector
960 * @param my the y component of the predicted motion vector
962 static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
964 const int top_ref= h->ref_cache[list][ scan8[0] - 8 ];
965 const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
967 tprintf("pred_16x8: (%2d %2d %2d) at %2d %2d %d list %d\n", top_ref, B[0], B[1], h->s.mb_x, h->s.mb_y, n, list);
975 const int left_ref= h->ref_cache[list][ scan8[8] - 1 ];
976 const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];
978 tprintf("pred_16x8: (%2d %2d %2d) at %2d %2d %d list %d\n", left_ref, A[0], A[1], h->s.mb_x, h->s.mb_y, n, list);
988 pred_motion(h, n, 4, list, ref, mx, my);
992 * gets the directionally predicted 8x16 MV.
993 * @param n the block index
994 * @param mx the x component of the predicted motion vector
995 * @param my the y component of the predicted motion vector
997 static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
999 const int left_ref= h->ref_cache[list][ scan8[0] - 1 ];
1000 const int16_t * const A= h->mv_cache[list][ scan8[0] - 1 ];
1002 tprintf("pred_8x16: (%2d %2d %2d) at %2d %2d %d list %d\n", left_ref, A[0], A[1], h->s.mb_x, h->s.mb_y, n, list);
1004 if(left_ref == ref){
1013 diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);
1015 tprintf("pred_8x16: (%2d %2d %2d) at %2d %2d %d list %d\n", diagonal_ref, C[0], C[1], h->s.mb_x, h->s.mb_y, n, list);
1017 if(diagonal_ref == ref){
1025 pred_motion(h, n, 2, list, ref, mx, my);
1028 static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
1029 const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
1030 const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];
1032 tprintf("pred_pskip: (%d) (%d) at %2d %2d\n", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
1034 if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
1035 || (top_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0)
1036 || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){
1042 pred_motion(h, 0, 4, 0, 0, mx, my);
1047 static inline void direct_dist_scale_factor(H264Context * const h){
1048 const int poc = h->s.current_picture_ptr->poc;
1049 const int poc1 = h->ref_list[1][0].poc;
1051 for(i=0; i<h->ref_count[0]; i++){
1052 int poc0 = h->ref_list[0][i].poc;
1053 int td = clip(poc1 - poc0, -128, 127);
1054 if(td == 0 /* FIXME || pic0 is a long-term ref */){
1055 h->dist_scale_factor[i] = 256;
1057 int tb = clip(poc - poc0, -128, 127);
1058 int tx = (16384 + (ABS(td) >> 1)) / td;
1059 h->dist_scale_factor[i] = clip((tb*tx + 32) >> 6, -1024, 1023);
1064 static inline void pred_direct_motion(H264Context * const h, int *mb_type){
1065 MpegEncContext * const s = &h->s;
1066 const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
1067 const int b8_xy = 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1068 const int b4_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1069 const int mb_type_col = h->ref_list[1][0].mb_type[mb_xy];
1070 const int16_t (*l1mv0)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[0][b4_xy];
1071 const int8_t *l1ref0 = &h->ref_list[1][0].ref_index[0][b8_xy];
1072 const int is_b8x8 = IS_8X8(*mb_type);
1076 if(IS_8X8(mb_type_col) && !h->sps.direct_8x8_inference_flag){
1077 /* FIXME save sub mb types from previous frames (or derive from MVs)
1078 * so we know exactly what block size to use */
1079 sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */
1080 *mb_type = MB_TYPE_8x8|MB_TYPE_L0L1;
1081 }else if(!is_b8x8 && (IS_16X16(mb_type_col) || IS_INTRA(mb_type_col))){
1082 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1083 *mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */
1085 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1086 *mb_type = MB_TYPE_8x8|MB_TYPE_L0L1;
1089 *mb_type |= MB_TYPE_DIRECT2;
1091 tprintf("mb_type = %08x, sub_mb_type = %08x, is_b8x8 = %d, mb_type_col = %08x\n", *mb_type, sub_mb_type, is_b8x8, mb_type_col);
1093 if(h->direct_spatial_mv_pred){
1098 /* ref = min(neighbors) */
1099 for(list=0; list<2; list++){
1100 int refa = h->ref_cache[list][scan8[0] - 1];
1101 int refb = h->ref_cache[list][scan8[0] - 8];
1102 int refc = h->ref_cache[list][scan8[0] - 8 + 4];
1104 refc = h->ref_cache[list][scan8[0] - 8 - 1];
1106 if(ref[list] < 0 || (refb < ref[list] && refb >= 0))
1108 if(ref[list] < 0 || (refc < ref[list] && refc >= 0))
1114 if(ref[0] < 0 && ref[1] < 0){
1115 ref[0] = ref[1] = 0;
1116 mv[0][0] = mv[0][1] =
1117 mv[1][0] = mv[1][1] = 0;
1119 for(list=0; list<2; list++){
1121 pred_motion(h, 0, 4, list, ref[list], &mv[list][0], &mv[list][1]);
1123 mv[list][0] = mv[list][1] = 0;
1128 *mb_type &= ~MB_TYPE_P0L1;
1129 sub_mb_type &= ~MB_TYPE_P0L1;
1130 }else if(ref[0] < 0){
1131 *mb_type &= ~MB_TYPE_P0L0;
1132 sub_mb_type &= ~MB_TYPE_P0L0;
1135 if(IS_16X16(*mb_type)){
1136 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref[0], 1);
1137 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, ref[1], 1);
1138 if(!IS_INTRA(mb_type_col) && l1ref0[0] == 0 &&
1139 ABS(l1mv0[0][0]) <= 1 && ABS(l1mv0[0][1]) <= 1){
1141 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1143 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
1145 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1147 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
1149 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1150 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1153 for(i8=0; i8<4; i8++){
1154 const int x8 = i8&1;
1155 const int y8 = i8>>1;
1157 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1159 h->sub_mb_type[i8] = sub_mb_type;
1161 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1162 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1163 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref[0], 1);
1164 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, ref[1], 1);
1167 if(!IS_INTRA(mb_type_col) && l1ref0[x8 + y8*h->b8_stride] == 0){
1168 for(i4=0; i4<4; i4++){
1169 const int16_t *mv_col = l1mv0[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1170 if(ABS(mv_col[0]) <= 1 && ABS(mv_col[1]) <= 1){
1172 *(uint32_t*)h->mv_cache[0][scan8[i8*4+i4]] = 0;
1174 *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = 0;
1180 }else{ /* direct temporal mv pred */
1181 /* FIXME assumes that L1ref0 used the same ref lists as current frame */
1182 if(IS_16X16(*mb_type)){
1183 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1);
1184 if(IS_INTRA(mb_type_col)){
1185 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
1186 fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
1187 fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
1189 const int ref0 = l1ref0[0];
1190 const int dist_scale_factor = h->dist_scale_factor[ref0];
1191 const int16_t *mv_col = l1mv0[0];
1193 mv_l0[0] = (dist_scale_factor * mv_col[0] + 128) >> 8;
1194 mv_l0[1] = (dist_scale_factor * mv_col[1] + 128) >> 8;
1195 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref0, 1);
1196 fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv_l0[0],mv_l0[1]), 4);
1197 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);
1200 for(i8=0; i8<4; i8++){
1201 const int x8 = i8&1;
1202 const int y8 = i8>>1;
1203 int ref0, dist_scale_factor;
1205 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1207 h->sub_mb_type[i8] = sub_mb_type;
1208 if(IS_INTRA(mb_type_col)){
1209 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1210 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1211 fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1212 fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1216 ref0 = l1ref0[x8 + y8*h->b8_stride];
1217 dist_scale_factor = h->dist_scale_factor[ref0];
1219 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1220 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1221 for(i4=0; i4<4; i4++){
1222 const int16_t *mv_col = l1mv0[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1223 int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]];
1224 mv_l0[0] = (dist_scale_factor * mv_col[0] + 128) >> 8;
1225 mv_l0[1] = (dist_scale_factor * mv_col[1] + 128) >> 8;
1226 *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] =
1227 pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
1234 static inline void write_back_motion(H264Context *h, int mb_type){
1235 MpegEncContext * const s = &h->s;
1236 const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1237 const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1240 for(list=0; list<2; list++){
1242 if(!USES_LIST(mb_type, list)){
1243 if(1){ //FIXME skip or never read if mb_type doesnt use it
1245 *(uint64_t*)s->current_picture.motion_val[list][b_xy + 0 + y*h->b_stride]=
1246 *(uint64_t*)s->current_picture.motion_val[list][b_xy + 2 + y*h->b_stride]= 0;
1248 if( h->pps.cabac ) {
1249 /* FIXME needed ? */
1251 *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]=
1252 *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= 0;
1256 *(uint16_t*)&s->current_picture.ref_index[list][b8_xy + y*h->b8_stride]= (LIST_NOT_USED&0xFF)*0x0101;
1263 *(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];
1264 *(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];
1266 if( h->pps.cabac ) {
1268 *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+0 + 8*y];
1269 *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+2 + 8*y];
1273 s->current_picture.ref_index[list][b8_xy + 0 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+0 + 16*y];
1274 s->current_picture.ref_index[list][b8_xy + 1 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+2 + 16*y];
1278 if(h->slice_type == B_TYPE && h->pps.cabac){
1279 if(IS_8X8(mb_type)){
1280 h->direct_table[b8_xy+1+0*h->b8_stride] = IS_DIRECT(h->sub_mb_type[1]) ? 1 : 0;
1281 h->direct_table[b8_xy+0+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[2]) ? 1 : 0;
1282 h->direct_table[b8_xy+1+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[3]) ? 1 : 0;
1288 * Decodes a network abstraction layer unit.
1289 * @param consumed is the number of bytes used as input
1290 * @param length is the length of the array
1291 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp ttailing?
1292 * @returns decoded bytes, might be src+1 if no escapes
1294 static uint8_t *decode_nal(H264Context *h, uint8_t *src, int *dst_length, int *consumed, int length){
1298 // src[0]&0x80; //forbidden bit
1299 h->nal_ref_idc= src[0]>>5;
1300 h->nal_unit_type= src[0]&0x1F;
1304 for(i=0; i<length; i++)
1305 printf("%2X ", src[i]);
1307 for(i=0; i+1<length; i+=2){
1308 if(src[i]) continue;
1309 if(i>0 && src[i-1]==0) i--;
1310 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1312 /* startcode, so we must be past the end */
1319 if(i>=length-1){ //no escaped 0
1320 *dst_length= length;
1321 *consumed= length+1; //+1 for the header
1325 h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length);
1326 dst= h->rbsp_buffer;
1328 //printf("deoding esc\n");
1331 //remove escapes (very rare 1:2^22)
1332 if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1333 if(src[si+2]==3){ //escape
1338 }else //next start code
1342 dst[di++]= src[si++];
1346 *consumed= si + 1;//+1 for the header
1347 //FIXME store exact number of bits in the getbitcontext (its needed for decoding)
1353 * @param src the data which should be escaped
1354 * @param dst the target buffer, dst+1 == src is allowed as a special case
1355 * @param length the length of the src data
1356 * @param dst_length the length of the dst array
1357 * @returns length of escaped data in bytes or -1 if an error occured
1359 static int encode_nal(H264Context *h, uint8_t *dst, uint8_t *src, int length, int dst_length){
1360 int i, escape_count, si, di;
1364 assert(dst_length>0);
1366 dst[0]= (h->nal_ref_idc<<5) + h->nal_unit_type;
1368 if(length==0) return 1;
1371 for(i=0; i<length; i+=2){
1372 if(src[i]) continue;
1373 if(i>0 && src[i-1]==0)
1375 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1381 if(escape_count==0){
1383 memcpy(dst+1, src, length);
1387 if(length + escape_count + 1> dst_length)
1390 //this should be damn rare (hopefully)
1392 h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length + escape_count);
1393 temp= h->rbsp_buffer;
1394 //printf("encoding esc\n");
1399 if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1400 temp[di++]= 0; si++;
1401 temp[di++]= 0; si++;
1403 temp[di++]= src[si++];
1406 temp[di++]= src[si++];
1408 memcpy(dst+1, temp, length+escape_count);
1410 assert(di == length+escape_count);
1416 * write 1,10,100,1000,... for alignment, yes its exactly inverse to mpeg4
1418 static void encode_rbsp_trailing(PutBitContext *pb){
1421 length= (-put_bits_count(pb))&7;
1422 if(length) put_bits(pb, length, 0);
1427 * identifies the exact end of the bitstream
1428 * @return the length of the trailing, or 0 if damaged
1430 static int decode_rbsp_trailing(uint8_t *src){
1434 tprintf("rbsp trailing %X\n", v);
1444 * idct tranforms the 16 dc values and dequantize them.
1445 * @param qp quantization parameter
1447 static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp){
1448 const int qmul= dequant_coeff[qp][0];
1451 int temp[16]; //FIXME check if this is a good idea
1452 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
1453 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1455 //memset(block, 64, 2*256);
1458 const int offset= y_offset[i];
1459 const int z0= block[offset+stride*0] + block[offset+stride*4];
1460 const int z1= block[offset+stride*0] - block[offset+stride*4];
1461 const int z2= block[offset+stride*1] - block[offset+stride*5];
1462 const int z3= block[offset+stride*1] + block[offset+stride*5];
1471 const int offset= x_offset[i];
1472 const int z0= temp[4*0+i] + temp[4*2+i];
1473 const int z1= temp[4*0+i] - temp[4*2+i];
1474 const int z2= temp[4*1+i] - temp[4*3+i];
1475 const int z3= temp[4*1+i] + temp[4*3+i];
1477 block[stride*0 +offset]= ((z0 + z3)*qmul + 2)>>2; //FIXME think about merging this into decode_resdual
1478 block[stride*2 +offset]= ((z1 + z2)*qmul + 2)>>2;
1479 block[stride*8 +offset]= ((z1 - z2)*qmul + 2)>>2;
1480 block[stride*10+offset]= ((z0 - z3)*qmul + 2)>>2;
1486 * dct tranforms the 16 dc values.
1487 * @param qp quantization parameter ??? FIXME
1489 static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
1490 // const int qmul= dequant_coeff[qp][0];
1492 int temp[16]; //FIXME check if this is a good idea
1493 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
1494 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1497 const int offset= y_offset[i];
1498 const int z0= block[offset+stride*0] + block[offset+stride*4];
1499 const int z1= block[offset+stride*0] - block[offset+stride*4];
1500 const int z2= block[offset+stride*1] - block[offset+stride*5];
1501 const int z3= block[offset+stride*1] + block[offset+stride*5];
1510 const int offset= x_offset[i];
1511 const int z0= temp[4*0+i] + temp[4*2+i];
1512 const int z1= temp[4*0+i] - temp[4*2+i];
1513 const int z2= temp[4*1+i] - temp[4*3+i];
1514 const int z3= temp[4*1+i] + temp[4*3+i];
1516 block[stride*0 +offset]= (z0 + z3)>>1;
1517 block[stride*2 +offset]= (z1 + z2)>>1;
1518 block[stride*8 +offset]= (z1 - z2)>>1;
1519 block[stride*10+offset]= (z0 - z3)>>1;
1527 static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp){
1528 const int qmul= dequant_coeff[qp][0];
1529 const int stride= 16*2;
1530 const int xStride= 16;
1533 a= block[stride*0 + xStride*0];
1534 b= block[stride*0 + xStride*1];
1535 c= block[stride*1 + xStride*0];
1536 d= block[stride*1 + xStride*1];
1543 block[stride*0 + xStride*0]= ((a+c)*qmul + 0)>>1;
1544 block[stride*0 + xStride*1]= ((e+b)*qmul + 0)>>1;
1545 block[stride*1 + xStride*0]= ((a-c)*qmul + 0)>>1;
1546 block[stride*1 + xStride*1]= ((e-b)*qmul + 0)>>1;
1550 static void chroma_dc_dct_c(DCTELEM *block){
1551 const int stride= 16*2;
1552 const int xStride= 16;
1555 a= block[stride*0 + xStride*0];
1556 b= block[stride*0 + xStride*1];
1557 c= block[stride*1 + xStride*0];
1558 d= block[stride*1 + xStride*1];
1565 block[stride*0 + xStride*0]= (a+c);
1566 block[stride*0 + xStride*1]= (e+b);
1567 block[stride*1 + xStride*0]= (a-c);
1568 block[stride*1 + xStride*1]= (e-b);
1573 * gets the chroma qp.
1575 static inline int get_chroma_qp(H264Context *h, int qscale){
1577 return chroma_qp[clip(qscale + h->pps.chroma_qp_index_offset, 0, 51)];
1582 static void h264_diff_dct_c(DCTELEM *block, uint8_t *src1, uint8_t *src2, int stride){
1584 //FIXME try int temp instead of block
1587 const int d0= src1[0 + i*stride] - src2[0 + i*stride];
1588 const int d1= src1[1 + i*stride] - src2[1 + i*stride];
1589 const int d2= src1[2 + i*stride] - src2[2 + i*stride];
1590 const int d3= src1[3 + i*stride] - src2[3 + i*stride];
1591 const int z0= d0 + d3;
1592 const int z3= d0 - d3;
1593 const int z1= d1 + d2;
1594 const int z2= d1 - d2;
1596 block[0 + 4*i]= z0 + z1;
1597 block[1 + 4*i]= 2*z3 + z2;
1598 block[2 + 4*i]= z0 - z1;
1599 block[3 + 4*i]= z3 - 2*z2;
1603 const int z0= block[0*4 + i] + block[3*4 + i];
1604 const int z3= block[0*4 + i] - block[3*4 + i];
1605 const int z1= block[1*4 + i] + block[2*4 + i];
1606 const int z2= block[1*4 + i] - block[2*4 + i];
1608 block[0*4 + i]= z0 + z1;
1609 block[1*4 + i]= 2*z3 + z2;
1610 block[2*4 + i]= z0 - z1;
1611 block[3*4 + i]= z3 - 2*z2;
1616 //FIXME need to check that this doesnt overflow signed 32 bit for low qp, iam not sure, its very close
1617 //FIXME check that gcc inlines this (and optimizes intra & seperate_dc stuff away)
1618 static inline int quantize_c(DCTELEM *block, uint8_t *scantable, int qscale, int intra, int seperate_dc){
1620 const int * const quant_table= quant_coeff[qscale];
1621 const int bias= intra ? (1<<QUANT_SHIFT)/3 : (1<<QUANT_SHIFT)/6;
1622 const unsigned int threshold1= (1<<QUANT_SHIFT) - bias - 1;
1623 const unsigned int threshold2= (threshold1<<1);
1629 const int dc_bias= intra ? (1<<(QUANT_SHIFT-2))/3 : (1<<(QUANT_SHIFT-2))/6;
1630 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT-2)) - dc_bias - 1;
1631 const unsigned int dc_threshold2= (dc_threshold1<<1);
1633 int level= block[0]*quant_coeff[qscale+18][0];
1634 if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1636 level= (dc_bias + level)>>(QUANT_SHIFT-2);
1639 level= (dc_bias - level)>>(QUANT_SHIFT-2);
1642 // last_non_zero = i;
1647 const int dc_bias= intra ? (1<<(QUANT_SHIFT+1))/3 : (1<<(QUANT_SHIFT+1))/6;
1648 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT+1)) - dc_bias - 1;
1649 const unsigned int dc_threshold2= (dc_threshold1<<1);
1651 int level= block[0]*quant_table[0];
1652 if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1654 level= (dc_bias + level)>>(QUANT_SHIFT+1);
1657 level= (dc_bias - level)>>(QUANT_SHIFT+1);
1660 // last_non_zero = i;
1673 const int j= scantable[i];
1674 int level= block[j]*quant_table[j];
1676 // if( bias+level >= (1<<(QMAT_SHIFT - 3))
1677 // || bias-level >= (1<<(QMAT_SHIFT - 3))){
1678 if(((unsigned)(level+threshold1))>threshold2){
1680 level= (bias + level)>>QUANT_SHIFT;
1683 level= (bias - level)>>QUANT_SHIFT;
1692 return last_non_zero;
1695 static void pred4x4_vertical_c(uint8_t *src, uint8_t *topright, int stride){
1696 const uint32_t a= ((uint32_t*)(src-stride))[0];
1697 ((uint32_t*)(src+0*stride))[0]= a;
1698 ((uint32_t*)(src+1*stride))[0]= a;
1699 ((uint32_t*)(src+2*stride))[0]= a;
1700 ((uint32_t*)(src+3*stride))[0]= a;
1703 static void pred4x4_horizontal_c(uint8_t *src, uint8_t *topright, int stride){
1704 ((uint32_t*)(src+0*stride))[0]= src[-1+0*stride]*0x01010101;
1705 ((uint32_t*)(src+1*stride))[0]= src[-1+1*stride]*0x01010101;
1706 ((uint32_t*)(src+2*stride))[0]= src[-1+2*stride]*0x01010101;
1707 ((uint32_t*)(src+3*stride))[0]= src[-1+3*stride]*0x01010101;
1710 static void pred4x4_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]
1712 + src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 4) >>3;
1714 ((uint32_t*)(src+0*stride))[0]=
1715 ((uint32_t*)(src+1*stride))[0]=
1716 ((uint32_t*)(src+2*stride))[0]=
1717 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1720 static void pred4x4_left_dc_c(uint8_t *src, uint8_t *topright, int stride){
1721 const int dc= ( src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 2) >>2;
1723 ((uint32_t*)(src+0*stride))[0]=
1724 ((uint32_t*)(src+1*stride))[0]=
1725 ((uint32_t*)(src+2*stride))[0]=
1726 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1729 static void pred4x4_top_dc_c(uint8_t *src, uint8_t *topright, int stride){
1730 const int dc= ( src[-stride] + src[1-stride] + src[2-stride] + src[3-stride] + 2) >>2;
1732 ((uint32_t*)(src+0*stride))[0]=
1733 ((uint32_t*)(src+1*stride))[0]=
1734 ((uint32_t*)(src+2*stride))[0]=
1735 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1738 static void pred4x4_128_dc_c(uint8_t *src, uint8_t *topright, int stride){
1739 ((uint32_t*)(src+0*stride))[0]=
1740 ((uint32_t*)(src+1*stride))[0]=
1741 ((uint32_t*)(src+2*stride))[0]=
1742 ((uint32_t*)(src+3*stride))[0]= 128U*0x01010101U;
1746 #define LOAD_TOP_RIGHT_EDGE\
1747 const int t4= topright[0];\
1748 const int t5= topright[1];\
1749 const int t6= topright[2];\
1750 const int t7= topright[3];\
1752 #define LOAD_LEFT_EDGE\
1753 const int l0= src[-1+0*stride];\
1754 const int l1= src[-1+1*stride];\
1755 const int l2= src[-1+2*stride];\
1756 const int l3= src[-1+3*stride];\
1758 #define LOAD_TOP_EDGE\
1759 const int t0= src[ 0-1*stride];\
1760 const int t1= src[ 1-1*stride];\
1761 const int t2= src[ 2-1*stride];\
1762 const int t3= src[ 3-1*stride];\
1764 static void pred4x4_down_right_c(uint8_t *src, uint8_t *topright, int stride){
1765 const int lt= src[-1-1*stride];
1769 src[0+3*stride]=(l3 + 2*l2 + l1 + 2)>>2;
1771 src[1+3*stride]=(l2 + 2*l1 + l0 + 2)>>2;
1774 src[2+3*stride]=(l1 + 2*l0 + lt + 2)>>2;
1778 src[3+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1781 src[3+2*stride]=(lt + 2*t0 + t1 + 2)>>2;
1783 src[3+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1784 src[3+0*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1787 static void pred4x4_down_left_c(uint8_t *src, uint8_t *topright, int stride){
1792 src[0+0*stride]=(t0 + t2 + 2*t1 + 2)>>2;
1794 src[0+1*stride]=(t1 + t3 + 2*t2 + 2)>>2;
1797 src[0+2*stride]=(t2 + t4 + 2*t3 + 2)>>2;
1801 src[0+3*stride]=(t3 + t5 + 2*t4 + 2)>>2;
1804 src[1+3*stride]=(t4 + t6 + 2*t5 + 2)>>2;
1806 src[2+3*stride]=(t5 + t7 + 2*t6 + 2)>>2;
1807 src[3+3*stride]=(t6 + 3*t7 + 2)>>2;
1810 static void pred4x4_vertical_right_c(uint8_t *src, uint8_t *topright, int stride){
1811 const int lt= src[-1-1*stride];
1814 const __attribute__((unused)) int unu= l3;
1817 src[1+2*stride]=(lt + t0 + 1)>>1;
1819 src[2+2*stride]=(t0 + t1 + 1)>>1;
1821 src[3+2*stride]=(t1 + t2 + 1)>>1;
1822 src[3+0*stride]=(t2 + t3 + 1)>>1;
1824 src[1+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1826 src[2+3*stride]=(lt + 2*t0 + t1 + 2)>>2;
1828 src[3+3*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1829 src[3+1*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1830 src[0+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1831 src[0+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1834 static void pred4x4_vertical_left_c(uint8_t *src, uint8_t *topright, int stride){
1837 const __attribute__((unused)) int unu= t7;
1839 src[0+0*stride]=(t0 + t1 + 1)>>1;
1841 src[0+2*stride]=(t1 + t2 + 1)>>1;
1843 src[1+2*stride]=(t2 + t3 + 1)>>1;
1845 src[2+2*stride]=(t3 + t4+ 1)>>1;
1846 src[3+2*stride]=(t4 + t5+ 1)>>1;
1847 src[0+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1849 src[0+3*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1851 src[1+3*stride]=(t2 + 2*t3 + t4 + 2)>>2;
1853 src[2+3*stride]=(t3 + 2*t4 + t5 + 2)>>2;
1854 src[3+3*stride]=(t4 + 2*t5 + t6 + 2)>>2;
1857 static void pred4x4_horizontal_up_c(uint8_t *src, uint8_t *topright, int stride){
1860 src[0+0*stride]=(l0 + l1 + 1)>>1;
1861 src[1+0*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1863 src[0+1*stride]=(l1 + l2 + 1)>>1;
1865 src[1+1*stride]=(l1 + 2*l2 + l3 + 2)>>2;
1867 src[0+2*stride]=(l2 + l3 + 1)>>1;
1869 src[1+2*stride]=(l2 + 2*l3 + l3 + 2)>>2;
1878 static void pred4x4_horizontal_down_c(uint8_t *src, uint8_t *topright, int stride){
1879 const int lt= src[-1-1*stride];
1882 const __attribute__((unused)) int unu= t3;
1885 src[2+1*stride]=(lt + l0 + 1)>>1;
1887 src[3+1*stride]=(l0 + 2*lt + t0 + 2)>>2;
1888 src[2+0*stride]=(lt + 2*t0 + t1 + 2)>>2;
1889 src[3+0*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1891 src[2+2*stride]=(l0 + l1 + 1)>>1;
1893 src[3+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1895 src[2+3*stride]=(l1 + l2+ 1)>>1;
1897 src[3+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1898 src[0+3*stride]=(l2 + l3 + 1)>>1;
1899 src[1+3*stride]=(l1 + 2*l2 + l3 + 2)>>2;
1902 static void pred16x16_vertical_c(uint8_t *src, int stride){
1904 const uint32_t a= ((uint32_t*)(src-stride))[0];
1905 const uint32_t b= ((uint32_t*)(src-stride))[1];
1906 const uint32_t c= ((uint32_t*)(src-stride))[2];
1907 const uint32_t d= ((uint32_t*)(src-stride))[3];
1909 for(i=0; i<16; i++){
1910 ((uint32_t*)(src+i*stride))[0]= a;
1911 ((uint32_t*)(src+i*stride))[1]= b;
1912 ((uint32_t*)(src+i*stride))[2]= c;
1913 ((uint32_t*)(src+i*stride))[3]= d;
1917 static void pred16x16_horizontal_c(uint8_t *src, int stride){
1920 for(i=0; i<16; i++){
1921 ((uint32_t*)(src+i*stride))[0]=
1922 ((uint32_t*)(src+i*stride))[1]=
1923 ((uint32_t*)(src+i*stride))[2]=
1924 ((uint32_t*)(src+i*stride))[3]= src[-1+i*stride]*0x01010101;
1928 static void pred16x16_dc_c(uint8_t *src, int stride){
1932 dc+= src[-1+i*stride];
1939 dc= 0x01010101*((dc + 16)>>5);
1941 for(i=0; i<16; i++){
1942 ((uint32_t*)(src+i*stride))[0]=
1943 ((uint32_t*)(src+i*stride))[1]=
1944 ((uint32_t*)(src+i*stride))[2]=
1945 ((uint32_t*)(src+i*stride))[3]= dc;
1949 static void pred16x16_left_dc_c(uint8_t *src, int stride){
1953 dc+= src[-1+i*stride];
1956 dc= 0x01010101*((dc + 8)>>4);
1958 for(i=0; i<16; i++){
1959 ((uint32_t*)(src+i*stride))[0]=
1960 ((uint32_t*)(src+i*stride))[1]=
1961 ((uint32_t*)(src+i*stride))[2]=
1962 ((uint32_t*)(src+i*stride))[3]= dc;
1966 static void pred16x16_top_dc_c(uint8_t *src, int stride){
1972 dc= 0x01010101*((dc + 8)>>4);
1974 for(i=0; i<16; i++){
1975 ((uint32_t*)(src+i*stride))[0]=
1976 ((uint32_t*)(src+i*stride))[1]=
1977 ((uint32_t*)(src+i*stride))[2]=
1978 ((uint32_t*)(src+i*stride))[3]= dc;
1982 static void pred16x16_128_dc_c(uint8_t *src, int stride){
1985 for(i=0; i<16; i++){
1986 ((uint32_t*)(src+i*stride))[0]=
1987 ((uint32_t*)(src+i*stride))[1]=
1988 ((uint32_t*)(src+i*stride))[2]=
1989 ((uint32_t*)(src+i*stride))[3]= 0x01010101U*128U;
1993 static inline void pred16x16_plane_compat_c(uint8_t *src, int stride, const int svq3){
1996 uint8_t *cm = cropTbl + MAX_NEG_CROP;
1997 const uint8_t * const src0 = src+7-stride;
1998 const uint8_t *src1 = src+8*stride-1;
1999 const uint8_t *src2 = src1-2*stride; // == src+6*stride-1;
2000 int H = src0[1] - src0[-1];
2001 int V = src1[0] - src2[ 0];
2002 for(k=2; k<=8; ++k) {
2003 src1 += stride; src2 -= stride;
2004 H += k*(src0[k] - src0[-k]);
2005 V += k*(src1[0] - src2[ 0]);
2008 H = ( 5*(H/4) ) / 16;
2009 V = ( 5*(V/4) ) / 16;
2011 /* required for 100% accuracy */
2012 i = H; H = V; V = i;
2014 H = ( 5*H+32 ) >> 6;
2015 V = ( 5*V+32 ) >> 6;
2018 a = 16*(src1[0] + src2[16] + 1) - 7*(V+H);
2019 for(j=16; j>0; --j) {
2022 for(i=-16; i<0; i+=4) {
2023 src[16+i] = cm[ (b ) >> 5 ];
2024 src[17+i] = cm[ (b+ H) >> 5 ];
2025 src[18+i] = cm[ (b+2*H) >> 5 ];
2026 src[19+i] = cm[ (b+3*H) >> 5 ];
2033 static void pred16x16_plane_c(uint8_t *src, int stride){
2034 pred16x16_plane_compat_c(src, stride, 0);
2037 static void pred8x8_vertical_c(uint8_t *src, int stride){
2039 const uint32_t a= ((uint32_t*)(src-stride))[0];
2040 const uint32_t b= ((uint32_t*)(src-stride))[1];
2043 ((uint32_t*)(src+i*stride))[0]= a;
2044 ((uint32_t*)(src+i*stride))[1]= b;
2048 static void pred8x8_horizontal_c(uint8_t *src, int stride){
2052 ((uint32_t*)(src+i*stride))[0]=
2053 ((uint32_t*)(src+i*stride))[1]= src[-1+i*stride]*0x01010101;
2057 static void pred8x8_128_dc_c(uint8_t *src, int stride){
2061 ((uint32_t*)(src+i*stride))[0]=
2062 ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
2065 ((uint32_t*)(src+i*stride))[0]=
2066 ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
2070 static void pred8x8_left_dc_c(uint8_t *src, int stride){
2076 dc0+= src[-1+i*stride];
2077 dc2+= src[-1+(i+4)*stride];
2079 dc0= 0x01010101*((dc0 + 2)>>2);
2080 dc2= 0x01010101*((dc2 + 2)>>2);
2083 ((uint32_t*)(src+i*stride))[0]=
2084 ((uint32_t*)(src+i*stride))[1]= dc0;
2087 ((uint32_t*)(src+i*stride))[0]=
2088 ((uint32_t*)(src+i*stride))[1]= dc2;
2092 static void pred8x8_top_dc_c(uint8_t *src, int stride){
2098 dc0+= src[i-stride];
2099 dc1+= src[4+i-stride];
2101 dc0= 0x01010101*((dc0 + 2)>>2);
2102 dc1= 0x01010101*((dc1 + 2)>>2);
2105 ((uint32_t*)(src+i*stride))[0]= dc0;
2106 ((uint32_t*)(src+i*stride))[1]= dc1;
2109 ((uint32_t*)(src+i*stride))[0]= dc0;
2110 ((uint32_t*)(src+i*stride))[1]= dc1;
2115 static void pred8x8_dc_c(uint8_t *src, int stride){
2117 int dc0, dc1, dc2, dc3;
2121 dc0+= src[-1+i*stride] + src[i-stride];
2122 dc1+= src[4+i-stride];
2123 dc2+= src[-1+(i+4)*stride];
2125 dc3= 0x01010101*((dc1 + dc2 + 4)>>3);
2126 dc0= 0x01010101*((dc0 + 4)>>3);
2127 dc1= 0x01010101*((dc1 + 2)>>2);
2128 dc2= 0x01010101*((dc2 + 2)>>2);
2131 ((uint32_t*)(src+i*stride))[0]= dc0;
2132 ((uint32_t*)(src+i*stride))[1]= dc1;
2135 ((uint32_t*)(src+i*stride))[0]= dc2;
2136 ((uint32_t*)(src+i*stride))[1]= dc3;
2140 static void pred8x8_plane_c(uint8_t *src, int stride){
2143 uint8_t *cm = cropTbl + MAX_NEG_CROP;
2144 const uint8_t * const src0 = src+3-stride;
2145 const uint8_t *src1 = src+4*stride-1;
2146 const uint8_t *src2 = src1-2*stride; // == src+2*stride-1;
2147 int H = src0[1] - src0[-1];
2148 int V = src1[0] - src2[ 0];
2149 for(k=2; k<=4; ++k) {
2150 src1 += stride; src2 -= stride;
2151 H += k*(src0[k] - src0[-k]);
2152 V += k*(src1[0] - src2[ 0]);
2154 H = ( 17*H+16 ) >> 5;
2155 V = ( 17*V+16 ) >> 5;
2157 a = 16*(src1[0] + src2[8]+1) - 3*(V+H);
2158 for(j=8; j>0; --j) {
2161 src[0] = cm[ (b ) >> 5 ];
2162 src[1] = cm[ (b+ H) >> 5 ];
2163 src[2] = cm[ (b+2*H) >> 5 ];
2164 src[3] = cm[ (b+3*H) >> 5 ];
2165 src[4] = cm[ (b+4*H) >> 5 ];
2166 src[5] = cm[ (b+5*H) >> 5 ];
2167 src[6] = cm[ (b+6*H) >> 5 ];
2168 src[7] = cm[ (b+7*H) >> 5 ];
2173 static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
2174 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2175 int src_x_offset, int src_y_offset,
2176 qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
2177 MpegEncContext * const s = &h->s;
2178 const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
2179 const int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
2180 const int luma_xy= (mx&3) + ((my&3)<<2);
2181 uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*s->linesize;
2182 uint8_t * src_cb= pic->data[1] + (mx>>3) + (my>>3)*s->uvlinesize;
2183 uint8_t * src_cr= pic->data[2] + (mx>>3) + (my>>3)*s->uvlinesize;
2184 int extra_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16; //FIXME increase edge?, IMHO not worth it
2185 int extra_height= extra_width;
2187 const int full_mx= mx>>2;
2188 const int full_my= my>>2;
2190 assert(pic->data[0]);
2192 if(mx&7) extra_width -= 3;
2193 if(my&7) extra_height -= 3;
2195 if( full_mx < 0-extra_width
2196 || full_my < 0-extra_height
2197 || full_mx + 16/*FIXME*/ > s->width + extra_width
2198 || full_my + 16/*FIXME*/ > s->height + extra_height){
2199 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);
2200 src_y= s->edge_emu_buffer + 2 + 2*s->linesize;
2204 qpix_op[luma_xy](dest_y, src_y, s->linesize); //FIXME try variable height perhaps?
2206 qpix_op[luma_xy](dest_y + delta, src_y + delta, s->linesize);
2209 if(s->flags&CODEC_FLAG_GRAY) return;
2212 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);
2213 src_cb= s->edge_emu_buffer;
2215 chroma_op(dest_cb, src_cb, s->uvlinesize, chroma_height, mx&7, my&7);
2218 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);
2219 src_cr= s->edge_emu_buffer;
2221 chroma_op(dest_cr, src_cr, s->uvlinesize, chroma_height, mx&7, my&7);
2224 static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
2225 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2226 int x_offset, int y_offset,
2227 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2228 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
2229 int list0, int list1){
2230 MpegEncContext * const s = &h->s;
2231 qpel_mc_func *qpix_op= qpix_put;
2232 h264_chroma_mc_func chroma_op= chroma_put;
2234 dest_y += 2*x_offset + 2*y_offset*s-> linesize;
2235 dest_cb += x_offset + y_offset*s->uvlinesize;
2236 dest_cr += x_offset + y_offset*s->uvlinesize;
2237 x_offset += 8*s->mb_x;
2238 y_offset += 8*s->mb_y;
2241 Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
2242 mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
2243 dest_y, dest_cb, dest_cr, x_offset, y_offset,
2244 qpix_op, chroma_op);
2247 chroma_op= chroma_avg;
2251 Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
2252 mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
2253 dest_y, dest_cb, dest_cr, x_offset, y_offset,
2254 qpix_op, chroma_op);
2258 static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
2259 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2260 int x_offset, int y_offset,
2261 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2262 h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
2263 h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
2264 int list0, int list1){
2265 MpegEncContext * const s = &h->s;
2267 dest_y += 2*x_offset + 2*y_offset*s-> linesize;
2268 dest_cb += x_offset + y_offset*s->uvlinesize;
2269 dest_cr += x_offset + y_offset*s->uvlinesize;
2270 x_offset += 8*s->mb_x;
2271 y_offset += 8*s->mb_y;
2274 /* don't optimize for luma-only case, since B-frames usually
2275 * use implicit weights => chroma too. */
2276 uint8_t *tmp_cb = s->obmc_scratchpad;
2277 uint8_t *tmp_cr = tmp_cb + 8*s->uvlinesize;
2278 uint8_t *tmp_y = tmp_cr + 8*s->uvlinesize;
2279 int refn0 = h->ref_cache[0][ scan8[n] ];
2280 int refn1 = h->ref_cache[1][ scan8[n] ];
2282 mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
2283 dest_y, dest_cb, dest_cr,
2284 x_offset, y_offset, qpix_put, chroma_put);
2285 mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
2286 tmp_y, tmp_cb, tmp_cr,
2287 x_offset, y_offset, qpix_put, chroma_put);
2289 if(h->use_weight == 2){
2290 int weight0 = h->implicit_weight[refn0][refn1];
2291 int weight1 = 64 - weight0;
2292 luma_weight_avg( dest_y, tmp_y, s-> linesize, 5, weight0, weight1, 0, 0);
2293 chroma_weight_avg(dest_cb, tmp_cb, s->uvlinesize, 5, weight0, weight1, 0, 0);
2294 chroma_weight_avg(dest_cr, tmp_cr, s->uvlinesize, 5, weight0, weight1, 0, 0);
2296 luma_weight_avg(dest_y, tmp_y, s->linesize, h->luma_log2_weight_denom,
2297 h->luma_weight[0][refn0], h->luma_weight[1][refn1],
2298 h->luma_offset[0][refn0], h->luma_offset[1][refn1]);
2299 chroma_weight_avg(dest_cb, tmp_cb, s->uvlinesize, h->chroma_log2_weight_denom,
2300 h->chroma_weight[0][refn0][0], h->chroma_weight[1][refn1][0],
2301 h->chroma_offset[0][refn0][0], h->chroma_offset[1][refn1][0]);
2302 chroma_weight_avg(dest_cr, tmp_cr, s->uvlinesize, h->chroma_log2_weight_denom,
2303 h->chroma_weight[0][refn0][1], h->chroma_weight[1][refn1][1],
2304 h->chroma_offset[0][refn0][1], h->chroma_offset[1][refn1][1]);
2307 int list = list1 ? 1 : 0;
2308 int refn = h->ref_cache[list][ scan8[n] ];
2309 Picture *ref= &h->ref_list[list][refn];
2310 mc_dir_part(h, ref, n, square, chroma_height, delta, list,
2311 dest_y, dest_cb, dest_cr, x_offset, y_offset,
2312 qpix_put, chroma_put);
2314 luma_weight_op(dest_y, s->linesize, h->luma_log2_weight_denom,
2315 h->luma_weight[list][refn], h->luma_offset[list][refn]);
2316 if(h->use_weight_chroma){
2317 chroma_weight_op(dest_cb, s->uvlinesize, h->chroma_log2_weight_denom,
2318 h->chroma_weight[list][refn][0], h->chroma_offset[list][refn][0]);
2319 chroma_weight_op(dest_cr, s->uvlinesize, h->chroma_log2_weight_denom,
2320 h->chroma_weight[list][refn][1], h->chroma_offset[list][refn][1]);
2325 static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
2326 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2327 int x_offset, int y_offset,
2328 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2329 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
2330 h264_weight_func *weight_op, h264_biweight_func *weight_avg,
2331 int list0, int list1){
2332 if((h->use_weight==2 && list0 && list1
2333 && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ] != 32))
2334 || h->use_weight==1)
2335 mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
2336 x_offset, y_offset, qpix_put, chroma_put,
2337 weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
2339 mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
2340 x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
2343 static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2344 qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
2345 qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
2346 h264_weight_func *weight_op, h264_biweight_func *weight_avg){
2347 MpegEncContext * const s = &h->s;
2348 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
2349 const int mb_type= s->current_picture.mb_type[mb_xy];
2351 assert(IS_INTER(mb_type));
2353 if(IS_16X16(mb_type)){
2354 mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
2355 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
2356 &weight_op[0], &weight_avg[0],
2357 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2358 }else if(IS_16X8(mb_type)){
2359 mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
2360 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
2361 &weight_op[1], &weight_avg[1],
2362 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2363 mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
2364 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
2365 &weight_op[1], &weight_avg[1],
2366 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2367 }else if(IS_8X16(mb_type)){
2368 mc_part(h, 0, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 0, 0,
2369 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2370 &weight_op[2], &weight_avg[2],
2371 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2372 mc_part(h, 4, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 4, 0,
2373 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2374 &weight_op[2], &weight_avg[2],
2375 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2379 assert(IS_8X8(mb_type));
2382 const int sub_mb_type= h->sub_mb_type[i];
2384 int x_offset= (i&1)<<2;
2385 int y_offset= (i&2)<<1;
2387 if(IS_SUB_8X8(sub_mb_type)){
2388 mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2389 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2390 &weight_op[3], &weight_avg[3],
2391 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2392 }else if(IS_SUB_8X4(sub_mb_type)){
2393 mc_part(h, n , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2394 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2395 &weight_op[4], &weight_avg[4],
2396 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2397 mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
2398 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2399 &weight_op[4], &weight_avg[4],
2400 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2401 }else if(IS_SUB_4X8(sub_mb_type)){
2402 mc_part(h, n , 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2403 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2404 &weight_op[5], &weight_avg[5],
2405 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2406 mc_part(h, n+1, 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
2407 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2408 &weight_op[5], &weight_avg[5],
2409 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2412 assert(IS_SUB_4X4(sub_mb_type));
2414 int sub_x_offset= x_offset + 2*(j&1);
2415 int sub_y_offset= y_offset + (j&2);
2416 mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
2417 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2418 &weight_op[6], &weight_avg[6],
2419 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2426 static void decode_init_vlc(H264Context *h){
2427 static int done = 0;
2433 init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5,
2434 &chroma_dc_coeff_token_len [0], 1, 1,
2435 &chroma_dc_coeff_token_bits[0], 1, 1, 1);
2438 init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17,
2439 &coeff_token_len [i][0], 1, 1,
2440 &coeff_token_bits[i][0], 1, 1, 1);
2444 init_vlc(&chroma_dc_total_zeros_vlc[i], CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
2445 &chroma_dc_total_zeros_len [i][0], 1, 1,
2446 &chroma_dc_total_zeros_bits[i][0], 1, 1, 1);
2448 for(i=0; i<15; i++){
2449 init_vlc(&total_zeros_vlc[i], TOTAL_ZEROS_VLC_BITS, 16,
2450 &total_zeros_len [i][0], 1, 1,
2451 &total_zeros_bits[i][0], 1, 1, 1);
2455 init_vlc(&run_vlc[i], RUN_VLC_BITS, 7,
2456 &run_len [i][0], 1, 1,
2457 &run_bits[i][0], 1, 1, 1);
2459 init_vlc(&run7_vlc, RUN7_VLC_BITS, 16,
2460 &run_len [6][0], 1, 1,
2461 &run_bits[6][0], 1, 1, 1);
2466 * Sets the intra prediction function pointers.
2468 static void init_pred_ptrs(H264Context *h){
2469 // MpegEncContext * const s = &h->s;
2471 h->pred4x4[VERT_PRED ]= pred4x4_vertical_c;
2472 h->pred4x4[HOR_PRED ]= pred4x4_horizontal_c;
2473 h->pred4x4[DC_PRED ]= pred4x4_dc_c;
2474 h->pred4x4[DIAG_DOWN_LEFT_PRED ]= pred4x4_down_left_c;
2475 h->pred4x4[DIAG_DOWN_RIGHT_PRED]= pred4x4_down_right_c;
2476 h->pred4x4[VERT_RIGHT_PRED ]= pred4x4_vertical_right_c;
2477 h->pred4x4[HOR_DOWN_PRED ]= pred4x4_horizontal_down_c;
2478 h->pred4x4[VERT_LEFT_PRED ]= pred4x4_vertical_left_c;
2479 h->pred4x4[HOR_UP_PRED ]= pred4x4_horizontal_up_c;
2480 h->pred4x4[LEFT_DC_PRED ]= pred4x4_left_dc_c;
2481 h->pred4x4[TOP_DC_PRED ]= pred4x4_top_dc_c;
2482 h->pred4x4[DC_128_PRED ]= pred4x4_128_dc_c;
2484 h->pred8x8[DC_PRED8x8 ]= pred8x8_dc_c;
2485 h->pred8x8[VERT_PRED8x8 ]= pred8x8_vertical_c;
2486 h->pred8x8[HOR_PRED8x8 ]= pred8x8_horizontal_c;
2487 h->pred8x8[PLANE_PRED8x8 ]= pred8x8_plane_c;
2488 h->pred8x8[LEFT_DC_PRED8x8]= pred8x8_left_dc_c;
2489 h->pred8x8[TOP_DC_PRED8x8 ]= pred8x8_top_dc_c;
2490 h->pred8x8[DC_128_PRED8x8 ]= pred8x8_128_dc_c;
2492 h->pred16x16[DC_PRED8x8 ]= pred16x16_dc_c;
2493 h->pred16x16[VERT_PRED8x8 ]= pred16x16_vertical_c;
2494 h->pred16x16[HOR_PRED8x8 ]= pred16x16_horizontal_c;
2495 h->pred16x16[PLANE_PRED8x8 ]= pred16x16_plane_c;
2496 h->pred16x16[LEFT_DC_PRED8x8]= pred16x16_left_dc_c;
2497 h->pred16x16[TOP_DC_PRED8x8 ]= pred16x16_top_dc_c;
2498 h->pred16x16[DC_128_PRED8x8 ]= pred16x16_128_dc_c;
2501 static void free_tables(H264Context *h){
2502 av_freep(&h->intra4x4_pred_mode);
2503 av_freep(&h->chroma_pred_mode_table);
2504 av_freep(&h->cbp_table);
2505 av_freep(&h->mvd_table[0]);
2506 av_freep(&h->mvd_table[1]);
2507 av_freep(&h->direct_table);
2508 av_freep(&h->non_zero_count);
2509 av_freep(&h->slice_table_base);
2510 av_freep(&h->top_border);
2511 h->slice_table= NULL;
2513 av_freep(&h->mb2b_xy);
2514 av_freep(&h->mb2b8_xy);
2516 av_freep(&h->s.obmc_scratchpad);
2521 * needs widzh/height
2523 static int alloc_tables(H264Context *h){
2524 MpegEncContext * const s = &h->s;
2525 const int big_mb_num= s->mb_stride * (s->mb_height+1);
2528 CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8 * sizeof(uint8_t))
2530 CHECKED_ALLOCZ(h->non_zero_count , big_mb_num * 16 * sizeof(uint8_t))
2531 CHECKED_ALLOCZ(h->slice_table_base , big_mb_num * sizeof(uint8_t))
2532 CHECKED_ALLOCZ(h->top_border , s->mb_width * (16+8+8) * sizeof(uint8_t))
2533 CHECKED_ALLOCZ(h->cbp_table, big_mb_num * sizeof(uint16_t))
2535 if( h->pps.cabac ) {
2536 CHECKED_ALLOCZ(h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t))
2537 CHECKED_ALLOCZ(h->mvd_table[0], 32*big_mb_num * sizeof(uint16_t));
2538 CHECKED_ALLOCZ(h->mvd_table[1], 32*big_mb_num * sizeof(uint16_t));
2539 CHECKED_ALLOCZ(h->direct_table, 32*big_mb_num * sizeof(uint8_t));
2542 memset(h->slice_table_base, -1, big_mb_num * sizeof(uint8_t));
2543 h->slice_table= h->slice_table_base + s->mb_stride + 1;
2545 CHECKED_ALLOCZ(h->mb2b_xy , big_mb_num * sizeof(uint16_t));
2546 CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint16_t));
2547 for(y=0; y<s->mb_height; y++){
2548 for(x=0; x<s->mb_width; x++){
2549 const int mb_xy= x + y*s->mb_stride;
2550 const int b_xy = 4*x + 4*y*h->b_stride;
2551 const int b8_xy= 2*x + 2*y*h->b8_stride;
2553 h->mb2b_xy [mb_xy]= b_xy;
2554 h->mb2b8_xy[mb_xy]= b8_xy;
2558 s->obmc_scratchpad = NULL;
2566 static void common_init(H264Context *h){
2567 MpegEncContext * const s = &h->s;
2569 s->width = s->avctx->width;
2570 s->height = s->avctx->height;
2571 s->codec_id= s->avctx->codec->id;
2575 s->unrestricted_mv=1;
2576 s->decode=1; //FIXME
2579 static int decode_init(AVCodecContext *avctx){
2580 H264Context *h= avctx->priv_data;
2581 MpegEncContext * const s = &h->s;
2583 MPV_decode_defaults(s);
2588 s->out_format = FMT_H264;
2589 s->workaround_bugs= avctx->workaround_bugs;
2592 // s->decode_mb= ff_h263_decode_mb;
2594 avctx->pix_fmt= PIX_FMT_YUV420P;
2598 if(avctx->codec_tag != 0x31637661 && avctx->codec_tag != 0x31435641) // avc1
2601 if((avctx->extradata_size == 0) || (avctx->extradata == NULL)) {
2602 av_log(avctx, AV_LOG_ERROR, "AVC codec requires avcC data\n");
2612 static void frame_start(H264Context *h){
2613 MpegEncContext * const s = &h->s;
2616 MPV_frame_start(s, s->avctx);
2617 ff_er_frame_start(s);
2619 assert(s->linesize && s->uvlinesize);
2621 for(i=0; i<16; i++){
2622 h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
2623 h->chroma_subblock_offset[i]= 2*((scan8[i] - scan8[0])&7) + 2*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2626 h->block_offset[16+i]=
2627 h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2630 /* can't be in alloc_tables because linesize isn't known there.
2631 * FIXME: redo bipred weight to not require extra buffer? */
2632 if(!s->obmc_scratchpad)
2633 s->obmc_scratchpad = av_malloc(16*s->linesize + 2*8*s->uvlinesize);
2635 // s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
2638 static inline void backup_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){
2639 MpegEncContext * const s = &h->s;
2643 src_cb -= uvlinesize;
2644 src_cr -= uvlinesize;
2646 h->left_border[0]= h->top_border[s->mb_x][15];
2647 for(i=1; i<17; i++){
2648 h->left_border[i]= src_y[15+i* linesize];
2651 *(uint64_t*)(h->top_border[s->mb_x]+0)= *(uint64_t*)(src_y + 16*linesize);
2652 *(uint64_t*)(h->top_border[s->mb_x]+8)= *(uint64_t*)(src_y +8+16*linesize);
2654 if(!(s->flags&CODEC_FLAG_GRAY)){
2655 h->left_border[17 ]= h->top_border[s->mb_x][16+7];
2656 h->left_border[17+9]= h->top_border[s->mb_x][24+7];
2658 h->left_border[i+17 ]= src_cb[7+i*uvlinesize];
2659 h->left_border[i+17+9]= src_cr[7+i*uvlinesize];
2661 *(uint64_t*)(h->top_border[s->mb_x]+16)= *(uint64_t*)(src_cb+8*uvlinesize);
2662 *(uint64_t*)(h->top_border[s->mb_x]+24)= *(uint64_t*)(src_cr+8*uvlinesize);
2666 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){
2667 MpegEncContext * const s = &h->s;
2670 int deblock_left = (s->mb_x > 0);
2671 int deblock_top = (s->mb_y > 0);
2673 src_y -= linesize + 1;
2674 src_cb -= uvlinesize + 1;
2675 src_cr -= uvlinesize + 1;
2677 #define XCHG(a,b,t,xchg)\
2684 for(i = !deblock_top; i<17; i++){
2685 XCHG(h->left_border[i ], src_y [i* linesize], temp8, xchg);
2690 XCHG(*(uint64_t*)(h->top_border[s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
2691 XCHG(*(uint64_t*)(h->top_border[s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
2694 if(!(s->flags&CODEC_FLAG_GRAY)){
2696 for(i = !deblock_top; i<9; i++){
2697 XCHG(h->left_border[i+17 ], src_cb[i*uvlinesize], temp8, xchg);
2698 XCHG(h->left_border[i+17+9], src_cr[i*uvlinesize], temp8, xchg);
2702 XCHG(*(uint64_t*)(h->top_border[s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
2703 XCHG(*(uint64_t*)(h->top_border[s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
2708 static void hl_decode_mb(H264Context *h){
2709 MpegEncContext * const s = &h->s;
2710 const int mb_x= s->mb_x;
2711 const int mb_y= s->mb_y;
2712 const int mb_xy= mb_x + mb_y*s->mb_stride;
2713 const int mb_type= s->current_picture.mb_type[mb_xy];
2714 uint8_t *dest_y, *dest_cb, *dest_cr;
2715 int linesize, uvlinesize /*dct_offset*/;
2724 dest_y = s->current_picture.data[0] + (mb_y * 16* s->linesize ) + mb_x * 16;
2725 dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2726 dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2728 if (h->mb_field_decoding_flag) {
2729 linesize = s->linesize * 2;
2730 uvlinesize = s->uvlinesize * 2;
2731 if(mb_y&1){ //FIXME move out of this func?
2732 dest_y -= s->linesize*15;
2733 dest_cb-= s->linesize*7;
2734 dest_cr-= s->linesize*7;
2737 linesize = s->linesize;
2738 uvlinesize = s->uvlinesize;
2739 // dct_offset = s->linesize * 16;
2742 if (IS_INTRA_PCM(mb_type)) {
2745 // The pixels are stored in h->mb array in the same order as levels,
2746 // copy them in output in the correct order.
2747 for(i=0; i<16; i++) {
2748 for (y=0; y<4; y++) {
2749 for (x=0; x<4; x++) {
2750 *(dest_y + h->block_offset[i] + y*linesize + x) = h->mb[i*16+y*4+x];
2754 for(i=16; i<16+4; i++) {
2755 for (y=0; y<4; y++) {
2756 for (x=0; x<4; x++) {
2757 *(dest_cb + h->block_offset[i] + y*uvlinesize + x) = h->mb[i*16+y*4+x];
2761 for(i=20; i<20+4; i++) {
2762 for (y=0; y<4; y++) {
2763 for (x=0; x<4; x++) {
2764 *(dest_cr + h->block_offset[i] + y*uvlinesize + x) = h->mb[i*16+y*4+x];
2769 if(IS_INTRA(mb_type)){
2770 if(h->deblocking_filter)
2771 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1);
2773 if(!(s->flags&CODEC_FLAG_GRAY)){
2774 h->pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
2775 h->pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
2778 if(IS_INTRA4x4(mb_type)){
2780 for(i=0; i<16; i++){
2781 uint8_t * const ptr= dest_y + h->block_offset[i];
2783 const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
2786 if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
2787 const int topright_avail= (h->topright_samples_available<<i)&0x8000;
2788 assert(mb_y || linesize <= h->block_offset[i]);
2789 if(!topright_avail){
2790 tr= ptr[3 - linesize]*0x01010101;
2791 topright= (uint8_t*) &tr;
2792 }else if(i==5 && h->deblocking_filter){
2793 tr= *(uint32_t*)h->top_border[mb_x+1];
2794 topright= (uint8_t*) &tr;
2796 topright= ptr + 4 - linesize;
2800 h->pred4x4[ dir ](ptr, topright, linesize);
2801 if(h->non_zero_count_cache[ scan8[i] ]){
2802 if(s->codec_id == CODEC_ID_H264)
2803 s->dsp.h264_idct_add(ptr, h->mb + i*16, linesize);
2805 svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
2810 h->pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
2811 if(s->codec_id == CODEC_ID_H264)
2812 h264_luma_dc_dequant_idct_c(h->mb, s->qscale);
2814 svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
2816 if(h->deblocking_filter)
2817 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);
2818 }else if(s->codec_id == CODEC_ID_H264){
2819 hl_motion(h, dest_y, dest_cb, dest_cr,
2820 s->dsp.put_h264_qpel_pixels_tab, s->dsp.put_h264_chroma_pixels_tab,
2821 s->dsp.avg_h264_qpel_pixels_tab, s->dsp.avg_h264_chroma_pixels_tab,
2822 s->dsp.weight_h264_pixels_tab, s->dsp.biweight_h264_pixels_tab);
2826 if(!IS_INTRA4x4(mb_type)){
2827 if(s->codec_id == CODEC_ID_H264){
2828 for(i=0; i<16; i++){
2829 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
2830 uint8_t * const ptr= dest_y + h->block_offset[i];
2831 s->dsp.h264_idct_add(ptr, h->mb + i*16, linesize);
2835 for(i=0; i<16; i++){
2836 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
2837 uint8_t * const ptr= dest_y + h->block_offset[i];
2838 svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
2844 if(!(s->flags&CODEC_FLAG_GRAY)){
2845 chroma_dc_dequant_idct_c(h->mb + 16*16, h->chroma_qp);
2846 chroma_dc_dequant_idct_c(h->mb + 16*16+4*16, h->chroma_qp);
2847 if(s->codec_id == CODEC_ID_H264){
2848 for(i=16; i<16+4; i++){
2849 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2850 uint8_t * const ptr= dest_cb + h->block_offset[i];
2851 s->dsp.h264_idct_add(ptr, h->mb + i*16, uvlinesize);
2854 for(i=20; i<20+4; i++){
2855 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2856 uint8_t * const ptr= dest_cr + h->block_offset[i];
2857 s->dsp.h264_idct_add(ptr, h->mb + i*16, uvlinesize);
2861 for(i=16; i<16+4; i++){
2862 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2863 uint8_t * const ptr= dest_cb + h->block_offset[i];
2864 svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
2867 for(i=20; i<20+4; i++){
2868 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2869 uint8_t * const ptr= dest_cr + h->block_offset[i];
2870 svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
2876 if(h->deblocking_filter) {
2877 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2878 fill_caches(h, mb_type, 1); //FIXME dont fill stuff which isnt used by filter_mb
2879 filter_mb(h, mb_x, mb_y, dest_y, dest_cb, dest_cr);
2884 * fills the default_ref_list.
2886 static int fill_default_ref_list(H264Context *h){
2887 MpegEncContext * const s = &h->s;
2889 int smallest_poc_greater_than_current = -1;
2890 Picture sorted_short_ref[16];
2892 if(h->slice_type==B_TYPE){
2896 /* sort frame according to poc in B slice */
2897 for(out_i=0; out_i<h->short_ref_count; out_i++){
2899 int best_poc=INT_MAX;
2901 for(i=0; i<h->short_ref_count; i++){
2902 const int poc= h->short_ref[i]->poc;
2903 if(poc > limit && poc < best_poc){
2909 assert(best_i != -1);
2912 sorted_short_ref[out_i]= *h->short_ref[best_i];
2913 tprintf("sorted poc: %d->%d poc:%d fn:%d\n", best_i, out_i, sorted_short_ref[out_i].poc, sorted_short_ref[out_i].frame_num);
2914 if (-1 == smallest_poc_greater_than_current) {
2915 if (h->short_ref[best_i]->poc >= s->current_picture_ptr->poc) {
2916 smallest_poc_greater_than_current = out_i;
2922 if(s->picture_structure == PICT_FRAME){
2923 if(h->slice_type==B_TYPE){
2925 tprintf("current poc: %d, smallest_poc_greater_than_current: %d\n", s->current_picture_ptr->poc, smallest_poc_greater_than_current);
2927 // find the largest poc
2928 for(list=0; list<2; list++){
2931 int step= list ? -1 : 1;
2933 for(i=0; i<h->short_ref_count && index < h->ref_count[list]; i++, j+=step) {
2934 while(j<0 || j>= h->short_ref_count){
2936 j= smallest_poc_greater_than_current + (step>>1);
2938 if(sorted_short_ref[j].reference != 3) continue;
2939 h->default_ref_list[list][index ]= sorted_short_ref[j];
2940 h->default_ref_list[list][index++].pic_id= sorted_short_ref[j].frame_num;
2943 for(i = 0; i < 16 && index < h->ref_count[ list ]; i++){
2944 if(h->long_ref[i] == NULL) continue;
2945 if(h->long_ref[i]->reference != 3) continue;
2947 h->default_ref_list[ list ][index ]= *h->long_ref[i];
2948 h->default_ref_list[ list ][index++].pic_id= i;;
2951 if(list && (smallest_poc_greater_than_current<=0 || smallest_poc_greater_than_current>=h->short_ref_count) && (1 < index)){
2952 // swap the two first elements of L1 when
2953 // L0 and L1 are identical
2954 Picture temp= h->default_ref_list[1][0];
2955 h->default_ref_list[1][0] = h->default_ref_list[1][1];
2956 h->default_ref_list[1][0] = temp;
2959 if(index < h->ref_count[ list ])
2960 memset(&h->default_ref_list[list][index], 0, sizeof(Picture)*(h->ref_count[ list ] - index));
2964 for(i=0; i<h->short_ref_count && index < h->ref_count[0]; i++){
2965 if(h->short_ref[i]->reference != 3) continue; //FIXME refernce field shit
2966 h->default_ref_list[0][index ]= *h->short_ref[i];
2967 h->default_ref_list[0][index++].pic_id= h->short_ref[i]->frame_num;
2969 for(i = 0; i < 16 && index < h->ref_count[0]; i++){
2970 if(h->long_ref[i] == NULL) continue;
2971 if(h->long_ref[i]->reference != 3) continue;
2972 h->default_ref_list[0][index ]= *h->long_ref[i];
2973 h->default_ref_list[0][index++].pic_id= i;;
2975 if(index < h->ref_count[0])
2976 memset(&h->default_ref_list[0][index], 0, sizeof(Picture)*(h->ref_count[0] - index));
2979 if(h->slice_type==B_TYPE){
2981 //FIXME second field balh
2985 for (i=0; i<h->ref_count[0]; i++) {
2986 tprintf("List0: %s fn:%d 0x%p\n", (h->default_ref_list[0][i].long_ref ? "LT" : "ST"), h->default_ref_list[0][i].pic_id, h->default_ref_list[0][i].data[0]);
2988 if(h->slice_type==B_TYPE){
2989 for (i=0; i<h->ref_count[1]; i++) {
2990 tprintf("List1: %s fn:%d 0x%p\n", (h->default_ref_list[1][i].long_ref ? "LT" : "ST"), h->default_ref_list[1][i].pic_id, h->default_ref_list[0][i].data[0]);
2997 static void print_short_term(H264Context *h);
2998 static void print_long_term(H264Context *h);
3000 static int decode_ref_pic_list_reordering(H264Context *h){
3001 MpegEncContext * const s = &h->s;
3004 print_short_term(h);
3006 if(h->slice_type==I_TYPE || h->slice_type==SI_TYPE) return 0; //FIXME move beofre func
3008 for(list=0; list<2; list++){
3009 memcpy(h->ref_list[list], h->default_ref_list[list], sizeof(Picture)*h->ref_count[list]);
3011 if(get_bits1(&s->gb)){
3012 int pred= h->curr_pic_num;
3015 for(index=0; ; index++){
3016 int reordering_of_pic_nums_idc= get_ue_golomb(&s->gb);
3020 if(reordering_of_pic_nums_idc==3)
3023 if(index >= h->ref_count[list]){
3024 av_log(h->s.avctx, AV_LOG_ERROR, "reference count overflow\n");
3028 if(reordering_of_pic_nums_idc<3){
3029 if(reordering_of_pic_nums_idc<2){
3030 const int abs_diff_pic_num= get_ue_golomb(&s->gb) + 1;
3032 if(abs_diff_pic_num >= h->max_pic_num){
3033 av_log(h->s.avctx, AV_LOG_ERROR, "abs_diff_pic_num overflow\n");
3037 if(reordering_of_pic_nums_idc == 0) pred-= abs_diff_pic_num;
3038 else pred+= abs_diff_pic_num;
3039 pred &= h->max_pic_num - 1;
3041 for(i= h->ref_count[list]-1; i>=0; i--){
3042 if(h->ref_list[list][i].data[0] != NULL && h->ref_list[list][i].pic_id == pred && h->ref_list[list][i].long_ref==0) // ignore non existing pictures by testing data[0] pointer
3046 pic_id= get_ue_golomb(&s->gb); //long_term_pic_idx
3048 for(i= h->ref_count[list]-1; i>=0; i--){
3049 if(h->ref_list[list][i].pic_id == pic_id && h->ref_list[list][i].long_ref==1) // no need to ignore non existing pictures as non existing pictures have long_ref==0
3055 av_log(h->s.avctx, AV_LOG_ERROR, "reference picture missing during reorder\n");
3056 memset(&h->ref_list[list][index], 0, sizeof(Picture)); //FIXME
3057 } else if (i != index) /* this test is not necessary, it is only an optimisation to skip double copy of Picture structure in this case */ {
3058 Picture tmp= h->ref_list[list][i];
3060 i = h->ref_count[list];
3062 for(; i > index; i--){
3063 h->ref_list[list][i]= h->ref_list[list][i-1];
3065 h->ref_list[list][index]= tmp;
3068 av_log(h->s.avctx, AV_LOG_ERROR, "illegal reordering_of_pic_nums_idc\n");
3074 if(h->slice_type!=B_TYPE) break;
3077 if(h->slice_type==B_TYPE && !h->direct_spatial_mv_pred)
3078 direct_dist_scale_factor(h);
3082 static int pred_weight_table(H264Context *h){
3083 MpegEncContext * const s = &h->s;
3085 int luma_def, chroma_def;
3088 h->use_weight_chroma= 0;
3089 h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
3090 h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
3091 luma_def = 1<<h->luma_log2_weight_denom;
3092 chroma_def = 1<<h->chroma_log2_weight_denom;
3094 for(list=0; list<2; list++){
3095 for(i=0; i<h->ref_count[list]; i++){
3096 int luma_weight_flag, chroma_weight_flag;
3098 luma_weight_flag= get_bits1(&s->gb);
3099 if(luma_weight_flag){
3100 h->luma_weight[list][i]= get_se_golomb(&s->gb);
3101 h->luma_offset[list][i]= get_se_golomb(&s->gb);
3102 if( h->luma_weight[list][i] != luma_def
3103 || h->luma_offset[list][i] != 0)
3106 h->luma_weight[list][i]= luma_def;
3107 h->luma_offset[list][i]= 0;
3110 chroma_weight_flag= get_bits1(&s->gb);
3111 if(chroma_weight_flag){
3114 h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);
3115 h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);
3116 if( h->chroma_weight[list][i][j] != chroma_def
3117 || h->chroma_offset[list][i][j] != 0)
3118 h->use_weight_chroma= 1;
3123 h->chroma_weight[list][i][j]= chroma_def;
3124 h->chroma_offset[list][i][j]= 0;
3128 if(h->slice_type != B_TYPE) break;
3130 h->use_weight= h->use_weight || h->use_weight_chroma;
3134 static void implicit_weight_table(H264Context *h){
3135 MpegEncContext * const s = &h->s;
3137 int cur_poc = s->current_picture_ptr->poc;
3139 if( h->ref_count[0] == 1 && h->ref_count[1] == 1
3140 && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
3142 h->use_weight_chroma= 0;
3147 h->use_weight_chroma= 2;
3148 h->luma_log2_weight_denom= 5;
3149 h->chroma_log2_weight_denom= 5;
3152 for(ref0=0; ref0 < h->ref_count[0]; ref0++){
3153 int poc0 = h->ref_list[0][ref0].poc;
3154 for(ref1=0; ref1 < h->ref_count[1]; ref1++){
3155 int poc1 = h->ref_list[1][ref1].poc;
3156 int td = clip(poc1 - poc0, -128, 127);
3158 int tb = clip(cur_poc - poc0, -128, 127);
3159 int tx = (16384 + (ABS(td) >> 1)) / td;
3160 int dist_scale_factor = clip((tb*tx + 32) >> 6, -1024, 1023) >> 2;
3161 if(dist_scale_factor < -64 || dist_scale_factor > 128)
3162 h->implicit_weight[ref0][ref1] = 32;
3164 h->implicit_weight[ref0][ref1] = 64 - dist_scale_factor;
3166 h->implicit_weight[ref0][ref1] = 32;
3172 * instantaneous decoder refresh.
3174 static void idr(H264Context *h){
3177 #define CHECK_DELAY(pic) \
3178 for(j = 0; h->delayed_pic[j]; j++) \
3179 if(pic == h->delayed_pic[j]){ \
3184 for(i=0; i<16; i++){
3185 if (h->long_ref[i] != NULL) {
3186 h->long_ref[i]->reference=0;
3187 CHECK_DELAY(h->long_ref[i]);
3188 h->long_ref[i]= NULL;
3191 h->long_ref_count=0;
3193 for(i=0; i<h->short_ref_count; i++){
3194 h->short_ref[i]->reference=0;
3195 CHECK_DELAY(h->short_ref[i]);
3196 h->short_ref[i]= NULL;
3198 h->short_ref_count=0;
3204 * @return the removed picture or NULL if an error occures
3206 static Picture * remove_short(H264Context *h, int frame_num){
3207 MpegEncContext * const s = &h->s;
3210 if(s->avctx->debug&FF_DEBUG_MMCO)
3211 av_log(h->s.avctx, AV_LOG_DEBUG, "remove short %d count %d\n", frame_num, h->short_ref_count);
3213 for(i=0; i<h->short_ref_count; i++){
3214 Picture *pic= h->short_ref[i];
3215 if(s->avctx->debug&FF_DEBUG_MMCO)
3216 av_log(h->s.avctx, AV_LOG_DEBUG, "%d %d %p\n", i, pic->frame_num, pic);
3217 if(pic->frame_num == frame_num){
3218 h->short_ref[i]= NULL;
3219 memmove(&h->short_ref[i], &h->short_ref[i+1], (h->short_ref_count - i - 1)*sizeof(Picture*));
3220 h->short_ref_count--;
3229 * @return the removed picture or NULL if an error occures
3231 static Picture * remove_long(H264Context *h, int i){
3234 pic= h->long_ref[i];
3235 h->long_ref[i]= NULL;
3236 if(pic) h->long_ref_count--;
3242 * print short term list
3244 static void print_short_term(H264Context *h) {
3246 if(h->s.avctx->debug&FF_DEBUG_MMCO) {
3247 av_log(h->s.avctx, AV_LOG_DEBUG, "short term list:\n");
3248 for(i=0; i<h->short_ref_count; i++){
3249 Picture *pic= h->short_ref[i];
3250 av_log(h->s.avctx, AV_LOG_DEBUG, "%d fn:%d poc:%d %p\n", i, pic->frame_num, pic->poc, pic->data[0]);
3256 * print long term list
3258 static void print_long_term(H264Context *h) {
3260 if(h->s.avctx->debug&FF_DEBUG_MMCO) {
3261 av_log(h->s.avctx, AV_LOG_DEBUG, "long term list:\n");
3262 for(i = 0; i < 16; i++){
3263 Picture *pic= h->long_ref[i];
3265 av_log(h->s.avctx, AV_LOG_DEBUG, "%d fn:%d poc:%d %p\n", i, pic->frame_num, pic->poc, pic->data[0]);
3272 * Executes the reference picture marking (memory management control operations).
3274 static int execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count){
3275 MpegEncContext * const s = &h->s;
3277 int current_is_long=0;
3280 if((s->avctx->debug&FF_DEBUG_MMCO) && mmco_count==0)
3281 av_log(h->s.avctx, AV_LOG_DEBUG, "no mmco here\n");
3283 for(i=0; i<mmco_count; i++){
3284 if(s->avctx->debug&FF_DEBUG_MMCO)
3285 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);
3287 switch(mmco[i].opcode){
3288 case MMCO_SHORT2UNUSED:
3289 pic= remove_short(h, mmco[i].short_frame_num);
3290 if(pic==NULL) return -1;
3293 case MMCO_SHORT2LONG:
3294 pic= remove_long(h, mmco[i].long_index);
3295 if(pic) pic->reference=0;
3297 h->long_ref[ mmco[i].long_index ]= remove_short(h, mmco[i].short_frame_num);
3298 h->long_ref[ mmco[i].long_index ]->long_ref=1;
3299 h->long_ref_count++;
3301 case MMCO_LONG2UNUSED:
3302 pic= remove_long(h, mmco[i].long_index);
3303 if(pic==NULL) return -1;
3307 pic= remove_long(h, mmco[i].long_index);
3308 if(pic) pic->reference=0;
3310 h->long_ref[ mmco[i].long_index ]= s->current_picture_ptr;
3311 h->long_ref[ mmco[i].long_index ]->long_ref=1;
3312 h->long_ref_count++;
3316 case MMCO_SET_MAX_LONG:
3317 assert(mmco[i].long_index <= 16);
3318 // just remove the long term which index is greater than new max
3319 for(j = mmco[i].long_index; j<16; j++){
3320 pic = remove_long(h, j);
3321 if (pic) pic->reference=0;
3325 while(h->short_ref_count){
3326 pic= remove_short(h, h->short_ref[0]->frame_num);
3329 for(j = 0; j < 16; j++) {
3330 pic= remove_long(h, j);
3331 if(pic) pic->reference=0;
3338 if(!current_is_long){
3339 pic= remove_short(h, s->current_picture_ptr->frame_num);
3342 av_log(h->s.avctx, AV_LOG_ERROR, "illegal short term buffer state detected\n");
3345 if(h->short_ref_count)
3346 memmove(&h->short_ref[1], &h->short_ref[0], h->short_ref_count*sizeof(Picture*));
3348 h->short_ref[0]= s->current_picture_ptr;
3349 h->short_ref[0]->long_ref=0;
3350 h->short_ref_count++;
3353 print_short_term(h);
3358 static int decode_ref_pic_marking(H264Context *h){
3359 MpegEncContext * const s = &h->s;
3362 if(h->nal_unit_type == NAL_IDR_SLICE){ //FIXME fields
3363 s->broken_link= get_bits1(&s->gb) -1;
3364 h->mmco[0].long_index= get_bits1(&s->gb) - 1; // current_long_term_idx
3365 if(h->mmco[0].long_index == -1)
3368 h->mmco[0].opcode= MMCO_LONG;
3372 if(get_bits1(&s->gb)){ // adaptive_ref_pic_marking_mode_flag
3373 for(i= 0; i<MAX_MMCO_COUNT; i++) {
3374 MMCOOpcode opcode= get_ue_golomb(&s->gb);;
3376 h->mmco[i].opcode= opcode;
3377 if(opcode==MMCO_SHORT2UNUSED || opcode==MMCO_SHORT2LONG){
3378 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
3379 /* if(h->mmco[i].short_frame_num >= h->short_ref_count || h->short_ref[ h->mmco[i].short_frame_num ] == NULL){
3380 fprintf(stderr, "illegal short ref in memory management control operation %d\n", mmco);
3384 if(opcode==MMCO_SHORT2LONG || opcode==MMCO_LONG2UNUSED || opcode==MMCO_LONG || opcode==MMCO_SET_MAX_LONG){
3385 h->mmco[i].long_index= get_ue_golomb(&s->gb);
3386 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){
3387 av_log(h->s.avctx, AV_LOG_ERROR, "illegal long ref in memory management control operation %d\n", opcode);
3392 if(opcode > MMCO_LONG){
3393 av_log(h->s.avctx, AV_LOG_ERROR, "illegal memory management control operation %d\n", opcode);
3396 if(opcode == MMCO_END)
3401 assert(h->long_ref_count + h->short_ref_count <= h->sps.ref_frame_count);
3403 if(h->long_ref_count + h->short_ref_count == h->sps.ref_frame_count){ //FIXME fields
3404 h->mmco[0].opcode= MMCO_SHORT2UNUSED;
3405 h->mmco[0].short_frame_num= h->short_ref[ h->short_ref_count - 1 ]->frame_num;
3415 static int init_poc(H264Context *h){
3416 MpegEncContext * const s = &h->s;
3417 const int max_frame_num= 1<<h->sps.log2_max_frame_num;
3420 if(h->nal_unit_type == NAL_IDR_SLICE){
3421 h->frame_num_offset= 0;
3423 if(h->frame_num < h->prev_frame_num)
3424 h->frame_num_offset= h->prev_frame_num_offset + max_frame_num;
3426 h->frame_num_offset= h->prev_frame_num_offset;
3429 if(h->sps.poc_type==0){
3430 const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
3432 if (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
3433 h->poc_msb = h->prev_poc_msb + max_poc_lsb;
3434 else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
3435 h->poc_msb = h->prev_poc_msb - max_poc_lsb;
3437 h->poc_msb = h->prev_poc_msb;
3438 //printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
3440 field_poc[1] = h->poc_msb + h->poc_lsb;
3441 if(s->picture_structure == PICT_FRAME)
3442 field_poc[1] += h->delta_poc_bottom;
3443 }else if(h->sps.poc_type==1){
3444 int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
3447 if(h->sps.poc_cycle_length != 0)
3448 abs_frame_num = h->frame_num_offset + h->frame_num;
3452 if(h->nal_ref_idc==0 && abs_frame_num > 0)
3455 expected_delta_per_poc_cycle = 0;
3456 for(i=0; i < h->sps.poc_cycle_length; i++)
3457 expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
3459 if(abs_frame_num > 0){
3460 int poc_cycle_cnt = (abs_frame_num - 1) / h->sps.poc_cycle_length;
3461 int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
3463 expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
3464 for(i = 0; i <= frame_num_in_poc_cycle; i++)
3465 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
3469 if(h->nal_ref_idc == 0)
3470 expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
3472 field_poc[0] = expectedpoc + h->delta_poc[0];
3473 field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
3475 if(s->picture_structure == PICT_FRAME)
3476 field_poc[1] += h->delta_poc[1];
3479 if(h->nal_unit_type == NAL_IDR_SLICE){
3482 if(h->nal_ref_idc) poc= 2*(h->frame_num_offset + h->frame_num);
3483 else poc= 2*(h->frame_num_offset + h->frame_num) - 1;
3489 if(s->picture_structure != PICT_BOTTOM_FIELD)
3490 s->current_picture_ptr->field_poc[0]= field_poc[0];
3491 if(s->picture_structure != PICT_TOP_FIELD)
3492 s->current_picture_ptr->field_poc[1]= field_poc[1];
3493 if(s->picture_structure == PICT_FRAME) // FIXME field pix?
3494 s->current_picture_ptr->poc= FFMIN(field_poc[0], field_poc[1]);
3500 * decodes a slice header.
3501 * this will allso call MPV_common_init() and frame_start() as needed
3503 static int decode_slice_header(H264Context *h){
3504 MpegEncContext * const s = &h->s;
3505 int first_mb_in_slice, pps_id;
3506 int num_ref_idx_active_override_flag;
3507 static const uint8_t slice_type_map[5]= {P_TYPE, B_TYPE, I_TYPE, SP_TYPE, SI_TYPE};
3509 int default_ref_list_done = 0;
3511 s->current_picture.reference= h->nal_ref_idc != 0;
3513 first_mb_in_slice= get_ue_golomb(&s->gb);
3515 slice_type= get_ue_golomb(&s->gb);
3517 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);
3522 h->slice_type_fixed=1;
3524 h->slice_type_fixed=0;
3526 slice_type= slice_type_map[ slice_type ];
3527 if (slice_type == I_TYPE
3528 || (h->slice_num != 0 && slice_type == h->slice_type) ) {
3529 default_ref_list_done = 1;
3531 h->slice_type= slice_type;
3533 s->pict_type= h->slice_type; // to make a few old func happy, its wrong though
3535 pps_id= get_ue_golomb(&s->gb);
3537 av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
3540 h->pps= h->pps_buffer[pps_id];
3541 if(h->pps.slice_group_count == 0){
3542 av_log(h->s.avctx, AV_LOG_ERROR, "non existing PPS referenced\n");
3546 h->sps= h->sps_buffer[ h->pps.sps_id ];
3547 if(h->sps.log2_max_frame_num == 0){
3548 av_log(h->s.avctx, AV_LOG_ERROR, "non existing SPS referenced\n");
3552 s->mb_width= h->sps.mb_width;
3553 s->mb_height= h->sps.mb_height;
3555 h->b_stride= s->mb_width*4 + 1;
3556 h->b8_stride= s->mb_width*2 + 1;
3558 s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
3559 s->resync_mb_y = s->mb_y = first_mb_in_slice / s->mb_width; //FIXME AFFW
3561 s->width = 16*s->mb_width - 2*(h->sps.crop_left + h->sps.crop_right );
3562 if(h->sps.frame_mbs_only_flag)
3563 s->height= 16*s->mb_height - 2*(h->sps.crop_top + h->sps.crop_bottom);
3565 s->height= 16*s->mb_height - 4*(h->sps.crop_top + h->sps.crop_bottom); //FIXME recheck
3567 if (s->context_initialized
3568 && ( s->width != s->avctx->width || s->height != s->avctx->height)) {
3572 if (!s->context_initialized) {
3573 if (MPV_common_init(s) < 0)
3578 s->avctx->width = s->width;
3579 s->avctx->height = s->height;
3580 s->avctx->sample_aspect_ratio= h->sps.sar;
3581 if(!s->avctx->sample_aspect_ratio.den)
3582 s->avctx->sample_aspect_ratio.den = 1;
3584 if(h->sps.timing_info_present_flag && h->sps.fixed_frame_rate_flag){
3585 s->avctx->frame_rate = h->sps.time_scale;
3586 s->avctx->frame_rate_base = h->sps.num_units_in_tick;
3590 if(h->slice_num == 0){
3594 s->current_picture_ptr->frame_num= //FIXME frame_num cleanup
3595 h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
3597 if(h->sps.frame_mbs_only_flag){
3598 s->picture_structure= PICT_FRAME;
3600 if(get_bits1(&s->gb)) //field_pic_flag
3601 s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
3603 s->picture_structure= PICT_FRAME;
3606 if(s->picture_structure==PICT_FRAME){
3607 h->curr_pic_num= h->frame_num;
3608 h->max_pic_num= 1<< h->sps.log2_max_frame_num;
3610 h->curr_pic_num= 2*h->frame_num;
3611 h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
3614 if(h->nal_unit_type == NAL_IDR_SLICE){
3615 get_ue_golomb(&s->gb); /* idr_pic_id */
3618 if(h->sps.poc_type==0){
3619 h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
3621 if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
3622 h->delta_poc_bottom= get_se_golomb(&s->gb);
3626 if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
3627 h->delta_poc[0]= get_se_golomb(&s->gb);
3629 if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
3630 h->delta_poc[1]= get_se_golomb(&s->gb);
3635 if(h->pps.redundant_pic_cnt_present){
3636 h->redundant_pic_count= get_ue_golomb(&s->gb);
3639 //set defaults, might be overriden a few line later
3640 h->ref_count[0]= h->pps.ref_count[0];
3641 h->ref_count[1]= h->pps.ref_count[1];
3643 if(h->slice_type == P_TYPE || h->slice_type == SP_TYPE || h->slice_type == B_TYPE){
3644 if(h->slice_type == B_TYPE){
3645 h->direct_spatial_mv_pred= get_bits1(&s->gb);
3647 num_ref_idx_active_override_flag= get_bits1(&s->gb);
3649 if(num_ref_idx_active_override_flag){
3650 h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
3651 if(h->slice_type==B_TYPE)
3652 h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
3654 if(h->ref_count[0] > 32 || h->ref_count[1] > 32){
3655 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
3661 if(!default_ref_list_done){
3662 fill_default_ref_list(h);
3665 decode_ref_pic_list_reordering(h);
3667 if( (h->pps.weighted_pred && (h->slice_type == P_TYPE || h->slice_type == SP_TYPE ))
3668 || (h->pps.weighted_bipred_idc==1 && h->slice_type==B_TYPE ) )
3669 pred_weight_table(h);
3670 else if(h->pps.weighted_bipred_idc==2 && h->slice_type==B_TYPE)
3671 implicit_weight_table(h);
3675 if(s->current_picture.reference)
3676 decode_ref_pic_marking(h);
3678 if( h->slice_type != I_TYPE && h->slice_type != SI_TYPE && h->pps.cabac )
3679 h->cabac_init_idc = get_ue_golomb(&s->gb);
3681 h->last_qscale_diff = 0;
3682 s->qscale = h->pps.init_qp + get_se_golomb(&s->gb);
3683 if(s->qscale<0 || s->qscale>51){
3684 av_log(s->avctx, AV_LOG_ERROR, "QP %d out of range\n", s->qscale);
3687 h->chroma_qp = get_chroma_qp(h, s->qscale);
3688 //FIXME qscale / qp ... stuff
3689 if(h->slice_type == SP_TYPE){
3690 get_bits1(&s->gb); /* sp_for_switch_flag */
3692 if(h->slice_type==SP_TYPE || h->slice_type == SI_TYPE){
3693 get_se_golomb(&s->gb); /* slice_qs_delta */
3696 h->deblocking_filter = 1;
3697 h->slice_alpha_c0_offset = 0;
3698 h->slice_beta_offset = 0;
3699 if( h->pps.deblocking_filter_parameters_present ) {
3700 h->deblocking_filter= get_ue_golomb(&s->gb);
3701 if(h->deblocking_filter < 2)
3702 h->deblocking_filter^= 1; // 1<->0
3704 if( h->deblocking_filter ) {
3705 h->slice_alpha_c0_offset = get_se_golomb(&s->gb) << 1;
3706 h->slice_beta_offset = get_se_golomb(&s->gb) << 1;
3711 if( h->pps.num_slice_groups > 1 && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
3712 slice_group_change_cycle= get_bits(&s->gb, ?);
3717 if(s->avctx->debug&FF_DEBUG_PICT_INFO){
3718 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 weight:%d%s\n",
3719 h->slice_num, first_mb_in_slice,
3720 av_get_pict_type_char(h->slice_type),
3721 pps_id, h->frame_num,
3722 s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
3723 h->ref_count[0], h->ref_count[1],
3725 h->deblocking_filter,
3727 h->use_weight==1 && h->use_weight_chroma ? "c" : ""
3737 static inline int get_level_prefix(GetBitContext *gb){
3741 OPEN_READER(re, gb);
3742 UPDATE_CACHE(re, gb);
3743 buf=GET_CACHE(re, gb);
3745 log= 32 - av_log2(buf);
3747 print_bin(buf>>(32-log), log);
3748 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__);
3751 LAST_SKIP_BITS(re, gb, log);
3752 CLOSE_READER(re, gb);
3758 * decodes a residual block.
3759 * @param n block index
3760 * @param scantable scantable
3761 * @param max_coeff number of coefficients in the block
3762 * @return <0 if an error occured
3764 static int decode_residual(H264Context *h, GetBitContext *gb, DCTELEM *block, int n, const uint8_t *scantable, int qp, int max_coeff){
3765 MpegEncContext * const s = &h->s;
3766 const uint16_t *qmul= dequant_coeff[qp];
3767 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};
3768 int level[16], run[16];
3769 int suffix_length, zeros_left, coeff_num, coeff_token, total_coeff, i, trailing_ones;
3771 //FIXME put trailing_onex into the context
3773 if(n == CHROMA_DC_BLOCK_INDEX){
3774 coeff_token= get_vlc2(gb, chroma_dc_coeff_token_vlc.table, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 1);
3775 total_coeff= coeff_token>>2;
3777 if(n == LUMA_DC_BLOCK_INDEX){
3778 total_coeff= pred_non_zero_count(h, 0);
3779 coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
3780 total_coeff= coeff_token>>2;
3782 total_coeff= pred_non_zero_count(h, n);
3783 coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
3784 total_coeff= coeff_token>>2;
3785 h->non_zero_count_cache[ scan8[n] ]= total_coeff;
3789 //FIXME set last_non_zero?
3794 trailing_ones= coeff_token&3;
3795 tprintf("trailing:%d, total:%d\n", trailing_ones, total_coeff);
3796 assert(total_coeff<=16);
3798 for(i=0; i<trailing_ones; i++){
3799 level[i]= 1 - 2*get_bits1(gb);
3802 suffix_length= total_coeff > 10 && trailing_ones < 3;
3804 for(; i<total_coeff; i++){
3805 const int prefix= get_level_prefix(gb);
3806 int level_code, mask;
3808 if(prefix<14){ //FIXME try to build a large unified VLC table for all this
3810 level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
3812 level_code= (prefix<<suffix_length); //part
3813 }else if(prefix==14){
3815 level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
3817 level_code= prefix + get_bits(gb, 4); //part
3818 }else if(prefix==15){
3819 level_code= (prefix<<suffix_length) + get_bits(gb, 12); //part
3820 if(suffix_length==0) level_code+=15; //FIXME doesnt make (much)sense
3822 av_log(h->s.avctx, AV_LOG_ERROR, "prefix too large at %d %d\n", s->mb_x, s->mb_y);
3826 if(i==trailing_ones && i<3) level_code+= 2; //FIXME split first iteration
3828 mask= -(level_code&1);
3829 level[i]= (((2+level_code)>>1) ^ mask) - mask;
3831 if(suffix_length==0) suffix_length=1; //FIXME split first iteration
3834 if(ABS(level[i]) > (3<<(suffix_length-1)) && suffix_length<6) suffix_length++;
3836 if((2+level_code)>>1) > (3<<(suffix_length-1)) && suffix_length<6) suffix_length++;
3837 /* ? == prefix > 2 or sth */
3839 tprintf("level: %d suffix_length:%d\n", level[i], suffix_length);
3842 if(total_coeff == max_coeff)
3845 if(n == CHROMA_DC_BLOCK_INDEX)
3846 zeros_left= get_vlc2(gb, chroma_dc_total_zeros_vlc[ total_coeff-1 ].table, CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 1);
3848 zeros_left= get_vlc2(gb, total_zeros_vlc[ total_coeff-1 ].table, TOTAL_ZEROS_VLC_BITS, 1);
3851 for(i=0; i<total_coeff-1; i++){
3854 else if(zeros_left < 7){
3855 run[i]= get_vlc2(gb, run_vlc[zeros_left-1].table, RUN_VLC_BITS, 1);
3857 run[i]= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2);
3859 zeros_left -= run[i];
3863 av_log(h->s.avctx, AV_LOG_ERROR, "negative number of zero coeffs at %d %d\n", s->mb_x, s->mb_y);
3867 for(; i<total_coeff-1; i++){
3875 for(i=total_coeff-1; i>=0; i--){ //FIXME merge into rundecode?
3878 coeff_num += run[i] + 1; //FIXME add 1 earlier ?
3879 j= scantable[ coeff_num ];
3884 for(i=total_coeff-1; i>=0; i--){ //FIXME merge into rundecode?
3887 coeff_num += run[i] + 1; //FIXME add 1 earlier ?
3888 j= scantable[ coeff_num ];
3890 block[j]= level[i] * qmul[j];
3891 // printf("%d %d ", block[j], qmul[j]);
3898 * decodes a P_SKIP or B_SKIP macroblock
3900 static void decode_mb_skip(H264Context *h){
3901 MpegEncContext * const s = &h->s;
3902 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
3905 memset(h->non_zero_count[mb_xy], 0, 16);
3906 memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui
3908 if( h->slice_type == B_TYPE )
3910 // just for fill_caches. pred_direct_motion will set the real mb_type
3911 mb_type= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2|MB_TYPE_SKIP;
3914 fill_caches(h, mb_type, 0); //FIXME check what is needed and what not ...
3915 pred_direct_motion(h, &mb_type);
3917 fill_rectangle(h->mvd_cache[0][scan8[0]], 4, 4, 8, 0, 4);
3918 fill_rectangle(h->mvd_cache[1][scan8[0]], 4, 4, 8, 0, 4);
3924 mb_type= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0|MB_TYPE_SKIP;
3926 if(h->sps.mb_aff && s->mb_skip_run==0 && (s->mb_y&1)==0){
3927 h->mb_field_decoding_flag= get_bits1(&s->gb);
3929 if(h->mb_field_decoding_flag)
3930 mb_type|= MB_TYPE_INTERLACED;
3932 fill_caches(h, mb_type, 0); //FIXME check what is needed and what not ...
3933 pred_pskip_motion(h, &mx, &my);
3934 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
3935 fill_rectangle( h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4);
3937 fill_rectangle(h->mvd_cache[0][scan8[0]], 4, 4, 8, 0, 4);
3940 write_back_motion(h, mb_type);
3941 s->current_picture.mb_type[mb_xy]= mb_type|MB_TYPE_SKIP;
3942 s->current_picture.qscale_table[mb_xy]= s->qscale;
3943 h->slice_table[ mb_xy ]= h->slice_num;
3944 h->prev_mb_skiped= 1;
3948 * decodes a macroblock
3949 * @returns 0 if ok, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
3951 static int decode_mb_cavlc(H264Context *h){
3952 MpegEncContext * const s = &h->s;
3953 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
3954 int mb_type, partition_count, cbp;
3956 s->dsp.clear_blocks(h->mb); //FIXME avoid if allready clear (move after skip handlong?
3958 tprintf("pic:%d mb:%d/%d\n", h->frame_num, s->mb_x, s->mb_y);
3959 cbp = 0; /* avoid warning. FIXME: find a solution without slowing
3961 if(h->slice_type != I_TYPE && h->slice_type != SI_TYPE){
3962 if(s->mb_skip_run==-1)
3963 s->mb_skip_run= get_ue_golomb(&s->gb);
3965 if (s->mb_skip_run--) {
3970 if(h->sps.mb_aff /* && !field pic FIXME needed? */){
3972 h->mb_field_decoding_flag = get_bits1(&s->gb);
3974 h->mb_field_decoding_flag=0; //FIXME som ed note ?!
3976 h->prev_mb_skiped= 0;
3978 mb_type= get_ue_golomb(&s->gb);
3979 if(h->slice_type == B_TYPE){
3981 partition_count= b_mb_type_info[mb_type].partition_count;
3982 mb_type= b_mb_type_info[mb_type].type;
3985 goto decode_intra_mb;
3987 }else if(h->slice_type == P_TYPE /*|| h->slice_type == SP_TYPE */){
3989 partition_count= p_mb_type_info[mb_type].partition_count;
3990 mb_type= p_mb_type_info[mb_type].type;
3993 goto decode_intra_mb;
3996 assert(h->slice_type == I_TYPE);
3999 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);
4003 cbp= i_mb_type_info[mb_type].cbp;
4004 h->intra16x16_pred_mode= i_mb_type_info[mb_type].pred_mode;
4005 mb_type= i_mb_type_info[mb_type].type;
4008 if(h->mb_field_decoding_flag)
4009 mb_type |= MB_TYPE_INTERLACED;
4011 s->current_picture.mb_type[mb_xy]= mb_type;
4012 h->slice_table[ mb_xy ]= h->slice_num;
4014 if(IS_INTRA_PCM(mb_type)){
4017 // we assume these blocks are very rare so we dont optimize it
4018 align_get_bits(&s->gb);
4020 // The pixels are stored in the same order as levels in h->mb array.
4021 for(y=0; y<16; y++){
4022 const int index= 4*(y&3) + 32*((y>>2)&1) + 128*(y>>3);
4023 for(x=0; x<16; x++){
4024 tprintf("LUMA ICPM LEVEL (%3d)\n", show_bits(&s->gb, 8));
4025 h->mb[index + (x&3) + 16*((x>>2)&1) + 64*(x>>3)]= get_bits(&s->gb, 8);
4029 const int index= 256 + 4*(y&3) + 32*(y>>2);
4031 tprintf("CHROMA U ICPM LEVEL (%3d)\n", show_bits(&s->gb, 8));
4032 h->mb[index + (x&3) + 16*(x>>2)]= get_bits(&s->gb, 8);
4036 const int index= 256 + 64 + 4*(y&3) + 32*(y>>2);
4038 tprintf("CHROMA V ICPM LEVEL (%3d)\n", show_bits(&s->gb, 8));
4039 h->mb[index + (x&3) + 16*(x>>2)]= get_bits(&s->gb, 8);
4043 // In deblocking, the quantiser is 0
4044 s->current_picture.qscale_table[mb_xy]= 0;
4045 h->chroma_qp = get_chroma_qp(h, 0);
4046 // All coeffs are presents
4047 memset(h->non_zero_count[mb_xy], 16, 16);
4052 fill_caches(h, mb_type, 0);
4055 if(IS_INTRA(mb_type)){
4056 // init_top_left_availability(h);
4057 if(IS_INTRA4x4(mb_type)){
4060 // fill_intra4x4_pred_table(h);
4061 for(i=0; i<16; i++){
4062 const int mode_coded= !get_bits1(&s->gb);
4063 const int predicted_mode= pred_intra_mode(h, i);
4067 const int rem_mode= get_bits(&s->gb, 3);
4068 if(rem_mode<predicted_mode)
4073 mode= predicted_mode;
4076 h->intra4x4_pred_mode_cache[ scan8[i] ] = mode;
4078 write_back_intra_pred_mode(h);
4079 if( check_intra4x4_pred_mode(h) < 0)
4082 h->intra16x16_pred_mode= check_intra_pred_mode(h, h->intra16x16_pred_mode);
4083 if(h->intra16x16_pred_mode < 0)
4086 h->chroma_pred_mode= get_ue_golomb(&s->gb);
4088 h->chroma_pred_mode= check_intra_pred_mode(h, h->chroma_pred_mode);
4089 if(h->chroma_pred_mode < 0)
4091 }else if(partition_count==4){
4092 int i, j, sub_partition_count[4], list, ref[2][4];
4094 if(h->slice_type == B_TYPE){
4096 h->sub_mb_type[i]= get_ue_golomb(&s->gb);
4097 if(h->sub_mb_type[i] >=13){
4098 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);
4101 sub_partition_count[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
4102 h->sub_mb_type[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].type;
4104 if( IS_DIRECT(h->sub_mb_type[0]) || IS_DIRECT(h->sub_mb_type[1])
4105 || IS_DIRECT(h->sub_mb_type[2]) || IS_DIRECT(h->sub_mb_type[3]))
4106 pred_direct_motion(h, &mb_type);
4108 assert(h->slice_type == P_TYPE || h->slice_type == SP_TYPE); //FIXME SP correct ?
4110 h->sub_mb_type[i]= get_ue_golomb(&s->gb);
4111 if(h->sub_mb_type[i] >=4){
4112 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);
4115 sub_partition_count[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
4116 h->sub_mb_type[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].type;
4120 for(list=0; list<2; list++){
4121 const int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list];
4122 if(ref_count == 0) continue;
4124 if(IS_DIRECT(h->sub_mb_type[i])) continue;
4125 if(IS_DIR(h->sub_mb_type[i], 0, list)){
4126 ref[list][i] = get_te0_golomb(&s->gb, ref_count); //FIXME init to 0 before and skip?
4134 for(list=0; list<2; list++){
4135 const int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list];
4136 if(ref_count == 0) continue;
4139 if(IS_DIRECT(h->sub_mb_type[i])) continue;
4140 h->ref_cache[list][ scan8[4*i] ]=h->ref_cache[list][ scan8[4*i]+1 ]=
4141 h->ref_cache[list][ scan8[4*i]+8 ]=h->ref_cache[list][ scan8[4*i]+9 ]= ref[list][i];
4143 if(IS_DIR(h->sub_mb_type[i], 0, list)){
4144 const int sub_mb_type= h->sub_mb_type[i];
4145 const int block_width= (sub_mb_type & (MB_TYPE_16x16|MB_TYPE_16x8)) ? 2 : 1;
4146 for(j=0; j<sub_partition_count[i]; j++){
4148 const int index= 4*i + block_width*j;
4149 int16_t (* mv_cache)[2]= &h->mv_cache[list][ scan8[index] ];
4150 pred_motion(h, index, block_width, list, h->ref_cache[list][ scan8[index] ], &mx, &my);
4151 mx += get_se_golomb(&s->gb);
4152 my += get_se_golomb(&s->gb);
4153 tprintf("final mv:%d %d\n", mx, my);
4155 if(IS_SUB_8X8(sub_mb_type)){
4156 mv_cache[ 0 ][0]= mv_cache[ 1 ][0]=
4157 mv_cache[ 8 ][0]= mv_cache[ 9 ][0]= mx;
4158 mv_cache[ 0 ][1]= mv_cache[ 1 ][1]=
4159 mv_cache[ 8 ][1]= mv_cache[ 9 ][1]= my;
4160 }else if(IS_SUB_8X4(sub_mb_type)){
4161 mv_cache[ 0 ][0]= mv_cache[ 1 ][0]= mx;
4162 mv_cache[ 0 ][1]= mv_cache[ 1 ][1]= my;
4163 }else if(IS_SUB_4X8(sub_mb_type)){
4164 mv_cache[ 0 ][0]= mv_cache[ 8 ][0]= mx;
4165 mv_cache[ 0 ][1]= mv_cache[ 8 ][1]= my;
4167 assert(IS_SUB_4X4(sub_mb_type));
4168 mv_cache[ 0 ][0]= mx;
4169 mv_cache[ 0 ][1]= my;
4173 uint32_t *p= (uint32_t *)&h->mv_cache[list][ scan8[4*i] ][0];
4179 }else if(IS_DIRECT(mb_type)){
4180 pred_direct_motion(h, &mb_type);
4181 s->current_picture.mb_type[mb_xy]= mb_type;
4183 int list, mx, my, i;
4184 //FIXME we should set ref_idx_l? to 0 if we use that later ...
4185 if(IS_16X16(mb_type)){
4186 for(list=0; list<2; list++){
4187 if(h->ref_count[list]>0){
4188 if(IS_DIR(mb_type, 0, list)){
4189 const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
4190 fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, val, 1);
4192 fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, (LIST_NOT_USED&0xFF), 1);
4195 for(list=0; list<2; list++){
4196 if(IS_DIR(mb_type, 0, list)){
4197 pred_motion(h, 0, 4, list, h->ref_cache[list][ scan8[0] ], &mx, &my);
4198 mx += get_se_golomb(&s->gb);
4199 my += get_se_golomb(&s->gb);
4200 tprintf("final mv:%d %d\n", mx, my);
4202 fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, pack16to32(mx,my), 4);
4204 fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, 0, 4);
4207 else if(IS_16X8(mb_type)){
4208 for(list=0; list<2; list++){
4209 if(h->ref_count[list]>0){
4211 if(IS_DIR(mb_type, i, list)){
4212 const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
4213 fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, val, 1);
4215 fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, (LIST_NOT_USED&0xFF), 1);
4219 for(list=0; list<2; list++){
4221 if(IS_DIR(mb_type, i, list)){
4222 pred_16x8_motion(h, 8*i, list, h->ref_cache[list][scan8[0] + 16*i], &mx, &my);
4223 mx += get_se_golomb(&s->gb);
4224 my += get_se_golomb(&s->gb);
4225 tprintf("final mv:%d %d\n", mx, my);
4227 fill_rectangle(h->mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, pack16to32(mx,my), 4);
4229 fill_rectangle(h->mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, 0, 4);
4233 assert(IS_8X16(mb_type));
4234 for(list=0; list<2; list++){
4235 if(h->ref_count[list]>0){
4237 if(IS_DIR(mb_type, i, list)){ //FIXME optimize
4238 const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
4239 fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, val, 1);
4241 fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, (LIST_NOT_USED&0xFF), 1);
4245 for(list=0; list<2; list++){
4247 if(IS_DIR(mb_type, i, list)){
4248 pred_8x16_motion(h, i*4, list, h->ref_cache[list][ scan8[0] + 2*i ], &mx, &my);
4249 mx += get_se_golomb(&s->gb);
4250 my += get_se_golomb(&s->gb);
4251 tprintf("final mv:%d %d\n", mx, my);
4253 fill_rectangle(h->mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, pack16to32(mx,my), 4);
4255 fill_rectangle(h->mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, 0, 4);
4261 if(IS_INTER(mb_type))
4262 write_back_motion(h, mb_type);
4264 if(!IS_INTRA16x16(mb_type)){
4265 cbp= get_ue_golomb(&s->gb);
4267 av_log(h->s.avctx, AV_LOG_ERROR, "cbp too large (%d) at %d %d\n", cbp, s->mb_x, s->mb_y);
4271 if(IS_INTRA4x4(mb_type))
4272 cbp= golomb_to_intra4x4_cbp[cbp];
4274 cbp= golomb_to_inter_cbp[cbp];
4277 if(cbp || IS_INTRA16x16(mb_type)){
4278 int i8x8, i4x4, chroma_idx;
4279 int chroma_qp, dquant;
4280 GetBitContext *gb= IS_INTRA(mb_type) ? h->intra_gb_ptr : h->inter_gb_ptr;
4281 const uint8_t *scan, *dc_scan;
4283 // fill_non_zero_count_cache(h);
4285 if(IS_INTERLACED(mb_type)){
4287 dc_scan= luma_dc_field_scan;
4290 dc_scan= luma_dc_zigzag_scan;
4293 dquant= get_se_golomb(&s->gb);
4295 if( dquant > 25 || dquant < -26 ){
4296 av_log(h->s.avctx, AV_LOG_ERROR, "dquant out of range (%d) at %d %d\n", dquant, s->mb_x, s->mb_y);
4300 s->qscale += dquant;
4301 if(((unsigned)s->qscale) > 51){
4302 if(s->qscale<0) s->qscale+= 52;
4303 else s->qscale-= 52;
4306 h->chroma_qp= chroma_qp= get_chroma_qp(h, s->qscale);
4307 if(IS_INTRA16x16(mb_type)){
4308 if( decode_residual(h, h->intra_gb_ptr, h->mb, LUMA_DC_BLOCK_INDEX, dc_scan, s->qscale, 16) < 0){
4309 return -1; //FIXME continue if partotioned and other retirn -1 too
4312 assert((cbp&15) == 0 || (cbp&15) == 15);
4315 for(i8x8=0; i8x8<4; i8x8++){
4316 for(i4x4=0; i4x4<4; i4x4++){
4317 const int index= i4x4 + 4*i8x8;
4318 if( decode_residual(h, h->intra_gb_ptr, h->mb + 16*index, index, scan + 1, s->qscale, 15) < 0 ){
4324 fill_rectangle(&h->non_zero_count_cache[scan8[0]], 4, 4, 8, 0, 1);
4327 for(i8x8=0; i8x8<4; i8x8++){
4328 if(cbp & (1<<i8x8)){
4329 for(i4x4=0; i4x4<4; i4x4++){
4330 const int index= i4x4 + 4*i8x8;
4332 if( decode_residual(h, gb, h->mb + 16*index, index, scan, s->qscale, 16) <0 ){
4337 uint8_t * const nnz= &h->non_zero_count_cache[ scan8[4*i8x8] ];
4338 nnz[0] = nnz[1] = nnz[8] = nnz[9] = 0;
4344 for(chroma_idx=0; chroma_idx<2; chroma_idx++)
4345 if( decode_residual(h, gb, h->mb + 256 + 16*4*chroma_idx, CHROMA_DC_BLOCK_INDEX, chroma_dc_scan, chroma_qp, 4) < 0){
4351 for(chroma_idx=0; chroma_idx<2; chroma_idx++){
4352 for(i4x4=0; i4x4<4; i4x4++){
4353 const int index= 16 + 4*chroma_idx + i4x4;
4354 if( decode_residual(h, gb, h->mb + 16*index, index, scan + 1, chroma_qp, 15) < 0){
4360 uint8_t * const nnz= &h->non_zero_count_cache[0];
4361 nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =
4362 nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
4365 uint8_t * const nnz= &h->non_zero_count_cache[0];
4366 fill_rectangle(&nnz[scan8[0]], 4, 4, 8, 0, 1);
4367 nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =
4368 nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
4370 s->current_picture.qscale_table[mb_xy]= s->qscale;
4371 write_back_non_zero_count(h);
4376 static int decode_cabac_intra_mb_type(H264Context *h, int ctx_base, int intra_slice) {
4377 uint8_t *state= &h->cabac_state[ctx_base];
4381 MpegEncContext * const s = &h->s;
4382 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
4383 const int mba_xy = mb_xy - 1;
4384 const int mbb_xy = mb_xy - s->mb_stride;
4386 if( h->slice_table[mba_xy] == h->slice_num && !IS_INTRA4x4( s->current_picture.mb_type[mba_xy] ) )
4388 if( h->slice_table[mbb_xy] == h->slice_num && !IS_INTRA4x4( s->current_picture.mb_type[mbb_xy] ) )
4390 if( get_cabac( &h->cabac, &state[ctx] ) == 0 )
4391 return 0; /* I4x4 */
4394 if( get_cabac( &h->cabac, &state[0] ) == 0 )
4395 return 0; /* I4x4 */
4398 if( get_cabac_terminate( &h->cabac ) )
4399 return 25; /* PCM */
4401 mb_type = 1; /* I16x16 */
4402 if( get_cabac( &h->cabac, &state[1] ) )
4403 mb_type += 12; /* cbp_luma != 0 */
4405 if( get_cabac( &h->cabac, &state[2] ) ) {
4406 if( get_cabac( &h->cabac, &state[2+intra_slice] ) )
4407 mb_type += 4 * 2; /* cbp_chroma == 2 */
4409 mb_type += 4 * 1; /* cbp_chroma == 1 */
4411 if( get_cabac( &h->cabac, &state[3+intra_slice] ) )
4413 if( get_cabac( &h->cabac, &state[3+2*intra_slice] ) )
4418 static int decode_cabac_mb_type( H264Context *h ) {
4419 MpegEncContext * const s = &h->s;
4421 if( h->slice_type == I_TYPE ) {
4422 return decode_cabac_intra_mb_type(h, 3, 1);
4423 } else if( h->slice_type == P_TYPE ) {
4424 if( get_cabac( &h->cabac, &h->cabac_state[14] ) == 0 ) {
4426 if( get_cabac( &h->cabac, &h->cabac_state[15] ) == 0 ) {
4427 if( get_cabac( &h->cabac, &h->cabac_state[16] ) == 0 )
4428 return 0; /* P_L0_D16x16; */
4430 return 3; /* P_8x8; */
4432 if( get_cabac( &h->cabac, &h->cabac_state[17] ) == 0 )
4433 return 2; /* P_L0_D8x16; */
4435 return 1; /* P_L0_D16x8; */
4438 return decode_cabac_intra_mb_type(h, 17, 0) + 5;
4440 } else if( h->slice_type == B_TYPE ) {
4441 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
4442 const int mba_xy = mb_xy - 1;
4443 const int mbb_xy = mb_xy - s->mb_stride;
4447 if( h->slice_table[mba_xy] == h->slice_num && !IS_SKIP( s->current_picture.mb_type[mba_xy] )
4448 && !IS_DIRECT( s->current_picture.mb_type[mba_xy] ) )
4450 if( h->slice_table[mbb_xy] == h->slice_num && !IS_SKIP( s->current_picture.mb_type[mbb_xy] )
4451 && !IS_DIRECT( s->current_picture.mb_type[mbb_xy] ) )
4454 if( !get_cabac( &h->cabac, &h->cabac_state[27+ctx] ) )
4455 return 0; /* B_Direct_16x16 */
4457 if( !get_cabac( &h->cabac, &h->cabac_state[27+3] ) ) {
4458 return 1 + get_cabac( &h->cabac, &h->cabac_state[27+5] ); /* B_L[01]_16x16 */
4461 bits = get_cabac( &h->cabac, &h->cabac_state[27+4] ) << 3;
4462 bits|= get_cabac( &h->cabac, &h->cabac_state[27+5] ) << 2;
4463 bits|= get_cabac( &h->cabac, &h->cabac_state[27+5] ) << 1;
4464 bits|= get_cabac( &h->cabac, &h->cabac_state[27+5] );
4466 return bits + 3; /* B_Bi_16x16 through B_L1_L0_16x8 */
4467 else if( bits == 13 ) {
4468 return decode_cabac_intra_mb_type(h, 32, 0) + 23;
4469 } else if( bits == 14 )
4470 return 11; /* B_L1_L0_8x16 */
4471 else if( bits == 15 )
4472 return 22; /* B_8x8 */
4474 bits= ( bits<<1 ) | get_cabac( &h->cabac, &h->cabac_state[27+5] );
4475 return bits - 4; /* B_L0_Bi_* through B_Bi_Bi_* */
4477 /* TODO SI/SP frames? */
4482 static int decode_cabac_mb_skip( H264Context *h) {
4483 MpegEncContext * const s = &h->s;
4484 const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
4485 const int mba_xy = mb_xy - 1;
4486 const int mbb_xy = mb_xy - s->mb_stride;
4489 if( h->slice_table[mba_xy] == h->slice_num && !IS_SKIP( s->current_picture.mb_type[mba_xy] ))
4491 if( h->slice_table[mbb_xy] == h->slice_num && !IS_SKIP( s->current_picture.mb_type[mbb_xy] ))
4494 if( h->slice_type == P_TYPE || h->slice_type == SP_TYPE)
4495 return get_cabac( &h->cabac, &h->cabac_state[11+ctx] );
4497 return get_cabac( &h->cabac, &h->cabac_state[24+ctx] );
4500 static int decode_cabac_mb_intra4x4_pred_mode( H264Context *h, int pred_mode ) {
4503 if( get_cabac( &h->cabac, &h->cabac_state[68] ) )
4506 if( get_cabac( &h->cabac, &h->cabac_state[69] ) )
4508 if( get_cabac( &h->cabac, &h->cabac_state[69] ) )
4510 if( get_cabac( &h->cabac, &h->cabac_state[69] ) )
4512 if( mode >= pred_mode )
4518 static int decode_cabac_mb_chroma_pre_mode( H264Context *h) {
4519 MpegEncContext * const s = &h->s;
4520 const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
4521 const int mba_xy = mb_xy - 1;
4522 const int mbb_xy = mb_xy - s->mb_stride;
4526 /* No need to test for IS_INTRA4x4 and IS_INTRA16x16, as we set chroma_pred_mode_table to 0 */
4527 if( h->slice_table[mba_xy] == h->slice_num && h->chroma_pred_mode_table[mba_xy] != 0 )
4530 if( h->slice_table[mbb_xy] == h->slice_num && h->chroma_pred_mode_table[mbb_xy] != 0 )
4533 if( get_cabac( &h->cabac, &h->cabac_state[64+ctx] ) == 0 )
4536 if( get_cabac( &h->cabac, &h->cabac_state[64+3] ) == 0 )
4538 if( get_cabac( &h->cabac, &h->cabac_state[64+3] ) == 0 )
4544 static const uint8_t block_idx_x[16] = {
4545 0, 1, 0, 1, 2, 3, 2, 3, 0, 1, 0, 1, 2, 3, 2, 3
4547 static const uint8_t block_idx_y[16] = {
4548 0, 0, 1, 1, 0, 0, 1, 1, 2, 2, 3, 3, 2, 2, 3, 3
4550 static const uint8_t block_idx_xy[4][4] = {
4557 static int decode_cabac_mb_cbp_luma( H264Context *h) {
4558 MpegEncContext * const s = &h->s;
4559 const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
4564 h->cbp_table[mb_xy] = 0; /* FIXME aaahahahah beurk */
4566 for( i8x8 = 0; i8x8 < 4; i8x8++ ) {
4572 x = block_idx_x[4*i8x8];
4573 y = block_idx_y[4*i8x8];
4577 else if( s->mb_x > 0 ) {
4579 if (h->slice_table[mba_xy] != h->slice_num) {
4586 else if( s->mb_y > 0 ) {
4587 mbb_xy = mb_xy - s->mb_stride;
4588 if (h->slice_table[mbb_xy] != h->slice_num) {
4593 /* No need to test for skip as we put 0 for skip block */
4595 int i8x8a = block_idx_xy[(x-1)&0x03][y]/4;
4596 if( ((h->cbp_table[mba_xy] >> i8x8a)&0x01) == 0 )
4601 int i8x8b = block_idx_xy[x][(y-1)&0x03]/4;
4602 if( ((h->cbp_table[mbb_xy] >> i8x8b)&0x01) == 0 )
4606 if( get_cabac( &h->cabac, &h->cabac_state[73 + ctx] ) ) {
4608 h->cbp_table[mb_xy] = cbp; /* FIXME aaahahahah beurk */
4613 static int decode_cabac_mb_cbp_chroma( H264Context *h) {
4617 cbp_a = (h->left_cbp>>4)&0x03;
4618 cbp_b = (h-> top_cbp>>4)&0x03;
4621 if( cbp_a > 0 ) ctx++;
4622 if( cbp_b > 0 ) ctx += 2;
4623 if( get_cabac( &h->cabac, &h->cabac_state[77 + ctx] ) == 0 )
4627 if( cbp_a == 2 ) ctx++;
4628 if( cbp_b == 2 ) ctx += 2;
4629 return 1 + get_cabac( &h->cabac, &h->cabac_state[77 + ctx] );
4631 static int decode_cabac_mb_dqp( H264Context *h) {
4632 MpegEncContext * const s = &h->s;
4638 mbn_xy = s->mb_x + s->mb_y*s->mb_stride - 1;
4640 mbn_xy = s->mb_width - 1 + (s->mb_y-1)*s->mb_stride;
4642 if( h->last_qscale_diff != 0 && ( IS_INTRA16x16(s->current_picture.mb_type[mbn_xy] ) || (h->cbp_table[mbn_xy]&0x3f) ) )
4645 while( get_cabac( &h->cabac, &h->cabac_state[60 + ctx] ) ) {
4656 return -(val + 1)/2;
4658 static int decode_cabac_p_mb_sub_type( H264Context *h ) {
4659 if( get_cabac( &h->cabac, &h->cabac_state[21] ) )
4661 if( !get_cabac( &h->cabac, &h->cabac_state[22] ) )
4663 if( get_cabac( &h->cabac, &h->cabac_state[23] ) )
4667 static int decode_cabac_b_mb_sub_type( H264Context *h ) {
4669 if( !get_cabac( &h->cabac, &h->cabac_state[36] ) )
4670 return 0; /* B_Direct_8x8 */
4671 if( !get_cabac( &h->cabac, &h->cabac_state[37] ) )
4672 return 1 + get_cabac( &h->cabac, &h->cabac_state[39] ); /* B_L0_8x8, B_L1_8x8 */
4674 if( get_cabac( &h->cabac, &h->cabac_state[38] ) ) {
4675 if( get_cabac( &h->cabac, &h->cabac_state[39] ) )
4676 return 11 + get_cabac( &h->cabac, &h->cabac_state[39] ); /* B_L1_4x4, B_Bi_4x4 */
4679 type += 2*get_cabac( &h->cabac, &h->cabac_state[39] );
4680 type += get_cabac( &h->cabac, &h->cabac_state[39] );
4684 static int decode_cabac_mb_ref( H264Context *h, int list, int n ) {
4685 int refa = h->ref_cache[list][scan8[n] - 1];
4686 int refb = h->ref_cache[list][scan8[n] - 8];
4690 if( h->slice_type == B_TYPE) {
4691 if( refa > 0 && !h->direct_cache[scan8[n] - 1] )
4693 if( refb > 0 && !h->direct_cache[scan8[n] - 8] )
4702 while( get_cabac( &h->cabac, &h->cabac_state[54+ctx] ) ) {
4712 static int decode_cabac_mb_mvd( H264Context *h, int list, int n, int l ) {
4713 int amvd = abs( h->mvd_cache[list][scan8[n] - 1][l] ) +
4714 abs( h->mvd_cache[list][scan8[n] - 8][l] );
4715 int ctxbase = (l == 0) ? 40 : 47;
4720 else if( amvd > 32 )
4725 if(!get_cabac(&h->cabac, &h->cabac_state[ctxbase+ctx]))
4730 while( mvd < 9 && get_cabac( &h->cabac, &h->cabac_state[ctxbase+ctx] ) ) {
4738 while( get_cabac_bypass( &h->cabac ) ) {
4743 if( get_cabac_bypass( &h->cabac ) )
4747 if( get_cabac_bypass( &h->cabac ) ) return -mvd;
4751 static int inline get_cabac_cbf_ctx( H264Context *h, int cat, int idx ) {
4756 nza = h->left_cbp&0x100;
4757 nzb = h-> top_cbp&0x100;
4758 } else if( cat == 1 || cat == 2 ) {
4759 nza = h->non_zero_count_cache[scan8[idx] - 1];
4760 nzb = h->non_zero_count_cache[scan8[idx] - 8];
4761 } else if( cat == 3 ) {
4762 nza = (h->left_cbp>>(6+idx))&0x01;
4763 nzb = (h-> top_cbp>>(6+idx))&0x01;
4766 nza = h->non_zero_count_cache[scan8[16+idx] - 1];
4767 nzb = h->non_zero_count_cache[scan8[16+idx] - 8];
4776 return ctx + 4 * cat;
4779 static int inline decode_cabac_residual( H264Context *h, DCTELEM *block, int cat, int n, const uint8_t *scantable, int qp, int max_coeff) {
4780 const int mb_xy = h->s.mb_x + h->s.mb_y*h->s.mb_stride;
4781 const uint16_t *qmul= dequant_coeff[qp];
4782 static const int significant_coeff_flag_offset[5] = { 0, 15, 29, 44, 47 };
4783 static const int coeff_abs_level_m1_offset[5] = {227+ 0, 227+10, 227+20, 227+30, 227+39 };
4788 int coeff_count = 0;
4791 int abslevelgt1 = 0;
4793 /* cat: 0-> DC 16x16 n = 0
4794 * 1-> AC 16x16 n = luma4x4idx
4795 * 2-> Luma4x4 n = luma4x4idx
4796 * 3-> DC Chroma n = iCbCr
4797 * 4-> AC Chroma n = 4 * iCbCr + chroma4x4idx
4800 /* read coded block flag */
4801 if( get_cabac( &h->cabac, &h->cabac_state[85 + get_cabac_cbf_ctx( h, cat, n ) ] ) == 0 ) {
4802 if( cat == 1 || cat == 2 )
4803 h->non_zero_count_cache[scan8[n]] = 0;
4805 h->non_zero_count_cache[scan8[16+n]] = 0;
4810 for(last= 0; last < max_coeff - 1; last++) {
4811 if( get_cabac( &h->cabac, &h->cabac_state[105+significant_coeff_flag_offset[cat]+last] )) {
4812 index[coeff_count++] = last;
4813 if( get_cabac( &h->cabac, &h->cabac_state[166+significant_coeff_flag_offset[cat]+last] ) ) {
4819 if( last == max_coeff -1 ) {
4820 index[coeff_count++] = last;
4822 assert(coeff_count > 0);
4825 h->cbp_table[mb_xy] |= 0x100;
4826 else if( cat == 1 || cat == 2 )
4827 h->non_zero_count_cache[scan8[n]] = coeff_count;
4829 h->cbp_table[mb_xy] |= 0x40 << n;
4832 h->non_zero_count_cache[scan8[16+n]] = coeff_count;
4835 for( i = coeff_count - 1; i >= 0; i-- ) {
4836 int ctx = (abslevelgt1 != 0 ? 0 : FFMIN( 4, abslevel1 )) + coeff_abs_level_m1_offset[cat];
4837 int j= scantable[index[i]];
4839 if( get_cabac( &h->cabac, &h->cabac_state[ctx] ) == 0 ) {
4840 if( cat == 0 || cat == 3 ) {
4841 if( get_cabac_bypass( &h->cabac ) ) block[j] = -1;
4844 if( get_cabac_bypass( &h->cabac ) ) block[j] = -qmul[j];
4845 else block[j] = qmul[j];
4851 ctx = 5 + FFMIN( 4, abslevelgt1 ) + coeff_abs_level_m1_offset[cat];
4852 while( coeff_abs < 15 && get_cabac( &h->cabac, &h->cabac_state[ctx] ) ) {
4856 if( coeff_abs >= 15 ) {
4858 while( get_cabac_bypass( &h->cabac ) ) {
4859 coeff_abs += 1 << j;
4864 if( get_cabac_bypass( &h->cabac ) )
4865 coeff_abs += 1 << j ;
4869 if( cat == 0 || cat == 3 ) {
4870 if( get_cabac_bypass( &h->cabac ) ) block[j] = -coeff_abs;
4871 else block[j] = coeff_abs;
4873 if( get_cabac_bypass( &h->cabac ) ) block[j] = -coeff_abs * qmul[j];
4874 else block[j] = coeff_abs * qmul[j];
4884 * decodes a macroblock
4885 * @returns 0 if ok, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
4887 static int decode_mb_cabac(H264Context *h) {
4888 MpegEncContext * const s = &h->s;
4889 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
4890 int mb_type, partition_count, cbp = 0;
4892 s->dsp.clear_blocks(h->mb); //FIXME avoid if allready clear (move after skip handlong?)
4894 if( h->sps.mb_aff ) {
4895 av_log( h->s.avctx, AV_LOG_ERROR, "Fields not supported with CABAC\n" );
4899 tprintf("pic:%d mb:%d/%d\n", h->frame_num, s->mb_x, s->mb_y);
4900 if( h->slice_type != I_TYPE && h->slice_type != SI_TYPE ) {
4901 /* read skip flags */
4902 if( decode_cabac_mb_skip( h ) ) {
4905 h->cbp_table[mb_xy] = 0;
4906 h->chroma_pred_mode_table[mb_xy] = 0;
4907 h->last_qscale_diff = 0;
4913 h->prev_mb_skiped = 0;
4915 if( ( mb_type = decode_cabac_mb_type( h ) ) < 0 ) {
4916 av_log( h->s.avctx, AV_LOG_ERROR, "decode_cabac_mb_type failed\n" );
4920 if( h->slice_type == B_TYPE ) {
4922 partition_count= b_mb_type_info[mb_type].partition_count;
4923 mb_type= b_mb_type_info[mb_type].type;
4926 goto decode_intra_mb;
4928 } else if( h->slice_type == P_TYPE ) {
4930 partition_count= p_mb_type_info[mb_type].partition_count;
4931 mb_type= p_mb_type_info[mb_type].type;
4934 goto decode_intra_mb;
4937 assert(h->slice_type == I_TYPE);
4939 partition_count = 0;
4940 cbp= i_mb_type_info[mb_type].cbp;
4941 h->intra16x16_pred_mode= i_mb_type_info[mb_type].pred_mode;
4942 mb_type= i_mb_type_info[mb_type].type;
4945 if(h->mb_field_decoding_flag)
4946 mb_type |= MB_TYPE_INTERLACED;
4949 s->current_picture.mb_type[mb_xy]= mb_type;
4950 h->slice_table[ mb_xy ]= h->slice_num;
4952 if(IS_INTRA_PCM(mb_type)) {
4956 // We assume these blocks are very rare so we dont optimize it.
4957 // FIXME The two following lines get the bitstream position in the cabac
4958 // decode, I think it should be done by a function in cabac.h (or cabac.c).
4959 ptr= h->cabac.bytestream;
4960 if (h->cabac.low&0x1) ptr-=CABAC_BITS/8;
4962 // The pixels are stored in the same order as levels in h->mb array.
4963 for(y=0; y<16; y++){
4964 const int index= 4*(y&3) + 32*((y>>2)&1) + 128*(y>>3);
4965 for(x=0; x<16; x++){
4966 tprintf("LUMA ICPM LEVEL (%3d)\n", *ptr);
4967 h->mb[index + (x&3) + 16*((x>>2)&1) + 64*(x>>3)]= *ptr++;
4971 const int index= 256 + 4*(y&3) + 32*(y>>2);
4973 tprintf("CHROMA U ICPM LEVEL (%3d)\n", *ptr);
4974 h->mb[index + (x&3) + 16*(x>>2)]= *ptr++;
4978 const int index= 256 + 64 + 4*(y&3) + 32*(y>>2);
4980 tprintf("CHROMA V ICPM LEVEL (%3d)\n", *ptr);
4981 h->mb[index + (x&3) + 16*(x>>2)]= *ptr++;
4985 ff_init_cabac_decoder(&h->cabac, ptr, h->cabac.bytestream_end - ptr);
4987 // All blocks are presents
4988 h->cbp_table[mb_xy] = 0x1ef;
4989 h->chroma_pred_mode_table[mb_xy] = 0;
4990 // In deblocking, the quantiser is 0
4991 s->current_picture.qscale_table[mb_xy]= 0;
4992 h->chroma_qp = get_chroma_qp(h, 0);
4993 // All coeffs are presents
4994 memset(h->non_zero_count[mb_xy], 16, 16);
4998 fill_caches(h, mb_type, 0);
5000 if( IS_INTRA( mb_type ) ) {
5001 if( IS_INTRA4x4( mb_type ) ) {
5003 for( i = 0; i < 16; i++ ) {
5004 int pred = pred_intra_mode( h, i );
5005 h->intra4x4_pred_mode_cache[ scan8[i] ] = decode_cabac_mb_intra4x4_pred_mode( h, pred );
5007 //av_log( s->avctx, AV_LOG_ERROR, "i4x4 pred=%d mode=%d\n", pred, h->intra4x4_pred_mode_cache[ scan8[i] ] );
5009 write_back_intra_pred_mode(h);
5010 if( check_intra4x4_pred_mode(h) < 0 ) return -1;
5012 h->intra16x16_pred_mode= check_intra_pred_mode( h, h->intra16x16_pred_mode );
5013 if( h->intra16x16_pred_mode < 0 ) return -1;
5015 h->chroma_pred_mode_table[mb_xy] =
5016 h->chroma_pred_mode = decode_cabac_mb_chroma_pre_mode( h );
5018 h->chroma_pred_mode= check_intra_pred_mode( h, h->chroma_pred_mode );
5019 if( h->chroma_pred_mode < 0 ) return -1;
5020 } else if( partition_count == 4 ) {
5021 int i, j, sub_partition_count[4], list, ref[2][4];
5023 if( h->slice_type == B_TYPE ) {
5024 for( i = 0; i < 4; i++ ) {
5025 h->sub_mb_type[i] = decode_cabac_b_mb_sub_type( h );
5026 sub_partition_count[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
5027 h->sub_mb_type[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].type;
5029 if( IS_DIRECT(h->sub_mb_type[0]) || IS_DIRECT(h->sub_mb_type[1])
5030 || IS_DIRECT(h->sub_mb_type[2]) || IS_DIRECT(h->sub_mb_type[3])) {
5031 pred_direct_motion(h, &mb_type);
5032 if( h->ref_count[0] > 1 || h->ref_count[1] > 1 ) {
5033 for( i = 0; i < 4; i++ )
5034 if( IS_DIRECT(h->sub_mb_type[i]) )
5035 fill_rectangle( &h->direct_cache[scan8[4*i]], 2, 2, 8, 1, 1 );
5039 for( i = 0; i < 4; i++ ) {
5040 h->sub_mb_type[i] = decode_cabac_p_mb_sub_type( h );
5041 sub_partition_count[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
5042 h->sub_mb_type[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].type;
5046 for( list = 0; list < 2; list++ ) {
5047 if( h->ref_count[list] > 0 ) {
5048 for( i = 0; i < 4; i++ ) {
5049 if(IS_DIRECT(h->sub_mb_type[i])) continue;
5050 if(IS_DIR(h->sub_mb_type[i], 0, list)){
5051 if( h->ref_count[list] > 1 )
5052 ref[list][i] = decode_cabac_mb_ref( h, list, 4*i );
5058 h->ref_cache[list][ scan8[4*i]+1 ]=
5059 h->ref_cache[list][ scan8[4*i]+8 ]=h->ref_cache[list][ scan8[4*i]+9 ]= ref[list][i];
5064 for(list=0; list<2; list++){
5066 if(IS_DIRECT(h->sub_mb_type[i])){
5067 fill_rectangle(h->mvd_cache[list][scan8[4*i]], 2, 2, 8, 0, 4);
5070 h->ref_cache[list][ scan8[4*i] ]=h->ref_cache[list][ scan8[4*i]+1 ];
5072 if(IS_DIR(h->sub_mb_type[i], 0, list) && !IS_DIRECT(h->sub_mb_type[i])){
5073 const int sub_mb_type= h->sub_mb_type[i];
5074 const int block_width= (sub_mb_type & (MB_TYPE_16x16|MB_TYPE_16x8)) ? 2 : 1;
5075 for(j=0; j<sub_partition_count[i]; j++){
5078 const int index= 4*i + block_width*j;
5079 int16_t (* mv_cache)[2]= &h->mv_cache[list][ scan8[index] ];
5080 int16_t (* mvd_cache)[2]= &h->mvd_cache[list][ scan8[index] ];
5081 pred_motion(h, index, block_width, list, h->ref_cache[list][ scan8[index] ], &mpx, &mpy);
5083 mx = mpx + decode_cabac_mb_mvd( h, list, index, 0 );
5084 my = mpy + decode_cabac_mb_mvd( h, list, index, 1 );
5085 tprintf("final mv:%d %d\n", mx, my);
5087 if(IS_SUB_8X8(sub_mb_type)){
5088 mv_cache[ 0 ][0]= mv_cache[ 1 ][0]=
5089 mv_cache[ 8 ][0]= mv_cache[ 9 ][0]= mx;
5090 mv_cache[ 0 ][1]= mv_cache[ 1 ][1]=
5091 mv_cache[ 8 ][1]= mv_cache[ 9 ][1]= my;
5093 mvd_cache[ 0 ][0]= mvd_cache[ 1 ][0]=
5094 mvd_cache[ 8 ][0]= mvd_cache[ 9 ][0]= mx - mpx;
5095 mvd_cache[ 0 ][1]= mvd_cache[ 1 ][1]=
5096 mvd_cache[ 8 ][1]= mvd_cache[ 9 ][1]= my - mpy;
5097 }else if(IS_SUB_8X4(sub_mb_type)){
5098 mv_cache[ 0 ][0]= mv_cache[ 1 ][0]= mx;
5099 mv_cache[ 0 ][1]= mv_cache[ 1 ][1]= my;
5101 mvd_cache[ 0 ][0]= mvd_cache[ 1 ][0]= mx- mpx;
5102 mvd_cache[ 0 ][1]= mvd_cache[ 1 ][1]= my - mpy;
5103 }else if(IS_SUB_4X8(sub_mb_type)){
5104 mv_cache[ 0 ][0]= mv_cache[ 8 ][0]= mx;
5105 mv_cache[ 0 ][1]= mv_cache[ 8 ][1]= my;
5107 mvd_cache[ 0 ][0]= mvd_cache[ 8 ][0]= mx - mpx;
5108 mvd_cache[ 0 ][1]= mvd_cache[ 8 ][1]= my - mpy;
5110 assert(IS_SUB_4X4(sub_mb_type));
5111 mv_cache[ 0 ][0]= mx;
5112 mv_cache[ 0 ][1]= my;
5114 mvd_cache[ 0 ][0]= mx - mpx;
5115 mvd_cache[ 0 ][1]= my - mpy;
5119 uint32_t *p= (uint32_t *)&h->mv_cache[list][ scan8[4*i] ][0];
5120 uint32_t *pd= (uint32_t *)&h->mvd_cache[list][ scan8[4*i] ][0];
5121 p[0] = p[1] = p[8] = p[9] = 0;
5122 pd[0]= pd[1]= pd[8]= pd[9]= 0;
5126 } else if( IS_DIRECT(mb_type) ) {
5127 pred_direct_motion(h, &mb_type);
5128 s->current_picture.mb_type[mb_xy]= mb_type;
5129 fill_rectangle(h->mvd_cache[0][scan8[0]], 4, 4, 8, 0, 4);
5130 fill_rectangle(h->mvd_cache[1][scan8[0]], 4, 4, 8, 0, 4);
5132 int list, mx, my, i, mpx, mpy;
5133 if(IS_16X16(mb_type)){
5134 for(list=0; list<2; list++){
5135 if(IS_DIR(mb_type, 0, list)){
5136 if(h->ref_count[list] > 0 ){
5137 const int ref = h->ref_count[list] > 1 ? decode_cabac_mb_ref( h, list, 0 ) : 0;
5138 fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, ref, 1);
5141 fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, (uint8_t)LIST_NOT_USED, 1);
5143 for(list=0; list<2; list++){
5144 if(IS_DIR(mb_type, 0, list)){
5145 pred_motion(h, 0, 4, list, h->ref_cache[list][ scan8[0] ], &mpx, &mpy);
5147 mx = mpx + decode_cabac_mb_mvd( h, list, 0, 0 );
5148 my = mpy + decode_cabac_mb_mvd( h, list, 0, 1 );
5149 tprintf("final mv:%d %d\n", mx, my);
5151 fill_rectangle(h->mvd_cache[list][ scan8[0] ], 4, 4, 8, pack16to32(mx-mpx,my-mpy), 4);
5152 fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, pack16to32(mx,my), 4);
5154 fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, 0, 4);
5157 else if(IS_16X8(mb_type)){
5158 for(list=0; list<2; list++){
5159 if(h->ref_count[list]>0){
5161 if(IS_DIR(mb_type, i, list)){
5162 const int ref= h->ref_count[list] > 1 ? decode_cabac_mb_ref( h, list, 8*i ) : 0;
5163 fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, ref, 1);
5165 fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, (LIST_NOT_USED&0xFF), 1);
5169 for(list=0; list<2; list++){
5171 if(IS_DIR(mb_type, i, list)){
5172 pred_16x8_motion(h, 8*i, list, h->ref_cache[list][scan8[0] + 16*i], &mpx, &mpy);
5173 mx = mpx + decode_cabac_mb_mvd( h, list, 8*i, 0 );
5174 my = mpy + decode_cabac_mb_mvd( h, list, 8*i, 1 );
5175 tprintf("final mv:%d %d\n", mx, my);
5177 fill_rectangle(h->mvd_cache[list][ scan8[0] + 16*i ], 4, 2, 8, pack16to32(mx-mpx,my-mpy), 4);
5178 fill_rectangle(h->mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, pack16to32(mx,my), 4);
5180 fill_rectangle(h->mvd_cache[list][ scan8[0] + 16*i ], 4, 2, 8, 0, 4);
5181 fill_rectangle(h-> mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, 0, 4);
5186 assert(IS_8X16(mb_type));
5187 for(list=0; list<2; list++){
5188 if(h->ref_count[list]>0){
5190 if(IS_DIR(mb_type, i, list)){ //FIXME optimize
5191 const int ref= h->ref_count[list] > 1 ? decode_cabac_mb_ref( h, list, 4*i ) : 0;
5192 fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, ref, 1);
5194 fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, (LIST_NOT_USED&0xFF), 1);
5198 for(list=0; list<2; list++){
5200 if(IS_DIR(mb_type, i, list)){
5201 pred_8x16_motion(h, i*4, list, h->ref_cache[list][ scan8[0] + 2*i ], &mpx, &mpy);
5202 mx = mpx + decode_cabac_mb_mvd( h, list, 4*i, 0 );
5203 my = mpy + decode_cabac_mb_mvd( h, list, 4*i, 1 );
5205 tprintf("final mv:%d %d\n", mx, my);
5206 fill_rectangle(h->mvd_cache[list][ scan8[0] + 2*i ], 2, 4, 8, pack16to32(mx-mpx,my-mpy), 4);
5207 fill_rectangle(h->mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, pack16to32(mx,my), 4);
5209 fill_rectangle(h->mvd_cache[list][ scan8[0] + 2*i ], 2, 4, 8, 0, 4);
5210 fill_rectangle(h-> mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, 0, 4);
5217 if( IS_INTER( mb_type ) ) {
5218 h->chroma_pred_mode_table[mb_xy] = 0;
5219 write_back_motion( h, mb_type );
5222 if( !IS_INTRA16x16( mb_type ) ) {
5223 cbp = decode_cabac_mb_cbp_luma( h );
5224 cbp |= decode_cabac_mb_cbp_chroma( h ) << 4;
5227 h->cbp_table[mb_xy] = cbp;
5229 if( cbp || IS_INTRA16x16( mb_type ) ) {
5230 const uint8_t *scan, *dc_scan;
5233 if(IS_INTERLACED(mb_type)){
5235 dc_scan= luma_dc_field_scan;
5238 dc_scan= luma_dc_zigzag_scan;
5241 h->last_qscale_diff = dqp = decode_cabac_mb_dqp( h );
5243 if(((unsigned)s->qscale) > 51){
5244 if(s->qscale<0) s->qscale+= 52;
5245 else s->qscale-= 52;
5247 h->chroma_qp = get_chroma_qp(h, s->qscale);
5249 if( IS_INTRA16x16( mb_type ) ) {
5251 //av_log( s->avctx, AV_LOG_ERROR, "INTRA16x16 DC\n" );
5252 if( decode_cabac_residual( h, h->mb, 0, 0, dc_scan, s->qscale, 16) < 0)
5255 for( i = 0; i < 16; i++ ) {
5256 //av_log( s->avctx, AV_LOG_ERROR, "INTRA16x16 AC:%d\n", i );
5257 if( decode_cabac_residual(h, h->mb + 16*i, 1, i, scan + 1, s->qscale, 15) < 0 )
5261 fill_rectangle(&h->non_zero_count_cache[scan8[0]], 4, 4, 8, 0, 1);
5265 for( i8x8 = 0; i8x8 < 4; i8x8++ ) {
5266 if( cbp & (1<<i8x8) ) {
5267 for( i4x4 = 0; i4x4 < 4; i4x4++ ) {
5268 const int index = 4*i8x8 + i4x4;
5269 //av_log( s->avctx, AV_LOG_ERROR, "Luma4x4: %d\n", index );
5270 if( decode_cabac_residual(h, h->mb + 16*index, 2, index, scan, s->qscale, 16) < 0 )
5274 uint8_t * const nnz= &h->non_zero_count_cache[ scan8[4*i8x8] ];
5275 nnz[0] = nnz[1] = nnz[8] = nnz[9] = 0;
5282 for( c = 0; c < 2; c++ ) {
5283 //av_log( s->avctx, AV_LOG_ERROR, "INTRA C%d-DC\n",c );
5284 if( decode_cabac_residual(h, h->mb + 256 + 16*4*c, 3, c, chroma_dc_scan, h->chroma_qp, 4) < 0)
5291 for( c = 0; c < 2; c++ ) {
5292 for( i = 0; i < 4; i++ ) {
5293 const int index = 16 + 4 * c + i;
5294 //av_log( s->avctx, AV_LOG_ERROR, "INTRA C%d-AC %d\n",c, index - 16 );
5295 if( decode_cabac_residual(h, h->mb + 16*index, 4, index - 16, scan + 1, h->chroma_qp, 15) < 0)
5300 uint8_t * const nnz= &h->non_zero_count_cache[0];
5301 nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =
5302 nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
5305 uint8_t * const nnz= &h->non_zero_count_cache[0];
5306 fill_rectangle(&nnz[scan8[0]], 4, 4, 8, 0, 1);
5307 nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =
5308 nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
5311 s->current_picture.qscale_table[mb_xy]= s->qscale;
5312 write_back_non_zero_count(h);
5318 static void filter_mb_edgev( H264Context *h, uint8_t *pix, int stride, int bS[4], int qp ) {
5320 const int index_a = clip( qp + h->slice_alpha_c0_offset, 0, 51 );
5321 const int alpha = alpha_table[index_a];
5322 const int beta = beta_table[clip( qp + h->slice_beta_offset, 0, 51 )];
5324 for( i = 0; i < 4; i++ ) {
5331 const int tc0 = tc0_table[index_a][bS[i] - 1];
5332 /* 4px edge length */
5333 for( d = 0; d < 4; d++ ) {
5334 const int p0 = pix[-1];
5335 const int p1 = pix[-2];
5336 const int p2 = pix[-3];
5337 const int q0 = pix[0];
5338 const int q1 = pix[1];
5339 const int q2 = pix[2];
5341 if( ABS( p0 - q0 ) < alpha &&
5342 ABS( p1 - p0 ) < beta &&
5343 ABS( q1 - q0 ) < beta ) {
5347 if( ABS( p2 - p0 ) < beta ) {
5348 pix[-2] = p1 + clip( ( p2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( p1 << 1 ) ) >> 1, -tc0, tc0 );
5351 if( ABS( q2 - q0 ) < beta ) {
5352 pix[1] = q1 + clip( ( q2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( q1 << 1 ) ) >> 1, -tc0, tc0 );
5356 i_delta = clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );
5357 pix[-1] = clip_uint8( p0 + i_delta ); /* p0' */
5358 pix[0] = clip_uint8( q0 - i_delta ); /* q0' */
5363 /* 4px edge length */
5364 for( d = 0; d < 4; d++ ) {
5365 const int p0 = pix[-1];
5366 const int p1 = pix[-2];
5367 const int p2 = pix[-3];
5369 const int q0 = pix[0];
5370 const int q1 = pix[1];
5371 const int q2 = pix[2];
5373 if( ABS( p0 - q0 ) < alpha &&
5374 ABS( p1 - p0 ) < beta &&
5375 ABS( q1 - q0 ) < beta ) {
5377 if(ABS( p0 - q0 ) < (( alpha >> 2 ) + 2 )){
5378 if( ABS( p2 - p0 ) < beta)
5380 const int p3 = pix[-4];
5382 pix[-1] = ( p2 + 2*p1 + 2*p0 + 2*q0 + q1 + 4 ) >> 3;
5383 pix[-2] = ( p2 + p1 + p0 + q0 + 2 ) >> 2;
5384 pix[-3] = ( 2*p3 + 3*p2 + p1 + p0 + q0 + 4 ) >> 3;
5387 pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
5389 if( ABS( q2 - q0 ) < beta)
5391 const int q3 = pix[3];
5393 pix[0] = ( p1 + 2*p0 + 2*q0 + 2*q1 + q2 + 4 ) >> 3;
5394 pix[1] = ( p0 + q0 + q1 + q2 + 2 ) >> 2;
5395 pix[2] = ( 2*q3 + 3*q2 + q1 + q0 + p0 + 4 ) >> 3;
5398 pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
5402 pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
5403 pix[ 0] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
5411 static void filter_mb_edgecv( H264Context *h, uint8_t *pix, int stride, int bS[4], int qp ) {
5413 const int index_a = clip( qp + h->slice_alpha_c0_offset, 0, 51 );
5414 const int alpha = alpha_table[index_a];
5415 const int beta = beta_table[clip( qp + h->slice_beta_offset, 0, 51 )];
5417 for( i = 0; i < 4; i++ ) {
5424 const int tc = tc0_table[index_a][bS[i] - 1] + 1;
5425 /* 2px edge length (because we use same bS than the one for luma) */
5426 for( d = 0; d < 2; d++ ){
5427 const int p0 = pix[-1];
5428 const int p1 = pix[-2];
5429 const int q0 = pix[0];
5430 const int q1 = pix[1];
5432 if( ABS( p0 - q0 ) < alpha &&
5433 ABS( p1 - p0 ) < beta &&
5434 ABS( q1 - q0 ) < beta ) {
5435 const int i_delta = clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );
5437 pix[-1] = clip_uint8( p0 + i_delta ); /* p0' */
5438 pix[0] = clip_uint8( q0 - i_delta ); /* q0' */
5439 //tprintf("filter_mb_edgecv i:%d d:%d, qp:%d, indexA:%d, alpha:%d, beta:%d, tc:%d\n# bS:%d -> [%02x, %02x, %02x, %02x, %02x, %02x] =>[%02x, %02x, %02x, %02x]\n", i, d, qp, index_a, alpha, beta, tc, bS[i], pix[-3], p1, p0, q0, q1, pix[2], p1, pix[-1], pix[0], q1);
5444 /* 2px edge length (because we use same bS than the one for luma) */
5445 for( d = 0; d < 2; d++ ){
5446 const int p0 = pix[-1];
5447 const int p1 = pix[-2];
5448 const int q0 = pix[0];
5449 const int q1 = pix[1];
5451 if( ABS( p0 - q0 ) < alpha &&
5452 ABS( p1 - p0 ) < beta &&
5453 ABS( q1 - q0 ) < beta ) {
5455 pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2; /* p0' */
5456 pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2; /* q0' */
5457 //tprintf("filter_mb_edgecv i:%d d:%d\n# bS:4 -> [%02x, %02x, %02x, %02x, %02x, %02x] =>[%02x, %02x, %02x, %02x]\n", i, d, pix[-3], p1, p0, q0, q1, pix[2], p1, pix[-1], pix[0], q1);
5465 static void filter_mb_edgeh( H264Context *h, uint8_t *pix, int stride, int bS[4], int qp ) {
5467 const int index_a = clip( qp + h->slice_alpha_c0_offset, 0, 51 );
5468 const int alpha = alpha_table[index_a];
5469 const int beta = beta_table[clip( qp + h->slice_beta_offset, 0, 51 )];
5470 const int pix_next = stride;
5472 for( i = 0; i < 4; i++ ) {
5479 const int tc0 = tc0_table[index_a][bS[i] - 1];
5480 /* 4px edge length */
5481 for( d = 0; d < 4; d++ ) {
5482 const int p0 = pix[-1*pix_next];
5483 const int p1 = pix[-2*pix_next];
5484 const int p2 = pix[-3*pix_next];
5485 const int q0 = pix[0];
5486 const int q1 = pix[1*pix_next];
5487 const int q2 = pix[2*pix_next];
5489 if( ABS( p0 - q0 ) < alpha &&
5490 ABS( p1 - p0 ) < beta &&
5491 ABS( q1 - q0 ) < beta ) {
5496 if( ABS( p2 - p0 ) < beta ) {
5497 pix[-2*pix_next] = p1 + clip( ( p2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( p1 << 1 ) ) >> 1, -tc0, tc0 );
5500 if( ABS( q2 - q0 ) < beta ) {
5501 pix[pix_next] = q1 + clip( ( q2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( q1 << 1 ) ) >> 1, -tc0, tc0 );
5505 i_delta = clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );
5506 pix[-pix_next] = clip_uint8( p0 + i_delta ); /* p0' */
5507 pix[0] = clip_uint8( q0 - i_delta ); /* q0' */
5512 /* 4px edge length */
5513 for( d = 0; d < 4; d++ ) {
5514 const int p0 = pix[-1*pix_next];
5515 const int p1 = pix[-2*pix_next];
5516 const int p2 = pix[-3*pix_next];
5517 const int q0 = pix[0];
5518 const int q1 = pix[1*pix_next];
5519 const int q2 = pix[2*pix_next];
5521 if( ABS( p0 - q0 ) < alpha &&
5522 ABS( p1 - p0 ) < beta &&
5523 ABS( q1 - q0 ) < beta ) {
5525 const int p3 = pix[-4*pix_next];
5526 const int q3 = pix[ 3*pix_next];
5528 if(ABS( p0 - q0 ) < (( alpha >> 2 ) + 2 )){
5529 if( ABS( p2 - p0 ) < beta) {
5531 pix[-1*pix_next] = ( p2 + 2*p1 + 2*p0 + 2*q0 + q1 + 4 ) >> 3;
5532 pix[-2*pix_next] = ( p2 + p1 + p0 + q0 + 2 ) >> 2;
5533 pix[-3*pix_next] = ( 2*p3 + 3*p2 + p1 + p0 + q0 + 4 ) >> 3;
5536 pix[-1*pix_next] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
5538 if( ABS( q2 - q0 ) < beta) {
5540 pix[0*pix_next] = ( p1 + 2*p0 + 2*q0 + 2*q1 + q2 + 4 ) >> 3;
5541 pix[1*pix_next] = ( p0 + q0 + q1 + q2 + 2 ) >> 2;
5542 pix[2*pix_next] = ( 2*q3 + 3*q2 + q1 + q0 + p0 + 4 ) >> 3;
5545 pix[0*pix_next] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
5549 pix[-1*pix_next] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
5550 pix[ 0*pix_next] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
5559 static void filter_mb_edgech( H264Context *h, uint8_t *pix, int stride, int bS[4], int qp ) {
5561 const int index_a = clip( qp + h->slice_alpha_c0_offset, 0, 51 );
5562 const int alpha = alpha_table[index_a];
5563 const int beta = beta_table[clip( qp + h->slice_beta_offset, 0, 51 )];
5564 const int pix_next = stride;
5566 for( i = 0; i < 4; i++ )
5574 int tc = tc0_table[index_a][bS[i] - 1] + 1;
5575 /* 2px edge length (see deblocking_filter_edgecv) */
5576 for( d = 0; d < 2; d++ ) {
5577 const int p0 = pix[-1*pix_next];
5578 const int p1 = pix[-2*pix_next];
5579 const int q0 = pix[0];
5580 const int q1 = pix[1*pix_next];
5582 if( ABS( p0 - q0 ) < alpha &&
5583 ABS( p1 - p0 ) < beta &&
5584 ABS( q1 - q0 ) < beta ) {
5586 int i_delta = clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );
5588 pix[-pix_next] = clip_uint8( p0 + i_delta ); /* p0' */
5589 pix[0] = clip_uint8( q0 - i_delta ); /* q0' */
5594 /* 2px edge length (see deblocking_filter_edgecv) */
5595 for( d = 0; d < 2; d++ ) {
5596 const int p0 = pix[-1*pix_next];
5597 const int p1 = pix[-2*pix_next];
5598 const int q0 = pix[0];
5599 const int q1 = pix[1*pix_next];
5601 if( ABS( p0 - q0 ) < alpha &&
5602 ABS( p1 - p0 ) < beta &&
5603 ABS( q1 - q0 ) < beta ) {
5605 pix[-pix_next] = ( 2*p1 + p0 + q1 + 2 ) >> 2; /* p0' */
5606 pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2; /* q0' */
5614 static void filter_mb( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr) {
5615 MpegEncContext * const s = &h->s;
5616 const int mb_xy= mb_x + mb_y*s->mb_stride;
5617 int linesize, uvlinesize;
5620 /* FIXME Implement deblocking filter for field MB */
5621 if( h->sps.mb_aff ) {
5624 linesize = s->linesize;
5625 uvlinesize = s->uvlinesize;
5627 /* dir : 0 -> vertical edge, 1 -> horizontal edge */
5628 for( dir = 0; dir < 2; dir++ )
5631 const int mbm_xy = dir == 0 ? mb_xy -1 : mb_xy - s->mb_stride;
5632 int start = h->slice_table[mbm_xy] == 255 ? 1 : 0;
5634 if (h->deblocking_filter==2 && h->slice_table[mbm_xy] != h->slice_table[mb_xy])
5638 for( edge = start; edge < 4; edge++ ) {
5639 /* mbn_xy: neighbour macroblock (how that works for field ?) */
5640 int mbn_xy = edge > 0 ? mb_xy : mbm_xy;
5644 if( IS_INTRA( s->current_picture.mb_type[mb_xy] ) ||
5645 IS_INTRA( s->current_picture.mb_type[mbn_xy] ) ) {
5646 bS[0] = bS[1] = bS[2] = bS[3] = ( edge == 0 ? 4 : 3 );
5649 for( i = 0; i < 4; i++ ) {
5650 int x = dir == 0 ? edge : i;
5651 int y = dir == 0 ? i : edge;
5652 int b_idx= 8 + 4 + x + 8*y;
5653 int bn_idx= b_idx - (dir ? 8:1);
5655 if( h->non_zero_count_cache[b_idx] != 0 ||
5656 h->non_zero_count_cache[bn_idx] != 0 ) {
5661 /* FIXME: A given frame may occupy more than one position in
5662 * the reference list. So we should compare the frame numbers,
5663 * not the indices in the ref list. */
5666 for( l = 0; l < 1 + (h->slice_type == B_TYPE); l++ ) {
5667 if( h->ref_cache[l][b_idx] != h->ref_cache[l][bn_idx] ||
5668 ABS( h->mv_cache[l][b_idx][0] - h->mv_cache[l][bn_idx][0] ) >= 4 ||
5669 ABS( h->mv_cache[l][b_idx][1] - h->mv_cache[l][bn_idx][1] ) >= 4 ) {
5677 if(bS[0]+bS[1]+bS[2]+bS[3] == 0)
5682 // Do not use s->qscale as luma quantiser because it has not the same
5683 // value in IPCM macroblocks.
5684 qp = ( s->current_picture.qscale_table[mb_xy] + s->current_picture.qscale_table[mbn_xy] + 1 ) >> 1;
5685 //tprintf("filter mb:%d/%d dir:%d edge:%d, QPy:%d, QPc:%d, QPcn:%d\n", mb_x, mb_y, dir, edge, qp, h->chroma_qp, s->current_picture.qscale_table[mbn_xy]);
5687 filter_mb_edgev( h, &img_y[4*edge], linesize, bS, qp );
5688 if( (edge&1) == 0 ) {
5689 int chroma_qp = ( h->chroma_qp +
5690 get_chroma_qp( h, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1;
5691 filter_mb_edgecv( h, &img_cb[2*edge], uvlinesize, bS, chroma_qp );
5692 filter_mb_edgecv( h, &img_cr[2*edge], uvlinesize, bS, chroma_qp );
5695 filter_mb_edgeh( h, &img_y[4*edge*linesize], linesize, bS, qp );
5696 if( (edge&1) == 0 ) {
5697 int chroma_qp = ( h->chroma_qp +
5698 get_chroma_qp( h, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1;
5699 filter_mb_edgech( h, &img_cb[2*edge*uvlinesize], uvlinesize, bS, chroma_qp );
5700 filter_mb_edgech( h, &img_cr[2*edge*uvlinesize], uvlinesize, bS, chroma_qp );
5707 static int decode_slice(H264Context *h){
5708 MpegEncContext * const s = &h->s;
5709 const int part_mask= s->partitioned_frame ? (AC_END|AC_ERROR) : 0x7F;
5713 if( h->pps.cabac ) {
5717 align_get_bits( &s->gb );
5720 ff_init_cabac_states( &h->cabac, ff_h264_lps_range, ff_h264_mps_state, ff_h264_lps_state, 64 );
5721 ff_init_cabac_decoder( &h->cabac,
5722 s->gb.buffer + get_bits_count(&s->gb)/8,
5723 ( s->gb.size_in_bits - get_bits_count(&s->gb) + 7)/8);
5724 /* calculate pre-state */
5725 for( i= 0; i < 399; i++ ) {
5727 if( h->slice_type == I_TYPE )
5728 pre = clip( ((cabac_context_init_I[i][0] * s->qscale) >>4 ) + cabac_context_init_I[i][1], 1, 126 );
5730 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 );
5733 h->cabac_state[i] = 2 * ( 63 - pre ) + 0;
5735 h->cabac_state[i] = 2 * ( pre - 64 ) + 1;
5739 int ret = decode_mb_cabac(h);
5740 int eos = get_cabac_terminate( &h->cabac ); /* End of Slice flag */
5742 if(ret>=0) hl_decode_mb(h);
5744 /* XXX: useless as decode_mb_cabac it doesn't support that ... */
5745 if( ret >= 0 && h->sps.mb_aff ) { //FIXME optimal? or let mb_decode decode 16x32 ?
5748 if(ret>=0) ret = decode_mb_cabac(h);
5749 eos = get_cabac_terminate( &h->cabac );
5755 if( ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 1) {
5756 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
5757 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);
5761 if( ++s->mb_x >= s->mb_width ) {
5763 ff_draw_horiz_band(s, 16*s->mb_y, 16);
5767 if( eos || s->mb_y >= s->mb_height ) {
5768 tprintf("slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
5769 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);
5773 /* TODO test over-reading in cabac code */
5774 else if( read too much in h->cabac ) {
5775 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);
5783 int ret = decode_mb_cavlc(h);
5785 if(ret>=0) hl_decode_mb(h);
5787 if(ret>=0 && h->sps.mb_aff){ //FIXME optimal? or let mb_decode decode 16x32 ?
5789 ret = decode_mb_cavlc(h);
5791 if(ret>=0) hl_decode_mb(h);
5796 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
5797 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);
5802 if(++s->mb_x >= s->mb_width){
5804 ff_draw_horiz_band(s, 16*s->mb_y, 16);
5805 if(++s->mb_y >= s->mb_height){
5806 tprintf("slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
5808 if(get_bits_count(&s->gb) == s->gb.size_in_bits ) {
5809 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);
5813 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);
5820 if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->mb_skip_run<=0){
5821 tprintf("slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
5822 if(get_bits_count(&s->gb) == s->gb.size_in_bits ){
5823 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);
5827 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);
5836 for(;s->mb_y < s->mb_height; s->mb_y++){
5837 for(;s->mb_x < s->mb_width; s->mb_x++){
5838 int ret= decode_mb(h);
5843 fprintf(stderr, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
5844 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);
5849 if(++s->mb_x >= s->mb_width){
5851 if(++s->mb_y >= s->mb_height){
5852 if(get_bits_count(s->gb) == s->gb.size_in_bits){
5853 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);
5857 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);
5864 if(get_bits_count(s->?gb) >= s->gb?.size_in_bits){
5865 if(get_bits_count(s->gb) == s->gb.size_in_bits){
5866 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);
5870 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);
5877 ff_draw_horiz_band(s, 16*s->mb_y, 16);
5880 return -1; //not reached
5883 static inline int decode_vui_parameters(H264Context *h, SPS *sps){
5884 MpegEncContext * const s = &h->s;
5885 int aspect_ratio_info_present_flag, aspect_ratio_idc;
5887 aspect_ratio_info_present_flag= get_bits1(&s->gb);
5889 if( aspect_ratio_info_present_flag ) {
5890 aspect_ratio_idc= get_bits(&s->gb, 8);
5891 if( aspect_ratio_idc == EXTENDED_SAR ) {
5892 sps->sar.num= get_bits(&s->gb, 16);
5893 sps->sar.den= get_bits(&s->gb, 16);
5894 }else if(aspect_ratio_idc < 16){
5895 sps->sar= pixel_aspect[aspect_ratio_idc];
5897 av_log(h->s.avctx, AV_LOG_ERROR, "illegal aspect ratio\n");
5904 // s->avctx->aspect_ratio= sar_width*s->width / (float)(s->height*sar_height);
5906 if(get_bits1(&s->gb)){ /* overscan_info_present_flag */
5907 get_bits1(&s->gb); /* overscan_appropriate_flag */
5910 if(get_bits1(&s->gb)){ /* video_signal_type_present_flag */
5911 get_bits(&s->gb, 3); /* video_format */
5912 get_bits1(&s->gb); /* video_full_range_flag */
5913 if(get_bits1(&s->gb)){ /* colour_description_present_flag */
5914 get_bits(&s->gb, 8); /* colour_primaries */
5915 get_bits(&s->gb, 8); /* transfer_characteristics */
5916 get_bits(&s->gb, 8); /* matrix_coefficients */
5920 if(get_bits1(&s->gb)){ /* chroma_location_info_present_flag */
5921 get_ue_golomb(&s->gb); /* chroma_sample_location_type_top_field */
5922 get_ue_golomb(&s->gb); /* chroma_sample_location_type_bottom_field */
5925 sps->timing_info_present_flag = get_bits1(&s->gb);
5926 if(sps->timing_info_present_flag){
5927 sps->num_units_in_tick = get_bits_long(&s->gb, 32);
5928 sps->time_scale = get_bits_long(&s->gb, 32);
5929 sps->fixed_frame_rate_flag = get_bits1(&s->gb);
5933 | nal_hrd_parameters_present_flag |0 |u(1) |
5934 | if( nal_hrd_parameters_present_flag = = 1) | | |
5935 | hrd_parameters( ) | | |
5936 | vcl_hrd_parameters_present_flag |0 |u(1) |
5937 | if( vcl_hrd_parameters_present_flag = = 1) | | |
5938 | hrd_parameters( ) | | |
5939 | if( ( nal_hrd_parameters_present_flag = = 1 | || | |
5941 |( vcl_hrd_parameters_present_flag = = 1 ) ) | | |
5942 | low_delay_hrd_flag |0 |u(1) |
5943 | bitstream_restriction_flag |0 |u(1) |
5944 | if( bitstream_restriction_flag ) { |0 |u(1) |
5945 | motion_vectors_over_pic_boundaries_flag |0 |u(1) |
5946 | max_bytes_per_pic_denom |0 |ue(v) |
5947 | max_bits_per_mb_denom |0 |ue(v) |
5948 | log2_max_mv_length_horizontal |0 |ue(v) |
5949 | log2_max_mv_length_vertical |0 |ue(v) |
5950 | num_reorder_frames |0 |ue(v) |
5951 | max_dec_frame_buffering |0 |ue(v) |
5958 static inline int decode_seq_parameter_set(H264Context *h){
5959 MpegEncContext * const s = &h->s;
5960 int profile_idc, level_idc;
5964 profile_idc= get_bits(&s->gb, 8);
5965 get_bits1(&s->gb); //constraint_set0_flag
5966 get_bits1(&s->gb); //constraint_set1_flag
5967 get_bits1(&s->gb); //constraint_set2_flag
5968 get_bits1(&s->gb); //constraint_set3_flag
5969 get_bits(&s->gb, 4); // reserved
5970 level_idc= get_bits(&s->gb, 8);
5971 sps_id= get_ue_golomb(&s->gb);
5973 sps= &h->sps_buffer[ sps_id ];
5974 sps->profile_idc= profile_idc;
5975 sps->level_idc= level_idc;
5977 sps->log2_max_frame_num= get_ue_golomb(&s->gb) + 4;
5978 sps->poc_type= get_ue_golomb(&s->gb);
5980 if(sps->poc_type == 0){ //FIXME #define
5981 sps->log2_max_poc_lsb= get_ue_golomb(&s->gb) + 4;
5982 } else if(sps->poc_type == 1){//FIXME #define
5983 sps->delta_pic_order_always_zero_flag= get_bits1(&s->gb);
5984 sps->offset_for_non_ref_pic= get_se_golomb(&s->gb);
5985 sps->offset_for_top_to_bottom_field= get_se_golomb(&s->gb);
5986 sps->poc_cycle_length= get_ue_golomb(&s->gb);
5988 for(i=0; i<sps->poc_cycle_length; i++)
5989 sps->offset_for_ref_frame[i]= get_se_golomb(&s->gb);
5991 if(sps->poc_type > 2){
5992 av_log(h->s.avctx, AV_LOG_ERROR, "illegal POC type %d\n", sps->poc_type);
5996 sps->ref_frame_count= get_ue_golomb(&s->gb);
5997 if(sps->ref_frame_count > MAX_PICTURE_COUNT-2){
5998 av_log(h->s.avctx, AV_LOG_ERROR, "too many reference frames\n");
6000 sps->gaps_in_frame_num_allowed_flag= get_bits1(&s->gb);
6001 sps->mb_width= get_ue_golomb(&s->gb) + 1;
6002 sps->mb_height= get_ue_golomb(&s->gb) + 1;
6003 if((unsigned)sps->mb_width >= INT_MAX/16 || (unsigned)sps->mb_height >= INT_MAX/16 ||
6004 avcodec_check_dimensions(NULL, 16*sps->mb_width, 16*sps->mb_height))
6007 sps->frame_mbs_only_flag= get_bits1(&s->gb);
6008 if(!sps->frame_mbs_only_flag)
6009 sps->mb_aff= get_bits1(&s->gb);
6013 sps->direct_8x8_inference_flag= get_bits1(&s->gb);
6015 sps->crop= get_bits1(&s->gb);
6017 sps->crop_left = get_ue_golomb(&s->gb);
6018 sps->crop_right = get_ue_golomb(&s->gb);
6019 sps->crop_top = get_ue_golomb(&s->gb);
6020 sps->crop_bottom= get_ue_golomb(&s->gb);
6021 if(sps->crop_left || sps->crop_top){
6022 av_log(h->s.avctx, AV_LOG_ERROR, "insane cropping not completly supported, this could look slightly wrong ...\n");
6028 sps->crop_bottom= 0;
6031 sps->vui_parameters_present_flag= get_bits1(&s->gb);
6032 if( sps->vui_parameters_present_flag )
6033 decode_vui_parameters(h, sps);
6035 if(s->avctx->debug&FF_DEBUG_PICT_INFO){
6036 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",
6037 sps_id, sps->profile_idc, sps->level_idc,
6039 sps->ref_frame_count,
6040 sps->mb_width, sps->mb_height,
6041 sps->frame_mbs_only_flag ? "FRM" : (sps->mb_aff ? "MB-AFF" : "PIC-AFF"),
6042 sps->direct_8x8_inference_flag ? "8B8" : "",
6043 sps->crop_left, sps->crop_right,
6044 sps->crop_top, sps->crop_bottom,
6045 sps->vui_parameters_present_flag ? "VUI" : ""
6051 static inline int decode_picture_parameter_set(H264Context *h){
6052 MpegEncContext * const s = &h->s;
6053 int pps_id= get_ue_golomb(&s->gb);
6054 PPS *pps= &h->pps_buffer[pps_id];
6056 pps->sps_id= get_ue_golomb(&s->gb);
6057 pps->cabac= get_bits1(&s->gb);
6058 pps->pic_order_present= get_bits1(&s->gb);
6059 pps->slice_group_count= get_ue_golomb(&s->gb) + 1;
6060 if(pps->slice_group_count > 1 ){
6061 pps->mb_slice_group_map_type= get_ue_golomb(&s->gb);
6062 av_log(h->s.avctx, AV_LOG_ERROR, "FMO not supported\n");
6063 switch(pps->mb_slice_group_map_type){
6066 | for( i = 0; i <= num_slice_groups_minus1; i++ ) | | |
6067 | run_length[ i ] |1 |ue(v) |
6072 | for( i = 0; i < num_slice_groups_minus1; i++ ) | | |
6074 | top_left_mb[ i ] |1 |ue(v) |
6075 | bottom_right_mb[ i ] |1 |ue(v) |
6083 | slice_group_change_direction_flag |1 |u(1) |
6084 | slice_group_change_rate_minus1 |1 |ue(v) |
6089 | slice_group_id_cnt_minus1 |1 |ue(v) |
6090 | for( i = 0; i <= slice_group_id_cnt_minus1; i++ | | |
6092 | slice_group_id[ i ] |1 |u(v) |
6097 pps->ref_count[0]= get_ue_golomb(&s->gb) + 1;
6098 pps->ref_count[1]= get_ue_golomb(&s->gb) + 1;
6099 if(pps->ref_count[0] > 32 || pps->ref_count[1] > 32){
6100 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow (pps)\n");
6104 pps->weighted_pred= get_bits1(&s->gb);
6105 pps->weighted_bipred_idc= get_bits(&s->gb, 2);
6106 pps->init_qp= get_se_golomb(&s->gb) + 26;
6107 pps->init_qs= get_se_golomb(&s->gb) + 26;
6108 pps->chroma_qp_index_offset= get_se_golomb(&s->gb);
6109 pps->deblocking_filter_parameters_present= get_bits1(&s->gb);
6110 pps->constrained_intra_pred= get_bits1(&s->gb);
6111 pps->redundant_pic_cnt_present = get_bits1(&s->gb);
6113 if(s->avctx->debug&FF_DEBUG_PICT_INFO){
6114 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",
6115 pps_id, pps->sps_id,
6116 pps->cabac ? "CABAC" : "CAVLC",
6117 pps->slice_group_count,
6118 pps->ref_count[0], pps->ref_count[1],
6119 pps->weighted_pred ? "weighted" : "",
6120 pps->init_qp, pps->init_qs, pps->chroma_qp_index_offset,
6121 pps->deblocking_filter_parameters_present ? "LPAR" : "",
6122 pps->constrained_intra_pred ? "CONSTR" : "",
6123 pps->redundant_pic_cnt_present ? "REDU" : ""
6131 * finds the end of the current frame in the bitstream.
6132 * @return the position of the first byte of the next frame, or -1
6134 static int find_frame_end(H264Context *h, const uint8_t *buf, int buf_size){
6137 ParseContext *pc = &(h->s.parse_context);
6138 //printf("first %02X%02X%02X%02X\n", buf[0], buf[1],buf[2],buf[3]);
6139 // mb_addr= pc->mb_addr - 1;
6141 for(i=0; i<=buf_size; i++){
6142 if((state&0xFFFFFF1F) == 0x101 || (state&0xFFFFFF1F) == 0x102 || (state&0xFFFFFF1F) == 0x105){
6143 tprintf("find_frame_end new startcode = %08x, frame_start_found = %d, pos = %d\n", state, pc->frame_start_found, i);
6144 if(pc->frame_start_found){
6145 // If there isn't one more byte in the buffer
6146 // the test on first_mb_in_slice cannot be done yet
6147 // do it at next call.
6148 if (i >= buf_size) break;
6149 if (buf[i] & 0x80) {
6150 // first_mb_in_slice is 0, probably the first nal of a new
6152 tprintf("find_frame_end frame_end_found, state = %08x, pos = %d\n", state, i);
6154 pc->frame_start_found= 0;
6158 pc->frame_start_found = 1;
6161 state= (state<<8) | buf[i];
6165 return END_NOT_FOUND;
6168 static int h264_parse(AVCodecParserContext *s,
6169 AVCodecContext *avctx,
6170 uint8_t **poutbuf, int *poutbuf_size,
6171 const uint8_t *buf, int buf_size)
6173 H264Context *h = s->priv_data;
6174 ParseContext *pc = &h->s.parse_context;
6177 next= find_frame_end(h, buf, buf_size);
6179 if (ff_combine_frame(pc, next, (uint8_t **)&buf, &buf_size) < 0) {
6185 *poutbuf = (uint8_t *)buf;
6186 *poutbuf_size = buf_size;
6190 static int decode_nal_units(H264Context *h, uint8_t *buf, int buf_size){
6191 MpegEncContext * const s = &h->s;
6192 AVCodecContext * const avctx= s->avctx;
6196 for(i=0; i<32; i++){
6197 printf("%X ", buf[i]);
6209 if(buf_index >= buf_size) break;
6211 for(i = 0; i < h->nal_length_size; i++)
6212 nalsize = (nalsize << 8) | buf[buf_index++];
6214 // start code prefix search
6215 for(; buf_index + 3 < buf_size; buf_index++){
6216 // this should allways succeed in the first iteration
6217 if(buf[buf_index] == 0 && buf[buf_index+1] == 0 && buf[buf_index+2] == 1)
6221 if(buf_index+3 >= buf_size) break;
6226 ptr= decode_nal(h, buf + buf_index, &dst_length, &consumed, h->is_avc ? nalsize : buf_size - buf_index);
6227 if(ptr[dst_length - 1] == 0) dst_length--;
6228 bit_length= 8*dst_length - decode_rbsp_trailing(ptr + dst_length - 1);
6230 if(s->avctx->debug&FF_DEBUG_STARTCODE){
6231 av_log(h->s.avctx, AV_LOG_DEBUG, "NAL %d at %d/%d length %d\n", h->nal_unit_type, buf_index, buf_size, dst_length);
6234 if (h->is_avc && (nalsize != consumed))
6235 av_log(h->s.avctx, AV_LOG_ERROR, "AVC: Consumed only %d bytes instead of %d\n", consumed, nalsize);
6237 buf_index += consumed;
6239 if( s->hurry_up == 1 && h->nal_ref_idc == 0 )
6242 switch(h->nal_unit_type){
6244 idr(h); //FIXME ensure we dont loose some frames if there is reordering
6246 init_get_bits(&s->gb, ptr, bit_length);
6248 h->inter_gb_ptr= &s->gb;
6249 s->data_partitioning = 0;
6251 if(decode_slice_header(h) < 0) return -1;
6252 if(h->redundant_pic_count==0 && s->hurry_up < 5 )
6256 init_get_bits(&s->gb, ptr, bit_length);
6258 h->inter_gb_ptr= NULL;
6259 s->data_partitioning = 1;
6261 if(decode_slice_header(h) < 0) return -1;
6264 init_get_bits(&h->intra_gb, ptr, bit_length);
6265 h->intra_gb_ptr= &h->intra_gb;
6268 init_get_bits(&h->inter_gb, ptr, bit_length);
6269 h->inter_gb_ptr= &h->inter_gb;
6271 if(h->redundant_pic_count==0 && h->intra_gb_ptr && s->data_partitioning && s->hurry_up < 5 )
6277 init_get_bits(&s->gb, ptr, bit_length);
6278 decode_seq_parameter_set(h);
6280 if(s->flags& CODEC_FLAG_LOW_DELAY)
6283 avctx->has_b_frames= !s->low_delay;
6286 init_get_bits(&s->gb, ptr, bit_length);
6288 decode_picture_parameter_set(h);
6291 case NAL_PICTURE_DELIMITER:
6293 case NAL_FILTER_DATA:
6296 av_log(avctx, AV_LOG_ERROR, "Unknown NAL code: %d\n", h->nal_unit_type);
6299 //FIXME move after where irt is set
6300 s->current_picture.pict_type= s->pict_type;
6301 s->current_picture.key_frame= s->pict_type == I_TYPE;
6304 if(!s->current_picture_ptr) return buf_index; //no frame
6306 h->prev_frame_num_offset= h->frame_num_offset;
6307 h->prev_frame_num= h->frame_num;
6308 if(s->current_picture_ptr->reference){
6309 h->prev_poc_msb= h->poc_msb;
6310 h->prev_poc_lsb= h->poc_lsb;
6312 if(s->current_picture_ptr->reference)
6313 execute_ref_pic_marking(h, h->mmco, h->mmco_index);
6323 * retunrs the number of bytes consumed for building the current frame
6325 static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size){
6326 if(s->flags&CODEC_FLAG_TRUNCATED){
6327 pos -= s->parse_context.last_index;
6328 if(pos<0) pos=0; // FIXME remove (uneeded?)
6332 if(pos==0) pos=1; //avoid infinite loops (i doubt thats needed but ...)
6333 if(pos+10>buf_size) pos=buf_size; // oops ;)
6339 static int decode_frame(AVCodecContext *avctx,
6340 void *data, int *data_size,
6341 uint8_t *buf, int buf_size)
6343 H264Context *h = avctx->priv_data;
6344 MpegEncContext *s = &h->s;
6345 AVFrame *pict = data;
6348 s->flags= avctx->flags;
6349 s->flags2= avctx->flags2;
6351 /* no supplementary picture */
6352 if (buf_size == 0) {
6356 if(s->flags&CODEC_FLAG_TRUNCATED){
6357 int next= find_frame_end(h, buf, buf_size);
6359 if( ff_combine_frame(&s->parse_context, next, &buf, &buf_size) < 0 )
6361 //printf("next:%d buf_size:%d last_index:%d\n", next, buf_size, s->parse_context.last_index);
6364 if(h->is_avc && !h->got_avcC) {
6365 int i, cnt, nalsize;
6366 unsigned char *p = avctx->extradata;
6367 if(avctx->extradata_size < 7) {
6368 av_log(avctx, AV_LOG_ERROR, "avcC too short\n");
6372 av_log(avctx, AV_LOG_ERROR, "Unknown avcC version %d\n", *p);
6375 /* sps and pps in the avcC always have length coded with 2 bytes,
6376 so put a fake nal_length_size = 2 while parsing them */
6377 h->nal_length_size = 2;
6378 // Decode sps from avcC
6379 cnt = *(p+5) & 0x1f; // Number of sps
6381 for (i = 0; i < cnt; i++) {
6382 nalsize = BE_16(p) + 2;
6383 if(decode_nal_units(h, p, nalsize) != nalsize) {
6384 av_log(avctx, AV_LOG_ERROR, "Decoding sps %d from avcC failed\n", i);
6389 // Decode pps from avcC
6390 cnt = *(p++); // Number of pps
6391 for (i = 0; i < cnt; i++) {
6392 nalsize = BE_16(p) + 2;
6393 if(decode_nal_units(h, p, nalsize) != nalsize) {
6394 av_log(avctx, AV_LOG_ERROR, "Decoding pps %d from avcC failed\n", i);
6399 // Now store right nal length size, that will be use to parse all other nals
6400 h->nal_length_size = ((*(((char*)(avctx->extradata))+4))&0x03)+1;
6401 // Do not reparse avcC
6405 if(!h->is_avc && s->avctx->extradata_size && s->picture_number==0){
6406 if(0 < decode_nal_units(h, s->avctx->extradata, s->avctx->extradata_size) )
6410 buf_index=decode_nal_units(h, buf, buf_size);
6414 //FIXME do something with unavailable reference frames
6416 // if(ret==FRAME_SKIPED) return get_consumed_bytes(s, buf_index, buf_size);
6417 if(!s->current_picture_ptr){
6418 av_log(h->s.avctx, AV_LOG_DEBUG, "error, NO frame\n");
6423 //#define DECODE_ORDER
6424 Picture *out = s->current_picture_ptr;
6425 #ifndef DECODE_ORDER
6426 /* Sort B-frames into display order
6427 * FIXME doesn't allow for multiple delayed frames */
6428 Picture *cur = s->current_picture_ptr;
6429 Picture *prev = h->delayed_pic[0];
6432 && (cur->pict_type == B_TYPE
6433 || (!h->sps.gaps_in_frame_num_allowed_flag
6434 && prev && cur->poc - prev->poc > 2))){
6436 s->avctx->has_b_frames = 1;
6437 if(prev && prev->poc > cur->poc)
6438 // too late to display this frame
6442 if(s->low_delay || !prev || cur->pict_type == B_TYPE)
6446 if(s->low_delay || !prev || out == prev){
6447 if(prev && prev->reference == 1)
6448 prev->reference = 0;
6449 h->delayed_pic[0] = cur;
6453 *pict= *(AVFrame*)out;
6456 ff_print_debug_info(s, pict);
6457 assert(pict->data[0]);
6458 //printf("out %d\n", (int)pict->data[0]);
6461 /* Return the Picture timestamp as the frame number */
6462 /* we substract 1 because it is added on utils.c */
6463 avctx->frame_number = s->picture_number - 1;
6466 /* dont output the last pic after seeking */
6467 if(s->last_picture_ptr || s->low_delay)
6468 //Note this isnt a issue as a IDR pic should flush the buffers
6470 *data_size = sizeof(AVFrame);
6471 return get_consumed_bytes(s, buf_index, buf_size);
6474 static inline void fill_mb_avail(H264Context *h){
6475 MpegEncContext * const s = &h->s;
6476 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
6479 h->mb_avail[0]= s->mb_x && h->slice_table[mb_xy - s->mb_stride - 1] == h->slice_num;
6480 h->mb_avail[1]= h->slice_table[mb_xy - s->mb_stride ] == h->slice_num;
6481 h->mb_avail[2]= s->mb_x+1 < s->mb_width && h->slice_table[mb_xy - s->mb_stride + 1] == h->slice_num;
6487 h->mb_avail[3]= s->mb_x && h->slice_table[mb_xy - 1] == h->slice_num;
6488 h->mb_avail[4]= 1; //FIXME move out
6489 h->mb_avail[5]= 0; //FIXME move out
6495 #define SIZE (COUNT*40)
6501 // int int_temp[10000];
6503 AVCodecContext avctx;
6505 dsputil_init(&dsp, &avctx);
6507 init_put_bits(&pb, temp, SIZE);
6508 printf("testing unsigned exp golomb\n");
6509 for(i=0; i<COUNT; i++){
6511 set_ue_golomb(&pb, i);
6512 STOP_TIMER("set_ue_golomb");
6514 flush_put_bits(&pb);
6516 init_get_bits(&gb, temp, 8*SIZE);
6517 for(i=0; i<COUNT; i++){
6520 s= show_bits(&gb, 24);
6523 j= get_ue_golomb(&gb);
6525 printf("missmatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
6528 STOP_TIMER("get_ue_golomb");
6532 init_put_bits(&pb, temp, SIZE);
6533 printf("testing signed exp golomb\n");
6534 for(i=0; i<COUNT; i++){
6536 set_se_golomb(&pb, i - COUNT/2);
6537 STOP_TIMER("set_se_golomb");
6539 flush_put_bits(&pb);
6541 init_get_bits(&gb, temp, 8*SIZE);
6542 for(i=0; i<COUNT; i++){
6545 s= show_bits(&gb, 24);
6548 j= get_se_golomb(&gb);
6549 if(j != i - COUNT/2){
6550 printf("missmatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
6553 STOP_TIMER("get_se_golomb");
6556 printf("testing 4x4 (I)DCT\n");
6559 uint8_t src[16], ref[16];
6560 uint64_t error= 0, max_error=0;
6562 for(i=0; i<COUNT; i++){
6564 // printf("%d %d %d\n", r1, r2, (r2-r1)*16);
6565 for(j=0; j<16; j++){
6566 ref[j]= random()%255;
6567 src[j]= random()%255;
6570 h264_diff_dct_c(block, src, ref, 4);
6573 for(j=0; j<16; j++){
6574 // printf("%d ", block[j]);
6575 block[j]= block[j]*4;
6576 if(j&1) block[j]= (block[j]*4 + 2)/5;
6577 if(j&4) block[j]= (block[j]*4 + 2)/5;
6581 s->dsp.h264_idct_add(ref, block, 4);
6582 /* for(j=0; j<16; j++){
6583 printf("%d ", ref[j]);
6587 for(j=0; j<16; j++){
6588 int diff= ABS(src[j] - ref[j]);
6591 max_error= FFMAX(max_error, diff);
6594 printf("error=%f max_error=%d\n", ((float)error)/COUNT/16, (int)max_error );
6596 printf("testing quantizer\n");
6597 for(qp=0; qp<52; qp++){
6599 src1_block[i]= src2_block[i]= random()%255;
6603 printf("Testing NAL layer\n");
6605 uint8_t bitstream[COUNT];
6606 uint8_t nal[COUNT*2];
6608 memset(&h, 0, sizeof(H264Context));
6610 for(i=0; i<COUNT; i++){
6618 for(j=0; j<COUNT; j++){
6619 bitstream[j]= (random() % 255) + 1;
6622 for(j=0; j<zeros; j++){
6623 int pos= random() % COUNT;
6624 while(bitstream[pos] == 0){
6633 nal_length= encode_nal(&h, nal, bitstream, COUNT, COUNT*2);
6635 printf("encoding failed\n");
6639 out= decode_nal(&h, nal, &out_length, &consumed, nal_length);
6643 if(out_length != COUNT){
6644 printf("incorrect length %d %d\n", out_length, COUNT);
6648 if(consumed != nal_length){
6649 printf("incorrect consumed length %d %d\n", nal_length, consumed);
6653 if(memcmp(bitstream, out, COUNT)){
6654 printf("missmatch\n");
6659 printf("Testing RBSP\n");
6667 static int decode_end(AVCodecContext *avctx)
6669 H264Context *h = avctx->priv_data;
6670 MpegEncContext *s = &h->s;
6672 free_tables(h); //FIXME cleanup init stuff perhaps
6675 // memset(h, 0, sizeof(H264Context));
6681 AVCodec h264_decoder = {
6685 sizeof(H264Context),
6690 /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 | CODEC_CAP_TRUNCATED | CODEC_CAP_DELAY,
6693 AVCodecParser h264_parser = {
6695 sizeof(H264Context),