2 * H.26L/H.264/AVC/JVT/14496-10/... encoder/decoder
3 * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
5 * This library is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU Lesser General Public
7 * License as published by the Free Software Foundation; either
8 * version 2 of the License, or (at your option) any later version.
10 * This library is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * Lesser General Public License for more details.
15 * You should have received a copy of the GNU Lesser General Public
16 * License along with this library; if not, write to the Free Software
17 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
23 * H.264 / AVC / MPEG4 part10 codec.
24 * @author Michael Niedermayer <michaelni@gmx.at>
30 #include "mpegvideo.h"
39 #define interlaced_dct interlaced_dct_is_a_bad_name
40 #define mb_intra mb_intra_isnt_initalized_see_mb_type
42 #define LUMA_DC_BLOCK_INDEX 25
43 #define CHROMA_DC_BLOCK_INDEX 26
45 #define CHROMA_DC_COEFF_TOKEN_VLC_BITS 8
46 #define COEFF_TOKEN_VLC_BITS 8
47 #define TOTAL_ZEROS_VLC_BITS 9
48 #define CHROMA_DC_TOTAL_ZEROS_VLC_BITS 3
49 #define RUN_VLC_BITS 3
50 #define RUN7_VLC_BITS 6
52 #define MAX_SPS_COUNT 32
53 #define MAX_PPS_COUNT 256
55 #define MAX_MMCO_COUNT 66
58 * Sequence parameter set
64 int log2_max_frame_num; ///< log2_max_frame_num_minus4 + 4
65 int poc_type; ///< pic_order_cnt_type
66 int log2_max_poc_lsb; ///< log2_max_pic_order_cnt_lsb_minus4
67 int delta_pic_order_always_zero_flag;
68 int offset_for_non_ref_pic;
69 int offset_for_top_to_bottom_field;
70 int poc_cycle_length; ///< num_ref_frames_in_pic_order_cnt_cycle
71 int ref_frame_count; ///< num_ref_frames
72 int gaps_in_frame_num_allowed_flag;
73 int mb_width; ///< frame_width_in_mbs_minus1 + 1
74 int mb_height; ///< frame_height_in_mbs_minus1 + 1
75 int frame_mbs_only_flag;
76 int mb_aff; ///<mb_adaptive_frame_field_flag
77 int direct_8x8_inference_flag;
78 int crop; ///< frame_cropping_flag
79 int crop_left; ///< frame_cropping_rect_left_offset
80 int crop_right; ///< frame_cropping_rect_right_offset
81 int crop_top; ///< frame_cropping_rect_top_offset
82 int crop_bottom; ///< frame_cropping_rect_bottom_offset
83 int vui_parameters_present_flag;
85 int timing_info_present_flag;
86 uint32_t num_units_in_tick;
88 int fixed_frame_rate_flag;
89 short offset_for_ref_frame[256]; //FIXME dyn aloc?
93 * Picture parameter set
97 int cabac; ///< entropy_coding_mode_flag
98 int pic_order_present; ///< pic_order_present_flag
99 int slice_group_count; ///< num_slice_groups_minus1 + 1
100 int mb_slice_group_map_type;
101 int ref_count[2]; ///< num_ref_idx_l0/1_active_minus1 + 1
102 int weighted_pred; ///< weighted_pred_flag
103 int weighted_bipred_idc;
104 int init_qp; ///< pic_init_qp_minus26 + 26
105 int init_qs; ///< pic_init_qs_minus26 + 26
106 int chroma_qp_index_offset;
107 int deblocking_filter_parameters_present; ///< deblocking_filter_parameters_present_flag
108 int constrained_intra_pred; ///< constrained_intra_pred_flag
109 int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag
113 * Memory management control operation opcode.
115 typedef enum MMCOOpcode{
126 * Memory management control operation.
137 typedef struct H264Context{
145 #define NAL_IDR_SLICE 5
149 #define NAL_PICTURE_DELIMITER 9
150 #define NAL_FILTER_DATA 10
151 uint8_t *rbsp_buffer;
152 int rbsp_buffer_size;
155 * Used to parse AVC variant of h264
157 int is_avc; ///< this flag is != 0 if codec is avc1
158 int got_avcC; ///< flag used to parse avcC data only once
159 int nal_length_size; ///< Number of bytes used for nal length (1, 2 or 4)
163 int prev_mb_skiped; //FIXME remove (IMHO not used)
166 int chroma_pred_mode;
167 int intra16x16_pred_mode;
169 int8_t intra4x4_pred_mode_cache[5*8];
170 int8_t (*intra4x4_pred_mode)[8];
171 void (*pred4x4 [9+3])(uint8_t *src, uint8_t *topright, int stride);//FIXME move to dsp?
172 void (*pred8x8 [4+3])(uint8_t *src, int stride);
173 void (*pred16x16[4+3])(uint8_t *src, int stride);
174 unsigned int topleft_samples_available;
175 unsigned int top_samples_available;
176 unsigned int topright_samples_available;
177 unsigned int left_samples_available;
178 uint8_t (*top_border)[16+2*8];
179 uint8_t left_border[17+2*9];
182 * non zero coeff count cache.
183 * is 64 if not available.
185 uint8_t non_zero_count_cache[6*8];
186 uint8_t (*non_zero_count)[16];
189 * Motion vector cache.
191 int16_t mv_cache[2][5*8][2];
192 int8_t ref_cache[2][5*8];
193 #define LIST_NOT_USED -1 //FIXME rename?
194 #define PART_NOT_AVAILABLE -2
197 * is 1 if the specific list MV&references are set to 0,0,-2.
199 int mv_cache_clean[2];
201 int block_offset[16+8];
202 int chroma_subblock_offset[16]; //FIXME remove
204 uint16_t *mb2b_xy; //FIXME are these 4 a good idea?
206 int b_stride; //FIXME use s->b4_stride
212 int unknown_svq3_flag;
213 int next_slice_index;
215 SPS sps_buffer[MAX_SPS_COUNT];
216 SPS sps; ///< current sps
218 PPS pps_buffer[MAX_PPS_COUNT];
222 PPS pps; //FIXME move tp Picture perhaps? (->no) do we need that?
225 uint8_t *slice_table_base;
226 uint8_t *slice_table; ///< slice_table_base + mb_stride + 1
228 int slice_type_fixed;
230 //interlacing specific flags
231 int mb_field_decoding_flag;
238 int delta_poc_bottom;
241 int prev_poc_msb; ///< poc_msb of the last reference pic for POC type 0
242 int prev_poc_lsb; ///< poc_lsb of the last reference pic for POC type 0
243 int frame_num_offset; ///< for POC type 2
244 int prev_frame_num_offset; ///< for POC type 2
245 int prev_frame_num; ///< frame_num of the last pic for POC type 1/2
248 * frame_num for frames or 2*frame_num for field pics.
253 * max_frame_num or 2*max_frame_num for field pics.
257 //Weighted pred stuff
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];
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 //FIXME check what gcc generates for 64 bit on x86 and possible write a 32 bit ver of it
364 *(uint16_t*)(p + stride)= size==4 ? val : val*0x0101;
365 }else if(w==2 && h==4){
366 *(uint16_t*)(p + 0*stride)=
367 *(uint16_t*)(p + 1*stride)=
368 *(uint16_t*)(p + 2*stride)=
369 *(uint16_t*)(p + 3*stride)= size==4 ? val : val*0x0101;
370 }else if(w==4 && h==1){
371 *(uint32_t*)(p + 0*stride)= size==4 ? val : val*0x01010101;
372 }else if(w==4 && h==2){
373 *(uint32_t*)(p + 0*stride)=
374 *(uint32_t*)(p + 1*stride)= size==4 ? val : val*0x01010101;
375 }else if(w==4 && h==4){
376 *(uint32_t*)(p + 0*stride)=
377 *(uint32_t*)(p + 1*stride)=
378 *(uint32_t*)(p + 2*stride)=
379 *(uint32_t*)(p + 3*stride)= size==4 ? val : val*0x01010101;
380 }else if(w==8 && h==1){
382 *(uint32_t*)(p + 4)= size==4 ? val : val*0x01010101;
383 }else if(w==8 && h==2){
384 *(uint32_t*)(p + 0 + 0*stride)=
385 *(uint32_t*)(p + 4 + 0*stride)=
386 *(uint32_t*)(p + 0 + 1*stride)=
387 *(uint32_t*)(p + 4 + 1*stride)= size==4 ? val : val*0x01010101;
388 }else if(w==8 && h==4){
389 *(uint64_t*)(p + 0*stride)=
390 *(uint64_t*)(p + 1*stride)=
391 *(uint64_t*)(p + 2*stride)=
392 *(uint64_t*)(p + 3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
393 }else if(w==16 && h==2){
394 *(uint64_t*)(p + 0+0*stride)=
395 *(uint64_t*)(p + 8+0*stride)=
396 *(uint64_t*)(p + 0+1*stride)=
397 *(uint64_t*)(p + 8+1*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
398 }else if(w==16 && h==4){
399 *(uint64_t*)(p + 0+0*stride)=
400 *(uint64_t*)(p + 8+0*stride)=
401 *(uint64_t*)(p + 0+1*stride)=
402 *(uint64_t*)(p + 8+1*stride)=
403 *(uint64_t*)(p + 0+2*stride)=
404 *(uint64_t*)(p + 8+2*stride)=
405 *(uint64_t*)(p + 0+3*stride)=
406 *(uint64_t*)(p + 8+3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
411 static inline void fill_caches(H264Context *h, int mb_type){
412 MpegEncContext * const s = &h->s;
413 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
414 int topleft_xy, top_xy, topright_xy, left_xy[2];
415 int topleft_type, top_type, topright_type, left_type[2];
419 //wow what a mess, why didnt they simplify the interlacing&intra stuff, i cant imagine that these complex rules are worth it
423 topleft_xy = 0; /* avoid warning */
424 top_xy = 0; /* avoid warning */
425 topright_xy = 0; /* avoid warning */
427 topleft_xy = mb_xy-1 - s->mb_stride;
428 top_xy = mb_xy - s->mb_stride;
429 topright_xy= mb_xy+1 - s->mb_stride;
430 left_xy[0] = mb_xy-1;
431 left_xy[1] = mb_xy-1;
438 topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
439 top_type = h->slice_table[top_xy ] == h->slice_num ? s->current_picture.mb_type[top_xy] : 0;
440 topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
441 left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
442 left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
444 if(IS_INTRA(mb_type)){
445 h->topleft_samples_available=
446 h->top_samples_available=
447 h->left_samples_available= 0xFFFF;
448 h->topright_samples_available= 0xEEEA;
450 if(!IS_INTRA(top_type) && (top_type==0 || h->pps.constrained_intra_pred)){
451 h->topleft_samples_available= 0xB3FF;
452 h->top_samples_available= 0x33FF;
453 h->topright_samples_available= 0x26EA;
456 if(!IS_INTRA(left_type[i]) && (left_type[i]==0 || h->pps.constrained_intra_pred)){
457 h->topleft_samples_available&= 0xDF5F;
458 h->left_samples_available&= 0x5F5F;
462 if(!IS_INTRA(topleft_type) && (topleft_type==0 || h->pps.constrained_intra_pred))
463 h->topleft_samples_available&= 0x7FFF;
465 if(!IS_INTRA(topright_type) && (topright_type==0 || h->pps.constrained_intra_pred))
466 h->topright_samples_available&= 0xFBFF;
468 if(IS_INTRA4x4(mb_type)){
469 if(IS_INTRA4x4(top_type)){
470 h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
471 h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
472 h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
473 h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
476 if(IS_INTRA16x16(top_type) || (IS_INTER(top_type) && !h->pps.constrained_intra_pred))
481 h->intra4x4_pred_mode_cache[4+8*0]=
482 h->intra4x4_pred_mode_cache[5+8*0]=
483 h->intra4x4_pred_mode_cache[6+8*0]=
484 h->intra4x4_pred_mode_cache[7+8*0]= pred;
487 if(IS_INTRA4x4(left_type[i])){
488 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
489 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
492 if(IS_INTRA16x16(left_type[i]) || (IS_INTER(left_type[i]) && !h->pps.constrained_intra_pred))
497 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
498 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
513 //FIXME constraint_intra_pred & partitioning & nnz (lets hope this is just a typo in the spec)
515 h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][0];
516 h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][1];
517 h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][2];
518 h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3];
520 h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][7];
521 h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8];
523 h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][10];
524 h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11];
526 h->top_cbp= h->cbp_table[top_xy];
528 h->non_zero_count_cache[4+8*0]=
529 h->non_zero_count_cache[5+8*0]=
530 h->non_zero_count_cache[6+8*0]=
531 h->non_zero_count_cache[7+8*0]=
533 h->non_zero_count_cache[1+8*0]=
534 h->non_zero_count_cache[2+8*0]=
536 h->non_zero_count_cache[1+8*3]=
537 h->non_zero_count_cache[2+8*3]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
539 if(IS_INTRA(mb_type)) h->top_cbp= 0x1C0;
544 h->non_zero_count_cache[3+8*1]= h->non_zero_count[left_xy[0]][6];
545 h->non_zero_count_cache[3+8*2]= h->non_zero_count[left_xy[0]][5];
546 h->non_zero_count_cache[0+8*1]= h->non_zero_count[left_xy[0]][9]; //FIXME left_block
547 h->non_zero_count_cache[0+8*4]= h->non_zero_count[left_xy[0]][12];
548 h->left_cbp= h->cbp_table[left_xy[0]]; //FIXME interlacing
550 h->non_zero_count_cache[3+8*1]=
551 h->non_zero_count_cache[3+8*2]=
552 h->non_zero_count_cache[0+8*1]=
553 h->non_zero_count_cache[0+8*4]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
555 if(IS_INTRA(mb_type)) h->left_cbp= 0x1C0;//FIXME interlacing
560 h->non_zero_count_cache[3+8*3]= h->non_zero_count[left_xy[1]][4];
561 h->non_zero_count_cache[3+8*4]= h->non_zero_count[left_xy[1]][3];
562 h->non_zero_count_cache[0+8*2]= h->non_zero_count[left_xy[1]][8];
563 h->non_zero_count_cache[0+8*5]= h->non_zero_count[left_xy[1]][11];
565 h->non_zero_count_cache[3+8*3]=
566 h->non_zero_count_cache[3+8*4]=
567 h->non_zero_count_cache[0+8*2]=
568 h->non_zero_count_cache[0+8*5]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
572 //FIXME direct mb can skip much of this
573 if(IS_INTER(mb_type) || (IS_DIRECT(mb_type) && h->direct_spatial_mv_pred)){
575 for(list=0; list<2; list++){
576 if((!IS_8X8(mb_type)) && !USES_LIST(mb_type, list) && !IS_DIRECT(mb_type)){
577 /*if(!h->mv_cache_clean[list]){
578 memset(h->mv_cache [list], 0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
579 memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
580 h->mv_cache_clean[list]= 1;
584 h->mv_cache_clean[list]= 0;
586 if(IS_INTER(topleft_type)){
587 const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
588 const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + h->b8_stride;
589 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
590 h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
592 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
593 h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
596 if(IS_INTER(top_type)){
597 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
598 const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
599 *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0];
600 *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];
601 *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];
602 *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];
603 h->ref_cache[list][scan8[0] + 0 - 1*8]=
604 h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
605 h->ref_cache[list][scan8[0] + 2 - 1*8]=
606 h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
608 *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]=
609 *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]=
610 *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]=
611 *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;
612 *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
615 if(IS_INTER(topright_type)){
616 const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
617 const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
618 *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
619 h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
621 *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
622 h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
625 //FIXME unify cleanup or sth
626 if(IS_INTER(left_type[0])){
627 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
628 const int b8_xy= h->mb2b8_xy[left_xy[0]] + 1;
629 *(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]];
630 *(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]];
631 h->ref_cache[list][scan8[0] - 1 + 0*8]=
632 h->ref_cache[list][scan8[0] - 1 + 1*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0]>>1)];
634 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 0*8]=
635 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 1*8]= 0;
636 h->ref_cache[list][scan8[0] - 1 + 0*8]=
637 h->ref_cache[list][scan8[0] - 1 + 1*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
640 if(IS_INTER(left_type[1])){
641 const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
642 const int b8_xy= h->mb2b8_xy[left_xy[1]] + 1;
643 *(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]];
644 *(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]];
645 h->ref_cache[list][scan8[0] - 1 + 2*8]=
646 h->ref_cache[list][scan8[0] - 1 + 3*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[2]>>1)];
648 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 2*8]=
649 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 3*8]= 0;
650 h->ref_cache[list][scan8[0] - 1 + 2*8]=
651 h->ref_cache[list][scan8[0] - 1 + 3*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
654 h->ref_cache[list][scan8[5 ]+1] =
655 h->ref_cache[list][scan8[7 ]+1] =
656 h->ref_cache[list][scan8[13]+1] = //FIXME remove past 3 (init somewher else)
657 h->ref_cache[list][scan8[4 ]] =
658 h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
659 *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
660 *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
661 *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewher else)
662 *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
663 *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
666 /* XXX beurk, Load mvd */
667 if(IS_INTER(topleft_type)){
668 const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
669 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy];
671 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 - 1*8]= 0;
674 if(IS_INTER(top_type)){
675 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
676 *(uint32_t*)h->mvd_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 0];
677 *(uint32_t*)h->mvd_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 1];
678 *(uint32_t*)h->mvd_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 2];
679 *(uint32_t*)h->mvd_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 3];
681 *(uint32_t*)h->mvd_cache [list][scan8[0] + 0 - 1*8]=
682 *(uint32_t*)h->mvd_cache [list][scan8[0] + 1 - 1*8]=
683 *(uint32_t*)h->mvd_cache [list][scan8[0] + 2 - 1*8]=
684 *(uint32_t*)h->mvd_cache [list][scan8[0] + 3 - 1*8]= 0;
686 if(IS_INTER(left_type[0])){
687 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
688 *(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]];
689 *(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]];
691 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 0*8]=
692 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 1*8]= 0;
694 if(IS_INTER(left_type[1])){
695 const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
696 *(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]];
697 *(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]];
699 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 2*8]=
700 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 3*8]= 0;
702 *(uint32_t*)h->mvd_cache [list][scan8[5 ]+1]=
703 *(uint32_t*)h->mvd_cache [list][scan8[7 ]+1]=
704 *(uint32_t*)h->mvd_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewher else)
705 *(uint32_t*)h->mvd_cache [list][scan8[4 ]]=
706 *(uint32_t*)h->mvd_cache [list][scan8[12]]= 0;
708 if(h->slice_type == B_TYPE){
709 fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, 0, 1);
711 if(IS_DIRECT(top_type)){
712 *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0x01010101;
713 }else if(IS_8X8(top_type)){
714 int b8_xy = h->mb2b8_xy[top_xy] + h->b8_stride;
715 h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy];
716 h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 1];
718 *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0;
722 if(IS_DIRECT(left_type[0])){
723 h->direct_cache[scan8[0] - 1 + 0*8]=
724 h->direct_cache[scan8[0] - 1 + 2*8]= 1;
725 }else if(IS_8X8(left_type[0])){
726 int b8_xy = h->mb2b8_xy[left_xy[0]] + 1;
727 h->direct_cache[scan8[0] - 1 + 0*8]= h->direct_table[b8_xy];
728 h->direct_cache[scan8[0] - 1 + 2*8]= h->direct_table[b8_xy + h->b8_stride];
730 h->direct_cache[scan8[0] - 1 + 0*8]=
731 h->direct_cache[scan8[0] - 1 + 2*8]= 0;
740 static inline void write_back_intra_pred_mode(H264Context *h){
741 MpegEncContext * const s = &h->s;
742 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
744 h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
745 h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
746 h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
747 h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
748 h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
749 h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
750 h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
754 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
756 static inline int check_intra4x4_pred_mode(H264Context *h){
757 MpegEncContext * const s = &h->s;
758 static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
759 static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
762 if(!(h->top_samples_available&0x8000)){
764 int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
766 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);
769 h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
774 if(!(h->left_samples_available&0x8000)){
776 int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
778 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);
781 h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
787 } //FIXME cleanup like next
790 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
792 static inline int check_intra_pred_mode(H264Context *h, int mode){
793 MpegEncContext * const s = &h->s;
794 static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
795 static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
797 if(mode < 0 || mode > 6) {
798 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);
802 if(!(h->top_samples_available&0x8000)){
805 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);
810 if(!(h->left_samples_available&0x8000)){
813 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);
822 * gets the predicted intra4x4 prediction mode.
824 static inline int pred_intra_mode(H264Context *h, int n){
825 const int index8= scan8[n];
826 const int left= h->intra4x4_pred_mode_cache[index8 - 1];
827 const int top = h->intra4x4_pred_mode_cache[index8 - 8];
828 const int min= FFMIN(left, top);
830 tprintf("mode:%d %d min:%d\n", left ,top, min);
832 if(min<0) return DC_PRED;
836 static inline void write_back_non_zero_count(H264Context *h){
837 MpegEncContext * const s = &h->s;
838 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
840 h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[4+8*4];
841 h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[5+8*4];
842 h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[6+8*4];
843 h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];
844 h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[7+8*3];
845 h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[7+8*2];
846 h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[7+8*1];
848 h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[1+8*2];
849 h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
850 h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[2+8*1];
852 h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[1+8*5];
853 h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
854 h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[2+8*4];
858 * gets the predicted number of non zero coefficients.
859 * @param n block index
861 static inline int pred_non_zero_count(H264Context *h, int n){
862 const int index8= scan8[n];
863 const int left= h->non_zero_count_cache[index8 - 1];
864 const int top = h->non_zero_count_cache[index8 - 8];
867 if(i<64) i= (i+1)>>1;
869 tprintf("pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
874 static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
875 const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];
877 if(topright_ref != PART_NOT_AVAILABLE){
878 *C= h->mv_cache[list][ i - 8 + part_width ];
881 tprintf("topright MV not available\n");
883 *C= h->mv_cache[list][ i - 8 - 1 ];
884 return h->ref_cache[list][ i - 8 - 1 ];
889 * gets the predicted MV.
890 * @param n the block index
891 * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)
892 * @param mx the x component of the predicted motion vector
893 * @param my the y component of the predicted motion vector
895 static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
896 const int index8= scan8[n];
897 const int top_ref= h->ref_cache[list][ index8 - 8 ];
898 const int left_ref= h->ref_cache[list][ index8 - 1 ];
899 const int16_t * const A= h->mv_cache[list][ index8 - 1 ];
900 const int16_t * const B= h->mv_cache[list][ index8 - 8 ];
902 int diagonal_ref, match_count;
904 assert(part_width==1 || part_width==2 || part_width==4);
914 diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);
915 match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);
916 if(match_count > 1){ //most common
917 *mx= mid_pred(A[0], B[0], C[0]);
918 *my= mid_pred(A[1], B[1], C[1]);
919 }else if(match_count==1){
923 }else if(top_ref==ref){
931 if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
935 *mx= mid_pred(A[0], B[0], C[0]);
936 *my= mid_pred(A[1], B[1], C[1]);
940 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);
944 * gets the directionally predicted 16x8 MV.
945 * @param n the block index
946 * @param mx the x component of the predicted motion vector
947 * @param my the y component of the predicted motion vector
949 static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
951 const int top_ref= h->ref_cache[list][ scan8[0] - 8 ];
952 const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
954 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);
962 const int left_ref= h->ref_cache[list][ scan8[8] - 1 ];
963 const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];
965 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);
975 pred_motion(h, n, 4, list, ref, mx, my);
979 * gets the directionally predicted 8x16 MV.
980 * @param n the block index
981 * @param mx the x component of the predicted motion vector
982 * @param my the y component of the predicted motion vector
984 static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
986 const int left_ref= h->ref_cache[list][ scan8[0] - 1 ];
987 const int16_t * const A= h->mv_cache[list][ scan8[0] - 1 ];
989 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);
1000 diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);
1002 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);
1004 if(diagonal_ref == ref){
1012 pred_motion(h, n, 2, list, ref, mx, my);
1015 static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
1016 const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
1017 const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];
1019 tprintf("pred_pskip: (%d) (%d) at %2d %2d\n", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
1021 if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
1022 || (top_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0)
1023 || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){
1029 pred_motion(h, 0, 4, 0, 0, mx, my);
1034 static inline void direct_dist_scale_factor(H264Context * const h){
1035 const int poc = h->s.current_picture_ptr->poc;
1036 const int poc1 = h->ref_list[1][0].poc;
1038 for(i=0; i<h->ref_count[0]; i++){
1039 int poc0 = h->ref_list[0][i].poc;
1040 int td = clip(poc1 - poc0, -128, 127);
1041 if(td == 0 /* FIXME || pic0 is a long-term ref */){
1042 h->dist_scale_factor[i] = 256;
1044 int tb = clip(poc - poc0, -128, 127);
1045 int tx = (16384 + (ABS(td) >> 1)) / td;
1046 h->dist_scale_factor[i] = clip((tb*tx + 32) >> 6, -1024, 1023);
1051 static inline void pred_direct_motion(H264Context * const h, int *mb_type){
1052 MpegEncContext * const s = &h->s;
1053 const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
1054 const int b8_xy = 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1055 const int b4_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1056 const int mb_type_col = h->ref_list[1][0].mb_type[mb_xy];
1057 const int16_t (*l1mv0)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[0][b4_xy];
1058 const int8_t *l1ref0 = &h->ref_list[1][0].ref_index[0][b8_xy];
1059 const int is_b8x8 = IS_8X8(*mb_type);
1063 if(IS_8X8(mb_type_col) && !h->sps.direct_8x8_inference_flag){
1064 /* FIXME save sub mb types from previous frames (or derive from MVs)
1065 * so we know exactly what block size to use */
1066 sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */
1067 *mb_type = MB_TYPE_8x8;
1068 }else if(!is_b8x8 && (IS_16X16(mb_type_col) || IS_INTRA(mb_type_col))){
1069 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1070 *mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */
1072 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1073 *mb_type = MB_TYPE_8x8;
1076 *mb_type |= MB_TYPE_DIRECT2;
1078 if(h->direct_spatial_mv_pred){
1083 /* ref = min(neighbors) */
1084 for(list=0; list<2; list++){
1085 int refa = h->ref_cache[list][scan8[0] - 1];
1086 int refb = h->ref_cache[list][scan8[0] - 8];
1087 int refc = h->ref_cache[list][scan8[0] - 8 + 4];
1089 refc = h->ref_cache[list][scan8[0] - 8 - 1];
1091 if(ref[list] < 0 || (refb < ref[list] && refb >= 0))
1093 if(ref[list] < 0 || (refc < ref[list] && refc >= 0))
1099 if(ref[0] < 0 && ref[1] < 0){
1100 ref[0] = ref[1] = 0;
1101 mv[0][0] = mv[0][1] =
1102 mv[1][0] = mv[1][1] = 0;
1104 for(list=0; list<2; list++){
1106 pred_motion(h, 0, 4, list, ref[list], &mv[list][0], &mv[list][1]);
1108 mv[list][0] = mv[list][1] = 0;
1113 *mb_type &= ~MB_TYPE_P0L1;
1114 sub_mb_type &= ~MB_TYPE_P0L1;
1115 }else if(ref[0] < 0){
1116 *mb_type &= ~MB_TYPE_P0L0;
1117 sub_mb_type &= ~MB_TYPE_P0L0;
1120 if(IS_16X16(*mb_type)){
1121 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref[0], 1);
1122 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, ref[1], 1);
1123 if(!IS_INTRA(mb_type_col) && l1ref0[0] == 0 &&
1124 ABS(l1mv0[0][0]) <= 1 && ABS(l1mv0[0][1]) <= 1){
1126 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1128 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
1130 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1132 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
1134 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1135 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1138 for(i8=0; i8<4; i8++){
1139 const int x8 = i8&1;
1140 const int y8 = i8>>1;
1142 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1144 h->sub_mb_type[i8] = sub_mb_type;
1146 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1147 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1148 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref[0], 1);
1149 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, ref[1], 1);
1152 if(!IS_INTRA(mb_type_col) && l1ref0[x8 + y8*h->b8_stride] == 0){
1153 for(i4=0; i4<4; i4++){
1154 const int16_t *mv_col = l1mv0[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1155 if(ABS(mv_col[0]) <= 1 && ABS(mv_col[1]) <= 1){
1157 *(uint32_t*)h->mv_cache[0][scan8[i8*4+i4]] = 0;
1159 *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = 0;
1165 }else{ /* direct temporal mv pred */
1166 /* FIXME assumes that L1ref0 used the same ref lists as current frame */
1167 if(IS_16X16(*mb_type)){
1168 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1);
1169 if(IS_INTRA(mb_type_col)){
1170 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
1171 fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
1172 fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
1174 const int ref0 = l1ref0[0];
1175 const int dist_scale_factor = h->dist_scale_factor[ref0];
1176 const int16_t *mv_col = l1mv0[0];
1178 mv_l0[0] = (dist_scale_factor * mv_col[0] + 128) >> 8;
1179 mv_l0[1] = (dist_scale_factor * mv_col[1] + 128) >> 8;
1180 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref0, 1);
1181 fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv_l0[0],mv_l0[1]), 4);
1182 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);
1185 for(i8=0; i8<4; i8++){
1186 const int x8 = i8&1;
1187 const int y8 = i8>>1;
1188 int ref0, dist_scale_factor;
1190 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1192 h->sub_mb_type[i8] = sub_mb_type;
1193 if(IS_INTRA(mb_type_col)){
1194 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1195 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1196 fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1197 fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1201 ref0 = l1ref0[x8 + y8*h->b8_stride];
1202 dist_scale_factor = h->dist_scale_factor[ref0];
1204 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1205 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1206 for(i4=0; i4<4; i4++){
1207 const int16_t *mv_col = l1mv0[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1208 int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]];
1209 mv_l0[0] = (dist_scale_factor * mv_col[0] + 128) >> 8;
1210 mv_l0[1] = (dist_scale_factor * mv_col[1] + 128) >> 8;
1211 *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] =
1212 pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
1219 static inline void write_back_motion(H264Context *h, int mb_type){
1220 MpegEncContext * const s = &h->s;
1221 const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1222 const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1225 for(list=0; list<2; list++){
1227 if((!IS_8X8(mb_type)) && !USES_LIST(mb_type, list)){
1228 if(1){ //FIXME skip or never read if mb_type doesnt use it
1230 *(uint64_t*)s->current_picture.motion_val[list][b_xy + 0 + y*h->b_stride]=
1231 *(uint64_t*)s->current_picture.motion_val[list][b_xy + 2 + y*h->b_stride]= 0;
1233 if( h->pps.cabac ) {
1234 /* FIXME needed ? */
1236 *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]=
1237 *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= 0;
1241 *(uint16_t*)&s->current_picture.ref_index[list][b8_xy + y*h->b8_stride]= (LIST_NOT_USED&0xFF)*0x0101;
1248 *(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];
1249 *(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];
1251 if( h->pps.cabac ) {
1253 *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+0 + 8*y];
1254 *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+2 + 8*y];
1258 s->current_picture.ref_index[list][b8_xy + 0 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+0 + 16*y];
1259 s->current_picture.ref_index[list][b8_xy + 1 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+2 + 16*y];
1263 if(h->slice_type == B_TYPE && h->pps.cabac){
1264 if(IS_8X8(mb_type)){
1265 h->direct_table[b8_xy+1+0*h->b8_stride] = IS_DIRECT(h->sub_mb_type[1]) ? 1 : 0;
1266 h->direct_table[b8_xy+0+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[2]) ? 1 : 0;
1267 h->direct_table[b8_xy+1+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[3]) ? 1 : 0;
1273 * Decodes a network abstraction layer unit.
1274 * @param consumed is the number of bytes used as input
1275 * @param length is the length of the array
1276 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp ttailing?
1277 * @returns decoded bytes, might be src+1 if no escapes
1279 static uint8_t *decode_nal(H264Context *h, uint8_t *src, int *dst_length, int *consumed, int length){
1283 // src[0]&0x80; //forbidden bit
1284 h->nal_ref_idc= src[0]>>5;
1285 h->nal_unit_type= src[0]&0x1F;
1289 for(i=0; i<length; i++)
1290 printf("%2X ", src[i]);
1292 for(i=0; i+1<length; i+=2){
1293 if(src[i]) continue;
1294 if(i>0 && src[i-1]==0) i--;
1295 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1297 /* startcode, so we must be past the end */
1304 if(i>=length-1){ //no escaped 0
1305 *dst_length= length;
1306 *consumed= length+1; //+1 for the header
1310 h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length);
1311 dst= h->rbsp_buffer;
1313 //printf("deoding esc\n");
1316 //remove escapes (very rare 1:2^22)
1317 if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1318 if(src[si+2]==3){ //escape
1323 }else //next start code
1327 dst[di++]= src[si++];
1331 *consumed= si + 1;//+1 for the header
1332 //FIXME store exact number of bits in the getbitcontext (its needed for decoding)
1338 * @param src the data which should be escaped
1339 * @param dst the target buffer, dst+1 == src is allowed as a special case
1340 * @param length the length of the src data
1341 * @param dst_length the length of the dst array
1342 * @returns length of escaped data in bytes or -1 if an error occured
1344 static int encode_nal(H264Context *h, uint8_t *dst, uint8_t *src, int length, int dst_length){
1345 int i, escape_count, si, di;
1349 assert(dst_length>0);
1351 dst[0]= (h->nal_ref_idc<<5) + h->nal_unit_type;
1353 if(length==0) return 1;
1356 for(i=0; i<length; i+=2){
1357 if(src[i]) continue;
1358 if(i>0 && src[i-1]==0)
1360 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1366 if(escape_count==0){
1368 memcpy(dst+1, src, length);
1372 if(length + escape_count + 1> dst_length)
1375 //this should be damn rare (hopefully)
1377 h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length + escape_count);
1378 temp= h->rbsp_buffer;
1379 //printf("encoding esc\n");
1384 if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1385 temp[di++]= 0; si++;
1386 temp[di++]= 0; si++;
1388 temp[di++]= src[si++];
1391 temp[di++]= src[si++];
1393 memcpy(dst+1, temp, length+escape_count);
1395 assert(di == length+escape_count);
1401 * write 1,10,100,1000,... for alignment, yes its exactly inverse to mpeg4
1403 static void encode_rbsp_trailing(PutBitContext *pb){
1406 length= (-put_bits_count(pb))&7;
1407 if(length) put_bits(pb, length, 0);
1412 * identifies the exact end of the bitstream
1413 * @return the length of the trailing, or 0 if damaged
1415 static int decode_rbsp_trailing(uint8_t *src){
1419 tprintf("rbsp trailing %X\n", v);
1429 * idct tranforms the 16 dc values and dequantize them.
1430 * @param qp quantization parameter
1432 static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp){
1433 const int qmul= dequant_coeff[qp][0];
1436 int temp[16]; //FIXME check if this is a good idea
1437 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
1438 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1440 //memset(block, 64, 2*256);
1443 const int offset= y_offset[i];
1444 const int z0= block[offset+stride*0] + block[offset+stride*4];
1445 const int z1= block[offset+stride*0] - block[offset+stride*4];
1446 const int z2= block[offset+stride*1] - block[offset+stride*5];
1447 const int z3= block[offset+stride*1] + block[offset+stride*5];
1456 const int offset= x_offset[i];
1457 const int z0= temp[4*0+i] + temp[4*2+i];
1458 const int z1= temp[4*0+i] - temp[4*2+i];
1459 const int z2= temp[4*1+i] - temp[4*3+i];
1460 const int z3= temp[4*1+i] + temp[4*3+i];
1462 block[stride*0 +offset]= ((z0 + z3)*qmul + 2)>>2; //FIXME think about merging this into decode_resdual
1463 block[stride*2 +offset]= ((z1 + z2)*qmul + 2)>>2;
1464 block[stride*8 +offset]= ((z1 - z2)*qmul + 2)>>2;
1465 block[stride*10+offset]= ((z0 - z3)*qmul + 2)>>2;
1471 * dct tranforms the 16 dc values.
1472 * @param qp quantization parameter ??? FIXME
1474 static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
1475 // const int qmul= dequant_coeff[qp][0];
1477 int temp[16]; //FIXME check if this is a good idea
1478 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
1479 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1482 const int offset= y_offset[i];
1483 const int z0= block[offset+stride*0] + block[offset+stride*4];
1484 const int z1= block[offset+stride*0] - block[offset+stride*4];
1485 const int z2= block[offset+stride*1] - block[offset+stride*5];
1486 const int z3= block[offset+stride*1] + block[offset+stride*5];
1495 const int offset= x_offset[i];
1496 const int z0= temp[4*0+i] + temp[4*2+i];
1497 const int z1= temp[4*0+i] - temp[4*2+i];
1498 const int z2= temp[4*1+i] - temp[4*3+i];
1499 const int z3= temp[4*1+i] + temp[4*3+i];
1501 block[stride*0 +offset]= (z0 + z3)>>1;
1502 block[stride*2 +offset]= (z1 + z2)>>1;
1503 block[stride*8 +offset]= (z1 - z2)>>1;
1504 block[stride*10+offset]= (z0 - z3)>>1;
1512 static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp){
1513 const int qmul= dequant_coeff[qp][0];
1514 const int stride= 16*2;
1515 const int xStride= 16;
1518 a= block[stride*0 + xStride*0];
1519 b= block[stride*0 + xStride*1];
1520 c= block[stride*1 + xStride*0];
1521 d= block[stride*1 + xStride*1];
1528 block[stride*0 + xStride*0]= ((a+c)*qmul + 0)>>1;
1529 block[stride*0 + xStride*1]= ((e+b)*qmul + 0)>>1;
1530 block[stride*1 + xStride*0]= ((a-c)*qmul + 0)>>1;
1531 block[stride*1 + xStride*1]= ((e-b)*qmul + 0)>>1;
1535 static void chroma_dc_dct_c(DCTELEM *block){
1536 const int stride= 16*2;
1537 const int xStride= 16;
1540 a= block[stride*0 + xStride*0];
1541 b= block[stride*0 + xStride*1];
1542 c= block[stride*1 + xStride*0];
1543 d= block[stride*1 + xStride*1];
1550 block[stride*0 + xStride*0]= (a+c);
1551 block[stride*0 + xStride*1]= (e+b);
1552 block[stride*1 + xStride*0]= (a-c);
1553 block[stride*1 + xStride*1]= (e-b);
1558 * gets the chroma qp.
1560 static inline int get_chroma_qp(H264Context *h, int qscale){
1562 return chroma_qp[clip(qscale + h->pps.chroma_qp_index_offset, 0, 51)];
1567 static void h264_diff_dct_c(DCTELEM *block, uint8_t *src1, uint8_t *src2, int stride){
1569 //FIXME try int temp instead of block
1572 const int d0= src1[0 + i*stride] - src2[0 + i*stride];
1573 const int d1= src1[1 + i*stride] - src2[1 + i*stride];
1574 const int d2= src1[2 + i*stride] - src2[2 + i*stride];
1575 const int d3= src1[3 + i*stride] - src2[3 + i*stride];
1576 const int z0= d0 + d3;
1577 const int z3= d0 - d3;
1578 const int z1= d1 + d2;
1579 const int z2= d1 - d2;
1581 block[0 + 4*i]= z0 + z1;
1582 block[1 + 4*i]= 2*z3 + z2;
1583 block[2 + 4*i]= z0 - z1;
1584 block[3 + 4*i]= z3 - 2*z2;
1588 const int z0= block[0*4 + i] + block[3*4 + i];
1589 const int z3= block[0*4 + i] - block[3*4 + i];
1590 const int z1= block[1*4 + i] + block[2*4 + i];
1591 const int z2= block[1*4 + i] - block[2*4 + i];
1593 block[0*4 + i]= z0 + z1;
1594 block[1*4 + i]= 2*z3 + z2;
1595 block[2*4 + i]= z0 - z1;
1596 block[3*4 + i]= z3 - 2*z2;
1601 //FIXME need to check that this doesnt overflow signed 32 bit for low qp, iam not sure, its very close
1602 //FIXME check that gcc inlines this (and optimizes intra & seperate_dc stuff away)
1603 static inline int quantize_c(DCTELEM *block, uint8_t *scantable, int qscale, int intra, int seperate_dc){
1605 const int * const quant_table= quant_coeff[qscale];
1606 const int bias= intra ? (1<<QUANT_SHIFT)/3 : (1<<QUANT_SHIFT)/6;
1607 const unsigned int threshold1= (1<<QUANT_SHIFT) - bias - 1;
1608 const unsigned int threshold2= (threshold1<<1);
1614 const int dc_bias= intra ? (1<<(QUANT_SHIFT-2))/3 : (1<<(QUANT_SHIFT-2))/6;
1615 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT-2)) - dc_bias - 1;
1616 const unsigned int dc_threshold2= (dc_threshold1<<1);
1618 int level= block[0]*quant_coeff[qscale+18][0];
1619 if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1621 level= (dc_bias + level)>>(QUANT_SHIFT-2);
1624 level= (dc_bias - level)>>(QUANT_SHIFT-2);
1627 // last_non_zero = i;
1632 const int dc_bias= intra ? (1<<(QUANT_SHIFT+1))/3 : (1<<(QUANT_SHIFT+1))/6;
1633 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT+1)) - dc_bias - 1;
1634 const unsigned int dc_threshold2= (dc_threshold1<<1);
1636 int level= block[0]*quant_table[0];
1637 if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1639 level= (dc_bias + level)>>(QUANT_SHIFT+1);
1642 level= (dc_bias - level)>>(QUANT_SHIFT+1);
1645 // last_non_zero = i;
1658 const int j= scantable[i];
1659 int level= block[j]*quant_table[j];
1661 // if( bias+level >= (1<<(QMAT_SHIFT - 3))
1662 // || bias-level >= (1<<(QMAT_SHIFT - 3))){
1663 if(((unsigned)(level+threshold1))>threshold2){
1665 level= (bias + level)>>QUANT_SHIFT;
1668 level= (bias - level)>>QUANT_SHIFT;
1677 return last_non_zero;
1680 static void pred4x4_vertical_c(uint8_t *src, uint8_t *topright, int stride){
1681 const uint32_t a= ((uint32_t*)(src-stride))[0];
1682 ((uint32_t*)(src+0*stride))[0]= a;
1683 ((uint32_t*)(src+1*stride))[0]= a;
1684 ((uint32_t*)(src+2*stride))[0]= a;
1685 ((uint32_t*)(src+3*stride))[0]= a;
1688 static void pred4x4_horizontal_c(uint8_t *src, uint8_t *topright, int stride){
1689 ((uint32_t*)(src+0*stride))[0]= src[-1+0*stride]*0x01010101;
1690 ((uint32_t*)(src+1*stride))[0]= src[-1+1*stride]*0x01010101;
1691 ((uint32_t*)(src+2*stride))[0]= src[-1+2*stride]*0x01010101;
1692 ((uint32_t*)(src+3*stride))[0]= src[-1+3*stride]*0x01010101;
1695 static void pred4x4_dc_c(uint8_t *src, uint8_t *topright, int stride){
1696 const int dc= ( src[-stride] + src[1-stride] + src[2-stride] + src[3-stride]
1697 + src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 4) >>3;
1699 ((uint32_t*)(src+0*stride))[0]=
1700 ((uint32_t*)(src+1*stride))[0]=
1701 ((uint32_t*)(src+2*stride))[0]=
1702 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1705 static void pred4x4_left_dc_c(uint8_t *src, uint8_t *topright, int stride){
1706 const int dc= ( src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 2) >>2;
1708 ((uint32_t*)(src+0*stride))[0]=
1709 ((uint32_t*)(src+1*stride))[0]=
1710 ((uint32_t*)(src+2*stride))[0]=
1711 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1714 static void pred4x4_top_dc_c(uint8_t *src, uint8_t *topright, int stride){
1715 const int dc= ( src[-stride] + src[1-stride] + src[2-stride] + src[3-stride] + 2) >>2;
1717 ((uint32_t*)(src+0*stride))[0]=
1718 ((uint32_t*)(src+1*stride))[0]=
1719 ((uint32_t*)(src+2*stride))[0]=
1720 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1723 static void pred4x4_128_dc_c(uint8_t *src, uint8_t *topright, int stride){
1724 ((uint32_t*)(src+0*stride))[0]=
1725 ((uint32_t*)(src+1*stride))[0]=
1726 ((uint32_t*)(src+2*stride))[0]=
1727 ((uint32_t*)(src+3*stride))[0]= 128U*0x01010101U;
1731 #define LOAD_TOP_RIGHT_EDGE\
1732 const int t4= topright[0];\
1733 const int t5= topright[1];\
1734 const int t6= topright[2];\
1735 const int t7= topright[3];\
1737 #define LOAD_LEFT_EDGE\
1738 const int l0= src[-1+0*stride];\
1739 const int l1= src[-1+1*stride];\
1740 const int l2= src[-1+2*stride];\
1741 const int l3= src[-1+3*stride];\
1743 #define LOAD_TOP_EDGE\
1744 const int t0= src[ 0-1*stride];\
1745 const int t1= src[ 1-1*stride];\
1746 const int t2= src[ 2-1*stride];\
1747 const int t3= src[ 3-1*stride];\
1749 static void pred4x4_down_right_c(uint8_t *src, uint8_t *topright, int stride){
1750 const int lt= src[-1-1*stride];
1754 src[0+3*stride]=(l3 + 2*l2 + l1 + 2)>>2;
1756 src[1+3*stride]=(l2 + 2*l1 + l0 + 2)>>2;
1759 src[2+3*stride]=(l1 + 2*l0 + lt + 2)>>2;
1763 src[3+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1766 src[3+2*stride]=(lt + 2*t0 + t1 + 2)>>2;
1768 src[3+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1769 src[3+0*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1772 static void pred4x4_down_left_c(uint8_t *src, uint8_t *topright, int stride){
1777 src[0+0*stride]=(t0 + t2 + 2*t1 + 2)>>2;
1779 src[0+1*stride]=(t1 + t3 + 2*t2 + 2)>>2;
1782 src[0+2*stride]=(t2 + t4 + 2*t3 + 2)>>2;
1786 src[0+3*stride]=(t3 + t5 + 2*t4 + 2)>>2;
1789 src[1+3*stride]=(t4 + t6 + 2*t5 + 2)>>2;
1791 src[2+3*stride]=(t5 + t7 + 2*t6 + 2)>>2;
1792 src[3+3*stride]=(t6 + 3*t7 + 2)>>2;
1795 static void pred4x4_vertical_right_c(uint8_t *src, uint8_t *topright, int stride){
1796 const int lt= src[-1-1*stride];
1799 const __attribute__((unused)) int unu= l3;
1802 src[1+2*stride]=(lt + t0 + 1)>>1;
1804 src[2+2*stride]=(t0 + t1 + 1)>>1;
1806 src[3+2*stride]=(t1 + t2 + 1)>>1;
1807 src[3+0*stride]=(t2 + t3 + 1)>>1;
1809 src[1+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1811 src[2+3*stride]=(lt + 2*t0 + t1 + 2)>>2;
1813 src[3+3*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1814 src[3+1*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1815 src[0+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1816 src[0+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1819 static void pred4x4_vertical_left_c(uint8_t *src, uint8_t *topright, int stride){
1822 const __attribute__((unused)) int unu= t7;
1824 src[0+0*stride]=(t0 + t1 + 1)>>1;
1826 src[0+2*stride]=(t1 + t2 + 1)>>1;
1828 src[1+2*stride]=(t2 + t3 + 1)>>1;
1830 src[2+2*stride]=(t3 + t4+ 1)>>1;
1831 src[3+2*stride]=(t4 + t5+ 1)>>1;
1832 src[0+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1834 src[0+3*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1836 src[1+3*stride]=(t2 + 2*t3 + t4 + 2)>>2;
1838 src[2+3*stride]=(t3 + 2*t4 + t5 + 2)>>2;
1839 src[3+3*stride]=(t4 + 2*t5 + t6 + 2)>>2;
1842 static void pred4x4_horizontal_up_c(uint8_t *src, uint8_t *topright, int stride){
1845 src[0+0*stride]=(l0 + l1 + 1)>>1;
1846 src[1+0*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1848 src[0+1*stride]=(l1 + l2 + 1)>>1;
1850 src[1+1*stride]=(l1 + 2*l2 + l3 + 2)>>2;
1852 src[0+2*stride]=(l2 + l3 + 1)>>1;
1854 src[1+2*stride]=(l2 + 2*l3 + l3 + 2)>>2;
1863 static void pred4x4_horizontal_down_c(uint8_t *src, uint8_t *topright, int stride){
1864 const int lt= src[-1-1*stride];
1867 const __attribute__((unused)) int unu= t3;
1870 src[2+1*stride]=(lt + l0 + 1)>>1;
1872 src[3+1*stride]=(l0 + 2*lt + t0 + 2)>>2;
1873 src[2+0*stride]=(lt + 2*t0 + t1 + 2)>>2;
1874 src[3+0*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1876 src[2+2*stride]=(l0 + l1 + 1)>>1;
1878 src[3+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1880 src[2+3*stride]=(l1 + l2+ 1)>>1;
1882 src[3+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1883 src[0+3*stride]=(l2 + l3 + 1)>>1;
1884 src[1+3*stride]=(l1 + 2*l2 + l3 + 2)>>2;
1887 static void pred16x16_vertical_c(uint8_t *src, int stride){
1889 const uint32_t a= ((uint32_t*)(src-stride))[0];
1890 const uint32_t b= ((uint32_t*)(src-stride))[1];
1891 const uint32_t c= ((uint32_t*)(src-stride))[2];
1892 const uint32_t d= ((uint32_t*)(src-stride))[3];
1894 for(i=0; i<16; i++){
1895 ((uint32_t*)(src+i*stride))[0]= a;
1896 ((uint32_t*)(src+i*stride))[1]= b;
1897 ((uint32_t*)(src+i*stride))[2]= c;
1898 ((uint32_t*)(src+i*stride))[3]= d;
1902 static void pred16x16_horizontal_c(uint8_t *src, int stride){
1905 for(i=0; i<16; i++){
1906 ((uint32_t*)(src+i*stride))[0]=
1907 ((uint32_t*)(src+i*stride))[1]=
1908 ((uint32_t*)(src+i*stride))[2]=
1909 ((uint32_t*)(src+i*stride))[3]= src[-1+i*stride]*0x01010101;
1913 static void pred16x16_dc_c(uint8_t *src, int stride){
1917 dc+= src[-1+i*stride];
1924 dc= 0x01010101*((dc + 16)>>5);
1926 for(i=0; i<16; i++){
1927 ((uint32_t*)(src+i*stride))[0]=
1928 ((uint32_t*)(src+i*stride))[1]=
1929 ((uint32_t*)(src+i*stride))[2]=
1930 ((uint32_t*)(src+i*stride))[3]= dc;
1934 static void pred16x16_left_dc_c(uint8_t *src, int stride){
1938 dc+= src[-1+i*stride];
1941 dc= 0x01010101*((dc + 8)>>4);
1943 for(i=0; i<16; i++){
1944 ((uint32_t*)(src+i*stride))[0]=
1945 ((uint32_t*)(src+i*stride))[1]=
1946 ((uint32_t*)(src+i*stride))[2]=
1947 ((uint32_t*)(src+i*stride))[3]= dc;
1951 static void pred16x16_top_dc_c(uint8_t *src, int stride){
1957 dc= 0x01010101*((dc + 8)>>4);
1959 for(i=0; i<16; i++){
1960 ((uint32_t*)(src+i*stride))[0]=
1961 ((uint32_t*)(src+i*stride))[1]=
1962 ((uint32_t*)(src+i*stride))[2]=
1963 ((uint32_t*)(src+i*stride))[3]= dc;
1967 static void pred16x16_128_dc_c(uint8_t *src, int stride){
1970 for(i=0; i<16; i++){
1971 ((uint32_t*)(src+i*stride))[0]=
1972 ((uint32_t*)(src+i*stride))[1]=
1973 ((uint32_t*)(src+i*stride))[2]=
1974 ((uint32_t*)(src+i*stride))[3]= 0x01010101U*128U;
1978 static inline void pred16x16_plane_compat_c(uint8_t *src, int stride, const int svq3){
1981 uint8_t *cm = cropTbl + MAX_NEG_CROP;
1982 const uint8_t * const src0 = src+7-stride;
1983 const uint8_t *src1 = src+8*stride-1;
1984 const uint8_t *src2 = src1-2*stride; // == src+6*stride-1;
1985 int H = src0[1] - src0[-1];
1986 int V = src1[0] - src2[ 0];
1987 for(k=2; k<=8; ++k) {
1988 src1 += stride; src2 -= stride;
1989 H += k*(src0[k] - src0[-k]);
1990 V += k*(src1[0] - src2[ 0]);
1993 H = ( 5*(H/4) ) / 16;
1994 V = ( 5*(V/4) ) / 16;
1996 /* required for 100% accuracy */
1997 i = H; H = V; V = i;
1999 H = ( 5*H+32 ) >> 6;
2000 V = ( 5*V+32 ) >> 6;
2003 a = 16*(src1[0] + src2[16] + 1) - 7*(V+H);
2004 for(j=16; j>0; --j) {
2007 for(i=-16; i<0; i+=4) {
2008 src[16+i] = cm[ (b ) >> 5 ];
2009 src[17+i] = cm[ (b+ H) >> 5 ];
2010 src[18+i] = cm[ (b+2*H) >> 5 ];
2011 src[19+i] = cm[ (b+3*H) >> 5 ];
2018 static void pred16x16_plane_c(uint8_t *src, int stride){
2019 pred16x16_plane_compat_c(src, stride, 0);
2022 static void pred8x8_vertical_c(uint8_t *src, int stride){
2024 const uint32_t a= ((uint32_t*)(src-stride))[0];
2025 const uint32_t b= ((uint32_t*)(src-stride))[1];
2028 ((uint32_t*)(src+i*stride))[0]= a;
2029 ((uint32_t*)(src+i*stride))[1]= b;
2033 static void pred8x8_horizontal_c(uint8_t *src, int stride){
2037 ((uint32_t*)(src+i*stride))[0]=
2038 ((uint32_t*)(src+i*stride))[1]= src[-1+i*stride]*0x01010101;
2042 static void pred8x8_128_dc_c(uint8_t *src, int stride){
2046 ((uint32_t*)(src+i*stride))[0]=
2047 ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
2050 ((uint32_t*)(src+i*stride))[0]=
2051 ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
2055 static void pred8x8_left_dc_c(uint8_t *src, int stride){
2061 dc0+= src[-1+i*stride];
2062 dc2+= src[-1+(i+4)*stride];
2064 dc0= 0x01010101*((dc0 + 2)>>2);
2065 dc2= 0x01010101*((dc2 + 2)>>2);
2068 ((uint32_t*)(src+i*stride))[0]=
2069 ((uint32_t*)(src+i*stride))[1]= dc0;
2072 ((uint32_t*)(src+i*stride))[0]=
2073 ((uint32_t*)(src+i*stride))[1]= dc2;
2077 static void pred8x8_top_dc_c(uint8_t *src, int stride){
2083 dc0+= src[i-stride];
2084 dc1+= src[4+i-stride];
2086 dc0= 0x01010101*((dc0 + 2)>>2);
2087 dc1= 0x01010101*((dc1 + 2)>>2);
2090 ((uint32_t*)(src+i*stride))[0]= dc0;
2091 ((uint32_t*)(src+i*stride))[1]= dc1;
2094 ((uint32_t*)(src+i*stride))[0]= dc0;
2095 ((uint32_t*)(src+i*stride))[1]= dc1;
2100 static void pred8x8_dc_c(uint8_t *src, int stride){
2102 int dc0, dc1, dc2, dc3;
2106 dc0+= src[-1+i*stride] + src[i-stride];
2107 dc1+= src[4+i-stride];
2108 dc2+= src[-1+(i+4)*stride];
2110 dc3= 0x01010101*((dc1 + dc2 + 4)>>3);
2111 dc0= 0x01010101*((dc0 + 4)>>3);
2112 dc1= 0x01010101*((dc1 + 2)>>2);
2113 dc2= 0x01010101*((dc2 + 2)>>2);
2116 ((uint32_t*)(src+i*stride))[0]= dc0;
2117 ((uint32_t*)(src+i*stride))[1]= dc1;
2120 ((uint32_t*)(src+i*stride))[0]= dc2;
2121 ((uint32_t*)(src+i*stride))[1]= dc3;
2125 static void pred8x8_plane_c(uint8_t *src, int stride){
2128 uint8_t *cm = cropTbl + MAX_NEG_CROP;
2129 const uint8_t * const src0 = src+3-stride;
2130 const uint8_t *src1 = src+4*stride-1;
2131 const uint8_t *src2 = src1-2*stride; // == src+2*stride-1;
2132 int H = src0[1] - src0[-1];
2133 int V = src1[0] - src2[ 0];
2134 for(k=2; k<=4; ++k) {
2135 src1 += stride; src2 -= stride;
2136 H += k*(src0[k] - src0[-k]);
2137 V += k*(src1[0] - src2[ 0]);
2139 H = ( 17*H+16 ) >> 5;
2140 V = ( 17*V+16 ) >> 5;
2142 a = 16*(src1[0] + src2[8]+1) - 3*(V+H);
2143 for(j=8; j>0; --j) {
2146 src[0] = cm[ (b ) >> 5 ];
2147 src[1] = cm[ (b+ H) >> 5 ];
2148 src[2] = cm[ (b+2*H) >> 5 ];
2149 src[3] = cm[ (b+3*H) >> 5 ];
2150 src[4] = cm[ (b+4*H) >> 5 ];
2151 src[5] = cm[ (b+5*H) >> 5 ];
2152 src[6] = cm[ (b+6*H) >> 5 ];
2153 src[7] = cm[ (b+7*H) >> 5 ];
2158 static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
2159 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2160 int src_x_offset, int src_y_offset,
2161 qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
2162 MpegEncContext * const s = &h->s;
2163 const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
2164 const int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
2165 const int luma_xy= (mx&3) + ((my&3)<<2);
2166 uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*s->linesize;
2167 uint8_t * src_cb= pic->data[1] + (mx>>3) + (my>>3)*s->uvlinesize;
2168 uint8_t * src_cr= pic->data[2] + (mx>>3) + (my>>3)*s->uvlinesize;
2169 int extra_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16; //FIXME increase edge?, IMHO not worth it
2170 int extra_height= extra_width;
2172 const int full_mx= mx>>2;
2173 const int full_my= my>>2;
2175 assert(pic->data[0]);
2177 if(mx&7) extra_width -= 3;
2178 if(my&7) extra_height -= 3;
2180 if( full_mx < 0-extra_width
2181 || full_my < 0-extra_height
2182 || full_mx + 16/*FIXME*/ > s->width + extra_width
2183 || full_my + 16/*FIXME*/ > s->height + extra_height){
2184 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);
2185 src_y= s->edge_emu_buffer + 2 + 2*s->linesize;
2189 qpix_op[luma_xy](dest_y, src_y, s->linesize); //FIXME try variable height perhaps?
2191 qpix_op[luma_xy](dest_y + delta, src_y + delta, s->linesize);
2194 if(s->flags&CODEC_FLAG_GRAY) return;
2197 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);
2198 src_cb= s->edge_emu_buffer;
2200 chroma_op(dest_cb, src_cb, s->uvlinesize, chroma_height, mx&7, my&7);
2203 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);
2204 src_cr= s->edge_emu_buffer;
2206 chroma_op(dest_cr, src_cr, s->uvlinesize, chroma_height, mx&7, my&7);
2209 static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
2210 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2211 int x_offset, int y_offset,
2212 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2213 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
2214 int list0, int list1){
2215 MpegEncContext * const s = &h->s;
2216 qpel_mc_func *qpix_op= qpix_put;
2217 h264_chroma_mc_func chroma_op= chroma_put;
2219 dest_y += 2*x_offset + 2*y_offset*s-> linesize;
2220 dest_cb += x_offset + y_offset*s->uvlinesize;
2221 dest_cr += x_offset + y_offset*s->uvlinesize;
2222 x_offset += 8*s->mb_x;
2223 y_offset += 8*s->mb_y;
2226 Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
2227 mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
2228 dest_y, dest_cb, dest_cr, x_offset, y_offset,
2229 qpix_op, chroma_op);
2232 chroma_op= chroma_avg;
2236 Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
2237 mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
2238 dest_y, dest_cb, dest_cr, x_offset, y_offset,
2239 qpix_op, chroma_op);
2243 static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
2244 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2245 int x_offset, int y_offset,
2246 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2247 h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
2248 h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
2249 int list0, int list1){
2250 MpegEncContext * const s = &h->s;
2252 dest_y += 2*x_offset + 2*y_offset*s-> linesize;
2253 dest_cb += x_offset + y_offset*s->uvlinesize;
2254 dest_cr += x_offset + y_offset*s->uvlinesize;
2255 x_offset += 8*s->mb_x;
2256 y_offset += 8*s->mb_y;
2259 /* don't optimize for luma-only case, since B-frames usually
2260 * use implicit weights => chroma too. */
2261 uint8_t *tmp_cb = s->obmc_scratchpad;
2262 uint8_t *tmp_cr = tmp_cb + 8*s->uvlinesize;
2263 uint8_t *tmp_y = tmp_cr + 8*s->uvlinesize;
2264 int refn0 = h->ref_cache[0][ scan8[n] ];
2265 int refn1 = h->ref_cache[1][ scan8[n] ];
2267 mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
2268 dest_y, dest_cb, dest_cr,
2269 x_offset, y_offset, qpix_put, chroma_put);
2270 mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
2271 tmp_y, tmp_cb, tmp_cr,
2272 x_offset, y_offset, qpix_put, chroma_put);
2274 if(h->use_weight == 2){
2275 int weight0 = h->implicit_weight[refn0][refn1];
2276 int weight1 = 64 - weight0;
2277 luma_weight_avg( dest_y, tmp_y, s-> linesize, 5, weight0, weight1, 0, 0);
2278 chroma_weight_avg(dest_cb, tmp_cb, s->uvlinesize, 5, weight0, weight1, 0, 0);
2279 chroma_weight_avg(dest_cr, tmp_cr, s->uvlinesize, 5, weight0, weight1, 0, 0);
2281 luma_weight_avg(dest_y, tmp_y, s->linesize, h->luma_log2_weight_denom,
2282 h->luma_weight[0][refn0], h->luma_weight[1][refn1],
2283 h->luma_offset[0][refn0], h->luma_offset[1][refn1]);
2284 chroma_weight_avg(dest_cb, tmp_cb, s->uvlinesize, h->chroma_log2_weight_denom,
2285 h->chroma_weight[0][refn0][0], h->chroma_weight[1][refn1][0],
2286 h->chroma_offset[0][refn0][0], h->chroma_offset[1][refn1][0]);
2287 chroma_weight_avg(dest_cr, tmp_cr, s->uvlinesize, h->chroma_log2_weight_denom,
2288 h->chroma_weight[0][refn0][1], h->chroma_weight[1][refn1][1],
2289 h->chroma_offset[0][refn0][1], h->chroma_offset[1][refn1][1]);
2292 int list = list1 ? 1 : 0;
2293 int refn = h->ref_cache[list][ scan8[n] ];
2294 Picture *ref= &h->ref_list[list][refn];
2295 mc_dir_part(h, ref, n, square, chroma_height, delta, list,
2296 dest_y, dest_cb, dest_cr, x_offset, y_offset,
2297 qpix_put, chroma_put);
2299 luma_weight_op(dest_y, s->linesize, h->luma_log2_weight_denom,
2300 h->luma_weight[list][refn], h->luma_offset[list][refn]);
2301 if(h->use_weight_chroma){
2302 chroma_weight_op(dest_cb, s->uvlinesize, h->chroma_log2_weight_denom,
2303 h->chroma_weight[list][refn][0], h->chroma_offset[list][refn][0]);
2304 chroma_weight_op(dest_cr, s->uvlinesize, h->chroma_log2_weight_denom,
2305 h->chroma_weight[list][refn][1], h->chroma_offset[list][refn][1]);
2310 static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
2311 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2312 int x_offset, int y_offset,
2313 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2314 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
2315 h264_weight_func *weight_op, h264_biweight_func *weight_avg,
2316 int list0, int list1){
2317 if((h->use_weight==2 && list0 && list1
2318 && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ] != 32))
2319 || h->use_weight==1)
2320 mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
2321 x_offset, y_offset, qpix_put, chroma_put,
2322 weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
2324 mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
2325 x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
2328 static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2329 qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
2330 qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
2331 h264_weight_func *weight_op, h264_biweight_func *weight_avg){
2332 MpegEncContext * const s = &h->s;
2333 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
2334 const int mb_type= s->current_picture.mb_type[mb_xy];
2336 assert(IS_INTER(mb_type));
2338 if(IS_16X16(mb_type)){
2339 mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
2340 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
2341 &weight_op[0], &weight_avg[0],
2342 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2343 }else if(IS_16X8(mb_type)){
2344 mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
2345 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
2346 &weight_op[1], &weight_avg[1],
2347 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2348 mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
2349 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
2350 &weight_op[1], &weight_avg[1],
2351 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2352 }else if(IS_8X16(mb_type)){
2353 mc_part(h, 0, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 0, 0,
2354 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2355 &weight_op[2], &weight_avg[2],
2356 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2357 mc_part(h, 4, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 4, 0,
2358 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2359 &weight_op[2], &weight_avg[2],
2360 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2364 assert(IS_8X8(mb_type));
2367 const int sub_mb_type= h->sub_mb_type[i];
2369 int x_offset= (i&1)<<2;
2370 int y_offset= (i&2)<<1;
2372 if(IS_SUB_8X8(sub_mb_type)){
2373 mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2374 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2375 &weight_op[3], &weight_avg[3],
2376 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2377 }else if(IS_SUB_8X4(sub_mb_type)){
2378 mc_part(h, n , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2379 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2380 &weight_op[4], &weight_avg[4],
2381 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2382 mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
2383 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2384 &weight_op[4], &weight_avg[4],
2385 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2386 }else if(IS_SUB_4X8(sub_mb_type)){
2387 mc_part(h, n , 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2388 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2389 &weight_op[5], &weight_avg[5],
2390 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2391 mc_part(h, n+1, 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
2392 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2393 &weight_op[5], &weight_avg[5],
2394 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2397 assert(IS_SUB_4X4(sub_mb_type));
2399 int sub_x_offset= x_offset + 2*(j&1);
2400 int sub_y_offset= y_offset + (j&2);
2401 mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
2402 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2403 &weight_op[6], &weight_avg[6],
2404 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2411 static void decode_init_vlc(H264Context *h){
2412 static int done = 0;
2418 init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5,
2419 &chroma_dc_coeff_token_len [0], 1, 1,
2420 &chroma_dc_coeff_token_bits[0], 1, 1, 1);
2423 init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17,
2424 &coeff_token_len [i][0], 1, 1,
2425 &coeff_token_bits[i][0], 1, 1, 1);
2429 init_vlc(&chroma_dc_total_zeros_vlc[i], CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
2430 &chroma_dc_total_zeros_len [i][0], 1, 1,
2431 &chroma_dc_total_zeros_bits[i][0], 1, 1, 1);
2433 for(i=0; i<15; i++){
2434 init_vlc(&total_zeros_vlc[i], TOTAL_ZEROS_VLC_BITS, 16,
2435 &total_zeros_len [i][0], 1, 1,
2436 &total_zeros_bits[i][0], 1, 1, 1);
2440 init_vlc(&run_vlc[i], RUN_VLC_BITS, 7,
2441 &run_len [i][0], 1, 1,
2442 &run_bits[i][0], 1, 1, 1);
2444 init_vlc(&run7_vlc, RUN7_VLC_BITS, 16,
2445 &run_len [6][0], 1, 1,
2446 &run_bits[6][0], 1, 1, 1);
2451 * Sets the intra prediction function pointers.
2453 static void init_pred_ptrs(H264Context *h){
2454 // MpegEncContext * const s = &h->s;
2456 h->pred4x4[VERT_PRED ]= pred4x4_vertical_c;
2457 h->pred4x4[HOR_PRED ]= pred4x4_horizontal_c;
2458 h->pred4x4[DC_PRED ]= pred4x4_dc_c;
2459 h->pred4x4[DIAG_DOWN_LEFT_PRED ]= pred4x4_down_left_c;
2460 h->pred4x4[DIAG_DOWN_RIGHT_PRED]= pred4x4_down_right_c;
2461 h->pred4x4[VERT_RIGHT_PRED ]= pred4x4_vertical_right_c;
2462 h->pred4x4[HOR_DOWN_PRED ]= pred4x4_horizontal_down_c;
2463 h->pred4x4[VERT_LEFT_PRED ]= pred4x4_vertical_left_c;
2464 h->pred4x4[HOR_UP_PRED ]= pred4x4_horizontal_up_c;
2465 h->pred4x4[LEFT_DC_PRED ]= pred4x4_left_dc_c;
2466 h->pred4x4[TOP_DC_PRED ]= pred4x4_top_dc_c;
2467 h->pred4x4[DC_128_PRED ]= pred4x4_128_dc_c;
2469 h->pred8x8[DC_PRED8x8 ]= pred8x8_dc_c;
2470 h->pred8x8[VERT_PRED8x8 ]= pred8x8_vertical_c;
2471 h->pred8x8[HOR_PRED8x8 ]= pred8x8_horizontal_c;
2472 h->pred8x8[PLANE_PRED8x8 ]= pred8x8_plane_c;
2473 h->pred8x8[LEFT_DC_PRED8x8]= pred8x8_left_dc_c;
2474 h->pred8x8[TOP_DC_PRED8x8 ]= pred8x8_top_dc_c;
2475 h->pred8x8[DC_128_PRED8x8 ]= pred8x8_128_dc_c;
2477 h->pred16x16[DC_PRED8x8 ]= pred16x16_dc_c;
2478 h->pred16x16[VERT_PRED8x8 ]= pred16x16_vertical_c;
2479 h->pred16x16[HOR_PRED8x8 ]= pred16x16_horizontal_c;
2480 h->pred16x16[PLANE_PRED8x8 ]= pred16x16_plane_c;
2481 h->pred16x16[LEFT_DC_PRED8x8]= pred16x16_left_dc_c;
2482 h->pred16x16[TOP_DC_PRED8x8 ]= pred16x16_top_dc_c;
2483 h->pred16x16[DC_128_PRED8x8 ]= pred16x16_128_dc_c;
2486 static void free_tables(H264Context *h){
2487 av_freep(&h->intra4x4_pred_mode);
2488 av_freep(&h->chroma_pred_mode_table);
2489 av_freep(&h->cbp_table);
2490 av_freep(&h->mvd_table[0]);
2491 av_freep(&h->mvd_table[1]);
2492 av_freep(&h->direct_table);
2493 av_freep(&h->non_zero_count);
2494 av_freep(&h->slice_table_base);
2495 av_freep(&h->top_border);
2496 h->slice_table= NULL;
2498 av_freep(&h->mb2b_xy);
2499 av_freep(&h->mb2b8_xy);
2501 av_freep(&h->s.obmc_scratchpad);
2506 * needs widzh/height
2508 static int alloc_tables(H264Context *h){
2509 MpegEncContext * const s = &h->s;
2510 const int big_mb_num= s->mb_stride * (s->mb_height+1);
2513 CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8 * sizeof(uint8_t))
2515 CHECKED_ALLOCZ(h->non_zero_count , big_mb_num * 16 * sizeof(uint8_t))
2516 CHECKED_ALLOCZ(h->slice_table_base , big_mb_num * sizeof(uint8_t))
2517 CHECKED_ALLOCZ(h->top_border , s->mb_width * (16+8+8) * sizeof(uint8_t))
2518 CHECKED_ALLOCZ(h->cbp_table, big_mb_num * sizeof(uint16_t))
2520 if( h->pps.cabac ) {
2521 CHECKED_ALLOCZ(h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t))
2522 CHECKED_ALLOCZ(h->mvd_table[0], 32*big_mb_num * sizeof(uint16_t));
2523 CHECKED_ALLOCZ(h->mvd_table[1], 32*big_mb_num * sizeof(uint16_t));
2524 CHECKED_ALLOCZ(h->direct_table, 32*big_mb_num * sizeof(uint8_t));
2527 memset(h->slice_table_base, -1, big_mb_num * sizeof(uint8_t));
2528 h->slice_table= h->slice_table_base + s->mb_stride + 1;
2530 CHECKED_ALLOCZ(h->mb2b_xy , big_mb_num * sizeof(uint16_t));
2531 CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint16_t));
2532 for(y=0; y<s->mb_height; y++){
2533 for(x=0; x<s->mb_width; x++){
2534 const int mb_xy= x + y*s->mb_stride;
2535 const int b_xy = 4*x + 4*y*h->b_stride;
2536 const int b8_xy= 2*x + 2*y*h->b8_stride;
2538 h->mb2b_xy [mb_xy]= b_xy;
2539 h->mb2b8_xy[mb_xy]= b8_xy;
2549 static void common_init(H264Context *h){
2550 MpegEncContext * const s = &h->s;
2552 s->width = s->avctx->width;
2553 s->height = s->avctx->height;
2554 s->codec_id= s->avctx->codec->id;
2558 s->unrestricted_mv=1;
2559 s->decode=1; //FIXME
2562 static int decode_init(AVCodecContext *avctx){
2563 H264Context *h= avctx->priv_data;
2564 MpegEncContext * const s = &h->s;
2566 MPV_decode_defaults(s);
2571 s->out_format = FMT_H264;
2572 s->workaround_bugs= avctx->workaround_bugs;
2575 // s->decode_mb= ff_h263_decode_mb;
2577 avctx->pix_fmt= PIX_FMT_YUV420P;
2581 if(avctx->codec_tag != 0x31637661 && avctx->codec_tag != 0x31435641) // avc1
2584 if((avctx->extradata_size == 0) || (avctx->extradata == NULL)) {
2585 av_log(avctx, AV_LOG_ERROR, "AVC codec requires avcC data\n");
2595 static void frame_start(H264Context *h){
2596 MpegEncContext * const s = &h->s;
2599 MPV_frame_start(s, s->avctx);
2600 ff_er_frame_start(s);
2603 assert(s->linesize && s->uvlinesize);
2605 for(i=0; i<16; i++){
2606 h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
2607 h->chroma_subblock_offset[i]= 2*((scan8[i] - scan8[0])&7) + 2*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2610 h->block_offset[16+i]=
2611 h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2614 /* can't be in alloc_tables because linesize isn't known there.
2615 * FIXME: redo bipred weight to not require extra buffer? */
2616 if(!s->obmc_scratchpad)
2617 s->obmc_scratchpad = av_malloc(16*s->linesize + 2*8*s->uvlinesize);
2619 // s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
2622 static inline void backup_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){
2623 MpegEncContext * const s = &h->s;
2627 src_cb -= uvlinesize;
2628 src_cr -= uvlinesize;
2630 h->left_border[0]= h->top_border[s->mb_x][15];
2631 for(i=1; i<17; i++){
2632 h->left_border[i]= src_y[15+i* linesize];
2635 *(uint64_t*)(h->top_border[s->mb_x]+0)= *(uint64_t*)(src_y + 16*linesize);
2636 *(uint64_t*)(h->top_border[s->mb_x]+8)= *(uint64_t*)(src_y +8+16*linesize);
2638 if(!(s->flags&CODEC_FLAG_GRAY)){
2639 h->left_border[17 ]= h->top_border[s->mb_x][16+7];
2640 h->left_border[17+9]= h->top_border[s->mb_x][24+7];
2642 h->left_border[i+17 ]= src_cb[7+i*uvlinesize];
2643 h->left_border[i+17+9]= src_cr[7+i*uvlinesize];
2645 *(uint64_t*)(h->top_border[s->mb_x]+16)= *(uint64_t*)(src_cb+8*uvlinesize);
2646 *(uint64_t*)(h->top_border[s->mb_x]+24)= *(uint64_t*)(src_cr+8*uvlinesize);
2650 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){
2651 MpegEncContext * const s = &h->s;
2654 int deblock_left = (s->mb_x > 0);
2655 int deblock_top = (s->mb_y > 0);
2657 src_y -= linesize + 1;
2658 src_cb -= uvlinesize + 1;
2659 src_cr -= uvlinesize + 1;
2661 #define XCHG(a,b,t,xchg)\
2668 for(i = !deblock_top; i<17; i++){
2669 XCHG(h->left_border[i ], src_y [i* linesize], temp8, xchg);
2674 XCHG(*(uint64_t*)(h->top_border[s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
2675 XCHG(*(uint64_t*)(h->top_border[s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
2678 if(!(s->flags&CODEC_FLAG_GRAY)){
2680 for(i = !deblock_top; i<9; i++){
2681 XCHG(h->left_border[i+17 ], src_cb[i*uvlinesize], temp8, xchg);
2682 XCHG(h->left_border[i+17+9], src_cr[i*uvlinesize], temp8, xchg);
2686 XCHG(*(uint64_t*)(h->top_border[s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
2687 XCHG(*(uint64_t*)(h->top_border[s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
2692 static void hl_decode_mb(H264Context *h){
2693 MpegEncContext * const s = &h->s;
2694 const int mb_x= s->mb_x;
2695 const int mb_y= s->mb_y;
2696 const int mb_xy= mb_x + mb_y*s->mb_stride;
2697 const int mb_type= s->current_picture.mb_type[mb_xy];
2698 uint8_t *dest_y, *dest_cb, *dest_cr;
2699 int linesize, uvlinesize /*dct_offset*/;
2708 dest_y = s->current_picture.data[0] + (mb_y * 16* s->linesize ) + mb_x * 16;
2709 dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2710 dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2712 if (h->mb_field_decoding_flag) {
2713 linesize = s->linesize * 2;
2714 uvlinesize = s->uvlinesize * 2;
2715 if(mb_y&1){ //FIXME move out of this func?
2716 dest_y -= s->linesize*15;
2717 dest_cb-= s->linesize*7;
2718 dest_cr-= s->linesize*7;
2721 linesize = s->linesize;
2722 uvlinesize = s->uvlinesize;
2723 // dct_offset = s->linesize * 16;
2726 if(IS_INTRA(mb_type)){
2727 if(h->deblocking_filter)
2728 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1);
2730 if(!(s->flags&CODEC_FLAG_GRAY)){
2731 h->pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
2732 h->pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
2735 if(IS_INTRA4x4(mb_type)){
2737 for(i=0; i<16; i++){
2738 uint8_t * const ptr= dest_y + h->block_offset[i];
2740 const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
2743 if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
2744 const int topright_avail= (h->topright_samples_available<<i)&0x8000;
2745 assert(mb_y || linesize <= h->block_offset[i]);
2746 if(!topright_avail){
2747 tr= ptr[3 - linesize]*0x01010101;
2748 topright= (uint8_t*) &tr;
2749 }else if(i==5 && h->deblocking_filter){
2750 tr= *(uint32_t*)h->top_border[mb_x+1];
2751 topright= (uint8_t*) &tr;
2753 topright= ptr + 4 - linesize;
2757 h->pred4x4[ dir ](ptr, topright, linesize);
2758 if(h->non_zero_count_cache[ scan8[i] ]){
2759 if(s->codec_id == CODEC_ID_H264)
2760 s->dsp.h264_idct_add(ptr, h->mb + i*16, linesize);
2762 svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
2767 h->pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
2768 if(s->codec_id == CODEC_ID_H264)
2769 h264_luma_dc_dequant_idct_c(h->mb, s->qscale);
2771 svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
2773 if(h->deblocking_filter)
2774 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);
2775 }else if(s->codec_id == CODEC_ID_H264){
2776 hl_motion(h, dest_y, dest_cb, dest_cr,
2777 s->dsp.put_h264_qpel_pixels_tab, s->dsp.put_h264_chroma_pixels_tab,
2778 s->dsp.avg_h264_qpel_pixels_tab, s->dsp.avg_h264_chroma_pixels_tab,
2779 s->dsp.weight_h264_pixels_tab, s->dsp.biweight_h264_pixels_tab);
2783 if(!IS_INTRA4x4(mb_type)){
2784 if(s->codec_id == CODEC_ID_H264){
2785 for(i=0; i<16; i++){
2786 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
2787 uint8_t * const ptr= dest_y + h->block_offset[i];
2788 s->dsp.h264_idct_add(ptr, h->mb + i*16, linesize);
2792 for(i=0; i<16; i++){
2793 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
2794 uint8_t * const ptr= dest_y + h->block_offset[i];
2795 svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
2801 if(!(s->flags&CODEC_FLAG_GRAY)){
2802 chroma_dc_dequant_idct_c(h->mb + 16*16, h->chroma_qp);
2803 chroma_dc_dequant_idct_c(h->mb + 16*16+4*16, h->chroma_qp);
2804 if(s->codec_id == CODEC_ID_H264){
2805 for(i=16; i<16+4; i++){
2806 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2807 uint8_t * const ptr= dest_cb + h->block_offset[i];
2808 s->dsp.h264_idct_add(ptr, h->mb + i*16, uvlinesize);
2811 for(i=20; i<20+4; i++){
2812 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2813 uint8_t * const ptr= dest_cr + h->block_offset[i];
2814 s->dsp.h264_idct_add(ptr, h->mb + i*16, uvlinesize);
2818 for(i=16; i<16+4; i++){
2819 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2820 uint8_t * const ptr= dest_cb + h->block_offset[i];
2821 svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
2824 for(i=20; i<20+4; i++){
2825 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2826 uint8_t * const ptr= dest_cr + h->block_offset[i];
2827 svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
2832 if(h->deblocking_filter) {
2833 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2834 filter_mb(h, mb_x, mb_y, dest_y, dest_cb, dest_cr);
2839 * fills the default_ref_list.
2841 static int fill_default_ref_list(H264Context *h){
2842 MpegEncContext * const s = &h->s;
2844 Picture sorted_short_ref[16];
2846 if(h->slice_type==B_TYPE){
2850 for(out_i=0; out_i<h->short_ref_count; out_i++){
2852 int best_poc=INT_MAX;
2854 for(i=0; i<h->short_ref_count; i++){
2855 const int poc= h->short_ref[i]->poc;
2856 if(poc > limit && poc < best_poc){
2862 assert(best_i != -1);
2865 sorted_short_ref[out_i]= *h->short_ref[best_i];
2869 if(s->picture_structure == PICT_FRAME){
2870 if(h->slice_type==B_TYPE){
2871 const int current_poc= s->current_picture_ptr->poc;
2874 for(list=0; list<2; list++){
2877 for(i=0; i<h->short_ref_count && index < h->ref_count[list]; i++){
2878 const int i2= list ? i : h->short_ref_count - i - 1;
2879 const int poc= sorted_short_ref[i2].poc;
2881 if(sorted_short_ref[i2].reference != 3) continue; //FIXME refernce field shit
2883 if((list==1 && poc > current_poc) || (list==0 && poc < current_poc)){
2884 h->default_ref_list[list][index ]= sorted_short_ref[i2];
2885 h->default_ref_list[list][index++].pic_id= sorted_short_ref[i2].frame_num;
2889 for(i=0; i<h->long_ref_count && index < h->ref_count[ list ]; i++){
2890 if(h->long_ref[i]->reference != 3) continue;
2892 h->default_ref_list[ list ][index ]= *h->long_ref[i];
2893 h->default_ref_list[ list ][index++].pic_id= i;;
2896 if(h->long_ref_count > 1 && h->short_ref_count==0){
2897 Picture temp= h->default_ref_list[1][0];
2898 h->default_ref_list[1][0] = h->default_ref_list[1][1];
2899 h->default_ref_list[1][0] = temp;
2902 if(index < h->ref_count[ list ])
2903 memset(&h->default_ref_list[list][index], 0, sizeof(Picture)*(h->ref_count[ list ] - index));
2907 for(i=0; i<h->short_ref_count && index < h->ref_count[0]; i++){
2908 if(h->short_ref[i]->reference != 3) continue; //FIXME refernce field shit
2909 h->default_ref_list[0][index ]= *h->short_ref[i];
2910 h->default_ref_list[0][index++].pic_id= h->short_ref[i]->frame_num;
2912 for(i=0; i<h->long_ref_count && index < h->ref_count[0]; i++){
2913 if(h->long_ref[i]->reference != 3) continue;
2914 h->default_ref_list[0][index ]= *h->long_ref[i];
2915 h->default_ref_list[0][index++].pic_id= i;;
2917 if(index < h->ref_count[0])
2918 memset(&h->default_ref_list[0][index], 0, sizeof(Picture)*(h->ref_count[0] - index));
2921 if(h->slice_type==B_TYPE){
2923 //FIXME second field balh
2929 static int decode_ref_pic_list_reordering(H264Context *h){
2930 MpegEncContext * const s = &h->s;
2933 if(h->slice_type==I_TYPE || h->slice_type==SI_TYPE) return 0; //FIXME move beofre func
2935 for(list=0; list<2; list++){
2936 memcpy(h->ref_list[list], h->default_ref_list[list], sizeof(Picture)*h->ref_count[list]);
2938 if(get_bits1(&s->gb)){
2939 int pred= h->curr_pic_num;
2942 for(index=0; ; index++){
2943 int reordering_of_pic_nums_idc= get_ue_golomb(&s->gb);
2947 if(reordering_of_pic_nums_idc==3)
2950 if(index >= h->ref_count[list]){
2951 av_log(h->s.avctx, AV_LOG_ERROR, "reference count overflow\n");
2955 if(reordering_of_pic_nums_idc<3){
2956 if(reordering_of_pic_nums_idc<2){
2957 const int abs_diff_pic_num= get_ue_golomb(&s->gb) + 1;
2959 if(abs_diff_pic_num >= h->max_pic_num){
2960 av_log(h->s.avctx, AV_LOG_ERROR, "abs_diff_pic_num overflow\n");
2964 if(reordering_of_pic_nums_idc == 0) pred-= abs_diff_pic_num;
2965 else pred+= abs_diff_pic_num;
2966 pred &= h->max_pic_num - 1;
2968 for(i= h->ref_count[list]-1; i>=index; i--){
2969 if(h->ref_list[list][i].pic_id == pred && h->ref_list[list][i].long_ref==0)
2973 pic_id= get_ue_golomb(&s->gb); //long_term_pic_idx
2975 for(i= h->ref_count[list]-1; i>=index; i--){
2976 if(h->ref_list[list][i].pic_id == pic_id && h->ref_list[list][i].long_ref==1)
2982 av_log(h->s.avctx, AV_LOG_ERROR, "reference picture missing during reorder\n");
2983 memset(&h->ref_list[list][index], 0, sizeof(Picture)); //FIXME
2984 }else if(i > index){
2985 Picture tmp= h->ref_list[list][i];
2986 for(; i>index; i--){
2987 h->ref_list[list][i]= h->ref_list[list][i-1];
2989 h->ref_list[list][index]= tmp;
2992 av_log(h->s.avctx, AV_LOG_ERROR, "illegal reordering_of_pic_nums_idc\n");
2998 if(h->slice_type!=B_TYPE) break;
3001 if(h->slice_type==B_TYPE && !h->direct_spatial_mv_pred)
3002 direct_dist_scale_factor(h);
3006 static int pred_weight_table(H264Context *h){
3007 MpegEncContext * const s = &h->s;
3009 int luma_def, chroma_def;
3012 h->use_weight_chroma= 0;
3013 h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
3014 h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
3015 luma_def = 1<<h->luma_log2_weight_denom;
3016 chroma_def = 1<<h->chroma_log2_weight_denom;
3018 for(list=0; list<2; list++){
3019 for(i=0; i<h->ref_count[list]; i++){
3020 int luma_weight_flag, chroma_weight_flag;
3022 luma_weight_flag= get_bits1(&s->gb);
3023 if(luma_weight_flag){
3024 h->luma_weight[list][i]= get_se_golomb(&s->gb);
3025 h->luma_offset[list][i]= get_se_golomb(&s->gb);
3026 if( h->luma_weight[list][i] != luma_def
3027 || h->luma_offset[list][i] != 0)
3030 h->luma_weight[list][i]= luma_def;
3031 h->luma_offset[list][i]= 0;
3034 chroma_weight_flag= get_bits1(&s->gb);
3035 if(chroma_weight_flag){
3038 h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);
3039 h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);
3040 if( h->chroma_weight[list][i][j] != chroma_def
3041 || h->chroma_offset[list][i][j] != 0)
3042 h->use_weight_chroma= 1;
3047 h->chroma_weight[list][i][j]= chroma_def;
3048 h->chroma_offset[list][i][j]= 0;
3052 if(h->slice_type != B_TYPE) break;
3054 h->use_weight= h->use_weight || h->use_weight_chroma;
3058 static void implicit_weight_table(H264Context *h){
3059 MpegEncContext * const s = &h->s;
3062 int cur_poc = s->current_picture_ptr->poc;
3064 if( h->ref_count[0] == 1 && h->ref_count[1] == 1
3065 && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
3067 h->use_weight_chroma= 0;
3072 h->use_weight_chroma= 2;
3073 h->luma_log2_weight_denom= 5;
3074 h->chroma_log2_weight_denom= 5;
3077 for(ref0=0; ref0 < h->ref_count[0]; ref0++){
3078 int poc0 = h->ref_list[0][ref0].poc;
3079 for(ref1=0; ref1 < h->ref_count[1]; ref1++){
3080 int poc1 = h->ref_list[0][ref1].poc;
3081 int td = clip(poc1 - poc0, -128, 127);
3083 int tb = clip(cur_poc - poc0, -128, 127);
3084 int tx = (16384 + (ABS(td) >> 1)) / td;
3085 int dist_scale_factor = clip((tb*tx + 32) >> 6, -1024, 1023) >> 2;
3086 if(dist_scale_factor < -64 || dist_scale_factor > 128)
3087 h->implicit_weight[ref0][ref1] = 32;
3089 h->implicit_weight[ref0][ref1] = 64 - dist_scale_factor;
3091 h->implicit_weight[ref0][ref1] = 32;
3097 * instantaneous decoder refresh.
3099 static void idr(H264Context *h){
3102 #define CHECK_DELAY(pic) \
3103 for(j = 0; h->delayed_pic[j]; j++) \
3104 if(pic == h->delayed_pic[j]){ \
3109 for(i=0; i<h->long_ref_count; i++){
3110 h->long_ref[i]->reference=0;
3111 CHECK_DELAY(h->long_ref[i]);
3112 h->long_ref[i]= NULL;
3114 h->long_ref_count=0;
3116 for(i=0; i<h->short_ref_count; i++){
3117 h->short_ref[i]->reference=0;
3118 CHECK_DELAY(h->short_ref[i]);
3119 h->short_ref[i]= NULL;
3121 h->short_ref_count=0;
3127 * @return the removed picture or NULL if an error occures
3129 static Picture * remove_short(H264Context *h, int frame_num){
3130 MpegEncContext * const s = &h->s;
3133 if(s->avctx->debug&FF_DEBUG_MMCO)
3134 av_log(h->s.avctx, AV_LOG_DEBUG, "remove short %d count %d\n", frame_num, h->short_ref_count);
3136 for(i=0; i<h->short_ref_count; i++){
3137 Picture *pic= h->short_ref[i];
3138 if(s->avctx->debug&FF_DEBUG_MMCO)
3139 av_log(h->s.avctx, AV_LOG_DEBUG, "%d %d %p\n", i, pic->frame_num, pic);
3140 if(pic->frame_num == frame_num){
3141 h->short_ref[i]= NULL;
3142 memmove(&h->short_ref[i], &h->short_ref[i+1], (h->short_ref_count - i - 1)*sizeof(Picture*));
3143 h->short_ref_count--;
3152 * @return the removed picture or NULL if an error occures
3154 static Picture * remove_long(H264Context *h, int i){
3157 if(i >= h->long_ref_count) return NULL;
3158 pic= h->long_ref[i];
3159 if(pic==NULL) return NULL;
3161 h->long_ref[i]= NULL;
3162 memmove(&h->long_ref[i], &h->long_ref[i+1], (h->long_ref_count - i - 1)*sizeof(Picture*));
3163 h->long_ref_count--;
3169 * Executes the reference picture marking (memory management control operations).
3171 static int execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count){
3172 MpegEncContext * const s = &h->s;
3174 int current_is_long=0;
3177 if((s->avctx->debug&FF_DEBUG_MMCO) && mmco_count==0)
3178 av_log(h->s.avctx, AV_LOG_DEBUG, "no mmco here\n");
3180 for(i=0; i<mmco_count; i++){
3181 if(s->avctx->debug&FF_DEBUG_MMCO)
3182 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);
3184 switch(mmco[i].opcode){
3185 case MMCO_SHORT2UNUSED:
3186 pic= remove_short(h, mmco[i].short_frame_num);
3187 if(pic==NULL) return -1;
3190 case MMCO_SHORT2LONG:
3191 pic= remove_long(h, mmco[i].long_index);
3192 if(pic) pic->reference=0;
3194 h->long_ref[ mmco[i].long_index ]= remove_short(h, mmco[i].short_frame_num);
3195 h->long_ref[ mmco[i].long_index ]->long_ref=1;
3197 case MMCO_LONG2UNUSED:
3198 pic= remove_long(h, mmco[i].long_index);
3199 if(pic==NULL) return -1;
3203 pic= remove_long(h, mmco[i].long_index);
3204 if(pic) pic->reference=0;
3206 h->long_ref[ mmco[i].long_index ]= s->current_picture_ptr;
3207 h->long_ref[ mmco[i].long_index ]->long_ref=1;
3208 h->long_ref_count++;
3212 case MMCO_SET_MAX_LONG:
3213 assert(mmco[i].long_index <= 16);
3214 while(mmco[i].long_index < h->long_ref_count){
3215 pic= remove_long(h, mmco[i].long_index);
3218 while(mmco[i].long_index > h->long_ref_count){
3219 h->long_ref[ h->long_ref_count++ ]= NULL;
3223 while(h->short_ref_count){
3224 pic= remove_short(h, h->short_ref[0]->frame_num);
3227 while(h->long_ref_count){
3228 pic= remove_long(h, h->long_ref_count-1);
3236 if(!current_is_long){
3237 pic= remove_short(h, s->current_picture_ptr->frame_num);
3240 av_log(h->s.avctx, AV_LOG_ERROR, "illegal short term buffer state detected\n");
3243 if(h->short_ref_count)
3244 memmove(&h->short_ref[1], &h->short_ref[0], h->short_ref_count*sizeof(Picture*));
3246 h->short_ref[0]= s->current_picture_ptr;
3247 h->short_ref[0]->long_ref=0;
3248 h->short_ref_count++;
3254 static int decode_ref_pic_marking(H264Context *h){
3255 MpegEncContext * const s = &h->s;
3258 if(h->nal_unit_type == NAL_IDR_SLICE){ //FIXME fields
3259 s->broken_link= get_bits1(&s->gb) -1;
3260 h->mmco[0].long_index= get_bits1(&s->gb) - 1; // current_long_term_idx
3261 if(h->mmco[0].long_index == -1)
3264 h->mmco[0].opcode= MMCO_LONG;
3268 if(get_bits1(&s->gb)){ // adaptive_ref_pic_marking_mode_flag
3269 for(i= 0; i<MAX_MMCO_COUNT; i++) {
3270 MMCOOpcode opcode= get_ue_golomb(&s->gb);;
3272 h->mmco[i].opcode= opcode;
3273 if(opcode==MMCO_SHORT2UNUSED || opcode==MMCO_SHORT2LONG){
3274 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
3275 /* if(h->mmco[i].short_frame_num >= h->short_ref_count || h->short_ref[ h->mmco[i].short_frame_num ] == NULL){
3276 fprintf(stderr, "illegal short ref in memory management control operation %d\n", mmco);
3280 if(opcode==MMCO_SHORT2LONG || opcode==MMCO_LONG2UNUSED || opcode==MMCO_LONG || opcode==MMCO_SET_MAX_LONG){
3281 h->mmco[i].long_index= get_ue_golomb(&s->gb);
3282 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){
3283 av_log(h->s.avctx, AV_LOG_ERROR, "illegal long ref in memory management control operation %d\n", opcode);
3288 if(opcode > MMCO_LONG){
3289 av_log(h->s.avctx, AV_LOG_ERROR, "illegal memory management control operation %d\n", opcode);
3292 if(opcode == MMCO_END)
3297 assert(h->long_ref_count + h->short_ref_count <= h->sps.ref_frame_count);
3299 if(h->long_ref_count + h->short_ref_count == h->sps.ref_frame_count){ //FIXME fields
3300 h->mmco[0].opcode= MMCO_SHORT2UNUSED;
3301 h->mmco[0].short_frame_num= h->short_ref[ h->short_ref_count - 1 ]->frame_num;
3311 static int init_poc(H264Context *h){
3312 MpegEncContext * const s = &h->s;
3313 const int max_frame_num= 1<<h->sps.log2_max_frame_num;
3316 if(h->nal_unit_type == NAL_IDR_SLICE){
3317 h->frame_num_offset= 0;
3319 if(h->frame_num < h->prev_frame_num)
3320 h->frame_num_offset= h->prev_frame_num_offset + max_frame_num;
3322 h->frame_num_offset= h->prev_frame_num_offset;
3325 if(h->sps.poc_type==0){
3326 const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
3328 if (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
3329 h->poc_msb = h->prev_poc_msb + max_poc_lsb;
3330 else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
3331 h->poc_msb = h->prev_poc_msb - max_poc_lsb;
3333 h->poc_msb = h->prev_poc_msb;
3334 //printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
3336 field_poc[1] = h->poc_msb + h->poc_lsb;
3337 if(s->picture_structure == PICT_FRAME)
3338 field_poc[1] += h->delta_poc_bottom;
3339 }else if(h->sps.poc_type==1){
3340 int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
3343 if(h->sps.poc_cycle_length != 0)
3344 abs_frame_num = h->frame_num_offset + h->frame_num;
3348 if(h->nal_ref_idc==0 && abs_frame_num > 0)
3351 expected_delta_per_poc_cycle = 0;
3352 for(i=0; i < h->sps.poc_cycle_length; i++)
3353 expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
3355 if(abs_frame_num > 0){
3356 int poc_cycle_cnt = (abs_frame_num - 1) / h->sps.poc_cycle_length;
3357 int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
3359 expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
3360 for(i = 0; i <= frame_num_in_poc_cycle; i++)
3361 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
3365 if(h->nal_ref_idc == 0)
3366 expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
3368 field_poc[0] = expectedpoc + h->delta_poc[0];
3369 field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
3371 if(s->picture_structure == PICT_FRAME)
3372 field_poc[1] += h->delta_poc[1];
3375 if(h->nal_unit_type == NAL_IDR_SLICE){
3378 if(h->nal_ref_idc) poc= 2*(h->frame_num_offset + h->frame_num);
3379 else poc= 2*(h->frame_num_offset + h->frame_num) - 1;
3385 if(s->picture_structure != PICT_BOTTOM_FIELD)
3386 s->current_picture_ptr->field_poc[0]= field_poc[0];
3387 if(s->picture_structure != PICT_TOP_FIELD)
3388 s->current_picture_ptr->field_poc[1]= field_poc[1];
3389 if(s->picture_structure == PICT_FRAME) // FIXME field pix?
3390 s->current_picture_ptr->poc= FFMIN(field_poc[0], field_poc[1]);
3396 * decodes a slice header.
3397 * this will allso call MPV_common_init() and frame_start() as needed
3399 static int decode_slice_header(H264Context *h){
3400 MpegEncContext * const s = &h->s;
3401 int first_mb_in_slice, pps_id;
3402 int num_ref_idx_active_override_flag;
3403 static const uint8_t slice_type_map[5]= {P_TYPE, B_TYPE, I_TYPE, SP_TYPE, SI_TYPE};
3405 s->current_picture.reference= h->nal_ref_idc != 0;
3407 first_mb_in_slice= get_ue_golomb(&s->gb);
3409 h->slice_type= get_ue_golomb(&s->gb);
3410 if(h->slice_type > 9){
3411 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);
3414 if(h->slice_type > 4){
3416 h->slice_type_fixed=1;
3418 h->slice_type_fixed=0;
3420 h->slice_type= slice_type_map[ h->slice_type ];
3422 s->pict_type= h->slice_type; // to make a few old func happy, its wrong though
3424 pps_id= get_ue_golomb(&s->gb);
3426 av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
3429 h->pps= h->pps_buffer[pps_id];
3430 if(h->pps.slice_group_count == 0){
3431 av_log(h->s.avctx, AV_LOG_ERROR, "non existing PPS referenced\n");
3435 h->sps= h->sps_buffer[ h->pps.sps_id ];
3436 if(h->sps.log2_max_frame_num == 0){
3437 av_log(h->s.avctx, AV_LOG_ERROR, "non existing SPS referenced\n");
3441 s->mb_width= h->sps.mb_width;
3442 s->mb_height= h->sps.mb_height;
3444 h->b_stride= s->mb_width*4 + 1;
3445 h->b8_stride= s->mb_width*2 + 1;
3447 s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
3448 s->resync_mb_y = s->mb_y = first_mb_in_slice / s->mb_width; //FIXME AFFW
3450 s->width = 16*s->mb_width - 2*(h->sps.crop_left + h->sps.crop_right );
3451 if(h->sps.frame_mbs_only_flag)
3452 s->height= 16*s->mb_height - 2*(h->sps.crop_top + h->sps.crop_bottom);
3454 s->height= 16*s->mb_height - 4*(h->sps.crop_top + h->sps.crop_bottom); //FIXME recheck
3456 if (s->context_initialized
3457 && ( s->width != s->avctx->width || s->height != s->avctx->height)) {
3461 if (!s->context_initialized) {
3462 if (MPV_common_init(s) < 0)
3467 s->avctx->width = s->width;
3468 s->avctx->height = s->height;
3469 s->avctx->sample_aspect_ratio= h->sps.sar;
3471 if(h->sps.timing_info_present_flag && h->sps.fixed_frame_rate_flag){
3472 s->avctx->frame_rate = h->sps.time_scale;
3473 s->avctx->frame_rate_base = h->sps.num_units_in_tick;
3477 if(h->slice_num == 0){
3481 s->current_picture_ptr->frame_num= //FIXME frame_num cleanup
3482 h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
3484 if(h->sps.frame_mbs_only_flag){
3485 s->picture_structure= PICT_FRAME;
3487 if(get_bits1(&s->gb)) //field_pic_flag
3488 s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
3490 s->picture_structure= PICT_FRAME;
3493 if(s->picture_structure==PICT_FRAME){
3494 h->curr_pic_num= h->frame_num;
3495 h->max_pic_num= 1<< h->sps.log2_max_frame_num;
3497 h->curr_pic_num= 2*h->frame_num;
3498 h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
3501 if(h->nal_unit_type == NAL_IDR_SLICE){
3502 get_ue_golomb(&s->gb); /* idr_pic_id */
3505 if(h->sps.poc_type==0){
3506 h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
3508 if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
3509 h->delta_poc_bottom= get_se_golomb(&s->gb);
3513 if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
3514 h->delta_poc[0]= get_se_golomb(&s->gb);
3516 if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
3517 h->delta_poc[1]= get_se_golomb(&s->gb);
3522 if(h->pps.redundant_pic_cnt_present){
3523 h->redundant_pic_count= get_ue_golomb(&s->gb);
3526 //set defaults, might be overriden a few line later
3527 h->ref_count[0]= h->pps.ref_count[0];
3528 h->ref_count[1]= h->pps.ref_count[1];
3530 if(h->slice_type == P_TYPE || h->slice_type == SP_TYPE || h->slice_type == B_TYPE){
3531 if(h->slice_type == B_TYPE){
3532 h->direct_spatial_mv_pred= get_bits1(&s->gb);
3534 num_ref_idx_active_override_flag= get_bits1(&s->gb);
3536 if(num_ref_idx_active_override_flag){
3537 h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
3538 if(h->slice_type==B_TYPE)
3539 h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
3541 if(h->ref_count[0] > 32 || h->ref_count[1] > 32){
3542 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
3548 if(h->slice_num == 0){
3549 fill_default_ref_list(h);
3552 decode_ref_pic_list_reordering(h);
3554 if( (h->pps.weighted_pred && (h->slice_type == P_TYPE || h->slice_type == SP_TYPE ))
3555 || (h->pps.weighted_bipred_idc==1 && h->slice_type==B_TYPE ) )
3556 pred_weight_table(h);
3557 else if(h->pps.weighted_bipred_idc==2 && h->slice_type==B_TYPE)
3558 implicit_weight_table(h);
3562 if(s->current_picture.reference)
3563 decode_ref_pic_marking(h);
3565 if( h->slice_type != I_TYPE && h->slice_type != SI_TYPE && h->pps.cabac )
3566 h->cabac_init_idc = get_ue_golomb(&s->gb);
3568 h->last_qscale_diff = 0;
3569 s->qscale = h->pps.init_qp + get_se_golomb(&s->gb);
3570 if(s->qscale<0 || s->qscale>51){
3571 av_log(s->avctx, AV_LOG_ERROR, "QP %d out of range\n", s->qscale);
3574 h->chroma_qp = get_chroma_qp(h, s->qscale);
3575 //FIXME qscale / qp ... stuff
3576 if(h->slice_type == SP_TYPE){
3577 get_bits1(&s->gb); /* sp_for_switch_flag */
3579 if(h->slice_type==SP_TYPE || h->slice_type == SI_TYPE){
3580 get_se_golomb(&s->gb); /* slice_qs_delta */
3583 h->deblocking_filter = 1;
3584 h->slice_alpha_c0_offset = 0;
3585 h->slice_beta_offset = 0;
3586 if( h->pps.deblocking_filter_parameters_present ) {
3587 h->deblocking_filter= get_ue_golomb(&s->gb);
3588 if(h->deblocking_filter < 2)
3589 h->deblocking_filter^= 1; // 1<->0
3591 if( h->deblocking_filter ) {
3592 h->slice_alpha_c0_offset = get_se_golomb(&s->gb) << 1;
3593 h->slice_beta_offset = get_se_golomb(&s->gb) << 1;
3598 if( h->pps.num_slice_groups > 1 && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
3599 slice_group_change_cycle= get_bits(&s->gb, ?);
3604 if(s->avctx->debug&FF_DEBUG_PICT_INFO){
3605 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",
3606 h->slice_num, first_mb_in_slice,
3607 av_get_pict_type_char(h->slice_type),
3608 pps_id, h->frame_num,
3609 s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
3610 h->ref_count[0], h->ref_count[1],
3612 h->deblocking_filter,
3614 h->use_weight==1 && h->use_weight_chroma ? "c" : ""
3624 static inline int get_level_prefix(GetBitContext *gb){
3628 OPEN_READER(re, gb);
3629 UPDATE_CACHE(re, gb);
3630 buf=GET_CACHE(re, gb);
3632 log= 32 - av_log2(buf);
3634 print_bin(buf>>(32-log), log);
3635 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__);
3638 LAST_SKIP_BITS(re, gb, log);
3639 CLOSE_READER(re, gb);
3645 * decodes a residual block.
3646 * @param n block index
3647 * @param scantable scantable
3648 * @param max_coeff number of coefficients in the block
3649 * @return <0 if an error occured
3651 static int decode_residual(H264Context *h, GetBitContext *gb, DCTELEM *block, int n, const uint8_t *scantable, int qp, int max_coeff){
3652 MpegEncContext * const s = &h->s;
3653 const uint16_t *qmul= dequant_coeff[qp];
3654 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};
3655 int level[16], run[16];
3656 int suffix_length, zeros_left, coeff_num, coeff_token, total_coeff, i, trailing_ones;
3658 //FIXME put trailing_onex into the context
3660 if(n == CHROMA_DC_BLOCK_INDEX){
3661 coeff_token= get_vlc2(gb, chroma_dc_coeff_token_vlc.table, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 1);
3662 total_coeff= coeff_token>>2;
3664 if(n == LUMA_DC_BLOCK_INDEX){
3665 total_coeff= pred_non_zero_count(h, 0);
3666 coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
3667 total_coeff= coeff_token>>2;
3669 total_coeff= pred_non_zero_count(h, n);
3670 coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
3671 total_coeff= coeff_token>>2;
3672 h->non_zero_count_cache[ scan8[n] ]= total_coeff;
3676 //FIXME set last_non_zero?
3681 trailing_ones= coeff_token&3;
3682 tprintf("trailing:%d, total:%d\n", trailing_ones, total_coeff);
3683 assert(total_coeff<=16);
3685 for(i=0; i<trailing_ones; i++){
3686 level[i]= 1 - 2*get_bits1(gb);
3689 suffix_length= total_coeff > 10 && trailing_ones < 3;
3691 for(; i<total_coeff; i++){
3692 const int prefix= get_level_prefix(gb);
3693 int level_code, mask;
3695 if(prefix<14){ //FIXME try to build a large unified VLC table for all this
3697 level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
3699 level_code= (prefix<<suffix_length); //part
3700 }else if(prefix==14){
3702 level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
3704 level_code= prefix + get_bits(gb, 4); //part
3705 }else if(prefix==15){
3706 level_code= (prefix<<suffix_length) + get_bits(gb, 12); //part
3707 if(suffix_length==0) level_code+=15; //FIXME doesnt make (much)sense
3709 av_log(h->s.avctx, AV_LOG_ERROR, "prefix too large at %d %d\n", s->mb_x, s->mb_y);
3713 if(i==trailing_ones && i<3) level_code+= 2; //FIXME split first iteration
3715 mask= -(level_code&1);
3716 level[i]= (((2+level_code)>>1) ^ mask) - mask;
3718 if(suffix_length==0) suffix_length=1; //FIXME split first iteration
3721 if(ABS(level[i]) > (3<<(suffix_length-1)) && suffix_length<6) suffix_length++;
3723 if((2+level_code)>>1) > (3<<(suffix_length-1)) && suffix_length<6) suffix_length++;
3724 /* ? == prefix > 2 or sth */
3726 tprintf("level: %d suffix_length:%d\n", level[i], suffix_length);
3729 if(total_coeff == max_coeff)
3732 if(n == CHROMA_DC_BLOCK_INDEX)
3733 zeros_left= get_vlc2(gb, chroma_dc_total_zeros_vlc[ total_coeff-1 ].table, CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 1);
3735 zeros_left= get_vlc2(gb, total_zeros_vlc[ total_coeff-1 ].table, TOTAL_ZEROS_VLC_BITS, 1);
3738 for(i=0; i<total_coeff-1; i++){
3741 else if(zeros_left < 7){
3742 run[i]= get_vlc2(gb, run_vlc[zeros_left-1].table, RUN_VLC_BITS, 1);
3744 run[i]= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2);
3746 zeros_left -= run[i];
3750 av_log(h->s.avctx, AV_LOG_ERROR, "negative number of zero coeffs at %d %d\n", s->mb_x, s->mb_y);
3754 for(; i<total_coeff-1; i++){
3762 for(i=total_coeff-1; i>=0; i--){ //FIXME merge into rundecode?
3765 coeff_num += run[i] + 1; //FIXME add 1 earlier ?
3766 j= scantable[ coeff_num ];
3771 for(i=total_coeff-1; i>=0; i--){ //FIXME merge into rundecode?
3774 coeff_num += run[i] + 1; //FIXME add 1 earlier ?
3775 j= scantable[ coeff_num ];
3777 block[j]= level[i] * qmul[j];
3778 // printf("%d %d ", block[j], qmul[j]);
3785 * decodes a P_SKIP or B_SKIP macroblock
3787 static void decode_mb_skip(H264Context *h){
3788 MpegEncContext * const s = &h->s;
3789 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
3792 memset(h->non_zero_count[mb_xy], 0, 16);
3793 memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui
3795 if( h->slice_type == B_TYPE )
3797 // just for fill_caches. pred_direct_motion will set the real mb_type
3798 mb_type= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2|MB_TYPE_SKIP;
3801 fill_caches(h, mb_type); //FIXME check what is needed and what not ...
3802 pred_direct_motion(h, &mb_type);
3804 fill_rectangle(h->mvd_cache[0][scan8[0]], 4, 4, 8, 0, 4);
3805 fill_rectangle(h->mvd_cache[1][scan8[0]], 4, 4, 8, 0, 4);
3811 mb_type= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0|MB_TYPE_SKIP;
3813 if(h->sps.mb_aff && s->mb_skip_run==0 && (s->mb_y&1)==0){
3814 h->mb_field_decoding_flag= get_bits1(&s->gb);
3816 if(h->mb_field_decoding_flag)
3817 mb_type|= MB_TYPE_INTERLACED;
3819 fill_caches(h, mb_type); //FIXME check what is needed and what not ...
3820 pred_pskip_motion(h, &mx, &my);
3821 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
3822 fill_rectangle( h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4);
3824 fill_rectangle(h->mvd_cache[0][scan8[0]], 4, 4, 8, 0, 4);
3827 write_back_motion(h, mb_type);
3828 s->current_picture.mb_type[mb_xy]= mb_type|MB_TYPE_SKIP;
3829 s->current_picture.qscale_table[mb_xy]= s->qscale;
3830 h->slice_table[ mb_xy ]= h->slice_num;
3831 h->prev_mb_skiped= 1;
3835 * decodes a macroblock
3836 * @returns 0 if ok, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
3838 static int decode_mb_cavlc(H264Context *h){
3839 MpegEncContext * const s = &h->s;
3840 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
3841 int mb_type, partition_count, cbp;
3843 s->dsp.clear_blocks(h->mb); //FIXME avoid if allready clear (move after skip handlong?
3845 tprintf("pic:%d mb:%d/%d\n", h->frame_num, s->mb_x, s->mb_y);
3846 cbp = 0; /* avoid warning. FIXME: find a solution without slowing
3848 if(h->slice_type != I_TYPE && h->slice_type != SI_TYPE){
3849 if(s->mb_skip_run==-1)
3850 s->mb_skip_run= get_ue_golomb(&s->gb);
3852 if (s->mb_skip_run--) {
3857 if(h->sps.mb_aff /* && !field pic FIXME needed? */){
3859 h->mb_field_decoding_flag = get_bits1(&s->gb);
3861 h->mb_field_decoding_flag=0; //FIXME som ed note ?!
3863 h->prev_mb_skiped= 0;
3865 mb_type= get_ue_golomb(&s->gb);
3866 if(h->slice_type == B_TYPE){
3868 partition_count= b_mb_type_info[mb_type].partition_count;
3869 mb_type= b_mb_type_info[mb_type].type;
3872 goto decode_intra_mb;
3874 }else if(h->slice_type == P_TYPE /*|| h->slice_type == SP_TYPE */){
3876 partition_count= p_mb_type_info[mb_type].partition_count;
3877 mb_type= p_mb_type_info[mb_type].type;
3880 goto decode_intra_mb;
3883 assert(h->slice_type == I_TYPE);
3886 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);
3890 cbp= i_mb_type_info[mb_type].cbp;
3891 h->intra16x16_pred_mode= i_mb_type_info[mb_type].pred_mode;
3892 mb_type= i_mb_type_info[mb_type].type;
3895 if(h->mb_field_decoding_flag)
3896 mb_type |= MB_TYPE_INTERLACED;
3898 s->current_picture.mb_type[mb_xy]= mb_type;
3899 h->slice_table[ mb_xy ]= h->slice_num;
3901 if(IS_INTRA_PCM(mb_type)){
3905 // we assume these blocks are very rare so we dont optimize it
3906 align_get_bits(&s->gb);
3908 ptr= s->gb.buffer + get_bits_count(&s->gb);
3910 for(y=0; y<16; y++){
3911 const int index= 4*(y&3) + 64*(y>>2);
3912 for(x=0; x<16; x++){
3913 h->mb[index + (x&3) + 16*(x>>2)]= *(ptr++);
3917 const int index= 256 + 4*(y&3) + 32*(y>>2);
3919 h->mb[index + (x&3) + 16*(x>>2)]= *(ptr++);
3923 const int index= 256 + 64 + 4*(y&3) + 32*(y>>2);
3925 h->mb[index + (x&3) + 16*(x>>2)]= *(ptr++);
3929 skip_bits(&s->gb, 384); //FIXME check /fix the bitstream readers
3931 //FIXME deblock filter, non_zero_count_cache init ...
3932 memset(h->non_zero_count[mb_xy], 16, 16);
3933 s->current_picture.qscale_table[mb_xy]= s->qscale;
3938 fill_caches(h, mb_type);
3941 if(IS_INTRA(mb_type)){
3942 // init_top_left_availability(h);
3943 if(IS_INTRA4x4(mb_type)){
3946 // fill_intra4x4_pred_table(h);
3947 for(i=0; i<16; i++){
3948 const int mode_coded= !get_bits1(&s->gb);
3949 const int predicted_mode= pred_intra_mode(h, i);
3953 const int rem_mode= get_bits(&s->gb, 3);
3954 if(rem_mode<predicted_mode)
3959 mode= predicted_mode;
3962 h->intra4x4_pred_mode_cache[ scan8[i] ] = mode;
3964 write_back_intra_pred_mode(h);
3965 if( check_intra4x4_pred_mode(h) < 0)
3968 h->intra16x16_pred_mode= check_intra_pred_mode(h, h->intra16x16_pred_mode);
3969 if(h->intra16x16_pred_mode < 0)
3972 h->chroma_pred_mode= get_ue_golomb(&s->gb);
3974 h->chroma_pred_mode= check_intra_pred_mode(h, h->chroma_pred_mode);
3975 if(h->chroma_pred_mode < 0)
3977 }else if(partition_count==4){
3978 int i, j, sub_partition_count[4], list, ref[2][4];
3980 if(h->slice_type == B_TYPE){
3982 h->sub_mb_type[i]= get_ue_golomb(&s->gb);
3983 if(h->sub_mb_type[i] >=13){
3984 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);
3987 sub_partition_count[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
3988 h->sub_mb_type[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].type;
3990 if( IS_DIRECT(h->sub_mb_type[0]) || IS_DIRECT(h->sub_mb_type[1])
3991 || IS_DIRECT(h->sub_mb_type[2]) || IS_DIRECT(h->sub_mb_type[3]))
3992 pred_direct_motion(h, &mb_type);
3994 assert(h->slice_type == P_TYPE || h->slice_type == SP_TYPE); //FIXME SP correct ?
3996 h->sub_mb_type[i]= get_ue_golomb(&s->gb);
3997 if(h->sub_mb_type[i] >=4){
3998 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);
4001 sub_partition_count[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
4002 h->sub_mb_type[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].type;
4006 for(list=0; list<2; list++){
4007 const int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list];
4008 if(ref_count == 0) continue;
4010 if(IS_DIRECT(h->sub_mb_type[i])) continue;
4011 if(IS_DIR(h->sub_mb_type[i], 0, list)){
4012 ref[list][i] = get_te0_golomb(&s->gb, ref_count); //FIXME init to 0 before and skip?
4020 for(list=0; list<2; list++){
4021 const int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list];
4022 if(ref_count == 0) continue;
4025 if(IS_DIRECT(h->sub_mb_type[i])) continue;
4026 h->ref_cache[list][ scan8[4*i] ]=h->ref_cache[list][ scan8[4*i]+1 ]=
4027 h->ref_cache[list][ scan8[4*i]+8 ]=h->ref_cache[list][ scan8[4*i]+9 ]= ref[list][i];
4029 if(IS_DIR(h->sub_mb_type[i], 0, list)){
4030 const int sub_mb_type= h->sub_mb_type[i];
4031 const int block_width= (sub_mb_type & (MB_TYPE_16x16|MB_TYPE_16x8)) ? 2 : 1;
4032 for(j=0; j<sub_partition_count[i]; j++){
4034 const int index= 4*i + block_width*j;
4035 int16_t (* mv_cache)[2]= &h->mv_cache[list][ scan8[index] ];
4036 pred_motion(h, index, block_width, list, h->ref_cache[list][ scan8[index] ], &mx, &my);
4037 mx += get_se_golomb(&s->gb);
4038 my += get_se_golomb(&s->gb);
4039 tprintf("final mv:%d %d\n", mx, my);
4041 if(IS_SUB_8X8(sub_mb_type)){
4042 mv_cache[ 0 ][0]= mv_cache[ 1 ][0]=
4043 mv_cache[ 8 ][0]= mv_cache[ 9 ][0]= mx;
4044 mv_cache[ 0 ][1]= mv_cache[ 1 ][1]=
4045 mv_cache[ 8 ][1]= mv_cache[ 9 ][1]= my;
4046 }else if(IS_SUB_8X4(sub_mb_type)){
4047 mv_cache[ 0 ][0]= mv_cache[ 1 ][0]= mx;
4048 mv_cache[ 0 ][1]= mv_cache[ 1 ][1]= my;
4049 }else if(IS_SUB_4X8(sub_mb_type)){
4050 mv_cache[ 0 ][0]= mv_cache[ 8 ][0]= mx;
4051 mv_cache[ 0 ][1]= mv_cache[ 8 ][1]= my;
4053 assert(IS_SUB_4X4(sub_mb_type));
4054 mv_cache[ 0 ][0]= mx;
4055 mv_cache[ 0 ][1]= my;
4059 uint32_t *p= (uint32_t *)&h->mv_cache[list][ scan8[4*i] ][0];
4065 }else if(IS_DIRECT(mb_type)){
4066 pred_direct_motion(h, &mb_type);
4067 s->current_picture.mb_type[mb_xy]= mb_type;
4069 int list, mx, my, i;
4070 //FIXME we should set ref_idx_l? to 0 if we use that later ...
4071 if(IS_16X16(mb_type)){
4072 for(list=0; list<2; list++){
4073 if(h->ref_count[list]>0){
4074 if(IS_DIR(mb_type, 0, list)){
4075 const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
4076 fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, val, 1);
4080 for(list=0; list<2; list++){
4081 if(IS_DIR(mb_type, 0, list)){
4082 pred_motion(h, 0, 4, list, h->ref_cache[list][ scan8[0] ], &mx, &my);
4083 mx += get_se_golomb(&s->gb);
4084 my += get_se_golomb(&s->gb);
4085 tprintf("final mv:%d %d\n", mx, my);
4087 fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, pack16to32(mx,my), 4);
4091 else if(IS_16X8(mb_type)){
4092 for(list=0; list<2; list++){
4093 if(h->ref_count[list]>0){
4095 if(IS_DIR(mb_type, i, list)){
4096 const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
4097 fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, val, 1);
4098 }else // needed only for mixed refs (e.g. B_L0_L1_16x8)
4099 fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, (LIST_NOT_USED&0xFF), 1);
4103 for(list=0; list<2; list++){
4105 if(IS_DIR(mb_type, i, list)){
4106 pred_16x8_motion(h, 8*i, list, h->ref_cache[list][scan8[0] + 16*i], &mx, &my);
4107 mx += get_se_golomb(&s->gb);
4108 my += get_se_golomb(&s->gb);
4109 tprintf("final mv:%d %d\n", mx, my);
4111 fill_rectangle(h->mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, pack16to32(mx,my), 4);
4113 fill_rectangle(h->mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, 0, 4);
4117 assert(IS_8X16(mb_type));
4118 for(list=0; list<2; list++){
4119 if(h->ref_count[list]>0){
4121 if(IS_DIR(mb_type, i, list)){ //FIXME optimize
4122 const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
4123 fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, val, 1);
4124 }else // needed only for mixed refs
4125 fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, (LIST_NOT_USED&0xFF), 1);
4129 for(list=0; list<2; list++){
4131 if(IS_DIR(mb_type, i, list)){
4132 pred_8x16_motion(h, i*4, list, h->ref_cache[list][ scan8[0] + 2*i ], &mx, &my);
4133 mx += get_se_golomb(&s->gb);
4134 my += get_se_golomb(&s->gb);
4135 tprintf("final mv:%d %d\n", mx, my);
4137 fill_rectangle(h->mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, pack16to32(mx,my), 4);
4139 fill_rectangle(h->mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, 0, 4);
4145 if(IS_INTER(mb_type))
4146 write_back_motion(h, mb_type);
4148 if(!IS_INTRA16x16(mb_type)){
4149 cbp= get_ue_golomb(&s->gb);
4151 av_log(h->s.avctx, AV_LOG_ERROR, "cbp too large (%d) at %d %d\n", cbp, s->mb_x, s->mb_y);
4155 if(IS_INTRA4x4(mb_type))
4156 cbp= golomb_to_intra4x4_cbp[cbp];
4158 cbp= golomb_to_inter_cbp[cbp];
4161 if(cbp || IS_INTRA16x16(mb_type)){
4162 int i8x8, i4x4, chroma_idx;
4163 int chroma_qp, dquant;
4164 GetBitContext *gb= IS_INTRA(mb_type) ? h->intra_gb_ptr : h->inter_gb_ptr;
4165 const uint8_t *scan, *dc_scan;
4167 // fill_non_zero_count_cache(h);
4169 if(IS_INTERLACED(mb_type)){
4171 dc_scan= luma_dc_field_scan;
4174 dc_scan= luma_dc_zigzag_scan;
4177 dquant= get_se_golomb(&s->gb);
4179 if( dquant > 25 || dquant < -26 ){
4180 av_log(h->s.avctx, AV_LOG_ERROR, "dquant out of range (%d) at %d %d\n", dquant, s->mb_x, s->mb_y);
4184 s->qscale += dquant;
4185 if(((unsigned)s->qscale) > 51){
4186 if(s->qscale<0) s->qscale+= 52;
4187 else s->qscale-= 52;
4190 h->chroma_qp= chroma_qp= get_chroma_qp(h, s->qscale);
4191 if(IS_INTRA16x16(mb_type)){
4192 if( decode_residual(h, h->intra_gb_ptr, h->mb, LUMA_DC_BLOCK_INDEX, dc_scan, s->qscale, 16) < 0){
4193 return -1; //FIXME continue if partotioned and other retirn -1 too
4196 assert((cbp&15) == 0 || (cbp&15) == 15);
4199 for(i8x8=0; i8x8<4; i8x8++){
4200 for(i4x4=0; i4x4<4; i4x4++){
4201 const int index= i4x4 + 4*i8x8;
4202 if( decode_residual(h, h->intra_gb_ptr, h->mb + 16*index, index, scan + 1, s->qscale, 15) < 0 ){
4208 fill_rectangle(&h->non_zero_count_cache[scan8[0]], 4, 4, 8, 0, 1);
4211 for(i8x8=0; i8x8<4; i8x8++){
4212 if(cbp & (1<<i8x8)){
4213 for(i4x4=0; i4x4<4; i4x4++){
4214 const int index= i4x4 + 4*i8x8;
4216 if( decode_residual(h, gb, h->mb + 16*index, index, scan, s->qscale, 16) <0 ){
4221 uint8_t * const nnz= &h->non_zero_count_cache[ scan8[4*i8x8] ];
4222 nnz[0] = nnz[1] = nnz[8] = nnz[9] = 0;
4228 for(chroma_idx=0; chroma_idx<2; chroma_idx++)
4229 if( decode_residual(h, gb, h->mb + 256 + 16*4*chroma_idx, CHROMA_DC_BLOCK_INDEX, chroma_dc_scan, chroma_qp, 4) < 0){
4235 for(chroma_idx=0; chroma_idx<2; chroma_idx++){
4236 for(i4x4=0; i4x4<4; i4x4++){
4237 const int index= 16 + 4*chroma_idx + i4x4;
4238 if( decode_residual(h, gb, h->mb + 16*index, index, scan + 1, chroma_qp, 15) < 0){
4244 uint8_t * const nnz= &h->non_zero_count_cache[0];
4245 nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =
4246 nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
4249 uint8_t * const nnz= &h->non_zero_count_cache[0];
4250 fill_rectangle(&nnz[scan8[0]], 4, 4, 8, 0, 1);
4251 nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =
4252 nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
4254 s->current_picture.qscale_table[mb_xy]= s->qscale;
4255 write_back_non_zero_count(h);
4260 static int decode_cabac_intra_mb_type(H264Context *h, int ctx_base, int intra_slice) {
4261 uint8_t *state= &h->cabac_state[ctx_base];
4265 MpegEncContext * const s = &h->s;
4266 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
4268 if( s->mb_x > 0 && !IS_INTRA4x4( s->current_picture.mb_type[mb_xy-1] ) )
4270 if( s->mb_y > 0 && !IS_INTRA4x4( s->current_picture.mb_type[mb_xy-s->mb_stride] ) )
4272 if( get_cabac( &h->cabac, &state[ctx] ) == 0 )
4273 return 0; /* I4x4 */
4276 if( get_cabac( &h->cabac, &state[0] ) == 0 )
4277 return 0; /* I4x4 */
4280 if( get_cabac_terminate( &h->cabac ) )
4281 return 25; /* PCM */
4283 mb_type = 1; /* I16x16 */
4284 if( get_cabac( &h->cabac, &state[1] ) )
4285 mb_type += 12; /* cbp_luma != 0 */
4287 if( get_cabac( &h->cabac, &state[2] ) ) {
4288 if( get_cabac( &h->cabac, &state[2+intra_slice] ) )
4289 mb_type += 4 * 2; /* cbp_chroma == 2 */
4291 mb_type += 4 * 1; /* cbp_chroma == 1 */
4293 if( get_cabac( &h->cabac, &state[3+intra_slice] ) )
4295 if( get_cabac( &h->cabac, &state[3+2*intra_slice] ) )
4300 static int decode_cabac_mb_type( H264Context *h ) {
4301 MpegEncContext * const s = &h->s;
4303 if( h->slice_type == I_TYPE ) {
4304 return decode_cabac_intra_mb_type(h, 3, 1);
4305 } else if( h->slice_type == P_TYPE ) {
4306 if( get_cabac( &h->cabac, &h->cabac_state[14] ) == 0 ) {
4308 if( get_cabac( &h->cabac, &h->cabac_state[15] ) == 0 ) {
4309 if( get_cabac( &h->cabac, &h->cabac_state[16] ) == 0 )
4310 return 0; /* P_L0_D16x16; */
4312 return 3; /* P_8x8; */
4314 if( get_cabac( &h->cabac, &h->cabac_state[17] ) == 0 )
4315 return 2; /* P_L0_D8x16; */
4317 return 1; /* P_L0_D16x8; */
4320 return decode_cabac_intra_mb_type(h, 17, 0) + 5;
4322 } else if( h->slice_type == B_TYPE ) {
4323 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
4327 if( s->mb_x > 0 && !IS_SKIP( s->current_picture.mb_type[mb_xy-1] )
4328 && !IS_DIRECT( s->current_picture.mb_type[mb_xy-1] ) )
4330 if( s->mb_y > 0 && !IS_SKIP( s->current_picture.mb_type[mb_xy-s->mb_stride] )
4331 && !IS_DIRECT( s->current_picture.mb_type[mb_xy-s->mb_stride] ) )
4334 if( !get_cabac( &h->cabac, &h->cabac_state[27+ctx] ) )
4335 return 0; /* B_Direct_16x16 */
4337 if( !get_cabac( &h->cabac, &h->cabac_state[27+3] ) ) {
4338 return 1 + get_cabac( &h->cabac, &h->cabac_state[27+5] ); /* B_L[01]_16x16 */
4341 bits = get_cabac( &h->cabac, &h->cabac_state[27+4] ) << 3;
4342 bits|= get_cabac( &h->cabac, &h->cabac_state[27+5] ) << 2;
4343 bits|= get_cabac( &h->cabac, &h->cabac_state[27+5] ) << 1;
4344 bits|= get_cabac( &h->cabac, &h->cabac_state[27+5] );
4346 return bits + 3; /* B_Bi_16x16 through B_L1_L0_16x8 */
4347 else if( bits == 13 ) {
4348 return decode_cabac_intra_mb_type(h, 32, 0) + 23;
4349 } else if( bits == 14 )
4350 return 11; /* B_L1_L0_8x16 */
4351 else if( bits == 15 )
4352 return 22; /* B_8x8 */
4354 bits= ( bits<<1 ) | get_cabac( &h->cabac, &h->cabac_state[27+5] );
4355 return bits - 4; /* B_L0_Bi_* through B_Bi_Bi_* */
4357 /* TODO SI/SP frames? */
4362 static int decode_cabac_mb_skip( H264Context *h) {
4363 MpegEncContext * const s = &h->s;
4364 const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
4365 const int mba_xy = mb_xy - 1;
4366 const int mbb_xy = mb_xy - s->mb_stride;
4369 if( s->mb_x > 0 && !IS_SKIP( s->current_picture.mb_type[mba_xy] ) )
4371 if( s->mb_y > 0 && !IS_SKIP( s->current_picture.mb_type[mbb_xy] ) )
4374 if( h->slice_type == P_TYPE || h->slice_type == SP_TYPE)
4375 return get_cabac( &h->cabac, &h->cabac_state[11+ctx] );
4377 return get_cabac( &h->cabac, &h->cabac_state[24+ctx] );
4380 static int decode_cabac_mb_intra4x4_pred_mode( H264Context *h, int pred_mode ) {
4383 if( get_cabac( &h->cabac, &h->cabac_state[68] ) )
4386 if( get_cabac( &h->cabac, &h->cabac_state[69] ) )
4388 if( get_cabac( &h->cabac, &h->cabac_state[69] ) )
4390 if( get_cabac( &h->cabac, &h->cabac_state[69] ) )
4392 if( mode >= pred_mode )
4398 static int decode_cabac_mb_chroma_pre_mode( H264Context *h) {
4399 MpegEncContext * const s = &h->s;
4400 const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
4401 const int mba_xy = mb_xy - 1;
4402 const int mbb_xy = mb_xy - s->mb_stride;
4406 /* No need to test for IS_INTRA4x4 and IS_INTRA16x16, as we set chroma_pred_mode_table to 0 */
4407 if( s->mb_x > 0 && h->chroma_pred_mode_table[mba_xy] != 0 )
4410 if( s->mb_y > 0 && h->chroma_pred_mode_table[mbb_xy] != 0 )
4413 if( get_cabac( &h->cabac, &h->cabac_state[64+ctx] ) == 0 )
4416 if( get_cabac( &h->cabac, &h->cabac_state[64+3] ) == 0 )
4418 if( get_cabac( &h->cabac, &h->cabac_state[64+3] ) == 0 )
4424 static const uint8_t block_idx_x[16] = {
4425 0, 1, 0, 1, 2, 3, 2, 3, 0, 1, 0, 1, 2, 3, 2, 3
4427 static const uint8_t block_idx_y[16] = {
4428 0, 0, 1, 1, 0, 0, 1, 1, 2, 2, 3, 3, 2, 2, 3, 3
4430 static const uint8_t block_idx_xy[4][4] = {
4437 static int decode_cabac_mb_cbp_luma( H264Context *h) {
4438 MpegEncContext * const s = &h->s;
4439 const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
4444 h->cbp_table[mb_xy] = 0; /* FIXME aaahahahah beurk */
4446 for( i8x8 = 0; i8x8 < 4; i8x8++ ) {
4452 x = block_idx_x[4*i8x8];
4453 y = block_idx_y[4*i8x8];
4457 else if( s->mb_x > 0 )
4462 else if( s->mb_y > 0 )
4463 mbb_xy = mb_xy - s->mb_stride;
4465 /* No need to test for skip as we put 0 for skip block */
4467 int i8x8a = block_idx_xy[(x-1)&0x03][y]/4;
4468 if( ((h->cbp_table[mba_xy] >> i8x8a)&0x01) == 0 )
4473 int i8x8b = block_idx_xy[x][(y-1)&0x03]/4;
4474 if( ((h->cbp_table[mbb_xy] >> i8x8b)&0x01) == 0 )
4478 if( get_cabac( &h->cabac, &h->cabac_state[73 + ctx] ) ) {
4480 h->cbp_table[mb_xy] = cbp; /* FIXME aaahahahah beurk */
4485 static int decode_cabac_mb_cbp_chroma( H264Context *h) {
4489 cbp_a = (h->left_cbp>>4)&0x03;
4490 cbp_b = (h-> top_cbp>>4)&0x03;
4493 if( cbp_a > 0 ) ctx++;
4494 if( cbp_b > 0 ) ctx += 2;
4495 if( get_cabac( &h->cabac, &h->cabac_state[77 + ctx] ) == 0 )
4499 if( cbp_a == 2 ) ctx++;
4500 if( cbp_b == 2 ) ctx += 2;
4501 return 1 + get_cabac( &h->cabac, &h->cabac_state[77 + ctx] );
4503 static int decode_cabac_mb_dqp( H264Context *h) {
4504 MpegEncContext * const s = &h->s;
4510 mbn_xy = s->mb_x + s->mb_y*s->mb_stride - 1;
4512 mbn_xy = s->mb_width - 1 + (s->mb_y-1)*s->mb_stride;
4514 if( mbn_xy >= 0 && h->last_qscale_diff != 0 && ( IS_INTRA16x16(s->current_picture.mb_type[mbn_xy] ) || (h->cbp_table[mbn_xy]&0x3f) ) )
4517 while( get_cabac( &h->cabac, &h->cabac_state[60 + ctx] ) ) {
4528 return -(val + 1)/2;
4530 static int decode_cabac_p_mb_sub_type( H264Context *h ) {
4531 if( get_cabac( &h->cabac, &h->cabac_state[21] ) )
4533 if( !get_cabac( &h->cabac, &h->cabac_state[22] ) )
4535 if( get_cabac( &h->cabac, &h->cabac_state[23] ) )
4539 static int decode_cabac_b_mb_sub_type( H264Context *h ) {
4541 if( !get_cabac( &h->cabac, &h->cabac_state[36] ) )
4542 return 0; /* B_Direct_8x8 */
4543 if( !get_cabac( &h->cabac, &h->cabac_state[37] ) )
4544 return 1 + get_cabac( &h->cabac, &h->cabac_state[39] ); /* B_L0_8x8, B_L1_8x8 */
4546 if( get_cabac( &h->cabac, &h->cabac_state[38] ) ) {
4547 if( get_cabac( &h->cabac, &h->cabac_state[39] ) )
4548 return 11 + get_cabac( &h->cabac, &h->cabac_state[39] ); /* B_L1_4x4, B_Bi_4x4 */
4551 type += 2*get_cabac( &h->cabac, &h->cabac_state[39] );
4552 type += get_cabac( &h->cabac, &h->cabac_state[39] );
4556 static int decode_cabac_mb_ref( H264Context *h, int list, int n ) {
4557 int refa = h->ref_cache[list][scan8[n] - 1];
4558 int refb = h->ref_cache[list][scan8[n] - 8];
4562 if( h->slice_type == B_TYPE) {
4563 if( refa > 0 && !h->direct_cache[scan8[n] - 1] )
4565 if( refb > 0 && !h->direct_cache[scan8[n] - 8] )
4574 while( get_cabac( &h->cabac, &h->cabac_state[54+ctx] ) ) {
4584 static int decode_cabac_mb_mvd( H264Context *h, int list, int n, int l ) {
4585 int amvd = abs( h->mvd_cache[list][scan8[n] - 1][l] ) +
4586 abs( h->mvd_cache[list][scan8[n] - 8][l] );
4587 int ctxbase = (l == 0) ? 40 : 47;
4592 else if( amvd > 32 )
4597 if(!get_cabac(&h->cabac, &h->cabac_state[ctxbase+ctx]))
4602 while( mvd < 9 && get_cabac( &h->cabac, &h->cabac_state[ctxbase+ctx] ) ) {
4610 while( get_cabac_bypass( &h->cabac ) ) {
4615 if( get_cabac_bypass( &h->cabac ) )
4619 if( get_cabac_bypass( &h->cabac ) ) return -mvd;
4623 static int inline get_cabac_cbf_ctx( H264Context *h, int cat, int idx ) {
4628 nza = h->left_cbp&0x100;
4629 nzb = h-> top_cbp&0x100;
4630 } else if( cat == 1 || cat == 2 ) {
4631 nza = h->non_zero_count_cache[scan8[idx] - 1];
4632 nzb = h->non_zero_count_cache[scan8[idx] - 8];
4633 } else if( cat == 3 ) {
4634 nza = (h->left_cbp>>(6+idx))&0x01;
4635 nzb = (h-> top_cbp>>(6+idx))&0x01;
4638 nza = h->non_zero_count_cache[scan8[16+idx] - 1];
4639 nzb = h->non_zero_count_cache[scan8[16+idx] - 8];
4648 return ctx + 4 * cat;
4651 static int inline decode_cabac_residual( H264Context *h, DCTELEM *block, int cat, int n, const uint8_t *scantable, int qp, int max_coeff) {
4652 const int mb_xy = h->s.mb_x + h->s.mb_y*h->s.mb_stride;
4653 const uint16_t *qmul= dequant_coeff[qp];
4654 static const int significant_coeff_flag_offset[5] = { 0, 15, 29, 44, 47 };
4655 static const int coeff_abs_level_m1_offset[5] = {227+ 0, 227+10, 227+20, 227+30, 227+39 };
4660 int coeff_count = 0;
4663 int abslevelgt1 = 0;
4665 /* cat: 0-> DC 16x16 n = 0
4666 * 1-> AC 16x16 n = luma4x4idx
4667 * 2-> Luma4x4 n = luma4x4idx
4668 * 3-> DC Chroma n = iCbCr
4669 * 4-> AC Chroma n = 4 * iCbCr + chroma4x4idx
4672 /* read coded block flag */
4673 if( get_cabac( &h->cabac, &h->cabac_state[85 + get_cabac_cbf_ctx( h, cat, n ) ] ) == 0 ) {
4674 if( cat == 1 || cat == 2 )
4675 h->non_zero_count_cache[scan8[n]] = 0;
4677 h->non_zero_count_cache[scan8[16+n]] = 0;
4682 for(last= 0; last < max_coeff - 1; last++) {
4683 if( get_cabac( &h->cabac, &h->cabac_state[105+significant_coeff_flag_offset[cat]+last] )) {
4684 index[coeff_count++] = last;
4685 if( get_cabac( &h->cabac, &h->cabac_state[166+significant_coeff_flag_offset[cat]+last] ) ) {
4691 if( last == max_coeff -1 ) {
4692 index[coeff_count++] = last;
4694 assert(coeff_count > 0);
4697 h->cbp_table[mb_xy] |= 0x100;
4698 else if( cat == 1 || cat == 2 )
4699 h->non_zero_count_cache[scan8[n]] = coeff_count;
4701 h->cbp_table[mb_xy] |= 0x40 << n;
4704 h->non_zero_count_cache[scan8[16+n]] = coeff_count;
4707 for( i = coeff_count - 1; i >= 0; i-- ) {
4708 int ctx = (abslevelgt1 != 0 ? 0 : FFMIN( 4, abslevel1 )) + coeff_abs_level_m1_offset[cat];
4709 int j= scantable[index[i]];
4711 if( get_cabac( &h->cabac, &h->cabac_state[ctx] ) == 0 ) {
4712 if( cat == 0 || cat == 3 ) {
4713 if( get_cabac_bypass( &h->cabac ) ) block[j] = -1;
4716 if( get_cabac_bypass( &h->cabac ) ) block[j] = -qmul[j];
4717 else block[j] = qmul[j];
4723 ctx = 5 + FFMIN( 4, abslevelgt1 ) + coeff_abs_level_m1_offset[cat];
4724 while( coeff_abs < 15 && get_cabac( &h->cabac, &h->cabac_state[ctx] ) ) {
4728 if( coeff_abs >= 15 ) {
4730 while( get_cabac_bypass( &h->cabac ) ) {
4731 coeff_abs += 1 << j;
4736 if( get_cabac_bypass( &h->cabac ) )
4737 coeff_abs += 1 << j ;
4741 if( cat == 0 || cat == 3 ) {
4742 if( get_cabac_bypass( &h->cabac ) ) block[j] = -coeff_abs;
4743 else block[j] = coeff_abs;
4745 if( get_cabac_bypass( &h->cabac ) ) block[j] = -coeff_abs * qmul[j];
4746 else block[j] = coeff_abs * qmul[j];
4756 * decodes a macroblock
4757 * @returns 0 if ok, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
4759 static int decode_mb_cabac(H264Context *h) {
4760 MpegEncContext * const s = &h->s;
4761 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
4762 int mb_type, partition_count, cbp = 0;
4764 s->dsp.clear_blocks(h->mb); //FIXME avoid if allready clear (move after skip handlong?)
4766 if( h->sps.mb_aff ) {
4767 av_log( h->s.avctx, AV_LOG_ERROR, "Fields not supported with CABAC\n" );
4771 if( h->slice_type != I_TYPE && h->slice_type != SI_TYPE ) {
4772 /* read skip flags */
4773 if( decode_cabac_mb_skip( h ) ) {
4776 h->cbp_table[mb_xy] = 0;
4777 h->chroma_pred_mode_table[mb_xy] = 0;
4778 h->last_qscale_diff = 0;
4784 h->prev_mb_skiped = 0;
4786 if( ( mb_type = decode_cabac_mb_type( h ) ) < 0 ) {
4787 av_log( h->s.avctx, AV_LOG_ERROR, "decode_cabac_mb_type failed\n" );
4791 if( h->slice_type == B_TYPE ) {
4793 partition_count= b_mb_type_info[mb_type].partition_count;
4794 mb_type= b_mb_type_info[mb_type].type;
4797 goto decode_intra_mb;
4799 } else if( h->slice_type == P_TYPE ) {
4801 partition_count= p_mb_type_info[mb_type].partition_count;
4802 mb_type= p_mb_type_info[mb_type].type;
4805 goto decode_intra_mb;
4808 assert(h->slice_type == I_TYPE);
4810 partition_count = 0;
4811 cbp= i_mb_type_info[mb_type].cbp;
4812 h->intra16x16_pred_mode= i_mb_type_info[mb_type].pred_mode;
4813 mb_type= i_mb_type_info[mb_type].type;
4816 if(h->mb_field_decoding_flag)
4817 mb_type |= MB_TYPE_INTERLACED;
4820 s->current_picture.mb_type[mb_xy]= mb_type;
4821 h->slice_table[ mb_xy ]= h->slice_num;
4823 if(IS_INTRA_PCM(mb_type)) {
4826 h->cbp_table[mb_xy] = 0xf +4*2; //FIXME ?!
4827 h->cbp_table[mb_xy] |= 0x1C0;
4828 h->chroma_pred_mode_table[mb_xy] = 0;
4829 s->current_picture.qscale_table[mb_xy]= s->qscale;
4833 fill_caches(h, mb_type);
4835 if( IS_INTRA( mb_type ) ) {
4836 if( IS_INTRA4x4( mb_type ) ) {
4838 for( i = 0; i < 16; i++ ) {
4839 int pred = pred_intra_mode( h, i );
4840 h->intra4x4_pred_mode_cache[ scan8[i] ] = decode_cabac_mb_intra4x4_pred_mode( h, pred );
4842 //av_log( s->avctx, AV_LOG_ERROR, "i4x4 pred=%d mode=%d\n", pred, h->intra4x4_pred_mode_cache[ scan8[i] ] );
4844 write_back_intra_pred_mode(h);
4845 if( check_intra4x4_pred_mode(h) < 0 ) return -1;
4847 h->intra16x16_pred_mode= check_intra_pred_mode( h, h->intra16x16_pred_mode );
4848 if( h->intra16x16_pred_mode < 0 ) return -1;
4850 h->chroma_pred_mode_table[mb_xy] =
4851 h->chroma_pred_mode = decode_cabac_mb_chroma_pre_mode( h );
4853 h->chroma_pred_mode= check_intra_pred_mode( h, h->chroma_pred_mode );
4854 if( h->chroma_pred_mode < 0 ) return -1;
4855 } else if( partition_count == 4 ) {
4856 int i, j, sub_partition_count[4], list, ref[2][4];
4858 if( h->slice_type == B_TYPE ) {
4859 for( i = 0; i < 4; i++ ) {
4860 h->sub_mb_type[i] = decode_cabac_b_mb_sub_type( h );
4861 sub_partition_count[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
4862 h->sub_mb_type[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].type;
4864 if( IS_DIRECT(h->sub_mb_type[0]) || IS_DIRECT(h->sub_mb_type[1])
4865 || IS_DIRECT(h->sub_mb_type[2]) || IS_DIRECT(h->sub_mb_type[3])) {
4866 pred_direct_motion(h, &mb_type);
4867 if( h->ref_count[0] > 1 || h->ref_count[1] > 1 ) {
4868 for( i = 0; i < 4; i++ )
4869 if( IS_DIRECT(h->sub_mb_type[i]) )
4870 fill_rectangle( &h->direct_cache[scan8[4*i]], 2, 2, 8, 1, 1 );
4874 for( i = 0; i < 4; i++ ) {
4875 h->sub_mb_type[i] = decode_cabac_p_mb_sub_type( h );
4876 sub_partition_count[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
4877 h->sub_mb_type[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].type;
4881 for( list = 0; list < 2; list++ ) {
4882 if( h->ref_count[list] > 0 ) {
4883 for( i = 0; i < 4; i++ ) {
4884 if(IS_DIRECT(h->sub_mb_type[i])) continue;
4885 if(IS_DIR(h->sub_mb_type[i], 0, list)){
4886 if( h->ref_count[list] > 1 )
4887 ref[list][i] = decode_cabac_mb_ref( h, list, 4*i );
4893 h->ref_cache[list][ scan8[4*i]+1 ]=
4894 h->ref_cache[list][ scan8[4*i]+8 ]=h->ref_cache[list][ scan8[4*i]+9 ]= ref[list][i];
4899 for(list=0; list<2; list++){
4901 if(IS_DIRECT(h->sub_mb_type[i])){
4902 fill_rectangle(h->mvd_cache[list][scan8[4*i]], 2, 2, 8, 0, 4);
4905 h->ref_cache[list][ scan8[4*i] ]=h->ref_cache[list][ scan8[4*i]+1 ];
4907 if(IS_DIR(h->sub_mb_type[i], 0, list) && !IS_DIRECT(h->sub_mb_type[i])){
4908 const int sub_mb_type= h->sub_mb_type[i];
4909 const int block_width= (sub_mb_type & (MB_TYPE_16x16|MB_TYPE_16x8)) ? 2 : 1;
4910 for(j=0; j<sub_partition_count[i]; j++){
4913 const int index= 4*i + block_width*j;
4914 int16_t (* mv_cache)[2]= &h->mv_cache[list][ scan8[index] ];
4915 int16_t (* mvd_cache)[2]= &h->mvd_cache[list][ scan8[index] ];
4916 pred_motion(h, index, block_width, list, h->ref_cache[list][ scan8[index] ], &mpx, &mpy);
4918 mx = mpx + decode_cabac_mb_mvd( h, list, index, 0 );
4919 my = mpy + decode_cabac_mb_mvd( h, list, index, 1 );
4920 tprintf("final mv:%d %d\n", mx, my);
4922 if(IS_SUB_8X8(sub_mb_type)){
4923 mv_cache[ 0 ][0]= mv_cache[ 1 ][0]=
4924 mv_cache[ 8 ][0]= mv_cache[ 9 ][0]= mx;
4925 mv_cache[ 0 ][1]= mv_cache[ 1 ][1]=
4926 mv_cache[ 8 ][1]= mv_cache[ 9 ][1]= my;
4928 mvd_cache[ 0 ][0]= mvd_cache[ 1 ][0]=
4929 mvd_cache[ 8 ][0]= mvd_cache[ 9 ][0]= mx - mpx;
4930 mvd_cache[ 0 ][1]= mvd_cache[ 1 ][1]=
4931 mvd_cache[ 8 ][1]= mvd_cache[ 9 ][1]= my - mpy;
4932 }else if(IS_SUB_8X4(sub_mb_type)){
4933 mv_cache[ 0 ][0]= mv_cache[ 1 ][0]= mx;
4934 mv_cache[ 0 ][1]= mv_cache[ 1 ][1]= my;
4936 mvd_cache[ 0 ][0]= mvd_cache[ 1 ][0]= mx- mpx;
4937 mvd_cache[ 0 ][1]= mvd_cache[ 1 ][1]= my - mpy;
4938 }else if(IS_SUB_4X8(sub_mb_type)){
4939 mv_cache[ 0 ][0]= mv_cache[ 8 ][0]= mx;
4940 mv_cache[ 0 ][1]= mv_cache[ 8 ][1]= my;
4942 mvd_cache[ 0 ][0]= mvd_cache[ 8 ][0]= mx - mpx;
4943 mvd_cache[ 0 ][1]= mvd_cache[ 8 ][1]= my - mpy;
4945 assert(IS_SUB_4X4(sub_mb_type));
4946 mv_cache[ 0 ][0]= mx;
4947 mv_cache[ 0 ][1]= my;
4949 mvd_cache[ 0 ][0]= mx - mpx;
4950 mvd_cache[ 0 ][1]= my - mpy;
4954 uint32_t *p= (uint32_t *)&h->mv_cache[list][ scan8[4*i] ][0];
4955 uint32_t *pd= (uint32_t *)&h->mvd_cache[list][ scan8[4*i] ][0];
4956 p[0] = p[1] = p[8] = p[9] = 0;
4957 pd[0]= pd[1]= pd[8]= pd[9]= 0;
4961 } else if( IS_DIRECT(mb_type) ) {
4962 pred_direct_motion(h, &mb_type);
4963 s->current_picture.mb_type[mb_xy]= mb_type;
4964 fill_rectangle(h->mvd_cache[0][scan8[0]], 4, 4, 8, 0, 4);
4965 fill_rectangle(h->mvd_cache[1][scan8[0]], 4, 4, 8, 0, 4);
4967 int list, mx, my, i, mpx, mpy;
4968 if(IS_16X16(mb_type)){
4969 for(list=0; list<2; list++){
4970 if(IS_DIR(mb_type, 0, list)){
4971 if(h->ref_count[list] > 0 ){
4972 const int ref = h->ref_count[list] > 1 ? decode_cabac_mb_ref( h, list, 0 ) : 0;
4973 fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, ref, 1);
4977 for(list=0; list<2; list++){
4978 if(IS_DIR(mb_type, 0, list)){
4979 pred_motion(h, 0, 4, list, h->ref_cache[list][ scan8[0] ], &mpx, &mpy);
4981 mx = mpx + decode_cabac_mb_mvd( h, list, 0, 0 );
4982 my = mpy + decode_cabac_mb_mvd( h, list, 0, 1 );
4983 tprintf("final mv:%d %d\n", mx, my);
4985 fill_rectangle(h->mvd_cache[list][ scan8[0] ], 4, 4, 8, pack16to32(mx-mpx,my-mpy), 4);
4986 fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, pack16to32(mx,my), 4);
4990 else if(IS_16X8(mb_type)){
4991 for(list=0; list<2; list++){
4992 if(h->ref_count[list]>0){
4994 if(IS_DIR(mb_type, i, list)){
4995 const int ref= h->ref_count[list] > 1 ? decode_cabac_mb_ref( h, list, 8*i ) : 0;
4996 fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, ref, 1);
4998 fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, (LIST_NOT_USED&0xFF), 1);
5002 for(list=0; list<2; list++){
5004 if(IS_DIR(mb_type, i, list)){
5005 pred_16x8_motion(h, 8*i, list, h->ref_cache[list][scan8[0] + 16*i], &mpx, &mpy);
5006 mx = mpx + decode_cabac_mb_mvd( h, list, 8*i, 0 );
5007 my = mpy + decode_cabac_mb_mvd( h, list, 8*i, 1 );
5008 tprintf("final mv:%d %d\n", mx, my);
5010 fill_rectangle(h->mvd_cache[list][ scan8[0] + 16*i ], 4, 2, 8, pack16to32(mx-mpx,my-mpy), 4);
5011 fill_rectangle(h->mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, pack16to32(mx,my), 4);
5012 }else{ // needed only for mixed refs
5013 fill_rectangle(h->mvd_cache[list][ scan8[0] + 16*i ], 4, 2, 8, 0, 4);
5014 fill_rectangle(h-> mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, 0, 4);
5019 assert(IS_8X16(mb_type));
5020 for(list=0; list<2; list++){
5021 if(h->ref_count[list]>0){
5023 if(IS_DIR(mb_type, i, list)){ //FIXME optimize
5024 const int ref= h->ref_count[list] > 1 ? decode_cabac_mb_ref( h, list, 4*i ) : 0;
5025 fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, ref, 1);
5027 fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, (LIST_NOT_USED&0xFF), 1);
5031 for(list=0; list<2; list++){
5033 if(IS_DIR(mb_type, i, list)){
5034 pred_8x16_motion(h, i*4, list, h->ref_cache[list][ scan8[0] + 2*i ], &mpx, &mpy);
5035 mx = mpx + decode_cabac_mb_mvd( h, list, 4*i, 0 );
5036 my = mpy + decode_cabac_mb_mvd( h, list, 4*i, 1 );
5038 tprintf("final mv:%d %d\n", mx, my);
5039 fill_rectangle(h->mvd_cache[list][ scan8[0] + 2*i ], 2, 4, 8, pack16to32(mx-mpx,my-mpy), 4);
5040 fill_rectangle(h->mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, pack16to32(mx,my), 4);
5041 }else{ // needed only for mixed refs
5042 fill_rectangle(h->mvd_cache[list][ scan8[0] + 2*i ], 2, 4, 8, 0, 4);
5043 fill_rectangle(h-> mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, 0, 4);
5050 if( IS_INTER( mb_type ) ) {
5051 h->chroma_pred_mode_table[mb_xy] = 0;
5052 write_back_motion( h, mb_type );
5055 if( !IS_INTRA16x16( mb_type ) ) {
5056 cbp = decode_cabac_mb_cbp_luma( h );
5057 cbp |= decode_cabac_mb_cbp_chroma( h ) << 4;
5060 h->cbp_table[mb_xy] = cbp;
5062 if( cbp || IS_INTRA16x16( mb_type ) ) {
5063 const uint8_t *scan, *dc_scan;
5066 if(IS_INTERLACED(mb_type)){
5068 dc_scan= luma_dc_field_scan;
5071 dc_scan= luma_dc_zigzag_scan;
5074 h->last_qscale_diff = dqp = decode_cabac_mb_dqp( h );
5076 if(((unsigned)s->qscale) > 51){
5077 if(s->qscale<0) s->qscale+= 52;
5078 else s->qscale-= 52;
5080 h->chroma_qp = get_chroma_qp(h, s->qscale);
5082 if( IS_INTRA16x16( mb_type ) ) {
5084 //av_log( s->avctx, AV_LOG_ERROR, "INTRA16x16 DC\n" );
5085 if( decode_cabac_residual( h, h->mb, 0, 0, dc_scan, s->qscale, 16) < 0)
5088 for( i = 0; i < 16; i++ ) {
5089 //av_log( s->avctx, AV_LOG_ERROR, "INTRA16x16 AC:%d\n", i );
5090 if( decode_cabac_residual(h, h->mb + 16*i, 1, i, scan + 1, s->qscale, 15) < 0 )
5094 fill_rectangle(&h->non_zero_count_cache[scan8[0]], 4, 4, 8, 0, 1);
5098 for( i8x8 = 0; i8x8 < 4; i8x8++ ) {
5099 if( cbp & (1<<i8x8) ) {
5100 for( i4x4 = 0; i4x4 < 4; i4x4++ ) {
5101 const int index = 4*i8x8 + i4x4;
5102 //av_log( s->avctx, AV_LOG_ERROR, "Luma4x4: %d\n", index );
5103 if( decode_cabac_residual(h, h->mb + 16*index, 2, index, scan, s->qscale, 16) < 0 )
5107 uint8_t * const nnz= &h->non_zero_count_cache[ scan8[4*i8x8] ];
5108 nnz[0] = nnz[1] = nnz[8] = nnz[9] = 0;
5115 for( c = 0; c < 2; c++ ) {
5116 //av_log( s->avctx, AV_LOG_ERROR, "INTRA C%d-DC\n",c );
5117 if( decode_cabac_residual(h, h->mb + 256 + 16*4*c, 3, c, chroma_dc_scan, h->chroma_qp, 4) < 0)
5124 for( c = 0; c < 2; c++ ) {
5125 for( i = 0; i < 4; i++ ) {
5126 const int index = 16 + 4 * c + i;
5127 //av_log( s->avctx, AV_LOG_ERROR, "INTRA C%d-AC %d\n",c, index - 16 );
5128 if( decode_cabac_residual(h, h->mb + 16*index, 4, index - 16, scan + 1, h->chroma_qp, 15) < 0)
5133 uint8_t * const nnz= &h->non_zero_count_cache[0];
5134 nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =
5135 nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
5138 uint8_t * const nnz= &h->non_zero_count_cache[0];
5139 fill_rectangle(&nnz[scan8[0]], 4, 4, 8, 0, 1);
5140 nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =
5141 nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
5144 s->current_picture.qscale_table[mb_xy]= s->qscale;
5145 write_back_non_zero_count(h);
5151 static void filter_mb_edgev( H264Context *h, uint8_t *pix, int stride, int bS[4], int qp ) {
5153 const int index_a = clip( qp + h->slice_alpha_c0_offset, 0, 51 );
5154 const int alpha = alpha_table[index_a];
5155 const int beta = beta_table[clip( qp + h->slice_beta_offset, 0, 51 )];
5157 for( i = 0; i < 4; i++ ) {
5164 const int tc0 = tc0_table[index_a][bS[i] - 1];
5165 /* 4px edge length */
5166 for( d = 0; d < 4; d++ ) {
5167 const int p0 = pix[-1];
5168 const int p1 = pix[-2];
5169 const int p2 = pix[-3];
5170 const int q0 = pix[0];
5171 const int q1 = pix[1];
5172 const int q2 = pix[2];
5174 if( ABS( p0 - q0 ) < alpha &&
5175 ABS( p1 - p0 ) < beta &&
5176 ABS( q1 - q0 ) < beta ) {
5180 if( ABS( p2 - p0 ) < beta ) {
5181 pix[-2] = p1 + clip( ( p2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( p1 << 1 ) ) >> 1, -tc0, tc0 );
5184 if( ABS( q2 - q0 ) < beta ) {
5185 pix[1] = q1 + clip( ( q2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( q1 << 1 ) ) >> 1, -tc0, tc0 );
5189 i_delta = clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );
5190 pix[-1] = clip_uint8( p0 + i_delta ); /* p0' */
5191 pix[0] = clip_uint8( q0 - i_delta ); /* q0' */
5196 /* 4px edge length */
5197 for( d = 0; d < 4; d++ ) {
5198 const int p0 = pix[-1];
5199 const int p1 = pix[-2];
5200 const int p2 = pix[-3];
5202 const int q0 = pix[0];
5203 const int q1 = pix[1];
5204 const int q2 = pix[2];
5206 if( ABS( p0 - q0 ) < alpha &&
5207 ABS( p1 - p0 ) < beta &&
5208 ABS( q1 - q0 ) < beta ) {
5210 if(ABS( p0 - q0 ) < (( alpha >> 2 ) + 2 )){
5211 if( ABS( p2 - p0 ) < beta)
5213 const int p3 = pix[-4];
5215 pix[-1] = ( p2 + 2*p1 + 2*p0 + 2*q0 + q1 + 4 ) >> 3;
5216 pix[-2] = ( p2 + p1 + p0 + q0 + 2 ) >> 2;
5217 pix[-3] = ( 2*p3 + 3*p2 + p1 + p0 + q0 + 4 ) >> 3;
5220 pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
5222 if( ABS( q2 - q0 ) < beta)
5224 const int q3 = pix[3];
5226 pix[0] = ( p1 + 2*p0 + 2*q0 + 2*q1 + q2 + 4 ) >> 3;
5227 pix[1] = ( p0 + q0 + q1 + q2 + 2 ) >> 2;
5228 pix[2] = ( 2*q3 + 3*q2 + q1 + q0 + p0 + 4 ) >> 3;
5231 pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
5235 pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
5236 pix[ 0] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
5244 static void filter_mb_edgecv( H264Context *h, uint8_t *pix, int stride, int bS[4], int qp ) {
5246 const int index_a = clip( qp + h->slice_alpha_c0_offset, 0, 51 );
5247 const int alpha = alpha_table[index_a];
5248 const int beta = beta_table[clip( qp + h->slice_beta_offset, 0, 51 )];
5250 for( i = 0; i < 4; i++ ) {
5257 const int tc = tc0_table[index_a][bS[i] - 1] + 1;
5258 /* 2px edge length (because we use same bS than the one for luma) */
5259 for( d = 0; d < 2; d++ ){
5260 const int p0 = pix[-1];
5261 const int p1 = pix[-2];
5262 const int q0 = pix[0];
5263 const int q1 = pix[1];
5265 if( ABS( p0 - q0 ) < alpha &&
5266 ABS( p1 - p0 ) < beta &&
5267 ABS( q1 - q0 ) < beta ) {
5268 const int i_delta = clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );
5270 pix[-1] = clip_uint8( p0 + i_delta ); /* p0' */
5271 pix[0] = clip_uint8( q0 - i_delta ); /* q0' */
5276 /* 2px edge length (because we use same bS than the one for luma) */
5277 for( d = 0; d < 2; d++ ){
5278 const int p0 = pix[-1];
5279 const int p1 = pix[-2];
5280 const int q0 = pix[0];
5281 const int q1 = pix[1];
5283 if( ABS( p0 - q0 ) < alpha &&
5284 ABS( p1 - p0 ) < beta &&
5285 ABS( q1 - q0 ) < beta ) {
5287 pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2; /* p0' */
5288 pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2; /* q0' */
5296 static void filter_mb_edgeh( H264Context *h, uint8_t *pix, int stride, int bS[4], int qp ) {
5298 const int index_a = clip( qp + h->slice_alpha_c0_offset, 0, 51 );
5299 const int alpha = alpha_table[index_a];
5300 const int beta = beta_table[clip( qp + h->slice_beta_offset, 0, 51 )];
5301 const int pix_next = stride;
5303 for( i = 0; i < 4; i++ ) {
5310 const int tc0 = tc0_table[index_a][bS[i] - 1];
5311 /* 4px edge length */
5312 for( d = 0; d < 4; d++ ) {
5313 const int p0 = pix[-1*pix_next];
5314 const int p1 = pix[-2*pix_next];
5315 const int p2 = pix[-3*pix_next];
5316 const int q0 = pix[0];
5317 const int q1 = pix[1*pix_next];
5318 const int q2 = pix[2*pix_next];
5320 if( ABS( p0 - q0 ) < alpha &&
5321 ABS( p1 - p0 ) < beta &&
5322 ABS( q1 - q0 ) < beta ) {
5327 if( ABS( p2 - p0 ) < beta ) {
5328 pix[-2*pix_next] = p1 + clip( ( p2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( p1 << 1 ) ) >> 1, -tc0, tc0 );
5331 if( ABS( q2 - q0 ) < beta ) {
5332 pix[pix_next] = q1 + clip( ( q2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( q1 << 1 ) ) >> 1, -tc0, tc0 );
5336 i_delta = clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );
5337 pix[-pix_next] = clip_uint8( p0 + i_delta ); /* p0' */
5338 pix[0] = clip_uint8( q0 - i_delta ); /* q0' */
5343 /* 4px edge length */
5344 for( d = 0; d < 4; d++ ) {
5345 const int p0 = pix[-1*pix_next];
5346 const int p1 = pix[-2*pix_next];
5347 const int p2 = pix[-3*pix_next];
5348 const int q0 = pix[0];
5349 const int q1 = pix[1*pix_next];
5350 const int q2 = pix[2*pix_next];
5352 if( ABS( p0 - q0 ) < alpha &&
5353 ABS( p1 - p0 ) < beta &&
5354 ABS( q1 - q0 ) < beta ) {
5356 const int p3 = pix[-4*pix_next];
5357 const int q3 = pix[ 3*pix_next];
5359 if(ABS( p0 - q0 ) < (( alpha >> 2 ) + 2 )){
5360 if( ABS( p2 - p0 ) < beta) {
5362 pix[-1*pix_next] = ( p2 + 2*p1 + 2*p0 + 2*q0 + q1 + 4 ) >> 3;
5363 pix[-2*pix_next] = ( p2 + p1 + p0 + q0 + 2 ) >> 2;
5364 pix[-3*pix_next] = ( 2*p3 + 3*p2 + p1 + p0 + q0 + 4 ) >> 3;
5367 pix[-1*pix_next] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
5369 if( ABS( q2 - q0 ) < beta) {
5371 pix[0*pix_next] = ( p1 + 2*p0 + 2*q0 + 2*q1 + q2 + 4 ) >> 3;
5372 pix[1*pix_next] = ( p0 + q0 + q1 + q2 + 2 ) >> 2;
5373 pix[2*pix_next] = ( 2*q3 + 3*q2 + q1 + q0 + p0 + 4 ) >> 3;
5376 pix[0*pix_next] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
5380 pix[-1*pix_next] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
5381 pix[ 0*pix_next] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
5390 static void filter_mb_edgech( H264Context *h, uint8_t *pix, int stride, int bS[4], int qp ) {
5392 const int index_a = clip( qp + h->slice_alpha_c0_offset, 0, 51 );
5393 const int alpha = alpha_table[index_a];
5394 const int beta = beta_table[clip( qp + h->slice_beta_offset, 0, 51 )];
5395 const int pix_next = stride;
5397 for( i = 0; i < 4; i++ )
5405 int tc = tc0_table[index_a][bS[i] - 1] + 1;
5406 /* 2px edge length (see deblocking_filter_edgecv) */
5407 for( d = 0; d < 2; d++ ) {
5408 const int p0 = pix[-1*pix_next];
5409 const int p1 = pix[-2*pix_next];
5410 const int q0 = pix[0];
5411 const int q1 = pix[1*pix_next];
5413 if( ABS( p0 - q0 ) < alpha &&
5414 ABS( p1 - p0 ) < beta &&
5415 ABS( q1 - q0 ) < beta ) {
5417 int i_delta = clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );
5419 pix[-pix_next] = clip_uint8( p0 + i_delta ); /* p0' */
5420 pix[0] = clip_uint8( q0 - i_delta ); /* q0' */
5425 /* 2px edge length (see deblocking_filter_edgecv) */
5426 for( d = 0; d < 2; d++ ) {
5427 const int p0 = pix[-1*pix_next];
5428 const int p1 = pix[-2*pix_next];
5429 const int q0 = pix[0];
5430 const int q1 = pix[1*pix_next];
5432 if( ABS( p0 - q0 ) < alpha &&
5433 ABS( p1 - p0 ) < beta &&
5434 ABS( q1 - q0 ) < beta ) {
5436 pix[-pix_next] = ( 2*p1 + p0 + q1 + 2 ) >> 2; /* p0' */
5437 pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2; /* q0' */
5445 static void filter_mb( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr) {
5446 MpegEncContext * const s = &h->s;
5447 const int mb_xy= mb_x + mb_y*s->mb_stride;
5448 int linesize, uvlinesize;
5451 /* FIXME Implement deblocking filter for field MB */
5452 if( h->sps.mb_aff ) {
5455 linesize = s->linesize;
5456 uvlinesize = s->uvlinesize;
5458 /* dir : 0 -> vertical edge, 1 -> horizontal edge */
5459 for( dir = 0; dir < 2; dir++ )
5464 /* test picture boundary */
5465 if( ( dir == 0 && mb_x == 0 ) || ( dir == 1 && mb_y == 0 ) ) {
5468 if( 0 == start && 2 == h->deblocking_filter) {
5469 const int mbn_xy = dir == 0 ? mb_xy -1 : mb_xy - s->mb_stride;
5470 if (h->slice_table[mbn_xy] != h->slice_table[mb_xy]) {
5476 for( edge = start; edge < 4; edge++ ) {
5477 /* mbn_xy: neighbour macroblock (how that works for field ?) */
5478 int mbn_xy = edge > 0 ? mb_xy : ( dir == 0 ? mb_xy -1 : mb_xy - s->mb_stride );
5482 if( IS_INTRA( s->current_picture.mb_type[mb_xy] ) ||
5483 IS_INTRA( s->current_picture.mb_type[mbn_xy] ) ) {
5484 bS[0] = bS[1] = bS[2] = bS[3] = ( edge == 0 ? 4 : 3 );
5487 const int slice_boundary = (h->slice_table[mbn_xy] != h->slice_table[mb_xy]);
5488 for( i = 0; i < 4; i++ ) {
5489 int x = dir == 0 ? edge : i;
5490 int y = dir == 0 ? i : edge;
5491 int b_idx= 8 + 4 + x + 8*y;
5492 int bn_idx= b_idx - (dir ? 8:1);
5493 uint8_t left_non_zero_count;
5494 if (slice_boundary) {
5495 // must not use non_zero_count_cache, it is not valid
5496 // across slice boundaries
5498 left_non_zero_count = h->non_zero_count[mbn_xy][6-i];
5500 left_non_zero_count = h->non_zero_count[mbn_xy][i];
5503 left_non_zero_count = h->non_zero_count_cache[bn_idx];
5506 if( h->non_zero_count_cache[b_idx] != 0 ||
5507 left_non_zero_count != 0 ) {
5510 else if( h->slice_type == P_TYPE ) {
5513 if (slice_boundary) {
5514 // must not use ref_cache and mv_cache, they are not
5515 // valid across slice boundaries
5517 left_ref = s->current_picture.ref_index[0][h->mb2b8_xy[mbn_xy] + (i>>1) * h->b8_stride + 1];
5518 *(uint32_t*)left_mv = *(uint32_t*)s->current_picture.motion_val[0][h->mb2b_xy[mbn_xy]+i*h->b_stride+3];
5520 left_ref = s->current_picture.ref_index[0][h->mb2b8_xy[mbn_xy] + (i>>1) + h->b8_stride];
5521 *(uint32_t*)left_mv = *(uint32_t*)s->current_picture.motion_val[0][h->mb2b_xy[mbn_xy]+3*h->b_stride+i];
5524 left_ref = h->ref_cache[0][bn_idx];
5525 *(uint32_t*)left_mv = *(uint32_t*)h->mv_cache[0][bn_idx];
5527 if( h->ref_cache[0][b_idx] != left_ref ||
5528 ABS( h->mv_cache[0][b_idx][0] - left_mv[0] ) >= 4 ||
5529 ABS( h->mv_cache[0][b_idx][1] - left_mv[1] ) >= 4 )
5535 /* FIXME Add support for B frame */
5540 if(bS[0]+bS[1]+bS[2]+bS[3] == 0)
5545 qp = ( s->qscale + s->current_picture.qscale_table[mbn_xy] + 1 ) >> 1;
5547 filter_mb_edgev( h, &img_y[4*edge], linesize, bS, qp );
5548 if( (edge&1) == 0 ) {
5549 int chroma_qp = ( h->chroma_qp +
5550 get_chroma_qp( h, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1;
5551 filter_mb_edgecv( h, &img_cb[2*edge], uvlinesize, bS, chroma_qp );
5552 filter_mb_edgecv( h, &img_cr[2*edge], uvlinesize, bS, chroma_qp );
5555 filter_mb_edgeh( h, &img_y[4*edge*linesize], linesize, bS, qp );
5556 if( (edge&1) == 0 ) {
5557 int chroma_qp = ( h->chroma_qp +
5558 get_chroma_qp( h, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1;
5559 filter_mb_edgech( h, &img_cb[2*edge*uvlinesize], uvlinesize, bS, chroma_qp );
5560 filter_mb_edgech( h, &img_cr[2*edge*uvlinesize], uvlinesize, bS, chroma_qp );
5567 static int decode_slice(H264Context *h){
5568 MpegEncContext * const s = &h->s;
5569 const int part_mask= s->partitioned_frame ? (AC_END|AC_ERROR) : 0x7F;
5573 if( h->pps.cabac ) {
5577 align_get_bits( &s->gb );
5580 ff_init_cabac_states( &h->cabac, ff_h264_lps_range, ff_h264_mps_state, ff_h264_lps_state, 64 );
5581 ff_init_cabac_decoder( &h->cabac,
5582 s->gb.buffer + get_bits_count(&s->gb)/8,
5583 ( s->gb.size_in_bits - get_bits_count(&s->gb) + 7)/8);
5584 /* calculate pre-state */
5585 for( i= 0; i < 399; i++ ) {
5587 if( h->slice_type == I_TYPE )
5588 pre = clip( ((cabac_context_init_I[i][0] * s->qscale) >>4 ) + cabac_context_init_I[i][1], 1, 126 );
5590 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 );
5593 h->cabac_state[i] = 2 * ( 63 - pre ) + 0;
5595 h->cabac_state[i] = 2 * ( pre - 64 ) + 1;
5599 int ret = decode_mb_cabac(h);
5600 int eos = get_cabac_terminate( &h->cabac ); /* End of Slice flag */
5602 if(ret>=0) hl_decode_mb(h);
5604 /* XXX: useless as decode_mb_cabac it doesn't support that ... */
5605 if( ret >= 0 && h->sps.mb_aff ) { //FIXME optimal? or let mb_decode decode 16x32 ?
5608 if(ret>=0) ret = decode_mb_cabac(h);
5609 eos = get_cabac_terminate( &h->cabac );
5615 if( ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 1) {
5616 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
5617 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);
5621 if( ++s->mb_x >= s->mb_width ) {
5623 ff_draw_horiz_band(s, 16*s->mb_y, 16);
5627 if( eos || s->mb_y >= s->mb_height ) {
5628 tprintf("slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
5629 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);
5633 /* TODO test over-reading in cabac code */
5634 else if( read too much in h->cabac ) {
5635 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);
5643 int ret = decode_mb_cavlc(h);
5645 if(ret>=0) hl_decode_mb(h);
5647 if(ret>=0 && h->sps.mb_aff){ //FIXME optimal? or let mb_decode decode 16x32 ?
5649 ret = decode_mb_cavlc(h);
5651 if(ret>=0) hl_decode_mb(h);
5656 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
5657 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);
5662 if(++s->mb_x >= s->mb_width){
5664 ff_draw_horiz_band(s, 16*s->mb_y, 16);
5665 if(++s->mb_y >= s->mb_height){
5666 tprintf("slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
5668 if(get_bits_count(&s->gb) == s->gb.size_in_bits ) {
5669 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);
5673 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);
5680 if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->mb_skip_run<=0){
5681 tprintf("slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
5682 if(get_bits_count(&s->gb) == s->gb.size_in_bits ){
5683 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);
5687 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);
5696 for(;s->mb_y < s->mb_height; s->mb_y++){
5697 for(;s->mb_x < s->mb_width; s->mb_x++){
5698 int ret= decode_mb(h);
5703 fprintf(stderr, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
5704 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);
5709 if(++s->mb_x >= s->mb_width){
5711 if(++s->mb_y >= s->mb_height){
5712 if(get_bits_count(s->gb) == s->gb.size_in_bits){
5713 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);
5717 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);
5724 if(get_bits_count(s->?gb) >= s->gb?.size_in_bits){
5725 if(get_bits_count(s->gb) == s->gb.size_in_bits){
5726 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);
5730 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);
5737 ff_draw_horiz_band(s, 16*s->mb_y, 16);
5740 return -1; //not reached
5743 static inline int decode_vui_parameters(H264Context *h, SPS *sps){
5744 MpegEncContext * const s = &h->s;
5745 int aspect_ratio_info_present_flag, aspect_ratio_idc;
5747 aspect_ratio_info_present_flag= get_bits1(&s->gb);
5749 if( aspect_ratio_info_present_flag ) {
5750 aspect_ratio_idc= get_bits(&s->gb, 8);
5751 if( aspect_ratio_idc == EXTENDED_SAR ) {
5752 sps->sar.num= get_bits(&s->gb, 16);
5753 sps->sar.den= get_bits(&s->gb, 16);
5754 }else if(aspect_ratio_idc < 16){
5755 sps->sar= pixel_aspect[aspect_ratio_idc];
5757 av_log(h->s.avctx, AV_LOG_ERROR, "illegal aspect ratio\n");
5764 // s->avctx->aspect_ratio= sar_width*s->width / (float)(s->height*sar_height);
5766 if(get_bits1(&s->gb)){ /* overscan_info_present_flag */
5767 get_bits1(&s->gb); /* overscan_appropriate_flag */
5770 if(get_bits1(&s->gb)){ /* video_signal_type_present_flag */
5771 get_bits(&s->gb, 3); /* video_format */
5772 get_bits1(&s->gb); /* video_full_range_flag */
5773 if(get_bits1(&s->gb)){ /* colour_description_present_flag */
5774 get_bits(&s->gb, 8); /* colour_primaries */
5775 get_bits(&s->gb, 8); /* transfer_characteristics */
5776 get_bits(&s->gb, 8); /* matrix_coefficients */
5780 if(get_bits1(&s->gb)){ /* chroma_location_info_present_flag */
5781 get_ue_golomb(&s->gb); /* chroma_sample_location_type_top_field */
5782 get_ue_golomb(&s->gb); /* chroma_sample_location_type_bottom_field */
5785 sps->timing_info_present_flag = get_bits1(&s->gb);
5786 if(sps->timing_info_present_flag){
5787 sps->num_units_in_tick = get_bits_long(&s->gb, 32);
5788 sps->time_scale = get_bits_long(&s->gb, 32);
5789 sps->fixed_frame_rate_flag = get_bits1(&s->gb);
5793 | nal_hrd_parameters_present_flag |0 |u(1) |
5794 | if( nal_hrd_parameters_present_flag = = 1) | | |
5795 | hrd_parameters( ) | | |
5796 | vcl_hrd_parameters_present_flag |0 |u(1) |
5797 | if( vcl_hrd_parameters_present_flag = = 1) | | |
5798 | hrd_parameters( ) | | |
5799 | if( ( nal_hrd_parameters_present_flag = = 1 | || | |
5801 |( vcl_hrd_parameters_present_flag = = 1 ) ) | | |
5802 | low_delay_hrd_flag |0 |u(1) |
5803 | bitstream_restriction_flag |0 |u(1) |
5804 | if( bitstream_restriction_flag ) { |0 |u(1) |
5805 | motion_vectors_over_pic_boundaries_flag |0 |u(1) |
5806 | max_bytes_per_pic_denom |0 |ue(v) |
5807 | max_bits_per_mb_denom |0 |ue(v) |
5808 | log2_max_mv_length_horizontal |0 |ue(v) |
5809 | log2_max_mv_length_vertical |0 |ue(v) |
5810 | num_reorder_frames |0 |ue(v) |
5811 | max_dec_frame_buffering |0 |ue(v) |
5818 static inline int decode_seq_parameter_set(H264Context *h){
5819 MpegEncContext * const s = &h->s;
5820 int profile_idc, level_idc;
5824 profile_idc= get_bits(&s->gb, 8);
5825 get_bits1(&s->gb); //constraint_set0_flag
5826 get_bits1(&s->gb); //constraint_set1_flag
5827 get_bits1(&s->gb); //constraint_set2_flag
5828 get_bits1(&s->gb); //constraint_set3_flag
5829 get_bits(&s->gb, 4); // reserved
5830 level_idc= get_bits(&s->gb, 8);
5831 sps_id= get_ue_golomb(&s->gb);
5833 sps= &h->sps_buffer[ sps_id ];
5834 sps->profile_idc= profile_idc;
5835 sps->level_idc= level_idc;
5837 sps->log2_max_frame_num= get_ue_golomb(&s->gb) + 4;
5838 sps->poc_type= get_ue_golomb(&s->gb);
5840 if(sps->poc_type == 0){ //FIXME #define
5841 sps->log2_max_poc_lsb= get_ue_golomb(&s->gb) + 4;
5842 } else if(sps->poc_type == 1){//FIXME #define
5843 sps->delta_pic_order_always_zero_flag= get_bits1(&s->gb);
5844 sps->offset_for_non_ref_pic= get_se_golomb(&s->gb);
5845 sps->offset_for_top_to_bottom_field= get_se_golomb(&s->gb);
5846 sps->poc_cycle_length= get_ue_golomb(&s->gb);
5848 for(i=0; i<sps->poc_cycle_length; i++)
5849 sps->offset_for_ref_frame[i]= get_se_golomb(&s->gb);
5851 if(sps->poc_type > 2){
5852 av_log(h->s.avctx, AV_LOG_ERROR, "illegal POC type %d\n", sps->poc_type);
5856 sps->ref_frame_count= get_ue_golomb(&s->gb);
5857 if(sps->ref_frame_count > MAX_PICTURE_COUNT-2){
5858 av_log(h->s.avctx, AV_LOG_ERROR, "too many reference frames\n");
5860 sps->gaps_in_frame_num_allowed_flag= get_bits1(&s->gb);
5861 sps->mb_width= get_ue_golomb(&s->gb) + 1;
5862 sps->mb_height= get_ue_golomb(&s->gb) + 1;
5863 sps->frame_mbs_only_flag= get_bits1(&s->gb);
5864 if(!sps->frame_mbs_only_flag)
5865 sps->mb_aff= get_bits1(&s->gb);
5869 sps->direct_8x8_inference_flag= get_bits1(&s->gb);
5871 sps->crop= get_bits1(&s->gb);
5873 sps->crop_left = get_ue_golomb(&s->gb);
5874 sps->crop_right = get_ue_golomb(&s->gb);
5875 sps->crop_top = get_ue_golomb(&s->gb);
5876 sps->crop_bottom= get_ue_golomb(&s->gb);
5877 if(sps->crop_left || sps->crop_top){
5878 av_log(h->s.avctx, AV_LOG_ERROR, "insane cropping not completly supported, this could look slightly wrong ...\n");
5884 sps->crop_bottom= 0;
5887 sps->vui_parameters_present_flag= get_bits1(&s->gb);
5888 if( sps->vui_parameters_present_flag )
5889 decode_vui_parameters(h, sps);
5891 if(s->avctx->debug&FF_DEBUG_PICT_INFO){
5892 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",
5893 sps_id, sps->profile_idc, sps->level_idc,
5895 sps->ref_frame_count,
5896 sps->mb_width, sps->mb_height,
5897 sps->frame_mbs_only_flag ? "FRM" : (sps->mb_aff ? "MB-AFF" : "PIC-AFF"),
5898 sps->direct_8x8_inference_flag ? "8B8" : "",
5899 sps->crop_left, sps->crop_right,
5900 sps->crop_top, sps->crop_bottom,
5901 sps->vui_parameters_present_flag ? "VUI" : ""
5907 static inline int decode_picture_parameter_set(H264Context *h){
5908 MpegEncContext * const s = &h->s;
5909 int pps_id= get_ue_golomb(&s->gb);
5910 PPS *pps= &h->pps_buffer[pps_id];
5912 pps->sps_id= get_ue_golomb(&s->gb);
5913 pps->cabac= get_bits1(&s->gb);
5914 pps->pic_order_present= get_bits1(&s->gb);
5915 pps->slice_group_count= get_ue_golomb(&s->gb) + 1;
5916 if(pps->slice_group_count > 1 ){
5917 pps->mb_slice_group_map_type= get_ue_golomb(&s->gb);
5918 av_log(h->s.avctx, AV_LOG_ERROR, "FMO not supported\n");
5919 switch(pps->mb_slice_group_map_type){
5922 | for( i = 0; i <= num_slice_groups_minus1; i++ ) | | |
5923 | run_length[ i ] |1 |ue(v) |
5928 | for( i = 0; i < num_slice_groups_minus1; i++ ) | | |
5930 | top_left_mb[ i ] |1 |ue(v) |
5931 | bottom_right_mb[ i ] |1 |ue(v) |
5939 | slice_group_change_direction_flag |1 |u(1) |
5940 | slice_group_change_rate_minus1 |1 |ue(v) |
5945 | slice_group_id_cnt_minus1 |1 |ue(v) |
5946 | for( i = 0; i <= slice_group_id_cnt_minus1; i++ | | |
5948 | slice_group_id[ i ] |1 |u(v) |
5953 pps->ref_count[0]= get_ue_golomb(&s->gb) + 1;
5954 pps->ref_count[1]= get_ue_golomb(&s->gb) + 1;
5955 if(pps->ref_count[0] > 32 || pps->ref_count[1] > 32){
5956 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow (pps)\n");
5960 pps->weighted_pred= get_bits1(&s->gb);
5961 pps->weighted_bipred_idc= get_bits(&s->gb, 2);
5962 pps->init_qp= get_se_golomb(&s->gb) + 26;
5963 pps->init_qs= get_se_golomb(&s->gb) + 26;
5964 pps->chroma_qp_index_offset= get_se_golomb(&s->gb);
5965 pps->deblocking_filter_parameters_present= get_bits1(&s->gb);
5966 pps->constrained_intra_pred= get_bits1(&s->gb);
5967 pps->redundant_pic_cnt_present = get_bits1(&s->gb);
5969 if(s->avctx->debug&FF_DEBUG_PICT_INFO){
5970 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",
5971 pps_id, pps->sps_id,
5972 pps->cabac ? "CABAC" : "CAVLC",
5973 pps->slice_group_count,
5974 pps->ref_count[0], pps->ref_count[1],
5975 pps->weighted_pred ? "weighted" : "",
5976 pps->init_qp, pps->init_qs, pps->chroma_qp_index_offset,
5977 pps->deblocking_filter_parameters_present ? "LPAR" : "",
5978 pps->constrained_intra_pred ? "CONSTR" : "",
5979 pps->redundant_pic_cnt_present ? "REDU" : ""
5987 * finds the end of the current frame in the bitstream.
5988 * @return the position of the first byte of the next frame, or -1
5990 static int find_frame_end(H264Context *h, const uint8_t *buf, int buf_size){
5993 ParseContext *pc = &(h->s.parse_context);
5994 //printf("first %02X%02X%02X%02X\n", buf[0], buf[1],buf[2],buf[3]);
5995 // mb_addr= pc->mb_addr - 1;
5997 for(i=0; i<=buf_size; i++){
5998 if((state&0xFFFFFF1F) == 0x101 || (state&0xFFFFFF1F) == 0x102 || (state&0xFFFFFF1F) == 0x105){
5999 tprintf("find_frame_end new startcode = %08x, frame_start_found = %d, pos = %d\n", state, pc->frame_start_found, i);
6000 if(pc->frame_start_found){
6001 // If there isn't one more byte in the buffer
6002 // the test on first_mb_in_slice cannot be done yet
6003 // do it at next call.
6004 if (i >= buf_size) break;
6005 if (buf[i] & 0x80) {
6006 // first_mb_in_slice is 0, probably the first nal of a new
6008 tprintf("find_frame_end frame_end_found, state = %08x, pos = %d\n", state, i);
6010 pc->frame_start_found= 0;
6014 pc->frame_start_found = 1;
6017 state= (state<<8) | buf[i];
6021 return END_NOT_FOUND;
6024 static int h264_parse(AVCodecParserContext *s,
6025 AVCodecContext *avctx,
6026 uint8_t **poutbuf, int *poutbuf_size,
6027 const uint8_t *buf, int buf_size)
6029 H264Context *h = s->priv_data;
6030 ParseContext *pc = &h->s.parse_context;
6033 next= find_frame_end(h, buf, buf_size);
6035 if (ff_combine_frame(pc, next, (uint8_t **)&buf, &buf_size) < 0) {
6041 *poutbuf = (uint8_t *)buf;
6042 *poutbuf_size = buf_size;
6046 static int decode_nal_units(H264Context *h, uint8_t *buf, int buf_size){
6047 MpegEncContext * const s = &h->s;
6048 AVCodecContext * const avctx= s->avctx;
6052 for(i=0; i<32; i++){
6053 printf("%X ", buf[i]);
6065 if(buf_index >= buf_size) break;
6067 for(i = 0; i < h->nal_length_size; i++)
6068 nalsize = (nalsize << 8) | buf[buf_index++];
6070 // start code prefix search
6071 for(; buf_index + 3 < buf_size; buf_index++){
6072 // this should allways succeed in the first iteration
6073 if(buf[buf_index] == 0 && buf[buf_index+1] == 0 && buf[buf_index+2] == 1)
6077 if(buf_index+3 >= buf_size) break;
6082 ptr= decode_nal(h, buf + buf_index, &dst_length, &consumed, h->is_avc ? nalsize : buf_size - buf_index);
6083 if(ptr[dst_length - 1] == 0) dst_length--;
6084 bit_length= 8*dst_length - decode_rbsp_trailing(ptr + dst_length - 1);
6086 if(s->avctx->debug&FF_DEBUG_STARTCODE){
6087 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);
6090 if (h->is_avc && (nalsize != consumed))
6091 av_log(h->s.avctx, AV_LOG_ERROR, "AVC: Consumed only %d bytes instead of %d\n", consumed, nalsize);
6093 buf_index += consumed;
6095 if( s->hurry_up == 1 && h->nal_ref_idc == 0 )
6098 switch(h->nal_unit_type){
6100 idr(h); //FIXME ensure we dont loose some frames if there is reordering
6102 init_get_bits(&s->gb, ptr, bit_length);
6104 h->inter_gb_ptr= &s->gb;
6105 s->data_partitioning = 0;
6107 if(decode_slice_header(h) < 0) return -1;
6108 if(h->redundant_pic_count==0 && s->hurry_up < 5 )
6112 init_get_bits(&s->gb, ptr, bit_length);
6114 h->inter_gb_ptr= NULL;
6115 s->data_partitioning = 1;
6117 if(decode_slice_header(h) < 0) return -1;
6120 init_get_bits(&h->intra_gb, ptr, bit_length);
6121 h->intra_gb_ptr= &h->intra_gb;
6124 init_get_bits(&h->inter_gb, ptr, bit_length);
6125 h->inter_gb_ptr= &h->inter_gb;
6127 if(h->redundant_pic_count==0 && h->intra_gb_ptr && s->data_partitioning && s->hurry_up < 5 )
6133 init_get_bits(&s->gb, ptr, bit_length);
6134 decode_seq_parameter_set(h);
6136 if(s->flags& CODEC_FLAG_LOW_DELAY)
6139 avctx->has_b_frames= !s->low_delay;
6142 init_get_bits(&s->gb, ptr, bit_length);
6144 decode_picture_parameter_set(h);
6147 case NAL_PICTURE_DELIMITER:
6149 case NAL_FILTER_DATA:
6152 av_log(avctx, AV_LOG_ERROR, "Unknown NAL code: %d\n", h->nal_unit_type);
6155 //FIXME move after where irt is set
6156 s->current_picture.pict_type= s->pict_type;
6157 s->current_picture.key_frame= s->pict_type == I_TYPE;
6160 if(!s->current_picture_ptr) return buf_index; //no frame
6162 h->prev_frame_num_offset= h->frame_num_offset;
6163 h->prev_frame_num= h->frame_num;
6164 if(s->current_picture_ptr->reference){
6165 h->prev_poc_msb= h->poc_msb;
6166 h->prev_poc_lsb= h->poc_lsb;
6168 if(s->current_picture_ptr->reference)
6169 execute_ref_pic_marking(h, h->mmco, h->mmco_index);
6171 assert(h->mmco_index==0);
6181 * retunrs the number of bytes consumed for building the current frame
6183 static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size){
6184 if(s->flags&CODEC_FLAG_TRUNCATED){
6185 pos -= s->parse_context.last_index;
6186 if(pos<0) pos=0; // FIXME remove (uneeded?)
6190 if(pos==0) pos=1; //avoid infinite loops (i doubt thats needed but ...)
6191 if(pos+10>buf_size) pos=buf_size; // oops ;)
6197 static int decode_frame(AVCodecContext *avctx,
6198 void *data, int *data_size,
6199 uint8_t *buf, int buf_size)
6201 H264Context *h = avctx->priv_data;
6202 MpegEncContext *s = &h->s;
6203 AVFrame *pict = data;
6206 s->flags= avctx->flags;
6207 s->flags2= avctx->flags2;
6209 /* no supplementary picture */
6210 if (buf_size == 0) {
6214 if(s->flags&CODEC_FLAG_TRUNCATED){
6215 int next= find_frame_end(h, buf, buf_size);
6217 if( ff_combine_frame(&s->parse_context, next, &buf, &buf_size) < 0 )
6219 //printf("next:%d buf_size:%d last_index:%d\n", next, buf_size, s->parse_context.last_index);
6222 if(h->is_avc && !h->got_avcC) {
6223 int i, cnt, nalsize;
6224 unsigned char *p = avctx->extradata;
6225 if(avctx->extradata_size < 7) {
6226 av_log(avctx, AV_LOG_ERROR, "avcC too short\n");
6230 av_log(avctx, AV_LOG_ERROR, "Unknown avcC version %d\n", *p);
6233 /* sps and pps in the avcC always have length coded with 2 bytes,
6234 so put a fake nal_length_size = 2 while parsing them */
6235 h->nal_length_size = 2;
6236 // Decode sps from avcC
6237 cnt = *(p+5) & 0x1f; // Number of sps
6239 for (i = 0; i < cnt; i++) {
6240 nalsize = BE_16(p) + 2;
6241 if(decode_nal_units(h, p, nalsize) != nalsize) {
6242 av_log(avctx, AV_LOG_ERROR, "Decoding sps %d from avcC failed\n", i);
6247 // Decode pps from avcC
6248 cnt = *(p++); // Number of pps
6249 for (i = 0; i < cnt; i++) {
6250 nalsize = BE_16(p) + 2;
6251 if(decode_nal_units(h, p, nalsize) != nalsize) {
6252 av_log(avctx, AV_LOG_ERROR, "Decoding pps %d from avcC failed\n", i);
6257 // Now store right nal length size, that will be use to parse all other nals
6258 h->nal_length_size = ((*(((char*)(avctx->extradata))+4))&0x03)+1;
6259 // Do not reparse avcC
6263 if(!h->is_avc && s->avctx->extradata_size && s->picture_number==0){
6264 if(0 < decode_nal_units(h, s->avctx->extradata, s->avctx->extradata_size) )
6268 buf_index=decode_nal_units(h, buf, buf_size);
6272 //FIXME do something with unavailable reference frames
6274 // if(ret==FRAME_SKIPED) return get_consumed_bytes(s, buf_index, buf_size);
6275 if(!s->current_picture_ptr){
6276 av_log(h->s.avctx, AV_LOG_DEBUG, "error, NO frame\n");
6281 /* Sort B-frames into display order
6282 * FIXME doesn't allow for multiple delayed frames */
6283 Picture *cur = s->current_picture_ptr;
6284 Picture *prev = h->delayed_pic[0];
6288 && (cur->pict_type == B_TYPE
6289 || (!h->sps.gaps_in_frame_num_allowed_flag
6290 && prev && cur->poc - prev->poc > 2))){
6292 s->avctx->has_b_frames = 1;
6293 if(prev && prev->poc > cur->poc)
6294 // too late to display this frame
6298 if(s->low_delay || !prev || cur->pict_type == B_TYPE)
6302 if(s->low_delay || !prev || out == prev){
6303 if(prev && prev->reference == 1)
6304 prev->reference = 0;
6305 h->delayed_pic[0] = cur;
6308 *pict= *(AVFrame*)out;
6311 ff_print_debug_info(s, pict);
6312 assert(pict->data[0]);
6313 //printf("out %d\n", (int)pict->data[0]);
6316 /* Return the Picture timestamp as the frame number */
6317 /* we substract 1 because it is added on utils.c */
6318 avctx->frame_number = s->picture_number - 1;
6321 /* dont output the last pic after seeking */
6322 if(s->last_picture_ptr || s->low_delay)
6323 //Note this isnt a issue as a IDR pic should flush the buffers
6325 *data_size = sizeof(AVFrame);
6326 return get_consumed_bytes(s, buf_index, buf_size);
6329 static inline void fill_mb_avail(H264Context *h){
6330 MpegEncContext * const s = &h->s;
6331 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
6334 h->mb_avail[0]= s->mb_x && h->slice_table[mb_xy - s->mb_stride - 1] == h->slice_num;
6335 h->mb_avail[1]= h->slice_table[mb_xy - s->mb_stride ] == h->slice_num;
6336 h->mb_avail[2]= s->mb_x+1 < s->mb_width && h->slice_table[mb_xy - s->mb_stride + 1] == h->slice_num;
6342 h->mb_avail[3]= s->mb_x && h->slice_table[mb_xy - 1] == h->slice_num;
6343 h->mb_avail[4]= 1; //FIXME move out
6344 h->mb_avail[5]= 0; //FIXME move out
6350 #define SIZE (COUNT*40)
6356 // int int_temp[10000];
6358 AVCodecContext avctx;
6360 dsputil_init(&dsp, &avctx);
6362 init_put_bits(&pb, temp, SIZE);
6363 printf("testing unsigned exp golomb\n");
6364 for(i=0; i<COUNT; i++){
6366 set_ue_golomb(&pb, i);
6367 STOP_TIMER("set_ue_golomb");
6369 flush_put_bits(&pb);
6371 init_get_bits(&gb, temp, 8*SIZE);
6372 for(i=0; i<COUNT; i++){
6375 s= show_bits(&gb, 24);
6378 j= get_ue_golomb(&gb);
6380 printf("missmatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
6383 STOP_TIMER("get_ue_golomb");
6387 init_put_bits(&pb, temp, SIZE);
6388 printf("testing signed exp golomb\n");
6389 for(i=0; i<COUNT; i++){
6391 set_se_golomb(&pb, i - COUNT/2);
6392 STOP_TIMER("set_se_golomb");
6394 flush_put_bits(&pb);
6396 init_get_bits(&gb, temp, 8*SIZE);
6397 for(i=0; i<COUNT; i++){
6400 s= show_bits(&gb, 24);
6403 j= get_se_golomb(&gb);
6404 if(j != i - COUNT/2){
6405 printf("missmatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
6408 STOP_TIMER("get_se_golomb");
6411 printf("testing 4x4 (I)DCT\n");
6414 uint8_t src[16], ref[16];
6415 uint64_t error= 0, max_error=0;
6417 for(i=0; i<COUNT; i++){
6419 // printf("%d %d %d\n", r1, r2, (r2-r1)*16);
6420 for(j=0; j<16; j++){
6421 ref[j]= random()%255;
6422 src[j]= random()%255;
6425 h264_diff_dct_c(block, src, ref, 4);
6428 for(j=0; j<16; j++){
6429 // printf("%d ", block[j]);
6430 block[j]= block[j]*4;
6431 if(j&1) block[j]= (block[j]*4 + 2)/5;
6432 if(j&4) block[j]= (block[j]*4 + 2)/5;
6436 s->dsp.h264_idct_add(ref, block, 4);
6437 /* for(j=0; j<16; j++){
6438 printf("%d ", ref[j]);
6442 for(j=0; j<16; j++){
6443 int diff= ABS(src[j] - ref[j]);
6446 max_error= FFMAX(max_error, diff);
6449 printf("error=%f max_error=%d\n", ((float)error)/COUNT/16, (int)max_error );
6451 printf("testing quantizer\n");
6452 for(qp=0; qp<52; qp++){
6454 src1_block[i]= src2_block[i]= random()%255;
6458 printf("Testing NAL layer\n");
6460 uint8_t bitstream[COUNT];
6461 uint8_t nal[COUNT*2];
6463 memset(&h, 0, sizeof(H264Context));
6465 for(i=0; i<COUNT; i++){
6473 for(j=0; j<COUNT; j++){
6474 bitstream[j]= (random() % 255) + 1;
6477 for(j=0; j<zeros; j++){
6478 int pos= random() % COUNT;
6479 while(bitstream[pos] == 0){
6488 nal_length= encode_nal(&h, nal, bitstream, COUNT, COUNT*2);
6490 printf("encoding failed\n");
6494 out= decode_nal(&h, nal, &out_length, &consumed, nal_length);
6498 if(out_length != COUNT){
6499 printf("incorrect length %d %d\n", out_length, COUNT);
6503 if(consumed != nal_length){
6504 printf("incorrect consumed length %d %d\n", nal_length, consumed);
6508 if(memcmp(bitstream, out, COUNT)){
6509 printf("missmatch\n");
6514 printf("Testing RBSP\n");
6522 static int decode_end(AVCodecContext *avctx)
6524 H264Context *h = avctx->priv_data;
6525 MpegEncContext *s = &h->s;
6527 free_tables(h); //FIXME cleanup init stuff perhaps
6530 // memset(h, 0, sizeof(H264Context));
6536 AVCodec h264_decoder = {
6540 sizeof(H264Context),
6545 /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 | CODEC_CAP_TRUNCATED,
6548 AVCodecParser h264_parser = {
6550 sizeof(H264Context),