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?
90 int bitstream_restriction_flag;
91 int num_reorder_frames;
95 * Picture parameter set
99 int cabac; ///< entropy_coding_mode_flag
100 int pic_order_present; ///< pic_order_present_flag
101 int slice_group_count; ///< num_slice_groups_minus1 + 1
102 int mb_slice_group_map_type;
103 int ref_count[2]; ///< num_ref_idx_l0/1_active_minus1 + 1
104 int weighted_pred; ///< weighted_pred_flag
105 int weighted_bipred_idc;
106 int init_qp; ///< pic_init_qp_minus26 + 26
107 int init_qs; ///< pic_init_qs_minus26 + 26
108 int chroma_qp_index_offset;
109 int deblocking_filter_parameters_present; ///< deblocking_filter_parameters_present_flag
110 int constrained_intra_pred; ///< constrained_intra_pred_flag
111 int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag
115 * Memory management control operation opcode.
117 typedef enum MMCOOpcode{
128 * Memory management control operation.
139 typedef struct H264Context{
147 #define NAL_IDR_SLICE 5
151 #define NAL_PICTURE_DELIMITER 9
152 #define NAL_FILTER_DATA 10
153 uint8_t *rbsp_buffer;
154 int rbsp_buffer_size;
157 * Used to parse AVC variant of h264
159 int is_avc; ///< this flag is != 0 if codec is avc1
160 int got_avcC; ///< flag used to parse avcC data only once
161 int nal_length_size; ///< Number of bytes used for nal length (1, 2 or 4)
165 int prev_mb_skipped; //FIXME remove (IMHO not used)
168 int chroma_pred_mode;
169 int intra16x16_pred_mode;
174 int8_t intra4x4_pred_mode_cache[5*8];
175 int8_t (*intra4x4_pred_mode)[8];
176 void (*pred4x4 [9+3])(uint8_t *src, uint8_t *topright, int stride);//FIXME move to dsp?
177 void (*pred8x8 [4+3])(uint8_t *src, int stride);
178 void (*pred16x16[4+3])(uint8_t *src, int stride);
179 unsigned int topleft_samples_available;
180 unsigned int top_samples_available;
181 unsigned int topright_samples_available;
182 unsigned int left_samples_available;
183 uint8_t (*top_borders[2])[16+2*8];
184 uint8_t left_border[2*(17+2*9)];
187 * non zero coeff count cache.
188 * is 64 if not available.
190 uint8_t non_zero_count_cache[6*8] __align8;
191 uint8_t (*non_zero_count)[16];
194 * Motion vector cache.
196 int16_t mv_cache[2][5*8][2] __align8;
197 int8_t ref_cache[2][5*8] __align8;
198 #define LIST_NOT_USED -1 //FIXME rename?
199 #define PART_NOT_AVAILABLE -2
202 * is 1 if the specific list MV&references are set to 0,0,-2.
204 int mv_cache_clean[2];
207 * block_offset[ 0..23] for frame macroblocks
208 * block_offset[24..47] for field macroblocks
210 int block_offset[2*(16+8)];
212 uint32_t *mb2b_xy; //FIXME are these 4 a good idea?
214 int b_stride; //FIXME use s->b4_stride
220 int unknown_svq3_flag;
221 int next_slice_index;
223 SPS sps_buffer[MAX_SPS_COUNT];
224 SPS sps; ///< current sps
226 PPS pps_buffer[MAX_PPS_COUNT];
230 PPS pps; //FIXME move to Picture perhaps? (->no) do we need that?
233 uint8_t *slice_table_base;
234 uint8_t *slice_table; ///< slice_table_base + mb_stride + 1
236 int slice_type_fixed;
238 //interlacing specific flags
240 int mb_field_decoding_flag;
247 int delta_poc_bottom;
250 int prev_poc_msb; ///< poc_msb of the last reference pic for POC type 0
251 int prev_poc_lsb; ///< poc_lsb of the last reference pic for POC type 0
252 int frame_num_offset; ///< for POC type 2
253 int prev_frame_num_offset; ///< for POC type 2
254 int prev_frame_num; ///< frame_num of the last pic for POC type 1/2
257 * frame_num for frames or 2*frame_num for field pics.
262 * max_frame_num or 2*max_frame_num for field pics.
266 //Weighted pred stuff
268 int use_weight_chroma;
269 int luma_log2_weight_denom;
270 int chroma_log2_weight_denom;
271 int luma_weight[2][16];
272 int luma_offset[2][16];
273 int chroma_weight[2][16][2];
274 int chroma_offset[2][16][2];
275 int implicit_weight[16][16];
278 int deblocking_filter; ///< disable_deblocking_filter_idc with 1<->0
279 int slice_alpha_c0_offset;
280 int slice_beta_offset;
282 int redundant_pic_count;
284 int direct_spatial_mv_pred;
285 int dist_scale_factor[16];
286 int map_col_to_list0[2][16];
289 * num_ref_idx_l0/1_active_minus1 + 1
291 int ref_count[2];// FIXME split for AFF
292 Picture *short_ref[32];
293 Picture *long_ref[32];
294 Picture default_ref_list[2][32];
295 Picture ref_list[2][32]; //FIXME size?
296 Picture field_ref_list[2][32]; //FIXME size?
297 Picture *delayed_pic[16]; //FIXME size?
298 Picture *delayed_output_pic;
301 * memory management control operations buffer.
303 MMCO mmco[MAX_MMCO_COUNT];
306 int long_ref_count; ///< number of actual long term references
307 int short_ref_count; ///< number of actual short term references
310 GetBitContext intra_gb;
311 GetBitContext inter_gb;
312 GetBitContext *intra_gb_ptr;
313 GetBitContext *inter_gb_ptr;
315 DCTELEM mb[16*24] __align8;
321 uint8_t cabac_state[399];
324 /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0,1,2), 0x0? luma_cbp */
328 /* chroma_pred_mode for i4x4 or i16x16, else 0 */
329 uint8_t *chroma_pred_mode_table;
330 int last_qscale_diff;
331 int16_t (*mvd_table[2])[2];
332 int16_t mvd_cache[2][5*8][2] __align8;
333 uint8_t *direct_table;
334 uint8_t direct_cache[5*8];
338 static VLC coeff_token_vlc[4];
339 static VLC chroma_dc_coeff_token_vlc;
341 static VLC total_zeros_vlc[15];
342 static VLC chroma_dc_total_zeros_vlc[3];
344 static VLC run_vlc[6];
347 static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
348 static void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
349 static void filter_mb( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr, unsigned int linesize, unsigned int uvlinesize);
351 static inline uint32_t pack16to32(int a, int b){
352 #ifdef WORDS_BIGENDIAN
353 return (b&0xFFFF) + (a<<16);
355 return (a&0xFFFF) + (b<<16);
361 * @param h height of the rectangle, should be a constant
362 * @param w width of the rectangle, should be a constant
363 * @param size the size of val (1 or 4), should be a constant
365 static inline void fill_rectangle(void *vp, int w, int h, int stride, uint32_t val, int size){ //FIXME ensure this IS inlined
366 uint8_t *p= (uint8_t*)vp;
367 assert(size==1 || size==4);
372 assert((((int)vp)&(FFMIN(w, STRIDE_ALIGN)-1)) == 0);
373 //FIXME check what gcc generates for 64 bit on x86 and possibly write a 32 bit ver of it
376 *(uint16_t*)(p + stride)= size==4 ? val : val*0x0101;
377 }else if(w==2 && h==4){
378 *(uint16_t*)(p + 0*stride)=
379 *(uint16_t*)(p + 1*stride)=
380 *(uint16_t*)(p + 2*stride)=
381 *(uint16_t*)(p + 3*stride)= size==4 ? val : val*0x0101;
382 }else if(w==4 && h==1){
383 *(uint32_t*)(p + 0*stride)= size==4 ? val : val*0x01010101;
384 }else if(w==4 && h==2){
385 *(uint32_t*)(p + 0*stride)=
386 *(uint32_t*)(p + 1*stride)= size==4 ? val : val*0x01010101;
387 }else if(w==4 && h==4){
388 *(uint32_t*)(p + 0*stride)=
389 *(uint32_t*)(p + 1*stride)=
390 *(uint32_t*)(p + 2*stride)=
391 *(uint32_t*)(p + 3*stride)= size==4 ? val : val*0x01010101;
392 }else if(w==8 && h==1){
394 *(uint32_t*)(p + 4)= size==4 ? val : val*0x01010101;
395 }else if(w==8 && h==2){
396 *(uint32_t*)(p + 0 + 0*stride)=
397 *(uint32_t*)(p + 4 + 0*stride)=
398 *(uint32_t*)(p + 0 + 1*stride)=
399 *(uint32_t*)(p + 4 + 1*stride)= size==4 ? val : val*0x01010101;
400 }else if(w==8 && h==4){
401 *(uint64_t*)(p + 0*stride)=
402 *(uint64_t*)(p + 1*stride)=
403 *(uint64_t*)(p + 2*stride)=
404 *(uint64_t*)(p + 3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
405 }else if(w==16 && h==2){
406 *(uint64_t*)(p + 0+0*stride)=
407 *(uint64_t*)(p + 8+0*stride)=
408 *(uint64_t*)(p + 0+1*stride)=
409 *(uint64_t*)(p + 8+1*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
410 }else if(w==16 && h==4){
411 *(uint64_t*)(p + 0+0*stride)=
412 *(uint64_t*)(p + 8+0*stride)=
413 *(uint64_t*)(p + 0+1*stride)=
414 *(uint64_t*)(p + 8+1*stride)=
415 *(uint64_t*)(p + 0+2*stride)=
416 *(uint64_t*)(p + 8+2*stride)=
417 *(uint64_t*)(p + 0+3*stride)=
418 *(uint64_t*)(p + 8+3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
423 static inline void fill_caches(H264Context *h, int mb_type, int for_deblock){
424 MpegEncContext * const s = &h->s;
425 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
426 int topleft_xy, top_xy, topright_xy, left_xy[2];
427 int topleft_type, top_type, topright_type, left_type[2];
431 //FIXME deblocking can skip fill_caches much of the time with multiple slices too.
432 // the actual condition is whether we're on the edge of a slice,
433 // and even then the intra and nnz parts are unnecessary.
434 if(for_deblock && h->slice_num == 1)
437 //wow what a mess, why didn't they simplify the interlacing&intra stuff, i can't imagine that these complex rules are worth it
439 top_xy = mb_xy - s->mb_stride;
440 topleft_xy = top_xy - 1;
441 topright_xy= top_xy + 1;
442 left_xy[1] = left_xy[0] = mb_xy-1;
452 const int pair_xy = s->mb_x + (s->mb_y & ~1)*s->mb_stride;
453 const int top_pair_xy = pair_xy - s->mb_stride;
454 const int topleft_pair_xy = top_pair_xy - 1;
455 const int topright_pair_xy = top_pair_xy + 1;
456 const int topleft_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[topleft_pair_xy]);
457 const int top_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[top_pair_xy]);
458 const int topright_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[topright_pair_xy]);
459 const int left_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[pair_xy-1]);
460 const int curr_mb_frame_flag = !IS_INTERLACED(mb_type);
461 const int bottom = (s->mb_y & 1);
462 tprintf("fill_caches: curr_mb_frame_flag:%d, left_mb_frame_flag:%d, topleft_mb_frame_flag:%d, top_mb_frame_flag:%d, topright_mb_frame_flag:%d\n", curr_mb_frame_flag, left_mb_frame_flag, topleft_mb_frame_flag, top_mb_frame_flag, topright_mb_frame_flag);
464 ? !curr_mb_frame_flag // bottom macroblock
465 : (!curr_mb_frame_flag && !top_mb_frame_flag) // top macroblock
467 top_xy -= s->mb_stride;
470 ? !curr_mb_frame_flag // bottom macroblock
471 : (!curr_mb_frame_flag && !topleft_mb_frame_flag) // top macroblock
473 topleft_xy -= s->mb_stride;
476 ? !curr_mb_frame_flag // bottom macroblock
477 : (!curr_mb_frame_flag && !topright_mb_frame_flag) // top macroblock
479 topright_xy -= s->mb_stride;
481 if (left_mb_frame_flag != curr_mb_frame_flag) {
482 left_xy[1] = left_xy[0] = pair_xy - 1;
483 if (curr_mb_frame_flag) {
504 left_xy[1] += s->mb_stride;
517 h->top_mb_xy = top_xy;
518 h->left_mb_xy[0] = left_xy[0];
519 h->left_mb_xy[1] = left_xy[1];
521 topleft_type = h->slice_table[topleft_xy ] < 255 ? s->current_picture.mb_type[topleft_xy] : 0;
522 top_type = h->slice_table[top_xy ] < 255 ? s->current_picture.mb_type[top_xy] : 0;
523 topright_type= h->slice_table[topright_xy] < 255 ? s->current_picture.mb_type[topright_xy]: 0;
524 left_type[0] = h->slice_table[left_xy[0] ] < 255 ? s->current_picture.mb_type[left_xy[0]] : 0;
525 left_type[1] = h->slice_table[left_xy[1] ] < 255 ? s->current_picture.mb_type[left_xy[1]] : 0;
527 topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
528 top_type = h->slice_table[top_xy ] == h->slice_num ? s->current_picture.mb_type[top_xy] : 0;
529 topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
530 left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
531 left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
534 if(IS_INTRA(mb_type)){
535 h->topleft_samples_available=
536 h->top_samples_available=
537 h->left_samples_available= 0xFFFF;
538 h->topright_samples_available= 0xEEEA;
540 if(!IS_INTRA(top_type) && (top_type==0 || h->pps.constrained_intra_pred)){
541 h->topleft_samples_available= 0xB3FF;
542 h->top_samples_available= 0x33FF;
543 h->topright_samples_available= 0x26EA;
546 if(!IS_INTRA(left_type[i]) && (left_type[i]==0 || h->pps.constrained_intra_pred)){
547 h->topleft_samples_available&= 0xDF5F;
548 h->left_samples_available&= 0x5F5F;
552 if(!IS_INTRA(topleft_type) && (topleft_type==0 || h->pps.constrained_intra_pred))
553 h->topleft_samples_available&= 0x7FFF;
555 if(!IS_INTRA(topright_type) && (topright_type==0 || h->pps.constrained_intra_pred))
556 h->topright_samples_available&= 0xFBFF;
558 if(IS_INTRA4x4(mb_type)){
559 if(IS_INTRA4x4(top_type)){
560 h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
561 h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
562 h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
563 h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
566 if(!top_type || (IS_INTER(top_type) && h->pps.constrained_intra_pred))
571 h->intra4x4_pred_mode_cache[4+8*0]=
572 h->intra4x4_pred_mode_cache[5+8*0]=
573 h->intra4x4_pred_mode_cache[6+8*0]=
574 h->intra4x4_pred_mode_cache[7+8*0]= pred;
577 if(IS_INTRA4x4(left_type[i])){
578 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
579 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
582 if(!left_type[i] || (IS_INTER(left_type[i]) && h->pps.constrained_intra_pred))
587 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
588 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
603 //FIXME constraint_intra_pred & partitioning & nnz (lets hope this is just a typo in the spec)
605 h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][4];
606 h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][5];
607 h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][6];
608 h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3];
610 h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][9];
611 h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8];
613 h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][12];
614 h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11];
617 h->non_zero_count_cache[4+8*0]=
618 h->non_zero_count_cache[5+8*0]=
619 h->non_zero_count_cache[6+8*0]=
620 h->non_zero_count_cache[7+8*0]=
622 h->non_zero_count_cache[1+8*0]=
623 h->non_zero_count_cache[2+8*0]=
625 h->non_zero_count_cache[1+8*3]=
626 h->non_zero_count_cache[2+8*3]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
630 for (i=0; i<2; i++) {
632 h->non_zero_count_cache[3+8*1 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[0+2*i]];
633 h->non_zero_count_cache[3+8*2 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[1+2*i]];
634 h->non_zero_count_cache[0+8*1 + 8*i]= h->non_zero_count[left_xy[i]][left_block[4+2*i]];
635 h->non_zero_count_cache[0+8*4 + 8*i]= h->non_zero_count[left_xy[i]][left_block[5+2*i]];
637 h->non_zero_count_cache[3+8*1 + 2*8*i]=
638 h->non_zero_count_cache[3+8*2 + 2*8*i]=
639 h->non_zero_count_cache[0+8*1 + 8*i]=
640 h->non_zero_count_cache[0+8*4 + 8*i]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
647 h->top_cbp = h->cbp_table[top_xy];
648 } else if(IS_INTRA(mb_type)) {
655 h->left_cbp = h->cbp_table[left_xy[0]] & 0x1f0;
656 } else if(IS_INTRA(mb_type)) {
662 h->left_cbp |= ((h->cbp_table[left_xy[0]]>>((left_block[0]&(~1))+1))&0x1) << 1;
665 h->left_cbp |= ((h->cbp_table[left_xy[1]]>>((left_block[2]&(~1))+1))&0x1) << 3;
670 //FIXME direct mb can skip much of this
671 if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
673 for(list=0; list<1+(h->slice_type==B_TYPE); list++){
674 if(!USES_LIST(mb_type, list) && !IS_DIRECT(mb_type) && !h->deblocking_filter){
675 /*if(!h->mv_cache_clean[list]){
676 memset(h->mv_cache [list], 0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
677 memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
678 h->mv_cache_clean[list]= 1;
682 h->mv_cache_clean[list]= 0;
684 if(IS_INTER(top_type)){
685 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
686 const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
687 *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0];
688 *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];
689 *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];
690 *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];
691 h->ref_cache[list][scan8[0] + 0 - 1*8]=
692 h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
693 h->ref_cache[list][scan8[0] + 2 - 1*8]=
694 h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
696 *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]=
697 *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]=
698 *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]=
699 *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;
700 *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
703 //FIXME unify cleanup or sth
704 if(IS_INTER(left_type[0])){
705 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
706 const int b8_xy= h->mb2b8_xy[left_xy[0]] + 1;
707 *(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]];
708 *(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]];
709 h->ref_cache[list][scan8[0] - 1 + 0*8]=
710 h->ref_cache[list][scan8[0] - 1 + 1*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0]>>1)];
712 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 0*8]=
713 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 1*8]= 0;
714 h->ref_cache[list][scan8[0] - 1 + 0*8]=
715 h->ref_cache[list][scan8[0] - 1 + 1*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
718 if(IS_INTER(left_type[1])){
719 const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
720 const int b8_xy= h->mb2b8_xy[left_xy[1]] + 1;
721 *(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]];
722 *(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]];
723 h->ref_cache[list][scan8[0] - 1 + 2*8]=
724 h->ref_cache[list][scan8[0] - 1 + 3*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[2]>>1)];
726 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 2*8]=
727 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 3*8]= 0;
728 h->ref_cache[list][scan8[0] - 1 + 2*8]=
729 h->ref_cache[list][scan8[0] - 1 + 3*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
732 if(for_deblock || (IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred))
735 if(IS_INTER(topleft_type)){
736 const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
737 const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + h->b8_stride;
738 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
739 h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
741 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
742 h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
745 if(IS_INTER(topright_type)){
746 const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
747 const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
748 *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
749 h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
751 *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
752 h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
756 h->ref_cache[list][scan8[5 ]+1] =
757 h->ref_cache[list][scan8[7 ]+1] =
758 h->ref_cache[list][scan8[13]+1] = //FIXME remove past 3 (init somewhere else)
759 h->ref_cache[list][scan8[4 ]] =
760 h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
761 *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
762 *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
763 *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
764 *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
765 *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
768 /* XXX beurk, Load mvd */
769 if(IS_INTER(topleft_type)){
770 const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
771 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy];
773 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 - 1*8]= 0;
776 if(IS_INTER(top_type)){
777 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
778 *(uint32_t*)h->mvd_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 0];
779 *(uint32_t*)h->mvd_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 1];
780 *(uint32_t*)h->mvd_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 2];
781 *(uint32_t*)h->mvd_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 3];
783 *(uint32_t*)h->mvd_cache [list][scan8[0] + 0 - 1*8]=
784 *(uint32_t*)h->mvd_cache [list][scan8[0] + 1 - 1*8]=
785 *(uint32_t*)h->mvd_cache [list][scan8[0] + 2 - 1*8]=
786 *(uint32_t*)h->mvd_cache [list][scan8[0] + 3 - 1*8]= 0;
788 if(IS_INTER(left_type[0])){
789 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
790 *(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]];
791 *(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]];
793 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 0*8]=
794 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 1*8]= 0;
796 if(IS_INTER(left_type[1])){
797 const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
798 *(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]];
799 *(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]];
801 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 2*8]=
802 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 3*8]= 0;
804 *(uint32_t*)h->mvd_cache [list][scan8[5 ]+1]=
805 *(uint32_t*)h->mvd_cache [list][scan8[7 ]+1]=
806 *(uint32_t*)h->mvd_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
807 *(uint32_t*)h->mvd_cache [list][scan8[4 ]]=
808 *(uint32_t*)h->mvd_cache [list][scan8[12]]= 0;
810 if(h->slice_type == B_TYPE){
811 fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, 0, 1);
813 if(IS_DIRECT(top_type)){
814 *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0x01010101;
815 }else if(IS_8X8(top_type)){
816 int b8_xy = h->mb2b8_xy[top_xy] + h->b8_stride;
817 h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy];
818 h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 1];
820 *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0;
824 if(IS_DIRECT(left_type[0])){
825 h->direct_cache[scan8[0] - 1 + 0*8]=
826 h->direct_cache[scan8[0] - 1 + 2*8]= 1;
827 }else if(IS_8X8(left_type[0])){
828 int b8_xy = h->mb2b8_xy[left_xy[0]] + 1;
829 h->direct_cache[scan8[0] - 1 + 0*8]= h->direct_table[b8_xy];
830 h->direct_cache[scan8[0] - 1 + 2*8]= h->direct_table[b8_xy + h->b8_stride];
832 h->direct_cache[scan8[0] - 1 + 0*8]=
833 h->direct_cache[scan8[0] - 1 + 2*8]= 0;
842 static inline void write_back_intra_pred_mode(H264Context *h){
843 MpegEncContext * const s = &h->s;
844 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
846 h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
847 h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
848 h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
849 h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
850 h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
851 h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
852 h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
856 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
858 static inline int check_intra4x4_pred_mode(H264Context *h){
859 MpegEncContext * const s = &h->s;
860 static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
861 static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
864 if(!(h->top_samples_available&0x8000)){
866 int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
868 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);
871 h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
876 if(!(h->left_samples_available&0x8000)){
878 int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
880 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);
883 h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
889 } //FIXME cleanup like next
892 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
894 static inline int check_intra_pred_mode(H264Context *h, int mode){
895 MpegEncContext * const s = &h->s;
896 static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
897 static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
899 if(mode < 0 || mode > 6) {
900 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);
904 if(!(h->top_samples_available&0x8000)){
907 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);
912 if(!(h->left_samples_available&0x8000)){
915 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);
924 * gets the predicted intra4x4 prediction mode.
926 static inline int pred_intra_mode(H264Context *h, int n){
927 const int index8= scan8[n];
928 const int left= h->intra4x4_pred_mode_cache[index8 - 1];
929 const int top = h->intra4x4_pred_mode_cache[index8 - 8];
930 const int min= FFMIN(left, top);
932 tprintf("mode:%d %d min:%d\n", left ,top, min);
934 if(min<0) return DC_PRED;
938 static inline void write_back_non_zero_count(H264Context *h){
939 MpegEncContext * const s = &h->s;
940 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
942 h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[7+8*1];
943 h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[7+8*2];
944 h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[7+8*3];
945 h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];
946 h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[4+8*4];
947 h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[5+8*4];
948 h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[6+8*4];
950 h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[1+8*2];
951 h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
952 h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[2+8*1];
954 h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[1+8*5];
955 h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
956 h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[2+8*4];
960 * gets the predicted number of non zero coefficients.
961 * @param n block index
963 static inline int pred_non_zero_count(H264Context *h, int n){
964 const int index8= scan8[n];
965 const int left= h->non_zero_count_cache[index8 - 1];
966 const int top = h->non_zero_count_cache[index8 - 8];
969 if(i<64) i= (i+1)>>1;
971 tprintf("pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
976 static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
977 const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];
979 if(topright_ref != PART_NOT_AVAILABLE){
980 *C= h->mv_cache[list][ i - 8 + part_width ];
983 tprintf("topright MV not available\n");
985 *C= h->mv_cache[list][ i - 8 - 1 ];
986 return h->ref_cache[list][ i - 8 - 1 ];
991 * gets the predicted MV.
992 * @param n the block index
993 * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)
994 * @param mx the x component of the predicted motion vector
995 * @param my the y component of the predicted motion vector
997 static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
998 const int index8= scan8[n];
999 const int top_ref= h->ref_cache[list][ index8 - 8 ];
1000 const int left_ref= h->ref_cache[list][ index8 - 1 ];
1001 const int16_t * const A= h->mv_cache[list][ index8 - 1 ];
1002 const int16_t * const B= h->mv_cache[list][ index8 - 8 ];
1004 int diagonal_ref, match_count;
1006 assert(part_width==1 || part_width==2 || part_width==4);
1016 diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);
1017 match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);
1018 tprintf("pred_motion match_count=%d\n", match_count);
1019 if(match_count > 1){ //most common
1020 *mx= mid_pred(A[0], B[0], C[0]);
1021 *my= mid_pred(A[1], B[1], C[1]);
1022 }else if(match_count==1){
1026 }else if(top_ref==ref){
1034 if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
1038 *mx= mid_pred(A[0], B[0], C[0]);
1039 *my= mid_pred(A[1], B[1], C[1]);
1043 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);
1047 * gets the directionally predicted 16x8 MV.
1048 * @param n the block index
1049 * @param mx the x component of the predicted motion vector
1050 * @param my the y component of the predicted motion vector
1052 static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
1054 const int top_ref= h->ref_cache[list][ scan8[0] - 8 ];
1055 const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
1057 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);
1065 const int left_ref= h->ref_cache[list][ scan8[8] - 1 ];
1066 const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];
1068 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);
1070 if(left_ref == ref){
1078 pred_motion(h, n, 4, list, ref, mx, my);
1082 * gets the directionally predicted 8x16 MV.
1083 * @param n the block index
1084 * @param mx the x component of the predicted motion vector
1085 * @param my the y component of the predicted motion vector
1087 static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
1089 const int left_ref= h->ref_cache[list][ scan8[0] - 1 ];
1090 const int16_t * const A= h->mv_cache[list][ scan8[0] - 1 ];
1092 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);
1094 if(left_ref == ref){
1103 diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);
1105 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);
1107 if(diagonal_ref == ref){
1115 pred_motion(h, n, 2, list, ref, mx, my);
1118 static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
1119 const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
1120 const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];
1122 tprintf("pred_pskip: (%d) (%d) at %2d %2d\n", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
1124 if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
1125 || (top_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0)
1126 || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){
1132 pred_motion(h, 0, 4, 0, 0, mx, my);
1137 static inline void direct_dist_scale_factor(H264Context * const h){
1138 const int poc = h->s.current_picture_ptr->poc;
1139 const int poc1 = h->ref_list[1][0].poc;
1141 for(i=0; i<h->ref_count[0]; i++){
1142 int poc0 = h->ref_list[0][i].poc;
1143 int td = clip(poc1 - poc0, -128, 127);
1144 if(td == 0 /* FIXME || pic0 is a long-term ref */){
1145 h->dist_scale_factor[i] = 256;
1147 int tb = clip(poc - poc0, -128, 127);
1148 int tx = (16384 + (ABS(td) >> 1)) / td;
1149 h->dist_scale_factor[i] = clip((tb*tx + 32) >> 6, -1024, 1023);
1153 static inline void direct_ref_list_init(H264Context * const h){
1154 MpegEncContext * const s = &h->s;
1155 Picture * const ref1 = &h->ref_list[1][0];
1156 Picture * const cur = s->current_picture_ptr;
1158 if(cur->pict_type == I_TYPE)
1159 cur->ref_count[0] = 0;
1160 if(cur->pict_type != B_TYPE)
1161 cur->ref_count[1] = 0;
1162 for(list=0; list<2; list++){
1163 cur->ref_count[list] = h->ref_count[list];
1164 for(j=0; j<h->ref_count[list]; j++)
1165 cur->ref_poc[list][j] = h->ref_list[list][j].poc;
1167 if(cur->pict_type != B_TYPE || h->direct_spatial_mv_pred)
1169 for(list=0; list<2; list++){
1170 for(i=0; i<ref1->ref_count[list]; i++){
1171 const int poc = ref1->ref_poc[list][i];
1172 h->map_col_to_list0[list][i] = PART_NOT_AVAILABLE;
1173 for(j=0; j<h->ref_count[list]; j++)
1174 if(h->ref_list[list][j].poc == poc){
1175 h->map_col_to_list0[list][i] = j;
1182 static inline void pred_direct_motion(H264Context * const h, int *mb_type){
1183 MpegEncContext * const s = &h->s;
1184 const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
1185 const int b8_xy = 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1186 const int b4_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1187 const int mb_type_col = h->ref_list[1][0].mb_type[mb_xy];
1188 const int16_t (*l1mv0)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[0][b4_xy];
1189 const int8_t *l1ref0 = &h->ref_list[1][0].ref_index[0][b8_xy];
1190 const int8_t *l1ref1 = &h->ref_list[1][0].ref_index[1][b8_xy];
1191 const int is_b8x8 = IS_8X8(*mb_type);
1195 if(IS_8X8(mb_type_col) && !h->sps.direct_8x8_inference_flag){
1196 /* FIXME save sub mb types from previous frames (or derive from MVs)
1197 * so we know exactly what block size to use */
1198 sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */
1199 *mb_type = MB_TYPE_8x8|MB_TYPE_L0L1;
1200 }else if(!is_b8x8 && (IS_16X16(mb_type_col) || IS_INTRA(mb_type_col))){
1201 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1202 *mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */
1204 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1205 *mb_type = MB_TYPE_8x8|MB_TYPE_L0L1;
1208 *mb_type |= MB_TYPE_DIRECT2;
1210 tprintf("mb_type = %08x, sub_mb_type = %08x, is_b8x8 = %d, mb_type_col = %08x\n", *mb_type, sub_mb_type, is_b8x8, mb_type_col);
1212 if(h->direct_spatial_mv_pred){
1217 /* ref = min(neighbors) */
1218 for(list=0; list<2; list++){
1219 int refa = h->ref_cache[list][scan8[0] - 1];
1220 int refb = h->ref_cache[list][scan8[0] - 8];
1221 int refc = h->ref_cache[list][scan8[0] - 8 + 4];
1223 refc = h->ref_cache[list][scan8[0] - 8 - 1];
1225 if(ref[list] < 0 || (refb < ref[list] && refb >= 0))
1227 if(ref[list] < 0 || (refc < ref[list] && refc >= 0))
1233 if(ref[0] < 0 && ref[1] < 0){
1234 ref[0] = ref[1] = 0;
1235 mv[0][0] = mv[0][1] =
1236 mv[1][0] = mv[1][1] = 0;
1238 for(list=0; list<2; list++){
1240 pred_motion(h, 0, 4, list, ref[list], &mv[list][0], &mv[list][1]);
1242 mv[list][0] = mv[list][1] = 0;
1247 *mb_type &= ~MB_TYPE_P0L1;
1248 sub_mb_type &= ~MB_TYPE_P0L1;
1249 }else if(ref[0] < 0){
1250 *mb_type &= ~MB_TYPE_P0L0;
1251 sub_mb_type &= ~MB_TYPE_P0L0;
1254 if(IS_16X16(*mb_type)){
1255 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref[0], 1);
1256 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, ref[1], 1);
1257 if(!IS_INTRA(mb_type_col) && l1ref0[0] == 0 &&
1258 ABS(l1mv0[0][0]) <= 1 && ABS(l1mv0[0][1]) <= 1){
1260 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1262 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
1264 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1266 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
1268 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1269 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1272 for(i8=0; i8<4; i8++){
1273 const int x8 = i8&1;
1274 const int y8 = i8>>1;
1276 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1278 h->sub_mb_type[i8] = sub_mb_type;
1280 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1281 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1282 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref[0], 1);
1283 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, ref[1], 1);
1286 if(!IS_INTRA(mb_type_col) && l1ref0[x8 + y8*h->b8_stride] == 0){
1287 for(i4=0; i4<4; i4++){
1288 const int16_t *mv_col = l1mv0[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1289 if(ABS(mv_col[0]) <= 1 && ABS(mv_col[1]) <= 1){
1291 *(uint32_t*)h->mv_cache[0][scan8[i8*4+i4]] = 0;
1293 *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = 0;
1299 }else{ /* direct temporal mv pred */
1300 if(IS_16X16(*mb_type)){
1301 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1);
1302 if(IS_INTRA(mb_type_col)){
1303 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
1304 fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
1305 fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
1307 const int ref0 = l1ref0[0] >= 0 ? h->map_col_to_list0[0][l1ref0[0]]
1308 : h->map_col_to_list0[1][l1ref1[0]];
1309 const int dist_scale_factor = h->dist_scale_factor[ref0];
1310 const int16_t *mv_col = l1mv0[0];
1312 mv_l0[0] = (dist_scale_factor * mv_col[0] + 128) >> 8;
1313 mv_l0[1] = (dist_scale_factor * mv_col[1] + 128) >> 8;
1314 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref0, 1);
1315 fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv_l0[0],mv_l0[1]), 4);
1316 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);
1319 for(i8=0; i8<4; i8++){
1320 const int x8 = i8&1;
1321 const int y8 = i8>>1;
1322 int ref0, dist_scale_factor;
1324 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1326 h->sub_mb_type[i8] = sub_mb_type;
1327 if(IS_INTRA(mb_type_col)){
1328 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1329 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1330 fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1331 fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1335 ref0 = l1ref0[x8 + y8*h->b8_stride];
1337 ref0 = h->map_col_to_list0[0][ref0];
1339 ref0 = h->map_col_to_list0[1][l1ref1[x8 + y8*h->b8_stride]];
1340 dist_scale_factor = h->dist_scale_factor[ref0];
1342 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1343 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1344 for(i4=0; i4<4; i4++){
1345 const int16_t *mv_col = l1mv0[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1346 int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]];
1347 mv_l0[0] = (dist_scale_factor * mv_col[0] + 128) >> 8;
1348 mv_l0[1] = (dist_scale_factor * mv_col[1] + 128) >> 8;
1349 *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] =
1350 pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
1357 static inline void write_back_motion(H264Context *h, int mb_type){
1358 MpegEncContext * const s = &h->s;
1359 const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1360 const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1363 for(list=0; list<2; list++){
1365 if(!USES_LIST(mb_type, list)){
1366 if(1){ //FIXME skip or never read if mb_type doesn't use it
1368 *(uint64_t*)s->current_picture.motion_val[list][b_xy + 0 + y*h->b_stride]=
1369 *(uint64_t*)s->current_picture.motion_val[list][b_xy + 2 + y*h->b_stride]= 0;
1371 if( h->pps.cabac ) {
1372 /* FIXME needed ? */
1374 *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]=
1375 *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= 0;
1379 *(uint16_t*)&s->current_picture.ref_index[list][b8_xy + y*h->b8_stride]= (LIST_NOT_USED&0xFF)*0x0101;
1386 *(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];
1387 *(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];
1389 if( h->pps.cabac ) {
1391 *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+0 + 8*y];
1392 *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+2 + 8*y];
1396 s->current_picture.ref_index[list][b8_xy + 0 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+0 + 16*y];
1397 s->current_picture.ref_index[list][b8_xy + 1 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+2 + 16*y];
1401 if(h->slice_type == B_TYPE && h->pps.cabac){
1402 if(IS_8X8(mb_type)){
1403 h->direct_table[b8_xy+1+0*h->b8_stride] = IS_DIRECT(h->sub_mb_type[1]) ? 1 : 0;
1404 h->direct_table[b8_xy+0+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[2]) ? 1 : 0;
1405 h->direct_table[b8_xy+1+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[3]) ? 1 : 0;
1411 * Decodes a network abstraction layer unit.
1412 * @param consumed is the number of bytes used as input
1413 * @param length is the length of the array
1414 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp tailing?
1415 * @returns decoded bytes, might be src+1 if no escapes
1417 static uint8_t *decode_nal(H264Context *h, uint8_t *src, int *dst_length, int *consumed, int length){
1421 // src[0]&0x80; //forbidden bit
1422 h->nal_ref_idc= src[0]>>5;
1423 h->nal_unit_type= src[0]&0x1F;
1427 for(i=0; i<length; i++)
1428 printf("%2X ", src[i]);
1430 for(i=0; i+1<length; i+=2){
1431 if(src[i]) continue;
1432 if(i>0 && src[i-1]==0) i--;
1433 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1435 /* startcode, so we must be past the end */
1442 if(i>=length-1){ //no escaped 0
1443 *dst_length= length;
1444 *consumed= length+1; //+1 for the header
1448 h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length);
1449 dst= h->rbsp_buffer;
1451 //printf("decoding esc\n");
1454 //remove escapes (very rare 1:2^22)
1455 if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1456 if(src[si+2]==3){ //escape
1461 }else //next start code
1465 dst[di++]= src[si++];
1469 *consumed= si + 1;//+1 for the header
1470 //FIXME store exact number of bits in the getbitcontext (its needed for decoding)
1476 * @param src the data which should be escaped
1477 * @param dst the target buffer, dst+1 == src is allowed as a special case
1478 * @param length the length of the src data
1479 * @param dst_length the length of the dst array
1480 * @returns length of escaped data in bytes or -1 if an error occured
1482 static int encode_nal(H264Context *h, uint8_t *dst, uint8_t *src, int length, int dst_length){
1483 int i, escape_count, si, di;
1487 assert(dst_length>0);
1489 dst[0]= (h->nal_ref_idc<<5) + h->nal_unit_type;
1491 if(length==0) return 1;
1494 for(i=0; i<length; i+=2){
1495 if(src[i]) continue;
1496 if(i>0 && src[i-1]==0)
1498 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1504 if(escape_count==0){
1506 memcpy(dst+1, src, length);
1510 if(length + escape_count + 1> dst_length)
1513 //this should be damn rare (hopefully)
1515 h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length + escape_count);
1516 temp= h->rbsp_buffer;
1517 //printf("encoding esc\n");
1522 if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1523 temp[di++]= 0; si++;
1524 temp[di++]= 0; si++;
1526 temp[di++]= src[si++];
1529 temp[di++]= src[si++];
1531 memcpy(dst+1, temp, length+escape_count);
1533 assert(di == length+escape_count);
1539 * write 1,10,100,1000,... for alignment, yes its exactly inverse to mpeg4
1541 static void encode_rbsp_trailing(PutBitContext *pb){
1544 length= (-put_bits_count(pb))&7;
1545 if(length) put_bits(pb, length, 0);
1550 * identifies the exact end of the bitstream
1551 * @return the length of the trailing, or 0 if damaged
1553 static int decode_rbsp_trailing(uint8_t *src){
1557 tprintf("rbsp trailing %X\n", v);
1567 * idct tranforms the 16 dc values and dequantize them.
1568 * @param qp quantization parameter
1570 static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp){
1571 const int qmul= dequant_coeff[qp][0];
1574 int temp[16]; //FIXME check if this is a good idea
1575 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
1576 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1578 //memset(block, 64, 2*256);
1581 const int offset= y_offset[i];
1582 const int z0= block[offset+stride*0] + block[offset+stride*4];
1583 const int z1= block[offset+stride*0] - block[offset+stride*4];
1584 const int z2= block[offset+stride*1] - block[offset+stride*5];
1585 const int z3= block[offset+stride*1] + block[offset+stride*5];
1594 const int offset= x_offset[i];
1595 const int z0= temp[4*0+i] + temp[4*2+i];
1596 const int z1= temp[4*0+i] - temp[4*2+i];
1597 const int z2= temp[4*1+i] - temp[4*3+i];
1598 const int z3= temp[4*1+i] + temp[4*3+i];
1600 block[stride*0 +offset]= ((z0 + z3)*qmul + 2)>>2; //FIXME think about merging this into decode_resdual
1601 block[stride*2 +offset]= ((z1 + z2)*qmul + 2)>>2;
1602 block[stride*8 +offset]= ((z1 - z2)*qmul + 2)>>2;
1603 block[stride*10+offset]= ((z0 - z3)*qmul + 2)>>2;
1609 * dct tranforms the 16 dc values.
1610 * @param qp quantization parameter ??? FIXME
1612 static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
1613 // const int qmul= dequant_coeff[qp][0];
1615 int temp[16]; //FIXME check if this is a good idea
1616 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
1617 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1620 const int offset= y_offset[i];
1621 const int z0= block[offset+stride*0] + block[offset+stride*4];
1622 const int z1= block[offset+stride*0] - block[offset+stride*4];
1623 const int z2= block[offset+stride*1] - block[offset+stride*5];
1624 const int z3= block[offset+stride*1] + block[offset+stride*5];
1633 const int offset= x_offset[i];
1634 const int z0= temp[4*0+i] + temp[4*2+i];
1635 const int z1= temp[4*0+i] - temp[4*2+i];
1636 const int z2= temp[4*1+i] - temp[4*3+i];
1637 const int z3= temp[4*1+i] + temp[4*3+i];
1639 block[stride*0 +offset]= (z0 + z3)>>1;
1640 block[stride*2 +offset]= (z1 + z2)>>1;
1641 block[stride*8 +offset]= (z1 - z2)>>1;
1642 block[stride*10+offset]= (z0 - z3)>>1;
1650 static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp){
1651 const int qmul= dequant_coeff[qp][0];
1652 const int stride= 16*2;
1653 const int xStride= 16;
1656 a= block[stride*0 + xStride*0];
1657 b= block[stride*0 + xStride*1];
1658 c= block[stride*1 + xStride*0];
1659 d= block[stride*1 + xStride*1];
1666 block[stride*0 + xStride*0]= ((a+c)*qmul + 0)>>1;
1667 block[stride*0 + xStride*1]= ((e+b)*qmul + 0)>>1;
1668 block[stride*1 + xStride*0]= ((a-c)*qmul + 0)>>1;
1669 block[stride*1 + xStride*1]= ((e-b)*qmul + 0)>>1;
1673 static void chroma_dc_dct_c(DCTELEM *block){
1674 const int stride= 16*2;
1675 const int xStride= 16;
1678 a= block[stride*0 + xStride*0];
1679 b= block[stride*0 + xStride*1];
1680 c= block[stride*1 + xStride*0];
1681 d= block[stride*1 + xStride*1];
1688 block[stride*0 + xStride*0]= (a+c);
1689 block[stride*0 + xStride*1]= (e+b);
1690 block[stride*1 + xStride*0]= (a-c);
1691 block[stride*1 + xStride*1]= (e-b);
1696 * gets the chroma qp.
1698 static inline int get_chroma_qp(int chroma_qp_index_offset, int qscale){
1700 return chroma_qp[clip(qscale + chroma_qp_index_offset, 0, 51)];
1705 static void h264_diff_dct_c(DCTELEM *block, uint8_t *src1, uint8_t *src2, int stride){
1707 //FIXME try int temp instead of block
1710 const int d0= src1[0 + i*stride] - src2[0 + i*stride];
1711 const int d1= src1[1 + i*stride] - src2[1 + i*stride];
1712 const int d2= src1[2 + i*stride] - src2[2 + i*stride];
1713 const int d3= src1[3 + i*stride] - src2[3 + i*stride];
1714 const int z0= d0 + d3;
1715 const int z3= d0 - d3;
1716 const int z1= d1 + d2;
1717 const int z2= d1 - d2;
1719 block[0 + 4*i]= z0 + z1;
1720 block[1 + 4*i]= 2*z3 + z2;
1721 block[2 + 4*i]= z0 - z1;
1722 block[3 + 4*i]= z3 - 2*z2;
1726 const int z0= block[0*4 + i] + block[3*4 + i];
1727 const int z3= block[0*4 + i] - block[3*4 + i];
1728 const int z1= block[1*4 + i] + block[2*4 + i];
1729 const int z2= block[1*4 + i] - block[2*4 + i];
1731 block[0*4 + i]= z0 + z1;
1732 block[1*4 + i]= 2*z3 + z2;
1733 block[2*4 + i]= z0 - z1;
1734 block[3*4 + i]= z3 - 2*z2;
1739 //FIXME need to check that this doesnt overflow signed 32 bit for low qp, i am not sure, it's very close
1740 //FIXME check that gcc inlines this (and optimizes intra & seperate_dc stuff away)
1741 static inline int quantize_c(DCTELEM *block, uint8_t *scantable, int qscale, int intra, int seperate_dc){
1743 const int * const quant_table= quant_coeff[qscale];
1744 const int bias= intra ? (1<<QUANT_SHIFT)/3 : (1<<QUANT_SHIFT)/6;
1745 const unsigned int threshold1= (1<<QUANT_SHIFT) - bias - 1;
1746 const unsigned int threshold2= (threshold1<<1);
1752 const int dc_bias= intra ? (1<<(QUANT_SHIFT-2))/3 : (1<<(QUANT_SHIFT-2))/6;
1753 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT-2)) - dc_bias - 1;
1754 const unsigned int dc_threshold2= (dc_threshold1<<1);
1756 int level= block[0]*quant_coeff[qscale+18][0];
1757 if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1759 level= (dc_bias + level)>>(QUANT_SHIFT-2);
1762 level= (dc_bias - level)>>(QUANT_SHIFT-2);
1765 // last_non_zero = i;
1770 const int dc_bias= intra ? (1<<(QUANT_SHIFT+1))/3 : (1<<(QUANT_SHIFT+1))/6;
1771 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT+1)) - dc_bias - 1;
1772 const unsigned int dc_threshold2= (dc_threshold1<<1);
1774 int level= block[0]*quant_table[0];
1775 if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1777 level= (dc_bias + level)>>(QUANT_SHIFT+1);
1780 level= (dc_bias - level)>>(QUANT_SHIFT+1);
1783 // last_non_zero = i;
1796 const int j= scantable[i];
1797 int level= block[j]*quant_table[j];
1799 // if( bias+level >= (1<<(QMAT_SHIFT - 3))
1800 // || bias-level >= (1<<(QMAT_SHIFT - 3))){
1801 if(((unsigned)(level+threshold1))>threshold2){
1803 level= (bias + level)>>QUANT_SHIFT;
1806 level= (bias - level)>>QUANT_SHIFT;
1815 return last_non_zero;
1818 static void pred4x4_vertical_c(uint8_t *src, uint8_t *topright, int stride){
1819 const uint32_t a= ((uint32_t*)(src-stride))[0];
1820 ((uint32_t*)(src+0*stride))[0]= a;
1821 ((uint32_t*)(src+1*stride))[0]= a;
1822 ((uint32_t*)(src+2*stride))[0]= a;
1823 ((uint32_t*)(src+3*stride))[0]= a;
1826 static void pred4x4_horizontal_c(uint8_t *src, uint8_t *topright, int stride){
1827 ((uint32_t*)(src+0*stride))[0]= src[-1+0*stride]*0x01010101;
1828 ((uint32_t*)(src+1*stride))[0]= src[-1+1*stride]*0x01010101;
1829 ((uint32_t*)(src+2*stride))[0]= src[-1+2*stride]*0x01010101;
1830 ((uint32_t*)(src+3*stride))[0]= src[-1+3*stride]*0x01010101;
1833 static void pred4x4_dc_c(uint8_t *src, uint8_t *topright, int stride){
1834 const int dc= ( src[-stride] + src[1-stride] + src[2-stride] + src[3-stride]
1835 + src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 4) >>3;
1837 ((uint32_t*)(src+0*stride))[0]=
1838 ((uint32_t*)(src+1*stride))[0]=
1839 ((uint32_t*)(src+2*stride))[0]=
1840 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1843 static void pred4x4_left_dc_c(uint8_t *src, uint8_t *topright, int stride){
1844 const int dc= ( src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 2) >>2;
1846 ((uint32_t*)(src+0*stride))[0]=
1847 ((uint32_t*)(src+1*stride))[0]=
1848 ((uint32_t*)(src+2*stride))[0]=
1849 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1852 static void pred4x4_top_dc_c(uint8_t *src, uint8_t *topright, int stride){
1853 const int dc= ( src[-stride] + src[1-stride] + src[2-stride] + src[3-stride] + 2) >>2;
1855 ((uint32_t*)(src+0*stride))[0]=
1856 ((uint32_t*)(src+1*stride))[0]=
1857 ((uint32_t*)(src+2*stride))[0]=
1858 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1861 static void pred4x4_128_dc_c(uint8_t *src, uint8_t *topright, int stride){
1862 ((uint32_t*)(src+0*stride))[0]=
1863 ((uint32_t*)(src+1*stride))[0]=
1864 ((uint32_t*)(src+2*stride))[0]=
1865 ((uint32_t*)(src+3*stride))[0]= 128U*0x01010101U;
1869 #define LOAD_TOP_RIGHT_EDGE\
1870 const int t4= topright[0];\
1871 const int t5= topright[1];\
1872 const int t6= topright[2];\
1873 const int t7= topright[3];\
1875 #define LOAD_LEFT_EDGE\
1876 const int l0= src[-1+0*stride];\
1877 const int l1= src[-1+1*stride];\
1878 const int l2= src[-1+2*stride];\
1879 const int l3= src[-1+3*stride];\
1881 #define LOAD_TOP_EDGE\
1882 const int t0= src[ 0-1*stride];\
1883 const int t1= src[ 1-1*stride];\
1884 const int t2= src[ 2-1*stride];\
1885 const int t3= src[ 3-1*stride];\
1887 static void pred4x4_down_right_c(uint8_t *src, uint8_t *topright, int stride){
1888 const int lt= src[-1-1*stride];
1892 src[0+3*stride]=(l3 + 2*l2 + l1 + 2)>>2;
1894 src[1+3*stride]=(l2 + 2*l1 + l0 + 2)>>2;
1897 src[2+3*stride]=(l1 + 2*l0 + lt + 2)>>2;
1901 src[3+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1904 src[3+2*stride]=(lt + 2*t0 + t1 + 2)>>2;
1906 src[3+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1907 src[3+0*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1910 static void pred4x4_down_left_c(uint8_t *src, uint8_t *topright, int stride){
1915 src[0+0*stride]=(t0 + t2 + 2*t1 + 2)>>2;
1917 src[0+1*stride]=(t1 + t3 + 2*t2 + 2)>>2;
1920 src[0+2*stride]=(t2 + t4 + 2*t3 + 2)>>2;
1924 src[0+3*stride]=(t3 + t5 + 2*t4 + 2)>>2;
1927 src[1+3*stride]=(t4 + t6 + 2*t5 + 2)>>2;
1929 src[2+3*stride]=(t5 + t7 + 2*t6 + 2)>>2;
1930 src[3+3*stride]=(t6 + 3*t7 + 2)>>2;
1933 static void pred4x4_vertical_right_c(uint8_t *src, uint8_t *topright, int stride){
1934 const int lt= src[-1-1*stride];
1937 const __attribute__((unused)) int unu= l3;
1940 src[1+2*stride]=(lt + t0 + 1)>>1;
1942 src[2+2*stride]=(t0 + t1 + 1)>>1;
1944 src[3+2*stride]=(t1 + t2 + 1)>>1;
1945 src[3+0*stride]=(t2 + t3 + 1)>>1;
1947 src[1+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1949 src[2+3*stride]=(lt + 2*t0 + t1 + 2)>>2;
1951 src[3+3*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1952 src[3+1*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1953 src[0+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1954 src[0+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1957 static void pred4x4_vertical_left_c(uint8_t *src, uint8_t *topright, int stride){
1960 const __attribute__((unused)) int unu= t7;
1962 src[0+0*stride]=(t0 + t1 + 1)>>1;
1964 src[0+2*stride]=(t1 + t2 + 1)>>1;
1966 src[1+2*stride]=(t2 + t3 + 1)>>1;
1968 src[2+2*stride]=(t3 + t4+ 1)>>1;
1969 src[3+2*stride]=(t4 + t5+ 1)>>1;
1970 src[0+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1972 src[0+3*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1974 src[1+3*stride]=(t2 + 2*t3 + t4 + 2)>>2;
1976 src[2+3*stride]=(t3 + 2*t4 + t5 + 2)>>2;
1977 src[3+3*stride]=(t4 + 2*t5 + t6 + 2)>>2;
1980 static void pred4x4_horizontal_up_c(uint8_t *src, uint8_t *topright, int stride){
1983 src[0+0*stride]=(l0 + l1 + 1)>>1;
1984 src[1+0*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1986 src[0+1*stride]=(l1 + l2 + 1)>>1;
1988 src[1+1*stride]=(l1 + 2*l2 + l3 + 2)>>2;
1990 src[0+2*stride]=(l2 + l3 + 1)>>1;
1992 src[1+2*stride]=(l2 + 2*l3 + l3 + 2)>>2;
2001 static void pred4x4_horizontal_down_c(uint8_t *src, uint8_t *topright, int stride){
2002 const int lt= src[-1-1*stride];
2005 const __attribute__((unused)) int unu= t3;
2008 src[2+1*stride]=(lt + l0 + 1)>>1;
2010 src[3+1*stride]=(l0 + 2*lt + t0 + 2)>>2;
2011 src[2+0*stride]=(lt + 2*t0 + t1 + 2)>>2;
2012 src[3+0*stride]=(t0 + 2*t1 + t2 + 2)>>2;
2014 src[2+2*stride]=(l0 + l1 + 1)>>1;
2016 src[3+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
2018 src[2+3*stride]=(l1 + l2+ 1)>>1;
2020 src[3+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
2021 src[0+3*stride]=(l2 + l3 + 1)>>1;
2022 src[1+3*stride]=(l1 + 2*l2 + l3 + 2)>>2;
2025 static void pred16x16_vertical_c(uint8_t *src, int stride){
2027 const uint32_t a= ((uint32_t*)(src-stride))[0];
2028 const uint32_t b= ((uint32_t*)(src-stride))[1];
2029 const uint32_t c= ((uint32_t*)(src-stride))[2];
2030 const uint32_t d= ((uint32_t*)(src-stride))[3];
2032 for(i=0; i<16; i++){
2033 ((uint32_t*)(src+i*stride))[0]= a;
2034 ((uint32_t*)(src+i*stride))[1]= b;
2035 ((uint32_t*)(src+i*stride))[2]= c;
2036 ((uint32_t*)(src+i*stride))[3]= d;
2040 static void pred16x16_horizontal_c(uint8_t *src, int stride){
2043 for(i=0; i<16; i++){
2044 ((uint32_t*)(src+i*stride))[0]=
2045 ((uint32_t*)(src+i*stride))[1]=
2046 ((uint32_t*)(src+i*stride))[2]=
2047 ((uint32_t*)(src+i*stride))[3]= src[-1+i*stride]*0x01010101;
2051 static void pred16x16_dc_c(uint8_t *src, int stride){
2055 dc+= src[-1+i*stride];
2062 dc= 0x01010101*((dc + 16)>>5);
2064 for(i=0; i<16; i++){
2065 ((uint32_t*)(src+i*stride))[0]=
2066 ((uint32_t*)(src+i*stride))[1]=
2067 ((uint32_t*)(src+i*stride))[2]=
2068 ((uint32_t*)(src+i*stride))[3]= dc;
2072 static void pred16x16_left_dc_c(uint8_t *src, int stride){
2076 dc+= src[-1+i*stride];
2079 dc= 0x01010101*((dc + 8)>>4);
2081 for(i=0; i<16; i++){
2082 ((uint32_t*)(src+i*stride))[0]=
2083 ((uint32_t*)(src+i*stride))[1]=
2084 ((uint32_t*)(src+i*stride))[2]=
2085 ((uint32_t*)(src+i*stride))[3]= dc;
2089 static void pred16x16_top_dc_c(uint8_t *src, int stride){
2095 dc= 0x01010101*((dc + 8)>>4);
2097 for(i=0; i<16; i++){
2098 ((uint32_t*)(src+i*stride))[0]=
2099 ((uint32_t*)(src+i*stride))[1]=
2100 ((uint32_t*)(src+i*stride))[2]=
2101 ((uint32_t*)(src+i*stride))[3]= dc;
2105 static void pred16x16_128_dc_c(uint8_t *src, int stride){
2108 for(i=0; i<16; i++){
2109 ((uint32_t*)(src+i*stride))[0]=
2110 ((uint32_t*)(src+i*stride))[1]=
2111 ((uint32_t*)(src+i*stride))[2]=
2112 ((uint32_t*)(src+i*stride))[3]= 0x01010101U*128U;
2116 static inline void pred16x16_plane_compat_c(uint8_t *src, int stride, const int svq3){
2119 uint8_t *cm = cropTbl + MAX_NEG_CROP;
2120 const uint8_t * const src0 = src+7-stride;
2121 const uint8_t *src1 = src+8*stride-1;
2122 const uint8_t *src2 = src1-2*stride; // == src+6*stride-1;
2123 int H = src0[1] - src0[-1];
2124 int V = src1[0] - src2[ 0];
2125 for(k=2; k<=8; ++k) {
2126 src1 += stride; src2 -= stride;
2127 H += k*(src0[k] - src0[-k]);
2128 V += k*(src1[0] - src2[ 0]);
2131 H = ( 5*(H/4) ) / 16;
2132 V = ( 5*(V/4) ) / 16;
2134 /* required for 100% accuracy */
2135 i = H; H = V; V = i;
2137 H = ( 5*H+32 ) >> 6;
2138 V = ( 5*V+32 ) >> 6;
2141 a = 16*(src1[0] + src2[16] + 1) - 7*(V+H);
2142 for(j=16; j>0; --j) {
2145 for(i=-16; i<0; i+=4) {
2146 src[16+i] = cm[ (b ) >> 5 ];
2147 src[17+i] = cm[ (b+ H) >> 5 ];
2148 src[18+i] = cm[ (b+2*H) >> 5 ];
2149 src[19+i] = cm[ (b+3*H) >> 5 ];
2156 static void pred16x16_plane_c(uint8_t *src, int stride){
2157 pred16x16_plane_compat_c(src, stride, 0);
2160 static void pred8x8_vertical_c(uint8_t *src, int stride){
2162 const uint32_t a= ((uint32_t*)(src-stride))[0];
2163 const uint32_t b= ((uint32_t*)(src-stride))[1];
2166 ((uint32_t*)(src+i*stride))[0]= a;
2167 ((uint32_t*)(src+i*stride))[1]= b;
2171 static void pred8x8_horizontal_c(uint8_t *src, int stride){
2175 ((uint32_t*)(src+i*stride))[0]=
2176 ((uint32_t*)(src+i*stride))[1]= src[-1+i*stride]*0x01010101;
2180 static void pred8x8_128_dc_c(uint8_t *src, int stride){
2184 ((uint32_t*)(src+i*stride))[0]=
2185 ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
2188 ((uint32_t*)(src+i*stride))[0]=
2189 ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
2193 static void pred8x8_left_dc_c(uint8_t *src, int stride){
2199 dc0+= src[-1+i*stride];
2200 dc2+= src[-1+(i+4)*stride];
2202 dc0= 0x01010101*((dc0 + 2)>>2);
2203 dc2= 0x01010101*((dc2 + 2)>>2);
2206 ((uint32_t*)(src+i*stride))[0]=
2207 ((uint32_t*)(src+i*stride))[1]= dc0;
2210 ((uint32_t*)(src+i*stride))[0]=
2211 ((uint32_t*)(src+i*stride))[1]= dc2;
2215 static void pred8x8_top_dc_c(uint8_t *src, int stride){
2221 dc0+= src[i-stride];
2222 dc1+= src[4+i-stride];
2224 dc0= 0x01010101*((dc0 + 2)>>2);
2225 dc1= 0x01010101*((dc1 + 2)>>2);
2228 ((uint32_t*)(src+i*stride))[0]= dc0;
2229 ((uint32_t*)(src+i*stride))[1]= dc1;
2232 ((uint32_t*)(src+i*stride))[0]= dc0;
2233 ((uint32_t*)(src+i*stride))[1]= dc1;
2238 static void pred8x8_dc_c(uint8_t *src, int stride){
2240 int dc0, dc1, dc2, dc3;
2244 dc0+= src[-1+i*stride] + src[i-stride];
2245 dc1+= src[4+i-stride];
2246 dc2+= src[-1+(i+4)*stride];
2248 dc3= 0x01010101*((dc1 + dc2 + 4)>>3);
2249 dc0= 0x01010101*((dc0 + 4)>>3);
2250 dc1= 0x01010101*((dc1 + 2)>>2);
2251 dc2= 0x01010101*((dc2 + 2)>>2);
2254 ((uint32_t*)(src+i*stride))[0]= dc0;
2255 ((uint32_t*)(src+i*stride))[1]= dc1;
2258 ((uint32_t*)(src+i*stride))[0]= dc2;
2259 ((uint32_t*)(src+i*stride))[1]= dc3;
2263 static void pred8x8_plane_c(uint8_t *src, int stride){
2266 uint8_t *cm = cropTbl + MAX_NEG_CROP;
2267 const uint8_t * const src0 = src+3-stride;
2268 const uint8_t *src1 = src+4*stride-1;
2269 const uint8_t *src2 = src1-2*stride; // == src+2*stride-1;
2270 int H = src0[1] - src0[-1];
2271 int V = src1[0] - src2[ 0];
2272 for(k=2; k<=4; ++k) {
2273 src1 += stride; src2 -= stride;
2274 H += k*(src0[k] - src0[-k]);
2275 V += k*(src1[0] - src2[ 0]);
2277 H = ( 17*H+16 ) >> 5;
2278 V = ( 17*V+16 ) >> 5;
2280 a = 16*(src1[0] + src2[8]+1) - 3*(V+H);
2281 for(j=8; j>0; --j) {
2284 src[0] = cm[ (b ) >> 5 ];
2285 src[1] = cm[ (b+ H) >> 5 ];
2286 src[2] = cm[ (b+2*H) >> 5 ];
2287 src[3] = cm[ (b+3*H) >> 5 ];
2288 src[4] = cm[ (b+4*H) >> 5 ];
2289 src[5] = cm[ (b+5*H) >> 5 ];
2290 src[6] = cm[ (b+6*H) >> 5 ];
2291 src[7] = cm[ (b+7*H) >> 5 ];
2296 static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
2297 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2298 int src_x_offset, int src_y_offset,
2299 qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
2300 MpegEncContext * const s = &h->s;
2301 const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
2302 const int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
2303 const int luma_xy= (mx&3) + ((my&3)<<2);
2304 uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*s->linesize;
2305 uint8_t * src_cb= pic->data[1] + (mx>>3) + (my>>3)*s->uvlinesize;
2306 uint8_t * src_cr= pic->data[2] + (mx>>3) + (my>>3)*s->uvlinesize;
2307 int extra_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16; //FIXME increase edge?, IMHO not worth it
2308 int extra_height= extra_width;
2310 const int full_mx= mx>>2;
2311 const int full_my= my>>2;
2313 assert(pic->data[0]);
2315 if(mx&7) extra_width -= 3;
2316 if(my&7) extra_height -= 3;
2318 if( full_mx < 0-extra_width
2319 || full_my < 0-extra_height
2320 || full_mx + 16/*FIXME*/ > s->width + extra_width
2321 || full_my + 16/*FIXME*/ > s->height + extra_height){
2322 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);
2323 src_y= s->edge_emu_buffer + 2 + 2*s->linesize;
2327 qpix_op[luma_xy](dest_y, src_y, s->linesize); //FIXME try variable height perhaps?
2329 qpix_op[luma_xy](dest_y + delta, src_y + delta, s->linesize);
2332 if(s->flags&CODEC_FLAG_GRAY) return;
2335 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);
2336 src_cb= s->edge_emu_buffer;
2338 chroma_op(dest_cb, src_cb, s->uvlinesize, chroma_height, mx&7, my&7);
2341 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);
2342 src_cr= s->edge_emu_buffer;
2344 chroma_op(dest_cr, src_cr, s->uvlinesize, chroma_height, mx&7, my&7);
2347 static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
2348 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2349 int x_offset, int y_offset,
2350 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2351 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
2352 int list0, int list1){
2353 MpegEncContext * const s = &h->s;
2354 qpel_mc_func *qpix_op= qpix_put;
2355 h264_chroma_mc_func chroma_op= chroma_put;
2357 dest_y += 2*x_offset + 2*y_offset*s-> linesize;
2358 dest_cb += x_offset + y_offset*s->uvlinesize;
2359 dest_cr += x_offset + y_offset*s->uvlinesize;
2360 x_offset += 8*s->mb_x;
2361 y_offset += 8*s->mb_y;
2364 Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
2365 mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
2366 dest_y, dest_cb, dest_cr, x_offset, y_offset,
2367 qpix_op, chroma_op);
2370 chroma_op= chroma_avg;
2374 Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
2375 mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
2376 dest_y, dest_cb, dest_cr, x_offset, y_offset,
2377 qpix_op, chroma_op);
2381 static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
2382 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2383 int x_offset, int y_offset,
2384 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2385 h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
2386 h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
2387 int list0, int list1){
2388 MpegEncContext * const s = &h->s;
2390 dest_y += 2*x_offset + 2*y_offset*s-> linesize;
2391 dest_cb += x_offset + y_offset*s->uvlinesize;
2392 dest_cr += x_offset + y_offset*s->uvlinesize;
2393 x_offset += 8*s->mb_x;
2394 y_offset += 8*s->mb_y;
2397 /* don't optimize for luma-only case, since B-frames usually
2398 * use implicit weights => chroma too. */
2399 uint8_t *tmp_cb = s->obmc_scratchpad;
2400 uint8_t *tmp_cr = tmp_cb + 8*s->uvlinesize;
2401 uint8_t *tmp_y = tmp_cr + 8*s->uvlinesize;
2402 int refn0 = h->ref_cache[0][ scan8[n] ];
2403 int refn1 = h->ref_cache[1][ scan8[n] ];
2405 mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
2406 dest_y, dest_cb, dest_cr,
2407 x_offset, y_offset, qpix_put, chroma_put);
2408 mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
2409 tmp_y, tmp_cb, tmp_cr,
2410 x_offset, y_offset, qpix_put, chroma_put);
2412 if(h->use_weight == 2){
2413 int weight0 = h->implicit_weight[refn0][refn1];
2414 int weight1 = 64 - weight0;
2415 luma_weight_avg( dest_y, tmp_y, s-> linesize, 5, weight0, weight1, 0, 0);
2416 chroma_weight_avg(dest_cb, tmp_cb, s->uvlinesize, 5, weight0, weight1, 0, 0);
2417 chroma_weight_avg(dest_cr, tmp_cr, s->uvlinesize, 5, weight0, weight1, 0, 0);
2419 luma_weight_avg(dest_y, tmp_y, s->linesize, h->luma_log2_weight_denom,
2420 h->luma_weight[0][refn0], h->luma_weight[1][refn1],
2421 h->luma_offset[0][refn0], h->luma_offset[1][refn1]);
2422 chroma_weight_avg(dest_cb, tmp_cb, s->uvlinesize, h->chroma_log2_weight_denom,
2423 h->chroma_weight[0][refn0][0], h->chroma_weight[1][refn1][0],
2424 h->chroma_offset[0][refn0][0], h->chroma_offset[1][refn1][0]);
2425 chroma_weight_avg(dest_cr, tmp_cr, s->uvlinesize, h->chroma_log2_weight_denom,
2426 h->chroma_weight[0][refn0][1], h->chroma_weight[1][refn1][1],
2427 h->chroma_offset[0][refn0][1], h->chroma_offset[1][refn1][1]);
2430 int list = list1 ? 1 : 0;
2431 int refn = h->ref_cache[list][ scan8[n] ];
2432 Picture *ref= &h->ref_list[list][refn];
2433 mc_dir_part(h, ref, n, square, chroma_height, delta, list,
2434 dest_y, dest_cb, dest_cr, x_offset, y_offset,
2435 qpix_put, chroma_put);
2437 luma_weight_op(dest_y, s->linesize, h->luma_log2_weight_denom,
2438 h->luma_weight[list][refn], h->luma_offset[list][refn]);
2439 if(h->use_weight_chroma){
2440 chroma_weight_op(dest_cb, s->uvlinesize, h->chroma_log2_weight_denom,
2441 h->chroma_weight[list][refn][0], h->chroma_offset[list][refn][0]);
2442 chroma_weight_op(dest_cr, s->uvlinesize, h->chroma_log2_weight_denom,
2443 h->chroma_weight[list][refn][1], h->chroma_offset[list][refn][1]);
2448 static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
2449 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2450 int x_offset, int y_offset,
2451 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2452 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
2453 h264_weight_func *weight_op, h264_biweight_func *weight_avg,
2454 int list0, int list1){
2455 if((h->use_weight==2 && list0 && list1
2456 && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ] != 32))
2457 || h->use_weight==1)
2458 mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
2459 x_offset, y_offset, qpix_put, chroma_put,
2460 weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
2462 mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
2463 x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
2466 static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2467 qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
2468 qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
2469 h264_weight_func *weight_op, h264_biweight_func *weight_avg){
2470 MpegEncContext * const s = &h->s;
2471 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
2472 const int mb_type= s->current_picture.mb_type[mb_xy];
2474 assert(IS_INTER(mb_type));
2476 if(IS_16X16(mb_type)){
2477 mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
2478 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
2479 &weight_op[0], &weight_avg[0],
2480 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2481 }else if(IS_16X8(mb_type)){
2482 mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
2483 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
2484 &weight_op[1], &weight_avg[1],
2485 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2486 mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
2487 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
2488 &weight_op[1], &weight_avg[1],
2489 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2490 }else if(IS_8X16(mb_type)){
2491 mc_part(h, 0, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 0, 0,
2492 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2493 &weight_op[2], &weight_avg[2],
2494 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2495 mc_part(h, 4, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 4, 0,
2496 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2497 &weight_op[2], &weight_avg[2],
2498 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2502 assert(IS_8X8(mb_type));
2505 const int sub_mb_type= h->sub_mb_type[i];
2507 int x_offset= (i&1)<<2;
2508 int y_offset= (i&2)<<1;
2510 if(IS_SUB_8X8(sub_mb_type)){
2511 mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2512 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2513 &weight_op[3], &weight_avg[3],
2514 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2515 }else if(IS_SUB_8X4(sub_mb_type)){
2516 mc_part(h, n , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2517 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2518 &weight_op[4], &weight_avg[4],
2519 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2520 mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
2521 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2522 &weight_op[4], &weight_avg[4],
2523 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2524 }else if(IS_SUB_4X8(sub_mb_type)){
2525 mc_part(h, n , 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2526 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2527 &weight_op[5], &weight_avg[5],
2528 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2529 mc_part(h, n+1, 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
2530 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2531 &weight_op[5], &weight_avg[5],
2532 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2535 assert(IS_SUB_4X4(sub_mb_type));
2537 int sub_x_offset= x_offset + 2*(j&1);
2538 int sub_y_offset= y_offset + (j&2);
2539 mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
2540 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2541 &weight_op[6], &weight_avg[6],
2542 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2549 static void decode_init_vlc(H264Context *h){
2550 static int done = 0;
2556 init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5,
2557 &chroma_dc_coeff_token_len [0], 1, 1,
2558 &chroma_dc_coeff_token_bits[0], 1, 1, 1);
2561 init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17,
2562 &coeff_token_len [i][0], 1, 1,
2563 &coeff_token_bits[i][0], 1, 1, 1);
2567 init_vlc(&chroma_dc_total_zeros_vlc[i], CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
2568 &chroma_dc_total_zeros_len [i][0], 1, 1,
2569 &chroma_dc_total_zeros_bits[i][0], 1, 1, 1);
2571 for(i=0; i<15; i++){
2572 init_vlc(&total_zeros_vlc[i], TOTAL_ZEROS_VLC_BITS, 16,
2573 &total_zeros_len [i][0], 1, 1,
2574 &total_zeros_bits[i][0], 1, 1, 1);
2578 init_vlc(&run_vlc[i], RUN_VLC_BITS, 7,
2579 &run_len [i][0], 1, 1,
2580 &run_bits[i][0], 1, 1, 1);
2582 init_vlc(&run7_vlc, RUN7_VLC_BITS, 16,
2583 &run_len [6][0], 1, 1,
2584 &run_bits[6][0], 1, 1, 1);
2589 * Sets the intra prediction function pointers.
2591 static void init_pred_ptrs(H264Context *h){
2592 // MpegEncContext * const s = &h->s;
2594 h->pred4x4[VERT_PRED ]= pred4x4_vertical_c;
2595 h->pred4x4[HOR_PRED ]= pred4x4_horizontal_c;
2596 h->pred4x4[DC_PRED ]= pred4x4_dc_c;
2597 h->pred4x4[DIAG_DOWN_LEFT_PRED ]= pred4x4_down_left_c;
2598 h->pred4x4[DIAG_DOWN_RIGHT_PRED]= pred4x4_down_right_c;
2599 h->pred4x4[VERT_RIGHT_PRED ]= pred4x4_vertical_right_c;
2600 h->pred4x4[HOR_DOWN_PRED ]= pred4x4_horizontal_down_c;
2601 h->pred4x4[VERT_LEFT_PRED ]= pred4x4_vertical_left_c;
2602 h->pred4x4[HOR_UP_PRED ]= pred4x4_horizontal_up_c;
2603 h->pred4x4[LEFT_DC_PRED ]= pred4x4_left_dc_c;
2604 h->pred4x4[TOP_DC_PRED ]= pred4x4_top_dc_c;
2605 h->pred4x4[DC_128_PRED ]= pred4x4_128_dc_c;
2607 h->pred8x8[DC_PRED8x8 ]= pred8x8_dc_c;
2608 h->pred8x8[VERT_PRED8x8 ]= pred8x8_vertical_c;
2609 h->pred8x8[HOR_PRED8x8 ]= pred8x8_horizontal_c;
2610 h->pred8x8[PLANE_PRED8x8 ]= pred8x8_plane_c;
2611 h->pred8x8[LEFT_DC_PRED8x8]= pred8x8_left_dc_c;
2612 h->pred8x8[TOP_DC_PRED8x8 ]= pred8x8_top_dc_c;
2613 h->pred8x8[DC_128_PRED8x8 ]= pred8x8_128_dc_c;
2615 h->pred16x16[DC_PRED8x8 ]= pred16x16_dc_c;
2616 h->pred16x16[VERT_PRED8x8 ]= pred16x16_vertical_c;
2617 h->pred16x16[HOR_PRED8x8 ]= pred16x16_horizontal_c;
2618 h->pred16x16[PLANE_PRED8x8 ]= pred16x16_plane_c;
2619 h->pred16x16[LEFT_DC_PRED8x8]= pred16x16_left_dc_c;
2620 h->pred16x16[TOP_DC_PRED8x8 ]= pred16x16_top_dc_c;
2621 h->pred16x16[DC_128_PRED8x8 ]= pred16x16_128_dc_c;
2624 static void free_tables(H264Context *h){
2625 av_freep(&h->intra4x4_pred_mode);
2626 av_freep(&h->chroma_pred_mode_table);
2627 av_freep(&h->cbp_table);
2628 av_freep(&h->mvd_table[0]);
2629 av_freep(&h->mvd_table[1]);
2630 av_freep(&h->direct_table);
2631 av_freep(&h->non_zero_count);
2632 av_freep(&h->slice_table_base);
2633 av_freep(&h->top_borders[1]);
2634 av_freep(&h->top_borders[0]);
2635 h->slice_table= NULL;
2637 av_freep(&h->mb2b_xy);
2638 av_freep(&h->mb2b8_xy);
2640 av_freep(&h->s.obmc_scratchpad);
2645 * needs width/height
2647 static int alloc_tables(H264Context *h){
2648 MpegEncContext * const s = &h->s;
2649 const int big_mb_num= s->mb_stride * (s->mb_height+1);
2652 CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8 * sizeof(uint8_t))
2654 CHECKED_ALLOCZ(h->non_zero_count , big_mb_num * 16 * sizeof(uint8_t))
2655 CHECKED_ALLOCZ(h->slice_table_base , big_mb_num * sizeof(uint8_t))
2656 CHECKED_ALLOCZ(h->top_borders[0] , s->mb_width * (16+8+8) * sizeof(uint8_t))
2657 CHECKED_ALLOCZ(h->top_borders[1] , s->mb_width * (16+8+8) * sizeof(uint8_t))
2658 CHECKED_ALLOCZ(h->cbp_table, big_mb_num * sizeof(uint16_t))
2660 if( h->pps.cabac ) {
2661 CHECKED_ALLOCZ(h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t))
2662 CHECKED_ALLOCZ(h->mvd_table[0], 32*big_mb_num * sizeof(uint16_t));
2663 CHECKED_ALLOCZ(h->mvd_table[1], 32*big_mb_num * sizeof(uint16_t));
2664 CHECKED_ALLOCZ(h->direct_table, 32*big_mb_num * sizeof(uint8_t));
2667 memset(h->slice_table_base, -1, big_mb_num * sizeof(uint8_t));
2668 h->slice_table= h->slice_table_base + s->mb_stride + 1;
2670 CHECKED_ALLOCZ(h->mb2b_xy , big_mb_num * sizeof(uint32_t));
2671 CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint32_t));
2672 for(y=0; y<s->mb_height; y++){
2673 for(x=0; x<s->mb_width; x++){
2674 const int mb_xy= x + y*s->mb_stride;
2675 const int b_xy = 4*x + 4*y*h->b_stride;
2676 const int b8_xy= 2*x + 2*y*h->b8_stride;
2678 h->mb2b_xy [mb_xy]= b_xy;
2679 h->mb2b8_xy[mb_xy]= b8_xy;
2683 s->obmc_scratchpad = NULL;
2691 static void common_init(H264Context *h){
2692 MpegEncContext * const s = &h->s;
2694 s->width = s->avctx->width;
2695 s->height = s->avctx->height;
2696 s->codec_id= s->avctx->codec->id;
2700 s->unrestricted_mv=1;
2701 s->decode=1; //FIXME
2704 static int decode_init(AVCodecContext *avctx){
2705 H264Context *h= avctx->priv_data;
2706 MpegEncContext * const s = &h->s;
2708 MPV_decode_defaults(s);
2713 s->out_format = FMT_H264;
2714 s->workaround_bugs= avctx->workaround_bugs;
2717 // s->decode_mb= ff_h263_decode_mb;
2719 avctx->pix_fmt= PIX_FMT_YUV420P;
2723 if(avctx->extradata_size > 0 && avctx->extradata &&
2724 *(char *)avctx->extradata == 1){
2734 static void frame_start(H264Context *h){
2735 MpegEncContext * const s = &h->s;
2738 MPV_frame_start(s, s->avctx);
2739 ff_er_frame_start(s);
2741 assert(s->linesize && s->uvlinesize);
2743 for(i=0; i<16; i++){
2744 h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
2745 h->block_offset[24+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
2748 h->block_offset[16+i]=
2749 h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2750 h->block_offset[24+16+i]=
2751 h->block_offset[24+20+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2754 /* can't be in alloc_tables because linesize isn't known there.
2755 * FIXME: redo bipred weight to not require extra buffer? */
2756 if(!s->obmc_scratchpad)
2757 s->obmc_scratchpad = av_malloc(16*s->linesize + 2*8*s->uvlinesize);
2759 // s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
2762 static inline void backup_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){
2763 MpegEncContext * const s = &h->s;
2767 src_cb -= uvlinesize;
2768 src_cr -= uvlinesize;
2770 // There are two lines saved, the line above the the top macroblock of a pair,
2771 // and the line above the bottom macroblock
2772 h->left_border[0]= h->top_borders[0][s->mb_x][15];
2773 for(i=1; i<17; i++){
2774 h->left_border[i]= src_y[15+i* linesize];
2777 *(uint64_t*)(h->top_borders[0][s->mb_x]+0)= *(uint64_t*)(src_y + 16*linesize);
2778 *(uint64_t*)(h->top_borders[0][s->mb_x]+8)= *(uint64_t*)(src_y +8+16*linesize);
2780 if(!(s->flags&CODEC_FLAG_GRAY)){
2781 h->left_border[17 ]= h->top_borders[0][s->mb_x][16+7];
2782 h->left_border[17+9]= h->top_borders[0][s->mb_x][24+7];
2784 h->left_border[i+17 ]= src_cb[7+i*uvlinesize];
2785 h->left_border[i+17+9]= src_cr[7+i*uvlinesize];
2787 *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+8*uvlinesize);
2788 *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+8*uvlinesize);
2792 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){
2793 MpegEncContext * const s = &h->s;
2796 int deblock_left = (s->mb_x > 0);
2797 int deblock_top = (s->mb_y > 0);
2799 src_y -= linesize + 1;
2800 src_cb -= uvlinesize + 1;
2801 src_cr -= uvlinesize + 1;
2803 #define XCHG(a,b,t,xchg)\
2810 for(i = !deblock_top; i<17; i++){
2811 XCHG(h->left_border[i ], src_y [i* linesize], temp8, xchg);
2816 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
2817 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
2820 if(!(s->flags&CODEC_FLAG_GRAY)){
2822 for(i = !deblock_top; i<9; i++){
2823 XCHG(h->left_border[i+17 ], src_cb[i*uvlinesize], temp8, xchg);
2824 XCHG(h->left_border[i+17+9], src_cr[i*uvlinesize], temp8, xchg);
2828 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
2829 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
2834 static inline void backup_pair_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){
2835 MpegEncContext * const s = &h->s;
2838 src_y -= 2 * linesize;
2839 src_cb -= 2 * uvlinesize;
2840 src_cr -= 2 * uvlinesize;
2842 // There are two lines saved, the line above the the top macroblock of a pair,
2843 // and the line above the bottom macroblock
2844 h->left_border[0]= h->top_borders[0][s->mb_x][15];
2845 h->left_border[1]= h->top_borders[1][s->mb_x][15];
2846 for(i=2; i<34; i++){
2847 h->left_border[i]= src_y[15+i* linesize];
2850 *(uint64_t*)(h->top_borders[0][s->mb_x]+0)= *(uint64_t*)(src_y + 32*linesize);
2851 *(uint64_t*)(h->top_borders[0][s->mb_x]+8)= *(uint64_t*)(src_y +8+32*linesize);
2852 *(uint64_t*)(h->top_borders[1][s->mb_x]+0)= *(uint64_t*)(src_y + 33*linesize);
2853 *(uint64_t*)(h->top_borders[1][s->mb_x]+8)= *(uint64_t*)(src_y +8+33*linesize);
2855 if(!(s->flags&CODEC_FLAG_GRAY)){
2856 h->left_border[34 ]= h->top_borders[0][s->mb_x][16+7];
2857 h->left_border[34+ 1]= h->top_borders[1][s->mb_x][16+7];
2858 h->left_border[34+18 ]= h->top_borders[0][s->mb_x][24+7];
2859 h->left_border[34+18+1]= h->top_borders[1][s->mb_x][24+7];
2860 for(i=2; i<18; i++){
2861 h->left_border[i+34 ]= src_cb[7+i*uvlinesize];
2862 h->left_border[i+34+18]= src_cr[7+i*uvlinesize];
2864 *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+16*uvlinesize);
2865 *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+16*uvlinesize);
2866 *(uint64_t*)(h->top_borders[1][s->mb_x]+16)= *(uint64_t*)(src_cb+17*uvlinesize);
2867 *(uint64_t*)(h->top_borders[1][s->mb_x]+24)= *(uint64_t*)(src_cr+17*uvlinesize);
2871 static inline void xchg_pair_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize, int xchg){
2872 MpegEncContext * const s = &h->s;
2875 int deblock_left = (s->mb_x > 0);
2876 int deblock_top = (s->mb_y > 0);
2878 tprintf("xchg_pair_border: src_y:%p src_cb:%p src_cr:%p ls:%d uvls:%d\n", src_y, src_cb, src_cr, linesize, uvlinesize);
2880 src_y -= 2 * linesize + 1;
2881 src_cb -= 2 * uvlinesize + 1;
2882 src_cr -= 2 * uvlinesize + 1;
2884 #define XCHG(a,b,t,xchg)\
2891 for(i = (!deblock_top)<<1; i<34; i++){
2892 XCHG(h->left_border[i ], src_y [i* linesize], temp8, xchg);
2897 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
2898 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
2899 XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+0), *(uint64_t*)(src_y +1 +linesize), temp64, xchg);
2900 XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+8), *(uint64_t*)(src_y +9 +linesize), temp64, 1);
2903 if(!(s->flags&CODEC_FLAG_GRAY)){
2905 for(i = (!deblock_top) << 1; i<18; i++){
2906 XCHG(h->left_border[i+34 ], src_cb[i*uvlinesize], temp8, xchg);
2907 XCHG(h->left_border[i+34+18], src_cr[i*uvlinesize], temp8, xchg);
2911 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
2912 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
2913 XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+16), *(uint64_t*)(src_cb+1 +uvlinesize), temp64, 1);
2914 XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+24), *(uint64_t*)(src_cr+1 +uvlinesize), temp64, 1);
2919 static void hl_decode_mb(H264Context *h){
2920 MpegEncContext * const s = &h->s;
2921 const int mb_x= s->mb_x;
2922 const int mb_y= s->mb_y;
2923 const int mb_xy= mb_x + mb_y*s->mb_stride;
2924 const int mb_type= s->current_picture.mb_type[mb_xy];
2925 uint8_t *dest_y, *dest_cb, *dest_cr;
2926 int linesize, uvlinesize /*dct_offset*/;
2928 int *block_offset = &h->block_offset[0];
2929 const unsigned int bottom = mb_y & 1;
2934 dest_y = s->current_picture.data[0] + (mb_y * 16* s->linesize ) + mb_x * 16;
2935 dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2936 dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2938 if (h->mb_field_decoding_flag) {
2939 linesize = s->linesize * 2;
2940 uvlinesize = s->uvlinesize * 2;
2941 block_offset = &h->block_offset[24];
2942 if(mb_y&1){ //FIXME move out of this func?
2943 dest_y -= s->linesize*15;
2944 dest_cb-= s->uvlinesize*7;
2945 dest_cr-= s->uvlinesize*7;
2948 linesize = s->linesize;
2949 uvlinesize = s->uvlinesize;
2950 // dct_offset = s->linesize * 16;
2953 if (IS_INTRA_PCM(mb_type)) {
2956 // The pixels are stored in h->mb array in the same order as levels,
2957 // copy them in output in the correct order.
2958 for(i=0; i<16; i++) {
2959 for (y=0; y<4; y++) {
2960 for (x=0; x<4; x++) {
2961 *(dest_y + block_offset[i] + y*linesize + x) = h->mb[i*16+y*4+x];
2965 for(i=16; i<16+4; i++) {
2966 for (y=0; y<4; y++) {
2967 for (x=0; x<4; x++) {
2968 *(dest_cb + block_offset[i] + y*uvlinesize + x) = h->mb[i*16+y*4+x];
2972 for(i=20; i<20+4; i++) {
2973 for (y=0; y<4; y++) {
2974 for (x=0; x<4; x++) {
2975 *(dest_cr + block_offset[i] + y*uvlinesize + x) = h->mb[i*16+y*4+x];
2980 if(IS_INTRA(mb_type)){
2981 if(h->deblocking_filter) {
2982 if (h->mb_aff_frame) {
2984 xchg_pair_border(h, dest_y, dest_cb, dest_cr, s->linesize, s->uvlinesize, 1);
2986 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1);
2990 if(!(s->flags&CODEC_FLAG_GRAY)){
2991 h->pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
2992 h->pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
2995 if(IS_INTRA4x4(mb_type)){
2997 for(i=0; i<16; i++){
2998 uint8_t * const ptr= dest_y + block_offset[i];
3000 const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
3003 if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
3004 const int topright_avail= (h->topright_samples_available<<i)&0x8000;
3005 assert(mb_y || linesize <= block_offset[i]);
3006 if(!topright_avail){
3007 tr= ptr[3 - linesize]*0x01010101;
3008 topright= (uint8_t*) &tr;
3009 }else if(i==5 && h->deblocking_filter){
3010 tr= *(uint32_t*)h->top_borders[h->mb_aff_frame ? IS_INTERLACED(mb_type) ? bottom : 1 : 0][mb_x+1];
3011 topright= (uint8_t*) &tr;
3013 topright= ptr + 4 - linesize;
3017 h->pred4x4[ dir ](ptr, topright, linesize);
3018 if(h->non_zero_count_cache[ scan8[i] ]){
3019 if(s->codec_id == CODEC_ID_H264)
3020 s->dsp.h264_idct_add(ptr, h->mb + i*16, linesize);
3022 svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
3027 h->pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
3028 if(s->codec_id == CODEC_ID_H264)
3029 h264_luma_dc_dequant_idct_c(h->mb, s->qscale);
3031 svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
3033 if(h->deblocking_filter) {
3034 if (h->mb_aff_frame) {
3036 uint8_t *pair_dest_y = s->current_picture.data[0] + ((mb_y-1) * 16* s->linesize ) + mb_x * 16;
3037 uint8_t *pair_dest_cb = s->current_picture.data[1] + ((mb_y-1) * 8 * s->uvlinesize) + mb_x * 8;
3038 uint8_t *pair_dest_cr = s->current_picture.data[2] + ((mb_y-1) * 8 * s->uvlinesize) + mb_x * 8;
3040 xchg_pair_border(h, pair_dest_y, pair_dest_cb, pair_dest_cr, s->linesize, s->uvlinesize, 0);
3044 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);
3047 }else if(s->codec_id == CODEC_ID_H264){
3048 hl_motion(h, dest_y, dest_cb, dest_cr,
3049 s->dsp.put_h264_qpel_pixels_tab, s->dsp.put_h264_chroma_pixels_tab,
3050 s->dsp.avg_h264_qpel_pixels_tab, s->dsp.avg_h264_chroma_pixels_tab,
3051 s->dsp.weight_h264_pixels_tab, s->dsp.biweight_h264_pixels_tab);
3055 if(!IS_INTRA4x4(mb_type)){
3056 if(s->codec_id == CODEC_ID_H264){
3057 for(i=0; i<16; i++){
3058 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
3059 uint8_t * const ptr= dest_y + block_offset[i];
3060 s->dsp.h264_idct_add(ptr, h->mb + i*16, linesize);
3064 for(i=0; i<16; i++){
3065 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
3066 uint8_t * const ptr= dest_y + block_offset[i];
3067 svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
3073 if(!(s->flags&CODEC_FLAG_GRAY)){
3074 chroma_dc_dequant_idct_c(h->mb + 16*16, h->chroma_qp);
3075 chroma_dc_dequant_idct_c(h->mb + 16*16+4*16, h->chroma_qp);
3076 if(s->codec_id == CODEC_ID_H264){
3077 for(i=16; i<16+4; i++){
3078 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
3079 uint8_t * const ptr= dest_cb + block_offset[i];
3080 s->dsp.h264_idct_add(ptr, h->mb + i*16, uvlinesize);
3083 for(i=20; i<20+4; i++){
3084 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
3085 uint8_t * const ptr= dest_cr + block_offset[i];
3086 s->dsp.h264_idct_add(ptr, h->mb + i*16, uvlinesize);
3090 for(i=16; i<16+4; i++){
3091 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
3092 uint8_t * const ptr= dest_cb + block_offset[i];
3093 svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
3096 for(i=20; i<20+4; i++){
3097 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
3098 uint8_t * const ptr= dest_cr + block_offset[i];
3099 svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
3105 if(h->deblocking_filter) {
3106 if (h->mb_aff_frame) {
3107 const int mb_y = s->mb_y - 1;
3108 uint8_t *pair_dest_y, *pair_dest_cb, *pair_dest_cr;
3109 const int mb_xy= mb_x + mb_y*s->mb_stride;
3110 const int mb_type_top = s->current_picture.mb_type[mb_xy];
3111 const int mb_type_bottom= s->current_picture.mb_type[mb_xy+s->mb_stride];
3112 uint8_t tmp = s->current_picture.data[1][384];
3113 if (!bottom) return;
3114 pair_dest_y = s->current_picture.data[0] + (mb_y * 16* s->linesize ) + mb_x * 16;
3115 pair_dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
3116 pair_dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
3118 backup_pair_border(h, pair_dest_y, pair_dest_cb, pair_dest_cr, s->linesize, s->uvlinesize);
3119 // TODO deblock a pair
3122 tprintf("call mbaff filter_mb mb_x:%d mb_y:%d pair_dest_y = %p, dest_y = %p\n", mb_x, mb_y, pair_dest_y, dest_y);
3123 fill_caches(h, mb_type_top, 1); //FIXME don't fill stuff which isn't used by filter_mb
3124 filter_mb(h, mb_x, mb_y, pair_dest_y, pair_dest_cb, pair_dest_cr, linesize, uvlinesize);
3125 if (tmp != s->current_picture.data[1][384]) {
3126 tprintf("modified pixel 8,1 (1)\n");
3130 tprintf("call mbaff filter_mb\n");
3131 fill_caches(h, mb_type_bottom, 1); //FIXME don't fill stuff which isn't used by filter_mb
3132 filter_mb(h, mb_x, mb_y+1, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
3133 if (tmp != s->current_picture.data[1][384]) {
3134 tprintf("modified pixel 8,1 (2)\n");
3137 tprintf("call filter_mb\n");
3138 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
3139 fill_caches(h, mb_type, 1); //FIXME don't fill stuff which isn't used by filter_mb
3140 filter_mb(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
3146 * fills the default_ref_list.
3148 static int fill_default_ref_list(H264Context *h){
3149 MpegEncContext * const s = &h->s;
3151 int smallest_poc_greater_than_current = -1;
3152 Picture sorted_short_ref[32];
3154 if(h->slice_type==B_TYPE){
3158 /* sort frame according to poc in B slice */
3159 for(out_i=0; out_i<h->short_ref_count; out_i++){
3161 int best_poc=INT_MAX;
3163 for(i=0; i<h->short_ref_count; i++){
3164 const int poc= h->short_ref[i]->poc;
3165 if(poc > limit && poc < best_poc){
3171 assert(best_i != -1);
3174 sorted_short_ref[out_i]= *h->short_ref[best_i];
3175 tprintf("sorted poc: %d->%d poc:%d fn:%d\n", best_i, out_i, sorted_short_ref[out_i].poc, sorted_short_ref[out_i].frame_num);
3176 if (-1 == smallest_poc_greater_than_current) {
3177 if (h->short_ref[best_i]->poc >= s->current_picture_ptr->poc) {
3178 smallest_poc_greater_than_current = out_i;
3184 if(s->picture_structure == PICT_FRAME){
3185 if(h->slice_type==B_TYPE){
3187 tprintf("current poc: %d, smallest_poc_greater_than_current: %d\n", s->current_picture_ptr->poc, smallest_poc_greater_than_current);
3189 // find the largest poc
3190 for(list=0; list<2; list++){
3193 int step= list ? -1 : 1;
3195 for(i=0; i<h->short_ref_count && index < h->ref_count[list]; i++, j+=step) {
3196 while(j<0 || j>= h->short_ref_count){
3198 j= smallest_poc_greater_than_current + (step>>1);
3200 if(sorted_short_ref[j].reference != 3) continue;
3201 h->default_ref_list[list][index ]= sorted_short_ref[j];
3202 h->default_ref_list[list][index++].pic_id= sorted_short_ref[j].frame_num;
3205 for(i = 0; i < 16 && index < h->ref_count[ list ]; i++){
3206 if(h->long_ref[i] == NULL) continue;
3207 if(h->long_ref[i]->reference != 3) continue;
3209 h->default_ref_list[ list ][index ]= *h->long_ref[i];
3210 h->default_ref_list[ list ][index++].pic_id= i;;
3213 if(list && (smallest_poc_greater_than_current<=0 || smallest_poc_greater_than_current>=h->short_ref_count) && (1 < index)){
3214 // swap the two first elements of L1 when
3215 // L0 and L1 are identical
3216 Picture temp= h->default_ref_list[1][0];
3217 h->default_ref_list[1][0] = h->default_ref_list[1][1];
3218 h->default_ref_list[1][0] = temp;
3221 if(index < h->ref_count[ list ])
3222 memset(&h->default_ref_list[list][index], 0, sizeof(Picture)*(h->ref_count[ list ] - index));
3226 for(i=0; i<h->short_ref_count; i++){
3227 if(h->short_ref[i]->reference != 3) continue; //FIXME refernce field shit
3228 h->default_ref_list[0][index ]= *h->short_ref[i];
3229 h->default_ref_list[0][index++].pic_id= h->short_ref[i]->frame_num;
3231 for(i = 0; i < 16; i++){
3232 if(h->long_ref[i] == NULL) continue;
3233 if(h->long_ref[i]->reference != 3) continue;
3234 h->default_ref_list[0][index ]= *h->long_ref[i];
3235 h->default_ref_list[0][index++].pic_id= i;;
3237 if(index < h->ref_count[0])
3238 memset(&h->default_ref_list[0][index], 0, sizeof(Picture)*(h->ref_count[0] - index));
3241 if(h->slice_type==B_TYPE){
3243 //FIXME second field balh
3247 for (i=0; i<h->ref_count[0]; i++) {
3248 tprintf("List0: %s fn:%d 0x%p\n", (h->default_ref_list[0][i].long_ref ? "LT" : "ST"), h->default_ref_list[0][i].pic_id, h->default_ref_list[0][i].data[0]);
3250 if(h->slice_type==B_TYPE){
3251 for (i=0; i<h->ref_count[1]; i++) {
3252 tprintf("List1: %s fn:%d 0x%p\n", (h->default_ref_list[1][i].long_ref ? "LT" : "ST"), h->default_ref_list[1][i].pic_id, h->default_ref_list[0][i].data[0]);
3259 static void print_short_term(H264Context *h);
3260 static void print_long_term(H264Context *h);
3262 static int decode_ref_pic_list_reordering(H264Context *h){
3263 MpegEncContext * const s = &h->s;
3266 print_short_term(h);
3268 if(h->slice_type==I_TYPE || h->slice_type==SI_TYPE) return 0; //FIXME move before func
3270 for(list=0; list<2; list++){
3271 memcpy(h->ref_list[list], h->default_ref_list[list], sizeof(Picture)*h->ref_count[list]);
3273 if(get_bits1(&s->gb)){
3274 int pred= h->curr_pic_num;
3277 for(index=0; ; index++){
3278 int reordering_of_pic_nums_idc= get_ue_golomb(&s->gb);
3281 Picture *ref = NULL;
3283 if(reordering_of_pic_nums_idc==3)
3286 if(index >= h->ref_count[list]){
3287 av_log(h->s.avctx, AV_LOG_ERROR, "reference count overflow\n");
3291 if(reordering_of_pic_nums_idc<3){
3292 if(reordering_of_pic_nums_idc<2){
3293 const int abs_diff_pic_num= get_ue_golomb(&s->gb) + 1;
3295 if(abs_diff_pic_num >= h->max_pic_num){
3296 av_log(h->s.avctx, AV_LOG_ERROR, "abs_diff_pic_num overflow\n");
3300 if(reordering_of_pic_nums_idc == 0) pred-= abs_diff_pic_num;
3301 else pred+= abs_diff_pic_num;
3302 pred &= h->max_pic_num - 1;
3304 for(i= h->ref_count[list]-1; i>=0; i--){
3305 if(h->ref_list[list][i].data[0] != NULL && h->ref_list[list][i].pic_id == pred && h->ref_list[list][i].long_ref==0) // ignore non existing pictures by testing data[0] pointer
3309 pic_id= get_ue_golomb(&s->gb); //long_term_pic_idx
3311 for(i= h->ref_count[list]-1; i>=0; i--){
3312 if(h->ref_list[list][i].pic_id == pic_id && h->ref_list[list][i].long_ref==1) // no need to ignore non existing pictures as non existing pictures have long_ref==0
3318 av_log(h->s.avctx, AV_LOG_ERROR, "reference picture missing during reorder\n");
3319 memset(&h->ref_list[list][index], 0, sizeof(Picture)); //FIXME
3320 } else if (i != index) /* this test is not necessary, it is only an optimisation to skip double copy of Picture structure in this case */ {
3321 Picture tmp= h->ref_list[list][i];
3323 i = h->ref_count[list];
3325 for(; i > index; i--){
3326 h->ref_list[list][i]= h->ref_list[list][i-1];
3328 h->ref_list[list][index]= tmp;
3331 av_log(h->s.avctx, AV_LOG_ERROR, "illegal reordering_of_pic_nums_idc\n");
3337 if(h->slice_type!=B_TYPE) break;
3340 if(h->slice_type==B_TYPE && !h->direct_spatial_mv_pred)
3341 direct_dist_scale_factor(h);
3342 direct_ref_list_init(h);
3346 static int pred_weight_table(H264Context *h){
3347 MpegEncContext * const s = &h->s;
3349 int luma_def, chroma_def;
3352 h->use_weight_chroma= 0;
3353 h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
3354 h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
3355 luma_def = 1<<h->luma_log2_weight_denom;
3356 chroma_def = 1<<h->chroma_log2_weight_denom;
3358 for(list=0; list<2; list++){
3359 for(i=0; i<h->ref_count[list]; i++){
3360 int luma_weight_flag, chroma_weight_flag;
3362 luma_weight_flag= get_bits1(&s->gb);
3363 if(luma_weight_flag){
3364 h->luma_weight[list][i]= get_se_golomb(&s->gb);
3365 h->luma_offset[list][i]= get_se_golomb(&s->gb);
3366 if( h->luma_weight[list][i] != luma_def
3367 || h->luma_offset[list][i] != 0)
3370 h->luma_weight[list][i]= luma_def;
3371 h->luma_offset[list][i]= 0;
3374 chroma_weight_flag= get_bits1(&s->gb);
3375 if(chroma_weight_flag){
3378 h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);
3379 h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);
3380 if( h->chroma_weight[list][i][j] != chroma_def
3381 || h->chroma_offset[list][i][j] != 0)
3382 h->use_weight_chroma= 1;
3387 h->chroma_weight[list][i][j]= chroma_def;
3388 h->chroma_offset[list][i][j]= 0;
3392 if(h->slice_type != B_TYPE) break;
3394 h->use_weight= h->use_weight || h->use_weight_chroma;
3398 static void implicit_weight_table(H264Context *h){
3399 MpegEncContext * const s = &h->s;
3401 int cur_poc = s->current_picture_ptr->poc;
3403 if( h->ref_count[0] == 1 && h->ref_count[1] == 1
3404 && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
3406 h->use_weight_chroma= 0;
3411 h->use_weight_chroma= 2;
3412 h->luma_log2_weight_denom= 5;
3413 h->chroma_log2_weight_denom= 5;
3416 for(ref0=0; ref0 < h->ref_count[0]; ref0++){
3417 int poc0 = h->ref_list[0][ref0].poc;
3418 for(ref1=0; ref1 < h->ref_count[1]; ref1++){
3419 int poc1 = h->ref_list[1][ref1].poc;
3420 int td = clip(poc1 - poc0, -128, 127);
3422 int tb = clip(cur_poc - poc0, -128, 127);
3423 int tx = (16384 + (ABS(td) >> 1)) / td;
3424 int dist_scale_factor = clip((tb*tx + 32) >> 6, -1024, 1023) >> 2;
3425 if(dist_scale_factor < -64 || dist_scale_factor > 128)
3426 h->implicit_weight[ref0][ref1] = 32;
3428 h->implicit_weight[ref0][ref1] = 64 - dist_scale_factor;
3430 h->implicit_weight[ref0][ref1] = 32;
3435 static inline void unreference_pic(H264Context *h, Picture *pic){
3438 if(pic == h->delayed_output_pic)
3441 for(i = 0; h->delayed_pic[i]; i++)
3442 if(pic == h->delayed_pic[i]){
3450 * instantaneous decoder refresh.
3452 static void idr(H264Context *h){
3455 for(i=0; i<16; i++){
3456 if (h->long_ref[i] != NULL) {
3457 unreference_pic(h, h->long_ref[i]);
3458 h->long_ref[i]= NULL;
3461 h->long_ref_count=0;
3463 for(i=0; i<h->short_ref_count; i++){
3464 unreference_pic(h, h->short_ref[i]);
3465 h->short_ref[i]= NULL;
3467 h->short_ref_count=0;
3470 /* forget old pics after a seek */
3471 static void flush_dpb(AVCodecContext *avctx){
3472 H264Context *h= avctx->priv_data;
3475 h->delayed_pic[i]= NULL;
3476 h->delayed_output_pic= NULL;
3482 * @return the removed picture or NULL if an error occurs
3484 static Picture * remove_short(H264Context *h, int frame_num){
3485 MpegEncContext * const s = &h->s;
3488 if(s->avctx->debug&FF_DEBUG_MMCO)
3489 av_log(h->s.avctx, AV_LOG_DEBUG, "remove short %d count %d\n", frame_num, h->short_ref_count);
3491 for(i=0; i<h->short_ref_count; i++){
3492 Picture *pic= h->short_ref[i];
3493 if(s->avctx->debug&FF_DEBUG_MMCO)
3494 av_log(h->s.avctx, AV_LOG_DEBUG, "%d %d %p\n", i, pic->frame_num, pic);
3495 if(pic->frame_num == frame_num){
3496 h->short_ref[i]= NULL;
3497 memmove(&h->short_ref[i], &h->short_ref[i+1], (h->short_ref_count - i - 1)*sizeof(Picture*));
3498 h->short_ref_count--;
3507 * @return the removed picture or NULL if an error occurs
3509 static Picture * remove_long(H264Context *h, int i){
3512 pic= h->long_ref[i];
3513 h->long_ref[i]= NULL;
3514 if(pic) h->long_ref_count--;
3520 * print short term list
3522 static void print_short_term(H264Context *h) {
3524 if(h->s.avctx->debug&FF_DEBUG_MMCO) {
3525 av_log(h->s.avctx, AV_LOG_DEBUG, "short term list:\n");
3526 for(i=0; i<h->short_ref_count; i++){
3527 Picture *pic= h->short_ref[i];
3528 av_log(h->s.avctx, AV_LOG_DEBUG, "%d fn:%d poc:%d %p\n", i, pic->frame_num, pic->poc, pic->data[0]);
3534 * print long term list
3536 static void print_long_term(H264Context *h) {
3538 if(h->s.avctx->debug&FF_DEBUG_MMCO) {
3539 av_log(h->s.avctx, AV_LOG_DEBUG, "long term list:\n");
3540 for(i = 0; i < 16; i++){
3541 Picture *pic= h->long_ref[i];
3543 av_log(h->s.avctx, AV_LOG_DEBUG, "%d fn:%d poc:%d %p\n", i, pic->frame_num, pic->poc, pic->data[0]);
3550 * Executes the reference picture marking (memory management control operations).
3552 static int execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count){
3553 MpegEncContext * const s = &h->s;
3555 int current_is_long=0;
3558 if((s->avctx->debug&FF_DEBUG_MMCO) && mmco_count==0)
3559 av_log(h->s.avctx, AV_LOG_DEBUG, "no mmco here\n");
3561 for(i=0; i<mmco_count; i++){
3562 if(s->avctx->debug&FF_DEBUG_MMCO)
3563 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);
3565 switch(mmco[i].opcode){
3566 case MMCO_SHORT2UNUSED:
3567 pic= remove_short(h, mmco[i].short_frame_num);
3568 if(pic==NULL) return -1;
3569 unreference_pic(h, pic);
3571 case MMCO_SHORT2LONG:
3572 pic= remove_long(h, mmco[i].long_index);
3573 if(pic) unreference_pic(h, pic);
3575 h->long_ref[ mmco[i].long_index ]= remove_short(h, mmco[i].short_frame_num);
3576 h->long_ref[ mmco[i].long_index ]->long_ref=1;
3577 h->long_ref_count++;
3579 case MMCO_LONG2UNUSED:
3580 pic= remove_long(h, mmco[i].long_index);
3581 if(pic==NULL) return -1;
3582 unreference_pic(h, pic);
3585 pic= remove_long(h, mmco[i].long_index);
3586 if(pic) unreference_pic(h, pic);
3588 h->long_ref[ mmco[i].long_index ]= s->current_picture_ptr;
3589 h->long_ref[ mmco[i].long_index ]->long_ref=1;
3590 h->long_ref_count++;
3594 case MMCO_SET_MAX_LONG:
3595 assert(mmco[i].long_index <= 16);
3596 // just remove the long term which index is greater than new max
3597 for(j = mmco[i].long_index; j<16; j++){
3598 pic = remove_long(h, j);
3599 if (pic) unreference_pic(h, pic);
3603 while(h->short_ref_count){
3604 pic= remove_short(h, h->short_ref[0]->frame_num);
3605 unreference_pic(h, pic);
3607 for(j = 0; j < 16; j++) {
3608 pic= remove_long(h, j);
3609 if(pic) unreference_pic(h, pic);
3616 if(!current_is_long){
3617 pic= remove_short(h, s->current_picture_ptr->frame_num);
3619 unreference_pic(h, pic);
3620 av_log(h->s.avctx, AV_LOG_ERROR, "illegal short term buffer state detected\n");
3623 if(h->short_ref_count)
3624 memmove(&h->short_ref[1], &h->short_ref[0], h->short_ref_count*sizeof(Picture*));
3626 h->short_ref[0]= s->current_picture_ptr;
3627 h->short_ref[0]->long_ref=0;
3628 h->short_ref_count++;
3631 print_short_term(h);
3636 static int decode_ref_pic_marking(H264Context *h){
3637 MpegEncContext * const s = &h->s;
3640 if(h->nal_unit_type == NAL_IDR_SLICE){ //FIXME fields
3641 s->broken_link= get_bits1(&s->gb) -1;
3642 h->mmco[0].long_index= get_bits1(&s->gb) - 1; // current_long_term_idx
3643 if(h->mmco[0].long_index == -1)
3646 h->mmco[0].opcode= MMCO_LONG;
3650 if(get_bits1(&s->gb)){ // adaptive_ref_pic_marking_mode_flag
3651 for(i= 0; i<MAX_MMCO_COUNT; i++) {
3652 MMCOOpcode opcode= get_ue_golomb(&s->gb);;
3654 h->mmco[i].opcode= opcode;
3655 if(opcode==MMCO_SHORT2UNUSED || opcode==MMCO_SHORT2LONG){
3656 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
3657 /* if(h->mmco[i].short_frame_num >= h->short_ref_count || h->short_ref[ h->mmco[i].short_frame_num ] == NULL){
3658 fprintf(stderr, "illegal short ref in memory management control operation %d\n", mmco);
3662 if(opcode==MMCO_SHORT2LONG || opcode==MMCO_LONG2UNUSED || opcode==MMCO_LONG || opcode==MMCO_SET_MAX_LONG){
3663 h->mmco[i].long_index= get_ue_golomb(&s->gb);
3664 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){
3665 av_log(h->s.avctx, AV_LOG_ERROR, "illegal long ref in memory management control operation %d\n", opcode);
3670 if(opcode > MMCO_LONG){
3671 av_log(h->s.avctx, AV_LOG_ERROR, "illegal memory management control operation %d\n", opcode);
3674 if(opcode == MMCO_END)
3679 assert(h->long_ref_count + h->short_ref_count <= h->sps.ref_frame_count);
3681 if(h->long_ref_count + h->short_ref_count == h->sps.ref_frame_count){ //FIXME fields
3682 h->mmco[0].opcode= MMCO_SHORT2UNUSED;
3683 h->mmco[0].short_frame_num= h->short_ref[ h->short_ref_count - 1 ]->frame_num;
3693 static int init_poc(H264Context *h){
3694 MpegEncContext * const s = &h->s;
3695 const int max_frame_num= 1<<h->sps.log2_max_frame_num;
3698 if(h->nal_unit_type == NAL_IDR_SLICE){
3699 h->frame_num_offset= 0;
3701 if(h->frame_num < h->prev_frame_num)
3702 h->frame_num_offset= h->prev_frame_num_offset + max_frame_num;
3704 h->frame_num_offset= h->prev_frame_num_offset;
3707 if(h->sps.poc_type==0){
3708 const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
3710 if(h->nal_unit_type == NAL_IDR_SLICE){
3715 if (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
3716 h->poc_msb = h->prev_poc_msb + max_poc_lsb;
3717 else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
3718 h->poc_msb = h->prev_poc_msb - max_poc_lsb;
3720 h->poc_msb = h->prev_poc_msb;
3721 //printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
3723 field_poc[1] = h->poc_msb + h->poc_lsb;
3724 if(s->picture_structure == PICT_FRAME)
3725 field_poc[1] += h->delta_poc_bottom;
3726 }else if(h->sps.poc_type==1){
3727 int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
3730 if(h->sps.poc_cycle_length != 0)
3731 abs_frame_num = h->frame_num_offset + h->frame_num;
3735 if(h->nal_ref_idc==0 && abs_frame_num > 0)
3738 expected_delta_per_poc_cycle = 0;
3739 for(i=0; i < h->sps.poc_cycle_length; i++)
3740 expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
3742 if(abs_frame_num > 0){
3743 int poc_cycle_cnt = (abs_frame_num - 1) / h->sps.poc_cycle_length;
3744 int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
3746 expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
3747 for(i = 0; i <= frame_num_in_poc_cycle; i++)
3748 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
3752 if(h->nal_ref_idc == 0)
3753 expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
3755 field_poc[0] = expectedpoc + h->delta_poc[0];
3756 field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
3758 if(s->picture_structure == PICT_FRAME)
3759 field_poc[1] += h->delta_poc[1];
3762 if(h->nal_unit_type == NAL_IDR_SLICE){
3765 if(h->nal_ref_idc) poc= 2*(h->frame_num_offset + h->frame_num);
3766 else poc= 2*(h->frame_num_offset + h->frame_num) - 1;
3772 if(s->picture_structure != PICT_BOTTOM_FIELD)
3773 s->current_picture_ptr->field_poc[0]= field_poc[0];
3774 if(s->picture_structure != PICT_TOP_FIELD)
3775 s->current_picture_ptr->field_poc[1]= field_poc[1];
3776 if(s->picture_structure == PICT_FRAME) // FIXME field pix?
3777 s->current_picture_ptr->poc= FFMIN(field_poc[0], field_poc[1]);
3783 * decodes a slice header.
3784 * this will allso call MPV_common_init() and frame_start() as needed
3786 static int decode_slice_header(H264Context *h){
3787 MpegEncContext * const s = &h->s;
3788 int first_mb_in_slice, pps_id;
3789 int num_ref_idx_active_override_flag;
3790 static const uint8_t slice_type_map[5]= {P_TYPE, B_TYPE, I_TYPE, SP_TYPE, SI_TYPE};
3792 int default_ref_list_done = 0;
3794 s->current_picture.reference= h->nal_ref_idc != 0;
3795 s->dropable= h->nal_ref_idc == 0;
3797 first_mb_in_slice= get_ue_golomb(&s->gb);
3799 slice_type= get_ue_golomb(&s->gb);
3801 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);
3806 h->slice_type_fixed=1;
3808 h->slice_type_fixed=0;
3810 slice_type= slice_type_map[ slice_type ];
3811 if (slice_type == I_TYPE
3812 || (h->slice_num != 0 && slice_type == h->slice_type) ) {
3813 default_ref_list_done = 1;
3815 h->slice_type= slice_type;
3817 s->pict_type= h->slice_type; // to make a few old func happy, it's wrong though
3819 pps_id= get_ue_golomb(&s->gb);
3821 av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
3824 h->pps= h->pps_buffer[pps_id];
3825 if(h->pps.slice_group_count == 0){
3826 av_log(h->s.avctx, AV_LOG_ERROR, "non existing PPS referenced\n");
3830 h->sps= h->sps_buffer[ h->pps.sps_id ];
3831 if(h->sps.log2_max_frame_num == 0){
3832 av_log(h->s.avctx, AV_LOG_ERROR, "non existing SPS referenced\n");
3836 s->mb_width= h->sps.mb_width;
3837 s->mb_height= h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
3839 h->b_stride= s->mb_width*4 + 1;
3840 h->b8_stride= s->mb_width*2 + 1;
3842 s->width = 16*s->mb_width - 2*(h->sps.crop_left + h->sps.crop_right );
3843 if(h->sps.frame_mbs_only_flag)
3844 s->height= 16*s->mb_height - 2*(h->sps.crop_top + h->sps.crop_bottom);
3846 s->height= 16*s->mb_height - 4*(h->sps.crop_top + h->sps.crop_bottom); //FIXME recheck
3848 if (s->context_initialized
3849 && ( s->width != s->avctx->width || s->height != s->avctx->height)) {
3853 if (!s->context_initialized) {
3854 if (MPV_common_init(s) < 0)
3859 s->avctx->width = s->width;
3860 s->avctx->height = s->height;
3861 s->avctx->sample_aspect_ratio= h->sps.sar;
3862 if(!s->avctx->sample_aspect_ratio.den)
3863 s->avctx->sample_aspect_ratio.den = 1;
3865 if(h->sps.timing_info_present_flag){
3866 s->avctx->time_base= (AVRational){h->sps.num_units_in_tick, h->sps.time_scale};
3870 if(h->slice_num == 0){
3874 s->current_picture_ptr->frame_num= //FIXME frame_num cleanup
3875 h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
3877 h->mb_aff_frame = 0;
3878 if(h->sps.frame_mbs_only_flag){
3879 s->picture_structure= PICT_FRAME;
3881 if(get_bits1(&s->gb)) { //field_pic_flag
3882 s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
3884 s->picture_structure= PICT_FRAME;
3885 first_mb_in_slice <<= 1;
3886 h->mb_aff_frame = h->sps.mb_aff;
3890 s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
3891 s->resync_mb_y = s->mb_y = first_mb_in_slice / s->mb_width;
3893 if(s->picture_structure==PICT_FRAME){
3894 h->curr_pic_num= h->frame_num;
3895 h->max_pic_num= 1<< h->sps.log2_max_frame_num;
3897 h->curr_pic_num= 2*h->frame_num;
3898 h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
3901 if(h->nal_unit_type == NAL_IDR_SLICE){
3902 get_ue_golomb(&s->gb); /* idr_pic_id */
3905 if(h->sps.poc_type==0){
3906 h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
3908 if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
3909 h->delta_poc_bottom= get_se_golomb(&s->gb);
3913 if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
3914 h->delta_poc[0]= get_se_golomb(&s->gb);
3916 if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
3917 h->delta_poc[1]= get_se_golomb(&s->gb);
3922 if(h->pps.redundant_pic_cnt_present){
3923 h->redundant_pic_count= get_ue_golomb(&s->gb);
3926 //set defaults, might be overriden a few line later
3927 h->ref_count[0]= h->pps.ref_count[0];
3928 h->ref_count[1]= h->pps.ref_count[1];
3930 if(h->slice_type == P_TYPE || h->slice_type == SP_TYPE || h->slice_type == B_TYPE){
3931 if(h->slice_type == B_TYPE){
3932 h->direct_spatial_mv_pred= get_bits1(&s->gb);
3934 num_ref_idx_active_override_flag= get_bits1(&s->gb);
3936 if(num_ref_idx_active_override_flag){
3937 h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
3938 if(h->slice_type==B_TYPE)
3939 h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
3941 if(h->ref_count[0] > 32 || h->ref_count[1] > 32){
3942 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
3948 if(!default_ref_list_done){
3949 fill_default_ref_list(h);
3952 decode_ref_pic_list_reordering(h);
3954 if( (h->pps.weighted_pred && (h->slice_type == P_TYPE || h->slice_type == SP_TYPE ))
3955 || (h->pps.weighted_bipred_idc==1 && h->slice_type==B_TYPE ) )
3956 pred_weight_table(h);
3957 else if(h->pps.weighted_bipred_idc==2 && h->slice_type==B_TYPE)
3958 implicit_weight_table(h);
3962 if(s->current_picture.reference)
3963 decode_ref_pic_marking(h);
3965 if( h->slice_type != I_TYPE && h->slice_type != SI_TYPE && h->pps.cabac )
3966 h->cabac_init_idc = get_ue_golomb(&s->gb);
3968 h->last_qscale_diff = 0;
3969 s->qscale = h->pps.init_qp + get_se_golomb(&s->gb);
3970 if(s->qscale<0 || s->qscale>51){
3971 av_log(s->avctx, AV_LOG_ERROR, "QP %d out of range\n", s->qscale);
3974 h->chroma_qp = get_chroma_qp(h->pps.chroma_qp_index_offset, s->qscale);
3975 //FIXME qscale / qp ... stuff
3976 if(h->slice_type == SP_TYPE){
3977 get_bits1(&s->gb); /* sp_for_switch_flag */
3979 if(h->slice_type==SP_TYPE || h->slice_type == SI_TYPE){
3980 get_se_golomb(&s->gb); /* slice_qs_delta */
3983 h->deblocking_filter = 1;
3984 h->slice_alpha_c0_offset = 0;
3985 h->slice_beta_offset = 0;
3986 if( h->pps.deblocking_filter_parameters_present ) {
3987 h->deblocking_filter= get_ue_golomb(&s->gb);
3988 if(h->deblocking_filter < 2)
3989 h->deblocking_filter^= 1; // 1<->0
3991 if( h->deblocking_filter ) {
3992 h->slice_alpha_c0_offset = get_se_golomb(&s->gb) << 1;
3993 h->slice_beta_offset = get_se_golomb(&s->gb) << 1;
3998 if( h->pps.num_slice_groups > 1 && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
3999 slice_group_change_cycle= get_bits(&s->gb, ?);
4004 if(s->avctx->debug&FF_DEBUG_PICT_INFO){
4005 av_log(h->s.avctx, AV_LOG_DEBUG, "slice:%d %s mb:%d %c pps:%d frame:%d poc:%d/%d ref:%d/%d qp:%d loop:%d:%d:%d weight:%d%s\n",
4007 (s->picture_structure==PICT_FRAME ? "F" : s->picture_structure==PICT_TOP_FIELD ? "T" : "B"),
4009 av_get_pict_type_char(h->slice_type),
4010 pps_id, h->frame_num,
4011 s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
4012 h->ref_count[0], h->ref_count[1],
4014 h->deblocking_filter, h->slice_alpha_c0_offset/2, h->slice_beta_offset/2,
4016 h->use_weight==1 && h->use_weight_chroma ? "c" : ""
4026 static inline int get_level_prefix(GetBitContext *gb){
4030 OPEN_READER(re, gb);
4031 UPDATE_CACHE(re, gb);
4032 buf=GET_CACHE(re, gb);
4034 log= 32 - av_log2(buf);
4036 print_bin(buf>>(32-log), log);
4037 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__);
4040 LAST_SKIP_BITS(re, gb, log);
4041 CLOSE_READER(re, gb);
4047 * decodes a residual block.
4048 * @param n block index
4049 * @param scantable scantable
4050 * @param max_coeff number of coefficients in the block
4051 * @return <0 if an error occured
4053 static int decode_residual(H264Context *h, GetBitContext *gb, DCTELEM *block, int n, const uint8_t *scantable, int qp, int max_coeff){
4054 MpegEncContext * const s = &h->s;
4055 const uint16_t *qmul= dequant_coeff[qp];
4056 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};
4057 int level[16], run[16];
4058 int suffix_length, zeros_left, coeff_num, coeff_token, total_coeff, i, trailing_ones;
4060 //FIXME put trailing_onex into the context
4062 if(n == CHROMA_DC_BLOCK_INDEX){
4063 coeff_token= get_vlc2(gb, chroma_dc_coeff_token_vlc.table, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 1);
4064 total_coeff= coeff_token>>2;
4066 if(n == LUMA_DC_BLOCK_INDEX){
4067 total_coeff= pred_non_zero_count(h, 0);
4068 coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
4069 total_coeff= coeff_token>>2;
4071 total_coeff= pred_non_zero_count(h, n);
4072 coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
4073 total_coeff= coeff_token>>2;
4074 h->non_zero_count_cache[ scan8[n] ]= total_coeff;
4078 //FIXME set last_non_zero?
4083 trailing_ones= coeff_token&3;
4084 tprintf("trailing:%d, total:%d\n", trailing_ones, total_coeff);
4085 assert(total_coeff<=16);
4087 for(i=0; i<trailing_ones; i++){
4088 level[i]= 1 - 2*get_bits1(gb);
4091 suffix_length= total_coeff > 10 && trailing_ones < 3;
4093 for(; i<total_coeff; i++){
4094 const int prefix= get_level_prefix(gb);
4095 int level_code, mask;
4097 if(prefix<14){ //FIXME try to build a large unified VLC table for all this
4099 level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
4101 level_code= (prefix<<suffix_length); //part
4102 }else if(prefix==14){
4104 level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
4106 level_code= prefix + get_bits(gb, 4); //part
4107 }else if(prefix==15){
4108 level_code= (prefix<<suffix_length) + get_bits(gb, 12); //part
4109 if(suffix_length==0) level_code+=15; //FIXME doesn't make (much)sense
4111 av_log(h->s.avctx, AV_LOG_ERROR, "prefix too large at %d %d\n", s->mb_x, s->mb_y);
4115 if(i==trailing_ones && i<3) level_code+= 2; //FIXME split first iteration
4117 mask= -(level_code&1);
4118 level[i]= (((2+level_code)>>1) ^ mask) - mask;
4120 if(suffix_length==0) suffix_length=1; //FIXME split first iteration
4123 if(ABS(level[i]) > (3<<(suffix_length-1)) && suffix_length<6) suffix_length++;
4125 if((2+level_code)>>1) > (3<<(suffix_length-1)) && suffix_length<6) suffix_length++;
4126 /* ? == prefix > 2 or sth */
4128 tprintf("level: %d suffix_length:%d\n", level[i], suffix_length);
4131 if(total_coeff == max_coeff)
4134 if(n == CHROMA_DC_BLOCK_INDEX)
4135 zeros_left= get_vlc2(gb, chroma_dc_total_zeros_vlc[ total_coeff-1 ].table, CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 1);
4137 zeros_left= get_vlc2(gb, total_zeros_vlc[ total_coeff-1 ].table, TOTAL_ZEROS_VLC_BITS, 1);
4140 for(i=0; i<total_coeff-1; i++){
4143 else if(zeros_left < 7){
4144 run[i]= get_vlc2(gb, run_vlc[zeros_left-1].table, RUN_VLC_BITS, 1);
4146 run[i]= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2);
4148 zeros_left -= run[i];
4152 av_log(h->s.avctx, AV_LOG_ERROR, "negative number of zero coeffs at %d %d\n", s->mb_x, s->mb_y);
4156 for(; i<total_coeff-1; i++){
4164 for(i=total_coeff-1; i>=0; i--){ //FIXME merge into rundecode?
4167 coeff_num += run[i] + 1; //FIXME add 1 earlier ?
4168 j= scantable[ coeff_num ];
4173 for(i=total_coeff-1; i>=0; i--){ //FIXME merge into rundecode?
4176 coeff_num += run[i] + 1; //FIXME add 1 earlier ?
4177 j= scantable[ coeff_num ];
4179 block[j]= level[i] * qmul[j];
4180 // printf("%d %d ", block[j], qmul[j]);
4187 * decodes a P_SKIP or B_SKIP macroblock
4189 static void decode_mb_skip(H264Context *h){
4190 MpegEncContext * const s = &h->s;
4191 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
4194 memset(h->non_zero_count[mb_xy], 0, 16);
4195 memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui
4197 if(h->mb_aff_frame && s->mb_skip_run==0 && (s->mb_y&1)==0){
4198 h->mb_field_decoding_flag= get_bits1(&s->gb);
4200 if(h->mb_field_decoding_flag)
4201 mb_type|= MB_TYPE_INTERLACED;
4203 if( h->slice_type == B_TYPE )
4205 // just for fill_caches. pred_direct_motion will set the real mb_type
4206 mb_type= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2|MB_TYPE_SKIP;
4208 fill_caches(h, mb_type, 0); //FIXME check what is needed and what not ...
4209 pred_direct_motion(h, &mb_type);
4211 fill_rectangle(h->mvd_cache[0][scan8[0]], 4, 4, 8, 0, 4);
4212 fill_rectangle(h->mvd_cache[1][scan8[0]], 4, 4, 8, 0, 4);
4218 mb_type= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0|MB_TYPE_SKIP;
4220 fill_caches(h, mb_type, 0); //FIXME check what is needed and what not ...
4221 pred_pskip_motion(h, &mx, &my);
4222 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
4223 fill_rectangle( h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4);
4225 fill_rectangle(h->mvd_cache[0][scan8[0]], 4, 4, 8, 0, 4);
4228 write_back_motion(h, mb_type);
4229 s->current_picture.mb_type[mb_xy]= mb_type|MB_TYPE_SKIP;
4230 s->current_picture.qscale_table[mb_xy]= s->qscale;
4231 h->slice_table[ mb_xy ]= h->slice_num;
4232 h->prev_mb_skipped= 1;
4236 * decodes a macroblock
4237 * @returns 0 if ok, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
4239 static int decode_mb_cavlc(H264Context *h){
4240 MpegEncContext * const s = &h->s;
4241 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
4242 int mb_type, partition_count, cbp;
4244 s->dsp.clear_blocks(h->mb); //FIXME avoid if already clear (move after skip handlong?
4246 tprintf("pic:%d mb:%d/%d\n", h->frame_num, s->mb_x, s->mb_y);
4247 cbp = 0; /* avoid warning. FIXME: find a solution without slowing
4249 if(h->slice_type != I_TYPE && h->slice_type != SI_TYPE){
4250 if(s->mb_skip_run==-1)
4251 s->mb_skip_run= get_ue_golomb(&s->gb);
4253 if (s->mb_skip_run--) {
4258 if(h->mb_aff_frame){
4259 if ( ((s->mb_y&1) == 0) || h->prev_mb_skipped)
4260 h->mb_field_decoding_flag = get_bits1(&s->gb);
4262 h->mb_field_decoding_flag= (s->picture_structure!=PICT_FRAME);
4264 h->prev_mb_skipped= 0;
4266 mb_type= get_ue_golomb(&s->gb);
4267 if(h->slice_type == B_TYPE){
4269 partition_count= b_mb_type_info[mb_type].partition_count;
4270 mb_type= b_mb_type_info[mb_type].type;
4273 goto decode_intra_mb;
4275 }else if(h->slice_type == P_TYPE /*|| h->slice_type == SP_TYPE */){
4277 partition_count= p_mb_type_info[mb_type].partition_count;
4278 mb_type= p_mb_type_info[mb_type].type;
4281 goto decode_intra_mb;
4284 assert(h->slice_type == I_TYPE);
4287 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);
4291 cbp= i_mb_type_info[mb_type].cbp;
4292 h->intra16x16_pred_mode= i_mb_type_info[mb_type].pred_mode;
4293 mb_type= i_mb_type_info[mb_type].type;
4296 if(h->mb_field_decoding_flag)
4297 mb_type |= MB_TYPE_INTERLACED;
4299 s->current_picture.mb_type[mb_xy]= mb_type;
4300 h->slice_table[ mb_xy ]= h->slice_num;
4302 if(IS_INTRA_PCM(mb_type)){
4305 // we assume these blocks are very rare so we dont optimize it
4306 align_get_bits(&s->gb);
4308 // The pixels are stored in the same order as levels in h->mb array.
4309 for(y=0; y<16; y++){
4310 const int index= 4*(y&3) + 32*((y>>2)&1) + 128*(y>>3);
4311 for(x=0; x<16; x++){
4312 tprintf("LUMA ICPM LEVEL (%3d)\n", show_bits(&s->gb, 8));
4313 h->mb[index + (x&3) + 16*((x>>2)&1) + 64*(x>>3)]= get_bits(&s->gb, 8);
4317 const int index= 256 + 4*(y&3) + 32*(y>>2);
4319 tprintf("CHROMA U ICPM LEVEL (%3d)\n", show_bits(&s->gb, 8));
4320 h->mb[index + (x&3) + 16*(x>>2)]= get_bits(&s->gb, 8);
4324 const int index= 256 + 64 + 4*(y&3) + 32*(y>>2);
4326 tprintf("CHROMA V ICPM LEVEL (%3d)\n", show_bits(&s->gb, 8));
4327 h->mb[index + (x&3) + 16*(x>>2)]= get_bits(&s->gb, 8);
4331 // In deblocking, the quantizer is 0
4332 s->current_picture.qscale_table[mb_xy]= 0;
4333 h->chroma_qp = get_chroma_qp(h->pps.chroma_qp_index_offset, 0);
4334 // All coeffs are present
4335 memset(h->non_zero_count[mb_xy], 16, 16);
4340 fill_caches(h, mb_type, 0);
4343 if(IS_INTRA(mb_type)){
4344 // init_top_left_availability(h);
4345 if(IS_INTRA4x4(mb_type)){
4348 // fill_intra4x4_pred_table(h);
4349 for(i=0; i<16; i++){
4350 const int mode_coded= !get_bits1(&s->gb);
4351 const int predicted_mode= pred_intra_mode(h, i);
4355 const int rem_mode= get_bits(&s->gb, 3);
4356 if(rem_mode<predicted_mode)
4361 mode= predicted_mode;
4364 h->intra4x4_pred_mode_cache[ scan8[i] ] = mode;
4366 write_back_intra_pred_mode(h);
4367 if( check_intra4x4_pred_mode(h) < 0)
4370 h->intra16x16_pred_mode= check_intra_pred_mode(h, h->intra16x16_pred_mode);
4371 if(h->intra16x16_pred_mode < 0)
4374 h->chroma_pred_mode= get_ue_golomb(&s->gb);
4376 h->chroma_pred_mode= check_intra_pred_mode(h, h->chroma_pred_mode);
4377 if(h->chroma_pred_mode < 0)
4379 }else if(partition_count==4){
4380 int i, j, sub_partition_count[4], list, ref[2][4];
4382 if(h->slice_type == B_TYPE){
4384 h->sub_mb_type[i]= get_ue_golomb(&s->gb);
4385 if(h->sub_mb_type[i] >=13){
4386 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);
4389 sub_partition_count[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
4390 h->sub_mb_type[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].type;
4392 if( IS_DIRECT(h->sub_mb_type[0]) || IS_DIRECT(h->sub_mb_type[1])
4393 || IS_DIRECT(h->sub_mb_type[2]) || IS_DIRECT(h->sub_mb_type[3]))
4394 pred_direct_motion(h, &mb_type);
4396 assert(h->slice_type == P_TYPE || h->slice_type == SP_TYPE); //FIXME SP correct ?
4398 h->sub_mb_type[i]= get_ue_golomb(&s->gb);
4399 if(h->sub_mb_type[i] >=4){
4400 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);
4403 sub_partition_count[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
4404 h->sub_mb_type[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].type;
4408 for(list=0; list<2; list++){
4409 int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list];
4410 if(ref_count == 0) continue;
4411 if (h->mb_aff_frame && h->mb_field_decoding_flag) {
4415 if(IS_DIRECT(h->sub_mb_type[i])) continue;
4416 if(IS_DIR(h->sub_mb_type[i], 0, list)){
4417 ref[list][i] = get_te0_golomb(&s->gb, ref_count); //FIXME init to 0 before and skip?
4425 for(list=0; list<2; list++){
4426 const int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list];
4427 if(ref_count == 0) continue;
4430 if(IS_DIRECT(h->sub_mb_type[i])) continue;
4431 h->ref_cache[list][ scan8[4*i] ]=h->ref_cache[list][ scan8[4*i]+1 ]=
4432 h->ref_cache[list][ scan8[4*i]+8 ]=h->ref_cache[list][ scan8[4*i]+9 ]= ref[list][i];
4434 if(IS_DIR(h->sub_mb_type[i], 0, list)){
4435 const int sub_mb_type= h->sub_mb_type[i];
4436 const int block_width= (sub_mb_type & (MB_TYPE_16x16|MB_TYPE_16x8)) ? 2 : 1;
4437 for(j=0; j<sub_partition_count[i]; j++){
4439 const int index= 4*i + block_width*j;
4440 int16_t (* mv_cache)[2]= &h->mv_cache[list][ scan8[index] ];
4441 pred_motion(h, index, block_width, list, h->ref_cache[list][ scan8[index] ], &mx, &my);
4442 mx += get_se_golomb(&s->gb);
4443 my += get_se_golomb(&s->gb);
4444 tprintf("final mv:%d %d\n", mx, my);
4446 if(IS_SUB_8X8(sub_mb_type)){
4447 mv_cache[ 0 ][0]= mv_cache[ 1 ][0]=
4448 mv_cache[ 8 ][0]= mv_cache[ 9 ][0]= mx;
4449 mv_cache[ 0 ][1]= mv_cache[ 1 ][1]=
4450 mv_cache[ 8 ][1]= mv_cache[ 9 ][1]= my;
4451 }else if(IS_SUB_8X4(sub_mb_type)){
4452 mv_cache[ 0 ][0]= mv_cache[ 1 ][0]= mx;
4453 mv_cache[ 0 ][1]= mv_cache[ 1 ][1]= my;
4454 }else if(IS_SUB_4X8(sub_mb_type)){
4455 mv_cache[ 0 ][0]= mv_cache[ 8 ][0]= mx;
4456 mv_cache[ 0 ][1]= mv_cache[ 8 ][1]= my;
4458 assert(IS_SUB_4X4(sub_mb_type));
4459 mv_cache[ 0 ][0]= mx;
4460 mv_cache[ 0 ][1]= my;
4464 uint32_t *p= (uint32_t *)&h->mv_cache[list][ scan8[4*i] ][0];
4470 }else if(IS_DIRECT(mb_type)){
4471 pred_direct_motion(h, &mb_type);
4472 s->current_picture.mb_type[mb_xy]= mb_type;
4474 int list, mx, my, i;
4475 //FIXME we should set ref_idx_l? to 0 if we use that later ...
4476 if(IS_16X16(mb_type)){
4477 for(list=0; list<2; list++){
4478 if(h->ref_count[list]>0){
4479 if(IS_DIR(mb_type, 0, list)){
4480 const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
4481 fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, val, 1);
4483 fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, (LIST_NOT_USED&0xFF), 1);
4486 for(list=0; list<2; list++){
4487 if(IS_DIR(mb_type, 0, list)){
4488 pred_motion(h, 0, 4, list, h->ref_cache[list][ scan8[0] ], &mx, &my);
4489 mx += get_se_golomb(&s->gb);
4490 my += get_se_golomb(&s->gb);
4491 tprintf("final mv:%d %d\n", mx, my);
4493 fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, pack16to32(mx,my), 4);
4495 fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, 0, 4);
4498 else if(IS_16X8(mb_type)){
4499 for(list=0; list<2; list++){
4500 if(h->ref_count[list]>0){
4502 if(IS_DIR(mb_type, i, list)){
4503 const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
4504 fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, val, 1);
4506 fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, (LIST_NOT_USED&0xFF), 1);
4510 for(list=0; list<2; list++){
4512 if(IS_DIR(mb_type, i, list)){
4513 pred_16x8_motion(h, 8*i, list, h->ref_cache[list][scan8[0] + 16*i], &mx, &my);
4514 mx += get_se_golomb(&s->gb);
4515 my += get_se_golomb(&s->gb);
4516 tprintf("final mv:%d %d\n", mx, my);
4518 fill_rectangle(h->mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, pack16to32(mx,my), 4);
4520 fill_rectangle(h->mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, 0, 4);
4524 assert(IS_8X16(mb_type));
4525 for(list=0; list<2; list++){
4526 if(h->ref_count[list]>0){
4528 if(IS_DIR(mb_type, i, list)){ //FIXME optimize
4529 const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
4530 fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, val, 1);
4532 fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, (LIST_NOT_USED&0xFF), 1);
4536 for(list=0; list<2; list++){
4538 if(IS_DIR(mb_type, i, list)){
4539 pred_8x16_motion(h, i*4, list, h->ref_cache[list][ scan8[0] + 2*i ], &mx, &my);
4540 mx += get_se_golomb(&s->gb);
4541 my += get_se_golomb(&s->gb);
4542 tprintf("final mv:%d %d\n", mx, my);
4544 fill_rectangle(h->mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, pack16to32(mx,my), 4);
4546 fill_rectangle(h->mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, 0, 4);
4552 if(IS_INTER(mb_type))
4553 write_back_motion(h, mb_type);
4555 if(!IS_INTRA16x16(mb_type)){
4556 cbp= get_ue_golomb(&s->gb);
4558 av_log(h->s.avctx, AV_LOG_ERROR, "cbp too large (%d) at %d %d\n", cbp, s->mb_x, s->mb_y);
4562 if(IS_INTRA4x4(mb_type))
4563 cbp= golomb_to_intra4x4_cbp[cbp];
4565 cbp= golomb_to_inter_cbp[cbp];
4568 if(cbp || IS_INTRA16x16(mb_type)){
4569 int i8x8, i4x4, chroma_idx;
4570 int chroma_qp, dquant;
4571 GetBitContext *gb= IS_INTRA(mb_type) ? h->intra_gb_ptr : h->inter_gb_ptr;
4572 const uint8_t *scan, *dc_scan;
4574 // fill_non_zero_count_cache(h);
4576 if(IS_INTERLACED(mb_type)){
4578 dc_scan= luma_dc_field_scan;
4581 dc_scan= luma_dc_zigzag_scan;
4584 dquant= get_se_golomb(&s->gb);
4586 if( dquant > 25 || dquant < -26 ){
4587 av_log(h->s.avctx, AV_LOG_ERROR, "dquant out of range (%d) at %d %d\n", dquant, s->mb_x, s->mb_y);
4591 s->qscale += dquant;
4592 if(((unsigned)s->qscale) > 51){
4593 if(s->qscale<0) s->qscale+= 52;
4594 else s->qscale-= 52;
4597 h->chroma_qp= chroma_qp= get_chroma_qp(h->pps.chroma_qp_index_offset, s->qscale);
4598 if(IS_INTRA16x16(mb_type)){
4599 if( decode_residual(h, h->intra_gb_ptr, h->mb, LUMA_DC_BLOCK_INDEX, dc_scan, s->qscale, 16) < 0){
4600 return -1; //FIXME continue if partitioned and other return -1 too
4603 assert((cbp&15) == 0 || (cbp&15) == 15);
4606 for(i8x8=0; i8x8<4; i8x8++){
4607 for(i4x4=0; i4x4<4; i4x4++){
4608 const int index= i4x4 + 4*i8x8;
4609 if( decode_residual(h, h->intra_gb_ptr, h->mb + 16*index, index, scan + 1, s->qscale, 15) < 0 ){
4615 fill_rectangle(&h->non_zero_count_cache[scan8[0]], 4, 4, 8, 0, 1);
4618 for(i8x8=0; i8x8<4; i8x8++){
4619 if(cbp & (1<<i8x8)){
4620 for(i4x4=0; i4x4<4; i4x4++){
4621 const int index= i4x4 + 4*i8x8;
4623 if( decode_residual(h, gb, h->mb + 16*index, index, scan, s->qscale, 16) <0 ){
4628 uint8_t * const nnz= &h->non_zero_count_cache[ scan8[4*i8x8] ];
4629 nnz[0] = nnz[1] = nnz[8] = nnz[9] = 0;
4635 for(chroma_idx=0; chroma_idx<2; chroma_idx++)
4636 if( decode_residual(h, gb, h->mb + 256 + 16*4*chroma_idx, CHROMA_DC_BLOCK_INDEX, chroma_dc_scan, chroma_qp, 4) < 0){
4642 for(chroma_idx=0; chroma_idx<2; chroma_idx++){
4643 for(i4x4=0; i4x4<4; i4x4++){
4644 const int index= 16 + 4*chroma_idx + i4x4;
4645 if( decode_residual(h, gb, h->mb + 16*index, index, scan + 1, chroma_qp, 15) < 0){
4651 uint8_t * const nnz= &h->non_zero_count_cache[0];
4652 nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =
4653 nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
4656 uint8_t * const nnz= &h->non_zero_count_cache[0];
4657 fill_rectangle(&nnz[scan8[0]], 4, 4, 8, 0, 1);
4658 nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =
4659 nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
4661 s->current_picture.qscale_table[mb_xy]= s->qscale;
4662 write_back_non_zero_count(h);
4667 static int decode_cabac_field_decoding_flag(H264Context *h) {
4668 MpegEncContext * const s = &h->s;
4669 const int mb_x = s->mb_x;
4670 const int mb_y = s->mb_y & ~1;
4671 const int mba_xy = mb_x - 1 + mb_y *s->mb_stride;
4672 const int mbb_xy = mb_x + (mb_y-2)*s->mb_stride;
4674 unsigned int ctx = 0;
4676 if( h->slice_table[mba_xy] == h->slice_num && IS_INTERLACED( s->current_picture.mb_type[mba_xy] ) ) {
4679 if( h->slice_table[mbb_xy] == h->slice_num && IS_INTERLACED( s->current_picture.mb_type[mbb_xy] ) ) {
4683 return get_cabac( &h->cabac, &h->cabac_state[70 + ctx] );
4686 static int decode_cabac_intra_mb_type(H264Context *h, int ctx_base, int intra_slice) {
4687 uint8_t *state= &h->cabac_state[ctx_base];
4691 MpegEncContext * const s = &h->s;
4692 const int mba_xy = h->left_mb_xy[0];
4693 const int mbb_xy = h->top_mb_xy;
4695 if( h->slice_table[mba_xy] == h->slice_num && !IS_INTRA4x4( s->current_picture.mb_type[mba_xy] ) )
4697 if( h->slice_table[mbb_xy] == h->slice_num && !IS_INTRA4x4( s->current_picture.mb_type[mbb_xy] ) )
4699 if( get_cabac( &h->cabac, &state[ctx] ) == 0 )
4700 return 0; /* I4x4 */
4703 if( get_cabac( &h->cabac, &state[0] ) == 0 )
4704 return 0; /* I4x4 */
4707 if( get_cabac_terminate( &h->cabac ) )
4708 return 25; /* PCM */
4710 mb_type = 1; /* I16x16 */
4711 if( get_cabac( &h->cabac, &state[1] ) )
4712 mb_type += 12; /* cbp_luma != 0 */
4714 if( get_cabac( &h->cabac, &state[2] ) ) {
4715 if( get_cabac( &h->cabac, &state[2+intra_slice] ) )
4716 mb_type += 4 * 2; /* cbp_chroma == 2 */
4718 mb_type += 4 * 1; /* cbp_chroma == 1 */
4720 if( get_cabac( &h->cabac, &state[3+intra_slice] ) )
4722 if( get_cabac( &h->cabac, &state[3+2*intra_slice] ) )
4727 static int decode_cabac_mb_type( H264Context *h ) {
4728 MpegEncContext * const s = &h->s;
4730 if( h->slice_type == I_TYPE ) {
4731 return decode_cabac_intra_mb_type(h, 3, 1);
4732 } else if( h->slice_type == P_TYPE ) {
4733 if( get_cabac( &h->cabac, &h->cabac_state[14] ) == 0 ) {
4735 if( get_cabac( &h->cabac, &h->cabac_state[15] ) == 0 ) {
4736 if( get_cabac( &h->cabac, &h->cabac_state[16] ) == 0 )
4737 return 0; /* P_L0_D16x16; */
4739 return 3; /* P_8x8; */
4741 if( get_cabac( &h->cabac, &h->cabac_state[17] ) == 0 )
4742 return 2; /* P_L0_D8x16; */
4744 return 1; /* P_L0_D16x8; */
4747 return decode_cabac_intra_mb_type(h, 17, 0) + 5;
4749 } else if( h->slice_type == B_TYPE ) {
4750 const int mba_xy = h->left_mb_xy[0];
4751 const int mbb_xy = h->top_mb_xy;
4755 if( h->slice_table[mba_xy] == h->slice_num && !IS_SKIP( s->current_picture.mb_type[mba_xy] )
4756 && !IS_DIRECT( s->current_picture.mb_type[mba_xy] ) )
4758 if( h->slice_table[mbb_xy] == h->slice_num && !IS_SKIP( s->current_picture.mb_type[mbb_xy] )
4759 && !IS_DIRECT( s->current_picture.mb_type[mbb_xy] ) )
4762 if( !get_cabac( &h->cabac, &h->cabac_state[27+ctx] ) )
4763 return 0; /* B_Direct_16x16 */
4765 if( !get_cabac( &h->cabac, &h->cabac_state[27+3] ) ) {
4766 return 1 + get_cabac( &h->cabac, &h->cabac_state[27+5] ); /* B_L[01]_16x16 */
4769 bits = get_cabac( &h->cabac, &h->cabac_state[27+4] ) << 3;
4770 bits|= get_cabac( &h->cabac, &h->cabac_state[27+5] ) << 2;
4771 bits|= get_cabac( &h->cabac, &h->cabac_state[27+5] ) << 1;
4772 bits|= get_cabac( &h->cabac, &h->cabac_state[27+5] );
4774 return bits + 3; /* B_Bi_16x16 through B_L1_L0_16x8 */
4775 else if( bits == 13 ) {
4776 return decode_cabac_intra_mb_type(h, 32, 0) + 23;
4777 } else if( bits == 14 )
4778 return 11; /* B_L1_L0_8x16 */
4779 else if( bits == 15 )
4780 return 22; /* B_8x8 */
4782 bits= ( bits<<1 ) | get_cabac( &h->cabac, &h->cabac_state[27+5] );
4783 return bits - 4; /* B_L0_Bi_* through B_Bi_Bi_* */
4785 /* TODO SI/SP frames? */
4790 static int decode_cabac_mb_skip( H264Context *h) {
4791 MpegEncContext * const s = &h->s;
4792 const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
4793 const int mba_xy = mb_xy - 1;
4794 const int mbb_xy = mb_xy - s->mb_stride;
4797 if( h->slice_table[mba_xy] == h->slice_num && !IS_SKIP( s->current_picture.mb_type[mba_xy] ))
4799 if( h->slice_table[mbb_xy] == h->slice_num && !IS_SKIP( s->current_picture.mb_type[mbb_xy] ))
4802 if( h->slice_type == P_TYPE || h->slice_type == SP_TYPE)
4803 return get_cabac( &h->cabac, &h->cabac_state[11+ctx] );
4805 return get_cabac( &h->cabac, &h->cabac_state[24+ctx] );
4808 static int decode_cabac_mb_intra4x4_pred_mode( H264Context *h, int pred_mode ) {
4811 if( get_cabac( &h->cabac, &h->cabac_state[68] ) )
4814 if( get_cabac( &h->cabac, &h->cabac_state[69] ) )
4816 if( get_cabac( &h->cabac, &h->cabac_state[69] ) )
4818 if( get_cabac( &h->cabac, &h->cabac_state[69] ) )
4820 if( mode >= pred_mode )
4826 static int decode_cabac_mb_chroma_pre_mode( H264Context *h) {
4827 const int mba_xy = h->left_mb_xy[0];
4828 const int mbb_xy = h->top_mb_xy;
4832 /* No need to test for IS_INTRA4x4 and IS_INTRA16x16, as we set chroma_pred_mode_table to 0 */
4833 if( h->slice_table[mba_xy] == h->slice_num && h->chroma_pred_mode_table[mba_xy] != 0 )
4836 if( h->slice_table[mbb_xy] == h->slice_num && h->chroma_pred_mode_table[mbb_xy] != 0 )
4839 if( get_cabac( &h->cabac, &h->cabac_state[64+ctx] ) == 0 )
4842 if( get_cabac( &h->cabac, &h->cabac_state[64+3] ) == 0 )
4844 if( get_cabac( &h->cabac, &h->cabac_state[64+3] ) == 0 )
4850 static const uint8_t block_idx_x[16] = {
4851 0, 1, 0, 1, 2, 3, 2, 3, 0, 1, 0, 1, 2, 3, 2, 3
4853 static const uint8_t block_idx_y[16] = {
4854 0, 0, 1, 1, 0, 0, 1, 1, 2, 2, 3, 3, 2, 2, 3, 3
4856 static const uint8_t block_idx_xy[4][4] = {
4863 static int decode_cabac_mb_cbp_luma( H264Context *h) {
4864 MpegEncContext * const s = &h->s;
4869 for( i8x8 = 0; i8x8 < 4; i8x8++ ) {
4875 x = block_idx_x[4*i8x8];
4876 y = block_idx_y[4*i8x8];
4880 else if( s->mb_x > 0 && (h->slice_table[h->left_mb_xy[0]] == h->slice_num)) {
4881 cbp_a = h->left_cbp;
4882 tprintf("cbp_a = left_cbp = %x\n", cbp_a);
4887 else if( s->mb_y > 0 && (h->slice_table[h->top_mb_xy] == h->slice_num)) {
4889 tprintf("cbp_b = top_cbp = %x\n", cbp_b);
4892 /* No need to test for skip as we put 0 for skip block */
4893 /* No need to test for IPCM as we put 1 for IPCM block */
4895 int i8x8a = block_idx_xy[(x-1)&0x03][y]/4;
4896 if( ((cbp_a >> i8x8a)&0x01) == 0 )
4901 int i8x8b = block_idx_xy[x][(y-1)&0x03]/4;
4902 if( ((cbp_b >> i8x8b)&0x01) == 0 )
4906 if( get_cabac( &h->cabac, &h->cabac_state[73 + ctx] ) ) {
4912 static int decode_cabac_mb_cbp_chroma( H264Context *h) {
4916 cbp_a = (h->left_cbp>>4)&0x03;
4917 cbp_b = (h-> top_cbp>>4)&0x03;
4920 if( cbp_a > 0 ) ctx++;
4921 if( cbp_b > 0 ) ctx += 2;
4922 if( get_cabac( &h->cabac, &h->cabac_state[77 + ctx] ) == 0 )
4926 if( cbp_a == 2 ) ctx++;
4927 if( cbp_b == 2 ) ctx += 2;
4928 return 1 + get_cabac( &h->cabac, &h->cabac_state[77 + ctx] );
4930 static int decode_cabac_mb_dqp( H264Context *h) {
4931 MpegEncContext * const s = &h->s;
4937 mbn_xy = s->mb_x + s->mb_y*s->mb_stride - 1;
4939 mbn_xy = s->mb_width - 1 + (s->mb_y-1)*s->mb_stride;
4941 if( h->last_qscale_diff != 0 && ( IS_INTRA16x16(s->current_picture.mb_type[mbn_xy] ) || (h->cbp_table[mbn_xy]&0x3f) ) )
4944 while( get_cabac( &h->cabac, &h->cabac_state[60 + ctx] ) ) {
4955 return -(val + 1)/2;
4957 static int decode_cabac_p_mb_sub_type( H264Context *h ) {
4958 if( get_cabac( &h->cabac, &h->cabac_state[21] ) )
4960 if( !get_cabac( &h->cabac, &h->cabac_state[22] ) )
4962 if( get_cabac( &h->cabac, &h->cabac_state[23] ) )
4966 static int decode_cabac_b_mb_sub_type( H264Context *h ) {
4968 if( !get_cabac( &h->cabac, &h->cabac_state[36] ) )
4969 return 0; /* B_Direct_8x8 */
4970 if( !get_cabac( &h->cabac, &h->cabac_state[37] ) )
4971 return 1 + get_cabac( &h->cabac, &h->cabac_state[39] ); /* B_L0_8x8, B_L1_8x8 */
4973 if( get_cabac( &h->cabac, &h->cabac_state[38] ) ) {
4974 if( get_cabac( &h->cabac, &h->cabac_state[39] ) )
4975 return 11 + get_cabac( &h->cabac, &h->cabac_state[39] ); /* B_L1_4x4, B_Bi_4x4 */
4978 type += 2*get_cabac( &h->cabac, &h->cabac_state[39] );
4979 type += get_cabac( &h->cabac, &h->cabac_state[39] );
4983 static int decode_cabac_mb_ref( H264Context *h, int list, int n ) {
4984 int refa = h->ref_cache[list][scan8[n] - 1];
4985 int refb = h->ref_cache[list][scan8[n] - 8];
4989 if( h->slice_type == B_TYPE) {
4990 if( refa > 0 && !h->direct_cache[scan8[n] - 1] )
4992 if( refb > 0 && !h->direct_cache[scan8[n] - 8] )
5001 while( get_cabac( &h->cabac, &h->cabac_state[54+ctx] ) ) {
5011 static int decode_cabac_mb_mvd( H264Context *h, int list, int n, int l ) {
5012 int amvd = abs( h->mvd_cache[list][scan8[n] - 1][l] ) +
5013 abs( h->mvd_cache[list][scan8[n] - 8][l] );
5014 int ctxbase = (l == 0) ? 40 : 47;
5019 else if( amvd > 32 )
5024 if(!get_cabac(&h->cabac, &h->cabac_state[ctxbase+ctx]))
5029 while( mvd < 9 && get_cabac( &h->cabac, &h->cabac_state[ctxbase+ctx] ) ) {
5037 while( get_cabac_bypass( &h->cabac ) ) {
5042 if( get_cabac_bypass( &h->cabac ) )
5046 if( get_cabac_bypass( &h->cabac ) ) return -mvd;
5050 static int inline get_cabac_cbf_ctx( H264Context *h, int cat, int idx ) {
5055 nza = h->left_cbp&0x100;
5056 nzb = h-> top_cbp&0x100;
5057 } else if( cat == 1 || cat == 2 ) {
5058 nza = h->non_zero_count_cache[scan8[idx] - 1];
5059 nzb = h->non_zero_count_cache[scan8[idx] - 8];
5060 } else if( cat == 3 ) {
5061 nza = (h->left_cbp>>(6+idx))&0x01;
5062 nzb = (h-> top_cbp>>(6+idx))&0x01;
5065 nza = h->non_zero_count_cache[scan8[16+idx] - 1];
5066 nzb = h->non_zero_count_cache[scan8[16+idx] - 8];
5075 return ctx + 4 * cat;
5078 static int inline decode_cabac_residual( H264Context *h, DCTELEM *block, int cat, int n, const uint8_t *scantable, int qp, int max_coeff) {
5079 const int mb_xy = h->s.mb_x + h->s.mb_y*h->s.mb_stride;
5080 const uint16_t *qmul= dequant_coeff[qp];
5081 static const int significant_coeff_flag_field_offset[2] = { 105, 277 };
5082 static const int last_significant_coeff_flag_field_offset[2] = { 166, 338 };
5083 static const int significant_coeff_flag_offset[5] = { 0, 15, 29, 44, 47 };
5084 static const int coeff_abs_level_m1_offset[5] = {227+ 0, 227+10, 227+20, 227+30, 227+39 };
5089 int coeff_count = 0;
5092 int abslevelgt1 = 0;
5094 /* cat: 0-> DC 16x16 n = 0
5095 * 1-> AC 16x16 n = luma4x4idx
5096 * 2-> Luma4x4 n = luma4x4idx
5097 * 3-> DC Chroma n = iCbCr
5098 * 4-> AC Chroma n = 4 * iCbCr + chroma4x4idx
5101 /* read coded block flag */
5102 if( get_cabac( &h->cabac, &h->cabac_state[85 + get_cabac_cbf_ctx( h, cat, n ) ] ) == 0 ) {
5103 if( cat == 1 || cat == 2 )
5104 h->non_zero_count_cache[scan8[n]] = 0;
5106 h->non_zero_count_cache[scan8[16+n]] = 0;
5111 for(last= 0; last < max_coeff - 1; last++) {
5112 if( get_cabac( &h->cabac, &h->cabac_state[significant_coeff_flag_field_offset[h->mb_field_decoding_flag]+significant_coeff_flag_offset[cat]+last] )) {
5113 index[coeff_count++] = last;
5114 if( get_cabac( &h->cabac, &h->cabac_state[last_significant_coeff_flag_field_offset[h->mb_field_decoding_flag]+significant_coeff_flag_offset[cat]+last] ) ) {
5120 if( last == max_coeff -1 ) {
5121 index[coeff_count++] = last;
5123 assert(coeff_count > 0);
5126 h->cbp_table[mb_xy] |= 0x100;
5127 else if( cat == 1 || cat == 2 )
5128 h->non_zero_count_cache[scan8[n]] = coeff_count;
5130 h->cbp_table[mb_xy] |= 0x40 << n;
5133 h->non_zero_count_cache[scan8[16+n]] = coeff_count;
5136 for( i = coeff_count - 1; i >= 0; i-- ) {
5137 int ctx = (abslevelgt1 != 0 ? 0 : FFMIN( 4, abslevel1 )) + coeff_abs_level_m1_offset[cat];
5138 int j= scantable[index[i]];
5140 if( get_cabac( &h->cabac, &h->cabac_state[ctx] ) == 0 ) {
5141 if( cat == 0 || cat == 3 ) {
5142 if( get_cabac_bypass( &h->cabac ) ) block[j] = -1;
5145 if( get_cabac_bypass( &h->cabac ) ) block[j] = -qmul[j];
5146 else block[j] = qmul[j];
5152 ctx = 5 + FFMIN( 4, abslevelgt1 ) + coeff_abs_level_m1_offset[cat];
5153 while( coeff_abs < 15 && get_cabac( &h->cabac, &h->cabac_state[ctx] ) ) {
5157 if( coeff_abs >= 15 ) {
5159 while( get_cabac_bypass( &h->cabac ) ) {
5160 coeff_abs += 1 << j;
5165 if( get_cabac_bypass( &h->cabac ) )
5166 coeff_abs += 1 << j ;
5170 if( cat == 0 || cat == 3 ) {
5171 if( get_cabac_bypass( &h->cabac ) ) block[j] = -coeff_abs;
5172 else block[j] = coeff_abs;
5174 if( get_cabac_bypass( &h->cabac ) ) block[j] = -coeff_abs * qmul[j];
5175 else block[j] = coeff_abs * qmul[j];
5184 void inline compute_mb_neighboors(H264Context *h)
5186 MpegEncContext * const s = &h->s;
5187 const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
5188 h->top_mb_xy = mb_xy - s->mb_stride;
5189 h->left_mb_xy[0] = mb_xy - 1;
5190 if(h->mb_aff_frame){
5191 const int pair_xy = s->mb_x + (s->mb_y & ~1)*s->mb_stride;
5192 const int top_pair_xy = pair_xy - s->mb_stride;
5193 const int top_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[top_pair_xy]);
5194 const int left_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[pair_xy-1]);
5195 const int curr_mb_frame_flag = !h->mb_field_decoding_flag;
5196 const int bottom = (s->mb_y & 1);
5198 ? !curr_mb_frame_flag // bottom macroblock
5199 : (!curr_mb_frame_flag && !top_mb_frame_flag) // top macroblock
5201 h->top_mb_xy -= s->mb_stride;
5203 if (left_mb_frame_flag != curr_mb_frame_flag) {
5204 h->left_mb_xy[0] = pair_xy - 1;
5211 * decodes a macroblock
5212 * @returns 0 if ok, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
5214 static int decode_mb_cabac(H264Context *h) {
5215 MpegEncContext * const s = &h->s;
5216 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
5217 int mb_type, partition_count, cbp = 0;
5219 s->dsp.clear_blocks(h->mb); //FIXME avoid if already clear (move after skip handlong?)
5221 tprintf("pic:%d mb:%d/%d\n", h->frame_num, s->mb_x, s->mb_y);
5222 if( h->slice_type != I_TYPE && h->slice_type != SI_TYPE ) {
5223 /* read skip flags */
5224 if( decode_cabac_mb_skip( h ) ) {
5227 h->cbp_table[mb_xy] = 0;
5228 h->chroma_pred_mode_table[mb_xy] = 0;
5229 h->last_qscale_diff = 0;
5235 if(h->mb_aff_frame){
5236 if ( ((s->mb_y&1) == 0) || h->prev_mb_skipped)
5237 h->mb_field_decoding_flag = decode_cabac_field_decoding_flag(h);
5239 h->mb_field_decoding_flag= (s->picture_structure!=PICT_FRAME);
5241 h->prev_mb_skipped = 0;
5243 compute_mb_neighboors(h);
5244 if( ( mb_type = decode_cabac_mb_type( h ) ) < 0 ) {
5245 av_log( h->s.avctx, AV_LOG_ERROR, "decode_cabac_mb_type failed\n" );
5249 if( h->slice_type == B_TYPE ) {
5251 partition_count= b_mb_type_info[mb_type].partition_count;
5252 mb_type= b_mb_type_info[mb_type].type;
5255 goto decode_intra_mb;
5257 } else if( h->slice_type == P_TYPE ) {
5259 partition_count= p_mb_type_info[mb_type].partition_count;
5260 mb_type= p_mb_type_info[mb_type].type;
5263 goto decode_intra_mb;
5266 assert(h->slice_type == I_TYPE);
5268 partition_count = 0;
5269 cbp= i_mb_type_info[mb_type].cbp;
5270 h->intra16x16_pred_mode= i_mb_type_info[mb_type].pred_mode;
5271 mb_type= i_mb_type_info[mb_type].type;
5273 if(h->mb_field_decoding_flag)
5274 mb_type |= MB_TYPE_INTERLACED;
5276 s->current_picture.mb_type[mb_xy]= mb_type;
5277 h->slice_table[ mb_xy ]= h->slice_num;
5279 if(IS_INTRA_PCM(mb_type)) {
5283 // We assume these blocks are very rare so we dont optimize it.
5284 // FIXME The two following lines get the bitstream position in the cabac
5285 // decode, I think it should be done by a function in cabac.h (or cabac.c).
5286 ptr= h->cabac.bytestream;
5287 if (h->cabac.low&0x1) ptr-=CABAC_BITS/8;
5289 // The pixels are stored in the same order as levels in h->mb array.
5290 for(y=0; y<16; y++){
5291 const int index= 4*(y&3) + 32*((y>>2)&1) + 128*(y>>3);
5292 for(x=0; x<16; x++){
5293 tprintf("LUMA ICPM LEVEL (%3d)\n", *ptr);
5294 h->mb[index + (x&3) + 16*((x>>2)&1) + 64*(x>>3)]= *ptr++;
5298 const int index= 256 + 4*(y&3) + 32*(y>>2);
5300 tprintf("CHROMA U ICPM LEVEL (%3d)\n", *ptr);
5301 h->mb[index + (x&3) + 16*(x>>2)]= *ptr++;
5305 const int index= 256 + 64 + 4*(y&3) + 32*(y>>2);
5307 tprintf("CHROMA V ICPM LEVEL (%3d)\n", *ptr);
5308 h->mb[index + (x&3) + 16*(x>>2)]= *ptr++;
5312 ff_init_cabac_decoder(&h->cabac, ptr, h->cabac.bytestream_end - ptr);
5314 // All blocks are present
5315 h->cbp_table[mb_xy] = 0x1ef;
5316 h->chroma_pred_mode_table[mb_xy] = 0;
5317 // In deblocking, the quantizer is 0
5318 s->current_picture.qscale_table[mb_xy]= 0;
5319 h->chroma_qp = get_chroma_qp(h->pps.chroma_qp_index_offset, 0);
5320 // All coeffs are present
5321 memset(h->non_zero_count[mb_xy], 16, 16);
5325 fill_caches(h, mb_type, 0);
5327 if( IS_INTRA( mb_type ) ) {
5328 if( IS_INTRA4x4( mb_type ) ) {
5330 for( i = 0; i < 16; i++ ) {
5331 int pred = pred_intra_mode( h, i );
5332 h->intra4x4_pred_mode_cache[ scan8[i] ] = decode_cabac_mb_intra4x4_pred_mode( h, pred );
5334 //av_log( s->avctx, AV_LOG_ERROR, "i4x4 pred=%d mode=%d\n", pred, h->intra4x4_pred_mode_cache[ scan8[i] ] );
5336 write_back_intra_pred_mode(h);
5337 if( check_intra4x4_pred_mode(h) < 0 ) return -1;
5339 h->intra16x16_pred_mode= check_intra_pred_mode( h, h->intra16x16_pred_mode );
5340 if( h->intra16x16_pred_mode < 0 ) return -1;
5342 h->chroma_pred_mode_table[mb_xy] =
5343 h->chroma_pred_mode = decode_cabac_mb_chroma_pre_mode( h );
5345 h->chroma_pred_mode= check_intra_pred_mode( h, h->chroma_pred_mode );
5346 if( h->chroma_pred_mode < 0 ) return -1;
5347 } else if( partition_count == 4 ) {
5348 int i, j, sub_partition_count[4], list, ref[2][4];
5350 if( h->slice_type == B_TYPE ) {
5351 for( i = 0; i < 4; i++ ) {
5352 h->sub_mb_type[i] = decode_cabac_b_mb_sub_type( h );
5353 sub_partition_count[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
5354 h->sub_mb_type[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].type;
5356 if( IS_DIRECT(h->sub_mb_type[0]) || IS_DIRECT(h->sub_mb_type[1])
5357 || IS_DIRECT(h->sub_mb_type[2]) || IS_DIRECT(h->sub_mb_type[3])) {
5358 pred_direct_motion(h, &mb_type);
5359 if( h->ref_count[0] > 1 || h->ref_count[1] > 1 ) {
5360 for( i = 0; i < 4; i++ )
5361 if( IS_DIRECT(h->sub_mb_type[i]) )
5362 fill_rectangle( &h->direct_cache[scan8[4*i]], 2, 2, 8, 1, 1 );
5366 for( i = 0; i < 4; i++ ) {
5367 h->sub_mb_type[i] = decode_cabac_p_mb_sub_type( h );
5368 sub_partition_count[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
5369 h->sub_mb_type[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].type;
5373 for( list = 0; list < 2; list++ ) {
5374 if( h->ref_count[list] > 0 ) {
5375 for( i = 0; i < 4; i++ ) {
5376 if(IS_DIRECT(h->sub_mb_type[i])) continue;
5377 if(IS_DIR(h->sub_mb_type[i], 0, list)){
5378 if( h->ref_count[list] > 1 )
5379 ref[list][i] = decode_cabac_mb_ref( h, list, 4*i );
5385 h->ref_cache[list][ scan8[4*i]+1 ]=
5386 h->ref_cache[list][ scan8[4*i]+8 ]=h->ref_cache[list][ scan8[4*i]+9 ]= ref[list][i];
5391 for(list=0; list<2; list++){
5393 if(IS_DIRECT(h->sub_mb_type[i])){
5394 fill_rectangle(h->mvd_cache[list][scan8[4*i]], 2, 2, 8, 0, 4);
5397 h->ref_cache[list][ scan8[4*i] ]=h->ref_cache[list][ scan8[4*i]+1 ];
5399 if(IS_DIR(h->sub_mb_type[i], 0, list) && !IS_DIRECT(h->sub_mb_type[i])){
5400 const int sub_mb_type= h->sub_mb_type[i];
5401 const int block_width= (sub_mb_type & (MB_TYPE_16x16|MB_TYPE_16x8)) ? 2 : 1;
5402 for(j=0; j<sub_partition_count[i]; j++){
5405 const int index= 4*i + block_width*j;
5406 int16_t (* mv_cache)[2]= &h->mv_cache[list][ scan8[index] ];
5407 int16_t (* mvd_cache)[2]= &h->mvd_cache[list][ scan8[index] ];
5408 pred_motion(h, index, block_width, list, h->ref_cache[list][ scan8[index] ], &mpx, &mpy);
5410 mx = mpx + decode_cabac_mb_mvd( h, list, index, 0 );
5411 my = mpy + decode_cabac_mb_mvd( h, list, index, 1 );
5412 tprintf("final mv:%d %d\n", mx, my);
5414 if(IS_SUB_8X8(sub_mb_type)){
5415 mv_cache[ 0 ][0]= mv_cache[ 1 ][0]=
5416 mv_cache[ 8 ][0]= mv_cache[ 9 ][0]= mx;
5417 mv_cache[ 0 ][1]= mv_cache[ 1 ][1]=
5418 mv_cache[ 8 ][1]= mv_cache[ 9 ][1]= my;
5420 mvd_cache[ 0 ][0]= mvd_cache[ 1 ][0]=
5421 mvd_cache[ 8 ][0]= mvd_cache[ 9 ][0]= mx - mpx;
5422 mvd_cache[ 0 ][1]= mvd_cache[ 1 ][1]=
5423 mvd_cache[ 8 ][1]= mvd_cache[ 9 ][1]= my - mpy;
5424 }else if(IS_SUB_8X4(sub_mb_type)){
5425 mv_cache[ 0 ][0]= mv_cache[ 1 ][0]= mx;
5426 mv_cache[ 0 ][1]= mv_cache[ 1 ][1]= my;
5428 mvd_cache[ 0 ][0]= mvd_cache[ 1 ][0]= mx- mpx;
5429 mvd_cache[ 0 ][1]= mvd_cache[ 1 ][1]= my - mpy;
5430 }else if(IS_SUB_4X8(sub_mb_type)){
5431 mv_cache[ 0 ][0]= mv_cache[ 8 ][0]= mx;
5432 mv_cache[ 0 ][1]= mv_cache[ 8 ][1]= my;
5434 mvd_cache[ 0 ][0]= mvd_cache[ 8 ][0]= mx - mpx;
5435 mvd_cache[ 0 ][1]= mvd_cache[ 8 ][1]= my - mpy;
5437 assert(IS_SUB_4X4(sub_mb_type));
5438 mv_cache[ 0 ][0]= mx;
5439 mv_cache[ 0 ][1]= my;
5441 mvd_cache[ 0 ][0]= mx - mpx;
5442 mvd_cache[ 0 ][1]= my - mpy;
5446 uint32_t *p= (uint32_t *)&h->mv_cache[list][ scan8[4*i] ][0];
5447 uint32_t *pd= (uint32_t *)&h->mvd_cache[list][ scan8[4*i] ][0];
5448 p[0] = p[1] = p[8] = p[9] = 0;
5449 pd[0]= pd[1]= pd[8]= pd[9]= 0;
5453 } else if( IS_DIRECT(mb_type) ) {
5454 pred_direct_motion(h, &mb_type);
5455 s->current_picture.mb_type[mb_xy]= mb_type;
5456 fill_rectangle(h->mvd_cache[0][scan8[0]], 4, 4, 8, 0, 4);
5457 fill_rectangle(h->mvd_cache[1][scan8[0]], 4, 4, 8, 0, 4);
5459 int list, mx, my, i, mpx, mpy;
5460 if(IS_16X16(mb_type)){
5461 for(list=0; list<2; list++){
5462 if(IS_DIR(mb_type, 0, list)){
5463 if(h->ref_count[list] > 0 ){
5464 const int ref = h->ref_count[list] > 1 ? decode_cabac_mb_ref( h, list, 0 ) : 0;
5465 fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, ref, 1);
5468 fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, (uint8_t)LIST_NOT_USED, 1);
5470 for(list=0; list<2; list++){
5471 if(IS_DIR(mb_type, 0, list)){
5472 pred_motion(h, 0, 4, list, h->ref_cache[list][ scan8[0] ], &mpx, &mpy);
5474 mx = mpx + decode_cabac_mb_mvd( h, list, 0, 0 );
5475 my = mpy + decode_cabac_mb_mvd( h, list, 0, 1 );
5476 tprintf("final mv:%d %d\n", mx, my);
5478 fill_rectangle(h->mvd_cache[list][ scan8[0] ], 4, 4, 8, pack16to32(mx-mpx,my-mpy), 4);
5479 fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, pack16to32(mx,my), 4);
5481 fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, 0, 4);
5484 else if(IS_16X8(mb_type)){
5485 for(list=0; list<2; list++){
5486 if(h->ref_count[list]>0){
5488 if(IS_DIR(mb_type, i, list)){
5489 const int ref= h->ref_count[list] > 1 ? decode_cabac_mb_ref( h, list, 8*i ) : 0;
5490 fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, ref, 1);
5492 fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, (LIST_NOT_USED&0xFF), 1);
5496 for(list=0; list<2; list++){
5498 if(IS_DIR(mb_type, i, list)){
5499 pred_16x8_motion(h, 8*i, list, h->ref_cache[list][scan8[0] + 16*i], &mpx, &mpy);
5500 mx = mpx + decode_cabac_mb_mvd( h, list, 8*i, 0 );
5501 my = mpy + decode_cabac_mb_mvd( h, list, 8*i, 1 );
5502 tprintf("final mv:%d %d\n", mx, my);
5504 fill_rectangle(h->mvd_cache[list][ scan8[0] + 16*i ], 4, 2, 8, pack16to32(mx-mpx,my-mpy), 4);
5505 fill_rectangle(h->mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, pack16to32(mx,my), 4);
5507 fill_rectangle(h->mvd_cache[list][ scan8[0] + 16*i ], 4, 2, 8, 0, 4);
5508 fill_rectangle(h-> mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, 0, 4);
5513 assert(IS_8X16(mb_type));
5514 for(list=0; list<2; list++){
5515 if(h->ref_count[list]>0){
5517 if(IS_DIR(mb_type, i, list)){ //FIXME optimize
5518 const int ref= h->ref_count[list] > 1 ? decode_cabac_mb_ref( h, list, 4*i ) : 0;
5519 fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, ref, 1);
5521 fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, (LIST_NOT_USED&0xFF), 1);
5525 for(list=0; list<2; list++){
5527 if(IS_DIR(mb_type, i, list)){
5528 pred_8x16_motion(h, i*4, list, h->ref_cache[list][ scan8[0] + 2*i ], &mpx, &mpy);
5529 mx = mpx + decode_cabac_mb_mvd( h, list, 4*i, 0 );
5530 my = mpy + decode_cabac_mb_mvd( h, list, 4*i, 1 );
5532 tprintf("final mv:%d %d\n", mx, my);
5533 fill_rectangle(h->mvd_cache[list][ scan8[0] + 2*i ], 2, 4, 8, pack16to32(mx-mpx,my-mpy), 4);
5534 fill_rectangle(h->mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, pack16to32(mx,my), 4);
5536 fill_rectangle(h->mvd_cache[list][ scan8[0] + 2*i ], 2, 4, 8, 0, 4);
5537 fill_rectangle(h-> mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, 0, 4);
5544 if( IS_INTER( mb_type ) ) {
5545 h->chroma_pred_mode_table[mb_xy] = 0;
5546 write_back_motion( h, mb_type );
5549 if( !IS_INTRA16x16( mb_type ) ) {
5550 cbp = decode_cabac_mb_cbp_luma( h );
5551 cbp |= decode_cabac_mb_cbp_chroma( h ) << 4;
5554 h->cbp_table[mb_xy] = cbp;
5556 if( cbp || IS_INTRA16x16( mb_type ) ) {
5557 const uint8_t *scan, *dc_scan;
5560 if(IS_INTERLACED(mb_type)){
5562 dc_scan= luma_dc_field_scan;
5565 dc_scan= luma_dc_zigzag_scan;
5568 h->last_qscale_diff = dqp = decode_cabac_mb_dqp( h );
5570 if(((unsigned)s->qscale) > 51){
5571 if(s->qscale<0) s->qscale+= 52;
5572 else s->qscale-= 52;
5574 h->chroma_qp = get_chroma_qp(h->pps.chroma_qp_index_offset, s->qscale);
5576 if( IS_INTRA16x16( mb_type ) ) {
5578 //av_log( s->avctx, AV_LOG_ERROR, "INTRA16x16 DC\n" );
5579 if( decode_cabac_residual( h, h->mb, 0, 0, dc_scan, s->qscale, 16) < 0)
5582 for( i = 0; i < 16; i++ ) {
5583 //av_log( s->avctx, AV_LOG_ERROR, "INTRA16x16 AC:%d\n", i );
5584 if( decode_cabac_residual(h, h->mb + 16*i, 1, i, scan + 1, s->qscale, 15) < 0 )
5588 fill_rectangle(&h->non_zero_count_cache[scan8[0]], 4, 4, 8, 0, 1);
5592 for( i8x8 = 0; i8x8 < 4; i8x8++ ) {
5593 if( cbp & (1<<i8x8) ) {
5594 for( i4x4 = 0; i4x4 < 4; i4x4++ ) {
5595 const int index = 4*i8x8 + i4x4;
5596 //av_log( s->avctx, AV_LOG_ERROR, "Luma4x4: %d\n", index );
5597 if( decode_cabac_residual(h, h->mb + 16*index, 2, index, scan, s->qscale, 16) < 0 )
5601 uint8_t * const nnz= &h->non_zero_count_cache[ scan8[4*i8x8] ];
5602 nnz[0] = nnz[1] = nnz[8] = nnz[9] = 0;
5609 for( c = 0; c < 2; c++ ) {
5610 //av_log( s->avctx, AV_LOG_ERROR, "INTRA C%d-DC\n",c );
5611 if( decode_cabac_residual(h, h->mb + 256 + 16*4*c, 3, c, chroma_dc_scan, h->chroma_qp, 4) < 0)
5618 for( c = 0; c < 2; c++ ) {
5619 for( i = 0; i < 4; i++ ) {
5620 const int index = 16 + 4 * c + i;
5621 //av_log( s->avctx, AV_LOG_ERROR, "INTRA C%d-AC %d\n",c, index - 16 );
5622 if( decode_cabac_residual(h, h->mb + 16*index, 4, index - 16, scan + 1, h->chroma_qp, 15) < 0)
5627 uint8_t * const nnz= &h->non_zero_count_cache[0];
5628 nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =
5629 nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
5632 uint8_t * const nnz= &h->non_zero_count_cache[0];
5633 fill_rectangle(&nnz[scan8[0]], 4, 4, 8, 0, 1);
5634 nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =
5635 nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
5638 s->current_picture.qscale_table[mb_xy]= s->qscale;
5639 write_back_non_zero_count(h);
5645 static void filter_mb_edgev( H264Context *h, uint8_t *pix, int stride, int bS[4], int qp ) {
5647 const int index_a = clip( qp + h->slice_alpha_c0_offset, 0, 51 );
5648 const int alpha = alpha_table[index_a];
5649 const int beta = beta_table[clip( qp + h->slice_beta_offset, 0, 51 )];
5654 tc[i] = bS[i] ? tc0_table[index_a][bS[i] - 1] : -1;
5655 h->s.dsp.h264_h_loop_filter_luma(pix, stride, alpha, beta, tc);
5657 /* 16px edge length, because bS=4 is triggered by being at
5658 * the edge of an intra MB, so all 4 bS are the same */
5659 for( d = 0; d < 16; d++ ) {
5660 const int p0 = pix[-1];
5661 const int p1 = pix[-2];
5662 const int p2 = pix[-3];
5664 const int q0 = pix[0];
5665 const int q1 = pix[1];
5666 const int q2 = pix[2];
5668 if( ABS( p0 - q0 ) < alpha &&
5669 ABS( p1 - p0 ) < beta &&
5670 ABS( q1 - q0 ) < beta ) {
5672 if(ABS( p0 - q0 ) < (( alpha >> 2 ) + 2 )){
5673 if( ABS( p2 - p0 ) < beta)
5675 const int p3 = pix[-4];
5677 pix[-1] = ( p2 + 2*p1 + 2*p0 + 2*q0 + q1 + 4 ) >> 3;
5678 pix[-2] = ( p2 + p1 + p0 + q0 + 2 ) >> 2;
5679 pix[-3] = ( 2*p3 + 3*p2 + p1 + p0 + q0 + 4 ) >> 3;
5682 pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
5684 if( ABS( q2 - q0 ) < beta)
5686 const int q3 = pix[3];
5688 pix[0] = ( p1 + 2*p0 + 2*q0 + 2*q1 + q2 + 4 ) >> 3;
5689 pix[1] = ( p0 + q0 + q1 + q2 + 2 ) >> 2;
5690 pix[2] = ( 2*q3 + 3*q2 + q1 + q0 + p0 + 4 ) >> 3;
5693 pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
5697 pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
5698 pix[ 0] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
5700 tprintf("filter_mb_edgev i:%d d:%d\n# bS:4 -> [%02x, %02x, %02x, %02x, %02x, %02x] =>[%02x, %02x, %02x, %02x]\n", i, d, p2, p1, p0, q0, q1, q2, pix[-2], pix[-1], pix[0], pix[1]);
5706 static void filter_mb_edgecv( H264Context *h, uint8_t *pix, int stride, int bS[4], int qp ) {
5708 const int index_a = clip( qp + h->slice_alpha_c0_offset, 0, 51 );
5709 const int alpha = alpha_table[index_a];
5710 const int beta = beta_table[clip( qp + h->slice_beta_offset, 0, 51 )];
5715 tc[i] = bS[i] ? tc0_table[index_a][bS[i] - 1] + 1 : 0;
5716 h->s.dsp.h264_h_loop_filter_chroma(pix, stride, alpha, beta, tc);
5718 /* 8px edge length, see filter_mb_edgev */
5719 for( d = 0; d < 8; d++ ){
5720 const int p0 = pix[-1];
5721 const int p1 = pix[-2];
5722 const int q0 = pix[0];
5723 const int q1 = pix[1];
5725 if( ABS( p0 - q0 ) < alpha &&
5726 ABS( p1 - p0 ) < beta &&
5727 ABS( q1 - q0 ) < beta ) {
5729 pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2; /* p0' */
5730 pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2; /* q0' */
5731 tprintf("filter_mb_edgecv i:%d d:%d\n# bS:4 -> [%02x, %02x, %02x, %02x, %02x, %02x] =>[%02x, %02x, %02x, %02x]\n", i, d, pix[-3], p1, p0, q0, q1, pix[2], p1, pix[-1], pix[0], q1);
5738 static void filter_mb_mbaff_edgev( H264Context *h, uint8_t *pix, int stride, int bS[8], int qp[2] ) {
5740 for( i = 0; i < 16; i++, pix += stride) {
5746 int bS_index = (i >> 1);
5747 if (h->mb_field_decoding_flag) {
5749 bS_index |= (i & 1);
5752 if( bS[bS_index] == 0 ) {
5756 qp_index = h->mb_field_decoding_flag ? (i & 1) : (i >> 3);
5757 index_a = clip( qp[qp_index] + h->slice_alpha_c0_offset, 0, 51 );
5758 alpha = alpha_table[index_a];
5759 beta = beta_table[clip( qp[qp_index] + h->slice_beta_offset, 0, 51 )];
5762 if( bS[bS_index] < 4 ) {
5763 const int tc0 = tc0_table[index_a][bS[bS_index] - 1];
5764 /* 4px edge length */
5765 const int p0 = pix[-1];
5766 const int p1 = pix[-2];
5767 const int p2 = pix[-3];
5768 const int q0 = pix[0];
5769 const int q1 = pix[1];
5770 const int q2 = pix[2];
5772 if( ABS( p0 - q0 ) < alpha &&
5773 ABS( p1 - p0 ) < beta &&
5774 ABS( q1 - q0 ) < beta ) {
5778 if( ABS( p2 - p0 ) < beta ) {
5779 pix[-2] = p1 + clip( ( p2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( p1 << 1 ) ) >> 1, -tc0, tc0 );
5782 if( ABS( q2 - q0 ) < beta ) {
5783 pix[1] = q1 + clip( ( q2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( q1 << 1 ) ) >> 1, -tc0, tc0 );
5787 i_delta = clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );
5788 pix[-1] = clip_uint8( p0 + i_delta ); /* p0' */
5789 pix[0] = clip_uint8( q0 - i_delta ); /* q0' */
5790 tprintf("filter_mb_mbaff_edgev i:%d, qp:%d, indexA:%d, alpha:%d, beta:%d, tc:%d\n# bS:%d -> [%02x, %02x, %02x, %02x, %02x, %02x] =>[%02x, %02x, %02x, %02x]\n", i, qp[qp_index], index_a, alpha, beta, tc, bS[bS_index], pix[-3], p1, p0, q0, q1, pix[2], p1, pix[-1], pix[0], q1);
5793 /* 4px edge length */
5794 const int p0 = pix[-1];
5795 const int p1 = pix[-2];
5796 const int p2 = pix[-3];
5798 const int q0 = pix[0];
5799 const int q1 = pix[1];
5800 const int q2 = pix[2];
5802 if( ABS( p0 - q0 ) < alpha &&
5803 ABS( p1 - p0 ) < beta &&
5804 ABS( q1 - q0 ) < beta ) {
5806 if(ABS( p0 - q0 ) < (( alpha >> 2 ) + 2 )){
5807 if( ABS( p2 - p0 ) < beta)
5809 const int p3 = pix[-4];
5811 pix[-1] = ( p2 + 2*p1 + 2*p0 + 2*q0 + q1 + 4 ) >> 3;
5812 pix[-2] = ( p2 + p1 + p0 + q0 + 2 ) >> 2;
5813 pix[-3] = ( 2*p3 + 3*p2 + p1 + p0 + q0 + 4 ) >> 3;
5816 pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
5818 if( ABS( q2 - q0 ) < beta)
5820 const int q3 = pix[3];
5822 pix[0] = ( p1 + 2*p0 + 2*q0 + 2*q1 + q2 + 4 ) >> 3;
5823 pix[1] = ( p0 + q0 + q1 + q2 + 2 ) >> 2;
5824 pix[2] = ( 2*q3 + 3*q2 + q1 + q0 + p0 + 4 ) >> 3;
5827 pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
5831 pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
5832 pix[ 0] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
5834 tprintf("filter_mb_mbaff_edgev i:%d, qp:%d, indexA:%d, alpha:%d, beta:%d\n# bS:4 -> [%02x, %02x, %02x, %02x, %02x, %02x] =>[%02x, %02x, %02x, %02x, %02x, %02x]\n", i, qp[qp_index], index_a, alpha, beta, p2, p1, p0, q0, q1, q2, pix[-3], pix[-2], pix[-1], pix[0], pix[1], pix[2]);
5839 static void filter_mb_mbaff_edgecv( H264Context *h, uint8_t *pix, int stride, int bS[4], int qp[2] ) {
5841 for( i = 0; i < 8; i++, pix += stride) {
5849 if( bS[bS_index] == 0 ) {
5853 qp_index = h->mb_field_decoding_flag ? (i & 1) : (i >> 3);
5854 index_a = clip( qp[qp_index] + h->slice_alpha_c0_offset, 0, 51 );
5855 alpha = alpha_table[index_a];
5856 beta = beta_table[clip( qp[qp_index] + h->slice_beta_offset, 0, 51 )];
5857 if( bS[bS_index] < 4 ) {
5858 const int tc = tc0_table[index_a][bS[bS_index] - 1] + 1;
5859 /* 2px edge length (because we use same bS than the one for luma) */
5860 const int p0 = pix[-1];
5861 const int p1 = pix[-2];
5862 const int q0 = pix[0];
5863 const int q1 = pix[1];
5865 if( ABS( p0 - q0 ) < alpha &&
5866 ABS( p1 - p0 ) < beta &&
5867 ABS( q1 - q0 ) < beta ) {
5868 const int i_delta = clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );
5870 pix[-1] = clip_uint8( p0 + i_delta ); /* p0' */
5871 pix[0] = clip_uint8( q0 - i_delta ); /* q0' */
5872 tprintf("filter_mb_mbaff_edgecv i:%d, qp:%d, indexA:%d, alpha:%d, beta:%d, tc:%d\n# bS:%d -> [%02x, %02x, %02x, %02x, %02x, %02x] =>[%02x, %02x, %02x, %02x]\n", i, qp[qp_index], index_a, alpha, beta, tc, bS[bS_index], pix[-3], p1, p0, q0, q1, pix[2], p1, pix[-1], pix[0], q1);
5875 const int p0 = pix[-1];
5876 const int p1 = pix[-2];
5877 const int q0 = pix[0];
5878 const int q1 = pix[1];
5880 if( ABS( p0 - q0 ) < alpha &&
5881 ABS( p1 - p0 ) < beta &&
5882 ABS( q1 - q0 ) < beta ) {
5884 pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2; /* p0' */
5885 pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2; /* q0' */
5886 tprintf("filter_mb_mbaff_edgecv i:%d\n# bS:4 -> [%02x, %02x, %02x, %02x, %02x, %02x] =>[%02x, %02x, %02x, %02x, %02x, %02x]\n", i, pix[-3], p1, p0, q0, q1, pix[2], pix[-3], pix[-2], pix[-1], pix[0], pix[1], pix[2]);
5892 static void filter_mb_edgeh( H264Context *h, uint8_t *pix, int stride, int bS[4], int qp ) {
5894 const int index_a = clip( qp + h->slice_alpha_c0_offset, 0, 51 );
5895 const int alpha = alpha_table[index_a];
5896 const int beta = beta_table[clip( qp + h->slice_beta_offset, 0, 51 )];
5897 const int pix_next = stride;
5902 tc[i] = bS[i] ? tc0_table[index_a][bS[i] - 1] : -1;
5903 h->s.dsp.h264_v_loop_filter_luma(pix, stride, alpha, beta, tc);
5905 /* 16px edge length, see filter_mb_edgev */
5906 for( d = 0; d < 16; d++ ) {
5907 const int p0 = pix[-1*pix_next];
5908 const int p1 = pix[-2*pix_next];
5909 const int p2 = pix[-3*pix_next];
5910 const int q0 = pix[0];
5911 const int q1 = pix[1*pix_next];
5912 const int q2 = pix[2*pix_next];
5914 if( ABS( p0 - q0 ) < alpha &&
5915 ABS( p1 - p0 ) < beta &&
5916 ABS( q1 - q0 ) < beta ) {
5918 const int p3 = pix[-4*pix_next];
5919 const int q3 = pix[ 3*pix_next];
5921 if(ABS( p0 - q0 ) < (( alpha >> 2 ) + 2 )){
5922 if( ABS( p2 - p0 ) < beta) {
5924 pix[-1*pix_next] = ( p2 + 2*p1 + 2*p0 + 2*q0 + q1 + 4 ) >> 3;
5925 pix[-2*pix_next] = ( p2 + p1 + p0 + q0 + 2 ) >> 2;
5926 pix[-3*pix_next] = ( 2*p3 + 3*p2 + p1 + p0 + q0 + 4 ) >> 3;
5929 pix[-1*pix_next] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
5931 if( ABS( q2 - q0 ) < beta) {
5933 pix[0*pix_next] = ( p1 + 2*p0 + 2*q0 + 2*q1 + q2 + 4 ) >> 3;
5934 pix[1*pix_next] = ( p0 + q0 + q1 + q2 + 2 ) >> 2;
5935 pix[2*pix_next] = ( 2*q3 + 3*q2 + q1 + q0 + p0 + 4 ) >> 3;
5938 pix[0*pix_next] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
5942 pix[-1*pix_next] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
5943 pix[ 0*pix_next] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
5945 tprintf("filter_mb_edgeh i:%d d:%d, qp:%d, indexA:%d, alpha:%d, beta:%d\n# bS:%d -> [%02x, %02x, %02x, %02x, %02x, %02x] =>[%02x, %02x, %02x, %02x]\n", i, d, qp, index_a, alpha, beta, bS[i], p2, p1, p0, q0, q1, q2, pix[-2*pix_next], pix[-pix_next], pix[0], pix[pix_next]);
5952 static void filter_mb_edgech( H264Context *h, uint8_t *pix, int stride, int bS[4], int qp ) {
5954 const int index_a = clip( qp + h->slice_alpha_c0_offset, 0, 51 );
5955 const int alpha = alpha_table[index_a];
5956 const int beta = beta_table[clip( qp + h->slice_beta_offset, 0, 51 )];
5957 const int pix_next = stride;
5962 tc[i] = bS[i] ? tc0_table[index_a][bS[i] - 1] + 1 : 0;
5963 h->s.dsp.h264_v_loop_filter_chroma(pix, stride, alpha, beta, tc);
5965 /* 8px edge length, see filter_mb_edgev */
5966 for( d = 0; d < 8; d++ ) {
5967 const int p0 = pix[-1*pix_next];
5968 const int p1 = pix[-2*pix_next];
5969 const int q0 = pix[0];
5970 const int q1 = pix[1*pix_next];
5972 if( ABS( p0 - q0 ) < alpha &&
5973 ABS( p1 - p0 ) < beta &&
5974 ABS( q1 - q0 ) < beta ) {
5976 pix[-pix_next] = ( 2*p1 + p0 + q1 + 2 ) >> 2; /* p0' */
5977 pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2; /* q0' */
5978 tprintf("filter_mb_edgech i:%d d:%d, qp:%d, indexA:%d, alpha:%d, beta:%d\n# bS:%d -> [%02x, %02x, %02x, %02x, %02x, %02x] =>[%02x, %02x, %02x, %02x]\n", i, d, qp, index_a, alpha, beta, bS[i], pix[-3*pix_next], p1, p0, q0, q1, pix[2*pix_next], pix[-2*pix_next], pix[-pix_next], pix[0], pix[pix_next]);
5985 static void filter_mb( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr, unsigned int linesize, unsigned int uvlinesize) {
5986 MpegEncContext * const s = &h->s;
5987 const int mb_xy= mb_x + mb_y*s->mb_stride;
5988 int first_vertical_edge_done = 0;
5990 /* FIXME: A given frame may occupy more than one position in
5991 * the reference list. So ref2frm should be populated with
5992 * frame numbers, not indices. */
5993 static const int ref2frm[18] = {-1,-1,0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15};
5996 // left mb is in picture
5997 && h->slice_table[mb_xy-1] != 255
5998 // and current and left pair do not have the same interlaced type
5999 && (IS_INTERLACED(s->current_picture.mb_type[mb_xy]) != IS_INTERLACED(s->current_picture.mb_type[mb_xy-1]))
6000 // and left mb is in the same slice if deblocking_filter == 2
6001 && (h->deblocking_filter!=2 || h->slice_table[mb_xy-1] == h->slice_table[mb_xy])) {
6002 /* First vertical edge is different in MBAFF frames
6003 * There are 8 different bS to compute and 2 different Qp
6010 first_vertical_edge_done = 1;
6011 for( i = 0; i < 8; i++ ) {
6013 int b_idx= 8 + 4 + 8*y;
6014 int bn_idx= b_idx - 1;
6016 int mbn_xy = h->mb_field_decoding_flag ? h->left_mb_xy[i>>2] : h->left_mb_xy[i&1];
6018 if( IS_INTRA( s->current_picture.mb_type[mb_xy] ) ||
6019 IS_INTRA( s->current_picture.mb_type[mbn_xy] ) ) {
6021 } else if( h->non_zero_count_cache[b_idx] != 0 ||
6022 h->non_zero_count_cache[bn_idx] != 0 ) {
6027 for( l = 0; l < 1 + (h->slice_type == B_TYPE); l++ ) {
6028 if( ref2frm[h->ref_cache[l][b_idx]+2] != ref2frm[h->ref_cache[l][bn_idx]+2] ||
6029 ABS( h->mv_cache[l][b_idx][0] - h->mv_cache[l][bn_idx][0] ) >= 4 ||
6030 ABS( h->mv_cache[l][b_idx][1] - h->mv_cache[l][bn_idx][1] ) >= 4 ) {
6037 if(bS[0]+bS[1]+bS[2]+bS[3] != 0) {
6038 // Do not use s->qscale as luma quantizer because it has not the same
6039 // value in IPCM macroblocks.
6040 qp[0] = ( s->current_picture.qscale_table[mb_xy] + s->current_picture.qscale_table[h->left_mb_xy[0]] + 1 ) >> 1;
6041 chroma_qp[0] = ( get_chroma_qp( h->pps.chroma_qp_index_offset, s->current_picture.qscale_table[mb_xy] ) +
6042 get_chroma_qp( h->pps.chroma_qp_index_offset, s->current_picture.qscale_table[h->left_mb_xy[0]] ) + 1 ) >> 1;
6043 qp[1] = ( s->current_picture.qscale_table[mb_xy] + s->current_picture.qscale_table[h->left_mb_xy[1]] + 1 ) >> 1;
6044 chroma_qp[1] = ( get_chroma_qp( h->pps.chroma_qp_index_offset, s->current_picture.qscale_table[mb_xy] ) +
6045 get_chroma_qp( h->pps.chroma_qp_index_offset, s->current_picture.qscale_table[h->left_mb_xy[1]] ) + 1 ) >> 1;
6048 tprintf("filter mb:%d/%d MBAFF, QPy:%d/%d, QPc:%d/%d ls:%d uvls:%d", mb_x, mb_y, qp[0], qp[1], chroma_qp[0], chroma_qp[1], linesize, uvlinesize);
6049 { int i; for (i = 0; i < 8; i++) tprintf(" bS[%d]:%d", i, bS[i]); tprintf("\n"); }
6050 filter_mb_mbaff_edgev ( h, &img_y [0], linesize, bS, qp );
6051 filter_mb_mbaff_edgecv( h, &img_cb[0], uvlinesize, bS, chroma_qp );
6052 filter_mb_mbaff_edgecv( h, &img_cr[0], uvlinesize, bS, chroma_qp );
6055 /* dir : 0 -> vertical edge, 1 -> horizontal edge */
6056 for( dir = 0; dir < 2; dir++ )
6059 const int mbm_xy = dir == 0 ? mb_xy -1 : h->top_mb_xy;
6060 int start = h->slice_table[mbm_xy] == 255 ? 1 : 0;
6062 if (first_vertical_edge_done) {
6064 first_vertical_edge_done = 0;
6067 if (h->deblocking_filter==2 && h->slice_table[mbm_xy] != h->slice_table[mb_xy])
6071 for( edge = start; edge < 4; edge++ ) {
6072 /* mbn_xy: neighbor macroblock */
6073 int mbn_xy = edge > 0 ? mb_xy : mbm_xy;
6077 if (h->mb_aff_frame && (dir == 1) && (edge == 0) && ((mb_y & 1) == 0)
6078 && !IS_INTERLACED(s->current_picture.mb_type[mb_xy])
6079 && IS_INTERLACED(s->current_picture.mb_type[mbn_xy])
6081 // This is a special case in the norm where the filtering must
6082 // be done twice (one each of the field) even if we are in a
6083 // frame macroblock.
6085 unsigned int tmp_linesize = 2 * linesize;
6086 unsigned int tmp_uvlinesize = 2 * uvlinesize;
6087 int mbn_xy = mb_xy - 2 * s->mb_stride;
6091 if( IS_INTRA( s->current_picture.mb_type[mb_xy] ) ||
6092 IS_INTRA( s->current_picture.mb_type[mbn_xy] ) ) {
6093 bS[0] = bS[1] = bS[2] = bS[3] = 3;
6099 // Do not use s->qscale as luma quantizer because it has not the same
6100 // value in IPCM macroblocks.
6101 qp = ( s->current_picture.qscale_table[mb_xy] + s->current_picture.qscale_table[mbn_xy] + 1 ) >> 1;
6102 tprintf("filter mb:%d/%d dir:%d edge:%d, QPy:%d ls:%d uvls:%d", mb_x, mb_y, dir, edge, qp, tmp_linesize, tmp_uvlinesize);
6103 { int i; for (i = 0; i < 4; i++) tprintf(" bS[%d]:%d", i, bS[i]); tprintf("\n"); }
6104 filter_mb_edgeh( h, &img_y[0], tmp_linesize, bS, qp );
6105 chroma_qp = ( h->chroma_qp +
6106 get_chroma_qp( h->pps.chroma_qp_index_offset, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1;
6107 filter_mb_edgech( h, &img_cb[0], tmp_uvlinesize, bS, chroma_qp );
6108 filter_mb_edgech( h, &img_cr[0], tmp_uvlinesize, bS, chroma_qp );
6111 mbn_xy += s->mb_stride;
6112 if( IS_INTRA( s->current_picture.mb_type[mb_xy] ) ||
6113 IS_INTRA( s->current_picture.mb_type[mbn_xy] ) ) {
6114 bS[0] = bS[1] = bS[2] = bS[3] = 3;
6120 // Do not use s->qscale as luma quantizer because it has not the same
6121 // value in IPCM macroblocks.
6122 qp = ( s->current_picture.qscale_table[mb_xy] + s->current_picture.qscale_table[mbn_xy] + 1 ) >> 1;
6123 tprintf("filter mb:%d/%d dir:%d edge:%d, QPy:%d ls:%d uvls:%d", mb_x, mb_y, dir, edge, qp, tmp_linesize, tmp_uvlinesize);
6124 { int i; for (i = 0; i < 4; i++) tprintf(" bS[%d]:%d", i, bS[i]); tprintf("\n"); }
6125 filter_mb_edgeh( h, &img_y[linesize], tmp_linesize, bS, qp );
6126 chroma_qp = ( h->chroma_qp +
6127 get_chroma_qp( h->pps.chroma_qp_index_offset, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1;
6128 filter_mb_edgech( h, &img_cb[uvlinesize], tmp_uvlinesize, bS, chroma_qp );
6129 filter_mb_edgech( h, &img_cr[uvlinesize], tmp_uvlinesize, bS, chroma_qp );
6132 if( IS_INTRA( s->current_picture.mb_type[mb_xy] ) ||
6133 IS_INTRA( s->current_picture.mb_type[mbn_xy] ) ) {
6136 if ( (!IS_INTERLACED(s->current_picture.mb_type[mb_xy]) && !IS_INTERLACED(s->current_picture.mb_type[mbm_xy]))
6137 || ((h->mb_aff_frame || (s->picture_structure != PICT_FRAME)) && (dir == 0))
6146 bS[0] = bS[1] = bS[2] = bS[3] = value;
6149 for( i = 0; i < 4; i++ ) {
6150 int x = dir == 0 ? edge : i;
6151 int y = dir == 0 ? i : edge;
6152 int b_idx= 8 + 4 + x + 8*y;
6153 int bn_idx= b_idx - (dir ? 8:1);
6155 if( h->non_zero_count_cache[b_idx] != 0 ||
6156 h->non_zero_count_cache[bn_idx] != 0 ) {
6163 for( l = 0; l < 1 + (h->slice_type == B_TYPE); l++ ) {
6164 if( ref2frm[h->ref_cache[l][b_idx]+2] != ref2frm[h->ref_cache[l][bn_idx]+2] ||
6165 ABS( h->mv_cache[l][b_idx][0] - h->mv_cache[l][bn_idx][0] ) >= 4 ||
6166 ABS( h->mv_cache[l][b_idx][1] - h->mv_cache[l][bn_idx][1] ) >= 4 ) {
6174 if(bS[0]+bS[1]+bS[2]+bS[3] == 0)
6179 // Do not use s->qscale as luma quantizer because it has not the same
6180 // value in IPCM macroblocks.
6181 qp = ( s->current_picture.qscale_table[mb_xy] + s->current_picture.qscale_table[mbn_xy] + 1 ) >> 1;
6182 //tprintf("filter mb:%d/%d dir:%d edge:%d, QPy:%d, QPc:%d, QPcn:%d\n", mb_x, mb_y, dir, edge, qp, h->chroma_qp, s->current_picture.qscale_table[mbn_xy]);
6183 tprintf("filter mb:%d/%d dir:%d edge:%d, QPy:%d ls:%d uvls:%d", mb_x, mb_y, dir, edge, qp, linesize, uvlinesize);
6184 { int i; for (i = 0; i < 4; i++) tprintf(" bS[%d]:%d", i, bS[i]); tprintf("\n"); }
6186 filter_mb_edgev( h, &img_y[4*edge], linesize, bS, qp );
6187 if( (edge&1) == 0 ) {
6188 int chroma_qp = ( h->chroma_qp +
6189 get_chroma_qp( h->pps.chroma_qp_index_offset, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1;
6190 filter_mb_edgecv( h, &img_cb[2*edge], uvlinesize, bS, chroma_qp );
6191 filter_mb_edgecv( h, &img_cr[2*edge], uvlinesize, bS, chroma_qp );
6194 filter_mb_edgeh( h, &img_y[4*edge*linesize], linesize, bS, qp );
6195 if( (edge&1) == 0 ) {
6196 int chroma_qp = ( h->chroma_qp +
6197 get_chroma_qp( h->pps.chroma_qp_index_offset, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1;
6198 filter_mb_edgech( h, &img_cb[2*edge*uvlinesize], uvlinesize, bS, chroma_qp );
6199 filter_mb_edgech( h, &img_cr[2*edge*uvlinesize], uvlinesize, bS, chroma_qp );
6206 static int decode_slice(H264Context *h){
6207 MpegEncContext * const s = &h->s;
6208 const int part_mask= s->partitioned_frame ? (AC_END|AC_ERROR) : 0x7F;
6212 if( h->pps.cabac ) {
6216 align_get_bits( &s->gb );
6219 ff_init_cabac_states( &h->cabac, ff_h264_lps_range, ff_h264_mps_state, ff_h264_lps_state, 64 );
6220 ff_init_cabac_decoder( &h->cabac,
6221 s->gb.buffer + get_bits_count(&s->gb)/8,
6222 ( s->gb.size_in_bits - get_bits_count(&s->gb) + 7)/8);
6223 /* calculate pre-state */
6224 for( i= 0; i < 399; i++ ) {
6226 if( h->slice_type == I_TYPE )
6227 pre = clip( ((cabac_context_init_I[i][0] * s->qscale) >>4 ) + cabac_context_init_I[i][1], 1, 126 );
6229 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 );
6232 h->cabac_state[i] = 2 * ( 63 - pre ) + 0;
6234 h->cabac_state[i] = 2 * ( pre - 64 ) + 1;
6238 int ret = decode_mb_cabac(h);
6241 if(ret>=0) hl_decode_mb(h);
6243 /* XXX: useless as decode_mb_cabac it doesn't support that ... */
6244 if( ret >= 0 && h->mb_aff_frame ) { //FIXME optimal? or let mb_decode decode 16x32 ?
6247 if(ret>=0) ret = decode_mb_cabac(h);
6252 eos = get_cabac_terminate( &h->cabac );
6254 if( ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 1) {
6255 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
6256 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);
6260 if( ++s->mb_x >= s->mb_width ) {
6262 ff_draw_horiz_band(s, 16*s->mb_y, 16);
6264 if(h->mb_aff_frame) {
6269 if( eos || s->mb_y >= s->mb_height ) {
6270 tprintf("slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
6271 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);
6278 int ret = decode_mb_cavlc(h);
6280 if(ret>=0) hl_decode_mb(h);
6282 if(ret>=0 && h->mb_aff_frame){ //FIXME optimal? or let mb_decode decode 16x32 ?
6284 ret = decode_mb_cavlc(h);
6286 if(ret>=0) hl_decode_mb(h);
6291 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
6292 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);
6297 if(++s->mb_x >= s->mb_width){
6299 ff_draw_horiz_band(s, 16*s->mb_y, 16);
6301 if(h->mb_aff_frame) {
6304 if(s->mb_y >= s->mb_height){
6305 tprintf("slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
6307 if(get_bits_count(&s->gb) == s->gb.size_in_bits ) {
6308 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);
6312 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);
6319 if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->mb_skip_run<=0){
6320 tprintf("slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
6321 if(get_bits_count(&s->gb) == s->gb.size_in_bits ){
6322 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);
6326 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);
6335 for(;s->mb_y < s->mb_height; s->mb_y++){
6336 for(;s->mb_x < s->mb_width; s->mb_x++){
6337 int ret= decode_mb(h);
6342 fprintf(stderr, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
6343 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);
6348 if(++s->mb_x >= s->mb_width){
6350 if(++s->mb_y >= s->mb_height){
6351 if(get_bits_count(s->gb) == s->gb.size_in_bits){
6352 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);
6356 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);
6363 if(get_bits_count(s->?gb) >= s->gb?.size_in_bits){
6364 if(get_bits_count(s->gb) == s->gb.size_in_bits){
6365 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);
6369 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);
6376 ff_draw_horiz_band(s, 16*s->mb_y, 16);
6379 return -1; //not reached
6382 static inline void decode_hrd_parameters(H264Context *h, SPS *sps){
6383 MpegEncContext * const s = &h->s;
6385 cpb_count = get_ue_golomb(&s->gb) + 1;
6386 get_bits(&s->gb, 4); /* bit_rate_scale */
6387 get_bits(&s->gb, 4); /* cpb_size_scale */
6388 for(i=0; i<cpb_count; i++){
6389 get_ue_golomb(&s->gb); /* bit_rate_value_minus1 */
6390 get_ue_golomb(&s->gb); /* cpb_size_value_minus1 */
6391 get_bits1(&s->gb); /* cbr_flag */
6393 get_bits(&s->gb, 5); /* initial_cpb_removal_delay_length_minus1 */
6394 get_bits(&s->gb, 5); /* cpb_removal_delay_length_minus1 */
6395 get_bits(&s->gb, 5); /* dpb_output_delay_length_minus1 */
6396 get_bits(&s->gb, 5); /* time_offset_length */
6399 static inline int decode_vui_parameters(H264Context *h, SPS *sps){
6400 MpegEncContext * const s = &h->s;
6401 int aspect_ratio_info_present_flag, aspect_ratio_idc;
6402 int nal_hrd_parameters_present_flag, vcl_hrd_parameters_present_flag;
6404 aspect_ratio_info_present_flag= get_bits1(&s->gb);
6406 if( aspect_ratio_info_present_flag ) {
6407 aspect_ratio_idc= get_bits(&s->gb, 8);
6408 if( aspect_ratio_idc == EXTENDED_SAR ) {
6409 sps->sar.num= get_bits(&s->gb, 16);
6410 sps->sar.den= get_bits(&s->gb, 16);
6411 }else if(aspect_ratio_idc < 16){
6412 sps->sar= pixel_aspect[aspect_ratio_idc];
6414 av_log(h->s.avctx, AV_LOG_ERROR, "illegal aspect ratio\n");
6421 // s->avctx->aspect_ratio= sar_width*s->width / (float)(s->height*sar_height);
6423 if(get_bits1(&s->gb)){ /* overscan_info_present_flag */
6424 get_bits1(&s->gb); /* overscan_appropriate_flag */
6427 if(get_bits1(&s->gb)){ /* video_signal_type_present_flag */
6428 get_bits(&s->gb, 3); /* video_format */
6429 get_bits1(&s->gb); /* video_full_range_flag */
6430 if(get_bits1(&s->gb)){ /* colour_description_present_flag */
6431 get_bits(&s->gb, 8); /* colour_primaries */
6432 get_bits(&s->gb, 8); /* transfer_characteristics */
6433 get_bits(&s->gb, 8); /* matrix_coefficients */
6437 if(get_bits1(&s->gb)){ /* chroma_location_info_present_flag */
6438 get_ue_golomb(&s->gb); /* chroma_sample_location_type_top_field */
6439 get_ue_golomb(&s->gb); /* chroma_sample_location_type_bottom_field */
6442 sps->timing_info_present_flag = get_bits1(&s->gb);
6443 if(sps->timing_info_present_flag){
6444 sps->num_units_in_tick = get_bits_long(&s->gb, 32);
6445 sps->time_scale = get_bits_long(&s->gb, 32);
6446 sps->fixed_frame_rate_flag = get_bits1(&s->gb);
6449 nal_hrd_parameters_present_flag = get_bits1(&s->gb);
6450 if(nal_hrd_parameters_present_flag)
6451 decode_hrd_parameters(h, sps);
6452 vcl_hrd_parameters_present_flag = get_bits1(&s->gb);
6453 if(vcl_hrd_parameters_present_flag)
6454 decode_hrd_parameters(h, sps);
6455 if(nal_hrd_parameters_present_flag || vcl_hrd_parameters_present_flag)
6456 get_bits1(&s->gb); /* low_delay_hrd_flag */
6457 get_bits1(&s->gb); /* pic_struct_present_flag */
6459 sps->bitstream_restriction_flag = get_bits1(&s->gb);
6460 if(sps->bitstream_restriction_flag){
6461 get_bits1(&s->gb); /* motion_vectors_over_pic_boundaries_flag */
6462 get_ue_golomb(&s->gb); /* max_bytes_per_pic_denom */
6463 get_ue_golomb(&s->gb); /* max_bits_per_mb_denom */
6464 get_ue_golomb(&s->gb); /* log2_max_mv_length_horizontal */
6465 get_ue_golomb(&s->gb); /* log2_max_mv_length_vertical */
6466 sps->num_reorder_frames = get_ue_golomb(&s->gb);
6467 get_ue_golomb(&s->gb); /* max_dec_frame_buffering */
6473 static inline int decode_seq_parameter_set(H264Context *h){
6474 MpegEncContext * const s = &h->s;
6475 int profile_idc, level_idc;
6479 profile_idc= get_bits(&s->gb, 8);
6480 get_bits1(&s->gb); //constraint_set0_flag
6481 get_bits1(&s->gb); //constraint_set1_flag
6482 get_bits1(&s->gb); //constraint_set2_flag
6483 get_bits1(&s->gb); //constraint_set3_flag
6484 get_bits(&s->gb, 4); // reserved
6485 level_idc= get_bits(&s->gb, 8);
6486 sps_id= get_ue_golomb(&s->gb);
6488 sps= &h->sps_buffer[ sps_id ];
6489 sps->profile_idc= profile_idc;
6490 sps->level_idc= level_idc;
6492 sps->log2_max_frame_num= get_ue_golomb(&s->gb) + 4;
6493 sps->poc_type= get_ue_golomb(&s->gb);
6495 if(sps->poc_type == 0){ //FIXME #define
6496 sps->log2_max_poc_lsb= get_ue_golomb(&s->gb) + 4;
6497 } else if(sps->poc_type == 1){//FIXME #define
6498 sps->delta_pic_order_always_zero_flag= get_bits1(&s->gb);
6499 sps->offset_for_non_ref_pic= get_se_golomb(&s->gb);
6500 sps->offset_for_top_to_bottom_field= get_se_golomb(&s->gb);
6501 sps->poc_cycle_length= get_ue_golomb(&s->gb);
6503 for(i=0; i<sps->poc_cycle_length; i++)
6504 sps->offset_for_ref_frame[i]= get_se_golomb(&s->gb);
6506 if(sps->poc_type > 2){
6507 av_log(h->s.avctx, AV_LOG_ERROR, "illegal POC type %d\n", sps->poc_type);
6511 sps->ref_frame_count= get_ue_golomb(&s->gb);
6512 if(sps->ref_frame_count > MAX_PICTURE_COUNT-2){
6513 av_log(h->s.avctx, AV_LOG_ERROR, "too many reference frames\n");
6515 sps->gaps_in_frame_num_allowed_flag= get_bits1(&s->gb);
6516 sps->mb_width= get_ue_golomb(&s->gb) + 1;
6517 sps->mb_height= get_ue_golomb(&s->gb) + 1;
6518 if((unsigned)sps->mb_width >= INT_MAX/16 || (unsigned)sps->mb_height >= INT_MAX/16 ||
6519 avcodec_check_dimensions(NULL, 16*sps->mb_width, 16*sps->mb_height))
6522 sps->frame_mbs_only_flag= get_bits1(&s->gb);
6523 if(!sps->frame_mbs_only_flag)
6524 sps->mb_aff= get_bits1(&s->gb);
6528 sps->direct_8x8_inference_flag= get_bits1(&s->gb);
6530 sps->crop= get_bits1(&s->gb);
6532 sps->crop_left = get_ue_golomb(&s->gb);
6533 sps->crop_right = get_ue_golomb(&s->gb);
6534 sps->crop_top = get_ue_golomb(&s->gb);
6535 sps->crop_bottom= get_ue_golomb(&s->gb);
6536 if(sps->crop_left || sps->crop_top){
6537 av_log(h->s.avctx, AV_LOG_ERROR, "insane cropping not completely supported, this could look slightly wrong ...\n");
6543 sps->crop_bottom= 0;
6546 sps->vui_parameters_present_flag= get_bits1(&s->gb);
6547 if( sps->vui_parameters_present_flag )
6548 decode_vui_parameters(h, sps);
6550 if(s->avctx->debug&FF_DEBUG_PICT_INFO){
6551 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",
6552 sps_id, sps->profile_idc, sps->level_idc,
6554 sps->ref_frame_count,
6555 sps->mb_width, sps->mb_height,
6556 sps->frame_mbs_only_flag ? "FRM" : (sps->mb_aff ? "MB-AFF" : "PIC-AFF"),
6557 sps->direct_8x8_inference_flag ? "8B8" : "",
6558 sps->crop_left, sps->crop_right,
6559 sps->crop_top, sps->crop_bottom,
6560 sps->vui_parameters_present_flag ? "VUI" : ""
6566 static inline int decode_picture_parameter_set(H264Context *h){
6567 MpegEncContext * const s = &h->s;
6568 int pps_id= get_ue_golomb(&s->gb);
6569 PPS *pps= &h->pps_buffer[pps_id];
6571 pps->sps_id= get_ue_golomb(&s->gb);
6572 pps->cabac= get_bits1(&s->gb);
6573 pps->pic_order_present= get_bits1(&s->gb);
6574 pps->slice_group_count= get_ue_golomb(&s->gb) + 1;
6575 if(pps->slice_group_count > 1 ){
6576 pps->mb_slice_group_map_type= get_ue_golomb(&s->gb);
6577 av_log(h->s.avctx, AV_LOG_ERROR, "FMO not supported\n");
6578 switch(pps->mb_slice_group_map_type){
6581 | for( i = 0; i <= num_slice_groups_minus1; i++ ) | | |
6582 | run_length[ i ] |1 |ue(v) |
6587 | for( i = 0; i < num_slice_groups_minus1; i++ ) | | |
6589 | top_left_mb[ i ] |1 |ue(v) |
6590 | bottom_right_mb[ i ] |1 |ue(v) |
6598 | slice_group_change_direction_flag |1 |u(1) |
6599 | slice_group_change_rate_minus1 |1 |ue(v) |
6604 | slice_group_id_cnt_minus1 |1 |ue(v) |
6605 | for( i = 0; i <= slice_group_id_cnt_minus1; i++ | | |
6607 | slice_group_id[ i ] |1 |u(v) |
6612 pps->ref_count[0]= get_ue_golomb(&s->gb) + 1;
6613 pps->ref_count[1]= get_ue_golomb(&s->gb) + 1;
6614 if(pps->ref_count[0] > 32 || pps->ref_count[1] > 32){
6615 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow (pps)\n");
6619 pps->weighted_pred= get_bits1(&s->gb);
6620 pps->weighted_bipred_idc= get_bits(&s->gb, 2);
6621 pps->init_qp= get_se_golomb(&s->gb) + 26;
6622 pps->init_qs= get_se_golomb(&s->gb) + 26;
6623 pps->chroma_qp_index_offset= get_se_golomb(&s->gb);
6624 pps->deblocking_filter_parameters_present= get_bits1(&s->gb);
6625 pps->constrained_intra_pred= get_bits1(&s->gb);
6626 pps->redundant_pic_cnt_present = get_bits1(&s->gb);
6628 if(s->avctx->debug&FF_DEBUG_PICT_INFO){
6629 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",
6630 pps_id, pps->sps_id,
6631 pps->cabac ? "CABAC" : "CAVLC",
6632 pps->slice_group_count,
6633 pps->ref_count[0], pps->ref_count[1],
6634 pps->weighted_pred ? "weighted" : "",
6635 pps->init_qp, pps->init_qs, pps->chroma_qp_index_offset,
6636 pps->deblocking_filter_parameters_present ? "LPAR" : "",
6637 pps->constrained_intra_pred ? "CONSTR" : "",
6638 pps->redundant_pic_cnt_present ? "REDU" : ""
6646 * finds the end of the current frame in the bitstream.
6647 * @return the position of the first byte of the next frame, or -1
6649 static int find_frame_end(H264Context *h, const uint8_t *buf, int buf_size){
6652 ParseContext *pc = &(h->s.parse_context);
6653 //printf("first %02X%02X%02X%02X\n", buf[0], buf[1],buf[2],buf[3]);
6654 // mb_addr= pc->mb_addr - 1;
6656 for(i=0; i<=buf_size; i++){
6657 if((state&0xFFFFFF1F) == 0x101 || (state&0xFFFFFF1F) == 0x102 || (state&0xFFFFFF1F) == 0x105){
6658 tprintf("find_frame_end new startcode = %08x, frame_start_found = %d, pos = %d\n", state, pc->frame_start_found, i);
6659 if(pc->frame_start_found){
6660 // If there isn't one more byte in the buffer
6661 // the test on first_mb_in_slice cannot be done yet
6662 // do it at next call.
6663 if (i >= buf_size) break;
6664 if (buf[i] & 0x80) {
6665 // first_mb_in_slice is 0, probably the first nal of a new
6667 tprintf("find_frame_end frame_end_found, state = %08x, pos = %d\n", state, i);
6669 pc->frame_start_found= 0;
6673 pc->frame_start_found = 1;
6676 state= (state<<8) | buf[i];
6680 return END_NOT_FOUND;
6683 static int h264_parse(AVCodecParserContext *s,
6684 AVCodecContext *avctx,
6685 uint8_t **poutbuf, int *poutbuf_size,
6686 const uint8_t *buf, int buf_size)
6688 H264Context *h = s->priv_data;
6689 ParseContext *pc = &h->s.parse_context;
6692 next= find_frame_end(h, buf, buf_size);
6694 if (ff_combine_frame(pc, next, (uint8_t **)&buf, &buf_size) < 0) {
6700 *poutbuf = (uint8_t *)buf;
6701 *poutbuf_size = buf_size;
6705 static int decode_nal_units(H264Context *h, uint8_t *buf, int buf_size){
6706 MpegEncContext * const s = &h->s;
6707 AVCodecContext * const avctx= s->avctx;
6711 for(i=0; i<32; i++){
6712 printf("%X ", buf[i]);
6724 if(buf_index >= buf_size) break;
6726 for(i = 0; i < h->nal_length_size; i++)
6727 nalsize = (nalsize << 8) | buf[buf_index++];
6729 // start code prefix search
6730 for(; buf_index + 3 < buf_size; buf_index++){
6731 // this should allways succeed in the first iteration
6732 if(buf[buf_index] == 0 && buf[buf_index+1] == 0 && buf[buf_index+2] == 1)
6736 if(buf_index+3 >= buf_size) break;
6741 ptr= decode_nal(h, buf + buf_index, &dst_length, &consumed, h->is_avc ? nalsize : buf_size - buf_index);
6742 if(ptr[dst_length - 1] == 0) dst_length--;
6743 bit_length= 8*dst_length - decode_rbsp_trailing(ptr + dst_length - 1);
6745 if(s->avctx->debug&FF_DEBUG_STARTCODE){
6746 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);
6749 if (h->is_avc && (nalsize != consumed))
6750 av_log(h->s.avctx, AV_LOG_ERROR, "AVC: Consumed only %d bytes instead of %d\n", consumed, nalsize);
6752 buf_index += consumed;
6754 if( s->hurry_up == 1 && h->nal_ref_idc == 0 )
6757 switch(h->nal_unit_type){
6759 idr(h); //FIXME ensure we don't loose some frames if there is reordering
6761 init_get_bits(&s->gb, ptr, bit_length);
6763 h->inter_gb_ptr= &s->gb;
6764 s->data_partitioning = 0;
6766 if(decode_slice_header(h) < 0) return -1;
6767 if(h->redundant_pic_count==0 && s->hurry_up < 5 )
6771 init_get_bits(&s->gb, ptr, bit_length);
6773 h->inter_gb_ptr= NULL;
6774 s->data_partitioning = 1;
6776 if(decode_slice_header(h) < 0) return -1;
6779 init_get_bits(&h->intra_gb, ptr, bit_length);
6780 h->intra_gb_ptr= &h->intra_gb;
6783 init_get_bits(&h->inter_gb, ptr, bit_length);
6784 h->inter_gb_ptr= &h->inter_gb;
6786 if(h->redundant_pic_count==0 && h->intra_gb_ptr && s->data_partitioning && s->hurry_up < 5 )
6792 init_get_bits(&s->gb, ptr, bit_length);
6793 decode_seq_parameter_set(h);
6795 if(s->flags& CODEC_FLAG_LOW_DELAY)
6798 if(avctx->has_b_frames < 2)
6799 avctx->has_b_frames= !s->low_delay;
6802 init_get_bits(&s->gb, ptr, bit_length);
6804 decode_picture_parameter_set(h);
6807 case NAL_PICTURE_DELIMITER:
6809 case NAL_FILTER_DATA:
6812 av_log(avctx, AV_LOG_ERROR, "Unknown NAL code: %d\n", h->nal_unit_type);
6816 if(!s->current_picture_ptr) return buf_index; //no frame
6818 s->current_picture_ptr->pict_type= s->pict_type;
6819 s->current_picture_ptr->key_frame= s->pict_type == I_TYPE && h->nal_unit_type == NAL_IDR_SLICE;
6821 h->prev_frame_num_offset= h->frame_num_offset;
6822 h->prev_frame_num= h->frame_num;
6823 if(s->current_picture_ptr->reference){
6824 h->prev_poc_msb= h->poc_msb;
6825 h->prev_poc_lsb= h->poc_lsb;
6827 if(s->current_picture_ptr->reference)
6828 execute_ref_pic_marking(h, h->mmco, h->mmco_index);
6838 * returns the number of bytes consumed for building the current frame
6840 static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size){
6841 if(s->flags&CODEC_FLAG_TRUNCATED){
6842 pos -= s->parse_context.last_index;
6843 if(pos<0) pos=0; // FIXME remove (unneeded?)
6847 if(pos==0) pos=1; //avoid infinite loops (i doubt thats needed but ...)
6848 if(pos+10>buf_size) pos=buf_size; // oops ;)
6854 static int decode_frame(AVCodecContext *avctx,
6855 void *data, int *data_size,
6856 uint8_t *buf, int buf_size)
6858 H264Context *h = avctx->priv_data;
6859 MpegEncContext *s = &h->s;
6860 AVFrame *pict = data;
6863 s->flags= avctx->flags;
6864 s->flags2= avctx->flags2;
6866 /* no supplementary picture */
6867 if (buf_size == 0) {
6871 if(s->flags&CODEC_FLAG_TRUNCATED){
6872 int next= find_frame_end(h, buf, buf_size);
6874 if( ff_combine_frame(&s->parse_context, next, &buf, &buf_size) < 0 )
6876 //printf("next:%d buf_size:%d last_index:%d\n", next, buf_size, s->parse_context.last_index);
6879 if(h->is_avc && !h->got_avcC) {
6880 int i, cnt, nalsize;
6881 unsigned char *p = avctx->extradata;
6882 if(avctx->extradata_size < 7) {
6883 av_log(avctx, AV_LOG_ERROR, "avcC too short\n");
6887 av_log(avctx, AV_LOG_ERROR, "Unknown avcC version %d\n", *p);
6890 /* sps and pps in the avcC always have length coded with 2 bytes,
6891 so put a fake nal_length_size = 2 while parsing them */
6892 h->nal_length_size = 2;
6893 // Decode sps from avcC
6894 cnt = *(p+5) & 0x1f; // Number of sps
6896 for (i = 0; i < cnt; i++) {
6897 nalsize = BE_16(p) + 2;
6898 if(decode_nal_units(h, p, nalsize) != nalsize) {
6899 av_log(avctx, AV_LOG_ERROR, "Decoding sps %d from avcC failed\n", i);
6904 // Decode pps from avcC
6905 cnt = *(p++); // Number of pps
6906 for (i = 0; i < cnt; i++) {
6907 nalsize = BE_16(p) + 2;
6908 if(decode_nal_units(h, p, nalsize) != nalsize) {
6909 av_log(avctx, AV_LOG_ERROR, "Decoding pps %d from avcC failed\n", i);
6914 // Now store right nal length size, that will be use to parse all other nals
6915 h->nal_length_size = ((*(((char*)(avctx->extradata))+4))&0x03)+1;
6916 // Do not reparse avcC
6920 if(!h->is_avc && s->avctx->extradata_size && s->picture_number==0){
6921 if(0 < decode_nal_units(h, s->avctx->extradata, s->avctx->extradata_size) )
6925 buf_index=decode_nal_units(h, buf, buf_size);
6929 //FIXME do something with unavailable reference frames
6931 // if(ret==FRAME_SKIPPED) return get_consumed_bytes(s, buf_index, buf_size);
6932 if(!s->current_picture_ptr){
6933 av_log(h->s.avctx, AV_LOG_DEBUG, "error, NO frame\n");
6938 Picture *out = s->current_picture_ptr;
6939 #if 0 //decode order
6940 *data_size = sizeof(AVFrame);
6942 /* Sort B-frames into display order */
6943 Picture *cur = s->current_picture_ptr;
6944 Picture *prev = h->delayed_output_pic;
6949 int dropped_frame = 0;
6952 if(h->sps.bitstream_restriction_flag
6953 && s->avctx->has_b_frames < h->sps.num_reorder_frames){
6954 s->avctx->has_b_frames = h->sps.num_reorder_frames;
6958 while(h->delayed_pic[pics]) pics++;
6959 h->delayed_pic[pics++] = cur;
6960 if(cur->reference == 0)
6963 for(i=0; h->delayed_pic[i]; i++)
6964 if(h->delayed_pic[i]->key_frame || h->delayed_pic[i]->poc==0)
6967 out = h->delayed_pic[0];
6968 for(i=1; h->delayed_pic[i] && !h->delayed_pic[i]->key_frame; i++)
6969 if(h->delayed_pic[i]->poc < out->poc){
6970 out = h->delayed_pic[i];
6974 out_of_order = !cross_idr && prev && out->poc < prev->poc;
6975 if(prev && pics <= s->avctx->has_b_frames)
6977 else if((out_of_order && pics-1 == s->avctx->has_b_frames)
6979 ((!cross_idr && prev && out->poc > prev->poc + 2)
6980 || cur->pict_type == B_TYPE)))
6983 s->avctx->has_b_frames++;
6986 else if(out_of_order)
6989 if(out_of_order || pics > s->avctx->has_b_frames){
6990 dropped_frame = (out != h->delayed_pic[out_idx]);
6991 for(i=out_idx; h->delayed_pic[i]; i++)
6992 h->delayed_pic[i] = h->delayed_pic[i+1];
6995 if(prev == out && !dropped_frame)
6998 *data_size = sizeof(AVFrame);
6999 if(prev && prev != out && prev->reference == 1)
7000 prev->reference = 0;
7001 h->delayed_output_pic = out;
7004 *pict= *(AVFrame*)out;
7007 assert(pict->data[0]);
7008 ff_print_debug_info(s, pict);
7009 //printf("out %d\n", (int)pict->data[0]);
7012 /* Return the Picture timestamp as the frame number */
7013 /* we substract 1 because it is added on utils.c */
7014 avctx->frame_number = s->picture_number - 1;
7016 return get_consumed_bytes(s, buf_index, buf_size);
7019 static inline void fill_mb_avail(H264Context *h){
7020 MpegEncContext * const s = &h->s;
7021 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
7024 h->mb_avail[0]= s->mb_x && h->slice_table[mb_xy - s->mb_stride - 1] == h->slice_num;
7025 h->mb_avail[1]= h->slice_table[mb_xy - s->mb_stride ] == h->slice_num;
7026 h->mb_avail[2]= s->mb_x+1 < s->mb_width && h->slice_table[mb_xy - s->mb_stride + 1] == h->slice_num;
7032 h->mb_avail[3]= s->mb_x && h->slice_table[mb_xy - 1] == h->slice_num;
7033 h->mb_avail[4]= 1; //FIXME move out
7034 h->mb_avail[5]= 0; //FIXME move out
7040 #define SIZE (COUNT*40)
7046 // int int_temp[10000];
7048 AVCodecContext avctx;
7050 dsputil_init(&dsp, &avctx);
7052 init_put_bits(&pb, temp, SIZE);
7053 printf("testing unsigned exp golomb\n");
7054 for(i=0; i<COUNT; i++){
7056 set_ue_golomb(&pb, i);
7057 STOP_TIMER("set_ue_golomb");
7059 flush_put_bits(&pb);
7061 init_get_bits(&gb, temp, 8*SIZE);
7062 for(i=0; i<COUNT; i++){
7065 s= show_bits(&gb, 24);
7068 j= get_ue_golomb(&gb);
7070 printf("missmatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
7073 STOP_TIMER("get_ue_golomb");
7077 init_put_bits(&pb, temp, SIZE);
7078 printf("testing signed exp golomb\n");
7079 for(i=0; i<COUNT; i++){
7081 set_se_golomb(&pb, i - COUNT/2);
7082 STOP_TIMER("set_se_golomb");
7084 flush_put_bits(&pb);
7086 init_get_bits(&gb, temp, 8*SIZE);
7087 for(i=0; i<COUNT; i++){
7090 s= show_bits(&gb, 24);
7093 j= get_se_golomb(&gb);
7094 if(j != i - COUNT/2){
7095 printf("missmatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
7098 STOP_TIMER("get_se_golomb");
7101 printf("testing 4x4 (I)DCT\n");
7104 uint8_t src[16], ref[16];
7105 uint64_t error= 0, max_error=0;
7107 for(i=0; i<COUNT; i++){
7109 // printf("%d %d %d\n", r1, r2, (r2-r1)*16);
7110 for(j=0; j<16; j++){
7111 ref[j]= random()%255;
7112 src[j]= random()%255;
7115 h264_diff_dct_c(block, src, ref, 4);
7118 for(j=0; j<16; j++){
7119 // printf("%d ", block[j]);
7120 block[j]= block[j]*4;
7121 if(j&1) block[j]= (block[j]*4 + 2)/5;
7122 if(j&4) block[j]= (block[j]*4 + 2)/5;
7126 s->dsp.h264_idct_add(ref, block, 4);
7127 /* for(j=0; j<16; j++){
7128 printf("%d ", ref[j]);
7132 for(j=0; j<16; j++){
7133 int diff= ABS(src[j] - ref[j]);
7136 max_error= FFMAX(max_error, diff);
7139 printf("error=%f max_error=%d\n", ((float)error)/COUNT/16, (int)max_error );
7141 printf("testing quantizer\n");
7142 for(qp=0; qp<52; qp++){
7144 src1_block[i]= src2_block[i]= random()%255;
7148 printf("Testing NAL layer\n");
7150 uint8_t bitstream[COUNT];
7151 uint8_t nal[COUNT*2];
7153 memset(&h, 0, sizeof(H264Context));
7155 for(i=0; i<COUNT; i++){
7163 for(j=0; j<COUNT; j++){
7164 bitstream[j]= (random() % 255) + 1;
7167 for(j=0; j<zeros; j++){
7168 int pos= random() % COUNT;
7169 while(bitstream[pos] == 0){
7178 nal_length= encode_nal(&h, nal, bitstream, COUNT, COUNT*2);
7180 printf("encoding failed\n");
7184 out= decode_nal(&h, nal, &out_length, &consumed, nal_length);
7188 if(out_length != COUNT){
7189 printf("incorrect length %d %d\n", out_length, COUNT);
7193 if(consumed != nal_length){
7194 printf("incorrect consumed length %d %d\n", nal_length, consumed);
7198 if(memcmp(bitstream, out, COUNT)){
7199 printf("missmatch\n");
7204 printf("Testing RBSP\n");
7212 static int decode_end(AVCodecContext *avctx)
7214 H264Context *h = avctx->priv_data;
7215 MpegEncContext *s = &h->s;
7217 free_tables(h); //FIXME cleanup init stuff perhaps
7220 // memset(h, 0, sizeof(H264Context));
7226 AVCodec h264_decoder = {
7230 sizeof(H264Context),
7235 /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 | CODEC_CAP_TRUNCATED | CODEC_CAP_DELAY,
7239 AVCodecParser h264_parser = {
7241 sizeof(H264Context),