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 assert((stride&(w-1))==0);
374 //FIXME check what gcc generates for 64 bit on x86 and possibly write a 32 bit ver of it
377 *(uint16_t*)(p + stride)= size==4 ? val : val*0x0101;
378 }else if(w==2 && h==4){
379 *(uint16_t*)(p + 0*stride)=
380 *(uint16_t*)(p + 1*stride)=
381 *(uint16_t*)(p + 2*stride)=
382 *(uint16_t*)(p + 3*stride)= size==4 ? val : val*0x0101;
383 }else if(w==4 && h==1){
384 *(uint32_t*)(p + 0*stride)= size==4 ? val : val*0x01010101;
385 }else if(w==4 && h==2){
386 *(uint32_t*)(p + 0*stride)=
387 *(uint32_t*)(p + 1*stride)= size==4 ? val : val*0x01010101;
388 }else if(w==4 && h==4){
389 *(uint32_t*)(p + 0*stride)=
390 *(uint32_t*)(p + 1*stride)=
391 *(uint32_t*)(p + 2*stride)=
392 *(uint32_t*)(p + 3*stride)= size==4 ? val : val*0x01010101;
393 }else if(w==8 && h==1){
395 *(uint32_t*)(p + 4)= size==4 ? val : val*0x01010101;
396 }else if(w==8 && h==2){
397 *(uint32_t*)(p + 0 + 0*stride)=
398 *(uint32_t*)(p + 4 + 0*stride)=
399 *(uint32_t*)(p + 0 + 1*stride)=
400 *(uint32_t*)(p + 4 + 1*stride)= size==4 ? val : val*0x01010101;
401 }else if(w==8 && h==4){
402 *(uint64_t*)(p + 0*stride)=
403 *(uint64_t*)(p + 1*stride)=
404 *(uint64_t*)(p + 2*stride)=
405 *(uint64_t*)(p + 3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
406 }else if(w==16 && h==2){
407 *(uint64_t*)(p + 0+0*stride)=
408 *(uint64_t*)(p + 8+0*stride)=
409 *(uint64_t*)(p + 0+1*stride)=
410 *(uint64_t*)(p + 8+1*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
411 }else if(w==16 && h==4){
412 *(uint64_t*)(p + 0+0*stride)=
413 *(uint64_t*)(p + 8+0*stride)=
414 *(uint64_t*)(p + 0+1*stride)=
415 *(uint64_t*)(p + 8+1*stride)=
416 *(uint64_t*)(p + 0+2*stride)=
417 *(uint64_t*)(p + 8+2*stride)=
418 *(uint64_t*)(p + 0+3*stride)=
419 *(uint64_t*)(p + 8+3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
424 static inline void fill_caches(H264Context *h, int mb_type, int for_deblock){
425 MpegEncContext * const s = &h->s;
426 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
427 int topleft_xy, top_xy, topright_xy, left_xy[2];
428 int topleft_type, top_type, topright_type, left_type[2];
432 //FIXME deblocking can skip fill_caches much of the time with multiple slices too.
433 // the actual condition is whether we're on the edge of a slice,
434 // and even then the intra and nnz parts are unnecessary.
435 if(for_deblock && h->slice_num == 1)
438 //wow what a mess, why didn't they simplify the interlacing&intra stuff, i can't imagine that these complex rules are worth it
440 top_xy = mb_xy - s->mb_stride;
441 topleft_xy = top_xy - 1;
442 topright_xy= top_xy + 1;
443 left_xy[1] = left_xy[0] = mb_xy-1;
453 const int pair_xy = s->mb_x + (s->mb_y & ~1)*s->mb_stride;
454 const int top_pair_xy = pair_xy - s->mb_stride;
455 const int topleft_pair_xy = top_pair_xy - 1;
456 const int topright_pair_xy = top_pair_xy + 1;
457 const int topleft_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[topleft_pair_xy]);
458 const int top_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[top_pair_xy]);
459 const int topright_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[topright_pair_xy]);
460 const int left_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[pair_xy-1]);
461 const int curr_mb_frame_flag = !IS_INTERLACED(mb_type);
462 const int bottom = (s->mb_y & 1);
463 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);
465 ? !curr_mb_frame_flag // bottom macroblock
466 : (!curr_mb_frame_flag && !top_mb_frame_flag) // top macroblock
468 top_xy -= s->mb_stride;
471 ? !curr_mb_frame_flag // bottom macroblock
472 : (!curr_mb_frame_flag && !topleft_mb_frame_flag) // top macroblock
474 topleft_xy -= s->mb_stride;
477 ? !curr_mb_frame_flag // bottom macroblock
478 : (!curr_mb_frame_flag && !topright_mb_frame_flag) // top macroblock
480 topright_xy -= s->mb_stride;
482 if (left_mb_frame_flag != curr_mb_frame_flag) {
483 left_xy[1] = left_xy[0] = pair_xy - 1;
484 if (curr_mb_frame_flag) {
505 left_xy[1] += s->mb_stride;
518 h->top_mb_xy = top_xy;
519 h->left_mb_xy[0] = left_xy[0];
520 h->left_mb_xy[1] = left_xy[1];
522 topleft_type = h->slice_table[topleft_xy ] < 255 ? s->current_picture.mb_type[topleft_xy] : 0;
523 top_type = h->slice_table[top_xy ] < 255 ? s->current_picture.mb_type[top_xy] : 0;
524 topright_type= h->slice_table[topright_xy] < 255 ? s->current_picture.mb_type[topright_xy]: 0;
525 left_type[0] = h->slice_table[left_xy[0] ] < 255 ? s->current_picture.mb_type[left_xy[0]] : 0;
526 left_type[1] = h->slice_table[left_xy[1] ] < 255 ? s->current_picture.mb_type[left_xy[1]] : 0;
528 topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
529 top_type = h->slice_table[top_xy ] == h->slice_num ? s->current_picture.mb_type[top_xy] : 0;
530 topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
531 left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
532 left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
535 if(IS_INTRA(mb_type)){
536 h->topleft_samples_available=
537 h->top_samples_available=
538 h->left_samples_available= 0xFFFF;
539 h->topright_samples_available= 0xEEEA;
541 if(!IS_INTRA(top_type) && (top_type==0 || h->pps.constrained_intra_pred)){
542 h->topleft_samples_available= 0xB3FF;
543 h->top_samples_available= 0x33FF;
544 h->topright_samples_available= 0x26EA;
547 if(!IS_INTRA(left_type[i]) && (left_type[i]==0 || h->pps.constrained_intra_pred)){
548 h->topleft_samples_available&= 0xDF5F;
549 h->left_samples_available&= 0x5F5F;
553 if(!IS_INTRA(topleft_type) && (topleft_type==0 || h->pps.constrained_intra_pred))
554 h->topleft_samples_available&= 0x7FFF;
556 if(!IS_INTRA(topright_type) && (topright_type==0 || h->pps.constrained_intra_pred))
557 h->topright_samples_available&= 0xFBFF;
559 if(IS_INTRA4x4(mb_type)){
560 if(IS_INTRA4x4(top_type)){
561 h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
562 h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
563 h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
564 h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
567 if(!top_type || (IS_INTER(top_type) && h->pps.constrained_intra_pred))
572 h->intra4x4_pred_mode_cache[4+8*0]=
573 h->intra4x4_pred_mode_cache[5+8*0]=
574 h->intra4x4_pred_mode_cache[6+8*0]=
575 h->intra4x4_pred_mode_cache[7+8*0]= pred;
578 if(IS_INTRA4x4(left_type[i])){
579 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
580 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
583 if(!left_type[i] || (IS_INTER(left_type[i]) && h->pps.constrained_intra_pred))
588 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
589 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
604 //FIXME constraint_intra_pred & partitioning & nnz (lets hope this is just a typo in the spec)
606 h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][4];
607 h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][5];
608 h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][6];
609 h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3];
611 h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][9];
612 h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8];
614 h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][12];
615 h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11];
618 h->non_zero_count_cache[4+8*0]=
619 h->non_zero_count_cache[5+8*0]=
620 h->non_zero_count_cache[6+8*0]=
621 h->non_zero_count_cache[7+8*0]=
623 h->non_zero_count_cache[1+8*0]=
624 h->non_zero_count_cache[2+8*0]=
626 h->non_zero_count_cache[1+8*3]=
627 h->non_zero_count_cache[2+8*3]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
631 for (i=0; i<2; i++) {
633 h->non_zero_count_cache[3+8*1 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[0+2*i]];
634 h->non_zero_count_cache[3+8*2 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[1+2*i]];
635 h->non_zero_count_cache[0+8*1 + 8*i]= h->non_zero_count[left_xy[i]][left_block[4+2*i]];
636 h->non_zero_count_cache[0+8*4 + 8*i]= h->non_zero_count[left_xy[i]][left_block[5+2*i]];
638 h->non_zero_count_cache[3+8*1 + 2*8*i]=
639 h->non_zero_count_cache[3+8*2 + 2*8*i]=
640 h->non_zero_count_cache[0+8*1 + 8*i]=
641 h->non_zero_count_cache[0+8*4 + 8*i]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
648 h->top_cbp = h->cbp_table[top_xy];
649 } else if(IS_INTRA(mb_type)) {
656 h->left_cbp = h->cbp_table[left_xy[0]] & 0x1f0;
657 } else if(IS_INTRA(mb_type)) {
663 h->left_cbp |= ((h->cbp_table[left_xy[0]]>>((left_block[0]&(~1))+1))&0x1) << 1;
666 h->left_cbp |= ((h->cbp_table[left_xy[1]]>>((left_block[2]&(~1))+1))&0x1) << 3;
671 //FIXME direct mb can skip much of this
672 if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
674 for(list=0; list<1+(h->slice_type==B_TYPE); list++){
675 if(!USES_LIST(mb_type, list) && !IS_DIRECT(mb_type) && !h->deblocking_filter){
676 /*if(!h->mv_cache_clean[list]){
677 memset(h->mv_cache [list], 0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
678 memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
679 h->mv_cache_clean[list]= 1;
683 h->mv_cache_clean[list]= 0;
685 if(IS_INTER(top_type)){
686 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
687 const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
688 *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0];
689 *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];
690 *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];
691 *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];
692 h->ref_cache[list][scan8[0] + 0 - 1*8]=
693 h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
694 h->ref_cache[list][scan8[0] + 2 - 1*8]=
695 h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
697 *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]=
698 *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]=
699 *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]=
700 *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;
701 *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
704 //FIXME unify cleanup or sth
705 if(IS_INTER(left_type[0])){
706 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
707 const int b8_xy= h->mb2b8_xy[left_xy[0]] + 1;
708 *(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]];
709 *(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]];
710 h->ref_cache[list][scan8[0] - 1 + 0*8]=
711 h->ref_cache[list][scan8[0] - 1 + 1*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0]>>1)];
713 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 0*8]=
714 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 1*8]= 0;
715 h->ref_cache[list][scan8[0] - 1 + 0*8]=
716 h->ref_cache[list][scan8[0] - 1 + 1*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
719 if(IS_INTER(left_type[1])){
720 const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
721 const int b8_xy= h->mb2b8_xy[left_xy[1]] + 1;
722 *(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]];
723 *(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]];
724 h->ref_cache[list][scan8[0] - 1 + 2*8]=
725 h->ref_cache[list][scan8[0] - 1 + 3*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[2]>>1)];
727 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 2*8]=
728 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 3*8]= 0;
729 h->ref_cache[list][scan8[0] - 1 + 2*8]=
730 h->ref_cache[list][scan8[0] - 1 + 3*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
731 assert((!left_type[0]) == (!left_type[1]));
734 if(for_deblock || (IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred))
737 if(IS_INTER(topleft_type)){
738 const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
739 const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + h->b8_stride;
740 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
741 h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
743 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
744 h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
747 if(IS_INTER(topright_type)){
748 const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
749 const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
750 *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
751 h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
753 *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
754 h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
758 h->ref_cache[list][scan8[5 ]+1] =
759 h->ref_cache[list][scan8[7 ]+1] =
760 h->ref_cache[list][scan8[13]+1] = //FIXME remove past 3 (init somewhere else)
761 h->ref_cache[list][scan8[4 ]] =
762 h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
763 *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
764 *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
765 *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
766 *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
767 *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
770 /* XXX beurk, Load mvd */
771 if(IS_INTER(topleft_type)){
772 const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
773 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy];
775 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 - 1*8]= 0;
778 if(IS_INTER(top_type)){
779 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
780 *(uint32_t*)h->mvd_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 0];
781 *(uint32_t*)h->mvd_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 1];
782 *(uint32_t*)h->mvd_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 2];
783 *(uint32_t*)h->mvd_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 3];
785 *(uint32_t*)h->mvd_cache [list][scan8[0] + 0 - 1*8]=
786 *(uint32_t*)h->mvd_cache [list][scan8[0] + 1 - 1*8]=
787 *(uint32_t*)h->mvd_cache [list][scan8[0] + 2 - 1*8]=
788 *(uint32_t*)h->mvd_cache [list][scan8[0] + 3 - 1*8]= 0;
790 if(IS_INTER(left_type[0])){
791 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
792 *(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]];
793 *(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]];
795 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 0*8]=
796 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 1*8]= 0;
798 if(IS_INTER(left_type[1])){
799 const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
800 *(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]];
801 *(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]];
803 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 2*8]=
804 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 3*8]= 0;
806 *(uint32_t*)h->mvd_cache [list][scan8[5 ]+1]=
807 *(uint32_t*)h->mvd_cache [list][scan8[7 ]+1]=
808 *(uint32_t*)h->mvd_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
809 *(uint32_t*)h->mvd_cache [list][scan8[4 ]]=
810 *(uint32_t*)h->mvd_cache [list][scan8[12]]= 0;
812 if(h->slice_type == B_TYPE){
813 fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, 0, 1);
815 if(IS_DIRECT(top_type)){
816 *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0x01010101;
817 }else if(IS_8X8(top_type)){
818 int b8_xy = h->mb2b8_xy[top_xy] + h->b8_stride;
819 h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy];
820 h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 1];
822 *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0;
826 if(IS_DIRECT(left_type[0])){
827 h->direct_cache[scan8[0] - 1 + 0*8]=
828 h->direct_cache[scan8[0] - 1 + 2*8]= 1;
829 }else if(IS_8X8(left_type[0])){
830 int b8_xy = h->mb2b8_xy[left_xy[0]] + 1;
831 h->direct_cache[scan8[0] - 1 + 0*8]= h->direct_table[b8_xy];
832 h->direct_cache[scan8[0] - 1 + 2*8]= h->direct_table[b8_xy + h->b8_stride];
834 h->direct_cache[scan8[0] - 1 + 0*8]=
835 h->direct_cache[scan8[0] - 1 + 2*8]= 0;
844 static inline void write_back_intra_pred_mode(H264Context *h){
845 MpegEncContext * const s = &h->s;
846 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
848 h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
849 h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
850 h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
851 h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
852 h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
853 h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
854 h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
858 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
860 static inline int check_intra4x4_pred_mode(H264Context *h){
861 MpegEncContext * const s = &h->s;
862 static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
863 static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
866 if(!(h->top_samples_available&0x8000)){
868 int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
870 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);
873 h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
878 if(!(h->left_samples_available&0x8000)){
880 int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
882 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);
885 h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
891 } //FIXME cleanup like next
894 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
896 static inline int check_intra_pred_mode(H264Context *h, int mode){
897 MpegEncContext * const s = &h->s;
898 static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
899 static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
901 if(mode < 0 || mode > 6) {
902 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);
906 if(!(h->top_samples_available&0x8000)){
909 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);
914 if(!(h->left_samples_available&0x8000)){
917 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);
926 * gets the predicted intra4x4 prediction mode.
928 static inline int pred_intra_mode(H264Context *h, int n){
929 const int index8= scan8[n];
930 const int left= h->intra4x4_pred_mode_cache[index8 - 1];
931 const int top = h->intra4x4_pred_mode_cache[index8 - 8];
932 const int min= FFMIN(left, top);
934 tprintf("mode:%d %d min:%d\n", left ,top, min);
936 if(min<0) return DC_PRED;
940 static inline void write_back_non_zero_count(H264Context *h){
941 MpegEncContext * const s = &h->s;
942 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
944 h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[7+8*1];
945 h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[7+8*2];
946 h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[7+8*3];
947 h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];
948 h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[4+8*4];
949 h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[5+8*4];
950 h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[6+8*4];
952 h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[1+8*2];
953 h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
954 h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[2+8*1];
956 h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[1+8*5];
957 h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
958 h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[2+8*4];
962 * gets the predicted number of non zero coefficients.
963 * @param n block index
965 static inline int pred_non_zero_count(H264Context *h, int n){
966 const int index8= scan8[n];
967 const int left= h->non_zero_count_cache[index8 - 1];
968 const int top = h->non_zero_count_cache[index8 - 8];
971 if(i<64) i= (i+1)>>1;
973 tprintf("pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
978 static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
979 const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];
981 if(topright_ref != PART_NOT_AVAILABLE){
982 *C= h->mv_cache[list][ i - 8 + part_width ];
985 tprintf("topright MV not available\n");
987 *C= h->mv_cache[list][ i - 8 - 1 ];
988 return h->ref_cache[list][ i - 8 - 1 ];
993 * gets the predicted MV.
994 * @param n the block index
995 * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)
996 * @param mx the x component of the predicted motion vector
997 * @param my the y component of the predicted motion vector
999 static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
1000 const int index8= scan8[n];
1001 const int top_ref= h->ref_cache[list][ index8 - 8 ];
1002 const int left_ref= h->ref_cache[list][ index8 - 1 ];
1003 const int16_t * const A= h->mv_cache[list][ index8 - 1 ];
1004 const int16_t * const B= h->mv_cache[list][ index8 - 8 ];
1006 int diagonal_ref, match_count;
1008 assert(part_width==1 || part_width==2 || part_width==4);
1018 diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);
1019 match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);
1020 tprintf("pred_motion match_count=%d\n", match_count);
1021 if(match_count > 1){ //most common
1022 *mx= mid_pred(A[0], B[0], C[0]);
1023 *my= mid_pred(A[1], B[1], C[1]);
1024 }else if(match_count==1){
1028 }else if(top_ref==ref){
1036 if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
1040 *mx= mid_pred(A[0], B[0], C[0]);
1041 *my= mid_pred(A[1], B[1], C[1]);
1045 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);
1049 * gets the directionally predicted 16x8 MV.
1050 * @param n the block index
1051 * @param mx the x component of the predicted motion vector
1052 * @param my the y component of the predicted motion vector
1054 static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
1056 const int top_ref= h->ref_cache[list][ scan8[0] - 8 ];
1057 const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
1059 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);
1067 const int left_ref= h->ref_cache[list][ scan8[8] - 1 ];
1068 const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];
1070 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);
1072 if(left_ref == ref){
1080 pred_motion(h, n, 4, list, ref, mx, my);
1084 * gets the directionally predicted 8x16 MV.
1085 * @param n the block index
1086 * @param mx the x component of the predicted motion vector
1087 * @param my the y component of the predicted motion vector
1089 static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
1091 const int left_ref= h->ref_cache[list][ scan8[0] - 1 ];
1092 const int16_t * const A= h->mv_cache[list][ scan8[0] - 1 ];
1094 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);
1096 if(left_ref == ref){
1105 diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);
1107 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);
1109 if(diagonal_ref == ref){
1117 pred_motion(h, n, 2, list, ref, mx, my);
1120 static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
1121 const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
1122 const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];
1124 tprintf("pred_pskip: (%d) (%d) at %2d %2d\n", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
1126 if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
1127 || (top_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0)
1128 || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){
1134 pred_motion(h, 0, 4, 0, 0, mx, my);
1139 static inline void direct_dist_scale_factor(H264Context * const h){
1140 const int poc = h->s.current_picture_ptr->poc;
1141 const int poc1 = h->ref_list[1][0].poc;
1143 for(i=0; i<h->ref_count[0]; i++){
1144 int poc0 = h->ref_list[0][i].poc;
1145 int td = clip(poc1 - poc0, -128, 127);
1146 if(td == 0 /* FIXME || pic0 is a long-term ref */){
1147 h->dist_scale_factor[i] = 256;
1149 int tb = clip(poc - poc0, -128, 127);
1150 int tx = (16384 + (ABS(td) >> 1)) / td;
1151 h->dist_scale_factor[i] = clip((tb*tx + 32) >> 6, -1024, 1023);
1155 static inline void direct_ref_list_init(H264Context * const h){
1156 MpegEncContext * const s = &h->s;
1157 Picture * const ref1 = &h->ref_list[1][0];
1158 Picture * const cur = s->current_picture_ptr;
1160 if(cur->pict_type == I_TYPE)
1161 cur->ref_count[0] = 0;
1162 if(cur->pict_type != B_TYPE)
1163 cur->ref_count[1] = 0;
1164 for(list=0; list<2; list++){
1165 cur->ref_count[list] = h->ref_count[list];
1166 for(j=0; j<h->ref_count[list]; j++)
1167 cur->ref_poc[list][j] = h->ref_list[list][j].poc;
1169 if(cur->pict_type != B_TYPE || h->direct_spatial_mv_pred)
1171 for(list=0; list<2; list++){
1172 for(i=0; i<ref1->ref_count[list]; i++){
1173 const int poc = ref1->ref_poc[list][i];
1174 h->map_col_to_list0[list][i] = PART_NOT_AVAILABLE;
1175 for(j=0; j<h->ref_count[list]; j++)
1176 if(h->ref_list[list][j].poc == poc){
1177 h->map_col_to_list0[list][i] = j;
1184 static inline void pred_direct_motion(H264Context * const h, int *mb_type){
1185 MpegEncContext * const s = &h->s;
1186 const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
1187 const int b8_xy = 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1188 const int b4_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1189 const int mb_type_col = h->ref_list[1][0].mb_type[mb_xy];
1190 const int16_t (*l1mv0)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[0][b4_xy];
1191 const int8_t *l1ref0 = &h->ref_list[1][0].ref_index[0][b8_xy];
1192 const int8_t *l1ref1 = &h->ref_list[1][0].ref_index[1][b8_xy];
1193 const int is_b8x8 = IS_8X8(*mb_type);
1197 if(IS_8X8(mb_type_col) && !h->sps.direct_8x8_inference_flag){
1198 /* FIXME save sub mb types from previous frames (or derive from MVs)
1199 * so we know exactly what block size to use */
1200 sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */
1201 *mb_type = MB_TYPE_8x8|MB_TYPE_L0L1;
1202 }else if(!is_b8x8 && (IS_16X16(mb_type_col) || IS_INTRA(mb_type_col))){
1203 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1204 *mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */
1206 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1207 *mb_type = MB_TYPE_8x8|MB_TYPE_L0L1;
1210 *mb_type |= MB_TYPE_DIRECT2;
1212 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);
1214 if(h->direct_spatial_mv_pred){
1219 /* ref = min(neighbors) */
1220 for(list=0; list<2; list++){
1221 int refa = h->ref_cache[list][scan8[0] - 1];
1222 int refb = h->ref_cache[list][scan8[0] - 8];
1223 int refc = h->ref_cache[list][scan8[0] - 8 + 4];
1225 refc = h->ref_cache[list][scan8[0] - 8 - 1];
1227 if(ref[list] < 0 || (refb < ref[list] && refb >= 0))
1229 if(ref[list] < 0 || (refc < ref[list] && refc >= 0))
1235 if(ref[0] < 0 && ref[1] < 0){
1236 ref[0] = ref[1] = 0;
1237 mv[0][0] = mv[0][1] =
1238 mv[1][0] = mv[1][1] = 0;
1240 for(list=0; list<2; list++){
1242 pred_motion(h, 0, 4, list, ref[list], &mv[list][0], &mv[list][1]);
1244 mv[list][0] = mv[list][1] = 0;
1249 *mb_type &= ~MB_TYPE_P0L1;
1250 sub_mb_type &= ~MB_TYPE_P0L1;
1251 }else if(ref[0] < 0){
1252 *mb_type &= ~MB_TYPE_P0L0;
1253 sub_mb_type &= ~MB_TYPE_P0L0;
1256 if(IS_16X16(*mb_type)){
1257 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref[0], 1);
1258 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, ref[1], 1);
1259 if(!IS_INTRA(mb_type_col) && l1ref0[0] == 0 &&
1260 ABS(l1mv0[0][0]) <= 1 && ABS(l1mv0[0][1]) <= 1){
1262 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1264 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
1266 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1268 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
1270 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1271 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1274 for(i8=0; i8<4; i8++){
1275 const int x8 = i8&1;
1276 const int y8 = i8>>1;
1278 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1280 h->sub_mb_type[i8] = sub_mb_type;
1282 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1283 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1284 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref[0], 1);
1285 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, ref[1], 1);
1288 if(!IS_INTRA(mb_type_col) && l1ref0[x8 + y8*h->b8_stride] == 0){
1289 for(i4=0; i4<4; i4++){
1290 const int16_t *mv_col = l1mv0[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1291 if(ABS(mv_col[0]) <= 1 && ABS(mv_col[1]) <= 1){
1293 *(uint32_t*)h->mv_cache[0][scan8[i8*4+i4]] = 0;
1295 *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = 0;
1301 }else{ /* direct temporal mv pred */
1302 if(IS_16X16(*mb_type)){
1303 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1);
1304 if(IS_INTRA(mb_type_col)){
1305 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
1306 fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
1307 fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
1309 const int ref0 = l1ref0[0] >= 0 ? h->map_col_to_list0[0][l1ref0[0]]
1310 : h->map_col_to_list0[1][l1ref1[0]];
1311 const int dist_scale_factor = h->dist_scale_factor[ref0];
1312 const int16_t *mv_col = l1mv0[0];
1314 mv_l0[0] = (dist_scale_factor * mv_col[0] + 128) >> 8;
1315 mv_l0[1] = (dist_scale_factor * mv_col[1] + 128) >> 8;
1316 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref0, 1);
1317 fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv_l0[0],mv_l0[1]), 4);
1318 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);
1321 for(i8=0; i8<4; i8++){
1322 const int x8 = i8&1;
1323 const int y8 = i8>>1;
1324 int ref0, dist_scale_factor;
1326 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1328 h->sub_mb_type[i8] = sub_mb_type;
1329 if(IS_INTRA(mb_type_col)){
1330 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1331 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1332 fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1333 fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1337 ref0 = l1ref0[x8 + y8*h->b8_stride];
1339 ref0 = h->map_col_to_list0[0][ref0];
1341 ref0 = h->map_col_to_list0[1][l1ref1[x8 + y8*h->b8_stride]];
1342 dist_scale_factor = h->dist_scale_factor[ref0];
1344 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1345 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1346 for(i4=0; i4<4; i4++){
1347 const int16_t *mv_col = l1mv0[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1348 int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]];
1349 mv_l0[0] = (dist_scale_factor * mv_col[0] + 128) >> 8;
1350 mv_l0[1] = (dist_scale_factor * mv_col[1] + 128) >> 8;
1351 *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] =
1352 pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
1359 static inline void write_back_motion(H264Context *h, int mb_type){
1360 MpegEncContext * const s = &h->s;
1361 const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1362 const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1365 for(list=0; list<2; list++){
1367 if(!USES_LIST(mb_type, list)){
1368 if(1){ //FIXME skip or never read if mb_type doesn't use it
1370 *(uint64_t*)s->current_picture.motion_val[list][b_xy + 0 + y*h->b_stride]=
1371 *(uint64_t*)s->current_picture.motion_val[list][b_xy + 2 + y*h->b_stride]= 0;
1373 if( h->pps.cabac ) {
1374 /* FIXME needed ? */
1376 *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]=
1377 *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= 0;
1381 s->current_picture.ref_index[list][b8_xy + 0 + y*h->b8_stride]=
1382 s->current_picture.ref_index[list][b8_xy + 1 + y*h->b8_stride]= LIST_NOT_USED;
1389 *(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];
1390 *(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];
1392 if( h->pps.cabac ) {
1394 *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+0 + 8*y];
1395 *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+2 + 8*y];
1399 s->current_picture.ref_index[list][b8_xy + 0 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+0 + 16*y];
1400 s->current_picture.ref_index[list][b8_xy + 1 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+2 + 16*y];
1404 if(h->slice_type == B_TYPE && h->pps.cabac){
1405 if(IS_8X8(mb_type)){
1406 h->direct_table[b8_xy+1+0*h->b8_stride] = IS_DIRECT(h->sub_mb_type[1]) ? 1 : 0;
1407 h->direct_table[b8_xy+0+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[2]) ? 1 : 0;
1408 h->direct_table[b8_xy+1+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[3]) ? 1 : 0;
1414 * Decodes a network abstraction layer unit.
1415 * @param consumed is the number of bytes used as input
1416 * @param length is the length of the array
1417 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp tailing?
1418 * @returns decoded bytes, might be src+1 if no escapes
1420 static uint8_t *decode_nal(H264Context *h, uint8_t *src, int *dst_length, int *consumed, int length){
1424 // src[0]&0x80; //forbidden bit
1425 h->nal_ref_idc= src[0]>>5;
1426 h->nal_unit_type= src[0]&0x1F;
1430 for(i=0; i<length; i++)
1431 printf("%2X ", src[i]);
1433 for(i=0; i+1<length; i+=2){
1434 if(src[i]) continue;
1435 if(i>0 && src[i-1]==0) i--;
1436 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1438 /* startcode, so we must be past the end */
1445 if(i>=length-1){ //no escaped 0
1446 *dst_length= length;
1447 *consumed= length+1; //+1 for the header
1451 h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length);
1452 dst= h->rbsp_buffer;
1454 //printf("decoding esc\n");
1457 //remove escapes (very rare 1:2^22)
1458 if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1459 if(src[si+2]==3){ //escape
1464 }else //next start code
1468 dst[di++]= src[si++];
1472 *consumed= si + 1;//+1 for the header
1473 //FIXME store exact number of bits in the getbitcontext (its needed for decoding)
1479 * @param src the data which should be escaped
1480 * @param dst the target buffer, dst+1 == src is allowed as a special case
1481 * @param length the length of the src data
1482 * @param dst_length the length of the dst array
1483 * @returns length of escaped data in bytes or -1 if an error occured
1485 static int encode_nal(H264Context *h, uint8_t *dst, uint8_t *src, int length, int dst_length){
1486 int i, escape_count, si, di;
1490 assert(dst_length>0);
1492 dst[0]= (h->nal_ref_idc<<5) + h->nal_unit_type;
1494 if(length==0) return 1;
1497 for(i=0; i<length; i+=2){
1498 if(src[i]) continue;
1499 if(i>0 && src[i-1]==0)
1501 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1507 if(escape_count==0){
1509 memcpy(dst+1, src, length);
1513 if(length + escape_count + 1> dst_length)
1516 //this should be damn rare (hopefully)
1518 h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length + escape_count);
1519 temp= h->rbsp_buffer;
1520 //printf("encoding esc\n");
1525 if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1526 temp[di++]= 0; si++;
1527 temp[di++]= 0; si++;
1529 temp[di++]= src[si++];
1532 temp[di++]= src[si++];
1534 memcpy(dst+1, temp, length+escape_count);
1536 assert(di == length+escape_count);
1542 * write 1,10,100,1000,... for alignment, yes its exactly inverse to mpeg4
1544 static void encode_rbsp_trailing(PutBitContext *pb){
1547 length= (-put_bits_count(pb))&7;
1548 if(length) put_bits(pb, length, 0);
1553 * identifies the exact end of the bitstream
1554 * @return the length of the trailing, or 0 if damaged
1556 static int decode_rbsp_trailing(uint8_t *src){
1560 tprintf("rbsp trailing %X\n", v);
1570 * idct tranforms the 16 dc values and dequantize them.
1571 * @param qp quantization parameter
1573 static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp){
1574 const int qmul= dequant_coeff[qp][0];
1577 int temp[16]; //FIXME check if this is a good idea
1578 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
1579 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1581 //memset(block, 64, 2*256);
1584 const int offset= y_offset[i];
1585 const int z0= block[offset+stride*0] + block[offset+stride*4];
1586 const int z1= block[offset+stride*0] - block[offset+stride*4];
1587 const int z2= block[offset+stride*1] - block[offset+stride*5];
1588 const int z3= block[offset+stride*1] + block[offset+stride*5];
1597 const int offset= x_offset[i];
1598 const int z0= temp[4*0+i] + temp[4*2+i];
1599 const int z1= temp[4*0+i] - temp[4*2+i];
1600 const int z2= temp[4*1+i] - temp[4*3+i];
1601 const int z3= temp[4*1+i] + temp[4*3+i];
1603 block[stride*0 +offset]= ((z0 + z3)*qmul + 2)>>2; //FIXME think about merging this into decode_resdual
1604 block[stride*2 +offset]= ((z1 + z2)*qmul + 2)>>2;
1605 block[stride*8 +offset]= ((z1 - z2)*qmul + 2)>>2;
1606 block[stride*10+offset]= ((z0 - z3)*qmul + 2)>>2;
1612 * dct tranforms the 16 dc values.
1613 * @param qp quantization parameter ??? FIXME
1615 static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
1616 // const int qmul= dequant_coeff[qp][0];
1618 int temp[16]; //FIXME check if this is a good idea
1619 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
1620 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1623 const int offset= y_offset[i];
1624 const int z0= block[offset+stride*0] + block[offset+stride*4];
1625 const int z1= block[offset+stride*0] - block[offset+stride*4];
1626 const int z2= block[offset+stride*1] - block[offset+stride*5];
1627 const int z3= block[offset+stride*1] + block[offset+stride*5];
1636 const int offset= x_offset[i];
1637 const int z0= temp[4*0+i] + temp[4*2+i];
1638 const int z1= temp[4*0+i] - temp[4*2+i];
1639 const int z2= temp[4*1+i] - temp[4*3+i];
1640 const int z3= temp[4*1+i] + temp[4*3+i];
1642 block[stride*0 +offset]= (z0 + z3)>>1;
1643 block[stride*2 +offset]= (z1 + z2)>>1;
1644 block[stride*8 +offset]= (z1 - z2)>>1;
1645 block[stride*10+offset]= (z0 - z3)>>1;
1653 static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp){
1654 const int qmul= dequant_coeff[qp][0];
1655 const int stride= 16*2;
1656 const int xStride= 16;
1659 a= block[stride*0 + xStride*0];
1660 b= block[stride*0 + xStride*1];
1661 c= block[stride*1 + xStride*0];
1662 d= block[stride*1 + xStride*1];
1669 block[stride*0 + xStride*0]= ((a+c)*qmul + 0)>>1;
1670 block[stride*0 + xStride*1]= ((e+b)*qmul + 0)>>1;
1671 block[stride*1 + xStride*0]= ((a-c)*qmul + 0)>>1;
1672 block[stride*1 + xStride*1]= ((e-b)*qmul + 0)>>1;
1676 static void chroma_dc_dct_c(DCTELEM *block){
1677 const int stride= 16*2;
1678 const int xStride= 16;
1681 a= block[stride*0 + xStride*0];
1682 b= block[stride*0 + xStride*1];
1683 c= block[stride*1 + xStride*0];
1684 d= block[stride*1 + xStride*1];
1691 block[stride*0 + xStride*0]= (a+c);
1692 block[stride*0 + xStride*1]= (e+b);
1693 block[stride*1 + xStride*0]= (a-c);
1694 block[stride*1 + xStride*1]= (e-b);
1699 * gets the chroma qp.
1701 static inline int get_chroma_qp(int chroma_qp_index_offset, int qscale){
1703 return chroma_qp[clip(qscale + chroma_qp_index_offset, 0, 51)];
1708 static void h264_diff_dct_c(DCTELEM *block, uint8_t *src1, uint8_t *src2, int stride){
1710 //FIXME try int temp instead of block
1713 const int d0= src1[0 + i*stride] - src2[0 + i*stride];
1714 const int d1= src1[1 + i*stride] - src2[1 + i*stride];
1715 const int d2= src1[2 + i*stride] - src2[2 + i*stride];
1716 const int d3= src1[3 + i*stride] - src2[3 + i*stride];
1717 const int z0= d0 + d3;
1718 const int z3= d0 - d3;
1719 const int z1= d1 + d2;
1720 const int z2= d1 - d2;
1722 block[0 + 4*i]= z0 + z1;
1723 block[1 + 4*i]= 2*z3 + z2;
1724 block[2 + 4*i]= z0 - z1;
1725 block[3 + 4*i]= z3 - 2*z2;
1729 const int z0= block[0*4 + i] + block[3*4 + i];
1730 const int z3= block[0*4 + i] - block[3*4 + i];
1731 const int z1= block[1*4 + i] + block[2*4 + i];
1732 const int z2= block[1*4 + i] - block[2*4 + i];
1734 block[0*4 + i]= z0 + z1;
1735 block[1*4 + i]= 2*z3 + z2;
1736 block[2*4 + i]= z0 - z1;
1737 block[3*4 + i]= z3 - 2*z2;
1742 //FIXME need to check that this doesnt overflow signed 32 bit for low qp, i am not sure, it's very close
1743 //FIXME check that gcc inlines this (and optimizes intra & seperate_dc stuff away)
1744 static inline int quantize_c(DCTELEM *block, uint8_t *scantable, int qscale, int intra, int seperate_dc){
1746 const int * const quant_table= quant_coeff[qscale];
1747 const int bias= intra ? (1<<QUANT_SHIFT)/3 : (1<<QUANT_SHIFT)/6;
1748 const unsigned int threshold1= (1<<QUANT_SHIFT) - bias - 1;
1749 const unsigned int threshold2= (threshold1<<1);
1755 const int dc_bias= intra ? (1<<(QUANT_SHIFT-2))/3 : (1<<(QUANT_SHIFT-2))/6;
1756 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT-2)) - dc_bias - 1;
1757 const unsigned int dc_threshold2= (dc_threshold1<<1);
1759 int level= block[0]*quant_coeff[qscale+18][0];
1760 if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1762 level= (dc_bias + level)>>(QUANT_SHIFT-2);
1765 level= (dc_bias - level)>>(QUANT_SHIFT-2);
1768 // last_non_zero = i;
1773 const int dc_bias= intra ? (1<<(QUANT_SHIFT+1))/3 : (1<<(QUANT_SHIFT+1))/6;
1774 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT+1)) - dc_bias - 1;
1775 const unsigned int dc_threshold2= (dc_threshold1<<1);
1777 int level= block[0]*quant_table[0];
1778 if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1780 level= (dc_bias + level)>>(QUANT_SHIFT+1);
1783 level= (dc_bias - level)>>(QUANT_SHIFT+1);
1786 // last_non_zero = i;
1799 const int j= scantable[i];
1800 int level= block[j]*quant_table[j];
1802 // if( bias+level >= (1<<(QMAT_SHIFT - 3))
1803 // || bias-level >= (1<<(QMAT_SHIFT - 3))){
1804 if(((unsigned)(level+threshold1))>threshold2){
1806 level= (bias + level)>>QUANT_SHIFT;
1809 level= (bias - level)>>QUANT_SHIFT;
1818 return last_non_zero;
1821 static void pred4x4_vertical_c(uint8_t *src, uint8_t *topright, int stride){
1822 const uint32_t a= ((uint32_t*)(src-stride))[0];
1823 ((uint32_t*)(src+0*stride))[0]= a;
1824 ((uint32_t*)(src+1*stride))[0]= a;
1825 ((uint32_t*)(src+2*stride))[0]= a;
1826 ((uint32_t*)(src+3*stride))[0]= a;
1829 static void pred4x4_horizontal_c(uint8_t *src, uint8_t *topright, int stride){
1830 ((uint32_t*)(src+0*stride))[0]= src[-1+0*stride]*0x01010101;
1831 ((uint32_t*)(src+1*stride))[0]= src[-1+1*stride]*0x01010101;
1832 ((uint32_t*)(src+2*stride))[0]= src[-1+2*stride]*0x01010101;
1833 ((uint32_t*)(src+3*stride))[0]= src[-1+3*stride]*0x01010101;
1836 static void pred4x4_dc_c(uint8_t *src, uint8_t *topright, int stride){
1837 const int dc= ( src[-stride] + src[1-stride] + src[2-stride] + src[3-stride]
1838 + src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 4) >>3;
1840 ((uint32_t*)(src+0*stride))[0]=
1841 ((uint32_t*)(src+1*stride))[0]=
1842 ((uint32_t*)(src+2*stride))[0]=
1843 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1846 static void pred4x4_left_dc_c(uint8_t *src, uint8_t *topright, int stride){
1847 const int dc= ( src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 2) >>2;
1849 ((uint32_t*)(src+0*stride))[0]=
1850 ((uint32_t*)(src+1*stride))[0]=
1851 ((uint32_t*)(src+2*stride))[0]=
1852 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1855 static void pred4x4_top_dc_c(uint8_t *src, uint8_t *topright, int stride){
1856 const int dc= ( src[-stride] + src[1-stride] + src[2-stride] + src[3-stride] + 2) >>2;
1858 ((uint32_t*)(src+0*stride))[0]=
1859 ((uint32_t*)(src+1*stride))[0]=
1860 ((uint32_t*)(src+2*stride))[0]=
1861 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1864 static void pred4x4_128_dc_c(uint8_t *src, uint8_t *topright, int stride){
1865 ((uint32_t*)(src+0*stride))[0]=
1866 ((uint32_t*)(src+1*stride))[0]=
1867 ((uint32_t*)(src+2*stride))[0]=
1868 ((uint32_t*)(src+3*stride))[0]= 128U*0x01010101U;
1872 #define LOAD_TOP_RIGHT_EDGE\
1873 const int t4= topright[0];\
1874 const int t5= topright[1];\
1875 const int t6= topright[2];\
1876 const int t7= topright[3];\
1878 #define LOAD_LEFT_EDGE\
1879 const int l0= src[-1+0*stride];\
1880 const int l1= src[-1+1*stride];\
1881 const int l2= src[-1+2*stride];\
1882 const int l3= src[-1+3*stride];\
1884 #define LOAD_TOP_EDGE\
1885 const int t0= src[ 0-1*stride];\
1886 const int t1= src[ 1-1*stride];\
1887 const int t2= src[ 2-1*stride];\
1888 const int t3= src[ 3-1*stride];\
1890 static void pred4x4_down_right_c(uint8_t *src, uint8_t *topright, int stride){
1891 const int lt= src[-1-1*stride];
1895 src[0+3*stride]=(l3 + 2*l2 + l1 + 2)>>2;
1897 src[1+3*stride]=(l2 + 2*l1 + l0 + 2)>>2;
1900 src[2+3*stride]=(l1 + 2*l0 + lt + 2)>>2;
1904 src[3+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1907 src[3+2*stride]=(lt + 2*t0 + t1 + 2)>>2;
1909 src[3+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1910 src[3+0*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1913 static void pred4x4_down_left_c(uint8_t *src, uint8_t *topright, int stride){
1918 src[0+0*stride]=(t0 + t2 + 2*t1 + 2)>>2;
1920 src[0+1*stride]=(t1 + t3 + 2*t2 + 2)>>2;
1923 src[0+2*stride]=(t2 + t4 + 2*t3 + 2)>>2;
1927 src[0+3*stride]=(t3 + t5 + 2*t4 + 2)>>2;
1930 src[1+3*stride]=(t4 + t6 + 2*t5 + 2)>>2;
1932 src[2+3*stride]=(t5 + t7 + 2*t6 + 2)>>2;
1933 src[3+3*stride]=(t6 + 3*t7 + 2)>>2;
1936 static void pred4x4_vertical_right_c(uint8_t *src, uint8_t *topright, int stride){
1937 const int lt= src[-1-1*stride];
1940 const __attribute__((unused)) int unu= l3;
1943 src[1+2*stride]=(lt + t0 + 1)>>1;
1945 src[2+2*stride]=(t0 + t1 + 1)>>1;
1947 src[3+2*stride]=(t1 + t2 + 1)>>1;
1948 src[3+0*stride]=(t2 + t3 + 1)>>1;
1950 src[1+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1952 src[2+3*stride]=(lt + 2*t0 + t1 + 2)>>2;
1954 src[3+3*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1955 src[3+1*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1956 src[0+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1957 src[0+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1960 static void pred4x4_vertical_left_c(uint8_t *src, uint8_t *topright, int stride){
1963 const __attribute__((unused)) int unu= t7;
1965 src[0+0*stride]=(t0 + t1 + 1)>>1;
1967 src[0+2*stride]=(t1 + t2 + 1)>>1;
1969 src[1+2*stride]=(t2 + t3 + 1)>>1;
1971 src[2+2*stride]=(t3 + t4+ 1)>>1;
1972 src[3+2*stride]=(t4 + t5+ 1)>>1;
1973 src[0+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1975 src[0+3*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1977 src[1+3*stride]=(t2 + 2*t3 + t4 + 2)>>2;
1979 src[2+3*stride]=(t3 + 2*t4 + t5 + 2)>>2;
1980 src[3+3*stride]=(t4 + 2*t5 + t6 + 2)>>2;
1983 static void pred4x4_horizontal_up_c(uint8_t *src, uint8_t *topright, int stride){
1986 src[0+0*stride]=(l0 + l1 + 1)>>1;
1987 src[1+0*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1989 src[0+1*stride]=(l1 + l2 + 1)>>1;
1991 src[1+1*stride]=(l1 + 2*l2 + l3 + 2)>>2;
1993 src[0+2*stride]=(l2 + l3 + 1)>>1;
1995 src[1+2*stride]=(l2 + 2*l3 + l3 + 2)>>2;
2004 static void pred4x4_horizontal_down_c(uint8_t *src, uint8_t *topright, int stride){
2005 const int lt= src[-1-1*stride];
2008 const __attribute__((unused)) int unu= t3;
2011 src[2+1*stride]=(lt + l0 + 1)>>1;
2013 src[3+1*stride]=(l0 + 2*lt + t0 + 2)>>2;
2014 src[2+0*stride]=(lt + 2*t0 + t1 + 2)>>2;
2015 src[3+0*stride]=(t0 + 2*t1 + t2 + 2)>>2;
2017 src[2+2*stride]=(l0 + l1 + 1)>>1;
2019 src[3+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
2021 src[2+3*stride]=(l1 + l2+ 1)>>1;
2023 src[3+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
2024 src[0+3*stride]=(l2 + l3 + 1)>>1;
2025 src[1+3*stride]=(l1 + 2*l2 + l3 + 2)>>2;
2028 static void pred16x16_vertical_c(uint8_t *src, int stride){
2030 const uint32_t a= ((uint32_t*)(src-stride))[0];
2031 const uint32_t b= ((uint32_t*)(src-stride))[1];
2032 const uint32_t c= ((uint32_t*)(src-stride))[2];
2033 const uint32_t d= ((uint32_t*)(src-stride))[3];
2035 for(i=0; i<16; i++){
2036 ((uint32_t*)(src+i*stride))[0]= a;
2037 ((uint32_t*)(src+i*stride))[1]= b;
2038 ((uint32_t*)(src+i*stride))[2]= c;
2039 ((uint32_t*)(src+i*stride))[3]= d;
2043 static void pred16x16_horizontal_c(uint8_t *src, int stride){
2046 for(i=0; i<16; i++){
2047 ((uint32_t*)(src+i*stride))[0]=
2048 ((uint32_t*)(src+i*stride))[1]=
2049 ((uint32_t*)(src+i*stride))[2]=
2050 ((uint32_t*)(src+i*stride))[3]= src[-1+i*stride]*0x01010101;
2054 static void pred16x16_dc_c(uint8_t *src, int stride){
2058 dc+= src[-1+i*stride];
2065 dc= 0x01010101*((dc + 16)>>5);
2067 for(i=0; i<16; i++){
2068 ((uint32_t*)(src+i*stride))[0]=
2069 ((uint32_t*)(src+i*stride))[1]=
2070 ((uint32_t*)(src+i*stride))[2]=
2071 ((uint32_t*)(src+i*stride))[3]= dc;
2075 static void pred16x16_left_dc_c(uint8_t *src, int stride){
2079 dc+= src[-1+i*stride];
2082 dc= 0x01010101*((dc + 8)>>4);
2084 for(i=0; i<16; i++){
2085 ((uint32_t*)(src+i*stride))[0]=
2086 ((uint32_t*)(src+i*stride))[1]=
2087 ((uint32_t*)(src+i*stride))[2]=
2088 ((uint32_t*)(src+i*stride))[3]= dc;
2092 static void pred16x16_top_dc_c(uint8_t *src, int stride){
2098 dc= 0x01010101*((dc + 8)>>4);
2100 for(i=0; i<16; i++){
2101 ((uint32_t*)(src+i*stride))[0]=
2102 ((uint32_t*)(src+i*stride))[1]=
2103 ((uint32_t*)(src+i*stride))[2]=
2104 ((uint32_t*)(src+i*stride))[3]= dc;
2108 static void pred16x16_128_dc_c(uint8_t *src, int stride){
2111 for(i=0; i<16; i++){
2112 ((uint32_t*)(src+i*stride))[0]=
2113 ((uint32_t*)(src+i*stride))[1]=
2114 ((uint32_t*)(src+i*stride))[2]=
2115 ((uint32_t*)(src+i*stride))[3]= 0x01010101U*128U;
2119 static inline void pred16x16_plane_compat_c(uint8_t *src, int stride, const int svq3){
2122 uint8_t *cm = cropTbl + MAX_NEG_CROP;
2123 const uint8_t * const src0 = src+7-stride;
2124 const uint8_t *src1 = src+8*stride-1;
2125 const uint8_t *src2 = src1-2*stride; // == src+6*stride-1;
2126 int H = src0[1] - src0[-1];
2127 int V = src1[0] - src2[ 0];
2128 for(k=2; k<=8; ++k) {
2129 src1 += stride; src2 -= stride;
2130 H += k*(src0[k] - src0[-k]);
2131 V += k*(src1[0] - src2[ 0]);
2134 H = ( 5*(H/4) ) / 16;
2135 V = ( 5*(V/4) ) / 16;
2137 /* required for 100% accuracy */
2138 i = H; H = V; V = i;
2140 H = ( 5*H+32 ) >> 6;
2141 V = ( 5*V+32 ) >> 6;
2144 a = 16*(src1[0] + src2[16] + 1) - 7*(V+H);
2145 for(j=16; j>0; --j) {
2148 for(i=-16; i<0; i+=4) {
2149 src[16+i] = cm[ (b ) >> 5 ];
2150 src[17+i] = cm[ (b+ H) >> 5 ];
2151 src[18+i] = cm[ (b+2*H) >> 5 ];
2152 src[19+i] = cm[ (b+3*H) >> 5 ];
2159 static void pred16x16_plane_c(uint8_t *src, int stride){
2160 pred16x16_plane_compat_c(src, stride, 0);
2163 static void pred8x8_vertical_c(uint8_t *src, int stride){
2165 const uint32_t a= ((uint32_t*)(src-stride))[0];
2166 const uint32_t b= ((uint32_t*)(src-stride))[1];
2169 ((uint32_t*)(src+i*stride))[0]= a;
2170 ((uint32_t*)(src+i*stride))[1]= b;
2174 static void pred8x8_horizontal_c(uint8_t *src, int stride){
2178 ((uint32_t*)(src+i*stride))[0]=
2179 ((uint32_t*)(src+i*stride))[1]= src[-1+i*stride]*0x01010101;
2183 static void pred8x8_128_dc_c(uint8_t *src, int stride){
2187 ((uint32_t*)(src+i*stride))[0]=
2188 ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
2191 ((uint32_t*)(src+i*stride))[0]=
2192 ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
2196 static void pred8x8_left_dc_c(uint8_t *src, int stride){
2202 dc0+= src[-1+i*stride];
2203 dc2+= src[-1+(i+4)*stride];
2205 dc0= 0x01010101*((dc0 + 2)>>2);
2206 dc2= 0x01010101*((dc2 + 2)>>2);
2209 ((uint32_t*)(src+i*stride))[0]=
2210 ((uint32_t*)(src+i*stride))[1]= dc0;
2213 ((uint32_t*)(src+i*stride))[0]=
2214 ((uint32_t*)(src+i*stride))[1]= dc2;
2218 static void pred8x8_top_dc_c(uint8_t *src, int stride){
2224 dc0+= src[i-stride];
2225 dc1+= src[4+i-stride];
2227 dc0= 0x01010101*((dc0 + 2)>>2);
2228 dc1= 0x01010101*((dc1 + 2)>>2);
2231 ((uint32_t*)(src+i*stride))[0]= dc0;
2232 ((uint32_t*)(src+i*stride))[1]= dc1;
2235 ((uint32_t*)(src+i*stride))[0]= dc0;
2236 ((uint32_t*)(src+i*stride))[1]= dc1;
2241 static void pred8x8_dc_c(uint8_t *src, int stride){
2243 int dc0, dc1, dc2, dc3;
2247 dc0+= src[-1+i*stride] + src[i-stride];
2248 dc1+= src[4+i-stride];
2249 dc2+= src[-1+(i+4)*stride];
2251 dc3= 0x01010101*((dc1 + dc2 + 4)>>3);
2252 dc0= 0x01010101*((dc0 + 4)>>3);
2253 dc1= 0x01010101*((dc1 + 2)>>2);
2254 dc2= 0x01010101*((dc2 + 2)>>2);
2257 ((uint32_t*)(src+i*stride))[0]= dc0;
2258 ((uint32_t*)(src+i*stride))[1]= dc1;
2261 ((uint32_t*)(src+i*stride))[0]= dc2;
2262 ((uint32_t*)(src+i*stride))[1]= dc3;
2266 static void pred8x8_plane_c(uint8_t *src, int stride){
2269 uint8_t *cm = cropTbl + MAX_NEG_CROP;
2270 const uint8_t * const src0 = src+3-stride;
2271 const uint8_t *src1 = src+4*stride-1;
2272 const uint8_t *src2 = src1-2*stride; // == src+2*stride-1;
2273 int H = src0[1] - src0[-1];
2274 int V = src1[0] - src2[ 0];
2275 for(k=2; k<=4; ++k) {
2276 src1 += stride; src2 -= stride;
2277 H += k*(src0[k] - src0[-k]);
2278 V += k*(src1[0] - src2[ 0]);
2280 H = ( 17*H+16 ) >> 5;
2281 V = ( 17*V+16 ) >> 5;
2283 a = 16*(src1[0] + src2[8]+1) - 3*(V+H);
2284 for(j=8; j>0; --j) {
2287 src[0] = cm[ (b ) >> 5 ];
2288 src[1] = cm[ (b+ H) >> 5 ];
2289 src[2] = cm[ (b+2*H) >> 5 ];
2290 src[3] = cm[ (b+3*H) >> 5 ];
2291 src[4] = cm[ (b+4*H) >> 5 ];
2292 src[5] = cm[ (b+5*H) >> 5 ];
2293 src[6] = cm[ (b+6*H) >> 5 ];
2294 src[7] = cm[ (b+7*H) >> 5 ];
2299 static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
2300 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2301 int src_x_offset, int src_y_offset,
2302 qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
2303 MpegEncContext * const s = &h->s;
2304 const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
2305 const int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
2306 const int luma_xy= (mx&3) + ((my&3)<<2);
2307 uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*s->linesize;
2308 uint8_t * src_cb= pic->data[1] + (mx>>3) + (my>>3)*s->uvlinesize;
2309 uint8_t * src_cr= pic->data[2] + (mx>>3) + (my>>3)*s->uvlinesize;
2310 int extra_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16; //FIXME increase edge?, IMHO not worth it
2311 int extra_height= extra_width;
2313 const int full_mx= mx>>2;
2314 const int full_my= my>>2;
2316 assert(pic->data[0]);
2318 if(mx&7) extra_width -= 3;
2319 if(my&7) extra_height -= 3;
2321 if( full_mx < 0-extra_width
2322 || full_my < 0-extra_height
2323 || full_mx + 16/*FIXME*/ > s->width + extra_width
2324 || full_my + 16/*FIXME*/ > s->height + extra_height){
2325 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);
2326 src_y= s->edge_emu_buffer + 2 + 2*s->linesize;
2330 qpix_op[luma_xy](dest_y, src_y, s->linesize); //FIXME try variable height perhaps?
2332 qpix_op[luma_xy](dest_y + delta, src_y + delta, s->linesize);
2335 if(s->flags&CODEC_FLAG_GRAY) return;
2338 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);
2339 src_cb= s->edge_emu_buffer;
2341 chroma_op(dest_cb, src_cb, s->uvlinesize, chroma_height, mx&7, my&7);
2344 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);
2345 src_cr= s->edge_emu_buffer;
2347 chroma_op(dest_cr, src_cr, s->uvlinesize, chroma_height, mx&7, my&7);
2350 static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
2351 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2352 int x_offset, int y_offset,
2353 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2354 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
2355 int list0, int list1){
2356 MpegEncContext * const s = &h->s;
2357 qpel_mc_func *qpix_op= qpix_put;
2358 h264_chroma_mc_func chroma_op= chroma_put;
2360 dest_y += 2*x_offset + 2*y_offset*s-> linesize;
2361 dest_cb += x_offset + y_offset*s->uvlinesize;
2362 dest_cr += x_offset + y_offset*s->uvlinesize;
2363 x_offset += 8*s->mb_x;
2364 y_offset += 8*s->mb_y;
2367 Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
2368 mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
2369 dest_y, dest_cb, dest_cr, x_offset, y_offset,
2370 qpix_op, chroma_op);
2373 chroma_op= chroma_avg;
2377 Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
2378 mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
2379 dest_y, dest_cb, dest_cr, x_offset, y_offset,
2380 qpix_op, chroma_op);
2384 static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
2385 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2386 int x_offset, int y_offset,
2387 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2388 h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
2389 h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
2390 int list0, int list1){
2391 MpegEncContext * const s = &h->s;
2393 dest_y += 2*x_offset + 2*y_offset*s-> linesize;
2394 dest_cb += x_offset + y_offset*s->uvlinesize;
2395 dest_cr += x_offset + y_offset*s->uvlinesize;
2396 x_offset += 8*s->mb_x;
2397 y_offset += 8*s->mb_y;
2400 /* don't optimize for luma-only case, since B-frames usually
2401 * use implicit weights => chroma too. */
2402 uint8_t *tmp_cb = s->obmc_scratchpad;
2403 uint8_t *tmp_cr = tmp_cb + 8*s->uvlinesize;
2404 uint8_t *tmp_y = tmp_cr + 8*s->uvlinesize;
2405 int refn0 = h->ref_cache[0][ scan8[n] ];
2406 int refn1 = h->ref_cache[1][ scan8[n] ];
2408 mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
2409 dest_y, dest_cb, dest_cr,
2410 x_offset, y_offset, qpix_put, chroma_put);
2411 mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
2412 tmp_y, tmp_cb, tmp_cr,
2413 x_offset, y_offset, qpix_put, chroma_put);
2415 if(h->use_weight == 2){
2416 int weight0 = h->implicit_weight[refn0][refn1];
2417 int weight1 = 64 - weight0;
2418 luma_weight_avg( dest_y, tmp_y, s-> linesize, 5, weight0, weight1, 0, 0);
2419 chroma_weight_avg(dest_cb, tmp_cb, s->uvlinesize, 5, weight0, weight1, 0, 0);
2420 chroma_weight_avg(dest_cr, tmp_cr, s->uvlinesize, 5, weight0, weight1, 0, 0);
2422 luma_weight_avg(dest_y, tmp_y, s->linesize, h->luma_log2_weight_denom,
2423 h->luma_weight[0][refn0], h->luma_weight[1][refn1],
2424 h->luma_offset[0][refn0], h->luma_offset[1][refn1]);
2425 chroma_weight_avg(dest_cb, tmp_cb, s->uvlinesize, h->chroma_log2_weight_denom,
2426 h->chroma_weight[0][refn0][0], h->chroma_weight[1][refn1][0],
2427 h->chroma_offset[0][refn0][0], h->chroma_offset[1][refn1][0]);
2428 chroma_weight_avg(dest_cr, tmp_cr, s->uvlinesize, h->chroma_log2_weight_denom,
2429 h->chroma_weight[0][refn0][1], h->chroma_weight[1][refn1][1],
2430 h->chroma_offset[0][refn0][1], h->chroma_offset[1][refn1][1]);
2433 int list = list1 ? 1 : 0;
2434 int refn = h->ref_cache[list][ scan8[n] ];
2435 Picture *ref= &h->ref_list[list][refn];
2436 mc_dir_part(h, ref, n, square, chroma_height, delta, list,
2437 dest_y, dest_cb, dest_cr, x_offset, y_offset,
2438 qpix_put, chroma_put);
2440 luma_weight_op(dest_y, s->linesize, h->luma_log2_weight_denom,
2441 h->luma_weight[list][refn], h->luma_offset[list][refn]);
2442 if(h->use_weight_chroma){
2443 chroma_weight_op(dest_cb, s->uvlinesize, h->chroma_log2_weight_denom,
2444 h->chroma_weight[list][refn][0], h->chroma_offset[list][refn][0]);
2445 chroma_weight_op(dest_cr, s->uvlinesize, h->chroma_log2_weight_denom,
2446 h->chroma_weight[list][refn][1], h->chroma_offset[list][refn][1]);
2451 static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
2452 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2453 int x_offset, int y_offset,
2454 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2455 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
2456 h264_weight_func *weight_op, h264_biweight_func *weight_avg,
2457 int list0, int list1){
2458 if((h->use_weight==2 && list0 && list1
2459 && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ] != 32))
2460 || h->use_weight==1)
2461 mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
2462 x_offset, y_offset, qpix_put, chroma_put,
2463 weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
2465 mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
2466 x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
2469 static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2470 qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
2471 qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
2472 h264_weight_func *weight_op, h264_biweight_func *weight_avg){
2473 MpegEncContext * const s = &h->s;
2474 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
2475 const int mb_type= s->current_picture.mb_type[mb_xy];
2477 assert(IS_INTER(mb_type));
2479 if(IS_16X16(mb_type)){
2480 mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
2481 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
2482 &weight_op[0], &weight_avg[0],
2483 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2484 }else if(IS_16X8(mb_type)){
2485 mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
2486 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
2487 &weight_op[1], &weight_avg[1],
2488 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2489 mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
2490 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
2491 &weight_op[1], &weight_avg[1],
2492 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2493 }else if(IS_8X16(mb_type)){
2494 mc_part(h, 0, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 0, 0,
2495 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2496 &weight_op[2], &weight_avg[2],
2497 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2498 mc_part(h, 4, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 4, 0,
2499 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2500 &weight_op[2], &weight_avg[2],
2501 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2505 assert(IS_8X8(mb_type));
2508 const int sub_mb_type= h->sub_mb_type[i];
2510 int x_offset= (i&1)<<2;
2511 int y_offset= (i&2)<<1;
2513 if(IS_SUB_8X8(sub_mb_type)){
2514 mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2515 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2516 &weight_op[3], &weight_avg[3],
2517 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2518 }else if(IS_SUB_8X4(sub_mb_type)){
2519 mc_part(h, n , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2520 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2521 &weight_op[4], &weight_avg[4],
2522 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2523 mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
2524 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2525 &weight_op[4], &weight_avg[4],
2526 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2527 }else if(IS_SUB_4X8(sub_mb_type)){
2528 mc_part(h, n , 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2529 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2530 &weight_op[5], &weight_avg[5],
2531 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2532 mc_part(h, n+1, 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
2533 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2534 &weight_op[5], &weight_avg[5],
2535 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2538 assert(IS_SUB_4X4(sub_mb_type));
2540 int sub_x_offset= x_offset + 2*(j&1);
2541 int sub_y_offset= y_offset + (j&2);
2542 mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
2543 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2544 &weight_op[6], &weight_avg[6],
2545 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2552 static void decode_init_vlc(H264Context *h){
2553 static int done = 0;
2559 init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5,
2560 &chroma_dc_coeff_token_len [0], 1, 1,
2561 &chroma_dc_coeff_token_bits[0], 1, 1, 1);
2564 init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17,
2565 &coeff_token_len [i][0], 1, 1,
2566 &coeff_token_bits[i][0], 1, 1, 1);
2570 init_vlc(&chroma_dc_total_zeros_vlc[i], CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
2571 &chroma_dc_total_zeros_len [i][0], 1, 1,
2572 &chroma_dc_total_zeros_bits[i][0], 1, 1, 1);
2574 for(i=0; i<15; i++){
2575 init_vlc(&total_zeros_vlc[i], TOTAL_ZEROS_VLC_BITS, 16,
2576 &total_zeros_len [i][0], 1, 1,
2577 &total_zeros_bits[i][0], 1, 1, 1);
2581 init_vlc(&run_vlc[i], RUN_VLC_BITS, 7,
2582 &run_len [i][0], 1, 1,
2583 &run_bits[i][0], 1, 1, 1);
2585 init_vlc(&run7_vlc, RUN7_VLC_BITS, 16,
2586 &run_len [6][0], 1, 1,
2587 &run_bits[6][0], 1, 1, 1);
2592 * Sets the intra prediction function pointers.
2594 static void init_pred_ptrs(H264Context *h){
2595 // MpegEncContext * const s = &h->s;
2597 h->pred4x4[VERT_PRED ]= pred4x4_vertical_c;
2598 h->pred4x4[HOR_PRED ]= pred4x4_horizontal_c;
2599 h->pred4x4[DC_PRED ]= pred4x4_dc_c;
2600 h->pred4x4[DIAG_DOWN_LEFT_PRED ]= pred4x4_down_left_c;
2601 h->pred4x4[DIAG_DOWN_RIGHT_PRED]= pred4x4_down_right_c;
2602 h->pred4x4[VERT_RIGHT_PRED ]= pred4x4_vertical_right_c;
2603 h->pred4x4[HOR_DOWN_PRED ]= pred4x4_horizontal_down_c;
2604 h->pred4x4[VERT_LEFT_PRED ]= pred4x4_vertical_left_c;
2605 h->pred4x4[HOR_UP_PRED ]= pred4x4_horizontal_up_c;
2606 h->pred4x4[LEFT_DC_PRED ]= pred4x4_left_dc_c;
2607 h->pred4x4[TOP_DC_PRED ]= pred4x4_top_dc_c;
2608 h->pred4x4[DC_128_PRED ]= pred4x4_128_dc_c;
2610 h->pred8x8[DC_PRED8x8 ]= pred8x8_dc_c;
2611 h->pred8x8[VERT_PRED8x8 ]= pred8x8_vertical_c;
2612 h->pred8x8[HOR_PRED8x8 ]= pred8x8_horizontal_c;
2613 h->pred8x8[PLANE_PRED8x8 ]= pred8x8_plane_c;
2614 h->pred8x8[LEFT_DC_PRED8x8]= pred8x8_left_dc_c;
2615 h->pred8x8[TOP_DC_PRED8x8 ]= pred8x8_top_dc_c;
2616 h->pred8x8[DC_128_PRED8x8 ]= pred8x8_128_dc_c;
2618 h->pred16x16[DC_PRED8x8 ]= pred16x16_dc_c;
2619 h->pred16x16[VERT_PRED8x8 ]= pred16x16_vertical_c;
2620 h->pred16x16[HOR_PRED8x8 ]= pred16x16_horizontal_c;
2621 h->pred16x16[PLANE_PRED8x8 ]= pred16x16_plane_c;
2622 h->pred16x16[LEFT_DC_PRED8x8]= pred16x16_left_dc_c;
2623 h->pred16x16[TOP_DC_PRED8x8 ]= pred16x16_top_dc_c;
2624 h->pred16x16[DC_128_PRED8x8 ]= pred16x16_128_dc_c;
2627 static void free_tables(H264Context *h){
2628 av_freep(&h->intra4x4_pred_mode);
2629 av_freep(&h->chroma_pred_mode_table);
2630 av_freep(&h->cbp_table);
2631 av_freep(&h->mvd_table[0]);
2632 av_freep(&h->mvd_table[1]);
2633 av_freep(&h->direct_table);
2634 av_freep(&h->non_zero_count);
2635 av_freep(&h->slice_table_base);
2636 av_freep(&h->top_borders[1]);
2637 av_freep(&h->top_borders[0]);
2638 h->slice_table= NULL;
2640 av_freep(&h->mb2b_xy);
2641 av_freep(&h->mb2b8_xy);
2643 av_freep(&h->s.obmc_scratchpad);
2648 * needs width/height
2650 static int alloc_tables(H264Context *h){
2651 MpegEncContext * const s = &h->s;
2652 const int big_mb_num= s->mb_stride * (s->mb_height+1);
2655 CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8 * sizeof(uint8_t))
2657 CHECKED_ALLOCZ(h->non_zero_count , big_mb_num * 16 * sizeof(uint8_t))
2658 CHECKED_ALLOCZ(h->slice_table_base , big_mb_num * sizeof(uint8_t))
2659 CHECKED_ALLOCZ(h->top_borders[0] , s->mb_width * (16+8+8) * sizeof(uint8_t))
2660 CHECKED_ALLOCZ(h->top_borders[1] , s->mb_width * (16+8+8) * sizeof(uint8_t))
2661 CHECKED_ALLOCZ(h->cbp_table, big_mb_num * sizeof(uint16_t))
2663 if( h->pps.cabac ) {
2664 CHECKED_ALLOCZ(h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t))
2665 CHECKED_ALLOCZ(h->mvd_table[0], 32*big_mb_num * sizeof(uint16_t));
2666 CHECKED_ALLOCZ(h->mvd_table[1], 32*big_mb_num * sizeof(uint16_t));
2667 CHECKED_ALLOCZ(h->direct_table, 32*big_mb_num * sizeof(uint8_t));
2670 memset(h->slice_table_base, -1, big_mb_num * sizeof(uint8_t));
2671 h->slice_table= h->slice_table_base + s->mb_stride + 1;
2673 CHECKED_ALLOCZ(h->mb2b_xy , big_mb_num * sizeof(uint32_t));
2674 CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint32_t));
2675 for(y=0; y<s->mb_height; y++){
2676 for(x=0; x<s->mb_width; x++){
2677 const int mb_xy= x + y*s->mb_stride;
2678 const int b_xy = 4*x + 4*y*h->b_stride;
2679 const int b8_xy= 2*x + 2*y*h->b8_stride;
2681 h->mb2b_xy [mb_xy]= b_xy;
2682 h->mb2b8_xy[mb_xy]= b8_xy;
2686 s->obmc_scratchpad = NULL;
2694 static void common_init(H264Context *h){
2695 MpegEncContext * const s = &h->s;
2697 s->width = s->avctx->width;
2698 s->height = s->avctx->height;
2699 s->codec_id= s->avctx->codec->id;
2703 s->unrestricted_mv=1;
2704 s->decode=1; //FIXME
2707 static int decode_init(AVCodecContext *avctx){
2708 H264Context *h= avctx->priv_data;
2709 MpegEncContext * const s = &h->s;
2711 MPV_decode_defaults(s);
2716 s->out_format = FMT_H264;
2717 s->workaround_bugs= avctx->workaround_bugs;
2720 // s->decode_mb= ff_h263_decode_mb;
2722 avctx->pix_fmt= PIX_FMT_YUV420P;
2726 if(avctx->extradata_size > 0 && avctx->extradata &&
2727 *(char *)avctx->extradata == 1){
2737 static void frame_start(H264Context *h){
2738 MpegEncContext * const s = &h->s;
2741 MPV_frame_start(s, s->avctx);
2742 ff_er_frame_start(s);
2744 assert(s->linesize && s->uvlinesize);
2746 for(i=0; i<16; i++){
2747 h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
2748 h->block_offset[24+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
2751 h->block_offset[16+i]=
2752 h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2753 h->block_offset[24+16+i]=
2754 h->block_offset[24+20+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2757 /* can't be in alloc_tables because linesize isn't known there.
2758 * FIXME: redo bipred weight to not require extra buffer? */
2759 if(!s->obmc_scratchpad)
2760 s->obmc_scratchpad = av_malloc(16*s->linesize + 2*8*s->uvlinesize);
2762 // s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
2765 static inline void backup_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){
2766 MpegEncContext * const s = &h->s;
2770 src_cb -= uvlinesize;
2771 src_cr -= uvlinesize;
2773 // There are two lines saved, the line above the the top macroblock of a pair,
2774 // and the line above the bottom macroblock
2775 h->left_border[0]= h->top_borders[0][s->mb_x][15];
2776 for(i=1; i<17; i++){
2777 h->left_border[i]= src_y[15+i* linesize];
2780 *(uint64_t*)(h->top_borders[0][s->mb_x]+0)= *(uint64_t*)(src_y + 16*linesize);
2781 *(uint64_t*)(h->top_borders[0][s->mb_x]+8)= *(uint64_t*)(src_y +8+16*linesize);
2783 if(!(s->flags&CODEC_FLAG_GRAY)){
2784 h->left_border[17 ]= h->top_borders[0][s->mb_x][16+7];
2785 h->left_border[17+9]= h->top_borders[0][s->mb_x][24+7];
2787 h->left_border[i+17 ]= src_cb[7+i*uvlinesize];
2788 h->left_border[i+17+9]= src_cr[7+i*uvlinesize];
2790 *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+8*uvlinesize);
2791 *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+8*uvlinesize);
2795 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){
2796 MpegEncContext * const s = &h->s;
2799 int deblock_left = (s->mb_x > 0);
2800 int deblock_top = (s->mb_y > 0);
2802 src_y -= linesize + 1;
2803 src_cb -= uvlinesize + 1;
2804 src_cr -= uvlinesize + 1;
2806 #define XCHG(a,b,t,xchg)\
2813 for(i = !deblock_top; i<17; i++){
2814 XCHG(h->left_border[i ], src_y [i* linesize], temp8, xchg);
2819 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
2820 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
2823 if(!(s->flags&CODEC_FLAG_GRAY)){
2825 for(i = !deblock_top; i<9; i++){
2826 XCHG(h->left_border[i+17 ], src_cb[i*uvlinesize], temp8, xchg);
2827 XCHG(h->left_border[i+17+9], src_cr[i*uvlinesize], temp8, xchg);
2831 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
2832 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
2837 static inline void backup_pair_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){
2838 MpegEncContext * const s = &h->s;
2841 src_y -= 2 * linesize;
2842 src_cb -= 2 * uvlinesize;
2843 src_cr -= 2 * uvlinesize;
2845 // There are two lines saved, the line above the the top macroblock of a pair,
2846 // and the line above the bottom macroblock
2847 h->left_border[0]= h->top_borders[0][s->mb_x][15];
2848 h->left_border[1]= h->top_borders[1][s->mb_x][15];
2849 for(i=2; i<34; i++){
2850 h->left_border[i]= src_y[15+i* linesize];
2853 *(uint64_t*)(h->top_borders[0][s->mb_x]+0)= *(uint64_t*)(src_y + 32*linesize);
2854 *(uint64_t*)(h->top_borders[0][s->mb_x]+8)= *(uint64_t*)(src_y +8+32*linesize);
2855 *(uint64_t*)(h->top_borders[1][s->mb_x]+0)= *(uint64_t*)(src_y + 33*linesize);
2856 *(uint64_t*)(h->top_borders[1][s->mb_x]+8)= *(uint64_t*)(src_y +8+33*linesize);
2858 if(!(s->flags&CODEC_FLAG_GRAY)){
2859 h->left_border[34 ]= h->top_borders[0][s->mb_x][16+7];
2860 h->left_border[34+ 1]= h->top_borders[1][s->mb_x][16+7];
2861 h->left_border[34+18 ]= h->top_borders[0][s->mb_x][24+7];
2862 h->left_border[34+18+1]= h->top_borders[1][s->mb_x][24+7];
2863 for(i=2; i<18; i++){
2864 h->left_border[i+34 ]= src_cb[7+i*uvlinesize];
2865 h->left_border[i+34+18]= src_cr[7+i*uvlinesize];
2867 *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+16*uvlinesize);
2868 *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+16*uvlinesize);
2869 *(uint64_t*)(h->top_borders[1][s->mb_x]+16)= *(uint64_t*)(src_cb+17*uvlinesize);
2870 *(uint64_t*)(h->top_borders[1][s->mb_x]+24)= *(uint64_t*)(src_cr+17*uvlinesize);
2874 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){
2875 MpegEncContext * const s = &h->s;
2878 int deblock_left = (s->mb_x > 0);
2879 int deblock_top = (s->mb_y > 0);
2881 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);
2883 src_y -= 2 * linesize + 1;
2884 src_cb -= 2 * uvlinesize + 1;
2885 src_cr -= 2 * uvlinesize + 1;
2887 #define XCHG(a,b,t,xchg)\
2894 for(i = (!deblock_top)<<1; i<34; i++){
2895 XCHG(h->left_border[i ], src_y [i* linesize], temp8, xchg);
2900 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
2901 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
2902 XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+0), *(uint64_t*)(src_y +1 +linesize), temp64, xchg);
2903 XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+8), *(uint64_t*)(src_y +9 +linesize), temp64, 1);
2906 if(!(s->flags&CODEC_FLAG_GRAY)){
2908 for(i = (!deblock_top) << 1; i<18; i++){
2909 XCHG(h->left_border[i+34 ], src_cb[i*uvlinesize], temp8, xchg);
2910 XCHG(h->left_border[i+34+18], src_cr[i*uvlinesize], temp8, xchg);
2914 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
2915 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
2916 XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+16), *(uint64_t*)(src_cb+1 +uvlinesize), temp64, 1);
2917 XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+24), *(uint64_t*)(src_cr+1 +uvlinesize), temp64, 1);
2922 static void hl_decode_mb(H264Context *h){
2923 MpegEncContext * const s = &h->s;
2924 const int mb_x= s->mb_x;
2925 const int mb_y= s->mb_y;
2926 const int mb_xy= mb_x + mb_y*s->mb_stride;
2927 const int mb_type= s->current_picture.mb_type[mb_xy];
2928 uint8_t *dest_y, *dest_cb, *dest_cr;
2929 int linesize, uvlinesize /*dct_offset*/;
2931 int *block_offset = &h->block_offset[0];
2932 const unsigned int bottom = mb_y & 1;
2937 dest_y = s->current_picture.data[0] + (mb_y * 16* s->linesize ) + mb_x * 16;
2938 dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2939 dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2941 if (h->mb_field_decoding_flag) {
2942 linesize = s->linesize * 2;
2943 uvlinesize = s->uvlinesize * 2;
2944 block_offset = &h->block_offset[24];
2945 if(mb_y&1){ //FIXME move out of this func?
2946 dest_y -= s->linesize*15;
2947 dest_cb-= s->uvlinesize*7;
2948 dest_cr-= s->uvlinesize*7;
2951 linesize = s->linesize;
2952 uvlinesize = s->uvlinesize;
2953 // dct_offset = s->linesize * 16;
2956 if (IS_INTRA_PCM(mb_type)) {
2959 // The pixels are stored in h->mb array in the same order as levels,
2960 // copy them in output in the correct order.
2961 for(i=0; i<16; i++) {
2962 for (y=0; y<4; y++) {
2963 for (x=0; x<4; x++) {
2964 *(dest_y + block_offset[i] + y*linesize + x) = h->mb[i*16+y*4+x];
2968 for(i=16; i<16+4; i++) {
2969 for (y=0; y<4; y++) {
2970 for (x=0; x<4; x++) {
2971 *(dest_cb + block_offset[i] + y*uvlinesize + x) = h->mb[i*16+y*4+x];
2975 for(i=20; i<20+4; i++) {
2976 for (y=0; y<4; y++) {
2977 for (x=0; x<4; x++) {
2978 *(dest_cr + block_offset[i] + y*uvlinesize + x) = h->mb[i*16+y*4+x];
2983 if(IS_INTRA(mb_type)){
2984 if(h->deblocking_filter) {
2985 if (h->mb_aff_frame) {
2987 xchg_pair_border(h, dest_y, dest_cb, dest_cr, s->linesize, s->uvlinesize, 1);
2989 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1);
2993 if(!(s->flags&CODEC_FLAG_GRAY)){
2994 h->pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
2995 h->pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
2998 if(IS_INTRA4x4(mb_type)){
3000 for(i=0; i<16; i++){
3001 uint8_t * const ptr= dest_y + block_offset[i];
3003 const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
3006 if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
3007 const int topright_avail= (h->topright_samples_available<<i)&0x8000;
3008 assert(mb_y || linesize <= block_offset[i]);
3009 if(!topright_avail){
3010 tr= ptr[3 - linesize]*0x01010101;
3011 topright= (uint8_t*) &tr;
3012 }else if(i==5 && h->deblocking_filter){
3013 tr= *(uint32_t*)h->top_borders[h->mb_aff_frame ? IS_INTERLACED(mb_type) ? bottom : 1 : 0][mb_x+1];
3014 topright= (uint8_t*) &tr;
3016 topright= ptr + 4 - linesize;
3020 h->pred4x4[ dir ](ptr, topright, linesize);
3021 if(h->non_zero_count_cache[ scan8[i] ]){
3022 if(s->codec_id == CODEC_ID_H264)
3023 s->dsp.h264_idct_add(ptr, h->mb + i*16, linesize);
3025 svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
3030 h->pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
3031 if(s->codec_id == CODEC_ID_H264)
3032 h264_luma_dc_dequant_idct_c(h->mb, s->qscale);
3034 svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
3036 if(h->deblocking_filter) {
3037 if (h->mb_aff_frame) {
3039 uint8_t *pair_dest_y = s->current_picture.data[0] + ((mb_y-1) * 16* s->linesize ) + mb_x * 16;
3040 uint8_t *pair_dest_cb = s->current_picture.data[1] + ((mb_y-1) * 8 * s->uvlinesize) + mb_x * 8;
3041 uint8_t *pair_dest_cr = s->current_picture.data[2] + ((mb_y-1) * 8 * s->uvlinesize) + mb_x * 8;
3043 xchg_pair_border(h, pair_dest_y, pair_dest_cb, pair_dest_cr, s->linesize, s->uvlinesize, 0);
3047 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);
3050 }else if(s->codec_id == CODEC_ID_H264){
3051 hl_motion(h, dest_y, dest_cb, dest_cr,
3052 s->dsp.put_h264_qpel_pixels_tab, s->dsp.put_h264_chroma_pixels_tab,
3053 s->dsp.avg_h264_qpel_pixels_tab, s->dsp.avg_h264_chroma_pixels_tab,
3054 s->dsp.weight_h264_pixels_tab, s->dsp.biweight_h264_pixels_tab);
3058 if(!IS_INTRA4x4(mb_type)){
3059 if(s->codec_id == CODEC_ID_H264){
3060 for(i=0; i<16; i++){
3061 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
3062 uint8_t * const ptr= dest_y + block_offset[i];
3063 s->dsp.h264_idct_add(ptr, h->mb + i*16, linesize);
3067 for(i=0; i<16; i++){
3068 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
3069 uint8_t * const ptr= dest_y + block_offset[i];
3070 svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
3076 if(!(s->flags&CODEC_FLAG_GRAY)){
3077 chroma_dc_dequant_idct_c(h->mb + 16*16, h->chroma_qp);
3078 chroma_dc_dequant_idct_c(h->mb + 16*16+4*16, h->chroma_qp);
3079 if(s->codec_id == CODEC_ID_H264){
3080 for(i=16; i<16+4; i++){
3081 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
3082 uint8_t * const ptr= dest_cb + block_offset[i];
3083 s->dsp.h264_idct_add(ptr, h->mb + i*16, uvlinesize);
3086 for(i=20; i<20+4; i++){
3087 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
3088 uint8_t * const ptr= dest_cr + block_offset[i];
3089 s->dsp.h264_idct_add(ptr, h->mb + i*16, uvlinesize);
3093 for(i=16; i<16+4; i++){
3094 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
3095 uint8_t * const ptr= dest_cb + block_offset[i];
3096 svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
3099 for(i=20; i<20+4; i++){
3100 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
3101 uint8_t * const ptr= dest_cr + block_offset[i];
3102 svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
3108 if(h->deblocking_filter) {
3109 if (h->mb_aff_frame) {
3110 const int mb_y = s->mb_y - 1;
3111 uint8_t *pair_dest_y, *pair_dest_cb, *pair_dest_cr;
3112 const int mb_xy= mb_x + mb_y*s->mb_stride;
3113 const int mb_type_top = s->current_picture.mb_type[mb_xy];
3114 const int mb_type_bottom= s->current_picture.mb_type[mb_xy+s->mb_stride];
3115 uint8_t tmp = s->current_picture.data[1][384];
3116 if (!bottom) return;
3117 pair_dest_y = s->current_picture.data[0] + (mb_y * 16* s->linesize ) + mb_x * 16;
3118 pair_dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
3119 pair_dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
3121 backup_pair_border(h, pair_dest_y, pair_dest_cb, pair_dest_cr, s->linesize, s->uvlinesize);
3122 // TODO deblock a pair
3125 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);
3126 fill_caches(h, mb_type_top, 1); //FIXME don't fill stuff which isn't used by filter_mb
3127 filter_mb(h, mb_x, mb_y, pair_dest_y, pair_dest_cb, pair_dest_cr, linesize, uvlinesize);
3128 if (tmp != s->current_picture.data[1][384]) {
3129 tprintf("modified pixel 8,1 (1)\n");
3133 tprintf("call mbaff filter_mb\n");
3134 fill_caches(h, mb_type_bottom, 1); //FIXME don't fill stuff which isn't used by filter_mb
3135 filter_mb(h, mb_x, mb_y+1, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
3136 if (tmp != s->current_picture.data[1][384]) {
3137 tprintf("modified pixel 8,1 (2)\n");
3140 tprintf("call filter_mb\n");
3141 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
3142 fill_caches(h, mb_type, 1); //FIXME don't fill stuff which isn't used by filter_mb
3143 filter_mb(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
3149 * fills the default_ref_list.
3151 static int fill_default_ref_list(H264Context *h){
3152 MpegEncContext * const s = &h->s;
3154 int smallest_poc_greater_than_current = -1;
3155 Picture sorted_short_ref[32];
3157 if(h->slice_type==B_TYPE){
3161 /* sort frame according to poc in B slice */
3162 for(out_i=0; out_i<h->short_ref_count; out_i++){
3164 int best_poc=INT_MAX;
3166 for(i=0; i<h->short_ref_count; i++){
3167 const int poc= h->short_ref[i]->poc;
3168 if(poc > limit && poc < best_poc){
3174 assert(best_i != INT_MIN);
3177 sorted_short_ref[out_i]= *h->short_ref[best_i];
3178 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);
3179 if (-1 == smallest_poc_greater_than_current) {
3180 if (h->short_ref[best_i]->poc >= s->current_picture_ptr->poc) {
3181 smallest_poc_greater_than_current = out_i;
3187 if(s->picture_structure == PICT_FRAME){
3188 if(h->slice_type==B_TYPE){
3190 tprintf("current poc: %d, smallest_poc_greater_than_current: %d\n", s->current_picture_ptr->poc, smallest_poc_greater_than_current);
3192 // find the largest poc
3193 for(list=0; list<2; list++){
3196 int step= list ? -1 : 1;
3198 for(i=0; i<h->short_ref_count && index < h->ref_count[list]; i++, j+=step) {
3199 while(j<0 || j>= h->short_ref_count){
3200 if(j != -99 && step == (list ? -1 : 1))
3203 j= smallest_poc_greater_than_current + (step>>1);
3205 if(sorted_short_ref[j].reference != 3) continue;
3206 h->default_ref_list[list][index ]= sorted_short_ref[j];
3207 h->default_ref_list[list][index++].pic_id= sorted_short_ref[j].frame_num;
3210 for(i = 0; i < 16 && index < h->ref_count[ list ]; i++){
3211 if(h->long_ref[i] == NULL) continue;
3212 if(h->long_ref[i]->reference != 3) continue;
3214 h->default_ref_list[ list ][index ]= *h->long_ref[i];
3215 h->default_ref_list[ list ][index++].pic_id= i;;
3218 if(list && (smallest_poc_greater_than_current<=0 || smallest_poc_greater_than_current>=h->short_ref_count) && (1 < index)){
3219 // swap the two first elements of L1 when
3220 // L0 and L1 are identical
3221 Picture temp= h->default_ref_list[1][0];
3222 h->default_ref_list[1][0] = h->default_ref_list[1][1];
3223 h->default_ref_list[1][1] = temp;
3226 if(index < h->ref_count[ list ])
3227 memset(&h->default_ref_list[list][index], 0, sizeof(Picture)*(h->ref_count[ list ] - index));
3231 for(i=0; i<h->short_ref_count; i++){
3232 if(h->short_ref[i]->reference != 3) continue; //FIXME refernce field shit
3233 h->default_ref_list[0][index ]= *h->short_ref[i];
3234 h->default_ref_list[0][index++].pic_id= h->short_ref[i]->frame_num;
3236 for(i = 0; i < 16; i++){
3237 if(h->long_ref[i] == NULL) continue;
3238 if(h->long_ref[i]->reference != 3) continue;
3239 h->default_ref_list[0][index ]= *h->long_ref[i];
3240 h->default_ref_list[0][index++].pic_id= i;;
3242 if(index < h->ref_count[0])
3243 memset(&h->default_ref_list[0][index], 0, sizeof(Picture)*(h->ref_count[0] - index));
3246 if(h->slice_type==B_TYPE){
3248 //FIXME second field balh
3252 for (i=0; i<h->ref_count[0]; i++) {
3253 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]);
3255 if(h->slice_type==B_TYPE){
3256 for (i=0; i<h->ref_count[1]; i++) {
3257 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]);
3264 static void print_short_term(H264Context *h);
3265 static void print_long_term(H264Context *h);
3267 static int decode_ref_pic_list_reordering(H264Context *h){
3268 MpegEncContext * const s = &h->s;
3271 print_short_term(h);
3273 if(h->slice_type==I_TYPE || h->slice_type==SI_TYPE) return 0; //FIXME move before func
3275 for(list=0; list<2; list++){
3276 memcpy(h->ref_list[list], h->default_ref_list[list], sizeof(Picture)*h->ref_count[list]);
3278 if(get_bits1(&s->gb)){
3279 int pred= h->curr_pic_num;
3281 for(index=0; ; index++){
3282 int reordering_of_pic_nums_idc= get_ue_golomb(&s->gb);
3285 Picture *ref = NULL;
3287 if(reordering_of_pic_nums_idc==3)
3290 if(index >= h->ref_count[list]){
3291 av_log(h->s.avctx, AV_LOG_ERROR, "reference count overflow\n");
3295 if(reordering_of_pic_nums_idc<3){
3296 if(reordering_of_pic_nums_idc<2){
3297 const int abs_diff_pic_num= get_ue_golomb(&s->gb) + 1;
3299 if(abs_diff_pic_num >= h->max_pic_num){
3300 av_log(h->s.avctx, AV_LOG_ERROR, "abs_diff_pic_num overflow\n");
3304 if(reordering_of_pic_nums_idc == 0) pred-= abs_diff_pic_num;
3305 else pred+= abs_diff_pic_num;
3306 pred &= h->max_pic_num - 1;
3308 for(i= h->short_ref_count-1; i>=0; i--){
3309 ref = h->short_ref[i];
3310 assert(ref->reference == 3);
3311 assert(!ref->long_ref);
3312 if(ref->data[0] != NULL && ref->frame_num == pred && ref->long_ref == 0) // ignore non existing pictures by testing data[0] pointer
3316 ref->pic_id= ref->frame_num;
3318 pic_id= get_ue_golomb(&s->gb); //long_term_pic_idx
3319 ref = h->long_ref[pic_id];
3320 ref->pic_id= pic_id;
3321 assert(ref->reference == 3);
3322 assert(ref->long_ref);
3327 av_log(h->s.avctx, AV_LOG_ERROR, "reference picture missing during reorder\n");
3328 memset(&h->ref_list[list][index], 0, sizeof(Picture)); //FIXME
3330 for(i=index; i+1<h->ref_count[list]; i++){
3331 if(ref->long_ref == h->ref_list[list][i].long_ref && ref->pic_id == h->ref_list[list][i].pic_id)
3334 for(; i > index; i--){
3335 h->ref_list[list][i]= h->ref_list[list][i-1];
3337 h->ref_list[list][index]= *ref;
3340 av_log(h->s.avctx, AV_LOG_ERROR, "illegal reordering_of_pic_nums_idc\n");
3346 if(h->slice_type!=B_TYPE) break;
3348 for(list=0; list<2; list++){
3349 for(index= 0; index < h->ref_count[list]; index++){
3350 if(!h->ref_list[list][index].data[0])
3351 h->ref_list[list][index]= s->current_picture;
3353 if(h->slice_type!=B_TYPE) break;
3356 if(h->slice_type==B_TYPE && !h->direct_spatial_mv_pred)
3357 direct_dist_scale_factor(h);
3358 direct_ref_list_init(h);
3362 static int pred_weight_table(H264Context *h){
3363 MpegEncContext * const s = &h->s;
3365 int luma_def, chroma_def;
3368 h->use_weight_chroma= 0;
3369 h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
3370 h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
3371 luma_def = 1<<h->luma_log2_weight_denom;
3372 chroma_def = 1<<h->chroma_log2_weight_denom;
3374 for(list=0; list<2; list++){
3375 for(i=0; i<h->ref_count[list]; i++){
3376 int luma_weight_flag, chroma_weight_flag;
3378 luma_weight_flag= get_bits1(&s->gb);
3379 if(luma_weight_flag){
3380 h->luma_weight[list][i]= get_se_golomb(&s->gb);
3381 h->luma_offset[list][i]= get_se_golomb(&s->gb);
3382 if( h->luma_weight[list][i] != luma_def
3383 || h->luma_offset[list][i] != 0)
3386 h->luma_weight[list][i]= luma_def;
3387 h->luma_offset[list][i]= 0;
3390 chroma_weight_flag= get_bits1(&s->gb);
3391 if(chroma_weight_flag){
3394 h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);
3395 h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);
3396 if( h->chroma_weight[list][i][j] != chroma_def
3397 || h->chroma_offset[list][i][j] != 0)
3398 h->use_weight_chroma= 1;
3403 h->chroma_weight[list][i][j]= chroma_def;
3404 h->chroma_offset[list][i][j]= 0;
3408 if(h->slice_type != B_TYPE) break;
3410 h->use_weight= h->use_weight || h->use_weight_chroma;
3414 static void implicit_weight_table(H264Context *h){
3415 MpegEncContext * const s = &h->s;
3417 int cur_poc = s->current_picture_ptr->poc;
3419 if( h->ref_count[0] == 1 && h->ref_count[1] == 1
3420 && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
3422 h->use_weight_chroma= 0;
3427 h->use_weight_chroma= 2;
3428 h->luma_log2_weight_denom= 5;
3429 h->chroma_log2_weight_denom= 5;
3432 for(ref0=0; ref0 < h->ref_count[0]; ref0++){
3433 int poc0 = h->ref_list[0][ref0].poc;
3434 for(ref1=0; ref1 < h->ref_count[1]; ref1++){
3435 int poc1 = h->ref_list[1][ref1].poc;
3436 int td = clip(poc1 - poc0, -128, 127);
3438 int tb = clip(cur_poc - poc0, -128, 127);
3439 int tx = (16384 + (ABS(td) >> 1)) / td;
3440 int dist_scale_factor = clip((tb*tx + 32) >> 6, -1024, 1023) >> 2;
3441 if(dist_scale_factor < -64 || dist_scale_factor > 128)
3442 h->implicit_weight[ref0][ref1] = 32;
3444 h->implicit_weight[ref0][ref1] = 64 - dist_scale_factor;
3446 h->implicit_weight[ref0][ref1] = 32;
3451 static inline void unreference_pic(H264Context *h, Picture *pic){
3454 if(pic == h->delayed_output_pic)
3457 for(i = 0; h->delayed_pic[i]; i++)
3458 if(pic == h->delayed_pic[i]){
3466 * instantaneous decoder refresh.
3468 static void idr(H264Context *h){
3471 for(i=0; i<16; i++){
3472 if (h->long_ref[i] != NULL) {
3473 unreference_pic(h, h->long_ref[i]);
3474 h->long_ref[i]= NULL;
3477 h->long_ref_count=0;
3479 for(i=0; i<h->short_ref_count; i++){
3480 unreference_pic(h, h->short_ref[i]);
3481 h->short_ref[i]= NULL;
3483 h->short_ref_count=0;
3486 /* forget old pics after a seek */
3487 static void flush_dpb(AVCodecContext *avctx){
3488 H264Context *h= avctx->priv_data;
3491 h->delayed_pic[i]= NULL;
3492 h->delayed_output_pic= NULL;
3498 * @return the removed picture or NULL if an error occurs
3500 static Picture * remove_short(H264Context *h, int frame_num){
3501 MpegEncContext * const s = &h->s;
3504 if(s->avctx->debug&FF_DEBUG_MMCO)
3505 av_log(h->s.avctx, AV_LOG_DEBUG, "remove short %d count %d\n", frame_num, h->short_ref_count);
3507 for(i=0; i<h->short_ref_count; i++){
3508 Picture *pic= h->short_ref[i];
3509 if(s->avctx->debug&FF_DEBUG_MMCO)
3510 av_log(h->s.avctx, AV_LOG_DEBUG, "%d %d %p\n", i, pic->frame_num, pic);
3511 if(pic->frame_num == frame_num){
3512 h->short_ref[i]= NULL;
3513 memmove(&h->short_ref[i], &h->short_ref[i+1], (h->short_ref_count - i - 1)*sizeof(Picture*));
3514 h->short_ref_count--;
3523 * @return the removed picture or NULL if an error occurs
3525 static Picture * remove_long(H264Context *h, int i){
3528 pic= h->long_ref[i];
3529 h->long_ref[i]= NULL;
3530 if(pic) h->long_ref_count--;
3536 * print short term list
3538 static void print_short_term(H264Context *h) {
3540 if(h->s.avctx->debug&FF_DEBUG_MMCO) {
3541 av_log(h->s.avctx, AV_LOG_DEBUG, "short term list:\n");
3542 for(i=0; i<h->short_ref_count; i++){
3543 Picture *pic= h->short_ref[i];
3544 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 * print long term list
3552 static void print_long_term(H264Context *h) {
3554 if(h->s.avctx->debug&FF_DEBUG_MMCO) {
3555 av_log(h->s.avctx, AV_LOG_DEBUG, "long term list:\n");
3556 for(i = 0; i < 16; i++){
3557 Picture *pic= h->long_ref[i];
3559 av_log(h->s.avctx, AV_LOG_DEBUG, "%d fn:%d poc:%d %p\n", i, pic->frame_num, pic->poc, pic->data[0]);
3566 * Executes the reference picture marking (memory management control operations).
3568 static int execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count){
3569 MpegEncContext * const s = &h->s;
3571 int current_is_long=0;
3574 if((s->avctx->debug&FF_DEBUG_MMCO) && mmco_count==0)
3575 av_log(h->s.avctx, AV_LOG_DEBUG, "no mmco here\n");
3577 for(i=0; i<mmco_count; i++){
3578 if(s->avctx->debug&FF_DEBUG_MMCO)
3579 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);
3581 switch(mmco[i].opcode){
3582 case MMCO_SHORT2UNUSED:
3583 pic= remove_short(h, mmco[i].short_frame_num);
3584 if(pic==NULL) return -1;
3585 unreference_pic(h, pic);
3587 case MMCO_SHORT2LONG:
3588 pic= remove_long(h, mmco[i].long_index);
3589 if(pic) unreference_pic(h, pic);
3591 h->long_ref[ mmco[i].long_index ]= remove_short(h, mmco[i].short_frame_num);
3592 h->long_ref[ mmco[i].long_index ]->long_ref=1;
3593 h->long_ref_count++;
3595 case MMCO_LONG2UNUSED:
3596 pic= remove_long(h, mmco[i].long_index);
3597 if(pic==NULL) return -1;
3598 unreference_pic(h, pic);
3601 pic= remove_long(h, mmco[i].long_index);
3602 if(pic) unreference_pic(h, pic);
3604 h->long_ref[ mmco[i].long_index ]= s->current_picture_ptr;
3605 h->long_ref[ mmco[i].long_index ]->long_ref=1;
3606 h->long_ref_count++;
3610 case MMCO_SET_MAX_LONG:
3611 assert(mmco[i].long_index <= 16);
3612 // just remove the long term which index is greater than new max
3613 for(j = mmco[i].long_index; j<16; j++){
3614 pic = remove_long(h, j);
3615 if (pic) unreference_pic(h, pic);
3619 while(h->short_ref_count){
3620 pic= remove_short(h, h->short_ref[0]->frame_num);
3621 unreference_pic(h, pic);
3623 for(j = 0; j < 16; j++) {
3624 pic= remove_long(h, j);
3625 if(pic) unreference_pic(h, pic);
3632 if(!current_is_long){
3633 pic= remove_short(h, s->current_picture_ptr->frame_num);
3635 unreference_pic(h, pic);
3636 av_log(h->s.avctx, AV_LOG_ERROR, "illegal short term buffer state detected\n");
3639 if(h->short_ref_count)
3640 memmove(&h->short_ref[1], &h->short_ref[0], h->short_ref_count*sizeof(Picture*));
3642 h->short_ref[0]= s->current_picture_ptr;
3643 h->short_ref[0]->long_ref=0;
3644 h->short_ref_count++;
3647 print_short_term(h);
3652 static int decode_ref_pic_marking(H264Context *h){
3653 MpegEncContext * const s = &h->s;
3656 if(h->nal_unit_type == NAL_IDR_SLICE){ //FIXME fields
3657 s->broken_link= get_bits1(&s->gb) -1;
3658 h->mmco[0].long_index= get_bits1(&s->gb) - 1; // current_long_term_idx
3659 if(h->mmco[0].long_index == -1)
3662 h->mmco[0].opcode= MMCO_LONG;
3666 if(get_bits1(&s->gb)){ // adaptive_ref_pic_marking_mode_flag
3667 for(i= 0; i<MAX_MMCO_COUNT; i++) {
3668 MMCOOpcode opcode= get_ue_golomb(&s->gb);;
3670 h->mmco[i].opcode= opcode;
3671 if(opcode==MMCO_SHORT2UNUSED || opcode==MMCO_SHORT2LONG){
3672 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
3673 /* if(h->mmco[i].short_frame_num >= h->short_ref_count || h->short_ref[ h->mmco[i].short_frame_num ] == NULL){
3674 fprintf(stderr, "illegal short ref in memory management control operation %d\n", mmco);
3678 if(opcode==MMCO_SHORT2LONG || opcode==MMCO_LONG2UNUSED || opcode==MMCO_LONG || opcode==MMCO_SET_MAX_LONG){
3679 h->mmco[i].long_index= get_ue_golomb(&s->gb);
3680 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){
3681 av_log(h->s.avctx, AV_LOG_ERROR, "illegal long ref in memory management control operation %d\n", opcode);
3686 if(opcode > MMCO_LONG){
3687 av_log(h->s.avctx, AV_LOG_ERROR, "illegal memory management control operation %d\n", opcode);
3690 if(opcode == MMCO_END)
3695 assert(h->long_ref_count + h->short_ref_count <= h->sps.ref_frame_count);
3697 if(h->long_ref_count + h->short_ref_count == h->sps.ref_frame_count){ //FIXME fields
3698 h->mmco[0].opcode= MMCO_SHORT2UNUSED;
3699 h->mmco[0].short_frame_num= h->short_ref[ h->short_ref_count - 1 ]->frame_num;
3709 static int init_poc(H264Context *h){
3710 MpegEncContext * const s = &h->s;
3711 const int max_frame_num= 1<<h->sps.log2_max_frame_num;
3714 if(h->nal_unit_type == NAL_IDR_SLICE){
3715 h->frame_num_offset= 0;
3717 if(h->frame_num < h->prev_frame_num)
3718 h->frame_num_offset= h->prev_frame_num_offset + max_frame_num;
3720 h->frame_num_offset= h->prev_frame_num_offset;
3723 if(h->sps.poc_type==0){
3724 const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
3726 if(h->nal_unit_type == NAL_IDR_SLICE){
3731 if (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
3732 h->poc_msb = h->prev_poc_msb + max_poc_lsb;
3733 else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
3734 h->poc_msb = h->prev_poc_msb - max_poc_lsb;
3736 h->poc_msb = h->prev_poc_msb;
3737 //printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
3739 field_poc[1] = h->poc_msb + h->poc_lsb;
3740 if(s->picture_structure == PICT_FRAME)
3741 field_poc[1] += h->delta_poc_bottom;
3742 }else if(h->sps.poc_type==1){
3743 int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
3746 if(h->sps.poc_cycle_length != 0)
3747 abs_frame_num = h->frame_num_offset + h->frame_num;
3751 if(h->nal_ref_idc==0 && abs_frame_num > 0)
3754 expected_delta_per_poc_cycle = 0;
3755 for(i=0; i < h->sps.poc_cycle_length; i++)
3756 expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
3758 if(abs_frame_num > 0){
3759 int poc_cycle_cnt = (abs_frame_num - 1) / h->sps.poc_cycle_length;
3760 int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
3762 expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
3763 for(i = 0; i <= frame_num_in_poc_cycle; i++)
3764 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
3768 if(h->nal_ref_idc == 0)
3769 expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
3771 field_poc[0] = expectedpoc + h->delta_poc[0];
3772 field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
3774 if(s->picture_structure == PICT_FRAME)
3775 field_poc[1] += h->delta_poc[1];
3778 if(h->nal_unit_type == NAL_IDR_SLICE){
3781 if(h->nal_ref_idc) poc= 2*(h->frame_num_offset + h->frame_num);
3782 else poc= 2*(h->frame_num_offset + h->frame_num) - 1;
3788 if(s->picture_structure != PICT_BOTTOM_FIELD)
3789 s->current_picture_ptr->field_poc[0]= field_poc[0];
3790 if(s->picture_structure != PICT_TOP_FIELD)
3791 s->current_picture_ptr->field_poc[1]= field_poc[1];
3792 if(s->picture_structure == PICT_FRAME) // FIXME field pix?
3793 s->current_picture_ptr->poc= FFMIN(field_poc[0], field_poc[1]);
3799 * decodes a slice header.
3800 * this will allso call MPV_common_init() and frame_start() as needed
3802 static int decode_slice_header(H264Context *h){
3803 MpegEncContext * const s = &h->s;
3804 int first_mb_in_slice, pps_id;
3805 int num_ref_idx_active_override_flag;
3806 static const uint8_t slice_type_map[5]= {P_TYPE, B_TYPE, I_TYPE, SP_TYPE, SI_TYPE};
3808 int default_ref_list_done = 0;
3810 s->current_picture.reference= h->nal_ref_idc != 0;
3811 s->dropable= h->nal_ref_idc == 0;
3813 first_mb_in_slice= get_ue_golomb(&s->gb);
3815 slice_type= get_ue_golomb(&s->gb);
3817 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);
3822 h->slice_type_fixed=1;
3824 h->slice_type_fixed=0;
3826 slice_type= slice_type_map[ slice_type ];
3827 if (slice_type == I_TYPE
3828 || (h->slice_num != 0 && slice_type == h->slice_type) ) {
3829 default_ref_list_done = 1;
3831 h->slice_type= slice_type;
3833 s->pict_type= h->slice_type; // to make a few old func happy, it's wrong though
3835 pps_id= get_ue_golomb(&s->gb);
3837 av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
3840 h->pps= h->pps_buffer[pps_id];
3841 if(h->pps.slice_group_count == 0){
3842 av_log(h->s.avctx, AV_LOG_ERROR, "non existing PPS referenced\n");
3846 h->sps= h->sps_buffer[ h->pps.sps_id ];
3847 if(h->sps.log2_max_frame_num == 0){
3848 av_log(h->s.avctx, AV_LOG_ERROR, "non existing SPS referenced\n");
3852 s->mb_width= h->sps.mb_width;
3853 s->mb_height= h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
3855 h->b_stride= s->mb_width*4 + 1;
3856 h->b8_stride= s->mb_width*2 + 1;
3858 s->width = 16*s->mb_width - 2*(h->sps.crop_left + h->sps.crop_right );
3859 if(h->sps.frame_mbs_only_flag)
3860 s->height= 16*s->mb_height - 2*(h->sps.crop_top + h->sps.crop_bottom);
3862 s->height= 16*s->mb_height - 4*(h->sps.crop_top + h->sps.crop_bottom); //FIXME recheck
3864 if (s->context_initialized
3865 && ( s->width != s->avctx->width || s->height != s->avctx->height)) {
3869 if (!s->context_initialized) {
3870 if (MPV_common_init(s) < 0)
3875 s->avctx->width = s->width;
3876 s->avctx->height = s->height;
3877 s->avctx->sample_aspect_ratio= h->sps.sar;
3878 if(!s->avctx->sample_aspect_ratio.den)
3879 s->avctx->sample_aspect_ratio.den = 1;
3881 if(h->sps.timing_info_present_flag){
3882 s->avctx->time_base= (AVRational){h->sps.num_units_in_tick, h->sps.time_scale};
3886 if(h->slice_num == 0){
3890 s->current_picture_ptr->frame_num= //FIXME frame_num cleanup
3891 h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
3893 h->mb_aff_frame = 0;
3894 if(h->sps.frame_mbs_only_flag){
3895 s->picture_structure= PICT_FRAME;
3897 if(get_bits1(&s->gb)) { //field_pic_flag
3898 s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
3900 s->picture_structure= PICT_FRAME;
3901 first_mb_in_slice <<= 1;
3902 h->mb_aff_frame = h->sps.mb_aff;
3906 s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
3907 s->resync_mb_y = s->mb_y = first_mb_in_slice / s->mb_width;
3909 if(s->picture_structure==PICT_FRAME){
3910 h->curr_pic_num= h->frame_num;
3911 h->max_pic_num= 1<< h->sps.log2_max_frame_num;
3913 h->curr_pic_num= 2*h->frame_num;
3914 h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
3917 if(h->nal_unit_type == NAL_IDR_SLICE){
3918 get_ue_golomb(&s->gb); /* idr_pic_id */
3921 if(h->sps.poc_type==0){
3922 h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
3924 if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
3925 h->delta_poc_bottom= get_se_golomb(&s->gb);
3929 if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
3930 h->delta_poc[0]= get_se_golomb(&s->gb);
3932 if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
3933 h->delta_poc[1]= get_se_golomb(&s->gb);
3938 if(h->pps.redundant_pic_cnt_present){
3939 h->redundant_pic_count= get_ue_golomb(&s->gb);
3942 //set defaults, might be overriden a few line later
3943 h->ref_count[0]= h->pps.ref_count[0];
3944 h->ref_count[1]= h->pps.ref_count[1];
3946 if(h->slice_type == P_TYPE || h->slice_type == SP_TYPE || h->slice_type == B_TYPE){
3947 if(h->slice_type == B_TYPE){
3948 h->direct_spatial_mv_pred= get_bits1(&s->gb);
3950 num_ref_idx_active_override_flag= get_bits1(&s->gb);
3952 if(num_ref_idx_active_override_flag){
3953 h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
3954 if(h->slice_type==B_TYPE)
3955 h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
3957 if(h->ref_count[0] > 32 || h->ref_count[1] > 32){
3958 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
3964 if(!default_ref_list_done){
3965 fill_default_ref_list(h);
3968 decode_ref_pic_list_reordering(h);
3970 if( (h->pps.weighted_pred && (h->slice_type == P_TYPE || h->slice_type == SP_TYPE ))
3971 || (h->pps.weighted_bipred_idc==1 && h->slice_type==B_TYPE ) )
3972 pred_weight_table(h);
3973 else if(h->pps.weighted_bipred_idc==2 && h->slice_type==B_TYPE)
3974 implicit_weight_table(h);
3978 if(s->current_picture.reference)
3979 decode_ref_pic_marking(h);
3981 if( h->slice_type != I_TYPE && h->slice_type != SI_TYPE && h->pps.cabac )
3982 h->cabac_init_idc = get_ue_golomb(&s->gb);
3984 h->last_qscale_diff = 0;
3985 s->qscale = h->pps.init_qp + get_se_golomb(&s->gb);
3986 if(s->qscale<0 || s->qscale>51){
3987 av_log(s->avctx, AV_LOG_ERROR, "QP %d out of range\n", s->qscale);
3990 h->chroma_qp = get_chroma_qp(h->pps.chroma_qp_index_offset, s->qscale);
3991 //FIXME qscale / qp ... stuff
3992 if(h->slice_type == SP_TYPE){
3993 get_bits1(&s->gb); /* sp_for_switch_flag */
3995 if(h->slice_type==SP_TYPE || h->slice_type == SI_TYPE){
3996 get_se_golomb(&s->gb); /* slice_qs_delta */
3999 h->deblocking_filter = 1;
4000 h->slice_alpha_c0_offset = 0;
4001 h->slice_beta_offset = 0;
4002 if( h->pps.deblocking_filter_parameters_present ) {
4003 h->deblocking_filter= get_ue_golomb(&s->gb);
4004 if(h->deblocking_filter < 2)
4005 h->deblocking_filter^= 1; // 1<->0
4007 if( h->deblocking_filter ) {
4008 h->slice_alpha_c0_offset = get_se_golomb(&s->gb) << 1;
4009 h->slice_beta_offset = get_se_golomb(&s->gb) << 1;
4014 if( h->pps.num_slice_groups > 1 && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
4015 slice_group_change_cycle= get_bits(&s->gb, ?);
4020 if(s->avctx->debug&FF_DEBUG_PICT_INFO){
4021 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",
4023 (s->picture_structure==PICT_FRAME ? "F" : s->picture_structure==PICT_TOP_FIELD ? "T" : "B"),
4025 av_get_pict_type_char(h->slice_type),
4026 pps_id, h->frame_num,
4027 s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
4028 h->ref_count[0], h->ref_count[1],
4030 h->deblocking_filter, h->slice_alpha_c0_offset/2, h->slice_beta_offset/2,
4032 h->use_weight==1 && h->use_weight_chroma ? "c" : ""
4042 static inline int get_level_prefix(GetBitContext *gb){
4046 OPEN_READER(re, gb);
4047 UPDATE_CACHE(re, gb);
4048 buf=GET_CACHE(re, gb);
4050 log= 32 - av_log2(buf);
4052 print_bin(buf>>(32-log), log);
4053 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__);
4056 LAST_SKIP_BITS(re, gb, log);
4057 CLOSE_READER(re, gb);
4063 * decodes a residual block.
4064 * @param n block index
4065 * @param scantable scantable
4066 * @param max_coeff number of coefficients in the block
4067 * @return <0 if an error occured
4069 static int decode_residual(H264Context *h, GetBitContext *gb, DCTELEM *block, int n, const uint8_t *scantable, int qp, int max_coeff){
4070 MpegEncContext * const s = &h->s;
4071 const uint16_t *qmul= dequant_coeff[qp];
4072 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};
4073 int level[16], run[16];
4074 int suffix_length, zeros_left, coeff_num, coeff_token, total_coeff, i, trailing_ones;
4076 //FIXME put trailing_onex into the context
4078 if(n == CHROMA_DC_BLOCK_INDEX){
4079 coeff_token= get_vlc2(gb, chroma_dc_coeff_token_vlc.table, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 1);
4080 total_coeff= coeff_token>>2;
4082 if(n == LUMA_DC_BLOCK_INDEX){
4083 total_coeff= pred_non_zero_count(h, 0);
4084 coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
4085 total_coeff= coeff_token>>2;
4087 total_coeff= pred_non_zero_count(h, n);
4088 coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
4089 total_coeff= coeff_token>>2;
4090 h->non_zero_count_cache[ scan8[n] ]= total_coeff;
4094 //FIXME set last_non_zero?
4099 trailing_ones= coeff_token&3;
4100 tprintf("trailing:%d, total:%d\n", trailing_ones, total_coeff);
4101 assert(total_coeff<=16);
4103 for(i=0; i<trailing_ones; i++){
4104 level[i]= 1 - 2*get_bits1(gb);
4107 suffix_length= total_coeff > 10 && trailing_ones < 3;
4109 for(; i<total_coeff; i++){
4110 const int prefix= get_level_prefix(gb);
4111 int level_code, mask;
4113 if(prefix<14){ //FIXME try to build a large unified VLC table for all this
4115 level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
4117 level_code= (prefix<<suffix_length); //part
4118 }else if(prefix==14){
4120 level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
4122 level_code= prefix + get_bits(gb, 4); //part
4123 }else if(prefix==15){
4124 level_code= (prefix<<suffix_length) + get_bits(gb, 12); //part
4125 if(suffix_length==0) level_code+=15; //FIXME doesn't make (much)sense
4127 av_log(h->s.avctx, AV_LOG_ERROR, "prefix too large at %d %d\n", s->mb_x, s->mb_y);
4131 if(i==trailing_ones && i<3) level_code+= 2; //FIXME split first iteration
4133 mask= -(level_code&1);
4134 level[i]= (((2+level_code)>>1) ^ mask) - mask;
4136 if(suffix_length==0) suffix_length=1; //FIXME split first iteration
4139 if(ABS(level[i]) > (3<<(suffix_length-1)) && suffix_length<6) suffix_length++;
4141 if((2+level_code)>>1) > (3<<(suffix_length-1)) && suffix_length<6) suffix_length++;
4142 /* ? == prefix > 2 or sth */
4144 tprintf("level: %d suffix_length:%d\n", level[i], suffix_length);
4147 if(total_coeff == max_coeff)
4150 if(n == CHROMA_DC_BLOCK_INDEX)
4151 zeros_left= get_vlc2(gb, chroma_dc_total_zeros_vlc[ total_coeff-1 ].table, CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 1);
4153 zeros_left= get_vlc2(gb, total_zeros_vlc[ total_coeff-1 ].table, TOTAL_ZEROS_VLC_BITS, 1);
4156 for(i=0; i<total_coeff-1; i++){
4159 else if(zeros_left < 7){
4160 run[i]= get_vlc2(gb, run_vlc[zeros_left-1].table, RUN_VLC_BITS, 1);
4162 run[i]= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2);
4164 zeros_left -= run[i];
4168 av_log(h->s.avctx, AV_LOG_ERROR, "negative number of zero coeffs at %d %d\n", s->mb_x, s->mb_y);
4172 for(; i<total_coeff-1; i++){
4180 for(i=total_coeff-1; i>=0; i--){ //FIXME merge into rundecode?
4183 coeff_num += run[i] + 1; //FIXME add 1 earlier ?
4184 j= scantable[ coeff_num ];
4189 for(i=total_coeff-1; i>=0; i--){ //FIXME merge into rundecode?
4192 coeff_num += run[i] + 1; //FIXME add 1 earlier ?
4193 j= scantable[ coeff_num ];
4195 block[j]= level[i] * qmul[j];
4196 // printf("%d %d ", block[j], qmul[j]);
4203 * decodes a P_SKIP or B_SKIP macroblock
4205 static void decode_mb_skip(H264Context *h){
4206 MpegEncContext * const s = &h->s;
4207 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
4210 memset(h->non_zero_count[mb_xy], 0, 16);
4211 memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui
4213 if(h->mb_aff_frame && s->mb_skip_run==0 && (s->mb_y&1)==0){
4214 h->mb_field_decoding_flag= get_bits1(&s->gb);
4216 if(h->mb_field_decoding_flag)
4217 mb_type|= MB_TYPE_INTERLACED;
4219 if( h->slice_type == B_TYPE )
4221 // just for fill_caches. pred_direct_motion will set the real mb_type
4222 mb_type|= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2|MB_TYPE_SKIP;
4224 fill_caches(h, mb_type, 0); //FIXME check what is needed and what not ...
4225 pred_direct_motion(h, &mb_type);
4227 fill_rectangle(h->mvd_cache[0][scan8[0]], 4, 4, 8, 0, 4);
4228 fill_rectangle(h->mvd_cache[1][scan8[0]], 4, 4, 8, 0, 4);
4234 mb_type|= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0|MB_TYPE_SKIP;
4236 fill_caches(h, mb_type, 0); //FIXME check what is needed and what not ...
4237 pred_pskip_motion(h, &mx, &my);
4238 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
4239 fill_rectangle( h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4);
4241 fill_rectangle(h->mvd_cache[0][scan8[0]], 4, 4, 8, 0, 4);
4244 write_back_motion(h, mb_type);
4245 s->current_picture.mb_type[mb_xy]= mb_type|MB_TYPE_SKIP;
4246 s->current_picture.qscale_table[mb_xy]= s->qscale;
4247 h->slice_table[ mb_xy ]= h->slice_num;
4248 h->prev_mb_skipped= 1;
4252 * decodes a macroblock
4253 * @returns 0 if ok, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
4255 static int decode_mb_cavlc(H264Context *h){
4256 MpegEncContext * const s = &h->s;
4257 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
4258 int mb_type, partition_count, cbp;
4260 s->dsp.clear_blocks(h->mb); //FIXME avoid if already clear (move after skip handlong?
4262 tprintf("pic:%d mb:%d/%d\n", h->frame_num, s->mb_x, s->mb_y);
4263 cbp = 0; /* avoid warning. FIXME: find a solution without slowing
4265 if(h->slice_type != I_TYPE && h->slice_type != SI_TYPE){
4266 if(s->mb_skip_run==-1)
4267 s->mb_skip_run= get_ue_golomb(&s->gb);
4269 if (s->mb_skip_run--) {
4274 if(h->mb_aff_frame){
4275 if ( ((s->mb_y&1) == 0) || h->prev_mb_skipped)
4276 h->mb_field_decoding_flag = get_bits1(&s->gb);
4278 h->mb_field_decoding_flag= (s->picture_structure!=PICT_FRAME);
4280 h->prev_mb_skipped= 0;
4282 mb_type= get_ue_golomb(&s->gb);
4283 if(h->slice_type == B_TYPE){
4285 partition_count= b_mb_type_info[mb_type].partition_count;
4286 mb_type= b_mb_type_info[mb_type].type;
4289 goto decode_intra_mb;
4291 }else if(h->slice_type == P_TYPE /*|| h->slice_type == SP_TYPE */){
4293 partition_count= p_mb_type_info[mb_type].partition_count;
4294 mb_type= p_mb_type_info[mb_type].type;
4297 goto decode_intra_mb;
4300 assert(h->slice_type == I_TYPE);
4303 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);
4307 cbp= i_mb_type_info[mb_type].cbp;
4308 h->intra16x16_pred_mode= i_mb_type_info[mb_type].pred_mode;
4309 mb_type= i_mb_type_info[mb_type].type;
4312 if(h->mb_field_decoding_flag)
4313 mb_type |= MB_TYPE_INTERLACED;
4315 s->current_picture.mb_type[mb_xy]= mb_type;
4316 h->slice_table[ mb_xy ]= h->slice_num;
4318 if(IS_INTRA_PCM(mb_type)){
4321 // we assume these blocks are very rare so we dont optimize it
4322 align_get_bits(&s->gb);
4324 // The pixels are stored in the same order as levels in h->mb array.
4325 for(y=0; y<16; y++){
4326 const int index= 4*(y&3) + 32*((y>>2)&1) + 128*(y>>3);
4327 for(x=0; x<16; x++){
4328 tprintf("LUMA ICPM LEVEL (%3d)\n", show_bits(&s->gb, 8));
4329 h->mb[index + (x&3) + 16*((x>>2)&1) + 64*(x>>3)]= get_bits(&s->gb, 8);
4333 const int index= 256 + 4*(y&3) + 32*(y>>2);
4335 tprintf("CHROMA U ICPM LEVEL (%3d)\n", show_bits(&s->gb, 8));
4336 h->mb[index + (x&3) + 16*(x>>2)]= get_bits(&s->gb, 8);
4340 const int index= 256 + 64 + 4*(y&3) + 32*(y>>2);
4342 tprintf("CHROMA V ICPM LEVEL (%3d)\n", show_bits(&s->gb, 8));
4343 h->mb[index + (x&3) + 16*(x>>2)]= get_bits(&s->gb, 8);
4347 // In deblocking, the quantizer is 0
4348 s->current_picture.qscale_table[mb_xy]= 0;
4349 h->chroma_qp = get_chroma_qp(h->pps.chroma_qp_index_offset, 0);
4350 // All coeffs are present
4351 memset(h->non_zero_count[mb_xy], 16, 16);
4356 fill_caches(h, mb_type, 0);
4359 if(IS_INTRA(mb_type)){
4360 // init_top_left_availability(h);
4361 if(IS_INTRA4x4(mb_type)){
4364 // fill_intra4x4_pred_table(h);
4365 for(i=0; i<16; i++){
4366 const int mode_coded= !get_bits1(&s->gb);
4367 const int predicted_mode= pred_intra_mode(h, i);
4371 const int rem_mode= get_bits(&s->gb, 3);
4372 if(rem_mode<predicted_mode)
4377 mode= predicted_mode;
4380 h->intra4x4_pred_mode_cache[ scan8[i] ] = mode;
4382 write_back_intra_pred_mode(h);
4383 if( check_intra4x4_pred_mode(h) < 0)
4386 h->intra16x16_pred_mode= check_intra_pred_mode(h, h->intra16x16_pred_mode);
4387 if(h->intra16x16_pred_mode < 0)
4390 h->chroma_pred_mode= get_ue_golomb(&s->gb);
4392 h->chroma_pred_mode= check_intra_pred_mode(h, h->chroma_pred_mode);
4393 if(h->chroma_pred_mode < 0)
4395 }else if(partition_count==4){
4396 int i, j, sub_partition_count[4], list, ref[2][4];
4398 if(h->slice_type == B_TYPE){
4400 h->sub_mb_type[i]= get_ue_golomb(&s->gb);
4401 if(h->sub_mb_type[i] >=13){
4402 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);
4405 sub_partition_count[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
4406 h->sub_mb_type[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].type;
4408 if( IS_DIRECT(h->sub_mb_type[0]) || IS_DIRECT(h->sub_mb_type[1])
4409 || IS_DIRECT(h->sub_mb_type[2]) || IS_DIRECT(h->sub_mb_type[3]))
4410 pred_direct_motion(h, &mb_type);
4412 assert(h->slice_type == P_TYPE || h->slice_type == SP_TYPE); //FIXME SP correct ?
4414 h->sub_mb_type[i]= get_ue_golomb(&s->gb);
4415 if(h->sub_mb_type[i] >=4){
4416 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);
4419 sub_partition_count[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
4420 h->sub_mb_type[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].type;
4424 for(list=0; list<2; list++){
4425 int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list];
4426 if(ref_count == 0) continue;
4427 if (h->mb_aff_frame && h->mb_field_decoding_flag) {
4431 if(IS_DIRECT(h->sub_mb_type[i])) continue;
4432 if(IS_DIR(h->sub_mb_type[i], 0, list)){
4433 ref[list][i] = get_te0_golomb(&s->gb, ref_count); //FIXME init to 0 before and skip?
4441 for(list=0; list<2; list++){
4442 const int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list];
4443 if(ref_count == 0) continue;
4446 if(IS_DIRECT(h->sub_mb_type[i])) continue;
4447 h->ref_cache[list][ scan8[4*i] ]=h->ref_cache[list][ scan8[4*i]+1 ]=
4448 h->ref_cache[list][ scan8[4*i]+8 ]=h->ref_cache[list][ scan8[4*i]+9 ]= ref[list][i];
4450 if(IS_DIR(h->sub_mb_type[i], 0, list)){
4451 const int sub_mb_type= h->sub_mb_type[i];
4452 const int block_width= (sub_mb_type & (MB_TYPE_16x16|MB_TYPE_16x8)) ? 2 : 1;
4453 for(j=0; j<sub_partition_count[i]; j++){
4455 const int index= 4*i + block_width*j;
4456 int16_t (* mv_cache)[2]= &h->mv_cache[list][ scan8[index] ];
4457 pred_motion(h, index, block_width, list, h->ref_cache[list][ scan8[index] ], &mx, &my);
4458 mx += get_se_golomb(&s->gb);
4459 my += get_se_golomb(&s->gb);
4460 tprintf("final mv:%d %d\n", mx, my);
4462 if(IS_SUB_8X8(sub_mb_type)){
4463 mv_cache[ 0 ][0]= mv_cache[ 1 ][0]=
4464 mv_cache[ 8 ][0]= mv_cache[ 9 ][0]= mx;
4465 mv_cache[ 0 ][1]= mv_cache[ 1 ][1]=
4466 mv_cache[ 8 ][1]= mv_cache[ 9 ][1]= my;
4467 }else if(IS_SUB_8X4(sub_mb_type)){
4468 mv_cache[ 0 ][0]= mv_cache[ 1 ][0]= mx;
4469 mv_cache[ 0 ][1]= mv_cache[ 1 ][1]= my;
4470 }else if(IS_SUB_4X8(sub_mb_type)){
4471 mv_cache[ 0 ][0]= mv_cache[ 8 ][0]= mx;
4472 mv_cache[ 0 ][1]= mv_cache[ 8 ][1]= my;
4474 assert(IS_SUB_4X4(sub_mb_type));
4475 mv_cache[ 0 ][0]= mx;
4476 mv_cache[ 0 ][1]= my;
4480 uint32_t *p= (uint32_t *)&h->mv_cache[list][ scan8[4*i] ][0];
4486 }else if(IS_DIRECT(mb_type)){
4487 pred_direct_motion(h, &mb_type);
4488 s->current_picture.mb_type[mb_xy]= mb_type;
4490 int list, mx, my, i;
4491 //FIXME we should set ref_idx_l? to 0 if we use that later ...
4492 if(IS_16X16(mb_type)){
4493 for(list=0; list<2; list++){
4494 if(h->ref_count[list]>0){
4495 if(IS_DIR(mb_type, 0, list)){
4496 const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
4497 fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, val, 1);
4499 fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, (LIST_NOT_USED&0xFF), 1);
4502 for(list=0; list<2; list++){
4503 if(IS_DIR(mb_type, 0, list)){
4504 pred_motion(h, 0, 4, list, h->ref_cache[list][ scan8[0] ], &mx, &my);
4505 mx += get_se_golomb(&s->gb);
4506 my += get_se_golomb(&s->gb);
4507 tprintf("final mv:%d %d\n", mx, my);
4509 fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, pack16to32(mx,my), 4);
4511 fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, 0, 4);
4514 else if(IS_16X8(mb_type)){
4515 for(list=0; list<2; list++){
4516 if(h->ref_count[list]>0){
4518 if(IS_DIR(mb_type, i, list)){
4519 const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
4520 fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, val, 1);
4522 fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, (LIST_NOT_USED&0xFF), 1);
4526 for(list=0; list<2; list++){
4528 if(IS_DIR(mb_type, i, list)){
4529 pred_16x8_motion(h, 8*i, list, h->ref_cache[list][scan8[0] + 16*i], &mx, &my);
4530 mx += get_se_golomb(&s->gb);
4531 my += get_se_golomb(&s->gb);
4532 tprintf("final mv:%d %d\n", mx, my);
4534 fill_rectangle(h->mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, pack16to32(mx,my), 4);
4536 fill_rectangle(h->mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, 0, 4);
4540 assert(IS_8X16(mb_type));
4541 for(list=0; list<2; list++){
4542 if(h->ref_count[list]>0){
4544 if(IS_DIR(mb_type, i, list)){ //FIXME optimize
4545 const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
4546 fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, val, 1);
4548 fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, (LIST_NOT_USED&0xFF), 1);
4552 for(list=0; list<2; list++){
4554 if(IS_DIR(mb_type, i, list)){
4555 pred_8x16_motion(h, i*4, list, h->ref_cache[list][ scan8[0] + 2*i ], &mx, &my);
4556 mx += get_se_golomb(&s->gb);
4557 my += get_se_golomb(&s->gb);
4558 tprintf("final mv:%d %d\n", mx, my);
4560 fill_rectangle(h->mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, pack16to32(mx,my), 4);
4562 fill_rectangle(h->mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, 0, 4);
4568 if(IS_INTER(mb_type))
4569 write_back_motion(h, mb_type);
4571 if(!IS_INTRA16x16(mb_type)){
4572 cbp= get_ue_golomb(&s->gb);
4574 av_log(h->s.avctx, AV_LOG_ERROR, "cbp too large (%d) at %d %d\n", cbp, s->mb_x, s->mb_y);
4578 if(IS_INTRA4x4(mb_type))
4579 cbp= golomb_to_intra4x4_cbp[cbp];
4581 cbp= golomb_to_inter_cbp[cbp];
4584 if(cbp || IS_INTRA16x16(mb_type)){
4585 int i8x8, i4x4, chroma_idx;
4586 int chroma_qp, dquant;
4587 GetBitContext *gb= IS_INTRA(mb_type) ? h->intra_gb_ptr : h->inter_gb_ptr;
4588 const uint8_t *scan, *dc_scan;
4590 // fill_non_zero_count_cache(h);
4592 if(IS_INTERLACED(mb_type)){
4594 dc_scan= luma_dc_field_scan;
4597 dc_scan= luma_dc_zigzag_scan;
4600 dquant= get_se_golomb(&s->gb);
4602 if( dquant > 25 || dquant < -26 ){
4603 av_log(h->s.avctx, AV_LOG_ERROR, "dquant out of range (%d) at %d %d\n", dquant, s->mb_x, s->mb_y);
4607 s->qscale += dquant;
4608 if(((unsigned)s->qscale) > 51){
4609 if(s->qscale<0) s->qscale+= 52;
4610 else s->qscale-= 52;
4613 h->chroma_qp= chroma_qp= get_chroma_qp(h->pps.chroma_qp_index_offset, s->qscale);
4614 if(IS_INTRA16x16(mb_type)){
4615 if( decode_residual(h, h->intra_gb_ptr, h->mb, LUMA_DC_BLOCK_INDEX, dc_scan, s->qscale, 16) < 0){
4616 return -1; //FIXME continue if partitioned and other return -1 too
4619 assert((cbp&15) == 0 || (cbp&15) == 15);
4622 for(i8x8=0; i8x8<4; i8x8++){
4623 for(i4x4=0; i4x4<4; i4x4++){
4624 const int index= i4x4 + 4*i8x8;
4625 if( decode_residual(h, h->intra_gb_ptr, h->mb + 16*index, index, scan + 1, s->qscale, 15) < 0 ){
4631 fill_rectangle(&h->non_zero_count_cache[scan8[0]], 4, 4, 8, 0, 1);
4634 for(i8x8=0; i8x8<4; i8x8++){
4635 if(cbp & (1<<i8x8)){
4636 for(i4x4=0; i4x4<4; i4x4++){
4637 const int index= i4x4 + 4*i8x8;
4639 if( decode_residual(h, gb, h->mb + 16*index, index, scan, s->qscale, 16) <0 ){
4644 uint8_t * const nnz= &h->non_zero_count_cache[ scan8[4*i8x8] ];
4645 nnz[0] = nnz[1] = nnz[8] = nnz[9] = 0;
4651 for(chroma_idx=0; chroma_idx<2; chroma_idx++)
4652 if( decode_residual(h, gb, h->mb + 256 + 16*4*chroma_idx, CHROMA_DC_BLOCK_INDEX, chroma_dc_scan, chroma_qp, 4) < 0){
4658 for(chroma_idx=0; chroma_idx<2; chroma_idx++){
4659 for(i4x4=0; i4x4<4; i4x4++){
4660 const int index= 16 + 4*chroma_idx + i4x4;
4661 if( decode_residual(h, gb, h->mb + 16*index, index, scan + 1, chroma_qp, 15) < 0){
4667 uint8_t * const nnz= &h->non_zero_count_cache[0];
4668 nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =
4669 nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
4672 uint8_t * const nnz= &h->non_zero_count_cache[0];
4673 fill_rectangle(&nnz[scan8[0]], 4, 4, 8, 0, 1);
4674 nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =
4675 nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
4677 s->current_picture.qscale_table[mb_xy]= s->qscale;
4678 write_back_non_zero_count(h);
4683 static int decode_cabac_field_decoding_flag(H264Context *h) {
4684 MpegEncContext * const s = &h->s;
4685 const int mb_x = s->mb_x;
4686 const int mb_y = s->mb_y & ~1;
4687 const int mba_xy = mb_x - 1 + mb_y *s->mb_stride;
4688 const int mbb_xy = mb_x + (mb_y-2)*s->mb_stride;
4690 unsigned int ctx = 0;
4692 if( h->slice_table[mba_xy] == h->slice_num && IS_INTERLACED( s->current_picture.mb_type[mba_xy] ) ) {
4695 if( h->slice_table[mbb_xy] == h->slice_num && IS_INTERLACED( s->current_picture.mb_type[mbb_xy] ) ) {
4699 return get_cabac( &h->cabac, &h->cabac_state[70 + ctx] );
4702 static int decode_cabac_intra_mb_type(H264Context *h, int ctx_base, int intra_slice) {
4703 uint8_t *state= &h->cabac_state[ctx_base];
4707 MpegEncContext * const s = &h->s;
4708 const int mba_xy = h->left_mb_xy[0];
4709 const int mbb_xy = h->top_mb_xy;
4711 if( h->slice_table[mba_xy] == h->slice_num && !IS_INTRA4x4( s->current_picture.mb_type[mba_xy] ) )
4713 if( h->slice_table[mbb_xy] == h->slice_num && !IS_INTRA4x4( s->current_picture.mb_type[mbb_xy] ) )
4715 if( get_cabac( &h->cabac, &state[ctx] ) == 0 )
4716 return 0; /* I4x4 */
4719 if( get_cabac( &h->cabac, &state[0] ) == 0 )
4720 return 0; /* I4x4 */
4723 if( get_cabac_terminate( &h->cabac ) )
4724 return 25; /* PCM */
4726 mb_type = 1; /* I16x16 */
4727 if( get_cabac( &h->cabac, &state[1] ) )
4728 mb_type += 12; /* cbp_luma != 0 */
4730 if( get_cabac( &h->cabac, &state[2] ) ) {
4731 if( get_cabac( &h->cabac, &state[2+intra_slice] ) )
4732 mb_type += 4 * 2; /* cbp_chroma == 2 */
4734 mb_type += 4 * 1; /* cbp_chroma == 1 */
4736 if( get_cabac( &h->cabac, &state[3+intra_slice] ) )
4738 if( get_cabac( &h->cabac, &state[3+2*intra_slice] ) )
4743 static int decode_cabac_mb_type( H264Context *h ) {
4744 MpegEncContext * const s = &h->s;
4746 if( h->slice_type == I_TYPE ) {
4747 return decode_cabac_intra_mb_type(h, 3, 1);
4748 } else if( h->slice_type == P_TYPE ) {
4749 if( get_cabac( &h->cabac, &h->cabac_state[14] ) == 0 ) {
4751 if( get_cabac( &h->cabac, &h->cabac_state[15] ) == 0 ) {
4752 if( get_cabac( &h->cabac, &h->cabac_state[16] ) == 0 )
4753 return 0; /* P_L0_D16x16; */
4755 return 3; /* P_8x8; */
4757 if( get_cabac( &h->cabac, &h->cabac_state[17] ) == 0 )
4758 return 2; /* P_L0_D8x16; */
4760 return 1; /* P_L0_D16x8; */
4763 return decode_cabac_intra_mb_type(h, 17, 0) + 5;
4765 } else if( h->slice_type == B_TYPE ) {
4766 const int mba_xy = h->left_mb_xy[0];
4767 const int mbb_xy = h->top_mb_xy;
4771 if( h->slice_table[mba_xy] == h->slice_num && !IS_SKIP( s->current_picture.mb_type[mba_xy] )
4772 && !IS_DIRECT( s->current_picture.mb_type[mba_xy] ) )
4774 if( h->slice_table[mbb_xy] == h->slice_num && !IS_SKIP( s->current_picture.mb_type[mbb_xy] )
4775 && !IS_DIRECT( s->current_picture.mb_type[mbb_xy] ) )
4778 if( !get_cabac( &h->cabac, &h->cabac_state[27+ctx] ) )
4779 return 0; /* B_Direct_16x16 */
4781 if( !get_cabac( &h->cabac, &h->cabac_state[27+3] ) ) {
4782 return 1 + get_cabac( &h->cabac, &h->cabac_state[27+5] ); /* B_L[01]_16x16 */
4785 bits = get_cabac( &h->cabac, &h->cabac_state[27+4] ) << 3;
4786 bits|= get_cabac( &h->cabac, &h->cabac_state[27+5] ) << 2;
4787 bits|= get_cabac( &h->cabac, &h->cabac_state[27+5] ) << 1;
4788 bits|= get_cabac( &h->cabac, &h->cabac_state[27+5] );
4790 return bits + 3; /* B_Bi_16x16 through B_L1_L0_16x8 */
4791 else if( bits == 13 ) {
4792 return decode_cabac_intra_mb_type(h, 32, 0) + 23;
4793 } else if( bits == 14 )
4794 return 11; /* B_L1_L0_8x16 */
4795 else if( bits == 15 )
4796 return 22; /* B_8x8 */
4798 bits= ( bits<<1 ) | get_cabac( &h->cabac, &h->cabac_state[27+5] );
4799 return bits - 4; /* B_L0_Bi_* through B_Bi_Bi_* */
4801 /* TODO SI/SP frames? */
4806 static int decode_cabac_mb_skip( H264Context *h) {
4807 MpegEncContext * const s = &h->s;
4808 const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
4809 const int mba_xy = mb_xy - 1;
4810 const int mbb_xy = mb_xy - s->mb_stride;
4813 if( h->slice_table[mba_xy] == h->slice_num && !IS_SKIP( s->current_picture.mb_type[mba_xy] ))
4815 if( h->slice_table[mbb_xy] == h->slice_num && !IS_SKIP( s->current_picture.mb_type[mbb_xy] ))
4818 if( h->slice_type == P_TYPE || h->slice_type == SP_TYPE)
4819 return get_cabac( &h->cabac, &h->cabac_state[11+ctx] );
4821 return get_cabac( &h->cabac, &h->cabac_state[24+ctx] );
4824 static int decode_cabac_mb_intra4x4_pred_mode( H264Context *h, int pred_mode ) {
4827 if( get_cabac( &h->cabac, &h->cabac_state[68] ) )
4830 if( get_cabac( &h->cabac, &h->cabac_state[69] ) )
4832 if( get_cabac( &h->cabac, &h->cabac_state[69] ) )
4834 if( get_cabac( &h->cabac, &h->cabac_state[69] ) )
4836 if( mode >= pred_mode )
4842 static int decode_cabac_mb_chroma_pre_mode( H264Context *h) {
4843 const int mba_xy = h->left_mb_xy[0];
4844 const int mbb_xy = h->top_mb_xy;
4848 /* No need to test for IS_INTRA4x4 and IS_INTRA16x16, as we set chroma_pred_mode_table to 0 */
4849 if( h->slice_table[mba_xy] == h->slice_num && h->chroma_pred_mode_table[mba_xy] != 0 )
4852 if( h->slice_table[mbb_xy] == h->slice_num && h->chroma_pred_mode_table[mbb_xy] != 0 )
4855 if( get_cabac( &h->cabac, &h->cabac_state[64+ctx] ) == 0 )
4858 if( get_cabac( &h->cabac, &h->cabac_state[64+3] ) == 0 )
4860 if( get_cabac( &h->cabac, &h->cabac_state[64+3] ) == 0 )
4866 static const uint8_t block_idx_x[16] = {
4867 0, 1, 0, 1, 2, 3, 2, 3, 0, 1, 0, 1, 2, 3, 2, 3
4869 static const uint8_t block_idx_y[16] = {
4870 0, 0, 1, 1, 0, 0, 1, 1, 2, 2, 3, 3, 2, 2, 3, 3
4872 static const uint8_t block_idx_xy[4][4] = {
4879 static int decode_cabac_mb_cbp_luma( H264Context *h) {
4880 MpegEncContext * const s = &h->s;
4885 for( i8x8 = 0; i8x8 < 4; i8x8++ ) {
4891 x = block_idx_x[4*i8x8];
4892 y = block_idx_y[4*i8x8];
4896 else if( s->mb_x > 0 && (h->slice_table[h->left_mb_xy[0]] == h->slice_num)) {
4897 cbp_a = h->left_cbp;
4898 tprintf("cbp_a = left_cbp = %x\n", cbp_a);
4903 else if( s->mb_y > 0 && (h->slice_table[h->top_mb_xy] == h->slice_num)) {
4905 tprintf("cbp_b = top_cbp = %x\n", cbp_b);
4908 /* No need to test for skip as we put 0 for skip block */
4909 /* No need to test for IPCM as we put 1 for IPCM block */
4911 int i8x8a = block_idx_xy[(x-1)&0x03][y]/4;
4912 if( ((cbp_a >> i8x8a)&0x01) == 0 )
4917 int i8x8b = block_idx_xy[x][(y-1)&0x03]/4;
4918 if( ((cbp_b >> i8x8b)&0x01) == 0 )
4922 if( get_cabac( &h->cabac, &h->cabac_state[73 + ctx] ) ) {
4928 static int decode_cabac_mb_cbp_chroma( H264Context *h) {
4932 cbp_a = (h->left_cbp>>4)&0x03;
4933 cbp_b = (h-> top_cbp>>4)&0x03;
4936 if( cbp_a > 0 ) ctx++;
4937 if( cbp_b > 0 ) ctx += 2;
4938 if( get_cabac( &h->cabac, &h->cabac_state[77 + ctx] ) == 0 )
4942 if( cbp_a == 2 ) ctx++;
4943 if( cbp_b == 2 ) ctx += 2;
4944 return 1 + get_cabac( &h->cabac, &h->cabac_state[77 + ctx] );
4946 static int decode_cabac_mb_dqp( H264Context *h) {
4947 MpegEncContext * const s = &h->s;
4953 mbn_xy = s->mb_x + s->mb_y*s->mb_stride - 1;
4955 mbn_xy = s->mb_width - 1 + (s->mb_y-1)*s->mb_stride;
4957 if( h->last_qscale_diff != 0 && ( IS_INTRA16x16(s->current_picture.mb_type[mbn_xy] ) || (h->cbp_table[mbn_xy]&0x3f) ) )
4960 while( get_cabac( &h->cabac, &h->cabac_state[60 + ctx] ) ) {
4971 return -(val + 1)/2;
4973 static int decode_cabac_p_mb_sub_type( H264Context *h ) {
4974 if( get_cabac( &h->cabac, &h->cabac_state[21] ) )
4976 if( !get_cabac( &h->cabac, &h->cabac_state[22] ) )
4978 if( get_cabac( &h->cabac, &h->cabac_state[23] ) )
4982 static int decode_cabac_b_mb_sub_type( H264Context *h ) {
4984 if( !get_cabac( &h->cabac, &h->cabac_state[36] ) )
4985 return 0; /* B_Direct_8x8 */
4986 if( !get_cabac( &h->cabac, &h->cabac_state[37] ) )
4987 return 1 + get_cabac( &h->cabac, &h->cabac_state[39] ); /* B_L0_8x8, B_L1_8x8 */
4989 if( get_cabac( &h->cabac, &h->cabac_state[38] ) ) {
4990 if( get_cabac( &h->cabac, &h->cabac_state[39] ) )
4991 return 11 + get_cabac( &h->cabac, &h->cabac_state[39] ); /* B_L1_4x4, B_Bi_4x4 */
4994 type += 2*get_cabac( &h->cabac, &h->cabac_state[39] );
4995 type += get_cabac( &h->cabac, &h->cabac_state[39] );
4999 static int decode_cabac_mb_ref( H264Context *h, int list, int n ) {
5000 int refa = h->ref_cache[list][scan8[n] - 1];
5001 int refb = h->ref_cache[list][scan8[n] - 8];
5005 if( h->slice_type == B_TYPE) {
5006 if( refa > 0 && !h->direct_cache[scan8[n] - 1] )
5008 if( refb > 0 && !h->direct_cache[scan8[n] - 8] )
5017 while( get_cabac( &h->cabac, &h->cabac_state[54+ctx] ) ) {
5027 static int decode_cabac_mb_mvd( H264Context *h, int list, int n, int l ) {
5028 int amvd = abs( h->mvd_cache[list][scan8[n] - 1][l] ) +
5029 abs( h->mvd_cache[list][scan8[n] - 8][l] );
5030 int ctxbase = (l == 0) ? 40 : 47;
5035 else if( amvd > 32 )
5040 if(!get_cabac(&h->cabac, &h->cabac_state[ctxbase+ctx]))
5045 while( mvd < 9 && get_cabac( &h->cabac, &h->cabac_state[ctxbase+ctx] ) ) {
5053 while( get_cabac_bypass( &h->cabac ) ) {
5058 if( get_cabac_bypass( &h->cabac ) )
5062 if( get_cabac_bypass( &h->cabac ) ) return -mvd;
5066 static int inline get_cabac_cbf_ctx( H264Context *h, int cat, int idx ) {
5071 nza = h->left_cbp&0x100;
5072 nzb = h-> top_cbp&0x100;
5073 } else if( cat == 1 || cat == 2 ) {
5074 nza = h->non_zero_count_cache[scan8[idx] - 1];
5075 nzb = h->non_zero_count_cache[scan8[idx] - 8];
5076 } else if( cat == 3 ) {
5077 nza = (h->left_cbp>>(6+idx))&0x01;
5078 nzb = (h-> top_cbp>>(6+idx))&0x01;
5081 nza = h->non_zero_count_cache[scan8[16+idx] - 1];
5082 nzb = h->non_zero_count_cache[scan8[16+idx] - 8];
5091 return ctx + 4 * cat;
5094 static int inline decode_cabac_residual( H264Context *h, DCTELEM *block, int cat, int n, const uint8_t *scantable, int qp, int max_coeff) {
5095 const int mb_xy = h->s.mb_x + h->s.mb_y*h->s.mb_stride;
5096 const uint16_t *qmul= dequant_coeff[qp];
5097 static const int significant_coeff_flag_field_offset[2] = { 105, 277 };
5098 static const int last_significant_coeff_flag_field_offset[2] = { 166, 338 };
5099 static const int significant_coeff_flag_offset[5] = { 0, 15, 29, 44, 47 };
5100 static const int coeff_abs_level_m1_offset[5] = {227+ 0, 227+10, 227+20, 227+30, 227+39 };
5105 int coeff_count = 0;
5108 int abslevelgt1 = 0;
5110 /* cat: 0-> DC 16x16 n = 0
5111 * 1-> AC 16x16 n = luma4x4idx
5112 * 2-> Luma4x4 n = luma4x4idx
5113 * 3-> DC Chroma n = iCbCr
5114 * 4-> AC Chroma n = 4 * iCbCr + chroma4x4idx
5117 /* read coded block flag */
5118 if( get_cabac( &h->cabac, &h->cabac_state[85 + get_cabac_cbf_ctx( h, cat, n ) ] ) == 0 ) {
5119 if( cat == 1 || cat == 2 )
5120 h->non_zero_count_cache[scan8[n]] = 0;
5122 h->non_zero_count_cache[scan8[16+n]] = 0;
5127 for(last= 0; last < max_coeff - 1; last++) {
5128 if( get_cabac( &h->cabac, &h->cabac_state[significant_coeff_flag_field_offset[h->mb_field_decoding_flag]+significant_coeff_flag_offset[cat]+last] )) {
5129 index[coeff_count++] = last;
5130 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] ) ) {
5136 if( last == max_coeff -1 ) {
5137 index[coeff_count++] = last;
5139 assert(coeff_count > 0);
5142 h->cbp_table[mb_xy] |= 0x100;
5143 else if( cat == 1 || cat == 2 )
5144 h->non_zero_count_cache[scan8[n]] = coeff_count;
5146 h->cbp_table[mb_xy] |= 0x40 << n;
5149 h->non_zero_count_cache[scan8[16+n]] = coeff_count;
5152 for( i = coeff_count - 1; i >= 0; i-- ) {
5153 int ctx = (abslevelgt1 != 0 ? 0 : FFMIN( 4, abslevel1 )) + coeff_abs_level_m1_offset[cat];
5154 int j= scantable[index[i]];
5156 if( get_cabac( &h->cabac, &h->cabac_state[ctx] ) == 0 ) {
5157 if( cat == 0 || cat == 3 ) {
5158 if( get_cabac_bypass( &h->cabac ) ) block[j] = -1;
5161 if( get_cabac_bypass( &h->cabac ) ) block[j] = -qmul[j];
5162 else block[j] = qmul[j];
5168 ctx = 5 + FFMIN( 4, abslevelgt1 ) + coeff_abs_level_m1_offset[cat];
5169 while( coeff_abs < 15 && get_cabac( &h->cabac, &h->cabac_state[ctx] ) ) {
5173 if( coeff_abs >= 15 ) {
5175 while( get_cabac_bypass( &h->cabac ) ) {
5176 coeff_abs += 1 << j;
5181 if( get_cabac_bypass( &h->cabac ) )
5182 coeff_abs += 1 << j ;
5186 if( cat == 0 || cat == 3 ) {
5187 if( get_cabac_bypass( &h->cabac ) ) block[j] = -coeff_abs;
5188 else block[j] = coeff_abs;
5190 if( get_cabac_bypass( &h->cabac ) ) block[j] = -coeff_abs * qmul[j];
5191 else block[j] = coeff_abs * qmul[j];
5200 void inline compute_mb_neighboors(H264Context *h)
5202 MpegEncContext * const s = &h->s;
5203 const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
5204 h->top_mb_xy = mb_xy - s->mb_stride;
5205 h->left_mb_xy[0] = mb_xy - 1;
5206 if(h->mb_aff_frame){
5207 const int pair_xy = s->mb_x + (s->mb_y & ~1)*s->mb_stride;
5208 const int top_pair_xy = pair_xy - s->mb_stride;
5209 const int top_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[top_pair_xy]);
5210 const int left_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[pair_xy-1]);
5211 const int curr_mb_frame_flag = !h->mb_field_decoding_flag;
5212 const int bottom = (s->mb_y & 1);
5214 ? !curr_mb_frame_flag // bottom macroblock
5215 : (!curr_mb_frame_flag && !top_mb_frame_flag) // top macroblock
5217 h->top_mb_xy -= s->mb_stride;
5219 if (left_mb_frame_flag != curr_mb_frame_flag) {
5220 h->left_mb_xy[0] = pair_xy - 1;
5227 * decodes a macroblock
5228 * @returns 0 if ok, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
5230 static int decode_mb_cabac(H264Context *h) {
5231 MpegEncContext * const s = &h->s;
5232 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
5233 int mb_type, partition_count, cbp = 0;
5235 s->dsp.clear_blocks(h->mb); //FIXME avoid if already clear (move after skip handlong?)
5237 tprintf("pic:%d mb:%d/%d\n", h->frame_num, s->mb_x, s->mb_y);
5238 if( h->slice_type != I_TYPE && h->slice_type != SI_TYPE ) {
5239 /* read skip flags */
5240 if( decode_cabac_mb_skip( h ) ) {
5243 h->cbp_table[mb_xy] = 0;
5244 h->chroma_pred_mode_table[mb_xy] = 0;
5245 h->last_qscale_diff = 0;
5251 if(h->mb_aff_frame){
5252 if ( ((s->mb_y&1) == 0) || h->prev_mb_skipped)
5253 h->mb_field_decoding_flag = decode_cabac_field_decoding_flag(h);
5255 h->mb_field_decoding_flag= (s->picture_structure!=PICT_FRAME);
5257 h->prev_mb_skipped = 0;
5259 compute_mb_neighboors(h);
5260 if( ( mb_type = decode_cabac_mb_type( h ) ) < 0 ) {
5261 av_log( h->s.avctx, AV_LOG_ERROR, "decode_cabac_mb_type failed\n" );
5265 if( h->slice_type == B_TYPE ) {
5267 partition_count= b_mb_type_info[mb_type].partition_count;
5268 mb_type= b_mb_type_info[mb_type].type;
5271 goto decode_intra_mb;
5273 } else if( h->slice_type == P_TYPE ) {
5275 partition_count= p_mb_type_info[mb_type].partition_count;
5276 mb_type= p_mb_type_info[mb_type].type;
5279 goto decode_intra_mb;
5282 assert(h->slice_type == I_TYPE);
5284 partition_count = 0;
5285 cbp= i_mb_type_info[mb_type].cbp;
5286 h->intra16x16_pred_mode= i_mb_type_info[mb_type].pred_mode;
5287 mb_type= i_mb_type_info[mb_type].type;
5289 if(h->mb_field_decoding_flag)
5290 mb_type |= MB_TYPE_INTERLACED;
5292 s->current_picture.mb_type[mb_xy]= mb_type;
5293 h->slice_table[ mb_xy ]= h->slice_num;
5295 if(IS_INTRA_PCM(mb_type)) {
5299 // We assume these blocks are very rare so we dont optimize it.
5300 // FIXME The two following lines get the bitstream position in the cabac
5301 // decode, I think it should be done by a function in cabac.h (or cabac.c).
5302 ptr= h->cabac.bytestream;
5303 if (h->cabac.low&0x1) ptr-=CABAC_BITS/8;
5305 // The pixels are stored in the same order as levels in h->mb array.
5306 for(y=0; y<16; y++){
5307 const int index= 4*(y&3) + 32*((y>>2)&1) + 128*(y>>3);
5308 for(x=0; x<16; x++){
5309 tprintf("LUMA ICPM LEVEL (%3d)\n", *ptr);
5310 h->mb[index + (x&3) + 16*((x>>2)&1) + 64*(x>>3)]= *ptr++;
5314 const int index= 256 + 4*(y&3) + 32*(y>>2);
5316 tprintf("CHROMA U ICPM LEVEL (%3d)\n", *ptr);
5317 h->mb[index + (x&3) + 16*(x>>2)]= *ptr++;
5321 const int index= 256 + 64 + 4*(y&3) + 32*(y>>2);
5323 tprintf("CHROMA V ICPM LEVEL (%3d)\n", *ptr);
5324 h->mb[index + (x&3) + 16*(x>>2)]= *ptr++;
5328 ff_init_cabac_decoder(&h->cabac, ptr, h->cabac.bytestream_end - ptr);
5330 // All blocks are present
5331 h->cbp_table[mb_xy] = 0x1ef;
5332 h->chroma_pred_mode_table[mb_xy] = 0;
5333 // In deblocking, the quantizer is 0
5334 s->current_picture.qscale_table[mb_xy]= 0;
5335 h->chroma_qp = get_chroma_qp(h->pps.chroma_qp_index_offset, 0);
5336 // All coeffs are present
5337 memset(h->non_zero_count[mb_xy], 16, 16);
5341 fill_caches(h, mb_type, 0);
5343 if( IS_INTRA( mb_type ) ) {
5344 if( IS_INTRA4x4( mb_type ) ) {
5346 for( i = 0; i < 16; i++ ) {
5347 int pred = pred_intra_mode( h, i );
5348 h->intra4x4_pred_mode_cache[ scan8[i] ] = decode_cabac_mb_intra4x4_pred_mode( h, pred );
5350 //av_log( s->avctx, AV_LOG_ERROR, "i4x4 pred=%d mode=%d\n", pred, h->intra4x4_pred_mode_cache[ scan8[i] ] );
5352 write_back_intra_pred_mode(h);
5353 if( check_intra4x4_pred_mode(h) < 0 ) return -1;
5355 h->intra16x16_pred_mode= check_intra_pred_mode( h, h->intra16x16_pred_mode );
5356 if( h->intra16x16_pred_mode < 0 ) return -1;
5358 h->chroma_pred_mode_table[mb_xy] =
5359 h->chroma_pred_mode = decode_cabac_mb_chroma_pre_mode( h );
5361 h->chroma_pred_mode= check_intra_pred_mode( h, h->chroma_pred_mode );
5362 if( h->chroma_pred_mode < 0 ) return -1;
5363 } else if( partition_count == 4 ) {
5364 int i, j, sub_partition_count[4], list, ref[2][4];
5366 if( h->slice_type == B_TYPE ) {
5367 for( i = 0; i < 4; i++ ) {
5368 h->sub_mb_type[i] = decode_cabac_b_mb_sub_type( h );
5369 sub_partition_count[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
5370 h->sub_mb_type[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].type;
5372 if( IS_DIRECT(h->sub_mb_type[0]) || IS_DIRECT(h->sub_mb_type[1])
5373 || IS_DIRECT(h->sub_mb_type[2]) || IS_DIRECT(h->sub_mb_type[3])) {
5374 pred_direct_motion(h, &mb_type);
5375 if( h->ref_count[0] > 1 || h->ref_count[1] > 1 ) {
5376 for( i = 0; i < 4; i++ )
5377 if( IS_DIRECT(h->sub_mb_type[i]) )
5378 fill_rectangle( &h->direct_cache[scan8[4*i]], 2, 2, 8, 1, 1 );
5382 for( i = 0; i < 4; i++ ) {
5383 h->sub_mb_type[i] = decode_cabac_p_mb_sub_type( h );
5384 sub_partition_count[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
5385 h->sub_mb_type[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].type;
5389 for( list = 0; list < 2; list++ ) {
5390 if( h->ref_count[list] > 0 ) {
5391 for( i = 0; i < 4; i++ ) {
5392 if(IS_DIRECT(h->sub_mb_type[i])) continue;
5393 if(IS_DIR(h->sub_mb_type[i], 0, list)){
5394 if( h->ref_count[list] > 1 )
5395 ref[list][i] = decode_cabac_mb_ref( h, list, 4*i );
5401 h->ref_cache[list][ scan8[4*i]+1 ]=
5402 h->ref_cache[list][ scan8[4*i]+8 ]=h->ref_cache[list][ scan8[4*i]+9 ]= ref[list][i];
5407 for(list=0; list<2; list++){
5409 if(IS_DIRECT(h->sub_mb_type[i])){
5410 fill_rectangle(h->mvd_cache[list][scan8[4*i]], 2, 2, 8, 0, 4);
5413 h->ref_cache[list][ scan8[4*i] ]=h->ref_cache[list][ scan8[4*i]+1 ];
5415 if(IS_DIR(h->sub_mb_type[i], 0, list) && !IS_DIRECT(h->sub_mb_type[i])){
5416 const int sub_mb_type= h->sub_mb_type[i];
5417 const int block_width= (sub_mb_type & (MB_TYPE_16x16|MB_TYPE_16x8)) ? 2 : 1;
5418 for(j=0; j<sub_partition_count[i]; j++){
5421 const int index= 4*i + block_width*j;
5422 int16_t (* mv_cache)[2]= &h->mv_cache[list][ scan8[index] ];
5423 int16_t (* mvd_cache)[2]= &h->mvd_cache[list][ scan8[index] ];
5424 pred_motion(h, index, block_width, list, h->ref_cache[list][ scan8[index] ], &mpx, &mpy);
5426 mx = mpx + decode_cabac_mb_mvd( h, list, index, 0 );
5427 my = mpy + decode_cabac_mb_mvd( h, list, index, 1 );
5428 tprintf("final mv:%d %d\n", mx, my);
5430 if(IS_SUB_8X8(sub_mb_type)){
5431 mv_cache[ 0 ][0]= mv_cache[ 1 ][0]=
5432 mv_cache[ 8 ][0]= mv_cache[ 9 ][0]= mx;
5433 mv_cache[ 0 ][1]= mv_cache[ 1 ][1]=
5434 mv_cache[ 8 ][1]= mv_cache[ 9 ][1]= my;
5436 mvd_cache[ 0 ][0]= mvd_cache[ 1 ][0]=
5437 mvd_cache[ 8 ][0]= mvd_cache[ 9 ][0]= mx - mpx;
5438 mvd_cache[ 0 ][1]= mvd_cache[ 1 ][1]=
5439 mvd_cache[ 8 ][1]= mvd_cache[ 9 ][1]= my - mpy;
5440 }else if(IS_SUB_8X4(sub_mb_type)){
5441 mv_cache[ 0 ][0]= mv_cache[ 1 ][0]= mx;
5442 mv_cache[ 0 ][1]= mv_cache[ 1 ][1]= my;
5444 mvd_cache[ 0 ][0]= mvd_cache[ 1 ][0]= mx- mpx;
5445 mvd_cache[ 0 ][1]= mvd_cache[ 1 ][1]= my - mpy;
5446 }else if(IS_SUB_4X8(sub_mb_type)){
5447 mv_cache[ 0 ][0]= mv_cache[ 8 ][0]= mx;
5448 mv_cache[ 0 ][1]= mv_cache[ 8 ][1]= my;
5450 mvd_cache[ 0 ][0]= mvd_cache[ 8 ][0]= mx - mpx;
5451 mvd_cache[ 0 ][1]= mvd_cache[ 8 ][1]= my - mpy;
5453 assert(IS_SUB_4X4(sub_mb_type));
5454 mv_cache[ 0 ][0]= mx;
5455 mv_cache[ 0 ][1]= my;
5457 mvd_cache[ 0 ][0]= mx - mpx;
5458 mvd_cache[ 0 ][1]= my - mpy;
5462 uint32_t *p= (uint32_t *)&h->mv_cache[list][ scan8[4*i] ][0];
5463 uint32_t *pd= (uint32_t *)&h->mvd_cache[list][ scan8[4*i] ][0];
5464 p[0] = p[1] = p[8] = p[9] = 0;
5465 pd[0]= pd[1]= pd[8]= pd[9]= 0;
5469 } else if( IS_DIRECT(mb_type) ) {
5470 pred_direct_motion(h, &mb_type);
5471 s->current_picture.mb_type[mb_xy]= mb_type;
5472 fill_rectangle(h->mvd_cache[0][scan8[0]], 4, 4, 8, 0, 4);
5473 fill_rectangle(h->mvd_cache[1][scan8[0]], 4, 4, 8, 0, 4);
5475 int list, mx, my, i, mpx, mpy;
5476 if(IS_16X16(mb_type)){
5477 for(list=0; list<2; list++){
5478 if(IS_DIR(mb_type, 0, list)){
5479 if(h->ref_count[list] > 0 ){
5480 const int ref = h->ref_count[list] > 1 ? decode_cabac_mb_ref( h, list, 0 ) : 0;
5481 fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, ref, 1);
5484 fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, (uint8_t)LIST_NOT_USED, 1);
5486 for(list=0; list<2; list++){
5487 if(IS_DIR(mb_type, 0, list)){
5488 pred_motion(h, 0, 4, list, h->ref_cache[list][ scan8[0] ], &mpx, &mpy);
5490 mx = mpx + decode_cabac_mb_mvd( h, list, 0, 0 );
5491 my = mpy + decode_cabac_mb_mvd( h, list, 0, 1 );
5492 tprintf("final mv:%d %d\n", mx, my);
5494 fill_rectangle(h->mvd_cache[list][ scan8[0] ], 4, 4, 8, pack16to32(mx-mpx,my-mpy), 4);
5495 fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, pack16to32(mx,my), 4);
5497 fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, 0, 4);
5500 else if(IS_16X8(mb_type)){
5501 for(list=0; list<2; list++){
5502 if(h->ref_count[list]>0){
5504 if(IS_DIR(mb_type, i, list)){
5505 const int ref= h->ref_count[list] > 1 ? decode_cabac_mb_ref( h, list, 8*i ) : 0;
5506 fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, ref, 1);
5508 fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, (LIST_NOT_USED&0xFF), 1);
5512 for(list=0; list<2; list++){
5514 if(IS_DIR(mb_type, i, list)){
5515 pred_16x8_motion(h, 8*i, list, h->ref_cache[list][scan8[0] + 16*i], &mpx, &mpy);
5516 mx = mpx + decode_cabac_mb_mvd( h, list, 8*i, 0 );
5517 my = mpy + decode_cabac_mb_mvd( h, list, 8*i, 1 );
5518 tprintf("final mv:%d %d\n", mx, my);
5520 fill_rectangle(h->mvd_cache[list][ scan8[0] + 16*i ], 4, 2, 8, pack16to32(mx-mpx,my-mpy), 4);
5521 fill_rectangle(h->mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, pack16to32(mx,my), 4);
5523 fill_rectangle(h->mvd_cache[list][ scan8[0] + 16*i ], 4, 2, 8, 0, 4);
5524 fill_rectangle(h-> mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, 0, 4);
5529 assert(IS_8X16(mb_type));
5530 for(list=0; list<2; list++){
5531 if(h->ref_count[list]>0){
5533 if(IS_DIR(mb_type, i, list)){ //FIXME optimize
5534 const int ref= h->ref_count[list] > 1 ? decode_cabac_mb_ref( h, list, 4*i ) : 0;
5535 fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, ref, 1);
5537 fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, (LIST_NOT_USED&0xFF), 1);
5541 for(list=0; list<2; list++){
5543 if(IS_DIR(mb_type, i, list)){
5544 pred_8x16_motion(h, i*4, list, h->ref_cache[list][ scan8[0] + 2*i ], &mpx, &mpy);
5545 mx = mpx + decode_cabac_mb_mvd( h, list, 4*i, 0 );
5546 my = mpy + decode_cabac_mb_mvd( h, list, 4*i, 1 );
5548 tprintf("final mv:%d %d\n", mx, my);
5549 fill_rectangle(h->mvd_cache[list][ scan8[0] + 2*i ], 2, 4, 8, pack16to32(mx-mpx,my-mpy), 4);
5550 fill_rectangle(h->mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, pack16to32(mx,my), 4);
5552 fill_rectangle(h->mvd_cache[list][ scan8[0] + 2*i ], 2, 4, 8, 0, 4);
5553 fill_rectangle(h-> mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, 0, 4);
5560 if( IS_INTER( mb_type ) ) {
5561 h->chroma_pred_mode_table[mb_xy] = 0;
5562 write_back_motion( h, mb_type );
5565 if( !IS_INTRA16x16( mb_type ) ) {
5566 cbp = decode_cabac_mb_cbp_luma( h );
5567 cbp |= decode_cabac_mb_cbp_chroma( h ) << 4;
5570 h->cbp_table[mb_xy] = cbp;
5572 if( cbp || IS_INTRA16x16( mb_type ) ) {
5573 const uint8_t *scan, *dc_scan;
5576 if(IS_INTERLACED(mb_type)){
5578 dc_scan= luma_dc_field_scan;
5581 dc_scan= luma_dc_zigzag_scan;
5584 h->last_qscale_diff = dqp = decode_cabac_mb_dqp( h );
5586 if(((unsigned)s->qscale) > 51){
5587 if(s->qscale<0) s->qscale+= 52;
5588 else s->qscale-= 52;
5590 h->chroma_qp = get_chroma_qp(h->pps.chroma_qp_index_offset, s->qscale);
5592 if( IS_INTRA16x16( mb_type ) ) {
5594 //av_log( s->avctx, AV_LOG_ERROR, "INTRA16x16 DC\n" );
5595 if( decode_cabac_residual( h, h->mb, 0, 0, dc_scan, s->qscale, 16) < 0)
5598 for( i = 0; i < 16; i++ ) {
5599 //av_log( s->avctx, AV_LOG_ERROR, "INTRA16x16 AC:%d\n", i );
5600 if( decode_cabac_residual(h, h->mb + 16*i, 1, i, scan + 1, s->qscale, 15) < 0 )
5604 fill_rectangle(&h->non_zero_count_cache[scan8[0]], 4, 4, 8, 0, 1);
5608 for( i8x8 = 0; i8x8 < 4; i8x8++ ) {
5609 if( cbp & (1<<i8x8) ) {
5610 for( i4x4 = 0; i4x4 < 4; i4x4++ ) {
5611 const int index = 4*i8x8 + i4x4;
5612 //av_log( s->avctx, AV_LOG_ERROR, "Luma4x4: %d\n", index );
5613 if( decode_cabac_residual(h, h->mb + 16*index, 2, index, scan, s->qscale, 16) < 0 )
5617 uint8_t * const nnz= &h->non_zero_count_cache[ scan8[4*i8x8] ];
5618 nnz[0] = nnz[1] = nnz[8] = nnz[9] = 0;
5625 for( c = 0; c < 2; c++ ) {
5626 //av_log( s->avctx, AV_LOG_ERROR, "INTRA C%d-DC\n",c );
5627 if( decode_cabac_residual(h, h->mb + 256 + 16*4*c, 3, c, chroma_dc_scan, h->chroma_qp, 4) < 0)
5634 for( c = 0; c < 2; c++ ) {
5635 for( i = 0; i < 4; i++ ) {
5636 const int index = 16 + 4 * c + i;
5637 //av_log( s->avctx, AV_LOG_ERROR, "INTRA C%d-AC %d\n",c, index - 16 );
5638 if( decode_cabac_residual(h, h->mb + 16*index, 4, index - 16, scan + 1, h->chroma_qp, 15) < 0)
5643 uint8_t * const nnz= &h->non_zero_count_cache[0];
5644 nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =
5645 nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
5648 uint8_t * const nnz= &h->non_zero_count_cache[0];
5649 fill_rectangle(&nnz[scan8[0]], 4, 4, 8, 0, 1);
5650 nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =
5651 nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
5654 s->current_picture.qscale_table[mb_xy]= s->qscale;
5655 write_back_non_zero_count(h);
5661 static void filter_mb_edgev( H264Context *h, uint8_t *pix, int stride, int bS[4], int qp ) {
5663 const int index_a = clip( qp + h->slice_alpha_c0_offset, 0, 51 );
5664 const int alpha = alpha_table[index_a];
5665 const int beta = beta_table[clip( qp + h->slice_beta_offset, 0, 51 )];
5670 tc[i] = bS[i] ? tc0_table[index_a][bS[i] - 1] : -1;
5671 h->s.dsp.h264_h_loop_filter_luma(pix, stride, alpha, beta, tc);
5673 /* 16px edge length, because bS=4 is triggered by being at
5674 * the edge of an intra MB, so all 4 bS are the same */
5675 for( d = 0; d < 16; d++ ) {
5676 const int p0 = pix[-1];
5677 const int p1 = pix[-2];
5678 const int p2 = pix[-3];
5680 const int q0 = pix[0];
5681 const int q1 = pix[1];
5682 const int q2 = pix[2];
5684 if( ABS( p0 - q0 ) < alpha &&
5685 ABS( p1 - p0 ) < beta &&
5686 ABS( q1 - q0 ) < beta ) {
5688 if(ABS( p0 - q0 ) < (( alpha >> 2 ) + 2 )){
5689 if( ABS( p2 - p0 ) < beta)
5691 const int p3 = pix[-4];
5693 pix[-1] = ( p2 + 2*p1 + 2*p0 + 2*q0 + q1 + 4 ) >> 3;
5694 pix[-2] = ( p2 + p1 + p0 + q0 + 2 ) >> 2;
5695 pix[-3] = ( 2*p3 + 3*p2 + p1 + p0 + q0 + 4 ) >> 3;
5698 pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
5700 if( ABS( q2 - q0 ) < beta)
5702 const int q3 = pix[3];
5704 pix[0] = ( p1 + 2*p0 + 2*q0 + 2*q1 + q2 + 4 ) >> 3;
5705 pix[1] = ( p0 + q0 + q1 + q2 + 2 ) >> 2;
5706 pix[2] = ( 2*q3 + 3*q2 + q1 + q0 + p0 + 4 ) >> 3;
5709 pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
5713 pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
5714 pix[ 0] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
5716 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]);
5722 static void filter_mb_edgecv( H264Context *h, uint8_t *pix, int stride, int bS[4], int qp ) {
5724 const int index_a = clip( qp + h->slice_alpha_c0_offset, 0, 51 );
5725 const int alpha = alpha_table[index_a];
5726 const int beta = beta_table[clip( qp + h->slice_beta_offset, 0, 51 )];
5731 tc[i] = bS[i] ? tc0_table[index_a][bS[i] - 1] + 1 : 0;
5732 h->s.dsp.h264_h_loop_filter_chroma(pix, stride, alpha, beta, tc);
5734 h->s.dsp.h264_h_loop_filter_chroma_intra(pix, stride, alpha, beta);
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 )];
5961 tc[i] = bS[i] ? tc0_table[index_a][bS[i] - 1] + 1 : 0;
5962 h->s.dsp.h264_v_loop_filter_chroma(pix, stride, alpha, beta, tc);
5964 h->s.dsp.h264_v_loop_filter_chroma_intra(pix, stride, alpha, beta);
5968 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) {
5969 MpegEncContext * const s = &h->s;
5970 const int mb_xy= mb_x + mb_y*s->mb_stride;
5971 int first_vertical_edge_done = 0;
5973 /* FIXME: A given frame may occupy more than one position in
5974 * the reference list. So ref2frm should be populated with
5975 * frame numbers, not indices. */
5976 static const int ref2frm[18] = {-1,-1,0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15};
5979 // left mb is in picture
5980 && h->slice_table[mb_xy-1] != 255
5981 // and current and left pair do not have the same interlaced type
5982 && (IS_INTERLACED(s->current_picture.mb_type[mb_xy]) != IS_INTERLACED(s->current_picture.mb_type[mb_xy-1]))
5983 // and left mb is in the same slice if deblocking_filter == 2
5984 && (h->deblocking_filter!=2 || h->slice_table[mb_xy-1] == h->slice_table[mb_xy])) {
5985 /* First vertical edge is different in MBAFF frames
5986 * There are 8 different bS to compute and 2 different Qp
5993 first_vertical_edge_done = 1;
5994 for( i = 0; i < 8; i++ ) {
5996 int b_idx= 8 + 4 + 8*y;
5997 int bn_idx= b_idx - 1;
5999 int mbn_xy = h->mb_field_decoding_flag ? h->left_mb_xy[i>>2] : h->left_mb_xy[i&1];
6001 if( IS_INTRA( s->current_picture.mb_type[mb_xy] ) ||
6002 IS_INTRA( s->current_picture.mb_type[mbn_xy] ) ) {
6004 } else if( h->non_zero_count_cache[b_idx] != 0 ||
6005 h->non_zero_count_cache[bn_idx] != 0 ) {
6010 for( l = 0; l < 1 + (h->slice_type == B_TYPE); l++ ) {
6011 if( ref2frm[h->ref_cache[l][b_idx]+2] != ref2frm[h->ref_cache[l][bn_idx]+2] ||
6012 ABS( h->mv_cache[l][b_idx][0] - h->mv_cache[l][bn_idx][0] ) >= 4 ||
6013 ABS( h->mv_cache[l][b_idx][1] - h->mv_cache[l][bn_idx][1] ) >= 4 ) {
6020 if(bS[0]+bS[1]+bS[2]+bS[3] != 0) {
6021 // Do not use s->qscale as luma quantizer because it has not the same
6022 // value in IPCM macroblocks.
6023 qp[0] = ( s->current_picture.qscale_table[mb_xy] + s->current_picture.qscale_table[h->left_mb_xy[0]] + 1 ) >> 1;
6024 chroma_qp[0] = ( get_chroma_qp( h->pps.chroma_qp_index_offset, s->current_picture.qscale_table[mb_xy] ) +
6025 get_chroma_qp( h->pps.chroma_qp_index_offset, s->current_picture.qscale_table[h->left_mb_xy[0]] ) + 1 ) >> 1;
6026 qp[1] = ( s->current_picture.qscale_table[mb_xy] + s->current_picture.qscale_table[h->left_mb_xy[1]] + 1 ) >> 1;
6027 chroma_qp[1] = ( get_chroma_qp( h->pps.chroma_qp_index_offset, s->current_picture.qscale_table[mb_xy] ) +
6028 get_chroma_qp( h->pps.chroma_qp_index_offset, s->current_picture.qscale_table[h->left_mb_xy[1]] ) + 1 ) >> 1;
6031 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);
6032 { int i; for (i = 0; i < 8; i++) tprintf(" bS[%d]:%d", i, bS[i]); tprintf("\n"); }
6033 filter_mb_mbaff_edgev ( h, &img_y [0], linesize, bS, qp );
6034 filter_mb_mbaff_edgecv( h, &img_cb[0], uvlinesize, bS, chroma_qp );
6035 filter_mb_mbaff_edgecv( h, &img_cr[0], uvlinesize, bS, chroma_qp );
6038 /* dir : 0 -> vertical edge, 1 -> horizontal edge */
6039 for( dir = 0; dir < 2; dir++ )
6042 const int mbm_xy = dir == 0 ? mb_xy -1 : h->top_mb_xy;
6043 int start = h->slice_table[mbm_xy] == 255 ? 1 : 0;
6045 if (first_vertical_edge_done) {
6047 first_vertical_edge_done = 0;
6050 if (h->deblocking_filter==2 && h->slice_table[mbm_xy] != h->slice_table[mb_xy])
6054 for( edge = start; edge < 4; edge++ ) {
6055 /* mbn_xy: neighbor macroblock */
6056 int mbn_xy = edge > 0 ? mb_xy : mbm_xy;
6060 if (h->mb_aff_frame && (dir == 1) && (edge == 0) && ((mb_y & 1) == 0)
6061 && !IS_INTERLACED(s->current_picture.mb_type[mb_xy])
6062 && IS_INTERLACED(s->current_picture.mb_type[mbn_xy])
6064 // This is a special case in the norm where the filtering must
6065 // be done twice (one each of the field) even if we are in a
6066 // frame macroblock.
6068 unsigned int tmp_linesize = 2 * linesize;
6069 unsigned int tmp_uvlinesize = 2 * uvlinesize;
6070 int mbn_xy = mb_xy - 2 * s->mb_stride;
6074 if( IS_INTRA( s->current_picture.mb_type[mb_xy] ) ||
6075 IS_INTRA( s->current_picture.mb_type[mbn_xy] ) ) {
6076 bS[0] = bS[1] = bS[2] = bS[3] = 3;
6082 // Do not use s->qscale as luma quantizer because it has not the same
6083 // value in IPCM macroblocks.
6084 qp = ( s->current_picture.qscale_table[mb_xy] + s->current_picture.qscale_table[mbn_xy] + 1 ) >> 1;
6085 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);
6086 { int i; for (i = 0; i < 4; i++) tprintf(" bS[%d]:%d", i, bS[i]); tprintf("\n"); }
6087 filter_mb_edgeh( h, &img_y[0], tmp_linesize, bS, qp );
6088 chroma_qp = ( h->chroma_qp +
6089 get_chroma_qp( h->pps.chroma_qp_index_offset, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1;
6090 filter_mb_edgech( h, &img_cb[0], tmp_uvlinesize, bS, chroma_qp );
6091 filter_mb_edgech( h, &img_cr[0], tmp_uvlinesize, bS, chroma_qp );
6094 mbn_xy += s->mb_stride;
6095 if( IS_INTRA( s->current_picture.mb_type[mb_xy] ) ||
6096 IS_INTRA( s->current_picture.mb_type[mbn_xy] ) ) {
6097 bS[0] = bS[1] = bS[2] = bS[3] = 3;
6103 // Do not use s->qscale as luma quantizer because it has not the same
6104 // value in IPCM macroblocks.
6105 qp = ( s->current_picture.qscale_table[mb_xy] + s->current_picture.qscale_table[mbn_xy] + 1 ) >> 1;
6106 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);
6107 { int i; for (i = 0; i < 4; i++) tprintf(" bS[%d]:%d", i, bS[i]); tprintf("\n"); }
6108 filter_mb_edgeh( h, &img_y[linesize], tmp_linesize, bS, qp );
6109 chroma_qp = ( h->chroma_qp +
6110 get_chroma_qp( h->pps.chroma_qp_index_offset, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1;
6111 filter_mb_edgech( h, &img_cb[uvlinesize], tmp_uvlinesize, bS, chroma_qp );
6112 filter_mb_edgech( h, &img_cr[uvlinesize], tmp_uvlinesize, bS, chroma_qp );
6115 if( IS_INTRA( s->current_picture.mb_type[mb_xy] ) ||
6116 IS_INTRA( s->current_picture.mb_type[mbn_xy] ) ) {
6119 if ( (!IS_INTERLACED(s->current_picture.mb_type[mb_xy]) && !IS_INTERLACED(s->current_picture.mb_type[mbm_xy]))
6120 || ((h->mb_aff_frame || (s->picture_structure != PICT_FRAME)) && (dir == 0))
6129 bS[0] = bS[1] = bS[2] = bS[3] = value;
6132 for( i = 0; i < 4; i++ ) {
6133 int x = dir == 0 ? edge : i;
6134 int y = dir == 0 ? i : edge;
6135 int b_idx= 8 + 4 + x + 8*y;
6136 int bn_idx= b_idx - (dir ? 8:1);
6138 if( h->non_zero_count_cache[b_idx] != 0 ||
6139 h->non_zero_count_cache[bn_idx] != 0 ) {
6146 for( l = 0; l < 1 + (h->slice_type == B_TYPE); l++ ) {
6147 if( ref2frm[h->ref_cache[l][b_idx]+2] != ref2frm[h->ref_cache[l][bn_idx]+2] ||
6148 ABS( h->mv_cache[l][b_idx][0] - h->mv_cache[l][bn_idx][0] ) >= 4 ||
6149 ABS( h->mv_cache[l][b_idx][1] - h->mv_cache[l][bn_idx][1] ) >= 4 ) {
6157 if(bS[0]+bS[1]+bS[2]+bS[3] == 0)
6162 // Do not use s->qscale as luma quantizer because it has not the same
6163 // value in IPCM macroblocks.
6164 qp = ( s->current_picture.qscale_table[mb_xy] + s->current_picture.qscale_table[mbn_xy] + 1 ) >> 1;
6165 //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]);
6166 tprintf("filter mb:%d/%d dir:%d edge:%d, QPy:%d ls:%d uvls:%d", mb_x, mb_y, dir, edge, qp, linesize, uvlinesize);
6167 { int i; for (i = 0; i < 4; i++) tprintf(" bS[%d]:%d", i, bS[i]); tprintf("\n"); }
6169 filter_mb_edgev( h, &img_y[4*edge], linesize, bS, qp );
6170 if( (edge&1) == 0 ) {
6171 int chroma_qp = ( h->chroma_qp +
6172 get_chroma_qp( h->pps.chroma_qp_index_offset, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1;
6173 filter_mb_edgecv( h, &img_cb[2*edge], uvlinesize, bS, chroma_qp );
6174 filter_mb_edgecv( h, &img_cr[2*edge], uvlinesize, bS, chroma_qp );
6177 filter_mb_edgeh( h, &img_y[4*edge*linesize], linesize, bS, qp );
6178 if( (edge&1) == 0 ) {
6179 int chroma_qp = ( h->chroma_qp +
6180 get_chroma_qp( h->pps.chroma_qp_index_offset, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1;
6181 filter_mb_edgech( h, &img_cb[2*edge*uvlinesize], uvlinesize, bS, chroma_qp );
6182 filter_mb_edgech( h, &img_cr[2*edge*uvlinesize], uvlinesize, bS, chroma_qp );
6189 static int decode_slice(H264Context *h){
6190 MpegEncContext * const s = &h->s;
6191 const int part_mask= s->partitioned_frame ? (AC_END|AC_ERROR) : 0x7F;
6195 if( h->pps.cabac ) {
6199 align_get_bits( &s->gb );
6202 ff_init_cabac_states( &h->cabac, ff_h264_lps_range, ff_h264_mps_state, ff_h264_lps_state, 64 );
6203 ff_init_cabac_decoder( &h->cabac,
6204 s->gb.buffer + get_bits_count(&s->gb)/8,
6205 ( s->gb.size_in_bits - get_bits_count(&s->gb) + 7)/8);
6206 /* calculate pre-state */
6207 for( i= 0; i < 399; i++ ) {
6209 if( h->slice_type == I_TYPE )
6210 pre = clip( ((cabac_context_init_I[i][0] * s->qscale) >>4 ) + cabac_context_init_I[i][1], 1, 126 );
6212 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 );
6215 h->cabac_state[i] = 2 * ( 63 - pre ) + 0;
6217 h->cabac_state[i] = 2 * ( pre - 64 ) + 1;
6221 int ret = decode_mb_cabac(h);
6224 if(ret>=0) hl_decode_mb(h);
6226 /* XXX: useless as decode_mb_cabac it doesn't support that ... */
6227 if( ret >= 0 && h->mb_aff_frame ) { //FIXME optimal? or let mb_decode decode 16x32 ?
6230 if(ret>=0) ret = decode_mb_cabac(h);
6235 eos = get_cabac_terminate( &h->cabac );
6237 if( ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 1) {
6238 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
6239 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);
6243 if( ++s->mb_x >= s->mb_width ) {
6245 ff_draw_horiz_band(s, 16*s->mb_y, 16);
6247 if(h->mb_aff_frame) {
6252 if( eos || s->mb_y >= s->mb_height ) {
6253 tprintf("slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
6254 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);
6261 int ret = decode_mb_cavlc(h);
6263 if(ret>=0) hl_decode_mb(h);
6265 if(ret>=0 && h->mb_aff_frame){ //FIXME optimal? or let mb_decode decode 16x32 ?
6267 ret = decode_mb_cavlc(h);
6269 if(ret>=0) hl_decode_mb(h);
6274 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
6275 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);
6280 if(++s->mb_x >= s->mb_width){
6282 ff_draw_horiz_band(s, 16*s->mb_y, 16);
6284 if(h->mb_aff_frame) {
6287 if(s->mb_y >= s->mb_height){
6288 tprintf("slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
6290 if(get_bits_count(&s->gb) == s->gb.size_in_bits ) {
6291 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);
6295 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);
6302 if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->mb_skip_run<=0){
6303 tprintf("slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
6304 if(get_bits_count(&s->gb) == s->gb.size_in_bits ){
6305 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);
6309 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);
6318 for(;s->mb_y < s->mb_height; s->mb_y++){
6319 for(;s->mb_x < s->mb_width; s->mb_x++){
6320 int ret= decode_mb(h);
6325 fprintf(stderr, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
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);
6331 if(++s->mb_x >= s->mb_width){
6333 if(++s->mb_y >= s->mb_height){
6334 if(get_bits_count(s->gb) == s->gb.size_in_bits){
6335 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);
6339 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);
6346 if(get_bits_count(s->?gb) >= s->gb?.size_in_bits){
6347 if(get_bits_count(s->gb) == s->gb.size_in_bits){
6348 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);
6352 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);
6359 ff_draw_horiz_band(s, 16*s->mb_y, 16);
6362 return -1; //not reached
6365 static inline void decode_hrd_parameters(H264Context *h, SPS *sps){
6366 MpegEncContext * const s = &h->s;
6368 cpb_count = get_ue_golomb(&s->gb) + 1;
6369 get_bits(&s->gb, 4); /* bit_rate_scale */
6370 get_bits(&s->gb, 4); /* cpb_size_scale */
6371 for(i=0; i<cpb_count; i++){
6372 get_ue_golomb(&s->gb); /* bit_rate_value_minus1 */
6373 get_ue_golomb(&s->gb); /* cpb_size_value_minus1 */
6374 get_bits1(&s->gb); /* cbr_flag */
6376 get_bits(&s->gb, 5); /* initial_cpb_removal_delay_length_minus1 */
6377 get_bits(&s->gb, 5); /* cpb_removal_delay_length_minus1 */
6378 get_bits(&s->gb, 5); /* dpb_output_delay_length_minus1 */
6379 get_bits(&s->gb, 5); /* time_offset_length */
6382 static inline int decode_vui_parameters(H264Context *h, SPS *sps){
6383 MpegEncContext * const s = &h->s;
6384 int aspect_ratio_info_present_flag, aspect_ratio_idc;
6385 int nal_hrd_parameters_present_flag, vcl_hrd_parameters_present_flag;
6387 aspect_ratio_info_present_flag= get_bits1(&s->gb);
6389 if( aspect_ratio_info_present_flag ) {
6390 aspect_ratio_idc= get_bits(&s->gb, 8);
6391 if( aspect_ratio_idc == EXTENDED_SAR ) {
6392 sps->sar.num= get_bits(&s->gb, 16);
6393 sps->sar.den= get_bits(&s->gb, 16);
6394 }else if(aspect_ratio_idc < 16){
6395 sps->sar= pixel_aspect[aspect_ratio_idc];
6397 av_log(h->s.avctx, AV_LOG_ERROR, "illegal aspect ratio\n");
6404 // s->avctx->aspect_ratio= sar_width*s->width / (float)(s->height*sar_height);
6406 if(get_bits1(&s->gb)){ /* overscan_info_present_flag */
6407 get_bits1(&s->gb); /* overscan_appropriate_flag */
6410 if(get_bits1(&s->gb)){ /* video_signal_type_present_flag */
6411 get_bits(&s->gb, 3); /* video_format */
6412 get_bits1(&s->gb); /* video_full_range_flag */
6413 if(get_bits1(&s->gb)){ /* colour_description_present_flag */
6414 get_bits(&s->gb, 8); /* colour_primaries */
6415 get_bits(&s->gb, 8); /* transfer_characteristics */
6416 get_bits(&s->gb, 8); /* matrix_coefficients */
6420 if(get_bits1(&s->gb)){ /* chroma_location_info_present_flag */
6421 get_ue_golomb(&s->gb); /* chroma_sample_location_type_top_field */
6422 get_ue_golomb(&s->gb); /* chroma_sample_location_type_bottom_field */
6425 sps->timing_info_present_flag = get_bits1(&s->gb);
6426 if(sps->timing_info_present_flag){
6427 sps->num_units_in_tick = get_bits_long(&s->gb, 32);
6428 sps->time_scale = get_bits_long(&s->gb, 32);
6429 sps->fixed_frame_rate_flag = get_bits1(&s->gb);
6432 nal_hrd_parameters_present_flag = get_bits1(&s->gb);
6433 if(nal_hrd_parameters_present_flag)
6434 decode_hrd_parameters(h, sps);
6435 vcl_hrd_parameters_present_flag = get_bits1(&s->gb);
6436 if(vcl_hrd_parameters_present_flag)
6437 decode_hrd_parameters(h, sps);
6438 if(nal_hrd_parameters_present_flag || vcl_hrd_parameters_present_flag)
6439 get_bits1(&s->gb); /* low_delay_hrd_flag */
6440 get_bits1(&s->gb); /* pic_struct_present_flag */
6442 sps->bitstream_restriction_flag = get_bits1(&s->gb);
6443 if(sps->bitstream_restriction_flag){
6444 get_bits1(&s->gb); /* motion_vectors_over_pic_boundaries_flag */
6445 get_ue_golomb(&s->gb); /* max_bytes_per_pic_denom */
6446 get_ue_golomb(&s->gb); /* max_bits_per_mb_denom */
6447 get_ue_golomb(&s->gb); /* log2_max_mv_length_horizontal */
6448 get_ue_golomb(&s->gb); /* log2_max_mv_length_vertical */
6449 sps->num_reorder_frames = get_ue_golomb(&s->gb);
6450 get_ue_golomb(&s->gb); /* max_dec_frame_buffering */
6456 static inline int decode_seq_parameter_set(H264Context *h){
6457 MpegEncContext * const s = &h->s;
6458 int profile_idc, level_idc;
6462 profile_idc= get_bits(&s->gb, 8);
6463 get_bits1(&s->gb); //constraint_set0_flag
6464 get_bits1(&s->gb); //constraint_set1_flag
6465 get_bits1(&s->gb); //constraint_set2_flag
6466 get_bits1(&s->gb); //constraint_set3_flag
6467 get_bits(&s->gb, 4); // reserved
6468 level_idc= get_bits(&s->gb, 8);
6469 sps_id= get_ue_golomb(&s->gb);
6471 sps= &h->sps_buffer[ sps_id ];
6472 sps->profile_idc= profile_idc;
6473 sps->level_idc= level_idc;
6475 sps->log2_max_frame_num= get_ue_golomb(&s->gb) + 4;
6476 sps->poc_type= get_ue_golomb(&s->gb);
6478 if(sps->poc_type == 0){ //FIXME #define
6479 sps->log2_max_poc_lsb= get_ue_golomb(&s->gb) + 4;
6480 } else if(sps->poc_type == 1){//FIXME #define
6481 sps->delta_pic_order_always_zero_flag= get_bits1(&s->gb);
6482 sps->offset_for_non_ref_pic= get_se_golomb(&s->gb);
6483 sps->offset_for_top_to_bottom_field= get_se_golomb(&s->gb);
6484 sps->poc_cycle_length= get_ue_golomb(&s->gb);
6486 for(i=0; i<sps->poc_cycle_length; i++)
6487 sps->offset_for_ref_frame[i]= get_se_golomb(&s->gb);
6489 if(sps->poc_type > 2){
6490 av_log(h->s.avctx, AV_LOG_ERROR, "illegal POC type %d\n", sps->poc_type);
6494 sps->ref_frame_count= get_ue_golomb(&s->gb);
6495 if(sps->ref_frame_count > MAX_PICTURE_COUNT-2){
6496 av_log(h->s.avctx, AV_LOG_ERROR, "too many reference frames\n");
6498 sps->gaps_in_frame_num_allowed_flag= get_bits1(&s->gb);
6499 sps->mb_width= get_ue_golomb(&s->gb) + 1;
6500 sps->mb_height= get_ue_golomb(&s->gb) + 1;
6501 if((unsigned)sps->mb_width >= INT_MAX/16 || (unsigned)sps->mb_height >= INT_MAX/16 ||
6502 avcodec_check_dimensions(NULL, 16*sps->mb_width, 16*sps->mb_height))
6505 sps->frame_mbs_only_flag= get_bits1(&s->gb);
6506 if(!sps->frame_mbs_only_flag)
6507 sps->mb_aff= get_bits1(&s->gb);
6511 sps->direct_8x8_inference_flag= get_bits1(&s->gb);
6513 sps->crop= get_bits1(&s->gb);
6515 sps->crop_left = get_ue_golomb(&s->gb);
6516 sps->crop_right = get_ue_golomb(&s->gb);
6517 sps->crop_top = get_ue_golomb(&s->gb);
6518 sps->crop_bottom= get_ue_golomb(&s->gb);
6519 if(sps->crop_left || sps->crop_top){
6520 av_log(h->s.avctx, AV_LOG_ERROR, "insane cropping not completely supported, this could look slightly wrong ...\n");
6526 sps->crop_bottom= 0;
6529 sps->vui_parameters_present_flag= get_bits1(&s->gb);
6530 if( sps->vui_parameters_present_flag )
6531 decode_vui_parameters(h, sps);
6533 if(s->avctx->debug&FF_DEBUG_PICT_INFO){
6534 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",
6535 sps_id, sps->profile_idc, sps->level_idc,
6537 sps->ref_frame_count,
6538 sps->mb_width, sps->mb_height,
6539 sps->frame_mbs_only_flag ? "FRM" : (sps->mb_aff ? "MB-AFF" : "PIC-AFF"),
6540 sps->direct_8x8_inference_flag ? "8B8" : "",
6541 sps->crop_left, sps->crop_right,
6542 sps->crop_top, sps->crop_bottom,
6543 sps->vui_parameters_present_flag ? "VUI" : ""
6549 static inline int decode_picture_parameter_set(H264Context *h){
6550 MpegEncContext * const s = &h->s;
6551 int pps_id= get_ue_golomb(&s->gb);
6552 PPS *pps= &h->pps_buffer[pps_id];
6554 pps->sps_id= get_ue_golomb(&s->gb);
6555 pps->cabac= get_bits1(&s->gb);
6556 pps->pic_order_present= get_bits1(&s->gb);
6557 pps->slice_group_count= get_ue_golomb(&s->gb) + 1;
6558 if(pps->slice_group_count > 1 ){
6559 pps->mb_slice_group_map_type= get_ue_golomb(&s->gb);
6560 av_log(h->s.avctx, AV_LOG_ERROR, "FMO not supported\n");
6561 switch(pps->mb_slice_group_map_type){
6564 | for( i = 0; i <= num_slice_groups_minus1; i++ ) | | |
6565 | run_length[ i ] |1 |ue(v) |
6570 | for( i = 0; i < num_slice_groups_minus1; i++ ) | | |
6572 | top_left_mb[ i ] |1 |ue(v) |
6573 | bottom_right_mb[ i ] |1 |ue(v) |
6581 | slice_group_change_direction_flag |1 |u(1) |
6582 | slice_group_change_rate_minus1 |1 |ue(v) |
6587 | slice_group_id_cnt_minus1 |1 |ue(v) |
6588 | for( i = 0; i <= slice_group_id_cnt_minus1; i++ | | |
6590 | slice_group_id[ i ] |1 |u(v) |
6595 pps->ref_count[0]= get_ue_golomb(&s->gb) + 1;
6596 pps->ref_count[1]= get_ue_golomb(&s->gb) + 1;
6597 if(pps->ref_count[0] > 32 || pps->ref_count[1] > 32){
6598 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow (pps)\n");
6602 pps->weighted_pred= get_bits1(&s->gb);
6603 pps->weighted_bipred_idc= get_bits(&s->gb, 2);
6604 pps->init_qp= get_se_golomb(&s->gb) + 26;
6605 pps->init_qs= get_se_golomb(&s->gb) + 26;
6606 pps->chroma_qp_index_offset= get_se_golomb(&s->gb);
6607 pps->deblocking_filter_parameters_present= get_bits1(&s->gb);
6608 pps->constrained_intra_pred= get_bits1(&s->gb);
6609 pps->redundant_pic_cnt_present = get_bits1(&s->gb);
6611 if(s->avctx->debug&FF_DEBUG_PICT_INFO){
6612 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",
6613 pps_id, pps->sps_id,
6614 pps->cabac ? "CABAC" : "CAVLC",
6615 pps->slice_group_count,
6616 pps->ref_count[0], pps->ref_count[1],
6617 pps->weighted_pred ? "weighted" : "",
6618 pps->init_qp, pps->init_qs, pps->chroma_qp_index_offset,
6619 pps->deblocking_filter_parameters_present ? "LPAR" : "",
6620 pps->constrained_intra_pred ? "CONSTR" : "",
6621 pps->redundant_pic_cnt_present ? "REDU" : ""
6629 * finds the end of the current frame in the bitstream.
6630 * @return the position of the first byte of the next frame, or -1
6632 static int find_frame_end(H264Context *h, const uint8_t *buf, int buf_size){
6635 ParseContext *pc = &(h->s.parse_context);
6636 //printf("first %02X%02X%02X%02X\n", buf[0], buf[1],buf[2],buf[3]);
6637 // mb_addr= pc->mb_addr - 1;
6639 for(i=0; i<=buf_size; i++){
6640 if((state&0xFFFFFF1F) == 0x101 || (state&0xFFFFFF1F) == 0x102 || (state&0xFFFFFF1F) == 0x105){
6641 tprintf("find_frame_end new startcode = %08x, frame_start_found = %d, pos = %d\n", state, pc->frame_start_found, i);
6642 if(pc->frame_start_found){
6643 // If there isn't one more byte in the buffer
6644 // the test on first_mb_in_slice cannot be done yet
6645 // do it at next call.
6646 if (i >= buf_size) break;
6647 if (buf[i] & 0x80) {
6648 // first_mb_in_slice is 0, probably the first nal of a new
6650 tprintf("find_frame_end frame_end_found, state = %08x, pos = %d\n", state, i);
6652 pc->frame_start_found= 0;
6656 pc->frame_start_found = 1;
6659 state= (state<<8) | buf[i];
6663 return END_NOT_FOUND;
6666 static int h264_parse(AVCodecParserContext *s,
6667 AVCodecContext *avctx,
6668 uint8_t **poutbuf, int *poutbuf_size,
6669 const uint8_t *buf, int buf_size)
6671 H264Context *h = s->priv_data;
6672 ParseContext *pc = &h->s.parse_context;
6675 next= find_frame_end(h, buf, buf_size);
6677 if (ff_combine_frame(pc, next, (uint8_t **)&buf, &buf_size) < 0) {
6683 *poutbuf = (uint8_t *)buf;
6684 *poutbuf_size = buf_size;
6688 static int decode_nal_units(H264Context *h, uint8_t *buf, int buf_size){
6689 MpegEncContext * const s = &h->s;
6690 AVCodecContext * const avctx= s->avctx;
6694 for(i=0; i<32; i++){
6695 printf("%X ", buf[i]);
6707 if(buf_index >= buf_size) break;
6709 for(i = 0; i < h->nal_length_size; i++)
6710 nalsize = (nalsize << 8) | buf[buf_index++];
6712 // start code prefix search
6713 for(; buf_index + 3 < buf_size; buf_index++){
6714 // this should allways succeed in the first iteration
6715 if(buf[buf_index] == 0 && buf[buf_index+1] == 0 && buf[buf_index+2] == 1)
6719 if(buf_index+3 >= buf_size) break;
6724 ptr= decode_nal(h, buf + buf_index, &dst_length, &consumed, h->is_avc ? nalsize : buf_size - buf_index);
6725 if(ptr[dst_length - 1] == 0) dst_length--;
6726 bit_length= 8*dst_length - decode_rbsp_trailing(ptr + dst_length - 1);
6728 if(s->avctx->debug&FF_DEBUG_STARTCODE){
6729 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);
6732 if (h->is_avc && (nalsize != consumed))
6733 av_log(h->s.avctx, AV_LOG_ERROR, "AVC: Consumed only %d bytes instead of %d\n", consumed, nalsize);
6735 buf_index += consumed;
6737 if( s->hurry_up == 1 && h->nal_ref_idc == 0 )
6740 switch(h->nal_unit_type){
6742 idr(h); //FIXME ensure we don't loose some frames if there is reordering
6744 init_get_bits(&s->gb, ptr, bit_length);
6746 h->inter_gb_ptr= &s->gb;
6747 s->data_partitioning = 0;
6749 if(decode_slice_header(h) < 0) return -1;
6750 if(h->redundant_pic_count==0 && s->hurry_up < 5 )
6754 init_get_bits(&s->gb, ptr, bit_length);
6756 h->inter_gb_ptr= NULL;
6757 s->data_partitioning = 1;
6759 if(decode_slice_header(h) < 0) return -1;
6762 init_get_bits(&h->intra_gb, ptr, bit_length);
6763 h->intra_gb_ptr= &h->intra_gb;
6766 init_get_bits(&h->inter_gb, ptr, bit_length);
6767 h->inter_gb_ptr= &h->inter_gb;
6769 if(h->redundant_pic_count==0 && h->intra_gb_ptr && s->data_partitioning && s->hurry_up < 5 )
6775 init_get_bits(&s->gb, ptr, bit_length);
6776 decode_seq_parameter_set(h);
6778 if(s->flags& CODEC_FLAG_LOW_DELAY)
6781 if(avctx->has_b_frames < 2)
6782 avctx->has_b_frames= !s->low_delay;
6785 init_get_bits(&s->gb, ptr, bit_length);
6787 decode_picture_parameter_set(h);
6790 case NAL_PICTURE_DELIMITER:
6792 case NAL_FILTER_DATA:
6795 av_log(avctx, AV_LOG_ERROR, "Unknown NAL code: %d\n", h->nal_unit_type);
6799 if(!s->current_picture_ptr) return buf_index; //no frame
6801 s->current_picture_ptr->pict_type= s->pict_type;
6802 s->current_picture_ptr->key_frame= s->pict_type == I_TYPE && h->nal_unit_type == NAL_IDR_SLICE;
6804 h->prev_frame_num_offset= h->frame_num_offset;
6805 h->prev_frame_num= h->frame_num;
6806 if(s->current_picture_ptr->reference){
6807 h->prev_poc_msb= h->poc_msb;
6808 h->prev_poc_lsb= h->poc_lsb;
6810 if(s->current_picture_ptr->reference)
6811 execute_ref_pic_marking(h, h->mmco, h->mmco_index);
6821 * returns the number of bytes consumed for building the current frame
6823 static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size){
6824 if(s->flags&CODEC_FLAG_TRUNCATED){
6825 pos -= s->parse_context.last_index;
6826 if(pos<0) pos=0; // FIXME remove (unneeded?)
6830 if(pos==0) pos=1; //avoid infinite loops (i doubt thats needed but ...)
6831 if(pos+10>buf_size) pos=buf_size; // oops ;)
6837 static int decode_frame(AVCodecContext *avctx,
6838 void *data, int *data_size,
6839 uint8_t *buf, int buf_size)
6841 H264Context *h = avctx->priv_data;
6842 MpegEncContext *s = &h->s;
6843 AVFrame *pict = data;
6846 s->flags= avctx->flags;
6847 s->flags2= avctx->flags2;
6849 /* no supplementary picture */
6850 if (buf_size == 0) {
6854 if(s->flags&CODEC_FLAG_TRUNCATED){
6855 int next= find_frame_end(h, buf, buf_size);
6857 if( ff_combine_frame(&s->parse_context, next, &buf, &buf_size) < 0 )
6859 //printf("next:%d buf_size:%d last_index:%d\n", next, buf_size, s->parse_context.last_index);
6862 if(h->is_avc && !h->got_avcC) {
6863 int i, cnt, nalsize;
6864 unsigned char *p = avctx->extradata;
6865 if(avctx->extradata_size < 7) {
6866 av_log(avctx, AV_LOG_ERROR, "avcC too short\n");
6870 av_log(avctx, AV_LOG_ERROR, "Unknown avcC version %d\n", *p);
6873 /* sps and pps in the avcC always have length coded with 2 bytes,
6874 so put a fake nal_length_size = 2 while parsing them */
6875 h->nal_length_size = 2;
6876 // Decode sps from avcC
6877 cnt = *(p+5) & 0x1f; // Number of sps
6879 for (i = 0; i < cnt; i++) {
6880 nalsize = BE_16(p) + 2;
6881 if(decode_nal_units(h, p, nalsize) != nalsize) {
6882 av_log(avctx, AV_LOG_ERROR, "Decoding sps %d from avcC failed\n", i);
6887 // Decode pps from avcC
6888 cnt = *(p++); // Number of pps
6889 for (i = 0; i < cnt; i++) {
6890 nalsize = BE_16(p) + 2;
6891 if(decode_nal_units(h, p, nalsize) != nalsize) {
6892 av_log(avctx, AV_LOG_ERROR, "Decoding pps %d from avcC failed\n", i);
6897 // Now store right nal length size, that will be use to parse all other nals
6898 h->nal_length_size = ((*(((char*)(avctx->extradata))+4))&0x03)+1;
6899 // Do not reparse avcC
6903 if(!h->is_avc && s->avctx->extradata_size && s->picture_number==0){
6904 if(0 < decode_nal_units(h, s->avctx->extradata, s->avctx->extradata_size) )
6908 buf_index=decode_nal_units(h, buf, buf_size);
6912 //FIXME do something with unavailable reference frames
6914 // if(ret==FRAME_SKIPPED) return get_consumed_bytes(s, buf_index, buf_size);
6915 if(!s->current_picture_ptr){
6916 av_log(h->s.avctx, AV_LOG_DEBUG, "error, NO frame\n");
6921 Picture *out = s->current_picture_ptr;
6922 #if 0 //decode order
6923 *data_size = sizeof(AVFrame);
6925 /* Sort B-frames into display order */
6926 Picture *cur = s->current_picture_ptr;
6927 Picture *prev = h->delayed_output_pic;
6932 int dropped_frame = 0;
6935 if(h->sps.bitstream_restriction_flag
6936 && s->avctx->has_b_frames < h->sps.num_reorder_frames){
6937 s->avctx->has_b_frames = h->sps.num_reorder_frames;
6941 while(h->delayed_pic[pics]) pics++;
6942 h->delayed_pic[pics++] = cur;
6943 if(cur->reference == 0)
6946 for(i=0; h->delayed_pic[i]; i++)
6947 if(h->delayed_pic[i]->key_frame || h->delayed_pic[i]->poc==0)
6950 out = h->delayed_pic[0];
6951 for(i=1; h->delayed_pic[i] && !h->delayed_pic[i]->key_frame; i++)
6952 if(h->delayed_pic[i]->poc < out->poc){
6953 out = h->delayed_pic[i];
6957 out_of_order = !cross_idr && prev && out->poc < prev->poc;
6958 if(prev && pics <= s->avctx->has_b_frames)
6960 else if((out_of_order && pics-1 == s->avctx->has_b_frames)
6962 ((!cross_idr && prev && out->poc > prev->poc + 2)
6963 || cur->pict_type == B_TYPE)))
6966 s->avctx->has_b_frames++;
6969 else if(out_of_order)
6972 if(out_of_order || pics > s->avctx->has_b_frames){
6973 dropped_frame = (out != h->delayed_pic[out_idx]);
6974 for(i=out_idx; h->delayed_pic[i]; i++)
6975 h->delayed_pic[i] = h->delayed_pic[i+1];
6978 if(prev == out && !dropped_frame)
6981 *data_size = sizeof(AVFrame);
6982 if(prev && prev != out && prev->reference == 1)
6983 prev->reference = 0;
6984 h->delayed_output_pic = out;
6987 *pict= *(AVFrame*)out;
6990 assert(pict->data[0]);
6991 ff_print_debug_info(s, pict);
6992 //printf("out %d\n", (int)pict->data[0]);
6995 /* Return the Picture timestamp as the frame number */
6996 /* we substract 1 because it is added on utils.c */
6997 avctx->frame_number = s->picture_number - 1;
6999 return get_consumed_bytes(s, buf_index, buf_size);
7002 static inline void fill_mb_avail(H264Context *h){
7003 MpegEncContext * const s = &h->s;
7004 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
7007 h->mb_avail[0]= s->mb_x && h->slice_table[mb_xy - s->mb_stride - 1] == h->slice_num;
7008 h->mb_avail[1]= h->slice_table[mb_xy - s->mb_stride ] == h->slice_num;
7009 h->mb_avail[2]= s->mb_x+1 < s->mb_width && h->slice_table[mb_xy - s->mb_stride + 1] == h->slice_num;
7015 h->mb_avail[3]= s->mb_x && h->slice_table[mb_xy - 1] == h->slice_num;
7016 h->mb_avail[4]= 1; //FIXME move out
7017 h->mb_avail[5]= 0; //FIXME move out
7023 #define SIZE (COUNT*40)
7029 // int int_temp[10000];
7031 AVCodecContext avctx;
7033 dsputil_init(&dsp, &avctx);
7035 init_put_bits(&pb, temp, SIZE);
7036 printf("testing unsigned exp golomb\n");
7037 for(i=0; i<COUNT; i++){
7039 set_ue_golomb(&pb, i);
7040 STOP_TIMER("set_ue_golomb");
7042 flush_put_bits(&pb);
7044 init_get_bits(&gb, temp, 8*SIZE);
7045 for(i=0; i<COUNT; i++){
7048 s= show_bits(&gb, 24);
7051 j= get_ue_golomb(&gb);
7053 printf("missmatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
7056 STOP_TIMER("get_ue_golomb");
7060 init_put_bits(&pb, temp, SIZE);
7061 printf("testing signed exp golomb\n");
7062 for(i=0; i<COUNT; i++){
7064 set_se_golomb(&pb, i - COUNT/2);
7065 STOP_TIMER("set_se_golomb");
7067 flush_put_bits(&pb);
7069 init_get_bits(&gb, temp, 8*SIZE);
7070 for(i=0; i<COUNT; i++){
7073 s= show_bits(&gb, 24);
7076 j= get_se_golomb(&gb);
7077 if(j != i - COUNT/2){
7078 printf("missmatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
7081 STOP_TIMER("get_se_golomb");
7084 printf("testing 4x4 (I)DCT\n");
7087 uint8_t src[16], ref[16];
7088 uint64_t error= 0, max_error=0;
7090 for(i=0; i<COUNT; i++){
7092 // printf("%d %d %d\n", r1, r2, (r2-r1)*16);
7093 for(j=0; j<16; j++){
7094 ref[j]= random()%255;
7095 src[j]= random()%255;
7098 h264_diff_dct_c(block, src, ref, 4);
7101 for(j=0; j<16; j++){
7102 // printf("%d ", block[j]);
7103 block[j]= block[j]*4;
7104 if(j&1) block[j]= (block[j]*4 + 2)/5;
7105 if(j&4) block[j]= (block[j]*4 + 2)/5;
7109 s->dsp.h264_idct_add(ref, block, 4);
7110 /* for(j=0; j<16; j++){
7111 printf("%d ", ref[j]);
7115 for(j=0; j<16; j++){
7116 int diff= ABS(src[j] - ref[j]);
7119 max_error= FFMAX(max_error, diff);
7122 printf("error=%f max_error=%d\n", ((float)error)/COUNT/16, (int)max_error );
7124 printf("testing quantizer\n");
7125 for(qp=0; qp<52; qp++){
7127 src1_block[i]= src2_block[i]= random()%255;
7131 printf("Testing NAL layer\n");
7133 uint8_t bitstream[COUNT];
7134 uint8_t nal[COUNT*2];
7136 memset(&h, 0, sizeof(H264Context));
7138 for(i=0; i<COUNT; i++){
7146 for(j=0; j<COUNT; j++){
7147 bitstream[j]= (random() % 255) + 1;
7150 for(j=0; j<zeros; j++){
7151 int pos= random() % COUNT;
7152 while(bitstream[pos] == 0){
7161 nal_length= encode_nal(&h, nal, bitstream, COUNT, COUNT*2);
7163 printf("encoding failed\n");
7167 out= decode_nal(&h, nal, &out_length, &consumed, nal_length);
7171 if(out_length != COUNT){
7172 printf("incorrect length %d %d\n", out_length, COUNT);
7176 if(consumed != nal_length){
7177 printf("incorrect consumed length %d %d\n", nal_length, consumed);
7181 if(memcmp(bitstream, out, COUNT)){
7182 printf("missmatch\n");
7187 printf("Testing RBSP\n");
7195 static int decode_end(AVCodecContext *avctx)
7197 H264Context *h = avctx->priv_data;
7198 MpegEncContext *s = &h->s;
7200 free_tables(h); //FIXME cleanup init stuff perhaps
7203 // memset(h, 0, sizeof(H264Context));
7209 AVCodec h264_decoder = {
7213 sizeof(H264Context),
7218 /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 | CODEC_CAP_TRUNCATED | CODEC_CAP_DELAY,
7222 AVCodecParser h264_parser = {
7224 sizeof(H264Context),