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_skiped; //FIXME remove (IMHO not used)
168 int chroma_pred_mode;
169 int intra16x16_pred_mode;
171 int8_t intra4x4_pred_mode_cache[5*8];
172 int8_t (*intra4x4_pred_mode)[8];
173 void (*pred4x4 [9+3])(uint8_t *src, uint8_t *topright, int stride);//FIXME move to dsp?
174 void (*pred8x8 [4+3])(uint8_t *src, int stride);
175 void (*pred16x16[4+3])(uint8_t *src, int stride);
176 unsigned int topleft_samples_available;
177 unsigned int top_samples_available;
178 unsigned int topright_samples_available;
179 unsigned int left_samples_available;
180 uint8_t (*top_border)[16+2*8];
181 uint8_t left_border[17+2*9];
184 * non zero coeff count cache.
185 * is 64 if not available.
187 uint8_t non_zero_count_cache[6*8] __align8;
188 uint8_t (*non_zero_count)[16];
191 * Motion vector cache.
193 int16_t mv_cache[2][5*8][2] __align8;
194 int8_t ref_cache[2][5*8] __align8;
195 #define LIST_NOT_USED -1 //FIXME rename?
196 #define PART_NOT_AVAILABLE -2
199 * is 1 if the specific list MV&references are set to 0,0,-2.
201 int mv_cache_clean[2];
204 * block_offset[ 0..23] for frame macroblocks
205 * block_offset[24..47] for field macroblocks
207 int block_offset[2*(16+8)];
209 uint16_t *mb2b_xy; //FIXME are these 4 a good idea?
211 int b_stride; //FIXME use s->b4_stride
217 int unknown_svq3_flag;
218 int next_slice_index;
220 SPS sps_buffer[MAX_SPS_COUNT];
221 SPS sps; ///< current sps
223 PPS pps_buffer[MAX_PPS_COUNT];
227 PPS pps; //FIXME move tp Picture perhaps? (->no) do we need that?
230 uint8_t *slice_table_base;
231 uint8_t *slice_table; ///< slice_table_base + mb_stride + 1
233 int slice_type_fixed;
235 //interlacing specific flags
236 int mb_field_decoding_flag;
243 int delta_poc_bottom;
246 int prev_poc_msb; ///< poc_msb of the last reference pic for POC type 0
247 int prev_poc_lsb; ///< poc_lsb of the last reference pic for POC type 0
248 int frame_num_offset; ///< for POC type 2
249 int prev_frame_num_offset; ///< for POC type 2
250 int prev_frame_num; ///< frame_num of the last pic for POC type 1/2
253 * frame_num for frames or 2*frame_num for field pics.
258 * max_frame_num or 2*max_frame_num for field pics.
262 //Weighted pred stuff
264 int use_weight_chroma;
265 int luma_log2_weight_denom;
266 int chroma_log2_weight_denom;
267 int luma_weight[2][16];
268 int luma_offset[2][16];
269 int chroma_weight[2][16][2];
270 int chroma_offset[2][16][2];
271 int implicit_weight[16][16];
274 int deblocking_filter; ///< disable_deblocking_filter_idc with 1<->0
275 int slice_alpha_c0_offset;
276 int slice_beta_offset;
278 int redundant_pic_count;
280 int direct_spatial_mv_pred;
281 int dist_scale_factor[16];
282 int map_col_to_list0[2][16];
285 * num_ref_idx_l0/1_active_minus1 + 1
287 int ref_count[2];// FIXME split for AFF
288 Picture *short_ref[16];
289 Picture *long_ref[16];
290 Picture default_ref_list[2][32];
291 Picture ref_list[2][32]; //FIXME size?
292 Picture field_ref_list[2][32]; //FIXME size?
293 Picture *delayed_pic[16]; //FIXME size?
294 Picture *delayed_output_pic;
297 * memory management control operations buffer.
299 MMCO mmco[MAX_MMCO_COUNT];
302 int long_ref_count; ///< number of actual long term references
303 int short_ref_count; ///< number of actual short term references
306 GetBitContext intra_gb;
307 GetBitContext inter_gb;
308 GetBitContext *intra_gb_ptr;
309 GetBitContext *inter_gb_ptr;
311 DCTELEM mb[16*24] __align8;
317 uint8_t cabac_state[399];
320 /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0,1,2), 0x0? luma_cbp */
324 /* chroma_pred_mode for i4x4 or i16x16, else 0 */
325 uint8_t *chroma_pred_mode_table;
326 int last_qscale_diff;
327 int16_t (*mvd_table[2])[2];
328 int16_t mvd_cache[2][5*8][2] __align8;
329 uint8_t *direct_table;
330 uint8_t direct_cache[5*8];
334 static VLC coeff_token_vlc[4];
335 static VLC chroma_dc_coeff_token_vlc;
337 static VLC total_zeros_vlc[15];
338 static VLC chroma_dc_total_zeros_vlc[3];
340 static VLC run_vlc[6];
343 static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
344 static void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
345 static void filter_mb( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr);
347 static inline uint32_t pack16to32(int a, int b){
348 #ifdef WORDS_BIGENDIAN
349 return (b&0xFFFF) + (a<<16);
351 return (a&0xFFFF) + (b<<16);
357 * @param h height of the rectangle, should be a constant
358 * @param w width of the rectangle, should be a constant
359 * @param size the size of val (1 or 4), should be a constant
361 static inline void fill_rectangle(void *vp, int w, int h, int stride, uint32_t val, int size){ //FIXME ensure this IS inlined
362 uint8_t *p= (uint8_t*)vp;
363 assert(size==1 || size==4);
368 assert((((int)vp)&(FFMIN(w, STRIDE_ALIGN)-1)) == 0);
369 //FIXME check what gcc generates for 64 bit on x86 and possible write a 32 bit ver of it
372 *(uint16_t*)(p + stride)= size==4 ? val : val*0x0101;
373 }else if(w==2 && h==4){
374 *(uint16_t*)(p + 0*stride)=
375 *(uint16_t*)(p + 1*stride)=
376 *(uint16_t*)(p + 2*stride)=
377 *(uint16_t*)(p + 3*stride)= size==4 ? val : val*0x0101;
378 }else if(w==4 && h==1){
379 *(uint32_t*)(p + 0*stride)= size==4 ? val : val*0x01010101;
380 }else if(w==4 && h==2){
381 *(uint32_t*)(p + 0*stride)=
382 *(uint32_t*)(p + 1*stride)= size==4 ? val : val*0x01010101;
383 }else if(w==4 && h==4){
384 *(uint32_t*)(p + 0*stride)=
385 *(uint32_t*)(p + 1*stride)=
386 *(uint32_t*)(p + 2*stride)=
387 *(uint32_t*)(p + 3*stride)= size==4 ? val : val*0x01010101;
388 }else if(w==8 && h==1){
390 *(uint32_t*)(p + 4)= size==4 ? val : val*0x01010101;
391 }else if(w==8 && h==2){
392 *(uint32_t*)(p + 0 + 0*stride)=
393 *(uint32_t*)(p + 4 + 0*stride)=
394 *(uint32_t*)(p + 0 + 1*stride)=
395 *(uint32_t*)(p + 4 + 1*stride)= size==4 ? val : val*0x01010101;
396 }else if(w==8 && h==4){
397 *(uint64_t*)(p + 0*stride)=
398 *(uint64_t*)(p + 1*stride)=
399 *(uint64_t*)(p + 2*stride)=
400 *(uint64_t*)(p + 3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
401 }else if(w==16 && h==2){
402 *(uint64_t*)(p + 0+0*stride)=
403 *(uint64_t*)(p + 8+0*stride)=
404 *(uint64_t*)(p + 0+1*stride)=
405 *(uint64_t*)(p + 8+1*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
406 }else if(w==16 && h==4){
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)=
411 *(uint64_t*)(p + 0+2*stride)=
412 *(uint64_t*)(p + 8+2*stride)=
413 *(uint64_t*)(p + 0+3*stride)=
414 *(uint64_t*)(p + 8+3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
419 static inline void fill_caches(H264Context *h, int mb_type, int for_deblock){
420 MpegEncContext * const s = &h->s;
421 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
422 int topleft_xy, top_xy, topright_xy, left_xy[2];
423 int topleft_type, top_type, topright_type, left_type[2];
427 //wow what a mess, why didnt they simplify the interlacing&intra stuff, i cant imagine that these complex rules are worth it
429 top_xy = mb_xy - s->mb_stride;
430 topleft_xy = top_xy - 1;
431 topright_xy= top_xy + 1;
432 left_xy[1] = left_xy[0] = mb_xy-1;
442 const int pair_xy = s->mb_x + (s->mb_y & ~1)*s->mb_stride;
443 const int top_pair_xy = pair_xy - s->mb_stride;
444 const int topleft_pair_xy = top_pair_xy - 1;
445 const int topright_pair_xy = top_pair_xy + 1;
446 const int topleft_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[topleft_pair_xy]);
447 const int top_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[top_pair_xy]);
448 const int topright_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[topright_pair_xy]);
449 const int left_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[pair_xy-1]);
450 const int curr_mb_frame_flag = !IS_INTERLACED(mb_type);
451 const int bottom = (s->mb_y & 1);
452 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);
454 ? !curr_mb_frame_flag // bottom macroblock
455 : (!curr_mb_frame_flag && !top_mb_frame_flag) // top macroblock
457 top_xy -= s->mb_stride;
460 ? !curr_mb_frame_flag // bottom macroblock
461 : (!curr_mb_frame_flag && !topleft_mb_frame_flag) // top macroblock
463 topleft_xy -= s->mb_stride;
466 ? !curr_mb_frame_flag // bottom macroblock
467 : (!curr_mb_frame_flag && !topright_mb_frame_flag) // top macroblock
469 topright_xy -= s->mb_stride;
471 if (left_mb_frame_flag != curr_mb_frame_flag) {
472 left_xy[1] = left_xy[0] = pair_xy - 1;
473 if (curr_mb_frame_flag) {
494 left_xy[1] += s->mb_stride;
508 topleft_type = h->slice_table[topleft_xy ] < 255 ? s->current_picture.mb_type[topleft_xy] : 0;
509 top_type = h->slice_table[top_xy ] < 255 ? s->current_picture.mb_type[top_xy] : 0;
510 topright_type= h->slice_table[topright_xy] < 255 ? s->current_picture.mb_type[topright_xy]: 0;
511 left_type[0] = h->slice_table[left_xy[0] ] < 255 ? s->current_picture.mb_type[left_xy[0]] : 0;
512 left_type[1] = h->slice_table[left_xy[1] ] < 255 ? s->current_picture.mb_type[left_xy[1]] : 0;
514 topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
515 top_type = h->slice_table[top_xy ] == h->slice_num ? s->current_picture.mb_type[top_xy] : 0;
516 topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
517 left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
518 left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
521 if(IS_INTRA(mb_type)){
522 h->topleft_samples_available=
523 h->top_samples_available=
524 h->left_samples_available= 0xFFFF;
525 h->topright_samples_available= 0xEEEA;
527 if(!IS_INTRA(top_type) && (top_type==0 || h->pps.constrained_intra_pred)){
528 h->topleft_samples_available= 0xB3FF;
529 h->top_samples_available= 0x33FF;
530 h->topright_samples_available= 0x26EA;
533 if(!IS_INTRA(left_type[i]) && (left_type[i]==0 || h->pps.constrained_intra_pred)){
534 h->topleft_samples_available&= 0xDF5F;
535 h->left_samples_available&= 0x5F5F;
539 if(!IS_INTRA(topleft_type) && (topleft_type==0 || h->pps.constrained_intra_pred))
540 h->topleft_samples_available&= 0x7FFF;
542 if(!IS_INTRA(topright_type) && (topright_type==0 || h->pps.constrained_intra_pred))
543 h->topright_samples_available&= 0xFBFF;
545 if(IS_INTRA4x4(mb_type)){
546 if(IS_INTRA4x4(top_type)){
547 h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
548 h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
549 h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
550 h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
553 if(!top_type || (IS_INTER(top_type) && h->pps.constrained_intra_pred))
558 h->intra4x4_pred_mode_cache[4+8*0]=
559 h->intra4x4_pred_mode_cache[5+8*0]=
560 h->intra4x4_pred_mode_cache[6+8*0]=
561 h->intra4x4_pred_mode_cache[7+8*0]= pred;
564 if(IS_INTRA4x4(left_type[i])){
565 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
566 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
569 if(!left_type[i] || (IS_INTER(left_type[i]) && h->pps.constrained_intra_pred))
574 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
575 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
590 //FIXME constraint_intra_pred & partitioning & nnz (lets hope this is just a typo in the spec)
592 h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][4];
593 h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][5];
594 h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][6];
595 h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3];
597 h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][9];
598 h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8];
600 h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][12];
601 h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11];
603 h->top_cbp= h->cbp_table[top_xy];
605 h->non_zero_count_cache[4+8*0]=
606 h->non_zero_count_cache[5+8*0]=
607 h->non_zero_count_cache[6+8*0]=
608 h->non_zero_count_cache[7+8*0]=
610 h->non_zero_count_cache[1+8*0]=
611 h->non_zero_count_cache[2+8*0]=
613 h->non_zero_count_cache[1+8*3]=
614 h->non_zero_count_cache[2+8*3]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
616 if(IS_INTRA(mb_type)) h->top_cbp= 0x1C0;
620 for (i=0; i<2; i++) {
622 h->non_zero_count_cache[3+8*1 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[0+2*i]];
623 h->non_zero_count_cache[3+8*2 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[1+2*i]];
624 h->non_zero_count_cache[0+8*1 + 8*i]= h->non_zero_count[left_xy[i]][left_block[4+2*i]];
625 h->non_zero_count_cache[0+8*4 + 8*i]= h->non_zero_count[left_xy[i]][left_block[5+2*i]];
626 h->left_cbp= h->cbp_table[left_xy[i]]; //FIXME interlacing
628 h->non_zero_count_cache[3+8*1 + 2*8*i]=
629 h->non_zero_count_cache[3+8*2 + 2*8*i]=
630 h->non_zero_count_cache[0+8*1 + 8*i]=
631 h->non_zero_count_cache[0+8*4 + 8*i]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
633 if(IS_INTRA(mb_type)) h->left_cbp= 0x1C0;//FIXME interlacing
639 //FIXME direct mb can skip much of this
640 if(IS_INTER(mb_type) || (IS_DIRECT(mb_type) && h->direct_spatial_mv_pred)){
642 for(list=0; list<2; list++){
643 if(!USES_LIST(mb_type, list) && !IS_DIRECT(mb_type) && !for_deblock){
644 /*if(!h->mv_cache_clean[list]){
645 memset(h->mv_cache [list], 0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
646 memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
647 h->mv_cache_clean[list]= 1;
651 h->mv_cache_clean[list]= 0;
653 if(IS_INTER(topleft_type)){
654 const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
655 const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + h->b8_stride;
656 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
657 h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
659 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
660 h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
663 if(IS_INTER(top_type)){
664 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
665 const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
666 *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0];
667 *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];
668 *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];
669 *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];
670 h->ref_cache[list][scan8[0] + 0 - 1*8]=
671 h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
672 h->ref_cache[list][scan8[0] + 2 - 1*8]=
673 h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
675 *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]=
676 *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]=
677 *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]=
678 *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;
679 *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
682 if(IS_INTER(topright_type)){
683 const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
684 const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
685 *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
686 h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
688 *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
689 h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
692 //FIXME unify cleanup or sth
693 if(IS_INTER(left_type[0])){
694 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
695 const int b8_xy= h->mb2b8_xy[left_xy[0]] + 1;
696 *(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]];
697 *(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]];
698 h->ref_cache[list][scan8[0] - 1 + 0*8]=
699 h->ref_cache[list][scan8[0] - 1 + 1*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0]>>1)];
701 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 0*8]=
702 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 1*8]= 0;
703 h->ref_cache[list][scan8[0] - 1 + 0*8]=
704 h->ref_cache[list][scan8[0] - 1 + 1*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
707 if(IS_INTER(left_type[1])){
708 const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
709 const int b8_xy= h->mb2b8_xy[left_xy[1]] + 1;
710 *(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]];
711 *(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]];
712 h->ref_cache[list][scan8[0] - 1 + 2*8]=
713 h->ref_cache[list][scan8[0] - 1 + 3*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[2]>>1)];
715 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 2*8]=
716 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 3*8]= 0;
717 h->ref_cache[list][scan8[0] - 1 + 2*8]=
718 h->ref_cache[list][scan8[0] - 1 + 3*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
724 h->ref_cache[list][scan8[5 ]+1] =
725 h->ref_cache[list][scan8[7 ]+1] =
726 h->ref_cache[list][scan8[13]+1] = //FIXME remove past 3 (init somewher else)
727 h->ref_cache[list][scan8[4 ]] =
728 h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
729 *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
730 *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
731 *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewher else)
732 *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
733 *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
736 /* XXX beurk, Load mvd */
737 if(IS_INTER(topleft_type)){
738 const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
739 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy];
741 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 - 1*8]= 0;
744 if(IS_INTER(top_type)){
745 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
746 *(uint32_t*)h->mvd_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 0];
747 *(uint32_t*)h->mvd_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 1];
748 *(uint32_t*)h->mvd_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 2];
749 *(uint32_t*)h->mvd_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 3];
751 *(uint32_t*)h->mvd_cache [list][scan8[0] + 0 - 1*8]=
752 *(uint32_t*)h->mvd_cache [list][scan8[0] + 1 - 1*8]=
753 *(uint32_t*)h->mvd_cache [list][scan8[0] + 2 - 1*8]=
754 *(uint32_t*)h->mvd_cache [list][scan8[0] + 3 - 1*8]= 0;
756 if(IS_INTER(left_type[0])){
757 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
758 *(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]];
759 *(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]];
761 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 0*8]=
762 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 1*8]= 0;
764 if(IS_INTER(left_type[1])){
765 const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
766 *(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]];
767 *(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]];
769 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 2*8]=
770 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 3*8]= 0;
772 *(uint32_t*)h->mvd_cache [list][scan8[5 ]+1]=
773 *(uint32_t*)h->mvd_cache [list][scan8[7 ]+1]=
774 *(uint32_t*)h->mvd_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewher else)
775 *(uint32_t*)h->mvd_cache [list][scan8[4 ]]=
776 *(uint32_t*)h->mvd_cache [list][scan8[12]]= 0;
778 if(h->slice_type == B_TYPE){
779 fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, 0, 1);
781 if(IS_DIRECT(top_type)){
782 *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0x01010101;
783 }else if(IS_8X8(top_type)){
784 int b8_xy = h->mb2b8_xy[top_xy] + h->b8_stride;
785 h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy];
786 h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 1];
788 *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0;
792 if(IS_DIRECT(left_type[0])){
793 h->direct_cache[scan8[0] - 1 + 0*8]=
794 h->direct_cache[scan8[0] - 1 + 2*8]= 1;
795 }else if(IS_8X8(left_type[0])){
796 int b8_xy = h->mb2b8_xy[left_xy[0]] + 1;
797 h->direct_cache[scan8[0] - 1 + 0*8]= h->direct_table[b8_xy];
798 h->direct_cache[scan8[0] - 1 + 2*8]= h->direct_table[b8_xy + h->b8_stride];
800 h->direct_cache[scan8[0] - 1 + 0*8]=
801 h->direct_cache[scan8[0] - 1 + 2*8]= 0;
810 static inline void write_back_intra_pred_mode(H264Context *h){
811 MpegEncContext * const s = &h->s;
812 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
814 h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
815 h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
816 h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
817 h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
818 h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
819 h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
820 h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
824 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
826 static inline int check_intra4x4_pred_mode(H264Context *h){
827 MpegEncContext * const s = &h->s;
828 static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
829 static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
832 if(!(h->top_samples_available&0x8000)){
834 int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
836 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);
839 h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
844 if(!(h->left_samples_available&0x8000)){
846 int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
848 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);
851 h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
857 } //FIXME cleanup like next
860 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
862 static inline int check_intra_pred_mode(H264Context *h, int mode){
863 MpegEncContext * const s = &h->s;
864 static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
865 static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
867 if(mode < 0 || mode > 6) {
868 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);
872 if(!(h->top_samples_available&0x8000)){
875 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);
880 if(!(h->left_samples_available&0x8000)){
883 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);
892 * gets the predicted intra4x4 prediction mode.
894 static inline int pred_intra_mode(H264Context *h, int n){
895 const int index8= scan8[n];
896 const int left= h->intra4x4_pred_mode_cache[index8 - 1];
897 const int top = h->intra4x4_pred_mode_cache[index8 - 8];
898 const int min= FFMIN(left, top);
900 tprintf("mode:%d %d min:%d\n", left ,top, min);
902 if(min<0) return DC_PRED;
906 static inline void write_back_non_zero_count(H264Context *h){
907 MpegEncContext * const s = &h->s;
908 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
910 h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[7+8*1];
911 h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[7+8*2];
912 h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[7+8*3];
913 h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];
914 h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[4+8*4];
915 h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[5+8*4];
916 h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[6+8*4];
918 h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[1+8*2];
919 h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
920 h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[2+8*1];
922 h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[1+8*5];
923 h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
924 h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[2+8*4];
928 * gets the predicted number of non zero coefficients.
929 * @param n block index
931 static inline int pred_non_zero_count(H264Context *h, int n){
932 const int index8= scan8[n];
933 const int left= h->non_zero_count_cache[index8 - 1];
934 const int top = h->non_zero_count_cache[index8 - 8];
937 if(i<64) i= (i+1)>>1;
939 tprintf("pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
944 static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
945 const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];
947 if(topright_ref != PART_NOT_AVAILABLE){
948 *C= h->mv_cache[list][ i - 8 + part_width ];
951 tprintf("topright MV not available\n");
953 *C= h->mv_cache[list][ i - 8 - 1 ];
954 return h->ref_cache[list][ i - 8 - 1 ];
959 * gets the predicted MV.
960 * @param n the block index
961 * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)
962 * @param mx the x component of the predicted motion vector
963 * @param my the y component of the predicted motion vector
965 static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
966 const int index8= scan8[n];
967 const int top_ref= h->ref_cache[list][ index8 - 8 ];
968 const int left_ref= h->ref_cache[list][ index8 - 1 ];
969 const int16_t * const A= h->mv_cache[list][ index8 - 1 ];
970 const int16_t * const B= h->mv_cache[list][ index8 - 8 ];
972 int diagonal_ref, match_count;
974 assert(part_width==1 || part_width==2 || part_width==4);
984 diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);
985 match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);
986 tprintf("pred_motion match_count=%d\n", match_count);
987 if(match_count > 1){ //most common
988 *mx= mid_pred(A[0], B[0], C[0]);
989 *my= mid_pred(A[1], B[1], C[1]);
990 }else if(match_count==1){
994 }else if(top_ref==ref){
1002 if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
1006 *mx= mid_pred(A[0], B[0], C[0]);
1007 *my= mid_pred(A[1], B[1], C[1]);
1011 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);
1015 * gets the directionally predicted 16x8 MV.
1016 * @param n the block index
1017 * @param mx the x component of the predicted motion vector
1018 * @param my the y component of the predicted motion vector
1020 static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
1022 const int top_ref= h->ref_cache[list][ scan8[0] - 8 ];
1023 const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
1025 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);
1033 const int left_ref= h->ref_cache[list][ scan8[8] - 1 ];
1034 const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];
1036 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);
1038 if(left_ref == ref){
1046 pred_motion(h, n, 4, list, ref, mx, my);
1050 * gets the directionally predicted 8x16 MV.
1051 * @param n the block index
1052 * @param mx the x component of the predicted motion vector
1053 * @param my the y component of the predicted motion vector
1055 static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
1057 const int left_ref= h->ref_cache[list][ scan8[0] - 1 ];
1058 const int16_t * const A= h->mv_cache[list][ scan8[0] - 1 ];
1060 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);
1062 if(left_ref == ref){
1071 diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);
1073 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);
1075 if(diagonal_ref == ref){
1083 pred_motion(h, n, 2, list, ref, mx, my);
1086 static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
1087 const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
1088 const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];
1090 tprintf("pred_pskip: (%d) (%d) at %2d %2d\n", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
1092 if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
1093 || (top_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0)
1094 || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){
1100 pred_motion(h, 0, 4, 0, 0, mx, my);
1105 static inline void direct_dist_scale_factor(H264Context * const h){
1106 const int poc = h->s.current_picture_ptr->poc;
1107 const int poc1 = h->ref_list[1][0].poc;
1109 for(i=0; i<h->ref_count[0]; i++){
1110 int poc0 = h->ref_list[0][i].poc;
1111 int td = clip(poc1 - poc0, -128, 127);
1112 if(td == 0 /* FIXME || pic0 is a long-term ref */){
1113 h->dist_scale_factor[i] = 256;
1115 int tb = clip(poc - poc0, -128, 127);
1116 int tx = (16384 + (ABS(td) >> 1)) / td;
1117 h->dist_scale_factor[i] = clip((tb*tx + 32) >> 6, -1024, 1023);
1121 static inline void direct_ref_list_init(H264Context * const h){
1122 MpegEncContext * const s = &h->s;
1123 Picture * const ref1 = &h->ref_list[1][0];
1124 Picture * const cur = s->current_picture_ptr;
1126 if(cur->pict_type == I_TYPE)
1127 cur->ref_count[0] = 0;
1128 if(cur->pict_type != B_TYPE)
1129 cur->ref_count[1] = 0;
1130 for(list=0; list<2; list++){
1131 cur->ref_count[list] = h->ref_count[list];
1132 for(j=0; j<h->ref_count[list]; j++)
1133 cur->ref_poc[list][j] = h->ref_list[list][j].poc;
1135 if(cur->pict_type != B_TYPE || h->direct_spatial_mv_pred)
1137 for(list=0; list<2; list++){
1138 for(i=0; i<ref1->ref_count[list]; i++){
1139 const int poc = ref1->ref_poc[list][i];
1140 h->map_col_to_list0[list][i] = PART_NOT_AVAILABLE;
1141 for(j=0; j<h->ref_count[list]; j++)
1142 if(h->ref_list[list][j].poc == poc){
1143 h->map_col_to_list0[list][i] = j;
1150 static inline void pred_direct_motion(H264Context * const h, int *mb_type){
1151 MpegEncContext * const s = &h->s;
1152 const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
1153 const int b8_xy = 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1154 const int b4_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1155 const int mb_type_col = h->ref_list[1][0].mb_type[mb_xy];
1156 const int16_t (*l1mv0)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[0][b4_xy];
1157 const int8_t *l1ref0 = &h->ref_list[1][0].ref_index[0][b8_xy];
1158 const int8_t *l1ref1 = &h->ref_list[1][0].ref_index[1][b8_xy];
1159 const int is_b8x8 = IS_8X8(*mb_type);
1163 if(IS_8X8(mb_type_col) && !h->sps.direct_8x8_inference_flag){
1164 /* FIXME save sub mb types from previous frames (or derive from MVs)
1165 * so we know exactly what block size to use */
1166 sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */
1167 *mb_type = MB_TYPE_8x8|MB_TYPE_L0L1;
1168 }else if(!is_b8x8 && (IS_16X16(mb_type_col) || IS_INTRA(mb_type_col))){
1169 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1170 *mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */
1172 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1173 *mb_type = MB_TYPE_8x8|MB_TYPE_L0L1;
1176 *mb_type |= MB_TYPE_DIRECT2;
1178 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);
1180 if(h->direct_spatial_mv_pred){
1185 /* ref = min(neighbors) */
1186 for(list=0; list<2; list++){
1187 int refa = h->ref_cache[list][scan8[0] - 1];
1188 int refb = h->ref_cache[list][scan8[0] - 8];
1189 int refc = h->ref_cache[list][scan8[0] - 8 + 4];
1191 refc = h->ref_cache[list][scan8[0] - 8 - 1];
1193 if(ref[list] < 0 || (refb < ref[list] && refb >= 0))
1195 if(ref[list] < 0 || (refc < ref[list] && refc >= 0))
1201 if(ref[0] < 0 && ref[1] < 0){
1202 ref[0] = ref[1] = 0;
1203 mv[0][0] = mv[0][1] =
1204 mv[1][0] = mv[1][1] = 0;
1206 for(list=0; list<2; list++){
1208 pred_motion(h, 0, 4, list, ref[list], &mv[list][0], &mv[list][1]);
1210 mv[list][0] = mv[list][1] = 0;
1215 *mb_type &= ~MB_TYPE_P0L1;
1216 sub_mb_type &= ~MB_TYPE_P0L1;
1217 }else if(ref[0] < 0){
1218 *mb_type &= ~MB_TYPE_P0L0;
1219 sub_mb_type &= ~MB_TYPE_P0L0;
1222 if(IS_16X16(*mb_type)){
1223 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref[0], 1);
1224 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, ref[1], 1);
1225 if(!IS_INTRA(mb_type_col) && l1ref0[0] == 0 &&
1226 ABS(l1mv0[0][0]) <= 1 && ABS(l1mv0[0][1]) <= 1){
1228 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1230 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
1232 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1234 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
1236 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1237 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1240 for(i8=0; i8<4; i8++){
1241 const int x8 = i8&1;
1242 const int y8 = i8>>1;
1244 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1246 h->sub_mb_type[i8] = sub_mb_type;
1248 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1249 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1250 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref[0], 1);
1251 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, ref[1], 1);
1254 if(!IS_INTRA(mb_type_col) && l1ref0[x8 + y8*h->b8_stride] == 0){
1255 for(i4=0; i4<4; i4++){
1256 const int16_t *mv_col = l1mv0[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1257 if(ABS(mv_col[0]) <= 1 && ABS(mv_col[1]) <= 1){
1259 *(uint32_t*)h->mv_cache[0][scan8[i8*4+i4]] = 0;
1261 *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = 0;
1267 }else{ /* direct temporal mv pred */
1268 if(IS_16X16(*mb_type)){
1269 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1);
1270 if(IS_INTRA(mb_type_col)){
1271 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
1272 fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
1273 fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
1275 const int ref0 = l1ref0[0] >= 0 ? h->map_col_to_list0[0][l1ref0[0]]
1276 : h->map_col_to_list0[1][l1ref1[0]];
1277 const int dist_scale_factor = h->dist_scale_factor[ref0];
1278 const int16_t *mv_col = l1mv0[0];
1280 mv_l0[0] = (dist_scale_factor * mv_col[0] + 128) >> 8;
1281 mv_l0[1] = (dist_scale_factor * mv_col[1] + 128) >> 8;
1282 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref0, 1);
1283 fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv_l0[0],mv_l0[1]), 4);
1284 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);
1287 for(i8=0; i8<4; i8++){
1288 const int x8 = i8&1;
1289 const int y8 = i8>>1;
1290 int ref0, dist_scale_factor;
1292 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1294 h->sub_mb_type[i8] = sub_mb_type;
1295 if(IS_INTRA(mb_type_col)){
1296 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1297 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1298 fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1299 fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1303 ref0 = l1ref0[x8 + y8*h->b8_stride];
1305 ref0 = h->map_col_to_list0[0][ref0];
1307 ref0 = h->map_col_to_list0[1][l1ref1[x8 + y8*h->b8_stride]];
1308 dist_scale_factor = h->dist_scale_factor[ref0];
1310 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1311 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1312 for(i4=0; i4<4; i4++){
1313 const int16_t *mv_col = l1mv0[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1314 int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]];
1315 mv_l0[0] = (dist_scale_factor * mv_col[0] + 128) >> 8;
1316 mv_l0[1] = (dist_scale_factor * mv_col[1] + 128) >> 8;
1317 *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] =
1318 pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
1325 static inline void write_back_motion(H264Context *h, int mb_type){
1326 MpegEncContext * const s = &h->s;
1327 const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1328 const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1331 for(list=0; list<2; list++){
1333 if(!USES_LIST(mb_type, list)){
1334 if(1){ //FIXME skip or never read if mb_type doesnt use it
1336 *(uint64_t*)s->current_picture.motion_val[list][b_xy + 0 + y*h->b_stride]=
1337 *(uint64_t*)s->current_picture.motion_val[list][b_xy + 2 + y*h->b_stride]= 0;
1339 if( h->pps.cabac ) {
1340 /* FIXME needed ? */
1342 *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]=
1343 *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= 0;
1347 *(uint16_t*)&s->current_picture.ref_index[list][b8_xy + y*h->b8_stride]= (LIST_NOT_USED&0xFF)*0x0101;
1354 *(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];
1355 *(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];
1357 if( h->pps.cabac ) {
1359 *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+0 + 8*y];
1360 *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+2 + 8*y];
1364 s->current_picture.ref_index[list][b8_xy + 0 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+0 + 16*y];
1365 s->current_picture.ref_index[list][b8_xy + 1 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+2 + 16*y];
1369 if(h->slice_type == B_TYPE && h->pps.cabac){
1370 if(IS_8X8(mb_type)){
1371 h->direct_table[b8_xy+1+0*h->b8_stride] = IS_DIRECT(h->sub_mb_type[1]) ? 1 : 0;
1372 h->direct_table[b8_xy+0+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[2]) ? 1 : 0;
1373 h->direct_table[b8_xy+1+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[3]) ? 1 : 0;
1379 * Decodes a network abstraction layer unit.
1380 * @param consumed is the number of bytes used as input
1381 * @param length is the length of the array
1382 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp ttailing?
1383 * @returns decoded bytes, might be src+1 if no escapes
1385 static uint8_t *decode_nal(H264Context *h, uint8_t *src, int *dst_length, int *consumed, int length){
1389 // src[0]&0x80; //forbidden bit
1390 h->nal_ref_idc= src[0]>>5;
1391 h->nal_unit_type= src[0]&0x1F;
1395 for(i=0; i<length; i++)
1396 printf("%2X ", src[i]);
1398 for(i=0; i+1<length; i+=2){
1399 if(src[i]) continue;
1400 if(i>0 && src[i-1]==0) i--;
1401 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1403 /* startcode, so we must be past the end */
1410 if(i>=length-1){ //no escaped 0
1411 *dst_length= length;
1412 *consumed= length+1; //+1 for the header
1416 h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length);
1417 dst= h->rbsp_buffer;
1419 //printf("deoding esc\n");
1422 //remove escapes (very rare 1:2^22)
1423 if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1424 if(src[si+2]==3){ //escape
1429 }else //next start code
1433 dst[di++]= src[si++];
1437 *consumed= si + 1;//+1 for the header
1438 //FIXME store exact number of bits in the getbitcontext (its needed for decoding)
1444 * @param src the data which should be escaped
1445 * @param dst the target buffer, dst+1 == src is allowed as a special case
1446 * @param length the length of the src data
1447 * @param dst_length the length of the dst array
1448 * @returns length of escaped data in bytes or -1 if an error occured
1450 static int encode_nal(H264Context *h, uint8_t *dst, uint8_t *src, int length, int dst_length){
1451 int i, escape_count, si, di;
1455 assert(dst_length>0);
1457 dst[0]= (h->nal_ref_idc<<5) + h->nal_unit_type;
1459 if(length==0) return 1;
1462 for(i=0; i<length; i+=2){
1463 if(src[i]) continue;
1464 if(i>0 && src[i-1]==0)
1466 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1472 if(escape_count==0){
1474 memcpy(dst+1, src, length);
1478 if(length + escape_count + 1> dst_length)
1481 //this should be damn rare (hopefully)
1483 h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length + escape_count);
1484 temp= h->rbsp_buffer;
1485 //printf("encoding esc\n");
1490 if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1491 temp[di++]= 0; si++;
1492 temp[di++]= 0; si++;
1494 temp[di++]= src[si++];
1497 temp[di++]= src[si++];
1499 memcpy(dst+1, temp, length+escape_count);
1501 assert(di == length+escape_count);
1507 * write 1,10,100,1000,... for alignment, yes its exactly inverse to mpeg4
1509 static void encode_rbsp_trailing(PutBitContext *pb){
1512 length= (-put_bits_count(pb))&7;
1513 if(length) put_bits(pb, length, 0);
1518 * identifies the exact end of the bitstream
1519 * @return the length of the trailing, or 0 if damaged
1521 static int decode_rbsp_trailing(uint8_t *src){
1525 tprintf("rbsp trailing %X\n", v);
1535 * idct tranforms the 16 dc values and dequantize them.
1536 * @param qp quantization parameter
1538 static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp){
1539 const int qmul= dequant_coeff[qp][0];
1542 int temp[16]; //FIXME check if this is a good idea
1543 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
1544 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1546 //memset(block, 64, 2*256);
1549 const int offset= y_offset[i];
1550 const int z0= block[offset+stride*0] + block[offset+stride*4];
1551 const int z1= block[offset+stride*0] - block[offset+stride*4];
1552 const int z2= block[offset+stride*1] - block[offset+stride*5];
1553 const int z3= block[offset+stride*1] + block[offset+stride*5];
1562 const int offset= x_offset[i];
1563 const int z0= temp[4*0+i] + temp[4*2+i];
1564 const int z1= temp[4*0+i] - temp[4*2+i];
1565 const int z2= temp[4*1+i] - temp[4*3+i];
1566 const int z3= temp[4*1+i] + temp[4*3+i];
1568 block[stride*0 +offset]= ((z0 + z3)*qmul + 2)>>2; //FIXME think about merging this into decode_resdual
1569 block[stride*2 +offset]= ((z1 + z2)*qmul + 2)>>2;
1570 block[stride*8 +offset]= ((z1 - z2)*qmul + 2)>>2;
1571 block[stride*10+offset]= ((z0 - z3)*qmul + 2)>>2;
1577 * dct tranforms the 16 dc values.
1578 * @param qp quantization parameter ??? FIXME
1580 static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
1581 // const int qmul= dequant_coeff[qp][0];
1583 int temp[16]; //FIXME check if this is a good idea
1584 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
1585 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1588 const int offset= y_offset[i];
1589 const int z0= block[offset+stride*0] + block[offset+stride*4];
1590 const int z1= block[offset+stride*0] - block[offset+stride*4];
1591 const int z2= block[offset+stride*1] - block[offset+stride*5];
1592 const int z3= block[offset+stride*1] + block[offset+stride*5];
1601 const int offset= x_offset[i];
1602 const int z0= temp[4*0+i] + temp[4*2+i];
1603 const int z1= temp[4*0+i] - temp[4*2+i];
1604 const int z2= temp[4*1+i] - temp[4*3+i];
1605 const int z3= temp[4*1+i] + temp[4*3+i];
1607 block[stride*0 +offset]= (z0 + z3)>>1;
1608 block[stride*2 +offset]= (z1 + z2)>>1;
1609 block[stride*8 +offset]= (z1 - z2)>>1;
1610 block[stride*10+offset]= (z0 - z3)>>1;
1618 static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp){
1619 const int qmul= dequant_coeff[qp][0];
1620 const int stride= 16*2;
1621 const int xStride= 16;
1624 a= block[stride*0 + xStride*0];
1625 b= block[stride*0 + xStride*1];
1626 c= block[stride*1 + xStride*0];
1627 d= block[stride*1 + xStride*1];
1634 block[stride*0 + xStride*0]= ((a+c)*qmul + 0)>>1;
1635 block[stride*0 + xStride*1]= ((e+b)*qmul + 0)>>1;
1636 block[stride*1 + xStride*0]= ((a-c)*qmul + 0)>>1;
1637 block[stride*1 + xStride*1]= ((e-b)*qmul + 0)>>1;
1641 static void chroma_dc_dct_c(DCTELEM *block){
1642 const int stride= 16*2;
1643 const int xStride= 16;
1646 a= block[stride*0 + xStride*0];
1647 b= block[stride*0 + xStride*1];
1648 c= block[stride*1 + xStride*0];
1649 d= block[stride*1 + xStride*1];
1656 block[stride*0 + xStride*0]= (a+c);
1657 block[stride*0 + xStride*1]= (e+b);
1658 block[stride*1 + xStride*0]= (a-c);
1659 block[stride*1 + xStride*1]= (e-b);
1664 * gets the chroma qp.
1666 static inline int get_chroma_qp(H264Context *h, int qscale){
1668 return chroma_qp[clip(qscale + h->pps.chroma_qp_index_offset, 0, 51)];
1673 static void h264_diff_dct_c(DCTELEM *block, uint8_t *src1, uint8_t *src2, int stride){
1675 //FIXME try int temp instead of block
1678 const int d0= src1[0 + i*stride] - src2[0 + i*stride];
1679 const int d1= src1[1 + i*stride] - src2[1 + i*stride];
1680 const int d2= src1[2 + i*stride] - src2[2 + i*stride];
1681 const int d3= src1[3 + i*stride] - src2[3 + i*stride];
1682 const int z0= d0 + d3;
1683 const int z3= d0 - d3;
1684 const int z1= d1 + d2;
1685 const int z2= d1 - d2;
1687 block[0 + 4*i]= z0 + z1;
1688 block[1 + 4*i]= 2*z3 + z2;
1689 block[2 + 4*i]= z0 - z1;
1690 block[3 + 4*i]= z3 - 2*z2;
1694 const int z0= block[0*4 + i] + block[3*4 + i];
1695 const int z3= block[0*4 + i] - block[3*4 + i];
1696 const int z1= block[1*4 + i] + block[2*4 + i];
1697 const int z2= block[1*4 + i] - block[2*4 + i];
1699 block[0*4 + i]= z0 + z1;
1700 block[1*4 + i]= 2*z3 + z2;
1701 block[2*4 + i]= z0 - z1;
1702 block[3*4 + i]= z3 - 2*z2;
1707 //FIXME need to check that this doesnt overflow signed 32 bit for low qp, iam not sure, its very close
1708 //FIXME check that gcc inlines this (and optimizes intra & seperate_dc stuff away)
1709 static inline int quantize_c(DCTELEM *block, uint8_t *scantable, int qscale, int intra, int seperate_dc){
1711 const int * const quant_table= quant_coeff[qscale];
1712 const int bias= intra ? (1<<QUANT_SHIFT)/3 : (1<<QUANT_SHIFT)/6;
1713 const unsigned int threshold1= (1<<QUANT_SHIFT) - bias - 1;
1714 const unsigned int threshold2= (threshold1<<1);
1720 const int dc_bias= intra ? (1<<(QUANT_SHIFT-2))/3 : (1<<(QUANT_SHIFT-2))/6;
1721 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT-2)) - dc_bias - 1;
1722 const unsigned int dc_threshold2= (dc_threshold1<<1);
1724 int level= block[0]*quant_coeff[qscale+18][0];
1725 if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1727 level= (dc_bias + level)>>(QUANT_SHIFT-2);
1730 level= (dc_bias - level)>>(QUANT_SHIFT-2);
1733 // last_non_zero = i;
1738 const int dc_bias= intra ? (1<<(QUANT_SHIFT+1))/3 : (1<<(QUANT_SHIFT+1))/6;
1739 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT+1)) - dc_bias - 1;
1740 const unsigned int dc_threshold2= (dc_threshold1<<1);
1742 int level= block[0]*quant_table[0];
1743 if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1745 level= (dc_bias + level)>>(QUANT_SHIFT+1);
1748 level= (dc_bias - level)>>(QUANT_SHIFT+1);
1751 // last_non_zero = i;
1764 const int j= scantable[i];
1765 int level= block[j]*quant_table[j];
1767 // if( bias+level >= (1<<(QMAT_SHIFT - 3))
1768 // || bias-level >= (1<<(QMAT_SHIFT - 3))){
1769 if(((unsigned)(level+threshold1))>threshold2){
1771 level= (bias + level)>>QUANT_SHIFT;
1774 level= (bias - level)>>QUANT_SHIFT;
1783 return last_non_zero;
1786 static void pred4x4_vertical_c(uint8_t *src, uint8_t *topright, int stride){
1787 const uint32_t a= ((uint32_t*)(src-stride))[0];
1788 ((uint32_t*)(src+0*stride))[0]= a;
1789 ((uint32_t*)(src+1*stride))[0]= a;
1790 ((uint32_t*)(src+2*stride))[0]= a;
1791 ((uint32_t*)(src+3*stride))[0]= a;
1794 static void pred4x4_horizontal_c(uint8_t *src, uint8_t *topright, int stride){
1795 ((uint32_t*)(src+0*stride))[0]= src[-1+0*stride]*0x01010101;
1796 ((uint32_t*)(src+1*stride))[0]= src[-1+1*stride]*0x01010101;
1797 ((uint32_t*)(src+2*stride))[0]= src[-1+2*stride]*0x01010101;
1798 ((uint32_t*)(src+3*stride))[0]= src[-1+3*stride]*0x01010101;
1801 static void pred4x4_dc_c(uint8_t *src, uint8_t *topright, int stride){
1802 const int dc= ( src[-stride] + src[1-stride] + src[2-stride] + src[3-stride]
1803 + src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 4) >>3;
1805 ((uint32_t*)(src+0*stride))[0]=
1806 ((uint32_t*)(src+1*stride))[0]=
1807 ((uint32_t*)(src+2*stride))[0]=
1808 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1811 static void pred4x4_left_dc_c(uint8_t *src, uint8_t *topright, int stride){
1812 const int dc= ( src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 2) >>2;
1814 ((uint32_t*)(src+0*stride))[0]=
1815 ((uint32_t*)(src+1*stride))[0]=
1816 ((uint32_t*)(src+2*stride))[0]=
1817 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1820 static void pred4x4_top_dc_c(uint8_t *src, uint8_t *topright, int stride){
1821 const int dc= ( src[-stride] + src[1-stride] + src[2-stride] + src[3-stride] + 2) >>2;
1823 ((uint32_t*)(src+0*stride))[0]=
1824 ((uint32_t*)(src+1*stride))[0]=
1825 ((uint32_t*)(src+2*stride))[0]=
1826 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1829 static void pred4x4_128_dc_c(uint8_t *src, uint8_t *topright, int stride){
1830 ((uint32_t*)(src+0*stride))[0]=
1831 ((uint32_t*)(src+1*stride))[0]=
1832 ((uint32_t*)(src+2*stride))[0]=
1833 ((uint32_t*)(src+3*stride))[0]= 128U*0x01010101U;
1837 #define LOAD_TOP_RIGHT_EDGE\
1838 const int t4= topright[0];\
1839 const int t5= topright[1];\
1840 const int t6= topright[2];\
1841 const int t7= topright[3];\
1843 #define LOAD_LEFT_EDGE\
1844 const int l0= src[-1+0*stride];\
1845 const int l1= src[-1+1*stride];\
1846 const int l2= src[-1+2*stride];\
1847 const int l3= src[-1+3*stride];\
1849 #define LOAD_TOP_EDGE\
1850 const int t0= src[ 0-1*stride];\
1851 const int t1= src[ 1-1*stride];\
1852 const int t2= src[ 2-1*stride];\
1853 const int t3= src[ 3-1*stride];\
1855 static void pred4x4_down_right_c(uint8_t *src, uint8_t *topright, int stride){
1856 const int lt= src[-1-1*stride];
1860 src[0+3*stride]=(l3 + 2*l2 + l1 + 2)>>2;
1862 src[1+3*stride]=(l2 + 2*l1 + l0 + 2)>>2;
1865 src[2+3*stride]=(l1 + 2*l0 + lt + 2)>>2;
1869 src[3+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1872 src[3+2*stride]=(lt + 2*t0 + t1 + 2)>>2;
1874 src[3+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1875 src[3+0*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1878 static void pred4x4_down_left_c(uint8_t *src, uint8_t *topright, int stride){
1883 src[0+0*stride]=(t0 + t2 + 2*t1 + 2)>>2;
1885 src[0+1*stride]=(t1 + t3 + 2*t2 + 2)>>2;
1888 src[0+2*stride]=(t2 + t4 + 2*t3 + 2)>>2;
1892 src[0+3*stride]=(t3 + t5 + 2*t4 + 2)>>2;
1895 src[1+3*stride]=(t4 + t6 + 2*t5 + 2)>>2;
1897 src[2+3*stride]=(t5 + t7 + 2*t6 + 2)>>2;
1898 src[3+3*stride]=(t6 + 3*t7 + 2)>>2;
1901 static void pred4x4_vertical_right_c(uint8_t *src, uint8_t *topright, int stride){
1902 const int lt= src[-1-1*stride];
1905 const __attribute__((unused)) int unu= l3;
1908 src[1+2*stride]=(lt + t0 + 1)>>1;
1910 src[2+2*stride]=(t0 + t1 + 1)>>1;
1912 src[3+2*stride]=(t1 + t2 + 1)>>1;
1913 src[3+0*stride]=(t2 + t3 + 1)>>1;
1915 src[1+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1917 src[2+3*stride]=(lt + 2*t0 + t1 + 2)>>2;
1919 src[3+3*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1920 src[3+1*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1921 src[0+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1922 src[0+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1925 static void pred4x4_vertical_left_c(uint8_t *src, uint8_t *topright, int stride){
1928 const __attribute__((unused)) int unu= t7;
1930 src[0+0*stride]=(t0 + t1 + 1)>>1;
1932 src[0+2*stride]=(t1 + t2 + 1)>>1;
1934 src[1+2*stride]=(t2 + t3 + 1)>>1;
1936 src[2+2*stride]=(t3 + t4+ 1)>>1;
1937 src[3+2*stride]=(t4 + t5+ 1)>>1;
1938 src[0+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1940 src[0+3*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1942 src[1+3*stride]=(t2 + 2*t3 + t4 + 2)>>2;
1944 src[2+3*stride]=(t3 + 2*t4 + t5 + 2)>>2;
1945 src[3+3*stride]=(t4 + 2*t5 + t6 + 2)>>2;
1948 static void pred4x4_horizontal_up_c(uint8_t *src, uint8_t *topright, int stride){
1951 src[0+0*stride]=(l0 + l1 + 1)>>1;
1952 src[1+0*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1954 src[0+1*stride]=(l1 + l2 + 1)>>1;
1956 src[1+1*stride]=(l1 + 2*l2 + l3 + 2)>>2;
1958 src[0+2*stride]=(l2 + l3 + 1)>>1;
1960 src[1+2*stride]=(l2 + 2*l3 + l3 + 2)>>2;
1969 static void pred4x4_horizontal_down_c(uint8_t *src, uint8_t *topright, int stride){
1970 const int lt= src[-1-1*stride];
1973 const __attribute__((unused)) int unu= t3;
1976 src[2+1*stride]=(lt + l0 + 1)>>1;
1978 src[3+1*stride]=(l0 + 2*lt + t0 + 2)>>2;
1979 src[2+0*stride]=(lt + 2*t0 + t1 + 2)>>2;
1980 src[3+0*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1982 src[2+2*stride]=(l0 + l1 + 1)>>1;
1984 src[3+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1986 src[2+3*stride]=(l1 + l2+ 1)>>1;
1988 src[3+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1989 src[0+3*stride]=(l2 + l3 + 1)>>1;
1990 src[1+3*stride]=(l1 + 2*l2 + l3 + 2)>>2;
1993 static void pred16x16_vertical_c(uint8_t *src, int stride){
1995 const uint32_t a= ((uint32_t*)(src-stride))[0];
1996 const uint32_t b= ((uint32_t*)(src-stride))[1];
1997 const uint32_t c= ((uint32_t*)(src-stride))[2];
1998 const uint32_t d= ((uint32_t*)(src-stride))[3];
2000 for(i=0; i<16; i++){
2001 ((uint32_t*)(src+i*stride))[0]= a;
2002 ((uint32_t*)(src+i*stride))[1]= b;
2003 ((uint32_t*)(src+i*stride))[2]= c;
2004 ((uint32_t*)(src+i*stride))[3]= d;
2008 static void pred16x16_horizontal_c(uint8_t *src, int stride){
2011 for(i=0; i<16; i++){
2012 ((uint32_t*)(src+i*stride))[0]=
2013 ((uint32_t*)(src+i*stride))[1]=
2014 ((uint32_t*)(src+i*stride))[2]=
2015 ((uint32_t*)(src+i*stride))[3]= src[-1+i*stride]*0x01010101;
2019 static void pred16x16_dc_c(uint8_t *src, int stride){
2023 dc+= src[-1+i*stride];
2030 dc= 0x01010101*((dc + 16)>>5);
2032 for(i=0; i<16; i++){
2033 ((uint32_t*)(src+i*stride))[0]=
2034 ((uint32_t*)(src+i*stride))[1]=
2035 ((uint32_t*)(src+i*stride))[2]=
2036 ((uint32_t*)(src+i*stride))[3]= dc;
2040 static void pred16x16_left_dc_c(uint8_t *src, int stride){
2044 dc+= src[-1+i*stride];
2047 dc= 0x01010101*((dc + 8)>>4);
2049 for(i=0; i<16; i++){
2050 ((uint32_t*)(src+i*stride))[0]=
2051 ((uint32_t*)(src+i*stride))[1]=
2052 ((uint32_t*)(src+i*stride))[2]=
2053 ((uint32_t*)(src+i*stride))[3]= dc;
2057 static void pred16x16_top_dc_c(uint8_t *src, int stride){
2063 dc= 0x01010101*((dc + 8)>>4);
2065 for(i=0; i<16; i++){
2066 ((uint32_t*)(src+i*stride))[0]=
2067 ((uint32_t*)(src+i*stride))[1]=
2068 ((uint32_t*)(src+i*stride))[2]=
2069 ((uint32_t*)(src+i*stride))[3]= dc;
2073 static void pred16x16_128_dc_c(uint8_t *src, int stride){
2076 for(i=0; i<16; i++){
2077 ((uint32_t*)(src+i*stride))[0]=
2078 ((uint32_t*)(src+i*stride))[1]=
2079 ((uint32_t*)(src+i*stride))[2]=
2080 ((uint32_t*)(src+i*stride))[3]= 0x01010101U*128U;
2084 static inline void pred16x16_plane_compat_c(uint8_t *src, int stride, const int svq3){
2087 uint8_t *cm = cropTbl + MAX_NEG_CROP;
2088 const uint8_t * const src0 = src+7-stride;
2089 const uint8_t *src1 = src+8*stride-1;
2090 const uint8_t *src2 = src1-2*stride; // == src+6*stride-1;
2091 int H = src0[1] - src0[-1];
2092 int V = src1[0] - src2[ 0];
2093 for(k=2; k<=8; ++k) {
2094 src1 += stride; src2 -= stride;
2095 H += k*(src0[k] - src0[-k]);
2096 V += k*(src1[0] - src2[ 0]);
2099 H = ( 5*(H/4) ) / 16;
2100 V = ( 5*(V/4) ) / 16;
2102 /* required for 100% accuracy */
2103 i = H; H = V; V = i;
2105 H = ( 5*H+32 ) >> 6;
2106 V = ( 5*V+32 ) >> 6;
2109 a = 16*(src1[0] + src2[16] + 1) - 7*(V+H);
2110 for(j=16; j>0; --j) {
2113 for(i=-16; i<0; i+=4) {
2114 src[16+i] = cm[ (b ) >> 5 ];
2115 src[17+i] = cm[ (b+ H) >> 5 ];
2116 src[18+i] = cm[ (b+2*H) >> 5 ];
2117 src[19+i] = cm[ (b+3*H) >> 5 ];
2124 static void pred16x16_plane_c(uint8_t *src, int stride){
2125 pred16x16_plane_compat_c(src, stride, 0);
2128 static void pred8x8_vertical_c(uint8_t *src, int stride){
2130 const uint32_t a= ((uint32_t*)(src-stride))[0];
2131 const uint32_t b= ((uint32_t*)(src-stride))[1];
2134 ((uint32_t*)(src+i*stride))[0]= a;
2135 ((uint32_t*)(src+i*stride))[1]= b;
2139 static void pred8x8_horizontal_c(uint8_t *src, int stride){
2143 ((uint32_t*)(src+i*stride))[0]=
2144 ((uint32_t*)(src+i*stride))[1]= src[-1+i*stride]*0x01010101;
2148 static void pred8x8_128_dc_c(uint8_t *src, int stride){
2152 ((uint32_t*)(src+i*stride))[0]=
2153 ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
2156 ((uint32_t*)(src+i*stride))[0]=
2157 ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
2161 static void pred8x8_left_dc_c(uint8_t *src, int stride){
2167 dc0+= src[-1+i*stride];
2168 dc2+= src[-1+(i+4)*stride];
2170 dc0= 0x01010101*((dc0 + 2)>>2);
2171 dc2= 0x01010101*((dc2 + 2)>>2);
2174 ((uint32_t*)(src+i*stride))[0]=
2175 ((uint32_t*)(src+i*stride))[1]= dc0;
2178 ((uint32_t*)(src+i*stride))[0]=
2179 ((uint32_t*)(src+i*stride))[1]= dc2;
2183 static void pred8x8_top_dc_c(uint8_t *src, int stride){
2189 dc0+= src[i-stride];
2190 dc1+= src[4+i-stride];
2192 dc0= 0x01010101*((dc0 + 2)>>2);
2193 dc1= 0x01010101*((dc1 + 2)>>2);
2196 ((uint32_t*)(src+i*stride))[0]= dc0;
2197 ((uint32_t*)(src+i*stride))[1]= dc1;
2200 ((uint32_t*)(src+i*stride))[0]= dc0;
2201 ((uint32_t*)(src+i*stride))[1]= dc1;
2206 static void pred8x8_dc_c(uint8_t *src, int stride){
2208 int dc0, dc1, dc2, dc3;
2212 dc0+= src[-1+i*stride] + src[i-stride];
2213 dc1+= src[4+i-stride];
2214 dc2+= src[-1+(i+4)*stride];
2216 dc3= 0x01010101*((dc1 + dc2 + 4)>>3);
2217 dc0= 0x01010101*((dc0 + 4)>>3);
2218 dc1= 0x01010101*((dc1 + 2)>>2);
2219 dc2= 0x01010101*((dc2 + 2)>>2);
2222 ((uint32_t*)(src+i*stride))[0]= dc0;
2223 ((uint32_t*)(src+i*stride))[1]= dc1;
2226 ((uint32_t*)(src+i*stride))[0]= dc2;
2227 ((uint32_t*)(src+i*stride))[1]= dc3;
2231 static void pred8x8_plane_c(uint8_t *src, int stride){
2234 uint8_t *cm = cropTbl + MAX_NEG_CROP;
2235 const uint8_t * const src0 = src+3-stride;
2236 const uint8_t *src1 = src+4*stride-1;
2237 const uint8_t *src2 = src1-2*stride; // == src+2*stride-1;
2238 int H = src0[1] - src0[-1];
2239 int V = src1[0] - src2[ 0];
2240 for(k=2; k<=4; ++k) {
2241 src1 += stride; src2 -= stride;
2242 H += k*(src0[k] - src0[-k]);
2243 V += k*(src1[0] - src2[ 0]);
2245 H = ( 17*H+16 ) >> 5;
2246 V = ( 17*V+16 ) >> 5;
2248 a = 16*(src1[0] + src2[8]+1) - 3*(V+H);
2249 for(j=8; j>0; --j) {
2252 src[0] = cm[ (b ) >> 5 ];
2253 src[1] = cm[ (b+ H) >> 5 ];
2254 src[2] = cm[ (b+2*H) >> 5 ];
2255 src[3] = cm[ (b+3*H) >> 5 ];
2256 src[4] = cm[ (b+4*H) >> 5 ];
2257 src[5] = cm[ (b+5*H) >> 5 ];
2258 src[6] = cm[ (b+6*H) >> 5 ];
2259 src[7] = cm[ (b+7*H) >> 5 ];
2264 static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
2265 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2266 int src_x_offset, int src_y_offset,
2267 qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
2268 MpegEncContext * const s = &h->s;
2269 const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
2270 const int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
2271 const int luma_xy= (mx&3) + ((my&3)<<2);
2272 uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*s->linesize;
2273 uint8_t * src_cb= pic->data[1] + (mx>>3) + (my>>3)*s->uvlinesize;
2274 uint8_t * src_cr= pic->data[2] + (mx>>3) + (my>>3)*s->uvlinesize;
2275 int extra_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16; //FIXME increase edge?, IMHO not worth it
2276 int extra_height= extra_width;
2278 const int full_mx= mx>>2;
2279 const int full_my= my>>2;
2281 assert(pic->data[0]);
2283 if(mx&7) extra_width -= 3;
2284 if(my&7) extra_height -= 3;
2286 if( full_mx < 0-extra_width
2287 || full_my < 0-extra_height
2288 || full_mx + 16/*FIXME*/ > s->width + extra_width
2289 || full_my + 16/*FIXME*/ > s->height + extra_height){
2290 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);
2291 src_y= s->edge_emu_buffer + 2 + 2*s->linesize;
2295 qpix_op[luma_xy](dest_y, src_y, s->linesize); //FIXME try variable height perhaps?
2297 qpix_op[luma_xy](dest_y + delta, src_y + delta, s->linesize);
2300 if(s->flags&CODEC_FLAG_GRAY) return;
2303 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);
2304 src_cb= s->edge_emu_buffer;
2306 chroma_op(dest_cb, src_cb, s->uvlinesize, chroma_height, mx&7, my&7);
2309 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);
2310 src_cr= s->edge_emu_buffer;
2312 chroma_op(dest_cr, src_cr, s->uvlinesize, chroma_height, mx&7, my&7);
2315 static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
2316 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2317 int x_offset, int y_offset,
2318 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2319 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
2320 int list0, int list1){
2321 MpegEncContext * const s = &h->s;
2322 qpel_mc_func *qpix_op= qpix_put;
2323 h264_chroma_mc_func chroma_op= chroma_put;
2325 dest_y += 2*x_offset + 2*y_offset*s-> linesize;
2326 dest_cb += x_offset + y_offset*s->uvlinesize;
2327 dest_cr += x_offset + y_offset*s->uvlinesize;
2328 x_offset += 8*s->mb_x;
2329 y_offset += 8*s->mb_y;
2332 Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
2333 mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
2334 dest_y, dest_cb, dest_cr, x_offset, y_offset,
2335 qpix_op, chroma_op);
2338 chroma_op= chroma_avg;
2342 Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
2343 mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
2344 dest_y, dest_cb, dest_cr, x_offset, y_offset,
2345 qpix_op, chroma_op);
2349 static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
2350 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2351 int x_offset, int y_offset,
2352 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2353 h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
2354 h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
2355 int list0, int list1){
2356 MpegEncContext * const s = &h->s;
2358 dest_y += 2*x_offset + 2*y_offset*s-> linesize;
2359 dest_cb += x_offset + y_offset*s->uvlinesize;
2360 dest_cr += x_offset + y_offset*s->uvlinesize;
2361 x_offset += 8*s->mb_x;
2362 y_offset += 8*s->mb_y;
2365 /* don't optimize for luma-only case, since B-frames usually
2366 * use implicit weights => chroma too. */
2367 uint8_t *tmp_cb = s->obmc_scratchpad;
2368 uint8_t *tmp_cr = tmp_cb + 8*s->uvlinesize;
2369 uint8_t *tmp_y = tmp_cr + 8*s->uvlinesize;
2370 int refn0 = h->ref_cache[0][ scan8[n] ];
2371 int refn1 = h->ref_cache[1][ scan8[n] ];
2373 mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
2374 dest_y, dest_cb, dest_cr,
2375 x_offset, y_offset, qpix_put, chroma_put);
2376 mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
2377 tmp_y, tmp_cb, tmp_cr,
2378 x_offset, y_offset, qpix_put, chroma_put);
2380 if(h->use_weight == 2){
2381 int weight0 = h->implicit_weight[refn0][refn1];
2382 int weight1 = 64 - weight0;
2383 luma_weight_avg( dest_y, tmp_y, s-> linesize, 5, weight0, weight1, 0, 0);
2384 chroma_weight_avg(dest_cb, tmp_cb, s->uvlinesize, 5, weight0, weight1, 0, 0);
2385 chroma_weight_avg(dest_cr, tmp_cr, s->uvlinesize, 5, weight0, weight1, 0, 0);
2387 luma_weight_avg(dest_y, tmp_y, s->linesize, h->luma_log2_weight_denom,
2388 h->luma_weight[0][refn0], h->luma_weight[1][refn1],
2389 h->luma_offset[0][refn0], h->luma_offset[1][refn1]);
2390 chroma_weight_avg(dest_cb, tmp_cb, s->uvlinesize, h->chroma_log2_weight_denom,
2391 h->chroma_weight[0][refn0][0], h->chroma_weight[1][refn1][0],
2392 h->chroma_offset[0][refn0][0], h->chroma_offset[1][refn1][0]);
2393 chroma_weight_avg(dest_cr, tmp_cr, s->uvlinesize, h->chroma_log2_weight_denom,
2394 h->chroma_weight[0][refn0][1], h->chroma_weight[1][refn1][1],
2395 h->chroma_offset[0][refn0][1], h->chroma_offset[1][refn1][1]);
2398 int list = list1 ? 1 : 0;
2399 int refn = h->ref_cache[list][ scan8[n] ];
2400 Picture *ref= &h->ref_list[list][refn];
2401 mc_dir_part(h, ref, n, square, chroma_height, delta, list,
2402 dest_y, dest_cb, dest_cr, x_offset, y_offset,
2403 qpix_put, chroma_put);
2405 luma_weight_op(dest_y, s->linesize, h->luma_log2_weight_denom,
2406 h->luma_weight[list][refn], h->luma_offset[list][refn]);
2407 if(h->use_weight_chroma){
2408 chroma_weight_op(dest_cb, s->uvlinesize, h->chroma_log2_weight_denom,
2409 h->chroma_weight[list][refn][0], h->chroma_offset[list][refn][0]);
2410 chroma_weight_op(dest_cr, s->uvlinesize, h->chroma_log2_weight_denom,
2411 h->chroma_weight[list][refn][1], h->chroma_offset[list][refn][1]);
2416 static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
2417 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2418 int x_offset, int y_offset,
2419 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2420 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
2421 h264_weight_func *weight_op, h264_biweight_func *weight_avg,
2422 int list0, int list1){
2423 if((h->use_weight==2 && list0 && list1
2424 && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ] != 32))
2425 || h->use_weight==1)
2426 mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
2427 x_offset, y_offset, qpix_put, chroma_put,
2428 weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
2430 mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
2431 x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
2434 static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2435 qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
2436 qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
2437 h264_weight_func *weight_op, h264_biweight_func *weight_avg){
2438 MpegEncContext * const s = &h->s;
2439 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
2440 const int mb_type= s->current_picture.mb_type[mb_xy];
2442 assert(IS_INTER(mb_type));
2444 if(IS_16X16(mb_type)){
2445 mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
2446 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
2447 &weight_op[0], &weight_avg[0],
2448 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2449 }else if(IS_16X8(mb_type)){
2450 mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
2451 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
2452 &weight_op[1], &weight_avg[1],
2453 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2454 mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
2455 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
2456 &weight_op[1], &weight_avg[1],
2457 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2458 }else if(IS_8X16(mb_type)){
2459 mc_part(h, 0, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 0, 0,
2460 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2461 &weight_op[2], &weight_avg[2],
2462 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2463 mc_part(h, 4, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 4, 0,
2464 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2465 &weight_op[2], &weight_avg[2],
2466 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2470 assert(IS_8X8(mb_type));
2473 const int sub_mb_type= h->sub_mb_type[i];
2475 int x_offset= (i&1)<<2;
2476 int y_offset= (i&2)<<1;
2478 if(IS_SUB_8X8(sub_mb_type)){
2479 mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2480 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2481 &weight_op[3], &weight_avg[3],
2482 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2483 }else if(IS_SUB_8X4(sub_mb_type)){
2484 mc_part(h, n , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2485 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2486 &weight_op[4], &weight_avg[4],
2487 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2488 mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
2489 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2490 &weight_op[4], &weight_avg[4],
2491 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2492 }else if(IS_SUB_4X8(sub_mb_type)){
2493 mc_part(h, n , 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2494 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2495 &weight_op[5], &weight_avg[5],
2496 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2497 mc_part(h, n+1, 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
2498 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2499 &weight_op[5], &weight_avg[5],
2500 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2503 assert(IS_SUB_4X4(sub_mb_type));
2505 int sub_x_offset= x_offset + 2*(j&1);
2506 int sub_y_offset= y_offset + (j&2);
2507 mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
2508 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2509 &weight_op[6], &weight_avg[6],
2510 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2517 static void decode_init_vlc(H264Context *h){
2518 static int done = 0;
2524 init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5,
2525 &chroma_dc_coeff_token_len [0], 1, 1,
2526 &chroma_dc_coeff_token_bits[0], 1, 1, 1);
2529 init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17,
2530 &coeff_token_len [i][0], 1, 1,
2531 &coeff_token_bits[i][0], 1, 1, 1);
2535 init_vlc(&chroma_dc_total_zeros_vlc[i], CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
2536 &chroma_dc_total_zeros_len [i][0], 1, 1,
2537 &chroma_dc_total_zeros_bits[i][0], 1, 1, 1);
2539 for(i=0; i<15; i++){
2540 init_vlc(&total_zeros_vlc[i], TOTAL_ZEROS_VLC_BITS, 16,
2541 &total_zeros_len [i][0], 1, 1,
2542 &total_zeros_bits[i][0], 1, 1, 1);
2546 init_vlc(&run_vlc[i], RUN_VLC_BITS, 7,
2547 &run_len [i][0], 1, 1,
2548 &run_bits[i][0], 1, 1, 1);
2550 init_vlc(&run7_vlc, RUN7_VLC_BITS, 16,
2551 &run_len [6][0], 1, 1,
2552 &run_bits[6][0], 1, 1, 1);
2557 * Sets the intra prediction function pointers.
2559 static void init_pred_ptrs(H264Context *h){
2560 // MpegEncContext * const s = &h->s;
2562 h->pred4x4[VERT_PRED ]= pred4x4_vertical_c;
2563 h->pred4x4[HOR_PRED ]= pred4x4_horizontal_c;
2564 h->pred4x4[DC_PRED ]= pred4x4_dc_c;
2565 h->pred4x4[DIAG_DOWN_LEFT_PRED ]= pred4x4_down_left_c;
2566 h->pred4x4[DIAG_DOWN_RIGHT_PRED]= pred4x4_down_right_c;
2567 h->pred4x4[VERT_RIGHT_PRED ]= pred4x4_vertical_right_c;
2568 h->pred4x4[HOR_DOWN_PRED ]= pred4x4_horizontal_down_c;
2569 h->pred4x4[VERT_LEFT_PRED ]= pred4x4_vertical_left_c;
2570 h->pred4x4[HOR_UP_PRED ]= pred4x4_horizontal_up_c;
2571 h->pred4x4[LEFT_DC_PRED ]= pred4x4_left_dc_c;
2572 h->pred4x4[TOP_DC_PRED ]= pred4x4_top_dc_c;
2573 h->pred4x4[DC_128_PRED ]= pred4x4_128_dc_c;
2575 h->pred8x8[DC_PRED8x8 ]= pred8x8_dc_c;
2576 h->pred8x8[VERT_PRED8x8 ]= pred8x8_vertical_c;
2577 h->pred8x8[HOR_PRED8x8 ]= pred8x8_horizontal_c;
2578 h->pred8x8[PLANE_PRED8x8 ]= pred8x8_plane_c;
2579 h->pred8x8[LEFT_DC_PRED8x8]= pred8x8_left_dc_c;
2580 h->pred8x8[TOP_DC_PRED8x8 ]= pred8x8_top_dc_c;
2581 h->pred8x8[DC_128_PRED8x8 ]= pred8x8_128_dc_c;
2583 h->pred16x16[DC_PRED8x8 ]= pred16x16_dc_c;
2584 h->pred16x16[VERT_PRED8x8 ]= pred16x16_vertical_c;
2585 h->pred16x16[HOR_PRED8x8 ]= pred16x16_horizontal_c;
2586 h->pred16x16[PLANE_PRED8x8 ]= pred16x16_plane_c;
2587 h->pred16x16[LEFT_DC_PRED8x8]= pred16x16_left_dc_c;
2588 h->pred16x16[TOP_DC_PRED8x8 ]= pred16x16_top_dc_c;
2589 h->pred16x16[DC_128_PRED8x8 ]= pred16x16_128_dc_c;
2592 static void free_tables(H264Context *h){
2593 av_freep(&h->intra4x4_pred_mode);
2594 av_freep(&h->chroma_pred_mode_table);
2595 av_freep(&h->cbp_table);
2596 av_freep(&h->mvd_table[0]);
2597 av_freep(&h->mvd_table[1]);
2598 av_freep(&h->direct_table);
2599 av_freep(&h->non_zero_count);
2600 av_freep(&h->slice_table_base);
2601 av_freep(&h->top_border);
2602 h->slice_table= NULL;
2604 av_freep(&h->mb2b_xy);
2605 av_freep(&h->mb2b8_xy);
2607 av_freep(&h->s.obmc_scratchpad);
2612 * needs widzh/height
2614 static int alloc_tables(H264Context *h){
2615 MpegEncContext * const s = &h->s;
2616 const int big_mb_num= s->mb_stride * (s->mb_height+1);
2619 CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8 * sizeof(uint8_t))
2621 CHECKED_ALLOCZ(h->non_zero_count , big_mb_num * 16 * sizeof(uint8_t))
2622 CHECKED_ALLOCZ(h->slice_table_base , big_mb_num * sizeof(uint8_t))
2623 CHECKED_ALLOCZ(h->top_border , s->mb_width * (16+8+8) * sizeof(uint8_t))
2624 CHECKED_ALLOCZ(h->cbp_table, big_mb_num * sizeof(uint16_t))
2626 if( h->pps.cabac ) {
2627 CHECKED_ALLOCZ(h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t))
2628 CHECKED_ALLOCZ(h->mvd_table[0], 32*big_mb_num * sizeof(uint16_t));
2629 CHECKED_ALLOCZ(h->mvd_table[1], 32*big_mb_num * sizeof(uint16_t));
2630 CHECKED_ALLOCZ(h->direct_table, 32*big_mb_num * sizeof(uint8_t));
2633 memset(h->slice_table_base, -1, big_mb_num * sizeof(uint8_t));
2634 h->slice_table= h->slice_table_base + s->mb_stride + 1;
2636 CHECKED_ALLOCZ(h->mb2b_xy , big_mb_num * sizeof(uint16_t));
2637 CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint16_t));
2638 for(y=0; y<s->mb_height; y++){
2639 for(x=0; x<s->mb_width; x++){
2640 const int mb_xy= x + y*s->mb_stride;
2641 const int b_xy = 4*x + 4*y*h->b_stride;
2642 const int b8_xy= 2*x + 2*y*h->b8_stride;
2644 h->mb2b_xy [mb_xy]= b_xy;
2645 h->mb2b8_xy[mb_xy]= b8_xy;
2649 s->obmc_scratchpad = NULL;
2657 static void common_init(H264Context *h){
2658 MpegEncContext * const s = &h->s;
2660 s->width = s->avctx->width;
2661 s->height = s->avctx->height;
2662 s->codec_id= s->avctx->codec->id;
2666 s->unrestricted_mv=1;
2667 s->decode=1; //FIXME
2670 static int decode_init(AVCodecContext *avctx){
2671 H264Context *h= avctx->priv_data;
2672 MpegEncContext * const s = &h->s;
2674 MPV_decode_defaults(s);
2679 s->out_format = FMT_H264;
2680 s->workaround_bugs= avctx->workaround_bugs;
2683 // s->decode_mb= ff_h263_decode_mb;
2685 avctx->pix_fmt= PIX_FMT_YUV420P;
2689 if(avctx->extradata_size > 0 && avctx->extradata &&
2690 *(char *)avctx->extradata == 1){
2700 static void frame_start(H264Context *h){
2701 MpegEncContext * const s = &h->s;
2704 MPV_frame_start(s, s->avctx);
2705 ff_er_frame_start(s);
2707 assert(s->linesize && s->uvlinesize);
2709 for(i=0; i<16; i++){
2710 h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
2711 h->block_offset[24+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
2714 h->block_offset[16+i]=
2715 h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2716 h->block_offset[24+16+i]=
2717 h->block_offset[24+20+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2720 /* can't be in alloc_tables because linesize isn't known there.
2721 * FIXME: redo bipred weight to not require extra buffer? */
2722 if(!s->obmc_scratchpad)
2723 s->obmc_scratchpad = av_malloc(16*s->linesize + 2*8*s->uvlinesize);
2725 // s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
2728 static inline void backup_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){
2729 MpegEncContext * const s = &h->s;
2733 src_cb -= uvlinesize;
2734 src_cr -= uvlinesize;
2736 // There is two lines saved, the line above the the top macroblock of a pair,
2737 // and the line above the bottom macroblock
2738 h->left_border[0]= h->top_border[s->mb_x][15];
2739 for(i=1; i<17; i++){
2740 h->left_border[i]= src_y[15+i* linesize];
2743 *(uint64_t*)(h->top_border[s->mb_x]+0)= *(uint64_t*)(src_y + 16*linesize);
2744 *(uint64_t*)(h->top_border[s->mb_x]+8)= *(uint64_t*)(src_y +8+16*linesize);
2746 if(!(s->flags&CODEC_FLAG_GRAY)){
2747 h->left_border[17 ]= h->top_border[s->mb_x][16+7];
2748 h->left_border[17+9]= h->top_border[s->mb_x][24+7];
2750 h->left_border[i+17 ]= src_cb[7+i*uvlinesize];
2751 h->left_border[i+17+9]= src_cr[7+i*uvlinesize];
2753 *(uint64_t*)(h->top_border[s->mb_x]+16)= *(uint64_t*)(src_cb+8*uvlinesize);
2754 *(uint64_t*)(h->top_border[s->mb_x]+24)= *(uint64_t*)(src_cr+8*uvlinesize);
2758 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){
2759 MpegEncContext * const s = &h->s;
2762 int deblock_left = (s->mb_x > 0);
2763 int deblock_top = (s->mb_y > 0);
2765 src_y -= linesize + 1;
2766 src_cb -= uvlinesize + 1;
2767 src_cr -= uvlinesize + 1;
2769 #define XCHG(a,b,t,xchg)\
2776 for(i = !deblock_top; i<17; i++){
2777 XCHG(h->left_border[i ], src_y [i* linesize], temp8, xchg);
2782 XCHG(*(uint64_t*)(h->top_border[s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
2783 XCHG(*(uint64_t*)(h->top_border[s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
2786 if(!(s->flags&CODEC_FLAG_GRAY)){
2788 for(i = !deblock_top; i<9; i++){
2789 XCHG(h->left_border[i+17 ], src_cb[i*uvlinesize], temp8, xchg);
2790 XCHG(h->left_border[i+17+9], src_cr[i*uvlinesize], temp8, xchg);
2794 XCHG(*(uint64_t*)(h->top_border[s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
2795 XCHG(*(uint64_t*)(h->top_border[s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
2800 static void hl_decode_mb(H264Context *h){
2801 MpegEncContext * const s = &h->s;
2802 const int mb_x= s->mb_x;
2803 const int mb_y= s->mb_y;
2804 const int mb_xy= mb_x + mb_y*s->mb_stride;
2805 const int mb_type= s->current_picture.mb_type[mb_xy];
2806 uint8_t *dest_y, *dest_cb, *dest_cr;
2807 int linesize, uvlinesize /*dct_offset*/;
2809 int *block_offset = &h->block_offset[0];
2814 dest_y = s->current_picture.data[0] + (mb_y * 16* s->linesize ) + mb_x * 16;
2815 dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2816 dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2818 if (h->mb_field_decoding_flag) {
2819 linesize = s->linesize * 2;
2820 uvlinesize = s->uvlinesize * 2;
2821 block_offset = &h->block_offset[24];
2822 if(mb_y&1){ //FIXME move out of this func?
2823 dest_y -= s->linesize*15;
2824 dest_cb-= s->uvlinesize*7;
2825 dest_cr-= s->uvlinesize*7;
2828 linesize = s->linesize;
2829 uvlinesize = s->uvlinesize;
2830 // dct_offset = s->linesize * 16;
2833 if (IS_INTRA_PCM(mb_type)) {
2836 // The pixels are stored in h->mb array in the same order as levels,
2837 // copy them in output in the correct order.
2838 for(i=0; i<16; i++) {
2839 for (y=0; y<4; y++) {
2840 for (x=0; x<4; x++) {
2841 *(dest_y + block_offset[i] + y*linesize + x) = h->mb[i*16+y*4+x];
2845 for(i=16; i<16+4; i++) {
2846 for (y=0; y<4; y++) {
2847 for (x=0; x<4; x++) {
2848 *(dest_cb + block_offset[i] + y*uvlinesize + x) = h->mb[i*16+y*4+x];
2852 for(i=20; i<20+4; i++) {
2853 for (y=0; y<4; y++) {
2854 for (x=0; x<4; x++) {
2855 *(dest_cr + block_offset[i] + y*uvlinesize + x) = h->mb[i*16+y*4+x];
2860 if(IS_INTRA(mb_type)){
2861 if(h->deblocking_filter)
2862 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1);
2864 if(!(s->flags&CODEC_FLAG_GRAY)){
2865 h->pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
2866 h->pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
2869 if(IS_INTRA4x4(mb_type)){
2871 for(i=0; i<16; i++){
2872 uint8_t * const ptr= dest_y + block_offset[i];
2874 const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
2877 if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
2878 const int topright_avail= (h->topright_samples_available<<i)&0x8000;
2879 assert(mb_y || linesize <= block_offset[i]);
2880 if(!topright_avail){
2881 tr= ptr[3 - linesize]*0x01010101;
2882 topright= (uint8_t*) &tr;
2883 }else if(i==5 && h->deblocking_filter){
2884 tr= *(uint32_t*)h->top_border[mb_x+1];
2885 topright= (uint8_t*) &tr;
2887 topright= ptr + 4 - linesize;
2891 h->pred4x4[ dir ](ptr, topright, linesize);
2892 if(h->non_zero_count_cache[ scan8[i] ]){
2893 if(s->codec_id == CODEC_ID_H264)
2894 s->dsp.h264_idct_add(ptr, h->mb + i*16, linesize);
2896 svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
2901 h->pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
2902 if(s->codec_id == CODEC_ID_H264)
2903 h264_luma_dc_dequant_idct_c(h->mb, s->qscale);
2905 svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
2907 if(h->deblocking_filter)
2908 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);
2909 }else if(s->codec_id == CODEC_ID_H264){
2910 hl_motion(h, dest_y, dest_cb, dest_cr,
2911 s->dsp.put_h264_qpel_pixels_tab, s->dsp.put_h264_chroma_pixels_tab,
2912 s->dsp.avg_h264_qpel_pixels_tab, s->dsp.avg_h264_chroma_pixels_tab,
2913 s->dsp.weight_h264_pixels_tab, s->dsp.biweight_h264_pixels_tab);
2917 if(!IS_INTRA4x4(mb_type)){
2918 if(s->codec_id == CODEC_ID_H264){
2919 for(i=0; i<16; i++){
2920 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
2921 uint8_t * const ptr= dest_y + block_offset[i];
2922 s->dsp.h264_idct_add(ptr, h->mb + i*16, linesize);
2926 for(i=0; i<16; i++){
2927 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
2928 uint8_t * const ptr= dest_y + block_offset[i];
2929 svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
2935 if(!(s->flags&CODEC_FLAG_GRAY)){
2936 chroma_dc_dequant_idct_c(h->mb + 16*16, h->chroma_qp);
2937 chroma_dc_dequant_idct_c(h->mb + 16*16+4*16, h->chroma_qp);
2938 if(s->codec_id == CODEC_ID_H264){
2939 for(i=16; i<16+4; i++){
2940 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2941 uint8_t * const ptr= dest_cb + block_offset[i];
2942 s->dsp.h264_idct_add(ptr, h->mb + i*16, uvlinesize);
2945 for(i=20; i<20+4; i++){
2946 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2947 uint8_t * const ptr= dest_cr + block_offset[i];
2948 s->dsp.h264_idct_add(ptr, h->mb + i*16, uvlinesize);
2952 for(i=16; i<16+4; i++){
2953 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2954 uint8_t * const ptr= dest_cb + block_offset[i];
2955 svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
2958 for(i=20; i<20+4; i++){
2959 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2960 uint8_t * const ptr= dest_cr + block_offset[i];
2961 svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
2967 if(h->deblocking_filter) {
2968 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2969 fill_caches(h, mb_type, 1); //FIXME dont fill stuff which isnt used by filter_mb
2970 filter_mb(h, mb_x, mb_y, dest_y, dest_cb, dest_cr);
2975 * fills the default_ref_list.
2977 static int fill_default_ref_list(H264Context *h){
2978 MpegEncContext * const s = &h->s;
2980 int smallest_poc_greater_than_current = -1;
2981 Picture sorted_short_ref[16];
2983 if(h->slice_type==B_TYPE){
2987 /* sort frame according to poc in B slice */
2988 for(out_i=0; out_i<h->short_ref_count; out_i++){
2990 int best_poc=INT_MAX;
2992 for(i=0; i<h->short_ref_count; i++){
2993 const int poc= h->short_ref[i]->poc;
2994 if(poc > limit && poc < best_poc){
3000 assert(best_i != -1);
3003 sorted_short_ref[out_i]= *h->short_ref[best_i];
3004 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);
3005 if (-1 == smallest_poc_greater_than_current) {
3006 if (h->short_ref[best_i]->poc >= s->current_picture_ptr->poc) {
3007 smallest_poc_greater_than_current = out_i;
3013 if(s->picture_structure == PICT_FRAME){
3014 if(h->slice_type==B_TYPE){
3016 tprintf("current poc: %d, smallest_poc_greater_than_current: %d\n", s->current_picture_ptr->poc, smallest_poc_greater_than_current);
3018 // find the largest poc
3019 for(list=0; list<2; list++){
3022 int step= list ? -1 : 1;
3024 for(i=0; i<h->short_ref_count && index < h->ref_count[list]; i++, j+=step) {
3025 while(j<0 || j>= h->short_ref_count){
3027 j= smallest_poc_greater_than_current + (step>>1);
3029 if(sorted_short_ref[j].reference != 3) continue;
3030 h->default_ref_list[list][index ]= sorted_short_ref[j];
3031 h->default_ref_list[list][index++].pic_id= sorted_short_ref[j].frame_num;
3034 for(i = 0; i < 16 && index < h->ref_count[ list ]; i++){
3035 if(h->long_ref[i] == NULL) continue;
3036 if(h->long_ref[i]->reference != 3) continue;
3038 h->default_ref_list[ list ][index ]= *h->long_ref[i];
3039 h->default_ref_list[ list ][index++].pic_id= i;;
3042 if(list && (smallest_poc_greater_than_current<=0 || smallest_poc_greater_than_current>=h->short_ref_count) && (1 < index)){
3043 // swap the two first elements of L1 when
3044 // L0 and L1 are identical
3045 Picture temp= h->default_ref_list[1][0];
3046 h->default_ref_list[1][0] = h->default_ref_list[1][1];
3047 h->default_ref_list[1][0] = temp;
3050 if(index < h->ref_count[ list ])
3051 memset(&h->default_ref_list[list][index], 0, sizeof(Picture)*(h->ref_count[ list ] - index));
3055 for(i=0; i<h->short_ref_count; i++){
3056 if(h->short_ref[i]->reference != 3) continue; //FIXME refernce field shit
3057 h->default_ref_list[0][index ]= *h->short_ref[i];
3058 h->default_ref_list[0][index++].pic_id= h->short_ref[i]->frame_num;
3060 for(i = 0; i < 16; i++){
3061 if(h->long_ref[i] == NULL) continue;
3062 if(h->long_ref[i]->reference != 3) continue;
3063 h->default_ref_list[0][index ]= *h->long_ref[i];
3064 h->default_ref_list[0][index++].pic_id= i;;
3066 if(index < h->ref_count[0])
3067 memset(&h->default_ref_list[0][index], 0, sizeof(Picture)*(h->ref_count[0] - index));
3070 if(h->slice_type==B_TYPE){
3072 //FIXME second field balh
3076 for (i=0; i<h->ref_count[0]; i++) {
3077 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]);
3079 if(h->slice_type==B_TYPE){
3080 for (i=0; i<h->ref_count[1]; i++) {
3081 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]);
3088 static void print_short_term(H264Context *h);
3089 static void print_long_term(H264Context *h);
3091 static int decode_ref_pic_list_reordering(H264Context *h){
3092 MpegEncContext * const s = &h->s;
3095 print_short_term(h);
3097 if(h->slice_type==I_TYPE || h->slice_type==SI_TYPE) return 0; //FIXME move beofre func
3099 for(list=0; list<2; list++){
3100 memcpy(h->ref_list[list], h->default_ref_list[list], sizeof(Picture)*h->ref_count[list]);
3102 if(get_bits1(&s->gb)){
3103 int pred= h->curr_pic_num;
3106 for(index=0; ; index++){
3107 int reordering_of_pic_nums_idc= get_ue_golomb(&s->gb);
3110 Picture *ref = NULL;
3112 if(reordering_of_pic_nums_idc==3)
3115 if(index >= h->ref_count[list]){
3116 av_log(h->s.avctx, AV_LOG_ERROR, "reference count overflow\n");
3120 if(reordering_of_pic_nums_idc<3){
3121 if(reordering_of_pic_nums_idc<2){
3122 const int abs_diff_pic_num= get_ue_golomb(&s->gb) + 1;
3124 if(abs_diff_pic_num >= h->max_pic_num){
3125 av_log(h->s.avctx, AV_LOG_ERROR, "abs_diff_pic_num overflow\n");
3129 if(reordering_of_pic_nums_idc == 0) pred-= abs_diff_pic_num;
3130 else pred+= abs_diff_pic_num;
3131 pred &= h->max_pic_num - 1;
3133 for(i= h->short_ref_count-1; i>=0; i--){
3134 ref = h->short_ref[i];
3135 if(ref->data[0] != NULL && ref->frame_num == pred && ref->long_ref == 0) // ignore non existing pictures by testing data[0] pointer
3139 pic_id= get_ue_golomb(&s->gb); //long_term_pic_idx
3140 ref = h->long_ref[pic_id];
3144 av_log(h->s.avctx, AV_LOG_ERROR, "reference picture missing during reorder\n");
3145 memset(&h->ref_list[list][index], 0, sizeof(Picture)); //FIXME
3147 h->ref_list[list][index]= *ref;
3150 av_log(h->s.avctx, AV_LOG_ERROR, "illegal reordering_of_pic_nums_idc\n");
3156 if(h->slice_type!=B_TYPE) break;
3159 if(h->slice_type==B_TYPE && !h->direct_spatial_mv_pred)
3160 direct_dist_scale_factor(h);
3161 direct_ref_list_init(h);
3165 static int pred_weight_table(H264Context *h){
3166 MpegEncContext * const s = &h->s;
3168 int luma_def, chroma_def;
3171 h->use_weight_chroma= 0;
3172 h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
3173 h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
3174 luma_def = 1<<h->luma_log2_weight_denom;
3175 chroma_def = 1<<h->chroma_log2_weight_denom;
3177 for(list=0; list<2; list++){
3178 for(i=0; i<h->ref_count[list]; i++){
3179 int luma_weight_flag, chroma_weight_flag;
3181 luma_weight_flag= get_bits1(&s->gb);
3182 if(luma_weight_flag){
3183 h->luma_weight[list][i]= get_se_golomb(&s->gb);
3184 h->luma_offset[list][i]= get_se_golomb(&s->gb);
3185 if( h->luma_weight[list][i] != luma_def
3186 || h->luma_offset[list][i] != 0)
3189 h->luma_weight[list][i]= luma_def;
3190 h->luma_offset[list][i]= 0;
3193 chroma_weight_flag= get_bits1(&s->gb);
3194 if(chroma_weight_flag){
3197 h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);
3198 h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);
3199 if( h->chroma_weight[list][i][j] != chroma_def
3200 || h->chroma_offset[list][i][j] != 0)
3201 h->use_weight_chroma= 1;
3206 h->chroma_weight[list][i][j]= chroma_def;
3207 h->chroma_offset[list][i][j]= 0;
3211 if(h->slice_type != B_TYPE) break;
3213 h->use_weight= h->use_weight || h->use_weight_chroma;
3217 static void implicit_weight_table(H264Context *h){
3218 MpegEncContext * const s = &h->s;
3220 int cur_poc = s->current_picture_ptr->poc;
3222 if( h->ref_count[0] == 1 && h->ref_count[1] == 1
3223 && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
3225 h->use_weight_chroma= 0;
3230 h->use_weight_chroma= 2;
3231 h->luma_log2_weight_denom= 5;
3232 h->chroma_log2_weight_denom= 5;
3235 for(ref0=0; ref0 < h->ref_count[0]; ref0++){
3236 int poc0 = h->ref_list[0][ref0].poc;
3237 for(ref1=0; ref1 < h->ref_count[1]; ref1++){
3238 int poc1 = h->ref_list[1][ref1].poc;
3239 int td = clip(poc1 - poc0, -128, 127);
3241 int tb = clip(cur_poc - poc0, -128, 127);
3242 int tx = (16384 + (ABS(td) >> 1)) / td;
3243 int dist_scale_factor = clip((tb*tx + 32) >> 6, -1024, 1023) >> 2;
3244 if(dist_scale_factor < -64 || dist_scale_factor > 128)
3245 h->implicit_weight[ref0][ref1] = 32;
3247 h->implicit_weight[ref0][ref1] = 64 - dist_scale_factor;
3249 h->implicit_weight[ref0][ref1] = 32;
3254 static inline void unreference_pic(H264Context *h, Picture *pic){
3257 if(pic == h->delayed_output_pic)
3260 for(i = 0; h->delayed_pic[i]; i++)
3261 if(pic == h->delayed_pic[i]){
3269 * instantaneous decoder refresh.
3271 static void idr(H264Context *h){
3274 for(i=0; i<16; i++){
3275 if (h->long_ref[i] != NULL) {
3276 unreference_pic(h, h->long_ref[i]);
3277 h->long_ref[i]= NULL;
3280 h->long_ref_count=0;
3282 for(i=0; i<h->short_ref_count; i++){
3283 unreference_pic(h, h->short_ref[i]);
3284 h->short_ref[i]= NULL;
3286 h->short_ref_count=0;
3291 * @return the removed picture or NULL if an error occures
3293 static Picture * remove_short(H264Context *h, int frame_num){
3294 MpegEncContext * const s = &h->s;
3297 if(s->avctx->debug&FF_DEBUG_MMCO)
3298 av_log(h->s.avctx, AV_LOG_DEBUG, "remove short %d count %d\n", frame_num, h->short_ref_count);
3300 for(i=0; i<h->short_ref_count; i++){
3301 Picture *pic= h->short_ref[i];
3302 if(s->avctx->debug&FF_DEBUG_MMCO)
3303 av_log(h->s.avctx, AV_LOG_DEBUG, "%d %d %p\n", i, pic->frame_num, pic);
3304 if(pic->frame_num == frame_num){
3305 h->short_ref[i]= NULL;
3306 memmove(&h->short_ref[i], &h->short_ref[i+1], (h->short_ref_count - i - 1)*sizeof(Picture*));
3307 h->short_ref_count--;
3316 * @return the removed picture or NULL if an error occures
3318 static Picture * remove_long(H264Context *h, int i){
3321 pic= h->long_ref[i];
3322 h->long_ref[i]= NULL;
3323 if(pic) h->long_ref_count--;
3329 * print short term list
3331 static void print_short_term(H264Context *h) {
3333 if(h->s.avctx->debug&FF_DEBUG_MMCO) {
3334 av_log(h->s.avctx, AV_LOG_DEBUG, "short term list:\n");
3335 for(i=0; i<h->short_ref_count; i++){
3336 Picture *pic= h->short_ref[i];
3337 av_log(h->s.avctx, AV_LOG_DEBUG, "%d fn:%d poc:%d %p\n", i, pic->frame_num, pic->poc, pic->data[0]);
3343 * print long term list
3345 static void print_long_term(H264Context *h) {
3347 if(h->s.avctx->debug&FF_DEBUG_MMCO) {
3348 av_log(h->s.avctx, AV_LOG_DEBUG, "long term list:\n");
3349 for(i = 0; i < 16; i++){
3350 Picture *pic= h->long_ref[i];
3352 av_log(h->s.avctx, AV_LOG_DEBUG, "%d fn:%d poc:%d %p\n", i, pic->frame_num, pic->poc, pic->data[0]);
3359 * Executes the reference picture marking (memory management control operations).
3361 static int execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count){
3362 MpegEncContext * const s = &h->s;
3364 int current_is_long=0;
3367 if((s->avctx->debug&FF_DEBUG_MMCO) && mmco_count==0)
3368 av_log(h->s.avctx, AV_LOG_DEBUG, "no mmco here\n");
3370 for(i=0; i<mmco_count; i++){
3371 if(s->avctx->debug&FF_DEBUG_MMCO)
3372 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);
3374 switch(mmco[i].opcode){
3375 case MMCO_SHORT2UNUSED:
3376 pic= remove_short(h, mmco[i].short_frame_num);
3377 if(pic==NULL) return -1;
3378 unreference_pic(h, pic);
3380 case MMCO_SHORT2LONG:
3381 pic= remove_long(h, mmco[i].long_index);
3382 if(pic) unreference_pic(h, pic);
3384 h->long_ref[ mmco[i].long_index ]= remove_short(h, mmco[i].short_frame_num);
3385 h->long_ref[ mmco[i].long_index ]->long_ref=1;
3386 h->long_ref_count++;
3388 case MMCO_LONG2UNUSED:
3389 pic= remove_long(h, mmco[i].long_index);
3390 if(pic==NULL) return -1;
3391 unreference_pic(h, pic);
3394 pic= remove_long(h, mmco[i].long_index);
3395 if(pic) unreference_pic(h, pic);
3397 h->long_ref[ mmco[i].long_index ]= s->current_picture_ptr;
3398 h->long_ref[ mmco[i].long_index ]->long_ref=1;
3399 h->long_ref_count++;
3403 case MMCO_SET_MAX_LONG:
3404 assert(mmco[i].long_index <= 16);
3405 // just remove the long term which index is greater than new max
3406 for(j = mmco[i].long_index; j<16; j++){
3407 pic = remove_long(h, j);
3408 if (pic) unreference_pic(h, pic);
3412 while(h->short_ref_count){
3413 pic= remove_short(h, h->short_ref[0]->frame_num);
3414 unreference_pic(h, pic);
3416 for(j = 0; j < 16; j++) {
3417 pic= remove_long(h, j);
3418 if(pic) unreference_pic(h, pic);
3425 if(!current_is_long){
3426 pic= remove_short(h, s->current_picture_ptr->frame_num);
3428 unreference_pic(h, pic);
3429 av_log(h->s.avctx, AV_LOG_ERROR, "illegal short term buffer state detected\n");
3432 if(h->short_ref_count)
3433 memmove(&h->short_ref[1], &h->short_ref[0], h->short_ref_count*sizeof(Picture*));
3435 h->short_ref[0]= s->current_picture_ptr;
3436 h->short_ref[0]->long_ref=0;
3437 h->short_ref_count++;
3440 print_short_term(h);
3445 static int decode_ref_pic_marking(H264Context *h){
3446 MpegEncContext * const s = &h->s;
3449 if(h->nal_unit_type == NAL_IDR_SLICE){ //FIXME fields
3450 s->broken_link= get_bits1(&s->gb) -1;
3451 h->mmco[0].long_index= get_bits1(&s->gb) - 1; // current_long_term_idx
3452 if(h->mmco[0].long_index == -1)
3455 h->mmco[0].opcode= MMCO_LONG;
3459 if(get_bits1(&s->gb)){ // adaptive_ref_pic_marking_mode_flag
3460 for(i= 0; i<MAX_MMCO_COUNT; i++) {
3461 MMCOOpcode opcode= get_ue_golomb(&s->gb);;
3463 h->mmco[i].opcode= opcode;
3464 if(opcode==MMCO_SHORT2UNUSED || opcode==MMCO_SHORT2LONG){
3465 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
3466 /* if(h->mmco[i].short_frame_num >= h->short_ref_count || h->short_ref[ h->mmco[i].short_frame_num ] == NULL){
3467 fprintf(stderr, "illegal short ref in memory management control operation %d\n", mmco);
3471 if(opcode==MMCO_SHORT2LONG || opcode==MMCO_LONG2UNUSED || opcode==MMCO_LONG || opcode==MMCO_SET_MAX_LONG){
3472 h->mmco[i].long_index= get_ue_golomb(&s->gb);
3473 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){
3474 av_log(h->s.avctx, AV_LOG_ERROR, "illegal long ref in memory management control operation %d\n", opcode);
3479 if(opcode > MMCO_LONG){
3480 av_log(h->s.avctx, AV_LOG_ERROR, "illegal memory management control operation %d\n", opcode);
3483 if(opcode == MMCO_END)
3488 assert(h->long_ref_count + h->short_ref_count <= h->sps.ref_frame_count);
3490 if(h->long_ref_count + h->short_ref_count == h->sps.ref_frame_count){ //FIXME fields
3491 h->mmco[0].opcode= MMCO_SHORT2UNUSED;
3492 h->mmco[0].short_frame_num= h->short_ref[ h->short_ref_count - 1 ]->frame_num;
3502 static int init_poc(H264Context *h){
3503 MpegEncContext * const s = &h->s;
3504 const int max_frame_num= 1<<h->sps.log2_max_frame_num;
3507 if(h->nal_unit_type == NAL_IDR_SLICE){
3508 h->frame_num_offset= 0;
3510 if(h->frame_num < h->prev_frame_num)
3511 h->frame_num_offset= h->prev_frame_num_offset + max_frame_num;
3513 h->frame_num_offset= h->prev_frame_num_offset;
3516 if(h->sps.poc_type==0){
3517 const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
3519 if (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
3520 h->poc_msb = h->prev_poc_msb + max_poc_lsb;
3521 else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
3522 h->poc_msb = h->prev_poc_msb - max_poc_lsb;
3524 h->poc_msb = h->prev_poc_msb;
3525 //printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
3527 field_poc[1] = h->poc_msb + h->poc_lsb;
3528 if(s->picture_structure == PICT_FRAME)
3529 field_poc[1] += h->delta_poc_bottom;
3530 }else if(h->sps.poc_type==1){
3531 int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
3534 if(h->sps.poc_cycle_length != 0)
3535 abs_frame_num = h->frame_num_offset + h->frame_num;
3539 if(h->nal_ref_idc==0 && abs_frame_num > 0)
3542 expected_delta_per_poc_cycle = 0;
3543 for(i=0; i < h->sps.poc_cycle_length; i++)
3544 expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
3546 if(abs_frame_num > 0){
3547 int poc_cycle_cnt = (abs_frame_num - 1) / h->sps.poc_cycle_length;
3548 int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
3550 expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
3551 for(i = 0; i <= frame_num_in_poc_cycle; i++)
3552 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
3556 if(h->nal_ref_idc == 0)
3557 expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
3559 field_poc[0] = expectedpoc + h->delta_poc[0];
3560 field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
3562 if(s->picture_structure == PICT_FRAME)
3563 field_poc[1] += h->delta_poc[1];
3566 if(h->nal_unit_type == NAL_IDR_SLICE){
3569 if(h->nal_ref_idc) poc= 2*(h->frame_num_offset + h->frame_num);
3570 else poc= 2*(h->frame_num_offset + h->frame_num) - 1;
3576 if(s->picture_structure != PICT_BOTTOM_FIELD)
3577 s->current_picture_ptr->field_poc[0]= field_poc[0];
3578 if(s->picture_structure != PICT_TOP_FIELD)
3579 s->current_picture_ptr->field_poc[1]= field_poc[1];
3580 if(s->picture_structure == PICT_FRAME) // FIXME field pix?
3581 s->current_picture_ptr->poc= FFMIN(field_poc[0], field_poc[1]);
3587 * decodes a slice header.
3588 * this will allso call MPV_common_init() and frame_start() as needed
3590 static int decode_slice_header(H264Context *h){
3591 MpegEncContext * const s = &h->s;
3592 int first_mb_in_slice, pps_id;
3593 int num_ref_idx_active_override_flag;
3594 static const uint8_t slice_type_map[5]= {P_TYPE, B_TYPE, I_TYPE, SP_TYPE, SI_TYPE};
3596 int default_ref_list_done = 0;
3598 s->current_picture.reference= h->nal_ref_idc != 0;
3599 s->dropable= h->nal_ref_idc == 0;
3601 first_mb_in_slice= get_ue_golomb(&s->gb);
3603 slice_type= get_ue_golomb(&s->gb);
3605 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);
3610 h->slice_type_fixed=1;
3612 h->slice_type_fixed=0;
3614 slice_type= slice_type_map[ slice_type ];
3615 if (slice_type == I_TYPE
3616 || (h->slice_num != 0 && slice_type == h->slice_type) ) {
3617 default_ref_list_done = 1;
3619 h->slice_type= slice_type;
3621 s->pict_type= h->slice_type; // to make a few old func happy, its wrong though
3623 pps_id= get_ue_golomb(&s->gb);
3625 av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
3628 h->pps= h->pps_buffer[pps_id];
3629 if(h->pps.slice_group_count == 0){
3630 av_log(h->s.avctx, AV_LOG_ERROR, "non existing PPS referenced\n");
3634 h->sps= h->sps_buffer[ h->pps.sps_id ];
3635 if(h->sps.log2_max_frame_num == 0){
3636 av_log(h->s.avctx, AV_LOG_ERROR, "non existing SPS referenced\n");
3640 s->mb_width= h->sps.mb_width;
3641 s->mb_height= h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
3643 h->b_stride= s->mb_width*4 + 1;
3644 h->b8_stride= s->mb_width*2 + 1;
3646 s->width = 16*s->mb_width - 2*(h->sps.crop_left + h->sps.crop_right );
3647 if(h->sps.frame_mbs_only_flag)
3648 s->height= 16*s->mb_height - 2*(h->sps.crop_top + h->sps.crop_bottom);
3650 s->height= 16*s->mb_height - 4*(h->sps.crop_top + h->sps.crop_bottom); //FIXME recheck
3652 if (s->context_initialized
3653 && ( s->width != s->avctx->width || s->height != s->avctx->height)) {
3657 if (!s->context_initialized) {
3658 if (MPV_common_init(s) < 0)
3663 s->avctx->width = s->width;
3664 s->avctx->height = s->height;
3665 s->avctx->sample_aspect_ratio= h->sps.sar;
3666 if(!s->avctx->sample_aspect_ratio.den)
3667 s->avctx->sample_aspect_ratio.den = 1;
3669 if(h->sps.timing_info_present_flag && h->sps.fixed_frame_rate_flag){
3670 s->avctx->frame_rate = h->sps.time_scale;
3671 s->avctx->frame_rate_base = h->sps.num_units_in_tick;
3675 if(h->slice_num == 0){
3679 s->current_picture_ptr->frame_num= //FIXME frame_num cleanup
3680 h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
3682 if(h->sps.frame_mbs_only_flag){
3683 s->picture_structure= PICT_FRAME;
3685 if(get_bits1(&s->gb)) //field_pic_flag
3686 s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
3688 s->picture_structure= PICT_FRAME;
3689 first_mb_in_slice <<= 1;
3693 s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
3694 s->resync_mb_y = s->mb_y = first_mb_in_slice / s->mb_width; //FIXME AFFW
3696 if(s->picture_structure==PICT_FRAME){
3697 h->curr_pic_num= h->frame_num;
3698 h->max_pic_num= 1<< h->sps.log2_max_frame_num;
3700 h->curr_pic_num= 2*h->frame_num;
3701 h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
3704 if(h->nal_unit_type == NAL_IDR_SLICE){
3705 get_ue_golomb(&s->gb); /* idr_pic_id */
3708 if(h->sps.poc_type==0){
3709 h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
3711 if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
3712 h->delta_poc_bottom= get_se_golomb(&s->gb);
3716 if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
3717 h->delta_poc[0]= get_se_golomb(&s->gb);
3719 if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
3720 h->delta_poc[1]= get_se_golomb(&s->gb);
3725 if(h->pps.redundant_pic_cnt_present){
3726 h->redundant_pic_count= get_ue_golomb(&s->gb);
3729 //set defaults, might be overriden a few line later
3730 h->ref_count[0]= h->pps.ref_count[0];
3731 h->ref_count[1]= h->pps.ref_count[1];
3733 if(h->slice_type == P_TYPE || h->slice_type == SP_TYPE || h->slice_type == B_TYPE){
3734 if(h->slice_type == B_TYPE){
3735 h->direct_spatial_mv_pred= get_bits1(&s->gb);
3737 num_ref_idx_active_override_flag= get_bits1(&s->gb);
3739 if(num_ref_idx_active_override_flag){
3740 h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
3741 if(h->slice_type==B_TYPE)
3742 h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
3744 if(h->ref_count[0] > 32 || h->ref_count[1] > 32){
3745 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
3751 if(!default_ref_list_done){
3752 fill_default_ref_list(h);
3755 decode_ref_pic_list_reordering(h);
3757 if( (h->pps.weighted_pred && (h->slice_type == P_TYPE || h->slice_type == SP_TYPE ))
3758 || (h->pps.weighted_bipred_idc==1 && h->slice_type==B_TYPE ) )
3759 pred_weight_table(h);
3760 else if(h->pps.weighted_bipred_idc==2 && h->slice_type==B_TYPE)
3761 implicit_weight_table(h);
3765 if(s->current_picture.reference == 3)
3766 decode_ref_pic_marking(h);
3768 if( h->slice_type != I_TYPE && h->slice_type != SI_TYPE && h->pps.cabac )
3769 h->cabac_init_idc = get_ue_golomb(&s->gb);
3771 h->last_qscale_diff = 0;
3772 s->qscale = h->pps.init_qp + get_se_golomb(&s->gb);
3773 if(s->qscale<0 || s->qscale>51){
3774 av_log(s->avctx, AV_LOG_ERROR, "QP %d out of range\n", s->qscale);
3777 h->chroma_qp = get_chroma_qp(h, s->qscale);
3778 //FIXME qscale / qp ... stuff
3779 if(h->slice_type == SP_TYPE){
3780 get_bits1(&s->gb); /* sp_for_switch_flag */
3782 if(h->slice_type==SP_TYPE || h->slice_type == SI_TYPE){
3783 get_se_golomb(&s->gb); /* slice_qs_delta */
3786 h->deblocking_filter = 1;
3787 h->slice_alpha_c0_offset = 0;
3788 h->slice_beta_offset = 0;
3789 if( h->pps.deblocking_filter_parameters_present ) {
3790 h->deblocking_filter= get_ue_golomb(&s->gb);
3791 if(h->deblocking_filter < 2)
3792 h->deblocking_filter^= 1; // 1<->0
3794 if( h->deblocking_filter ) {
3795 h->slice_alpha_c0_offset = get_se_golomb(&s->gb) << 1;
3796 h->slice_beta_offset = get_se_golomb(&s->gb) << 1;
3801 if( h->pps.num_slice_groups > 1 && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
3802 slice_group_change_cycle= get_bits(&s->gb, ?);
3807 if(s->avctx->debug&FF_DEBUG_PICT_INFO){
3808 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 weight:%d%s\n",
3810 (s->picture_structure==PICT_FRAME ? "F" : s->picture_structure==PICT_TOP_FIELD ? "T" : "B"),
3812 av_get_pict_type_char(h->slice_type),
3813 pps_id, h->frame_num,
3814 s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
3815 h->ref_count[0], h->ref_count[1],
3817 h->deblocking_filter,
3819 h->use_weight==1 && h->use_weight_chroma ? "c" : ""
3829 static inline int get_level_prefix(GetBitContext *gb){
3833 OPEN_READER(re, gb);
3834 UPDATE_CACHE(re, gb);
3835 buf=GET_CACHE(re, gb);
3837 log= 32 - av_log2(buf);
3839 print_bin(buf>>(32-log), log);
3840 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__);
3843 LAST_SKIP_BITS(re, gb, log);
3844 CLOSE_READER(re, gb);
3850 * decodes a residual block.
3851 * @param n block index
3852 * @param scantable scantable
3853 * @param max_coeff number of coefficients in the block
3854 * @return <0 if an error occured
3856 static int decode_residual(H264Context *h, GetBitContext *gb, DCTELEM *block, int n, const uint8_t *scantable, int qp, int max_coeff){
3857 MpegEncContext * const s = &h->s;
3858 const uint16_t *qmul= dequant_coeff[qp];
3859 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};
3860 int level[16], run[16];
3861 int suffix_length, zeros_left, coeff_num, coeff_token, total_coeff, i, trailing_ones;
3863 //FIXME put trailing_onex into the context
3865 if(n == CHROMA_DC_BLOCK_INDEX){
3866 coeff_token= get_vlc2(gb, chroma_dc_coeff_token_vlc.table, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 1);
3867 total_coeff= coeff_token>>2;
3869 if(n == LUMA_DC_BLOCK_INDEX){
3870 total_coeff= pred_non_zero_count(h, 0);
3871 coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
3872 total_coeff= coeff_token>>2;
3874 total_coeff= pred_non_zero_count(h, n);
3875 coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
3876 total_coeff= coeff_token>>2;
3877 h->non_zero_count_cache[ scan8[n] ]= total_coeff;
3881 //FIXME set last_non_zero?
3886 trailing_ones= coeff_token&3;
3887 tprintf("trailing:%d, total:%d\n", trailing_ones, total_coeff);
3888 assert(total_coeff<=16);
3890 for(i=0; i<trailing_ones; i++){
3891 level[i]= 1 - 2*get_bits1(gb);
3894 suffix_length= total_coeff > 10 && trailing_ones < 3;
3896 for(; i<total_coeff; i++){
3897 const int prefix= get_level_prefix(gb);
3898 int level_code, mask;
3900 if(prefix<14){ //FIXME try to build a large unified VLC table for all this
3902 level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
3904 level_code= (prefix<<suffix_length); //part
3905 }else if(prefix==14){
3907 level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
3909 level_code= prefix + get_bits(gb, 4); //part
3910 }else if(prefix==15){
3911 level_code= (prefix<<suffix_length) + get_bits(gb, 12); //part
3912 if(suffix_length==0) level_code+=15; //FIXME doesnt make (much)sense
3914 av_log(h->s.avctx, AV_LOG_ERROR, "prefix too large at %d %d\n", s->mb_x, s->mb_y);
3918 if(i==trailing_ones && i<3) level_code+= 2; //FIXME split first iteration
3920 mask= -(level_code&1);
3921 level[i]= (((2+level_code)>>1) ^ mask) - mask;
3923 if(suffix_length==0) suffix_length=1; //FIXME split first iteration
3926 if(ABS(level[i]) > (3<<(suffix_length-1)) && suffix_length<6) suffix_length++;
3928 if((2+level_code)>>1) > (3<<(suffix_length-1)) && suffix_length<6) suffix_length++;
3929 /* ? == prefix > 2 or sth */
3931 tprintf("level: %d suffix_length:%d\n", level[i], suffix_length);
3934 if(total_coeff == max_coeff)
3937 if(n == CHROMA_DC_BLOCK_INDEX)
3938 zeros_left= get_vlc2(gb, chroma_dc_total_zeros_vlc[ total_coeff-1 ].table, CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 1);
3940 zeros_left= get_vlc2(gb, total_zeros_vlc[ total_coeff-1 ].table, TOTAL_ZEROS_VLC_BITS, 1);
3943 for(i=0; i<total_coeff-1; i++){
3946 else if(zeros_left < 7){
3947 run[i]= get_vlc2(gb, run_vlc[zeros_left-1].table, RUN_VLC_BITS, 1);
3949 run[i]= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2);
3951 zeros_left -= run[i];
3955 av_log(h->s.avctx, AV_LOG_ERROR, "negative number of zero coeffs at %d %d\n", s->mb_x, s->mb_y);
3959 for(; i<total_coeff-1; i++){
3967 for(i=total_coeff-1; i>=0; i--){ //FIXME merge into rundecode?
3970 coeff_num += run[i] + 1; //FIXME add 1 earlier ?
3971 j= scantable[ coeff_num ];
3976 for(i=total_coeff-1; i>=0; i--){ //FIXME merge into rundecode?
3979 coeff_num += run[i] + 1; //FIXME add 1 earlier ?
3980 j= scantable[ coeff_num ];
3982 block[j]= level[i] * qmul[j];
3983 // printf("%d %d ", block[j], qmul[j]);
3990 * decodes a P_SKIP or B_SKIP macroblock
3992 static void decode_mb_skip(H264Context *h){
3993 MpegEncContext * const s = &h->s;
3994 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
3997 memset(h->non_zero_count[mb_xy], 0, 16);
3998 memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui
4000 if( h->slice_type == B_TYPE )
4002 // just for fill_caches. pred_direct_motion will set the real mb_type
4003 mb_type= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2|MB_TYPE_SKIP;
4006 fill_caches(h, mb_type, 0); //FIXME check what is needed and what not ...
4007 pred_direct_motion(h, &mb_type);
4009 fill_rectangle(h->mvd_cache[0][scan8[0]], 4, 4, 8, 0, 4);
4010 fill_rectangle(h->mvd_cache[1][scan8[0]], 4, 4, 8, 0, 4);
4016 mb_type= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0|MB_TYPE_SKIP;
4018 if(h->sps.mb_aff && s->mb_skip_run==0 && (s->mb_y&1)==0){
4019 h->mb_field_decoding_flag= get_bits1(&s->gb);
4021 if(h->mb_field_decoding_flag)
4022 mb_type|= MB_TYPE_INTERLACED;
4024 fill_caches(h, mb_type, 0); //FIXME check what is needed and what not ...
4025 pred_pskip_motion(h, &mx, &my);
4026 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
4027 fill_rectangle( h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4);
4029 fill_rectangle(h->mvd_cache[0][scan8[0]], 4, 4, 8, 0, 4);
4032 write_back_motion(h, mb_type);
4033 s->current_picture.mb_type[mb_xy]= mb_type|MB_TYPE_SKIP;
4034 s->current_picture.qscale_table[mb_xy]= s->qscale;
4035 h->slice_table[ mb_xy ]= h->slice_num;
4036 h->prev_mb_skiped= 1;
4040 * decodes a macroblock
4041 * @returns 0 if ok, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
4043 static int decode_mb_cavlc(H264Context *h){
4044 MpegEncContext * const s = &h->s;
4045 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
4046 int mb_type, partition_count, cbp;
4048 s->dsp.clear_blocks(h->mb); //FIXME avoid if allready clear (move after skip handlong?
4050 tprintf("pic:%d mb:%d/%d\n", h->frame_num, s->mb_x, s->mb_y);
4051 cbp = 0; /* avoid warning. FIXME: find a solution without slowing
4053 if(h->slice_type != I_TYPE && h->slice_type != SI_TYPE){
4054 if(s->mb_skip_run==-1)
4055 s->mb_skip_run= get_ue_golomb(&s->gb);
4057 if (s->mb_skip_run--) {
4062 if(h->sps.mb_aff && s->picture_structure==PICT_FRAME ){
4063 if ( ((s->mb_y&1) == 0) || h->prev_mb_skiped)
4064 h->mb_field_decoding_flag = get_bits1(&s->gb);
4066 h->mb_field_decoding_flag= (s->picture_structure!=PICT_FRAME);
4068 h->prev_mb_skiped= 0;
4070 mb_type= get_ue_golomb(&s->gb);
4071 if(h->slice_type == B_TYPE){
4073 partition_count= b_mb_type_info[mb_type].partition_count;
4074 mb_type= b_mb_type_info[mb_type].type;
4077 goto decode_intra_mb;
4079 }else if(h->slice_type == P_TYPE /*|| h->slice_type == SP_TYPE */){
4081 partition_count= p_mb_type_info[mb_type].partition_count;
4082 mb_type= p_mb_type_info[mb_type].type;
4085 goto decode_intra_mb;
4088 assert(h->slice_type == I_TYPE);
4091 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);
4095 cbp= i_mb_type_info[mb_type].cbp;
4096 h->intra16x16_pred_mode= i_mb_type_info[mb_type].pred_mode;
4097 mb_type= i_mb_type_info[mb_type].type;
4100 if(h->mb_field_decoding_flag)
4101 mb_type |= MB_TYPE_INTERLACED;
4103 s->current_picture.mb_type[mb_xy]= mb_type;
4104 h->slice_table[ mb_xy ]= h->slice_num;
4106 if(IS_INTRA_PCM(mb_type)){
4109 // we assume these blocks are very rare so we dont optimize it
4110 align_get_bits(&s->gb);
4112 // The pixels are stored in the same order as levels in h->mb array.
4113 for(y=0; y<16; y++){
4114 const int index= 4*(y&3) + 32*((y>>2)&1) + 128*(y>>3);
4115 for(x=0; x<16; x++){
4116 tprintf("LUMA ICPM LEVEL (%3d)\n", show_bits(&s->gb, 8));
4117 h->mb[index + (x&3) + 16*((x>>2)&1) + 64*(x>>3)]= get_bits(&s->gb, 8);
4121 const int index= 256 + 4*(y&3) + 32*(y>>2);
4123 tprintf("CHROMA U ICPM LEVEL (%3d)\n", show_bits(&s->gb, 8));
4124 h->mb[index + (x&3) + 16*(x>>2)]= get_bits(&s->gb, 8);
4128 const int index= 256 + 64 + 4*(y&3) + 32*(y>>2);
4130 tprintf("CHROMA V ICPM LEVEL (%3d)\n", show_bits(&s->gb, 8));
4131 h->mb[index + (x&3) + 16*(x>>2)]= get_bits(&s->gb, 8);
4135 // In deblocking, the quantiser is 0
4136 s->current_picture.qscale_table[mb_xy]= 0;
4137 h->chroma_qp = get_chroma_qp(h, 0);
4138 // All coeffs are presents
4139 memset(h->non_zero_count[mb_xy], 16, 16);
4144 fill_caches(h, mb_type, 0);
4147 if(IS_INTRA(mb_type)){
4148 // init_top_left_availability(h);
4149 if(IS_INTRA4x4(mb_type)){
4152 // fill_intra4x4_pred_table(h);
4153 for(i=0; i<16; i++){
4154 const int mode_coded= !get_bits1(&s->gb);
4155 const int predicted_mode= pred_intra_mode(h, i);
4159 const int rem_mode= get_bits(&s->gb, 3);
4160 if(rem_mode<predicted_mode)
4165 mode= predicted_mode;
4168 h->intra4x4_pred_mode_cache[ scan8[i] ] = mode;
4170 write_back_intra_pred_mode(h);
4171 if( check_intra4x4_pred_mode(h) < 0)
4174 h->intra16x16_pred_mode= check_intra_pred_mode(h, h->intra16x16_pred_mode);
4175 if(h->intra16x16_pred_mode < 0)
4178 h->chroma_pred_mode= get_ue_golomb(&s->gb);
4180 h->chroma_pred_mode= check_intra_pred_mode(h, h->chroma_pred_mode);
4181 if(h->chroma_pred_mode < 0)
4183 }else if(partition_count==4){
4184 int i, j, sub_partition_count[4], list, ref[2][4];
4186 if(h->slice_type == B_TYPE){
4188 h->sub_mb_type[i]= get_ue_golomb(&s->gb);
4189 if(h->sub_mb_type[i] >=13){
4190 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);
4193 sub_partition_count[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
4194 h->sub_mb_type[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].type;
4196 if( IS_DIRECT(h->sub_mb_type[0]) || IS_DIRECT(h->sub_mb_type[1])
4197 || IS_DIRECT(h->sub_mb_type[2]) || IS_DIRECT(h->sub_mb_type[3]))
4198 pred_direct_motion(h, &mb_type);
4200 assert(h->slice_type == P_TYPE || h->slice_type == SP_TYPE); //FIXME SP correct ?
4202 h->sub_mb_type[i]= get_ue_golomb(&s->gb);
4203 if(h->sub_mb_type[i] >=4){
4204 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);
4207 sub_partition_count[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
4208 h->sub_mb_type[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].type;
4212 for(list=0; list<2; list++){
4213 const int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list];
4214 if(ref_count == 0) continue;
4216 if(IS_DIRECT(h->sub_mb_type[i])) continue;
4217 if(IS_DIR(h->sub_mb_type[i], 0, list)){
4218 ref[list][i] = get_te0_golomb(&s->gb, ref_count); //FIXME init to 0 before and skip?
4226 for(list=0; list<2; list++){
4227 const int ref_count= IS_REF0(mb_type) ? 1 : h->ref_count[list];
4228 if(ref_count == 0) continue;
4231 if(IS_DIRECT(h->sub_mb_type[i])) continue;
4232 h->ref_cache[list][ scan8[4*i] ]=h->ref_cache[list][ scan8[4*i]+1 ]=
4233 h->ref_cache[list][ scan8[4*i]+8 ]=h->ref_cache[list][ scan8[4*i]+9 ]= ref[list][i];
4235 if(IS_DIR(h->sub_mb_type[i], 0, list)){
4236 const int sub_mb_type= h->sub_mb_type[i];
4237 const int block_width= (sub_mb_type & (MB_TYPE_16x16|MB_TYPE_16x8)) ? 2 : 1;
4238 for(j=0; j<sub_partition_count[i]; j++){
4240 const int index= 4*i + block_width*j;
4241 int16_t (* mv_cache)[2]= &h->mv_cache[list][ scan8[index] ];
4242 pred_motion(h, index, block_width, list, h->ref_cache[list][ scan8[index] ], &mx, &my);
4243 mx += get_se_golomb(&s->gb);
4244 my += get_se_golomb(&s->gb);
4245 tprintf("final mv:%d %d\n", mx, my);
4247 if(IS_SUB_8X8(sub_mb_type)){
4248 mv_cache[ 0 ][0]= mv_cache[ 1 ][0]=
4249 mv_cache[ 8 ][0]= mv_cache[ 9 ][0]= mx;
4250 mv_cache[ 0 ][1]= mv_cache[ 1 ][1]=
4251 mv_cache[ 8 ][1]= mv_cache[ 9 ][1]= my;
4252 }else if(IS_SUB_8X4(sub_mb_type)){
4253 mv_cache[ 0 ][0]= mv_cache[ 1 ][0]= mx;
4254 mv_cache[ 0 ][1]= mv_cache[ 1 ][1]= my;
4255 }else if(IS_SUB_4X8(sub_mb_type)){
4256 mv_cache[ 0 ][0]= mv_cache[ 8 ][0]= mx;
4257 mv_cache[ 0 ][1]= mv_cache[ 8 ][1]= my;
4259 assert(IS_SUB_4X4(sub_mb_type));
4260 mv_cache[ 0 ][0]= mx;
4261 mv_cache[ 0 ][1]= my;
4265 uint32_t *p= (uint32_t *)&h->mv_cache[list][ scan8[4*i] ][0];
4271 }else if(IS_DIRECT(mb_type)){
4272 pred_direct_motion(h, &mb_type);
4273 s->current_picture.mb_type[mb_xy]= mb_type;
4275 int list, mx, my, i;
4276 //FIXME we should set ref_idx_l? to 0 if we use that later ...
4277 if(IS_16X16(mb_type)){
4278 for(list=0; list<2; list++){
4279 if(h->ref_count[list]>0){
4280 if(IS_DIR(mb_type, 0, list)){
4281 const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
4282 fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, val, 1);
4284 fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, (LIST_NOT_USED&0xFF), 1);
4287 for(list=0; list<2; list++){
4288 if(IS_DIR(mb_type, 0, list)){
4289 pred_motion(h, 0, 4, list, h->ref_cache[list][ scan8[0] ], &mx, &my);
4290 mx += get_se_golomb(&s->gb);
4291 my += get_se_golomb(&s->gb);
4292 tprintf("final mv:%d %d\n", mx, my);
4294 fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, pack16to32(mx,my), 4);
4296 fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, 0, 4);
4299 else if(IS_16X8(mb_type)){
4300 for(list=0; list<2; list++){
4301 if(h->ref_count[list]>0){
4303 if(IS_DIR(mb_type, i, list)){
4304 const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
4305 fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, val, 1);
4307 fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, (LIST_NOT_USED&0xFF), 1);
4311 for(list=0; list<2; list++){
4313 if(IS_DIR(mb_type, i, list)){
4314 pred_16x8_motion(h, 8*i, list, h->ref_cache[list][scan8[0] + 16*i], &mx, &my);
4315 mx += get_se_golomb(&s->gb);
4316 my += get_se_golomb(&s->gb);
4317 tprintf("final mv:%d %d\n", mx, my);
4319 fill_rectangle(h->mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, pack16to32(mx,my), 4);
4321 fill_rectangle(h->mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, 0, 4);
4325 assert(IS_8X16(mb_type));
4326 for(list=0; list<2; list++){
4327 if(h->ref_count[list]>0){
4329 if(IS_DIR(mb_type, i, list)){ //FIXME optimize
4330 const int val= get_te0_golomb(&s->gb, h->ref_count[list]);
4331 fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, val, 1);
4333 fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, (LIST_NOT_USED&0xFF), 1);
4337 for(list=0; list<2; list++){
4339 if(IS_DIR(mb_type, i, list)){
4340 pred_8x16_motion(h, i*4, list, h->ref_cache[list][ scan8[0] + 2*i ], &mx, &my);
4341 mx += get_se_golomb(&s->gb);
4342 my += get_se_golomb(&s->gb);
4343 tprintf("final mv:%d %d\n", mx, my);
4345 fill_rectangle(h->mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, pack16to32(mx,my), 4);
4347 fill_rectangle(h->mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, 0, 4);
4353 if(IS_INTER(mb_type))
4354 write_back_motion(h, mb_type);
4356 if(!IS_INTRA16x16(mb_type)){
4357 cbp= get_ue_golomb(&s->gb);
4359 av_log(h->s.avctx, AV_LOG_ERROR, "cbp too large (%d) at %d %d\n", cbp, s->mb_x, s->mb_y);
4363 if(IS_INTRA4x4(mb_type))
4364 cbp= golomb_to_intra4x4_cbp[cbp];
4366 cbp= golomb_to_inter_cbp[cbp];
4369 if(cbp || IS_INTRA16x16(mb_type)){
4370 int i8x8, i4x4, chroma_idx;
4371 int chroma_qp, dquant;
4372 GetBitContext *gb= IS_INTRA(mb_type) ? h->intra_gb_ptr : h->inter_gb_ptr;
4373 const uint8_t *scan, *dc_scan;
4375 // fill_non_zero_count_cache(h);
4377 if(IS_INTERLACED(mb_type)){
4379 dc_scan= luma_dc_field_scan;
4382 dc_scan= luma_dc_zigzag_scan;
4385 dquant= get_se_golomb(&s->gb);
4387 if( dquant > 25 || dquant < -26 ){
4388 av_log(h->s.avctx, AV_LOG_ERROR, "dquant out of range (%d) at %d %d\n", dquant, s->mb_x, s->mb_y);
4392 s->qscale += dquant;
4393 if(((unsigned)s->qscale) > 51){
4394 if(s->qscale<0) s->qscale+= 52;
4395 else s->qscale-= 52;
4398 h->chroma_qp= chroma_qp= get_chroma_qp(h, s->qscale);
4399 if(IS_INTRA16x16(mb_type)){
4400 if( decode_residual(h, h->intra_gb_ptr, h->mb, LUMA_DC_BLOCK_INDEX, dc_scan, s->qscale, 16) < 0){
4401 return -1; //FIXME continue if partotioned and other retirn -1 too
4404 assert((cbp&15) == 0 || (cbp&15) == 15);
4407 for(i8x8=0; i8x8<4; i8x8++){
4408 for(i4x4=0; i4x4<4; i4x4++){
4409 const int index= i4x4 + 4*i8x8;
4410 if( decode_residual(h, h->intra_gb_ptr, h->mb + 16*index, index, scan + 1, s->qscale, 15) < 0 ){
4416 fill_rectangle(&h->non_zero_count_cache[scan8[0]], 4, 4, 8, 0, 1);
4419 for(i8x8=0; i8x8<4; i8x8++){
4420 if(cbp & (1<<i8x8)){
4421 for(i4x4=0; i4x4<4; i4x4++){
4422 const int index= i4x4 + 4*i8x8;
4424 if( decode_residual(h, gb, h->mb + 16*index, index, scan, s->qscale, 16) <0 ){
4429 uint8_t * const nnz= &h->non_zero_count_cache[ scan8[4*i8x8] ];
4430 nnz[0] = nnz[1] = nnz[8] = nnz[9] = 0;
4436 for(chroma_idx=0; chroma_idx<2; chroma_idx++)
4437 if( decode_residual(h, gb, h->mb + 256 + 16*4*chroma_idx, CHROMA_DC_BLOCK_INDEX, chroma_dc_scan, chroma_qp, 4) < 0){
4443 for(chroma_idx=0; chroma_idx<2; chroma_idx++){
4444 for(i4x4=0; i4x4<4; i4x4++){
4445 const int index= 16 + 4*chroma_idx + i4x4;
4446 if( decode_residual(h, gb, h->mb + 16*index, index, scan + 1, chroma_qp, 15) < 0){
4452 uint8_t * const nnz= &h->non_zero_count_cache[0];
4453 nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =
4454 nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
4457 uint8_t * const nnz= &h->non_zero_count_cache[0];
4458 fill_rectangle(&nnz[scan8[0]], 4, 4, 8, 0, 1);
4459 nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =
4460 nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
4462 s->current_picture.qscale_table[mb_xy]= s->qscale;
4463 write_back_non_zero_count(h);
4468 static int decode_cabac_intra_mb_type(H264Context *h, int ctx_base, int intra_slice) {
4469 uint8_t *state= &h->cabac_state[ctx_base];
4473 MpegEncContext * const s = &h->s;
4474 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
4475 const int mba_xy = mb_xy - 1;
4476 const int mbb_xy = mb_xy - s->mb_stride;
4478 if( h->slice_table[mba_xy] == h->slice_num && !IS_INTRA4x4( s->current_picture.mb_type[mba_xy] ) )
4480 if( h->slice_table[mbb_xy] == h->slice_num && !IS_INTRA4x4( s->current_picture.mb_type[mbb_xy] ) )
4482 if( get_cabac( &h->cabac, &state[ctx] ) == 0 )
4483 return 0; /* I4x4 */
4486 if( get_cabac( &h->cabac, &state[0] ) == 0 )
4487 return 0; /* I4x4 */
4490 if( get_cabac_terminate( &h->cabac ) )
4491 return 25; /* PCM */
4493 mb_type = 1; /* I16x16 */
4494 if( get_cabac( &h->cabac, &state[1] ) )
4495 mb_type += 12; /* cbp_luma != 0 */
4497 if( get_cabac( &h->cabac, &state[2] ) ) {
4498 if( get_cabac( &h->cabac, &state[2+intra_slice] ) )
4499 mb_type += 4 * 2; /* cbp_chroma == 2 */
4501 mb_type += 4 * 1; /* cbp_chroma == 1 */
4503 if( get_cabac( &h->cabac, &state[3+intra_slice] ) )
4505 if( get_cabac( &h->cabac, &state[3+2*intra_slice] ) )
4510 static int decode_cabac_mb_type( H264Context *h ) {
4511 MpegEncContext * const s = &h->s;
4513 if( h->slice_type == I_TYPE ) {
4514 return decode_cabac_intra_mb_type(h, 3, 1);
4515 } else if( h->slice_type == P_TYPE ) {
4516 if( get_cabac( &h->cabac, &h->cabac_state[14] ) == 0 ) {
4518 if( get_cabac( &h->cabac, &h->cabac_state[15] ) == 0 ) {
4519 if( get_cabac( &h->cabac, &h->cabac_state[16] ) == 0 )
4520 return 0; /* P_L0_D16x16; */
4522 return 3; /* P_8x8; */
4524 if( get_cabac( &h->cabac, &h->cabac_state[17] ) == 0 )
4525 return 2; /* P_L0_D8x16; */
4527 return 1; /* P_L0_D16x8; */
4530 return decode_cabac_intra_mb_type(h, 17, 0) + 5;
4532 } else if( h->slice_type == B_TYPE ) {
4533 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
4534 const int mba_xy = mb_xy - 1;
4535 const int mbb_xy = mb_xy - s->mb_stride;
4539 if( h->slice_table[mba_xy] == h->slice_num && !IS_SKIP( s->current_picture.mb_type[mba_xy] )
4540 && !IS_DIRECT( s->current_picture.mb_type[mba_xy] ) )
4542 if( h->slice_table[mbb_xy] == h->slice_num && !IS_SKIP( s->current_picture.mb_type[mbb_xy] )
4543 && !IS_DIRECT( s->current_picture.mb_type[mbb_xy] ) )
4546 if( !get_cabac( &h->cabac, &h->cabac_state[27+ctx] ) )
4547 return 0; /* B_Direct_16x16 */
4549 if( !get_cabac( &h->cabac, &h->cabac_state[27+3] ) ) {
4550 return 1 + get_cabac( &h->cabac, &h->cabac_state[27+5] ); /* B_L[01]_16x16 */
4553 bits = get_cabac( &h->cabac, &h->cabac_state[27+4] ) << 3;
4554 bits|= get_cabac( &h->cabac, &h->cabac_state[27+5] ) << 2;
4555 bits|= get_cabac( &h->cabac, &h->cabac_state[27+5] ) << 1;
4556 bits|= get_cabac( &h->cabac, &h->cabac_state[27+5] );
4558 return bits + 3; /* B_Bi_16x16 through B_L1_L0_16x8 */
4559 else if( bits == 13 ) {
4560 return decode_cabac_intra_mb_type(h, 32, 0) + 23;
4561 } else if( bits == 14 )
4562 return 11; /* B_L1_L0_8x16 */
4563 else if( bits == 15 )
4564 return 22; /* B_8x8 */
4566 bits= ( bits<<1 ) | get_cabac( &h->cabac, &h->cabac_state[27+5] );
4567 return bits - 4; /* B_L0_Bi_* through B_Bi_Bi_* */
4569 /* TODO SI/SP frames? */
4574 static int decode_cabac_mb_skip( H264Context *h) {
4575 MpegEncContext * const s = &h->s;
4576 const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
4577 const int mba_xy = mb_xy - 1;
4578 const int mbb_xy = mb_xy - s->mb_stride;
4581 if( h->slice_table[mba_xy] == h->slice_num && !IS_SKIP( s->current_picture.mb_type[mba_xy] ))
4583 if( h->slice_table[mbb_xy] == h->slice_num && !IS_SKIP( s->current_picture.mb_type[mbb_xy] ))
4586 if( h->slice_type == P_TYPE || h->slice_type == SP_TYPE)
4587 return get_cabac( &h->cabac, &h->cabac_state[11+ctx] );
4589 return get_cabac( &h->cabac, &h->cabac_state[24+ctx] );
4592 static int decode_cabac_mb_intra4x4_pred_mode( H264Context *h, int pred_mode ) {
4595 if( get_cabac( &h->cabac, &h->cabac_state[68] ) )
4598 if( get_cabac( &h->cabac, &h->cabac_state[69] ) )
4600 if( get_cabac( &h->cabac, &h->cabac_state[69] ) )
4602 if( get_cabac( &h->cabac, &h->cabac_state[69] ) )
4604 if( mode >= pred_mode )
4610 static int decode_cabac_mb_chroma_pre_mode( H264Context *h) {
4611 MpegEncContext * const s = &h->s;
4612 const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
4613 const int mba_xy = mb_xy - 1;
4614 const int mbb_xy = mb_xy - s->mb_stride;
4618 /* No need to test for IS_INTRA4x4 and IS_INTRA16x16, as we set chroma_pred_mode_table to 0 */
4619 if( h->slice_table[mba_xy] == h->slice_num && h->chroma_pred_mode_table[mba_xy] != 0 )
4622 if( h->slice_table[mbb_xy] == h->slice_num && h->chroma_pred_mode_table[mbb_xy] != 0 )
4625 if( get_cabac( &h->cabac, &h->cabac_state[64+ctx] ) == 0 )
4628 if( get_cabac( &h->cabac, &h->cabac_state[64+3] ) == 0 )
4630 if( get_cabac( &h->cabac, &h->cabac_state[64+3] ) == 0 )
4636 static const uint8_t block_idx_x[16] = {
4637 0, 1, 0, 1, 2, 3, 2, 3, 0, 1, 0, 1, 2, 3, 2, 3
4639 static const uint8_t block_idx_y[16] = {
4640 0, 0, 1, 1, 0, 0, 1, 1, 2, 2, 3, 3, 2, 2, 3, 3
4642 static const uint8_t block_idx_xy[4][4] = {
4649 static int decode_cabac_mb_cbp_luma( H264Context *h) {
4650 MpegEncContext * const s = &h->s;
4651 const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
4656 h->cbp_table[mb_xy] = 0; /* FIXME aaahahahah beurk */
4658 for( i8x8 = 0; i8x8 < 4; i8x8++ ) {
4664 x = block_idx_x[4*i8x8];
4665 y = block_idx_y[4*i8x8];
4669 else if( s->mb_x > 0 ) {
4671 if (h->slice_table[mba_xy] != h->slice_num) {
4678 else if( s->mb_y > 0 ) {
4679 mbb_xy = mb_xy - s->mb_stride;
4680 if (h->slice_table[mbb_xy] != h->slice_num) {
4685 /* No need to test for skip as we put 0 for skip block */
4686 /* No need to test for IPCM as we put 1 for IPCM block */
4688 int i8x8a = block_idx_xy[(x-1)&0x03][y]/4;
4689 if( ((h->cbp_table[mba_xy] >> i8x8a)&0x01) == 0 )
4694 int i8x8b = block_idx_xy[x][(y-1)&0x03]/4;
4695 if( ((h->cbp_table[mbb_xy] >> i8x8b)&0x01) == 0 )
4699 if( get_cabac( &h->cabac, &h->cabac_state[73 + ctx] ) ) {
4701 h->cbp_table[mb_xy] = cbp; /* FIXME aaahahahah beurk */
4706 static int decode_cabac_mb_cbp_chroma( H264Context *h) {
4710 cbp_a = (h->left_cbp>>4)&0x03;
4711 cbp_b = (h-> top_cbp>>4)&0x03;
4714 if( cbp_a > 0 ) ctx++;
4715 if( cbp_b > 0 ) ctx += 2;
4716 if( get_cabac( &h->cabac, &h->cabac_state[77 + ctx] ) == 0 )
4720 if( cbp_a == 2 ) ctx++;
4721 if( cbp_b == 2 ) ctx += 2;
4722 return 1 + get_cabac( &h->cabac, &h->cabac_state[77 + ctx] );
4724 static int decode_cabac_mb_dqp( H264Context *h) {
4725 MpegEncContext * const s = &h->s;
4731 mbn_xy = s->mb_x + s->mb_y*s->mb_stride - 1;
4733 mbn_xy = s->mb_width - 1 + (s->mb_y-1)*s->mb_stride;
4735 if( h->last_qscale_diff != 0 && ( IS_INTRA16x16(s->current_picture.mb_type[mbn_xy] ) || (h->cbp_table[mbn_xy]&0x3f) ) )
4738 while( get_cabac( &h->cabac, &h->cabac_state[60 + ctx] ) ) {
4749 return -(val + 1)/2;
4751 static int decode_cabac_p_mb_sub_type( H264Context *h ) {
4752 if( get_cabac( &h->cabac, &h->cabac_state[21] ) )
4754 if( !get_cabac( &h->cabac, &h->cabac_state[22] ) )
4756 if( get_cabac( &h->cabac, &h->cabac_state[23] ) )
4760 static int decode_cabac_b_mb_sub_type( H264Context *h ) {
4762 if( !get_cabac( &h->cabac, &h->cabac_state[36] ) )
4763 return 0; /* B_Direct_8x8 */
4764 if( !get_cabac( &h->cabac, &h->cabac_state[37] ) )
4765 return 1 + get_cabac( &h->cabac, &h->cabac_state[39] ); /* B_L0_8x8, B_L1_8x8 */
4767 if( get_cabac( &h->cabac, &h->cabac_state[38] ) ) {
4768 if( get_cabac( &h->cabac, &h->cabac_state[39] ) )
4769 return 11 + get_cabac( &h->cabac, &h->cabac_state[39] ); /* B_L1_4x4, B_Bi_4x4 */
4772 type += 2*get_cabac( &h->cabac, &h->cabac_state[39] );
4773 type += get_cabac( &h->cabac, &h->cabac_state[39] );
4777 static int decode_cabac_mb_ref( H264Context *h, int list, int n ) {
4778 int refa = h->ref_cache[list][scan8[n] - 1];
4779 int refb = h->ref_cache[list][scan8[n] - 8];
4783 if( h->slice_type == B_TYPE) {
4784 if( refa > 0 && !h->direct_cache[scan8[n] - 1] )
4786 if( refb > 0 && !h->direct_cache[scan8[n] - 8] )
4795 while( get_cabac( &h->cabac, &h->cabac_state[54+ctx] ) ) {
4805 static int decode_cabac_mb_mvd( H264Context *h, int list, int n, int l ) {
4806 int amvd = abs( h->mvd_cache[list][scan8[n] - 1][l] ) +
4807 abs( h->mvd_cache[list][scan8[n] - 8][l] );
4808 int ctxbase = (l == 0) ? 40 : 47;
4813 else if( amvd > 32 )
4818 if(!get_cabac(&h->cabac, &h->cabac_state[ctxbase+ctx]))
4823 while( mvd < 9 && get_cabac( &h->cabac, &h->cabac_state[ctxbase+ctx] ) ) {
4831 while( get_cabac_bypass( &h->cabac ) ) {
4836 if( get_cabac_bypass( &h->cabac ) )
4840 if( get_cabac_bypass( &h->cabac ) ) return -mvd;
4844 static int inline get_cabac_cbf_ctx( H264Context *h, int cat, int idx ) {
4849 nza = h->left_cbp&0x100;
4850 nzb = h-> top_cbp&0x100;
4851 } else if( cat == 1 || cat == 2 ) {
4852 nza = h->non_zero_count_cache[scan8[idx] - 1];
4853 nzb = h->non_zero_count_cache[scan8[idx] - 8];
4854 } else if( cat == 3 ) {
4855 nza = (h->left_cbp>>(6+idx))&0x01;
4856 nzb = (h-> top_cbp>>(6+idx))&0x01;
4859 nza = h->non_zero_count_cache[scan8[16+idx] - 1];
4860 nzb = h->non_zero_count_cache[scan8[16+idx] - 8];
4869 return ctx + 4 * cat;
4872 static int inline decode_cabac_residual( H264Context *h, DCTELEM *block, int cat, int n, const uint8_t *scantable, int qp, int max_coeff) {
4873 const int mb_xy = h->s.mb_x + h->s.mb_y*h->s.mb_stride;
4874 const uint16_t *qmul= dequant_coeff[qp];
4875 static const int significant_coeff_flag_offset[5] = { 0, 15, 29, 44, 47 };
4876 static const int coeff_abs_level_m1_offset[5] = {227+ 0, 227+10, 227+20, 227+30, 227+39 };
4881 int coeff_count = 0;
4884 int abslevelgt1 = 0;
4886 /* cat: 0-> DC 16x16 n = 0
4887 * 1-> AC 16x16 n = luma4x4idx
4888 * 2-> Luma4x4 n = luma4x4idx
4889 * 3-> DC Chroma n = iCbCr
4890 * 4-> AC Chroma n = 4 * iCbCr + chroma4x4idx
4893 /* read coded block flag */
4894 if( get_cabac( &h->cabac, &h->cabac_state[85 + get_cabac_cbf_ctx( h, cat, n ) ] ) == 0 ) {
4895 if( cat == 1 || cat == 2 )
4896 h->non_zero_count_cache[scan8[n]] = 0;
4898 h->non_zero_count_cache[scan8[16+n]] = 0;
4903 for(last= 0; last < max_coeff - 1; last++) {
4904 if( get_cabac( &h->cabac, &h->cabac_state[105+significant_coeff_flag_offset[cat]+last] )) {
4905 index[coeff_count++] = last;
4906 if( get_cabac( &h->cabac, &h->cabac_state[166+significant_coeff_flag_offset[cat]+last] ) ) {
4912 if( last == max_coeff -1 ) {
4913 index[coeff_count++] = last;
4915 assert(coeff_count > 0);
4918 h->cbp_table[mb_xy] |= 0x100;
4919 else if( cat == 1 || cat == 2 )
4920 h->non_zero_count_cache[scan8[n]] = coeff_count;
4922 h->cbp_table[mb_xy] |= 0x40 << n;
4925 h->non_zero_count_cache[scan8[16+n]] = coeff_count;
4928 for( i = coeff_count - 1; i >= 0; i-- ) {
4929 int ctx = (abslevelgt1 != 0 ? 0 : FFMIN( 4, abslevel1 )) + coeff_abs_level_m1_offset[cat];
4930 int j= scantable[index[i]];
4932 if( get_cabac( &h->cabac, &h->cabac_state[ctx] ) == 0 ) {
4933 if( cat == 0 || cat == 3 ) {
4934 if( get_cabac_bypass( &h->cabac ) ) block[j] = -1;
4937 if( get_cabac_bypass( &h->cabac ) ) block[j] = -qmul[j];
4938 else block[j] = qmul[j];
4944 ctx = 5 + FFMIN( 4, abslevelgt1 ) + coeff_abs_level_m1_offset[cat];
4945 while( coeff_abs < 15 && get_cabac( &h->cabac, &h->cabac_state[ctx] ) ) {
4949 if( coeff_abs >= 15 ) {
4951 while( get_cabac_bypass( &h->cabac ) ) {
4952 coeff_abs += 1 << j;
4957 if( get_cabac_bypass( &h->cabac ) )
4958 coeff_abs += 1 << j ;
4962 if( cat == 0 || cat == 3 ) {
4963 if( get_cabac_bypass( &h->cabac ) ) block[j] = -coeff_abs;
4964 else block[j] = coeff_abs;
4966 if( get_cabac_bypass( &h->cabac ) ) block[j] = -coeff_abs * qmul[j];
4967 else block[j] = coeff_abs * qmul[j];
4977 * decodes a macroblock
4978 * @returns 0 if ok, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
4980 static int decode_mb_cabac(H264Context *h) {
4981 MpegEncContext * const s = &h->s;
4982 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
4983 int mb_type, partition_count, cbp = 0;
4985 s->dsp.clear_blocks(h->mb); //FIXME avoid if allready clear (move after skip handlong?)
4987 if( h->sps.mb_aff ) {
4988 av_log( h->s.avctx, AV_LOG_ERROR, "Fields not supported with CABAC\n" );
4992 tprintf("pic:%d mb:%d/%d\n", h->frame_num, s->mb_x, s->mb_y);
4993 if( h->slice_type != I_TYPE && h->slice_type != SI_TYPE ) {
4994 /* read skip flags */
4995 if( decode_cabac_mb_skip( h ) ) {
4998 h->cbp_table[mb_xy] = 0;
4999 h->chroma_pred_mode_table[mb_xy] = 0;
5000 h->last_qscale_diff = 0;
5006 h->prev_mb_skiped = 0;
5008 if( ( mb_type = decode_cabac_mb_type( h ) ) < 0 ) {
5009 av_log( h->s.avctx, AV_LOG_ERROR, "decode_cabac_mb_type failed\n" );
5013 if( h->slice_type == B_TYPE ) {
5015 partition_count= b_mb_type_info[mb_type].partition_count;
5016 mb_type= b_mb_type_info[mb_type].type;
5019 goto decode_intra_mb;
5021 } else if( h->slice_type == P_TYPE ) {
5023 partition_count= p_mb_type_info[mb_type].partition_count;
5024 mb_type= p_mb_type_info[mb_type].type;
5027 goto decode_intra_mb;
5030 assert(h->slice_type == I_TYPE);
5032 partition_count = 0;
5033 cbp= i_mb_type_info[mb_type].cbp;
5034 h->intra16x16_pred_mode= i_mb_type_info[mb_type].pred_mode;
5035 mb_type= i_mb_type_info[mb_type].type;
5038 if(h->mb_field_decoding_flag)
5039 mb_type |= MB_TYPE_INTERLACED;
5042 s->current_picture.mb_type[mb_xy]= mb_type;
5043 h->slice_table[ mb_xy ]= h->slice_num;
5045 if(IS_INTRA_PCM(mb_type)) {
5049 // We assume these blocks are very rare so we dont optimize it.
5050 // FIXME The two following lines get the bitstream position in the cabac
5051 // decode, I think it should be done by a function in cabac.h (or cabac.c).
5052 ptr= h->cabac.bytestream;
5053 if (h->cabac.low&0x1) ptr-=CABAC_BITS/8;
5055 // The pixels are stored in the same order as levels in h->mb array.
5056 for(y=0; y<16; y++){
5057 const int index= 4*(y&3) + 32*((y>>2)&1) + 128*(y>>3);
5058 for(x=0; x<16; x++){
5059 tprintf("LUMA ICPM LEVEL (%3d)\n", *ptr);
5060 h->mb[index + (x&3) + 16*((x>>2)&1) + 64*(x>>3)]= *ptr++;
5064 const int index= 256 + 4*(y&3) + 32*(y>>2);
5066 tprintf("CHROMA U ICPM LEVEL (%3d)\n", *ptr);
5067 h->mb[index + (x&3) + 16*(x>>2)]= *ptr++;
5071 const int index= 256 + 64 + 4*(y&3) + 32*(y>>2);
5073 tprintf("CHROMA V ICPM LEVEL (%3d)\n", *ptr);
5074 h->mb[index + (x&3) + 16*(x>>2)]= *ptr++;
5078 ff_init_cabac_decoder(&h->cabac, ptr, h->cabac.bytestream_end - ptr);
5080 // All blocks are presents
5081 h->cbp_table[mb_xy] = 0x1ef;
5082 h->chroma_pred_mode_table[mb_xy] = 0;
5083 // In deblocking, the quantiser is 0
5084 s->current_picture.qscale_table[mb_xy]= 0;
5085 h->chroma_qp = get_chroma_qp(h, 0);
5086 // All coeffs are presents
5087 memset(h->non_zero_count[mb_xy], 16, 16);
5091 fill_caches(h, mb_type, 0);
5093 if( IS_INTRA( mb_type ) ) {
5094 if( IS_INTRA4x4( mb_type ) ) {
5096 for( i = 0; i < 16; i++ ) {
5097 int pred = pred_intra_mode( h, i );
5098 h->intra4x4_pred_mode_cache[ scan8[i] ] = decode_cabac_mb_intra4x4_pred_mode( h, pred );
5100 //av_log( s->avctx, AV_LOG_ERROR, "i4x4 pred=%d mode=%d\n", pred, h->intra4x4_pred_mode_cache[ scan8[i] ] );
5102 write_back_intra_pred_mode(h);
5103 if( check_intra4x4_pred_mode(h) < 0 ) return -1;
5105 h->intra16x16_pred_mode= check_intra_pred_mode( h, h->intra16x16_pred_mode );
5106 if( h->intra16x16_pred_mode < 0 ) return -1;
5108 h->chroma_pred_mode_table[mb_xy] =
5109 h->chroma_pred_mode = decode_cabac_mb_chroma_pre_mode( h );
5111 h->chroma_pred_mode= check_intra_pred_mode( h, h->chroma_pred_mode );
5112 if( h->chroma_pred_mode < 0 ) return -1;
5113 } else if( partition_count == 4 ) {
5114 int i, j, sub_partition_count[4], list, ref[2][4];
5116 if( h->slice_type == B_TYPE ) {
5117 for( i = 0; i < 4; i++ ) {
5118 h->sub_mb_type[i] = decode_cabac_b_mb_sub_type( h );
5119 sub_partition_count[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
5120 h->sub_mb_type[i]= b_sub_mb_type_info[ h->sub_mb_type[i] ].type;
5122 if( IS_DIRECT(h->sub_mb_type[0]) || IS_DIRECT(h->sub_mb_type[1])
5123 || IS_DIRECT(h->sub_mb_type[2]) || IS_DIRECT(h->sub_mb_type[3])) {
5124 pred_direct_motion(h, &mb_type);
5125 if( h->ref_count[0] > 1 || h->ref_count[1] > 1 ) {
5126 for( i = 0; i < 4; i++ )
5127 if( IS_DIRECT(h->sub_mb_type[i]) )
5128 fill_rectangle( &h->direct_cache[scan8[4*i]], 2, 2, 8, 1, 1 );
5132 for( i = 0; i < 4; i++ ) {
5133 h->sub_mb_type[i] = decode_cabac_p_mb_sub_type( h );
5134 sub_partition_count[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].partition_count;
5135 h->sub_mb_type[i]= p_sub_mb_type_info[ h->sub_mb_type[i] ].type;
5139 for( list = 0; list < 2; list++ ) {
5140 if( h->ref_count[list] > 0 ) {
5141 for( i = 0; i < 4; i++ ) {
5142 if(IS_DIRECT(h->sub_mb_type[i])) continue;
5143 if(IS_DIR(h->sub_mb_type[i], 0, list)){
5144 if( h->ref_count[list] > 1 )
5145 ref[list][i] = decode_cabac_mb_ref( h, list, 4*i );
5151 h->ref_cache[list][ scan8[4*i]+1 ]=
5152 h->ref_cache[list][ scan8[4*i]+8 ]=h->ref_cache[list][ scan8[4*i]+9 ]= ref[list][i];
5157 for(list=0; list<2; list++){
5159 if(IS_DIRECT(h->sub_mb_type[i])){
5160 fill_rectangle(h->mvd_cache[list][scan8[4*i]], 2, 2, 8, 0, 4);
5163 h->ref_cache[list][ scan8[4*i] ]=h->ref_cache[list][ scan8[4*i]+1 ];
5165 if(IS_DIR(h->sub_mb_type[i], 0, list) && !IS_DIRECT(h->sub_mb_type[i])){
5166 const int sub_mb_type= h->sub_mb_type[i];
5167 const int block_width= (sub_mb_type & (MB_TYPE_16x16|MB_TYPE_16x8)) ? 2 : 1;
5168 for(j=0; j<sub_partition_count[i]; j++){
5171 const int index= 4*i + block_width*j;
5172 int16_t (* mv_cache)[2]= &h->mv_cache[list][ scan8[index] ];
5173 int16_t (* mvd_cache)[2]= &h->mvd_cache[list][ scan8[index] ];
5174 pred_motion(h, index, block_width, list, h->ref_cache[list][ scan8[index] ], &mpx, &mpy);
5176 mx = mpx + decode_cabac_mb_mvd( h, list, index, 0 );
5177 my = mpy + decode_cabac_mb_mvd( h, list, index, 1 );
5178 tprintf("final mv:%d %d\n", mx, my);
5180 if(IS_SUB_8X8(sub_mb_type)){
5181 mv_cache[ 0 ][0]= mv_cache[ 1 ][0]=
5182 mv_cache[ 8 ][0]= mv_cache[ 9 ][0]= mx;
5183 mv_cache[ 0 ][1]= mv_cache[ 1 ][1]=
5184 mv_cache[ 8 ][1]= mv_cache[ 9 ][1]= my;
5186 mvd_cache[ 0 ][0]= mvd_cache[ 1 ][0]=
5187 mvd_cache[ 8 ][0]= mvd_cache[ 9 ][0]= mx - mpx;
5188 mvd_cache[ 0 ][1]= mvd_cache[ 1 ][1]=
5189 mvd_cache[ 8 ][1]= mvd_cache[ 9 ][1]= my - mpy;
5190 }else if(IS_SUB_8X4(sub_mb_type)){
5191 mv_cache[ 0 ][0]= mv_cache[ 1 ][0]= mx;
5192 mv_cache[ 0 ][1]= mv_cache[ 1 ][1]= my;
5194 mvd_cache[ 0 ][0]= mvd_cache[ 1 ][0]= mx- mpx;
5195 mvd_cache[ 0 ][1]= mvd_cache[ 1 ][1]= my - mpy;
5196 }else if(IS_SUB_4X8(sub_mb_type)){
5197 mv_cache[ 0 ][0]= mv_cache[ 8 ][0]= mx;
5198 mv_cache[ 0 ][1]= mv_cache[ 8 ][1]= my;
5200 mvd_cache[ 0 ][0]= mvd_cache[ 8 ][0]= mx - mpx;
5201 mvd_cache[ 0 ][1]= mvd_cache[ 8 ][1]= my - mpy;
5203 assert(IS_SUB_4X4(sub_mb_type));
5204 mv_cache[ 0 ][0]= mx;
5205 mv_cache[ 0 ][1]= my;
5207 mvd_cache[ 0 ][0]= mx - mpx;
5208 mvd_cache[ 0 ][1]= my - mpy;
5212 uint32_t *p= (uint32_t *)&h->mv_cache[list][ scan8[4*i] ][0];
5213 uint32_t *pd= (uint32_t *)&h->mvd_cache[list][ scan8[4*i] ][0];
5214 p[0] = p[1] = p[8] = p[9] = 0;
5215 pd[0]= pd[1]= pd[8]= pd[9]= 0;
5219 } else if( IS_DIRECT(mb_type) ) {
5220 pred_direct_motion(h, &mb_type);
5221 s->current_picture.mb_type[mb_xy]= mb_type;
5222 fill_rectangle(h->mvd_cache[0][scan8[0]], 4, 4, 8, 0, 4);
5223 fill_rectangle(h->mvd_cache[1][scan8[0]], 4, 4, 8, 0, 4);
5225 int list, mx, my, i, mpx, mpy;
5226 if(IS_16X16(mb_type)){
5227 for(list=0; list<2; list++){
5228 if(IS_DIR(mb_type, 0, list)){
5229 if(h->ref_count[list] > 0 ){
5230 const int ref = h->ref_count[list] > 1 ? decode_cabac_mb_ref( h, list, 0 ) : 0;
5231 fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, ref, 1);
5234 fill_rectangle(&h->ref_cache[list][ scan8[0] ], 4, 4, 8, (uint8_t)LIST_NOT_USED, 1);
5236 for(list=0; list<2; list++){
5237 if(IS_DIR(mb_type, 0, list)){
5238 pred_motion(h, 0, 4, list, h->ref_cache[list][ scan8[0] ], &mpx, &mpy);
5240 mx = mpx + decode_cabac_mb_mvd( h, list, 0, 0 );
5241 my = mpy + decode_cabac_mb_mvd( h, list, 0, 1 );
5242 tprintf("final mv:%d %d\n", mx, my);
5244 fill_rectangle(h->mvd_cache[list][ scan8[0] ], 4, 4, 8, pack16to32(mx-mpx,my-mpy), 4);
5245 fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, pack16to32(mx,my), 4);
5247 fill_rectangle(h->mv_cache[list][ scan8[0] ], 4, 4, 8, 0, 4);
5250 else if(IS_16X8(mb_type)){
5251 for(list=0; list<2; list++){
5252 if(h->ref_count[list]>0){
5254 if(IS_DIR(mb_type, i, list)){
5255 const int ref= h->ref_count[list] > 1 ? decode_cabac_mb_ref( h, list, 8*i ) : 0;
5256 fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, ref, 1);
5258 fill_rectangle(&h->ref_cache[list][ scan8[0] + 16*i ], 4, 2, 8, (LIST_NOT_USED&0xFF), 1);
5262 for(list=0; list<2; list++){
5264 if(IS_DIR(mb_type, i, list)){
5265 pred_16x8_motion(h, 8*i, list, h->ref_cache[list][scan8[0] + 16*i], &mpx, &mpy);
5266 mx = mpx + decode_cabac_mb_mvd( h, list, 8*i, 0 );
5267 my = mpy + decode_cabac_mb_mvd( h, list, 8*i, 1 );
5268 tprintf("final mv:%d %d\n", mx, my);
5270 fill_rectangle(h->mvd_cache[list][ scan8[0] + 16*i ], 4, 2, 8, pack16to32(mx-mpx,my-mpy), 4);
5271 fill_rectangle(h->mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, pack16to32(mx,my), 4);
5273 fill_rectangle(h->mvd_cache[list][ scan8[0] + 16*i ], 4, 2, 8, 0, 4);
5274 fill_rectangle(h-> mv_cache[list][ scan8[0] + 16*i ], 4, 2, 8, 0, 4);
5279 assert(IS_8X16(mb_type));
5280 for(list=0; list<2; list++){
5281 if(h->ref_count[list]>0){
5283 if(IS_DIR(mb_type, i, list)){ //FIXME optimize
5284 const int ref= h->ref_count[list] > 1 ? decode_cabac_mb_ref( h, list, 4*i ) : 0;
5285 fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, ref, 1);
5287 fill_rectangle(&h->ref_cache[list][ scan8[0] + 2*i ], 2, 4, 8, (LIST_NOT_USED&0xFF), 1);
5291 for(list=0; list<2; list++){
5293 if(IS_DIR(mb_type, i, list)){
5294 pred_8x16_motion(h, i*4, list, h->ref_cache[list][ scan8[0] + 2*i ], &mpx, &mpy);
5295 mx = mpx + decode_cabac_mb_mvd( h, list, 4*i, 0 );
5296 my = mpy + decode_cabac_mb_mvd( h, list, 4*i, 1 );
5298 tprintf("final mv:%d %d\n", mx, my);
5299 fill_rectangle(h->mvd_cache[list][ scan8[0] + 2*i ], 2, 4, 8, pack16to32(mx-mpx,my-mpy), 4);
5300 fill_rectangle(h->mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, pack16to32(mx,my), 4);
5302 fill_rectangle(h->mvd_cache[list][ scan8[0] + 2*i ], 2, 4, 8, 0, 4);
5303 fill_rectangle(h-> mv_cache[list][ scan8[0] + 2*i ], 2, 4, 8, 0, 4);
5310 if( IS_INTER( mb_type ) ) {
5311 h->chroma_pred_mode_table[mb_xy] = 0;
5312 write_back_motion( h, mb_type );
5315 if( !IS_INTRA16x16( mb_type ) ) {
5316 cbp = decode_cabac_mb_cbp_luma( h );
5317 cbp |= decode_cabac_mb_cbp_chroma( h ) << 4;
5320 h->cbp_table[mb_xy] = cbp;
5322 if( cbp || IS_INTRA16x16( mb_type ) ) {
5323 const uint8_t *scan, *dc_scan;
5326 if(IS_INTERLACED(mb_type)){
5328 dc_scan= luma_dc_field_scan;
5331 dc_scan= luma_dc_zigzag_scan;
5334 h->last_qscale_diff = dqp = decode_cabac_mb_dqp( h );
5336 if(((unsigned)s->qscale) > 51){
5337 if(s->qscale<0) s->qscale+= 52;
5338 else s->qscale-= 52;
5340 h->chroma_qp = get_chroma_qp(h, s->qscale);
5342 if( IS_INTRA16x16( mb_type ) ) {
5344 //av_log( s->avctx, AV_LOG_ERROR, "INTRA16x16 DC\n" );
5345 if( decode_cabac_residual( h, h->mb, 0, 0, dc_scan, s->qscale, 16) < 0)
5348 for( i = 0; i < 16; i++ ) {
5349 //av_log( s->avctx, AV_LOG_ERROR, "INTRA16x16 AC:%d\n", i );
5350 if( decode_cabac_residual(h, h->mb + 16*i, 1, i, scan + 1, s->qscale, 15) < 0 )
5354 fill_rectangle(&h->non_zero_count_cache[scan8[0]], 4, 4, 8, 0, 1);
5358 for( i8x8 = 0; i8x8 < 4; i8x8++ ) {
5359 if( cbp & (1<<i8x8) ) {
5360 for( i4x4 = 0; i4x4 < 4; i4x4++ ) {
5361 const int index = 4*i8x8 + i4x4;
5362 //av_log( s->avctx, AV_LOG_ERROR, "Luma4x4: %d\n", index );
5363 if( decode_cabac_residual(h, h->mb + 16*index, 2, index, scan, s->qscale, 16) < 0 )
5367 uint8_t * const nnz= &h->non_zero_count_cache[ scan8[4*i8x8] ];
5368 nnz[0] = nnz[1] = nnz[8] = nnz[9] = 0;
5375 for( c = 0; c < 2; c++ ) {
5376 //av_log( s->avctx, AV_LOG_ERROR, "INTRA C%d-DC\n",c );
5377 if( decode_cabac_residual(h, h->mb + 256 + 16*4*c, 3, c, chroma_dc_scan, h->chroma_qp, 4) < 0)
5384 for( c = 0; c < 2; c++ ) {
5385 for( i = 0; i < 4; i++ ) {
5386 const int index = 16 + 4 * c + i;
5387 //av_log( s->avctx, AV_LOG_ERROR, "INTRA C%d-AC %d\n",c, index - 16 );
5388 if( decode_cabac_residual(h, h->mb + 16*index, 4, index - 16, scan + 1, h->chroma_qp, 15) < 0)
5393 uint8_t * const nnz= &h->non_zero_count_cache[0];
5394 nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =
5395 nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
5398 uint8_t * const nnz= &h->non_zero_count_cache[0];
5399 fill_rectangle(&nnz[scan8[0]], 4, 4, 8, 0, 1);
5400 nnz[ scan8[16]+0 ] = nnz[ scan8[16]+1 ] =nnz[ scan8[16]+8 ] =nnz[ scan8[16]+9 ] =
5401 nnz[ scan8[20]+0 ] = nnz[ scan8[20]+1 ] =nnz[ scan8[20]+8 ] =nnz[ scan8[20]+9 ] = 0;
5404 s->current_picture.qscale_table[mb_xy]= s->qscale;
5405 write_back_non_zero_count(h);
5411 static void filter_mb_edgev( H264Context *h, uint8_t *pix, int stride, int bS[4], int qp ) {
5413 const int index_a = clip( qp + h->slice_alpha_c0_offset, 0, 51 );
5414 const int alpha = alpha_table[index_a];
5415 const int beta = beta_table[clip( qp + h->slice_beta_offset, 0, 51 )];
5417 for( i = 0; i < 4; i++ ) {
5424 const int tc0 = tc0_table[index_a][bS[i] - 1];
5425 /* 4px edge length */
5426 for( d = 0; d < 4; d++ ) {
5427 const int p0 = pix[-1];
5428 const int p1 = pix[-2];
5429 const int p2 = pix[-3];
5430 const int q0 = pix[0];
5431 const int q1 = pix[1];
5432 const int q2 = pix[2];
5434 if( ABS( p0 - q0 ) < alpha &&
5435 ABS( p1 - p0 ) < beta &&
5436 ABS( q1 - q0 ) < beta ) {
5440 if( ABS( p2 - p0 ) < beta ) {
5441 pix[-2] = p1 + clip( ( p2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( p1 << 1 ) ) >> 1, -tc0, tc0 );
5444 if( ABS( q2 - q0 ) < beta ) {
5445 pix[1] = q1 + clip( ( q2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( q1 << 1 ) ) >> 1, -tc0, tc0 );
5449 i_delta = clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );
5450 pix[-1] = clip_uint8( p0 + i_delta ); /* p0' */
5451 pix[0] = clip_uint8( q0 - i_delta ); /* q0' */
5456 /* 4px edge length */
5457 for( d = 0; d < 4; d++ ) {
5458 const int p0 = pix[-1];
5459 const int p1 = pix[-2];
5460 const int p2 = pix[-3];
5462 const int q0 = pix[0];
5463 const int q1 = pix[1];
5464 const int q2 = pix[2];
5466 if( ABS( p0 - q0 ) < alpha &&
5467 ABS( p1 - p0 ) < beta &&
5468 ABS( q1 - q0 ) < beta ) {
5470 if(ABS( p0 - q0 ) < (( alpha >> 2 ) + 2 )){
5471 if( ABS( p2 - p0 ) < beta)
5473 const int p3 = pix[-4];
5475 pix[-1] = ( p2 + 2*p1 + 2*p0 + 2*q0 + q1 + 4 ) >> 3;
5476 pix[-2] = ( p2 + p1 + p0 + q0 + 2 ) >> 2;
5477 pix[-3] = ( 2*p3 + 3*p2 + p1 + p0 + q0 + 4 ) >> 3;
5480 pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
5482 if( ABS( q2 - q0 ) < beta)
5484 const int q3 = pix[3];
5486 pix[0] = ( p1 + 2*p0 + 2*q0 + 2*q1 + q2 + 4 ) >> 3;
5487 pix[1] = ( p0 + q0 + q1 + q2 + 2 ) >> 2;
5488 pix[2] = ( 2*q3 + 3*q2 + q1 + q0 + p0 + 4 ) >> 3;
5491 pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
5495 pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
5496 pix[ 0] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
5504 static void filter_mb_edgecv( H264Context *h, uint8_t *pix, int stride, int bS[4], int qp ) {
5506 const int index_a = clip( qp + h->slice_alpha_c0_offset, 0, 51 );
5507 const int alpha = alpha_table[index_a];
5508 const int beta = beta_table[clip( qp + h->slice_beta_offset, 0, 51 )];
5510 for( i = 0; i < 4; i++ ) {
5517 const int tc = tc0_table[index_a][bS[i] - 1] + 1;
5518 /* 2px edge length (because we use same bS than the one for luma) */
5519 for( d = 0; d < 2; d++ ){
5520 const int p0 = pix[-1];
5521 const int p1 = pix[-2];
5522 const int q0 = pix[0];
5523 const int q1 = pix[1];
5525 if( ABS( p0 - q0 ) < alpha &&
5526 ABS( p1 - p0 ) < beta &&
5527 ABS( q1 - q0 ) < beta ) {
5528 const int i_delta = clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );
5530 pix[-1] = clip_uint8( p0 + i_delta ); /* p0' */
5531 pix[0] = clip_uint8( q0 - i_delta ); /* q0' */
5532 //tprintf("filter_mb_edgecv i:%d d:%d, qp:%d, indexA:%d, alpha:%d, beta:%d, tc:%d\n# bS:%d -> [%02x, %02x, %02x, %02x, %02x, %02x] =>[%02x, %02x, %02x, %02x]\n", i, d, qp, index_a, alpha, beta, tc, bS[i], pix[-3], p1, p0, q0, q1, pix[2], p1, pix[-1], pix[0], q1);
5537 /* 2px edge length (because we use same bS than the one for luma) */
5538 for( d = 0; d < 2; d++ ){
5539 const int p0 = pix[-1];
5540 const int p1 = pix[-2];
5541 const int q0 = pix[0];
5542 const int q1 = pix[1];
5544 if( ABS( p0 - q0 ) < alpha &&
5545 ABS( p1 - p0 ) < beta &&
5546 ABS( q1 - q0 ) < beta ) {
5548 pix[-1] = ( 2*p1 + p0 + q1 + 2 ) >> 2; /* p0' */
5549 pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2; /* q0' */
5550 //tprintf("filter_mb_edgecv i:%d d:%d\n# bS:4 -> [%02x, %02x, %02x, %02x, %02x, %02x] =>[%02x, %02x, %02x, %02x]\n", i, d, pix[-3], p1, p0, q0, q1, pix[2], p1, pix[-1], pix[0], q1);
5558 static void filter_mb_edgeh( H264Context *h, uint8_t *pix, int stride, int bS[4], int qp ) {
5560 const int index_a = clip( qp + h->slice_alpha_c0_offset, 0, 51 );
5561 const int alpha = alpha_table[index_a];
5562 const int beta = beta_table[clip( qp + h->slice_beta_offset, 0, 51 )];
5563 const int pix_next = stride;
5565 for( i = 0; i < 4; i++ ) {
5572 const int tc0 = tc0_table[index_a][bS[i] - 1];
5573 /* 4px edge length */
5574 for( d = 0; d < 4; d++ ) {
5575 const int p0 = pix[-1*pix_next];
5576 const int p1 = pix[-2*pix_next];
5577 const int p2 = pix[-3*pix_next];
5578 const int q0 = pix[0];
5579 const int q1 = pix[1*pix_next];
5580 const int q2 = pix[2*pix_next];
5582 if( ABS( p0 - q0 ) < alpha &&
5583 ABS( p1 - p0 ) < beta &&
5584 ABS( q1 - q0 ) < beta ) {
5589 if( ABS( p2 - p0 ) < beta ) {
5590 pix[-2*pix_next] = p1 + clip( ( p2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( p1 << 1 ) ) >> 1, -tc0, tc0 );
5593 if( ABS( q2 - q0 ) < beta ) {
5594 pix[pix_next] = q1 + clip( ( q2 + ( ( p0 + q0 + 1 ) >> 1 ) - ( q1 << 1 ) ) >> 1, -tc0, tc0 );
5598 i_delta = clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );
5599 pix[-pix_next] = clip_uint8( p0 + i_delta ); /* p0' */
5600 pix[0] = clip_uint8( q0 - i_delta ); /* q0' */
5605 /* 4px edge length */
5606 for( d = 0; d < 4; d++ ) {
5607 const int p0 = pix[-1*pix_next];
5608 const int p1 = pix[-2*pix_next];
5609 const int p2 = pix[-3*pix_next];
5610 const int q0 = pix[0];
5611 const int q1 = pix[1*pix_next];
5612 const int q2 = pix[2*pix_next];
5614 if( ABS( p0 - q0 ) < alpha &&
5615 ABS( p1 - p0 ) < beta &&
5616 ABS( q1 - q0 ) < beta ) {
5618 const int p3 = pix[-4*pix_next];
5619 const int q3 = pix[ 3*pix_next];
5621 if(ABS( p0 - q0 ) < (( alpha >> 2 ) + 2 )){
5622 if( ABS( p2 - p0 ) < beta) {
5624 pix[-1*pix_next] = ( p2 + 2*p1 + 2*p0 + 2*q0 + q1 + 4 ) >> 3;
5625 pix[-2*pix_next] = ( p2 + p1 + p0 + q0 + 2 ) >> 2;
5626 pix[-3*pix_next] = ( 2*p3 + 3*p2 + p1 + p0 + q0 + 4 ) >> 3;
5629 pix[-1*pix_next] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
5631 if( ABS( q2 - q0 ) < beta) {
5633 pix[0*pix_next] = ( p1 + 2*p0 + 2*q0 + 2*q1 + q2 + 4 ) >> 3;
5634 pix[1*pix_next] = ( p0 + q0 + q1 + q2 + 2 ) >> 2;
5635 pix[2*pix_next] = ( 2*q3 + 3*q2 + q1 + q0 + p0 + 4 ) >> 3;
5638 pix[0*pix_next] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
5642 pix[-1*pix_next] = ( 2*p1 + p0 + q1 + 2 ) >> 2;
5643 pix[ 0*pix_next] = ( 2*q1 + q0 + p1 + 2 ) >> 2;
5652 static void filter_mb_edgech( H264Context *h, uint8_t *pix, int stride, int bS[4], int qp ) {
5654 const int index_a = clip( qp + h->slice_alpha_c0_offset, 0, 51 );
5655 const int alpha = alpha_table[index_a];
5656 const int beta = beta_table[clip( qp + h->slice_beta_offset, 0, 51 )];
5657 const int pix_next = stride;
5659 for( i = 0; i < 4; i++ )
5667 int tc = tc0_table[index_a][bS[i] - 1] + 1;
5668 /* 2px edge length (see deblocking_filter_edgecv) */
5669 for( d = 0; d < 2; d++ ) {
5670 const int p0 = pix[-1*pix_next];
5671 const int p1 = pix[-2*pix_next];
5672 const int q0 = pix[0];
5673 const int q1 = pix[1*pix_next];
5675 if( ABS( p0 - q0 ) < alpha &&
5676 ABS( p1 - p0 ) < beta &&
5677 ABS( q1 - q0 ) < beta ) {
5679 int i_delta = clip( (((q0 - p0 ) << 2) + (p1 - q1) + 4) >> 3, -tc, tc );
5681 pix[-pix_next] = clip_uint8( p0 + i_delta ); /* p0' */
5682 pix[0] = clip_uint8( q0 - i_delta ); /* q0' */
5687 /* 2px edge length (see deblocking_filter_edgecv) */
5688 for( d = 0; d < 2; d++ ) {
5689 const int p0 = pix[-1*pix_next];
5690 const int p1 = pix[-2*pix_next];
5691 const int q0 = pix[0];
5692 const int q1 = pix[1*pix_next];
5694 if( ABS( p0 - q0 ) < alpha &&
5695 ABS( p1 - p0 ) < beta &&
5696 ABS( q1 - q0 ) < beta ) {
5698 pix[-pix_next] = ( 2*p1 + p0 + q1 + 2 ) >> 2; /* p0' */
5699 pix[0] = ( 2*q1 + q0 + p1 + 2 ) >> 2; /* q0' */
5707 static void filter_mb( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr) {
5708 MpegEncContext * const s = &h->s;
5709 const int mb_xy= mb_x + mb_y*s->mb_stride;
5710 int linesize, uvlinesize;
5713 /* FIXME Implement deblocking filter for field MB */
5714 if( h->sps.mb_aff ) {
5717 linesize = s->linesize;
5718 uvlinesize = s->uvlinesize;
5720 /* dir : 0 -> vertical edge, 1 -> horizontal edge */
5721 for( dir = 0; dir < 2; dir++ )
5724 const int mbm_xy = dir == 0 ? mb_xy -1 : mb_xy - s->mb_stride;
5725 int start = h->slice_table[mbm_xy] == 255 ? 1 : 0;
5727 if (h->deblocking_filter==2 && h->slice_table[mbm_xy] != h->slice_table[mb_xy])
5731 for( edge = start; edge < 4; edge++ ) {
5732 /* mbn_xy: neighbour macroblock (how that works for field ?) */
5733 int mbn_xy = edge > 0 ? mb_xy : mbm_xy;
5737 if( IS_INTRA( s->current_picture.mb_type[mb_xy] ) ||
5738 IS_INTRA( s->current_picture.mb_type[mbn_xy] ) ) {
5739 bS[0] = bS[1] = bS[2] = bS[3] = ( edge == 0 ? 4 : 3 );
5742 for( i = 0; i < 4; i++ ) {
5743 int x = dir == 0 ? edge : i;
5744 int y = dir == 0 ? i : edge;
5745 int b_idx= 8 + 4 + x + 8*y;
5746 int bn_idx= b_idx - (dir ? 8:1);
5748 if( h->non_zero_count_cache[b_idx] != 0 ||
5749 h->non_zero_count_cache[bn_idx] != 0 ) {
5754 /* FIXME: A given frame may occupy more than one position in
5755 * the reference list. So we should compare the frame numbers,
5756 * not the indices in the ref list. */
5759 for( l = 0; l < 1 + (h->slice_type == B_TYPE); l++ ) {
5760 if( h->ref_cache[l][b_idx] != h->ref_cache[l][bn_idx] ||
5761 ABS( h->mv_cache[l][b_idx][0] - h->mv_cache[l][bn_idx][0] ) >= 4 ||
5762 ABS( h->mv_cache[l][b_idx][1] - h->mv_cache[l][bn_idx][1] ) >= 4 ) {
5770 if(bS[0]+bS[1]+bS[2]+bS[3] == 0)
5775 // Do not use s->qscale as luma quantiser because it has not the same
5776 // value in IPCM macroblocks.
5777 qp = ( s->current_picture.qscale_table[mb_xy] + s->current_picture.qscale_table[mbn_xy] + 1 ) >> 1;
5778 //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]);
5780 filter_mb_edgev( h, &img_y[4*edge], linesize, bS, qp );
5781 if( (edge&1) == 0 ) {
5782 int chroma_qp = ( h->chroma_qp +
5783 get_chroma_qp( h, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1;
5784 filter_mb_edgecv( h, &img_cb[2*edge], uvlinesize, bS, chroma_qp );
5785 filter_mb_edgecv( h, &img_cr[2*edge], uvlinesize, bS, chroma_qp );
5788 filter_mb_edgeh( h, &img_y[4*edge*linesize], linesize, bS, qp );
5789 if( (edge&1) == 0 ) {
5790 int chroma_qp = ( h->chroma_qp +
5791 get_chroma_qp( h, s->current_picture.qscale_table[mbn_xy] ) + 1 ) >> 1;
5792 filter_mb_edgech( h, &img_cb[2*edge*uvlinesize], uvlinesize, bS, chroma_qp );
5793 filter_mb_edgech( h, &img_cr[2*edge*uvlinesize], uvlinesize, bS, chroma_qp );
5800 static int decode_slice(H264Context *h){
5801 MpegEncContext * const s = &h->s;
5802 const int part_mask= s->partitioned_frame ? (AC_END|AC_ERROR) : 0x7F;
5806 if( h->pps.cabac ) {
5810 align_get_bits( &s->gb );
5813 ff_init_cabac_states( &h->cabac, ff_h264_lps_range, ff_h264_mps_state, ff_h264_lps_state, 64 );
5814 ff_init_cabac_decoder( &h->cabac,
5815 s->gb.buffer + get_bits_count(&s->gb)/8,
5816 ( s->gb.size_in_bits - get_bits_count(&s->gb) + 7)/8);
5817 /* calculate pre-state */
5818 for( i= 0; i < 399; i++ ) {
5820 if( h->slice_type == I_TYPE )
5821 pre = clip( ((cabac_context_init_I[i][0] * s->qscale) >>4 ) + cabac_context_init_I[i][1], 1, 126 );
5823 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 );
5826 h->cabac_state[i] = 2 * ( 63 - pre ) + 0;
5828 h->cabac_state[i] = 2 * ( pre - 64 ) + 1;
5832 int ret = decode_mb_cabac(h);
5835 if(ret>=0) hl_decode_mb(h);
5837 /* XXX: useless as decode_mb_cabac it doesn't support that ... */
5838 if( ret >= 0 && h->sps.mb_aff ) { //FIXME optimal? or let mb_decode decode 16x32 ?
5841 if(ret>=0) ret = decode_mb_cabac(h);
5846 eos = get_cabac_terminate( &h->cabac );
5848 if( ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 1) {
5849 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
5850 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);
5854 if( ++s->mb_x >= s->mb_width ) {
5856 ff_draw_horiz_band(s, 16*s->mb_y, 16);
5858 if(h->sps.mb_aff && s->picture_structure==PICT_FRAME ) {
5863 if( eos || s->mb_y >= s->mb_height ) {
5864 tprintf("slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
5865 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);
5869 /* TODO test over-reading in cabac code */
5870 else if( read too much in h->cabac ) {
5871 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);
5879 int ret = decode_mb_cavlc(h);
5881 if(ret>=0) hl_decode_mb(h);
5883 if(ret>=0 && h->sps.mb_aff){ //FIXME optimal? or let mb_decode decode 16x32 ?
5885 ret = decode_mb_cavlc(h);
5887 if(ret>=0) hl_decode_mb(h);
5892 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
5893 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);
5898 if(++s->mb_x >= s->mb_width){
5900 ff_draw_horiz_band(s, 16*s->mb_y, 16);
5902 if(h->sps.mb_aff && s->picture_structure==PICT_FRAME ) {
5905 if(s->mb_y >= s->mb_height){
5906 tprintf("slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
5908 if(get_bits_count(&s->gb) == s->gb.size_in_bits ) {
5909 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);
5913 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);
5920 if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->mb_skip_run<=0){
5921 tprintf("slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
5922 if(get_bits_count(&s->gb) == s->gb.size_in_bits ){
5923 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);
5927 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);
5936 for(;s->mb_y < s->mb_height; s->mb_y++){
5937 for(;s->mb_x < s->mb_width; s->mb_x++){
5938 int ret= decode_mb(h);
5943 fprintf(stderr, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
5944 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);
5949 if(++s->mb_x >= s->mb_width){
5951 if(++s->mb_y >= s->mb_height){
5952 if(get_bits_count(s->gb) == s->gb.size_in_bits){
5953 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);
5957 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);
5964 if(get_bits_count(s->?gb) >= s->gb?.size_in_bits){
5965 if(get_bits_count(s->gb) == s->gb.size_in_bits){
5966 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);
5970 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);
5977 ff_draw_horiz_band(s, 16*s->mb_y, 16);
5980 return -1; //not reached
5983 static inline void decode_hrd_parameters(H264Context *h, SPS *sps){
5984 MpegEncContext * const s = &h->s;
5986 cpb_count = get_ue_golomb(&s->gb) + 1;
5987 get_bits(&s->gb, 4); /* bit_rate_scale */
5988 get_bits(&s->gb, 4); /* cpb_size_scale */
5989 for(i=0; i<cpb_count; i++){
5990 get_ue_golomb(&s->gb); /* bit_rate_value_minus1 */
5991 get_ue_golomb(&s->gb); /* cpb_size_value_minus1 */
5992 get_bits1(&s->gb); /* cbr_flag */
5994 get_bits(&s->gb, 5); /* initial_cpb_removal_delay_length_minus1 */
5995 get_bits(&s->gb, 5); /* cpb_removal_delay_length_minus1 */
5996 get_bits(&s->gb, 5); /* dpb_output_delay_length_minus1 */
5997 get_bits(&s->gb, 5); /* time_offset_length */
6000 static inline int decode_vui_parameters(H264Context *h, SPS *sps){
6001 MpegEncContext * const s = &h->s;
6002 int aspect_ratio_info_present_flag, aspect_ratio_idc;
6003 int nal_hrd_parameters_present_flag, vcl_hrd_parameters_present_flag;
6005 aspect_ratio_info_present_flag= get_bits1(&s->gb);
6007 if( aspect_ratio_info_present_flag ) {
6008 aspect_ratio_idc= get_bits(&s->gb, 8);
6009 if( aspect_ratio_idc == EXTENDED_SAR ) {
6010 sps->sar.num= get_bits(&s->gb, 16);
6011 sps->sar.den= get_bits(&s->gb, 16);
6012 }else if(aspect_ratio_idc < 16){
6013 sps->sar= pixel_aspect[aspect_ratio_idc];
6015 av_log(h->s.avctx, AV_LOG_ERROR, "illegal aspect ratio\n");
6022 // s->avctx->aspect_ratio= sar_width*s->width / (float)(s->height*sar_height);
6024 if(get_bits1(&s->gb)){ /* overscan_info_present_flag */
6025 get_bits1(&s->gb); /* overscan_appropriate_flag */
6028 if(get_bits1(&s->gb)){ /* video_signal_type_present_flag */
6029 get_bits(&s->gb, 3); /* video_format */
6030 get_bits1(&s->gb); /* video_full_range_flag */
6031 if(get_bits1(&s->gb)){ /* colour_description_present_flag */
6032 get_bits(&s->gb, 8); /* colour_primaries */
6033 get_bits(&s->gb, 8); /* transfer_characteristics */
6034 get_bits(&s->gb, 8); /* matrix_coefficients */
6038 if(get_bits1(&s->gb)){ /* chroma_location_info_present_flag */
6039 get_ue_golomb(&s->gb); /* chroma_sample_location_type_top_field */
6040 get_ue_golomb(&s->gb); /* chroma_sample_location_type_bottom_field */
6043 sps->timing_info_present_flag = get_bits1(&s->gb);
6044 if(sps->timing_info_present_flag){
6045 sps->num_units_in_tick = get_bits_long(&s->gb, 32);
6046 sps->time_scale = get_bits_long(&s->gb, 32);
6047 sps->fixed_frame_rate_flag = get_bits1(&s->gb);
6050 nal_hrd_parameters_present_flag = get_bits1(&s->gb);
6051 if(nal_hrd_parameters_present_flag)
6052 decode_hrd_parameters(h, sps);
6053 vcl_hrd_parameters_present_flag = get_bits1(&s->gb);
6054 if(vcl_hrd_parameters_present_flag)
6055 decode_hrd_parameters(h, sps);
6056 if(nal_hrd_parameters_present_flag || vcl_hrd_parameters_present_flag)
6057 get_bits1(&s->gb); /* low_delay_hrd_flag */
6058 get_bits1(&s->gb); /* pic_struct_present_flag */
6060 sps->bitstream_restriction_flag = get_bits1(&s->gb);
6061 if(sps->bitstream_restriction_flag){
6062 get_bits1(&s->gb); /* motion_vectors_over_pic_boundaries_flag */
6063 get_ue_golomb(&s->gb); /* max_bytes_per_pic_denom */
6064 get_ue_golomb(&s->gb); /* max_bits_per_mb_denom */
6065 get_ue_golomb(&s->gb); /* log2_max_mv_length_horizontal */
6066 get_ue_golomb(&s->gb); /* log2_max_mv_length_vertical */
6067 sps->num_reorder_frames = get_ue_golomb(&s->gb);
6068 get_ue_golomb(&s->gb); /* max_dec_frame_buffering */
6074 static inline int decode_seq_parameter_set(H264Context *h){
6075 MpegEncContext * const s = &h->s;
6076 int profile_idc, level_idc;
6080 profile_idc= get_bits(&s->gb, 8);
6081 get_bits1(&s->gb); //constraint_set0_flag
6082 get_bits1(&s->gb); //constraint_set1_flag
6083 get_bits1(&s->gb); //constraint_set2_flag
6084 get_bits1(&s->gb); //constraint_set3_flag
6085 get_bits(&s->gb, 4); // reserved
6086 level_idc= get_bits(&s->gb, 8);
6087 sps_id= get_ue_golomb(&s->gb);
6089 sps= &h->sps_buffer[ sps_id ];
6090 sps->profile_idc= profile_idc;
6091 sps->level_idc= level_idc;
6093 sps->log2_max_frame_num= get_ue_golomb(&s->gb) + 4;
6094 sps->poc_type= get_ue_golomb(&s->gb);
6096 if(sps->poc_type == 0){ //FIXME #define
6097 sps->log2_max_poc_lsb= get_ue_golomb(&s->gb) + 4;
6098 } else if(sps->poc_type == 1){//FIXME #define
6099 sps->delta_pic_order_always_zero_flag= get_bits1(&s->gb);
6100 sps->offset_for_non_ref_pic= get_se_golomb(&s->gb);
6101 sps->offset_for_top_to_bottom_field= get_se_golomb(&s->gb);
6102 sps->poc_cycle_length= get_ue_golomb(&s->gb);
6104 for(i=0; i<sps->poc_cycle_length; i++)
6105 sps->offset_for_ref_frame[i]= get_se_golomb(&s->gb);
6107 if(sps->poc_type > 2){
6108 av_log(h->s.avctx, AV_LOG_ERROR, "illegal POC type %d\n", sps->poc_type);
6112 sps->ref_frame_count= get_ue_golomb(&s->gb);
6113 if(sps->ref_frame_count > MAX_PICTURE_COUNT-2){
6114 av_log(h->s.avctx, AV_LOG_ERROR, "too many reference frames\n");
6116 sps->gaps_in_frame_num_allowed_flag= get_bits1(&s->gb);
6117 sps->mb_width= get_ue_golomb(&s->gb) + 1;
6118 sps->mb_height= get_ue_golomb(&s->gb) + 1;
6119 if((unsigned)sps->mb_width >= INT_MAX/16 || (unsigned)sps->mb_height >= INT_MAX/16 ||
6120 avcodec_check_dimensions(NULL, 16*sps->mb_width, 16*sps->mb_height))
6123 sps->frame_mbs_only_flag= get_bits1(&s->gb);
6124 if(!sps->frame_mbs_only_flag)
6125 sps->mb_aff= get_bits1(&s->gb);
6129 sps->direct_8x8_inference_flag= get_bits1(&s->gb);
6131 sps->crop= get_bits1(&s->gb);
6133 sps->crop_left = get_ue_golomb(&s->gb);
6134 sps->crop_right = get_ue_golomb(&s->gb);
6135 sps->crop_top = get_ue_golomb(&s->gb);
6136 sps->crop_bottom= get_ue_golomb(&s->gb);
6137 if(sps->crop_left || sps->crop_top){
6138 av_log(h->s.avctx, AV_LOG_ERROR, "insane cropping not completly supported, this could look slightly wrong ...\n");
6144 sps->crop_bottom= 0;
6147 sps->vui_parameters_present_flag= get_bits1(&s->gb);
6148 if( sps->vui_parameters_present_flag )
6149 decode_vui_parameters(h, sps);
6151 if(s->avctx->debug&FF_DEBUG_PICT_INFO){
6152 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",
6153 sps_id, sps->profile_idc, sps->level_idc,
6155 sps->ref_frame_count,
6156 sps->mb_width, sps->mb_height,
6157 sps->frame_mbs_only_flag ? "FRM" : (sps->mb_aff ? "MB-AFF" : "PIC-AFF"),
6158 sps->direct_8x8_inference_flag ? "8B8" : "",
6159 sps->crop_left, sps->crop_right,
6160 sps->crop_top, sps->crop_bottom,
6161 sps->vui_parameters_present_flag ? "VUI" : ""
6167 static inline int decode_picture_parameter_set(H264Context *h){
6168 MpegEncContext * const s = &h->s;
6169 int pps_id= get_ue_golomb(&s->gb);
6170 PPS *pps= &h->pps_buffer[pps_id];
6172 pps->sps_id= get_ue_golomb(&s->gb);
6173 pps->cabac= get_bits1(&s->gb);
6174 pps->pic_order_present= get_bits1(&s->gb);
6175 pps->slice_group_count= get_ue_golomb(&s->gb) + 1;
6176 if(pps->slice_group_count > 1 ){
6177 pps->mb_slice_group_map_type= get_ue_golomb(&s->gb);
6178 av_log(h->s.avctx, AV_LOG_ERROR, "FMO not supported\n");
6179 switch(pps->mb_slice_group_map_type){
6182 | for( i = 0; i <= num_slice_groups_minus1; i++ ) | | |
6183 | run_length[ i ] |1 |ue(v) |
6188 | for( i = 0; i < num_slice_groups_minus1; i++ ) | | |
6190 | top_left_mb[ i ] |1 |ue(v) |
6191 | bottom_right_mb[ i ] |1 |ue(v) |
6199 | slice_group_change_direction_flag |1 |u(1) |
6200 | slice_group_change_rate_minus1 |1 |ue(v) |
6205 | slice_group_id_cnt_minus1 |1 |ue(v) |
6206 | for( i = 0; i <= slice_group_id_cnt_minus1; i++ | | |
6208 | slice_group_id[ i ] |1 |u(v) |
6213 pps->ref_count[0]= get_ue_golomb(&s->gb) + 1;
6214 pps->ref_count[1]= get_ue_golomb(&s->gb) + 1;
6215 if(pps->ref_count[0] > 32 || pps->ref_count[1] > 32){
6216 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow (pps)\n");
6220 pps->weighted_pred= get_bits1(&s->gb);
6221 pps->weighted_bipred_idc= get_bits(&s->gb, 2);
6222 pps->init_qp= get_se_golomb(&s->gb) + 26;
6223 pps->init_qs= get_se_golomb(&s->gb) + 26;
6224 pps->chroma_qp_index_offset= get_se_golomb(&s->gb);
6225 pps->deblocking_filter_parameters_present= get_bits1(&s->gb);
6226 pps->constrained_intra_pred= get_bits1(&s->gb);
6227 pps->redundant_pic_cnt_present = get_bits1(&s->gb);
6229 if(s->avctx->debug&FF_DEBUG_PICT_INFO){
6230 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",
6231 pps_id, pps->sps_id,
6232 pps->cabac ? "CABAC" : "CAVLC",
6233 pps->slice_group_count,
6234 pps->ref_count[0], pps->ref_count[1],
6235 pps->weighted_pred ? "weighted" : "",
6236 pps->init_qp, pps->init_qs, pps->chroma_qp_index_offset,
6237 pps->deblocking_filter_parameters_present ? "LPAR" : "",
6238 pps->constrained_intra_pred ? "CONSTR" : "",
6239 pps->redundant_pic_cnt_present ? "REDU" : ""
6247 * finds the end of the current frame in the bitstream.
6248 * @return the position of the first byte of the next frame, or -1
6250 static int find_frame_end(H264Context *h, const uint8_t *buf, int buf_size){
6253 ParseContext *pc = &(h->s.parse_context);
6254 //printf("first %02X%02X%02X%02X\n", buf[0], buf[1],buf[2],buf[3]);
6255 // mb_addr= pc->mb_addr - 1;
6257 for(i=0; i<=buf_size; i++){
6258 if((state&0xFFFFFF1F) == 0x101 || (state&0xFFFFFF1F) == 0x102 || (state&0xFFFFFF1F) == 0x105){
6259 tprintf("find_frame_end new startcode = %08x, frame_start_found = %d, pos = %d\n", state, pc->frame_start_found, i);
6260 if(pc->frame_start_found){
6261 // If there isn't one more byte in the buffer
6262 // the test on first_mb_in_slice cannot be done yet
6263 // do it at next call.
6264 if (i >= buf_size) break;
6265 if (buf[i] & 0x80) {
6266 // first_mb_in_slice is 0, probably the first nal of a new
6268 tprintf("find_frame_end frame_end_found, state = %08x, pos = %d\n", state, i);
6270 pc->frame_start_found= 0;
6274 pc->frame_start_found = 1;
6277 state= (state<<8) | buf[i];
6281 return END_NOT_FOUND;
6284 static int h264_parse(AVCodecParserContext *s,
6285 AVCodecContext *avctx,
6286 uint8_t **poutbuf, int *poutbuf_size,
6287 const uint8_t *buf, int buf_size)
6289 H264Context *h = s->priv_data;
6290 ParseContext *pc = &h->s.parse_context;
6293 next= find_frame_end(h, buf, buf_size);
6295 if (ff_combine_frame(pc, next, (uint8_t **)&buf, &buf_size) < 0) {
6301 *poutbuf = (uint8_t *)buf;
6302 *poutbuf_size = buf_size;
6306 static int decode_nal_units(H264Context *h, uint8_t *buf, int buf_size){
6307 MpegEncContext * const s = &h->s;
6308 AVCodecContext * const avctx= s->avctx;
6312 for(i=0; i<32; i++){
6313 printf("%X ", buf[i]);
6325 if(buf_index >= buf_size) break;
6327 for(i = 0; i < h->nal_length_size; i++)
6328 nalsize = (nalsize << 8) | buf[buf_index++];
6330 // start code prefix search
6331 for(; buf_index + 3 < buf_size; buf_index++){
6332 // this should allways succeed in the first iteration
6333 if(buf[buf_index] == 0 && buf[buf_index+1] == 0 && buf[buf_index+2] == 1)
6337 if(buf_index+3 >= buf_size) break;
6342 ptr= decode_nal(h, buf + buf_index, &dst_length, &consumed, h->is_avc ? nalsize : buf_size - buf_index);
6343 if(ptr[dst_length - 1] == 0) dst_length--;
6344 bit_length= 8*dst_length - decode_rbsp_trailing(ptr + dst_length - 1);
6346 if(s->avctx->debug&FF_DEBUG_STARTCODE){
6347 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);
6350 if (h->is_avc && (nalsize != consumed))
6351 av_log(h->s.avctx, AV_LOG_ERROR, "AVC: Consumed only %d bytes instead of %d\n", consumed, nalsize);
6353 buf_index += consumed;
6355 if( s->hurry_up == 1 && h->nal_ref_idc == 0 )
6358 switch(h->nal_unit_type){
6360 idr(h); //FIXME ensure we dont loose some frames if there is reordering
6362 init_get_bits(&s->gb, ptr, bit_length);
6364 h->inter_gb_ptr= &s->gb;
6365 s->data_partitioning = 0;
6367 if(decode_slice_header(h) < 0) return -1;
6368 if(h->redundant_pic_count==0 && s->hurry_up < 5 )
6372 init_get_bits(&s->gb, ptr, bit_length);
6374 h->inter_gb_ptr= NULL;
6375 s->data_partitioning = 1;
6377 if(decode_slice_header(h) < 0) return -1;
6380 init_get_bits(&h->intra_gb, ptr, bit_length);
6381 h->intra_gb_ptr= &h->intra_gb;
6384 init_get_bits(&h->inter_gb, ptr, bit_length);
6385 h->inter_gb_ptr= &h->inter_gb;
6387 if(h->redundant_pic_count==0 && h->intra_gb_ptr && s->data_partitioning && s->hurry_up < 5 )
6393 init_get_bits(&s->gb, ptr, bit_length);
6394 decode_seq_parameter_set(h);
6396 if(s->flags& CODEC_FLAG_LOW_DELAY)
6399 if(avctx->has_b_frames < 2)
6400 avctx->has_b_frames= !s->low_delay;
6403 init_get_bits(&s->gb, ptr, bit_length);
6405 decode_picture_parameter_set(h);
6408 case NAL_PICTURE_DELIMITER:
6410 case NAL_FILTER_DATA:
6413 av_log(avctx, AV_LOG_ERROR, "Unknown NAL code: %d\n", h->nal_unit_type);
6417 if(!s->current_picture_ptr) return buf_index; //no frame
6419 s->current_picture_ptr->pict_type= s->pict_type;
6420 s->current_picture_ptr->key_frame= s->pict_type == I_TYPE && h->nal_unit_type == NAL_IDR_SLICE;
6422 h->prev_frame_num_offset= h->frame_num_offset;
6423 h->prev_frame_num= h->frame_num;
6424 if(s->current_picture_ptr->reference){
6425 h->prev_poc_msb= h->poc_msb;
6426 h->prev_poc_lsb= h->poc_lsb;
6428 if(s->current_picture_ptr->reference)
6429 execute_ref_pic_marking(h, h->mmco, h->mmco_index);
6439 * retunrs the number of bytes consumed for building the current frame
6441 static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size){
6442 if(s->flags&CODEC_FLAG_TRUNCATED){
6443 pos -= s->parse_context.last_index;
6444 if(pos<0) pos=0; // FIXME remove (uneeded?)
6448 if(pos==0) pos=1; //avoid infinite loops (i doubt thats needed but ...)
6449 if(pos+10>buf_size) pos=buf_size; // oops ;)
6455 static int decode_frame(AVCodecContext *avctx,
6456 void *data, int *data_size,
6457 uint8_t *buf, int buf_size)
6459 H264Context *h = avctx->priv_data;
6460 MpegEncContext *s = &h->s;
6461 AVFrame *pict = data;
6464 s->flags= avctx->flags;
6465 s->flags2= avctx->flags2;
6467 /* no supplementary picture */
6468 if (buf_size == 0) {
6472 if(s->flags&CODEC_FLAG_TRUNCATED){
6473 int next= find_frame_end(h, buf, buf_size);
6475 if( ff_combine_frame(&s->parse_context, next, &buf, &buf_size) < 0 )
6477 //printf("next:%d buf_size:%d last_index:%d\n", next, buf_size, s->parse_context.last_index);
6480 if(h->is_avc && !h->got_avcC) {
6481 int i, cnt, nalsize;
6482 unsigned char *p = avctx->extradata;
6483 if(avctx->extradata_size < 7) {
6484 av_log(avctx, AV_LOG_ERROR, "avcC too short\n");
6488 av_log(avctx, AV_LOG_ERROR, "Unknown avcC version %d\n", *p);
6491 /* sps and pps in the avcC always have length coded with 2 bytes,
6492 so put a fake nal_length_size = 2 while parsing them */
6493 h->nal_length_size = 2;
6494 // Decode sps from avcC
6495 cnt = *(p+5) & 0x1f; // Number of sps
6497 for (i = 0; i < cnt; i++) {
6498 nalsize = BE_16(p) + 2;
6499 if(decode_nal_units(h, p, nalsize) != nalsize) {
6500 av_log(avctx, AV_LOG_ERROR, "Decoding sps %d from avcC failed\n", i);
6505 // Decode pps from avcC
6506 cnt = *(p++); // Number of pps
6507 for (i = 0; i < cnt; i++) {
6508 nalsize = BE_16(p) + 2;
6509 if(decode_nal_units(h, p, nalsize) != nalsize) {
6510 av_log(avctx, AV_LOG_ERROR, "Decoding pps %d from avcC failed\n", i);
6515 // Now store right nal length size, that will be use to parse all other nals
6516 h->nal_length_size = ((*(((char*)(avctx->extradata))+4))&0x03)+1;
6517 // Do not reparse avcC
6521 if(!h->is_avc && s->avctx->extradata_size && s->picture_number==0){
6522 if(0 < decode_nal_units(h, s->avctx->extradata, s->avctx->extradata_size) )
6526 buf_index=decode_nal_units(h, buf, buf_size);
6530 //FIXME do something with unavailable reference frames
6532 // if(ret==FRAME_SKIPED) return get_consumed_bytes(s, buf_index, buf_size);
6533 if(!s->current_picture_ptr){
6534 av_log(h->s.avctx, AV_LOG_DEBUG, "error, NO frame\n");
6539 Picture *out = s->current_picture_ptr;
6540 #if 0 //decode order
6541 *data_size = sizeof(AVFrame);
6543 /* Sort B-frames into display order */
6544 Picture *cur = s->current_picture_ptr;
6545 Picture *prev = h->delayed_output_pic;
6550 int dropped_frame = 0;
6553 if(h->sps.bitstream_restriction_flag
6554 && s->avctx->has_b_frames < h->sps.num_reorder_frames){
6555 s->avctx->has_b_frames = h->sps.num_reorder_frames;
6559 while(h->delayed_pic[pics]) pics++;
6560 h->delayed_pic[pics++] = cur;
6561 if(cur->reference == 0)
6564 for(i=0; h->delayed_pic[i]; i++)
6565 if(h->delayed_pic[i]->key_frame)
6568 out = h->delayed_pic[0];
6569 for(i=1; h->delayed_pic[i] && !h->delayed_pic[i]->key_frame; i++)
6570 if(h->delayed_pic[i]->poc < out->poc){
6571 out = h->delayed_pic[i];
6575 out_of_order = !cross_idr && prev && out->poc < prev->poc;
6576 if(prev && pics <= s->avctx->has_b_frames)
6578 else if((out_of_order && pics-1 == s->avctx->has_b_frames)
6580 ((!cross_idr && prev && out->poc > prev->poc + 2)
6581 || cur->pict_type == B_TYPE)))
6584 s->avctx->has_b_frames++;
6587 else if(out_of_order)
6590 if(out_of_order || pics > s->avctx->has_b_frames){
6591 dropped_frame = (out != h->delayed_pic[out_idx]);
6592 for(i=out_idx; h->delayed_pic[i]; i++)
6593 h->delayed_pic[i] = h->delayed_pic[i+1];
6596 if(prev == out && !dropped_frame)
6599 *data_size = sizeof(AVFrame);
6600 if(prev && prev != out && prev->reference == 1)
6601 prev->reference = 0;
6602 h->delayed_output_pic = out;
6605 *pict= *(AVFrame*)out;
6608 assert(pict->data[0]);
6609 ff_print_debug_info(s, pict);
6610 //printf("out %d\n", (int)pict->data[0]);
6613 /* Return the Picture timestamp as the frame number */
6614 /* we substract 1 because it is added on utils.c */
6615 avctx->frame_number = s->picture_number - 1;
6617 return get_consumed_bytes(s, buf_index, buf_size);
6620 static inline void fill_mb_avail(H264Context *h){
6621 MpegEncContext * const s = &h->s;
6622 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
6625 h->mb_avail[0]= s->mb_x && h->slice_table[mb_xy - s->mb_stride - 1] == h->slice_num;
6626 h->mb_avail[1]= h->slice_table[mb_xy - s->mb_stride ] == h->slice_num;
6627 h->mb_avail[2]= s->mb_x+1 < s->mb_width && h->slice_table[mb_xy - s->mb_stride + 1] == h->slice_num;
6633 h->mb_avail[3]= s->mb_x && h->slice_table[mb_xy - 1] == h->slice_num;
6634 h->mb_avail[4]= 1; //FIXME move out
6635 h->mb_avail[5]= 0; //FIXME move out
6641 #define SIZE (COUNT*40)
6647 // int int_temp[10000];
6649 AVCodecContext avctx;
6651 dsputil_init(&dsp, &avctx);
6653 init_put_bits(&pb, temp, SIZE);
6654 printf("testing unsigned exp golomb\n");
6655 for(i=0; i<COUNT; i++){
6657 set_ue_golomb(&pb, i);
6658 STOP_TIMER("set_ue_golomb");
6660 flush_put_bits(&pb);
6662 init_get_bits(&gb, temp, 8*SIZE);
6663 for(i=0; i<COUNT; i++){
6666 s= show_bits(&gb, 24);
6669 j= get_ue_golomb(&gb);
6671 printf("missmatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
6674 STOP_TIMER("get_ue_golomb");
6678 init_put_bits(&pb, temp, SIZE);
6679 printf("testing signed exp golomb\n");
6680 for(i=0; i<COUNT; i++){
6682 set_se_golomb(&pb, i - COUNT/2);
6683 STOP_TIMER("set_se_golomb");
6685 flush_put_bits(&pb);
6687 init_get_bits(&gb, temp, 8*SIZE);
6688 for(i=0; i<COUNT; i++){
6691 s= show_bits(&gb, 24);
6694 j= get_se_golomb(&gb);
6695 if(j != i - COUNT/2){
6696 printf("missmatch! at %d (%d should be %d) bits:%6X\n", i, j, i, s);
6699 STOP_TIMER("get_se_golomb");
6702 printf("testing 4x4 (I)DCT\n");
6705 uint8_t src[16], ref[16];
6706 uint64_t error= 0, max_error=0;
6708 for(i=0; i<COUNT; i++){
6710 // printf("%d %d %d\n", r1, r2, (r2-r1)*16);
6711 for(j=0; j<16; j++){
6712 ref[j]= random()%255;
6713 src[j]= random()%255;
6716 h264_diff_dct_c(block, src, ref, 4);
6719 for(j=0; j<16; j++){
6720 // printf("%d ", block[j]);
6721 block[j]= block[j]*4;
6722 if(j&1) block[j]= (block[j]*4 + 2)/5;
6723 if(j&4) block[j]= (block[j]*4 + 2)/5;
6727 s->dsp.h264_idct_add(ref, block, 4);
6728 /* for(j=0; j<16; j++){
6729 printf("%d ", ref[j]);
6733 for(j=0; j<16; j++){
6734 int diff= ABS(src[j] - ref[j]);
6737 max_error= FFMAX(max_error, diff);
6740 printf("error=%f max_error=%d\n", ((float)error)/COUNT/16, (int)max_error );
6742 printf("testing quantizer\n");
6743 for(qp=0; qp<52; qp++){
6745 src1_block[i]= src2_block[i]= random()%255;
6749 printf("Testing NAL layer\n");
6751 uint8_t bitstream[COUNT];
6752 uint8_t nal[COUNT*2];
6754 memset(&h, 0, sizeof(H264Context));
6756 for(i=0; i<COUNT; i++){
6764 for(j=0; j<COUNT; j++){
6765 bitstream[j]= (random() % 255) + 1;
6768 for(j=0; j<zeros; j++){
6769 int pos= random() % COUNT;
6770 while(bitstream[pos] == 0){
6779 nal_length= encode_nal(&h, nal, bitstream, COUNT, COUNT*2);
6781 printf("encoding failed\n");
6785 out= decode_nal(&h, nal, &out_length, &consumed, nal_length);
6789 if(out_length != COUNT){
6790 printf("incorrect length %d %d\n", out_length, COUNT);
6794 if(consumed != nal_length){
6795 printf("incorrect consumed length %d %d\n", nal_length, consumed);
6799 if(memcmp(bitstream, out, COUNT)){
6800 printf("missmatch\n");
6805 printf("Testing RBSP\n");
6813 static int decode_end(AVCodecContext *avctx)
6815 H264Context *h = avctx->priv_data;
6816 MpegEncContext *s = &h->s;
6818 free_tables(h); //FIXME cleanup init stuff perhaps
6821 // memset(h, 0, sizeof(H264Context));
6827 AVCodec h264_decoder = {
6831 sizeof(H264Context),
6836 /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 | CODEC_CAP_TRUNCATED | CODEC_CAP_DELAY,
6839 AVCodecParser h264_parser = {
6841 sizeof(H264Context),