2 * H.26L/H.264/AVC/JVT/14496-10/... encoder/decoder
3 * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
5 * This library is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU Lesser General Public
7 * License as published by the Free Software Foundation; either
8 * version 2 of the License, or (at your option) any later version.
10 * This library is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * Lesser General Public License for more details.
15 * You should have received a copy of the GNU Lesser General Public
16 * License along with this library; if not, write to the Free Software
17 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
23 * H.264 / AVC / MPEG4 part10 codec.
24 * @author Michael Niedermayer <michaelni@gmx.at>
30 #include "mpegvideo.h"
39 #define interlaced_dct interlaced_dct_is_a_bad_name
40 #define mb_intra mb_intra_isnt_initalized_see_mb_type
42 #define LUMA_DC_BLOCK_INDEX 25
43 #define CHROMA_DC_BLOCK_INDEX 26
45 #define CHROMA_DC_COEFF_TOKEN_VLC_BITS 8
46 #define COEFF_TOKEN_VLC_BITS 8
47 #define TOTAL_ZEROS_VLC_BITS 9
48 #define CHROMA_DC_TOTAL_ZEROS_VLC_BITS 3
49 #define RUN_VLC_BITS 3
50 #define RUN7_VLC_BITS 6
52 #define MAX_SPS_COUNT 32
53 #define MAX_PPS_COUNT 256
55 #define MAX_MMCO_COUNT 66
58 * Sequence parameter set
64 int log2_max_frame_num; ///< log2_max_frame_num_minus4 + 4
65 int poc_type; ///< pic_order_cnt_type
66 int log2_max_poc_lsb; ///< log2_max_pic_order_cnt_lsb_minus4
67 int delta_pic_order_always_zero_flag;
68 int offset_for_non_ref_pic;
69 int offset_for_top_to_bottom_field;
70 int poc_cycle_length; ///< num_ref_frames_in_pic_order_cnt_cycle
71 int ref_frame_count; ///< num_ref_frames
72 int gaps_in_frame_num_allowed_flag;
73 int mb_width; ///< frame_width_in_mbs_minus1 + 1
74 int mb_height; ///< frame_height_in_mbs_minus1 + 1
75 int frame_mbs_only_flag;
76 int mb_aff; ///<mb_adaptive_frame_field_flag
77 int direct_8x8_inference_flag;
78 int crop; ///< frame_cropping_flag
79 int crop_left; ///< frame_cropping_rect_left_offset
80 int crop_right; ///< frame_cropping_rect_right_offset
81 int crop_top; ///< frame_cropping_rect_top_offset
82 int crop_bottom; ///< frame_cropping_rect_bottom_offset
83 int vui_parameters_present_flag;
85 int timing_info_present_flag;
86 uint32_t num_units_in_tick;
88 int fixed_frame_rate_flag;
89 short offset_for_ref_frame[256]; //FIXME dyn aloc?
93 * Picture parameter set
97 int cabac; ///< entropy_coding_mode_flag
98 int pic_order_present; ///< pic_order_present_flag
99 int slice_group_count; ///< num_slice_groups_minus1 + 1
100 int mb_slice_group_map_type;
101 int ref_count[2]; ///< num_ref_idx_l0/1_active_minus1 + 1
102 int weighted_pred; ///< weighted_pred_flag
103 int weighted_bipred_idc;
104 int init_qp; ///< pic_init_qp_minus26 + 26
105 int init_qs; ///< pic_init_qs_minus26 + 26
106 int chroma_qp_index_offset;
107 int deblocking_filter_parameters_present; ///< deblocking_filter_parameters_present_flag
108 int constrained_intra_pred; ///< constrained_intra_pred_flag
109 int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag
113 * Memory management control operation opcode.
115 typedef enum MMCOOpcode{
126 * Memory management control operation.
137 typedef struct H264Context{
145 #define NAL_IDR_SLICE 5
149 #define NAL_PICTURE_DELIMITER 9
150 #define NAL_FILTER_DATA 10
151 uint8_t *rbsp_buffer;
152 int rbsp_buffer_size;
155 * Used to parse AVC variant of h264
157 int is_avc; ///< this flag is != 0 if codec is avc1
158 int got_avcC; ///< flag used to parse avcC data only once
159 int nal_length_size; ///< Number of bytes used for nal length (1, 2 or 4)
163 int prev_mb_skiped; //FIXME remove (IMHO not used)
166 int chroma_pred_mode;
167 int intra16x16_pred_mode;
169 int8_t intra4x4_pred_mode_cache[5*8];
170 int8_t (*intra4x4_pred_mode)[8];
171 void (*pred4x4 [9+3])(uint8_t *src, uint8_t *topright, int stride);//FIXME move to dsp?
172 void (*pred8x8 [4+3])(uint8_t *src, int stride);
173 void (*pred16x16[4+3])(uint8_t *src, int stride);
174 unsigned int topleft_samples_available;
175 unsigned int top_samples_available;
176 unsigned int topright_samples_available;
177 unsigned int left_samples_available;
178 uint8_t (*top_border)[16+2*8];
179 uint8_t left_border[17+2*9];
182 * non zero coeff count cache.
183 * is 64 if not available.
185 uint8_t non_zero_count_cache[6*8];
186 uint8_t (*non_zero_count)[16];
189 * Motion vector cache.
191 int16_t mv_cache[2][5*8][2];
192 int8_t ref_cache[2][5*8];
193 #define LIST_NOT_USED -1 //FIXME rename?
194 #define PART_NOT_AVAILABLE -2
197 * is 1 if the specific list MV&references are set to 0,0,-2.
199 int mv_cache_clean[2];
201 int block_offset[16+8];
202 int chroma_subblock_offset[16]; //FIXME remove
204 uint16_t *mb2b_xy; //FIXME are these 4 a good idea?
206 int b_stride; //FIXME use s->b4_stride
212 int unknown_svq3_flag;
213 int next_slice_index;
215 SPS sps_buffer[MAX_SPS_COUNT];
216 SPS sps; ///< current sps
218 PPS pps_buffer[MAX_PPS_COUNT];
222 PPS pps; //FIXME move tp Picture perhaps? (->no) do we need that?
225 uint8_t *slice_table_base;
226 uint8_t *slice_table; ///< slice_table_base + mb_stride + 1
228 int slice_type_fixed;
230 //interlacing specific flags
231 int mb_field_decoding_flag;
238 int delta_poc_bottom;
241 int prev_poc_msb; ///< poc_msb of the last reference pic for POC type 0
242 int prev_poc_lsb; ///< poc_lsb of the last reference pic for POC type 0
243 int frame_num_offset; ///< for POC type 2
244 int prev_frame_num_offset; ///< for POC type 2
245 int prev_frame_num; ///< frame_num of the last pic for POC type 1/2
248 * frame_num for frames or 2*frame_num for field pics.
253 * max_frame_num or 2*max_frame_num for field pics.
257 //Weighted pred stuff
258 int luma_log2_weight_denom;
259 int chroma_log2_weight_denom;
260 int luma_weight[2][16];
261 int luma_offset[2][16];
262 int chroma_weight[2][16][2];
263 int chroma_offset[2][16][2];
266 int deblocking_filter; ///< disable_deblocking_filter_idc with 1<->0
267 int slice_alpha_c0_offset;
268 int slice_beta_offset;
270 int redundant_pic_count;
272 int direct_spatial_mv_pred;
273 int dist_scale_factor[16];
276 * num_ref_idx_l0/1_active_minus1 + 1
278 int ref_count[2];// FIXME split for AFF
279 Picture *short_ref[16];
280 Picture *long_ref[16];
281 Picture default_ref_list[2][32];
282 Picture ref_list[2][32]; //FIXME size?
283 Picture field_ref_list[2][32]; //FIXME size?
284 Picture *delayed_pic[16]; //FIXME size?
287 * memory management control operations buffer.
289 MMCO mmco[MAX_MMCO_COUNT];
292 int long_ref_count; ///< number of actual long term references
293 int short_ref_count; ///< number of actual short term references
296 GetBitContext intra_gb;
297 GetBitContext inter_gb;
298 GetBitContext *intra_gb_ptr;
299 GetBitContext *inter_gb_ptr;
301 DCTELEM mb[16*24] __align8;
307 uint8_t cabac_state[399];
310 /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0,1,2), 0x0? luma_cbp */
314 /* chroma_pred_mode for i4x4 or i16x16, else 0 */
315 uint8_t *chroma_pred_mode_table;
316 int last_qscale_diff;
317 int16_t (*mvd_table[2])[2];
318 int16_t mvd_cache[2][5*8][2];
319 uint8_t *direct_table;
320 uint8_t direct_cache[5*8];
324 static VLC coeff_token_vlc[4];
325 static VLC chroma_dc_coeff_token_vlc;
327 static VLC total_zeros_vlc[15];
328 static VLC chroma_dc_total_zeros_vlc[3];
330 static VLC run_vlc[6];
333 static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
334 static void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
335 static void filter_mb( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr);
337 static inline uint32_t pack16to32(int a, int b){
338 #ifdef WORDS_BIGENDIAN
339 return (b&0xFFFF) + (a<<16);
341 return (a&0xFFFF) + (b<<16);
347 * @param h height of the rectangle, should be a constant
348 * @param w width of the rectangle, should be a constant
349 * @param size the size of val (1 or 4), should be a constant
351 static inline void fill_rectangle(void *vp, int w, int h, int stride, uint32_t val, int size){ //FIXME ensure this IS inlined
352 uint8_t *p= (uint8_t*)vp;
353 assert(size==1 || size==4);
358 //FIXME check what gcc generates for 64 bit on x86 and possible write a 32 bit ver of it
361 *(uint16_t*)(p + stride)= size==4 ? val : val*0x0101;
362 }else if(w==2 && h==4){
363 *(uint16_t*)(p + 0*stride)=
364 *(uint16_t*)(p + 1*stride)=
365 *(uint16_t*)(p + 2*stride)=
366 *(uint16_t*)(p + 3*stride)= size==4 ? val : val*0x0101;
367 }else if(w==4 && h==1){
368 *(uint32_t*)(p + 0*stride)= size==4 ? val : val*0x01010101;
369 }else if(w==4 && h==2){
370 *(uint32_t*)(p + 0*stride)=
371 *(uint32_t*)(p + 1*stride)= size==4 ? val : val*0x01010101;
372 }else if(w==4 && h==4){
373 *(uint32_t*)(p + 0*stride)=
374 *(uint32_t*)(p + 1*stride)=
375 *(uint32_t*)(p + 2*stride)=
376 *(uint32_t*)(p + 3*stride)= size==4 ? val : val*0x01010101;
377 }else if(w==8 && h==1){
379 *(uint32_t*)(p + 4)= size==4 ? val : val*0x01010101;
380 }else if(w==8 && h==2){
381 *(uint32_t*)(p + 0 + 0*stride)=
382 *(uint32_t*)(p + 4 + 0*stride)=
383 *(uint32_t*)(p + 0 + 1*stride)=
384 *(uint32_t*)(p + 4 + 1*stride)= size==4 ? val : val*0x01010101;
385 }else if(w==8 && h==4){
386 *(uint64_t*)(p + 0*stride)=
387 *(uint64_t*)(p + 1*stride)=
388 *(uint64_t*)(p + 2*stride)=
389 *(uint64_t*)(p + 3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
390 }else if(w==16 && h==2){
391 *(uint64_t*)(p + 0+0*stride)=
392 *(uint64_t*)(p + 8+0*stride)=
393 *(uint64_t*)(p + 0+1*stride)=
394 *(uint64_t*)(p + 8+1*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
395 }else if(w==16 && h==4){
396 *(uint64_t*)(p + 0+0*stride)=
397 *(uint64_t*)(p + 8+0*stride)=
398 *(uint64_t*)(p + 0+1*stride)=
399 *(uint64_t*)(p + 8+1*stride)=
400 *(uint64_t*)(p + 0+2*stride)=
401 *(uint64_t*)(p + 8+2*stride)=
402 *(uint64_t*)(p + 0+3*stride)=
403 *(uint64_t*)(p + 8+3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
408 static inline void fill_caches(H264Context *h, int mb_type){
409 MpegEncContext * const s = &h->s;
410 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
411 int topleft_xy, top_xy, topright_xy, left_xy[2];
412 int topleft_type, top_type, topright_type, left_type[2];
416 //wow what a mess, why didnt they simplify the interlacing&intra stuff, i cant imagine that these complex rules are worth it
420 topleft_xy = 0; /* avoid warning */
421 top_xy = 0; /* avoid warning */
422 topright_xy = 0; /* avoid warning */
424 topleft_xy = mb_xy-1 - s->mb_stride;
425 top_xy = mb_xy - s->mb_stride;
426 topright_xy= mb_xy+1 - s->mb_stride;
427 left_xy[0] = mb_xy-1;
428 left_xy[1] = mb_xy-1;
435 topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
436 top_type = h->slice_table[top_xy ] == h->slice_num ? s->current_picture.mb_type[top_xy] : 0;
437 topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
438 left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
439 left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
441 if(IS_INTRA(mb_type)){
442 h->topleft_samples_available=
443 h->top_samples_available=
444 h->left_samples_available= 0xFFFF;
445 h->topright_samples_available= 0xEEEA;
447 if(!IS_INTRA(top_type) && (top_type==0 || h->pps.constrained_intra_pred)){
448 h->topleft_samples_available= 0xB3FF;
449 h->top_samples_available= 0x33FF;
450 h->topright_samples_available= 0x26EA;
453 if(!IS_INTRA(left_type[i]) && (left_type[i]==0 || h->pps.constrained_intra_pred)){
454 h->topleft_samples_available&= 0xDF5F;
455 h->left_samples_available&= 0x5F5F;
459 if(!IS_INTRA(topleft_type) && (topleft_type==0 || h->pps.constrained_intra_pred))
460 h->topleft_samples_available&= 0x7FFF;
462 if(!IS_INTRA(topright_type) && (topright_type==0 || h->pps.constrained_intra_pred))
463 h->topright_samples_available&= 0xFBFF;
465 if(IS_INTRA4x4(mb_type)){
466 if(IS_INTRA4x4(top_type)){
467 h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
468 h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
469 h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
470 h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
473 if(IS_INTRA16x16(top_type) || (IS_INTER(top_type) && !h->pps.constrained_intra_pred))
478 h->intra4x4_pred_mode_cache[4+8*0]=
479 h->intra4x4_pred_mode_cache[5+8*0]=
480 h->intra4x4_pred_mode_cache[6+8*0]=
481 h->intra4x4_pred_mode_cache[7+8*0]= pred;
484 if(IS_INTRA4x4(left_type[i])){
485 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
486 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
489 if(IS_INTRA16x16(left_type[i]) || (IS_INTER(left_type[i]) && !h->pps.constrained_intra_pred))
494 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
495 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
510 //FIXME constraint_intra_pred & partitioning & nnz (lets hope this is just a typo in the spec)
512 h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][0];
513 h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][1];
514 h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][2];
515 h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3];
517 h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][7];
518 h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8];
520 h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][10];
521 h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11];
523 h->top_cbp= h->cbp_table[top_xy];
525 h->non_zero_count_cache[4+8*0]=
526 h->non_zero_count_cache[5+8*0]=
527 h->non_zero_count_cache[6+8*0]=
528 h->non_zero_count_cache[7+8*0]=
530 h->non_zero_count_cache[1+8*0]=
531 h->non_zero_count_cache[2+8*0]=
533 h->non_zero_count_cache[1+8*3]=
534 h->non_zero_count_cache[2+8*3]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
536 if(IS_INTRA(mb_type)) h->top_cbp= 0x1C0;
541 h->non_zero_count_cache[3+8*1]= h->non_zero_count[left_xy[0]][6];
542 h->non_zero_count_cache[3+8*2]= h->non_zero_count[left_xy[0]][5];
543 h->non_zero_count_cache[0+8*1]= h->non_zero_count[left_xy[0]][9]; //FIXME left_block
544 h->non_zero_count_cache[0+8*4]= h->non_zero_count[left_xy[0]][12];
545 h->left_cbp= h->cbp_table[left_xy[0]]; //FIXME interlacing
547 h->non_zero_count_cache[3+8*1]=
548 h->non_zero_count_cache[3+8*2]=
549 h->non_zero_count_cache[0+8*1]=
550 h->non_zero_count_cache[0+8*4]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
552 if(IS_INTRA(mb_type)) h->left_cbp= 0x1C0;//FIXME interlacing
557 h->non_zero_count_cache[3+8*3]= h->non_zero_count[left_xy[1]][4];
558 h->non_zero_count_cache[3+8*4]= h->non_zero_count[left_xy[1]][3];
559 h->non_zero_count_cache[0+8*2]= h->non_zero_count[left_xy[1]][8];
560 h->non_zero_count_cache[0+8*5]= h->non_zero_count[left_xy[1]][11];
562 h->non_zero_count_cache[3+8*3]=
563 h->non_zero_count_cache[3+8*4]=
564 h->non_zero_count_cache[0+8*2]=
565 h->non_zero_count_cache[0+8*5]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
569 //FIXME direct mb can skip much of this
570 if(IS_INTER(mb_type) || (IS_DIRECT(mb_type) && h->direct_spatial_mv_pred)){
572 for(list=0; list<2; list++){
573 if((!IS_8X8(mb_type)) && !USES_LIST(mb_type, list) && !IS_DIRECT(mb_type)){
574 /*if(!h->mv_cache_clean[list]){
575 memset(h->mv_cache [list], 0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
576 memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
577 h->mv_cache_clean[list]= 1;
581 h->mv_cache_clean[list]= 0;
583 if(IS_INTER(topleft_type)){
584 const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
585 const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + h->b8_stride;
586 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
587 h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
589 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
590 h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
593 if(IS_INTER(top_type)){
594 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
595 const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
596 *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0];
597 *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];
598 *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];
599 *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];
600 h->ref_cache[list][scan8[0] + 0 - 1*8]=
601 h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
602 h->ref_cache[list][scan8[0] + 2 - 1*8]=
603 h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
605 *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]=
606 *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]=
607 *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]=
608 *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;
609 *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
612 if(IS_INTER(topright_type)){
613 const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
614 const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
615 *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
616 h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
618 *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
619 h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
622 //FIXME unify cleanup or sth
623 if(IS_INTER(left_type[0])){
624 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
625 const int b8_xy= h->mb2b8_xy[left_xy[0]] + 1;
626 *(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]];
627 *(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]];
628 h->ref_cache[list][scan8[0] - 1 + 0*8]=
629 h->ref_cache[list][scan8[0] - 1 + 1*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0]>>1)];
631 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 0*8]=
632 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 1*8]= 0;
633 h->ref_cache[list][scan8[0] - 1 + 0*8]=
634 h->ref_cache[list][scan8[0] - 1 + 1*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
637 if(IS_INTER(left_type[1])){
638 const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
639 const int b8_xy= h->mb2b8_xy[left_xy[1]] + 1;
640 *(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]];
641 *(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]];
642 h->ref_cache[list][scan8[0] - 1 + 2*8]=
643 h->ref_cache[list][scan8[0] - 1 + 3*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[2]>>1)];
645 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 2*8]=
646 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 3*8]= 0;
647 h->ref_cache[list][scan8[0] - 1 + 2*8]=
648 h->ref_cache[list][scan8[0] - 1 + 3*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
651 h->ref_cache[list][scan8[5 ]+1] =
652 h->ref_cache[list][scan8[7 ]+1] =
653 h->ref_cache[list][scan8[13]+1] = //FIXME remove past 3 (init somewher else)
654 h->ref_cache[list][scan8[4 ]] =
655 h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
656 *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
657 *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
658 *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewher else)
659 *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
660 *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
663 /* XXX beurk, Load mvd */
664 if(IS_INTER(topleft_type)){
665 const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
666 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy];
668 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 - 1*8]= 0;
671 if(IS_INTER(top_type)){
672 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
673 *(uint32_t*)h->mvd_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 0];
674 *(uint32_t*)h->mvd_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 1];
675 *(uint32_t*)h->mvd_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 2];
676 *(uint32_t*)h->mvd_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 3];
678 *(uint32_t*)h->mvd_cache [list][scan8[0] + 0 - 1*8]=
679 *(uint32_t*)h->mvd_cache [list][scan8[0] + 1 - 1*8]=
680 *(uint32_t*)h->mvd_cache [list][scan8[0] + 2 - 1*8]=
681 *(uint32_t*)h->mvd_cache [list][scan8[0] + 3 - 1*8]= 0;
683 if(IS_INTER(left_type[0])){
684 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
685 *(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]];
686 *(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]];
688 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 0*8]=
689 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 1*8]= 0;
691 if(IS_INTER(left_type[1])){
692 const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
693 *(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]];
694 *(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]];
696 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 2*8]=
697 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 3*8]= 0;
699 *(uint32_t*)h->mvd_cache [list][scan8[5 ]+1]=
700 *(uint32_t*)h->mvd_cache [list][scan8[7 ]+1]=
701 *(uint32_t*)h->mvd_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewher else)
702 *(uint32_t*)h->mvd_cache [list][scan8[4 ]]=
703 *(uint32_t*)h->mvd_cache [list][scan8[12]]= 0;
705 if(h->slice_type == B_TYPE){
706 fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, 0, 1);
708 if(IS_DIRECT(top_type)){
709 *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0x01010101;
710 }else if(IS_8X8(top_type)){
711 int b8_xy = h->mb2b8_xy[top_xy] + h->b8_stride;
712 h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy];
713 h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 1];
715 *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0;
719 if(IS_DIRECT(left_type[0])){
720 h->direct_cache[scan8[0] - 1 + 0*8]=
721 h->direct_cache[scan8[0] - 1 + 2*8]= 1;
722 }else if(IS_8X8(left_type[0])){
723 int b8_xy = h->mb2b8_xy[left_xy[0]] + 1;
724 h->direct_cache[scan8[0] - 1 + 0*8]= h->direct_table[b8_xy];
725 h->direct_cache[scan8[0] - 1 + 2*8]= h->direct_table[b8_xy + h->b8_stride];
727 h->direct_cache[scan8[0] - 1 + 0*8]=
728 h->direct_cache[scan8[0] - 1 + 2*8]= 0;
737 static inline void write_back_intra_pred_mode(H264Context *h){
738 MpegEncContext * const s = &h->s;
739 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
741 h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
742 h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
743 h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
744 h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
745 h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
746 h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
747 h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
751 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
753 static inline int check_intra4x4_pred_mode(H264Context *h){
754 MpegEncContext * const s = &h->s;
755 static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
756 static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
759 if(!(h->top_samples_available&0x8000)){
761 int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
763 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);
766 h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
771 if(!(h->left_samples_available&0x8000)){
773 int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
775 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);
778 h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
784 } //FIXME cleanup like next
787 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
789 static inline int check_intra_pred_mode(H264Context *h, int mode){
790 MpegEncContext * const s = &h->s;
791 static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
792 static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
794 if(mode < 0 || mode > 6) {
795 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);
799 if(!(h->top_samples_available&0x8000)){
802 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);
807 if(!(h->left_samples_available&0x8000)){
810 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);
819 * gets the predicted intra4x4 prediction mode.
821 static inline int pred_intra_mode(H264Context *h, int n){
822 const int index8= scan8[n];
823 const int left= h->intra4x4_pred_mode_cache[index8 - 1];
824 const int top = h->intra4x4_pred_mode_cache[index8 - 8];
825 const int min= FFMIN(left, top);
827 tprintf("mode:%d %d min:%d\n", left ,top, min);
829 if(min<0) return DC_PRED;
833 static inline void write_back_non_zero_count(H264Context *h){
834 MpegEncContext * const s = &h->s;
835 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
837 h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[4+8*4];
838 h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[5+8*4];
839 h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[6+8*4];
840 h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];
841 h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[7+8*3];
842 h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[7+8*2];
843 h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[7+8*1];
845 h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[1+8*2];
846 h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
847 h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[2+8*1];
849 h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[1+8*5];
850 h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
851 h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[2+8*4];
855 * gets the predicted number of non zero coefficients.
856 * @param n block index
858 static inline int pred_non_zero_count(H264Context *h, int n){
859 const int index8= scan8[n];
860 const int left= h->non_zero_count_cache[index8 - 1];
861 const int top = h->non_zero_count_cache[index8 - 8];
864 if(i<64) i= (i+1)>>1;
866 tprintf("pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
871 static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
872 const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];
874 if(topright_ref != PART_NOT_AVAILABLE){
875 *C= h->mv_cache[list][ i - 8 + part_width ];
878 tprintf("topright MV not available\n");
880 *C= h->mv_cache[list][ i - 8 - 1 ];
881 return h->ref_cache[list][ i - 8 - 1 ];
886 * gets the predicted MV.
887 * @param n the block index
888 * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)
889 * @param mx the x component of the predicted motion vector
890 * @param my the y component of the predicted motion vector
892 static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
893 const int index8= scan8[n];
894 const int top_ref= h->ref_cache[list][ index8 - 8 ];
895 const int left_ref= h->ref_cache[list][ index8 - 1 ];
896 const int16_t * const A= h->mv_cache[list][ index8 - 1 ];
897 const int16_t * const B= h->mv_cache[list][ index8 - 8 ];
899 int diagonal_ref, match_count;
901 assert(part_width==1 || part_width==2 || part_width==4);
911 diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);
912 match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);
913 if(match_count > 1){ //most common
914 *mx= mid_pred(A[0], B[0], C[0]);
915 *my= mid_pred(A[1], B[1], C[1]);
916 }else if(match_count==1){
920 }else if(top_ref==ref){
928 if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
932 *mx= mid_pred(A[0], B[0], C[0]);
933 *my= mid_pred(A[1], B[1], C[1]);
937 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);
941 * gets the directionally predicted 16x8 MV.
942 * @param n the block index
943 * @param mx the x component of the predicted motion vector
944 * @param my the y component of the predicted motion vector
946 static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
948 const int top_ref= h->ref_cache[list][ scan8[0] - 8 ];
949 const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
951 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);
959 const int left_ref= h->ref_cache[list][ scan8[8] - 1 ];
960 const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];
962 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);
972 pred_motion(h, n, 4, list, ref, mx, my);
976 * gets the directionally predicted 8x16 MV.
977 * @param n the block index
978 * @param mx the x component of the predicted motion vector
979 * @param my the y component of the predicted motion vector
981 static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
983 const int left_ref= h->ref_cache[list][ scan8[0] - 1 ];
984 const int16_t * const A= h->mv_cache[list][ scan8[0] - 1 ];
986 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);
997 diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);
999 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);
1001 if(diagonal_ref == ref){
1009 pred_motion(h, n, 2, list, ref, mx, my);
1012 static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
1013 const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
1014 const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];
1016 tprintf("pred_pskip: (%d) (%d) at %2d %2d\n", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
1018 if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
1019 || (top_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0)
1020 || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){
1026 pred_motion(h, 0, 4, 0, 0, mx, my);
1031 static inline void direct_dist_scale_factor(H264Context * const h){
1032 const int poc = h->s.current_picture_ptr->poc;
1033 const int poc1 = h->ref_list[1][0].poc;
1035 for(i=0; i<h->ref_count[0]; i++){
1036 int poc0 = h->ref_list[0][i].poc;
1037 int td = clip(poc1 - poc0, -128, 127);
1038 if(td == 0 /* FIXME || pic0 is a long-term ref */){
1039 h->dist_scale_factor[i] = 256;
1041 int tb = clip(poc - poc0, -128, 127);
1042 int tx = (16384 + (ABS(td) >> 1)) / td;
1043 h->dist_scale_factor[i] = clip((tb*tx + 32) >> 6, -1024, 1023);
1048 static inline void pred_direct_motion(H264Context * const h, int *mb_type){
1049 MpegEncContext * const s = &h->s;
1050 const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
1051 const int b8_xy = 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1052 const int b4_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1053 const int mb_type_col = h->ref_list[1][0].mb_type[mb_xy];
1054 const int16_t (*l1mv0)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[0][b4_xy];
1055 const int8_t *l1ref0 = &h->ref_list[1][0].ref_index[0][b8_xy];
1056 const int is_b8x8 = IS_8X8(*mb_type);
1060 if(IS_8X8(mb_type_col) && !h->sps.direct_8x8_inference_flag){
1061 /* FIXME save sub mb types from previous frames (or derive from MVs)
1062 * so we know exactly what block size to use */
1063 sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */
1064 *mb_type = MB_TYPE_8x8;
1065 }else if(!is_b8x8 && (IS_16X16(mb_type_col) || IS_INTRA(mb_type_col))){
1066 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1067 *mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */
1069 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1070 *mb_type = MB_TYPE_8x8;
1073 *mb_type |= MB_TYPE_DIRECT2;
1075 if(h->direct_spatial_mv_pred){
1080 /* ref = min(neighbors) */
1081 for(list=0; list<2; list++){
1082 int refa = h->ref_cache[list][scan8[0] - 1];
1083 int refb = h->ref_cache[list][scan8[0] - 8];
1084 int refc = h->ref_cache[list][scan8[0] - 8 + 4];
1086 refc = h->ref_cache[list][scan8[0] - 8 - 1];
1088 if(ref[list] < 0 || (refb < ref[list] && refb >= 0))
1090 if(ref[list] < 0 || (refc < ref[list] && refc >= 0))
1096 if(ref[0] < 0 && ref[1] < 0){
1097 ref[0] = ref[1] = 0;
1098 mv[0][0] = mv[0][1] =
1099 mv[1][0] = mv[1][1] = 0;
1101 for(list=0; list<2; list++){
1103 pred_motion(h, 0, 4, list, ref[list], &mv[list][0], &mv[list][1]);
1105 mv[list][0] = mv[list][1] = 0;
1110 *mb_type &= ~MB_TYPE_P0L1;
1111 sub_mb_type &= ~MB_TYPE_P0L1;
1112 }else if(ref[0] < 0){
1113 *mb_type &= ~MB_TYPE_P0L0;
1114 sub_mb_type &= ~MB_TYPE_P0L0;
1117 if(IS_16X16(*mb_type)){
1118 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref[0], 1);
1119 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, ref[1], 1);
1120 if(!IS_INTRA(mb_type_col) && l1ref0[0] == 0 &&
1121 ABS(l1mv0[0][0]) <= 1 && ABS(l1mv0[0][1]) <= 1){
1123 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1125 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
1127 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1129 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
1131 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1132 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1135 for(i8=0; i8<4; i8++){
1136 const int x8 = i8&1;
1137 const int y8 = i8>>1;
1139 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1141 h->sub_mb_type[i8] = sub_mb_type;
1143 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1144 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1145 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref[0], 1);
1146 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, ref[1], 1);
1149 if(!IS_INTRA(mb_type_col) && l1ref0[x8 + y8*h->b8_stride] == 0){
1150 for(i4=0; i4<4; i4++){
1151 const int16_t *mv_col = l1mv0[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1152 if(ABS(mv_col[0]) <= 1 && ABS(mv_col[1]) <= 1){
1154 *(uint32_t*)h->mv_cache[0][scan8[i8*4+i4]] = 0;
1156 *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = 0;
1162 }else{ /* direct temporal mv pred */
1163 /* FIXME assumes that L1ref0 used the same ref lists as current frame */
1164 if(IS_16X16(*mb_type)){
1165 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1);
1166 if(IS_INTRA(mb_type_col)){
1167 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
1168 fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
1169 fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
1171 const int ref0 = l1ref0[0];
1172 const int dist_scale_factor = h->dist_scale_factor[ref0];
1173 const int16_t *mv_col = l1mv0[0];
1175 mv_l0[0] = (dist_scale_factor * mv_col[0] + 128) >> 8;
1176 mv_l0[1] = (dist_scale_factor * mv_col[1] + 128) >> 8;
1177 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref0, 1);
1178 fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv_l0[0],mv_l0[1]), 4);
1179 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);
1182 for(i8=0; i8<4; i8++){
1183 const int x8 = i8&1;
1184 const int y8 = i8>>1;
1185 int ref0, dist_scale_factor;
1187 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1189 h->sub_mb_type[i8] = sub_mb_type;
1190 if(IS_INTRA(mb_type_col)){
1191 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1192 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1193 fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1194 fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1198 ref0 = l1ref0[x8 + y8*h->b8_stride];
1199 dist_scale_factor = h->dist_scale_factor[ref0];
1201 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1202 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1203 for(i4=0; i4<4; i4++){
1204 const int16_t *mv_col = l1mv0[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1205 int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]];
1206 mv_l0[0] = (dist_scale_factor * mv_col[0] + 128) >> 8;
1207 mv_l0[1] = (dist_scale_factor * mv_col[1] + 128) >> 8;
1208 *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] =
1209 pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
1216 static inline void write_back_motion(H264Context *h, int mb_type){
1217 MpegEncContext * const s = &h->s;
1218 const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1219 const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1222 for(list=0; list<2; list++){
1224 if((!IS_8X8(mb_type)) && !USES_LIST(mb_type, list)){
1225 if(1){ //FIXME skip or never read if mb_type doesnt use it
1227 *(uint64_t*)s->current_picture.motion_val[list][b_xy + 0 + y*h->b_stride]=
1228 *(uint64_t*)s->current_picture.motion_val[list][b_xy + 2 + y*h->b_stride]= 0;
1230 if( h->pps.cabac ) {
1231 /* FIXME needed ? */
1233 *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]=
1234 *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= 0;
1238 *(uint16_t*)&s->current_picture.ref_index[list][b8_xy + y*h->b8_stride]= (LIST_NOT_USED&0xFF)*0x0101;
1245 *(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];
1246 *(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];
1248 if( h->pps.cabac ) {
1250 *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+0 + 8*y];
1251 *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+2 + 8*y];
1255 s->current_picture.ref_index[list][b8_xy + 0 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+0 + 16*y];
1256 s->current_picture.ref_index[list][b8_xy + 1 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+2 + 16*y];
1260 if(h->slice_type == B_TYPE && h->pps.cabac){
1261 if(IS_8X8(mb_type)){
1262 h->direct_table[b8_xy+1+0*h->b8_stride] = IS_DIRECT(h->sub_mb_type[1]) ? 1 : 0;
1263 h->direct_table[b8_xy+0+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[2]) ? 1 : 0;
1264 h->direct_table[b8_xy+1+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[3]) ? 1 : 0;
1270 * Decodes a network abstraction layer unit.
1271 * @param consumed is the number of bytes used as input
1272 * @param length is the length of the array
1273 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp ttailing?
1274 * @returns decoded bytes, might be src+1 if no escapes
1276 static uint8_t *decode_nal(H264Context *h, uint8_t *src, int *dst_length, int *consumed, int length){
1280 // src[0]&0x80; //forbidden bit
1281 h->nal_ref_idc= src[0]>>5;
1282 h->nal_unit_type= src[0]&0x1F;
1286 for(i=0; i<length; i++)
1287 printf("%2X ", src[i]);
1289 for(i=0; i+1<length; i+=2){
1290 if(src[i]) continue;
1291 if(i>0 && src[i-1]==0) i--;
1292 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1294 /* startcode, so we must be past the end */
1301 if(i>=length-1){ //no escaped 0
1302 *dst_length= length;
1303 *consumed= length+1; //+1 for the header
1307 h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length);
1308 dst= h->rbsp_buffer;
1310 //printf("deoding esc\n");
1313 //remove escapes (very rare 1:2^22)
1314 if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1315 if(src[si+2]==3){ //escape
1320 }else //next start code
1324 dst[di++]= src[si++];
1328 *consumed= si + 1;//+1 for the header
1329 //FIXME store exact number of bits in the getbitcontext (its needed for decoding)
1335 * @param src the data which should be escaped
1336 * @param dst the target buffer, dst+1 == src is allowed as a special case
1337 * @param length the length of the src data
1338 * @param dst_length the length of the dst array
1339 * @returns length of escaped data in bytes or -1 if an error occured
1341 static int encode_nal(H264Context *h, uint8_t *dst, uint8_t *src, int length, int dst_length){
1342 int i, escape_count, si, di;
1346 assert(dst_length>0);
1348 dst[0]= (h->nal_ref_idc<<5) + h->nal_unit_type;
1350 if(length==0) return 1;
1353 for(i=0; i<length; i+=2){
1354 if(src[i]) continue;
1355 if(i>0 && src[i-1]==0)
1357 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1363 if(escape_count==0){
1365 memcpy(dst+1, src, length);
1369 if(length + escape_count + 1> dst_length)
1372 //this should be damn rare (hopefully)
1374 h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length + escape_count);
1375 temp= h->rbsp_buffer;
1376 //printf("encoding esc\n");
1381 if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1382 temp[di++]= 0; si++;
1383 temp[di++]= 0; si++;
1385 temp[di++]= src[si++];
1388 temp[di++]= src[si++];
1390 memcpy(dst+1, temp, length+escape_count);
1392 assert(di == length+escape_count);
1398 * write 1,10,100,1000,... for alignment, yes its exactly inverse to mpeg4
1400 static void encode_rbsp_trailing(PutBitContext *pb){
1403 length= (-put_bits_count(pb))&7;
1404 if(length) put_bits(pb, length, 0);
1409 * identifies the exact end of the bitstream
1410 * @return the length of the trailing, or 0 if damaged
1412 static int decode_rbsp_trailing(uint8_t *src){
1416 tprintf("rbsp trailing %X\n", v);
1426 * idct tranforms the 16 dc values and dequantize them.
1427 * @param qp quantization parameter
1429 static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp){
1430 const int qmul= dequant_coeff[qp][0];
1433 int temp[16]; //FIXME check if this is a good idea
1434 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
1435 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1437 //memset(block, 64, 2*256);
1440 const int offset= y_offset[i];
1441 const int z0= block[offset+stride*0] + block[offset+stride*4];
1442 const int z1= block[offset+stride*0] - block[offset+stride*4];
1443 const int z2= block[offset+stride*1] - block[offset+stride*5];
1444 const int z3= block[offset+stride*1] + block[offset+stride*5];
1453 const int offset= x_offset[i];
1454 const int z0= temp[4*0+i] + temp[4*2+i];
1455 const int z1= temp[4*0+i] - temp[4*2+i];
1456 const int z2= temp[4*1+i] - temp[4*3+i];
1457 const int z3= temp[4*1+i] + temp[4*3+i];
1459 block[stride*0 +offset]= ((z0 + z3)*qmul + 2)>>2; //FIXME think about merging this into decode_resdual
1460 block[stride*2 +offset]= ((z1 + z2)*qmul + 2)>>2;
1461 block[stride*8 +offset]= ((z1 - z2)*qmul + 2)>>2;
1462 block[stride*10+offset]= ((z0 - z3)*qmul + 2)>>2;
1468 * dct tranforms the 16 dc values.
1469 * @param qp quantization parameter ??? FIXME
1471 static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
1472 // const int qmul= dequant_coeff[qp][0];
1474 int temp[16]; //FIXME check if this is a good idea
1475 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
1476 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1479 const int offset= y_offset[i];
1480 const int z0= block[offset+stride*0] + block[offset+stride*4];
1481 const int z1= block[offset+stride*0] - block[offset+stride*4];
1482 const int z2= block[offset+stride*1] - block[offset+stride*5];
1483 const int z3= block[offset+stride*1] + block[offset+stride*5];
1492 const int offset= x_offset[i];
1493 const int z0= temp[4*0+i] + temp[4*2+i];
1494 const int z1= temp[4*0+i] - temp[4*2+i];
1495 const int z2= temp[4*1+i] - temp[4*3+i];
1496 const int z3= temp[4*1+i] + temp[4*3+i];
1498 block[stride*0 +offset]= (z0 + z3)>>1;
1499 block[stride*2 +offset]= (z1 + z2)>>1;
1500 block[stride*8 +offset]= (z1 - z2)>>1;
1501 block[stride*10+offset]= (z0 - z3)>>1;
1509 static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp){
1510 const int qmul= dequant_coeff[qp][0];
1511 const int stride= 16*2;
1512 const int xStride= 16;
1515 a= block[stride*0 + xStride*0];
1516 b= block[stride*0 + xStride*1];
1517 c= block[stride*1 + xStride*0];
1518 d= block[stride*1 + xStride*1];
1525 block[stride*0 + xStride*0]= ((a+c)*qmul + 0)>>1;
1526 block[stride*0 + xStride*1]= ((e+b)*qmul + 0)>>1;
1527 block[stride*1 + xStride*0]= ((a-c)*qmul + 0)>>1;
1528 block[stride*1 + xStride*1]= ((e-b)*qmul + 0)>>1;
1532 static void chroma_dc_dct_c(DCTELEM *block){
1533 const int stride= 16*2;
1534 const int xStride= 16;
1537 a= block[stride*0 + xStride*0];
1538 b= block[stride*0 + xStride*1];
1539 c= block[stride*1 + xStride*0];
1540 d= block[stride*1 + xStride*1];
1547 block[stride*0 + xStride*0]= (a+c);
1548 block[stride*0 + xStride*1]= (e+b);
1549 block[stride*1 + xStride*0]= (a-c);
1550 block[stride*1 + xStride*1]= (e-b);
1555 * gets the chroma qp.
1557 static inline int get_chroma_qp(H264Context *h, int qscale){
1559 return chroma_qp[clip(qscale + h->pps.chroma_qp_index_offset, 0, 51)];
1564 static void h264_diff_dct_c(DCTELEM *block, uint8_t *src1, uint8_t *src2, int stride){
1566 //FIXME try int temp instead of block
1569 const int d0= src1[0 + i*stride] - src2[0 + i*stride];
1570 const int d1= src1[1 + i*stride] - src2[1 + i*stride];
1571 const int d2= src1[2 + i*stride] - src2[2 + i*stride];
1572 const int d3= src1[3 + i*stride] - src2[3 + i*stride];
1573 const int z0= d0 + d3;
1574 const int z3= d0 - d3;
1575 const int z1= d1 + d2;
1576 const int z2= d1 - d2;
1578 block[0 + 4*i]= z0 + z1;
1579 block[1 + 4*i]= 2*z3 + z2;
1580 block[2 + 4*i]= z0 - z1;
1581 block[3 + 4*i]= z3 - 2*z2;
1585 const int z0= block[0*4 + i] + block[3*4 + i];
1586 const int z3= block[0*4 + i] - block[3*4 + i];
1587 const int z1= block[1*4 + i] + block[2*4 + i];
1588 const int z2= block[1*4 + i] - block[2*4 + i];
1590 block[0*4 + i]= z0 + z1;
1591 block[1*4 + i]= 2*z3 + z2;
1592 block[2*4 + i]= z0 - z1;
1593 block[3*4 + i]= z3 - 2*z2;
1598 //FIXME need to check that this doesnt overflow signed 32 bit for low qp, iam not sure, its very close
1599 //FIXME check that gcc inlines this (and optimizes intra & seperate_dc stuff away)
1600 static inline int quantize_c(DCTELEM *block, uint8_t *scantable, int qscale, int intra, int seperate_dc){
1602 const int * const quant_table= quant_coeff[qscale];
1603 const int bias= intra ? (1<<QUANT_SHIFT)/3 : (1<<QUANT_SHIFT)/6;
1604 const unsigned int threshold1= (1<<QUANT_SHIFT) - bias - 1;
1605 const unsigned int threshold2= (threshold1<<1);
1611 const int dc_bias= intra ? (1<<(QUANT_SHIFT-2))/3 : (1<<(QUANT_SHIFT-2))/6;
1612 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT-2)) - dc_bias - 1;
1613 const unsigned int dc_threshold2= (dc_threshold1<<1);
1615 int level= block[0]*quant_coeff[qscale+18][0];
1616 if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1618 level= (dc_bias + level)>>(QUANT_SHIFT-2);
1621 level= (dc_bias - level)>>(QUANT_SHIFT-2);
1624 // last_non_zero = i;
1629 const int dc_bias= intra ? (1<<(QUANT_SHIFT+1))/3 : (1<<(QUANT_SHIFT+1))/6;
1630 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT+1)) - dc_bias - 1;
1631 const unsigned int dc_threshold2= (dc_threshold1<<1);
1633 int level= block[0]*quant_table[0];
1634 if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1636 level= (dc_bias + level)>>(QUANT_SHIFT+1);
1639 level= (dc_bias - level)>>(QUANT_SHIFT+1);
1642 // last_non_zero = i;
1655 const int j= scantable[i];
1656 int level= block[j]*quant_table[j];
1658 // if( bias+level >= (1<<(QMAT_SHIFT - 3))
1659 // || bias-level >= (1<<(QMAT_SHIFT - 3))){
1660 if(((unsigned)(level+threshold1))>threshold2){
1662 level= (bias + level)>>QUANT_SHIFT;
1665 level= (bias - level)>>QUANT_SHIFT;
1674 return last_non_zero;
1677 static void pred4x4_vertical_c(uint8_t *src, uint8_t *topright, int stride){
1678 const uint32_t a= ((uint32_t*)(src-stride))[0];
1679 ((uint32_t*)(src+0*stride))[0]= a;
1680 ((uint32_t*)(src+1*stride))[0]= a;
1681 ((uint32_t*)(src+2*stride))[0]= a;
1682 ((uint32_t*)(src+3*stride))[0]= a;
1685 static void pred4x4_horizontal_c(uint8_t *src, uint8_t *topright, int stride){
1686 ((uint32_t*)(src+0*stride))[0]= src[-1+0*stride]*0x01010101;
1687 ((uint32_t*)(src+1*stride))[0]= src[-1+1*stride]*0x01010101;
1688 ((uint32_t*)(src+2*stride))[0]= src[-1+2*stride]*0x01010101;
1689 ((uint32_t*)(src+3*stride))[0]= src[-1+3*stride]*0x01010101;
1692 static void pred4x4_dc_c(uint8_t *src, uint8_t *topright, int stride){
1693 const int dc= ( src[-stride] + src[1-stride] + src[2-stride] + src[3-stride]
1694 + src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 4) >>3;
1696 ((uint32_t*)(src+0*stride))[0]=
1697 ((uint32_t*)(src+1*stride))[0]=
1698 ((uint32_t*)(src+2*stride))[0]=
1699 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1702 static void pred4x4_left_dc_c(uint8_t *src, uint8_t *topright, int stride){
1703 const int dc= ( src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 2) >>2;
1705 ((uint32_t*)(src+0*stride))[0]=
1706 ((uint32_t*)(src+1*stride))[0]=
1707 ((uint32_t*)(src+2*stride))[0]=
1708 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1711 static void pred4x4_top_dc_c(uint8_t *src, uint8_t *topright, int stride){
1712 const int dc= ( src[-stride] + src[1-stride] + src[2-stride] + src[3-stride] + 2) >>2;
1714 ((uint32_t*)(src+0*stride))[0]=
1715 ((uint32_t*)(src+1*stride))[0]=
1716 ((uint32_t*)(src+2*stride))[0]=
1717 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1720 static void pred4x4_128_dc_c(uint8_t *src, uint8_t *topright, int stride){
1721 ((uint32_t*)(src+0*stride))[0]=
1722 ((uint32_t*)(src+1*stride))[0]=
1723 ((uint32_t*)(src+2*stride))[0]=
1724 ((uint32_t*)(src+3*stride))[0]= 128U*0x01010101U;
1728 #define LOAD_TOP_RIGHT_EDGE\
1729 const int t4= topright[0];\
1730 const int t5= topright[1];\
1731 const int t6= topright[2];\
1732 const int t7= topright[3];\
1734 #define LOAD_LEFT_EDGE\
1735 const int l0= src[-1+0*stride];\
1736 const int l1= src[-1+1*stride];\
1737 const int l2= src[-1+2*stride];\
1738 const int l3= src[-1+3*stride];\
1740 #define LOAD_TOP_EDGE\
1741 const int t0= src[ 0-1*stride];\
1742 const int t1= src[ 1-1*stride];\
1743 const int t2= src[ 2-1*stride];\
1744 const int t3= src[ 3-1*stride];\
1746 static void pred4x4_down_right_c(uint8_t *src, uint8_t *topright, int stride){
1747 const int lt= src[-1-1*stride];
1751 src[0+3*stride]=(l3 + 2*l2 + l1 + 2)>>2;
1753 src[1+3*stride]=(l2 + 2*l1 + l0 + 2)>>2;
1756 src[2+3*stride]=(l1 + 2*l0 + lt + 2)>>2;
1760 src[3+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1763 src[3+2*stride]=(lt + 2*t0 + t1 + 2)>>2;
1765 src[3+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1766 src[3+0*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1769 static void pred4x4_down_left_c(uint8_t *src, uint8_t *topright, int stride){
1774 src[0+0*stride]=(t0 + t2 + 2*t1 + 2)>>2;
1776 src[0+1*stride]=(t1 + t3 + 2*t2 + 2)>>2;
1779 src[0+2*stride]=(t2 + t4 + 2*t3 + 2)>>2;
1783 src[0+3*stride]=(t3 + t5 + 2*t4 + 2)>>2;
1786 src[1+3*stride]=(t4 + t6 + 2*t5 + 2)>>2;
1788 src[2+3*stride]=(t5 + t7 + 2*t6 + 2)>>2;
1789 src[3+3*stride]=(t6 + 3*t7 + 2)>>2;
1792 static void pred4x4_vertical_right_c(uint8_t *src, uint8_t *topright, int stride){
1793 const int lt= src[-1-1*stride];
1796 const __attribute__((unused)) int unu= l3;
1799 src[1+2*stride]=(lt + t0 + 1)>>1;
1801 src[2+2*stride]=(t0 + t1 + 1)>>1;
1803 src[3+2*stride]=(t1 + t2 + 1)>>1;
1804 src[3+0*stride]=(t2 + t3 + 1)>>1;
1806 src[1+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1808 src[2+3*stride]=(lt + 2*t0 + t1 + 2)>>2;
1810 src[3+3*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1811 src[3+1*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1812 src[0+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1813 src[0+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1816 static void pred4x4_vertical_left_c(uint8_t *src, uint8_t *topright, int stride){
1819 const __attribute__((unused)) int unu= t7;
1821 src[0+0*stride]=(t0 + t1 + 1)>>1;
1823 src[0+2*stride]=(t1 + t2 + 1)>>1;
1825 src[1+2*stride]=(t2 + t3 + 1)>>1;
1827 src[2+2*stride]=(t3 + t4+ 1)>>1;
1828 src[3+2*stride]=(t4 + t5+ 1)>>1;
1829 src[0+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1831 src[0+3*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1833 src[1+3*stride]=(t2 + 2*t3 + t4 + 2)>>2;
1835 src[2+3*stride]=(t3 + 2*t4 + t5 + 2)>>2;
1836 src[3+3*stride]=(t4 + 2*t5 + t6 + 2)>>2;
1839 static void pred4x4_horizontal_up_c(uint8_t *src, uint8_t *topright, int stride){
1842 src[0+0*stride]=(l0 + l1 + 1)>>1;
1843 src[1+0*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1845 src[0+1*stride]=(l1 + l2 + 1)>>1;
1847 src[1+1*stride]=(l1 + 2*l2 + l3 + 2)>>2;
1849 src[0+2*stride]=(l2 + l3 + 1)>>1;
1851 src[1+2*stride]=(l2 + 2*l3 + l3 + 2)>>2;
1860 static void pred4x4_horizontal_down_c(uint8_t *src, uint8_t *topright, int stride){
1861 const int lt= src[-1-1*stride];
1864 const __attribute__((unused)) int unu= t3;
1867 src[2+1*stride]=(lt + l0 + 1)>>1;
1869 src[3+1*stride]=(l0 + 2*lt + t0 + 2)>>2;
1870 src[2+0*stride]=(lt + 2*t0 + t1 + 2)>>2;
1871 src[3+0*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1873 src[2+2*stride]=(l0 + l1 + 1)>>1;
1875 src[3+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1877 src[2+3*stride]=(l1 + l2+ 1)>>1;
1879 src[3+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1880 src[0+3*stride]=(l2 + l3 + 1)>>1;
1881 src[1+3*stride]=(l1 + 2*l2 + l3 + 2)>>2;
1884 static void pred16x16_vertical_c(uint8_t *src, int stride){
1886 const uint32_t a= ((uint32_t*)(src-stride))[0];
1887 const uint32_t b= ((uint32_t*)(src-stride))[1];
1888 const uint32_t c= ((uint32_t*)(src-stride))[2];
1889 const uint32_t d= ((uint32_t*)(src-stride))[3];
1891 for(i=0; i<16; i++){
1892 ((uint32_t*)(src+i*stride))[0]= a;
1893 ((uint32_t*)(src+i*stride))[1]= b;
1894 ((uint32_t*)(src+i*stride))[2]= c;
1895 ((uint32_t*)(src+i*stride))[3]= d;
1899 static void pred16x16_horizontal_c(uint8_t *src, int stride){
1902 for(i=0; i<16; i++){
1903 ((uint32_t*)(src+i*stride))[0]=
1904 ((uint32_t*)(src+i*stride))[1]=
1905 ((uint32_t*)(src+i*stride))[2]=
1906 ((uint32_t*)(src+i*stride))[3]= src[-1+i*stride]*0x01010101;
1910 static void pred16x16_dc_c(uint8_t *src, int stride){
1914 dc+= src[-1+i*stride];
1921 dc= 0x01010101*((dc + 16)>>5);
1923 for(i=0; i<16; i++){
1924 ((uint32_t*)(src+i*stride))[0]=
1925 ((uint32_t*)(src+i*stride))[1]=
1926 ((uint32_t*)(src+i*stride))[2]=
1927 ((uint32_t*)(src+i*stride))[3]= dc;
1931 static void pred16x16_left_dc_c(uint8_t *src, int stride){
1935 dc+= src[-1+i*stride];
1938 dc= 0x01010101*((dc + 8)>>4);
1940 for(i=0; i<16; i++){
1941 ((uint32_t*)(src+i*stride))[0]=
1942 ((uint32_t*)(src+i*stride))[1]=
1943 ((uint32_t*)(src+i*stride))[2]=
1944 ((uint32_t*)(src+i*stride))[3]= dc;
1948 static void pred16x16_top_dc_c(uint8_t *src, int stride){
1954 dc= 0x01010101*((dc + 8)>>4);
1956 for(i=0; i<16; i++){
1957 ((uint32_t*)(src+i*stride))[0]=
1958 ((uint32_t*)(src+i*stride))[1]=
1959 ((uint32_t*)(src+i*stride))[2]=
1960 ((uint32_t*)(src+i*stride))[3]= dc;
1964 static void pred16x16_128_dc_c(uint8_t *src, int stride){
1967 for(i=0; i<16; i++){
1968 ((uint32_t*)(src+i*stride))[0]=
1969 ((uint32_t*)(src+i*stride))[1]=
1970 ((uint32_t*)(src+i*stride))[2]=
1971 ((uint32_t*)(src+i*stride))[3]= 0x01010101U*128U;
1975 static inline void pred16x16_plane_compat_c(uint8_t *src, int stride, const int svq3){
1978 uint8_t *cm = cropTbl + MAX_NEG_CROP;
1979 const uint8_t * const src0 = src+7-stride;
1980 const uint8_t *src1 = src+8*stride-1;
1981 const uint8_t *src2 = src1-2*stride; // == src+6*stride-1;
1982 int H = src0[1] - src0[-1];
1983 int V = src1[0] - src2[ 0];
1984 for(k=2; k<=8; ++k) {
1985 src1 += stride; src2 -= stride;
1986 H += k*(src0[k] - src0[-k]);
1987 V += k*(src1[0] - src2[ 0]);
1990 H = ( 5*(H/4) ) / 16;
1991 V = ( 5*(V/4) ) / 16;
1993 /* required for 100% accuracy */
1994 i = H; H = V; V = i;
1996 H = ( 5*H+32 ) >> 6;
1997 V = ( 5*V+32 ) >> 6;
2000 a = 16*(src1[0] + src2[16] + 1) - 7*(V+H);
2001 for(j=16; j>0; --j) {
2004 for(i=-16; i<0; i+=4) {
2005 src[16+i] = cm[ (b ) >> 5 ];
2006 src[17+i] = cm[ (b+ H) >> 5 ];
2007 src[18+i] = cm[ (b+2*H) >> 5 ];
2008 src[19+i] = cm[ (b+3*H) >> 5 ];
2015 static void pred16x16_plane_c(uint8_t *src, int stride){
2016 pred16x16_plane_compat_c(src, stride, 0);
2019 static void pred8x8_vertical_c(uint8_t *src, int stride){
2021 const uint32_t a= ((uint32_t*)(src-stride))[0];
2022 const uint32_t b= ((uint32_t*)(src-stride))[1];
2025 ((uint32_t*)(src+i*stride))[0]= a;
2026 ((uint32_t*)(src+i*stride))[1]= b;
2030 static void pred8x8_horizontal_c(uint8_t *src, int stride){
2034 ((uint32_t*)(src+i*stride))[0]=
2035 ((uint32_t*)(src+i*stride))[1]= src[-1+i*stride]*0x01010101;
2039 static void pred8x8_128_dc_c(uint8_t *src, int stride){
2043 ((uint32_t*)(src+i*stride))[0]=
2044 ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
2047 ((uint32_t*)(src+i*stride))[0]=
2048 ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
2052 static void pred8x8_left_dc_c(uint8_t *src, int stride){
2058 dc0+= src[-1+i*stride];
2059 dc2+= src[-1+(i+4)*stride];
2061 dc0= 0x01010101*((dc0 + 2)>>2);
2062 dc2= 0x01010101*((dc2 + 2)>>2);
2065 ((uint32_t*)(src+i*stride))[0]=
2066 ((uint32_t*)(src+i*stride))[1]= dc0;
2069 ((uint32_t*)(src+i*stride))[0]=
2070 ((uint32_t*)(src+i*stride))[1]= dc2;
2074 static void pred8x8_top_dc_c(uint8_t *src, int stride){
2080 dc0+= src[i-stride];
2081 dc1+= src[4+i-stride];
2083 dc0= 0x01010101*((dc0 + 2)>>2);
2084 dc1= 0x01010101*((dc1 + 2)>>2);
2087 ((uint32_t*)(src+i*stride))[0]= dc0;
2088 ((uint32_t*)(src+i*stride))[1]= dc1;
2091 ((uint32_t*)(src+i*stride))[0]= dc0;
2092 ((uint32_t*)(src+i*stride))[1]= dc1;
2097 static void pred8x8_dc_c(uint8_t *src, int stride){
2099 int dc0, dc1, dc2, dc3;
2103 dc0+= src[-1+i*stride] + src[i-stride];
2104 dc1+= src[4+i-stride];
2105 dc2+= src[-1+(i+4)*stride];
2107 dc3= 0x01010101*((dc1 + dc2 + 4)>>3);
2108 dc0= 0x01010101*((dc0 + 4)>>3);
2109 dc1= 0x01010101*((dc1 + 2)>>2);
2110 dc2= 0x01010101*((dc2 + 2)>>2);
2113 ((uint32_t*)(src+i*stride))[0]= dc0;
2114 ((uint32_t*)(src+i*stride))[1]= dc1;
2117 ((uint32_t*)(src+i*stride))[0]= dc2;
2118 ((uint32_t*)(src+i*stride))[1]= dc3;
2122 static void pred8x8_plane_c(uint8_t *src, int stride){
2125 uint8_t *cm = cropTbl + MAX_NEG_CROP;
2126 const uint8_t * const src0 = src+3-stride;
2127 const uint8_t *src1 = src+4*stride-1;
2128 const uint8_t *src2 = src1-2*stride; // == src+2*stride-1;
2129 int H = src0[1] - src0[-1];
2130 int V = src1[0] - src2[ 0];
2131 for(k=2; k<=4; ++k) {
2132 src1 += stride; src2 -= stride;
2133 H += k*(src0[k] - src0[-k]);
2134 V += k*(src1[0] - src2[ 0]);
2136 H = ( 17*H+16 ) >> 5;
2137 V = ( 17*V+16 ) >> 5;
2139 a = 16*(src1[0] + src2[8]+1) - 3*(V+H);
2140 for(j=8; j>0; --j) {
2143 src[0] = cm[ (b ) >> 5 ];
2144 src[1] = cm[ (b+ H) >> 5 ];
2145 src[2] = cm[ (b+2*H) >> 5 ];
2146 src[3] = cm[ (b+3*H) >> 5 ];
2147 src[4] = cm[ (b+4*H) >> 5 ];
2148 src[5] = cm[ (b+5*H) >> 5 ];
2149 src[6] = cm[ (b+6*H) >> 5 ];
2150 src[7] = cm[ (b+7*H) >> 5 ];
2155 static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
2156 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2157 int src_x_offset, int src_y_offset,
2158 qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
2159 MpegEncContext * const s = &h->s;
2160 const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
2161 const int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
2162 const int luma_xy= (mx&3) + ((my&3)<<2);
2163 uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*s->linesize;
2164 uint8_t * src_cb= pic->data[1] + (mx>>3) + (my>>3)*s->uvlinesize;
2165 uint8_t * src_cr= pic->data[2] + (mx>>3) + (my>>3)*s->uvlinesize;
2166 int extra_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16; //FIXME increase edge?, IMHO not worth it
2167 int extra_height= extra_width;
2169 const int full_mx= mx>>2;
2170 const int full_my= my>>2;
2172 assert(pic->data[0]);
2174 if(mx&7) extra_width -= 3;
2175 if(my&7) extra_height -= 3;
2177 if( full_mx < 0-extra_width
2178 || full_my < 0-extra_height
2179 || full_mx + 16/*FIXME*/ > s->width + extra_width
2180 || full_my + 16/*FIXME*/ > s->height + extra_height){
2181 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);
2182 src_y= s->edge_emu_buffer + 2 + 2*s->linesize;
2186 qpix_op[luma_xy](dest_y, src_y, s->linesize); //FIXME try variable height perhaps?
2188 qpix_op[luma_xy](dest_y + delta, src_y + delta, s->linesize);
2191 if(s->flags&CODEC_FLAG_GRAY) return;
2194 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);
2195 src_cb= s->edge_emu_buffer;
2197 chroma_op(dest_cb, src_cb, s->uvlinesize, chroma_height, mx&7, my&7);
2200 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);
2201 src_cr= s->edge_emu_buffer;
2203 chroma_op(dest_cr, src_cr, s->uvlinesize, chroma_height, mx&7, my&7);
2206 static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
2207 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2208 int x_offset, int y_offset,
2209 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2210 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
2211 int list0, int list1){
2212 MpegEncContext * const s = &h->s;
2213 qpel_mc_func *qpix_op= qpix_put;
2214 h264_chroma_mc_func chroma_op= chroma_put;
2216 dest_y += 2*x_offset + 2*y_offset*s-> linesize;
2217 dest_cb += x_offset + y_offset*s->uvlinesize;
2218 dest_cr += x_offset + y_offset*s->uvlinesize;
2219 x_offset += 8*s->mb_x;
2220 y_offset += 8*s->mb_y;
2223 Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
2224 mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
2225 dest_y, dest_cb, dest_cr, x_offset, y_offset,
2226 qpix_op, chroma_op);
2229 chroma_op= chroma_avg;
2233 Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
2234 mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
2235 dest_y, dest_cb, dest_cr, x_offset, y_offset,
2236 qpix_op, chroma_op);
2240 static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2241 qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
2242 qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg)){
2243 MpegEncContext * const s = &h->s;
2244 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
2245 const int mb_type= s->current_picture.mb_type[mb_xy];
2247 assert(IS_INTER(mb_type));
2249 if(IS_16X16(mb_type)){
2250 mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
2251 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
2252 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2253 }else if(IS_16X8(mb_type)){
2254 mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
2255 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
2256 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2257 mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
2258 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
2259 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2260 }else if(IS_8X16(mb_type)){
2261 mc_part(h, 0, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 0, 0,
2262 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2263 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2264 mc_part(h, 4, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 4, 0,
2265 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2266 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2270 assert(IS_8X8(mb_type));
2273 const int sub_mb_type= h->sub_mb_type[i];
2275 int x_offset= (i&1)<<2;
2276 int y_offset= (i&2)<<1;
2278 if(IS_SUB_8X8(sub_mb_type)){
2279 mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2280 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2281 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2282 }else if(IS_SUB_8X4(sub_mb_type)){
2283 mc_part(h, n , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2284 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2285 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2286 mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
2287 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2288 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2289 }else if(IS_SUB_4X8(sub_mb_type)){
2290 mc_part(h, n , 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2291 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2292 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2293 mc_part(h, n+1, 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
2294 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2295 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2298 assert(IS_SUB_4X4(sub_mb_type));
2300 int sub_x_offset= x_offset + 2*(j&1);
2301 int sub_y_offset= y_offset + (j&2);
2302 mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
2303 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2304 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2311 static void decode_init_vlc(H264Context *h){
2312 static int done = 0;
2318 init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5,
2319 &chroma_dc_coeff_token_len [0], 1, 1,
2320 &chroma_dc_coeff_token_bits[0], 1, 1, 1);
2323 init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17,
2324 &coeff_token_len [i][0], 1, 1,
2325 &coeff_token_bits[i][0], 1, 1, 1);
2329 init_vlc(&chroma_dc_total_zeros_vlc[i], CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
2330 &chroma_dc_total_zeros_len [i][0], 1, 1,
2331 &chroma_dc_total_zeros_bits[i][0], 1, 1, 1);
2333 for(i=0; i<15; i++){
2334 init_vlc(&total_zeros_vlc[i], TOTAL_ZEROS_VLC_BITS, 16,
2335 &total_zeros_len [i][0], 1, 1,
2336 &total_zeros_bits[i][0], 1, 1, 1);
2340 init_vlc(&run_vlc[i], RUN_VLC_BITS, 7,
2341 &run_len [i][0], 1, 1,
2342 &run_bits[i][0], 1, 1, 1);
2344 init_vlc(&run7_vlc, RUN7_VLC_BITS, 16,
2345 &run_len [6][0], 1, 1,
2346 &run_bits[6][0], 1, 1, 1);
2351 * Sets the intra prediction function pointers.
2353 static void init_pred_ptrs(H264Context *h){
2354 // MpegEncContext * const s = &h->s;
2356 h->pred4x4[VERT_PRED ]= pred4x4_vertical_c;
2357 h->pred4x4[HOR_PRED ]= pred4x4_horizontal_c;
2358 h->pred4x4[DC_PRED ]= pred4x4_dc_c;
2359 h->pred4x4[DIAG_DOWN_LEFT_PRED ]= pred4x4_down_left_c;
2360 h->pred4x4[DIAG_DOWN_RIGHT_PRED]= pred4x4_down_right_c;
2361 h->pred4x4[VERT_RIGHT_PRED ]= pred4x4_vertical_right_c;
2362 h->pred4x4[HOR_DOWN_PRED ]= pred4x4_horizontal_down_c;
2363 h->pred4x4[VERT_LEFT_PRED ]= pred4x4_vertical_left_c;
2364 h->pred4x4[HOR_UP_PRED ]= pred4x4_horizontal_up_c;
2365 h->pred4x4[LEFT_DC_PRED ]= pred4x4_left_dc_c;
2366 h->pred4x4[TOP_DC_PRED ]= pred4x4_top_dc_c;
2367 h->pred4x4[DC_128_PRED ]= pred4x4_128_dc_c;
2369 h->pred8x8[DC_PRED8x8 ]= pred8x8_dc_c;
2370 h->pred8x8[VERT_PRED8x8 ]= pred8x8_vertical_c;
2371 h->pred8x8[HOR_PRED8x8 ]= pred8x8_horizontal_c;
2372 h->pred8x8[PLANE_PRED8x8 ]= pred8x8_plane_c;
2373 h->pred8x8[LEFT_DC_PRED8x8]= pred8x8_left_dc_c;
2374 h->pred8x8[TOP_DC_PRED8x8 ]= pred8x8_top_dc_c;
2375 h->pred8x8[DC_128_PRED8x8 ]= pred8x8_128_dc_c;
2377 h->pred16x16[DC_PRED8x8 ]= pred16x16_dc_c;
2378 h->pred16x16[VERT_PRED8x8 ]= pred16x16_vertical_c;
2379 h->pred16x16[HOR_PRED8x8 ]= pred16x16_horizontal_c;
2380 h->pred16x16[PLANE_PRED8x8 ]= pred16x16_plane_c;
2381 h->pred16x16[LEFT_DC_PRED8x8]= pred16x16_left_dc_c;
2382 h->pred16x16[TOP_DC_PRED8x8 ]= pred16x16_top_dc_c;
2383 h->pred16x16[DC_128_PRED8x8 ]= pred16x16_128_dc_c;
2386 static void free_tables(H264Context *h){
2387 av_freep(&h->intra4x4_pred_mode);
2388 av_freep(&h->chroma_pred_mode_table);
2389 av_freep(&h->cbp_table);
2390 av_freep(&h->mvd_table[0]);
2391 av_freep(&h->mvd_table[1]);
2392 av_freep(&h->direct_table);
2393 av_freep(&h->non_zero_count);
2394 av_freep(&h->slice_table_base);
2395 av_freep(&h->top_border);
2396 h->slice_table= NULL;
2398 av_freep(&h->mb2b_xy);
2399 av_freep(&h->mb2b8_xy);
2404 * needs widzh/height
2406 static int alloc_tables(H264Context *h){
2407 MpegEncContext * const s = &h->s;
2408 const int big_mb_num= s->mb_stride * (s->mb_height+1);
2411 CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8 * sizeof(uint8_t))
2413 CHECKED_ALLOCZ(h->non_zero_count , big_mb_num * 16 * sizeof(uint8_t))
2414 CHECKED_ALLOCZ(h->slice_table_base , big_mb_num * sizeof(uint8_t))
2415 CHECKED_ALLOCZ(h->top_border , s->mb_width * (16+8+8) * sizeof(uint8_t))
2416 CHECKED_ALLOCZ(h->cbp_table, big_mb_num * sizeof(uint16_t))
2418 if( h->pps.cabac ) {
2419 CHECKED_ALLOCZ(h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t))
2420 CHECKED_ALLOCZ(h->mvd_table[0], 32*big_mb_num * sizeof(uint16_t));
2421 CHECKED_ALLOCZ(h->mvd_table[1], 32*big_mb_num * sizeof(uint16_t));
2422 CHECKED_ALLOCZ(h->direct_table, 32*big_mb_num * sizeof(uint8_t));
2425 memset(h->slice_table_base, -1, big_mb_num * sizeof(uint8_t));
2426 h->slice_table= h->slice_table_base + s->mb_stride + 1;
2428 CHECKED_ALLOCZ(h->mb2b_xy , big_mb_num * sizeof(uint16_t));
2429 CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint16_t));
2430 for(y=0; y<s->mb_height; y++){
2431 for(x=0; x<s->mb_width; x++){
2432 const int mb_xy= x + y*s->mb_stride;
2433 const int b_xy = 4*x + 4*y*h->b_stride;
2434 const int b8_xy= 2*x + 2*y*h->b8_stride;
2436 h->mb2b_xy [mb_xy]= b_xy;
2437 h->mb2b8_xy[mb_xy]= b8_xy;
2447 static void common_init(H264Context *h){
2448 MpegEncContext * const s = &h->s;
2450 s->width = s->avctx->width;
2451 s->height = s->avctx->height;
2452 s->codec_id= s->avctx->codec->id;
2456 s->unrestricted_mv=1;
2457 s->decode=1; //FIXME
2460 static int decode_init(AVCodecContext *avctx){
2461 H264Context *h= avctx->priv_data;
2462 MpegEncContext * const s = &h->s;
2464 MPV_decode_defaults(s);
2469 s->out_format = FMT_H264;
2470 s->workaround_bugs= avctx->workaround_bugs;
2473 // s->decode_mb= ff_h263_decode_mb;
2475 avctx->pix_fmt= PIX_FMT_YUV420P;
2479 if(avctx->codec_tag != 0x31637661 && avctx->codec_tag != 0x31435641) // avc1
2482 if((avctx->extradata_size == 0) || (avctx->extradata == NULL)) {
2483 av_log(avctx, AV_LOG_ERROR, "AVC codec requires avcC data\n");
2493 static void frame_start(H264Context *h){
2494 MpegEncContext * const s = &h->s;
2497 MPV_frame_start(s, s->avctx);
2498 ff_er_frame_start(s);
2501 assert(s->linesize && s->uvlinesize);
2503 for(i=0; i<16; i++){
2504 h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
2505 h->chroma_subblock_offset[i]= 2*((scan8[i] - scan8[0])&7) + 2*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2508 h->block_offset[16+i]=
2509 h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2512 // s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
2515 static inline void backup_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){
2516 MpegEncContext * const s = &h->s;
2520 src_cb -= uvlinesize;
2521 src_cr -= uvlinesize;
2523 h->left_border[0]= h->top_border[s->mb_x][15];
2524 for(i=1; i<17; i++){
2525 h->left_border[i]= src_y[15+i* linesize];
2528 *(uint64_t*)(h->top_border[s->mb_x]+0)= *(uint64_t*)(src_y + 16*linesize);
2529 *(uint64_t*)(h->top_border[s->mb_x]+8)= *(uint64_t*)(src_y +8+16*linesize);
2531 if(!(s->flags&CODEC_FLAG_GRAY)){
2532 h->left_border[17 ]= h->top_border[s->mb_x][16+7];
2533 h->left_border[17+9]= h->top_border[s->mb_x][24+7];
2535 h->left_border[i+17 ]= src_cb[7+i*uvlinesize];
2536 h->left_border[i+17+9]= src_cr[7+i*uvlinesize];
2538 *(uint64_t*)(h->top_border[s->mb_x]+16)= *(uint64_t*)(src_cb+8*uvlinesize);
2539 *(uint64_t*)(h->top_border[s->mb_x]+24)= *(uint64_t*)(src_cr+8*uvlinesize);
2543 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){
2544 MpegEncContext * const s = &h->s;
2547 int deblock_left = (s->mb_x > 0);
2548 int deblock_top = (s->mb_y > 0);
2550 src_y -= linesize + 1;
2551 src_cb -= uvlinesize + 1;
2552 src_cr -= uvlinesize + 1;
2554 #define XCHG(a,b,t,xchg)\
2561 for(i = !deblock_top; i<17; i++){
2562 XCHG(h->left_border[i ], src_y [i* linesize], temp8, xchg);
2567 XCHG(*(uint64_t*)(h->top_border[s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
2568 XCHG(*(uint64_t*)(h->top_border[s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
2571 if(!(s->flags&CODEC_FLAG_GRAY)){
2573 for(i = !deblock_top; i<9; i++){
2574 XCHG(h->left_border[i+17 ], src_cb[i*uvlinesize], temp8, xchg);
2575 XCHG(h->left_border[i+17+9], src_cr[i*uvlinesize], temp8, xchg);
2579 XCHG(*(uint64_t*)(h->top_border[s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
2580 XCHG(*(uint64_t*)(h->top_border[s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
2585 static void hl_decode_mb(H264Context *h){
2586 MpegEncContext * const s = &h->s;
2587 const int mb_x= s->mb_x;
2588 const int mb_y= s->mb_y;
2589 const int mb_xy= mb_x + mb_y*s->mb_stride;
2590 const int mb_type= s->current_picture.mb_type[mb_xy];
2591 uint8_t *dest_y, *dest_cb, *dest_cr;
2592 int linesize, uvlinesize /*dct_offset*/;
2601 dest_y = s->current_picture.data[0] + (mb_y * 16* s->linesize ) + mb_x * 16;
2602 dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2603 dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
2605 if (h->mb_field_decoding_flag) {
2606 linesize = s->linesize * 2;
2607 uvlinesize = s->uvlinesize * 2;
2608 if(mb_y&1){ //FIXME move out of this func?
2609 dest_y -= s->linesize*15;
2610 dest_cb-= s->linesize*7;
2611 dest_cr-= s->linesize*7;
2614 linesize = s->linesize;
2615 uvlinesize = s->uvlinesize;
2616 // dct_offset = s->linesize * 16;
2619 if(IS_INTRA(mb_type)){
2620 if(h->deblocking_filter)
2621 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1);
2623 if(!(s->flags&CODEC_FLAG_GRAY)){
2624 h->pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
2625 h->pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
2628 if(IS_INTRA4x4(mb_type)){
2630 for(i=0; i<16; i++){
2631 uint8_t * const ptr= dest_y + h->block_offset[i];
2633 const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
2636 if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
2637 const int topright_avail= (h->topright_samples_available<<i)&0x8000;
2638 assert(mb_y || linesize <= h->block_offset[i]);
2639 if(!topright_avail){
2640 tr= ptr[3 - linesize]*0x01010101;
2641 topright= (uint8_t*) &tr;
2642 }else if(i==5 && h->deblocking_filter){
2643 tr= *(uint32_t*)h->top_border[mb_x+1];
2644 topright= (uint8_t*) &tr;
2646 topright= ptr + 4 - linesize;
2650 h->pred4x4[ dir ](ptr, topright, linesize);
2651 if(h->non_zero_count_cache[ scan8[i] ]){
2652 if(s->codec_id == CODEC_ID_H264)
2653 s->dsp.h264_idct_add(ptr, h->mb + i*16, linesize);
2655 svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
2660 h->pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
2661 if(s->codec_id == CODEC_ID_H264)
2662 h264_luma_dc_dequant_idct_c(h->mb, s->qscale);
2664 svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
2666 if(h->deblocking_filter)
2667 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);
2668 }else if(s->codec_id == CODEC_ID_H264){
2669 hl_motion(h, dest_y, dest_cb, dest_cr,
2670 s->dsp.put_h264_qpel_pixels_tab, s->dsp.put_h264_chroma_pixels_tab,
2671 s->dsp.avg_h264_qpel_pixels_tab, s->dsp.avg_h264_chroma_pixels_tab);
2675 if(!IS_INTRA4x4(mb_type)){
2676 if(s->codec_id == CODEC_ID_H264){
2677 for(i=0; i<16; i++){
2678 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
2679 uint8_t * const ptr= dest_y + h->block_offset[i];
2680 s->dsp.h264_idct_add(ptr, h->mb + i*16, linesize);
2684 for(i=0; i<16; i++){
2685 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
2686 uint8_t * const ptr= dest_y + h->block_offset[i];
2687 svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
2693 if(!(s->flags&CODEC_FLAG_GRAY)){
2694 chroma_dc_dequant_idct_c(h->mb + 16*16, h->chroma_qp);
2695 chroma_dc_dequant_idct_c(h->mb + 16*16+4*16, h->chroma_qp);
2696 if(s->codec_id == CODEC_ID_H264){
2697 for(i=16; i<16+4; i++){
2698 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2699 uint8_t * const ptr= dest_cb + h->block_offset[i];
2700 s->dsp.h264_idct_add(ptr, h->mb + i*16, uvlinesize);
2703 for(i=20; i<20+4; i++){
2704 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2705 uint8_t * const ptr= dest_cr + h->block_offset[i];
2706 s->dsp.h264_idct_add(ptr, h->mb + i*16, uvlinesize);
2710 for(i=16; i<16+4; i++){
2711 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2712 uint8_t * const ptr= dest_cb + h->block_offset[i];
2713 svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
2716 for(i=20; i<20+4; i++){
2717 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
2718 uint8_t * const ptr= dest_cr + h->block_offset[i];
2719 svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
2724 if(h->deblocking_filter) {
2725 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
2726 filter_mb(h, mb_x, mb_y, dest_y, dest_cb, dest_cr);
2731 * fills the default_ref_list.
2733 static int fill_default_ref_list(H264Context *h){
2734 MpegEncContext * const s = &h->s;
2736 Picture sorted_short_ref[16];
2738 if(h->slice_type==B_TYPE){
2742 for(out_i=0; out_i<h->short_ref_count; out_i++){
2744 int best_poc=INT_MAX;
2746 for(i=0; i<h->short_ref_count; i++){
2747 const int poc= h->short_ref[i]->poc;
2748 if(poc > limit && poc < best_poc){
2754 assert(best_i != -1);
2757 sorted_short_ref[out_i]= *h->short_ref[best_i];
2761 if(s->picture_structure == PICT_FRAME){
2762 if(h->slice_type==B_TYPE){
2763 const int current_poc= s->current_picture_ptr->poc;
2766 for(list=0; list<2; list++){
2769 for(i=0; i<h->short_ref_count && index < h->ref_count[list]; i++){
2770 const int i2= list ? i : h->short_ref_count - i - 1;
2771 const int poc= sorted_short_ref[i2].poc;
2773 if(sorted_short_ref[i2].reference != 3) continue; //FIXME refernce field shit
2775 if((list==1 && poc > current_poc) || (list==0 && poc < current_poc)){
2776 h->default_ref_list[list][index ]= sorted_short_ref[i2];
2777 h->default_ref_list[list][index++].pic_id= sorted_short_ref[i2].frame_num;
2781 for(i=0; i<h->long_ref_count && index < h->ref_count[ list ]; i++){
2782 if(h->long_ref[i]->reference != 3) continue;
2784 h->default_ref_list[ list ][index ]= *h->long_ref[i];
2785 h->default_ref_list[ list ][index++].pic_id= i;;
2788 if(h->long_ref_count > 1 && h->short_ref_count==0){
2789 Picture temp= h->default_ref_list[1][0];
2790 h->default_ref_list[1][0] = h->default_ref_list[1][1];
2791 h->default_ref_list[1][0] = temp;
2794 if(index < h->ref_count[ list ])
2795 memset(&h->default_ref_list[list][index], 0, sizeof(Picture)*(h->ref_count[ list ] - index));
2799 for(i=0; i<h->short_ref_count && index < h->ref_count[0]; i++){
2800 if(h->short_ref[i]->reference != 3) continue; //FIXME refernce field shit
2801 h->default_ref_list[0][index ]= *h->short_ref[i];
2802 h->default_ref_list[0][index++].pic_id= h->short_ref[i]->frame_num;
2804 for(i=0; i<h->long_ref_count && index < h->ref_count[0]; i++){
2805 if(h->long_ref[i]->reference != 3) continue;
2806 h->default_ref_list[0][index ]= *h->long_ref[i];
2807 h->default_ref_list[0][index++].pic_id= i;;
2809 if(index < h->ref_count[0])
2810 memset(&h->default_ref_list[0][index], 0, sizeof(Picture)*(h->ref_count[0] - index));
2813 if(h->slice_type==B_TYPE){
2815 //FIXME second field balh
2821 static int decode_ref_pic_list_reordering(H264Context *h){
2822 MpegEncContext * const s = &h->s;
2825 if(h->slice_type==I_TYPE || h->slice_type==SI_TYPE) return 0; //FIXME move beofre func
2827 for(list=0; list<2; list++){
2828 memcpy(h->ref_list[list], h->default_ref_list[list], sizeof(Picture)*h->ref_count[list]);
2830 if(get_bits1(&s->gb)){
2831 int pred= h->curr_pic_num;
2834 for(index=0; ; index++){
2835 int reordering_of_pic_nums_idc= get_ue_golomb(&s->gb);
2839 if(reordering_of_pic_nums_idc==3)
2842 if(index >= h->ref_count[list]){
2843 av_log(h->s.avctx, AV_LOG_ERROR, "reference count overflow\n");
2847 if(reordering_of_pic_nums_idc<3){
2848 if(reordering_of_pic_nums_idc<2){
2849 const int abs_diff_pic_num= get_ue_golomb(&s->gb) + 1;
2851 if(abs_diff_pic_num >= h->max_pic_num){
2852 av_log(h->s.avctx, AV_LOG_ERROR, "abs_diff_pic_num overflow\n");
2856 if(reordering_of_pic_nums_idc == 0) pred-= abs_diff_pic_num;
2857 else pred+= abs_diff_pic_num;
2858 pred &= h->max_pic_num - 1;
2860 for(i= h->ref_count[list]-1; i>=index; i--){
2861 if(h->ref_list[list][i].pic_id == pred && h->ref_list[list][i].long_ref==0)
2865 pic_id= get_ue_golomb(&s->gb); //long_term_pic_idx
2867 for(i= h->ref_count[list]-1; i>=index; i--){
2868 if(h->ref_list[list][i].pic_id == pic_id && h->ref_list[list][i].long_ref==1)
2874 av_log(h->s.avctx, AV_LOG_ERROR, "reference picture missing during reorder\n");
2875 memset(&h->ref_list[list][index], 0, sizeof(Picture)); //FIXME
2876 }else if(i > index){
2877 Picture tmp= h->ref_list[list][i];
2878 for(; i>index; i--){
2879 h->ref_list[list][i]= h->ref_list[list][i-1];
2881 h->ref_list[list][index]= tmp;
2884 av_log(h->s.avctx, AV_LOG_ERROR, "illegal reordering_of_pic_nums_idc\n");
2890 if(h->slice_type!=B_TYPE) break;
2893 if(h->slice_type==B_TYPE && !h->direct_spatial_mv_pred)
2894 direct_dist_scale_factor(h);
2898 static int pred_weight_table(H264Context *h){
2899 MpegEncContext * const s = &h->s;
2902 h->luma_log2_weight_denom= get_ue_golomb(&s->gb);
2903 h->chroma_log2_weight_denom= get_ue_golomb(&s->gb);
2905 for(list=0; list<2; list++){
2906 for(i=0; i<h->ref_count[list]; i++){
2907 int luma_weight_flag, chroma_weight_flag;
2909 luma_weight_flag= get_bits1(&s->gb);
2910 if(luma_weight_flag){
2911 h->luma_weight[list][i]= get_se_golomb(&s->gb);
2912 h->luma_offset[list][i]= get_se_golomb(&s->gb);
2915 chroma_weight_flag= get_bits1(&s->gb);
2916 if(chroma_weight_flag){
2919 h->chroma_weight[list][i][j]= get_se_golomb(&s->gb);
2920 h->chroma_offset[list][i][j]= get_se_golomb(&s->gb);
2924 if(h->slice_type != B_TYPE) break;
2930 * instantaneous decoder refresh.
2932 static void idr(H264Context *h){
2935 #define CHECK_DELAY(pic) \
2936 for(j = 0; h->delayed_pic[j]; j++) \
2937 if(pic == h->delayed_pic[j]){ \
2942 for(i=0; i<h->long_ref_count; i++){
2943 h->long_ref[i]->reference=0;
2944 CHECK_DELAY(h->long_ref[i]);
2945 h->long_ref[i]= NULL;
2947 h->long_ref_count=0;
2949 for(i=0; i<h->short_ref_count; i++){
2950 h->short_ref[i]->reference=0;
2951 CHECK_DELAY(h->short_ref[i]);
2952 h->short_ref[i]= NULL;
2954 h->short_ref_count=0;
2960 * @return the removed picture or NULL if an error occures
2962 static Picture * remove_short(H264Context *h, int frame_num){
2963 MpegEncContext * const s = &h->s;
2966 if(s->avctx->debug&FF_DEBUG_MMCO)
2967 av_log(h->s.avctx, AV_LOG_DEBUG, "remove short %d count %d\n", frame_num, h->short_ref_count);
2969 for(i=0; i<h->short_ref_count; i++){
2970 Picture *pic= h->short_ref[i];
2971 if(s->avctx->debug&FF_DEBUG_MMCO)
2972 av_log(h->s.avctx, AV_LOG_DEBUG, "%d %d %p\n", i, pic->frame_num, pic);
2973 if(pic->frame_num == frame_num){
2974 h->short_ref[i]= NULL;
2975 memmove(&h->short_ref[i], &h->short_ref[i+1], (h->short_ref_count - i - 1)*sizeof(Picture*));
2976 h->short_ref_count--;
2985 * @return the removed picture or NULL if an error occures
2987 static Picture * remove_long(H264Context *h, int i){
2990 if(i >= h->long_ref_count) return NULL;
2991 pic= h->long_ref[i];
2992 if(pic==NULL) return NULL;
2994 h->long_ref[i]= NULL;
2995 memmove(&h->long_ref[i], &h->long_ref[i+1], (h->long_ref_count - i - 1)*sizeof(Picture*));
2996 h->long_ref_count--;
3002 * Executes the reference picture marking (memory management control operations).
3004 static int execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count){
3005 MpegEncContext * const s = &h->s;
3007 int current_is_long=0;
3010 if((s->avctx->debug&FF_DEBUG_MMCO) && mmco_count==0)
3011 av_log(h->s.avctx, AV_LOG_DEBUG, "no mmco here\n");
3013 for(i=0; i<mmco_count; i++){
3014 if(s->avctx->debug&FF_DEBUG_MMCO)
3015 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);
3017 switch(mmco[i].opcode){
3018 case MMCO_SHORT2UNUSED:
3019 pic= remove_short(h, mmco[i].short_frame_num);
3020 if(pic==NULL) return -1;
3023 case MMCO_SHORT2LONG:
3024 pic= remove_long(h, mmco[i].long_index);
3025 if(pic) pic->reference=0;
3027 h->long_ref[ mmco[i].long_index ]= remove_short(h, mmco[i].short_frame_num);
3028 h->long_ref[ mmco[i].long_index ]->long_ref=1;
3030 case MMCO_LONG2UNUSED:
3031 pic= remove_long(h, mmco[i].long_index);
3032 if(pic==NULL) return -1;
3036 pic= remove_long(h, mmco[i].long_index);
3037 if(pic) pic->reference=0;
3039 h->long_ref[ mmco[i].long_index ]= s->current_picture_ptr;
3040 h->long_ref[ mmco[i].long_index ]->long_ref=1;
3041 h->long_ref_count++;
3045 case MMCO_SET_MAX_LONG:
3046 assert(mmco[i].long_index <= 16);
3047 while(mmco[i].long_index < h->long_ref_count){
3048 pic= remove_long(h, mmco[i].long_index);
3051 while(mmco[i].long_index > h->long_ref_count){
3052 h->long_ref[ h->long_ref_count++ ]= NULL;
3056 while(h->short_ref_count){
3057 pic= remove_short(h, h->short_ref[0]->frame_num);
3060 while(h->long_ref_count){
3061 pic= remove_long(h, h->long_ref_count-1);
3069 if(!current_is_long){
3070 pic= remove_short(h, s->current_picture_ptr->frame_num);
3073 av_log(h->s.avctx, AV_LOG_ERROR, "illegal short term buffer state detected\n");
3076 if(h->short_ref_count)
3077 memmove(&h->short_ref[1], &h->short_ref[0], h->short_ref_count*sizeof(Picture*));
3079 h->short_ref[0]= s->current_picture_ptr;
3080 h->short_ref[0]->long_ref=0;
3081 h->short_ref_count++;
3087 static int decode_ref_pic_marking(H264Context *h){
3088 MpegEncContext * const s = &h->s;
3091 if(h->nal_unit_type == NAL_IDR_SLICE){ //FIXME fields
3092 s->broken_link= get_bits1(&s->gb) -1;
3093 h->mmco[0].long_index= get_bits1(&s->gb) - 1; // current_long_term_idx
3094 if(h->mmco[0].long_index == -1)
3097 h->mmco[0].opcode= MMCO_LONG;
3101 if(get_bits1(&s->gb)){ // adaptive_ref_pic_marking_mode_flag
3102 for(i= 0; i<MAX_MMCO_COUNT; i++) {
3103 MMCOOpcode opcode= get_ue_golomb(&s->gb);;
3105 h->mmco[i].opcode= opcode;
3106 if(opcode==MMCO_SHORT2UNUSED || opcode==MMCO_SHORT2LONG){
3107 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
3108 /* if(h->mmco[i].short_frame_num >= h->short_ref_count || h->short_ref[ h->mmco[i].short_frame_num ] == NULL){
3109 fprintf(stderr, "illegal short ref in memory management control operation %d\n", mmco);
3113 if(opcode==MMCO_SHORT2LONG || opcode==MMCO_LONG2UNUSED || opcode==MMCO_LONG || opcode==MMCO_SET_MAX_LONG){
3114 h->mmco[i].long_index= get_ue_golomb(&s->gb);
3115 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){
3116 av_log(h->s.avctx, AV_LOG_ERROR, "illegal long ref in memory management control operation %d\n", opcode);
3121 if(opcode > MMCO_LONG){
3122 av_log(h->s.avctx, AV_LOG_ERROR, "illegal memory management control operation %d\n", opcode);
3125 if(opcode == MMCO_END)
3130 assert(h->long_ref_count + h->short_ref_count <= h->sps.ref_frame_count);
3132 if(h->long_ref_count + h->short_ref_count == h->sps.ref_frame_count){ //FIXME fields
3133 h->mmco[0].opcode= MMCO_SHORT2UNUSED;
3134 h->mmco[0].short_frame_num= h->short_ref[ h->short_ref_count - 1 ]->frame_num;
3144 static int init_poc(H264Context *h){
3145 MpegEncContext * const s = &h->s;
3146 const int max_frame_num= 1<<h->sps.log2_max_frame_num;
3149 if(h->nal_unit_type == NAL_IDR_SLICE){
3150 h->frame_num_offset= 0;
3152 if(h->frame_num < h->prev_frame_num)
3153 h->frame_num_offset= h->prev_frame_num_offset + max_frame_num;
3155 h->frame_num_offset= h->prev_frame_num_offset;
3158 if(h->sps.poc_type==0){
3159 const int max_poc_lsb= 1<<h->sps.log2_max_poc_lsb;
3161 if (h->poc_lsb < h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb >= max_poc_lsb/2)
3162 h->poc_msb = h->prev_poc_msb + max_poc_lsb;
3163 else if(h->poc_lsb > h->prev_poc_lsb && h->prev_poc_lsb - h->poc_lsb < -max_poc_lsb/2)
3164 h->poc_msb = h->prev_poc_msb - max_poc_lsb;
3166 h->poc_msb = h->prev_poc_msb;
3167 //printf("poc: %d %d\n", h->poc_msb, h->poc_lsb);
3169 field_poc[1] = h->poc_msb + h->poc_lsb;
3170 if(s->picture_structure == PICT_FRAME)
3171 field_poc[1] += h->delta_poc_bottom;
3172 }else if(h->sps.poc_type==1){
3173 int abs_frame_num, expected_delta_per_poc_cycle, expectedpoc;
3176 if(h->sps.poc_cycle_length != 0)
3177 abs_frame_num = h->frame_num_offset + h->frame_num;
3181 if(h->nal_ref_idc==0 && abs_frame_num > 0)
3184 expected_delta_per_poc_cycle = 0;
3185 for(i=0; i < h->sps.poc_cycle_length; i++)
3186 expected_delta_per_poc_cycle += h->sps.offset_for_ref_frame[ i ]; //FIXME integrate during sps parse
3188 if(abs_frame_num > 0){
3189 int poc_cycle_cnt = (abs_frame_num - 1) / h->sps.poc_cycle_length;
3190 int frame_num_in_poc_cycle = (abs_frame_num - 1) % h->sps.poc_cycle_length;
3192 expectedpoc = poc_cycle_cnt * expected_delta_per_poc_cycle;
3193 for(i = 0; i <= frame_num_in_poc_cycle; i++)
3194 expectedpoc = expectedpoc + h->sps.offset_for_ref_frame[ i ];
3198 if(h->nal_ref_idc == 0)
3199 expectedpoc = expectedpoc + h->sps.offset_for_non_ref_pic;
3201 field_poc[0] = expectedpoc + h->delta_poc[0];
3202 field_poc[1] = field_poc[0] + h->sps.offset_for_top_to_bottom_field;
3204 if(s->picture_structure == PICT_FRAME)
3205 field_poc[1] += h->delta_poc[1];
3208 if(h->nal_unit_type == NAL_IDR_SLICE){
3211 if(h->nal_ref_idc) poc= 2*(h->frame_num_offset + h->frame_num);
3212 else poc= 2*(h->frame_num_offset + h->frame_num) - 1;
3218 if(s->picture_structure != PICT_BOTTOM_FIELD)
3219 s->current_picture_ptr->field_poc[0]= field_poc[0];
3220 if(s->picture_structure != PICT_TOP_FIELD)
3221 s->current_picture_ptr->field_poc[1]= field_poc[1];
3222 if(s->picture_structure == PICT_FRAME) // FIXME field pix?
3223 s->current_picture_ptr->poc= FFMIN(field_poc[0], field_poc[1]);
3229 * decodes a slice header.
3230 * this will allso call MPV_common_init() and frame_start() as needed
3232 static int decode_slice_header(H264Context *h){
3233 MpegEncContext * const s = &h->s;
3234 int first_mb_in_slice, pps_id;
3235 int num_ref_idx_active_override_flag;
3236 static const uint8_t slice_type_map[5]= {P_TYPE, B_TYPE, I_TYPE, SP_TYPE, SI_TYPE};
3238 s->current_picture.reference= h->nal_ref_idc != 0;
3240 first_mb_in_slice= get_ue_golomb(&s->gb);
3242 h->slice_type= get_ue_golomb(&s->gb);
3243 if(h->slice_type > 9){
3244 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);
3247 if(h->slice_type > 4){
3249 h->slice_type_fixed=1;
3251 h->slice_type_fixed=0;
3253 h->slice_type= slice_type_map[ h->slice_type ];
3255 s->pict_type= h->slice_type; // to make a few old func happy, its wrong though
3257 pps_id= get_ue_golomb(&s->gb);
3259 av_log(h->s.avctx, AV_LOG_ERROR, "pps_id out of range\n");
3262 h->pps= h->pps_buffer[pps_id];
3263 if(h->pps.slice_group_count == 0){
3264 av_log(h->s.avctx, AV_LOG_ERROR, "non existing PPS referenced\n");
3268 h->sps= h->sps_buffer[ h->pps.sps_id ];
3269 if(h->sps.log2_max_frame_num == 0){
3270 av_log(h->s.avctx, AV_LOG_ERROR, "non existing SPS referenced\n");
3274 s->mb_width= h->sps.mb_width;
3275 s->mb_height= h->sps.mb_height;
3277 h->b_stride= s->mb_width*4 + 1;
3278 h->b8_stride= s->mb_width*2 + 1;
3280 s->resync_mb_x = s->mb_x = first_mb_in_slice % s->mb_width;
3281 s->resync_mb_y = s->mb_y = first_mb_in_slice / s->mb_width; //FIXME AFFW
3283 s->width = 16*s->mb_width - 2*(h->sps.crop_left + h->sps.crop_right );
3284 if(h->sps.frame_mbs_only_flag)
3285 s->height= 16*s->mb_height - 2*(h->sps.crop_top + h->sps.crop_bottom);
3287 s->height= 16*s->mb_height - 4*(h->sps.crop_top + h->sps.crop_bottom); //FIXME recheck
3289 if (s->context_initialized
3290 && ( s->width != s->avctx->width || s->height != s->avctx->height)) {
3294 if (!s->context_initialized) {
3295 if (MPV_common_init(s) < 0)
3300 s->avctx->width = s->width;
3301 s->avctx->height = s->height;
3302 s->avctx->sample_aspect_ratio= h->sps.sar;
3304 if(h->sps.timing_info_present_flag && h->sps.fixed_frame_rate_flag){
3305 s->avctx->frame_rate = h->sps.time_scale;
3306 s->avctx->frame_rate_base = h->sps.num_units_in_tick;
3310 if(h->slice_num == 0){
3314 s->current_picture_ptr->frame_num= //FIXME frame_num cleanup
3315 h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
3317 if(h->sps.frame_mbs_only_flag){
3318 s->picture_structure= PICT_FRAME;
3320 if(get_bits1(&s->gb)) //field_pic_flag
3321 s->picture_structure= PICT_TOP_FIELD + get_bits1(&s->gb); //bottom_field_flag
3323 s->picture_structure= PICT_FRAME;
3326 if(s->picture_structure==PICT_FRAME){
3327 h->curr_pic_num= h->frame_num;
3328 h->max_pic_num= 1<< h->sps.log2_max_frame_num;
3330 h->curr_pic_num= 2*h->frame_num;
3331 h->max_pic_num= 1<<(h->sps.log2_max_frame_num + 1);
3334 if(h->nal_unit_type == NAL_IDR_SLICE){
3335 get_ue_golomb(&s->gb); /* idr_pic_id */
3338 if(h->sps.poc_type==0){
3339 h->poc_lsb= get_bits(&s->gb, h->sps.log2_max_poc_lsb);
3341 if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME){
3342 h->delta_poc_bottom= get_se_golomb(&s->gb);
3346 if(h->sps.poc_type==1 && !h->sps.delta_pic_order_always_zero_flag){
3347 h->delta_poc[0]= get_se_golomb(&s->gb);
3349 if(h->pps.pic_order_present==1 && s->picture_structure==PICT_FRAME)
3350 h->delta_poc[1]= get_se_golomb(&s->gb);
3355 if(h->pps.redundant_pic_cnt_present){
3356 h->redundant_pic_count= get_ue_golomb(&s->gb);
3359 //set defaults, might be overriden a few line later
3360 h->ref_count[0]= h->pps.ref_count[0];
3361 h->ref_count[1]= h->pps.ref_count[1];
3363 if(h->slice_type == P_TYPE || h->slice_type == SP_TYPE || h->slice_type == B_TYPE){
3364 if(h->slice_type == B_TYPE){
3365 h->direct_spatial_mv_pred= get_bits1(&s->gb);
3367 num_ref_idx_active_override_flag= get_bits1(&s->gb);
3369 if(num_ref_idx_active_override_flag){
3370 h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
3371 if(h->slice_type==B_TYPE)
3372 h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
3374 if(h->ref_count[0] > 32 || h->ref_count[1] > 32){
3375 av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
3381 if(h->slice_num == 0){
3382 fill_default_ref_list(h);
3385 decode_ref_pic_list_reordering(h);
3387 if( (h->pps.weighted_pred && (h->slice_type == P_TYPE || h->slice_type == SP_TYPE ))
3388 || (h->pps.weighted_bipred_idc==1 && h->slice_type==B_TYPE ) )
3389 pred_weight_table(h);
3391 if(s->current_picture.reference)
3392 decode_ref_pic_marking(h);
3394 if( h->slice_type != I_TYPE && h->slice_type != SI_TYPE && h->pps.cabac )
3395 h->cabac_init_idc = get_ue_golomb(&s->gb);
3397 h->last_qscale_diff = 0;
3398 s->qscale = h->pps.init_qp + get_se_golomb(&s->gb);
3399 if(s->qscale<0 || s->qscale>51){
3400 av_log(s->avctx, AV_LOG_ERROR, "QP %d out of range\n", s->qscale);
3403 h->chroma_qp = get_chroma_qp(h, s->qscale);
3404 //FIXME qscale / qp ... stuff
3405 if(h->slice_type == SP_TYPE){
3406 get_bits1(&s->gb); /* sp_for_switch_flag */
3408 if(h->slice_type==SP_TYPE || h->slice_type == SI_TYPE){
3409 get_se_golomb(&s->gb); /* slice_qs_delta */
3412 h->deblocking_filter = 1;
3413 h->slice_alpha_c0_offset = 0;
3414 h->slice_beta_offset = 0;
3415 if( h->pps.deblocking_filter_parameters_present ) {
3416 h->deblocking_filter= get_ue_golomb(&s->gb);
3417 if(h->deblocking_filter < 2)
3418 h->deblocking_filter^= 1; // 1<->0
3420 if( h->deblocking_filter ) {
3421 h->slice_alpha_c0_offset = get_se_golomb(&s->gb) << 1;
3422 h->slice_beta_offset = get_se_golomb(&s->gb) << 1;
3427 if( h->pps.num_slice_groups > 1 && h->pps.mb_slice_group_map_type >= 3 && h->pps.mb_slice_group_map_type <= 5)
3428 slice_group_change_cycle= get_bits(&s->gb, ?);
3433 if(s->avctx->debug&FF_DEBUG_PICT_INFO){
3434 av_log(h->s.avctx, AV_LOG_DEBUG, "slice:%d mb:%d %c pps:%d frame:%d poc:%d/%d ref:%d/%d qp:%d loop:%d\n",
3435 h->slice_num, first_mb_in_slice,
3436 av_get_pict_type_char(h->slice_type),
3437 pps_id, h->frame_num,
3438 s->current_picture_ptr->field_poc[0], s->current_picture_ptr->field_poc[1],
3439 h->ref_count[0], h->ref_count[1],
3441 h->deblocking_filter
3451 static inline int get_level_prefix(GetBitContext *gb){
3455 OPEN_READER(re, gb);
3456 UPDATE_CACHE(re, gb);
3457 buf=GET_CACHE(re, gb);
3459 log= 32 - av_log2(buf);
3461 print_bin(buf>>(32-log), log);
3462 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__);
3465 LAST_SKIP_BITS(re, gb, log);
3466 CLOSE_READER(re, gb);
3472 * decodes a residual block.
3473 * @param n block index
3474 * @param scantable scantable
3475 * @param max_coeff number of coefficients in the block
3476 * @return <0 if an error occured
3478 static int decode_residual(H264Context *h, GetBitContext *gb, DCTELEM *block, int n, const uint8_t *scantable, int qp, int max_coeff){
3479 MpegEncContext * const s = &h->s;
3480 const uint16_t *qmul= dequant_coeff[qp];
3481 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};
3482 int level[16], run[16];
3483 int suffix_length, zeros_left, coeff_num, coeff_token, total_coeff, i, trailing_ones;
3485 //FIXME put trailing_onex into the context
3487 if(n == CHROMA_DC_BLOCK_INDEX){
3488 coeff_token= get_vlc2(gb, chroma_dc_coeff_token_vlc.table, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 1);
3489 total_coeff= coeff_token>>2;
3491 if(n == LUMA_DC_BLOCK_INDEX){
3492 total_coeff= pred_non_zero_count(h, 0);
3493 coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
3494 total_coeff= coeff_token>>2;
3496 total_coeff= pred_non_zero_count(h, n);
3497 coeff_token= get_vlc2(gb, coeff_token_vlc[ coeff_token_table_index[total_coeff] ].table, COEFF_TOKEN_VLC_BITS, 2);
3498 total_coeff= coeff_token>>2;
3499 h->non_zero_count_cache[ scan8[n] ]= total_coeff;
3503 //FIXME set last_non_zero?
3508 trailing_ones= coeff_token&3;
3509 tprintf("trailing:%d, total:%d\n", trailing_ones, total_coeff);
3510 assert(total_coeff<=16);
3512 for(i=0; i<trailing_ones; i++){
3513 level[i]= 1 - 2*get_bits1(gb);
3516 suffix_length= total_coeff > 10 && trailing_ones < 3;
3518 for(; i<total_coeff; i++){
3519 const int prefix= get_level_prefix(gb);
3520 int level_code, mask;
3522 if(prefix<14){ //FIXME try to build a large unified VLC table for all this
3524 level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
3526 level_code= (prefix<<suffix_length); //part
3527 }else if(prefix==14){
3529 level_code= (prefix<<suffix_length) + get_bits(gb, suffix_length); //part
3531 level_code= prefix + get_bits(gb, 4); //part
3532 }else if(prefix==15){
3533 level_code= (prefix<<suffix_length) + get_bits(gb, 12); //part
3534 if(suffix_length==0) level_code+=15; //FIXME doesnt make (much)sense
3536 av_log(h->s.avctx, AV_LOG_ERROR, "prefix too large at %d %d\n", s->mb_x, s->mb_y);
3540 if(i==trailing_ones && i<3) level_code+= 2; //FIXME split first iteration
3542 mask= -(level_code&1);
3543 level[i]= (((2+level_code)>>1) ^ mask) - mask;
3545 if(suffix_length==0) suffix_length=1; //FIXME split first iteration
3548 if(ABS(level[i]) > (3<<(suffix_length-1)) && suffix_length<6) suffix_length++;
3550 if((2+level_code)>>1) > (3<<(suffix_length-1)) && suffix_length<6) suffix_length++;
3551 /* ? == prefix > 2 or sth */
3553 tprintf("level: %d suffix_length:%d\n", level[i], suffix_length);
3556 if(total_coeff == max_coeff)
3559 if(n == CHROMA_DC_BLOCK_INDEX)
3560 zeros_left= get_vlc2(gb, chroma_dc_total_zeros_vlc[ total_coeff-1 ].table, CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 1);
3562 zeros_left= get_vlc2(gb, total_zeros_vlc[ total_coeff-1 ].table, TOTAL_ZEROS_VLC_BITS, 1);
3565 for(i=0; i<total_coeff-1; i++){
3568 else if(zeros_left < 7){
3569 run[i]= get_vlc2(gb, run_vlc[zeros_left-1].table, RUN_VLC_BITS, 1);
3571 run[i]= get_vlc2(gb, run7_vlc.table, RUN7_VLC_BITS, 2);
3573 zeros_left -= run[i];
3577 av_log(h->s.avctx, AV_LOG_ERROR, "negative number of zero coeffs at %d %d\n", s->mb_x, s->mb_y);
3581 for(; i<total_coeff-1; i++){
3589 for(i=total_coeff-1; i>=0; i--){ //FIXME merge into rundecode?
3592 coeff_num += run[i] + 1; //FIXME add 1 earlier ?
3593 j= scantable[ coeff_num ];
3598 for(i=total_coeff-1; i>=0; i--){ //FIXME merge into rundecode?
3601 coeff_num += run[i] + 1; //FIXME add 1 earlier ?
3602 j= scantable[ coeff_num ];
3604 block[j]= level[i] * qmul[j];
3605 // printf("%d %d ", block[j], qmul[j]);
3612 * decodes a P_SKIP or B_SKIP macroblock
3614 static void decode_mb_skip(H264Context *h){
3615 MpegEncContext * const s = &h->s;
3616 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
3619 memset(h->non_zero_count[mb_xy], 0, 16);
3620 memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui
3622 if( h->slice_type == B_TYPE )
3624 // just for fill_caches. pred_direct_motion will set the real mb_type
3625 mb_type= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2|MB_TYPE_SKIP;
3628 fill_caches(h, mb_type); //FIXME check what is needed and what not ...
3629 pred_direct_motion(h, &mb_type);