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 transform_bypass; ///< qpprime_y_zero_transform_bypass_flag
65 int log2_max_frame_num; ///< log2_max_frame_num_minus4 + 4
66 int poc_type; ///< pic_order_cnt_type
67 int log2_max_poc_lsb; ///< log2_max_pic_order_cnt_lsb_minus4
68 int delta_pic_order_always_zero_flag;
69 int offset_for_non_ref_pic;
70 int offset_for_top_to_bottom_field;
71 int poc_cycle_length; ///< num_ref_frames_in_pic_order_cnt_cycle
72 int ref_frame_count; ///< num_ref_frames
73 int gaps_in_frame_num_allowed_flag;
74 int mb_width; ///< frame_width_in_mbs_minus1 + 1
75 int mb_height; ///< frame_height_in_mbs_minus1 + 1
76 int frame_mbs_only_flag;
77 int mb_aff; ///<mb_adaptive_frame_field_flag
78 int direct_8x8_inference_flag;
79 int crop; ///< frame_cropping_flag
80 int crop_left; ///< frame_cropping_rect_left_offset
81 int crop_right; ///< frame_cropping_rect_right_offset
82 int crop_top; ///< frame_cropping_rect_top_offset
83 int crop_bottom; ///< frame_cropping_rect_bottom_offset
84 int vui_parameters_present_flag;
86 int timing_info_present_flag;
87 uint32_t num_units_in_tick;
89 int fixed_frame_rate_flag;
90 short offset_for_ref_frame[256]; //FIXME dyn aloc?
91 int bitstream_restriction_flag;
92 int num_reorder_frames;
96 * Picture parameter set
100 int cabac; ///< entropy_coding_mode_flag
101 int pic_order_present; ///< pic_order_present_flag
102 int slice_group_count; ///< num_slice_groups_minus1 + 1
103 int mb_slice_group_map_type;
104 int ref_count[2]; ///< num_ref_idx_l0/1_active_minus1 + 1
105 int weighted_pred; ///< weighted_pred_flag
106 int weighted_bipred_idc;
107 int init_qp; ///< pic_init_qp_minus26 + 26
108 int init_qs; ///< pic_init_qs_minus26 + 26
109 int chroma_qp_index_offset;
110 int deblocking_filter_parameters_present; ///< deblocking_filter_parameters_present_flag
111 int constrained_intra_pred; ///< constrained_intra_pred_flag
112 int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag
113 int transform_8x8_mode; ///< transform_8x8_mode_flag
117 * Memory management control operation opcode.
119 typedef enum MMCOOpcode{
130 * Memory management control operation.
141 typedef struct H264Context{
149 #define NAL_IDR_SLICE 5
153 #define NAL_PICTURE_DELIMITER 9
154 #define NAL_FILTER_DATA 10
155 uint8_t *rbsp_buffer;
156 int rbsp_buffer_size;
159 * Used to parse AVC variant of h264
161 int is_avc; ///< this flag is != 0 if codec is avc1
162 int got_avcC; ///< flag used to parse avcC data only once
163 int nal_length_size; ///< Number of bytes used for nal length (1, 2 or 4)
167 int prev_mb_skipped; //FIXME remove (IMHO not used)
170 int chroma_pred_mode;
171 int intra16x16_pred_mode;
176 int8_t intra4x4_pred_mode_cache[5*8];
177 int8_t (*intra4x4_pred_mode)[8];
178 void (*pred4x4 [9+3])(uint8_t *src, uint8_t *topright, int stride);//FIXME move to dsp?
179 void (*pred8x8l [9+3])(uint8_t *src, int topleft, int topright, int stride);
180 void (*pred8x8 [4+3])(uint8_t *src, int stride);
181 void (*pred16x16[4+3])(uint8_t *src, int stride);
182 unsigned int topleft_samples_available;
183 unsigned int top_samples_available;
184 unsigned int topright_samples_available;
185 unsigned int left_samples_available;
186 uint8_t (*top_borders[2])[16+2*8];
187 uint8_t left_border[2*(17+2*9)];
190 * non zero coeff count cache.
191 * is 64 if not available.
193 uint8_t non_zero_count_cache[6*8] __align8;
194 uint8_t (*non_zero_count)[16];
197 * Motion vector cache.
199 int16_t mv_cache[2][5*8][2] __align8;
200 int8_t ref_cache[2][5*8] __align8;
201 #define LIST_NOT_USED -1 //FIXME rename?
202 #define PART_NOT_AVAILABLE -2
205 * is 1 if the specific list MV&references are set to 0,0,-2.
207 int mv_cache_clean[2];
210 * number of neighbors (top and/or left) that used 8x8 dct
212 int neighbor_transform_size;
215 * block_offset[ 0..23] for frame macroblocks
216 * block_offset[24..47] for field macroblocks
218 int block_offset[2*(16+8)];
220 uint32_t *mb2b_xy; //FIXME are these 4 a good idea?
222 int b_stride; //FIXME use s->b4_stride
228 int unknown_svq3_flag;
229 int next_slice_index;
231 SPS sps_buffer[MAX_SPS_COUNT];
232 SPS sps; ///< current sps
234 PPS pps_buffer[MAX_PPS_COUNT];
238 PPS pps; //FIXME move to Picture perhaps? (->no) do we need that?
240 uint16_t (*dequant4_coeff)[16]; // FIXME quant matrices should be per SPS or PPS
241 uint16_t (*dequant8_coeff)[64];
244 uint8_t *slice_table_base;
245 uint8_t *slice_table; ///< slice_table_base + mb_stride + 1
247 int slice_type_fixed;
249 //interlacing specific flags
251 int mb_field_decoding_flag;
258 int delta_poc_bottom;
261 int prev_poc_msb; ///< poc_msb of the last reference pic for POC type 0
262 int prev_poc_lsb; ///< poc_lsb of the last reference pic for POC type 0
263 int frame_num_offset; ///< for POC type 2
264 int prev_frame_num_offset; ///< for POC type 2
265 int prev_frame_num; ///< frame_num of the last pic for POC type 1/2
268 * frame_num for frames or 2*frame_num for field pics.
273 * max_frame_num or 2*max_frame_num for field pics.
277 //Weighted pred stuff
279 int use_weight_chroma;
280 int luma_log2_weight_denom;
281 int chroma_log2_weight_denom;
282 int luma_weight[2][16];
283 int luma_offset[2][16];
284 int chroma_weight[2][16][2];
285 int chroma_offset[2][16][2];
286 int implicit_weight[16][16];
289 int deblocking_filter; ///< disable_deblocking_filter_idc with 1<->0
290 int slice_alpha_c0_offset;
291 int slice_beta_offset;
293 int redundant_pic_count;
295 int direct_spatial_mv_pred;
296 int dist_scale_factor[16];
297 int map_col_to_list0[2][16];
300 * num_ref_idx_l0/1_active_minus1 + 1
302 int ref_count[2];// FIXME split for AFF
303 Picture *short_ref[32];
304 Picture *long_ref[32];
305 Picture default_ref_list[2][32];
306 Picture ref_list[2][32]; //FIXME size?
307 Picture field_ref_list[2][32]; //FIXME size?
308 Picture *delayed_pic[16]; //FIXME size?
309 Picture *delayed_output_pic;
312 * memory management control operations buffer.
314 MMCO mmco[MAX_MMCO_COUNT];
317 int long_ref_count; ///< number of actual long term references
318 int short_ref_count; ///< number of actual short term references
321 GetBitContext intra_gb;
322 GetBitContext inter_gb;
323 GetBitContext *intra_gb_ptr;
324 GetBitContext *inter_gb_ptr;
326 DCTELEM mb[16*24] __align8;
332 uint8_t cabac_state[460];
335 /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0,1,2), 0x0? luma_cbp */
339 /* chroma_pred_mode for i4x4 or i16x16, else 0 */
340 uint8_t *chroma_pred_mode_table;
341 int last_qscale_diff;
342 int16_t (*mvd_table[2])[2];
343 int16_t mvd_cache[2][5*8][2] __align8;
344 uint8_t *direct_table;
345 uint8_t direct_cache[5*8];
347 uint8_t zigzag_scan[16];
348 uint8_t field_scan[16];
349 const uint8_t *zigzag_scan_q0;
350 const uint8_t *field_scan_q0;
353 static VLC coeff_token_vlc[4];
354 static VLC chroma_dc_coeff_token_vlc;
356 static VLC total_zeros_vlc[15];
357 static VLC chroma_dc_total_zeros_vlc[3];
359 static VLC run_vlc[6];
362 static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
363 static void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
364 static void filter_mb( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr, unsigned int linesize, unsigned int uvlinesize);
366 static inline uint32_t pack16to32(int a, int b){
367 #ifdef WORDS_BIGENDIAN
368 return (b&0xFFFF) + (a<<16);
370 return (a&0xFFFF) + (b<<16);
376 * @param h height of the rectangle, should be a constant
377 * @param w width of the rectangle, should be a constant
378 * @param size the size of val (1 or 4), should be a constant
380 static inline void fill_rectangle(void *vp, int w, int h, int stride, uint32_t val, int size){ //FIXME ensure this IS inlined
381 uint8_t *p= (uint8_t*)vp;
382 assert(size==1 || size==4);
387 assert((((int)vp)&(FFMIN(w, STRIDE_ALIGN)-1)) == 0);
388 assert((stride&(w-1))==0);
389 //FIXME check what gcc generates for 64 bit on x86 and possibly write a 32 bit ver of it
392 *(uint16_t*)(p + stride)= size==4 ? val : val*0x0101;
393 }else if(w==2 && h==4){
394 *(uint16_t*)(p + 0*stride)=
395 *(uint16_t*)(p + 1*stride)=
396 *(uint16_t*)(p + 2*stride)=
397 *(uint16_t*)(p + 3*stride)= size==4 ? val : val*0x0101;
398 }else if(w==4 && h==1){
399 *(uint32_t*)(p + 0*stride)= size==4 ? val : val*0x01010101;
400 }else if(w==4 && h==2){
401 *(uint32_t*)(p + 0*stride)=
402 *(uint32_t*)(p + 1*stride)= size==4 ? val : val*0x01010101;
403 }else if(w==4 && h==4){
404 *(uint32_t*)(p + 0*stride)=
405 *(uint32_t*)(p + 1*stride)=
406 *(uint32_t*)(p + 2*stride)=
407 *(uint32_t*)(p + 3*stride)= size==4 ? val : val*0x01010101;
408 }else if(w==8 && h==1){
410 *(uint32_t*)(p + 4)= size==4 ? val : val*0x01010101;
411 }else if(w==8 && h==2){
412 *(uint32_t*)(p + 0 + 0*stride)=
413 *(uint32_t*)(p + 4 + 0*stride)=
414 *(uint32_t*)(p + 0 + 1*stride)=
415 *(uint32_t*)(p + 4 + 1*stride)= size==4 ? val : val*0x01010101;
416 }else if(w==8 && h==4){
417 *(uint64_t*)(p + 0*stride)=
418 *(uint64_t*)(p + 1*stride)=
419 *(uint64_t*)(p + 2*stride)=
420 *(uint64_t*)(p + 3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
421 }else if(w==16 && h==2){
422 *(uint64_t*)(p + 0+0*stride)=
423 *(uint64_t*)(p + 8+0*stride)=
424 *(uint64_t*)(p + 0+1*stride)=
425 *(uint64_t*)(p + 8+1*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
426 }else if(w==16 && h==4){
427 *(uint64_t*)(p + 0+0*stride)=
428 *(uint64_t*)(p + 8+0*stride)=
429 *(uint64_t*)(p + 0+1*stride)=
430 *(uint64_t*)(p + 8+1*stride)=
431 *(uint64_t*)(p + 0+2*stride)=
432 *(uint64_t*)(p + 8+2*stride)=
433 *(uint64_t*)(p + 0+3*stride)=
434 *(uint64_t*)(p + 8+3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
439 static inline void fill_caches(H264Context *h, int mb_type, int for_deblock){
440 MpegEncContext * const s = &h->s;
441 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
442 int topleft_xy, top_xy, topright_xy, left_xy[2];
443 int topleft_type, top_type, topright_type, left_type[2];
447 //FIXME deblocking can skip fill_caches much of the time with multiple slices too.
448 // the actual condition is whether we're on the edge of a slice,
449 // and even then the intra and nnz parts are unnecessary.
450 if(for_deblock && h->slice_num == 1)
453 //wow what a mess, why didn't they simplify the interlacing&intra stuff, i can't imagine that these complex rules are worth it
455 top_xy = mb_xy - s->mb_stride;
456 topleft_xy = top_xy - 1;
457 topright_xy= top_xy + 1;
458 left_xy[1] = left_xy[0] = mb_xy-1;
468 const int pair_xy = s->mb_x + (s->mb_y & ~1)*s->mb_stride;
469 const int top_pair_xy = pair_xy - s->mb_stride;
470 const int topleft_pair_xy = top_pair_xy - 1;
471 const int topright_pair_xy = top_pair_xy + 1;
472 const int topleft_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[topleft_pair_xy]);
473 const int top_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[top_pair_xy]);
474 const int topright_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[topright_pair_xy]);
475 const int left_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[pair_xy-1]);
476 const int curr_mb_frame_flag = !IS_INTERLACED(mb_type);
477 const int bottom = (s->mb_y & 1);
478 tprintf("fill_caches: curr_mb_frame_flag:%d, left_mb_frame_flag:%d, topleft_mb_frame_flag:%d, top_mb_frame_flag:%d, topright_mb_frame_flag:%d\n", curr_mb_frame_flag, left_mb_frame_flag, topleft_mb_frame_flag, top_mb_frame_flag, topright_mb_frame_flag);
480 ? !curr_mb_frame_flag // bottom macroblock
481 : (!curr_mb_frame_flag && !top_mb_frame_flag) // top macroblock
483 top_xy -= s->mb_stride;
486 ? !curr_mb_frame_flag // bottom macroblock
487 : (!curr_mb_frame_flag && !topleft_mb_frame_flag) // top macroblock
489 topleft_xy -= s->mb_stride;
492 ? !curr_mb_frame_flag // bottom macroblock
493 : (!curr_mb_frame_flag && !topright_mb_frame_flag) // top macroblock
495 topright_xy -= s->mb_stride;
497 if (left_mb_frame_flag != curr_mb_frame_flag) {
498 left_xy[1] = left_xy[0] = pair_xy - 1;
499 if (curr_mb_frame_flag) {
520 left_xy[1] += s->mb_stride;
533 h->top_mb_xy = top_xy;
534 h->left_mb_xy[0] = left_xy[0];
535 h->left_mb_xy[1] = left_xy[1];
537 topleft_type = h->slice_table[topleft_xy ] < 255 ? s->current_picture.mb_type[topleft_xy] : 0;
538 top_type = h->slice_table[top_xy ] < 255 ? s->current_picture.mb_type[top_xy] : 0;
539 topright_type= h->slice_table[topright_xy] < 255 ? s->current_picture.mb_type[topright_xy]: 0;
540 left_type[0] = h->slice_table[left_xy[0] ] < 255 ? s->current_picture.mb_type[left_xy[0]] : 0;
541 left_type[1] = h->slice_table[left_xy[1] ] < 255 ? s->current_picture.mb_type[left_xy[1]] : 0;
543 topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
544 top_type = h->slice_table[top_xy ] == h->slice_num ? s->current_picture.mb_type[top_xy] : 0;
545 topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
546 left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
547 left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
550 if(IS_INTRA(mb_type)){
551 h->topleft_samples_available=
552 h->top_samples_available=
553 h->left_samples_available= 0xFFFF;
554 h->topright_samples_available= 0xEEEA;
556 if(!IS_INTRA(top_type) && (top_type==0 || h->pps.constrained_intra_pred)){
557 h->topleft_samples_available= 0xB3FF;
558 h->top_samples_available= 0x33FF;
559 h->topright_samples_available= 0x26EA;
562 if(!IS_INTRA(left_type[i]) && (left_type[i]==0 || h->pps.constrained_intra_pred)){
563 h->topleft_samples_available&= 0xDF5F;
564 h->left_samples_available&= 0x5F5F;
568 if(!IS_INTRA(topleft_type) && (topleft_type==0 || h->pps.constrained_intra_pred))
569 h->topleft_samples_available&= 0x7FFF;
571 if(!IS_INTRA(topright_type) && (topright_type==0 || h->pps.constrained_intra_pred))
572 h->topright_samples_available&= 0xFBFF;
574 if(IS_INTRA4x4(mb_type)){
575 if(IS_INTRA4x4(top_type)){
576 h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
577 h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
578 h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
579 h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
582 if(!top_type || (IS_INTER(top_type) && h->pps.constrained_intra_pred))
587 h->intra4x4_pred_mode_cache[4+8*0]=
588 h->intra4x4_pred_mode_cache[5+8*0]=
589 h->intra4x4_pred_mode_cache[6+8*0]=
590 h->intra4x4_pred_mode_cache[7+8*0]= pred;
593 if(IS_INTRA4x4(left_type[i])){
594 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
595 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
598 if(!left_type[i] || (IS_INTER(left_type[i]) && h->pps.constrained_intra_pred))
603 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
604 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
619 //FIXME constraint_intra_pred & partitioning & nnz (lets hope this is just a typo in the spec)
621 h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][4];
622 h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][5];
623 h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][6];
624 h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3];
626 h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][9];
627 h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8];
629 h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][12];
630 h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11];
633 h->non_zero_count_cache[4+8*0]=
634 h->non_zero_count_cache[5+8*0]=
635 h->non_zero_count_cache[6+8*0]=
636 h->non_zero_count_cache[7+8*0]=
638 h->non_zero_count_cache[1+8*0]=
639 h->non_zero_count_cache[2+8*0]=
641 h->non_zero_count_cache[1+8*3]=
642 h->non_zero_count_cache[2+8*3]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
646 for (i=0; i<2; i++) {
648 h->non_zero_count_cache[3+8*1 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[0+2*i]];
649 h->non_zero_count_cache[3+8*2 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[1+2*i]];
650 h->non_zero_count_cache[0+8*1 + 8*i]= h->non_zero_count[left_xy[i]][left_block[4+2*i]];
651 h->non_zero_count_cache[0+8*4 + 8*i]= h->non_zero_count[left_xy[i]][left_block[5+2*i]];
653 h->non_zero_count_cache[3+8*1 + 2*8*i]=
654 h->non_zero_count_cache[3+8*2 + 2*8*i]=
655 h->non_zero_count_cache[0+8*1 + 8*i]=
656 h->non_zero_count_cache[0+8*4 + 8*i]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
663 h->top_cbp = h->cbp_table[top_xy];
664 } else if(IS_INTRA(mb_type)) {
671 h->left_cbp = h->cbp_table[left_xy[0]] & 0x1f0;
672 } else if(IS_INTRA(mb_type)) {
678 h->left_cbp |= ((h->cbp_table[left_xy[0]]>>((left_block[0]&(~1))+1))&0x1) << 1;
681 h->left_cbp |= ((h->cbp_table[left_xy[1]]>>((left_block[2]&(~1))+1))&0x1) << 3;
686 //FIXME direct mb can skip much of this
687 if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
689 for(list=0; list<1+(h->slice_type==B_TYPE); list++){
690 if(!USES_LIST(mb_type, list) && !IS_DIRECT(mb_type) && !h->deblocking_filter){
691 /*if(!h->mv_cache_clean[list]){
692 memset(h->mv_cache [list], 0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
693 memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
694 h->mv_cache_clean[list]= 1;
698 h->mv_cache_clean[list]= 0;
700 if(IS_INTER(top_type)){
701 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
702 const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
703 *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0];
704 *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];
705 *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];
706 *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];
707 h->ref_cache[list][scan8[0] + 0 - 1*8]=
708 h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
709 h->ref_cache[list][scan8[0] + 2 - 1*8]=
710 h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
712 *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]=
713 *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]=
714 *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]=
715 *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;
716 *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
719 //FIXME unify cleanup or sth
720 if(IS_INTER(left_type[0])){
721 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
722 const int b8_xy= h->mb2b8_xy[left_xy[0]] + 1;
723 *(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]];
724 *(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]];
725 h->ref_cache[list][scan8[0] - 1 + 0*8]=
726 h->ref_cache[list][scan8[0] - 1 + 1*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0]>>1)];
728 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 0*8]=
729 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 1*8]= 0;
730 h->ref_cache[list][scan8[0] - 1 + 0*8]=
731 h->ref_cache[list][scan8[0] - 1 + 1*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
734 if(IS_INTER(left_type[1])){
735 const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
736 const int b8_xy= h->mb2b8_xy[left_xy[1]] + 1;
737 *(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]];
738 *(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]];
739 h->ref_cache[list][scan8[0] - 1 + 2*8]=
740 h->ref_cache[list][scan8[0] - 1 + 3*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[2]>>1)];
742 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 2*8]=
743 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 3*8]= 0;
744 h->ref_cache[list][scan8[0] - 1 + 2*8]=
745 h->ref_cache[list][scan8[0] - 1 + 3*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
746 assert((!left_type[0]) == (!left_type[1]));
749 if(for_deblock || (IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred))
752 if(IS_INTER(topleft_type)){
753 const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
754 const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + h->b8_stride;
755 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
756 h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
758 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
759 h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
762 if(IS_INTER(topright_type)){
763 const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
764 const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
765 *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
766 h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
768 *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
769 h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
773 h->ref_cache[list][scan8[5 ]+1] =
774 h->ref_cache[list][scan8[7 ]+1] =
775 h->ref_cache[list][scan8[13]+1] = //FIXME remove past 3 (init somewhere else)
776 h->ref_cache[list][scan8[4 ]] =
777 h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
778 *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
779 *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
780 *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
781 *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
782 *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
785 /* XXX beurk, Load mvd */
786 if(IS_INTER(topleft_type)){
787 const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
788 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy];
790 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 - 1*8]= 0;
793 if(IS_INTER(top_type)){
794 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
795 *(uint32_t*)h->mvd_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 0];
796 *(uint32_t*)h->mvd_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 1];
797 *(uint32_t*)h->mvd_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 2];
798 *(uint32_t*)h->mvd_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 3];
800 *(uint32_t*)h->mvd_cache [list][scan8[0] + 0 - 1*8]=
801 *(uint32_t*)h->mvd_cache [list][scan8[0] + 1 - 1*8]=
802 *(uint32_t*)h->mvd_cache [list][scan8[0] + 2 - 1*8]=
803 *(uint32_t*)h->mvd_cache [list][scan8[0] + 3 - 1*8]= 0;
805 if(IS_INTER(left_type[0])){
806 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
807 *(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]];
808 *(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]];
810 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 0*8]=
811 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 1*8]= 0;
813 if(IS_INTER(left_type[1])){
814 const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
815 *(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]];
816 *(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]];
818 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 2*8]=
819 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 3*8]= 0;
821 *(uint32_t*)h->mvd_cache [list][scan8[5 ]+1]=
822 *(uint32_t*)h->mvd_cache [list][scan8[7 ]+1]=
823 *(uint32_t*)h->mvd_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
824 *(uint32_t*)h->mvd_cache [list][scan8[4 ]]=
825 *(uint32_t*)h->mvd_cache [list][scan8[12]]= 0;
827 if(h->slice_type == B_TYPE){
828 fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, 0, 1);
830 if(IS_DIRECT(top_type)){
831 *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0x01010101;
832 }else if(IS_8X8(top_type)){
833 int b8_xy = h->mb2b8_xy[top_xy] + h->b8_stride;
834 h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy];
835 h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 1];
837 *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0;
841 if(IS_DIRECT(left_type[0])){
842 h->direct_cache[scan8[0] - 1 + 0*8]=
843 h->direct_cache[scan8[0] - 1 + 2*8]= 1;
844 }else if(IS_8X8(left_type[0])){
845 int b8_xy = h->mb2b8_xy[left_xy[0]] + 1;
846 h->direct_cache[scan8[0] - 1 + 0*8]= h->direct_table[b8_xy];
847 h->direct_cache[scan8[0] - 1 + 2*8]= h->direct_table[b8_xy + h->b8_stride];
849 h->direct_cache[scan8[0] - 1 + 0*8]=
850 h->direct_cache[scan8[0] - 1 + 2*8]= 0;
858 h->neighbor_transform_size= !!IS_8x8DCT(top_type) + !!IS_8x8DCT(left_type[0]);
861 static inline void write_back_intra_pred_mode(H264Context *h){
862 MpegEncContext * const s = &h->s;
863 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
865 h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
866 h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
867 h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
868 h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
869 h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
870 h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
871 h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
875 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
877 static inline int check_intra4x4_pred_mode(H264Context *h){
878 MpegEncContext * const s = &h->s;
879 static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
880 static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
883 if(!(h->top_samples_available&0x8000)){
885 int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
887 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);
890 h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
895 if(!(h->left_samples_available&0x8000)){
897 int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
899 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);
902 h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
908 } //FIXME cleanup like next
911 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
913 static inline int check_intra_pred_mode(H264Context *h, int mode){
914 MpegEncContext * const s = &h->s;
915 static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
916 static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
918 if(mode < 0 || mode > 6) {
919 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);
923 if(!(h->top_samples_available&0x8000)){
926 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);
931 if(!(h->left_samples_available&0x8000)){
934 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);
943 * gets the predicted intra4x4 prediction mode.
945 static inline int pred_intra_mode(H264Context *h, int n){
946 const int index8= scan8[n];
947 const int left= h->intra4x4_pred_mode_cache[index8 - 1];
948 const int top = h->intra4x4_pred_mode_cache[index8 - 8];
949 const int min= FFMIN(left, top);
951 tprintf("mode:%d %d min:%d\n", left ,top, min);
953 if(min<0) return DC_PRED;
957 static inline void write_back_non_zero_count(H264Context *h){
958 MpegEncContext * const s = &h->s;
959 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
961 h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[7+8*1];
962 h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[7+8*2];
963 h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[7+8*3];
964 h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];
965 h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[4+8*4];
966 h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[5+8*4];
967 h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[6+8*4];
969 h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[1+8*2];
970 h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
971 h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[2+8*1];
973 h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[1+8*5];
974 h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
975 h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[2+8*4];
979 * gets the predicted number of non zero coefficients.
980 * @param n block index
982 static inline int pred_non_zero_count(H264Context *h, int n){
983 const int index8= scan8[n];
984 const int left= h->non_zero_count_cache[index8 - 1];
985 const int top = h->non_zero_count_cache[index8 - 8];
988 if(i<64) i= (i+1)>>1;
990 tprintf("pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
995 static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
996 const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];
998 if(topright_ref != PART_NOT_AVAILABLE){
999 *C= h->mv_cache[list][ i - 8 + part_width ];
1000 return topright_ref;
1002 tprintf("topright MV not available\n");
1004 *C= h->mv_cache[list][ i - 8 - 1 ];
1005 return h->ref_cache[list][ i - 8 - 1 ];
1010 * gets the predicted MV.
1011 * @param n the block index
1012 * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)
1013 * @param mx the x component of the predicted motion vector
1014 * @param my the y component of the predicted motion vector
1016 static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
1017 const int index8= scan8[n];
1018 const int top_ref= h->ref_cache[list][ index8 - 8 ];
1019 const int left_ref= h->ref_cache[list][ index8 - 1 ];
1020 const int16_t * const A= h->mv_cache[list][ index8 - 1 ];
1021 const int16_t * const B= h->mv_cache[list][ index8 - 8 ];
1023 int diagonal_ref, match_count;
1025 assert(part_width==1 || part_width==2 || part_width==4);
1035 diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);
1036 match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);
1037 tprintf("pred_motion match_count=%d\n", match_count);
1038 if(match_count > 1){ //most common
1039 *mx= mid_pred(A[0], B[0], C[0]);
1040 *my= mid_pred(A[1], B[1], C[1]);
1041 }else if(match_count==1){
1045 }else if(top_ref==ref){
1053 if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
1057 *mx= mid_pred(A[0], B[0], C[0]);
1058 *my= mid_pred(A[1], B[1], C[1]);
1062 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);
1066 * gets the directionally predicted 16x8 MV.
1067 * @param n the block index
1068 * @param mx the x component of the predicted motion vector
1069 * @param my the y component of the predicted motion vector
1071 static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
1073 const int top_ref= h->ref_cache[list][ scan8[0] - 8 ];
1074 const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
1076 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);
1084 const int left_ref= h->ref_cache[list][ scan8[8] - 1 ];
1085 const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];
1087 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);
1089 if(left_ref == ref){
1097 pred_motion(h, n, 4, list, ref, mx, my);
1101 * gets the directionally predicted 8x16 MV.
1102 * @param n the block index
1103 * @param mx the x component of the predicted motion vector
1104 * @param my the y component of the predicted motion vector
1106 static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
1108 const int left_ref= h->ref_cache[list][ scan8[0] - 1 ];
1109 const int16_t * const A= h->mv_cache[list][ scan8[0] - 1 ];
1111 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);
1113 if(left_ref == ref){
1122 diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);
1124 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);
1126 if(diagonal_ref == ref){
1134 pred_motion(h, n, 2, list, ref, mx, my);
1137 static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
1138 const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
1139 const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];
1141 tprintf("pred_pskip: (%d) (%d) at %2d %2d\n", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
1143 if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
1144 || (top_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0)
1145 || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){
1151 pred_motion(h, 0, 4, 0, 0, mx, my);
1156 static inline void direct_dist_scale_factor(H264Context * const h){
1157 const int poc = h->s.current_picture_ptr->poc;
1158 const int poc1 = h->ref_list[1][0].poc;
1160 for(i=0; i<h->ref_count[0]; i++){
1161 int poc0 = h->ref_list[0][i].poc;
1162 int td = clip(poc1 - poc0, -128, 127);
1163 if(td == 0 /* FIXME || pic0 is a long-term ref */){
1164 h->dist_scale_factor[i] = 256;
1166 int tb = clip(poc - poc0, -128, 127);
1167 int tx = (16384 + (ABS(td) >> 1)) / td;
1168 h->dist_scale_factor[i] = clip((tb*tx + 32) >> 6, -1024, 1023);
1172 static inline void direct_ref_list_init(H264Context * const h){
1173 MpegEncContext * const s = &h->s;
1174 Picture * const ref1 = &h->ref_list[1][0];
1175 Picture * const cur = s->current_picture_ptr;
1177 if(cur->pict_type == I_TYPE)
1178 cur->ref_count[0] = 0;
1179 if(cur->pict_type != B_TYPE)
1180 cur->ref_count[1] = 0;
1181 for(list=0; list<2; list++){
1182 cur->ref_count[list] = h->ref_count[list];
1183 for(j=0; j<h->ref_count[list]; j++)
1184 cur->ref_poc[list][j] = h->ref_list[list][j].poc;
1186 if(cur->pict_type != B_TYPE || h->direct_spatial_mv_pred)
1188 for(list=0; list<2; list++){
1189 for(i=0; i<ref1->ref_count[list]; i++){
1190 const int poc = ref1->ref_poc[list][i];
1191 h->map_col_to_list0[list][i] = PART_NOT_AVAILABLE;
1192 for(j=0; j<h->ref_count[list]; j++)
1193 if(h->ref_list[list][j].poc == poc){
1194 h->map_col_to_list0[list][i] = j;
1201 static inline void pred_direct_motion(H264Context * const h, int *mb_type){
1202 MpegEncContext * const s = &h->s;
1203 const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
1204 const int b8_xy = 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1205 const int b4_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1206 const int mb_type_col = h->ref_list[1][0].mb_type[mb_xy];
1207 const int16_t (*l1mv0)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[0][b4_xy];
1208 const int8_t *l1ref0 = &h->ref_list[1][0].ref_index[0][b8_xy];
1209 const int8_t *l1ref1 = &h->ref_list[1][0].ref_index[1][b8_xy];
1210 const int is_b8x8 = IS_8X8(*mb_type);
1214 if(IS_8X8(mb_type_col) && !h->sps.direct_8x8_inference_flag){
1215 /* FIXME save sub mb types from previous frames (or derive from MVs)
1216 * so we know exactly what block size to use */
1217 sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */
1218 *mb_type = MB_TYPE_8x8|MB_TYPE_L0L1;
1219 }else if(!is_b8x8 && (IS_16X16(mb_type_col) || IS_INTRA(mb_type_col))){
1220 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1221 *mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */
1223 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1224 *mb_type = MB_TYPE_8x8|MB_TYPE_L0L1;
1227 *mb_type |= MB_TYPE_DIRECT2;
1229 tprintf("mb_type = %08x, sub_mb_type = %08x, is_b8x8 = %d, mb_type_col = %08x\n", *mb_type, sub_mb_type, is_b8x8, mb_type_col);
1231 if(h->direct_spatial_mv_pred){
1236 /* ref = min(neighbors) */
1237 for(list=0; list<2; list++){
1238 int refa = h->ref_cache[list][scan8[0] - 1];
1239 int refb = h->ref_cache[list][scan8[0] - 8];
1240 int refc = h->ref_cache[list][scan8[0] - 8 + 4];
1242 refc = h->ref_cache[list][scan8[0] - 8 - 1];
1244 if(ref[list] < 0 || (refb < ref[list] && refb >= 0))
1246 if(ref[list] < 0 || (refc < ref[list] && refc >= 0))
1252 if(ref[0] < 0 && ref[1] < 0){
1253 ref[0] = ref[1] = 0;
1254 mv[0][0] = mv[0][1] =
1255 mv[1][0] = mv[1][1] = 0;
1257 for(list=0; list<2; list++){
1259 pred_motion(h, 0, 4, list, ref[list], &mv[list][0], &mv[list][1]);
1261 mv[list][0] = mv[list][1] = 0;
1266 *mb_type &= ~MB_TYPE_P0L1;
1267 sub_mb_type &= ~MB_TYPE_P0L1;
1268 }else if(ref[0] < 0){
1269 *mb_type &= ~MB_TYPE_P0L0;
1270 sub_mb_type &= ~MB_TYPE_P0L0;
1273 if(IS_16X16(*mb_type)){
1274 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref[0], 1);
1275 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, ref[1], 1);
1276 if(!IS_INTRA(mb_type_col) && l1ref0[0] == 0 &&
1277 ABS(l1mv0[0][0]) <= 1 && ABS(l1mv0[0][1]) <= 1){
1279 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1281 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
1283 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1285 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
1287 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1288 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1291 for(i8=0; i8<4; i8++){
1292 const int x8 = i8&1;
1293 const int y8 = i8>>1;
1295 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1297 h->sub_mb_type[i8] = sub_mb_type;
1299 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1300 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1301 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref[0], 1);
1302 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, ref[1], 1);
1305 if(!IS_INTRA(mb_type_col) && l1ref0[x8 + y8*h->b8_stride] == 0){
1306 for(i4=0; i4<4; i4++){
1307 const int16_t *mv_col = l1mv0[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1308 if(ABS(mv_col[0]) <= 1 && ABS(mv_col[1]) <= 1){
1310 *(uint32_t*)h->mv_cache[0][scan8[i8*4+i4]] = 0;
1312 *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = 0;
1318 }else{ /* direct temporal mv pred */
1319 if(IS_16X16(*mb_type)){
1320 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1);
1321 if(IS_INTRA(mb_type_col)){
1322 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
1323 fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
1324 fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
1326 const int ref0 = l1ref0[0] >= 0 ? h->map_col_to_list0[0][l1ref0[0]]
1327 : h->map_col_to_list0[1][l1ref1[0]];
1328 const int dist_scale_factor = h->dist_scale_factor[ref0];
1329 const int16_t *mv_col = l1mv0[0];
1331 mv_l0[0] = (dist_scale_factor * mv_col[0] + 128) >> 8;
1332 mv_l0[1] = (dist_scale_factor * mv_col[1] + 128) >> 8;
1333 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref0, 1);
1334 fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv_l0[0],mv_l0[1]), 4);
1335 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);
1338 for(i8=0; i8<4; i8++){
1339 const int x8 = i8&1;
1340 const int y8 = i8>>1;
1341 int ref0, dist_scale_factor;
1343 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1345 h->sub_mb_type[i8] = sub_mb_type;
1346 if(IS_INTRA(mb_type_col)){
1347 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1348 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1349 fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1350 fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1354 ref0 = l1ref0[x8 + y8*h->b8_stride];
1356 ref0 = h->map_col_to_list0[0][ref0];
1358 ref0 = h->map_col_to_list0[1][l1ref1[x8 + y8*h->b8_stride]];
1359 dist_scale_factor = h->dist_scale_factor[ref0];
1361 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1362 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1363 for(i4=0; i4<4; i4++){
1364 const int16_t *mv_col = l1mv0[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1365 int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]];
1366 mv_l0[0] = (dist_scale_factor * mv_col[0] + 128) >> 8;
1367 mv_l0[1] = (dist_scale_factor * mv_col[1] + 128) >> 8;
1368 *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] =
1369 pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
1376 static inline void write_back_motion(H264Context *h, int mb_type){
1377 MpegEncContext * const s = &h->s;
1378 const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1379 const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1382 for(list=0; list<2; list++){
1384 if(!USES_LIST(mb_type, list)){
1385 if(1){ //FIXME skip or never read if mb_type doesn't use it
1387 *(uint64_t*)s->current_picture.motion_val[list][b_xy + 0 + y*h->b_stride]=
1388 *(uint64_t*)s->current_picture.motion_val[list][b_xy + 2 + y*h->b_stride]= 0;
1390 if( h->pps.cabac ) {
1391 /* FIXME needed ? */
1393 *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]=
1394 *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= 0;
1398 s->current_picture.ref_index[list][b8_xy + 0 + y*h->b8_stride]=
1399 s->current_picture.ref_index[list][b8_xy + 1 + y*h->b8_stride]= LIST_NOT_USED;
1406 *(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];
1407 *(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];
1409 if( h->pps.cabac ) {
1411 *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+0 + 8*y];
1412 *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+2 + 8*y];
1416 s->current_picture.ref_index[list][b8_xy + 0 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+0 + 16*y];
1417 s->current_picture.ref_index[list][b8_xy + 1 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+2 + 16*y];
1421 if(h->slice_type == B_TYPE && h->pps.cabac){
1422 if(IS_8X8(mb_type)){
1423 h->direct_table[b8_xy+1+0*h->b8_stride] = IS_DIRECT(h->sub_mb_type[1]) ? 1 : 0;
1424 h->direct_table[b8_xy+0+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[2]) ? 1 : 0;
1425 h->direct_table[b8_xy+1+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[3]) ? 1 : 0;
1431 * Decodes a network abstraction layer unit.
1432 * @param consumed is the number of bytes used as input
1433 * @param length is the length of the array
1434 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp tailing?
1435 * @returns decoded bytes, might be src+1 if no escapes
1437 static uint8_t *decode_nal(H264Context *h, uint8_t *src, int *dst_length, int *consumed, int length){
1441 // src[0]&0x80; //forbidden bit
1442 h->nal_ref_idc= src[0]>>5;
1443 h->nal_unit_type= src[0]&0x1F;
1447 for(i=0; i<length; i++)
1448 printf("%2X ", src[i]);
1450 for(i=0; i+1<length; i+=2){
1451 if(src[i]) continue;
1452 if(i>0 && src[i-1]==0) i--;
1453 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1455 /* startcode, so we must be past the end */
1462 if(i>=length-1){ //no escaped 0
1463 *dst_length= length;
1464 *consumed= length+1; //+1 for the header
1468 h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length);
1469 dst= h->rbsp_buffer;
1471 //printf("decoding esc\n");
1474 //remove escapes (very rare 1:2^22)
1475 if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1476 if(src[si+2]==3){ //escape
1481 }else //next start code
1485 dst[di++]= src[si++];
1489 *consumed= si + 1;//+1 for the header
1490 //FIXME store exact number of bits in the getbitcontext (its needed for decoding)
1496 * @param src the data which should be escaped
1497 * @param dst the target buffer, dst+1 == src is allowed as a special case
1498 * @param length the length of the src data
1499 * @param dst_length the length of the dst array
1500 * @returns length of escaped data in bytes or -1 if an error occured
1502 static int encode_nal(H264Context *h, uint8_t *dst, uint8_t *src, int length, int dst_length){
1503 int i, escape_count, si, di;
1507 assert(dst_length>0);
1509 dst[0]= (h->nal_ref_idc<<5) + h->nal_unit_type;
1511 if(length==0) return 1;
1514 for(i=0; i<length; i+=2){
1515 if(src[i]) continue;
1516 if(i>0 && src[i-1]==0)
1518 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1524 if(escape_count==0){
1526 memcpy(dst+1, src, length);
1530 if(length + escape_count + 1> dst_length)
1533 //this should be damn rare (hopefully)
1535 h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length + escape_count);
1536 temp= h->rbsp_buffer;
1537 //printf("encoding esc\n");
1542 if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1543 temp[di++]= 0; si++;
1544 temp[di++]= 0; si++;
1546 temp[di++]= src[si++];
1549 temp[di++]= src[si++];
1551 memcpy(dst+1, temp, length+escape_count);
1553 assert(di == length+escape_count);
1559 * write 1,10,100,1000,... for alignment, yes its exactly inverse to mpeg4
1561 static void encode_rbsp_trailing(PutBitContext *pb){
1564 length= (-put_bits_count(pb))&7;
1565 if(length) put_bits(pb, length, 0);
1570 * identifies the exact end of the bitstream
1571 * @return the length of the trailing, or 0 if damaged
1573 static int decode_rbsp_trailing(uint8_t *src){
1577 tprintf("rbsp trailing %X\n", v);
1587 * idct tranforms the 16 dc values and dequantize them.
1588 * @param qp quantization parameter
1590 static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp){
1591 const int qmul= dequant_coeff[qp][0];
1594 int temp[16]; //FIXME check if this is a good idea
1595 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
1596 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1598 //memset(block, 64, 2*256);
1601 const int offset= y_offset[i];
1602 const int z0= block[offset+stride*0] + block[offset+stride*4];
1603 const int z1= block[offset+stride*0] - block[offset+stride*4];
1604 const int z2= block[offset+stride*1] - block[offset+stride*5];
1605 const int z3= block[offset+stride*1] + block[offset+stride*5];
1614 const int offset= x_offset[i];
1615 const int z0= temp[4*0+i] + temp[4*2+i];
1616 const int z1= temp[4*0+i] - temp[4*2+i];
1617 const int z2= temp[4*1+i] - temp[4*3+i];
1618 const int z3= temp[4*1+i] + temp[4*3+i];
1620 block[stride*0 +offset]= ((z0 + z3)*qmul + 2)>>2; //FIXME think about merging this into decode_resdual
1621 block[stride*2 +offset]= ((z1 + z2)*qmul + 2)>>2;
1622 block[stride*8 +offset]= ((z1 - z2)*qmul + 2)>>2;
1623 block[stride*10+offset]= ((z0 - z3)*qmul + 2)>>2;
1629 * dct tranforms the 16 dc values.
1630 * @param qp quantization parameter ??? FIXME
1632 static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
1633 // const int qmul= dequant_coeff[qp][0];
1635 int temp[16]; //FIXME check if this is a good idea
1636 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
1637 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1640 const int offset= y_offset[i];
1641 const int z0= block[offset+stride*0] + block[offset+stride*4];
1642 const int z1= block[offset+stride*0] - block[offset+stride*4];
1643 const int z2= block[offset+stride*1] - block[offset+stride*5];
1644 const int z3= block[offset+stride*1] + block[offset+stride*5];
1653 const int offset= x_offset[i];
1654 const int z0= temp[4*0+i] + temp[4*2+i];
1655 const int z1= temp[4*0+i] - temp[4*2+i];
1656 const int z2= temp[4*1+i] - temp[4*3+i];
1657 const int z3= temp[4*1+i] + temp[4*3+i];
1659 block[stride*0 +offset]= (z0 + z3)>>1;
1660 block[stride*2 +offset]= (z1 + z2)>>1;
1661 block[stride*8 +offset]= (z1 - z2)>>1;
1662 block[stride*10+offset]= (z0 - z3)>>1;
1670 static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp){
1671 const int qmul= dequant_coeff[qp][0];
1672 const int stride= 16*2;
1673 const int xStride= 16;
1676 a= block[stride*0 + xStride*0];
1677 b= block[stride*0 + xStride*1];
1678 c= block[stride*1 + xStride*0];
1679 d= block[stride*1 + xStride*1];
1686 block[stride*0 + xStride*0]= ((a+c)*qmul + 0)>>1;
1687 block[stride*0 + xStride*1]= ((e+b)*qmul + 0)>>1;
1688 block[stride*1 + xStride*0]= ((a-c)*qmul + 0)>>1;
1689 block[stride*1 + xStride*1]= ((e-b)*qmul + 0)>>1;
1693 static void chroma_dc_dct_c(DCTELEM *block){
1694 const int stride= 16*2;
1695 const int xStride= 16;
1698 a= block[stride*0 + xStride*0];
1699 b= block[stride*0 + xStride*1];
1700 c= block[stride*1 + xStride*0];
1701 d= block[stride*1 + xStride*1];
1708 block[stride*0 + xStride*0]= (a+c);
1709 block[stride*0 + xStride*1]= (e+b);
1710 block[stride*1 + xStride*0]= (a-c);
1711 block[stride*1 + xStride*1]= (e-b);
1716 * gets the chroma qp.
1718 static inline int get_chroma_qp(int chroma_qp_index_offset, int qscale){
1720 return chroma_qp[clip(qscale + chroma_qp_index_offset, 0, 51)];
1725 static void h264_diff_dct_c(DCTELEM *block, uint8_t *src1, uint8_t *src2, int stride){
1727 //FIXME try int temp instead of block
1730 const int d0= src1[0 + i*stride] - src2[0 + i*stride];
1731 const int d1= src1[1 + i*stride] - src2[1 + i*stride];
1732 const int d2= src1[2 + i*stride] - src2[2 + i*stride];
1733 const int d3= src1[3 + i*stride] - src2[3 + i*stride];
1734 const int z0= d0 + d3;
1735 const int z3= d0 - d3;
1736 const int z1= d1 + d2;
1737 const int z2= d1 - d2;
1739 block[0 + 4*i]= z0 + z1;
1740 block[1 + 4*i]= 2*z3 + z2;
1741 block[2 + 4*i]= z0 - z1;
1742 block[3 + 4*i]= z3 - 2*z2;
1746 const int z0= block[0*4 + i] + block[3*4 + i];
1747 const int z3= block[0*4 + i] - block[3*4 + i];
1748 const int z1= block[1*4 + i] + block[2*4 + i];
1749 const int z2= block[1*4 + i] - block[2*4 + i];
1751 block[0*4 + i]= z0 + z1;
1752 block[1*4 + i]= 2*z3 + z2;
1753 block[2*4 + i]= z0 - z1;
1754 block[3*4 + i]= z3 - 2*z2;
1759 //FIXME need to check that this doesnt overflow signed 32 bit for low qp, i am not sure, it's very close
1760 //FIXME check that gcc inlines this (and optimizes intra & seperate_dc stuff away)
1761 static inline int quantize_c(DCTELEM *block, uint8_t *scantable, int qscale, int intra, int seperate_dc){
1763 const int * const quant_table= quant_coeff[qscale];
1764 const int bias= intra ? (1<<QUANT_SHIFT)/3 : (1<<QUANT_SHIFT)/6;
1765 const unsigned int threshold1= (1<<QUANT_SHIFT) - bias - 1;
1766 const unsigned int threshold2= (threshold1<<1);
1772 const int dc_bias= intra ? (1<<(QUANT_SHIFT-2))/3 : (1<<(QUANT_SHIFT-2))/6;
1773 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT-2)) - dc_bias - 1;
1774 const unsigned int dc_threshold2= (dc_threshold1<<1);
1776 int level= block[0]*quant_coeff[qscale+18][0];
1777 if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1779 level= (dc_bias + level)>>(QUANT_SHIFT-2);
1782 level= (dc_bias - level)>>(QUANT_SHIFT-2);
1785 // last_non_zero = i;
1790 const int dc_bias= intra ? (1<<(QUANT_SHIFT+1))/3 : (1<<(QUANT_SHIFT+1))/6;
1791 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT+1)) - dc_bias - 1;
1792 const unsigned int dc_threshold2= (dc_threshold1<<1);
1794 int level= block[0]*quant_table[0];
1795 if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1797 level= (dc_bias + level)>>(QUANT_SHIFT+1);
1800 level= (dc_bias - level)>>(QUANT_SHIFT+1);
1803 // last_non_zero = i;
1816 const int j= scantable[i];
1817 int level= block[j]*quant_table[j];
1819 // if( bias+level >= (1<<(QMAT_SHIFT - 3))
1820 // || bias-level >= (1<<(QMAT_SHIFT - 3))){
1821 if(((unsigned)(level+threshold1))>threshold2){
1823 level= (bias + level)>>QUANT_SHIFT;
1826 level= (bias - level)>>QUANT_SHIFT;
1835 return last_non_zero;
1838 static void pred4x4_vertical_c(uint8_t *src, uint8_t *topright, int stride){
1839 const uint32_t a= ((uint32_t*)(src-stride))[0];
1840 ((uint32_t*)(src+0*stride))[0]= a;
1841 ((uint32_t*)(src+1*stride))[0]= a;
1842 ((uint32_t*)(src+2*stride))[0]= a;
1843 ((uint32_t*)(src+3*stride))[0]= a;
1846 static void pred4x4_horizontal_c(uint8_t *src, uint8_t *topright, int stride){
1847 ((uint32_t*)(src+0*stride))[0]= src[-1+0*stride]*0x01010101;
1848 ((uint32_t*)(src+1*stride))[0]= src[-1+1*stride]*0x01010101;
1849 ((uint32_t*)(src+2*stride))[0]= src[-1+2*stride]*0x01010101;
1850 ((uint32_t*)(src+3*stride))[0]= src[-1+3*stride]*0x01010101;
1853 static void pred4x4_dc_c(uint8_t *src, uint8_t *topright, int stride){
1854 const int dc= ( src[-stride] + src[1-stride] + src[2-stride] + src[3-stride]
1855 + src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 4) >>3;
1857 ((uint32_t*)(src+0*stride))[0]=
1858 ((uint32_t*)(src+1*stride))[0]=
1859 ((uint32_t*)(src+2*stride))[0]=
1860 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1863 static void pred4x4_left_dc_c(uint8_t *src, uint8_t *topright, int stride){
1864 const int dc= ( src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 2) >>2;
1866 ((uint32_t*)(src+0*stride))[0]=
1867 ((uint32_t*)(src+1*stride))[0]=
1868 ((uint32_t*)(src+2*stride))[0]=
1869 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1872 static void pred4x4_top_dc_c(uint8_t *src, uint8_t *topright, int stride){
1873 const int dc= ( src[-stride] + src[1-stride] + src[2-stride] + src[3-stride] + 2) >>2;
1875 ((uint32_t*)(src+0*stride))[0]=
1876 ((uint32_t*)(src+1*stride))[0]=
1877 ((uint32_t*)(src+2*stride))[0]=
1878 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1881 static void pred4x4_128_dc_c(uint8_t *src, uint8_t *topright, int stride){
1882 ((uint32_t*)(src+0*stride))[0]=
1883 ((uint32_t*)(src+1*stride))[0]=
1884 ((uint32_t*)(src+2*stride))[0]=
1885 ((uint32_t*)(src+3*stride))[0]= 128U*0x01010101U;
1889 #define LOAD_TOP_RIGHT_EDGE\
1890 const int t4= topright[0];\
1891 const int t5= topright[1];\
1892 const int t6= topright[2];\
1893 const int t7= topright[3];\
1895 #define LOAD_LEFT_EDGE\
1896 const int l0= src[-1+0*stride];\
1897 const int l1= src[-1+1*stride];\
1898 const int l2= src[-1+2*stride];\
1899 const int l3= src[-1+3*stride];\
1901 #define LOAD_TOP_EDGE\
1902 const int t0= src[ 0-1*stride];\
1903 const int t1= src[ 1-1*stride];\
1904 const int t2= src[ 2-1*stride];\
1905 const int t3= src[ 3-1*stride];\
1907 static void pred4x4_down_right_c(uint8_t *src, uint8_t *topright, int stride){
1908 const int lt= src[-1-1*stride];
1912 src[0+3*stride]=(l3 + 2*l2 + l1 + 2)>>2;
1914 src[1+3*stride]=(l2 + 2*l1 + l0 + 2)>>2;
1917 src[2+3*stride]=(l1 + 2*l0 + lt + 2)>>2;
1921 src[3+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1924 src[3+2*stride]=(lt + 2*t0 + t1 + 2)>>2;
1926 src[3+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1927 src[3+0*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1930 static void pred4x4_down_left_c(uint8_t *src, uint8_t *topright, int stride){
1935 src[0+0*stride]=(t0 + t2 + 2*t1 + 2)>>2;
1937 src[0+1*stride]=(t1 + t3 + 2*t2 + 2)>>2;
1940 src[0+2*stride]=(t2 + t4 + 2*t3 + 2)>>2;
1944 src[0+3*stride]=(t3 + t5 + 2*t4 + 2)>>2;
1947 src[1+3*stride]=(t4 + t6 + 2*t5 + 2)>>2;
1949 src[2+3*stride]=(t5 + t7 + 2*t6 + 2)>>2;
1950 src[3+3*stride]=(t6 + 3*t7 + 2)>>2;
1953 static void pred4x4_vertical_right_c(uint8_t *src, uint8_t *topright, int stride){
1954 const int lt= src[-1-1*stride];
1957 const __attribute__((unused)) int unu= l3;
1960 src[1+2*stride]=(lt + t0 + 1)>>1;
1962 src[2+2*stride]=(t0 + t1 + 1)>>1;
1964 src[3+2*stride]=(t1 + t2 + 1)>>1;
1965 src[3+0*stride]=(t2 + t3 + 1)>>1;
1967 src[1+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1969 src[2+3*stride]=(lt + 2*t0 + t1 + 2)>>2;
1971 src[3+3*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1972 src[3+1*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1973 src[0+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1974 src[0+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1977 static void pred4x4_vertical_left_c(uint8_t *src, uint8_t *topright, int stride){
1980 const __attribute__((unused)) int unu= t7;
1982 src[0+0*stride]=(t0 + t1 + 1)>>1;
1984 src[0+2*stride]=(t1 + t2 + 1)>>1;
1986 src[1+2*stride]=(t2 + t3 + 1)>>1;
1988 src[2+2*stride]=(t3 + t4+ 1)>>1;
1989 src[3+2*stride]=(t4 + t5+ 1)>>1;
1990 src[0+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1992 src[0+3*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1994 src[1+3*stride]=(t2 + 2*t3 + t4 + 2)>>2;
1996 src[2+3*stride]=(t3 + 2*t4 + t5 + 2)>>2;
1997 src[3+3*stride]=(t4 + 2*t5 + t6 + 2)>>2;
2000 static void pred4x4_horizontal_up_c(uint8_t *src, uint8_t *topright, int stride){
2003 src[0+0*stride]=(l0 + l1 + 1)>>1;
2004 src[1+0*stride]=(l0 + 2*l1 + l2 + 2)>>2;
2006 src[0+1*stride]=(l1 + l2 + 1)>>1;
2008 src[1+1*stride]=(l1 + 2*l2 + l3 + 2)>>2;
2010 src[0+2*stride]=(l2 + l3 + 1)>>1;
2012 src[1+2*stride]=(l2 + 2*l3 + l3 + 2)>>2;
2021 static void pred4x4_horizontal_down_c(uint8_t *src, uint8_t *topright, int stride){
2022 const int lt= src[-1-1*stride];
2025 const __attribute__((unused)) int unu= t3;
2028 src[2+1*stride]=(lt + l0 + 1)>>1;
2030 src[3+1*stride]=(l0 + 2*lt + t0 + 2)>>2;
2031 src[2+0*stride]=(lt + 2*t0 + t1 + 2)>>2;
2032 src[3+0*stride]=(t0 + 2*t1 + t2 + 2)>>2;
2034 src[2+2*stride]=(l0 + l1 + 1)>>1;
2036 src[3+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
2038 src[2+3*stride]=(l1 + l2+ 1)>>1;
2040 src[3+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
2041 src[0+3*stride]=(l2 + l3 + 1)>>1;
2042 src[1+3*stride]=(l1 + 2*l2 + l3 + 2)>>2;
2045 static void pred16x16_vertical_c(uint8_t *src, int stride){
2047 const uint32_t a= ((uint32_t*)(src-stride))[0];
2048 const uint32_t b= ((uint32_t*)(src-stride))[1];
2049 const uint32_t c= ((uint32_t*)(src-stride))[2];
2050 const uint32_t d= ((uint32_t*)(src-stride))[3];
2052 for(i=0; i<16; i++){
2053 ((uint32_t*)(src+i*stride))[0]= a;
2054 ((uint32_t*)(src+i*stride))[1]= b;
2055 ((uint32_t*)(src+i*stride))[2]= c;
2056 ((uint32_t*)(src+i*stride))[3]= d;
2060 static void pred16x16_horizontal_c(uint8_t *src, int stride){
2063 for(i=0; i<16; i++){
2064 ((uint32_t*)(src+i*stride))[0]=
2065 ((uint32_t*)(src+i*stride))[1]=
2066 ((uint32_t*)(src+i*stride))[2]=
2067 ((uint32_t*)(src+i*stride))[3]= src[-1+i*stride]*0x01010101;
2071 static void pred16x16_dc_c(uint8_t *src, int stride){
2075 dc+= src[-1+i*stride];
2082 dc= 0x01010101*((dc + 16)>>5);
2084 for(i=0; i<16; i++){
2085 ((uint32_t*)(src+i*stride))[0]=
2086 ((uint32_t*)(src+i*stride))[1]=
2087 ((uint32_t*)(src+i*stride))[2]=
2088 ((uint32_t*)(src+i*stride))[3]= dc;
2092 static void pred16x16_left_dc_c(uint8_t *src, int stride){
2096 dc+= src[-1+i*stride];
2099 dc= 0x01010101*((dc + 8)>>4);
2101 for(i=0; i<16; i++){
2102 ((uint32_t*)(src+i*stride))[0]=
2103 ((uint32_t*)(src+i*stride))[1]=
2104 ((uint32_t*)(src+i*stride))[2]=
2105 ((uint32_t*)(src+i*stride))[3]= dc;
2109 static void pred16x16_top_dc_c(uint8_t *src, int stride){
2115 dc= 0x01010101*((dc + 8)>>4);
2117 for(i=0; i<16; i++){
2118 ((uint32_t*)(src+i*stride))[0]=
2119 ((uint32_t*)(src+i*stride))[1]=
2120 ((uint32_t*)(src+i*stride))[2]=
2121 ((uint32_t*)(src+i*stride))[3]= dc;
2125 static void pred16x16_128_dc_c(uint8_t *src, int stride){
2128 for(i=0; i<16; i++){
2129 ((uint32_t*)(src+i*stride))[0]=
2130 ((uint32_t*)(src+i*stride))[1]=
2131 ((uint32_t*)(src+i*stride))[2]=
2132 ((uint32_t*)(src+i*stride))[3]= 0x01010101U*128U;
2136 static inline void pred16x16_plane_compat_c(uint8_t *src, int stride, const int svq3){
2139 uint8_t *cm = cropTbl + MAX_NEG_CROP;
2140 const uint8_t * const src0 = src+7-stride;
2141 const uint8_t *src1 = src+8*stride-1;
2142 const uint8_t *src2 = src1-2*stride; // == src+6*stride-1;
2143 int H = src0[1] - src0[-1];
2144 int V = src1[0] - src2[ 0];
2145 for(k=2; k<=8; ++k) {
2146 src1 += stride; src2 -= stride;
2147 H += k*(src0[k] - src0[-k]);
2148 V += k*(src1[0] - src2[ 0]);
2151 H = ( 5*(H/4) ) / 16;
2152 V = ( 5*(V/4) ) / 16;
2154 /* required for 100% accuracy */
2155 i = H; H = V; V = i;
2157 H = ( 5*H+32 ) >> 6;
2158 V = ( 5*V+32 ) >> 6;
2161 a = 16*(src1[0] + src2[16] + 1) - 7*(V+H);
2162 for(j=16; j>0; --j) {
2165 for(i=-16; i<0; i+=4) {
2166 src[16+i] = cm[ (b ) >> 5 ];
2167 src[17+i] = cm[ (b+ H) >> 5 ];
2168 src[18+i] = cm[ (b+2*H) >> 5 ];
2169 src[19+i] = cm[ (b+3*H) >> 5 ];
2176 static void pred16x16_plane_c(uint8_t *src, int stride){
2177 pred16x16_plane_compat_c(src, stride, 0);
2180 static void pred8x8_vertical_c(uint8_t *src, int stride){
2182 const uint32_t a= ((uint32_t*)(src-stride))[0];
2183 const uint32_t b= ((uint32_t*)(src-stride))[1];
2186 ((uint32_t*)(src+i*stride))[0]= a;
2187 ((uint32_t*)(src+i*stride))[1]= b;
2191 static void pred8x8_horizontal_c(uint8_t *src, int stride){
2195 ((uint32_t*)(src+i*stride))[0]=
2196 ((uint32_t*)(src+i*stride))[1]= src[-1+i*stride]*0x01010101;
2200 static void pred8x8_128_dc_c(uint8_t *src, int stride){
2204 ((uint32_t*)(src+i*stride))[0]=
2205 ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
2209 static void pred8x8_left_dc_c(uint8_t *src, int stride){
2215 dc0+= src[-1+i*stride];
2216 dc2+= src[-1+(i+4)*stride];
2218 dc0= 0x01010101*((dc0 + 2)>>2);
2219 dc2= 0x01010101*((dc2 + 2)>>2);
2222 ((uint32_t*)(src+i*stride))[0]=
2223 ((uint32_t*)(src+i*stride))[1]= dc0;
2226 ((uint32_t*)(src+i*stride))[0]=
2227 ((uint32_t*)(src+i*stride))[1]= dc2;
2231 static void pred8x8_top_dc_c(uint8_t *src, int stride){
2237 dc0+= src[i-stride];
2238 dc1+= src[4+i-stride];
2240 dc0= 0x01010101*((dc0 + 2)>>2);
2241 dc1= 0x01010101*((dc1 + 2)>>2);
2244 ((uint32_t*)(src+i*stride))[0]= dc0;
2245 ((uint32_t*)(src+i*stride))[1]= dc1;
2248 ((uint32_t*)(src+i*stride))[0]= dc0;
2249 ((uint32_t*)(src+i*stride))[1]= dc1;
2254 static void pred8x8_dc_c(uint8_t *src, int stride){
2256 int dc0, dc1, dc2, dc3;
2260 dc0+= src[-1+i*stride] + src[i-stride];
2261 dc1+= src[4+i-stride];
2262 dc2+= src[-1+(i+4)*stride];
2264 dc3= 0x01010101*((dc1 + dc2 + 4)>>3);
2265 dc0= 0x01010101*((dc0 + 4)>>3);
2266 dc1= 0x01010101*((dc1 + 2)>>2);
2267 dc2= 0x01010101*((dc2 + 2)>>2);
2270 ((uint32_t*)(src+i*stride))[0]= dc0;
2271 ((uint32_t*)(src+i*stride))[1]= dc1;
2274 ((uint32_t*)(src+i*stride))[0]= dc2;
2275 ((uint32_t*)(src+i*stride))[1]= dc3;
2279 static void pred8x8_plane_c(uint8_t *src, int stride){
2282 uint8_t *cm = cropTbl + MAX_NEG_CROP;
2283 const uint8_t * const src0 = src+3-stride;
2284 const uint8_t *src1 = src+4*stride-1;
2285 const uint8_t *src2 = src1-2*stride; // == src+2*stride-1;
2286 int H = src0[1] - src0[-1];
2287 int V = src1[0] - src2[ 0];
2288 for(k=2; k<=4; ++k) {
2289 src1 += stride; src2 -= stride;
2290 H += k*(src0[k] - src0[-k]);
2291 V += k*(src1[0] - src2[ 0]);
2293 H = ( 17*H+16 ) >> 5;
2294 V = ( 17*V+16 ) >> 5;
2296 a = 16*(src1[0] + src2[8]+1) - 3*(V+H);
2297 for(j=8; j>0; --j) {
2300 src[0] = cm[ (b ) >> 5 ];
2301 src[1] = cm[ (b+ H) >> 5 ];
2302 src[2] = cm[ (b+2*H) >> 5 ];
2303 src[3] = cm[ (b+3*H) >> 5 ];
2304 src[4] = cm[ (b+4*H) >> 5 ];
2305 src[5] = cm[ (b+5*H) >> 5 ];
2306 src[6] = cm[ (b+6*H) >> 5 ];
2307 src[7] = cm[ (b+7*H) >> 5 ];
2312 #define SRC(x,y) src[(x)+(y)*stride]
2314 const int l##y = (SRC(-1,y-1) + 2*SRC(-1,y) + SRC(-1,y+1) + 2) >> 2;
2315 #define PREDICT_8x8_LOAD_LEFT \
2316 const int l0 = ((has_topleft ? SRC(-1,-1) : SRC(-1,0)) \
2317 + 2*SRC(-1,0) + SRC(-1,1) + 2) >> 2; \
2318 PL(1) PL(2) PL(3) PL(4) PL(5) PL(6) \
2319 const int l7 = (SRC(-1,6) + 3*SRC(-1,7) + 2) >> 2
2322 const int t##x = (SRC(x-1,-1) + 2*SRC(x,-1) + SRC(x+1,-1) + 2) >> 2;
2323 #define PREDICT_8x8_LOAD_TOP \
2324 const int t0 = ((has_topleft ? SRC(-1,-1) : SRC(0,-1)) \
2325 + 2*SRC(0,-1) + SRC(1,-1) + 2) >> 2; \
2326 PT(1) PT(2) PT(3) PT(4) PT(5) PT(6) \
2327 const int t7 = ((has_topright ? SRC(8,-1) : SRC(7,-1)) \
2328 + 2*SRC(7,-1) + SRC(6,-1) + 2) >> 2
2331 t##x = (SRC(x-1,-1) + 2*SRC(x,-1) + SRC(x+1,-1) + 2) >> 2;
2332 #define PREDICT_8x8_LOAD_TOPRIGHT \
2333 int t8, t9, t10, t11, t12, t13, t14, t15; \
2334 if(has_topright) { \
2335 PTR(8) PTR(9) PTR(10) PTR(11) PTR(12) PTR(13) PTR(14) \
2336 t15 = (SRC(14,-1) + 3*SRC(15,-1) + 2) >> 2; \
2337 } else t8=t9=t10=t11=t12=t13=t14=t15= SRC(7,-1);
2339 #define PREDICT_8x8_LOAD_TOPLEFT \
2340 const int lt = (SRC(-1,0) + 2*SRC(-1,-1) + SRC(0,-1) + 2) >> 2
2342 #define PREDICT_8x8_DC(v) \
2344 for( y = 0; y < 8; y++ ) { \
2345 ((uint32_t*)src)[0] = \
2346 ((uint32_t*)src)[1] = v; \
2350 static void pred8x8l_128_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2352 PREDICT_8x8_DC(0x80808080);
2354 static void pred8x8l_left_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2356 PREDICT_8x8_LOAD_LEFT;
2357 const uint32_t dc = ((l0+l1+l2+l3+l4+l5+l6+l7+4) >> 3) * 0x01010101;
2360 static void pred8x8l_top_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2362 PREDICT_8x8_LOAD_TOP;
2363 const uint32_t dc = ((t0+t1+t2+t3+t4+t5+t6+t7+4) >> 3) * 0x01010101;
2366 static void pred8x8l_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2368 PREDICT_8x8_LOAD_LEFT;
2369 PREDICT_8x8_LOAD_TOP;
2370 const uint32_t dc = ((l0+l1+l2+l3+l4+l5+l6+l7
2371 +t0+t1+t2+t3+t4+t5+t6+t7+8) >> 4) * 0x01010101;
2374 static void pred8x8l_horizontal_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2376 PREDICT_8x8_LOAD_LEFT;
2377 #define ROW(y) ((uint32_t*)(src+y*stride))[0] =\
2378 ((uint32_t*)(src+y*stride))[1] = 0x01010101 * l##y
2379 ROW(0); ROW(1); ROW(2); ROW(3); ROW(4); ROW(5); ROW(6); ROW(7);
2382 static void pred8x8l_vertical_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2385 PREDICT_8x8_LOAD_TOP;
2394 for( y = 1; y < 8; y++ )
2395 *(uint64_t*)(src+y*stride) = *(uint64_t*)src;
2397 static void pred8x8l_down_left_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2399 PREDICT_8x8_LOAD_TOP;
2400 PREDICT_8x8_LOAD_TOPRIGHT;
2401 SRC(0,0)= (t0 + 2*t1 + t2 + 2) >> 2;
2402 SRC(0,1)=SRC(1,0)= (t1 + 2*t2 + t3 + 2) >> 2;
2403 SRC(0,2)=SRC(1,1)=SRC(2,0)= (t2 + 2*t3 + t4 + 2) >> 2;
2404 SRC(0,3)=SRC(1,2)=SRC(2,1)=SRC(3,0)= (t3 + 2*t4 + t5 + 2) >> 2;
2405 SRC(0,4)=SRC(1,3)=SRC(2,2)=SRC(3,1)=SRC(4,0)= (t4 + 2*t5 + t6 + 2) >> 2;
2406 SRC(0,5)=SRC(1,4)=SRC(2,3)=SRC(3,2)=SRC(4,1)=SRC(5,0)= (t5 + 2*t6 + t7 + 2) >> 2;
2407 SRC(0,6)=SRC(1,5)=SRC(2,4)=SRC(3,3)=SRC(4,2)=SRC(5,1)=SRC(6,0)= (t6 + 2*t7 + t8 + 2) >> 2;
2408 SRC(0,7)=SRC(1,6)=SRC(2,5)=SRC(3,4)=SRC(4,3)=SRC(5,2)=SRC(6,1)=SRC(7,0)= (t7 + 2*t8 + t9 + 2) >> 2;
2409 SRC(1,7)=SRC(2,6)=SRC(3,5)=SRC(4,4)=SRC(5,3)=SRC(6,2)=SRC(7,1)= (t8 + 2*t9 + t10 + 2) >> 2;
2410 SRC(2,7)=SRC(3,6)=SRC(4,5)=SRC(5,4)=SRC(6,3)=SRC(7,2)= (t9 + 2*t10 + t11 + 2) >> 2;
2411 SRC(3,7)=SRC(4,6)=SRC(5,5)=SRC(6,4)=SRC(7,3)= (t10 + 2*t11 + t12 + 2) >> 2;
2412 SRC(4,7)=SRC(5,6)=SRC(6,5)=SRC(7,4)= (t11 + 2*t12 + t13 + 2) >> 2;
2413 SRC(5,7)=SRC(6,6)=SRC(7,5)= (t12 + 2*t13 + t14 + 2) >> 2;
2414 SRC(6,7)=SRC(7,6)= (t13 + 2*t14 + t15 + 2) >> 2;
2415 SRC(7,7)= (t14 + 3*t15 + 2) >> 2;
2417 static void pred8x8l_down_right_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2419 PREDICT_8x8_LOAD_TOP;
2420 PREDICT_8x8_LOAD_LEFT;
2421 PREDICT_8x8_LOAD_TOPLEFT;
2422 SRC(0,7)= (l7 + 2*l6 + l5 + 2) >> 2;
2423 SRC(0,6)=SRC(1,7)= (l6 + 2*l5 + l4 + 2) >> 2;
2424 SRC(0,5)=SRC(1,6)=SRC(2,7)= (l5 + 2*l4 + l3 + 2) >> 2;
2425 SRC(0,4)=SRC(1,5)=SRC(2,6)=SRC(3,7)= (l4 + 2*l3 + l2 + 2) >> 2;
2426 SRC(0,3)=SRC(1,4)=SRC(2,5)=SRC(3,6)=SRC(4,7)= (l3 + 2*l2 + l1 + 2) >> 2;
2427 SRC(0,2)=SRC(1,3)=SRC(2,4)=SRC(3,5)=SRC(4,6)=SRC(5,7)= (l2 + 2*l1 + l0 + 2) >> 2;
2428 SRC(0,1)=SRC(1,2)=SRC(2,3)=SRC(3,4)=SRC(4,5)=SRC(5,6)=SRC(6,7)= (l1 + 2*l0 + lt + 2) >> 2;
2429 SRC(0,0)=SRC(1,1)=SRC(2,2)=SRC(3,3)=SRC(4,4)=SRC(5,5)=SRC(6,6)=SRC(7,7)= (l0 + 2*lt + t0 + 2) >> 2;
2430 SRC(1,0)=SRC(2,1)=SRC(3,2)=SRC(4,3)=SRC(5,4)=SRC(6,5)=SRC(7,6)= (lt + 2*t0 + t1 + 2) >> 2;
2431 SRC(2,0)=SRC(3,1)=SRC(4,2)=SRC(5,3)=SRC(6,4)=SRC(7,5)= (t0 + 2*t1 + t2 + 2) >> 2;
2432 SRC(3,0)=SRC(4,1)=SRC(5,2)=SRC(6,3)=SRC(7,4)= (t1 + 2*t2 + t3 + 2) >> 2;
2433 SRC(4,0)=SRC(5,1)=SRC(6,2)=SRC(7,3)= (t2 + 2*t3 + t4 + 2) >> 2;
2434 SRC(5,0)=SRC(6,1)=SRC(7,2)= (t3 + 2*t4 + t5 + 2) >> 2;
2435 SRC(6,0)=SRC(7,1)= (t4 + 2*t5 + t6 + 2) >> 2;
2436 SRC(7,0)= (t5 + 2*t6 + t7 + 2) >> 2;
2439 static void pred8x8l_vertical_right_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2441 PREDICT_8x8_LOAD_TOP;
2442 PREDICT_8x8_LOAD_LEFT;
2443 PREDICT_8x8_LOAD_TOPLEFT;
2444 SRC(0,6)= (l5 + 2*l4 + l3 + 2) >> 2;
2445 SRC(0,7)= (l6 + 2*l5 + l4 + 2) >> 2;
2446 SRC(0,4)=SRC(1,6)= (l3 + 2*l2 + l1 + 2) >> 2;
2447 SRC(0,5)=SRC(1,7)= (l4 + 2*l3 + l2 + 2) >> 2;
2448 SRC(0,2)=SRC(1,4)=SRC(2,6)= (l1 + 2*l0 + lt + 2) >> 2;
2449 SRC(0,3)=SRC(1,5)=SRC(2,7)= (l2 + 2*l1 + l0 + 2) >> 2;
2450 SRC(0,1)=SRC(1,3)=SRC(2,5)=SRC(3,7)= (l0 + 2*lt + t0 + 2) >> 2;
2451 SRC(0,0)=SRC(1,2)=SRC(2,4)=SRC(3,6)= (lt + t0 + 1) >> 1;
2452 SRC(1,1)=SRC(2,3)=SRC(3,5)=SRC(4,7)= (lt + 2*t0 + t1 + 2) >> 2;
2453 SRC(1,0)=SRC(2,2)=SRC(3,4)=SRC(4,6)= (t0 + t1 + 1) >> 1;
2454 SRC(2,1)=SRC(3,3)=SRC(4,5)=SRC(5,7)= (t0 + 2*t1 + t2 + 2) >> 2;
2455 SRC(2,0)=SRC(3,2)=SRC(4,4)=SRC(5,6)= (t1 + t2 + 1) >> 1;
2456 SRC(3,1)=SRC(4,3)=SRC(5,5)=SRC(6,7)= (t1 + 2*t2 + t3 + 2) >> 2;
2457 SRC(3,0)=SRC(4,2)=SRC(5,4)=SRC(6,6)= (t2 + t3 + 1) >> 1;
2458 SRC(4,1)=SRC(5,3)=SRC(6,5)=SRC(7,7)= (t2 + 2*t3 + t4 + 2) >> 2;
2459 SRC(4,0)=SRC(5,2)=SRC(6,4)=SRC(7,6)= (t3 + t4 + 1) >> 1;
2460 SRC(5,1)=SRC(6,3)=SRC(7,5)= (t3 + 2*t4 + t5 + 2) >> 2;
2461 SRC(5,0)=SRC(6,2)=SRC(7,4)= (t4 + t5 + 1) >> 1;
2462 SRC(6,1)=SRC(7,3)= (t4 + 2*t5 + t6 + 2) >> 2;
2463 SRC(6,0)=SRC(7,2)= (t5 + t6 + 1) >> 1;
2464 SRC(7,1)= (t5 + 2*t6 + t7 + 2) >> 2;
2465 SRC(7,0)= (t6 + t7 + 1) >> 1;
2467 static void pred8x8l_horizontal_down_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2469 PREDICT_8x8_LOAD_TOP;
2470 PREDICT_8x8_LOAD_LEFT;
2471 PREDICT_8x8_LOAD_TOPLEFT;
2472 SRC(0,7)= (l6 + l7 + 1) >> 1;
2473 SRC(1,7)= (l5 + 2*l6 + l7 + 2) >> 2;
2474 SRC(0,6)=SRC(2,7)= (l5 + l6 + 1) >> 1;
2475 SRC(1,6)=SRC(3,7)= (l4 + 2*l5 + l6 + 2) >> 2;
2476 SRC(0,5)=SRC(2,6)=SRC(4,7)= (l4 + l5 + 1) >> 1;
2477 SRC(1,5)=SRC(3,6)=SRC(5,7)= (l3 + 2*l4 + l5 + 2) >> 2;
2478 SRC(0,4)=SRC(2,5)=SRC(4,6)=SRC(6,7)= (l3 + l4 + 1) >> 1;
2479 SRC(1,4)=SRC(3,5)=SRC(5,6)=SRC(7,7)= (l2 + 2*l3 + l4 + 2) >> 2;
2480 SRC(0,3)=SRC(2,4)=SRC(4,5)=SRC(6,6)= (l2 + l3 + 1) >> 1;
2481 SRC(1,3)=SRC(3,4)=SRC(5,5)=SRC(7,6)= (l1 + 2*l2 + l3 + 2) >> 2;
2482 SRC(0,2)=SRC(2,3)=SRC(4,4)=SRC(6,5)= (l1 + l2 + 1) >> 1;
2483 SRC(1,2)=SRC(3,3)=SRC(5,4)=SRC(7,5)= (l0 + 2*l1 + l2 + 2) >> 2;
2484 SRC(0,1)=SRC(2,2)=SRC(4,3)=SRC(6,4)= (l0 + l1 + 1) >> 1;
2485 SRC(1,1)=SRC(3,2)=SRC(5,3)=SRC(7,4)= (lt + 2*l0 + l1 + 2) >> 2;
2486 SRC(0,0)=SRC(2,1)=SRC(4,2)=SRC(6,3)= (lt + l0 + 1) >> 1;
2487 SRC(1,0)=SRC(3,1)=SRC(5,2)=SRC(7,3)= (l0 + 2*lt + t0 + 2) >> 2;
2488 SRC(2,0)=SRC(4,1)=SRC(6,2)= (t1 + 2*t0 + lt + 2) >> 2;
2489 SRC(3,0)=SRC(5,1)=SRC(7,2)= (t2 + 2*t1 + t0 + 2) >> 2;
2490 SRC(4,0)=SRC(6,1)= (t3 + 2*t2 + t1 + 2) >> 2;
2491 SRC(5,0)=SRC(7,1)= (t4 + 2*t3 + t2 + 2) >> 2;
2492 SRC(6,0)= (t5 + 2*t4 + t3 + 2) >> 2;
2493 SRC(7,0)= (t6 + 2*t5 + t4 + 2) >> 2;
2495 static void pred8x8l_vertical_left_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2497 PREDICT_8x8_LOAD_TOP;
2498 PREDICT_8x8_LOAD_TOPRIGHT;
2499 SRC(0,0)= (t0 + t1 + 1) >> 1;
2500 SRC(0,1)= (t0 + 2*t1 + t2 + 2) >> 2;
2501 SRC(0,2)=SRC(1,0)= (t1 + t2 + 1) >> 1;
2502 SRC(0,3)=SRC(1,1)= (t1 + 2*t2 + t3 + 2) >> 2;
2503 SRC(0,4)=SRC(1,2)=SRC(2,0)= (t2 + t3 + 1) >> 1;
2504 SRC(0,5)=SRC(1,3)=SRC(2,1)= (t2 + 2*t3 + t4 + 2) >> 2;
2505 SRC(0,6)=SRC(1,4)=SRC(2,2)=SRC(3,0)= (t3 + t4 + 1) >> 1;
2506 SRC(0,7)=SRC(1,5)=SRC(2,3)=SRC(3,1)= (t3 + 2*t4 + t5 + 2) >> 2;
2507 SRC(1,6)=SRC(2,4)=SRC(3,2)=SRC(4,0)= (t4 + t5 + 1) >> 1;
2508 SRC(1,7)=SRC(2,5)=SRC(3,3)=SRC(4,1)= (t4 + 2*t5 + t6 + 2) >> 2;
2509 SRC(2,6)=SRC(3,4)=SRC(4,2)=SRC(5,0)= (t5 + t6 + 1) >> 1;
2510 SRC(2,7)=SRC(3,5)=SRC(4,3)=SRC(5,1)= (t5 + 2*t6 + t7 + 2) >> 2;
2511 SRC(3,6)=SRC(4,4)=SRC(5,2)=SRC(6,0)= (t6 + t7 + 1) >> 1;
2512 SRC(3,7)=SRC(4,5)=SRC(5,3)=SRC(6,1)= (t6 + 2*t7 + t8 + 2) >> 2;
2513 SRC(4,6)=SRC(5,4)=SRC(6,2)=SRC(7,0)= (t7 + t8 + 1) >> 1;
2514 SRC(4,7)=SRC(5,5)=SRC(6,3)=SRC(7,1)= (t7 + 2*t8 + t9 + 2) >> 2;
2515 SRC(5,6)=SRC(6,4)=SRC(7,2)= (t8 + t9 + 1) >> 1;
2516 SRC(5,7)=SRC(6,5)=SRC(7,3)= (t8 + 2*t9 + t10 + 2) >> 2;
2517 SRC(6,6)=SRC(7,4)= (t9 + t10 + 1) >> 1;
2518 SRC(6,7)=SRC(7,5)= (t9 + 2*t10 + t11 + 2) >> 2;
2519 SRC(7,6)= (t10 + t11 + 1) >> 1;
2520 SRC(7,7)= (t10 + 2*t11 + t12 + 2) >> 2;
2522 static void pred8x8l_horizontal_up_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2524 PREDICT_8x8_LOAD_LEFT;
2525 SRC(0,0)= (l0 + l1 + 1) >> 1;
2526 SRC(1,0)= (l0 + 2*l1 + l2 + 2) >> 2;
2527 SRC(0,1)=SRC(2,0)= (l1 + l2 + 1) >> 1;
2528 SRC(1,1)=SRC(3,0)= (l1 + 2*l2 + l3 + 2) >> 2;
2529 SRC(0,2)=SRC(2,1)=SRC(4,0)= (l2 + l3 + 1) >> 1;
2530 SRC(1,2)=SRC(3,1)=SRC(5,0)= (l2 + 2*l3 + l4 + 2) >> 2;
2531 SRC(0,3)=SRC(2,2)=SRC(4,1)=SRC(6,0)= (l3 + l4 + 1) >> 1;
2532 SRC(1,3)=SRC(3,2)=SRC(5,1)=SRC(7,0)= (l3 + 2*l4 + l5 + 2) >> 2;
2533 SRC(0,4)=SRC(2,3)=SRC(4,2)=SRC(6,1)= (l4 + l5 + 1) >> 1;
2534 SRC(1,4)=SRC(3,3)=SRC(5,2)=SRC(7,1)= (l4 + 2*l5 + l6 + 2) >> 2;
2535 SRC(0,5)=SRC(2,4)=SRC(4,3)=SRC(6,2)= (l5 + l6 + 1) >> 1;
2536 SRC(1,5)=SRC(3,4)=SRC(5,3)=SRC(7,2)= (l5 + 2*l6 + l7 + 2) >> 2;
2537 SRC(0,6)=SRC(2,5)=SRC(4,4)=SRC(6,3)= (l6 + l7 + 1) >> 1;
2538 SRC(1,6)=SRC(3,5)=SRC(5,4)=SRC(7,3)= (l6 + 3*l7 + 2) >> 2;
2539 SRC(0,7)=SRC(1,7)=SRC(2,6)=SRC(2,7)=SRC(3,6)=
2540 SRC(3,7)=SRC(4,5)=SRC(4,6)=SRC(4,7)=SRC(5,5)=
2541 SRC(5,6)=SRC(5,7)=SRC(6,4)=SRC(6,5)=SRC(6,6)=
2542 SRC(6,7)=SRC(7,4)=SRC(7,5)=SRC(7,6)=SRC(7,7)= l7;
2544 #undef PREDICT_8x8_LOAD_LEFT
2545 #undef PREDICT_8x8_LOAD_TOP
2546 #undef PREDICT_8x8_LOAD_TOPLEFT
2547 #undef PREDICT_8x8_LOAD_TOPRIGHT
2548 #undef PREDICT_8x8_DC
2554 static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
2555 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2556 int src_x_offset, int src_y_offset,
2557 qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
2558 MpegEncContext * const s = &h->s;
2559 const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
2560 const int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
2561 const int luma_xy= (mx&3) + ((my&3)<<2);
2562 uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*s->linesize;
2563 uint8_t * src_cb= pic->data[1] + (mx>>3) + (my>>3)*s->uvlinesize;
2564 uint8_t * src_cr= pic->data[2] + (mx>>3) + (my>>3)*s->uvlinesize;
2565 int extra_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16; //FIXME increase edge?, IMHO not worth it
2566 int extra_height= extra_width;
2568 const int full_mx= mx>>2;
2569 const int full_my= my>>2;
2571 assert(pic->data[0]);
2573 if(mx&7) extra_width -= 3;
2574 if(my&7) extra_height -= 3;
2576 if( full_mx < 0-extra_width
2577 || full_my < 0-extra_height
2578 || full_mx + 16/*FIXME*/ > s->width + extra_width
2579 || full_my + 16/*FIXME*/ > s->height + extra_height){
2580 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);
2581 src_y= s->edge_emu_buffer + 2 + 2*s->linesize;
2585 qpix_op[luma_xy](dest_y, src_y, s->linesize); //FIXME try variable height perhaps?
2587 qpix_op[luma_xy](dest_y + delta, src_y + delta, s->linesize);
2590 if(s->flags&CODEC_FLAG_GRAY) return;
2593 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);
2594 src_cb= s->edge_emu_buffer;
2596 chroma_op(dest_cb, src_cb, s->uvlinesize, chroma_height, mx&7, my&7);
2599 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);
2600 src_cr= s->edge_emu_buffer;
2602 chroma_op(dest_cr, src_cr, s->uvlinesize, chroma_height, mx&7, my&7);
2605 static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
2606 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2607 int x_offset, int y_offset,
2608 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2609 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
2610 int list0, int list1){
2611 MpegEncContext * const s = &h->s;
2612 qpel_mc_func *qpix_op= qpix_put;
2613 h264_chroma_mc_func chroma_op= chroma_put;
2615 dest_y += 2*x_offset + 2*y_offset*s-> linesize;
2616 dest_cb += x_offset + y_offset*s->uvlinesize;
2617 dest_cr += x_offset + y_offset*s->uvlinesize;
2618 x_offset += 8*s->mb_x;
2619 y_offset += 8*s->mb_y;
2622 Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
2623 mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
2624 dest_y, dest_cb, dest_cr, x_offset, y_offset,
2625 qpix_op, chroma_op);
2628 chroma_op= chroma_avg;
2632 Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
2633 mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
2634 dest_y, dest_cb, dest_cr, x_offset, y_offset,
2635 qpix_op, chroma_op);
2639 static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
2640 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2641 int x_offset, int y_offset,
2642 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2643 h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
2644 h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
2645 int list0, int list1){
2646 MpegEncContext * const s = &h->s;
2648 dest_y += 2*x_offset + 2*y_offset*s-> linesize;
2649 dest_cb += x_offset + y_offset*s->uvlinesize;
2650 dest_cr += x_offset + y_offset*s->uvlinesize;
2651 x_offset += 8*s->mb_x;
2652 y_offset += 8*s->mb_y;
2655 /* don't optimize for luma-only case, since B-frames usually
2656 * use implicit weights => chroma too. */
2657 uint8_t *tmp_cb = s->obmc_scratchpad;
2658 uint8_t *tmp_cr = tmp_cb + 8*s->uvlinesize;
2659 uint8_t *tmp_y = tmp_cr + 8*s->uvlinesize;
2660 int refn0 = h->ref_cache[0][ scan8[n] ];
2661 int refn1 = h->ref_cache[1][ scan8[n] ];
2663 mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
2664 dest_y, dest_cb, dest_cr,
2665 x_offset, y_offset, qpix_put, chroma_put);
2666 mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
2667 tmp_y, tmp_cb, tmp_cr,
2668 x_offset, y_offset, qpix_put, chroma_put);
2670 if(h->use_weight == 2){
2671 int weight0 = h->implicit_weight[refn0][refn1];
2672 int weight1 = 64 - weight0;
2673 luma_weight_avg( dest_y, tmp_y, s-> linesize, 5, weight0, weight1, 0, 0);
2674 chroma_weight_avg(dest_cb, tmp_cb, s->uvlinesize, 5, weight0, weight1, 0, 0);
2675 chroma_weight_avg(dest_cr, tmp_cr, s->uvlinesize, 5, weight0, weight1, 0, 0);
2677 luma_weight_avg(dest_y, tmp_y, s->linesize, h->luma_log2_weight_denom,
2678 h->luma_weight[0][refn0], h->luma_weight[1][refn1],
2679 h->luma_offset[0][refn0], h->luma_offset[1][refn1]);
2680 chroma_weight_avg(dest_cb, tmp_cb, s->uvlinesize, h->chroma_log2_weight_denom,
2681 h->chroma_weight[0][refn0][0], h->chroma_weight[1][refn1][0],
2682 h->chroma_offset[0][refn0][0], h->chroma_offset[1][refn1][0]);
2683 chroma_weight_avg(dest_cr, tmp_cr, s->uvlinesize, h->chroma_log2_weight_denom,
2684 h->chroma_weight[0][refn0][1], h->chroma_weight[1][refn1][1],
2685 h->chroma_offset[0][refn0][1], h->chroma_offset[1][refn1][1]);
2688 int list = list1 ? 1 : 0;
2689 int refn = h->ref_cache[list][ scan8[n] ];
2690 Picture *ref= &h->ref_list[list][refn];
2691 mc_dir_part(h, ref, n, square, chroma_height, delta, list,
2692 dest_y, dest_cb, dest_cr, x_offset, y_offset,
2693 qpix_put, chroma_put);
2695 luma_weight_op(dest_y, s->linesize, h->luma_log2_weight_denom,
2696 h->luma_weight[list][refn], h->luma_offset[list][refn]);
2697 if(h->use_weight_chroma){
2698 chroma_weight_op(dest_cb, s->uvlinesize, h->chroma_log2_weight_denom,
2699 h->chroma_weight[list][refn][0], h->chroma_offset[list][refn][0]);
2700 chroma_weight_op(dest_cr, s->uvlinesize, h->chroma_log2_weight_denom,
2701 h->chroma_weight[list][refn][1], h->chroma_offset[list][refn][1]);
2706 static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
2707 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2708 int x_offset, int y_offset,
2709 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2710 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
2711 h264_weight_func *weight_op, h264_biweight_func *weight_avg,
2712 int list0, int list1){
2713 if((h->use_weight==2 && list0 && list1
2714 && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ] != 32))
2715 || h->use_weight==1)
2716 mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
2717 x_offset, y_offset, qpix_put, chroma_put,
2718 weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
2720 mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
2721 x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
2724 static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2725 qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
2726 qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
2727 h264_weight_func *weight_op, h264_biweight_func *weight_avg){
2728 MpegEncContext * const s = &h->s;
2729 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
2730 const int mb_type= s->current_picture.mb_type[mb_xy];
2732 assert(IS_INTER(mb_type));
2734 if(IS_16X16(mb_type)){
2735 mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
2736 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
2737 &weight_op[0], &weight_avg[0],
2738 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2739 }else if(IS_16X8(mb_type)){
2740 mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
2741 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
2742 &weight_op[1], &weight_avg[1],
2743 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2744 mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
2745 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
2746 &weight_op[1], &weight_avg[1],
2747 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2748 }else if(IS_8X16(mb_type)){
2749 mc_part(h, 0, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 0, 0,
2750 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2751 &weight_op[2], &weight_avg[2],
2752 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2753 mc_part(h, 4, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 4, 0,
2754 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2755 &weight_op[2], &weight_avg[2],
2756 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2760 assert(IS_8X8(mb_type));
2763 const int sub_mb_type= h->sub_mb_type[i];
2765 int x_offset= (i&1)<<2;
2766 int y_offset= (i&2)<<1;
2768 if(IS_SUB_8X8(sub_mb_type)){
2769 mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2770 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2771 &weight_op[3], &weight_avg[3],
2772 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2773 }else if(IS_SUB_8X4(sub_mb_type)){
2774 mc_part(h, n , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2775 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2776 &weight_op[4], &weight_avg[4],
2777 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2778 mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
2779 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2780 &weight_op[4], &weight_avg[4],
2781 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2782 }else if(IS_SUB_4X8(sub_mb_type)){
2783 mc_part(h, n , 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2784 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2785 &weight_op[5], &weight_avg[5],
2786 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2787 mc_part(h, n+1, 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
2788 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2789 &weight_op[5], &weight_avg[5],
2790 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2793 assert(IS_SUB_4X4(sub_mb_type));
2795 int sub_x_offset= x_offset + 2*(j&1);
2796 int sub_y_offset= y_offset + (j&2);
2797 mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
2798 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2799 &weight_op[6], &weight_avg[6],
2800 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2807 static void decode_init_vlc(H264Context *h){
2808 static int done = 0;
2814 init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5,
2815 &chroma_dc_coeff_token_len [0], 1, 1,
2816 &chroma_dc_coeff_token_bits[0], 1, 1, 1);
2819 init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17,
2820 &coeff_token_len [i][0], 1, 1,
2821 &coeff_token_bits[i][0], 1, 1, 1);
2825 init_vlc(&chroma_dc_total_zeros_vlc[i], CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
2826 &chroma_dc_total_zeros_len [i][0], 1, 1,
2827 &chroma_dc_total_zeros_bits[i][0], 1, 1, 1);
2829 for(i=0; i<15; i++){
2830 init_vlc(&total_zeros_vlc[i], TOTAL_ZEROS_VLC_BITS, 16,
2831 &total_zeros_len [i][0], 1, 1,
2832 &total_zeros_bits[i][0], 1, 1, 1);
2836 init_vlc(&run_vlc[i], RUN_VLC_BITS, 7,
2837 &run_len [i][0], 1, 1,
2838 &run_bits[i][0], 1, 1, 1);
2840 init_vlc(&run7_vlc, RUN7_VLC_BITS, 16,
2841 &run_len [6][0], 1, 1,
2842 &run_bits[6][0], 1, 1, 1);
2847 * Sets the intra prediction function pointers.
2849 static void init_pred_ptrs(H264Context *h){
2850 // MpegEncContext * const s = &h->s;
2852 h->pred4x4[VERT_PRED ]= pred4x4_vertical_c;
2853 h->pred4x4[HOR_PRED ]= pred4x4_horizontal_c;
2854 h->pred4x4[DC_PRED ]= pred4x4_dc_c;
2855 h->pred4x4[DIAG_DOWN_LEFT_PRED ]= pred4x4_down_left_c;
2856 h->pred4x4[DIAG_DOWN_RIGHT_PRED]= pred4x4_down_right_c;
2857 h->pred4x4[VERT_RIGHT_PRED ]= pred4x4_vertical_right_c;
2858 h->pred4x4[HOR_DOWN_PRED ]= pred4x4_horizontal_down_c;
2859 h->pred4x4[VERT_LEFT_PRED ]= pred4x4_vertical_left_c;
2860 h->pred4x4[HOR_UP_PRED ]= pred4x4_horizontal_up_c;
2861 h->pred4x4[LEFT_DC_PRED ]= pred4x4_left_dc_c;
2862 h->pred4x4[TOP_DC_PRED ]= pred4x4_top_dc_c;
2863 h->pred4x4[DC_128_PRED ]= pred4x4_128_dc_c;
2865 h->pred8x8l[VERT_PRED ]= pred8x8l_vertical_c;
2866 h->pred8x8l[HOR_PRED ]= pred8x8l_horizontal_c;
2867 h->pred8x8l[DC_PRED ]= pred8x8l_dc_c;
2868 h->pred8x8l[DIAG_DOWN_LEFT_PRED ]= pred8x8l_down_left_c;
2869 h->pred8x8l[DIAG_DOWN_RIGHT_PRED]= pred8x8l_down_right_c;
2870 h->pred8x8l[VERT_RIGHT_PRED ]= pred8x8l_vertical_right_c;
2871 h->pred8x8l[HOR_DOWN_PRED ]= pred8x8l_horizontal_down_c;
2872 h->pred8x8l[VERT_LEFT_PRED ]= pred8x8l_vertical_left_c;
2873 h->pred8x8l[HOR_UP_PRED ]= pred8x8l_horizontal_up_c;
2874 h->pred8x8l[LEFT_DC_PRED ]= pred8x8l_left_dc_c;
2875 h->pred8x8l[TOP_DC_PRED ]= pred8x8l_top_dc_c;
2876 h->pred8x8l[DC_128_PRED ]= pred8x8l_128_dc_c;
2878 h->pred8x8[DC_PRED8x8 ]= pred8x8_dc_c;
2879 h->pred8x8[VERT_PRED8x8 ]= pred8x8_vertical_c;
2880 h->pred8x8[HOR_PRED8x8 ]= pred8x8_horizontal_c;
2881 h->pred8x8[PLANE_PRED8x8 ]= pred8x8_plane_c;
2882 h->pred8x8[LEFT_DC_PRED8x8]= pred8x8_left_dc_c;
2883 h->pred8x8[TOP_DC_PRED8x8 ]= pred8x8_top_dc_c;
2884 h->pred8x8[DC_128_PRED8x8 ]= pred8x8_128_dc_c;
2886 h->pred16x16[DC_PRED8x8 ]= pred16x16_dc_c;
2887 h->pred16x16[VERT_PRED8x8 ]= pred16x16_vertical_c;
2888 h->pred16x16[HOR_PRED8x8 ]= pred16x16_horizontal_c;
2889 h->pred16x16[PLANE_PRED8x8 ]= pred16x16_plane_c;
2890 h->pred16x16[LEFT_DC_PRED8x8]= pred16x16_left_dc_c;
2891 h->pred16x16[TOP_DC_PRED8x8 ]= pred16x16_top_dc_c;
2892 h->pred16x16[DC_128_PRED8x8 ]= pred16x16_128_dc_c;
2895 static void free_tables(H264Context *h){
2896 av_freep(&h->intra4x4_pred_mode);
2897 av_freep(&h->chroma_pred_mode_table);
2898 av_freep(&h->cbp_table);
2899 av_freep(&h->mvd_table[0]);
2900 av_freep(&h->mvd_table[1]);
2901 av_freep(&h->direct_table);
2902 av_freep(&h->non_zero_count);
2903 av_freep(&h->slice_table_base);
2904 av_freep(&h->top_borders[1]);
2905 av_freep(&h->top_borders[0]);
2906 h->slice_table= NULL;
2908 av_freep(&h->mb2b_xy);
2909 av_freep(&h->mb2b8_xy);
2911 av_freep(&h->dequant4_coeff);
2912 av_freep(&h->dequant8_coeff);
2914 av_freep(&h->s.obmc_scratchpad);
2919 * needs width/height
2921 static int alloc_tables(H264Context *h){
2922 MpegEncContext * const s = &h->s;
2923 const int big_mb_num= s->mb_stride * (s->mb_height+1);
2926 CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8 * sizeof(uint8_t))
2928 CHECKED_ALLOCZ(h->non_zero_count , big_mb_num * 16 * sizeof(uint8_t))
2929 CHECKED_ALLOCZ(h->slice_table_base , big_mb_num * sizeof(uint8_t))
2930 CHECKED_ALLOCZ(h->top_borders[0] , s->mb_width * (16+8+8) * sizeof(uint8_t))
2931 CHECKED_ALLOCZ(h->top_borders[1] , s->mb_width * (16+8+8) * sizeof(uint8_t))
2932 CHECKED_ALLOCZ(h->cbp_table, big_mb_num * sizeof(uint16_t))
2934 if( h->pps.cabac ) {
2935 CHECKED_ALLOCZ(h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t))
2936 CHECKED_ALLOCZ(h->mvd_table[0], 32*big_mb_num * sizeof(uint16_t));
2937 CHECKED_ALLOCZ(h->mvd_table[1], 32*big_mb_num * sizeof(uint16_t));
2938 CHECKED_ALLOCZ(h->direct_table, 32*big_mb_num * sizeof(uint8_t));
2941 memset(h->slice_table_base, -1, big_mb_num * sizeof(uint8_t));
2942 h->slice_table= h->slice_table_base + s->mb_stride + 1;
2944 CHECKED_ALLOCZ(h->mb2b_xy , big_mb_num * sizeof(uint32_t));
2945 CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint32_t));
2946 for(y=0; y<s->mb_height; y++){
2947 for(x=0; x<s->mb_width; x++){
2948 const int mb_xy= x + y*s->mb_stride;
2949 const int b_xy = 4*x + 4*y*h->b_stride;
2950 const int b8_xy= 2*x + 2*y*h->b8_stride;
2952 h->mb2b_xy [mb_xy]= b_xy;
2953 h->mb2b8_xy[mb_xy]= b8_xy;
2957 CHECKED_ALLOCZ(h->dequant4_coeff, 52*16 * sizeof(uint16_t));
2958 CHECKED_ALLOCZ(h->dequant8_coeff, 52*64 * sizeof(uint16_t));
2959 memcpy(h->dequant4_coeff, dequant_coeff, 52*16 * sizeof(uint16_t));
2960 for(q=0; q<52; q++){
2961 int shift = div6[q];
2963 if(shift >= 2) // qp<12 are shifted during dequant
2966 h->dequant8_coeff[q][x] = dequant8_coeff_init[idx][
2967 dequant8_coeff_init_scan[(x>>1)&12 | x&3] ] << shift;
2969 if(h->sps.transform_bypass){
2971 h->dequant4_coeff[0][x] = 1;
2973 h->dequant8_coeff[0][x] = 1<<2;
2976 s->obmc_scratchpad = NULL;
2984 static void common_init(H264Context *h){
2985 MpegEncContext * const s = &h->s;
2987 s->width = s->avctx->width;
2988 s->height = s->avctx->height;
2989 s->codec_id= s->avctx->codec->id;
2993 s->unrestricted_mv=1;
2994 s->decode=1; //FIXME
2997 static int decode_init(AVCodecContext *avctx){
2998 H264Context *h= avctx->priv_data;
2999 MpegEncContext * const s = &h->s;
3001 MPV_decode_defaults(s);
3006 s->out_format = FMT_H264;
3007 s->workaround_bugs= avctx->workaround_bugs;
3010 // s->decode_mb= ff_h263_decode_mb;
3012 avctx->pix_fmt= PIX_FMT_YUV420P;
3016 if(avctx->extradata_size > 0 && avctx->extradata &&
3017 *(char *)avctx->extradata == 1){
3027 static void frame_start(H264Context *h){
3028 MpegEncContext * const s = &h->s;
3031 MPV_frame_start(s, s->avctx);
3032 ff_er_frame_start(s);
3034 assert(s->linesize && s->uvlinesize);
3036 for(i=0; i<16; i++){
3037 h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
3038 h->block_offset[24+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->linesize*((scan8[i] - scan8[0])>>3);
3041 h->block_offset[16+i]=
3042 h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);
3043 h->block_offset[24+16+i]=
3044 h->block_offset[24+20+i]= 4*((scan8[i] - scan8[0])&7) + 8*s->uvlinesize*((scan8[i] - scan8[0])>>3);
3047 /* can't be in alloc_tables because linesize isn't known there.
3048 * FIXME: redo bipred weight to not require extra buffer? */
3049 if(!s->obmc_scratchpad)
3050 s->obmc_scratchpad = av_malloc(16*s->linesize + 2*8*s->uvlinesize);
3052 // s->decode= (s->flags&CODEC_FLAG_PSNR) || !s->encoding || s->current_picture.reference /*|| h->contains_intra*/ || 1;
3055 static inline void backup_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){
3056 MpegEncContext * const s = &h->s;
3060 src_cb -= uvlinesize;
3061 src_cr -= uvlinesize;
3063 // There are two lines saved, the line above the the top macroblock of a pair,
3064 // and the line above the bottom macroblock
3065 h->left_border[0]= h->top_borders[0][s->mb_x][15];
3066 for(i=1; i<17; i++){
3067 h->left_border[i]= src_y[15+i* linesize];
3070 *(uint64_t*)(h->top_borders[0][s->mb_x]+0)= *(uint64_t*)(src_y + 16*linesize);
3071 *(uint64_t*)(h->top_borders[0][s->mb_x]+8)= *(uint64_t*)(src_y +8+16*linesize);
3073 if(!(s->flags&CODEC_FLAG_GRAY)){
3074 h->left_border[17 ]= h->top_borders[0][s->mb_x][16+7];
3075 h->left_border[17+9]= h->top_borders[0][s->mb_x][24+7];
3077 h->left_border[i+17 ]= src_cb[7+i*uvlinesize];
3078 h->left_border[i+17+9]= src_cr[7+i*uvlinesize];
3080 *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+8*uvlinesize);
3081 *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+8*uvlinesize);
3085 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){
3086 MpegEncContext * const s = &h->s;
3089 int deblock_left = (s->mb_x > 0);
3090 int deblock_top = (s->mb_y > 0);
3092 src_y -= linesize + 1;
3093 src_cb -= uvlinesize + 1;
3094 src_cr -= uvlinesize + 1;
3096 #define XCHG(a,b,t,xchg)\
3103 for(i = !deblock_top; i<17; i++){
3104 XCHG(h->left_border[i ], src_y [i* linesize], temp8, xchg);
3109 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
3110 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
3111 if(s->mb_x < s->mb_width){
3112 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x+1]), *(uint64_t*)(src_y +17), temp64, 1);
3116 if(!(s->flags&CODEC_FLAG_GRAY)){
3118 for(i = !deblock_top; i<9; i++){
3119 XCHG(h->left_border[i+17 ], src_cb[i*uvlinesize], temp8, xchg);
3120 XCHG(h->left_border[i+17+9], src_cr[i*uvlinesize], temp8, xchg);
3124 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
3125 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
3130 static inline void backup_pair_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize){
3131 MpegEncContext * const s = &h->s;
3134 src_y -= 2 * linesize;
3135 src_cb -= 2 * uvlinesize;
3136 src_cr -= 2 * uvlinesize;
3138 // There are two lines saved, the line above the the top macroblock of a pair,
3139 // and the line above the bottom macroblock
3140 h->left_border[0]= h->top_borders[0][s->mb_x][15];
3141 h->left_border[1]= h->top_borders[1][s->mb_x][15];
3142 for(i=2; i<34; i++){
3143 h->left_border[i]= src_y[15+i* linesize];
3146 *(uint64_t*)(h->top_borders[0][s->mb_x]+0)= *(uint64_t*)(src_y + 32*linesize);
3147 *(uint64_t*)(h->top_borders[0][s->mb_x]+8)= *(uint64_t*)(src_y +8+32*linesize);
3148 *(uint64_t*)(h->top_borders[1][s->mb_x]+0)= *(uint64_t*)(src_y + 33*linesize);
3149 *(uint64_t*)(h->top_borders[1][s->mb_x]+8)= *(uint64_t*)(src_y +8+33*linesize);
3151 if(!(s->flags&CODEC_FLAG_GRAY)){
3152 h->left_border[34 ]= h->top_borders[0][s->mb_x][16+7];
3153 h->left_border[34+ 1]= h->top_borders[1][s->mb_x][16+7];
3154 h->left_border[34+18 ]= h->top_borders[0][s->mb_x][24+7];
3155 h->left_border[34+18+1]= h->top_borders[1][s->mb_x][24+7];
3156 for(i=2; i<18; i++){
3157 h->left_border[i+34 ]= src_cb[7+i*uvlinesize];
3158 h->left_border[i+34+18]= src_cr[7+i*uvlinesize];
3160 *(uint64_t*)(h->top_borders[0][s->mb_x]+16)= *(uint64_t*)(src_cb+16*uvlinesize);
3161 *(uint64_t*)(h->top_borders[0][s->mb_x]+24)= *(uint64_t*)(src_cr+16*uvlinesize);
3162 *(uint64_t*)(h->top_borders[1][s->mb_x]+16)= *(uint64_t*)(src_cb+17*uvlinesize);
3163 *(uint64_t*)(h->top_borders[1][s->mb_x]+24)= *(uint64_t*)(src_cr+17*uvlinesize);
3167 static inline void xchg_pair_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize, int xchg){
3168 MpegEncContext * const s = &h->s;
3171 int deblock_left = (s->mb_x > 0);
3172 int deblock_top = (s->mb_y > 0);
3174 tprintf("xchg_pair_border: src_y:%p src_cb:%p src_cr:%p ls:%d uvls:%d\n", src_y, src_cb, src_cr, linesize, uvlinesize);
3176 src_y -= 2 * linesize + 1;
3177 src_cb -= 2 * uvlinesize + 1;
3178 src_cr -= 2 * uvlinesize + 1;
3180 #define XCHG(a,b,t,xchg)\
3187 for(i = (!deblock_top)<<1; i<34; i++){
3188 XCHG(h->left_border[i ], src_y [i* linesize], temp8, xchg);
3193 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+0), *(uint64_t*)(src_y +1), temp64, xchg);
3194 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+8), *(uint64_t*)(src_y +9), temp64, 1);
3195 XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+0), *(uint64_t*)(src_y +1 +linesize), temp64, xchg);
3196 XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+8), *(uint64_t*)(src_y +9 +linesize), temp64, 1);
3199 if(!(s->flags&CODEC_FLAG_GRAY)){
3201 for(i = (!deblock_top) << 1; i<18; i++){
3202 XCHG(h->left_border[i+34 ], src_cb[i*uvlinesize], temp8, xchg);
3203 XCHG(h->left_border[i+34+18], src_cr[i*uvlinesize], temp8, xchg);
3207 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+16), *(uint64_t*)(src_cb+1), temp64, 1);
3208 XCHG(*(uint64_t*)(h->top_borders[0][s->mb_x]+24), *(uint64_t*)(src_cr+1), temp64, 1);
3209 XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+16), *(uint64_t*)(src_cb+1 +uvlinesize), temp64, 1);
3210 XCHG(*(uint64_t*)(h->top_borders[1][s->mb_x]+24), *(uint64_t*)(src_cr+1 +uvlinesize), temp64, 1);
3215 static void hl_decode_mb(H264Context *h){
3216 MpegEncContext * const s = &h->s;
3217 const int mb_x= s->mb_x;
3218 const int mb_y= s->mb_y;
3219 const int mb_xy= mb_x + mb_y*s->mb_stride;
3220 const int mb_type= s->current_picture.mb_type[mb_xy];
3221 uint8_t *dest_y, *dest_cb, *dest_cr;
3222 int linesize, uvlinesize /*dct_offset*/;
3224 int *block_offset = &h->block_offset[0];
3225 const unsigned int bottom = mb_y & 1;
3226 const int transform_bypass = (s->qscale == 0 && h->sps.transform_bypass);
3227 void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
3232 dest_y = s->current_picture.data[0] + (mb_y * 16* s->linesize ) + mb_x * 16;
3233 dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
3234 dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
3236 if (h->mb_field_decoding_flag) {
3237 linesize = s->linesize * 2;
3238 uvlinesize = s->uvlinesize * 2;
3239 block_offset = &h->block_offset[24];
3240 if(mb_y&1){ //FIXME move out of this func?
3241 dest_y -= s->linesize*15;
3242 dest_cb-= s->uvlinesize*7;
3243 dest_cr-= s->uvlinesize*7;
3246 linesize = s->linesize;
3247 uvlinesize = s->uvlinesize;
3248 // dct_offset = s->linesize * 16;
3251 idct_add = transform_bypass
3252 ? IS_8x8DCT(mb_type) ? s->dsp.add_pixels8 : s->dsp.add_pixels4
3253 : IS_8x8DCT(mb_type) ? s->dsp.h264_idct8_add : s->dsp.h264_idct_add;
3255 if (IS_INTRA_PCM(mb_type)) {
3258 // The pixels are stored in h->mb array in the same order as levels,
3259 // copy them in output in the correct order.
3260 for(i=0; i<16; i++) {
3261 for (y=0; y<4; y++) {
3262 for (x=0; x<4; x++) {
3263 *(dest_y + block_offset[i] + y*linesize + x) = h->mb[i*16+y*4+x];
3267 for(i=16; i<16+4; i++) {
3268 for (y=0; y<4; y++) {
3269 for (x=0; x<4; x++) {
3270 *(dest_cb + block_offset[i] + y*uvlinesize + x) = h->mb[i*16+y*4+x];
3274 for(i=20; i<20+4; i++) {
3275 for (y=0; y<4; y++) {
3276 for (x=0; x<4; x++) {
3277 *(dest_cr + block_offset[i] + y*uvlinesize + x) = h->mb[i*16+y*4+x];
3282 if(IS_INTRA(mb_type)){
3283 if(h->deblocking_filter) {
3284 if (h->mb_aff_frame) {
3286 xchg_pair_border(h, dest_y, dest_cb, dest_cr, s->linesize, s->uvlinesize, 1);
3288 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 1);
3292 if(!(s->flags&CODEC_FLAG_GRAY)){
3293 h->pred8x8[ h->chroma_pred_mode ](dest_cb, uvlinesize);
3294 h->pred8x8[ h->chroma_pred_mode ](dest_cr, uvlinesize);
3297 if(IS_INTRA4x4(mb_type)){
3299 if(IS_8x8DCT(mb_type)){
3300 for(i=0; i<16; i+=4){
3301 uint8_t * const ptr= dest_y + block_offset[i];
3302 const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
3303 h->pred8x8l[ dir ](ptr, (h->topleft_samples_available<<i)&0x8000,
3304 (h->topright_samples_available<<(i+1))&0x8000, linesize);
3305 if(h->non_zero_count_cache[ scan8[i] ])
3306 idct_add(ptr, h->mb + i*16, linesize);
3309 for(i=0; i<16; i++){
3310 uint8_t * const ptr= dest_y + block_offset[i];
3312 const int dir= h->intra4x4_pred_mode_cache[ scan8[i] ];
3315 if(dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED){
3316 const int topright_avail= (h->topright_samples_available<<i)&0x8000;
3317 assert(mb_y || linesize <= block_offset[i]);
3318 if(!topright_avail){
3319 tr= ptr[3 - linesize]*0x01010101;
3320 topright= (uint8_t*) &tr;
3322 topright= ptr + 4 - linesize;
3326 h->pred4x4[ dir ](ptr, topright, linesize);
3327 if(h->non_zero_count_cache[ scan8[i] ]){
3328 if(s->codec_id == CODEC_ID_H264)
3329 idct_add(ptr, h->mb + i*16, linesize);
3331 svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, 0);
3336 h->pred16x16[ h->intra16x16_pred_mode ](dest_y , linesize);
3337 if(s->codec_id == CODEC_ID_H264){
3338 if(!transform_bypass)
3339 h264_luma_dc_dequant_idct_c(h->mb, s->qscale);
3341 svq3_luma_dc_dequant_idct_c(h->mb, s->qscale);
3343 if(h->deblocking_filter) {
3344 if (h->mb_aff_frame) {
3346 uint8_t *pair_dest_y = s->current_picture.data[0] + ((mb_y-1) * 16* s->linesize ) + mb_x * 16;
3347 uint8_t *pair_dest_cb = s->current_picture.data[1] + ((mb_y-1) * 8 * s->uvlinesize) + mb_x * 8;
3348 uint8_t *pair_dest_cr = s->current_picture.data[2] + ((mb_y-1) * 8 * s->uvlinesize) + mb_x * 8;
3350 xchg_pair_border(h, pair_dest_y, pair_dest_cb, pair_dest_cr, s->linesize, s->uvlinesize, 0);
3354 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);
3357 }else if(s->codec_id == CODEC_ID_H264){
3358 hl_motion(h, dest_y, dest_cb, dest_cr,
3359 s->dsp.put_h264_qpel_pixels_tab, s->dsp.put_h264_chroma_pixels_tab,
3360 s->dsp.avg_h264_qpel_pixels_tab, s->dsp.avg_h264_chroma_pixels_tab,
3361 s->dsp.weight_h264_pixels_tab, s->dsp.biweight_h264_pixels_tab);
3365 if(!IS_INTRA4x4(mb_type)){
3366 if(s->codec_id == CODEC_ID_H264){
3367 const int di = IS_8x8DCT(mb_type) ? 4 : 1;
3368 for(i=0; i<16; i+=di){
3369 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
3370 uint8_t * const ptr= dest_y + block_offset[i];
3371 idct_add(ptr, h->mb + i*16, linesize);
3375 for(i=0; i<16; i++){
3376 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){ //FIXME benchmark weird rule, & below
3377 uint8_t * const ptr= dest_y + block_offset[i];
3378 svq3_add_idct_c(ptr, h->mb + i*16, linesize, s->qscale, IS_INTRA(mb_type) ? 1 : 0);
3384 if(!(s->flags&CODEC_FLAG_GRAY)){
3385 idct_add = transform_bypass ? s->dsp.add_pixels4 : s->dsp.h264_idct_add;
3386 if(!transform_bypass){
3387 chroma_dc_dequant_idct_c(h->mb + 16*16, h->chroma_qp);
3388 chroma_dc_dequant_idct_c(h->mb + 16*16+4*16, h->chroma_qp);
3390 if(s->codec_id == CODEC_ID_H264){
3391 for(i=16; i<16+4; i++){
3392 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
3393 uint8_t * const ptr= dest_cb + block_offset[i];
3394 idct_add(ptr, h->mb + i*16, uvlinesize);
3397 for(i=20; i<20+4; i++){
3398 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
3399 uint8_t * const ptr= dest_cr + block_offset[i];
3400 idct_add(ptr, h->mb + i*16, uvlinesize);
3404 for(i=16; i<16+4; i++){
3405 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
3406 uint8_t * const ptr= dest_cb + block_offset[i];
3407 svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
3410 for(i=20; i<20+4; i++){
3411 if(h->non_zero_count_cache[ scan8[i] ] || h->mb[i*16]){
3412 uint8_t * const ptr= dest_cr + block_offset[i];
3413 svq3_add_idct_c(ptr, h->mb + i*16, uvlinesize, chroma_qp[s->qscale + 12] - 12, 2);
3419 if(h->deblocking_filter) {
3420 if (h->mb_aff_frame) {
3421 const int mb_y = s->mb_y - 1;
3422 uint8_t *pair_dest_y, *pair_dest_cb, *pair_dest_cr;
3423 const int mb_xy= mb_x + mb_y*s->mb_stride;
3424 const int mb_type_top = s->current_picture.mb_type[mb_xy];
3425 const int mb_type_bottom= s->current_picture.mb_type[mb_xy+s->mb_stride];
3426 uint8_t tmp = s->current_picture.data[1][384];
3427 if (!bottom) return;
3428 pair_dest_y = s->current_picture.data[0] + (mb_y * 16* s->linesize ) + mb_x * 16;
3429 pair_dest_cb = s->current_picture.data[1] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
3430 pair_dest_cr = s->current_picture.data[2] + (mb_y * 8 * s->uvlinesize) + mb_x * 8;
3432 backup_pair_border(h, pair_dest_y, pair_dest_cb, pair_dest_cr, s->linesize, s->uvlinesize);
3433 // TODO deblock a pair
3436 tprintf("call mbaff filter_mb mb_x:%d mb_y:%d pair_dest_y = %p, dest_y = %p\n", mb_x, mb_y, pair_dest_y, dest_y);
3437 fill_caches(h, mb_type_top, 1); //FIXME don't fill stuff which isn't used by filter_mb
3438 filter_mb(h, mb_x, mb_y, pair_dest_y, pair_dest_cb, pair_dest_cr, linesize, uvlinesize);
3439 if (tmp != s->current_picture.data[1][384]) {
3440 tprintf("modified pixel 8,1 (1)\n");
3444 tprintf("call mbaff filter_mb\n");
3445 fill_caches(h, mb_type_bottom, 1); //FIXME don't fill stuff which isn't used by filter_mb
3446 filter_mb(h, mb_x, mb_y+1, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
3447 if (tmp != s->current_picture.data[1][384]) {
3448 tprintf("modified pixel 8,1 (2)\n");
3451 tprintf("call filter_mb\n");
3452 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
3453 fill_caches(h, mb_type, 1); //FIXME don't fill stuff which isn't used by filter_mb
3454 filter_mb(h, mb_x, mb_y, dest_y, dest_cb, dest_cr, linesize, uvlinesize);
3460 * fills the default_ref_list.
3462 static int fill_default_ref_list(H264Context *h){
3463 MpegEncContext * const s = &h->s;
3465 int smallest_poc_greater_than_current = -1;
3466 Picture sorted_short_ref[32];
3468 if(h->slice_type==B_TYPE){
3472 /* sort frame according to poc in B slice */
3473 for(out_i=0; out_i<h->short_ref_count; out_i++){
3475 int best_poc=INT_MAX;
3477 for(i=0; i<h->short_ref_count; i++){
3478 const int poc= h->short_ref[i]->poc;
3479 if(poc > limit && poc < best_poc){
3485 assert(best_i != INT_MIN);
3488 sorted_short_ref[out_i]= *h->short_ref[best_i];
3489 tprintf("sorted poc: %d->%d poc:%d fn:%d\n", best_i, out_i, sorted_short_ref[out_i].poc, sorted_short_ref[out_i].frame_num);
3490 if (-1 == smallest_poc_greater_than_current) {
3491 if (h->short_ref[best_i]->poc >= s->current_picture_ptr->poc) {
3492 smallest_poc_greater_than_current = out_i;
3498 if(s->picture_structure == PICT_FRAME){
3499 if(h->slice_type==B_TYPE){
3501 tprintf("current poc: %d, smallest_poc_greater_than_current: %d\n", s->current_picture_ptr->poc, smallest_poc_greater_than_current);
3503 // find the largest poc
3504 for(list=0; list<2; list++){
3507 int step= list ? -1 : 1;
3509 for(i=0; i<h->short_ref_count && index < h->ref_count[list]; i++, j+=step) {
3510 while(j<0 || j>= h->short_ref_count){
3511 if(j != -99 && step == (list ? -1 : 1))
3514 j= smallest_poc_greater_than_current + (step>>1);
3516 if(sorted_short_ref[j].reference != 3) continue;
3517 h->default_ref_list[list][index ]= sorted_short_ref[j];
3518 h->default_ref_list[list][index++].pic_id= sorted_short_ref[j].frame_num;
3521 for(i = 0; i < 16 && index < h->ref_count[ list ]; i++){
3522 if(h->long_ref[i] == NULL) continue;
3523 if(h->long_ref[i]->reference != 3) continue;
3525 h->default_ref_list[ list ][index ]= *h->long_ref[i];
3526 h->default_ref_list[ list ][index++].pic_id= i;;
3529 if(list && (smallest_poc_greater_than_current<=0 || smallest_poc_greater_than_current>=h->short_ref_count) && (1 < index)){
3530 // swap the two first elements of L1 when
3531 // L0 and L1 are identical
3532 Picture temp= h->default_ref_list[1][0];
3533 h->default_ref_list[1][0] = h->default_ref_list[1][1];
3534 h->default_ref_list[1][1] = temp;
3537 if(index < h->ref_count[ list ])
3538 memset(&h->default_ref_list[list][index], 0, sizeof(Picture)*(h->ref_count[ list ] - index));
3542 for(i=0; i<h->short_ref_count; i++){
3543 if(h->short_ref[i]->reference != 3) continue; //FIXME refernce field shit
3544 h->default_ref_list[0][index ]= *h->short_ref[i];
3545 h->default_ref_list[0][index++].pic_id= h->short_ref[i]->frame_num;
3547 for(i = 0; i < 16; i++){
3548 if(h->long_ref[i] == NULL) continue;
3549 if(h->long_ref[i]->reference != 3) continue;
3550 h->default_ref_list[0][index ]= *h->long_ref[i];
3551 h->default_ref_list[0][index++].pic_id= i;;