MMX for H.264 deblocking filter
[libav.git] / libavcodec / h264.c
1 /*
2 * H.26L/H.264/AVC/JVT/14496-10/... encoder/decoder
3 * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
4 *
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.
9 *
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.
14 *
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
18 *
19 */
20
21 /**
22 * @file h264.c
23 * H.264 / AVC / MPEG4 part10 codec.
24 * @author Michael Niedermayer <michaelni@gmx.at>
25 */
26
27 #include "common.h"
28 #include "dsputil.h"
29 #include "avcodec.h"
30 #include "mpegvideo.h"
31 #include "h264data.h"
32 #include "golomb.h"
33
34 #include "cabac.h"
35
36 #undef NDEBUG
37 #include <assert.h>
38
39 #define interlaced_dct interlaced_dct_is_a_bad_name
40 #define mb_intra mb_intra_isnt_initalized_see_mb_type
41
42 #define LUMA_DC_BLOCK_INDEX 25
43 #define CHROMA_DC_BLOCK_INDEX 26
44
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
51
52 #define MAX_SPS_COUNT 32
53 #define MAX_PPS_COUNT 256
54
55 #define MAX_MMCO_COUNT 66
56
57 /**
58 * Sequence parameter set
59 */
60 typedef struct SPS{
61
62 int profile_idc;
63 int level_idc;
64 int log2_max_frame_num; ///< log2_max_frame_num_minus4 + 4
65 int poc_type; ///< pic_order_cnt_type
66 int log2_max_poc_lsb; ///< log2_max_pic_order_cnt_lsb_minus4
67 int delta_pic_order_always_zero_flag;
68 int offset_for_non_ref_pic;
69 int offset_for_top_to_bottom_field;
70 int poc_cycle_length; ///< num_ref_frames_in_pic_order_cnt_cycle
71 int ref_frame_count; ///< num_ref_frames
72 int gaps_in_frame_num_allowed_flag;
73 int mb_width; ///< frame_width_in_mbs_minus1 + 1
74 int mb_height; ///< frame_height_in_mbs_minus1 + 1
75 int frame_mbs_only_flag;
76 int mb_aff; ///<mb_adaptive_frame_field_flag
77 int direct_8x8_inference_flag;
78 int crop; ///< frame_cropping_flag
79 int crop_left; ///< frame_cropping_rect_left_offset
80 int crop_right; ///< frame_cropping_rect_right_offset
81 int crop_top; ///< frame_cropping_rect_top_offset
82 int crop_bottom; ///< frame_cropping_rect_bottom_offset
83 int vui_parameters_present_flag;
84 AVRational sar;
85 int timing_info_present_flag;
86 uint32_t num_units_in_tick;
87 uint32_t time_scale;
88 int fixed_frame_rate_flag;
89 short offset_for_ref_frame[256]; //FIXME dyn aloc?
90 int bitstream_restriction_flag;
91 int num_reorder_frames;
92 }SPS;
93
94 /**
95 * Picture parameter set
96 */
97 typedef struct PPS{
98 int sps_id;
99 int cabac; ///< entropy_coding_mode_flag
100 int pic_order_present; ///< pic_order_present_flag
101 int slice_group_count; ///< num_slice_groups_minus1 + 1
102 int mb_slice_group_map_type;
103 int ref_count[2]; ///< num_ref_idx_l0/1_active_minus1 + 1
104 int weighted_pred; ///< weighted_pred_flag
105 int weighted_bipred_idc;
106 int init_qp; ///< pic_init_qp_minus26 + 26
107 int init_qs; ///< pic_init_qs_minus26 + 26
108 int chroma_qp_index_offset;
109 int deblocking_filter_parameters_present; ///< deblocking_filter_parameters_present_flag
110 int constrained_intra_pred; ///< constrained_intra_pred_flag
111 int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag
112 }PPS;
113
114 /**
115 * Memory management control operation opcode.
116 */
117 typedef enum MMCOOpcode{
118 MMCO_END=0,
119 MMCO_SHORT2UNUSED,
120 MMCO_LONG2UNUSED,
121 MMCO_SHORT2LONG,
122 MMCO_SET_MAX_LONG,
123 MMCO_RESET,
124 MMCO_LONG,
125 } MMCOOpcode;
126
127 /**
128 * Memory management control operation.
129 */
130 typedef struct MMCO{
131 MMCOOpcode opcode;
132 int short_frame_num;
133 int long_index;
134 } MMCO;
135
136 /**
137 * H264Context
138 */
139 typedef struct H264Context{
140 MpegEncContext s;
141 int nal_ref_idc;
142 int nal_unit_type;
143 #define NAL_SLICE 1
144 #define NAL_DPA 2
145 #define NAL_DPB 3
146 #define NAL_DPC 4
147 #define NAL_IDR_SLICE 5
148 #define NAL_SEI 6
149 #define NAL_SPS 7
150 #define NAL_PPS 8
151 #define NAL_PICTURE_DELIMITER 9
152 #define NAL_FILTER_DATA 10
153 uint8_t *rbsp_buffer;
154 int rbsp_buffer_size;
155
156 /**
157 * Used to parse AVC variant of h264
158 */
159 int is_avc; ///< this flag is != 0 if codec is avc1
160 int got_avcC; ///< flag used to parse avcC data only once
161 int nal_length_size; ///< Number of bytes used for nal length (1, 2 or 4)
162
163 int chroma_qp; //QPc
164
165 int prev_mb_skipped; //FIXME remove (IMHO not used)
166
167 //prediction stuff
168 int chroma_pred_mode;
169 int intra16x16_pred_mode;
170
171 int top_mb_xy;
172 int left_mb_xy[2];
173
174 int8_t intra4x4_pred_mode_cache[5*8];
175 int8_t (*intra4x4_pred_mode)[8];
176 void (*pred4x4 [9+3])(uint8_t *src, uint8_t *topright, int stride);//FIXME move to dsp?
177 void (*pred8x8 [4+3])(uint8_t *src, int stride);
178 void (*pred16x16[4+3])(uint8_t *src, int stride);
179 unsigned int topleft_samples_available;
180 unsigned int top_samples_available;
181 unsigned int topright_samples_available;
182 unsigned int left_samples_available;
183 uint8_t (*top_borders[2])[16+2*8];
184 uint8_t left_border[2*(17+2*9)];
185
186 /**
187 * non zero coeff count cache.
188 * is 64 if not available.
189 */
190 uint8_t non_zero_count_cache[6*8] __align8;
191 uint8_t (*non_zero_count)[16];
192
193 /**
194 * Motion vector cache.
195 */
196 int16_t mv_cache[2][5*8][2] __align8;
197 int8_t ref_cache[2][5*8] __align8;
198 #define LIST_NOT_USED -1 //FIXME rename?
199 #define PART_NOT_AVAILABLE -2
200
201 /**
202 * is 1 if the specific list MV&references are set to 0,0,-2.
203 */
204 int mv_cache_clean[2];
205
206 /**
207 * block_offset[ 0..23] for frame macroblocks
208 * block_offset[24..47] for field macroblocks
209 */
210 int block_offset[2*(16+8)];
211
212 uint16_t *mb2b_xy; //FIXME are these 4 a good idea?
213 uint16_t *mb2b8_xy;
214 int b_stride; //FIXME use s->b4_stride
215 int b8_stride;
216
217 int halfpel_flag;
218 int thirdpel_flag;
219
220 int unknown_svq3_flag;
221 int next_slice_index;
222
223 SPS sps_buffer[MAX_SPS_COUNT];
224 SPS sps; ///< current sps
225
226 PPS pps_buffer[MAX_PPS_COUNT];
227 /**
228 * current pps
229 */
230 PPS pps; //FIXME move tp Picture perhaps? (->no) do we need that?
231
232 int slice_num;
233 uint8_t *slice_table_base;
234 uint8_t *slice_table; ///< slice_table_base + mb_stride + 1
235 int slice_type;
236 int slice_type_fixed;
237
238 //interlacing specific flags
239 int mb_aff_frame;
240 int mb_field_decoding_flag;
241
242 int sub_mb_type[4];
243
244 //POC stuff
245 int poc_lsb;
246 int poc_msb;
247 int delta_poc_bottom;
248 int delta_poc[2];
249 int frame_num;
250 int prev_poc_msb; ///< poc_msb of the last reference pic for POC type 0
251 int prev_poc_lsb; ///< poc_lsb of the last reference pic for POC type 0
252 int frame_num_offset; ///< for POC type 2
253 int prev_frame_num_offset; ///< for POC type 2
254 int prev_frame_num; ///< frame_num of the last pic for POC type 1/2
255
256 /**
257 * frame_num for frames or 2*frame_num for field pics.
258 */
259 int curr_pic_num;
260
261 /**
262 * max_frame_num or 2*max_frame_num for field pics.
263 */
264 int max_pic_num;
265
266 //Weighted pred stuff
267 int use_weight;
268 int use_weight_chroma;
269 int luma_log2_weight_denom;
270 int chroma_log2_weight_denom;
271 int luma_weight[2][16];
272 int luma_offset[2][16];
273 int chroma_weight[2][16][2];
274 int chroma_offset[2][16][2];
275 int implicit_weight[16][16];
276
277 //deblock
278 int deblocking_filter; ///< disable_deblocking_filter_idc with 1<->0
279 int slice_alpha_c0_offset;
280 int slice_beta_offset;
281
282 int redundant_pic_count;
283
284 int direct_spatial_mv_pred;
285 int dist_scale_factor[16];
286 int map_col_to_list0[2][16];
287
288 /**
289 * num_ref_idx_l0/1_active_minus1 + 1
290 */
291 int ref_count[2];// FIXME split for AFF
292 Picture *short_ref[32];
293 Picture *long_ref[32];
294 Picture default_ref_list[2][32];
295 Picture ref_list[2][32]; //FIXME size?
296 Picture field_ref_list[2][32]; //FIXME size?
297 Picture *delayed_pic[16]; //FIXME size?
298 Picture *delayed_output_pic;
299
300 /**
301 * memory management control operations buffer.
302 */
303 MMCO mmco[MAX_MMCO_COUNT];
304 int mmco_index;
305
306 int long_ref_count; ///< number of actual long term references
307 int short_ref_count; ///< number of actual short term references
308
309 //data partitioning
310 GetBitContext intra_gb;
311 GetBitContext inter_gb;
312 GetBitContext *intra_gb_ptr;
313 GetBitContext *inter_gb_ptr;
314
315 DCTELEM mb[16*24] __align8;
316
317 /**
318 * Cabac
319 */
320 CABACContext cabac;
321 uint8_t cabac_state[399];
322 int cabac_init_idc;
323
324 /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0,1,2), 0x0? luma_cbp */
325 uint16_t *cbp_table;
326 int top_cbp;
327 int left_cbp;
328 /* chroma_pred_mode for i4x4 or i16x16, else 0 */
329 uint8_t *chroma_pred_mode_table;
330 int last_qscale_diff;
331 int16_t (*mvd_table[2])[2];
332 int16_t mvd_cache[2][5*8][2] __align8;
333 uint8_t *direct_table;
334 uint8_t direct_cache[5*8];
335
336 }H264Context;
337
338 static VLC coeff_token_vlc[4];
339 static VLC chroma_dc_coeff_token_vlc;
340
341 static VLC total_zeros_vlc[15];
342 static VLC chroma_dc_total_zeros_vlc[3];
343
344 static VLC run_vlc[6];
345 static VLC run7_vlc;
346
347 static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
348 static void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
349 static void filter_mb( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr, unsigned int linesize, unsigned int uvlinesize);
350
351 static inline uint32_t pack16to32(int a, int b){
352 #ifdef WORDS_BIGENDIAN
353 return (b&0xFFFF) + (a<<16);
354 #else
355 return (a&0xFFFF) + (b<<16);
356 #endif
357 }
358
359 /**
360 * fill a rectangle.
361 * @param h height of the rectangle, should be a constant
362 * @param w width of the rectangle, should be a constant
363 * @param size the size of val (1 or 4), should be a constant
364 */
365 static inline void fill_rectangle(void *vp, int w, int h, int stride, uint32_t val, int size){ //FIXME ensure this IS inlined
366 uint8_t *p= (uint8_t*)vp;
367 assert(size==1 || size==4);
368
369 w *= size;
370 stride *= size;
371
372 assert((((int)vp)&(FFMIN(w, STRIDE_ALIGN)-1)) == 0);
373 //FIXME check what gcc generates for 64 bit on x86 and possible write a 32 bit ver of it
374 if(w==2 && h==2){
375 *(uint16_t*)(p + 0)=
376 *(uint16_t*)(p + stride)= size==4 ? val : val*0x0101;
377 }else if(w==2 && h==4){
378 *(uint16_t*)(p + 0*stride)=
379 *(uint16_t*)(p + 1*stride)=
380 *(uint16_t*)(p + 2*stride)=
381 *(uint16_t*)(p + 3*stride)= size==4 ? val : val*0x0101;
382 }else if(w==4 && h==1){
383 *(uint32_t*)(p + 0*stride)= size==4 ? val : val*0x01010101;
384 }else if(w==4 && h==2){
385 *(uint32_t*)(p + 0*stride)=
386 *(uint32_t*)(p + 1*stride)= size==4 ? val : val*0x01010101;
387 }else if(w==4 && h==4){
388 *(uint32_t*)(p + 0*stride)=
389 *(uint32_t*)(p + 1*stride)=
390 *(uint32_t*)(p + 2*stride)=
391 *(uint32_t*)(p + 3*stride)= size==4 ? val : val*0x01010101;
392 }else if(w==8 && h==1){
393 *(uint32_t*)(p + 0)=
394 *(uint32_t*)(p + 4)= size==4 ? val : val*0x01010101;
395 }else if(w==8 && h==2){
396 *(uint32_t*)(p + 0 + 0*stride)=
397 *(uint32_t*)(p + 4 + 0*stride)=
398 *(uint32_t*)(p + 0 + 1*stride)=
399 *(uint32_t*)(p + 4 + 1*stride)= size==4 ? val : val*0x01010101;
400 }else if(w==8 && h==4){
401 *(uint64_t*)(p + 0*stride)=
402 *(uint64_t*)(p + 1*stride)=
403 *(uint64_t*)(p + 2*stride)=
404 *(uint64_t*)(p + 3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
405 }else if(w==16 && h==2){
406 *(uint64_t*)(p + 0+0*stride)=
407 *(uint64_t*)(p + 8+0*stride)=
408 *(uint64_t*)(p + 0+1*stride)=
409 *(uint64_t*)(p + 8+1*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
410 }else if(w==16 && h==4){
411 *(uint64_t*)(p + 0+0*stride)=
412 *(uint64_t*)(p + 8+0*stride)=
413 *(uint64_t*)(p + 0+1*stride)=
414 *(uint64_t*)(p + 8+1*stride)=
415 *(uint64_t*)(p + 0+2*stride)=
416 *(uint64_t*)(p + 8+2*stride)=
417 *(uint64_t*)(p + 0+3*stride)=
418 *(uint64_t*)(p + 8+3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
419 }else
420 assert(0);
421 }
422
423 static inline void fill_caches(H264Context *h, int mb_type, int for_deblock){
424 MpegEncContext * const s = &h->s;
425 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
426 int topleft_xy, top_xy, topright_xy, left_xy[2];
427 int topleft_type, top_type, topright_type, left_type[2];
428 int left_block[8];
429 int i;
430
431 //FIXME deblocking can skip fill_caches much of the time with multiple slices too.
432 // the actual condition is whether we're on the edge of a slice,
433 // and even then the intra and nnz parts are unnecessary.
434 if(for_deblock && h->slice_num == 1)
435 return;
436
437 //wow what a mess, why didnt they simplify the interlacing&intra stuff, i cant imagine that these complex rules are worth it
438
439 top_xy = mb_xy - s->mb_stride;
440 topleft_xy = top_xy - 1;
441 topright_xy= top_xy + 1;
442 left_xy[1] = left_xy[0] = mb_xy-1;
443 left_block[0]= 0;
444 left_block[1]= 1;
445 left_block[2]= 2;
446 left_block[3]= 3;
447 left_block[4]= 7;
448 left_block[5]= 10;
449 left_block[6]= 8;
450 left_block[7]= 11;
451 if(h->mb_aff_frame){
452 const int pair_xy = s->mb_x + (s->mb_y & ~1)*s->mb_stride;
453 const int top_pair_xy = pair_xy - s->mb_stride;
454 const int topleft_pair_xy = top_pair_xy - 1;
455 const int topright_pair_xy = top_pair_xy + 1;
456 const int topleft_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[topleft_pair_xy]);
457 const int top_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[top_pair_xy]);
458 const int topright_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[topright_pair_xy]);
459 const int left_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[pair_xy-1]);
460 const int curr_mb_frame_flag = !IS_INTERLACED(mb_type);
461 const int bottom = (s->mb_y & 1);
462 tprintf("fill_caches: curr_mb_frame_flag:%d, left_mb_frame_flag:%d, topleft_mb_frame_flag:%d, top_mb_frame_flag:%d, topright_mb_frame_flag:%d\n", curr_mb_frame_flag, left_mb_frame_flag, topleft_mb_frame_flag, top_mb_frame_flag, topright_mb_frame_flag);
463 if (bottom
464 ? !curr_mb_frame_flag // bottom macroblock
465 : (!curr_mb_frame_flag && !top_mb_frame_flag) // top macroblock
466 ) {
467 top_xy -= s->mb_stride;
468 }
469 if (bottom
470 ? !curr_mb_frame_flag // bottom macroblock
471 : (!curr_mb_frame_flag && !topleft_mb_frame_flag) // top macroblock
472 ) {
473 topleft_xy -= s->mb_stride;
474 }
475 if (bottom
476 ? !curr_mb_frame_flag // bottom macroblock
477 : (!curr_mb_frame_flag && !topright_mb_frame_flag) // top macroblock
478 ) {
479 topright_xy -= s->mb_stride;
480 }
481 if (left_mb_frame_flag != curr_mb_frame_flag) {
482 left_xy[1] = left_xy[0] = pair_xy - 1;
483 if (curr_mb_frame_flag) {
484 if (bottom) {
485 left_block[0]= 2;
486 left_block[1]= 2;
487 left_block[2]= 3;
488 left_block[3]= 3;
489 left_block[4]= 8;
490 left_block[5]= 11;
491 left_block[6]= 8;
492 left_block[7]= 11;
493 } else {
494 left_block[0]= 0;
495 left_block[1]= 0;
496 left_block[2]= 1;
497 left_block[3]= 1;
498 left_block[4]= 7;
499 left_block[5]= 10;
500 left_block[6]= 7;
501 left_block[7]= 10;
502 }
503 } else {
504 left_xy[1] += s->mb_stride;
505 //left_block[0]= 0;
506 left_block[1]= 2;
507 left_block[2]= 0;
508 left_block[3]= 2;
509 //left_block[4]= 7;
510 left_block[5]= 10;
511 left_block[6]= 7;
512 left_block[7]= 10;
513 }
514 }
515 }
516
517 h->top_mb_xy = top_xy;
518 h->left_mb_xy[0] = left_xy[0];
519 h->left_mb_xy[1] = left_xy[1];
520 if(for_deblock){
521 topleft_type = h->slice_table[topleft_xy ] < 255 ? s->current_picture.mb_type[topleft_xy] : 0;
522 top_type = h->slice_table[top_xy ] < 255 ? s->current_picture.mb_type[top_xy] : 0;
523 topright_type= h->slice_table[topright_xy] < 255 ? s->current_picture.mb_type[topright_xy]: 0;
524 left_type[0] = h->slice_table[left_xy[0] ] < 255 ? s->current_picture.mb_type[left_xy[0]] : 0;
525 left_type[1] = h->slice_table[left_xy[1] ] < 255 ? s->current_picture.mb_type[left_xy[1]] : 0;
526 }else{
527 topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
528 top_type = h->slice_table[top_xy ] == h->slice_num ? s->current_picture.mb_type[top_xy] : 0;
529 topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
530 left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
531 left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
532 }
533
534 if(IS_INTRA(mb_type)){
535 h->topleft_samples_available=
536 h->top_samples_available=
537 h->left_samples_available= 0xFFFF;
538 h->topright_samples_available= 0xEEEA;
539
540 if(!IS_INTRA(top_type) && (top_type==0 || h->pps.constrained_intra_pred)){
541 h->topleft_samples_available= 0xB3FF;
542 h->top_samples_available= 0x33FF;
543 h->topright_samples_available= 0x26EA;
544 }
545 for(i=0; i<2; i++){
546 if(!IS_INTRA(left_type[i]) && (left_type[i]==0 || h->pps.constrained_intra_pred)){
547 h->topleft_samples_available&= 0xDF5F;
548 h->left_samples_available&= 0x5F5F;
549 }
550 }
551
552 if(!IS_INTRA(topleft_type) && (topleft_type==0 || h->pps.constrained_intra_pred))
553 h->topleft_samples_available&= 0x7FFF;
554
555 if(!IS_INTRA(topright_type) && (topright_type==0 || h->pps.constrained_intra_pred))
556 h->topright_samples_available&= 0xFBFF;
557
558 if(IS_INTRA4x4(mb_type)){
559 if(IS_INTRA4x4(top_type)){
560 h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
561 h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
562 h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
563 h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
564 }else{
565 int pred;
566 if(!top_type || (IS_INTER(top_type) && h->pps.constrained_intra_pred))
567 pred= -1;
568 else{
569 pred= 2;
570 }
571 h->intra4x4_pred_mode_cache[4+8*0]=
572 h->intra4x4_pred_mode_cache[5+8*0]=
573 h->intra4x4_pred_mode_cache[6+8*0]=
574 h->intra4x4_pred_mode_cache[7+8*0]= pred;
575 }
576 for(i=0; i<2; i++){
577 if(IS_INTRA4x4(left_type[i])){
578 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
579 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
580 }else{
581 int pred;
582 if(!left_type[i] || (IS_INTER(left_type[i]) && h->pps.constrained_intra_pred))
583 pred= -1;
584 else{
585 pred= 2;
586 }
587 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
588 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
589 }
590 }
591 }
592 }
593
594
595 /*
596 0 . T T. T T T T
597 1 L . .L . . . .
598 2 L . .L . . . .
599 3 . T TL . . . .
600 4 L . .L . . . .
601 5 L . .. . . . .
602 */
603 //FIXME constraint_intra_pred & partitioning & nnz (lets hope this is just a typo in the spec)
604 if(top_type){
605 h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][4];
606 h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][5];
607 h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][6];
608 h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3];
609
610 h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][9];
611 h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8];
612
613 h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][12];
614 h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11];
615
616 }else{
617 h->non_zero_count_cache[4+8*0]=
618 h->non_zero_count_cache[5+8*0]=
619 h->non_zero_count_cache[6+8*0]=
620 h->non_zero_count_cache[7+8*0]=
621
622 h->non_zero_count_cache[1+8*0]=
623 h->non_zero_count_cache[2+8*0]=
624
625 h->non_zero_count_cache[1+8*3]=
626 h->non_zero_count_cache[2+8*3]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
627
628 }
629
630 for (i=0; i<2; i++) {
631 if(left_type[i]){
632 h->non_zero_count_cache[3+8*1 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[0+2*i]];
633 h->non_zero_count_cache[3+8*2 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[1+2*i]];
634 h->non_zero_count_cache[0+8*1 + 8*i]= h->non_zero_count[left_xy[i]][left_block[4+2*i]];
635 h->non_zero_count_cache[0+8*4 + 8*i]= h->non_zero_count[left_xy[i]][left_block[5+2*i]];
636 }else{
637 h->non_zero_count_cache[3+8*1 + 2*8*i]=
638 h->non_zero_count_cache[3+8*2 + 2*8*i]=
639 h->non_zero_count_cache[0+8*1 + 8*i]=
640 h->non_zero_count_cache[0+8*4 + 8*i]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
641 }
642 }
643
644 if( h->pps.cabac ) {
645 // top_cbp
646 if(top_type) {
647 h->top_cbp = h->cbp_table[top_xy];
648 } else if(IS_INTRA(mb_type)) {
649 h->top_cbp = 0x1C0;
650 } else {
651 h->top_cbp = 0;
652 }
653 // left_cbp
654 if (left_type[0]) {
655 h->left_cbp = h->cbp_table[left_xy[0]] & 0x1f0;
656 } else if(IS_INTRA(mb_type)) {
657 h->left_cbp = 0x1C0;
658 } else {
659 h->left_cbp = 0;
660 }
661 if (left_type[0]) {
662 h->left_cbp |= ((h->cbp_table[left_xy[0]]>>((left_block[0]&(~1))+1))&0x1) << 1;
663 }
664 if (left_type[1]) {
665 h->left_cbp |= ((h->cbp_table[left_xy[1]]>>((left_block[2]&(~1))+1))&0x1) << 3;
666 }
667 }
668
669 #if 1
670 //FIXME direct mb can skip much of this
671 if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
672 int list;
673 for(list=0; list<1+(h->slice_type==B_TYPE); list++){
674 if(!USES_LIST(mb_type, list) && !IS_DIRECT(mb_type) && !h->deblocking_filter){
675 /*if(!h->mv_cache_clean[list]){
676 memset(h->mv_cache [list], 0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
677 memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
678 h->mv_cache_clean[list]= 1;
679 }*/
680 continue;
681 }
682 h->mv_cache_clean[list]= 0;
683
684 if(IS_INTER(top_type)){
685 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
686 const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
687 *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0];
688 *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];
689 *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];
690 *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];
691 h->ref_cache[list][scan8[0] + 0 - 1*8]=
692 h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
693 h->ref_cache[list][scan8[0] + 2 - 1*8]=
694 h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
695 }else{
696 *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]=
697 *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]=
698 *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]=
699 *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;
700 *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
701 }
702
703 //FIXME unify cleanup or sth
704 if(IS_INTER(left_type[0])){
705 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
706 const int b8_xy= h->mb2b8_xy[left_xy[0]] + 1;
707 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 + 0*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[0]];
708 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 + 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[1]];
709 h->ref_cache[list][scan8[0] - 1 + 0*8]=
710 h->ref_cache[list][scan8[0] - 1 + 1*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0]>>1)];
711 }else{
712 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 0*8]=
713 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 1*8]= 0;
714 h->ref_cache[list][scan8[0] - 1 + 0*8]=
715 h->ref_cache[list][scan8[0] - 1 + 1*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
716 }
717
718 if(IS_INTER(left_type[1])){
719 const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
720 const int b8_xy= h->mb2b8_xy[left_xy[1]] + 1;
721 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 + 2*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[2]];
722 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 + 3*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[3]];
723 h->ref_cache[list][scan8[0] - 1 + 2*8]=
724 h->ref_cache[list][scan8[0] - 1 + 3*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[2]>>1)];
725 }else{
726 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 2*8]=
727 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 3*8]= 0;
728 h->ref_cache[list][scan8[0] - 1 + 2*8]=
729 h->ref_cache[list][scan8[0] - 1 + 3*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
730 }
731
732 if(for_deblock || (IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred))
733 continue;
734
735 if(IS_INTER(topleft_type)){
736 const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
737 const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + h->b8_stride;
738 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
739 h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
740 }else{
741 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
742 h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
743 }
744
745 if(IS_INTER(topright_type)){
746 const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
747 const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
748 *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
749 h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
750 }else{
751 *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
752 h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
753 }
754
755
756 h->ref_cache[list][scan8[5 ]+1] =
757 h->ref_cache[list][scan8[7 ]+1] =
758 h->ref_cache[list][scan8[13]+1] = //FIXME remove past 3 (init somewher else)
759 h->ref_cache[list][scan8[4 ]] =
760 h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
761 *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
762 *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
763 *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewher else)
764 *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
765 *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
766
767 if( h->pps.cabac ) {
768 /* XXX beurk, Load mvd */
769 if(IS_INTER(topleft_type)){
770 const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
771 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy];
772 }else{
773 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 - 1*8]= 0;
774 }
775
776 if(IS_INTER(top_type)){
777 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
778 *(uint32_t*)h->mvd_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 0];
779 *(uint32_t*)h->mvd_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 1];
780 *(uint32_t*)h->mvd_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 2];
781 *(uint32_t*)h->mvd_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 3];
782 }else{
783 *(uint32_t*)h->mvd_cache [list][scan8[0] + 0 - 1*8]=
784 *(uint32_t*)h->mvd_cache [list][scan8[0] + 1 - 1*8]=
785 *(uint32_t*)h->mvd_cache [list][scan8[0] + 2 - 1*8]=
786 *(uint32_t*)h->mvd_cache [list][scan8[0] + 3 - 1*8]= 0;
787 }
788 if(IS_INTER(left_type[0])){
789 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
790 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 0*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[0]];
791 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[1]];
792 }else{
793 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 0*8]=
794 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 1*8]= 0;
795 }
796 if(IS_INTER(left_type[1])){
797 const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
798 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 2*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[2]];
799 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 3*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[3]];
800 }else{
801 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 2*8]=
802 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 3*8]= 0;
803 }
804 *(uint32_t*)h->mvd_cache [list][scan8[5 ]+1]=
805 *(uint32_t*)h->mvd_cache [list][scan8[7 ]+1]=
806 *(uint32_t*)h->mvd_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewher else)
807 *(uint32_t*)h->mvd_cache [list][scan8[4 ]]=
808 *(uint32_t*)h->mvd_cache [list][scan8[12]]= 0;
809
810 if(h->slice_type == B_TYPE){
811 fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, 0, 1);
812
813 if(IS_DIRECT(top_type)){
814 *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0x01010101;
815 }else if(IS_8X8(top_type)){
816 int b8_xy = h->mb2b8_xy[top_xy] + h->b8_stride;
817 h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy];
818 h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 1];
819 }else{
820 *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0;
821 }
822
823 //FIXME interlacing
824 if(IS_DIRECT(left_type[0])){
825 h->direct_cache[scan8[0] - 1 + 0*8]=
826 h->direct_cache[scan8[0] - 1 + 2*8]= 1;
827 }else if(IS_8X8(left_type[0])){
828 int b8_xy = h->mb2b8_xy[left_xy[0]] + 1;
829 h->direct_cache[scan8[0] - 1 + 0*8]= h->direct_table[b8_xy];
830 h->direct_cache[scan8[0] - 1 + 2*8]= h->direct_table[b8_xy + h->b8_stride];
831 }else{
832 h->direct_cache[scan8[0] - 1 + 0*8]=
833 h->direct_cache[scan8[0] - 1 + 2*8]= 0;
834 }
835 }
836 }
837 }
838 }
839 #endif
840 }
841
842 static inline void write_back_intra_pred_mode(H264Context *h){
843 MpegEncContext * const s = &h->s;
844 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
845
846 h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
847 h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
848 h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
849 h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
850 h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
851 h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
852 h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
853 }
854
855 /**
856 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
857 */
858 static inline int check_intra4x4_pred_mode(H264Context *h){
859 MpegEncContext * const s = &h->s;
860 static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
861 static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
862 int i;
863
864 if(!(h->top_samples_available&0x8000)){
865 for(i=0; i<4; i++){
866 int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
867 if(status<0){
868 av_log(h->s.avctx, AV_LOG_ERROR, "top block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y);
869 return -1;
870 } else if(status){
871 h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
872 }
873 }
874 }
875
876 if(!(h->left_samples_available&0x8000)){
877 for(i=0; i<4; i++){
878 int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
879 if(status<0){
880 av_log(h->s.avctx, AV_LOG_ERROR, "left block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y);
881 return -1;
882 } else if(status){
883 h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
884 }
885 }
886 }
887
888 return 0;
889 } //FIXME cleanup like next
890
891 /**
892 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
893 */
894 static inline int check_intra_pred_mode(H264Context *h, int mode){
895 MpegEncContext * const s = &h->s;
896 static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
897 static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
898
899 if(mode < 0 || mode > 6) {
900 av_log(h->s.avctx, AV_LOG_ERROR, "out of range intra chroma pred mode at %d %d\n", s->mb_x, s->mb_y);
901 return -1;
902 }
903
904 if(!(h->top_samples_available&0x8000)){
905 mode= top[ mode ];
906 if(mode<0){
907 av_log(h->s.avctx, AV_LOG_ERROR, "top block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);
908 return -1;
909 }
910 }
911
912 if(!(h->left_samples_available&0x8000)){
913 mode= left[ mode ];
914 if(mode<0){
915 av_log(h->s.avctx, AV_LOG_ERROR, "left block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);
916 return -1;
917 }
918 }
919
920 return mode;
921 }
922
923 /**
924 * gets the predicted intra4x4 prediction mode.
925 */
926 static inline int pred_intra_mode(H264Context *h, int n){
927 const int index8= scan8[n];
928 const int left= h->intra4x4_pred_mode_cache[index8 - 1];
929 const int top = h->intra4x4_pred_mode_cache[index8 - 8];
930 const int min= FFMIN(left, top);
931
932 tprintf("mode:%d %d min:%d\n", left ,top, min);
933
934 if(min<0) return DC_PRED;
935 else return min;
936 }
937
938 static inline void write_back_non_zero_count(H264Context *h){
939 MpegEncContext * const s = &h->s;
940 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
941
942 h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[7+8*1];
943 h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[7+8*2];
944 h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[7+8*3];
945 h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];
946 h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[4+8*4];
947 h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[5+8*4];
948 h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[6+8*4];
949
950 h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[1+8*2];
951 h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
952 h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[2+8*1];
953
954 h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[1+8*5];
955 h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
956 h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[2+8*4];
957 }
958
959 /**
960 * gets the predicted number of non zero coefficients.
961 * @param n block index
962 */
963 static inline int pred_non_zero_count(H264Context *h, int n){
964 const int index8= scan8[n];
965 const int left= h->non_zero_count_cache[index8 - 1];
966 const int top = h->non_zero_count_cache[index8 - 8];
967 int i= left + top;
968
969 if(i<64) i= (i+1)>>1;
970
971 tprintf("pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
972
973 return i&31;
974 }
975
976 static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
977 const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];
978
979 if(topright_ref != PART_NOT_AVAILABLE){
980 *C= h->mv_cache[list][ i - 8 + part_width ];
981 return topright_ref;
982 }else{
983 tprintf("topright MV not available\n");
984
985 *C= h->mv_cache[list][ i - 8 - 1 ];
986 return h->ref_cache[list][ i - 8 - 1 ];
987 }
988 }
989
990 /**
991 * gets the predicted MV.
992 * @param n the block index
993 * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)
994 * @param mx the x component of the predicted motion vector
995 * @param my the y component of the predicted motion vector
996 */
997 static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
998 const int index8= scan8[n];
999 const int top_ref= h->ref_cache[list][ index8 - 8 ];
1000 const int left_ref= h->ref_cache[list][ index8 - 1 ];
1001 const int16_t * const A= h->mv_cache[list][ index8 - 1 ];
1002 const int16_t * const B= h->mv_cache[list][ index8 - 8 ];
1003 const int16_t * C;
1004 int diagonal_ref, match_count;
1005
1006 assert(part_width==1 || part_width==2 || part_width==4);
1007
1008 /* mv_cache
1009 B . . A T T T T
1010 U . . L . . , .
1011 U . . L . . . .
1012 U . . L . . , .
1013 . . . L . . . .
1014 */
1015
1016 diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);
1017 match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);
1018 tprintf("pred_motion match_count=%d\n", match_count);
1019 if(match_count > 1){ //most common
1020 *mx= mid_pred(A[0], B[0], C[0]);
1021 *my= mid_pred(A[1], B[1], C[1]);
1022 }else if(match_count==1){
1023 if(left_ref==ref){
1024 *mx= A[0];
1025 *my= A[1];
1026 }else if(top_ref==ref){
1027 *mx= B[0];
1028 *my= B[1];
1029 }else{
1030 *mx= C[0];
1031 *my= C[1];
1032 }
1033 }else{
1034 if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
1035 *mx= A[0];
1036 *my= A[1];
1037 }else{
1038 *mx= mid_pred(A[0], B[0], C[0]);
1039 *my= mid_pred(A[1], B[1], C[1]);
1040 }
1041 }
1042
1043 tprintf("pred_motion (%2d %2d %2d) (%2d %2d %2d) (%2d %2d %2d) -> (%2d %2d %2d) at %2d %2d %d list %d\n", top_ref, B[0], B[1], diagonal_ref, C[0], C[1], left_ref, A[0], A[1], ref, *mx, *my, h->s.mb_x, h->s.mb_y, n, list);
1044 }
1045
1046 /**
1047 * gets the directionally predicted 16x8 MV.
1048 * @param n the block index
1049 * @param mx the x component of the predicted motion vector
1050 * @param my the y component of the predicted motion vector
1051 */
1052 static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
1053 if(n==0){
1054 const int top_ref= h->ref_cache[list][ scan8[0] - 8 ];
1055 const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
1056
1057 tprintf("pred_16x8: (%2d %2d %2d) at %2d %2d %d list %d\n", top_ref, B[0], B[1], h->s.mb_x, h->s.mb_y, n, list);
1058
1059 if(top_ref == ref){
1060 *mx= B[0];
1061 *my= B[1];
1062 return;
1063 }
1064 }else{
1065 const int left_ref= h->ref_cache[list][ scan8[8] - 1 ];
1066 const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];
1067
1068 tprintf("pred_16x8: (%2d %2d %2d) at %2d %2d %d list %d\n", left_ref, A[0], A[1], h->s.mb_x, h->s.mb_y, n, list);
1069
1070 if(left_ref == ref){
1071 *mx= A[0];
1072 *my= A[1];
1073 return;
1074 }
1075 }
1076
1077 //RARE
1078 pred_motion(h, n, 4, list, ref, mx, my);
1079 }
1080
1081 /**
1082 * gets the directionally predicted 8x16 MV.
1083 * @param n the block index
1084 * @param mx the x component of the predicted motion vector
1085 * @param my the y component of the predicted motion vector
1086 */
1087 static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
1088 if(n==0){
1089 const int left_ref= h->ref_cache[list][ scan8[0] - 1 ];
1090 const int16_t * const A= h->mv_cache[list][ scan8[0] - 1 ];
1091
1092 tprintf("pred_8x16: (%2d %2d %2d) at %2d %2d %d list %d\n", left_ref, A[0], A[1], h->s.mb_x, h->s.mb_y, n, list);
1093
1094 if(left_ref == ref){
1095 *mx= A[0];
1096 *my= A[1];
1097 return;
1098 }
1099 }else{
1100 const int16_t * C;
1101 int diagonal_ref;
1102
1103 diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);
1104
1105 tprintf("pred_8x16: (%2d %2d %2d) at %2d %2d %d list %d\n", diagonal_ref, C[0], C[1], h->s.mb_x, h->s.mb_y, n, list);
1106
1107 if(diagonal_ref == ref){
1108 *mx= C[0];
1109 *my= C[1];
1110 return;
1111 }
1112 }
1113
1114 //RARE
1115 pred_motion(h, n, 2, list, ref, mx, my);
1116 }
1117
1118 static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
1119 const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
1120 const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];
1121
1122 tprintf("pred_pskip: (%d) (%d) at %2d %2d\n", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
1123
1124 if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
1125 || (top_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0)
1126 || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){
1127
1128 *mx = *my = 0;
1129 return;
1130 }
1131
1132 pred_motion(h, 0, 4, 0, 0, mx, my);
1133
1134 return;
1135 }
1136
1137 static inline void direct_dist_scale_factor(H264Context * const h){
1138 const int poc = h->s.current_picture_ptr->poc;
1139 const int poc1 = h->ref_list[1][0].poc;
1140 int i;
1141 for(i=0; i<h->ref_count[0]; i++){
1142 int poc0 = h->ref_list[0][i].poc;
1143 int td = clip(poc1 - poc0, -128, 127);
1144 if(td == 0 /* FIXME || pic0 is a long-term ref */){
1145 h->dist_scale_factor[i] = 256;
1146 }else{
1147 int tb = clip(poc - poc0, -128, 127);
1148 int tx = (16384 + (ABS(td) >> 1)) / td;
1149 h->dist_scale_factor[i] = clip((tb*tx + 32) >> 6, -1024, 1023);
1150 }
1151 }
1152 }
1153 static inline void direct_ref_list_init(H264Context * const h){
1154 MpegEncContext * const s = &h->s;
1155 Picture * const ref1 = &h->ref_list[1][0];
1156 Picture * const cur = s->current_picture_ptr;
1157 int list, i, j;
1158 if(cur->pict_type == I_TYPE)
1159 cur->ref_count[0] = 0;
1160 if(cur->pict_type != B_TYPE)
1161 cur->ref_count[1] = 0;
1162 for(list=0; list<2; list++){
1163 cur->ref_count[list] = h->ref_count[list];
1164 for(j=0; j<h->ref_count[list]; j++)
1165 cur->ref_poc[list][j] = h->ref_list[list][j].poc;
1166 }
1167 if(cur->pict_type != B_TYPE || h->direct_spatial_mv_pred)
1168 return;
1169 for(list=0; list<2; list++){
1170 for(i=0; i<ref1->ref_count[list]; i++){
1171 const int poc = ref1->ref_poc[list][i];
1172 h->map_col_to_list0[list][i] = PART_NOT_AVAILABLE;
1173 for(j=0; j<h->ref_count[list]; j++)
1174 if(h->ref_list[list][j].poc == poc){
1175 h->map_col_to_list0[list][i] = j;
1176 break;
1177 }
1178 }
1179 }
1180 }
1181
1182 static inline void pred_direct_motion(H264Context * const h, int *mb_type){
1183 MpegEncContext * const s = &h->s;
1184 const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
1185 const int b8_xy = 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1186 const int b4_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1187 const int mb_type_col = h->ref_list[1][0].mb_type[mb_xy];
1188 const int16_t (*l1mv0)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[0][b4_xy];
1189 const int8_t *l1ref0 = &h->ref_list[1][0].ref_index[0][b8_xy];
1190 const int8_t *l1ref1 = &h->ref_list[1][0].ref_index[1][b8_xy];
1191 const int is_b8x8 = IS_8X8(*mb_type);
1192 int sub_mb_type;
1193 int i8, i4;
1194
1195 if(IS_8X8(mb_type_col) && !h->sps.direct_8x8_inference_flag){
1196 /* FIXME save sub mb types from previous frames (or derive from MVs)
1197 * so we know exactly what block size to use */
1198 sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */
1199 *mb_type = MB_TYPE_8x8|MB_TYPE_L0L1;
1200 }else if(!is_b8x8 && (IS_16X16(mb_type_col) || IS_INTRA(mb_type_col))){
1201 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1202 *mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */
1203 }else{
1204 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1205 *mb_type = MB_TYPE_8x8|MB_TYPE_L0L1;
1206 }
1207 if(!is_b8x8)
1208 *mb_type |= MB_TYPE_DIRECT2;
1209
1210 tprintf("mb_type = %08x, sub_mb_type = %08x, is_b8x8 = %d, mb_type_col = %08x\n", *mb_type, sub_mb_type, is_b8x8, mb_type_col);
1211
1212 if(h->direct_spatial_mv_pred){
1213 int ref[2];
1214 int mv[2][2];
1215 int list;
1216
1217 /* ref = min(neighbors) */
1218 for(list=0; list<2; list++){
1219 int refa = h->ref_cache[list][scan8[0] - 1];
1220 int refb = h->ref_cache[list][scan8[0] - 8];
1221 int refc = h->ref_cache[list][scan8[0] - 8 + 4];
1222 if(refc == -2)
1223 refc = h->ref_cache[list][scan8[0] - 8 - 1];
1224 ref[list] = refa;
1225 if(ref[list] < 0 || (refb < ref[list] && refb >= 0))
1226 ref[list] = refb;
1227 if(ref[list] < 0 || (refc < ref[list] && refc >= 0))
1228 ref[list] = refc;
1229 if(ref[list] < 0)
1230 ref[list] = -1;
1231 }
1232
1233 if(ref[0] < 0 && ref[1] < 0){
1234 ref[0] = ref[1] = 0;
1235 mv[0][0] = mv[0][1] =
1236 mv[1][0] = mv[1][1] = 0;
1237 }else{
1238 for(list=0; list<2; list++){
1239 if(ref[list] >= 0)
1240 pred_motion(h, 0, 4, list, ref[list], &mv[list][0], &mv[list][1]);
1241 else
1242 mv[list][0] = mv[list][1] = 0;
1243 }
1244 }
1245
1246 if(ref[1] < 0){
1247 *mb_type &= ~MB_TYPE_P0L1;
1248 sub_mb_type &= ~MB_TYPE_P0L1;
1249 }else if(ref[0] < 0){
1250 *mb_type &= ~MB_TYPE_P0L0;
1251 sub_mb_type &= ~MB_TYPE_P0L0;
1252 }
1253
1254 if(IS_16X16(*mb_type)){
1255 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref[0], 1);
1256 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, ref[1], 1);
1257 if(!IS_INTRA(mb_type_col) && l1ref0[0] == 0 &&
1258 ABS(l1mv0[0][0]) <= 1 && ABS(l1mv0[0][1]) <= 1){
1259 if(ref[0] > 0)
1260 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1261 else
1262 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
1263 if(ref[1] > 0)
1264 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1265 else
1266 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
1267 }else{
1268 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1269 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1270 }
1271 }else{
1272 for(i8=0; i8<4; i8++){
1273 const int x8 = i8&1;
1274 const int y8 = i8>>1;
1275
1276 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1277 continue;
1278 h->sub_mb_type[i8] = sub_mb_type;
1279
1280 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1281 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1282 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref[0], 1);
1283 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, ref[1], 1);
1284
1285 /* col_zero_flag */
1286 if(!IS_INTRA(mb_type_col) && l1ref0[x8 + y8*h->b8_stride] == 0){
1287 for(i4=0; i4<4; i4++){
1288 const int16_t *mv_col = l1mv0[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1289 if(ABS(mv_col[0]) <= 1 && ABS(mv_col[1]) <= 1){
1290 if(ref[0] == 0)
1291 *(uint32_t*)h->mv_cache[0][scan8[i8*4+i4]] = 0;
1292 if(ref[1] == 0)
1293 *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = 0;
1294 }
1295 }
1296 }
1297 }
1298 }
1299 }else{ /* direct temporal mv pred */
1300 if(IS_16X16(*mb_type)){
1301 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1);
1302 if(IS_INTRA(mb_type_col)){
1303 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
1304 fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
1305 fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
1306 }else{
1307 const int ref0 = l1ref0[0] >= 0 ? h->map_col_to_list0[0][l1ref0[0]]
1308 : h->map_col_to_list0[1][l1ref1[0]];
1309 const int dist_scale_factor = h->dist_scale_factor[ref0];
1310 const int16_t *mv_col = l1mv0[0];
1311 int mv_l0[2];
1312 mv_l0[0] = (dist_scale_factor * mv_col[0] + 128) >> 8;
1313 mv_l0[1] = (dist_scale_factor * mv_col[1] + 128) >> 8;
1314 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref0, 1);
1315 fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv_l0[0],mv_l0[1]), 4);
1316 fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]), 4);
1317 }
1318 }else{
1319 for(i8=0; i8<4; i8++){
1320 const int x8 = i8&1;
1321 const int y8 = i8>>1;
1322 int ref0, dist_scale_factor;
1323
1324 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1325 continue;
1326 h->sub_mb_type[i8] = sub_mb_type;
1327 if(IS_INTRA(mb_type_col)){
1328 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1329 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1330 fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1331 fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1332 continue;
1333 }
1334
1335 ref0 = l1ref0[x8 + y8*h->b8_stride];
1336 if(ref0 >= 0)
1337 ref0 = h->map_col_to_list0[0][ref0];
1338 else
1339 ref0 = h->map_col_to_list0[1][l1ref1[x8 + y8*h->b8_stride]];
1340 dist_scale_factor = h->dist_scale_factor[ref0];
1341
1342 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1343 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1344 for(i4=0; i4<4; i4++){
1345 const int16_t *mv_col = l1mv0[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1346 int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]];
1347 mv_l0[0] = (dist_scale_factor * mv_col[0] + 128) >> 8;
1348 mv_l0[1] = (dist_scale_factor * mv_col[1] + 128) >> 8;
1349 *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] =
1350 pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
1351 }
1352 }
1353 }
1354 }
1355 }
1356
1357 static inline void write_back_motion(H264Context *h, int mb_type){
1358 MpegEncContext * const s = &h->s;
1359 const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1360 const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1361 int list;
1362
1363 for(list=0; list<2; list++){
1364 int y;
1365 if(!USES_LIST(mb_type, list)){
1366 if(1){ //FIXME skip or never read if mb_type doesnt use it
1367 for(y=0; y<4; y++){
1368 *(uint64_t*)s->current_picture.motion_val[list][b_xy + 0 + y*h->b_stride]=
1369 *(uint64_t*)s->current_picture.motion_val[list][b_xy + 2 + y*h->b_stride]= 0;
1370 }
1371 if( h->pps.cabac ) {
1372 /* FIXME needed ? */
1373 for(y=0; y<4; y++){
1374 *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]=
1375 *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= 0;
1376 }
1377 }
1378 for(y=0; y<2; y++){
1379 *(uint16_t*)&s->current_picture.ref_index[list][b8_xy + y*h->b8_stride]= (LIST_NOT_USED&0xFF)*0x0101;
1380 }
1381 }
1382 continue;
1383 }
1384
1385 for(y=0; y<4; y++){
1386 *(uint64_t*)s->current_picture.motion_val[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mv_cache[list][scan8[0]+0 + 8*y];
1387 *(uint64_t*)s->current_picture.motion_val[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mv_cache[list][scan8[0]+2 + 8*y];
1388 }
1389 if( h->pps.cabac ) {
1390 for(y=0; y<4; y++){
1391 *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+0 + 8*y];
1392 *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+2 + 8*y];
1393 }
1394 }
1395 for(y=0; y<2; y++){
1396 s->current_picture.ref_index[list][b8_xy + 0 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+0 + 16*y];
1397 s->current_picture.ref_index[list][b8_xy + 1 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+2 + 16*y];
1398 }
1399 }
1400
1401 if(h->slice_type == B_TYPE && h->pps.cabac){
1402 if(IS_8X8(mb_type)){
1403 h->direct_table[b8_xy+1+0*h->b8_stride] = IS_DIRECT(h->sub_mb_type[1]) ? 1 : 0;
1404 h->direct_table[b8_xy+0+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[2]) ? 1 : 0;
1405 h->direct_table[b8_xy+1+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[3]) ? 1 : 0;
1406 }
1407 }
1408 }
1409
1410 /**
1411 * Decodes a network abstraction layer unit.
1412 * @param consumed is the number of bytes used as input
1413 * @param length is the length of the array
1414 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp ttailing?
1415 * @returns decoded bytes, might be src+1 if no escapes
1416 */
1417 static uint8_t *decode_nal(H264Context *h, uint8_t *src, int *dst_length, int *consumed, int length){
1418 int i, si, di;
1419 uint8_t *dst;
1420
1421 // src[0]&0x80; //forbidden bit
1422 h->nal_ref_idc= src[0]>>5;
1423 h->nal_unit_type= src[0]&0x1F;
1424
1425 src++; length--;
1426 #if 0
1427 for(i=0; i<length; i++)
1428 printf("%2X ", src[i]);
1429 #endif
1430 for(i=0; i+1<length; i+=2){
1431 if(src[i]) continue;
1432 if(i>0 && src[i-1]==0) i--;
1433 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1434 if(src[i+2]!=3){
1435 /* startcode, so we must be past the end */
1436 length=i;
1437 }
1438 break;
1439 }
1440 }
1441
1442 if(i>=length-1){ //no escaped 0
1443 *dst_length= length;
1444 *consumed= length+1; //+1 for the header
1445 return src;
1446 }
1447
1448 h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length);
1449 dst= h->rbsp_buffer;
1450
1451 //printf("deoding esc\n");
1452 si=di=0;
1453 while(si<length){
1454 //remove escapes (very rare 1:2^22)
1455 if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1456 if(src[si+2]==3){ //escape
1457 dst[di++]= 0;
1458 dst[di++]= 0;
1459 si+=3;
1460 continue;
1461 }else //next start code
1462 break;
1463 }
1464
1465 dst[di++]= src[si++];
1466 }
1467
1468 *dst_length= di;
1469 *consumed= si + 1;//+1 for the header
1470 //FIXME store exact number of bits in the getbitcontext (its needed for decoding)
1471 return dst;
1472 }
1473
1474 #if 0
1475 /**
1476 * @param src the data which should be escaped
1477 * @param dst the target buffer, dst+1 == src is allowed as a special case
1478 * @param length the length of the src data
1479 * @param dst_length the length of the dst array
1480 * @returns length of escaped data in bytes or -1 if an error occured
1481 */
1482 static int encode_nal(H264Context *h, uint8_t *dst, uint8_t *src, int length, int dst_length){
1483 int i, escape_count, si, di;
1484 uint8_t *temp;
1485
1486 assert(length>=0);
1487 assert(dst_length>0);
1488
1489 dst[0]= (h->nal_ref_idc<<5) + h->nal_unit_type;
1490
1491 if(length==0) return 1;
1492
1493 escape_count= 0;
1494 for(i=0; i<length; i+=2){
1495 if(src[i]) continue;
1496 if(i>0 && src[i-1]==0)
1497 i--;
1498 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1499 escape_count++;
1500 i+=2;
1501 }
1502 }
1503
1504 if(escape_count==0){
1505 if(dst+1 != src)
1506 memcpy(dst+1, src, length);
1507 return length + 1;
1508 }
1509
1510 if(length + escape_count + 1> dst_length)
1511 return -1;
1512
1513 //this should be damn rare (hopefully)
1514
1515 h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length + escape_count);
1516 temp= h->rbsp_buffer;
1517 //printf("encoding esc\n");
1518
1519 si= 0;
1520 di= 0;
1521 while(si < length){
1522 if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1523 temp[di++]= 0; si++;
1524 temp[di++]= 0; si++;
1525 temp[di++]= 3;
1526 temp[di++]= src[si++];
1527 }
1528 else
1529 temp[di++]= src[si++];
1530 }
1531 memcpy(dst+1, temp, length+escape_count);
1532
1533 assert(di == length+escape_count);
1534
1535 return di + 1;
1536 }
1537
1538 /**
1539 * write 1,10,100,1000,... for alignment, yes its exactly inverse to mpeg4
1540 */
1541 static void encode_rbsp_trailing(PutBitContext *pb){
1542 int length;
1543 put_bits(pb, 1, 1);
1544 length= (-put_bits_count(pb))&7;
1545 if(length) put_bits(pb, length, 0);
1546 }
1547 #endif
1548
1549 /**
1550 * identifies the exact end of the bitstream
1551 * @return the length of the trailing, or 0 if damaged
1552 */
1553 static int decode_rbsp_trailing(uint8_t *src){
1554 int v= *src;
1555 int r;
1556
1557 tprintf("rbsp trailing %X\n", v);
1558
1559 for(r=1; r<9; r++){
1560 if(v&1) return r;
1561 v>>=1;
1562 }
1563 return 0;
1564 }
1565
1566 /**
1567 * idct tranforms the 16 dc values and dequantize them.
1568 * @param qp quantization parameter
1569 */
1570 static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp){
1571 const int qmul= dequant_coeff[qp][0];
1572 #define stride 16
1573 int i;
1574 int temp[16]; //FIXME check if this is a good idea
1575 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
1576 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1577
1578 //memset(block, 64, 2*256);
1579 //return;
1580 for(i=0; i<4; i++){
1581 const int offset= y_offset[i];
1582 const int z0= block[offset+stride*0] + block[offset+stride*4];
1583 const int z1= block[offset+stride*0] - block[offset+stride*4];
1584 const int z2= block[offset+stride*1] - block[offset+stride*5];
1585 const int z3= block[offset+stride*1] + block[offset+stride*5];
1586
1587 temp[4*i+0]= z0+z3;
1588 temp[4*i+1]= z1+z2;
1589 temp[4*i+2]= z1-z2;
1590 temp[4*i+3]= z0-z3;
1591 }
1592
1593 for(i=0; i<4; i++){
1594 const int offset= x_offset[i];
1595 const int z0= temp[4*0+i] + temp[4*2+i];
1596 const int z1= temp[4*0+i] - temp[4*2+i];
1597 const int z2= temp[4*1+i] - temp[4*3+i];
1598 const int z3= temp[4*1+i] + temp[4*3+i];
1599
1600 block[stride*0 +offset]= ((z0 + z3)*qmul + 2)>>2; //FIXME think about merging this into decode_resdual
1601 block[stride*2 +offset]= ((z1 + z2)*qmul + 2)>>2;
1602 block[stride*8 +offset]= ((z1 - z2)*qmul + 2)>>2;
1603 block[stride*10+offset]= ((z0 - z3)*qmul + 2)>>2;
1604 }
1605 }
1606
1607 #if 0
1608 /**
1609 * dct tranforms the 16 dc values.
1610 * @param qp quantization parameter ??? FIXME
1611 */
1612 static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
1613 // const int qmul= dequant_coeff[qp][0];
1614 int i;
1615 int temp[16]; //FIXME check if this is a good idea
1616 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
1617 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1618
1619 for(i=0; i<4; i++){
1620 const int offset= y_offset[i];
1621 const int z0= block[offset+stride*0] + block[offset+stride*4];
1622 const int z1= block[offset+stride*0] - block[offset+stride*4];
1623 const int z2= block[offset+stride*1] - block[offset+stride*5];
1624 const int z3= block[offset+stride*1] + block[offset+stride*5];
1625
1626 temp[4*i+0]= z0+z3;
1627 temp[4*i+1]= z1+z2;
1628 temp[4*i+2]= z1-z2;
1629 temp[4*i+3]= z0-z3;
1630 }
1631
1632 for(i=0; i<4; i++){
1633 const int offset= x_offset[i];
1634 const int z0= temp[4*0+i] + temp[4*2+i];
1635 const int z1= temp[4*0+i] - temp[4*2+i];
1636 const int z2= temp[4*1+i] - temp[4*3+i];
1637 const int z3= temp[4*1+i] + temp[4*3+i];
1638
1639 block[stride*0 +offset]= (z0 + z3)>>1;
1640 block[stride*2 +offset]= (z1 + z2)>>1;
1641 block[stride*8 +offset]= (z1 - z2)>>1;
1642 block[stride*10+offset]= (z0 - z3)>>1;
1643 }
1644 }
1645 #endif
1646
1647 #undef xStride
1648 #undef stride
1649
1650 static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp){
1651 const int qmul= dequant_coeff[qp][0];
1652 const int stride= 16*2;
1653 const int xStride= 16;
1654 int a,b,c,d,e;
1655
1656 a= block[stride*0 + xStride*0];
1657 b= block[stride*0 + xStride*1];
1658 c= block[stride*1 + xStride*0];
1659 d= block[stride*1 + xStride*1];
1660
1661 e= a-b;
1662 a= a+b;
1663 b= c-d;
1664 c= c+d;
1665
1666 block[stride*0 + xStride*0]= ((a+c)*qmul + 0)>>1;
1667 block[stride*0 + xStride*1]= ((e+b)*qmul + 0)>>1;
1668 block[stride*1 + xStride*0]= ((a-c)*qmul + 0)>>1;
1669 block[stride*1 + xStride*1]= ((e-b)*qmul + 0)>>1;
1670 }
1671
1672 #if 0
1673 static void chroma_dc_dct_c(DCTELEM *block){
1674 const int stride= 16*2;
1675 const int xStride= 16;
1676 int a,b,c,d,e;
1677
1678 a= block[stride*0 + xStride*0];
1679 b= block[stride*0 + xStride*1];
1680 c= block[stride*1 + xStride*0];
1681 d= block[stride*1 + xStride*1];
1682
1683 e= a-b;
1684 a= a+b;
1685 b= c-d;
1686 c= c+d;
1687
1688 block[stride*0 + xStride*0]= (a+c);
1689 block[stride*0 + xStride*1]= (e+b);
1690 block[stride*1 + xStride*0]= (a-c);
1691 block[stride*1 + xStride*1]= (e-b);
1692 }
1693 #endif
1694
1695 /**
1696 * gets the chroma qp.
1697 */
1698 static inline int get_chroma_qp(int chroma_qp_index_offset, int qscale){
1699
1700 return chroma_qp[clip(qscale + chroma_qp_index_offset, 0, 51)];
1701 }
1702
1703
1704 #if 0
1705 static void h264_diff_dct_c(DCTELEM *block, uint8_t *src1, uint8_t *src2, int stride){
1706 int i;
1707 //FIXME try int temp instead of block
1708
1709 for(i=0; i<4; i++){
1710 const int d0= src1[0 + i*stride] - src2[0 + i*stride];
1711 const int d1= src1[1 + i*stride] - src2[1 + i*stride];
1712 const int d2= src1[2 + i*stride] - src2[2 + i*stride];
1713 const int d3= src1[3 + i*stride] - src2[3 + i*stride];
1714 const int z0= d0 + d3;
1715 const int z3= d0 - d3;
1716 const int z1= d1 + d2;
1717 const int z2= d1 - d2;
1718
1719 block[0 + 4*i]= z0 + z1;
1720 block[1 + 4*i]= 2*z3 + z2;
1721 block[2 + 4*i]= z0 - z1;
1722 block[3 + 4*i]= z3 - 2*z2;
1723 }
1724
1725 for(i=0; i<4; i++){
1726 const int z0= block[0*4 + i] + block[3*4 + i];
1727 const int z3= block[0*4 + i] - block[3*4 + i];
1728 const int z1= block[1*4 + i] + block[2*4 + i];
1729 const int z2= block[1*4 + i] - block[2*4 + i];
1730
1731 block[0*4 + i]= z0 + z1;
1732 block[1*4 + i]= 2*z3 + z2;
1733 block[2*4 + i]= z0 - z1;
1734 block[3*4 + i]= z3 - 2*z2;
1735 }
1736 }
1737 #endif
1738
1739 //FIXME need to check that this doesnt overflow signed 32 bit for low qp, iam not sure, its very close
1740 //FIXME check that gcc inlines this (and optimizes intra & seperate_dc stuff away)
1741 static inline int quantize_c(DCTELEM *block, uint8_t *scantable, int qscale, int intra, int seperate_dc){
1742 int i;
1743 const int * const quant_table= quant_coeff[qscale];
1744 const int bias= intra ? (1<<QUANT_SHIFT)/3 : (1<<QUANT_SHIFT)/6;
1745 const unsigned int threshold1= (1<<QUANT_SHIFT) - bias - 1;
1746 const unsigned int threshold2= (threshold1<<1);
1747 int last_non_zero;
1748
1749 if(seperate_dc){
1750 if(qscale<=18){
1751 //avoid overflows
1752 const int dc_bias= intra ? (1<<(QUANT_SHIFT-2))/3 : (1<<(QUANT_SHIFT-2))/6;
1753 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT-2)) - dc_bias - 1;
1754 const unsigned int dc_threshold2= (dc_threshold1<<1);
1755
1756 int level= block[0]*quant_coeff[qscale+18][0];
1757 if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1758 if(level>0){
1759 level= (dc_bias + level)>>(QUANT_SHIFT-2);
1760 block[0]= level;
1761 }else{
1762 level= (dc_bias - level)>>(QUANT_SHIFT-2);
1763 block[0]= -level;
1764 }
1765 // last_non_zero = i;
1766 }else{
1767 block[0]=0;
1768 }
1769 }else{
1770 const int dc_bias= intra ? (1<<(QUANT_SHIFT+1))/3 : (1<<(QUANT_SHIFT+1))/6;
1771 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT+1)) - dc_bias - 1;
1772 const unsigned int dc_threshold2= (dc_threshold1<<1);
1773
1774 int level= block[0]*quant_table[0];
1775 if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1776 if(level>0){
1777 level= (dc_bias + level)>>(QUANT_SHIFT+1);
1778 block[0]= level;
1779 }else{
1780 level= (dc_bias - level)>>(QUANT_SHIFT+1);
1781 block[0]= -level;
1782 }
1783 // last_non_zero = i;
1784 }else{
1785 block[0]=0;
1786 }
1787 }
1788 last_non_zero= 0;
1789 i=1;
1790 }else{
1791 last_non_zero= -1;
1792 i=0;
1793 }
1794
1795 for(; i<16; i++){
1796 const int j= scantable[i];
1797 int level= block[j]*quant_table[j];
1798
1799 // if( bias+level >= (1<<(QMAT_SHIFT - 3))
1800 // || bias-level >= (1<<(QMAT_SHIFT - 3))){
1801 if(((unsigned)(level+threshold1))>threshold2){
1802 if(level>0){
1803 level= (bias + level)>>QUANT_SHIFT;
1804 block[j]= level;
1805 }else{
1806 level= (bias - level)>>QUANT_SHIFT;
1807 block[j]= -level;
1808 }
1809 last_non_zero = i;
1810 }else{
1811 block[j]=0;
1812 }
1813 }
1814
1815 return last_non_zero;
1816 }
1817
1818 static void pred4x4_vertical_c(uint8_t *src, uint8_t *topright, int stride){
1819 const uint32_t a= ((uint32_t*)(src-stride))[0];
1820 ((uint32_t*)(src+0*stride))[0]= a;
1821 ((uint32_t*)(src+1*stride))[0]= a;
1822 ((uint32_t*)(src+2*stride))[0]= a;
1823 ((uint32_t*)(src+3*stride))[0]= a;
1824 }
1825
1826 static void pred4x4_horizontal_c(uint8_t *src, uint8_t *topright, int stride){
1827 ((uint32_t*)(src+0*stride))[0]= src[-1+0*stride]*0x01010101;
1828 ((uint32_t*)(src+1*stride))[0]= src[-1+1*stride]*0x01010101;
1829 ((uint32_t*)(src+2*stride))[0]= src[-1+2*stride]*0x01010101;
1830 ((uint32_t*)(src+3*stride))[0]= src[-1+3*stride]*0x01010101;
1831 }
1832
1833 static void pred4x4_dc_c(uint8_t *src, uint8_t *topright, int stride){
1834 const int dc= ( src[-stride] + src[1-stride] + src[2-stride] + src[3-stride]
1835 + src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 4) >>3;
1836
1837 ((uint32_t*)(src+0*stride))[0]=
1838 ((uint32_t*)(src+1*stride))[0]=
1839 ((uint32_t*)(src+2*stride))[0]=
1840 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1841 }
1842
1843 static void pred4x4_left_dc_c(uint8_t *src, uint8_t *topright, int stride){
1844 const int dc= ( src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 2) >>2;
1845
1846 ((uint32_t*)(src+0*stride))[0]=
1847 ((uint32_t*)(src+1*stride))[0]=
1848 ((uint32_t*)(src+2*stride))[0]=
1849 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1850 }
1851
1852 static void pred4x4_top_dc_c(uint8_t *src, uint8_t *topright, int stride){
1853 const int dc= ( src[-stride] + src[1-stride] + src[2-stride] + src[3-stride] + 2) >>2;
1854
1855 ((uint32_t*)(src+0*stride))[0]=
1856 ((uint32_t*)(src+1*stride))[0]=
1857 ((uint32_t*)(src+2*stride))[0]=
1858 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1859 }
1860
1861 static void pred4x4_128_dc_c(uint8_t *src, uint8_t *topright, int stride){
1862 ((uint32_t*)(src+0*stride))[0]=
1863 ((uint32_t*)(src+1*stride))[0]=
1864 ((uint32_t*)(src+2*stride))[0]=
1865 ((uint32_t*)(src+3*stride))[0]= 128U*0x01010101U;
1866 }
1867
1868
1869 #define LOAD_TOP_RIGHT_EDGE\
1870 const int t4= topright[0];\
1871 const int t5= topright[1];\
1872 const int t6= topright[2];\
1873 const int t7= topright[3];\
1874
1875 #define LOAD_LEFT_EDGE\
1876 const int l0= src[-1+0*stride];\
1877 const int l1= src[-1+1*stride];\
1878 const int l2= src[-1+2*stride];\
1879 const int l3= src[-1+3*stride];\
1880
1881 #define LOAD_TOP_EDGE\
1882 const int t0= src[ 0-1*stride];\
1883 const int t1= src[ 1-1*stride];\
1884 const int t2= src[ 2-1*stride];\
1885 const int t3= src[ 3-1*stride];\
1886
1887 static void pred4x4_down_right_c(uint8_t *src, uint8_t *topright, int stride){
1888 const int lt= src[-1-1*stride];
1889 LOAD_TOP_EDGE
1890 LOAD_LEFT_EDGE
1891
1892 src[0+3*stride]=(l3 + 2*l2 + l1 + 2)>>2;
1893 src[0+2*stride]=
1894 src[1+3*stride]=(l2 + 2*l1 + l0 + 2)>>2;
1895 src[0+1*stride]=
1896 src[1+2*stride]=
1897 src[2+3*stride]=(l1 + 2*l0 + lt + 2)>>2;
1898 src[0+0*stride]=
1899 src[1+1*stride]=
1900 src[2+2*stride]=
1901 src[3+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1902 src[1+0*stride]=
1903 src[2+1*stride]=
1904 src[3+2*stride]=(lt + 2*t0 + t1 + 2)>>2;
1905 src[2+0*stride]=
1906 src[3+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1907 src[3+0*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1908 }
1909
1910 static void pred4x4_down_left_c(uint8_t *src, uint8_t *topright, int stride){
1911 LOAD_TOP_EDGE
1912 LOAD_TOP_RIGHT_EDGE
1913 // LOAD_LEFT_EDGE
1914
1915 src[0+0*stride]=(t0 + t2 + 2*t1 + 2)>>2;
1916 src[1+0*stride]=
1917 src[0+1*stride]=(t1 + t3 + 2*t2 + 2)>>2;
1918 src[2+0*stride]=
1919 src[1+1*stride]=
1920 src[0+2*stride]=(t2 + t4 + 2*t3 + 2)>>2;
1921 src[3+0*stride]=
1922 src[2+1*stride]=
1923 src[1+2*stride]=
1924 src[0+3*stride]=(t3 + t5 + 2*t4 + 2)>>2;
1925 src[3+1*stride]=
1926 src[2+2*stride]=
1927 src[1+3*stride]=(t4 + t6 + 2*t5 + 2)>>2;
1928 src[3+2*stride]=
1929 src[2+3*stride]=(t5 + t7 + 2*t6 + 2)>>2;
1930 src[3+3*stride]=(t6 + 3*t7 + 2)>>2;
1931 }
1932
1933 static void pred4x4_vertical_right_c(uint8_t *src, uint8_t *topright, int stride){
1934 const int lt= src[-1-1*stride];
1935 LOAD_TOP_EDGE
1936 LOAD_LEFT_EDGE
1937 const __attribute__((unused)) int unu= l3;
1938
1939 src[0+0*stride]=
1940 src[1+2*stride]=(lt + t0 + 1)>>1;
1941 src[1+0*stride]=
1942 src[2+2*stride]=(t0 + t1 + 1)>>1;
1943 src[2+0*stride]=
1944 src[3+2*stride]=(t1 + t2 + 1)>>1;
1945 src[3+0*stride]=(t2 + t3 + 1)>>1;
1946 src[0+1*stride]=
1947 src[1+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1948 src[1+1*stride]=
1949 src[2+3*stride]=(lt + 2*t0 + t1 + 2)>>2;
1950 src[2+1*stride]=
1951 src[3+3*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1952 src[3+1*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1953 src[0+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1954 src[0+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1955 }
1956
1957 static void pred4x4_vertical_left_c(uint8_t *src, uint8_t *topright, int stride){
1958 LOAD_TOP_EDGE
1959 LOAD_TOP_RIGHT_EDGE
1960 const __attribute__((unused)) int unu= t7;
1961
1962 src[0+0*stride]=(t0 + t1 + 1)>>1;
1963 src[1+0*stride]=
1964 src[0+2*stride]=(t1 + t2 + 1)>>1;
1965 src[2+0*stride]=
1966 src[1+2*stride]=(t2 + t3 + 1)>>1;
1967 src[3+0*stride]=
1968 src[2+2*stride]=(t3 + t4+ 1)>>1;
1969 src[3+2*stride]=(t4 + t5+ 1)>>1;
1970 src[0+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1971 src[1+1*stride]=
1972 src[0+3*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1973 src[2+1*stride]=
1974 src[1+3*stride]=(t2 + 2*t3 + t4 + 2)>>2;
1975 src[3+1*stride]=
1976 src[2+3*stride]=(t3 + 2*t4 + t5 + 2)>>2;
1977 src[3+3*stride]=(t4 + 2*t5 + t6 + 2)>>2;
1978 }
1979
1980 static void pred4x4_horizontal_up_c(uint8_t *src, uint8_t *topright, int stride){
1981 LOAD_LEFT_EDGE
1982
1983 src[0+0*stride]=(l0 + l1 + 1)>>1;
1984 src[1+0*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1985 src[2+0*stride]=
1986 src[0+1*stride]=(l1 + l2 + 1)>>1;
1987 src[3+0*stride]=
1988 src[1+1*stride]=(l1 + 2*l2 + l3 + 2)>>2;
1989 src[2+1*stride]=
1990 src[0+2*stride]=(l2 + l3 + 1)>>1;
1991 src[3+1*stride]=
1992 src[1+2*stride]=(l2 + 2*l3 + l3 + 2)>>2;
1993 src[3+2*stride]=
1994 src[1+3*stride]=
1995 src[0+3*stride]=
1996 src[2+2*stride]=
1997 src[2+3*stride]=
1998 src[3+3*stride]=l3;
1999 }
2000
2001 static void pred4x4_horizontal_down_c(uint8_t *src, uint8_t *topright, int stride){
2002 const int lt= src[-1-1*stride];
2003 LOAD_TOP_EDGE
2004 LOAD_LEFT_EDGE
2005 const __attribute__((unused)) int unu= t3;
2006
2007 src[0+0*stride]=
2008 src[2+1*stride]=(lt + l0 + 1)>>1;
2009 src[1+0*stride]=
2010 src[3+1*stride]=(l0 + 2*lt + t0 + 2)>>2;
2011 src[2+0*stride]=(lt + 2*t0 + t1 + 2)>>2;
2012 src[3+0*stride]=(t0 + 2*t1 + t2 + 2)>>2;
2013 src[0+1*stride]=
2014 src[2+2*stride]=(l0 + l1 + 1)>>1;
2015 src[1+1*stride]=
2016 src[3+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
2017 src[0+2*stride]=
2018 src[2+3*stride]=(l1 + l2+ 1)>>1;
2019 src[1+2*stride]=
2020 src[3+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
2021 src[0+3*stride]=(l2 + l3 + 1)>>1;
2022 src[1+3*stride]=(l1 + 2*l2 + l3 + 2)>>2;
2023 }
2024
2025 static void pred16x16_vertical_c(uint8_t *src, int stride){
2026 int i;
2027 const uint32_t a= ((uint32_t*)(src-stride))[0];
2028 const uint32_t b= ((uint32_t*)(src-stride))[1];
2029 const uint32_t c= ((uint32_t*)(src-stride))[2];
2030 const uint32_t d= ((uint32_t*)(src-stride))[3];
2031
2032 for(i=0; i<16; i++){
2033 ((uint32_t*)(src+i*stride))[0]= a;
2034 ((uint32_t*)(src+i*stride))[1]= b;
2035 ((uint32_t*)(src+i*stride))[2]= c;
2036 ((uint32_t*)(src+i*stride))[3]= d;
2037 }
2038 }
2039
2040 static void pred16x16_horizontal_c(uint8_t *src, int stride){
2041 int i;
2042
2043 for(i=0; i<16; i++){
2044 ((uint32_t*)(src+i*stride))[0]=
2045 ((uint32_t*)(src+i*stride))[1]=
2046 ((uint32_t*)(src+i*stride))[2]=
2047 ((uint32_t*)(src+i*stride))[3]= src[-1+i*stride]*0x01010101;
2048 }
2049 }
2050
2051 static void pred16x16_dc_c(uint8_t *src, int stride){
2052 int i, dc=0;
2053
2054 for(i=0;i<16; i++){
2055 dc+= src[-1+i*stride];
2056 }
2057
2058 for(i=0;i<16; i++){
2059 dc+= src[i-stride];
2060 }
2061
2062 dc= 0x01010101*((dc + 16)>>5);
2063
2064 for(i=0; i<16; i++){
2065 ((uint32_t*)(src+i*stride))[0]=
2066 ((uint32_t*)(src+i*stride))[1]=
2067 ((uint32_t*)(src+i*stride))[2]=
2068 ((uint32_t*)(src+i*stride))[3]= dc;
2069 }
2070 }
2071
2072 static void pred16x16_left_dc_c(uint8_t *src, int stride){
2073 int i, dc=0;
2074
2075 for(i=0;i<16; i++){
2076 dc+= src[-1+i*stride];
2077 }
2078
2079 dc= 0x01010101*((dc + 8)>>4);
2080
2081 for(i=0; i<16; i++){
2082 ((uint32_t*)(src+i*stride))[0]=
2083 ((uint32_t*)(src+i*stride))[1]=
2084 ((uint32_t*)(src+i*stride))[2]=
2085 ((uint32_t*)(src+i*stride))[3]= dc;
2086 }
2087 }
2088
2089 static void pred16x16_top_dc_c(uint8_t *src, int stride){
2090 int i, dc=0;
2091
2092 for(i=0;i<16; i++){
2093 dc+= src[i-stride];
2094 }
2095 dc= 0x01010101*((dc + 8)>>4);
2096
2097 for(i=0; i<16; i++){
2098 ((uint32_t*)(src+i*stride))[0]=
2099 ((uint32_t*)(src+i*stride))[1]=
2100 ((uint32_t*)(src+i*stride))[2]=
2101 ((uint32_t*)(src+i*stride))[3]= dc;
2102 }
2103 }
2104
2105 static void pred16x16_128_dc_c(uint8_t *src, int stride){
2106 int i;
2107
2108 for(i=0; i<16; i++){
2109 ((uint32_t*)(src+i*stride))[0]=
2110 ((uint32_t*)(src+i*stride))[1]=
2111 ((uint32_t*)(src+i*stride))[2]=
2112 ((uint32_t*)(src+i*stride))[3]= 0x01010101U*128U;
2113 }
2114 }
2115
2116 static inline void pred16x16_plane_compat_c(uint8_t *src, int stride, const int svq3){
2117 int i, j, k;
2118 int a;
2119 uint8_t *cm = cropTbl + MAX_NEG_CROP;
2120 const uint8_t * const src0 = src+7-stride;
2121 const uint8_t *src1 = src+8*stride-1;
2122 const uint8_t *src2 = src1-2*stride; // == src+6*stride-1;
2123 int H = src0[1] - src0[-1];
2124 int V = src1[0] - src2[ 0];
2125 for(k=2; k<=8; ++k) {
2126 src1 += stride; src2 -= stride;
2127 H += k*(src0[k] - src0[-k]);
2128 V += k*(src1[0] - src2[ 0]);
2129 }
2130 if(svq3){
2131 H = ( 5*(H/4) ) / 16;
2132 V = ( 5*(V/4) ) / 16;
2133
2134 /* required for 100% accuracy */
2135 i = H; H = V; V = i;
2136 }else{
2137 H = ( 5*H+32 ) >> 6;
2138 V = ( 5*V+32 ) >> 6;
2139 }
2140
2141 a = 16*(src1[0] + src2[16] + 1) - 7*(V+H);
2142 for(j=16; j>0; --j) {
2143 int b = a;
2144 a += V;
2145 for(i=-16; i<0; i+=4) {
2146 src[16+i] = cm[ (b ) >> 5 ];
2147 src[17+i] = cm[ (b+ H) >> 5 ];
2148 src[18+i] = cm[ (b+2*H) >> 5 ];
2149 src[19+i] = cm[ (b+3*H) >> 5 ];
2150 b += 4*H;
2151 }
2152 src += stride;
2153 }
2154 }
2155
2156 static void pred16x16_plane_c(uint8_t *src, int stride){
2157 pred16x16_plane_compat_c(src, stride, 0);
2158 }
2159
2160 static void pred8x8_vertical_c(uint8_t *src, int stride){
2161 int i;
2162 const uint32_t a= ((uint32_t*)(src-stride))[0];
2163 const uint32_t b= ((uint32_t*)(src-stride))[1];
2164
2165 for(i=0; i<8; i++){
2166 ((uint32_t*)(src+i*stride))[0]= a;
2167 ((uint32_t*)(src+i*stride))[1]= b;
2168 }
2169 }
2170
2171 static void pred8x8_horizontal_c(uint8_t *src, int stride){
2172 int i;
2173
2174 for(i=0; i<8; i++){
2175 ((uint32_t*)(src+i*stride))[0]=
2176 ((uint32_t*)(src+i*stride))[1]= src[-1+i*stride]*0x01010101;
2177 }
2178 }
2179
2180 static void pred8x8_128_dc_c(uint8_t *src, int stride){
2181 int i;
2182
2183 for(i=0; i<4; i++){
2184 ((uint32_t*)(src+i*stride))[0]=
2185 ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
2186 }
2187 for(i=4; i<8; i++){
2188 ((uint32_t*)(src+i*stride))[0]=
2189 ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
2190 }
2191 }
2192
2193 static void pred8x8_left_dc_c(uint8_t *src, int stride){
2194 int i;
2195 int dc0, dc2;
2196
2197 dc0=dc2=0;
2198 for(i=0;i<4; i++){
2199 dc0+= src[-1+i*stride];
2200 dc2+= src[-1+(i+4)*stride];
2201 }
2202 dc0= 0x01010101*((dc0 + 2)>>2);
2203 dc2= 0x01010101*((dc2 + 2)>>2);
2204
2205 for(i=0; i<4; i++){
2206 ((uint32_t*)(src+i*stride))[0]=
2207 ((uint32_t*)(src+i*stride))[1]= dc0;
2208 }
2209 for(i=4; i<8; i++){
2210 ((uint32_t*)(src+i*stride))[0]=
2211 ((uint32_t*)(src+i*stride))[1]= dc2;
2212 }
2213 }
2214
2215 static void pred8x8_top_dc_c(uint8_t *src, int stride){
2216 int i;
2217 int dc0, dc1;
2218
2219 dc0=dc1=0;
2220 for(i=0;i<4; i++){
2221 dc0+= src[i-stride];
2222 dc1+= src[4+i-stride];
2223 }
2224 dc0= 0x01010101*((dc0 + 2)>>2);
2225 dc1= 0x01010101*((dc1 + 2)>>2);
2226
2227 for(i=0; i<4; i++){
2228 ((uint32_t*)(src+i*stride))[0]= dc0;
2229 ((uint32_t*)(src+i*stride))[1]= dc1;
2230 }
2231 for(i=4; i<8; i++){
2232 ((uint32_t*)(src+i*stride))[0]= dc0;
2233 ((uint32_t*)(src+i*stride))[1]= dc1;
2234 }
2235 }
2236
2237
2238 static void pred8x8_dc_c(uint8_t *src, int stride){
2239 int i;
2240 int dc0, dc1, dc2, dc3;
2241
2242 dc0=dc1=dc2=0;
2243 for(i=0;i<4; i++){
2244 dc0+= src[-1+i*stride] + src[i-stride];
2245 dc1+= src[4+i-stride];
2246 dc2+= src[-1+(i+4)*stride];
2247 }
2248 dc3= 0x01010101*((dc1 + dc2 + 4)>>3);
2249 dc0= 0x01010101*((dc0 + 4)>>3);
2250 dc1= 0x01010101*((dc1 + 2)>>2);
2251 dc2= 0x01010101*((dc2 + 2)>>2);
2252
2253 for(i=0; i<4; i++){
2254 ((uint32_t*)(src+i*stride))[0]= dc0;
2255 ((uint32_t*)(src+i*stride))[1]= dc1;
2256 }
2257 for(i=4; i<8; i++){
2258 ((uint32_t*)(src+i*stride))[0]= dc2;
2259 ((uint32_t*)(src+i*stride))[1]= dc3;
2260 }
2261 }
2262
2263 static void pred8x8_plane_c(uint8_t *src, int stride){
2264 int j, k;
2265 int a;
2266 uint8_t *cm = cropTbl + MAX_NEG_CROP;
2267 const uint8_t * const src0 = src+3-stride;
2268 const uint8_t *src1 = src+4*stride-1;
2269 const uint8_t *src2 = src1-2*stride; // == src+2*stride-1;
2270 int H = src0[1] - src0[-1];
2271 int V = src1[0] - src2[ 0];
2272 for(k=2; k<=4; ++k) {
2273 src1 += stride; src2 -= stride;
2274 H += k*(src0[k] - src0[-k]);
2275 V += k*(src1[0] - src2[ 0]);
2276 }
2277 H = ( 17*H+16 ) >> 5;
2278 V = ( 17*V+16 ) >> 5;
2279
2280 a = 16*(src1[0] + src2[8]+1) - 3*(V+H);
2281 for(j=8; j>0; --j) {
2282 int b = a;
2283 a += V;
2284 src[0] = cm[ (b ) >> 5 ];
2285 src[1] = cm[ (b+ H) >> 5 ];
2286 src[2] = cm[ (b+2*H) >> 5 ];
2287 src[3] = cm[ (b+3*H) >> 5 ];
2288 src[4] = cm[ (b+4*H) >> 5 ];
2289 src[5] = cm[ (b+5*H) >> 5 ];
2290 src[6] = cm[ (b+6*H) >> 5 ];
2291 src[7] = cm[ (b+7*H) >> 5 ];
2292 src += stride;
2293 }
2294 }
2295
2296 static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
2297 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2298 int src_x_offset, int src_y_offset,
2299 qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
2300 MpegEncContext * const s = &h->s;
2301 const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
2302 const int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
2303 const int luma_xy= (mx&3) + ((my&3)<<2);
2304 uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*s->linesize;
2305 uint8_t * src_cb= pic->data[1] + (mx>>3) + (my>>3)*s->uvlinesize;
2306 uint8_t * src_cr= pic->data[2] + (mx>>3) + (my>>3)*s->uvlinesize;
2307 int extra_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16; //FIXME increase edge?, IMHO not worth it
2308 int extra_height= extra_width;
2309 int emu=0;
2310 const int full_mx= mx>>2;
2311 const int full_my= my>>2;
2312
2313 assert(pic->data[0]);
2314
2315 if(mx&7) extra_width -= 3;
2316 if(my&7) extra_height -= 3;
2317
2318 if( full_mx < 0-extra_width
2319 || full_my < 0-extra_height
2320 || full_mx + 16/*FIXME*/ > s->width + extra_width
2321 || full_my + 16/*FIXME*/ > s->height + extra_height){
2322 ff_emulated_edge_mc(s->edge_emu_buffer, src_y - 2 - 2*s->linesize, s->linesize, 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, s->width, s->height);
2323 src_y= s->edge_emu_buffer + 2 + 2*s->linesize;
2324 emu=1;
2325 }
2326
2327 qpix_op[luma_xy](dest_y, src_y, s->linesize); //FIXME try variable height perhaps?
2328 if(!square){
2329 qpix_op[luma_xy](dest_y + delta, src_y + delta, s->linesize);
2330 }
2331
2332 if(s->flags&CODEC_FLAG_GRAY) return;
2333
2334 if(emu){
2335 ff_emulated_edge_mc(s->edge_emu_buffer, src_cb, s->uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), s->width>>1, s->height>>1);
2336 src_cb= s->edge_emu_buffer;
2337 }
2338 chroma_op(dest_cb, src_cb, s->uvlinesize, chroma_height, mx&7, my&7);
2339
2340 if(emu){
2341 ff_emulated_edge_mc(s->edge_emu_buffer, src_cr, s->uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), s->width>>1, s->height>>1);
2342 src_cr= s->edge_emu_buffer;
2343 }
2344 chroma_op(dest_cr, src_cr, s->uvlinesize, chroma_height, mx&7, my&7);
2345 }
2346
2347 static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
2348 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2349 int x_offset, int y_offset,
2350 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2351 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
2352 int list0, int list1){
2353 MpegEncContext * const s = &h->s;
2354 qpel_mc_func *qpix_op= qpix_put;
2355 h264_chroma_mc_func chroma_op= chroma_put;
2356
2357 dest_y += 2*x_offset + 2*y_offset*s-> linesize;
2358 dest_cb += x_offset + y_offset*s->uvlinesize;
2359 dest_cr += x_offset + y_offset*s->uvlinesize;
2360 x_offset += 8*s->mb_x;
2361 y_offset += 8*s->mb_y;
2362
2363 if(list0){
2364 Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
2365 mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
2366 dest_y, dest_cb, dest_cr, x_offset, y_offset,
2367 qpix_op, chroma_op);
2368
2369 qpix_op= qpix_avg;
2370 chroma_op= chroma_avg;
2371 }
2372
2373 if(list1){
2374 Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
2375 mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
2376 dest_y, dest_cb, dest_cr, x_offset, y_offset,
2377 qpix_op, chroma_op);
2378 }
2379 }
2380
2381 static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
2382 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2383 int x_offset, int y_offset,
2384 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2385 h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
2386 h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
2387 int list0, int list1){
2388 MpegEncContext * const s = &h->s;
2389
2390 dest_y += 2*x_offset + 2*y_offset*s-> linesize;
2391 dest_cb += x_offset + y_offset*s->uvlinesize;
2392 dest_cr += x_offset + y_offset*s->uvlinesize;
2393 x_offset += 8*s->mb_x;
2394 y_offset += 8*s->mb_y;
2395
2396 if(list0 && list1){
2397 /* don't optimize for luma-only case, since B-frames usually
2398 * use implicit weights => chroma too. */
2399 uint8_t *tmp_cb = s->obmc_scratchpad;
2400 uint8_t *tmp_cr = tmp_cb + 8*s->uvlinesize;
2401 uint8_t *tmp_y = tmp_cr + 8*s->uvlinesize;
2402 int refn0 = h->ref_cache[0][ scan8[n] ];
2403 int refn1 = h->ref_cache[1][ scan8[n] ];
2404
2405 mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
2406 dest_y, dest_cb, dest_cr,
2407 x_offset, y_offset, qpix_put, chroma_put);
2408 mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
2409 tmp_y, tmp_cb, tmp_cr,
2410 x_offset, y_offset, qpix_put, chroma_put);
2411
2412 if(h->use_weight == 2){
2413 int weight0 = h->implicit_weight[refn0][refn1];
2414 int weight1 = 64 - weight0;
2415 luma_weight_avg( dest_y, tmp_y, s-> linesize, 5, weight0, weight1, 0, 0);
2416 chroma_weight_avg(dest_cb, tmp_cb, s->uvlinesize, 5, weight0, weight1, 0, 0);
2417 chroma_weight_avg(dest_cr, tmp_cr, s->uvlinesize, 5, weight0, weight1, 0, 0);
2418 }else{
2419 luma_weight_avg(dest_y, tmp_y, s->linesize, h->luma_log2_weight_denom,
2420 h->luma_weight[0][refn0], h->luma_weight[1][refn1],
2421 h->luma_offset[0][refn0], h->luma_offset[1][refn1]);
2422 chroma_weight_avg(dest_cb, tmp_cb, s->uvlinesize, h->chroma_log2_weight_denom,
2423 h->chroma_weight[0][refn0][0], h->chroma_weight[1][refn1][0],
2424 h->chroma_offset[0][refn0][0], h->chroma_offset[1][refn1][0]);
2425 chroma_weight_avg(dest_cr, tmp_cr, s->uvlinesize, h->chroma_log2_weight_denom,
2426 h->chroma_weight[0][refn0][1], h->chroma_weight[1][refn1][1],
2427 h->chroma_offset[0][refn0][1], h->chroma_offset[1][refn1][1]);
2428 }
2429 }else{
2430 int list = list1 ? 1 : 0;
2431 int refn = h->ref_cache[list][ scan8[n] ];
2432 Picture *ref= &h->ref_list[list][refn];
2433 mc_dir_part(h, ref, n, square, chroma_height, delta, list,
2434 dest_y, dest_cb, dest_cr, x_offset, y_offset,
2435 qpix_put, chroma_put);
2436
2437 luma_weight_op(dest_y, s->linesize, h->luma_log2_weight_denom,
2438 h->luma_weight[list][refn], h->luma_offset[list][refn]);
2439 if(h->use_weight_chroma){
2440 chroma_weight_op(dest_cb, s->uvlinesize, h->chroma_log2_weight_denom,
2441 h->chroma_weight[list][refn][0], h->chroma_offset[list][refn][0]);
2442 chroma_weight_op(dest_cr, s->uvlinesize, h->chroma_log2_weight_denom,
2443 h->chroma_weight[list][refn][1], h->chroma_offset[list][refn][1]);
2444 }
2445 }
2446 }
2447
2448 static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
2449 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2450 int x_offset, int y_offset,
2451 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2452 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
2453 h264_weight_func *weight_op, h264_biweight_func *weight_avg,
2454 int list0, int list1){
2455 if((h->use_weight==2 && list0 && list1
2456 && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ] != 32))
2457 || h->use_weight==1)
2458 mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
2459 x_offset, y_offset, qpix_put, chroma_put,
2460 weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
2461 else
2462 mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
2463 x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
2464 }
2465
2466 static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2467 qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
2468 qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
2469 h264_weight_func *weight_op, h264_biweight_func *weight_avg){
2470 MpegEncContext * const s = &h->s;
2471 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
2472 const int mb_type= s->current_picture.mb_type[mb_xy];
2473
2474 assert(IS_INTER(mb_type));
2475
2476 if(IS_16X16(mb_type)){
2477 mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
2478 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
2479 &weight_op[0], &weight_avg[0],
2480 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2481 }else if(IS_16X8(mb_type)){
2482 mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
2483 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
2484 &weight_op[1], &weight_avg[1],
2485 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2486 mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
2487 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
2488 &weight_op[1], &weight_avg[1],
2489 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2490 }else if(IS_8X16(mb_type)){
2491 mc_part(h, 0, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 0, 0,
2492 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2493 &weight_op[2], &weight_avg[2],
2494 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2495 mc_part(h, 4, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 4, 0,
2496 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2497 &weight_op[2], &weight_avg[2],
2498 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2499 }else{
2500 int i;
2501
2502 assert(IS_8X8(mb_type));
2503
2504 for(i=0; i<4; i++){
2505 const int sub_mb_type= h->sub_mb_type[i];
2506 const int n= 4*i;
2507 int x_offset= (i&1)<<2;
2508 int y_offset= (i&2)<<1;
2509
2510 if(IS_SUB_8X8(sub_mb_type)){
2511 mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2512 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2513 &weight_op[3], &weight_avg[3],
2514 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2515 }else if(IS_SUB_8X4(sub_mb_type)){
2516 mc_part(h, n , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2517 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2518 &weight_op[4], &weight_avg[4],
2519 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2520 mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
2521 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2522 &weight_op[4], &weight_avg[4],
2523 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2524 }else if(IS_SUB_4X8(sub_mb_type)){
2525 mc_part(h, n , 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2526 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2527 &weight_op[5], &weight_avg[5],
2528 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2529 mc_part(h, n+1, 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
2530 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2531 &weight_op[5], &weight_avg[5],
2532 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2533 }else{
2534 int j;
2535 assert(IS_SUB_4X4(sub_mb_type));
2536 for(j=0; j<4; j++){
2537 int sub_x_offset= x_offset + 2*(j&1);
2538 int sub_y_offset= y_offset + (j&2);
2539 mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
2540 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2541 &weight_op[6], &weight_avg[6],
2542 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2543 }
2544 }
2545 }
2546 }
2547 }
2548
2549 static void decode_init_vlc(H264Context *h){
2550 static int done = 0;
2551
2552 if (!done) {
2553 int i;
2554 done = 1;
2555
2556 init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5,
2557 &chroma_dc_coeff_token_len [0], 1, 1,
2558 &chroma_dc_coeff_token_bits[0], 1, 1, 1);
2559
2560 for(i=0; i<4; i++){
2561 init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17,
2562 &coeff_token_len [i][0], 1, 1,
2563 &coeff_token_bits[i][0], 1, 1, 1);
2564 }
2565
2566 for(i=0; i<3; i++){
2567 init_vlc(&chroma_dc_total_zeros_vlc[i], CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
2568 &chroma_dc_total_zeros_len [i][0], 1, 1,
2569 &chroma_dc_total_zeros_bits[i][0], 1, 1, 1);
2570 }
2571 for(i=0; i<15; i++){
2572 init_vlc(&total_zeros_vlc[i], TOTAL_ZEROS_VLC_BITS, 16,
2573 &total_zeros_len [i][0], 1, 1,
2574 &total_zeros_bits[i][0], 1, 1, 1);
2575 }
2576
2577 for(i=0; i<6; i++){
2578 init_vlc(&run_vlc[i], RUN_VLC_BITS, 7,
2579 &run_len [i][0], 1, 1,
2580 &run_bits[i][0], 1, 1, 1);
2581 }
2582 init_vlc(&run7_vlc, RUN7_VLC_BITS, 16,
2583 &run_len [6][0], 1, 1,
2584 &run_bits[6][0], 1, 1, 1);
2585 }
2586 }
2587
2588 /**
2589 * Sets the intra prediction function pointers.
2590 */
2591 static void init_pred_ptrs(H264Context *h){
2592 // MpegEncContext * const s = &h->s;
2593
2594 h->pred4x4[VERT_PRED ]= pred4x4_vertical_c;
2595 h->pred4x4[HOR_PRED ]= pred4x4_horizontal_c;
2596 h->pred4x4[DC_PRED ]= pred4x4_dc_c;
2597 h->pred4x4[DIAG_DOWN_LEFT_PRED ]= pred4x4_down_left_c;
2598 h->pred4x4[DIAG_DOWN_RIGHT_PRED]= pred4x4_down_right_c;
2599 h->pred4x4[VERT_RIGHT_PRED ]= pred4x4_vertical_right_c;
2600 h->pred4x4[HOR_DOWN_PRED ]= pred4x4_horizontal_down_c;
2601 h->pred4x4[VERT_LEFT_PRED ]= pred4x4_vertical_left_c;
2602 h->pred4x4[HOR_UP_PRED ]= pred4x4_horizontal_up_c;
2603 h->pred4x4[LEFT_DC_PRED ]= pred4x4_left_dc_c;
2604 h->pred4x4[TOP_DC_PRED ]= pred4x4_top_dc_c;
2605 h->pred4x4[DC_128_PRED ]= pred4x4_128_dc_c