decode custom quant matrices.
[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 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;
85 AVRational sar;
86 int timing_info_present_flag;
87 uint32_t num_units_in_tick;
88 uint32_t time_scale;
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;
93 int scaling_matrix_present;
94 uint8_t scaling_matrix4[6][16];
95 uint8_t scaling_matrix8[2][64];
96 }SPS;
97
98 /**
99 * Picture parameter set
100 */
101 typedef struct PPS{
102 int sps_id;
103 int cabac; ///< entropy_coding_mode_flag
104 int pic_order_present; ///< pic_order_present_flag
105 int slice_group_count; ///< num_slice_groups_minus1 + 1
106 int mb_slice_group_map_type;
107 int ref_count[2]; ///< num_ref_idx_l0/1_active_minus1 + 1
108 int weighted_pred; ///< weighted_pred_flag
109 int weighted_bipred_idc;
110 int init_qp; ///< pic_init_qp_minus26 + 26
111 int init_qs; ///< pic_init_qs_minus26 + 26
112 int chroma_qp_index_offset;
113 int deblocking_filter_parameters_present; ///< deblocking_filter_parameters_present_flag
114 int constrained_intra_pred; ///< constrained_intra_pred_flag
115 int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag
116 int transform_8x8_mode; ///< transform_8x8_mode_flag
117 uint8_t scaling_matrix4[6][16];
118 uint8_t scaling_matrix8[2][64];
119 }PPS;
120
121 /**
122 * Memory management control operation opcode.
123 */
124 typedef enum MMCOOpcode{
125 MMCO_END=0,
126 MMCO_SHORT2UNUSED,
127 MMCO_LONG2UNUSED,
128 MMCO_SHORT2LONG,
129 MMCO_SET_MAX_LONG,
130 MMCO_RESET,
131 MMCO_LONG,
132 } MMCOOpcode;
133
134 /**
135 * Memory management control operation.
136 */
137 typedef struct MMCO{
138 MMCOOpcode opcode;
139 int short_frame_num;
140 int long_index;
141 } MMCO;
142
143 /**
144 * H264Context
145 */
146 typedef struct H264Context{
147 MpegEncContext s;
148 int nal_ref_idc;
149 int nal_unit_type;
150 #define NAL_SLICE 1
151 #define NAL_DPA 2
152 #define NAL_DPB 3
153 #define NAL_DPC 4
154 #define NAL_IDR_SLICE 5
155 #define NAL_SEI 6
156 #define NAL_SPS 7
157 #define NAL_PPS 8
158 #define NAL_PICTURE_DELIMITER 9
159 #define NAL_FILTER_DATA 10
160 uint8_t *rbsp_buffer;
161 int rbsp_buffer_size;
162
163 /**
164 * Used to parse AVC variant of h264
165 */
166 int is_avc; ///< this flag is != 0 if codec is avc1
167 int got_avcC; ///< flag used to parse avcC data only once
168 int nal_length_size; ///< Number of bytes used for nal length (1, 2 or 4)
169
170 int chroma_qp; //QPc
171
172 int prev_mb_skipped; //FIXME remove (IMHO not used)
173
174 //prediction stuff
175 int chroma_pred_mode;
176 int intra16x16_pred_mode;
177
178 int top_mb_xy;
179 int left_mb_xy[2];
180
181 int8_t intra4x4_pred_mode_cache[5*8];
182 int8_t (*intra4x4_pred_mode)[8];
183 void (*pred4x4 [9+3])(uint8_t *src, uint8_t *topright, int stride);//FIXME move to dsp?
184 void (*pred8x8l [9+3])(uint8_t *src, int topleft, int topright, int stride);
185 void (*pred8x8 [4+3])(uint8_t *src, int stride);
186 void (*pred16x16[4+3])(uint8_t *src, int stride);
187 unsigned int topleft_samples_available;
188 unsigned int top_samples_available;
189 unsigned int topright_samples_available;
190 unsigned int left_samples_available;
191 uint8_t (*top_borders[2])[16+2*8];
192 uint8_t left_border[2*(17+2*9)];
193
194 /**
195 * non zero coeff count cache.
196 * is 64 if not available.
197 */
198 uint8_t non_zero_count_cache[6*8] __align8;
199 uint8_t (*non_zero_count)[16];
200
201 /**
202 * Motion vector cache.
203 */
204 int16_t mv_cache[2][5*8][2] __align8;
205 int8_t ref_cache[2][5*8] __align8;
206 #define LIST_NOT_USED -1 //FIXME rename?
207 #define PART_NOT_AVAILABLE -2
208
209 /**
210 * is 1 if the specific list MV&references are set to 0,0,-2.
211 */
212 int mv_cache_clean[2];
213
214 /**
215 * number of neighbors (top and/or left) that used 8x8 dct
216 */
217 int neighbor_transform_size;
218
219 /**
220 * block_offset[ 0..23] for frame macroblocks
221 * block_offset[24..47] for field macroblocks
222 */
223 int block_offset[2*(16+8)];
224
225 uint32_t *mb2b_xy; //FIXME are these 4 a good idea?
226 uint32_t *mb2b8_xy;
227 int b_stride; //FIXME use s->b4_stride
228 int b8_stride;
229
230 int halfpel_flag;
231 int thirdpel_flag;
232
233 int unknown_svq3_flag;
234 int next_slice_index;
235
236 SPS sps_buffer[MAX_SPS_COUNT];
237 SPS sps; ///< current sps
238
239 PPS pps_buffer[MAX_PPS_COUNT];
240 /**
241 * current pps
242 */
243 PPS pps; //FIXME move to Picture perhaps? (->no) do we need that?
244
245 uint32_t dequant4_buffer[6][52][16];
246 uint32_t dequant8_buffer[2][52][64];
247 uint32_t (*dequant4_coeff[6])[16];
248 uint32_t (*dequant8_coeff[2])[64];
249 int dequant_coeff_pps; ///< reinit tables when pps changes
250
251 int slice_num;
252 uint8_t *slice_table_base;
253 uint8_t *slice_table; ///< slice_table_base + mb_stride + 1
254 int slice_type;
255 int slice_type_fixed;
256
257 //interlacing specific flags
258 int mb_aff_frame;
259 int mb_field_decoding_flag;
260
261 int sub_mb_type[4];
262
263 //POC stuff
264 int poc_lsb;
265 int poc_msb;
266 int delta_poc_bottom;
267 int delta_poc[2];
268 int frame_num;
269 int prev_poc_msb; ///< poc_msb of the last reference pic for POC type 0
270 int prev_poc_lsb; ///< poc_lsb of the last reference pic for POC type 0
271 int frame_num_offset; ///< for POC type 2
272 int prev_frame_num_offset; ///< for POC type 2
273 int prev_frame_num; ///< frame_num of the last pic for POC type 1/2
274
275 /**
276 * frame_num for frames or 2*frame_num for field pics.
277 */
278 int curr_pic_num;
279
280 /**
281 * max_frame_num or 2*max_frame_num for field pics.
282 */
283 int max_pic_num;
284
285 //Weighted pred stuff
286 int use_weight;
287 int use_weight_chroma;
288 int luma_log2_weight_denom;
289 int chroma_log2_weight_denom;
290 int luma_weight[2][16];
291 int luma_offset[2][16];
292 int chroma_weight[2][16][2];
293 int chroma_offset[2][16][2];
294 int implicit_weight[16][16];
295
296 //deblock
297 int deblocking_filter; ///< disable_deblocking_filter_idc with 1<->0
298 int slice_alpha_c0_offset;
299 int slice_beta_offset;
300
301 int redundant_pic_count;
302
303 int direct_spatial_mv_pred;
304 int dist_scale_factor[16];
305 int map_col_to_list0[2][16];
306
307 /**
308 * num_ref_idx_l0/1_active_minus1 + 1
309 */
310 int ref_count[2];// FIXME split for AFF
311 Picture *short_ref[32];
312 Picture *long_ref[32];
313 Picture default_ref_list[2][32];
314 Picture ref_list[2][32]; //FIXME size?
315 Picture field_ref_list[2][32]; //FIXME size?
316 Picture *delayed_pic[16]; //FIXME size?
317 Picture *delayed_output_pic;
318
319 /**
320 * memory management control operations buffer.
321 */
322 MMCO mmco[MAX_MMCO_COUNT];
323 int mmco_index;
324
325 int long_ref_count; ///< number of actual long term references
326 int short_ref_count; ///< number of actual short term references
327
328 //data partitioning
329 GetBitContext intra_gb;
330 GetBitContext inter_gb;
331 GetBitContext *intra_gb_ptr;
332 GetBitContext *inter_gb_ptr;
333
334 DCTELEM mb[16*24] __align8;
335
336 /**
337 * Cabac
338 */
339 CABACContext cabac;
340 uint8_t cabac_state[460];
341 int cabac_init_idc;
342
343 /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0,1,2), 0x0? luma_cbp */
344 uint16_t *cbp_table;
345 int top_cbp;
346 int left_cbp;
347 /* chroma_pred_mode for i4x4 or i16x16, else 0 */
348 uint8_t *chroma_pred_mode_table;
349 int last_qscale_diff;
350 int16_t (*mvd_table[2])[2];
351 int16_t mvd_cache[2][5*8][2] __align8;
352 uint8_t *direct_table;
353 uint8_t direct_cache[5*8];
354
355 uint8_t zigzag_scan[16];
356 uint8_t field_scan[16];
357 const uint8_t *zigzag_scan_q0;
358 const uint8_t *field_scan_q0;
359
360 int x264_build;
361 }H264Context;
362
363 static VLC coeff_token_vlc[4];
364 static VLC chroma_dc_coeff_token_vlc;
365
366 static VLC total_zeros_vlc[15];
367 static VLC chroma_dc_total_zeros_vlc[3];
368
369 static VLC run_vlc[6];
370 static VLC run7_vlc;
371
372 static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
373 static void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
374 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);
375
376 static inline uint32_t pack16to32(int a, int b){
377 #ifdef WORDS_BIGENDIAN
378 return (b&0xFFFF) + (a<<16);
379 #else
380 return (a&0xFFFF) + (b<<16);
381 #endif
382 }
383
384 /**
385 * fill a rectangle.
386 * @param h height of the rectangle, should be a constant
387 * @param w width of the rectangle, should be a constant
388 * @param size the size of val (1 or 4), should be a constant
389 */
390 static inline void fill_rectangle(void *vp, int w, int h, int stride, uint32_t val, int size){ //FIXME ensure this IS inlined
391 uint8_t *p= (uint8_t*)vp;
392 assert(size==1 || size==4);
393
394 w *= size;
395 stride *= size;
396
397 assert((((int)vp)&(FFMIN(w, STRIDE_ALIGN)-1)) == 0);
398 assert((stride&(w-1))==0);
399 //FIXME check what gcc generates for 64 bit on x86 and possibly write a 32 bit ver of it
400 if(w==2 && h==2){
401 *(uint16_t*)(p + 0)=
402 *(uint16_t*)(p + stride)= size==4 ? val : val*0x0101;
403 }else if(w==2 && h==4){
404 *(uint16_t*)(p + 0*stride)=
405 *(uint16_t*)(p + 1*stride)=
406 *(uint16_t*)(p + 2*stride)=
407 *(uint16_t*)(p + 3*stride)= size==4 ? val : val*0x0101;
408 }else if(w==4 && h==1){
409 *(uint32_t*)(p + 0*stride)= size==4 ? val : val*0x01010101;
410 }else if(w==4 && h==2){
411 *(uint32_t*)(p + 0*stride)=
412 *(uint32_t*)(p + 1*stride)= size==4 ? val : val*0x01010101;
413 }else if(w==4 && h==4){
414 *(uint32_t*)(p + 0*stride)=
415 *(uint32_t*)(p + 1*stride)=
416 *(uint32_t*)(p + 2*stride)=
417 *(uint32_t*)(p + 3*stride)= size==4 ? val : val*0x01010101;
418 }else if(w==8 && h==1){
419 *(uint32_t*)(p + 0)=
420 *(uint32_t*)(p + 4)= size==4 ? val : val*0x01010101;
421 }else if(w==8 && h==2){
422 *(uint32_t*)(p + 0 + 0*stride)=
423 *(uint32_t*)(p + 4 + 0*stride)=
424 *(uint32_t*)(p + 0 + 1*stride)=
425 *(uint32_t*)(p + 4 + 1*stride)= size==4 ? val : val*0x01010101;
426 }else if(w==8 && h==4){
427 *(uint64_t*)(p + 0*stride)=
428 *(uint64_t*)(p + 1*stride)=
429 *(uint64_t*)(p + 2*stride)=
430 *(uint64_t*)(p + 3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
431 }else if(w==16 && h==2){
432 *(uint64_t*)(p + 0+0*stride)=
433 *(uint64_t*)(p + 8+0*stride)=
434 *(uint64_t*)(p + 0+1*stride)=
435 *(uint64_t*)(p + 8+1*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
436 }else if(w==16 && h==4){
437 *(uint64_t*)(p + 0+0*stride)=
438 *(uint64_t*)(p + 8+0*stride)=
439 *(uint64_t*)(p + 0+1*stride)=
440 *(uint64_t*)(p + 8+1*stride)=
441 *(uint64_t*)(p + 0+2*stride)=
442 *(uint64_t*)(p + 8+2*stride)=
443 *(uint64_t*)(p + 0+3*stride)=
444 *(uint64_t*)(p + 8+3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
445 }else
446 assert(0);
447 }
448
449 static inline void fill_caches(H264Context *h, int mb_type, int for_deblock){
450 MpegEncContext * const s = &h->s;
451 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
452 int topleft_xy, top_xy, topright_xy, left_xy[2];
453 int topleft_type, top_type, topright_type, left_type[2];
454 int left_block[8];
455 int i;
456
457 //FIXME deblocking can skip fill_caches much of the time with multiple slices too.
458 // the actual condition is whether we're on the edge of a slice,
459 // and even then the intra and nnz parts are unnecessary.
460 if(for_deblock && h->slice_num == 1)
461 return;
462
463 //wow what a mess, why didn't they simplify the interlacing&intra stuff, i can't imagine that these complex rules are worth it
464
465 top_xy = mb_xy - s->mb_stride;
466 topleft_xy = top_xy - 1;
467 topright_xy= top_xy + 1;
468 left_xy[1] = left_xy[0] = mb_xy-1;
469 left_block[0]= 0;
470 left_block[1]= 1;
471 left_block[2]= 2;
472 left_block[3]= 3;
473 left_block[4]= 7;
474 left_block[5]= 10;
475 left_block[6]= 8;
476 left_block[7]= 11;
477 if(h->mb_aff_frame){
478 const int pair_xy = s->mb_x + (s->mb_y & ~1)*s->mb_stride;
479 const int top_pair_xy = pair_xy - s->mb_stride;
480 const int topleft_pair_xy = top_pair_xy - 1;
481 const int topright_pair_xy = top_pair_xy + 1;
482 const int topleft_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[topleft_pair_xy]);
483 const int top_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[top_pair_xy]);
484 const int topright_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[topright_pair_xy]);
485 const int left_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[pair_xy-1]);
486 const int curr_mb_frame_flag = !IS_INTERLACED(mb_type);
487 const int bottom = (s->mb_y & 1);
488 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);
489 if (bottom
490 ? !curr_mb_frame_flag // bottom macroblock
491 : (!curr_mb_frame_flag && !top_mb_frame_flag) // top macroblock
492 ) {
493 top_xy -= s->mb_stride;
494 }
495 if (bottom
496 ? !curr_mb_frame_flag // bottom macroblock
497 : (!curr_mb_frame_flag && !topleft_mb_frame_flag) // top macroblock
498 ) {
499 topleft_xy -= s->mb_stride;
500 }
501 if (bottom
502 ? !curr_mb_frame_flag // bottom macroblock
503 : (!curr_mb_frame_flag && !topright_mb_frame_flag) // top macroblock
504 ) {
505 topright_xy -= s->mb_stride;
506 }
507 if (left_mb_frame_flag != curr_mb_frame_flag) {
508 left_xy[1] = left_xy[0] = pair_xy - 1;
509 if (curr_mb_frame_flag) {
510 if (bottom) {
511 left_block[0]= 2;
512 left_block[1]= 2;
513 left_block[2]= 3;
514 left_block[3]= 3;
515 left_block[4]= 8;
516 left_block[5]= 11;
517 left_block[6]= 8;
518 left_block[7]= 11;
519 } else {
520 left_block[0]= 0;
521 left_block[1]= 0;
522 left_block[2]= 1;
523 left_block[3]= 1;
524 left_block[4]= 7;
525 left_block[5]= 10;
526 left_block[6]= 7;
527 left_block[7]= 10;
528 }
529 } else {
530 left_xy[1] += s->mb_stride;
531 //left_block[0]= 0;
532 left_block[1]= 2;
533 left_block[2]= 0;
534 left_block[3]= 2;
535 //left_block[4]= 7;
536 left_block[5]= 10;
537 left_block[6]= 7;
538 left_block[7]= 10;
539 }
540 }
541 }
542
543 h->top_mb_xy = top_xy;
544 h->left_mb_xy[0] = left_xy[0];
545 h->left_mb_xy[1] = left_xy[1];
546 if(for_deblock){
547 topleft_type = h->slice_table[topleft_xy ] < 255 ? s->current_picture.mb_type[topleft_xy] : 0;
548 top_type = h->slice_table[top_xy ] < 255 ? s->current_picture.mb_type[top_xy] : 0;
549 topright_type= h->slice_table[topright_xy] < 255 ? s->current_picture.mb_type[topright_xy]: 0;
550 left_type[0] = h->slice_table[left_xy[0] ] < 255 ? s->current_picture.mb_type[left_xy[0]] : 0;
551 left_type[1] = h->slice_table[left_xy[1] ] < 255 ? s->current_picture.mb_type[left_xy[1]] : 0;
552 }else{
553 topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
554 top_type = h->slice_table[top_xy ] == h->slice_num ? s->current_picture.mb_type[top_xy] : 0;
555 topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
556 left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
557 left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
558 }
559
560 if(IS_INTRA(mb_type)){
561 h->topleft_samples_available=
562 h->top_samples_available=
563 h->left_samples_available= 0xFFFF;
564 h->topright_samples_available= 0xEEEA;
565
566 if(!IS_INTRA(top_type) && (top_type==0 || h->pps.constrained_intra_pred)){
567 h->topleft_samples_available= 0xB3FF;
568 h->top_samples_available= 0x33FF;
569 h->topright_samples_available= 0x26EA;
570 }
571 for(i=0; i<2; i++){
572 if(!IS_INTRA(left_type[i]) && (left_type[i]==0 || h->pps.constrained_intra_pred)){
573 h->topleft_samples_available&= 0xDF5F;
574 h->left_samples_available&= 0x5F5F;
575 }
576 }
577
578 if(!IS_INTRA(topleft_type) && (topleft_type==0 || h->pps.constrained_intra_pred))
579 h->topleft_samples_available&= 0x7FFF;
580
581 if(!IS_INTRA(topright_type) && (topright_type==0 || h->pps.constrained_intra_pred))
582 h->topright_samples_available&= 0xFBFF;
583
584 if(IS_INTRA4x4(mb_type)){
585 if(IS_INTRA4x4(top_type)){
586 h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
587 h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
588 h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
589 h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
590 }else{
591 int pred;
592 if(!top_type || (IS_INTER(top_type) && h->pps.constrained_intra_pred))
593 pred= -1;
594 else{
595 pred= 2;
596 }
597 h->intra4x4_pred_mode_cache[4+8*0]=
598 h->intra4x4_pred_mode_cache[5+8*0]=
599 h->intra4x4_pred_mode_cache[6+8*0]=
600 h->intra4x4_pred_mode_cache[7+8*0]= pred;
601 }
602 for(i=0; i<2; i++){
603 if(IS_INTRA4x4(left_type[i])){
604 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
605 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
606 }else{
607 int pred;
608 if(!left_type[i] || (IS_INTER(left_type[i]) && h->pps.constrained_intra_pred))
609 pred= -1;
610 else{
611 pred= 2;
612 }
613 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
614 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
615 }
616 }
617 }
618 }
619
620
621 /*
622 0 . T T. T T T T
623 1 L . .L . . . .
624 2 L . .L . . . .
625 3 . T TL . . . .
626 4 L . .L . . . .
627 5 L . .. . . . .
628 */
629 //FIXME constraint_intra_pred & partitioning & nnz (lets hope this is just a typo in the spec)
630 if(top_type){
631 h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][4];
632 h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][5];
633 h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][6];
634 h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3];
635
636 h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][9];
637 h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8];
638
639 h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][12];
640 h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11];
641
642 }else{
643 h->non_zero_count_cache[4+8*0]=
644 h->non_zero_count_cache[5+8*0]=
645 h->non_zero_count_cache[6+8*0]=
646 h->non_zero_count_cache[7+8*0]=
647
648 h->non_zero_count_cache[1+8*0]=
649 h->non_zero_count_cache[2+8*0]=
650
651 h->non_zero_count_cache[1+8*3]=
652 h->non_zero_count_cache[2+8*3]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
653
654 }
655
656 for (i=0; i<2; i++) {
657 if(left_type[i]){
658 h->non_zero_count_cache[3+8*1 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[0+2*i]];
659 h->non_zero_count_cache[3+8*2 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[1+2*i]];
660 h->non_zero_count_cache[0+8*1 + 8*i]= h->non_zero_count[left_xy[i]][left_block[4+2*i]];
661 h->non_zero_count_cache[0+8*4 + 8*i]= h->non_zero_count[left_xy[i]][left_block[5+2*i]];
662 }else{
663 h->non_zero_count_cache[3+8*1 + 2*8*i]=
664 h->non_zero_count_cache[3+8*2 + 2*8*i]=
665 h->non_zero_count_cache[0+8*1 + 8*i]=
666 h->non_zero_count_cache[0+8*4 + 8*i]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
667 }
668 }
669
670 if( h->pps.cabac ) {
671 // top_cbp
672 if(top_type) {
673 h->top_cbp = h->cbp_table[top_xy];
674 } else if(IS_INTRA(mb_type)) {
675 h->top_cbp = 0x1C0;
676 } else {
677 h->top_cbp = 0;
678 }
679 // left_cbp
680 if (left_type[0]) {
681 h->left_cbp = h->cbp_table[left_xy[0]] & 0x1f0;
682 } else if(IS_INTRA(mb_type)) {
683 h->left_cbp = 0x1C0;
684 } else {
685 h->left_cbp = 0;
686 }
687 if (left_type[0]) {
688 h->left_cbp |= ((h->cbp_table[left_xy[0]]>>((left_block[0]&(~1))+1))&0x1) << 1;
689 }
690 if (left_type[1]) {
691 h->left_cbp |= ((h->cbp_table[left_xy[1]]>>((left_block[2]&(~1))+1))&0x1) << 3;
692 }
693 }
694
695 #if 1
696 //FIXME direct mb can skip much of this
697 if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
698 int list;
699 for(list=0; list<1+(h->slice_type==B_TYPE); list++){
700 if(!USES_LIST(mb_type, list) && !IS_DIRECT(mb_type) && !h->deblocking_filter){
701 /*if(!h->mv_cache_clean[list]){
702 memset(h->mv_cache [list], 0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
703 memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
704 h->mv_cache_clean[list]= 1;
705 }*/
706 continue;
707 }
708 h->mv_cache_clean[list]= 0;
709
710 if(IS_INTER(top_type)){
711 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
712 const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
713 *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0];
714 *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];
715 *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];
716 *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];
717 h->ref_cache[list][scan8[0] + 0 - 1*8]=
718 h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
719 h->ref_cache[list][scan8[0] + 2 - 1*8]=
720 h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
721 }else{
722 *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]=
723 *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]=
724 *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]=
725 *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;
726 *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
727 }
728
729 //FIXME unify cleanup or sth
730 if(IS_INTER(left_type[0])){
731 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
732 const int b8_xy= h->mb2b8_xy[left_xy[0]] + 1;
733 *(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]];
734 *(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]];
735 h->ref_cache[list][scan8[0] - 1 + 0*8]=
736 h->ref_cache[list][scan8[0] - 1 + 1*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0]>>1)];
737 }else{
738 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 0*8]=
739 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 1*8]= 0;
740 h->ref_cache[list][scan8[0] - 1 + 0*8]=
741 h->ref_cache[list][scan8[0] - 1 + 1*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
742 }
743
744 if(IS_INTER(left_type[1])){
745 const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
746 const int b8_xy= h->mb2b8_xy[left_xy[1]] + 1;
747 *(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]];
748 *(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]];
749 h->ref_cache[list][scan8[0] - 1 + 2*8]=
750 h->ref_cache[list][scan8[0] - 1 + 3*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[2]>>1)];
751 }else{
752 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 2*8]=
753 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 3*8]= 0;
754 h->ref_cache[list][scan8[0] - 1 + 2*8]=
755 h->ref_cache[list][scan8[0] - 1 + 3*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
756 assert((!left_type[0]) == (!left_type[1]));
757 }
758
759 if(for_deblock || (IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred))
760 continue;
761
762 if(IS_INTER(topleft_type)){
763 const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
764 const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + h->b8_stride;
765 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
766 h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
767 }else{
768 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
769 h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
770 }
771
772 if(IS_INTER(topright_type)){
773 const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
774 const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
775 *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
776 h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
777 }else{
778 *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
779 h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
780 }
781
782
783 h->ref_cache[list][scan8[5 ]+1] =
784 h->ref_cache[list][scan8[7 ]+1] =
785 h->ref_cache[list][scan8[13]+1] = //FIXME remove past 3 (init somewhere else)
786 h->ref_cache[list][scan8[4 ]] =
787 h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
788 *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
789 *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
790 *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
791 *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
792 *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
793
794 if( h->pps.cabac ) {
795 /* XXX beurk, Load mvd */
796 if(IS_INTER(topleft_type)){
797 const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
798 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy];
799 }else{
800 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 - 1*8]= 0;
801 }
802
803 if(IS_INTER(top_type)){
804 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
805 *(uint32_t*)h->mvd_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 0];
806 *(uint32_t*)h->mvd_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 1];
807 *(uint32_t*)h->mvd_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 2];
808 *(uint32_t*)h->mvd_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 3];
809 }else{
810 *(uint32_t*)h->mvd_cache [list][scan8[0] + 0 - 1*8]=
811 *(uint32_t*)h->mvd_cache [list][scan8[0] + 1 - 1*8]=
812 *(uint32_t*)h->mvd_cache [list][scan8[0] + 2 - 1*8]=
813 *(uint32_t*)h->mvd_cache [list][scan8[0] + 3 - 1*8]= 0;
814 }
815 if(IS_INTER(left_type[0])){
816 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
817 *(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]];
818 *(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]];
819 }else{
820 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 0*8]=
821 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 1*8]= 0;
822 }
823 if(IS_INTER(left_type[1])){
824 const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
825 *(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]];
826 *(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]];
827 }else{
828 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 2*8]=
829 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 3*8]= 0;
830 }
831 *(uint32_t*)h->mvd_cache [list][scan8[5 ]+1]=
832 *(uint32_t*)h->mvd_cache [list][scan8[7 ]+1]=
833 *(uint32_t*)h->mvd_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
834 *(uint32_t*)h->mvd_cache [list][scan8[4 ]]=
835 *(uint32_t*)h->mvd_cache [list][scan8[12]]= 0;
836
837 if(h->slice_type == B_TYPE){
838 fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, 0, 1);
839
840 if(IS_DIRECT(top_type)){
841 *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0x01010101;
842 }else if(IS_8X8(top_type)){
843 int b8_xy = h->mb2b8_xy[top_xy] + h->b8_stride;
844 h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy];
845 h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 1];
846 }else{
847 *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0;
848 }
849
850 //FIXME interlacing
851 if(IS_DIRECT(left_type[0])){
852 h->direct_cache[scan8[0] - 1 + 0*8]=
853 h->direct_cache[scan8[0] - 1 + 2*8]= 1;
854 }else if(IS_8X8(left_type[0])){
855 int b8_xy = h->mb2b8_xy[left_xy[0]] + 1;
856 h->direct_cache[scan8[0] - 1 + 0*8]= h->direct_table[b8_xy];
857 h->direct_cache[scan8[0] - 1 + 2*8]= h->direct_table[b8_xy + h->b8_stride];
858 }else{
859 h->direct_cache[scan8[0] - 1 + 0*8]=
860 h->direct_cache[scan8[0] - 1 + 2*8]= 0;
861 }
862 }
863 }
864 }
865 }
866 #endif
867
868 h->neighbor_transform_size= !!IS_8x8DCT(top_type) + !!IS_8x8DCT(left_type[0]);
869 }
870
871 static inline void write_back_intra_pred_mode(H264Context *h){
872 MpegEncContext * const s = &h->s;
873 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
874
875 h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
876 h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
877 h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
878 h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
879 h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
880 h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
881 h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
882 }
883
884 /**
885 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
886 */
887 static inline int check_intra4x4_pred_mode(H264Context *h){
888 MpegEncContext * const s = &h->s;
889 static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
890 static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
891 int i;
892
893 if(!(h->top_samples_available&0x8000)){
894 for(i=0; i<4; i++){
895 int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
896 if(status<0){
897 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);
898 return -1;
899 } else if(status){
900 h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
901 }
902 }
903 }
904
905 if(!(h->left_samples_available&0x8000)){
906 for(i=0; i<4; i++){
907 int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
908 if(status<0){
909 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);
910 return -1;
911 } else if(status){
912 h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
913 }
914 }
915 }
916
917 return 0;
918 } //FIXME cleanup like next
919
920 /**
921 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
922 */
923 static inline int check_intra_pred_mode(H264Context *h, int mode){
924 MpegEncContext * const s = &h->s;
925 static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
926 static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
927
928 if(mode < 0 || mode > 6) {
929 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);
930 return -1;
931 }
932
933 if(!(h->top_samples_available&0x8000)){
934 mode= top[ mode ];
935 if(mode<0){
936 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);
937 return -1;
938 }
939 }
940
941 if(!(h->left_samples_available&0x8000)){
942 mode= left[ mode ];
943 if(mode<0){
944 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);
945 return -1;
946 }
947 }
948
949 return mode;
950 }
951
952 /**
953 * gets the predicted intra4x4 prediction mode.
954 */
955 static inline int pred_intra_mode(H264Context *h, int n){
956 const int index8= scan8[n];
957 const int left= h->intra4x4_pred_mode_cache[index8 - 1];
958 const int top = h->intra4x4_pred_mode_cache[index8 - 8];
959 const int min= FFMIN(left, top);
960
961 tprintf("mode:%d %d min:%d\n", left ,top, min);
962
963 if(min<0) return DC_PRED;
964 else return min;
965 }
966
967 static inline void write_back_non_zero_count(H264Context *h){
968 MpegEncContext * const s = &h->s;
969 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
970
971 h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[7+8*1];
972 h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[7+8*2];
973 h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[7+8*3];
974 h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];
975 h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[4+8*4];
976 h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[5+8*4];
977 h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[6+8*4];
978
979 h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[1+8*2];
980 h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
981 h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[2+8*1];
982
983 h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[1+8*5];
984 h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
985 h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[2+8*4];
986 }
987
988 /**
989 * gets the predicted number of non zero coefficients.
990 * @param n block index
991 */
992 static inline int pred_non_zero_count(H264Context *h, int n){
993 const int index8= scan8[n];
994 const int left= h->non_zero_count_cache[index8 - 1];
995 const int top = h->non_zero_count_cache[index8 - 8];
996 int i= left + top;
997
998 if(i<64) i= (i+1)>>1;
999
1000 tprintf("pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
1001
1002 return i&31;
1003 }
1004
1005 static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
1006 const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];
1007
1008 if(topright_ref != PART_NOT_AVAILABLE){
1009 *C= h->mv_cache[list][ i - 8 + part_width ];
1010 return topright_ref;
1011 }else{
1012 tprintf("topright MV not available\n");
1013
1014 *C= h->mv_cache[list][ i - 8 - 1 ];
1015 return h->ref_cache[list][ i - 8 - 1 ];
1016 }
1017 }
1018
1019 /**
1020 * gets the predicted MV.
1021 * @param n the block index
1022 * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)
1023 * @param mx the x component of the predicted motion vector
1024 * @param my the y component of the predicted motion vector
1025 */
1026 static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
1027 const int index8= scan8[n];
1028 const int top_ref= h->ref_cache[list][ index8 - 8 ];
1029 const int left_ref= h->ref_cache[list][ index8 - 1 ];
1030 const int16_t * const A= h->mv_cache[list][ index8 - 1 ];
1031 const int16_t * const B= h->mv_cache[list][ index8 - 8 ];
1032 const int16_t * C;
1033 int diagonal_ref, match_count;
1034
1035 assert(part_width==1 || part_width==2 || part_width==4);
1036
1037 /* mv_cache
1038 B . . A T T T T
1039 U . . L . . , .
1040 U . . L . . . .
1041 U . . L . . , .
1042 . . . L . . . .
1043 */
1044
1045 diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);
1046 match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);
1047 tprintf("pred_motion match_count=%d\n", match_count);
1048 if(match_count > 1){ //most common
1049 *mx= mid_pred(A[0], B[0], C[0]);
1050 *my= mid_pred(A[1], B[1], C[1]);
1051 }else if(match_count==1){
1052 if(left_ref==ref){
1053 *mx= A[0];
1054 *my= A[1];
1055 }else if(top_ref==ref){
1056 *mx= B[0];
1057 *my= B[1];
1058 }else{
1059 *mx= C[0];
1060 *my= C[1];
1061 }
1062 }else{
1063 if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
1064 *mx= A[0];
1065 *my= A[1];
1066 }else{
1067 *mx= mid_pred(A[0], B[0], C[0]);
1068 *my= mid_pred(A[1], B[1], C[1]);
1069 }
1070 }
1071
1072 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);
1073 }
1074
1075 /**
1076 * gets the directionally predicted 16x8 MV.
1077 * @param n the block index
1078 * @param mx the x component of the predicted motion vector
1079 * @param my the y component of the predicted motion vector
1080 */
1081 static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
1082 if(n==0){
1083 const int top_ref= h->ref_cache[list][ scan8[0] - 8 ];
1084 const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
1085
1086 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);
1087
1088 if(top_ref == ref){
1089 *mx= B[0];
1090 *my= B[1];
1091 return;
1092 }
1093 }else{
1094 const int left_ref= h->ref_cache[list][ scan8[8] - 1 ];
1095 const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];
1096
1097 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);
1098
1099 if(left_ref == ref){
1100 *mx= A[0];
1101 *my= A[1];
1102 return;
1103 }
1104 }
1105
1106 //RARE
1107 pred_motion(h, n, 4, list, ref, mx, my);
1108 }
1109
1110 /**
1111 * gets the directionally predicted 8x16 MV.
1112 * @param n the block index
1113 * @param mx the x component of the predicted motion vector
1114 * @param my the y component of the predicted motion vector
1115 */
1116 static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
1117 if(n==0){
1118 const int left_ref= h->ref_cache[list][ scan8[0] - 1 ];
1119 const int16_t * const A= h->mv_cache[list][ scan8[0] - 1 ];
1120
1121 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);
1122
1123 if(left_ref == ref){
1124 *mx= A[0];
1125 *my= A[1];
1126 return;
1127 }
1128 }else{
1129 const int16_t * C;
1130 int diagonal_ref;
1131
1132 diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);
1133
1134 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);
1135
1136 if(diagonal_ref == ref){
1137 *mx= C[0];
1138 *my= C[1];
1139 return;
1140 }
1141 }
1142
1143 //RARE
1144 pred_motion(h, n, 2, list, ref, mx, my);
1145 }
1146
1147 static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
1148 const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
1149 const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];
1150
1151 tprintf("pred_pskip: (%d) (%d) at %2d %2d\n", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
1152
1153 if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
1154 || (top_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0)
1155 || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){
1156
1157 *mx = *my = 0;
1158 return;
1159 }
1160
1161 pred_motion(h, 0, 4, 0, 0, mx, my);
1162
1163 return;
1164 }
1165
1166 static inline void direct_dist_scale_factor(H264Context * const h){
1167 const int poc = h->s.current_picture_ptr->poc;
1168 const int poc1 = h->ref_list[1][0].poc;
1169 int i;
1170 for(i=0; i<h->ref_count[0]; i++){
1171 int poc0 = h->ref_list[0][i].poc;
1172 int td = clip(poc1 - poc0, -128, 127);
1173 if(td == 0 /* FIXME || pic0 is a long-term ref */){
1174 h->dist_scale_factor[i] = 256;
1175 }else{
1176 int tb = clip(poc - poc0, -128, 127);
1177 int tx = (16384 + (ABS(td) >> 1)) / td;
1178 h->dist_scale_factor[i] = clip((tb*tx + 32) >> 6, -1024, 1023);
1179 }
1180 }
1181 }
1182 static inline void direct_ref_list_init(H264Context * const h){
1183 MpegEncContext * const s = &h->s;
1184 Picture * const ref1 = &h->ref_list[1][0];
1185 Picture * const cur = s->current_picture_ptr;
1186 int list, i, j;
1187 if(cur->pict_type == I_TYPE)
1188 cur->ref_count[0] = 0;
1189 if(cur->pict_type != B_TYPE)
1190 cur->ref_count[1] = 0;
1191 for(list=0; list<2; list++){
1192 cur->ref_count[list] = h->ref_count[list];
1193 for(j=0; j<h->ref_count[list]; j++)
1194 cur->ref_poc[list][j] = h->ref_list[list][j].poc;
1195 }
1196 if(cur->pict_type != B_TYPE || h->direct_spatial_mv_pred)
1197 return;
1198 for(list=0; list<2; list++){
1199 for(i=0; i<ref1->ref_count[list]; i++){
1200 const int poc = ref1->ref_poc[list][i];
1201 h->map_col_to_list0[list][i] = PART_NOT_AVAILABLE;
1202 for(j=0; j<h->ref_count[list]; j++)
1203 if(h->ref_list[list][j].poc == poc){
1204 h->map_col_to_list0[list][i] = j;
1205 break;
1206 }
1207 }
1208 }
1209 }
1210
1211 static inline void pred_direct_motion(H264Context * const h, int *mb_type){
1212 MpegEncContext * const s = &h->s;
1213 const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
1214 const int b8_xy = 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1215 const int b4_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1216 const int mb_type_col = h->ref_list[1][0].mb_type[mb_xy];
1217 const int16_t (*l1mv0)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[0][b4_xy];
1218 const int16_t (*l1mv1)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[1][b4_xy];
1219 const int8_t *l1ref0 = &h->ref_list[1][0].ref_index[0][b8_xy];
1220 const int8_t *l1ref1 = &h->ref_list[1][0].ref_index[1][b8_xy];
1221 const int is_b8x8 = IS_8X8(*mb_type);
1222 int sub_mb_type;
1223 int i8, i4;
1224
1225 if(IS_8X8(mb_type_col) && !h->sps.direct_8x8_inference_flag){
1226 /* FIXME save sub mb types from previous frames (or derive from MVs)
1227 * so we know exactly what block size to use */
1228 sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */
1229 *mb_type = MB_TYPE_8x8|MB_TYPE_L0L1;
1230 }else if(!is_b8x8 && (IS_16X16(mb_type_col) || IS_INTRA(mb_type_col))){
1231 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1232 *mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */
1233 }else{
1234 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1235 *mb_type = MB_TYPE_8x8|MB_TYPE_L0L1;
1236 }
1237 if(!is_b8x8)
1238 *mb_type |= MB_TYPE_DIRECT2;
1239
1240 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);
1241
1242 if(h->direct_spatial_mv_pred){
1243 int ref[2];
1244 int mv[2][2];
1245 int list;
1246
1247 /* ref = min(neighbors) */
1248 for(list=0; list<2; list++){
1249 int refa = h->ref_cache[list][scan8[0] - 1];
1250 int refb = h->ref_cache[list][scan8[0] - 8];
1251 int refc = h->ref_cache[list][scan8[0] - 8 + 4];
1252 if(refc == -2)
1253 refc = h->ref_cache[list][scan8[0] - 8 - 1];
1254 ref[list] = refa;
1255 if(ref[list] < 0 || (refb < ref[list] && refb >= 0))
1256 ref[list] = refb;
1257 if(ref[list] < 0 || (refc < ref[list] && refc >= 0))
1258 ref[list] = refc;
1259 if(ref[list] < 0)
1260 ref[list] = -1;
1261 }
1262
1263 if(ref[0] < 0 && ref[1] < 0){
1264 ref[0] = ref[1] = 0;
1265 mv[0][0] = mv[0][1] =
1266 mv[1][0] = mv[1][1] = 0;
1267 }else{
1268 for(list=0; list<2; list++){
1269 if(ref[list] >= 0)
1270 pred_motion(h, 0, 4, list, ref[list], &mv[list][0], &mv[list][1]);
1271 else
1272 mv[list][0] = mv[list][1] = 0;
1273 }
1274 }
1275
1276 if(ref[1] < 0){
1277 *mb_type &= ~MB_TYPE_P0L1;
1278 sub_mb_type &= ~MB_TYPE_P0L1;
1279 }else if(ref[0] < 0){
1280 *mb_type &= ~MB_TYPE_P0L0;
1281 sub_mb_type &= ~MB_TYPE_P0L0;
1282 }
1283
1284 if(IS_16X16(*mb_type)){
1285 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref[0], 1);
1286 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, ref[1], 1);
1287 if(!IS_INTRA(mb_type_col)
1288 && ( (l1ref0[0] == 0 && ABS(l1mv0[0][0]) <= 1 && ABS(l1mv0[0][1]) <= 1)
1289 || (l1ref0[0] < 0 && l1ref1[0] == 0 && ABS(l1mv1[0][0]) <= 1 && ABS(l1mv1[0][1]) <= 1
1290 && (h->x264_build>33 || !h->x264_build)))){
1291 if(ref[0] > 0)
1292 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1293 else
1294 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
1295 if(ref[1] > 0)
1296 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1297 else
1298 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
1299 }else{
1300 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1301 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1302 }
1303 }else{
1304 for(i8=0; i8<4; i8++){
1305 const int x8 = i8&1;
1306 const int y8 = i8>>1;
1307
1308 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1309 continue;
1310 h->sub_mb_type[i8] = sub_mb_type;
1311
1312 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1313 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1314 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref[0], 1);
1315 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, ref[1], 1);
1316
1317 /* col_zero_flag */
1318 if(!IS_INTRA(mb_type_col) && ( l1ref0[x8 + y8*h->b8_stride] == 0
1319 || (l1ref0[x8 + y8*h->b8_stride] < 0 && l1ref1[x8 + y8*h->b8_stride] == 0
1320 && (h->x264_build>33 || !h->x264_build)))){
1321 const int16_t (*l1mv)[2]= l1ref0[x8 + y8*h->b8_stride] == 0 ? l1mv0 : l1mv1;
1322 for(i4=0; i4<4; i4++){
1323 const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1324 if(ABS(mv_col[0]) <= 1 && ABS(mv_col[1]) <= 1){
1325 if(ref[0] == 0)
1326 *(uint32_t*)h->mv_cache[0][scan8[i8*4+i4]] = 0;
1327 if(ref[1] == 0)
1328 *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = 0;
1329 }
1330 }
1331 }
1332 }
1333 }
1334 }else{ /* direct temporal mv pred */
1335 if(IS_16X16(*mb_type)){
1336 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1);
1337 if(IS_INTRA(mb_type_col)){
1338 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
1339 fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
1340 fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
1341 }else{
1342 const int ref0 = l1ref0[0] >= 0 ? h->map_col_to_list0[0][l1ref0[0]]
1343 : h->map_col_to_list0[1][l1ref1[0]];
1344 const int dist_scale_factor = h->dist_scale_factor[ref0];
1345 const int16_t *mv_col = l1ref0[0] >= 0 ? l1mv0[0] : l1mv1[0];
1346 int mv_l0[2];
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 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref0, 1);
1350 fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv_l0[0],mv_l0[1]), 4);
1351 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);
1352 }
1353 }else{
1354 for(i8=0; i8<4; i8++){
1355 const int x8 = i8&1;
1356 const int y8 = i8>>1;
1357 int ref0, dist_scale_factor;
1358 const int16_t (*l1mv)[2]= l1mv0;
1359
1360 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1361 continue;
1362 h->sub_mb_type[i8] = sub_mb_type;
1363 if(IS_INTRA(mb_type_col)){
1364 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1365 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1366 fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1367 fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1368 continue;
1369 }
1370
1371 ref0 = l1ref0[x8 + y8*h->b8_stride];
1372 if(ref0 >= 0)
1373 ref0 = h->map_col_to_list0[0][ref0];
1374 else{
1375 ref0 = h->map_col_to_list0[1][l1ref1[x8 + y8*h->b8_stride]];
1376 l1mv= l1mv1;
1377 }
1378 dist_scale_factor = h->dist_scale_factor[ref0];
1379
1380 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1381 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1382 for(i4=0; i4<4; i4++){
1383 const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1384 int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]];
1385 mv_l0[0] = (dist_scale_factor * mv_col[0] + 128) >> 8;
1386 mv_l0[1] = (dist_scale_factor * mv_col[1] + 128) >> 8;
1387 *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] =
1388 pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
1389 }
1390 }
1391 }
1392 }
1393 }
1394
1395 static inline void write_back_motion(H264Context *h, int mb_type){
1396 MpegEncContext * const s = &h->s;
1397 const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1398 const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1399 int list;
1400
1401 for(list=0; list<2; list++){
1402 int y;
1403 if(!USES_LIST(mb_type, list)){
1404 if(1){ //FIXME skip or never read if mb_type doesn't use it
1405 for(y=0; y<4; y++){
1406 *(uint64_t*)s->current_picture.motion_val[list][b_xy + 0 + y*h->b_stride]=
1407 *(uint64_t*)s->current_picture.motion_val[list][b_xy + 2 + y*h->b_stride]= 0;
1408 }
1409 if( h->pps.cabac ) {
1410 /* FIXME needed ? */
1411 for(y=0; y<4; y++){
1412 *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]=
1413 *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= 0;
1414 }
1415 }
1416 for(y=0; y<2; y++){
1417 s->current_picture.ref_index[list][b8_xy + 0 + y*h->b8_stride]=
1418 s->current_picture.ref_index[list][b8_xy + 1 + y*h->b8_stride]= LIST_NOT_USED;
1419 }
1420 }
1421 continue;
1422 }
1423
1424 for(y=0; y<4; y++){
1425 *(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];
1426 *(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];
1427 }
1428 if( h->pps.cabac ) {
1429 for(y=0; y<4; y++){
1430 *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+0 + 8*y];
1431 *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+2 + 8*y];
1432 }
1433 }
1434 for(y=0; y<2; y++){
1435 s->current_picture.ref_index[list][b8_xy + 0 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+0 + 16*y];
1436 s->current_picture.ref_index[list][b8_xy + 1 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+2 + 16*y];
1437 }
1438 }
1439
1440 if(h->slice_type == B_TYPE && h->pps.cabac){
1441 if(IS_8X8(mb_type)){
1442 h->direct_table[b8_xy+1+0*h->b8_stride] = IS_DIRECT(h->sub_mb_type[1]) ? 1 : 0;
1443 h->direct_table[b8_xy+0+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[2]) ? 1 : 0;
1444 h->direct_table[b8_xy+1+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[3]) ? 1 : 0;
1445 }
1446 }
1447 }
1448
1449 /**
1450 * Decodes a network abstraction layer unit.
1451 * @param consumed is the number of bytes used as input
1452 * @param length is the length of the array
1453 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp tailing?
1454 * @returns decoded bytes, might be src+1 if no escapes
1455 */
1456 static uint8_t *decode_nal(H264Context *h, uint8_t *src, int *dst_length, int *consumed, int length){
1457 int i, si, di;
1458 uint8_t *dst;
1459
1460 // src[0]&0x80; //forbidden bit
1461 h->nal_ref_idc= src[0]>>5;
1462 h->nal_unit_type= src[0]&0x1F;
1463
1464 src++; length--;
1465 #if 0
1466 for(i=0; i<length; i++)
1467 printf("%2X ", src[i]);
1468 #endif
1469 for(i=0; i+1<length; i+=2){
1470 if(src[i]) continue;
1471 if(i>0 && src[i-1]==0) i--;
1472 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1473 if(src[i+2]!=3){
1474 /* startcode, so we must be past the end */
1475 length=i;
1476 }
1477 break;
1478 }
1479 }
1480
1481 if(i>=length-1){ //no escaped 0
1482 *dst_length= length;
1483 *consumed= length+1; //+1 for the header
1484 return src;
1485 }
1486
1487 h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length);
1488 dst= h->rbsp_buffer;
1489
1490 //printf("decoding esc\n");
1491 si=di=0;
1492 while(si<length){
1493 //remove escapes (very rare 1:2^22)
1494 if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1495 if(src[si+2]==3){ //escape
1496 dst[di++]= 0;
1497 dst[di++]= 0;
1498 si+=3;
1499 continue;
1500 }else //next start code
1501 break;
1502 }
1503
1504 dst[di++]= src[si++];
1505 }
1506
1507 *dst_length= di;
1508 *consumed= si + 1;//+1 for the header
1509 //FIXME store exact number of bits in the getbitcontext (its needed for decoding)
1510 return dst;
1511 }
1512
1513 #if 0
1514 /**
1515 * @param src the data which should be escaped
1516 * @param dst the target buffer, dst+1 == src is allowed as a special case
1517 * @param length the length of the src data
1518 * @param dst_length the length of the dst array
1519 * @returns length of escaped data in bytes or -1 if an error occured
1520 */
1521 static int encode_nal(H264Context *h, uint8_t *dst, uint8_t *src, int length, int dst_length){
1522 int i, escape_count, si, di;
1523 uint8_t *temp;
1524
1525 assert(length>=0);
1526 assert(dst_length>0);
1527
1528 dst[0]= (h->nal_ref_idc<<5) + h->nal_unit_type;
1529
1530 if(length==0) return 1;
1531
1532 escape_count= 0;
1533 for(i=0; i<length; i+=2){
1534 if(src[i]) continue;
1535 if(i>0 && src[i-1]==0)
1536 i--;
1537 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1538 escape_count++;
1539 i+=2;
1540 }
1541 }
1542
1543 if(escape_count==0){
1544 if(dst+1 != src)
1545 memcpy(dst+1, src, length);
1546 return length + 1;
1547 }
1548
1549 if(length + escape_count + 1> dst_length)
1550 return -1;
1551
1552 //this should be damn rare (hopefully)
1553
1554 h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length + escape_count);
1555 temp= h->rbsp_buffer;
1556 //printf("encoding esc\n");
1557
1558 si= 0;
1559 di= 0;
1560 while(si < length){
1561 if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1562 temp[di++]= 0; si++;
1563 temp[di++]= 0; si++;
1564 temp[di++]= 3;
1565 temp[di++]= src[si++];
1566 }
1567 else
1568 temp[di++]= src[si++];
1569 }
1570 memcpy(dst+1, temp, length+escape_count);
1571
1572 assert(di == length+escape_count);
1573
1574 return di + 1;
1575 }
1576
1577 /**
1578 * write 1,10,100,1000,... for alignment, yes its exactly inverse to mpeg4
1579 */
1580 static void encode_rbsp_trailing(PutBitContext *pb){
1581 int length;
1582 put_bits(pb, 1, 1);
1583 length= (-put_bits_count(pb))&7;
1584 if(length) put_bits(pb, length, 0);
1585 }
1586 #endif
1587
1588 /**
1589 * identifies the exact end of the bitstream
1590 * @return the length of the trailing, or 0 if damaged
1591 */
1592 static int decode_rbsp_trailing(uint8_t *src){
1593 int v= *src;
1594 int r;
1595
1596 tprintf("rbsp trailing %X\n", v);
1597
1598 for(r=1; r<9; r++){
1599 if(v&1) return r;
1600 v>>=1;
1601 }
1602 return 0;
1603 }
1604
1605 /**
1606 * idct tranforms the 16 dc values and dequantize them.
1607 * @param qp quantization parameter
1608 */
1609 static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
1610 #define stride 16
1611 int i;
1612 int temp[16]; //FIXME check if this is a good idea
1613 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
1614 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1615
1616 //memset(block, 64, 2*256);
1617 //return;
1618 for(i=0; i<4; i++){
1619 const int offset= y_offset[i];
1620 const int z0= block[offset+stride*0] + block[offset+stride*4];
1621 const int z1= block[offset+stride*0] - block[offset+stride*4];
1622 const int z2= block[offset+stride*1] - block[offset+stride*5];
1623 const int z3= block[offset+stride*1] + block[offset+stride*5];
1624
1625 temp[4*i+0]= z0+z3;
1626 temp[4*i+1]= z1+z2;
1627 temp[4*i+2]= z1-z2;
1628 temp[4*i+3]= z0-z3;
1629 }
1630
1631 for(i=0; i<4; i++){
1632 const int offset= x_offset[i];
1633 const int z0= temp[4*0+i] + temp[4*2+i];
1634 const int z1= temp[4*0+i] - temp[4*2+i];
1635 const int z2= temp[4*1+i] - temp[4*3+i];
1636 const int z3= temp[4*1+i] + temp[4*3+i];
1637
1638 block[stride*0 +offset]= ((((z0 + z3)*qmul + 128 ) >> 8)); //FIXME think about merging this into decode_resdual
1639 block[stride*2 +offset]= ((((z1 + z2)*qmul + 128 ) >> 8));
1640 block[stride*8 +offset]= ((((z1 - z2)*qmul + 128 ) >> 8));
1641 block[stride*10+offset]= ((((z0 - z3)*qmul + 128 ) >> 8));
1642 }
1643 }
1644
1645 #if 0
1646 /**
1647 * dct tranforms the 16 dc values.
1648 * @param qp quantization parameter ??? FIXME
1649 */
1650 static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
1651 // const int qmul= dequant_coeff[qp][0];
1652 int i;
1653 int temp[16]; //FIXME check if this is a good idea
1654 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
1655 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1656
1657 for(i=0; i<4; i++){
1658 const int offset= y_offset[i];
1659 const int z0= block[offset+stride*0] + block[offset+stride*4];
1660 const int z1= block[offset+stride*0] - block[offset+stride*4];
1661 const int z2= block[offset+stride*1] - block[offset+stride*5];
1662 const int z3= block[offset+stride*1] + block[offset+stride*5];
1663
1664 temp[4*i+0]= z0+z3;
1665 temp[4*i+1]= z1+z2;
1666 temp[4*i+2]= z1-z2;
1667 temp[4*i+3]= z0-z3;
1668 }
1669
1670 for(i=0; i<4; i++){
1671 const int offset= x_offset[i];
1672 const int z0= temp[4*0+i] + temp[4*2+i];
1673 const int z1= temp[4*0+i] - temp[4*2+i];
1674 const int z2= temp[4*1+i] - temp[4*3+i];
1675 const int z3= temp[4*1+i] + temp[4*3+i];
1676
1677 block[stride*0 +offset]= (z0 + z3)>>1;
1678 block[stride*2 +offset]= (z1 + z2)>>1;
1679 block[stride*8 +offset]= (z1 - z2)>>1;
1680 block[stride*10+offset]= (z0 - z3)>>1;
1681 }
1682 }
1683 #endif
1684
1685 #undef xStride
1686 #undef stride
1687
1688 static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
1689 const int stride= 16*2;
1690 const int xStride= 16;
1691 int a,b,c,d,e;
1692
1693 a= block[stride*0 + xStride*0];
1694 b= block[stride*0 + xStride*1];
1695 c= block[stride*1 + xStride*0];
1696 d= block[stride*1 + xStride*1];
1697
1698 e= a-b;
1699 a= a+b;
1700 b= c-d;
1701 c= c+d;
1702
1703 block[stride*0 + xStride*0]= ((a+c)*qmul) >> 7;
1704 block[stride*0 + xStride*1]= ((e+b)*qmul) >> 7;
1705 block[stride*1 + xStride*0]= ((a-c)*qmul) >> 7;
1706 block[stride*1 + xStride*1]= ((e-b)*qmul) >> 7;
1707 }
1708
1709 #if 0
1710 static void chroma_dc_dct_c(DCTELEM *block){
1711 const int stride= 16*2;
1712 const int xStride= 16;
1713 int a,b,c,d,e;
1714
1715 a= block[stride*0 + xStride*0];
1716 b= block[stride*0 + xStride*1];
1717 c= block[stride*1 + xStride*0];
1718 d= block[stride*1 + xStride*1];
1719
1720 e= a-b;
1721 a= a+b;
1722 b= c-d;
1723 c= c+d;
1724
1725 block[stride*0 + xStride*0]= (a+c);
1726 block[stride*0 + xStride*1]= (e+b);
1727 block[stride*1 + xStride*0]= (a-c);
1728 block[stride*1 + xStride*1]= (e-b);
1729 }
1730 #endif
1731
1732 /**
1733 * gets the chroma qp.
1734 */
1735 static inline int get_chroma_qp(int chroma_qp_index_offset, int qscale){
1736
1737 return chroma_qp[clip(qscale + chroma_qp_index_offset, 0, 51)];
1738 }
1739
1740
1741 #if 0
1742 static void h264_diff_dct_c(DCTELEM *block, uint8_t *src1, uint8_t *src2, int stride){
1743 int i;
1744 //FIXME try int temp instead of block
1745
1746 for(i=0; i<4; i++){
1747 const int d0= src1[0 + i*stride] - src2[0 + i*stride];
1748 const int d1= src1[1 + i*stride] - src2[1 + i*stride];
1749 const int d2= src1[2 + i*stride] - src2[2 + i*stride];
1750 const int d3= src1[3 + i*stride] - src2[3 + i*stride];
1751 const int z0= d0 + d3;
1752 const int z3= d0 - d3;
1753 const int z1= d1 + d2;
1754 const int z2= d1 - d2;
1755
1756 block[0 + 4*i]= z0 + z1;
1757 block[1 + 4*i]= 2*z3 + z2;
1758 block[2 + 4*i]= z0 - z1;
1759 block[3 + 4*i]= z3 - 2*z2;
1760 }
1761
1762 for(i=0; i<4; i++){
1763 const int z0= block[0*4 + i] + block[3*4 + i];
1764 const int z3= block[0*4 + i] - block[3*4 + i];
1765 const int z1= block[1*4 + i] + block[2*4 + i];
1766 const int z2= block[1*4 + i] - block[2*4 + i];
1767
1768 block[0*4 + i]= z0 + z1;
1769 block[1*4 + i]= 2*z3 + z2;
1770 block[2*4 + i]= z0 - z1;
1771 block[3*4 + i]= z3 - 2*z2;
1772 }
1773 }
1774 #endif
1775
1776 //FIXME need to check that this doesnt overflow signed 32 bit for low qp, i am not sure, it's very close
1777 //FIXME check that gcc inlines this (and optimizes intra & seperate_dc stuff away)
1778 static inline int quantize_c(DCTELEM *block, uint8_t *scantable, int qscale, int intra, int seperate_dc){
1779 int i;
1780 const int * const quant_table= quant_coeff[qscale];
1781 const int bias= intra ? (1<<QUANT_SHIFT)/3 : (1<<QUANT_SHIFT)/6;
1782 const unsigned int threshold1= (1<<QUANT_SHIFT) - bias - 1;
1783 const unsigned int threshold2= (threshold1<<1);
1784 int last_non_zero;
1785
1786 if(seperate_dc){
1787 if(qscale<=18){
1788 //avoid overflows
1789 const int dc_bias= intra ? (1<<(QUANT_SHIFT-2))/3 : (1<<(QUANT_SHIFT-2))/6;
1790 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT-2)) - dc_bias - 1;
1791 const unsigned int dc_threshold2= (dc_threshold1<<1);
1792
1793 int level= block[0]*quant_coeff[qscale+18][0];
1794 if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1795 if(level>0){
1796 level= (dc_bias + level)>>(QUANT_SHIFT-2);
1797 block[0]= level;
1798 }else{
1799 level= (dc_bias - level)>>(QUANT_SHIFT-2);
1800 block[0]= -level;
1801 }
1802 // last_non_zero = i;
1803 }else{
1804 block[0]=0;
1805 }
1806 }else{
1807 const int dc_bias= intra ? (1<<(QUANT_SHIFT+1))/3 : (1<<(QUANT_SHIFT+1))/6;
1808 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT+1)) - dc_bias - 1;
1809 const unsigned int dc_threshold2= (dc_threshold1<<1);
1810
1811 int level= block[0]*quant_table[0];
1812 if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1813 if(level>0){
1814 level= (dc_bias + level)>>(QUANT_SHIFT+1);
1815 block[0]= level;
1816 }else{
1817 level= (dc_bias - level)>>(QUANT_SHIFT+1);
1818 block[0]= -level;
1819 }
1820 // last_non_zero = i;
1821 }else{
1822 block[0]=0;
1823 }
1824 }
1825 last_non_zero= 0;
1826 i=1;
1827 }else{
1828 last_non_zero= -1;
1829 i=0;
1830 }
1831
1832 for(; i<16; i++){
1833 const int j= scantable[i];
1834 int level= block[j]*quant_table[j];
1835
1836 // if( bias+level >= (1<<(QMAT_SHIFT - 3))
1837 // || bias-level >= (1<<(QMAT_SHIFT - 3))){
1838 if(((unsigned)(level+threshold1))>threshold2){
1839 if(level>0){
1840 level= (bias + level)>>QUANT_SHIFT;
1841 block[j]= level;
1842 }else{
1843 level= (bias - level)>>QUANT_SHIFT;
1844 block[j]= -level;
1845 }
1846 last_non_zero = i;
1847 }else{
1848 block[j]=0;
1849 }
1850 }
1851
1852 return last_non_zero;
1853 }
1854
1855 static void pred4x4_vertical_c(uint8_t *src, uint8_t *topright, int stride){
1856 const uint32_t a= ((uint32_t*)(src-stride))[0];
1857 ((uint32_t*)(src+0*stride))[0]= a;
1858 ((uint32_t*)(src+1*stride))[0]= a;
1859 ((uint32_t*)(src+2*stride))[0]= a;
1860 ((uint32_t*)(src+3*stride))[0]= a;
1861 }
1862
1863 static void pred4x4_horizontal_c(uint8_t *src, uint8_t *topright, int stride){
1864 ((uint32_t*)(src+0*stride))[0]= src[-1+0*stride]*0x01010101;
1865 ((uint32_t*)(src+1*stride))[0]= src[-1+1*stride]*0x01010101;
1866 ((uint32_t*)(src+2*stride))[0]= src[-1+2*stride]*0x01010101;
1867 ((uint32_t*)(src+3*stride))[0]= src[-1+3*stride]*0x01010101;
1868 }
1869
1870 static void pred4x4_dc_c(uint8_t *src, uint8_t *topright, int stride){
1871 const int dc= ( src[-stride] + src[1-stride] + src[2-stride] + src[3-stride]
1872 + src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 4) >>3;
1873
1874 ((uint32_t*)(src+0*stride))[0]=
1875 ((uint32_t*)(src+1*stride))[0]=
1876 ((uint32_t*)(src+2*stride))[0]=
1877 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1878 }
1879
1880 static void pred4x4_left_dc_c(uint8_t *src, uint8_t *topright, int stride){
1881 const int dc= ( src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 2) >>2;
1882
1883 ((uint32_t*)(src+0*stride))[0]=
1884 ((uint32_t*)(src+1*stride))[0]=
1885 ((uint32_t*)(src+2*stride))[0]=
1886 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1887 }
1888
1889 static void pred4x4_top_dc_c(uint8_t *src, uint8_t *topright, int stride){
1890 const int dc= ( src[-stride] + src[1-stride] + src[2-stride] + src[3-stride] + 2) >>2;
1891
1892 ((uint32_t*)(src+0*stride))[0]=
1893 ((uint32_t*)(src+1*stride))[0]=
1894 ((uint32_t*)(src+2*stride))[0]=
1895 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1896 }
1897
1898 static void pred4x4_128_dc_c(uint8_t *src, uint8_t *topright, int stride){
1899 ((uint32_t*)(src+0*stride))[0]=
1900 ((uint32_t*)(src+1*stride))[0]=
1901 ((uint32_t*)(src+2*stride))[0]=
1902 ((uint32_t*)(src+3*stride))[0]= 128U*0x01010101U;
1903 }
1904
1905
1906 #define LOAD_TOP_RIGHT_EDGE\
1907 const int t4= topright[0];\
1908 const int t5= topright[1];\
1909 const int t6= topright[2];\
1910 const int t7= topright[3];\
1911
1912 #define LOAD_LEFT_EDGE\
1913 const int l0= src[-1+0*stride];\
1914 const int l1= src[-1+1*stride];\
1915 const int l2= src[-1+2*stride];\
1916 const int l3= src[-1+3*stride];\
1917
1918 #define LOAD_TOP_EDGE\
1919 const int t0= src[ 0-1*stride];\
1920 const int t1= src[ 1-1*stride];\
1921 const int t2= src[ 2-1*stride];\
1922 const int t3= src[ 3-1*stride];\
1923
1924 static void pred4x4_down_right_c(uint8_t *src, uint8_t *topright, int stride){
1925 const int lt= src[-1-1*stride];
1926 LOAD_TOP_EDGE
1927 LOAD_LEFT_EDGE
1928
1929 src[0+3*stride]=(l3 + 2*l2 + l1 + 2)>>2;
1930 src[0+2*stride]=
1931 src[1+3*stride]=(l2 + 2*l1 + l0 + 2)>>2;
1932 src[0+1*stride]=
1933 src[1+2*stride]=
1934 src[2+3*stride]=(l1 + 2*l0 + lt + 2)>>2;
1935 src[0+0*stride]=
1936 src[1+1*stride]=
1937 src[2+2*stride]=
1938 src[3+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1939 src[1+0*stride]=
1940 src[2+1*stride]=
1941 src[3+2*stride]=(lt + 2*t0 + t1 + 2)>>2;
1942 src[2+0*stride]=
1943 src[3+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1944 src[3+0*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1945 }
1946
1947 static void pred4x4_down_left_c(uint8_t *src, uint8_t *topright, int stride){
1948 LOAD_TOP_EDGE
1949 LOAD_TOP_RIGHT_EDGE
1950 // LOAD_LEFT_EDGE
1951
1952 src[0+0*stride]=(t0 + t2 + 2*t1 + 2)>>2;
1953 src[1+0*stride]=
1954 src[0+1*stride]=(t1 + t3 + 2*t2 + 2)>>2;
1955 src[2+0*stride]=
1956 src[1+1*stride]=
1957 src[0+2*stride]=(t2 + t4 + 2*t3 + 2)>>2;
1958 src[3+0*stride]=
1959 src[2+1*stride]=
1960 src[1+2*stride]=
1961 src[0+3*stride]=(t3 + t5 + 2*t4 + 2)>>2;
1962 src[3+1*stride]=
1963 src[2+2*stride]=
1964 src[1+3*stride]=(t4 + t6 + 2*t5 + 2)>>2;
1965 src[3+2*stride]=
1966 src[2+3*stride]=(t5 + t7 + 2*t6 + 2)>>2;
1967 src[3+3*stride]=(t6 + 3*t7 + 2)>>2;
1968 }
1969
1970 static void pred4x4_vertical_right_c(uint8_t *src, uint8_t *topright, int stride){
1971 const int lt= src[-1-1*stride];
1972 LOAD_TOP_EDGE
1973 LOAD_LEFT_EDGE
1974 const __attribute__((unused)) int unu= l3;
1975
1976 src[0+0*stride]=
1977 src[1+2*stride]=(lt + t0 + 1)>>1;
1978 src[1+0*stride]=
1979 src[2+2*stride]=(t0 + t1 + 1)>>1;
1980 src[2+0*stride]=
1981 src[3+2*stride]=(t1 + t2 + 1)>>1;
1982 src[3+0*stride]=(t2 + t3 + 1)>>1;
1983 src[0+1*stride]=
1984 src[1+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1985 src[1+1*stride]=
1986 src[2+3*stride]=(lt + 2*t0 + t1 + 2)>>2;
1987 src[2+1*stride]=
1988 src[3+3*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1989 src[3+1*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1990 src[0+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1991 src[0+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1992 }
1993
1994 static void pred4x4_vertical_left_c(uint8_t *src, uint8_t *topright, int stride){
1995 LOAD_TOP_EDGE
1996 LOAD_TOP_RIGHT_EDGE
1997 const __attribute__((unused)) int unu= t7;
1998
1999 src[0+0*stride]=(t0 + t1 + 1)>>1;
2000 src[1+0*stride]=
2001 src[0+2*stride]=(t1 + t2 + 1)>>1;
2002 src[2+0*stride]=
2003 src[1+2*stride]=(t2 + t3 + 1)>>1;
2004 src[3+0*stride]=
2005 src[2+2*stride]=(t3 + t4+ 1)>>1;
2006 src[3+2*stride]=(t4 + t5+ 1)>>1;
2007 src[0+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
2008 src[1+1*stride]=
2009 src[0+3*stride]=(t1 + 2*t2 + t3 + 2)>>2;
2010 src[2+1*stride]=
2011 src[1+3*stride]=(t2 + 2*t3 + t4 + 2)>>2;
2012 src[3+1*stride]=
2013 src[2+3*stride]=(t3 + 2*t4 + t5 + 2)>>2;
2014 src[3+3*stride]=(t4 + 2*t5 + t6 + 2)>>2;
2015 }
2016
2017 static void pred4x4_horizontal_up_c(uint8_t *src, uint8_t *topright, int stride){
2018 LOAD_LEFT_EDGE
2019
2020 src[0+0*stride]=(l0 + l1 + 1)>>1;
2021 src[1+0*stride]=(l0 + 2*l1 + l2 + 2)>>2;
2022 src[2+0*stride]=
2023 src[0+1*stride]=(l1 + l2 + 1)>>1;
2024 src[3+0*stride]=
2025 src[1+1*stride]=(l1 + 2*l2 + l3 + 2)>>2;
2026 src[2+1*stride]=
2027 src[0+2*stride]=(l2 + l3 + 1)>>1;
2028 src[3+1*stride]=
2029 src[1+2*stride]=(l2 + 2*l3 + l3 + 2)>>2;
2030 src[3+2*stride]=
2031 src[1+3*stride]=
2032 src[0+3*stride]=
2033 src[2+2*stride]=
2034 src[2+3*stride]=
2035 src[3+3*stride]=l3;
2036 }
2037
2038 static void pred4x4_horizontal_down_c(uint8_t *src, uint8_t *topright, int stride){
2039 const int lt= src[-1-1*stride];
2040 LOAD_TOP_EDGE
2041 LOAD_LEFT_EDGE
2042 const __attribute__((unused)) int unu= t3;
2043
2044 src[0+0*stride]=
2045 src[2+1*stride]=(lt + l0 + 1)>>1;
2046 src[1+0*stride]=
2047 src[3+1*stride]=(l0 + 2*lt + t0 + 2)>>2;
2048 src[2+0*stride]=(lt + 2*t0 + t1 + 2)>>2;
2049 src[3+0*stride]=(t0 + 2*t1 + t2 + 2)>>2;
2050 src[0+1*stride]=
2051 src[2+2*stride]=(l0 + l1 + 1)>>1;
2052 src[1+1*stride]=
2053 src[3+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
2054 src[0+2*stride]=
2055 src[2+3*stride]=(l1 + l2+ 1)>>1;
2056 src[1+2*stride]=
2057 src[3+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
2058 src[0+3*stride]=(l2 + l3 + 1)>>1;
2059 src[1+3*stride]=(l1 + 2*l2 + l3 + 2)>>2;
2060 }
2061
2062 static void pred16x16_vertical_c(uint8_t *src, int stride){
2063 int i;
2064 const uint32_t a= ((uint32_t*)(src-stride))[0];
2065 const uint32_t b= ((uint32_t*)(src-stride))[1];
2066 const uint32_t c= ((uint32_t*)(src-stride))[2];
2067 const uint32_t d= ((uint32_t*)(src-stride))[3];
2068
2069 for(i=0; i<16; i++){
2070 ((uint32_t*)(src+i*stride))[0]= a;
2071 ((uint32_t*)(src+i*stride))[1]= b;
2072 ((uint32_t*)(src+i*stride))[2]= c;
2073 ((uint32_t*)(src+i*stride))[3]= d;
2074 }
2075 }
2076
2077 static void pred16x16_horizontal_c(uint8_t *src, int stride){
2078 int i;
2079
2080 for(i=0; i<16; i++){
2081 ((uint32_t*)(src+i*stride))[0]=
2082 ((uint32_t*)(src+i*stride))[1]=
2083 ((uint32_t*)(src+i*stride))[2]=
2084 ((uint32_t*)(src+i*stride))[3]= src[-1+i*stride]*0x01010101;
2085 }
2086 }
2087
2088 static void pred16x16_dc_c(uint8_t *src, int stride){
2089 int i, dc=0;
2090
2091 for(i=0;i<16; i++){
2092 dc+= src[-1+i*stride];
2093 }
2094
2095 for(i=0;i<16; i++){
2096 dc+= src[i-stride];
2097 }
2098
2099 dc= 0x01010101*((dc + 16)>>5);
2100
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;
2106 }
2107 }
2108
2109 static void pred16x16_left_dc_c(uint8_t *src, int stride){
2110 int i, dc=0;
2111
2112 for(i=0;i<16; i++){
2113 dc+= src[-1+i*stride];
2114 }
2115
2116 dc= 0x01010101*((dc + 8)>>4);
2117
2118 for(i=0; i<16; i++){
2119 ((uint32_t*)(src+i*stride))[0]=
2120 ((uint32_t*)(src+i*stride))[1]=
2121 ((uint32_t*)(src+i*stride))[2]=
2122 ((uint32_t*)(src+i*stride))[3]= dc;
2123 }
2124 }
2125
2126 static void pred16x16_top_dc_c(uint8_t *src, int stride){
2127 int i, dc=0;
2128
2129 for(i=0;i<16; i++){
2130 dc+= src[i-stride];
2131 }
2132 dc= 0x01010101*((dc + 8)>>4);
2133
2134 for(i=0; i<16; i++){
2135 ((uint32_t*)(src+i*stride))[0]=
2136 ((uint32_t*)(src+i*stride))[1]=
2137 ((uint32_t*)(src+i*stride))[2]=
2138 ((uint32_t*)(src+i*stride))[3]= dc;
2139 }
2140 }
2141
2142 static void pred16x16_128_dc_c(uint8_t *src, int stride){
2143 int i;
2144
2145 for(i=0; i<16; i++){
2146 ((uint32_t*)(src+i*stride))[0]=
2147 ((uint32_t*)(src+i*stride))[1]=
2148 ((uint32_t*)(src+i*stride))[2]=
2149 ((uint32_t*)(src+i*stride))[3]= 0x01010101U*128U;
2150 }
2151 }
2152
2153 static inline void pred16x16_plane_compat_c(uint8_t *src, int stride, const int svq3){
2154 int i, j, k;
2155 int a;
2156 uint8_t *cm = cropTbl + MAX_NEG_CROP;
2157 const uint8_t * const src0 = src+7-stride;
2158 const uint8_t *src1 = src+8*stride-1;
2159 const uint8_t *src2 = src1-2*stride; // == src+6*stride-1;
2160 int H = src0[1] - src0[-1];
2161 int V = src1[0] - src2[ 0];
2162 for(k=2; k<=8; ++k) {
2163 src1 += stride; src2 -= stride;
2164 H += k*(src0[k] - src0[-k]);
2165 V += k*(src1[0] - src2[ 0]);
2166 }
2167 if(svq3){
2168 H = ( 5*(H/4) ) / 16;
2169 V = ( 5*(V/4) ) / 16;
2170
2171 /* required for 100% accuracy */
2172 i = H; H = V; V = i;
2173 }else{
2174 H = ( 5*H+32 ) >> 6;
2175 V = ( 5*V+32 ) >> 6;
2176 }
2177
2178 a = 16*(src1[0] + src2[16] + 1) - 7*(V+H);
2179 for(j=16; j>0; --j) {
2180 int b = a;
2181 a += V;
2182 for(i=-16; i<0; i+=4) {
2183 src[16+i] = cm[ (b ) >> 5 ];
2184 src[17+i] = cm[ (b+ H) >> 5 ];
2185 src[18+i] = cm[ (b+2*H) >> 5 ];
2186 src[19+i] = cm[ (b+3*H) >> 5 ];
2187 b += 4*H;
2188 }
2189 src += stride;
2190 }
2191 }
2192
2193 static void pred16x16_plane_c(uint8_t *src, int stride){
2194 pred16x16_plane_compat_c(src, stride, 0);
2195 }
2196
2197 static void pred8x8_vertical_c(uint8_t *src, int stride){
2198 int i;
2199 const uint32_t a= ((uint32_t*)(src-stride))[0];
2200 const uint32_t b= ((uint32_t*)(src-stride))[1];
2201
2202 for(i=0; i<8; i++){
2203 ((uint32_t*)(src+i*stride))[0]= a;
2204 ((uint32_t*)(src+i*stride))[1]= b;
2205 }
2206 }
2207
2208 static void pred8x8_horizontal_c(uint8_t *src, int stride){
2209 int i;
2210
2211 for(i=0; i<8; i++){
2212 ((uint32_t*)(src+i*stride))[0]=
2213 ((uint32_t*)(src+i*stride))[1]= src[-1+i*stride]*0x01010101;
2214 }
2215 }
2216
2217 static void pred8x8_128_dc_c(uint8_t *src, int stride){
2218 int i;
2219
2220 for(i=0; i<8; i++){
2221 ((uint32_t*)(src+i*stride))[0]=
2222 ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
2223 }
2224 }
2225
2226 static void pred8x8_left_dc_c(uint8_t *src, int stride){
2227 int i;
2228 int dc0, dc2;
2229
2230 dc0=dc2=0;
2231 for(i=0;i<4; i++){
2232 dc0+= src[-1+i*stride];
2233 dc2+= src[-1+(i+4)*stride];
2234 }
2235 dc0= 0x01010101*((dc0 + 2)>>2);
2236 dc2= 0x01010101*((dc2 + 2)>>2);
2237
2238 for(i=0; i<4; i++){
2239 ((uint32_t*)(src+i*stride))[0]=
2240 ((uint32_t*)(src+i*stride))[1]= dc0;
2241 }
2242 for(i=4; i<8; i++){
2243 ((uint32_t*)(src+i*stride))[0]=
2244 ((uint32_t*)(src+i*stride))[1]= dc2;
2245 }
2246 }
2247
2248 static void pred8x8_top_dc_c(uint8_t *src, int stride){
2249 int i;
2250 int dc0, dc1;
2251
2252 dc0=dc1=0;
2253 for(i=0;i<4; i++){
2254 dc0+= src[i-stride];
2255 dc1+= src[4+i-stride];
2256 }
2257 dc0= 0x01010101*((dc0 + 2)>>2);
2258 dc1= 0x01010101*((dc1 + 2)>>2);
2259
2260 for(i=0; i<4; i++){
2261 ((uint32_t*)(src+i*stride))[0]= dc0;
2262 ((uint32_t*)(src+i*stride))[1]= dc1;
2263 }
2264 for(i=4; i<8; i++){
2265 ((uint32_t*)(src+i*stride))[0]= dc0;
2266 ((uint32_t*)(src+i*stride))[1]= dc1;
2267 }
2268 }
2269
2270
2271 static void pred8x8_dc_c(uint8_t *src, int stride){
2272 int i;
2273 int dc0, dc1, dc2, dc3;
2274
2275 dc0=dc1=dc2=0;
2276 for(i=0;i<4; i++){
2277 dc0+= src[-1+i*stride] + src[i-stride];
2278 dc1+= src[4+i-stride];
2279 dc2+= src[-1+(i+4)*stride];
2280 }
2281 dc3= 0x01010101*((dc1 + dc2 + 4)>>3);
2282 dc0= 0x01010101*((dc0 + 4)>>3);
2283 dc1= 0x01010101*((dc1 + 2)>>2);
2284 dc2= 0x01010101*((dc2 + 2)>>2);
2285
2286 for(i=0; i<4; i++){
2287 ((uint32_t*)(src+i*stride))[0]= dc0;
2288 ((uint32_t*)(src+i*stride))[1]= dc1;
2289 }
2290 for(i=4; i<8; i++){
2291 ((uint32_t*)(src+i*stride))[0]= dc2;
2292 ((uint32_t*)(src+i*stride))[1]= dc3;
2293 }
2294 }
2295
2296 static void pred8x8_plane_c(uint8_t *src, int stride){
2297 int j, k;
2298 int a;
2299 uint8_t *cm = cropTbl + MAX_NEG_CROP;
2300 const uint8_t * const src0 = src+3-stride;
2301 const uint8_t *src1 = src+4*stride-1;
2302 const uint8_t *src2 = src1-2*stride; // == src+2*stride-1;
2303 int H = src0[1] - src0[-1];
2304 int V = src1[0] - src2[ 0];
2305 for(k=2; k<=4; ++k) {
2306 src1 += stride; src2 -= stride;
2307 H += k*(src0[k] - src0[-k]);
2308 V += k*(src1[0] - src2[ 0]);
2309 }
2310 H = ( 17*H+16 ) >> 5;
2311 V = ( 17*V+16 ) >> 5;
2312
2313 a = 16*(src1[0] + src2[8]+1) - 3*(V+H);
2314 for(j=8; j>0; --j) {
2315 int b = a;
2316 a += V;
2317 src[0] = cm[ (b ) >> 5 ];
2318 src[1] = cm[ (b+ H) >> 5 ];
2319 src[2] = cm[ (b+2*H) >> 5 ];
2320 src[3] = cm[ (b+3*H) >> 5 ];
2321 src[4] = cm[ (b+4*H) >> 5 ];
2322 src[5] = cm[ (b+5*H) >> 5 ];
2323 src[6] = cm[ (b+6*H) >> 5 ];
2324 src[7] = cm[ (b+7*H) >> 5 ];
2325 src += stride;
2326 }
2327 }
2328
2329 #define SRC(x,y) src[(x)+(y)*stride]
2330 #define PL(y) \
2331 const int l##y = (SRC(-1,y-1) + 2*SRC(-1,y) + SRC(-1,y+1) + 2) >> 2;
2332 #define PREDICT_8x8_LOAD_LEFT \
2333 const int l0 = ((has_topleft ? SRC(-1,-1) : SRC(-1,0)) \
2334 + 2*SRC(-1,0) + SRC(-1,1) + 2) >> 2; \
2335 PL(1) PL(2) PL(3) PL(4) PL(5) PL(6) \
2336 const int l7 attribute_unused = (SRC(-1,6) + 3*SRC(-1,7) + 2) >> 2
2337
2338 #define PT(x) \
2339 const int t##x = (SRC(x-1,-1) + 2*SRC(x,-1) + SRC(x+1,-1) + 2) >> 2;
2340 #define PREDICT_8x8_LOAD_TOP \
2341 const int t0 = ((has_topleft ? SRC(-1,-1) : SRC(0,-1)) \
2342 + 2*SRC(0,-1) + SRC(1,-1) + 2) >> 2; \
2343 PT(1) PT(2) PT(3) PT(4) PT(5) PT(6) \
2344 const int t7 attribute_unused = ((has_topright ? SRC(8,-1) : SRC(7,-1)) \
2345 + 2*SRC(7,-1) + SRC(6,-1) + 2) >> 2
2346
2347 #define PTR(x) \
2348 t##x = (SRC(x-1,-1) + 2*SRC(x,-1) + SRC(x+1,-1) + 2) >> 2;
2349 #define PREDICT_8x8_LOAD_TOPRIGHT \
2350 int t8, t9, t10, t11, t12, t13, t14, t15; \
2351 if(has_topright) { \
2352 PTR(8) PTR(9) PTR(10) PTR(11) PTR(12) PTR(13) PTR(14) \
2353 t15 = (SRC(14,-1) + 3*SRC(15,-1) + 2) >> 2; \
2354 } else t8=t9=t10=t11=t12=t13=t14=t15= SRC(7,-1);
2355
2356 #define PREDICT_8x8_LOAD_TOPLEFT \
2357 const int lt = (SRC(-1,0) + 2*SRC(-1,-1) + SRC(0,-1) + 2) >> 2
2358
2359 #define PREDICT_8x8_DC(v) \
2360 int y; \
2361 for( y = 0; y < 8; y++ ) { \
2362 ((uint32_t*)src)[0] = \
2363 ((uint32_t*)src)[1] = v; \
2364 src += stride; \
2365 }
2366
2367 static void pred8x8l_128_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2368 {
2369 PREDICT_8x8_DC(0x80808080);
2370 }
2371 static void pred8x8l_left_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2372 {
2373 PREDICT_8x8_LOAD_LEFT;
2374 const uint32_t dc = ((l0+l1+l2+l3+l4+l5+l6+l7+4) >> 3) * 0x01010101;
2375 PREDICT_8x8_DC(dc);
2376 }
2377 static void pred8x8l_top_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2378 {
2379 PREDICT_8x8_LOAD_TOP;
2380 const uint32_t dc = ((t0+t1+t2+t3+t4+t5+t6+t7+4) >> 3) * 0x01010101;
2381 PREDICT_8x8_DC(dc);
2382 }
2383 static void pred8x8l_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2384 {
2385 PREDICT_8x8_LOAD_LEFT;
2386 PREDICT_8x8_LOAD_TOP;
2387 const uint32_t dc = ((l0+l1+l2+l3+l4+l5+l6+l7
2388 +t0+t1+t2+t3+t4+t5+t6+t7+8) >> 4) * 0x01010101;
2389 PREDICT_8x8_DC(dc);
2390 }
2391 static void pred8x8l_horizontal_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2392 {
2393 PREDICT_8x8_LOAD_LEFT;
2394 #define ROW(y) ((uint32_t*)(src+y*stride))[0] =\
2395 ((uint32_t*)(src+y*stride))[1] = 0x01010101 * l##y
2396 ROW(0); ROW(1); ROW(2); ROW(3); ROW(4); ROW(5); ROW(6); ROW(7);
2397 #undef ROW
2398 }
2399 static void pred8x8l_vertical_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2400 {
2401 int y;
2402 PREDICT_8x8_LOAD_TOP;
2403 src[0] = t0;
2404 src[1] = t1;
2405 src[2] = t2;
2406 src[3] = t3;
2407 src[4] = t4;
2408 src[5] = t5;
2409 src[6] = t6;
2410 src[7] = t7;
2411 for( y = 1; y < 8; y++ )
2412 *(uint64_t*)(src+y*stride) = *(uint64_t*)src;
2413 }
2414 static void pred8x8l_down_left_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2415 {
2416 PREDICT_8x8_LOAD_TOP;
2417 PREDICT_8x8_LOAD_TOPRIGHT;
2418 SRC(0,0)= (t0 + 2*t1 + t2 + 2) >> 2;
2419 SRC(0,1)=SRC(1,0)= (t1 + 2*t2 + t3 + 2) >> 2;
2420 SRC(0,2)=SRC(1,1)=SRC(2,0)= (t2 + 2*t3 + t4 + 2) >> 2;
2421 SRC(0,3)=SRC(1,2)=SRC(2,1)=SRC(3,0)= (t3 + 2*t4 + t5 + 2) >> 2;
2422 SRC(0,4)=SRC(1,3)=SRC(2,2)=SRC(3,1)=SRC(4,0)= (t4 + 2*t5 + t6 + 2) >> 2;
2423 SRC(0,5)=SRC(1,4)=SRC(2,3)=SRC(3,2)=SRC(4,1)=SRC(5,0)= (t5 + 2*t6 + t7 + 2) >> 2;
2424 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;
2425 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;
2426 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;
2427 SRC(2,7)=SRC(3,6)=SRC(4,5)=SRC(5,4)=SRC(6,3)=SRC(7,2)= (t9 + 2*t10 + t11 + 2) >> 2;
2428 SRC(3,7)=SRC(4,6)=SRC(5,5)=SRC(6,4)=SRC(7,3)= (t10 + 2*t11 + t12 + 2) >> 2;
2429 SRC(4,7)=SRC(5,6)=SRC(6,5)=SRC(7,4)= (t11 + 2*t12 + t13 + 2) >> 2;
2430 SRC(5,7)=SRC(6,6)=SRC(7,5)= (t12 + 2*t13 + t14 + 2) >> 2;
2431 SRC(6,7)=SRC(7,6)= (t13 + 2*t14 + t15 + 2) >> 2;
2432 SRC(7,7)= (t14 + 3*t15 + 2) >> 2;
2433 }
2434 static void pred8x8l_down_right_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2435 {
2436 PREDICT_8x8_LOAD_TOP;
2437 PREDICT_8x8_LOAD_LEFT;
2438 PREDICT_8x8_LOAD_TOPLEFT;
2439 SRC(0,7)= (l7 + 2*l6 + l5 + 2) >> 2;
2440 SRC(0,6)=SRC(1,7)= (l6 + 2*l5 + l4 + 2) >> 2;
2441 SRC(0,5)=SRC(1,6)=SRC(2,7)= (l5 + 2*l4 + l3 + 2) >> 2;
2442 SRC(0,4)=SRC(1,5)=SRC(2,6)=SRC(3,7)= (l4 + 2*l3 + l2 + 2) >> 2;
2443 SRC(0,3)=SRC(1,4)=SRC(2,5)=SRC(3,6)=SRC(4,7)= (l3 + 2*l2 + l1 + 2) >> 2;
2444 SRC(0,2)=SRC(1,3)=SRC(2,4)=SRC(3,5)=SRC(4,6)=SRC(5,7)= (l2 + 2*l1 + l0 + 2) >> 2;
2445 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;
2446 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;
2447 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;
2448 SRC(2,0)=SRC(3,1)=SRC(4,2)=SRC(5,3)=SRC(6,4)=SRC(7,5)= (t0 + 2*t1 + t2 + 2) >> 2;
2449 SRC(3,0)=SRC(4,1)=SRC(5,2)=SRC(6,3)=SRC(7,4)= (t1 + 2*t2 + t3 + 2) >> 2;
2450 SRC(4,0)=SRC(5,1)=SRC(6,2)=SRC(7,3)= (t2 + 2*t3 + t4 + 2) >> 2;
2451 SRC(5,0)=SRC(6,1)=SRC(7,2)= (t3 + 2*t4 + t5 + 2) >> 2;
2452 SRC(6,0)=SRC(7,1)= (t4 + 2*t5 + t6 + 2) >> 2;
2453 SRC(7,0)= (t5 + 2*t6 + t7 + 2) >> 2;
2454
2455 }
2456 static void pred8x8l_vertical_right_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2457 {
2458 PREDICT_8x8_LOAD_TOP;
2459 PREDICT_8x8_LOAD_LEFT;
2460 PREDICT_8x8_LOAD_TOPLEFT;
2461 SRC(0,6)= (l5 + 2*l4 + l3 + 2) >> 2;
2462 SRC(0,7)= (l6 + 2*l5 + l4 + 2) >> 2;
2463 SRC(0,4)=SRC(1,6)= (l3 + 2*l2 + l1 + 2) >> 2;
2464 SRC(0,5)=SRC(1,7)= (l4 + 2*l3 + l2 + 2) >> 2;
2465 SRC(0,2)=SRC(1,4)=SRC(2,6)= (l1 + 2*l0 + lt + 2) >> 2;
2466 SRC(0,3)=SRC(1,5)=SRC(2,7)= (l2 + 2*l1 + l0 + 2) >> 2;
2467 SRC(0,1)=SRC(1,3)=SRC(2,5)=SRC(3,7)= (l0 + 2*lt + t0 + 2) >> 2;
2468 SRC(0,0)=SRC(1,2)=SRC(2,4)=SRC(3,6)= (lt + t0 + 1) >> 1;
2469 SRC(1,1)=SRC(2,3)=SRC(3,5)=SRC(4,7)= (lt + 2*t0 + t1 + 2) >> 2;
2470 SRC(1,0)=SRC(2,2)=SRC(3,4)=SRC(4,6)= (t0 + t1 + 1) >> 1;
2471 SRC(2,1)=SRC(3,3)=SRC(4,5)=SRC(5,7)= (t0 + 2*t1 + t2 + 2) >> 2;
2472 SRC(2,0)=SRC(3,2)=SRC(4,4)=SRC(5,6)= (t1 + t2 + 1) >> 1;
2473 SRC(3,1)=SRC(4,3)=SRC(5,5)=SRC(6,7)= (t1 + 2*t2 + t3 + 2) >> 2;
2474 SRC(3,0)=SRC(4,2)=SRC(5,4)=SRC(6,6)= (t2 + t3 + 1) >> 1;
2475 SRC(4,1)=SRC(5,3)=SRC(6,5)=SRC(7,7)= (t2 + 2*t3 + t4 + 2) >> 2;
2476 SRC(4,0)=SRC(5,2)=SRC(6,4)=SRC(7,6)= (t3 + t4 + 1) >> 1;
2477 SRC(5,1)=SRC(6,3)=SRC(7,5)= (t3 + 2*t4 + t5 + 2) >> 2;
2478 SRC(5,0)=SRC(6,2)=SRC(7,4)= (t4 + t5 + 1) >> 1;
2479 SRC(6,1)=SRC(7,3)= (t4 + 2*t5 + t6 + 2) >> 2;
2480 SRC(6,0)=SRC(7,2)= (t5 + t6 + 1) >> 1;
2481 SRC(7,1)= (t5 + 2*t6 + t7 + 2) >> 2;
2482 SRC(7,0)= (t6 + t7 + 1) >> 1;
2483 }
2484 static void pred8x8l_horizontal_down_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2485 {
2486 PREDICT_8x8_LOAD_TOP;
2487 PREDICT_8x8_LOAD_LEFT;
2488 PREDICT_8x8_LOAD_TOPLEFT;
2489 SRC(0,7)= (l6 + l7 + 1) >> 1;
2490 SRC(1,7)= (l5 + 2*l6 + l7 + 2) >> 2;
2491 SRC(0,6)=SRC(2,7)= (l5 + l6 + 1) >> 1;
2492 SRC(1,6)=SRC(3,7)= (l4 + 2*l5 + l6 + 2) >> 2;
2493 SRC(0,5)=SRC(2,6)=SRC(4,7)= (l4 + l5 + 1) >> 1;
2494 SRC(1,5)=SRC(3,6)=SRC(5,7)= (l3 + 2*l4 + l5 + 2) >> 2;
2495 SRC(0,4)=SRC(2,5)=SRC(4,6)=SRC(6,7)= (l3 + l4 + 1) >> 1;
2496 SRC(1,4)=SRC(3,5)=SRC(5,6)=SRC(7,7)= (l2 + 2*l3 + l4 + 2) >> 2;
2497 SRC(0,3)=SRC(2,4)=SRC(4,5)=SRC(6,6)= (l2 + l3 + 1) >> 1;
2498 SRC(1,3)=SRC(3,4)=SRC(5,5)=SRC(7,6)= (l1 + 2*l2 + l3 + 2) >> 2;
2499 SRC(0,2)=SRC(2,3)=SRC(4,4)=SRC(6,5)= (l1 + l2 + 1) >> 1;
2500 SRC(1,2)=SRC(3,3)=SRC(5,4)=SRC(7,5)= (l0 + 2*l1 + l2 + 2) >> 2;
2501 SRC(0,1)=SRC(2,2)=SRC(4,3)=SRC(6,4)= (l0 + l1 + 1) >> 1;
2502 SRC(1,1)=SRC(3,2)=SRC(5,3)=SRC(7,4)= (lt + 2*l0 + l1 + 2) >> 2;
2503 SRC(0,0)=SRC(2,1)=SRC(4,2)=SRC(6,3)= (lt + l0 + 1) >> 1;
2504 SRC(1,0)=SRC(3,1)=SRC(5,2)=SRC(7,3)= (l0 + 2*lt + t0 + 2) >> 2;
2505 SRC(2,0)=SRC(4,1)=SRC(6,2)= (t1 + 2*t0 + lt + 2) >> 2;
2506 SRC(3,0)=SRC(5,1)=SRC(7,2)= (t2 + 2*t1 + t0 + 2) >> 2;
2507 SRC(4,0)=SRC(6,1)= (t3 + 2*t2 + t1 + 2) >> 2;
2508 SRC(5,0)=SRC(7,1)= (t4 + 2*t3 + t2 + 2) >> 2;
2509 SRC(6,0)= (t5 + 2*t4 + t3 + 2) >> 2;
2510 SRC(7,0)= (t6 + 2*t5 + t4 + 2) >> 2;
2511 }
2512 static void pred8x8l_vertical_left_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2513 {
2514 PREDICT_8x8_LOAD_TOP;
2515 PREDICT_8x8_LOAD_TOPRIGHT;
2516 SRC(0,0)= (t0 + t1 + 1) >> 1;
2517 SRC(0,1)= (t0 + 2*t1 + t2 + 2) >> 2;
2518 SRC(0,2)=SRC(1,0)= (t1 + t2 + 1) >> 1;
2519 SRC(0,3)=SRC(1,1)= (t1 + 2*t2 + t3 + 2) >> 2;
2520 SRC(0,4)=SRC(1,2)=SRC(2,0)= (t2 + t3 + 1) >> 1;
2521 SRC(0,5)=SRC(1,3)=SRC(2,1)= (t2 + 2*t3 + t4 + 2) >> 2;
2522 SRC(