43e349e36bd0d006d683b4eafb2ead4488a8c758
[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., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 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_AUD 9
159 #define NAL_END_SEQUENCE 10
160 #define NAL_END_STREAM 11
161 #define NAL_FILLER_DATA 12
162 #define NAL_SPS_EXT 13
163 #define NAL_AUXILIARY_SLICE 19
164 uint8_t *rbsp_buffer;
165 unsigned int rbsp_buffer_size;
166
167 /**
168 * Used to parse AVC variant of h264
169 */
170 int is_avc; ///< this flag is != 0 if codec is avc1
171 int got_avcC; ///< flag used to parse avcC data only once
172 int nal_length_size; ///< Number of bytes used for nal length (1, 2 or 4)
173
174 int chroma_qp; //QPc
175
176 int prev_mb_skipped; //FIXME remove (IMHO not used)
177
178 //prediction stuff
179 int chroma_pred_mode;
180 int intra16x16_pred_mode;
181
182 int top_mb_xy;
183 int left_mb_xy[2];
184
185 int8_t intra4x4_pred_mode_cache[5*8];
186 int8_t (*intra4x4_pred_mode)[8];
187 void (*pred4x4 [9+3])(uint8_t *src, uint8_t *topright, int stride);//FIXME move to dsp?
188 void (*pred8x8l [9+3])(uint8_t *src, int topleft, int topright, int stride);
189 void (*pred8x8 [4+3])(uint8_t *src, int stride);
190 void (*pred16x16[4+3])(uint8_t *src, int stride);
191 unsigned int topleft_samples_available;
192 unsigned int top_samples_available;
193 unsigned int topright_samples_available;
194 unsigned int left_samples_available;
195 uint8_t (*top_borders[2])[16+2*8];
196 uint8_t left_border[2*(17+2*9)];
197
198 /**
199 * non zero coeff count cache.
200 * is 64 if not available.
201 */
202 uint8_t non_zero_count_cache[6*8] __align8;
203 uint8_t (*non_zero_count)[16];
204
205 /**
206 * Motion vector cache.
207 */
208 int16_t mv_cache[2][5*8][2] __align8;
209 int8_t ref_cache[2][5*8] __align8;
210 #define LIST_NOT_USED -1 //FIXME rename?
211 #define PART_NOT_AVAILABLE -2
212
213 /**
214 * is 1 if the specific list MV&references are set to 0,0,-2.
215 */
216 int mv_cache_clean[2];
217
218 /**
219 * number of neighbors (top and/or left) that used 8x8 dct
220 */
221 int neighbor_transform_size;
222
223 /**
224 * block_offset[ 0..23] for frame macroblocks
225 * block_offset[24..47] for field macroblocks
226 */
227 int block_offset[2*(16+8)];
228
229 uint32_t *mb2b_xy; //FIXME are these 4 a good idea?
230 uint32_t *mb2b8_xy;
231 int b_stride; //FIXME use s->b4_stride
232 int b8_stride;
233
234 int halfpel_flag;
235 int thirdpel_flag;
236
237 int unknown_svq3_flag;
238 int next_slice_index;
239
240 SPS sps_buffer[MAX_SPS_COUNT];
241 SPS sps; ///< current sps
242
243 PPS pps_buffer[MAX_PPS_COUNT];
244 /**
245 * current pps
246 */
247 PPS pps; //FIXME move to Picture perhaps? (->no) do we need that?
248
249 uint32_t dequant4_buffer[6][52][16];
250 uint32_t dequant8_buffer[2][52][64];
251 uint32_t (*dequant4_coeff[6])[16];
252 uint32_t (*dequant8_coeff[2])[64];
253 int dequant_coeff_pps; ///< reinit tables when pps changes
254
255 int slice_num;
256 uint8_t *slice_table_base;
257 uint8_t *slice_table; ///< slice_table_base + mb_stride + 1
258 int slice_type;
259 int slice_type_fixed;
260
261 //interlacing specific flags
262 int mb_aff_frame;
263 int mb_field_decoding_flag;
264
265 int sub_mb_type[4];
266
267 //POC stuff
268 int poc_lsb;
269 int poc_msb;
270 int delta_poc_bottom;
271 int delta_poc[2];
272 int frame_num;
273 int prev_poc_msb; ///< poc_msb of the last reference pic for POC type 0
274 int prev_poc_lsb; ///< poc_lsb of the last reference pic for POC type 0
275 int frame_num_offset; ///< for POC type 2
276 int prev_frame_num_offset; ///< for POC type 2
277 int prev_frame_num; ///< frame_num of the last pic for POC type 1/2
278
279 /**
280 * frame_num for frames or 2*frame_num for field pics.
281 */
282 int curr_pic_num;
283
284 /**
285 * max_frame_num or 2*max_frame_num for field pics.
286 */
287 int max_pic_num;
288
289 //Weighted pred stuff
290 int use_weight;
291 int use_weight_chroma;
292 int luma_log2_weight_denom;
293 int chroma_log2_weight_denom;
294 int luma_weight[2][16];
295 int luma_offset[2][16];
296 int chroma_weight[2][16][2];
297 int chroma_offset[2][16][2];
298 int implicit_weight[16][16];
299
300 //deblock
301 int deblocking_filter; ///< disable_deblocking_filter_idc with 1<->0
302 int slice_alpha_c0_offset;
303 int slice_beta_offset;
304
305 int redundant_pic_count;
306
307 int direct_spatial_mv_pred;
308 int dist_scale_factor[16];
309 int map_col_to_list0[2][16];
310
311 /**
312 * num_ref_idx_l0/1_active_minus1 + 1
313 */
314 int ref_count[2];// FIXME split for AFF
315 Picture *short_ref[32];
316 Picture *long_ref[32];
317 Picture default_ref_list[2][32];
318 Picture ref_list[2][32]; //FIXME size?
319 Picture field_ref_list[2][32]; //FIXME size?
320 Picture *delayed_pic[16]; //FIXME size?
321 Picture *delayed_output_pic;
322
323 /**
324 * memory management control operations buffer.
325 */
326 MMCO mmco[MAX_MMCO_COUNT];
327 int mmco_index;
328
329 int long_ref_count; ///< number of actual long term references
330 int short_ref_count; ///< number of actual short term references
331
332 //data partitioning
333 GetBitContext intra_gb;
334 GetBitContext inter_gb;
335 GetBitContext *intra_gb_ptr;
336 GetBitContext *inter_gb_ptr;
337
338 DCTELEM mb[16*24] __align8;
339
340 /**
341 * Cabac
342 */
343 CABACContext cabac;
344 uint8_t cabac_state[460];
345 int cabac_init_idc;
346
347 /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0,1,2), 0x0? luma_cbp */
348 uint16_t *cbp_table;
349 int top_cbp;
350 int left_cbp;
351 /* chroma_pred_mode for i4x4 or i16x16, else 0 */
352 uint8_t *chroma_pred_mode_table;
353 int last_qscale_diff;
354 int16_t (*mvd_table[2])[2];
355 int16_t mvd_cache[2][5*8][2] __align8;
356 uint8_t *direct_table;
357 uint8_t direct_cache[5*8];
358
359 uint8_t zigzag_scan[16];
360 uint8_t field_scan[16];
361 const uint8_t *zigzag_scan_q0;
362 const uint8_t *field_scan_q0;
363
364 int x264_build;
365 }H264Context;
366
367 static VLC coeff_token_vlc[4];
368 static VLC chroma_dc_coeff_token_vlc;
369
370 static VLC total_zeros_vlc[15];
371 static VLC chroma_dc_total_zeros_vlc[3];
372
373 static VLC run_vlc[6];
374 static VLC run7_vlc;
375
376 static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
377 static void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
378 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);
379
380 static always_inline uint32_t pack16to32(int a, int b){
381 #ifdef WORDS_BIGENDIAN
382 return (b&0xFFFF) + (a<<16);
383 #else
384 return (a&0xFFFF) + (b<<16);
385 #endif
386 }
387
388 /**
389 * fill a rectangle.
390 * @param h height of the rectangle, should be a constant
391 * @param w width of the rectangle, should be a constant
392 * @param size the size of val (1 or 4), should be a constant
393 */
394 static always_inline void fill_rectangle(void *vp, int w, int h, int stride, uint32_t val, int size){
395 uint8_t *p= (uint8_t*)vp;
396 assert(size==1 || size==4);
397
398 w *= size;
399 stride *= size;
400
401 assert((((long)vp)&(FFMIN(w, STRIDE_ALIGN)-1)) == 0);
402 assert((stride&(w-1))==0);
403 //FIXME check what gcc generates for 64 bit on x86 and possibly write a 32 bit ver of it
404 if(w==2 && h==2){
405 *(uint16_t*)(p + 0)=
406 *(uint16_t*)(p + stride)= size==4 ? val : val*0x0101;
407 }else if(w==2 && h==4){
408 *(uint16_t*)(p + 0*stride)=
409 *(uint16_t*)(p + 1*stride)=
410 *(uint16_t*)(p + 2*stride)=
411 *(uint16_t*)(p + 3*stride)= size==4 ? val : val*0x0101;
412 }else if(w==4 && h==1){
413 *(uint32_t*)(p + 0*stride)= size==4 ? val : val*0x01010101;
414 }else if(w==4 && h==2){
415 *(uint32_t*)(p + 0*stride)=
416 *(uint32_t*)(p + 1*stride)= size==4 ? val : val*0x01010101;
417 }else if(w==4 && h==4){
418 *(uint32_t*)(p + 0*stride)=
419 *(uint32_t*)(p + 1*stride)=
420 *(uint32_t*)(p + 2*stride)=
421 *(uint32_t*)(p + 3*stride)= size==4 ? val : val*0x01010101;
422 }else if(w==8 && h==1){
423 *(uint32_t*)(p + 0)=
424 *(uint32_t*)(p + 4)= size==4 ? val : val*0x01010101;
425 }else if(w==8 && h==2){
426 *(uint32_t*)(p + 0 + 0*stride)=
427 *(uint32_t*)(p + 4 + 0*stride)=
428 *(uint32_t*)(p + 0 + 1*stride)=
429 *(uint32_t*)(p + 4 + 1*stride)= size==4 ? val : val*0x01010101;
430 }else if(w==8 && h==4){
431 *(uint64_t*)(p + 0*stride)=
432 *(uint64_t*)(p + 1*stride)=
433 *(uint64_t*)(p + 2*stride)=
434 *(uint64_t*)(p + 3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
435 }else if(w==16 && h==2){
436 *(uint64_t*)(p + 0+0*stride)=
437 *(uint64_t*)(p + 8+0*stride)=
438 *(uint64_t*)(p + 0+1*stride)=
439 *(uint64_t*)(p + 8+1*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
440 }else if(w==16 && h==4){
441 *(uint64_t*)(p + 0+0*stride)=
442 *(uint64_t*)(p + 8+0*stride)=
443 *(uint64_t*)(p + 0+1*stride)=
444 *(uint64_t*)(p + 8+1*stride)=
445 *(uint64_t*)(p + 0+2*stride)=
446 *(uint64_t*)(p + 8+2*stride)=
447 *(uint64_t*)(p + 0+3*stride)=
448 *(uint64_t*)(p + 8+3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
449 }else
450 assert(0);
451 }
452
453 static void fill_caches(H264Context *h, int mb_type, int for_deblock){
454 MpegEncContext * const s = &h->s;
455 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
456 int topleft_xy, top_xy, topright_xy, left_xy[2];
457 int topleft_type, top_type, topright_type, left_type[2];
458 int left_block[8];
459 int i;
460
461 //FIXME deblocking can skip fill_caches much of the time with multiple slices too.
462 // the actual condition is whether we're on the edge of a slice,
463 // and even then the intra and nnz parts are unnecessary.
464 if(for_deblock && h->slice_num == 1)
465 return;
466
467 //wow what a mess, why didn't they simplify the interlacing&intra stuff, i can't imagine that these complex rules are worth it
468
469 top_xy = mb_xy - s->mb_stride;
470 topleft_xy = top_xy - 1;
471 topright_xy= top_xy + 1;
472 left_xy[1] = left_xy[0] = mb_xy-1;
473 left_block[0]= 0;
474 left_block[1]= 1;
475 left_block[2]= 2;
476 left_block[3]= 3;
477 left_block[4]= 7;
478 left_block[5]= 10;
479 left_block[6]= 8;
480 left_block[7]= 11;
481 if(h->mb_aff_frame){
482 const int pair_xy = s->mb_x + (s->mb_y & ~1)*s->mb_stride;
483 const int top_pair_xy = pair_xy - s->mb_stride;
484 const int topleft_pair_xy = top_pair_xy - 1;
485 const int topright_pair_xy = top_pair_xy + 1;
486 const int topleft_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[topleft_pair_xy]);
487 const int top_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[top_pair_xy]);
488 const int topright_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[topright_pair_xy]);
489 const int left_mb_frame_flag = !IS_INTERLACED(s->current_picture.mb_type[pair_xy-1]);
490 const int curr_mb_frame_flag = !IS_INTERLACED(mb_type);
491 const int bottom = (s->mb_y & 1);
492 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);
493 if (bottom
494 ? !curr_mb_frame_flag // bottom macroblock
495 : (!curr_mb_frame_flag && !top_mb_frame_flag) // top macroblock
496 ) {
497 top_xy -= s->mb_stride;
498 }
499 if (bottom
500 ? !curr_mb_frame_flag // bottom macroblock
501 : (!curr_mb_frame_flag && !topleft_mb_frame_flag) // top macroblock
502 ) {
503 topleft_xy -= s->mb_stride;
504 }
505 if (bottom
506 ? !curr_mb_frame_flag // bottom macroblock
507 : (!curr_mb_frame_flag && !topright_mb_frame_flag) // top macroblock
508 ) {
509 topright_xy -= s->mb_stride;
510 }
511 if (left_mb_frame_flag != curr_mb_frame_flag) {
512 left_xy[1] = left_xy[0] = pair_xy - 1;
513 if (curr_mb_frame_flag) {
514 if (bottom) {
515 left_block[0]= 2;
516 left_block[1]= 2;
517 left_block[2]= 3;
518 left_block[3]= 3;
519 left_block[4]= 8;
520 left_block[5]= 11;
521 left_block[6]= 8;
522 left_block[7]= 11;
523 } else {
524 left_block[0]= 0;
525 left_block[1]= 0;
526 left_block[2]= 1;
527 left_block[3]= 1;
528 left_block[4]= 7;
529 left_block[5]= 10;
530 left_block[6]= 7;
531 left_block[7]= 10;
532 }
533 } else {
534 left_xy[1] += s->mb_stride;
535 //left_block[0]= 0;
536 left_block[1]= 2;
537 left_block[2]= 0;
538 left_block[3]= 2;
539 //left_block[4]= 7;
540 left_block[5]= 10;
541 left_block[6]= 7;
542 left_block[7]= 10;
543 }
544 }
545 }
546
547 h->top_mb_xy = top_xy;
548 h->left_mb_xy[0] = left_xy[0];
549 h->left_mb_xy[1] = left_xy[1];
550 if(for_deblock){
551 topleft_type = h->slice_table[topleft_xy ] < 255 ? s->current_picture.mb_type[topleft_xy] : 0;
552 top_type = h->slice_table[top_xy ] < 255 ? s->current_picture.mb_type[top_xy] : 0;
553 topright_type= h->slice_table[topright_xy] < 255 ? s->current_picture.mb_type[topright_xy]: 0;
554 left_type[0] = h->slice_table[left_xy[0] ] < 255 ? s->current_picture.mb_type[left_xy[0]] : 0;
555 left_type[1] = h->slice_table[left_xy[1] ] < 255 ? s->current_picture.mb_type[left_xy[1]] : 0;
556 }else{
557 topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
558 top_type = h->slice_table[top_xy ] == h->slice_num ? s->current_picture.mb_type[top_xy] : 0;
559 topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
560 left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
561 left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
562 }
563
564 if(IS_INTRA(mb_type)){
565 h->topleft_samples_available=
566 h->top_samples_available=
567 h->left_samples_available= 0xFFFF;
568 h->topright_samples_available= 0xEEEA;
569
570 if(!IS_INTRA(top_type) && (top_type==0 || h->pps.constrained_intra_pred)){
571 h->topleft_samples_available= 0xB3FF;
572 h->top_samples_available= 0x33FF;
573 h->topright_samples_available= 0x26EA;
574 }
575 for(i=0; i<2; i++){
576 if(!IS_INTRA(left_type[i]) && (left_type[i]==0 || h->pps.constrained_intra_pred)){
577 h->topleft_samples_available&= 0xDF5F;
578 h->left_samples_available&= 0x5F5F;
579 }
580 }
581
582 if(!IS_INTRA(topleft_type) && (topleft_type==0 || h->pps.constrained_intra_pred))
583 h->topleft_samples_available&= 0x7FFF;
584
585 if(!IS_INTRA(topright_type) && (topright_type==0 || h->pps.constrained_intra_pred))
586 h->topright_samples_available&= 0xFBFF;
587
588 if(IS_INTRA4x4(mb_type)){
589 if(IS_INTRA4x4(top_type)){
590 h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
591 h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
592 h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
593 h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
594 }else{
595 int pred;
596 if(!top_type || (IS_INTER(top_type) && h->pps.constrained_intra_pred))
597 pred= -1;
598 else{
599 pred= 2;
600 }
601 h->intra4x4_pred_mode_cache[4+8*0]=
602 h->intra4x4_pred_mode_cache[5+8*0]=
603 h->intra4x4_pred_mode_cache[6+8*0]=
604 h->intra4x4_pred_mode_cache[7+8*0]= pred;
605 }
606 for(i=0; i<2; i++){
607 if(IS_INTRA4x4(left_type[i])){
608 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
609 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
610 }else{
611 int pred;
612 if(!left_type[i] || (IS_INTER(left_type[i]) && h->pps.constrained_intra_pred))
613 pred= -1;
614 else{
615 pred= 2;
616 }
617 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
618 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
619 }
620 }
621 }
622 }
623
624
625 /*
626 0 . T T. T T T T
627 1 L . .L . . . .
628 2 L . .L . . . .
629 3 . T TL . . . .
630 4 L . .L . . . .
631 5 L . .. . . . .
632 */
633 //FIXME constraint_intra_pred & partitioning & nnz (lets hope this is just a typo in the spec)
634 if(top_type){
635 h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][4];
636 h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][5];
637 h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][6];
638 h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3];
639
640 h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][9];
641 h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8];
642
643 h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][12];
644 h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11];
645
646 }else{
647 h->non_zero_count_cache[4+8*0]=
648 h->non_zero_count_cache[5+8*0]=
649 h->non_zero_count_cache[6+8*0]=
650 h->non_zero_count_cache[7+8*0]=
651
652 h->non_zero_count_cache[1+8*0]=
653 h->non_zero_count_cache[2+8*0]=
654
655 h->non_zero_count_cache[1+8*3]=
656 h->non_zero_count_cache[2+8*3]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
657
658 }
659
660 for (i=0; i<2; i++) {
661 if(left_type[i]){
662 h->non_zero_count_cache[3+8*1 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[0+2*i]];
663 h->non_zero_count_cache[3+8*2 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[1+2*i]];
664 h->non_zero_count_cache[0+8*1 + 8*i]= h->non_zero_count[left_xy[i]][left_block[4+2*i]];
665 h->non_zero_count_cache[0+8*4 + 8*i]= h->non_zero_count[left_xy[i]][left_block[5+2*i]];
666 }else{
667 h->non_zero_count_cache[3+8*1 + 2*8*i]=
668 h->non_zero_count_cache[3+8*2 + 2*8*i]=
669 h->non_zero_count_cache[0+8*1 + 8*i]=
670 h->non_zero_count_cache[0+8*4 + 8*i]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
671 }
672 }
673
674 if( h->pps.cabac ) {
675 // top_cbp
676 if(top_type) {
677 h->top_cbp = h->cbp_table[top_xy];
678 } else if(IS_INTRA(mb_type)) {
679 h->top_cbp = 0x1C0;
680 } else {
681 h->top_cbp = 0;
682 }
683 // left_cbp
684 if (left_type[0]) {
685 h->left_cbp = h->cbp_table[left_xy[0]] & 0x1f0;
686 } else if(IS_INTRA(mb_type)) {
687 h->left_cbp = 0x1C0;
688 } else {
689 h->left_cbp = 0;
690 }
691 if (left_type[0]) {
692 h->left_cbp |= ((h->cbp_table[left_xy[0]]>>((left_block[0]&(~1))+1))&0x1) << 1;
693 }
694 if (left_type[1]) {
695 h->left_cbp |= ((h->cbp_table[left_xy[1]]>>((left_block[2]&(~1))+1))&0x1) << 3;
696 }
697 }
698
699 #if 1
700 //FIXME direct mb can skip much of this
701 if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
702 int list;
703 for(list=0; list<1+(h->slice_type==B_TYPE); list++){
704 if(!USES_LIST(mb_type, list) && !IS_DIRECT(mb_type) && !h->deblocking_filter){
705 /*if(!h->mv_cache_clean[list]){
706 memset(h->mv_cache [list], 0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
707 memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
708 h->mv_cache_clean[list]= 1;
709 }*/
710 continue;
711 }
712 h->mv_cache_clean[list]= 0;
713
714 if(IS_INTER(top_type)){
715 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
716 const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
717 *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0];
718 *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];
719 *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];
720 *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];
721 h->ref_cache[list][scan8[0] + 0 - 1*8]=
722 h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
723 h->ref_cache[list][scan8[0] + 2 - 1*8]=
724 h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
725 }else{
726 *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]=
727 *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]=
728 *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]=
729 *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;
730 *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
731 }
732
733 //FIXME unify cleanup or sth
734 if(IS_INTER(left_type[0])){
735 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
736 const int b8_xy= h->mb2b8_xy[left_xy[0]] + 1;
737 *(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]];
738 *(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]];
739 h->ref_cache[list][scan8[0] - 1 + 0*8]=
740 h->ref_cache[list][scan8[0] - 1 + 1*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0]>>1)];
741 }else{
742 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 0*8]=
743 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 1*8]= 0;
744 h->ref_cache[list][scan8[0] - 1 + 0*8]=
745 h->ref_cache[list][scan8[0] - 1 + 1*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
746 }
747
748 if(IS_INTER(left_type[1])){
749 const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
750 const int b8_xy= h->mb2b8_xy[left_xy[1]] + 1;
751 *(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]];
752 *(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]];
753 h->ref_cache[list][scan8[0] - 1 + 2*8]=
754 h->ref_cache[list][scan8[0] - 1 + 3*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[2]>>1)];
755 }else{
756 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 2*8]=
757 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 3*8]= 0;
758 h->ref_cache[list][scan8[0] - 1 + 2*8]=
759 h->ref_cache[list][scan8[0] - 1 + 3*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
760 assert((!left_type[0]) == (!left_type[1]));
761 }
762
763 if(for_deblock || (IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred))
764 continue;
765
766 if(IS_INTER(topleft_type)){
767 const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
768 const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + h->b8_stride;
769 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
770 h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
771 }else{
772 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
773 h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
774 }
775
776 if(IS_INTER(topright_type)){
777 const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
778 const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
779 *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
780 h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
781 }else{
782 *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
783 h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
784 }
785
786
787 h->ref_cache[list][scan8[5 ]+1] =
788 h->ref_cache[list][scan8[7 ]+1] =
789 h->ref_cache[list][scan8[13]+1] = //FIXME remove past 3 (init somewhere else)
790 h->ref_cache[list][scan8[4 ]] =
791 h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
792 *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
793 *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
794 *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
795 *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
796 *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
797
798 if( h->pps.cabac ) {
799 /* XXX beurk, Load mvd */
800 if(IS_INTER(topleft_type)){
801 const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
802 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy];
803 }else{
804 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 - 1*8]= 0;
805 }
806
807 if(IS_INTER(top_type)){
808 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
809 *(uint32_t*)h->mvd_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 0];
810 *(uint32_t*)h->mvd_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 1];
811 *(uint32_t*)h->mvd_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 2];
812 *(uint32_t*)h->mvd_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 3];
813 }else{
814 *(uint32_t*)h->mvd_cache [list][scan8[0] + 0 - 1*8]=
815 *(uint32_t*)h->mvd_cache [list][scan8[0] + 1 - 1*8]=
816 *(uint32_t*)h->mvd_cache [list][scan8[0] + 2 - 1*8]=
817 *(uint32_t*)h->mvd_cache [list][scan8[0] + 3 - 1*8]= 0;
818 }
819 if(IS_INTER(left_type[0])){
820 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
821 *(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]];
822 *(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]];
823 }else{
824 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 0*8]=
825 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 1*8]= 0;
826 }
827 if(IS_INTER(left_type[1])){
828 const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
829 *(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]];
830 *(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]];
831 }else{
832 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 2*8]=
833 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 3*8]= 0;
834 }
835 *(uint32_t*)h->mvd_cache [list][scan8[5 ]+1]=
836 *(uint32_t*)h->mvd_cache [list][scan8[7 ]+1]=
837 *(uint32_t*)h->mvd_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewhere else)
838 *(uint32_t*)h->mvd_cache [list][scan8[4 ]]=
839 *(uint32_t*)h->mvd_cache [list][scan8[12]]= 0;
840
841 if(h->slice_type == B_TYPE){
842 fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, 0, 1);
843
844 if(IS_DIRECT(top_type)){
845 *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0x01010101;
846 }else if(IS_8X8(top_type)){
847 int b8_xy = h->mb2b8_xy[top_xy] + h->b8_stride;
848 h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy];
849 h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 1];
850 }else{
851 *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0;
852 }
853
854 //FIXME interlacing
855 if(IS_DIRECT(left_type[0])){
856 h->direct_cache[scan8[0] - 1 + 0*8]=
857 h->direct_cache[scan8[0] - 1 + 2*8]= 1;
858 }else if(IS_8X8(left_type[0])){
859 int b8_xy = h->mb2b8_xy[left_xy[0]] + 1;
860 h->direct_cache[scan8[0] - 1 + 0*8]= h->direct_table[b8_xy];
861 h->direct_cache[scan8[0] - 1 + 2*8]= h->direct_table[b8_xy + h->b8_stride];
862 }else{
863 h->direct_cache[scan8[0] - 1 + 0*8]=
864 h->direct_cache[scan8[0] - 1 + 2*8]= 0;
865 }
866 }
867 }
868 }
869 }
870 #endif
871
872 h->neighbor_transform_size= !!IS_8x8DCT(top_type) + !!IS_8x8DCT(left_type[0]);
873 }
874
875 static inline void write_back_intra_pred_mode(H264Context *h){
876 MpegEncContext * const s = &h->s;
877 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
878
879 h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
880 h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
881 h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
882 h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
883 h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
884 h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
885 h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
886 }
887
888 /**
889 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
890 */
891 static inline int check_intra4x4_pred_mode(H264Context *h){
892 MpegEncContext * const s = &h->s;
893 static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
894 static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
895 int i;
896
897 if(!(h->top_samples_available&0x8000)){
898 for(i=0; i<4; i++){
899 int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
900 if(status<0){
901 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);
902 return -1;
903 } else if(status){
904 h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
905 }
906 }
907 }
908
909 if(!(h->left_samples_available&0x8000)){
910 for(i=0; i<4; i++){
911 int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
912 if(status<0){
913 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);
914 return -1;
915 } else if(status){
916 h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
917 }
918 }
919 }
920
921 return 0;
922 } //FIXME cleanup like next
923
924 /**
925 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
926 */
927 static inline int check_intra_pred_mode(H264Context *h, int mode){
928 MpegEncContext * const s = &h->s;
929 static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
930 static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
931
932 if(mode < 0 || mode > 6) {
933 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);
934 return -1;
935 }
936
937 if(!(h->top_samples_available&0x8000)){
938 mode= top[ mode ];
939 if(mode<0){
940 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);
941 return -1;
942 }
943 }
944
945 if(!(h->left_samples_available&0x8000)){
946 mode= left[ mode ];
947 if(mode<0){
948 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);
949 return -1;
950 }
951 }
952
953 return mode;
954 }
955
956 /**
957 * gets the predicted intra4x4 prediction mode.
958 */
959 static inline int pred_intra_mode(H264Context *h, int n){
960 const int index8= scan8[n];
961 const int left= h->intra4x4_pred_mode_cache[index8 - 1];
962 const int top = h->intra4x4_pred_mode_cache[index8 - 8];
963 const int min= FFMIN(left, top);
964
965 tprintf("mode:%d %d min:%d\n", left ,top, min);
966
967 if(min<0) return DC_PRED;
968 else return min;
969 }
970
971 static inline void write_back_non_zero_count(H264Context *h){
972 MpegEncContext * const s = &h->s;
973 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
974
975 h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[7+8*1];
976 h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[7+8*2];
977 h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[7+8*3];
978 h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];
979 h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[4+8*4];
980 h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[5+8*4];
981 h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[6+8*4];
982
983 h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[1+8*2];
984 h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
985 h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[2+8*1];
986
987 h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[1+8*5];
988 h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
989 h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[2+8*4];
990 }
991
992 /**
993 * gets the predicted number of non zero coefficients.
994 * @param n block index
995 */
996 static inline int pred_non_zero_count(H264Context *h, int n){
997 const int index8= scan8[n];
998 const int left= h->non_zero_count_cache[index8 - 1];
999 const int top = h->non_zero_count_cache[index8 - 8];
1000 int i= left + top;
1001
1002 if(i<64) i= (i+1)>>1;
1003
1004 tprintf("pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
1005
1006 return i&31;
1007 }
1008
1009 static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
1010 const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];
1011
1012 if(topright_ref != PART_NOT_AVAILABLE){
1013 *C= h->mv_cache[list][ i - 8 + part_width ];
1014 return topright_ref;
1015 }else{
1016 tprintf("topright MV not available\n");
1017
1018 *C= h->mv_cache[list][ i - 8 - 1 ];
1019 return h->ref_cache[list][ i - 8 - 1 ];
1020 }
1021 }
1022
1023 /**
1024 * gets the predicted MV.
1025 * @param n the block index
1026 * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)
1027 * @param mx the x component of the predicted motion vector
1028 * @param my the y component of the predicted motion vector
1029 */
1030 static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
1031 const int index8= scan8[n];
1032 const int top_ref= h->ref_cache[list][ index8 - 8 ];
1033 const int left_ref= h->ref_cache[list][ index8 - 1 ];
1034 const int16_t * const A= h->mv_cache[list][ index8 - 1 ];
1035 const int16_t * const B= h->mv_cache[list][ index8 - 8 ];
1036 const int16_t * C;
1037 int diagonal_ref, match_count;
1038
1039 assert(part_width==1 || part_width==2 || part_width==4);
1040
1041 /* mv_cache
1042 B . . A T T T T
1043 U . . L . . , .
1044 U . . L . . . .
1045 U . . L . . , .
1046 . . . L . . . .
1047 */
1048
1049 diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);
1050 match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);
1051 tprintf("pred_motion match_count=%d\n", match_count);
1052 if(match_count > 1){ //most common
1053 *mx= mid_pred(A[0], B[0], C[0]);
1054 *my= mid_pred(A[1], B[1], C[1]);
1055 }else if(match_count==1){
1056 if(left_ref==ref){
1057 *mx= A[0];
1058 *my= A[1];
1059 }else if(top_ref==ref){
1060 *mx= B[0];
1061 *my= B[1];
1062 }else{
1063 *mx= C[0];
1064 *my= C[1];
1065 }
1066 }else{
1067 if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
1068 *mx= A[0];
1069 *my= A[1];
1070 }else{
1071 *mx= mid_pred(A[0], B[0], C[0]);
1072 *my= mid_pred(A[1], B[1], C[1]);
1073 }
1074 }
1075
1076 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);
1077 }
1078
1079 /**
1080 * gets the directionally predicted 16x8 MV.
1081 * @param n the block index
1082 * @param mx the x component of the predicted motion vector
1083 * @param my the y component of the predicted motion vector
1084 */
1085 static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
1086 if(n==0){
1087 const int top_ref= h->ref_cache[list][ scan8[0] - 8 ];
1088 const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
1089
1090 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);
1091
1092 if(top_ref == ref){
1093 *mx= B[0];
1094 *my= B[1];
1095 return;
1096 }
1097 }else{
1098 const int left_ref= h->ref_cache[list][ scan8[8] - 1 ];
1099 const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];
1100
1101 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);
1102
1103 if(left_ref == ref){
1104 *mx= A[0];
1105 *my= A[1];
1106 return;
1107 }
1108 }
1109
1110 //RARE
1111 pred_motion(h, n, 4, list, ref, mx, my);
1112 }
1113
1114 /**
1115 * gets the directionally predicted 8x16 MV.
1116 * @param n the block index
1117 * @param mx the x component of the predicted motion vector
1118 * @param my the y component of the predicted motion vector
1119 */
1120 static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
1121 if(n==0){
1122 const int left_ref= h->ref_cache[list][ scan8[0] - 1 ];
1123 const int16_t * const A= h->mv_cache[list][ scan8[0] - 1 ];
1124
1125 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);
1126
1127 if(left_ref == ref){
1128 *mx= A[0];
1129 *my= A[1];
1130 return;
1131 }
1132 }else{
1133 const int16_t * C;
1134 int diagonal_ref;
1135
1136 diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);
1137
1138 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);
1139
1140 if(diagonal_ref == ref){
1141 *mx= C[0];
1142 *my= C[1];
1143 return;
1144 }
1145 }
1146
1147 //RARE
1148 pred_motion(h, n, 2, list, ref, mx, my);
1149 }
1150
1151 static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
1152 const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
1153 const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];
1154
1155 tprintf("pred_pskip: (%d) (%d) at %2d %2d\n", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
1156
1157 if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
1158 || (top_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0)
1159 || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){
1160
1161 *mx = *my = 0;
1162 return;
1163 }
1164
1165 pred_motion(h, 0, 4, 0, 0, mx, my);
1166
1167 return;
1168 }
1169
1170 static inline void direct_dist_scale_factor(H264Context * const h){
1171 const int poc = h->s.current_picture_ptr->poc;
1172 const int poc1 = h->ref_list[1][0].poc;
1173 int i;
1174 for(i=0; i<h->ref_count[0]; i++){
1175 int poc0 = h->ref_list[0][i].poc;
1176 int td = clip(poc1 - poc0, -128, 127);
1177 if(td == 0 /* FIXME || pic0 is a long-term ref */){
1178 h->dist_scale_factor[i] = 256;
1179 }else{
1180 int tb = clip(poc - poc0, -128, 127);
1181 int tx = (16384 + (ABS(td) >> 1)) / td;
1182 h->dist_scale_factor[i] = clip((tb*tx + 32) >> 6, -1024, 1023);
1183 }
1184 }
1185 }
1186 static inline void direct_ref_list_init(H264Context * const h){
1187 MpegEncContext * const s = &h->s;
1188 Picture * const ref1 = &h->ref_list[1][0];
1189 Picture * const cur = s->current_picture_ptr;
1190 int list, i, j;
1191 if(cur->pict_type == I_TYPE)
1192 cur->ref_count[0] = 0;
1193 if(cur->pict_type != B_TYPE)
1194 cur->ref_count[1] = 0;
1195 for(list=0; list<2; list++){
1196 cur->ref_count[list] = h->ref_count[list];
1197 for(j=0; j<h->ref_count[list]; j++)
1198 cur->ref_poc[list][j] = h->ref_list[list][j].poc;
1199 }
1200 if(cur->pict_type != B_TYPE || h->direct_spatial_mv_pred)
1201 return;
1202 for(list=0; list<2; list++){
1203 for(i=0; i<ref1->ref_count[list]; i++){
1204 const int poc = ref1->ref_poc[list][i];
1205 h->map_col_to_list0[list][i] = PART_NOT_AVAILABLE;
1206 for(j=0; j<h->ref_count[list]; j++)
1207 if(h->ref_list[list][j].poc == poc){
1208 h->map_col_to_list0[list][i] = j;
1209 break;
1210 }
1211 }
1212 }
1213 }
1214
1215 static inline void pred_direct_motion(H264Context * const h, int *mb_type){
1216 MpegEncContext * const s = &h->s;
1217 const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
1218 const int b8_xy = 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1219 const int b4_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1220 const int mb_type_col = h->ref_list[1][0].mb_type[mb_xy];
1221 const int16_t (*l1mv0)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[0][b4_xy];
1222 const int16_t (*l1mv1)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[1][b4_xy];
1223 const int8_t *l1ref0 = &h->ref_list[1][0].ref_index[0][b8_xy];
1224 const int8_t *l1ref1 = &h->ref_list[1][0].ref_index[1][b8_xy];
1225 const int is_b8x8 = IS_8X8(*mb_type);
1226 int sub_mb_type;
1227 int i8, i4;
1228
1229 if(IS_8X8(mb_type_col) && !h->sps.direct_8x8_inference_flag){
1230 /* FIXME save sub mb types from previous frames (or derive from MVs)
1231 * so we know exactly what block size to use */
1232 sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */
1233 *mb_type = MB_TYPE_8x8|MB_TYPE_L0L1;
1234 }else if(!is_b8x8 && (IS_16X16(mb_type_col) || IS_INTRA(mb_type_col))){
1235 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1236 *mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */
1237 }else{
1238 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1239 *mb_type = MB_TYPE_8x8|MB_TYPE_L0L1;
1240 }
1241 if(!is_b8x8)
1242 *mb_type |= MB_TYPE_DIRECT2;
1243
1244 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);
1245
1246 if(h->direct_spatial_mv_pred){
1247 int ref[2];
1248 int mv[2][2];
1249 int list;
1250
1251 /* ref = min(neighbors) */
1252 for(list=0; list<2; list++){
1253 int refa = h->ref_cache[list][scan8[0] - 1];
1254 int refb = h->ref_cache[list][scan8[0] - 8];
1255 int refc = h->ref_cache[list][scan8[0] - 8 + 4];
1256 if(refc == -2)
1257 refc = h->ref_cache[list][scan8[0] - 8 - 1];
1258 ref[list] = refa;
1259 if(ref[list] < 0 || (refb < ref[list] && refb >= 0))
1260 ref[list] = refb;
1261 if(ref[list] < 0 || (refc < ref[list] && refc >= 0))
1262 ref[list] = refc;
1263 if(ref[list] < 0)
1264 ref[list] = -1;
1265 }
1266
1267 if(ref[0] < 0 && ref[1] < 0){
1268 ref[0] = ref[1] = 0;
1269 mv[0][0] = mv[0][1] =
1270 mv[1][0] = mv[1][1] = 0;
1271 }else{
1272 for(list=0; list<2; list++){
1273 if(ref[list] >= 0)
1274 pred_motion(h, 0, 4, list, ref[list], &mv[list][0], &mv[list][1]);
1275 else
1276 mv[list][0] = mv[list][1] = 0;
1277 }
1278 }
1279
1280 if(ref[1] < 0){
1281 *mb_type &= ~MB_TYPE_P0L1;
1282 sub_mb_type &= ~MB_TYPE_P0L1;
1283 }else if(ref[0] < 0){
1284 *mb_type &= ~MB_TYPE_P0L0;
1285 sub_mb_type &= ~MB_TYPE_P0L0;
1286 }
1287
1288 if(IS_16X16(*mb_type)){
1289 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, (uint8_t)ref[0], 1);
1290 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, (uint8_t)ref[1], 1);
1291 if(!IS_INTRA(mb_type_col)
1292 && ( (l1ref0[0] == 0 && ABS(l1mv0[0][0]) <= 1 && ABS(l1mv0[0][1]) <= 1)
1293 || (l1ref0[0] < 0 && l1ref1[0] == 0 && ABS(l1mv1[0][0]) <= 1 && ABS(l1mv1[0][1]) <= 1
1294 && (h->x264_build>33 || !h->x264_build)))){
1295 if(ref[0] > 0)
1296 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1297 else
1298 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
1299 if(ref[1] > 0)
1300 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1301 else
1302 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
1303 }else{
1304 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1305 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1306 }
1307 }else{
1308 for(i8=0; i8<4; i8++){
1309 const int x8 = i8&1;
1310 const int y8 = i8>>1;
1311
1312 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1313 continue;
1314 h->sub_mb_type[i8] = sub_mb_type;
1315
1316 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1317 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1318 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[0], 1);
1319 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, (uint8_t)ref[1], 1);
1320
1321 /* col_zero_flag */
1322 if(!IS_INTRA(mb_type_col) && ( l1ref0[x8 + y8*h->b8_stride] == 0
1323 || (l1ref0[x8 + y8*h->b8_stride] < 0 && l1ref1[x8 + y8*h->b8_stride] == 0
1324 && (h->x264_build>33 || !h->x264_build)))){
1325 const int16_t (*l1mv)[2]= l1ref0[x8 + y8*h->b8_stride] == 0 ? l1mv0 : l1mv1;
1326 if(IS_SUB_8X8(sub_mb_type)){
1327 const int16_t *mv_col = l1mv[x8*3 + y8*3*h->b_stride];
1328 if(ABS(mv_col[0]) <= 1 && ABS(mv_col[1]) <= 1){
1329 if(ref[0] == 0)
1330 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1331 if(ref[1] == 0)
1332 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1333 }
1334 }else
1335 for(i4=0; i4<4; i4++){
1336 const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1337 if(ABS(mv_col[0]) <= 1 && ABS(mv_col[1]) <= 1){
1338 if(ref[0] == 0)
1339 *(uint32_t*)h->mv_cache[0][scan8[i8*4+i4]] = 0;
1340 if(ref[1] == 0)
1341 *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = 0;
1342 }
1343 }
1344 }
1345 }
1346 }
1347 }else{ /* direct temporal mv pred */
1348 if(IS_16X16(*mb_type)){
1349 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1);
1350 if(IS_INTRA(mb_type_col)){
1351 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
1352 fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
1353 fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
1354 }else{
1355 const int ref0 = l1ref0[0] >= 0 ? h->map_col_to_list0[0][l1ref0[0]]
1356 : h->map_col_to_list0[1][l1ref1[0]];
1357 const int dist_scale_factor = h->dist_scale_factor[ref0];
1358 const int16_t *mv_col = l1ref0[0] >= 0 ? l1mv0[0] : l1mv1[0];
1359 int mv_l0[2];
1360 mv_l0[0] = (dist_scale_factor * mv_col[0] + 128) >> 8;
1361 mv_l0[1] = (dist_scale_factor * mv_col[1] + 128) >> 8;
1362 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref0, 1);
1363 fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv_l0[0],mv_l0[1]), 4);
1364 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);
1365 }
1366 }else{
1367 for(i8=0; i8<4; i8++){
1368 const int x8 = i8&1;
1369 const int y8 = i8>>1;
1370 int ref0, dist_scale_factor;
1371 const int16_t (*l1mv)[2]= l1mv0;
1372
1373 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1374 continue;
1375 h->sub_mb_type[i8] = sub_mb_type;
1376 if(IS_INTRA(mb_type_col)){
1377 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1378 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1379 fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1380 fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1381 continue;
1382 }
1383
1384 ref0 = l1ref0[x8 + y8*h->b8_stride];
1385 if(ref0 >= 0)
1386 ref0 = h->map_col_to_list0[0][ref0];
1387 else{
1388 ref0 = h->map_col_to_list0[1][l1ref1[x8 + y8*h->b8_stride]];
1389 l1mv= l1mv1;
1390 }
1391 dist_scale_factor = h->dist_scale_factor[ref0];
1392
1393 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1394 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1395 if(IS_SUB_8X8(sub_mb_type)){
1396 const int16_t *mv_col = l1mv[x8*3 + y8*3*h->b_stride];
1397 int mx = (dist_scale_factor * mv_col[0] + 128) >> 8;
1398 int my = (dist_scale_factor * mv_col[1] + 128) >> 8;
1399 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mx,my), 4);
1400 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mx-mv_col[0],my-mv_col[1]), 4);
1401 }else
1402 for(i4=0; i4<4; i4++){
1403 const int16_t *mv_col = l1mv[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1404 int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]];
1405 mv_l0[0] = (dist_scale_factor * mv_col[0] + 128) >> 8;
1406 mv_l0[1] = (dist_scale_factor * mv_col[1] + 128) >> 8;
1407 *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] =
1408 pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
1409 }
1410 }
1411 }
1412 }
1413 }
1414
1415 static inline void write_back_motion(H264Context *h, int mb_type){
1416 MpegEncContext * const s = &h->s;
1417 const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1418 const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1419 int list;
1420
1421 for(list=0; list<2; list++){
1422 int y;
1423 if(!USES_LIST(mb_type, list)){
1424 if(1){ //FIXME skip or never read if mb_type doesn't use it
1425 for(y=0; y<4; y++){
1426 *(uint64_t*)s->current_picture.motion_val[list][b_xy + 0 + y*h->b_stride]=
1427 *(uint64_t*)s->current_picture.motion_val[list][b_xy + 2 + y*h->b_stride]= 0;
1428 }
1429 if( h->pps.cabac ) {
1430 /* FIXME needed ? */
1431 for(y=0; y<4; y++){
1432 *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]=
1433 *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= 0;
1434 }
1435 }
1436 for(y=0; y<2; y++){
1437 s->current_picture.ref_index[list][b8_xy + 0 + y*h->b8_stride]=
1438 s->current_picture.ref_index[list][b8_xy + 1 + y*h->b8_stride]= LIST_NOT_USED;
1439 }
1440 }
1441 continue;
1442 }
1443
1444 for(y=0; y<4; y++){
1445 *(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];
1446 *(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];
1447 }
1448 if( h->pps.cabac ) {
1449 for(y=0; y<4; y++){
1450 *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+0 + 8*y];
1451 *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+2 + 8*y];
1452 }
1453 }
1454 for(y=0; y<2; y++){
1455 s->current_picture.ref_index[list][b8_xy + 0 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+0 + 16*y];
1456 s->current_picture.ref_index[list][b8_xy + 1 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+2 + 16*y];
1457 }
1458 }
1459
1460 if(h->slice_type == B_TYPE && h->pps.cabac){
1461 if(IS_8X8(mb_type)){
1462 h->direct_table[b8_xy+1+0*h->b8_stride] = IS_DIRECT(h->sub_mb_type[1]) ? 1 : 0;
1463 h->direct_table[b8_xy+0+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[2]) ? 1 : 0;
1464 h->direct_table[b8_xy+1+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[3]) ? 1 : 0;
1465 }
1466 }
1467 }
1468
1469 /**
1470 * Decodes a network abstraction layer unit.
1471 * @param consumed is the number of bytes used as input
1472 * @param length is the length of the array
1473 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp tailing?
1474 * @returns decoded bytes, might be src+1 if no escapes
1475 */
1476 static uint8_t *decode_nal(H264Context *h, uint8_t *src, int *dst_length, int *consumed, int length){
1477 int i, si, di;
1478 uint8_t *dst;
1479
1480 // src[0]&0x80; //forbidden bit
1481 h->nal_ref_idc= src[0]>>5;
1482 h->nal_unit_type= src[0]&0x1F;
1483
1484 src++; length--;
1485 #if 0
1486 for(i=0; i<length; i++)
1487 printf("%2X ", src[i]);
1488 #endif
1489 for(i=0; i+1<length; i+=2){
1490 if(src[i]) continue;
1491 if(i>0 && src[i-1]==0) i--;
1492 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1493 if(src[i+2]!=3){
1494 /* startcode, so we must be past the end */
1495 length=i;
1496 }
1497 break;
1498 }
1499 }
1500
1501 if(i>=length-1){ //no escaped 0
1502 *dst_length= length;
1503 *consumed= length+1; //+1 for the header
1504 return src;
1505 }
1506
1507 h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length);
1508 dst= h->rbsp_buffer;
1509
1510 //printf("decoding esc\n");
1511 si=di=0;
1512 while(si<length){
1513 //remove escapes (very rare 1:2^22)
1514 if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1515 if(src[si+2]==3){ //escape
1516 dst[di++]= 0;
1517 dst[di++]= 0;
1518 si+=3;
1519 continue;
1520 }else //next start code
1521 break;
1522 }
1523
1524 dst[di++]= src[si++];
1525 }
1526
1527 *dst_length= di;
1528 *consumed= si + 1;//+1 for the header
1529 //FIXME store exact number of bits in the getbitcontext (its needed for decoding)
1530 return dst;
1531 }
1532
1533 #if 0
1534 /**
1535 * @param src the data which should be escaped
1536 * @param dst the target buffer, dst+1 == src is allowed as a special case
1537 * @param length the length of the src data
1538 * @param dst_length the length of the dst array
1539 * @returns length of escaped data in bytes or -1 if an error occured
1540 */
1541 static int encode_nal(H264Context *h, uint8_t *dst, uint8_t *src, int length, int dst_length){
1542 int i, escape_count, si, di;
1543 uint8_t *temp;
1544
1545 assert(length>=0);
1546 assert(dst_length>0);
1547
1548 dst[0]= (h->nal_ref_idc<<5) + h->nal_unit_type;
1549
1550 if(length==0) return 1;
1551
1552 escape_count= 0;
1553 for(i=0; i<length; i+=2){
1554 if(src[i]) continue;
1555 if(i>0 && src[i-1]==0)
1556 i--;
1557 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1558 escape_count++;
1559 i+=2;
1560 }
1561 }
1562
1563 if(escape_count==0){
1564 if(dst+1 != src)
1565 memcpy(dst+1, src, length);
1566 return length + 1;
1567 }
1568
1569 if(length + escape_count + 1> dst_length)
1570 return -1;
1571
1572 //this should be damn rare (hopefully)
1573
1574 h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length + escape_count);
1575 temp= h->rbsp_buffer;
1576 //printf("encoding esc\n");
1577
1578 si= 0;
1579 di= 0;
1580 while(si < length){
1581 if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1582 temp[di++]= 0; si++;
1583 temp[di++]= 0; si++;
1584 temp[di++]= 3;
1585 temp[di++]= src[si++];
1586 }
1587 else
1588 temp[di++]= src[si++];
1589 }
1590 memcpy(dst+1, temp, length+escape_count);
1591
1592 assert(di == length+escape_count);
1593
1594 return di + 1;
1595 }
1596
1597 /**
1598 * write 1,10,100,1000,... for alignment, yes its exactly inverse to mpeg4
1599 */
1600 static void encode_rbsp_trailing(PutBitContext *pb){
1601 int length;
1602 put_bits(pb, 1, 1);
1603 length= (-put_bits_count(pb))&7;
1604 if(length) put_bits(pb, length, 0);
1605 }
1606 #endif
1607
1608 /**
1609 * identifies the exact end of the bitstream
1610 * @return the length of the trailing, or 0 if damaged
1611 */
1612 static int decode_rbsp_trailing(uint8_t *src){
1613 int v= *src;
1614 int r;
1615
1616 tprintf("rbsp trailing %X\n", v);
1617
1618 for(r=1; r<9; r++){
1619 if(v&1) return r;
1620 v>>=1;
1621 }
1622 return 0;
1623 }
1624
1625 /**
1626 * idct tranforms the 16 dc values and dequantize them.
1627 * @param qp quantization parameter
1628 */
1629 static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
1630 #define stride 16
1631 int i;
1632 int temp[16]; //FIXME check if this is a good idea
1633 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
1634 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1635
1636 //memset(block, 64, 2*256);
1637 //return;
1638 for(i=0; i<4; i++){
1639 const int offset= y_offset[i];
1640 const int z0= block[offset+stride*0] + block[offset+stride*4];
1641 const int z1= block[offset+stride*0] - block[offset+stride*4];
1642 const int z2= block[offset+stride*1] - block[offset+stride*5];
1643 const int z3= block[offset+stride*1] + block[offset+stride*5];
1644
1645 temp[4*i+0]= z0+z3;
1646 temp[4*i+1]= z1+z2;
1647 temp[4*i+2]= z1-z2;
1648 temp[4*i+3]= z0-z3;
1649 }
1650
1651 for(i=0; i<4; i++){
1652 const int offset= x_offset[i];
1653 const int z0= temp[4*0+i] + temp[4*2+i];
1654 const int z1= temp[4*0+i] - temp[4*2+i];
1655 const int z2= temp[4*1+i] - temp[4*3+i];
1656 const int z3= temp[4*1+i] + temp[4*3+i];
1657
1658 block[stride*0 +offset]= ((((z0 + z3)*qmul + 128 ) >> 8)); //FIXME think about merging this into decode_resdual
1659 block[stride*2 +offset]= ((((z1 + z2)*qmul + 128 ) >> 8));
1660 block[stride*8 +offset]= ((((z1 - z2)*qmul + 128 ) >> 8));
1661 block[stride*10+offset]= ((((z0 - z3)*qmul + 128 ) >> 8));
1662 }
1663 }
1664
1665 #if 0
1666 /**
1667 * dct tranforms the 16 dc values.
1668 * @param qp quantization parameter ??? FIXME
1669 */
1670 static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
1671 // const int qmul= dequant_coeff[qp][0];
1672 int i;
1673 int temp[16]; //FIXME check if this is a good idea
1674 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
1675 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1676
1677 for(i=0; i<4; i++){
1678 const int offset= y_offset[i];
1679 const int z0= block[offset+stride*0] + block[offset+stride*4];
1680 const int z1= block[offset+stride*0] - block[offset+stride*4];
1681 const int z2= block[offset+stride*1] - block[offset+stride*5];
1682 const int z3= block[offset+stride*1] + block[offset+stride*5];
1683
1684 temp[4*i+0]= z0+z3;
1685 temp[4*i+1]= z1+z2;
1686 temp[4*i+2]= z1-z2;
1687 temp[4*i+3]= z0-z3;
1688 }
1689
1690 for(i=0; i<4; i++){
1691 const int offset= x_offset[i];
1692 const int z0= temp[4*0+i] + temp[4*2+i];
1693 const int z1= temp[4*0+i] - temp[4*2+i];
1694 const int z2= temp[4*1+i] - temp[4*3+i];
1695 const int z3= temp[4*1+i] + temp[4*3+i];
1696
1697 block[stride*0 +offset]= (z0 + z3)>>1;
1698 block[stride*2 +offset]= (z1 + z2)>>1;
1699 block[stride*8 +offset]= (z1 - z2)>>1;
1700 block[stride*10+offset]= (z0 - z3)>>1;
1701 }
1702 }
1703 #endif
1704
1705 #undef xStride
1706 #undef stride
1707
1708 static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp, int qmul){
1709 const int stride= 16*2;
1710 const int xStride= 16;
1711 int a,b,c,d,e;
1712
1713 a= block[stride*0 + xStride*0];
1714 b= block[stride*0 + xStride*1];
1715 c= block[stride*1 + xStride*0];
1716 d= block[stride*1 + xStride*1];
1717
1718 e= a-b;
1719 a= a+b;
1720 b= c-d;
1721 c= c+d;
1722
1723 block[stride*0 + xStride*0]= ((a+c)*qmul) >> 7;
1724 block[stride*0 + xStride*1]= ((e+b)*qmul) >> 7;
1725 block[stride*1 + xStride*0]= ((a-c)*qmul) >> 7;
1726 block[stride*1 + xStride*1]= ((e-b)*qmul) >> 7;
1727 }
1728
1729 #if 0
1730 static void chroma_dc_dct_c(DCTELEM *block){
1731 const int stride= 16*2;
1732 const int xStride= 16;
1733 int a,b,c,d,e;
1734
1735 a= block[stride*0 + xStride*0];
1736 b= block[stride*0 + xStride*1];
1737 c= block[stride*1 + xStride*0];
1738 d= block[stride*1 + xStride*1];
1739
1740 e= a-b;
1741 a= a+b;
1742 b= c-d;
1743 c= c+d;
1744
1745 block[stride*0 + xStride*0]= (a+c);
1746 block[stride*0 + xStride*1]= (e+b);
1747 block[stride*1 + xStride*0]= (a-c);
1748 block[stride*1 + xStride*1]= (e-b);
1749 }
1750 #endif
1751
1752 /**
1753 * gets the chroma qp.
1754 */
1755 static inline int get_chroma_qp(int chroma_qp_index_offset, int qscale){
1756
1757 return chroma_qp[clip(qscale + chroma_qp_index_offset, 0, 51)];
1758 }
1759
1760
1761 #if 0
1762 static void h264_diff_dct_c(DCTELEM *block, uint8_t *src1, uint8_t *src2, int stride){
1763 int i;
1764 //FIXME try int temp instead of block
1765
1766 for(i=0; i<4; i++){
1767 const int d0= src1[0 + i*stride] - src2[0 + i*stride];
1768 const int d1= src1[1 + i*stride] - src2[1 + i*stride];
1769 const int d2= src1[2 + i*stride] - src2[2 + i*stride];
1770 const int d3= src1[3 + i*stride] - src2[3 + i*stride];
1771 const int z0= d0 + d3;
1772 const int z3= d0 - d3;
1773 const int z1= d1 + d2;
1774 const int z2= d1 - d2;
1775
1776 block[0 + 4*i]= z0 + z1;
1777 block[1 + 4*i]= 2*z3 + z2;
1778 block[2 + 4*i]= z0 - z1;
1779 block[3 + 4*i]= z3 - 2*z2;
1780 }
1781
1782 for(i=0; i<4; i++){
1783 const int z0= block[0*4 + i] + block[3*4 + i];
1784 const int z3= block[0*4 + i] - block[3*4 + i];
1785 const int z1= block[1*4 + i] + block[2*4 + i];
1786 const int z2= block[1*4 + i] - block[2*4 + i];
1787
1788 block[0*4 + i]= z0 + z1;
1789 block[1*4 + i]= 2*z3 + z2;
1790 block[2*4 + i]= z0 - z1;
1791 block[3*4 + i]= z3 - 2*z2;
1792 }
1793 }
1794 #endif
1795
1796 //FIXME need to check that this doesnt overflow signed 32 bit for low qp, i am not sure, it's very close
1797 //FIXME check that gcc inlines this (and optimizes intra & seperate_dc stuff away)
1798 static inline int quantize_c(DCTELEM *block, uint8_t *scantable, int qscale, int intra, int seperate_dc){
1799 int i;
1800 const int * const quant_table= quant_coeff[qscale];
1801 const int bias= intra ? (1<<QUANT_SHIFT)/3 : (1<<QUANT_SHIFT)/6;
1802 const unsigned int threshold1= (1<<QUANT_SHIFT) - bias - 1;
1803 const unsigned int threshold2= (threshold1<<1);
1804 int last_non_zero;
1805
1806 if(seperate_dc){
1807 if(qscale<=18){
1808 //avoid overflows
1809 const int dc_bias= intra ? (1<<(QUANT_SHIFT-2))/3 : (1<<(QUANT_SHIFT-2))/6;
1810 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT-2)) - dc_bias - 1;
1811 const unsigned int dc_threshold2= (dc_threshold1<<1);
1812
1813 int level= block[0]*quant_coeff[qscale+18][0];
1814 if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1815 if(level>0){
1816 level= (dc_bias + level)>>(QUANT_SHIFT-2);
1817 block[0]= level;
1818 }else{
1819 level= (dc_bias - level)>>(QUANT_SHIFT-2);
1820 block[0]= -level;
1821 }
1822 // last_non_zero = i;
1823 }else{
1824 block[0]=0;
1825 }
1826 }else{
1827 const int dc_bias= intra ? (1<<(QUANT_SHIFT+1))/3 : (1<<(QUANT_SHIFT+1))/6;
1828 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT+1)) - dc_bias - 1;
1829 const unsigned int dc_threshold2= (dc_threshold1<<1);
1830
1831 int level= block[0]*quant_table[0];
1832 if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1833 if(level>0){
1834 level= (dc_bias + level)>>(QUANT_SHIFT+1);
1835 block[0]= level;
1836 }else{
1837 level= (dc_bias - level)>>(QUANT_SHIFT+1);
1838 block[0]= -level;
1839 }
1840 // last_non_zero = i;
1841 }else{
1842 block[0]=0;
1843 }
1844 }
1845 last_non_zero= 0;
1846 i=1;
1847 }else{
1848 last_non_zero= -1;
1849 i=0;
1850 }
1851
1852 for(; i<16; i++){
1853 const int j= scantable[i];
1854 int level= block[j]*quant_table[j];
1855
1856 // if( bias+level >= (1<<(QMAT_SHIFT - 3))
1857 // || bias-level >= (1<<(QMAT_SHIFT - 3))){
1858 if(((unsigned)(level+threshold1))>threshold2){
1859 if(level>0){
1860 level= (bias + level)>>QUANT_SHIFT;
1861 block[j]= level;
1862 }else{
1863 level= (bias - level)>>QUANT_SHIFT;
1864 block[j]= -level;
1865 }
1866 last_non_zero = i;
1867 }else{
1868 block[j]=0;
1869 }
1870 }
1871
1872 return last_non_zero;
1873 }
1874
1875 static void pred4x4_vertical_c(uint8_t *src, uint8_t *topright, int stride){
1876 const uint32_t a= ((uint32_t*)(src-stride))[0];
1877 ((uint32_t*)(src+0*stride))[0]= a;
1878 ((uint32_t*)(src+1*stride))[0]= a;
1879 ((uint32_t*)(src+2*stride))[0]= a;
1880 ((uint32_t*)(src+3*stride))[0]= a;
1881 }
1882
1883 static void pred4x4_horizontal_c(uint8_t *src, uint8_t *topright, int stride){
1884 ((uint32_t*)(src+0*stride))[0]= src[-1+0*stride]*0x01010101;
1885 ((uint32_t*)(src+1*stride))[0]= src[-1+1*stride]*0x01010101;
1886 ((uint32_t*)(src+2*stride))[0]= src[-1+2*stride]*0x01010101;
1887 ((uint32_t*)(src+3*stride))[0]= src[-1+3*stride]*0x01010101;
1888 }
1889
1890 static void pred4x4_dc_c(uint8_t *src, uint8_t *topright, int stride){
1891 const int dc= ( src[-stride] + src[1-stride] + src[2-stride] + src[3-stride]
1892 + src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 4) >>3;
1893
1894 ((uint32_t*)(src+0*stride))[0]=
1895 ((uint32_t*)(src+1*stride))[0]=
1896 ((uint32_t*)(src+2*stride))[0]=
1897 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1898 }
1899
1900 static void pred4x4_left_dc_c(uint8_t *src, uint8_t *topright, int stride){
1901 const int dc= ( src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 2) >>2;
1902
1903 ((uint32_t*)(src+0*stride))[0]=
1904 ((uint32_t*)(src+1*stride))[0]=
1905 ((uint32_t*)(src+2*stride))[0]=
1906 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1907 }
1908
1909 static void pred4x4_top_dc_c(uint8_t *src, uint8_t *topright, int stride){
1910 const int dc= ( src[-stride] + src[1-stride] + src[2-stride] + src[3-stride] + 2) >>2;
1911
1912 ((uint32_t*)(src+0*stride))[0]=
1913 ((uint32_t*)(src+1*stride))[0]=
1914 ((uint32_t*)(src+2*stride))[0]=
1915 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1916 }
1917
1918 static void pred4x4_128_dc_c(uint8_t *src, uint8_t *topright, int stride){
1919 ((uint32_t*)(src+0*stride))[0]=
1920 ((uint32_t*)(src+1*stride))[0]=
1921 ((uint32_t*)(src+2*stride))[0]=
1922 ((uint32_t*)(src+3*stride))[0]= 128U*0x01010101U;
1923 }
1924
1925
1926 #define LOAD_TOP_RIGHT_EDGE\
1927 const int t4= topright[0];\
1928 const int t5= topright[1];\
1929 const int t6= topright[2];\
1930 const int t7= topright[3];\
1931
1932 #define LOAD_LEFT_EDGE\
1933 const int l0= src[-1+0*stride];\
1934 const int l1= src[-1+1*stride];\
1935 const int l2= src[-1+2*stride];\
1936 const int l3= src[-1+3*stride];\
1937
1938 #define LOAD_TOP_EDGE\
1939 const int t0= src[ 0-1*stride];\
1940 const int t1= src[ 1-1*stride];\
1941 const int t2= src[ 2-1*stride];\
1942 const int t3= src[ 3-1*stride];\
1943
1944 static void pred4x4_down_right_c(uint8_t *src, uint8_t *topright, int stride){
1945 const int lt= src[-1-1*stride];
1946 LOAD_TOP_EDGE
1947 LOAD_LEFT_EDGE
1948
1949 src[0+3*stride]=(l3 + 2*l2 + l1 + 2)>>2;
1950 src[0+2*stride]=
1951 src[1+3*stride]=(l2 + 2*l1 + l0 + 2)>>2;
1952 src[0+1*stride]=
1953 src[1+2*stride]=
1954 src[2+3*stride]=(l1 + 2*l0 + lt + 2)>>2;
1955 src[0+0*stride]=
1956 src[1+1*stride]=
1957 src[2+2*stride]=
1958 src[3+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1959 src[1+0*stride]=
1960 src[2+1*stride]=
1961 src[3+2*stride]=(lt + 2*t0 + t1 + 2)>>2;
1962 src[2+0*stride]=
1963 src[3+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1964 src[3+0*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1965 }
1966
1967 static void pred4x4_down_left_c(uint8_t *src, uint8_t *topright, int stride){
1968 LOAD_TOP_EDGE
1969 LOAD_TOP_RIGHT_EDGE
1970 // LOAD_LEFT_EDGE
1971
1972 src[0+0*stride]=(t0 + t2 + 2*t1 + 2)>>2;
1973 src[1+0*stride]=
1974 src[0+1*stride]=(t1 + t3 + 2*t2 + 2)>>2;
1975 src[2+0*stride]=
1976 src[1+1*stride]=
1977 src[0+2*stride]=(t2 + t4 + 2*t3 + 2)>>2;
1978 src[3+0*stride]=
1979 src[2+1*stride]=
1980 src[1+2*stride]=
1981 src[0+3*stride]=(t3 + t5 + 2*t4 + 2)>>2;
1982 src[3+1*stride]=
1983 src[2+2*stride]=
1984 src[1+3*stride]=(t4 + t6 + 2*t5 + 2)>>2;
1985 src[3+2*stride]=
1986 src[2+3*stride]=(t5 + t7 + 2*t6 + 2)>>2;
1987 src[3+3*stride]=(t6 + 3*t7 + 2)>>2;
1988 }
1989
1990 static void pred4x4_vertical_right_c(uint8_t *src, uint8_t *topright, int stride){
1991 const int lt= src[-1-1*stride];
1992 LOAD_TOP_EDGE
1993 LOAD_LEFT_EDGE
1994 const __attribute__((unused)) int unu= l3;
1995
1996 src[0+0*stride]=
1997 src[1+2*stride]=(lt + t0 + 1)>>1;
1998 src[1+0*stride]=
1999 src[2+2*stride]=(t0 + t1 + 1)>>1;
2000 src[2+0*stride]=
2001 src[3+2*stride]=(t1 + t2 + 1)>>1;
2002 src[3+0*stride]=(t2 + t3 + 1)>>1;
2003 src[0+1*stride]=
2004 src[1+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
2005 src[1+1*stride]=
2006 src[2+3*stride]=(lt + 2*t0 + t1 + 2)>>2;
2007 src[2+1*stride]=
2008 src[3+3*stride]=(t0 + 2*t1 + t2 + 2)>>2;
2009 src[3+1*stride]=(t1 + 2*t2 + t3 + 2)>>2;
2010 src[0+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
2011 src[0+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
2012 }
2013
2014 static void pred4x4_vertical_left_c(uint8_t *src, uint8_t *topright, int stride){
2015 LOAD_TOP_EDGE
2016 LOAD_TOP_RIGHT_EDGE
2017 const __attribute__((unused)) int unu= t7;
2018
2019 src[0+0*stride]=(t0 + t1 + 1)>>1;
2020 src[1+0*stride]=
2021 src[0+2*stride]=(t1 + t2 + 1)>>1;
2022 src[2+0*stride]=
2023 src[1+2*stride]=(t2 + t3 + 1)>>1;
2024 src[3+0*stride]=
2025 src[2+2*stride]=(t3 + t4+ 1)>>1;
2026 src[3+2*stride]=(t4 + t5+ 1)>>1;
2027 src[0+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
2028 src[1+1*stride]=
2029 src[0+3*stride]=(t1 + 2*t2 + t3 + 2)>>2;
2030 src[2+1*stride]=
2031 src[1+3*stride]=(t2 + 2*t3 + t4 + 2)>>2;
2032 src[3+1*stride]=
2033 src[2+3*stride]=(t3 + 2*t4 + t5 + 2)>>2;
2034 src[3+3*stride]=(t4 + 2*t5 + t6 + 2)>>2;
2035 }
2036
2037 static void pred4x4_horizontal_up_c(uint8_t *src, uint8_t *topright, int stride){
2038 LOAD_LEFT_EDGE
2039
2040 src[0+0*stride]=(l0 + l1 + 1)>>1;
2041 src[1+0*stride]=(l0 + 2*l1 + l2 + 2)>>2;
2042 src[2+0*stride]=
2043 src[0+1*stride]=(l1 + l2 + 1)>>1;
2044 src[3+0*stride]=
2045 src[1+1*stride]=(l1 + 2*l2 + l3 + 2)>>2;
2046 src[2+1*stride]=
2047 src[0+2*stride]=(l2 + l3 + 1)>>1;
2048 src[3+1*stride]=
2049 src[1+2*stride]=(l2 + 2*l3 + l3 + 2)>>2;
2050 src[3+2*stride]=
2051 src[1+3*stride]=
2052 src[0+3*stride]=
2053 src[2+2*stride]=
2054 src[2+3*stride]=
2055 src[3+3*stride]=l3;
2056 }
2057
2058 static void pred4x4_horizontal_down_c(uint8_t *src, uint8_t *topright, int stride){
2059 const int lt= src[-1-1*stride];
2060 LOAD_TOP_EDGE
2061 LOAD_LEFT_EDGE
2062 const __attribute__((unused)) int unu= t3;
2063
2064 src[0+0*stride]=
2065 src[2+1*stride]=(lt + l0 + 1)>>1;
2066 src[1+0*stride]=
2067 src[3+1*stride]=(l0 + 2*lt + t0 + 2)>>2;
2068 src[2+0*stride]=(lt + 2*t0 + t1 + 2)>>2;
2069 src[3+0*stride]=(t0 + 2*t1 + t2 + 2)>>2;
2070 src[0+1*stride]=
2071 src[2+2*stride]=(l0 + l1 + 1)>>1;
2072 src[1+1*stride]=
2073 src[3+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
2074 src[0+2*stride]=
2075 src[2+3*stride]=(l1 + l2+ 1)>>1;
2076 src[1+2*stride]=
2077 src[3+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
2078 src[0+3*stride]=(l2 + l3 + 1)>>1;
2079 src[1+3*stride]=(l1 + 2*l2 + l3 + 2)>>2;
2080 }
2081
2082 static void pred16x16_vertical_c(uint8_t *src, int stride){
2083 int i;
2084 const uint32_t a= ((uint32_t*)(src-stride))[0];
2085 const uint32_t b= ((uint32_t*)(src-stride))[1];
2086 const uint32_t c= ((uint32_t*)(src-stride))[2];
2087 const uint32_t d= ((uint32_t*)(src-stride))[3];
2088
2089 for(i=0; i<16; i++){
2090 ((uint32_t*)(src+i*stride))[0]= a;
2091 ((uint32_t*)(src+i*stride))[1]= b;
2092 ((uint32_t*)(src+i*stride))[2]= c;
2093 ((uint32_t*)(src+i*stride))[3]= d;
2094 }
2095 }
2096
2097 static void pred16x16_horizontal_c(uint8_t *src, int stride){
2098 int i;
2099
2100 for(i=0; i<16; i++){
2101 ((uint32_t*)(src+i*stride))[0]=
2102 ((uint32_t*)(src+i*stride))[1]=
2103 ((uint32_t*)(src+i*stride))[2]=
2104 ((uint32_t*)(src+i*stride))[3]= src[-1+i*stride]*0x01010101;
2105 }
2106 }
2107
2108 static void pred16x16_dc_c(uint8_t *src, int stride){
2109 int i, dc=0;
2110
2111 for(i=0;i<16; i++){
2112 dc+= src[-1+i*stride];
2113 }
2114
2115 for(i=0;i<16; i++){
2116 dc+= src[i-stride];
2117 }
2118
2119 dc= 0x01010101*((dc + 16)>>5);
2120
2121 for(i=0; i<16; i++){
2122 ((uint32_t*)(src+i*stride))[0]=
2123 ((uint32_t*)(src+i*stride))[1]=
2124 ((uint32_t*)(src+i*stride))[2]=
2125 ((uint32_t*)(src+i*stride))[3]= dc;
2126 }
2127 }
2128
2129 static void pred16x16_left_dc_c(uint8_t *src, int stride){
2130 int i, dc=0;
2131
2132 for(i=0;i<16; i++){
2133 dc+= src[-1+i*stride];
2134 }
2135
2136 dc= 0x01010101*((dc + 8)>>4);
2137
2138 for(i=0; i<16; i++){
2139 ((uint32_t*)(src+i*stride))[0]=
2140 ((uint32_t*)(src+i*stride))[1]=
2141 ((uint32_t*)(src+i*stride))[2]=
2142 ((uint32_t*)(src+i*stride))[3]= dc;
2143 }
2144 }
2145
2146 static void pred16x16_top_dc_c(uint8_t *src, int stride){
2147 int i, dc=0;
2148
2149 for(i=0;i<16; i++){
2150 dc+= src[i-stride];
2151 }
2152 dc= 0x01010101*((dc + 8)>>4);
2153
2154 for(i=0; i<16; i++){
2155 ((uint32_t*)(src+i*stride))[0]=
2156 ((uint32_t*)(src+i*stride))[1]=
2157 ((uint32_t*)(src+i*stride))[2]=
2158 ((uint32_t*)(src+i*stride))[3]= dc;
2159 }
2160 }
2161
2162 static void pred16x16_128_dc_c(uint8_t *src, int stride){
2163 int i;
2164
2165 for(i=0; i<16; i++){
2166 ((uint32_t*)(src+i*stride))[0]=
2167 ((uint32_t*)(src+i*stride))[1]=
2168 ((uint32_t*)(src+i*stride))[2]=
2169 ((uint32_t*)(src+i*stride))[3]= 0x01010101U*128U;
2170 }
2171 }
2172
2173 static inline void pred16x16_plane_compat_c(uint8_t *src, int stride, const int svq3){
2174 int i, j, k;
2175 int a;
2176 uint8_t *cm = cropTbl + MAX_NEG_CROP;
2177 const uint8_t * const src0 = src+7-stride;
2178 const uint8_t *src1 = src+8*stride-1;
2179 const uint8_t *src2 = src1-2*stride; // == src+6*stride-1;
2180 int H = src0[1] - src0[-1];
2181 int V = src1[0] - src2[ 0];
2182 for(k=2; k<=8; ++k) {
2183 src1 += stride; src2 -= stride;
2184 H += k*(src0[k] - src0[-k]);
2185 V += k*(src1[0] - src2[ 0]);
2186 }
2187 if(svq3){
2188 H = ( 5*(H/4) ) / 16;
2189 V = ( 5*(V/4) ) / 16;
2190
2191 /* required for 100% accuracy */
2192 i = H; H = V; V = i;
2193 }else{
2194 H = ( 5*H+32 ) >> 6;
2195 V = ( 5*V+32 ) >> 6;
2196 }
2197
2198 a = 16*(src1[0] + src2[16] + 1) - 7*(V+H);
2199 for(j=16; j>0; --j) {
2200 int b = a;
2201 a += V;
2202 for(i=-16; i<0; i+=4) {
2203 src[16+i] = cm[ (b ) >> 5 ];
2204 src[17+i] = cm[ (b+ H) >> 5 ];
2205 src[18+i] = cm[ (b+2*H) >> 5 ];
2206 src[19+i] = cm[ (b+3*H) >> 5 ];
2207 b += 4*H;
2208 }
2209 src += stride;
2210 }
2211 }
2212
2213 static void pred16x16_plane_c(uint8_t *src, int stride){
2214 pred16x16_plane_compat_c(src, stride, 0);
2215 }
2216
2217 static void pred8x8_vertical_c(uint8_t *src, int stride){
2218 int i;
2219 const uint32_t a= ((uint32_t*)(src-stride))[0];
2220 const uint32_t b= ((uint32_t*)(src-stride))[1];
2221
2222 for(i=0; i<8; i++){
2223 ((uint32_t*)(src+i*stride))[0]= a;
2224 ((uint32_t*)(src+i*stride))[1]= b;
2225 }
2226 }
2227
2228 static void pred8x8_horizontal_c(uint8_t *src, int stride){
2229 int i;
2230
2231 for(i=0; i<8; i++){
2232 ((uint32_t*)(src+i*stride))[0]=
2233 ((uint32_t*)(src+i*stride))[1]= src[-1+i*stride]*0x01010101;
2234 }
2235 }
2236
2237 static void pred8x8_128_dc_c(uint8_t *src, int stride){
2238 int i;
2239
2240 for(i=0; i<8; i++){
2241 ((uint32_t*)(src+i*stride))[0]=
2242 ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
2243 }
2244 }
2245
2246 static void pred8x8_left_dc_c(uint8_t *src, int stride){
2247 int i;
2248 int dc0, dc2;
2249
2250 dc0=dc2=0;
2251 for(i=0;i<4; i++){
2252 dc0+= src[-1+i*stride];
2253 dc2+= src[-1+(i+4)*stride];
2254 }
2255 dc0= 0x01010101*((dc0 + 2)>>2);
2256 dc2= 0x01010101*((dc2 + 2)>>2);
2257
2258 for(i=0; i<4; i++){
2259 ((uint32_t*)(src+i*stride))[0]=
2260 ((uint32_t*)(src+i*stride))[1]= dc0;
2261 }
2262 for(i=4; i<8; i++){
2263 ((uint32_t*)(src+i*stride))[0]=
2264 ((uint32_t*)(src+i*stride))[1]= dc2;
2265 }
2266 }
2267
2268 static void pred8x8_top_dc_c(uint8_t *src, int stride){
2269 int i;
2270 int dc0, dc1;
2271
2272 dc0=dc1=0;
2273 for(i=0;i<4; i++){
2274 dc0+= src[i-stride];
2275 dc1+= src[4+i-stride];
2276 }
2277 dc0= 0x01010101*((dc0 + 2)>>2);
2278 dc1= 0x01010101*((dc1 + 2)>>2);
2279
2280 for(i=0; i<4; i++){
2281 ((uint32_t*)(src+i*stride))[0]= dc0;
2282 ((uint32_t*)(src+i*stride))[1]= dc1;
2283 }
2284 for(i=4; i<8; i++){
2285 ((uint32_t*)(src+i*stride))[0]= dc0;
2286 ((uint32_t*)(src+i*stride))[1]= dc1;
2287 }
2288 }
2289
2290
2291 static void pred8x8_dc_c(uint8_t *src, int stride){
2292 int i;
2293 int dc0, dc1, dc2, dc3;
2294
2295 dc0=dc1=dc2=0;
2296 for(i=0;i<4; i++){
2297 dc0+= src[-1+i*stride] + src[i-stride];
2298 dc1+= src[4+i-stride];
2299 dc2+= src[-1+(i+4)*stride];
2300 }
2301 dc3= 0x01010101*((dc1 + dc2 + 4)>>3);
2302 dc0= 0x01010101*((dc0 + 4)>>3);
2303 dc1= 0x01010101*((dc1 + 2)>>2);
2304 dc2= 0x01010101*((dc2 + 2)>>2);
2305
2306 for(i=0; i<4; i++){
2307 ((uint32_t*)(src+i*stride))[0]= dc0;
2308 ((uint32_t*)(src+i*stride))[1]= dc1;
2309 }
2310 for(i=4; i<8; i++){
2311 ((uint32_t*)(src+i*stride))[0]= dc2;
2312 ((uint32_t*)(src+i*stride))[1]= dc3;
2313 }
2314 }
2315
2316 static void pred8x8_plane_c(uint8_t *src, int stride){
2317 int j, k;
2318 int a;
2319 uint8_t *cm = cropTbl + MAX_NEG_CROP;
2320 const uint8_t * const src0 = src+3-stride;
2321 const uint8_t *src1 = src+4*stride-1;
2322 const uint8_t *src2 = src1-2*stride; // == src+2*stride-1;
2323 int H = src0[1] - src0[-1];
2324 int V = src1[0] - src2[ 0];
2325 for(k=2; k<=4; ++k) {
2326 src1 += stride; src2 -= stride;
2327 H += k*(src0[k] - src0[-k]);
2328 V += k*(src1[0] - src2[ 0]);
2329 }
2330 H = ( 17*H+16 ) >> 5;
2331 V = ( 17*V+16 ) >> 5;
2332
2333 a = 16*(src1[0] + src2[8]+1) - 3*(V+H);
2334 for(j=8; j>0; --j) {
2335 int b = a;
2336 a += V;
2337 src[0] = cm[ (b ) >> 5 ];
2338 src[1] = cm[ (b+ H) >> 5 ];
2339 src[2] = cm[ (b+2*H) >> 5 ];
2340 src[3] = cm[ (b+3*H) >> 5 ];
2341 src[4] = cm[ (b+4*H) >> 5 ];
2342 src[5] = cm[ (b+5*H) >> 5 ];
2343 src[6] = cm[ (b+6*H) >> 5 ];
2344 src[7] = cm[ (b+7*H) >> 5 ];
2345 src += stride;
2346 }
2347 }
2348
2349 #define SRC(x,y) src[(x)+(y)*stride]
2350 #define PL(y) \
2351 const int l##y = (SRC(-1,y-1) + 2*SRC(-1,y) + SRC(-1,y+1) + 2) >> 2;
2352 #define PREDICT_8x8_LOAD_LEFT \
2353 const int l0 = ((has_topleft ? SRC(-1,-1) : SRC(-1,0)) \
2354 + 2*SRC(-1,0) + SRC(-1,1) + 2) >> 2; \
2355 PL(1) PL(2) PL(3) PL(4) PL(5) PL(6) \
2356 const int l7 attribute_unused = (SRC(-1,6) + 3*SRC(-1,7) + 2) >> 2
2357
2358 #define PT(x) \
2359 const int t##x = (SRC(x-1,-1) + 2*SRC(x,-1) + SRC(x+1,-1) + 2) >> 2;
2360 #define PREDICT_8x8_LOAD_TOP \
2361 const int t0 = ((has_topleft ? SRC(-1,-1) : SRC(0,-1)) \
2362 + 2*SRC(0,-1) + SRC(1,-1) + 2) >> 2; \
2363 PT(1) PT(2) PT(3) PT(4) PT(5) PT(6) \
2364 const int t7 attribute_unused = ((has_topright ? SRC(8,-1) : SRC(7,-1)) \
2365 + 2*SRC(7,-1) + SRC(6,-1) + 2) >> 2
2366
2367 #define PTR(x) \
2368 t##x = (SRC(x-1,-1) + 2*SRC(x,-1) + SRC(x+1,-1) + 2) >> 2;
2369 #define PREDICT_8x8_LOAD_TOPRIGHT \
2370 int t8, t9, t10, t11, t12, t13, t14, t15; \
2371 if(has_topright) { \
2372 PTR(8) PTR(9) PTR(10) PTR(11) PTR(12) PTR(13) PTR(14) \
2373 t15 = (SRC(14,-1) + 3*SRC(15,-1) + 2) >> 2; \
2374 } else t8=t9=t10=t11=t12=t13=t14=t15= SRC(7,-1);
2375
2376 #define PREDICT_8x8_LOAD_TOPLEFT \
2377 const int lt = (SRC(-1,0) + 2*SRC(-1,-1) + SRC(0,-1) + 2) >> 2
2378
2379 #define PREDICT_8x8_DC(v) \
2380 int y; \
2381 for( y = 0; y < 8; y++ ) { \
2382 ((uint32_t*)src)[0] = \
2383 ((uint32_t*)src)[1] = v; \
2384 src += stride; \
2385 }
2386
2387 static void pred8x8l_128_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2388 {
2389 PREDICT_8x8_DC(0x80808080);
2390 }
2391 static void pred8x8l_left_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2392 {
2393 PREDICT_8x8_LOAD_LEFT;
2394 const uint32_t dc = ((l0+l1+l2+l3+l4+l5+l6+l7+4) >> 3) * 0x01010101;
2395 PREDICT_8x8_DC(dc);
2396 }
2397 static void pred8x8l_top_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2398 {
2399 PREDICT_8x8_LOAD_TOP;
2400 const uint32_t dc = ((t0+t1+t2+t3+t4+t5+t6+t7+4) >> 3) * 0x01010101;
2401 PREDICT_8x8_DC(dc);
2402 }
2403 static void pred8x8l_dc_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2404 {
2405 PREDICT_8x8_LOAD_LEFT;
2406 PREDICT_8x8_LOAD_TOP;
2407 const uint32_t dc = ((l0+l1+l2+l3+l4+l5+l6+l7
2408 +t0+t1+t2+t3+t4+t5+t6+t7+8) >> 4) * 0x01010101;
2409 PREDICT_8x8_DC(dc);
2410 }
2411 static void pred8x8l_horizontal_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2412 {
2413 PREDICT_8x8_LOAD_LEFT;
2414 #define ROW(y) ((uint32_t*)(src+y*stride))[0] =\
2415 ((uint32_t*)(src+y*stride))[1] = 0x01010101 * l##y
2416 ROW(0); ROW(1); ROW(2); ROW(3); ROW(4); ROW(5); ROW(6); ROW(7);
2417 #undef ROW
2418 }
2419 static void pred8x8l_vertical_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2420 {
2421 int y;
2422 PREDICT_8x8_LOAD_TOP;
2423 src[0] = t0;
2424 src[1] = t1;
2425 src[2] = t2;
2426 src[3] = t3;
2427 src[4] = t4;
2428 src[5] = t5;
2429 src[6] = t6;
2430 src[7] = t7;
2431 for( y = 1; y < 8; y++ )
2432 *(uint64_t*)(src+y*stride) = *(uint64_t*)src;
2433 }
2434 static void pred8x8l_down_left_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2435 {
2436 PREDICT_8x8_LOAD_TOP;
2437 PREDICT_8x8_LOAD_TOPRIGHT;
2438 SRC(0,0)= (t0 + 2*t1 + t2 + 2) >> 2;
2439 SRC(0,1)=SRC(1,0)= (t1 + 2*t2 + t3 + 2) >> 2;
2440 SRC(0,2)=SRC(1,1)=SRC(2,0)= (t2 + 2*t3 + t4 + 2) >> 2;
2441 SRC(0,3)=SRC(1,2)=SRC(2,1)=SRC(3,0)= (t3 + 2*t4 + t5 + 2) >> 2;
2442 SRC(0,4)=SRC(1,3)=SRC(2,2)=SRC(3,1)=SRC(4,0)= (t4 + 2*t5 + t6 + 2) >> 2;
2443 SRC(0,5)=SRC(1,4)=SRC(2,3)=SRC(3,2)=SRC(4,1)=SRC(5,0)= (t5 + 2*t6 + t7 + 2) >> 2;
2444 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;
2445 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;
2446 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;
2447 SRC(2,7)=SRC(3,6)=SRC(4,5)=SRC(5,4)=SRC(6,3)=SRC(7,2)= (t9 + 2*t10 + t11 + 2) >> 2;
2448 SRC(3,7)=SRC(4,6)=SRC(5,5)=SRC(6,4)=SRC(7,3)= (t10 + 2*t11 + t12 + 2) >> 2;
2449 SRC(4,7)=SRC(5,6)=SRC(6,5)=SRC(7,4)= (t11 + 2*t12 + t13 + 2) >> 2;
2450 SRC(5,7)=SRC(6,6)=SRC(7,5)= (t12 + 2*t13 + t14 + 2) >> 2;
2451 SRC(6,7)=SRC(7,6)= (t13 + 2*t14 + t15 + 2) >> 2;
2452 SRC(7,7)= (t14 + 3*t15 + 2) >> 2;
2453 }
2454 static void pred8x8l_down_right_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2455 {
2456 PREDICT_8x8_LOAD_TOP;
2457 PREDICT_8x8_LOAD_LEFT;
2458 PREDICT_8x8_LOAD_TOPLEFT;
2459 SRC(0,7)= (l7 + 2*l6 + l5 + 2) >> 2;
2460 SRC(0,6)=SRC(1,7)= (l6 + 2*l5 + l4 + 2) >> 2;
2461 SRC(0,5)=SRC(1,6)=SRC(2,7)= (l5 + 2*l4 + l3 + 2) >> 2;
2462 SRC(0,4)=SRC(1,5)=SRC(2,6)=SRC(3,7)= (l4 + 2*l3 + l2 + 2) >> 2;
2463 SRC(0,3)=SRC(1,4)=SRC(2,5)=SRC(3,6)=SRC(4,7)= (l3 + 2*l2 + l1 + 2) >> 2;
2464 SRC(0,2)=SRC(1,3)=SRC(2,4)=SRC(3,5)=SRC(4,6)=SRC(5,7)= (l2 + 2*l1 + l0 + 2) >> 2;
2465 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;
2466 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;
2467 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;
2468 SRC(2,0)=SRC(3,1)=SRC(4,2)=SRC(5,3)=SRC(6,4)=SRC(7,5)= (t0 + 2*t1 + t2 + 2) >> 2;
2469 SRC(3,0)=SRC(4,1)=SRC(5,2)=SRC(6,3)=SRC(7,4)= (t1 + 2*t2 + t3 + 2) >> 2;
2470 SRC(4,0)=SRC(5,1)=SRC(6,2)=SRC(7,3)= (t2 + 2*t3 + t4 + 2) >> 2;
2471 SRC(5,0)=SRC(6,1)=SRC(7,2)= (t3 + 2*t4 + t5 + 2) >> 2;
2472 SRC(6,0)=SRC(7,1)= (t4 + 2*t5 + t6 + 2) >> 2;
2473 SRC(7,0)= (t5 + 2*t6 + t7 + 2) >> 2;
2474
2475 }
2476 static void pred8x8l_vertical_right_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2477 {
2478 PREDICT_8x8_LOAD_TOP;
2479 PREDICT_8x8_LOAD_LEFT;
2480 PREDICT_8x8_LOAD_TOPLEFT;
2481 SRC(0,6)= (l5 + 2*l4 + l3 + 2) >> 2;
2482 SRC(0,7)= (l6 + 2*l5 + l4 + 2) >> 2;
2483 SRC(0,4)=SRC(1,6)= (l3 + 2*l2 + l1 + 2) >> 2;
2484 SRC(0,5)=SRC(1,7)= (l4 + 2*l3 + l2 + 2) >> 2;
2485 SRC(0,2)=SRC(1,4)=SRC(2,6)= (l1 + 2*l0 + lt + 2) >> 2;
2486 SRC(0,3)=SRC(1,5)=SRC(2,7)= (l2 + 2*l1 + l0 + 2) >> 2;
2487 SRC(0,1)=SRC(1,3)=SRC(2,5)=SRC(3,7)= (l0 + 2*lt + t0 + 2) >> 2;
2488 SRC(0,0)=SRC(1,2)=SRC(2,4)=SRC(3,6)= (lt + t0 + 1) >> 1;
2489 SRC(1,1)=SRC(2,3)=SRC(3,5)=SRC(4,7)= (lt + 2*t0 + t1 + 2) >> 2;
2490 SRC(1,0)=SRC(2,2)=SRC(3,4)=SRC(4,6)= (t0 + t1 + 1) >> 1;
2491 SRC(2,1)=SRC(3,3)=SRC(4,5)=SRC(5,7)= (t0 + 2*t1 + t2 + 2) >> 2;
2492 SRC(2,0)=SRC(3,2)=SRC(4,4)=SRC(5,6)= (t1 + t2 + 1) >> 1;
2493 SRC(3,1)=SRC(4,3)=SRC(5,5)=SRC(6,7)= (t1 + 2*t2 + t3 + 2) >> 2;
2494 SRC(3,0)=SRC(4,2)=SRC(5,4)=SRC(6,6)= (t2 + t3 + 1) >> 1;
2495 SRC(4,1)=SRC(5,3)=SRC(6,5)=SRC(7,7)= (t2 + 2*t3 + t4 + 2) >> 2;
2496 SRC(4,0)=SRC(5,2)=SRC(6,4)=SRC(7,6)= (t3 + t4 + 1) >> 1;
2497 SRC(5,1)=SRC(6,3)=SRC(7,5)= (t3 + 2*t4 + t5 + 2) >> 2;
2498 SRC(5,0)=SRC(6,2)=SRC(7,4)= (t4 + t5 + 1) >> 1;
2499 SRC(6,1)=SRC(7,3)= (t4 + 2*t5 + t6 + 2) >> 2;
2500 SRC(6,0)=SRC(7,2)= (t5 + t6 + 1) >> 1;
2501 SRC(7,1)= (t5 + 2*t6 + t7 + 2) >> 2;
2502 SRC(7,0)= (t6 + t7 + 1) >> 1;
2503 }
2504 static void pred8x8l_horizontal_down_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2505 {
2506 PREDICT_8x8_LOAD_TOP;
2507 PREDICT_8x8_LOAD_LEFT;
2508 PREDICT_8x8_LOAD_TOPLEFT;
2509 SRC(0,7)= (l6 + l7 + 1) >> 1;
2510 SRC(1,7)= (l5 + 2*l6 + l7 + 2) >> 2;
2511 SRC(0,6)=SRC(2,7)= (l5 + l6 + 1) >> 1;
2512 SRC(1,6)=SRC(3,7)= (l4 + 2*l5 + l6 + 2) >> 2;
2513 SRC(0,5)=SRC(2,6)=SRC(4,7)= (l4 + l5 + 1) >> 1;
2514 SRC(1,5)=SRC(3,6)=SRC(5,7)= (l3 + 2*l4 + l5 + 2) >> 2;
2515 SRC(0,4)=SRC(2,5)=SRC(4,6)=SRC(6,7)= (l3 + l4 + 1) >> 1;
2516 SRC(1,4)=SRC(3,5)=SRC(5,6)=SRC(7,7)= (l2 + 2*l3 + l4 + 2) >> 2;
2517 SRC(0,3)=SRC(2,4)=SRC(4,5)=SRC(6,6)= (l2 + l3 + 1) >> 1;
2518 SRC(1,3)=SRC(3,4)=SRC(5,5)=SRC(7,6)= (l1 + 2*l2 + l3 + 2) >> 2;
2519 SRC(0,2)=SRC(2,3)=SRC(4,4)=SRC(6,5)= (l1 + l2 + 1) >> 1;
2520 SRC(1,2)=SRC(3,3)=SRC(5,4)=SRC(7,5)= (l0 + 2*l1 + l2 + 2) >> 2;
2521 SRC(0,1)=SRC(2,2)=SRC(4,3)=SRC(6,4)= (l0 + l1 + 1) >> 1;
2522 SRC(1,1)=SRC(3,2)=SRC(5,3)=SRC(7,4)= (lt + 2*l0 + l1 + 2) >> 2;
2523 SRC(0,0)=SRC(2,1)=SRC(4,2)=SRC(6,3)= (lt + l0 + 1) >> 1;
2524 SRC(1,0)=SRC(3,1)=SRC(5,2)=SRC(7,3)= (l0 + 2*lt + t0 + 2) >> 2;
2525 SRC(2,0)=SRC(4,1)=SRC(6,2)= (t1 + 2*t0 + lt + 2) >> 2;
2526 SRC(3,0)=SRC(5,1)=SRC(7,2)= (t2 + 2*t1 + t0 + 2) >> 2;
2527 SRC(4,0)=SRC(6,1)= (t3 + 2*t2 + t1 + 2) >> 2;
2528 SRC(5,0)=SRC(7,1)= (t4 + 2*t3 + t2 + 2) >> 2;
2529 SRC(6,0)= (t5 + 2*t4 + t3 + 2) >> 2;
2530 SRC(7,0)= (t6 + 2*t5 + t4 + 2) >> 2;
2531 }
2532 static void pred8x8l_vertical_left_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2533 {
2534 PREDICT_8x8_LOAD_TOP;
2535 PREDICT_8x8_LOAD_TOPRIGHT;
2536 SRC(0,0)= (t0 + t1 + 1) >> 1;
2537 SRC(0,1)= (t0 + 2*t1 + t2 + 2) >> 2;
2538 SRC(0,2)=SRC(1,0)= (t1 + t2 + 1) >> 1;
2539 SRC(0,3)=SRC(1,1)= (t1 + 2*t2 + t3 + 2) >> 2;
2540 SRC(0,4)=SRC(1,2)=SRC(2,0)= (t2 + t3 + 1) >> 1;
2541 SRC(0,5)=SRC(1,3)=SRC(2,1)= (t2 + 2*t3 + t4 + 2) >> 2;
2542 SRC(0,6)=SRC(1,4)=SRC(2,2)=SRC(3,0)= (t3 + t4 + 1) >> 1;
2543 SRC(0,7)=SRC(1,5)=SRC(2,3)=SRC(3,1)= (t3 + 2*t4 + t5 + 2) >> 2;
2544 SRC(1,6)=SRC(2,4)=SRC(3,2)=SRC(4,0)= (t4 + t5 + 1) >> 1;
2545 SRC(1,7)=SRC(2,5)=SRC(3,3)=SRC(4,1)= (t4 + 2*t5 + t6 + 2) >> 2;
2546 SRC(2,6)=SRC(3,4)=SRC(4,2)=SRC(5,0)= (t5 + t6 + 1) >> 1;
2547 SRC(2,7)=SRC(3,5)=SRC(4,3)=SRC(5,1)= (t5 + 2*t6 + t7 + 2) >> 2;
2548 SRC(3,6)=SRC(4,4)=SRC(5,2)=SRC(6,0)= (t6 + t7 + 1) >> 1;
2549 SRC(3,7)=SRC(4,5)=SRC(5,3)=SRC(6,1)= (t6 + 2*t7 + t8 + 2) >> 2;
2550 SRC(4,6)=SRC(5,4)=SRC(6,2)=SRC(7,0)= (t7 + t8 + 1) >> 1;
2551 SRC(4,7)=SRC(5,5)=SRC(6,3)=SRC(7,1)= (t7 + 2*t8 + t9 + 2) >> 2;
2552 SRC(5,6)=SRC(6,4)=SRC(7,2)= (t8 + t9 + 1) >> 1;
2553 SRC(5,7)=SRC(6,5)=SRC(7,3)= (t8 + 2*t9 + t10 + 2) >> 2;
2554 SRC(6,6)=SRC(7,4)= (t9 + t10 + 1) >> 1;
2555 SRC(6,7)=SRC(7,5)= (t9 + 2*t10 + t11 + 2) >> 2;
2556 SRC(7,6)= (t10 + t11 + 1) >> 1;
2557 SRC(7,7)= (t10 + 2*t11 + t12 + 2) >> 2;
2558 }
2559 static void pred8x8l_horizontal_up_c(uint8_t *src, int has_topleft, int has_topright, int stride)
2560 {
2561 PREDICT_8x8_LOAD_LEFT;
2562 SRC(0,0)= (l0 + l1 + 1) >> 1;
2563 SRC(1,0)= (l0 + 2*l1 + l2 + 2) >> 2;
2564 SRC(0,1)=SRC(2,0)= (l1 + l2 + 1) >> 1;
2565 SRC(1,1)=SRC(3,0)= (l1 + 2*l2 + l3 + 2) >> 2;
2566 SRC(0,2)=SRC(2,1)=SRC(4,0)= (l2 + l3 + 1) >> 1;
2567 SRC(1,2)=SRC(3,1)=SRC(5,0)= (l2 + 2*l3 + l4 + 2) >> 2;
2568 SRC(0,3)=SRC(2,2)=SRC(4,1)=SRC(6,0)= (l3 + l4 + 1) >> 1;
2569 SRC(1,3)=SRC(3,2)=SRC(5,1)=SRC(7,0)= (l3 + 2*l4 + l5 + 2) >> 2;
2570 SRC(0,4)=SRC(2,3)=SRC(4,2)=SRC(6,1)= (l4 + l5 + 1) >> 1;
2571 SRC(1,4)=SRC(3,3)=SRC(5,2)=SRC(7,1)= (l4 + 2*l5 + l6 + 2) >> 2;
2572 SRC(0,5)=SRC(2,4)=SRC(4,3)=SRC(6,2)= (l5 + l6 + 1) >> 1;
2573 SRC(1,5)=SRC(3,4)=SRC(5,3)=SRC(7,2)= (l5 + 2*l6 + l7 + 2) >> 2;
2574 SRC(0,6)=SRC(2,5)=SRC(4,4)=SRC(6,3)= (l6 + l7 + 1) >> 1;
2575 SRC(1,6)=SRC(3,5)=SRC(5,4)=SRC(7,3)= (l6 + 3*l7 + 2) >> 2;
2576 SRC(0,7)=SRC(1,7)=SRC(2,6)=SRC(2,7)=SRC(3,6)=
2577 SRC(3,7)=SRC(4,5)=SRC(4,6)=SRC(4,7)=SRC(5,5)=
2578 SRC(5,6)=SRC(5,7)=SRC(6,4)=SRC(6,5)=SRC(6,6)=
2579 SRC(6,7)=SRC(7,4)=SRC(7,5)=SRC(7,6)=SRC(7,7)= l7;
2580 }
2581 #undef PREDICT_8x8_LOAD_LEFT
2582 #undef PREDICT_8x8_LOAD_TOP
2583 #undef PREDICT_8x8_LOAD_TOPLEFT
2584 #undef PREDICT_8x8_LOAD_TOPRIGHT
2585 #undef PREDICT_8x8_DC
2586 #undef PTR
2587 #undef PT
2588 #undef PL
2589 #undef SRC
2590
2591 static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
2592 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2593 int src_x_offset, int src_y_offset,
2594 qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
2595 MpegEncContext * const s = &h->s;
2596 const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
2597 const int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
2598 const int luma_xy= (mx&3) + ((my&3)<<2);
2599 uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*s->linesize;
2600 uint8_t * src_cb= pic->data[1] + (mx>>3) + (my>>3)*s->uvlinesize;
2601 uint8_t * src_cr= pic->data[2] + (mx>>3) + (my>>3)*s->uvlinesize;
2602 int extra_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16; //FIXME increase edge?, IMHO not worth it
2603 int extra_height= extra_width;
2604 int emu=0;
2605 const int full_mx= mx>>2;
2606 const int full_my= my>>2;
2607 const int pic_width = 16*s->mb_width;
2608 const int pic_height = 16*s->mb_height;
2609
2610 assert(pic->data[0]);
2611
2612 if(mx&7) extra_width -= 3;
2613 if(my&7) extra_height -= 3;
2614
2615 if( full_mx < 0-extra_width
2616 || full_my < 0-extra_height
2617 || full_mx + 16/*FIXME*/ > pic_width + extra_width
2618 || full_my + 16/*FIXME*/ > pic_height + extra_height){
2619 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, pic_width, pic_height);
2620 src_y= s->edge_emu_buffer + 2 + 2*s->linesize;
2621 emu=1;
2622 }
2623
2624 qpix_op[luma_xy](dest_y, src_y, s->linesize); //FIXME try variable height perhaps?
2625 if(!square){
2626 qpix_op[luma_xy](dest_y + delta, src_y + delta, s->linesize);
2627 }
2628
2629 if(s->flags&CODEC_FLAG_GRAY) return;
2630
2631 if(emu){
2632 ff_emulated_edge_mc(s->edge_emu_buffer, src_cb, s->uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
2633 src_cb= s->edge_emu_buffer;
2634 }
2635 chroma_op(dest_cb, src_cb, s->uvlinesize, chroma_height, mx&7, my&7);
2636
2637 if(emu){
2638 ff_emulated_edge_mc(s->edge_emu_buffer, src_cr, s->uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
2639 src_cr= s->edge_emu_buffer;
2640 }
2641 chroma_op(dest_cr, src_cr, s->uvlinesize, chroma_height, mx&7, my&7);
2642 }
2643
2644 static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
2645 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2646 int x_offset, int y_offset,
2647 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2648 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
2649 int list0, int list1){
2650 MpegEncContext * const s = &h->s;
2651 qpel_mc_func *qpix_op= qpix_put;
2652 h264_chroma_mc_func chroma_op= chroma_put;
2653
2654 dest_y += 2*x_offset + 2*y_offset*s-> linesize;
2655 dest_cb += x_offset + y_offset*s->uvlinesize;
2656 dest_cr += x_offset + y_offset*s->uvlinesize;
2657 x_offset += 8*s->mb_x;
2658 y_offset += 8*s->mb_y;
2659
2660 if(list0){
2661 Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
2662 mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
2663 dest_y, dest_cb, dest_cr, x_offset, y_offset,
2664 qpix_op, chroma_op);
2665
2666 qpix_op= qpix_avg;
2667 chroma_op= chroma_avg;
2668 }
2669
2670 if(list1){
2671 Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
2672 mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
2673 dest_y, dest_cb, dest_cr, x_offset, y_offset,
2674 qpix_op, chroma_op);
2675 }
2676 }
2677
2678 static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
2679 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2680 int x_offset, int y_offset,
2681 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2682 h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
2683 h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
2684 int list0, int list1){
2685 MpegEncContext * const s = &h->s;
2686
2687 dest_y += 2*x_offset + 2*y_offset*s-> linesize;
2688 dest_cb += x_offset + y_offset*s->uvlinesize;
2689 dest_cr += x_offset + y_offset*s->uvlinesize;
2690 x_offset += 8*s->mb_x;
2691 y_offset += 8*s->mb_y;
2692
2693 if(list0 && list1){
2694 /* don't optimize for luma-only case, since B-frames usually
2695 * use implicit weights => chroma too. */
2696 uint8_t *tmp_cb = s->obmc_scratchpad;
2697 uint8_t *tmp_cr = tmp_cb + 8*s->uvlinesize;
2698 uint8_t *tmp_y = tmp_cr + 8*s->uvlinesize;
2699 int refn0 = h->ref_cache[0][ scan8[n] ];
2700 int refn1 = h->ref_cache[1][ scan8[n] ];
2701
2702 mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
2703 dest_y, dest_cb, dest_cr,
2704 x_offset, y_offset, qpix_put, chroma_put);
2705 mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
2706 tmp_y, tmp_cb, tmp_cr,
2707 x_offset, y_offset, qpix_put, chroma_put);
2708
2709 if(h->use_weight == 2){
2710 int weight0 = h->implicit_weight[refn0][refn1];
2711 int weight1 = 64 - weight0;
2712 luma_weight_avg( dest_y, tmp_y, s-> linesize, 5, weight0, weight1, 0);
2713 chroma_weight_avg(dest_cb, tmp_cb, s->uvlinesize, 5, weight0, weight1, 0);
2714 chroma_weight_avg(dest_cr, tmp_cr, s->uvlinesize, 5, weight0, weight1, 0);
2715 }else{
2716 luma_weight_avg(dest_y, tmp_y, s->linesize, h->luma_log2_weight_denom,
2717 h->luma_weight[0][refn0], h->luma_weight[1][refn1],
2718 h->luma_offset[0][refn0] + h->luma_offset[1][refn1]);
2719 chroma_weight_avg(dest_cb, tmp_cb, s->uvlinesize, h->chroma_log2_weight_denom,
2720 h->chroma_weight[0][refn0][0], h->chroma_weight[1][refn1][0],
2721 h->chroma_offset[0][refn0][0] + h->chroma_offset[1][refn1][0]);
2722 chroma_weight_avg(dest_cr, tmp_cr, s->uvlinesize, h->chroma_log2_weight_denom,
2723 h->chroma_weight[0][refn0][1], h->chroma_weight[1][refn1][1],
2724 h->chroma_offset[0][refn0][1] + h->chroma_offset[1][refn1][1]);
2725 }
2726 }else{
2727 int list = list1 ? 1 : 0;
2728 int refn = h->ref_cache[list][ scan8[n] ];
2729 Picture *ref= &h->ref_list[list][refn];
2730 mc_dir_part(h, ref, n, square, chroma_height, delta, list,
2731 dest_y, dest_cb, dest_cr, x_offset, y_offset,
2732 qpix_put, chroma_put);
2733
2734 luma_weight_op(dest_y, s->linesize, h->luma_log2_weight_denom,
2735 h->luma_weight[list][refn], h->luma_offset[list][refn]);
2736 if(h->use_weight_chroma){
2737 chroma_weight_op(dest_cb, s->uvlinesize, h->chroma_log2_weight_denom,
2738 h->chroma_weight[list][refn][0], h->chroma_offset[list][refn][0]);
2739 chroma_weight_op(dest_cr, s->uvlinesize, h->chroma_log2_weight_denom,
2740 h->chroma_weight[list][refn][1], h->chroma_offset[list][refn][1]);
2741 }
2742 }
2743 }
2744
2745 static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
2746 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2747 int x_offset, int y_offset,
2748 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2749 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
2750 h264_weight_func *weight_op, h264_biweight_func *weight_avg,
2751 int list0, int list1){
2752 if((h->use_weight==2 && list0 && list1
2753 && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ] != 32))
2754 || h->use_weight==1)
2755 mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
2756 x_offset, y_offset, qpix_put, chroma_put,
2757 weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
2758 else
2759 mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
2760 x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
2761 }
2762
2763 static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2764 qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
2765